Date:	Mon, 17 Jan 2011 17:29:42 +0900
From:	Akira Fujita <a-fujita@...jp.nec.com>
To:	Theodore Tso <tytso@....edu>
CC:	ext4 development <linux-ext4@...r.kernel.org>
Subject: [RFC][PATCH] ext4: Fix overflow of metadata block reservation counter

Hi,

The metadata block reservation counter overflows with data write
on ext4 (indirect block map) when its disk space is almost full.
This overflow triggers following BUG_ON.

Jan 14 09:36:48 TNESG9423 kernel: ------------[ cut here ]------------
Jan 14 09:36:48 TNESG9423 kernel: kernel BUG at fs/ext4/inode.c:2170!
Jan 14 09:36:48 TNESG9423 kernel: invalid opcode: 0000 [#1] SMP
Jan 14 09:36:48 TNESG9423 kernel: last sysfs file: /sys/kernel/mm/ksm/run
Jan 14 09:36:48 TNESG9423 kernel: CPU 0
Jan 14 09:36:48 TNESG9423 kernel: Modules linked in: ip6table_filter ip6_tables ebtable_nat ebtables xt_CHECKSUM iptable_mangle ipt_MASQUERADE iptable_nat nf_nat nf_conntrack_ipv4 nf_defrag_ipv4
xt_state nf_conntrack ipt_REJECT bridge stp llc autofs4 sunrpc p4_clockmod freq_table speedstep_lib ipv6 xt_physdev iptable_filter ip_tables nls_utf8 dm_mirror dm_region_hash dm_log dm_mod kvm_intel
kvm uinput ppdev parport_pc parport sg pcspkr i2c_i801 iTCO_wdt iTCO_vendor_support snd_hda_codec_realtek snd_hda_intel snd_hda_codec snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore
snd_page_alloc e1000e ext3 jbd sd_mod crc_t10dif sr_mod cdrom pata_via pata_acpi ata_generic ata_piix i915 drm_kms_helper drm i2c_algo_bit i2c_core video output [last unloaded: mperf]
Jan 14 09:36:48 TNESG9423 kernel:
Jan 14 09:36:48 TNESG9423 kernel: Pid: 937, comm: flush-8:0 Not tainted 2.6.37 #1 MS-7264BLM/PC-MJ18ABZR4
Jan 14 09:36:48 TNESG9423 kernel: RIP: 0010:[<ffffffff811c5298>]  [<ffffffff811c5298>] ext4_da_block_invalidatepages+0x168/0x180
Jan 14 09:36:48 TNESG9423 kernel: RSP: 0018:ffff88007613f780  EFLAGS: 00010246
Jan 14 09:36:48 TNESG9423 kernel: RAX: 0010000000000024 RBX: 0000000000008cf2 RCX: 000000000000000e
Jan 14 09:36:48 TNESG9423 kernel: RDX: 000000000000000e RSI: 0000000000000001 RDI: ffffea0000a70d30
Jan 14 09:36:48 TNESG9423 kernel: RBP: ffff88007613f850 R08: 0000000000000001 R09: 0000000000000002
Jan 14 09:36:48 TNESG9423 kernel: R10: ffffea0000a70d38 R11: ffff880035f01b58 R12: ffff88007613f7a0
Jan 14 09:36:48 TNESG9423 kernel: R13: ffff880065eecd68 R14: ffff88007613f7b8 R15: ffffea0000a70a58
Jan 14 09:36:48 TNESG9423 kernel: FS:  0000000000000000(0000) GS:ffff88007f400000(0000) knlGS:0000000000000000
Jan 14 09:36:48 TNESG9423 kernel: CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
Jan 14 09:36:48 TNESG9423 kernel: CR2: 0000003680ae1560 CR3: 000000004da45000 CR4: 00000000000006f0
Jan 14 09:36:48 TNESG9423 kernel: DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
Jan 14 09:36:48 TNESG9423 kernel: DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Jan 14 09:36:48 TNESG9423 kernel: Process flush-8:0 (pid: 937, threadinfo ffff88007613e000, task ffff8800371b54e0)
Jan 14 09:36:48 TNESG9423 kernel: Stack:
Jan 14 09:36:48 TNESG9423 kernel: ffff88007613f7e0 ffffffff814ee3d6 0000000000000008 0000000e7613f7f0
Jan 14 09:36:48 TNESG9423 kernel: 000000000000000e 000000003741a4b9 ffffea0000a70a58 ffffea0000a70a90
Jan 14 09:36:48 TNESG9423 kernel: ffffea0000a70ac8 ffffea0000a70b00 ffffea0000a70b38 ffffea0000a70b70
Jan 14 09:36:48 TNESG9423 kernel: Call Trace:
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff814ee3d6>] ? printk+0x41/0x43
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff811c9fe4>] mpage_da_map_and_submit+0x274/0x470
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff811ca24d>] mpage_add_bh_to_extent+0x6d/0xf0
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff811ca5a0>] write_cache_pages_da+0x2d0/0x4a0
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff811caa4c>] ext4_da_writepages+0x2dc/0x650
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff81105321>] do_writepages+0x21/0x40
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff810fb55b>] __filemap_fdatawrite_range+0x5b/0x60
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff810fb833>] filemap_fdatawrite_range+0x13/0x20
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff811fb6ce>] jbd2_journal_begin_ordered_truncate+0x8e/0xb0
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff811cd02b>] ext4_evict_inode+0x23b/0x3b0
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff81167477>] evict+0x27/0xc0
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff81167a4b>] iput+0x1bb/0x2a0
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff811734a4>] writeback_sb_inodes+0x104/0x180
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff81173cfd>] writeback_inodes_wb+0x9d/0x160
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff8117404b>] wb_writeback+0x28b/0x400
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff810714cc>] ? lock_timer_base+0x3c/0x70
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff81072092>] ? del_timer_sync+0x22/0x30
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff81174257>] wb_do_writeback+0x97/0x1e0
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff81174452>] bdi_writeback_thread+0xb2/0x270
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff811743a0>] ? bdi_writeback_thread+0x0/0x270
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff811743a0>] ? bdi_writeback_thread+0x0/0x270
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff81083356>] kthread+0x96/0xa0
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff8100ce84>] kernel_thread_helper+0x4/0x10
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff810832c0>] ? kthread+0x0/0xa0
Jan 14 09:36:48 TNESG9423 kernel: [<ffffffff8100ce80>] ? kernel_thread_helper+0x0/0x10
Jan 14 09:36:48 TNESG9423 kernel: Code: a8 00 00 00 5b 41 5c 41 5d 41 5e 41 5f c9 c3 0f 1f 40 00 4c 89 e7 48 89 95 40 ff ff ff e8 01 1b f4 ff 48 8b 95 40 ff ff ff eb c9 <0f> 0b eb fe 0f 0b 66 90 eb fc
66 66 66 66 66 2e 0f 1f 84 00 00
Jan 14 09:36:48 TNESG9423 kernel: RIP  [<ffffffff811c5298>] ext4_da_block_invalidatepages+0x168/0x180
Jan 14 09:36:48 TNESG9423 kernel: RSP <ffff88007613f780>
Jan 14 09:36:48 TNESG9423 kernel: ---[ end trace 0496eaed3b9ec629 ]---

To fix this, I referred to the patch which is for data blocks
reservation counter (commit: ef627929781c98113e6ae93f159dd3c12a884ad8)
and made a following patch which prints metadata block inconsistency
and corrects it.
My patch is trial, if you have better idea, feel free to fix this bug.

# You can reproduce this problem with attached programs.
# In my environment, this occurs in 1 minute.

Signed-off-by: Akira Fujita <a-fujita@...jp.nec.com>
---
 fs/ext4/inode.c |   10 ++++++++++
 1 file changed, 10 insertions(+)
diff -Nrup -X linux-2.6.37-org/Documentation/dontdiff linux-2.6.37-org/fs/ext4/inode.c linux-2.6.37/fs/ext4/inode.c
--- linux-2.6.37-org/fs/ext4/inode.c	2011-01-17 15:47:59.000000000 +0900
+++ linux-2.6.37/fs/ext4/inode.c	2011-01-17 15:52:25.000000000 +0900
@@ -1127,6 +1127,16 @@ void ext4_da_update_reserve_space(struct
 		used = ei->i_reserved_data_blocks;
 	}

+	if (unlikely(ei->i_allocated_meta_blocks >
+			ei->i_reserved_meta_blocks)) {
+		ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, "
+			"meta blocks %d with only %d reserved meta blocks\n",
+			__func__, inode->i_ino, ei->i_allocated_meta_blocks,
+			ei->i_reserved_meta_blocks);
+		WARN_ON(1);
+		ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
+	}
+
 	/* Update per-inode reservations */
 	ei->i_reserved_data_blocks -= used;
 	ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;

Download attachment "metadata_blocks_overflow.sh" of type "application/x-shellscript" (542 bytes)

View attachment "fs_write.c" of type "text/plain" (5703 bytes)

Powered by blists - more mailing lists