kernel_optimize_test/fs/btrfs/tree-defrag.c
Filipe Manana 0376374a98 Btrfs: fix locking bugs when defragging leaves
When running fstests btrfs/070, with a higher number of fsstress
operations, I ran frequently into two different locking bugs when
defragging directories.

The first bug produced the following traces:

[133860.229792] ------------[ cut here ]------------
[133860.251062] WARNING: CPU: 2 PID: 26057 at fs/btrfs/locking.c:46 btrfs_set_lock_blocking_rw+0x57/0xbd [btrfs]()
[133860.253576] Modules linked in: btrfs crc32c_generic xor raid6_pq nfsd auth_rpcgss oid_registry nfs_acl nfs lockd grace fscache sunrpc loop fuse parport_pc i2c_piix4 psmouse parport
[133860.282566] CPU: 2 PID: 26057 Comm: btrfs Tainted: G        W       4.3.0-rc5-btrfs-next-17+ #1
[133860.284393] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.1-0-g4adadbd-20150316_085822-nilsson.home.kraxel.org 04/01/2014
[133860.286827]  0000000000000000 ffff880207697b78 ffffffff812566f4 0000000000000000
[133860.288341]  ffff880207697bb0 ffffffff8104d0a6 ffffffffa052d4c1 ffff880178f60e00
[133860.294219]  ffff880178f60e00 0000000000000000 00000000000000f6 ffff880207697bc0
[133860.295831] Call Trace:
[133860.306518]  [<ffffffff812566f4>] dump_stack+0x4e/0x79
[133860.307473]  [<ffffffff8104d0a6>] warn_slowpath_common+0x9f/0xb8
[133860.308619]  [<ffffffffa052d4c1>] ? btrfs_set_lock_blocking_rw+0x57/0xbd [btrfs]
[133860.310068]  [<ffffffff8104d172>] warn_slowpath_null+0x1a/0x1c
[133860.312552]  [<ffffffffa052d4c1>] btrfs_set_lock_blocking_rw+0x57/0xbd [btrfs]
[133860.314630]  [<ffffffffa04d5787>] btrfs_set_lock_blocking+0xe/0x10 [btrfs]
[133860.323596]  [<ffffffffa04d99cb>] btrfs_realloc_node+0xb3/0x341 [btrfs]
[133860.325233]  [<ffffffffa050e396>] btrfs_defrag_leaves+0x239/0x2fa [btrfs]
[133860.332427]  [<ffffffffa04fc2ce>] btrfs_defrag_root+0x63/0xca [btrfs]
[133860.337259]  [<ffffffffa052a34e>] btrfs_ioctl_defrag+0x78/0x14e [btrfs]
[133860.340147]  [<ffffffffa052b00b>] btrfs_ioctl+0x746/0x24c6 [btrfs]
[133860.344833]  [<ffffffff81087481>] ? arch_local_irq_save+0x9/0xc
[133860.346343]  [<ffffffff8113ad61>] ? __might_fault+0x4c/0xa7
[133860.353248]  [<ffffffff8113ad61>] ? __might_fault+0x4c/0xa7
[133860.354242]  [<ffffffff8113adba>] ? __might_fault+0xa5/0xa7
[133860.355232]  [<ffffffff81171139>] ? cp_new_stat+0x15d/0x174
[133860.356237]  [<ffffffff8117c610>] do_vfs_ioctl+0x427/0x4e6
[133860.358587]  [<ffffffff81171175>] ? SYSC_newfstat+0x25/0x2e
[133860.360195]  [<ffffffff8118574d>] ? __fget_light+0x4d/0x71
[133860.361380]  [<ffffffff8117c726>] SyS_ioctl+0x57/0x79
[133860.363578]  [<ffffffff8147cd97>] entry_SYSCALL_64_fastpath+0x12/0x6f
[133860.366217] ---[ end trace 2cadb2f653437e49 ]---
[133860.367399] ------------[ cut here ]------------
[133860.368162] kernel BUG at fs/btrfs/locking.c:307!
[133860.369430] invalid opcode: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
[133860.370205] Modules linked in: btrfs crc32c_generic xor raid6_pq nfsd auth_rpcgss oid_registry nfs_acl nfs lockd grace fscache sunrpc loop fuse parport_pc i2c_piix4 psmouse parport
[133860.370205] CPU: 2 PID: 26057 Comm: btrfs Tainted: G        W       4.3.0-rc5-btrfs-next-17+ #1
[133860.370205] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.1-0-g4adadbd-20150316_085822-nilsson.home.kraxel.org 04/01/2014
[133860.370205] task: ffff8800aec6db40 ti: ffff880207694000 task.ti: ffff880207694000
[133860.370205] RIP: 0010:[<ffffffffa052d466>]  [<ffffffffa052d466>] btrfs_assert_tree_locked+0x10/0x14 [btrfs]
[133860.370205] RSP: 0018:ffff880207697bc0  EFLAGS: 00010246
[133860.370205] RAX: 0000000000000000 RBX: ffff880178f60e00 RCX: 0000000000000000
[133860.370205] RDX: ffff88023ec4fb50 RSI: 00000000ffffffff RDI: ffff880178f60e00
[133860.370205] RBP: ffff880207697bc0 R08: 0000000000000001 R09: 0000000000000000
[133860.370205] R10: 0000160000000000 R11: ffffffff81651000 R12: ffff880178f60e00
[133860.370205] R13: 0000000000000000 R14: 00000000000000f6 R15: ffff8801ff409000
[133860.370205] FS:  00007f763efd48c0(0000) GS:ffff88023ec40000(0000) knlGS:0000000000000000
[133860.370205] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[133860.370205] CR2: 0000000002158048 CR3: 000000003fd6c000 CR4: 00000000000006e0
[133860.370205] Stack:
[133860.370205]  ffff880207697bd8 ffffffffa052d4d0 0000000000000000 ffff880207697be8
[133860.370205]  ffffffffa04d5787 ffff880207697c80 ffffffffa04d99cb ffff8801ff409590
[133860.370205]  ffff880207697ca8 000000f507697c80 ffff880183c11bb8 0000000000000000
[133860.370205] Call Trace:
[133860.370205]  [<ffffffffa052d4d0>] btrfs_set_lock_blocking_rw+0x66/0xbd [btrfs]
[133860.370205]  [<ffffffffa04d5787>] btrfs_set_lock_blocking+0xe/0x10 [btrfs]
[133860.370205]  [<ffffffffa04d99cb>] btrfs_realloc_node+0xb3/0x341 [btrfs]
[133860.370205]  [<ffffffffa050e396>] btrfs_defrag_leaves+0x239/0x2fa [btrfs]
[133860.370205]  [<ffffffffa04fc2ce>] btrfs_defrag_root+0x63/0xca [btrfs]
[133860.370205]  [<ffffffffa052a34e>] btrfs_ioctl_defrag+0x78/0x14e [btrfs]
[133860.370205]  [<ffffffffa052b00b>] btrfs_ioctl+0x746/0x24c6 [btrfs]
[133860.370205]  [<ffffffff81087481>] ? arch_local_irq_save+0x9/0xc
[133860.370205]  [<ffffffff8113ad61>] ? __might_fault+0x4c/0xa7
[133860.370205]  [<ffffffff8113ad61>] ? __might_fault+0x4c/0xa7
[133860.370205]  [<ffffffff8113adba>] ? __might_fault+0xa5/0xa7
[133860.370205]  [<ffffffff81171139>] ? cp_new_stat+0x15d/0x174
[133860.370205]  [<ffffffff8117c610>] do_vfs_ioctl+0x427/0x4e6
[133860.370205]  [<ffffffff81171175>] ? SYSC_newfstat+0x25/0x2e
[133860.370205]  [<ffffffff8118574d>] ? __fget_light+0x4d/0x71
[133860.370205]  [<ffffffff8117c726>] SyS_ioctl+0x57/0x79
[133860.370205]  [<ffffffff8147cd97>] entry_SYSCALL_64_fastpath+0x12/0x6f

This bug happened because we assumed that by setting keep_locks to 1 in
our search path, our path after a call to btrfs_search_slot() would have
all nodes locked, which is not always true because unlock_up() (called by
btrfs_search_slot()) will unlock a node in a path if the slot of the node
below it doesn't point to the last item or beyond the last item. For
example, when the tree has a heigth of 2 and path->slots[0] has a value
smaller than btrfs_header_nritems(path->nodes[0]) - 1, the node at level 2
will be unlocked (also because lowest_unlock is set to 1 due to the fact
that the value passed as ins_len to btrfs_search_slot is 0).
This resulted in btrfs_find_next_key(), called before btrfs_realloc_node(),
to release out path and call again btrfs_search_slot(), but this time with
the cow parameter set to 0, meaning the resulting path got only read locks.
Therefore when we called btrfs_realloc_node(), with path->nodes[1] having
a read lock, it resulted in the warning and BUG_ON when calling
btrfs_set_lock_blocking() against the node, as that function expects the
node to have a write lock.

The second bug happened often when the first bug didn't happen, and made
us hang and hitting the following warning at fs/btrfs/locking.c:

   251  void btrfs_tree_lock(struct extent_buffer *eb)
   252  {
   253          WARN_ON(eb->lock_owner == current->pid);

This happened because the tree search we made at btrfs_defrag_leaves()
before calling btrfs_find_next_key() locked a leaf and all the other
nodes in the path, so btrfs_find_next_key() had no need to release the
path and make a new search (with path->lowest_level set to 1). This
made btrfs_realloc_node() attempt to write lock the same leaf again,
resulting in a hang/deadlock.

So fix these issues by calling btrfs_find_next_key() after calling
btrfs_realloc_node() and setting the search path's lowest_level to 1
to avoid the hang/deadlock when attempting to write lock the leaves
at btrfs_realloc_node().

Signed-off-by: Filipe Manana <fdmanana@suse.com>
2015-12-18 02:51:32 +00:00

157 lines
4.2 KiB
C

/*
* Copyright (C) 2007 Oracle. All rights reserved.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License v2 as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public
* License along with this program; if not, write to the
* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
* Boston, MA 021110-1307, USA.
*/
#include <linux/sched.h>
#include "ctree.h"
#include "disk-io.h"
#include "print-tree.h"
#include "transaction.h"
#include "locking.h"
/*
* Defrag all the leaves in a given btree.
* Read all the leaves and try to get key order to
* better reflect disk order
*/
int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
struct btrfs_path *path = NULL;
struct btrfs_key key;
int ret = 0;
int wret;
int level;
int next_key_ret = 0;
u64 last_ret = 0;
u64 min_trans = 0;
if (root->fs_info->extent_root == root) {
/*
* there's recursion here right now in the tree locking,
* we can't defrag the extent root without deadlock
*/
goto out;
}
if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
goto out;
path = btrfs_alloc_path();
if (!path)
return -ENOMEM;
level = btrfs_header_level(root->node);
if (level == 0)
goto out;
if (root->defrag_progress.objectid == 0) {
struct extent_buffer *root_node;
u32 nritems;
root_node = btrfs_lock_root_node(root);
btrfs_set_lock_blocking(root_node);
nritems = btrfs_header_nritems(root_node);
root->defrag_max.objectid = 0;
/* from above we know this is not a leaf */
btrfs_node_key_to_cpu(root_node, &root->defrag_max,
nritems - 1);
btrfs_tree_unlock(root_node);
free_extent_buffer(root_node);
memset(&key, 0, sizeof(key));
} else {
memcpy(&key, &root->defrag_progress, sizeof(key));
}
path->keep_locks = 1;
ret = btrfs_search_forward(root, &key, path, min_trans);
if (ret < 0)
goto out;
if (ret > 0) {
ret = 0;
goto out;
}
btrfs_release_path(path);
/*
* We don't need a lock on a leaf. btrfs_realloc_node() will lock all
* leafs from path->nodes[1], so set lowest_level to 1 to avoid later
* a deadlock (attempting to write lock an already write locked leaf).
*/
path->lowest_level = 1;
wret = btrfs_search_slot(trans, root, &key, path, 0, 1);
if (wret < 0) {
ret = wret;
goto out;
}
if (!path->nodes[1]) {
ret = 0;
goto out;
}
/*
* The node at level 1 must always be locked when our path has
* keep_locks set and lowest_level is 1, regardless of the value of
* path->slots[1].
*/
BUG_ON(path->locks[1] == 0);
ret = btrfs_realloc_node(trans, root,
path->nodes[1], 0,
&last_ret,
&root->defrag_progress);
if (ret) {
WARN_ON(ret == -EAGAIN);
goto out;
}
/*
* Now that we reallocated the node we can find the next key. Note that
* btrfs_find_next_key() can release our path and do another search
* without COWing, this is because even with path->keep_locks = 1,
* btrfs_search_slot() / ctree.c:unlock_up() does not keeps a lock on a
* node when path->slots[node_level - 1] does not point to the last
* item or a slot beyond the last item (ctree.c:unlock_up()). Therefore
* we search for the next key after reallocating our node.
*/
path->slots[1] = btrfs_header_nritems(path->nodes[1]);
next_key_ret = btrfs_find_next_key(root, path, &key, 1,
min_trans);
if (next_key_ret == 0) {
memcpy(&root->defrag_progress, &key, sizeof(key));
ret = -EAGAIN;
}
out:
btrfs_free_path(path);
if (ret == -EAGAIN) {
if (root->defrag_max.objectid > root->defrag_progress.objectid)
goto done;
if (root->defrag_max.type > root->defrag_progress.type)
goto done;
if (root->defrag_max.offset > root->defrag_progress.offset)
goto done;
ret = 0;
}
done:
if (ret != -EAGAIN) {
memset(&root->defrag_progress, 0,
sizeof(root->defrag_progress));
root->defrag_trans_start = trans->transid;
}
return ret;
}