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_cmp(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         *io_opts = bch2_opts_to_inode_opts(trans->c->opts);
 438
 439         ret = lookup_inode(trans,
 440                            SPOS(0, k.k->p.inode, k.k->p.snapshot),
 441                            &inode);
 442         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 443                 return ret;
 444
 445         if (!ret)
 446                 bch2_io_opts_apply(io_opts, bch2_inode_opts_get(&inode));
 447
 448         *cur_inum = k.k->p.inode;
 449         return 0;
 450 }
 451
 452 static int __bch2_move_data(struct moving_context *ctxt,
 453                             struct bpos start,
 454                             struct bpos end,
 455                             move_pred_fn pred, void *arg,
 456                             enum btree_id btree_id)
 457 {
 458         struct bch_fs *c = ctxt->c;
 459         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 460         struct bkey_buf sk;
 461         struct btree_trans trans;
 462         struct btree_iter iter;
 463         struct bkey_s_c k;
 464         struct data_update_opts data_opts;
 465         u64 cur_inum = U64_MAX;
 466         int ret = 0, ret2;
 467
 468         bch2_bkey_buf_init(&sk);
 469         bch2_trans_init(&trans, c, 0, 0);
 470
 471         ctxt->stats->data_type  = BCH_DATA_user;
 472         ctxt->stats->btree_id   = btree_id;
 473         ctxt->stats->pos        = start;
 474
 475         bch2_trans_iter_init(&trans, &iter, btree_id, start,
 476                              BTREE_ITER_PREFETCH|
 477                              BTREE_ITER_ALL_SNAPSHOTS);
 478
 479         if (ctxt->rate)
 480                 bch2_ratelimit_reset(ctxt->rate);
 481
 482         while (!move_ratelimit(&trans, ctxt)) {
 483                 bch2_trans_begin(&trans);
 484
 485                 k = bch2_btree_iter_peek(&iter);
 486                 if (!k.k)
 487                         break;
 488
 489                 ret = bkey_err(k);
 490                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 491                         continue;
 492                 if (ret)
 493                         break;
 494
 495                 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
 496                         break;
 497
 498                 ctxt->stats->pos = iter.pos;
 499
 500                 if (!bkey_extent_is_direct_data(k.k))
 501                         goto next_nondata;
 502
 503                 ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 504                 if (ret)
 505                         continue;
 506
 507                 memset(&data_opts, 0, sizeof(data_opts));
 508                 if (!pred(c, arg, k, &io_opts, &data_opts))
 509                         goto next;
 510
 511                 /*
 512                  * The iterator gets unlocked by __bch2_read_extent - need to
 513                  * save a copy of @k elsewhere:
 514                  */
 515                 bch2_bkey_buf_reassemble(&sk, c, k);
 516                 k = bkey_i_to_s_c(sk.k);
 517                 bch2_trans_unlock(&trans);
 518
 519                 ret2 = bch2_move_extent(&trans, &iter, ctxt, io_opts,
 520                                         btree_id, k, data_opts);
 521                 if (ret2) {
 522                         if (bch2_err_matches(ret2, BCH_ERR_transaction_restart))
 523                                 continue;
 524
 525                         if (ret2 == -ENOMEM) {
 526                                 /* memory allocation failure, wait for some IO to finish */
 527                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 528                                 continue;
 529                         }
 530
 531                         /* XXX signal failure */
 532                         goto next;
 533                 }
 534
 535                 if (ctxt->rate)
 536                         bch2_ratelimit_increment(ctxt->rate, k.k->size);
 537 next:
 538                 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 539 next_nondata:
 540                 bch2_btree_iter_advance(&iter);
 541         }
 542
 543         bch2_trans_iter_exit(&trans, &iter);
 544         bch2_trans_exit(&trans);
 545         bch2_bkey_buf_exit(&sk, c);
 546
 547         return ret;
 548 }
 549
 550 int bch2_move_data(struct bch_fs *c,
 551                    enum btree_id start_btree_id, struct bpos start_pos,
 552                    enum btree_id end_btree_id,   struct bpos end_pos,
 553                    struct bch_ratelimit *rate,
 554                    struct bch_move_stats *stats,
 555                    struct write_point_specifier wp,
 556                    bool wait_on_copygc,
 557                    move_pred_fn pred, void *arg)
 558 {
 559         struct moving_context ctxt;
 560         enum btree_id id;
 561         int ret;
 562
 563         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 564
 565         for (id = start_btree_id;
 566              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 567              id++) {
 568                 stats->btree_id = id;
 569
 570                 if (id != BTREE_ID_extents &&
 571                     id != BTREE_ID_reflink)
 572                         continue;
 573
 574                 ret = __bch2_move_data(&ctxt,
 575                                        id == start_btree_id ? start_pos : POS_MIN,
 576                                        id == end_btree_id   ? end_pos   : POS_MAX,
 577                                        pred, arg, id);
 578                 if (ret)
 579                         break;
 580         }
 581
 582         bch2_moving_ctxt_exit(&ctxt);
 583
 584         return ret;
 585 }
 586
 587 static int verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
 588 {
 589         struct bch_fs *c = trans->c;
 590         struct btree_iter iter;
 591         struct bkey_s_c k;
 592         struct printbuf buf = PRINTBUF;
 593         struct bch_backpointer bp;
 594         u64 bp_offset = 0;
 595         int ret;
 596
 597         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
 598                              bucket, BTREE_ITER_CACHED);
 599 again:
 600         k = bch2_btree_iter_peek_slot(&iter);
 601         ret = bkey_err(k);
 602
 603         if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
 604                 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
 605
 606                 if (a.v->gen == gen &&
 607                     a.v->dirty_sectors) {
 608                         if (a.v->data_type == BCH_DATA_btree) {
 609                                 bch2_trans_unlock(trans);
 610                                 if (bch2_btree_interior_updates_flush(c))
 611                                         goto again;
 612                                 goto failed_to_evacuate;
 613                         }
 614                 }
 615         }
 616
 617         bch2_trans_iter_exit(trans, &iter);
 618         return ret;
 619 failed_to_evacuate:
 620         bch2_trans_iter_exit(trans, &iter);
 621
 622         prt_printf(&buf, bch2_log_msg(c, "failed to evacuate bucket "));
 623         bch2_bkey_val_to_text(&buf, c, k);
 624
 625         while (1) {
 626                 bch2_trans_begin(trans);
 627
 628                 ret = bch2_get_next_backpointer(trans, bucket, gen,
 629                                                 &bp_offset, &bp,
 630                                                 BTREE_ITER_CACHED);
 631                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 632                         continue;
 633                 if (ret)
 634                         break;
 635                 if (bp_offset == U64_MAX)
 636                         break;
 637
 638                 k = bch2_backpointer_get_key(trans, &iter,
 639                                              bucket, bp_offset, bp);
 640                 ret = bkey_err(k);
 641                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 642                         continue;
 643                 if (ret)
 644                         break;
 645                 if (!k.k)
 646                         continue;
 647                 prt_newline(&buf);
 648                 bch2_bkey_val_to_text(&buf, c, k);
 649                 bch2_trans_iter_exit(trans, &iter);
 650         }
 651
 652         bch2_print_string_as_lines(KERN_ERR, buf.buf);
 653         printbuf_exit(&buf);
 654         return 0;
 655 }
 656
 657 int __bch2_evacuate_bucket(struct moving_context *ctxt,
 658                            struct bpos bucket, int gen,
 659                            struct data_update_opts _data_opts)
 660 {
 661         struct bch_fs *c = ctxt->c;
 662         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 663         struct btree_trans trans;
 664         struct btree_iter iter;
 665         struct bkey_buf sk;
 666         struct bch_backpointer bp;
 667         struct data_update_opts data_opts;
 668         u64 bp_offset = 0, cur_inum = U64_MAX;
 669         int ret = 0;
 670
 671         bch2_bkey_buf_init(&sk);
 672         bch2_trans_init(&trans, c, 0, 0);
 673
 674         while (!(ret = move_ratelimit(&trans, ctxt))) {
 675                 bch2_trans_begin(&trans);
 676
 677                 ret = bch2_get_next_backpointer(&trans, bucket, gen,
 678                                                 &bp_offset, &bp,
 679                                                 BTREE_ITER_CACHED);
 680                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 681                         continue;
 682                 if (ret)
 683                         goto err;
 684                 if (bp_offset == U64_MAX)
 685                         break;
 686
 687                 if (!bp.level) {
 688                         const struct bch_extent_ptr *ptr;
 689                         struct bkey_s_c k;
 690                         unsigned i = 0;
 691
 692                         k = bch2_backpointer_get_key(&trans, &iter,
 693                                                 bucket, bp_offset, bp);
 694                         ret = bkey_err(k);
 695                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 696                                 continue;
 697                         if (ret)
 698                                 goto err;
 699                         if (!k.k)
 700                                 continue;
 701
 702                         bch2_bkey_buf_reassemble(&sk, c, k);
 703                         k = bkey_i_to_s_c(sk.k);
 704
 705                         ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 706                         if (ret) {
 707                                 bch2_trans_iter_exit(&trans, &iter);
 708                                 continue;
 709                         }
 710
 711                         data_opts = _data_opts;
 712                         data_opts.target        = io_opts.background_target;
 713                         data_opts.rewrite_ptrs = 0;
 714
 715                         bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
 716                                 if (ptr->dev == bucket.inode)
 717                                         data_opts.rewrite_ptrs |= 1U << i;
 718                                 i++;
 719                         }
 720
 721                         ret = bch2_move_extent(&trans, &iter, ctxt, io_opts,
 722                                                bp.btree_id, k, data_opts);
 723                         bch2_trans_iter_exit(&trans, &iter);
 724
 725                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 726                                 continue;
 727                         if (ret == -ENOMEM) {
 728                                 /* memory allocation failure, wait for some IO to finish */
 729                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 730                                 continue;
 731                         }
 732                         if (ret)
 733                                 goto err;
 734
 735                         if (ctxt->rate)
 736                                 bch2_ratelimit_increment(ctxt->rate, k.k->size);
 737                         atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 738                 } else {
 739                         struct btree *b;
 740
 741                         b = bch2_backpointer_get_node(&trans, &iter,
 742                                                 bucket, bp_offset, bp);
 743                         ret = PTR_ERR_OR_ZERO(b);
 744                         if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
 745                                 continue;
 746                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 747                                 continue;
 748                         if (ret)
 749                                 goto err;
 750                         if (!b)
 751                                 continue;
 752
 753                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0);
 754                         bch2_trans_iter_exit(&trans, &iter);
 755
 756                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 757                                 continue;
 758                         if (ret)
 759                                 goto err;
 760
 761                         if (ctxt->rate)
 762                                 bch2_ratelimit_increment(ctxt->rate,
 763                                                          c->opts.btree_node_size >> 9);
 764                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_seen);
 765                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_moved);
 766                 }
 767
 768                 bp_offset++;
 769         }
 770
 771         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
 772                 bch2_trans_unlock(&trans);
 773                 move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
 774                 closure_sync(&ctxt->cl);
 775                 if (!ctxt->write_error)
 776                         lockrestart_do(&trans, verify_bucket_evacuated(&trans, bucket, gen));
 777         }
 778 err:
 779         bch2_trans_exit(&trans);
 780         bch2_bkey_buf_exit(&sk, c);
 781         return ret;
 782 }
 783
 784 int bch2_evacuate_bucket(struct bch_fs *c,
 785                          struct bpos bucket, int gen,
 786                          struct data_update_opts data_opts,
 787                          struct bch_ratelimit *rate,
 788                          struct bch_move_stats *stats,
 789                          struct write_point_specifier wp,
 790                          bool wait_on_copygc)
 791 {
 792         struct moving_context ctxt;
 793         int ret;
 794
 795         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 796         ret = __bch2_evacuate_bucket(&ctxt, bucket, gen, data_opts);
 797         bch2_moving_ctxt_exit(&ctxt);
 798
 799         return ret;
 800 }
 801
 802 typedef bool (*move_btree_pred)(struct bch_fs *, void *,
 803                                 struct btree *, struct bch_io_opts *,
 804                                 struct data_update_opts *);
 805
 806 static int bch2_move_btree(struct bch_fs *c,
 807                            enum btree_id start_btree_id, struct bpos start_pos,
 808                            enum btree_id end_btree_id,   struct bpos end_pos,
 809                            move_btree_pred pred, void *arg,
 810                            struct bch_move_stats *stats)
 811 {
 812         bool kthread = (current->flags & PF_KTHREAD) != 0;
 813         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 814         struct btree_trans trans;
 815         struct btree_iter iter;
 816         struct btree *b;
 817         enum btree_id id;
 818         struct data_update_opts data_opts;
 819         int ret = 0;
 820
 821         bch2_trans_init(&trans, c, 0, 0);
 822         progress_list_add(c, stats);
 823
 824         stats->data_type = BCH_DATA_btree;
 825
 826         for (id = start_btree_id;
 827              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 828              id++) {
 829                 stats->btree_id = id;
 830
 831                 bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
 832                                           BTREE_ITER_PREFETCH);
 833 retry:
 834                 ret = 0;
 835                 while (bch2_trans_begin(&trans),
 836                        (b = bch2_btree_iter_peek_node(&iter)) &&
 837                        !(ret = PTR_ERR_OR_ZERO(b))) {
 838                         if (kthread && kthread_should_stop())
 839                                 break;
 840
 841                         if ((cmp_int(id, end_btree_id) ?:
 842                              bpos_cmp(b->key.k.p, end_pos)) > 0)
 843                                 break;
 844
 845                         stats->pos = iter.pos;
 846
 847                         if (!pred(c, arg, b, &io_opts, &data_opts))
 848                                 goto next;
 849
 850                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
 851                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 852                                 continue;
 853                         if (ret)
 854                                 break;
 855 next:
 856                         bch2_btree_iter_next_node(&iter);
 857                 }
 858                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 859                         goto retry;
 860
 861                 bch2_trans_iter_exit(&trans, &iter);
 862
 863                 if (kthread && kthread_should_stop())
 864                         break;
 865         }
 866
 867         bch2_trans_exit(&trans);
 868
 869         if (ret)
 870                 bch_err(c, "error in %s(): %s", __func__, bch2_err_str(ret));
 871
 872         bch2_btree_interior_updates_flush(c);
 873
 874         progress_list_del(c, stats);
 875         return ret;
 876 }
 877
 878 static bool rereplicate_pred(struct bch_fs *c, void *arg,
 879                              struct bkey_s_c k,
 880                              struct bch_io_opts *io_opts,
 881                              struct data_update_opts *data_opts)
 882 {
 883         unsigned nr_good = bch2_bkey_durability(c, k);
 884         unsigned replicas = bkey_is_btree_ptr(k.k)
 885                 ? c->opts.metadata_replicas
 886                 : io_opts->data_replicas;
 887
 888         if (!nr_good || nr_good >= replicas)
 889                 return false;
 890
 891         data_opts->target               = 0;
 892         data_opts->extra_replicas       = replicas - nr_good;
 893         data_opts->btree_insert_flags   = 0;
 894         return true;
 895 }
 896
 897 static bool migrate_pred(struct bch_fs *c, void *arg,
 898                          struct bkey_s_c k,
 899                          struct bch_io_opts *io_opts,
 900                          struct data_update_opts *data_opts)
 901 {
 902         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 903         const struct bch_extent_ptr *ptr;
 904         struct bch_ioctl_data *op = arg;
 905         unsigned i = 0;
 906
 907         data_opts->rewrite_ptrs         = 0;
 908         data_opts->target               = 0;
 909         data_opts->extra_replicas       = 0;
 910         data_opts->btree_insert_flags   = 0;
 911
 912         bkey_for_each_ptr(ptrs, ptr) {
 913                 if (ptr->dev == op->migrate.dev)
 914                         data_opts->rewrite_ptrs |= 1U << i;
 915                 i++;
 916         }
 917
 918         return data_opts->rewrite_ptrs != 0;
 919 }
 920
 921 static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
 922                                    struct btree *b,
 923                                    struct bch_io_opts *io_opts,
 924                                    struct data_update_opts *data_opts)
 925 {
 926         return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 927 }
 928
 929 static bool migrate_btree_pred(struct bch_fs *c, void *arg,
 930                                struct btree *b,
 931                                struct bch_io_opts *io_opts,
 932                                struct data_update_opts *data_opts)
 933 {
 934         return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 935 }
 936
 937 static bool bformat_needs_redo(struct bkey_format *f)
 938 {
 939         unsigned i;
 940
 941         for (i = 0; i < f->nr_fields; i++) {
 942                 unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
 943                 u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
 944                 u64 field_offset = le64_to_cpu(f->field_offset[i]);
 945
 946                 if (f->bits_per_field[i] > unpacked_bits)
 947                         return true;
 948
 949                 if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
 950                         return true;
 951
 952                 if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
 953                      unpacked_mask) <
 954                     field_offset)
 955                         return true;
 956         }
 957
 958         return false;
 959 }
 960
 961 static bool rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
 962                                    struct btree *b,
 963                                    struct bch_io_opts *io_opts,
 964                                    struct data_update_opts *data_opts)
 965 {
 966         if (b->version_ondisk != c->sb.version ||
 967             btree_node_need_rewrite(b) ||
 968             bformat_needs_redo(&b->format)) {
 969                 data_opts->target               = 0;
 970                 data_opts->extra_replicas       = 0;
 971                 data_opts->btree_insert_flags   = 0;
 972                 return true;
 973         }
 974
 975         return false;
 976 }
 977
 978 int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
 979 {
 980         int ret;
 981
 982         ret = bch2_move_btree(c,
 983                               0,                POS_MIN,
 984                               BTREE_ID_NR,      SPOS_MAX,
 985                               rewrite_old_nodes_pred, c, stats);
 986         if (!ret) {
 987                 mutex_lock(&c->sb_lock);
 988                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 989                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 990                 c->disk_sb.sb->version_min = c->disk_sb.sb->version;
 991                 bch2_write_super(c);
 992                 mutex_unlock(&c->sb_lock);
 993         }
 994
 995         return ret;
 996 }
 997
 998 int bch2_data_job(struct bch_fs *c,
 999                   struct bch_move_stats *stats,
1000                   struct bch_ioctl_data op)
1001 {
1002         int ret = 0;
1003
1004         switch (op.op) {
1005         case BCH_DATA_OP_REREPLICATE:
1006                 bch2_move_stats_init(stats, "rereplicate");
1007                 stats->data_type = BCH_DATA_journal;
1008                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
1009
1010                 ret = bch2_move_btree(c,
1011                                       op.start_btree,   op.start_pos,
1012                                       op.end_btree,     op.end_pos,
1013                                       rereplicate_btree_pred, c, stats) ?: ret;
1014                 ret = bch2_replicas_gc2(c) ?: ret;
1015
1016                 ret = bch2_move_data(c,
1017                                      op.start_btree,    op.start_pos,
1018                                      op.end_btree,      op.end_pos,
1019                                      NULL,
1020                                      stats,
1021                                      writepoint_hashed((unsigned long) current),
1022                                      true,
1023                                      rereplicate_pred, c) ?: ret;
1024                 ret = bch2_replicas_gc2(c) ?: ret;
1025                 break;
1026         case BCH_DATA_OP_MIGRATE:
1027                 if (op.migrate.dev >= c->sb.nr_devices)
1028                         return -EINVAL;
1029
1030                 bch2_move_stats_init(stats, "migrate");
1031                 stats->data_type = BCH_DATA_journal;
1032                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
1033
1034                 ret = bch2_move_btree(c,
1035                                       op.start_btree,   op.start_pos,
1036                                       op.end_btree,     op.end_pos,
1037                                       migrate_btree_pred, &op, stats) ?: ret;
1038                 ret = bch2_replicas_gc2(c) ?: ret;
1039
1040                 ret = bch2_move_data(c,
1041                                      op.start_btree,    op.start_pos,
1042                                      op.end_btree,      op.end_pos,
1043                                      NULL,
1044                                      stats,
1045                                      writepoint_hashed((unsigned long) current),
1046                                      true,
1047                                      migrate_pred, &op) ?: ret;
1048                 ret = bch2_replicas_gc2(c) ?: ret;
1049                 break;
1050         case BCH_DATA_OP_REWRITE_OLD_NODES:
1051                 bch2_move_stats_init(stats, "rewrite_old_nodes");
1052                 ret = bch2_scan_old_btree_nodes(c, stats);
1053                 break;
1054         default:
1055                 ret = -EINVAL;
1056         }
1057
1058         return ret;
1059 }