Hello Axel,

attached is a small tool to test fchown on a freshly created file: <>

These are the tests I've made using this tool:

0. compile

   ...

   xlc /tmp/t.c -o /tmp/t

   environemnt

   ...

   mount | grep -E "(filer0|hd4)" /dev/hd4 / jfs Jun 11 23:25 rw,log=/dev/hd8 filer0 /vol/mail /net/var_mail nfs3 Jun 11 23:26 rw,proto=tcp,port=2049,wsize=65534,rsize=65534 filer0 /vol/home /u/f0 nfs4 Jul 06 08:09 rw,proto=tcp,port=2049,vers=4,wsize=65534,rsize=65534

1. local filesystem => SUCCESS

   ...

   /tmp/t /tmp/test.txt fchown returns 0

2. NFS3 filesystem => SUCCESS

   ...

   /tmp/t /net/var_mail/spool/test.txt fchown returns 0

3. NFS4 filesystem => ERROR

   ...

   /tmp/t /u/f0/test.txt fchown returns -1 errno=22 (Invalid argument)

So I should alter the to subject to ... with AIX, NFS4 and netapp ... :-)

Back to your question:

Wouldn't it be worth to check what kind of entity gets created under your environment?

yes and it's just a raw file

and as suspected conclusion

EINVAL is returned because the owner or group ID is not a value supported by the implementation (of NFS4 on netapp filers?)

Ralf

Axel Luttgens schrieb:

According to the posix specification, fchown may return EINVAL when the owner or group ID is not a value supported by the implementation, or when the fildes argument refers to a pipe or socket or an fattach()-ed STREAM and the implementation disallows execution of fchown() on a pipe.

Wouldn't it be worth to check what kind of entity gets created under your environment? I ask because I wouldn't exclude without further investigations the possibility of encountering other side effects wrt files throughout the code.

--

Dipl.-Inform. (FH) Ralf Becker Rechenzentrum (r/ft) der FH Trier (Network|Mail|Web|Firewall) University of applied sciences Administrator Schneidershof, D-54293 Trier

Mail: beckerr@fh-trier.de Fon: +49 651 8103 499 WWW: http://www.fh-trier.de/~beckerr Fax: +49 651 8103 214

Hello Axel,

...

attached is a small tool to test fchown on a freshly created file: <>

Damn... didn't go through... ;-)

ok... maybe i missed it... let's do it inline:

#include #include #include #include #include

main(int argc, char **argv) { int f = open(argv[1],O_CREAT|O_TRUNC); printf("fchown returns %i\n",fchown(f,-1L,-1L)); if (errno) printf("errno=%i (%s)\n", errno, strerror(errno)); close(f); unlink(argv[1]); }

As a result, I'm not sure whether entirely hiding the log message is a good idea; perhaps just change the logging level would be better, so that one keeps the ability to track possibly problematic file systems...

It's not just the log message. If you have a look on the entire function, you'll see that it fails if fchown fails:

static int file_dotlock_open_mode_full(<...>,uid_t uid, gid_t gid,<...>) { <...>

if (fd != -1) {
    if (fchown(fd, uid, gid) < 0) {
        if (errno == EPERM && uid == (uid_t)-1) {
            i_error("%s", eperm_error_get_chgrp("fchown",

                file_dotlock_get_lock_path(dotlock),
                gid, gid_origin));
         } else {
            i_error("fchown(%s, %ld, %ld) failed: %m",
                file_dotlock_get_lock_path(dotlock),
                (long)uid, (long)gid);
         }
         file_dotlock_delete(&dotlock);
         return -1;
    }
}
*dotlock_r = dotlock;
return fd;

}

While this function seems to create all dotlock files (not just for the .subscribtions file) this means that on same NFS(4) file systems dotlocking is actually not working.

The linux man page of chown(3) (in place of fchown(3)) says:

-----------------------------8<-------------------------------- If owner or group is specified as ( uid_t)-1 or ( gid_t)-1, respectively, the corresponding ID of the file shall not be changed. If both owner and group are -1, the times need not be updated. Upon successful completion, chown() shall mark for update the st_ctime field of the file. ------------------->8------------------------------------------

Is my understanding of these sentences correct? "If owner and group are -1, nothing is done?"

In this case it should be save to skip the call, shouldn't it?

Ralf

--

Dipl.-Inform. (FH) Ralf Becker Rechenzentrum (r/ft) der FH Trier (Network|Mail|Web|Firewall) University of applied sciences Administrator Schneidershof, D-54293 Trier

Mail: beckerr@fh-trier.de Fon: +49 651 8103 499 Web: http://www.fh-trier.de/~beckerr Fax: +49 651 8103 214 PubKey: http://www.fh-trier.de/~beckerr Crypto: GnuPG, S/MIME

Wenn Gott gewollt haette, dass E-Mail in HTML geschrieben wuerden, endeten Gebete traditionell mit </amen>. (Tom Listen)

On Tue, 2009-07-07 at 17:58 +0200, Axel Luttgens wrote:

...

Is my understanding of these sentences correct? "If owner and group are -1, nothing is done?"

In this case it should be save to skip the call, shouldn't it?

Yes, I guess so.

Yes. Committed: http://hg.dovecot.org/dovecot-1.2/rev/d6337be8ae30

Unless the rationale for that call is to ensure a correct cache
flushing for NFS clients, while being some kind of (costly) no-op
otherwise?

In that case I would have used those nfs_flush_*() functions.