Re: Problem with inotify

[Daniel Drake <dsd@gentoo.org>]

Hi Anton,

Anton Altaparmakov wrote:
> Thinking about it some more made me realize that there may be a problem in 
> inotify after all...  Could you try the below patch to fs/inotify.c and 
> tell me if it cures the lockup you are seeing?  (Note patch compiles but 
> is otherwise untested.  But given it locks up without the patch it can't 
> do much worse with it!)

Thanks for writing that patch, the effort is much appreciated. Unfortunately
it does not help :(

I've done a bit more investigating for you though. I did some tests purely on
the console, using inotify-test (from inotify-utils) which is the most
simplistic way you can use inotify: it just prints out the recieved events to
the screen.

I found out that unmount works perfectly well as long as there are no active
inotify watches (this might be quite obvious though!) - i.e. closing
inotify-test before unmounting results in a clean unmount. Unmounting while
inotify-test is watching the NTFS partition causes the freeze.

When the machine freezes, it still responds to ping, but not to ssh. Sysrq
works, so I got a sysrq-p trace:

Pid 8997 comm umount
EIP is at inotify_unmount_inodes+0x38/0x140

stack trace:
invalidate_inodes+0x40/0x90
generic_shutdown_super+0x59/0x140
kill_block_super+0x2d/0x50
deactivate_super+0x5a/0x90
sys_umount+0x3f/0x90
filp_close+0x52/0xa0
sys_oldumount+0x17/0x20
sysenter_past_esp+0x54/0x75

Investigating that function:

(gdb) list *inotify_unmount_inodes+0x38
0x9c8 is in inotify_unmount_inodes (inotify.c:565).
560      */
561     void inotify_unmount_inodes(struct list_head *list)
562     {
563             struct inode *inode, *next_i, *need_iput = NULL;
564
565             list_for_each_entry_safe(inode, next_i, list, i_sb_list) {
566                     struct inode *need_iput_tmp;
567                     struct inotify_watch *watch, *next_w;
568                     struct list_head *watches;
569

I then added a loop counter printk in at line 569 above. It shows that the
loop iterates 8 times on a clean unmount, and goes into a seemingly infinite
loop (i.e. freeze) when unmounting with inotify watches active.

I don't know much about filesystem internals so I'm pretty stuck here, haven't
reached that chapter of Robert's book yet ;)

Please let me know if theres any other info I can provide.

Thanks,
Daniel
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/