Rewrite the data and ack handling code such that:
(1) Parsing of received ACK and ABORT packets and the distribution and the
filing of DATA packets happens entirely within the data_ready context
called from the UDP socket. This allows us to process and discard ACK
and ABORT packets much more quickly (they're no longer stashed on a
queue for a background thread to process).
(2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead
keep track of the offset and length of the content of each packet in
the sk_buff metadata. This means we don't do any allocation in the
receive path.
(3) Jumbo DATA packet parsing is now done in data_ready context. Rather
than cloning the packet once for each subpacket and pulling/trimming
it, we file the packet multiple times with an annotation for each
indicating which subpacket is there. From that we can directly
calculate the offset and length.
(4) A call's receive queue can be accessed without taking locks (memory
barriers do have to be used, though).
(5) Incoming calls are set up from preallocated resources and immediately
made live. They can than have packets queued upon them and ACKs
generated. If insufficient resources exist, DATA packet #1 is given a
BUSY reply and other DATA packets are discarded).
(6) sk_buffs no longer take a ref on their parent call.
To make this work, the following changes are made:
(1) Each call's receive buffer is now a circular buffer of sk_buff
pointers (rxtx_buffer) rather than a number of sk_buff_heads spread
between the call and the socket. This permits each sk_buff to be in
the buffer multiple times. The receive buffer is reused for the
transmit buffer.
(2) A circular buffer of annotations (rxtx_annotations) is kept parallel
to the data buffer. Transmission phase annotations indicate whether a
buffered packet has been ACK'd or not and whether it needs
retransmission.
Receive phase annotations indicate whether a slot holds a whole packet
or a jumbo subpacket and, if the latter, which subpacket. They also
note whether the packet has been decrypted in place.
(3) DATA packet window tracking is much simplified. Each phase has just
two numbers representing the window (rx_hard_ack/rx_top and
tx_hard_ack/tx_top).
The hard_ack number is the sequence number before base of the window,
representing the last packet the other side says it has consumed.
hard_ack starts from 0 and the first packet is sequence number 1.
The top number is the sequence number of the highest-numbered packet
residing in the buffer. Packets between hard_ack+1 and top are
soft-ACK'd to indicate they've been received, but not yet consumed.
Four macros, before(), before_eq(), after() and after_eq() are added
to compare sequence numbers within the window. This allows for the
top of the window to wrap when the hard-ack sequence number gets close
to the limit.
Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also
to indicate when rx_top and tx_top point at the packets with the
LAST_PACKET bit set, indicating the end of the phase.
(4) Calls are queued on the socket 'receive queue' rather than packets.
This means that we don't need have to invent dummy packets to queue to
indicate abnormal/terminal states and we don't have to keep metadata
packets (such as ABORTs) around
(5) The offset and length of a (sub)packet's content are now passed to
the verify_packet security op. This is currently expected to decrypt
the packet in place and validate it.
However, there's now nowhere to store the revised offset and length of
the actual data within the decrypted blob (there may be a header and
padding to skip) because an sk_buff may represent multiple packets, so
a locate_data security op is added to retrieve these details from the
sk_buff content when needed.
(6) recvmsg() now has to handle jumbo subpackets, where each subpacket is
individually secured and needs to be individually decrypted. The code
to do this is broken out into rxrpc_recvmsg_data() and shared with the
kernel API. It now iterates over the call's receive buffer rather
than walking the socket receive queue.
Additional changes:
(1) The timers are condensed to a single timer that is set for the soonest
of three timeouts (delayed ACK generation, DATA retransmission and
call lifespan).
(2) Transmission of ACK and ABORT packets is effected immediately from
process-context socket ops/kernel API calls that cause them instead of
them being punted off to a background work item. The data_ready
handler still has to defer to the background, though.
(3) A shutdown op is added to the AF_RXRPC socket so that the AFS
filesystem can shut down the socket and flush its own work items
before closing the socket to deal with any in-progress service calls.
Future additional changes that will need to be considered:
(1) Make sure that a call doesn't hog the front of the queue by receiving
data from the network as fast as userspace is consuming it to the
exclusion of other calls.
(2) Transmit delayed ACKs from within recvmsg() when we've consumed
sufficiently more packets to avoid the background work item needing to
run.
Signed-off-by: David Howells <dhowells@redhat.com>
Make it possible for the data_ready handler called from the UDP transport
socket to completely instantiate an rxrpc_call structure and make it
immediately live by preallocating all the memory it might need. The idea
is to cut out the background thread usage as much as possible.
[Note that the preallocated structs are not actually used in this patch -
that will be done in a future patch.]
If insufficient resources are available in the preallocation buffers, it
will be possible to discard the DATA packet in the data_ready handler or
schedule a BUSY packet without the need to schedule an attempt at
allocation in a background thread.
To this end:
(1) Preallocate rxrpc_peer, rxrpc_connection and rxrpc_call structs to a
maximum number each of the listen backlog size. The backlog size is
limited to a maxmimum of 32. Only this many of each can be in the
preallocation buffer.
(2) For userspace sockets, the preallocation is charged initially by
listen() and will be recharged by accepting or rejecting pending
new incoming calls.
(3) For kernel services {,re,dis}charging of the preallocation buffers is
handled manually. Two notifier callbacks have to be provided before
kernel_listen() is invoked:
(a) An indication that a new call has been instantiated. This can be
used to trigger background recharging.
(b) An indication that a call is being discarded. This is used when
the socket is being released.
A function, rxrpc_kernel_charge_accept() is called by the kernel
service to preallocate a single call. It should be passed the user ID
to be used for that call and a callback to associate the rxrpc call
with the kernel service's side of the ID.
(4) Discard the preallocation when the socket is closed.
(5) Temporarily bump the refcount on the call allocated in
rxrpc_incoming_call() so that rxrpc_release_call() can ditch the
preallocation ref on service calls unconditionally. This will no
longer be necessary once the preallocation is used.
Note that this does not yet control the number of active service calls on a
client - that will come in a later patch.
A future development would be to provide a setsockopt() call that allows a
userspace server to manually charge the preallocation buffer. This would
allow user call IDs to be provided in advance and the awkward manual accept
stage to be bypassed.
Signed-off-by: David Howells <dhowells@redhat.com>
Add a tracepoint for working out where local aborts happen. Each
tracepoint call is labelled with a 3-letter code so that they can be
distinguished - and the DATA sequence number is added too where available.
rxrpc_kernel_abort_call() also takes a 3-letter code so that AFS can
indicate the circumstances when it aborts a call.
Signed-off-by: David Howells <dhowells@redhat.com>
The workqueue "afs_lock_manager" queues work item &vnode->lock_work,
per vnode. Since there can be multiple vnodes and since their work items
can be executed concurrently, alloc_workqueue has been used to replace
the deprecated create_singlethread_workqueue instance.
The WQ_MEM_RECLAIM flag has been set to ensure forward progress under
memory pressure because the workqueue is being used on a memory reclaim
path.
Since there are fixed number of work items, explicit concurrency
limit is unnecessary here.
Signed-off-by: Bhaktipriya Shridhar <bhaktipriya96@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
The workqueue "afs_callback_update_worker" queues multiple work items
viz &vnode->cb_broken_work, &server->cb_break_work which require strict
execution ordering. Hence, an ordered dedicated workqueue has been used.
Since the workqueue is being used on a memory reclaim path, WQ_MEM_RECLAIM
has been set to ensure forward progress under memory pressure.
Signed-off-by: Bhaktipriya Shridhar <bhaktipriya96@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
The workqueue "afs_async_calls" queues work item
&call->async_work per afs_call. Since there could be multiple calls and since
these calls can be run concurrently, alloc_workqueue has been used to replace
the deprecated create_singlethread_workqueue instance.
The WQ_MEM_RECLAIM flag has been set to ensure forward progress under
memory pressure because the workqueue is being used on a memory reclaim
path.
Since there are fixed number of work items, explicit concurrency
limit is unnecessary here.
Signed-off-by: Bhaktipriya Shridhar <bhaktipriya96@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
The workqueue "afs_vlocation_update_worker" queues a single work item
&afs_vlocation_update and hence it doesn't require execution ordering.
Hence, alloc_workqueue has been used to replace the deprecated
create_singlethread_workqueue instance.
Since the workqueue is being used on a memory reclaim path, WQ_MEM_RECLAIM
flag has been set to ensure forward progress under memory pressure.
Since there are fixed number of work items, explicit concurrency
limit is unnecessary here.
Signed-off-by: Bhaktipriya Shridhar <bhaktipriya96@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Don't expose skbs to in-kernel users, such as the AFS filesystem, but
instead provide a notification hook the indicates that a call needs
attention and another that indicates that there's a new call to be
collected.
This makes the following possibilities more achievable:
(1) Call refcounting can be made simpler if skbs don't hold refs to calls.
(2) skbs referring to non-data events will be able to be freed much sooner
rather than being queued for AFS to pick up as rxrpc_kernel_recv_data
will be able to consult the call state.
(3) We can shortcut the receive phase when a call is remotely aborted
because we don't have to go through all the packets to get to the one
cancelling the operation.
(4) It makes it easier to do encryption/decryption directly between AFS's
buffers and sk_buffs.
(5) Encryption/decryption can more easily be done in the AFS's thread
contexts - usually that of the userspace process that issued a syscall
- rather than in one of rxrpc's background threads on a workqueue.
(6) AFS will be able to wait synchronously on a call inside AF_RXRPC.
To make this work, the following interface function has been added:
int rxrpc_kernel_recv_data(
struct socket *sock, struct rxrpc_call *call,
void *buffer, size_t bufsize, size_t *_offset,
bool want_more, u32 *_abort_code);
This is the recvmsg equivalent. It allows the caller to find out about the
state of a specific call and to transfer received data into a buffer
piecemeal.
afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction
logic between them. They don't wait synchronously yet because the socket
lock needs to be dealt with.
Five interface functions have been removed:
rxrpc_kernel_is_data_last()
rxrpc_kernel_get_abort_code()
rxrpc_kernel_get_error_number()
rxrpc_kernel_free_skb()
rxrpc_kernel_data_consumed()
As a temporary hack, sk_buffs going to an in-kernel call are queued on the
rxrpc_call struct (->knlrecv_queue) rather than being handed over to the
in-kernel user. To process the queue internally, a temporary function,
temp_deliver_data() has been added. This will be replaced with common code
between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a
future patch.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pass struct socket * to more rxrpc kernel interface functions. They should
be starting from this rather than the socket pointer in the rxrpc_call
struct if they need to access the socket.
I have left:
rxrpc_kernel_is_data_last()
rxrpc_kernel_get_abort_code()
rxrpc_kernel_get_error_number()
rxrpc_kernel_free_skb()
rxrpc_kernel_data_consumed()
unmodified as they're all about to be removed (and, in any case, don't
touch the socket).
Signed-off-by: David Howells <dhowells@redhat.com>
Provide a function so that kernel users, such as AFS, can ask for the peer
address of a call:
void rxrpc_kernel_get_peer(struct rxrpc_call *call,
struct sockaddr_rxrpc *_srx);
In the future the kernel service won't get sk_buffs to look inside.
Further, this allows us to hide any canonicalisation inside AF_RXRPC for
when IPv6 support is added.
Also propagate this through to afs_find_server() and issue a warning if we
can't handle the address family yet.
Signed-off-by: David Howells <dhowells@redhat.com>
while moving xattrs to expand the extended inode. Also add some
sanity checks to the block group descriptors to make sure we don't end
up overwriting the superblock.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
iQEcBAABCAAGBQJXw7i2AAoJEPL5WVaVDYGj96gH/A8rNgx7BoqPx3kanVEamblT
tM0X9JcEGmKHN4enRts2b78EWbR0/U0SOP92+fg9SSq2MDJ0/kdaKLWmbUwx8jUi
B7HMEqCprlCdigK7wwt3xF+6edyZRhtzlWy3bhxJ40f0KT5CuriSQbxogr931uKl
hUKW2h5JtUqHtINzTt4oWjVm8xwrScxuYHYAcpw0G42ZzfO6xQOzQdowcx4m3cE9
PrtTbU5MwW8/wgsdLiClScQq30MK/GCbHh5heyRt1BcNo9+MDsZDOgdavh9StfnW
Bl1N6zwRtRBJNcpKWfTfwU4NTIvStCTyA8BJgKgE95YIHDsstJVl4MO7ot25qbM=
=pXe+
-----END PGP SIGNATURE-----
Merge tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
Pull ext4 fixes from Ted Ts'o:
"Fix bugs that could cause kernel deadlocks or file system corruption
while moving xattrs to expand the extended inode.
Also add some sanity checks to the block group descriptors to make
sure we don't end up overwriting the superblock"
* tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
ext4: avoid deadlock when expanding inode size
ext4: properly align shifted xattrs when expanding inodes
ext4: fix xattr shifting when expanding inodes part 2
ext4: fix xattr shifting when expanding inodes
ext4: validate that metadata blocks do not overlap superblock
ext4: reserve xattr index for the Hurd
Pull btrfs fixes from Chris Mason:
"We've queued up a few different fixes in here. These range from
enospc corners to fsync and quota fixes, and a few targeted at error
handling for corrupt metadata/fuzzing"
* 'for-linus-4.8' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs:
Btrfs: fix lockdep warning on deadlock against an inode's log mutex
Btrfs: detect corruption when non-root leaf has zero item
Btrfs: check btree node's nritems
btrfs: don't create or leak aliased root while cleaning up orphans
Btrfs: fix em leak in find_first_block_group
btrfs: do not background blkdev_put()
Btrfs: clarify do_chunk_alloc()'s return value
btrfs: fix fsfreeze hang caused by delayed iputs deal
btrfs: update btrfs_space_info's bytes_may_use timely
btrfs: divide btrfs_update_reserved_bytes() into two functions
btrfs: use correct offset for reloc_inode in prealloc_file_extent_cluster()
btrfs: qgroup: Fix qgroup incorrectness caused by log replay
btrfs: relocation: Fix leaking qgroups numbers on data extents
btrfs: qgroup: Refactor btrfs_qgroup_insert_dirty_extent()
btrfs: waiting on qgroup rescan should not always be interruptible
btrfs: properly track when rescan worker is running
btrfs: flush_space: treat return value of do_chunk_alloc properly
Btrfs: add ASSERT for block group's memory leak
btrfs: backref: Fix soft lockup in __merge_refs function
Btrfs: fix memory leak of reloc_root
This fixes a bug introduced by recent debugfs cleanup.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1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=9igI
-----END PGP SIGNATURE-----
Merge tag 'dlm-4.8-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/linux-dlm
Pull dlm fix from David Teigland:
"This fixes a bug introduced by recent debugfs cleanup"
* tag 'dlm-4.8-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/linux-dlm:
dlm: fix malfunction of dlm_tool caused by debugfs changes
Pull block fixes from Jens Axboe:
"Here's a set of block fixes for the current 4.8-rc release. This
contains:
- a fix for a secure erase regression, from Adrian.
- a fix for an mmc use-after-free bug regression, also from Adrian.
- potential zero pointer deference in bdev freezing, from Andrey.
- a race fix for blk_set_queue_dying() from Bart.
- a set of xen blkfront fixes from Bob Liu.
- three small fixes for bcache, from Eric and Kent.
- a fix for a potential invalid NVMe state transition, from Gabriel.
- blk-mq CPU offline fix, preventing us from issuing and completing a
request on the wrong queue. From me.
- revert two previous floppy changes, since they caused a user
visibile regression. A better fix is in the works.
- ensure that we don't send down bios that have more than 256
elements in them. Fixes a crash with bcache, for example. From
Ming.
- a fix for deferencing an error pointer with cgroup writeback.
Fixes a regression. From Vegard"
* 'for-linus' of git://git.kernel.dk/linux-block:
mmc: fix use-after-free of struct request
Revert "floppy: refactor open() flags handling"
Revert "floppy: fix open(O_ACCMODE) for ioctl-only open"
fs/block_dev: fix potential NULL ptr deref in freeze_bdev()
blk-mq: improve warning for running a queue on the wrong CPU
blk-mq: don't overwrite rq->mq_ctx
block: make sure a big bio is split into at most 256 bvecs
nvme: Fix nvme_get/set_features() with a NULL result pointer
bdev: fix NULL pointer dereference
xen-blkfront: free resources if xlvbd_alloc_gendisk fails
xen-blkfront: introduce blkif_set_queue_limits()
xen-blkfront: fix places not updated after introducing 64KB page granularity
bcache: pr_err: more meaningful error message when nr_stripes is invalid
bcache: RESERVE_PRIO is too small by one when prio_buckets() is a power of two.
bcache: register_bcache(): call blkdev_put() when cache_alloc() fails
block: Fix race triggered by blk_set_queue_dying()
block: Fix secure erase
nvme: Prevent controller state invalid transition
seq_read() is a nasty piece of work, not to mention buggy.
It has (I think) an old bug which allows unprivileged userspace to read
beyond the end of m->buf.
I was getting these:
BUG: KASAN: slab-out-of-bounds in seq_read+0xcd2/0x1480 at addr ffff880116889880
Read of size 2713 by task trinity-c2/1329
CPU: 2 PID: 1329 Comm: trinity-c2 Not tainted 4.8.0-rc1+ #96
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.3-0-ge2fc41e-prebuilt.qemu-project.org 04/01/2014
Call Trace:
kasan_object_err+0x1c/0x80
kasan_report_error+0x2cb/0x7e0
kasan_report+0x4e/0x80
check_memory_region+0x13e/0x1a0
kasan_check_read+0x11/0x20
seq_read+0xcd2/0x1480
proc_reg_read+0x10b/0x260
do_loop_readv_writev.part.5+0x140/0x2c0
do_readv_writev+0x589/0x860
vfs_readv+0x7b/0xd0
do_readv+0xd8/0x2c0
SyS_readv+0xb/0x10
do_syscall_64+0x1b3/0x4b0
entry_SYSCALL64_slow_path+0x25/0x25
Object at ffff880116889100, in cache kmalloc-4096 size: 4096
Allocated:
PID = 1329
save_stack_trace+0x26/0x80
save_stack+0x46/0xd0
kasan_kmalloc+0xad/0xe0
__kmalloc+0x1aa/0x4a0
seq_buf_alloc+0x35/0x40
seq_read+0x7d8/0x1480
proc_reg_read+0x10b/0x260
do_loop_readv_writev.part.5+0x140/0x2c0
do_readv_writev+0x589/0x860
vfs_readv+0x7b/0xd0
do_readv+0xd8/0x2c0
SyS_readv+0xb/0x10
do_syscall_64+0x1b3/0x4b0
return_from_SYSCALL_64+0x0/0x6a
Freed:
PID = 0
(stack is not available)
Memory state around the buggy address:
ffff88011688a000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
ffff88011688a080: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>ffff88011688a100: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
^
ffff88011688a180: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
ffff88011688a200: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
==================================================================
Disabling lock debugging due to kernel taint
This seems to be the same thing that Dave Jones was seeing here:
https://lkml.org/lkml/2016/8/12/334
There are multiple issues here:
1) If we enter the function with a non-empty buffer, there is an attempt
to flush it. But it was not clearing m->from after doing so, which
means that if we try to do this flush twice in a row without any call
to traverse() in between, we are going to be reading from the wrong
place -- the splat above, fixed by this patch.
2) If there's a short write to userspace because of page faults, the
buffer may already contain multiple lines (i.e. pos has advanced by
more than 1), but we don't save the progress that was made so the
next call will output what we've already returned previously. Since
that is a much less serious issue (and I have a headache after
staring at seq_read() for the past 8 hours), I'll leave that for now.
Link: http://lkml.kernel.org/r/1471447270-32093-1-git-send-email-vegard.nossum@oracle.com
Signed-off-by: Vegard Nossum <vegard.nossum@oracle.com>
Reported-by: Dave Jones <davej@codemonkey.org.uk>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
With the current kernel, `dlm_tool lockdebug` fails as below:
"dlm_tool lockdebug ED0BD86DCE724393918A1AE8FDBF1EE3
can't open /sys/kernel/debug/dlm/ED0BD86DCE724393918A1AE8FDBF1EE3:
Operation not permitted"
This is because table_open() depends on file->f_op to tell which
seq_file ops should be passed down. But, the original file ops in
file->f_op is replaced by "debugfs_full_proxy_file_operations" with
commit 49d200deaa ("debugfs: prevent access to removed files'
private data").
Currently, I can think up 2 solutions: 1st, replace
debugfs_create_file() with debugfs_create_file_unsafe();
2nd, make different table_open#() accordingly. The 1st one
is neat, but I don't thoroughly understand its risk. Maybe
someone has a better one.
Signed-off-by: Eric Ren <zren@suse.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Calling freeze_bdev() twice on the same block device without mounted
filesystem get_super() will return NULL, which will lead to NULL-ptr
dereference later in drop_super().
Check get_super() result to fix that.
Note, that this is a purely theoretical issue. We have only 3
freeze_bdev() callers. 2 of them are in filesystem code and used on a
device with mounted fs. The third one in lock_fs() has protection in
upper-layer code against freezing block device the second time without
thawing it first.
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
Commit 44f714dae5 ("Btrfs: improve performance on fsync against new
inode after rename/unlink"), which landed in 4.8-rc2, introduced a
possibility for a deadlock due to double locking of an inode's log mutex
by the same task, which lockdep reports with:
[23045.433975] =============================================
[23045.434748] [ INFO: possible recursive locking detected ]
[23045.435426] 4.7.0-rc6-btrfs-next-34+ #1 Not tainted
[23045.436044] ---------------------------------------------
[23045.436044] xfs_io/3688 is trying to acquire lock:
[23045.436044] (&ei->log_mutex){+.+...}, at: [<ffffffffa038552d>] btrfs_log_inode+0x13a/0xc95 [btrfs]
[23045.436044]
but task is already holding lock:
[23045.436044] (&ei->log_mutex){+.+...}, at: [<ffffffffa038552d>] btrfs_log_inode+0x13a/0xc95 [btrfs]
[23045.436044]
other info that might help us debug this:
[23045.436044] Possible unsafe locking scenario:
[23045.436044] CPU0
[23045.436044] ----
[23045.436044] lock(&ei->log_mutex);
[23045.436044] lock(&ei->log_mutex);
[23045.436044]
*** DEADLOCK ***
[23045.436044] May be due to missing lock nesting notation
[23045.436044] 3 locks held by xfs_io/3688:
[23045.436044] #0: (&sb->s_type->i_mutex_key#15){+.+...}, at: [<ffffffffa035f2ae>] btrfs_sync_file+0x14e/0x425 [btrfs]
[23045.436044] #1: (sb_internal#2){.+.+.+}, at: [<ffffffff8118446b>] __sb_start_write+0x5f/0xb0
[23045.436044] #2: (&ei->log_mutex){+.+...}, at: [<ffffffffa038552d>] btrfs_log_inode+0x13a/0xc95 [btrfs]
[23045.436044]
stack backtrace:
[23045.436044] CPU: 4 PID: 3688 Comm: xfs_io Not tainted 4.7.0-rc6-btrfs-next-34+ #1
[23045.436044] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.1-0-gb3ef39f-prebuilt.qemu-project.org 04/01/2014
[23045.436044] 0000000000000000 ffff88022f5f7860 ffffffff8127074d ffffffff82a54b70
[23045.436044] ffffffff82a54b70 ffff88022f5f7920 ffffffff81092897 ffff880228015d68
[23045.436044] 0000000000000000 ffffffff82a54b70 ffffffff829c3f00 ffff880228015d68
[23045.436044] Call Trace:
[23045.436044] [<ffffffff8127074d>] dump_stack+0x67/0x90
[23045.436044] [<ffffffff81092897>] __lock_acquire+0xcbb/0xe4e
[23045.436044] [<ffffffff8109155f>] ? mark_lock+0x24/0x201
[23045.436044] [<ffffffff8109179a>] ? mark_held_locks+0x5e/0x74
[23045.436044] [<ffffffff81092de0>] lock_acquire+0x12f/0x1c3
[23045.436044] [<ffffffff81092de0>] ? lock_acquire+0x12f/0x1c3
[23045.436044] [<ffffffffa038552d>] ? btrfs_log_inode+0x13a/0xc95 [btrfs]
[23045.436044] [<ffffffffa038552d>] ? btrfs_log_inode+0x13a/0xc95 [btrfs]
[23045.436044] [<ffffffff814a51a4>] mutex_lock_nested+0x77/0x3a7
[23045.436044] [<ffffffffa038552d>] ? btrfs_log_inode+0x13a/0xc95 [btrfs]
[23045.436044] [<ffffffffa039705e>] ? btrfs_release_delayed_node+0xb/0xd [btrfs]
[23045.436044] [<ffffffffa038552d>] btrfs_log_inode+0x13a/0xc95 [btrfs]
[23045.436044] [<ffffffffa038552d>] ? btrfs_log_inode+0x13a/0xc95 [btrfs]
[23045.436044] [<ffffffff810a0ed1>] ? vprintk_emit+0x453/0x465
[23045.436044] [<ffffffffa0385a61>] btrfs_log_inode+0x66e/0xc95 [btrfs]
[23045.436044] [<ffffffffa03c084d>] log_new_dir_dentries+0x26c/0x359 [btrfs]
[23045.436044] [<ffffffffa03865aa>] btrfs_log_inode_parent+0x4a6/0x628 [btrfs]
[23045.436044] [<ffffffffa0387552>] btrfs_log_dentry_safe+0x5a/0x75 [btrfs]
[23045.436044] [<ffffffffa035f464>] btrfs_sync_file+0x304/0x425 [btrfs]
[23045.436044] [<ffffffff811acaf4>] vfs_fsync_range+0x8c/0x9e
[23045.436044] [<ffffffff811acb22>] vfs_fsync+0x1c/0x1e
[23045.436044] [<ffffffff811acc79>] do_fsync+0x31/0x4a
[23045.436044] [<ffffffff811ace99>] SyS_fsync+0x10/0x14
[23045.436044] [<ffffffff814a88e5>] entry_SYSCALL_64_fastpath+0x18/0xa8
[23045.436044] [<ffffffff8108f039>] ? trace_hardirqs_off_caller+0x3f/0xaa
An example reproducer for this is:
$ mkfs.btrfs -f /dev/sdb
$ mount /dev/sdb /mnt
$ mkdir /mnt/dir
$ touch /mnt/dir/foo
$ sync
$ mv /mnt/dir/foo /mnt/dir/bar
$ touch /mnt/dir/foo
$ xfs_io -c "fsync" /mnt/dir/bar
This is because while logging the inode of file bar we end up logging its
parent directory (since its inode has an unlink_trans field matching the
current transaction id due to the rename operation), which in turn logs
the inodes for all its new dentries, so that the new inode for the new
file named foo gets logged which in turn triggered another logging attempt
for the inode we are fsync'ing, since that inode had an old name that
corresponds to the name of the new inode.
So fix this by ensuring that when logging the inode for a new dentry that
has a name matching an old name of some other inode, we don't log again
the original inode that we are fsync'ing.
Fixes: 44f714dae5 ("Btrfs: improve performance on fsync against new inode after rename/unlink")
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
Right now we treat leaf which has zero item as a valid one
because we could have an empty tree, that is, a root that is
also a leaf without any item, however, in the same case but
when the leaf is not a root, we can end up with hitting the
BUG_ON(1) in btrfs_extend_item() called by
setup_inline_extent_backref().
This makes us check the situation as a corruption if leaf is
not its own root.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
When btree node (level = 1) has nritems which equals to zero,
we can end up with panic due to insert_ptr()'s
BUG_ON(slot > nritems);
where slot is 1 and nritems is 0, as copy_for_split() calls
insert_ptr(.., path->slots[1] + 1, ...);
A invalid value results in the whole mess, this adds the check
for btree's node nritems so that we stop reading block when
when something is wrong.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
commit 909c3a22da (Btrfs: fix loading of orphan roots leading to BUG_ON)
avoids the BUG_ON but can add an aliased root to the dead_roots list or
leak the root.
Since we've already been loading roots into the radix tree, we should
use it before looking the root up on disk.
Cc: <stable@vger.kernel.org> # 4.5
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
We need to call free_extent_map() on the em we look up.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
At the end of unmount/dev-delete, if the device exclusive open is not
actually closed, then there might be a race with another program in
the userland who is trying to open the device in exclusive mode and
it may fail for eg:
unmount /btrfs; fsck /dev/x
btrfs dev del /dev/x /btrfs; fsck /dev/x
so here background blkdev_put() is not a choice
Signed-off-by: Anand Jain <Anand.Jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
Function start_transaction() can return ERR_PTR(1) when flush is
BTRFS_RESERVE_FLUSH_LIMIT, so the call graph is
start_transaction (return ERR_PTR(1))
-> btrfs_block_rsv_add (return 1)
-> reserve_metadata_bytes (return 1)
-> flush_space (return 1)
-> do_chunk_alloc (return 1)
With BTRFS_RESERVE_FLUSH_LIMIT, if flush_space is already on the
flush_state of ALLOC_CHUNK and it successfully allocates a new
chunk, then instead of trying to reserve space again,
reserve_metadata_bytes returns 1 immediately.
Eventually the callers who call start_transaction() usually just
do the IS_ERR() check which ERR_PTR(1) can pass, then it'll get
a panic when dereferencing a pointer which is ERR_PTR(1).
The following patch fixes the above problem.
"btrfs: flush_space: treat return value of do_chunk_alloc properly"
https://patchwork.kernel.org/patch/7778651/
This add comments to clarify do_chunk_alloc()'s return value.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
When running fstests generic/068, sometimes we got below deadlock:
xfs_io D ffff8800331dbb20 0 6697 6693 0x00000080
ffff8800331dbb20 ffff88007acfc140 ffff880034d895c0 ffff8800331dc000
ffff880032d243e8 fffffffeffffffff ffff880032d24400 0000000000000001
ffff8800331dbb38 ffffffff816a9045 ffff880034d895c0 ffff8800331dbba8
Call Trace:
[<ffffffff816a9045>] schedule+0x35/0x80
[<ffffffff816abab2>] rwsem_down_read_failed+0xf2/0x140
[<ffffffff8118f5e1>] ? __filemap_fdatawrite_range+0xd1/0x100
[<ffffffff8134f978>] call_rwsem_down_read_failed+0x18/0x30
[<ffffffffa06631fc>] ? btrfs_alloc_block_rsv+0x2c/0xb0 [btrfs]
[<ffffffff810d32b5>] percpu_down_read+0x35/0x50
[<ffffffff81217dfc>] __sb_start_write+0x2c/0x40
[<ffffffffa067f5d5>] start_transaction+0x2a5/0x4d0 [btrfs]
[<ffffffffa067f857>] btrfs_join_transaction+0x17/0x20 [btrfs]
[<ffffffffa068ba34>] btrfs_evict_inode+0x3c4/0x5d0 [btrfs]
[<ffffffff81230a1a>] evict+0xba/0x1a0
[<ffffffff812316b6>] iput+0x196/0x200
[<ffffffffa06851d0>] btrfs_run_delayed_iputs+0x70/0xc0 [btrfs]
[<ffffffffa067f1d8>] btrfs_commit_transaction+0x928/0xa80 [btrfs]
[<ffffffffa0646df0>] btrfs_freeze+0x30/0x40 [btrfs]
[<ffffffff81218040>] freeze_super+0xf0/0x190
[<ffffffff81229275>] do_vfs_ioctl+0x4a5/0x5c0
[<ffffffff81003176>] ? do_audit_syscall_entry+0x66/0x70
[<ffffffff810038cf>] ? syscall_trace_enter_phase1+0x11f/0x140
[<ffffffff81229409>] SyS_ioctl+0x79/0x90
[<ffffffff81003c12>] do_syscall_64+0x62/0x110
[<ffffffff816acbe1>] entry_SYSCALL64_slow_path+0x25/0x25
>From this warning, freeze_super() already holds SB_FREEZE_FS, but
btrfs_freeze() will call btrfs_commit_transaction() again, if
btrfs_commit_transaction() finds that it has delayed iputs to handle,
it'll start_transaction(), which will try to get SB_FREEZE_FS lock
again, then deadlock occurs.
The root cause is that in btrfs, sync_filesystem(sb) does not make
sure all metadata is updated. There still maybe some codes adding
delayed iputs, see below sample race window:
CPU1 | CPU2
|-> freeze_super() |
|-> sync_filesystem(sb); |
| |-> cleaner_kthread()
| | |-> btrfs_delete_unused_bgs()
| | |-> btrfs_remove_chunk()
| | |-> btrfs_remove_block_group()
| | |-> btrfs_add_delayed_iput()
| |
|-> sb->s_writers.frozen = SB_FREEZE_FS; |
|-> sb_wait_write(sb, SB_FREEZE_FS); |
| acquire SB_FREEZE_FS lock. |
| |
|-> btrfs_freeze() |
|-> btrfs_commit_transaction() |
|-> btrfs_run_delayed_iputs() |
| will handle delayed iputs, |
| that means start_transaction() |
| will be called, which will try |
| to get SB_FREEZE_FS lock. |
To fix this issue, introduce a "int fs_frozen" to record internally whether
fs has been frozen. If fs has been frozen, we can not handle delayed iputs.
Signed-off-by: Wang Xiaoguang <wangxg.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add comment to btrfs_freeze ]
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
This patch can fix some false ENOSPC errors, below test script can
reproduce one false ENOSPC error:
#!/bin/bash
dd if=/dev/zero of=fs.img bs=$((1024*1024)) count=128
dev=$(losetup --show -f fs.img)
mkfs.btrfs -f -M $dev
mkdir /tmp/mntpoint
mount $dev /tmp/mntpoint
cd /tmp/mntpoint
xfs_io -f -c "falloc 0 $((64*1024*1024))" testfile
Above script will fail for ENOSPC reason, but indeed fs still has free
space to satisfy this request. Please see call graph:
btrfs_fallocate()
|-> btrfs_alloc_data_chunk_ondemand()
| bytes_may_use += 64M
|-> btrfs_prealloc_file_range()
|-> btrfs_reserve_extent()
|-> btrfs_add_reserved_bytes()
| alloc_type is RESERVE_ALLOC_NO_ACCOUNT, so it does not
| change bytes_may_use, and bytes_reserved += 64M. Now
| bytes_may_use + bytes_reserved == 128M, which is greater
| than btrfs_space_info's total_bytes, false enospc occurs.
| Note, the bytes_may_use decrease operation will be done in
| end of btrfs_fallocate(), which is too late.
Here is another simple case for buffered write:
CPU 1 | CPU 2
|
|-> cow_file_range() |-> __btrfs_buffered_write()
|-> btrfs_reserve_extent() | |
| | |
| | |
| ..... | |-> btrfs_check_data_free_space()
| |
| |
|-> extent_clear_unlock_delalloc() |
In CPU 1, btrfs_reserve_extent()->find_free_extent()->
btrfs_add_reserved_bytes() do not decrease bytes_may_use, the decrease
operation will be delayed to be done in extent_clear_unlock_delalloc().
Assume in this case, btrfs_reserve_extent() reserved 128MB data, CPU2's
btrfs_check_data_free_space() tries to reserve 100MB data space.
If
100MB > data_sinfo->total_bytes - data_sinfo->bytes_used -
data_sinfo->bytes_reserved - data_sinfo->bytes_pinned -
data_sinfo->bytes_readonly - data_sinfo->bytes_may_use
btrfs_check_data_free_space() will try to allcate new data chunk or call
btrfs_start_delalloc_roots(), or commit current transaction in order to
reserve some free space, obviously a lot of work. But indeed it's not
necessary as long as decreasing bytes_may_use timely, we still have
free space, decreasing 128M from bytes_may_use.
To fix this issue, this patch chooses to update bytes_may_use for both
data and metadata in btrfs_add_reserved_bytes(). For compress path, real
extent length may not be equal to file content length, so introduce a
ram_bytes argument for btrfs_reserve_extent(), find_free_extent() and
btrfs_add_reserved_bytes(), it's becasue bytes_may_use is increased by
file content length. Then compress path can update bytes_may_use
correctly. Also now we can discard RESERVE_ALLOC_NO_ACCOUNT, RESERVE_ALLOC
and RESERVE_FREE.
As we know, usually EXTENT_DO_ACCOUNTING is used for error path. In
run_delalloc_nocow(), for inode marked as NODATACOW or extent marked as
PREALLOC, we also need to update bytes_may_use, but can not pass
EXTENT_DO_ACCOUNTING, because it also clears metadata reservation, so
here we introduce EXTENT_CLEAR_DATA_RESV flag to indicate btrfs_clear_bit_hook()
to update btrfs_space_info's bytes_may_use.
Meanwhile __btrfs_prealloc_file_range() will call
btrfs_free_reserved_data_space() internally for both sucessful and failed
path, btrfs_prealloc_file_range()'s callers does not need to call
btrfs_free_reserved_data_space() any more.
Signed-off-by: Wang Xiaoguang <wangxg.fnst@cn.fujitsu.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
This patch divides btrfs_update_reserved_bytes() into
btrfs_add_reserved_bytes() and btrfs_free_reserved_bytes(), and
next patch will extend btrfs_add_reserved_bytes（）to fix some
false ENOSPC error, please see later patch for detailed info.
Signed-off-by: Wang Xiaoguang <wangxg.fnst@cn.fujitsu.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
In prealloc_file_extent_cluster(), btrfs_check_data_free_space() uses
wrong file offset for reloc_inode, it uses cluster->start and cluster->end,
which indeed are extent's bytenr. The correct value should be
cluster->[start|end] minus block group's start bytenr.
start bytenr cluster->start
| | extent | extent | ...| extent |
|----------------------------------------------------------------|
| block group reloc_inode |
Signed-off-by: Wang Xiaoguang <wangxg.fnst@cn.fujitsu.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
When doing log replay at mount time(after power loss), qgroup will leak
numbers of replayed data extents.
The cause is almost the same of balance.
So fix it by manually informing qgroup for owner changed extents.
The bug can be detected by btrfs/119 test case.
Cc: Mark Fasheh <mfasheh@suse.de>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Reviewed-and-Tested-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
This patch fixes a REGRESSION introduced in 4.2, caused by the big quota
rework.
When balancing data extents, qgroup will leak all its numbers for
relocated data extents.
The relocation is done in the following steps for data extents:
1) Create data reloc tree and inode
2) Copy all data extents to data reloc tree
And commit transaction
3) Create tree reloc tree(special snapshot) for any related subvolumes
4) Replace file extent in tree reloc tree with new extents in data reloc
tree
And commit transaction
5) Merge tree reloc tree with original fs, by swapping tree blocks
For 1)~4), since tree reloc tree and data reloc tree doesn't count to
qgroup, everything is OK.
But for 5), the swapping of tree blocks will only info qgroup to track
metadata extents.
If metadata extents contain file extents, qgroup number for file extents
will get lost, leading to corrupted qgroup accounting.
The fix is, before commit transaction of step 5), manually info qgroup to
track all file extents in data reloc tree.
Since at commit transaction time, the tree swapping is done, and qgroup
will account these data extents correctly.
Cc: Mark Fasheh <mfasheh@suse.de>
Reported-by: Mark Fasheh <mfasheh@suse.de>
Reported-by: Filipe Manana <fdmanana@gmail.com>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Tested-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
Refactor btrfs_qgroup_insert_dirty_extent() function, to two functions:
1. btrfs_qgroup_insert_dirty_extent_nolock()
Almost the same with original code.
For delayed_ref usage, which has delayed refs locked.
Change the return value type to int, since caller never needs the
pointer, but only needs to know if they need to free the allocated
memory.
2. btrfs_qgroup_insert_dirty_extent()
The more encapsulated version.
Will do the delayed_refs lock, memory allocation, quota enabled check
and other things.
The original design is to keep exported functions to minimal, but since
more btrfs hacks exposed, like replacing path in balance, we need to
record dirty extents manually, so we have to add such functions.
Also, add comment for both functions, to info developers how to keep
qgroup correct when doing hacks.
Cc: Mark Fasheh <mfasheh@suse.de>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Reviewed-and-Tested-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
We wait on qgroup rescan completion in three places: file system
shutdown, the quota disable ioctl, and the rescan wait ioctl. If the
user sends a signal while we're waiting, we continue happily along. This
is expected behavior for the rescan wait ioctl. It's racy in the shutdown
path but mostly works due to other unrelated synchronization points.
In the quota disable path, it Oopses the kernel pretty much immediately.
Cc: <stable@vger.kernel.org> # v4.4+
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
The qgroup_flags field is overloaded such that it reflects the on-disk
status of qgroups and the runtime state. The BTRFS_QGROUP_STATUS_FLAG_RESCAN
flag is used to indicate that a rescan operation is in progress, but if
the file system is unmounted while a rescan is running, the rescan
operation is paused. If the file system is then mounted read-only,
the flag will still be present but the rescan operation will not have
been resumed. When we go to umount, btrfs_qgroup_wait_for_completion
will see the flag and interpret it to mean that the rescan worker is
still running and will wait for a completion that will never come.
This patch uses a separate flag to indicate when the worker is
running. The locking and state surrounding the qgroup rescan worker
needs a lot of attention beyond this patch but this is enough to
avoid a hung umount.
Cc: <stable@vger.kernel.org> # v4.4+
Signed-off-by; Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
do_chunk_alloc returns 1 when it succeeds to allocate a new chunk.
But flush_space will not convert this to 0, and will also return 1.
As a result, reserve_metadata_bytes will think that flush_space failed,
and may potentially return this value "1" to the caller (depends how
reserve_metadata_bytes was called). The caller will also treat this as an error.
For example, btrfs_block_rsv_refill does:
int ret = -ENOSPC;
...
ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
if (!ret) {
block_rsv_add_bytes(block_rsv, num_bytes, 0);
return 0;
}
return ret;
So it will return -ENOSPC.
Signed-off-by: Alex Lyakas <alex@zadarastorage.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
This adds several ASSERT()' s to report memory leak of block group cache.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
When over 1000 file extents refers to one extent, find_parent_nodes()
will be obviously slow, due to the O(n^2)~O(n^3) loops inside
__merge_refs().
The following ftrace shows the cubic growth of execution time:
256 refs
5) + 91.768 us | __add_keyed_refs.isra.12 [btrfs]();
5) 1.447 us | __add_missing_keys.isra.13 [btrfs]();
5) ! 114.544 us | __merge_refs [btrfs]();
5) ! 136.399 us | __merge_refs [btrfs]();
512 refs
6) ! 279.859 us | __add_keyed_refs.isra.12 [btrfs]();
6) 3.164 us | __add_missing_keys.isra.13 [btrfs]();
6) ! 442.498 us | __merge_refs [btrfs]();
6) # 2091.073 us | __merge_refs [btrfs]();
and 1024 refs
7) ! 368.683 us | __add_keyed_refs.isra.12 [btrfs]();
7) 4.810 us | __add_missing_keys.isra.13 [btrfs]();
7) # 2043.428 us | __merge_refs [btrfs]();
7) * 18964.23 us | __merge_refs [btrfs]();
And sort them into the following char:
(Unit: us)
------------------------------------------------------------------------
Trace function | 256 ref | 512 refs | 1024 refs |
------------------------------------------------------------------------
__add_keyed_refs | 91 | 249 | 368 |
__add_missing_keys | 1 | 3 | 4 |
__merge_refs 1st call | 114 | 442 | 2043 |
__merge_refs 2nd call | 136 | 2091 | 18964 |
------------------------------------------------------------------------
We can see the that __add_keyed_refs() grows almost in linear behavior.
And __add_missing_keys() in this case doesn't change much or takes much
time.
While for the 1st __merge_refs() it's square growth
for the 2nd __merge_refs() call it's cubic growth.
It's no doubt that merge_refs() will take a long long time to execute if
the number of refs continues its grows.
So add a cond_resced() into the loop of __merge_refs().
Although this will solve the problem of soft lockup, we need to use the
new rb_tree based structure introduced by Lu Fengqi to really solve the
long execution time.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
When some critical errors occur and FS would be flipped into RO,
if we have an on-going balance, we can end up with a memory leak
of root->reloc_root since btrfs_drop_snapshots() bails out
without freeing reloc_root at the very early start.
However, we're not able to free reloc_root in btrfs_drop_snapshots()
because its caller, merge_reloc_roots(), still needs to access it to
cleanup reloc_root's rbtree.
This makes us free reloc_root when we're going to free fs/file roots.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
UBIFS uses full names to work with xattrs, therefore we have to use
xattr_full_name() to obtain the xattr prefix as string.
Cc: <stable@vger.kernel.org>
Cc: Andreas Gruenbacher <agruenba@redhat.com>
Fixes: 2b88fc21ca ("ubifs: Switch to generic xattr handlers")
Signed-off-by: Richard Weinberger <richard@nod.at>
Reviewed-by: Andreas Gruenbacher <agruenba@redhat.com>
Tested-by: Dongsheng Yang <dongsheng081251@gmail.com>
An assertion in layout_in_gaps() verifies that the gap_lebs pointer is
below the maximum bound. When computing this maximum bound the idx_lebs
count is multiplied by sizeof(int), while C pointers arithmetic does take
into account the size of the pointed elements implicitly already. Remove
the multiplication to fix the assertion.
Fixes: 1e51764a3c ("UBIFS: add new flash file system")
Cc: <stable@vger.kernel.org>
Signed-off-by: Vincent Stehlé <vincent.stehle@intel.com>
Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
Changes in this update
- regression fixes for XFS changes introduce in 4.8-rc1
- buffer IO accounting assert failure
- ENOSPC block accounting reservation issue
- DAX IO path page cache invalidation fix
- rmapbt on-disk block count in agf
- correct classification of rmap block type when updating AGFL.
- iomap support for attribute fork mapping
- regression fixes for iomap infrastructure in 4.8-rc1
- fiemap: honor FIEMAP_FLAG_SYNC
- fiemap: implement FIEMAP_FLAG_XATTR support to fix XFS regression
- make mark_page_accessed and pagefault_disable usage consistent with
other IO paths
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1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=h8VB
-----END PGP SIGNATURE-----
Merge tag 'xfs-iomap-for-linus-4.8-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs
Pull xfs and iomap fixes from Dave Chinner:
"Changes in this update:
Regression fixes for XFS changes introduce in 4.8-rc1:
- buffer IO accounting assert failure
- ENOSPC block accounting reservation issue
- DAX IO path page cache invalidation fix
- rmapbt on-disk block count in agf
- correct classification of rmap block type when updating AGFL.
- iomap support for attribute fork mapping
Regression fixes for iomap infrastructure in 4.8-rc1:
- fiemap: honor FIEMAP_FLAG_SYNC
- fiemap: implement FIEMAP_FLAG_XATTR support to fix XFS regression
- make mark_page_accessed and pagefault_disable usage consistent with
other IO paths"
* tag 'xfs-iomap-for-linus-4.8-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs:
xfs: remove OWN_AG rmap when allocating a block from the AGFL
xfs: (re-)implement FIEMAP_FLAG_XATTR
xfs: simplify xfs_file_iomap_begin
iomap: mark ->iomap_end as optional
iomap: prepare iomap_fiemap for attribute mappings
iomap: fiemap should honor the FIEMAP_FLAG_SYNC flag
iomap: remove superflous pagefault_disable from iomap_write_actor
iomap: remove superflous mark_page_accessed from iomap_write_actor
xfs: store rmapbt block count in the AGF
xfs: don't invalidate whole file on DAX read/write
xfs: fix bogus space reservation in xfs_iomap_write_allocate
xfs: don't assert fail on non-async buffers on ioacct decrement
Thread A Thread B
- inode_lock fileA
- inode_lock fileB
- inode_lock fileA
- inode_lock fileB
We may encounter above potential deadlock during moving file range in
concurrent scenario. This patch fixes the issue by using inode_trylock
instead.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Only if two input files are regular files, we allow copying data in
range of them, otherwise, deny it.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This reverts commit a2ee0a3003.
When testing with generic/032 of xfstest suit, failure message will be
reported as below:
generic/032 8s ... [failed, exit status 1] - output mismatch (see results/generic/032.out.bad)
--- tests/generic/032.out 2015-01-11 16:52:27.643681072 +0800
+++ results/generic/032.out.bad 2016-08-06 13:44:43.861330500 +0800
@@ -1,5 +1,5 @@
QA output created by 032
-100 iterations
-0000000 cdcd cdcd cdcd cdcd cdcd cdcd cdcd cdcd
-*
-0100000
+1: [768..775]: unwritten
+Unwritten extents found!
...
(Run 'diff -u tests/generic/032.out results/generic/032.out.bad' to see the entire diff)
Ran: generic/032
Failures: generic/032
Failed 1 of 1 tests
In write_end(), we should update i_size of inode before unlock page,
otherwise, we will lose newly updated data in following race condition.
Thread A Thread B
- write_end
- unlock page
- writepages
- lock_page
- writepage
if page is out-of-range of file size,
we will skip writting the page.
- update i_size
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
