On Sun, Aug 26, 2012 at 05:56:30PM +0300, Idan Kedar wrote: > On Fri, Aug 24, 2012 at 5:19 PM, Tetsuo Handa > <penguin-kernel@xxxxxxxxxxxxxxxxxxx> wrote: > > Hello. > > > > I noticed a strange delay when I was running a test program for > > http://serverfault.com/questions/235059/vfs-file-max-limit-1231582-reached . > > > > ---------- test program start ---------- > > #include <sys/types.h> > > #include <sys/stat.h> > > #include <sys/file.h> > > #include <stdio.h> > > #include <unistd.h> > > #include <errno.h> > > > > int main(int argc, char *argv[]) { > > while (1) { > > struct flock lock = { }; > > int fd = open("/mnt/file", O_CREAT | O_RDWR, 0600); > > lock.l_type = F_WRLCK; > > lock.l_whence = SEEK_SET; > > if (fcntl(fd, F_SETLK, &lock)) { > > int err = errno; > > fprintf(stderr, "Error %d\n", err); > > return 1; > > } > > close(fd); > > fprintf(stderr, "."); > > } > > } > > ---------- test program end ---------- > > > > When above program is executed after "service nfs start", above program sleeps > > for about 90-100 seconds at the first fcntl() attempt. But after once the first > > '.' is printed, above program never sleeps even after above program was > > re-executed. > > > > # cat /etc/exports > > /mountdir *(rw,all_squash,sync,fsid=0) > > # service nfs start > > # mount -t nfs 127.0.0.1:/mountdir /mnt/ > > # time ./a.out > > > > sysrq on 3.6.0-rc2-00400-g23dcfa6 kernel shows: > > > > [ 728.514014] a.out S 00000000 5656 4400 3765 0x00000080 > > [ 728.514014] d9a0dd7c 00000046 d9a0dcf0 00000000 00000001 dcf0aec8 00000023 00000000 > > [ 728.514014] 00000000 dcf0aa60 c17d2220 c17d2220 9fac0f21 000000a8 c17d2220 c17d2220 > > [ 728.514014] 9fa96cf7 000000a8 dcf0aa60 000000a8 9fac0995 dcf0aa60 00000000 00000000 > > [ 728.514014] Call Trace: > > [ 728.514014] [<c143fc55>] ? _raw_spin_unlock_irqrestore+0x35/0x60 > > [ 728.514014] [<c143e45e>] schedule+0x1e/0x50 > > [ 728.514014] [<c143c6ea>] schedule_timeout+0xea/0x170 > > [ 728.514014] [<c1040a10>] ? lock_timer_base+0x50/0x50 > > [ 728.514014] [<c11e1f27>] nlmclnt_call+0x97/0x1f0 > > [ 728.514014] [<c1053210>] ? wake_up_bit+0x30/0x30 > > [ 728.514014] [<c11e282e>] nlmclnt_proc+0x22e/0x6f0 > > [ 728.514014] [<c1062e1e>] ? local_clock+0x4e/0x60 > > [ 728.514014] [<c11a567a>] nfs3_proc_lock+0x1a/0x20 > > [ 728.514014] [<c1196069>] do_setlk+0xc9/0xf0 > > [ 728.514014] [<c11961b8>] nfs_lock+0x98/0x170 > > [ 728.514014] [<c1196120>] ? nfs_flock+0x90/0x90 > > [ 728.514014] [<c1122161>] vfs_lock_file+0x21/0x50 > > [ 728.514014] [<c112227a>] do_lock_file_wait+0x4a/0xd0 > > [ 728.514014] [<c10c7426>] ? might_fault+0x46/0xa0 > > [ 728.514014] [<c1230304>] ? _copy_from_user+0x34/0x80 > > [ 728.514014] [<c112278e>] fcntl_setlk+0xce/0x2c0 > > [ 728.514014] [<c107ba3b>] ? trace_hardirqs_on+0xb/0x10 > > [ 728.514014] [<c10f0b82>] do_fcntl+0x1e2/0x5f0 > > [ 728.514014] [<c10f1013>] sys_fcntl64+0x83/0xd0 > > [ 728.514014] [<c144048c>] sysenter_do_call+0x12/0x32 > > > > Same result for CentOS 6.3's 2.6.32-279.5.1.el6.i686 kernel: > > > > a.out S 00000286 0 1290 1160 0x00000080 > > df4d1aa0 00000086 e0d20aac 00000286 00000007 df57c24c dcd94544 0000ad6f > > 00000000 df8393c0 0000001e d1daf53c 0000001e c0b25680 c0b25680 df4d1d48 > > c0b25680 c0b21024 c0b25680 df4d1d48 de01e000 00000246 e0d25b24 e0d3dd88 > > Call Trace: > > [<e0d20aac>] ? xs_sendpages+0x1cc/0x240 [sunrpc] > > [<e0d25b24>] ? rpc_wake_up_next+0xc4/0x180 [sunrpc] > > [<c0440372>] ? __wake_up+0x42/0x60 > > [<e0d1cecb>] ? rpc_release_client+0x7b/0x90 [sunrpc] > > [<c0466137>] ? lock_timer_base+0x27/0x50 > > [<c083d34c>] ? schedule_timeout+0x12c/0x250 > > [<c04e3591>] ? mempool_free+0x81/0x90 > > [<c0466210>] ? process_timeout+0x0/0x10 > > [<e0daf8a6>] ? nlmclnt_call+0xd6/0x290 [lockd] > > [<c0476a10>] ? autoremove_wake_function+0x0/0x40 > > [<e0db0349>] ? nlmclnt_proc+0x219/0x760 [lockd] > > [<c04769f3>] ? wake_up_bit+0x13/0x30 > > [<e0eeda0e>] ? do_setlk+0xbe/0xe0 [nfs] > > [<e0eedbd3>] ? nfs_lock+0xb3/0x1b0 [nfs] > > [<c053be35>] ? do_filp_open+0x635/0xb00 > > [<c051f102>] ? cache_alloc_refill+0x172/0x510 > > [<e0eedb20>] ? nfs_lock+0x0/0x1b0 [nfs] > > [<c056e1d2>] ? vfs_lock_file+0x22/0x50 > > [<c056e2eb>] ? do_lock_file_wait+0x4b/0xd0 > > [<c056e64f>] ? fcntl_setlk+0xff/0x1e0 > > [<c053cf52>] ? do_fcntl+0x92/0x440 > > [<c04b14cc>] ? audit_syscall_entry+0x21c/0x240 > > [<c053d35a>] ? sys_fcntl64+0x5a/0xd0 > > [<c0409a9f>] ? sysenter_do_call+0x12/0x28 > > > > I wonder what this sleeping period is for. ... > After the NFS server restarts it has a grace period of 90 seconds, for > recovery purposes. You can read about it online. > You can also look it up on the NFS 4.1 RFC 5661: > http://tools.ietf.org/rfcmarkup?doc=5661 Yes. In this particular case it's NLM, not NFSv4, but it's the same basic idea: no new locks (opens too in the case of NFSv4) are allowed for a while at the start, to give previous clients a chance to reclaim their locks first. --b. -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
