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_update.h"
  15 #include "buckets.h"
  16 #include "debug.h"
  17 #include "error.h"
  18 #include "extents.h"
  19 #include "fsck.h"
  20 #include "inode.h"
  21 #include "io.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;
  42
  43         if (!bch2_dev_get_ioref(ca, READ))
  44                 return false;
  45
  46         bio = bio_alloc_bioset(GFP_NOIO,
  47                         buf_pages(n_sorted, btree_bytes(c)),
  48                         &c->btree_bio);
  49         bio_set_dev(bio, ca->disk_sb.bdev);
  50         bio->bi_opf             = REQ_OP_READ|REQ_META;
  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))
  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_next(k))
 156                 if (k->type == KEY_TYPE_btree_ptr_v2) {
 157                         struct bch_btree_ptr_v2 *v = (void *) bkeyp_val(&b->format, k);
 158                         v->mem_ptr = 0;
 159                 }
 160
 161         v = c->verify_data;
 162         bkey_copy(&v->key, &b->key);
 163         v->c.level      = b->c.level;
 164         v->c.btree_id   = b->c.btree_id;
 165         bch2_btree_keys_init(v);
 166
 167         ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(&b->key));
 168         bkey_for_each_ptr_decode(&b->key.k, ptrs, p, entry)
 169                 failed |= bch2_btree_verify_replica(c, b, p);
 170
 171         if (failed) {
 172                 struct printbuf buf = PRINTBUF;
 173
 174                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
 175                 bch2_fs_fatal_error(c, "btree node verify failed for : %s\n", buf.buf);
 176                 printbuf_exit(&buf);
 177         }
 178 out:
 179         mutex_unlock(&c->verify_lock);
 180         bch2_btree_node_io_unlock(b);
 181 }
 182
 183 #ifdef CONFIG_DEBUG_FS
 184
 185 /* XXX: bch_fs refcounting */
 186
 187 struct dump_iter {
 188         struct bch_fs           *c;
 189         enum btree_id           id;
 190         struct bpos             from;
 191         u64                     iter;
 192
 193         struct printbuf         buf;
 194
 195         char __user             *ubuf;  /* destination user buffer */
 196         size_t                  size;   /* size of requested read */
 197         ssize_t                 ret;    /* bytes read so far */
 198 };
 199
 200 static int flush_buf(struct dump_iter *i)
 201 {
 202         if (i->buf.pos) {
 203                 size_t bytes = min_t(size_t, i->buf.pos, i->size);
 204                 int err = copy_to_user(i->ubuf, i->buf.buf, bytes);
 205
 206                 if (err)
 207                         return err;
 208
 209                 i->ret   += bytes;
 210                 i->ubuf  += bytes;
 211                 i->size  -= bytes;
 212                 i->buf.pos -= bytes;
 213                 memmove(i->buf.buf, i->buf.buf + bytes, i->buf.pos);
 214         }
 215
 216         return 0;
 217 }
 218
 219 static int bch2_dump_open(struct inode *inode, struct file *file)
 220 {
 221         struct btree_debug *bd = inode->i_private;
 222         struct dump_iter *i;
 223
 224         i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
 225         if (!i)
 226                 return -ENOMEM;
 227
 228         file->private_data = i;
 229         i->from = POS_MIN;
 230         i->iter = 0;
 231         i->c    = container_of(bd, struct bch_fs, btree_debug[bd->id]);
 232         i->id   = bd->id;
 233         i->buf  = PRINTBUF;
 234
 235         return 0;
 236 }
 237
 238 static int bch2_dump_release(struct inode *inode, struct file *file)
 239 {
 240         struct dump_iter *i = file->private_data;
 241
 242         printbuf_exit(&i->buf);
 243         kfree(i);
 244         return 0;
 245 }
 246
 247 static ssize_t bch2_read_btree(struct file *file, char __user *buf,
 248                                size_t size, loff_t *ppos)
 249 {
 250         struct dump_iter *i = file->private_data;
 251         struct btree_trans trans;
 252         struct btree_iter iter;
 253         struct bkey_s_c k;
 254         int err;
 255
 256         i->ubuf = buf;
 257         i->size = size;
 258         i->ret  = 0;
 259
 260         err = flush_buf(i);
 261         if (err)
 262                 return err;
 263
 264         if (!i->size)
 265                 return i->ret;
 266
 267         bch2_trans_init(&trans, i->c, 0, 0);
 268
 269         bch2_trans_iter_init(&trans, &iter, i->id, i->from,
 270                              BTREE_ITER_PREFETCH|
 271                              BTREE_ITER_ALL_SNAPSHOTS);
 272         k = bch2_btree_iter_peek(&iter);
 273
 274         while (k.k && !(err = bkey_err(k))) {
 275                 bch2_bkey_val_to_text(&i->buf, i->c, k);
 276                 pr_char(&i->buf, '\n');
 277
 278                 k = bch2_btree_iter_next(&iter);
 279                 i->from = iter.pos;
 280
 281                 err = flush_buf(i);
 282                 if (err)
 283                         break;
 284
 285                 if (!i->size)
 286                         break;
 287         }
 288         bch2_trans_iter_exit(&trans, &iter);
 289
 290         bch2_trans_exit(&trans);
 291
 292         return err < 0 ? err : i->ret;
 293 }
 294
 295 static const struct file_operations btree_debug_ops = {
 296         .owner          = THIS_MODULE,
 297         .open           = bch2_dump_open,
 298         .release        = bch2_dump_release,
 299         .read           = bch2_read_btree,
 300 };
 301
 302 static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
 303                                        size_t size, loff_t *ppos)
 304 {
 305         struct dump_iter *i = file->private_data;
 306         struct btree_trans trans;
 307         struct btree_iter iter;
 308         struct btree *b;
 309         int err;
 310
 311         i->ubuf = buf;
 312         i->size = size;
 313         i->ret  = 0;
 314
 315         err = flush_buf(i);
 316         if (err)
 317                 return err;
 318
 319         if (!i->size || !bpos_cmp(SPOS_MAX, i->from))
 320                 return i->ret;
 321
 322         bch2_trans_init(&trans, i->c, 0, 0);
 323
 324         for_each_btree_node(&trans, iter, i->id, i->from, 0, b, err) {
 325                 bch2_btree_node_to_text(&i->buf, i->c, b);
 326                 err = flush_buf(i);
 327                 if (err)
 328                         break;
 329
 330                 /*
 331                  * can't easily correctly restart a btree node traversal across
 332                  * all nodes, meh
 333                  */
 334                 i->from = bpos_cmp(SPOS_MAX, b->key.k.p)
 335                         ? bpos_successor(b->key.k.p)
 336                         : b->key.k.p;
 337
 338                 if (!i->size)
 339                         break;
 340         }
 341         bch2_trans_iter_exit(&trans, &iter);
 342
 343         bch2_trans_exit(&trans);
 344
 345         return err < 0 ? err : i->ret;
 346 }
 347
 348 static const struct file_operations btree_format_debug_ops = {
 349         .owner          = THIS_MODULE,
 350         .open           = bch2_dump_open,
 351         .release        = bch2_dump_release,
 352         .read           = bch2_read_btree_formats,
 353 };
 354
 355 static ssize_t bch2_read_bfloat_failed(struct file *file, char __user *buf,
 356                                        size_t size, loff_t *ppos)
 357 {
 358         struct dump_iter *i = file->private_data;
 359         struct btree_trans trans;
 360         struct btree_iter iter;
 361         struct bkey_s_c k;
 362         struct btree *prev_node = NULL;
 363         int err;
 364
 365         i->ubuf = buf;
 366         i->size = size;
 367         i->ret  = 0;
 368
 369         err = flush_buf(i);
 370         if (err)
 371                 return err;
 372
 373         if (!i->size)
 374                 return i->ret;
 375
 376         bch2_trans_init(&trans, i->c, 0, 0);
 377
 378         bch2_trans_iter_init(&trans, &iter, i->id, i->from,
 379                              BTREE_ITER_PREFETCH|
 380                              BTREE_ITER_ALL_SNAPSHOTS);
 381
 382         while ((k = bch2_btree_iter_peek(&iter)).k &&
 383                !(err = bkey_err(k))) {
 384                 struct btree_path_level *l = &iter.path->l[0];
 385                 struct bkey_packed *_k =
 386                         bch2_btree_node_iter_peek(&l->iter, l->b);
 387
 388                 if (l->b != prev_node) {
 389                         bch2_btree_node_to_text(&i->buf, i->c, l->b);
 390                         err = flush_buf(i);
 391                         if (err)
 392                                 break;
 393                 }
 394                 prev_node = l->b;
 395
 396                 bch2_bfloat_to_text(&i->buf, l->b, _k);
 397                 err = flush_buf(i);
 398                 if (err)
 399                         break;
 400
 401                 bch2_btree_iter_advance(&iter);
 402                 i->from = iter.pos;
 403
 404                 err = flush_buf(i);
 405                 if (err)
 406                         break;
 407
 408                 if (!i->size)
 409                         break;
 410         }
 411         bch2_trans_iter_exit(&trans, &iter);
 412
 413         bch2_trans_exit(&trans);
 414
 415         return err < 0 ? err : i->ret;
 416 }
 417
 418 static const struct file_operations bfloat_failed_debug_ops = {
 419         .owner          = THIS_MODULE,
 420         .open           = bch2_dump_open,
 421         .release        = bch2_dump_release,
 422         .read           = bch2_read_bfloat_failed,
 423 };
 424
 425 static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
 426                                            struct btree *b)
 427 {
 428         out->tabstops[0] = 32;
 429
 430         pr_buf(out, "%px btree=%s l=%u ",
 431                b,
 432                bch2_btree_ids[b->c.btree_id],
 433                b->c.level);
 434         pr_newline(out);
 435
 436         pr_indent_push(out, 2);
 437
 438         bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
 439         pr_newline(out);
 440
 441         pr_buf(out, "flags: ");
 442         pr_tab(out);
 443         bch2_flags_to_text(out, bch2_btree_node_flags, b->flags);
 444         pr_newline(out);
 445
 446         pr_buf(out, "pcpu read locks: ");
 447         pr_tab(out);
 448         pr_buf(out, "%u", b->c.lock.readers != NULL);
 449         pr_newline(out);
 450
 451         pr_buf(out, "written:");
 452         pr_tab(out);
 453         pr_buf(out, "%u", b->written);
 454         pr_newline(out);
 455
 456         pr_buf(out, "writes blocked:");
 457         pr_tab(out);
 458         pr_buf(out, "%u", !list_empty_careful(&b->write_blocked));
 459         pr_newline(out);
 460
 461         pr_buf(out, "will make reachable:");
 462         pr_tab(out);
 463         pr_buf(out, "%lx", b->will_make_reachable);
 464         pr_newline(out);
 465
 466         pr_buf(out, "journal pin %px:", &b->writes[0].journal);
 467         pr_tab(out);
 468         pr_buf(out, "%llu", b->writes[0].journal.seq);
 469         pr_newline(out);
 470
 471         pr_buf(out, "journal pin %px:", &b->writes[1].journal);
 472         pr_tab(out);
 473         pr_buf(out, "%llu", b->writes[1].journal.seq);
 474         pr_newline(out);
 475
 476         pr_indent_pop(out, 2);
 477 }
 478
 479 static ssize_t bch2_cached_btree_nodes_read(struct file *file, char __user *buf,
 480                                             size_t size, loff_t *ppos)
 481 {
 482         struct dump_iter *i = file->private_data;
 483         struct bch_fs *c = i->c;
 484         bool done = false;
 485         int err;
 486
 487         i->ubuf = buf;
 488         i->size = size;
 489         i->ret  = 0;
 490
 491         do {
 492                 struct bucket_table *tbl;
 493                 struct rhash_head *pos;
 494                 struct btree *b;
 495
 496                 err = flush_buf(i);
 497                 if (err)
 498                         return err;
 499
 500                 if (!i->size)
 501                         break;
 502
 503                 rcu_read_lock();
 504                 i->buf.atomic++;
 505                 tbl = rht_dereference_rcu(c->btree_cache.table.tbl,
 506                                           &c->btree_cache.table);
 507                 if (i->iter < tbl->size) {
 508                         rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
 509                                 bch2_cached_btree_node_to_text(&i->buf, c, b);
 510                         i->iter++;;
 511                 } else {
 512                         done = true;
 513                 }
 514                 --i->buf.atomic;
 515                 rcu_read_unlock();
 516         } while (!done);
 517
 518         if (i->buf.allocation_failure)
 519                 return -ENOMEM;
 520
 521         return i->ret;
 522 }
 523
 524 static const struct file_operations cached_btree_nodes_ops = {
 525         .owner          = THIS_MODULE,
 526         .open           = bch2_dump_open,
 527         .release        = bch2_dump_release,
 528         .read           = bch2_cached_btree_nodes_read,
 529 };
 530
 531 static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
 532                                       size_t size, loff_t *ppos)
 533 {
 534         struct dump_iter *i = file->private_data;
 535         struct bch_fs *c = i->c;
 536         bool done = false;
 537         int err;
 538
 539         i->ubuf = buf;
 540         i->size = size;
 541         i->ret  = 0;
 542
 543         do {
 544                 err = flush_buf(i);
 545                 if (err)
 546                         return err;
 547
 548                 if (!i->size)
 549                         break;
 550
 551                 done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
 552                 i->iter++;
 553         } while (!done);
 554
 555         if (i->buf.allocation_failure)
 556                 return -ENOMEM;
 557
 558         return i->ret;
 559 }
 560
 561 static const struct file_operations journal_pins_ops = {
 562         .owner          = THIS_MODULE,
 563         .open           = bch2_dump_open,
 564         .release        = bch2_dump_release,
 565         .read           = bch2_journal_pins_read,
 566 };
 567
 568 void bch2_fs_debug_exit(struct bch_fs *c)
 569 {
 570         if (!IS_ERR_OR_NULL(c->fs_debug_dir))
 571                 debugfs_remove_recursive(c->fs_debug_dir);
 572 }
 573
 574 void bch2_fs_debug_init(struct bch_fs *c)
 575 {
 576         struct btree_debug *bd;
 577         char name[100];
 578
 579         if (IS_ERR_OR_NULL(bch_debug))
 580                 return;
 581
 582         snprintf(name, sizeof(name), "%pU", c->sb.user_uuid.b);
 583         c->fs_debug_dir = debugfs_create_dir(name, bch_debug);
 584         if (IS_ERR_OR_NULL(c->fs_debug_dir))
 585                 return;
 586
 587         debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
 588                             c->btree_debug, &cached_btree_nodes_ops);
 589
 590         debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
 591                             c->btree_debug, &journal_pins_ops);
 592
 593         c->btree_debug_dir = debugfs_create_dir("btrees", c->fs_debug_dir);
 594         if (IS_ERR_OR_NULL(c->btree_debug_dir))
 595                 return;
 596
 597         for (bd = c->btree_debug;
 598              bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
 599              bd++) {
 600                 bd->id = bd - c->btree_debug;
 601                 debugfs_create_file(bch2_btree_ids[bd->id],
 602                                     0400, c->btree_debug_dir, bd,
 603                                     &btree_debug_ops);
 604
 605                 snprintf(name, sizeof(name), "%s-formats",
 606                          bch2_btree_ids[bd->id]);
 607
 608                 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
 609                                     &btree_format_debug_ops);
 610
 611                 snprintf(name, sizeof(name), "%s-bfloat-failed",
 612                          bch2_btree_ids[bd->id]);
 613
 614                 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
 615                                     &bfloat_failed_debug_ops);
 616         }
 617 }
 618
 619 #endif
 620
 621 void bch2_debug_exit(void)
 622 {
 623         if (!IS_ERR_OR_NULL(bch_debug))
 624                 debugfs_remove_recursive(bch_debug);
 625 }
 626
 627 int __init bch2_debug_init(void)
 628 {
 629         int ret = 0;
 630
 631         bch_debug = debugfs_create_dir("bcachefs", NULL);
 632         return ret;
 633 }