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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "alloc_foreground.h"
   5 #include "bkey_buf.h"
   6 #include "btree_update.h"
   7 #include "buckets.h"
   8 #include "data_update.h"
   9 #include "ec.h"
  10 #include "error.h"
  11 #include "extents.h"
  12 #include "io.h"
  13 #include "keylist.h"
  14 #include "move.h"
  15 #include "nocow_locking.h"
  16 #include "subvolume.h"
  17 #include "trace.h"
  18
  19 static int insert_snapshot_whiteouts(struct btree_trans *trans,
  20                                      enum btree_id id,
  21                                      struct bpos old_pos,
  22                                      struct bpos new_pos)
  23 {
  24         struct bch_fs *c = trans->c;
  25         struct btree_iter iter, iter2;
  26         struct bkey_s_c k, k2;
  27         snapshot_id_list s;
  28         struct bkey_i *update;
  29         int ret;
  30
  31         if (!btree_type_has_snapshots(id))
  32                 return 0;
  33
  34         darray_init(&s);
  35
  36         if (!bch2_snapshot_has_children(c, old_pos.snapshot))
  37                 return 0;
  38
  39         bch2_trans_iter_init(trans, &iter, id, old_pos,
  40                              BTREE_ITER_NOT_EXTENTS|
  41                              BTREE_ITER_ALL_SNAPSHOTS);
  42         while (1) {
  43                 k = bch2_btree_iter_prev(&iter);
  44                 ret = bkey_err(k);
  45                 if (ret)
  46                         break;
  47
  48                 if (!k.k)
  49                         break;
  50
  51                 if (!bkey_eq(old_pos, k.k->p))
  52                         break;
  53
  54                 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, old_pos.snapshot) &&
  55                     !snapshot_list_has_ancestor(c, &s, k.k->p.snapshot)) {
  56                         struct bpos whiteout_pos = new_pos;
  57
  58                         whiteout_pos.snapshot = k.k->p.snapshot;
  59
  60                         k2 = bch2_bkey_get_iter(trans, &iter2, id, whiteout_pos,
  61                                                 BTREE_ITER_NOT_EXTENTS|
  62                                                 BTREE_ITER_INTENT);
  63                         ret = bkey_err(k2);
  64
  65                         if (!ret && k2.k->type == KEY_TYPE_deleted) {
  66                                 update = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
  67                                 ret = PTR_ERR_OR_ZERO(update);
  68                                 if (ret)
  69                                         break;
  70
  71                                 bkey_init(&update->k);
  72                                 update->k.p             = whiteout_pos;
  73                                 update->k.type          = KEY_TYPE_whiteout;
  74
  75                                 ret = bch2_trans_update(trans, &iter2, update,
  76                                                         BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
  77                         }
  78                         bch2_trans_iter_exit(trans, &iter2);
  79
  80                         if (ret)
  81                                 break;
  82
  83                         ret = snapshot_list_add(c, &s, k.k->p.snapshot);
  84                         if (ret)
  85                                 break;
  86                 }
  87         }
  88         bch2_trans_iter_exit(trans, &iter);
  89         darray_exit(&s);
  90
  91         return ret;
  92 }
  93
  94 static void trace_move_extent_finish2(struct bch_fs *c, struct bkey_s_c k)
  95 {
  96         if (trace_move_extent_finish_enabled()) {
  97                 struct printbuf buf = PRINTBUF;
  98
  99                 bch2_bkey_val_to_text(&buf, c, k);
 100                 trace_move_extent_finish(c, buf.buf);
 101                 printbuf_exit(&buf);
 102         }
 103 }
 104
 105 static void trace_move_extent_fail2(struct data_update *m,
 106                          struct bkey_s_c new,
 107                          struct bkey_s_c wrote,
 108                          struct bkey_i *insert,
 109                          const char *msg)
 110 {
 111         struct bch_fs *c = m->op.c;
 112         struct bkey_s_c old = bkey_i_to_s_c(m->k.k);
 113         const union bch_extent_entry *entry;
 114         struct bch_extent_ptr *ptr;
 115         struct extent_ptr_decoded p;
 116         struct printbuf buf = PRINTBUF;
 117         unsigned i, rewrites_found = 0;
 118
 119         if (!trace_move_extent_fail_enabled())
 120                 return;
 121
 122         prt_str(&buf, msg);
 123
 124         if (insert) {
 125                 i = 0;
 126                 bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs_c(old), p, entry) {
 127                         struct bkey_s new_s;
 128                         new_s.k = (void *) new.k;
 129                         new_s.v = (void *) new.v;
 130
 131                         if (((1U << i) & m->data_opts.rewrite_ptrs) &&
 132                             (ptr = bch2_extent_has_ptr(old, p, bkey_i_to_s(insert))) &&
 133                             !ptr->cached)
 134                                 rewrites_found |= 1U << i;
 135                         i++;
 136                 }
 137         }
 138
 139         prt_printf(&buf, "\nrewrite ptrs:   %u%u%u%u",
 140                    (m->data_opts.rewrite_ptrs & (1 << 0)) != 0,
 141                    (m->data_opts.rewrite_ptrs & (1 << 1)) != 0,
 142                    (m->data_opts.rewrite_ptrs & (1 << 2)) != 0,
 143                    (m->data_opts.rewrite_ptrs & (1 << 3)) != 0);
 144
 145         prt_printf(&buf, "\nrewrites found: %u%u%u%u",
 146                    (rewrites_found & (1 << 0)) != 0,
 147                    (rewrites_found & (1 << 1)) != 0,
 148                    (rewrites_found & (1 << 2)) != 0,
 149                    (rewrites_found & (1 << 3)) != 0);
 150
 151         prt_str(&buf, "\nold:    ");
 152         bch2_bkey_val_to_text(&buf, c, old);
 153
 154         prt_str(&buf, "\nnew:    ");
 155         bch2_bkey_val_to_text(&buf, c, new);
 156
 157         prt_str(&buf, "\nwrote:  ");
 158         bch2_bkey_val_to_text(&buf, c, wrote);
 159
 160         if (insert) {
 161                 prt_str(&buf, "\ninsert: ");
 162                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(insert));
 163         }
 164
 165         trace_move_extent_fail(c, buf.buf);
 166         printbuf_exit(&buf);
 167 }
 168
 169 static int __bch2_data_update_index_update(struct btree_trans *trans,
 170                                            struct bch_write_op *op)
 171 {
 172         struct bch_fs *c = op->c;
 173         struct btree_iter iter;
 174         struct data_update *m =
 175                 container_of(op, struct data_update, op);
 176         struct keylist *keys = &op->insert_keys;
 177         struct bkey_buf _new, _insert;
 178         int ret = 0;
 179
 180         bch2_bkey_buf_init(&_new);
 181         bch2_bkey_buf_init(&_insert);
 182         bch2_bkey_buf_realloc(&_insert, c, U8_MAX);
 183
 184         bch2_trans_iter_init(trans, &iter, m->btree_id,
 185                              bkey_start_pos(&bch2_keylist_front(keys)->k),
 186                              BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 187
 188         while (1) {
 189                 struct bkey_s_c k;
 190                 struct bkey_s_c old = bkey_i_to_s_c(m->k.k);
 191                 struct bkey_i *insert = NULL;
 192                 struct bkey_i_extent *new;
 193                 const union bch_extent_entry *entry_c;
 194                 union bch_extent_entry *entry;
 195                 struct extent_ptr_decoded p;
 196                 struct bch_extent_ptr *ptr;
 197                 const struct bch_extent_ptr *ptr_c;
 198                 struct bpos next_pos;
 199                 bool should_check_enospc;
 200                 s64 i_sectors_delta = 0, disk_sectors_delta = 0;
 201                 unsigned rewrites_found = 0, durability, i;
 202
 203                 bch2_trans_begin(trans);
 204
 205                 k = bch2_btree_iter_peek_slot(&iter);
 206                 ret = bkey_err(k);
 207                 if (ret)
 208                         goto err;
 209
 210                 new = bkey_i_to_extent(bch2_keylist_front(keys));
 211
 212                 if (!bch2_extents_match(k, old)) {
 213                         trace_move_extent_fail2(m, k, bkey_i_to_s_c(&new->k_i),
 214                                                 NULL, "no match:");
 215                         goto nowork;
 216                 }
 217
 218                 bkey_reassemble(_insert.k, k);
 219                 insert = _insert.k;
 220
 221                 bch2_bkey_buf_copy(&_new, c, bch2_keylist_front(keys));
 222                 new = bkey_i_to_extent(_new.k);
 223                 bch2_cut_front(iter.pos, &new->k_i);
 224
 225                 bch2_cut_front(iter.pos,        insert);
 226                 bch2_cut_back(new->k.p,         insert);
 227                 bch2_cut_back(insert->k.p,      &new->k_i);
 228
 229                 /*
 230                  * @old: extent that we read from
 231                  * @insert: key that we're going to update, initialized from
 232                  * extent currently in btree - same as @old unless we raced with
 233                  * other updates
 234                  * @new: extent with new pointers that we'll be adding to @insert
 235                  *
 236                  * Fist, drop rewrite_ptrs from @new:
 237                  */
 238                 i = 0;
 239                 bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs_c(old), p, entry_c) {
 240                         if (((1U << i) & m->data_opts.rewrite_ptrs) &&
 241                             (ptr = bch2_extent_has_ptr(old, p, bkey_i_to_s(insert))) &&
 242                             !ptr->cached) {
 243                                 bch2_bkey_drop_ptr_noerror(bkey_i_to_s(insert), ptr);
 244                                 /*
 245                                  * See comment below:
 246                                 bch2_extent_ptr_set_cached(bkey_i_to_s(insert), ptr);
 247                                 */
 248                                 rewrites_found |= 1U << i;
 249                         }
 250                         i++;
 251                 }
 252
 253                 if (m->data_opts.rewrite_ptrs &&
 254                     !rewrites_found &&
 255                     bch2_bkey_durability(c, k) >= m->op.opts.data_replicas) {
 256                         trace_move_extent_fail2(m, k, bkey_i_to_s_c(&new->k_i), insert, "no rewrites found:");
 257                         goto nowork;
 258                 }
 259
 260                 /*
 261                  * A replica that we just wrote might conflict with a replica
 262                  * that we want to keep, due to racing with another move:
 263                  */
 264 restart_drop_conflicting_replicas:
 265                 extent_for_each_ptr(extent_i_to_s(new), ptr)
 266                         if ((ptr_c = bch2_bkey_has_device_c(bkey_i_to_s_c(insert), ptr->dev)) &&
 267                             !ptr_c->cached) {
 268                                 bch2_bkey_drop_ptr_noerror(bkey_i_to_s(&new->k_i), ptr);
 269                                 goto restart_drop_conflicting_replicas;
 270                         }
 271
 272                 if (!bkey_val_u64s(&new->k)) {
 273                         trace_move_extent_fail2(m, k, bkey_i_to_s_c(&new->k_i), insert, "new replicas conflicted:");
 274                         goto nowork;
 275                 }
 276
 277                 /* Now, drop pointers that conflict with what we just wrote: */
 278                 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry)
 279                         if ((ptr = bch2_bkey_has_device(bkey_i_to_s(insert), p.ptr.dev)))
 280                                 bch2_bkey_drop_ptr_noerror(bkey_i_to_s(insert), ptr);
 281
 282                 durability = bch2_bkey_durability(c, bkey_i_to_s_c(insert)) +
 283                         bch2_bkey_durability(c, bkey_i_to_s_c(&new->k_i));
 284
 285                 /* Now, drop excess replicas: */
 286 restart_drop_extra_replicas:
 287                 bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs(bkey_i_to_s(insert)), p, entry) {
 288                         unsigned ptr_durability = bch2_extent_ptr_durability(c, &p);
 289
 290                         if (!p.ptr.cached &&
 291                             durability - ptr_durability >= m->op.opts.data_replicas) {
 292                                 durability -= ptr_durability;
 293                                 bch2_bkey_drop_ptr_noerror(bkey_i_to_s(insert), &entry->ptr);
 294                                 /*
 295                                  * Currently, we're dropping unneeded replicas
 296                                  * instead of marking them as cached, since
 297                                  * cached data in stripe buckets prevents them
 298                                  * from being reused:
 299                                 bch2_extent_ptr_set_cached(bkey_i_to_s(insert), &entry->ptr);
 300                                 */
 301                                 goto restart_drop_extra_replicas;
 302                         }
 303                 }
 304
 305                 /* Finally, add the pointers we just wrote: */
 306                 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry)
 307                         bch2_extent_ptr_decoded_append(insert, &p);
 308
 309                 bch2_bkey_narrow_crcs(insert, (struct bch_extent_crc_unpacked) { 0 });
 310                 bch2_extent_normalize(c, bkey_i_to_s(insert));
 311
 312                 ret = bch2_sum_sector_overwrites(trans, &iter, insert,
 313                                                  &should_check_enospc,
 314                                                  &i_sectors_delta,
 315                                                  &disk_sectors_delta);
 316                 if (ret)
 317                         goto err;
 318
 319                 if (disk_sectors_delta > (s64) op->res.sectors) {
 320                         ret = bch2_disk_reservation_add(c, &op->res,
 321                                                 disk_sectors_delta - op->res.sectors,
 322                                                 !should_check_enospc
 323                                                 ? BCH_DISK_RESERVATION_NOFAIL : 0);
 324                         if (ret)
 325                                 goto out;
 326                 }
 327
 328                 next_pos = insert->k.p;
 329
 330                 if (!bkey_eq(bkey_start_pos(&insert->k), bkey_start_pos(k.k))) {
 331                         ret = insert_snapshot_whiteouts(trans, m->btree_id, k.k->p,
 332                                                         bkey_start_pos(&insert->k));
 333                         if (ret)
 334                                 goto err;
 335                 }
 336
 337                 if (!bkey_eq(insert->k.p, k.k->p)) {
 338                         ret = insert_snapshot_whiteouts(trans, m->btree_id,
 339                                                         k.k->p, insert->k.p);
 340                         if (ret)
 341                                 goto err;
 342                 }
 343
 344                 ret   = bch2_trans_update(trans, &iter, insert,
 345                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 346                         bch2_trans_commit(trans, &op->res,
 347                                 NULL,
 348                                 BTREE_INSERT_NOCHECK_RW|
 349                                 BTREE_INSERT_NOFAIL|
 350                                 m->data_opts.btree_insert_flags);
 351                 if (!ret) {
 352                         bch2_btree_iter_set_pos(&iter, next_pos);
 353
 354                         this_cpu_add(c->counters[BCH_COUNTER_move_extent_finish], new->k.size);
 355                         trace_move_extent_finish2(c, bkey_i_to_s_c(&new->k_i));
 356                 }
 357 err:
 358                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 359                         ret = 0;
 360                 if (ret)
 361                         break;
 362 next:
 363                 while (bkey_ge(iter.pos, bch2_keylist_front(keys)->k.p)) {
 364                         bch2_keylist_pop_front(keys);
 365                         if (bch2_keylist_empty(keys))
 366                                 goto out;
 367                 }
 368                 continue;
 369 nowork:
 370                 if (m->ctxt && m->ctxt->stats) {
 371                         BUG_ON(k.k->p.offset <= iter.pos.offset);
 372                         atomic64_inc(&m->ctxt->stats->keys_raced);
 373                         atomic64_add(k.k->p.offset - iter.pos.offset,
 374                                      &m->ctxt->stats->sectors_raced);
 375                 }
 376
 377                 this_cpu_add(c->counters[BCH_COUNTER_move_extent_fail], new->k.size);
 378
 379                 bch2_btree_iter_advance(&iter);
 380                 goto next;
 381         }
 382 out:
 383         bch2_trans_iter_exit(trans, &iter);
 384         bch2_bkey_buf_exit(&_insert, c);
 385         bch2_bkey_buf_exit(&_new, c);
 386         BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart));
 387         return ret;
 388 }
 389
 390 int bch2_data_update_index_update(struct bch_write_op *op)
 391 {
 392         return bch2_trans_run(op->c, __bch2_data_update_index_update(&trans, op));
 393 }
 394
 395 void bch2_data_update_read_done(struct data_update *m,
 396                                 struct bch_extent_crc_unpacked crc)
 397 {
 398         /* write bio must own pages: */
 399         BUG_ON(!m->op.wbio.bio.bi_vcnt);
 400
 401         m->op.crc = crc;
 402         m->op.wbio.bio.bi_iter.bi_size = crc.compressed_size << 9;
 403
 404         closure_call(&m->op.cl, bch2_write, NULL, NULL);
 405 }
 406
 407 void bch2_data_update_exit(struct data_update *update)
 408 {
 409         struct bch_fs *c = update->op.c;
 410         struct bkey_ptrs_c ptrs =
 411                 bch2_bkey_ptrs_c(bkey_i_to_s_c(update->k.k));
 412         const struct bch_extent_ptr *ptr;
 413
 414         bkey_for_each_ptr(ptrs, ptr) {
 415                 if (c->opts.nocow_enabled)
 416                         bch2_bucket_nocow_unlock(&c->nocow_locks,
 417                                                  PTR_BUCKET_POS(c, ptr), 0);
 418                 percpu_ref_put(&bch_dev_bkey_exists(c, ptr->dev)->ref);
 419         }
 420
 421         bch2_bkey_buf_exit(&update->k, c);
 422         bch2_disk_reservation_put(c, &update->op.res);
 423         bch2_bio_free_pages_pool(c, &update->op.wbio.bio);
 424 }
 425
 426 void bch2_update_unwritten_extent(struct btree_trans *trans,
 427                                   struct data_update *update)
 428 {
 429         struct bch_fs *c = update->op.c;
 430         struct bio *bio = &update->op.wbio.bio;
 431         struct bkey_i_extent *e;
 432         struct write_point *wp;
 433         struct bch_extent_ptr *ptr;
 434         struct closure cl;
 435         struct btree_iter iter;
 436         struct bkey_s_c k;
 437         int ret;
 438
 439         closure_init_stack(&cl);
 440         bch2_keylist_init(&update->op.insert_keys, update->op.inline_keys);
 441
 442         while (bio_sectors(bio)) {
 443                 unsigned sectors = bio_sectors(bio);
 444
 445                 bch2_trans_iter_init(trans, &iter, update->btree_id, update->op.pos,
 446                                      BTREE_ITER_SLOTS);
 447                 ret = lockrestart_do(trans, ({
 448                         k = bch2_btree_iter_peek_slot(&iter);
 449                         bkey_err(k);
 450                 }));
 451                 bch2_trans_iter_exit(trans, &iter);
 452
 453                 if (ret || !bch2_extents_match(k, bkey_i_to_s_c(update->k.k)))
 454                         break;
 455
 456                 e = bkey_extent_init(update->op.insert_keys.top);
 457                 e->k.p = update->op.pos;
 458
 459                 ret = bch2_alloc_sectors_start_trans(trans,
 460                                 update->op.target,
 461                                 false,
 462                                 update->op.write_point,
 463                                 &update->op.devs_have,
 464                                 update->op.nr_replicas,
 465                                 update->op.nr_replicas,
 466                                 update->op.alloc_reserve,
 467                                 0, &cl, &wp);
 468                 if (bch2_err_matches(ret, BCH_ERR_operation_blocked)) {
 469                         bch2_trans_unlock(trans);
 470                         closure_sync(&cl);
 471                         continue;
 472                 }
 473
 474                 if (ret)
 475                         return;
 476
 477                 sectors = min(sectors, wp->sectors_free);
 478
 479                 bch2_key_resize(&e->k, sectors);
 480
 481                 bch2_open_bucket_get(c, wp, &update->op.open_buckets);
 482                 bch2_alloc_sectors_append_ptrs(c, wp, &e->k_i, sectors, false);
 483                 bch2_alloc_sectors_done(c, wp);
 484
 485                 bio_advance(bio, sectors << 9);
 486                 update->op.pos.offset += sectors;
 487
 488                 extent_for_each_ptr(extent_i_to_s(e), ptr)
 489                         ptr->unwritten = true;
 490                 bch2_keylist_push(&update->op.insert_keys);
 491
 492                 ret = __bch2_data_update_index_update(trans, &update->op);
 493
 494                 bch2_open_buckets_put(c, &update->op.open_buckets);
 495
 496                 if (ret)
 497                         break;
 498         }
 499
 500         if ((atomic_read(&cl.remaining) & CLOSURE_REMAINING_MASK) != 1) {
 501                 bch2_trans_unlock(trans);
 502                 closure_sync(&cl);
 503         }
 504 }
 505
 506 int bch2_data_update_init(struct btree_trans *trans,
 507                           struct moving_context *ctxt,
 508                           struct data_update *m,
 509                           struct write_point_specifier wp,
 510                           struct bch_io_opts io_opts,
 511                           struct data_update_opts data_opts,
 512                           enum btree_id btree_id,
 513                           struct bkey_s_c k)
 514 {
 515         struct bch_fs *c = trans->c;
 516         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 517         const union bch_extent_entry *entry;
 518         struct extent_ptr_decoded p;
 519         const struct bch_extent_ptr *ptr;
 520         unsigned i, reserve_sectors = k.k->size * data_opts.extra_replicas;
 521         unsigned ptrs_locked = 0;
 522         int ret;
 523
 524         bch2_bkey_buf_init(&m->k);
 525         bch2_bkey_buf_reassemble(&m->k, c, k);
 526         m->btree_id     = btree_id;
 527         m->data_opts    = data_opts;
 528
 529         bch2_write_op_init(&m->op, c, io_opts);
 530         m->op.pos       = bkey_start_pos(k.k);
 531         m->op.version   = k.k->version;
 532         m->op.target    = data_opts.target;
 533         m->op.write_point = wp;
 534         m->op.nr_replicas = 0;
 535         m->op.flags     |= BCH_WRITE_PAGES_STABLE|
 536                 BCH_WRITE_PAGES_OWNED|
 537                 BCH_WRITE_DATA_ENCODED|
 538                 BCH_WRITE_MOVE|
 539                 m->data_opts.write_flags;
 540         m->op.compression_type =
 541                 bch2_compression_opt_to_type[io_opts.background_compression ?:
 542                                              io_opts.compression];
 543         if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE)
 544                 m->op.alloc_reserve = RESERVE_movinggc;
 545
 546         bkey_for_each_ptr(ptrs, ptr)
 547                 percpu_ref_get(&bch_dev_bkey_exists(c, ptr->dev)->ref);
 548
 549         i = 0;
 550         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 551                 bool locked;
 552
 553                 if (((1U << i) & m->data_opts.rewrite_ptrs)) {
 554                         BUG_ON(p.ptr.cached);
 555
 556                         if (crc_is_compressed(p.crc))
 557                                 reserve_sectors += k.k->size;
 558
 559                         m->op.nr_replicas += bch2_extent_ptr_durability(c, &p);
 560                 } else if (!p.ptr.cached) {
 561                         bch2_dev_list_add_dev(&m->op.devs_have, p.ptr.dev);
 562                 }
 563
 564                 /*
 565                  * op->csum_type is normally initialized from the fs/file's
 566                  * current options - but if an extent is encrypted, we require
 567                  * that it stays encrypted:
 568                  */
 569                 if (bch2_csum_type_is_encryption(p.crc.csum_type)) {
 570                         m->op.nonce     = p.crc.nonce + p.crc.offset;
 571                         m->op.csum_type = p.crc.csum_type;
 572                 }
 573
 574                 if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
 575                         m->op.incompressible = true;
 576
 577                 if (c->opts.nocow_enabled) {
 578                         if (ctxt) {
 579                                 move_ctxt_wait_event(ctxt, trans,
 580                                                 (locked = bch2_bucket_nocow_trylock(&c->nocow_locks,
 581                                                                           PTR_BUCKET_POS(c, &p.ptr), 0)) ||
 582                                                 !atomic_read(&ctxt->read_sectors));
 583
 584                                 if (!locked)
 585                                         bch2_bucket_nocow_lock(&c->nocow_locks,
 586                                                                PTR_BUCKET_POS(c, &p.ptr), 0);
 587                         } else {
 588                                 if (!bch2_bucket_nocow_trylock(&c->nocow_locks,
 589                                                                PTR_BUCKET_POS(c, &p.ptr), 0)) {
 590                                         ret = -BCH_ERR_nocow_lock_blocked;
 591                                         goto err;
 592                                 }
 593                         }
 594                         ptrs_locked |= (1U << i);
 595                 }
 596
 597                 i++;
 598         }
 599
 600         if (reserve_sectors) {
 601                 ret = bch2_disk_reservation_add(c, &m->op.res, reserve_sectors,
 602                                 m->data_opts.extra_replicas
 603                                 ? 0
 604                                 : BCH_DISK_RESERVATION_NOFAIL);
 605                 if (ret)
 606                         goto err;
 607         }
 608
 609         m->op.nr_replicas += m->data_opts.extra_replicas;
 610         m->op.nr_replicas_required = m->op.nr_replicas;
 611
 612         BUG_ON(!m->op.nr_replicas);
 613
 614         /* Special handling required: */
 615         if (bkey_extent_is_unwritten(k))
 616                 return -BCH_ERR_unwritten_extent_update;
 617         return 0;
 618 err:
 619         i = 0;
 620         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 621                 if ((1U << i) & ptrs_locked)
 622                         bch2_bucket_nocow_unlock(&c->nocow_locks,
 623                                                  PTR_BUCKET_POS(c, &p.ptr), 0);
 624                 percpu_ref_put(&bch_dev_bkey_exists(c, p.ptr.dev)->ref);
 625                 i++;
 626         }
 627
 628         bch2_bkey_buf_exit(&m->k, c);
 629         bch2_bio_free_pages_pool(c, &m->op.wbio.bio);
 630         return ret;
 631 }
 632
 633 void bch2_data_update_opts_normalize(struct bkey_s_c k, struct data_update_opts *opts)
 634 {
 635         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 636         const struct bch_extent_ptr *ptr;
 637         unsigned i = 0;
 638
 639         bkey_for_each_ptr(ptrs, ptr) {
 640                 if ((opts->rewrite_ptrs & (1U << i)) && ptr->cached) {
 641                         opts->kill_ptrs |= 1U << i;
 642                         opts->rewrite_ptrs ^= 1U << i;
 643                 }
 644
 645                 i++;
 646         }
 647 }