fs/bcachefs/btree_update.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "btree_update.h"
   5 #include "btree_iter.h"
   6 #include "btree_journal_iter.h"
   7 #include "btree_locking.h"
   8 #include "buckets.h"
   9 #include "debug.h"
  10 #include "errcode.h"
  11 #include "error.h"
  12 #include "extents.h"
  13 #include "keylist.h"
  14 #include "snapshot.h"
  15 #include "trace.h"
  16
  17 static inline int btree_insert_entry_cmp(const struct btree_insert_entry *l,
  18                                          const struct btree_insert_entry *r)
  19 {
  20         return   cmp_int(l->btree_id,   r->btree_id) ?:
  21                  cmp_int(l->cached,     r->cached) ?:
  22                  -cmp_int(l->level,     r->level) ?:
  23                  bpos_cmp(l->k->k.p,    r->k->k.p);
  24 }
  25
  26 static int __must_check
  27 bch2_trans_update_by_path(struct btree_trans *, btree_path_idx_t,
  28                           struct bkey_i *, enum btree_iter_update_trigger_flags,
  29                           unsigned long ip);
  30
  31 static noinline int extent_front_merge(struct btree_trans *trans,
  32                                        struct btree_iter *iter,
  33                                        struct bkey_s_c k,
  34                                        struct bkey_i **insert,
  35                                        enum btree_iter_update_trigger_flags flags)
  36 {
  37         struct bch_fs *c = trans->c;
  38         struct bkey_i *update;
  39         int ret;
  40
  41         if (unlikely(trans->journal_replay_not_finished))
  42                 return 0;
  43
  44         update = bch2_bkey_make_mut_noupdate(trans, k);
  45         ret = PTR_ERR_OR_ZERO(update);
  46         if (ret)
  47                 return ret;
  48
  49         if (!bch2_bkey_merge(c, bkey_i_to_s(update), bkey_i_to_s_c(*insert)))
  50                 return 0;
  51
  52         ret =   bch2_key_has_snapshot_overwrites(trans, iter->btree_id, k.k->p) ?:
  53                 bch2_key_has_snapshot_overwrites(trans, iter->btree_id, (*insert)->k.p);
  54         if (ret < 0)
  55                 return ret;
  56         if (ret)
  57                 return 0;
  58
  59         ret = bch2_btree_delete_at(trans, iter, flags);
  60         if (ret)
  61                 return ret;
  62
  63         *insert = update;
  64         return 0;
  65 }
  66
  67 static noinline int extent_back_merge(struct btree_trans *trans,
  68                                       struct btree_iter *iter,
  69                                       struct bkey_i *insert,
  70                                       struct bkey_s_c k)
  71 {
  72         struct bch_fs *c = trans->c;
  73         int ret;
  74
  75         if (unlikely(trans->journal_replay_not_finished))
  76                 return 0;
  77
  78         ret =   bch2_key_has_snapshot_overwrites(trans, iter->btree_id, insert->k.p) ?:
  79                 bch2_key_has_snapshot_overwrites(trans, iter->btree_id, k.k->p);
  80         if (ret < 0)
  81                 return ret;
  82         if (ret)
  83                 return 0;
  84
  85         bch2_bkey_merge(c, bkey_i_to_s(insert), k);
  86         return 0;
  87 }
  88
  89 /*
  90  * When deleting, check if we need to emit a whiteout (because we're overwriting
  91  * something in an ancestor snapshot)
  92  */
  93 static int need_whiteout_for_snapshot(struct btree_trans *trans,
  94                                       enum btree_id btree_id, struct bpos pos)
  95 {
  96         struct btree_iter iter;
  97         struct bkey_s_c k;
  98         u32 snapshot = pos.snapshot;
  99         int ret;
 100
 101         if (!bch2_snapshot_parent(trans->c, pos.snapshot))
 102                 return 0;
 103
 104         pos.snapshot++;
 105
 106         for_each_btree_key_norestart(trans, iter, btree_id, pos,
 107                            BTREE_ITER_all_snapshots|
 108                            BTREE_ITER_nopreserve, k, ret) {
 109                 if (!bkey_eq(k.k->p, pos))
 110                         break;
 111
 112                 if (bch2_snapshot_is_ancestor(trans->c, snapshot,
 113                                               k.k->p.snapshot)) {
 114                         ret = !bkey_whiteout(k.k);
 115                         break;
 116                 }
 117         }
 118         bch2_trans_iter_exit(trans, &iter);
 119
 120         return ret;
 121 }
 122
 123 int __bch2_insert_snapshot_whiteouts(struct btree_trans *trans,
 124                                    enum btree_id id,
 125                                    struct bpos old_pos,
 126                                    struct bpos new_pos)
 127 {
 128         struct bch_fs *c = trans->c;
 129         struct btree_iter old_iter, new_iter = { NULL };
 130         struct bkey_s_c old_k, new_k;
 131         snapshot_id_list s;
 132         struct bkey_i *update;
 133         int ret = 0;
 134
 135         if (!bch2_snapshot_has_children(c, old_pos.snapshot))
 136                 return 0;
 137
 138         darray_init(&s);
 139
 140         bch2_trans_iter_init(trans, &old_iter, id, old_pos,
 141                              BTREE_ITER_not_extents|
 142                              BTREE_ITER_all_snapshots);
 143         while ((old_k = bch2_btree_iter_prev(&old_iter)).k &&
 144                !(ret = bkey_err(old_k)) &&
 145                bkey_eq(old_pos, old_k.k->p)) {
 146                 struct bpos whiteout_pos =
 147                         SPOS(new_pos.inode, new_pos.offset, old_k.k->p.snapshot);
 148
 149                 if (!bch2_snapshot_is_ancestor(c, old_k.k->p.snapshot, old_pos.snapshot) ||
 150                     snapshot_list_has_ancestor(c, &s, old_k.k->p.snapshot))
 151                         continue;
 152
 153                 new_k = bch2_bkey_get_iter(trans, &new_iter, id, whiteout_pos,
 154                                            BTREE_ITER_not_extents|
 155                                            BTREE_ITER_intent);
 156                 ret = bkey_err(new_k);
 157                 if (ret)
 158                         break;
 159
 160                 if (new_k.k->type == KEY_TYPE_deleted) {
 161                         update = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
 162                         ret = PTR_ERR_OR_ZERO(update);
 163                         if (ret)
 164                                 break;
 165
 166                         bkey_init(&update->k);
 167                         update->k.p             = whiteout_pos;
 168                         update->k.type          = KEY_TYPE_whiteout;
 169
 170                         ret = bch2_trans_update(trans, &new_iter, update,
 171                                                 BTREE_UPDATE_internal_snapshot_node);
 172                 }
 173                 bch2_trans_iter_exit(trans, &new_iter);
 174
 175                 ret = snapshot_list_add(c, &s, old_k.k->p.snapshot);
 176                 if (ret)
 177                         break;
 178         }
 179         bch2_trans_iter_exit(trans, &new_iter);
 180         bch2_trans_iter_exit(trans, &old_iter);
 181         darray_exit(&s);
 182
 183         return ret;
 184 }
 185
 186 int bch2_trans_update_extent_overwrite(struct btree_trans *trans,
 187                                        struct btree_iter *iter,
 188                                        enum btree_iter_update_trigger_flags flags,
 189                                        struct bkey_s_c old,
 190                                        struct bkey_s_c new)
 191 {
 192         enum btree_id btree_id = iter->btree_id;
 193         struct bkey_i *update;
 194         struct bpos new_start = bkey_start_pos(new.k);
 195         unsigned front_split = bkey_lt(bkey_start_pos(old.k), new_start);
 196         unsigned back_split  = bkey_gt(old.k->p, new.k->p);
 197         unsigned middle_split = (front_split || back_split) &&
 198                 old.k->p.snapshot != new.k->p.snapshot;
 199         unsigned nr_splits = front_split + back_split + middle_split;
 200         int ret = 0, compressed_sectors;
 201
 202         /*
 203          * If we're going to be splitting a compressed extent, note it
 204          * so that __bch2_trans_commit() can increase our disk
 205          * reservation:
 206          */
 207         if (nr_splits > 1 &&
 208             (compressed_sectors = bch2_bkey_sectors_compressed(old)))
 209                 trans->extra_disk_res += compressed_sectors * (nr_splits - 1);
 210
 211         if (front_split) {
 212                 update = bch2_bkey_make_mut_noupdate(trans, old);
 213                 if ((ret = PTR_ERR_OR_ZERO(update)))
 214                         return ret;
 215
 216                 bch2_cut_back(new_start, update);
 217
 218                 ret =   bch2_insert_snapshot_whiteouts(trans, btree_id,
 219                                         old.k->p, update->k.p) ?:
 220                         bch2_btree_insert_nonextent(trans, btree_id, update,
 221                                         BTREE_UPDATE_internal_snapshot_node|flags);
 222                 if (ret)
 223                         return ret;
 224         }
 225
 226         /* If we're overwriting in a different snapshot - middle split: */
 227         if (middle_split) {
 228                 update = bch2_bkey_make_mut_noupdate(trans, old);
 229                 if ((ret = PTR_ERR_OR_ZERO(update)))
 230                         return ret;
 231
 232                 bch2_cut_front(new_start, update);
 233                 bch2_cut_back(new.k->p, update);
 234
 235                 ret =   bch2_insert_snapshot_whiteouts(trans, btree_id,
 236                                         old.k->p, update->k.p) ?:
 237                         bch2_btree_insert_nonextent(trans, btree_id, update,
 238                                           BTREE_UPDATE_internal_snapshot_node|flags);
 239                 if (ret)
 240                         return ret;
 241         }
 242
 243         if (bkey_le(old.k->p, new.k->p)) {
 244                 update = bch2_trans_kmalloc(trans, sizeof(*update));
 245                 if ((ret = PTR_ERR_OR_ZERO(update)))
 246                         return ret;
 247
 248                 bkey_init(&update->k);
 249                 update->k.p = old.k->p;
 250                 update->k.p.snapshot = new.k->p.snapshot;
 251
 252                 if (new.k->p.snapshot != old.k->p.snapshot) {
 253                         update->k.type = KEY_TYPE_whiteout;
 254                 } else if (btree_type_has_snapshots(btree_id)) {
 255                         ret = need_whiteout_for_snapshot(trans, btree_id, update->k.p);
 256                         if (ret < 0)
 257                                 return ret;
 258                         if (ret)
 259                                 update->k.type = KEY_TYPE_whiteout;
 260                 }
 261
 262                 ret = bch2_btree_insert_nonextent(trans, btree_id, update,
 263                                           BTREE_UPDATE_internal_snapshot_node|flags);
 264                 if (ret)
 265                         return ret;
 266         }
 267
 268         if (back_split) {
 269                 update = bch2_bkey_make_mut_noupdate(trans, old);
 270                 if ((ret = PTR_ERR_OR_ZERO(update)))
 271                         return ret;
 272
 273                 bch2_cut_front(new.k->p, update);
 274
 275                 ret = bch2_trans_update_by_path(trans, iter->path, update,
 276                                           BTREE_UPDATE_internal_snapshot_node|
 277                                           flags, _RET_IP_);
 278                 if (ret)
 279                         return ret;
 280         }
 281
 282         return 0;
 283 }
 284
 285 static int bch2_trans_update_extent(struct btree_trans *trans,
 286                                     struct btree_iter *orig_iter,
 287                                     struct bkey_i *insert,
 288                                     enum btree_iter_update_trigger_flags flags)
 289 {
 290         struct btree_iter iter;
 291         struct bkey_s_c k;
 292         enum btree_id btree_id = orig_iter->btree_id;
 293         int ret = 0;
 294
 295         bch2_trans_iter_init(trans, &iter, btree_id, bkey_start_pos(&insert->k),
 296                              BTREE_ITER_intent|
 297                              BTREE_ITER_with_updates|
 298                              BTREE_ITER_not_extents);
 299         k = bch2_btree_iter_peek_max(&iter, POS(insert->k.p.inode, U64_MAX));
 300         if ((ret = bkey_err(k)))
 301                 goto err;
 302         if (!k.k)
 303                 goto out;
 304
 305         if (bkey_eq(k.k->p, bkey_start_pos(&insert->k))) {
 306                 if (bch2_bkey_maybe_mergable(k.k, &insert->k)) {
 307                         ret = extent_front_merge(trans, &iter, k, &insert, flags);
 308                         if (ret)
 309                                 goto err;
 310                 }
 311
 312                 goto next;
 313         }
 314
 315         while (bkey_gt(insert->k.p, bkey_start_pos(k.k))) {
 316                 bool done = bkey_lt(insert->k.p, k.k->p);
 317
 318                 ret = bch2_trans_update_extent_overwrite(trans, &iter, flags, k, bkey_i_to_s_c(insert));
 319                 if (ret)
 320                         goto err;
 321
 322                 if (done)
 323                         goto out;
 324 next:
 325                 bch2_btree_iter_advance(&iter);
 326                 k = bch2_btree_iter_peek_max(&iter, POS(insert->k.p.inode, U64_MAX));
 327                 if ((ret = bkey_err(k)))
 328                         goto err;
 329                 if (!k.k)
 330                         goto out;
 331         }
 332
 333         if (bch2_bkey_maybe_mergable(&insert->k, k.k)) {
 334                 ret = extent_back_merge(trans, &iter, insert, k);
 335                 if (ret)
 336                         goto err;
 337         }
 338 out:
 339         if (!bkey_deleted(&insert->k))
 340                 ret = bch2_btree_insert_nonextent(trans, btree_id, insert, flags);
 341 err:
 342         bch2_trans_iter_exit(trans, &iter);
 343
 344         return ret;
 345 }
 346
 347 static noinline int flush_new_cached_update(struct btree_trans *trans,
 348                                             struct btree_insert_entry *i,
 349                                             enum btree_iter_update_trigger_flags flags,
 350                                             unsigned long ip)
 351 {
 352         struct bkey k;
 353         int ret;
 354
 355         btree_path_idx_t path_idx =
 356                 bch2_path_get(trans, i->btree_id, i->old_k.p, 1, 0,
 357                               BTREE_ITER_intent, _THIS_IP_);
 358         ret = bch2_btree_path_traverse(trans, path_idx, 0);
 359         if (ret)
 360                 goto out;
 361
 362         struct btree_path *btree_path = trans->paths + path_idx;
 363
 364         /*
 365          * The old key in the insert entry might actually refer to an existing
 366          * key in the btree that has been deleted from cache and not yet
 367          * flushed. Check for this and skip the flush so we don't run triggers
 368          * against a stale key.
 369          */
 370         bch2_btree_path_peek_slot_exact(btree_path, &k);
 371         if (!bkey_deleted(&k))
 372                 goto out;
 373
 374         i->key_cache_already_flushed = true;
 375         i->flags |= BTREE_TRIGGER_norun;
 376
 377         btree_path_set_should_be_locked(trans, btree_path);
 378         ret = bch2_trans_update_by_path(trans, path_idx, i->k, flags, ip);
 379 out:
 380         bch2_path_put(trans, path_idx, true);
 381         return ret;
 382 }
 383
 384 static int __must_check
 385 bch2_trans_update_by_path(struct btree_trans *trans, btree_path_idx_t path_idx,
 386                           struct bkey_i *k, enum btree_iter_update_trigger_flags flags,
 387                           unsigned long ip)
 388 {
 389         struct bch_fs *c = trans->c;
 390         struct btree_insert_entry *i, n;
 391         int cmp;
 392
 393         struct btree_path *path = trans->paths + path_idx;
 394         EBUG_ON(!path->should_be_locked);
 395         EBUG_ON(trans->nr_updates >= trans->nr_paths);
 396         EBUG_ON(!bpos_eq(k->k.p, path->pos));
 397
 398         n = (struct btree_insert_entry) {
 399                 .flags          = flags,
 400                 .bkey_type      = __btree_node_type(path->level, path->btree_id),
 401                 .btree_id       = path->btree_id,
 402                 .level          = path->level,
 403                 .cached         = path->cached,
 404                 .path           = path_idx,
 405                 .k              = k,
 406                 .ip_allocated   = ip,
 407         };
 408
 409 #ifdef CONFIG_BCACHEFS_DEBUG
 410         trans_for_each_update(trans, i)
 411                 BUG_ON(i != trans->updates &&
 412                        btree_insert_entry_cmp(i - 1, i) >= 0);
 413 #endif
 414
 415         /*
 416          * Pending updates are kept sorted: first, find position of new update,
 417          * then delete/trim any updates the new update overwrites:
 418          */
 419         for (i = trans->updates; i < trans->updates + trans->nr_updates; i++) {
 420                 cmp = btree_insert_entry_cmp(&n, i);
 421                 if (cmp <= 0)
 422                         break;
 423         }
 424
 425         bool overwrite = !cmp && i < trans->updates + trans->nr_updates;
 426
 427         if (overwrite) {
 428                 EBUG_ON(i->insert_trigger_run || i->overwrite_trigger_run);
 429
 430                 bch2_path_put(trans, i->path, true);
 431                 i->flags        = n.flags;
 432                 i->cached       = n.cached;
 433                 i->k            = n.k;
 434                 i->path         = n.path;
 435                 i->ip_allocated = n.ip_allocated;
 436         } else {
 437                 array_insert_item(trans->updates, trans->nr_updates,
 438                                   i - trans->updates, n);
 439
 440                 i->old_v = bch2_btree_path_peek_slot_exact(path, &i->old_k).v;
 441                 i->old_btree_u64s = !bkey_deleted(&i->old_k) ? i->old_k.u64s : 0;
 442
 443                 if (unlikely(trans->journal_replay_not_finished)) {
 444                         struct bkey_i *j_k =
 445                                 bch2_journal_keys_peek_slot(c, n.btree_id, n.level, k->k.p);
 446
 447                         if (j_k) {
 448                                 i->old_k = j_k->k;
 449                                 i->old_v = &j_k->v;
 450                         }
 451                 }
 452         }
 453
 454         __btree_path_get(trans, trans->paths + i->path, true);
 455
 456         trace_update_by_path(trans, path, i, overwrite);
 457
 458         /*
 459          * If a key is present in the key cache, it must also exist in the
 460          * btree - this is necessary for cache coherency. When iterating over
 461          * a btree that's cached in the key cache, the btree iter code checks
 462          * the key cache - but the key has to exist in the btree for that to
 463          * work:
 464          */
 465         if (path->cached && !i->old_btree_u64s)
 466                 return flush_new_cached_update(trans, i, flags, ip);
 467
 468         return 0;
 469 }
 470
 471 static noinline int bch2_trans_update_get_key_cache(struct btree_trans *trans,
 472                                                     struct btree_iter *iter,
 473                                                     struct btree_path *path)
 474 {
 475         struct btree_path *key_cache_path = btree_iter_key_cache_path(trans, iter);
 476
 477         if (!key_cache_path ||
 478             !key_cache_path->should_be_locked ||
 479             !bpos_eq(key_cache_path->pos, iter->pos)) {
 480                 struct bkey_cached *ck;
 481                 int ret;
 482
 483                 if (!iter->key_cache_path)
 484                         iter->key_cache_path =
 485                                 bch2_path_get(trans, path->btree_id, path->pos, 1, 0,
 486                                               BTREE_ITER_intent|
 487                                               BTREE_ITER_cached, _THIS_IP_);
 488
 489                 iter->key_cache_path =
 490                         bch2_btree_path_set_pos(trans, iter->key_cache_path, path->pos,
 491                                                 iter->flags & BTREE_ITER_intent,
 492                                                 _THIS_IP_);
 493
 494                 ret = bch2_btree_path_traverse(trans, iter->key_cache_path, BTREE_ITER_cached);
 495                 if (unlikely(ret))
 496                         return ret;
 497
 498                 ck = (void *) trans->paths[iter->key_cache_path].l[0].b;
 499
 500                 if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 501                         trace_and_count(trans->c, trans_restart_key_cache_raced, trans, _RET_IP_);
 502                         return btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_raced);
 503                 }
 504
 505                 btree_path_set_should_be_locked(trans, trans->paths + iter->key_cache_path);
 506         }
 507
 508         return 0;
 509 }
 510
 511 int __must_check bch2_trans_update(struct btree_trans *trans, struct btree_iter *iter,
 512                                    struct bkey_i *k, enum btree_iter_update_trigger_flags flags)
 513 {
 514         btree_path_idx_t path_idx = iter->update_path ?: iter->path;
 515         int ret;
 516
 517         if (iter->flags & BTREE_ITER_is_extents)
 518                 return bch2_trans_update_extent(trans, iter, k, flags);
 519
 520         if (bkey_deleted(&k->k) &&
 521             !(flags & BTREE_UPDATE_key_cache_reclaim) &&
 522             (iter->flags & BTREE_ITER_filter_snapshots)) {
 523                 ret = need_whiteout_for_snapshot(trans, iter->btree_id, k->k.p);
 524                 if (unlikely(ret < 0))
 525                         return ret;
 526
 527                 if (ret)
 528                         k->k.type = KEY_TYPE_whiteout;
 529         }
 530
 531         /*
 532          * Ensure that updates to cached btrees go to the key cache:
 533          */
 534         struct btree_path *path = trans->paths + path_idx;
 535         if (!(flags & BTREE_UPDATE_key_cache_reclaim) &&
 536             !path->cached &&
 537             !path->level &&
 538             btree_id_cached(trans->c, path->btree_id)) {
 539                 ret = bch2_trans_update_get_key_cache(trans, iter, path);
 540                 if (ret)
 541                         return ret;
 542
 543                 path_idx = iter->key_cache_path;
 544         }
 545
 546         return bch2_trans_update_by_path(trans, path_idx, k, flags, _RET_IP_);
 547 }
 548
 549 int bch2_btree_insert_clone_trans(struct btree_trans *trans,
 550                                   enum btree_id btree,
 551                                   struct bkey_i *k)
 552 {
 553         struct bkey_i *n = bch2_trans_kmalloc(trans, bkey_bytes(&k->k));
 554         int ret = PTR_ERR_OR_ZERO(n);
 555         if (ret)
 556                 return ret;
 557
 558         bkey_copy(n, k);
 559         return bch2_btree_insert_trans(trans, btree, n, 0);
 560 }
 561
 562 struct jset_entry *__bch2_trans_jset_entry_alloc(struct btree_trans *trans, unsigned u64s)
 563 {
 564         unsigned new_top = trans->journal_entries_u64s + u64s;
 565         unsigned old_size = trans->journal_entries_size;
 566
 567         if (new_top > trans->journal_entries_size) {
 568                 trans->journal_entries_size = roundup_pow_of_two(new_top);
 569
 570                 btree_trans_stats(trans)->journal_entries_size = trans->journal_entries_size;
 571         }
 572
 573         struct jset_entry *n =
 574                 bch2_trans_kmalloc_nomemzero(trans,
 575                                 trans->journal_entries_size * sizeof(u64));
 576         if (IS_ERR(n))
 577                 return ERR_CAST(n);
 578
 579         if (trans->journal_entries)
 580                 memcpy(n, trans->journal_entries, old_size * sizeof(u64));
 581         trans->journal_entries = n;
 582
 583         struct jset_entry *e = btree_trans_journal_entries_top(trans);
 584         trans->journal_entries_u64s = new_top;
 585         return e;
 586 }
 587
 588 int bch2_bkey_get_empty_slot(struct btree_trans *trans, struct btree_iter *iter,
 589                              enum btree_id btree, struct bpos end)
 590 {
 591         bch2_trans_iter_init(trans, iter, btree, end, BTREE_ITER_intent);
 592         struct bkey_s_c k = bch2_btree_iter_peek_prev(iter);
 593         int ret = bkey_err(k);
 594         if (ret)
 595                 goto err;
 596
 597         bch2_btree_iter_advance(iter);
 598         k = bch2_btree_iter_peek_slot(iter);
 599         ret = bkey_err(k);
 600         if (ret)
 601                 goto err;
 602
 603         BUG_ON(k.k->type != KEY_TYPE_deleted);
 604
 605         if (bkey_gt(k.k->p, end)) {
 606                 ret = -BCH_ERR_ENOSPC_btree_slot;
 607                 goto err;
 608         }
 609
 610         return 0;
 611 err:
 612         bch2_trans_iter_exit(trans, iter);
 613         return ret;
 614 }
 615
 616 void bch2_trans_commit_hook(struct btree_trans *trans,
 617                             struct btree_trans_commit_hook *h)
 618 {
 619         h->next = trans->hooks;
 620         trans->hooks = h;
 621 }
 622
 623 int bch2_btree_insert_nonextent(struct btree_trans *trans,
 624                                 enum btree_id btree, struct bkey_i *k,
 625                                 enum btree_iter_update_trigger_flags flags)
 626 {
 627         struct btree_iter iter;
 628         int ret;
 629
 630         bch2_trans_iter_init(trans, &iter, btree, k->k.p,
 631                              BTREE_ITER_cached|
 632                              BTREE_ITER_not_extents|
 633                              BTREE_ITER_intent);
 634         ret   = bch2_btree_iter_traverse(&iter) ?:
 635                 bch2_trans_update(trans, &iter, k, flags);
 636         bch2_trans_iter_exit(trans, &iter);
 637         return ret;
 638 }
 639
 640 int bch2_btree_insert_trans(struct btree_trans *trans, enum btree_id id,
 641                             struct bkey_i *k, enum btree_iter_update_trigger_flags flags)
 642 {
 643         struct btree_iter iter;
 644         bch2_trans_iter_init(trans, &iter, id, bkey_start_pos(&k->k),
 645                              BTREE_ITER_intent|flags);
 646         int ret = bch2_btree_iter_traverse(&iter) ?:
 647                   bch2_trans_update(trans, &iter, k, flags);
 648         bch2_trans_iter_exit(trans, &iter);
 649         return ret;
 650 }
 651
 652 /**
 653  * bch2_btree_insert - insert keys into the extent btree
 654  * @c:                  pointer to struct bch_fs
 655  * @id:                 btree to insert into
 656  * @k:                  key to insert
 657  * @disk_res:           must be non-NULL whenever inserting or potentially
 658  *                      splitting data extents
 659  * @flags:              transaction commit flags
 660  * @iter_flags:         btree iter update trigger flags
 661  *
 662  * Returns:             0 on success, error code on failure
 663  */
 664 int bch2_btree_insert(struct bch_fs *c, enum btree_id id, struct bkey_i *k,
 665                       struct disk_reservation *disk_res, int flags,
 666                       enum btree_iter_update_trigger_flags iter_flags)
 667 {
 668         return bch2_trans_commit_do(c, disk_res, NULL, flags,
 669                              bch2_btree_insert_trans(trans, id, k, iter_flags));
 670 }
 671
 672 int bch2_btree_delete_at(struct btree_trans *trans,
 673                          struct btree_iter *iter, unsigned update_flags)
 674 {
 675         struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(*k));
 676         int ret = PTR_ERR_OR_ZERO(k);
 677         if (ret)
 678                 return ret;
 679
 680         bkey_init(&k->k);
 681         k->k.p = iter->pos;
 682         return bch2_trans_update(trans, iter, k, update_flags);
 683 }
 684
 685 int bch2_btree_delete(struct btree_trans *trans,
 686                       enum btree_id btree, struct bpos pos,
 687                       unsigned update_flags)
 688 {
 689         struct btree_iter iter;
 690         int ret;
 691
 692         bch2_trans_iter_init(trans, &iter, btree, pos,
 693                              BTREE_ITER_cached|
 694                              BTREE_ITER_intent);
 695         ret   = bch2_btree_iter_traverse(&iter) ?:
 696                 bch2_btree_delete_at(trans, &iter, update_flags);
 697         bch2_trans_iter_exit(trans, &iter);
 698
 699         return ret;
 700 }
 701
 702 int bch2_btree_delete_range_trans(struct btree_trans *trans, enum btree_id id,
 703                                   struct bpos start, struct bpos end,
 704                                   unsigned update_flags,
 705                                   u64 *journal_seq)
 706 {
 707         u32 restart_count = trans->restart_count;
 708         struct btree_iter iter;
 709         struct bkey_s_c k;
 710         int ret = 0;
 711
 712         bch2_trans_iter_init(trans, &iter, id, start, BTREE_ITER_intent);
 713         while ((k = bch2_btree_iter_peek_max(&iter, end)).k) {
 714                 struct disk_reservation disk_res =
 715                         bch2_disk_reservation_init(trans->c, 0);
 716                 struct bkey_i delete;
 717
 718                 ret = bkey_err(k);
 719                 if (ret)
 720                         goto err;
 721
 722                 bkey_init(&delete.k);
 723
 724                 /*
 725                  * This could probably be more efficient for extents:
 726                  */
 727
 728                 /*
 729                  * For extents, iter.pos won't necessarily be the same as
 730                  * bkey_start_pos(k.k) (for non extents they always will be the
 731                  * same). It's important that we delete starting from iter.pos
 732                  * because the range we want to delete could start in the middle
 733                  * of k.
 734                  *
 735                  * (bch2_btree_iter_peek() does guarantee that iter.pos >=
 736                  * bkey_start_pos(k.k)).
 737                  */
 738                 delete.k.p = iter.pos;
 739
 740                 if (iter.flags & BTREE_ITER_is_extents)
 741                         bch2_key_resize(&delete.k,
 742                                         bpos_min(end, k.k->p).offset -
 743                                         iter.pos.offset);
 744
 745                 ret   = bch2_trans_update(trans, &iter, &delete, update_flags) ?:
 746                         bch2_trans_commit(trans, &disk_res, journal_seq,
 747                                           BCH_TRANS_COMMIT_no_enospc);
 748                 bch2_disk_reservation_put(trans->c, &disk_res);
 749 err:
 750                 /*
 751                  * the bch2_trans_begin() call is in a weird place because we
 752                  * need to call it after every transaction commit, to avoid path
 753                  * overflow, but don't want to call it if the delete operation
 754                  * is a no-op and we have no work to do:
 755                  */
 756                 bch2_trans_begin(trans);
 757
 758                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 759                         ret = 0;
 760                 if (ret)
 761                         break;
 762         }
 763         bch2_trans_iter_exit(trans, &iter);
 764
 765         return ret ?: trans_was_restarted(trans, restart_count);
 766 }
 767
 768 /*
 769  * bch_btree_delete_range - delete everything within a given range
 770  *
 771  * Range is a half open interval - [start, end)
 772  */
 773 int bch2_btree_delete_range(struct bch_fs *c, enum btree_id id,
 774                             struct bpos start, struct bpos end,
 775                             unsigned update_flags,
 776                             u64 *journal_seq)
 777 {
 778         int ret = bch2_trans_run(c,
 779                         bch2_btree_delete_range_trans(trans, id, start, end,
 780                                                       update_flags, journal_seq));
 781         if (ret == -BCH_ERR_transaction_restart_nested)
 782                 ret = 0;
 783         return ret;
 784 }
 785
 786 int bch2_btree_bit_mod_iter(struct btree_trans *trans, struct btree_iter *iter, bool set)
 787 {
 788         struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(*k));
 789         int ret = PTR_ERR_OR_ZERO(k);
 790         if (ret)
 791                 return ret;
 792
 793         bkey_init(&k->k);
 794         k->k.type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
 795         k->k.p = iter->pos;
 796         if (iter->flags & BTREE_ITER_is_extents)
 797                 bch2_key_resize(&k->k, 1);
 798
 799         return bch2_trans_update(trans, iter, k, 0);
 800 }
 801
 802 int bch2_btree_bit_mod(struct btree_trans *trans, enum btree_id btree,
 803                        struct bpos pos, bool set)
 804 {
 805         struct btree_iter iter;
 806         bch2_trans_iter_init(trans, &iter, btree, pos, BTREE_ITER_intent);
 807
 808         int ret = bch2_btree_iter_traverse(&iter) ?:
 809                   bch2_btree_bit_mod_iter(trans, &iter, set);
 810         bch2_trans_iter_exit(trans, &iter);
 811         return ret;
 812 }
 813
 814 int bch2_btree_bit_mod_buffered(struct btree_trans *trans, enum btree_id btree,
 815                                 struct bpos pos, bool set)
 816 {
 817         struct bkey_i k;
 818
 819         bkey_init(&k.k);
 820         k.k.type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
 821         k.k.p = pos;
 822
 823         return bch2_trans_update_buffered(trans, btree, &k);
 824 }
 825
 826 int bch2_trans_log_msg(struct btree_trans *trans, struct printbuf *buf)
 827 {
 828         unsigned u64s = DIV_ROUND_UP(buf->pos, sizeof(u64));
 829         prt_chars(buf, '\0', u64s * sizeof(u64) - buf->pos);
 830
 831         int ret = buf->allocation_failure ? -BCH_ERR_ENOMEM_trans_log_msg : 0;
 832         if (ret)
 833                 return ret;
 834
 835         struct jset_entry *e = bch2_trans_jset_entry_alloc(trans, jset_u64s(u64s));
 836         ret = PTR_ERR_OR_ZERO(e);
 837         if (ret)
 838                 return ret;
 839
 840         struct jset_entry_log *l = container_of(e, struct jset_entry_log, entry);
 841         journal_entry_init(e, BCH_JSET_ENTRY_log, 0, 1, u64s);
 842         memcpy(l->d, buf->buf, buf->pos);
 843         return 0;
 844 }
 845
 846 __printf(3, 0)
 847 static int
 848 __bch2_fs_log_msg(struct bch_fs *c, unsigned commit_flags, const char *fmt,
 849                   va_list args)
 850 {
 851         struct printbuf buf = PRINTBUF;
 852         prt_vprintf(&buf, fmt, args);
 853
 854         unsigned u64s = DIV_ROUND_UP(buf.pos, sizeof(u64));
 855         prt_chars(&buf, '\0', u64s * sizeof(u64) - buf.pos);
 856
 857         int ret = buf.allocation_failure ? -BCH_ERR_ENOMEM_trans_log_msg : 0;
 858         if (ret)
 859                 goto err;
 860
 861         if (!test_bit(JOURNAL_running, &c->journal.flags)) {
 862                 ret = darray_make_room(&c->journal.early_journal_entries, jset_u64s(u64s));
 863                 if (ret)
 864                         goto err;
 865
 866                 struct jset_entry_log *l = (void *) &darray_top(c->journal.early_journal_entries);
 867                 journal_entry_init(&l->entry, BCH_JSET_ENTRY_log, 0, 1, u64s);
 868                 memcpy(l->d, buf.buf, buf.pos);
 869                 c->journal.early_journal_entries.nr += jset_u64s(u64s);
 870         } else {
 871                 ret = bch2_trans_commit_do(c, NULL, NULL, commit_flags,
 872                         bch2_trans_log_msg(trans, &buf));
 873         }
 874 err:
 875         printbuf_exit(&buf);
 876         return ret;
 877 }
 878
 879 __printf(2, 3)
 880 int bch2_fs_log_msg(struct bch_fs *c, const char *fmt, ...)
 881 {
 882         va_list args;
 883         int ret;
 884
 885         va_start(args, fmt);
 886         ret = __bch2_fs_log_msg(c, 0, fmt, args);
 887         va_end(args);
 888         return ret;
 889 }
 890
 891 /*
 892  * Use for logging messages during recovery to enable reserved space and avoid
 893  * blocking.
 894  */
 895 __printf(2, 3)
 896 int bch2_journal_log_msg(struct bch_fs *c, const char *fmt, ...)
 897 {
 898         va_list args;
 899         int ret;
 900
 901         va_start(args, fmt);
 902         ret = __bch2_fs_log_msg(c, BCH_WATERMARK_reclaim, fmt, args);
 903         va_end(args);
 904         return ret;
 905 }