git/fsck.c

1417 lines
37 KiB
C
Raw Normal View History

#include "git-compat-util.h"
#include "alloc.h"
#include "date.h"
#include "hex.h"
#include "object-store.h"
#include "repository.h"
#include "object.h"
#include "attr.h"
#include "blob.h"
#include "tree.h"
#include "tree-walk.h"
#include "commit.h"
#include "tag.h"
#include "fsck.h"
#include "refs.h"
#include "url.h"
#include "utf8.h"
#include "decorate.h"
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
#include "oidset.h"
#include "packfile.h"
#include "submodule-config.h"
#include "config.h"
#include "credential.h"
#include "help.h"
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
#define STR(x) #x
#define MSG_ID(id, msg_type) { STR(id), NULL, NULL, FSCK_##msg_type },
static struct {
const char *id_string;
const char *downcased;
const char *camelcased;
enum fsck_msg_type msg_type;
} msg_id_info[FSCK_MSG_MAX + 1] = {
FOREACH_FSCK_MSG_ID(MSG_ID)
{ NULL, NULL, NULL, -1 }
};
#undef MSG_ID
#undef STR
static void prepare_msg_ids(void)
{
int i;
if (msg_id_info[0].downcased)
return;
/* convert id_string to lower case, without underscores. */
for (i = 0; i < FSCK_MSG_MAX; i++) {
const char *p = msg_id_info[i].id_string;
int len = strlen(p);
char *q = xmalloc(len);
msg_id_info[i].downcased = q;
while (*p)
if (*p == '_')
p++;
else
*(q)++ = tolower(*(p)++);
*q = '\0';
p = msg_id_info[i].id_string;
q = xmalloc(len);
msg_id_info[i].camelcased = q;
while (*p) {
if (*p == '_') {
p++;
if (*p)
*q++ = *p++;
} else {
*q++ = tolower(*p++);
}
}
*q = '\0';
}
}
static int parse_msg_id(const char *text)
{
int i;
prepare_msg_ids();
for (i = 0; i < FSCK_MSG_MAX; i++)
if (!strcmp(text, msg_id_info[i].downcased))
return i;
return -1;
}
void list_config_fsck_msg_ids(struct string_list *list, const char *prefix)
{
int i;
prepare_msg_ids();
for (i = 0; i < FSCK_MSG_MAX; i++)
list_config_item(list, prefix, msg_id_info[i].camelcased);
}
static enum fsck_msg_type fsck_msg_type(enum fsck_msg_id msg_id,
struct fsck_options *options)
{
assert(msg_id >= 0 && msg_id < FSCK_MSG_MAX);
if (!options->msg_type) {
enum fsck_msg_type msg_type = msg_id_info[msg_id].msg_type;
if (options->strict && msg_type == FSCK_WARN)
msg_type = FSCK_ERROR;
return msg_type;
}
return options->msg_type[msg_id];
}
static enum fsck_msg_type parse_msg_type(const char *str)
{
if (!strcmp(str, "error"))
return FSCK_ERROR;
else if (!strcmp(str, "warn"))
return FSCK_WARN;
else if (!strcmp(str, "ignore"))
return FSCK_IGNORE;
else
die("Unknown fsck message type: '%s'", str);
}
int is_valid_msg_type(const char *msg_id, const char *msg_type)
{
if (parse_msg_id(msg_id) < 0)
return 0;
parse_msg_type(msg_type);
return 1;
}
void fsck_set_msg_type_from_ids(struct fsck_options *options,
enum fsck_msg_id msg_id,
enum fsck_msg_type msg_type)
{
if (!options->msg_type) {
int i;
enum fsck_msg_type *severity;
ALLOC_ARRAY(severity, FSCK_MSG_MAX);
for (i = 0; i < FSCK_MSG_MAX; i++)
severity[i] = fsck_msg_type(i, options);
options->msg_type = severity;
}
options->msg_type[msg_id] = msg_type;
}
void fsck_set_msg_type(struct fsck_options *options,
const char *msg_id_str, const char *msg_type_str)
{
int msg_id = parse_msg_id(msg_id_str);
enum fsck_msg_type msg_type = parse_msg_type(msg_type_str);
if (msg_id < 0)
die("Unhandled message id: %s", msg_id_str);
if (msg_type != FSCK_ERROR && msg_id_info[msg_id].msg_type == FSCK_FATAL)
die("Cannot demote %s to %s", msg_id_str, msg_type_str);
fsck_set_msg_type_from_ids(options, msg_id, msg_type);
}
void fsck_set_msg_types(struct fsck_options *options, const char *values)
{
char *buf = xstrdup(values), *to_free = buf;
int done = 0;
while (!done) {
int len = strcspn(buf, " ,|"), equal;
done = !buf[len];
if (!len) {
buf++;
continue;
}
buf[len] = '\0';
for (equal = 0;
equal < len && buf[equal] != '=' && buf[equal] != ':';
equal++)
buf[equal] = tolower(buf[equal]);
buf[equal] = '\0';
if (!strcmp(buf, "skiplist")) {
if (equal == len)
die("skiplist requires a path");
oidset_parse_file(&options->skiplist, buf + equal + 1);
buf += len + 1;
continue;
}
if (equal == len)
die("Missing '=': '%s'", buf);
fsck_set_msg_type(options, buf, buf + equal + 1);
buf += len + 1;
}
free(to_free);
}
static int object_on_skiplist(struct fsck_options *opts,
const struct object_id *oid)
fsck: check skiplist for object in fsck_blob() Since commit ed8b10f631 ("fsck: check .gitmodules content", 2018-05-02), fsck will issue an error message for '.gitmodules' content that cannot be parsed correctly. This is the case, even when the corresponding blob object has been included on the skiplist. For example, using the cgit repository, we see the following: $ git fsck Checking object directories: 100% (256/256), done. error: bad config line 5 in blob .gitmodules error in blob 51dd1eff1edc663674df9ab85d2786a40f7ae3a5: gitmodulesParse: could not parse gitmodules blob Checking objects: 100% (6626/6626), done. $ $ git config fsck.skiplist '.git/skip' $ echo 51dd1eff1edc663674df9ab85d2786a40f7ae3a5 >.git/skip $ $ git fsck Checking object directories: 100% (256/256), done. error: bad config line 5 in blob .gitmodules Checking objects: 100% (6626/6626), done. $ Note that the error message issued by the config parser is still present, despite adding the object-id of the blob to the skiplist. One solution would be to provide a means of suppressing the messages issued by the config parser. However, given that (logically) we are asking fsck to ignore this object, a simpler approach is to just not call the config parser if the object is to be skipped. Add a check to the 'fsck_blob()' processing function, to determine if the object is on the skiplist and, if so, exit the function early. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-27 18:39:53 +00:00
{
return opts && oid && oidset_contains(&opts->skiplist, oid);
fsck: check skiplist for object in fsck_blob() Since commit ed8b10f631 ("fsck: check .gitmodules content", 2018-05-02), fsck will issue an error message for '.gitmodules' content that cannot be parsed correctly. This is the case, even when the corresponding blob object has been included on the skiplist. For example, using the cgit repository, we see the following: $ git fsck Checking object directories: 100% (256/256), done. error: bad config line 5 in blob .gitmodules error in blob 51dd1eff1edc663674df9ab85d2786a40f7ae3a5: gitmodulesParse: could not parse gitmodules blob Checking objects: 100% (6626/6626), done. $ $ git config fsck.skiplist '.git/skip' $ echo 51dd1eff1edc663674df9ab85d2786a40f7ae3a5 >.git/skip $ $ git fsck Checking object directories: 100% (256/256), done. error: bad config line 5 in blob .gitmodules Checking objects: 100% (6626/6626), done. $ Note that the error message issued by the config parser is still present, despite adding the object-id of the blob to the skiplist. One solution would be to provide a means of suppressing the messages issued by the config parser. However, given that (logically) we are asking fsck to ignore this object, a simpler approach is to just not call the config parser if the object is to be skipped. Add a check to the 'fsck_blob()' processing function, to determine if the object is on the skiplist and, if so, exit the function early. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-27 18:39:53 +00:00
}
__attribute__((format (printf, 5, 6)))
static int report(struct fsck_options *options,
const struct object_id *oid, enum object_type object_type,
enum fsck_msg_id msg_id, const char *fmt, ...)
{
va_list ap;
struct strbuf sb = STRBUF_INIT;
enum fsck_msg_type msg_type = fsck_msg_type(msg_id, options);
int result;
if (msg_type == FSCK_IGNORE)
return 0;
if (object_on_skiplist(options, oid))
return 0;
if (msg_type == FSCK_FATAL)
msg_type = FSCK_ERROR;
else if (msg_type == FSCK_INFO)
msg_type = FSCK_WARN;
prepare_msg_ids();
strbuf_addf(&sb, "%s: ", msg_id_info[msg_id].camelcased);
va_start(ap, fmt);
strbuf_vaddf(&sb, fmt, ap);
result = options->error_func(options, oid, object_type,
msg_type, msg_id, sb.buf);
strbuf_release(&sb);
va_end(ap);
return result;
}
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
void fsck_enable_object_names(struct fsck_options *options)
{
if (!options->object_names)
options->object_names = kh_init_oid_map();
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
}
const char *fsck_get_object_name(struct fsck_options *options,
const struct object_id *oid)
{
khiter_t pos;
if (!options->object_names)
return NULL;
pos = kh_get_oid_map(options->object_names, *oid);
if (pos >= kh_end(options->object_names))
return NULL;
return kh_value(options->object_names, pos);
}
void fsck_put_object_name(struct fsck_options *options,
const struct object_id *oid,
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
const char *fmt, ...)
{
va_list ap;
struct strbuf buf = STRBUF_INIT;
khiter_t pos;
int hashret;
if (!options->object_names)
return;
pos = kh_put_oid_map(options->object_names, *oid, &hashret);
if (!hashret)
return;
va_start(ap, fmt);
strbuf_vaddf(&buf, fmt, ap);
kh_value(options->object_names, pos) = strbuf_detach(&buf, NULL);
va_end(ap);
}
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
const char *fsck_describe_object(struct fsck_options *options,
const struct object_id *oid)
{
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
static struct strbuf bufs[] = {
STRBUF_INIT, STRBUF_INIT, STRBUF_INIT, STRBUF_INIT
};
static int b = 0;
struct strbuf *buf;
const char *name = fsck_get_object_name(options, oid);
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
buf = bufs + b;
b = (b + 1) % ARRAY_SIZE(bufs);
strbuf_reset(buf);
strbuf_addstr(buf, oid_to_hex(oid));
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
if (name)
strbuf_addf(buf, " (%s)", name);
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
return buf->buf;
}
static int fsck_walk_tree(struct tree *tree, void *data, struct fsck_options *options)
{
struct tree_desc desc;
struct name_entry entry;
int res = 0;
const char *name;
if (parse_tree(tree))
return -1;
name = fsck_get_object_name(options, &tree->object.oid);
if (init_tree_desc_gently(&desc, tree->buffer, tree->size, 0))
return -1;
while (tree_entry_gently(&desc, &entry)) {
struct object *obj;
int result;
if (S_ISGITLINK(entry.mode))
continue;
if (S_ISDIR(entry.mode)) {
obj = (struct object *)lookup_tree(the_repository, &entry.oid);
if (name && obj)
fsck_put_object_name(options, &entry.oid, "%s%s/",
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
name, entry.path);
result = options->walk(obj, OBJ_TREE, data, options);
}
else if (S_ISREG(entry.mode) || S_ISLNK(entry.mode)) {
obj = (struct object *)lookup_blob(the_repository, &entry.oid);
if (name && obj)
fsck_put_object_name(options, &entry.oid, "%s%s",
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
name, entry.path);
result = options->walk(obj, OBJ_BLOB, data, options);
}
else {
result = error("in tree %s: entry %s has bad mode %.6o",
fsck_describe_object(options, &tree->object.oid),
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
entry.path, entry.mode);
}
if (result < 0)
return result;
if (!res)
res = result;
}
return res;
}
static int fsck_walk_commit(struct commit *commit, void *data, struct fsck_options *options)
{
int counter = 0, generation = 0, name_prefix_len = 0;
struct commit_list *parents;
int res;
int result;
const char *name;
if (repo_parse_commit(the_repository, commit))
return -1;
name = fsck_get_object_name(options, &commit->object.oid);
if (name)
fsck_put_object_name(options, get_commit_tree_oid(commit),
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
"%s:", name);
result = options->walk((struct object *) repo_get_commit_tree(the_repository, commit),
OBJ_TREE, data, options);
if (result < 0)
return result;
res = result;
parents = commit->parents;
if (name && parents) {
int len = strlen(name), power;
if (len && name[len - 1] == '^') {
generation = 1;
name_prefix_len = len - 1;
}
else { /* parse ~<generation> suffix */
for (generation = 0, power = 1;
len && isdigit(name[len - 1]);
power *= 10)
generation += power * (name[--len] - '0');
if (power > 1 && len && name[len - 1] == '~')
name_prefix_len = len - 1;
else {
/* Maybe a non-first parent, e.g. HEAD^2 */
generation = 0;
name_prefix_len = len;
}
}
}
while (parents) {
if (name) {
struct object_id *oid = &parents->item->object.oid;
if (counter++)
fsck_put_object_name(options, oid, "%s^%d",
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
name, counter);
else if (generation > 0)
fsck_put_object_name(options, oid, "%.*s~%d",
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
name_prefix_len, name,
generation + 1);
else
fsck_put_object_name(options, oid, "%s^", name);
}
result = options->walk((struct object *)parents->item, OBJ_COMMIT, data, options);
if (result < 0)
return result;
if (!res)
res = result;
parents = parents->next;
}
return res;
}
static int fsck_walk_tag(struct tag *tag, void *data, struct fsck_options *options)
{
const char *name = fsck_get_object_name(options, &tag->object.oid);
if (parse_tag(tag))
return -1;
if (name)
fsck_put_object_name(options, &tag->tagged->oid, "%s", name);
return options->walk(tag->tagged, OBJ_ANY, data, options);
}
int fsck_walk(struct object *obj, void *data, struct fsck_options *options)
{
if (!obj)
return -1;
fsck: lazily load types under --connectivity-only The recent fixes to "fsck --connectivity-only" load all of the objects with their correct types. This keeps the connectivity-only code path close to the regular one, but it also introduces some unnecessary inefficiency. While getting the type of an object is cheap compared to actually opening and parsing the object (as the non-connectivity-only case would do), it's still not free. For reachable non-blob objects, we end up having to parse them later anyway (to see what they point to), making our type lookup here redundant. For unreachable objects, we might never hit them at all in the reachability traversal, making the lookup completely wasted. And in some cases, we might have quite a few unreachable objects (e.g., when alternates are used for shared object storage between repositories, it's normal for there to be objects reachable from other repositories but not the one running fsck). The comment in mark_object_for_connectivity() claims two benefits to getting the type up front: 1. We need to know the types during fsck_walk(). (And not explicitly mentioned, but we also need them when printing the types of broken or dangling commits). We can address this by lazy-loading the types as necessary. Most objects never need this lazy-load at all, because they fall into one of these categories: a. Reachable from our tips, and are coerced into the correct type as we traverse (e.g., a parent link will call lookup_commit(), which converts OBJ_NONE to OBJ_COMMIT). b. Unreachable, but not at the tip of a chunk of unreachable history. We only mention the tips as "dangling", so an unreachable commit which links to hundreds of other objects needs only report the type of the tip commit. 2. It serves as a cross-check that the coercion in (1a) is correct (i.e., we'll complain about a parent link that points to a blob). But we get most of this for free already, because right after coercing, we'll parse any non-blob objects. So we'd notice then if we expected a commit and got a blob. The one exception is when we expect a blob, in which case we never actually read the object contents. So this is a slight weakening, but given that the whole point of --connectivity-only is to sacrifice some data integrity checks for speed, this seems like an acceptable tradeoff. Here are before and after timings for an extreme case with ~5M reachable objects and another ~12M unreachable (it's the torvalds/linux repository on GitHub, connected to shared storage for all of the other kernel forks): [before] $ time git fsck --no-dangling --connectivity-only real 3m4.323s user 1m25.121s sys 1m38.710s [after] $ time git fsck --no-dangling --connectivity-only real 0m51.497s user 0m49.575s sys 0m1.776s Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-01-26 04:12:07 +00:00
if (obj->type == OBJ_NONE)
parse_object(the_repository, &obj->oid);
fsck: lazily load types under --connectivity-only The recent fixes to "fsck --connectivity-only" load all of the objects with their correct types. This keeps the connectivity-only code path close to the regular one, but it also introduces some unnecessary inefficiency. While getting the type of an object is cheap compared to actually opening and parsing the object (as the non-connectivity-only case would do), it's still not free. For reachable non-blob objects, we end up having to parse them later anyway (to see what they point to), making our type lookup here redundant. For unreachable objects, we might never hit them at all in the reachability traversal, making the lookup completely wasted. And in some cases, we might have quite a few unreachable objects (e.g., when alternates are used for shared object storage between repositories, it's normal for there to be objects reachable from other repositories but not the one running fsck). The comment in mark_object_for_connectivity() claims two benefits to getting the type up front: 1. We need to know the types during fsck_walk(). (And not explicitly mentioned, but we also need them when printing the types of broken or dangling commits). We can address this by lazy-loading the types as necessary. Most objects never need this lazy-load at all, because they fall into one of these categories: a. Reachable from our tips, and are coerced into the correct type as we traverse (e.g., a parent link will call lookup_commit(), which converts OBJ_NONE to OBJ_COMMIT). b. Unreachable, but not at the tip of a chunk of unreachable history. We only mention the tips as "dangling", so an unreachable commit which links to hundreds of other objects needs only report the type of the tip commit. 2. It serves as a cross-check that the coercion in (1a) is correct (i.e., we'll complain about a parent link that points to a blob). But we get most of this for free already, because right after coercing, we'll parse any non-blob objects. So we'd notice then if we expected a commit and got a blob. The one exception is when we expect a blob, in which case we never actually read the object contents. So this is a slight weakening, but given that the whole point of --connectivity-only is to sacrifice some data integrity checks for speed, this seems like an acceptable tradeoff. Here are before and after timings for an extreme case with ~5M reachable objects and another ~12M unreachable (it's the torvalds/linux repository on GitHub, connected to shared storage for all of the other kernel forks): [before] $ time git fsck --no-dangling --connectivity-only real 3m4.323s user 1m25.121s sys 1m38.710s [after] $ time git fsck --no-dangling --connectivity-only real 0m51.497s user 0m49.575s sys 0m1.776s Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-01-26 04:12:07 +00:00
switch (obj->type) {
case OBJ_BLOB:
return 0;
case OBJ_TREE:
return fsck_walk_tree((struct tree *)obj, data, options);
case OBJ_COMMIT:
return fsck_walk_commit((struct commit *)obj, data, options);
case OBJ_TAG:
return fsck_walk_tag((struct tag *)obj, data, options);
default:
fsck: unify object-name code Commit 90cf590f53 (fsck: optionally show more helpful info for broken links, 2016-07-17) added a system for decorating objects with names. The code is split across builtin/fsck.c (which gives the initial names) and fsck.c (which adds to the names as it traverses the object graph). This leads to some duplication, where both sites have near-identical describe_object() functions (the difference being that the one in builtin/fsck.c uses a circular array of buffers to allow multiple calls in a single printf). Let's provide a unified object_name API for fsck. That lets us drop the duplication, as well as making the interface boundaries more clear (which will let us refactor the implementation more in a future patch). We'll leave describe_object() in builtin/fsck.c as a thin wrapper around the new API, as it relies on a static global to make its many callers a bit shorter. We'll also convert the bare add_decoration() calls in builtin/fsck.c to put_object_name(). This fixes two minor bugs: 1. We leak many small strings. add_decoration() has a last-one-wins approach: it updates the decoration to the new string and returns the old one. But we ignore the return value, leaking the old string. This is quite common to trigger, since we look at reflogs: the tip of any ref will be described both by looking at the actual ref, as well as the latest reflog entry. So we'd always end up leaking one of those strings. 2. The last-one-wins approach gives us lousy names. For instance, we first look at all of the refs, and then all of the reflogs. So rather than seeing "refs/heads/master", we're likely to overwrite it with "HEAD@{12345678}". We're generally better off using the first name we find. And indeed, the test in t1450 expects this ugly HEAD@{} name. After this patch, we've switched to using fsck_put_object_name()'s first-one-wins semantics, and we output the more human-friendly "refs/tags/julius" (and the test is updated accordingly). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:56:13 +00:00
error("Unknown object type for %s",
fsck_describe_object(options, &obj->oid));
return -1;
}
}
struct name_stack {
const char **names;
size_t nr, alloc;
};
static void name_stack_push(struct name_stack *stack, const char *name)
{
ALLOC_GROW(stack->names, stack->nr + 1, stack->alloc);
stack->names[stack->nr++] = name;
}
static const char *name_stack_pop(struct name_stack *stack)
{
return stack->nr ? stack->names[--stack->nr] : NULL;
}
static void name_stack_clear(struct name_stack *stack)
{
FREE_AND_NULL(stack->names);
stack->nr = stack->alloc = 0;
}
/*
* The entries in a tree are ordered in the _path_ order,
* which means that a directory entry is ordered by adding
* a slash to the end of it.
*
* So a directory called "a" is ordered _after_ a file
* called "a.c", because "a/" sorts after "a.c".
*/
#define TREE_UNORDERED (-1)
#define TREE_HAS_DUPS (-2)
static int is_less_than_slash(unsigned char c)
{
return '\0' < c && c < '/';
}
static int verify_ordered(unsigned mode1, const char *name1,
unsigned mode2, const char *name2,
struct name_stack *candidates)
{
int len1 = strlen(name1);
int len2 = strlen(name2);
int len = len1 < len2 ? len1 : len2;
unsigned char c1, c2;
int cmp;
cmp = memcmp(name1, name2, len);
if (cmp < 0)
return 0;
if (cmp > 0)
return TREE_UNORDERED;
/*
* Ok, the first <len> characters are the same.
* Now we need to order the next one, but turn
* a '\0' into a '/' for a directory entry.
*/
c1 = name1[len];
c2 = name2[len];
if (!c1 && !c2)
/*
* git-write-tree used to write out a nonsense tree that has
* entries with the same name, one blob and one tree. Make
* sure we do not have duplicate entries.
*/
return TREE_HAS_DUPS;
if (!c1 && S_ISDIR(mode1))
c1 = '/';
if (!c2 && S_ISDIR(mode2))
c2 = '/';
/*
* There can be non-consecutive duplicates due to the implicitly
* added slash, e.g.:
*
* foo
* foo.bar
* foo.bar.baz
* foo.bar/
* foo/
*
* Record non-directory candidates (like "foo" and "foo.bar" in
* the example) on a stack and check directory candidates (like
* foo/" and "foo.bar/") against that stack.
*/
if (!c1 && is_less_than_slash(c2)) {
name_stack_push(candidates, name1);
} else if (c2 == '/' && is_less_than_slash(c1)) {
for (;;) {
const char *p;
const char *f_name = name_stack_pop(candidates);
if (!f_name)
break;
if (!skip_prefix(name2, f_name, &p))
continue;
if (!*p)
return TREE_HAS_DUPS;
if (is_less_than_slash(*p)) {
name_stack_push(candidates, f_name);
break;
}
}
}
return c1 < c2 ? 0 : TREE_UNORDERED;
}
static int fsck_tree(const struct object_id *tree_oid,
fsck: require an actual buffer for non-blobs The fsck_object() function takes in a buffer, but also a "struct object". The rules for using these vary between types: - for a commit, we'll use the provided buffer; if it's NULL, we'll fall back to get_commit_buffer(), which loads from either an in-memory cache or from disk. If the latter fails, we'd die(), which is non-ideal for fsck. - for a tag, a NULL buffer will fall back to loading the object from disk (and failure would lead to an fsck error) - for a tree, we _never_ look at the provided buffer, and always use tree->buffer - for a blob, we usually don't look at the buffer at all, unless it has been marked as a .gitmodule file. In that case we check the buffer given to us, or assume a NULL buffer is a very large blob (and complain about it) This is much more complex than it needs to be. It turns out that nobody ever feeds a NULL buffer that isn't a blob: - git-fsck calls fsck_object() only from fsck_obj(). That in turn is called by one of: - fsck_obj_buffer(), which is a callback to verify_pack(), which unpacks everything except large blobs into a buffer (see pack-check.c, lines 131-141). - fsck_loose(), which hits a BUG() on non-blobs with a NULL buffer (builtin/fsck.c, lines 639-640) And in either case, we'll have just called parse_object_buffer() anyway, which would segfault on a NULL buffer for commits or tags (not for trees, but it would install a NULL tree->buffer which would later cause a segfault) - git-index-pack asserts that the buffer is non-NULL unless the object is a blob (see builtin/index-pack.c, line 832) - git-unpack-objects always writes a non-NULL buffer into its obj_buffer hash, which is then fed to fsck_object(). (There is actually a funny thing here where it does not store blob buffers at all, nor does it call fsck on them; it does check any needed blobs via fsck_finish() though). Let's make the rules simpler, which reduces the amount of code and gives us more flexibility in refactoring the fsck code. The new rules are: - only blobs are allowed to pass a NULL buffer - we always use the provided buffer, never pulling information from the object struct We don't have to adjust any callers, because they were already adhering to these. Note that we do drop a few fsck identifiers for missing tags, but that was all dead code (because nobody passed a NULL tag buffer). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:54:12 +00:00
const char *buffer, unsigned long size,
struct fsck_options *options)
{
int retval = 0;
int has_null_sha1 = 0;
int has_full_path = 0;
int has_empty_name = 0;
int has_dot = 0;
int has_dotdot = 0;
int has_dotgit = 0;
int has_zero_pad = 0;
int has_bad_modes = 0;
int has_dup_entries = 0;
int not_properly_sorted = 0;
struct tree_desc desc;
unsigned o_mode;
const char *o_name;
struct name_stack df_dup_candidates = { NULL };
if (init_tree_desc_gently(&desc, buffer, size, TREE_DESC_RAW_MODES)) {
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_BAD_TREE,
"cannot be parsed as a tree");
return retval;
}
o_mode = 0;
o_name = NULL;
while (desc.size) {
unsigned short mode;
is_ntfs_dotgit(): only verify the leading segment The config setting `core.protectNTFS` is specifically designed to work not only on Windows, but anywhere, to allow for repositories hosted on, say, Linux servers to be protected against NTFS-specific attack vectors. As a consequence, `is_ntfs_dotgit()` manually splits backslash-separated paths (but does not do the same for paths separated by forward slashes), under the assumption that the backslash might not be a valid directory separator on the _current_ Operating System. However, the two callers, `verify_path()` and `fsck_tree()`, are supposed to feed only individual path segments to the `is_ntfs_dotgit()` function. This causes a lot of duplicate scanning (and very inefficient scanning, too, as the inner loop of `is_ntfs_dotgit()` was optimized for readability rather than for speed. Let's simplify the design of `is_ntfs_dotgit()` by putting the burden of splitting the paths by backslashes as directory separators on the callers of said function. Consequently, the `verify_path()` function, which already splits the path by directory separators, now treats backslashes as directory separators _explicitly_ when `core.protectNTFS` is turned on, even on platforms where the backslash is _not_ a directory separator. Note that we have to repeat some code in `verify_path()`: if the backslash is not a directory separator on the current Operating System, we want to allow file names like `\`, but we _do_ want to disallow paths that are clearly intended to cause harm when the repository is cloned on Windows. The `fsck_tree()` function (the other caller of `is_ntfs_dotgit()`) now needs to look for backslashes in tree entries' names specifically when `core.protectNTFS` is turned on. While it would be tempting to completely disallow backslashes in that case (much like `fsck` reports names containing forward slashes as "full paths"), this would be overzealous: when `core.protectNTFS` is turned on in a non-Windows setup, backslashes are perfectly valid characters in file names while we _still_ want to disallow tree entries that are clearly designed to exploit NTFS-specific behavior. This simplification will make subsequent changes easier to implement, such as turning `core.protectNTFS` on by default (not only on Windows) or protecting against attack vectors involving NTFS Alternate Data Streams. Incidentally, this change allows for catching malicious repositories that contain tree entries of the form `dir\.gitmodules` already on the server side rather than only on the client side (and previously only on Windows): in contrast to `is_ntfs_dotgit()`, the `is_ntfs_dotgitmodules()` function already expects the caller to split the paths by directory separators. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
2019-09-23 06:58:11 +00:00
const char *name, *backslash;
const struct object_id *entry_oid;
entry_oid = tree_entry_extract(&desc, &name, &mode);
has_null_sha1 |= is_null_oid(entry_oid);
has_full_path |= !!strchr(name, '/');
has_empty_name |= !*name;
has_dot |= !strcmp(name, ".");
has_dotdot |= !strcmp(name, "..");
has_dotgit |= is_hfs_dotgit(name) || is_ntfs_dotgit(name);
has_zero_pad |= *(char *)desc.buffer == '0';
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
if (is_hfs_dotgitmodules(name) || is_ntfs_dotgitmodules(name)) {
if (!S_ISLNK(mode))
oidset_insert(&options->gitmodules_found,
entry_oid);
else
retval += report(options,
tree_oid, OBJ_TREE,
FSCK_MSG_GITMODULES_SYMLINK,
".gitmodules is a symbolic link");
}
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
if (is_hfs_dotgitattributes(name) || is_ntfs_dotgitattributes(name)) {
if (!S_ISLNK(mode))
oidset_insert(&options->gitattributes_found,
entry_oid);
else
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_GITATTRIBUTES_SYMLINK,
".gitattributes is a symlink");
}
fsck: warn about symlinked dotfiles we'll open with O_NOFOLLOW In the commits merged in via 204333b015 (Merge branch 'jk/open-dotgitx-with-nofollow', 2021-03-22), we stopped following symbolic links for .gitattributes, .gitignore, and .mailmap files. Let's teach fsck to warn that these symlinks are not going to do anything. Note that this is just a warning, and won't block the objects via transfer.fsckObjects, since there are reported to be cases of this in the wild (and even once fixed, they will continue to exist in the commit history of those projects, but are not particularly dangerous). Note that we won't add these to the existing gitmodules block in the fsck code. The logic for gitmodules is a bit more complicated, as we also check the content of non-symlink instances we find. But for these new files, there is no content check; we're just looking at the name and mode of the tree entry (and we can avoid even the complicated name checks in the common case that the mode doesn't indicate a symlink). We can reuse the test helper function we defined for .gitmodules, though (it needs some slight adjustments for the fsck error code, and because we don't block these symlinks via verify_path()). Note that I didn't explicitly test the transfer.fsckObjects case here (nor does the existing .gitmodules test that it blocks a push). The translation of fsck severities to outcomes is covered in general in t5504. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-05-03 20:43:25 +00:00
if (S_ISLNK(mode)) {
if (is_hfs_dotgitignore(name) ||
is_ntfs_dotgitignore(name))
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_GITIGNORE_SYMLINK,
".gitignore is a symlink");
if (is_hfs_dotmailmap(name) ||
is_ntfs_dotmailmap(name))
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_MAILMAP_SYMLINK,
".mailmap is a symlink");
}
is_ntfs_dotgit(): only verify the leading segment The config setting `core.protectNTFS` is specifically designed to work not only on Windows, but anywhere, to allow for repositories hosted on, say, Linux servers to be protected against NTFS-specific attack vectors. As a consequence, `is_ntfs_dotgit()` manually splits backslash-separated paths (but does not do the same for paths separated by forward slashes), under the assumption that the backslash might not be a valid directory separator on the _current_ Operating System. However, the two callers, `verify_path()` and `fsck_tree()`, are supposed to feed only individual path segments to the `is_ntfs_dotgit()` function. This causes a lot of duplicate scanning (and very inefficient scanning, too, as the inner loop of `is_ntfs_dotgit()` was optimized for readability rather than for speed. Let's simplify the design of `is_ntfs_dotgit()` by putting the burden of splitting the paths by backslashes as directory separators on the callers of said function. Consequently, the `verify_path()` function, which already splits the path by directory separators, now treats backslashes as directory separators _explicitly_ when `core.protectNTFS` is turned on, even on platforms where the backslash is _not_ a directory separator. Note that we have to repeat some code in `verify_path()`: if the backslash is not a directory separator on the current Operating System, we want to allow file names like `\`, but we _do_ want to disallow paths that are clearly intended to cause harm when the repository is cloned on Windows. The `fsck_tree()` function (the other caller of `is_ntfs_dotgit()`) now needs to look for backslashes in tree entries' names specifically when `core.protectNTFS` is turned on. While it would be tempting to completely disallow backslashes in that case (much like `fsck` reports names containing forward slashes as "full paths"), this would be overzealous: when `core.protectNTFS` is turned on in a non-Windows setup, backslashes are perfectly valid characters in file names while we _still_ want to disallow tree entries that are clearly designed to exploit NTFS-specific behavior. This simplification will make subsequent changes easier to implement, such as turning `core.protectNTFS` on by default (not only on Windows) or protecting against attack vectors involving NTFS Alternate Data Streams. Incidentally, this change allows for catching malicious repositories that contain tree entries of the form `dir\.gitmodules` already on the server side rather than only on the client side (and previously only on Windows): in contrast to `is_ntfs_dotgit()`, the `is_ntfs_dotgitmodules()` function already expects the caller to split the paths by directory separators. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
2019-09-23 06:58:11 +00:00
if ((backslash = strchr(name, '\\'))) {
while (backslash) {
backslash++;
has_dotgit |= is_ntfs_dotgit(backslash);
if (is_ntfs_dotgitmodules(backslash)) {
if (!S_ISLNK(mode))
oidset_insert(&options->gitmodules_found,
entry_oid);
else
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_GITMODULES_SYMLINK,
".gitmodules is a symbolic link");
}
is_ntfs_dotgit(): only verify the leading segment The config setting `core.protectNTFS` is specifically designed to work not only on Windows, but anywhere, to allow for repositories hosted on, say, Linux servers to be protected against NTFS-specific attack vectors. As a consequence, `is_ntfs_dotgit()` manually splits backslash-separated paths (but does not do the same for paths separated by forward slashes), under the assumption that the backslash might not be a valid directory separator on the _current_ Operating System. However, the two callers, `verify_path()` and `fsck_tree()`, are supposed to feed only individual path segments to the `is_ntfs_dotgit()` function. This causes a lot of duplicate scanning (and very inefficient scanning, too, as the inner loop of `is_ntfs_dotgit()` was optimized for readability rather than for speed. Let's simplify the design of `is_ntfs_dotgit()` by putting the burden of splitting the paths by backslashes as directory separators on the callers of said function. Consequently, the `verify_path()` function, which already splits the path by directory separators, now treats backslashes as directory separators _explicitly_ when `core.protectNTFS` is turned on, even on platforms where the backslash is _not_ a directory separator. Note that we have to repeat some code in `verify_path()`: if the backslash is not a directory separator on the current Operating System, we want to allow file names like `\`, but we _do_ want to disallow paths that are clearly intended to cause harm when the repository is cloned on Windows. The `fsck_tree()` function (the other caller of `is_ntfs_dotgit()`) now needs to look for backslashes in tree entries' names specifically when `core.protectNTFS` is turned on. While it would be tempting to completely disallow backslashes in that case (much like `fsck` reports names containing forward slashes as "full paths"), this would be overzealous: when `core.protectNTFS` is turned on in a non-Windows setup, backslashes are perfectly valid characters in file names while we _still_ want to disallow tree entries that are clearly designed to exploit NTFS-specific behavior. This simplification will make subsequent changes easier to implement, such as turning `core.protectNTFS` on by default (not only on Windows) or protecting against attack vectors involving NTFS Alternate Data Streams. Incidentally, this change allows for catching malicious repositories that contain tree entries of the form `dir\.gitmodules` already on the server side rather than only on the client side (and previously only on Windows): in contrast to `is_ntfs_dotgit()`, the `is_ntfs_dotgitmodules()` function already expects the caller to split the paths by directory separators. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
2019-09-23 06:58:11 +00:00
backslash = strchr(backslash, '\\');
}
}
if (update_tree_entry_gently(&desc)) {
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_BAD_TREE,
"cannot be parsed as a tree");
break;
}
switch (mode) {
/*
* Standard modes..
*/
case S_IFREG | 0755:
case S_IFREG | 0644:
case S_IFLNK:
case S_IFDIR:
case S_IFGITLINK:
break;
/*
* This is nonstandard, but we had a few of these
* early on when we honored the full set of mode
* bits..
*/
case S_IFREG | 0664:
if (!options->strict)
break;
consistently use "fallthrough" comments in switches Gcc 7 adds -Wimplicit-fallthrough, which can warn when a switch case falls through to the next case. The general idea is that the compiler can't tell if this was intentional or not, so you should annotate any intentional fall-throughs as such, leaving it to complain about any unannotated ones. There's a GNU __attribute__ which can be used for annotation, but of course we'd have to #ifdef it away on non-gcc compilers. Gcc will also recognize specially-formatted comments, which matches our current practice. Let's extend that practice to all of the unannotated sites (which I did look over and verify that they were behaving as intended). Ideally in each case we'd actually give some reasons in the comment about why we're falling through, or what we're falling through to. And gcc does support that with -Wimplicit-fallthrough=2, which relaxes the comment pattern matching to anything that contains "fallthrough" (or a variety of spelling variants). However, this isn't the default for -Wimplicit-fallthrough, nor for -Wextra. In the name of simplicity, it's probably better for us to support the default level, which requires "fallthrough" to be the only thing in the comment (modulo some window dressing like "else" and some punctuation; see the gcc manual for the complete set of patterns). This patch suppresses all warnings due to -Wimplicit-fallthrough. We might eventually want to add that to the DEVELOPER Makefile knob, but we should probably wait until gcc 7 is more widely adopted (since earlier versions will complain about the unknown warning type). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-21 06:25:41 +00:00
/* fallthrough */
default:
has_bad_modes = 1;
}
if (o_name) {
switch (verify_ordered(o_mode, o_name, mode, name,
&df_dup_candidates)) {
case TREE_UNORDERED:
not_properly_sorted = 1;
break;
case TREE_HAS_DUPS:
has_dup_entries = 1;
break;
default:
break;
}
}
o_mode = mode;
o_name = name;
}
name_stack_clear(&df_dup_candidates);
if (has_null_sha1)
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_NULL_SHA1,
"contains entries pointing to null sha1");
if (has_full_path)
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_FULL_PATHNAME,
"contains full pathnames");
if (has_empty_name)
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_EMPTY_NAME,
"contains empty pathname");
if (has_dot)
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_HAS_DOT,
"contains '.'");
if (has_dotdot)
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_HAS_DOTDOT,
"contains '..'");
if (has_dotgit)
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_HAS_DOTGIT,
"contains '.git'");
if (has_zero_pad)
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_ZERO_PADDED_FILEMODE,
"contains zero-padded file modes");
if (has_bad_modes)
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_BAD_FILEMODE,
"contains bad file modes");
if (has_dup_entries)
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_DUPLICATE_ENTRIES,
"contains duplicate file entries");
if (not_properly_sorted)
retval += report(options, tree_oid, OBJ_TREE,
FSCK_MSG_TREE_NOT_SORTED,
"not properly sorted");
return retval;
}
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
/*
* Confirm that the headers of a commit or tag object end in a reasonable way,
* either with the usual "\n\n" separator, or at least with a trailing newline
* on the final header line.
*
* This property is important for the memory safety of our callers. It allows
* them to scan the buffer linewise without constantly checking the remaining
* size as long as:
*
* - they check that there are bytes left in the buffer at the start of any
* line (i.e., that the last newline they saw was not the final one we
* found here)
*
* - any intra-line scanning they do will stop at a newline, which will worst
* case hit the newline we found here as the end-of-header. This makes it
* OK for them to use helpers like parse_oid_hex(), or even skip_prefix().
*/
fsck: it is OK for a tag and a commit to lack the body When fsck validates a commit or a tag, it scans each line in the header of the object using helper functions such as "start_with()", etc. that work on a NUL terminated buffer, but before a1e920a0 (index-pack: terminate object buffers with NUL, 2014-12-08), the validation functions were fed the object data in a piece of memory that is not necessarily terminated with a NUL. We added a helper function require_end_of_header() to be called at the beginning of these validation functions to insist that the object data contains an empty line before its end. The theory is that the validating functions will notice and stop when it hits an empty line as a normal end of header (or a required header line that is missing) without scanning past the end of potentially not NUL-terminated buffer. But the theory forgot that in the older days, Git itself happily created objects with only the header lines without a body. This caused Git 2.2 and later to issue an unnecessary warning in some existing repositories. With a1e920a0, we do not need to require an empty line (or the body) in these objects to safely parse and validate them. Drop the offending "must have an empty line" check from this helper function, while keeping the other check to make sure that there is no NUL in the header part of the object, and adjust the name of the helper to what it does accordingly. Noticed-by: Wolfgang Denk <wd@denx.de> Helped-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-06-28 18:18:31 +00:00
static int verify_headers(const void *data, unsigned long size,
const struct object_id *oid, enum object_type type,
struct fsck_options *options)
{
const char *buffer = (const char *)data;
unsigned long i;
for (i = 0; i < size; i++) {
switch (buffer[i]) {
case '\0':
return report(options, oid, type,
FSCK_MSG_NUL_IN_HEADER,
"unterminated header: NUL at offset %ld", i);
case '\n':
if (i + 1 < size && buffer[i + 1] == '\n')
return 0;
}
}
fsck: it is OK for a tag and a commit to lack the body When fsck validates a commit or a tag, it scans each line in the header of the object using helper functions such as "start_with()", etc. that work on a NUL terminated buffer, but before a1e920a0 (index-pack: terminate object buffers with NUL, 2014-12-08), the validation functions were fed the object data in a piece of memory that is not necessarily terminated with a NUL. We added a helper function require_end_of_header() to be called at the beginning of these validation functions to insist that the object data contains an empty line before its end. The theory is that the validating functions will notice and stop when it hits an empty line as a normal end of header (or a required header line that is missing) without scanning past the end of potentially not NUL-terminated buffer. But the theory forgot that in the older days, Git itself happily created objects with only the header lines without a body. This caused Git 2.2 and later to issue an unnecessary warning in some existing repositories. With a1e920a0, we do not need to require an empty line (or the body) in these objects to safely parse and validate them. Drop the offending "must have an empty line" check from this helper function, while keeping the other check to make sure that there is no NUL in the header part of the object, and adjust the name of the helper to what it does accordingly. Noticed-by: Wolfgang Denk <wd@denx.de> Helped-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-06-28 18:18:31 +00:00
/*
* We did not find double-LF that separates the header
* and the body. Not having a body is not a crime but
* we do want to see the terminating LF for the last header
* line.
*/
if (size && buffer[size - 1] == '\n')
return 0;
return report(options, oid, type,
FSCK_MSG_UNTERMINATED_HEADER, "unterminated header");
}
static int fsck_ident(const char **ident,
const struct object_id *oid, enum object_type type,
struct fsck_options *options)
{
const char *p = *ident;
char *end;
*ident = strchrnul(*ident, '\n');
if (**ident == '\n')
(*ident)++;
if (*p == '<')
return report(options, oid, type, FSCK_MSG_MISSING_NAME_BEFORE_EMAIL, "invalid author/committer line - missing space before email");
p += strcspn(p, "<>\n");
if (*p == '>')
return report(options, oid, type, FSCK_MSG_BAD_NAME, "invalid author/committer line - bad name");
if (*p != '<')
return report(options, oid, type, FSCK_MSG_MISSING_EMAIL, "invalid author/committer line - missing email");
if (p[-1] != ' ')
return report(options, oid, type, FSCK_MSG_MISSING_SPACE_BEFORE_EMAIL, "invalid author/committer line - missing space before email");
p++;
p += strcspn(p, "<>\n");
if (*p != '>')
return report(options, oid, type, FSCK_MSG_BAD_EMAIL, "invalid author/committer line - bad email");
p++;
if (*p != ' ')
return report(options, oid, type, FSCK_MSG_MISSING_SPACE_BEFORE_DATE, "invalid author/committer line - missing space before date");
p++;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
/*
* Our timestamp parser is based on the C strto*() functions, which
* will happily eat whitespace, including the newline that is supposed
* to prevent us walking past the end of the buffer. So do our own
* scan, skipping linear whitespace but not newlines, and then
* confirming we found a digit. We _could_ be even more strict here,
* as we really expect only a single space, but since we have
* traditionally allowed extra whitespace, we'll continue to do so.
*/
while (*p == ' ' || *p == '\t')
p++;
if (!isdigit(*p))
return report(options, oid, type, FSCK_MSG_BAD_DATE,
"invalid author/committer line - bad date");
if (*p == '0' && p[1] != ' ')
return report(options, oid, type, FSCK_MSG_ZERO_PADDED_DATE, "invalid author/committer line - zero-padded date");
if (date_overflows(parse_timestamp(p, &end, 10)))
return report(options, oid, type, FSCK_MSG_BAD_DATE_OVERFLOW, "invalid author/committer line - date causes integer overflow");
if ((end == p || *end != ' '))
return report(options, oid, type, FSCK_MSG_BAD_DATE, "invalid author/committer line - bad date");
p = end + 1;
if ((*p != '+' && *p != '-') ||
!isdigit(p[1]) ||
!isdigit(p[2]) ||
!isdigit(p[3]) ||
!isdigit(p[4]) ||
(p[5] != '\n'))
return report(options, oid, type, FSCK_MSG_BAD_TIMEZONE, "invalid author/committer line - bad time zone");
p += 6;
return 0;
}
static int fsck_commit(const struct object_id *oid,
const char *buffer, unsigned long size,
struct fsck_options *options)
{
struct object_id tree_oid, parent_oid;
unsigned author_count;
int err;
const char *buffer_begin = buffer;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
const char *buffer_end = buffer + size;
const char *p;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
/*
* We _must_ stop parsing immediately if this reports failure, as the
* memory safety of the rest of the function depends on it. See the
* comment above the definition of verify_headers() for more details.
*/
if (verify_headers(buffer, size, oid, OBJ_COMMIT, options))
return -1;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
if (buffer >= buffer_end || !skip_prefix(buffer, "tree ", &buffer))
return report(options, oid, OBJ_COMMIT, FSCK_MSG_MISSING_TREE, "invalid format - expected 'tree' line");
if (parse_oid_hex(buffer, &tree_oid, &p) || *p != '\n') {
err = report(options, oid, OBJ_COMMIT, FSCK_MSG_BAD_TREE_SHA1, "invalid 'tree' line format - bad sha1");
if (err)
return err;
}
buffer = p + 1;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
while (buffer < buffer_end && skip_prefix(buffer, "parent ", &buffer)) {
if (parse_oid_hex(buffer, &parent_oid, &p) || *p != '\n') {
err = report(options, oid, OBJ_COMMIT, FSCK_MSG_BAD_PARENT_SHA1, "invalid 'parent' line format - bad sha1");
if (err)
return err;
}
buffer = p + 1;
}
author_count = 0;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
while (buffer < buffer_end && skip_prefix(buffer, "author ", &buffer)) {
author_count++;
err = fsck_ident(&buffer, oid, OBJ_COMMIT, options);
if (err)
return err;
}
if (author_count < 1)
err = report(options, oid, OBJ_COMMIT, FSCK_MSG_MISSING_AUTHOR, "invalid format - expected 'author' line");
else if (author_count > 1)
err = report(options, oid, OBJ_COMMIT, FSCK_MSG_MULTIPLE_AUTHORS, "invalid format - multiple 'author' lines");
if (err)
return err;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
if (buffer >= buffer_end || !skip_prefix(buffer, "committer ", &buffer))
return report(options, oid, OBJ_COMMIT, FSCK_MSG_MISSING_COMMITTER, "invalid format - expected 'committer' line");
err = fsck_ident(&buffer, oid, OBJ_COMMIT, options);
if (err)
return err;
if (memchr(buffer_begin, '\0', size)) {
err = report(options, oid, OBJ_COMMIT, FSCK_MSG_NUL_IN_COMMIT,
"NUL byte in the commit object body");
if (err)
return err;
}
return 0;
}
static int fsck_tag(const struct object_id *oid, const char *buffer,
unsigned long size, struct fsck_options *options)
{
struct object_id tagged_oid;
mktag: use fsck instead of custom verify_tag() Change the validation logic in "mktag" to use fsck's fsck_tag() instead of its own custom parser. Curiously the logic for both dates back to the same commit[1]. Let's unify them so we're not maintaining two sets functions to verify that a tag is OK. The behavior of fsck_tag() and the old "mktag" code being removed here is different in few aspects. I think it makes sense to remove some of those checks, namely: A. fsck only cares that the timezone matches [-+][0-9]{4}. The mktag code disallowed values larger than 1400. Yes there's currently no timezone with a greater offset[2], but since we allow any number of non-offical timezones (e.g. +1234) passing this through seems fine. Git also won't break in the future if e.g. French Polynesia decides it needs to outdo the Line Islands when it comes to timezone extravagance. B. fsck allows missing author names such as "tagger <email>", mktag wouldn't, but would allow e.g. "tagger [2 spaces] <email>" (but not "tagger [1 space] <email>"). Now we allow all of these. C. Like B, but "mktag" disallowed spaces in the <email> part, fsck allows it. In some ways fsck_tag() is stricter than "mktag" was, namely: D. fsck disallows zero-padded dates, but mktag didn't care. So e.g. the timestamp "0000000000 +0000" produces an error now. A test in "t1006-cat-file.sh" relied on this, it's been changed to use "hash-object" (without fsck) instead. There was one check I deemed worth keeping by porting it over to fsck_tag(): E. "mktag" did not allow any custom headers, and by extension (as an empty commit is allowed) also forbade an extra stray trailing newline after the headers it knew about. Add a new check in the "ignore" category to fsck and use it. This somewhat abuses the facility added in efaba7cc77f (fsck: optionally ignore specific fsck issues completely, 2015-06-22). This is somewhat of hack, but probably the least invasive change we can make here. The fsck command will shuffle these categories around, e.g. under --strict the "info" becomes a "warn" and "warn" becomes "error". Existing users of fsck's (and others, e.g. index-pack) --strict option rely on this. So we need to put something into a category that'll be ignored by all existing users of the API. Pretending that fsck.extraHeaderEntry=error ("ignore" by default) was set serves to do this for us. 1. ec4465adb38 (Add "tag" objects that can be used to sign other objects., 2005-04-25) 2. https://en.wikipedia.org/wiki/List_of_UTC_time_offsets Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-05 19:42:46 +00:00
int tagged_type;
return fsck_tag_standalone(oid, buffer, size, options, &tagged_oid,
&tagged_type);
}
int fsck_tag_standalone(const struct object_id *oid, const char *buffer,
unsigned long size, struct fsck_options *options,
struct object_id *tagged_oid,
int *tagged_type)
{
int ret = 0;
fsck: require an actual buffer for non-blobs The fsck_object() function takes in a buffer, but also a "struct object". The rules for using these vary between types: - for a commit, we'll use the provided buffer; if it's NULL, we'll fall back to get_commit_buffer(), which loads from either an in-memory cache or from disk. If the latter fails, we'd die(), which is non-ideal for fsck. - for a tag, a NULL buffer will fall back to loading the object from disk (and failure would lead to an fsck error) - for a tree, we _never_ look at the provided buffer, and always use tree->buffer - for a blob, we usually don't look at the buffer at all, unless it has been marked as a .gitmodule file. In that case we check the buffer given to us, or assume a NULL buffer is a very large blob (and complain about it) This is much more complex than it needs to be. It turns out that nobody ever feeds a NULL buffer that isn't a blob: - git-fsck calls fsck_object() only from fsck_obj(). That in turn is called by one of: - fsck_obj_buffer(), which is a callback to verify_pack(), which unpacks everything except large blobs into a buffer (see pack-check.c, lines 131-141). - fsck_loose(), which hits a BUG() on non-blobs with a NULL buffer (builtin/fsck.c, lines 639-640) And in either case, we'll have just called parse_object_buffer() anyway, which would segfault on a NULL buffer for commits or tags (not for trees, but it would install a NULL tree->buffer which would later cause a segfault) - git-index-pack asserts that the buffer is non-NULL unless the object is a blob (see builtin/index-pack.c, line 832) - git-unpack-objects always writes a non-NULL buffer into its obj_buffer hash, which is then fed to fsck_object(). (There is actually a funny thing here where it does not store blob buffers at all, nor does it call fsck on them; it does check any needed blobs via fsck_finish() though). Let's make the rules simpler, which reduces the amount of code and gives us more flexibility in refactoring the fsck code. The new rules are: - only blobs are allowed to pass a NULL buffer - we always use the provided buffer, never pulling information from the object struct We don't have to adjust any callers, because they were already adhering to these. Note that we do drop a few fsck identifiers for missing tags, but that was all dead code (because nobody passed a NULL tag buffer). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-18 04:54:12 +00:00
char *eol;
struct strbuf sb = STRBUF_INIT;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
const char *buffer_end = buffer + size;
const char *p;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
/*
* We _must_ stop parsing immediately if this reports failure, as the
* memory safety of the rest of the function depends on it. See the
* comment above the definition of verify_headers() for more details.
*/
ret = verify_headers(buffer, size, oid, OBJ_TAG, options);
if (ret)
goto done;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
if (buffer >= buffer_end || !skip_prefix(buffer, "object ", &buffer)) {
ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_OBJECT, "invalid format - expected 'object' line");
goto done;
}
mktag: use fsck instead of custom verify_tag() Change the validation logic in "mktag" to use fsck's fsck_tag() instead of its own custom parser. Curiously the logic for both dates back to the same commit[1]. Let's unify them so we're not maintaining two sets functions to verify that a tag is OK. The behavior of fsck_tag() and the old "mktag" code being removed here is different in few aspects. I think it makes sense to remove some of those checks, namely: A. fsck only cares that the timezone matches [-+][0-9]{4}. The mktag code disallowed values larger than 1400. Yes there's currently no timezone with a greater offset[2], but since we allow any number of non-offical timezones (e.g. +1234) passing this through seems fine. Git also won't break in the future if e.g. French Polynesia decides it needs to outdo the Line Islands when it comes to timezone extravagance. B. fsck allows missing author names such as "tagger <email>", mktag wouldn't, but would allow e.g. "tagger [2 spaces] <email>" (but not "tagger [1 space] <email>"). Now we allow all of these. C. Like B, but "mktag" disallowed spaces in the <email> part, fsck allows it. In some ways fsck_tag() is stricter than "mktag" was, namely: D. fsck disallows zero-padded dates, but mktag didn't care. So e.g. the timestamp "0000000000 +0000" produces an error now. A test in "t1006-cat-file.sh" relied on this, it's been changed to use "hash-object" (without fsck) instead. There was one check I deemed worth keeping by porting it over to fsck_tag(): E. "mktag" did not allow any custom headers, and by extension (as an empty commit is allowed) also forbade an extra stray trailing newline after the headers it knew about. Add a new check in the "ignore" category to fsck and use it. This somewhat abuses the facility added in efaba7cc77f (fsck: optionally ignore specific fsck issues completely, 2015-06-22). This is somewhat of hack, but probably the least invasive change we can make here. The fsck command will shuffle these categories around, e.g. under --strict the "info" becomes a "warn" and "warn" becomes "error". Existing users of fsck's (and others, e.g. index-pack) --strict option rely on this. So we need to put something into a category that'll be ignored by all existing users of the API. Pretending that fsck.extraHeaderEntry=error ("ignore" by default) was set serves to do this for us. 1. ec4465adb38 (Add "tag" objects that can be used to sign other objects., 2005-04-25) 2. https://en.wikipedia.org/wiki/List_of_UTC_time_offsets Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-05 19:42:46 +00:00
if (parse_oid_hex(buffer, tagged_oid, &p) || *p != '\n') {
ret = report(options, oid, OBJ_TAG, FSCK_MSG_BAD_OBJECT_SHA1, "invalid 'object' line format - bad sha1");
if (ret)
goto done;
}
buffer = p + 1;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
if (buffer >= buffer_end || !skip_prefix(buffer, "type ", &buffer)) {
ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TYPE_ENTRY, "invalid format - expected 'type' line");
goto done;
}
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
eol = memchr(buffer, '\n', buffer_end - buffer);
if (!eol) {
ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TYPE, "invalid format - unexpected end after 'type' line");
goto done;
}
mktag: use fsck instead of custom verify_tag() Change the validation logic in "mktag" to use fsck's fsck_tag() instead of its own custom parser. Curiously the logic for both dates back to the same commit[1]. Let's unify them so we're not maintaining two sets functions to verify that a tag is OK. The behavior of fsck_tag() and the old "mktag" code being removed here is different in few aspects. I think it makes sense to remove some of those checks, namely: A. fsck only cares that the timezone matches [-+][0-9]{4}. The mktag code disallowed values larger than 1400. Yes there's currently no timezone with a greater offset[2], but since we allow any number of non-offical timezones (e.g. +1234) passing this through seems fine. Git also won't break in the future if e.g. French Polynesia decides it needs to outdo the Line Islands when it comes to timezone extravagance. B. fsck allows missing author names such as "tagger <email>", mktag wouldn't, but would allow e.g. "tagger [2 spaces] <email>" (but not "tagger [1 space] <email>"). Now we allow all of these. C. Like B, but "mktag" disallowed spaces in the <email> part, fsck allows it. In some ways fsck_tag() is stricter than "mktag" was, namely: D. fsck disallows zero-padded dates, but mktag didn't care. So e.g. the timestamp "0000000000 +0000" produces an error now. A test in "t1006-cat-file.sh" relied on this, it's been changed to use "hash-object" (without fsck) instead. There was one check I deemed worth keeping by porting it over to fsck_tag(): E. "mktag" did not allow any custom headers, and by extension (as an empty commit is allowed) also forbade an extra stray trailing newline after the headers it knew about. Add a new check in the "ignore" category to fsck and use it. This somewhat abuses the facility added in efaba7cc77f (fsck: optionally ignore specific fsck issues completely, 2015-06-22). This is somewhat of hack, but probably the least invasive change we can make here. The fsck command will shuffle these categories around, e.g. under --strict the "info" becomes a "warn" and "warn" becomes "error". Existing users of fsck's (and others, e.g. index-pack) --strict option rely on this. So we need to put something into a category that'll be ignored by all existing users of the API. Pretending that fsck.extraHeaderEntry=error ("ignore" by default) was set serves to do this for us. 1. ec4465adb38 (Add "tag" objects that can be used to sign other objects., 2005-04-25) 2. https://en.wikipedia.org/wiki/List_of_UTC_time_offsets Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-05 19:42:46 +00:00
*tagged_type = type_from_string_gently(buffer, eol - buffer, 1);
if (*tagged_type < 0)
ret = report(options, oid, OBJ_TAG, FSCK_MSG_BAD_TYPE, "invalid 'type' value");
if (ret)
goto done;
buffer = eol + 1;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
if (buffer >= buffer_end || !skip_prefix(buffer, "tag ", &buffer)) {
ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TAG_ENTRY, "invalid format - expected 'tag' line");
goto done;
}
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
eol = memchr(buffer, '\n', buffer_end - buffer);
if (!eol) {
ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TAG, "invalid format - unexpected end after 'type' line");
goto done;
}
strbuf_addf(&sb, "refs/tags/%.*s", (int)(eol - buffer), buffer);
if (check_refname_format(sb.buf, 0)) {
ret = report(options, oid, OBJ_TAG,
FSCK_MSG_BAD_TAG_NAME,
"invalid 'tag' name: %.*s",
(int)(eol - buffer), buffer);
if (ret)
goto done;
}
buffer = eol + 1;
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
if (buffer >= buffer_end || !skip_prefix(buffer, "tagger ", &buffer)) {
/* early tags do not contain 'tagger' lines; warn only */
ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TAGGER_ENTRY, "invalid format - expected 'tagger' line");
if (ret)
goto done;
}
else
ret = fsck_ident(&buffer, oid, OBJ_TAG, options);
fsck: do not assume NUL-termination of buffers The fsck code operates on an object buffer represented as a pointer/len combination. However, the parsing of commits and tags is a little bit loose; we mostly scan left-to-right through the buffer, without checking whether we've gone past the length we were given. This has traditionally been OK because the buffers we feed to fsck always have an extra NUL after the end of the object content, which ends any left-to-right scan. That has always been true for objects we read from the odb, and we made it true for incoming index-pack/unpack-objects checks in a1e920a0a7 (index-pack: terminate object buffers with NUL, 2014-12-08). However, we recently added an exception: hash-object asks index_fd() to do fsck checks. That _may_ have an extra NUL (if we read from a pipe into a strbuf), but it might not (if we read the contents from the file). Nor can we just teach it to always add a NUL. We may mmap the on-disk file, which will not have any extra bytes (if it's a multiple of the page size). Not to mention that this is a rather subtle assumption for the fsck code to make. Instead, let's make sure that the fsck parsers don't ever look past the size of the buffer they've been given. This _almost_ works already, thanks to earlier work in 4d0d89755e (Make sure fsck_commit_buffer() does not run out of the buffer, 2014-09-11). The theory there is that we check up front whether we have the end of header double-newline separator. And then any left-to-right scanning we do is OK as long as it stops when it hits that boundary. However, we later softened that in 84d18c0bcf (fsck: it is OK for a tag and a commit to lack the body, 2015-06-28), which allows the double-newline header to be missing, but does require that the header ends in a newline. That was OK back then, because of the NUL-termination guarantees (including the one from a1e920a0a7 mentioned above). Because 84d18c0bcf guarantees that any header line does end in a newline, we are still OK with most of the left-to-right scanning. We only need to take care after completing a line, to check that there is another line (and we didn't run out of buffer). Most of these checks are just need to check "buffer < buffer_end" (where buffer is advanced as we parse) before scanning for the next header line. But here are a few notes: - we don't technically need to check for remaining buffer before parsing the very first line ("tree" for a commit, or "object" for a tag), because verify_headers() rejects a totally empty buffer. But we'll do so in the name of consistency and defensiveness. - there are some calls to strchr('\n'). These are actually OK by the "the final header line must end in a newline" guarantee from verify_headers(). They will always find that rather than run off the end of the buffer. Curiously, they do check for a NULL return and complain, but I believe that condition can never be reached. However, I converted them to use memchr() with a proper size and retained the NULL checks. Using memchr() is not much longer and makes it more obvious what is going on. Likewise, retaining the NULL checks serves as a defensive measure in case my analysis is wrong. - commit 9a1a3a4d4c (mktag: allow omitting the header/body \n separator, 2021-01-05), does check for the end-of-buffer condition, but does so with "!*buffer", relying explicitly on the NUL termination. We can accomplish the same thing with a pointer comparison. I also folded it into the follow-on conditional that checks the contents of the buffer, for consistency with the other checks. - fsck_ident() uses parse_timestamp(), which is based on strtoumax(). That function will happily skip past leading whitespace, including newlines, which makes it a risk. We can fix this by scanning to the first digit ourselves, and then using parse_timestamp() to do the actual numeric conversion. Note that as a side effect this fixes the fact that we missed zero-padded timestamps like "<email> 0123" (whereas we would complain about "<email> 0123"). I doubt anybody cares, but I mention it here for completeness. - fsck_tree() does not need any modifications. It relies on decode_tree_entry() to do the actual parsing, and that function checks both that there are enough bytes in the buffer to represent an entry, and that there is a NUL at the appropriate spot (one hash-length from the end; this may not be the NUL for the entry we are parsing, but we know that in the worst case, everything from our current position to that NUL is a filename, so we won't run out of bytes). In addition to fixing the code itself, we'd like to make sure our rather subtle assumptions are not violated in the future. So this patch does two more things: - add comments around verify_headers() documenting the link between what it checks and the memory safety of the callers. I don't expect this code to be modified frequently, but this may help somebody from accidentally breaking things. - add a thorough set of tests covering truncations at various key spots (e.g., for a "tree $oid" line, in the middle of the word "tree", right after it, after the space, in the middle of the $oid, and right at the end of the line. Most of these are fine already (it is only truncating right at the end of the line that is currently broken). And some of them are not even possible with the current code (we parse "tree " as a unit, so truncating before the space is equivalent). But I aimed here to consider the code a black box and look for any truncations that would be a problem for a left-to-right parser. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-19 23:13:29 +00:00
if (buffer < buffer_end && !starts_with(buffer, "\n")) {
mktag: use fsck instead of custom verify_tag() Change the validation logic in "mktag" to use fsck's fsck_tag() instead of its own custom parser. Curiously the logic for both dates back to the same commit[1]. Let's unify them so we're not maintaining two sets functions to verify that a tag is OK. The behavior of fsck_tag() and the old "mktag" code being removed here is different in few aspects. I think it makes sense to remove some of those checks, namely: A. fsck only cares that the timezone matches [-+][0-9]{4}. The mktag code disallowed values larger than 1400. Yes there's currently no timezone with a greater offset[2], but since we allow any number of non-offical timezones (e.g. +1234) passing this through seems fine. Git also won't break in the future if e.g. French Polynesia decides it needs to outdo the Line Islands when it comes to timezone extravagance. B. fsck allows missing author names such as "tagger <email>", mktag wouldn't, but would allow e.g. "tagger [2 spaces] <email>" (but not "tagger [1 space] <email>"). Now we allow all of these. C. Like B, but "mktag" disallowed spaces in the <email> part, fsck allows it. In some ways fsck_tag() is stricter than "mktag" was, namely: D. fsck disallows zero-padded dates, but mktag didn't care. So e.g. the timestamp "0000000000 +0000" produces an error now. A test in "t1006-cat-file.sh" relied on this, it's been changed to use "hash-object" (without fsck) instead. There was one check I deemed worth keeping by porting it over to fsck_tag(): E. "mktag" did not allow any custom headers, and by extension (as an empty commit is allowed) also forbade an extra stray trailing newline after the headers it knew about. Add a new check in the "ignore" category to fsck and use it. This somewhat abuses the facility added in efaba7cc77f (fsck: optionally ignore specific fsck issues completely, 2015-06-22). This is somewhat of hack, but probably the least invasive change we can make here. The fsck command will shuffle these categories around, e.g. under --strict the "info" becomes a "warn" and "warn" becomes "error". Existing users of fsck's (and others, e.g. index-pack) --strict option rely on this. So we need to put something into a category that'll be ignored by all existing users of the API. Pretending that fsck.extraHeaderEntry=error ("ignore" by default) was set serves to do this for us. 1. ec4465adb38 (Add "tag" objects that can be used to sign other objects., 2005-04-25) 2. https://en.wikipedia.org/wiki/List_of_UTC_time_offsets Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-05 19:42:46 +00:00
/*
* The verify_headers() check will allow
* e.g. "[...]tagger <tagger>\nsome
* garbage\n\nmessage" to pass, thinking "some
* garbage" could be a custom header. E.g. "mktag"
* doesn't want any unknown headers.
*/
ret = report(options, oid, OBJ_TAG, FSCK_MSG_EXTRA_HEADER_ENTRY, "invalid format - extra header(s) after 'tagger'");
if (ret)
goto done;
}
done:
strbuf_release(&sb);
return ret;
}
dir API: add a generalized path_match_flags() function Add a path_match_flags() function and have the two sets of starts_with_dot_{,dot_}slash() functions added in 63e95beb085 (submodule: port resolve_relative_url from shell to C, 2016-04-15) and a2b26ffb1a8 (fsck: convert gitmodules url to URL passed to curl, 2020-04-18) be thin wrappers for it. As the latter of those notes the fsck version was copied from the initial builtin/submodule--helper.c version. Since the code added in a2b26ffb1a8 was doing really doing the same as win32_is_dir_sep() added in 1cadad6f658 (git clone <url> C:\cygwin\home\USER\repo' is working (again), 2018-12-15) let's move the latter to git-compat-util.h is a is_xplatform_dir_sep(). We can then call either it or the platform-specific is_dir_sep() from this new function. Let's likewise change code in various other places that was hardcoding checks for "'/' || '\\'" with the new is_xplatform_dir_sep(). As can be seen in those callers some of them still concern themselves with ':' (Mac OS classic?), but let's leave the question of whether that should be consolidated for some other time. As we expect to make wider use of the "native" case in the future, define and use two starts_with_dot_{,dot_}slash_native() convenience wrappers. This makes the diff in builtin/submodule--helper.c much smaller. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Derrick Stolee <derrickstolee@github.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-05-16 20:10:59 +00:00
static int starts_with_dot_slash(const char *const path)
{
dir API: add a generalized path_match_flags() function Add a path_match_flags() function and have the two sets of starts_with_dot_{,dot_}slash() functions added in 63e95beb085 (submodule: port resolve_relative_url from shell to C, 2016-04-15) and a2b26ffb1a8 (fsck: convert gitmodules url to URL passed to curl, 2020-04-18) be thin wrappers for it. As the latter of those notes the fsck version was copied from the initial builtin/submodule--helper.c version. Since the code added in a2b26ffb1a8 was doing really doing the same as win32_is_dir_sep() added in 1cadad6f658 (git clone <url> C:\cygwin\home\USER\repo' is working (again), 2018-12-15) let's move the latter to git-compat-util.h is a is_xplatform_dir_sep(). We can then call either it or the platform-specific is_dir_sep() from this new function. Let's likewise change code in various other places that was hardcoding checks for "'/' || '\\'" with the new is_xplatform_dir_sep(). As can be seen in those callers some of them still concern themselves with ':' (Mac OS classic?), but let's leave the question of whether that should be consolidated for some other time. As we expect to make wider use of the "native" case in the future, define and use two starts_with_dot_{,dot_}slash_native() convenience wrappers. This makes the diff in builtin/submodule--helper.c much smaller. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Derrick Stolee <derrickstolee@github.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-05-16 20:10:59 +00:00
return path_match_flags(path, PATH_MATCH_STARTS_WITH_DOT_SLASH |
PATH_MATCH_XPLATFORM);
}
dir API: add a generalized path_match_flags() function Add a path_match_flags() function and have the two sets of starts_with_dot_{,dot_}slash() functions added in 63e95beb085 (submodule: port resolve_relative_url from shell to C, 2016-04-15) and a2b26ffb1a8 (fsck: convert gitmodules url to URL passed to curl, 2020-04-18) be thin wrappers for it. As the latter of those notes the fsck version was copied from the initial builtin/submodule--helper.c version. Since the code added in a2b26ffb1a8 was doing really doing the same as win32_is_dir_sep() added in 1cadad6f658 (git clone <url> C:\cygwin\home\USER\repo' is working (again), 2018-12-15) let's move the latter to git-compat-util.h is a is_xplatform_dir_sep(). We can then call either it or the platform-specific is_dir_sep() from this new function. Let's likewise change code in various other places that was hardcoding checks for "'/' || '\\'" with the new is_xplatform_dir_sep(). As can be seen in those callers some of them still concern themselves with ':' (Mac OS classic?), but let's leave the question of whether that should be consolidated for some other time. As we expect to make wider use of the "native" case in the future, define and use two starts_with_dot_{,dot_}slash_native() convenience wrappers. This makes the diff in builtin/submodule--helper.c much smaller. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Derrick Stolee <derrickstolee@github.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-05-16 20:10:59 +00:00
static int starts_with_dot_dot_slash(const char *const path)
{
dir API: add a generalized path_match_flags() function Add a path_match_flags() function and have the two sets of starts_with_dot_{,dot_}slash() functions added in 63e95beb085 (submodule: port resolve_relative_url from shell to C, 2016-04-15) and a2b26ffb1a8 (fsck: convert gitmodules url to URL passed to curl, 2020-04-18) be thin wrappers for it. As the latter of those notes the fsck version was copied from the initial builtin/submodule--helper.c version. Since the code added in a2b26ffb1a8 was doing really doing the same as win32_is_dir_sep() added in 1cadad6f658 (git clone <url> C:\cygwin\home\USER\repo' is working (again), 2018-12-15) let's move the latter to git-compat-util.h is a is_xplatform_dir_sep(). We can then call either it or the platform-specific is_dir_sep() from this new function. Let's likewise change code in various other places that was hardcoding checks for "'/' || '\\'" with the new is_xplatform_dir_sep(). As can be seen in those callers some of them still concern themselves with ':' (Mac OS classic?), but let's leave the question of whether that should be consolidated for some other time. As we expect to make wider use of the "native" case in the future, define and use two starts_with_dot_{,dot_}slash_native() convenience wrappers. This makes the diff in builtin/submodule--helper.c much smaller. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Derrick Stolee <derrickstolee@github.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-05-16 20:10:59 +00:00
return path_match_flags(path, PATH_MATCH_STARTS_WITH_DOT_DOT_SLASH |
PATH_MATCH_XPLATFORM);
}
static int submodule_url_is_relative(const char *url)
{
return starts_with_dot_slash(url) || starts_with_dot_dot_slash(url);
}
credential: treat URL without scheme as invalid libcurl permits making requests without a URL scheme specified. In this case, it guesses the URL from the hostname, so I can run git ls-remote http::ftp.example.com/path/to/repo and it would make an FTP request. Any user intentionally using such a URL is likely to have made a typo. Unfortunately, credential_from_url is not able to determine the host and protocol in order to determine appropriate credentials to send, and until "credential: refuse to operate when missing host or protocol", this resulted in another host's credentials being leaked to the named host. Teach credential_from_url_gently to consider such a URL to be invalid so that fsck can detect and block gitmodules files with such URLs, allowing server operators to avoid serving them to downstream users running older versions of Git. This also means that when such URLs are passed on the command line, Git will print a clearer error so affected users can switch to the simpler URL that explicitly specifies the host and protocol they intend. One subtlety: .gitmodules files can contain relative URLs, representing a URL relative to the URL they were cloned from. The relative URL resolver used for .gitmodules can follow ".." components out of the path part and past the host part of a URL, meaning that such a relative URL can be used to traverse from a https://foo.example.com/innocent superproject to a https::attacker.example.com/exploit submodule. Fortunately a leading ':' in the first path component after a series of leading './' and '../' components is unlikely to show up in other contexts, so we can catch this by detecting that pattern. Reported-by: Jeff King <peff@peff.net> Signed-off-by: Jonathan Nieder <jrnieder@gmail.com> Reviewed-by: Jeff King <peff@peff.net>
2020-04-19 03:54:13 +00:00
/*
* Count directory components that a relative submodule URL should chop
* from the remote_url it is to be resolved against.
*
* In other words, this counts "../" components at the start of a
* submodule URL.
*
* Returns the number of directory components to chop and writes a
* pointer to the next character of url after all leading "./" and
* "../" components to out.
*/
static int count_leading_dotdots(const char *url, const char **out)
{
int result = 0;
while (1) {
if (starts_with_dot_dot_slash(url)) {
result++;
url += strlen("../");
continue;
}
if (starts_with_dot_slash(url)) {
url += strlen("./");
continue;
}
*out = url;
return result;
}
}
/*
* Check whether a transport is implemented by git-remote-curl.
*
* If it is, returns 1 and writes the URL that would be passed to
* git-remote-curl to the "out" parameter.
*
* Otherwise, returns 0 and leaves "out" untouched.
*
* Examples:
* http::https://example.com/repo.git -> 1, https://example.com/repo.git
* https://example.com/repo.git -> 1, https://example.com/repo.git
* git://example.com/repo.git -> 0
*
* This is for use in checking for previously exploitable bugs that
* required a submodule URL to be passed to git-remote-curl.
*/
static int url_to_curl_url(const char *url, const char **out)
{
/*
* We don't need to check for case-aliases, "http.exe", and so
* on because in the default configuration, is_transport_allowed
* prevents URLs with those schemes from being cloned
* automatically.
*/
if (skip_prefix(url, "http::", out) ||
skip_prefix(url, "https::", out) ||
skip_prefix(url, "ftp::", out) ||
skip_prefix(url, "ftps::", out))
return 1;
if (starts_with(url, "http://") ||
starts_with(url, "https://") ||
starts_with(url, "ftp://") ||
starts_with(url, "ftps://")) {
*out = url;
return 1;
}
return 0;
}
static int check_submodule_url(const char *url)
{
const char *curl_url;
if (looks_like_command_line_option(url))
return -1;
if (submodule_url_is_relative(url) || starts_with(url, "git://")) {
credential: treat URL without scheme as invalid libcurl permits making requests without a URL scheme specified. In this case, it guesses the URL from the hostname, so I can run git ls-remote http::ftp.example.com/path/to/repo and it would make an FTP request. Any user intentionally using such a URL is likely to have made a typo. Unfortunately, credential_from_url is not able to determine the host and protocol in order to determine appropriate credentials to send, and until "credential: refuse to operate when missing host or protocol", this resulted in another host's credentials being leaked to the named host. Teach credential_from_url_gently to consider such a URL to be invalid so that fsck can detect and block gitmodules files with such URLs, allowing server operators to avoid serving them to downstream users running older versions of Git. This also means that when such URLs are passed on the command line, Git will print a clearer error so affected users can switch to the simpler URL that explicitly specifies the host and protocol they intend. One subtlety: .gitmodules files can contain relative URLs, representing a URL relative to the URL they were cloned from. The relative URL resolver used for .gitmodules can follow ".." components out of the path part and past the host part of a URL, meaning that such a relative URL can be used to traverse from a https://foo.example.com/innocent superproject to a https::attacker.example.com/exploit submodule. Fortunately a leading ':' in the first path component after a series of leading './' and '../' components is unlikely to show up in other contexts, so we can catch this by detecting that pattern. Reported-by: Jeff King <peff@peff.net> Signed-off-by: Jonathan Nieder <jrnieder@gmail.com> Reviewed-by: Jeff King <peff@peff.net>
2020-04-19 03:54:13 +00:00
char *decoded;
const char *next;
int has_nl;
/*
* This could be appended to an http URL and url-decoded;
* check for malicious characters.
*/
credential: treat URL without scheme as invalid libcurl permits making requests without a URL scheme specified. In this case, it guesses the URL from the hostname, so I can run git ls-remote http::ftp.example.com/path/to/repo and it would make an FTP request. Any user intentionally using such a URL is likely to have made a typo. Unfortunately, credential_from_url is not able to determine the host and protocol in order to determine appropriate credentials to send, and until "credential: refuse to operate when missing host or protocol", this resulted in another host's credentials being leaked to the named host. Teach credential_from_url_gently to consider such a URL to be invalid so that fsck can detect and block gitmodules files with such URLs, allowing server operators to avoid serving them to downstream users running older versions of Git. This also means that when such URLs are passed on the command line, Git will print a clearer error so affected users can switch to the simpler URL that explicitly specifies the host and protocol they intend. One subtlety: .gitmodules files can contain relative URLs, representing a URL relative to the URL they were cloned from. The relative URL resolver used for .gitmodules can follow ".." components out of the path part and past the host part of a URL, meaning that such a relative URL can be used to traverse from a https://foo.example.com/innocent superproject to a https::attacker.example.com/exploit submodule. Fortunately a leading ':' in the first path component after a series of leading './' and '../' components is unlikely to show up in other contexts, so we can catch this by detecting that pattern. Reported-by: Jeff King <peff@peff.net> Signed-off-by: Jonathan Nieder <jrnieder@gmail.com> Reviewed-by: Jeff King <peff@peff.net>
2020-04-19 03:54:13 +00:00
decoded = url_decode(url);
has_nl = !!strchr(decoded, '\n');
free(decoded);
if (has_nl)
return -1;
credential: treat URL without scheme as invalid libcurl permits making requests without a URL scheme specified. In this case, it guesses the URL from the hostname, so I can run git ls-remote http::ftp.example.com/path/to/repo and it would make an FTP request. Any user intentionally using such a URL is likely to have made a typo. Unfortunately, credential_from_url is not able to determine the host and protocol in order to determine appropriate credentials to send, and until "credential: refuse to operate when missing host or protocol", this resulted in another host's credentials being leaked to the named host. Teach credential_from_url_gently to consider such a URL to be invalid so that fsck can detect and block gitmodules files with such URLs, allowing server operators to avoid serving them to downstream users running older versions of Git. This also means that when such URLs are passed on the command line, Git will print a clearer error so affected users can switch to the simpler URL that explicitly specifies the host and protocol they intend. One subtlety: .gitmodules files can contain relative URLs, representing a URL relative to the URL they were cloned from. The relative URL resolver used for .gitmodules can follow ".." components out of the path part and past the host part of a URL, meaning that such a relative URL can be used to traverse from a https://foo.example.com/innocent superproject to a https::attacker.example.com/exploit submodule. Fortunately a leading ':' in the first path component after a series of leading './' and '../' components is unlikely to show up in other contexts, so we can catch this by detecting that pattern. Reported-by: Jeff King <peff@peff.net> Signed-off-by: Jonathan Nieder <jrnieder@gmail.com> Reviewed-by: Jeff King <peff@peff.net>
2020-04-19 03:54:13 +00:00
/*
* URLs which escape their root via "../" can overwrite
* the host field and previous components, resolving to
* URLs like https::example.com/submodule.git and
* https:///example.com/submodule.git that were
credential: treat URL without scheme as invalid libcurl permits making requests without a URL scheme specified. In this case, it guesses the URL from the hostname, so I can run git ls-remote http::ftp.example.com/path/to/repo and it would make an FTP request. Any user intentionally using such a URL is likely to have made a typo. Unfortunately, credential_from_url is not able to determine the host and protocol in order to determine appropriate credentials to send, and until "credential: refuse to operate when missing host or protocol", this resulted in another host's credentials being leaked to the named host. Teach credential_from_url_gently to consider such a URL to be invalid so that fsck can detect and block gitmodules files with such URLs, allowing server operators to avoid serving them to downstream users running older versions of Git. This also means that when such URLs are passed on the command line, Git will print a clearer error so affected users can switch to the simpler URL that explicitly specifies the host and protocol they intend. One subtlety: .gitmodules files can contain relative URLs, representing a URL relative to the URL they were cloned from. The relative URL resolver used for .gitmodules can follow ".." components out of the path part and past the host part of a URL, meaning that such a relative URL can be used to traverse from a https://foo.example.com/innocent superproject to a https::attacker.example.com/exploit submodule. Fortunately a leading ':' in the first path component after a series of leading './' and '../' components is unlikely to show up in other contexts, so we can catch this by detecting that pattern. Reported-by: Jeff King <peff@peff.net> Signed-off-by: Jonathan Nieder <jrnieder@gmail.com> Reviewed-by: Jeff King <peff@peff.net>
2020-04-19 03:54:13 +00:00
* susceptible to CVE-2020-11008.
*/
if (count_leading_dotdots(url, &next) > 0 &&
(*next == ':' || *next == '/'))
credential: treat URL without scheme as invalid libcurl permits making requests without a URL scheme specified. In this case, it guesses the URL from the hostname, so I can run git ls-remote http::ftp.example.com/path/to/repo and it would make an FTP request. Any user intentionally using such a URL is likely to have made a typo. Unfortunately, credential_from_url is not able to determine the host and protocol in order to determine appropriate credentials to send, and until "credential: refuse to operate when missing host or protocol", this resulted in another host's credentials being leaked to the named host. Teach credential_from_url_gently to consider such a URL to be invalid so that fsck can detect and block gitmodules files with such URLs, allowing server operators to avoid serving them to downstream users running older versions of Git. This also means that when such URLs are passed on the command line, Git will print a clearer error so affected users can switch to the simpler URL that explicitly specifies the host and protocol they intend. One subtlety: .gitmodules files can contain relative URLs, representing a URL relative to the URL they were cloned from. The relative URL resolver used for .gitmodules can follow ".." components out of the path part and past the host part of a URL, meaning that such a relative URL can be used to traverse from a https://foo.example.com/innocent superproject to a https::attacker.example.com/exploit submodule. Fortunately a leading ':' in the first path component after a series of leading './' and '../' components is unlikely to show up in other contexts, so we can catch this by detecting that pattern. Reported-by: Jeff King <peff@peff.net> Signed-off-by: Jonathan Nieder <jrnieder@gmail.com> Reviewed-by: Jeff King <peff@peff.net>
2020-04-19 03:54:13 +00:00
return -1;
}
else if (url_to_curl_url(url, &curl_url)) {
struct credential c = CREDENTIAL_INIT;
int ret = 0;
if (credential_from_url_gently(&c, curl_url, 1) ||
!*c.host)
ret = -1;
credential_clear(&c);
return ret;
}
return 0;
}
struct fsck_gitmodules_data {
const struct object_id *oid;
struct fsck_options *options;
int ret;
};
static int fsck_gitmodules_fn(const char *var, const char *value, void *vdata)
{
struct fsck_gitmodules_data *data = vdata;
const char *subsection, *key;
size_t subsection_len;
char *name;
if (parse_config_key(var, "submodule", &subsection, &subsection_len, &key) < 0 ||
!subsection)
return 0;
name = xmemdupz(subsection, subsection_len);
if (check_submodule_name(name) < 0)
data->ret |= report(data->options,
data->oid, OBJ_BLOB,
FSCK_MSG_GITMODULES_NAME,
"disallowed submodule name: %s",
name);
if (!strcmp(key, "url") && value &&
check_submodule_url(value) < 0)
data->ret |= report(data->options,
data->oid, OBJ_BLOB,
FSCK_MSG_GITMODULES_URL,
"disallowed submodule url: %s",
value);
if (!strcmp(key, "path") && value &&
looks_like_command_line_option(value))
data->ret |= report(data->options,
data->oid, OBJ_BLOB,
FSCK_MSG_GITMODULES_PATH,
"disallowed submodule path: %s",
value);
if (!strcmp(key, "update") && value &&
parse_submodule_update_type(value) == SM_UPDATE_COMMAND)
2019-12-10 06:17:55 +00:00
data->ret |= report(data->options, data->oid, OBJ_BLOB,
FSCK_MSG_GITMODULES_UPDATE,
"disallowed submodule update setting: %s",
value);
free(name);
return 0;
}
static int fsck_blob(const struct object_id *oid, const char *buf,
unsigned long size, struct fsck_options *options)
{
int ret = 0;
if (object_on_skiplist(options, oid))
fsck: check skiplist for object in fsck_blob() Since commit ed8b10f631 ("fsck: check .gitmodules content", 2018-05-02), fsck will issue an error message for '.gitmodules' content that cannot be parsed correctly. This is the case, even when the corresponding blob object has been included on the skiplist. For example, using the cgit repository, we see the following: $ git fsck Checking object directories: 100% (256/256), done. error: bad config line 5 in blob .gitmodules error in blob 51dd1eff1edc663674df9ab85d2786a40f7ae3a5: gitmodulesParse: could not parse gitmodules blob Checking objects: 100% (6626/6626), done. $ $ git config fsck.skiplist '.git/skip' $ echo 51dd1eff1edc663674df9ab85d2786a40f7ae3a5 >.git/skip $ $ git fsck Checking object directories: 100% (256/256), done. error: bad config line 5 in blob .gitmodules Checking objects: 100% (6626/6626), done. $ Note that the error message issued by the config parser is still present, despite adding the object-id of the blob to the skiplist. One solution would be to provide a means of suppressing the messages issued by the config parser. However, given that (logically) we are asking fsck to ignore this object, a simpler approach is to just not call the config parser if the object is to be skipped. Add a check to the 'fsck_blob()' processing function, to determine if the object is on the skiplist and, if so, exit the function early. Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-27 18:39:53 +00:00
return 0;
if (oidset_contains(&options->gitmodules_found, oid)) {
struct config_options config_opts = { 0 };
struct fsck_gitmodules_data data;
oidset_insert(&options->gitmodules_done, oid);
if (!buf) {
/*
* A missing buffer here is a sign that the caller found the
* blob too gigantic to load into memory. Let's just consider
* that an error.
*/
return report(options, oid, OBJ_BLOB,
FSCK_MSG_GITMODULES_LARGE,
".gitmodules too large to parse");
}
data.oid = oid;
data.options = options;
data.ret = 0;
config_opts.error_action = CONFIG_ERROR_SILENT;
if (git_config_from_mem(fsck_gitmodules_fn, CONFIG_ORIGIN_BLOB,
".gitmodules", buf, size, &data, &config_opts))
data.ret |= report(options, oid, OBJ_BLOB,
FSCK_MSG_GITMODULES_PARSE,
"could not parse gitmodules blob");
ret |= data.ret;
}
if (oidset_contains(&options->gitattributes_found, oid)) {
const char *ptr;
oidset_insert(&options->gitattributes_done, oid);
if (!buf || size > ATTR_MAX_FILE_SIZE) {
/*
* A missing buffer here is a sign that the caller found the
* blob too gigantic to load into memory. Let's just consider
* that an error.
*/
return report(options, oid, OBJ_BLOB,
FSCK_MSG_GITATTRIBUTES_LARGE,
".gitattributes too large to parse");
}
for (ptr = buf; *ptr; ) {
const char *eol = strchrnul(ptr, '\n');
if (eol - ptr >= ATTR_MAX_LINE_LENGTH) {
ret |= report(options, oid, OBJ_BLOB,
FSCK_MSG_GITATTRIBUTES_LINE_LENGTH,
".gitattributes has too long lines to parse");
break;
}
ptr = *eol ? eol + 1 : eol;
}
}
return ret;
}
int fsck_object(struct object *obj, void *data, unsigned long size,
struct fsck_options *options)
{
if (!obj)
return report(options, NULL, OBJ_NONE, FSCK_MSG_BAD_OBJECT_SHA1, "no valid object to fsck");
return fsck_buffer(&obj->oid, obj->type, data, size, options);
}
int fsck_buffer(const struct object_id *oid, enum object_type type,
void *data, unsigned long size,
struct fsck_options *options)
{
if (type == OBJ_BLOB)
return fsck_blob(oid, data, size, options);
if (type == OBJ_TREE)
return fsck_tree(oid, data, size, options);
if (type == OBJ_COMMIT)
return fsck_commit(oid, data, size, options);
if (type == OBJ_TAG)
return fsck_tag(oid, data, size, options);
return report(options, oid, type,
FSCK_MSG_UNKNOWN_TYPE,
"unknown type '%d' (internal fsck error)",
type);
}
int fsck_error_function(struct fsck_options *o,
const struct object_id *oid,
enum object_type object_type,
enum fsck_msg_type msg_type,
enum fsck_msg_id msg_id,
const char *message)
{
if (msg_type == FSCK_WARN) {
warning("object %s: %s", fsck_describe_object(o, oid), message);
return 0;
}
error("object %s: %s", fsck_describe_object(o, oid), message);
return 1;
}
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
static int fsck_blobs(struct oidset *blobs_found, struct oidset *blobs_done,
enum fsck_msg_id msg_missing, enum fsck_msg_id msg_type,
struct fsck_options *options, const char *blob_type)
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
{
int ret = 0;
struct oidset_iter iter;
const struct object_id *oid;
oidset_iter_init(blobs_found, &iter);
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
while ((oid = oidset_iter_next(&iter))) {
enum object_type type;
unsigned long size;
char *buf;
if (oidset_contains(blobs_done, oid))
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
continue;
buf = repo_read_object_file(the_repository, oid, &type, &size);
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
if (!buf) {
if (is_promisor_object(oid))
continue;
ret |= report(options,
oid, OBJ_BLOB, msg_missing,
"unable to read %s blob", blob_type);
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
continue;
}
if (type == OBJ_BLOB)
ret |= fsck_blob(oid, buf, size, options);
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
else
ret |= report(options, oid, type, msg_type,
"non-blob found at %s", blob_type);
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
free(buf);
}
oidset_clear(blobs_found);
oidset_clear(blobs_done);
return ret;
}
int fsck_finish(struct fsck_options *options)
{
int ret = 0;
ret |= fsck_blobs(&options->gitmodules_found, &options->gitmodules_done,
FSCK_MSG_GITMODULES_MISSING, FSCK_MSG_GITMODULES_BLOB,
options, ".gitmodules");
ret |= fsck_blobs(&options->gitattributes_found, &options->gitattributes_done,
FSCK_MSG_GITATTRIBUTES_MISSING, FSCK_MSG_GITATTRIBUTES_BLOB,
options, ".gitattributes");
fsck: detect gitmodules files In preparation for performing fsck checks on .gitmodules files, this commit plumbs in the actual detection of the files. Note that unlike most other fsck checks, this cannot be a property of a single object: we must know that the object is found at a ".gitmodules" path at the root tree of a commit. Since the fsck code only sees one object at a time, we have to mark the related objects to fit the puzzle together. When we see a commit we mark its tree as a root tree, and when we see a root tree with a .gitmodules file, we mark the corresponding blob to be checked. In an ideal world, we'd check the objects in topological order: commits followed by trees followed by blobs. In that case we can avoid ever loading an object twice, since all markings would be complete by the time we get to the marked objects. And indeed, if we are checking a single packfile, this is the order in which Git will generally write the objects. But we can't count on that: 1. git-fsck may show us the objects in arbitrary order (loose objects are fed in sha1 order, but we may also have multiple packs, and we process each pack fully in sequence). 2. The type ordering is just what git-pack-objects happens to write now. The pack format does not require a specific order, and it's possible that future versions of Git (or a custom version trying to fool official Git's fsck checks!) may order it differently. 3. We may not even be fscking all of the relevant objects at once. Consider pushing with transfer.fsckObjects, where one push adds a blob at path "foo", and then a second push adds the same blob at path ".gitmodules". The blob is not part of the second push at all, but we need to mark and check it. So in the general case, we need to make up to three passes over the objects: once to make sure we've seen all commits, then once to cover any trees we might have missed, and then a final pass to cover any .gitmodules blobs we found in the second pass. We can simplify things a bit by loosening the requirement that we find .gitmodules only at root trees. Technically a file like "subdir/.gitmodules" is not parsed by Git, but it's not unreasonable for us to declare that Git is aware of all ".gitmodules" files and make them eligible for checking. That lets us drop the root-tree requirement, which eliminates one pass entirely. And it makes our worst case much better: instead of potentially queueing every root tree to be re-examined, the worst case is that we queue each unique .gitmodules blob for a second look. This patch just adds the boilerplate to find .gitmodules files. The actual content checks will come in a subsequent commit. Signed-off-by: Jeff King <peff@peff.net>
2018-05-02 21:20:08 +00:00
return ret;
}
int git_fsck_config(const char *var, const char *value, void *cb)
{
struct fsck_options *options = cb;
if (strcmp(var, "fsck.skiplist") == 0) {
const char *path;
struct strbuf sb = STRBUF_INIT;
if (git_config_pathname(&path, var, value))
return 1;
strbuf_addf(&sb, "skiplist=%s", path);
free((char *)path);
fsck_set_msg_types(options, sb.buf);
strbuf_release(&sb);
return 0;
}
if (skip_prefix(var, "fsck.", &var)) {
fsck_set_msg_type(options, var, value);
return 0;
}
return git_default_config(var, value, cb);
}
/*
* Custom error callbacks that are used in more than one place.
*/
int fsck_error_cb_print_missing_gitmodules(struct fsck_options *o,
const struct object_id *oid,
enum object_type object_type,
enum fsck_msg_type msg_type,
enum fsck_msg_id msg_id,
const char *message)
{
if (msg_id == FSCK_MSG_GITMODULES_MISSING) {
puts(oid_to_hex(oid));
return 0;
}
return fsck_error_function(o, oid, object_type, msg_type, msg_id, message);
}