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