Skip to contentSkip to conversation

sendmail "cannot fsync directory" after 2.4 -> 2.6

DISCUSSION Frank Steiner 3 participants 6 messages Sep 9, 2004 – Sep 10, 2004

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi,

we've been running SuSE 9.0 with kernel 2.4.21 for some months. Two weeks ago we switched to 2.6.8.1 from kernel.org.

Our diskless clients get (due to lack of a local hd) everything via nfs, including /var/spool/mqueue. The /var directories for the clients (one seperate for each client) are exported with "sync" and mounted with "tcp,hard,intr,lock,sync", and this has not changed when switching from 2.4 to 2.6. But after we switched to 2.6.8.1, we get messages

Sep  9 08:03:13 galois sendmail[21596]: sync_dir: cannot fsync directory .: Invalid argument
Sep  9 08:04:02 galois sendmail[21649]: i89641NZ021648: SYSERR(root): sync_dir: cannot fsync directory .: Invalid argument

every time we send a mail with "mail -s test root" or similar. Note that sendmail is running as daemon on every client, but only listening on localhost. The sendmail package has REQUIRES_DIR_FSYNC enabled.

I tried mounting the nfs dir with "dirsync" additionally, but it doesn't help. Also, it doesn't make a difference if the exported fs on the server is ext3 or reiserfs. Since it works on local disks and worked also over NFS with kernel 2.4.21, I guess it is a bug in 2.6? Or should it not work with nfs and 2.4 just didn't notice?

cu,
Frank

Signature · 15 lines
Dipl.-Inform. Frank Steiner   Web:  http://www.bio.ifi.lmu.de/~steiner/
Lehrstuhl f. Bioinformatik    Mail: http://www.bio.ifi.lmu.de/~steiner/m/
LMU, Amalienstr. 17           Phone: +49 89 2180-4049
80333 Muenchen, Germany       Fax:   +49 89 2180-99-4049
* Rekursion kann man erst verstehen, wenn man Rekursion verstanden hat. *


-------------------------------------------------------
This SF.Net email is sponsored by BEA Weblogic Workshop
FREE Java Enterprise J2EE developer tools!
Get your free copy of BEA WebLogic Workshop 8.1 today.
http://ads.osdn.com/?ad_id=5047&alloc_id=10808&op=click
_______________________________________________
NFS maillist  -  [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
#1 lore raw

I wrote a little c program to test fsync on nfs directories (must be called in the directory you want to check):

#include <fcntl.h>

int main() {
int dirfd;
dirfd = open(".",O_RDONLY, 0700);
if (fsync(dirfd) < 0)
printf("failed\n");
else
printf("succeeded\n");
}

Indeed it fails for all nfs-mounted directories when the client is running kernel 2.6 (tried 2.6.6 and 2.6.8.1), and succeeds for clients running kernel 2.4, with the same nfs server (regardless if the server is running 2.6 or 2.4).

cu,
Frank

Signature · 15 lines
Dipl.-Inform. Frank Steiner   Web:  http://www.bio.ifi.lmu.de/~steiner/
Lehrstuhl f. Bioinformatik    Mail: http://www.bio.ifi.lmu.de/~steiner/m/
LMU, Amalienstr. 17           Phone: +49 89 2180-4049
80333 Muenchen, Germany       Fax:   +49 89 2180-99-4049



-------------------------------------------------------
This SF.Net email is sponsored by BEA Weblogic Workshop
FREE Java Enterprise J2EE developer tools!
Get your free copy of BEA WebLogic Workshop 8.1 today.
http://ads.osdn.com/?ad_id=5047&alloc_id=10808&op=click
_______________________________________________
NFS maillist  -  [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
#2 lore raw
In reply to Frank Steiner

På to , 09/09/2004 klokka 02:15, skreiv Frank Steiner:

Quoted 2 lines
Sep  9 08:03:13 galois sendmail[21596]: sync_dir: cannot fsync directory .: Invalid argument
Sep  9 08:04:02 galois sendmail[21649]: i89641NZ021648: SYSERR(root): sync_dir: cannot fsync directory .: Invalid argument

Yep. We forgot to add the stub to Linux 2.6.x. Please confirm that the following patch fixes the problem for you.

Cheers,
Trond

NFS: add an fsync() stub for directories

---
 dir.c |   11 +++++++++++
 1 files changed, 11 insertions(+)
fs/nfs/dir.c +11 -0
Index: linux-2.6.9-rc1/fs/nfs/dir.c
===================================================================
@@ -51,12 +51,14 @@
 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
 static int nfs_rename(struct inode *, struct dentry *,
 		      struct inode *, struct dentry *);
+static int nfs_fsync_dir(struct file *, struct dentry *, int);
 
 struct file_operations nfs_dir_operations = {
 	.read		= generic_read_dir,
 	.readdir	= nfs_readdir,
 	.open		= nfs_opendir,
 	.release	= nfs_release,
+	.fsync		= nfs_fsync_dir,
 };
 
 struct inode_operations nfs_dir_inode_operations = {
@@ -493,6 +495,15 @@
 }
 
 /*
+ * All directory operations under NFS are synchronous, so fsync()
+ * is a dummy operation.
+ */
+int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
+{
+	return 0;
+}
+
+/*
  * A check for whether or not the parent directory has changed.
  * In the case it has, we assume that the dentries are untrustworthy
  * and may need to be looked up again.
#3 lore raw
Quoted 7 lines · Trond Myklebust

P� to , 09/09/2004 klokka 02:15, skreiv Frank Steiner:

Quoted 2 lines
Sep  9 08:03:13 galois sendmail[21596]: sync_dir: cannot fsync directory .: Invalid argument
Sep  9 08:04:02 galois sendmail[21649]: i89641NZ021648: SYSERR(root): sync_dir: cannot fsync directory .: Invalid argument

Yep. We forgot to add the stub to Linux 2.6.x. Please confirm that the following patch fixes the problem for you.

will fsync'ing a directory have any affect on the client inode cache for that dir? eg. will it invalidate that cache?

cheers.

-a
--
===============================================================================

| EMAIL   :: Ara [dot] T [dot] Howard [at] noaa [dot] gov
| PHONE   :: 303.497.6469
| A flower falls, even though we love it;
| and a weed grows, even though we do not love it. |   --Dogen

===============================================================================

#4 lore raw

P=E5 to , 09/09/2004 klokka 13:54, skreiv Ara.T.Howard:

Quoted 1 line

will fsync'ing a directory have any affect on the client inode cache for =

that

Quoted 1 line

dir? eg. will it invalidate that cache?

No. It should be a no-op.

Cheers,
Trond

-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170 Project Admins to receive an Apple iPod Mini FREE for your judgement on who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 13. Go here: http://sf.net/ppc_contest.php
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs

#5 lore raw
In reply to Trond Myklebust
Quoted 2 lines · Trond Myklebust

Yep. We forgot to add the stub to Linux 2.6.x. Please confirm that the following patch fixes the problem for you.

Yes, with that patch it works fine and sendmail stops complaining.

Thanks a lot!
cu,
Frank

Signature · 14 lines
Dipl.-Inform. Frank Steiner   Web:  http://www.bio.ifi.lmu.de/~steiner/
Lehrstuhl f. Bioinformatik    Mail: http://www.bio.ifi.lmu.de/~steiner/m/
LMU, Amalienstr. 17           Phone: +49 89 2180-4049
80333 Muenchen, Germany       Fax:   +49 89 2180-99-4049



-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM. Deadline: Sept. 13. Go here: http://sf.net/ppc_contest.php
_______________________________________________
NFS maillist  -  [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs