diff options
author | Kent Overstreet <kent.overstreet@linux.dev> | 2023-05-22 00:49:06 -0400 |
---|---|---|
committer | Kent Overstreet <kent.overstreet@linux.dev> | 2023-10-22 17:10:01 -0400 |
commit | 0b438c5bfaebda3fdf6edc35d9572d4e2f66aef1 (patch) | |
tree | 21fe07b14d0b4cc365034eac529fae0f94c1fa65 /fs/bcachefs/btree_cache.c | |
parent | faa62a2036a491a919deffd980abc867be51b6f1 (diff) |
bcachefs: Clear btree_node_just_written() when node reused or evicted
This fixes the following bug:
Journal reclaim attempts to flush a node, but races with the node being
evicted from the btree node cache; when we lock the node, the data
buffers have already been freed.
We don't evict a node that's dirty, so calling btree_node_write() is
fine - it's a noop - except that the btree_node_just_written bit causes
bch2_btree_post_write_cleanup() to run (resorting the node), which then
causes a null ptr deref.
00078 Unable to handle kernel NULL pointer dereference at virtual address 000000000000009e
00078 Mem abort info:
00078 ESR = 0x0000000096000005
00078 EC = 0x25: DABT (current EL), IL = 32 bits
00078 SET = 0, FnV = 0
00078 EA = 0, S1PTW = 0
00078 FSC = 0x05: level 1 translation fault
00078 Data abort info:
00078 ISV = 0, ISS = 0x00000005
00078 CM = 0, WnR = 0
00078 user pgtable: 4k pages, 39-bit VAs, pgdp=000000007ed64000
00078 [000000000000009e] pgd=0000000000000000, p4d=0000000000000000, pud=0000000000000000
00078 Internal error: Oops: 0000000096000005 [#1] SMP
00078 Modules linked in:
00078 CPU: 75 PID: 1170 Comm: stress-ng-utime Not tainted 6.3.0-ktest-g5ef5b466e77e #2078
00078 Hardware name: linux,dummy-virt (DT)
00078 pstate: 80001005 (Nzcv daif -PAN -UAO -TCO -DIT +SSBS BTYPE=--)
00078 pc : btree_node_sort+0xc4/0x568
00078 lr : bch2_btree_post_write_cleanup+0x6c/0x1c0
00078 sp : ffffff803e30b350
00078 x29: ffffff803e30b350 x28: 0000000000000001 x27: ffffff80076e52a8
00078 x26: 0000000000000002 x25: 0000000000000000 x24: ffffffc00912e000
00078 x23: ffffff80076e52a8 x22: 0000000000000000 x21: ffffff80076e52bc
00078 x20: ffffff80076e5200 x19: 0000000000000000 x18: 0000000000000000
00078 x17: fffffffff8000000 x16: 0000000008000000 x15: 0000000008000000
00078 x14: 0000000000000002 x13: 0000000000000000 x12: 00000000000000a0
00078 x11: ffffff803e30b400 x10: ffffff803e30b408 x9 : 0000000000000001
00078 x8 : 0000000000000000 x7 : ffffff803e480000 x6 : 00000000000000a0
00078 x5 : 0000000000000088 x4 : 0000000000000000 x3 : 0000000000000010
00078 x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffffff80076e52a8
00078 Call trace:
00078 btree_node_sort+0xc4/0x568
00078 bch2_btree_post_write_cleanup+0x6c/0x1c0
00078 bch2_btree_node_write+0x108/0x148
00078 __btree_node_flush+0x104/0x160
00078 bch2_btree_node_flush0+0x1c/0x30
00078 journal_flush_pins.constprop.0+0x184/0x2d0
00078 __bch2_journal_reclaim+0x4d4/0x508
00078 bch2_journal_reclaim+0x1c/0x30
00078 __bch2_journal_preres_get+0x244/0x268
00078 bch2_trans_journal_preres_get_cold+0xa4/0x180
00078 __bch2_trans_commit+0x61c/0x1bb0
00078 bch2_setattr_nonsize+0x254/0x318
00078 bch2_setattr+0x5c/0x78
00078 notify_change+0x2bc/0x408
00078 vfs_utimes+0x11c/0x218
00078 do_utimes+0x84/0x140
00078 __arm64_sys_utimensat+0x68/0xa8
00078 invoke_syscall.constprop.0+0x54/0xf0
00078 do_el0_svc+0x48/0xd8
00078 el0_svc+0x14/0x48
00078 el0t_64_sync_handler+0xb0/0xb8
00078 el0t_64_sync+0x14c/0x150
00078 Code: 8b050265 910020c6 8b060266 910060ac (79402cad)
00078 ---[ end trace 0000000000000000 ]---
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Diffstat (limited to 'fs/bcachefs/btree_cache.c')
-rw-r--r-- | fs/bcachefs/btree_cache.c | 3 |
1 files changed, 3 insertions, 0 deletions
diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c index 46a8a29ddef7..76e08f2f6689 100644 --- a/fs/bcachefs/btree_cache.c +++ b/fs/bcachefs/btree_cache.c @@ -55,6 +55,8 @@ static void btree_node_data_free(struct bch_fs *c, struct btree *b) EBUG_ON(btree_node_write_in_flight(b)); + clear_btree_node_just_written(b); + kvpfree(b->data, btree_bytes(c)); b->data = NULL; #ifdef __KERNEL__ @@ -648,6 +650,7 @@ err: /* Try to cannibalize another cached btree node: */ if (bc->alloc_lock == current) { b2 = btree_node_cannibalize(c); + clear_btree_node_just_written(b2); bch2_btree_node_hash_remove(bc, b2); if (b) { |