Regular for-next build test #1157
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
kdave
force-pushed
the
for-next
branch
6 times, most recently
from
2d4aefb
to
c9e380a
Compare
kdave
force-pushed
the
for-next
branch
6 times, most recently
from
c56343b
to
1cab137
Compare
kdave
force-pushed
the
for-next
branch
2 times, most recently
from
6613f3c
to
b30a0ce
Compare
kdave
force-pushed
the
for-next
branch
6 times, most recently
from
d205ebd
to
c0bd9d9
Compare
kdave
force-pushed
the
for-next
branch
4 times, most recently
from
15022b1
to
c22750c
Compare
kdave
force-pushed
the
for-next
branch
3 times, most recently
from
28d9855
to
e18d8ce
Compare
After the previous patch, which converted the rb-tree used to track delayed ref heads into an xarray, the find_ref_head() function is now used only by one caller which always passes false to the 'return_bigger' argument. So remove the 'return_bigger' logic, simplifying the function, and move all the function code to the single caller. Reviewed-by: Boris Burkov <boris@bur.io> Reviewed-by: Qu Wenruo <wqu@suse.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
iocb->ki_pos isn't used after this function, so there's no point in changing its value. Signed-off-by: Mark Harmstone <maharmstone@fb.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
…by caller Change the behaviour of btrfs_encoded_read() so that if it needs to read an extent from disk, it leaves the extent and inode locked and returns -EIOCBQUEUED. The caller is then responsible for doing the I/O via btrfs_encoded_read_regular() and unlocking the extent and inode. Signed-off-by: Mark Harmstone <maharmstone@fb.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
Change btrfs_encoded_read() so that it returns -EAGAIN rather than sleeps if IOCB_NOWAIT is set in iocb->ki_flags. The conditions that require sleeping are: inode lock, writeback, extent lock, ordered range. Signed-off-by: Mark Harmstone <maharmstone@fb.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
Change btrfs_encoded_read_regular_fill_pages() so that the priv struct is allocated rather than stored on the stack, in preparation for adding an asynchronous mode to the function. Signed-off-by: Mark Harmstone <maharmstone@fb.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
Add an io_uring command for encoded reads, using the same interface as the existing BTRFS_IOC_ENCODED_READ ioctl. btrfs_uring_encoded_read() is an io_uring version of btrfs_ioctl_encoded_read(), which validates the user input and calls btrfs_encoded_read() to read the appropriate metadata. If we determine that we need to read an extent from disk, we call btrfs_encoded_read_regular_fill_pages() through btrfs_uring_read_extent() to prepare the bio. The existing btrfs_encoded_read_regular_fill_pages() is changed so that if it is passed a valid uring_ctx, rather than waking up any waiting threads it calls btrfs_uring_read_extent_endio(). This in turn copies the read data back to userspace, and calls io_uring_cmd_done() to complete the io_uring command. Because we're potentially doing a non-blocking read, btrfs_uring_read_extent() doesn't clean up after itself if it returns -EIOCBQUEUED. Instead, it allocates a priv struct, populates the fields there that we will need to unlock the inode and free our allocations, and defers this to the btrfs_uring_read_finished() that gets called when the bio completes. Signed-off-by: Mark Harmstone <maharmstone@fb.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
Add struct io_btrfs_cmd as a wrapper type for io_uring_cmd_to_pdu(), rather than using a raw pointer. Suggested-by: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Mark Harmstone <maharmstone@fb.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
When the taks that submitted a request is dying, a task work for that request might get run by a kernel thread or even worse by a half dismantled task. We can't just cancel the task work without running the callback as the cmd might need to do some clean up, so pass a flag instead. If set, it's not safe to access any task resources and the callback is expected to cancel the cmd ASAP. Reviewed-by: Jens Axboe <axboe@kernel.dk> Reviewed-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: David Sterba <dsterba@suse.com>
Move btrfs_add_inode_to_root() so it can be called from btrfs_read_locked_inode(), no changes were made to the function. Move cleanup code from btrfs_iget_path() to btrfs_read_locked_inode. This improves readability and improves a leaky abstraction. Previously btrfs_iget_path() had to handle a positive error case as a result of a call to btrfs_search_slot(), but it makes more sense to handle this closer to the source of the call. Signed-off-by: Leo Martins <loemra.dev@gmail.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
Remove conditional path allocation from btrfs_read_locked_inode(). Add an ASSERT(path) to indicate it should never be called with a NULL path. Call btrfs_read_locked_inode() directly from btrfs_iget(). This causes code duplication between btrfs_iget() and btrfs_iget_path(), but I think this is justifiable as it removes the need for conditionally allocating the path inside of btrfs_read_locked_inode(). This makes the code easier to reason about and makes it clear who has the responsibility of allocating and freeing the path. Signed-off-by: Leo Martins <loemra.dev@gmail.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
Simplify tracking of the range processed by using cur_alloc_size only to store the reserved part that may fail to the allocated extent. Remove the ram_size as well since it is always equal to cur_alloc_size in the context. Advance the start in normal path until extent allocation succeeds and keep the start unchanged in the error handling path. Passed the fstest generic/475 test for a hundred times with quota enabled. And a modified generic/475 test by removing the sleep time for a hundred times. About one tenth of the tests do enter the error handling path due to fail to reserve extent. Suggested-by: Qu Wenruo <wqu@suse.com> Signed-off-by: Haisu Wang <haisuwang@tencent.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
Add a new unprivileged ioctl that will let the command 'btrfs subvolume sync' work without the (privileged) SEARCH_TREE ioctl. There are several modes of operation, where the most common ones are to wait on a specific subvolume or all currently queued for cleaning. This is utilized e.g. in backup applications that delete subvolumes and wait until they're cleaned to check for remaining space. The other modes are for flexibility, e.g. for monitoring or checkpoints in the queue of deleted subvolumes, again without the need to use SEARCH_TREE. Notes: - waiting is interruptible, the timeout is set to 1 second and is not configurable - repeated calls to the ioctl see a different state, so this is inherently racy when using e.g. the count or peek next/last Use cases: - a subvolume A was deleted, wait for cleaning (WAIT_FOR_ONE) - a bunch of subvolumes were deleted, wait for all (WAIT_FOR_QUEUED or PEEK_LAST + WAIT_FOR_ONE) - count how many are queued (not blocking), for monitoring purposes - report progress (PEEK_NEXT), may miss some if cleaning is quick - own waiting in user space (PEEK_LAST until it's 0) Signed-off-by: David Sterba <dsterba@suse.com>
The comment refers to a list in the respective delayed ref head that no longer exists (ref_list), it was replaced with a rbtree (ref_tree) in commit 0e0adbc ("btrfs: track refs in a rb_tree instead of a list"). So update the stale comment to refer to the rbtree instead of the old list. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
On x86_64 and a release kernel, there's a 4 bytes hole in the structure
after the ref count field:
struct btrfs_delayed_node {
u64 inode_id; /* 0 8 */
u64 bytes_reserved; /* 8 8 */
struct btrfs_root * root; /* 16 8 */
struct list_head n_list; /* 24 16 */
struct list_head p_list; /* 40 16 */
struct rb_root_cached ins_root; /* 56 16 */
/* --- cacheline 1 boundary (64 bytes) was 8 bytes ago --- */
struct rb_root_cached del_root; /* 72 16 */
struct mutex mutex; /* 88 32 */
struct btrfs_inode_item inode_item; /* 120 160 */
/* --- cacheline 4 boundary (256 bytes) was 24 bytes ago --- */
refcount_t refs; /* 280 4 */
/* XXX 4 bytes hole, try to pack */
u64 index_cnt; /* 288 8 */
long unsigned int flags; /* 296 8 */
int count; /* 304 4 */
u32 curr_index_batch_size; /* 308 4 */
u32 index_item_leaves; /* 312 4 */
/* size: 320, cachelines: 5, members: 15 */
/* sum members: 312, holes: 1, sum holes: 4 */
/* padding: 4 */
};
Move the 'count' field, which is 4 bytes long, to just below the ref count
field, so we eliminate the hole and reduce the structure size from 320
bytes down to 312 bytes:
struct btrfs_delayed_node {
u64 inode_id; /* 0 8 */
u64 bytes_reserved; /* 8 8 */
struct btrfs_root * root; /* 16 8 */
struct list_head n_list; /* 24 16 */
struct list_head p_list; /* 40 16 */
struct rb_root_cached ins_root; /* 56 16 */
/* --- cacheline 1 boundary (64 bytes) was 8 bytes ago --- */
struct rb_root_cached del_root; /* 72 16 */
struct mutex mutex; /* 88 32 */
struct btrfs_inode_item inode_item; /* 120 160 */
/* --- cacheline 4 boundary (256 bytes) was 24 bytes ago --- */
refcount_t refs; /* 280 4 */
int count; /* 284 4 */
u64 index_cnt; /* 288 8 */
long unsigned int flags; /* 296 8 */
u32 curr_index_batch_size; /* 304 4 */
u32 index_item_leaves; /* 308 4 */
/* size: 312, cachelines: 5, members: 15 */
/* last cacheline: 56 bytes */
};
This now allows to have 13 delayed nodes per 4K page instead of 12.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
…ot() There's no point in having a 'snapshot_force_cow' variable to track if we need to decrement the root->snapshot_force_cow counter, as we never jump to the 'out' label after incrementing the counter. Simplify this by removing the variable and always decrementing the counter before the 'out' label, right after the call to btrfs_mksubvol(). Reviewed-by: Qu Wenruo <wqu@suse.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
…read() Change the control flow of btrfs_encoded_read() so that it doesn't call free_extent_map() when we know that this has already been done. Reviewed-by: Anand Jain <anand.jain@oracle.com> Signed-off-by: Mark Harmstone <maharmstone@fb.com> Suggested-by: Anand Jain <anand.jain@oracle.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
REQ_OP_ZONE_APPNED -> REQ_OP_ZONE_APPEND. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
…ioctl() Smatch complains about calling PTR_ERR() against a NULL pointer: fs/btrfs/super.c:2272 btrfs_control_ioctl() warn: passing zero to 'PTR_ERR' Fix this by calling PTR_ERR() against the device pointer only if it contains an error. Reviewed-by: Qu Wenruo <wqu@suse.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
Smatch complains about possibly dereferencing a NULL fs_info at btrfs_folio_end_lock_bitmap(): fs/btrfs/subpage.c:332 btrfs_folio_end_lock_bitmap() warn: variable dereferenced before check 'fs_info' (see line 326) because we access fs_info to set the 'start_bit' variable before doing the check for a NULL fs_info. However fs_info is never NULL, since in the only caller of btrfs_folio_end_lock_bitmap() is extent_writepage(), where we have an inode which always as a non-NULL fs_info. So remove the check for a NULL fs_info at btrfs_folio_end_lock_bitmap(). Reviewed-by: Qu Wenruo <wqu@suse.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
We're checking if the send root is dead without the protection of the root's root_item_lock spinlock, which is what protects the root's flags. The inverse, setting the dead flag on a root, is done under the protection of that lock, at btrfs_delete_subvolume(). Also checking and updating the root's send_in_progress counter is supposed to be done in the same critical section as checking for or setting the root dead flag, so that these operations are done atomically as a single step (which is correctly done by btrfs_delete_subvolume()). So fix this by checking if the send root is dead in the same critical section that updates the send_in_progress counter, which is protected by the root's root_item_lock spinlock. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
We're checking if the send root is read-only without being under the protection of the root's root_item_lock spinlock, which is what protects the root's flags when clearing the read-only flag, done at btrfs_ioctl_subvol_setflags(). Furthermore, it should be done in the same critical section that increments the root's send_in_progress counter, as btrfs_ioctl_subvol_setflags() clears the read-only flag in the same critical section that checks the counter's value. So fix this by moving the read-only check under the critical section delimited by the root's root_item_lock which also increments the root's send_in_progress counter. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
Shinichiro reported the following use-after free that sometimes is happening in our CI system when running fstests' btrfs/284 on a TCMU runner device: ================================================================== BUG: KASAN: slab-use-after-free in lock_release+0x708/0x780 Read of size 8 at addr ffff888106a83f18 by task kworker/u80:6/219 CPU: 8 UID: 0 PID: 219 Comm: kworker/u80:6 Not tainted 6.12.0-rc6-kts+ #15 Hardware name: Supermicro Super Server/X11SPi-TF, BIOS 3.3 02/21/2020 Workqueue: btrfs-endio btrfs_end_bio_work [btrfs] Call Trace: <TASK> dump_stack_lvl+0x6e/0xa0 ? lock_release+0x708/0x780 print_report+0x174/0x505 ? lock_release+0x708/0x780 ? __virt_addr_valid+0x224/0x410 ? lock_release+0x708/0x780 kasan_report+0xda/0x1b0 ? lock_release+0x708/0x780 ? __wake_up+0x44/0x60 lock_release+0x708/0x780 ? __pfx_lock_release+0x10/0x10 ? __pfx_do_raw_spin_lock+0x10/0x10 ? lock_is_held_type+0x9a/0x110 _raw_spin_unlock_irqrestore+0x1f/0x60 __wake_up+0x44/0x60 btrfs_encoded_read_endio+0x14b/0x190 [btrfs] btrfs_check_read_bio+0x8d9/0x1360 [btrfs] ? lock_release+0x1b0/0x780 ? trace_lock_acquire+0x12f/0x1a0 ? __pfx_btrfs_check_read_bio+0x10/0x10 [btrfs] ? process_one_work+0x7e3/0x1460 ? lock_acquire+0x31/0xc0 ? process_one_work+0x7e3/0x1460 process_one_work+0x85c/0x1460 ? __pfx_process_one_work+0x10/0x10 ? assign_work+0x16c/0x240 worker_thread+0x5e6/0xfc0 ? __pfx_worker_thread+0x10/0x10 kthread+0x2c3/0x3a0 ? __pfx_kthread+0x10/0x10 ret_from_fork+0x31/0x70 ? __pfx_kthread+0x10/0x10 ret_from_fork_asm+0x1a/0x30 </TASK> Allocated by task 3661: kasan_save_stack+0x30/0x50 kasan_save_track+0x14/0x30 __kasan_kmalloc+0xaa/0xb0 btrfs_encoded_read_regular_fill_pages+0x16c/0x6d0 [btrfs] send_extent_data+0xf0f/0x24a0 [btrfs] process_extent+0x48a/0x1830 [btrfs] changed_cb+0x178b/0x2ea0 [btrfs] btrfs_ioctl_send+0x3bf9/0x5c20 [btrfs] _btrfs_ioctl_send+0x117/0x330 [btrfs] btrfs_ioctl+0x184a/0x60a0 [btrfs] __x64_sys_ioctl+0x12e/0x1a0 do_syscall_64+0x95/0x180 entry_SYSCALL_64_after_hwframe+0x76/0x7e Freed by task 3661: kasan_save_stack+0x30/0x50 kasan_save_track+0x14/0x30 kasan_save_free_info+0x3b/0x70 __kasan_slab_free+0x4f/0x70 kfree+0x143/0x490 btrfs_encoded_read_regular_fill_pages+0x531/0x6d0 [btrfs] send_extent_data+0xf0f/0x24a0 [btrfs] process_extent+0x48a/0x1830 [btrfs] changed_cb+0x178b/0x2ea0 [btrfs] btrfs_ioctl_send+0x3bf9/0x5c20 [btrfs] _btrfs_ioctl_send+0x117/0x330 [btrfs] btrfs_ioctl+0x184a/0x60a0 [btrfs] __x64_sys_ioctl+0x12e/0x1a0 do_syscall_64+0x95/0x180 entry_SYSCALL_64_after_hwframe+0x76/0x7e The buggy address belongs to the object at ffff888106a83f00 which belongs to the cache kmalloc-rnd-07-96 of size 96 The buggy address is located 24 bytes inside of freed 96-byte region [ffff888106a83f00, ffff888106a83f60) The buggy address belongs to the physical page: page: refcount:1 mapcount:0 mapping:0000000000000000 index:0xffff888106a83800 pfn:0x106a83 flags: 0x17ffffc0000000(node=0|zone=2|lastcpupid=0x1fffff) page_type: f5(slab) raw: 0017ffffc0000000 ffff888100053680 ffffea0004917200 0000000000000004 raw: ffff888106a83800 0000000080200019 00000001f5000000 0000000000000000 page dumped because: kasan: bad access detected Memory state around the buggy address: ffff888106a83e00: fa fb fb fb fb fb fb fb fb fb fb fb fc fc fc fc ffff888106a83e80: fa fb fb fb fb fb fb fb fb fb fb fb fc fc fc fc >ffff888106a83f00: fa fb fb fb fb fb fb fb fb fb fb fb fc fc fc fc ^ ffff888106a83f80: fa fb fb fb fb fb fb fb fb fb fb fb fc fc fc fc ffff888106a84000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ================================================================== Further analyzing the trace and the crash dump's vmcore file shows that the wake_up() call in btrfs_encoded_read_endio() is calling wake_up() on the wait_queue that is in the private data passed to the end_io handler. Commit 4ff47df ("btrfs: move priv off stack in btrfs_encoded_read_regular_fill_pages()") moved 'struct btrfs_encoded_read_private' off the stack. Before that commit one can see a corruption of the private data when analyzing the vmcore after a crash: *(struct btrfs_encoded_read_private *)0xffff88815626eec8 = { .wait = (wait_queue_head_t){ .lock = (spinlock_t){ .rlock = (struct raw_spinlock){ .raw_lock = (arch_spinlock_t){ .val = (atomic_t){ .counter = (int)-2005885696, }, .locked = (u8)0, .pending = (u8)157, .locked_pending = (u16)40192, .tail = (u16)34928, }, .magic = (unsigned int)536325682, .owner_cpu = (unsigned int)29, .owner = (void *)__SCT__tp_func_btrfs_transaction_commit+0x0 = 0x0, .dep_map = (struct lockdep_map){ .key = (struct lock_class_key *)0xffff8881575a3b6c, .class_cache = (struct lock_class *[2]){ 0xffff8882a71985c0, 0xffffea00066f5d40 }, .name = (const char *)0xffff88815626f100 = "", .wait_type_outer = (u8)37, .wait_type_inner = (u8)178, .lock_type = (u8)154, }, }, .__padding = (u8 [24]){ 0, 157, 112, 136, 50, 174, 247, 31, 29 }, .dep_map = (struct lockdep_map){ .key = (struct lock_class_key *)0xffff8881575a3b6c, .class_cache = (struct lock_class *[2]){ 0xffff8882a71985c0, 0xffffea00066f5d40 }, .name = (const char *)0xffff88815626f100 = "", .wait_type_outer = (u8)37, .wait_type_inner = (u8)178, .lock_type = (u8)154, }, }, .head = (struct list_head){ .next = (struct list_head *)0x112cca, .prev = (struct list_head *)0x47, }, }, .pending = (atomic_t){ .counter = (int)-1491499288, }, .status = (blk_status_t)130, } Here we can see several indicators of in-memory data corruption, e.g. the large negative atomic values of ->pending or ->wait->lock->rlock->raw_lock->val, as well as the bogus spinlock magic 0x1ff7ae32 (decimal 536325682 above) instead of 0xdead4ead or the bogus pointer values for ->wait->head. To fix this, change atomic_dec_return() to atomic_dec_and_test() to fix the corruption, as atomic_dec_return() is defined as two instructions on x86_64, whereas atomic_dec_and_test() is defined as a single atomic operation. This can lead to a situation where counter value is already decremented but the if statement in btrfs_encoded_read_endio() is not completely processed, i.e. the 0 test has not completed. If another thread continues executing btrfs_encoded_read_regular_fill_pages() the atomic_dec_return() there can see an already updated ->pending counter and continues by freeing the private data. Continuing in the endio handler the test for 0 succeeds and the wait_queue is woken up, resulting in a use-after-free. Reported-by: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com> Suggested-by: Damien Le Moal <Damien.LeMoal@wdc.com> Fixes: 1881fba ("btrfs: add BTRFS_IOC_ENCODED_READ ioctl") Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: Qu Wenruo <wqu@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. Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: Qu Wenruo <wqu@suse.com>
When running a workload with fsstress and duperemove (generic/561) we can hit a deadlock related to transaction commits and locking extent ranges, as described below. Task A hanging during a transaction commit, waiting for all other writers to complete: [178317.334817] INFO: task fsstress:555623 blocked for more than 120 seconds. [178317.335693] Not tainted 6.12.0-rc6-btrfs-next-179+ #1 [178317.336528] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [178317.337673] task:fsstress state:D stack:0 pid:555623 tgid:555623 ppid:555620 flags:0x00004002 [178317.337679] Call Trace: [178317.337681] <TASK> [178317.337685] __schedule+0x364/0xbe0 [178317.337691] schedule+0x26/0xa0 [178317.337695] btrfs_commit_transaction+0x5c5/0x1050 [btrfs] [178317.337769] ? start_transaction+0xc4/0x800 [btrfs] [178317.337815] ? __pfx_autoremove_wake_function+0x10/0x10 [178317.337819] btrfs_mksubvol+0x381/0x640 [btrfs] [178317.337878] btrfs_mksnapshot+0x7a/0xb0 [btrfs] [178317.337935] __btrfs_ioctl_snap_create+0x1bb/0x1d0 [btrfs] [178317.337995] btrfs_ioctl_snap_create_v2+0x103/0x130 [btrfs] [178317.338053] btrfs_ioctl+0x29b/0x2a90 [btrfs] [178317.338118] ? kmem_cache_alloc_noprof+0x5f/0x2c0 [178317.338126] ? getname_flags+0x45/0x1f0 [178317.338133] ? _raw_spin_unlock+0x15/0x30 [178317.338145] ? __x64_sys_ioctl+0x88/0xc0 [178317.338149] __x64_sys_ioctl+0x88/0xc0 [178317.338152] do_syscall_64+0x4a/0x110 [178317.338160] entry_SYSCALL_64_after_hwframe+0x76/0x7e [178317.338190] RIP: 0033:0x7f13c28e271b Which corresponds to line 2361 of transaction.c: $ cat -n fs/btrfs/transaction.c (...) 2162 int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 2163 { (...) 2349 spin_lock(&fs_info->trans_lock); 2350 add_pending_snapshot(trans); 2351 cur_trans->state = TRANS_STATE_COMMIT_DOING; 2352 spin_unlock(&fs_info->trans_lock); 2353 2354 /* 2355 * The thread has started/joined the transaction thus it holds the 2356 * lockdep map as a reader. It has to release it before acquiring the 2357 * lockdep map as a writer. 2358 */ 2359 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2360 btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2361 wait_event(cur_trans->writer_wait, 2362 atomic_read(&cur_trans->num_writers) == 1); (...) The transaction is in the TRANS_STATE_COMMIT_DOING state and so it's waiting for all other existing writers to complete and release their transaction handle. Task B is running ordered extent completion and blocked waiting to lock an extent range in an inode's io tree: [178317.327411] INFO: task kworker/u48:8:554545 blocked for more than 120 seconds. [178317.328630] Not tainted 6.12.0-rc6-btrfs-next-179+ #1 [178317.329635] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [178317.330872] task:kworker/u48:8 state:D stack:0 pid:554545 tgid:554545 ppid:2 flags:0x00004000 [178317.330878] Workqueue: btrfs-endio-write btrfs_work_helper [btrfs] [178317.330944] Call Trace: [178317.330945] <TASK> [178317.330947] __schedule+0x364/0xbe0 [178317.330952] schedule+0x26/0xa0 [178317.330955] __lock_extent+0x337/0x3a0 [btrfs] [178317.331014] ? __pfx_autoremove_wake_function+0x10/0x10 [178317.331017] btrfs_finish_one_ordered+0x47a/0xaa0 [btrfs] [178317.331074] ? psi_group_change+0x132/0x2d0 [178317.331078] btrfs_work_helper+0xbd/0x370 [btrfs] [178317.331140] process_scheduled_works+0xd3/0x460 [178317.331144] ? __pfx_worker_thread+0x10/0x10 [178317.331146] worker_thread+0x121/0x250 [178317.331149] ? __pfx_worker_thread+0x10/0x10 [178317.331151] kthread+0xe9/0x120 [178317.331154] ? __pfx_kthread+0x10/0x10 [178317.331157] ret_from_fork+0x2d/0x50 [178317.331159] ? __pfx_kthread+0x10/0x10 [178317.331162] ret_from_fork_asm+0x1a/0x30 This extent range locking happens after joining the current transaction, so task A is waiting for task B to release its transaction handle (decrementing the transaction's num_writers counter). Task C while doing a fiemap it tries to join the current transaction: [242682.812815] task:pool state:D stack:0 pid:560767 tgid:560724 ppid:555622 flags:0x00004006 [242682.812827] Call Trace: [242682.812856] <TASK> [242682.812864] __schedule+0x364/0xbe0 [242682.812879] ? _raw_spin_unlock_irqrestore+0x23/0x40 [242682.812897] schedule+0x26/0xa0 [242682.812909] wait_current_trans+0xd6/0x130 [btrfs] [242682.813148] ? __pfx_autoremove_wake_function+0x10/0x10 [242682.813162] start_transaction+0x3d4/0x800 [btrfs] [242682.813399] btrfs_is_data_extent_shared+0xd2/0x440 [btrfs] [242682.813723] fiemap_process_hole+0x2a2/0x300 [btrfs] [242682.813995] extent_fiemap+0x9b8/0xb80 [btrfs] [242682.814249] btrfs_fiemap+0x78/0xc0 [btrfs] [242682.814501] do_vfs_ioctl+0x2db/0xa50 [242682.814519] __x64_sys_ioctl+0x6a/0xc0 [242682.814531] do_syscall_64+0x4a/0x110 [242682.814544] entry_SYSCALL_64_after_hwframe+0x76/0x7e [242682.814556] RIP: 0033:0x7efff595e71b It tries to join the current transaction, but it can't because the transaction is in the TRANS_STATE_COMMIT_DOING state, so join_transaction() returns -EBUSY to start_transaction() and makes it wait for the current transaction to complete. And while it's waiting for the transaction to complete, it's holding an extent range locked in the same inode that task B is operating, which causes a deadlock between these 3 tasks. The extent range for the inode was locked at the start of the fiemap operation, early at extent_fiemap(). In short these tasks deadlock because: 1) Task A is waiting for task B to release its transaction handle; 2) Task B is waiting to lock an extent range for an inode while holding a transaction handle open; 3) Task C is waiting for the current transaction to complete (for task A to finish the transaction commit) while holding the extent range for the inode locked, so task B can't progress and release its transaction handle. This results in an ABBA deadlock involving transaction commits and extent locks. Extent locks are higher level locks, like inode VFS locks, and should always be acquired before joining or starting a transaction, but recently commit 2206265 ("btrfs: remove code duplication in ordered extent finishing") accidentally changed btrfs_finish_one_ordered() to do the transaction join before locking the extent range. Fix this by making sure that btrfs_finish_one_ordered() always locks the extent before joining a transaction and add an explicit comment about the need for this order. Fixes: 2206265 ("btrfs: remove code duplication in ordered extent finishing") Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by: Filipe Manana <fdmanana@suse.com>
…PERIMENTAL=y We are advertising experimental features through sysfs if CONFIG_BTRFS_DEBUG is set, without looking if CONFIG_BTRFS_EXPERIMENTAL is set. This is wrong as it will result in reporting experimental features as supported when CONFIG_BTRFS_EXPERIMENTAL is not set but CONFIG_BTRFS_DEBUG is set. Fix this by checking for CONFIG_BTRFS_EXPERIMENTAL instead of CONFIG_BTRFS_DEBUG. Fixes: 67cd3f2 ("btrfs: split out CONFIG_BTRFS_EXPERIMENTAL from CONFIG_BTRFS_DEBUG") Reviewed-by: Neal Gompa <neal@gompa.dev> Reviewed-by: Qu Wenruo <wqu@suse.com> Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by: Filipe Manana <fdmanana@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>
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>
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>
When checking for delayed refs when verifying if there are cross references for a data extent, we stop if the path has nowait set and we can't try lock the delayed ref head's mutex, returning -EAGAIN with the goal of making a write fallback to a blocking context. However we ignore the -EAGAIN at btrfs_cross_ref_exist() when check_delayed_ref() returns it, and keep looping instead of immediately returning the -EAGAIN to the caller. Fix this by not looping if we get -EAGAIN and we have a nowait path. Fixes: 26ce911 ("btrfs: make can_nocow_extent nowait compatible") Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Keep this open, the build tests are hosted on github CI.