[PATCH] Optimisation for readdir using fstatat.
Jeremy Allison
jra at samba.org
Fri May 24 14:37:35 MDT 2013
On Fri, May 24, 2013 at 01:31:17PM -0700, Jeremy Allison wrote:
> On Fri, May 24, 2013 at 01:23:20PM -0700, Jeremy Allison wrote:
> > On Fri, May 24, 2013 at 04:19:59PM -0400, Ira Cooper wrote:
> > > A concern raised by Richard, which I haven't had the chance to look at... Is
> > > the use of "stat" here premature?
> >
> > No.
> >
> > > (For every readdir, are we guaranteed to stat, and not lstat etc...)
> >
> > If we pass in a SMB_STRUCT_STAT *sbuf then yes, we are
> > guaranteed to stat(). If it's a POSIX path then we will
> > ignore the reply from readdir() here and re-stat using
> > lstat() on top (see the comment in the patch). I checked
> > out all codepaths before adding this.
>
> If you're really worried about this I can add a:
>
> int flags = (lp_posix_pathnames() ? AT_SYMLINK_NOFOLLOW : 0);
>
> arg to the fstatat() call just to make sure we can't
> mess up here.
Updated for taste :-). Bless whichever version you're
happy with :-).
Jeremy.
-------------- next part --------------
From a1044b693603e24c86efb5384172abb7727abd3a Mon Sep 17 00:00:00 2001
From: Jeremy Allison <jra at samba.org>
Date: Fri, 24 May 2013 10:33:16 -0700
Subject: [PATCH 1/2] Add HAVE_DIRFD to waf/autoconf. Detect fstatat.
Signed-off-by: Jeremy Allison <jra at samba.org>
---
source3/configure.in | 6 ++++++
source3/wscript | 4 ++++
2 files changed, 10 insertions(+)
diff --git a/source3/configure.in b/source3/configure.in
index 0e2f126..b7a9620 100644
--- a/source3/configure.in
+++ b/source3/configure.in
@@ -834,9 +834,15 @@ fi
AC_CHECK_FUNCS(dirfd)
if test x"$ac_cv_func_dirfd" = x"yes"; then
+ AC_DEFINE(HAVE_DIRFD,1,[Whether we have dirfd])
default_shared_modules="$default_shared_modules vfs_syncops vfs_dirsort"
fi
+AC_CHECK_FUNCS(fstatat)
+if test x"$ac_cv_func_fstatat" = x"yes"; then
+ AC_DEFINE(HAVE_FSTATAT,1,[Whether we have fstatat])
+fi
+
AC_CACHE_CHECK([for struct sigevent type],samba_cv_struct_sigevent, [
AC_TRY_COMPILE([
#include <sys/types.h>
diff --git a/source3/wscript b/source3/wscript
index dba6cdc..92207ef 100644
--- a/source3/wscript
+++ b/source3/wscript
@@ -452,6 +452,10 @@ return acl_get_perm_np(permset_d, perm);
if conf.CHECK_FUNCS('dirfd'):
conf.DEFINE('HAVE_DIRFD_DECL', 1)
+ conf.DEFINE('HAVE_DIRFD', 1)
+
+ if conf.CHECK_FUNCS('fstatat'):
+ conf.DEFINE('HAVE_fstatat', 1)
conf.CHECK_CODE('struct statfs fsd; fsid_t fsid = fsd.f_fsid; return statfs(".", &fsd);',
'HAVE_STATFS_F_FSID',
--
1.8.2.1
From fb7da6aa1456b47c8b80861dc5d11ebd9d9bbb6f Mon Sep 17 00:00:00 2001
From: Jeremy Allison <jra at samba.org>
Date: Fri, 24 May 2013 10:33:38 -0700
Subject: [PATCH 2/2] Optimization on POSIX platforms that have fstatat.
Tests show significant speedup in directory listings
by using fstatat instead of a full pathname walk.
Signed-off-by: Jeremy Allison <jra at samba.org>
---
source3/modules/vfs_default.c | 27 +++++++++++++++++++++++----
1 file changed, 23 insertions(+), 4 deletions(-)
diff --git a/source3/modules/vfs_default.c b/source3/modules/vfs_default.c
index 8804e62..82d059c 100644
--- a/source3/modules/vfs_default.c
+++ b/source3/modules/vfs_default.c
@@ -376,11 +376,30 @@ static struct dirent *vfswrap_readdir(vfs_handle_struct *handle,
START_PROFILE(syscall_readdir);
result = readdir(dirp);
- /* Default Posix readdir() does not give us stat info.
- * Set to invalid to indicate we didn't return this info. */
- if (sbuf)
- SET_STAT_INVALID(*sbuf);
END_PROFILE(syscall_readdir);
+ if (sbuf) {
+ /* Default Posix readdir() does not give us stat info.
+ * Set to invalid to indicate we didn't return this info. */
+ SET_STAT_INVALID(*sbuf);
+#if defined(HAVE_DIRFD) && defined(HAVE_FSTATAT)
+ if (result != NULL) {
+ /* See if we can efficiently return this. */
+ struct stat st;
+ int flags = (lp_posix_pathnames() ?
+ AT_SYMLINK_NOFOLLOW : 0);
+ int ret = fstatat(dirfd(dirp),
+ result->d_name,
+ &st,
+ flags);
+ if (ret == 0) {
+ init_stat_ex_from_stat(sbuf,
+ &st,
+ lp_fake_dir_create_times(
+ SNUM(handle->conn)));
+ }
+ }
+#endif
+ }
return result;
}
--
1.8.2.1
More information about the samba-technical
mailing list