1 // SPDX-License-Identifier: GPL-2.0
3 * Assorted bcachefs debug code
5 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6 * Copyright 2012 Google, Inc.
10 #include "bkey_methods.h"
11 #include "btree_cache.h"
13 #include "btree_iter.h"
14 #include "btree_locking.h"
15 #include "btree_update.h"
25 #include <linux/console.h>
26 #include <linux/debugfs.h>
27 #include <linux/module.h>
28 #include <linux/random.h>
29 #include <linux/seq_file.h>
31 static struct dentry *bch_debug;
33 static bool bch2_btree_verify_replica(struct bch_fs *c, struct btree *b,
34 struct extent_ptr_decoded pick)
36 struct btree *v = c->verify_data;
37 struct btree_node *n_ondisk = c->verify_ondisk;
38 struct btree_node *n_sorted = c->verify_data->data;
39 struct bset *sorted, *inmemory = &b->data->keys;
40 struct bch_dev *ca = bch_dev_bkey_exists(c, pick.ptr.dev);
42 bool failed = false, saw_error = false;
44 if (!bch2_dev_get_ioref(ca, READ))
47 bio = bio_alloc_bioset(ca->disk_sb.bdev,
48 buf_pages(n_sorted, btree_bytes(c)),
52 bio->bi_iter.bi_sector = pick.ptr.offset;
53 bch2_bio_map(bio, n_sorted, btree_bytes(c));
58 percpu_ref_put(&ca->io_ref);
60 memcpy(n_ondisk, n_sorted, btree_bytes(c));
63 if (bch2_btree_node_read_done(c, ca, v, false, &saw_error) || saw_error)
66 n_sorted = c->verify_data->data;
67 sorted = &n_sorted->keys;
69 if (inmemory->u64s != sorted->u64s ||
70 memcmp(inmemory->start,
72 vstruct_end(inmemory) - (void *) inmemory->start)) {
73 unsigned offset = 0, sectors;
79 printk(KERN_ERR "*** in memory:\n");
80 bch2_dump_bset(c, b, inmemory, 0);
82 printk(KERN_ERR "*** read back in:\n");
83 bch2_dump_bset(c, v, sorted, 0);
85 while (offset < v->written) {
88 sectors = vstruct_blocks(n_ondisk, c->block_bits) <<
91 struct btree_node_entry *bne =
92 (void *) n_ondisk + (offset << 9);
95 sectors = vstruct_blocks(bne, c->block_bits) <<
99 printk(KERN_ERR "*** on disk block %u:\n", offset);
100 bch2_dump_bset(c, b, i, offset);
105 for (j = 0; j < le16_to_cpu(inmemory->u64s); j++)
106 if (inmemory->_data[j] != sorted->_data[j])
110 bch_err(c, "verify failed at key %u", j);
115 if (v->written != b->written) {
116 bch_err(c, "written wrong: expected %u, got %u",
117 b->written, v->written);
124 void __bch2_btree_verify(struct bch_fs *c, struct btree *b)
126 struct bkey_ptrs_c ptrs;
127 struct extent_ptr_decoded p;
128 const union bch_extent_entry *entry;
130 struct bset *inmemory = &b->data->keys;
131 struct bkey_packed *k;
134 if (c->opts.nochanges)
137 bch2_btree_node_io_lock(b);
138 mutex_lock(&c->verify_lock);
140 if (!c->verify_ondisk) {
141 c->verify_ondisk = kvpmalloc(btree_bytes(c), GFP_KERNEL);
142 if (!c->verify_ondisk)
146 if (!c->verify_data) {
147 c->verify_data = __bch2_btree_node_mem_alloc(c);
151 list_del_init(&c->verify_data->list);
154 BUG_ON(b->nsets != 1);
156 for (k = inmemory->start; k != vstruct_last(inmemory); k = bkey_next(k))
157 if (k->type == KEY_TYPE_btree_ptr_v2) {
158 struct bch_btree_ptr_v2 *v = (void *) bkeyp_val(&b->format, k);
163 bkey_copy(&v->key, &b->key);
164 v->c.level = b->c.level;
165 v->c.btree_id = b->c.btree_id;
166 bch2_btree_keys_init(v);
168 ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(&b->key));
169 bkey_for_each_ptr_decode(&b->key.k, ptrs, p, entry)
170 failed |= bch2_btree_verify_replica(c, b, p);
173 struct printbuf buf = PRINTBUF;
175 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
176 bch2_fs_fatal_error(c, "btree node verify failed for : %s\n", buf.buf);
180 mutex_unlock(&c->verify_lock);
181 bch2_btree_node_io_unlock(b);
184 #ifdef CONFIG_DEBUG_FS
186 /* XXX: bch_fs refcounting */
192 struct bpos prev_node;
197 char __user *ubuf; /* destination user buffer */
198 size_t size; /* size of requested read */
199 ssize_t ret; /* bytes read so far */
202 static ssize_t flush_buf(struct dump_iter *i)
205 size_t bytes = min_t(size_t, i->buf.pos, i->size);
206 int err = copy_to_user(i->ubuf, i->buf.buf, bytes);
215 memmove(i->buf.buf, i->buf.buf + bytes, i->buf.pos);
218 return i->size ? 0 : i->ret;
221 static int bch2_dump_open(struct inode *inode, struct file *file)
223 struct btree_debug *bd = inode->i_private;
226 i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
230 file->private_data = i;
233 i->c = container_of(bd, struct bch_fs, btree_debug[bd->id]);
240 static int bch2_dump_release(struct inode *inode, struct file *file)
242 struct dump_iter *i = file->private_data;
244 printbuf_exit(&i->buf);
249 static ssize_t bch2_read_btree(struct file *file, char __user *buf,
250 size_t size, loff_t *ppos)
252 struct dump_iter *i = file->private_data;
253 struct btree_trans trans;
254 struct btree_iter iter;
262 bch2_trans_init(&trans, i->c, 0, 0);
264 ret = for_each_btree_key2(&trans, iter, i->id, i->from,
266 BTREE_ITER_ALL_SNAPSHOTS, k, ({
271 bch2_bkey_val_to_text(&i->buf, i->c, k);
272 prt_newline(&i->buf);
280 bch2_trans_exit(&trans);
282 return ret ?: i->ret;
285 static const struct file_operations btree_debug_ops = {
286 .owner = THIS_MODULE,
287 .open = bch2_dump_open,
288 .release = bch2_dump_release,
289 .read = bch2_read_btree,
292 static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
293 size_t size, loff_t *ppos)
295 struct dump_iter *i = file->private_data;
296 struct btree_trans trans;
297 struct btree_iter iter;
309 if (bpos_eq(SPOS_MAX, i->from))
312 bch2_trans_init(&trans, i->c, 0, 0);
314 for_each_btree_node(&trans, iter, i->id, i->from, 0, b, ret) {
319 bch2_btree_node_to_text(&i->buf, i->c, b);
320 i->from = !bpos_eq(SPOS_MAX, b->key.k.p)
321 ? bpos_successor(b->key.k.p)
324 bch2_trans_iter_exit(&trans, &iter);
326 bch2_trans_exit(&trans);
331 return ret ?: i->ret;
334 static const struct file_operations btree_format_debug_ops = {
335 .owner = THIS_MODULE,
336 .open = bch2_dump_open,
337 .release = bch2_dump_release,
338 .read = bch2_read_btree_formats,
341 static ssize_t bch2_read_bfloat_failed(struct file *file, char __user *buf,
342 size_t size, loff_t *ppos)
344 struct dump_iter *i = file->private_data;
345 struct btree_trans trans;
346 struct btree_iter iter;
358 bch2_trans_init(&trans, i->c, 0, 0);
360 ret = for_each_btree_key2(&trans, iter, i->id, i->from,
362 BTREE_ITER_ALL_SNAPSHOTS, k, ({
363 struct btree_path_level *l = &iter.path->l[0];
364 struct bkey_packed *_k =
365 bch2_btree_node_iter_peek(&l->iter, l->b);
371 if (bpos_gt(l->b->key.k.p, i->prev_node)) {
372 bch2_btree_node_to_text(&i->buf, i->c, l->b);
373 i->prev_node = l->b->key.k.p;
376 bch2_bfloat_to_text(&i->buf, l->b, _k);
381 bch2_trans_exit(&trans);
386 return ret ?: i->ret;
389 static const struct file_operations bfloat_failed_debug_ops = {
390 .owner = THIS_MODULE,
391 .open = bch2_dump_open,
392 .release = bch2_dump_release,
393 .read = bch2_read_bfloat_failed,
396 static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
399 if (!out->nr_tabstops)
400 printbuf_tabstop_push(out, 32);
402 prt_printf(out, "%px btree=%s l=%u ",
404 bch2_btree_ids[b->c.btree_id],
408 printbuf_indent_add(out, 2);
410 bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
413 prt_printf(out, "flags: ");
415 prt_bitflags(out, bch2_btree_node_flags, b->flags);
418 prt_printf(out, "pcpu read locks: ");
420 prt_printf(out, "%u", b->c.lock.readers != NULL);
423 prt_printf(out, "written:");
425 prt_printf(out, "%u", b->written);
428 prt_printf(out, "writes blocked:");
430 prt_printf(out, "%u", !list_empty_careful(&b->write_blocked));
433 prt_printf(out, "will make reachable:");
435 prt_printf(out, "%lx", b->will_make_reachable);
438 prt_printf(out, "journal pin %px:", &b->writes[0].journal);
440 prt_printf(out, "%llu", b->writes[0].journal.seq);
443 prt_printf(out, "journal pin %px:", &b->writes[1].journal);
445 prt_printf(out, "%llu", b->writes[1].journal.seq);
448 printbuf_indent_sub(out, 2);
451 static ssize_t bch2_cached_btree_nodes_read(struct file *file, char __user *buf,
452 size_t size, loff_t *ppos)
454 struct dump_iter *i = file->private_data;
455 struct bch_fs *c = i->c;
464 struct bucket_table *tbl;
465 struct rhash_head *pos;
474 tbl = rht_dereference_rcu(c->btree_cache.table.tbl,
475 &c->btree_cache.table);
476 if (i->iter < tbl->size) {
477 rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
478 bch2_cached_btree_node_to_text(&i->buf, c, b);
487 if (i->buf.allocation_failure)
493 return ret ?: i->ret;
496 static const struct file_operations cached_btree_nodes_ops = {
497 .owner = THIS_MODULE,
498 .open = bch2_dump_open,
499 .release = bch2_dump_release,
500 .read = bch2_cached_btree_nodes_read,
503 #ifdef CONFIG_BCACHEFS_DEBUG_TRANSACTIONS
504 static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
505 size_t size, loff_t *ppos)
507 struct dump_iter *i = file->private_data;
508 struct bch_fs *c = i->c;
509 struct btree_trans *trans;
516 mutex_lock(&c->btree_trans_lock);
517 list_for_each_entry(trans, &c->btree_trans_list, list) {
518 if (trans->locking_wait.task->pid <= i->iter)
525 bch2_btree_trans_to_text(&i->buf, trans);
527 prt_printf(&i->buf, "backtrace:");
528 prt_newline(&i->buf);
529 printbuf_indent_add(&i->buf, 2);
530 bch2_prt_task_backtrace(&i->buf, trans->locking_wait.task);
531 printbuf_indent_sub(&i->buf, 2);
532 prt_newline(&i->buf);
534 i->iter = trans->locking_wait.task->pid;
536 mutex_unlock(&c->btree_trans_lock);
538 if (i->buf.allocation_failure)
544 return ret ?: i->ret;
547 static const struct file_operations btree_transactions_ops = {
548 .owner = THIS_MODULE,
549 .open = bch2_dump_open,
550 .release = bch2_dump_release,
551 .read = bch2_btree_transactions_read,
553 #endif /* CONFIG_BCACHEFS_DEBUG_TRANSACTIONS */
555 static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
556 size_t size, loff_t *ppos)
558 struct dump_iter *i = file->private_data;
559 struct bch_fs *c = i->c;
575 done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
579 if (i->buf.allocation_failure)
585 static const struct file_operations journal_pins_ops = {
586 .owner = THIS_MODULE,
587 .open = bch2_dump_open,
588 .release = bch2_dump_release,
589 .read = bch2_journal_pins_read,
592 static int lock_held_stats_open(struct inode *inode, struct file *file)
594 struct bch_fs *c = inode->i_private;
597 i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
605 file->private_data = i;
610 static int lock_held_stats_release(struct inode *inode, struct file *file)
612 struct dump_iter *i = file->private_data;
614 printbuf_exit(&i->buf);
620 static ssize_t lock_held_stats_read(struct file *file, char __user *buf,
621 size_t size, loff_t *ppos)
623 struct dump_iter *i = file->private_data;
624 struct bch_fs *c = i->c;
632 struct btree_transaction_stats *s = &c->btree_transaction_stats[i->iter];
641 if (i->iter == ARRAY_SIZE(bch2_btree_transaction_fns) ||
642 !bch2_btree_transaction_fns[i->iter])
645 prt_printf(&i->buf, "%s: ", bch2_btree_transaction_fns[i->iter]);
646 prt_newline(&i->buf);
647 printbuf_indent_add(&i->buf, 2);
649 mutex_lock(&s->lock);
651 prt_printf(&i->buf, "Max mem used: %u", s->max_mem);
652 prt_newline(&i->buf);
654 if (IS_ENABLED(CONFIG_BCACHEFS_LOCK_TIME_STATS)) {
655 prt_printf(&i->buf, "Lock hold times:");
656 prt_newline(&i->buf);
658 printbuf_indent_add(&i->buf, 2);
659 bch2_time_stats_to_text(&i->buf, &s->lock_hold_times);
660 printbuf_indent_sub(&i->buf, 2);
663 if (s->max_paths_text) {
664 prt_printf(&i->buf, "Maximum allocated btree paths (%u):", s->nr_max_paths);
665 prt_newline(&i->buf);
667 printbuf_indent_add(&i->buf, 2);
668 prt_str_indented(&i->buf, s->max_paths_text);
669 printbuf_indent_sub(&i->buf, 2);
672 mutex_unlock(&s->lock);
674 printbuf_indent_sub(&i->buf, 2);
675 prt_newline(&i->buf);
679 if (i->buf.allocation_failure)
685 static const struct file_operations lock_held_stats_op = {
686 .owner = THIS_MODULE,
687 .open = lock_held_stats_open,
688 .release = lock_held_stats_release,
689 .read = lock_held_stats_read,
692 static ssize_t bch2_btree_deadlock_read(struct file *file, char __user *buf,
693 size_t size, loff_t *ppos)
695 struct dump_iter *i = file->private_data;
696 struct bch_fs *c = i->c;
697 struct btree_trans *trans;
707 mutex_lock(&c->btree_trans_lock);
708 list_for_each_entry(trans, &c->btree_trans_list, list) {
709 if (trans->locking_wait.task->pid <= i->iter)
716 bch2_check_for_deadlock(trans, &i->buf);
718 i->iter = trans->locking_wait.task->pid;
720 mutex_unlock(&c->btree_trans_lock);
722 if (i->buf.allocation_failure)
728 return ret ?: i->ret;
731 static const struct file_operations btree_deadlock_ops = {
732 .owner = THIS_MODULE,
733 .open = bch2_dump_open,
734 .release = bch2_dump_release,
735 .read = bch2_btree_deadlock_read,
738 void bch2_fs_debug_exit(struct bch_fs *c)
740 if (!IS_ERR_OR_NULL(c->fs_debug_dir))
741 debugfs_remove_recursive(c->fs_debug_dir);
744 void bch2_fs_debug_init(struct bch_fs *c)
746 struct btree_debug *bd;
749 if (IS_ERR_OR_NULL(bch_debug))
752 snprintf(name, sizeof(name), "%pU", c->sb.user_uuid.b);
753 c->fs_debug_dir = debugfs_create_dir(name, bch_debug);
754 if (IS_ERR_OR_NULL(c->fs_debug_dir))
757 debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
758 c->btree_debug, &cached_btree_nodes_ops);
760 #ifdef CONFIG_BCACHEFS_DEBUG_TRANSACTIONS
761 debugfs_create_file("btree_transactions", 0400, c->fs_debug_dir,
762 c->btree_debug, &btree_transactions_ops);
765 debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
766 c->btree_debug, &journal_pins_ops);
768 debugfs_create_file("btree_transaction_stats", 0400, c->fs_debug_dir,
769 c, &lock_held_stats_op);
771 debugfs_create_file("btree_deadlock", 0400, c->fs_debug_dir,
772 c->btree_debug, &btree_deadlock_ops);
774 c->btree_debug_dir = debugfs_create_dir("btrees", c->fs_debug_dir);
775 if (IS_ERR_OR_NULL(c->btree_debug_dir))
778 for (bd = c->btree_debug;
779 bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
781 bd->id = bd - c->btree_debug;
782 debugfs_create_file(bch2_btree_ids[bd->id],
783 0400, c->btree_debug_dir, bd,
786 snprintf(name, sizeof(name), "%s-formats",
787 bch2_btree_ids[bd->id]);
789 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
790 &btree_format_debug_ops);
792 snprintf(name, sizeof(name), "%s-bfloat-failed",
793 bch2_btree_ids[bd->id]);
795 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
796 &bfloat_failed_debug_ops);
802 void bch2_debug_exit(void)
804 if (!IS_ERR_OR_NULL(bch_debug))
805 debugfs_remove_recursive(bch_debug);
808 int __init bch2_debug_init(void)
812 bch_debug = debugfs_create_dir("bcachefs", NULL);