Macro nr_to_fifo_front() is only used once in btree_flush_write(),
it is unncessary indeed. This patch removes this macro and does
calculation directly in place.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This reverts commit 1df3877ff6.
In my testing, sometimes even all the cached btree nodes are freed,
creating gc and allocator kernel threads may still fail. Finally it
turns out that kthread_run() may fail if there is pending signal for
current task. And the pending signal is sent from OOM killer which
is triggered by memory consuption in bch_btree_check().
Therefore explicitly shrinking bcache btree node here does not help,
and after the shrinker callback is improved, as well as pending signals
are ignored before creating kernel threads, now such operation is
unncessary anymore.
This patch reverts the commit 1df3877ff6 ("bcache: shrink btree node
cache after bch_btree_check()") because we have better improvement now.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When run a cache set, all the bcache btree node of this cache set will
be checked by bch_btree_check(). If the bcache btree is very large,
iterating all the btree nodes will occupy too much system memory and
the bcache registering process might be selected and killed by system
OOM killer. kthread_run() will fail if current process has pending
signal, therefore the kthread creating in run_cache_set() for gc and
allocator kernel threads are very probably failed for a very large
bcache btree.
Indeed such OOM is safe and the registering process will exit after
the registration done. Therefore this patch flushes pending signals
during the cache set start up, specificly in bch_cache_allocator_start()
and bch_gc_thread_start(), to make sure run_cache_set() won't fail for
large cahced data set.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Now if prio_read() failed during starting a cache set, we can print
out error message in run_cache_set() and handle the failure properly.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Dan Carpenter points out that from commit 2aa8c52938 ("bcache: avoid
unnecessary btree nodes flushing in btree_flush_write()"), there is a
incorrect data type usage which leads to the following static checker
warning:
drivers/md/bcache/journal.c:444 btree_flush_write()
warn: 'ref_nr' unsigned <= 0
drivers/md/bcache/journal.c
422 static void btree_flush_write(struct cache_set *c)
423 {
424 struct btree *b, *t, *btree_nodes[BTREE_FLUSH_NR];
425 unsigned int i, nr, ref_nr;
^^^^^^
426 atomic_t *fifo_front_p, *now_fifo_front_p;
427 size_t mask;
428
429 if (c->journal.btree_flushing)
430 return;
431
432 spin_lock(&c->journal.flush_write_lock);
433 if (c->journal.btree_flushing) {
434 spin_unlock(&c->journal.flush_write_lock);
435 return;
436 }
437 c->journal.btree_flushing = true;
438 spin_unlock(&c->journal.flush_write_lock);
439
440 /* get the oldest journal entry and check its refcount */
441 spin_lock(&c->journal.lock);
442 fifo_front_p = &fifo_front(&c->journal.pin);
443 ref_nr = atomic_read(fifo_front_p);
444 if (ref_nr <= 0) {
^^^^^^^^^^^
Unsigned can't be less than zero.
445 /*
446 * do nothing if no btree node references
447 * the oldest journal entry
448 */
449 spin_unlock(&c->journal.lock);
450 goto out;
451 }
452 spin_unlock(&c->journal.lock);
As the warning information indicates, local varaible ref_nr in unsigned
int type is wrong, which does not matche atomic_read() and the "<= 0"
checking.
This patch fixes the above error by defining local variable ref_nr as
int type.
Fixes: 2aa8c52938 ("bcache: avoid unnecessary btree nodes flushing in btree_flush_write()")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In year 2007 high performance SSD was still expensive, in order to
save more space for real workload or meta data, the readahead I/Os
for non-meta data was bypassed and not cached on SSD.
In now days, SSD price drops a lot and people can find larger size
SSD with more comfortable price. It is unncessary to alway bypass
normal readahead I/Os to save SSD space for now.
This patch adds options for readahead data cache policies via sysfs
file /sys/block/bcache<N>/readahead_cache_policy, the options are,
- "all": cache all readahead data I/Os.
- "meta-only": only cache meta data, and bypass other regular I/Os.
If users want to make bcache continue to only cache readahead request
for metadata and bypass regular data readahead, please set "meta-only"
to this sysfs file. By default, bcache will back to cache all read-
ahead requests now.
Cc: stable@vger.kernel.org
Signed-off-by: Coly Li <colyli@suse.de>
Acked-by: Eric Wheeler <bcache@linux.ewheeler.net>
Cc: Michael Lyle <mlyle@lyle.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In bset.h, macro bset_bkey_last() is defined as,
bkey_idx((struct bkey *) (i)->d, (i)->keys)
Parameter i can be variable type of data structure, the macro always
works once the type of struct i has member 'd' and 'keys'.
bset_bkey_last() is also used in macro csum_set() to calculate the
checksum of a on-disk data structure. When csum_set() is used to
calculate checksum of on-disk bcache super block, the parameter 'i'
data type is struct cache_sb_disk. Inside struct cache_sb_disk (also in
struct cache_sb) the member keys is __u16 type. But bkey_idx() expects
unsigned int (a 32bit width), so there is problem when sending
parameters via stack to call bkey_idx().
Sparse tool from Intel 0day kbuild system reports this incompatible
problem. bkey_idx() is part of user space API, so the simplest fix is
to cast the (i)->keys to unsigned int type in macro bset_bkey_last().
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When shrink btree node cache from c->btree_cache in bch_mca_scan(),
no matter the selected node is reaped or not, it will be rotated from
the head to the tail of c->btree_cache list. But in bcache journal
code, when flushing the btree nodes with oldest journal entry, btree
nodes are iterated and slected from the tail of c->btree_cache list in
btree_flush_write(). The list_rotate_left() in bch_mca_scan() will
make btree_flush_write() iterate more nodes in c->btree_list in reverse
order.
This patch just reaps the selected btree node cache, and not move it
from the head to the tail of c->btree_cache list. Then bch_mca_scan()
will not mess up c->btree_cache list to btree_flush_write().
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In order to skip the most recently freed btree node cahce, currently
in bch_mca_scan() the first 3 caches in c->btree_cache_freeable list
are skipped when shrinking bcache node caches in bch_mca_scan(). The
related code in bch_mca_scan() is,
737 list_for_each_entry_safe(b, t, &c->btree_cache_freeable, list) {
738 if (nr <= 0)
739 goto out;
740
741 if (++i > 3 &&
742 !mca_reap(b, 0, false)) {
lines free cache memory
746 }
747 nr--;
748 }
The problem is, if virtual memory code calls bch_mca_scan() and
the calculated 'nr' is 1 or 2, then in the above loop, nothing will
be shunk. In such case, if slub/slab manager calls bch_mca_scan()
for many times with small scan number, it does not help to shrink
cache memory and just wasts CPU cycles.
This patch just selects btree node caches from tail of the
c->btree_cache_freeable list, then the newly freed host cache can
still be allocated by mca_alloc(), and at least 1 node can be shunk.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The member 'accessed' of struct btree is used in bch_mca_scan() when
shrinking btree node caches. The original idea is, if b->accessed is
set, clean it and look at next btree node cache from c->btree_cache
list, and only shrink the caches whose b->accessed is cleaned. Then
only cold btree node cache will be shrunk.
But when I/O pressure is high, it is very probably that b->accessed
of a btree node cache will be set again in bch_btree_node_get()
before bch_mca_scan() selects it again. Then there is no chance for
bch_mca_scan() to shrink enough memory back to slub or slab system.
This patch removes member accessed from struct btree, then once a
btree node ache is selected, it will be immediately shunk. By this
change, bch_mca_scan() may release btree node cahce more efficiently.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
the commit 91be66e131 ("bcache: performance improvement for
btree_flush_write()") was an effort to flushing btree node with oldest
btree node faster in following methods,
- Only iterate dirty btree nodes in c->btree_cache, avoid scanning a lot
of clean btree nodes.
- Take c->btree_cache as a LRU-like list, aggressively flushing all
dirty nodes from tail of c->btree_cache util the btree node with
oldest journal entry is flushed. This is to reduce the time of holding
c->bucket_lock.
Guoju Fang and Shuang Li reported that they observe unexptected extra
write I/Os on cache device after applying the above patch. Guoju Fang
provideed more detailed diagnose information that the aggressive
btree nodes flushing may cause 10x more btree nodes to flush in his
workload. He points out when system memory is large enough to hold all
btree nodes in memory, c->btree_cache is not a LRU-like list any more.
Then the btree node with oldest journal entry is very probably not-
close to the tail of c->btree_cache list. In such situation much more
dirty btree nodes will be aggressively flushed before the target node
is flushed. When slow SATA SSD is used as cache device, such over-
aggressive flushing behavior will cause performance regression.
After spending a lot of time on debug and diagnose, I find the real
condition is more complicated, aggressive flushing dirty btree nodes
from tail of c->btree_cache list is not a good solution.
- When all btree nodes are cached in memory, c->btree_cache is not
a LRU-like list, the btree nodes with oldest journal entry won't
be close to the tail of the list.
- There can be hundreds dirty btree nodes reference the oldest journal
entry, before flushing all the nodes the oldest journal entry cannot
be reclaimed.
When the above two conditions mixed together, a simply flushing from
tail of c->btree_cache list is really NOT a good idea.
Fortunately there is still chance to make btree_flush_write() work
better. Here is how this patch avoids unnecessary btree nodes flushing,
- Only acquire c->journal.lock when getting oldest journal entry of
fifo c->journal.pin. In rested locations check the journal entries
locklessly, so their values can be changed on other cores
in parallel.
- In loop list_for_each_entry_safe_reverse(), checking latest front
point of fifo c->journal.pin. If it is different from the original
point which we get with locking c->journal.lock, it means the oldest
journal entry is reclaim on other cores. At this moment, all selected
dirty nodes recorded in array btree_nodes[] are all flushed and clean
on other CPU cores, it is unncessary to iterate c->btree_cache any
longer. Just quit the list_for_each_entry_safe_reverse() loop and
the following for-loop will skip all the selected clean nodes.
- Find a proper time to quit the list_for_each_entry_safe_reverse()
loop. Check the refcount value of orignial fifo front point, if the
value is larger than selected node number of btree_nodes[], it means
more matching btree nodes should be scanned. Otherwise it means no
more matching btee nodes in rest of c->btree_cache list, the loop
can be quit. If the original oldest journal entry is reclaimed and
fifo front point is updated, the refcount of original fifo front point
will be 0, then the loop will be quit too.
- Not hold c->bucket_lock too long time. c->bucket_lock is also required
for space allocation for cached data, hold it for too long time will
block regular I/O requests. When iterating list c->btree_cache, even
there are a lot of maching btree nodes, in order to not holding
c->bucket_lock for too long time, only BTREE_FLUSH_NR nodes are
selected and to flush in following for-loop.
With this patch, only btree nodes referencing oldest journal entry
are flushed to cache device, no aggressive flushing for unnecessary
btree node any more. And in order to avoid blocking regluar I/O
requests, each time when btree_flush_write() called, at most only
BTREE_FLUSH_NR btree nodes are selected to flush, even there are more
maching btree nodes in list c->btree_cache.
At last, one more thing to explain: Why it is safe to read front point
of c->journal.pin without holding c->journal.lock inside the
list_for_each_entry_safe_reverse() loop ?
Here is my answer: When reading the front point of fifo c->journal.pin,
we don't need to know the exact value of front point, we just want to
check whether the value is different from the original front point
(which is accurate value because we get it while c->jouranl.lock is
held). For such purpose, it works as expected without holding
c->journal.lock. Even the front point is changed on other CPU core and
not updated to local core, and current iterating btree node has
identical journal entry local as original fetched fifo front point, it
is still safe. Because after holding mutex b->write_lock (with memory
barrier) this btree node can be found as clean and skipped, the loop
will quite latter when iterate on next node of list c->btree_cache.
Fixes: 91be66e131 ("bcache: performance improvement for btree_flush_write()")
Reported-by: Guoju Fang <fangguoju@gmail.com>
Reported-by: Shuang Li <psymon@bonuscloud.io>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
To explain the pages allocated from mempool state->pool can be
swapped in __btree_sort(), because state->pool is a page pool,
which allocates pages by alloc_pages() indeed.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Avoid a pointless dependency on buffer heads in bcache by simply open
coding reading a single page. Also add a SB_OFFSET define for the
byte offset of the superblock instead of using magic numbers.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This allows to properly build the superblock bio including the offset in
the page using the normal bio helpers. This fixes writing the superblock
for page sizes larger than 4k where the sb write bio would need an offset
in the bio_vec.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Returning the properly typed actual data structure insteaf of the
containing struct page will save the callers some work going
forward.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Avoid an extra reference count roundtrip by transferring the sb_page
ownership to the lower level register helpers.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The patch "bcache: rework error unwinding in register_bcache" introduces
a use-after-free regression in register_bcache(). Here are current code,
2510 out_free_path:
2511 kfree(path);
2512 out_module_put:
2513 module_put(THIS_MODULE);
2514 out:
2515 pr_info("error %s: %s", path, err);
2516 return ret;
If some error happens and the above code path is executed, at line 2511
path is released, but referenced at line 2515. Then KASAN reports a use-
after-free error message.
This patch changes line 2515 in the following way to fix the problem,
2515 pr_info("error %s: %s", path?path:"", err);
Signed-off-by: Coly Li <colyli@suse.de>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Patch "bcache: rework error unwinding in register_bcache" from
Christoph Hellwig changes the local variables 'path' and 'err'
in undefined initial state. If the code in register_bcache() jumps
to label 'out:' or 'out_module_put:' by goto, these two variables
might be reference with undefined value by the following line,
out_module_put:
module_put(THIS_MODULE);
out:
pr_info("error %s: %s", path, err);
return ret;
Therefore this patch initializes these two local variables properly
in register_bcache() to avoid such issue.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Split the successful and error return path, and use one goto label for each
resource to unwind. This also fixes some small errors like leaking the
module reference count in the reboot case (which seems entirely harmless)
or printing the wrong warning messages for early failures.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Split out an on-disk version struct cache_sb with the proper endianness
annotations. This fixes a fair chunk of sparse warnings, but there are
some left due to the way the checksum is defined.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Same as cache device, the buffer page needs to be put while
freeing cached_dev. Otherwise a page would be leaked every
time a cached_dev is stopped.
Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Coly says:
"Guoju Fang talked to me today, he told me this change was unnecessary
and I was over-thought.
Then I realize fifo_idx() uses a mask to handle the array index overflow
condition, so the index swap in journal_pin_cmp() won't happen. And yes,
Guoju and Kent are correct.
Since you already applied this patch, can you please to remove this
patch from your for-next branch? This single patch does not break
thing, but it is unecessary at this moment."
This reverts commit c0e0954e90.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
None of the exported bcache symbols are actually used anywhere.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There is no block directory this file needs includes from.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In bch_mca_scan(), the number of shrinking btree node is calculated
by code like this,
unsigned long nr = sc->nr_to_scan;
nr /= c->btree_pages;
nr = min_t(unsigned long, nr, mca_can_free(c));
variable sc->nr_to_scan is number of objects (here is bcache B+tree
nodes' number) to shrink, and pointer variable sc is sent from memory
management code as parametr of a callback.
If sc->nr_to_scan is smaller than c->btree_pages, after the above
calculation, variable 'nr' will be 0 and nothing will be shrunk. It is
frequeently observed that only 1 or 2 is set to sc->nr_to_scan and make
nr to be zero. Then bch_mca_scan() will do nothing more then acquiring
and releasing mutex c->bucket_lock.
This patch checkes whether nr is 0 after the above calculation, if 0
is the result then set 1 to variable 'n'. Then at least bch_mca_scan()
will try to shrink a single B+tree node.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For writeback mode, if there is no regular I/O request for a while,
the writeback rate will be set to the maximum value (1TB/s for now).
This is good for most of the storage workload, but there are still
people don't what the maximum writeback rate in I/O idle time.
This patch adds a sysfs interface file idle_max_writeback_rate to
permit people to disable maximum writeback rate. Then the minimum
writeback rate can be advised by writeback_rate_minimum in the
bcache device's sysfs interface.
Reported-by: Christian Balzer <chibi@gol.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This patch adds code comments in bch_btree_leaf_dirty() to explain
why w->journal should always reference the eldest journal pin of
all the writing bkeys in the btree node. To make the bcache journal
code to be easier to be understood.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
bcache_allocator can call the following:
bch_allocator_thread()
-> bch_prio_write()
-> bch_bucket_alloc()
-> wait on &ca->set->bucket_wait
But the wake up event on bucket_wait is supposed to come from
bch_allocator_thread() itself => deadlock:
[ 1158.490744] INFO: task bcache_allocato:15861 blocked for more than 10 seconds.
[ 1158.495929] Not tainted 5.3.0-050300rc3-generic #201908042232
[ 1158.500653] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 1158.504413] bcache_allocato D 0 15861 2 0x80004000
[ 1158.504419] Call Trace:
[ 1158.504429] __schedule+0x2a8/0x670
[ 1158.504432] schedule+0x2d/0x90
[ 1158.504448] bch_bucket_alloc+0xe5/0x370 [bcache]
[ 1158.504453] ? wait_woken+0x80/0x80
[ 1158.504466] bch_prio_write+0x1dc/0x390 [bcache]
[ 1158.504476] bch_allocator_thread+0x233/0x490 [bcache]
[ 1158.504491] kthread+0x121/0x140
[ 1158.504503] ? invalidate_buckets+0x890/0x890 [bcache]
[ 1158.504506] ? kthread_park+0xb0/0xb0
[ 1158.504510] ret_from_fork+0x35/0x40
Fix by making the call to bch_prio_write() non-blocking, so that
bch_allocator_thread() never waits on itself.
Moreover, make sure to wake up the garbage collector thread when
bch_prio_write() is failing to allocate buckets.
BugLink: https://bugs.launchpad.net/bugs/1784665
BugLink: https://bugs.launchpad.net/bugs/1796292
Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This patch adds simple code comments for bch_keylist_pop() and
bch_keylist_pop_front() in bset.c, to make the code more easier to
be understand.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In request.c:bch_data_insert_keys(), there is code comment for a piece
of dead code. This patch deletes the dead code and its code comment
since they are useless in practice.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Previous code only returns "Not a bcache superblock" for both bcache
super block offset and magic error. This patch addss more accurate error
messages,
- for super block unmatched offset:
"Not a bcache superblock (bad offset)"
- for super block unmatched magic number:
"Not a bcache superblock (bad magic)"
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit cafe563591 ("bcache: A block layer cache") leads to the
following static checker warning:
./drivers/md/bcache/super.c:770 bcache_device_free()
warn: variable dereferenced before check 'd->disk' (see line 766)
drivers/md/bcache/super.c
762 static void bcache_device_free(struct bcache_device *d)
763 {
764 lockdep_assert_held(&bch_register_lock);
765
766 pr_info("%s stopped", d->disk->disk_name);
^^^^^^^^^
Unchecked dereference.
767
768 if (d->c)
769 bcache_device_detach(d);
770 if (d->disk && d->disk->flags & GENHD_FL_UP)
^^^^^^^
Check too late.
771 del_gendisk(d->disk);
772 if (d->disk && d->disk->queue)
773 blk_cleanup_queue(d->disk->queue);
774 if (d->disk) {
775 ida_simple_remove(&bcache_device_idx,
776 first_minor_to_idx(d->disk->first_minor));
777 put_disk(d->disk);
778 }
779
It is not 100% sure that the gendisk struct of bcache device will always
be there, the warning makes sense when there is problem in block core.
This patch tries to remove the static checking warning by checking
d->disk to avoid NULL pointer deferences.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This patch fix a lost wake-up problem caused by the race between
mca_cannibalize_lock and bch_cannibalize_unlock.
Consider two processes, A and B. Process A is executing
mca_cannibalize_lock, while process B takes c->btree_cache_alloc_lock
and is executing bch_cannibalize_unlock. The problem happens that after
process A executes cmpxchg and will execute prepare_to_wait. In this
timeslice process B executes wake_up, but after that process A executes
prepare_to_wait and set the state to TASK_INTERRUPTIBLE. Then process A
goes to sleep but no one will wake up it. This problem may cause bcache
device to dead.
Signed-off-by: Guoju Fang <fangguoju@gmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Fifo structure journal.pin is implemented by a cycle buffer, if the back
index reaches highest location of the cycle buffer, it will be swapped
to 0. Once the swapping happens, it means a smaller fifo index might be
associated to a newer journal entry. So the btree node with oldest
journal entry won't be selected in bch_btree_leaf_dirty() to reference
the dirty B+tree leaf node. This problem may cause bcache journal won't
protect unflushed oldest B+tree dirty leaf node in power failure, and
this B+tree leaf node is possible to beinconsistent after reboot from
power failure.
This patch fixes the fifo index comparing logic in journal_pin_cmp(),
to avoid potential corrupted B+tree leaf node when the back index of
journal pin is swapped.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
-----BEGIN PGP SIGNATURE-----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=Qt+h
-----END PGP SIGNATURE-----
Merge tag 'for-5.4/block-2019-09-16' of git://git.kernel.dk/linux-block
Pull block updates from Jens Axboe:
- Two NVMe pull requests:
- ana log parse fix from Anton
- nvme quirks support for Apple devices from Ben
- fix missing bio completion tracing for multipath stack devices
from Hannes and Mikhail
- IP TOS settings for nvme rdma and tcp transports from Israel
- rq_dma_dir cleanups from Israel
- tracing for Get LBA Status command from Minwoo
- Some nvme-tcp cleanups from Minwoo, Potnuri and Myself
- Some consolidation between the fabrics transports for handling
the CAP register
- reset race with ns scanning fix for fabrics (move fabrics
commands to a dedicated request queue with a different lifetime
from the admin request queue)."
- controller reset and namespace scan races fixes
- nvme discovery log change uevent support
- naming improvements from Keith
- multiple discovery controllers reject fix from James
- some regular cleanups from various people
- Series fixing (and re-fixing) null_blk debug printing and nr_devices
checks (André)
- A few pull requests from Song, with fixes from Andy, Guoqing,
Guilherme, Neil, Nigel, and Yufen.
- REQ_OP_ZONE_RESET_ALL support (Chaitanya)
- Bio merge handling unification (Christoph)
- Pick default elevator correctly for devices with special needs
(Damien)
- Block stats fixes (Hou)
- Timeout and support devices nbd fixes (Mike)
- Series fixing races around elevator switching and device add/remove
(Ming)
- sed-opal cleanups (Revanth)
- Per device weight support for BFQ (Fam)
- Support for blk-iocost, a new model that can properly account cost of
IO workloads. (Tejun)
- blk-cgroup writeback fixes (Tejun)
- paride queue init fixes (zhengbin)
- blk_set_runtime_active() cleanup (Stanley)
- Block segment mapping optimizations (Bart)
- lightnvm fixes (Hans/Minwoo/YueHaibing)
- Various little fixes and cleanups
* tag 'for-5.4/block-2019-09-16' of git://git.kernel.dk/linux-block: (186 commits)
null_blk: format pr_* logs with pr_fmt
null_blk: match the type of parameter nr_devices
null_blk: do not fail the module load with zero devices
block: also check RQF_STATS in blk_mq_need_time_stamp()
block: make rq sector size accessible for block stats
bfq: Fix bfq linkage error
raid5: use bio_end_sector in r5_next_bio
raid5: remove STRIPE_OPS_REQ_PENDING
md: add feature flag MD_FEATURE_RAID0_LAYOUT
md/raid0: avoid RAID0 data corruption due to layout confusion.
raid5: don't set STRIPE_HANDLE to stripe which is in batch list
raid5: don't increment read_errors on EILSEQ return
nvmet: fix a wrong error status returned in error log page
nvme: send discovery log page change events to userspace
nvme: add uevent variables for controller devices
nvme: enable aen regardless of the presence of I/O queues
nvme-fabrics: allow discovery subsystems accept a kato
nvmet: Use PTR_ERR_OR_ZERO() in nvmet_init_discovery()
nvme: Remove redundant assignment of cq vector
nvme: Assign subsys instance from first ctrl
...
The race was when a thread using closure_sync() notices cl->s->done == 1
before the thread calling closure_put() calls wake_up_process(). Then,
it's possible for that thread to return and exit just before
wake_up_process() is called - so we're trying to wake up a process that
no longer exists.
rcu_read_lock() is sufficient to protect against this, as there's an rcu
barrier somewhere in the process teardown path.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Acked-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The copy_to_user() function returns the number of bytes remaining to be
copied, but the intention here was to return -EFAULT if the copy fails.
Fixes: cafe563591 ("bcache: A block layer cache")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Read /sys/fs/bcache/<uuid>/cacheN/priority_stats can take very long
time with huge cache after long run.
Signed-off-by: Shile Zhang <shile.zhang@linux.alibaba.com>
Tested-by: Heitor Alves de Siqueira <halves@canonical.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
memory malloced in bch_cached_dev_run() and should be freed before
leaving from the error handling cases, otherwise it will cause
memory leak.
Fixes: 0b13efecf5 ("bcache: add return value check to bch_cached_dev_run()")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Now we have counters for how many times jouranl is reclaimed, how many
times cached dirty btree nodes are flushed, but we don't know how many
jouranl buckets are really reclaimed.
This patch adds reclaimed_journal_buckets into struct cache_set, this
is an increasing only counter, to tell how many journal buckets are
reclaimed since cache set runs. From all these three counters (reclaim,
reclaimed_journal_buckets, flush_write), we can have idea how well
current journal space reclaim code works.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This patch improves performance for btree_flush_write() in following
ways,
- Use another spinlock journal.flush_write_lock to replace the very
hot journal.lock. We don't have to use journal.lock here, selecting
candidate btree nodes takes a lot of time, hold journal.lock here will
block other jouranling threads and drop the overall I/O performance.
- Only select flushing btree node from c->btree_cache list. When the
machine has a large system memory, mca cache may have a huge number of
cached btree nodes. Iterating all the cached nodes will take a lot
of CPU time, and most of the nodes on c->btree_cache_freeable and
c->btree_cache_freed lists are cleared and have need to flush. So only
travel mca list c->btree_cache to select flushing btree node should be
enough for most of the cases.
- Don't iterate whole c->btree_cache list, only reversely select first
BTREE_FLUSH_NR btree nodes to flush. Iterate all btree nodes from
c->btree_cache and select the oldest journal pin btree nodes consumes
huge number of CPU cycles if the list is huge (push and pop a node
into/out of a heap is expensive). The last several dirty btree nodes
on the tail of c->btree_cache list are earlest allocated and cached
btree nodes, they are relative to the oldest journal pin btree nodes.
Therefore only flushing BTREE_FLUSH_NR btree nodes from tail of
c->btree_cache probably includes the oldest journal pin btree nodes.
In my testing, the above change decreases 50%+ CPU consumption when
journal space is full. Some times IOPS drops to 0 for 5-8 seconds,
comparing blocking I/O for 120+ seconds in previous code, this is much
better. Maybe there is room to improve in future, but at this momment
the fix looks fine and performs well in my testing.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There is a race between mca_reap(), btree_node_free() and journal code
btree_flush_write(), which results very rare and strange deadlock or
panic and are very hard to reproduce.
Let me explain how the race happens. In btree_flush_write() one btree
node with oldest journal pin is selected, then it is flushed to cache
device, the select-and-flush is a two steps operation. Between these two
steps, there are something may happen inside the race window,
- The selected btree node was reaped by mca_reap() and allocated to
other requesters for other btree node.
- The slected btree node was selected, flushed and released by mca
shrink callback bch_mca_scan().
When btree_flush_write() tries to flush the selected btree node, firstly
b->write_lock is held by mutex_lock(). If the race happens and the
memory of selected btree node is allocated to other btree node, if that
btree node's write_lock is held already, a deadlock very probably
happens here. A worse case is the memory of the selected btree node is
released, then all references to this btree node (e.g. b->write_lock)
will trigger NULL pointer deference panic.
This race was introduced in commit cafe563591 ("bcache: A block layer
cache"), and enlarged by commit c4dc2497d5 ("bcache: fix high CPU
occupancy during journal"), which selected 128 btree nodes and flushed
them one-by-one in a quite long time period.
Such race is not easy to reproduce before. On a Lenovo SR650 server with
48 Xeon cores, and configure 1 NVMe SSD as cache device, a MD raid0
device assembled by 3 NVMe SSDs as backing device, this race can be
observed around every 10,000 times btree_flush_write() gets called. Both
deadlock and kernel panic all happened as aftermath of the race.
The idea of the fix is to add a btree flag BTREE_NODE_journal_flush. It
is set when selecting btree nodes, and cleared after btree nodes
flushed. Then when mca_reap() selects a btree node with this bit set,
this btree node will be skipped. Since mca_reap() only reaps btree node
without BTREE_NODE_journal_flush flag, such race is avoided.
Once corner case should be noticed, that is btree_node_free(). It might
be called in some error handling code path. For example the following
code piece from btree_split(),
2149 err_free2:
2150 bkey_put(b->c, &n2->key);
2151 btree_node_free(n2);
2152 rw_unlock(true, n2);
2153 err_free1:
2154 bkey_put(b->c, &n1->key);
2155 btree_node_free(n1);
2156 rw_unlock(true, n1);
At line 2151 and 2155, the btree node n2 and n1 are released without
mac_reap(), so BTREE_NODE_journal_flush also needs to be checked here.
If btree_node_free() is called directly in such error handling path,
and the selected btree node has BTREE_NODE_journal_flush bit set, just
delay for 1 us and retry again. In this case this btree node won't
be skipped, just retry until the BTREE_NODE_journal_flush bit cleared,
and free the btree node memory.
Fixes: cafe563591 ("bcache: A block layer cache")
Signed-off-by: Coly Li <colyli@suse.de>
Reported-and-tested-by: kbuild test robot <lkp@intel.com>
Cc: stable@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In struct cache_set, retry_flush_write is added for commit c4dc2497d5
("bcache: fix high CPU occupancy during journal") which is reverted in
previous patch.
Now it is useless anymore, and this patch removes it from bcache code.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When accessing or modifying BTREE_NODE_dirty bit, it is not always
necessary to acquire b->write_lock. In bch_btree_cache_free() and
mca_reap() acquiring b->write_lock is necessary, and this patch adds
comments to explain why mutex_lock(&b->write_lock) is necessary for
checking or clearing BTREE_NODE_dirty bit there.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In bch_btree_cache_free() and btree_node_free(), BTREE_NODE_dirty is
always set no matter btree node is dirty or not. The code looks like
this,
if (btree_node_dirty(b))
btree_complete_write(b, btree_current_write(b));
clear_bit(BTREE_NODE_dirty, &b->flags);
Indeed if btree_node_dirty(b) returns false, it means BTREE_NODE_dirty
bit is cleared, then it is unnecessary to clear the bit again.
This patch only clears BTREE_NODE_dirty when btree_node_dirty(b) is
true (the bit is set), to save a few CPU cycles.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This reverts commit c4dc2497d5.
This patch enlarges a race between normal btree flush code path and
flush_btree_write(), which causes deadlock when journal space is
exhausted. Reverts this patch makes the race window from 128 btree
nodes to only 1 btree nodes.
Fixes: c4dc2497d5 ("bcache: fix high CPU occupancy during journal")
Signed-off-by: Coly Li <colyli@suse.de>
Cc: stable@vger.kernel.org
Cc: Tang Junhui <tang.junhui.linux@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This reverts commit 6268dc2c47.
This patch depends on commit c4dc2497d5 ("bcache: fix high CPU
occupancy during journal") which is reverted in previous patch. So
revert this one too.
Fixes: 6268dc2c47 ("bcache: free heap cache_set->flush_btree in bch_journal_free")
Signed-off-by: Coly Li <colyli@suse.de>
Cc: stable@vger.kernel.org
Cc: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When cache set starts, bch_btree_check() will check all bkeys on cache
device by calculating the checksum. This operation will consume a huge
number of system memory if there are a lot of data cached. Since bcache
uses its own mca cache to maintain all its read-in btree nodes, and only
releases the cache space when system memory manage code starts to shrink
caches. Then before memory manager code to call the mca cache shrinker
callback, bcache mca cache will compete memory resource with user space
application, which may have nagive effect to performance of user space
workloads (e.g. data base, or I/O service of distributed storage node).
This patch tries to call bcache mca shrinker routine to proactively
release mca cache memory, to decrease the memory pressure of system and
avoid negative effort of the overall system I/O performance.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>