From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752968AbbGAQ5v (ORCPT ); Wed, 1 Jul 2015 12:57:51 -0400 Received: from mx1.redhat.com ([209.132.183.28]:60250 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751614AbbGAQ5n (ORCPT ); Wed, 1 Jul 2015 12:57:43 -0400 Date: Wed, 1 Jul 2015 12:57:40 -0400 From: Mike Snitzer To: "Martin K. Petersen" Cc: axboe@kernel.dk, dm-devel@redhat.com, linux-kernel@vger.kernel.org, mpatocka@redhat.com Subject: [RFC PATCH] bio integrity: do not assume bio_integrity_pool exists if bioset exists Message-ID: <20150701165740.GA1573@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.5.23 (2014-03-12) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org bio_integrity_alloc() and bio_integrity_free() assume that if a bio was allocated from a bioset that that bioset also had its bio_integrity_pool allocated using bioset_integrity_create(). This is a very bad assumption given that bioset_create() and bioset_integrity_create() are completely disjoint. Not all callers of bioset_create() have been trained to also call bioset_integrity_create() -- and they may not care to be. Fix this by falling back to kmalloc'ing 'struct bio_integrity_payload' rather than force all bioset consumers to (wastefully) preallocate a bio_integrity_pool that they very likely won't actually need (given the niche nature of the current block integrity support). Otherwise, a NULL pointer "Kernel BUG" with a trace like the following will be observed (as seen on s390x using zfcp storage) because dm-io doesn't use bioset_integrity_create() when creating its bioset: [ 791.643338] Call Trace: [ 791.643339] ([<00000003df98b848>] 0x3df98b848) [ 791.643341] [<00000000002c5de8>] bio_integrity_alloc+0x48/0xf8 [ 791.643348] [<00000000002c6486>] bio_integrity_prep+0xae/0x2f0 [ 791.643349] [<0000000000371e38>] blk_queue_bio+0x1c8/0x3d8 [ 791.643355] [<000000000036f8d0>] generic_make_request+0xc0/0x100 [ 791.643357] [<000000000036f9b2>] submit_bio+0xa2/0x198 [ 791.643406] [<000003ff801f9774>] dispatch_io+0x15c/0x3b0 [dm_mod] [ 791.643419] [<000003ff801f9b3e>] dm_io+0x176/0x2f0 [dm_mod] [ 791.643423] [<000003ff8074b28a>] do_reads+0x13a/0x1a8 [dm_mirror] [ 791.643425] [<000003ff8074b43a>] do_mirror+0x142/0x298 [dm_mirror] [ 791.643428] [<0000000000154fca>] process_one_work+0x18a/0x3f8 [ 791.643432] [<000000000015598a>] worker_thread+0x132/0x3b0 [ 791.643435] [<000000000015d49a>] kthread+0xd2/0xd8 [ 791.643438] [<00000000005bc0ca>] kernel_thread_starter+0x6/0xc [ 791.643446] [<00000000005bc0c4>] kernel_thread_starter+0x0/0xc Signed-off-by: Mike Snitzer Cc: stable@vger.kernel.org --- block/bio-integrity.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) NOTE: this serves as a more generic fix in the block layer rather than the dm-io specific fix which isn't ideal (due to potential for memory waste), see: https://git.kernel.org/cgit/linux/kernel/git/device-mapper/linux-dm.git/commit/?h=for-next&id=17dbe96d4f8a6f87004e6cfb5944872dfe2edb9f diff --git a/block/bio-integrity.c b/block/bio-integrity.c index 0436c21..719b715 100644 --- a/block/bio-integrity.c +++ b/block/bio-integrity.c @@ -51,7 +51,7 @@ struct bio_integrity_payload *bio_integrity_alloc(struct bio *bio, unsigned long idx = BIO_POOL_NONE; unsigned inline_vecs; - if (!bs) { + if (!bs || !bs->bio_integrity_pool) { bip = kmalloc(sizeof(struct bio_integrity_payload) + sizeof(struct bio_vec) * nr_vecs, gfp_mask); inline_vecs = nr_vecs; @@ -104,7 +104,7 @@ void bio_integrity_free(struct bio *bio) kfree(page_address(bip->bip_vec->bv_page) + bip->bip_vec->bv_offset); - if (bs) { + if (bs && bs->bio_integrity_pool) { if (bip->bip_slab != BIO_POOL_NONE) bvec_free(bs->bvec_integrity_pool, bip->bip_vec, bip->bip_slab); -- 2.3.2 (Apple Git-55)