fs/bcachefs/debug.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Assorted bcachefs debug code
   4  *
   5  * Copyright 2010, 2011 Kent Overstreet <[email protected]>
   6  * Copyright 2012 Google, Inc.
   7  */
   8
   9 #include "bcachefs.h"
  10 #include "bkey_methods.h"
  11 #include "btree_cache.h"
  12 #include "btree_io.h"
  13 #include "btree_iter.h"
  14 #include "btree_locking.h"
  15 #include "btree_update.h"
  16 #include "btree_update_interior.h"
  17 #include "buckets.h"
  18 #include "debug.h"
  19 #include "error.h"
  20 #include "extents.h"
  21 #include "fsck.h"
  22 #include "inode.h"
  23 #include "super.h"
  24
  25 #include <linux/console.h>
  26 #include <linux/debugfs.h>
  27 #include <linux/module.h>
  28 #include <linux/random.h>
  29 #include <linux/seq_file.h>
  30
  31 static struct dentry *bch_debug;
  32
  33 static bool bch2_btree_verify_replica(struct bch_fs *c, struct btree *b,
  34                                       struct extent_ptr_decoded pick)
  35 {
  36         struct btree *v = c->verify_data;
  37         struct btree_node *n_ondisk = c->verify_ondisk;
  38         struct btree_node *n_sorted = c->verify_data->data;
  39         struct bset *sorted, *inmemory = &b->data->keys;
  40         struct bio *bio;
  41         bool failed = false, saw_error = false;
  42
  43         struct bch_dev *ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ);
  44         if (!ca)
  45                 return false;
  46
  47         bio = bio_alloc_bioset(ca->disk_sb.bdev,
  48                                buf_pages(n_sorted, btree_buf_bytes(b)),
  49                                REQ_OP_READ|REQ_META,
  50                                GFP_NOFS,
  51                                &c->btree_bio);
  52         bio->bi_iter.bi_sector  = pick.ptr.offset;
  53         bch2_bio_map(bio, n_sorted, btree_buf_bytes(b));
  54
  55         submit_bio_wait(bio);
  56
  57         bio_put(bio);
  58         percpu_ref_put(&ca->io_ref);
  59
  60         memcpy(n_ondisk, n_sorted, btree_buf_bytes(b));
  61
  62         v->written = 0;
  63         if (bch2_btree_node_read_done(c, ca, v, false, &saw_error) || saw_error)
  64                 return false;
  65
  66         n_sorted = c->verify_data->data;
  67         sorted = &n_sorted->keys;
  68
  69         if (inmemory->u64s != sorted->u64s ||
  70             memcmp(inmemory->start,
  71                    sorted->start,
  72                    vstruct_end(inmemory) - (void *) inmemory->start)) {
  73                 unsigned offset = 0, sectors;
  74                 struct bset *i;
  75                 unsigned j;
  76
  77                 console_lock();
  78
  79                 printk(KERN_ERR "*** in memory:\n");
  80                 bch2_dump_bset(c, b, inmemory, 0);
  81
  82                 printk(KERN_ERR "*** read back in:\n");
  83                 bch2_dump_bset(c, v, sorted, 0);
  84
  85                 while (offset < v->written) {
  86                         if (!offset) {
  87                                 i = &n_ondisk->keys;
  88                                 sectors = vstruct_blocks(n_ondisk, c->block_bits) <<
  89                                         c->block_bits;
  90                         } else {
  91                                 struct btree_node_entry *bne =
  92                                         (void *) n_ondisk + (offset << 9);
  93                                 i = &bne->keys;
  94
  95                                 sectors = vstruct_blocks(bne, c->block_bits) <<
  96                                         c->block_bits;
  97                         }
  98
  99                         printk(KERN_ERR "*** on disk block %u:\n", offset);
 100                         bch2_dump_bset(c, b, i, offset);
 101
 102                         offset += sectors;
 103                 }
 104
 105                 for (j = 0; j < le16_to_cpu(inmemory->u64s); j++)
 106                         if (inmemory->_data[j] != sorted->_data[j])
 107                                 break;
 108
 109                 console_unlock();
 110                 bch_err(c, "verify failed at key %u", j);
 111
 112                 failed = true;
 113         }
 114
 115         if (v->written != b->written) {
 116                 bch_err(c, "written wrong: expected %u, got %u",
 117                         b->written, v->written);
 118                 failed = true;
 119         }
 120
 121         return failed;
 122 }
 123
 124 void __bch2_btree_verify(struct bch_fs *c, struct btree *b)
 125 {
 126         struct bkey_ptrs_c ptrs;
 127         struct extent_ptr_decoded p;
 128         const union bch_extent_entry *entry;
 129         struct btree *v;
 130         struct bset *inmemory = &b->data->keys;
 131         struct bkey_packed *k;
 132         bool failed = false;
 133
 134         if (c->opts.nochanges)
 135                 return;
 136
 137         bch2_btree_node_io_lock(b);
 138         mutex_lock(&c->verify_lock);
 139
 140         if (!c->verify_ondisk) {
 141                 c->verify_ondisk = kvmalloc(btree_buf_bytes(b), GFP_KERNEL);
 142                 if (!c->verify_ondisk)
 143                         goto out;
 144         }
 145
 146         if (!c->verify_data) {
 147                 c->verify_data = __bch2_btree_node_mem_alloc(c);
 148                 if (!c->verify_data)
 149                         goto out;
 150
 151                 list_del_init(&c->verify_data->list);
 152         }
 153
 154         BUG_ON(b->nsets != 1);
 155
 156         for (k = inmemory->start; k != vstruct_last(inmemory); k = bkey_p_next(k))
 157                 if (k->type == KEY_TYPE_btree_ptr_v2)
 158                         ((struct bch_btree_ptr_v2 *) bkeyp_val(&b->format, k))->mem_ptr = 0;
 159
 160         v = c->verify_data;
 161         bkey_copy(&v->key, &b->key);
 162         v->c.level      = b->c.level;
 163         v->c.btree_id   = b->c.btree_id;
 164         bch2_btree_keys_init(v);
 165
 166         ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(&b->key));
 167         bkey_for_each_ptr_decode(&b->key.k, ptrs, p, entry)
 168                 failed |= bch2_btree_verify_replica(c, b, p);
 169
 170         if (failed) {
 171                 struct printbuf buf = PRINTBUF;
 172
 173                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
 174                 bch2_fs_fatal_error(c, ": btree node verify failed for: %s\n", buf.buf);
 175                 printbuf_exit(&buf);
 176         }
 177 out:
 178         mutex_unlock(&c->verify_lock);
 179         bch2_btree_node_io_unlock(b);
 180 }
 181
 182 void bch2_btree_node_ondisk_to_text(struct printbuf *out, struct bch_fs *c,
 183                                     const struct btree *b)
 184 {
 185         struct btree_node *n_ondisk = NULL;
 186         struct extent_ptr_decoded pick;
 187         struct bch_dev *ca;
 188         struct bio *bio = NULL;
 189         unsigned offset = 0;
 190         int ret;
 191
 192         if (bch2_bkey_pick_read_device(c, bkey_i_to_s_c(&b->key), NULL, &pick) <= 0) {
 193                 prt_printf(out, "error getting device to read from: invalid device\n");
 194                 return;
 195         }
 196
 197         ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ);
 198         if (!ca) {
 199                 prt_printf(out, "error getting device to read from: not online\n");
 200                 return;
 201         }
 202
 203         n_ondisk = kvmalloc(btree_buf_bytes(b), GFP_KERNEL);
 204         if (!n_ondisk) {
 205                 prt_printf(out, "memory allocation failure\n");
 206                 goto out;
 207         }
 208
 209         bio = bio_alloc_bioset(ca->disk_sb.bdev,
 210                                buf_pages(n_ondisk, btree_buf_bytes(b)),
 211                                REQ_OP_READ|REQ_META,
 212                                GFP_NOFS,
 213                                &c->btree_bio);
 214         bio->bi_iter.bi_sector  = pick.ptr.offset;
 215         bch2_bio_map(bio, n_ondisk, btree_buf_bytes(b));
 216
 217         ret = submit_bio_wait(bio);
 218         if (ret) {
 219                 prt_printf(out, "IO error reading btree node: %s\n", bch2_err_str(ret));
 220                 goto out;
 221         }
 222
 223         while (offset < btree_sectors(c)) {
 224                 struct bset *i;
 225                 struct nonce nonce;
 226                 struct bch_csum csum;
 227                 struct bkey_packed *k;
 228                 unsigned sectors;
 229
 230                 if (!offset) {
 231                         i = &n_ondisk->keys;
 232
 233                         if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
 234                                 prt_printf(out, "unknown checksum type at offset %u: %llu\n",
 235                                            offset, BSET_CSUM_TYPE(i));
 236                                 goto out;
 237                         }
 238
 239                         nonce = btree_nonce(i, offset << 9);
 240                         csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, n_ondisk);
 241
 242                         if (bch2_crc_cmp(csum, n_ondisk->csum)) {
 243                                 prt_printf(out, "invalid checksum\n");
 244                                 goto out;
 245                         }
 246
 247                         bset_encrypt(c, i, offset << 9);
 248
 249                         sectors = vstruct_sectors(n_ondisk, c->block_bits);
 250                 } else {
 251                         struct btree_node_entry *bne = (void *) n_ondisk + (offset << 9);
 252
 253                         i = &bne->keys;
 254
 255                         if (i->seq != n_ondisk->keys.seq)
 256                                 break;
 257
 258                         if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
 259                                 prt_printf(out, "unknown checksum type at offset %u: %llu\n",
 260                                            offset, BSET_CSUM_TYPE(i));
 261                                 goto out;
 262                         }
 263
 264                         nonce = btree_nonce(i, offset << 9);
 265                         csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne);
 266
 267                         if (bch2_crc_cmp(csum, bne->csum)) {
 268                                 prt_printf(out, "invalid checksum");
 269                                 goto out;
 270                         }
 271
 272                         bset_encrypt(c, i, offset << 9);
 273
 274                         sectors = vstruct_sectors(bne, c->block_bits);
 275                 }
 276
 277                 prt_printf(out, "  offset %u version %u, journal seq %llu\n",
 278                            offset,
 279                            le16_to_cpu(i->version),
 280                            le64_to_cpu(i->journal_seq));
 281                 offset += sectors;
 282
 283                 printbuf_indent_add(out, 4);
 284
 285                 for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k)) {
 286                         struct bkey u;
 287
 288                         bch2_bkey_val_to_text(out, c, bkey_disassemble(b, k, &u));
 289                         prt_newline(out);
 290                 }
 291
 292                 printbuf_indent_sub(out, 4);
 293         }
 294 out:
 295         if (bio)
 296                 bio_put(bio);
 297         kvfree(n_ondisk);
 298         percpu_ref_put(&ca->io_ref);
 299 }
 300
 301 #ifdef CONFIG_DEBUG_FS
 302
 303 /* XXX: bch_fs refcounting */
 304
 305 struct dump_iter {
 306         struct bch_fs           *c;
 307         enum btree_id           id;
 308         struct bpos             from;
 309         struct bpos             prev_node;
 310         u64                     iter;
 311
 312         struct printbuf         buf;
 313
 314         char __user             *ubuf;  /* destination user buffer */
 315         size_t                  size;   /* size of requested read */
 316         ssize_t                 ret;    /* bytes read so far */
 317 };
 318
 319 static ssize_t flush_buf(struct dump_iter *i)
 320 {
 321         if (i->buf.pos) {
 322                 size_t bytes = min_t(size_t, i->buf.pos, i->size);
 323                 int copied = bytes - copy_to_user(i->ubuf, i->buf.buf, bytes);
 324
 325                 i->ret   += copied;
 326                 i->ubuf  += copied;
 327                 i->size  -= copied;
 328                 i->buf.pos -= copied;
 329                 memmove(i->buf.buf, i->buf.buf + copied, i->buf.pos);
 330
 331                 if (copied != bytes)
 332                         return -EFAULT;
 333         }
 334
 335         return i->size ? 0 : i->ret;
 336 }
 337
 338 static int bch2_dump_open(struct inode *inode, struct file *file)
 339 {
 340         struct btree_debug *bd = inode->i_private;
 341         struct dump_iter *i;
 342
 343         i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
 344         if (!i)
 345                 return -ENOMEM;
 346
 347         file->private_data = i;
 348         i->from = POS_MIN;
 349         i->iter = 0;
 350         i->c    = container_of(bd, struct bch_fs, btree_debug[bd->id]);
 351         i->id   = bd->id;
 352         i->buf  = PRINTBUF;
 353
 354         return 0;
 355 }
 356
 357 static int bch2_dump_release(struct inode *inode, struct file *file)
 358 {
 359         struct dump_iter *i = file->private_data;
 360
 361         printbuf_exit(&i->buf);
 362         kfree(i);
 363         return 0;
 364 }
 365
 366 static ssize_t bch2_read_btree(struct file *file, char __user *buf,
 367                                size_t size, loff_t *ppos)
 368 {
 369         struct dump_iter *i = file->private_data;
 370
 371         i->ubuf = buf;
 372         i->size = size;
 373         i->ret  = 0;
 374
 375         return flush_buf(i) ?:
 376                 bch2_trans_run(i->c,
 377                         for_each_btree_key(trans, iter, i->id, i->from,
 378                                            BTREE_ITER_prefetch|
 379                                            BTREE_ITER_all_snapshots, k, ({
 380                                 bch2_bkey_val_to_text(&i->buf, i->c, k);
 381                                 prt_newline(&i->buf);
 382                                 bch2_trans_unlock(trans);
 383                                 i->from = bpos_successor(iter.pos);
 384                                 flush_buf(i);
 385                         }))) ?:
 386                 i->ret;
 387 }
 388
 389 static const struct file_operations btree_debug_ops = {
 390         .owner          = THIS_MODULE,
 391         .open           = bch2_dump_open,
 392         .release        = bch2_dump_release,
 393         .read           = bch2_read_btree,
 394 };
 395
 396 static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
 397                                        size_t size, loff_t *ppos)
 398 {
 399         struct dump_iter *i = file->private_data;
 400
 401         i->ubuf = buf;
 402         i->size = size;
 403         i->ret  = 0;
 404
 405         ssize_t ret = flush_buf(i);
 406         if (ret)
 407                 return ret;
 408
 409         if (bpos_eq(SPOS_MAX, i->from))
 410                 return i->ret;
 411
 412         return bch2_trans_run(i->c,
 413                 for_each_btree_node(trans, iter, i->id, i->from, 0, b, ({
 414                         bch2_btree_node_to_text(&i->buf, i->c, b);
 415                         i->from = !bpos_eq(SPOS_MAX, b->key.k.p)
 416                                 ? bpos_successor(b->key.k.p)
 417                                 : b->key.k.p;
 418
 419                         drop_locks_do(trans, flush_buf(i));
 420                 }))) ?: i->ret;
 421 }
 422
 423 static const struct file_operations btree_format_debug_ops = {
 424         .owner          = THIS_MODULE,
 425         .open           = bch2_dump_open,
 426         .release        = bch2_dump_release,
 427         .read           = bch2_read_btree_formats,
 428 };
 429
 430 static ssize_t bch2_read_bfloat_failed(struct file *file, char __user *buf,
 431                                        size_t size, loff_t *ppos)
 432 {
 433         struct dump_iter *i = file->private_data;
 434
 435         i->ubuf = buf;
 436         i->size = size;
 437         i->ret  = 0;
 438
 439         return flush_buf(i) ?:
 440                 bch2_trans_run(i->c,
 441                         for_each_btree_key(trans, iter, i->id, i->from,
 442                                            BTREE_ITER_prefetch|
 443                                            BTREE_ITER_all_snapshots, k, ({
 444                                 struct btree_path_level *l =
 445                                         &btree_iter_path(trans, &iter)->l[0];
 446                                 struct bkey_packed *_k =
 447                                         bch2_btree_node_iter_peek(&l->iter, l->b);
 448
 449                                 if (bpos_gt(l->b->key.k.p, i->prev_node)) {
 450                                         bch2_btree_node_to_text(&i->buf, i->c, l->b);
 451                                         i->prev_node = l->b->key.k.p;
 452                                 }
 453
 454                                 bch2_bfloat_to_text(&i->buf, l->b, _k);
 455                                 bch2_trans_unlock(trans);
 456                                 i->from = bpos_successor(iter.pos);
 457                                 flush_buf(i);
 458                         }))) ?:
 459                 i->ret;
 460 }
 461
 462 static const struct file_operations bfloat_failed_debug_ops = {
 463         .owner          = THIS_MODULE,
 464         .open           = bch2_dump_open,
 465         .release        = bch2_dump_release,
 466         .read           = bch2_read_bfloat_failed,
 467 };
 468
 469 static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
 470                                            struct btree *b)
 471 {
 472         if (!out->nr_tabstops)
 473                 printbuf_tabstop_push(out, 32);
 474
 475         prt_printf(out, "%px btree=%s l=%u\n", b, bch2_btree_id_str(b->c.btree_id), b->c.level);
 476
 477         printbuf_indent_add(out, 2);
 478
 479         bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
 480         prt_newline(out);
 481
 482         prt_printf(out, "flags:\t");
 483         prt_bitflags(out, bch2_btree_node_flags, b->flags);
 484         prt_newline(out);
 485
 486         prt_printf(out, "pcpu read locks:\t%u\n",       b->c.lock.readers != NULL);
 487         prt_printf(out, "written:\t%u\n",               b->written);
 488         prt_printf(out, "writes blocked:\t%u\n",        !list_empty_careful(&b->write_blocked));
 489         prt_printf(out, "will make reachable:\t%lx\n",  b->will_make_reachable);
 490
 491         prt_printf(out, "journal pin %px:\t%llu\n",
 492                    &b->writes[0].journal, b->writes[0].journal.seq);
 493         prt_printf(out, "journal pin %px:\t%llu\n",
 494                    &b->writes[1].journal, b->writes[1].journal.seq);
 495
 496         printbuf_indent_sub(out, 2);
 497 }
 498
 499 static ssize_t bch2_cached_btree_nodes_read(struct file *file, char __user *buf,
 500                                             size_t size, loff_t *ppos)
 501 {
 502         struct dump_iter *i = file->private_data;
 503         struct bch_fs *c = i->c;
 504         bool done = false;
 505         ssize_t ret = 0;
 506
 507         i->ubuf = buf;
 508         i->size = size;
 509         i->ret  = 0;
 510
 511         do {
 512                 struct bucket_table *tbl;
 513                 struct rhash_head *pos;
 514                 struct btree *b;
 515
 516                 ret = flush_buf(i);
 517                 if (ret)
 518                         return ret;
 519
 520                 rcu_read_lock();
 521                 i->buf.atomic++;
 522                 tbl = rht_dereference_rcu(c->btree_cache.table.tbl,
 523                                           &c->btree_cache.table);
 524                 if (i->iter < tbl->size) {
 525                         rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
 526                                 bch2_cached_btree_node_to_text(&i->buf, c, b);
 527                         i->iter++;
 528                 } else {
 529                         done = true;
 530                 }
 531                 --i->buf.atomic;
 532                 rcu_read_unlock();
 533         } while (!done);
 534
 535         if (i->buf.allocation_failure)
 536                 ret = -ENOMEM;
 537
 538         if (!ret)
 539                 ret = flush_buf(i);
 540
 541         return ret ?: i->ret;
 542 }
 543
 544 static const struct file_operations cached_btree_nodes_ops = {
 545         .owner          = THIS_MODULE,
 546         .open           = bch2_dump_open,
 547         .release        = bch2_dump_release,
 548         .read           = bch2_cached_btree_nodes_read,
 549 };
 550
 551 typedef int (*list_cmp_fn)(const struct list_head *l, const struct list_head *r);
 552
 553 static void list_sort(struct list_head *head, list_cmp_fn cmp)
 554 {
 555         struct list_head *pos;
 556
 557         list_for_each(pos, head)
 558                 while (!list_is_last(pos, head) &&
 559                        cmp(pos, pos->next) > 0) {
 560                         struct list_head *pos2, *next = pos->next;
 561
 562                         list_del(next);
 563                         list_for_each(pos2, head)
 564                                 if (cmp(next, pos2) < 0)
 565                                         goto pos_found;
 566                         BUG();
 567 pos_found:
 568                         list_add_tail(next, pos2);
 569                 }
 570 }
 571
 572 static int list_ptr_order_cmp(const struct list_head *l, const struct list_head *r)
 573 {
 574         return cmp_int(l, r);
 575 }
 576
 577 static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
 578                                             size_t size, loff_t *ppos)
 579 {
 580         struct dump_iter *i = file->private_data;
 581         struct bch_fs *c = i->c;
 582         struct btree_trans *trans;
 583         ssize_t ret = 0;
 584
 585         i->ubuf = buf;
 586         i->size = size;
 587         i->ret  = 0;
 588 restart:
 589         seqmutex_lock(&c->btree_trans_lock);
 590         list_sort(&c->btree_trans_list, list_ptr_order_cmp);
 591
 592         list_for_each_entry(trans, &c->btree_trans_list, list) {
 593                 if ((ulong) trans <= i->iter)
 594                         continue;
 595
 596                 i->iter = (ulong) trans;
 597
 598                 if (!closure_get_not_zero(&trans->ref))
 599                         continue;
 600
 601                 u32 seq = seqmutex_unlock(&c->btree_trans_lock);
 602
 603                 bch2_btree_trans_to_text(&i->buf, trans);
 604
 605                 prt_printf(&i->buf, "backtrace:\n");
 606                 printbuf_indent_add(&i->buf, 2);
 607                 bch2_prt_task_backtrace(&i->buf, trans->locking_wait.task, 0, GFP_KERNEL);
 608                 printbuf_indent_sub(&i->buf, 2);
 609                 prt_newline(&i->buf);
 610
 611                 closure_put(&trans->ref);
 612
 613                 ret = flush_buf(i);
 614                 if (ret)
 615                         goto unlocked;
 616
 617                 if (!seqmutex_relock(&c->btree_trans_lock, seq))
 618                         goto restart;
 619         }
 620         seqmutex_unlock(&c->btree_trans_lock);
 621 unlocked:
 622         if (i->buf.allocation_failure)
 623                 ret = -ENOMEM;
 624
 625         if (!ret)
 626                 ret = flush_buf(i);
 627
 628         return ret ?: i->ret;
 629 }
 630
 631 static const struct file_operations btree_transactions_ops = {
 632         .owner          = THIS_MODULE,
 633         .open           = bch2_dump_open,
 634         .release        = bch2_dump_release,
 635         .read           = bch2_btree_transactions_read,
 636 };
 637
 638 static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
 639                                       size_t size, loff_t *ppos)
 640 {
 641         struct dump_iter *i = file->private_data;
 642         struct bch_fs *c = i->c;
 643         bool done = false;
 644         int err;
 645
 646         i->ubuf = buf;
 647         i->size = size;
 648         i->ret  = 0;
 649
 650         while (1) {
 651                 err = flush_buf(i);
 652                 if (err)
 653                         return err;
 654
 655                 if (!i->size)
 656                         break;
 657
 658                 if (done)
 659                         break;
 660
 661                 done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
 662                 i->iter++;
 663         }
 664
 665         if (i->buf.allocation_failure)
 666                 return -ENOMEM;
 667
 668         return i->ret;
 669 }
 670
 671 static const struct file_operations journal_pins_ops = {
 672         .owner          = THIS_MODULE,
 673         .open           = bch2_dump_open,
 674         .release        = bch2_dump_release,
 675         .read           = bch2_journal_pins_read,
 676 };
 677
 678 static ssize_t bch2_btree_updates_read(struct file *file, char __user *buf,
 679                                        size_t size, loff_t *ppos)
 680 {
 681         struct dump_iter *i = file->private_data;
 682         struct bch_fs *c = i->c;
 683         int err;
 684
 685         i->ubuf = buf;
 686         i->size = size;
 687         i->ret  = 0;
 688
 689         if (!i->iter) {
 690                 bch2_btree_updates_to_text(&i->buf, c);
 691                 i->iter++;
 692         }
 693
 694         err = flush_buf(i);
 695         if (err)
 696                 return err;
 697
 698         if (i->buf.allocation_failure)
 699                 return -ENOMEM;
 700
 701         return i->ret;
 702 }
 703
 704 static const struct file_operations btree_updates_ops = {
 705         .owner          = THIS_MODULE,
 706         .open           = bch2_dump_open,
 707         .release        = bch2_dump_release,
 708         .read           = bch2_btree_updates_read,
 709 };
 710
 711 static int btree_transaction_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         if (!i)
 718                 return -ENOMEM;
 719
 720         i->iter = 1;
 721         i->c    = c;
 722         i->buf  = PRINTBUF;
 723         file->private_data = i;
 724
 725         return 0;
 726 }
 727
 728 static int btree_transaction_stats_release(struct inode *inode, struct file *file)
 729 {
 730         struct dump_iter *i = file->private_data;
 731
 732         printbuf_exit(&i->buf);
 733         kfree(i);
 734
 735         return 0;
 736 }
 737
 738 static ssize_t btree_transaction_stats_read(struct file *file, char __user *buf,
 739                                             size_t size, loff_t *ppos)
 740 {
 741         struct dump_iter        *i = file->private_data;
 742         struct bch_fs *c = i->c;
 743         int err;
 744
 745         i->ubuf = buf;
 746         i->size = size;
 747         i->ret  = 0;
 748
 749         while (1) {
 750                 struct btree_transaction_stats *s = &c->btree_transaction_stats[i->iter];
 751
 752                 err = flush_buf(i);
 753                 if (err)
 754                         return err;
 755
 756                 if (!i->size)
 757                         break;
 758
 759                 if (i->iter == ARRAY_SIZE(bch2_btree_transaction_fns) ||
 760                     !bch2_btree_transaction_fns[i->iter])
 761                         break;
 762
 763                 prt_printf(&i->buf, "%s:\n", bch2_btree_transaction_fns[i->iter]);
 764                 printbuf_indent_add(&i->buf, 2);
 765
 766                 mutex_lock(&s->lock);
 767
 768                 prt_printf(&i->buf, "Max mem used: %u\n", s->max_mem);
 769                 prt_printf(&i->buf, "Transaction duration:\n");
 770
 771                 printbuf_indent_add(&i->buf, 2);
 772                 bch2_time_stats_to_text(&i->buf, &s->duration);
 773                 printbuf_indent_sub(&i->buf, 2);
 774
 775                 if (IS_ENABLED(CONFIG_BCACHEFS_LOCK_TIME_STATS)) {
 776                         prt_printf(&i->buf, "Lock hold times:\n");
 777
 778                         printbuf_indent_add(&i->buf, 2);
 779                         bch2_time_stats_to_text(&i->buf, &s->lock_hold_times);
 780                         printbuf_indent_sub(&i->buf, 2);
 781                 }
 782
 783                 if (s->max_paths_text) {
 784                         prt_printf(&i->buf, "Maximum allocated btree paths (%u):\n", s->nr_max_paths);
 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 btree_transaction_stats_op = {
 805         .owner          = THIS_MODULE,
 806         .open           = btree_transaction_stats_open,
 807         .release        = btree_transaction_stats_release,
 808         .read           = btree_transaction_stats_read,
 809 };
 810
 811 /* walk btree transactions until we find a deadlock and print it */
 812 static void btree_deadlock_to_text(struct printbuf *out, struct bch_fs *c)
 813 {
 814         struct btree_trans *trans;
 815         ulong iter = 0;
 816 restart:
 817         seqmutex_lock(&c->btree_trans_lock);
 818         list_sort(&c->btree_trans_list, list_ptr_order_cmp);
 819
 820         list_for_each_entry(trans, &c->btree_trans_list, list) {
 821                 if ((ulong) trans <= iter)
 822                         continue;
 823
 824                 iter = (ulong) trans;
 825
 826                 if (!closure_get_not_zero(&trans->ref))
 827                         continue;
 828
 829                 u32 seq = seqmutex_unlock(&c->btree_trans_lock);
 830
 831                 bool found = bch2_check_for_deadlock(trans, out) != 0;
 832
 833                 closure_put(&trans->ref);
 834
 835                 if (found)
 836                         return;
 837
 838                 if (!seqmutex_relock(&c->btree_trans_lock, seq))
 839                         goto restart;
 840         }
 841         seqmutex_unlock(&c->btree_trans_lock);
 842 }
 843
 844 static ssize_t bch2_btree_deadlock_read(struct file *file, char __user *buf,
 845                                             size_t size, loff_t *ppos)
 846 {
 847         struct dump_iter *i = file->private_data;
 848         struct bch_fs *c = i->c;
 849         ssize_t ret = 0;
 850
 851         i->ubuf = buf;
 852         i->size = size;
 853         i->ret  = 0;
 854
 855         if (!i->iter) {
 856                 btree_deadlock_to_text(&i->buf, c);
 857                 i->iter++;
 858         }
 859
 860         if (i->buf.allocation_failure)
 861                 ret = -ENOMEM;
 862
 863         if (!ret)
 864                 ret = flush_buf(i);
 865
 866         return ret ?: i->ret;
 867 }
 868
 869 static const struct file_operations btree_deadlock_ops = {
 870         .owner          = THIS_MODULE,
 871         .open           = bch2_dump_open,
 872         .release        = bch2_dump_release,
 873         .read           = bch2_btree_deadlock_read,
 874 };
 875
 876 void bch2_fs_debug_exit(struct bch_fs *c)
 877 {
 878         if (!IS_ERR_OR_NULL(c->fs_debug_dir))
 879                 debugfs_remove_recursive(c->fs_debug_dir);
 880 }
 881
 882 static void bch2_fs_debug_btree_init(struct bch_fs *c, struct btree_debug *bd)
 883 {
 884         struct dentry *d;
 885
 886         d = debugfs_create_dir(bch2_btree_id_str(bd->id), c->btree_debug_dir);
 887
 888         debugfs_create_file("keys", 0400, d, bd, &btree_debug_ops);
 889
 890         debugfs_create_file("formats", 0400, d, bd, &btree_format_debug_ops);
 891
 892         debugfs_create_file("bfloat-failed", 0400, d, bd,
 893                             &bfloat_failed_debug_ops);
 894 }
 895
 896 void bch2_fs_debug_init(struct bch_fs *c)
 897 {
 898         struct btree_debug *bd;
 899         char name[100];
 900
 901         if (IS_ERR_OR_NULL(bch_debug))
 902                 return;
 903
 904         snprintf(name, sizeof(name), "%pU", c->sb.user_uuid.b);
 905         c->fs_debug_dir = debugfs_create_dir(name, bch_debug);
 906         if (IS_ERR_OR_NULL(c->fs_debug_dir))
 907                 return;
 908
 909         debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
 910                             c->btree_debug, &cached_btree_nodes_ops);
 911
 912         debugfs_create_file("btree_transactions", 0400, c->fs_debug_dir,
 913                             c->btree_debug, &btree_transactions_ops);
 914
 915         debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
 916                             c->btree_debug, &journal_pins_ops);
 917
 918         debugfs_create_file("btree_updates", 0400, c->fs_debug_dir,
 919                             c->btree_debug, &btree_updates_ops);
 920
 921         debugfs_create_file("btree_transaction_stats", 0400, c->fs_debug_dir,
 922                             c, &btree_transaction_stats_op);
 923
 924         debugfs_create_file("btree_deadlock", 0400, c->fs_debug_dir,
 925                             c->btree_debug, &btree_deadlock_ops);
 926
 927         c->btree_debug_dir = debugfs_create_dir("btrees", c->fs_debug_dir);
 928         if (IS_ERR_OR_NULL(c->btree_debug_dir))
 929                 return;
 930
 931         for (bd = c->btree_debug;
 932              bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
 933              bd++) {
 934                 bd->id = bd - c->btree_debug;
 935                 bch2_fs_debug_btree_init(c, bd);
 936         }
 937 }
 938
 939 #endif
 940
 941 void bch2_debug_exit(void)
 942 {
 943         if (!IS_ERR_OR_NULL(bch_debug))
 944                 debugfs_remove_recursive(bch_debug);
 945 }
 946
 947 int __init bch2_debug_init(void)
 948 {
 949         bch_debug = debugfs_create_dir("bcachefs", NULL);
 950         return 0;
 951 }