Re: processes stuck in llseek

[David Sterba <dsterba@suse.cz>]

On Mon, Aug 15, 2011 at 11:59:50PM -0400, Dan Merillat wrote:
> I noticed a series of hung_task notifications in dmesg, so I went
> poking at it.   Process is 'dropbox', and it's stuck trying to llseek
> it's library.zip file.

looks like mutex lock is waiting, in btrfs_file_llseek. somebody else is
holding it for that file. it could be a forgotten unlock (of
inode->i_mutex) on some error path, but this does not look very
promising to me, or the process holding the lock still exists and will
be visible in process listings. locdep would tell more exactly.

> 
> strace of dropbox:
> ...
> stat("/home/x/.dropbox-dist/library.zip", {st_mode=S_IFREG|0755,
> st_size=11575179, ...}) = 0
> open("/home/x/.dropbox-dist/library.zip", O_RDONLY) = 3
> fstat(3, {st_mode=S_IFREG|0755, st_size=11575179, ...}) = 0
> mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1,
> 0) = 0x7fa766ea8000
> fstat(3, {st_mode=S_IFREG|0755, st_size=11575179, ...}) = 0
> lseek(3, 11571200, SEEK_SET
> 
> SEEK_SET is less than st_size
> 
> strace of dd if=library.zip of=/dev/null bs=1 seek=11571200:
> 
> open("library.zip", O_RDONLY)           = 3
> dup2(3, 0)                              = 0
> close(3)                                = 0
> lseek(0, 0, SEEK_CUR
> 
> [72960.716080] INFO: task dropbox:1348 blocked for more than 120 seconds.
> [72960.716084] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs"
> disables this message.
> [72960.716087] dropbox         D ffff8800762d8a78     0  1348      1 0x00000004
> [72960.716092]  ffff880069cede18 0000000000000086 0000000000096000
> 0000000000000000
> [72960.716097]  ffff880069cec000 00000000000119c0 ffff8800762d8700
> 00000000000119c0
> [72960.716101]  ffff880069cedfd8 0000000000004000 ffff880069cedfd8
> 00000000000119c0
> [72960.716106] Call Trace:
> [72960.716113]  [<ffffffff81331f9e>] ? trace_hardirqs_on_thunk+0x3a/0x3c
> [72960.716119]  [<ffffffff8155710a>] ? retint_restore_args+0xe/0xe
> [72960.716124]  [<ffffffff810e8ef1>] ? noop_llseek+0xa/0xa
> [72960.716129]  [<ffffffff810340ff>] ? mutex_spin_on_owner+0x1c/0x45
> [72960.716133]  [<ffffffff81555be0>] __mutex_lock_slowpath+0xd2/0x116
> [72960.716137]  [<ffffffff81559edd>] ? do_page_fault+0x374/0x3e6
> [72960.716140]  [<ffffffff81555a87>] mutex_lock+0x16/0x27
> [72960.716146]  [<ffffffff812c8ad7>] btrfs_file_llseek+0x38/0x297
> [72960.716150]  [<ffffffff81331fda>] ? trace_hardirqs_off_thunk+0x3a/0x6c
> [72960.716153]  [<ffffffff810e8f2c>] vfs_llseek+0x2e/0x30
> [72960.716155]  [<ffffffff810e9311>] sys_lseek+0x3e/0x5d
> [72960.716159]  [<ffffffff8155d4fb>] system_call_fastpath+0x16/0x1b

just a guess, as it touches file pages, can you try Sage's fix

"truncate pages from clone ioctl target range"
https://patchwork.kernel.org/patch/1054552/

under assumption that the file on which seek is run has been reflinked
or at least written to during the running transaction (ie no sync in
between).

> Oddly enough, it's just llseek.  I can copy/read the file sequentially
> just fine, but llseek fails on a copy as well.

incorrect or unexpected contents of the same-aged files would point to
the missing truncate.


david

> Once I get physically
> to the system I'll recompile to an unmodified kernel and try again.
> Other files of the same approximate age llseek correctly.
> 
> Kernel is 3.1-rc1  with the fglrx module from AMD and the following
> patch:  (I was playing with cross-subvolume reflinking, but not on
> this file)
> 
> diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
> index 7cf0133..a13b5e2 100644
> --- a/fs/btrfs/ioctl.c
> +++ b/fs/btrfs/ioctl.c
> @@ -2182,7 +2182,7 @@ static noinline long btrfs_ioctl_clone(struct
> file *file, unsigned long srcfd,
>  		goto out_fput;
> 
>  	ret = -EXDEV;
> -	if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
> +	if (src->i_sb != inode->i_sb)
>  		goto out_fput;
> 
>  	ret = -ENOMEM;
> @@ -2246,13 +2246,13 @@ static noinline long btrfs_ioctl_clone(struct
> file *file, unsigned long srcfd,
>  		 * note the key will change type as we walk through the
>  		 * tree.
>  		 */
> -		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
> +		ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path, 0, 0);
>  		if (ret < 0)
>  			goto out;
> 
>  		nritems = btrfs_header_nritems(path->nodes[0]);
>  		if (path->slots[0] >= nritems) {
> -			ret = btrfs_next_leaf(root, path);
> +			ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
>  			if (ret < 0)
>  				goto out;
>  			if (ret > 0)
> @@ -2313,7 +2313,7 @@ static noinline long btrfs_ioctl_clone(struct
> file *file, unsigned long srcfd,
>  			else
>  				new_key.offset = destoff;
> 
> -			trans = btrfs_start_transaction(root, 1);
> +			trans = btrfs_start_transaction(root, 3);
>  			if (IS_ERR(trans)) {
>  				ret = PTR_ERR(trans);
>  				goto out;
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html