1Use of index and Racy git problem 2================================= 3 4Background 5---------- 6 7The index is one of the most important data structures in git. 8It represents a virtual working tree state by recording list of 9paths and their object names and serves as a staging area to 10write out the next tree object to be committed. The state is 11"virtual" in the sense that it does not necessarily have to, and 12often does not, match the files in the working tree. 13 14There are cases git needs to examine the differences between the 15virtual working tree state in the index and the files in the 16working tree. The most obvious case is when the user asks `git 17diff` (or its low level implementation, `git diff-files`) or 18`git-ls-files --modified`. In addition, git internally checks 19if the files in the working tree are different from what are 20recorded in the index to avoid stomping on local changes in them 21during patch application, switching branches, and merging. 22 23In order to speed up this comparison between the files in the 24working tree and the index entries, the index entries record the 25information obtained from the filesystem via `lstat(2)` system 26call when they were last updated. When checking if they differ, 27git first runs `lstat(2)` on the files and compares the result 28with 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 31stat information" fields do not match, git can tell that the 32files are modified without even looking at their contents. 33 34Note: not all members in `struct stat` obtained via `lstat(2)` 35are used for this comparison. For example, `st_atime` obviously 36is not useful. Currently, git compares the file type (regular 37files vs symbolic links) and executable bits (only for regular 38files) from `st_mode` member, `st_mtime` and `st_ctime` 39timestamps, `st_uid`, `st_gid`, `st_ino`, and `st_size` members. 40With a `USE_STDEV` compile-time option, `st_dev` is also 41compared, but this is not enabled by default because this member 42is not stable on network filesystems. With `USE_NSEC` 43compile-time option, `st_mtim.tv_nsec` and `st_ctim.tv_nsec` 44members are also compared, but this is not enabled by default 45because in-core timestamps can have finer granularity than 46on-disk timestamps, resulting in meaningless changes when an 47inode is evicted from the inode cache. See commit 8ce13b0 48of git://git.kernel.org/pub/scm/linux/kernel/git/tglx/history.git 49([PATCH] Sync in core time granuality with filesystems, 502005-01-04). 51 52Racy git 53-------- 54 55There is one slight problem with the optimization based on the 56cached stat information. Consider this sequence: 57 58 : modify 'foo' 59 $ git update-index 'foo' 60 : modify 'foo' again, in-place, without changing its size 61 62The first `update-index` computes the object name of the 63contents of file `foo` and updates the index entry for `foo` 64along with the `struct stat` information. If the modification 65that follows it happens very fast so that the file's `st_mtime` 66timestamp does not change, after this sequence, the cached stat 67information the index entry records still exactly match what you 68would see in the filesystem, even though the file `foo` is now 69different. 70This way, git can incorrectly think files in the working tree 71are unmodified even though they actually are. This is called 72the "racy git" problem (discovered by Pasky), and the entries 73that appear clean when they may not be because of this problem 74are called "racily clean". 75 76To avoid this problem, git does two things: 77 78. When the cached stat information says the file has not been 79 modified, and the `st_mtime` is the same as (or newer than) 80 the timestamp of the index file itself (which is the time `git 81 update-index foo` finished running in the above example), it 82 also compares the contents with the object registered in the 83 index entry to make sure they match. 84 85. When the index file is updated that contains racily clean 86 entries, cached `st_size` information is truncated to zero 87 before writing a new version of the index file. 88 89Because the index file itself is written after collecting all 90the stat information from updated paths, `st_mtime` timestamp of 91it is usually the same as or newer than any of the paths the 92index contains. And no matter how quick the modification that 93follows `git update-index foo` finishes, the resulting 94`st_mtime` timestamp on `foo` cannot get a value earlier 95than the index file. Therefore, index entries that can be 96racily clean are limited to the ones that have the same 97timestamp as the index file itself. 98 99The callers that want to check if an index entry matches the 100corresponding file in the working tree continue to call 101`ce_match_stat()`, but with this change, `ce_match_stat()` uses 102`ce_modified_check_fs()` to see if racily clean ones are 103actually clean after comparing the cached stat information using 104`ce_match_stat_basic()`. 105 106The problem the latter solves is this sequence: 107 108 $ git update-index 'foo' 109 : modify 'foo' in-place without changing its size 110 : wait for enough time 111 $ git update-index 'bar' 112 113Without the latter, the timestamp of the index file gets a newer 114value, and falsely clean entry `foo` would not be caught by the 115timestamp comparison check done with the former logic anymore. 116The latter makes sure that the cached stat information for `foo` 117would never match with the file in the working tree, so later 118checks by `ce_match_stat_basic()` would report that the index entry 119does not match the file and git does not have to fall back on more 120expensive `ce_modified_check_fs()`. 121 122 123Runtime penalty 124--------------- 125 126The runtime penalty of falling back to `ce_modified_check_fs()` 127from `ce_match_stat()` can be very expensive when there are many 128racily clean entries. An obvious way to artificially create 129this situation is to give the same timestamp to all the files in 130the working tree in a large project, run `git update-index` on 131them, and give the same timestamp to the index file: 132 133 $ date >.datestamp 134 $ git ls-files | xargs touch -r .datestamp 135 $ git ls-files | git update-index --stdin 136 $ touch -r .datestamp .git/index 137 138This will make all index entries racily clean. The linux-2.6 139project, for example, there are over 20,000 files in the working 140tree. On my Athlon 64 X2 3800+, after the above: 141 142 $ /usr/bin/time git diff-files 143 1.68user 0.54system 0:02.22elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k 144 0inputs+0outputs (0major+67111minor)pagefaults 0swaps 145 $ git update-index MAINTAINERS 146 $ /usr/bin/time git diff-files 147 0.02user 0.12system 0:00.14elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k 148 0inputs+0outputs (0major+935minor)pagefaults 0swaps 149 150Running `git update-index` in the middle checked the racily 151clean entries, and left the cached `st_mtime` for all the paths 152intact because they were actually clean (so this step took about 153the same amount of time as the first `git diff-files`). After 154that, they are not racily clean anymore but are truly clean, so 155the second invocation of `git diff-files` fully took advantage 156of the cached stat information. 157 158 159Avoiding runtime penalty 160------------------------ 161 162In order to avoid the above runtime penalty, post 1.4.2 git used 163to have a code that made sure the index file 164got timestamp newer than the youngest files in the index when 165there are many young files with the same timestamp as the 166resulting index file would otherwise would have by waiting 167before finishing writing the index file out. 168 169I suspected that in practice the situation where many paths in the 170index are all racily clean was quite rare. The only code paths 171that can record recent timestamp for large number of paths are: 172 173. Initial `git add .` of a large project. 174 175. `git checkout` of a large project from an empty index into an 176 unpopulated working tree. 177 178Note: switching branches with `git checkout` keeps the cached 179stat information of existing working tree files that are the 180same between the current branch and the new branch, which are 181all older than the resulting index file, and they will not 182become racily clean. Only the files that are actually checked 183out can become racily clean. 184 185In a large project where raciness avoidance cost really matters, 186however, the initial computation of all object names in the 187index takes more than one second, and the index file is written 188out after all that happens. Therefore the timestamp of the 189index file will be more than one seconds later than the 190youngest file in the working tree. This means that in these 191cases there actually will not be any racily clean entry in 192the resulting index. 193 194Based on this discussion, the current code does not use the 195"workaround" to avoid the runtime penalty that does not exist in 196practice anymore. This was done with commit 0fc82cff on Aug 15, 1972006.