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 "journal_reclaim.h"
  24 #include "super.h"
  25
  26 #include <linux/console.h>
  27 #include <linux/debugfs.h>
  28 #include <linux/module.h>
  29 #include <linux/random.h>
  30 #include <linux/seq_file.h>
  31
  32 static struct dentry *bch_debug;
  33
  34 static bool bch2_btree_verify_replica(struct bch_fs *c, struct btree *b,
  35                                       struct extent_ptr_decoded pick)
  36 {
  37         struct btree *v = c->verify_data;
  38         struct btree_node *n_ondisk = c->verify_ondisk;
  39         struct btree_node *n_sorted = c->verify_data->data;
  40         struct bset *sorted, *inmemory = &b->data->keys;
  41         struct bio *bio;
  42         bool failed = false, saw_error = false;
  43
  44         struct bch_dev *ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ);
  45         if (!ca)
  46                 return false;
  47
  48         bio = bio_alloc_bioset(ca->disk_sb.bdev,
  49                                buf_pages(n_sorted, btree_buf_bytes(b)),
  50                                REQ_OP_READ|REQ_META,
  51                                GFP_NOFS,
  52                                &c->btree_bio);
  53         bio->bi_iter.bi_sector  = pick.ptr.offset;
  54         bch2_bio_map(bio, n_sorted, btree_buf_bytes(b));
  55
  56         submit_bio_wait(bio);
  57
  58         bio_put(bio);
  59         percpu_ref_put(&ca->io_ref);
  60
  61         memcpy(n_ondisk, n_sorted, btree_buf_bytes(b));
  62
  63         v->written = 0;
  64         if (bch2_btree_node_read_done(c, ca, v, false, &saw_error) || saw_error)
  65                 return false;
  66
  67         n_sorted = c->verify_data->data;
  68         sorted = &n_sorted->keys;
  69
  70         if (inmemory->u64s != sorted->u64s ||
  71             memcmp(inmemory->start,
  72                    sorted->start,
  73                    vstruct_end(inmemory) - (void *) inmemory->start)) {
  74                 unsigned offset = 0, sectors;
  75                 struct bset *i;
  76                 unsigned j;
  77
  78                 console_lock();
  79
  80                 printk(KERN_ERR "*** in memory:\n");
  81                 bch2_dump_bset(c, b, inmemory, 0);
  82
  83                 printk(KERN_ERR "*** read back in:\n");
  84                 bch2_dump_bset(c, v, sorted, 0);
  85
  86                 while (offset < v->written) {
  87                         if (!offset) {
  88                                 i = &n_ondisk->keys;
  89                                 sectors = vstruct_blocks(n_ondisk, c->block_bits) <<
  90                                         c->block_bits;
  91                         } else {
  92                                 struct btree_node_entry *bne =
  93                                         (void *) n_ondisk + (offset << 9);
  94                                 i = &bne->keys;
  95
  96                                 sectors = vstruct_blocks(bne, c->block_bits) <<
  97                                         c->block_bits;
  98                         }
  99
 100                         printk(KERN_ERR "*** on disk block %u:\n", offset);
 101                         bch2_dump_bset(c, b, i, offset);
 102
 103                         offset += sectors;
 104                 }
 105
 106                 for (j = 0; j < le16_to_cpu(inmemory->u64s); j++)
 107                         if (inmemory->_data[j] != sorted->_data[j])
 108                                 break;
 109
 110                 console_unlock();
 111                 bch_err(c, "verify failed at key %u", j);
 112
 113                 failed = true;
 114         }
 115
 116         if (v->written != b->written) {
 117                 bch_err(c, "written wrong: expected %u, got %u",
 118                         b->written, v->written);
 119                 failed = true;
 120         }
 121
 122         return failed;
 123 }
 124
 125 void __bch2_btree_verify(struct bch_fs *c, struct btree *b)
 126 {
 127         struct bkey_ptrs_c ptrs;
 128         struct extent_ptr_decoded p;
 129         const union bch_extent_entry *entry;
 130         struct btree *v;
 131         struct bset *inmemory = &b->data->keys;
 132         struct bkey_packed *k;
 133         bool failed = false;
 134
 135         if (c->opts.nochanges)
 136                 return;
 137
 138         bch2_btree_node_io_lock(b);
 139         mutex_lock(&c->verify_lock);
 140
 141         if (!c->verify_ondisk) {
 142                 c->verify_ondisk = kvmalloc(btree_buf_bytes(b), GFP_KERNEL);
 143                 if (!c->verify_ondisk)
 144                         goto out;
 145         }
 146
 147         if (!c->verify_data) {
 148                 c->verify_data = __bch2_btree_node_mem_alloc(c);
 149                 if (!c->verify_data)
 150                         goto out;
 151
 152                 list_del_init(&c->verify_data->list);
 153         }
 154
 155         BUG_ON(b->nsets != 1);
 156
 157         for (k = inmemory->start; k != vstruct_last(inmemory); k = bkey_p_next(k))
 158                 if (k->type == KEY_TYPE_btree_ptr_v2)
 159                         ((struct bch_btree_ptr_v2 *) bkeyp_val(&b->format, k))->mem_ptr = 0;
 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 void bch2_btree_node_ondisk_to_text(struct printbuf *out, struct bch_fs *c,
 184                                     const struct btree *b)
 185 {
 186         struct btree_node *n_ondisk = NULL;
 187         struct extent_ptr_decoded pick;
 188         struct bch_dev *ca;
 189         struct bio *bio = NULL;
 190         unsigned offset = 0;
 191         int ret;
 192
 193         if (bch2_bkey_pick_read_device(c, bkey_i_to_s_c(&b->key), NULL, &pick) <= 0) {
 194                 prt_printf(out, "error getting device to read from: invalid device\n");
 195                 return;
 196         }
 197
 198         ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ);
 199         if (!ca) {
 200                 prt_printf(out, "error getting device to read from: not online\n");
 201                 return;
 202         }
 203
 204         n_ondisk = kvmalloc(btree_buf_bytes(b), GFP_KERNEL);
 205         if (!n_ondisk) {
 206                 prt_printf(out, "memory allocation failure\n");
 207                 goto out;
 208         }
 209
 210         bio = bio_alloc_bioset(ca->disk_sb.bdev,
 211                                buf_pages(n_ondisk, btree_buf_bytes(b)),
 212                                REQ_OP_READ|REQ_META,
 213                                GFP_NOFS,
 214                                &c->btree_bio);
 215         bio->bi_iter.bi_sector  = pick.ptr.offset;
 216         bch2_bio_map(bio, n_ondisk, btree_buf_bytes(b));
 217
 218         ret = submit_bio_wait(bio);
 219         if (ret) {
 220                 prt_printf(out, "IO error reading btree node: %s\n", bch2_err_str(ret));
 221                 goto out;
 222         }
 223
 224         while (offset < btree_sectors(c)) {
 225                 struct bset *i;
 226                 struct nonce nonce;
 227                 struct bch_csum csum;
 228                 struct bkey_packed *k;
 229                 unsigned sectors;
 230
 231                 if (!offset) {
 232                         i = &n_ondisk->keys;
 233
 234                         if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
 235                                 prt_printf(out, "unknown checksum type at offset %u: %llu\n",
 236                                            offset, BSET_CSUM_TYPE(i));
 237                                 goto out;
 238                         }
 239
 240                         nonce = btree_nonce(i, offset << 9);
 241                         csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, n_ondisk);
 242
 243                         if (bch2_crc_cmp(csum, n_ondisk->csum)) {
 244                                 prt_printf(out, "invalid checksum\n");
 245                                 goto out;
 246                         }
 247
 248                         bset_encrypt(c, i, offset << 9);
 249
 250                         sectors = vstruct_sectors(n_ondisk, c->block_bits);
 251                 } else {
 252                         struct btree_node_entry *bne = (void *) n_ondisk + (offset << 9);
 253
 254                         i = &bne->keys;
 255
 256                         if (i->seq != n_ondisk->keys.seq)
 257                                 break;
 258
 259                         if (!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i))) {
 260                                 prt_printf(out, "unknown checksum type at offset %u: %llu\n",
 261                                            offset, BSET_CSUM_TYPE(i));
 262                                 goto out;
 263                         }
 264
 265                         nonce = btree_nonce(i, offset << 9);
 266                         csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne);
 267
 268                         if (bch2_crc_cmp(csum, bne->csum)) {
 269                                 prt_printf(out, "invalid checksum");
 270                                 goto out;
 271                         }
 272
 273                         bset_encrypt(c, i, offset << 9);
 274
 275                         sectors = vstruct_sectors(bne, c->block_bits);
 276                 }
 277
 278                 prt_printf(out, "  offset %u version %u, journal seq %llu\n",
 279                            offset,
 280                            le16_to_cpu(i->version),
 281                            le64_to_cpu(i->journal_seq));
 282                 offset += sectors;
 283
 284                 printbuf_indent_add(out, 4);
 285
 286                 for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k)) {
 287                         struct bkey u;
 288
 289                         bch2_bkey_val_to_text(out, c, bkey_disassemble(b, k, &u));
 290                         prt_newline(out);
 291                 }
 292
 293                 printbuf_indent_sub(out, 4);
 294         }
 295 out:
 296         if (bio)
 297                 bio_put(bio);
 298         kvfree(n_ondisk);
 299         percpu_ref_put(&ca->io_ref);
 300 }
 301
 302 #ifdef CONFIG_DEBUG_FS
 303
 304 /* XXX: bch_fs refcounting */
 305
 306 struct dump_iter {
 307         struct bch_fs           *c;
 308         enum btree_id           id;
 309         struct bpos             from;
 310         struct bpos             prev_node;
 311         u64                     iter;
 312
 313         struct printbuf         buf;
 314
 315         char __user             *ubuf;  /* destination user buffer */
 316         size_t                  size;   /* size of requested read */
 317         ssize_t                 ret;    /* bytes read so far */
 318 };
 319
 320 static ssize_t flush_buf(struct dump_iter *i)
 321 {
 322         if (i->buf.pos) {
 323                 size_t bytes = min_t(size_t, i->buf.pos, i->size);
 324                 int copied = bytes - copy_to_user(i->ubuf, i->buf.buf, bytes);
 325
 326                 i->ret   += copied;
 327                 i->ubuf  += copied;
 328                 i->size  -= copied;
 329                 i->buf.pos -= copied;
 330                 memmove(i->buf.buf, i->buf.buf + copied, i->buf.pos);
 331
 332                 if (copied != bytes)
 333                         return -EFAULT;
 334         }
 335
 336         return i->size ? 0 : i->ret;
 337 }
 338
 339 static int bch2_dump_open(struct inode *inode, struct file *file)
 340 {
 341         struct btree_debug *bd = inode->i_private;
 342         struct dump_iter *i;
 343
 344         i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
 345         if (!i)
 346                 return -ENOMEM;
 347
 348         file->private_data = i;
 349         i->from = POS_MIN;
 350         i->iter = 0;
 351         i->c    = container_of(bd, struct bch_fs, btree_debug[bd->id]);
 352         i->id   = bd->id;
 353         i->buf  = PRINTBUF;
 354
 355         return 0;
 356 }
 357
 358 static int bch2_dump_release(struct inode *inode, struct file *file)
 359 {
 360         struct dump_iter *i = file->private_data;
 361
 362         printbuf_exit(&i->buf);
 363         kfree(i);
 364         return 0;
 365 }
 366
 367 static ssize_t bch2_read_btree(struct file *file, char __user *buf,
 368                                size_t size, loff_t *ppos)
 369 {
 370         struct dump_iter *i = file->private_data;
 371
 372         i->ubuf = buf;
 373         i->size = size;
 374         i->ret  = 0;
 375
 376         return flush_buf(i) ?:
 377                 bch2_trans_run(i->c,
 378                         for_each_btree_key(trans, iter, i->id, i->from,
 379                                            BTREE_ITER_prefetch|
 380                                            BTREE_ITER_all_snapshots, k, ({
 381                                 bch2_bkey_val_to_text(&i->buf, i->c, k);
 382                                 prt_newline(&i->buf);
 383                                 bch2_trans_unlock(trans);
 384                                 i->from = bpos_successor(iter.pos);
 385                                 flush_buf(i);
 386                         }))) ?:
 387                 i->ret;
 388 }
 389
 390 static const struct file_operations btree_debug_ops = {
 391         .owner          = THIS_MODULE,
 392         .open           = bch2_dump_open,
 393         .release        = bch2_dump_release,
 394         .read           = bch2_read_btree,
 395 };
 396
 397 static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
 398                                        size_t size, loff_t *ppos)
 399 {
 400         struct dump_iter *i = file->private_data;
 401
 402         i->ubuf = buf;
 403         i->size = size;
 404         i->ret  = 0;
 405
 406         ssize_t ret = flush_buf(i);
 407         if (ret)
 408                 return ret;
 409
 410         if (bpos_eq(SPOS_MAX, i->from))
 411                 return i->ret;
 412
 413         return bch2_trans_run(i->c,
 414                 for_each_btree_node(trans, iter, i->id, i->from, 0, b, ({
 415                         bch2_btree_node_to_text(&i->buf, i->c, b);
 416                         i->from = !bpos_eq(SPOS_MAX, b->key.k.p)
 417                                 ? bpos_successor(b->key.k.p)
 418                                 : b->key.k.p;
 419
 420                         drop_locks_do(trans, flush_buf(i));
 421                 }))) ?: i->ret;
 422 }
 423
 424 static const struct file_operations btree_format_debug_ops = {
 425         .owner          = THIS_MODULE,
 426         .open           = bch2_dump_open,
 427         .release        = bch2_dump_release,
 428         .read           = bch2_read_btree_formats,
 429 };
 430
 431 static ssize_t bch2_read_bfloat_failed(struct file *file, char __user *buf,
 432                                        size_t size, loff_t *ppos)
 433 {
 434         struct dump_iter *i = file->private_data;
 435
 436         i->ubuf = buf;
 437         i->size = size;
 438         i->ret  = 0;
 439
 440         return flush_buf(i) ?:
 441                 bch2_trans_run(i->c,
 442                         for_each_btree_key(trans, iter, i->id, i->from,
 443                                            BTREE_ITER_prefetch|
 444                                            BTREE_ITER_all_snapshots, k, ({
 445                                 struct btree_path_level *l =
 446                                         &btree_iter_path(trans, &iter)->l[0];
 447                                 struct bkey_packed *_k =
 448                                         bch2_btree_node_iter_peek(&l->iter, l->b);
 449
 450                                 if (bpos_gt(l->b->key.k.p, i->prev_node)) {
 451                                         bch2_btree_node_to_text(&i->buf, i->c, l->b);
 452                                         i->prev_node = l->b->key.k.p;
 453                                 }
 454
 455                                 bch2_bfloat_to_text(&i->buf, l->b, _k);
 456                                 bch2_trans_unlock(trans);
 457                                 i->from = bpos_successor(iter.pos);
 458                                 flush_buf(i);
 459                         }))) ?:
 460                 i->ret;
 461 }
 462
 463 static const struct file_operations bfloat_failed_debug_ops = {
 464         .owner          = THIS_MODULE,
 465         .open           = bch2_dump_open,
 466         .release        = bch2_dump_release,
 467         .read           = bch2_read_bfloat_failed,
 468 };
 469
 470 static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
 471                                            struct btree *b)
 472 {
 473         if (!out->nr_tabstops)
 474                 printbuf_tabstop_push(out, 32);
 475
 476         prt_printf(out, "%px ", b);
 477         bch2_btree_id_level_to_text(out, b->c.btree_id, b->c.level);
 478         prt_printf(out, "\n");
 479
 480         printbuf_indent_add(out, 2);
 481
 482         bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
 483         prt_newline(out);
 484
 485         prt_printf(out, "flags:\t");
 486         prt_bitflags(out, bch2_btree_node_flags, b->flags);
 487         prt_newline(out);
 488
 489         prt_printf(out, "pcpu read locks:\t%u\n",       b->c.lock.readers != NULL);
 490         prt_printf(out, "written:\t%u\n",               b->written);
 491         prt_printf(out, "writes blocked:\t%u\n",        !list_empty_careful(&b->write_blocked));
 492         prt_printf(out, "will make reachable:\t%lx\n",  b->will_make_reachable);
 493
 494         prt_printf(out, "journal pin %px:\t%llu\n",
 495                    &b->writes[0].journal, b->writes[0].journal.seq);
 496         prt_printf(out, "journal pin %px:\t%llu\n",
 497                    &b->writes[1].journal, b->writes[1].journal.seq);
 498
 499         printbuf_indent_sub(out, 2);
 500 }
 501
 502 static ssize_t bch2_cached_btree_nodes_read(struct file *file, char __user *buf,
 503                                             size_t size, loff_t *ppos)
 504 {
 505         struct dump_iter *i = file->private_data;
 506         struct bch_fs *c = i->c;
 507         bool done = false;
 508         ssize_t ret = 0;
 509
 510         i->ubuf = buf;
 511         i->size = size;
 512         i->ret  = 0;
 513
 514         do {
 515                 struct bucket_table *tbl;
 516                 struct rhash_head *pos;
 517                 struct btree *b;
 518
 519                 ret = flush_buf(i);
 520                 if (ret)
 521                         return ret;
 522
 523                 rcu_read_lock();
 524                 i->buf.atomic++;
 525                 tbl = rht_dereference_rcu(c->btree_cache.table.tbl,
 526                                           &c->btree_cache.table);
 527                 if (i->iter < tbl->size) {
 528                         rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
 529                                 bch2_cached_btree_node_to_text(&i->buf, c, b);
 530                         i->iter++;
 531                 } else {
 532                         done = true;
 533                 }
 534                 --i->buf.atomic;
 535                 rcu_read_unlock();
 536         } while (!done);
 537
 538         if (i->buf.allocation_failure)
 539                 ret = -ENOMEM;
 540
 541         if (!ret)
 542                 ret = flush_buf(i);
 543
 544         return ret ?: i->ret;
 545 }
 546
 547 static const struct file_operations cached_btree_nodes_ops = {
 548         .owner          = THIS_MODULE,
 549         .open           = bch2_dump_open,
 550         .release        = bch2_dump_release,
 551         .read           = bch2_cached_btree_nodes_read,
 552 };
 553
 554 typedef int (*list_cmp_fn)(const struct list_head *l, const struct list_head *r);
 555
 556 static void list_sort(struct list_head *head, list_cmp_fn cmp)
 557 {
 558         struct list_head *pos;
 559
 560         list_for_each(pos, head)
 561                 while (!list_is_last(pos, head) &&
 562                        cmp(pos, pos->next) > 0) {
 563                         struct list_head *pos2, *next = pos->next;
 564
 565                         list_del(next);
 566                         list_for_each(pos2, head)
 567                                 if (cmp(next, pos2) < 0)
 568                                         goto pos_found;
 569                         BUG();
 570 pos_found:
 571                         list_add_tail(next, pos2);
 572                 }
 573 }
 574
 575 static int list_ptr_order_cmp(const struct list_head *l, const struct list_head *r)
 576 {
 577         return cmp_int(l, r);
 578 }
 579
 580 static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
 581                                             size_t size, loff_t *ppos)
 582 {
 583         struct dump_iter *i = file->private_data;
 584         struct bch_fs *c = i->c;
 585         struct btree_trans *trans;
 586         ssize_t ret = 0;
 587
 588         i->ubuf = buf;
 589         i->size = size;
 590         i->ret  = 0;
 591 restart:
 592         seqmutex_lock(&c->btree_trans_lock);
 593         list_sort(&c->btree_trans_list, list_ptr_order_cmp);
 594
 595         list_for_each_entry(trans, &c->btree_trans_list, list) {
 596                 if ((ulong) trans <= i->iter)
 597                         continue;
 598
 599                 i->iter = (ulong) trans;
 600
 601                 if (!closure_get_not_zero(&trans->ref))
 602                         continue;
 603
 604                 u32 seq = seqmutex_unlock(&c->btree_trans_lock);
 605
 606                 bch2_btree_trans_to_text(&i->buf, trans);
 607
 608                 prt_printf(&i->buf, "backtrace:\n");
 609                 printbuf_indent_add(&i->buf, 2);
 610                 bch2_prt_task_backtrace(&i->buf, trans->locking_wait.task, 0, GFP_KERNEL);
 611                 printbuf_indent_sub(&i->buf, 2);
 612                 prt_newline(&i->buf);
 613
 614                 closure_put(&trans->ref);
 615
 616                 ret = flush_buf(i);
 617                 if (ret)
 618                         goto unlocked;
 619
 620                 if (!seqmutex_relock(&c->btree_trans_lock, seq))
 621                         goto restart;
 622         }
 623         seqmutex_unlock(&c->btree_trans_lock);
 624 unlocked:
 625         if (i->buf.allocation_failure)
 626                 ret = -ENOMEM;
 627
 628         if (!ret)
 629                 ret = flush_buf(i);
 630
 631         return ret ?: i->ret;
 632 }
 633
 634 static const struct file_operations btree_transactions_ops = {
 635         .owner          = THIS_MODULE,
 636         .open           = bch2_dump_open,
 637         .release        = bch2_dump_release,
 638         .read           = bch2_btree_transactions_read,
 639 };
 640
 641 static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
 642                                       size_t size, loff_t *ppos)
 643 {
 644         struct dump_iter *i = file->private_data;
 645         struct bch_fs *c = i->c;
 646         bool done = false;
 647         int err;
 648
 649         i->ubuf = buf;
 650         i->size = size;
 651         i->ret  = 0;
 652
 653         while (1) {
 654                 err = flush_buf(i);
 655                 if (err)
 656                         return err;
 657
 658                 if (!i->size)
 659                         break;
 660
 661                 if (done)
 662                         break;
 663
 664                 done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
 665                 i->iter++;
 666         }
 667
 668         if (i->buf.allocation_failure)
 669                 return -ENOMEM;
 670
 671         return i->ret;
 672 }
 673
 674 static const struct file_operations journal_pins_ops = {
 675         .owner          = THIS_MODULE,
 676         .open           = bch2_dump_open,
 677         .release        = bch2_dump_release,
 678         .read           = bch2_journal_pins_read,
 679 };
 680
 681 static ssize_t bch2_btree_updates_read(struct file *file, char __user *buf,
 682                                        size_t size, loff_t *ppos)
 683 {
 684         struct dump_iter *i = file->private_data;
 685         struct bch_fs *c = i->c;
 686         int err;
 687
 688         i->ubuf = buf;
 689         i->size = size;
 690         i->ret  = 0;
 691
 692         if (!i->iter) {
 693                 bch2_btree_updates_to_text(&i->buf, c);
 694                 i->iter++;
 695         }
 696
 697         err = flush_buf(i);
 698         if (err)
 699                 return err;
 700
 701         if (i->buf.allocation_failure)
 702                 return -ENOMEM;
 703
 704         return i->ret;
 705 }
 706
 707 static const struct file_operations btree_updates_ops = {
 708         .owner          = THIS_MODULE,
 709         .open           = bch2_dump_open,
 710         .release        = bch2_dump_release,
 711         .read           = bch2_btree_updates_read,
 712 };
 713
 714 static int btree_transaction_stats_open(struct inode *inode, struct file *file)
 715 {
 716         struct bch_fs *c = inode->i_private;
 717         struct dump_iter *i;
 718
 719         i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
 720         if (!i)
 721                 return -ENOMEM;
 722
 723         i->iter = 1;
 724         i->c    = c;
 725         i->buf  = PRINTBUF;
 726         file->private_data = i;
 727
 728         return 0;
 729 }
 730
 731 static int btree_transaction_stats_release(struct inode *inode, struct file *file)
 732 {
 733         struct dump_iter *i = file->private_data;
 734
 735         printbuf_exit(&i->buf);
 736         kfree(i);
 737
 738         return 0;
 739 }
 740
 741 static ssize_t btree_transaction_stats_read(struct file *file, char __user *buf,
 742                                             size_t size, loff_t *ppos)
 743 {
 744         struct dump_iter        *i = file->private_data;
 745         struct bch_fs *c = i->c;
 746         int err;
 747
 748         i->ubuf = buf;
 749         i->size = size;
 750         i->ret  = 0;
 751
 752         while (1) {
 753                 struct btree_transaction_stats *s = &c->btree_transaction_stats[i->iter];
 754
 755                 err = flush_buf(i);
 756                 if (err)
 757                         return err;
 758
 759                 if (!i->size)
 760                         break;
 761
 762                 if (i->iter == ARRAY_SIZE(bch2_btree_transaction_fns) ||
 763                     !bch2_btree_transaction_fns[i->iter])
 764                         break;
 765
 766                 prt_printf(&i->buf, "%s:\n", bch2_btree_transaction_fns[i->iter]);
 767                 printbuf_indent_add(&i->buf, 2);
 768
 769                 mutex_lock(&s->lock);
 770
 771                 prt_printf(&i->buf, "Max mem used: %u\n", s->max_mem);
 772                 prt_printf(&i->buf, "Transaction duration:\n");
 773
 774                 printbuf_indent_add(&i->buf, 2);
 775                 bch2_time_stats_to_text(&i->buf, &s->duration);
 776                 printbuf_indent_sub(&i->buf, 2);
 777
 778                 if (IS_ENABLED(CONFIG_BCACHEFS_LOCK_TIME_STATS)) {
 779                         prt_printf(&i->buf, "Lock hold times:\n");
 780
 781                         printbuf_indent_add(&i->buf, 2);
 782                         bch2_time_stats_to_text(&i->buf, &s->lock_hold_times);
 783                         printbuf_indent_sub(&i->buf, 2);
 784                 }
 785
 786                 if (s->max_paths_text) {
 787                         prt_printf(&i->buf, "Maximum allocated btree paths (%u):\n", s->nr_max_paths);
 788
 789                         printbuf_indent_add(&i->buf, 2);
 790                         prt_str_indented(&i->buf, s->max_paths_text);
 791                         printbuf_indent_sub(&i->buf, 2);
 792                 }
 793
 794                 mutex_unlock(&s->lock);
 795
 796                 printbuf_indent_sub(&i->buf, 2);
 797                 prt_newline(&i->buf);
 798                 i->iter++;
 799         }
 800
 801         if (i->buf.allocation_failure)
 802                 return -ENOMEM;
 803
 804         return i->ret;
 805 }
 806
 807 static const struct file_operations btree_transaction_stats_op = {
 808         .owner          = THIS_MODULE,
 809         .open           = btree_transaction_stats_open,
 810         .release        = btree_transaction_stats_release,
 811         .read           = btree_transaction_stats_read,
 812 };
 813
 814 /* walk btree transactions until we find a deadlock and print it */
 815 static void btree_deadlock_to_text(struct printbuf *out, struct bch_fs *c)
 816 {
 817         struct btree_trans *trans;
 818         ulong iter = 0;
 819 restart:
 820         seqmutex_lock(&c->btree_trans_lock);
 821         list_sort(&c->btree_trans_list, list_ptr_order_cmp);
 822
 823         list_for_each_entry(trans, &c->btree_trans_list, list) {
 824                 if ((ulong) trans <= iter)
 825                         continue;
 826
 827                 iter = (ulong) trans;
 828
 829                 if (!closure_get_not_zero(&trans->ref))
 830                         continue;
 831
 832                 u32 seq = seqmutex_unlock(&c->btree_trans_lock);
 833
 834                 bool found = bch2_check_for_deadlock(trans, out) != 0;
 835
 836                 closure_put(&trans->ref);
 837
 838                 if (found)
 839                         return;
 840
 841                 if (!seqmutex_relock(&c->btree_trans_lock, seq))
 842                         goto restart;
 843         }
 844         seqmutex_unlock(&c->btree_trans_lock);
 845 }
 846
 847 static ssize_t bch2_btree_deadlock_read(struct file *file, char __user *buf,
 848                                             size_t size, loff_t *ppos)
 849 {
 850         struct dump_iter *i = file->private_data;
 851         struct bch_fs *c = i->c;
 852         ssize_t ret = 0;
 853
 854         i->ubuf = buf;
 855         i->size = size;
 856         i->ret  = 0;
 857
 858         if (!i->iter) {
 859                 btree_deadlock_to_text(&i->buf, c);
 860                 i->iter++;
 861         }
 862
 863         if (i->buf.allocation_failure)
 864                 ret = -ENOMEM;
 865
 866         if (!ret)
 867                 ret = flush_buf(i);
 868
 869         return ret ?: i->ret;
 870 }
 871
 872 static const struct file_operations btree_deadlock_ops = {
 873         .owner          = THIS_MODULE,
 874         .open           = bch2_dump_open,
 875         .release        = bch2_dump_release,
 876         .read           = bch2_btree_deadlock_read,
 877 };
 878
 879 void bch2_fs_debug_exit(struct bch_fs *c)
 880 {
 881         if (!IS_ERR_OR_NULL(c->fs_debug_dir))
 882                 debugfs_remove_recursive(c->fs_debug_dir);
 883 }
 884
 885 static void bch2_fs_debug_btree_init(struct bch_fs *c, struct btree_debug *bd)
 886 {
 887         struct dentry *d;
 888
 889         d = debugfs_create_dir(bch2_btree_id_str(bd->id), c->btree_debug_dir);
 890
 891         debugfs_create_file("keys", 0400, d, bd, &btree_debug_ops);
 892
 893         debugfs_create_file("formats", 0400, d, bd, &btree_format_debug_ops);
 894
 895         debugfs_create_file("bfloat-failed", 0400, d, bd,
 896                             &bfloat_failed_debug_ops);
 897 }
 898
 899 void bch2_fs_debug_init(struct bch_fs *c)
 900 {
 901         struct btree_debug *bd;
 902         char name[100];
 903
 904         if (IS_ERR_OR_NULL(bch_debug))
 905                 return;
 906
 907         snprintf(name, sizeof(name), "%pU", c->sb.user_uuid.b);
 908         c->fs_debug_dir = debugfs_create_dir(name, bch_debug);
 909         if (IS_ERR_OR_NULL(c->fs_debug_dir))
 910                 return;
 911
 912         debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
 913                             c->btree_debug, &cached_btree_nodes_ops);
 914
 915         debugfs_create_file("btree_transactions", 0400, c->fs_debug_dir,
 916                             c->btree_debug, &btree_transactions_ops);
 917
 918         debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
 919                             c->btree_debug, &journal_pins_ops);
 920
 921         debugfs_create_file("btree_updates", 0400, c->fs_debug_dir,
 922                             c->btree_debug, &btree_updates_ops);
 923
 924         debugfs_create_file("btree_transaction_stats", 0400, c->fs_debug_dir,
 925                             c, &btree_transaction_stats_op);
 926
 927         debugfs_create_file("btree_deadlock", 0400, c->fs_debug_dir,
 928                             c->btree_debug, &btree_deadlock_ops);
 929
 930         c->btree_debug_dir = debugfs_create_dir("btrees", c->fs_debug_dir);
 931         if (IS_ERR_OR_NULL(c->btree_debug_dir))
 932                 return;
 933
 934         for (bd = c->btree_debug;
 935              bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
 936              bd++) {
 937                 bd->id = bd - c->btree_debug;
 938                 bch2_fs_debug_btree_init(c, bd);
 939         }
 940 }
 941
 942 #endif
 943
 944 void bch2_debug_exit(void)
 945 {
 946         if (!IS_ERR_OR_NULL(bch_debug))
 947                 debugfs_remove_recursive(bch_debug);
 948 }
 949
 950 int __init bch2_debug_init(void)
 951 {
 952         bch_debug = debugfs_create_dir("bcachefs", NULL);
 953         return 0;
 954 }