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_p_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 void bch2_btree_node_ondisk_to_text(struct printbuf *out, struct bch_fs *c,
185 const struct btree *b)
187 struct btree_node *n_ondisk = NULL;
188 struct extent_ptr_decoded pick;
190 struct bio *bio = NULL;
194 if (bch2_bkey_pick_read_device(c, bkey_i_to_s_c(&b->key), NULL, &pick) <= 0) {
195 prt_printf(out, "error getting device to read from: invalid device\n");
199 ca = bch_dev_bkey_exists(c, pick.ptr.dev);
200 if (!bch2_dev_get_ioref(ca, READ)) {
201 prt_printf(out, "error getting device to read from: not online\n");
205 n_ondisk = kvpmalloc(btree_bytes(c), GFP_KERNEL);
207 prt_printf(out, "memory allocation failure\n");
211 bio = bio_alloc_bioset(ca->disk_sb.bdev,
212 buf_pages(n_ondisk, btree_bytes(c)),
213 REQ_OP_READ|REQ_META,
216 bio->bi_iter.bi_sector = pick.ptr.offset;
217 bch2_bio_map(bio, n_ondisk, btree_bytes(c));
219 ret = submit_bio_wait(bio);
221 prt_printf(out, "IO error reading btree node: %s\n", bch2_err_str(ret));
225 while (offset < btree_sectors(c)) {
228 struct bch_csum csum;
229 struct bkey_packed *k;
235 if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
236 prt_printf(out, "unknown checksum type at offset %u: %llu\n",
237 offset, BSET_CSUM_TYPE(i));
241 nonce = btree_nonce(i, offset << 9);
242 csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, n_ondisk);
244 if (bch2_crc_cmp(csum, n_ondisk->csum)) {
245 prt_printf(out, "invalid checksum\n");
249 bset_encrypt(c, i, offset << 9);
251 sectors = vstruct_sectors(n_ondisk, c->block_bits);
253 struct btree_node_entry *bne = (void *) n_ondisk + (offset << 9);
257 if (i->seq != n_ondisk->keys.seq)
260 if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
261 prt_printf(out, "unknown checksum type at offset %u: %llu\n",
262 offset, BSET_CSUM_TYPE(i));
266 nonce = btree_nonce(i, offset << 9);
267 csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne);
269 if (bch2_crc_cmp(csum, bne->csum)) {
270 prt_printf(out, "invalid checksum");
274 bset_encrypt(c, i, offset << 9);
276 sectors = vstruct_sectors(bne, c->block_bits);
279 prt_printf(out, " offset %u version %u, journal seq %llu\n",
281 le16_to_cpu(i->version),
282 le64_to_cpu(i->journal_seq));
285 printbuf_indent_add(out, 4);
287 for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k)) {
290 bch2_bkey_val_to_text(out, c, bkey_disassemble(b, k, &u));
294 printbuf_indent_sub(out, 4);
299 kvpfree(n_ondisk, btree_bytes(c));
300 percpu_ref_put(&ca->io_ref);
303 #ifdef CONFIG_DEBUG_FS
305 /* XXX: bch_fs refcounting */
311 struct bpos prev_node;
316 char __user *ubuf; /* destination user buffer */
317 size_t size; /* size of requested read */
318 ssize_t ret; /* bytes read so far */
321 static ssize_t flush_buf(struct dump_iter *i)
324 size_t bytes = min_t(size_t, i->buf.pos, i->size);
325 int err = copy_to_user(i->ubuf, i->buf.buf, bytes);
334 memmove(i->buf.buf, i->buf.buf + bytes, i->buf.pos);
337 return i->size ? 0 : i->ret;
340 static int bch2_dump_open(struct inode *inode, struct file *file)
342 struct btree_debug *bd = inode->i_private;
345 i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
349 file->private_data = i;
352 i->c = container_of(bd, struct bch_fs, btree_debug[bd->id]);
359 static int bch2_dump_release(struct inode *inode, struct file *file)
361 struct dump_iter *i = file->private_data;
363 printbuf_exit(&i->buf);
368 static ssize_t bch2_read_btree(struct file *file, char __user *buf,
369 size_t size, loff_t *ppos)
371 struct dump_iter *i = file->private_data;
372 struct btree_trans trans;
373 struct btree_iter iter;
385 bch2_trans_init(&trans, i->c, 0, 0);
386 ret = for_each_btree_key2(&trans, iter, i->id, i->from,
388 BTREE_ITER_ALL_SNAPSHOTS, k, ({
389 bch2_bkey_val_to_text(&i->buf, i->c, k);
390 prt_newline(&i->buf);
391 drop_locks_do(&trans, flush_buf(i));
395 bch2_trans_exit(&trans);
400 return ret ?: i->ret;
403 static const struct file_operations btree_debug_ops = {
404 .owner = THIS_MODULE,
405 .open = bch2_dump_open,
406 .release = bch2_dump_release,
407 .read = bch2_read_btree,
410 static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
411 size_t size, loff_t *ppos)
413 struct dump_iter *i = file->private_data;
414 struct btree_trans trans;
415 struct btree_iter iter;
427 if (bpos_eq(SPOS_MAX, i->from))
430 bch2_trans_init(&trans, i->c, 0, 0);
432 bch2_trans_begin(&trans);
434 for_each_btree_node(&trans, iter, i->id, i->from, 0, b, ret) {
435 bch2_btree_node_to_text(&i->buf, i->c, b);
436 i->from = !bpos_eq(SPOS_MAX, b->key.k.p)
437 ? bpos_successor(b->key.k.p)
440 ret = drop_locks_do(&trans, flush_buf(i));
444 bch2_trans_iter_exit(&trans, &iter);
446 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
449 bch2_trans_exit(&trans);
454 return ret ?: i->ret;
457 static const struct file_operations btree_format_debug_ops = {
458 .owner = THIS_MODULE,
459 .open = bch2_dump_open,
460 .release = bch2_dump_release,
461 .read = bch2_read_btree_formats,
464 static ssize_t bch2_read_bfloat_failed(struct file *file, char __user *buf,
465 size_t size, loff_t *ppos)
467 struct dump_iter *i = file->private_data;
468 struct btree_trans trans;
469 struct btree_iter iter;
481 bch2_trans_init(&trans, i->c, 0, 0);
483 ret = for_each_btree_key2(&trans, iter, i->id, i->from,
485 BTREE_ITER_ALL_SNAPSHOTS, k, ({
486 struct btree_path_level *l = &iter.path->l[0];
487 struct bkey_packed *_k =
488 bch2_btree_node_iter_peek(&l->iter, l->b);
490 if (bpos_gt(l->b->key.k.p, i->prev_node)) {
491 bch2_btree_node_to_text(&i->buf, i->c, l->b);
492 i->prev_node = l->b->key.k.p;
495 bch2_bfloat_to_text(&i->buf, l->b, _k);
496 drop_locks_do(&trans, flush_buf(i));
500 bch2_trans_exit(&trans);
505 return ret ?: i->ret;
508 static const struct file_operations bfloat_failed_debug_ops = {
509 .owner = THIS_MODULE,
510 .open = bch2_dump_open,
511 .release = bch2_dump_release,
512 .read = bch2_read_bfloat_failed,
515 static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
518 if (!out->nr_tabstops)
519 printbuf_tabstop_push(out, 32);
521 prt_printf(out, "%px btree=%s l=%u ",
523 bch2_btree_ids[b->c.btree_id],
527 printbuf_indent_add(out, 2);
529 bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
532 prt_printf(out, "flags: ");
534 prt_bitflags(out, bch2_btree_node_flags, b->flags);
537 prt_printf(out, "pcpu read locks: ");
539 prt_printf(out, "%u", b->c.lock.readers != NULL);
542 prt_printf(out, "written:");
544 prt_printf(out, "%u", b->written);
547 prt_printf(out, "writes blocked:");
549 prt_printf(out, "%u", !list_empty_careful(&b->write_blocked));
552 prt_printf(out, "will make reachable:");
554 prt_printf(out, "%lx", b->will_make_reachable);
557 prt_printf(out, "journal pin %px:", &b->writes[0].journal);
559 prt_printf(out, "%llu", b->writes[0].journal.seq);
562 prt_printf(out, "journal pin %px:", &b->writes[1].journal);
564 prt_printf(out, "%llu", b->writes[1].journal.seq);
567 printbuf_indent_sub(out, 2);
570 static ssize_t bch2_cached_btree_nodes_read(struct file *file, char __user *buf,
571 size_t size, loff_t *ppos)
573 struct dump_iter *i = file->private_data;
574 struct bch_fs *c = i->c;
583 struct bucket_table *tbl;
584 struct rhash_head *pos;
593 tbl = rht_dereference_rcu(c->btree_cache.table.tbl,
594 &c->btree_cache.table);
595 if (i->iter < tbl->size) {
596 rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
597 bch2_cached_btree_node_to_text(&i->buf, c, b);
606 if (i->buf.allocation_failure)
612 return ret ?: i->ret;
615 static const struct file_operations cached_btree_nodes_ops = {
616 .owner = THIS_MODULE,
617 .open = bch2_dump_open,
618 .release = bch2_dump_release,
619 .read = bch2_cached_btree_nodes_read,
622 #ifdef CONFIG_BCACHEFS_DEBUG_TRANSACTIONS
623 static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
624 size_t size, loff_t *ppos)
626 struct dump_iter *i = file->private_data;
627 struct bch_fs *c = i->c;
628 struct btree_trans *trans;
636 seqmutex_lock(&c->btree_trans_lock);
637 list_for_each_entry(trans, &c->btree_trans_list, list) {
638 if (trans->locking_wait.task->pid <= i->iter)
641 closure_get(&trans->ref);
642 seq = seqmutex_seq(&c->btree_trans_lock);
643 seqmutex_unlock(&c->btree_trans_lock);
647 closure_put(&trans->ref);
651 bch2_btree_trans_to_text(&i->buf, trans);
653 prt_printf(&i->buf, "backtrace:");
654 prt_newline(&i->buf);
655 printbuf_indent_add(&i->buf, 2);
656 bch2_prt_task_backtrace(&i->buf, trans->locking_wait.task);
657 printbuf_indent_sub(&i->buf, 2);
658 prt_newline(&i->buf);
660 i->iter = trans->locking_wait.task->pid;
662 closure_put(&trans->ref);
664 if (!seqmutex_relock(&c->btree_trans_lock, seq))
667 seqmutex_unlock(&c->btree_trans_lock);
669 if (i->buf.allocation_failure)
675 return ret ?: i->ret;
678 static const struct file_operations btree_transactions_ops = {
679 .owner = THIS_MODULE,
680 .open = bch2_dump_open,
681 .release = bch2_dump_release,
682 .read = bch2_btree_transactions_read,
684 #endif /* CONFIG_BCACHEFS_DEBUG_TRANSACTIONS */
686 static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
687 size_t size, loff_t *ppos)
689 struct dump_iter *i = file->private_data;
690 struct bch_fs *c = i->c;
706 done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
710 if (i->buf.allocation_failure)
716 static const struct file_operations journal_pins_ops = {
717 .owner = THIS_MODULE,
718 .open = bch2_dump_open,
719 .release = bch2_dump_release,
720 .read = bch2_journal_pins_read,
723 static int lock_held_stats_open(struct inode *inode, struct file *file)
725 struct bch_fs *c = inode->i_private;
728 i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
736 file->private_data = i;
741 static int lock_held_stats_release(struct inode *inode, struct file *file)
743 struct dump_iter *i = file->private_data;
745 printbuf_exit(&i->buf);
751 static ssize_t lock_held_stats_read(struct file *file, char __user *buf,
752 size_t size, loff_t *ppos)
754 struct dump_iter *i = file->private_data;
755 struct bch_fs *c = i->c;
763 struct btree_transaction_stats *s = &c->btree_transaction_stats[i->iter];
772 if (i->iter == ARRAY_SIZE(bch2_btree_transaction_fns) ||
773 !bch2_btree_transaction_fns[i->iter])
776 prt_printf(&i->buf, "%s: ", bch2_btree_transaction_fns[i->iter]);
777 prt_newline(&i->buf);
778 printbuf_indent_add(&i->buf, 2);
780 mutex_lock(&s->lock);
782 prt_printf(&i->buf, "Max mem used: %u", s->max_mem);
783 prt_newline(&i->buf);
785 if (IS_ENABLED(CONFIG_BCACHEFS_LOCK_TIME_STATS)) {
786 prt_printf(&i->buf, "Lock hold times:");
787 prt_newline(&i->buf);
789 printbuf_indent_add(&i->buf, 2);
790 bch2_time_stats_to_text(&i->buf, &s->lock_hold_times);
791 printbuf_indent_sub(&i->buf, 2);
794 if (s->max_paths_text) {
795 prt_printf(&i->buf, "Maximum allocated btree paths (%u):", s->nr_max_paths);
796 prt_newline(&i->buf);
798 printbuf_indent_add(&i->buf, 2);
799 prt_str_indented(&i->buf, s->max_paths_text);
800 printbuf_indent_sub(&i->buf, 2);
803 mutex_unlock(&s->lock);
805 printbuf_indent_sub(&i->buf, 2);
806 prt_newline(&i->buf);
810 if (i->buf.allocation_failure)
816 static const struct file_operations lock_held_stats_op = {
817 .owner = THIS_MODULE,
818 .open = lock_held_stats_open,
819 .release = lock_held_stats_release,
820 .read = lock_held_stats_read,
823 static ssize_t bch2_btree_deadlock_read(struct file *file, char __user *buf,
824 size_t size, loff_t *ppos)
826 struct dump_iter *i = file->private_data;
827 struct bch_fs *c = i->c;
828 struct btree_trans *trans;
839 seqmutex_lock(&c->btree_trans_lock);
840 list_for_each_entry(trans, &c->btree_trans_list, list) {
841 if (trans->locking_wait.task->pid <= i->iter)
844 closure_get(&trans->ref);
845 seq = seqmutex_seq(&c->btree_trans_lock);
846 seqmutex_unlock(&c->btree_trans_lock);
850 closure_put(&trans->ref);
854 bch2_check_for_deadlock(trans, &i->buf);
856 i->iter = trans->locking_wait.task->pid;
858 closure_put(&trans->ref);
860 if (!seqmutex_relock(&c->btree_trans_lock, seq))
863 seqmutex_unlock(&c->btree_trans_lock);
865 if (i->buf.allocation_failure)
871 return ret ?: i->ret;
874 static const struct file_operations btree_deadlock_ops = {
875 .owner = THIS_MODULE,
876 .open = bch2_dump_open,
877 .release = bch2_dump_release,
878 .read = bch2_btree_deadlock_read,
881 void bch2_fs_debug_exit(struct bch_fs *c)
883 if (!IS_ERR_OR_NULL(c->fs_debug_dir))
884 debugfs_remove_recursive(c->fs_debug_dir);
887 void bch2_fs_debug_init(struct bch_fs *c)
889 struct btree_debug *bd;
892 if (IS_ERR_OR_NULL(bch_debug))
895 snprintf(name, sizeof(name), "%pU", c->sb.user_uuid.b);
896 c->fs_debug_dir = debugfs_create_dir(name, bch_debug);
897 if (IS_ERR_OR_NULL(c->fs_debug_dir))
900 debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
901 c->btree_debug, &cached_btree_nodes_ops);
903 #ifdef CONFIG_BCACHEFS_DEBUG_TRANSACTIONS
904 debugfs_create_file("btree_transactions", 0400, c->fs_debug_dir,
905 c->btree_debug, &btree_transactions_ops);
908 debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
909 c->btree_debug, &journal_pins_ops);
911 debugfs_create_file("btree_transaction_stats", 0400, c->fs_debug_dir,
912 c, &lock_held_stats_op);
914 debugfs_create_file("btree_deadlock", 0400, c->fs_debug_dir,
915 c->btree_debug, &btree_deadlock_ops);
917 c->btree_debug_dir = debugfs_create_dir("btrees", c->fs_debug_dir);
918 if (IS_ERR_OR_NULL(c->btree_debug_dir))
921 for (bd = c->btree_debug;
922 bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
924 bd->id = bd - c->btree_debug;
925 debugfs_create_file(bch2_btree_ids[bd->id],
926 0400, c->btree_debug_dir, bd,
929 snprintf(name, sizeof(name), "%s-formats",
930 bch2_btree_ids[bd->id]);
932 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
933 &btree_format_debug_ops);
935 snprintf(name, sizeof(name), "%s-bfloat-failed",
936 bch2_btree_ids[bd->id]);
938 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
939 &bfloat_failed_debug_ops);
945 void bch2_debug_exit(void)
947 if (!IS_ERR_OR_NULL(bch_debug))
948 debugfs_remove_recursive(bch_debug);
951 int __init bch2_debug_init(void)
955 bch_debug = debugfs_create_dir("bcachefs", NULL);