git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH 0/9] refs: stop using `chdir_notify_reparent()`

From
Jeff King <peff@peff.net>
Date
Jun 13, 2026, 14:00 UTC
Message-ID
<20260613140024.GA766297@coredump.intra.peff.net>
In-Reply-To
<aiukqI0Nj_RRn-wZ@pks.im>
On Fri, Jun 12, 2026 at 08:18:16AM +0200, Patrick Steinhardt wrote:
Show 6 quoted lines
> > If we move to a world of all absolute paths where chdir-notify is not
> > necessary, will we lose that optimization?
> 
> Probably. Unfortunately, the commit doesn't have any repeatable
> benchmarks in there, so it's hard to say whether we could still
> reproduce those issues or not.
Here's an easy-ish reproduction specific to the ref code:
  rm -rf a/
  dir=$(perl -e 'print "a/" x 1024')
  mkdir -p $dir &&
  cd $dir &&
  git init &&
  git commit --allow-empty -m foo &&
  seq -f 'create refs/heads/foo%05g HEAD' 10000 |
  git update-ref --stdin &&
  time git show-ref
Before your series, I get timings like this:
  real	0m0.078s
  user	0m0.020s
  sys	0m0.057s
After, I get:
  real	0m0.876s
  user	0m0.004s
  sys	0m0.872s

So it really is measurable (and I did not expect the effect to be nearly so large). Unsurprisingly the extra CPU goes to system time.

But obviously that case is quite silly. It's an absurdly deep hierarchy, and 10,000 loose refs is a lot. Just running "git pack-refs --all" brings the before/after to roughly the same timings (around 40ms -- faster even than the before timing).

So it _can_ matter, but I think ultimately the better direction is probably "make fewer syscalls". Which we do via packfiles, and via packed-refs, and eventually via reftables, all of which put more data into a single file.

I offer the script above more as food for thought, and not necessarily an argument against your series.

Show 16 quoted lines
> Ideally, we'd have the best of both worlds: absolute paths everywhere
> without the performance hit. A while back I had a discussion with
> Torvalds on the securiy mailing list around this issue, and ultimately
> the conclusion was that the best way forward would be to use openat(3p).
> 
> This wouldn't only allow us to optimize cases like this, but it also has
> the added benefit that we're much less prone to TOCTOU-style issues and
> we might even be able to use flags like O_BENEATH. So it would basically
> be win-win. The only problem is of course that Windows doesn't have
> openat(3p), so we'd have to emulate it, and that's where I always lost
> the desire to do this.
> 
> When waking up this morning though I had the thought that we shouldn't
> try to emulate openat(3p) directly, but instead create a higher-level
> interface.
> [...]

Yeah, I think given a decent interface it might not be so bad. It would mean code thinking about filesystem syscalls in a different way, but if done subsystem-by-subsystem it might be OK to do incrementally. Much of the code that would want to switch to this is using repo_git_path() or similar already (and getting rid of those remaining static-buffer functions would be a nice bonus).

I do wonder if your series here to move to absolute paths makes the TOCTOU situation a little worse. With a relative path, once we are "inside" the repo then we are only susceptible to changes within it. Whereas with an absolute path, if one of the intermediate paths changes from under us, there may be confusion.

Without thinking on it too hard, though, I'd guess if any such case is a security problem, it already was during the "open" part (because it implies that the attacker controls paths below you in the hierarchy, and you had to get to your cwd _somehow_, at which point they could have attacked you then).

-Peff
Previous: Patrick SteinhardtNext: Patrick Steinhardt
Message 22 of 88 in “refs: stop using `chdir_notify_reparent()`”
  1. 0/9 refs: stop using `chdir_notify_reparent()`Patrick Steinhardt, Jun 10, 2026
  2. 1/9 setup: inline `check_and_apply_repository_format()`Patrick Steinhardt, Jun 10, 2026
  3. 2/9 setup: stop applying repository format twicePatrick Steinhardt, Jun 10, 2026
  4. Karthik NayakJun 12, 2026
  5. Patrick SteinhardtJun 15, 2026
  6. 3/9 setup: don't apply "GIT_REFERENCE_BACKEND" without a repositoryPatrick Steinhardt, Jun 10, 2026
  7. Junio C HamanoJun 10, 2026
  8. Patrick SteinhardtJun 12, 2026
  9. 4/9 refs: unregister reference stores from "chdir_notify"Patrick Steinhardt, Jun 10, 2026
  10. Karthik NayakJun 12, 2026
  11. Patrick SteinhardtJun 15, 2026
  12. 5/9 chdir-notify: drop unused `chdir_notify_reparent()`Patrick Steinhardt, Jun 10, 2026
  13. 6/9 repository: free main reference databasePatrick Steinhardt, Jun 10, 2026
  14. Karthik NayakJun 12, 2026
  15. 7/9 refs: fix recursing `get_main_ref_store()` with "onbranch" configPatrick Steinhardt, Jun 10, 2026
  16. 8/9 refs: drop local buffer in `refs_compute_filesystem_location()`Patrick Steinhardt, Jun 10, 2026
  17. 9/9 refs: always use absolute paths for reference storesPatrick Steinhardt, Jun 10, 2026
  18. Karthik NayakJun 12, 2026
  19. Patrick SteinhardtJun 15, 2026
  20. Jeff KingJun 11, 2026
  21. Patrick SteinhardtJun 12, 2026
  22. Jeff KingJun 13, 2026
  23. Patrick SteinhardtJun 15, 2026
  24. 0/8 refs: stop using `chdir_notify_reparent()`Patrick Steinhardt, Jun 15, 2026
  25. 1/8 setup: inline `check_and_apply_repository_format()`Patrick Steinhardt, Jun 15, 2026
  26. 2/8 setup: stop applying repository format twicePatrick Steinhardt, Jun 15, 2026
  27. Justin ToblerJun 17, 2026
  28. 3/8 setup: don't apply "GIT_REFERENCE_BACKEND" without a repositoryPatrick Steinhardt, Jun 15, 2026
  29. Justin ToblerJun 17, 2026
  30. Patrick SteinhardtJun 18, 2026
  31. 4/8 refs: unregister reference stores from "chdir_notify"Patrick Steinhardt, Jun 15, 2026
  32. Justin ToblerJun 17, 2026
  33. Justin ToblerJun 17, 2026
  34. Patrick SteinhardtJun 18, 2026
  35. 5/8 chdir-notify: drop unused `chdir_notify_reparent()`Patrick Steinhardt, Jun 15, 2026
  36. 6/8 repository: free main reference databasePatrick Steinhardt, Jun 15, 2026
  37. Justin ToblerJun 17, 2026
  38. 7/8 refs: fix recursing `get_main_ref_store()` with "onbranch" configPatrick Steinhardt, Jun 15, 2026
  39. Justin ToblerJun 17, 2026
  40. Patrick SteinhardtJun 18, 2026
  41. Justin ToblerJun 18, 2026
  42. Patrick SteinhardtJun 18, 2026
  43. Justin ToblerJun 18, 2026
  44. Jeff KingJun 18, 2026
  45. Patrick SteinhardtJun 19, 2026
  46. Jeff KingJun 21, 2026
  47. Patrick SteinhardtJun 22, 2026
  48. 8/8 refs: drop local buffer in `refs_compute_filesystem_location()`Patrick Steinhardt, Jun 15, 2026
  49. 0/8 refs: stop using `chdir_notify_reparent()`Patrick Steinhardt, Jun 18, 2026
  50. 1/8 setup: inline `check_and_apply_repository_format()`Patrick Steinhardt, Jun 18, 2026
  51. 2/8 setup: stop applying repository format twicePatrick Steinhardt, Jun 18, 2026
  52. 3/8 setup: don't apply "GIT_REFERENCE_BACKEND" without a repositoryPatrick Steinhardt, Jun 18, 2026
  53. 4/8 refs: unregister reference stores from "chdir_notify"Patrick Steinhardt, Jun 18, 2026
  54. 5/8 chdir-notify: drop unused `chdir_notify_reparent()`Patrick Steinhardt, Jun 18, 2026
  55. 6/8 repository: free main reference databasePatrick Steinhardt, Jun 18, 2026
  56. 7/8 refs: fix recursing `get_main_ref_store()` with "onbranch" configPatrick Steinhardt, Jun 18, 2026
  57. 8/8 refs: drop local buffer in `refs_compute_filesystem_location()`Patrick Steinhardt, Jun 18, 2026
  58. 00/10 refs: stop using `chdir_notify_reparent()`Patrick Steinhardt, Jun 19, 2026
  59. 01/10 setup: inline `check_and_apply_repository_format()`Patrick Steinhardt, Jun 19, 2026
  60. 02/10 setup: stop applying repository format twicePatrick Steinhardt, Jun 19, 2026
  61. 03/10 setup: don't apply "GIT_REFERENCE_BACKEND" without a repositoryPatrick Steinhardt, Jun 19, 2026
  62. 04/10 refs: unregister reference stores from "chdir_notify"Patrick Steinhardt, Jun 19, 2026
  63. 05/10 chdir-notify: drop unused `chdir_notify_reparent()`Patrick Steinhardt, Jun 19, 2026
  64. 06/10 repository: free main reference databasePatrick Steinhardt, Jun 19, 2026
  65. 07/10 refs: move parsing of "core.logAllRefUpdates" back into ref storesPatrick Steinhardt, Jun 19, 2026
  66. 08/10 refs/reftable-backend: manually parse "core.sharedRepository"Patrick Steinhardt, Jun 19, 2026
  67. 09/10 refs: fix recursing `get_main_ref_store()` with "onbranch" configPatrick Steinhardt, Jun 19, 2026
  68. 10/10 refs: drop local buffer in `refs_compute_filesystem_location()`Patrick Steinhardt, Jun 19, 2026
  69. 00/11 refs: fix "onbranch" conditionsPatrick Steinhardt, Jun 22, 2026
  70. 01/11 setup: inline `check_and_apply_repository_format()`Patrick Steinhardt, Jun 22, 2026
  71. 02/11 setup: stop applying repository format twicePatrick Steinhardt, Jun 22, 2026
  72. 03/11 setup: don't apply "GIT_REFERENCE_BACKEND" without a repositoryPatrick Steinhardt, Jun 22, 2026
  73. 04/11 refs: unregister reference stores from "chdir_notify"Patrick Steinhardt, Jun 22, 2026
  74. 05/11 chdir-notify: drop unused `chdir_notify_reparent()`Patrick Steinhardt, Jun 22, 2026
  75. 06/11 repository: free main reference databasePatrick Steinhardt, Jun 22, 2026
  76. 07/11 refs: move parsing of "core.logAllRefUpdates" back into ref storesPatrick Steinhardt, Jun 22, 2026
  77. Justin ToblerJun 24, 2026
  78. Patrick SteinhardtJun 25, 2026
  79. 08/11 refs/files: lazy-load configuration to fix chicken-and-eggPatrick Steinhardt, Jun 22, 2026
  80. Justin ToblerJun 24, 2026
  81. Patrick SteinhardtJun 25, 2026
  82. 09/11 reftable: split up write optionsPatrick Steinhardt, Jun 22, 2026
  83. Justin ToblerJun 24, 2026
  84. Patrick SteinhardtJun 25, 2026
  85. 10/11 refs/reftable: lazy-load configuration to fix chicken-and-eggPatrick Steinhardt, Jun 22, 2026
  86. Justin ToblerJun 24, 2026
  87. Patrick SteinhardtJun 25, 2026
  88. 11/11 refs: protect against chicken-and-egg recursionPatrick Steinhardt, Jun 22, 2026

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.