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/pretty-printers.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;
  43
  44         if (!bch2_dev_get_ioref(ca, READ))
  45                 return false;
  46
  47         bio = bio_alloc_bioset(ca->disk_sb.bdev,
  48                                buf_pages(n_sorted, btree_bytes(c)),
  49                                REQ_OP_READ|REQ_META,
  50                                GFP_NOIO,
  51                                &c->btree_bio);
  52         bio->bi_iter.bi_sector  = pick.ptr.offset;
  53         bch2_bio_map(bio, n_sorted, btree_bytes(c));
  54
  55         submit_bio_wait(bio);
  56
  57         bio_put(bio);
  58         percpu_ref_put(&ca->io_ref);
  59
  60         memcpy(n_ondisk, n_sorted, btree_bytes(c));
  61
  62         v->written = 0;
  63         if (bch2_btree_node_read_done(c, ca, v, false))
  64                 return false;
  65
  66         n_sorted = c->verify_data->data;
  67         sorted = &n_sorted->keys;
  68
  69         if (inmemory->u64s != sorted->u64s ||
  70             memcmp(inmemory->start,
  71                    sorted->start,
  72                    vstruct_end(inmemory) - (void *) inmemory->start)) {
  73                 unsigned offset = 0, sectors;
  74                 struct bset *i;
  75                 unsigned j;
  76
  77                 console_lock();
  78
  79                 printk(KERN_ERR "*** in memory:\n");
  80                 bch2_dump_bset(c, b, inmemory, 0);
  81
  82                 printk(KERN_ERR "*** read back in:\n");
  83                 bch2_dump_bset(c, v, sorted, 0);
  84
  85                 while (offset < v->written) {
  86                         if (!offset) {
  87                                 i = &n_ondisk->keys;
  88                                 sectors = vstruct_blocks(n_ondisk, c->block_bits) <<
  89                                         c->block_bits;
  90                         } else {
  91                                 struct btree_node_entry *bne =
  92                                         (void *) n_ondisk + (offset << 9);
  93                                 i = &bne->keys;
  94
  95                                 sectors = vstruct_blocks(bne, c->block_bits) <<
  96                                         c->block_bits;
  97                         }
  98
  99                         printk(KERN_ERR "*** on disk block %u:\n", offset);
 100                         bch2_dump_bset(c, b, i, offset);
 101
 102                         offset += sectors;
 103                 }
 104
 105                 for (j = 0; j < le16_to_cpu(inmemory->u64s); j++)
 106                         if (inmemory->_data[j] != sorted->_data[j])
 107                                 break;
 108
 109                 console_unlock();
 110                 bch_err(c, "verify failed at key %u", j);
 111
 112                 failed = true;
 113         }
 114
 115         if (v->written != b->written) {
 116                 bch_err(c, "written wrong: expected %u, got %u",
 117                         b->written, v->written);
 118                 failed = true;
 119         }
 120
 121         return failed;
 122 }
 123
 124 void __bch2_btree_verify(struct bch_fs *c, struct btree *b)
 125 {
 126         struct bkey_ptrs_c ptrs;
 127         struct extent_ptr_decoded p;
 128         const union bch_extent_entry *entry;
 129         struct btree *v;
 130         struct bset *inmemory = &b->data->keys;
 131         struct bkey_packed *k;
 132         bool failed = false;
 133
 134         if (c->opts.nochanges)
 135                 return;
 136
 137         bch2_btree_node_io_lock(b);
 138         mutex_lock(&c->verify_lock);
 139
 140         if (!c->verify_ondisk) {
 141                 c->verify_ondisk = kvpmalloc(btree_bytes(c), GFP_KERNEL);
 142                 if (!c->verify_ondisk)
 143                         goto out;
 144         }
 145
 146         if (!c->verify_data) {
 147                 c->verify_data = __bch2_btree_node_mem_alloc(c);
 148                 if (!c->verify_data)
 149                         goto out;
 150
 151                 list_del_init(&c->verify_data->list);
 152         }
 153
 154         BUG_ON(b->nsets != 1);
 155
 156         for (k = inmemory->start; k != vstruct_last(inmemory); k = bkey_next(k))
 157                 if (k->type == KEY_TYPE_btree_ptr_v2) {
 158                         struct bch_btree_ptr_v2 *v = (void *) bkeyp_val(&b->format, k);
 159                         v->mem_ptr = 0;
 160                 }
 161
 162         v = c->verify_data;
 163         bkey_copy(&v->key, &b->key);
 164         v->c.level      = b->c.level;
 165         v->c.btree_id   = b->c.btree_id;
 166         bch2_btree_keys_init(v);
 167
 168         ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(&b->key));
 169         bkey_for_each_ptr_decode(&b->key.k, ptrs, p, entry)
 170                 failed |= bch2_btree_verify_replica(c, b, p);
 171
 172         if (failed) {
 173                 struct printbuf buf = PRINTBUF;
 174
 175                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
 176                 bch2_fs_fatal_error(c, "btree node verify failed for : %s\n", buf.buf);
 177                 printbuf_exit(&buf);
 178         }
 179 out:
 180         mutex_unlock(&c->verify_lock);
 181         bch2_btree_node_io_unlock(b);
 182 }
 183
 184 #ifdef CONFIG_DEBUG_FS
 185
 186 /* XXX: bch_fs refcounting */
 187
 188 struct dump_iter {
 189         struct bch_fs           *c;
 190         enum btree_id           id;
 191         struct bpos             from;
 192         u64                     iter;
 193
 194         struct printbuf         buf;
 195
 196         char __user             *ubuf;  /* destination user buffer */
 197         size_t                  size;   /* size of requested read */
 198         ssize_t                 ret;    /* bytes read so far */
 199 };
 200
 201 static int flush_buf(struct dump_iter *i)
 202 {
 203         if (i->buf.pos) {
 204                 size_t bytes = min_t(size_t, i->buf.pos, i->size);
 205                 int err = copy_to_user(i->ubuf, i->buf.buf, bytes);
 206
 207                 if (err)
 208                         return err;
 209
 210                 i->ret   += bytes;
 211                 i->ubuf  += bytes;
 212                 i->size  -= bytes;
 213                 i->buf.pos -= bytes;
 214                 memmove(i->buf.buf, i->buf.buf + bytes, i->buf.pos);
 215         }
 216
 217         return 0;
 218 }
 219
 220 static int bch2_dump_open(struct inode *inode, struct file *file)
 221 {
 222         struct btree_debug *bd = inode->i_private;
 223         struct dump_iter *i;
 224
 225         i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
 226         if (!i)
 227                 return -ENOMEM;
 228
 229         file->private_data = i;
 230         i->from = POS_MIN;
 231         i->iter = 0;
 232         i->c    = container_of(bd, struct bch_fs, btree_debug[bd->id]);
 233         i->id   = bd->id;
 234         i->buf  = PRINTBUF;
 235
 236         return 0;
 237 }
 238
 239 static int bch2_dump_release(struct inode *inode, struct file *file)
 240 {
 241         struct dump_iter *i = file->private_data;
 242
 243         printbuf_exit(&i->buf);
 244         kfree(i);
 245         return 0;
 246 }
 247
 248 static ssize_t bch2_read_btree(struct file *file, char __user *buf,
 249                                size_t size, loff_t *ppos)
 250 {
 251         struct dump_iter *i = file->private_data;
 252         struct btree_trans trans;
 253         struct btree_iter iter;
 254         struct bkey_s_c k;
 255         int err;
 256
 257         i->ubuf = buf;
 258         i->size = size;
 259         i->ret  = 0;
 260
 261         err = flush_buf(i);
 262         if (err)
 263                 return err;
 264
 265         if (!i->size)
 266                 return i->ret;
 267
 268         bch2_trans_init(&trans, i->c, 0, 0);
 269
 270         bch2_trans_iter_init(&trans, &iter, i->id, i->from,
 271                              BTREE_ITER_PREFETCH|
 272                              BTREE_ITER_ALL_SNAPSHOTS);
 273         k = bch2_btree_iter_peek(&iter);
 274
 275         while (k.k && !(err = bkey_err(k))) {
 276                 bch2_bkey_val_to_text(&i->buf, i->c, k);
 277                 prt_char(&i->buf, '\n');
 278
 279                 k = bch2_btree_iter_next(&iter);
 280                 i->from = iter.pos;
 281
 282                 err = flush_buf(i);
 283                 if (err)
 284                         break;
 285
 286                 if (!i->size)
 287                         break;
 288         }
 289         bch2_trans_iter_exit(&trans, &iter);
 290
 291         bch2_trans_exit(&trans);
 292
 293         return err < 0 ? err : i->ret;
 294 }
 295
 296 static const struct file_operations btree_debug_ops = {
 297         .owner          = THIS_MODULE,
 298         .open           = bch2_dump_open,
 299         .release        = bch2_dump_release,
 300         .read           = bch2_read_btree,
 301 };
 302
 303 static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
 304                                        size_t size, loff_t *ppos)
 305 {
 306         struct dump_iter *i = file->private_data;
 307         struct btree_trans trans;
 308         struct btree_iter iter;
 309         struct btree *b;
 310         int err;
 311
 312         i->ubuf = buf;
 313         i->size = size;
 314         i->ret  = 0;
 315
 316         err = flush_buf(i);
 317         if (err)
 318                 return err;
 319
 320         if (!i->size || !bpos_cmp(SPOS_MAX, i->from))
 321                 return i->ret;
 322
 323         bch2_trans_init(&trans, i->c, 0, 0);
 324
 325         for_each_btree_node(&trans, iter, i->id, i->from, 0, b, err) {
 326                 bch2_btree_node_to_text(&i->buf, i->c, b);
 327                 err = flush_buf(i);
 328                 if (err)
 329                         break;
 330
 331                 /*
 332                  * can't easily correctly restart a btree node traversal across
 333                  * all nodes, meh
 334                  */
 335                 i->from = bpos_cmp(SPOS_MAX, b->key.k.p)
 336                         ? bpos_successor(b->key.k.p)
 337                         : b->key.k.p;
 338
 339                 if (!i->size)
 340                         break;
 341         }
 342         bch2_trans_iter_exit(&trans, &iter);
 343
 344         bch2_trans_exit(&trans);
 345
 346         return err < 0 ? err : i->ret;
 347 }
 348
 349 static const struct file_operations btree_format_debug_ops = {
 350         .owner          = THIS_MODULE,
 351         .open           = bch2_dump_open,
 352         .release        = bch2_dump_release,
 353         .read           = bch2_read_btree_formats,
 354 };
 355
 356 static ssize_t bch2_read_bfloat_failed(struct file *file, char __user *buf,
 357                                        size_t size, loff_t *ppos)
 358 {
 359         struct dump_iter *i = file->private_data;
 360         struct btree_trans trans;
 361         struct btree_iter iter;
 362         struct bkey_s_c k;
 363         struct btree *prev_node = NULL;
 364         int err;
 365
 366         i->ubuf = buf;
 367         i->size = size;
 368         i->ret  = 0;
 369
 370         err = flush_buf(i);
 371         if (err)
 372                 return err;
 373
 374         if (!i->size)
 375                 return i->ret;
 376
 377         bch2_trans_init(&trans, i->c, 0, 0);
 378
 379         bch2_trans_iter_init(&trans, &iter, i->id, i->from,
 380                              BTREE_ITER_PREFETCH|
 381                              BTREE_ITER_ALL_SNAPSHOTS);
 382
 383         while ((k = bch2_btree_iter_peek(&iter)).k &&
 384                !(err = bkey_err(k))) {
 385                 struct btree_path_level *l = &iter.path->l[0];
 386                 struct bkey_packed *_k =
 387                         bch2_btree_node_iter_peek(&l->iter, l->b);
 388
 389                 if (l->b != prev_node) {
 390                         bch2_btree_node_to_text(&i->buf, i->c, l->b);
 391                         err = flush_buf(i);
 392                         if (err)
 393                                 break;
 394                 }
 395                 prev_node = l->b;
 396
 397                 bch2_bfloat_to_text(&i->buf, l->b, _k);
 398                 err = flush_buf(i);
 399                 if (err)
 400                         break;
 401
 402                 bch2_btree_iter_advance(&iter);
 403                 i->from = iter.pos;
 404
 405                 err = flush_buf(i);
 406                 if (err)
 407                         break;
 408
 409                 if (!i->size)
 410                         break;
 411         }
 412         bch2_trans_iter_exit(&trans, &iter);
 413
 414         bch2_trans_exit(&trans);
 415
 416         return err < 0 ? err : i->ret;
 417 }
 418
 419 static const struct file_operations bfloat_failed_debug_ops = {
 420         .owner          = THIS_MODULE,
 421         .open           = bch2_dump_open,
 422         .release        = bch2_dump_release,
 423         .read           = bch2_read_bfloat_failed,
 424 };
 425
 426 static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
 427                                            struct btree *b)
 428 {
 429         out->tabstops[0] = 32;
 430
 431         prt_printf(out, "%px btree=%s l=%u ",
 432                b,
 433                bch2_btree_ids[b->c.btree_id],
 434                b->c.level);
 435         prt_newline(out);
 436
 437         printbuf_indent_add(out, 2);
 438
 439         bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
 440         prt_newline(out);
 441
 442         prt_printf(out, "flags: ");
 443         prt_tab(out);
 444         prt_bitflags(out, bch2_btree_node_flags, b->flags);
 445         prt_newline(out);
 446
 447         prt_printf(out, "pcpu read locks: ");
 448         prt_tab(out);
 449         prt_printf(out, "%u", b->c.lock.readers != NULL);
 450         prt_newline(out);
 451
 452         prt_printf(out, "written:");
 453         prt_tab(out);
 454         prt_printf(out, "%u", b->written);
 455         prt_newline(out);
 456
 457         prt_printf(out, "writes blocked:");
 458         prt_tab(out);
 459         prt_printf(out, "%u", !list_empty_careful(&b->write_blocked));
 460         prt_newline(out);
 461
 462         prt_printf(out, "will make reachable:");
 463         prt_tab(out);
 464         prt_printf(out, "%lx", b->will_make_reachable);
 465         prt_newline(out);
 466
 467         prt_printf(out, "journal pin %px:", &b->writes[0].journal);
 468         prt_tab(out);
 469         prt_printf(out, "%llu", b->writes[0].journal.seq);
 470         prt_newline(out);
 471
 472         prt_printf(out, "journal pin %px:", &b->writes[1].journal);
 473         prt_tab(out);
 474         prt_printf(out, "%llu", b->writes[1].journal.seq);
 475         prt_newline(out);
 476
 477         printbuf_indent_sub(out, 2);
 478 }
 479
 480 static ssize_t bch2_cached_btree_nodes_read(struct file *file, char __user *buf,
 481                                             size_t size, loff_t *ppos)
 482 {
 483         struct dump_iter *i = file->private_data;
 484         struct bch_fs *c = i->c;
 485         bool done = false;
 486         int err;
 487
 488         i->ubuf = buf;
 489         i->size = size;
 490         i->ret  = 0;
 491
 492         do {
 493                 struct bucket_table *tbl;
 494                 struct rhash_head *pos;
 495                 struct btree *b;
 496
 497                 err = flush_buf(i);
 498                 if (err)
 499                         return err;
 500
 501                 if (!i->size)
 502                         break;
 503
 504                 rcu_read_lock();
 505                 i->buf.atomic++;
 506                 tbl = rht_dereference_rcu(c->btree_cache.table.tbl,
 507                                           &c->btree_cache.table);
 508                 if (i->iter < tbl->size) {
 509                         rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
 510                                 bch2_cached_btree_node_to_text(&i->buf, c, b);
 511                         i->iter++;;
 512                 } else {
 513                         done = true;
 514                 }
 515                 --i->buf.atomic;
 516                 rcu_read_unlock();
 517         } while (!done);
 518
 519         if (i->buf.allocation_failure)
 520                 return -ENOMEM;
 521
 522         return i->ret;
 523 }
 524
 525 static const struct file_operations cached_btree_nodes_ops = {
 526         .owner          = THIS_MODULE,
 527         .open           = bch2_dump_open,
 528         .release        = bch2_dump_release,
 529         .read           = bch2_cached_btree_nodes_read,
 530 };
 531
 532 static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
 533                                       size_t size, loff_t *ppos)
 534 {
 535         struct dump_iter *i = file->private_data;
 536         struct bch_fs *c = i->c;
 537         bool done = false;
 538         int err;
 539
 540         i->ubuf = buf;
 541         i->size = size;
 542         i->ret  = 0;
 543
 544         do {
 545                 err = flush_buf(i);
 546                 if (err)
 547                         return err;
 548
 549                 if (!i->size)
 550                         break;
 551
 552                 done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
 553                 i->iter++;
 554         } while (!done);
 555
 556         if (i->buf.allocation_failure)
 557                 return -ENOMEM;
 558
 559         return i->ret;
 560 }
 561
 562 static const struct file_operations journal_pins_ops = {
 563         .owner          = THIS_MODULE,
 564         .open           = bch2_dump_open,
 565         .release        = bch2_dump_release,
 566         .read           = bch2_journal_pins_read,
 567 };
 568
 569 void bch2_fs_debug_exit(struct bch_fs *c)
 570 {
 571         if (!IS_ERR_OR_NULL(c->fs_debug_dir))
 572                 debugfs_remove_recursive(c->fs_debug_dir);
 573 }
 574
 575 void bch2_fs_debug_init(struct bch_fs *c)
 576 {
 577         struct btree_debug *bd;
 578         char name[100];
 579
 580         if (IS_ERR_OR_NULL(bch_debug))
 581                 return;
 582
 583         snprintf(name, sizeof(name), "%pU", c->sb.user_uuid.b);
 584         c->fs_debug_dir = debugfs_create_dir(name, bch_debug);
 585         if (IS_ERR_OR_NULL(c->fs_debug_dir))
 586                 return;
 587
 588         debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
 589                             c->btree_debug, &cached_btree_nodes_ops);
 590
 591         debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
 592                             c->btree_debug, &journal_pins_ops);
 593
 594         c->btree_debug_dir = debugfs_create_dir("btrees", c->fs_debug_dir);
 595         if (IS_ERR_OR_NULL(c->btree_debug_dir))
 596                 return;
 597
 598         for (bd = c->btree_debug;
 599              bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
 600              bd++) {
 601                 bd->id = bd - c->btree_debug;
 602                 debugfs_create_file(bch2_btree_ids[bd->id],
 603                                     0400, c->btree_debug_dir, bd,
 604                                     &btree_debug_ops);
 605
 606                 snprintf(name, sizeof(name), "%s-formats",
 607                          bch2_btree_ids[bd->id]);
 608
 609                 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
 610                                     &btree_format_debug_ops);
 611
 612                 snprintf(name, sizeof(name), "%s-bfloat-failed",
 613                          bch2_btree_ids[bd->id]);
 614
 615                 debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
 616                                     &bfloat_failed_debug_ops);
 617         }
 618 }
 619
 620 #endif
 621
 622 void bch2_debug_exit(void)
 623 {
 624         if (!IS_ERR_OR_NULL(bch_debug))
 625                 debugfs_remove_recursive(bch_debug);
 626 }
 627
 628 int __init bch2_debug_init(void)
 629 {
 630         int ret = 0;
 631
 632         bch_debug = debugfs_create_dir("bcachefs", NULL);
 633         return ret;
 634 }