1============================
2A block layer cache (bcache)
3============================
4
5Say you've got a big slow raid 6, and an ssd or three. Wouldn't it be
6nice if you could use them as cache... Hence bcache.
7
8Wiki and git repositories are at:
9
10  - https://bcache.evilpiepirate.org
11  - http://evilpiepirate.org/git/linux-bcache.git
12  - https://evilpiepirate.org/git/bcache-tools.git
13
14It's designed around the performance characteristics of SSDs - it only allocates
15in erase block sized buckets, and it uses a hybrid btree/log to track cached
16extents (which can be anywhere from a single sector to the bucket size). It's
17designed to avoid random writes at all costs; it fills up an erase block
18sequentially, then issues a discard before reusing it.
19
20Both writethrough and writeback caching are supported. Writeback defaults to
21off, but can be switched on and off arbitrarily at runtime. Bcache goes to
22great lengths to protect your data - it reliably handles unclean shutdown. (It
23doesn't even have a notion of a clean shutdown; bcache simply doesn't return
24writes as completed until they're on stable storage).
25
26Writeback caching can use most of the cache for buffering writes - writing
27dirty data to the backing device is always done sequentially, scanning from the
28start to the end of the index.
29
30Since random IO is what SSDs excel at, there generally won't be much benefit
31to caching large sequential IO. Bcache detects sequential IO and skips it;
32it also keeps a rolling average of the IO sizes per task, and as long as the
33average is above the cutoff it will skip all IO from that task - instead of
34caching the first 512k after every seek. Backups and large file copies should
35thus entirely bypass the cache.
36
37In the event of a data IO error on the flash it will try to recover by reading
38from disk or invalidating cache entries.  For unrecoverable errors (meta data
39or dirty data), caching is automatically disabled; if dirty data was present
40in the cache it first disables writeback caching and waits for all dirty data
41to be flushed.
42
43Getting started:
44You'll need make-bcache from the bcache-tools repository. Both the cache device
45and backing device must be formatted before use::
46
47  make-bcache -B /dev/sdb
48  make-bcache -C /dev/sdc
49
50make-bcache has the ability to format multiple devices at the same time - if
51you format your backing devices and cache device at the same time, you won't
52have to manually attach::
53
54  make-bcache -B /dev/sda /dev/sdb -C /dev/sdc
55
56bcache-tools now ships udev rules, and bcache devices are known to the kernel
57immediately.  Without udev, you can manually register devices like this::
58
59  echo /dev/sdb > /sys/fs/bcache/register
60  echo /dev/sdc > /sys/fs/bcache/register
61
62Registering the backing device makes the bcache device show up in /dev; you can
63now format it and use it as normal. But the first time using a new bcache
64device, it'll be running in passthrough mode until you attach it to a cache.
65If you are thinking about using bcache later, it is recommended to setup all your
66slow devices as bcache backing devices without a cache, and you can choose to add
67a caching device later.
68See 'ATTACHING' section below.
69
70The devices show up as::
71
72  /dev/bcache<N>
73
74As well as (with udev)::
75
76  /dev/bcache/by-uuid/<uuid>
77  /dev/bcache/by-label/<label>
78
79To get started::
80
81  mkfs.ext4 /dev/bcache0
82  mount /dev/bcache0 /mnt
83
84You can control bcache devices through sysfs at /sys/block/bcache<N>/bcache .
85You can also control them through /sys/fs//bcache/<cset-uuid>/ .
86
87Cache devices are managed as sets; multiple caches per set isn't supported yet
88but will allow for mirroring of metadata and dirty data in the future. Your new
89cache set shows up as /sys/fs/bcache/<UUID>
90
91Attaching
92---------
93
94After your cache device and backing device are registered, the backing device
95must be attached to your cache set to enable caching. Attaching a backing
96device to a cache set is done thusly, with the UUID of the cache set in
97/sys/fs/bcache::
98
99  echo <CSET-UUID> > /sys/block/bcache0/bcache/attach
100
101This only has to be done once. The next time you reboot, just reregister all
102your bcache devices. If a backing device has data in a cache somewhere, the
103/dev/bcache<N> device won't be created until the cache shows up - particularly
104important if you have writeback caching turned on.
105
106If you're booting up and your cache device is gone and never coming back, you
107can force run the backing device::
108
109  echo 1 > /sys/block/sdb/bcache/running
110
111(You need to use /sys/block/sdb (or whatever your backing device is called), not
112/sys/block/bcache0, because bcache0 doesn't exist yet. If you're using a
113partition, the bcache directory would be at /sys/block/sdb/sdb2/bcache)
114
115The backing device will still use that cache set if it shows up in the future,
116but all the cached data will be invalidated. If there was dirty data in the
117cache, don't expect the filesystem to be recoverable - you will have massive
118filesystem corruption, though ext4's fsck does work miracles.
119
120Error Handling
121--------------
122
123Bcache tries to transparently handle IO errors to/from the cache device without
124affecting normal operation; if it sees too many errors (the threshold is
125configurable, and defaults to 0) it shuts down the cache device and switches all
126the backing devices to passthrough mode.
127
128 - For reads from the cache, if they error we just retry the read from the
129   backing device.
130
131 - For writethrough writes, if the write to the cache errors we just switch to
132   invalidating the data at that lba in the cache (i.e. the same thing we do for
133   a write that bypasses the cache)
134
135 - For writeback writes, we currently pass that error back up to the
136   filesystem/userspace. This could be improved - we could retry it as a write
137   that skips the cache so we don't have to error the write.
138
139 - When we detach, we first try to flush any dirty data (if we were running in
140   writeback mode). It currently doesn't do anything intelligent if it fails to
141   read some of the dirty data, though.
142
143
144Howto/cookbook
145--------------
146
147A) Starting a bcache with a missing caching device
148
149If registering the backing device doesn't help, it's already there, you just need
150to force it to run without the cache::
151
152	host:~# echo /dev/sdb1 > /sys/fs/bcache/register
153	[  119.844831] bcache: register_bcache() error opening /dev/sdb1: device already registered
154
155Next, you try to register your caching device if it's present. However
156if it's absent, or registration fails for some reason, you can still
157start your bcache without its cache, like so::
158
159	host:/sys/block/sdb/sdb1/bcache# echo 1 > running
160
161Note that this may cause data loss if you were running in writeback mode.
162
163
164B) Bcache does not find its cache::
165
166	host:/sys/block/md5/bcache# echo 0226553a-37cf-41d5-b3ce-8b1e944543a8 > attach
167	[ 1933.455082] bcache: bch_cached_dev_attach() Couldn't find uuid for md5 in set
168	[ 1933.478179] bcache: __cached_dev_store() Can't attach 0226553a-37cf-41d5-b3ce-8b1e944543a8
169	[ 1933.478179] : cache set not found
170
171In this case, the caching device was simply not registered at boot
172or disappeared and came back, and needs to be (re-)registered::
173
174	host:/sys/block/md5/bcache# echo /dev/sdh2 > /sys/fs/bcache/register
175
176
177C) Corrupt bcache crashes the kernel at device registration time:
178
179This should never happen.  If it does happen, then you have found a bug!
180Please report it to the bcache development list: linux-bcache@vger.kernel.org
181
182Be sure to provide as much information that you can including kernel dmesg
183output if available so that we may assist.
184
185
186D) Recovering data without bcache:
187
188If bcache is not available in the kernel, a filesystem on the backing
189device is still available at an 8KiB offset. So either via a loopdev
190of the backing device created with --offset 8K, or any value defined by
191--data-offset when you originally formatted bcache with `make-bcache`.
192
193For example::
194
195	losetup -o 8192 /dev/loop0 /dev/your_bcache_backing_dev
196
197This should present your unmodified backing device data in /dev/loop0
198
199If your cache is in writethrough mode, then you can safely discard the
200cache device without loosing data.
201
202
203E) Wiping a cache device
204
205::
206
207	host:~# wipefs -a /dev/sdh2
208	16 bytes were erased at offset 0x1018 (bcache)
209	they were: c6 85 73 f6 4e 1a 45 ca 82 65 f5 7f 48 ba 6d 81
210
211After you boot back with bcache enabled, you recreate the cache and attach it::
212
213	host:~# make-bcache -C /dev/sdh2
214	UUID:                   7be7e175-8f4c-4f99-94b2-9c904d227045
215	Set UUID:               5bc072a8-ab17-446d-9744-e247949913c1
216	version:                0
217	nbuckets:               106874
218	block_size:             1
219	bucket_size:            1024
220	nr_in_set:              1
221	nr_this_dev:            0
222	first_bucket:           1
223	[  650.511912] bcache: run_cache_set() invalidating existing data
224	[  650.549228] bcache: register_cache() registered cache device sdh2
225
226start backing device with missing cache::
227
228	host:/sys/block/md5/bcache# echo 1 > running
229
230attach new cache::
231
232	host:/sys/block/md5/bcache# echo 5bc072a8-ab17-446d-9744-e247949913c1 > attach
233	[  865.276616] bcache: bch_cached_dev_attach() Caching md5 as bcache0 on set 5bc072a8-ab17-446d-9744-e247949913c1
234
235
236F) Remove or replace a caching device::
237
238	host:/sys/block/sda/sda7/bcache# echo 1 > detach
239	[  695.872542] bcache: cached_dev_detach_finish() Caching disabled for sda7
240
241	host:~# wipefs -a /dev/nvme0n1p4
242	wipefs: error: /dev/nvme0n1p4: probing initialization failed: Device or resource busy
243	Ooops, it's disabled, but not unregistered, so it's still protected
244
245We need to go and unregister it::
246
247	host:/sys/fs/bcache/b7ba27a1-2398-4649-8ae3-0959f57ba128# ls -l cache0
248	lrwxrwxrwx 1 root root 0 Feb 25 18:33 cache0 -> ../../../devices/pci0000:00/0000:00:1d.0/0000:70:00.0/nvme/nvme0/nvme0n1/nvme0n1p4/bcache/
249	host:/sys/fs/bcache/b7ba27a1-2398-4649-8ae3-0959f57ba128# echo 1 > stop
250	kernel: [  917.041908] bcache: cache_set_free() Cache set b7ba27a1-2398-4649-8ae3-0959f57ba128 unregistered
251
252Now we can wipe it::
253
254	host:~# wipefs -a /dev/nvme0n1p4
255	/dev/nvme0n1p4: 16 bytes were erased at offset 0x00001018 (bcache): c6 85 73 f6 4e 1a 45 ca 82 65 f5 7f 48 ba 6d 81
256
257
258G) dm-crypt and bcache
259
260First setup bcache unencrypted and then install dmcrypt on top of
261/dev/bcache<N> This will work faster than if you dmcrypt both the backing
262and caching devices and then install bcache on top. [benchmarks?]
263
264
265H) Stop/free a registered bcache to wipe and/or recreate it
266
267Suppose that you need to free up all bcache references so that you can
268fdisk run and re-register a changed partition table, which won't work
269if there are any active backing or caching devices left on it:
270
2711) Is it present in /dev/bcache* ? (there are times where it won't be)
272
273   If so, it's easy::
274
275	host:/sys/block/bcache0/bcache# echo 1 > stop
276
2772) But if your backing device is gone, this won't work::
278
279	host:/sys/block/bcache0# cd bcache
280	bash: cd: bcache: No such file or directory
281
282   In this case, you may have to unregister the dmcrypt block device that
283   references this bcache to free it up::
284
285	host:~# dmsetup remove oldds1
286	bcache: bcache_device_free() bcache0 stopped
287	bcache: cache_set_free() Cache set 5bc072a8-ab17-446d-9744-e247949913c1 unregistered
288
289   This causes the backing bcache to be removed from /sys/fs/bcache and
290   then it can be reused.  This would be true of any block device stacking
291   where bcache is a lower device.
292
2933) In other cases, you can also look in /sys/fs/bcache/::
294
295	host:/sys/fs/bcache# ls -l */{cache?,bdev?}
296	lrwxrwxrwx 1 root root 0 Mar  5 09:39 0226553a-37cf-41d5-b3ce-8b1e944543a8/bdev1 -> ../../../devices/virtual/block/dm-1/bcache/
297	lrwxrwxrwx 1 root root 0 Mar  5 09:39 0226553a-37cf-41d5-b3ce-8b1e944543a8/cache0 -> ../../../devices/virtual/block/dm-4/bcache/
298	lrwxrwxrwx 1 root root 0 Mar  5 09:39 5bc072a8-ab17-446d-9744-e247949913c1/cache0 -> ../../../devices/pci0000:00/0000:00:01.0/0000:01:00.0/ata10/host9/target9:0:0/9:0:0:0/block/sdl/sdl2/bcache/
299
300   The device names will show which UUID is relevant, cd in that directory
301   and stop the cache::
302
303	host:/sys/fs/bcache/5bc072a8-ab17-446d-9744-e247949913c1# echo 1 > stop
304
305   This will free up bcache references and let you reuse the partition for
306   other purposes.
307
308
309
310Troubleshooting performance
311---------------------------
312
313Bcache has a bunch of config options and tunables. The defaults are intended to
314be reasonable for typical desktop and server workloads, but they're not what you
315want for getting the best possible numbers when benchmarking.
316
317 - Backing device alignment
318
319   The default metadata size in bcache is 8k.  If your backing device is
320   RAID based, then be sure to align this by a multiple of your stride
321   width using `make-bcache --data-offset`. If you intend to expand your
322   disk array in the future, then multiply a series of primes by your
323   raid stripe size to get the disk multiples that you would like.
324
325   For example:  If you have a 64k stripe size, then the following offset
326   would provide alignment for many common RAID5 data spindle counts::
327
328	64k * 2*2*2*3*3*5*7 bytes = 161280k
329
330   That space is wasted, but for only 157.5MB you can grow your RAID 5
331   volume to the following data-spindle counts without re-aligning::
332
333	3,4,5,6,7,8,9,10,12,14,15,18,20,21 ...
334
335 - Bad write performance
336
337   If write performance is not what you expected, you probably wanted to be
338   running in writeback mode, which isn't the default (not due to a lack of
339   maturity, but simply because in writeback mode you'll lose data if something
340   happens to your SSD)::
341
342	# echo writeback > /sys/block/bcache0/bcache/cache_mode
343
344 - Bad performance, or traffic not going to the SSD that you'd expect
345
346   By default, bcache doesn't cache everything. It tries to skip sequential IO -
347   because you really want to be caching the random IO, and if you copy a 10
348   gigabyte file you probably don't want that pushing 10 gigabytes of randomly
349   accessed data out of your cache.
350
351   But if you want to benchmark reads from cache, and you start out with fio
352   writing an 8 gigabyte test file - so you want to disable that::
353
354	# echo 0 > /sys/block/bcache0/bcache/sequential_cutoff
355
356   To set it back to the default (4 mb), do::
357
358	# echo 4M > /sys/block/bcache0/bcache/sequential_cutoff
359
360 - Traffic's still going to the spindle/still getting cache misses
361
362   In the real world, SSDs don't always keep up with disks - particularly with
363   slower SSDs, many disks being cached by one SSD, or mostly sequential IO. So
364   you want to avoid being bottlenecked by the SSD and having it slow everything
365   down.
366
367   To avoid that bcache tracks latency to the cache device, and gradually
368   throttles traffic if the latency exceeds a threshold (it does this by
369   cranking down the sequential bypass).
370
371   You can disable this if you need to by setting the thresholds to 0::
372
373	# echo 0 > /sys/fs/bcache/<cache set>/congested_read_threshold_us
374	# echo 0 > /sys/fs/bcache/<cache set>/congested_write_threshold_us
375
376   The default is 2000 us (2 milliseconds) for reads, and 20000 for writes.
377
378 - Still getting cache misses, of the same data
379
380   One last issue that sometimes trips people up is actually an old bug, due to
381   the way cache coherency is handled for cache misses. If a btree node is full,
382   a cache miss won't be able to insert a key for the new data and the data
383   won't be written to the cache.
384
385   In practice this isn't an issue because as soon as a write comes along it'll
386   cause the btree node to be split, and you need almost no write traffic for
387   this to not show up enough to be noticeable (especially since bcache's btree
388   nodes are huge and index large regions of the device). But when you're
389   benchmarking, if you're trying to warm the cache by reading a bunch of data
390   and there's no other traffic - that can be a problem.
391
392   Solution: warm the cache by doing writes, or use the testing branch (there's
393   a fix for the issue there).
394
395
396Sysfs - backing device
397----------------------
398
399Available at /sys/block/<bdev>/bcache, /sys/block/bcache*/bcache and
400(if attached) /sys/fs/bcache/<cset-uuid>/bdev*
401
402attach
403  Echo the UUID of a cache set to this file to enable caching.
404
405cache_mode
406  Can be one of either writethrough, writeback, writearound or none.
407
408clear_stats
409  Writing to this file resets the running total stats (not the day/hour/5 minute
410  decaying versions).
411
412detach
413  Write to this file to detach from a cache set. If there is dirty data in the
414  cache, it will be flushed first.
415
416dirty_data
417  Amount of dirty data for this backing device in the cache. Continuously
418  updated unlike the cache set's version, but may be slightly off.
419
420label
421  Name of underlying device.
422
423readahead
424  Size of readahead that should be performed.  Defaults to 0.  If set to e.g.
425  1M, it will round cache miss reads up to that size, but without overlapping
426  existing cache entries.
427
428running
429  1 if bcache is running (i.e. whether the /dev/bcache device exists, whether
430  it's in passthrough mode or caching).
431
432sequential_cutoff
433  A sequential IO will bypass the cache once it passes this threshold; the
434  most recent 128 IOs are tracked so sequential IO can be detected even when
435  it isn't all done at once.
436
437sequential_merge
438  If non zero, bcache keeps a list of the last 128 requests submitted to compare
439  against all new requests to determine which new requests are sequential
440  continuations of previous requests for the purpose of determining sequential
441  cutoff. This is necessary if the sequential cutoff value is greater than the
442  maximum acceptable sequential size for any single request.
443
444state
445  The backing device can be in one of four different states:
446
447  no cache: Has never been attached to a cache set.
448
449  clean: Part of a cache set, and there is no cached dirty data.
450
451  dirty: Part of a cache set, and there is cached dirty data.
452
453  inconsistent: The backing device was forcibly run by the user when there was
454  dirty data cached but the cache set was unavailable; whatever data was on the
455  backing device has likely been corrupted.
456
457stop
458  Write to this file to shut down the bcache device and close the backing
459  device.
460
461writeback_delay
462  When dirty data is written to the cache and it previously did not contain
463  any, waits some number of seconds before initiating writeback. Defaults to
464  30.
465
466writeback_percent
467  If nonzero, bcache tries to keep around this percentage of the cache dirty by
468  throttling background writeback and using a PD controller to smoothly adjust
469  the rate.
470
471writeback_rate
472  Rate in sectors per second - if writeback_percent is nonzero, background
473  writeback is throttled to this rate. Continuously adjusted by bcache but may
474  also be set by the user.
475
476writeback_running
477  If off, writeback of dirty data will not take place at all. Dirty data will
478  still be added to the cache until it is mostly full; only meant for
479  benchmarking. Defaults to on.
480
481Sysfs - backing device stats
482~~~~~~~~~~~~~~~~~~~~~~~~~~~~
483
484There are directories with these numbers for a running total, as well as
485versions that decay over the past day, hour and 5 minutes; they're also
486aggregated in the cache set directory as well.
487
488bypassed
489  Amount of IO (both reads and writes) that has bypassed the cache
490
491cache_hits, cache_misses, cache_hit_ratio
492  Hits and misses are counted per individual IO as bcache sees them; a
493  partial hit is counted as a miss.
494
495cache_bypass_hits, cache_bypass_misses
496  Hits and misses for IO that is intended to skip the cache are still counted,
497  but broken out here.
498
499cache_miss_collisions
500  Counts instances where data was going to be inserted into the cache from a
501  cache miss, but raced with a write and data was already present (usually 0
502  since the synchronization for cache misses was rewritten)
503
504cache_readaheads
505  Count of times readahead occurred.
506
507Sysfs - cache set
508~~~~~~~~~~~~~~~~~
509
510Available at /sys/fs/bcache/<cset-uuid>
511
512average_key_size
513  Average data per key in the btree.
514
515bdev<0..n>
516  Symlink to each of the attached backing devices.
517
518block_size
519  Block size of the cache devices.
520
521btree_cache_size
522  Amount of memory currently used by the btree cache
523
524bucket_size
525  Size of buckets
526
527cache<0..n>
528  Symlink to each of the cache devices comprising this cache set.
529
530cache_available_percent
531  Percentage of cache device which doesn't contain dirty data, and could
532  potentially be used for writeback.  This doesn't mean this space isn't used
533  for clean cached data; the unused statistic (in priority_stats) is typically
534  much lower.
535
536clear_stats
537  Clears the statistics associated with this cache
538
539dirty_data
540  Amount of dirty data is in the cache (updated when garbage collection runs).
541
542flash_vol_create
543  Echoing a size to this file (in human readable units, k/M/G) creates a thinly
544  provisioned volume backed by the cache set.
545
546io_error_halflife, io_error_limit
547  These determines how many errors we accept before disabling the cache.
548  Each error is decayed by the half life (in # ios).  If the decaying count
549  reaches io_error_limit dirty data is written out and the cache is disabled.
550
551journal_delay_ms
552  Journal writes will delay for up to this many milliseconds, unless a cache
553  flush happens sooner. Defaults to 100.
554
555root_usage_percent
556  Percentage of the root btree node in use.  If this gets too high the node
557  will split, increasing the tree depth.
558
559stop
560  Write to this file to shut down the cache set - waits until all attached
561  backing devices have been shut down.
562
563tree_depth
564  Depth of the btree (A single node btree has depth 0).
565
566unregister
567  Detaches all backing devices and closes the cache devices; if dirty data is
568  present it will disable writeback caching and wait for it to be flushed.
569
570Sysfs - cache set internal
571~~~~~~~~~~~~~~~~~~~~~~~~~~
572
573This directory also exposes timings for a number of internal operations, with
574separate files for average duration, average frequency, last occurrence and max
575duration: garbage collection, btree read, btree node sorts and btree splits.
576
577active_journal_entries
578  Number of journal entries that are newer than the index.
579
580btree_nodes
581  Total nodes in the btree.
582
583btree_used_percent
584  Average fraction of btree in use.
585
586bset_tree_stats
587  Statistics about the auxiliary search trees
588
589btree_cache_max_chain
590  Longest chain in the btree node cache's hash table
591
592cache_read_races
593  Counts instances where while data was being read from the cache, the bucket
594  was reused and invalidated - i.e. where the pointer was stale after the read
595  completed. When this occurs the data is reread from the backing device.
596
597trigger_gc
598  Writing to this file forces garbage collection to run.
599
600Sysfs - Cache device
601~~~~~~~~~~~~~~~~~~~~
602
603Available at /sys/block/<cdev>/bcache
604
605block_size
606  Minimum granularity of writes - should match hardware sector size.
607
608btree_written
609  Sum of all btree writes, in (kilo/mega/giga) bytes
610
611bucket_size
612  Size of buckets
613
614cache_replacement_policy
615  One of either lru, fifo or random.
616
617discard
618  Boolean; if on a discard/TRIM will be issued to each bucket before it is
619  reused. Defaults to off, since SATA TRIM is an unqueued command (and thus
620  slow).
621
622freelist_percent
623  Size of the freelist as a percentage of nbuckets. Can be written to to
624  increase the number of buckets kept on the freelist, which lets you
625  artificially reduce the size of the cache at runtime. Mostly for testing
626  purposes (i.e. testing how different size caches affect your hit rate), but
627  since buckets are discarded when they move on to the freelist will also make
628  the SSD's garbage collection easier by effectively giving it more reserved
629  space.
630
631io_errors
632  Number of errors that have occurred, decayed by io_error_halflife.
633
634metadata_written
635  Sum of all non data writes (btree writes and all other metadata).
636
637nbuckets
638  Total buckets in this cache
639
640priority_stats
641  Statistics about how recently data in the cache has been accessed.
642  This can reveal your working set size.  Unused is the percentage of
643  the cache that doesn't contain any data.  Metadata is bcache's
644  metadata overhead.  Average is the average priority of cache buckets.
645  Next is a list of quantiles with the priority threshold of each.
646
647written
648  Sum of all data that has been written to the cache; comparison with
649  btree_written gives the amount of write inflation in bcache.
650