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 *, struct btree_path *,
  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_journal_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_path *path,
 342                                             struct btree_insert_entry *i,
 343                                             enum btree_update_flags flags,
 344                                             unsigned long ip)
 345 {
 346         struct btree_path *btree_path;
 347         struct bkey k;
 348         int ret;
 349
 350         btree_path = bch2_path_get(trans, path->btree_id, path->pos, 1, 0,
 351                                    BTREE_ITER_INTENT, _THIS_IP_);
 352         ret = bch2_btree_path_traverse(trans, btree_path, 0);
 353         if (ret)
 354                 goto out;
 355
 356         /*
 357          * The old key in the insert entry might actually refer to an existing
 358          * key in the btree that has been deleted from cache and not yet
 359          * flushed. Check for this and skip the flush so we don't run triggers
 360          * against a stale key.
 361          */
 362         bch2_btree_path_peek_slot_exact(btree_path, &k);
 363         if (!bkey_deleted(&k))
 364                 goto out;
 365
 366         i->key_cache_already_flushed = true;
 367         i->flags |= BTREE_TRIGGER_NORUN;
 368
 369         btree_path_set_should_be_locked(btree_path);
 370         ret = bch2_trans_update_by_path(trans, btree_path, i->k, flags, ip);
 371 out:
 372         bch2_path_put(trans, btree_path, true);
 373         return ret;
 374 }
 375
 376 static int __must_check
 377 bch2_trans_update_by_path(struct btree_trans *trans, struct btree_path *path,
 378                           struct bkey_i *k, enum btree_update_flags flags,
 379                           unsigned long ip)
 380 {
 381         struct bch_fs *c = trans->c;
 382         struct btree_insert_entry *i, n;
 383         u64 seq = 0;
 384         int cmp;
 385
 386         EBUG_ON(!path->should_be_locked);
 387         EBUG_ON(trans->nr_updates >= BTREE_ITER_MAX);
 388         EBUG_ON(!bpos_eq(k->k.p, path->pos));
 389
 390         /*
 391          * The transaction journal res hasn't been allocated at this point.
 392          * That occurs at commit time. Reuse the seq field to pass in the seq
 393          * of a prejournaled key.
 394          */
 395         if (flags & BTREE_UPDATE_PREJOURNAL)
 396                 seq = trans->journal_res.seq;
 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,
 405                 .k              = k,
 406                 .seq            = seq,
 407                 .ip_allocated   = ip,
 408         };
 409
 410 #ifdef CONFIG_BCACHEFS_DEBUG
 411         trans_for_each_update(trans, i)
 412                 BUG_ON(i != trans->updates &&
 413                        btree_insert_entry_cmp(i - 1, i) >= 0);
 414 #endif
 415
 416         /*
 417          * Pending updates are kept sorted: first, find position of new update,
 418          * then delete/trim any updates the new update overwrites:
 419          */
 420         trans_for_each_update(trans, i) {
 421                 cmp = btree_insert_entry_cmp(&n, i);
 422                 if (cmp <= 0)
 423                         break;
 424         }
 425
 426         if (!cmp && i < trans->updates + trans->nr_updates) {
 427                 EBUG_ON(i->insert_trigger_run || i->overwrite_trigger_run);
 428
 429                 bch2_path_put(trans, i->path, true);
 430                 i->flags        = n.flags;
 431                 i->cached       = n.cached;
 432                 i->k            = n.k;
 433                 i->path         = n.path;
 434                 i->seq          = n.seq;
 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(i->path, true);
 455
 456         /*
 457          * If a key is present in the key cache, it must also exist in the
 458          * btree - this is necessary for cache coherency. When iterating over
 459          * a btree that's cached in the key cache, the btree iter code checks
 460          * the key cache - but the key has to exist in the btree for that to
 461          * work:
 462          */
 463         if (path->cached && bkey_deleted(&i->old_k))
 464                 return flush_new_cached_update(trans, path, i, flags, ip);
 465
 466         return 0;
 467 }
 468
 469 static noinline int bch2_trans_update_get_key_cache(struct btree_trans *trans,
 470                                                     struct btree_iter *iter,
 471                                                     struct btree_path *path)
 472 {
 473         if (!iter->key_cache_path ||
 474             !iter->key_cache_path->should_be_locked ||
 475             !bpos_eq(iter->key_cache_path->pos, iter->pos)) {
 476                 struct bkey_cached *ck;
 477                 int ret;
 478
 479                 if (!iter->key_cache_path)
 480                         iter->key_cache_path =
 481                                 bch2_path_get(trans, path->btree_id, path->pos, 1, 0,
 482                                               BTREE_ITER_INTENT|
 483                                               BTREE_ITER_CACHED, _THIS_IP_);
 484
 485                 iter->key_cache_path =
 486                         bch2_btree_path_set_pos(trans, iter->key_cache_path, path->pos,
 487                                                 iter->flags & BTREE_ITER_INTENT,
 488                                                 _THIS_IP_);
 489
 490                 ret = bch2_btree_path_traverse(trans, iter->key_cache_path,
 491                                                BTREE_ITER_CACHED);
 492                 if (unlikely(ret))
 493                         return ret;
 494
 495                 ck = (void *) iter->key_cache_path->l[0].b;
 496
 497                 if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 498                         trace_and_count(trans->c, trans_restart_key_cache_raced, trans, _RET_IP_);
 499                         return btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_raced);
 500                 }
 501
 502                 btree_path_set_should_be_locked(iter->key_cache_path);
 503         }
 504
 505         return 0;
 506 }
 507
 508 int __must_check bch2_trans_update(struct btree_trans *trans, struct btree_iter *iter,
 509                                    struct bkey_i *k, enum btree_update_flags flags)
 510 {
 511         struct btree_path *path = iter->update_path ?: iter->path;
 512         int ret;
 513
 514         if (iter->flags & BTREE_ITER_IS_EXTENTS)
 515                 return bch2_trans_update_extent(trans, iter, k, flags);
 516
 517         if (bkey_deleted(&k->k) &&
 518             !(flags & BTREE_UPDATE_KEY_CACHE_RECLAIM) &&
 519             (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)) {
 520                 ret = need_whiteout_for_snapshot(trans, iter->btree_id, k->k.p);
 521                 if (unlikely(ret < 0))
 522                         return ret;
 523
 524                 if (ret)
 525                         k->k.type = KEY_TYPE_whiteout;
 526         }
 527
 528         /*
 529          * Ensure that updates to cached btrees go to the key cache:
 530          */
 531         if (!(flags & BTREE_UPDATE_KEY_CACHE_RECLAIM) &&
 532             !path->cached &&
 533             !path->level &&
 534             btree_id_cached(trans->c, path->btree_id)) {
 535                 ret = bch2_trans_update_get_key_cache(trans, iter, path);
 536                 if (ret)
 537                         return ret;
 538
 539                 path = iter->key_cache_path;
 540         }
 541
 542         return bch2_trans_update_by_path(trans, path, k, flags, _RET_IP_);
 543 }
 544
 545 /*
 546  * Add a transaction update for a key that has already been journaled.
 547  */
 548 int __must_check bch2_trans_update_seq(struct btree_trans *trans, u64 seq,
 549                                        struct btree_iter *iter, struct bkey_i *k,
 550                                        enum btree_update_flags flags)
 551 {
 552         trans->journal_res.seq = seq;
 553         return bch2_trans_update(trans, iter, k, flags|BTREE_UPDATE_NOJOURNAL|
 554                                                  BTREE_UPDATE_PREJOURNAL);
 555 }
 556
 557 int __must_check bch2_trans_update_buffered(struct btree_trans *trans,
 558                                             enum btree_id btree,
 559                                             struct bkey_i *k)
 560 {
 561         struct btree_write_buffered_key *i;
 562         int ret;
 563
 564         EBUG_ON(trans->nr_wb_updates > trans->wb_updates_size);
 565         EBUG_ON(k->k.u64s > BTREE_WRITE_BUFERED_U64s_MAX);
 566
 567         trans_for_each_wb_update(trans, i) {
 568                 if (i->btree == btree && bpos_eq(i->k.k.p, k->k.p)) {
 569                         bkey_copy(&i->k, k);
 570                         return 0;
 571                 }
 572         }
 573
 574         if (!trans->wb_updates ||
 575             trans->nr_wb_updates == trans->wb_updates_size) {
 576                 struct btree_write_buffered_key *u;
 577
 578                 if (trans->nr_wb_updates == trans->wb_updates_size) {
 579                         struct btree_transaction_stats *s = btree_trans_stats(trans);
 580
 581                         BUG_ON(trans->wb_updates_size > U8_MAX / 2);
 582                         trans->wb_updates_size = max(1, trans->wb_updates_size * 2);
 583                         if (s)
 584                                 s->wb_updates_size = trans->wb_updates_size;
 585                 }
 586
 587                 u = bch2_trans_kmalloc_nomemzero(trans,
 588                                         trans->wb_updates_size *
 589                                         sizeof(struct btree_write_buffered_key));
 590                 ret = PTR_ERR_OR_ZERO(u);
 591                 if (ret)
 592                         return ret;
 593
 594                 if (trans->nr_wb_updates)
 595                         memcpy(u, trans->wb_updates, trans->nr_wb_updates *
 596                                sizeof(struct btree_write_buffered_key));
 597                 trans->wb_updates = u;
 598         }
 599
 600         trans->wb_updates[trans->nr_wb_updates] = (struct btree_write_buffered_key) {
 601                 .btree  = btree,
 602         };
 603
 604         bkey_copy(&trans->wb_updates[trans->nr_wb_updates].k, k);
 605         trans->nr_wb_updates++;
 606
 607         return 0;
 608 }
 609
 610 int bch2_bkey_get_empty_slot(struct btree_trans *trans, struct btree_iter *iter,
 611                              enum btree_id btree, struct bpos end)
 612 {
 613         struct bkey_s_c k;
 614         int ret = 0;
 615
 616         bch2_trans_iter_init(trans, iter, btree, POS_MAX, BTREE_ITER_INTENT);
 617         k = bch2_btree_iter_prev(iter);
 618         ret = bkey_err(k);
 619         if (ret)
 620                 goto err;
 621
 622         bch2_btree_iter_advance(iter);
 623         k = bch2_btree_iter_peek_slot(iter);
 624         ret = bkey_err(k);
 625         if (ret)
 626                 goto err;
 627
 628         BUG_ON(k.k->type != KEY_TYPE_deleted);
 629
 630         if (bkey_gt(k.k->p, end)) {
 631                 ret = -BCH_ERR_ENOSPC_btree_slot;
 632                 goto err;
 633         }
 634
 635         return 0;
 636 err:
 637         bch2_trans_iter_exit(trans, iter);
 638         return ret;
 639 }
 640
 641 void bch2_trans_commit_hook(struct btree_trans *trans,
 642                             struct btree_trans_commit_hook *h)
 643 {
 644         h->next = trans->hooks;
 645         trans->hooks = h;
 646 }
 647
 648 int bch2_btree_insert_nonextent(struct btree_trans *trans,
 649                                 enum btree_id btree, struct bkey_i *k,
 650                                 enum btree_update_flags flags)
 651 {
 652         struct btree_iter iter;
 653         int ret;
 654
 655         bch2_trans_iter_init(trans, &iter, btree, k->k.p,
 656                              BTREE_ITER_CACHED|
 657                              BTREE_ITER_NOT_EXTENTS|
 658                              BTREE_ITER_INTENT);
 659         ret   = bch2_btree_iter_traverse(&iter) ?:
 660                 bch2_trans_update(trans, &iter, k, flags);
 661         bch2_trans_iter_exit(trans, &iter);
 662         return ret;
 663 }
 664
 665 int bch2_btree_insert_trans(struct btree_trans *trans, enum btree_id id,
 666                             struct bkey_i *k, enum btree_update_flags flags)
 667 {
 668         struct btree_iter iter;
 669         int ret;
 670
 671         bch2_trans_iter_init(trans, &iter, id, bkey_start_pos(&k->k),
 672                              BTREE_ITER_CACHED|
 673                              BTREE_ITER_INTENT);
 674         ret   = bch2_btree_iter_traverse(&iter) ?:
 675                 bch2_trans_update(trans, &iter, k, flags);
 676         bch2_trans_iter_exit(trans, &iter);
 677         return ret;
 678 }
 679
 680 /**
 681  * bch2_btree_insert - insert keys into the extent btree
 682  * @c:                  pointer to struct bch_fs
 683  * @id:                 btree to insert into
 684  * @k:                  key to insert
 685  * @disk_res:           must be non-NULL whenever inserting or potentially
 686  *                      splitting data extents
 687  * @flags:              transaction commit flags
 688  *
 689  * Returns:             0 on success, error code on failure
 690  */
 691 int bch2_btree_insert(struct bch_fs *c, enum btree_id id, struct bkey_i *k,
 692                       struct disk_reservation *disk_res, int flags)
 693 {
 694         return bch2_trans_do(c, disk_res, NULL, flags,
 695                              bch2_btree_insert_trans(trans, id, k, 0));
 696 }
 697
 698 int bch2_btree_delete_extent_at(struct btree_trans *trans, struct btree_iter *iter,
 699                                 unsigned len, unsigned update_flags)
 700 {
 701         struct bkey_i *k;
 702
 703         k = bch2_trans_kmalloc(trans, sizeof(*k));
 704         if (IS_ERR(k))
 705                 return PTR_ERR(k);
 706
 707         bkey_init(&k->k);
 708         k->k.p = iter->pos;
 709         bch2_key_resize(&k->k, len);
 710         return bch2_trans_update(trans, iter, k, update_flags);
 711 }
 712
 713 int bch2_btree_delete_at(struct btree_trans *trans,
 714                          struct btree_iter *iter, unsigned update_flags)
 715 {
 716         return bch2_btree_delete_extent_at(trans, iter, 0, update_flags);
 717 }
 718
 719 int bch2_btree_delete_at_buffered(struct btree_trans *trans,
 720                                   enum btree_id btree, struct bpos pos)
 721 {
 722         struct bkey_i *k;
 723
 724         k = bch2_trans_kmalloc(trans, sizeof(*k));
 725         if (IS_ERR(k))
 726                 return PTR_ERR(k);
 727
 728         bkey_init(&k->k);
 729         k->k.p = pos;
 730         return bch2_trans_update_buffered(trans, btree, k);
 731 }
 732
 733 int bch2_btree_delete(struct btree_trans *trans,
 734                       enum btree_id btree, struct bpos pos,
 735                       unsigned update_flags)
 736 {
 737         struct btree_iter iter;
 738         int ret;
 739
 740         bch2_trans_iter_init(trans, &iter, btree, pos,
 741                              BTREE_ITER_CACHED|
 742                              BTREE_ITER_INTENT);
 743         ret   = bch2_btree_iter_traverse(&iter) ?:
 744                 bch2_btree_delete_at(trans, &iter, update_flags);
 745         bch2_trans_iter_exit(trans, &iter);
 746
 747         return ret;
 748 }
 749
 750 int bch2_btree_delete_range_trans(struct btree_trans *trans, enum btree_id id,
 751                                   struct bpos start, struct bpos end,
 752                                   unsigned update_flags,
 753                                   u64 *journal_seq)
 754 {
 755         u32 restart_count = trans->restart_count;
 756         struct btree_iter iter;
 757         struct bkey_s_c k;
 758         int ret = 0;
 759
 760         bch2_trans_iter_init(trans, &iter, id, start, BTREE_ITER_INTENT);
 761         while ((k = bch2_btree_iter_peek_upto(&iter, end)).k) {
 762                 struct disk_reservation disk_res =
 763                         bch2_disk_reservation_init(trans->c, 0);
 764                 struct bkey_i delete;
 765
 766                 ret = bkey_err(k);
 767                 if (ret)
 768                         goto err;
 769
 770                 bkey_init(&delete.k);
 771
 772                 /*
 773                  * This could probably be more efficient for extents:
 774                  */
 775
 776                 /*
 777                  * For extents, iter.pos won't necessarily be the same as
 778                  * bkey_start_pos(k.k) (for non extents they always will be the
 779                  * same). It's important that we delete starting from iter.pos
 780                  * because the range we want to delete could start in the middle
 781                  * of k.
 782                  *
 783                  * (bch2_btree_iter_peek() does guarantee that iter.pos >=
 784                  * bkey_start_pos(k.k)).
 785                  */
 786                 delete.k.p = iter.pos;
 787
 788                 if (iter.flags & BTREE_ITER_IS_EXTENTS)
 789                         bch2_key_resize(&delete.k,
 790                                         bpos_min(end, k.k->p).offset -
 791                                         iter.pos.offset);
 792
 793                 ret   = bch2_trans_update(trans, &iter, &delete, update_flags) ?:
 794                         bch2_trans_commit(trans, &disk_res, journal_seq,
 795                                           BTREE_INSERT_NOFAIL);
 796                 bch2_disk_reservation_put(trans->c, &disk_res);
 797 err:
 798                 /*
 799                  * the bch2_trans_begin() call is in a weird place because we
 800                  * need to call it after every transaction commit, to avoid path
 801                  * overflow, but don't want to call it if the delete operation
 802                  * is a no-op and we have no work to do:
 803                  */
 804                 bch2_trans_begin(trans);
 805
 806                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 807                         ret = 0;
 808                 if (ret)
 809                         break;
 810         }
 811         bch2_trans_iter_exit(trans, &iter);
 812
 813         return ret ?: trans_was_restarted(trans, restart_count);
 814 }
 815
 816 /*
 817  * bch_btree_delete_range - delete everything within a given range
 818  *
 819  * Range is a half open interval - [start, end)
 820  */
 821 int bch2_btree_delete_range(struct bch_fs *c, enum btree_id id,
 822                             struct bpos start, struct bpos end,
 823                             unsigned update_flags,
 824                             u64 *journal_seq)
 825 {
 826         int ret = bch2_trans_run(c,
 827                         bch2_btree_delete_range_trans(trans, id, start, end,
 828                                                       update_flags, journal_seq));
 829         if (ret == -BCH_ERR_transaction_restart_nested)
 830                 ret = 0;
 831         return ret;
 832 }
 833
 834 int bch2_btree_bit_mod(struct btree_trans *trans, enum btree_id btree,
 835                        struct bpos pos, bool set)
 836 {
 837         struct bkey_i *k;
 838         int ret = 0;
 839
 840         k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
 841         ret = PTR_ERR_OR_ZERO(k);
 842         if (unlikely(ret))
 843                 return ret;
 844
 845         bkey_init(&k->k);
 846         k->k.type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
 847         k->k.p = pos;
 848
 849         return bch2_trans_update_buffered(trans, btree, k);
 850 }
 851
 852 __printf(2, 0)
 853 static int __bch2_trans_log_msg(darray_u64 *entries, const char *fmt, va_list args)
 854 {
 855         struct printbuf buf = PRINTBUF;
 856         struct jset_entry_log *l;
 857         unsigned u64s;
 858         int ret;
 859
 860         prt_vprintf(&buf, fmt, args);
 861         ret = buf.allocation_failure ? -BCH_ERR_ENOMEM_trans_log_msg : 0;
 862         if (ret)
 863                 goto err;
 864
 865         u64s = DIV_ROUND_UP(buf.pos, sizeof(u64));
 866
 867         ret = darray_make_room(entries, jset_u64s(u64s));
 868         if (ret)
 869                 goto err;
 870
 871         l = (void *) &darray_top(*entries);
 872         l->entry.u64s           = cpu_to_le16(u64s);
 873         l->entry.btree_id       = 0;
 874         l->entry.level          = 1;
 875         l->entry.type           = BCH_JSET_ENTRY_log;
 876         l->entry.pad[0]         = 0;
 877         l->entry.pad[1]         = 0;
 878         l->entry.pad[2]         = 0;
 879         memcpy(l->d, buf.buf, buf.pos);
 880         while (buf.pos & 7)
 881                 l->d[buf.pos++] = '\0';
 882
 883         entries->nr += jset_u64s(u64s);
 884 err:
 885         printbuf_exit(&buf);
 886         return ret;
 887 }
 888
 889 __printf(3, 0)
 890 static int
 891 __bch2_fs_log_msg(struct bch_fs *c, unsigned commit_flags, const char *fmt,
 892                   va_list args)
 893 {
 894         int ret;
 895
 896         if (!test_bit(JOURNAL_STARTED, &c->journal.flags)) {
 897                 ret = __bch2_trans_log_msg(&c->journal.early_journal_entries, fmt, args);
 898         } else {
 899                 ret = bch2_trans_do(c, NULL, NULL,
 900                         BTREE_INSERT_LAZY_RW|commit_flags,
 901                         __bch2_trans_log_msg(&trans->extra_journal_entries, fmt, args));
 902         }
 903
 904         return ret;
 905 }
 906
 907 __printf(2, 3)
 908 int bch2_fs_log_msg(struct bch_fs *c, const char *fmt, ...)
 909 {
 910         va_list args;
 911         int ret;
 912
 913         va_start(args, fmt);
 914         ret = __bch2_fs_log_msg(c, 0, fmt, args);
 915         va_end(args);
 916         return ret;
 917 }
 918
 919 /*
 920  * Use for logging messages during recovery to enable reserved space and avoid
 921  * blocking.
 922  */
 923 __printf(2, 3)
 924 int bch2_journal_log_msg(struct bch_fs *c, const char *fmt, ...)
 925 {
 926         va_list args;
 927         int ret;
 928
 929         va_start(args, fmt);
 930         ret = __bch2_fs_log_msg(c, BCH_WATERMARK_reclaim, fmt, args);
 931         va_end(args);
 932         return ret;
 933 }