This patch reduces to call them across the whole tree.
- sync_inode_page()
- update_inode_page()
- update_inode()
- f2fs_write_inode()
Instead, checkpoint will flush all the dirty inode metadata before syncing
node pages.
Note that, this is doable, since we call mark_inode_dirty_sync() for all
inode's field change which needs to update on-disk inode as well.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch calls mark_inode_dirty_sync() for the following on-disk inode
changes.
-> largest
-> ctime/mtime/atime
-> i_current_depth
-> i_xattr_nid
-> i_pino
-> i_advise
-> i_flags
-> i_mode
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch introduces f2fs_i_links_write() to call mark_inode_dirty_sync() when
changing inode->i_links.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch introduces f2fs_i_blocks_write() to call mark_inode_dirty_sync() when
changing inode->i_blocks.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In ee1d267423 ("pstore: add pstore unregister") I added:
.owner = THIS_MODULE,
in both pstore_fs_type and pstore_file_operations to increase a reference
count when pstore filesystem is mounted and pstore file is opened.
But, it's repetitive. There is no need to increase the opened reference
count. We only need to increase the mounted reference count. When a file
is opened, the filesystem can't be unmounted. Hence the pstore module
can't be unloaded either.
So I drop the opened reference count in this patch.
Fixes: ee1d267423 ("pstore: add pstore unregister")
Signed-off-by: Geliang Tang <geliangtang@163.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Like zlib compression in pstore, this patch added lzo and lz4
compression support so that users can have more options and better
compression ratio.
The original code treats the compressed data together with the
uncompressed ECC correction notice by using zlib decompress. The
ECC correction notice is missing in the decompression process. The
treatment also makes lzo and lz4 not working. So I treat them
separately by using pstore_decompress() to treat the compressed
data, and memcpy() to treat the uncompressed ECC correction notice.
Signed-off-by: Geliang Tang <geliangtang@163.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
self-tests code assumes 4k as the sectorsize and nodesize. This commit
fix hardcoded 4K. Enables the self-tests code to be executed on non-4k
page sized systems (e.g. ppc64).
Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Feifei Xu <xufeifei@linux.vnet.ibm.com>
Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
Signed-off-by: David Sterba <dsterba@suse.com>
On a ppc64 machine using 64K as the block size, assume that the RB
tree at btrfs_free_space_ctl->free_space_offset contains following
two entries:
1. A bitmap entry having an offset value of 0 and having the bits
corresponding to the address range [128M+512K, 128M+768K] set.
2. An extent entry corresponding to the address range
[128M-256K, 128M-128K]
In such a scenario, test_check_exists() invoked for checking the
existence of address range [128M+768K, 256M] can lead to an
infinite loop as explained below:
- Checking for the extent entry fails.
- Checking for a bitmap entry results in the free space info in
range [128M+512K, 128M+768K] beng returned.
- rb_prev(info) returns NULL because the bitmap entry starting from
offset 0 comes first in the RB tree.
- current_node = bitmap node.
- while (current_node)
tmp = rb_next(bitmap_node);/*tmp is extent based free space entry*/
Since extent based free space entry's last address is smaller
than the address being searched for (i.e. 128M+768K) we
incorrectly again obtain the extent node as the "next right node"
of the RB tree and thus end up looping infinitely.
This patch fixes the issue by checking the "tmp" variable which point
to the most recently searched free space node.
Reviewed-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
Signed-off-by: Feifei Xu <xufeifei@linux.vnet.ibm.com>
Signed-off-by: David Sterba <dsterba@suse.com>
There are several issues in fscache revalidation code.
- In ceph_revalidate_work(), fscache_invalidate() is called when
fscache_check_consistency() return 0. This is complete wrong
because 0 means cache is valid.
- Handle_cap_grant() calls ceph_queue_revalidate() if client
already has CAP_FILE_CACHE. This code is confusing. Client
should revalidate the cache each time it got CAP_FILE_CACHE
anew.
- In Handle_cap_grant(), fscache_invalidate() is called if MDS
revokes CAP_FILE_CACHE. This is inconsistency with the case
that inode get evicted. In the later case, the cache is not
discarded. Client may use the cache when inode is reloaded.
This patch moves the fscache revalidation into ceph_get_caps().
Client revalidates the cache after it gets CAP_FILE_CACHE.
i_rdcache_gen should keep constance while CAP_FILE_CACHE is
used. If i_fscache_gen is not equal to i_rdcache_gen, client
needs to check cache's consistency.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
All other filesystems do not add dirty pages to fscache. They all
disable fscache when inode is opened for write. Only ceph adds
dirty pages to fscache, but the code is buggy.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
__fscache_check_consistency() calls check_consistency() callback
and return the callback's return value. But the return type of
check_consistency() is bool. So __fscache_check_consistency()
return 1 if the cache is inconsistent. This is inconsistent with
the document.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Acked-by: David Howells <dhowells@redhat.com>
When we have a lot of metadata to flush from the AIL, the buffer
list can get very long. The current submission code tries to batch
submission to optimise IO order of the metadata (i.e. ascending
block order) to maximise block layer merging or IO to adjacent
metadata blocks.
Unfortunately, the method used can result in long lock times
occurring as buffers locked early on in the buffer list might not be
dispatched until the end of the IO licst processing. This is because
sorting does not occur util after the buffer list has been processed
and the buffers that are going to be submitted are locked. Hence
when the buffer list is several thousand buffers long, the lock hold
times before IO dispatch can be significant.
To fix this, sort the buffer list before we start trying to lock and
submit buffers. This means we can now submit buffers immediately
after they are locked, allowing merging to occur immediately on the
plug and dispatch to occur as quickly as possible. This means there
is minimal delay between locking the buffer and IO submission
occuring, hence reducing the worst case lock hold times seen during
delayed write buffer IO submission signficantly.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
And the same for XFS_IOC_THAW. Just because we now have a common
version of the ioctl we still need to provide the old name for it
for anyone using those.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Al Viro noticed that xfs_lock_inodes should be static, and
that led to ... a few more.
These are just the easy ones, others require moving functions
higher in source files, so that's not done here to keep
this review simple.
Signed-off-by: Eric Sandeen <sandeen@sandeen.net>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The static checker reports that after commit 8d99fe92fe ("xfs: fix
efi/efd error handling to avoid fs shutdown hangs"), the code has been
reworked such that error == -EFSCORRUPTED is not possible in this
codepath.
Remove the spurious error check and just use SHUTDOWN_META_IO_ERROR
unconditionally.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Multi-block buffers are logged based on buffer offset in
xfs_trans_log_buf(). xfs_buf_item_log() ultimately walks each mapping in
the buffer and marks the associated range to be logged in the
xfs_buf_log_format bitmap for that mapping. This code is broken,
however, in that it marks the actual buffer offsets of the associated
range in each bitmap rather than shifting to the byte range for that
particular mapping.
For example, on a 4k fsb fs, buffer offset 4096 refers to the first byte
of the second mapping in the buffer. This means byte 0 of the second log
format bitmap should be tagged as dirty. Instead, the current code marks
byte offset 4096 of the second log format bitmap, which is invalid and
potentially out of range of the mapping.
As a result of this, the log item format code invoked at transaction
commit time is not be able to correctly identify what parts of the
buffer to copy into log vectors. This can lead to NULL log vector
pointer dereferences in CIL push context if the item format code was not
able to locate any dirty ranges at all. This crash has been reproduced
on a 4k FSB filesystem using 16k directory blocks where an unlink
operation happened not to log anything in the first block of the
mapping. The logged offsets were all over 4k, marked as such in the
subsequent log format mappings, and thus left the transaction with an
xfs_log_item that is marked DIRTY but without any logged regions.
Further, even when the logged regions are marked correctly in the buffer
log format bitmaps, the format code doesn't copy the correct ranges of
the buffer into the log. This means that any logged region beyond the
first block of a multi-block buffer is subject to corruption after a
crash and log recovery sequence. This is due to a failure to convert the
mapping bm_len field from basic blocks to bytes in the buffer offset
tracking code in xfs_buf_item_format().
Update xfs_buf_item_log() to convert buffer offsets to segment relative
offsets when logging multi-block buffers. This ensures that the modified
regions of a buffer are logged correctly and avoids the aforementioned
crash. Also update xfs_buf_item_format() to correctly track the source
offset into the buffer for the log vector formatting code. This ensures
that the correct data is copied into the log.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
This driver predates those methods and was trying to be clever allocating
it's own private data. Switch to the generic scheme used by other file
systems.
Based on an earlier patch from Krzysztof Błaszkowski <kb@sysmikro.com.pl>.
Signed-off-by: Christoph Hellwig <hch@lst.de>
To support VxFS filesystems from HP-UX on x86 systems we need to
implement byte swapping, and to keep support for Unixware filesystems
it needs to be the complicated dual-endian kind ala sysvfs.
To do this properly we have to split the on disk and in-core inode
so that we can keep the in-core one in native endianness. All other
structures are byteswapped on demand.
Signed-off-by: Krzysztof Błaszkowski <kb@sysmikro.com.pl>
[hch: make spare happy]
Signed-off-by: Christoph Hellwig <hch@lst.de>
We still need to call btrfs_end_transaction if we call btrfs_abort_transaction,
otherwise we hang and make me super grumpy. Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The commit f0e2efcfd2 ("pstore: do not use message compression
without lock") added a check to 'is_locked' to avoid breakage in
concurrent accesses. But it has a side-effect of disabling compression
on normal path since 'is_locked' variable is not set. As normal path
always takes the lock, it should be initialized to 1.
This also makes the unlock code a bit simpler.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
While none of the "fragile" pstore backends unregister yet, if they
ever did, the unregistering code for the non-dump targets might get
confused. This adds a check for fragile backends on unregister.
Signed-off-by: Kees Cook <keescook@chromium.org>
While we are finishing a device replace operation we can have a concurrent
task trying to do a read repair operation, in which case it will call
btrfs_map_block() to get a struct btrfs_bio which can have a stripe that
points to the source device of the device replace operation. This allows
for the read repair task to dereference the stripe's device pointer after
the device replace operation has freed the source device, resulting in
an invalid memory access. This is similar to the problem solved by my
previous patch in the same series and named "Btrfs: fix race between
device replace and discard".
So fix this by surrounding the call to btrfs_map_block() and the code
that uses the returned struct btrfs_bio with calls to
btrfs_bio_counter_inc_blocked() and btrfs_bio_counter_dec(), giving the
proper serialization with the finishing phase of the device replace
operation.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
While we are finishing a device replace operation, we can make a discard
operation (fs mounted with -o discard) do an invalid memory access like
the one reported by the following trace:
[ 3206.384654] general protection fault: 0000 [#1] PREEMPT SMP
[ 3206.387520] Modules linked in: dm_mod btrfs crc32c_generic xor raid6_pq acpi_cpufreq tpm_tis psmouse tpm ppdev sg parport_pc evdev i2c_piix4 parport
processor serio_raw i2c_core pcspkr button loop autofs4 ext4 crc16 jbd2 mbcache sr_mod cdrom ata_generic sd_mod virtio_scsi ata_piix libata virtio_pci
virtio_ring scsi_mod e1000 virtio floppy [last unloaded: btrfs]
[ 3206.388595] CPU: 14 PID: 29194 Comm: fsstress Not tainted 4.6.0-rc7-btrfs-next-29+ #1
[ 3206.388595] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS by qemu-project.org 04/01/2014
[ 3206.388595] task: ffff88017ace0100 ti: ffff880171b98000 task.ti: ffff880171b98000
[ 3206.388595] RIP: 0010:[<ffffffff8124d233>] [<ffffffff8124d233>] blkdev_issue_discard+0x5c/0x2a7
[ 3206.388595] RSP: 0018:ffff880171b9bb80 EFLAGS: 00010246
[ 3206.388595] RAX: ffff880171b9bc28 RBX: 000000000090d000 RCX: 0000000000000000
[ 3206.388595] RDX: ffffffff82fa1b48 RSI: ffffffff8179f46c RDI: ffffffff82fa1b48
[ 3206.388595] RBP: ffff880171b9bcc0 R08: 0000000000000000 R09: 0000000000000001
[ 3206.388595] R10: ffff880171b9bce0 R11: 000000000090f000 R12: ffff880171b9bbe8
[ 3206.388595] R13: 0000000000000010 R14: 0000000000004868 R15: 6b6b6b6b6b6b6b6b
[ 3206.388595] FS: 00007f6182e4e700(0000) GS:ffff88023fdc0000(0000) knlGS:0000000000000000
[ 3206.388595] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 3206.388595] CR2: 00007f617c2bbb18 CR3: 000000017ad9c000 CR4: 00000000000006e0
[ 3206.388595] Stack:
[ 3206.388595] 0000000000004878 0000000000000000 0000000002400040 0000000000000000
[ 3206.388595] 0000000000000000 ffff880171b9bbe8 ffff880171b9bbb0 ffff880171b9bbb0
[ 3206.388595] ffff880171b9bbc0 ffff880171b9bbc0 ffff880171b9bbd0 ffff880171b9bbd0
[ 3206.388595] Call Trace:
[ 3206.388595] [<ffffffffa042899e>] btrfs_issue_discard+0x12f/0x143 [btrfs]
[ 3206.388595] [<ffffffffa042899e>] ? btrfs_issue_discard+0x12f/0x143 [btrfs]
[ 3206.388595] [<ffffffffa042e862>] btrfs_discard_extent+0x87/0xde [btrfs]
[ 3206.388595] [<ffffffffa04303b5>] btrfs_finish_extent_commit+0xb2/0x1df [btrfs]
[ 3206.388595] [<ffffffff8149c246>] ? __mutex_unlock_slowpath+0x150/0x15b
[ 3206.388595] [<ffffffffa04464c4>] btrfs_commit_transaction+0x7fc/0x980 [btrfs]
[ 3206.388595] [<ffffffff8149c246>] ? __mutex_unlock_slowpath+0x150/0x15b
[ 3206.388595] [<ffffffffa0459af6>] btrfs_sync_file+0x38f/0x428 [btrfs]
[ 3206.388595] [<ffffffff811a8292>] vfs_fsync_range+0x8c/0x9e
[ 3206.388595] [<ffffffff811a82c0>] vfs_fsync+0x1c/0x1e
[ 3206.388595] [<ffffffff811a8417>] do_fsync+0x31/0x4a
[ 3206.388595] [<ffffffff811a8637>] SyS_fsync+0x10/0x14
[ 3206.388595] [<ffffffff8149e025>] entry_SYSCALL_64_fastpath+0x18/0xa8
[ 3206.388595] [<ffffffff81100c6b>] ? time_hardirqs_off+0x9/0x14
[ 3206.388595] [<ffffffff8108e87d>] ? trace_hardirqs_off_caller+0x1f/0xaa
This happens because when we call btrfs_map_block() from
btrfs_discard_extent() to get a btrfs_bio structure, the device replace
operation has not finished yet, but before we use the device of one of the
stripes from the returned btrfs_bio structure, the device object is freed.
This is illustrated by the following diagram.
CPU 1 CPU 2
btrfs_dev_replace_start()
(...)
btrfs_dev_replace_finishing()
btrfs_start_transaction()
btrfs_commit_transaction()
(...)
btrfs_sync_file()
btrfs_start_transaction()
(...)
btrfs_commit_transaction()
btrfs_finish_extent_commit()
btrfs_discard_extent()
btrfs_map_block()
--> returns a struct btrfs_bio
with a stripe that has a
device field pointing to
source device of the replace
operation (the device that
is being replaced)
mutex_lock(&uuid_mutex)
mutex_lock(&fs_info->fs_devices->device_list_mutex)
mutex_lock(&fs_info->chunk_mutex)
btrfs_dev_replace_update_device_in_mapping_tree()
--> iterates the mapping tree and for each
extent map that has a stripe pointing to
the source device, it updates the stripe
to point to the target device instead
btrfs_rm_dev_replace_blocked()
--> waits for fs_info->bio_counter to go down to 0
btrfs_rm_dev_replace_remove_srcdev()
--> removes source device from the list of devices
mutex_unlock(&fs_info->chunk_mutex)
mutex_unlock(&fs_info->fs_devices->device_list_mutex)
mutex_unlock(&uuid_mutex)
btrfs_rm_dev_replace_free_srcdev()
--> frees the source device
--> iterates over all stripes
of the returned struct
btrfs_bio
--> for each stripe it
dereferences its device
pointer
--> it ends up finding a
pointer to the device
used as the source
device for the replace
operation and that was
already freed
So fix this by surrounding the call to btrfs_map_block(), and the code
that uses the returned struct btrfs_bio, with calls to
btrfs_bio_counter_inc_blocked() and btrfs_bio_counter_dec(), so that
the finishing phase of the device replace operation blocks until the
the bio counter decreases to zero before it frees the source device.
This is the same approach we do at btrfs_map_bio() for example.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
For the benefit of every single caller, take osdc instead of map.
Also, now that osdc->osdmap can't ever be NULL, drop the check.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
While iterating and copying extents from the source device, the device
replace code keeps adjusting a left cursor that is used to make sure that
once we finish processing a device extent, any future writes to extents
from the corresponding block group will get into both the source and
target devices. This left cursor is also used for resuming the device
replace operation at mount time.
However using this left cursor to decide whether writes go into both
devices or only the source device is not enough to guarantee we don't
miss copying extents into the target device. There are two cases where
the current approach fails. The first one is related to when there are
holes in the device and they get allocated for new block groups while
the device replace operation is iterating the device extents (more on
this explained below). The second one is that when that loop over the
device extents finishes, we start dellaloc, wait for all ordered extents
and then commit the current transaction, we might have got new block
groups allocated that are now using a device extent that has an offset
greater then or equals to the value of the left cursor, in which case
writes to extents belonging to these new block groups will get issued
only to the source device.
For the first case where the current approach of using a left cursor
fails, consider the source device currently has the following layout:
[ extent bg A ] [ hole, unallocated space ] [extent bg B ]
3Gb 4Gb 5Gb
While we are iterating the device extents from the source device using
the commit root of the device tree, the following happens:
CPU 1 CPU 2
<we are at transaction N>
scrub_enumerate_chunks()
--> searches the device tree for
extents belonging to the source
device using the device tree's
commit root
--> 1st iteration finds extent belonging to
block group A
--> sets block group A to RO mode
(btrfs_inc_block_group_ro)
--> sets cursor left to found_key.offset
which is 3Gb
--> scrub_chunk() starts
copies all allocated extents from
block group's A stripe at source
device into target device
btrfs_alloc_chunk()
--> allocates device extent
in the range [4Gb, 5Gb[
from the source device for
a new block group C
extent allocated from block
group C for a direct IO,
buffered write or btree node/leaf
extent is written to, perhaps
in response to a writepages()
call from the VM or directly
through direct IO
the write is made only against
the source device and not against
the target device because the
extent's offset is in the interval
[4Gb, 5Gb[ which is larger then
the value of cursor_left (3Gb)
--> scrub_chunks() finishes
--> updates left cursor from 3Gb to
4Gb
--> btrfs_dec_block_group_ro() sets
block group A back to RW mode
<we are still at transaction N>
--> 2nd iteration finds extent belonging to
block group B - it did not find the new
extent in the range [4Gb, 5Gb[ for block
group C because we are using the device
tree's commit root or even because the
block group's items are not all yet
inserted in the respective btrees, that is,
the block group is still attached to some
transaction handle's new_bgs list and
btrfs_create_pending_block_groups() was
not called yet against that transaction
handle, so the device extent items were
not yet inserted into the devices tree
<we are still at transaction N>
--> so we end not copying anything from the newly
allocated device extent from the source device
to the target device
So fix this by making __btrfs_map_block() always redirect writes to the
target device as well, independently of the left cursor's value. With
this change the left cursor is now used only for the purpose of tracking
progress and allow a mount operation to resume a device replace.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
After it finishes processing a device extent, the device replace code sets
back the block group to RW mode and then after that it sets the left cursor
to match the logical end address of the block group, so that future writes
into extents belonging to the block group go both the source (old) and
target (new) devices. However from the moment we turn the block group
back to RW mode we have a short time window, that lasts until we update
the left cursor's value, where extents can be allocated from the block
group and written to, in which case they will not be copied/written to
the target (new) device. Fix this by updating the left cursor's value
before turning the block group back to RW mode.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
We were assigning new values to fields of the device replace object
without holding the respective lock after processing each device extent.
This is important for the left cursor field which can be accessed by a
concurrent task running __btrfs_map_block (which, correctly, takes the
device replace lock).
So change these fields while holding the device replace lock.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
When we do a device replace, for each device extent we find from the
source device, we set the corresponding block group to readonly mode to
prevent writes into it from happening while we are copying the device
extent from the source to the target device. However just before we set
the block group to readonly mode some concurrent task might have already
allocated an extent from it or decided it could perform a nocow write
into one of its extents, which can make the device replace process to
miss copying an extent since it uses the extent tree's commit root to
search for extents and only once it finishes searching for all extents
belonging to the block group it does set the left cursor to the logical
end address of the block group - this is a problem if the respective
ordered extents finish while we are searching for extents using the
extent tree's commit root and no transaction commit happens while we
are iterating the tree, since it's the delayed references created by the
ordered extents (when they complete) that insert the extent items into
the extent tree (using the non-commit root of course).
Example:
CPU 1 CPU 2
btrfs_dev_replace_start()
btrfs_scrub_dev()
scrub_enumerate_chunks()
--> finds device extent belonging
to block group X
<transaction N starts>
starts buffered write
against some inode
writepages is run against
that inode forcing dellaloc
to run
btrfs_writepages()
extent_writepages()
extent_write_cache_pages()
__extent_writepage()
writepage_delalloc()
run_delalloc_range()
cow_file_range()
btrfs_reserve_extent()
--> allocates an extent
from block group X
(which is not yet
in RO mode)
btrfs_add_ordered_extent()
--> creates ordered extent Y
flush_epd_write_bio()
--> bio against the extent from
block group X is submitted
btrfs_inc_block_group_ro(bg X)
--> sets block group X to readonly
scrub_chunk(bg X)
scrub_stripe(device extent from srcdev)
--> keeps searching for extent items
belonging to the block group using
the extent tree's commit root
--> it never blocks due to
fs_info->scrub_pause_req as no
one tries to commit transaction N
--> copies all extents found from the
source device into the target device
--> finishes search loop
bio completes
ordered extent Y completes
and creates delayed data
reference which will add an
extent item to the extent
tree when run (typically
at transaction commit time)
--> so the task doing the
scrub/device replace
at CPU 1 misses this
and does not copy this
extent into the new/target
device
btrfs_dec_block_group_ro(bg X)
--> turns block group X back to RW mode
dev_replace->cursor_left is set to the
logical end offset of block group X
So fix this by waiting for all cow and nocow writes after setting a block
group to readonly mode.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
When it's finishing, the device replace code iterates all extent maps
representing block group and for each one that has a stripe that refers
to the source device, it replaces its device with the target device.
However when it replaces the source device with the target device it,
the target device still has an ID of 0ULL (BTRFS_DEV_REPLACE_DEVID),
only after its ID is changed to match the one from the source device.
This leads to races with the chunk removal code that can temporarly see
a device with an ID of 0ULL and then attempt to use that ID to remove
items from the device tree and fail, causing a transaction abort:
[ 9238.594364] BTRFS info (device sdf): dev_replace from /dev/sdf (devid 3) to /dev/sde finished
[ 9238.594377] ------------[ cut here ]------------
[ 9238.594402] WARNING: CPU: 14 PID: 21566 at fs/btrfs/volumes.c:2771 btrfs_remove_chunk+0x2e5/0x793 [btrfs]
[ 9238.594403] BTRFS: Transaction aborted (error 1)
[ 9238.594416] Modules linked in: btrfs crc32c_generic acpi_cpufreq xor tpm_tis tpm raid6_pq ppdev parport_pc processor psmouse parport i2c_piix4 evdev sg i2c_core se
rio_raw pcspkr button loop autofs4 ext4 crc16 jbd2 mbcache sr_mod cdrom sd_mod ata_generic virtio_scsi ata_piix virtio_pci libata virtio_ring virtio e1000 scsi_mod fl
oppy [last unloaded: btrfs]
[ 9238.594418] CPU: 14 PID: 21566 Comm: btrfs-cleaner Not tainted 4.6.0-rc7-btrfs-next-29+ #1
[ 9238.594419] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS by qemu-project.org 04/01/2014
[ 9238.594421] 0000000000000000 ffff88017f1dbc60 ffffffff8126b42c ffff88017f1dbcb0
[ 9238.594422] 0000000000000000 ffff88017f1dbca0 ffffffff81052b14 00000ad37f1dbd18
[ 9238.594423] 0000000000000001 ffff88018068a558 ffff88005c4b9c00 ffff880233f60db0
[ 9238.594424] Call Trace:
[ 9238.594428] [<ffffffff8126b42c>] dump_stack+0x67/0x90
[ 9238.594430] [<ffffffff81052b14>] __warn+0xc2/0xdd
[ 9238.594432] [<ffffffff81052b7a>] warn_slowpath_fmt+0x4b/0x53
[ 9238.594434] [<ffffffff8116c311>] ? kmem_cache_free+0x128/0x188
[ 9238.594450] [<ffffffffa04d43f5>] btrfs_remove_chunk+0x2e5/0x793 [btrfs]
[ 9238.594452] [<ffffffff8108e456>] ? arch_local_irq_save+0x9/0xc
[ 9238.594464] [<ffffffffa04a26fa>] btrfs_delete_unused_bgs+0x317/0x382 [btrfs]
[ 9238.594476] [<ffffffffa04a961d>] cleaner_kthread+0x1ad/0x1c7 [btrfs]
[ 9238.594489] [<ffffffffa04a9470>] ? btree_invalidatepage+0x8e/0x8e [btrfs]
[ 9238.594490] [<ffffffff8106f403>] kthread+0xd4/0xdc
[ 9238.594494] [<ffffffff8149e242>] ret_from_fork+0x22/0x40
[ 9238.594495] [<ffffffff8106f32f>] ? kthread_stop+0x286/0x286
[ 9238.594496] ---[ end trace 183efbe50275f059 ]---
The sequence of steps leading to this is like the following:
CPU 1 CPU 2
btrfs_dev_replace_finishing()
at this point
dev_replace->tgtdev->devid ==
BTRFS_DEV_REPLACE_DEVID (0ULL)
...
btrfs_start_transaction()
btrfs_commit_transaction()
btrfs_delete_unused_bgs()
btrfs_remove_chunk()
looks up for the extent map
corresponding to the chunk
lock_chunks() (chunk_mutex)
check_system_chunk()
unlock_chunks() (chunk_mutex)
locks fs_info->chunk_mutex
btrfs_dev_replace_update_device_in_mapping_tree()
--> iterates fs_info->mapping_tree and
replaces the device in every extent
map's map->stripes[] with
dev_replace->tgtdev, which still has
an id of 0ULL (BTRFS_DEV_REPLACE_DEVID)
iterates over all stripes from
the extent map
--> calls btrfs_free_dev_extent()
passing it the target device
that still has an ID of 0ULL
--> btrfs_free_dev_extent() fails
--> aborts current transaction
finishes setting up the target device,
namely it sets tgtdev->devid to the value
of srcdev->devid (which is necessarily > 0)
frees the srcdev
unlocks fs_info->chunk_mutex
So fix this by taking the device list mutex while processing the stripes
for the chunk's extent map. This is similar to the race between device
replace and block group creation that was fixed by commit 50460e3718
("Btrfs: fix race when finishing dev replace leading to transaction abort").
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
There is a lot of duplication between dentry_unlink_inode() and dentry_iput().
The only real difference is that dentry_unlink_inode() bumps ->d_seq and
dentry_iput() doesn't. The argument of the latter is known to have been
unhashed, so anybody who might've found it in RCU lookup would already be
doomed to a ->d_seq mismatch. And we want to avoid pointless smp_rmb() there.
This patch makes dentry_unlink_inode() bump ->d_seq only for hashed dentries.
It's safe (d_delete() calls that sucker only if we are holding the only
reference to dentry, so rehash is not going to happen) and it allows
to use dentry_unlink_inode() in __dentry_kill() and get rid of dentry_iput().
The interesting question here is profiling; it *is* a hot path, and extra
conditional jumps in there might or might not be painful.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
it's not needed for file_operations of inodes located on fs defined
in the hosting module and for file_operations that go into procfs.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
