1 // SPDX-License-Identifier: GPL-2.0
3 * bcache sysfs interfaces
5 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6 * Copyright 2012 Google, Inc.
9 #ifndef NO_BCACHEFS_SYSFS
12 #include "alloc_background.h"
13 #include "alloc_foreground.h"
15 #include "btree_cache.h"
17 #include "btree_iter.h"
18 #include "btree_key_cache.h"
19 #include "btree_update.h"
20 #include "btree_update_interior.h"
24 #include "disk_groups.h"
31 #include "rebalance.h"
36 #include <linux/blkdev.h>
37 #include <linux/sort.h>
38 #include <linux/sched/clock.h>
42 #define SYSFS_OPS(type) \
43 struct sysfs_ops type ## _sysfs_ops = { \
44 .show = type ## _show, \
45 .store = type ## _store \
49 static ssize_t fn ## _to_text(struct printbuf *, \
50 struct kobject *, struct attribute *);\
52 static ssize_t fn ## _show(struct kobject *kobj, struct attribute *attr,\
55 struct printbuf out = PRINTBUF; \
56 ssize_t ret = fn ## _to_text(&out, kobj, attr); \
58 if (!ret && out.allocation_failure) \
62 ret = min_t(size_t, out.pos, PAGE_SIZE - 1); \
63 memcpy(buf, out.buf, ret); \
65 printbuf_exit(&out); \
69 static ssize_t fn ## _to_text(struct printbuf *out, struct kobject *kobj,\
70 struct attribute *attr)
73 static ssize_t fn ## _store(struct kobject *kobj, struct attribute *attr,\
74 const char *buf, size_t size) \
76 #define __sysfs_attribute(_name, _mode) \
77 static struct attribute sysfs_##_name = \
78 { .name = #_name, .mode = _mode }
80 #define write_attribute(n) __sysfs_attribute(n, S_IWUSR)
81 #define read_attribute(n) __sysfs_attribute(n, S_IRUGO)
82 #define rw_attribute(n) __sysfs_attribute(n, S_IRUGO|S_IWUSR)
84 #define sysfs_printf(file, fmt, ...) \
86 if (attr == &sysfs_ ## file) \
87 pr_buf(out, fmt "\n", __VA_ARGS__); \
90 #define sysfs_print(file, var) \
92 if (attr == &sysfs_ ## file) \
96 #define sysfs_hprint(file, val) \
98 if (attr == &sysfs_ ## file) \
99 bch2_hprint(out, val); \
102 #define var_printf(_var, fmt) sysfs_printf(_var, fmt, var(_var))
103 #define var_print(_var) sysfs_print(_var, var(_var))
104 #define var_hprint(_var) sysfs_hprint(_var, var(_var))
106 #define sysfs_strtoul(file, var) \
108 if (attr == &sysfs_ ## file) \
109 return strtoul_safe(buf, var) ?: (ssize_t) size; \
112 #define sysfs_strtoul_clamp(file, var, min, max) \
114 if (attr == &sysfs_ ## file) \
115 return strtoul_safe_clamp(buf, var, min, max) \
119 #define strtoul_or_return(cp) \
122 int _r = kstrtoul(cp, 10, &_v); \
128 #define strtoul_restrict_or_return(cp, min, max) \
130 unsigned long __v = 0; \
131 int _r = strtoul_safe_restrict(cp, __v, min, max); \
137 #define strtoi_h_or_return(cp) \
140 int _r = strtoi_h(cp, &_v); \
146 #define sysfs_hatoi(file, var) \
148 if (attr == &sysfs_ ## file) \
149 return strtoi_h(buf, &var) ?: (ssize_t) size; \
152 write_attribute(trigger_gc);
153 write_attribute(prune_cache);
154 rw_attribute(btree_gc_periodic);
155 rw_attribute(gc_gens_pos);
157 read_attribute(uuid);
158 read_attribute(minor);
159 read_attribute(bucket_size);
160 read_attribute(first_bucket);
161 read_attribute(nbuckets);
162 read_attribute(durability);
163 read_attribute(iodone);
165 read_attribute(io_latency_read);
166 read_attribute(io_latency_write);
167 read_attribute(io_latency_stats_read);
168 read_attribute(io_latency_stats_write);
169 read_attribute(congested);
171 read_attribute(btree_avg_write_size);
173 read_attribute(reserve_stats);
174 read_attribute(btree_cache_size);
175 read_attribute(compression_stats);
176 read_attribute(journal_debug);
177 read_attribute(btree_updates);
178 read_attribute(btree_cache);
179 read_attribute(btree_key_cache);
180 read_attribute(btree_transactions);
181 read_attribute(stripes_heap);
182 read_attribute(open_buckets);
184 read_attribute(internal_uuid);
186 read_attribute(has_data);
187 read_attribute(alloc_debug);
188 write_attribute(wake_allocator);
190 read_attribute(read_realloc_races);
191 read_attribute(extent_migrate_done);
192 read_attribute(extent_migrate_raced);
194 rw_attribute(discard);
197 rw_attribute(copy_gc_enabled);
198 read_attribute(copy_gc_wait);
200 rw_attribute(rebalance_enabled);
201 sysfs_pd_controller_attribute(rebalance);
202 read_attribute(rebalance_work);
203 rw_attribute(promote_whole_extents);
205 read_attribute(new_stripes);
207 read_attribute(io_timers_read);
208 read_attribute(io_timers_write);
210 read_attribute(data_jobs);
212 #ifdef CONFIG_BCACHEFS_TESTS
213 write_attribute(perf_test);
214 #endif /* CONFIG_BCACHEFS_TESTS */
217 static struct attribute sysfs_time_stat_##_name = \
218 { .name = #_name, .mode = S_IRUGO };
222 static struct attribute sysfs_state_rw = {
227 static size_t bch2_btree_cache_size(struct bch_fs *c)
232 mutex_lock(&c->btree_cache.lock);
233 list_for_each_entry(b, &c->btree_cache.live, list)
234 ret += btree_bytes(c);
236 mutex_unlock(&c->btree_cache.lock);
240 static size_t bch2_btree_avg_write_size(struct bch_fs *c)
242 u64 nr = atomic64_read(&c->btree_writes_nr);
243 u64 sectors = atomic64_read(&c->btree_writes_sectors);
245 return nr ? div64_u64(sectors, nr) : 0;
248 static long data_progress_to_text(struct printbuf *out, struct bch_fs *c)
251 struct bch_move_stats *stats;
253 mutex_lock(&c->data_progress_lock);
254 list_for_each_entry(stats, &c->data_progress_list, list) {
255 pr_buf(out, "%s: data type %s btree_id %s position: ",
257 bch2_data_types[stats->data_type],
258 bch2_btree_ids[stats->btree_id]);
259 bch2_bpos_to_text(out, stats->pos);
260 pr_buf(out, "%s", "\n");
263 mutex_unlock(&c->data_progress_lock);
267 static int bch2_compression_stats_to_text(struct printbuf *out, struct bch_fs *c)
269 struct btree_trans trans;
270 struct btree_iter iter;
273 u64 nr_uncompressed_extents = 0,
274 nr_compressed_extents = 0,
275 nr_incompressible_extents = 0,
276 uncompressed_sectors = 0,
277 incompressible_sectors = 0,
278 compressed_sectors_compressed = 0,
279 compressed_sectors_uncompressed = 0;
282 if (!test_bit(BCH_FS_STARTED, &c->flags))
285 bch2_trans_init(&trans, c, 0, 0);
287 for (id = 0; id < BTREE_ID_NR; id++) {
288 if (!((1U << id) & BTREE_ID_HAS_PTRS))
291 for_each_btree_key(&trans, iter, id, POS_MIN,
292 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
293 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
294 const union bch_extent_entry *entry;
295 struct extent_ptr_decoded p;
296 bool compressed = false, uncompressed = false, incompressible = false;
298 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
299 switch (p.crc.compression_type) {
300 case BCH_COMPRESSION_TYPE_none:
302 uncompressed_sectors += k.k->size;
304 case BCH_COMPRESSION_TYPE_incompressible:
305 incompressible = true;
306 incompressible_sectors += k.k->size;
309 compressed_sectors_compressed +=
310 p.crc.compressed_size;
311 compressed_sectors_uncompressed +=
312 p.crc.uncompressed_size;
319 nr_incompressible_extents++;
320 else if (uncompressed)
321 nr_uncompressed_extents++;
323 nr_compressed_extents++;
325 bch2_trans_iter_exit(&trans, &iter);
328 bch2_trans_exit(&trans);
333 pr_buf(out, "uncompressed:\n");
334 pr_buf(out, " nr extents: %llu\n", nr_uncompressed_extents);
335 pr_buf(out, " size: ");
336 bch2_hprint(out, uncompressed_sectors << 9);
339 pr_buf(out, "compressed:\n");
340 pr_buf(out, " nr extents: %llu\n", nr_compressed_extents);
341 pr_buf(out, " compressed size: ");
342 bch2_hprint(out, compressed_sectors_compressed << 9);
344 pr_buf(out, " uncompressed size: ");
345 bch2_hprint(out, compressed_sectors_uncompressed << 9);
348 pr_buf(out, "incompressible:\n");
349 pr_buf(out, " nr extents: %llu\n", nr_incompressible_extents);
350 pr_buf(out, " size: ");
351 bch2_hprint(out, incompressible_sectors << 9);
356 static void bch2_gc_gens_pos_to_text(struct printbuf *out, struct bch_fs *c)
358 pr_buf(out, "%s: ", bch2_btree_ids[c->gc_gens_btree]);
359 bch2_bpos_to_text(out, c->gc_gens_pos);
365 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
367 sysfs_print(minor, c->minor);
368 sysfs_printf(internal_uuid, "%pU", c->sb.uuid.b);
370 sysfs_hprint(btree_cache_size, bch2_btree_cache_size(c));
371 sysfs_hprint(btree_avg_write_size, bch2_btree_avg_write_size(c));
373 sysfs_print(read_realloc_races,
374 atomic_long_read(&c->read_realloc_races));
375 sysfs_print(extent_migrate_done,
376 atomic_long_read(&c->extent_migrate_done));
377 sysfs_print(extent_migrate_raced,
378 atomic_long_read(&c->extent_migrate_raced));
380 sysfs_printf(btree_gc_periodic, "%u", (int) c->btree_gc_periodic);
382 if (attr == &sysfs_gc_gens_pos)
383 bch2_gc_gens_pos_to_text(out, c);
385 sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
387 sysfs_printf(rebalance_enabled, "%i", c->rebalance.enabled);
388 sysfs_pd_controller_show(rebalance, &c->rebalance.pd); /* XXX */
389 sysfs_hprint(copy_gc_wait,
390 max(0LL, c->copygc_wait -
391 atomic64_read(&c->io_clock[WRITE].now)) << 9);
393 if (attr == &sysfs_rebalance_work)
394 bch2_rebalance_work_to_text(out, c);
396 sysfs_print(promote_whole_extents, c->promote_whole_extents);
400 if (attr == &sysfs_journal_debug)
401 bch2_journal_debug_to_text(out, &c->journal);
403 if (attr == &sysfs_btree_updates)
404 bch2_btree_updates_to_text(out, c);
406 if (attr == &sysfs_btree_cache)
407 bch2_btree_cache_to_text(out, c);
409 if (attr == &sysfs_btree_key_cache)
410 bch2_btree_key_cache_to_text(out, &c->btree_key_cache);
412 if (attr == &sysfs_btree_transactions)
413 bch2_btree_trans_to_text(out, c);
415 if (attr == &sysfs_stripes_heap)
416 bch2_stripes_heap_to_text(out, c);
418 if (attr == &sysfs_open_buckets)
419 bch2_open_buckets_to_text(out, c);
421 if (attr == &sysfs_compression_stats)
422 bch2_compression_stats_to_text(out, c);
424 if (attr == &sysfs_new_stripes)
425 bch2_new_stripes_to_text(out, c);
427 if (attr == &sysfs_io_timers_read)
428 bch2_io_timers_to_text(out, &c->io_clock[READ]);
430 if (attr == &sysfs_io_timers_write)
431 bch2_io_timers_to_text(out, &c->io_clock[WRITE]);
433 if (attr == &sysfs_data_jobs)
434 data_progress_to_text(out, c);
441 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
443 if (attr == &sysfs_btree_gc_periodic) {
444 ssize_t ret = strtoul_safe(buf, c->btree_gc_periodic)
447 wake_up_process(c->gc_thread);
451 if (attr == &sysfs_copy_gc_enabled) {
452 ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
455 if (c->copygc_thread)
456 wake_up_process(c->copygc_thread);
460 if (attr == &sysfs_rebalance_enabled) {
461 ssize_t ret = strtoul_safe(buf, c->rebalance.enabled)
468 sysfs_pd_controller_store(rebalance, &c->rebalance.pd);
470 sysfs_strtoul(promote_whole_extents, c->promote_whole_extents);
474 if (!test_bit(BCH_FS_STARTED, &c->flags))
479 if (!test_bit(BCH_FS_RW, &c->flags))
482 if (attr == &sysfs_prune_cache) {
483 struct shrink_control sc;
485 sc.gfp_mask = GFP_KERNEL;
486 sc.nr_to_scan = strtoul_or_return(buf);
487 c->btree_cache.shrink.scan_objects(&c->btree_cache.shrink, &sc);
490 if (attr == &sysfs_trigger_gc) {
492 * Full gc is currently incompatible with btree key cache:
495 down_read(&c->state_lock);
496 bch2_gc(c, false, false);
497 up_read(&c->state_lock);
503 #ifdef CONFIG_BCACHEFS_TESTS
504 if (attr == &sysfs_perf_test) {
505 char *tmp = kstrdup(buf, GFP_KERNEL), *p = tmp;
506 char *test = strsep(&p, " \t\n");
507 char *nr_str = strsep(&p, " \t\n");
508 char *threads_str = strsep(&p, " \t\n");
514 !(ret = kstrtouint(threads_str, 10, &threads)) &&
515 !(ret = bch2_strtoull_h(nr_str, &nr)))
516 ret = bch2_btree_perf_test(c, test, nr, threads);
527 struct attribute *bch2_fs_files[] = {
529 &sysfs_btree_cache_size,
530 &sysfs_btree_avg_write_size,
532 &sysfs_promote_whole_extents,
534 &sysfs_compression_stats,
536 #ifdef CONFIG_BCACHEFS_TESTS
542 /* internal dir - just a wrapper */
544 SHOW(bch2_fs_internal)
546 struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
547 return bch2_fs_to_text(out, &c->kobj, attr);
550 STORE(bch2_fs_internal)
552 struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
553 return bch2_fs_store(&c->kobj, attr, buf, size);
555 SYSFS_OPS(bch2_fs_internal);
557 struct attribute *bch2_fs_internal_files[] = {
558 &sysfs_journal_debug,
559 &sysfs_btree_updates,
561 &sysfs_btree_key_cache,
562 &sysfs_btree_transactions,
566 &sysfs_io_timers_read,
567 &sysfs_io_timers_write,
572 &sysfs_read_realloc_races,
573 &sysfs_extent_migrate_done,
574 &sysfs_extent_migrate_raced,
578 &sysfs_copy_gc_enabled,
581 &sysfs_rebalance_enabled,
582 &sysfs_rebalance_work,
583 sysfs_pd_controller_files(rebalance),
587 &sysfs_internal_uuid,
593 SHOW(bch2_fs_opts_dir)
595 struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
596 const struct bch_option *opt = container_of(attr, struct bch_option, attr);
597 int id = opt - bch2_opt_table;
598 u64 v = bch2_opt_get_by_id(&c->opts, id);
600 bch2_opt_to_text(out, c, c->disk_sb.sb, opt, v, OPT_SHOW_FULL_LIST);
606 STORE(bch2_fs_opts_dir)
608 struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
609 const struct bch_option *opt = container_of(attr, struct bch_option, attr);
610 int ret = size, id = opt - bch2_opt_table;
615 * We don't need to take c->writes for correctness, but it eliminates an
616 * unsightly error message in the dmesg log when we're RO:
618 if (unlikely(!percpu_ref_tryget(&c->writes)))
621 tmp = kstrdup(buf, GFP_KERNEL);
627 ret = bch2_opt_parse(c, NULL, opt, strim(tmp), &v);
633 ret = bch2_opt_check_may_set(c, id, v);
637 bch2_opt_set_sb(c, opt, v);
638 bch2_opt_set_by_id(&c->opts, id, v);
640 if ((id == Opt_background_target ||
641 id == Opt_background_compression) && v) {
642 bch2_rebalance_add_work(c, S64_MAX);
646 percpu_ref_put(&c->writes);
649 SYSFS_OPS(bch2_fs_opts_dir);
651 struct attribute *bch2_fs_opts_dir_files[] = { NULL };
653 int bch2_opts_create_sysfs_files(struct kobject *kobj)
655 const struct bch_option *i;
658 for (i = bch2_opt_table;
659 i < bch2_opt_table + bch2_opts_nr;
661 if (!(i->flags & OPT_FS))
664 ret = sysfs_create_file(kobj, &i->attr);
674 SHOW(bch2_fs_time_stats)
676 struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
679 if (attr == &sysfs_time_stat_##name) \
680 bch2_time_stats_to_text(out, &c->times[BCH_TIME_##name]);
687 STORE(bch2_fs_time_stats)
691 SYSFS_OPS(bch2_fs_time_stats);
693 struct attribute *bch2_fs_time_stats_files[] = {
695 &sysfs_time_stat_##name,
701 static void reserve_stats_to_text(struct printbuf *out, struct bch_dev *ca)
703 enum alloc_reserve i;
705 spin_lock(&ca->fs->freelist_lock);
707 pr_buf(out, "free_inc:\t%zu\t%zu\n",
708 fifo_used(&ca->free_inc),
711 for (i = 0; i < RESERVE_NR; i++)
712 pr_buf(out, "free[%u]:\t%zu\t%zu\n", i,
713 fifo_used(&ca->free[i]),
716 spin_unlock(&ca->fs->freelist_lock);
719 static void dev_alloc_debug_to_text(struct printbuf *out, struct bch_dev *ca)
721 struct bch_fs *c = ca->fs;
722 struct bch_dev_usage stats = bch2_dev_usage_read(ca);
723 unsigned i, nr[BCH_DATA_NR];
725 memset(nr, 0, sizeof(nr));
727 for (i = 0; i < ARRAY_SIZE(c->open_buckets); i++)
728 nr[c->open_buckets[i].data_type]++;
731 "\t\t buckets\t sectors fragmented\n"
733 ca->mi.nbuckets - ca->mi.first_bucket);
735 for (i = 1; i < BCH_DATA_NR; i++)
736 pr_buf(out, "%-8s%16llu%16llu%16llu\n",
737 bch2_data_types[i], stats.d[i].buckets,
738 stats.d[i].sectors, stats.d[i].fragmented);
744 "free_inc\t\t%zu/%zu\n"
745 "free[RESERVE_MOVINGGC]\t%zu/%zu\n"
746 "free[RESERVE_NONE]\t%zu/%zu\n"
747 "freelist_wait\t\t%s\n"
748 "open buckets allocated\t%u\n"
749 "open buckets this dev\t%u\n"
750 "open buckets total\t%u\n"
751 "open_buckets_wait\t%s\n"
752 "open_buckets_btree\t%u\n"
753 "open_buckets_user\t%u\n"
754 "btree reserve cache\t%u\n"
755 "thread state:\t\t%s\n",
757 __dev_buckets_available(ca, stats),
758 fifo_used(&ca->free_inc), ca->free_inc.size,
759 fifo_used(&ca->free[RESERVE_MOVINGGC]), ca->free[RESERVE_MOVINGGC].size,
760 fifo_used(&ca->free[RESERVE_NONE]), ca->free[RESERVE_NONE].size,
761 c->freelist_wait.list.first ? "waiting" : "empty",
762 OPEN_BUCKETS_COUNT - c->open_buckets_nr_free,
765 c->open_buckets_wait.list.first ? "waiting" : "empty",
768 c->btree_reserve_cache_nr,
769 bch2_allocator_states[ca->allocator_state]);
772 static const char * const bch2_rw[] = {
778 static void dev_iodone_to_text(struct printbuf *out, struct bch_dev *ca)
782 for (rw = 0; rw < 2; rw++) {
783 pr_buf(out, "%s:\n", bch2_rw[rw]);
785 for (i = 1; i < BCH_DATA_NR; i++)
786 pr_buf(out, "%-12s:%12llu\n",
788 percpu_u64_get(&ca->io_done->sectors[rw][i]) << 9);
794 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
795 struct bch_fs *c = ca->fs;
797 sysfs_printf(uuid, "%pU\n", ca->uuid.b);
799 sysfs_print(bucket_size, bucket_bytes(ca));
800 sysfs_print(first_bucket, ca->mi.first_bucket);
801 sysfs_print(nbuckets, ca->mi.nbuckets);
802 sysfs_print(durability, ca->mi.durability);
803 sysfs_print(discard, ca->mi.discard);
805 if (attr == &sysfs_label) {
807 mutex_lock(&c->sb_lock);
808 bch2_disk_path_to_text(out, c->disk_sb.sb,
810 mutex_unlock(&c->sb_lock);
816 if (attr == &sysfs_has_data) {
817 bch2_flags_to_text(out, bch2_data_types,
818 bch2_dev_has_data(c, ca));
822 if (attr == &sysfs_state_rw) {
823 bch2_string_opt_to_text(out, bch2_member_states,
828 if (attr == &sysfs_iodone)
829 dev_iodone_to_text(out, ca);
831 sysfs_print(io_latency_read, atomic64_read(&ca->cur_latency[READ]));
832 sysfs_print(io_latency_write, atomic64_read(&ca->cur_latency[WRITE]));
834 if (attr == &sysfs_io_latency_stats_read)
835 bch2_time_stats_to_text(out, &ca->io_latency[READ]);
837 if (attr == &sysfs_io_latency_stats_write)
838 bch2_time_stats_to_text(out, &ca->io_latency[WRITE]);
840 sysfs_printf(congested, "%u%%",
841 clamp(atomic_read(&ca->congested), 0, CONGESTED_MAX)
842 * 100 / CONGESTED_MAX);
844 if (attr == &sysfs_reserve_stats)
845 reserve_stats_to_text(out, ca);
847 if (attr == &sysfs_alloc_debug)
848 dev_alloc_debug_to_text(out, ca);
855 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
856 struct bch_fs *c = ca->fs;
857 struct bch_member *mi;
859 if (attr == &sysfs_discard) {
860 bool v = strtoul_or_return(buf);
862 mutex_lock(&c->sb_lock);
863 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
865 if (v != BCH_MEMBER_DISCARD(mi)) {
866 SET_BCH_MEMBER_DISCARD(mi, v);
869 mutex_unlock(&c->sb_lock);
872 if (attr == &sysfs_label) {
876 tmp = kstrdup(buf, GFP_KERNEL);
880 ret = bch2_dev_group_set(c, ca, strim(tmp));
886 if (attr == &sysfs_wake_allocator)
887 bch2_wake_allocator(ca);
893 struct attribute *bch2_dev_files[] = {
908 &sysfs_io_latency_read,
909 &sysfs_io_latency_write,
910 &sysfs_io_latency_stats_read,
911 &sysfs_io_latency_stats_write,
914 &sysfs_reserve_stats,
918 &sysfs_wake_allocator,
922 #endif /* _BCACHEFS_SYSFS_H_ */