Hello everybody,
During test of NFS software I've found the problem in Linux kernel (nfsd)
On the NFS4_WRITE request (data len=44), NFS server is responding with NFS4_OK (bytes written count = 0) that follows to
cycliing in write request on client side.
REQUEST DUMP:
0000 00 00 00 01 00 06 00 09 6b 1a ed 41 00 00 08 00 ........ k..A....
0010 45 00 01 24 00 ae 00 00 40 06 7b da 09 0b 75 66 E..$.... @.{...uf
0020 09 0b 75 d0 1e 2c 08 01 ec 30 eb 6d 2e 0e 1d f8 ..u..,.. .0.m....
0030 80 18 7f dd 17 96 00 00 01 01 08 0a e4 d5 5a 09 ........ ......Z.
0040 02 e5 a3 f9 80 00 00 ec 39 35 0c f8 00 00 00 00 ........ 95......
0050 00 00 00 02 00 01 86 a3 00 00 00 04 00 00 00 01 ........ ........
0060 00 00 00 06 00 00 00 18 00 00 00 01 00 00 00 00 ........ ........
0070 00 00 00 79 00 00 00 01 00 00 00 04 00 00 00 16 ...y.... ........
0080 00 00 00 06 00 00 00 25 60 23 06 09 2a 86 48 86 .......% `#..*.H.
0090 f7 12 01 02 02 01 01 00 00 ff ff ff ff a3 ef 9f ........ ........
00a0 60 d6 0f 43 48 bd 3a d4 0b 7f d1 13 85 00 00 00 `..CH.:. ........
00b0 00 00 00 00 00 00 00 00 00 00 00 03 00 00 00 16 ........ ........
00c0 00 00 00 10 01 00 01 01 00 00 00 00 00 0a 78 ee ........ ......x.
00d0 81 d9 43 8a 00 00 00 26 00 00 00 00 4a 13 df 23 ..C....& ....J..#
00e0 00 00 00 13 00 00 01 db 00 00 00 00 00 00 00 00 ........ ........
00f0 00 00 00 01 00 00 00 2c e3 c8 c9 e2 40 c9 e2 40 ......., ....@..@
0100 c1 40 e3 c5 e2 e3 40 c2 e4 c6 c6 c5 d9 4b e3 c8 .@....@. .....K..
0110 c9 e2 40 c9 e2 40 c1 40 e3 c5 e2 e3 40 c2 e4 c6 ..@..@.@ ....@...
0120 c6 c5 d9 4b 00 00 00 09 00 00 00 02 00 00 00 18 ...K.... ........
0130 00 30 a0 3a .0.:
ANSWER DUMP:
0000 00 04 00 01 00 06 62 a2 90 00 80 04 00 00 08 00 ......b. ........
0010 45 00 01 48 34 f2 40 00 40 06 07 72 09 0b 75 d0 E..H4.@. @..r..u.
0020 09 0b 75 66 08 01 1e 2c 2e 0e 1d f8 ec 30 ec 5d ..uf..., .....0.]
0030 80 18 04 c8 00 00 00 00 01 01 08 0a 02 e5 a3 fe ........ ........
0040 e4 d5 5a 09 80 00 01 10 39 35 0c f8 00 00 00 01 ..Z..... 95......
0050 00 00 00 00 00 00 00 06 00 00 00 25 60 23 06 09 ........ ...%`#..
0060 2a 86 48 86 f7 12 01 02 02 01 01 00 00 ff ff ff *.H..... ........
0070 ff 3f 01 f5 2c f4 b1 ad 03 29 4f ef 8b f8 1f ea .?..,... .)O.....
0080 d8 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ........ ........
0090 00 00 00 03 00 00 00 16 00 00 00 00 00 00 00 26 ........ .......&
00a0 00 00 00 00 00 00 00 00 00 00 00 01 4a 13 df 23 ........ ....J..#
00b0 00 0b 33 42 00 00 00 09 00 00 00 00 00 00 00 02 ..3B.... ........
00c0 00 00 00 18 00 30 a0 3a 00 00 00 8c 4a 16 d7 b0 .....0.: ....J...
00d0 00 00 00 00 00 00 00 00 00 00 00 2c 00 00 01 b6 ........ ...,....
00e0 00 00 00 01 00 00 00 1d 75 73 65 72 31 31 40 73 ........ user11@s
00f0 74 6f 72 61 67 65 2e 74 75 63 73 6f 6e 2e 69 62 torage.t ucson.ib
0100 6d 2e 63 6f 6d 00 00 00 00 00 00 1f 74 73 6f 75 m.com... ....tsou
0110 73 65 72 31 40 73 74 6f 72 61 67 65 2e 74 75 63 ser1@sto rage.tuc
0120 73 6f 6e 2e 69 62 6d 2e 63 6f 6d 00 00 00 00 00 son.ibm. com.....
0130 00 00 10 00 00 00 00 00 00 00 00 01 00 00 00 00 ........ ........
0140 00 00 00 00 4a 16 d7 b0 00 00 00 00 00 00 00 00 ....J... ........
0150 4a 16 d7 b0 00 00 00 00 J.......
I can suppose that problem is located in:
linux-2.6.30-rc7\fs\nfsd\vfs.c (nfsd_vfs_write):
....
host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); In the stable release (2.6.29) host_err
means OK == 0 or error code if host_err < 0
Current implementation (RC7) host_err
is NUMBER OF BYTES WRITTEN (>=0) or error code (<0)
....
if (inode->i_state & I_DIRTY) {
dprintk("nfsd: write sync %d\n", task_pid_nr(current));
host_err=nfsd_sync(file); Here host_err seems to have old meaning
so host_err again just 0 for OK and <0 for error
....
dprintk("nfsd: write complete host_err=%d\n", host_err);
if (host_err >= 0) {
err = 0;
*cnt = host_err; Here is place to fill 'written bytes' field for
RPC answer. It is filled sometimes to 'written
bytes' count and sometimes to zero.
} else
err = nfserrno(host_err);
out:
return err;
}
Is it's the case I can propose to move the assignment "*cnt = host_err;" upper:
host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
if( host_err >= 0 )
*cnt = host_err;
with best regards,
Alexander Smirnov,
Luxoft,
IBM zOS NFS Development team
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
