git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/btree_update_leaf.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "btree_update.h"
   5 #include "btree_update_interior.h"
   6 #include "btree_gc.h"
   7 #include "btree_io.h"
   8 #include "btree_iter.h"
   9 #include "btree_key_cache.h"
  10 #include "btree_locking.h"
  11 #include "btree_write_buffer.h"
  12 #include "buckets.h"
  13 #include "debug.h"
  14 #include "errcode.h"
  15 #include "error.h"
  16 #include "extent_update.h"
  17 #include "journal.h"
  18 #include "journal_reclaim.h"
  19 #include "keylist.h"
  20 #include "recovery.h"
  21 #include "subvolume.h"
  22 #include "replicas.h"
  23
  24 #include <linux/prefetch.h>
  25 #include <linux/sort.h>
  26 #include <trace/events/bcachefs.h>
  27
  28 /*
  29  * bch2_btree_path_peek_slot() for a cached iterator might return a key in a
  30  * different snapshot:
  31  */
  32 struct bkey_s_c bch2_btree_path_peek_slot_exact(struct btree_path *path, struct bkey *u)
  33 {
  34         struct bkey_s_c k = bch2_btree_path_peek_slot(path, u);
  35
  36         if (k.k && bpos_eq(path->pos, k.k->p))
  37                 return k;
  38
  39         bkey_init(u);
  40         u->p = path->pos;
  41         return (struct bkey_s_c) { u, NULL };
  42 }
  43
  44 static void verify_update_old_key(struct btree_trans *trans, struct btree_insert_entry *i)
  45 {
  46 #ifdef CONFIG_BCACHEFS_DEBUG
  47         struct bch_fs *c = trans->c;
  48         struct bkey u;
  49         struct bkey_s_c k = bch2_btree_path_peek_slot_exact(i->path, &u);
  50
  51         if (unlikely(trans->journal_replay_not_finished)) {
  52                 struct bkey_i *j_k =
  53                         bch2_journal_keys_peek_slot(c, i->btree_id, i->level, i->k->k.p);
  54
  55                 if (j_k)
  56                         k = bkey_i_to_s_c(j_k);
  57         }
  58
  59         u = *k.k;
  60         u.needs_whiteout = i->old_k.needs_whiteout;
  61
  62         BUG_ON(memcmp(&i->old_k, &u, sizeof(struct bkey)));
  63         BUG_ON(i->old_v != k.v);
  64 #endif
  65 }
  66
  67 static int __must_check
  68 bch2_trans_update_by_path(struct btree_trans *, struct btree_path *,
  69                           struct bkey_i *, enum btree_update_flags);
  70
  71 static inline int btree_insert_entry_cmp(const struct btree_insert_entry *l,
  72                                          const struct btree_insert_entry *r)
  73 {
  74         return   cmp_int(l->btree_id,   r->btree_id) ?:
  75                  cmp_int(l->cached,     r->cached) ?:
  76                  -cmp_int(l->level,     r->level) ?:
  77                  bpos_cmp(l->k->k.p,    r->k->k.p);
  78 }
  79
  80 static inline struct btree_path_level *insert_l(struct btree_insert_entry *i)
  81 {
  82         return i->path->l + i->level;
  83 }
  84
  85 static inline bool same_leaf_as_prev(struct btree_trans *trans,
  86                                      struct btree_insert_entry *i)
  87 {
  88         return i != trans->updates &&
  89                 insert_l(&i[0])->b == insert_l(&i[-1])->b;
  90 }
  91
  92 static inline bool same_leaf_as_next(struct btree_trans *trans,
  93                                      struct btree_insert_entry *i)
  94 {
  95         return i + 1 < trans->updates + trans->nr_updates &&
  96                 insert_l(&i[0])->b == insert_l(&i[1])->b;
  97 }
  98
  99 inline void bch2_btree_node_prep_for_write(struct btree_trans *trans,
 100                                            struct btree_path *path,
 101                                            struct btree *b)
 102 {
 103         struct bch_fs *c = trans->c;
 104
 105         if (unlikely(btree_node_just_written(b)) &&
 106             bch2_btree_post_write_cleanup(c, b))
 107                 bch2_trans_node_reinit_iter(trans, b);
 108
 109         /*
 110          * If the last bset has been written, or if it's gotten too big - start
 111          * a new bset to insert into:
 112          */
 113         if (want_new_bset(c, b))
 114                 bch2_btree_init_next(trans, b);
 115 }
 116
 117 /* Inserting into a given leaf node (last stage of insert): */
 118
 119 /* Handle overwrites and do insert, for non extents: */
 120 bool bch2_btree_bset_insert_key(struct btree_trans *trans,
 121                                 struct btree_path *path,
 122                                 struct btree *b,
 123                                 struct btree_node_iter *node_iter,
 124                                 struct bkey_i *insert)
 125 {
 126         struct bkey_packed *k;
 127         unsigned clobber_u64s = 0, new_u64s = 0;
 128
 129         EBUG_ON(btree_node_just_written(b));
 130         EBUG_ON(bset_written(b, btree_bset_last(b)));
 131         EBUG_ON(bkey_deleted(&insert->k) && bkey_val_u64s(&insert->k));
 132         EBUG_ON(bpos_lt(insert->k.p, b->data->min_key));
 133         EBUG_ON(bpos_gt(insert->k.p, b->data->max_key));
 134         EBUG_ON(insert->k.u64s >
 135                 bch_btree_keys_u64s_remaining(trans->c, b));
 136
 137         k = bch2_btree_node_iter_peek_all(node_iter, b);
 138         if (k && bkey_cmp_left_packed(b, k, &insert->k.p))
 139                 k = NULL;
 140
 141         /* @k is the key being overwritten/deleted, if any: */
 142         EBUG_ON(k && bkey_deleted(k));
 143
 144         /* Deleting, but not found? nothing to do: */
 145         if (bkey_deleted(&insert->k) && !k)
 146                 return false;
 147
 148         if (bkey_deleted(&insert->k)) {
 149                 /* Deleting: */
 150                 btree_account_key_drop(b, k);
 151                 k->type = KEY_TYPE_deleted;
 152
 153                 if (k->needs_whiteout)
 154                         push_whiteout(trans->c, b, insert->k.p);
 155                 k->needs_whiteout = false;
 156
 157                 if (k >= btree_bset_last(b)->start) {
 158                         clobber_u64s = k->u64s;
 159                         bch2_bset_delete(b, k, clobber_u64s);
 160                         goto fix_iter;
 161                 } else {
 162                         bch2_btree_path_fix_key_modified(trans, b, k);
 163                 }
 164
 165                 return true;
 166         }
 167
 168         if (k) {
 169                 /* Overwriting: */
 170                 btree_account_key_drop(b, k);
 171                 k->type = KEY_TYPE_deleted;
 172
 173                 insert->k.needs_whiteout = k->needs_whiteout;
 174                 k->needs_whiteout = false;
 175
 176                 if (k >= btree_bset_last(b)->start) {
 177                         clobber_u64s = k->u64s;
 178                         goto overwrite;
 179                 } else {
 180                         bch2_btree_path_fix_key_modified(trans, b, k);
 181                 }
 182         }
 183
 184         k = bch2_btree_node_iter_bset_pos(node_iter, b, bset_tree_last(b));
 185 overwrite:
 186         bch2_bset_insert(b, node_iter, k, insert, clobber_u64s);
 187         new_u64s = k->u64s;
 188 fix_iter:
 189         if (clobber_u64s != new_u64s)
 190                 bch2_btree_node_iter_fix(trans, path, b, node_iter, k,
 191                                          clobber_u64s, new_u64s);
 192         return true;
 193 }
 194
 195 static int __btree_node_flush(struct journal *j, struct journal_entry_pin *pin,
 196                                unsigned i, u64 seq)
 197 {
 198         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 199         struct btree_write *w = container_of(pin, struct btree_write, journal);
 200         struct btree *b = container_of(w, struct btree, writes[i]);
 201         struct btree_trans trans;
 202         unsigned long old, new, v;
 203         unsigned idx = w - b->writes;
 204
 205         bch2_trans_init(&trans, c, 0, 0);
 206
 207         btree_node_lock_nopath_nofail(&trans, &b->c, SIX_LOCK_read);
 208         v = READ_ONCE(b->flags);
 209
 210         do {
 211                 old = new = v;
 212
 213                 if (!(old & (1 << BTREE_NODE_dirty)) ||
 214                     !!(old & (1 << BTREE_NODE_write_idx)) != idx ||
 215                     w->journal.seq != seq)
 216                         break;
 217
 218                 new &= ~BTREE_WRITE_TYPE_MASK;
 219                 new |= BTREE_WRITE_journal_reclaim;
 220                 new |= 1 << BTREE_NODE_need_write;
 221         } while ((v = cmpxchg(&b->flags, old, new)) != old);
 222
 223         btree_node_write_if_need(c, b, SIX_LOCK_read);
 224         six_unlock_read(&b->c.lock);
 225
 226         bch2_trans_exit(&trans);
 227         return 0;
 228 }
 229
 230 static int btree_node_flush0(struct journal *j, struct journal_entry_pin *pin, u64 seq)
 231 {
 232         return __btree_node_flush(j, pin, 0, seq);
 233 }
 234
 235 static int btree_node_flush1(struct journal *j, struct journal_entry_pin *pin, u64 seq)
 236 {
 237         return __btree_node_flush(j, pin, 1, seq);
 238 }
 239
 240 inline void bch2_btree_add_journal_pin(struct bch_fs *c,
 241                                        struct btree *b, u64 seq)
 242 {
 243         struct btree_write *w = btree_current_write(b);
 244
 245         bch2_journal_pin_add(&c->journal, seq, &w->journal,
 246                              btree_node_write_idx(b) == 0
 247                              ? btree_node_flush0
 248                              : btree_node_flush1);
 249 }
 250
 251 /**
 252  * btree_insert_key - insert a key one key into a leaf node
 253  */
 254 inline void bch2_btree_insert_key_leaf(struct btree_trans *trans,
 255                                        struct btree_path *path,
 256                                        struct bkey_i *insert,
 257                                        u64 journal_seq)
 258 {
 259         struct bch_fs *c = trans->c;
 260         struct btree *b = path_l(path)->b;
 261         struct bset_tree *t = bset_tree_last(b);
 262         struct bset *i = bset(b, t);
 263         int old_u64s = bset_u64s(t);
 264         int old_live_u64s = b->nr.live_u64s;
 265         int live_u64s_added, u64s_added;
 266
 267         if (unlikely(!bch2_btree_bset_insert_key(trans, path, b,
 268                                         &path_l(path)->iter, insert)))
 269                 return;
 270
 271         i->journal_seq = cpu_to_le64(max(journal_seq, le64_to_cpu(i->journal_seq)));
 272
 273         bch2_btree_add_journal_pin(c, b, journal_seq);
 274
 275         if (unlikely(!btree_node_dirty(b)))
 276                 set_btree_node_dirty_acct(c, b);
 277
 278         live_u64s_added = (int) b->nr.live_u64s - old_live_u64s;
 279         u64s_added = (int) bset_u64s(t) - old_u64s;
 280
 281         if (b->sib_u64s[0] != U16_MAX && live_u64s_added < 0)
 282                 b->sib_u64s[0] = max(0, (int) b->sib_u64s[0] + live_u64s_added);
 283         if (b->sib_u64s[1] != U16_MAX && live_u64s_added < 0)
 284                 b->sib_u64s[1] = max(0, (int) b->sib_u64s[1] + live_u64s_added);
 285
 286         if (u64s_added > live_u64s_added &&
 287             bch2_maybe_compact_whiteouts(c, b))
 288                 bch2_trans_node_reinit_iter(trans, b);
 289 }
 290
 291 static void btree_insert_key_leaf(struct btree_trans *trans,
 292                                   struct btree_insert_entry *insert)
 293 {
 294         bch2_btree_insert_key_leaf(trans, insert->path, insert->k, trans->journal_res.seq);
 295 }
 296
 297 /* Cached btree updates: */
 298
 299 /* Normal update interface: */
 300
 301 static inline void btree_insert_entry_checks(struct btree_trans *trans,
 302                                              struct btree_insert_entry *i)
 303 {
 304         BUG_ON(!bpos_eq(i->k->k.p, i->path->pos));
 305         BUG_ON(i->cached        != i->path->cached);
 306         BUG_ON(i->level         != i->path->level);
 307         BUG_ON(i->btree_id      != i->path->btree_id);
 308         EBUG_ON(!i->level &&
 309                 !(i->flags & BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) &&
 310                 test_bit(JOURNAL_REPLAY_DONE, &trans->c->journal.flags) &&
 311                 i->k->k.p.snapshot &&
 312                 bch2_snapshot_internal_node(trans->c, i->k->k.p.snapshot));
 313 }
 314
 315 static noinline int
 316 bch2_trans_journal_preres_get_cold(struct btree_trans *trans, unsigned flags,
 317                                    unsigned long trace_ip)
 318 {
 319         struct bch_fs *c = trans->c;
 320         int ret;
 321
 322         bch2_trans_unlock(trans);
 323
 324         ret = bch2_journal_preres_get(&c->journal,
 325                         &trans->journal_preres,
 326                         trans->journal_preres_u64s,
 327                         (flags & JOURNAL_WATERMARK_MASK));
 328         if (ret)
 329                 return ret;
 330
 331         ret = bch2_trans_relock(trans);
 332         if (ret) {
 333                 trace_and_count(c, trans_restart_journal_preres_get, trans, trace_ip, 0);
 334                 return ret;
 335         }
 336
 337         return 0;
 338 }
 339
 340 static __always_inline int bch2_trans_journal_res_get(struct btree_trans *trans,
 341                                                       unsigned flags)
 342 {
 343         return bch2_journal_res_get(&trans->c->journal, &trans->journal_res,
 344                                     trans->journal_u64s, flags);
 345 }
 346
 347 #define JSET_ENTRY_LOG_U64s             4
 348
 349 static noinline void journal_transaction_name(struct btree_trans *trans)
 350 {
 351         struct bch_fs *c = trans->c;
 352         struct journal *j = &c->journal;
 353         struct jset_entry *entry =
 354                 bch2_journal_add_entry(j, &trans->journal_res,
 355                                        BCH_JSET_ENTRY_log, 0, 0,
 356                                        JSET_ENTRY_LOG_U64s);
 357         struct jset_entry_log *l =
 358                 container_of(entry, struct jset_entry_log, entry);
 359
 360         strncpy(l->d, trans->fn, JSET_ENTRY_LOG_U64s * sizeof(u64));
 361 }
 362
 363 static inline int btree_key_can_insert(struct btree_trans *trans,
 364                                        struct btree *b, unsigned u64s)
 365 {
 366         struct bch_fs *c = trans->c;
 367
 368         if (!bch2_btree_node_insert_fits(c, b, u64s))
 369                 return -BCH_ERR_btree_insert_btree_node_full;
 370
 371         return 0;
 372 }
 373
 374 static int btree_key_can_insert_cached(struct btree_trans *trans, unsigned flags,
 375                                        struct btree_path *path, unsigned u64s)
 376 {
 377         struct bch_fs *c = trans->c;
 378         struct bkey_cached *ck = (void *) path->l[0].b;
 379         struct btree_insert_entry *i;
 380         unsigned new_u64s;
 381         struct bkey_i *new_k;
 382
 383         EBUG_ON(path->level);
 384
 385         if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags) &&
 386             bch2_btree_key_cache_must_wait(c) &&
 387             !(flags & BTREE_INSERT_JOURNAL_RECLAIM))
 388                 return -BCH_ERR_btree_insert_need_journal_reclaim;
 389
 390         /*
 391          * bch2_varint_decode can read past the end of the buffer by at most 7
 392          * bytes (it won't be used):
 393          */
 394         u64s += 1;
 395
 396         if (u64s <= ck->u64s)
 397                 return 0;
 398
 399         new_u64s        = roundup_pow_of_two(u64s);
 400         new_k           = krealloc(ck->k, new_u64s * sizeof(u64), GFP_NOFS);
 401         if (!new_k) {
 402                 bch_err(c, "error allocating memory for key cache key, btree %s u64s %u",
 403                         bch2_btree_ids[path->btree_id], new_u64s);
 404                 return -ENOMEM;
 405         }
 406
 407         trans_for_each_update(trans, i)
 408                 if (i->old_v == &ck->k->v)
 409                         i->old_v = &new_k->v;
 410
 411         ck->u64s        = new_u64s;
 412         ck->k           = new_k;
 413         return 0;
 414 }
 415
 416 /* Triggers: */
 417
 418 static int run_one_mem_trigger(struct btree_trans *trans,
 419                                struct btree_insert_entry *i,
 420                                unsigned flags)
 421 {
 422         struct bkey_s_c old = { &i->old_k, i->old_v };
 423         struct bkey_i *new = i->k;
 424         int ret;
 425
 426         verify_update_old_key(trans, i);
 427
 428         if (unlikely(flags & BTREE_TRIGGER_NORUN))
 429                 return 0;
 430
 431         if (!btree_node_type_needs_gc(i->btree_id))
 432                 return 0;
 433
 434         if (bch2_bkey_ops[old.k->type].atomic_trigger ==
 435             bch2_bkey_ops[i->k->k.type].atomic_trigger &&
 436             ((1U << old.k->type) & BTREE_TRIGGER_WANTS_OLD_AND_NEW)) {
 437                 ret   = bch2_mark_key(trans, old, bkey_i_to_s_c(new),
 438                                 BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|flags);
 439         } else {
 440                 struct bkey             _deleted = KEY(0, 0, 0);
 441                 struct bkey_s_c         deleted = (struct bkey_s_c) { &_deleted, NULL };
 442
 443                 _deleted.p = i->path->pos;
 444
 445                 ret   = bch2_mark_key(trans, deleted, bkey_i_to_s_c(new),
 446                                 BTREE_TRIGGER_INSERT|flags) ?:
 447                         bch2_mark_key(trans, old, deleted,
 448                                 BTREE_TRIGGER_OVERWRITE|flags);
 449         }
 450
 451         return ret;
 452 }
 453
 454 static int run_one_trans_trigger(struct btree_trans *trans, struct btree_insert_entry *i,
 455                                  bool overwrite)
 456 {
 457         /*
 458          * Transactional triggers create new btree_insert_entries, so we can't
 459          * pass them a pointer to a btree_insert_entry, that memory is going to
 460          * move:
 461          */
 462         struct bkey old_k = i->old_k;
 463         struct bkey_s_c old = { &old_k, i->old_v };
 464
 465         verify_update_old_key(trans, i);
 466
 467         if ((i->flags & BTREE_TRIGGER_NORUN) ||
 468             !(BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->bkey_type)))
 469                 return 0;
 470
 471         if (!i->insert_trigger_run &&
 472             !i->overwrite_trigger_run &&
 473             bch2_bkey_ops[old.k->type].trans_trigger ==
 474             bch2_bkey_ops[i->k->k.type].trans_trigger &&
 475             ((1U << old.k->type) & BTREE_TRIGGER_WANTS_OLD_AND_NEW)) {
 476                 i->overwrite_trigger_run = true;
 477                 i->insert_trigger_run = true;
 478                 return bch2_trans_mark_key(trans, i->btree_id, i->level, old, i->k,
 479                                            BTREE_TRIGGER_INSERT|
 480                                            BTREE_TRIGGER_OVERWRITE|
 481                                            i->flags) ?: 1;
 482         } else if (overwrite && !i->overwrite_trigger_run) {
 483                 i->overwrite_trigger_run = true;
 484                 return bch2_trans_mark_old(trans, i->btree_id, i->level, old, i->flags) ?: 1;
 485         } else if (!overwrite && !i->insert_trigger_run) {
 486                 i->insert_trigger_run = true;
 487                 return bch2_trans_mark_new(trans, i->btree_id, i->level, i->k, i->flags) ?: 1;
 488         } else {
 489                 return 0;
 490         }
 491 }
 492
 493 static int run_btree_triggers(struct btree_trans *trans, enum btree_id btree_id,
 494                               struct btree_insert_entry *btree_id_start)
 495 {
 496         struct btree_insert_entry *i;
 497         bool trans_trigger_run;
 498         int ret, overwrite;
 499
 500         for (overwrite = 1; overwrite >= 0; --overwrite) {
 501
 502                 /*
 503                  * Running triggers will append more updates to the list of updates as
 504                  * we're walking it:
 505                  */
 506                 do {
 507                         trans_trigger_run = false;
 508
 509                         for (i = btree_id_start;
 510                              i < trans->updates + trans->nr_updates && i->btree_id <= btree_id;
 511                              i++) {
 512                                 if (i->btree_id != btree_id)
 513                                         continue;
 514
 515                                 ret = run_one_trans_trigger(trans, i, overwrite);
 516                                 if (ret < 0)
 517                                         return ret;
 518                                 if (ret)
 519                                         trans_trigger_run = true;
 520                         }
 521                 } while (trans_trigger_run);
 522         }
 523
 524         return 0;
 525 }
 526
 527 static int bch2_trans_commit_run_triggers(struct btree_trans *trans)
 528 {
 529         struct btree_insert_entry *i = NULL, *btree_id_start = trans->updates;
 530         unsigned btree_id = 0;
 531         int ret = 0;
 532
 533         /*
 534          *
 535          * For a given btree, this algorithm runs insert triggers before
 536          * overwrite triggers: this is so that when extents are being moved
 537          * (e.g. by FALLOCATE_FL_INSERT_RANGE), we don't drop references before
 538          * they are re-added.
 539          */
 540         for (btree_id = 0; btree_id < BTREE_ID_NR; btree_id++) {
 541                 if (btree_id == BTREE_ID_alloc)
 542                         continue;
 543
 544                 while (btree_id_start < trans->updates + trans->nr_updates &&
 545                        btree_id_start->btree_id < btree_id)
 546                         btree_id_start++;
 547
 548                 ret = run_btree_triggers(trans, btree_id, btree_id_start);
 549                 if (ret)
 550                         return ret;
 551         }
 552
 553         trans_for_each_update(trans, i) {
 554                 if (i->btree_id > BTREE_ID_alloc)
 555                         break;
 556                 if (i->btree_id == BTREE_ID_alloc) {
 557                         ret = run_btree_triggers(trans, BTREE_ID_alloc, i);
 558                         if (ret)
 559                                 return ret;
 560                         break;
 561                 }
 562         }
 563
 564 #ifdef CONFIG_BCACHEFS_DEBUG
 565         trans_for_each_update(trans, i)
 566                 BUG_ON(!(i->flags & BTREE_TRIGGER_NORUN) &&
 567                        (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->bkey_type)) &&
 568                        (!i->insert_trigger_run || !i->overwrite_trigger_run));
 569 #endif
 570         return 0;
 571 }
 572
 573 static noinline int bch2_trans_commit_run_gc_triggers(struct btree_trans *trans)
 574 {
 575         struct bch_fs *c = trans->c;
 576         struct btree_insert_entry *i;
 577         int ret = 0;
 578
 579         trans_for_each_update(trans, i) {
 580                 /*
 581                  * XXX: synchronization of cached update triggers with gc
 582                  * XXX: synchronization of interior node updates with gc
 583                  */
 584                 BUG_ON(i->cached || i->level);
 585
 586                 if (gc_visited(c, gc_pos_btree_node(insert_l(i)->b))) {
 587                         ret = run_one_mem_trigger(trans, i, i->flags|BTREE_TRIGGER_GC);
 588                         if (ret)
 589                                 break;
 590                 }
 591         }
 592
 593         return ret;
 594 }
 595
 596 static inline int
 597 bch2_trans_commit_write_locked(struct btree_trans *trans, unsigned flags,
 598                                struct btree_insert_entry **stopped_at,
 599                                unsigned long trace_ip)
 600 {
 601         struct bch_fs *c = trans->c;
 602         struct btree_insert_entry *i;
 603         struct btree_write_buffered_key *wb;
 604         struct btree_trans_commit_hook *h;
 605         unsigned u64s = 0;
 606         bool marking = false;
 607         int ret;
 608
 609         if (race_fault()) {
 610                 trace_and_count(c, trans_restart_fault_inject, trans, trace_ip);
 611                 return btree_trans_restart_nounlock(trans, BCH_ERR_transaction_restart_fault_inject);
 612         }
 613
 614         /*
 615          * Check if the insert will fit in the leaf node with the write lock
 616          * held, otherwise another thread could write the node changing the
 617          * amount of space available:
 618          */
 619
 620         prefetch(&trans->c->journal.flags);
 621
 622         h = trans->hooks;
 623         while (h) {
 624                 ret = h->fn(trans, h);
 625                 if (ret)
 626                         return ret;
 627                 h = h->next;
 628         }
 629
 630         trans_for_each_update(trans, i) {
 631                 /* Multiple inserts might go to same leaf: */
 632                 if (!same_leaf_as_prev(trans, i))
 633                         u64s = 0;
 634
 635                 u64s += i->k->k.u64s;
 636                 ret = !i->cached
 637                         ? btree_key_can_insert(trans, insert_l(i)->b, u64s)
 638                         : btree_key_can_insert_cached(trans, flags, i->path, u64s);
 639                 if (ret) {
 640                         *stopped_at = i;
 641                         return ret;
 642                 }
 643
 644                 if (btree_node_type_needs_gc(i->bkey_type))
 645                         marking = true;
 646         }
 647
 648         if (trans->nr_wb_updates &&
 649             trans->nr_wb_updates + c->btree_write_buffer.state.nr > c->btree_write_buffer.size)
 650                 return -BCH_ERR_btree_insert_need_flush_buffer;
 651
 652         /*
 653          * Don't get journal reservation until after we know insert will
 654          * succeed:
 655          */
 656         if (likely(!(flags & BTREE_INSERT_JOURNAL_REPLAY))) {
 657                 ret = bch2_trans_journal_res_get(trans,
 658                                 (flags & JOURNAL_WATERMARK_MASK)|
 659                                 JOURNAL_RES_GET_NONBLOCK);
 660                 if (ret)
 661                         return ret;
 662
 663                 if (unlikely(trans->journal_transaction_names))
 664                         journal_transaction_name(trans);
 665         } else {
 666                 trans->journal_res.seq = c->journal.replay_journal_seq;
 667         }
 668
 669         /*
 670          * Not allowed to fail after we've gotten our journal reservation - we
 671          * have to use it:
 672          */
 673
 674         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
 675             !(flags & BTREE_INSERT_JOURNAL_REPLAY)) {
 676                 if (bch2_journal_seq_verify)
 677                         trans_for_each_update(trans, i)
 678                                 i->k->k.version.lo = trans->journal_res.seq;
 679                 else if (bch2_inject_invalid_keys)
 680                         trans_for_each_update(trans, i)
 681                                 i->k->k.version = MAX_VERSION;
 682         }
 683
 684         if (trans->fs_usage_deltas &&
 685             bch2_trans_fs_usage_apply(trans, trans->fs_usage_deltas))
 686                 return -BCH_ERR_btree_insert_need_mark_replicas;
 687
 688         if (trans->nr_wb_updates) {
 689                 EBUG_ON(flags & BTREE_INSERT_JOURNAL_REPLAY);
 690
 691                 ret = bch2_btree_insert_keys_write_buffer(trans);
 692                 if (ret)
 693                         goto revert_fs_usage;
 694         }
 695
 696         trans_for_each_update(trans, i)
 697                 if (BTREE_NODE_TYPE_HAS_MEM_TRIGGERS & (1U << i->bkey_type)) {
 698                         ret = run_one_mem_trigger(trans, i, i->flags);
 699                         if (ret)
 700                                 goto fatal_err;
 701                 }
 702
 703         if (unlikely(c->gc_pos.phase)) {
 704                 ret = bch2_trans_commit_run_gc_triggers(trans);
 705                 if  (ret)
 706                         goto fatal_err;
 707         }
 708
 709         if (unlikely(trans->extra_journal_entries.nr)) {
 710                 memcpy_u64s_small(journal_res_entry(&c->journal, &trans->journal_res),
 711                                   trans->extra_journal_entries.data,
 712                                   trans->extra_journal_entries.nr);
 713
 714                 trans->journal_res.offset       += trans->extra_journal_entries.nr;
 715                 trans->journal_res.u64s         -= trans->extra_journal_entries.nr;
 716         }
 717
 718         if (likely(!(flags & BTREE_INSERT_JOURNAL_REPLAY))) {
 719                 struct journal *j = &c->journal;
 720                 struct jset_entry *entry;
 721
 722                 trans_for_each_update(trans, i) {
 723                         if (i->key_cache_already_flushed)
 724                                 continue;
 725
 726                         if (i->flags & BTREE_UPDATE_NOJOURNAL)
 727                                 continue;
 728
 729                         verify_update_old_key(trans, i);
 730
 731                         if (trans->journal_transaction_names) {
 732                                 entry = bch2_journal_add_entry(j, &trans->journal_res,
 733                                                        BCH_JSET_ENTRY_overwrite,
 734                                                        i->btree_id, i->level,
 735                                                        i->old_k.u64s);
 736                                 bkey_reassemble(&entry->start[0],
 737                                                 (struct bkey_s_c) { &i->old_k, i->old_v });
 738                         }
 739
 740                         entry = bch2_journal_add_entry(j, &trans->journal_res,
 741                                                BCH_JSET_ENTRY_btree_keys,
 742                                                i->btree_id, i->level,
 743                                                i->k->k.u64s);
 744                         bkey_copy(&entry->start[0], i->k);
 745                 }
 746
 747                 trans_for_each_wb_update(trans, wb) {
 748                         entry = bch2_journal_add_entry(j, &trans->journal_res,
 749                                                BCH_JSET_ENTRY_btree_keys,
 750                                                wb->btree, 0,
 751                                                wb->k.k.u64s);
 752                         bkey_copy(&entry->start[0], &wb->k);
 753                 }
 754
 755                 if (trans->journal_seq)
 756                         *trans->journal_seq = trans->journal_res.seq;
 757         }
 758
 759         trans_for_each_update(trans, i) {
 760                 i->k->k.needs_whiteout = false;
 761
 762                 if (!i->cached)
 763                         btree_insert_key_leaf(trans, i);
 764                 else if (!i->key_cache_already_flushed)
 765                         bch2_btree_insert_key_cached(trans, flags, i->path, i->k);
 766                 else {
 767                         bch2_btree_key_cache_drop(trans, i->path);
 768                         btree_path_set_dirty(i->path, BTREE_ITER_NEED_TRAVERSE);
 769                 }
 770         }
 771
 772         return 0;
 773 fatal_err:
 774         bch2_fatal_error(c);
 775 revert_fs_usage:
 776         if (trans->fs_usage_deltas)
 777                 bch2_trans_fs_usage_revert(trans, trans->fs_usage_deltas);
 778         return ret;
 779 }
 780
 781 static noinline int trans_lock_write_fail(struct btree_trans *trans, struct btree_insert_entry *i)
 782 {
 783         while (--i >= trans->updates) {
 784                 if (same_leaf_as_prev(trans, i))
 785                         continue;
 786
 787                 bch2_btree_node_unlock_write(trans, i->path, insert_l(i)->b);
 788         }
 789
 790         trace_and_count(trans->c, trans_restart_would_deadlock_write, trans);
 791         return btree_trans_restart(trans, BCH_ERR_transaction_restart_would_deadlock_write);
 792 }
 793
 794 static inline int trans_lock_write(struct btree_trans *trans)
 795 {
 796         struct btree_insert_entry *i;
 797
 798         trans_for_each_update(trans, i) {
 799                 if (same_leaf_as_prev(trans, i))
 800                         continue;
 801
 802                 if (bch2_btree_node_lock_write(trans, i->path, &insert_l(i)->b->c))
 803                         return trans_lock_write_fail(trans, i);
 804
 805                 if (!i->cached)
 806                         bch2_btree_node_prep_for_write(trans, i->path, insert_l(i)->b);
 807         }
 808
 809         return 0;
 810 }
 811
 812 static noinline void bch2_drop_overwrites_from_journal(struct btree_trans *trans)
 813 {
 814         struct btree_insert_entry *i;
 815         struct btree_write_buffered_key *wb;
 816
 817         trans_for_each_update(trans, i)
 818                 bch2_journal_key_overwritten(trans->c, i->btree_id, i->level, i->k->k.p);
 819
 820         trans_for_each_wb_update(trans, wb)
 821                 bch2_journal_key_overwritten(trans->c, wb->btree, 0, wb->k.k.p);
 822 }
 823
 824 #ifdef CONFIG_BCACHEFS_DEBUG
 825 static noinline int bch2_trans_commit_bkey_invalid(struct btree_trans *trans, unsigned flags,
 826                                                    struct btree_insert_entry *i,
 827                                                    struct printbuf *err)
 828 {
 829         struct bch_fs *c = trans->c;
 830         int rw = (flags & BTREE_INSERT_JOURNAL_REPLAY) ? READ : WRITE;
 831
 832         printbuf_reset(err);
 833         prt_printf(err, "invalid bkey on insert from %s -> %ps",
 834                    trans->fn, (void *) i->ip_allocated);
 835         prt_newline(err);
 836         printbuf_indent_add(err, 2);
 837
 838         bch2_bkey_val_to_text(err, c, bkey_i_to_s_c(i->k));
 839         prt_newline(err);
 840
 841         bch2_bkey_invalid(c, bkey_i_to_s_c(i->k),
 842                           i->bkey_type, rw, err);
 843         bch2_print_string_as_lines(KERN_ERR, err->buf);
 844
 845         bch2_inconsistent_error(c);
 846         bch2_dump_trans_updates(trans);
 847         printbuf_exit(err);
 848
 849         return -EINVAL;
 850 }
 851 #endif
 852
 853 /*
 854  * Get journal reservation, take write locks, and attempt to do btree update(s):
 855  */
 856 static inline int do_bch2_trans_commit(struct btree_trans *trans, unsigned flags,
 857                                        struct btree_insert_entry **stopped_at,
 858                                        unsigned long trace_ip)
 859 {
 860         struct bch_fs *c = trans->c;
 861         struct btree_insert_entry *i;
 862         int ret, u64s_delta = 0;
 863
 864 #ifdef CONFIG_BCACHEFS_DEBUG
 865         struct printbuf buf = PRINTBUF;
 866
 867         trans_for_each_update(trans, i) {
 868                 int rw = (flags & BTREE_INSERT_JOURNAL_REPLAY) ? READ : WRITE;
 869
 870                 if (unlikely(bch2_bkey_invalid(c, bkey_i_to_s_c(i->k),
 871                                                i->bkey_type, rw, &buf)))
 872                         return bch2_trans_commit_bkey_invalid(trans, flags, i, &buf);
 873                 btree_insert_entry_checks(trans, i);
 874         }
 875         printbuf_exit(&buf);
 876 #endif
 877
 878         trans_for_each_update(trans, i) {
 879                 if (i->cached)
 880                         continue;
 881
 882                 u64s_delta += !bkey_deleted(&i->k->k) ? i->k->k.u64s : 0;
 883                 u64s_delta -= i->old_btree_u64s;
 884
 885                 if (!same_leaf_as_next(trans, i)) {
 886                         if (u64s_delta <= 0) {
 887                                 ret = bch2_foreground_maybe_merge(trans, i->path,
 888                                                         i->level, flags);
 889                                 if (unlikely(ret))
 890                                         return ret;
 891                         }
 892
 893                         u64s_delta = 0;
 894                 }
 895         }
 896
 897         ret = bch2_journal_preres_get(&c->journal,
 898                         &trans->journal_preres, trans->journal_preres_u64s,
 899                         (flags & JOURNAL_WATERMARK_MASK)|JOURNAL_RES_GET_NONBLOCK);
 900         if (unlikely(ret == -BCH_ERR_journal_preres_get_blocked))
 901                 ret = bch2_trans_journal_preres_get_cold(trans, flags, trace_ip);
 902         if (unlikely(ret))
 903                 return ret;
 904
 905         ret = trans_lock_write(trans);
 906         if (unlikely(ret))
 907                 return ret;
 908
 909         ret = bch2_trans_commit_write_locked(trans, flags, stopped_at, trace_ip);
 910
 911         if (!ret && unlikely(trans->journal_replay_not_finished))
 912                 bch2_drop_overwrites_from_journal(trans);
 913
 914         trans_for_each_update(trans, i)
 915                 if (!same_leaf_as_prev(trans, i))
 916                         bch2_btree_node_unlock_write_inlined(trans, i->path,
 917                                                         insert_l(i)->b);
 918
 919         if (!ret && trans->journal_pin)
 920                 bch2_journal_pin_add(&c->journal, trans->journal_res.seq,
 921                                      trans->journal_pin, NULL);
 922
 923         /*
 924          * Drop journal reservation after dropping write locks, since dropping
 925          * the journal reservation may kick off a journal write:
 926          */
 927         bch2_journal_res_put(&c->journal, &trans->journal_res);
 928
 929         if (unlikely(ret))
 930                 return ret;
 931
 932         bch2_trans_downgrade(trans);
 933
 934         return 0;
 935 }
 936
 937 static int journal_reclaim_wait_done(struct bch_fs *c)
 938 {
 939         int ret = bch2_journal_error(&c->journal) ?:
 940                 !bch2_btree_key_cache_must_wait(c);
 941
 942         if (!ret)
 943                 journal_reclaim_kick(&c->journal);
 944         return ret;
 945 }
 946
 947 static noinline
 948 int bch2_trans_commit_error(struct btree_trans *trans, unsigned flags,
 949                             struct btree_insert_entry *i,
 950                             int ret, unsigned long trace_ip)
 951 {
 952         struct bch_fs *c = trans->c;
 953
 954         switch (ret) {
 955         case -BCH_ERR_btree_insert_btree_node_full:
 956                 ret = bch2_btree_split_leaf(trans, i->path, flags);
 957                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 958                         trace_and_count(c, trans_restart_btree_node_split, trans, trace_ip, i->path);
 959                 break;
 960         case -BCH_ERR_btree_insert_need_mark_replicas:
 961                 bch2_trans_unlock(trans);
 962
 963                 ret = bch2_replicas_delta_list_mark(c, trans->fs_usage_deltas);
 964                 if (ret)
 965                         break;
 966
 967                 ret = bch2_trans_relock(trans);
 968                 if (ret)
 969                         trace_and_count(c, trans_restart_mark_replicas, trans, trace_ip);
 970                 break;
 971         case -BCH_ERR_journal_res_get_blocked:
 972                 bch2_trans_unlock(trans);
 973
 974                 if ((flags & BTREE_INSERT_JOURNAL_RECLAIM) &&
 975                     !(flags & JOURNAL_WATERMARK_reserved)) {
 976                         ret = -BCH_ERR_journal_reclaim_would_deadlock;
 977                         break;
 978                 }
 979
 980                 ret = bch2_trans_journal_res_get(trans,
 981                                         (flags & JOURNAL_WATERMARK_MASK)|
 982                                         JOURNAL_RES_GET_CHECK);
 983                 if (ret)
 984                         break;
 985
 986                 ret = bch2_trans_relock(trans);
 987                 if (ret)
 988                         trace_and_count(c, trans_restart_journal_res_get, trans, trace_ip);
 989                 break;
 990         case -BCH_ERR_btree_insert_need_journal_reclaim:
 991                 bch2_trans_unlock(trans);
 992
 993                 trace_and_count(c, trans_blocked_journal_reclaim, trans, trace_ip);
 994
 995                 wait_event_freezable(c->journal.reclaim_wait,
 996                                      (ret = journal_reclaim_wait_done(c)));
 997                 if (ret < 0)
 998                         break;
 999
1000                 ret = bch2_trans_relock(trans);
1001                 if (ret)
1002                         trace_and_count(c, trans_restart_journal_reclaim, trans, trace_ip);
1003                 break;
1004         case -BCH_ERR_btree_insert_need_flush_buffer: {
1005                 struct btree_write_buffer *wb = &c->btree_write_buffer;
1006
1007                 ret = 0;
1008
1009                 if (wb->state.nr > wb->size * 3 / 4) {
1010                         bch2_trans_reset_updates(trans);
1011                         bch2_trans_unlock(trans);
1012
1013                         mutex_lock(&wb->flush_lock);
1014
1015                         if (wb->state.nr > wb->size * 3 / 4)
1016                                 ret = __bch2_btree_write_buffer_flush(trans,
1017                                                 flags|BTREE_INSERT_NOCHECK_RW, true);
1018                         else
1019                                 mutex_unlock(&wb->flush_lock);
1020
1021                         if (!ret) {
1022                                 trace_and_count(c, trans_restart_write_buffer_flush, trans, _THIS_IP_);
1023                                 ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_write_buffer_flush);
1024                         }
1025                 }
1026                 break;
1027         }
1028         default:
1029                 BUG_ON(ret >= 0);
1030                 break;
1031         }
1032
1033         BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart) != !!trans->restarted);
1034
1035         bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOSPC) &&
1036                                 !(flags & BTREE_INSERT_NOWAIT) &&
1037                                 (flags & BTREE_INSERT_NOFAIL), c,
1038                 "%s: incorrectly got %s\n", __func__, bch2_err_str(ret));
1039
1040         return ret;
1041 }
1042
1043 static noinline int
1044 bch2_trans_commit_get_rw_cold(struct btree_trans *trans, unsigned flags)
1045 {
1046         struct bch_fs *c = trans->c;
1047         int ret;
1048
1049         if (likely(!(flags & BTREE_INSERT_LAZY_RW)) ||
1050             test_bit(BCH_FS_STARTED, &c->flags))
1051                 return -BCH_ERR_erofs_trans_commit;
1052
1053         bch2_trans_unlock(trans);
1054
1055         ret =   bch2_fs_read_write_early(c) ?:
1056                 bch2_trans_relock(trans);
1057         if (ret)
1058                 return ret;
1059
1060         bch2_write_ref_get(c, BCH_WRITE_REF_trans);
1061         return 0;
1062 }
1063
1064 /*
1065  * This is for updates done in the early part of fsck - btree_gc - before we've
1066  * gone RW. we only add the new key to the list of keys for journal replay to
1067  * do.
1068  */
1069 static noinline int
1070 do_bch2_trans_commit_to_journal_replay(struct btree_trans *trans)
1071 {
1072         struct bch_fs *c = trans->c;
1073         struct btree_insert_entry *i;
1074         int ret = 0;
1075
1076         trans_for_each_update(trans, i) {
1077                 ret = bch2_journal_key_insert(c, i->btree_id, i->level, i->k);
1078                 if (ret)
1079                         break;
1080         }
1081
1082         return ret;
1083 }
1084
1085 int __bch2_trans_commit(struct btree_trans *trans, unsigned flags)
1086 {
1087         struct bch_fs *c = trans->c;
1088         struct btree_insert_entry *i = NULL;
1089         struct btree_write_buffered_key *wb;
1090         unsigned u64s;
1091         int ret = 0;
1092
1093         if (!trans->nr_updates &&
1094             !trans->nr_wb_updates &&
1095             !trans->extra_journal_entries.nr)
1096                 goto out_reset;
1097
1098         if (flags & BTREE_INSERT_GC_LOCK_HELD)
1099                 lockdep_assert_held(&c->gc_lock);
1100
1101         ret = bch2_trans_commit_run_triggers(trans);
1102         if (ret)
1103                 goto out_reset;
1104
1105         if (unlikely(!test_bit(BCH_FS_MAY_GO_RW, &c->flags))) {
1106                 ret = do_bch2_trans_commit_to_journal_replay(trans);
1107                 goto out_reset;
1108         }
1109
1110         if (!(flags & BTREE_INSERT_NOCHECK_RW) &&
1111             unlikely(!bch2_write_ref_tryget(c, BCH_WRITE_REF_trans))) {
1112                 ret = bch2_trans_commit_get_rw_cold(trans, flags);
1113                 if (ret)
1114                         goto out_reset;
1115         }
1116
1117         if (c->btree_write_buffer.state.nr > c->btree_write_buffer.size / 2 &&
1118             mutex_trylock(&c->btree_write_buffer.flush_lock)) {
1119                 bch2_trans_begin(trans);
1120                 bch2_trans_unlock(trans);
1121
1122                 ret = __bch2_btree_write_buffer_flush(trans,
1123                                         flags|BTREE_INSERT_NOCHECK_RW, true);
1124                 if (!ret) {
1125                         trace_and_count(c, trans_restart_write_buffer_flush, trans, _THIS_IP_);
1126                         ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_write_buffer_flush);
1127                 }
1128                 goto out;
1129         }
1130
1131         EBUG_ON(test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags));
1132
1133         memset(&trans->journal_preres, 0, sizeof(trans->journal_preres));
1134
1135         trans->journal_u64s             = trans->extra_journal_entries.nr;
1136         trans->journal_preres_u64s      = 0;
1137
1138         trans->journal_transaction_names = READ_ONCE(c->opts.journal_transaction_names);
1139
1140         if (trans->journal_transaction_names)
1141                 trans->journal_u64s += jset_u64s(JSET_ENTRY_LOG_U64s);
1142
1143         trans_for_each_update(trans, i) {
1144                 EBUG_ON(!i->path->should_be_locked);
1145
1146                 ret = bch2_btree_path_upgrade(trans, i->path, i->level + 1);
1147                 if (unlikely(ret))
1148                         goto out;
1149
1150                 EBUG_ON(!btree_node_intent_locked(i->path, i->level));
1151
1152                 if (i->key_cache_already_flushed)
1153                         continue;
1154
1155                 /* we're going to journal the key being updated: */
1156                 u64s = jset_u64s(i->k->k.u64s);
1157                 if (i->cached &&
1158                     likely(!(flags & BTREE_INSERT_JOURNAL_REPLAY)))
1159                         trans->journal_preres_u64s += u64s;
1160
1161                 if (i->flags & BTREE_UPDATE_NOJOURNAL)
1162                         continue;
1163
1164                 trans->journal_u64s += u64s;
1165
1166                 /* and we're also going to log the overwrite: */
1167                 if (trans->journal_transaction_names)
1168                         trans->journal_u64s += jset_u64s(i->old_k.u64s);
1169         }
1170
1171         trans_for_each_wb_update(trans, wb)
1172                 trans->journal_u64s += jset_u64s(wb->k.k.u64s);
1173
1174         if (trans->extra_journal_res) {
1175                 ret = bch2_disk_reservation_add(c, trans->disk_res,
1176                                 trans->extra_journal_res,
1177                                 (flags & BTREE_INSERT_NOFAIL)
1178                                 ? BCH_DISK_RESERVATION_NOFAIL : 0);
1179                 if (ret)
1180                         goto err;
1181         }
1182 retry:
1183         bch2_trans_verify_not_in_restart(trans);
1184         memset(&trans->journal_res, 0, sizeof(trans->journal_res));
1185
1186         ret = do_bch2_trans_commit(trans, flags, &i, _RET_IP_);
1187
1188         /* make sure we didn't drop or screw up locks: */
1189         bch2_trans_verify_locks(trans);
1190
1191         if (ret)
1192                 goto err;
1193
1194         trace_and_count(c, transaction_commit, trans, _RET_IP_);
1195 out:
1196         bch2_journal_preres_put(&c->journal, &trans->journal_preres);
1197
1198         if (likely(!(flags & BTREE_INSERT_NOCHECK_RW)))
1199                 bch2_write_ref_put(c, BCH_WRITE_REF_trans);
1200 out_reset:
1201         bch2_trans_reset_updates(trans);
1202
1203         return ret;
1204 err:
1205         ret = bch2_trans_commit_error(trans, flags, i, ret, _RET_IP_);
1206         if (ret)
1207                 goto out;
1208
1209         goto retry;
1210 }
1211
1212 static noinline int __check_pos_snapshot_overwritten(struct btree_trans *trans,
1213                                           enum btree_id id,
1214                                           struct bpos pos)
1215 {
1216         struct bch_fs *c = trans->c;
1217         struct btree_iter iter;
1218         struct bkey_s_c k;
1219         int ret;
1220
1221         bch2_trans_iter_init(trans, &iter, id, pos,
1222                              BTREE_ITER_NOT_EXTENTS|
1223                              BTREE_ITER_ALL_SNAPSHOTS);
1224         while (1) {
1225                 k = bch2_btree_iter_prev(&iter);
1226                 ret = bkey_err(k);
1227                 if (ret)
1228                         break;
1229
1230                 if (!k.k)
1231                         break;
1232
1233                 if (!bkey_eq(pos, k.k->p))
1234                         break;
1235
1236                 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, pos.snapshot)) {
1237                         ret = 1;
1238                         break;
1239                 }
1240         }
1241         bch2_trans_iter_exit(trans, &iter);
1242
1243         return ret;
1244 }
1245
1246 static inline int check_pos_snapshot_overwritten(struct btree_trans *trans,
1247                                           enum btree_id id,
1248                                           struct bpos pos)
1249 {
1250         if (!btree_type_has_snapshots(id) ||
1251             pos.snapshot == U32_MAX ||
1252             !snapshot_t(trans->c, pos.snapshot)->children[0])
1253                 return 0;
1254
1255         return __check_pos_snapshot_overwritten(trans, id, pos);
1256 }
1257
1258 static noinline int extent_front_merge(struct btree_trans *trans,
1259                                        struct btree_iter *iter,
1260                                        struct bkey_s_c k,
1261                                        struct bkey_i **insert,
1262                                        enum btree_update_flags flags)
1263 {
1264         struct bch_fs *c = trans->c;
1265         struct bkey_i *update;
1266         int ret;
1267
1268         update = bch2_bkey_make_mut(trans, k);
1269         ret = PTR_ERR_OR_ZERO(update);
1270         if (ret)
1271                 return ret;
1272
1273         if (!bch2_bkey_merge(c, bkey_i_to_s(update), bkey_i_to_s_c(*insert)))
1274                 return 0;
1275
1276         ret =   check_pos_snapshot_overwritten(trans, iter->btree_id, k.k->p) ?:
1277                 check_pos_snapshot_overwritten(trans, iter->btree_id, (*insert)->k.p);
1278         if (ret < 0)
1279                 return ret;
1280         if (ret)
1281                 return 0;
1282
1283         ret = bch2_btree_delete_at(trans, iter, flags);
1284         if (ret)
1285                 return ret;
1286
1287         *insert = update;
1288         return 0;
1289 }
1290
1291 static noinline int extent_back_merge(struct btree_trans *trans,
1292                                       struct btree_iter *iter,
1293                                       struct bkey_i *insert,
1294                                       struct bkey_s_c k)
1295 {
1296         struct bch_fs *c = trans->c;
1297         int ret;
1298
1299         ret =   check_pos_snapshot_overwritten(trans, iter->btree_id, insert->k.p) ?:
1300                 check_pos_snapshot_overwritten(trans, iter->btree_id, k.k->p);
1301         if (ret < 0)
1302                 return ret;
1303         if (ret)
1304                 return 0;
1305
1306         bch2_bkey_merge(c, bkey_i_to_s(insert), k);
1307         return 0;
1308 }
1309
1310 /*
1311  * When deleting, check if we need to emit a whiteout (because we're overwriting
1312  * something in an ancestor snapshot)
1313  */
1314 static int need_whiteout_for_snapshot(struct btree_trans *trans,
1315                                       enum btree_id btree_id, struct bpos pos)
1316 {
1317         struct btree_iter iter;
1318         struct bkey_s_c k;
1319         u32 snapshot = pos.snapshot;
1320         int ret;
1321
1322         if (!bch2_snapshot_parent(trans->c, pos.snapshot))
1323                 return 0;
1324
1325         pos.snapshot++;
1326
1327         for_each_btree_key_norestart(trans, iter, btree_id, pos,
1328                            BTREE_ITER_ALL_SNAPSHOTS|
1329                            BTREE_ITER_NOPRESERVE, k, ret) {
1330                 if (!bkey_eq(k.k->p, pos))
1331                         break;
1332
1333                 if (bch2_snapshot_is_ancestor(trans->c, snapshot,
1334                                               k.k->p.snapshot)) {
1335                         ret = !bkey_whiteout(k.k);
1336                         break;
1337                 }
1338         }
1339         bch2_trans_iter_exit(trans, &iter);
1340
1341         return ret;
1342 }
1343 int bch2_trans_update_extent(struct btree_trans *trans,
1344                              struct btree_iter *orig_iter,
1345                              struct bkey_i *insert,
1346                              enum btree_update_flags flags)
1347 {
1348         struct btree_iter iter;
1349         struct bpos start = bkey_start_pos(&insert->k);
1350         struct bkey_i *update;
1351         struct bkey_s_c k;
1352         enum btree_id btree_id = orig_iter->btree_id;
1353         int ret = 0, compressed_sectors;
1354
1355         bch2_trans_iter_init(trans, &iter, btree_id, start,
1356                              BTREE_ITER_INTENT|
1357                              BTREE_ITER_WITH_UPDATES|
1358                              BTREE_ITER_NOT_EXTENTS);
1359         k = bch2_btree_iter_peek_upto(&iter, POS(insert->k.p.inode, U64_MAX));
1360         if ((ret = bkey_err(k)))
1361                 goto err;
1362         if (!k.k)
1363                 goto out;
1364
1365         if (bkey_eq(k.k->p, bkey_start_pos(&insert->k))) {
1366                 if (bch2_bkey_maybe_mergable(k.k, &insert->k)) {
1367                         ret = extent_front_merge(trans, &iter, k, &insert, flags);
1368                         if (ret)
1369                                 goto err;
1370                 }
1371
1372                 goto next;
1373         }
1374
1375         while (bkey_gt(insert->k.p, bkey_start_pos(k.k))) {
1376                 bool front_split = bkey_lt(bkey_start_pos(k.k), start);
1377                 bool back_split  = bkey_gt(k.k->p, insert->k.p);
1378
1379                 /*
1380                  * If we're going to be splitting a compressed extent, note it
1381                  * so that __bch2_trans_commit() can increase our disk
1382                  * reservation:
1383                  */
1384                 if (((front_split && back_split) ||
1385                      ((front_split || back_split) && k.k->p.snapshot != insert->k.p.snapshot)) &&
1386                     (compressed_sectors = bch2_bkey_sectors_compressed(k)))
1387                         trans->extra_journal_res += compressed_sectors;
1388
1389                 if (front_split) {
1390                         update = bch2_bkey_make_mut(trans, k);
1391                         if ((ret = PTR_ERR_OR_ZERO(update)))
1392                                 goto err;
1393
1394                         bch2_cut_back(start, update);
1395
1396                         ret = bch2_btree_insert_nonextent(trans, btree_id, update,
1397                                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|flags);
1398                         if (ret)
1399                                 goto err;
1400                 }
1401
1402                 if (k.k->p.snapshot != insert->k.p.snapshot &&
1403                     (front_split || back_split)) {
1404                         update = bch2_bkey_make_mut(trans, k);
1405                         if ((ret = PTR_ERR_OR_ZERO(update)))
1406                                 goto err;
1407
1408                         bch2_cut_front(start, update);
1409                         bch2_cut_back(insert->k.p, update);
1410
1411                         ret = bch2_btree_insert_nonextent(trans, btree_id, update,
1412                                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|flags);
1413                         if (ret)
1414                                 goto err;
1415                 }
1416
1417                 if (bkey_le(k.k->p, insert->k.p)) {
1418                         update = bch2_trans_kmalloc(trans, sizeof(*update));
1419                         if ((ret = PTR_ERR_OR_ZERO(update)))
1420                                 goto err;
1421
1422                         bkey_init(&update->k);
1423                         update->k.p = k.k->p;
1424                         update->k.p.snapshot = insert->k.p.snapshot;
1425
1426                         if (insert->k.p.snapshot != k.k->p.snapshot ||
1427                             (btree_type_has_snapshots(btree_id) &&
1428                              need_whiteout_for_snapshot(trans, btree_id, update->k.p)))
1429                                 update->k.type = KEY_TYPE_whiteout;
1430
1431                         ret = bch2_btree_insert_nonextent(trans, btree_id, update,
1432                                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|flags);
1433                         if (ret)
1434                                 goto err;
1435                 }
1436
1437                 if (back_split) {
1438                         update = bch2_bkey_make_mut(trans, k);
1439                         if ((ret = PTR_ERR_OR_ZERO(update)))
1440                                 goto err;
1441
1442                         bch2_cut_front(insert->k.p, update);
1443
1444                         ret = bch2_trans_update_by_path(trans, iter.path, update,
1445                                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|
1446                                                   flags);
1447                         if (ret)
1448                                 goto err;
1449                         goto out;
1450                 }
1451 next:
1452                 bch2_btree_iter_advance(&iter);
1453                 k = bch2_btree_iter_peek_upto(&iter, POS(insert->k.p.inode, U64_MAX));
1454                 if ((ret = bkey_err(k)))
1455                         goto err;
1456                 if (!k.k)
1457                         goto out;
1458         }
1459
1460         if (bch2_bkey_maybe_mergable(&insert->k, k.k)) {
1461                 ret = extent_back_merge(trans, &iter, insert, k);
1462                 if (ret)
1463                         goto err;
1464         }
1465 out:
1466         if (!bkey_deleted(&insert->k)) {
1467                 /*
1468                  * Rewinding iterators is expensive: get a new one and the one
1469                  * that points to the start of insert will be cloned from:
1470                  */
1471                 bch2_trans_iter_exit(trans, &iter);
1472                 bch2_trans_iter_init(trans, &iter, btree_id, insert->k.p,
1473                                      BTREE_ITER_NOT_EXTENTS|
1474                                      BTREE_ITER_INTENT);
1475                 ret   = bch2_btree_iter_traverse(&iter) ?:
1476                         bch2_trans_update(trans, &iter, insert, flags);
1477         }
1478 err:
1479         bch2_trans_iter_exit(trans, &iter);
1480
1481         return ret;
1482 }
1483
1484 static int __must_check
1485 bch2_trans_update_by_path_trace(struct btree_trans *trans, struct btree_path *path,
1486                                 struct bkey_i *k, enum btree_update_flags flags,
1487                                 unsigned long ip);
1488
1489 static noinline int flush_new_cached_update(struct btree_trans *trans,
1490                                             struct btree_path *path,
1491                                             struct btree_insert_entry *i,
1492                                             enum btree_update_flags flags,
1493                                             unsigned long ip)
1494 {
1495         struct btree_path *btree_path;
1496         int ret;
1497
1498         i->key_cache_already_flushed = true;
1499         i->flags |= BTREE_TRIGGER_NORUN;
1500
1501         btree_path = bch2_path_get(trans, path->btree_id, path->pos, 1, 0,
1502                                    BTREE_ITER_INTENT, _THIS_IP_);
1503
1504         ret = bch2_btree_path_traverse(trans, btree_path, 0);
1505         if (ret)
1506                 goto err;
1507
1508         btree_path_set_should_be_locked(btree_path);
1509         ret = bch2_trans_update_by_path_trace(trans, btree_path, i->k, flags, ip);
1510 err:
1511         bch2_path_put(trans, btree_path, true);
1512         return ret;
1513 }
1514
1515 static int __must_check
1516 bch2_trans_update_by_path_trace(struct btree_trans *trans, struct btree_path *path,
1517                                 struct bkey_i *k, enum btree_update_flags flags,
1518                                 unsigned long ip)
1519 {
1520         struct bch_fs *c = trans->c;
1521         struct btree_insert_entry *i, n;
1522         int cmp;
1523
1524         EBUG_ON(!path->should_be_locked);
1525         EBUG_ON(trans->nr_updates >= BTREE_ITER_MAX);
1526         EBUG_ON(!bpos_eq(k->k.p, path->pos));
1527
1528         n = (struct btree_insert_entry) {
1529                 .flags          = flags,
1530                 .bkey_type      = __btree_node_type(path->level, path->btree_id),
1531                 .btree_id       = path->btree_id,
1532                 .level          = path->level,
1533                 .cached         = path->cached,
1534                 .path           = path,
1535                 .k              = k,
1536                 .ip_allocated   = ip,
1537         };
1538
1539 #ifdef CONFIG_BCACHEFS_DEBUG
1540         trans_for_each_update(trans, i)
1541                 BUG_ON(i != trans->updates &&
1542                        btree_insert_entry_cmp(i - 1, i) >= 0);
1543 #endif
1544
1545         /*
1546          * Pending updates are kept sorted: first, find position of new update,
1547          * then delete/trim any updates the new update overwrites:
1548          */
1549         trans_for_each_update(trans, i) {
1550                 cmp = btree_insert_entry_cmp(&n, i);
1551                 if (cmp <= 0)
1552                         break;
1553         }
1554
1555         if (!cmp && i < trans->updates + trans->nr_updates) {
1556                 EBUG_ON(i->insert_trigger_run || i->overwrite_trigger_run);
1557
1558                 bch2_path_put(trans, i->path, true);
1559                 i->flags        = n.flags;
1560                 i->cached       = n.cached;
1561                 i->k            = n.k;
1562                 i->path         = n.path;
1563                 i->ip_allocated = n.ip_allocated;
1564         } else {
1565                 array_insert_item(trans->updates, trans->nr_updates,
1566                                   i - trans->updates, n);
1567
1568                 i->old_v = bch2_btree_path_peek_slot_exact(path, &i->old_k).v;
1569                 i->old_btree_u64s = !bkey_deleted(&i->old_k) ? i->old_k.u64s : 0;
1570
1571                 if (unlikely(trans->journal_replay_not_finished)) {
1572                         struct bkey_i *j_k =
1573                                 bch2_journal_keys_peek_slot(c, n.btree_id, n.level, k->k.p);
1574
1575                         if (j_k) {
1576                                 i->old_k = j_k->k;
1577                                 i->old_v = &j_k->v;
1578                         }
1579                 }
1580         }
1581
1582         __btree_path_get(i->path, true);
1583
1584         /*
1585          * If a key is present in the key cache, it must also exist in the
1586          * btree - this is necessary for cache coherency. When iterating over
1587          * a btree that's cached in the key cache, the btree iter code checks
1588          * the key cache - but the key has to exist in the btree for that to
1589          * work:
1590          */
1591         if (path->cached &&
1592             bkey_deleted(&i->old_k) &&
1593             !(flags & BTREE_UPDATE_NO_KEY_CACHE_COHERENCY))
1594                 return flush_new_cached_update(trans, path, i, flags, ip);
1595
1596         return 0;
1597 }
1598
1599 static inline int __must_check
1600 bch2_trans_update_by_path(struct btree_trans *trans, struct btree_path *path,
1601                           struct bkey_i *k, enum btree_update_flags flags)
1602 {
1603         return bch2_trans_update_by_path_trace(trans, path, k, flags, _RET_IP_);
1604 }
1605
1606 int __must_check bch2_trans_update(struct btree_trans *trans, struct btree_iter *iter,
1607                                    struct bkey_i *k, enum btree_update_flags flags)
1608 {
1609         struct btree_path *path = iter->update_path ?: iter->path;
1610         struct bkey_cached *ck;
1611         int ret;
1612
1613         if (iter->flags & BTREE_ITER_IS_EXTENTS)
1614                 return bch2_trans_update_extent(trans, iter, k, flags);
1615
1616         if (bkey_deleted(&k->k) &&
1617             !(flags & BTREE_UPDATE_KEY_CACHE_RECLAIM) &&
1618             (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)) {
1619                 ret = need_whiteout_for_snapshot(trans, iter->btree_id, k->k.p);
1620                 if (unlikely(ret < 0))
1621                         return ret;
1622
1623                 if (ret)
1624                         k->k.type = KEY_TYPE_whiteout;
1625         }
1626
1627         /*
1628          * Ensure that updates to cached btrees go to the key cache:
1629          */
1630         if (!(flags & BTREE_UPDATE_KEY_CACHE_RECLAIM) &&
1631             !path->cached &&
1632             !path->level &&
1633             btree_id_cached(trans->c, path->btree_id)) {
1634                 if (!iter->key_cache_path ||
1635                     !iter->key_cache_path->should_be_locked ||
1636                     !bpos_eq(iter->key_cache_path->pos, k->k.p)) {
1637                         if (!iter->key_cache_path)
1638                                 iter->key_cache_path =
1639                                         bch2_path_get(trans, path->btree_id, path->pos, 1, 0,
1640                                                       BTREE_ITER_INTENT|
1641                                                       BTREE_ITER_CACHED, _THIS_IP_);
1642
1643                         iter->key_cache_path =
1644                                 bch2_btree_path_set_pos(trans, iter->key_cache_path, path->pos,
1645                                                         iter->flags & BTREE_ITER_INTENT,
1646                                                         _THIS_IP_);
1647
1648                         ret = bch2_btree_path_traverse(trans, iter->key_cache_path,
1649                                                        BTREE_ITER_CACHED);
1650                         if (unlikely(ret))
1651                                 return ret;
1652
1653                         ck = (void *) iter->key_cache_path->l[0].b;
1654
1655                         if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
1656                                 trace_and_count(trans->c, trans_restart_key_cache_raced, trans, _RET_IP_);
1657                                 return btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_raced);
1658                         }
1659
1660                         btree_path_set_should_be_locked(iter->key_cache_path);
1661                 }
1662
1663                 path = iter->key_cache_path;
1664         }
1665
1666         return bch2_trans_update_by_path(trans, path, k, flags);
1667 }
1668
1669 int __must_check bch2_trans_update_buffered(struct btree_trans *trans,
1670                                             enum btree_id btree,
1671                                             struct bkey_i *k)
1672 {
1673         struct btree_write_buffered_key *i;
1674         int ret;
1675
1676         EBUG_ON(trans->nr_wb_updates > trans->wb_updates_size);
1677         EBUG_ON(k->k.u64s > BTREE_WRITE_BUFERED_U64s_MAX);
1678
1679         trans_for_each_wb_update(trans, i) {
1680                 if (i->btree == btree && bpos_eq(i->k.k.p, k->k.p)) {
1681                         bkey_copy(&i->k, k);
1682                         return 0;
1683                 }
1684         }
1685
1686         if (!trans->wb_updates ||
1687             trans->nr_wb_updates == trans->wb_updates_size) {
1688                 struct btree_write_buffered_key *u;
1689
1690                 if (trans->nr_wb_updates == trans->wb_updates_size) {
1691                         struct btree_transaction_stats *s = btree_trans_stats(trans);
1692
1693                         BUG_ON(trans->wb_updates_size > U8_MAX / 2);
1694                         trans->wb_updates_size = max(1, trans->wb_updates_size * 2);
1695                         if (s)
1696                                 s->wb_updates_size = trans->wb_updates_size;
1697                 }
1698
1699                 u = bch2_trans_kmalloc_nomemzero(trans,
1700                                         trans->wb_updates_size *
1701                                         sizeof(struct btree_write_buffered_key));
1702                 ret = PTR_ERR_OR_ZERO(u);
1703                 if (ret)
1704                         return ret;
1705
1706                 if (trans->nr_wb_updates)
1707                         memcpy(u, trans->wb_updates, trans->nr_wb_updates *
1708                                sizeof(struct btree_write_buffered_key));
1709                 trans->wb_updates = u;
1710         }
1711
1712         trans->wb_updates[trans->nr_wb_updates] = (struct btree_write_buffered_key) {
1713                 .btree  = btree,
1714         };
1715
1716         bkey_copy(&trans->wb_updates[trans->nr_wb_updates].k, k);
1717         trans->nr_wb_updates++;
1718
1719         return 0;
1720 }
1721
1722 void bch2_trans_commit_hook(struct btree_trans *trans,
1723                             struct btree_trans_commit_hook *h)
1724 {
1725         h->next = trans->hooks;
1726         trans->hooks = h;
1727 }
1728
1729 int bch2_btree_insert_nonextent(struct btree_trans *trans,
1730                                 enum btree_id btree, struct bkey_i *k,
1731                                 enum btree_update_flags flags)
1732 {
1733         struct btree_iter iter;
1734         int ret;
1735
1736         bch2_trans_iter_init(trans, &iter, btree, k->k.p,
1737                              BTREE_ITER_NOT_EXTENTS|
1738                              BTREE_ITER_INTENT);
1739         ret   = bch2_btree_iter_traverse(&iter) ?:
1740                 bch2_trans_update(trans, &iter, k, flags);
1741         bch2_trans_iter_exit(trans, &iter);
1742         return ret;
1743 }
1744
1745 int __bch2_btree_insert(struct btree_trans *trans, enum btree_id id,
1746                         struct bkey_i *k, enum btree_update_flags flags)
1747 {
1748         struct btree_iter iter;
1749         int ret;
1750
1751         bch2_trans_iter_init(trans, &iter, id, bkey_start_pos(&k->k),
1752                              BTREE_ITER_CACHED|
1753                              BTREE_ITER_INTENT);
1754         ret   = bch2_btree_iter_traverse(&iter) ?:
1755                 bch2_trans_update(trans, &iter, k, flags);
1756         bch2_trans_iter_exit(trans, &iter);
1757         return ret;
1758 }
1759
1760 /**
1761  * bch2_btree_insert - insert keys into the extent btree
1762  * @c:                  pointer to struct bch_fs
1763  * @id:                 btree to insert into
1764  * @insert_keys:        list of keys to insert
1765  * @hook:               insert callback
1766  */
1767 int bch2_btree_insert(struct bch_fs *c, enum btree_id id,
1768                       struct bkey_i *k,
1769                       struct disk_reservation *disk_res,
1770                       u64 *journal_seq, int flags)
1771 {
1772         return bch2_trans_do(c, disk_res, journal_seq, flags,
1773                              __bch2_btree_insert(&trans, id, k, 0));
1774 }
1775
1776 int bch2_btree_delete_extent_at(struct btree_trans *trans, struct btree_iter *iter,
1777                                 unsigned len, unsigned update_flags)
1778 {
1779         struct bkey_i *k;
1780
1781         k = bch2_trans_kmalloc(trans, sizeof(*k));
1782         if (IS_ERR(k))
1783                 return PTR_ERR(k);
1784
1785         bkey_init(&k->k);
1786         k->k.p = iter->pos;
1787         bch2_key_resize(&k->k, len);
1788         return bch2_trans_update(trans, iter, k, update_flags);
1789 }
1790
1791 int bch2_btree_delete_at(struct btree_trans *trans,
1792                          struct btree_iter *iter, unsigned update_flags)
1793 {
1794         return bch2_btree_delete_extent_at(trans, iter, 0, update_flags);
1795 }
1796
1797 int bch2_btree_delete_range_trans(struct btree_trans *trans, enum btree_id id,
1798                                   struct bpos start, struct bpos end,
1799                                   unsigned update_flags,
1800                                   u64 *journal_seq)
1801 {
1802         u32 restart_count = trans->restart_count;
1803         struct btree_iter iter;
1804         struct bkey_s_c k;
1805         int ret = 0;
1806
1807         bch2_trans_iter_init(trans, &iter, id, start, BTREE_ITER_INTENT);
1808         while ((k = bch2_btree_iter_peek_upto(&iter, end)).k) {
1809                 struct disk_reservation disk_res =
1810                         bch2_disk_reservation_init(trans->c, 0);
1811                 struct bkey_i delete;
1812
1813                 ret = bkey_err(k);
1814                 if (ret)
1815                         goto err;
1816
1817                 bkey_init(&delete.k);
1818
1819                 /*
1820                  * This could probably be more efficient for extents:
1821                  */
1822
1823                 /*
1824                  * For extents, iter.pos won't necessarily be the same as
1825                  * bkey_start_pos(k.k) (for non extents they always will be the
1826                  * same). It's important that we delete starting from iter.pos
1827                  * because the range we want to delete could start in the middle
1828                  * of k.
1829                  *
1830                  * (bch2_btree_iter_peek() does guarantee that iter.pos >=
1831                  * bkey_start_pos(k.k)).
1832                  */
1833                 delete.k.p = iter.pos;
1834
1835                 if (iter.flags & BTREE_ITER_IS_EXTENTS)
1836                         bch2_key_resize(&delete.k,
1837                                         bpos_min(end, k.k->p).offset -
1838                                         iter.pos.offset);
1839
1840                 ret   = bch2_trans_update(trans, &iter, &delete, update_flags) ?:
1841                         bch2_trans_commit(trans, &disk_res, journal_seq,
1842                                           BTREE_INSERT_NOFAIL);
1843                 bch2_disk_reservation_put(trans->c, &disk_res);
1844 err:
1845                 /*
1846                  * the bch2_trans_begin() call is in a weird place because we
1847                  * need to call it after every transaction commit, to avoid path
1848                  * overflow, but don't want to call it if the delete operation
1849                  * is a no-op and we have no work to do:
1850                  */
1851                 bch2_trans_begin(trans);
1852
1853                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1854                         ret = 0;
1855                 if (ret)
1856                         break;
1857         }
1858         bch2_trans_iter_exit(trans, &iter);
1859
1860         if (!ret && trans_was_restarted(trans, restart_count))
1861                 ret = -BCH_ERR_transaction_restart_nested;
1862         return ret;
1863 }
1864
1865 /*
1866  * bch_btree_delete_range - delete everything within a given range
1867  *
1868  * Range is a half open interval - [start, end)
1869  */
1870 int bch2_btree_delete_range(struct bch_fs *c, enum btree_id id,
1871                             struct bpos start, struct bpos end,
1872                             unsigned update_flags,
1873                             u64 *journal_seq)
1874 {
1875         int ret = bch2_trans_run(c,
1876                         bch2_btree_delete_range_trans(&trans, id, start, end,
1877                                                       update_flags, journal_seq));
1878         if (ret == -BCH_ERR_transaction_restart_nested)
1879                 ret = 0;
1880         return ret;
1881 }
1882
1883 static int __bch2_trans_log_msg(darray_u64 *entries, const char *fmt, va_list args)
1884 {
1885         struct printbuf buf = PRINTBUF;
1886         struct jset_entry_log *l;
1887         unsigned u64s;
1888         int ret;
1889
1890         prt_vprintf(&buf, fmt, args);
1891         ret = buf.allocation_failure ? -ENOMEM : 0;
1892         if (ret)
1893                 goto err;
1894
1895         u64s = DIV_ROUND_UP(buf.pos, sizeof(u64));
1896
1897         ret = darray_make_room(entries, jset_u64s(u64s));
1898         if (ret)
1899                 goto err;
1900
1901         l = (void *) &darray_top(*entries);
1902         l->entry.u64s           = cpu_to_le16(u64s);
1903         l->entry.btree_id       = 0;
1904         l->entry.level          = 1;
1905         l->entry.type           = BCH_JSET_ENTRY_log;
1906         l->entry.pad[0]         = 0;
1907         l->entry.pad[1]         = 0;
1908         l->entry.pad[2]         = 0;
1909         memcpy(l->d, buf.buf, buf.pos);
1910         while (buf.pos & 7)
1911                 l->d[buf.pos++] = '\0';
1912
1913         entries->nr += jset_u64s(u64s);
1914 err:
1915         printbuf_exit(&buf);
1916         return ret;
1917 }
1918
1919 int bch2_trans_log_msg(struct btree_trans *trans, const char *fmt, ...)
1920 {
1921         va_list args;
1922         int ret;
1923
1924         va_start(args, fmt);
1925         ret = __bch2_trans_log_msg(&trans->extra_journal_entries, fmt, args);
1926         va_end(args);
1927
1928         return ret;
1929 }
1930
1931 int bch2_fs_log_msg(struct bch_fs *c, const char *fmt, ...)
1932 {
1933         va_list args;
1934         int ret;
1935
1936         va_start(args, fmt);
1937
1938         if (!test_bit(JOURNAL_STARTED, &c->journal.flags)) {
1939                 ret = __bch2_trans_log_msg(&c->journal.early_journal_entries, fmt, args);
1940         } else {
1941                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1942                         __bch2_trans_log_msg(&trans.extra_journal_entries, fmt, args));
1943         }
1944
1945         va_end(args);
1946
1947         return ret;
1948
1949 }