git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/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_update_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_update_flags flags)
  36 {
  37         struct bch_fs *c = trans->c;
  38         struct bkey_i *update;
  39         int ret;
  40
  41         update = bch2_bkey_make_mut_noupdate(trans, k);
  42         ret = PTR_ERR_OR_ZERO(update);
  43         if (ret)
  44                 return ret;
  45
  46         if (!bch2_bkey_merge(c, bkey_i_to_s(update), bkey_i_to_s_c(*insert)))
  47                 return 0;
  48
  49         ret =   bch2_key_has_snapshot_overwrites(trans, iter->btree_id, k.k->p) ?:
  50                 bch2_key_has_snapshot_overwrites(trans, iter->btree_id, (*insert)->k.p);
  51         if (ret < 0)
  52                 return ret;
  53         if (ret)
  54                 return 0;
  55
  56         ret = bch2_btree_delete_at(trans, iter, flags);
  57         if (ret)
  58                 return ret;
  59
  60         *insert = update;
  61         return 0;
  62 }
  63
  64 static noinline int extent_back_merge(struct btree_trans *trans,
  65                                       struct btree_iter *iter,
  66                                       struct bkey_i *insert,
  67                                       struct bkey_s_c k)
  68 {
  69         struct bch_fs *c = trans->c;
  70         int ret;
  71
  72         ret =   bch2_key_has_snapshot_overwrites(trans, iter->btree_id, insert->k.p) ?:
  73                 bch2_key_has_snapshot_overwrites(trans, iter->btree_id, k.k->p);
  74         if (ret < 0)
  75                 return ret;
  76         if (ret)
  77                 return 0;
  78
  79         bch2_bkey_merge(c, bkey_i_to_s(insert), k);
  80         return 0;
  81 }
  82
  83 /*
  84  * When deleting, check if we need to emit a whiteout (because we're overwriting
  85  * something in an ancestor snapshot)
  86  */
  87 static int need_whiteout_for_snapshot(struct btree_trans *trans,
  88                                       enum btree_id btree_id, struct bpos pos)
  89 {
  90         struct btree_iter iter;
  91         struct bkey_s_c k;
  92         u32 snapshot = pos.snapshot;
  93         int ret;
  94
  95         if (!bch2_snapshot_parent(trans->c, pos.snapshot))
  96                 return 0;
  97
  98         pos.snapshot++;
  99
 100         for_each_btree_key_norestart(trans, iter, btree_id, pos,
 101                            BTREE_ITER_ALL_SNAPSHOTS|
 102                            BTREE_ITER_NOPRESERVE, k, ret) {
 103                 if (!bkey_eq(k.k->p, pos))
 104                         break;
 105
 106                 if (bch2_snapshot_is_ancestor(trans->c, snapshot,
 107                                               k.k->p.snapshot)) {
 108                         ret = !bkey_whiteout(k.k);
 109                         break;
 110                 }
 111         }
 112         bch2_trans_iter_exit(trans, &iter);
 113
 114         return ret;
 115 }
 116
 117 int __bch2_insert_snapshot_whiteouts(struct btree_trans *trans,
 118                                    enum btree_id id,
 119                                    struct bpos old_pos,
 120                                    struct bpos new_pos)
 121 {
 122         struct bch_fs *c = trans->c;
 123         struct btree_iter old_iter, new_iter = { NULL };
 124         struct bkey_s_c old_k, new_k;
 125         snapshot_id_list s;
 126         struct bkey_i *update;
 127         int ret = 0;
 128
 129         if (!bch2_snapshot_has_children(c, old_pos.snapshot))
 130                 return 0;
 131
 132         darray_init(&s);
 133
 134         bch2_trans_iter_init(trans, &old_iter, id, old_pos,
 135                              BTREE_ITER_NOT_EXTENTS|
 136                              BTREE_ITER_ALL_SNAPSHOTS);
 137         while ((old_k = bch2_btree_iter_prev(&old_iter)).k &&
 138                !(ret = bkey_err(old_k)) &&
 139                bkey_eq(old_pos, old_k.k->p)) {
 140                 struct bpos whiteout_pos =
 141                         SPOS(new_pos.inode, new_pos.offset, old_k.k->p.snapshot);;
 142
 143                 if (!bch2_snapshot_is_ancestor(c, old_k.k->p.snapshot, old_pos.snapshot) ||
 144                     snapshot_list_has_ancestor(c, &s, old_k.k->p.snapshot))
 145                         continue;
 146
 147                 new_k = bch2_bkey_get_iter(trans, &new_iter, id, whiteout_pos,
 148                                            BTREE_ITER_NOT_EXTENTS|
 149                                            BTREE_ITER_INTENT);
 150                 ret = bkey_err(new_k);
 151                 if (ret)
 152                         break;
 153
 154                 if (new_k.k->type == KEY_TYPE_deleted) {
 155                         update = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
 156                         ret = PTR_ERR_OR_ZERO(update);
 157                         if (ret)
 158                                 break;
 159
 160                         bkey_init(&update->k);
 161                         update->k.p             = whiteout_pos;
 162                         update->k.type          = KEY_TYPE_whiteout;
 163
 164                         ret = bch2_trans_update(trans, &new_iter, update,
 165                                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 166                 }
 167                 bch2_trans_iter_exit(trans, &new_iter);
 168
 169                 ret = snapshot_list_add(c, &s, old_k.k->p.snapshot);
 170                 if (ret)
 171                         break;
 172         }
 173         bch2_trans_iter_exit(trans, &new_iter);
 174         bch2_trans_iter_exit(trans, &old_iter);
 175         darray_exit(&s);
 176
 177         return ret;
 178 }
 179
 180 int bch2_trans_update_extent_overwrite(struct btree_trans *trans,
 181                                        struct btree_iter *iter,
 182                                        enum btree_update_flags flags,
 183                                        struct bkey_s_c old,
 184                                        struct bkey_s_c new)
 185 {
 186         enum btree_id btree_id = iter->btree_id;
 187         struct bkey_i *update;
 188         struct bpos new_start = bkey_start_pos(new.k);
 189         bool front_split = bkey_lt(bkey_start_pos(old.k), new_start);
 190         bool back_split  = bkey_gt(old.k->p, new.k->p);
 191         int ret = 0, compressed_sectors;
 192
 193         /*
 194          * If we're going to be splitting a compressed extent, note it
 195          * so that __bch2_trans_commit() can increase our disk
 196          * reservation:
 197          */
 198         if (((front_split && back_split) ||
 199              ((front_split || back_split) && old.k->p.snapshot != new.k->p.snapshot)) &&
 200             (compressed_sectors = bch2_bkey_sectors_compressed(old)))
 201                 trans->extra_disk_res += compressed_sectors;
 202
 203         if (front_split) {
 204                 update = bch2_bkey_make_mut_noupdate(trans, old);
 205                 if ((ret = PTR_ERR_OR_ZERO(update)))
 206                         return ret;
 207
 208                 bch2_cut_back(new_start, update);
 209
 210                 ret =   bch2_insert_snapshot_whiteouts(trans, btree_id,
 211                                         old.k->p, update->k.p) ?:
 212                         bch2_btree_insert_nonextent(trans, btree_id, update,
 213                                         BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|flags);
 214                 if (ret)
 215                         return ret;
 216         }
 217
 218         /* If we're overwriting in a different snapshot - middle split: */
 219         if (old.k->p.snapshot != new.k->p.snapshot &&
 220             (front_split || back_split)) {
 221                 update = bch2_bkey_make_mut_noupdate(trans, old);
 222                 if ((ret = PTR_ERR_OR_ZERO(update)))
 223                         return ret;
 224
 225                 bch2_cut_front(new_start, update);
 226                 bch2_cut_back(new.k->p, update);
 227
 228                 ret =   bch2_insert_snapshot_whiteouts(trans, btree_id,
 229                                         old.k->p, update->k.p) ?:
 230                         bch2_btree_insert_nonextent(trans, btree_id, update,
 231                                           BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|flags);
 232                 if (ret)
 233                         return ret;
 234         }
 235
 236         if (bkey_le(old.k->p, new.k->p)) {
 237                 update = bch2_trans_kmalloc(trans, sizeof(*update));
 238                 if ((ret = PTR_ERR_OR_ZERO(update)))
 239                         return ret;
 240
 241                 bkey_init(&update->k);
 242                 update->k.p = old.k->p;
 243                 update->k.p.snapshot = new.k->p.snapshot;
 244
 245                 if (new.k->p.snapshot != old.k->p.snapshot) {
 246                         update->k.type = KEY_TYPE_whiteout;
 247                 } else if (btree_type_has_snapshots(btree_id)) {
 248                         ret = need_whiteout_for_snapshot(trans, btree_id, update->k.p);
 249                         if (ret < 0)
 250                                 return ret;
 251                         if (ret)
 252                                 update->k.type = KEY_TYPE_whiteout;
 253                 }
 254
 255                 ret = bch2_btree_insert_nonextent(trans, btree_id, update,
 256                                           BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|flags);
 257                 if (ret)
 258                         return ret;
 259         }
 260
 261         if (back_split) {
 262                 update = bch2_bkey_make_mut_noupdate(trans, old);
 263                 if ((ret = PTR_ERR_OR_ZERO(update)))
 264                         return ret;
 265
 266                 bch2_cut_front(new.k->p, update);
 267
 268                 ret = bch2_trans_update_by_path(trans, iter->path, update,
 269                                           BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|
 270                                           flags, _RET_IP_);
 271                 if (ret)
 272                         return ret;
 273         }
 274
 275         return 0;
 276 }
 277
 278 static int bch2_trans_update_extent(struct btree_trans *trans,
 279                                     struct btree_iter *orig_iter,
 280                                     struct bkey_i *insert,
 281                                     enum btree_update_flags flags)
 282 {
 283         struct btree_iter iter;
 284         struct bkey_s_c k;
 285         enum btree_id btree_id = orig_iter->btree_id;
 286         int ret = 0;
 287
 288         bch2_trans_iter_init(trans, &iter, btree_id, bkey_start_pos(&insert->k),
 289                              BTREE_ITER_INTENT|
 290                              BTREE_ITER_WITH_UPDATES|
 291                              BTREE_ITER_NOT_EXTENTS);
 292         k = bch2_btree_iter_peek_upto(&iter, POS(insert->k.p.inode, U64_MAX));
 293         if ((ret = bkey_err(k)))
 294                 goto err;
 295         if (!k.k)
 296                 goto out;
 297
 298         if (bkey_eq(k.k->p, bkey_start_pos(&insert->k))) {
 299                 if (bch2_bkey_maybe_mergable(k.k, &insert->k)) {
 300                         ret = extent_front_merge(trans, &iter, k, &insert, flags);
 301                         if (ret)
 302                                 goto err;
 303                 }
 304
 305                 goto next;
 306         }
 307
 308         while (bkey_gt(insert->k.p, bkey_start_pos(k.k))) {
 309                 bool done = bkey_lt(insert->k.p, k.k->p);
 310
 311                 ret = bch2_trans_update_extent_overwrite(trans, &iter, flags, k, bkey_i_to_s_c(insert));
 312                 if (ret)
 313                         goto err;
 314
 315                 if (done)
 316                         goto out;
 317 next:
 318                 bch2_btree_iter_advance(&iter);
 319                 k = bch2_btree_iter_peek_upto(&iter, POS(insert->k.p.inode, U64_MAX));
 320                 if ((ret = bkey_err(k)))
 321                         goto err;
 322                 if (!k.k)
 323                         goto out;
 324         }
 325
 326         if (bch2_bkey_maybe_mergable(&insert->k, k.k)) {
 327                 ret = extent_back_merge(trans, &iter, insert, k);
 328                 if (ret)
 329                         goto err;
 330         }
 331 out:
 332         if (!bkey_deleted(&insert->k))
 333                 ret = bch2_btree_insert_nonextent(trans, btree_id, insert, flags);
 334 err:
 335         bch2_trans_iter_exit(trans, &iter);
 336
 337         return ret;
 338 }
 339
 340 static noinline int flush_new_cached_update(struct btree_trans *trans,
 341                                             struct btree_insert_entry *i,
 342                                             enum btree_update_flags flags,
 343                                             unsigned long ip)
 344 {
 345         struct bkey k;
 346         int ret;
 347
 348         btree_path_idx_t path_idx =
 349                 bch2_path_get(trans, i->btree_id, i->old_k.p, 1, 0,
 350                               BTREE_ITER_INTENT, _THIS_IP_);
 351         ret = bch2_btree_path_traverse(trans, path_idx, 0);
 352         if (ret)
 353                 goto out;
 354
 355         struct btree_path *btree_path = trans->paths + path_idx;
 356
 357         /*
 358          * The old key in the insert entry might actually refer to an existing
 359          * key in the btree that has been deleted from cache and not yet
 360          * flushed. Check for this and skip the flush so we don't run triggers
 361          * against a stale key.
 362          */
 363         bch2_btree_path_peek_slot_exact(btree_path, &k);
 364         if (!bkey_deleted(&k))
 365                 goto out;
 366
 367         i->key_cache_already_flushed = true;
 368         i->flags |= BTREE_TRIGGER_NORUN;
 369
 370         btree_path_set_should_be_locked(btree_path);
 371         ret = bch2_trans_update_by_path(trans, path_idx, i->k, flags, ip);
 372 out:
 373         bch2_path_put(trans, path_idx, true);
 374         return ret;
 375 }
 376
 377 static int __must_check
 378 bch2_trans_update_by_path(struct btree_trans *trans, btree_path_idx_t path_idx,
 379                           struct bkey_i *k, enum btree_update_flags flags,
 380                           unsigned long ip)
 381 {
 382         struct bch_fs *c = trans->c;
 383         struct btree_insert_entry *i, n;
 384         int cmp;
 385
 386         struct btree_path *path = trans->paths + path_idx;
 387         EBUG_ON(!path->should_be_locked);
 388         EBUG_ON(trans->nr_updates >= trans->nr_paths);
 389         EBUG_ON(!bpos_eq(k->k.p, path->pos));
 390
 391         n = (struct btree_insert_entry) {
 392                 .flags          = flags,
 393                 .bkey_type      = __btree_node_type(path->level, path->btree_id),
 394                 .btree_id       = path->btree_id,
 395                 .level          = path->level,
 396                 .cached         = path->cached,
 397                 .path           = path_idx,
 398                 .k              = k,
 399                 .ip_allocated   = ip,
 400         };
 401
 402 #ifdef CONFIG_BCACHEFS_DEBUG
 403         trans_for_each_update(trans, i)
 404                 BUG_ON(i != trans->updates &&
 405                        btree_insert_entry_cmp(i - 1, i) >= 0);
 406 #endif
 407
 408         /*
 409          * Pending updates are kept sorted: first, find position of new update,
 410          * then delete/trim any updates the new update overwrites:
 411          */
 412         trans_for_each_update(trans, i) {
 413                 cmp = btree_insert_entry_cmp(&n, i);
 414                 if (cmp <= 0)
 415                         break;
 416         }
 417
 418         if (!cmp && i < trans->updates + trans->nr_updates) {
 419                 EBUG_ON(i->insert_trigger_run || i->overwrite_trigger_run);
 420
 421                 bch2_path_put(trans, i->path, true);
 422                 i->flags        = n.flags;
 423                 i->cached       = n.cached;
 424                 i->k            = n.k;
 425                 i->path         = n.path;
 426                 i->ip_allocated = n.ip_allocated;
 427         } else {
 428                 array_insert_item(trans->updates, trans->nr_updates,
 429                                   i - trans->updates, n);
 430
 431                 i->old_v = bch2_btree_path_peek_slot_exact(path, &i->old_k).v;
 432                 i->old_btree_u64s = !bkey_deleted(&i->old_k) ? i->old_k.u64s : 0;
 433
 434                 if (unlikely(trans->journal_replay_not_finished)) {
 435                         struct bkey_i *j_k =
 436                                 bch2_journal_keys_peek_slot(c, n.btree_id, n.level, k->k.p);
 437
 438                         if (j_k) {
 439                                 i->old_k = j_k->k;
 440                                 i->old_v = &j_k->v;
 441                         }
 442                 }
 443         }
 444
 445         __btree_path_get(trans->paths + i->path, true);
 446
 447         /*
 448          * If a key is present in the key cache, it must also exist in the
 449          * btree - this is necessary for cache coherency. When iterating over
 450          * a btree that's cached in the key cache, the btree iter code checks
 451          * the key cache - but the key has to exist in the btree for that to
 452          * work:
 453          */
 454         if (path->cached && bkey_deleted(&i->old_k))
 455                 return flush_new_cached_update(trans, i, flags, ip);
 456
 457         return 0;
 458 }
 459
 460 static noinline int bch2_trans_update_get_key_cache(struct btree_trans *trans,
 461                                                     struct btree_iter *iter,
 462                                                     struct btree_path *path)
 463 {
 464         struct btree_path *key_cache_path = btree_iter_key_cache_path(trans, iter);
 465
 466         if (!key_cache_path ||
 467             !key_cache_path->should_be_locked ||
 468             !bpos_eq(key_cache_path->pos, iter->pos)) {
 469                 struct bkey_cached *ck;
 470                 int ret;
 471
 472                 if (!iter->key_cache_path)
 473                         iter->key_cache_path =
 474                                 bch2_path_get(trans, path->btree_id, path->pos, 1, 0,
 475                                               BTREE_ITER_INTENT|
 476                                               BTREE_ITER_CACHED, _THIS_IP_);
 477
 478                 iter->key_cache_path =
 479                         bch2_btree_path_set_pos(trans, iter->key_cache_path, path->pos,
 480                                                 iter->flags & BTREE_ITER_INTENT,
 481                                                 _THIS_IP_);
 482
 483                 ret = bch2_btree_path_traverse(trans, iter->key_cache_path, BTREE_ITER_CACHED);
 484                 if (unlikely(ret))
 485                         return ret;
 486
 487                 ck = (void *) trans->paths[iter->key_cache_path].l[0].b;
 488
 489                 if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 490                         trace_and_count(trans->c, trans_restart_key_cache_raced, trans, _RET_IP_);
 491                         return btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_raced);
 492                 }
 493
 494                 btree_path_set_should_be_locked(trans->paths + iter->key_cache_path);
 495         }
 496
 497         return 0;
 498 }
 499
 500 int __must_check bch2_trans_update(struct btree_trans *trans, struct btree_iter *iter,
 501                                    struct bkey_i *k, enum btree_update_flags flags)
 502 {
 503         btree_path_idx_t path_idx = iter->update_path ?: iter->path;
 504         int ret;
 505
 506         if (iter->flags & BTREE_ITER_IS_EXTENTS)
 507                 return bch2_trans_update_extent(trans, iter, k, flags);
 508
 509         if (bkey_deleted(&k->k) &&
 510             !(flags & BTREE_UPDATE_KEY_CACHE_RECLAIM) &&
 511             (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)) {
 512                 ret = need_whiteout_for_snapshot(trans, iter->btree_id, k->k.p);
 513                 if (unlikely(ret < 0))
 514                         return ret;
 515
 516                 if (ret)
 517                         k->k.type = KEY_TYPE_whiteout;
 518         }
 519
 520         /*
 521          * Ensure that updates to cached btrees go to the key cache:
 522          */
 523         struct btree_path *path = trans->paths + path_idx;
 524         if (!(flags & BTREE_UPDATE_KEY_CACHE_RECLAIM) &&
 525             !path->cached &&
 526             !path->level &&
 527             btree_id_cached(trans->c, path->btree_id)) {
 528                 ret = bch2_trans_update_get_key_cache(trans, iter, path);
 529                 if (ret)
 530                         return ret;
 531
 532                 path_idx = iter->key_cache_path;
 533         }
 534
 535         return bch2_trans_update_by_path(trans, path_idx, k, flags, _RET_IP_);
 536 }
 537
 538 int bch2_btree_insert_clone_trans(struct btree_trans *trans,
 539                                   enum btree_id btree,
 540                                   struct bkey_i *k)
 541 {
 542         struct bkey_i *n = bch2_trans_kmalloc(trans, bkey_bytes(&k->k));
 543         int ret = PTR_ERR_OR_ZERO(n);
 544         if (ret)
 545                 return ret;
 546
 547         bkey_copy(n, k);
 548         return bch2_btree_insert_trans(trans, btree, n, 0);
 549 }
 550
 551 struct jset_entry *__bch2_trans_jset_entry_alloc(struct btree_trans *trans, unsigned u64s)
 552 {
 553         unsigned new_top = trans->journal_entries_u64s + u64s;
 554         unsigned old_size = trans->journal_entries_size;
 555
 556         if (new_top > trans->journal_entries_size) {
 557                 trans->journal_entries_size = roundup_pow_of_two(new_top);
 558
 559                 struct btree_transaction_stats *s = btree_trans_stats(trans);
 560                 if (s)
 561                         s->journal_entries_size = trans->journal_entries_size;
 562         }
 563
 564         struct jset_entry *n =
 565                 bch2_trans_kmalloc_nomemzero(trans,
 566                                 trans->journal_entries_size * sizeof(u64));
 567         if (IS_ERR(n))
 568                 return ERR_CAST(n);
 569
 570         if (trans->journal_entries)
 571                 memcpy(n, trans->journal_entries, old_size * sizeof(u64));
 572         trans->journal_entries = n;
 573
 574         struct jset_entry *e = btree_trans_journal_entries_top(trans);
 575         trans->journal_entries_u64s = new_top;
 576         return e;
 577 }
 578
 579 int bch2_bkey_get_empty_slot(struct btree_trans *trans, struct btree_iter *iter,
 580                              enum btree_id btree, struct bpos end)
 581 {
 582         struct bkey_s_c k;
 583         int ret = 0;
 584
 585         bch2_trans_iter_init(trans, iter, btree, POS_MAX, BTREE_ITER_INTENT);
 586         k = bch2_btree_iter_prev(iter);
 587         ret = bkey_err(k);
 588         if (ret)
 589                 goto err;
 590
 591         bch2_btree_iter_advance(iter);
 592         k = bch2_btree_iter_peek_slot(iter);
 593         ret = bkey_err(k);
 594         if (ret)
 595                 goto err;
 596
 597         BUG_ON(k.k->type != KEY_TYPE_deleted);
 598
 599         if (bkey_gt(k.k->p, end)) {
 600                 ret = -BCH_ERR_ENOSPC_btree_slot;
 601                 goto err;
 602         }
 603
 604         return 0;
 605 err:
 606         bch2_trans_iter_exit(trans, iter);
 607         return ret;
 608 }
 609
 610 void bch2_trans_commit_hook(struct btree_trans *trans,
 611                             struct btree_trans_commit_hook *h)
 612 {
 613         h->next = trans->hooks;
 614         trans->hooks = h;
 615 }
 616
 617 int bch2_btree_insert_nonextent(struct btree_trans *trans,
 618                                 enum btree_id btree, struct bkey_i *k,
 619                                 enum btree_update_flags flags)
 620 {
 621         struct btree_iter iter;
 622         int ret;
 623
 624         bch2_trans_iter_init(trans, &iter, btree, k->k.p,
 625                              BTREE_ITER_CACHED|
 626                              BTREE_ITER_NOT_EXTENTS|
 627                              BTREE_ITER_INTENT);
 628         ret   = bch2_btree_iter_traverse(&iter) ?:
 629                 bch2_trans_update(trans, &iter, k, flags);
 630         bch2_trans_iter_exit(trans, &iter);
 631         return ret;
 632 }
 633
 634 int bch2_btree_insert_trans(struct btree_trans *trans, enum btree_id id,
 635                             struct bkey_i *k, enum btree_update_flags flags)
 636 {
 637         struct btree_iter iter;
 638         int ret;
 639
 640         bch2_trans_iter_init(trans, &iter, id, bkey_start_pos(&k->k),
 641                              BTREE_ITER_CACHED|
 642                              BTREE_ITER_INTENT);
 643         ret   = bch2_btree_iter_traverse(&iter) ?:
 644                 bch2_trans_update(trans, &iter, k, flags);
 645         bch2_trans_iter_exit(trans, &iter);
 646         return ret;
 647 }
 648
 649 /**
 650  * bch2_btree_insert - insert keys into the extent btree
 651  * @c:                  pointer to struct bch_fs
 652  * @id:                 btree to insert into
 653  * @k:                  key to insert
 654  * @disk_res:           must be non-NULL whenever inserting or potentially
 655  *                      splitting data extents
 656  * @flags:              transaction commit flags
 657  *
 658  * Returns:             0 on success, error code on failure
 659  */
 660 int bch2_btree_insert(struct bch_fs *c, enum btree_id id, struct bkey_i *k,
 661                       struct disk_reservation *disk_res, int flags)
 662 {
 663         return bch2_trans_do(c, disk_res, NULL, flags,
 664                              bch2_btree_insert_trans(trans, id, k, 0));
 665 }
 666
 667 int bch2_btree_delete_extent_at(struct btree_trans *trans, struct btree_iter *iter,
 668                                 unsigned len, unsigned update_flags)
 669 {
 670         struct bkey_i *k;
 671
 672         k = bch2_trans_kmalloc(trans, sizeof(*k));
 673         if (IS_ERR(k))
 674                 return PTR_ERR(k);
 675
 676         bkey_init(&k->k);
 677         k->k.p = iter->pos;
 678         bch2_key_resize(&k->k, len);
 679         return bch2_trans_update(trans, iter, k, update_flags);
 680 }
 681
 682 int bch2_btree_delete_at(struct btree_trans *trans,
 683                          struct btree_iter *iter, unsigned update_flags)
 684 {
 685         return bch2_btree_delete_extent_at(trans, iter, 0, update_flags);
 686 }
 687
 688 int bch2_btree_delete(struct btree_trans *trans,
 689                       enum btree_id btree, struct bpos pos,
 690                       unsigned update_flags)
 691 {
 692         struct btree_iter iter;
 693         int ret;
 694
 695         bch2_trans_iter_init(trans, &iter, btree, pos,
 696                              BTREE_ITER_CACHED|
 697                              BTREE_ITER_INTENT);
 698         ret   = bch2_btree_iter_traverse(&iter) ?:
 699                 bch2_btree_delete_at(trans, &iter, update_flags);
 700         bch2_trans_iter_exit(trans, &iter);
 701
 702         return ret;
 703 }
 704
 705 int bch2_btree_delete_range_trans(struct btree_trans *trans, enum btree_id id,
 706                                   struct bpos start, struct bpos end,
 707                                   unsigned update_flags,
 708                                   u64 *journal_seq)
 709 {
 710         u32 restart_count = trans->restart_count;
 711         struct btree_iter iter;
 712         struct bkey_s_c k;
 713         int ret = 0;
 714
 715         bch2_trans_iter_init(trans, &iter, id, start, BTREE_ITER_INTENT);
 716         while ((k = bch2_btree_iter_peek_upto(&iter, end)).k) {
 717                 struct disk_reservation disk_res =
 718                         bch2_disk_reservation_init(trans->c, 0);
 719                 struct bkey_i delete;
 720
 721                 ret = bkey_err(k);
 722                 if (ret)
 723                         goto err;
 724
 725                 bkey_init(&delete.k);
 726
 727                 /*
 728                  * This could probably be more efficient for extents:
 729                  */
 730
 731                 /*
 732                  * For extents, iter.pos won't necessarily be the same as
 733                  * bkey_start_pos(k.k) (for non extents they always will be the
 734                  * same). It's important that we delete starting from iter.pos
 735                  * because the range we want to delete could start in the middle
 736                  * of k.
 737                  *
 738                  * (bch2_btree_iter_peek() does guarantee that iter.pos >=
 739                  * bkey_start_pos(k.k)).
 740                  */
 741                 delete.k.p = iter.pos;
 742
 743                 if (iter.flags & BTREE_ITER_IS_EXTENTS)
 744                         bch2_key_resize(&delete.k,
 745                                         bpos_min(end, k.k->p).offset -
 746                                         iter.pos.offset);
 747
 748                 ret   = bch2_trans_update(trans, &iter, &delete, update_flags) ?:
 749                         bch2_trans_commit(trans, &disk_res, journal_seq,
 750                                           BCH_TRANS_COMMIT_no_enospc);
 751                 bch2_disk_reservation_put(trans->c, &disk_res);
 752 err:
 753                 /*
 754                  * the bch2_trans_begin() call is in a weird place because we
 755                  * need to call it after every transaction commit, to avoid path
 756                  * overflow, but don't want to call it if the delete operation
 757                  * is a no-op and we have no work to do:
 758                  */
 759                 bch2_trans_begin(trans);
 760
 761                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 762                         ret = 0;
 763                 if (ret)
 764                         break;
 765         }
 766         bch2_trans_iter_exit(trans, &iter);
 767
 768         return ret ?: trans_was_restarted(trans, restart_count);
 769 }
 770
 771 /*
 772  * bch_btree_delete_range - delete everything within a given range
 773  *
 774  * Range is a half open interval - [start, end)
 775  */
 776 int bch2_btree_delete_range(struct bch_fs *c, enum btree_id id,
 777                             struct bpos start, struct bpos end,
 778                             unsigned update_flags,
 779                             u64 *journal_seq)
 780 {
 781         int ret = bch2_trans_run(c,
 782                         bch2_btree_delete_range_trans(trans, id, start, end,
 783                                                       update_flags, journal_seq));
 784         if (ret == -BCH_ERR_transaction_restart_nested)
 785                 ret = 0;
 786         return ret;
 787 }
 788
 789 int bch2_btree_bit_mod(struct btree_trans *trans, enum btree_id btree,
 790                        struct bpos pos, bool set)
 791 {
 792         struct bkey_i k;
 793
 794         bkey_init(&k.k);
 795         k.k.type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
 796         k.k.p = pos;
 797
 798         return bch2_trans_update_buffered(trans, btree, &k);
 799 }
 800
 801 static int __bch2_trans_log_msg(struct btree_trans *trans, struct printbuf *buf, unsigned u64s)
 802 {
 803         struct jset_entry *e = bch2_trans_jset_entry_alloc(trans, jset_u64s(u64s));
 804         int ret = PTR_ERR_OR_ZERO(e);
 805         if (ret)
 806                 return ret;
 807
 808         struct jset_entry_log *l = container_of(e, struct jset_entry_log, entry);
 809         journal_entry_init(e, BCH_JSET_ENTRY_log, 0, 1, u64s);
 810         memcpy(l->d, buf->buf, buf->pos);
 811         return 0;
 812 }
 813
 814 __printf(3, 0)
 815 static int
 816 __bch2_fs_log_msg(struct bch_fs *c, unsigned commit_flags, const char *fmt,
 817                   va_list args)
 818 {
 819         struct printbuf buf = PRINTBUF;
 820         prt_vprintf(&buf, fmt, args);
 821
 822         unsigned u64s = DIV_ROUND_UP(buf.pos, sizeof(u64));
 823         prt_chars(&buf, '\0', u64s * sizeof(u64) - buf.pos);
 824
 825         int ret = buf.allocation_failure ? -BCH_ERR_ENOMEM_trans_log_msg : 0;
 826         if (ret)
 827                 goto err;
 828
 829         if (!test_bit(JOURNAL_STARTED, &c->journal.flags)) {
 830                 ret = darray_make_room(&c->journal.early_journal_entries, jset_u64s(u64s));
 831                 if (ret)
 832                         goto err;
 833
 834                 struct jset_entry_log *l = (void *) &darray_top(c->journal.early_journal_entries);
 835                 journal_entry_init(&l->entry, BCH_JSET_ENTRY_log, 0, 1, u64s);
 836                 memcpy(l->d, buf.buf, buf.pos);
 837                 c->journal.early_journal_entries.nr += jset_u64s(u64s);
 838         } else {
 839                 ret = bch2_trans_do(c, NULL, NULL,
 840                         BCH_TRANS_COMMIT_lazy_rw|commit_flags,
 841                         __bch2_trans_log_msg(trans, &buf, u64s));
 842         }
 843 err:
 844         printbuf_exit(&buf);
 845         return ret;
 846 }
 847
 848 __printf(2, 3)
 849 int bch2_fs_log_msg(struct bch_fs *c, const char *fmt, ...)
 850 {
 851         va_list args;
 852         int ret;
 853
 854         va_start(args, fmt);
 855         ret = __bch2_fs_log_msg(c, 0, fmt, args);
 856         va_end(args);
 857         return ret;
 858 }
 859
 860 /*
 861  * Use for logging messages during recovery to enable reserved space and avoid
 862  * blocking.
 863  */
 864 __printf(2, 3)
 865 int bch2_journal_log_msg(struct bch_fs *c, const char *fmt, ...)
 866 {
 867         va_list args;
 868         int ret;
 869
 870         va_start(args, fmt);
 871         ret = __bch2_fs_log_msg(c, BCH_WATERMARK_reclaim, fmt, args);
 872         va_end(args);
 873         return ret;
 874 }