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 "alloc_foreground.h"
11 #include "async_objs.h"
12 #include "bkey_methods.h"
13 #include "btree_cache.h"
14 #include "btree_io.h"
15 #include "btree_iter.h"
16 #include "btree_locking.h"
17 #include "btree_update.h"
18 #include "btree_update_interior.h"
19 #include "buckets.h"
20 #include "data_update.h"
21 #include "debug.h"
22 #include "error.h"
23 #include "extents.h"
24 #include "fsck.h"
25 #include "inode.h"
26 #include "journal_reclaim.h"
27 #include "super.h"
28
29 #include <linux/console.h>
30 #include <linux/debugfs.h>
31 #include <linux/module.h>
32 #include <linux/random.h>
33 #include <linux/seq_file.h>
34
35 static struct dentry *bch_debug;
36
bch2_btree_verify_replica(struct bch_fs * c,struct btree * b,struct extent_ptr_decoded pick)37 static bool bch2_btree_verify_replica(struct bch_fs *c, struct btree *b,
38 struct extent_ptr_decoded pick)
39 {
40 struct btree *v = c->verify_data;
41 struct btree_node *n_ondisk = c->verify_ondisk;
42 struct btree_node *n_sorted = c->verify_data->data;
43 struct bset *sorted, *inmemory = &b->data->keys;
44 struct bio *bio;
45 bool failed = false;
46
47 struct bch_dev *ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ,
48 BCH_DEV_READ_REF_btree_verify_replicas);
49 if (!ca)
50 return false;
51
52 bio = bio_alloc_bioset(ca->disk_sb.bdev,
53 buf_pages(n_sorted, btree_buf_bytes(b)),
54 REQ_OP_READ|REQ_META,
55 GFP_NOFS,
56 &c->btree_bio);
57 bio->bi_iter.bi_sector = pick.ptr.offset;
58 bch2_bio_map(bio, n_sorted, btree_buf_bytes(b));
59
60 submit_bio_wait(bio);
61
62 bio_put(bio);
63 enumerated_ref_put(&ca->io_ref[READ],
64 BCH_DEV_READ_REF_btree_verify_replicas);
65
66 memcpy(n_ondisk, n_sorted, btree_buf_bytes(b));
67
68 v->written = 0;
69 if (bch2_btree_node_read_done(c, ca, v, NULL, NULL))
70 return false;
71
72 n_sorted = c->verify_data->data;
73 sorted = &n_sorted->keys;
74
75 if (inmemory->u64s != sorted->u64s ||
76 memcmp(inmemory->start,
77 sorted->start,
78 vstruct_end(inmemory) - (void *) inmemory->start)) {
79 unsigned offset = 0, sectors;
80 struct bset *i;
81 unsigned j;
82
83 console_lock();
84
85 printk(KERN_ERR "*** in memory:\n");
86 bch2_dump_bset(c, b, inmemory, 0);
87
88 printk(KERN_ERR "*** read back in:\n");
89 bch2_dump_bset(c, v, sorted, 0);
90
91 while (offset < v->written) {
92 if (!offset) {
93 i = &n_ondisk->keys;
94 sectors = vstruct_blocks(n_ondisk, c->block_bits) <<
95 c->block_bits;
96 } else {
97 struct btree_node_entry *bne =
98 (void *) n_ondisk + (offset << 9);
99 i = &bne->keys;
100
101 sectors = vstruct_blocks(bne, c->block_bits) <<
102 c->block_bits;
103 }
104
105 printk(KERN_ERR "*** on disk block %u:\n", offset);
106 bch2_dump_bset(c, b, i, offset);
107
108 offset += sectors;
109 }
110
111 for (j = 0; j < le16_to_cpu(inmemory->u64s); j++)
112 if (inmemory->_data[j] != sorted->_data[j])
113 break;
114
115 console_unlock();
116 bch_err(c, "verify failed at key %u", j);
117
118 failed = true;
119 }
120
121 if (v->written != b->written) {
122 bch_err(c, "written wrong: expected %u, got %u",
123 b->written, v->written);
124 failed = true;
125 }
126
127 return failed;
128 }
129
__bch2_btree_verify(struct bch_fs * c,struct btree * b)130 void __bch2_btree_verify(struct bch_fs *c, struct btree *b)
131 {
132 struct bkey_ptrs_c ptrs;
133 struct extent_ptr_decoded p;
134 const union bch_extent_entry *entry;
135 struct btree *v;
136 struct bset *inmemory = &b->data->keys;
137 struct bkey_packed *k;
138 bool failed = false;
139
140 if (c->opts.nochanges)
141 return;
142
143 bch2_btree_node_io_lock(b);
144 mutex_lock(&c->verify_lock);
145
146 if (!c->verify_ondisk) {
147 c->verify_ondisk = kvmalloc(btree_buf_bytes(b), GFP_KERNEL);
148 if (!c->verify_ondisk)
149 goto out;
150 }
151
152 if (!c->verify_data) {
153 c->verify_data = __bch2_btree_node_mem_alloc(c);
154 if (!c->verify_data)
155 goto out;
156 }
157
158 BUG_ON(b->nsets != 1);
159
160 for (k = inmemory->start; k != vstruct_last(inmemory); k = bkey_p_next(k))
161 if (k->type == KEY_TYPE_btree_ptr_v2)
162 ((struct bch_btree_ptr_v2 *) bkeyp_val(&b->format, k))->mem_ptr = 0;
163
164 v = c->verify_data;
165 bkey_copy(&v->key, &b->key);
166 v->c.level = b->c.level;
167 v->c.btree_id = b->c.btree_id;
168 bch2_btree_keys_init(v);
169
170 ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(&b->key));
171 bkey_for_each_ptr_decode(&b->key.k, ptrs, p, entry)
172 failed |= bch2_btree_verify_replica(c, b, p);
173
174 if (failed) {
175 struct printbuf buf = PRINTBUF;
176
177 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
178 bch2_fs_fatal_error(c, ": btree node verify failed for: %s\n", buf.buf);
179 printbuf_exit(&buf);
180 }
181 out:
182 mutex_unlock(&c->verify_lock);
183 bch2_btree_node_io_unlock(b);
184 }
185
bch2_btree_node_ondisk_to_text(struct printbuf * out,struct bch_fs * c,const struct btree * b)186 void bch2_btree_node_ondisk_to_text(struct printbuf *out, struct bch_fs *c,
187 const struct btree *b)
188 {
189 struct btree_node *n_ondisk = NULL;
190 struct extent_ptr_decoded pick;
191 struct bch_dev *ca;
192 struct bio *bio = NULL;
193 unsigned offset = 0;
194 int ret;
195
196 if (bch2_bkey_pick_read_device(c, bkey_i_to_s_c(&b->key), NULL, &pick, -1) <= 0) {
197 prt_printf(out, "error getting device to read from: invalid device\n");
198 return;
199 }
200
201 ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ,
202 BCH_DEV_READ_REF_btree_node_ondisk_to_text);
203 if (!ca) {
204 prt_printf(out, "error getting device to read from: not online\n");
205 return;
206 }
207
208 n_ondisk = kvmalloc(btree_buf_bytes(b), GFP_KERNEL);
209 if (!n_ondisk) {
210 prt_printf(out, "memory allocation failure\n");
211 goto out;
212 }
213
214 bio = bio_alloc_bioset(ca->disk_sb.bdev,
215 buf_pages(n_ondisk, btree_buf_bytes(b)),
216 REQ_OP_READ|REQ_META,
217 GFP_NOFS,
218 &c->btree_bio);
219 bio->bi_iter.bi_sector = pick.ptr.offset;
220 bch2_bio_map(bio, n_ondisk, btree_buf_bytes(b));
221
222 ret = submit_bio_wait(bio);
223 if (ret) {
224 prt_printf(out, "IO error reading btree node: %s\n", bch2_err_str(ret));
225 goto out;
226 }
227
228 while (offset < btree_sectors(c)) {
229 struct bset *i;
230 struct nonce nonce;
231 struct bch_csum csum;
232 struct bkey_packed *k;
233 unsigned sectors;
234
235 if (!offset) {
236 i = &n_ondisk->keys;
237
238 if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
239 prt_printf(out, "unknown checksum type at offset %u: %llu\n",
240 offset, BSET_CSUM_TYPE(i));
241 goto out;
242 }
243
244 nonce = btree_nonce(i, offset << 9);
245 csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, n_ondisk);
246
247 if (bch2_crc_cmp(csum, n_ondisk->csum)) {
248 prt_printf(out, "invalid checksum\n");
249 goto out;
250 }
251
252 bset_encrypt(c, i, offset << 9);
253
254 sectors = vstruct_sectors(n_ondisk, c->block_bits);
255 } else {
256 struct btree_node_entry *bne = (void *) n_ondisk + (offset << 9);
257
258 i = &bne->keys;
259
260 if (i->seq != n_ondisk->keys.seq)
261 break;
262
263 if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
264 prt_printf(out, "unknown checksum type at offset %u: %llu\n",
265 offset, BSET_CSUM_TYPE(i));
266 goto out;
267 }
268
269 nonce = btree_nonce(i, offset << 9);
270 csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne);
271
272 if (bch2_crc_cmp(csum, bne->csum)) {
273 prt_printf(out, "invalid checksum");
274 goto out;
275 }
276
277 bset_encrypt(c, i, offset << 9);
278
279 sectors = vstruct_sectors(bne, c->block_bits);
280 }
281
282 prt_printf(out, " offset %u version %u, journal seq %llu\n",
283 offset,
284 le16_to_cpu(i->version),
285 le64_to_cpu(i->journal_seq));
286 offset += sectors;
287
288 printbuf_indent_add(out, 4);
289
290 for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k)) {
291 struct bkey u;
292
293 bch2_bkey_val_to_text(out, c, bkey_disassemble(b, k, &u));
294 prt_newline(out);
295 }
296
297 printbuf_indent_sub(out, 4);
298 }
299 out:
300 if (bio)
301 bio_put(bio);
302 kvfree(n_ondisk);
303 enumerated_ref_put(&ca->io_ref[READ],
304 BCH_DEV_READ_REF_btree_node_ondisk_to_text);
305 }
306
307 #ifdef CONFIG_DEBUG_FS
308
bch2_debugfs_flush_buf(struct dump_iter * i)309 ssize_t bch2_debugfs_flush_buf(struct dump_iter *i)
310 {
311 if (i->buf.pos) {
312 size_t bytes = min_t(size_t, i->buf.pos, i->size);
313 int copied = bytes - copy_to_user(i->ubuf, i->buf.buf, bytes);
314
315 i->ret += copied;
316 i->ubuf += copied;
317 i->size -= copied;
318 i->buf.pos -= copied;
319 memmove(i->buf.buf, i->buf.buf + copied, i->buf.pos);
320
321 if (i->buf.last_newline >= copied)
322 i->buf.last_newline -= copied;
323 if (i->buf.last_field >= copied)
324 i->buf.last_field -= copied;
325
326 if (copied != bytes)
327 return -EFAULT;
328 }
329
330 return i->size ? 0 : i->ret;
331 }
332
bch2_dump_open(struct inode * inode,struct file * file)333 static int bch2_dump_open(struct inode *inode, struct file *file)
334 {
335 struct btree_debug *bd = inode->i_private;
336 struct dump_iter *i;
337
338 i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
339 if (!i)
340 return -ENOMEM;
341
342 file->private_data = i;
343 i->from = POS_MIN;
344 i->iter = 0;
345 i->c = container_of(bd, struct bch_fs, btree_debug[bd->id]);
346 i->id = bd->id;
347 i->buf = PRINTBUF;
348
349 return 0;
350 }
351
bch2_dump_release(struct inode * inode,struct file * file)352 int bch2_dump_release(struct inode *inode, struct file *file)
353 {
354 struct dump_iter *i = file->private_data;
355
356 printbuf_exit(&i->buf);
357 kfree(i);
358 return 0;
359 }
360
bch2_read_btree(struct file * file,char __user * buf,size_t size,loff_t * ppos)361 static ssize_t bch2_read_btree(struct file *file, char __user *buf,
362 size_t size, loff_t *ppos)
363 {
364 struct dump_iter *i = file->private_data;
365
366 i->ubuf = buf;
367 i->size = size;
368 i->ret = 0;
369
370 return bch2_debugfs_flush_buf(i) ?:
371 bch2_trans_run(i->c,
372 for_each_btree_key(trans, iter, i->id, i->from,
373 BTREE_ITER_prefetch|
374 BTREE_ITER_all_snapshots, k, ({
375 bch2_bkey_val_to_text(&i->buf, i->c, k);
376 prt_newline(&i->buf);
377 bch2_trans_unlock(trans);
378 i->from = bpos_successor(iter.pos);
379 bch2_debugfs_flush_buf(i);
380 }))) ?:
381 i->ret;
382 }
383
384 static const struct file_operations btree_debug_ops = {
385 .owner = THIS_MODULE,
386 .open = bch2_dump_open,
387 .release = bch2_dump_release,
388 .read = bch2_read_btree,
389 };
390
bch2_read_btree_formats(struct file * file,char __user * buf,size_t size,loff_t * ppos)391 static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
392 size_t size, loff_t *ppos)
393 {
394 struct dump_iter *i = file->private_data;
395
396 i->ubuf = buf;
397 i->size = size;
398 i->ret = 0;
399
400 ssize_t ret = bch2_debugfs_flush_buf(i);
401 if (ret)
402 return ret;
403
404 if (bpos_eq(SPOS_MAX, i->from))
405 return i->ret;
406
407 return bch2_trans_run(i->c,
408 for_each_btree_node(trans, iter, i->id, i->from, 0, b, ({
409 bch2_btree_node_to_text(&i->buf, i->c, b);
410 i->from = !bpos_eq(SPOS_MAX, b->key.k.p)
411 ? bpos_successor(b->key.k.p)
412 : b->key.k.p;
413
414 drop_locks_do(trans, bch2_debugfs_flush_buf(i));
415 }))) ?: i->ret;
416 }
417
418 static const struct file_operations btree_format_debug_ops = {
419 .owner = THIS_MODULE,
420 .open = bch2_dump_open,
421 .release = bch2_dump_release,
422 .read = bch2_read_btree_formats,
423 };
424
bch2_read_bfloat_failed(struct file * file,char __user * buf,size_t size,loff_t * ppos)425 static ssize_t bch2_read_bfloat_failed(struct file *file, char __user *buf,
426 size_t size, loff_t *ppos)
427 {
428 struct dump_iter *i = file->private_data;
429
430 i->ubuf = buf;
431 i->size = size;
432 i->ret = 0;
433
434 return bch2_debugfs_flush_buf(i) ?:
435 bch2_trans_run(i->c,
436 for_each_btree_key(trans, iter, i->id, i->from,
437 BTREE_ITER_prefetch|
438 BTREE_ITER_all_snapshots, k, ({
439 struct btree_path_level *l =
440 &btree_iter_path(trans, &iter)->l[0];
441 struct bkey_packed *_k =
442 bch2_btree_node_iter_peek(&l->iter, l->b);
443
444 if (bpos_gt(l->b->key.k.p, i->prev_node)) {
445 bch2_btree_node_to_text(&i->buf, i->c, l->b);
446 i->prev_node = l->b->key.k.p;
447 }
448
449 bch2_bfloat_to_text(&i->buf, l->b, _k);
450 bch2_trans_unlock(trans);
451 i->from = bpos_successor(iter.pos);
452 bch2_debugfs_flush_buf(i);
453 }))) ?:
454 i->ret;
455 }
456
457 static const struct file_operations bfloat_failed_debug_ops = {
458 .owner = THIS_MODULE,
459 .open = bch2_dump_open,
460 .release = bch2_dump_release,
461 .read = bch2_read_bfloat_failed,
462 };
463
bch2_cached_btree_node_to_text(struct printbuf * out,struct bch_fs * c,struct btree * b)464 static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
465 struct btree *b)
466 {
467 if (!out->nr_tabstops)
468 printbuf_tabstop_push(out, 32);
469
470 prt_printf(out, "%px ", b);
471 bch2_btree_id_level_to_text(out, b->c.btree_id, b->c.level);
472 prt_printf(out, "\n");
473
474 printbuf_indent_add(out, 2);
475
476 bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
477 prt_newline(out);
478
479 prt_printf(out, "flags:\t");
480 prt_bitflags(out, bch2_btree_node_flags, b->flags);
481 prt_newline(out);
482
483 prt_printf(out, "pcpu read locks:\t%u\n", b->c.lock.readers != NULL);
484 prt_printf(out, "written:\t%u\n", b->written);
485 prt_printf(out, "writes blocked:\t%u\n", !list_empty_careful(&b->write_blocked));
486 prt_printf(out, "will make reachable:\t%lx\n", b->will_make_reachable);
487
488 prt_printf(out, "journal pin %px:\t%llu\n",
489 &b->writes[0].journal, b->writes[0].journal.seq);
490 prt_printf(out, "journal pin %px:\t%llu\n",
491 &b->writes[1].journal, b->writes[1].journal.seq);
492
493 prt_printf(out, "ob:\t%u\n", b->ob.nr);
494
495 printbuf_indent_sub(out, 2);
496 }
497
bch2_cached_btree_nodes_read(struct file * file,char __user * buf,size_t size,loff_t * ppos)498 static ssize_t bch2_cached_btree_nodes_read(struct file *file, char __user *buf,
499 size_t size, loff_t *ppos)
500 {
501 struct dump_iter *i = file->private_data;
502 struct bch_fs *c = i->c;
503 bool done = false;
504 ssize_t ret = 0;
505
506 i->ubuf = buf;
507 i->size = size;
508 i->ret = 0;
509
510 do {
511 ret = bch2_debugfs_flush_buf(i);
512 if (ret)
513 return ret;
514
515 i->buf.atomic++;
516 scoped_guard(rcu) {
517 struct bucket_table *tbl =
518 rht_dereference_rcu(c->btree_cache.table.tbl,
519 &c->btree_cache.table);
520 if (i->iter < tbl->size) {
521 struct rhash_head *pos;
522 struct btree *b;
523
524 rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
525 bch2_cached_btree_node_to_text(&i->buf, c, b);
526 i->iter++;
527 } else {
528 done = true;
529 }
530 }
531 --i->buf.atomic;
532 } while (!done);
533
534 if (i->buf.allocation_failure)
535 ret = -ENOMEM;
536
537 if (!ret)
538 ret = bch2_debugfs_flush_buf(i);
539
540 return ret ?: i->ret;
541 }
542
543 static const struct file_operations cached_btree_nodes_ops = {
544 .owner = THIS_MODULE,
545 .open = bch2_dump_open,
546 .release = bch2_dump_release,
547 .read = bch2_cached_btree_nodes_read,
548 };
549
550 typedef int (*list_cmp_fn)(const struct list_head *l, const struct list_head *r);
551
list_sort(struct list_head * head,list_cmp_fn cmp)552 static void list_sort(struct list_head *head, list_cmp_fn cmp)
553 {
554 struct list_head *pos;
555
556 list_for_each(pos, head)
557 while (!list_is_last(pos, head) &&
558 cmp(pos, pos->next) > 0) {
559 struct list_head *pos2, *next = pos->next;
560
561 list_del(next);
562 list_for_each(pos2, head)
563 if (cmp(next, pos2) < 0)
564 goto pos_found;
565 BUG();
566 pos_found:
567 list_add_tail(next, pos2);
568 }
569 }
570
list_ptr_order_cmp(const struct list_head * l,const struct list_head * r)571 static int list_ptr_order_cmp(const struct list_head *l, const struct list_head *r)
572 {
573 return cmp_int(l, r);
574 }
575
bch2_btree_transactions_read(struct file * file,char __user * buf,size_t size,loff_t * ppos)576 static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
577 size_t size, loff_t *ppos)
578 {
579 struct dump_iter *i = file->private_data;
580 struct bch_fs *c = i->c;
581 struct btree_trans *trans;
582 ssize_t ret = 0;
583
584 i->ubuf = buf;
585 i->size = size;
586 i->ret = 0;
587
588 int srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
589 restart:
590 seqmutex_lock(&c->btree_trans_lock);
591 list_sort(&c->btree_trans_list, list_ptr_order_cmp);
592
593 list_for_each_entry(trans, &c->btree_trans_list, list) {
594 if ((ulong) trans <= i->iter)
595 continue;
596
597 i->iter = (ulong) trans;
598
599 if (!closure_get_not_zero(&trans->ref))
600 continue;
601
602 if (!trans->srcu_held) {
603 closure_put(&trans->ref);
604 continue;
605 }
606
607 u32 seq = seqmutex_unlock(&c->btree_trans_lock);
608
609 bch2_btree_trans_to_text(&i->buf, trans);
610
611 prt_printf(&i->buf, "backtrace:\n");
612 printbuf_indent_add(&i->buf, 2);
613 bch2_prt_task_backtrace(&i->buf, trans->locking_wait.task, 0, GFP_KERNEL);
614 printbuf_indent_sub(&i->buf, 2);
615 prt_newline(&i->buf);
616
617 closure_put(&trans->ref);
618
619 ret = bch2_debugfs_flush_buf(i);
620 if (ret)
621 goto unlocked;
622
623 if (!seqmutex_relock(&c->btree_trans_lock, seq))
624 goto restart;
625 }
626 seqmutex_unlock(&c->btree_trans_lock);
627 unlocked:
628 srcu_read_unlock(&c->btree_trans_barrier, srcu_idx);
629
630 if (i->buf.allocation_failure)
631 ret = -ENOMEM;
632
633 if (!ret)
634 ret = bch2_debugfs_flush_buf(i);
635
636 return ret ?: i->ret;
637 }
638
639 static const struct file_operations btree_transactions_ops = {
640 .owner = THIS_MODULE,
641 .open = bch2_dump_open,
642 .release = bch2_dump_release,
643 .read = bch2_btree_transactions_read,
644 };
645
bch2_journal_pins_read(struct file * file,char __user * buf,size_t size,loff_t * ppos)646 static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
647 size_t size, loff_t *ppos)
648 {
649 struct dump_iter *i = file->private_data;
650 struct bch_fs *c = i->c;
651 bool done = false;
652 int err;
653
654 i->ubuf = buf;
655 i->size = size;
656 i->ret = 0;
657
658 while (1) {
659 err = bch2_debugfs_flush_buf(i);
660 if (err)
661 return err;
662
663 if (!i->size)
664 break;
665
666 if (done)
667 break;
668
669 done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
670 i->iter++;
671 }
672
673 if (i->buf.allocation_failure)
674 return -ENOMEM;
675
676 return i->ret;
677 }
678
679 static const struct file_operations journal_pins_ops = {
680 .owner = THIS_MODULE,
681 .open = bch2_dump_open,
682 .release = bch2_dump_release,
683 .read = bch2_journal_pins_read,
684 };
685
bch2_btree_updates_read(struct file * file,char __user * buf,size_t size,loff_t * ppos)686 static ssize_t bch2_btree_updates_read(struct file *file, char __user *buf,
687 size_t size, loff_t *ppos)
688 {
689 struct dump_iter *i = file->private_data;
690 struct bch_fs *c = i->c;
691 int err;
692
693 i->ubuf = buf;
694 i->size = size;
695 i->ret = 0;
696
697 if (!i->iter) {
698 bch2_btree_updates_to_text(&i->buf, c);
699 i->iter++;
700 }
701
702 err = bch2_debugfs_flush_buf(i);
703 if (err)
704 return err;
705
706 if (i->buf.allocation_failure)
707 return -ENOMEM;
708
709 return i->ret;
710 }
711
712 static const struct file_operations btree_updates_ops = {
713 .owner = THIS_MODULE,
714 .open = bch2_dump_open,
715 .release = bch2_dump_release,
716 .read = bch2_btree_updates_read,
717 };
718
btree_transaction_stats_open(struct inode * inode,struct file * file)719 static int btree_transaction_stats_open(struct inode *inode, struct file *file)
720 {
721 struct bch_fs *c = inode->i_private;
722 struct dump_iter *i;
723
724 i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
725 if (!i)
726 return -ENOMEM;
727
728 i->iter = 1;
729 i->c = c;
730 i->buf = PRINTBUF;
731 file->private_data = i;
732
733 return 0;
734 }
735
btree_transaction_stats_release(struct inode * inode,struct file * file)736 static int btree_transaction_stats_release(struct inode *inode, struct file *file)
737 {
738 struct dump_iter *i = file->private_data;
739
740 printbuf_exit(&i->buf);
741 kfree(i);
742
743 return 0;
744 }
745
btree_transaction_stats_read(struct file * file,char __user * buf,size_t size,loff_t * ppos)746 static ssize_t btree_transaction_stats_read(struct file *file, char __user *buf,
747 size_t size, loff_t *ppos)
748 {
749 struct dump_iter *i = file->private_data;
750 struct bch_fs *c = i->c;
751 int err;
752
753 i->ubuf = buf;
754 i->size = size;
755 i->ret = 0;
756
757 while (1) {
758 struct btree_transaction_stats *s = &c->btree_transaction_stats[i->iter];
759
760 err = bch2_debugfs_flush_buf(i);
761 if (err)
762 return err;
763
764 if (!i->size)
765 break;
766
767 if (i->iter == ARRAY_SIZE(bch2_btree_transaction_fns) ||
768 !bch2_btree_transaction_fns[i->iter])
769 break;
770
771 prt_printf(&i->buf, "%s:\n", bch2_btree_transaction_fns[i->iter]);
772 printbuf_indent_add(&i->buf, 2);
773
774 mutex_lock(&s->lock);
775
776 prt_printf(&i->buf, "Max mem used: %u\n", s->max_mem);
777 #ifdef CONFIG_BCACHEFS_TRANS_KMALLOC_TRACE
778 printbuf_indent_add(&i->buf, 2);
779 bch2_trans_kmalloc_trace_to_text(&i->buf, &s->trans_kmalloc_trace);
780 printbuf_indent_sub(&i->buf, 2);
781 #endif
782
783 prt_printf(&i->buf, "Transaction duration:\n");
784
785 printbuf_indent_add(&i->buf, 2);
786 bch2_time_stats_to_text(&i->buf, &s->duration);
787 printbuf_indent_sub(&i->buf, 2);
788
789 if (IS_ENABLED(CONFIG_BCACHEFS_LOCK_TIME_STATS)) {
790 prt_printf(&i->buf, "Lock hold times:\n");
791
792 printbuf_indent_add(&i->buf, 2);
793 bch2_time_stats_to_text(&i->buf, &s->lock_hold_times);
794 printbuf_indent_sub(&i->buf, 2);
795 }
796
797 if (s->max_paths_text) {
798 prt_printf(&i->buf, "Maximum allocated btree paths (%u):\n", s->nr_max_paths);
799
800 printbuf_indent_add(&i->buf, 2);
801 prt_str_indented(&i->buf, s->max_paths_text);
802 printbuf_indent_sub(&i->buf, 2);
803 }
804
805 mutex_unlock(&s->lock);
806
807 printbuf_indent_sub(&i->buf, 2);
808 prt_newline(&i->buf);
809 i->iter++;
810 }
811
812 if (i->buf.allocation_failure)
813 return -ENOMEM;
814
815 return i->ret;
816 }
817
818 static const struct file_operations btree_transaction_stats_op = {
819 .owner = THIS_MODULE,
820 .open = btree_transaction_stats_open,
821 .release = btree_transaction_stats_release,
822 .read = btree_transaction_stats_read,
823 };
824
825 /* walk btree transactions until we find a deadlock and print it */
btree_deadlock_to_text(struct printbuf * out,struct bch_fs * c)826 static void btree_deadlock_to_text(struct printbuf *out, struct bch_fs *c)
827 {
828 struct btree_trans *trans;
829 ulong iter = 0;
830 restart:
831 seqmutex_lock(&c->btree_trans_lock);
832 list_sort(&c->btree_trans_list, list_ptr_order_cmp);
833
834 list_for_each_entry(trans, &c->btree_trans_list, list) {
835 if ((ulong) trans <= iter)
836 continue;
837
838 iter = (ulong) trans;
839
840 if (!closure_get_not_zero(&trans->ref))
841 continue;
842
843 u32 seq = seqmutex_unlock(&c->btree_trans_lock);
844
845 bool found = bch2_check_for_deadlock(trans, out) != 0;
846
847 closure_put(&trans->ref);
848
849 if (found)
850 return;
851
852 if (!seqmutex_relock(&c->btree_trans_lock, seq))
853 goto restart;
854 }
855 seqmutex_unlock(&c->btree_trans_lock);
856 }
857
858 typedef void (*fs_to_text_fn)(struct printbuf *, struct bch_fs *);
859
bch2_simple_print(struct file * file,char __user * buf,size_t size,loff_t * ppos,fs_to_text_fn fn)860 static ssize_t bch2_simple_print(struct file *file, char __user *buf,
861 size_t size, loff_t *ppos,
862 fs_to_text_fn fn)
863 {
864 struct dump_iter *i = file->private_data;
865 struct bch_fs *c = i->c;
866 ssize_t ret = 0;
867
868 i->ubuf = buf;
869 i->size = size;
870 i->ret = 0;
871
872 if (!i->iter) {
873 fn(&i->buf, c);
874 i->iter++;
875 }
876
877 if (i->buf.allocation_failure)
878 ret = -ENOMEM;
879
880 if (!ret)
881 ret = bch2_debugfs_flush_buf(i);
882
883 return ret ?: i->ret;
884 }
885
bch2_btree_deadlock_read(struct file * file,char __user * buf,size_t size,loff_t * ppos)886 static ssize_t bch2_btree_deadlock_read(struct file *file, char __user *buf,
887 size_t size, loff_t *ppos)
888 {
889 return bch2_simple_print(file, buf, size, ppos, btree_deadlock_to_text);
890 }
891
892 static const struct file_operations btree_deadlock_ops = {
893 .owner = THIS_MODULE,
894 .open = bch2_dump_open,
895 .release = bch2_dump_release,
896 .read = bch2_btree_deadlock_read,
897 };
898
bch2_write_points_read(struct file * file,char __user * buf,size_t size,loff_t * ppos)899 static ssize_t bch2_write_points_read(struct file *file, char __user *buf,
900 size_t size, loff_t *ppos)
901 {
902 return bch2_simple_print(file, buf, size, ppos, bch2_write_points_to_text);
903 }
904
905 static const struct file_operations write_points_ops = {
906 .owner = THIS_MODULE,
907 .open = bch2_dump_open,
908 .release = bch2_dump_release,
909 .read = bch2_write_points_read,
910 };
911
bch2_fs_debug_exit(struct bch_fs * c)912 void bch2_fs_debug_exit(struct bch_fs *c)
913 {
914 if (!IS_ERR_OR_NULL(c->fs_debug_dir))
915 debugfs_remove_recursive(c->fs_debug_dir);
916 }
917
bch2_fs_debug_btree_init(struct bch_fs * c,struct btree_debug * bd)918 static void bch2_fs_debug_btree_init(struct bch_fs *c, struct btree_debug *bd)
919 {
920 struct dentry *d;
921
922 d = debugfs_create_dir(bch2_btree_id_str(bd->id), c->btree_debug_dir);
923
924 debugfs_create_file("keys", 0400, d, bd, &btree_debug_ops);
925
926 debugfs_create_file("formats", 0400, d, bd, &btree_format_debug_ops);
927
928 debugfs_create_file("bfloat-failed", 0400, d, bd,
929 &bfloat_failed_debug_ops);
930 }
931
bch2_fs_debug_init(struct bch_fs * c)932 void bch2_fs_debug_init(struct bch_fs *c)
933 {
934 struct btree_debug *bd;
935 char name[100];
936
937 if (IS_ERR_OR_NULL(bch_debug))
938 return;
939
940 if (c->sb.multi_device)
941 snprintf(name, sizeof(name), "%pU", c->sb.user_uuid.b);
942 else
943 strscpy(name, c->name, sizeof(name));
944
945 c->fs_debug_dir = debugfs_create_dir(name, bch_debug);
946 if (IS_ERR_OR_NULL(c->fs_debug_dir))
947 return;
948
949 debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
950 c->btree_debug, &cached_btree_nodes_ops);
951
952 debugfs_create_file("btree_transactions", 0400, c->fs_debug_dir,
953 c->btree_debug, &btree_transactions_ops);
954
955 debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
956 c->btree_debug, &journal_pins_ops);
957
958 debugfs_create_file("btree_updates", 0400, c->fs_debug_dir,
959 c->btree_debug, &btree_updates_ops);
960
961 debugfs_create_file("btree_transaction_stats", 0400, c->fs_debug_dir,
962 c, &btree_transaction_stats_op);
963
964 debugfs_create_file("btree_deadlock", 0400, c->fs_debug_dir,
965 c->btree_debug, &btree_deadlock_ops);
966
967 debugfs_create_file("write_points", 0400, c->fs_debug_dir,
968 c->btree_debug, &write_points_ops);
969
970 bch2_fs_async_obj_debugfs_init(c);
971
972 c->btree_debug_dir = debugfs_create_dir("btrees", c->fs_debug_dir);
973 if (IS_ERR_OR_NULL(c->btree_debug_dir))
974 return;
975
976 for (bd = c->btree_debug;
977 bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
978 bd++) {
979 bd->id = bd - c->btree_debug;
980 bch2_fs_debug_btree_init(c, bd);
981 }
982 }
983
984 #endif
985
bch2_debug_exit(void)986 void bch2_debug_exit(void)
987 {
988 if (!IS_ERR_OR_NULL(bch_debug))
989 debugfs_remove_recursive(bch_debug);
990 }
991
bch2_debug_init(void)992 int __init bch2_debug_init(void)
993 {
994 bch_debug = debugfs_create_dir("bcachefs", NULL);
995 return 0;
996 }
997