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(btree_cache_size);
174 read_attribute(compression_stats);
175 read_attribute(journal_debug);
176 read_attribute(btree_updates);
177 read_attribute(btree_cache);
178 read_attribute(btree_key_cache);
179 read_attribute(btree_transactions);
180 read_attribute(stripes_heap);
181 read_attribute(open_buckets);
183 read_attribute(internal_uuid);
185 read_attribute(has_data);
186 read_attribute(alloc_debug);
188 read_attribute(read_realloc_races);
189 read_attribute(extent_migrate_done);
190 read_attribute(extent_migrate_raced);
191 read_attribute(bucket_alloc_fail);
193 rw_attribute(discard);
196 rw_attribute(copy_gc_enabled);
197 read_attribute(copy_gc_wait);
199 rw_attribute(rebalance_enabled);
200 sysfs_pd_controller_attribute(rebalance);
201 read_attribute(rebalance_work);
202 rw_attribute(promote_whole_extents);
204 read_attribute(new_stripes);
206 read_attribute(io_timers_read);
207 read_attribute(io_timers_write);
209 read_attribute(data_jobs);
211 #ifdef CONFIG_BCACHEFS_TESTS
212 write_attribute(perf_test);
213 #endif /* CONFIG_BCACHEFS_TESTS */
216 static struct attribute sysfs_time_stat_##_name = \
217 { .name = #_name, .mode = S_IRUGO };
221 static struct attribute sysfs_state_rw = {
226 static size_t bch2_btree_cache_size(struct bch_fs *c)
231 mutex_lock(&c->btree_cache.lock);
232 list_for_each_entry(b, &c->btree_cache.live, list)
233 ret += btree_bytes(c);
235 mutex_unlock(&c->btree_cache.lock);
239 static size_t bch2_btree_avg_write_size(struct bch_fs *c)
241 u64 nr = atomic64_read(&c->btree_writes_nr);
242 u64 sectors = atomic64_read(&c->btree_writes_sectors);
244 return nr ? div64_u64(sectors, nr) : 0;
247 static long data_progress_to_text(struct printbuf *out, struct bch_fs *c)
250 struct bch_move_stats *stats;
252 mutex_lock(&c->data_progress_lock);
253 list_for_each_entry(stats, &c->data_progress_list, list) {
254 pr_buf(out, "%s: data type %s btree_id %s position: ",
256 bch2_data_types[stats->data_type],
257 bch2_btree_ids[stats->btree_id]);
258 bch2_bpos_to_text(out, stats->pos);
259 pr_buf(out, "%s", "\n");
262 mutex_unlock(&c->data_progress_lock);
266 static int bch2_compression_stats_to_text(struct printbuf *out, struct bch_fs *c)
268 struct btree_trans trans;
269 struct btree_iter iter;
272 u64 nr_uncompressed_extents = 0,
273 nr_compressed_extents = 0,
274 nr_incompressible_extents = 0,
275 uncompressed_sectors = 0,
276 incompressible_sectors = 0,
277 compressed_sectors_compressed = 0,
278 compressed_sectors_uncompressed = 0;
281 if (!test_bit(BCH_FS_STARTED, &c->flags))
284 bch2_trans_init(&trans, c, 0, 0);
286 for (id = 0; id < BTREE_ID_NR; id++) {
287 if (!((1U << id) & BTREE_ID_HAS_PTRS))
290 for_each_btree_key(&trans, iter, id, POS_MIN,
291 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
292 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
293 const union bch_extent_entry *entry;
294 struct extent_ptr_decoded p;
295 bool compressed = false, uncompressed = false, incompressible = false;
297 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
298 switch (p.crc.compression_type) {
299 case BCH_COMPRESSION_TYPE_none:
301 uncompressed_sectors += k.k->size;
303 case BCH_COMPRESSION_TYPE_incompressible:
304 incompressible = true;
305 incompressible_sectors += k.k->size;
308 compressed_sectors_compressed +=
309 p.crc.compressed_size;
310 compressed_sectors_uncompressed +=
311 p.crc.uncompressed_size;
318 nr_incompressible_extents++;
319 else if (uncompressed)
320 nr_uncompressed_extents++;
322 nr_compressed_extents++;
324 bch2_trans_iter_exit(&trans, &iter);
327 bch2_trans_exit(&trans);
332 pr_buf(out, "uncompressed:\n");
333 pr_buf(out, " nr extents: %llu\n", nr_uncompressed_extents);
334 pr_buf(out, " size: ");
335 bch2_hprint(out, uncompressed_sectors << 9);
338 pr_buf(out, "compressed:\n");
339 pr_buf(out, " nr extents: %llu\n", nr_compressed_extents);
340 pr_buf(out, " compressed size: ");
341 bch2_hprint(out, compressed_sectors_compressed << 9);
343 pr_buf(out, " uncompressed size: ");
344 bch2_hprint(out, compressed_sectors_uncompressed << 9);
347 pr_buf(out, "incompressible:\n");
348 pr_buf(out, " nr extents: %llu\n", nr_incompressible_extents);
349 pr_buf(out, " size: ");
350 bch2_hprint(out, incompressible_sectors << 9);
355 static void bch2_gc_gens_pos_to_text(struct printbuf *out, struct bch_fs *c)
357 pr_buf(out, "%s: ", bch2_btree_ids[c->gc_gens_btree]);
358 bch2_bpos_to_text(out, c->gc_gens_pos);
364 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
366 sysfs_print(minor, c->minor);
367 sysfs_printf(internal_uuid, "%pU", c->sb.uuid.b);
369 sysfs_hprint(btree_cache_size, bch2_btree_cache_size(c));
370 sysfs_hprint(btree_avg_write_size, bch2_btree_avg_write_size(c));
372 sysfs_print(read_realloc_races,
373 atomic_long_read(&c->read_realloc_races));
374 sysfs_print(extent_migrate_done,
375 atomic_long_read(&c->extent_migrate_done));
376 sysfs_print(extent_migrate_raced,
377 atomic_long_read(&c->extent_migrate_raced));
378 sysfs_print(bucket_alloc_fail,
379 atomic_long_read(&c->bucket_alloc_fail));
381 sysfs_printf(btree_gc_periodic, "%u", (int) c->btree_gc_periodic);
383 if (attr == &sysfs_gc_gens_pos)
384 bch2_gc_gens_pos_to_text(out, c);
386 sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
388 sysfs_printf(rebalance_enabled, "%i", c->rebalance.enabled);
389 sysfs_pd_controller_show(rebalance, &c->rebalance.pd); /* XXX */
390 sysfs_hprint(copy_gc_wait,
391 max(0LL, c->copygc_wait -
392 atomic64_read(&c->io_clock[WRITE].now)) << 9);
394 if (attr == &sysfs_rebalance_work)
395 bch2_rebalance_work_to_text(out, c);
397 sysfs_print(promote_whole_extents, c->promote_whole_extents);
401 if (attr == &sysfs_journal_debug)
402 bch2_journal_debug_to_text(out, &c->journal);
404 if (attr == &sysfs_btree_updates)
405 bch2_btree_updates_to_text(out, c);
407 if (attr == &sysfs_btree_cache)
408 bch2_btree_cache_to_text(out, c);
410 if (attr == &sysfs_btree_key_cache)
411 bch2_btree_key_cache_to_text(out, &c->btree_key_cache);
413 if (attr == &sysfs_btree_transactions)
414 bch2_btree_trans_to_text(out, c);
416 if (attr == &sysfs_stripes_heap)
417 bch2_stripes_heap_to_text(out, c);
419 if (attr == &sysfs_open_buckets)
420 bch2_open_buckets_to_text(out, c);
422 if (attr == &sysfs_compression_stats)
423 bch2_compression_stats_to_text(out, c);
425 if (attr == &sysfs_new_stripes)
426 bch2_new_stripes_to_text(out, c);
428 if (attr == &sysfs_io_timers_read)
429 bch2_io_timers_to_text(out, &c->io_clock[READ]);
431 if (attr == &sysfs_io_timers_write)
432 bch2_io_timers_to_text(out, &c->io_clock[WRITE]);
434 if (attr == &sysfs_data_jobs)
435 data_progress_to_text(out, c);
442 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
444 if (attr == &sysfs_btree_gc_periodic) {
445 ssize_t ret = strtoul_safe(buf, c->btree_gc_periodic)
448 wake_up_process(c->gc_thread);
452 if (attr == &sysfs_copy_gc_enabled) {
453 ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
456 if (c->copygc_thread)
457 wake_up_process(c->copygc_thread);
461 if (attr == &sysfs_rebalance_enabled) {
462 ssize_t ret = strtoul_safe(buf, c->rebalance.enabled)
469 sysfs_pd_controller_store(rebalance, &c->rebalance.pd);
471 sysfs_strtoul(promote_whole_extents, c->promote_whole_extents);
475 if (!test_bit(BCH_FS_STARTED, &c->flags))
480 if (!test_bit(BCH_FS_RW, &c->flags))
483 if (attr == &sysfs_prune_cache) {
484 struct shrink_control sc;
486 sc.gfp_mask = GFP_KERNEL;
487 sc.nr_to_scan = strtoul_or_return(buf);
488 c->btree_cache.shrink.scan_objects(&c->btree_cache.shrink, &sc);
491 if (attr == &sysfs_trigger_gc) {
493 * Full gc is currently incompatible with btree key cache:
496 down_read(&c->state_lock);
497 bch2_gc(c, false, false);
498 up_read(&c->state_lock);
504 #ifdef CONFIG_BCACHEFS_TESTS
505 if (attr == &sysfs_perf_test) {
506 char *tmp = kstrdup(buf, GFP_KERNEL), *p = tmp;
507 char *test = strsep(&p, " \t\n");
508 char *nr_str = strsep(&p, " \t\n");
509 char *threads_str = strsep(&p, " \t\n");
515 !(ret = kstrtouint(threads_str, 10, &threads)) &&
516 !(ret = bch2_strtoull_h(nr_str, &nr)))
517 ret = bch2_btree_perf_test(c, test, nr, threads);
528 struct attribute *bch2_fs_files[] = {
530 &sysfs_btree_cache_size,
531 &sysfs_btree_avg_write_size,
533 &sysfs_promote_whole_extents,
535 &sysfs_compression_stats,
537 #ifdef CONFIG_BCACHEFS_TESTS
543 /* internal dir - just a wrapper */
545 SHOW(bch2_fs_internal)
547 struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
548 return bch2_fs_to_text(out, &c->kobj, attr);
551 STORE(bch2_fs_internal)
553 struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
554 return bch2_fs_store(&c->kobj, attr, buf, size);
556 SYSFS_OPS(bch2_fs_internal);
558 struct attribute *bch2_fs_internal_files[] = {
559 &sysfs_journal_debug,
560 &sysfs_btree_updates,
562 &sysfs_btree_key_cache,
563 &sysfs_btree_transactions,
567 &sysfs_io_timers_read,
568 &sysfs_io_timers_write,
573 &sysfs_read_realloc_races,
574 &sysfs_extent_migrate_done,
575 &sysfs_extent_migrate_raced,
576 &sysfs_bucket_alloc_fail,
580 &sysfs_copy_gc_enabled,
583 &sysfs_rebalance_enabled,
584 &sysfs_rebalance_work,
585 sysfs_pd_controller_files(rebalance),
589 &sysfs_internal_uuid,
595 SHOW(bch2_fs_opts_dir)
597 struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
598 const struct bch_option *opt = container_of(attr, struct bch_option, attr);
599 int id = opt - bch2_opt_table;
600 u64 v = bch2_opt_get_by_id(&c->opts, id);
602 bch2_opt_to_text(out, c, c->disk_sb.sb, opt, v, OPT_SHOW_FULL_LIST);
608 STORE(bch2_fs_opts_dir)
610 struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
611 const struct bch_option *opt = container_of(attr, struct bch_option, attr);
612 int ret = size, id = opt - bch2_opt_table;
617 * We don't need to take c->writes for correctness, but it eliminates an
618 * unsightly error message in the dmesg log when we're RO:
620 if (unlikely(!percpu_ref_tryget(&c->writes)))
623 tmp = kstrdup(buf, GFP_KERNEL);
629 ret = bch2_opt_parse(c, opt, strim(tmp), &v, NULL);
635 ret = bch2_opt_check_may_set(c, id, v);
639 bch2_opt_set_sb(c, opt, v);
640 bch2_opt_set_by_id(&c->opts, id, v);
642 if ((id == Opt_background_target ||
643 id == Opt_background_compression) && v) {
644 bch2_rebalance_add_work(c, S64_MAX);
648 percpu_ref_put(&c->writes);
651 SYSFS_OPS(bch2_fs_opts_dir);
653 struct attribute *bch2_fs_opts_dir_files[] = { NULL };
655 int bch2_opts_create_sysfs_files(struct kobject *kobj)
657 const struct bch_option *i;
660 for (i = bch2_opt_table;
661 i < bch2_opt_table + bch2_opts_nr;
663 if (!(i->flags & OPT_FS))
666 ret = sysfs_create_file(kobj, &i->attr);
676 SHOW(bch2_fs_time_stats)
678 struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
681 if (attr == &sysfs_time_stat_##name) \
682 bch2_time_stats_to_text(out, &c->times[BCH_TIME_##name]);
689 STORE(bch2_fs_time_stats)
693 SYSFS_OPS(bch2_fs_time_stats);
695 struct attribute *bch2_fs_time_stats_files[] = {
697 &sysfs_time_stat_##name,
703 static void dev_alloc_debug_to_text(struct printbuf *out, struct bch_dev *ca)
705 struct bch_fs *c = ca->fs;
706 struct bch_dev_usage stats = bch2_dev_usage_read(ca);
707 unsigned i, nr[BCH_DATA_NR];
709 memset(nr, 0, sizeof(nr));
711 for (i = 0; i < ARRAY_SIZE(c->open_buckets); i++)
712 nr[c->open_buckets[i].data_type]++;
715 "\t\t buckets\t sectors fragmented\n"
717 ca->mi.nbuckets - ca->mi.first_bucket);
719 for (i = 1; i < BCH_DATA_NR; i++)
720 pr_buf(out, "%-8s%16llu%16llu%16llu\n",
721 bch2_data_types[i], stats.d[i].buckets,
722 stats.d[i].sectors, stats.d[i].fragmented);
728 "freelist_wait\t\t%s\n"
729 "open buckets allocated\t%u\n"
730 "open buckets this dev\t%u\n"
731 "open buckets total\t%u\n"
732 "open_buckets_wait\t%s\n"
733 "open_buckets_btree\t%u\n"
734 "open_buckets_user\t%u\n"
735 "btree reserve cache\t%u\n",
737 __dev_buckets_available(ca, stats, RESERVE_none),
738 c->freelist_wait.list.first ? "waiting" : "empty",
739 OPEN_BUCKETS_COUNT - c->open_buckets_nr_free,
742 c->open_buckets_wait.list.first ? "waiting" : "empty",
745 c->btree_reserve_cache_nr);
748 static const char * const bch2_rw[] = {
754 static void dev_iodone_to_text(struct printbuf *out, struct bch_dev *ca)
758 for (rw = 0; rw < 2; rw++) {
759 pr_buf(out, "%s:\n", bch2_rw[rw]);
761 for (i = 1; i < BCH_DATA_NR; i++)
762 pr_buf(out, "%-12s:%12llu\n",
764 percpu_u64_get(&ca->io_done->sectors[rw][i]) << 9);
770 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
771 struct bch_fs *c = ca->fs;
773 sysfs_printf(uuid, "%pU\n", ca->uuid.b);
775 sysfs_print(bucket_size, bucket_bytes(ca));
776 sysfs_print(first_bucket, ca->mi.first_bucket);
777 sysfs_print(nbuckets, ca->mi.nbuckets);
778 sysfs_print(durability, ca->mi.durability);
779 sysfs_print(discard, ca->mi.discard);
781 if (attr == &sysfs_label) {
783 mutex_lock(&c->sb_lock);
784 bch2_disk_path_to_text(out, c->disk_sb.sb,
786 mutex_unlock(&c->sb_lock);
792 if (attr == &sysfs_has_data) {
793 bch2_flags_to_text(out, bch2_data_types,
794 bch2_dev_has_data(c, ca));
798 if (attr == &sysfs_state_rw) {
799 bch2_string_opt_to_text(out, bch2_member_states,
804 if (attr == &sysfs_iodone)
805 dev_iodone_to_text(out, ca);
807 sysfs_print(io_latency_read, atomic64_read(&ca->cur_latency[READ]));
808 sysfs_print(io_latency_write, atomic64_read(&ca->cur_latency[WRITE]));
810 if (attr == &sysfs_io_latency_stats_read)
811 bch2_time_stats_to_text(out, &ca->io_latency[READ]);
813 if (attr == &sysfs_io_latency_stats_write)
814 bch2_time_stats_to_text(out, &ca->io_latency[WRITE]);
816 sysfs_printf(congested, "%u%%",
817 clamp(atomic_read(&ca->congested), 0, CONGESTED_MAX)
818 * 100 / CONGESTED_MAX);
820 if (attr == &sysfs_alloc_debug)
821 dev_alloc_debug_to_text(out, ca);
828 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
829 struct bch_fs *c = ca->fs;
830 struct bch_member *mi;
832 if (attr == &sysfs_discard) {
833 bool v = strtoul_or_return(buf);
835 mutex_lock(&c->sb_lock);
836 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
838 if (v != BCH_MEMBER_DISCARD(mi)) {
839 SET_BCH_MEMBER_DISCARD(mi, v);
842 mutex_unlock(&c->sb_lock);
845 if (attr == &sysfs_label) {
849 tmp = kstrdup(buf, GFP_KERNEL);
853 ret = bch2_dev_group_set(c, ca, strim(tmp));
863 struct attribute *bch2_dev_files[] = {
878 &sysfs_io_latency_read,
879 &sysfs_io_latency_write,
880 &sysfs_io_latency_stats_read,
881 &sysfs_io_latency_stats_write,
889 #endif /* _BCACHEFS_SYSFS_H_ */