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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "alloc_foreground.h"
   5 #include "backpointers.h"
   6 #include "bkey_buf.h"
   7 #include "btree_gc.h"
   8 #include "btree_update.h"
   9 #include "btree_update_interior.h"
  10 #include "disk_groups.h"
  11 #include "ec.h"
  12 #include "errcode.h"
  13 #include "error.h"
  14 #include "inode.h"
  15 #include "io.h"
  16 #include "journal_reclaim.h"
  17 #include "move.h"
  18 #include "replicas.h"
  19 #include "super-io.h"
  20 #include "keylist.h"
  21
  22 #include <linux/ioprio.h>
  23 #include <linux/kthread.h>
  24
  25 #include <trace/events/bcachefs.h>
  26
  27 static void progress_list_add(struct bch_fs *c, struct bch_move_stats *stats)
  28 {
  29         mutex_lock(&c->data_progress_lock);
  30         list_add(&stats->list, &c->data_progress_list);
  31         mutex_unlock(&c->data_progress_lock);
  32 }
  33
  34 static void progress_list_del(struct bch_fs *c, struct bch_move_stats *stats)
  35 {
  36         mutex_lock(&c->data_progress_lock);
  37         list_del(&stats->list);
  38         mutex_unlock(&c->data_progress_lock);
  39 }
  40
  41 struct moving_io {
  42         struct list_head        list;
  43         struct closure          cl;
  44         bool                    read_completed;
  45
  46         unsigned                read_sectors;
  47         unsigned                write_sectors;
  48
  49         struct bch_read_bio     rbio;
  50
  51         struct data_update      write;
  52         /* Must be last since it is variable size */
  53         struct bio_vec          bi_inline_vecs[0];
  54 };
  55
  56 static void move_free(struct moving_io *io)
  57 {
  58         struct moving_context *ctxt = io->write.ctxt;
  59         struct bch_fs *c = ctxt->c;
  60
  61         bch2_data_update_exit(&io->write);
  62         wake_up(&ctxt->wait);
  63         percpu_ref_put(&c->writes);
  64         kfree(io);
  65 }
  66
  67 static void move_write_done(struct bch_write_op *op)
  68 {
  69         struct moving_io *io = container_of(op, struct moving_io, write.op);
  70         struct moving_context *ctxt = io->write.ctxt;
  71
  72         if (io->write.op.error)
  73                 ctxt->write_error = true;
  74
  75         atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
  76         move_free(io);
  77         closure_put(&ctxt->cl);
  78 }
  79
  80 static void move_write(struct moving_io *io)
  81 {
  82         if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
  83                 move_free(io);
  84                 return;
  85         }
  86
  87         closure_get(&io->write.ctxt->cl);
  88         atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
  89
  90         bch2_data_update_read_done(&io->write, io->rbio.pick.crc);
  91 }
  92
  93 static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
  94 {
  95         struct moving_io *io =
  96                 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
  97
  98         return io && io->read_completed ? io : NULL;
  99 }
 100
 101 static void move_read_endio(struct bio *bio)
 102 {
 103         struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
 104         struct moving_context *ctxt = io->write.ctxt;
 105
 106         atomic_sub(io->read_sectors, &ctxt->read_sectors);
 107         io->read_completed = true;
 108
 109         wake_up(&ctxt->wait);
 110         closure_put(&ctxt->cl);
 111 }
 112
 113 static void do_pending_writes(struct moving_context *ctxt, struct btree_trans *trans)
 114 {
 115         struct moving_io *io;
 116
 117         if (trans)
 118                 bch2_trans_unlock(trans);
 119
 120         while ((io = next_pending_write(ctxt))) {
 121                 list_del(&io->list);
 122                 move_write(io);
 123         }
 124 }
 125
 126 #define move_ctxt_wait_event(_ctxt, _trans, _cond)              \
 127 do {                                                            \
 128         do_pending_writes(_ctxt, _trans);                       \
 129                                                                 \
 130         if (_cond)                                              \
 131                 break;                                          \
 132         __wait_event((_ctxt)->wait,                             \
 133                      next_pending_write(_ctxt) || (_cond));     \
 134 } while (1)
 135
 136 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt,
 137                                        struct btree_trans *trans)
 138 {
 139         unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
 140
 141         move_ctxt_wait_event(ctxt, trans,
 142                 !atomic_read(&ctxt->write_sectors) ||
 143                 atomic_read(&ctxt->write_sectors) != sectors_pending);
 144 }
 145
 146 void bch2_moving_ctxt_exit(struct moving_context *ctxt)
 147 {
 148         move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
 149         closure_sync(&ctxt->cl);
 150         EBUG_ON(atomic_read(&ctxt->write_sectors));
 151
 152         if (ctxt->stats) {
 153                 progress_list_del(ctxt->c, ctxt->stats);
 154
 155                 trace_move_data(ctxt->c,
 156                                 atomic64_read(&ctxt->stats->sectors_moved),
 157                                 atomic64_read(&ctxt->stats->keys_moved));
 158         }
 159 }
 160
 161 void bch2_moving_ctxt_init(struct moving_context *ctxt,
 162                            struct bch_fs *c,
 163                            struct bch_ratelimit *rate,
 164                            struct bch_move_stats *stats,
 165                            struct write_point_specifier wp,
 166                            bool wait_on_copygc)
 167 {
 168         memset(ctxt, 0, sizeof(*ctxt));
 169
 170         ctxt->c         = c;
 171         ctxt->rate      = rate;
 172         ctxt->stats     = stats;
 173         ctxt->wp        = wp;
 174         ctxt->wait_on_copygc = wait_on_copygc;
 175
 176         closure_init_stack(&ctxt->cl);
 177         INIT_LIST_HEAD(&ctxt->reads);
 178         init_waitqueue_head(&ctxt->wait);
 179
 180         if (stats) {
 181                 progress_list_add(c, stats);
 182                 stats->data_type = BCH_DATA_user;
 183         }
 184 }
 185
 186 void bch2_move_stats_init(struct bch_move_stats *stats, char *name)
 187 {
 188         memset(stats, 0, sizeof(*stats));
 189         scnprintf(stats->name, sizeof(stats->name), "%s", name);
 190 }
 191
 192 static int bch2_extent_drop_ptrs(struct btree_trans *trans,
 193                                  struct btree_iter *iter,
 194                                  struct bkey_s_c k,
 195                                  struct data_update_opts data_opts)
 196 {
 197         struct bch_fs *c = trans->c;
 198         struct bkey_i *n;
 199         int ret;
 200
 201         n = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
 202         ret = PTR_ERR_OR_ZERO(n);
 203         if (ret)
 204                 return ret;
 205
 206         bkey_reassemble(n, k);
 207
 208         while (data_opts.kill_ptrs) {
 209                 unsigned i = 0, drop = __fls(data_opts.kill_ptrs);
 210                 struct bch_extent_ptr *ptr;
 211
 212                 bch2_bkey_drop_ptrs(bkey_i_to_s(n), ptr, i++ == drop);
 213                 data_opts.kill_ptrs ^= 1U << drop;
 214         }
 215
 216         /*
 217          * If the new extent no longer has any pointers, bch2_extent_normalize()
 218          * will do the appropriate thing with it (turning it into a
 219          * KEY_TYPE_error key, or just a discard if it was a cached extent)
 220          */
 221         bch2_extent_normalize(c, bkey_i_to_s(n));
 222
 223         /*
 224          * Since we're not inserting through an extent iterator
 225          * (BTREE_ITER_ALL_SNAPSHOTS iterators aren't extent iterators),
 226          * we aren't using the extent overwrite path to delete, we're
 227          * just using the normal key deletion path:
 228          */
 229         if (bkey_deleted(&n->k))
 230                 n->k.size = 0;
 231
 232         return bch2_trans_update(trans, iter, n, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 233                 bch2_trans_commit(trans, NULL, NULL, BTREE_INSERT_NOFAIL);
 234 }
 235
 236 static int bch2_move_extent(struct btree_trans *trans,
 237                             struct btree_iter *iter,
 238                             struct moving_context *ctxt,
 239                             struct bch_io_opts io_opts,
 240                             enum btree_id btree_id,
 241                             struct bkey_s_c k,
 242                             struct data_update_opts data_opts)
 243 {
 244         struct bch_fs *c = trans->c;
 245         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 246         struct moving_io *io;
 247         const union bch_extent_entry *entry;
 248         struct extent_ptr_decoded p;
 249         unsigned sectors = k.k->size, pages;
 250         int ret = -ENOMEM;
 251
 252         bch2_data_update_opts_normalize(k, &data_opts);
 253
 254         if (!data_opts.rewrite_ptrs &&
 255             !data_opts.extra_replicas) {
 256                 if (data_opts.kill_ptrs)
 257                         return bch2_extent_drop_ptrs(trans, iter, k, data_opts);
 258                 return 0;
 259         }
 260
 261         if (!percpu_ref_tryget_live(&c->writes))
 262                 return -EROFS;
 263
 264         /* write path might have to decompress data: */
 265         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 266                 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
 267
 268         pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
 269         io = kzalloc(sizeof(struct moving_io) +
 270                      sizeof(struct bio_vec) * pages, GFP_KERNEL);
 271         if (!io)
 272                 goto err;
 273
 274         io->write.ctxt          = ctxt;
 275         io->read_sectors        = k.k->size;
 276         io->write_sectors       = k.k->size;
 277
 278         bio_init(&io->write.op.wbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 279         bio_set_prio(&io->write.op.wbio.bio,
 280                      IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 281
 282         if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
 283                                  GFP_KERNEL))
 284                 goto err_free;
 285
 286         io->rbio.c              = c;
 287         io->rbio.opts           = io_opts;
 288         bio_init(&io->rbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 289         io->rbio.bio.bi_vcnt = pages;
 290         bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 291         io->rbio.bio.bi_iter.bi_size = sectors << 9;
 292
 293         bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
 294         io->rbio.bio.bi_iter.bi_sector  = bkey_start_offset(k.k);
 295         io->rbio.bio.bi_end_io          = move_read_endio;
 296
 297         ret = bch2_data_update_init(c, &io->write, ctxt->wp, io_opts,
 298                                     data_opts, btree_id, k);
 299         if (ret)
 300                 goto err_free_pages;
 301
 302         io->write.ctxt = ctxt;
 303         io->write.op.end_io = move_write_done;
 304
 305         atomic64_inc(&ctxt->stats->keys_moved);
 306         atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
 307         this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
 308         this_cpu_add(c->counters[BCH_COUNTER_move_extent_read], k.k->size);
 309         trace_move_extent_read(k.k);
 310
 311         atomic_add(io->read_sectors, &ctxt->read_sectors);
 312         list_add_tail(&io->list, &ctxt->reads);
 313
 314         /*
 315          * dropped by move_read_endio() - guards against use after free of
 316          * ctxt when doing wakeup
 317          */
 318         closure_get(&ctxt->cl);
 319         bch2_read_extent(trans, &io->rbio,
 320                          bkey_start_pos(k.k),
 321                          btree_id, k, 0,
 322                          BCH_READ_NODECODE|
 323                          BCH_READ_LAST_FRAGMENT);
 324         return 0;
 325 err_free_pages:
 326         bio_free_pages(&io->write.op.wbio.bio);
 327 err_free:
 328         kfree(io);
 329 err:
 330         percpu_ref_put(&c->writes);
 331         trace_and_count(c, move_extent_alloc_mem_fail, k.k);
 332         return ret;
 333 }
 334
 335 static int lookup_inode(struct btree_trans *trans, struct bpos pos,
 336                         struct bch_inode_unpacked *inode)
 337 {
 338         struct btree_iter iter;
 339         struct bkey_s_c k;
 340         int ret;
 341
 342         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
 343                              BTREE_ITER_ALL_SNAPSHOTS);
 344         k = bch2_btree_iter_peek(&iter);
 345         ret = bkey_err(k);
 346         if (ret)
 347                 goto err;
 348
 349         if (!k.k || bkey_cmp(k.k->p, pos)) {
 350                 ret = -ENOENT;
 351                 goto err;
 352         }
 353
 354         ret = bkey_is_inode(k.k) ? 0 : -EIO;
 355         if (ret)
 356                 goto err;
 357
 358         ret = bch2_inode_unpack(k, inode);
 359         if (ret)
 360                 goto err;
 361 err:
 362         bch2_trans_iter_exit(trans, &iter);
 363         return ret;
 364 }
 365
 366 static int move_ratelimit(struct btree_trans *trans,
 367                           struct moving_context *ctxt)
 368 {
 369         struct bch_fs *c = trans->c;
 370         u64 delay;
 371
 372         if (ctxt->wait_on_copygc) {
 373                 bch2_trans_unlock(trans);
 374                 wait_event_killable(c->copygc_running_wq,
 375                                     !c->copygc_running ||
 376                                     kthread_should_stop());
 377         }
 378
 379         do {
 380                 delay = ctxt->rate ? bch2_ratelimit_delay(ctxt->rate) : 0;
 381
 382                 if (delay) {
 383                         bch2_trans_unlock(trans);
 384                         set_current_state(TASK_INTERRUPTIBLE);
 385                 }
 386
 387                 if ((current->flags & PF_KTHREAD) && kthread_should_stop()) {
 388                         __set_current_state(TASK_RUNNING);
 389                         return 1;
 390                 }
 391
 392                 if (delay)
 393                         schedule_timeout(delay);
 394
 395                 if (unlikely(freezing(current))) {
 396                         move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
 397                         try_to_freeze();
 398                 }
 399         } while (delay);
 400
 401         move_ctxt_wait_event(ctxt, trans,
 402                 atomic_read(&ctxt->write_sectors) <
 403                 c->opts.move_bytes_in_flight >> 9);
 404
 405         move_ctxt_wait_event(ctxt, trans,
 406                 atomic_read(&ctxt->read_sectors) <
 407                 c->opts.move_bytes_in_flight >> 9);
 408
 409         return 0;
 410 }
 411
 412 static int move_get_io_opts(struct btree_trans *trans,
 413                             struct bch_io_opts *io_opts,
 414                             struct bkey_s_c k, u64 *cur_inum)
 415 {
 416         struct bch_inode_unpacked inode;
 417         int ret;
 418
 419         if (*cur_inum == k.k->p.inode)
 420                 return 0;
 421
 422         *io_opts = bch2_opts_to_inode_opts(trans->c->opts);
 423
 424         ret = lookup_inode(trans,
 425                            SPOS(0, k.k->p.inode, k.k->p.snapshot),
 426                            &inode);
 427         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 428                 return ret;
 429
 430         if (!ret)
 431                 bch2_io_opts_apply(io_opts, bch2_inode_opts_get(&inode));
 432
 433         *cur_inum = k.k->p.inode;
 434         return 0;
 435 }
 436
 437 static int __bch2_move_data(struct moving_context *ctxt,
 438                             struct bpos start,
 439                             struct bpos end,
 440                             move_pred_fn pred, void *arg,
 441                             enum btree_id btree_id)
 442 {
 443         struct bch_fs *c = ctxt->c;
 444         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 445         struct bkey_buf sk;
 446         struct btree_trans trans;
 447         struct btree_iter iter;
 448         struct bkey_s_c k;
 449         struct data_update_opts data_opts;
 450         u64 cur_inum = U64_MAX;
 451         int ret = 0, ret2;
 452
 453         bch2_bkey_buf_init(&sk);
 454         bch2_trans_init(&trans, c, 0, 0);
 455
 456         ctxt->stats->data_type  = BCH_DATA_user;
 457         ctxt->stats->btree_id   = btree_id;
 458         ctxt->stats->pos        = start;
 459
 460         bch2_trans_iter_init(&trans, &iter, btree_id, start,
 461                              BTREE_ITER_PREFETCH|
 462                              BTREE_ITER_ALL_SNAPSHOTS);
 463
 464         if (ctxt->rate)
 465                 bch2_ratelimit_reset(ctxt->rate);
 466
 467         while (!move_ratelimit(&trans, ctxt)) {
 468                 bch2_trans_begin(&trans);
 469
 470                 k = bch2_btree_iter_peek(&iter);
 471                 if (!k.k)
 472                         break;
 473
 474                 ret = bkey_err(k);
 475                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 476                         continue;
 477                 if (ret)
 478                         break;
 479
 480                 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
 481                         break;
 482
 483                 ctxt->stats->pos = iter.pos;
 484
 485                 if (!bkey_extent_is_direct_data(k.k))
 486                         goto next_nondata;
 487
 488                 ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 489                 if (ret)
 490                         continue;
 491
 492                 memset(&data_opts, 0, sizeof(data_opts));
 493                 if (!pred(c, arg, k, &io_opts, &data_opts))
 494                         goto next;
 495
 496                 /*
 497                  * The iterator gets unlocked by __bch2_read_extent - need to
 498                  * save a copy of @k elsewhere:
 499                  */
 500                 bch2_bkey_buf_reassemble(&sk, c, k);
 501                 k = bkey_i_to_s_c(sk.k);
 502
 503                 ret2 = bch2_move_extent(&trans, &iter, ctxt, io_opts,
 504                                         btree_id, k, data_opts);
 505                 if (ret2) {
 506                         if (bch2_err_matches(ret2, BCH_ERR_transaction_restart))
 507                                 continue;
 508
 509                         if (ret2 == -ENOMEM) {
 510                                 /* memory allocation failure, wait for some IO to finish */
 511                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 512                                 continue;
 513                         }
 514
 515                         /* XXX signal failure */
 516                         goto next;
 517                 }
 518
 519                 if (ctxt->rate)
 520                         bch2_ratelimit_increment(ctxt->rate, k.k->size);
 521 next:
 522                 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 523 next_nondata:
 524                 bch2_btree_iter_advance(&iter);
 525         }
 526
 527         bch2_trans_iter_exit(&trans, &iter);
 528         bch2_trans_exit(&trans);
 529         bch2_bkey_buf_exit(&sk, c);
 530
 531         return ret;
 532 }
 533
 534 int bch2_move_data(struct bch_fs *c,
 535                    enum btree_id start_btree_id, struct bpos start_pos,
 536                    enum btree_id end_btree_id,   struct bpos end_pos,
 537                    struct bch_ratelimit *rate,
 538                    struct bch_move_stats *stats,
 539                    struct write_point_specifier wp,
 540                    bool wait_on_copygc,
 541                    move_pred_fn pred, void *arg)
 542 {
 543         struct moving_context ctxt;
 544         enum btree_id id;
 545         int ret;
 546
 547         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 548
 549         for (id = start_btree_id;
 550              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 551              id++) {
 552                 stats->btree_id = id;
 553
 554                 if (id != BTREE_ID_extents &&
 555                     id != BTREE_ID_reflink)
 556                         continue;
 557
 558                 ret = __bch2_move_data(&ctxt,
 559                                        id == start_btree_id ? start_pos : POS_MIN,
 560                                        id == end_btree_id   ? end_pos   : POS_MAX,
 561                                        pred, arg, id);
 562                 if (ret)
 563                         break;
 564         }
 565
 566         bch2_moving_ctxt_exit(&ctxt);
 567
 568         return ret;
 569 }
 570
 571 static int verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
 572 {
 573         struct bch_fs *c = trans->c;
 574         struct btree_iter iter;
 575         struct bkey_s_c k;
 576         int ret;
 577
 578         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
 579                              bucket, BTREE_ITER_CACHED);
 580 again:
 581         k = bch2_btree_iter_peek_slot(&iter);
 582         ret = bkey_err(k);
 583
 584         if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
 585                 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
 586
 587                 if (a.v->gen == gen &&
 588                     a.v->dirty_sectors) {
 589                         struct printbuf buf = PRINTBUF;
 590
 591                         if (a.v->data_type == BCH_DATA_btree) {
 592                                 bch2_trans_unlock(trans);
 593                                 if (bch2_btree_interior_updates_flush(c))
 594                                         goto again;
 595                         }
 596
 597                         prt_str(&buf, "failed to evacuate bucket ");
 598                         bch2_bkey_val_to_text(&buf, c, k);
 599
 600                         bch_err(c, "%s", buf.buf);
 601                         printbuf_exit(&buf);
 602                 }
 603         }
 604
 605         bch2_trans_iter_exit(trans, &iter);
 606         return ret;
 607 }
 608
 609 int __bch2_evacuate_bucket(struct moving_context *ctxt,
 610                            struct bpos bucket, int gen,
 611                            struct data_update_opts _data_opts)
 612 {
 613         struct bch_fs *c = ctxt->c;
 614         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 615         struct btree_trans trans;
 616         struct btree_iter iter;
 617         struct bkey_buf sk;
 618         struct bch_backpointer bp;
 619         struct data_update_opts data_opts;
 620         u64 bp_offset = 0, cur_inum = U64_MAX;
 621         int ret = 0;
 622
 623         bch2_bkey_buf_init(&sk);
 624         bch2_trans_init(&trans, c, 0, 0);
 625
 626         while (!(ret = move_ratelimit(&trans, ctxt))) {
 627                 bch2_trans_begin(&trans);
 628
 629                 ret = bch2_get_next_backpointer(&trans, bucket, gen,
 630                                                 &bp_offset, &bp,
 631                                                 BTREE_ITER_CACHED);
 632                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 633                         continue;
 634                 if (ret)
 635                         goto err;
 636                 if (bp_offset == U64_MAX)
 637                         break;
 638
 639                 if (!bp.level) {
 640                         const struct bch_extent_ptr *ptr;
 641                         struct bkey_s_c k;
 642                         unsigned i = 0;
 643
 644                         k = bch2_backpointer_get_key(&trans, &iter,
 645                                                 bucket, bp_offset, bp);
 646                         ret = bkey_err(k);
 647                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 648                                 continue;
 649                         if (ret)
 650                                 goto err;
 651                         if (!k.k)
 652                                 continue;
 653
 654                         bch2_bkey_buf_reassemble(&sk, c, k);
 655                         k = bkey_i_to_s_c(sk.k);
 656
 657                         ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 658                         if (ret) {
 659                                 bch2_trans_iter_exit(&trans, &iter);
 660                                 continue;
 661                         }
 662
 663                         data_opts = _data_opts;
 664                         data_opts.target        = io_opts.background_target;
 665                         data_opts.rewrite_ptrs = 0;
 666
 667                         bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
 668                                 if (ptr->dev == bucket.inode)
 669                                         data_opts.rewrite_ptrs |= 1U << i;
 670                                 i++;
 671                         }
 672
 673                         ret = bch2_move_extent(&trans, &iter, ctxt, io_opts,
 674                                                bp.btree_id, k, data_opts);
 675                         bch2_trans_iter_exit(&trans, &iter);
 676
 677                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 678                                 continue;
 679                         if (ret == -ENOMEM) {
 680                                 /* memory allocation failure, wait for some IO to finish */
 681                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 682                                 continue;
 683                         }
 684                         if (ret)
 685                                 goto err;
 686
 687                         if (ctxt->rate)
 688                                 bch2_ratelimit_increment(ctxt->rate, k.k->size);
 689                         atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 690                 } else {
 691                         struct btree *b;
 692
 693                         b = bch2_backpointer_get_node(&trans, &iter,
 694                                                 bucket, bp_offset, bp);
 695                         ret = PTR_ERR_OR_ZERO(b);
 696                         if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
 697                                 continue;
 698                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 699                                 continue;
 700                         if (ret)
 701                                 goto err;
 702                         if (!b)
 703                                 continue;
 704
 705                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0);
 706                         bch2_trans_iter_exit(&trans, &iter);
 707
 708                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 709                                 continue;
 710                         if (ret)
 711                                 goto err;
 712
 713                         if (ctxt->rate)
 714                                 bch2_ratelimit_increment(ctxt->rate,
 715                                                          c->opts.btree_node_size >> 9);
 716                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_seen);
 717                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_moved);
 718                 }
 719
 720                 bp_offset++;
 721         }
 722
 723         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
 724                 bch2_trans_unlock(&trans);
 725                 move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
 726                 closure_sync(&ctxt->cl);
 727                 if (!ctxt->write_error)
 728                         lockrestart_do(&trans, verify_bucket_evacuated(&trans, bucket, gen));
 729         }
 730 err:
 731         bch2_trans_exit(&trans);
 732         bch2_bkey_buf_exit(&sk, c);
 733         return ret;
 734 }
 735
 736 int bch2_evacuate_bucket(struct bch_fs *c,
 737                          struct bpos bucket, int gen,
 738                          struct data_update_opts data_opts,
 739                          struct bch_ratelimit *rate,
 740                          struct bch_move_stats *stats,
 741                          struct write_point_specifier wp,
 742                          bool wait_on_copygc)
 743 {
 744         struct moving_context ctxt;
 745         int ret;
 746
 747         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 748         ret = __bch2_evacuate_bucket(&ctxt, bucket, gen, data_opts);
 749         bch2_moving_ctxt_exit(&ctxt);
 750
 751         return ret;
 752 }
 753
 754 typedef bool (*move_btree_pred)(struct bch_fs *, void *,
 755                                 struct btree *, struct bch_io_opts *,
 756                                 struct data_update_opts *);
 757
 758 static int bch2_move_btree(struct bch_fs *c,
 759                            enum btree_id start_btree_id, struct bpos start_pos,
 760                            enum btree_id end_btree_id,   struct bpos end_pos,
 761                            move_btree_pred pred, void *arg,
 762                            struct bch_move_stats *stats)
 763 {
 764         bool kthread = (current->flags & PF_KTHREAD) != 0;
 765         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 766         struct btree_trans trans;
 767         struct btree_iter iter;
 768         struct btree *b;
 769         enum btree_id id;
 770         struct data_update_opts data_opts;
 771         int ret = 0;
 772
 773         bch2_trans_init(&trans, c, 0, 0);
 774         progress_list_add(c, stats);
 775
 776         stats->data_type = BCH_DATA_btree;
 777
 778         for (id = start_btree_id;
 779              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 780              id++) {
 781                 stats->btree_id = id;
 782
 783                 bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
 784                                           BTREE_ITER_PREFETCH);
 785 retry:
 786                 ret = 0;
 787                 while (bch2_trans_begin(&trans),
 788                        (b = bch2_btree_iter_peek_node(&iter)) &&
 789                        !(ret = PTR_ERR_OR_ZERO(b))) {
 790                         if (kthread && kthread_should_stop())
 791                                 break;
 792
 793                         if ((cmp_int(id, end_btree_id) ?:
 794                              bpos_cmp(b->key.k.p, end_pos)) > 0)
 795                                 break;
 796
 797                         stats->pos = iter.pos;
 798
 799                         if (!pred(c, arg, b, &io_opts, &data_opts))
 800                                 goto next;
 801
 802                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
 803                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 804                                 continue;
 805                         if (ret)
 806                                 break;
 807 next:
 808                         bch2_btree_iter_next_node(&iter);
 809                 }
 810                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 811                         goto retry;
 812
 813                 bch2_trans_iter_exit(&trans, &iter);
 814
 815                 if (kthread && kthread_should_stop())
 816                         break;
 817         }
 818
 819         bch2_trans_exit(&trans);
 820
 821         if (ret)
 822                 bch_err(c, "error in %s(): %s", __func__, bch2_err_str(ret));
 823
 824         bch2_btree_interior_updates_flush(c);
 825
 826         progress_list_del(c, stats);
 827         return ret;
 828 }
 829
 830 static bool rereplicate_pred(struct bch_fs *c, void *arg,
 831                              struct bkey_s_c k,
 832                              struct bch_io_opts *io_opts,
 833                              struct data_update_opts *data_opts)
 834 {
 835         unsigned nr_good = bch2_bkey_durability(c, k);
 836         unsigned replicas = bkey_is_btree_ptr(k.k)
 837                 ? c->opts.metadata_replicas
 838                 : io_opts->data_replicas;
 839
 840         if (!nr_good || nr_good >= replicas)
 841                 return false;
 842
 843         data_opts->target               = 0;
 844         data_opts->extra_replicas       = replicas - nr_good;
 845         data_opts->btree_insert_flags   = 0;
 846         return true;
 847 }
 848
 849 static bool migrate_pred(struct bch_fs *c, void *arg,
 850                          struct bkey_s_c k,
 851                          struct bch_io_opts *io_opts,
 852                          struct data_update_opts *data_opts)
 853 {
 854         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 855         const struct bch_extent_ptr *ptr;
 856         struct bch_ioctl_data *op = arg;
 857         unsigned i = 0;
 858
 859         data_opts->rewrite_ptrs         = 0;
 860         data_opts->target               = 0;
 861         data_opts->extra_replicas       = 0;
 862         data_opts->btree_insert_flags   = 0;
 863
 864         bkey_for_each_ptr(ptrs, ptr) {
 865                 if (ptr->dev == op->migrate.dev)
 866                         data_opts->rewrite_ptrs |= 1U << i;
 867                 i++;
 868         }
 869
 870         return data_opts->rewrite_ptrs != 0;
 871 }
 872
 873 static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
 874                                    struct btree *b,
 875                                    struct bch_io_opts *io_opts,
 876                                    struct data_update_opts *data_opts)
 877 {
 878         return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 879 }
 880
 881 static bool migrate_btree_pred(struct bch_fs *c, void *arg,
 882                                struct btree *b,
 883                                struct bch_io_opts *io_opts,
 884                                struct data_update_opts *data_opts)
 885 {
 886         return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 887 }
 888
 889 static bool bformat_needs_redo(struct bkey_format *f)
 890 {
 891         unsigned i;
 892
 893         for (i = 0; i < f->nr_fields; i++) {
 894                 unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
 895                 u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
 896                 u64 field_offset = le64_to_cpu(f->field_offset[i]);
 897
 898                 if (f->bits_per_field[i] > unpacked_bits)
 899                         return true;
 900
 901                 if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
 902                         return true;
 903
 904                 if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
 905                      unpacked_mask) <
 906                     field_offset)
 907                         return true;
 908         }
 909
 910         return false;
 911 }
 912
 913 static bool rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
 914                                    struct btree *b,
 915                                    struct bch_io_opts *io_opts,
 916                                    struct data_update_opts *data_opts)
 917 {
 918         if (b->version_ondisk != c->sb.version ||
 919             btree_node_need_rewrite(b) ||
 920             bformat_needs_redo(&b->format)) {
 921                 data_opts->target               = 0;
 922                 data_opts->extra_replicas       = 0;
 923                 data_opts->btree_insert_flags   = 0;
 924                 return true;
 925         }
 926
 927         return false;
 928 }
 929
 930 int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
 931 {
 932         int ret;
 933
 934         ret = bch2_move_btree(c,
 935                               0,                POS_MIN,
 936                               BTREE_ID_NR,      SPOS_MAX,
 937                               rewrite_old_nodes_pred, c, stats);
 938         if (!ret) {
 939                 mutex_lock(&c->sb_lock);
 940                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 941                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 942                 c->disk_sb.sb->version_min = c->disk_sb.sb->version;
 943                 bch2_write_super(c);
 944                 mutex_unlock(&c->sb_lock);
 945         }
 946
 947         return ret;
 948 }
 949
 950 int bch2_data_job(struct bch_fs *c,
 951                   struct bch_move_stats *stats,
 952                   struct bch_ioctl_data op)
 953 {
 954         int ret = 0;
 955
 956         switch (op.op) {
 957         case BCH_DATA_OP_REREPLICATE:
 958                 bch2_move_stats_init(stats, "rereplicate");
 959                 stats->data_type = BCH_DATA_journal;
 960                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
 961
 962                 ret = bch2_move_btree(c,
 963                                       op.start_btree,   op.start_pos,
 964                                       op.end_btree,     op.end_pos,
 965                                       rereplicate_btree_pred, c, stats) ?: ret;
 966                 ret = bch2_replicas_gc2(c) ?: ret;
 967
 968                 ret = bch2_move_data(c,
 969                                      op.start_btree,    op.start_pos,
 970                                      op.end_btree,      op.end_pos,
 971                                      NULL,
 972                                      stats,
 973                                      writepoint_hashed((unsigned long) current),
 974                                      true,
 975                                      rereplicate_pred, c) ?: ret;
 976                 ret = bch2_replicas_gc2(c) ?: ret;
 977                 break;
 978         case BCH_DATA_OP_MIGRATE:
 979                 if (op.migrate.dev >= c->sb.nr_devices)
 980                         return -EINVAL;
 981
 982                 bch2_move_stats_init(stats, "migrate");
 983                 stats->data_type = BCH_DATA_journal;
 984                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
 985
 986                 ret = bch2_move_btree(c,
 987                                       op.start_btree,   op.start_pos,
 988                                       op.end_btree,     op.end_pos,
 989                                       migrate_btree_pred, &op, stats) ?: ret;
 990                 ret = bch2_replicas_gc2(c) ?: ret;
 991
 992                 ret = bch2_move_data(c,
 993                                      op.start_btree,    op.start_pos,
 994                                      op.end_btree,      op.end_pos,
 995                                      NULL,
 996                                      stats,
 997                                      writepoint_hashed((unsigned long) current),
 998                                      true,
 999                                      migrate_pred, &op) ?: ret;
1000                 ret = bch2_replicas_gc2(c) ?: ret;
1001                 break;
1002         case BCH_DATA_OP_REWRITE_OLD_NODES:
1003                 bch2_move_stats_init(stats, "rewrite_old_nodes");
1004                 ret = bch2_scan_old_btree_nodes(c, stats);
1005                 break;
1006         default:
1007                 ret = -EINVAL;
1008         }
1009
1010         return ret;
1011 }