From mboxrd@z Thu Jan 1 00:00:00 1970 From: Andrew Morton Date: Wed, 11 Mar 2020 18:40:31 -0700 Subject: [Ocfs2-devel] [PATCH] ocfs2: fix a null pointer derefrence in ocfs2_block_group_clear_bits() In-Reply-To: <32e99737-b992-d9b7-5a97-a17168dde88f@huawei.com> References: <1d38573d-61c7-be60-334e-c263caf7465c@huawei.com> <32e99737-b992-d9b7-5a97-a17168dde88f@huawei.com> Message-ID: <20200311184031.a850f8e3dbd664b599654303@linux-foundation.org> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: ocfs2-devel@oss.oracle.com On Tue, 10 Mar 2020 21:07:06 +0800 lishan wrote: > On 2020/3/9 16:26, lishan wrote: > > A NULL pointer panic dereference in ocfs2_block_group_clear_bits() happen again. > > The information of NULL pointer stack as follows: > > > > PID: 81866 TASK: ffffa07c3c21ae80 CPU: 66 COMMAND: "fallocate" > > #0 [ffff0000b4d6b0b0] machine_kexec at ffff0000800a2954 > > #1 [ffff0000b4d6b110] __crash_kexec at ffff0000801bab34 > > #2 [ffff0000b4d6b2a0] panic at ffff0000800f02cc > > #3 [ffff0000b4d6b380] die at ffff00008008f6ac > > #4 [ffff0000b4d6b3d0] bug_handler at ffff00008008f744 > > #5 [ffff0000b4d6b400] brk_handler at ffff000080085d1c > > #6 [ffff0000b4d6b420] do_debug_exception at ffff000080081194 > > #7 [ffff0000b4d6b630] el1_dbg at ffff00008008332c > > PC: ffff00000190e9c0 [_ocfs2_free_suballoc_bits+1608] > > LR: ffff00000190e990 [_ocfs2_free_suballoc_bits+1560] > > SP: ffff0000b4d6b640 PSTATE: 60400009 > > X29: ffff0000b4d6b650 X28: 0000000000000000 X27: 00000000000052f3 > > X26: ffff807c511a9570 X25: ffff807ca0054000 X24: 00000000000052f2 > > X23: 0000000000000001 X22: ffff807c7cde7a90 X21: ffff0000811d9000 > > X20: ffff807c5e7d2000 X19: ffff00000190c768 X18: 0000000000000000 > > X17: 0000000000000000 X16: ffff000080a032f0 X15: 0000000000000000 > > X14: ffffffffffffffff X13: fffffffffffffff7 X12: ffffffffffffffff > > X11: 0000000000000038 X10: 0101010101010101 X9: ffffffffffffffff > > X8: 7f7f7f7f7f7f7f7f X7: 0000000000000000 X6: 0000000000000080 > > X5: 0000000000000000 X4: 0000000000000002 X3: ffff00000199f390 > > X2: a603c08321456e00 X1: ffff807c7cde7a90 X0: 0000000000000000 > > #8 [ffff0000b4d6b650] _ocfs2_free_suballoc_bits at ffff00000190e9bc [ocfs2] > > #9 [ffff0000b4d6b710] _ocfs2_free_clusters at ffff0000019110d4 [ocfs2] > > #10 [ffff0000b4d6b790] ocfs2_free_clusters at ffff000001913e94 [ocfs2] > > #11 [ffff0000b4d6b7d0] __ocfs2_flush_truncate_log at ffff0000018b5294 [ocfs2] > > #12 [ffff0000b4d6b8a0] ocfs2_remove_btree_range at ffff0000018bb34c [ocfs2] > > #13 [ffff0000b4d6b960] ocfs2_commit_truncate at ffff0000018bc76c [ocfs2] > > #14 [ffff0000b4d6ba60] ocfs2_wipe_inode at ffff0000018e57bc [ocfs2] > > #15 [ffff0000b4d6bb00] ocfs2_evict_inode at ffff0000018e5db8 [ocfs2] > > #16 [ffff0000b4d6bb70] evict at ffff000080365040 > > #17 [ffff0000b4d6bba0] iput at ffff0000803655d8 > > #18 [ffff0000b4d6bbe0] ocfs2_dentry_iput at ffff0000018c60a0 [ocfs2] > > #19 [ffff0000b4d6bc30] dentry_unlink_inode at ffff00008035ef58 > > #20 [ffff0000b4d6bc50] __dentry_kill at ffff000080360384 > > #21 [ffff0000b4d6bc80] dentry_kill at ffff000080360670 > > #22 [ffff0000b4d6bcb0] dput at ffff00008036093c > > #23 [ffff0000b4d6bcf0] __fput at ffff000080343930 > > #24 [ffff0000b4d6bd40] ____fput at ffff000080343aac > > #25 [ffff0000b4d6bd60] task_work_run at ffff0000801172fc > > > > The direct panic reason is that bh2jh (group_bh)-> b_committed_data is null. > > It is presumed that the network was disconnected during the write process, > > causing the transaction abort. as follows: > > jbd2_journal_abort > > ....... > > jbd2_journal_commit_transaction > > jh->b_committed_data = NULL; > > > > _ocfs2_free_suballoc_bits > > ocfs2_block_group_clear_bits > > // undo_bg is now set to null > > BUG_ON(!undo_bg); > > > > When applying for free space, if b_committed_data is null, > > it will be directly occupied, as follows: > > ocfs2_cluster_group_search > > ocfs2_block_group_find_clear_bits > > ocfs2_test_bg_bit_allocatable: > > bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data; > > if (bg) > > ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap); > > else > > ret = 1; > > b_committed_data is an intermediate state backup for bitmap transaction commits, > > newly applied space can overwrite previous dirty data, > > so, I think, while free clusters, if b_committed_data is null, ignore it. > > Host panic directly, too violent. (top-posting repaired) > ping ? > cc Andrew Morton There's something dreadfully wrong with the ocfs2-devel mailing list. I almost never receive patches when people add me to cc. I've never seen much of a pattern to it - it just drops stuff everywhere. Who is the admin for this list? Can we please move it to kernel.org? Regarding this patch, I did actually receive one email from the server. >From Joseph: : > + if (!undo_bg) : > + mlog(ML_NOTICE, "%s: group descriptor # %llu (device %s) journal " : > + "b_committed_data had been cleared.\n", : > + OCFS2_SB(alloc_inode->i_sb)->uuid_str, : > + (unsigned long long)le64_to_cpu(bg->bg_blkno), : > + alloc_inode->i_sb->s_id); : : Seems a kind of workaround. : I am worrying about other abnormal cases of NULL b_committed_data, it : may lead to a corrupt filesystem. : So how about isolating the journal abort case?