We should always call check_delayed_ref() with a path having a locked leaf
from the extent tree where either the extent item is located or where it
should be located in case it doesn't exist yet (when there's a pending
unflushed delayed ref to do it), as we need to lock any existing delayed
ref head while holding such leaf locked in order to avoid races with
flushing delayed references, which could make us think an extent is not
shared when it really is.
So add some assertions and a comment about such expectations to
btrfs_cross_ref_exist(), which is the only caller of check_delayed_ref().
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
There are some not immediately obvious details about the operation of
check_committed_ref(), namely that when it returns 0 it must return with
the path having a locked leaf from the extent tree that contains the
extent's extent item, so that we can later check for delayed refs when
calling check_delayed_ref() in a way that doesn't race with a task running
delayed references. For similar reasons, it must also return with a locked
leaf when the extent item is not found, and that leaf is where the extent
item should be located, because we may have delayed references that are
going to create the extent item. Also document that the function can
return false positives in order to not be too slow, and that the most
important is to not return false negatives.
So add a function comment to check_committed_ref().
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Instead of passing a root and an objectid which matches an inode number,
pass the inode instead, since the root is always the root associated to
the inode and the objectid is the number of that inode.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Instead of setting the value to return in a local variable 'ret' and then
jumping into a label named 'out' that does nothing but return that value,
simplify everything by getting rid of the label and directly returning a
value.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
At check_committed_ref() we are calling btrfs_get_extent_inline_ref_type()
twice, once before we check if have an inline extent owner ref (for simple
qgroups) and then once again sometime after that check. This second call
is redundant when we have simple quotas disabled or we found an inline ref
that is not of the owner ref type. So avoid this second call unless we
have simple quotas enabled and found an owner ref, saving a function call
that does inline ref validation again.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
At check_committed_ref() we have this check to see if the data extent was
created in a generation lower than or equals to the generation where the
last snapshot for the root was created, and if so we return immediately
with 1, since it's very likely the extent is shared, referenced by other
root.
The only call chain for check_committed_ref() is the following:
can_nocow_file_extent()
btrfs_cross_ref_exist()
check_committed_ref()
And we already do that snapshot check at can_nocow_file_extent(), before
we call btrfs_cross_ref_exist(). This makes the check done at
check_committed_ref() redundant, so remove it.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
All callers of can_nocow_extent() now pass a value of false for its
'strict' argument, making it redundant. So remove the argument from
can_nocow_extent() as well as can_nocow_file_extent(),
btrfs_cross_ref_exist() and check_committed_ref(), because this
argument was used just to influence the behavior of check_committed_ref().
Also remove the 'strict' field from struct can_nocow_file_extent_args,
which is now always false as well, as its value is taken from the
argument to can_nocow_extent().
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
During swap activation we iterate over the extents of a file and we can
have many thounsands of them, so we can end up in a busy loop monopolizing
a core. Avoid this by doing a voluntary reschedule after processing each
extent.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
During swap activation we iterate over the extents of a file, then do
several checks for each extent, some of which may take some significant
time such as checking if an extent is shared. Since a file can have
many thousands of extents, this can be a very slow operation and it's
currently not interruptible. I had a bug during development of a previous
patch that resulted in an infinite loop when iterating the extents, so
a core was busy looping and I couldn't cancel the operation, which is very
annoying and requires a reboot. So make the loop interruptible by checking
for fatal signals at the end of each iteration and stopping immediately if
there is one.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When activating a swap file, to determine if an extent is shared we use
can_nocow_extent(), which ends up at btrfs_cross_ref_exist(). That helper
is meant to be quick because it's used in the NOCOW write path, when
flushing delalloc and when doing a direct IO write, however it does return
some false positives, meaning it may indicate that an extent is shared
even if it's no longer the case. For the write path this is fine, we just
do a unnecessary COW operation instead of doing a more rigorous check
which would be too heavy (calling btrfs_is_data_extent_shared()).
However when activating a swap file, the false positives simply result
in a failure, which is confusing for users/applications. One particular
case where this happens is when a data extent only has 1 reference but
that reference is not inlined in the extent item located in the extent
tree - this happens when we create more than 33 references for an extent
and then delete those 33 references plus every other non-inline reference
except one. The function check_committed_ref() assumes that if the size
of an extent item doesn't match the size of struct btrfs_extent_item
plus the size of an inline reference (plus an owner reference in case
simple quotas are enabled), then the extent is shared - that is not the
case however, we can have a single reference but it's not inlined - the
reason we do this is to be fast and avoid inspecting non-inline references
which may be located in another leaf of the extent tree, slowing down
write paths.
The following test script reproduces the bug:
$ cat test.sh
#!/bin/bash
DEV=/dev/sdi
MNT=/mnt/sdi
NUM_CLONES=50
umount $DEV &> /dev/null
run_test()
{
local sync_after_add_reflinks=$1
local sync_after_remove_reflinks=$2
mkfs.btrfs -f $DEV > /dev/null
#mkfs.xfs -f $DEV > /dev/null
mount $DEV $MNT
touch $MNT/foo
chmod 0600 $MNT/foo
# On btrfs the file must be NOCOW.
chattr +C $MNT/foo &> /dev/null
xfs_io -s -c "pwrite -b 1M 0 1M" $MNT/foo
mkswap $MNT/foo
for ((i = 1; i <= $NUM_CLONES; i++)); do
touch $MNT/foo_clone_$i
chmod 0600 $MNT/foo_clone_$i
# On btrfs the file must be NOCOW.
chattr +C $MNT/foo_clone_$i &> /dev/null
cp --reflink=always $MNT/foo $MNT/foo_clone_$i
done
if [ $sync_after_add_reflinks -ne 0 ]; then
# Flush delayed refs and commit current transaction.
sync -f $MNT
fi
# Remove the original file and all clones except the last.
rm -f $MNT/foo
for ((i = 1; i < $NUM_CLONES; i++)); do
rm -f $MNT/foo_clone_$i
done
if [ $sync_after_remove_reflinks -ne 0 ]; then
# Flush delayed refs and commit current transaction.
sync -f $MNT
fi
# Now use the last clone as a swap file. It should work since
# its extent are not shared anymore.
swapon $MNT/foo_clone_${NUM_CLONES}
swapoff $MNT/foo_clone_${NUM_CLONES}
umount $MNT
}
echo -e "\nTest without sync after creating and removing clones"
run_test 0 0
echo -e "\nTest with sync after creating clones"
run_test 1 0
echo -e "\nTest with sync after removing clones"
run_test 0 1
echo -e "\nTest with sync after creating and removing clones"
run_test 1 1
Running the test:
$ ./test.sh
Test without sync after creating and removing clones
wrote 1048576/1048576 bytes at offset 0
1 MiB, 1 ops; 0.0017 sec (556.793 MiB/sec and 556.7929 ops/sec)
Setting up swapspace version 1, size = 1020 KiB (1044480 bytes)
no label, UUID=a6b9c29e-5ef4-4689-a8ac-bc199c750f02
swapon: /mnt/sdi/foo_clone_50: swapon failed: Invalid argument
swapoff: /mnt/sdi/foo_clone_50: swapoff failed: Invalid argument
Test with sync after creating clones
wrote 1048576/1048576 bytes at offset 0
1 MiB, 1 ops; 0.0036 sec (271.739 MiB/sec and 271.7391 ops/sec)
Setting up swapspace version 1, size = 1020 KiB (1044480 bytes)
no label, UUID=5e9008d6-1f7a-4948-a1b4-3f30aba20a33
swapon: /mnt/sdi/foo_clone_50: swapon failed: Invalid argument
swapoff: /mnt/sdi/foo_clone_50: swapoff failed: Invalid argument
Test with sync after removing clones
wrote 1048576/1048576 bytes at offset 0
1 MiB, 1 ops; 0.0103 sec (96.665 MiB/sec and 96.6651 ops/sec)
Setting up swapspace version 1, size = 1020 KiB (1044480 bytes)
no label, UUID=916c2740-fa9f-4385-9f06-29c3f89e4764
Test with sync after creating and removing clones
wrote 1048576/1048576 bytes at offset 0
1 MiB, 1 ops; 0.0031 sec (314.268 MiB/sec and 314.2678 ops/sec)
Setting up swapspace version 1, size = 1020 KiB (1044480 bytes)
no label, UUID=06aab1dd-4d90-49c0-bd9f-3a8db4e2f912
swapon: /mnt/sdi/foo_clone_50: swapon failed: Invalid argument
swapoff: /mnt/sdi/foo_clone_50: swapoff failed: Invalid argument
Fix this by reworking btrfs_swap_activate() to instead of using extent
maps and checking for shared extents with can_nocow_extent(), iterate
over the inode's file extent items and use the accurate
btrfs_is_data_extent_shared().
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When activating the swap file we flush all delalloc and wait for ordered
extent completion, so that we don't miss any delalloc and extents before
we check that the file's extent layout is usable for a swap file and
activate the swap file. We are called with the inode's VFS lock acquired,
so we won't race with buffered and direct IO writes, however we can still
race with memory mapped writes since they don't acquire the inode's VFS
lock. The race window is between flushing all delalloc and locking the
whole file's extent range, since memory mapped writes lock an extent range
with the length of a page.
Fix this by acquiring the inode's mmap lock before we flush delalloc.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When we call btrfs_read_folio() we get an unlocked folio, so it is possible
for a different thread to concurrently modify folio->mapping. We must
check that this hasn't happened once we do have the lock.
CC: stable@vger.kernel.org # 6.12+
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Boris Burkov <boris@bur.io>
Signed-off-by: David Sterba <dsterba@suse.com>
When we call btrfs_read_folio() to bring a folio uptodate, we unlock the
folio. The result of that is that a different thread can modify the
mapping (like remove it with invalidate) before we call folio_lock().
This results in an invalid page and we need to try again.
In particular, if we are relocating concurrently with aborting a
transaction, this can result in a crash like the following:
BUG: kernel NULL pointer dereference, address: 0000000000000000
PGD 0 P4D 0
Oops: 0000 [#1] SMP
CPU: 76 PID: 1411631 Comm: kworker/u322:5
Workqueue: events_unbound btrfs_reclaim_bgs_work
RIP: 0010:set_page_extent_mapped+0x20/0xb0
RSP: 0018:ffffc900516a7be8 EFLAGS: 00010246
RAX: ffffea009e851d08 RBX: ffffea009e0b1880 RCX: 0000000000000000
RDX: 0000000000000000 RSI: ffffc900516a7b90 RDI: ffffea009e0b1880
RBP: 0000000003573000 R08: 0000000000000001 R09: ffff88c07fd2f3f0
R10: 0000000000000000 R11: 0000194754b575be R12: 0000000003572000
R13: 0000000003572fff R14: 0000000000100cca R15: 0000000005582fff
FS: 0000000000000000(0000) GS:ffff88c07fd00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 000000407d00f002 CR4: 00000000007706f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
<TASK>
? __die+0x78/0xc0
? page_fault_oops+0x2a8/0x3a0
? __switch_to+0x133/0x530
? wq_worker_running+0xa/0x40
? exc_page_fault+0x63/0x130
? asm_exc_page_fault+0x22/0x30
? set_page_extent_mapped+0x20/0xb0
relocate_file_extent_cluster+0x1a7/0x940
relocate_data_extent+0xaf/0x120
relocate_block_group+0x20f/0x480
btrfs_relocate_block_group+0x152/0x320
btrfs_relocate_chunk+0x3d/0x120
btrfs_reclaim_bgs_work+0x2ae/0x4e0
process_scheduled_works+0x184/0x370
worker_thread+0xc6/0x3e0
? blk_add_timer+0xb0/0xb0
kthread+0xae/0xe0
? flush_tlb_kernel_range+0x90/0x90
ret_from_fork+0x2f/0x40
? flush_tlb_kernel_range+0x90/0x90
ret_from_fork_asm+0x11/0x20
</TASK>
This occurs because cleanup_one_transaction() calls
destroy_delalloc_inodes() which calls invalidate_inode_pages2() which
takes the folio_lock before setting mapping to NULL. We fail to check
this, and subsequently call set_extent_mapping(), which assumes that
mapping != NULL (in fact it asserts that in debug mode)
Note that the "fixes" patch here is not the one that introduced the
race (the very first iteration of this code from 2009) but a more recent
change that made this particular crash happen in practice.
Fixes: e7f1326cc2 ("btrfs: set page extent mapped after read_folio in relocate_one_page")
CC: stable@vger.kernel.org # 6.1+
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Boris Burkov <boris@bur.io>
Signed-off-by: David Sterba <dsterba@suse.com>
Remove the variable length in btrfs_insert_one_raid_extent() as it is
unused.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When a COWing a tree block, at btrfs_cow_block(), and we have the
tracepoint trace_btrfs_cow_block() enabled and preemption is also enabled
(CONFIG_PREEMPT=y), we can trigger a use-after-free in the COWed extent
buffer while inside the tracepoint code. This is because in some paths
that call btrfs_cow_block(), such as btrfs_search_slot(), we are holding
the last reference on the extent buffer @buf so btrfs_force_cow_block()
drops the last reference on the @buf extent buffer when it calls
free_extent_buffer_stale(buf), which schedules the release of the extent
buffer with RCU. This means that if we are on a kernel with preemption,
the current task may be preempted before calling trace_btrfs_cow_block()
and the extent buffer already released by the time trace_btrfs_cow_block()
is called, resulting in a use-after-free.
Fix this by moving the trace_btrfs_cow_block() from btrfs_cow_block() to
btrfs_force_cow_block() before the COWed extent buffer is freed.
This also has a side effect of invoking the tracepoint in the tree defrag
code, at defrag.c:btrfs_realloc_node(), since btrfs_force_cow_block() is
called there, but this is fine and it was actually missing there.
Reported-by: syzbot+8517da8635307182c8a5@syzkaller.appspotmail.com
Link: https://lore.kernel.org/linux-btrfs/6759a9b9.050a0220.1ac542.000d.GAE@google.com/
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
There is a recent ML report that mounting a large fs backed by hardware
RAID56 controller (with one device missing) took too much time, and
systemd seems to kill the mount attempt.
In that case, the only error message is:
BTRFS error (device sdj): open_ctree failed
There is no reason on why the failure happened, making it very hard to
understand the reason.
At least output the error number (in the particular case it should be
-EINTR) to provide some clue.
Link: https://lore.kernel.org/linux-btrfs/9b9c4d2810abcca2f9f76e32220ed9a90febb235.camel@scientia.org/
Reported-by: Christoph Anton Mitterer <calestyo@scientia.org>
Cc: stable@vger.kernel.org
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The function btrfs_copy_from_user() handles the folio dirtying for
buffered write. The original design is to allow that function to handle
multiple folios, but since commit c87c299776 ("btrfs: make buffered
write to copy one page a time") there is no need to support multiple
folios.
So here open-code btrfs_copy_from_user() to
copy_folio_from_iter_atomic() and flush_dcache_folio() calls.
The short-copy check and revert are still kept as-is.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
[WARNING]
There are several warnings about the recently introduced qgroup
auto-removal that it triggers WARN_ON() for the non-zero rfer/excl
numbers, e.g:
------------[ cut here ]------------
WARNING: CPU: 67 PID: 2882 at fs/btrfs/qgroup.c:1854 btrfs_remove_qgroup+0x3df/0x450
CPU: 67 UID: 0 PID: 2882 Comm: btrfs-cleaner Kdump: loaded Not tainted 6.11.6-300.fc41.x86_64 #1
RIP: 0010:btrfs_remove_qgroup+0x3df/0x450
Call Trace:
<TASK>
btrfs_qgroup_cleanup_dropped_subvolume+0x97/0xc0
btrfs_drop_snapshot+0x44e/0xa80
btrfs_clean_one_deleted_snapshot+0xc3/0x110
cleaner_kthread+0xd8/0x130
kthread+0xd2/0x100
ret_from_fork+0x34/0x50
ret_from_fork_asm+0x1a/0x30
</TASK>
---[ end trace 0000000000000000 ]---
BTRFS warning (device sda): to be deleted qgroup 0/319 has non-zero numbers, rfer 258478080 rfer_cmpr 258478080 excl 0 excl_cmpr 0
[CAUSE]
Although the root cause is still unclear, as if qgroup is consistent a
fully dropped subvolume (with extra transaction committed) should lead
to all zero numbers for the qgroup.
My current guess is the subvolume drop triggered the new subtree drop
threshold thus marked qgroup inconsistent, then rescan cleared it but
some corner case is not properly handled during subvolume dropping.
But at least for this particular case, since it's only the rfer/excl not
properly reset to 0, and qgroup is already marked inconsistent, there is
nothing to be worried for the end users.
The user space tool utilizing qgroup would queue a rescan to handle
everything, so the kernel wanring is a little overkilled.
[ENHANCEMENT]
Enhance the warning inside btrfs_remove_qgroup() by:
- Only do WARN() if CONFIG_BTRFS_DEBUG is enabled
As explained the kernel can handle inconsistent qgroups by simply do a
rescan, there is nothing to bother the end users.
- Treat the reserved space leak the same as non-zero numbers
By outputting the values and trigger a WARN() if it's a debug build.
So far I haven't experienced any case related to reserved space so I
hope we will never need to bother them.
Fixes: 839d6ea4f8 ("btrfs: automatically remove the subvolume qgroup")
Link: https://github.com/kdave/btrfs-progs/issues/922
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We don't ever look at this list, remove it.
Reviewed-by: Boris Burkov <boris@bur.io>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Before we were keeping all of our nodes on various lists in order to
make sure everything got cleaned up correctly. We used node->lowest to
indicate that node->lower was linked into the cache->leaves list. Now
that we do cleanup based on the rb-tree both the list and the flag are
useless, so delete them both.
Reviewed-by: Boris Burkov <boris@bur.io>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We rely on finding all our nodes on the various lists in the backref
cache, when they are all also in the rbtree. Instead just search
through the rbtree and free everything.
Reviewed-by: Boris Burkov <boris@bur.io>
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 handle relocation for non-shareable roots without using the
backref cache, remove the ->cowonly field from the backref nodes and
update the handling to throw an error.
Reviewed-by: Boris Burkov <boris@bur.io>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We already determine the owner for any blocks we find when we're
relocating, and for COW-only blocks (and the data reloc tree) we COW
down to the block and call it good enough. However we still build a
whole backref tree for them, even though we're not going to use it, and
then just don't put these blocks in the cache.
Rework the code to check if the block belongs to a COW-only root or the
data reloc root, and then just cow down to the block, skipping the
backref cache generation.
Reviewed-by: Boris Burkov <boris@bur.io>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Since we no longer maintain backref cache across transactions, and this
is only called when we're creating the reloc root for a newly created
snapshot in the transaction critical section, we will end up doing a
bunch of work that will just get thrown away when we start the
transaction in the relocation loop. Delete this code as it no longer
does anything for us.
Reviewed-by: Boris Burkov <boris@bur.io>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We have this setup as a loop, but in reality we will never walk back up
the backref tree, if we do then it's a bug. Get rid of the loop and
handle the case where we have node->new_bytenr set at all. Previous
check was only if node->new_bytenr != root->node->start, but if it did
then we would hit the WARN_ON() and walk back up the tree.
Instead we want to just return error if ->new_bytenr is set, and then do
the normal updating of the node for the reloc root and carry on.
Reviewed-by: Boris Burkov <boris@bur.io>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Add a comment for this field so we know what it is used for. Previously
we used it to update the backref cache, so people may mistakenly think
it is useless, but in fact exists to make sure the backref cache makes
sense.
Reviewed-by: Boris Burkov <boris@bur.io>
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 not updating the backref cache when we switch transids we
can remove the changed list.
We're going to keep the new_bytenr field because it serves as a good
sanity check for the backref cache and relocation, and can prevent us
from making extent tree corruption worse.
Reviewed-by: Boris Burkov <boris@bur.io>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This BUG_ON is meant to catch backref cache problems, but these can
arise from either bugs in the backref cache or corruption in the extent
tree. Fix it to be a proper error.
Reviewed-by: Boris Burkov <boris@bur.io>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
[BUG]
There is a bug report in the mailing list where btrfs_run_delayed_refs()
failed to drop the ref count for logical 25870311358464 num_bytes
2113536.
The involved leaf dump looks like this:
item 166 key (25870311358464 168 2113536) itemoff 10091 itemsize 50
extent refs 1 gen 84178 flags 1
ref#0: shared data backref parent 32399126528000 count 0 <<<
ref#1: shared data backref parent 31808973717504 count 1
Notice the count number is 0.
[CAUSE]
There is no concrete evidence yet, but considering 0 -> 1 is also a
single bit flipped, it's possible that hardware memory bitflip is
involved, causing the on-disk extent tree to be corrupted.
[FIX]
To prevent us reading such corrupted extent item, or writing such
damaged extent item back to disk, enhance the handling of
BTRFS_EXTENT_DATA_REF_KEY and BTRFS_SHARED_DATA_REF_KEY keys for both
inlined and key items, to detect such 0 ref count and reject them.
Link: https://lore.kernel.org/linux-btrfs/7c69dd49-c346-4806-86e7-e6f863a66f48@app.fastmail.com/
Reported-by: Frankie Fisher <frankie@terrorise.me.uk>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Btrfs like other file systems can't really deal with I/O not aligned to
it's internal block size (which strangely is called sector size in
btrfs, for historical reasons), but the block layer split helper doesn't
even know about that.
Round down the split boundary so that all I/Os are aligned.
Fixes: d5e4377d50 ("btrfs: split zone append bios in btrfs_submit_bio")
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Otherwise it won't catch bios turned into regular writes by the block
level zone write plugging. The additional test it adds is for emulated
zone append.
Fixes: 9b1ce7f0c6 ("block: Implement zone append emulation")
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
A data race occurs when the function `insert_ordered_extent_file_extent()`
and the function `btrfs_inode_safe_disk_i_size_write()` are executed
concurrently. The function `insert_ordered_extent_file_extent()` is not
locked when reading inode->disk_i_size, causing
`btrfs_inode_safe_disk_i_size_write()` to cause data competition when
writing inode->disk_i_size, thus affecting the value of `modify_tree`.
The specific call stack that appears during testing is as follows:
============DATA_RACE============
btrfs_drop_extents+0x89a/0xa060 [btrfs]
insert_reserved_file_extent+0xb54/0x2960 [btrfs]
insert_ordered_extent_file_extent+0xff5/0x1760 [btrfs]
btrfs_finish_one_ordered+0x1b85/0x36a0 [btrfs]
btrfs_finish_ordered_io+0x37/0x60 [btrfs]
finish_ordered_fn+0x3e/0x50 [btrfs]
btrfs_work_helper+0x9c9/0x27a0 [btrfs]
process_scheduled_works+0x716/0xf10
worker_thread+0xb6a/0x1190
kthread+0x292/0x330
ret_from_fork+0x4d/0x80
ret_from_fork_asm+0x1a/0x30
============OTHER_INFO============
btrfs_inode_safe_disk_i_size_write+0x4ec/0x600 [btrfs]
btrfs_finish_one_ordered+0x24c7/0x36a0 [btrfs]
btrfs_finish_ordered_io+0x37/0x60 [btrfs]
finish_ordered_fn+0x3e/0x50 [btrfs]
btrfs_work_helper+0x9c9/0x27a0 [btrfs]
process_scheduled_works+0x716/0xf10
worker_thread+0xb6a/0x1190
kthread+0x292/0x330
ret_from_fork+0x4d/0x80
ret_from_fork_asm+0x1a/0x30
=================================
The main purpose of the check of the inode's disk_i_size is to avoid
taking write locks on a btree path when we have a write at or beyond
eof, since in these cases we don't expect to find extent items in the
root to drop. However if we end up taking write locks due to a data
race on disk_i_size, everything is still correct, we only add extra
lock contention on the tree in case there's concurrency from other tasks.
If the race causes us to not take write locks when we actually need them,
then everything is functionally correct as well, since if we find out we
have extent items to drop and we took read locks (modify_tree set to 0),
we release the path and retry again with write locks.
Since this data race does not affect the correctness of the function,
it is a harmless data race, use data_race() to check inode->disk_i_size.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Hao-ran Zheng <zhenghaoran154@gmail.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_drop_extents() calls BUG_ON() in case the counter of to be deleted
extents is greater than 0. But all of these code paths can handle errors,
so there's no need to crash the kernel. Instead WARN() that the condition
has been met and gracefully bail out.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
On the zoned mode, once used and freed region is still not reusable after the
freeing. The underlying zone needs to be reset before reusing. Btrfs resets a
zone when it removes a block group, and then new block group is allocated on
the zones to reuse the zones. But, it is sometime too late to catch up with a
write side.
This commit introduces a new space-info reclaim method ZONE_RESET. That will
pick a block group from the unused list and reset its zone to reuse the
zone_unusable space. It is faster than removing the block group and re-creating
a new block group on the same zones.
For the first implementation, the ZONE_RESET is only applied to a block group
whose region is fully zone_unusable. Reclaiming partial zone_unusable block
group could be implemented later.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Since commit e1e577aafe41 ("btrfs: store fs_info in space_info"), we have
the fs_info in a space_info. So, we can drop fs_info argument from
btrfs_update_space_info_*. There is no behavior change.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Factor out a part of unpin_extent_range() that returns space back to the
space info, prioritizing global block reserve. Also, move the "len"
variable into the loop to clarify we don't need to carry it beyond an
iteration.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Commit 146054090b ("btrfs: initial fsverity support") introduced
fs-verity support for btrfs, but didn't add support for
FS_IOC_READ_VERITY_METADATA to directly query the Merkle tree,
descriptor and signature blocks for fs-verity enabled files.
Add the (trival) implementation: we just need to wire it through to the
fs-verity code, the same way as is done in the other two filesystems
which support this ioctl (ext4, f2fs). The fs-verity code already has
access to the required data.
This is also safe to backport to older stable trees (5.15+) if needed.
Signed-off-by: Allison Karlitskaya <allison.karlitskaya@redhat.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The variable ret is being initialized to zero and also later re-assigned
to zero. In both cases the assignment is redundant since the value is
never read after the assignment and hence they can be removed.
Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The function btrfs_get_extent() will only return an PTR_ERR() or a valid
extent map pointer. It will not return NULL.
Thus the usage of PTR_ERR_OR_ZERO() inside submit_one_sector() is not
needed, use plain PTR_ERR() instead, and that is the only usage of
PTR_ERR_OR_ZERO() after btrfs_get_extent().
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Use kmemdup() in btrfs_uring_encoded_read() rather than kmalloc() followed by
memcpy().
Link: https://lore.kernel.org/oe-kbuild-all/202411050846.GI8oh5IK-lkp@intel.com/
Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Mark Harmstone <maharmstone@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The recent fix for a stupid mistake I made uncovered the fact that we
don't have adequate testing in the delayed refs code, as it took a
pretty extensive and long running stress test to uncover something that
a unit test would have uncovered right away.
Fix this by adding a delayed refs self test suite. This will validate
that the btrfs_ref transformation does the correct thing, that we do the
correct thing when merging delayed refs, and that we get the delayed
refs in the order that we expect. These are all crucial to how the
delayed refs operate.
I introduced various bugs (including the original bug) into the delayed
refs code to validate that these tests caught all of the shenanigans
that I could think of.
Reviewed-by: Boris Burkov <boris@bur.io>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This helper is how we select the delayed ref to run once we've selected
the delayed ref head. I need this exported to add a unit test for
delayed refs, and it's more natural home is in delayed-ref.c. Rename it
to btrfs_select_delayed_ref and move it into delayed-ref.c.
Reviewed-by: Boris Burkov <boris@bur.io>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We have been using the following check
if (generation <= root->root_key.offset)
to make decisions about whether or not to visit a node during snapshot
delete. This is because for normal subvolumes this is set to 0, and for
snapshots it's set to the creation generation. The idea being that if
the generation of the node is less than or equal to our creation
generation then we don't need to visit that node, because it doesn't
belong to us, we can simply drop our reference and move on.
However reloc roots don't have their generation stored in
root->root_key.offset, instead that is the objectid of their
corresponding fs root. This means we can incorrectly not walk into
nodes that need to be dropped when deleting a reloc root.
There are a variety of consequences to making the wrong choice in two
distinct areas.
visit_node_for_delete()
1. False positive. We think we are newer than the block when we really
aren't. We don't visit the node and drop our reference to the node
and carry on. This would result in leaked space.
2. False negative. We do decide to walk down into a block that we
should have just dropped our reference to. However this means that
the child node will have refs > 1, so we will switch to
UPDATE_BACKREF, and then the subsequent walk_down_proc() will notice
that btrfs_header_owner(node) != root->root_key.objectid and it'll
break out of the loop, and then walk_up_proc() will drop our reference,
so this appears to be ok.
do_walk_down()
1. False positive. We are in UPDATE_BACKREF and incorrectly decide that
we are done and don't need to update the backref for our lower nodes.
This is another case that simply won't happen with relocation, as we
only have to do UPDATE_BACKREF if the node below us was shared and
didn't have FULL_BACKREF set, and since we don't own that node
because we're a reloc root we actually won't end up in this case.
2. False negative. Again this is tricky because as described above, we
simply wouldn't be here from relocation, because we don't own any of
the nodes because we never set btrfs_header_owner() to the reloc root
objectid, and we always use FULL_BACKREF, we never actually need to
set FULL_BACKREF on any children.
Having spent a lot of time stressing relocation/snapshot delete recently
I've not seen this pop in practice. But this is objectively incorrect,
so fix this to get the correct starting generation based on the root
we're dropping to keep me from thinking there's a problem here.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Simplify the I/O completion path for encoded reads by using a
completion instead of a wait_queue.
Furthermore use refcount_t instead of atomic_t for reference counting the
private data.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
-----BEGIN PGP SIGNATURE-----
iQIzBAABCgAdFiEE8rQSAMVO+zA4DBdWxWXV+ddtWDsFAmdYzmoACgkQxWXV+ddt
WDv5GxAAnCsGctNax89x/VpCDZynRghrkxlzu/4kG/pqxsJyzlgXDFtzHAEewSMs
MYL+WCZLYpeKB5FpZq98mDJVLGNMG+9wqkx1bH/xy2ajBGZTeQe5pnkXMNlv9U1O
SX34t8nzOdTCENDnQeRc5I2vTcsQRhgHoVjJkAYdWdhcD9fs6xHKZRe+himlstSn
46ioKzEKSR3ztEUW4ycPF379g7d4kTR0hkk3pu5Nxe7ER8iq+jNSWXj0mzKg7mpJ
KxP56VgY0OrsiUcJr2qFZ1hQIp810puaAuM4C1lLgRplECHxtLbP9JvL9Rr7a8Ox
68tuThyLEpQtR59078jIX3RK6CwVi15rKb/ZkLZkW19TNSAAfM5qrB146hLBUM4T
16WaiJ0x9lVkH2oYQv8zbNZiqDxPhPUdS/JArNAcQYk9ma+C1hCsxPQ/N5yoWH/C
OABJddNR83sm4VTXu3Nci1EB8QuEoOuihYO6CdRkJ3PPNDuQiG6gwnoA2zqSihhy
L5fQaLSWAUsLczarHZrvAi9Y0rfG66QzqGR+A1K/8qMTQ8pSCupd+LfqVa21QpI1
Awx/wVFzsAm7z9CrnPTRJe+JSlBDQdeXWX7pDhhkXgwbCsMVSf3dbBweCD3o1EiM
BVI7SfEgImlbatd0QvDp9FcsnEqp90SCi+99U+zZCmQ1SW8CEC0=
=+DUB
-----END PGP SIGNATURE-----
Merge tag 'for-6.13-rc2-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs fixes from David Sterba:
"A few more fixes. Apart from the one liners and updated bio splitting
error handling there's a fix for subvolume mount with different flags.
This was known and fixed for some time but I've delayed it to give it
more testing.
- fix unbalanced locking when swapfile activation fails when the
subvolume gets deleted in the meantime
- add btrfs error handling after bio_split() calls that got error
handling recently
- during unmount, flush delalloc workers at the right time before the
cleaner thread is shut down
- fix regression in buffered write folio conversion, explicitly wait
for writeback as FGP_STABLE flag is currently a no-op on btrfs
- handle race in subvolume mount with different flags, the conversion
to the new mount API did not handle the case where multiple
subvolumes get mounted in parallel, which is a distro use case"
* tag 'for-6.13-rc2-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
btrfs: flush delalloc workers queue before stopping cleaner kthread during unmount
btrfs: handle bio_split() errors
btrfs: properly wait for writeback before buffered write
btrfs: fix missing snapshot drew unlock when root is dead during swap activation
btrfs: fix mount failure due to remount races
Commit e546fe1da9 ("block: Rework bio_split() return value") changed
bio_split() so that it can return errors.
Add error handling for it in btrfs_split_bio() and ultimately
btrfs_submit_chunk(). As the bio is not submitted, the bio counter must
be decremented to pair btrfs_bio_counter_inc_blocked().
Reviewed-by: John Garry <john.g.garry@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
[BUG]
Before commit e820dbeb6a ("btrfs: convert btrfs_buffered_write() to
use folios"), function prepare_one_folio() will always wait for folio
writeback to finish before returning the folio.
However commit e820dbeb6a ("btrfs: convert btrfs_buffered_write() to
use folios") changed to use FGP_STABLE to do the writeback wait, but
FGP_STABLE is calling folio_wait_stable(), which only calls
folio_wait_writeback() if the address space has AS_STABLE_WRITES, which
is not set for btrfs inodes.
This means we will not wait for the folio writeback at all.
[CAUSE]
The cause is FGP_STABLE is not waiting for writeback unconditionally, but
only for address spaces with AS_STABLE_WRITES, normally such flag is set
when the super block has SB_I_STABLE_WRITES flag.
Such super block flag is set when the block device has hardware digest
support or has internal checksum requirement.
I'd argue btrfs should set such super block due to its default data
checksum behavior, but it is not set yet, so this means FGP_STABLE flag
will have no effect at all.
(For NODATASUM inodes, we can skip the waiting in theory but that should
be an optimization in the future.)
This can lead to data checksum mismatch, as we can modify the folio
while it's still under writeback, this will make the contents differ
from the contents at submission and checksum calculation.
[FIX]
Instead of fully relying on FGP_STABLE, manually do the folio writeback
waiting, until we set the address space or super flag.
Fixes: e820dbeb6a ("btrfs: convert btrfs_buffered_write() to use folios")
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When activating a swap file we acquire the root's snapshot drew lock and
then check if the root is dead, failing and returning with -EPERM if it's
dead but without unlocking the root's snapshot lock. Fix this by adding
the missing unlock.
Fixes: 60021bd754 ("btrfs: prevent subvol with swapfile from being deleted")
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
[BUG]
The following reproducer can cause btrfs mount to fail:
dev="/dev/test/scratch1"
mnt1="/mnt/test"
mnt2="/mnt/scratch"
mkfs.btrfs -f $dev
mount $dev $mnt1
btrfs subvolume create $mnt1/subvol1
btrfs subvolume create $mnt1/subvol2
umount $mnt1
mount $dev $mnt1 -o subvol=subvol1
while mount -o remount,ro $mnt1; do mount -o remount,rw $mnt1; done &
bg=$!
while mount $dev $mnt2 -o subvol=subvol2; do umount $mnt2; done
kill $bg
wait
umount -R $mnt1
umount -R $mnt2
The script will fail with the following error:
mount: /mnt/scratch: /dev/mapper/test-scratch1 already mounted on /mnt/test.
dmesg(1) may have more information after failed mount system call.
umount: /mnt/test: target is busy.
umount: /mnt/scratch/: not mounted
And there is no kernel error message.
[CAUSE]
During the btrfs mount, to support mounting different subvolumes with
different RO/RW flags, we need to detect that and retry if needed:
Retry with matching RO flags if the initial mount fail with -EBUSY.
The problem is, during that retry we do not hold any super block lock
(s_umount), this means there can be a remount process changing the RO
flags of the original fs super block.
If so, we can have an EBUSY error during retry. And this time we treat
any failure as an error, without any retry and cause the above EBUSY
mount failure.
[FIX]
The current retry behavior is racy because we do not have a super block
thus no way to hold s_umount to prevent the race with remount.
Solve the root problem by allowing fc->sb_flags to mismatch from the
sb->s_flags at btrfs_get_tree_super().
Then at the re-entry point btrfs_get_tree_subvol(), manually check the
fc->s_flags against sb->s_flags, if it's a RO->RW mismatch, then
reconfigure with s_umount lock hold.
Reported-by: Enno Gotthold <egotthold@suse.com>
Reported-by: Fabian Vogt <fvogt@suse.com>
[ Special thanks for the reproducer and early analysis pointing to btrfs. ]
Fixes: f044b31867 ("btrfs: handle the ro->rw transition for mounting different subvolumes")
Link: https://bugzilla.suse.com/show_bug.cgi?id=1231836
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
