This has been reported and discussed for a long time, and this hang occurs in
both 3.15 and 3.16.
Btrfs now migrates to use kernel workqueue, but it introduces this hang problem.
Btrfs has a kind of work queued as an ordered way, which means that its
ordered_func() must be processed in the way of FIFO, so it usually looks like --
normal_work_helper(arg)
work = container_of(arg, struct btrfs_work, normal_work);
work->func() <---- (we name it work X)
for ordered_work in wq->ordered_list
ordered_work->ordered_func()
ordered_work->ordered_free()
The hang is a rare case, first when we find free space, we get an uncached block
group, then we go to read its free space cache inode for free space information,
so it will
file a readahead request
btrfs_readpages()
for page that is not in page cache
__do_readpage()
submit_extent_page()
btrfs_submit_bio_hook()
btrfs_bio_wq_end_io()
submit_bio()
end_workqueue_bio() <--(ret by the 1st endio)
queue a work(named work Y) for the 2nd
also the real endio()
So the hang occurs when work Y's work_struct and work X's work_struct happens
to share the same address.
A bit more explanation,
A,B,C -- struct btrfs_work
arg -- struct work_struct
kthread:
worker_thread()
pick up a work_struct from @worklist
process_one_work(arg)
worker->current_work = arg; <-- arg is A->normal_work
worker->current_func(arg)
normal_work_helper(arg)
A = container_of(arg, struct btrfs_work, normal_work);
A->func()
A->ordered_func()
A->ordered_free() <-- A gets freed
B->ordered_func()
submit_compressed_extents()
find_free_extent()
load_free_space_inode()
... <-- (the above readhead stack)
end_workqueue_bio()
btrfs_queue_work(work C)
B->ordered_free()
As if work A has a high priority in wq->ordered_list and there are more ordered
works queued after it, such as B->ordered_func(), its memory could have been
freed before normal_work_helper() returns, which means that kernel workqueue
code worker_thread() still has worker->current_work pointer to be work
A->normal_work's, ie. arg's address.
Meanwhile, work C is allocated after work A is freed, work C->normal_work
and work A->normal_work are likely to share the same address(I confirmed this
with ftrace output, so I'm not just guessing, it's rare though).
When another kthread picks up work C->normal_work to process, and finds our
kthread is processing it(see find_worker_executing_work()), it'll think
work C as a collision and skip then, which ends up nobody processing work C.
So the situation is that our kthread is waiting forever on work C.
Besides, there're other cases that can lead to deadlock, but the real problem
is that all btrfs workqueue shares one work->func, -- normal_work_helper,
so this makes each workqueue to have its own helper function, but only a
wraper pf normal_work_helper.
With this patch, I no long hit the above hang.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
Steps to reproduce:
# mkfs.btrfs -f /dev/sd[b-f] -m raid5 -d raid5
# mkfs.ext4 /dev/sdc --->corrupt one of btrfs device
# mount /dev/sdb /mnt -o degraded
# btrfs scrub start -BRd /mnt
This is because readahead would skip missing device, this is not true
for RAID5/6, because REQ_GET_READ_MIRRORS return 1 for RAID5/6 block
mapping. If expected data locates in missing device, readahead thread
would not call __readahead_hook() which makes event @rc->elems=0
wait forever.
Fix this problem by checking return value of btrfs_map_block(),we
can only skip missing device safely if there are several mirrors.
Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
Since the "_struct" suffix is mainly used for distinguish the differnt
btrfs_work between the original and the newly created one,
there is no need using the suffix since all btrfs_workers are changed
into btrfs_workqueue.
Also this patch fixed some codes whose code style is changed due to the
too long "_struct" suffix.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Tested-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Replace the fs_info->readahead_workers with the newly created
btrfs_workqueue.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Tested-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Convert all applicable cases of printk and pr_* to the btrfs_* macros.
Fix all uses of the BTRFS prefix.
Signed-off-by: Frank Holton <fholton@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
This fixes the following errors:
fs/btrfs/reada.c: In function ‘btrfs_reada_wait’:
fs/btrfs/reada.c:958:42: error: invalid operands to binary < (have ‘atomic_t’ and ‘int’)
fs/btrfs/reada.c:961:41: error: invalid operands to binary < (have ‘atomic_t’ and ‘int’)
Signed-off-by: Vincent Stehlé <vincent.stehle@laposte.net>
Cc: Chris Mason <chris.mason@fusionio.com>
Cc: linux-btrfs@vger.kernel.org
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Before this commit, btrfs_map_block() was called with REQ_WRITE
in order to retrieve the list of mirrors for a disk block.
This needs to be changed for the device replace procedure since
it makes a difference whether you are asking for read mirrors
or for locations to write to.
GET_READ_MIRRORS is introduced as a new interface to call
btrfs_map_block().
In the current commit, the functionality is not yet changed,
only the interface for GET_READ_MIRRORS is introduced and all
the places that should use this new interface are adapted.
The reason that REQ_WRITE cannot be abused anymore to retrieve
a list of read mirrors is that during a running dev replace
operation all write requests to the live filesystem are
duplicated to also write to the target drive.
Keep in mind that the target disk is only partially a valid
copy of the source disk while the operation is ongoing. All
writes go to the target disk, but not all reads would return
valid data on the target disk. Therefore it is not possible
anymore to abuse a REQ_WRITE interface to find valid mirrors
for a REQ_READ.
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
This commit contains all the essential changes to the core code
of Btrfs for support of the device replace procedure.
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
The device replace procedure makes use of the scrub code. The scrub
code is the most efficient code to read the allocated data of a disk,
i.e. it reads sequentially in order to avoid disk head movements, it
skips unallocated blocks, it uses read ahead mechanisms, and it
contains all the code to detect and repair defects.
This commit adds code to scrub to allow the scrub code to copy read
data to another disk.
One goal is to be able to perform as fast as possible. Therefore the
write requests are collected until huge bios are built, and the
write process is decoupled from the read process with some kind of
flow control, of course, in order to limit the allocated memory.
The best performance on spinning disks could by reached when the
head movements are avoided as much as possible. Therefore a single
worker is used to interface the read process with the write process.
The regular scrub operation works as fast as before, it is not
negatively influenced and actually it is more or less unchanged.
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
This is required for the device replace procedure in a later step.
Two calling functions also had to be changed to have the fs_info
pointer: repair_io_failure() and scrub_setup_recheck_block().
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
All increments and decrements are under the same spinlock - have to be,
since they need to protect the radix_tree it's found in. Just use
int, no need to wank with kref...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Reduce ioprio class of scrub readahead threads to idle priority.
This setting is fixed. This priority has shown the best performance
during all measurements.
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Normally when there are 2 copies of a block, we add both to the
reada extent tree and prefetch only the one that is easier to reach.
This way we can better utilize multiple devices.
In case of DUP this makes no sense as both copies reside on the
same device.
Signed-off-by: Arne Jansen <sensille@gmx.net>
When inserting into the radix tree returns EEXIST, get the existing
entry without giving up the spinlock in between.
There was a race for both the zones trees and the extent tree.
Signed-off-by: Arne Jansen <sensille@gmx.net>
Readahead already has a define for the max number of mirrors. Scrub
needs such a define now, the rest of the code will need something
like this soon. Therefore the define was added to ctree.h and removed
from the readahead code.
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
The reada code from scrub was casting down a u64 to
an unsigned long so it could insert it into a radix tree.
What it really wanted to do was cast down the result of a shift, instead
of casting down the u64. The bug resulted in trying to insert our
reada struct into the wrong place, which caused soft lockups and other
problems.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
This is the implementation for the generic read ahead framework.
To trigger a readahead, btrfs_reada_add must be called. It will start
a read ahead for the given range [start, end) on tree root. The returned
handle can either be used to wait on the readahead to finish
(btrfs_reada_wait), or to send it to the background (btrfs_reada_detach).
The read ahead works as follows:
On btrfs_reada_add, the root of the tree is inserted into a radix_tree.
reada_start_machine will then search for extents to prefetch and trigger
some reads. When a read finishes for a node, all contained node/leaf
pointers that lie in the given range will also be enqueued. The reads will
be triggered in sequential order, thus giving a big win over a naive
enumeration. It will also make use of multi-device layouts. Each disk
will have its on read pointer and all disks will by utilized in parallel.
Also will no two disks read both sides of a mirror simultaneously, as this
would waste seeking capacity. Instead both disks will read different parts
of the filesystem.
Any number of readaheads can be started in parallel. The read order will be
determined globally, i.e. 2 parallel readaheads will normally finish faster
than the 2 started one after another.
Changes v2:
- protect root->node by transaction instead of node_lock
- fix missed branches:
The readahead had a too simple check to determine if a branch from
a node should be checked or not. It now also records the upper bound
of each node to see if the requested RA range lies within.
- use KERN_CONT to debug output, to avoid line breaks
- defer reada_start_machine to worker to avoid deadlock
Changes v3:
- protect root->node by rcu
Changes v5:
- changed EIO-semantics of reada_tree_block_flagged
- remove spin_lock from reada_control and make elems an atomic_t
- remove unused read_total from reada_control
- kill reada_key_cmp, use btrfs_comp_cpu_keys instead
- use kref-style release functions where possible
- return struct reada_control * instead of void * from btrfs_reada_add
Signed-off-by: Arne Jansen <sensille@gmx.net>
