Skip to contentSkip to conversation

oops in close when exiting fsx

DISCUSSION Chuck Ebbert 2 participants 2 messages Aug 13, 2006

lore
Share & tools
lore
#0 lore raw

In-Reply-To: <[email protected]>

Quoted 7 lines · Steve French

ctl-c exiting fsx after a few hours with 2.6.18-rc4 got the following oops - anyone recognize it?
Although I didn't see cifs symbols on the call stack it is running on a cifs mount, but it is not
one I have seen before.

EIP is at __down+0x56/0xc5

  1a:   8d 43 08                  lea    0x8(%ebx),%eax  <= addr of sema wait queue list_head
  1d:   8b 48 04                  mov    0x4(%eax),%ecx  <= list->prev
  20:   8d 54 24 2c               lea    0x2c(%esp),%edx
  24:   89 50 04                  mov    %edx,0x4(%eax)
  27:   89 44 24 2c               mov    %eax,0x2c(%esp)
   0:   89 11                     mov    %edx,(%ecx)   <===== list->prev->next = new

The semaphore's wait queue head is corrupted: 'prev' is 0.

Quoted 11 lines
 [<c1038908>] mempool_free+0x43/0x46
 [<c1013678>] default_wake_function+0x0/0xc
 [<c132ed37>] __down_failed+0x7/0xc
 [<fa2da685>] .text.lock.file+0x87/0x9a [cifs]      <=====
 [<c104e807>] __fput+0xab/0x148
 [<c104c453>] filp_close+0x4e/0x54
 [<c101773a>] put_files_struct+0x64/0xa6
 [<c1018581>] do_exit+0x1c7/0x675
 [<c10052b0>] do_syscall_trace+0x12b/0x172
 [<c1018a8b>] sys_exit_group+0x0/0xd
 [<c1002abf>] syscall_call+0x7/0xb

It came from a lock section in the cifs code. If you disassemble .text.lock.file in cifs.o, at offset 0x87 (or shortly after) you will see a jump back to the code that's trying to get the semaphore.

Signature · 1 line
Chuck
#1 lore raw
Quoted 37 lines · Chuck Ebbert

In-Reply-To: <[email protected]>

Quoted 8 lines · Steve French

ctl-c exiting fsx after a few hours with 2.6.18-rc4 got the following oops - anyone recognize it? Although I didn't see cifs symbols on the call stack it is running on a cifs mount, but it is not one I have seen before.

EIP is at __down+0x56/0xc5

  1a:   8d 43 08                  lea    0x8(%ebx),%eax  <= addr of sema wait queue list_head
  1d:   8b 48 04                  mov    0x4(%eax),%ecx  <= list->prev
  20:   8d 54 24 2c               lea    0x2c(%esp),%edx
  24:   89 50 04                  mov    %edx,0x4(%eax)
  27:   89 44 24 2c               mov    %eax,0x2c(%esp)
   0:   89 11                     mov    %edx,(%ecx)   <===== list->prev->next = new

The semaphore's wait queue head is corrupted: 'prev' is 0.

Quoted 12 lines
 [<c1038908>] mempool_free+0x43/0x46
 [<c1013678>] default_wake_function+0x0/0xc
 [<c132ed37>] __down_failed+0x7/0xc
 [<fa2da685>] .text.lock.file+0x87/0x9a [cifs]      <=====
 [<c104e807>] __fput+0xab/0x148
 [<c104c453>] filp_close+0x4e/0x54
 [<c101773a>] put_files_struct+0x64/0xa6
 [<c1018581>] do_exit+0x1c7/0x675
 [<c10052b0>] do_syscall_trace+0x12b/0x172
 [<c1018a8b>] sys_exit_group+0x0/0xd
 [<c1002abf>] syscall_call+0x7/0xb
Quoted 5 lines

It came from a lock section in the cifs code. If you disassemble .text.lock.file in cifs.o, at offset 0x87 (or shortly after) you will see a jump back to the code that's trying to get the semaphore.

Thanks - This is a part of new cifs code recently added to handle posix locks (it has not pushed to mainline yet) better

               down(&pSMBFile->lock_sem);
               list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
                       list_del(&li->llist);
                       kfree(li);
               }
               up(&pSMBFile->lock_sem);

My guess is that there is a path in which the lock_sem is not initialized - will trace that.