1 Use of index and Racy Git problem
2 =================================
7 The index is one of the most important data structures in Git.
8 It represents a virtual working tree state by recording list of
9 paths and their object names and serves as a staging area to
10 write out the next tree object to be committed. The state is
11 "virtual" in the sense that it does not necessarily have to, and
12 often does not, match the files in the working tree.
14 There are cases Git needs to examine the differences between the
15 virtual working tree state in the index and the files in the
16 working tree. The most obvious case is when the user asks `git
17 diff` (or its low level implementation, `git diff-files`) or
18 `git-ls-files --modified`. In addition, Git internally checks
19 if the files in the working tree are different from what are
20 recorded in the index to avoid stomping on local changes in them
21 during patch application, switching branches, and merging.
23 In order to speed up this comparison between the files in the
24 working tree and the index entries, the index entries record the
25 information obtained from the filesystem via `lstat(2)` system
26 call when they were last updated. When checking if they differ,
27 Git first runs `lstat(2)` on the files and compares the result
28 with this information (this is what was originally done by the
29 `ce_match_stat()` function, but the current code does it in
30 `ce_match_stat_basic()` function). If some of these "cached
31 stat information" fields do not match, Git can tell that the
32 files are modified without even looking at their contents.
34 Note: not all members in `struct stat` obtained via `lstat(2)`
35 are used for this comparison. For example, `st_atime` obviously
36 is not useful. Currently, Git compares the file type (regular
37 files vs symbolic links) and executable bits (only for regular
38 files) from `st_mode` member, `st_mtime` and `st_ctime`
39 timestamps, `st_uid`, `st_gid`, `st_ino`, and `st_size` members.
40 With a `USE_STDEV` compile-time option, `st_dev` is also
41 compared, but this is not enabled by default because this member
42 is not stable on network filesystems. With `USE_NSEC`
43 compile-time option, `st_mtim.tv_nsec` and `st_ctim.tv_nsec`
44 members are also compared. On Linux, this is not enabled by default
45 because in-core timestamps can have finer granularity than
46 on-disk timestamps, resulting in meaningless changes when an
47 inode is evicted from the inode cache. See commit 8ce13b0
48 of git://git.kernel.org/pub/scm/linux/kernel/git/tglx/history.git
49 ([PATCH] Sync in core time granularity with filesystems,
50 2005-01-04). This patch is included in kernel 2.6.11 and newer, but
51 only fixes the issue for file systems with exactly 1 ns or 1 s
52 resolution. Other file systems are still broken in current Linux
53 kernels (e.g. CEPH, CIFS, NTFS, UDF), see
54 https://lore.kernel.org/lkml/5577240D.7020309@gmail.com/
59 There is one slight problem with the optimization based on the
60 cached stat information. Consider this sequence:
63 $ git update-index 'foo'
64 : modify 'foo' again, in-place, without changing its size
66 The first `update-index` computes the object name of the
67 contents of file `foo` and updates the index entry for `foo`
68 along with the `struct stat` information. If the modification
69 that follows it happens very fast so that the file's `st_mtime`
70 timestamp does not change, after this sequence, the cached stat
71 information the index entry records still exactly match what you
72 would see in the filesystem, even though the file `foo` is now
74 This way, Git can incorrectly think files in the working tree
75 are unmodified even though they actually are. This is called
76 the "racy Git" problem (discovered by Pasky), and the entries
77 that appear clean when they may not be because of this problem
78 are called "racily clean".
80 To avoid this problem, Git does two things:
82 . When the cached stat information says the file has not been
83 modified, and the `st_mtime` is the same as (or newer than)
84 the timestamp of the index file itself (which is the time `git
85 update-index foo` finished running in the above example), it
86 also compares the contents with the object registered in the
87 index entry to make sure they match.
89 . When the index file is updated that contains racily clean
90 entries, cached `st_size` information is truncated to zero
91 before writing a new version of the index file.
93 Because the index file itself is written after collecting all
94 the stat information from updated paths, `st_mtime` timestamp of
95 it is usually the same as or newer than any of the paths the
96 index contains. And no matter how quick the modification that
97 follows `git update-index foo` finishes, the resulting
98 `st_mtime` timestamp on `foo` cannot get a value earlier
99 than the index file. Therefore, index entries that can be
100 racily clean are limited to the ones that have the same
101 timestamp as the index file itself.
103 The callers that want to check if an index entry matches the
104 corresponding file in the working tree continue to call
105 `ce_match_stat()`, but with this change, `ce_match_stat()` uses
106 `ce_modified_check_fs()` to see if racily clean ones are
107 actually clean after comparing the cached stat information using
108 `ce_match_stat_basic()`.
110 The problem the latter solves is this sequence:
112 $ git update-index 'foo'
113 : modify 'foo' in-place without changing its size
114 : wait for enough time
115 $ git update-index 'bar'
117 Without the latter, the timestamp of the index file gets a newer
118 value, and falsely clean entry `foo` would not be caught by the
119 timestamp comparison check done with the former logic anymore.
120 The latter makes sure that the cached stat information for `foo`
121 would never match with the file in the working tree, so later
122 checks by `ce_match_stat_basic()` would report that the index entry
123 does not match the file and Git does not have to fall back on more
124 expensive `ce_modified_check_fs()`.
130 The runtime penalty of falling back to `ce_modified_check_fs()`
131 from `ce_match_stat()` can be very expensive when there are many
132 racily clean entries. An obvious way to artificially create
133 this situation is to give the same timestamp to all the files in
134 the working tree in a large project, run `git update-index` on
135 them, and give the same timestamp to the index file:
138 $ git ls-files | xargs touch -r .datestamp
139 $ git ls-files | git update-index --stdin
140 $ touch -r .datestamp .git/index
142 This will make all index entries racily clean. The linux project, for
143 example, there are over 20,000 files in the working tree. On my
144 Athlon 64 X2 3800+, after the above:
146 $ /usr/bin/time git diff-files
147 1.68user 0.54system 0:02.22elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k
148 0inputs+0outputs (0major+67111minor)pagefaults 0swaps
149 $ git update-index MAINTAINERS
150 $ /usr/bin/time git diff-files
151 0.02user 0.12system 0:00.14elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k
152 0inputs+0outputs (0major+935minor)pagefaults 0swaps
154 Running `git update-index` in the middle checked the racily
155 clean entries, and left the cached `st_mtime` for all the paths
156 intact because they were actually clean (so this step took about
157 the same amount of time as the first `git diff-files`). After
158 that, they are not racily clean anymore but are truly clean, so
159 the second invocation of `git diff-files` fully took advantage
160 of the cached stat information.
163 Avoiding runtime penalty
164 ------------------------
166 In order to avoid the above runtime penalty, post 1.4.2 Git used
167 to have a code that made sure the index file
168 got timestamp newer than the youngest files in the index when
169 there are many young files with the same timestamp as the
170 resulting index file would otherwise would have by waiting
171 before finishing writing the index file out.
173 I suspected that in practice the situation where many paths in the
174 index are all racily clean was quite rare. The only code paths
175 that can record recent timestamp for large number of paths are:
177 . Initial `git add .` of a large project.
179 . `git checkout` of a large project from an empty index into an
180 unpopulated working tree.
182 Note: switching branches with `git checkout` keeps the cached
183 stat information of existing working tree files that are the
184 same between the current branch and the new branch, which are
185 all older than the resulting index file, and they will not
186 become racily clean. Only the files that are actually checked
187 out can become racily clean.
189 In a large project where raciness avoidance cost really matters,
190 however, the initial computation of all object names in the
191 index takes more than one second, and the index file is written
192 out after all that happens. Therefore the timestamp of the
193 index file will be more than one seconds later than the
194 youngest file in the working tree. This means that in these
195 cases there actually will not be any racily clean entry in
198 Based on this discussion, the current code does not use the
199 "workaround" to avoid the runtime penalty that does not exist in
200 practice anymore. This was done with commit 0fc82cff on Aug 15,
projects / git / blob