git/read-cache.c

696 lines
17 KiB
C
Raw Normal View History

/*
* GIT - The information manager from hell
*
* Copyright (C) Linus Torvalds, 2005
*/
#include "cache.h"
struct cache_entry **active_cache = NULL;
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 08:02:15 +00:00
static time_t index_file_timestamp;
unsigned int active_nr = 0, active_alloc = 0, active_cache_changed = 0;
/*
* This only updates the "non-critical" parts of the directory
* cache, ie the parts that aren't tracked by GIT, and only used
* to validate the cache.
*/
void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
{
ce->ce_ctime.sec = htonl(st->st_ctime);
ce->ce_mtime.sec = htonl(st->st_mtime);
#ifdef USE_NSEC
ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
#endif
ce->ce_dev = htonl(st->st_dev);
ce->ce_ino = htonl(st->st_ino);
ce->ce_uid = htonl(st->st_uid);
ce->ce_gid = htonl(st->st_gid);
ce->ce_size = htonl(st->st_size);
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 05:15:24 +00:00
if (assume_unchanged)
ce->ce_flags |= htons(CE_VALID);
}
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 08:02:15 +00:00
static int ce_compare_data(struct cache_entry *ce, struct stat *st)
{
int match = -1;
int fd = open(ce->name, O_RDONLY);
if (fd >= 0) {
unsigned char sha1[20];
if (!index_fd(sha1, fd, st, 0, NULL))
match = memcmp(sha1, ce->sha1, 20);
close(fd);
}
return match;
}
static int ce_compare_link(struct cache_entry *ce, unsigned long expected_size)
{
int match = -1;
char *target;
void *buffer;
unsigned long size;
char type[10];
int len;
target = xmalloc(expected_size);
len = readlink(ce->name, target, expected_size);
if (len != expected_size) {
free(target);
return -1;
}
buffer = read_sha1_file(ce->sha1, type, &size);
if (!buffer) {
free(target);
return -1;
}
if (size == expected_size)
match = memcmp(buffer, target, size);
free(buffer);
free(target);
return match;
}
static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
{
switch (st->st_mode & S_IFMT) {
case S_IFREG:
if (ce_compare_data(ce, st))
return DATA_CHANGED;
break;
case S_IFLNK:
if (ce_compare_link(ce, st->st_size))
return DATA_CHANGED;
break;
default:
return TYPE_CHANGED;
}
return 0;
}
static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
{
unsigned int changed = 0;
switch (ntohl(ce->ce_mode) & S_IFMT) {
case S_IFREG:
changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
/* We consider only the owner x bit to be relevant for
* "mode changes"
*/
if (trust_executable_bit &&
(0100 & (ntohl(ce->ce_mode) ^ st->st_mode)))
changed |= MODE_CHANGED;
break;
case S_IFLNK:
changed |= !S_ISLNK(st->st_mode) ? TYPE_CHANGED : 0;
break;
default:
die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
}
if (ce->ce_mtime.sec != htonl(st->st_mtime))
changed |= MTIME_CHANGED;
if (ce->ce_ctime.sec != htonl(st->st_ctime))
changed |= CTIME_CHANGED;
#ifdef USE_NSEC
/*
* nsec seems unreliable - not all filesystems support it, so
* as long as it is in the inode cache you get right nsec
* but after it gets flushed, you get zero nsec.
*/
if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
changed |= MTIME_CHANGED;
if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
changed |= CTIME_CHANGED;
#endif
if (ce->ce_uid != htonl(st->st_uid) ||
ce->ce_gid != htonl(st->st_gid))
changed |= OWNER_CHANGED;
if (ce->ce_ino != htonl(st->st_ino))
changed |= INODE_CHANGED;
#ifdef USE_STDEV
/*
* st_dev breaks on network filesystems where different
* clients will have different views of what "device"
* the filesystem is on
*/
if (ce->ce_dev != htonl(st->st_dev))
changed |= INODE_CHANGED;
#endif
if (ce->ce_size != htonl(st->st_size))
changed |= DATA_CHANGED;
return changed;
}
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 05:15:24 +00:00
int ce_match_stat(struct cache_entry *ce, struct stat *st, int ignore_valid)
{
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 05:15:24 +00:00
unsigned int changed;
/*
* If it's marked as always valid in the index, it's
* valid whatever the checked-out copy says.
*/
if (!ignore_valid && (ce->ce_flags & htons(CE_VALID)))
return 0;
changed = ce_match_stat_basic(ce, st);
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 08:02:15 +00:00
/*
* Within 1 second of this sequence:
* echo xyzzy >file && git-update-index --add file
* running this command:
* echo frotz >file
* would give a falsely clean cache entry. The mtime and
* length match the cache, and other stat fields do not change.
*
* We could detect this at update-index time (the cache entry
* being registered/updated records the same time as "now")
* and delay the return from git-update-index, but that would
* effectively mean we can make at most one commit per second,
* which is not acceptable. Instead, we check cache entries
* whose mtime are the same as the index file timestamp more
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 05:15:24 +00:00
* carefully than others.
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 08:02:15 +00:00
*/
if (!changed &&
index_file_timestamp &&
index_file_timestamp <= ntohl(ce->ce_mtime.sec))
changed |= ce_modified_check_fs(ce, st);
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 08:02:15 +00:00
return changed;
}
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 05:15:24 +00:00
int ce_modified(struct cache_entry *ce, struct stat *st, int really)
{
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 08:02:15 +00:00
int changed, changed_fs;
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 05:15:24 +00:00
changed = ce_match_stat(ce, st, really);
if (!changed)
return 0;
/*
* If the mode or type has changed, there's no point in trying
* to refresh the entry - it's not going to match
*/
if (changed & (MODE_CHANGED | TYPE_CHANGED))
return changed;
/* Immediately after read-tree or update-index --cacheinfo,
* the length field is zero. For other cases the ce_size
* should match the SHA1 recorded in the index entry.
*/
if ((changed & DATA_CHANGED) && ce->ce_size != htonl(0))
return changed;
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 08:02:15 +00:00
changed_fs = ce_modified_check_fs(ce, st);
if (changed_fs)
return changed | changed_fs;
return 0;
}
int base_name_compare(const char *name1, int len1, int mode1,
const char *name2, int len2, int mode2)
{
unsigned char c1, c2;
int len = len1 < len2 ? len1 : len2;
int cmp;
cmp = memcmp(name1, name2, len);
if (cmp)
return cmp;
c1 = name1[len];
c2 = name2[len];
if (!c1 && S_ISDIR(mode1))
c1 = '/';
if (!c2 && S_ISDIR(mode2))
c2 = '/';
return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
}
int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
{
int len1 = flags1 & CE_NAMEMASK;
int len2 = flags2 & CE_NAMEMASK;
int len = len1 < len2 ? len1 : len2;
int cmp;
cmp = memcmp(name1, name2, len);
if (cmp)
return cmp;
if (len1 < len2)
return -1;
if (len1 > len2)
return 1;
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 05:15:24 +00:00
/* Compare stages */
flags1 &= CE_STAGEMASK;
flags2 &= CE_STAGEMASK;
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 05:15:24 +00:00
if (flags1 < flags2)
return -1;
if (flags1 > flags2)
return 1;
return 0;
}
int cache_name_pos(const char *name, int namelen)
{
int first, last;
first = 0;
last = active_nr;
while (last > first) {
int next = (last + first) >> 1;
struct cache_entry *ce = active_cache[next];
int cmp = cache_name_compare(name, namelen, ce->name, ntohs(ce->ce_flags));
if (!cmp)
return next;
if (cmp < 0) {
last = next;
continue;
}
first = next+1;
}
return -first-1;
}
/* Remove entry, return true if there are more entries to go.. */
int remove_cache_entry_at(int pos)
{
active_cache_changed = 1;
active_nr--;
if (pos >= active_nr)
return 0;
memmove(active_cache + pos, active_cache + pos + 1, (active_nr - pos) * sizeof(struct cache_entry *));
return 1;
}
int remove_file_from_cache(const char *path)
{
int pos = cache_name_pos(path, strlen(path));
if (pos < 0)
pos = -pos-1;
while (pos < active_nr && !strcmp(active_cache[pos]->name, path))
remove_cache_entry_at(pos);
return 0;
}
int ce_same_name(struct cache_entry *a, struct cache_entry *b)
{
int len = ce_namelen(a);
return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
}
int ce_path_match(const struct cache_entry *ce, const char **pathspec)
{
const char *match, *name;
int len;
if (!pathspec)
return 1;
len = ce_namelen(ce);
name = ce->name;
while ((match = *pathspec++) != NULL) {
int matchlen = strlen(match);
if (matchlen > len)
continue;
if (memcmp(name, match, matchlen))
continue;
if (matchlen && name[matchlen-1] == '/')
return 1;
if (name[matchlen] == '/' || !name[matchlen])
return 1;
if (!matchlen)
return 1;
}
return 0;
}
/*
* Do we have another file that has the beginning components being a
* proper superset of the name we're trying to add?
*/
static int has_file_name(const struct cache_entry *ce, int pos, int ok_to_replace)
{
int retval = 0;
int len = ce_namelen(ce);
int stage = ce_stage(ce);
const char *name = ce->name;
while (pos < active_nr) {
struct cache_entry *p = active_cache[pos++];
if (len >= ce_namelen(p))
break;
if (memcmp(name, p->name, len))
break;
if (ce_stage(p) != stage)
continue;
if (p->name[len] != '/')
continue;
retval = -1;
if (!ok_to_replace)
break;
remove_cache_entry_at(--pos);
}
return retval;
}
/*
* Do we have another file with a pathname that is a proper
* subset of the name we're trying to add?
*/
static int has_dir_name(const struct cache_entry *ce, int pos, int ok_to_replace)
{
int retval = 0;
int stage = ce_stage(ce);
const char *name = ce->name;
const char *slash = name + ce_namelen(ce);
for (;;) {
int len;
for (;;) {
if (*--slash == '/')
break;
if (slash <= ce->name)
return retval;
}
len = slash - name;
pos = cache_name_pos(name, ntohs(create_ce_flags(len, stage)));
if (pos >= 0) {
retval = -1;
if (ok_to_replace)
break;
remove_cache_entry_at(pos);
continue;
}
/*
* Trivial optimization: if we find an entry that
* already matches the sub-directory, then we know
* we're ok, and we can exit.
*/
pos = -pos-1;
while (pos < active_nr) {
struct cache_entry *p = active_cache[pos];
if ((ce_namelen(p) <= len) ||
(p->name[len] != '/') ||
memcmp(p->name, name, len))
break; /* not our subdirectory */
if (ce_stage(p) == stage)
/* p is at the same stage as our entry, and
* is a subdirectory of what we are looking
* at, so we cannot have conflicts at our
* level or anything shorter.
*/
return retval;
pos++;
}
}
return retval;
}
/* We may be in a situation where we already have path/file and path
* is being added, or we already have path and path/file is being
* added. Either one would result in a nonsense tree that has path
* twice when git-write-tree tries to write it out. Prevent it.
*
* If ok-to-replace is specified, we remove the conflicting entries
* from the cache so the caller should recompute the insert position.
* When this happens, we return non-zero.
*/
static int check_file_directory_conflict(const struct cache_entry *ce, int pos, int ok_to_replace)
{
/*
* We check if the path is a sub-path of a subsequent pathname
* first, since removing those will not change the position
* in the array
*/
int retval = has_file_name(ce, pos, ok_to_replace);
/*
* Then check if the path might have a clashing sub-directory
* before it.
*/
return retval + has_dir_name(ce, pos, ok_to_replace);
}
int add_cache_entry(struct cache_entry *ce, int option)
{
int pos;
int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 05:15:24 +00:00
pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
/* existing match? Just replace it. */
if (pos >= 0) {
active_cache_changed = 1;
active_cache[pos] = ce;
return 0;
}
pos = -pos-1;
/*
* Inserting a merged entry ("stage 0") into the index
* will always replace all non-merged entries..
*/
if (pos < active_nr && ce_stage(ce) == 0) {
while (ce_same_name(active_cache[pos], ce)) {
ok_to_add = 1;
if (!remove_cache_entry_at(pos))
break;
}
}
if (!ok_to_add)
return -1;
if (!skip_df_check &&
check_file_directory_conflict(ce, pos, ok_to_replace)) {
if (!ok_to_replace)
return -1;
pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
pos = -pos-1;
}
/* Make sure the array is big enough .. */
if (active_nr == active_alloc) {
active_alloc = alloc_nr(active_alloc);
active_cache = xrealloc(active_cache, active_alloc * sizeof(struct cache_entry *));
}
/* Add it in.. */
active_nr++;
if (active_nr > pos)
memmove(active_cache + pos + 1, active_cache + pos, (active_nr - pos - 1) * sizeof(ce));
active_cache[pos] = ce;
active_cache_changed = 1;
return 0;
}
static int verify_hdr(struct cache_header *hdr, unsigned long size)
{
SHA_CTX c;
unsigned char sha1[20];
if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
return error("bad signature");
if (hdr->hdr_version != htonl(2))
return error("bad index version");
SHA1_Init(&c);
SHA1_Update(&c, hdr, size - 20);
SHA1_Final(sha1, &c);
if (memcmp(sha1, (void *)hdr + size - 20, 20))
return error("bad index file sha1 signature");
return 0;
}
int read_cache(void)
{
int fd, i;
struct stat st;
unsigned long size, offset;
void *map;
struct cache_header *hdr;
errno = EBUSY;
if (active_cache)
return active_nr;
errno = ENOENT;
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 08:02:15 +00:00
index_file_timestamp = 0;
fd = open(get_index_file(), O_RDONLY);
if (fd < 0) {
if (errno == ENOENT)
return 0;
die("index file open failed (%s)", strerror(errno));
}
size = 0; // avoid gcc warning
map = MAP_FAILED;
if (!fstat(fd, &st)) {
size = st.st_size;
errno = EINVAL;
if (size >= sizeof(struct cache_header) + 20)
map = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
}
close(fd);
if (map == MAP_FAILED)
die("index file mmap failed (%s)", strerror(errno));
hdr = map;
if (verify_hdr(hdr, size) < 0)
goto unmap;
active_nr = ntohl(hdr->hdr_entries);
active_alloc = alloc_nr(active_nr);
active_cache = calloc(active_alloc, sizeof(struct cache_entry *));
offset = sizeof(*hdr);
for (i = 0; i < active_nr; i++) {
struct cache_entry *ce = map + offset;
offset = offset + ce_size(ce);
active_cache[i] = ce;
}
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 08:02:15 +00:00
index_file_timestamp = st.st_mtime;
return active_nr;
unmap:
munmap(map, size);
errno = EINVAL;
die("index file corrupt");
}
#define WRITE_BUFFER_SIZE 8192
static unsigned char write_buffer[WRITE_BUFFER_SIZE];
static unsigned long write_buffer_len;
static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
{
while (len) {
unsigned int buffered = write_buffer_len;
unsigned int partial = WRITE_BUFFER_SIZE - buffered;
if (partial > len)
partial = len;
memcpy(write_buffer + buffered, data, partial);
buffered += partial;
if (buffered == WRITE_BUFFER_SIZE) {
SHA1_Update(context, write_buffer, WRITE_BUFFER_SIZE);
if (write(fd, write_buffer, WRITE_BUFFER_SIZE) != WRITE_BUFFER_SIZE)
return -1;
buffered = 0;
}
write_buffer_len = buffered;
len -= partial;
data += partial;
}
return 0;
}
static int ce_flush(SHA_CTX *context, int fd)
{
unsigned int left = write_buffer_len;
if (left) {
write_buffer_len = 0;
SHA1_Update(context, write_buffer, left);
}
/* Flush first if not enough space for SHA1 signature */
if (left + 20 > WRITE_BUFFER_SIZE) {
if (write(fd, write_buffer, left) != left)
return -1;
left = 0;
}
/* Append the SHA1 signature at the end */
SHA1_Final(write_buffer + left, context);
left += 20;
if (write(fd, write_buffer, left) != left)
return -1;
return 0;
}
static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
{
/*
* The only thing we care about in this function is to smudge the
* falsely clean entry due to touch-update-touch race, so we leave
* everything else as they are. We are called for entries whose
* ce_mtime match the index file mtime.
*/
struct stat st;
if (lstat(ce->name, &st) < 0)
return;
if (ce_match_stat_basic(ce, &st))
return;
if (ce_modified_check_fs(ce, &st)) {
/* This is "racily clean"; smudge it. Note that this
* is a tricky code. At first glance, it may appear
* that it can break with this sequence:
*
* $ echo xyzzy >frotz
* $ git-update-index --add frotz
* $ : >frotz
* $ sleep 3
* $ echo filfre >nitfol
* $ git-update-index --add nitfol
*
* but it does not. Whe the second update-index runs,
* it notices that the entry "frotz" has the same timestamp
* as index, and if we were to smudge it by resetting its
* size to zero here, then the object name recorded
* in index is the 6-byte file but the cached stat information
* becomes zero --- which would then match what we would
* obtain from the filesystem next time we stat("frotz").
*
* However, the second update-index, before calling
* this function, notices that the cached size is 6
* bytes and what is on the filesystem is an empty
* file, and never calls us, so the cached size information
* for "frotz" stays 6 which does not match the filesystem.
*/
ce->ce_size = htonl(0);
}
}
int write_cache(int newfd, struct cache_entry **cache, int entries)
{
SHA_CTX c;
struct cache_header hdr;
int i, removed;
for (i = removed = 0; i < entries; i++)
if (!cache[i]->ce_mode)
removed++;
hdr.hdr_signature = htonl(CACHE_SIGNATURE);
hdr.hdr_version = htonl(2);
hdr.hdr_entries = htonl(entries - removed);
SHA1_Init(&c);
if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
return -1;
for (i = 0; i < entries; i++) {
struct cache_entry *ce = cache[i];
if (!ce->ce_mode)
continue;
if (index_file_timestamp &&
index_file_timestamp <= ntohl(ce->ce_mtime.sec))
ce_smudge_racily_clean_entry(ce);
if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
return -1;
}
return ce_flush(&c, newfd);
}