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/pretty-printers.h>
29 #include <linux/random.h>
30 #include <linux/seq_file.h>
32 static struct dentry *bch_debug;
34 static bool bch2_btree_verify_replica(struct bch_fs *c, struct btree *b,
35 struct extent_ptr_decoded pick)
37 struct btree *v = c->verify_data;
38 struct btree_node *n_ondisk = c->verify_ondisk;
39 struct btree_node *n_sorted = c->verify_data->data;
40 struct bset *sorted, *inmemory = &b->data->keys;
41 struct bch_dev *ca = bch_dev_bkey_exists(c, pick.ptr.dev);
45 if (!bch2_dev_get_ioref(ca, READ))
48 bio = bio_alloc_bioset(ca->disk_sb.bdev,
49 buf_pages(n_sorted, btree_bytes(c)),
53 bio->bi_iter.bi_sector = pick.ptr.offset;
54 bch2_bio_map(bio, n_sorted, btree_bytes(c));
59 percpu_ref_put(&ca->io_ref);
61 memcpy(n_ondisk, n_sorted, btree_bytes(c));
64 if (bch2_btree_node_read_done(c, ca, v, false))
67 n_sorted = c->verify_data->data;
68 sorted = &n_sorted->keys;
70 if (inmemory->u64s != sorted->u64s ||
71 memcmp(inmemory->start,
73 vstruct_end(inmemory) - (void *) inmemory->start)) {
74 unsigned offset = 0, sectors;
80 printk(KERN_ERR "*** in memory:\n");
81 bch2_dump_bset(c, b, inmemory, 0);
83 printk(KERN_ERR "*** read back in:\n");
84 bch2_dump_bset(c, v, sorted, 0);
86 while (offset < v->written) {
89 sectors = vstruct_blocks(n_ondisk, c->block_bits) <<
92 struct btree_node_entry *bne =
93 (void *) n_ondisk + (offset << 9);
96 sectors = vstruct_blocks(bne, c->block_bits) <<
100 printk(KERN_ERR "*** on disk block %u:\n", offset);
101 bch2_dump_bset(c, b, i, offset);
106 for (j = 0; j < le16_to_cpu(inmemory->u64s); j++)
107 if (inmemory->_data[j] != sorted->_data[j])
111 bch_err(c, "verify failed at key %u", j);
116 if (v->written != b->written) {
117 bch_err(c, "written wrong: expected %u, got %u",
118 b->written, v->written);
125 void __bch2_btree_verify(struct bch_fs *c, struct btree *b)
127 struct bkey_ptrs_c ptrs;
128 struct extent_ptr_decoded p;
129 const union bch_extent_entry *entry;
131 struct bset *inmemory = &b->data->keys;
132 struct bkey_packed *k;
135 if (c->opts.nochanges)
138 bch2_btree_node_io_lock(b);
139 mutex_lock(&c->verify_lock);
141 if (!c->verify_ondisk) {
142 c->verify_ondisk = kvpmalloc(btree_bytes(c), GFP_KERNEL);
143 if (!c->verify_ondisk)
147 if (!c->verify_data) {
148 c->verify_data = __bch2_btree_node_mem_alloc(c);
152 list_del_init(&c->verify_data->list);
155 BUG_ON(b->nsets != 1);
157 for (k = inmemory->start; k != vstruct_last(inmemory); k = bkey_next(k))
158 if (k->type == KEY_TYPE_btree_ptr_v2) {
159 struct bch_btree_ptr_v2 *v = (void *) bkeyp_val(&b->format, k);
164 bkey_copy(&v->key, &b->key);
165 v->c.level = b->c.level;
166 v->c.btree_id = b->c.btree_id;
167 bch2_btree_keys_init(v);
169 ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(&b->key));
170 bkey_for_each_ptr_decode(&b->key.k, ptrs, p, entry)
171 failed |= bch2_btree_verify_replica(c, b, p);
174 struct printbuf buf = PRINTBUF;
176 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
177 bch2_fs_fatal_error(c, "btree node verify failed for : %s\n", buf.buf);
181 mutex_unlock(&c->verify_lock);
182 bch2_btree_node_io_unlock(b);
185 #ifdef CONFIG_DEBUG_FS
187 /* XXX: bch_fs refcounting */
193 struct bpos prev_node;
198 char __user *ubuf; /* destination user buffer */
199 size_t size; /* size of requested read */
200 ssize_t ret; /* bytes read so far */
203 static ssize_t flush_buf(struct dump_iter *i)
206 size_t bytes = min_t(size_t, i->buf.pos, i->size);
207 int err = copy_to_user(i->ubuf, i->buf.buf, bytes);
216 memmove(i->buf.buf, i->buf.buf + bytes, i->buf.pos);
219 return i->size ? 0 : i->ret;
222 static int bch2_dump_open(struct inode *inode, struct file *file)
224 struct btree_debug *bd = inode->i_private;
227 i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
231 file->private_data = i;
234 i->c = container_of(bd, struct bch_fs, btree_debug[bd->id]);
241 static int bch2_dump_release(struct inode *inode, struct file *file)
243 struct dump_iter *i = file->private_data;
245 printbuf_exit(&i->buf);
250 static ssize_t bch2_read_btree(struct file *file, char __user *buf,
251 size_t size, loff_t *ppos)
253 struct dump_iter *i = file->private_data;
254 struct btree_trans trans;
255 struct btree_iter iter;
263 bch2_trans_init(&trans, i->c, 0, 0);
265 ret = for_each_btree_key2(&trans, iter, i->id, i->from,
267 BTREE_ITER_ALL_SNAPSHOTS, k, ({
272 bch2_bkey_val_to_text(&i->buf, i->c, k);
273 prt_newline(&i->buf);
281 bch2_trans_exit(&trans);
283 return ret ?: i->ret;
286 static const struct file_operations btree_debug_ops = {
287 .owner = THIS_MODULE,
288 .open = bch2_dump_open,
289 .release = bch2_dump_release,
290 .read = bch2_read_btree,
293 static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
294 size_t size, loff_t *ppos)
296 struct dump_iter *i = file->private_data;
297 struct btree_trans trans;
298 struct btree_iter iter;
310 if (!bpos_cmp(SPOS_MAX, i->from))
313 bch2_trans_init(&trans, i->c, 0, 0);
315 for_each_btree_node(&trans, iter, i->id, i->from, 0, b, ret) {
320 bch2_btree_node_to_text(&i->buf, i->c, b);
321 i->from = bpos_cmp(SPOS_MAX, b->key.k.p)
322 ? bpos_successor(b->key.k.p)
325 bch2_trans_iter_exit(&trans, &iter);
327 bch2_trans_exit(&trans);
332 return ret ?: i->ret;
335 static const struct file_operations btree_format_debug_ops = {
336 .owner = THIS_MODULE,
337 .open = bch2_dump_open,
338 .release = bch2_dump_release,
339 .read = bch2_read_btree_formats,
342 static ssize_t bch2_read_bfloat_failed(struct file *file, char __user *buf,
343 size_t size, loff_t *ppos)
345 struct dump_iter *i = file->private_data;
346 struct btree_trans trans;
347 struct btree_iter iter;
359 bch2_trans_init(&trans, i->c, 0, 0);
361 ret = for_each_btree_key2(&trans, iter, i->id, i->from,
363 BTREE_ITER_ALL_SNAPSHOTS, k, ({
364 struct btree_path_level *l = &iter.path->l[0];
365 struct bkey_packed *_k =
366 bch2_btree_node_iter_peek(&l->iter, l->b);
372 if (bpos_cmp(l->b->key.k.p, i->prev_node) > 0) {
373 bch2_btree_node_to_text(&i->buf, i->c, l->b);
374 i->prev_node = l->b->key.k.p;
377 bch2_bfloat_to_text(&i->buf, l->b, _k);
382 bch2_trans_exit(&trans);
387 return ret ?: i->ret;
390 static const struct file_operations bfloat_failed_debug_ops = {
391 .owner = THIS_MODULE,
392 .open = bch2_dump_open,
393 .release = bch2_dump_release,
394 .read = bch2_read_bfloat_failed,
397 static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
400 if (!out->nr_tabstops)
401 printbuf_tabstop_push(out, 32);
403 prt_printf(out, "%px btree=%s l=%u ",
405 bch2_btree_ids[b->c.btree_id],
409 printbuf_indent_add(out, 2);
411 bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
414 prt_printf(out, "flags: ");
416 prt_bitflags(out, bch2_btree_node_flags, b->flags);
419 prt_printf(out, "pcpu read locks: ");
421 prt_printf(out, "%u", b->c.lock.readers != NULL);
424 prt_printf(out, "written:");
426 prt_printf(out, "%u", b->written);
429 prt_printf(out, "writes blocked:");
431 prt_printf(out, "%u", !list_empty_careful(&b->write_blocked));
434 prt_printf(out, "will make reachable:");
436 prt_printf(out, "%lx", b->will_make_reachable);
439 prt_printf(out, "journal pin %px:", &b->writes[0].journal);
441 prt_printf(out, "%llu", b->writes[0].journal.seq);
444 prt_printf(out, "journal pin %px:", &b->writes[1].journal);
446 prt_printf(out, "%llu", b->writes[1].journal.seq);
449 printbuf_indent_sub(out, 2);
452 static ssize_t bch2_cached_btree_nodes_read(struct file *file, char __user *buf,
453 size_t size, loff_t *ppos)
455 struct dump_iter *i = file->private_data;
456 struct bch_fs *c = i->c;
465 struct bucket_table *tbl;
466 struct rhash_head *pos;
475 tbl = rht_dereference_rcu(c->btree_cache.table.tbl,
476 &c->btree_cache.table);
477 if (i->iter < tbl->size) {
478 rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
479 bch2_cached_btree_node_to_text(&i->buf, c, b);
488 if (i->buf.allocation_failure)
494 return ret ?: i->ret;
497 static const struct file_operations cached_btree_nodes_ops = {
498 .owner = THIS_MODULE,
499 .open = bch2_dump_open,
500 .release = bch2_dump_release,
501 .read = bch2_cached_btree_nodes_read,
504 static int prt_backtrace(struct printbuf *out, struct task_struct *task)
506 unsigned long entries[32];
507 unsigned i, nr_entries;
510 ret = down_read_killable(&task->signal->exec_update_lock);
514 nr_entries = stack_trace_save_tsk(task, entries, ARRAY_SIZE(entries), 0);
515 for (i = 0; i < nr_entries; i++) {
516 prt_printf(out, "[<0>] %pB", (void *)entries[i]);
520 up_read(&task->signal->exec_update_lock);
524 static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
525 size_t size, loff_t *ppos)
527 struct dump_iter *i = file->private_data;
528 struct bch_fs *c = i->c;
529 struct btree_trans *trans;
536 mutex_lock(&c->btree_trans_lock);
537 list_for_each_entry(trans, &c->btree_trans_list, list) {
538 if (trans->locking_wait.task->pid <= i->iter)
545 bch2_btree_trans_to_text(&i->buf, trans);
547 prt_printf(&i->buf, "backtrace:");
548 prt_newline(&i->buf);
549 printbuf_indent_add(&i->buf, 2);
550 prt_backtrace(&i->buf, trans->locking_wait.task);
551 printbuf_indent_sub(&i->buf, 2);
552 prt_newline(&i->buf);
554 i->iter = trans->locking_wait.task->pid;
556 mutex_unlock(&c->btree_trans_lock);
558 if (i->buf.allocation_failure)
564 return ret ?: i->ret;
567 static const struct file_operations btree_transactions_ops = {
568 .owner = THIS_MODULE,
569 .open = bch2_dump_open,
570 .release = bch2_dump_release,
571 .read = bch2_btree_transactions_read,
574 static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
575 size_t size, loff_t *ppos)
577 struct dump_iter *i = file->private_data;
578 struct bch_fs *c = i->c;
594 done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
598 if (i->buf.allocation_failure)
604 static const struct file_operations journal_pins_ops = {
605 .owner = THIS_MODULE,
606 .open = bch2_dump_open,
607 .release = bch2_dump_release,
608 .read = bch2_journal_pins_read,
611 static int lock_held_stats_open(struct inode *inode, struct file *file)
613 struct bch_fs *c = inode->i_private;
616 i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
624 file->private_data = i;
629 static int lock_held_stats_release(struct inode *inode, struct file *file)
631 struct dump_iter *i = file->private_data;
633 printbuf_exit(&i->buf);
639 static ssize_t lock_held_stats_read(struct file *file, char __user *buf,
640 size_t size, loff_t *ppos)
642 struct dump_iter *i = file->private_data;
643 struct bch_fs *c = i->c;
651 struct btree_transaction_stats *s = &c->btree_transaction_stats[i->iter];
660 if (i->iter == ARRAY_SIZE(c->btree_transaction_fns) ||
661 !c->btree_transaction_fns[i->iter])
664 prt_printf(&i->buf, "%s: ", c->btree_transaction_fns[i->iter]);
665 prt_newline(&i->buf);
666 printbuf_indent_add(&i->buf, 2);
668 mutex_lock(&s->lock);
670 prt_printf(&i->buf, "Max mem used: %u", s->max_mem);
671 prt_newline(&i->buf);
673 if (IS_ENABLED(CONFIG_BCACHEFS_LOCK_TIME_STATS)) {
674 prt_printf(&i->buf, "Lock hold times:");
675 prt_newline(&i->buf);
677 printbuf_indent_add(&i->buf, 2);
678 bch2_time_stats_to_text(&i->buf, &s->lock_hold_times);
679 printbuf_indent_sub(&i->buf, 2);
682 if (s->max_paths_text) {
683 prt_printf(&i->buf, "Maximum allocated btree paths (%u):", s->nr_max_paths);
684 prt_newline(&i->buf);
686 printbuf_indent_add(&i->buf, 2);
687 prt_str_indented(&i->buf, s->max_paths_text);
688 printbuf_indent_sub(&i->buf, 2);
691 mutex_unlock(&s->lock);
693 printbuf_indent_sub(&i->buf, 2);
694 prt_newline(&i->buf);
698 if (i->buf.allocation_failure)
704 static const struct file_operations lock_held_stats_op = {
705 .owner = THIS_MODULE,
706 .open = lock_held_stats_open,
707 .release = lock_held_stats_release,
708 .read = lock_held_stats_read,
711 static ssize_t bch2_btree_deadlock_read(struct file *file, char __user *buf,
712 size_t size, loff_t *ppos)
714 struct dump_iter *i = file->private_data;
715 struct bch_fs *c = i->c;
716 struct btree_trans *trans;
726 mutex_lock(&c->btree_trans_lock);
727 list_for_each_entry(trans, &c->btree_trans_list, list)
728 if (bch2_check_for_deadlock(trans, &i->buf)) {
732 mutex_unlock(&c->btree_trans_lock);
734 if (i->buf.allocation_failure)
740 return ret ?: i->ret;
743 static const struct file_operations btree_deadlock_ops = {
744 .owner = THIS_MODULE,
745 .open = bch2_dump_open,
746 .release = bch2_dump_release,
747 .read = bch2_btree_deadlock_read,
750 void bch2_fs_debug_exit(struct bch_fs *c)
752 if (!IS_ERR_OR_NULL(c->fs_debug_dir))
753 debugfs_remove_recursive(c->fs_debug_dir);
756 void bch2_fs_debug_init(struct bch_fs *c)
758 struct btree_debug *bd;
761 if (IS_ERR_OR_NULL(bch_debug))
764 snprintf(name, sizeof(name), "%pU", c->sb.user_uuid.b);
765 c->fs_debug_dir = debugfs_create_dir(name, bch_debug);
766 if (IS_ERR_OR_NULL(c->fs_debug_dir))
769 debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
770 c->btree_debug, &cached_btree_nodes_ops);
772 debugfs_create_file("btree_transactions", 0400, c->fs_debug_dir,
773 c->btree_debug, &btree_transactions_ops);
775 debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
776 c->btree_debug, &journal_pins_ops);
778 debugfs_create_file("btree_transaction_stats", 0400, c->fs_debug_dir,
779 c, &lock_held_stats_op);
781 debugfs_create_file("btree_deadlock", 0400, c->fs_debug_dir,
782 c->btree_debug, &btree_deadlock_ops);
784 c->btree_debug_dir = debugfs_create_dir("btrees", c->fs_debug_dir);
785 if (IS_ERR_OR_NULL(c->btree_debug_dir))
788 for (bd = c->btree_debug;
789 bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
791 bd->id = bd - c->btree_debug;
792 debugfs_create_file(bch2_btree_ids[bd->id],
793 0400, c->btree_debug_dir, bd,
796 snprintf(name, sizeof(name), "%s-formats",
797 bch2_btree_ids[bd->id]);
799 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
800 &btree_format_debug_ops);
802 snprintf(name, sizeof(name), "%s-bfloat-failed",
803 bch2_btree_ids[bd->id]);
805 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
806 &bfloat_failed_debug_ops);
812 void bch2_debug_exit(void)
814 if (!IS_ERR_OR_NULL(bch_debug))
815 debugfs_remove_recursive(bch_debug);
818 int __init bch2_debug_init(void)
822 bch_debug = debugfs_create_dir("bcachefs", NULL);