1 // SPDX-License-Identifier: GPL-2.0
4 #include "alloc_foreground.h"
7 #include "btree_update.h"
8 #include "btree_update_interior.h"
10 #include "disk_groups.h"
13 #include "journal_reclaim.h"
19 #include <linux/ioprio.h>
20 #include <linux/kthread.h>
22 #include <trace/events/bcachefs.h>
24 #define SECTORS_IN_FLIGHT_PER_DEVICE 2048
27 struct list_head list;
31 unsigned read_sectors;
32 unsigned write_sectors;
34 struct bch_read_bio rbio;
36 struct migrate_write write;
37 /* Must be last since it is variable size */
38 struct bio_vec bi_inline_vecs[0];
41 struct moving_context {
42 /* Closure for waiting on all reads and writes to complete */
45 struct bch_move_stats *stats;
47 struct list_head reads;
49 /* in flight sectors: */
50 atomic_t read_sectors;
51 atomic_t write_sectors;
53 wait_queue_head_t wait;
56 static int bch2_migrate_index_update(struct bch_write_op *op)
58 struct bch_fs *c = op->c;
59 struct btree_trans trans;
60 struct btree_iter *iter;
61 struct migrate_write *m =
62 container_of(op, struct migrate_write, op);
63 struct keylist *keys = &op->insert_keys;
64 struct bkey_buf _new, _insert;
67 bch2_bkey_buf_init(&_new);
68 bch2_bkey_buf_init(&_insert);
69 bch2_bkey_buf_realloc(&_insert, c, U8_MAX);
71 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024);
73 iter = bch2_trans_get_iter(&trans, m->btree_id,
74 bkey_start_pos(&bch2_keylist_front(keys)->k),
75 BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
79 struct bkey_i *insert;
80 struct bkey_i_extent *new;
81 const union bch_extent_entry *entry;
82 struct extent_ptr_decoded p;
83 bool did_work = false;
84 bool extending = false, should_check_enospc;
85 s64 i_sectors_delta = 0, disk_sectors_delta = 0;
87 bch2_trans_reset(&trans, 0);
89 k = bch2_btree_iter_peek_slot(iter);
94 new = bkey_i_to_extent(bch2_keylist_front(keys));
96 if (bversion_cmp(k.k->version, new->k.version) ||
97 !bch2_bkey_matches_ptr(c, k, m->ptr, m->offset))
100 bkey_reassemble(_insert.k, k);
103 bch2_bkey_buf_copy(&_new, c, bch2_keylist_front(keys));
104 new = bkey_i_to_extent(_new.k);
105 bch2_cut_front(iter->pos, &new->k_i);
107 bch2_cut_front(iter->pos, insert);
108 bch2_cut_back(new->k.p, insert);
109 bch2_cut_back(insert->k.p, &new->k_i);
111 if (m->data_cmd == DATA_REWRITE) {
112 struct bch_extent_ptr *new_ptr, *old_ptr = (void *)
113 bch2_bkey_has_device(bkey_i_to_s_c(insert),
114 m->data_opts.rewrite_dev);
119 extent_for_each_ptr(extent_i_to_s(new), new_ptr)
120 new_ptr->cached = true;
122 bch2_bkey_drop_ptr(bkey_i_to_s(insert), old_ptr);
125 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
126 if (bch2_bkey_has_device(bkey_i_to_s_c(insert), p.ptr.dev)) {
128 * raced with another move op? extent already
129 * has a pointer to the device we just wrote
135 bch2_extent_ptr_decoded_append(insert, &p);
142 bch2_bkey_narrow_crcs(insert,
143 (struct bch_extent_crc_unpacked) { 0 });
144 bch2_extent_normalize(c, bkey_i_to_s(insert));
145 bch2_bkey_mark_replicas_cached(c, bkey_i_to_s(insert),
146 op->opts.background_target,
147 op->opts.data_replicas);
149 ret = bch2_sum_sector_overwrites(&trans, iter, insert,
151 &should_check_enospc,
153 &disk_sectors_delta);
157 if (disk_sectors_delta > (s64) op->res.sectors) {
158 ret = bch2_disk_reservation_add(c, &op->res,
159 disk_sectors_delta - op->res.sectors,
161 ? BCH_DISK_RESERVATION_NOFAIL : 0);
166 ret = bch2_trans_update(&trans, iter, insert, 0) ?:
167 bch2_trans_commit(&trans, &op->res,
170 m->data_opts.btree_insert_flags);
173 atomic_long_inc(&c->extent_migrate_done);
179 while (bkey_cmp(iter->pos, bch2_keylist_front(keys)->k.p) >= 0) {
180 bch2_keylist_pop_front(keys);
181 if (bch2_keylist_empty(keys))
187 BUG_ON(k.k->p.offset <= iter->pos.offset);
188 atomic64_inc(&m->ctxt->stats->keys_raced);
189 atomic64_add(k.k->p.offset - iter->pos.offset,
190 &m->ctxt->stats->sectors_raced);
192 atomic_long_inc(&c->extent_migrate_raced);
193 trace_move_race(&new->k);
194 bch2_btree_iter_next_slot(iter);
198 bch2_trans_iter_put(&trans, iter);
199 bch2_trans_exit(&trans);
200 bch2_bkey_buf_exit(&_insert, c);
201 bch2_bkey_buf_exit(&_new, c);
202 BUG_ON(ret == -EINTR);
206 void bch2_migrate_read_done(struct migrate_write *m, struct bch_read_bio *rbio)
208 /* write bio must own pages: */
209 BUG_ON(!m->op.wbio.bio.bi_vcnt);
211 m->ptr = rbio->pick.ptr;
212 m->offset = rbio->data_pos.offset - rbio->pick.crc.offset;
213 m->op.devs_have = rbio->devs_have;
214 m->op.pos = rbio->data_pos;
215 m->op.version = rbio->version;
216 m->op.crc = rbio->pick.crc;
217 m->op.wbio.bio.bi_iter.bi_size = m->op.crc.compressed_size << 9;
219 if (bch2_csum_type_is_encryption(m->op.crc.csum_type)) {
220 m->op.nonce = m->op.crc.nonce + m->op.crc.offset;
221 m->op.csum_type = m->op.crc.csum_type;
224 if (m->data_cmd == DATA_REWRITE)
225 bch2_dev_list_drop_dev(&m->op.devs_have, m->data_opts.rewrite_dev);
228 int bch2_migrate_write_init(struct bch_fs *c, struct migrate_write *m,
229 struct write_point_specifier wp,
230 struct bch_io_opts io_opts,
231 enum data_cmd data_cmd,
232 struct data_opts data_opts,
233 enum btree_id btree_id,
236 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
237 const union bch_extent_entry *entry;
238 struct extent_ptr_decoded p;
241 m->btree_id = btree_id;
242 m->data_cmd = data_cmd;
243 m->data_opts = data_opts;
244 m->nr_ptrs_reserved = 0;
246 bch2_write_op_init(&m->op, c, io_opts);
248 if (!bch2_bkey_is_incompressible(k))
249 m->op.compression_type =
250 bch2_compression_opt_to_type[io_opts.background_compression ?:
251 io_opts.compression];
253 m->op.incompressible = true;
255 m->op.target = data_opts.target,
256 m->op.write_point = wp;
258 if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE) {
259 m->op.alloc_reserve = RESERVE_MOVINGGC;
260 m->op.flags |= BCH_WRITE_ALLOC_NOWAIT;
262 /* XXX: this should probably be passed in */
263 m->op.flags |= BCH_WRITE_ONLY_SPECIFIED_DEVS;
266 m->op.flags |= BCH_WRITE_PAGES_STABLE|
267 BCH_WRITE_PAGES_OWNED|
268 BCH_WRITE_DATA_ENCODED|
269 BCH_WRITE_FROM_INTERNAL;
271 m->op.nr_replicas = data_opts.nr_replicas;
272 m->op.nr_replicas_required = data_opts.nr_replicas;
273 m->op.index_update_fn = bch2_migrate_index_update;
276 case DATA_ADD_REPLICAS: {
278 * DATA_ADD_REPLICAS is used for moving data to a different
279 * device in the background, and due to compression the new copy
280 * might take up more space than the old copy:
283 int nr = (int) io_opts.data_replicas -
284 bch2_bkey_nr_ptrs_allocated(k);
286 int nr = (int) io_opts.data_replicas;
289 m->op.nr_replicas = m->nr_ptrs_reserved = nr;
291 ret = bch2_disk_reservation_get(c, &m->op.res,
292 k.k->size, m->op.nr_replicas, 0);
299 unsigned compressed_sectors = 0;
301 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
302 if (p.ptr.dev == data_opts.rewrite_dev &&
304 crc_is_compressed(p.crc))
305 compressed_sectors += p.crc.compressed_size;
307 if (compressed_sectors) {
308 ret = bch2_disk_reservation_add(c, &m->op.res,
309 k.k->size * m->op.nr_replicas,
310 BCH_DISK_RESERVATION_NOFAIL);
317 m->op.flags |= BCH_WRITE_ALLOC_NOWAIT;
318 m->op.flags |= BCH_WRITE_CACHED;
327 static void move_free(struct closure *cl)
329 struct moving_io *io = container_of(cl, struct moving_io, cl);
330 struct moving_context *ctxt = io->write.ctxt;
331 struct bvec_iter_all iter;
334 bch2_disk_reservation_put(io->write.op.c, &io->write.op.res);
336 bio_for_each_segment_all(bv, &io->write.op.wbio.bio, iter)
338 __free_page(bv->bv_page);
340 wake_up(&ctxt->wait);
345 static void move_write_done(struct closure *cl)
347 struct moving_io *io = container_of(cl, struct moving_io, cl);
349 atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
350 closure_return_with_destructor(cl, move_free);
353 static void move_write(struct closure *cl)
355 struct moving_io *io = container_of(cl, struct moving_io, cl);
357 if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
358 closure_return_with_destructor(cl, move_free);
362 bch2_migrate_read_done(&io->write, &io->rbio);
364 atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
365 closure_call(&io->write.op.cl, bch2_write, NULL, cl);
366 continue_at(cl, move_write_done, NULL);
369 static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
371 struct moving_io *io =
372 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
374 return io && io->read_completed ? io : NULL;
377 static void move_read_endio(struct bio *bio)
379 struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
380 struct moving_context *ctxt = io->write.ctxt;
382 atomic_sub(io->read_sectors, &ctxt->read_sectors);
383 io->read_completed = true;
385 if (next_pending_write(ctxt))
386 wake_up(&ctxt->wait);
388 closure_put(&ctxt->cl);
391 static void do_pending_writes(struct moving_context *ctxt)
393 struct moving_io *io;
395 while ((io = next_pending_write(ctxt))) {
397 closure_call(&io->cl, move_write, NULL, &ctxt->cl);
401 #define move_ctxt_wait_event(_ctxt, _cond) \
403 do_pending_writes(_ctxt); \
407 __wait_event((_ctxt)->wait, \
408 next_pending_write(_ctxt) || (_cond)); \
411 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt)
413 unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
415 move_ctxt_wait_event(ctxt,
416 !atomic_read(&ctxt->write_sectors) ||
417 atomic_read(&ctxt->write_sectors) != sectors_pending);
420 static int bch2_move_extent(struct btree_trans *trans,
421 struct moving_context *ctxt,
422 struct write_point_specifier wp,
423 struct bch_io_opts io_opts,
424 enum btree_id btree_id,
426 enum data_cmd data_cmd,
427 struct data_opts data_opts)
429 struct bch_fs *c = trans->c;
430 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
431 struct moving_io *io;
432 const union bch_extent_entry *entry;
433 struct extent_ptr_decoded p;
434 unsigned sectors = k.k->size, pages;
437 move_ctxt_wait_event(ctxt,
438 atomic_read(&ctxt->write_sectors) <
439 SECTORS_IN_FLIGHT_PER_DEVICE);
441 move_ctxt_wait_event(ctxt,
442 atomic_read(&ctxt->read_sectors) <
443 SECTORS_IN_FLIGHT_PER_DEVICE);
445 /* write path might have to decompress data: */
446 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
447 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
449 pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
450 io = kzalloc(sizeof(struct moving_io) +
451 sizeof(struct bio_vec) * pages, GFP_KERNEL);
455 io->write.ctxt = ctxt;
456 io->read_sectors = k.k->size;
457 io->write_sectors = k.k->size;
459 bio_init(&io->write.op.wbio.bio, io->bi_inline_vecs, pages);
460 bio_set_prio(&io->write.op.wbio.bio,
461 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
463 if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
468 io->rbio.opts = io_opts;
469 bio_init(&io->rbio.bio, io->bi_inline_vecs, pages);
470 io->rbio.bio.bi_vcnt = pages;
471 bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
472 io->rbio.bio.bi_iter.bi_size = sectors << 9;
474 bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
475 io->rbio.bio.bi_iter.bi_sector = bkey_start_offset(k.k);
476 io->rbio.bio.bi_end_io = move_read_endio;
478 ret = bch2_migrate_write_init(c, &io->write, wp, io_opts,
479 data_cmd, data_opts, btree_id, k);
483 atomic64_inc(&ctxt->stats->keys_moved);
484 atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
486 trace_move_extent(k.k);
488 atomic_add(io->read_sectors, &ctxt->read_sectors);
489 list_add_tail(&io->list, &ctxt->reads);
492 * dropped by move_read_endio() - guards against use after free of
493 * ctxt when doing wakeup
495 closure_get(&ctxt->cl);
496 bch2_read_extent(trans, &io->rbio,
500 BCH_READ_LAST_FRAGMENT);
503 bio_free_pages(&io->write.op.wbio.bio);
507 trace_move_alloc_fail(k.k);
511 static int lookup_inode(struct btree_trans *trans, struct bpos pos,
512 struct bch_inode_unpacked *inode)
514 struct btree_iter *iter;
518 iter = bch2_trans_get_iter(trans, BTREE_ID_inodes, pos,
519 BTREE_ITER_ALL_SNAPSHOTS);
520 k = bch2_btree_iter_peek(iter);
525 if (!k.k || bkey_cmp(k.k->p, pos)) {
530 ret = k.k->type == KEY_TYPE_inode ? 0 : -EIO;
534 ret = bch2_inode_unpack(bkey_s_c_to_inode(k), inode);
538 bch2_trans_iter_put(trans, iter);
542 static int __bch2_move_data(struct bch_fs *c,
543 struct moving_context *ctxt,
544 struct bch_ratelimit *rate,
545 struct write_point_specifier wp,
548 move_pred_fn pred, void *arg,
549 struct bch_move_stats *stats,
550 enum btree_id btree_id)
552 bool kthread = (current->flags & PF_KTHREAD) != 0;
553 struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
555 struct btree_trans trans;
556 struct btree_iter *iter;
558 struct data_opts data_opts;
559 enum data_cmd data_cmd;
560 u64 delay, cur_inum = U64_MAX;
563 bch2_bkey_buf_init(&sk);
564 bch2_trans_init(&trans, c, 0, 0);
566 stats->data_type = BCH_DATA_user;
567 stats->btree_id = btree_id;
570 iter = bch2_trans_get_iter(&trans, btree_id, start,
571 BTREE_ITER_PREFETCH);
574 bch2_ratelimit_reset(rate);
578 delay = rate ? bch2_ratelimit_delay(rate) : 0;
581 bch2_trans_unlock(&trans);
582 set_current_state(TASK_INTERRUPTIBLE);
585 if (kthread && (ret = kthread_should_stop())) {
586 __set_current_state(TASK_RUNNING);
591 schedule_timeout(delay);
593 if (unlikely(freezing(current))) {
594 bch2_trans_unlock(&trans);
595 move_ctxt_wait_event(ctxt, list_empty(&ctxt->reads));
600 k = bch2_btree_iter_peek(iter);
602 stats->pos = iter->pos;
609 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
612 if (!bkey_extent_is_direct_data(k.k))
615 if (btree_id == BTREE_ID_extents &&
616 cur_inum != k.k->p.inode) {
617 struct bch_inode_unpacked inode;
619 io_opts = bch2_opts_to_inode_opts(c->opts);
621 ret = lookup_inode(&trans,
622 SPOS(0, k.k->p.inode, k.k->p.snapshot),
628 bch2_io_opts_apply(&io_opts, bch2_inode_opts_get(&inode));
630 cur_inum = k.k->p.inode;
633 switch ((data_cmd = pred(c, arg, k, &io_opts, &data_opts))) {
638 case DATA_ADD_REPLICAS:
646 /* unlock before doing IO: */
647 bch2_bkey_buf_reassemble(&sk, c, k);
648 k = bkey_i_to_s_c(sk.k);
649 bch2_trans_unlock(&trans);
651 ret2 = bch2_move_extent(&trans, ctxt, wp, io_opts, btree_id, k,
652 data_cmd, data_opts);
654 if (ret2 == -EINTR) {
655 bch2_trans_reset(&trans, 0);
656 bch2_trans_cond_resched(&trans);
660 if (ret2 == -ENOMEM) {
661 /* memory allocation failure, wait for some IO to finish */
662 bch2_move_ctxt_wait_for_io(ctxt);
666 /* XXX signal failure */
671 bch2_ratelimit_increment(rate, k.k->size);
673 atomic64_add(k.k->size * bch2_bkey_nr_ptrs_allocated(k),
674 &stats->sectors_seen);
676 bch2_btree_iter_advance(iter);
677 bch2_trans_cond_resched(&trans);
681 bch2_trans_iter_put(&trans, iter);
682 ret = bch2_trans_exit(&trans) ?: ret;
683 bch2_bkey_buf_exit(&sk, c);
688 int bch2_move_data(struct bch_fs *c,
689 enum btree_id start_btree_id, struct bpos start_pos,
690 enum btree_id end_btree_id, struct bpos end_pos,
691 struct bch_ratelimit *rate,
692 struct write_point_specifier wp,
693 move_pred_fn pred, void *arg,
694 struct bch_move_stats *stats)
696 struct moving_context ctxt = { .stats = stats };
700 closure_init_stack(&ctxt.cl);
701 INIT_LIST_HEAD(&ctxt.reads);
702 init_waitqueue_head(&ctxt.wait);
704 stats->data_type = BCH_DATA_user;
706 for (id = start_btree_id;
707 id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
709 stats->btree_id = id;
711 if (id != BTREE_ID_extents &&
712 id != BTREE_ID_reflink)
715 ret = __bch2_move_data(c, &ctxt, rate, wp,
716 id == start_btree_id ? start_pos : POS_MIN,
717 id == end_btree_id ? end_pos : POS_MAX,
718 pred, arg, stats, id);
724 move_ctxt_wait_event(&ctxt, list_empty(&ctxt.reads));
725 closure_sync(&ctxt.cl);
727 EBUG_ON(atomic_read(&ctxt.write_sectors));
730 atomic64_read(&stats->sectors_moved),
731 atomic64_read(&stats->keys_moved));
736 typedef enum data_cmd (*move_btree_pred)(struct bch_fs *, void *,
737 struct btree *, struct bch_io_opts *,
740 static int bch2_move_btree(struct bch_fs *c,
741 enum btree_id start_btree_id, struct bpos start_pos,
742 enum btree_id end_btree_id, struct bpos end_pos,
743 move_btree_pred pred, void *arg,
744 struct bch_move_stats *stats)
746 bool kthread = (current->flags & PF_KTHREAD) != 0;
747 struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
748 struct btree_trans trans;
749 struct btree_iter *iter;
752 struct data_opts data_opts;
756 bch2_trans_init(&trans, c, 0, 0);
758 stats->data_type = BCH_DATA_btree;
760 for (id = start_btree_id;
761 id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
763 stats->btree_id = id;
765 for_each_btree_node(&trans, iter, id,
766 id == start_btree_id ? start_pos : POS_MIN,
767 BTREE_ITER_PREFETCH, b) {
768 if (kthread && kthread_should_stop())
771 if ((cmp_int(id, end_btree_id) ?:
772 bpos_cmp(b->key.k.p, end_pos)) > 0)
775 stats->pos = iter->pos;
777 switch ((cmd = pred(c, arg, b, &io_opts, &data_opts))) {
782 case DATA_ADD_REPLICAS:
789 ret = bch2_btree_node_rewrite(&trans, iter,
790 b->data->keys.seq, 0) ?: ret;
792 bch2_trans_cond_resched(&trans);
795 ret = bch2_trans_iter_free(&trans, iter) ?: ret;
796 if (kthread && kthread_should_stop())
800 bch2_trans_exit(&trans);
803 bch_err(c, "error %i in bch2_move_btree", ret);
809 static enum data_cmd scrub_pred(struct bch_fs *c, void *arg,
811 struct bch_io_opts *io_opts,
812 struct data_opts *data_opts)
818 static enum data_cmd rereplicate_pred(struct bch_fs *c, void *arg,
820 struct bch_io_opts *io_opts,
821 struct data_opts *data_opts)
823 unsigned nr_good = bch2_bkey_durability(c, k);
824 unsigned replicas = 0;
827 case KEY_TYPE_btree_ptr:
828 replicas = c->opts.metadata_replicas;
830 case KEY_TYPE_extent:
831 replicas = io_opts->data_replicas;
835 if (!nr_good || nr_good >= replicas)
838 data_opts->target = 0;
839 data_opts->nr_replicas = 1;
840 data_opts->btree_insert_flags = 0;
841 return DATA_ADD_REPLICAS;
844 static enum data_cmd migrate_pred(struct bch_fs *c, void *arg,
846 struct bch_io_opts *io_opts,
847 struct data_opts *data_opts)
849 struct bch_ioctl_data *op = arg;
851 if (!bch2_bkey_has_device(k, op->migrate.dev))
854 data_opts->target = 0;
855 data_opts->nr_replicas = 1;
856 data_opts->btree_insert_flags = 0;
857 data_opts->rewrite_dev = op->migrate.dev;
861 static enum data_cmd rereplicate_btree_pred(struct bch_fs *c, void *arg,
863 struct bch_io_opts *io_opts,
864 struct data_opts *data_opts)
866 return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
869 static enum data_cmd migrate_btree_pred(struct bch_fs *c, void *arg,
871 struct bch_io_opts *io_opts,
872 struct data_opts *data_opts)
874 return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
877 static bool bformat_needs_redo(struct bkey_format *f)
881 for (i = 0; i < f->nr_fields; i++) {
882 unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
883 u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
884 u64 field_offset = le64_to_cpu(f->field_offset[i]);
886 if (f->bits_per_field[i] > unpacked_bits)
889 if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
892 if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
901 static enum data_cmd rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
903 struct bch_io_opts *io_opts,
904 struct data_opts *data_opts)
906 if (b->version_ondisk != c->sb.version ||
907 btree_node_need_rewrite(b) ||
908 bformat_needs_redo(&b->format)) {
909 data_opts->target = 0;
910 data_opts->nr_replicas = 1;
911 data_opts->btree_insert_flags = 0;
918 int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
922 ret = bch2_move_btree(c,
924 BTREE_ID_NR, SPOS_MAX,
925 rewrite_old_nodes_pred, c, stats);
927 mutex_lock(&c->sb_lock);
928 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
929 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
930 c->disk_sb.sb->version_min = c->disk_sb.sb->version;
932 mutex_unlock(&c->sb_lock);
938 int bch2_data_job(struct bch_fs *c,
939 struct bch_move_stats *stats,
940 struct bch_ioctl_data op)
945 case BCH_DATA_OP_REREPLICATE:
946 stats->data_type = BCH_DATA_journal;
947 ret = bch2_journal_flush_device_pins(&c->journal, -1);
949 ret = bch2_move_btree(c,
950 op.start_btree, op.start_pos,
951 op.end_btree, op.end_pos,
952 rereplicate_btree_pred, c, stats) ?: ret;
954 closure_wait_event(&c->btree_interior_update_wait,
955 !bch2_btree_interior_updates_nr_pending(c));
957 ret = bch2_replicas_gc2(c) ?: ret;
959 ret = bch2_move_data(c,
960 op.start_btree, op.start_pos,
961 op.end_btree, op.end_pos,
962 NULL, writepoint_hashed((unsigned long) current),
963 rereplicate_pred, c, stats) ?: ret;
964 ret = bch2_replicas_gc2(c) ?: ret;
966 case BCH_DATA_OP_MIGRATE:
967 if (op.migrate.dev >= c->sb.nr_devices)
970 stats->data_type = BCH_DATA_journal;
971 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
973 ret = bch2_move_btree(c,
974 op.start_btree, op.start_pos,
975 op.end_btree, op.end_pos,
976 migrate_btree_pred, &op, stats) ?: ret;
977 ret = bch2_replicas_gc2(c) ?: ret;
979 ret = bch2_move_data(c,
980 op.start_btree, op.start_pos,
981 op.end_btree, op.end_pos,
982 NULL, writepoint_hashed((unsigned long) current),
983 migrate_pred, &op, stats) ?: ret;
984 ret = bch2_replicas_gc2(c) ?: ret;
986 case BCH_DATA_OP_REWRITE_OLD_NODES:
987 ret = bch2_scan_old_btree_nodes(c, stats);