Currently, inode flags are fully backwards incompatible in btrfs. If we
introduce a new inode flag, then tree-checker will detect it and fail.
This can even cause us to fail to mount entirely. To make it possible to
introduce new flags which can be read-only compatible, like VERITY, we
add new ro flags to btrfs without treating them quite so harshly in
tree-checker. A read-only file system can survive an unexpected flag,
and can be mounted.
As for the implementation, it unfortunately gets a little complicated.
The on-disk representation of the inode, btrfs_inode_item, has an __le64
for flags but the in-memory representation, btrfs_inode, uses a u32.
David Sterba had the nice idea that we could reclaim those wasted 32 bits
on disk and use them for the new ro_compat flags.
It turns out that the tree-checker code which checks for unknown flags
is broken, and ignores the upper 32 bits we are hoping to use. The issue
is that the flags use the literal 1 rather than 1ULL, so the flags are
signed ints, and one of them is specifically (1 << 31). As a result, the
mask which ORs the flags is a negative integer on machines where int is
32 bit twos complement. When tree-checker evaluates the expression:
btrfs_inode_flags(leaf, iitem) & ~BTRFS_INODE_FLAG_MASK)
The mask is something like 0x80000abc, which gets promoted to u64 with
sign extension to 0xffffffff80000abc. Negating that 64 bit mask leaves
all the upper bits zeroed, and we can't detect unexpected flags.
This suggests that we can't use those bits after all. Luckily, we have
good reason to believe that they are zero anyway. Inode flags are
metadata, which is always checksummed, so any bit flips that would
introduce 1s would cause a checksum failure anyway (excluding the
improbable case of the checksum getting corrupted exactly badly).
Further, unless the 1 << 31 flag is used, the cast to u64 of the 32 bit
inode flag should preserve its value and not add leading zeroes
(at least for twos complement). The only place that flag
(BTRFS_INODE_ROOT_ITEM_INIT) is used is in a special inode embedded in
the root item, and indeed for that inode we see 0xffffffff80000000 as
the flags on disk. However, that inode is never seen by tree checker,
nor is it used in a context where verity might be meaningful.
Theoretically, a future ro flag might cause trouble on that inode, so we
should proactively clean up that mess before it does.
With the introduction of the new ro flags, keep two separate unsigned
masks and check them against the appropriate u32. Since we no longer run
afoul of sign extension, this also stops writing out 0xffffffff80000000
in root_item inodes going forward.
Signed-off-by: Boris Burkov <boris@bur.io>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
In commit 351cbf6e44 ("btrfs: use nofs allocations for running delayed
items") we wrapped all btree updates when running delayed items with
memalloc_nofs_save() and memalloc_nofs_restore(), due to a lock inversion
detected by lockdep involving reclaim and the mutex of delayed nodes.
The problem is because the ref verify tool does some memory allocations
with GFP_KERNEL, which can trigger reclaim and reclaim can trigger inode
eviction, which requires locking the mutex of an inode's delayed node.
On the other hand the ref verify tool is called when allocating metadata
extents as part of operations that modify a btree, which is a problem when
running delayed nodes, where we do btree updates while holding the mutex
of a delayed node. This is what caused the lockdep warning.
Instead of wrapping every btree update when running delayed nodes, change
the ref verify tool to never do GFP_KERNEL allocations, because:
1) We get less repeated code, which at the moment does not even have a
comment mentioning why we need to setup the NOFS context, which is a
recommended good practice as mentioned at
Documentation/core-api/gfp_mask-from-fs-io.rst
2) The ref verify tool is something meant only for debugging and not
something that should be enabled on non-debug / non-development
kernels;
3) We may have yet more places outside delayed-inode.c where we have
similar problem: doing btree updates while holding some lock and
then having the GFP_KERNEL memory allocations, from the ref verify
tool, trigger reclaim and trying again to acquire the same lock
through the reclaim path.
Or we could get more such cases in the future, therefore this change
prevents getting into similar cases when using the ref verify tool.
Curiously most of the memory allocations done by the ref verify tool
were already using GFP_NOFS, except a few ones for no apparent reason.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When we insert the delayed items of an inode, which corresponds to the
directory index keys for a directory (key type BTRFS_DIR_INDEX_KEY), we
do the following:
1) Pick the first delayed item from the rbtree and insert it into the
fs/subvolume btree, using btrfs_insert_empty_item() for that;
2) Without releasing the path returned by btrfs_insert_empty_item(),
keep collecting as many consecutive delayed items from the rbtree
as possible, as long as each one's BTRFS_DIR_INDEX_KEY key is the
immediate successor of the previously picked item and as long as
they fit in the available space of the leaf the path points to;
3) Then insert all the collected items into the leaf;
4) Release the reserve metadata space for each collected item and
release each item (implies deleting from the rbtree);
5) Unlock the path.
While this is much better than inserting items one by one, it can be
improved in a few aspects:
1) Instead of adding items based on the remaining free space of the
leaf, collect as many items that can fit in a leaf and bulk insert
them. This results in less and larger batches, reducing the total
amount of time to insert the delayed items. For example when adding
100K files to a directory, we ended up creating 1658 batches with
very variable sizes ranging from 1 item to 118 items, on a filesystem
with a node/leaf size of 16K. After this change, we end up with 839
batches, with the vast majority of them having exactly 120 items;
2) We do the search for more items to batch, by iterating the rbtree,
while holding a write lock on the leaf;
3) While still holding the leaf locked, we are releasing the reserved
metadata for each item and then deleting each item, keeping a write
lock on the leaf for longer than necessary. Releasing the delayed items
one by one can take a significant amount of time, because deleting
them from the rbtree can often be a bit slow when the deletion results
in rebalancing the rbtree.
So change this so that we try to create larger batches, with a total
item size up to the maximum a leaf can support, and by unlocking the leaf
immediately after inserting the items, releasing the reserved metadata
space of each item and releasing each item without holding the write lock
on the leaf.
The following script that runs fs_mark was used to test this change:
$ cat test.sh
#!/bin/bash
DEV=/dev/nvme0n1
MNT=/mnt/nvme0n1
MOUNT_OPTIONS="-o ssd"
MKFS_OPTIONS="-m single -d single"
FILES=1000000
THREADS=16
FILE_SIZE=0
echo "performance" | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
umount $DEV &> /dev/null
mkfs.btrfs -f $MKFS_OPTIONS $DEV
mount $MOUNT_OPTIONS $DEV $MNT
OPTS="-S 0 -L 5 -n $FILES -s $FILE_SIZE -t 16"
for ((i = 1; i <= $THREADS; i++)); do
OPTS="$OPTS -d $MNT/d$i"
done
fs_mark $OPTS
umount $MNT
It was run on machine with 12 cores, 64G of ram, using a NVMe device and
using a non-debug kernel config (Debian's default config).
Results before this change:
FSUse% Count Size Files/sec App Overhead
1 16000000 0 76182.1 72223046
3 32000000 0 62746.9 80776528
5 48000000 0 77029.0 93022381
6 64000000 0 73691.6 95251075
8 80000000 0 66288.0 85089634
Results after this change:
FSUse% Count Size Files/sec App Overhead
1 16000000 0 79049.5 (+3.7%) 69700824
3 32000000 0 65248.9 (+3.9%) 80583693
5 48000000 0 77991.4 (+1.2%) 90040908
6 64000000 0 75096.8 (+1.9%) 89862241
8 80000000 0 66926.8 (+1.0%) 84429169
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
clang warns:
fs/btrfs/delayed-inode.c:684:6: warning: variable 'total_data_size' set
but not used [-Wunused-but-set-variable]
int total_data_size = 0, total_size = 0;
^
1 warning generated.
This variable's value has been unused since commit fc0d82e103 ("btrfs:
sink total_data parameter in setup_items_for_insert"). Eliminate it.
Link: https://github.com/ClangBuiltLinux/linux/issues/1391
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
If we fail to update the delayed inode we need to abort the transaction,
because we could leave an inode with the improper counts or some other
such corruption behind.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
If we get an error while looking up the inode item we'll simply bail
without cleaning up the delayed node. This results in this style of
warning happening on commit:
WARNING: CPU: 0 PID: 76403 at fs/btrfs/delayed-inode.c:1365 btrfs_assert_delayed_root_empty+0x5b/0x90
CPU: 0 PID: 76403 Comm: fsstress Tainted: G W 5.13.0-rc1+ #373
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.13.0-2.fc32 04/01/2014
RIP: 0010:btrfs_assert_delayed_root_empty+0x5b/0x90
RSP: 0018:ffffb8bb815a7e50 EFLAGS: 00010286
RAX: 0000000000000000 RBX: ffff95d6d07e1888 RCX: ffff95d6c0fa3000
RDX: 0000000000000002 RSI: 000000000029e91c RDI: ffff95d6c0fc8060
RBP: ffff95d6c0fc8060 R08: 00008d6d701a2c1d R09: 0000000000000000
R10: ffff95d6d1760ea0 R11: 0000000000000001 R12: ffff95d6c15a4d00
R13: ffff95d6c0fa3000 R14: 0000000000000000 R15: ffffb8bb815a7e90
FS: 00007f490e8dbb80(0000) GS:ffff95d73bc00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007f6e75555cb0 CR3: 00000001101ce001 CR4: 0000000000370ef0
Call Trace:
btrfs_commit_transaction+0x43c/0xb00
? finish_wait+0x80/0x80
? vfs_fsync_range+0x90/0x90
iterate_supers+0x8c/0x100
ksys_sync+0x50/0x90
__do_sys_sync+0xa/0x10
do_syscall_64+0x3d/0x80
entry_SYSCALL_64_after_hwframe+0x44/0xae
Because the iref isn't dropped and this leaves an elevated node->count,
so any release just re-queues it onto the delayed inodes list. Fix this
by going to the out label to handle the proper cleanup of the delayed
node.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Right now we only cleanup the delayed iref if we have
BTRFS_DELAYED_NODE_DEL_IREF set on the node. However we have some error
conditions that need to cleanup the iref if it still exists, so to make
this code cleaner move the test_bit into btrfs_release_delayed_iref
itself and unconditionally call it in each of the cases instead.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
A few places we intermix btrfs_inode_lock with a inode_unlock, and some
places we just use inode_lock/inode_unlock instead of btrfs_inode_lock.
None of these places are using this incorrectly, but as we adjust some
of these callers it would be nice to keep everything consistent, so
convert everybody to use btrfs_inode_lock/btrfs_inode_unlock.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_block_rsv_add can return only ENOSPC since it's called with
NO_FLUSH modifier. This so simplify the logic in
btrfs_delayed_inode_reserve_metadata to exploit this invariant.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add assert and comment ]
Signed-off-by: David Sterba <dsterba@suse.com>
It's only used for tracepoint to obtain the inode number, but we already
have the ino from btrfs_delayed_node::inode_id.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Calling btrfs_qgroup_reserve_meta_prealloc from
btrfs_delayed_inode_reserve_metadata can result in flushing delalloc
while holding a transaction and delayed node locks. This is deadlock
prone. In the past multiple commits:
* ae5e070eac ("btrfs: qgroup: don't try to wait flushing if we're
already holding a transaction")
* 6f23277a49 ("btrfs: qgroup: don't commit transaction when we already
hold the handle")
Tried to solve various aspects of this but this was always a
whack-a-mole game. Unfortunately those 2 fixes don't solve a deadlock
scenario involving btrfs_delayed_node::mutex. Namely, one thread
can call btrfs_dirty_inode as a result of reading a file and modifying
its atime:
PID: 6963 TASK: ffff8c7f3f94c000 CPU: 2 COMMAND: "test"
#0 __schedule at ffffffffa529e07d
#1 schedule at ffffffffa529e4ff
#2 schedule_timeout at ffffffffa52a1bdd
#3 wait_for_completion at ffffffffa529eeea <-- sleeps with delayed node mutex held
#4 start_delalloc_inodes at ffffffffc0380db5
#5 btrfs_start_delalloc_snapshot at ffffffffc0393836
#6 try_flush_qgroup at ffffffffc03f04b2
#7 __btrfs_qgroup_reserve_meta at ffffffffc03f5bb6 <-- tries to reserve space and starts delalloc inodes.
#8 btrfs_delayed_update_inode at ffffffffc03e31aa <-- acquires delayed node mutex
#9 btrfs_update_inode at ffffffffc0385ba8
#10 btrfs_dirty_inode at ffffffffc038627b <-- TRANSACTIION OPENED
#11 touch_atime at ffffffffa4cf0000
#12 generic_file_read_iter at ffffffffa4c1f123
#13 new_sync_read at ffffffffa4ccdc8a
#14 vfs_read at ffffffffa4cd0849
#15 ksys_read at ffffffffa4cd0bd1
#16 do_syscall_64 at ffffffffa4a052eb
#17 entry_SYSCALL_64_after_hwframe at ffffffffa540008c
This will cause an asynchronous work to flush the delalloc inodes to
happen which can try to acquire the same delayed_node mutex:
PID: 455 TASK: ffff8c8085fa4000 CPU: 5 COMMAND: "kworker/u16:30"
#0 __schedule at ffffffffa529e07d
#1 schedule at ffffffffa529e4ff
#2 schedule_preempt_disabled at ffffffffa529e80a
#3 __mutex_lock at ffffffffa529fdcb <-- goes to sleep, never wakes up.
#4 btrfs_delayed_update_inode at ffffffffc03e3143 <-- tries to acquire the mutex
#5 btrfs_update_inode at ffffffffc0385ba8 <-- this is the same inode that pid 6963 is holding
#6 cow_file_range_inline.constprop.78 at ffffffffc0386be7
#7 cow_file_range at ffffffffc03879c1
#8 btrfs_run_delalloc_range at ffffffffc038894c
#9 writepage_delalloc at ffffffffc03a3c8f
#10 __extent_writepage at ffffffffc03a4c01
#11 extent_write_cache_pages at ffffffffc03a500b
#12 extent_writepages at ffffffffc03a6de2
#13 do_writepages at ffffffffa4c277eb
#14 __filemap_fdatawrite_range at ffffffffa4c1e5bb
#15 btrfs_run_delalloc_work at ffffffffc0380987 <-- starts running delayed nodes
#16 normal_work_helper at ffffffffc03b706c
#17 process_one_work at ffffffffa4aba4e4
#18 worker_thread at ffffffffa4aba6fd
#19 kthread at ffffffffa4ac0a3d
#20 ret_from_fork at ffffffffa54001ff
To fully address those cases the complete fix is to never issue any
flushing while holding the transaction or the delayed node lock. This
patch achieves it by calling qgroup_reserve_meta directly which will
either succeed without flushing or will fail and return -EDQUOT. In the
latter case that return value is going to be propagated to
btrfs_dirty_inode which will fallback to start a new transaction. That's
fine as the majority of time we expect the inode will have
BTRFS_DELAYED_NODE_INODE_DIRTY flag set which will result in directly
copying the in-memory state.
Fixes: c53e965360 ("btrfs: qgroup: try to flush qgroup space when we get -EDQUOT")
CC: stable@vger.kernel.org # 5.10+
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Following commit f218ea6c47 ("btrfs: delayed-inode: Remove wrong
qgroup meta reservation calls") this function now reserves num_bytes,
rather than the fixed amount of nodesize. As such this requires the
same amount to be freed in case of failure. Fix this by adjusting
the amount we are freeing.
Fixes: f218ea6c47 ("btrfs: delayed-inode: Remove wrong qgroup meta reservation calls")
CC: stable@vger.kernel.org # 4.19+
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Fix the following coccicheck warnings:
./fs/btrfs/delayed-inode.c:1157:39-41: WARNING !A || A && B is
equivalent to !A || B.
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Suggested-by: Jiapeng Zhong <oswb@linux.alibaba.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Abaci Team <abaci-bugfix@linux.alibaba.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We no longer distinguish between blocking and spinning, so rip out all
this code.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Now that we're using a rw_semaphore we no longer need to indicate if a
lock is blocking or not, nor do we need to flip the entire path from
blocking to spinning. Remove these helpers and all the places they are
called.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
That parameter can easily be derived based on the "data_size" and "nr"
parameters exploit this fact to simply the function's signature. No
functional changes.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The value of this argument can be derived from the total_data as it's
simply the value of the data size + size of btrfs_items being touched.
Move the parameter calculation inside the function. This results in a
simpler interface and also a minor size reduction:
./scripts/bloat-o-meter ctree.original fs/btrfs/ctree.o
add/remove: 0/0 grow/shrink: 0/3 up/down: 0/-34 (-34)
Function old new delta
btrfs_duplicate_item 260 259 -1
setup_items_for_insert 1200 1190 -10
btrfs_insert_empty_items 177 154 -23
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
For delayed inode facility, qgroup metadata is reserved for it, and
later freed.
However we're freeing more bytes than we reserved.
In btrfs_delayed_inode_reserve_metadata():
num_bytes = btrfs_calc_metadata_size(fs_info, 1);
...
ret = btrfs_qgroup_reserve_meta_prealloc(root,
fs_info->nodesize, true);
...
if (!ret) {
node->bytes_reserved = num_bytes;
But in btrfs_delayed_inode_release_metadata():
if (qgroup_free)
btrfs_qgroup_free_meta_prealloc(node->root,
node->bytes_reserved);
else
btrfs_qgroup_convert_reserved_meta(node->root,
node->bytes_reserved);
This means, we're always releasing more qgroup metadata rsv than we have
reserved.
This won't trigger selftest warning, as btrfs qgroup metadata rsv has
extra protection against cases like quota enabled half-way.
But we still need to fix this problem any way.
This patch will use the same num_bytes for qgroup metadata rsv so we
could handle it correctly.
Fixes: f218ea6c47 ("btrfs: delayed-inode: Remove wrong qgroup meta reservation calls")
CC: stable@vger.kernel.org # 4.19+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Currently the non-prefixed version is a simple wrapper used to hide
the 4th argument of the prefixed version. This doesn't bring much value
in practice and only makes the code harder to follow by adding another
level of indirection. Rectify this by removing the __ prefix and
have only one public function to release bytes from a block reservation.
No semantic changes.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The status of aborted transaction can change between calls and it needs
to be accessed by READ_ONCE. Add a helper that also wraps the unlikely
hint.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We want to use this everywhere we modify the file extent items
permanently. These include:
1) Inserting new file extents for writes and prealloc extents.
2) Truncating inode items.
3) btrfs_cont_expand().
4) Insert inline extents.
5) Insert new extents from log replay.
6) Insert a new extent for clone, as it could be past i_size.
7) Hole punching
For hole punching in particular it might seem it's not necessary because
anybody extending would use btrfs_cont_expand, however there is a corner
that still can give us trouble. Start with an empty file and
fallocate KEEP_SIZE 1M-2M
We now have a 0 length file, and a hole file extent from 0-1M, and a
prealloc extent from 1M-2M. Now
punch 1M-1.5M
Because this is past i_size we have
[HOLE EXTENT][ NOTHING ][PREALLOC]
[0 1M][1M 1.5M][1.5M 2M]
with an i_size of 0. Now if we pwrite 0-1.5M we'll increas our i_size
to 1.5M, but our disk_i_size is still 0 until the ordered extent
completes.
However if we now immediately truncate 2M on the file we'll just call
btrfs_cont_expand(inode, 1.5M, 2M), since our old i_size is 1.5M. If we
commit the transaction here and crash we'll expose the gap.
To fix this we need to clear the file extent mapping for the range that
we punched but didn't insert a corresponding file extent for. This will
mean the truncate will only get an disk_i_size set to 1M if we crash
before the finish ordered io happens.
I've written an xfstest to reproduce the problem and validate this fix.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We hit the following warning while running down a different problem
[ 6197.175850] ------------[ cut here ]------------
[ 6197.185082] refcount_t: underflow; use-after-free.
[ 6197.194704] WARNING: CPU: 47 PID: 966 at lib/refcount.c:190 refcount_sub_and_test_checked+0x53/0x60
[ 6197.521792] Call Trace:
[ 6197.526687] __btrfs_release_delayed_node+0x76/0x1c0
[ 6197.536615] btrfs_kill_all_delayed_nodes+0xec/0x130
[ 6197.546532] ? __btrfs_btree_balance_dirty+0x60/0x60
[ 6197.556482] btrfs_clean_one_deleted_snapshot+0x71/0xd0
[ 6197.566910] cleaner_kthread+0xfa/0x120
[ 6197.574573] kthread+0x111/0x130
[ 6197.581022] ? kthread_create_on_node+0x60/0x60
[ 6197.590086] ret_from_fork+0x1f/0x30
[ 6197.597228] ---[ end trace 424bb7ae00509f56 ]---
This is because the free side drops the ref without the lock, and then
takes the lock if our refcount is 0. So you can have nodes on the tree
that have a refcount of 0. Fix this by zero'ing out that element in our
temporary array so we don't try to kill it again.
CC: stable@vger.kernel.org # 4.14+
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add comment ]
Signed-off-by: David Sterba <dsterba@suse.com>
The function belongs to the family of locking functions, so move it
there. The 'noinline' keyword is dropped as it's now an exported
function that does not need it.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Commit 9e0af23764 ("Btrfs: fix task hang under heavy compressed
write") worked around the issue that a recycled work item could get a
false dependency on the original work item due to how the workqueue code
guarantees non-reentrancy. It did so by giving different work functions
to different types of work.
However, the fixes in the previous few patches are more complete, as
they prevent a work item from being recycled at all (except for a tiny
window that the kernel workqueue code handles for us). This obsoletes
the previous fix, so we don't need the unique helpers for correctness.
The only other reason to keep them would be so they show up in stack
traces, but they always seem to be optimized to a tail call, so they
don't show up anyways. So, let's just get rid of the extra indirection.
While we're here, rename normal_work_helper() to the more informative
btrfs_work_helper().
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The file ctree.h serves as a header for everything and has become quite
bloated. Split some helpers that are generic and create a new file that
should be the catch-all for code that's not btrfs-specific.
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: David Sterba <dsterba@suse.com>
Historically we reserved worst case for every btree operation, and
generally speaking we want to do that in cases where it could be the
worst case. However for updating inodes we know the inode items are
already in the tree, so it will only be an update operation and never an
insert operation. This allows us to always reserve only the
metadata_size amount for inode updates rather than the
insert_metadata_size amount.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_calc_trunc_metadata_size differs from trans_metadata_size in that
it doesn't take into account any splitting at the levels, because
truncate will never split nodes. However truncate _and_ changing will
never split nodes, so rename btrfs_calc_trunc_metadata_size to
btrfs_calc_metadata_size. Also btrfs_calc_trans_metadata_size is purely
for inserting items, so rename this to btrfs_calc_insert_metadata_size.
Making these clearer will help when I start using them differently in
upcoming patches.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
There is one report of fuzzed image which leads to BUG_ON() in
btrfs_delete_delayed_dir_index().
Although that fuzzed image can already be addressed by enhanced
extent-tree error handler, it's still better to hunt down more BUG_ON().
This patch will hunt down two BUG_ON()s in
btrfs_delete_delayed_dir_index():
- One for error from btrfs_delayed_item_reserve_metadata()
Instead of BUG_ON(), we output an error message and free the item.
And return the error.
All callers of this function handles the error by aborting current
trasaction.
- One for possible EEXIST from __btrfs_add_delayed_deletion_item()
That function can return -EEXIST.
We already have a good enough error message for that, only need to
clean up the reserved metadata space and allocated item.
To help above cleanup, also modifiy __btrfs_remove_delayed_item() called
in btrfs_release_delayed_item(), to skip unassociated item.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=203253
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Deduplicate the btrfs file type conversion implementation - file systems
that use the same file types as defined by POSIX do not need to define
their own versions and can use the common helper functions decared in
fs_types.h and implemented in fs_types.c
Common implementation can be found via commit:
bbe7449e25 "fs: common implementation of file type"
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Phillip Potter <phil@philpotter.co.uk>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Btrfs's btree locking has two modes, spinning mode and blocking mode,
while searching btree, locking is always acquired in spinning mode and
then converted to blocking mode if necessary, and in some hot paths we may
switch the locking back to spinning mode by btrfs_clear_path_blocking().
When acquiring locks, both of reader and writer need to wait for blocking
readers and writers to complete before doing read_lock()/write_lock().
The problem is that btrfs_clear_path_blocking() needs to switch nodes
in the path to blocking mode at first (by btrfs_set_path_blocking) to
make lockdep happy before doing its actual clearing blocking job.
When switching to blocking mode from spinning mode, it consists of
step 1) bumping up blocking readers counter and
step 2) read_unlock()/write_unlock(),
this has caused serious ping-pong effect if there're a great amount of
concurrent readers/writers, as waiters will be woken up and go to
sleep immediately.
1) Killing this kind of ping-pong results in a big improvement in my 1600k
files creation script,
MNT=/mnt/btrfs
mkfs.btrfs -f /dev/sdf
mount /dev/def $MNT
time fsmark -D 10000 -S0 -n 100000 -s 0 -L 1 -l /tmp/fs_log.txt \
-d $MNT/0 -d $MNT/1 \
-d $MNT/2 -d $MNT/3 \
-d $MNT/4 -d $MNT/5 \
-d $MNT/6 -d $MNT/7 \
-d $MNT/8 -d $MNT/9 \
-d $MNT/10 -d $MNT/11 \
-d $MNT/12 -d $MNT/13 \
-d $MNT/14 -d $MNT/15
w/o patch:
real 2m27.307s
user 0m12.839s
sys 13m42.831s
w/ patch:
real 1m2.273s
user 0m15.802s
sys 8m16.495s
1.1) latency histogram from funclatency[1]
Overall with the patch, there're ~50% less write lock acquisition and
the 95% max latency that write lock takes also reduces to ~100ms from
>500ms.
--------------------------------------------
w/o patch:
--------------------------------------------
Function = btrfs_tree_lock
msecs : count distribution
0 -> 1 : 2385222 |****************************************|
2 -> 3 : 37147 | |
4 -> 7 : 20452 | |
8 -> 15 : 13131 | |
16 -> 31 : 3877 | |
32 -> 63 : 3900 | |
64 -> 127 : 2612 | |
128 -> 255 : 974 | |
256 -> 511 : 165 | |
512 -> 1023 : 13 | |
Function = btrfs_tree_read_lock
msecs : count distribution
0 -> 1 : 6743860 |****************************************|
2 -> 3 : 2146 | |
4 -> 7 : 190 | |
8 -> 15 : 38 | |
16 -> 31 : 4 | |
--------------------------------------------
w/ patch:
--------------------------------------------
Function = btrfs_tree_lock
msecs : count distribution
0 -> 1 : 1318454 |****************************************|
2 -> 3 : 6800 | |
4 -> 7 : 3664 | |
8 -> 15 : 2145 | |
16 -> 31 : 809 | |
32 -> 63 : 219 | |
64 -> 127 : 10 | |
Function = btrfs_tree_read_lock
msecs : count distribution
0 -> 1 : 6854317 |****************************************|
2 -> 3 : 2383 | |
4 -> 7 : 601 | |
8 -> 15 : 92 | |
2) dbench also proves the improvement,
dbench -t 120 -D /mnt/btrfs 16
w/o patch:
Throughput 158.363 MB/sec
w/ patch:
Throughput 449.52 MB/sec
3) xfstests didn't show any additional failures.
One thing to note is that callers may set path->leave_spinning to have
all nodes in the path stay in spinning mode, which means callers are
ready to not sleep before releasing the path, but it won't cause
problems if they don't want to sleep in blocking mode.
[1]: https://github.com/iovisor/bcc/blob/master/tools/funclatency.py
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: David Sterba <dsterba@suse.com>
rb_first_cached() trades an extra pointer "leftmost" for doing the same job as
rb_first() but in O(1).
Functions manipulating delayed_item need to get the first entry, this converts
it to use rb_first_cached().
For more details about the optimization see patch "Btrfs: delayed-refs:
use rb_first_cached for href_root".
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
There are two members in struct btrfs_root which indicate root's
objectid: objectid and root_key.objectid.
They are both set to the same value in __setup_root():
static void __setup_root(struct btrfs_root *root,
struct btrfs_fs_info *fs_info,
u64 objectid)
{
...
root->objectid = objectid;
...
root->root_key.objectid = objecitd;
...
}
and not changed to other value after initialization.
grep in btrfs directory shows both are used in many places:
$ grep -rI "root->root_key.objectid" | wc -l
133
$ grep -rI "root->objectid" | wc -l
55
(4.17, inc. some noise)
It is confusing to have two similar variable names and it seems
that there is no rule about which should be used in a certain case.
Since ->root_key itself is needed for tree reloc tree, let's remove
'objecitd' member and unify code to use ->root_key.objectid in all places.
Signed-off-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Using true and false here is closer to the expected semantic than using
0 and 1. No functional change.
Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
It can be referenced from the passed transaction handle.
Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
It can be referenced from the passed transaction handle.
Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Functions that get btrfs inode can simply reach the fs_info by
dereferencing the root and this looks a bit more straightforward
compared to the btrfs_sb(...) indirection.
If the transaction handle is available and not NULL it's used instead.
Signed-off-by: David Sterba <dsterba@suse.com>
Use the wrappers and reduce the amount of low-level details about the
waitqueue management.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Commit 4f5427ccce ("btrfs: delayed-inode: Use new qgroup meta rsv for
delayed inode and item") merged into mainline was not latest version
submitted to the mail list in Dec 2017.
Which lacks the following fixes:
1) Remove btrfs_qgroup_convert_reserved_meta() call in
btrfs_delayed_item_release_metadata()
2) Remove btrfs_qgroup_reserve_meta_prealloc() call in
btrfs_delayed_inode_reserve_metadata()
Those fixes will resolve unexpected EDQUOT problems.
Fixes: 4f5427ccce ("btrfs: delayed-inode: Use new qgroup meta rsv for delayed inode and item")
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Remove GPL boilerplate text (long, short, one-line) and keep the rest,
ie. personal, company or original source copyright statements. Add the
SPDX header.
Signed-off-by: David Sterba <dsterba@suse.com>
Quite similar for delalloc, some modification to delayed-inode and
delayed-item reservation. Also needs extra parameter for release case
to distinguish normal release and error release.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The __cold functions are placed to a special section, as they're
expected to be called rarely. This could help i-cache prefetches or help
compiler to decide which branches are more/less likely to be taken
without any other annotations needed.
Though we can't add more __exit annotations, it's still possible to add
__cold (that's also added with __exit). That way the following function
categories are tagged:
- printf wrappers, error messages
- exit helpers
Signed-off-by: David Sterba <dsterba@suse.com>
We already pass the transaction which has a reference to the fs_info,
so use that. No functional changes.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We already pass the transaction handle, which contains a refrence to
the fs_info so grab it from there. No functional changes.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
-----BEGIN PGP SIGNATURE-----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=Irul
-----END PGP SIGNATURE-----
Merge tag 'for-4.16-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs updates from David Sterba:
"Features or user visible changes:
- fallocate: implement zero range mode
- avoid losing data raid profile when deleting a device
- tree item checker: more checks for directory items and xattrs
Notable fixes:
- raid56 recovery: don't use cached stripes, that could be
potentially changed and a later RMW or recovery would lead to
corruptions or failures
- let raid56 try harder to rebuild damaged data, reading from all
stripes if necessary
- fix scrub to repair raid56 in a similar way as in the case above
Other:
- cleanups: device freeing, removed some call indirections, redundant
bio_put/_get, unused parameters, refactorings and renames
- RCU list traversal fixups
- simplify mount callchain, remove recursing back when mounting a
subvolume
- plug for fsync, may improve bio merging on multiple devices
- compression heurisic: replace heap sort with radix sort, gains some
performance
- add extent map selftests, buffered write vs dio"
* tag 'for-4.16-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux: (155 commits)
btrfs: drop devid as device_list_add() arg
btrfs: get device pointer from device_list_add()
btrfs: set the total_devices in device_list_add()
btrfs: move pr_info into device_list_add
btrfs: make btrfs_free_stale_devices() to match the path
btrfs: rename btrfs_free_stale_devices() arg to skip_dev
btrfs: make btrfs_free_stale_devices() argument optional
btrfs: make btrfs_free_stale_device() to iterate all stales
btrfs: no need to check for btrfs_fs_devices::seeding
btrfs: Use IS_ALIGNED in btrfs_truncate_block instead of opencoding it
Btrfs: noinline merge_extent_mapping
Btrfs: add WARN_ONCE to detect unexpected error from merge_extent_mapping
Btrfs: extent map selftest: dio write vs dio read
Btrfs: extent map selftest: buffered write vs dio read
Btrfs: add extent map selftests
Btrfs: move extent map specific code to extent_map.c
Btrfs: add helper for em merge logic
Btrfs: fix unexpected EEXIST from btrfs_get_extent
Btrfs: fix incorrect block_len in merge_extent_mapping
btrfs: Remove unused readahead spinlock
...
-----BEGIN PGP SIGNATURE-----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=iihN
-----END PGP SIGNATURE-----
Merge tag 'iversion-v4.16-1' of git://git.kernel.org/pub/scm/linux/kernel/git/jlayton/linux
Pull inode->i_version rework from Jeff Layton:
"This pile of patches is a rework of the inode->i_version field. We
have traditionally incremented that field on every inode data or
metadata change. Typically this increment needs to be logged on disk
even when nothing else has changed, which is rather expensive.
It turns out though that none of the consumers of that field actually
require this behavior. The only real requirement for all of them is
that it be different iff the inode has changed since the last time the
field was checked.
Given that, we can optimize away most of the i_version increments and
avoid dirtying inode metadata when the only change is to the i_version
and no one is querying it. Queries of the i_version field are rather
rare, so we can help write performance under many common workloads.
This patch series converts existing accesses of the i_version field to
a new API, and then converts all of the in-kernel filesystems to use
it. The last patch in the series then converts the backend
implementation to a scheme that optimizes away a large portion of the
metadata updates when no one is looking at it.
In my own testing this series significantly helps performance with
small I/O sizes. I also got this email for Christmas this year from
the kernel test robot (a 244% r/w bandwidth improvement with XFS over
DAX, with 4k writes):
https://lkml.org/lkml/2017/12/25/8
A few of the earlier patches in this pile are also flowing to you via
other trees (mm, integrity, and nfsd trees in particular)".
* tag 'iversion-v4.16-1' of git://git.kernel.org/pub/scm/linux/kernel/git/jlayton/linux: (22 commits)
fs: handle inode->i_version more efficiently
btrfs: only dirty the inode in btrfs_update_time if something was changed
xfs: avoid setting XFS_ILOG_CORE if i_version doesn't need incrementing
fs: only set S_VERSION when updating times if necessary
IMA: switch IMA over to new i_version API
xfs: convert to new i_version API
ufs: use new i_version API
ocfs2: convert to new i_version API
nfsd: convert to new i_version API
nfs: convert to new i_version API
ext4: convert to new i_version API
ext2: convert to new i_version API
exofs: switch to new i_version API
btrfs: convert to new i_version API
afs: convert to new i_version API
affs: convert to new i_version API
fat: convert to new i_version API
fs: don't take the i_lock in inode_inc_iversion
fs: new API for handling inode->i_version
ntfs: remove i_version handling
...
In fixing the readdir+pagefault deadlock I accidentally introduced a
stale entry regression in readdir. If we get close to full for the
temporary buffer, and then skip a few delayed deletions, and then try to
add another entry that won't fit, we will emit the entries we found and
retry. Unfortunately we delete entries from our del_list as we find
them, assuming we won't need them. However our pos will be with
whatever our last entry was, which could be before the delayed deletions
we skipped, so the next search will add the deleted entries back into
our readdir buffer. So instead don't delete entries we find in our
del_list so we can make sure we always find our delayed deletions. This
is a slight perf hit for readdir with lots of pending deletions, but
hopefully this isn't a common occurrence. If it is we can revist this
and optimize it.
cc: stable@vger.kernel.org
Fixes: 23b5ec7494 ("btrfs: fix readdir deadlock with pagefault")
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
