network lock manager (lockd) deadlocked in 'rpcrecv'

[John Hein]

John Hein wrote at 15:31 -0600 on Jul  8, 2009:
 > I have a home directory on FreeBSD 7.2-stable (20090705), amd64.
 > It is serving up the directory over nfs (v3, tcp), and now
 > I'm seeing lots of 'lockd not responding' on Fedora 10 & 11 systems.
 > 
 > USER    PID  PPID   SID NI %CPU %MEM   VSZ   RSS  TT  WCHAN  STAT STARTED      TIME COMMAND
 > root    791     1   791  0  0.0  0.0  6748  1500  ??  rpcrec Ds    2:45PM   0:05.80 /usr/sbin/rpc.lockd
 > 
 > Once lockd gets in this state, doing a test lock on a file
 > from a FreeBSD box locks with 'lockd not responding', too
 > (and ctrl-c and kill -9 does nothing).
 > 
 > USER         PID  PPID   SID NI %CPU %MEM   VSZ   RSS  TT  WCHAN  STAT STARTED      TIME COMMAND
 > jhein       6297  3491  3491  0  0.0  0.0  1412   604  p5  nlmrcv T+    3:18PM   0:00.00 /h/jhein/nfslocktest /nfs/locktest
 > 
 > 
 > I see this on an i386 6.4-stable, too.

Also in dmesg:

NLM: failed to contact remote rpcbind, stat = 5, port = 28416


And from ddb...

Tracing command rpc.lockd pid 791 tid 100176 td 0xffffff00069dd720
sched_switch() at 0xffffffff8037df95 = sched_switch+0x1d5
mi_switch() at 0xffffffff803656fb = mi_switch+0x18b
sleepq_timedwait() at 0xffffffff80390aeb = sleepq_timedwait+0x3b
_sleep() at 0xffffffff80365cd4 = _sleep+0x324
clnt_dg_call() at 0xffffffff80504a0b = clnt_dg_call+0x4fb
nlm_get_rpc() at 0xffffffff804f3ef7 = nlm_get_rpc+0x147
nlm_host_get_rpc() at 0xffffffff804f430e = nlm_host_get_rpc+0x10e
nlm_do_lock() at 0xffffffff804f58be = nlm_do_lock+0x1ce
nlm4_lock_4_svc() at 0xffffffff804f6c91 = nlm4_lock_4_svc+0x11
nlm_prog_4() at 0xffffffff804f8098 = nlm_prog_4+0x308
svc_run() at 0xffffffff8050c1f3 = svc_run+0x293
nlm_syscall() at 0xffffffff804f675c = nlm_syscall+0x79c
syscall() at 0xffffffff805818f4 = syscall+0x1b4
Xfast_syscall() at 0xffffffff8056d35b = Xfast_syscall+0xab
--- syscall (154, FreeBSD ELF64, nlm_syscall), rip = 0x8008a91ec, rsp = 0x7fffffffed08, rbp = 0x7fffffffe
e20 ---