git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/debug.c

   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 "buckets.h"
  17 #include "debug.h"
  18 #include "error.h"
  19 #include "extents.h"
  20 #include "fsck.h"
  21 #include "inode.h"
  22 #include "io.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 bch_dev *ca = bch_dev_bkey_exists(c, pick.ptr.dev);
  41         struct bio *bio;
  42         bool failed = false, saw_error = false;
  43
  44         if (!bch2_dev_get_ioref(ca, READ))
  45                 return false;
  46
  47         bio = bio_alloc_bioset(ca->disk_sb.bdev,
  48                                buf_pages(n_sorted, btree_bytes(c)),
  49                                REQ_OP_READ|REQ_META,
  50                                GFP_NOIO,
  51                                &c->btree_bio);
  52         bio->bi_iter.bi_sector  = pick.ptr.offset;
  53         bch2_bio_map(bio, n_sorted, btree_bytes(c));
  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_bytes(c));
  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 = kvpmalloc(btree_bytes(c), 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 *v = (void *) bkeyp_val(&b->format, k);
 159                         v->mem_ptr = 0;
 160                 }
 161
 162         v = c->verify_data;
 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);
 167
 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);
 171
 172         if (failed) {
 173                 struct printbuf buf = PRINTBUF;
 174
 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);
 177                 printbuf_exit(&buf);
 178         }
 179 out:
 180         mutex_unlock(&c->verify_lock);
 181         bch2_btree_node_io_unlock(b);
 182 }
 183
 184 void bch2_btree_node_ondisk_to_text(struct printbuf *out, struct bch_fs *c,
 185                                     const struct btree *b)
 186 {
 187         struct btree_node *n_ondisk = NULL;
 188         struct extent_ptr_decoded pick;
 189         struct bch_dev *ca;
 190         struct bio *bio = NULL;
 191         unsigned offset = 0;
 192         int ret;
 193
 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");
 196                 return;
 197         }
 198
 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");
 202                 return;
 203         }
 204
 205         n_ondisk = kvpmalloc(btree_bytes(c), GFP_KERNEL);
 206         if (!n_ondisk) {
 207                 prt_printf(out, "memory allocation failure\n");
 208                 goto out;
 209         }
 210
 211         bio = bio_alloc_bioset(ca->disk_sb.bdev,
 212                                buf_pages(n_ondisk, btree_bytes(c)),
 213                                REQ_OP_READ|REQ_META,
 214                                GFP_NOIO,
 215                                &c->btree_bio);
 216         bio->bi_iter.bi_sector  = pick.ptr.offset;
 217         bch2_bio_map(bio, n_ondisk, btree_bytes(c));
 218
 219         ret = submit_bio_wait(bio);
 220         if (ret) {
 221                 prt_printf(out, "IO error reading btree node: %s\n", bch2_err_str(ret));
 222                 goto out;
 223         }
 224
 225         while (offset < btree_sectors(c)) {
 226                 struct bset *i;
 227                 struct nonce nonce;
 228                 struct bch_csum csum;
 229                 struct bkey_packed *k;
 230                 unsigned sectors;
 231
 232                 if (!offset) {
 233                         i = &n_ondisk->keys;
 234
 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));
 238                                 goto out;
 239                         }
 240
 241                         nonce = btree_nonce(i, offset << 9);
 242                         csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, n_ondisk);
 243
 244                         if (bch2_crc_cmp(csum, n_ondisk->csum)) {
 245                                 prt_printf(out, "invalid checksum\n");
 246                                 goto out;
 247                         }
 248
 249                         bset_encrypt(c, i, offset << 9);
 250
 251                         sectors = vstruct_sectors(n_ondisk, c->block_bits);
 252                 } else {
 253                         struct btree_node_entry *bne = (void *) n_ondisk + (offset << 9);
 254
 255                         i = &bne->keys;
 256
 257                         if (i->seq != n_ondisk->keys.seq)
 258                                 break;
 259
 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));
 263                                 goto out;
 264                         }
 265
 266                         nonce = btree_nonce(i, offset << 9);
 267                         csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne);
 268
 269                         if (bch2_crc_cmp(csum, bne->csum)) {
 270                                 prt_printf(out, "invalid checksum");
 271                                 goto out;
 272                         }
 273
 274                         bset_encrypt(c, i, offset << 9);
 275
 276                         sectors = vstruct_sectors(bne, c->block_bits);
 277                 }
 278
 279                 prt_printf(out, "  offset %u version %u, journal seq %llu\n",
 280                            offset,
 281                            le16_to_cpu(i->version),
 282                            le64_to_cpu(i->journal_seq));
 283                 offset += sectors;
 284
 285                 printbuf_indent_add(out, 4);
 286
 287                 for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k)) {
 288                         struct bkey u;
 289
 290                         bch2_bkey_val_to_text(out, c, bkey_disassemble(b, k, &u));
 291                         prt_newline(out);
 292                 }
 293
 294                 printbuf_indent_sub(out, 4);
 295         }
 296 out:
 297         if (bio)
 298                 bio_put(bio);
 299         kvpfree(n_ondisk, btree_bytes(c));
 300         percpu_ref_put(&ca->io_ref);
 301 }
 302
 303 #ifdef CONFIG_DEBUG_FS
 304
 305 /* XXX: bch_fs refcounting */
 306
 307 struct dump_iter {
 308         struct bch_fs           *c;
 309         enum btree_id           id;
 310         struct bpos             from;
 311         struct bpos             prev_node;
 312         u64                     iter;
 313
 314         struct printbuf         buf;
 315
 316         char __user             *ubuf;  /* destination user buffer */
 317         size_t                  size;   /* size of requested read */
 318         ssize_t                 ret;    /* bytes read so far */
 319 };
 320
 321 static ssize_t flush_buf(struct dump_iter *i)
 322 {
 323         if (i->buf.pos) {
 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);
 326
 327                 if (err)
 328                         return err;
 329
 330                 i->ret   += bytes;
 331                 i->ubuf  += bytes;
 332                 i->size  -= bytes;
 333                 i->buf.pos -= bytes;
 334                 memmove(i->buf.buf, i->buf.buf + bytes, i->buf.pos);
 335         }
 336
 337         return i->size ? 0 : i->ret;
 338 }
 339
 340 static int bch2_dump_open(struct inode *inode, struct file *file)
 341 {
 342         struct btree_debug *bd = inode->i_private;
 343         struct dump_iter *i;
 344
 345         i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
 346         if (!i)
 347                 return -ENOMEM;
 348
 349         file->private_data = i;
 350         i->from = POS_MIN;
 351         i->iter = 0;
 352         i->c    = container_of(bd, struct bch_fs, btree_debug[bd->id]);
 353         i->id   = bd->id;
 354         i->buf  = PRINTBUF;
 355
 356         return 0;
 357 }
 358
 359 static int bch2_dump_release(struct inode *inode, struct file *file)
 360 {
 361         struct dump_iter *i = file->private_data;
 362
 363         printbuf_exit(&i->buf);
 364         kfree(i);
 365         return 0;
 366 }
 367
 368 static ssize_t bch2_read_btree(struct file *file, char __user *buf,
 369                                size_t size, loff_t *ppos)
 370 {
 371         struct dump_iter *i = file->private_data;
 372         struct btree_trans trans;
 373         struct btree_iter iter;
 374         struct bkey_s_c k;
 375         ssize_t ret;
 376
 377         i->ubuf = buf;
 378         i->size = size;
 379         i->ret  = 0;
 380
 381         bch2_trans_init(&trans, i->c, 0, 0);
 382
 383         ret = for_each_btree_key2(&trans, iter, i->id, i->from,
 384                                   BTREE_ITER_PREFETCH|
 385                                   BTREE_ITER_ALL_SNAPSHOTS, k, ({
 386                 ret = flush_buf(i);
 387                 if (ret)
 388                         break;
 389
 390                 bch2_bkey_val_to_text(&i->buf, i->c, k);
 391                 prt_newline(&i->buf);
 392                 0;
 393         }));
 394         i->from = iter.pos;
 395
 396         if (!ret)
 397                 ret = flush_buf(i);
 398
 399         bch2_trans_exit(&trans);
 400
 401         return ret ?: i->ret;
 402 }
 403
 404 static const struct file_operations btree_debug_ops = {
 405         .owner          = THIS_MODULE,
 406         .open           = bch2_dump_open,
 407         .release        = bch2_dump_release,
 408         .read           = bch2_read_btree,
 409 };
 410
 411 static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
 412                                        size_t size, loff_t *ppos)
 413 {
 414         struct dump_iter *i = file->private_data;
 415         struct btree_trans trans;
 416         struct btree_iter iter;
 417         struct btree *b;
 418         ssize_t ret;
 419
 420         i->ubuf = buf;
 421         i->size = size;
 422         i->ret  = 0;
 423
 424         ret = flush_buf(i);
 425         if (ret)
 426                 return ret;
 427
 428         if (bpos_eq(SPOS_MAX, i->from))
 429                 return i->ret;
 430
 431         bch2_trans_init(&trans, i->c, 0, 0);
 432
 433         for_each_btree_node(&trans, iter, i->id, i->from, 0, b, ret) {
 434                 ret = flush_buf(i);
 435                 if (ret)
 436                         break;
 437
 438                 bch2_btree_node_to_text(&i->buf, i->c, b);
 439                 i->from = !bpos_eq(SPOS_MAX, b->key.k.p)
 440                         ? bpos_successor(b->key.k.p)
 441                         : b->key.k.p;
 442         }
 443         bch2_trans_iter_exit(&trans, &iter);
 444
 445         bch2_trans_exit(&trans);
 446
 447         if (!ret)
 448                 ret = flush_buf(i);
 449
 450         return ret ?: i->ret;
 451 }
 452
 453 static const struct file_operations btree_format_debug_ops = {
 454         .owner          = THIS_MODULE,
 455         .open           = bch2_dump_open,
 456         .release        = bch2_dump_release,
 457         .read           = bch2_read_btree_formats,
 458 };
 459
 460 static ssize_t bch2_read_bfloat_failed(struct file *file, char __user *buf,
 461                                        size_t size, loff_t *ppos)
 462 {
 463         struct dump_iter *i = file->private_data;
 464         struct btree_trans trans;
 465         struct btree_iter iter;
 466         struct bkey_s_c k;
 467         ssize_t ret;
 468
 469         i->ubuf = buf;
 470         i->size = size;
 471         i->ret  = 0;
 472
 473         ret = flush_buf(i);
 474         if (ret)
 475                 return ret;
 476
 477         bch2_trans_init(&trans, i->c, 0, 0);
 478
 479         ret = for_each_btree_key2(&trans, iter, i->id, i->from,
 480                                   BTREE_ITER_PREFETCH|
 481                                   BTREE_ITER_ALL_SNAPSHOTS, k, ({
 482                 struct btree_path_level *l = &iter.path->l[0];
 483                 struct bkey_packed *_k =
 484                         bch2_btree_node_iter_peek(&l->iter, l->b);
 485
 486                 ret = flush_buf(i);
 487                 if (ret)
 488                         break;
 489
 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;
 493                 }
 494
 495                 bch2_bfloat_to_text(&i->buf, l->b, _k);
 496                 0;
 497         }));
 498         i->from = iter.pos;
 499
 500         bch2_trans_exit(&trans);
 501
 502         if (!ret)
 503                 ret = flush_buf(i);
 504
 505         return ret ?: i->ret;
 506 }
 507
 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,
 513 };
 514
 515 static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
 516                                            struct btree *b)
 517 {
 518         if (!out->nr_tabstops)
 519                 printbuf_tabstop_push(out, 32);
 520
 521         prt_printf(out, "%px btree=%s l=%u ",
 522                b,
 523                bch2_btree_ids[b->c.btree_id],
 524                b->c.level);
 525         prt_newline(out);
 526
 527         printbuf_indent_add(out, 2);
 528
 529         bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
 530         prt_newline(out);
 531
 532         prt_printf(out, "flags: ");
 533         prt_tab(out);
 534         prt_bitflags(out, bch2_btree_node_flags, b->flags);
 535         prt_newline(out);
 536
 537         prt_printf(out, "pcpu read locks: ");
 538         prt_tab(out);
 539         prt_printf(out, "%u", b->c.lock.readers != NULL);
 540         prt_newline(out);
 541
 542         prt_printf(out, "written:");
 543         prt_tab(out);
 544         prt_printf(out, "%u", b->written);
 545         prt_newline(out);
 546
 547         prt_printf(out, "writes blocked:");
 548         prt_tab(out);
 549         prt_printf(out, "%u", !list_empty_careful(&b->write_blocked));
 550         prt_newline(out);
 551
 552         prt_printf(out, "will make reachable:");
 553         prt_tab(out);
 554         prt_printf(out, "%lx", b->will_make_reachable);
 555         prt_newline(out);
 556
 557         prt_printf(out, "journal pin %px:", &b->writes[0].journal);
 558         prt_tab(out);
 559         prt_printf(out, "%llu", b->writes[0].journal.seq);
 560         prt_newline(out);
 561
 562         prt_printf(out, "journal pin %px:", &b->writes[1].journal);
 563         prt_tab(out);
 564         prt_printf(out, "%llu", b->writes[1].journal.seq);
 565         prt_newline(out);
 566
 567         printbuf_indent_sub(out, 2);
 568 }
 569
 570 static ssize_t bch2_cached_btree_nodes_read(struct file *file, char __user *buf,
 571                                             size_t size, loff_t *ppos)
 572 {
 573         struct dump_iter *i = file->private_data;
 574         struct bch_fs *c = i->c;
 575         bool done = false;
 576         ssize_t ret = 0;
 577
 578         i->ubuf = buf;
 579         i->size = size;
 580         i->ret  = 0;
 581
 582         do {
 583                 struct bucket_table *tbl;
 584                 struct rhash_head *pos;
 585                 struct btree *b;
 586
 587                 ret = flush_buf(i);
 588                 if (ret)
 589                         return ret;
 590
 591                 rcu_read_lock();
 592                 i->buf.atomic++;
 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);
 598                         i->iter++;
 599                 } else {
 600                         done = true;
 601                 }
 602                 --i->buf.atomic;
 603                 rcu_read_unlock();
 604         } while (!done);
 605
 606         if (i->buf.allocation_failure)
 607                 ret = -ENOMEM;
 608
 609         if (!ret)
 610                 ret = flush_buf(i);
 611
 612         return ret ?: i->ret;
 613 }
 614
 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,
 620 };
 621
 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)
 625 {
 626         struct dump_iter *i = file->private_data;
 627         struct bch_fs *c = i->c;
 628         struct btree_trans *trans;
 629         ssize_t ret = 0;
 630
 631         i->ubuf = buf;
 632         i->size = size;
 633         i->ret  = 0;
 634
 635         mutex_lock(&c->btree_trans_lock);
 636         list_for_each_entry(trans, &c->btree_trans_list, list) {
 637                 if (trans->locking_wait.task->pid <= i->iter)
 638                         continue;
 639
 640                 ret = flush_buf(i);
 641                 if (ret)
 642                         break;
 643
 644                 bch2_btree_trans_to_text(&i->buf, trans);
 645
 646                 prt_printf(&i->buf, "backtrace:");
 647                 prt_newline(&i->buf);
 648                 printbuf_indent_add(&i->buf, 2);
 649                 bch2_prt_task_backtrace(&i->buf, trans->locking_wait.task);
 650                 printbuf_indent_sub(&i->buf, 2);
 651                 prt_newline(&i->buf);
 652
 653                 i->iter = trans->locking_wait.task->pid;
 654         }
 655         mutex_unlock(&c->btree_trans_lock);
 656
 657         if (i->buf.allocation_failure)
 658                 ret = -ENOMEM;
 659
 660         if (!ret)
 661                 ret = flush_buf(i);
 662
 663         return ret ?: i->ret;
 664 }
 665
 666 static const struct file_operations btree_transactions_ops = {
 667         .owner          = THIS_MODULE,
 668         .open           = bch2_dump_open,
 669         .release        = bch2_dump_release,
 670         .read           = bch2_btree_transactions_read,
 671 };
 672 #endif /* CONFIG_BCACHEFS_DEBUG_TRANSACTIONS */
 673
 674 static ssize_t bch2_journal_pins_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         bool done = false;
 680         int err;
 681
 682         i->ubuf = buf;
 683         i->size = size;
 684         i->ret  = 0;
 685
 686         do {
 687                 err = flush_buf(i);
 688                 if (err)
 689                         return err;
 690
 691                 if (!i->size)
 692                         break;
 693
 694                 done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
 695                 i->iter++;
 696         } while (!done);
 697
 698         if (i->buf.allocation_failure)
 699                 return -ENOMEM;
 700
 701         return i->ret;
 702 }
 703
 704 static const struct file_operations journal_pins_ops = {
 705         .owner          = THIS_MODULE,
 706         .open           = bch2_dump_open,
 707         .release        = bch2_dump_release,
 708         .read           = bch2_journal_pins_read,
 709 };
 710
 711 static int lock_held_stats_open(struct inode *inode, struct file *file)
 712 {
 713         struct bch_fs *c = inode->i_private;
 714         struct dump_iter *i;
 715
 716         i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
 717
 718         if (!i)
 719                 return -ENOMEM;
 720
 721         i->iter = 0;
 722         i->c    = c;
 723         i->buf  = PRINTBUF;
 724         file->private_data = i;
 725
 726         return 0;
 727 }
 728
 729 static int lock_held_stats_release(struct inode *inode, struct file *file)
 730 {
 731         struct dump_iter *i = file->private_data;
 732
 733         printbuf_exit(&i->buf);
 734         kfree(i);
 735
 736         return 0;
 737 }
 738
 739 static ssize_t lock_held_stats_read(struct file *file, char __user *buf,
 740                                       size_t size, loff_t *ppos)
 741 {
 742         struct dump_iter        *i = file->private_data;
 743         struct bch_fs *c = i->c;
 744         int err;
 745
 746         i->ubuf = buf;
 747         i->size = size;
 748         i->ret  = 0;
 749
 750         while (1) {
 751                 struct btree_transaction_stats *s = &c->btree_transaction_stats[i->iter];
 752
 753                 err = flush_buf(i);
 754                 if (err)
 755                         return err;
 756
 757                 if (!i->size)
 758                         break;
 759
 760                 if (i->iter == ARRAY_SIZE(bch2_btree_transaction_fns) ||
 761                     !bch2_btree_transaction_fns[i->iter])
 762                         break;
 763
 764                 prt_printf(&i->buf, "%s: ", bch2_btree_transaction_fns[i->iter]);
 765                 prt_newline(&i->buf);
 766                 printbuf_indent_add(&i->buf, 2);
 767
 768                 mutex_lock(&s->lock);
 769
 770                 prt_printf(&i->buf, "Max mem used: %u", s->max_mem);
 771                 prt_newline(&i->buf);
 772
 773                 if (IS_ENABLED(CONFIG_BCACHEFS_LOCK_TIME_STATS)) {
 774                         prt_printf(&i->buf, "Lock hold times:");
 775                         prt_newline(&i->buf);
 776
 777                         printbuf_indent_add(&i->buf, 2);
 778                         bch2_time_stats_to_text(&i->buf, &s->lock_hold_times);
 779                         printbuf_indent_sub(&i->buf, 2);
 780                 }
 781
 782                 if (s->max_paths_text) {
 783                         prt_printf(&i->buf, "Maximum allocated btree paths (%u):", s->nr_max_paths);
 784                         prt_newline(&i->buf);
 785
 786                         printbuf_indent_add(&i->buf, 2);
 787                         prt_str_indented(&i->buf, s->max_paths_text);
 788                         printbuf_indent_sub(&i->buf, 2);
 789                 }
 790
 791                 mutex_unlock(&s->lock);
 792
 793                 printbuf_indent_sub(&i->buf, 2);
 794                 prt_newline(&i->buf);
 795                 i->iter++;
 796         }
 797
 798         if (i->buf.allocation_failure)
 799                 return -ENOMEM;
 800
 801         return i->ret;
 802 }
 803
 804 static const struct file_operations lock_held_stats_op = {
 805         .owner = THIS_MODULE,
 806         .open = lock_held_stats_open,
 807         .release = lock_held_stats_release,
 808         .read = lock_held_stats_read,
 809 };
 810
 811 static ssize_t bch2_btree_deadlock_read(struct file *file, char __user *buf,
 812                                             size_t size, loff_t *ppos)
 813 {
 814         struct dump_iter *i = file->private_data;
 815         struct bch_fs *c = i->c;
 816         struct btree_trans *trans;
 817         ssize_t ret = 0;
 818
 819         i->ubuf = buf;
 820         i->size = size;
 821         i->ret  = 0;
 822
 823         if (i->iter)
 824                 goto out;
 825
 826         mutex_lock(&c->btree_trans_lock);
 827         list_for_each_entry(trans, &c->btree_trans_list, list) {
 828                 if (trans->locking_wait.task->pid <= i->iter)
 829                         continue;
 830
 831                 ret = flush_buf(i);
 832                 if (ret)
 833                         break;
 834
 835                 bch2_check_for_deadlock(trans, &i->buf);
 836
 837                 i->iter = trans->locking_wait.task->pid;
 838         }
 839         mutex_unlock(&c->btree_trans_lock);
 840 out:
 841         if (i->buf.allocation_failure)
 842                 ret = -ENOMEM;
 843
 844         if (!ret)
 845                 ret = flush_buf(i);
 846
 847         return ret ?: i->ret;
 848 }
 849
 850 static const struct file_operations btree_deadlock_ops = {
 851         .owner          = THIS_MODULE,
 852         .open           = bch2_dump_open,
 853         .release        = bch2_dump_release,
 854         .read           = bch2_btree_deadlock_read,
 855 };
 856
 857 void bch2_fs_debug_exit(struct bch_fs *c)
 858 {
 859         if (!IS_ERR_OR_NULL(c->fs_debug_dir))
 860                 debugfs_remove_recursive(c->fs_debug_dir);
 861 }
 862
 863 void bch2_fs_debug_init(struct bch_fs *c)
 864 {
 865         struct btree_debug *bd;
 866         char name[100];
 867
 868         if (IS_ERR_OR_NULL(bch_debug))
 869                 return;
 870
 871         snprintf(name, sizeof(name), "%pU", c->sb.user_uuid.b);
 872         c->fs_debug_dir = debugfs_create_dir(name, bch_debug);
 873         if (IS_ERR_OR_NULL(c->fs_debug_dir))
 874                 return;
 875
 876         debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
 877                             c->btree_debug, &cached_btree_nodes_ops);
 878
 879 #ifdef CONFIG_BCACHEFS_DEBUG_TRANSACTIONS
 880         debugfs_create_file("btree_transactions", 0400, c->fs_debug_dir,
 881                             c->btree_debug, &btree_transactions_ops);
 882 #endif
 883
 884         debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
 885                             c->btree_debug, &journal_pins_ops);
 886
 887         debugfs_create_file("btree_transaction_stats", 0400, c->fs_debug_dir,
 888                             c, &lock_held_stats_op);
 889
 890         debugfs_create_file("btree_deadlock", 0400, c->fs_debug_dir,
 891                             c->btree_debug, &btree_deadlock_ops);
 892
 893         c->btree_debug_dir = debugfs_create_dir("btrees", c->fs_debug_dir);
 894         if (IS_ERR_OR_NULL(c->btree_debug_dir))
 895                 return;
 896
 897         for (bd = c->btree_debug;
 898              bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
 899              bd++) {
 900                 bd->id = bd - c->btree_debug;
 901                 debugfs_create_file(bch2_btree_ids[bd->id],
 902                                     0400, c->btree_debug_dir, bd,
 903                                     &btree_debug_ops);
 904
 905                 snprintf(name, sizeof(name), "%s-formats",
 906                          bch2_btree_ids[bd->id]);
 907
 908                 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
 909                                     &btree_format_debug_ops);
 910
 911                 snprintf(name, sizeof(name), "%s-bfloat-failed",
 912                          bch2_btree_ids[bd->id]);
 913
 914                 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
 915                                     &bfloat_failed_debug_ops);
 916         }
 917 }
 918
 919 #endif
 920
 921 void bch2_debug_exit(void)
 922 {
 923         if (!IS_ERR_OR_NULL(bch_debug))
 924                 debugfs_remove_recursive(bch_debug);
 925 }
 926
 927 int __init bch2_debug_init(void)
 928 {
 929         int ret = 0;
 930
 931         bch_debug = debugfs_create_dir("bcachefs", NULL);
 932         return ret;
 933 }