fs/bcachefs/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 "btree_update_interior.h"
  17 #include "buckets.h"
  18 #include "debug.h"
  19 #include "error.h"
  20 #include "extents.h"
  21 #include "fsck.h"
  22 #include "inode.h"
  23 #include "super.h"
  24
  25 #include <linux/console.h>
  26 #include <linux/debugfs.h>
  27 #include <linux/module.h>
  28 #include <linux/random.h>
  29 #include <linux/seq_file.h>
  30
  31 static struct dentry *bch_debug;
  32
  33 static bool bch2_btree_verify_replica(struct bch_fs *c, struct btree *b,
  34                                       struct extent_ptr_decoded pick)
  35 {
  36         struct btree *v = c->verify_data;
  37         struct btree_node *n_ondisk = c->verify_ondisk;
  38         struct btree_node *n_sorted = c->verify_data->data;
  39         struct bset *sorted, *inmemory = &b->data->keys;
  40         struct 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_buf_bytes(b)),
  49                                REQ_OP_READ|REQ_META,
  50                                GFP_NOFS,
  51                                &c->btree_bio);
  52         bio->bi_iter.bi_sector  = pick.ptr.offset;
  53         bch2_bio_map(bio, n_sorted, btree_buf_bytes(b));
  54
  55         submit_bio_wait(bio);
  56
  57         bio_put(bio);
  58         percpu_ref_put(&ca->io_ref);
  59
  60         memcpy(n_ondisk, n_sorted, btree_buf_bytes(b));
  61
  62         v->written = 0;
  63         if (bch2_btree_node_read_done(c, ca, v, false, &saw_error) || saw_error)
  64                 return false;
  65
  66         n_sorted = c->verify_data->data;
  67         sorted = &n_sorted->keys;
  68
  69         if (inmemory->u64s != sorted->u64s ||
  70             memcmp(inmemory->start,
  71                    sorted->start,
  72                    vstruct_end(inmemory) - (void *) inmemory->start)) {
  73                 unsigned offset = 0, sectors;
  74                 struct bset *i;
  75                 unsigned j;
  76
  77                 console_lock();
  78
  79                 printk(KERN_ERR "*** in memory:\n");
  80                 bch2_dump_bset(c, b, inmemory, 0);
  81
  82                 printk(KERN_ERR "*** read back in:\n");
  83                 bch2_dump_bset(c, v, sorted, 0);
  84
  85                 while (offset < v->written) {
  86                         if (!offset) {
  87                                 i = &n_ondisk->keys;
  88                                 sectors = vstruct_blocks(n_ondisk, c->block_bits) <<
  89                                         c->block_bits;
  90                         } else {
  91                                 struct btree_node_entry *bne =
  92                                         (void *) n_ondisk + (offset << 9);
  93                                 i = &bne->keys;
  94
  95                                 sectors = vstruct_blocks(bne, c->block_bits) <<
  96                                         c->block_bits;
  97                         }
  98
  99                         printk(KERN_ERR "*** on disk block %u:\n", offset);
 100                         bch2_dump_bset(c, b, i, offset);
 101
 102                         offset += sectors;
 103                 }
 104
 105                 for (j = 0; j < le16_to_cpu(inmemory->u64s); j++)
 106                         if (inmemory->_data[j] != sorted->_data[j])
 107                                 break;
 108
 109                 console_unlock();
 110                 bch_err(c, "verify failed at key %u", j);
 111
 112                 failed = true;
 113         }
 114
 115         if (v->written != b->written) {
 116                 bch_err(c, "written wrong: expected %u, got %u",
 117                         b->written, v->written);
 118                 failed = true;
 119         }
 120
 121         return failed;
 122 }
 123
 124 void __bch2_btree_verify(struct bch_fs *c, struct btree *b)
 125 {
 126         struct bkey_ptrs_c ptrs;
 127         struct extent_ptr_decoded p;
 128         const union bch_extent_entry *entry;
 129         struct btree *v;
 130         struct bset *inmemory = &b->data->keys;
 131         struct bkey_packed *k;
 132         bool failed = false;
 133
 134         if (c->opts.nochanges)
 135                 return;
 136
 137         bch2_btree_node_io_lock(b);
 138         mutex_lock(&c->verify_lock);
 139
 140         if (!c->verify_ondisk) {
 141                 c->verify_ondisk = kvmalloc(btree_buf_bytes(b), GFP_KERNEL);
 142                 if (!c->verify_ondisk)
 143                         goto out;
 144         }
 145
 146         if (!c->verify_data) {
 147                 c->verify_data = __bch2_btree_node_mem_alloc(c);
 148                 if (!c->verify_data)
 149                         goto out;
 150
 151                 list_del_init(&c->verify_data->list);
 152         }
 153
 154         BUG_ON(b->nsets != 1);
 155
 156         for (k = inmemory->start; k != vstruct_last(inmemory); k = bkey_p_next(k))
 157                 if (k->type == KEY_TYPE_btree_ptr_v2)
 158                         ((struct bch_btree_ptr_v2 *) bkeyp_val(&b->format, k))->mem_ptr = 0;
 159
 160         v = c->verify_data;
 161         bkey_copy(&v->key, &b->key);
 162         v->c.level      = b->c.level;
 163         v->c.btree_id   = b->c.btree_id;
 164         bch2_btree_keys_init(v);
 165
 166         ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(&b->key));
 167         bkey_for_each_ptr_decode(&b->key.k, ptrs, p, entry)
 168                 failed |= bch2_btree_verify_replica(c, b, p);
 169
 170         if (failed) {
 171                 struct printbuf buf = PRINTBUF;
 172
 173                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
 174                 bch2_fs_fatal_error(c, ": btree node verify failed for: %s\n", buf.buf);
 175                 printbuf_exit(&buf);
 176         }
 177 out:
 178         mutex_unlock(&c->verify_lock);
 179         bch2_btree_node_io_unlock(b);
 180 }
 181
 182 void bch2_btree_node_ondisk_to_text(struct printbuf *out, struct bch_fs *c,
 183                                     const struct btree *b)
 184 {
 185         struct btree_node *n_ondisk = NULL;
 186         struct extent_ptr_decoded pick;
 187         struct bch_dev *ca;
 188         struct bio *bio = NULL;
 189         unsigned offset = 0;
 190         int ret;
 191
 192         if (bch2_bkey_pick_read_device(c, bkey_i_to_s_c(&b->key), NULL, &pick) <= 0) {
 193                 prt_printf(out, "error getting device to read from: invalid device\n");
 194                 return;
 195         }
 196
 197         ca = bch_dev_bkey_exists(c, pick.ptr.dev);
 198         if (!bch2_dev_get_ioref(ca, READ)) {
 199                 prt_printf(out, "error getting device to read from: not online\n");
 200                 return;
 201         }
 202
 203         n_ondisk = kvmalloc(btree_buf_bytes(b), GFP_KERNEL);
 204         if (!n_ondisk) {
 205                 prt_printf(out, "memory allocation failure\n");
 206                 goto out;
 207         }
 208
 209         bio = bio_alloc_bioset(ca->disk_sb.bdev,
 210                                buf_pages(n_ondisk, btree_buf_bytes(b)),
 211                                REQ_OP_READ|REQ_META,
 212                                GFP_NOFS,
 213                                &c->btree_bio);
 214         bio->bi_iter.bi_sector  = pick.ptr.offset;
 215         bch2_bio_map(bio, n_ondisk, btree_buf_bytes(b));
 216
 217         ret = submit_bio_wait(bio);
 218         if (ret) {
 219                 prt_printf(out, "IO error reading btree node: %s\n", bch2_err_str(ret));
 220                 goto out;
 221         }
 222
 223         while (offset < btree_sectors(c)) {
 224                 struct bset *i;
 225                 struct nonce nonce;
 226                 struct bch_csum csum;
 227                 struct bkey_packed *k;
 228                 unsigned sectors;
 229
 230                 if (!offset) {
 231                         i = &n_ondisk->keys;
 232
 233                         if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
 234                                 prt_printf(out, "unknown checksum type at offset %u: %llu\n",
 235                                            offset, BSET_CSUM_TYPE(i));
 236                                 goto out;
 237                         }
 238
 239                         nonce = btree_nonce(i, offset << 9);
 240                         csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, n_ondisk);
 241
 242                         if (bch2_crc_cmp(csum, n_ondisk->csum)) {
 243                                 prt_printf(out, "invalid checksum\n");
 244                                 goto out;
 245                         }
 246
 247                         bset_encrypt(c, i, offset << 9);
 248
 249                         sectors = vstruct_sectors(n_ondisk, c->block_bits);
 250                 } else {
 251                         struct btree_node_entry *bne = (void *) n_ondisk + (offset << 9);
 252
 253                         i = &bne->keys;
 254
 255                         if (i->seq != n_ondisk->keys.seq)
 256                                 break;
 257
 258                         if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
 259                                 prt_printf(out, "unknown checksum type at offset %u: %llu\n",
 260                                            offset, BSET_CSUM_TYPE(i));
 261                                 goto out;
 262                         }
 263
 264                         nonce = btree_nonce(i, offset << 9);
 265                         csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne);
 266
 267                         if (bch2_crc_cmp(csum, bne->csum)) {
 268                                 prt_printf(out, "invalid checksum");
 269                                 goto out;
 270                         }
 271
 272                         bset_encrypt(c, i, offset << 9);
 273
 274                         sectors = vstruct_sectors(bne, c->block_bits);
 275                 }
 276
 277                 prt_printf(out, "  offset %u version %u, journal seq %llu\n",
 278                            offset,
 279                            le16_to_cpu(i->version),
 280                            le64_to_cpu(i->journal_seq));
 281                 offset += sectors;
 282
 283                 printbuf_indent_add(out, 4);
 284
 285                 for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k)) {
 286                         struct bkey u;
 287
 288                         bch2_bkey_val_to_text(out, c, bkey_disassemble(b, k, &u));
 289                         prt_newline(out);
 290                 }
 291
 292                 printbuf_indent_sub(out, 4);
 293         }
 294 out:
 295         if (bio)
 296                 bio_put(bio);
 297         kvfree(n_ondisk);
 298         percpu_ref_put(&ca->io_ref);
 299 }
 300
 301 #ifdef CONFIG_DEBUG_FS
 302
 303 /* XXX: bch_fs refcounting */
 304
 305 struct dump_iter {
 306         struct bch_fs           *c;
 307         enum btree_id           id;
 308         struct bpos             from;
 309         struct bpos             prev_node;
 310         u64                     iter;
 311
 312         struct printbuf         buf;
 313
 314         char __user             *ubuf;  /* destination user buffer */
 315         size_t                  size;   /* size of requested read */
 316         ssize_t                 ret;    /* bytes read so far */
 317 };
 318
 319 static ssize_t flush_buf(struct dump_iter *i)
 320 {
 321         if (i->buf.pos) {
 322                 size_t bytes = min_t(size_t, i->buf.pos, i->size);
 323                 int copied = bytes - copy_to_user(i->ubuf, i->buf.buf, bytes);
 324
 325                 i->ret   += copied;
 326                 i->ubuf  += copied;
 327                 i->size  -= copied;
 328                 i->buf.pos -= copied;
 329                 memmove(i->buf.buf, i->buf.buf + copied, i->buf.pos);
 330
 331                 if (copied != bytes)
 332                         return -EFAULT;
 333         }
 334
 335         return i->size ? 0 : i->ret;
 336 }
 337
 338 static int bch2_dump_open(struct inode *inode, struct file *file)
 339 {
 340         struct btree_debug *bd = inode->i_private;
 341         struct dump_iter *i;
 342
 343         i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
 344         if (!i)
 345                 return -ENOMEM;
 346
 347         file->private_data = i;
 348         i->from = POS_MIN;
 349         i->iter = 0;
 350         i->c    = container_of(bd, struct bch_fs, btree_debug[bd->id]);
 351         i->id   = bd->id;
 352         i->buf  = PRINTBUF;
 353
 354         return 0;
 355 }
 356
 357 static int bch2_dump_release(struct inode *inode, struct file *file)
 358 {
 359         struct dump_iter *i = file->private_data;
 360
 361         printbuf_exit(&i->buf);
 362         kfree(i);
 363         return 0;
 364 }
 365
 366 static ssize_t bch2_read_btree(struct file *file, char __user *buf,
 367                                size_t size, loff_t *ppos)
 368 {
 369         struct dump_iter *i = file->private_data;
 370
 371         i->ubuf = buf;
 372         i->size = size;
 373         i->ret  = 0;
 374
 375         return flush_buf(i) ?:
 376                 bch2_trans_run(i->c,
 377                         for_each_btree_key(trans, iter, i->id, i->from,
 378                                            BTREE_ITER_PREFETCH|
 379                                            BTREE_ITER_ALL_SNAPSHOTS, k, ({
 380                                 bch2_bkey_val_to_text(&i->buf, i->c, k);
 381                                 prt_newline(&i->buf);
 382                                 bch2_trans_unlock(trans);
 383                                 i->from = bpos_successor(iter.pos);
 384                                 flush_buf(i);
 385                         }))) ?:
 386                 i->ret;
 387 }
 388
 389 static const struct file_operations btree_debug_ops = {
 390         .owner          = THIS_MODULE,
 391         .open           = bch2_dump_open,
 392         .release        = bch2_dump_release,
 393         .read           = bch2_read_btree,
 394 };
 395
 396 static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
 397                                        size_t size, loff_t *ppos)
 398 {
 399         struct dump_iter *i = file->private_data;
 400         struct btree_trans *trans;
 401         struct btree_iter iter;
 402         struct btree *b;
 403         ssize_t ret;
 404
 405         i->ubuf = buf;
 406         i->size = size;
 407         i->ret  = 0;
 408
 409         ret = flush_buf(i);
 410         if (ret)
 411                 return ret;
 412
 413         if (bpos_eq(SPOS_MAX, i->from))
 414                 return i->ret;
 415
 416         trans = bch2_trans_get(i->c);
 417 retry:
 418         bch2_trans_begin(trans);
 419
 420         for_each_btree_node(trans, iter, i->id, i->from, 0, b, ret) {
 421                 bch2_btree_node_to_text(&i->buf, i->c, b);
 422                 i->from = !bpos_eq(SPOS_MAX, b->key.k.p)
 423                         ? bpos_successor(b->key.k.p)
 424                         : b->key.k.p;
 425
 426                 ret = drop_locks_do(trans, flush_buf(i));
 427                 if (ret)
 428                         break;
 429         }
 430         bch2_trans_iter_exit(trans, &iter);
 431
 432         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 433                 goto retry;
 434
 435         bch2_trans_put(trans);
 436
 437         if (!ret)
 438                 ret = flush_buf(i);
 439
 440         return ret ?: i->ret;
 441 }
 442
 443 static const struct file_operations btree_format_debug_ops = {
 444         .owner          = THIS_MODULE,
 445         .open           = bch2_dump_open,
 446         .release        = bch2_dump_release,
 447         .read           = bch2_read_btree_formats,
 448 };
 449
 450 static ssize_t bch2_read_bfloat_failed(struct file *file, char __user *buf,
 451                                        size_t size, loff_t *ppos)
 452 {
 453         struct dump_iter *i = file->private_data;
 454
 455         i->ubuf = buf;
 456         i->size = size;
 457         i->ret  = 0;
 458
 459         return flush_buf(i) ?:
 460                 bch2_trans_run(i->c,
 461                         for_each_btree_key(trans, iter, i->id, i->from,
 462                                            BTREE_ITER_PREFETCH|
 463                                            BTREE_ITER_ALL_SNAPSHOTS, k, ({
 464                                 struct btree_path_level *l =
 465                                         &btree_iter_path(trans, &iter)->l[0];
 466                                 struct bkey_packed *_k =
 467                                         bch2_btree_node_iter_peek(&l->iter, l->b);
 468
 469                                 if (bpos_gt(l->b->key.k.p, i->prev_node)) {
 470                                         bch2_btree_node_to_text(&i->buf, i->c, l->b);
 471                                         i->prev_node = l->b->key.k.p;
 472                                 }
 473
 474                                 bch2_bfloat_to_text(&i->buf, l->b, _k);
 475                                 bch2_trans_unlock(trans);
 476                                 i->from = bpos_successor(iter.pos);
 477                                 flush_buf(i);
 478                         }))) ?:
 479                 i->ret;
 480 }
 481
 482 static const struct file_operations bfloat_failed_debug_ops = {
 483         .owner          = THIS_MODULE,
 484         .open           = bch2_dump_open,
 485         .release        = bch2_dump_release,
 486         .read           = bch2_read_bfloat_failed,
 487 };
 488
 489 static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
 490                                            struct btree *b)
 491 {
 492         if (!out->nr_tabstops)
 493                 printbuf_tabstop_push(out, 32);
 494
 495         prt_printf(out, "%px btree=%s l=%u ",
 496                b,
 497                bch2_btree_id_str(b->c.btree_id),
 498                b->c.level);
 499         prt_newline(out);
 500
 501         printbuf_indent_add(out, 2);
 502
 503         bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
 504         prt_newline(out);
 505
 506         prt_printf(out, "flags: ");
 507         prt_tab(out);
 508         prt_bitflags(out, bch2_btree_node_flags, b->flags);
 509         prt_newline(out);
 510
 511         prt_printf(out, "pcpu read locks: ");
 512         prt_tab(out);
 513         prt_printf(out, "%u", b->c.lock.readers != NULL);
 514         prt_newline(out);
 515
 516         prt_printf(out, "written:");
 517         prt_tab(out);
 518         prt_printf(out, "%u", b->written);
 519         prt_newline(out);
 520
 521         prt_printf(out, "writes blocked:");
 522         prt_tab(out);
 523         prt_printf(out, "%u", !list_empty_careful(&b->write_blocked));
 524         prt_newline(out);
 525
 526         prt_printf(out, "will make reachable:");
 527         prt_tab(out);
 528         prt_printf(out, "%lx", b->will_make_reachable);
 529         prt_newline(out);
 530
 531         prt_printf(out, "journal pin %px:", &b->writes[0].journal);
 532         prt_tab(out);
 533         prt_printf(out, "%llu", b->writes[0].journal.seq);
 534         prt_newline(out);
 535
 536         prt_printf(out, "journal pin %px:", &b->writes[1].journal);
 537         prt_tab(out);
 538         prt_printf(out, "%llu", b->writes[1].journal.seq);
 539         prt_newline(out);
 540
 541         printbuf_indent_sub(out, 2);
 542 }
 543
 544 static ssize_t bch2_cached_btree_nodes_read(struct file *file, char __user *buf,
 545                                             size_t size, loff_t *ppos)
 546 {
 547         struct dump_iter *i = file->private_data;
 548         struct bch_fs *c = i->c;
 549         bool done = false;
 550         ssize_t ret = 0;
 551
 552         i->ubuf = buf;
 553         i->size = size;
 554         i->ret  = 0;
 555
 556         do {
 557                 struct bucket_table *tbl;
 558                 struct rhash_head *pos;
 559                 struct btree *b;
 560
 561                 ret = flush_buf(i);
 562                 if (ret)
 563                         return ret;
 564
 565                 rcu_read_lock();
 566                 i->buf.atomic++;
 567                 tbl = rht_dereference_rcu(c->btree_cache.table.tbl,
 568                                           &c->btree_cache.table);
 569                 if (i->iter < tbl->size) {
 570                         rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
 571                                 bch2_cached_btree_node_to_text(&i->buf, c, b);
 572                         i->iter++;
 573                 } else {
 574                         done = true;
 575                 }
 576                 --i->buf.atomic;
 577                 rcu_read_unlock();
 578         } while (!done);
 579
 580         if (i->buf.allocation_failure)
 581                 ret = -ENOMEM;
 582
 583         if (!ret)
 584                 ret = flush_buf(i);
 585
 586         return ret ?: i->ret;
 587 }
 588
 589 static const struct file_operations cached_btree_nodes_ops = {
 590         .owner          = THIS_MODULE,
 591         .open           = bch2_dump_open,
 592         .release        = bch2_dump_release,
 593         .read           = bch2_cached_btree_nodes_read,
 594 };
 595
 596 static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
 597                                             size_t size, loff_t *ppos)
 598 {
 599         struct dump_iter *i = file->private_data;
 600         struct bch_fs *c = i->c;
 601         struct btree_trans *trans;
 602         ssize_t ret = 0;
 603         u32 seq;
 604
 605         i->ubuf = buf;
 606         i->size = size;
 607         i->ret  = 0;
 608 restart:
 609         seqmutex_lock(&c->btree_trans_lock);
 610         list_for_each_entry(trans, &c->btree_trans_list, list) {
 611                 struct task_struct *task = READ_ONCE(trans->locking_wait.task);
 612
 613                 if (!task || task->pid <= i->iter)
 614                         continue;
 615
 616                 closure_get(&trans->ref);
 617                 seq = seqmutex_seq(&c->btree_trans_lock);
 618                 seqmutex_unlock(&c->btree_trans_lock);
 619
 620                 ret = flush_buf(i);
 621                 if (ret) {
 622                         closure_put(&trans->ref);
 623                         goto unlocked;
 624                 }
 625
 626                 bch2_btree_trans_to_text(&i->buf, trans);
 627
 628                 prt_printf(&i->buf, "backtrace:");
 629                 prt_newline(&i->buf);
 630                 printbuf_indent_add(&i->buf, 2);
 631                 bch2_prt_task_backtrace(&i->buf, task, 0, GFP_KERNEL);
 632                 printbuf_indent_sub(&i->buf, 2);
 633                 prt_newline(&i->buf);
 634
 635                 i->iter = task->pid;
 636
 637                 closure_put(&trans->ref);
 638
 639                 if (!seqmutex_relock(&c->btree_trans_lock, seq))
 640                         goto restart;
 641         }
 642         seqmutex_unlock(&c->btree_trans_lock);
 643 unlocked:
 644         if (i->buf.allocation_failure)
 645                 ret = -ENOMEM;
 646
 647         if (!ret)
 648                 ret = flush_buf(i);
 649
 650         return ret ?: i->ret;
 651 }
 652
 653 static const struct file_operations btree_transactions_ops = {
 654         .owner          = THIS_MODULE,
 655         .open           = bch2_dump_open,
 656         .release        = bch2_dump_release,
 657         .read           = bch2_btree_transactions_read,
 658 };
 659
 660 static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
 661                                       size_t size, loff_t *ppos)
 662 {
 663         struct dump_iter *i = file->private_data;
 664         struct bch_fs *c = i->c;
 665         bool done = false;
 666         int err;
 667
 668         i->ubuf = buf;
 669         i->size = size;
 670         i->ret  = 0;
 671
 672         while (1) {
 673                 err = flush_buf(i);
 674                 if (err)
 675                         return err;
 676
 677                 if (!i->size)
 678                         break;
 679
 680                 if (done)
 681                         break;
 682
 683                 done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
 684                 i->iter++;
 685         }
 686
 687         if (i->buf.allocation_failure)
 688                 return -ENOMEM;
 689
 690         return i->ret;
 691 }
 692
 693 static const struct file_operations journal_pins_ops = {
 694         .owner          = THIS_MODULE,
 695         .open           = bch2_dump_open,
 696         .release        = bch2_dump_release,
 697         .read           = bch2_journal_pins_read,
 698 };
 699
 700 static ssize_t bch2_btree_updates_read(struct file *file, char __user *buf,
 701                                        size_t size, loff_t *ppos)
 702 {
 703         struct dump_iter *i = file->private_data;
 704         struct bch_fs *c = i->c;
 705         int err;
 706
 707         i->ubuf = buf;
 708         i->size = size;
 709         i->ret  = 0;
 710
 711         if (!i->iter) {
 712                 bch2_btree_updates_to_text(&i->buf, c);
 713                 i->iter++;
 714         }
 715
 716         err = flush_buf(i);
 717         if (err)
 718                 return err;
 719
 720         if (i->buf.allocation_failure)
 721                 return -ENOMEM;
 722
 723         return i->ret;
 724 }
 725
 726 static const struct file_operations btree_updates_ops = {
 727         .owner          = THIS_MODULE,
 728         .open           = bch2_dump_open,
 729         .release        = bch2_dump_release,
 730         .read           = bch2_btree_updates_read,
 731 };
 732
 733 static int btree_transaction_stats_open(struct inode *inode, struct file *file)
 734 {
 735         struct bch_fs *c = inode->i_private;
 736         struct dump_iter *i;
 737
 738         i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
 739         if (!i)
 740                 return -ENOMEM;
 741
 742         i->iter = 1;
 743         i->c    = c;
 744         i->buf  = PRINTBUF;
 745         file->private_data = i;
 746
 747         return 0;
 748 }
 749
 750 static int btree_transaction_stats_release(struct inode *inode, struct file *file)
 751 {
 752         struct dump_iter *i = file->private_data;
 753
 754         printbuf_exit(&i->buf);
 755         kfree(i);
 756
 757         return 0;
 758 }
 759
 760 static ssize_t btree_transaction_stats_read(struct file *file, char __user *buf,
 761                                             size_t size, loff_t *ppos)
 762 {
 763         struct dump_iter        *i = file->private_data;
 764         struct bch_fs *c = i->c;
 765         int err;
 766
 767         i->ubuf = buf;
 768         i->size = size;
 769         i->ret  = 0;
 770
 771         while (1) {
 772                 struct btree_transaction_stats *s = &c->btree_transaction_stats[i->iter];
 773
 774                 err = flush_buf(i);
 775                 if (err)
 776                         return err;
 777
 778                 if (!i->size)
 779                         break;
 780
 781                 if (i->iter == ARRAY_SIZE(bch2_btree_transaction_fns) ||
 782                     !bch2_btree_transaction_fns[i->iter])
 783                         break;
 784
 785                 prt_printf(&i->buf, "%s: ", bch2_btree_transaction_fns[i->iter]);
 786                 prt_newline(&i->buf);
 787                 printbuf_indent_add(&i->buf, 2);
 788
 789                 mutex_lock(&s->lock);
 790
 791                 prt_printf(&i->buf, "Max mem used: %u", s->max_mem);
 792                 prt_newline(&i->buf);
 793
 794                 prt_printf(&i->buf, "Transaction duration:");
 795                 prt_newline(&i->buf);
 796
 797                 printbuf_indent_add(&i->buf, 2);
 798                 bch2_time_stats_to_text(&i->buf, &s->duration);
 799                 printbuf_indent_sub(&i->buf, 2);
 800
 801                 if (IS_ENABLED(CONFIG_BCACHEFS_LOCK_TIME_STATS)) {
 802                         prt_printf(&i->buf, "Lock hold times:");
 803                         prt_newline(&i->buf);
 804
 805                         printbuf_indent_add(&i->buf, 2);
 806                         bch2_time_stats_to_text(&i->buf, &s->lock_hold_times);
 807                         printbuf_indent_sub(&i->buf, 2);
 808                 }
 809
 810                 if (s->max_paths_text) {
 811                         prt_printf(&i->buf, "Maximum allocated btree paths (%u):", s->nr_max_paths);
 812                         prt_newline(&i->buf);
 813
 814                         printbuf_indent_add(&i->buf, 2);
 815                         prt_str_indented(&i->buf, s->max_paths_text);
 816                         printbuf_indent_sub(&i->buf, 2);
 817                 }
 818
 819                 mutex_unlock(&s->lock);
 820
 821                 printbuf_indent_sub(&i->buf, 2);
 822                 prt_newline(&i->buf);
 823                 i->iter++;
 824         }
 825
 826         if (i->buf.allocation_failure)
 827                 return -ENOMEM;
 828
 829         return i->ret;
 830 }
 831
 832 static const struct file_operations btree_transaction_stats_op = {
 833         .owner          = THIS_MODULE,
 834         .open           = btree_transaction_stats_open,
 835         .release        = btree_transaction_stats_release,
 836         .read           = btree_transaction_stats_read,
 837 };
 838
 839 static ssize_t bch2_btree_deadlock_read(struct file *file, char __user *buf,
 840                                             size_t size, loff_t *ppos)
 841 {
 842         struct dump_iter *i = file->private_data;
 843         struct bch_fs *c = i->c;
 844         struct btree_trans *trans;
 845         ssize_t ret = 0;
 846         u32 seq;
 847
 848         i->ubuf = buf;
 849         i->size = size;
 850         i->ret  = 0;
 851
 852         if (i->iter)
 853                 goto out;
 854 restart:
 855         seqmutex_lock(&c->btree_trans_lock);
 856         list_for_each_entry(trans, &c->btree_trans_list, list) {
 857                 struct task_struct *task = READ_ONCE(trans->locking_wait.task);
 858
 859                 if (!task || task->pid <= i->iter)
 860                         continue;
 861
 862                 closure_get(&trans->ref);
 863                 seq = seqmutex_seq(&c->btree_trans_lock);
 864                 seqmutex_unlock(&c->btree_trans_lock);
 865
 866                 ret = flush_buf(i);
 867                 if (ret) {
 868                         closure_put(&trans->ref);
 869                         goto out;
 870                 }
 871
 872                 bch2_check_for_deadlock(trans, &i->buf);
 873
 874                 i->iter = task->pid;
 875
 876                 closure_put(&trans->ref);
 877
 878                 if (!seqmutex_relock(&c->btree_trans_lock, seq))
 879                         goto restart;
 880         }
 881         seqmutex_unlock(&c->btree_trans_lock);
 882 out:
 883         if (i->buf.allocation_failure)
 884                 ret = -ENOMEM;
 885
 886         if (!ret)
 887                 ret = flush_buf(i);
 888
 889         return ret ?: i->ret;
 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
 899 void bch2_fs_debug_exit(struct bch_fs *c)
 900 {
 901         if (!IS_ERR_OR_NULL(c->fs_debug_dir))
 902                 debugfs_remove_recursive(c->fs_debug_dir);
 903 }
 904
 905 static void bch2_fs_debug_btree_init(struct bch_fs *c, struct btree_debug *bd)
 906 {
 907         struct dentry *d;
 908
 909         d = debugfs_create_dir(bch2_btree_id_str(bd->id), c->btree_debug_dir);
 910
 911         debugfs_create_file("keys", 0400, d, bd, &btree_debug_ops);
 912
 913         debugfs_create_file("formats", 0400, d, bd, &btree_format_debug_ops);
 914
 915         debugfs_create_file("bfloat-failed", 0400, d, bd,
 916                             &bfloat_failed_debug_ops);
 917 }
 918
 919 void bch2_fs_debug_init(struct bch_fs *c)
 920 {
 921         struct btree_debug *bd;
 922         char name[100];
 923
 924         if (IS_ERR_OR_NULL(bch_debug))
 925                 return;
 926
 927         snprintf(name, sizeof(name), "%pU", c->sb.user_uuid.b);
 928         c->fs_debug_dir = debugfs_create_dir(name, bch_debug);
 929         if (IS_ERR_OR_NULL(c->fs_debug_dir))
 930                 return;
 931
 932         debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
 933                             c->btree_debug, &cached_btree_nodes_ops);
 934
 935         debugfs_create_file("btree_transactions", 0400, c->fs_debug_dir,
 936                             c->btree_debug, &btree_transactions_ops);
 937
 938         debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
 939                             c->btree_debug, &journal_pins_ops);
 940
 941         debugfs_create_file("btree_updates", 0400, c->fs_debug_dir,
 942                             c->btree_debug, &btree_updates_ops);
 943
 944         debugfs_create_file("btree_transaction_stats", 0400, c->fs_debug_dir,
 945                             c, &btree_transaction_stats_op);
 946
 947         debugfs_create_file("btree_deadlock", 0400, c->fs_debug_dir,
 948                             c->btree_debug, &btree_deadlock_ops);
 949
 950         c->btree_debug_dir = debugfs_create_dir("btrees", c->fs_debug_dir);
 951         if (IS_ERR_OR_NULL(c->btree_debug_dir))
 952                 return;
 953
 954         for (bd = c->btree_debug;
 955              bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
 956              bd++) {
 957                 bd->id = bd - c->btree_debug;
 958                 bch2_fs_debug_btree_init(c, bd);
 959         }
 960 }
 961
 962 #endif
 963
 964 void bch2_debug_exit(void)
 965 {
 966         if (!IS_ERR_OR_NULL(bch_debug))
 967                 debugfs_remove_recursive(bch_debug);
 968 }
 969
 970 int __init bch2_debug_init(void)
 971 {
 972         bch_debug = debugfs_create_dir("bcachefs", NULL);
 973         return 0;
 974 }