From: Dmitry Potapov <dpotapov@gmail.com>
To: Zygo Blaxell <zblaxell@esightcorp.com>
Cc: Ilari Liusvaara <ilari.liusvaara@elisanet.fi>,
Thomas Rast <trast@student.ethz.ch>,
Jonathan Nieder <jrnieder@gmail.com>,
git@vger.kernel.org
Subject: [PATCH] don't use mmap() to hash files
Date: Sun, 14 Feb 2010 04:18:12 +0300 [thread overview]
Message-ID: <20100214011812.GA2175@dpotapov.dyndns.org> (raw)
In-Reply-To: <20100213223733.GP24809@gibbs.hungrycats.org>
If a mmapped file is changed by another program during git-add, it
causes the repository corruption. Disabling mmap() in index_fd() does
not have any negative impact on the overall speed of Git. In fact, it
makes git hash-object to work slightly faster. Here is the best result
before and after patch based on 5 runs on the Linix kernel repository:
Before:
$ git ls-files | time git hash-object --stdin-path > /dev/null
2.15user 0.36system 0:02.52elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+103248minor)pagefaults 0swaps
After:
$ git ls-files | time ../git/git hash-object --stdin-path > /dev/null
2.09user 0.33system 0:02.42elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+1073minor)pagefaults 0swaps
Signed-off-by: Dmitry Potapov <dpotapov@gmail.com>
---
I think more people should test this change to see its impact on
performance. For me, it was positive. Here is my results:
Using mmap (git version 1.7.0)
2.18user 0.33system 0:02.52elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+103248minor)pagefaults 0swaps
$ git ls-files | time git hash-object --stdin-path > /dev/null
2.23user 0.28system 0:02.53elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+103249minor)pagefaults 0swaps
$ git ls-files | time git hash-object --stdin-path > /dev/null
2.20user 0.31system 0:02.52elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+103248minor)pagefaults 0swaps
$ git ls-files | time git hash-object --stdin-path > /dev/null
2.21user 0.30system 0:02.51elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+103248minor)pagefaults 0swaps
$ git ls-files | time git hash-object --stdin-path > /dev/null
2.15user 0.36system 0:02.52elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+103248minor)pagefaults 0swaps
Using read() instead of mmap() (1.7.0 with the above patch)
$ git ls-files | time ../git/git hash-object --stdin-path > /dev/null
2.19user 0.24system 0:02.42elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+1073minor)pagefaults 0swaps
$ git ls-files | time ../git/git hash-object --stdin-path > /dev/null
2.15user 0.26system 0:02.42elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+1073minor)pagefaults 0swaps
$ git ls-files | time ../git/git hash-object --stdin-path > /dev/null
2.18user 0.24system 0:02.42elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+1073minor)pagefaults 0swaps
$ git ls-files | time ../git/git hash-object --stdin-path > /dev/null
2.18user 0.25system 0:02.42elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+1073minor)pagefaults 0swaps
$ git ls-files | time ../git/git hash-object --stdin-path > /dev/null
2.09user 0.33system 0:02.42elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+1073minor)pagefaults 0swaps
sha1_file.c | 22 +++++++---------------
1 files changed, 7 insertions(+), 15 deletions(-)
diff --git a/sha1_file.c b/sha1_file.c
index 657825e..83f82a2 100644
--- a/sha1_file.c
+++ b/sha1_file.c
@@ -2438,22 +2438,14 @@ int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
enum object_type type, const char *path)
{
int ret;
- size_t size = xsize_t(st->st_size);
+ struct strbuf sbuf = STRBUF_INIT;
- if (!S_ISREG(st->st_mode)) {
- struct strbuf sbuf = STRBUF_INIT;
- if (strbuf_read(&sbuf, fd, 4096) >= 0)
- ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
- type, path);
- else
- ret = -1;
- strbuf_release(&sbuf);
- } else if (size) {
- void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
- ret = index_mem(sha1, buf, size, write_object, type, path);
- munmap(buf, size);
- } else
- ret = index_mem(sha1, NULL, size, write_object, type, path);
+ if (strbuf_read(&sbuf, fd, 4096) >= 0)
+ ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
+ type, path);
+ else
+ ret = -1;
+ strbuf_release(&sbuf);
close(fd);
return ret;
}
--
1.7.0
next prev parent reply other threads:[~2010-02-14 1:23 UTC|newest]
Thread overview: 84+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <20100211234753.22574.48799.reportbug@gibbs.hungrycats.org>
2010-02-12 0:27 ` Bug#569505: git-core: 'git add' corrupts repository if the working directory is modified as it runs Jonathan Nieder
2010-02-12 1:23 ` Zygo Blaxell
2010-02-13 12:12 ` Jonathan Nieder
2010-02-13 13:39 ` Ilari Liusvaara
2010-02-13 14:39 ` Thomas Rast
2010-02-13 16:29 ` Ilari Liusvaara
2010-02-13 22:09 ` Dmitry Potapov
2010-02-13 22:37 ` Zygo Blaxell
2010-02-14 1:18 ` Dmitry Potapov [this message]
2010-02-14 1:37 ` [PATCH] don't use mmap() to hash files Junio C Hamano
2010-02-14 2:18 ` Dmitry Potapov
2010-02-14 3:14 ` Junio C Hamano
2010-02-14 11:14 ` Thomas Rast
2010-02-14 11:46 ` Junio C Hamano
2010-02-14 1:53 ` Johannes Schindelin
2010-02-14 2:00 ` Junio C Hamano
2010-02-14 2:42 ` Dmitry Potapov
2010-02-14 11:07 ` Jakub Narebski
2010-02-14 11:55 ` Paolo Bonzini
2010-02-14 18:10 ` Johannes Schindelin
2010-02-14 19:06 ` Dmitry Potapov
2010-02-14 19:22 ` Johannes Schindelin
2010-02-14 19:28 ` Johannes Schindelin
2010-02-14 19:56 ` Dmitry Potapov
2010-02-14 23:52 ` Zygo Blaxell
2010-02-15 5:05 ` Nicolas Pitre
2010-02-15 12:23 ` Dmitry Potapov
2010-02-15 7:48 ` Paolo Bonzini
2010-02-15 12:25 ` Dmitry Potapov
2010-02-14 19:55 ` Dmitry Potapov
2010-02-14 23:13 ` Avery Pennarun
2010-02-15 4:16 ` Nicolas Pitre
2010-02-15 5:01 ` Avery Pennarun
2010-02-15 5:48 ` Nicolas Pitre
2010-02-15 19:19 ` Avery Pennarun
2010-02-15 19:29 ` Nicolas Pitre
2010-02-14 3:05 ` [PATCH v2] " Dmitry Potapov
2010-02-18 1:16 ` [PATCH] Teach "git add" and friends to be paranoid Junio C Hamano
2010-02-18 1:20 ` Junio C Hamano
2010-02-18 15:32 ` Zygo Blaxell
2010-02-19 17:51 ` Junio C Hamano
2010-02-18 1:38 ` Jeff King
2010-02-18 4:55 ` Nicolas Pitre
2010-02-18 5:36 ` Junio C Hamano
2010-02-18 7:27 ` Wincent Colaiuta
2010-02-18 16:18 ` Zygo Blaxell
2010-02-18 18:12 ` Jonathan Nieder
2010-02-18 18:35 ` Junio C Hamano
2010-02-22 12:59 ` Paolo Bonzini
2010-02-22 13:33 ` Dmitry Potapov
2010-02-18 10:14 ` Thomas Rast
2010-02-18 18:16 ` Junio C Hamano
2010-02-18 19:58 ` Nicolas Pitre
2010-02-18 20:11 ` 16 gig, 350,000 file repository Bill Lear
2010-02-18 20:58 ` Nicolas Pitre
2010-02-19 9:27 ` Erik Faye-Lund
2010-02-22 22:20 ` Bill Lear
2010-02-22 22:31 ` Nicolas Pitre
2010-02-18 20:14 ` [PATCH] Teach "git add" and friends to be paranoid Peter Harris
2010-02-18 20:17 ` Junio C Hamano
2010-02-18 21:30 ` Nicolas Pitre
2010-02-19 1:04 ` Jonathan Nieder
2010-02-19 15:26 ` Zygo Blaxell
2010-02-19 17:52 ` Junio C Hamano
2010-02-19 19:08 ` Zygo Blaxell
2010-02-19 8:28 ` Dmitry Potapov
2010-02-19 17:52 ` Junio C Hamano
2010-02-20 19:23 ` Junio C Hamano
2010-02-21 7:21 ` Dmitry Potapov
2010-02-21 19:32 ` Junio C Hamano
2010-02-22 3:35 ` Dmitry Potapov
2010-02-22 6:59 ` Junio C Hamano
2010-02-22 12:25 ` Dmitry Potapov
2010-02-22 15:40 ` Nicolas Pitre
2010-02-22 16:01 ` Dmitry Potapov
2010-02-22 17:31 ` Zygo Blaxell
2010-02-22 18:01 ` Nicolas Pitre
2010-02-22 19:56 ` Junio C Hamano
2010-02-22 20:52 ` Nicolas Pitre
2010-02-22 18:05 ` Dmitry Potapov
2010-02-22 18:14 ` Nicolas Pitre
2010-02-14 1:36 ` mmap with MAP_PRIVATE is useless (was Re: Bug#569505: git-core: 'git add' corrupts repository if the working directory is modified as it runs) Paolo Bonzini
2010-02-14 1:53 ` mmap with MAP_PRIVATE is useless Junio C Hamano
2010-02-14 2:11 ` Paolo Bonzini
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=20100214011812.GA2175@dpotapov.dyndns.org \
--to=dpotapov@gmail.com \
--cc=git@vger.kernel.org \
--cc=ilari.liusvaara@elisanet.fi \
--cc=jrnieder@gmail.com \
--cc=trast@student.ethz.ch \
--cc=zblaxell@esightcorp.com \
/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).