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 "extents.h"
  11 #include "io.h"
  12 #include "keylist.h"
  13 #include "move.h"
  14 #include "nocow_locking.h"
  15 #include "subvolume.h"
  16
  17 #include <trace/events/bcachefs.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, update_iter;
  26         struct bkey_s_c k;
  27         snapshot_id_list s;
  28         int ret;
  29
  30         if (!btree_type_has_snapshots(id))
  31                 return 0;
  32
  33         darray_init(&s);
  34
  35         if (bkey_eq(old_pos, new_pos))
  36                 return 0;
  37
  38         if (!snapshot_t(c, old_pos.snapshot)->children[0])
  39                 return 0;
  40
  41         bch2_trans_iter_init(trans, &iter, id, old_pos,
  42                              BTREE_ITER_NOT_EXTENTS|
  43                              BTREE_ITER_ALL_SNAPSHOTS);
  44         while (1) {
  45                 k = bch2_btree_iter_prev(&iter);
  46                 ret = bkey_err(k);
  47                 if (ret)
  48                         break;
  49
  50                 if (!bkey_eq(old_pos, k.k->p))
  51                         break;
  52
  53                 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, old_pos.snapshot)) {
  54                         struct bkey_i *update;
  55
  56                         if (snapshot_list_has_ancestor(c, &s, k.k->p.snapshot))
  57                                 continue;
  58
  59                         update = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
  60
  61                         ret = PTR_ERR_OR_ZERO(update);
  62                         if (ret)
  63                                 break;
  64
  65                         bkey_init(&update->k);
  66                         update->k.p = new_pos;
  67                         update->k.p.snapshot = k.k->p.snapshot;
  68
  69                         bch2_trans_iter_init(trans, &update_iter, id, update->k.p,
  70                                              BTREE_ITER_NOT_EXTENTS|
  71                                              BTREE_ITER_ALL_SNAPSHOTS|
  72                                              BTREE_ITER_INTENT);
  73                         ret   = bch2_btree_iter_traverse(&update_iter) ?:
  74                                 bch2_trans_update(trans, &update_iter, update,
  75                                           BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
  76                         bch2_trans_iter_exit(trans, &update_iter);
  77                         if (ret)
  78                                 break;
  79
  80                         ret = snapshot_list_add(c, &s, k.k->p.snapshot);
  81                         if (ret)
  82                                 break;
  83                 }
  84         }
  85         bch2_trans_iter_exit(trans, &iter);
  86         darray_exit(&s);
  87
  88         return ret;
  89 }
  90
  91 static void bch2_bkey_mark_dev_cached(struct bkey_s k, unsigned dev)
  92 {
  93         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
  94         struct bch_extent_ptr *ptr;
  95
  96         bkey_for_each_ptr(ptrs, ptr)
  97                 if (ptr->dev == dev)
  98                         ptr->cached = true;
  99 }
 100
 101 static int __bch2_data_update_index_update(struct btree_trans *trans,
 102                                            struct bch_write_op *op)
 103 {
 104         struct bch_fs *c = op->c;
 105         struct btree_iter iter;
 106         struct data_update *m =
 107                 container_of(op, struct data_update, op);
 108         struct keylist *keys = &op->insert_keys;
 109         struct bkey_buf _new, _insert;
 110         int ret = 0;
 111
 112         bch2_bkey_buf_init(&_new);
 113         bch2_bkey_buf_init(&_insert);
 114         bch2_bkey_buf_realloc(&_insert, c, U8_MAX);
 115
 116         bch2_trans_iter_init(trans, &iter, m->btree_id,
 117                              bkey_start_pos(&bch2_keylist_front(keys)->k),
 118                              BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 119
 120         while (1) {
 121                 struct bkey_s_c k;
 122                 struct bkey_s_c old = bkey_i_to_s_c(m->k.k);
 123                 struct bkey_i *insert;
 124                 struct bkey_i_extent *new;
 125                 const union bch_extent_entry *entry;
 126                 struct extent_ptr_decoded p;
 127                 struct bpos next_pos;
 128                 bool did_work = false;
 129                 bool should_check_enospc;
 130                 s64 i_sectors_delta = 0, disk_sectors_delta = 0;
 131                 unsigned i;
 132
 133                 bch2_trans_begin(trans);
 134
 135                 k = bch2_btree_iter_peek_slot(&iter);
 136                 ret = bkey_err(k);
 137                 if (ret)
 138                         goto err;
 139
 140                 new = bkey_i_to_extent(bch2_keylist_front(keys));
 141
 142                 if (!bch2_extents_match(k, old))
 143                         goto nomatch;
 144
 145                 bkey_reassemble(_insert.k, k);
 146                 insert = _insert.k;
 147
 148                 bch2_bkey_buf_copy(&_new, c, bch2_keylist_front(keys));
 149                 new = bkey_i_to_extent(_new.k);
 150                 bch2_cut_front(iter.pos, &new->k_i);
 151
 152                 bch2_cut_front(iter.pos,        insert);
 153                 bch2_cut_back(new->k.p,         insert);
 154                 bch2_cut_back(insert->k.p,      &new->k_i);
 155
 156                 /*
 157                  * @old: extent that we read from
 158                  * @insert: key that we're going to update, initialized from
 159                  * extent currently in btree - same as @old unless we raced with
 160                  * other updates
 161                  * @new: extent with new pointers that we'll be adding to @insert
 162                  *
 163                  * Fist, drop rewrite_ptrs from @new:
 164                  */
 165                 i = 0;
 166                 bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs_c(old), p, entry) {
 167                         if (((1U << i) & m->data_opts.rewrite_ptrs) &&
 168                             bch2_extent_has_ptr(old, p, bkey_i_to_s_c(insert))) {
 169                                 /*
 170                                  * If we're going to be adding a pointer to the
 171                                  * same device, we have to drop the old one -
 172                                  * otherwise, we can just mark it cached:
 173                                  */
 174                                 if (bch2_bkey_has_device(bkey_i_to_s_c(&new->k_i), p.ptr.dev))
 175                                         bch2_bkey_drop_device_noerror(bkey_i_to_s(insert), p.ptr.dev);
 176                                 else
 177                                         bch2_bkey_mark_dev_cached(bkey_i_to_s(insert), p.ptr.dev);
 178                         }
 179                         i++;
 180                 }
 181
 182
 183                 /* Add new ptrs: */
 184                 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
 185                         if (bch2_bkey_has_device(bkey_i_to_s_c(insert), p.ptr.dev)) {
 186                                 /*
 187                                  * raced with another move op? extent already
 188                                  * has a pointer to the device we just wrote
 189                                  * data to
 190                                  */
 191                                 continue;
 192                         }
 193
 194                         bch2_extent_ptr_decoded_append(insert, &p);
 195                         did_work = true;
 196                 }
 197
 198                 if (!did_work)
 199                         goto nomatch;
 200
 201                 bch2_bkey_narrow_crcs(insert, (struct bch_extent_crc_unpacked) { 0 });
 202                 bch2_extent_normalize(c, bkey_i_to_s(insert));
 203
 204                 ret = bch2_sum_sector_overwrites(trans, &iter, insert,
 205                                                  &should_check_enospc,
 206                                                  &i_sectors_delta,
 207                                                  &disk_sectors_delta);
 208                 if (ret)
 209                         goto err;
 210
 211                 if (disk_sectors_delta > (s64) op->res.sectors) {
 212                         ret = bch2_disk_reservation_add(c, &op->res,
 213                                                 disk_sectors_delta - op->res.sectors,
 214                                                 !should_check_enospc
 215                                                 ? BCH_DISK_RESERVATION_NOFAIL : 0);
 216                         if (ret)
 217                                 goto out;
 218                 }
 219
 220                 next_pos = insert->k.p;
 221
 222                 ret   = insert_snapshot_whiteouts(trans, m->btree_id,
 223                                                   k.k->p, insert->k.p) ?:
 224                         bch2_trans_update(trans, &iter, insert,
 225                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 226                         bch2_trans_commit(trans, &op->res,
 227                                 NULL,
 228                                 BTREE_INSERT_NOFAIL|
 229                                 m->data_opts.btree_insert_flags);
 230                 if (!ret) {
 231                         bch2_btree_iter_set_pos(&iter, next_pos);
 232
 233                         this_cpu_add(c->counters[BCH_COUNTER_move_extent_finish], new->k.size);
 234                         trace_move_extent_finish(&new->k);
 235                 }
 236 err:
 237                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 238                         ret = 0;
 239                 if (ret)
 240                         break;
 241 next:
 242                 while (bkey_ge(iter.pos, bch2_keylist_front(keys)->k.p)) {
 243                         bch2_keylist_pop_front(keys);
 244                         if (bch2_keylist_empty(keys))
 245                                 goto out;
 246                 }
 247                 continue;
 248 nomatch:
 249                 if (m->ctxt) {
 250                         BUG_ON(k.k->p.offset <= iter.pos.offset);
 251                         atomic64_inc(&m->ctxt->stats->keys_raced);
 252                         atomic64_add(k.k->p.offset - iter.pos.offset,
 253                                      &m->ctxt->stats->sectors_raced);
 254                 }
 255
 256                 this_cpu_add(c->counters[BCH_COUNTER_move_extent_race], new->k.size);
 257                 trace_move_extent_race(&new->k);
 258
 259                 bch2_btree_iter_advance(&iter);
 260                 goto next;
 261         }
 262 out:
 263         bch2_trans_iter_exit(trans, &iter);
 264         bch2_bkey_buf_exit(&_insert, c);
 265         bch2_bkey_buf_exit(&_new, c);
 266         BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart));
 267         return ret;
 268 }
 269
 270 int bch2_data_update_index_update(struct bch_write_op *op)
 271 {
 272         struct bch_fs *c = op->c;
 273         struct btree_trans trans;
 274         int ret;
 275
 276         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024);
 277         ret = __bch2_data_update_index_update(&trans, op);
 278         bch2_trans_exit(&trans);
 279
 280         return ret;
 281 }
 282
 283 void bch2_data_update_read_done(struct data_update *m,
 284                                 struct bch_extent_crc_unpacked crc)
 285 {
 286         /* write bio must own pages: */
 287         BUG_ON(!m->op.wbio.bio.bi_vcnt);
 288
 289         m->op.crc = crc;
 290         m->op.wbio.bio.bi_iter.bi_size = crc.compressed_size << 9;
 291
 292         closure_call(&m->op.cl, bch2_write, NULL, NULL);
 293 }
 294
 295 void bch2_data_update_exit(struct data_update *update)
 296 {
 297         struct bch_fs *c = update->op.c;
 298         struct bkey_ptrs_c ptrs =
 299                 bch2_bkey_ptrs_c(bkey_i_to_s_c(update->k.k));
 300         const struct bch_extent_ptr *ptr;
 301
 302         bkey_for_each_ptr(ptrs, ptr)
 303                 bch2_bucket_nocow_unlock(&c->nocow_locks,
 304                                        PTR_BUCKET_POS(c, ptr), 0);
 305
 306         bch2_bkey_buf_exit(&update->k, c);
 307         bch2_disk_reservation_put(c, &update->op.res);
 308         bch2_bio_free_pages_pool(c, &update->op.wbio.bio);
 309 }
 310
 311 void bch2_update_unwritten_extent(struct btree_trans *trans,
 312                                   struct data_update *update)
 313 {
 314         struct bch_fs *c = update->op.c;
 315         struct bio *bio = &update->op.wbio.bio;
 316         struct bkey_i_extent *e;
 317         struct write_point *wp;
 318         struct bch_extent_ptr *ptr;
 319         struct closure cl;
 320         struct btree_iter iter;
 321         struct bkey_s_c k;
 322         int ret;
 323
 324         closure_init_stack(&cl);
 325         bch2_keylist_init(&update->op.insert_keys, update->op.inline_keys);
 326
 327         while (bio_sectors(bio)) {
 328                 unsigned sectors = bio_sectors(bio);
 329
 330                 bch2_trans_iter_init(trans, &iter, update->btree_id, update->op.pos,
 331                                      BTREE_ITER_SLOTS);
 332                 ret = lockrestart_do(trans, ({
 333                         k = bch2_btree_iter_peek_slot(&iter);
 334                         bkey_err(k);
 335                 }));
 336                 bch2_trans_iter_exit(trans, &iter);
 337
 338                 if (ret || !bch2_extents_match(k, bkey_i_to_s_c(update->k.k)))
 339                         break;
 340
 341                 e = bkey_extent_init(update->op.insert_keys.top);
 342                 e->k.p = update->op.pos;
 343
 344                 ret = bch2_alloc_sectors_start_trans(trans,
 345                                 update->op.target,
 346                                 false,
 347                                 update->op.write_point,
 348                                 &update->op.devs_have,
 349                                 update->op.nr_replicas,
 350                                 update->op.nr_replicas,
 351                                 update->op.alloc_reserve,
 352                                 0, &cl, &wp);
 353                 if (bch2_err_matches(ret, BCH_ERR_operation_blocked)) {
 354                         bch2_trans_unlock(trans);
 355                         closure_sync(&cl);
 356                         continue;
 357                 }
 358
 359                 if (ret)
 360                         return;
 361
 362                 sectors = min(sectors, wp->sectors_free);
 363
 364                 bch2_key_resize(&e->k, sectors);
 365
 366                 bch2_open_bucket_get(c, wp, &update->op.open_buckets);
 367                 bch2_alloc_sectors_append_ptrs(c, wp, &e->k_i, sectors, false);
 368                 bch2_alloc_sectors_done(c, wp);
 369
 370                 bio_advance(bio, sectors << 9);
 371                 update->op.pos.offset += sectors;
 372
 373                 extent_for_each_ptr(extent_i_to_s(e), ptr)
 374                         ptr->unwritten = true;
 375                 bch2_keylist_push(&update->op.insert_keys);
 376
 377                 ret = __bch2_data_update_index_update(trans, &update->op);
 378
 379                 bch2_open_buckets_put(c, &update->op.open_buckets);
 380
 381                 if (ret)
 382                         break;
 383         }
 384
 385         if ((atomic_read(&cl.remaining) & CLOSURE_REMAINING_MASK) != 1) {
 386                 bch2_trans_unlock(trans);
 387                 closure_sync(&cl);
 388         }
 389 }
 390
 391 int bch2_data_update_init(struct bch_fs *c, struct data_update *m,
 392                           struct write_point_specifier wp,
 393                           struct bch_io_opts io_opts,
 394                           struct data_update_opts data_opts,
 395                           enum btree_id btree_id,
 396                           struct bkey_s_c k)
 397 {
 398         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 399         const union bch_extent_entry *entry;
 400         struct extent_ptr_decoded p;
 401         unsigned i, reserve_sectors = k.k->size * data_opts.extra_replicas;
 402         int ret;
 403
 404         bch2_bkey_buf_init(&m->k);
 405         bch2_bkey_buf_reassemble(&m->k, c, k);
 406         m->btree_id     = btree_id;
 407         m->data_opts    = data_opts;
 408
 409         bch2_write_op_init(&m->op, c, io_opts);
 410         m->op.pos       = bkey_start_pos(k.k);
 411         m->op.version   = k.k->version;
 412         m->op.target    = data_opts.target;
 413         m->op.write_point = wp;
 414         m->op.flags     |= BCH_WRITE_PAGES_STABLE|
 415                 BCH_WRITE_PAGES_OWNED|
 416                 BCH_WRITE_DATA_ENCODED|
 417                 BCH_WRITE_MOVE|
 418                 m->data_opts.write_flags;
 419         m->op.compression_type =
 420                 bch2_compression_opt_to_type[io_opts.background_compression ?:
 421                                              io_opts.compression];
 422         if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE)
 423                 m->op.alloc_reserve = RESERVE_movinggc;
 424
 425         i = 0;
 426         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 427                 if (((1U << i) & m->data_opts.rewrite_ptrs) &&
 428                     p.ptr.cached)
 429                         BUG();
 430
 431                 if (!((1U << i) & m->data_opts.rewrite_ptrs))
 432                         bch2_dev_list_add_dev(&m->op.devs_have, p.ptr.dev);
 433
 434                 if (((1U << i) & m->data_opts.rewrite_ptrs) &&
 435                     crc_is_compressed(p.crc))
 436                         reserve_sectors += k.k->size;
 437
 438                 /*
 439                  * op->csum_type is normally initialized from the fs/file's
 440                  * current options - but if an extent is encrypted, we require
 441                  * that it stays encrypted:
 442                  */
 443                 if (bch2_csum_type_is_encryption(p.crc.csum_type)) {
 444                         m->op.nonce     = p.crc.nonce + p.crc.offset;
 445                         m->op.csum_type = p.crc.csum_type;
 446                 }
 447
 448                 if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
 449                         m->op.incompressible = true;
 450
 451                 i++;
 452
 453                 bch2_bucket_nocow_lock(&c->nocow_locks,
 454                                        PTR_BUCKET_POS(c, &p.ptr), 0);
 455         }
 456
 457         if (reserve_sectors) {
 458                 ret = bch2_disk_reservation_add(c, &m->op.res, reserve_sectors,
 459                                 m->data_opts.extra_replicas
 460                                 ? 0
 461                                 : BCH_DISK_RESERVATION_NOFAIL);
 462                 if (ret)
 463                         goto err;
 464         }
 465
 466         m->op.nr_replicas = m->op.nr_replicas_required =
 467                 hweight32(m->data_opts.rewrite_ptrs) + m->data_opts.extra_replicas;
 468
 469         BUG_ON(!m->op.nr_replicas);
 470
 471         /* Special handling required: */
 472         if (bkey_extent_is_unwritten(k))
 473                 return -BCH_ERR_unwritten_extent_update;
 474         return 0;
 475 err:
 476         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 477                 bch2_bucket_nocow_unlock(&c->nocow_locks,
 478                                        PTR_BUCKET_POS(c, &p.ptr), 0);
 479
 480         bch2_bkey_buf_exit(&m->k, c);
 481         bch2_bio_free_pages_pool(c, &m->op.wbio.bio);
 482         return ret;
 483 }
 484
 485 void bch2_data_update_opts_normalize(struct bkey_s_c k, struct data_update_opts *opts)
 486 {
 487         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 488         const struct bch_extent_ptr *ptr;
 489         unsigned i = 0;
 490
 491         bkey_for_each_ptr(ptrs, ptr) {
 492                 if ((opts->rewrite_ptrs & (1U << i)) && ptr->cached) {
 493                         opts->kill_ptrs |= 1U << i;
 494                         opts->rewrite_ptrs ^= 1U << i;
 495                 }
 496
 497                 i++;
 498         }
 499 }