2006-12-09 23:16:08

by Catalin Marinas

Subject: Possible memory leak in block/ll_rw_blk.c

Hi,

After 2.6.19, kmemleak reports several (few tens) orphan blocks
allocated in bio_alloc_bioset() via __blk_rq_map_user() in
block/ll_rq_blk.c. I think these came with commit
0e75f9063f5c55fb0b0b546a7c356f8ec186825e (support larger block pc
requests). The allocation backtrace for the "bio" structure is:

unreferenced object 0xdd9162b0 (size 64):
[<c018d46f>] kmem_cache_alloc
[<c0170b2e>] mempool_alloc_slab
[<c01709cb>] mempool_alloc
[<c01b7baa>] bio_alloc_bioset
[<c01b7d0e>] bio_alloc
[<c01b83f8>] bio_copy_user
[<c021a380>] __blk_rq_map_user
[<c021a4ff>] blk_rq_map_user
[<c021e687>] sg_io
[<c021ed3e>] scsi_cmd_ioctl
[<c02bcc13>] sd_ioctl
[<c021ca65>] blkdev_driver_ioctl
[<c021cc27>] blkdev_ioctl
[<c01ba72b>] block_ioctl
[<c019ea36>] do_ioctl

Because the above objects cannot be tracked, kmemleak also reports the
bio_map_data structures allocated in bio_alloc_map_data (called from
bio_copy_user via the above backtrace).

Thanks.

--
Catalin

2006-12-13 10:00:07

by Catalin Marinas

Subject: Re: Possible memory leak in block/ll_rw_blk.c

On 09/12/06, Catalin Marinas <[email protected]> wrote:
> unreferenced object 0xdd9162b0 (size 64):
> [<c018d46f>] kmem_cache_alloc
> [<c0170b2e>] mempool_alloc_slab
> [<c01709cb>] mempool_alloc
> [<c01b7baa>] bio_alloc_bioset
> [<c01b7d0e>] bio_alloc
> [<c01b83f8>] bio_copy_user
> [<c021a380>] __blk_rq_map_user
> [<c021a4ff>] blk_rq_map_user
> [<c021e687>] sg_io
> [<c021ed3e>] scsi_cmd_ioctl
> [<c02bcc13>] sd_ioctl
> [<c021ca65>] blkdev_driver_ioctl
> [<c021cc27>] blkdev_ioctl
> [<c01ba72b>] block_ioctl
> [<c019ea36>] do_ioctl

I can confirm that the memory leak disappeared with commit
77d172ce2719b5ad2dc0637452c8871d9cba344c (by Fujita Tomonori).

Regards.

--
Catalin

2006-12-13 15:48:52

by FUJITA Tomonori

Subject: Re: Possible memory leak in block/ll_rw_blk.c

From: "Catalin Marinas" <[email protected]>
Subject: Re: Possible memory leak in block/ll_rw_blk.c
Date: Wed, 13 Dec 2006 10:00:05 +0000

> On 09/12/06, Catalin Marinas <[email protected]> wrote:
> > unreferenced object 0xdd9162b0 (size 64):
> > [<c018d46f>] kmem_cache_alloc
> > [<c0170b2e>] mempool_alloc_slab
> > [<c01709cb>] mempool_alloc
> > [<c01b7baa>] bio_alloc_bioset
> > [<c01b7d0e>] bio_alloc
> > [<c01b83f8>] bio_copy_user
> > [<c021a380>] __blk_rq_map_user
> > [<c021a4ff>] blk_rq_map_user
> > [<c021e687>] sg_io
> > [<c021ed3e>] scsi_cmd_ioctl
> > [<c02bcc13>] sd_ioctl
> > [<c021ca65>] blkdev_driver_ioctl
> > [<c021cc27>] blkdev_ioctl
> > [<c01ba72b>] block_ioctl
> > [<c019ea36>] do_ioctl
>
> I can confirm that the memory leak disappeared with commit
> 77d172ce2719b5ad2dc0637452c8871d9cba344c (by Fujita Tomonori).

Thanks. And sorry for introducing that bug.