From: "Ævar Arnfjörð Bjarmason" <avarab@gmail.com>
To: git@vger.kernel.org
Cc: "Junio C Hamano" <gitster@pobox.com>, "Jeff King" <peff@peff.net>,
"Taylor Blau" <me@ttaylorr.com>,
"Ævar Arnfjörð Bjarmason" <avarab@gmail.com>
Subject: Re: [PATCH v2 4/4] pack-write: rename *.idx file into place last (really!)
Date: Wed, 08 Sep 2021 03:14:12 +0200 [thread overview]
Message-ID: <87lf47kgkh.fsf@evledraar.gmail.com> (raw)
In-Reply-To: <patch-v2-4.4-70f4a9767d-20210908T003631Z-avarab@gmail.com>
On Wed, Sep 08 2021, Ævar Arnfjörð Bjarmason wrote:
> Follow-up a preceding commit (pack-write.c: rename `.idx` file into
> place last, 2021-08-16)[1] and rename the *.idx file in-place after we
> write the *.bitmap. The preceding commit fixed the issue of *.idx
> being written before *.rev files, but did not do so for *.idx files.
>
> See 7cc8f971085 (pack-objects: implement bitmap writing, 2013-12-21)
> for commentary at the time when *.bitmap was implemented about how
> those files are written out, nothing in that commit contradicts what's
> being done here.
>
> Note that the referenced earlier commit[1] is overly optimistic about
> "clos[ing the] race", i.e. yes we'll now write the files in the right
> order, but we might still race due to our sloppy use of fsync(). See
> the thread at [2] for a rabbit hole of various discussions about
> filesystem races in the face of doing and not doing fsync() (and if
> doing fsync(), not doing it properly).
Actually I think it's a bit worse than that, we will unconditionally
fsync() the *.pack we write out, but in stage_tmp_packfiles() (the
behavior pre-dates both this series and its parent, I just think my
stage_tmp_packfiles() is easier to follow) we'll not write the *.idx
file with fsync() since we won't pass WRITE_IDX_VERIFY.
The same goes for *.rev (which oddly makes its fsync() conditional on
WRITE_IDX_VERIFY), but not *.bitmap, which fsyncs unconditionally just
like *.pack does.
And then of course we'll do all these in-place renames but nothing
fsyncs the fd of the directory, so the metadata and new names being
committed to disk & visible to other processes is anyone's guess.
But not only is that metadata commit not made, but due to the
inconsistent fsync() we might end up with an *.idx that's partial and
renamed in-place.
In any case, any such issues pre-date this series and the series by
Taylor it depends on, just adding some #leftoverbits for future fsync()
fixes since I spent time looking into it.
next prev parent reply other threads:[~2021-09-08 1:21 UTC|newest]
Thread overview: 81+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-09-01 2:05 [PATCH 0/2] pack-write,repack: prevent opening packs too early Taylor Blau
2021-09-01 2:06 ` [PATCH 1/2] pack-write.c: rename `.idx` file into place last Taylor Blau
2021-09-01 2:06 ` [PATCH 2/2] builtin/repack.c: move `.idx` files " Taylor Blau
2021-09-01 3:53 ` [PATCH 0/2] pack-write,repack: prevent opening packs too early Jeff King
2021-09-01 4:29 ` Taylor Blau
2021-09-01 4:59 ` Jeff King
2021-09-01 5:12 ` Taylor Blau
2021-09-01 6:08 ` Jeff King
2021-09-01 21:40 ` Taylor Blau
2021-09-07 16:07 ` Jeff King
2021-09-07 19:42 ` [PATCH 0/3] rename *.idx file into place last (also after *.bitmap) Ævar Arnfjörð Bjarmason
2021-09-07 19:42 ` [PATCH 1/3] pack-write: use more idiomatic strbuf usage for packname construction Ævar Arnfjörð Bjarmason
2021-09-07 22:21 ` Taylor Blau
2021-09-07 23:22 ` Ævar Arnfjörð Bjarmason
2021-09-07 19:42 ` [PATCH 2/3] pack-write: split up finish_tmp_packfile() function Ævar Arnfjörð Bjarmason
2021-09-07 22:28 ` Taylor Blau
2021-09-07 19:42 ` [PATCH 3/3] pack-write: rename *.idx file into place last (really!) Ævar Arnfjörð Bjarmason
2021-09-07 22:31 ` Taylor Blau
2021-09-07 22:36 ` [PATCH 0/3] rename *.idx file into place last (also after *.bitmap) Taylor Blau
2021-09-07 19:48 ` [PATCH 0/2] pack-write,repack: prevent opening packs too early Ævar Arnfjörð Bjarmason
2021-09-08 0:38 ` [PATCH v2 0/4] rename *.idx file into place last (also after *.bitmap) Ævar Arnfjörð Bjarmason
2021-09-08 0:38 ` [PATCH v2 1/4] pack.h: line-wrap the definition of finish_tmp_packfile() Ævar Arnfjörð Bjarmason
2021-09-08 0:38 ` [PATCH v2 2/4] pack-write: refactor renaming in finish_tmp_packfile() Ævar Arnfjörð Bjarmason
2021-09-08 4:22 ` Taylor Blau
2021-09-08 0:38 ` [PATCH v2 3/4] pack-write: split up finish_tmp_packfile() function Ævar Arnfjörð Bjarmason
2021-09-08 0:38 ` [PATCH v2 4/4] pack-write: rename *.idx file into place last (really!) Ævar Arnfjörð Bjarmason
2021-09-08 1:14 ` Ævar Arnfjörð Bjarmason [this message]
2021-09-08 9:18 ` Ævar Arnfjörð Bjarmason
2021-09-08 4:24 ` Taylor Blau
2021-09-08 22:17 ` [PATCH v2 0/3] prevent opening packs too early Taylor Blau
2021-09-08 22:17 ` [PATCH v2 1/3] pack-write.c: rename `.idx` files into place last Taylor Blau
2021-09-08 22:17 ` [PATCH v2 2/3] builtin/repack.c: move " Taylor Blau
2021-09-08 22:17 ` [PATCH v2 3/3] builtin/index-pack.c: " Taylor Blau
2021-09-08 23:52 ` [PATCH v2 0/3] prevent opening packs too early Ævar Arnfjörð Bjarmason
2021-09-09 0:50 ` Ævar Arnfjörð Bjarmason
2021-09-09 1:13 ` Taylor Blau
2021-09-09 1:33 ` Ævar Arnfjörð Bjarmason
2021-09-09 2:36 ` Ævar Arnfjörð Bjarmason
2021-09-09 2:49 ` Taylor Blau
2021-09-09 3:24 ` [PATCH 0/9] packfile: avoid .idx rename races Taylor Blau
2021-09-09 3:24 ` [PATCH 1/9] pack.h: line-wrap the definition of finish_tmp_packfile() Taylor Blau
2021-09-09 3:24 ` [PATCH 2/9] bulk-checkin.c: store checksum directly Taylor Blau
2021-09-09 7:38 ` Ævar Arnfjörð Bjarmason
2021-09-09 3:24 ` [PATCH 3/9] pack-write: refactor renaming in finish_tmp_packfile() Taylor Blau
2021-09-09 19:29 ` Junio C Hamano
2021-09-09 21:07 ` Taylor Blau
2021-09-09 23:30 ` Junio C Hamano
2021-09-09 23:31 ` Taylor Blau
2021-09-10 1:29 ` Junio C Hamano
2021-09-09 3:24 ` [PATCH 4/9] pack-write.c: rename `.idx` files after `*.rev` Taylor Blau
2021-09-09 7:46 ` Ævar Arnfjörð Bjarmason
2021-09-09 14:37 ` Taylor Blau
2021-09-09 19:32 ` Junio C Hamano
2021-09-09 3:25 ` [PATCH 5/9] builtin/repack.c: move `.idx` files into place last Taylor Blau
2021-09-09 19:38 ` Junio C Hamano
2021-09-09 21:08 ` Taylor Blau
2021-09-09 3:25 ` [PATCH 6/9] index-pack: refactor renaming in final() Taylor Blau
2021-09-09 19:45 ` Junio C Hamano
2021-09-09 21:11 ` Taylor Blau
2021-09-09 3:25 ` [PATCH 7/9] builtin/index-pack.c: move `.idx` files into place last Taylor Blau
2021-09-09 7:52 ` Ævar Arnfjörð Bjarmason
2021-09-09 19:45 ` Junio C Hamano
2021-09-09 3:25 ` [PATCH 8/9] pack-write: split up finish_tmp_packfile() function Taylor Blau
2021-09-09 3:25 ` [PATCH 9/9] pack-objects: rename .idx files into place after .bitmap files Taylor Blau
2021-09-09 7:54 ` Ævar Arnfjörð Bjarmason
2021-09-09 19:52 ` Junio C Hamano
2021-09-09 21:13 ` Taylor Blau
2021-09-09 8:06 ` [PATCH 0/9] packfile: avoid .idx rename races Ævar Arnfjörð Bjarmason
2021-09-09 14:40 ` Taylor Blau
2021-09-09 19:52 ` Junio C Hamano
2021-09-09 23:24 ` [PATCH v2 " Taylor Blau
2021-09-09 23:24 ` [PATCH v2 1/9] pack.h: line-wrap the definition of finish_tmp_packfile() Taylor Blau
2021-09-09 23:24 ` [PATCH v2 2/9] bulk-checkin.c: store checksum directly Taylor Blau
2021-09-09 23:24 ` [PATCH v2 3/9] pack-write: refactor renaming in finish_tmp_packfile() Taylor Blau
2021-09-09 23:24 ` [PATCH v2 4/9] pack-write.c: rename `.idx` files after `*.rev` Taylor Blau
2021-09-09 23:24 ` [PATCH v2 5/9] builtin/repack.c: move `.idx` files into place last Taylor Blau
2021-09-09 23:24 ` [PATCH v2 6/9] index-pack: refactor renaming in final() Taylor Blau
2021-09-09 23:24 ` [PATCH v2 7/9] builtin/index-pack.c: move `.idx` files into place last Taylor Blau
2021-09-09 23:24 ` [PATCH v2 8/9] pack-write: split up finish_tmp_packfile() function Taylor Blau
2021-09-09 23:25 ` [PATCH v2 9/9] pack-objects: rename .idx files into place after .bitmap files Taylor Blau
2021-09-10 1:35 ` [PATCH v2 0/9] packfile: avoid .idx rename races Junio C Hamano
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
List information: http://vger.kernel.org/majordomo-info.html
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=87lf47kgkh.fsf@evledraar.gmail.com \
--to=avarab@gmail.com \
--cc=git@vger.kernel.org \
--cc=gitster@pobox.com \
--cc=me@ttaylorr.com \
--cc=peff@peff.net \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
Code repositories for project(s) associated with this public inbox
https://80x24.org/mirrors/git.git
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).