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 "error.h"
  13 #include "inode.h"
  14 #include "io.h"
  15 #include "journal_reclaim.h"
  16 #include "move.h"
  17 #include "replicas.h"
  18 #include "super-io.h"
  19 #include "keylist.h"
  20
  21 #include <linux/ioprio.h>
  22 #include <linux/kthread.h>
  23
  24 #include <trace/events/bcachefs.h>
  25
  26 static void progress_list_add(struct bch_fs *c, struct bch_move_stats *stats)
  27 {
  28         mutex_lock(&c->data_progress_lock);
  29         list_add(&stats->list, &c->data_progress_list);
  30         mutex_unlock(&c->data_progress_lock);
  31 }
  32
  33 static void progress_list_del(struct bch_fs *c, struct bch_move_stats *stats)
  34 {
  35         mutex_lock(&c->data_progress_lock);
  36         list_del(&stats->list);
  37         mutex_unlock(&c->data_progress_lock);
  38 }
  39
  40 struct moving_io {
  41         struct list_head        list;
  42         struct closure          cl;
  43         bool                    read_completed;
  44
  45         unsigned                read_sectors;
  46         unsigned                write_sectors;
  47
  48         struct bch_read_bio     rbio;
  49
  50         struct data_update      write;
  51         /* Must be last since it is variable size */
  52         struct bio_vec          bi_inline_vecs[0];
  53 };
  54
  55 static void move_free(struct closure *cl)
  56 {
  57         struct moving_io *io = container_of(cl, struct moving_io, cl);
  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 closure *cl)
  68 {
  69         struct moving_io *io = container_of(cl, struct moving_io, cl);
  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         closure_return_with_destructor(cl, move_free);
  77 }
  78
  79 static void move_write(struct closure *cl)
  80 {
  81         struct moving_io *io = container_of(cl, struct moving_io, cl);
  82
  83         if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
  84                 closure_return_with_destructor(cl, move_free);
  85                 return;
  86         }
  87
  88         atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
  89
  90         bch2_data_update_read_done(&io->write, io->rbio.pick.crc, cl);
  91         continue_at(cl, move_write_done, NULL);
  92 }
  93
  94 static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
  95 {
  96         struct moving_io *io =
  97                 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
  98
  99         return io && io->read_completed ? io : NULL;
 100 }
 101
 102 static void move_read_endio(struct bio *bio)
 103 {
 104         struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
 105         struct moving_context *ctxt = io->write.ctxt;
 106
 107         atomic_sub(io->read_sectors, &ctxt->read_sectors);
 108         io->read_completed = true;
 109
 110         wake_up(&ctxt->wait);
 111         closure_put(&ctxt->cl);
 112 }
 113
 114 static void do_pending_writes(struct moving_context *ctxt, struct btree_trans *trans)
 115 {
 116         struct moving_io *io;
 117
 118         if (trans)
 119                 bch2_trans_unlock(trans);
 120
 121         while ((io = next_pending_write(ctxt))) {
 122                 list_del(&io->list);
 123                 closure_call(&io->cl, move_write, NULL, &ctxt->cl);
 124         }
 125 }
 126
 127 #define move_ctxt_wait_event(_ctxt, _trans, _cond)              \
 128 do {                                                            \
 129         do_pending_writes(_ctxt, _trans);                       \
 130                                                                 \
 131         if (_cond)                                              \
 132                 break;                                          \
 133         __wait_event((_ctxt)->wait,                             \
 134                      next_pending_write(_ctxt) || (_cond));     \
 135 } while (1)
 136
 137 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt,
 138                                        struct btree_trans *trans)
 139 {
 140         unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
 141
 142         move_ctxt_wait_event(ctxt, trans,
 143                 !atomic_read(&ctxt->write_sectors) ||
 144                 atomic_read(&ctxt->write_sectors) != sectors_pending);
 145 }
 146
 147 void bch2_moving_ctxt_exit(struct moving_context *ctxt)
 148 {
 149         move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
 150         closure_sync(&ctxt->cl);
 151         progress_list_del(ctxt->c, ctxt->stats);
 152
 153         EBUG_ON(atomic_read(&ctxt->write_sectors));
 154
 155         trace_move_data(ctxt->c,
 156                         atomic64_read(&ctxt->stats->sectors_moved),
 157                         atomic64_read(&ctxt->stats->keys_moved));
 158 }
 159
 160 void bch2_moving_ctxt_init(struct moving_context *ctxt,
 161                            struct bch_fs *c,
 162                            struct bch_ratelimit *rate,
 163                            struct bch_move_stats *stats,
 164                            struct write_point_specifier wp,
 165                            bool wait_on_copygc)
 166 {
 167         memset(ctxt, 0, sizeof(*ctxt));
 168
 169         ctxt->c         = c;
 170         ctxt->rate      = rate;
 171         ctxt->stats     = stats;
 172         ctxt->wp        = wp;
 173         ctxt->wait_on_copygc = wait_on_copygc;
 174
 175         progress_list_add(c, stats);
 176         closure_init_stack(&ctxt->cl);
 177         INIT_LIST_HEAD(&ctxt->reads);
 178         init_waitqueue_head(&ctxt->wait);
 179
 180         if (stats)
 181                 stats->data_type = BCH_DATA_user;
 182 }
 183
 184 void bch_move_stats_init(struct bch_move_stats *stats, char *name)
 185 {
 186         memset(stats, 0, sizeof(*stats));
 187         scnprintf(stats->name, sizeof(stats->name), "%s", name);
 188 }
 189
 190 static int bch2_move_extent(struct btree_trans *trans,
 191                             struct moving_context *ctxt,
 192                             struct bch_io_opts io_opts,
 193                             enum btree_id btree_id,
 194                             struct bkey_s_c k,
 195                             struct data_update_opts data_opts)
 196 {
 197         struct bch_fs *c = trans->c;
 198         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 199         struct moving_io *io;
 200         const union bch_extent_entry *entry;
 201         struct extent_ptr_decoded p;
 202         unsigned sectors = k.k->size, pages;
 203         int ret = -ENOMEM;
 204
 205         if (!percpu_ref_tryget_live(&c->writes))
 206                 return -EROFS;
 207
 208         /* write path might have to decompress data: */
 209         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 210                 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
 211
 212         pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
 213         io = kzalloc(sizeof(struct moving_io) +
 214                      sizeof(struct bio_vec) * pages, GFP_KERNEL);
 215         if (!io)
 216                 goto err;
 217
 218         io->write.ctxt          = ctxt;
 219         io->read_sectors        = k.k->size;
 220         io->write_sectors       = k.k->size;
 221
 222         bio_init(&io->write.op.wbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 223         bio_set_prio(&io->write.op.wbio.bio,
 224                      IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 225
 226         if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
 227                                  GFP_KERNEL))
 228                 goto err_free;
 229
 230         io->rbio.c              = c;
 231         io->rbio.opts           = io_opts;
 232         bio_init(&io->rbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 233         io->rbio.bio.bi_vcnt = pages;
 234         bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 235         io->rbio.bio.bi_iter.bi_size = sectors << 9;
 236
 237         bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
 238         io->rbio.bio.bi_iter.bi_sector  = bkey_start_offset(k.k);
 239         io->rbio.bio.bi_end_io          = move_read_endio;
 240
 241         ret = bch2_data_update_init(c, &io->write, ctxt->wp, io_opts,
 242                                     data_opts, btree_id, k);
 243         if (ret)
 244                 goto err_free_pages;
 245
 246         io->write.ctxt = ctxt;
 247
 248         atomic64_inc(&ctxt->stats->keys_moved);
 249         atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
 250         this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
 251
 252         trace_move_extent(k.k);
 253
 254         atomic_add(io->read_sectors, &ctxt->read_sectors);
 255         list_add_tail(&io->list, &ctxt->reads);
 256
 257         /*
 258          * dropped by move_read_endio() - guards against use after free of
 259          * ctxt when doing wakeup
 260          */
 261         closure_get(&ctxt->cl);
 262         bch2_read_extent(trans, &io->rbio,
 263                          bkey_start_pos(k.k),
 264                          btree_id, k, 0,
 265                          BCH_READ_NODECODE|
 266                          BCH_READ_LAST_FRAGMENT);
 267         return 0;
 268 err_free_pages:
 269         bio_free_pages(&io->write.op.wbio.bio);
 270 err_free:
 271         kfree(io);
 272 err:
 273         percpu_ref_put(&c->writes);
 274         trace_move_alloc_mem_fail(k.k);
 275         return ret;
 276 }
 277
 278 static int lookup_inode(struct btree_trans *trans, struct bpos pos,
 279                         struct bch_inode_unpacked *inode)
 280 {
 281         struct btree_iter iter;
 282         struct bkey_s_c k;
 283         int ret;
 284
 285         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
 286                              BTREE_ITER_ALL_SNAPSHOTS);
 287         k = bch2_btree_iter_peek(&iter);
 288         ret = bkey_err(k);
 289         if (ret)
 290                 goto err;
 291
 292         if (!k.k || bkey_cmp(k.k->p, pos)) {
 293                 ret = -ENOENT;
 294                 goto err;
 295         }
 296
 297         ret = bkey_is_inode(k.k) ? 0 : -EIO;
 298         if (ret)
 299                 goto err;
 300
 301         ret = bch2_inode_unpack(k, inode);
 302         if (ret)
 303                 goto err;
 304 err:
 305         bch2_trans_iter_exit(trans, &iter);
 306         return ret;
 307 }
 308
 309 static int move_ratelimit(struct btree_trans *trans,
 310                           struct moving_context *ctxt)
 311 {
 312         struct bch_fs *c = trans->c;
 313         u64 delay;
 314
 315         if (ctxt->wait_on_copygc) {
 316                 bch2_trans_unlock(trans);
 317                 wait_event_killable(c->copygc_running_wq,
 318                                     !c->copygc_running ||
 319                                     kthread_should_stop());
 320         }
 321
 322         do {
 323                 delay = ctxt->rate ? bch2_ratelimit_delay(ctxt->rate) : 0;
 324
 325                 if (delay) {
 326                         bch2_trans_unlock(trans);
 327                         set_current_state(TASK_INTERRUPTIBLE);
 328                 }
 329
 330                 if ((current->flags & PF_KTHREAD) && kthread_should_stop()) {
 331                         __set_current_state(TASK_RUNNING);
 332                         return 1;
 333                 }
 334
 335                 if (delay)
 336                         schedule_timeout(delay);
 337
 338                 if (unlikely(freezing(current))) {
 339                         move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
 340                         try_to_freeze();
 341                 }
 342         } while (delay);
 343
 344         move_ctxt_wait_event(ctxt, trans,
 345                 atomic_read(&ctxt->write_sectors) <
 346                 c->opts.move_bytes_in_flight >> 9);
 347
 348         move_ctxt_wait_event(ctxt, trans,
 349                 atomic_read(&ctxt->read_sectors) <
 350                 c->opts.move_bytes_in_flight >> 9);
 351
 352         return 0;
 353 }
 354
 355 static int move_get_io_opts(struct btree_trans *trans,
 356                             struct bch_io_opts *io_opts,
 357                             struct bkey_s_c k, u64 *cur_inum)
 358 {
 359         struct bch_inode_unpacked inode;
 360         int ret;
 361
 362         if (*cur_inum == k.k->p.inode)
 363                 return 0;
 364
 365         *io_opts = bch2_opts_to_inode_opts(trans->c->opts);
 366
 367         ret = lookup_inode(trans,
 368                            SPOS(0, k.k->p.inode, k.k->p.snapshot),
 369                            &inode);
 370         if (ret == -EINTR)
 371                 return ret;
 372
 373         if (!ret)
 374                 bch2_io_opts_apply(io_opts, bch2_inode_opts_get(&inode));
 375
 376         *cur_inum = k.k->p.inode;
 377         return 0;
 378 }
 379
 380 static int __bch2_move_data(struct moving_context *ctxt,
 381                             struct bpos start,
 382                             struct bpos end,
 383                             move_pred_fn pred, void *arg,
 384                             enum btree_id btree_id)
 385 {
 386         struct bch_fs *c = ctxt->c;
 387         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 388         struct bkey_buf sk;
 389         struct btree_trans trans;
 390         struct btree_iter iter;
 391         struct bkey_s_c k;
 392         struct data_update_opts data_opts;
 393         u64 cur_inum = U64_MAX;
 394         int ret = 0, ret2;
 395
 396         bch2_bkey_buf_init(&sk);
 397         bch2_trans_init(&trans, c, 0, 0);
 398
 399         ctxt->stats->data_type  = BCH_DATA_user;
 400         ctxt->stats->btree_id   = btree_id;
 401         ctxt->stats->pos        = start;
 402
 403         bch2_trans_iter_init(&trans, &iter, btree_id, start,
 404                              BTREE_ITER_PREFETCH|
 405                              BTREE_ITER_ALL_SNAPSHOTS);
 406
 407         if (ctxt->rate)
 408                 bch2_ratelimit_reset(ctxt->rate);
 409
 410         while (!move_ratelimit(&trans, ctxt)) {
 411                 bch2_trans_begin(&trans);
 412
 413                 k = bch2_btree_iter_peek(&iter);
 414                 if (!k.k)
 415                         break;
 416
 417                 ret = bkey_err(k);
 418                 if (ret == -EINTR)
 419                         continue;
 420                 if (ret)
 421                         break;
 422
 423                 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
 424                         break;
 425
 426                 ctxt->stats->pos = iter.pos;
 427
 428                 if (!bkey_extent_is_direct_data(k.k))
 429                         goto next_nondata;
 430
 431                 ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 432                 if (ret)
 433                         continue;
 434
 435                 memset(&data_opts, 0, sizeof(data_opts));
 436                 if (!pred(c, arg, k, &io_opts, &data_opts))
 437                         goto next;
 438
 439                 /*
 440                  * The iterator gets unlocked by __bch2_read_extent - need to
 441                  * save a copy of @k elsewhere:
 442                   */
 443                 bch2_bkey_buf_reassemble(&sk, c, k);
 444                 k = bkey_i_to_s_c(sk.k);
 445
 446                 ret2 = bch2_move_extent(&trans, ctxt, io_opts,
 447                                         btree_id, k, data_opts);
 448                 if (ret2) {
 449                         if (ret2 == -EINTR)
 450                                 continue;
 451
 452                         if (ret2 == -ENOMEM) {
 453                                 /* memory allocation failure, wait for some IO to finish */
 454                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 455                                 continue;
 456                         }
 457
 458                         /* XXX signal failure */
 459                         goto next;
 460                 }
 461
 462                 if (ctxt->rate)
 463                         bch2_ratelimit_increment(ctxt->rate, k.k->size);
 464 next:
 465                 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 466 next_nondata:
 467                 bch2_btree_iter_advance(&iter);
 468         }
 469
 470         bch2_trans_iter_exit(&trans, &iter);
 471         bch2_trans_exit(&trans);
 472         bch2_bkey_buf_exit(&sk, c);
 473
 474         return ret;
 475 }
 476
 477 int bch2_move_data(struct bch_fs *c,
 478                    enum btree_id start_btree_id, struct bpos start_pos,
 479                    enum btree_id end_btree_id,   struct bpos end_pos,
 480                    struct bch_ratelimit *rate,
 481                    struct bch_move_stats *stats,
 482                    struct write_point_specifier wp,
 483                    bool wait_on_copygc,
 484                    move_pred_fn pred, void *arg)
 485 {
 486         struct moving_context ctxt;
 487         enum btree_id id;
 488         int ret;
 489
 490         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 491
 492         for (id = start_btree_id;
 493              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 494              id++) {
 495                 stats->btree_id = id;
 496
 497                 if (id != BTREE_ID_extents &&
 498                     id != BTREE_ID_reflink)
 499                         continue;
 500
 501                 ret = __bch2_move_data(&ctxt,
 502                                        id == start_btree_id ? start_pos : POS_MIN,
 503                                        id == end_btree_id   ? end_pos   : POS_MAX,
 504                                        pred, arg, id);
 505                 if (ret)
 506                         break;
 507         }
 508
 509         bch2_moving_ctxt_exit(&ctxt);
 510
 511         return ret;
 512 }
 513
 514 static int verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
 515 {
 516         struct bch_fs *c = trans->c;
 517         struct btree_iter iter;
 518         struct bkey_s_c k;
 519         int ret;
 520
 521         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
 522                              bucket, BTREE_ITER_CACHED);
 523 again:
 524         k = bch2_btree_iter_peek_slot(&iter);
 525         ret = bkey_err(k);
 526
 527         if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
 528                 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
 529
 530                 if (a.v->gen == gen &&
 531                     a.v->dirty_sectors) {
 532                         struct printbuf buf = PRINTBUF;
 533
 534                         if (a.v->data_type == BCH_DATA_btree) {
 535                                 bch2_trans_unlock(trans);
 536                                 if (bch2_btree_interior_updates_flush(c))
 537                                         goto again;
 538                         }
 539
 540                         prt_str(&buf, "failed to evacuate bucket ");
 541                         bch2_bkey_val_to_text(&buf, c, k);
 542
 543                         bch2_trans_inconsistent(trans, "%s", buf.buf);
 544                         printbuf_exit(&buf);
 545                 }
 546         }
 547
 548         bch2_trans_iter_exit(trans, &iter);
 549         return ret;
 550 }
 551
 552 int __bch2_evacuate_bucket(struct moving_context *ctxt,
 553                            struct bpos bucket, int gen,
 554                            struct data_update_opts _data_opts)
 555 {
 556         struct bch_fs *c = ctxt->c;
 557         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 558         struct btree_trans trans;
 559         struct btree_iter iter;
 560         struct bkey_buf sk;
 561         struct bch_backpointer bp;
 562         struct data_update_opts data_opts;
 563         u64 bp_offset = 0, cur_inum = U64_MAX;
 564         int ret = 0;
 565
 566         bch2_bkey_buf_init(&sk);
 567         bch2_trans_init(&trans, c, 0, 0);
 568
 569         while (!(ret = move_ratelimit(&trans, ctxt))) {
 570                 bch2_trans_begin(&trans);
 571
 572                 ret = bch2_get_next_backpointer(&trans, bucket, gen,
 573                                                 &bp_offset, &bp);
 574                 if (ret == -EINTR)
 575                         continue;
 576                 if (ret)
 577                         goto err;
 578                 if (bp_offset == U64_MAX)
 579                         break;
 580
 581                 if (!bp.level) {
 582                         const struct bch_extent_ptr *ptr;
 583                         struct bkey_s_c k;
 584                         unsigned i = 0;
 585
 586                         k = bch2_backpointer_get_key(&trans, &iter,
 587                                                 bucket, bp_offset, bp);
 588                         ret = bkey_err(k);
 589                         if (ret == -EINTR)
 590                                 continue;
 591                         if (ret)
 592                                 goto err;
 593                         if (!k.k)
 594                                 continue;
 595
 596                         bch2_bkey_buf_reassemble(&sk, c, k);
 597                         k = bkey_i_to_s_c(sk.k);
 598                         bch2_trans_iter_exit(&trans, &iter);
 599
 600                         ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 601                         if (ret)
 602                                 continue;
 603
 604                         data_opts = _data_opts;
 605                         data_opts.target        = io_opts.background_target;
 606                         data_opts.rewrite_ptrs = 0;
 607
 608                         bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
 609                                 if (ptr->dev == bucket.inode)
 610                                         data_opts.rewrite_ptrs |= 1U << i;
 611                                 i++;
 612                         }
 613
 614                         ret = bch2_move_extent(&trans, ctxt, io_opts,
 615                                                bp.btree_id, k, data_opts);
 616                         if (ret == -EINTR)
 617                                 continue;
 618                         if (ret == -ENOMEM) {
 619                                 /* memory allocation failure, wait for some IO to finish */
 620                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 621                                 continue;
 622                         }
 623                         if (ret)
 624                                 goto err;
 625
 626                         if (ctxt->rate)
 627                                 bch2_ratelimit_increment(ctxt->rate, k.k->size);
 628                         atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 629                 } else {
 630                         struct btree *b;
 631
 632                         b = bch2_backpointer_get_node(&trans, &iter,
 633                                                 bucket, bp_offset, bp);
 634                         ret = PTR_ERR_OR_ZERO(b);
 635                         if (ret == -EINTR)
 636                                 continue;
 637                         if (ret)
 638                                 goto err;
 639                         if (!b)
 640                                 continue;
 641
 642                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0);
 643                         bch2_trans_iter_exit(&trans, &iter);
 644
 645                         if (ret == -EINTR)
 646                                 continue;
 647                         if (ret)
 648                                 goto err;
 649
 650                         if (ctxt->rate)
 651                                 bch2_ratelimit_increment(ctxt->rate,
 652                                                          c->opts.btree_node_size >> 9);
 653                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_seen);
 654                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_moved);
 655                 }
 656
 657                 bp_offset++;
 658         }
 659
 660         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
 661                 bch2_trans_unlock(&trans);
 662                 move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
 663                 closure_sync(&ctxt->cl);
 664                 if (!ctxt->write_error)
 665                         lockrestart_do(&trans, verify_bucket_evacuated(&trans, bucket, gen));
 666         }
 667 err:
 668         bch2_trans_exit(&trans);
 669         bch2_bkey_buf_exit(&sk, c);
 670         return ret;
 671 }
 672
 673 int bch2_evacuate_bucket(struct bch_fs *c,
 674                          struct bpos bucket, int gen,
 675                          struct data_update_opts data_opts,
 676                          struct bch_ratelimit *rate,
 677                          struct bch_move_stats *stats,
 678                          struct write_point_specifier wp,
 679                          bool wait_on_copygc)
 680 {
 681         struct moving_context ctxt;
 682         int ret;
 683
 684         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 685         ret = __bch2_evacuate_bucket(&ctxt, bucket, gen, data_opts);
 686         bch2_moving_ctxt_exit(&ctxt);
 687
 688         return ret;
 689 }
 690
 691 typedef bool (*move_btree_pred)(struct bch_fs *, void *,
 692                                 struct btree *, struct bch_io_opts *,
 693                                 struct data_update_opts *);
 694
 695 static int bch2_move_btree(struct bch_fs *c,
 696                            enum btree_id start_btree_id, struct bpos start_pos,
 697                            enum btree_id end_btree_id,   struct bpos end_pos,
 698                            move_btree_pred pred, void *arg,
 699                            struct bch_move_stats *stats)
 700 {
 701         bool kthread = (current->flags & PF_KTHREAD) != 0;
 702         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 703         struct btree_trans trans;
 704         struct btree_iter iter;
 705         struct btree *b;
 706         enum btree_id id;
 707         struct data_update_opts data_opts;
 708         int ret = 0;
 709
 710         bch2_trans_init(&trans, c, 0, 0);
 711         progress_list_add(c, stats);
 712
 713         stats->data_type = BCH_DATA_btree;
 714
 715         for (id = start_btree_id;
 716              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 717              id++) {
 718                 stats->btree_id = id;
 719
 720                 bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
 721                                           BTREE_ITER_PREFETCH);
 722 retry:
 723                 ret = 0;
 724                 while (bch2_trans_begin(&trans),
 725                        (b = bch2_btree_iter_peek_node(&iter)) &&
 726                        !(ret = PTR_ERR_OR_ZERO(b))) {
 727                         if (kthread && kthread_should_stop())
 728                                 break;
 729
 730                         if ((cmp_int(id, end_btree_id) ?:
 731                              bpos_cmp(b->key.k.p, end_pos)) > 0)
 732                                 break;
 733
 734                         stats->pos = iter.pos;
 735
 736                         if (!pred(c, arg, b, &io_opts, &data_opts))
 737                                 goto next;
 738
 739                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
 740                         if (ret == -EINTR)
 741                                 continue;
 742                         if (ret)
 743                                 break;
 744 next:
 745                         bch2_btree_iter_next_node(&iter);
 746                 }
 747                 if (ret == -EINTR)
 748                         goto retry;
 749
 750                 bch2_trans_iter_exit(&trans, &iter);
 751
 752                 if (kthread && kthread_should_stop())
 753                         break;
 754         }
 755
 756         bch2_trans_exit(&trans);
 757
 758         if (ret)
 759                 bch_err(c, "error %i in bch2_move_btree", ret);
 760
 761         bch2_btree_interior_updates_flush(c);
 762
 763         progress_list_del(c, stats);
 764         return ret;
 765 }
 766
 767 static bool rereplicate_pred(struct bch_fs *c, void *arg,
 768                              struct bkey_s_c k,
 769                              struct bch_io_opts *io_opts,
 770                              struct data_update_opts *data_opts)
 771 {
 772         unsigned nr_good = bch2_bkey_durability(c, k);
 773         unsigned replicas = bkey_is_btree_ptr(k.k)
 774                 ? c->opts.metadata_replicas
 775                 : io_opts->data_replicas;
 776
 777         if (!nr_good || nr_good >= replicas)
 778                 return false;
 779
 780         data_opts->target               = 0;
 781         data_opts->extra_replicas       = replicas - nr_good;
 782         data_opts->btree_insert_flags   = 0;
 783         return true;
 784 }
 785
 786 static bool migrate_pred(struct bch_fs *c, void *arg,
 787                          struct bkey_s_c k,
 788                          struct bch_io_opts *io_opts,
 789                          struct data_update_opts *data_opts)
 790 {
 791         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 792         const struct bch_extent_ptr *ptr;
 793         struct bch_ioctl_data *op = arg;
 794         unsigned i = 0;
 795
 796         data_opts->rewrite_ptrs         = 0;
 797         data_opts->target               = 0;
 798         data_opts->extra_replicas       = 0;
 799         data_opts->btree_insert_flags   = 0;
 800
 801         bkey_for_each_ptr(ptrs, ptr) {
 802                 if (ptr->dev == op->migrate.dev)
 803                         data_opts->rewrite_ptrs |= 1U << i;
 804                 i++;
 805         }
 806
 807         return data_opts->rewrite_ptrs != 0;;
 808 }
 809
 810 static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
 811                                    struct btree *b,
 812                                    struct bch_io_opts *io_opts,
 813                                    struct data_update_opts *data_opts)
 814 {
 815         return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 816 }
 817
 818 static bool migrate_btree_pred(struct bch_fs *c, void *arg,
 819                                struct btree *b,
 820                                struct bch_io_opts *io_opts,
 821                                struct data_update_opts *data_opts)
 822 {
 823         return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 824 }
 825
 826 static bool bformat_needs_redo(struct bkey_format *f)
 827 {
 828         unsigned i;
 829
 830         for (i = 0; i < f->nr_fields; i++) {
 831                 unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
 832                 u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
 833                 u64 field_offset = le64_to_cpu(f->field_offset[i]);
 834
 835                 if (f->bits_per_field[i] > unpacked_bits)
 836                         return true;
 837
 838                 if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
 839                         return true;
 840
 841                 if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
 842                      unpacked_mask) <
 843                     field_offset)
 844                         return true;
 845         }
 846
 847         return false;
 848 }
 849
 850 static bool rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
 851                                    struct btree *b,
 852                                    struct bch_io_opts *io_opts,
 853                                    struct data_update_opts *data_opts)
 854 {
 855         if (b->version_ondisk != c->sb.version ||
 856             btree_node_need_rewrite(b) ||
 857             bformat_needs_redo(&b->format)) {
 858                 data_opts->target               = 0;
 859                 data_opts->extra_replicas       = 0;
 860                 data_opts->btree_insert_flags   = 0;
 861                 return true;
 862         }
 863
 864         return false;
 865 }
 866
 867 int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
 868 {
 869         int ret;
 870
 871         ret = bch2_move_btree(c,
 872                               0,                POS_MIN,
 873                               BTREE_ID_NR,      SPOS_MAX,
 874                               rewrite_old_nodes_pred, c, stats);
 875         if (!ret) {
 876                 mutex_lock(&c->sb_lock);
 877                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 878                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 879                 c->disk_sb.sb->version_min = c->disk_sb.sb->version;
 880                 bch2_write_super(c);
 881                 mutex_unlock(&c->sb_lock);
 882         }
 883
 884         return ret;
 885 }
 886
 887 int bch2_data_job(struct bch_fs *c,
 888                   struct bch_move_stats *stats,
 889                   struct bch_ioctl_data op)
 890 {
 891         int ret = 0;
 892
 893         switch (op.op) {
 894         case BCH_DATA_OP_REREPLICATE:
 895                 bch_move_stats_init(stats, "rereplicate");
 896                 stats->data_type = BCH_DATA_journal;
 897                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
 898
 899                 ret = bch2_move_btree(c,
 900                                       op.start_btree,   op.start_pos,
 901                                       op.end_btree,     op.end_pos,
 902                                       rereplicate_btree_pred, c, stats) ?: ret;
 903                 ret = bch2_replicas_gc2(c) ?: ret;
 904
 905                 ret = bch2_move_data(c,
 906                                      op.start_btree,    op.start_pos,
 907                                      op.end_btree,      op.end_pos,
 908                                      NULL,
 909                                      stats,
 910                                      writepoint_hashed((unsigned long) current),
 911                                      true,
 912                                      rereplicate_pred, c) ?: ret;
 913                 ret = bch2_replicas_gc2(c) ?: ret;
 914                 break;
 915         case BCH_DATA_OP_MIGRATE:
 916                 if (op.migrate.dev >= c->sb.nr_devices)
 917                         return -EINVAL;
 918
 919                 bch_move_stats_init(stats, "migrate");
 920                 stats->data_type = BCH_DATA_journal;
 921                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
 922
 923                 ret = bch2_move_btree(c,
 924                                       op.start_btree,   op.start_pos,
 925                                       op.end_btree,     op.end_pos,
 926                                       migrate_btree_pred, &op, stats) ?: ret;
 927                 ret = bch2_replicas_gc2(c) ?: ret;
 928
 929                 ret = bch2_move_data(c,
 930                                      op.start_btree,    op.start_pos,
 931                                      op.end_btree,      op.end_pos,
 932                                      NULL,
 933                                      stats,
 934                                      writepoint_hashed((unsigned long) current),
 935                                      true,
 936                                      migrate_pred, &op) ?: ret;
 937                 ret = bch2_replicas_gc2(c) ?: ret;
 938                 break;
 939         case BCH_DATA_OP_REWRITE_OLD_NODES:
 940                 bch_move_stats_init(stats, "rewrite_old_nodes");
 941                 ret = bch2_scan_old_btree_nodes(c, stats);
 942                 break;
 943         default:
 944                 ret = -EINVAL;
 945         }
 946
 947         return ret;
 948 }