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

Re: [PATCH v5 18/35] commit_lock_file(): if close fails, roll back

From
Michael Haggerty <mhagger@alum.mit.edu>
Date
Sep 23, 2014, 11:30 UTC
Message-ID
<542159CF.7020105@alum.mit.edu>
In-Reply-To
<20140916221956.GR29050@google.com>
On 09/17/2014 12:19 AM, Jonathan Nieder wrote:
Show 8 quoted lines
> Michael Haggerty wrote:
> 
>> If closing an open lockfile fails, then we cannot be sure of the
>> contents of the lockfile
> 
> Is that true?  It seems more like a bug in close_lock_file: if it
> fails, perhaps it should either set lk->fd back to fd or unlink the
> lockfile itself.
>From close(2) (note especially the last sentence):
Show 7 quoted lines
> Not checking the return value of close() is a common but nevertheless serious  programming
> error.   It  is  quite  possible  that  errors  on a previous write(2) operation are first
> reported at the final close().  Not checking the return value when closing  the  file  may
> lead  to  silent  loss  of  data.   This can especially be observed with NFS and with disk
> quota.  Note that the return value should only be used  for  diagnostics.   In  particular
> close() should not be retried after an EINTR since this may cause a reused descriptor from
> another thread to be closed.
>From that I conclude that if close() fails, pretty much all bets are off

about the contents of the lockfile. So we wouldn't want to set lk->fd back to fd.

But finishing the rollback itself might be an alternative...
> What do other callers do on close_lock_file failure?
>From what I can see, the only callers that don't die() immediately are

the following (which call close_lock_file() directly or indirectly via write_locked_index()):

try_merge_strategy(): returns an error. It looks like this could end up reusing the same lock_file object before it had been rolled back -> would be improved if close_lock_file() would rollback on failure.

write_cache_as_tree(): does a rollback -> wouldn't mind an automatic rollback.

merge_recursive_generic(): returns an error, and caller exits immediately -> wouldn't mind an automatic rollback.

So, I will change close_lock_file() to roll back on errors.
Michael
-- 
Michael Haggerty
mhagger@alum.mit.edu
Previous: Jonathan NiederNext: Michael Haggerty
Message 47 of 73 in “Lockfile correctness and refactoring”
  1. 00/35 Lockfile correctness and refactoringMichael Haggerty, Sep 16, 2014
  2. 01/35 unable_to_lock_die(): rename function from unable_to_lock_index_die()Michael Haggerty, Sep 16, 2014
  3. Jonathan NiederSep 16, 2014
  4. 02/35 api-lockfile: expand the documentationMichael Haggerty, Sep 16, 2014
  5. Jonathan NiederSep 16, 2014
  6. Michael HaggertySep 22, 2014
  7. 03/35 rollback_lock_file(): do not clear filename redundantlyMichael Haggerty, Sep 16, 2014
  8. Jonathan NiederSep 16, 2014
  9. 04/35 rollback_lock_file(): exit early if lock is not activeMichael Haggerty, Sep 16, 2014
  10. Jonathan NiederSep 16, 2014
  11. 05/35 rollback_lock_file(): set fd to -1Michael Haggerty, Sep 16, 2014
  12. Jonathan NiederSep 16, 2014
  13. Jonathan NiederSep 16, 2014
  14. Michael HaggertySep 17, 2014
  15. 06/35 lockfile: unlock file if lockfile permissions cannot be adjustedMichael Haggerty, Sep 16, 2014
  16. Jonathan NiederSep 16, 2014
  17. 07/35 hold_lock_file_for_append(): release lock on errorsMichael Haggerty, Sep 16, 2014
  18. Jonathan NiederSep 16, 2014
  19. Michael HaggertySep 17, 2014
  20. 08/35 lock_file(): always add lock_file object to lock_file_listMichael Haggerty, Sep 16, 2014
  21. Jonathan NiederSep 16, 2014
  22. Michael HaggertySep 17, 2014
  23. Torsten BögershausenSep 18, 2014
  24. Michael HaggertySep 18, 2014
  25. 09/35 lockfile.c: document the various states of lock_file objectsMichael Haggerty, Sep 16, 2014
  26. Jonathan NiederSep 16, 2014
  27. Michael HaggertySep 22, 2014
  28. Jonathan NiederSep 22, 2014
  29. 10/35 cache.h: define constants LOCK_SUFFIX and LOCK_SUFFIX_LENMichael Haggerty, Sep 16, 2014
  30. Jonathan NiederSep 16, 2014
  31. Michael HaggertySep 22, 2014
  32. 11/35 delete_ref_loose(): don't muck around in the lock_file's filenameMichael Haggerty, Sep 16, 2014
  33. Jonathan NiederSep 16, 2014
  34. 12/35 prepare_index(): declare return value to be (const char *)Michael Haggerty, Sep 16, 2014
  35. Jonathan NiederSep 16, 2014
  36. 13/35 write_packed_entry_fn(): convert cb_data into a (const int *)Michael Haggerty, Sep 16, 2014
  37. 14/35 lock_file(): exit early if lockfile cannot be openedMichael Haggerty, Sep 16, 2014
  38. Jonathan NiederSep 16, 2014
  39. 15/35 remove_lock_file(): call rollback_lock_file()Michael Haggerty, Sep 16, 2014
  40. Jonathan NiederSep 16, 2014
  41. 16/35 commit_lock_file(): inline temporary variableMichael Haggerty, Sep 16, 2014
  42. Jonathan NiederSep 16, 2014
  43. 17/35 commit_lock_file(): die() if called for unlocked lockfile objectMichael Haggerty, Sep 16, 2014
  44. Jonathan NiederSep 16, 2014
  45. 18/35 commit_lock_file(): if close fails, roll backMichael Haggerty, Sep 16, 2014
  46. Jonathan NiederSep 16, 2014
  47. Michael HaggertySep 23, 2014
  48. 19/35 commit_lock_file(): rollback lock file on failure to renameMichael Haggerty, Sep 16, 2014
  49. Jonathan NiederSep 16, 2014
  50. Michael HaggertySep 23, 2014
  51. 20/35 api-lockfile: document edge casesMichael Haggerty, Sep 16, 2014
  52. Jonathan NiederSep 16, 2014
  53. Michael HaggertySep 23, 2014
  54. 21/35 dump_marks(): remove a redundant call to rollback_lock_file()Michael Haggerty, Sep 16, 2014
  55. Jonathan NiederSep 16, 2014
  56. 22/35 git_config_set_multivar_in_file(): avoid call to rollback_lock_file()Michael Haggerty, Sep 16, 2014
  57. Jonathan NiederSep 16, 2014
  58. Michael HaggertySep 23, 2014
  59. 23/35 lockfile: avoid transitory invalid statesMichael Haggerty, Sep 16, 2014
  60. Jonathan NiederSep 16, 2014
  61. Michael HaggertySep 23, 2014
  62. 24/35 struct lock_file: declare some fields volatileMichael Haggerty, Sep 16, 2014
  63. 25/35 try_merge_strategy(): remove redundant lock_file allocationMichael Haggerty, Sep 16, 2014
  64. 26/35 try_merge_strategy(): use a statically-allocated lock_file objectMichael Haggerty, Sep 16, 2014
  65. 27/35 commit_lock_file(): use a strbuf to manage temporary spaceMichael Haggerty, Sep 16, 2014
  66. 28/35 Change lock_file::filename into a strbufMichael Haggerty, Sep 16, 2014
  67. 29/35 resolve_symlink(): use a strbuf for internal scratch spaceMichael Haggerty, Sep 16, 2014
  68. 30/35 resolve_symlink(): take a strbuf parameterMichael Haggerty, Sep 16, 2014
  69. 31/35 trim_last_path_component(): replace last_path_elm()Michael Haggerty, Sep 16, 2014
  70. 32/35 Extract a function commit_lock_file_to()Michael Haggerty, Sep 16, 2014
  71. 33/35 Rename LOCK_NODEREF to LOCK_NO_DEREFMichael Haggerty, Sep 16, 2014
  72. 34/35 lockfile.c: rename static functionsMichael Haggerty, Sep 16, 2014
  73. 35/35 get_locked_file_path(): new functionMichael Haggerty, Sep 16, 2014

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.