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         /*
 265          * Before memory allocations & taking nocow locks in
 266          * bch2_data_update_init():
 267          */
 268         bch2_trans_unlock(trans);
 269
 270         /* write path might have to decompress data: */
 271         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 272                 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
 273
 274         pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
 275         io = kzalloc(sizeof(struct moving_io) +
 276                      sizeof(struct bio_vec) * pages, GFP_KERNEL);
 277         if (!io)
 278                 goto err;
 279
 280         io->write.ctxt          = ctxt;
 281         io->read_sectors        = k.k->size;
 282         io->write_sectors       = k.k->size;
 283
 284         bio_init(&io->write.op.wbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 285         bio_set_prio(&io->write.op.wbio.bio,
 286                      IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 287
 288         if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
 289                                  GFP_KERNEL))
 290                 goto err_free;
 291
 292         io->rbio.c              = c;
 293         io->rbio.opts           = io_opts;
 294         bio_init(&io->rbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 295         io->rbio.bio.bi_vcnt = pages;
 296         bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 297         io->rbio.bio.bi_iter.bi_size = sectors << 9;
 298
 299         bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
 300         io->rbio.bio.bi_iter.bi_sector  = bkey_start_offset(k.k);
 301         io->rbio.bio.bi_end_io          = move_read_endio;
 302
 303         ret = bch2_data_update_init(c, &io->write, ctxt->wp, io_opts,
 304                                     data_opts, btree_id, k);
 305         if (ret && ret != -BCH_ERR_unwritten_extent_update)
 306                 goto err_free_pages;
 307
 308         io->write.ctxt = ctxt;
 309         io->write.op.end_io = move_write_done;
 310
 311         atomic64_inc(&ctxt->stats->keys_moved);
 312         atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
 313
 314         if (ret == -BCH_ERR_unwritten_extent_update) {
 315                 bch2_update_unwritten_extent(trans, &io->write);
 316                 move_free(io);
 317                 return 0;
 318         }
 319
 320         BUG_ON(ret);
 321
 322         this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
 323         this_cpu_add(c->counters[BCH_COUNTER_move_extent_read], k.k->size);
 324         trace_move_extent_read(k.k);
 325
 326         atomic_add(io->read_sectors, &ctxt->read_sectors);
 327         list_add_tail(&io->list, &ctxt->reads);
 328
 329         /*
 330          * dropped by move_read_endio() - guards against use after free of
 331          * ctxt when doing wakeup
 332          */
 333         closure_get(&ctxt->cl);
 334         bch2_read_extent(trans, &io->rbio,
 335                          bkey_start_pos(k.k),
 336                          btree_id, k, 0,
 337                          BCH_READ_NODECODE|
 338                          BCH_READ_LAST_FRAGMENT);
 339         return 0;
 340 err_free_pages:
 341         bio_free_pages(&io->write.op.wbio.bio);
 342 err_free:
 343         kfree(io);
 344 err:
 345         percpu_ref_put(&c->writes);
 346         trace_and_count(c, move_extent_alloc_mem_fail, k.k);
 347         return ret;
 348 }
 349
 350 static int lookup_inode(struct btree_trans *trans, struct bpos pos,
 351                         struct bch_inode_unpacked *inode)
 352 {
 353         struct btree_iter iter;
 354         struct bkey_s_c k;
 355         int ret;
 356
 357         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
 358                              BTREE_ITER_ALL_SNAPSHOTS);
 359         k = bch2_btree_iter_peek(&iter);
 360         ret = bkey_err(k);
 361         if (ret)
 362                 goto err;
 363
 364         if (!k.k || !bkey_eq(k.k->p, pos)) {
 365                 ret = -ENOENT;
 366                 goto err;
 367         }
 368
 369         ret = bkey_is_inode(k.k) ? 0 : -EIO;
 370         if (ret)
 371                 goto err;
 372
 373         ret = bch2_inode_unpack(k, inode);
 374         if (ret)
 375                 goto err;
 376 err:
 377         bch2_trans_iter_exit(trans, &iter);
 378         return ret;
 379 }
 380
 381 static int move_ratelimit(struct btree_trans *trans,
 382                           struct moving_context *ctxt)
 383 {
 384         struct bch_fs *c = trans->c;
 385         u64 delay;
 386
 387         if (ctxt->wait_on_copygc) {
 388                 bch2_trans_unlock(trans);
 389                 wait_event_killable(c->copygc_running_wq,
 390                                     !c->copygc_running ||
 391                                     kthread_should_stop());
 392         }
 393
 394         do {
 395                 delay = ctxt->rate ? bch2_ratelimit_delay(ctxt->rate) : 0;
 396
 397                 if (delay) {
 398                         bch2_trans_unlock(trans);
 399                         set_current_state(TASK_INTERRUPTIBLE);
 400                 }
 401
 402                 if ((current->flags & PF_KTHREAD) && kthread_should_stop()) {
 403                         __set_current_state(TASK_RUNNING);
 404                         return 1;
 405                 }
 406
 407                 if (delay)
 408                         schedule_timeout(delay);
 409
 410                 if (unlikely(freezing(current))) {
 411                         move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
 412                         try_to_freeze();
 413                 }
 414         } while (delay);
 415
 416         move_ctxt_wait_event(ctxt, trans,
 417                 atomic_read(&ctxt->write_sectors) <
 418                 c->opts.move_bytes_in_flight >> 9);
 419
 420         move_ctxt_wait_event(ctxt, trans,
 421                 atomic_read(&ctxt->read_sectors) <
 422                 c->opts.move_bytes_in_flight >> 9);
 423
 424         return 0;
 425 }
 426
 427 static int move_get_io_opts(struct btree_trans *trans,
 428                             struct bch_io_opts *io_opts,
 429                             struct bkey_s_c k, u64 *cur_inum)
 430 {
 431         struct bch_inode_unpacked inode;
 432         int ret;
 433
 434         if (*cur_inum == k.k->p.inode)
 435                 return 0;
 436
 437         ret = lookup_inode(trans,
 438                            SPOS(0, k.k->p.inode, k.k->p.snapshot),
 439                            &inode);
 440         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 441                 return ret;
 442
 443         if (!ret)
 444                 bch2_inode_opts_get(io_opts, trans->c, &inode);
 445         else
 446                 *io_opts = bch2_opts_to_inode_opts(trans->c->opts);
 447         *cur_inum = k.k->p.inode;
 448         return 0;
 449 }
 450
 451 static int __bch2_move_data(struct moving_context *ctxt,
 452                             struct bpos start,
 453                             struct bpos end,
 454                             move_pred_fn pred, void *arg,
 455                             enum btree_id btree_id)
 456 {
 457         struct bch_fs *c = ctxt->c;
 458         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 459         struct bkey_buf sk;
 460         struct btree_trans trans;
 461         struct btree_iter iter;
 462         struct bkey_s_c k;
 463         struct data_update_opts data_opts;
 464         u64 cur_inum = U64_MAX;
 465         int ret = 0, ret2;
 466
 467         bch2_bkey_buf_init(&sk);
 468         bch2_trans_init(&trans, c, 0, 0);
 469
 470         ctxt->stats->data_type  = BCH_DATA_user;
 471         ctxt->stats->btree_id   = btree_id;
 472         ctxt->stats->pos        = start;
 473
 474         bch2_trans_iter_init(&trans, &iter, btree_id, start,
 475                              BTREE_ITER_PREFETCH|
 476                              BTREE_ITER_ALL_SNAPSHOTS);
 477
 478         if (ctxt->rate)
 479                 bch2_ratelimit_reset(ctxt->rate);
 480
 481         while (!move_ratelimit(&trans, ctxt)) {
 482                 bch2_trans_begin(&trans);
 483
 484                 k = bch2_btree_iter_peek(&iter);
 485                 if (!k.k)
 486                         break;
 487
 488                 ret = bkey_err(k);
 489                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 490                         continue;
 491                 if (ret)
 492                         break;
 493
 494                 if (bkey_ge(bkey_start_pos(k.k), end))
 495                         break;
 496
 497                 ctxt->stats->pos = iter.pos;
 498
 499                 if (!bkey_extent_is_direct_data(k.k))
 500                         goto next_nondata;
 501
 502                 ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 503                 if (ret)
 504                         continue;
 505
 506                 memset(&data_opts, 0, sizeof(data_opts));
 507                 if (!pred(c, arg, k, &io_opts, &data_opts))
 508                         goto next;
 509
 510                 /*
 511                  * The iterator gets unlocked by __bch2_read_extent - need to
 512                  * save a copy of @k elsewhere:
 513                  */
 514                 bch2_bkey_buf_reassemble(&sk, c, k);
 515                 k = bkey_i_to_s_c(sk.k);
 516                 bch2_trans_unlock(&trans);
 517
 518                 ret2 = bch2_move_extent(&trans, &iter, ctxt, io_opts,
 519                                         btree_id, k, data_opts);
 520                 if (ret2) {
 521                         if (bch2_err_matches(ret2, BCH_ERR_transaction_restart))
 522                                 continue;
 523
 524                         if (ret2 == -ENOMEM) {
 525                                 /* memory allocation failure, wait for some IO to finish */
 526                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 527                                 continue;
 528                         }
 529
 530                         /* XXX signal failure */
 531                         goto next;
 532                 }
 533
 534                 if (ctxt->rate)
 535                         bch2_ratelimit_increment(ctxt->rate, k.k->size);
 536 next:
 537                 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 538 next_nondata:
 539                 bch2_btree_iter_advance(&iter);
 540         }
 541
 542         bch2_trans_iter_exit(&trans, &iter);
 543         bch2_trans_exit(&trans);
 544         bch2_bkey_buf_exit(&sk, c);
 545
 546         return ret;
 547 }
 548
 549 int bch2_move_data(struct bch_fs *c,
 550                    enum btree_id start_btree_id, struct bpos start_pos,
 551                    enum btree_id end_btree_id,   struct bpos end_pos,
 552                    struct bch_ratelimit *rate,
 553                    struct bch_move_stats *stats,
 554                    struct write_point_specifier wp,
 555                    bool wait_on_copygc,
 556                    move_pred_fn pred, void *arg)
 557 {
 558         struct moving_context ctxt;
 559         enum btree_id id;
 560         int ret;
 561
 562         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 563
 564         for (id = start_btree_id;
 565              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 566              id++) {
 567                 stats->btree_id = id;
 568
 569                 if (id != BTREE_ID_extents &&
 570                     id != BTREE_ID_reflink)
 571                         continue;
 572
 573                 ret = __bch2_move_data(&ctxt,
 574                                        id == start_btree_id ? start_pos : POS_MIN,
 575                                        id == end_btree_id   ? end_pos   : POS_MAX,
 576                                        pred, arg, id);
 577                 if (ret)
 578                         break;
 579         }
 580
 581         bch2_moving_ctxt_exit(&ctxt);
 582
 583         return ret;
 584 }
 585
 586 static int verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
 587 {
 588         struct bch_fs *c = trans->c;
 589         struct btree_iter iter;
 590         struct bkey_s_c k;
 591         struct printbuf buf = PRINTBUF;
 592         struct bch_backpointer bp;
 593         u64 bp_offset = 0;
 594         int ret;
 595
 596         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
 597                              bucket, BTREE_ITER_CACHED);
 598 again:
 599         k = bch2_btree_iter_peek_slot(&iter);
 600         ret = bkey_err(k);
 601
 602         if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
 603                 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
 604
 605                 if (a.v->gen == gen &&
 606                     a.v->dirty_sectors) {
 607                         if (a.v->data_type == BCH_DATA_btree) {
 608                                 bch2_trans_unlock(trans);
 609                                 if (bch2_btree_interior_updates_flush(c))
 610                                         goto again;
 611                                 goto failed_to_evacuate;
 612                         }
 613                 }
 614         }
 615
 616         bch2_trans_iter_exit(trans, &iter);
 617         return ret;
 618 failed_to_evacuate:
 619         bch2_trans_iter_exit(trans, &iter);
 620
 621         prt_printf(&buf, bch2_log_msg(c, "failed to evacuate bucket "));
 622         bch2_bkey_val_to_text(&buf, c, k);
 623
 624         while (1) {
 625                 bch2_trans_begin(trans);
 626
 627                 ret = bch2_get_next_backpointer(trans, bucket, gen,
 628                                                 &bp_offset, &bp,
 629                                                 BTREE_ITER_CACHED);
 630                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 631                         continue;
 632                 if (ret)
 633                         break;
 634                 if (bp_offset == U64_MAX)
 635                         break;
 636
 637                 k = bch2_backpointer_get_key(trans, &iter,
 638                                              bucket, bp_offset, bp);
 639                 ret = bkey_err(k);
 640                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 641                         continue;
 642                 if (ret)
 643                         break;
 644                 if (!k.k)
 645                         continue;
 646                 prt_newline(&buf);
 647                 bch2_bkey_val_to_text(&buf, c, k);
 648                 bch2_trans_iter_exit(trans, &iter);
 649         }
 650
 651         bch2_print_string_as_lines(KERN_ERR, buf.buf);
 652         printbuf_exit(&buf);
 653         return 0;
 654 }
 655
 656 int __bch2_evacuate_bucket(struct moving_context *ctxt,
 657                            struct bpos bucket, int gen,
 658                            struct data_update_opts _data_opts)
 659 {
 660         struct bch_fs *c = ctxt->c;
 661         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 662         struct btree_trans trans;
 663         struct btree_iter iter;
 664         struct bkey_buf sk;
 665         struct bch_backpointer bp;
 666         struct data_update_opts data_opts;
 667         u64 bp_offset = 0, cur_inum = U64_MAX;
 668         int ret = 0;
 669
 670         bch2_bkey_buf_init(&sk);
 671         bch2_trans_init(&trans, c, 0, 0);
 672
 673         while (!(ret = move_ratelimit(&trans, ctxt))) {
 674                 bch2_trans_begin(&trans);
 675
 676                 ret = bch2_get_next_backpointer(&trans, bucket, gen,
 677                                                 &bp_offset, &bp,
 678                                                 BTREE_ITER_CACHED);
 679                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 680                         continue;
 681                 if (ret)
 682                         goto err;
 683                 if (bp_offset == U64_MAX)
 684                         break;
 685
 686                 if (!bp.level) {
 687                         const struct bch_extent_ptr *ptr;
 688                         struct bkey_s_c k;
 689                         unsigned i = 0;
 690
 691                         k = bch2_backpointer_get_key(&trans, &iter,
 692                                                 bucket, bp_offset, bp);
 693                         ret = bkey_err(k);
 694                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 695                                 continue;
 696                         if (ret)
 697                                 goto err;
 698                         if (!k.k)
 699                                 continue;
 700
 701                         bch2_bkey_buf_reassemble(&sk, c, k);
 702                         k = bkey_i_to_s_c(sk.k);
 703
 704                         ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 705                         if (ret) {
 706                                 bch2_trans_iter_exit(&trans, &iter);
 707                                 continue;
 708                         }
 709
 710                         data_opts = _data_opts;
 711                         data_opts.target        = io_opts.background_target;
 712                         data_opts.rewrite_ptrs = 0;
 713
 714                         bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
 715                                 if (ptr->dev == bucket.inode)
 716                                         data_opts.rewrite_ptrs |= 1U << i;
 717                                 i++;
 718                         }
 719
 720                         ret = bch2_move_extent(&trans, &iter, ctxt, io_opts,
 721                                                bp.btree_id, k, data_opts);
 722                         bch2_trans_iter_exit(&trans, &iter);
 723
 724                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 725                                 continue;
 726                         if (ret == -ENOMEM) {
 727                                 /* memory allocation failure, wait for some IO to finish */
 728                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 729                                 continue;
 730                         }
 731                         if (ret)
 732                                 goto err;
 733
 734                         if (ctxt->rate)
 735                                 bch2_ratelimit_increment(ctxt->rate, k.k->size);
 736                         atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 737                 } else {
 738                         struct btree *b;
 739
 740                         b = bch2_backpointer_get_node(&trans, &iter,
 741                                                 bucket, bp_offset, bp);
 742                         ret = PTR_ERR_OR_ZERO(b);
 743                         if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
 744                                 continue;
 745                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 746                                 continue;
 747                         if (ret)
 748                                 goto err;
 749                         if (!b)
 750                                 continue;
 751
 752                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0);
 753                         bch2_trans_iter_exit(&trans, &iter);
 754
 755                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 756                                 continue;
 757                         if (ret)
 758                                 goto err;
 759
 760                         if (ctxt->rate)
 761                                 bch2_ratelimit_increment(ctxt->rate,
 762                                                          c->opts.btree_node_size >> 9);
 763                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_seen);
 764                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_moved);
 765                 }
 766
 767                 bp_offset++;
 768         }
 769
 770         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
 771                 bch2_trans_unlock(&trans);
 772                 move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
 773                 closure_sync(&ctxt->cl);
 774                 if (!ctxt->write_error)
 775                         lockrestart_do(&trans, verify_bucket_evacuated(&trans, bucket, gen));
 776         }
 777 err:
 778         bch2_trans_exit(&trans);
 779         bch2_bkey_buf_exit(&sk, c);
 780         return ret;
 781 }
 782
 783 int bch2_evacuate_bucket(struct bch_fs *c,
 784                          struct bpos bucket, int gen,
 785                          struct data_update_opts data_opts,
 786                          struct bch_ratelimit *rate,
 787                          struct bch_move_stats *stats,
 788                          struct write_point_specifier wp,
 789                          bool wait_on_copygc)
 790 {
 791         struct moving_context ctxt;
 792         int ret;
 793
 794         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 795         ret = __bch2_evacuate_bucket(&ctxt, bucket, gen, data_opts);
 796         bch2_moving_ctxt_exit(&ctxt);
 797
 798         return ret;
 799 }
 800
 801 typedef bool (*move_btree_pred)(struct bch_fs *, void *,
 802                                 struct btree *, struct bch_io_opts *,
 803                                 struct data_update_opts *);
 804
 805 static int bch2_move_btree(struct bch_fs *c,
 806                            enum btree_id start_btree_id, struct bpos start_pos,
 807                            enum btree_id end_btree_id,   struct bpos end_pos,
 808                            move_btree_pred pred, void *arg,
 809                            struct bch_move_stats *stats)
 810 {
 811         bool kthread = (current->flags & PF_KTHREAD) != 0;
 812         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 813         struct btree_trans trans;
 814         struct btree_iter iter;
 815         struct btree *b;
 816         enum btree_id id;
 817         struct data_update_opts data_opts;
 818         int ret = 0;
 819
 820         bch2_trans_init(&trans, c, 0, 0);
 821         progress_list_add(c, stats);
 822
 823         stats->data_type = BCH_DATA_btree;
 824
 825         for (id = start_btree_id;
 826              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 827              id++) {
 828                 stats->btree_id = id;
 829
 830                 bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
 831                                           BTREE_ITER_PREFETCH);
 832 retry:
 833                 ret = 0;
 834                 while (bch2_trans_begin(&trans),
 835                        (b = bch2_btree_iter_peek_node(&iter)) &&
 836                        !(ret = PTR_ERR_OR_ZERO(b))) {
 837                         if (kthread && kthread_should_stop())
 838                                 break;
 839
 840                         if ((cmp_int(id, end_btree_id) ?:
 841                              bpos_cmp(b->key.k.p, end_pos)) > 0)
 842                                 break;
 843
 844                         stats->pos = iter.pos;
 845
 846                         if (!pred(c, arg, b, &io_opts, &data_opts))
 847                                 goto next;
 848
 849                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
 850                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 851                                 continue;
 852                         if (ret)
 853                                 break;
 854 next:
 855                         bch2_btree_iter_next_node(&iter);
 856                 }
 857                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 858                         goto retry;
 859
 860                 bch2_trans_iter_exit(&trans, &iter);
 861
 862                 if (kthread && kthread_should_stop())
 863                         break;
 864         }
 865
 866         bch2_trans_exit(&trans);
 867
 868         if (ret)
 869                 bch_err(c, "error in %s(): %s", __func__, bch2_err_str(ret));
 870
 871         bch2_btree_interior_updates_flush(c);
 872
 873         progress_list_del(c, stats);
 874         return ret;
 875 }
 876
 877 static bool rereplicate_pred(struct bch_fs *c, void *arg,
 878                              struct bkey_s_c k,
 879                              struct bch_io_opts *io_opts,
 880                              struct data_update_opts *data_opts)
 881 {
 882         unsigned nr_good = bch2_bkey_durability(c, k);
 883         unsigned replicas = bkey_is_btree_ptr(k.k)
 884                 ? c->opts.metadata_replicas
 885                 : io_opts->data_replicas;
 886
 887         if (!nr_good || nr_good >= replicas)
 888                 return false;
 889
 890         data_opts->target               = 0;
 891         data_opts->extra_replicas       = replicas - nr_good;
 892         data_opts->btree_insert_flags   = 0;
 893         return true;
 894 }
 895
 896 static bool migrate_pred(struct bch_fs *c, void *arg,
 897                          struct bkey_s_c k,
 898                          struct bch_io_opts *io_opts,
 899                          struct data_update_opts *data_opts)
 900 {
 901         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 902         const struct bch_extent_ptr *ptr;
 903         struct bch_ioctl_data *op = arg;
 904         unsigned i = 0;
 905
 906         data_opts->rewrite_ptrs         = 0;
 907         data_opts->target               = 0;
 908         data_opts->extra_replicas       = 0;
 909         data_opts->btree_insert_flags   = 0;
 910
 911         bkey_for_each_ptr(ptrs, ptr) {
 912                 if (ptr->dev == op->migrate.dev)
 913                         data_opts->rewrite_ptrs |= 1U << i;
 914                 i++;
 915         }
 916
 917         return data_opts->rewrite_ptrs != 0;
 918 }
 919
 920 static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
 921                                    struct btree *b,
 922                                    struct bch_io_opts *io_opts,
 923                                    struct data_update_opts *data_opts)
 924 {
 925         return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 926 }
 927
 928 static bool migrate_btree_pred(struct bch_fs *c, void *arg,
 929                                struct btree *b,
 930                                struct bch_io_opts *io_opts,
 931                                struct data_update_opts *data_opts)
 932 {
 933         return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 934 }
 935
 936 static bool bformat_needs_redo(struct bkey_format *f)
 937 {
 938         unsigned i;
 939
 940         for (i = 0; i < f->nr_fields; i++) {
 941                 unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
 942                 u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
 943                 u64 field_offset = le64_to_cpu(f->field_offset[i]);
 944
 945                 if (f->bits_per_field[i] > unpacked_bits)
 946                         return true;
 947
 948                 if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
 949                         return true;
 950
 951                 if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
 952                      unpacked_mask) <
 953                     field_offset)
 954                         return true;
 955         }
 956
 957         return false;
 958 }
 959
 960 static bool rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
 961                                    struct btree *b,
 962                                    struct bch_io_opts *io_opts,
 963                                    struct data_update_opts *data_opts)
 964 {
 965         if (b->version_ondisk != c->sb.version ||
 966             btree_node_need_rewrite(b) ||
 967             bformat_needs_redo(&b->format)) {
 968                 data_opts->target               = 0;
 969                 data_opts->extra_replicas       = 0;
 970                 data_opts->btree_insert_flags   = 0;
 971                 return true;
 972         }
 973
 974         return false;
 975 }
 976
 977 int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
 978 {
 979         int ret;
 980
 981         ret = bch2_move_btree(c,
 982                               0,                POS_MIN,
 983                               BTREE_ID_NR,      SPOS_MAX,
 984                               rewrite_old_nodes_pred, c, stats);
 985         if (!ret) {
 986                 mutex_lock(&c->sb_lock);
 987                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 988                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 989                 c->disk_sb.sb->version_min = c->disk_sb.sb->version;
 990                 bch2_write_super(c);
 991                 mutex_unlock(&c->sb_lock);
 992         }
 993
 994         return ret;
 995 }
 996
 997 int bch2_data_job(struct bch_fs *c,
 998                   struct bch_move_stats *stats,
 999                   struct bch_ioctl_data op)
1000 {
1001         int ret = 0;
1002
1003         switch (op.op) {
1004         case BCH_DATA_OP_REREPLICATE:
1005                 bch2_move_stats_init(stats, "rereplicate");
1006                 stats->data_type = BCH_DATA_journal;
1007                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
1008
1009                 ret = bch2_move_btree(c,
1010                                       op.start_btree,   op.start_pos,
1011                                       op.end_btree,     op.end_pos,
1012                                       rereplicate_btree_pred, c, stats) ?: ret;
1013                 ret = bch2_replicas_gc2(c) ?: ret;
1014
1015                 ret = bch2_move_data(c,
1016                                      op.start_btree,    op.start_pos,
1017                                      op.end_btree,      op.end_pos,
1018                                      NULL,
1019                                      stats,
1020                                      writepoint_hashed((unsigned long) current),
1021                                      true,
1022                                      rereplicate_pred, c) ?: ret;
1023                 ret = bch2_replicas_gc2(c) ?: ret;
1024                 break;
1025         case BCH_DATA_OP_MIGRATE:
1026                 if (op.migrate.dev >= c->sb.nr_devices)
1027                         return -EINVAL;
1028
1029                 bch2_move_stats_init(stats, "migrate");
1030                 stats->data_type = BCH_DATA_journal;
1031                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
1032
1033                 ret = bch2_move_btree(c,
1034                                       op.start_btree,   op.start_pos,
1035                                       op.end_btree,     op.end_pos,
1036                                       migrate_btree_pred, &op, stats) ?: ret;
1037                 ret = bch2_replicas_gc2(c) ?: ret;
1038
1039                 ret = bch2_move_data(c,
1040                                      op.start_btree,    op.start_pos,
1041                                      op.end_btree,      op.end_pos,
1042                                      NULL,
1043                                      stats,
1044                                      writepoint_hashed((unsigned long) current),
1045                                      true,
1046                                      migrate_pred, &op) ?: ret;
1047                 ret = bch2_replicas_gc2(c) ?: ret;
1048                 break;
1049         case BCH_DATA_OP_REWRITE_OLD_NODES:
1050                 bch2_move_stats_init(stats, "rewrite_old_nodes");
1051                 ret = bch2_scan_old_btree_nodes(c, stats);
1052                 break;
1053         default:
1054                 ret = -EINVAL;
1055         }
1056
1057         return ret;
1058 }