linux-stable.git/drivers/md/bcache/writeback.c, branch v5.18.2

bcache: fixup multiple threads crash

2022-03-06T14:33:45+00:00

When multiple threads to check btree nodes in parallel, the main
thread wait for all threads to stop or CACHE_SET_IO_DISABLE flag:

wait_event_interruptible(check_state->wait,
                         atomic_read(&check_state->started) == 0 ||
                         test_bit(CACHE_SET_IO_DISABLE, &c->flags));

However, the bch_btree_node_read and bch_btree_node_read_done
maybe call bch_cache_set_error, then the CACHE_SET_IO_DISABLE
will be set. If the flag already set, the main thread return
error. At the same time, maybe some threads still running and
read NULL pointer, the kernel will crash.

This patch change the event wait condition, the main thread must
wait for all threads to stop.

Fixes: 8e7102273f597 ("bcache: make bch_btree_check() to be multithreaded")
Signed-off-by: Mingzhe Zou 
Cc: stable@vger.kernel.org # v5.7+
Signed-off-by: Coly Li

bcache: fixup bcache_dev_sectors_dirty_add() multithreaded CPU false sharing

2022-03-06T14:33:37+00:00

When attaching a cached device (a.k.a backing device) to a cache
device, bch_sectors_dirty_init() is called to count dirty sectors
and stripes (see what bcache_dev_sectors_dirty_add() does) on the
cache device.

When bcache_dev_sectors_dirty_add() is called, set_bit(stripe,
d->full_dirty_stripes) or clear_bit(stripe, d->full_dirty_stripes)
operation will always be performed. In full_dirty_stripes, each 1bit
represents stripe_size (8192) sectors (512B), so 1bit=4MB (8192*512),
and each CPU cache line=64B=512bit=2048MB. When 20 threads process
a cached disk with 100G dirty data, a single thread processes about
23M at a time, and 20 threads total 460M. These full_dirty_stripes
bits corresponding to the 460M data is likely to fall in the same CPU
cache line. When one of these threads performs a set_bit or clear_bit
operation, the same CPU cache line of other threads will become invalid
and must read the full_dirty_stripes from the main memory again. Compared
with single thread, the time of a bcache_dev_sectors_dirty_add()
call is increased by about 50 times in our test (100G dirty data,
20 threads, bcache_dev_sectors_dirty_add() is called more than
20 million times).

This patch tries to test_bit before set_bit or clear_bit operation.
Therefore, a lot of force set and clear operations will be avoided,
and most of bcache_dev_sectors_dirty_add() calls will only read CPU
cache line.

Signed-off-by: Mingzhe Zou 
Signed-off-by: Coly Li

block: pass a block_device and opf to bio_init

2022-02-02T14:49:59+00:00

Pass the block_device that we plan to use this bio for and the
operation to bio_init to optimize the assignment.  A NULL block_device
can be passed, both for the passthrough case on a raw request_queue and
to temporarily avoid refactoring some nasty code.

Signed-off-by: Christoph Hellwig 
Reviewed-by: Chaitanya Kulkarni 
Link: https://lore.kernel.org/r/20220124091107.642561-19-hch@lst.de
Signed-off-by: Jens Axboe

bcache: remove bdev_sectors

2021-10-18T20:43:22+00:00

Use the equivalent block layer helper instead.

Signed-off-by: Christoph Hellwig 
Reviewed-by: Kees Cook 
Reviewed-by: Chaitanya Kulkarni 
Acked-by: Coly Li 
Link: https://lore.kernel.org/r/20211018101130.1838532-4-hch@lst.de
Signed-off-by: Jens Axboe

bcache: Use 64-bit arithmetic instead of 32-bit

2021-04-11T14:37:56+00:00

Cast multiple variables to (int64_t) in order to give the compiler
complete information about the proper arithmetic to use. Notice that
these variables are being used in contexts that expect expressions of
type int64_t  (64 bit, signed). And currently, such expressions are
being evaluated using 32-bit arithmetic.

Fixes: d0cf9503e908 ("octeontx2-pf: ethtool fec mode support")
Addresses-Coverity-ID: 1501724 ("Unintentional integer overflow")
Addresses-Coverity-ID: 1501725 ("Unintentional integer overflow")
Addresses-Coverity-ID: 1501726 ("Unintentional integer overflow")
Signed-off-by: Gustavo A. R. Silva 
Signed-off-by: Coly Li 
Link: https://lore.kernel.org/r/20210411134316.80274-7-colyli@suse.de
Signed-off-by: Jens Axboe

bcache: remove PTR_CACHE

2021-04-11T14:37:55+00:00

Remove the PTR_CACHE inline and replace it with a direct dereference
of c->cache.

(Coly Li: fix the typo from PTR_BUCKET to PTR_CACHE in commit log)

Signed-off-by: Christoph Hellwig 
Signed-off-by: Coly Li 
Link: https://lore.kernel.org/r/20210411134316.80274-3-colyli@suse.de
Signed-off-by: Jens Axboe

bcache: consider the fragmentation when update the writeback rate

2021-02-10T15:05:59+00:00

Current way to calculate the writeback rate only considered the
dirty sectors, this usually works fine when the fragmentation
is not high, but it will give us unreasonable small rate when
we are under a situation that very few dirty sectors consumed
a lot dirty buckets. In some case, the dirty bucekts can reached
to CUTOFF_WRITEBACK_SYNC while the dirty data(sectors) not even
reached the writeback_percent, the writeback rate will still
be the minimum value (4k), thus it will cause all the writes to be
stucked in a non-writeback mode because of the slow writeback.

We accelerate the rate in 3 stages with different aggressiveness,
the first stage starts when dirty buckets percent reach above
BCH_WRITEBACK_FRAGMENT_THRESHOLD_LOW (50), the second is
BCH_WRITEBACK_FRAGMENT_THRESHOLD_MID (57), the third is
BCH_WRITEBACK_FRAGMENT_THRESHOLD_HIGH (64). By default
the first stage tries to writeback the amount of dirty data
in one bucket (on average) in (1 / (dirty_buckets_percent - 50)) second,
the second stage tries to writeback the amount of dirty data in one bucket
in (1 / (dirty_buckets_percent - 57)) * 100 millisecond, the third
stage tries to writeback the amount of dirty data in one bucket in
(1 / (dirty_buckets_percent - 64)) millisecond.

the initial rate at each stage can be controlled by 3 configurable
parameters writeback_rate_fp_term_{low|mid|high}, they are by default
1, 10, 1000, the hint of IO throughput that these values are trying
to achieve is described by above paragraph, the reason that
I choose those value as default is based on the testing and the
production data, below is some details:

A. When it comes to the low stage, there is still a bit far from the 70
   threshold, so we only want to give it a little bit push by setting the
   term to 1, it means the initial rate will be 170 if the fragment is 6,
   it is calculated by bucket_size/fragment, this rate is very small,
   but still much reasonable than the minimum 8.
   For a production bcache with unheavy workload, if the cache device
   is bigger than 1 TB, it may take hours to consume 1% buckets,
   so it is very possible to reclaim enough dirty buckets in this stage,
   thus to avoid entering the next stage.

B. If the dirty buckets ratio didn't turn around during the first stage,
   it comes to the mid stage, then it is necessary for mid stage
   to be more aggressive than low stage, so i choose the initial rate
   to be 10 times more than low stage, that means 1700 as the initial
   rate if the fragment is 6. This is some normal rate
   we usually see for a normal workload when writeback happens
   because of writeback_percent.

C. If the dirty buckets ratio didn't turn around during the low and mid
   stages, it comes to the third stage, and it is the last chance that
   we can turn around to avoid the horrible cutoff writeback sync issue,
   then we choose 100 times more aggressive than the mid stage, that
   means 170000 as the initial rate if the fragment is 6. This is also
   inferred from a production bcache, I've got one week's writeback rate
   data from a production bcache which has quite heavy workloads,
   again, the writeback is triggered by the writeback percent,
   the highest rate area is around 100000 to 240000, so I believe this
   kind aggressiveness at this stage is reasonable for production.
   And it should be mostly enough because the hint is trying to reclaim
   1000 bucket per second, and from that heavy production env,
   it is consuming 50 bucket per second on average in one week's data.

Option writeback_consider_fragment is to control whether we want
this feature to be on or off, it's on by default.

Lastly, below is the performance data for all the testing result,
including the data from production env:
https://docs.google.com/document/d/1AmbIEa_2MhB9bqhC3rfga9tp7n9YX9PLn0jSUxscVW0/edit?usp=sharing

Signed-off-by: dongdong tao 
Signed-off-by: Coly Li 
Signed-off-by: Jens Axboe

bcache: fix race between setting bdev state to none and new write request direct to backing

2020-12-07T20:25:16+00:00

There is a race condition in detaching as below:
A. detaching			B. Write request
(1) writing back
(2) write back done, set bdev
    state to clean.
(3) cached_dev_put() and
    schedule_work(&dc->detach);
				(4) write data [0 - 4K] directly
				    into backing and ack to user.
(5) power-failure...

When we restart this bcache device, this bdev is clean but not detached,
and read [0 - 4K], we will get unexpected old data from cache device.

To fix this problem, set the bdev state to none when we writeback done
in detaching, and then if power-failure happened as above, the data in
cache will not be used in next bcache device starting, it's detached, we
will read the correct data from backing derectly.

Signed-off-by: Dongsheng Yang 
Signed-off-by: Coly Li 
Signed-off-by: Jens Axboe

bcache: remove embedded struct cache_sb from struct cache_set

2020-10-02T20:25:30+00:00

Since bcache code was merged into mainline kerrnel, each cache set only
as one single cache in it. The multiple caches framework is here but the
code is far from completed. Considering the multiple copies of cached
data can also be stored on e.g. md raid1 devices, it is unnecessary to
support multiple caches in one cache set indeed.

The previous preparation patches fix the dependencies of explicitly
making a cache set only have single cache. Now we don't have to maintain
an embedded partial super block in struct cache_set, the in-memory super
block can be directly referenced from struct cache.

This patch removes the embedded struct cache_sb from struct cache_set,
and fixes all locations where the superb lock was referenced from this
removed super block by referencing the in-memory super block of struct
cache.

Signed-off-by: Coly Li 
Reviewed-by: Hannes Reinecke 
Signed-off-by: Jens Axboe

bcache: fix overflow in offset_to_stripe()

2020-07-25T13:38:20+00:00

offset_to_stripe() returns the stripe number (in type unsigned int) from
an offset (in type uint64_t) by the following calculation,
	do_div(offset, d->stripe_size);
For large capacity backing device (e.g. 18TB) with small stripe size
(e.g. 4KB), the result is 4831838208 and exceeds UINT_MAX. The actual
returned value which caller receives is 536870912, due to the overflow.

Indeed in bcache_device_init(), bcache_device->nr_stripes is limited in
range [1, INT_MAX]. Therefore all valid stripe numbers in bcache are
in range [0, bcache_dev->nr_stripes - 1].

This patch adds a upper limition check in offset_to_stripe(): the max
valid stripe number should be less than bcache_device->nr_stripes. If
the calculated stripe number from do_div() is equal to or larger than
bcache_device->nr_stripe, -EINVAL will be returned. (Normally nr_stripes
is less than INT_MAX, exceeding upper limitation doesn't mean overflow,
therefore -EOVERFLOW is not used as error code.)

This patch also changes nr_stripes' type of struct bcache_device from
'unsigned int' to 'int', and return value type of offset_to_stripe()
from 'unsigned int' to 'int', to match their exact data ranges.

All locations where bcache_device->nr_stripes and offset_to_stripe() are
referenced also get updated for the above type change.

Reported-and-tested-by: Ken Raeburn 
Signed-off-by: Coly Li 
Cc: stable@vger.kernel.org
Link: https://bugzilla.redhat.com/show_bug.cgi?id=1783075
Signed-off-by: Jens Axboe