From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from titan.nuclearwinter.com ([174.136.96.186]:36501 "EHLO mail.nuclearwinter.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1751450AbaIXP3J (ORCPT ); Wed, 24 Sep 2014 11:29:09 -0400 Received: from [IPv6:2601:e:1200:11d0:b2c5:54ff:feff:f401] ([IPv6:2601:e:1200:11d0:b2c5:54ff:feff:f401]) (authenticated bits=0) by mail.nuclearwinter.com (8.14.4/8.14.4) with ESMTP id s8OFT7rs032074 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES128-SHA bits=128 verify=NO) for ; Wed, 24 Sep 2014 10:29:08 -0500 Message-ID: <5422E342.8060000@nuclearwinter.com> Date: Wed, 24 Sep 2014 10:29:06 -0500 From: Larkin Lowrey MIME-Version: 1.0 To: linux-btrfs@vger.kernel.org Subject: Re: btrfsck check infinite loop References: <5422D4E0.8090605@nuclearwinter.com> In-Reply-To: <5422D4E0.8090605@nuclearwinter.com> Content-Type: text/plain; charset=ISO-8859-1 Sender: linux-btrfs-owner@vger.kernel.org List-ID: I noticed the following: (gdb) print nrscan $19 = 1680726970 (gdb) print tree->cache_size $20 = 1073741824 (gdb) print cache_hard_max $21 = 1073741824 It appears that cache_size can not shrink below cache_hard_max so we never end up breaking out of the loop. The FS in question is 30TB with ~26TB in use. Perhaps cache_hard_max (1GB) is too small for this size FS? I just bumped it to 2GB and am re-running to see if that helps. --Larkin On 9/24/2014 9:27 AM, Larkin Lowrey wrote: > I ran 'btrfs check --repair --init-extent-tree' and appear to be in an > infinite loop. It performed heavy IO for about 1.5 hours then the IO > stopped and the CPU stayed at 100%. It's been like that for more than 12 > hours now. > > I made a hardware change last week that resulted in unstable RAM so I > suspect some corrupt data was written to disk. I tried mounting with > -orecovery,clear_cache,nospace_cache but I would get a panic shortly > thereafter. I tried 'btrfs check --repair' but also got a panic. I > finally tried 'btrfs check --repair --init-extent-tree' and hit an > assertion failed error with btrfs-progs 3.16. > > After noticing some promising commits, I built from the integration repo > (kdave), re-ran (v3.16.1) and got further (2hrs) but then got stuck in > this infinite loop. > > Here's the backtrace of where it is now and has been for hours: > > #0 0x0000000000438f01 in free_some_buffers (tree=0xda3078) at > extent_io.c:553 > #1 __alloc_extent_buffer (blocksize=4096, bytenr=, > tree=0xda3078) at extent_io.c:592 > #2 alloc_extent_buffer (tree=0xda3078, bytenr=, > blocksize=4096) at extent_io.c:671 > #3 0x000000000042be29 in btrfs_find_create_tree_block > (root=root@entry=0xda34a0, bytenr=, blocksize= out>) at disk-io.c:133 > #4 0x000000000042d683 in read_tree_block (root=0xda34a0, > bytenr=, blocksize=, > parent_transid=161580) at disk-io.c:260 > #5 0x0000000000427c58 in read_node_slot (root=root@entry=0xda34a0, > parent=parent@entry=0x165ab88c0, slot=slot@entry=43) at ctree.c:634 > #6 0x0000000000428558 in push_leaf_right (trans=trans@entry=0xe709b0, > root=root@entry=0xda34a0, path=path@entry=0xde317a0, > data_size=data_size@entry=67, empty=empty@entry=0) > at ctree.c:1608 > #7 0x0000000000428e4c in split_leaf (trans=trans@entry=0xe709b0, > root=root@entry=0xda34a0, ins_key=ins_key@entry=0x7fff24da24b0, > path=path@entry=0xde317a0, > data_size=data_size@entry=67, extend=extend@entry=0) at ctree.c:1977 > #8 0x000000000042aa54 in btrfs_search_slot (trans=0xe709b0, > root=root@entry=0xda34a0, key=key@entry=0x7fff24da24b0, > p=p@entry=0xde317a0, ins_len=ins_len@entry=67, > cow=cow@entry=1) at ctree.c:1120 > #9 0x000000000042af51 in btrfs_insert_empty_items > (trans=trans@entry=0xe709b0, root=root@entry=0xda34a0, > path=path@entry=0xde317a0, cpu_key=cpu_key@entry=0x7fff24da24b0, > data_size=data_size@entry=0x7fff24da24a0, nr=nr@entry=1) at ctree.c:2412 > #10 0x00000000004175f6 in btrfs_insert_empty_item (data_size=42, > key=0x7fff24da24b0, path=0xde317a0, root=0xda34a0, trans=0xe709b0) at > ctree.h:2312 > #11 record_extent (flags=0, allocated=, back=0x95cb3d90, > rec=0x95cb3cc0, path=0xde317a0, info=0xda3010, trans=0xe709b0) at > cmds-check.c:4438 > #12 fixup_extent_refs (trans=trans@entry=0xe709b0, info=, > extent_cache=extent_cache@entry=0x7fff24da2970, > rec=rec@entry=0x95cb3cc0) at cmds-check.c:5287 > #13 0x000000000041ac01 in check_extent_refs > (extent_cache=0x7fff24da2970, root=, trans= out>) at cmds-check.c:5511 > #14 check_chunks_and_extents (root=root@entry=0xfa7c70) at cmds-check.c:5978 > #15 0x000000000041bdd9 in cmd_check (argc=, > argv=) at cmds-check.c:6723 > #16 0x0000000000404481 in main (argc=4, argv=0x7fff24da2fe0) at btrfs.c:247 > > I checked node, node->next, node->next->next, node->next->prev, etc. and > saw no obvious loop, at least not in the immediate vicinity of node. The > value of node is different each time I check it. > > I'll periodically see the following backtrace: > > #0 __list_del (next=0x1326fe820, prev=0xda3088) at list.h:113 > #1 list_move_tail (head=0xda3088, list=0x1514b40f0) at list.h:183 > #2 free_some_buffers (tree=0xda3078) at extent_io.c:560 > #3 __alloc_extent_buffer (blocksize=4096, bytenr=, > tree=0xda3078) at extent_io.c:592 > #4 alloc_extent_buffer (tree=0xda3078, bytenr=, > blocksize=4096) at extent_io.c:671 > #5 0x000000000042be29 in btrfs_find_create_tree_block > (root=root@entry=0xda34a0, bytenr=, blocksize= out>) at disk-io.c:133 > #6 0x000000000042d683 in read_tree_block (root=0xda34a0, > bytenr=, blocksize=, > parent_transid=161580) at disk-io.c:260 > #7 0x0000000000427c58 in read_node_slot (root=root@entry=0xda34a0, > parent=parent@entry=0x165ab88c0, slot=slot@entry=43) at ctree.c:634 > #8 0x0000000000428558 in push_leaf_right (trans=trans@entry=0xe709b0, > root=root@entry=0xda34a0, path=path@entry=0xde317a0, > data_size=data_size@entry=67, empty=empty@entry=0) > at ctree.c:1608 > #9 0x0000000000428e4c in split_leaf (trans=trans@entry=0xe709b0, > root=root@entry=0xda34a0, ins_key=ins_key@entry=0x7fff24da24b0, > path=path@entry=0xde317a0, > data_size=data_size@entry=67, extend=extend@entry=0) at ctree.c:1977 > #10 0x000000000042aa54 in btrfs_search_slot (trans=0xe709b0, > root=root@entry=0xda34a0, key=key@entry=0x7fff24da24b0, > p=p@entry=0xde317a0, ins_len=ins_len@entry=67, > cow=cow@entry=1) at ctree.c:1120 > #11 0x000000000042af51 in btrfs_insert_empty_items > (trans=trans@entry=0xe709b0, root=root@entry=0xda34a0, > path=path@entry=0xde317a0, cpu_key=cpu_key@entry=0x7fff24da24b0, > data_size=data_size@entry=0x7fff24da24a0, nr=nr@entry=1) at ctree.c:2412 > #12 0x00000000004175f6 in btrfs_insert_empty_item (data_size=42, > key=0x7fff24da24b0, path=0xde317a0, root=0xda34a0, trans=0xe709b0) at > ctree.h:2312 > #13 record_extent (flags=0, allocated=, back=0x95cb3d90, > rec=0x95cb3cc0, path=0xde317a0, info=0xda3010, trans=0xe709b0) at > cmds-check.c:4438 > #14 fixup_extent_refs (trans=trans@entry=0xe709b0, info=, > extent_cache=extent_cache@entry=0x7fff24da2970, > rec=rec@entry=0x95cb3cc0) at cmds-check.c:5287 > #15 0x000000000041ac01 in check_extent_refs > (extent_cache=0x7fff24da2970, root=, trans= out>) at cmds-check.c:5511 > #16 check_chunks_and_extents (root=root@entry=0xfa7c70) at cmds-check.c:5978 > #17 0x000000000041bdd9 in cmd_check (argc=, > argv=) at cmds-check.c:6723 > #18 0x0000000000404481 in main (argc=4, argv=0x7fff24da2fe0) at btrfs.c:247 > > If there's interest in debugging I can leave this machine in this > condition for a few days. It's just a backup server so losing the fs > won't be the end of the world. > > --Larkin >