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 "super.h"
  23
  24 #include <linux/console.h>
  25 #include <linux/debugfs.h>
  26 #include <linux/module.h>
  27 #include <linux/random.h>
  28 #include <linux/seq_file.h>
  29
  30 static struct dentry *bch_debug;
  31
  32 static bool bch2_btree_verify_replica(struct bch_fs *c, struct btree *b,
  33                                       struct extent_ptr_decoded pick)
  34 {
  35         struct btree *v = c->verify_data;
  36         struct btree_node *n_ondisk = c->verify_ondisk;
  37         struct btree_node *n_sorted = c->verify_data->data;
  38         struct bset *sorted, *inmemory = &b->data->keys;
  39         struct bch_dev *ca = bch_dev_bkey_exists(c, pick.ptr.dev);
  40         struct bio *bio;
  41         bool failed = false, saw_error = false;
  42
  43         if (!bch2_dev_get_ioref(ca, READ))
  44                 return false;
  45
  46         bio = bio_alloc_bioset(ca->disk_sb.bdev,
  47                                buf_pages(n_sorted, btree_bytes(c)),
  48                                REQ_OP_READ|REQ_META,
  49                                GFP_NOFS,
  50                                &c->btree_bio);
  51         bio->bi_iter.bi_sector  = pick.ptr.offset;
  52         bch2_bio_map(bio, n_sorted, btree_bytes(c));
  53
  54         submit_bio_wait(bio);
  55
  56         bio_put(bio);
  57         percpu_ref_put(&ca->io_ref);
  58
  59         memcpy(n_ondisk, n_sorted, btree_bytes(c));
  60
  61         v->written = 0;
  62         if (bch2_btree_node_read_done(c, ca, v, false, &saw_error) || saw_error)
  63                 return false;
  64
  65         n_sorted = c->verify_data->data;
  66         sorted = &n_sorted->keys;
  67
  68         if (inmemory->u64s != sorted->u64s ||
  69             memcmp(inmemory->start,
  70                    sorted->start,
  71                    vstruct_end(inmemory) - (void *) inmemory->start)) {
  72                 unsigned offset = 0, sectors;
  73                 struct bset *i;
  74                 unsigned j;
  75
  76                 console_lock();
  77
  78                 printk(KERN_ERR "*** in memory:\n");
  79                 bch2_dump_bset(c, b, inmemory, 0);
  80
  81                 printk(KERN_ERR "*** read back in:\n");
  82                 bch2_dump_bset(c, v, sorted, 0);
  83
  84                 while (offset < v->written) {
  85                         if (!offset) {
  86                                 i = &n_ondisk->keys;
  87                                 sectors = vstruct_blocks(n_ondisk, c->block_bits) <<
  88                                         c->block_bits;
  89                         } else {
  90                                 struct btree_node_entry *bne =
  91                                         (void *) n_ondisk + (offset << 9);
  92                                 i = &bne->keys;
  93
  94                                 sectors = vstruct_blocks(bne, c->block_bits) <<
  95                                         c->block_bits;
  96                         }
  97
  98                         printk(KERN_ERR "*** on disk block %u:\n", offset);
  99                         bch2_dump_bset(c, b, i, offset);
 100
 101                         offset += sectors;
 102                 }
 103
 104                 for (j = 0; j < le16_to_cpu(inmemory->u64s); j++)
 105                         if (inmemory->_data[j] != sorted->_data[j])
 106                                 break;
 107
 108                 console_unlock();
 109                 bch_err(c, "verify failed at key %u", j);
 110
 111                 failed = true;
 112         }
 113
 114         if (v->written != b->written) {
 115                 bch_err(c, "written wrong: expected %u, got %u",
 116                         b->written, v->written);
 117                 failed = true;
 118         }
 119
 120         return failed;
 121 }
 122
 123 void __bch2_btree_verify(struct bch_fs *c, struct btree *b)
 124 {
 125         struct bkey_ptrs_c ptrs;
 126         struct extent_ptr_decoded p;
 127         const union bch_extent_entry *entry;
 128         struct btree *v;
 129         struct bset *inmemory = &b->data->keys;
 130         struct bkey_packed *k;
 131         bool failed = false;
 132
 133         if (c->opts.nochanges)
 134                 return;
 135
 136         bch2_btree_node_io_lock(b);
 137         mutex_lock(&c->verify_lock);
 138
 139         if (!c->verify_ondisk) {
 140                 c->verify_ondisk = kvpmalloc(btree_bytes(c), GFP_KERNEL);
 141                 if (!c->verify_ondisk)
 142                         goto out;
 143         }
 144
 145         if (!c->verify_data) {
 146                 c->verify_data = __bch2_btree_node_mem_alloc(c);
 147                 if (!c->verify_data)
 148                         goto out;
 149
 150                 list_del_init(&c->verify_data->list);
 151         }
 152
 153         BUG_ON(b->nsets != 1);
 154
 155         for (k = inmemory->start; k != vstruct_last(inmemory); k = bkey_p_next(k))
 156                 if (k->type == KEY_TYPE_btree_ptr_v2)
 157                         ((struct bch_btree_ptr_v2 *) bkeyp_val(&b->format, k))->mem_ptr = 0;
 158
 159         v = c->verify_data;
 160         bkey_copy(&v->key, &b->key);
 161         v->c.level      = b->c.level;
 162         v->c.btree_id   = b->c.btree_id;
 163         bch2_btree_keys_init(v);
 164
 165         ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(&b->key));
 166         bkey_for_each_ptr_decode(&b->key.k, ptrs, p, entry)
 167                 failed |= bch2_btree_verify_replica(c, b, p);
 168
 169         if (failed) {
 170                 struct printbuf buf = PRINTBUF;
 171
 172                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
 173                 bch2_fs_fatal_error(c, "btree node verify failed for : %s\n", buf.buf);
 174                 printbuf_exit(&buf);
 175         }
 176 out:
 177         mutex_unlock(&c->verify_lock);
 178         bch2_btree_node_io_unlock(b);
 179 }
 180
 181 void bch2_btree_node_ondisk_to_text(struct printbuf *out, struct bch_fs *c,
 182                                     const struct btree *b)
 183 {
 184         struct btree_node *n_ondisk = NULL;
 185         struct extent_ptr_decoded pick;
 186         struct bch_dev *ca;
 187         struct bio *bio = NULL;
 188         unsigned offset = 0;
 189         int ret;
 190
 191         if (bch2_bkey_pick_read_device(c, bkey_i_to_s_c(&b->key), NULL, &pick) <= 0) {
 192                 prt_printf(out, "error getting device to read from: invalid device\n");
 193                 return;
 194         }
 195
 196         ca = bch_dev_bkey_exists(c, pick.ptr.dev);
 197         if (!bch2_dev_get_ioref(ca, READ)) {
 198                 prt_printf(out, "error getting device to read from: not online\n");
 199                 return;
 200         }
 201
 202         n_ondisk = kvpmalloc(btree_bytes(c), GFP_KERNEL);
 203         if (!n_ondisk) {
 204                 prt_printf(out, "memory allocation failure\n");
 205                 goto out;
 206         }
 207
 208         bio = bio_alloc_bioset(ca->disk_sb.bdev,
 209                                buf_pages(n_ondisk, btree_bytes(c)),
 210                                REQ_OP_READ|REQ_META,
 211                                GFP_NOFS,
 212                                &c->btree_bio);
 213         bio->bi_iter.bi_sector  = pick.ptr.offset;
 214         bch2_bio_map(bio, n_ondisk, btree_bytes(c));
 215
 216         ret = submit_bio_wait(bio);
 217         if (ret) {
 218                 prt_printf(out, "IO error reading btree node: %s\n", bch2_err_str(ret));
 219                 goto out;
 220         }
 221
 222         while (offset < btree_sectors(c)) {
 223                 struct bset *i;
 224                 struct nonce nonce;
 225                 struct bch_csum csum;
 226                 struct bkey_packed *k;
 227                 unsigned sectors;
 228
 229                 if (!offset) {
 230                         i = &n_ondisk->keys;
 231
 232                         if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
 233                                 prt_printf(out, "unknown checksum type at offset %u: %llu\n",
 234                                            offset, BSET_CSUM_TYPE(i));
 235                                 goto out;
 236                         }
 237
 238                         nonce = btree_nonce(i, offset << 9);
 239                         csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, n_ondisk);
 240
 241                         if (bch2_crc_cmp(csum, n_ondisk->csum)) {
 242                                 prt_printf(out, "invalid checksum\n");
 243                                 goto out;
 244                         }
 245
 246                         bset_encrypt(c, i, offset << 9);
 247
 248                         sectors = vstruct_sectors(n_ondisk, c->block_bits);
 249                 } else {
 250                         struct btree_node_entry *bne = (void *) n_ondisk + (offset << 9);
 251
 252                         i = &bne->keys;
 253
 254                         if (i->seq != n_ondisk->keys.seq)
 255                                 break;
 256
 257                         if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
 258                                 prt_printf(out, "unknown checksum type at offset %u: %llu\n",
 259                                            offset, BSET_CSUM_TYPE(i));
 260                                 goto out;
 261                         }
 262
 263                         nonce = btree_nonce(i, offset << 9);
 264                         csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne);
 265
 266                         if (bch2_crc_cmp(csum, bne->csum)) {
 267                                 prt_printf(out, "invalid checksum");
 268                                 goto out;
 269                         }
 270
 271                         bset_encrypt(c, i, offset << 9);
 272
 273                         sectors = vstruct_sectors(bne, c->block_bits);
 274                 }
 275
 276                 prt_printf(out, "  offset %u version %u, journal seq %llu\n",
 277                            offset,
 278                            le16_to_cpu(i->version),
 279                            le64_to_cpu(i->journal_seq));
 280                 offset += sectors;
 281
 282                 printbuf_indent_add(out, 4);
 283
 284                 for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k)) {
 285                         struct bkey u;
 286
 287                         bch2_bkey_val_to_text(out, c, bkey_disassemble(b, k, &u));
 288                         prt_newline(out);
 289                 }
 290
 291                 printbuf_indent_sub(out, 4);
 292         }
 293 out:
 294         if (bio)
 295                 bio_put(bio);
 296         kvpfree(n_ondisk, btree_bytes(c));
 297         percpu_ref_put(&ca->io_ref);
 298 }
 299
 300 #ifdef CONFIG_DEBUG_FS
 301
 302 /* XXX: bch_fs refcounting */
 303
 304 struct dump_iter {
 305         struct bch_fs           *c;
 306         enum btree_id           id;
 307         struct bpos             from;
 308         struct bpos             prev_node;
 309         u64                     iter;
 310
 311         struct printbuf         buf;
 312
 313         char __user             *ubuf;  /* destination user buffer */
 314         size_t                  size;   /* size of requested read */
 315         ssize_t                 ret;    /* bytes read so far */
 316 };
 317
 318 static ssize_t flush_buf(struct dump_iter *i)
 319 {
 320         if (i->buf.pos) {
 321                 size_t bytes = min_t(size_t, i->buf.pos, i->size);
 322                 int copied = bytes - copy_to_user(i->ubuf, i->buf.buf, bytes);
 323
 324                 i->ret   += copied;
 325                 i->ubuf  += copied;
 326                 i->size  -= copied;
 327                 i->buf.pos -= copied;
 328                 memmove(i->buf.buf, i->buf.buf + copied, i->buf.pos);
 329
 330                 if (copied != bytes)
 331                         return -EFAULT;
 332         }
 333
 334         return i->size ? 0 : i->ret;
 335 }
 336
 337 static int bch2_dump_open(struct inode *inode, struct file *file)
 338 {
 339         struct btree_debug *bd = inode->i_private;
 340         struct dump_iter *i;
 341
 342         i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
 343         if (!i)
 344                 return -ENOMEM;
 345
 346         file->private_data = i;
 347         i->from = POS_MIN;
 348         i->iter = 0;
 349         i->c    = container_of(bd, struct bch_fs, btree_debug[bd->id]);
 350         i->id   = bd->id;
 351         i->buf  = PRINTBUF;
 352
 353         return 0;
 354 }
 355
 356 static int bch2_dump_release(struct inode *inode, struct file *file)
 357 {
 358         struct dump_iter *i = file->private_data;
 359
 360         printbuf_exit(&i->buf);
 361         kfree(i);
 362         return 0;
 363 }
 364
 365 static ssize_t bch2_read_btree(struct file *file, char __user *buf,
 366                                size_t size, loff_t *ppos)
 367 {
 368         struct dump_iter *i = file->private_data;
 369         struct btree_trans *trans;
 370         struct btree_iter iter;
 371         struct bkey_s_c k;
 372         ssize_t ret;
 373
 374         i->ubuf = buf;
 375         i->size = size;
 376         i->ret  = 0;
 377
 378         ret = flush_buf(i);
 379         if (ret)
 380                 return ret;
 381
 382         trans = bch2_trans_get(i->c);
 383         ret = for_each_btree_key2(trans, iter, i->id, i->from,
 384                                   BTREE_ITER_PREFETCH|
 385                                   BTREE_ITER_ALL_SNAPSHOTS, k, ({
 386                 bch2_bkey_val_to_text(&i->buf, i->c, k);
 387                 prt_newline(&i->buf);
 388                 drop_locks_do(trans, flush_buf(i));
 389         }));
 390         i->from = iter.pos;
 391
 392         bch2_trans_put(trans);
 393
 394         if (!ret)
 395                 ret = flush_buf(i);
 396
 397         return ret ?: i->ret;
 398 }
 399
 400 static const struct file_operations btree_debug_ops = {
 401         .owner          = THIS_MODULE,
 402         .open           = bch2_dump_open,
 403         .release        = bch2_dump_release,
 404         .read           = bch2_read_btree,
 405 };
 406
 407 static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
 408                                        size_t size, loff_t *ppos)
 409 {
 410         struct dump_iter *i = file->private_data;
 411         struct btree_trans *trans;
 412         struct btree_iter iter;
 413         struct btree *b;
 414         ssize_t ret;
 415
 416         i->ubuf = buf;
 417         i->size = size;
 418         i->ret  = 0;
 419
 420         ret = flush_buf(i);
 421         if (ret)
 422                 return ret;
 423
 424         if (bpos_eq(SPOS_MAX, i->from))
 425                 return i->ret;
 426
 427         trans = bch2_trans_get(i->c);
 428 retry:
 429         bch2_trans_begin(trans);
 430
 431         for_each_btree_node(trans, iter, i->id, i->from, 0, b, ret) {
 432                 bch2_btree_node_to_text(&i->buf, i->c, b);
 433                 i->from = !bpos_eq(SPOS_MAX, b->key.k.p)
 434                         ? bpos_successor(b->key.k.p)
 435                         : b->key.k.p;
 436
 437                 ret = drop_locks_do(trans, flush_buf(i));
 438                 if (ret)
 439                         break;
 440         }
 441         bch2_trans_iter_exit(trans, &iter);
 442
 443         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 444                 goto retry;
 445
 446         bch2_trans_put(trans);
 447
 448         if (!ret)
 449                 ret = flush_buf(i);
 450
 451         return ret ?: i->ret;
 452 }
 453
 454 static const struct file_operations btree_format_debug_ops = {
 455         .owner          = THIS_MODULE,
 456         .open           = bch2_dump_open,
 457         .release        = bch2_dump_release,
 458         .read           = bch2_read_btree_formats,
 459 };
 460
 461 static ssize_t bch2_read_bfloat_failed(struct file *file, char __user *buf,
 462                                        size_t size, loff_t *ppos)
 463 {
 464         struct dump_iter *i = file->private_data;
 465         struct btree_trans *trans;
 466         struct btree_iter iter;
 467         struct bkey_s_c k;
 468         ssize_t ret;
 469
 470         i->ubuf = buf;
 471         i->size = size;
 472         i->ret  = 0;
 473
 474         ret = flush_buf(i);
 475         if (ret)
 476                 return ret;
 477
 478         trans = bch2_trans_get(i->c);
 479
 480         ret = for_each_btree_key2(trans, iter, i->id, i->from,
 481                                   BTREE_ITER_PREFETCH|
 482                                   BTREE_ITER_ALL_SNAPSHOTS, k, ({
 483                 struct btree_path_level *l = &iter.path->l[0];
 484                 struct bkey_packed *_k =
 485                         bch2_btree_node_iter_peek(&l->iter, l->b);
 486
 487                 if (bpos_gt(l->b->key.k.p, i->prev_node)) {
 488                         bch2_btree_node_to_text(&i->buf, i->c, l->b);
 489                         i->prev_node = l->b->key.k.p;
 490                 }
 491
 492                 bch2_bfloat_to_text(&i->buf, l->b, _k);
 493                 drop_locks_do(trans, flush_buf(i));
 494         }));
 495         i->from = iter.pos;
 496
 497         bch2_trans_put(trans);
 498
 499         if (!ret)
 500                 ret = flush_buf(i);
 501
 502         return ret ?: i->ret;
 503 }
 504
 505 static const struct file_operations bfloat_failed_debug_ops = {
 506         .owner          = THIS_MODULE,
 507         .open           = bch2_dump_open,
 508         .release        = bch2_dump_release,
 509         .read           = bch2_read_bfloat_failed,
 510 };
 511
 512 static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
 513                                            struct btree *b)
 514 {
 515         if (!out->nr_tabstops)
 516                 printbuf_tabstop_push(out, 32);
 517
 518         prt_printf(out, "%px btree=%s l=%u ",
 519                b,
 520                bch2_btree_id_str(b->c.btree_id),
 521                b->c.level);
 522         prt_newline(out);
 523
 524         printbuf_indent_add(out, 2);
 525
 526         bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
 527         prt_newline(out);
 528
 529         prt_printf(out, "flags: ");
 530         prt_tab(out);
 531         prt_bitflags(out, bch2_btree_node_flags, b->flags);
 532         prt_newline(out);
 533
 534         prt_printf(out, "pcpu read locks: ");
 535         prt_tab(out);
 536         prt_printf(out, "%u", b->c.lock.readers != NULL);
 537         prt_newline(out);
 538
 539         prt_printf(out, "written:");
 540         prt_tab(out);
 541         prt_printf(out, "%u", b->written);
 542         prt_newline(out);
 543
 544         prt_printf(out, "writes blocked:");
 545         prt_tab(out);
 546         prt_printf(out, "%u", !list_empty_careful(&b->write_blocked));
 547         prt_newline(out);
 548
 549         prt_printf(out, "will make reachable:");
 550         prt_tab(out);
 551         prt_printf(out, "%lx", b->will_make_reachable);
 552         prt_newline(out);
 553
 554         prt_printf(out, "journal pin %px:", &b->writes[0].journal);
 555         prt_tab(out);
 556         prt_printf(out, "%llu", b->writes[0].journal.seq);
 557         prt_newline(out);
 558
 559         prt_printf(out, "journal pin %px:", &b->writes[1].journal);
 560         prt_tab(out);
 561         prt_printf(out, "%llu", b->writes[1].journal.seq);
 562         prt_newline(out);
 563
 564         printbuf_indent_sub(out, 2);
 565 }
 566
 567 static ssize_t bch2_cached_btree_nodes_read(struct file *file, char __user *buf,
 568                                             size_t size, loff_t *ppos)
 569 {
 570         struct dump_iter *i = file->private_data;
 571         struct bch_fs *c = i->c;
 572         bool done = false;
 573         ssize_t ret = 0;
 574
 575         i->ubuf = buf;
 576         i->size = size;
 577         i->ret  = 0;
 578
 579         do {
 580                 struct bucket_table *tbl;
 581                 struct rhash_head *pos;
 582                 struct btree *b;
 583
 584                 ret = flush_buf(i);
 585                 if (ret)
 586                         return ret;
 587
 588                 rcu_read_lock();
 589                 i->buf.atomic++;
 590                 tbl = rht_dereference_rcu(c->btree_cache.table.tbl,
 591                                           &c->btree_cache.table);
 592                 if (i->iter < tbl->size) {
 593                         rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
 594                                 bch2_cached_btree_node_to_text(&i->buf, c, b);
 595                         i->iter++;
 596                 } else {
 597                         done = true;
 598                 }
 599                 --i->buf.atomic;
 600                 rcu_read_unlock();
 601         } while (!done);
 602
 603         if (i->buf.allocation_failure)
 604                 ret = -ENOMEM;
 605
 606         if (!ret)
 607                 ret = flush_buf(i);
 608
 609         return ret ?: i->ret;
 610 }
 611
 612 static const struct file_operations cached_btree_nodes_ops = {
 613         .owner          = THIS_MODULE,
 614         .open           = bch2_dump_open,
 615         .release        = bch2_dump_release,
 616         .read           = bch2_cached_btree_nodes_read,
 617 };
 618
 619 #ifdef CONFIG_BCACHEFS_DEBUG_TRANSACTIONS
 620 static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
 621                                             size_t size, loff_t *ppos)
 622 {
 623         struct dump_iter *i = file->private_data;
 624         struct bch_fs *c = i->c;
 625         struct btree_trans *trans;
 626         ssize_t ret = 0;
 627         u32 seq;
 628
 629         i->ubuf = buf;
 630         i->size = size;
 631         i->ret  = 0;
 632 restart:
 633         seqmutex_lock(&c->btree_trans_lock);
 634         list_for_each_entry(trans, &c->btree_trans_list, list) {
 635                 if (trans->locking_wait.task->pid <= i->iter)
 636                         continue;
 637
 638                 closure_get(&trans->ref);
 639                 seq = seqmutex_seq(&c->btree_trans_lock);
 640                 seqmutex_unlock(&c->btree_trans_lock);
 641
 642                 ret = flush_buf(i);
 643                 if (ret) {
 644                         closure_put(&trans->ref);
 645                         goto unlocked;
 646                 }
 647
 648                 bch2_btree_trans_to_text(&i->buf, trans);
 649
 650                 prt_printf(&i->buf, "backtrace:");
 651                 prt_newline(&i->buf);
 652                 printbuf_indent_add(&i->buf, 2);
 653                 bch2_prt_task_backtrace(&i->buf, trans->locking_wait.task);
 654                 printbuf_indent_sub(&i->buf, 2);
 655                 prt_newline(&i->buf);
 656
 657                 i->iter = trans->locking_wait.task->pid;
 658
 659                 closure_put(&trans->ref);
 660
 661                 if (!seqmutex_relock(&c->btree_trans_lock, seq))
 662                         goto restart;
 663         }
 664         seqmutex_unlock(&c->btree_trans_lock);
 665 unlocked:
 666         if (i->buf.allocation_failure)
 667                 ret = -ENOMEM;
 668
 669         if (!ret)
 670                 ret = flush_buf(i);
 671
 672         return ret ?: i->ret;
 673 }
 674
 675 static const struct file_operations btree_transactions_ops = {
 676         .owner          = THIS_MODULE,
 677         .open           = bch2_dump_open,
 678         .release        = bch2_dump_release,
 679         .read           = bch2_btree_transactions_read,
 680 };
 681 #endif /* CONFIG_BCACHEFS_DEBUG_TRANSACTIONS */
 682
 683 static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
 684                                       size_t size, loff_t *ppos)
 685 {
 686         struct dump_iter *i = file->private_data;
 687         struct bch_fs *c = i->c;
 688         bool done = false;
 689         int err;
 690
 691         i->ubuf = buf;
 692         i->size = size;
 693         i->ret  = 0;
 694
 695         do {
 696                 err = flush_buf(i);
 697                 if (err)
 698                         return err;
 699
 700                 if (!i->size)
 701                         break;
 702
 703                 done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
 704                 i->iter++;
 705         } while (!done);
 706
 707         if (i->buf.allocation_failure)
 708                 return -ENOMEM;
 709
 710         return i->ret;
 711 }
 712
 713 static const struct file_operations journal_pins_ops = {
 714         .owner          = THIS_MODULE,
 715         .open           = bch2_dump_open,
 716         .release        = bch2_dump_release,
 717         .read           = bch2_journal_pins_read,
 718 };
 719
 720 static int lock_held_stats_open(struct inode *inode, struct file *file)
 721 {
 722         struct bch_fs *c = inode->i_private;
 723         struct dump_iter *i;
 724
 725         i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
 726
 727         if (!i)
 728                 return -ENOMEM;
 729
 730         i->iter = 0;
 731         i->c    = c;
 732         i->buf  = PRINTBUF;
 733         file->private_data = i;
 734
 735         return 0;
 736 }
 737
 738 static int lock_held_stats_release(struct inode *inode, struct file *file)
 739 {
 740         struct dump_iter *i = file->private_data;
 741
 742         printbuf_exit(&i->buf);
 743         kfree(i);
 744
 745         return 0;
 746 }
 747
 748 static ssize_t lock_held_stats_read(struct file *file, char __user *buf,
 749                                       size_t size, loff_t *ppos)
 750 {
 751         struct dump_iter        *i = file->private_data;
 752         struct bch_fs *c = i->c;
 753         int err;
 754
 755         i->ubuf = buf;
 756         i->size = size;
 757         i->ret  = 0;
 758
 759         while (1) {
 760                 struct btree_transaction_stats *s = &c->btree_transaction_stats[i->iter];
 761
 762                 err = flush_buf(i);
 763                 if (err)
 764                         return err;
 765
 766                 if (!i->size)
 767                         break;
 768
 769                 if (i->iter == ARRAY_SIZE(bch2_btree_transaction_fns) ||
 770                     !bch2_btree_transaction_fns[i->iter])
 771                         break;
 772
 773                 prt_printf(&i->buf, "%s: ", bch2_btree_transaction_fns[i->iter]);
 774                 prt_newline(&i->buf);
 775                 printbuf_indent_add(&i->buf, 2);
 776
 777                 mutex_lock(&s->lock);
 778
 779                 prt_printf(&i->buf, "Max mem used: %u", s->max_mem);
 780                 prt_newline(&i->buf);
 781
 782                 if (IS_ENABLED(CONFIG_BCACHEFS_LOCK_TIME_STATS)) {
 783                         prt_printf(&i->buf, "Lock hold times:");
 784                         prt_newline(&i->buf);
 785
 786                         printbuf_indent_add(&i->buf, 2);
 787                         bch2_time_stats_to_text(&i->buf, &s->lock_hold_times);
 788                         printbuf_indent_sub(&i->buf, 2);
 789                 }
 790
 791                 if (s->max_paths_text) {
 792                         prt_printf(&i->buf, "Maximum allocated btree paths (%u):", s->nr_max_paths);
 793                         prt_newline(&i->buf);
 794
 795                         printbuf_indent_add(&i->buf, 2);
 796                         prt_str_indented(&i->buf, s->max_paths_text);
 797                         printbuf_indent_sub(&i->buf, 2);
 798                 }
 799
 800                 mutex_unlock(&s->lock);
 801
 802                 printbuf_indent_sub(&i->buf, 2);
 803                 prt_newline(&i->buf);
 804                 i->iter++;
 805         }
 806
 807         if (i->buf.allocation_failure)
 808                 return -ENOMEM;
 809
 810         return i->ret;
 811 }
 812
 813 static const struct file_operations lock_held_stats_op = {
 814         .owner = THIS_MODULE,
 815         .open = lock_held_stats_open,
 816         .release = lock_held_stats_release,
 817         .read = lock_held_stats_read,
 818 };
 819
 820 static ssize_t bch2_btree_deadlock_read(struct file *file, char __user *buf,
 821                                             size_t size, loff_t *ppos)
 822 {
 823         struct dump_iter *i = file->private_data;
 824         struct bch_fs *c = i->c;
 825         struct btree_trans *trans;
 826         ssize_t ret = 0;
 827         u32 seq;
 828
 829         i->ubuf = buf;
 830         i->size = size;
 831         i->ret  = 0;
 832
 833         if (i->iter)
 834                 goto out;
 835 restart:
 836         seqmutex_lock(&c->btree_trans_lock);
 837         list_for_each_entry(trans, &c->btree_trans_list, list) {
 838                 if (trans->locking_wait.task->pid <= i->iter)
 839                         continue;
 840
 841                 closure_get(&trans->ref);
 842                 seq = seqmutex_seq(&c->btree_trans_lock);
 843                 seqmutex_unlock(&c->btree_trans_lock);
 844
 845                 ret = flush_buf(i);
 846                 if (ret) {
 847                         closure_put(&trans->ref);
 848                         goto out;
 849                 }
 850
 851                 bch2_check_for_deadlock(trans, &i->buf);
 852
 853                 i->iter = trans->locking_wait.task->pid;
 854
 855                 closure_put(&trans->ref);
 856
 857                 if (!seqmutex_relock(&c->btree_trans_lock, seq))
 858                         goto restart;
 859         }
 860         seqmutex_unlock(&c->btree_trans_lock);
 861 out:
 862         if (i->buf.allocation_failure)
 863                 ret = -ENOMEM;
 864
 865         if (!ret)
 866                 ret = flush_buf(i);
 867
 868         return ret ?: i->ret;
 869 }
 870
 871 static const struct file_operations btree_deadlock_ops = {
 872         .owner          = THIS_MODULE,
 873         .open           = bch2_dump_open,
 874         .release        = bch2_dump_release,
 875         .read           = bch2_btree_deadlock_read,
 876 };
 877
 878 void bch2_fs_debug_exit(struct bch_fs *c)
 879 {
 880         if (!IS_ERR_OR_NULL(c->fs_debug_dir))
 881                 debugfs_remove_recursive(c->fs_debug_dir);
 882 }
 883
 884 void bch2_fs_debug_init(struct bch_fs *c)
 885 {
 886         struct btree_debug *bd;
 887         char name[100];
 888
 889         if (IS_ERR_OR_NULL(bch_debug))
 890                 return;
 891
 892         snprintf(name, sizeof(name), "%pU", c->sb.user_uuid.b);
 893         c->fs_debug_dir = debugfs_create_dir(name, bch_debug);
 894         if (IS_ERR_OR_NULL(c->fs_debug_dir))
 895                 return;
 896
 897         debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
 898                             c->btree_debug, &cached_btree_nodes_ops);
 899
 900 #ifdef CONFIG_BCACHEFS_DEBUG_TRANSACTIONS
 901         debugfs_create_file("btree_transactions", 0400, c->fs_debug_dir,
 902                             c->btree_debug, &btree_transactions_ops);
 903 #endif
 904
 905         debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
 906                             c->btree_debug, &journal_pins_ops);
 907
 908         debugfs_create_file("btree_transaction_stats", 0400, c->fs_debug_dir,
 909                             c, &lock_held_stats_op);
 910
 911         debugfs_create_file("btree_deadlock", 0400, c->fs_debug_dir,
 912                             c->btree_debug, &btree_deadlock_ops);
 913
 914         c->btree_debug_dir = debugfs_create_dir("btrees", c->fs_debug_dir);
 915         if (IS_ERR_OR_NULL(c->btree_debug_dir))
 916                 return;
 917
 918         for (bd = c->btree_debug;
 919              bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
 920              bd++) {
 921                 bd->id = bd - c->btree_debug;
 922                 debugfs_create_file(bch2_btree_id_str(bd->id),
 923                                     0400, c->btree_debug_dir, bd,
 924                                     &btree_debug_ops);
 925
 926                 snprintf(name, sizeof(name), "%s-formats",
 927                          bch2_btree_id_str(bd->id));
 928
 929                 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
 930                                     &btree_format_debug_ops);
 931
 932                 snprintf(name, sizeof(name), "%s-bfloat-failed",
 933                          bch2_btree_id_str(bd->id));
 934
 935                 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
 936                                     &bfloat_failed_debug_ops);
 937         }
 938 }
 939
 940 #endif
 941
 942 void bch2_debug_exit(void)
 943 {
 944         if (!IS_ERR_OR_NULL(bch_debug))
 945                 debugfs_remove_recursive(bch_debug);
 946 }
 947
 948 int __init bch2_debug_init(void)
 949 {
 950         int ret = 0;
 951
 952         bch_debug = debugfs_create_dir("bcachefs", NULL);
 953         return ret;
 954 }