From: Kent Overstreet Date: Thu, 18 Aug 2016 01:21:24 +0000 (-0700) Subject: bcache: RESERVE_PRIO is too small by one when prio_buckets() is a power of two. X-Git-Url: http://drtracing.org/?a=commitdiff_plain;h=acc9cf8c66c66b2cbbdb4a375537edee72be64df;p=deliverable%2Flinux.git bcache: RESERVE_PRIO is too small by one when prio_buckets() is a power of two. This patch fixes a cachedev registration-time allocation deadlock. This can deadlock on boot if your initrd auto-registeres bcache devices: Allocator thread: [ 720.727614] INFO: task bcache_allocato:3833 blocked for more than 120 seconds. [ 720.732361] [] schedule+0x37/0x90 [ 720.732963] [] bch_bucket_alloc+0x188/0x360 [bcache] [ 720.733538] [] ? prepare_to_wait_event+0xf0/0xf0 [ 720.734137] [] bch_prio_write+0x19d/0x340 [bcache] [ 720.734715] [] bch_allocator_thread+0x3ff/0x470 [bcache] [ 720.735311] [] ? __schedule+0x2dc/0x950 [ 720.735884] [] ? invalidate_buckets+0x980/0x980 [bcache] Registration thread: [ 720.710403] INFO: task bash:3531 blocked for more than 120 seconds. [ 720.715226] [] schedule+0x37/0x90 [ 720.715805] [] __bch_btree_map_nodes+0x12d/0x150 [bcache] [ 720.716409] [] ? bch_btree_insert_check_key+0x1c0/0x1c0 [bcache] [ 720.717008] [] bch_btree_insert+0xf4/0x170 [bcache] [ 720.717586] [] ? prepare_to_wait_event+0xf0/0xf0 [ 720.718191] [] bch_journal_replay+0x14a/0x290 [bcache] [ 720.718766] [] ? ttwu_do_activate.constprop.94+0x5d/0x70 [ 720.719369] [] ? try_to_wake_up+0x1d4/0x350 [ 720.719968] [] run_cache_set+0x580/0x8e0 [bcache] [ 720.720553] [] register_bcache+0xe2e/0x13b0 [bcache] [ 720.721153] [] kobj_attr_store+0xf/0x20 [ 720.721730] [] sysfs_kf_write+0x3d/0x50 [ 720.722327] [] kernfs_fop_write+0x12a/0x180 [ 720.722904] [] __vfs_write+0x37/0x110 [ 720.723503] [] ? __sb_start_write+0x58/0x110 [ 720.724100] [] ? security_file_permission+0x23/0xa0 [ 720.724675] [] vfs_write+0xa9/0x1b0 [ 720.725275] [] ? do_audit_syscall_entry+0x6c/0x70 [ 720.725849] [] SyS_write+0x55/0xd0 [ 720.726451] [] ? do_page_fault+0x30/0x80 [ 720.727045] [] system_call_fastpath+0x12/0x71 The fifo code in upstream bcache can't use the last element in the buffer, which was the cause of the bug: if you asked for a power of two size, it'd give you a fifo that could hold one less than what you asked for rather than allocating a buffer twice as big. Signed-off-by: Kent Overstreet Tested-by: Eric Wheeler Cc: stable@vger.kernel.org --- diff --git a/drivers/md/bcache/super.c b/drivers/md/bcache/super.c index 6ada14b9a157..6b93e1b77767 100644 --- a/drivers/md/bcache/super.c +++ b/drivers/md/bcache/super.c @@ -1820,7 +1820,7 @@ static int cache_alloc(struct cache *ca) free = roundup_pow_of_two(ca->sb.nbuckets) >> 10; if (!init_fifo(&ca->free[RESERVE_BTREE], 8, GFP_KERNEL) || - !init_fifo(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) || + !init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) || !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) || !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) || !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) ||