Re: [PATCH] ext4: fix suboptimal seek_{data,hole} extents traversial

From: Dmitry Monakhov
Date: Fri Nov 28 2014 - 10:02:36 EST


Theodore Ts'o <tytso@xxxxxxx> writes:

> On Fri, Oct 17, 2014 at 05:25:34PM +0400, Dmitry Monakhov wrote:
>> It is rediculus practice to scan inode block by block, this technique
>> applicable only for old indirect files. This takes signifficant amount
>> of time for really large files. Let's reuse ext4_fiemap which already
>> traverse inode-tree in most optimal meaner.
>>
>> TESTCASE:
>> ftruncate64(fd, 0);
>> ftruncate64(fd, 1ULL << 40);
>> /* lseek will spin very long time */
>> lseek64(fd, 0, SEEK_DATA);
>> lseek64(fd, 0, SEEK_HOLE);
>>
>>
>> Original report: https://lkml.org/lkml/2014/10/16/620
>>
>> ##################################
>> BTW: Why do we need i_mutex here?
>>
>> Signed-off-by: Dmitry Monakhov <dmonakhov@xxxxxxxxxx>
>
> Note: this patch causes generic/285 to loop forever in inline-data
> mode. My guess is in the special case handling of inline data in
> ext4_fiemap not playing well with this change, but I haven't had a
> chance to look deeply into this yet.
FYI: inline_data feature is fatally broken

1) incorrect lock order (journal_start vs grab_cache_page_write_begin )
ext4_da_write_inline_data_begin
->start_journal
->grab_cache_page_write_begin
->pagecache_get_page -> FS_REENTRANCE->DEADLOCK

2) ext4_inline_data_fiemap(0 ignores start and len arguments from ext4_filemap()
which obviously result in endless loop for anyone who want to use
fiemap for inline files (which my patch try to do)

I'll send patches soon.
>
> - Ted

Attachment: signature.asc
Description: PGP signature