xref: /linux/fs/bcachefs/debug.c (revision c6144a21169fe7d0d70f1a0dae6f6301e5918d30)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Assorted bcachefs debug code
4  *
5  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6  * Copyright 2012 Google, Inc.
7  */
8 
9 #include "bcachefs.h"
10 #include "bkey_methods.h"
11 #include "btree_cache.h"
12 #include "btree_io.h"
13 #include "btree_iter.h"
14 #include "btree_locking.h"
15 #include "btree_update.h"
16 #include "btree_update_interior.h"
17 #include "buckets.h"
18 #include "debug.h"
19 #include "error.h"
20 #include "extents.h"
21 #include "fsck.h"
22 #include "inode.h"
23 #include "super.h"
24 
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>
30 
31 static struct dentry *bch_debug;
32 
33 static bool bch2_btree_verify_replica(struct bch_fs *c, struct btree *b,
34 				      struct extent_ptr_decoded pick)
35 {
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 bio *bio;
41 	bool failed = false, saw_error = false;
42 
43 	struct bch_dev *ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ);
44 	if (!ca)
45 		return false;
46 
47 	bio = bio_alloc_bioset(ca->disk_sb.bdev,
48 			       buf_pages(n_sorted, btree_buf_bytes(b)),
49 			       REQ_OP_READ|REQ_META,
50 			       GFP_NOFS,
51 			       &c->btree_bio);
52 	bio->bi_iter.bi_sector	= pick.ptr.offset;
53 	bch2_bio_map(bio, n_sorted, btree_buf_bytes(b));
54 
55 	submit_bio_wait(bio);
56 
57 	bio_put(bio);
58 	percpu_ref_put(&ca->io_ref);
59 
60 	memcpy(n_ondisk, n_sorted, btree_buf_bytes(b));
61 
62 	v->written = 0;
63 	if (bch2_btree_node_read_done(c, ca, v, false, &saw_error) || saw_error)
64 		return false;
65 
66 	n_sorted = c->verify_data->data;
67 	sorted = &n_sorted->keys;
68 
69 	if (inmemory->u64s != sorted->u64s ||
70 	    memcmp(inmemory->start,
71 		   sorted->start,
72 		   vstruct_end(inmemory) - (void *) inmemory->start)) {
73 		unsigned offset = 0, sectors;
74 		struct bset *i;
75 		unsigned j;
76 
77 		console_lock();
78 
79 		printk(KERN_ERR "*** in memory:\n");
80 		bch2_dump_bset(c, b, inmemory, 0);
81 
82 		printk(KERN_ERR "*** read back in:\n");
83 		bch2_dump_bset(c, v, sorted, 0);
84 
85 		while (offset < v->written) {
86 			if (!offset) {
87 				i = &n_ondisk->keys;
88 				sectors = vstruct_blocks(n_ondisk, c->block_bits) <<
89 					c->block_bits;
90 			} else {
91 				struct btree_node_entry *bne =
92 					(void *) n_ondisk + (offset << 9);
93 				i = &bne->keys;
94 
95 				sectors = vstruct_blocks(bne, c->block_bits) <<
96 					c->block_bits;
97 			}
98 
99 			printk(KERN_ERR "*** on disk block %u:\n", offset);
100 			bch2_dump_bset(c, b, i, offset);
101 
102 			offset += sectors;
103 		}
104 
105 		for (j = 0; j < le16_to_cpu(inmemory->u64s); j++)
106 			if (inmemory->_data[j] != sorted->_data[j])
107 				break;
108 
109 		console_unlock();
110 		bch_err(c, "verify failed at key %u", j);
111 
112 		failed = true;
113 	}
114 
115 	if (v->written != b->written) {
116 		bch_err(c, "written wrong: expected %u, got %u",
117 			b->written, v->written);
118 		failed = true;
119 	}
120 
121 	return failed;
122 }
123 
124 void __bch2_btree_verify(struct bch_fs *c, struct btree *b)
125 {
126 	struct bkey_ptrs_c ptrs;
127 	struct extent_ptr_decoded p;
128 	const union bch_extent_entry *entry;
129 	struct btree *v;
130 	struct bset *inmemory = &b->data->keys;
131 	struct bkey_packed *k;
132 	bool failed = false;
133 
134 	if (c->opts.nochanges)
135 		return;
136 
137 	bch2_btree_node_io_lock(b);
138 	mutex_lock(&c->verify_lock);
139 
140 	if (!c->verify_ondisk) {
141 		c->verify_ondisk = kvmalloc(btree_buf_bytes(b), GFP_KERNEL);
142 		if (!c->verify_ondisk)
143 			goto out;
144 	}
145 
146 	if (!c->verify_data) {
147 		c->verify_data = __bch2_btree_node_mem_alloc(c);
148 		if (!c->verify_data)
149 			goto out;
150 
151 		list_del_init(&c->verify_data->list);
152 	}
153 
154 	BUG_ON(b->nsets != 1);
155 
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 *) bkeyp_val(&b->format, k))->mem_ptr = 0;
159 
160 	v = c->verify_data;
161 	bkey_copy(&v->key, &b->key);
162 	v->c.level	= b->c.level;
163 	v->c.btree_id	= b->c.btree_id;
164 	bch2_btree_keys_init(v);
165 
166 	ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(&b->key));
167 	bkey_for_each_ptr_decode(&b->key.k, ptrs, p, entry)
168 		failed |= bch2_btree_verify_replica(c, b, p);
169 
170 	if (failed) {
171 		struct printbuf buf = PRINTBUF;
172 
173 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
174 		bch2_fs_fatal_error(c, ": btree node verify failed for: %s\n", buf.buf);
175 		printbuf_exit(&buf);
176 	}
177 out:
178 	mutex_unlock(&c->verify_lock);
179 	bch2_btree_node_io_unlock(b);
180 }
181 
182 void bch2_btree_node_ondisk_to_text(struct printbuf *out, struct bch_fs *c,
183 				    const struct btree *b)
184 {
185 	struct btree_node *n_ondisk = NULL;
186 	struct extent_ptr_decoded pick;
187 	struct bch_dev *ca;
188 	struct bio *bio = NULL;
189 	unsigned offset = 0;
190 	int ret;
191 
192 	if (bch2_bkey_pick_read_device(c, bkey_i_to_s_c(&b->key), NULL, &pick) <= 0) {
193 		prt_printf(out, "error getting device to read from: invalid device\n");
194 		return;
195 	}
196 
197 	ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ);
198 	if (!ca) {
199 		prt_printf(out, "error getting device to read from: not online\n");
200 		return;
201 	}
202 
203 	n_ondisk = kvmalloc(btree_buf_bytes(b), GFP_KERNEL);
204 	if (!n_ondisk) {
205 		prt_printf(out, "memory allocation failure\n");
206 		goto out;
207 	}
208 
209 	bio = bio_alloc_bioset(ca->disk_sb.bdev,
210 			       buf_pages(n_ondisk, btree_buf_bytes(b)),
211 			       REQ_OP_READ|REQ_META,
212 			       GFP_NOFS,
213 			       &c->btree_bio);
214 	bio->bi_iter.bi_sector	= pick.ptr.offset;
215 	bch2_bio_map(bio, n_ondisk, btree_buf_bytes(b));
216 
217 	ret = submit_bio_wait(bio);
218 	if (ret) {
219 		prt_printf(out, "IO error reading btree node: %s\n", bch2_err_str(ret));
220 		goto out;
221 	}
222 
223 	while (offset < btree_sectors(c)) {
224 		struct bset *i;
225 		struct nonce nonce;
226 		struct bch_csum csum;
227 		struct bkey_packed *k;
228 		unsigned sectors;
229 
230 		if (!offset) {
231 			i = &n_ondisk->keys;
232 
233 			if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
234 				prt_printf(out, "unknown checksum type at offset %u: %llu\n",
235 					   offset, BSET_CSUM_TYPE(i));
236 				goto out;
237 			}
238 
239 			nonce = btree_nonce(i, offset << 9);
240 			csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, n_ondisk);
241 
242 			if (bch2_crc_cmp(csum, n_ondisk->csum)) {
243 				prt_printf(out, "invalid checksum\n");
244 				goto out;
245 			}
246 
247 			bset_encrypt(c, i, offset << 9);
248 
249 			sectors = vstruct_sectors(n_ondisk, c->block_bits);
250 		} else {
251 			struct btree_node_entry *bne = (void *) n_ondisk + (offset << 9);
252 
253 			i = &bne->keys;
254 
255 			if (i->seq != n_ondisk->keys.seq)
256 				break;
257 
258 			if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
259 				prt_printf(out, "unknown checksum type at offset %u: %llu\n",
260 					   offset, BSET_CSUM_TYPE(i));
261 				goto out;
262 			}
263 
264 			nonce = btree_nonce(i, offset << 9);
265 			csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne);
266 
267 			if (bch2_crc_cmp(csum, bne->csum)) {
268 				prt_printf(out, "invalid checksum");
269 				goto out;
270 			}
271 
272 			bset_encrypt(c, i, offset << 9);
273 
274 			sectors = vstruct_sectors(bne, c->block_bits);
275 		}
276 
277 		prt_printf(out, "  offset %u version %u, journal seq %llu\n",
278 			   offset,
279 			   le16_to_cpu(i->version),
280 			   le64_to_cpu(i->journal_seq));
281 		offset += sectors;
282 
283 		printbuf_indent_add(out, 4);
284 
285 		for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k)) {
286 			struct bkey u;
287 
288 			bch2_bkey_val_to_text(out, c, bkey_disassemble(b, k, &u));
289 			prt_newline(out);
290 		}
291 
292 		printbuf_indent_sub(out, 4);
293 	}
294 out:
295 	if (bio)
296 		bio_put(bio);
297 	kvfree(n_ondisk);
298 	percpu_ref_put(&ca->io_ref);
299 }
300 
301 #ifdef CONFIG_DEBUG_FS
302 
303 /* XXX: bch_fs refcounting */
304 
305 struct dump_iter {
306 	struct bch_fs		*c;
307 	enum btree_id		id;
308 	struct bpos		from;
309 	struct bpos		prev_node;
310 	u64			iter;
311 
312 	struct printbuf		buf;
313 
314 	char __user		*ubuf;	/* destination user buffer */
315 	size_t			size;	/* size of requested read */
316 	ssize_t			ret;	/* bytes read so far */
317 };
318 
319 static ssize_t flush_buf(struct dump_iter *i)
320 {
321 	if (i->buf.pos) {
322 		size_t bytes = min_t(size_t, i->buf.pos, i->size);
323 		int copied = bytes - copy_to_user(i->ubuf, i->buf.buf, bytes);
324 
325 		i->ret	 += copied;
326 		i->ubuf	 += copied;
327 		i->size	 -= copied;
328 		i->buf.pos -= copied;
329 		memmove(i->buf.buf, i->buf.buf + copied, i->buf.pos);
330 
331 		if (copied != bytes)
332 			return -EFAULT;
333 	}
334 
335 	return i->size ? 0 : i->ret;
336 }
337 
338 static int bch2_dump_open(struct inode *inode, struct file *file)
339 {
340 	struct btree_debug *bd = inode->i_private;
341 	struct dump_iter *i;
342 
343 	i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
344 	if (!i)
345 		return -ENOMEM;
346 
347 	file->private_data = i;
348 	i->from = POS_MIN;
349 	i->iter	= 0;
350 	i->c	= container_of(bd, struct bch_fs, btree_debug[bd->id]);
351 	i->id	= bd->id;
352 	i->buf	= PRINTBUF;
353 
354 	return 0;
355 }
356 
357 static int bch2_dump_release(struct inode *inode, struct file *file)
358 {
359 	struct dump_iter *i = file->private_data;
360 
361 	printbuf_exit(&i->buf);
362 	kfree(i);
363 	return 0;
364 }
365 
366 static ssize_t bch2_read_btree(struct file *file, char __user *buf,
367 			       size_t size, loff_t *ppos)
368 {
369 	struct dump_iter *i = file->private_data;
370 
371 	i->ubuf = buf;
372 	i->size	= size;
373 	i->ret	= 0;
374 
375 	return flush_buf(i) ?:
376 		bch2_trans_run(i->c,
377 			for_each_btree_key(trans, iter, i->id, i->from,
378 					   BTREE_ITER_prefetch|
379 					   BTREE_ITER_all_snapshots, k, ({
380 				bch2_bkey_val_to_text(&i->buf, i->c, k);
381 				prt_newline(&i->buf);
382 				bch2_trans_unlock(trans);
383 				i->from = bpos_successor(iter.pos);
384 				flush_buf(i);
385 			}))) ?:
386 		i->ret;
387 }
388 
389 static const struct file_operations btree_debug_ops = {
390 	.owner		= THIS_MODULE,
391 	.open		= bch2_dump_open,
392 	.release	= bch2_dump_release,
393 	.read		= bch2_read_btree,
394 };
395 
396 static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
397 				       size_t size, loff_t *ppos)
398 {
399 	struct dump_iter *i = file->private_data;
400 	struct btree_trans *trans;
401 	struct btree_iter iter;
402 	struct btree *b;
403 	ssize_t ret;
404 
405 	i->ubuf = buf;
406 	i->size	= size;
407 	i->ret	= 0;
408 
409 	ret = flush_buf(i);
410 	if (ret)
411 		return ret;
412 
413 	if (bpos_eq(SPOS_MAX, i->from))
414 		return i->ret;
415 
416 	trans = bch2_trans_get(i->c);
417 retry:
418 	bch2_trans_begin(trans);
419 
420 	for_each_btree_node(trans, iter, i->id, i->from, 0, b, ret) {
421 		bch2_btree_node_to_text(&i->buf, i->c, b);
422 		i->from = !bpos_eq(SPOS_MAX, b->key.k.p)
423 			? bpos_successor(b->key.k.p)
424 			: b->key.k.p;
425 
426 		ret = drop_locks_do(trans, flush_buf(i));
427 		if (ret)
428 			break;
429 	}
430 	bch2_trans_iter_exit(trans, &iter);
431 
432 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
433 		goto retry;
434 
435 	bch2_trans_put(trans);
436 
437 	if (!ret)
438 		ret = flush_buf(i);
439 
440 	return ret ?: i->ret;
441 }
442 
443 static const struct file_operations btree_format_debug_ops = {
444 	.owner		= THIS_MODULE,
445 	.open		= bch2_dump_open,
446 	.release	= bch2_dump_release,
447 	.read		= bch2_read_btree_formats,
448 };
449 
450 static ssize_t bch2_read_bfloat_failed(struct file *file, char __user *buf,
451 				       size_t size, loff_t *ppos)
452 {
453 	struct dump_iter *i = file->private_data;
454 
455 	i->ubuf = buf;
456 	i->size	= size;
457 	i->ret	= 0;
458 
459 	return flush_buf(i) ?:
460 		bch2_trans_run(i->c,
461 			for_each_btree_key(trans, iter, i->id, i->from,
462 					   BTREE_ITER_prefetch|
463 					   BTREE_ITER_all_snapshots, k, ({
464 				struct btree_path_level *l =
465 					&btree_iter_path(trans, &iter)->l[0];
466 				struct bkey_packed *_k =
467 					bch2_btree_node_iter_peek(&l->iter, l->b);
468 
469 				if (bpos_gt(l->b->key.k.p, i->prev_node)) {
470 					bch2_btree_node_to_text(&i->buf, i->c, l->b);
471 					i->prev_node = l->b->key.k.p;
472 				}
473 
474 				bch2_bfloat_to_text(&i->buf, l->b, _k);
475 				bch2_trans_unlock(trans);
476 				i->from = bpos_successor(iter.pos);
477 				flush_buf(i);
478 			}))) ?:
479 		i->ret;
480 }
481 
482 static const struct file_operations bfloat_failed_debug_ops = {
483 	.owner		= THIS_MODULE,
484 	.open		= bch2_dump_open,
485 	.release	= bch2_dump_release,
486 	.read		= bch2_read_bfloat_failed,
487 };
488 
489 static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
490 					   struct btree *b)
491 {
492 	if (!out->nr_tabstops)
493 		printbuf_tabstop_push(out, 32);
494 
495 	prt_printf(out, "%px btree=%s l=%u\n", b, bch2_btree_id_str(b->c.btree_id), b->c.level);
496 
497 	printbuf_indent_add(out, 2);
498 
499 	bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
500 	prt_newline(out);
501 
502 	prt_printf(out, "flags:\t");
503 	prt_bitflags(out, bch2_btree_node_flags, b->flags);
504 	prt_newline(out);
505 
506 	prt_printf(out, "pcpu read locks:\t%u\n",	b->c.lock.readers != NULL);
507 	prt_printf(out, "written:\t%u\n",		b->written);
508 	prt_printf(out, "writes blocked:\t%u\n",	!list_empty_careful(&b->write_blocked));
509 	prt_printf(out, "will make reachable:\t%lx\n",	b->will_make_reachable);
510 
511 	prt_printf(out, "journal pin %px:\t%llu\n",
512 		   &b->writes[0].journal, b->writes[0].journal.seq);
513 	prt_printf(out, "journal pin %px:\t%llu\n",
514 		   &b->writes[1].journal, b->writes[1].journal.seq);
515 
516 	printbuf_indent_sub(out, 2);
517 }
518 
519 static ssize_t bch2_cached_btree_nodes_read(struct file *file, char __user *buf,
520 					    size_t size, loff_t *ppos)
521 {
522 	struct dump_iter *i = file->private_data;
523 	struct bch_fs *c = i->c;
524 	bool done = false;
525 	ssize_t ret = 0;
526 
527 	i->ubuf = buf;
528 	i->size	= size;
529 	i->ret	= 0;
530 
531 	do {
532 		struct bucket_table *tbl;
533 		struct rhash_head *pos;
534 		struct btree *b;
535 
536 		ret = flush_buf(i);
537 		if (ret)
538 			return ret;
539 
540 		rcu_read_lock();
541 		i->buf.atomic++;
542 		tbl = rht_dereference_rcu(c->btree_cache.table.tbl,
543 					  &c->btree_cache.table);
544 		if (i->iter < tbl->size) {
545 			rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
546 				bch2_cached_btree_node_to_text(&i->buf, c, b);
547 			i->iter++;
548 		} else {
549 			done = true;
550 		}
551 		--i->buf.atomic;
552 		rcu_read_unlock();
553 	} while (!done);
554 
555 	if (i->buf.allocation_failure)
556 		ret = -ENOMEM;
557 
558 	if (!ret)
559 		ret = flush_buf(i);
560 
561 	return ret ?: i->ret;
562 }
563 
564 static const struct file_operations cached_btree_nodes_ops = {
565 	.owner		= THIS_MODULE,
566 	.open		= bch2_dump_open,
567 	.release	= bch2_dump_release,
568 	.read		= bch2_cached_btree_nodes_read,
569 };
570 
571 static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
572 					    size_t size, loff_t *ppos)
573 {
574 	struct dump_iter *i = file->private_data;
575 	struct bch_fs *c = i->c;
576 	struct btree_trans *trans;
577 	ssize_t ret = 0;
578 	u32 seq;
579 
580 	i->ubuf = buf;
581 	i->size	= size;
582 	i->ret	= 0;
583 restart:
584 	seqmutex_lock(&c->btree_trans_lock);
585 	list_for_each_entry(trans, &c->btree_trans_list, list) {
586 		struct task_struct *task = READ_ONCE(trans->locking_wait.task);
587 
588 		if (!task || task->pid <= i->iter)
589 			continue;
590 
591 		closure_get(&trans->ref);
592 		seq = seqmutex_seq(&c->btree_trans_lock);
593 		seqmutex_unlock(&c->btree_trans_lock);
594 
595 		ret = flush_buf(i);
596 		if (ret) {
597 			closure_put(&trans->ref);
598 			goto unlocked;
599 		}
600 
601 		bch2_btree_trans_to_text(&i->buf, trans);
602 
603 		prt_printf(&i->buf, "backtrace:\n");
604 		printbuf_indent_add(&i->buf, 2);
605 		bch2_prt_task_backtrace(&i->buf, task, 0, GFP_KERNEL);
606 		printbuf_indent_sub(&i->buf, 2);
607 		prt_newline(&i->buf);
608 
609 		i->iter = task->pid;
610 
611 		closure_put(&trans->ref);
612 
613 		if (!seqmutex_relock(&c->btree_trans_lock, seq))
614 			goto restart;
615 	}
616 	seqmutex_unlock(&c->btree_trans_lock);
617 unlocked:
618 	if (i->buf.allocation_failure)
619 		ret = -ENOMEM;
620 
621 	if (!ret)
622 		ret = flush_buf(i);
623 
624 	return ret ?: i->ret;
625 }
626 
627 static const struct file_operations btree_transactions_ops = {
628 	.owner		= THIS_MODULE,
629 	.open		= bch2_dump_open,
630 	.release	= bch2_dump_release,
631 	.read		= bch2_btree_transactions_read,
632 };
633 
634 static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
635 				      size_t size, loff_t *ppos)
636 {
637 	struct dump_iter *i = file->private_data;
638 	struct bch_fs *c = i->c;
639 	bool done = false;
640 	int err;
641 
642 	i->ubuf = buf;
643 	i->size	= size;
644 	i->ret	= 0;
645 
646 	while (1) {
647 		err = flush_buf(i);
648 		if (err)
649 			return err;
650 
651 		if (!i->size)
652 			break;
653 
654 		if (done)
655 			break;
656 
657 		done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
658 		i->iter++;
659 	}
660 
661 	if (i->buf.allocation_failure)
662 		return -ENOMEM;
663 
664 	return i->ret;
665 }
666 
667 static const struct file_operations journal_pins_ops = {
668 	.owner		= THIS_MODULE,
669 	.open		= bch2_dump_open,
670 	.release	= bch2_dump_release,
671 	.read		= bch2_journal_pins_read,
672 };
673 
674 static ssize_t bch2_btree_updates_read(struct file *file, char __user *buf,
675 				       size_t size, loff_t *ppos)
676 {
677 	struct dump_iter *i = file->private_data;
678 	struct bch_fs *c = i->c;
679 	int err;
680 
681 	i->ubuf = buf;
682 	i->size	= size;
683 	i->ret	= 0;
684 
685 	if (!i->iter) {
686 		bch2_btree_updates_to_text(&i->buf, c);
687 		i->iter++;
688 	}
689 
690 	err = flush_buf(i);
691 	if (err)
692 		return err;
693 
694 	if (i->buf.allocation_failure)
695 		return -ENOMEM;
696 
697 	return i->ret;
698 }
699 
700 static const struct file_operations btree_updates_ops = {
701 	.owner		= THIS_MODULE,
702 	.open		= bch2_dump_open,
703 	.release	= bch2_dump_release,
704 	.read		= bch2_btree_updates_read,
705 };
706 
707 static int btree_transaction_stats_open(struct inode *inode, struct file *file)
708 {
709 	struct bch_fs *c = inode->i_private;
710 	struct dump_iter *i;
711 
712 	i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
713 	if (!i)
714 		return -ENOMEM;
715 
716 	i->iter = 1;
717 	i->c    = c;
718 	i->buf  = PRINTBUF;
719 	file->private_data = i;
720 
721 	return 0;
722 }
723 
724 static int btree_transaction_stats_release(struct inode *inode, struct file *file)
725 {
726 	struct dump_iter *i = file->private_data;
727 
728 	printbuf_exit(&i->buf);
729 	kfree(i);
730 
731 	return 0;
732 }
733 
734 static ssize_t btree_transaction_stats_read(struct file *file, char __user *buf,
735 					    size_t size, loff_t *ppos)
736 {
737 	struct dump_iter        *i = file->private_data;
738 	struct bch_fs *c = i->c;
739 	int err;
740 
741 	i->ubuf = buf;
742 	i->size = size;
743 	i->ret  = 0;
744 
745 	while (1) {
746 		struct btree_transaction_stats *s = &c->btree_transaction_stats[i->iter];
747 
748 		err = flush_buf(i);
749 		if (err)
750 			return err;
751 
752 		if (!i->size)
753 			break;
754 
755 		if (i->iter == ARRAY_SIZE(bch2_btree_transaction_fns) ||
756 		    !bch2_btree_transaction_fns[i->iter])
757 			break;
758 
759 		prt_printf(&i->buf, "%s:\n", bch2_btree_transaction_fns[i->iter]);
760 		printbuf_indent_add(&i->buf, 2);
761 
762 		mutex_lock(&s->lock);
763 
764 		prt_printf(&i->buf, "Max mem used: %u\n", s->max_mem);
765 		prt_printf(&i->buf, "Transaction duration:\n");
766 
767 		printbuf_indent_add(&i->buf, 2);
768 		bch2_time_stats_to_text(&i->buf, &s->duration);
769 		printbuf_indent_sub(&i->buf, 2);
770 
771 		if (IS_ENABLED(CONFIG_BCACHEFS_LOCK_TIME_STATS)) {
772 			prt_printf(&i->buf, "Lock hold times:\n");
773 
774 			printbuf_indent_add(&i->buf, 2);
775 			bch2_time_stats_to_text(&i->buf, &s->lock_hold_times);
776 			printbuf_indent_sub(&i->buf, 2);
777 		}
778 
779 		if (s->max_paths_text) {
780 			prt_printf(&i->buf, "Maximum allocated btree paths (%u):\n", s->nr_max_paths);
781 
782 			printbuf_indent_add(&i->buf, 2);
783 			prt_str_indented(&i->buf, s->max_paths_text);
784 			printbuf_indent_sub(&i->buf, 2);
785 		}
786 
787 		mutex_unlock(&s->lock);
788 
789 		printbuf_indent_sub(&i->buf, 2);
790 		prt_newline(&i->buf);
791 		i->iter++;
792 	}
793 
794 	if (i->buf.allocation_failure)
795 		return -ENOMEM;
796 
797 	return i->ret;
798 }
799 
800 static const struct file_operations btree_transaction_stats_op = {
801 	.owner		= THIS_MODULE,
802 	.open		= btree_transaction_stats_open,
803 	.release	= btree_transaction_stats_release,
804 	.read		= btree_transaction_stats_read,
805 };
806 
807 static ssize_t bch2_btree_deadlock_read(struct file *file, char __user *buf,
808 					    size_t size, loff_t *ppos)
809 {
810 	struct dump_iter *i = file->private_data;
811 	struct bch_fs *c = i->c;
812 	struct btree_trans *trans;
813 	ssize_t ret = 0;
814 	u32 seq;
815 
816 	i->ubuf = buf;
817 	i->size	= size;
818 	i->ret	= 0;
819 
820 	if (i->iter)
821 		goto out;
822 restart:
823 	seqmutex_lock(&c->btree_trans_lock);
824 	list_for_each_entry(trans, &c->btree_trans_list, list) {
825 		struct task_struct *task = READ_ONCE(trans->locking_wait.task);
826 
827 		if (!task || task->pid <= i->iter)
828 			continue;
829 
830 		closure_get(&trans->ref);
831 		seq = seqmutex_seq(&c->btree_trans_lock);
832 		seqmutex_unlock(&c->btree_trans_lock);
833 
834 		ret = flush_buf(i);
835 		if (ret) {
836 			closure_put(&trans->ref);
837 			goto out;
838 		}
839 
840 		bch2_check_for_deadlock(trans, &i->buf);
841 
842 		i->iter = task->pid;
843 
844 		closure_put(&trans->ref);
845 
846 		if (!seqmutex_relock(&c->btree_trans_lock, seq))
847 			goto restart;
848 	}
849 	seqmutex_unlock(&c->btree_trans_lock);
850 out:
851 	if (i->buf.allocation_failure)
852 		ret = -ENOMEM;
853 
854 	if (!ret)
855 		ret = flush_buf(i);
856 
857 	return ret ?: i->ret;
858 }
859 
860 static const struct file_operations btree_deadlock_ops = {
861 	.owner		= THIS_MODULE,
862 	.open		= bch2_dump_open,
863 	.release	= bch2_dump_release,
864 	.read		= bch2_btree_deadlock_read,
865 };
866 
867 void bch2_fs_debug_exit(struct bch_fs *c)
868 {
869 	if (!IS_ERR_OR_NULL(c->fs_debug_dir))
870 		debugfs_remove_recursive(c->fs_debug_dir);
871 }
872 
873 static void bch2_fs_debug_btree_init(struct bch_fs *c, struct btree_debug *bd)
874 {
875 	struct dentry *d;
876 
877 	d = debugfs_create_dir(bch2_btree_id_str(bd->id), c->btree_debug_dir);
878 
879 	debugfs_create_file("keys", 0400, d, bd, &btree_debug_ops);
880 
881 	debugfs_create_file("formats", 0400, d, bd, &btree_format_debug_ops);
882 
883 	debugfs_create_file("bfloat-failed", 0400, d, bd,
884 			    &bfloat_failed_debug_ops);
885 }
886 
887 void bch2_fs_debug_init(struct bch_fs *c)
888 {
889 	struct btree_debug *bd;
890 	char name[100];
891 
892 	if (IS_ERR_OR_NULL(bch_debug))
893 		return;
894 
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))
898 		return;
899 
900 	debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
901 			    c->btree_debug, &cached_btree_nodes_ops);
902 
903 	debugfs_create_file("btree_transactions", 0400, c->fs_debug_dir,
904 			    c->btree_debug, &btree_transactions_ops);
905 
906 	debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
907 			    c->btree_debug, &journal_pins_ops);
908 
909 	debugfs_create_file("btree_updates", 0400, c->fs_debug_dir,
910 			    c->btree_debug, &btree_updates_ops);
911 
912 	debugfs_create_file("btree_transaction_stats", 0400, c->fs_debug_dir,
913 			    c, &btree_transaction_stats_op);
914 
915 	debugfs_create_file("btree_deadlock", 0400, c->fs_debug_dir,
916 			    c->btree_debug, &btree_deadlock_ops);
917 
918 	c->btree_debug_dir = debugfs_create_dir("btrees", c->fs_debug_dir);
919 	if (IS_ERR_OR_NULL(c->btree_debug_dir))
920 		return;
921 
922 	for (bd = c->btree_debug;
923 	     bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
924 	     bd++) {
925 		bd->id = bd - c->btree_debug;
926 		bch2_fs_debug_btree_init(c, bd);
927 	}
928 }
929 
930 #endif
931 
932 void bch2_debug_exit(void)
933 {
934 	if (!IS_ERR_OR_NULL(bch_debug))
935 		debugfs_remove_recursive(bch_debug);
936 }
937 
938 int __init bch2_debug_init(void)
939 {
940 	bch_debug = debugfs_create_dir("bcachefs", NULL);
941 	return 0;
942 }
943