lkml.org 
[lkml]   [2003]   [Aug]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectFile system synchronous completion poll.
Hi,

it's time again to poll for file system semantics with respect to
synchronous completion, for MTA use.

Define: directory operations := link, rename, mkdir (symlink, rmdir, unlink)

Assume the drive's write cache is off (if your file system is safe with
write cache on, because it queues ordered tags on SCSI, please state
so):

For each of XFS, JFS, ext3fs, ext2fs, reiserfs-3.6, reiserfs-4, for
both of 2.4.21 and 2.6.0-testX, please state (rationale below):

1. Is mount -o sync implemented?
If not, does mount -o sync fail?
If yes, does mount -o sync mean a) "all operations synchronous" or
b) "directory operations" synchronous?

2. Is mount -o dirsync implemented?

3. Is chattr +S implemented, and inherited by freshly-created subdirectories?

4. Is chattr +D implemented, and inherited by freshly-created subdirectories?

5. Does
fd=open("a", ...); rename("a", "b"); fsync(fd); close(fd);
guarantee that "b" is on disk after fsync() has returned?

6. Which code fragment makes sure a directory is on disk?
a. mkdir("parent/sub"); fd=open("parent"); fsync(fd);
b. mkdir("parent/sub"); fd=open("parent/sub"); fsync(fd);
This question may not be relevant if the system supports -o sync or
-o dirsync.

7. Is rename("a", "b") truly atomic or can it cause a file to either
end up in lost+found or be there as "a" and "b" in a crash?

The results should be recorded (with date of finding) in man mount(8).

I am aware that XFS, reiserfs and ext3fs as of Linux 2.4 commit all
pending transactions within a file system when an fsync() is requested,
however, this isn't always sufficient (particularly in the mkdir case,
what do you open to fsync it?).

The rationale is to find out how much Linux file systems differ from BSD
UFS + softupdates code, and how well which file system suits the
opposite of the common tuning (data base): mail transfer agent. Their
authors are still reluctant to add explicit fd=open("directory");
fsync(fd) code and assume traditional BSD "noasync" (= dirsync)
semantics.

--
Matthias Andree
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:47    [W:0.035 / U:0.840 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site