git/refs/files-backend.c

3293 lines
89 KiB
C
Raw Normal View History

#include "../cache.h"
#include "../config.h"
#include "../refs.h"
#include "refs-internal.h"
#include "ref-cache.h"
#include "packed-backend.h"
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
#include "../iterator.h"
#include "../dir-iterator.h"
#include "../lockfile.h"
#include "../object.h"
#include "../dir.h"
#include "../chdir-notify.h"
refs: new ref types to make per-worktree refs visible to all worktrees One of the problems with multiple worktree is accessing per-worktree refs of one worktree from another worktree. This was sort of solved by multiple ref store, where the code can open the ref store of another worktree and has access to the ref space of that worktree. The problem with this is reporting. "HEAD" in another ref space is also called "HEAD" like in the current ref space. In order to differentiate them, all the code must somehow carry the ref store around and print something like "HEAD from this ref store". But that is not feasible (or possible with a _lot_ of work). With the current design, we pass a reference around as a string (so called "refname"). Extending this design to pass a string _and_ a ref store is a nightmare, especially when handling extended SHA-1 syntax. So we do it another way. Instead of entering a separate ref space, we make refs from other worktrees available in the current ref space. So "HEAD" is always HEAD of the current worktree, but then we can have "worktrees/blah/HEAD" to denote HEAD from a worktree named "blah". This syntax coincidentally matches the underlying directory structure which makes implementation a bit easier. The main worktree has to be treated specially because well... it's special from the beginning. So HEAD from the main worktree is acccessible via the name "main-worktree/HEAD" instead of "worktrees/main/HEAD" because "main" could be just another secondary worktree. This patch also makes it possible to specify refs from one worktree in another one, e.g. git log worktrees/foo/HEAD Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-10-21 08:08:54 +00:00
#include "worktree.h"
/*
* This backend uses the following flags in `ref_update::flags` for
* internal bookkeeping purposes. Their numerical values must not
* conflict with REF_NO_DEREF, REF_FORCE_CREATE_REFLOG, REF_HAVE_NEW,
* REF_HAVE_OLD, or REF_IS_PRUNING, which are also stored in
* `ref_update::flags`.
*/
/*
* Used as a flag in ref_update::flags when a loose ref is being
* pruned. This flag must only be used when REF_NO_DEREF is set.
*/
#define REF_IS_PRUNING (1 << 4)
/*
* Flag passed to lock_ref_sha1_basic() telling it to tolerate broken
* refs (i.e., because the reference is about to be deleted anyway).
*/
#define REF_DELETING (1 << 5)
/*
* Used as a flag in ref_update::flags when the lockfile needs to be
* committed.
*/
#define REF_NEEDS_COMMIT (1 << 6)
/*
* Used as a flag in ref_update::flags when the ref_update was via an
* update to HEAD.
*/
#define REF_UPDATE_VIA_HEAD (1 << 8)
/*
* Used as a flag in ref_update::flags when a reference has been
* deleted and the ref's parent directories may need cleanup.
*/
#define REF_DELETED_RMDIR (1 << 9)
struct ref_lock {
char *ref_name;
struct lock_file lk;
struct object_id old_oid;
};
struct files_ref_store {
struct ref_store base;
unsigned int store_flags;
char *gitcommondir;
struct ref_cache *loose;
struct ref_store *packed_ref_store;
};
static void clear_loose_ref_cache(struct files_ref_store *refs)
{
if (refs->loose) {
free_ref_cache(refs->loose);
refs->loose = NULL;
}
}
resolve_gitlink_ref: ignore non-repository paths When we want to look up a submodule ref, we use get_ref_cache(path) to find or auto-create its ref cache. But if we feed a path that isn't actually a git repository, we blindly create the ref cache, and then may die deeper in the code when we try to access it. This is a problem because many callers speculatively feed us a path that looks vaguely like a repository, and expect us to tell them when it is not. This patch teaches resolve_gitlink_ref to reject non-repository paths without creating a ref_cache. This avoids the die(), and also performs better if you have a large number of these faux-submodule directories (because the ref_cache lookup is linear, under the assumption that there won't be a large number of submodules). To accomplish this, we also break get_ref_cache into two pieces: the lookup and auto-creation (the latter is lumped into create_ref_cache). This lets us first cheaply ask our cache "is it a submodule we know about?" If so, we can avoid repeating our filesystem lookup. So lookups of real submodules are not penalized; they examine the submodule's .git directory only once. The test in t3000 demonstrates a case where this improves correctness (we used to just die). The new perf case in p7300 shows off the speed improvement in an admittedly pathological repository: Test HEAD^ HEAD ---------------------------------------------------------------- 7300.4: ls-files -o 66.97(66.15+0.87) 0.33(0.08+0.24) -99.5% Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-01-22 22:29:30 +00:00
/*
* Create a new submodule ref cache and add it to the internal
* set of caches.
*/
static struct ref_store *files_ref_store_create(const char *gitdir,
unsigned int flags)
{
struct files_ref_store *refs = xcalloc(1, sizeof(*refs));
struct ref_store *ref_store = (struct ref_store *)refs;
struct strbuf sb = STRBUF_INIT;
ref_store->gitdir = xstrdup(gitdir);
base_ref_store_init(ref_store, &refs_be_files);
refs->store_flags = flags;
get_common_dir_noenv(&sb, gitdir);
refs->gitcommondir = strbuf_detach(&sb, NULL);
strbuf_addf(&sb, "%s/packed-refs", refs->gitcommondir);
refs->packed_ref_store = packed_ref_store_create(sb.buf, flags);
strbuf_release(&sb);
chdir_notify_reparent("files-backend $GIT_DIR", &refs->base.gitdir);
chdir_notify_reparent("files-backend $GIT_COMMONDIR",
&refs->gitcommondir);
return ref_store;
resolve_gitlink_ref: ignore non-repository paths When we want to look up a submodule ref, we use get_ref_cache(path) to find or auto-create its ref cache. But if we feed a path that isn't actually a git repository, we blindly create the ref cache, and then may die deeper in the code when we try to access it. This is a problem because many callers speculatively feed us a path that looks vaguely like a repository, and expect us to tell them when it is not. This patch teaches resolve_gitlink_ref to reject non-repository paths without creating a ref_cache. This avoids the die(), and also performs better if you have a large number of these faux-submodule directories (because the ref_cache lookup is linear, under the assumption that there won't be a large number of submodules). To accomplish this, we also break get_ref_cache into two pieces: the lookup and auto-creation (the latter is lumped into create_ref_cache). This lets us first cheaply ask our cache "is it a submodule we know about?" If so, we can avoid repeating our filesystem lookup. So lookups of real submodules are not penalized; they examine the submodule's .git directory only once. The test in t3000 demonstrates a case where this improves correctness (we used to just die). The new perf case in p7300 shows off the speed improvement in an admittedly pathological repository: Test HEAD^ HEAD ---------------------------------------------------------------- 7300.4: ls-files -o 66.97(66.15+0.87) 0.33(0.08+0.24) -99.5% Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-01-22 22:29:30 +00:00
}
/*
* Die if refs is not the main ref store. caller is used in any
* necessary error messages.
*/
static void files_assert_main_repository(struct files_ref_store *refs,
const char *caller)
{
if (refs->store_flags & REF_STORE_MAIN)
return;
BUG("operation %s only allowed for main ref store", caller);
}
resolve_gitlink_ref: ignore non-repository paths When we want to look up a submodule ref, we use get_ref_cache(path) to find or auto-create its ref cache. But if we feed a path that isn't actually a git repository, we blindly create the ref cache, and then may die deeper in the code when we try to access it. This is a problem because many callers speculatively feed us a path that looks vaguely like a repository, and expect us to tell them when it is not. This patch teaches resolve_gitlink_ref to reject non-repository paths without creating a ref_cache. This avoids the die(), and also performs better if you have a large number of these faux-submodule directories (because the ref_cache lookup is linear, under the assumption that there won't be a large number of submodules). To accomplish this, we also break get_ref_cache into two pieces: the lookup and auto-creation (the latter is lumped into create_ref_cache). This lets us first cheaply ask our cache "is it a submodule we know about?" If so, we can avoid repeating our filesystem lookup. So lookups of real submodules are not penalized; they examine the submodule's .git directory only once. The test in t3000 demonstrates a case where this improves correctness (we used to just die). The new perf case in p7300 shows off the speed improvement in an admittedly pathological repository: Test HEAD^ HEAD ---------------------------------------------------------------- 7300.4: ls-files -o 66.97(66.15+0.87) 0.33(0.08+0.24) -99.5% Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-01-22 22:29:30 +00:00
/*
* Downcast ref_store to files_ref_store. Die if ref_store is not a
* files_ref_store. required_flags is compared with ref_store's
* store_flags to ensure the ref_store has all required capabilities.
* "caller" is used in any necessary error messages.
resolve_gitlink_ref: ignore non-repository paths When we want to look up a submodule ref, we use get_ref_cache(path) to find or auto-create its ref cache. But if we feed a path that isn't actually a git repository, we blindly create the ref cache, and then may die deeper in the code when we try to access it. This is a problem because many callers speculatively feed us a path that looks vaguely like a repository, and expect us to tell them when it is not. This patch teaches resolve_gitlink_ref to reject non-repository paths without creating a ref_cache. This avoids the die(), and also performs better if you have a large number of these faux-submodule directories (because the ref_cache lookup is linear, under the assumption that there won't be a large number of submodules). To accomplish this, we also break get_ref_cache into two pieces: the lookup and auto-creation (the latter is lumped into create_ref_cache). This lets us first cheaply ask our cache "is it a submodule we know about?" If so, we can avoid repeating our filesystem lookup. So lookups of real submodules are not penalized; they examine the submodule's .git directory only once. The test in t3000 demonstrates a case where this improves correctness (we used to just die). The new perf case in p7300 shows off the speed improvement in an admittedly pathological repository: Test HEAD^ HEAD ---------------------------------------------------------------- 7300.4: ls-files -o 66.97(66.15+0.87) 0.33(0.08+0.24) -99.5% Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-01-22 22:29:30 +00:00
*/
static struct files_ref_store *files_downcast(struct ref_store *ref_store,
unsigned int required_flags,
const char *caller)
resolve_gitlink_ref: ignore non-repository paths When we want to look up a submodule ref, we use get_ref_cache(path) to find or auto-create its ref cache. But if we feed a path that isn't actually a git repository, we blindly create the ref cache, and then may die deeper in the code when we try to access it. This is a problem because many callers speculatively feed us a path that looks vaguely like a repository, and expect us to tell them when it is not. This patch teaches resolve_gitlink_ref to reject non-repository paths without creating a ref_cache. This avoids the die(), and also performs better if you have a large number of these faux-submodule directories (because the ref_cache lookup is linear, under the assumption that there won't be a large number of submodules). To accomplish this, we also break get_ref_cache into two pieces: the lookup and auto-creation (the latter is lumped into create_ref_cache). This lets us first cheaply ask our cache "is it a submodule we know about?" If so, we can avoid repeating our filesystem lookup. So lookups of real submodules are not penalized; they examine the submodule's .git directory only once. The test in t3000 demonstrates a case where this improves correctness (we used to just die). The new perf case in p7300 shows off the speed improvement in an admittedly pathological repository: Test HEAD^ HEAD ---------------------------------------------------------------- 7300.4: ls-files -o 66.97(66.15+0.87) 0.33(0.08+0.24) -99.5% Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-01-22 22:29:30 +00:00
{
struct files_ref_store *refs;
if (ref_store->be != &refs_be_files)
BUG("ref_store is type \"%s\" not \"files\" in %s",
ref_store->be->name, caller);
refs = (struct files_ref_store *)ref_store;
if ((refs->store_flags & required_flags) != required_flags)
BUG("operation %s requires abilities 0x%x, but only have 0x%x",
caller, required_flags, refs->store_flags);
return refs;
}
refs: new ref types to make per-worktree refs visible to all worktrees One of the problems with multiple worktree is accessing per-worktree refs of one worktree from another worktree. This was sort of solved by multiple ref store, where the code can open the ref store of another worktree and has access to the ref space of that worktree. The problem with this is reporting. "HEAD" in another ref space is also called "HEAD" like in the current ref space. In order to differentiate them, all the code must somehow carry the ref store around and print something like "HEAD from this ref store". But that is not feasible (or possible with a _lot_ of work). With the current design, we pass a reference around as a string (so called "refname"). Extending this design to pass a string _and_ a ref store is a nightmare, especially when handling extended SHA-1 syntax. So we do it another way. Instead of entering a separate ref space, we make refs from other worktrees available in the current ref space. So "HEAD" is always HEAD of the current worktree, but then we can have "worktrees/blah/HEAD" to denote HEAD from a worktree named "blah". This syntax coincidentally matches the underlying directory structure which makes implementation a bit easier. The main worktree has to be treated specially because well... it's special from the beginning. So HEAD from the main worktree is acccessible via the name "main-worktree/HEAD" instead of "worktrees/main/HEAD" because "main" could be just another secondary worktree. This patch also makes it possible to specify refs from one worktree in another one, e.g. git log worktrees/foo/HEAD Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-10-21 08:08:54 +00:00
static void files_reflog_path_other_worktrees(struct files_ref_store *refs,
struct strbuf *sb,
const char *refname)
{
const char *real_ref;
const char *worktree_name;
int length;
if (parse_worktree_ref(refname, &worktree_name, &length, &real_ref))
BUG("refname %s is not a other-worktree ref", refname);
if (worktree_name)
strbuf_addf(sb, "%s/worktrees/%.*s/logs/%s", refs->gitcommondir,
length, worktree_name, real_ref);
else
strbuf_addf(sb, "%s/logs/%s", refs->gitcommondir,
real_ref);
}
static void files_reflog_path(struct files_ref_store *refs,
struct strbuf *sb,
const char *refname)
{
switch (ref_type(refname)) {
case REF_TYPE_PER_WORKTREE:
case REF_TYPE_PSEUDOREF:
strbuf_addf(sb, "%s/logs/%s", refs->base.gitdir, refname);
break;
refs: new ref types to make per-worktree refs visible to all worktrees One of the problems with multiple worktree is accessing per-worktree refs of one worktree from another worktree. This was sort of solved by multiple ref store, where the code can open the ref store of another worktree and has access to the ref space of that worktree. The problem with this is reporting. "HEAD" in another ref space is also called "HEAD" like in the current ref space. In order to differentiate them, all the code must somehow carry the ref store around and print something like "HEAD from this ref store". But that is not feasible (or possible with a _lot_ of work). With the current design, we pass a reference around as a string (so called "refname"). Extending this design to pass a string _and_ a ref store is a nightmare, especially when handling extended SHA-1 syntax. So we do it another way. Instead of entering a separate ref space, we make refs from other worktrees available in the current ref space. So "HEAD" is always HEAD of the current worktree, but then we can have "worktrees/blah/HEAD" to denote HEAD from a worktree named "blah". This syntax coincidentally matches the underlying directory structure which makes implementation a bit easier. The main worktree has to be treated specially because well... it's special from the beginning. So HEAD from the main worktree is acccessible via the name "main-worktree/HEAD" instead of "worktrees/main/HEAD" because "main" could be just another secondary worktree. This patch also makes it possible to specify refs from one worktree in another one, e.g. git log worktrees/foo/HEAD Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-10-21 08:08:54 +00:00
case REF_TYPE_OTHER_PSEUDOREF:
case REF_TYPE_MAIN_PSEUDOREF:
files_reflog_path_other_worktrees(refs, sb, refname);
break;
case REF_TYPE_NORMAL:
strbuf_addf(sb, "%s/logs/%s", refs->gitcommondir, refname);
break;
default:
BUG("unknown ref type %d of ref %s",
ref_type(refname), refname);
}
}
static void files_ref_path(struct files_ref_store *refs,
struct strbuf *sb,
const char *refname)
{
switch (ref_type(refname)) {
case REF_TYPE_PER_WORKTREE:
case REF_TYPE_PSEUDOREF:
strbuf_addf(sb, "%s/%s", refs->base.gitdir, refname);
break;
refs: new ref types to make per-worktree refs visible to all worktrees One of the problems with multiple worktree is accessing per-worktree refs of one worktree from another worktree. This was sort of solved by multiple ref store, where the code can open the ref store of another worktree and has access to the ref space of that worktree. The problem with this is reporting. "HEAD" in another ref space is also called "HEAD" like in the current ref space. In order to differentiate them, all the code must somehow carry the ref store around and print something like "HEAD from this ref store". But that is not feasible (or possible with a _lot_ of work). With the current design, we pass a reference around as a string (so called "refname"). Extending this design to pass a string _and_ a ref store is a nightmare, especially when handling extended SHA-1 syntax. So we do it another way. Instead of entering a separate ref space, we make refs from other worktrees available in the current ref space. So "HEAD" is always HEAD of the current worktree, but then we can have "worktrees/blah/HEAD" to denote HEAD from a worktree named "blah". This syntax coincidentally matches the underlying directory structure which makes implementation a bit easier. The main worktree has to be treated specially because well... it's special from the beginning. So HEAD from the main worktree is acccessible via the name "main-worktree/HEAD" instead of "worktrees/main/HEAD" because "main" could be just another secondary worktree. This patch also makes it possible to specify refs from one worktree in another one, e.g. git log worktrees/foo/HEAD Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-10-21 08:08:54 +00:00
case REF_TYPE_MAIN_PSEUDOREF:
if (!skip_prefix(refname, "main-worktree/", &refname))
BUG("ref %s is not a main pseudoref", refname);
/* fallthrough */
case REF_TYPE_OTHER_PSEUDOREF:
case REF_TYPE_NORMAL:
strbuf_addf(sb, "%s/%s", refs->gitcommondir, refname);
break;
default:
BUG("unknown ref type %d of ref %s",
ref_type(refname), refname);
}
}
/*
* Manually add refs/bisect, refs/rewritten and refs/worktree, which, being
* per-worktree, might not appear in the directory listing for
* refs/ in the main repo.
*/
static void add_per_worktree_entries_to_dir(struct ref_dir *dir, const char *dirname)
{
const char *prefixes[] = { "refs/bisect/", "refs/worktree/", "refs/rewritten/" };
int ip;
if (strcmp(dirname, "refs/"))
return;
for (ip = 0; ip < ARRAY_SIZE(prefixes); ip++) {
const char *prefix = prefixes[ip];
int prefix_len = strlen(prefix);
struct ref_entry *child_entry;
int pos;
pos = search_ref_dir(dir, prefix, prefix_len);
if (pos >= 0)
continue;
child_entry = create_dir_entry(dir->cache, prefix, prefix_len);
add_entry_to_dir(dir, child_entry);
}
}
/*
* Read the loose references from the namespace dirname into dir
* (without recursing). dirname must end with '/'. dir must be the
* directory entry corresponding to dirname.
*/
static void loose_fill_ref_dir(struct ref_store *ref_store,
struct ref_dir *dir, const char *dirname)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_READ, "fill_ref_dir");
DIR *d;
struct dirent *de;
int dirnamelen = strlen(dirname);
struct strbuf refname;
struct strbuf path = STRBUF_INIT;
size_t path_baselen;
files_ref_path(refs, &path, dirname);
path_baselen = path.len;
d = opendir(path.buf);
if (!d) {
strbuf_release(&path);
return;
}
strbuf_init(&refname, dirnamelen + 257);
strbuf_add(&refname, dirname, dirnamelen);
while ((de = readdir(d)) != NULL) {
struct object_id oid;
struct stat st;
int flag;
if (de->d_name[0] == '.')
continue;
if (ends_with(de->d_name, ".lock"))
continue;
strbuf_addstr(&refname, de->d_name);
strbuf_addstr(&path, de->d_name);
if (stat(path.buf, &st) < 0) {
; /* silently ignore */
} else if (S_ISDIR(st.st_mode)) {
strbuf_addch(&refname, '/');
add_entry_to_dir(dir,
create_dir_entry(dir->cache, refname.buf,
refname.len));
} else {
if (!refs_resolve_ref_unsafe(&refs->base,
refname.buf,
RESOLVE_REF_READING,
&oid, &flag)) {
oidclr(&oid);
flag |= REF_ISBROKEN;
} else if (is_null_oid(&oid)) {
/*
* It is so astronomically unlikely
* that null_oid is the OID of an
* actual object that we consider its
* appearance in a loose reference
* file to be repo corruption
* (probably due to a software bug).
*/
flag |= REF_ISBROKEN;
}
if (check_refname_format(refname.buf,
REFNAME_ALLOW_ONELEVEL)) {
if (!refname_is_safe(refname.buf))
die("loose refname is dangerous: %s", refname.buf);
oidclr(&oid);
flag |= REF_BAD_NAME | REF_ISBROKEN;
}
add_entry_to_dir(dir,
create_ref_entry(refname.buf, &oid, flag));
}
strbuf_setlen(&refname, dirnamelen);
strbuf_setlen(&path, path_baselen);
}
strbuf_release(&refname);
strbuf_release(&path);
closedir(d);
add_per_worktree_entries_to_dir(dir, dirname);
}
static struct ref_cache *get_loose_ref_cache(struct files_ref_store *refs)
{
if (!refs->loose) {
/*
* Mark the top-level directory complete because we
* are about to read the only subdirectory that can
* hold references:
*/
refs->loose = create_ref_cache(&refs->base, loose_fill_ref_dir);
/* We're going to fill the top level ourselves: */
refs->loose->root->flag &= ~REF_INCOMPLETE;
/*
* Add an incomplete entry for "refs/" (to be filled
* lazily):
*/
add_entry_to_dir(get_ref_dir(refs->loose->root),
create_dir_entry(refs->loose, "refs/", 5));
}
return refs->loose;
}
static int files_read_raw_ref(struct ref_store *ref_store, const char *refname,
struct object_id *oid, struct strbuf *referent,
unsigned int *type, int *failure_errno)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_READ, "read_raw_ref");
struct strbuf sb_contents = STRBUF_INIT;
struct strbuf sb_path = STRBUF_INIT;
const char *path;
const char *buf;
struct stat st;
int fd;
int ret = -1;
int remaining_retries = 3;
int myerr = 0;
*type = 0;
strbuf_reset(&sb_path);
files_ref_path(refs, &sb_path, refname);
path = sb_path.buf;
stat_ref:
/*
* We might have to loop back here to avoid a race
* condition: first we lstat() the file, then we try
* to read it as a link or as a file. But if somebody
* changes the type of the file (file <-> directory
* <-> symlink) between the lstat() and reading, then
* we don't want to report that as an error but rather
* try again starting with the lstat().
*
* We'll keep a count of the retries, though, just to avoid
* any confusing situation sending us into an infinite loop.
*/
if (remaining_retries-- <= 0)
goto out;
if (lstat(path, &st) < 0) {
int ignore_errno;
myerr = errno;
errno = 0;
if (myerr != ENOENT)
goto out;
if (refs_read_raw_ref(refs->packed_ref_store, refname, oid,
referent, type, &ignore_errno)) {
myerr = ENOENT;
goto out;
}
ret = 0;
goto out;
}
/* Follow "normalized" - ie "refs/.." symlinks by hand */
if (S_ISLNK(st.st_mode)) {
strbuf_reset(&sb_contents);
if (strbuf_readlink(&sb_contents, path, st.st_size) < 0) {
myerr = errno;
errno = 0;
if (myerr == ENOENT || myerr == EINVAL)
/* inconsistent with lstat; retry */
goto stat_ref;
else
goto out;
}
if (starts_with(sb_contents.buf, "refs/") &&
!check_refname_format(sb_contents.buf, 0)) {
strbuf_swap(&sb_contents, referent);
*type |= REF_ISSYMREF;
ret = 0;
goto out;
}
/*
* It doesn't look like a refname; fall through to just
* treating it like a non-symlink, and reading whatever it
* points to.
*/
}
/* Is it a directory? */
if (S_ISDIR(st.st_mode)) {
int ignore_errno;
/*
* Even though there is a directory where the loose
* ref is supposed to be, there could still be a
* packed ref:
*/
if (refs_read_raw_ref(refs->packed_ref_store, refname, oid,
referent, type, &ignore_errno)) {
myerr = EISDIR;
goto out;
}
ret = 0;
goto out;
}
/*
* Anything else, just open it and try to use it as
* a ref
*/
fd = open(path, O_RDONLY);
if (fd < 0) {
myerr = errno;
if (myerr == ENOENT && !S_ISLNK(st.st_mode))
/* inconsistent with lstat; retry */
goto stat_ref;
else
goto out;
}
strbuf_reset(&sb_contents);
if (strbuf_read(&sb_contents, fd, 256) < 0) {
myerr = errno;
close(fd);
goto out;
}
close(fd);
strbuf_rtrim(&sb_contents);
buf = sb_contents.buf;
ret = parse_loose_ref_contents(buf, oid, referent, type, &myerr);
out:
if (ret && !myerr)
BUG("returning non-zero %d, should have set myerr!", ret);
*failure_errno = myerr;
strbuf_release(&sb_path);
strbuf_release(&sb_contents);
return ret;
}
int parse_loose_ref_contents(const char *buf, struct object_id *oid,
struct strbuf *referent, unsigned int *type,
int *failure_errno)
{
const char *p;
if (skip_prefix(buf, "ref:", &buf)) {
while (isspace(*buf))
buf++;
strbuf_reset(referent);
strbuf_addstr(referent, buf);
*type |= REF_ISSYMREF;
return 0;
}
/*
* FETCH_HEAD has additional data after the sha.
*/
if (parse_oid_hex(buf, oid, &p) ||
(*p != '\0' && !isspace(*p))) {
*type |= REF_ISBROKEN;
*failure_errno = EINVAL;
return -1;
}
return 0;
}
static void unlock_ref(struct ref_lock *lock)
{
rollback_lock_file(&lock->lk);
free(lock->ref_name);
free(lock);
}
/*
* Lock refname, without following symrefs, and set *lock_p to point
* at a newly-allocated lock object. Fill in lock->old_oid, referent,
* and type similarly to read_raw_ref().
*
* The caller must verify that refname is a "safe" reference name (in
* the sense of refname_is_safe()) before calling this function.
*
* If the reference doesn't already exist, verify that refname doesn't
* have a D/F conflict with any existing references. extras and skip
* are passed to refs_verify_refname_available() for this check.
*
* If mustexist is not set and the reference is not found or is
* broken, lock the reference anyway but clear old_oid.
*
* Return 0 on success. On failure, write an error message to err and
* return TRANSACTION_NAME_CONFLICT or TRANSACTION_GENERIC_ERROR.
*
* Implementation note: This function is basically
*
* lock reference
* read_raw_ref()
*
* but it includes a lot more code to
* - Deal with possible races with other processes
* - Avoid calling refs_verify_refname_available() when it can be
* avoided, namely if we were successfully able to read the ref
* - Generate informative error messages in the case of failure
*/
static int lock_raw_ref(struct files_ref_store *refs,
const char *refname, int mustexist,
const struct string_list *extras,
struct ref_lock **lock_p,
struct strbuf *referent,
unsigned int *type,
struct strbuf *err)
{
struct ref_lock *lock;
struct strbuf ref_file = STRBUF_INIT;
int attempts_remaining = 3;
int ret = TRANSACTION_GENERIC_ERROR;
int failure_errno;
assert(err);
files_assert_main_repository(refs, "lock_raw_ref");
*type = 0;
/* First lock the file so it can't change out from under us. */
*lock_p = CALLOC_ARRAY(lock, 1);
lock->ref_name = xstrdup(refname);
files_ref_path(refs, &ref_file, refname);
retry:
switch (safe_create_leading_directories(ref_file.buf)) {
case SCLD_OK:
break; /* success */
case SCLD_EXISTS:
/*
* Suppose refname is "refs/foo/bar". We just failed
* to create the containing directory, "refs/foo",
* because there was a non-directory in the way. This
* indicates a D/F conflict, probably because of
* another reference such as "refs/foo". There is no
* reason to expect this error to be transitory.
*/
if (refs_verify_refname_available(&refs->base, refname,
extras, NULL, err)) {
if (mustexist) {
/*
* To the user the relevant error is
* that the "mustexist" reference is
* missing:
*/
strbuf_reset(err);
strbuf_addf(err, "unable to resolve reference '%s'",
refname);
} else {
/*
* The error message set by
* refs_verify_refname_available() is
* OK.
*/
ret = TRANSACTION_NAME_CONFLICT;
}
} else {
/*
* The file that is in the way isn't a loose
* reference. Report it as a low-level
* failure.
*/
strbuf_addf(err, "unable to create lock file %s.lock; "
"non-directory in the way",
ref_file.buf);
}
goto error_return;
case SCLD_VANISHED:
/* Maybe another process was tidying up. Try again. */
if (--attempts_remaining > 0)
goto retry;
/* fall through */
default:
strbuf_addf(err, "unable to create directory for %s",
ref_file.buf);
goto error_return;
}
if (hold_lock_file_for_update_timeout(
&lock->lk, ref_file.buf, LOCK_NO_DEREF,
get_files_ref_lock_timeout_ms()) < 0) {
int myerr = errno;
errno = 0;
if (myerr == ENOENT && --attempts_remaining > 0) {
/*
* Maybe somebody just deleted one of the
* directories leading to ref_file. Try
* again:
*/
goto retry;
} else {
unable_to_lock_message(ref_file.buf, myerr, err);
goto error_return;
}
}
/*
* Now we hold the lock and can read the reference without
* fear that its value will change.
*/
if (files_read_raw_ref(&refs->base, refname, &lock->old_oid, referent,
type, &failure_errno)) {
if (failure_errno == ENOENT) {
if (mustexist) {
/* Garden variety missing reference. */
strbuf_addf(err, "unable to resolve reference '%s'",
refname);
goto error_return;
} else {
/*
* Reference is missing, but that's OK. We
* know that there is not a conflict with
* another loose reference because
* (supposing that we are trying to lock
* reference "refs/foo/bar"):
*
* - We were successfully able to create
* the lockfile refs/foo/bar.lock, so we
* know there cannot be a loose reference
* named "refs/foo".
*
* - We got ENOENT and not EISDIR, so we
* know that there cannot be a loose
* reference named "refs/foo/bar/baz".
*/
}
} else if (failure_errno == EISDIR) {
/*
* There is a directory in the way. It might have
* contained references that have been deleted. If
* we don't require that the reference already
* exists, try to remove the directory so that it
* doesn't cause trouble when we want to rename the
* lockfile into place later.
*/
if (mustexist) {
/* Garden variety missing reference. */
strbuf_addf(err, "unable to resolve reference '%s'",
refname);
goto error_return;
} else if (remove_dir_recursively(&ref_file,
REMOVE_DIR_EMPTY_ONLY)) {
if (refs_verify_refname_available(
&refs->base, refname,
extras, NULL, err)) {
/*
* The error message set by
* verify_refname_available() is OK.
*/
ret = TRANSACTION_NAME_CONFLICT;
goto error_return;
} else {
/*
* We can't delete the directory,
* but we also don't know of any
* references that it should
* contain.
*/
strbuf_addf(err, "there is a non-empty directory '%s' "
"blocking reference '%s'",
ref_file.buf, refname);
goto error_return;
}
}
} else if (failure_errno == EINVAL && (*type & REF_ISBROKEN)) {
strbuf_addf(err, "unable to resolve reference '%s': "
"reference broken", refname);
goto error_return;
} else {
strbuf_addf(err, "unable to resolve reference '%s': %s",
refname, strerror(failure_errno));
goto error_return;
}
/*
* If the ref did not exist and we are creating it,
files-backend: cheapen refname_available check when locking refs When locking references in preparation for updating them, we need to check that none of the newly added references D/F conflict with existing references (e.g., we don't allow `refs/foo` to be added if `refs/foo/bar` already exists, or vice versa). Prior to 524a9fdb51 (refs_verify_refname_available(): use function in more places, 2017-04-16), conflicts with existing loose references were checked by looking directly in the filesystem, and then conflicts with existing packed references were checked by running `verify_refname_available_dir()` against the packed-refs cache. But that commit changed the final check to call `refs_verify_refname_available()` against the *whole* files ref-store, including both loose and packed references, with the following comment: > This means that those callsites now check for conflicts with all > references rather than just packed refs, but the performance cost > shouldn't be significant (and will be regained later). That comment turned out to be too sanguine. User s@kazlauskas.me reported that fetches involving a very large number of references in neighboring directories were slowed down by that change. The problem is that when fetching, each reference is updated individually, within its own reference transaction. This is done because some reference updates might succeed even though others fail. But every time a reference update transaction is finished, `clear_loose_ref_cache()` is called. So when it is time to update the next reference, part of the loose ref cache has to be repopulated for the `refs_verify_refname_available()` call. If the references are all in neighboring directories, then the cost of repopulating the reference cache increases with the number of references, resulting in O(N²) effort. The comment above also claims that the performance cost "will be regained later". The idea was that once the packed-refs were finished being split out into a separate ref-store, we could limit the `refs_verify_refname_available()` call to the packed references again. That is what we do now. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-08-17 15:12:50 +00:00
* make sure there is no existing packed ref that
* conflicts with refname:
*/
if (refs_verify_refname_available(
files-backend: cheapen refname_available check when locking refs When locking references in preparation for updating them, we need to check that none of the newly added references D/F conflict with existing references (e.g., we don't allow `refs/foo` to be added if `refs/foo/bar` already exists, or vice versa). Prior to 524a9fdb51 (refs_verify_refname_available(): use function in more places, 2017-04-16), conflicts with existing loose references were checked by looking directly in the filesystem, and then conflicts with existing packed references were checked by running `verify_refname_available_dir()` against the packed-refs cache. But that commit changed the final check to call `refs_verify_refname_available()` against the *whole* files ref-store, including both loose and packed references, with the following comment: > This means that those callsites now check for conflicts with all > references rather than just packed refs, but the performance cost > shouldn't be significant (and will be regained later). That comment turned out to be too sanguine. User s@kazlauskas.me reported that fetches involving a very large number of references in neighboring directories were slowed down by that change. The problem is that when fetching, each reference is updated individually, within its own reference transaction. This is done because some reference updates might succeed even though others fail. But every time a reference update transaction is finished, `clear_loose_ref_cache()` is called. So when it is time to update the next reference, part of the loose ref cache has to be repopulated for the `refs_verify_refname_available()` call. If the references are all in neighboring directories, then the cost of repopulating the reference cache increases with the number of references, resulting in O(N²) effort. The comment above also claims that the performance cost "will be regained later". The idea was that once the packed-refs were finished being split out into a separate ref-store, we could limit the `refs_verify_refname_available()` call to the packed references again. That is what we do now. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-08-17 15:12:50 +00:00
refs->packed_ref_store, refname,
extras, NULL, err))
goto error_return;
}
ret = 0;
goto out;
error_return:
unlock_ref(lock);
*lock_p = NULL;
out:
strbuf_release(&ref_file);
return ret;
}
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
struct files_ref_iterator {
struct ref_iterator base;
struct ref_iterator *iter0;
unsigned int flags;
};
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
static int files_ref_iterator_advance(struct ref_iterator *ref_iterator)
{
struct files_ref_iterator *iter =
(struct files_ref_iterator *)ref_iterator;
int ok;
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
while ((ok = ref_iterator_advance(iter->iter0)) == ITER_OK) {
if (iter->flags & DO_FOR_EACH_PER_WORKTREE_ONLY &&
ref_type(iter->iter0->refname) != REF_TYPE_PER_WORKTREE)
continue;
refs: add DO_FOR_EACH_OMIT_DANGLING_SYMREFS flag When the DO_FOR_EACH_INCLUDE_BROKEN flag is used, we include both actual corrupt refs (illegal names, missing objects), but also symrefs that point to nothing. This latter is not really a corruption, but just something that may happen normally. For example, the symref at refs/remotes/origin/HEAD may point to a tracking branch which is later deleted. (The local HEAD may also be unborn, of course, but we do not access it through ref iteration). Most callers of for_each_ref() etc, do not care. They don't pass INCLUDE_BROKEN, so don't see it at all. But for those which do pass it, this somewhat-normal state causes extra warnings (e.g., from for-each-ref) or even aborts operations (destructive repacks with GIT_REF_PARANOIA set). This patch just introduces the flag and the mechanism; there are no callers yet (and hence no tests). Two things to note on the implementation: - we actually skip any symref that does not resolve to a ref. This includes ones which point to an invalidly-named ref. You could argue this is a more serious breakage than simple dangling. But the overall effect is the same (we could not follow the symref), as well as the impact on things like REF_PARANOIA (either way, a symref we can't follow won't impact reachability, because we'll see the ref itself during iteration). The underlying resolution function doesn't distinguish these two cases (they both get REF_ISBROKEN). - we change the iterator in refs/files-backend.c where we check INCLUDE_BROKEN. There's a matching spot in refs/packed-backend.c, but we don't know need to do anything there. The packed backend does not support symrefs at all. The resulting set of flags might be a bit easier to follow if we broke this down into "INCLUDE_CORRUPT_REFS" and "INCLUDE_DANGLING_SYMREFS". But there are a few reasons not do so: - adding a new OMIT_DANGLING_SYMREFS flag lets us leave existing callers intact, without changing their behavior (and some of them really do want to see the dangling symrefs; e.g., t5505 has a test which expects us to report when a symref becomes dangling) - they're not actually independent. You cannot say "include dangling symrefs" without also including refs whose objects are not reachable, because dangling symrefs by definition do not have an object. We could tweak the implementation to distinguish this, but in practice nobody wants to ask for that. Adding the OMIT flag keeps the implementation simple and makes sure we don't regress the current behavior. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Jonathan Tan <jonathantanmy@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-09-24 18:41:32 +00:00
if ((iter->flags & DO_FOR_EACH_OMIT_DANGLING_SYMREFS) &&
(iter->iter0->flags & REF_ISSYMREF) &&
(iter->iter0->flags & REF_ISBROKEN))
continue;
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
if (!(iter->flags & DO_FOR_EACH_INCLUDE_BROKEN) &&
!ref_resolves_to_object(iter->iter0->refname,
iter->iter0->oid,
iter->iter0->flags))
continue;
iter->base.refname = iter->iter0->refname;
iter->base.oid = iter->iter0->oid;
iter->base.flags = iter->iter0->flags;
return ITER_OK;
}
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
iter->iter0 = NULL;
if (ref_iterator_abort(ref_iterator) != ITER_DONE)
ok = ITER_ERROR;
return ok;
}
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
static int files_ref_iterator_peel(struct ref_iterator *ref_iterator,
struct object_id *peeled)
{
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
struct files_ref_iterator *iter =
(struct files_ref_iterator *)ref_iterator;
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
return ref_iterator_peel(iter->iter0, peeled);
}
static int files_ref_iterator_abort(struct ref_iterator *ref_iterator)
{
struct files_ref_iterator *iter =
(struct files_ref_iterator *)ref_iterator;
int ok = ITER_DONE;
if (iter->iter0)
ok = ref_iterator_abort(iter->iter0);
base_ref_iterator_free(ref_iterator);
return ok;
}
static struct ref_iterator_vtable files_ref_iterator_vtable = {
files_ref_iterator_advance,
files_ref_iterator_peel,
files_ref_iterator_abort
};
static struct ref_iterator *files_ref_iterator_begin(
struct ref_store *ref_store,
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
const char *prefix, unsigned int flags)
{
struct files_ref_store *refs;
struct ref_iterator *loose_iter, *packed_iter, *overlay_iter;
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
struct files_ref_iterator *iter;
struct ref_iterator *ref_iterator;
unsigned int required_flags = REF_STORE_READ;
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
if (!(flags & DO_FOR_EACH_INCLUDE_BROKEN))
required_flags |= REF_STORE_ODB;
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
refs = files_downcast(ref_store, required_flags, "ref_iterator_begin");
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
/*
* We must make sure that all loose refs are read before
* accessing the packed-refs file; this avoids a race
* condition if loose refs are migrated to the packed-refs
* file by a simultaneous process, but our in-memory view is
* from before the migration. We ensure this as follows:
* First, we call start the loose refs iteration with its
* `prime_ref` argument set to true. This causes the loose
* references in the subtree to be pre-read into the cache.
* (If they've already been read, that's OK; we only need to
* guarantee that they're read before the packed refs, not
* *how much* before.) After that, we call
* packed_ref_iterator_begin(), which internally checks
* whether the packed-ref cache is up to date with what is on
* disk, and re-reads it if not.
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
*/
loose_iter = cache_ref_iterator_begin(get_loose_ref_cache(refs),
prefix, 1);
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
/*
* The packed-refs file might contain broken references, for
* example an old version of a reference that points at an
* object that has since been garbage-collected. This is OK as
* long as there is a corresponding loose reference that
* overrides it, and we don't want to emit an error message in
* this case. So ask the packed_ref_store for all of its
* references, and (if needed) do our own check for broken
* ones in files_ref_iterator_advance(), after we have merged
* the packed and loose references.
*/
packed_iter = refs_ref_iterator_begin(
refs->packed_ref_store, prefix, 0,
DO_FOR_EACH_INCLUDE_BROKEN);
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
overlay_iter = overlay_ref_iterator_begin(loose_iter, packed_iter);
CALLOC_ARRAY(iter, 1);
ref_iterator = &iter->base;
base_ref_iterator_init(ref_iterator, &files_ref_iterator_vtable,
overlay_iter->ordered);
iter->iter0 = overlay_iter;
refs: introduce an iterator interface Currently, the API for iterating over references is via a family of for_each_ref()-type functions that invoke a callback function for each selected reference. All of these eventually call do_for_each_ref(), which knows how to do one thing: iterate in parallel through two ref_caches, one for loose and one for packed refs, giving loose references precedence over packed refs. This is rather complicated code, and is quite specialized to the files backend. It also requires callers to encapsulate their work into a callback function, which often means that they have to define and use a "cb_data" struct to manage their context. The current design is already bursting at the seams, and will become even more awkward in the upcoming world of multiple reference storage backends: * Per-worktree vs. shared references are currently handled via a kludge in git_path() rather than iterating over each part of the reference namespace separately and merging the results. This kludge will cease to work when we have multiple reference storage backends. * The current scheme is inflexible. What if we sometimes want to bypass the ref_cache, or use it only for packed or only for loose refs? What if we want to store symbolic refs in one type of storage backend and non-symbolic ones in another? In the future, each reference backend will need to define its own way of iterating over references. The crux of the problem with the current design is that it is impossible to compose for_each_ref()-style iterations, because the flow of control is owned by the for_each_ref() function. There is nothing that a caller can do but iterate through all references in a single burst, so there is no way for it to interleave references from multiple backends and present the result to the rest of the world as a single compound backend. This commit introduces a new iteration primitive for references: a ref_iterator. A ref_iterator is a polymorphic object that a reference storage backend can be asked to instantiate. There are three functions that can be applied to a ref_iterator: * ref_iterator_advance(): move to the next reference in the iteration * ref_iterator_abort(): end the iteration before it is exhausted * ref_iterator_peel(): peel the reference currently being looked at Iterating using a ref_iterator leaves the flow of control in the hands of the caller, which means that ref_iterators from multiple sources (e.g., loose and packed refs) can be composed and presented to the world as a single compound ref_iterator. It also means that the backend code for implementing reference iteration will sometimes be more complicated. For example, the cache_ref_iterator (which iterates over a ref_cache) can't use the C stack to recurse; instead, it must manage its own stack internally as explicit data structures. There is also a lot of boilerplate connected with object-oriented programming in C. Eventually, end-user callers will be able to be written in a more natural way—managing their own flow of control rather than having to work via callbacks. Since there will only be a few reference backends but there are many consumers of this API, this is a good tradeoff. More importantly, we gain composability, and especially the possibility of writing interchangeable parts that can work with any ref_iterator. For example, merge_ref_iterator implements a generic way of merging the contents of any two ref_iterators. It is used to merge loose + packed refs as part of the implementation of the files_ref_iterator. But it will also be possible to use it to merge other pairs of reference sources (e.g., per-worktree vs. shared refs). Another example is prefix_ref_iterator, which can be used to trim a prefix off the front of reference names before presenting them to the caller (e.g., "refs/heads/master" -> "master"). In this patch, we introduce the iterator abstraction and many utilities, and implement a reference iterator for the files ref storage backend. (I've written several other obvious utilities, for example a generic way to filter references being iterated over. These will probably be useful in the future. But they are not needed for this patch series, so I am not including them at this time.) In a moment we will rewrite do_for_each_ref() to work via reference iterators (allowing some special-purpose code to be discarded), and do something similar for reflogs. In future patch series, we will expose the ref_iterator abstraction in the public refs API so that callers can use it directly. Implementation note: I tried abstracting this a layer further to allow generic iterators (over arbitrary types of objects) and generic utilities like a generic merge_iterator. But the implementation in C was very cumbersome, involving (in my opinion) too much boilerplate and too much unsafe casting, some of which would have had to be done on the caller side. However, I did put a few iterator-related constants in a top-level header file, iterator.h, as they will be useful in a moment to implement iteration over directory trees and possibly other types of iterators in the future. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-06-18 04:15:15 +00:00
iter->flags = flags;
return ref_iterator;
}
/*
* Callback function for raceproof_create_file(). This function is
* expected to do something that makes dirname(path) permanent despite
* the fact that other processes might be cleaning up empty
* directories at the same time. Usually it will create a file named
* path, but alternatively it could create another file in that
* directory, or even chdir() into that directory. The function should
* return 0 if the action was completed successfully. On error, it
* should return a nonzero result and set errno.
* raceproof_create_file() treats two errno values specially:
*
* - ENOENT -- dirname(path) does not exist. In this case,
* raceproof_create_file() tries creating dirname(path)
* (and any parent directories, if necessary) and calls
* the function again.
*
* - EISDIR -- the file already exists and is a directory. In this
* case, raceproof_create_file() removes the directory if
* it is empty (and recursively any empty directories that
* it contains) and calls the function again.
*
* Any other errno causes raceproof_create_file() to fail with the
* callback's return value and errno.
*
* Obviously, this function should be OK with being called again if it
* fails with ENOENT or EISDIR. In other scenarios it will not be
* called again.
*/
typedef int create_file_fn(const char *path, void *cb);
/*
* Create a file in dirname(path) by calling fn, creating leading
* directories if necessary. Retry a few times in case we are racing
* with another process that is trying to clean up the directory that
* contains path. See the documentation for create_file_fn for more
* details.
*
* Return the value and set the errno that resulted from the most
* recent call of fn. fn is always called at least once, and will be
* called more than once if it returns ENOENT or EISDIR.
*/
static int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
{
/*
* The number of times we will try to remove empty directories
* in the way of path. This is only 1 because if another
* process is racily creating directories that conflict with
* us, we don't want to fight against them.
*/
int remove_directories_remaining = 1;
/*
* The number of times that we will try to create the
* directories containing path. We are willing to attempt this
* more than once, because another process could be trying to
* clean up empty directories at the same time as we are
* trying to create them.
*/
int create_directories_remaining = 3;
/* A scratch copy of path, filled lazily if we need it: */
struct strbuf path_copy = STRBUF_INIT;
int ret, save_errno;
/* Sanity check: */
assert(*path);
retry_fn:
ret = fn(path, cb);
save_errno = errno;
if (!ret)
goto out;
if (errno == EISDIR && remove_directories_remaining-- > 0) {
/*
* A directory is in the way. Maybe it is empty; try
* to remove it:
*/
if (!path_copy.len)
strbuf_addstr(&path_copy, path);
if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
goto retry_fn;
} else if (errno == ENOENT && create_directories_remaining-- > 0) {
/*
* Maybe the containing directory didn't exist, or
* maybe it was just deleted by a process that is
* racing with us to clean up empty directories. Try
* to create it:
*/
enum scld_error scld_result;
if (!path_copy.len)
strbuf_addstr(&path_copy, path);
do {
scld_result = safe_create_leading_directories(path_copy.buf);
if (scld_result == SCLD_OK)
goto retry_fn;
} while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
}
out:
strbuf_release(&path_copy);
errno = save_errno;
return ret;
}
static int remove_empty_directories(struct strbuf *path)
{
/*
* we want to create a file but there is a directory there;
* if that is an empty directory (or a directory that contains
* only empty directories), remove them.
*/
return remove_dir_recursively(path, REMOVE_DIR_EMPTY_ONLY);
}
static int create_reflock(const char *path, void *cb)
{
struct lock_file *lk = cb;
return hold_lock_file_for_update_timeout(
lk, path, LOCK_NO_DEREF,
get_files_ref_lock_timeout_ms()) < 0 ? -1 : 0;
}
/*
* Locks a ref returning the lock on success and NULL on failure.
*/
static struct ref_lock *lock_ref_oid_basic(struct files_ref_store *refs,
refs/files: remove "name exist?" check in lock_ref_oid_basic() In lock_ref_oid_basic() we'll happily lock a reference that doesn't exist yet. That's normal, and is how references are initially born, but we don't need to retain checks here in lock_ref_oid_basic() about the state of the ref, when what we're checking is either checked already, or something we're about to discover by trying to lock the ref with raceproof_create_file(). The one exception is the caller in files_reflog_expire(), who passes us a "type" to find out if the reference is a symref or not. We can move the that logic over to that caller, which can now defer its discovery of whether or not the ref is a symref until it's needed. In the preceding commit an exhaustive regression test was added for that case in a new test in "t1417-reflog-updateref.sh". The improved diagnostics here were added in 5b2d8d6f218 (lock_ref_sha1_basic(): improve diagnostics for ref D/F conflicts, 2015-05-11), and then much of the surrounding code went away recently in my 245fbba46d6 (refs/files: remove unused "errno == EISDIR" code, 2021-08-23). The refs_resolve_ref_unsafe() code being removed here looks like it should be tasked with doing that, but it's actually redundant to other code. The reason for that is as noted in 245fbba46d6 this once widely used function now only has a handful of callers left, which all handle this case themselves. To the extent that we're racy between their check and ours removing this check actually improves the situation, as we'll be doing fewer things between the not-under-lock initial check and acquiring the lock. Why this is OK for all the remaining callers of lock_ref_oid_basic() is noted below. There are only two of those callers: * "git branch -[cm] <oldbranch> <newbranch>": In files_copy_or_rename_ref() we'll call this when we copy or rename refs via rename_ref() and copy_ref(). but only after we've checked if the refname exists already via its own call to refs_resolve_ref_unsafe() and refs_rename_ref_available(). As the updated comment to the latter here notes neither of those are actually needed. If we delete not only this code but also refs_rename_ref_available() we'll do just fine, we'll just emit a less friendly error message if e.g. "git branch -m A B/C" would have a D/F conflict with a "B" file. Actually we'd probably die before that in case reflogs for the branch existed, i.e. when the try to rename() or copy_file() the relevant reflog, since if we've got a D/F conflict with a branch name we'll probably also have the same with its reflogs (but not necessarily, we might have reflogs, but it might not). As some #leftoverbits that code seems buggy to me, i.e. the reflog "protocol" should be to get a lock on the main ref, and then perform ref and/or reflog operations. That code dates back to c976d415e53 (git-branch: add options and tests for branch renaming, 2006-11-28) and probably pre-dated the solidifying of that convention. But in any case, that edge case is not our bug or problem right now. * "git reflog expire <ref>": In files_reflog_expire() we'll call this without previous ref existence checking in files-backend.c, but that code is in turn called by code that's just finished checking if the refname whose reflog we're expiring exists. See ae35e16cd43 (reflog expire: don't lock reflogs using previously seen OID, 2021-08-23) for the current state of that code, and 5e6f003ca8a (reflog_expire(): ignore --updateref for symbolic references, 2015-03-03) for the code we'd break if we only did a "update = !!ref" here, which is covered by the aforementioned regression test in "t1417-reflog-updateref.sh". Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-16 09:39:13 +00:00
const char *refname,
struct strbuf *err)
{
struct strbuf ref_file = STRBUF_INIT;
struct ref_lock *lock;
int ignore_errno;
files_assert_main_repository(refs, "lock_ref_oid_basic");
assert(err);
CALLOC_ARRAY(lock, 1);
files_ref_path(refs, &ref_file, refname);
lock_ref_sha1_basic: handle REF_NODEREF with invalid refs We sometimes call lock_ref_sha1_basic with REF_NODEREF to operate directly on a symbolic ref. This is used, for example, to move to a detached HEAD, or when updating the contents of HEAD via checkout or symbolic-ref. However, the first step of the function is to resolve the refname to get the "old" sha1, and we do so without telling resolve_ref_unsafe() that we are only interested in the symref. As a result, we may detect a problem there not with the symref itself, but with something it points to. The real-world example I found (and what is used in the test suite) is a HEAD pointing to a ref that cannot exist, because it would cause a directory/file conflict with other existing refs. This situation is somewhat broken, of course, as trying to _commit_ on that HEAD would fail. But it's not explicitly forbidden, and we should be able to move away from it. However, neither "git checkout" nor "git symbolic-ref" can do so. We try to take the lock on HEAD, which is pointing to a non-existent ref. We bail from resolve_ref_unsafe() with errno set to EISDIR, and the lock code thinks we are attempting to create a d/f conflict. Of course we're not. The problem is that the lock code has no idea what level we were at when we got EISDIR, so trying to diagnose or remove empty directories for HEAD is not useful. To make things even more complicated, we only get EISDIR in the loose-ref case. If the refs are packed, the resolution may "succeed", giving us the pointed-to ref in "refname", but a null oid. Later, we say "ah, the null oid means we are creating; let's make sure there is room for it", but mistakenly check against the _resolved_ refname, not the original. We can fix this by making two tweaks: 1. Call resolve_ref_unsafe() with RESOLVE_REF_NO_RECURSE when REF_NODEREF is set. This means any errors we get will be from the orig_refname, and we can act accordingly. We already do this in the REF_DELETING case, but we should do it for update, too. 2. If we do get a "refname" return from resolve_ref_unsafe(), even with RESOLVE_REF_NO_RECURSE it may be the name of the ref pointed-to by a symref. We already normalize this back to orig_refname before taking the lockfile, but we need to do so before the null_oid check. While we're rearranging the REF_NODEREF handling, we can also bump the initialization of lflags to the top of the function, where we are setting up other flags. This saves us from having yet another conditional block on REF_NODEREF just to set it later. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2016-01-12 21:45:09 +00:00
/*
* If the ref did not exist and we are creating it, make sure
* there is no existing packed ref whose name begins with our
* refname, nor a packed ref whose name is a proper prefix of
* our refname.
*/
if (is_null_oid(&lock->old_oid) &&
files-backend: cheapen refname_available check when locking refs When locking references in preparation for updating them, we need to check that none of the newly added references D/F conflict with existing references (e.g., we don't allow `refs/foo` to be added if `refs/foo/bar` already exists, or vice versa). Prior to 524a9fdb51 (refs_verify_refname_available(): use function in more places, 2017-04-16), conflicts with existing loose references were checked by looking directly in the filesystem, and then conflicts with existing packed references were checked by running `verify_refname_available_dir()` against the packed-refs cache. But that commit changed the final check to call `refs_verify_refname_available()` against the *whole* files ref-store, including both loose and packed references, with the following comment: > This means that those callsites now check for conflicts with all > references rather than just packed refs, but the performance cost > shouldn't be significant (and will be regained later). That comment turned out to be too sanguine. User s@kazlauskas.me reported that fetches involving a very large number of references in neighboring directories were slowed down by that change. The problem is that when fetching, each reference is updated individually, within its own reference transaction. This is done because some reference updates might succeed even though others fail. But every time a reference update transaction is finished, `clear_loose_ref_cache()` is called. So when it is time to update the next reference, part of the loose ref cache has to be repopulated for the `refs_verify_refname_available()` call. If the references are all in neighboring directories, then the cost of repopulating the reference cache increases with the number of references, resulting in O(N²) effort. The comment above also claims that the performance cost "will be regained later". The idea was that once the packed-refs were finished being split out into a separate ref-store, we could limit the `refs_verify_refname_available()` call to the packed references again. That is what we do now. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-08-17 15:12:50 +00:00
refs_verify_refname_available(refs->packed_ref_store, refname,
NULL, NULL, err))
goto error_return;
lock->ref_name = xstrdup(refname);
if (raceproof_create_file(ref_file.buf, create_reflock, &lock->lk)) {
unable_to_lock_message(ref_file.buf, errno, err);
goto error_return;
}
if (!refs_werrres_ref_unsafe(&refs->base, lock->ref_name, 0,
&lock->old_oid, NULL, &ignore_errno))
oidclr(&lock->old_oid);
goto out;
error_return:
unlock_ref(lock);
lock = NULL;
out:
strbuf_release(&ref_file);
return lock;
}
struct ref_to_prune {
struct ref_to_prune *next;
struct object_id oid;
char name[FLEX_ARRAY];
};
enum {
REMOVE_EMPTY_PARENTS_REF = 0x01,
REMOVE_EMPTY_PARENTS_REFLOG = 0x02
};
/*
* Remove empty parent directories associated with the specified
* reference and/or its reflog, but spare [logs/]refs/ and immediate
* subdirs. flags is a combination of REMOVE_EMPTY_PARENTS_REF and/or
* REMOVE_EMPTY_PARENTS_REFLOG.
*/
static void try_remove_empty_parents(struct files_ref_store *refs,
const char *refname,
unsigned int flags)
{
struct strbuf buf = STRBUF_INIT;
struct strbuf sb = STRBUF_INIT;
char *p, *q;
int i;
strbuf_addstr(&buf, refname);
p = buf.buf;
for (i = 0; i < 2; i++) { /* refs/{heads,tags,...}/ */
while (*p && *p != '/')
p++;
/* tolerate duplicate slashes; see check_refname_format() */
while (*p == '/')
p++;
}
q = buf.buf + buf.len;
while (flags & (REMOVE_EMPTY_PARENTS_REF | REMOVE_EMPTY_PARENTS_REFLOG)) {
while (q > p && *q != '/')
q--;
while (q > p && *(q-1) == '/')
q--;
if (q == p)
break;
strbuf_setlen(&buf, q - buf.buf);
strbuf_reset(&sb);
files_ref_path(refs, &sb, buf.buf);
if ((flags & REMOVE_EMPTY_PARENTS_REF) && rmdir(sb.buf))
flags &= ~REMOVE_EMPTY_PARENTS_REF;
strbuf_reset(&sb);
files_reflog_path(refs, &sb, buf.buf);
if ((flags & REMOVE_EMPTY_PARENTS_REFLOG) && rmdir(sb.buf))
flags &= ~REMOVE_EMPTY_PARENTS_REFLOG;
}
strbuf_release(&buf);
strbuf_release(&sb);
}
/* make sure nobody touched the ref, and unlink */
static void prune_ref(struct files_ref_store *refs, struct ref_to_prune *r)
{
struct ref_transaction *transaction;
struct strbuf err = STRBUF_INIT;
int ret = -1;
if (check_refname_format(r->name, 0))
return;
transaction = ref_store_transaction_begin(&refs->base, &err);
if (!transaction)
goto cleanup;
ref_transaction_add_update(
transaction, r->name,
REF_NO_DEREF | REF_HAVE_NEW | REF_HAVE_OLD | REF_IS_PRUNING,
null_oid(), &r->oid, NULL);
if (ref_transaction_commit(transaction, &err))
goto cleanup;
ret = 0;
cleanup:
if (ret)
error("%s", err.buf);
strbuf_release(&err);
ref_transaction_free(transaction);
return;
}
/*
* Prune the loose versions of the references in the linked list
* `*refs_to_prune`, freeing the entries in the list as we go.
*/
static void prune_refs(struct files_ref_store *refs, struct ref_to_prune **refs_to_prune)
{
while (*refs_to_prune) {
struct ref_to_prune *r = *refs_to_prune;
*refs_to_prune = r->next;
prune_ref(refs, r);
free(r);
}
}
/*
* Return true if the specified reference should be packed.
*/
static int should_pack_ref(const char *refname,
const struct object_id *oid, unsigned int ref_flags,
unsigned int pack_flags)
{
/* Do not pack per-worktree refs: */
if (ref_type(refname) != REF_TYPE_NORMAL)
return 0;
/* Do not pack non-tags unless PACK_REFS_ALL is set: */
if (!(pack_flags & PACK_REFS_ALL) && !starts_with(refname, "refs/tags/"))
return 0;
/* Do not pack symbolic refs: */
if (ref_flags & REF_ISSYMREF)
return 0;
/* Do not pack broken refs: */
if (!ref_resolves_to_object(refname, oid, ref_flags))
return 0;
return 1;
}
static int files_pack_refs(struct ref_store *ref_store, unsigned int flags)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_WRITE | REF_STORE_ODB,
"pack_refs");
struct ref_iterator *iter;
int ok;
struct ref_to_prune *refs_to_prune = NULL;
struct strbuf err = STRBUF_INIT;
struct ref_transaction *transaction;
transaction = ref_store_transaction_begin(refs->packed_ref_store, &err);
if (!transaction)
return -1;
packed_refs_lock(refs->packed_ref_store, LOCK_DIE_ON_ERROR, &err);
iter = cache_ref_iterator_begin(get_loose_ref_cache(refs), NULL, 0);
while ((ok = ref_iterator_advance(iter)) == ITER_OK) {
/*
* If the loose reference can be packed, add an entry
* in the packed ref cache. If the reference should be
* pruned, also add it to refs_to_prune.
*/
if (!should_pack_ref(iter->refname, iter->oid, iter->flags,
flags))
continue;
/*
* Add a reference creation for this reference to the
* packed-refs transaction:
*/
if (ref_transaction_update(transaction, iter->refname,
iter->oid, NULL,
REF_NO_DEREF, NULL, &err))
die("failure preparing to create packed reference %s: %s",
iter->refname, err.buf);
/* Schedule the loose reference for pruning if requested. */
if ((flags & PACK_REFS_PRUNE)) {
struct ref_to_prune *n;
FLEX_ALLOC_STR(n, name, iter->refname);
oidcpy(&n->oid, iter->oid);
n->next = refs_to_prune;
refs_to_prune = n;
}
}
if (ok != ITER_DONE)
die("error while iterating over references");
if (ref_transaction_commit(transaction, &err))
die("unable to write new packed-refs: %s", err.buf);
ref_transaction_free(transaction);
packed_refs_unlock(refs->packed_ref_store);
prune_refs(refs, &refs_to_prune);
strbuf_release(&err);
return 0;
}
static int files_delete_refs(struct ref_store *ref_store, const char *msg,
struct string_list *refnames, unsigned int flags)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_WRITE, "delete_refs");
struct strbuf err = STRBUF_INIT;
int i, result = 0;
if (!refnames->nr)
return 0;
if (packed_refs_lock(refs->packed_ref_store, 0, &err))
goto error;
if (refs_delete_refs(refs->packed_ref_store, msg, refnames, flags)) {
packed_refs_unlock(refs->packed_ref_store);
goto error;
}
packed_refs_unlock(refs->packed_ref_store);
for (i = 0; i < refnames->nr; i++) {
const char *refname = refnames->items[i].string;
if (refs_delete_ref(&refs->base, msg, refname, NULL, flags))
result |= error(_("could not remove reference %s"), refname);
}
strbuf_release(&err);
return result;
error:
/*
* If we failed to rewrite the packed-refs file, then it is
* unsafe to try to remove loose refs, because doing so might
* expose an obsolete packed value for a reference that might
* even point at an object that has been garbage collected.
*/
if (refnames->nr == 1)
error(_("could not delete reference %s: %s"),
refnames->items[0].string, err.buf);
else
error(_("could not delete references: %s"), err.buf);
strbuf_release(&err);
return -1;
}
/*
* People using contrib's git-new-workdir have .git/logs/refs ->
* /some/other/path/.git/logs/refs, and that may live on another device.
*
* IOW, to avoid cross device rename errors, the temporary renamed log must
* live into logs/refs.
*/
#define TMP_RENAMED_LOG "refs/.tmp-renamed-log"
struct rename_cb {
const char *tmp_renamed_log;
int true_errno;
};
static int rename_tmp_log_callback(const char *path, void *cb_data)
{
struct rename_cb *cb = cb_data;
if (rename(cb->tmp_renamed_log, path)) {
/*
* rename(a, b) when b is an existing directory ought
* to result in ISDIR, but Solaris 5.8 gives ENOTDIR.
* Sheesh. Record the true errno for error reporting,
* but report EISDIR to raceproof_create_file() so
* that it knows to retry.
*/
cb->true_errno = errno;
if (errno == ENOTDIR)
errno = EISDIR;
return -1;
} else {
return 0;
}
}
static int rename_tmp_log(struct files_ref_store *refs, const char *newrefname)
{
struct strbuf path = STRBUF_INIT;
struct strbuf tmp = STRBUF_INIT;
struct rename_cb cb;
int ret;
files_reflog_path(refs, &path, newrefname);
files_reflog_path(refs, &tmp, TMP_RENAMED_LOG);
cb.tmp_renamed_log = tmp.buf;
ret = raceproof_create_file(path.buf, rename_tmp_log_callback, &cb);
if (ret) {
if (errno == EISDIR)
error("directory not empty: %s", path.buf);
else
error("unable to move logfile %s to %s: %s",
tmp.buf, path.buf,
strerror(cb.true_errno));
}
strbuf_release(&path);
strbuf_release(&tmp);
return ret;
}
static int write_ref_to_lockfile(struct ref_lock *lock,
const struct object_id *oid, struct strbuf *err);
static int commit_ref_update(struct files_ref_store *refs,
struct ref_lock *lock,
const struct object_id *oid, const char *logmsg,
struct strbuf *err);
/*
refs/files: remove "name exist?" check in lock_ref_oid_basic() In lock_ref_oid_basic() we'll happily lock a reference that doesn't exist yet. That's normal, and is how references are initially born, but we don't need to retain checks here in lock_ref_oid_basic() about the state of the ref, when what we're checking is either checked already, or something we're about to discover by trying to lock the ref with raceproof_create_file(). The one exception is the caller in files_reflog_expire(), who passes us a "type" to find out if the reference is a symref or not. We can move the that logic over to that caller, which can now defer its discovery of whether or not the ref is a symref until it's needed. In the preceding commit an exhaustive regression test was added for that case in a new test in "t1417-reflog-updateref.sh". The improved diagnostics here were added in 5b2d8d6f218 (lock_ref_sha1_basic(): improve diagnostics for ref D/F conflicts, 2015-05-11), and then much of the surrounding code went away recently in my 245fbba46d6 (refs/files: remove unused "errno == EISDIR" code, 2021-08-23). The refs_resolve_ref_unsafe() code being removed here looks like it should be tasked with doing that, but it's actually redundant to other code. The reason for that is as noted in 245fbba46d6 this once widely used function now only has a handful of callers left, which all handle this case themselves. To the extent that we're racy between their check and ours removing this check actually improves the situation, as we'll be doing fewer things between the not-under-lock initial check and acquiring the lock. Why this is OK for all the remaining callers of lock_ref_oid_basic() is noted below. There are only two of those callers: * "git branch -[cm] <oldbranch> <newbranch>": In files_copy_or_rename_ref() we'll call this when we copy or rename refs via rename_ref() and copy_ref(). but only after we've checked if the refname exists already via its own call to refs_resolve_ref_unsafe() and refs_rename_ref_available(). As the updated comment to the latter here notes neither of those are actually needed. If we delete not only this code but also refs_rename_ref_available() we'll do just fine, we'll just emit a less friendly error message if e.g. "git branch -m A B/C" would have a D/F conflict with a "B" file. Actually we'd probably die before that in case reflogs for the branch existed, i.e. when the try to rename() or copy_file() the relevant reflog, since if we've got a D/F conflict with a branch name we'll probably also have the same with its reflogs (but not necessarily, we might have reflogs, but it might not). As some #leftoverbits that code seems buggy to me, i.e. the reflog "protocol" should be to get a lock on the main ref, and then perform ref and/or reflog operations. That code dates back to c976d415e53 (git-branch: add options and tests for branch renaming, 2006-11-28) and probably pre-dated the solidifying of that convention. But in any case, that edge case is not our bug or problem right now. * "git reflog expire <ref>": In files_reflog_expire() we'll call this without previous ref existence checking in files-backend.c, but that code is in turn called by code that's just finished checking if the refname whose reflog we're expiring exists. See ae35e16cd43 (reflog expire: don't lock reflogs using previously seen OID, 2021-08-23) for the current state of that code, and 5e6f003ca8a (reflog_expire(): ignore --updateref for symbolic references, 2015-03-03) for the code we'd break if we only did a "update = !!ref" here, which is covered by the aforementioned regression test in "t1417-reflog-updateref.sh". Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-16 09:39:13 +00:00
* Emit a better error message than lockfile.c's
* unable_to_lock_message() would in case there is a D/F conflict with
* another existing reference. If there would be a conflict, emit an error
* message and return false; otherwise, return true.
*
* Note that this function is not safe against all races with other
refs/files: remove "name exist?" check in lock_ref_oid_basic() In lock_ref_oid_basic() we'll happily lock a reference that doesn't exist yet. That's normal, and is how references are initially born, but we don't need to retain checks here in lock_ref_oid_basic() about the state of the ref, when what we're checking is either checked already, or something we're about to discover by trying to lock the ref with raceproof_create_file(). The one exception is the caller in files_reflog_expire(), who passes us a "type" to find out if the reference is a symref or not. We can move the that logic over to that caller, which can now defer its discovery of whether or not the ref is a symref until it's needed. In the preceding commit an exhaustive regression test was added for that case in a new test in "t1417-reflog-updateref.sh". The improved diagnostics here were added in 5b2d8d6f218 (lock_ref_sha1_basic(): improve diagnostics for ref D/F conflicts, 2015-05-11), and then much of the surrounding code went away recently in my 245fbba46d6 (refs/files: remove unused "errno == EISDIR" code, 2021-08-23). The refs_resolve_ref_unsafe() code being removed here looks like it should be tasked with doing that, but it's actually redundant to other code. The reason for that is as noted in 245fbba46d6 this once widely used function now only has a handful of callers left, which all handle this case themselves. To the extent that we're racy between their check and ours removing this check actually improves the situation, as we'll be doing fewer things between the not-under-lock initial check and acquiring the lock. Why this is OK for all the remaining callers of lock_ref_oid_basic() is noted below. There are only two of those callers: * "git branch -[cm] <oldbranch> <newbranch>": In files_copy_or_rename_ref() we'll call this when we copy or rename refs via rename_ref() and copy_ref(). but only after we've checked if the refname exists already via its own call to refs_resolve_ref_unsafe() and refs_rename_ref_available(). As the updated comment to the latter here notes neither of those are actually needed. If we delete not only this code but also refs_rename_ref_available() we'll do just fine, we'll just emit a less friendly error message if e.g. "git branch -m A B/C" would have a D/F conflict with a "B" file. Actually we'd probably die before that in case reflogs for the branch existed, i.e. when the try to rename() or copy_file() the relevant reflog, since if we've got a D/F conflict with a branch name we'll probably also have the same with its reflogs (but not necessarily, we might have reflogs, but it might not). As some #leftoverbits that code seems buggy to me, i.e. the reflog "protocol" should be to get a lock on the main ref, and then perform ref and/or reflog operations. That code dates back to c976d415e53 (git-branch: add options and tests for branch renaming, 2006-11-28) and probably pre-dated the solidifying of that convention. But in any case, that edge case is not our bug or problem right now. * "git reflog expire <ref>": In files_reflog_expire() we'll call this without previous ref existence checking in files-backend.c, but that code is in turn called by code that's just finished checking if the refname whose reflog we're expiring exists. See ae35e16cd43 (reflog expire: don't lock reflogs using previously seen OID, 2021-08-23) for the current state of that code, and 5e6f003ca8a (reflog_expire(): ignore --updateref for symbolic references, 2015-03-03) for the code we'd break if we only did a "update = !!ref" here, which is covered by the aforementioned regression test in "t1417-reflog-updateref.sh". Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-16 09:39:13 +00:00
* processes, and that's not its job. We'll emit a more verbose error on D/f
* conflicts if we get past it into lock_ref_oid_basic().
*/
static int refs_rename_ref_available(struct ref_store *refs,
const char *old_refname,
const char *new_refname)
{
struct string_list skip = STRING_LIST_INIT_NODUP;
struct strbuf err = STRBUF_INIT;
int ok;
string_list_insert(&skip, old_refname);
ok = !refs_verify_refname_available(refs, new_refname,
NULL, &skip, &err);
if (!ok)
error("%s", err.buf);
string_list_clear(&skip, 0);
strbuf_release(&err);
return ok;
}
branch: add a --copy (-c) option to go with --move (-m) Add the ability to --copy a branch and its reflog and configuration, this uses the same underlying machinery as the --move (-m) option except the reflog and configuration is copied instead of being moved. This is useful for e.g. copying a topic branch to a new version, e.g. work to work-2 after submitting the work topic to the list, while preserving all the tracking info and other configuration that goes with the branch, and unlike --move keeping the other already-submitted branch around for reference. Like --move, when the source branch is the currently checked out branch the HEAD is moved to the destination branch. In the case of --move we don't really have a choice (other than remaining on a detached HEAD) and in order to keep the functionality consistent, we are doing it in similar way for --copy too. The most common usage of this feature is expected to be moving to a new topic branch which is a copy of the current one, in that case moving to the target branch is what the user wants, and doesn't unexpectedly behave differently than --move would. One outstanding caveat of this implementation is that: git checkout maint && git checkout master && git branch -c topic && git checkout - Will check out 'maint' instead of 'master'. This is because the @{-N} feature (or its -1 shorthand "-") relies on HEAD reflogs created by the checkout command, so in this case we'll checkout maint instead of master, as the user might expect. What to do about that is left to a future change. Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Sahil Dua <sahildua2305@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-18 21:19:16 +00:00
static int files_copy_or_rename_ref(struct ref_store *ref_store,
const char *oldrefname, const char *newrefname,
branch: add a --copy (-c) option to go with --move (-m) Add the ability to --copy a branch and its reflog and configuration, this uses the same underlying machinery as the --move (-m) option except the reflog and configuration is copied instead of being moved. This is useful for e.g. copying a topic branch to a new version, e.g. work to work-2 after submitting the work topic to the list, while preserving all the tracking info and other configuration that goes with the branch, and unlike --move keeping the other already-submitted branch around for reference. Like --move, when the source branch is the currently checked out branch the HEAD is moved to the destination branch. In the case of --move we don't really have a choice (other than remaining on a detached HEAD) and in order to keep the functionality consistent, we are doing it in similar way for --copy too. The most common usage of this feature is expected to be moving to a new topic branch which is a copy of the current one, in that case moving to the target branch is what the user wants, and doesn't unexpectedly behave differently than --move would. One outstanding caveat of this implementation is that: git checkout maint && git checkout master && git branch -c topic && git checkout - Will check out 'maint' instead of 'master'. This is because the @{-N} feature (or its -1 shorthand "-") relies on HEAD reflogs created by the checkout command, so in this case we'll checkout maint instead of master, as the user might expect. What to do about that is left to a future change. Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Sahil Dua <sahildua2305@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-18 21:19:16 +00:00
const char *logmsg, int copy)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_WRITE, "rename_ref");
struct object_id orig_oid;
int flag = 0, logmoved = 0;
struct ref_lock *lock;
struct stat loginfo;
struct strbuf sb_oldref = STRBUF_INIT;
struct strbuf sb_newref = STRBUF_INIT;
struct strbuf tmp_renamed_log = STRBUF_INIT;
int log, ret;
struct strbuf err = STRBUF_INIT;
int ignore_errno;
files_reflog_path(refs, &sb_oldref, oldrefname);
files_reflog_path(refs, &sb_newref, newrefname);
files_reflog_path(refs, &tmp_renamed_log, TMP_RENAMED_LOG);
log = !lstat(sb_oldref.buf, &loginfo);
if (log && S_ISLNK(loginfo.st_mode)) {
ret = error("reflog for %s is a symlink", oldrefname);
goto out;
}
if (!refs_resolve_ref_unsafe(&refs->base, oldrefname,
RESOLVE_REF_READING | RESOLVE_REF_NO_RECURSE,
&orig_oid, &flag)) {
ret = error("refname %s not found", oldrefname);
goto out;
}
if (flag & REF_ISSYMREF) {
branch: add a --copy (-c) option to go with --move (-m) Add the ability to --copy a branch and its reflog and configuration, this uses the same underlying machinery as the --move (-m) option except the reflog and configuration is copied instead of being moved. This is useful for e.g. copying a topic branch to a new version, e.g. work to work-2 after submitting the work topic to the list, while preserving all the tracking info and other configuration that goes with the branch, and unlike --move keeping the other already-submitted branch around for reference. Like --move, when the source branch is the currently checked out branch the HEAD is moved to the destination branch. In the case of --move we don't really have a choice (other than remaining on a detached HEAD) and in order to keep the functionality consistent, we are doing it in similar way for --copy too. The most common usage of this feature is expected to be moving to a new topic branch which is a copy of the current one, in that case moving to the target branch is what the user wants, and doesn't unexpectedly behave differently than --move would. One outstanding caveat of this implementation is that: git checkout maint && git checkout master && git branch -c topic && git checkout - Will check out 'maint' instead of 'master'. This is because the @{-N} feature (or its -1 shorthand "-") relies on HEAD reflogs created by the checkout command, so in this case we'll checkout maint instead of master, as the user might expect. What to do about that is left to a future change. Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Sahil Dua <sahildua2305@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-18 21:19:16 +00:00
if (copy)
ret = error("refname %s is a symbolic ref, copying it is not supported",
oldrefname);
else
ret = error("refname %s is a symbolic ref, renaming it is not supported",
oldrefname);
goto out;
}
if (!refs_rename_ref_available(&refs->base, oldrefname, newrefname)) {
ret = 1;
goto out;
}
branch: add a --copy (-c) option to go with --move (-m) Add the ability to --copy a branch and its reflog and configuration, this uses the same underlying machinery as the --move (-m) option except the reflog and configuration is copied instead of being moved. This is useful for e.g. copying a topic branch to a new version, e.g. work to work-2 after submitting the work topic to the list, while preserving all the tracking info and other configuration that goes with the branch, and unlike --move keeping the other already-submitted branch around for reference. Like --move, when the source branch is the currently checked out branch the HEAD is moved to the destination branch. In the case of --move we don't really have a choice (other than remaining on a detached HEAD) and in order to keep the functionality consistent, we are doing it in similar way for --copy too. The most common usage of this feature is expected to be moving to a new topic branch which is a copy of the current one, in that case moving to the target branch is what the user wants, and doesn't unexpectedly behave differently than --move would. One outstanding caveat of this implementation is that: git checkout maint && git checkout master && git branch -c topic && git checkout - Will check out 'maint' instead of 'master'. This is because the @{-N} feature (or its -1 shorthand "-") relies on HEAD reflogs created by the checkout command, so in this case we'll checkout maint instead of master, as the user might expect. What to do about that is left to a future change. Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Sahil Dua <sahildua2305@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-18 21:19:16 +00:00
if (!copy && log && rename(sb_oldref.buf, tmp_renamed_log.buf)) {
ret = error("unable to move logfile logs/%s to logs/"TMP_RENAMED_LOG": %s",
oldrefname, strerror(errno));
goto out;
}
branch: add a --copy (-c) option to go with --move (-m) Add the ability to --copy a branch and its reflog and configuration, this uses the same underlying machinery as the --move (-m) option except the reflog and configuration is copied instead of being moved. This is useful for e.g. copying a topic branch to a new version, e.g. work to work-2 after submitting the work topic to the list, while preserving all the tracking info and other configuration that goes with the branch, and unlike --move keeping the other already-submitted branch around for reference. Like --move, when the source branch is the currently checked out branch the HEAD is moved to the destination branch. In the case of --move we don't really have a choice (other than remaining on a detached HEAD) and in order to keep the functionality consistent, we are doing it in similar way for --copy too. The most common usage of this feature is expected to be moving to a new topic branch which is a copy of the current one, in that case moving to the target branch is what the user wants, and doesn't unexpectedly behave differently than --move would. One outstanding caveat of this implementation is that: git checkout maint && git checkout master && git branch -c topic && git checkout - Will check out 'maint' instead of 'master'. This is because the @{-N} feature (or its -1 shorthand "-") relies on HEAD reflogs created by the checkout command, so in this case we'll checkout maint instead of master, as the user might expect. What to do about that is left to a future change. Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Sahil Dua <sahildua2305@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-18 21:19:16 +00:00
if (copy && log && copy_file(tmp_renamed_log.buf, sb_oldref.buf, 0644)) {
ret = error("unable to copy logfile logs/%s to logs/"TMP_RENAMED_LOG": %s",
oldrefname, strerror(errno));
goto out;
}
if (!copy && refs_delete_ref(&refs->base, logmsg, oldrefname,
&orig_oid, REF_NO_DEREF)) {
error("unable to delete old %s", oldrefname);
goto rollback;
}
/*
* Since we are doing a shallow lookup, oid is not the
* correct value to pass to delete_ref as old_oid. But that
* doesn't matter, because an old_oid check wouldn't add to
* the safety anyway; we want to delete the reference whatever
* its current value.
*/
if (!copy && refs_werrres_ref_unsafe(&refs->base, newrefname,
RESOLVE_REF_READING | RESOLVE_REF_NO_RECURSE,
NULL, NULL, &ignore_errno) &&
refs_delete_ref(&refs->base, NULL, newrefname,
NULL, REF_NO_DEREF)) {
if (errno == EISDIR) {
struct strbuf path = STRBUF_INIT;
int result;
files_ref_path(refs, &path, newrefname);
result = remove_empty_directories(&path);
strbuf_release(&path);
if (result) {
error("Directory not empty: %s", newrefname);
goto rollback;
}
} else {
error("unable to delete existing %s", newrefname);
goto rollback;
}
}
if (log && rename_tmp_log(refs, newrefname))
goto rollback;
logmoved = log;
refs/files: remove "name exist?" check in lock_ref_oid_basic() In lock_ref_oid_basic() we'll happily lock a reference that doesn't exist yet. That's normal, and is how references are initially born, but we don't need to retain checks here in lock_ref_oid_basic() about the state of the ref, when what we're checking is either checked already, or something we're about to discover by trying to lock the ref with raceproof_create_file(). The one exception is the caller in files_reflog_expire(), who passes us a "type" to find out if the reference is a symref or not. We can move the that logic over to that caller, which can now defer its discovery of whether or not the ref is a symref until it's needed. In the preceding commit an exhaustive regression test was added for that case in a new test in "t1417-reflog-updateref.sh". The improved diagnostics here were added in 5b2d8d6f218 (lock_ref_sha1_basic(): improve diagnostics for ref D/F conflicts, 2015-05-11), and then much of the surrounding code went away recently in my 245fbba46d6 (refs/files: remove unused "errno == EISDIR" code, 2021-08-23). The refs_resolve_ref_unsafe() code being removed here looks like it should be tasked with doing that, but it's actually redundant to other code. The reason for that is as noted in 245fbba46d6 this once widely used function now only has a handful of callers left, which all handle this case themselves. To the extent that we're racy between their check and ours removing this check actually improves the situation, as we'll be doing fewer things between the not-under-lock initial check and acquiring the lock. Why this is OK for all the remaining callers of lock_ref_oid_basic() is noted below. There are only two of those callers: * "git branch -[cm] <oldbranch> <newbranch>": In files_copy_or_rename_ref() we'll call this when we copy or rename refs via rename_ref() and copy_ref(). but only after we've checked if the refname exists already via its own call to refs_resolve_ref_unsafe() and refs_rename_ref_available(). As the updated comment to the latter here notes neither of those are actually needed. If we delete not only this code but also refs_rename_ref_available() we'll do just fine, we'll just emit a less friendly error message if e.g. "git branch -m A B/C" would have a D/F conflict with a "B" file. Actually we'd probably die before that in case reflogs for the branch existed, i.e. when the try to rename() or copy_file() the relevant reflog, since if we've got a D/F conflict with a branch name we'll probably also have the same with its reflogs (but not necessarily, we might have reflogs, but it might not). As some #leftoverbits that code seems buggy to me, i.e. the reflog "protocol" should be to get a lock on the main ref, and then perform ref and/or reflog operations. That code dates back to c976d415e53 (git-branch: add options and tests for branch renaming, 2006-11-28) and probably pre-dated the solidifying of that convention. But in any case, that edge case is not our bug or problem right now. * "git reflog expire <ref>": In files_reflog_expire() we'll call this without previous ref existence checking in files-backend.c, but that code is in turn called by code that's just finished checking if the refname whose reflog we're expiring exists. See ae35e16cd43 (reflog expire: don't lock reflogs using previously seen OID, 2021-08-23) for the current state of that code, and 5e6f003ca8a (reflog_expire(): ignore --updateref for symbolic references, 2015-03-03) for the code we'd break if we only did a "update = !!ref" here, which is covered by the aforementioned regression test in "t1417-reflog-updateref.sh". Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-16 09:39:13 +00:00
lock = lock_ref_oid_basic(refs, newrefname, &err);
if (!lock) {
branch: add a --copy (-c) option to go with --move (-m) Add the ability to --copy a branch and its reflog and configuration, this uses the same underlying machinery as the --move (-m) option except the reflog and configuration is copied instead of being moved. This is useful for e.g. copying a topic branch to a new version, e.g. work to work-2 after submitting the work topic to the list, while preserving all the tracking info and other configuration that goes with the branch, and unlike --move keeping the other already-submitted branch around for reference. Like --move, when the source branch is the currently checked out branch the HEAD is moved to the destination branch. In the case of --move we don't really have a choice (other than remaining on a detached HEAD) and in order to keep the functionality consistent, we are doing it in similar way for --copy too. The most common usage of this feature is expected to be moving to a new topic branch which is a copy of the current one, in that case moving to the target branch is what the user wants, and doesn't unexpectedly behave differently than --move would. One outstanding caveat of this implementation is that: git checkout maint && git checkout master && git branch -c topic && git checkout - Will check out 'maint' instead of 'master'. This is because the @{-N} feature (or its -1 shorthand "-") relies on HEAD reflogs created by the checkout command, so in this case we'll checkout maint instead of master, as the user might expect. What to do about that is left to a future change. Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Sahil Dua <sahildua2305@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-18 21:19:16 +00:00
if (copy)
error("unable to copy '%s' to '%s': %s", oldrefname, newrefname, err.buf);
else
error("unable to rename '%s' to '%s': %s", oldrefname, newrefname, err.buf);
strbuf_release(&err);
goto rollback;
}
oidcpy(&lock->old_oid, &orig_oid);
if (write_ref_to_lockfile(lock, &orig_oid, &err) ||
commit_ref_update(refs, lock, &orig_oid, logmsg, &err)) {
error("unable to write current sha1 into %s: %s", newrefname, err.buf);
strbuf_release(&err);
goto rollback;
}
ret = 0;
goto out;
rollback:
refs/files: remove "name exist?" check in lock_ref_oid_basic() In lock_ref_oid_basic() we'll happily lock a reference that doesn't exist yet. That's normal, and is how references are initially born, but we don't need to retain checks here in lock_ref_oid_basic() about the state of the ref, when what we're checking is either checked already, or something we're about to discover by trying to lock the ref with raceproof_create_file(). The one exception is the caller in files_reflog_expire(), who passes us a "type" to find out if the reference is a symref or not. We can move the that logic over to that caller, which can now defer its discovery of whether or not the ref is a symref until it's needed. In the preceding commit an exhaustive regression test was added for that case in a new test in "t1417-reflog-updateref.sh". The improved diagnostics here were added in 5b2d8d6f218 (lock_ref_sha1_basic(): improve diagnostics for ref D/F conflicts, 2015-05-11), and then much of the surrounding code went away recently in my 245fbba46d6 (refs/files: remove unused "errno == EISDIR" code, 2021-08-23). The refs_resolve_ref_unsafe() code being removed here looks like it should be tasked with doing that, but it's actually redundant to other code. The reason for that is as noted in 245fbba46d6 this once widely used function now only has a handful of callers left, which all handle this case themselves. To the extent that we're racy between their check and ours removing this check actually improves the situation, as we'll be doing fewer things between the not-under-lock initial check and acquiring the lock. Why this is OK for all the remaining callers of lock_ref_oid_basic() is noted below. There are only two of those callers: * "git branch -[cm] <oldbranch> <newbranch>": In files_copy_or_rename_ref() we'll call this when we copy or rename refs via rename_ref() and copy_ref(). but only after we've checked if the refname exists already via its own call to refs_resolve_ref_unsafe() and refs_rename_ref_available(). As the updated comment to the latter here notes neither of those are actually needed. If we delete not only this code but also refs_rename_ref_available() we'll do just fine, we'll just emit a less friendly error message if e.g. "git branch -m A B/C" would have a D/F conflict with a "B" file. Actually we'd probably die before that in case reflogs for the branch existed, i.e. when the try to rename() or copy_file() the relevant reflog, since if we've got a D/F conflict with a branch name we'll probably also have the same with its reflogs (but not necessarily, we might have reflogs, but it might not). As some #leftoverbits that code seems buggy to me, i.e. the reflog "protocol" should be to get a lock on the main ref, and then perform ref and/or reflog operations. That code dates back to c976d415e53 (git-branch: add options and tests for branch renaming, 2006-11-28) and probably pre-dated the solidifying of that convention. But in any case, that edge case is not our bug or problem right now. * "git reflog expire <ref>": In files_reflog_expire() we'll call this without previous ref existence checking in files-backend.c, but that code is in turn called by code that's just finished checking if the refname whose reflog we're expiring exists. See ae35e16cd43 (reflog expire: don't lock reflogs using previously seen OID, 2021-08-23) for the current state of that code, and 5e6f003ca8a (reflog_expire(): ignore --updateref for symbolic references, 2015-03-03) for the code we'd break if we only did a "update = !!ref" here, which is covered by the aforementioned regression test in "t1417-reflog-updateref.sh". Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-16 09:39:13 +00:00
lock = lock_ref_oid_basic(refs, oldrefname, &err);
if (!lock) {
error("unable to lock %s for rollback: %s", oldrefname, err.buf);
strbuf_release(&err);
goto rollbacklog;
}
flag = log_all_ref_updates;
log_all_ref_updates = LOG_REFS_NONE;
if (write_ref_to_lockfile(lock, &orig_oid, &err) ||
commit_ref_update(refs, lock, &orig_oid, NULL, &err)) {
error("unable to write current sha1 into %s: %s", oldrefname, err.buf);
strbuf_release(&err);
}
log_all_ref_updates = flag;
rollbacklog:
if (logmoved && rename(sb_newref.buf, sb_oldref.buf))
error("unable to restore logfile %s from %s: %s",
oldrefname, newrefname, strerror(errno));
if (!logmoved && log &&
rename(tmp_renamed_log.buf, sb_oldref.buf))
error("unable to restore logfile %s from logs/"TMP_RENAMED_LOG": %s",
oldrefname, strerror(errno));
ret = 1;
out:
strbuf_release(&sb_newref);
strbuf_release(&sb_oldref);
strbuf_release(&tmp_renamed_log);
return ret;
}
branch: add a --copy (-c) option to go with --move (-m) Add the ability to --copy a branch and its reflog and configuration, this uses the same underlying machinery as the --move (-m) option except the reflog and configuration is copied instead of being moved. This is useful for e.g. copying a topic branch to a new version, e.g. work to work-2 after submitting the work topic to the list, while preserving all the tracking info and other configuration that goes with the branch, and unlike --move keeping the other already-submitted branch around for reference. Like --move, when the source branch is the currently checked out branch the HEAD is moved to the destination branch. In the case of --move we don't really have a choice (other than remaining on a detached HEAD) and in order to keep the functionality consistent, we are doing it in similar way for --copy too. The most common usage of this feature is expected to be moving to a new topic branch which is a copy of the current one, in that case moving to the target branch is what the user wants, and doesn't unexpectedly behave differently than --move would. One outstanding caveat of this implementation is that: git checkout maint && git checkout master && git branch -c topic && git checkout - Will check out 'maint' instead of 'master'. This is because the @{-N} feature (or its -1 shorthand "-") relies on HEAD reflogs created by the checkout command, so in this case we'll checkout maint instead of master, as the user might expect. What to do about that is left to a future change. Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Sahil Dua <sahildua2305@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-18 21:19:16 +00:00
static int files_rename_ref(struct ref_store *ref_store,
const char *oldrefname, const char *newrefname,
const char *logmsg)
{
return files_copy_or_rename_ref(ref_store, oldrefname,
newrefname, logmsg, 0);
}
static int files_copy_ref(struct ref_store *ref_store,
const char *oldrefname, const char *newrefname,
const char *logmsg)
{
return files_copy_or_rename_ref(ref_store, oldrefname,
newrefname, logmsg, 1);
}
static int close_ref_gently(struct ref_lock *lock)
{
if (close_lock_file_gently(&lock->lk))
return -1;
return 0;
}
static int commit_ref(struct ref_lock *lock)
{
char *path = get_locked_file_path(&lock->lk);
struct stat st;
if (!lstat(path, &st) && S_ISDIR(st.st_mode)) {
/*
* There is a directory at the path we want to rename
* the lockfile to. Hopefully it is empty; try to
* delete it.
*/
size_t len = strlen(path);
struct strbuf sb_path = STRBUF_INIT;
strbuf_attach(&sb_path, path, len, len);
/*
* If this fails, commit_lock_file() will also fail
* and will report the problem.
*/
remove_empty_directories(&sb_path);
strbuf_release(&sb_path);
} else {
free(path);
}
if (commit_lock_file(&lock->lk))
return -1;
return 0;
}
static int open_or_create_logfile(const char *path, void *cb)
{
int *fd = cb;
*fd = open(path, O_APPEND | O_WRONLY | O_CREAT, 0666);
return (*fd < 0) ? -1 : 0;
}
/*
* Create a reflog for a ref. If force_create = 0, only create the
* reflog for certain refs (those for which should_autocreate_reflog
* returns non-zero). Otherwise, create it regardless of the reference
* name. If the logfile already existed or was created, return 0 and
* set *logfd to the file descriptor opened for appending to the file.
* If no logfile exists and we decided not to create one, return 0 and
* set *logfd to -1. On failure, fill in *err, set *logfd to -1, and
* return -1.
*/
static int log_ref_setup(struct files_ref_store *refs,
const char *refname, int force_create,
int *logfd, struct strbuf *err)
{
struct strbuf logfile_sb = STRBUF_INIT;
char *logfile;
files_reflog_path(refs, &logfile_sb, refname);
logfile = strbuf_detach(&logfile_sb, NULL);
if (force_create || should_autocreate_reflog(refname)) {
if (raceproof_create_file(logfile, open_or_create_logfile, logfd)) {
if (errno == ENOENT)
strbuf_addf(err, "unable to create directory for '%s': "
"%s", logfile, strerror(errno));
else if (errno == EISDIR)
strbuf_addf(err, "there are still logs under '%s'",
logfile);
else
strbuf_addf(err, "unable to append to '%s': %s",
logfile, strerror(errno));
goto error;
}
} else {
*logfd = open(logfile, O_APPEND | O_WRONLY);
if (*logfd < 0) {
if (errno == ENOENT || errno == EISDIR) {
/*
* The logfile doesn't already exist,
* but that is not an error; it only
* means that we won't write log
* entries to it.
*/
;
} else {
strbuf_addf(err, "unable to append to '%s': %s",
logfile, strerror(errno));
goto error;
}
}
}
if (*logfd >= 0)
adjust_shared_perm(logfile);
free(logfile);
return 0;
error:
free(logfile);
return -1;
}
static int files_create_reflog(struct ref_store *ref_store,
const char *refname, int force_create,
struct strbuf *err)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_WRITE, "create_reflog");
int fd;
if (log_ref_setup(refs, refname, force_create, &fd, err))
return -1;
if (fd >= 0)
close(fd);
return 0;
}
static int log_ref_write_fd(int fd, const struct object_id *old_oid,
const struct object_id *new_oid,
const char *committer, const char *msg)
{
struct strbuf sb = STRBUF_INIT;
int ret = 0;
strbuf_addf(&sb, "%s %s %s", oid_to_hex(old_oid), oid_to_hex(new_oid), committer);
if (msg && *msg) {
strbuf_addch(&sb, '\t');
reflog: cleanse messages in the refs.c layer Regarding reflog messages: - We expect that a reflog message consists of a single line. The file format used by the files backend may add a LF after the message as a delimiter, and output by commands like "git log -g" may complete such an incomplete line by adding a LF at the end, but philosophically, the terminating LF is not a part of the message. - We however allow callers of refs API to supply a random sequence of NUL terminated bytes. We cleanse caller-supplied message by squashing a run of whitespaces into a SP, and by trimming trailing whitespace, before storing the message. This is how we tolerate, instead of erring out, a message with LF in it (be it at the end, in the middle, or both). Currently, the cleansing of the reflog message is done by the files backend, before the log is written out. This is sufficient with the current code, as that is the only backend that writes reflogs. But new backends can be added that write reflogs, and we'd want the resulting log message we would read out of "log -g" the same no matter what backend is used, and moving the code to do so to the generic layer is a way to do so. An added benefit is that the "cleansing" function could be updated later, independent from individual backends, to e.g. allow multi-line log messages if we wanted to, and when that happens, it would help a lot to ensure we covered all bases if the cleansing function (which would be updated) is called from the generic layer. Side note: I am not interested in supporting multi-line reflog messages right at the moment (nobody is asking for it), but I envision that instead of the "squash a run of whitespaces into a SP and rtrim" cleansing, we can %urlencode problematic bytes in the message *AND* append a SP at the end, when a new version of Git that supports multi-line and/or verbatim reflog messages writes a reflog record. The reading side can detect the presense of SP at the end (which should have been rtrimmed out if it were written by existing versions of Git) as a signal that decoding %urlencode recovers the original reflog message. Signed-off-by: Han-Wen Nienhuys <hanwen@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-07-10 17:19:53 +00:00
strbuf_addstr(&sb, msg);
}
strbuf_addch(&sb, '\n');
if (write_in_full(fd, sb.buf, sb.len) < 0)
ret = -1;
strbuf_release(&sb);
return ret;
}
static int files_log_ref_write(struct files_ref_store *refs,
const char *refname, const struct object_id *old_oid,
const struct object_id *new_oid, const char *msg,
int flags, struct strbuf *err)
{
int logfd, result;
if (log_all_ref_updates == LOG_REFS_UNSET)
log_all_ref_updates = is_bare_repository() ? LOG_REFS_NONE : LOG_REFS_NORMAL;
result = log_ref_setup(refs, refname,
flags & REF_FORCE_CREATE_REFLOG,
&logfd, err);
if (result)
return result;
if (logfd < 0)
return 0;
result = log_ref_write_fd(logfd, old_oid, new_oid,
git_committer_info(0), msg);
if (result) {
struct strbuf sb = STRBUF_INIT;
int save_errno = errno;
files_reflog_path(refs, &sb, refname);
strbuf_addf(err, "unable to append to '%s': %s",
sb.buf, strerror(save_errno));
strbuf_release(&sb);
close(logfd);
return -1;
}
if (close(logfd)) {
struct strbuf sb = STRBUF_INIT;
int save_errno = errno;
files_reflog_path(refs, &sb, refname);
strbuf_addf(err, "unable to append to '%s': %s",
sb.buf, strerror(save_errno));
strbuf_release(&sb);
return -1;
}
return 0;
}
/*
* Write oid into the open lockfile, then close the lockfile. On
* errors, rollback the lockfile, fill in *err and return -1.
*/
static int write_ref_to_lockfile(struct ref_lock *lock,
const struct object_id *oid, struct strbuf *err)
{
static char term = '\n';
struct object *o;
int fd;
o = parse_object(the_repository, oid);
if (!o) {
strbuf_addf(err,
"trying to write ref '%s' with nonexistent object %s",
lock->ref_name, oid_to_hex(oid));
unlock_ref(lock);
return -1;
}
if (o->type != OBJ_COMMIT && is_branch(lock->ref_name)) {
strbuf_addf(err,
"trying to write non-commit object %s to branch '%s'",
oid_to_hex(oid), lock->ref_name);
unlock_ref(lock);
return -1;
}
fd = get_lock_file_fd(&lock->lk);
if (write_in_full(fd, oid_to_hex(oid), the_hash_algo->hexsz) < 0 ||
avoid "write_in_full(fd, buf, len) != len" pattern The return value of write_in_full() is either "-1", or the requested number of bytes[1]. If we make a partial write before seeing an error, we still return -1, not a partial value. This goes back to f6aa66cb95 (write_in_full: really write in full or return error on disk full., 2007-01-11). So checking anything except "was the return value negative" is pointless. And there are a couple of reasons not to do so: 1. It can do a funny signed/unsigned comparison. If your "len" is signed (e.g., a size_t) then the compiler will promote the "-1" to its unsigned variant. This works out for "!= len" (unless you really were trying to write the maximum size_t bytes), but is a bug if you check "< len" (an example of which was fixed recently in config.c). We should avoid promoting the mental model that you need to check the length at all, so that new sites are not tempted to copy us. 2. Checking for a negative value is shorter to type, especially when the length is an expression. 3. Linus says so. In d34cf19b89 (Clean up write_in_full() users, 2007-01-11), right after the write_in_full() semantics were changed, he wrote: I really wish every "write_in_full()" user would just check against "<0" now, but this fixes the nasty and stupid ones. Appeals to authority aside, this makes it clear that writing it this way does not have an intentional benefit. It's a historical curiosity that we never bothered to clean up (and which was undoubtedly cargo-culted into new sites). So let's convert these obviously-correct cases (this includes write_str_in_full(), which is just a wrapper for write_in_full()). [1] A careful reader may notice there is one way that write_in_full() can return a different value. If we ask write() to write N bytes and get a return value that is _larger_ than N, we could return a larger total. But besides the fact that this would imply a totally broken version of write(), it would already invoke undefined behavior. Our internal remaining counter is an unsigned size_t, which means that subtracting too many byte will wrap it around to a very large number. So we'll instantly begin reading off the end of the buffer, trying to write gigabytes (or petabytes) of data. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Jonathan Nieder <jrnieder@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-13 17:16:03 +00:00
write_in_full(fd, &term, 1) < 0 ||
close_ref_gently(lock) < 0) {
strbuf_addf(err,
"couldn't write '%s'", get_lock_file_path(&lock->lk));
unlock_ref(lock);
return -1;
}
return 0;
}
/*
* Commit a change to a loose reference that has already been written
* to the loose reference lockfile. Also update the reflogs if
* necessary, using the specified lockmsg (which can be NULL).
*/
static int commit_ref_update(struct files_ref_store *refs,
struct ref_lock *lock,
const struct object_id *oid, const char *logmsg,
struct strbuf *err)
{
files_assert_main_repository(refs, "commit_ref_update");
clear_loose_ref_cache(refs);
if (files_log_ref_write(refs, lock->ref_name,
&lock->old_oid, oid,
logmsg, 0, err)) {
char *old_msg = strbuf_detach(err, NULL);
strbuf_addf(err, "cannot update the ref '%s': %s",
lock->ref_name, old_msg);
free(old_msg);
unlock_ref(lock);
return -1;
}
if (strcmp(lock->ref_name, "HEAD") != 0) {
/*
* Special hack: If a branch is updated directly and HEAD
* points to it (may happen on the remote side of a push
* for example) then logically the HEAD reflog should be
* updated too.
* A generic solution implies reverse symref information,
* but finding all symrefs pointing to the given branch
* would be rather costly for this rare event (the direct
* update of a branch) to be worth it. So let's cheat and
* check with HEAD only which should cover 99% of all usage
* scenarios (even 100% of the default ones).
*/
int head_flag;
const char *head_ref;
head_ref = refs_resolve_ref_unsafe(&refs->base, "HEAD",
RESOLVE_REF_READING,
NULL, &head_flag);
if (head_ref && (head_flag & REF_ISSYMREF) &&
!strcmp(head_ref, lock->ref_name)) {
struct strbuf log_err = STRBUF_INIT;
if (files_log_ref_write(refs, "HEAD",
&lock->old_oid, oid,
logmsg, 0, &log_err)) {
error("%s", log_err.buf);
strbuf_release(&log_err);
}
}
}
if (commit_ref(lock)) {
strbuf_addf(err, "couldn't set '%s'", lock->ref_name);
unlock_ref(lock);
return -1;
}
unlock_ref(lock);
return 0;
}
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
static int create_ref_symlink(struct ref_lock *lock, const char *target)
{
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
int ret = -1;
#ifndef NO_SYMLINK_HEAD
char *ref_path = get_locked_file_path(&lock->lk);
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
unlink(ref_path);
ret = symlink(target, ref_path);
free(ref_path);
if (ret)
fprintf(stderr, "no symlink - falling back to symbolic ref\n");
#endif
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
return ret;
}
static void update_symref_reflog(struct files_ref_store *refs,
struct ref_lock *lock, const char *refname,
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
const char *target, const char *logmsg)
{
struct strbuf err = STRBUF_INIT;
struct object_id new_oid;
int ignore_errno;
if (logmsg &&
refs_werrres_ref_unsafe(&refs->base, target,
RESOLVE_REF_READING, &new_oid, NULL,
&ignore_errno) &&
files_log_ref_write(refs, refname, &lock->old_oid,
&new_oid, logmsg, 0, &err)) {
error("%s", err.buf);
strbuf_release(&err);
}
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
}
static int create_symref_locked(struct files_ref_store *refs,
struct ref_lock *lock, const char *refname,
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
const char *target, const char *logmsg)
{
if (prefer_symlink_refs && !create_ref_symlink(lock, target)) {
update_symref_reflog(refs, lock, refname, target, logmsg);
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
return 0;
}
if (!fdopen_lock_file(&lock->lk, "w"))
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
return error("unable to fdopen %s: %s",
get_lock_file_path(&lock->lk), strerror(errno));
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
update_symref_reflog(refs, lock, refname, target, logmsg);
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
/* no error check; commit_ref will check ferror */
fprintf(get_lock_file_fp(&lock->lk), "ref: %s\n", target);
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
if (commit_ref(lock) < 0)
return error("unable to write symref for %s: %s", refname,
strerror(errno));
return 0;
}
static int files_create_symref(struct ref_store *ref_store,
const char *refname, const char *target,
const char *logmsg)
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_WRITE, "create_symref");
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
struct strbuf err = STRBUF_INIT;
struct ref_lock *lock;
int ret;
refs/files: remove "name exist?" check in lock_ref_oid_basic() In lock_ref_oid_basic() we'll happily lock a reference that doesn't exist yet. That's normal, and is how references are initially born, but we don't need to retain checks here in lock_ref_oid_basic() about the state of the ref, when what we're checking is either checked already, or something we're about to discover by trying to lock the ref with raceproof_create_file(). The one exception is the caller in files_reflog_expire(), who passes us a "type" to find out if the reference is a symref or not. We can move the that logic over to that caller, which can now defer its discovery of whether or not the ref is a symref until it's needed. In the preceding commit an exhaustive regression test was added for that case in a new test in "t1417-reflog-updateref.sh". The improved diagnostics here were added in 5b2d8d6f218 (lock_ref_sha1_basic(): improve diagnostics for ref D/F conflicts, 2015-05-11), and then much of the surrounding code went away recently in my 245fbba46d6 (refs/files: remove unused "errno == EISDIR" code, 2021-08-23). The refs_resolve_ref_unsafe() code being removed here looks like it should be tasked with doing that, but it's actually redundant to other code. The reason for that is as noted in 245fbba46d6 this once widely used function now only has a handful of callers left, which all handle this case themselves. To the extent that we're racy between their check and ours removing this check actually improves the situation, as we'll be doing fewer things between the not-under-lock initial check and acquiring the lock. Why this is OK for all the remaining callers of lock_ref_oid_basic() is noted below. There are only two of those callers: * "git branch -[cm] <oldbranch> <newbranch>": In files_copy_or_rename_ref() we'll call this when we copy or rename refs via rename_ref() and copy_ref(). but only after we've checked if the refname exists already via its own call to refs_resolve_ref_unsafe() and refs_rename_ref_available(). As the updated comment to the latter here notes neither of those are actually needed. If we delete not only this code but also refs_rename_ref_available() we'll do just fine, we'll just emit a less friendly error message if e.g. "git branch -m A B/C" would have a D/F conflict with a "B" file. Actually we'd probably die before that in case reflogs for the branch existed, i.e. when the try to rename() or copy_file() the relevant reflog, since if we've got a D/F conflict with a branch name we'll probably also have the same with its reflogs (but not necessarily, we might have reflogs, but it might not). As some #leftoverbits that code seems buggy to me, i.e. the reflog "protocol" should be to get a lock on the main ref, and then perform ref and/or reflog operations. That code dates back to c976d415e53 (git-branch: add options and tests for branch renaming, 2006-11-28) and probably pre-dated the solidifying of that convention. But in any case, that edge case is not our bug or problem right now. * "git reflog expire <ref>": In files_reflog_expire() we'll call this without previous ref existence checking in files-backend.c, but that code is in turn called by code that's just finished checking if the refname whose reflog we're expiring exists. See ae35e16cd43 (reflog expire: don't lock reflogs using previously seen OID, 2021-08-23) for the current state of that code, and 5e6f003ca8a (reflog_expire(): ignore --updateref for symbolic references, 2015-03-03) for the code we'd break if we only did a "update = !!ref" here, which is covered by the aforementioned regression test in "t1417-reflog-updateref.sh". Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-16 09:39:13 +00:00
lock = lock_ref_oid_basic(refs, refname, &err);
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
if (!lock) {
error("%s", err.buf);
strbuf_release(&err);
return -1;
}
ret = create_symref_locked(refs, lock, refname, target, logmsg);
create_symref: use existing ref-lock code The create_symref() function predates the existence of "struct lock_file", let alone the more recent "struct ref_lock". Instead, it just does its own manual dot-locking. Besides being more code, this has a few downsides: - if git is interrupted while holding the lock, we don't clean up the lockfile - we don't do the usual directory/filename conflict check. So you can sometimes create a symref "refs/heads/foo/bar", even if "refs/heads/foo" exists (namely, if the refs are packed and we do not hit the d/f conflict in the filesystem). This patch refactors create_symref() to use the "struct ref_lock" interface, which handles both of these things. There are a few bonus cleanups that come along with it: - we leaked ref_path in some error cases - the symref contents were stored in a fixed-size buffer, putting an artificial (albeit large) limitation on the length of the refname. We now write through fprintf, and handle refnames of any size. - we called adjust_shared_perm only after the file was renamed into place, creating a potential race with readers in a shared repository. The lockfile code now handles this when creating the lockfile, making it atomic. - the legacy prefer_symlink_refs path did not do any locking at all. Admittedly, it is not atomic from a reader's perspective (as it unlinks and re-creates the symlink to overwrite), but at least it cannot conflict with other writers now. - the result of this patch is hopefully more readable. It eliminates three goto labels. Two were for error checking that is now simplified, and the third was to reach shared code that has been pulled into its own function. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-12-29 05:57:01 +00:00
unlock_ref(lock);
return ret;
}
static int files_reflog_exists(struct ref_store *ref_store,
const char *refname)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_READ, "reflog_exists");
struct strbuf sb = STRBUF_INIT;
struct stat st;
int ret;
files_reflog_path(refs, &sb, refname);
ret = !lstat(sb.buf, &st) && S_ISREG(st.st_mode);
strbuf_release(&sb);
return ret;
}
static int files_delete_reflog(struct ref_store *ref_store,
const char *refname)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_WRITE, "delete_reflog");
struct strbuf sb = STRBUF_INIT;
int ret;
files_reflog_path(refs, &sb, refname);
ret = remove_path(sb.buf);
strbuf_release(&sb);
return ret;
}
static int show_one_reflog_ent(struct strbuf *sb, each_reflog_ent_fn fn, void *cb_data)
{
struct object_id ooid, noid;
char *email_end, *message;
timestamp_t timestamp;
int tz;
const char *p = sb->buf;
/* old SP new SP name <email> SP time TAB msg LF */
if (!sb->len || sb->buf[sb->len - 1] != '\n' ||
parse_oid_hex(p, &ooid, &p) || *p++ != ' ' ||
parse_oid_hex(p, &noid, &p) || *p++ != ' ' ||
!(email_end = strchr(p, '>')) ||
email_end[1] != ' ' ||
!(timestamp = parse_timestamp(email_end + 2, &message, 10)) ||
!message || message[0] != ' ' ||
(message[1] != '+' && message[1] != '-') ||
!isdigit(message[2]) || !isdigit(message[3]) ||
!isdigit(message[4]) || !isdigit(message[5]))
return 0; /* corrupt? */
email_end[1] = '\0';
tz = strtol(message + 1, NULL, 10);
if (message[6] != '\t')
message += 6;
else
message += 7;
return fn(&ooid, &noid, p, timestamp, tz, message, cb_data);
}
static char *find_beginning_of_line(char *bob, char *scan)
{
while (bob < scan && *(--scan) != '\n')
; /* keep scanning backwards */
/*
* Return either beginning of the buffer, or LF at the end of
* the previous line.
*/
return scan;
}
static int files_for_each_reflog_ent_reverse(struct ref_store *ref_store,
const char *refname,
each_reflog_ent_fn fn,
void *cb_data)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_READ,
"for_each_reflog_ent_reverse");
struct strbuf sb = STRBUF_INIT;
FILE *logfp;
long pos;
int ret = 0, at_tail = 1;
files_reflog_path(refs, &sb, refname);
logfp = fopen(sb.buf, "r");
strbuf_release(&sb);
if (!logfp)
return -1;
/* Jump to the end */
if (fseek(logfp, 0, SEEK_END) < 0)
ret = error("cannot seek back reflog for %s: %s",
refname, strerror(errno));
pos = ftell(logfp);
while (!ret && 0 < pos) {
int cnt;
size_t nread;
char buf[BUFSIZ];
char *endp, *scanp;
/* Fill next block from the end */
cnt = (sizeof(buf) < pos) ? sizeof(buf) : pos;
if (fseek(logfp, pos - cnt, SEEK_SET)) {
ret = error("cannot seek back reflog for %s: %s",
refname, strerror(errno));
break;
}
nread = fread(buf, cnt, 1, logfp);
if (nread != 1) {
ret = error("cannot read %d bytes from reflog for %s: %s",
cnt, refname, strerror(errno));
break;
}
pos -= cnt;
scanp = endp = buf + cnt;
if (at_tail && scanp[-1] == '\n')
/* Looking at the final LF at the end of the file */
scanp--;
at_tail = 0;
while (buf < scanp) {
/*
* terminating LF of the previous line, or the beginning
* of the buffer.
*/
char *bp;
bp = find_beginning_of_line(buf, scanp);
if (*bp == '\n') {
/*
* The newline is the end of the previous line,
* so we know we have complete line starting
* at (bp + 1). Prefix it onto any prior data
* we collected for the line and process it.
*/
strbuf_splice(&sb, 0, 0, bp + 1, endp - (bp + 1));
scanp = bp;
endp = bp + 1;
ret = show_one_reflog_ent(&sb, fn, cb_data);
strbuf_reset(&sb);
if (ret)
break;
} else if (!pos) {
/*
* We are at the start of the buffer, and the
* start of the file; there is no previous
* line, and we have everything for this one.
* Process it, and we can end the loop.
*/
strbuf_splice(&sb, 0, 0, buf, endp - buf);
ret = show_one_reflog_ent(&sb, fn, cb_data);
strbuf_reset(&sb);
break;
}
if (bp == buf) {
/*
* We are at the start of the buffer, and there
* is more file to read backwards. Which means
* we are in the middle of a line. Note that we
* may get here even if *bp was a newline; that
* just means we are at the exact end of the
* previous line, rather than some spot in the
* middle.
*
* Save away what we have to be combined with
* the data from the next read.
*/
strbuf_splice(&sb, 0, 0, buf, endp - buf);
break;
}
}
}
if (!ret && sb.len)
BUG("reverse reflog parser had leftover data");
fclose(logfp);
strbuf_release(&sb);
return ret;
}
static int files_for_each_reflog_ent(struct ref_store *ref_store,
const char *refname,
each_reflog_ent_fn fn, void *cb_data)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_READ,
"for_each_reflog_ent");
FILE *logfp;
struct strbuf sb = STRBUF_INIT;
int ret = 0;
files_reflog_path(refs, &sb, refname);
logfp = fopen(sb.buf, "r");
strbuf_release(&sb);
if (!logfp)
return -1;
while (!ret && !strbuf_getwholeline(&sb, logfp, '\n'))
ret = show_one_reflog_ent(&sb, fn, cb_data);
fclose(logfp);
strbuf_release(&sb);
return ret;
}
struct files_reflog_iterator {
struct ref_iterator base;
struct ref_store *ref_store;
struct dir_iterator *dir_iterator;
struct object_id oid;
};
static int files_reflog_iterator_advance(struct ref_iterator *ref_iterator)
{
struct files_reflog_iterator *iter =
(struct files_reflog_iterator *)ref_iterator;
struct dir_iterator *diter = iter->dir_iterator;
int ok;
int ignore_errno;
while ((ok = dir_iterator_advance(diter)) == ITER_OK) {
int flags;
if (!S_ISREG(diter->st.st_mode))
continue;
if (diter->basename[0] == '.')
continue;
if (ends_with(diter->basename, ".lock"))
continue;
if (!refs_werrres_ref_unsafe(iter->ref_store,
diter->relative_path, 0,
&iter->oid, &flags,
&ignore_errno)) {
error("bad ref for %s", diter->path.buf);
continue;
}
iter->base.refname = diter->relative_path;
iter->base.oid = &iter->oid;
iter->base.flags = flags;
return ITER_OK;
}
iter->dir_iterator = NULL;
if (ref_iterator_abort(ref_iterator) == ITER_ERROR)
ok = ITER_ERROR;
return ok;
}
static int files_reflog_iterator_peel(struct ref_iterator *ref_iterator,
struct object_id *peeled)
{
BUG("ref_iterator_peel() called for reflog_iterator");
}
static int files_reflog_iterator_abort(struct ref_iterator *ref_iterator)
{
struct files_reflog_iterator *iter =
(struct files_reflog_iterator *)ref_iterator;
int ok = ITER_DONE;
if (iter->dir_iterator)
ok = dir_iterator_abort(iter->dir_iterator);
base_ref_iterator_free(ref_iterator);
return ok;
}
static struct ref_iterator_vtable files_reflog_iterator_vtable = {
files_reflog_iterator_advance,
files_reflog_iterator_peel,
files_reflog_iterator_abort
};
static struct ref_iterator *reflog_iterator_begin(struct ref_store *ref_store,
const char *gitdir)
{
struct dir_iterator *diter;
struct files_reflog_iterator *iter;
struct ref_iterator *ref_iterator;
struct strbuf sb = STRBUF_INIT;
strbuf_addf(&sb, "%s/logs", gitdir);
diter = dir_iterator_begin(sb.buf, 0);
if (!diter) {
strbuf_release(&sb);
return empty_ref_iterator_begin();
}
CALLOC_ARRAY(iter, 1);
ref_iterator = &iter->base;
base_ref_iterator_init(ref_iterator, &files_reflog_iterator_vtable, 0);
iter->dir_iterator = diter;
iter->ref_store = ref_store;
strbuf_release(&sb);
return ref_iterator;
}
static enum iterator_selection reflog_iterator_select(
struct ref_iterator *iter_worktree,
struct ref_iterator *iter_common,
void *cb_data)
{
if (iter_worktree) {
/*
* We're a bit loose here. We probably should ignore
* common refs if they are accidentally added as
* per-worktree refs.
*/
return ITER_SELECT_0;
} else if (iter_common) {
if (ref_type(iter_common->refname) == REF_TYPE_NORMAL)
return ITER_SELECT_1;
/*
* The main ref store may contain main worktree's
* per-worktree refs, which should be ignored
*/
return ITER_SKIP_1;
} else
return ITER_DONE;
}
static struct ref_iterator *files_reflog_iterator_begin(struct ref_store *ref_store)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_READ,
"reflog_iterator_begin");
if (!strcmp(refs->base.gitdir, refs->gitcommondir)) {
return reflog_iterator_begin(ref_store, refs->gitcommondir);
} else {
return merge_ref_iterator_begin(
0, reflog_iterator_begin(ref_store, refs->base.gitdir),
reflog_iterator_begin(ref_store, refs->gitcommondir),
reflog_iterator_select, refs);
}
}
/*
* If update is a direct update of head_ref (the reference pointed to
* by HEAD), then add an extra REF_LOG_ONLY update for HEAD.
*/
static int split_head_update(struct ref_update *update,
struct ref_transaction *transaction,
const char *head_ref,
struct string_list *affected_refnames,
struct strbuf *err)
{
struct string_list_item *item;
struct ref_update *new_update;
if ((update->flags & REF_LOG_ONLY) ||
(update->flags & REF_IS_PRUNING) ||
(update->flags & REF_UPDATE_VIA_HEAD))
return 0;
if (strcmp(update->refname, head_ref))
return 0;
/*
* First make sure that HEAD is not already in the
* transaction. This check is O(lg N) in the transaction
* size, but it happens at most once per transaction.
*/
if (string_list_has_string(affected_refnames, "HEAD")) {
/* An entry already existed */
strbuf_addf(err,
"multiple updates for 'HEAD' (including one "
"via its referent '%s') are not allowed",
update->refname);
return TRANSACTION_NAME_CONFLICT;
}
new_update = ref_transaction_add_update(
transaction, "HEAD",
update->flags | REF_LOG_ONLY | REF_NO_DEREF,
&update->new_oid, &update->old_oid,
update->msg);
/*
* Add "HEAD". This insertion is O(N) in the transaction
* size, but it happens at most once per transaction.
* Add new_update->refname instead of a literal "HEAD".
*/
if (strcmp(new_update->refname, "HEAD"))
BUG("%s unexpectedly not 'HEAD'", new_update->refname);
item = string_list_insert(affected_refnames, new_update->refname);
item->util = new_update;
return 0;
}
/*
* update is for a symref that points at referent and doesn't have
* REF_NO_DEREF set. Split it into two updates:
* - The original update, but with REF_LOG_ONLY and REF_NO_DEREF set
* - A new, separate update for the referent reference
* Note that the new update will itself be subject to splitting when
* the iteration gets to it.
*/
static int split_symref_update(struct ref_update *update,
const char *referent,
struct ref_transaction *transaction,
struct string_list *affected_refnames,
struct strbuf *err)
{
struct string_list_item *item;
struct ref_update *new_update;
unsigned int new_flags;
/*
* First make sure that referent is not already in the
refs/files-backend: add longer-scoped copy of string to list split_symref_update() receives a string-pointer `referent` and adds it to the list of `affected_refnames`. The list simply holds on to the pointers it is given, it does not copy the strings and it does not ever free them. The `referent` string in split_symref_update() belongs to a string buffer in the caller. After we return, the string will be leaked. In the next patch, we want to properly release the string buffer in the caller, but we can't safely do so until we've made sure that `affected_refnames` will not be holding on to a pointer to the string. We could configure the list to handle its own resources, but it would mean some alloc/free-churning. The list is already handling other strings (through other code paths) which we do not need to worry about, and we'd be memory-churning those strings too, completely unnecessary. Observe that split_symref_update() creates a `new_update`-object through ref_transaction_add_update(), after which `new_update->refname` is a copy of `referent`. The difference is, this copy will be freed, and it will be freed *after* `affected_refnames` has been cleared. Rearrange the handling of `referent`, so that we don't add it directly to `affected_refnames`. Instead, first just check whether `referent` exists in the string list, and later add `new_update->refname`. Helped-by: Michael Haggerty <mhagger@alum.mit.edu> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Martin Ågren <martin.agren@gmail.com> Reviewed-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-09 06:57:15 +00:00
* transaction. This check is O(lg N) in the transaction
* size, but it happens at most once per symref in a
* transaction.
*/
refs/files-backend: add longer-scoped copy of string to list split_symref_update() receives a string-pointer `referent` and adds it to the list of `affected_refnames`. The list simply holds on to the pointers it is given, it does not copy the strings and it does not ever free them. The `referent` string in split_symref_update() belongs to a string buffer in the caller. After we return, the string will be leaked. In the next patch, we want to properly release the string buffer in the caller, but we can't safely do so until we've made sure that `affected_refnames` will not be holding on to a pointer to the string. We could configure the list to handle its own resources, but it would mean some alloc/free-churning. The list is already handling other strings (through other code paths) which we do not need to worry about, and we'd be memory-churning those strings too, completely unnecessary. Observe that split_symref_update() creates a `new_update`-object through ref_transaction_add_update(), after which `new_update->refname` is a copy of `referent`. The difference is, this copy will be freed, and it will be freed *after* `affected_refnames` has been cleared. Rearrange the handling of `referent`, so that we don't add it directly to `affected_refnames`. Instead, first just check whether `referent` exists in the string list, and later add `new_update->refname`. Helped-by: Michael Haggerty <mhagger@alum.mit.edu> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Martin Ågren <martin.agren@gmail.com> Reviewed-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-09 06:57:15 +00:00
if (string_list_has_string(affected_refnames, referent)) {
/* An entry already exists */
strbuf_addf(err,
"multiple updates for '%s' (including one "
"via symref '%s') are not allowed",
referent, update->refname);
return TRANSACTION_NAME_CONFLICT;
}
new_flags = update->flags;
if (!strcmp(update->refname, "HEAD")) {
/*
* Record that the new update came via HEAD, so that
* when we process it, split_head_update() doesn't try
* to add another reflog update for HEAD. Note that
* this bit will be propagated if the new_update
* itself needs to be split.
*/
new_flags |= REF_UPDATE_VIA_HEAD;
}
new_update = ref_transaction_add_update(
transaction, referent, new_flags,
&update->new_oid, &update->old_oid,
update->msg);
new_update->parent_update = update;
/*
* Change the symbolic ref update to log only. Also, it
* doesn't need to check its old OID value, as that will be
* done when new_update is processed.
*/
update->flags |= REF_LOG_ONLY | REF_NO_DEREF;
update->flags &= ~REF_HAVE_OLD;
refs/files-backend: add longer-scoped copy of string to list split_symref_update() receives a string-pointer `referent` and adds it to the list of `affected_refnames`. The list simply holds on to the pointers it is given, it does not copy the strings and it does not ever free them. The `referent` string in split_symref_update() belongs to a string buffer in the caller. After we return, the string will be leaked. In the next patch, we want to properly release the string buffer in the caller, but we can't safely do so until we've made sure that `affected_refnames` will not be holding on to a pointer to the string. We could configure the list to handle its own resources, but it would mean some alloc/free-churning. The list is already handling other strings (through other code paths) which we do not need to worry about, and we'd be memory-churning those strings too, completely unnecessary. Observe that split_symref_update() creates a `new_update`-object through ref_transaction_add_update(), after which `new_update->refname` is a copy of `referent`. The difference is, this copy will be freed, and it will be freed *after* `affected_refnames` has been cleared. Rearrange the handling of `referent`, so that we don't add it directly to `affected_refnames`. Instead, first just check whether `referent` exists in the string list, and later add `new_update->refname`. Helped-by: Michael Haggerty <mhagger@alum.mit.edu> Reviewed-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Martin Ågren <martin.agren@gmail.com> Reviewed-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-09 06:57:15 +00:00
/*
* Add the referent. This insertion is O(N) in the transaction
* size, but it happens at most once per symref in a
* transaction. Make sure to add new_update->refname, which will
* be valid as long as affected_refnames is in use, and NOT
* referent, which might soon be freed by our caller.
*/
item = string_list_insert(affected_refnames, new_update->refname);
if (item->util)
BUG("%s unexpectedly found in affected_refnames",
new_update->refname);
item->util = new_update;
return 0;
}
/*
* Return the refname under which update was originally requested.
*/
static const char *original_update_refname(struct ref_update *update)
{
while (update->parent_update)
update = update->parent_update;
return update->refname;
}
/*
* Check whether the REF_HAVE_OLD and old_oid values stored in update
* are consistent with oid, which is the reference's current value. If
* everything is OK, return 0; otherwise, write an error message to
* err and return -1.
*/
static int check_old_oid(struct ref_update *update, struct object_id *oid,
struct strbuf *err)
{
if (!(update->flags & REF_HAVE_OLD) ||
oideq(oid, &update->old_oid))
return 0;
if (is_null_oid(&update->old_oid))
strbuf_addf(err, "cannot lock ref '%s': "
"reference already exists",
original_update_refname(update));
else if (is_null_oid(oid))
strbuf_addf(err, "cannot lock ref '%s': "
"reference is missing but expected %s",
original_update_refname(update),
oid_to_hex(&update->old_oid));
else
strbuf_addf(err, "cannot lock ref '%s': "
"is at %s but expected %s",
original_update_refname(update),
oid_to_hex(oid),
oid_to_hex(&update->old_oid));
return -1;
}
/*
* Prepare for carrying out update:
* - Lock the reference referred to by update.
* - Read the reference under lock.
* - Check that its old OID value (if specified) is correct, and in
* any case record it in update->lock->old_oid for later use when
* writing the reflog.
* - If it is a symref update without REF_NO_DEREF, split it up into a
* REF_LOG_ONLY update of the symref and add a separate update for
* the referent to transaction.
* - If it is an update of head_ref, add a corresponding REF_LOG_ONLY
* update of HEAD.
*/
static int lock_ref_for_update(struct files_ref_store *refs,
struct ref_update *update,
struct ref_transaction *transaction,
const char *head_ref,
struct string_list *affected_refnames,
struct strbuf *err)
{
struct strbuf referent = STRBUF_INIT;
int mustexist = (update->flags & REF_HAVE_OLD) &&
!is_null_oid(&update->old_oid);
int ret = 0;
struct ref_lock *lock;
files_assert_main_repository(refs, "lock_ref_for_update");
if ((update->flags & REF_HAVE_NEW) && is_null_oid(&update->new_oid))
update->flags |= REF_DELETING;
if (head_ref) {
ret = split_head_update(update, transaction, head_ref,
affected_refnames, err);
if (ret)
goto out;
}
ret = lock_raw_ref(refs, update->refname, mustexist,
affected_refnames,
&lock, &referent,
&update->type, err);
if (ret) {
char *reason;
reason = strbuf_detach(err, NULL);
strbuf_addf(err, "cannot lock ref '%s': %s",
original_update_refname(update), reason);
free(reason);
goto out;
}
update->backend_data = lock;
if (update->type & REF_ISSYMREF) {
if (update->flags & REF_NO_DEREF) {
/*
* We won't be reading the referent as part of
* the transaction, so we have to read it here
* to record and possibly check old_oid:
*/
int ignore_errno;
if (!refs_werrres_ref_unsafe(&refs->base,
referent.buf, 0,
&lock->old_oid, NULL,
&ignore_errno)) {
if (update->flags & REF_HAVE_OLD) {
strbuf_addf(err, "cannot lock ref '%s': "
"error reading reference",
original_update_refname(update));
ret = TRANSACTION_GENERIC_ERROR;
goto out;
}
} else if (check_old_oid(update, &lock->old_oid, err)) {
ret = TRANSACTION_GENERIC_ERROR;
goto out;
}
} else {
/*
* Create a new update for the reference this
* symref is pointing at. Also, we will record
* and verify old_oid for this update as part
* of processing the split-off update, so we
* don't have to do it here.
*/
ret = split_symref_update(update,
referent.buf, transaction,
affected_refnames, err);
if (ret)
goto out;
}
} else {
struct ref_update *parent_update;
if (check_old_oid(update, &lock->old_oid, err)) {
ret = TRANSACTION_GENERIC_ERROR;
goto out;
}
/*
* If this update is happening indirectly because of a
* symref update, record the old OID in the parent
* update:
*/
for (parent_update = update->parent_update;
parent_update;
parent_update = parent_update->parent_update) {
struct ref_lock *parent_lock = parent_update->backend_data;
oidcpy(&parent_lock->old_oid, &lock->old_oid);
}
}
if ((update->flags & REF_HAVE_NEW) &&
!(update->flags & REF_DELETING) &&
!(update->flags & REF_LOG_ONLY)) {
if (!(update->type & REF_ISSYMREF) &&
oideq(&lock->old_oid, &update->new_oid)) {
/*
* The reference already has the desired
* value, so we don't need to write it.
*/
} else if (write_ref_to_lockfile(lock, &update->new_oid,
err)) {
char *write_err = strbuf_detach(err, NULL);
/*
* The lock was freed upon failure of
* write_ref_to_lockfile():
*/
update->backend_data = NULL;
strbuf_addf(err,
"cannot update ref '%s': %s",
update->refname, write_err);
free(write_err);
ret = TRANSACTION_GENERIC_ERROR;
goto out;
} else {
update->flags |= REF_NEEDS_COMMIT;
}
}
if (!(update->flags & REF_NEEDS_COMMIT)) {
/*
* We didn't call write_ref_to_lockfile(), so
* the lockfile is still open. Close it to
* free up the file descriptor:
*/
if (close_ref_gently(lock)) {
strbuf_addf(err, "couldn't close '%s.lock'",
update->refname);
ret = TRANSACTION_GENERIC_ERROR;
goto out;
}
}
out:
strbuf_release(&referent);
return ret;
}
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
struct files_transaction_backend_data {
struct ref_transaction *packed_transaction;
int packed_refs_locked;
};
/*
* Unlock any references in `transaction` that are still locked, and
* mark the transaction closed.
*/
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
static void files_transaction_cleanup(struct files_ref_store *refs,
struct ref_transaction *transaction)
{
size_t i;
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
struct files_transaction_backend_data *backend_data =
transaction->backend_data;
struct strbuf err = STRBUF_INIT;
for (i = 0; i < transaction->nr; i++) {
struct ref_update *update = transaction->updates[i];
struct ref_lock *lock = update->backend_data;
if (lock) {
unlock_ref(lock);
update->backend_data = NULL;
}
}
if (backend_data) {
if (backend_data->packed_transaction &&
ref_transaction_abort(backend_data->packed_transaction, &err)) {
error("error aborting transaction: %s", err.buf);
strbuf_release(&err);
}
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
if (backend_data->packed_refs_locked)
packed_refs_unlock(refs->packed_ref_store);
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
free(backend_data);
}
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
transaction->state = REF_TRANSACTION_CLOSED;
}
static int files_transaction_prepare(struct ref_store *ref_store,
struct ref_transaction *transaction,
struct strbuf *err)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_WRITE,
"ref_transaction_prepare");
size_t i;
int ret = 0;
struct string_list affected_refnames = STRING_LIST_INIT_NODUP;
char *head_ref = NULL;
int head_type;
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
struct files_transaction_backend_data *backend_data;
struct ref_transaction *packed_transaction = NULL;
assert(err);
if (!transaction->nr)
goto cleanup;
CALLOC_ARRAY(backend_data, 1);
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
transaction->backend_data = backend_data;
/*
* Fail if a refname appears more than once in the
* transaction. (If we end up splitting up any updates using
* split_symref_update() or split_head_update(), those
* functions will check that the new updates don't have the
* same refname as any existing ones.) Also fail if any of the
* updates use REF_IS_PRUNING without REF_NO_DEREF.
*/
for (i = 0; i < transaction->nr; i++) {
struct ref_update *update = transaction->updates[i];
struct string_list_item *item =
string_list_append(&affected_refnames, update->refname);
if ((update->flags & REF_IS_PRUNING) &&
!(update->flags & REF_NO_DEREF))
BUG("REF_IS_PRUNING set without REF_NO_DEREF");
/*
* We store a pointer to update in item->util, but at
* the moment we never use the value of this field
* except to check whether it is non-NULL.
*/
item->util = update;
}
string_list_sort(&affected_refnames);
if (ref_update_reject_duplicates(&affected_refnames, err)) {
ret = TRANSACTION_GENERIC_ERROR;
goto cleanup;
}
/*
* Special hack: If a branch is updated directly and HEAD
* points to it (may happen on the remote side of a push
* for example) then logically the HEAD reflog should be
* updated too.
*
* A generic solution would require reverse symref lookups,
* but finding all symrefs pointing to a given branch would be
* rather costly for this rare event (the direct update of a
* branch) to be worth it. So let's cheat and check with HEAD
* only, which should cover 99% of all usage scenarios (even
* 100% of the default ones).
*
* So if HEAD is a symbolic reference, then record the name of
* the reference that it points to. If we see an update of
* head_ref within the transaction, then split_head_update()
* arranges for the reflog of HEAD to be updated, too.
*/
head_ref = refs_resolve_refdup(ref_store, "HEAD",
RESOLVE_REF_NO_RECURSE,
NULL, &head_type);
if (head_ref && !(head_type & REF_ISSYMREF)) {
FREE_AND_NULL(head_ref);
}
/*
* Acquire all locks, verify old values if provided, check
* that new values are valid, and write new values to the
* lockfiles, ready to be activated. Only keep one lockfile
* open at a time to avoid running out of file descriptors.
* Note that lock_ref_for_update() might append more updates
* to the transaction.
*/
for (i = 0; i < transaction->nr; i++) {
struct ref_update *update = transaction->updates[i];
ret = lock_ref_for_update(refs, update, transaction,
head_ref, &affected_refnames, err);
if (ret)
files_transaction_prepare(): fix handling of ref lock failure Since dc39e09942 (files_ref_store: use a transaction to update packed refs, 2017-09-08), failure to lock a reference has been handled incorrectly by `files_transaction_prepare()`. If `lock_ref_for_update()` fails in the lock-acquisition loop of that function, it sets `ret` then breaks out of that loop. Prior to dc39e09942, that was OK, because the only thing following the loop was the cleanup code. But dc39e09942 added another blurb of code between the loop and the cleanup. That blurb sometimes resets `ret` to zero, making the cleanup code think that the locking was successful. Specifically, whenever * One or more reference deletions have been processed successfully in the lock-acquisition loop. (Processing the first such reference causes a packed-ref transaction to be initialized.) * Then `lock_ref_for_update()` fails for a subsequent reference. Such a failure can happen for a number of reasons, such as the old SHA-1 not being correct, lock contention, etc. This causes a `break` out of the lock-acquisition loop. * The `packed-refs` lock is acquired successfully and `ref_transaction_prepare()` succeeds for the packed-ref transaction. This has the effect of resetting `ret` back to 0, and making the cleanup code think that lock acquisition was successful. In that case, any reference updates that were processed prior to breaking out of the loop would be carried out (loose and packed), but the reference that couldn't be locked and any subsequent references would silently be ignored. This can easily cause data loss if, for example, the user was trying to push a new name for an existing branch while deleting the old name. After the push, the branch could be left unreachable, and could even subsequently be garbage-collected. This problem was noticed in the context of deleting one reference and creating another in a single transaction, when the two references D/F conflict with each other, like git update-ref --stdin <<EOF delete refs/foo create refs/foo/bar HEAD EOF This triggers the above bug because the deletion is processed successfully for `refs/foo`, then the D/F conflict causes `lock_ref_for_update()` to fail when `refs/foo/bar` is processed. In this case the transaction *should* fail, but instead it causes `refs/foo` to be deleted without creating `refs/foo`. This could easily result in data loss. The fix is simple: instead of just breaking out of the loop, jump directly to the cleanup code. This fixes some tests in t1404 that were added in the previous commit. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-10-24 15:16:25 +00:00
goto cleanup;
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
if (update->flags & REF_DELETING &&
!(update->flags & REF_LOG_ONLY) &&
!(update->flags & REF_IS_PRUNING)) {
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
/*
* This reference has to be deleted from
* packed-refs if it exists there.
*/
if (!packed_transaction) {
packed_transaction = ref_store_transaction_begin(
refs->packed_ref_store, err);
if (!packed_transaction) {
ret = TRANSACTION_GENERIC_ERROR;
goto cleanup;
}
backend_data->packed_transaction =
packed_transaction;
}
ref_transaction_add_update(
packed_transaction, update->refname,
REF_HAVE_NEW | REF_NO_DEREF,
&update->new_oid, NULL,
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
NULL);
}
}
if (packed_transaction) {
if (packed_refs_lock(refs->packed_ref_store, 0, err)) {
ret = TRANSACTION_GENERIC_ERROR;
goto cleanup;
}
backend_data->packed_refs_locked = 1;
files-backend: don't rewrite the `packed-refs` file unnecessarily Even when we are deleting references, we needn't overwrite the `packed-refs` file if the references that we are deleting only exist as loose references. Implement this optimization as follows: * Add a function `is_packed_transaction_needed()`, which checks whether a given packed-refs transaction actually needs to be carried out (i.e., it returns false if the transaction obviously wouldn't have any effect). This function must be called while holding the `packed-refs` lock to avoid races. * Change `files_transaction_prepare()` to check whether the packed-refs transaction is actually needed. If not, squelch it, but continue holding the `packed-refs` lock until the end of the transaction to avoid races. This fixes a mild regression caused by dc39e09942 (files_ref_store: use a transaction to update packed refs, 2017-09-08). Before that commit, unnecessary rewrites of `packed-refs` were suppressed by `repack_without_refs()`. But the transaction-based writing introduced by that commit didn't perform that optimization. Note that the pre-dc39e09942 code still had to *read* the whole `packed-refs` file to determine that the rewrite could be skipped, so the performance for the cases that the write could be elided was `O(N)` in the number of packed references both before and after dc39e09942. But after that commit the constant factor increased. This commit reimplements the optimization of eliding unnecessary `packed-refs` rewrites. That, plus the fact that since cfa2e29c34 (packed_ref_store: get rid of the `ref_cache` entirely, 2017-03-17) we don't necessarily have to read the whole `packed-refs` file at all, means that deletes of one or a few loose references can now be done with `O(n lg N)` effort, where `n` is the number of loose references being deleted and `N` is the total number of packed references. This commit fixes two tests in t1409. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-10-28 09:16:02 +00:00
if (is_packed_transaction_needed(refs->packed_ref_store,
packed_transaction)) {
ret = ref_transaction_prepare(packed_transaction, err);
refs/files-backend: handle packed transaction prepare failure In files_transaction_prepare(), if we have to delete some refs, we use a subordinate packed_transaction to do so. It's rare for that sub-transaction's prepare step to fail, since we hold the packed-refs lock. But if it does, we trigger a BUG() due to these steps: - we've attached the packed transaction to the files transaction as backend_data->packed_transaction - when the prepare step fails, the packed transaction cleans itself up, putting itself into the CLOSED state - the error value from preparing the packed transaction lets us know in files_transaction_prepare() that we should also clean up and return an error. We call files_transaction_cleanup(), which tries to abort backend_data->packed_transaction. Since it's already CLOSED, that triggers an assertion in ref_transaction_abort(). We can fix that by disconnecting the packed transaction from the outer files transaction, and then free-ing (not aborting!) it ourselves. A few other options/alternatives I considered: - we could just make it a noop to abort a CLOSED transaction. But that seems less safe, since clearly this code expects (and enforces) a particular set of state transitions. - we could have files_transaction_cleanup() selectively call abort() vs free() based on the state of the on the packed transaction. That's basically a more restricted version of the above, but also potentially unsafe. - instead of disconnecting backend_data->packed_transaction on error, we could wait to install it until we successfully prepare. That might make the flow a little simpler, but it introduces a hassle. Earlier parts of files_transaction_prepare() that encounter an error will jump to the cleanup label, and expect that cleaning up the outer transaction will clean up the packed transaction, too. We'd have to adjust those sites to clean up the packed transaction. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-03-21 09:28:44 +00:00
/*
* A failure during the prepare step will abort
* itself, but not free. Do that now, and disconnect
* from the files_transaction so it does not try to
* abort us when we hit the cleanup code below.
*/
if (ret) {
ref_transaction_free(packed_transaction);
backend_data->packed_transaction = NULL;
}
files-backend: don't rewrite the `packed-refs` file unnecessarily Even when we are deleting references, we needn't overwrite the `packed-refs` file if the references that we are deleting only exist as loose references. Implement this optimization as follows: * Add a function `is_packed_transaction_needed()`, which checks whether a given packed-refs transaction actually needs to be carried out (i.e., it returns false if the transaction obviously wouldn't have any effect). This function must be called while holding the `packed-refs` lock to avoid races. * Change `files_transaction_prepare()` to check whether the packed-refs transaction is actually needed. If not, squelch it, but continue holding the `packed-refs` lock until the end of the transaction to avoid races. This fixes a mild regression caused by dc39e09942 (files_ref_store: use a transaction to update packed refs, 2017-09-08). Before that commit, unnecessary rewrites of `packed-refs` were suppressed by `repack_without_refs()`. But the transaction-based writing introduced by that commit didn't perform that optimization. Note that the pre-dc39e09942 code still had to *read* the whole `packed-refs` file to determine that the rewrite could be skipped, so the performance for the cases that the write could be elided was `O(N)` in the number of packed references both before and after dc39e09942. But after that commit the constant factor increased. This commit reimplements the optimization of eliding unnecessary `packed-refs` rewrites. That, plus the fact that since cfa2e29c34 (packed_ref_store: get rid of the `ref_cache` entirely, 2017-03-17) we don't necessarily have to read the whole `packed-refs` file at all, means that deletes of one or a few loose references can now be done with `O(n lg N)` effort, where `n` is the number of loose references being deleted and `N` is the total number of packed references. This commit fixes two tests in t1409. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-10-28 09:16:02 +00:00
} else {
/*
* We can skip rewriting the `packed-refs`
* file. But we do need to leave it locked, so
* that somebody else doesn't pack a reference
* that we are trying to delete.
*
* We need to disconnect our transaction from
* backend_data, since the abort (whether successful or
* not) will free it.
files-backend: don't rewrite the `packed-refs` file unnecessarily Even when we are deleting references, we needn't overwrite the `packed-refs` file if the references that we are deleting only exist as loose references. Implement this optimization as follows: * Add a function `is_packed_transaction_needed()`, which checks whether a given packed-refs transaction actually needs to be carried out (i.e., it returns false if the transaction obviously wouldn't have any effect). This function must be called while holding the `packed-refs` lock to avoid races. * Change `files_transaction_prepare()` to check whether the packed-refs transaction is actually needed. If not, squelch it, but continue holding the `packed-refs` lock until the end of the transaction to avoid races. This fixes a mild regression caused by dc39e09942 (files_ref_store: use a transaction to update packed refs, 2017-09-08). Before that commit, unnecessary rewrites of `packed-refs` were suppressed by `repack_without_refs()`. But the transaction-based writing introduced by that commit didn't perform that optimization. Note that the pre-dc39e09942 code still had to *read* the whole `packed-refs` file to determine that the rewrite could be skipped, so the performance for the cases that the write could be elided was `O(N)` in the number of packed references both before and after dc39e09942. But after that commit the constant factor increased. This commit reimplements the optimization of eliding unnecessary `packed-refs` rewrites. That, plus the fact that since cfa2e29c34 (packed_ref_store: get rid of the `ref_cache` entirely, 2017-03-17) we don't necessarily have to read the whole `packed-refs` file at all, means that deletes of one or a few loose references can now be done with `O(n lg N)` effort, where `n` is the number of loose references being deleted and `N` is the total number of packed references. This commit fixes two tests in t1409. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-10-28 09:16:02 +00:00
*/
backend_data->packed_transaction = NULL;
files-backend: don't rewrite the `packed-refs` file unnecessarily Even when we are deleting references, we needn't overwrite the `packed-refs` file if the references that we are deleting only exist as loose references. Implement this optimization as follows: * Add a function `is_packed_transaction_needed()`, which checks whether a given packed-refs transaction actually needs to be carried out (i.e., it returns false if the transaction obviously wouldn't have any effect). This function must be called while holding the `packed-refs` lock to avoid races. * Change `files_transaction_prepare()` to check whether the packed-refs transaction is actually needed. If not, squelch it, but continue holding the `packed-refs` lock until the end of the transaction to avoid races. This fixes a mild regression caused by dc39e09942 (files_ref_store: use a transaction to update packed refs, 2017-09-08). Before that commit, unnecessary rewrites of `packed-refs` were suppressed by `repack_without_refs()`. But the transaction-based writing introduced by that commit didn't perform that optimization. Note that the pre-dc39e09942 code still had to *read* the whole `packed-refs` file to determine that the rewrite could be skipped, so the performance for the cases that the write could be elided was `O(N)` in the number of packed references both before and after dc39e09942. But after that commit the constant factor increased. This commit reimplements the optimization of eliding unnecessary `packed-refs` rewrites. That, plus the fact that since cfa2e29c34 (packed_ref_store: get rid of the `ref_cache` entirely, 2017-03-17) we don't necessarily have to read the whole `packed-refs` file at all, means that deletes of one or a few loose references can now be done with `O(n lg N)` effort, where `n` is the number of loose references being deleted and `N` is the total number of packed references. This commit fixes two tests in t1409. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-10-28 09:16:02 +00:00
if (ref_transaction_abort(packed_transaction, err)) {
ret = TRANSACTION_GENERIC_ERROR;
goto cleanup;
}
}
}
cleanup:
free(head_ref);
string_list_clear(&affected_refnames, 0);
if (ret)
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
files_transaction_cleanup(refs, transaction);
else
transaction->state = REF_TRANSACTION_PREPARED;
return ret;
}
static int files_transaction_finish(struct ref_store *ref_store,
struct ref_transaction *transaction,
struct strbuf *err)
{
struct files_ref_store *refs =
files_downcast(ref_store, 0, "ref_transaction_finish");
size_t i;
int ret = 0;
struct strbuf sb = STRBUF_INIT;
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
struct files_transaction_backend_data *backend_data;
struct ref_transaction *packed_transaction;
assert(err);
if (!transaction->nr) {
transaction->state = REF_TRANSACTION_CLOSED;
return 0;
}
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
backend_data = transaction->backend_data;
packed_transaction = backend_data->packed_transaction;
/* Perform updates first so live commits remain referenced */
for (i = 0; i < transaction->nr; i++) {
struct ref_update *update = transaction->updates[i];
struct ref_lock *lock = update->backend_data;
if (update->flags & REF_NEEDS_COMMIT ||
update->flags & REF_LOG_ONLY) {
if (files_log_ref_write(refs,
lock->ref_name,
&lock->old_oid,
&update->new_oid,
update->msg, update->flags,
err)) {
char *old_msg = strbuf_detach(err, NULL);
strbuf_addf(err, "cannot update the ref '%s': %s",
lock->ref_name, old_msg);
free(old_msg);
unlock_ref(lock);
update->backend_data = NULL;
ret = TRANSACTION_GENERIC_ERROR;
goto cleanup;
}
}
if (update->flags & REF_NEEDS_COMMIT) {
clear_loose_ref_cache(refs);
if (commit_ref(lock)) {
strbuf_addf(err, "couldn't set '%s'", lock->ref_name);
unlock_ref(lock);
update->backend_data = NULL;
ret = TRANSACTION_GENERIC_ERROR;
goto cleanup;
}
}
}
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
/*
* Now that updates are safely completed, we can perform
* deletes. First delete the reflogs of any references that
* will be deleted, since (in the unexpected event of an
* error) leaving a reference without a reflog is less bad
* than leaving a reflog without a reference (the latter is a
* mildly invalid repository state):
*/
for (i = 0; i < transaction->nr; i++) {
struct ref_update *update = transaction->updates[i];
if (update->flags & REF_DELETING &&
!(update->flags & REF_LOG_ONLY) &&
!(update->flags & REF_IS_PRUNING)) {
strbuf_reset(&sb);
files_reflog_path(refs, &sb, update->refname);
if (!unlink_or_warn(sb.buf))
try_remove_empty_parents(refs, update->refname,
REMOVE_EMPTY_PARENTS_REFLOG);
}
}
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
/*
* Perform deletes now that updates are safely completed.
*
* First delete any packed versions of the references, while
* retaining the packed-refs lock:
*/
if (packed_transaction) {
ret = ref_transaction_commit(packed_transaction, err);
ref_transaction_free(packed_transaction);
packed_transaction = NULL;
backend_data->packed_transaction = NULL;
if (ret)
goto cleanup;
}
/* Now delete the loose versions of the references: */
for (i = 0; i < transaction->nr; i++) {
struct ref_update *update = transaction->updates[i];
struct ref_lock *lock = update->backend_data;
if (update->flags & REF_DELETING &&
!(update->flags & REF_LOG_ONLY)) {
update->flags |= REF_DELETED_RMDIR;
if (!(update->type & REF_ISPACKED) ||
update->type & REF_ISSYMREF) {
/* It is a loose reference. */
strbuf_reset(&sb);
files_ref_path(refs, &sb, lock->ref_name);
if (unlink_or_msg(sb.buf, err)) {
ret = TRANSACTION_GENERIC_ERROR;
goto cleanup;
}
}
}
}
clear_loose_ref_cache(refs);
cleanup:
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
files_transaction_cleanup(refs, transaction);
for (i = 0; i < transaction->nr; i++) {
struct ref_update *update = transaction->updates[i];
if (update->flags & REF_DELETED_RMDIR) {
/*
* The reference was deleted. Delete any
* empty parent directories. (Note that this
* can only work because we have already
* removed the lockfile.)
*/
try_remove_empty_parents(refs, update->refname,
REMOVE_EMPTY_PARENTS_REF);
}
}
strbuf_release(&sb);
return ret;
}
static int files_transaction_abort(struct ref_store *ref_store,
struct ref_transaction *transaction,
struct strbuf *err)
{
files_ref_store: use a transaction to update packed refs When processing a `files_ref_store` transaction, it is sometimes necessary to delete some references from the "packed-refs" file. Do that using a reference transaction conducted against the `packed_ref_store`. This change further decouples `files_ref_store` from `packed_ref_store`. It also fixes multiple problems, including the two revealed by test cases added in the previous commit. First, the old code didn't obtain the `packed-refs` lock until `files_transaction_finish()`. This means that a failure to acquire the `packed-refs` lock (e.g., due to contention with another process) wasn't detected until it was too late (problems like this are supposed to be detected in the "prepare" phase). The new code acquires the `packed-refs` lock in `files_transaction_prepare()`, the same stage of the processing when the loose reference locks are being acquired, removing another reason why the "prepare" phase might succeed and the "finish" phase might nevertheless fail. Second, the old code deleted the loose version of a reference before deleting any packed version of the same reference. This left a moment when another process might think that the packed version of the reference is current, which is incorrect. (Even worse, the packed version of the reference can be arbitrarily old, and might even point at an object that has since been garbage-collected.) Third, if a reference deletion fails to acquire the `packed-refs` lock altogether, then the old code might leave the repository in the incorrect state (possibly corrupt) described in the previous paragraph. Now we activate the new "packed-refs" file (sans any references that are being deleted) *before* deleting the corresponding loose references. But we hold the "packed-refs" lock until after the loose references have been finalized, thus preventing a simultaneous "pack-refs" process from packing the loose version of the reference in the time gap, which would otherwise defeat our attempt to delete it. Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 13:51:51 +00:00
struct files_ref_store *refs =
files_downcast(ref_store, 0, "ref_transaction_abort");
files_transaction_cleanup(refs, transaction);
return 0;
}
static int ref_present(const char *refname,
const struct object_id *oid, int flags, void *cb_data)
{
struct string_list *affected_refnames = cb_data;
return string_list_has_string(affected_refnames, refname);
}
static int files_initial_transaction_commit(struct ref_store *ref_store,
struct ref_transaction *transaction,
struct strbuf *err)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_WRITE,
"initial_ref_transaction_commit");
size_t i;
int ret = 0;
struct string_list affected_refnames = STRING_LIST_INIT_NODUP;
struct ref_transaction *packed_transaction = NULL;
assert(err);
if (transaction->state != REF_TRANSACTION_OPEN)
BUG("commit called for transaction that is not open");
/* Fail if a refname appears more than once in the transaction: */
for (i = 0; i < transaction->nr; i++)
string_list_append(&affected_refnames,
transaction->updates[i]->refname);
string_list_sort(&affected_refnames);
if (ref_update_reject_duplicates(&affected_refnames, err)) {
ret = TRANSACTION_GENERIC_ERROR;
goto cleanup;
}
/*
* It's really undefined to call this function in an active
* repository or when there are existing references: we are
* only locking and changing packed-refs, so (1) any
* simultaneous processes might try to change a reference at
* the same time we do, and (2) any existing loose versions of
* the references that we are setting would have precedence
* over our values. But some remote helpers create the remote
* "HEAD" and "master" branches before calling this function,
* so here we really only check that none of the references
* that we are creating already exists.
*/
if (refs_for_each_rawref(&refs->base, ref_present,
&affected_refnames))
BUG("initial ref transaction called with existing refs");
packed_transaction = ref_store_transaction_begin(refs->packed_ref_store, err);
if (!packed_transaction) {
ret = TRANSACTION_GENERIC_ERROR;
goto cleanup;
}
for (i = 0; i < transaction->nr; i++) {
struct ref_update *update = transaction->updates[i];
if ((update->flags & REF_HAVE_OLD) &&
!is_null_oid(&update->old_oid))
BUG("initial ref transaction with old_sha1 set");
if (refs_verify_refname_available(&refs->base, update->refname,
&affected_refnames, NULL,
err)) {
ret = TRANSACTION_NAME_CONFLICT;
goto cleanup;
}
/*
* Add a reference creation for this reference to the
* packed-refs transaction:
*/
ref_transaction_add_update(packed_transaction, update->refname,
update->flags & ~REF_HAVE_OLD,
&update->new_oid, &update->old_oid,
NULL);
}
if (packed_refs_lock(refs->packed_ref_store, 0, err)) {
ret = TRANSACTION_GENERIC_ERROR;
goto cleanup;
}
if (initial_ref_transaction_commit(packed_transaction, err)) {
ret = TRANSACTION_GENERIC_ERROR;
}
packed_refs_unlock(refs->packed_ref_store);
cleanup:
if (packed_transaction)
ref_transaction_free(packed_transaction);
transaction->state = REF_TRANSACTION_CLOSED;
string_list_clear(&affected_refnames, 0);
return ret;
}
struct expire_reflog_cb {
unsigned int flags;
reflog_expiry_should_prune_fn *should_prune_fn;
void *policy_cb;
FILE *newlog;
struct object_id last_kept_oid;
};
static int expire_reflog_ent(struct object_id *ooid, struct object_id *noid,
const char *email, timestamp_t timestamp, int tz,
const char *message, void *cb_data)
{
struct expire_reflog_cb *cb = cb_data;
struct expire_reflog_policy_cb *policy_cb = cb->policy_cb;
if (cb->flags & EXPIRE_REFLOGS_REWRITE)
ooid = &cb->last_kept_oid;
if ((*cb->should_prune_fn)(ooid, noid, email, timestamp, tz,
message, policy_cb)) {
if (!cb->newlog)
printf("would prune %s", message);
else if (cb->flags & EXPIRE_REFLOGS_VERBOSE)
printf("prune %s", message);
} else {
if (cb->newlog) {
fprintf(cb->newlog, "%s %s %s %"PRItime" %+05d\t%s",
oid_to_hex(ooid), oid_to_hex(noid),
email, timestamp, tz, message);
oidcpy(&cb->last_kept_oid, noid);
}
if (cb->flags & EXPIRE_REFLOGS_VERBOSE)
printf("keep %s", message);
}
return 0;
}
static int files_reflog_expire(struct ref_store *ref_store,
const char *refname,
unsigned int flags,
reflog_expiry_prepare_fn prepare_fn,
reflog_expiry_should_prune_fn should_prune_fn,
reflog_expiry_cleanup_fn cleanup_fn,
void *policy_cb_data)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_WRITE, "reflog_expire");
struct lock_file reflog_lock = LOCK_INIT;
struct expire_reflog_cb cb;
struct ref_lock *lock;
struct strbuf log_file_sb = STRBUF_INIT;
char *log_file;
int status = 0;
struct strbuf err = STRBUF_INIT;
reflog expire: don't lock reflogs using previously seen OID During reflog expiry, the cmd_reflog_expire() function first iterates over all reflogs in logs/*, and then one-by-one acquires the lock for each one and expires it. This behavior has been with us since this command was implemented in 4264dc15e1 ("git reflog expire", 2006-12-19). Change this to stop calling lock_ref_oid_basic() with the OID we saw when we looped over the logs, instead have it pass the OID it managed to lock. This mostly mitigates a race condition where e.g. "git gc" will fail in a concurrently updated repository because the branch moved since "git reflog expire --all" was started. I.e. with: error: cannot lock ref '<refname>': ref '<refname>' is at <OID-A> but expected <OID-B> This behavior of passing in an "oid" was needed for an edge-case that I've untangled in this and preceding commits though, namely that we needed this OID because we'd: 1. Lookup the reflog name/OID via dwim_log() 2. With that OID, lock the reflog 3. Later in builtin/reflog.c we use the OID we looked as input to lookup_commit_reference_gently(), assured that it's equal to the OID we got from dwim_log(). We can be sure that this change is safe to make because between dwim_log (step #1) and lock_ref_oid_basic (step #2) there was no other logic relevant to the OID or expiry run in the cmd_reflog_expire() caller. We can thus treat that code as a black box, before and after this change it would get an OID that's been locked, the only difference is that now we mostly won't be failing to get the lock due to the TOCTOU race[0]. That failure was purely an implementation detail in how the "current OID" was looked up, it was divorced from the locking mechanism. What do we mean with "mostly"? It mostly mitigates it because we'll still run into cases where the ref is locked and being updated as we want to expire it, and other git processes wanting to update the refs will in turn race with us as we expire the reflog. That remaining race can in turn be mitigated with the core.filesRefLockTimeout setting, see 4ff0f01cb7 ("refs: retry acquiring reference locks for 100ms", 2017-08-21). In practice if that value is high enough we'll probably never have ref updates or reflog expiry failing, since the clients involved will retry for far longer than the time any of those operations could take. See [1] for an initial report of how this impacted "git gc" and a large discussion about this change in early 2019. In particular patch looked good to Michael Haggerty, see his[2]. That message seems to not have made it to the ML archive, its content is quoted in full in my [3]. I'm leaving behind now-unused code the refs API etc. that takes the now-NULL "unused_oid" argument, and other code that can be simplified now that we never have on OID in that context, that'll be cleaned up in subsequent commits, but for now let's narrowly focus on fixing the "git gc" issue. As the modified assert() shows we always pass a NULL oid to reflog_expire() now. Unfortunately this sort of probabilistic contention is hard to turn into a test. I've tested this by running the following three subshells in concurrent terminals: ( rm -rf /tmp/git && git init /tmp/git && while true do head -c 10 /dev/urandom | hexdump >/tmp/git/out && git -C /tmp/git add out && git -C /tmp/git commit -m"out" done ) ( rm -rf /tmp/git-clone && git clone file:///tmp/git /tmp/git-clone && while git -C /tmp/git-clone pull do date done ) ( while git -C /tmp/git-clone reflog expire --all do date done ) Before this change the "reflog expire" would fail really quickly with the "but expected" error noted above. After this change both the "pull" and "reflog expire" will run for a while, but eventually fail because I get unlucky with core.filesRefLockTimeout (the "reflog expire" is in a really tight loop). As noted above that can in turn be mitigated with higher values of core.filesRefLockTimeout than the 100ms default. As noted in the commentary added in the preceding commit there's also the case of branches being racily deleted, that can be tested by adding this to the above: ( while git -C /tmp/git-clone branch topic master && git -C /tmp/git-clone branch -D topic do date done ) With core.filesRefLockTimeout set to 10 seconds (it can probably be a lot lower) I managed to run all four of these concurrently for about an hour, and accumulated ~125k commits, auto-gc's and all, and didn't have a single failure. The loops visibly stall while waiting for the lock, but that's expected and desired behavior. 0. https://en.wikipedia.org/wiki/Time-of-check_to_time-of-use 1. https://lore.kernel.org/git/87tvg7brlm.fsf@evledraar.gmail.com/ 2. http://lore.kernel.org/git/b870a17d-2103-41b8-3cbc-7389d5fff33a@alum.mit.edu 3. https://lore.kernel.org/git/87pnqkco8v.fsf@evledraar.gmail.com/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-08-23 11:36:10 +00:00
const struct object_id *oid;
memset(&cb, 0, sizeof(cb));
cb.flags = flags;
cb.policy_cb = policy_cb_data;
cb.should_prune_fn = should_prune_fn;
/*
* The reflog file is locked by holding the lock on the
* reference itself, plus we might need to update the
* reference if --updateref was specified:
*/
refs/files: remove "name exist?" check in lock_ref_oid_basic() In lock_ref_oid_basic() we'll happily lock a reference that doesn't exist yet. That's normal, and is how references are initially born, but we don't need to retain checks here in lock_ref_oid_basic() about the state of the ref, when what we're checking is either checked already, or something we're about to discover by trying to lock the ref with raceproof_create_file(). The one exception is the caller in files_reflog_expire(), who passes us a "type" to find out if the reference is a symref or not. We can move the that logic over to that caller, which can now defer its discovery of whether or not the ref is a symref until it's needed. In the preceding commit an exhaustive regression test was added for that case in a new test in "t1417-reflog-updateref.sh". The improved diagnostics here were added in 5b2d8d6f218 (lock_ref_sha1_basic(): improve diagnostics for ref D/F conflicts, 2015-05-11), and then much of the surrounding code went away recently in my 245fbba46d6 (refs/files: remove unused "errno == EISDIR" code, 2021-08-23). The refs_resolve_ref_unsafe() code being removed here looks like it should be tasked with doing that, but it's actually redundant to other code. The reason for that is as noted in 245fbba46d6 this once widely used function now only has a handful of callers left, which all handle this case themselves. To the extent that we're racy between their check and ours removing this check actually improves the situation, as we'll be doing fewer things between the not-under-lock initial check and acquiring the lock. Why this is OK for all the remaining callers of lock_ref_oid_basic() is noted below. There are only two of those callers: * "git branch -[cm] <oldbranch> <newbranch>": In files_copy_or_rename_ref() we'll call this when we copy or rename refs via rename_ref() and copy_ref(). but only after we've checked if the refname exists already via its own call to refs_resolve_ref_unsafe() and refs_rename_ref_available(). As the updated comment to the latter here notes neither of those are actually needed. If we delete not only this code but also refs_rename_ref_available() we'll do just fine, we'll just emit a less friendly error message if e.g. "git branch -m A B/C" would have a D/F conflict with a "B" file. Actually we'd probably die before that in case reflogs for the branch existed, i.e. when the try to rename() or copy_file() the relevant reflog, since if we've got a D/F conflict with a branch name we'll probably also have the same with its reflogs (but not necessarily, we might have reflogs, but it might not). As some #leftoverbits that code seems buggy to me, i.e. the reflog "protocol" should be to get a lock on the main ref, and then perform ref and/or reflog operations. That code dates back to c976d415e53 (git-branch: add options and tests for branch renaming, 2006-11-28) and probably pre-dated the solidifying of that convention. But in any case, that edge case is not our bug or problem right now. * "git reflog expire <ref>": In files_reflog_expire() we'll call this without previous ref existence checking in files-backend.c, but that code is in turn called by code that's just finished checking if the refname whose reflog we're expiring exists. See ae35e16cd43 (reflog expire: don't lock reflogs using previously seen OID, 2021-08-23) for the current state of that code, and 5e6f003ca8a (reflog_expire(): ignore --updateref for symbolic references, 2015-03-03) for the code we'd break if we only did a "update = !!ref" here, which is covered by the aforementioned regression test in "t1417-reflog-updateref.sh". Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-16 09:39:13 +00:00
lock = lock_ref_oid_basic(refs, refname, &err);
if (!lock) {
error("cannot lock ref '%s': %s", refname, err.buf);
strbuf_release(&err);
return -1;
}
reflog expire: don't lock reflogs using previously seen OID During reflog expiry, the cmd_reflog_expire() function first iterates over all reflogs in logs/*, and then one-by-one acquires the lock for each one and expires it. This behavior has been with us since this command was implemented in 4264dc15e1 ("git reflog expire", 2006-12-19). Change this to stop calling lock_ref_oid_basic() with the OID we saw when we looped over the logs, instead have it pass the OID it managed to lock. This mostly mitigates a race condition where e.g. "git gc" will fail in a concurrently updated repository because the branch moved since "git reflog expire --all" was started. I.e. with: error: cannot lock ref '<refname>': ref '<refname>' is at <OID-A> but expected <OID-B> This behavior of passing in an "oid" was needed for an edge-case that I've untangled in this and preceding commits though, namely that we needed this OID because we'd: 1. Lookup the reflog name/OID via dwim_log() 2. With that OID, lock the reflog 3. Later in builtin/reflog.c we use the OID we looked as input to lookup_commit_reference_gently(), assured that it's equal to the OID we got from dwim_log(). We can be sure that this change is safe to make because between dwim_log (step #1) and lock_ref_oid_basic (step #2) there was no other logic relevant to the OID or expiry run in the cmd_reflog_expire() caller. We can thus treat that code as a black box, before and after this change it would get an OID that's been locked, the only difference is that now we mostly won't be failing to get the lock due to the TOCTOU race[0]. That failure was purely an implementation detail in how the "current OID" was looked up, it was divorced from the locking mechanism. What do we mean with "mostly"? It mostly mitigates it because we'll still run into cases where the ref is locked and being updated as we want to expire it, and other git processes wanting to update the refs will in turn race with us as we expire the reflog. That remaining race can in turn be mitigated with the core.filesRefLockTimeout setting, see 4ff0f01cb7 ("refs: retry acquiring reference locks for 100ms", 2017-08-21). In practice if that value is high enough we'll probably never have ref updates or reflog expiry failing, since the clients involved will retry for far longer than the time any of those operations could take. See [1] for an initial report of how this impacted "git gc" and a large discussion about this change in early 2019. In particular patch looked good to Michael Haggerty, see his[2]. That message seems to not have made it to the ML archive, its content is quoted in full in my [3]. I'm leaving behind now-unused code the refs API etc. that takes the now-NULL "unused_oid" argument, and other code that can be simplified now that we never have on OID in that context, that'll be cleaned up in subsequent commits, but for now let's narrowly focus on fixing the "git gc" issue. As the modified assert() shows we always pass a NULL oid to reflog_expire() now. Unfortunately this sort of probabilistic contention is hard to turn into a test. I've tested this by running the following three subshells in concurrent terminals: ( rm -rf /tmp/git && git init /tmp/git && while true do head -c 10 /dev/urandom | hexdump >/tmp/git/out && git -C /tmp/git add out && git -C /tmp/git commit -m"out" done ) ( rm -rf /tmp/git-clone && git clone file:///tmp/git /tmp/git-clone && while git -C /tmp/git-clone pull do date done ) ( while git -C /tmp/git-clone reflog expire --all do date done ) Before this change the "reflog expire" would fail really quickly with the "but expected" error noted above. After this change both the "pull" and "reflog expire" will run for a while, but eventually fail because I get unlucky with core.filesRefLockTimeout (the "reflog expire" is in a really tight loop). As noted above that can in turn be mitigated with higher values of core.filesRefLockTimeout than the 100ms default. As noted in the commentary added in the preceding commit there's also the case of branches being racily deleted, that can be tested by adding this to the above: ( while git -C /tmp/git-clone branch topic master && git -C /tmp/git-clone branch -D topic do date done ) With core.filesRefLockTimeout set to 10 seconds (it can probably be a lot lower) I managed to run all four of these concurrently for about an hour, and accumulated ~125k commits, auto-gc's and all, and didn't have a single failure. The loops visibly stall while waiting for the lock, but that's expected and desired behavior. 0. https://en.wikipedia.org/wiki/Time-of-check_to_time-of-use 1. https://lore.kernel.org/git/87tvg7brlm.fsf@evledraar.gmail.com/ 2. http://lore.kernel.org/git/b870a17d-2103-41b8-3cbc-7389d5fff33a@alum.mit.edu 3. https://lore.kernel.org/git/87pnqkco8v.fsf@evledraar.gmail.com/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-08-23 11:36:10 +00:00
oid = &lock->old_oid;
/*
* When refs are deleted, their reflog is deleted before the
* ref itself is deleted. This is because there is no separate
* lock for reflog; instead we take a lock on the ref with
* lock_ref_oid_basic().
*
* If a race happens and the reflog doesn't exist after we've
* acquired the lock that's OK. We've got nothing more to do;
* We were asked to delete the reflog, but someone else
* deleted it! The caller doesn't care that we deleted it,
* just that it is deleted. So we can return successfully.
*/
if (!refs_reflog_exists(ref_store, refname)) {
unlock_ref(lock);
return 0;
}
files_reflog_path(refs, &log_file_sb, refname);
log_file = strbuf_detach(&log_file_sb, NULL);
if (!(flags & EXPIRE_REFLOGS_DRY_RUN)) {
/*
* Even though holding $GIT_DIR/logs/$reflog.lock has
* no locking implications, we use the lock_file
* machinery here anyway because it does a lot of the
* work we need, including cleaning up if the program
* exits unexpectedly.
*/
if (hold_lock_file_for_update(&reflog_lock, log_file, 0) < 0) {
struct strbuf err = STRBUF_INIT;
unable_to_lock_message(log_file, errno, &err);
error("%s", err.buf);
strbuf_release(&err);
goto failure;
}
cb.newlog = fdopen_lock_file(&reflog_lock, "w");
if (!cb.newlog) {
error("cannot fdopen %s (%s)",
get_lock_file_path(&reflog_lock), strerror(errno));
goto failure;
}
}
(*prepare_fn)(refname, oid, cb.policy_cb);
refs_for_each_reflog_ent(ref_store, refname, expire_reflog_ent, &cb);
(*cleanup_fn)(cb.policy_cb);
if (!(flags & EXPIRE_REFLOGS_DRY_RUN)) {
/*
* It doesn't make sense to adjust a reference pointed
* to by a symbolic ref based on expiring entries in
* the symbolic reference's reflog. Nor can we update
* a reference if there are no remaining reflog
* entries.
*/
refs/files: remove "name exist?" check in lock_ref_oid_basic() In lock_ref_oid_basic() we'll happily lock a reference that doesn't exist yet. That's normal, and is how references are initially born, but we don't need to retain checks here in lock_ref_oid_basic() about the state of the ref, when what we're checking is either checked already, or something we're about to discover by trying to lock the ref with raceproof_create_file(). The one exception is the caller in files_reflog_expire(), who passes us a "type" to find out if the reference is a symref or not. We can move the that logic over to that caller, which can now defer its discovery of whether or not the ref is a symref until it's needed. In the preceding commit an exhaustive regression test was added for that case in a new test in "t1417-reflog-updateref.sh". The improved diagnostics here were added in 5b2d8d6f218 (lock_ref_sha1_basic(): improve diagnostics for ref D/F conflicts, 2015-05-11), and then much of the surrounding code went away recently in my 245fbba46d6 (refs/files: remove unused "errno == EISDIR" code, 2021-08-23). The refs_resolve_ref_unsafe() code being removed here looks like it should be tasked with doing that, but it's actually redundant to other code. The reason for that is as noted in 245fbba46d6 this once widely used function now only has a handful of callers left, which all handle this case themselves. To the extent that we're racy between their check and ours removing this check actually improves the situation, as we'll be doing fewer things between the not-under-lock initial check and acquiring the lock. Why this is OK for all the remaining callers of lock_ref_oid_basic() is noted below. There are only two of those callers: * "git branch -[cm] <oldbranch> <newbranch>": In files_copy_or_rename_ref() we'll call this when we copy or rename refs via rename_ref() and copy_ref(). but only after we've checked if the refname exists already via its own call to refs_resolve_ref_unsafe() and refs_rename_ref_available(). As the updated comment to the latter here notes neither of those are actually needed. If we delete not only this code but also refs_rename_ref_available() we'll do just fine, we'll just emit a less friendly error message if e.g. "git branch -m A B/C" would have a D/F conflict with a "B" file. Actually we'd probably die before that in case reflogs for the branch existed, i.e. when the try to rename() or copy_file() the relevant reflog, since if we've got a D/F conflict with a branch name we'll probably also have the same with its reflogs (but not necessarily, we might have reflogs, but it might not). As some #leftoverbits that code seems buggy to me, i.e. the reflog "protocol" should be to get a lock on the main ref, and then perform ref and/or reflog operations. That code dates back to c976d415e53 (git-branch: add options and tests for branch renaming, 2006-11-28) and probably pre-dated the solidifying of that convention. But in any case, that edge case is not our bug or problem right now. * "git reflog expire <ref>": In files_reflog_expire() we'll call this without previous ref existence checking in files-backend.c, but that code is in turn called by code that's just finished checking if the refname whose reflog we're expiring exists. See ae35e16cd43 (reflog expire: don't lock reflogs using previously seen OID, 2021-08-23) for the current state of that code, and 5e6f003ca8a (reflog_expire(): ignore --updateref for symbolic references, 2015-03-03) for the code we'd break if we only did a "update = !!ref" here, which is covered by the aforementioned regression test in "t1417-reflog-updateref.sh". Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-16 09:39:13 +00:00
int update = 0;
if ((flags & EXPIRE_REFLOGS_UPDATE_REF) &&
!is_null_oid(&cb.last_kept_oid)) {
int ignore_errno;
int type;
const char *ref;
ref = refs_werrres_ref_unsafe(&refs->base, refname,
RESOLVE_REF_NO_RECURSE,
NULL, &type,
&ignore_errno);
update = !!(ref && !(type & REF_ISSYMREF));
}
if (close_lock_file_gently(&reflog_lock)) {
status |= error("couldn't write %s: %s", log_file,
strerror(errno));
rollback_lock_file(&reflog_lock);
} else if (update &&
(write_in_full(get_lock_file_fd(&lock->lk),
oid_to_hex(&cb.last_kept_oid), the_hash_algo->hexsz) < 0 ||
write_str_in_full(get_lock_file_fd(&lock->lk), "\n") < 0 ||
close_ref_gently(lock) < 0)) {
status |= error("couldn't write %s",
get_lock_file_path(&lock->lk));
rollback_lock_file(&reflog_lock);
} else if (commit_lock_file(&reflog_lock)) {
status |= error("unable to write reflog '%s' (%s)",
log_file, strerror(errno));
} else if (update && commit_ref(lock)) {
status |= error("couldn't set %s", lock->ref_name);
}
}
free(log_file);
unlock_ref(lock);
return status;
failure:
rollback_lock_file(&reflog_lock);
free(log_file);
unlock_ref(lock);
return -1;
}
static int files_init_db(struct ref_store *ref_store, struct strbuf *err)
{
struct files_ref_store *refs =
files_downcast(ref_store, REF_STORE_WRITE, "init_db");
struct strbuf sb = STRBUF_INIT;
/*
* Create .git/refs/{heads,tags}
*/
files_ref_path(refs, &sb, "refs/heads");
safe_create_dir(sb.buf, 1);
strbuf_reset(&sb);
files_ref_path(refs, &sb, "refs/tags");
safe_create_dir(sb.buf, 1);
strbuf_release(&sb);
return 0;
}
struct ref_storage_be refs_be_files = {
NULL,
"files",
files_ref_store_create,
files_init_db,
files_transaction_prepare,
files_transaction_finish,
files_transaction_abort,
files_initial_transaction_commit,
files_pack_refs,
files_create_symref,
files_delete_refs,
files_rename_ref,
branch: add a --copy (-c) option to go with --move (-m) Add the ability to --copy a branch and its reflog and configuration, this uses the same underlying machinery as the --move (-m) option except the reflog and configuration is copied instead of being moved. This is useful for e.g. copying a topic branch to a new version, e.g. work to work-2 after submitting the work topic to the list, while preserving all the tracking info and other configuration that goes with the branch, and unlike --move keeping the other already-submitted branch around for reference. Like --move, when the source branch is the currently checked out branch the HEAD is moved to the destination branch. In the case of --move we don't really have a choice (other than remaining on a detached HEAD) and in order to keep the functionality consistent, we are doing it in similar way for --copy too. The most common usage of this feature is expected to be moving to a new topic branch which is a copy of the current one, in that case moving to the target branch is what the user wants, and doesn't unexpectedly behave differently than --move would. One outstanding caveat of this implementation is that: git checkout maint && git checkout master && git branch -c topic && git checkout - Will check out 'maint' instead of 'master'. This is because the @{-N} feature (or its -1 shorthand "-") relies on HEAD reflogs created by the checkout command, so in this case we'll checkout maint instead of master, as the user might expect. What to do about that is left to a future change. Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Sahil Dua <sahildua2305@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-18 21:19:16 +00:00
files_copy_ref,
files_ref_iterator_begin,
files_read_raw_ref,
files_reflog_iterator_begin,
files_for_each_reflog_ent,
files_for_each_reflog_ent_reverse,
files_reflog_exists,
files_create_reflog,
files_delete_reflog,
files_reflog_expire
};