1 // SPDX-License-Identifier: GPL-2.0
4 #include "alloc_background.h"
5 #include "alloc_foreground.h"
6 #include "backpointers.h"
9 #include "btree_update.h"
10 #include "btree_update_interior.h"
11 #include "btree_write_buffer.h"
12 #include "disk_groups.h"
18 #include "journal_reclaim.h"
24 #include <linux/ioprio.h>
25 #include <linux/kthread.h>
27 #include <trace/events/bcachefs.h>
29 static void progress_list_add(struct bch_fs *c, struct bch_move_stats *stats)
31 mutex_lock(&c->data_progress_lock);
32 list_add(&stats->list, &c->data_progress_list);
33 mutex_unlock(&c->data_progress_lock);
36 static void progress_list_del(struct bch_fs *c, struct bch_move_stats *stats)
38 mutex_lock(&c->data_progress_lock);
39 list_del(&stats->list);
40 mutex_unlock(&c->data_progress_lock);
44 struct list_head list;
48 unsigned read_sectors;
49 unsigned write_sectors;
51 struct bch_read_bio rbio;
53 struct data_update write;
54 /* Must be last since it is variable size */
55 struct bio_vec bi_inline_vecs[0];
58 static void move_free(struct moving_io *io)
60 struct moving_context *ctxt = io->write.ctxt;
61 struct bch_fs *c = ctxt->c;
63 bch2_data_update_exit(&io->write);
65 bch2_write_ref_put(c, BCH_WRITE_REF_move);
69 static void move_write_done(struct bch_write_op *op)
71 struct moving_io *io = container_of(op, struct moving_io, write.op);
72 struct moving_context *ctxt = io->write.ctxt;
74 if (io->write.op.error)
75 ctxt->write_error = true;
77 atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
78 atomic_dec(&io->write.ctxt->write_ios);
80 closure_put(&ctxt->cl);
83 static void move_write(struct moving_io *io)
85 if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
90 closure_get(&io->write.ctxt->cl);
91 atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
92 atomic_inc(&io->write.ctxt->write_ios);
94 bch2_data_update_read_done(&io->write, io->rbio.pick.crc);
97 struct moving_io *bch2_moving_ctxt_next_pending_write(struct moving_context *ctxt)
99 struct moving_io *io =
100 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
102 return io && io->read_completed ? io : NULL;
105 static void move_read_endio(struct bio *bio)
107 struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
108 struct moving_context *ctxt = io->write.ctxt;
110 atomic_sub(io->read_sectors, &ctxt->read_sectors);
111 atomic_dec(&ctxt->read_ios);
112 io->read_completed = true;
114 wake_up(&ctxt->wait);
115 closure_put(&ctxt->cl);
118 void bch2_moving_ctxt_do_pending_writes(struct moving_context *ctxt,
119 struct btree_trans *trans)
121 struct moving_io *io;
124 bch2_trans_unlock(trans);
126 while ((io = bch2_moving_ctxt_next_pending_write(ctxt))) {
132 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt,
133 struct btree_trans *trans)
135 unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
137 move_ctxt_wait_event(ctxt, trans,
138 !atomic_read(&ctxt->write_sectors) ||
139 atomic_read(&ctxt->write_sectors) != sectors_pending);
142 void bch2_moving_ctxt_exit(struct moving_context *ctxt)
144 move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
145 closure_sync(&ctxt->cl);
147 EBUG_ON(atomic_read(&ctxt->write_sectors));
148 EBUG_ON(atomic_read(&ctxt->write_ios));
149 EBUG_ON(atomic_read(&ctxt->read_sectors));
150 EBUG_ON(atomic_read(&ctxt->read_ios));
153 progress_list_del(ctxt->c, ctxt->stats);
155 trace_move_data(ctxt->c,
156 atomic64_read(&ctxt->stats->sectors_moved),
157 atomic64_read(&ctxt->stats->keys_moved));
161 void bch2_moving_ctxt_init(struct moving_context *ctxt,
163 struct bch_ratelimit *rate,
164 struct bch_move_stats *stats,
165 struct write_point_specifier wp,
168 memset(ctxt, 0, sizeof(*ctxt));
174 ctxt->wait_on_copygc = wait_on_copygc;
176 closure_init_stack(&ctxt->cl);
177 INIT_LIST_HEAD(&ctxt->reads);
178 init_waitqueue_head(&ctxt->wait);
181 progress_list_add(c, stats);
182 stats->data_type = BCH_DATA_user;
186 void bch2_move_stats_init(struct bch_move_stats *stats, char *name)
188 memset(stats, 0, sizeof(*stats));
189 scnprintf(stats->name, sizeof(stats->name), "%s", name);
192 static int bch2_extent_drop_ptrs(struct btree_trans *trans,
193 struct btree_iter *iter,
195 struct data_update_opts data_opts)
197 struct bch_fs *c = trans->c;
201 n = bch2_bkey_make_mut(trans, k);
202 ret = PTR_ERR_OR_ZERO(n);
206 while (data_opts.kill_ptrs) {
207 unsigned i = 0, drop = __fls(data_opts.kill_ptrs);
208 struct bch_extent_ptr *ptr;
210 bch2_bkey_drop_ptrs(bkey_i_to_s(n), ptr, i++ == drop);
211 data_opts.kill_ptrs ^= 1U << drop;
215 * If the new extent no longer has any pointers, bch2_extent_normalize()
216 * will do the appropriate thing with it (turning it into a
217 * KEY_TYPE_error key, or just a discard if it was a cached extent)
219 bch2_extent_normalize(c, bkey_i_to_s(n));
222 * Since we're not inserting through an extent iterator
223 * (BTREE_ITER_ALL_SNAPSHOTS iterators aren't extent iterators),
224 * we aren't using the extent overwrite path to delete, we're
225 * just using the normal key deletion path:
227 if (bkey_deleted(&n->k))
230 return bch2_trans_relock(trans) ?:
231 bch2_trans_update(trans, iter, n, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
232 bch2_trans_commit(trans, NULL, NULL, BTREE_INSERT_NOFAIL);
235 static int bch2_move_extent(struct btree_trans *trans,
236 struct btree_iter *iter,
237 struct moving_context *ctxt,
238 struct bch_io_opts io_opts,
239 enum btree_id btree_id,
241 struct data_update_opts data_opts)
243 struct bch_fs *c = trans->c;
244 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
245 struct moving_io *io;
246 const union bch_extent_entry *entry;
247 struct extent_ptr_decoded p;
248 unsigned sectors = k.k->size, pages;
251 bch2_data_update_opts_normalize(k, &data_opts);
253 if (!data_opts.rewrite_ptrs &&
254 !data_opts.extra_replicas) {
255 if (data_opts.kill_ptrs)
256 return bch2_extent_drop_ptrs(trans, iter, k, data_opts);
260 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_move))
261 return -BCH_ERR_erofs_no_writes;
264 * Before memory allocations & taking nocow locks in
265 * bch2_data_update_init():
267 bch2_trans_unlock(trans);
269 /* write path might have to decompress data: */
270 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
271 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
273 pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
274 io = kzalloc(sizeof(struct moving_io) +
275 sizeof(struct bio_vec) * pages, GFP_KERNEL);
279 io->write.ctxt = ctxt;
280 io->read_sectors = k.k->size;
281 io->write_sectors = k.k->size;
283 bio_init(&io->write.op.wbio.bio, NULL, io->bi_inline_vecs, pages, 0);
284 bio_set_prio(&io->write.op.wbio.bio,
285 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
287 if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
292 io->rbio.opts = io_opts;
293 bio_init(&io->rbio.bio, NULL, io->bi_inline_vecs, pages, 0);
294 io->rbio.bio.bi_vcnt = pages;
295 bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
296 io->rbio.bio.bi_iter.bi_size = sectors << 9;
298 bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
299 io->rbio.bio.bi_iter.bi_sector = bkey_start_offset(k.k);
300 io->rbio.bio.bi_end_io = move_read_endio;
302 ret = bch2_data_update_init(trans, ctxt, &io->write, ctxt->wp,
303 io_opts, data_opts, btree_id, k);
304 if (ret && ret != -BCH_ERR_unwritten_extent_update)
307 if (ret == -BCH_ERR_unwritten_extent_update) {
308 bch2_update_unwritten_extent(trans, &io->write);
315 io->write.ctxt = ctxt;
316 io->write.op.end_io = move_write_done;
319 atomic64_inc(&ctxt->stats->keys_moved);
320 atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
323 this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
324 this_cpu_add(c->counters[BCH_COUNTER_move_extent_read], k.k->size);
325 trace_move_extent_read(k.k);
327 atomic_add(io->read_sectors, &ctxt->read_sectors);
328 atomic_inc(&ctxt->read_ios);
329 list_add_tail(&io->list, &ctxt->reads);
332 * dropped by move_read_endio() - guards against use after free of
333 * ctxt when doing wakeup
335 closure_get(&ctxt->cl);
336 bch2_read_extent(trans, &io->rbio,
340 BCH_READ_LAST_FRAGMENT);
343 bio_free_pages(&io->write.op.wbio.bio);
347 bch2_write_ref_put(c, BCH_WRITE_REF_move);
348 trace_and_count(c, move_extent_alloc_mem_fail, k.k);
352 static int lookup_inode(struct btree_trans *trans, struct bpos pos,
353 struct bch_inode_unpacked *inode)
355 struct btree_iter iter;
359 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
360 BTREE_ITER_ALL_SNAPSHOTS);
361 k = bch2_btree_iter_peek(&iter);
366 if (!k.k || !bkey_eq(k.k->p, pos)) {
371 ret = bkey_is_inode(k.k) ? 0 : -EIO;
375 ret = bch2_inode_unpack(k, inode);
379 bch2_trans_iter_exit(trans, &iter);
383 static int move_ratelimit(struct btree_trans *trans,
384 struct moving_context *ctxt)
386 struct bch_fs *c = trans->c;
389 if (ctxt->wait_on_copygc) {
390 bch2_trans_unlock(trans);
391 wait_event_killable(c->copygc_running_wq,
392 !c->copygc_running ||
393 kthread_should_stop());
397 delay = ctxt->rate ? bch2_ratelimit_delay(ctxt->rate) : 0;
400 bch2_trans_unlock(trans);
401 set_current_state(TASK_INTERRUPTIBLE);
404 if ((current->flags & PF_KTHREAD) && kthread_should_stop()) {
405 __set_current_state(TASK_RUNNING);
410 schedule_timeout(delay);
412 if (unlikely(freezing(current))) {
413 move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
419 * XXX: these limits really ought to be per device, SSDs and hard drives
420 * will want different limits
422 move_ctxt_wait_event(ctxt, trans,
423 atomic_read(&ctxt->write_sectors) < c->opts.move_bytes_in_flight >> 9 &&
424 atomic_read(&ctxt->read_sectors) < c->opts.move_bytes_in_flight >> 9 &&
425 atomic_read(&ctxt->write_ios) < c->opts.move_ios_in_flight &&
426 atomic_read(&ctxt->read_ios) < c->opts.move_ios_in_flight);
431 static int move_get_io_opts(struct btree_trans *trans,
432 struct bch_io_opts *io_opts,
433 struct bkey_s_c k, u64 *cur_inum)
435 struct bch_inode_unpacked inode;
438 if (*cur_inum == k.k->p.inode)
441 ret = lookup_inode(trans,
442 SPOS(0, k.k->p.inode, k.k->p.snapshot),
444 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
448 bch2_inode_opts_get(io_opts, trans->c, &inode);
450 *io_opts = bch2_opts_to_inode_opts(trans->c->opts);
451 *cur_inum = k.k->p.inode;
455 static int __bch2_move_data(struct moving_context *ctxt,
458 move_pred_fn pred, void *arg,
459 enum btree_id btree_id)
461 struct bch_fs *c = ctxt->c;
462 struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
464 struct btree_trans trans;
465 struct btree_iter iter;
467 struct data_update_opts data_opts;
468 u64 cur_inum = U64_MAX;
471 bch2_bkey_buf_init(&sk);
472 bch2_trans_init(&trans, c, 0, 0);
475 ctxt->stats->data_type = BCH_DATA_user;
476 ctxt->stats->btree_id = btree_id;
477 ctxt->stats->pos = start;
480 bch2_trans_iter_init(&trans, &iter, btree_id, start,
482 BTREE_ITER_ALL_SNAPSHOTS);
485 bch2_ratelimit_reset(ctxt->rate);
487 while (!move_ratelimit(&trans, ctxt)) {
488 bch2_trans_begin(&trans);
490 k = bch2_btree_iter_peek(&iter);
495 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
500 if (bkey_ge(bkey_start_pos(k.k), end))
504 ctxt->stats->pos = iter.pos;
506 if (!bkey_extent_is_direct_data(k.k))
509 ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
513 memset(&data_opts, 0, sizeof(data_opts));
514 if (!pred(c, arg, k, &io_opts, &data_opts))
518 * The iterator gets unlocked by __bch2_read_extent - need to
519 * save a copy of @k elsewhere:
521 bch2_bkey_buf_reassemble(&sk, c, k);
522 k = bkey_i_to_s_c(sk.k);
523 bch2_trans_unlock(&trans);
525 ret2 = bch2_move_extent(&trans, &iter, ctxt, io_opts,
526 btree_id, k, data_opts);
528 if (bch2_err_matches(ret2, BCH_ERR_transaction_restart))
531 if (ret2 == -ENOMEM) {
532 /* memory allocation failure, wait for some IO to finish */
533 bch2_move_ctxt_wait_for_io(ctxt, &trans);
537 /* XXX signal failure */
542 bch2_ratelimit_increment(ctxt->rate, k.k->size);
545 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
547 bch2_btree_iter_advance(&iter);
550 bch2_trans_iter_exit(&trans, &iter);
551 bch2_trans_exit(&trans);
552 bch2_bkey_buf_exit(&sk, c);
557 int bch2_move_data(struct bch_fs *c,
558 enum btree_id start_btree_id, struct bpos start_pos,
559 enum btree_id end_btree_id, struct bpos end_pos,
560 struct bch_ratelimit *rate,
561 struct bch_move_stats *stats,
562 struct write_point_specifier wp,
564 move_pred_fn pred, void *arg)
566 struct moving_context ctxt;
570 bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
572 for (id = start_btree_id;
573 id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
575 stats->btree_id = id;
577 if (id != BTREE_ID_extents &&
578 id != BTREE_ID_reflink)
581 ret = __bch2_move_data(&ctxt,
582 id == start_btree_id ? start_pos : POS_MIN,
583 id == end_btree_id ? end_pos : POS_MAX,
589 bch2_moving_ctxt_exit(&ctxt);
594 void bch2_verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
596 struct bch_fs *c = trans->c;
597 struct btree_iter iter;
599 struct printbuf buf = PRINTBUF;
600 struct bch_backpointer bp;
604 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
605 bucket, BTREE_ITER_CACHED);
607 ret = lockrestart_do(trans,
608 bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
610 if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
611 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
613 if (a.v->gen == gen &&
614 a.v->dirty_sectors) {
615 if (a.v->data_type == BCH_DATA_btree) {
616 bch2_trans_unlock(trans);
617 if (bch2_btree_interior_updates_flush(c))
619 goto failed_to_evacuate;
624 bch2_trans_iter_exit(trans, &iter);
627 bch2_trans_iter_exit(trans, &iter);
629 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
632 prt_printf(&buf, bch2_log_msg(c, "failed to evacuate bucket "));
633 bch2_bkey_val_to_text(&buf, c, k);
636 bch2_trans_begin(trans);
638 ret = bch2_get_next_backpointer(trans, bucket, gen,
641 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
645 if (bp_offset == U64_MAX)
648 k = bch2_backpointer_get_key(trans, &iter,
649 bucket, bp_offset, bp);
651 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
658 bch2_bkey_val_to_text(&buf, c, k);
659 bch2_trans_iter_exit(trans, &iter);
662 bch2_print_string_as_lines(KERN_ERR, buf.buf);
666 int __bch2_evacuate_bucket(struct btree_trans *trans,
667 struct moving_context *ctxt,
668 struct bpos bucket, int gen,
669 struct data_update_opts _data_opts)
671 struct bch_fs *c = ctxt->c;
672 struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
673 struct btree_iter iter;
675 struct bch_backpointer bp;
676 struct bch_alloc_v4 a_convert;
677 const struct bch_alloc_v4 *a;
679 struct data_update_opts data_opts;
680 unsigned dirty_sectors, bucket_size;
682 u64 bp_offset = 0, cur_inum = U64_MAX;
685 bch2_bkey_buf_init(&sk);
687 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
688 bucket, BTREE_ITER_CACHED);
689 ret = lockrestart_do(trans,
690 bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
691 bch2_trans_iter_exit(trans, &iter);
694 bch_err(c, "%s: error looking up alloc key: %s", __func__, bch2_err_str(ret));
698 a = bch2_alloc_to_v4(k, &a_convert);
699 dirty_sectors = a->dirty_sectors;
700 bucket_size = bch_dev_bkey_exists(c, bucket.inode)->mi.bucket_size;
701 fragmentation = a->fragmentation_lru;
703 ret = bch2_btree_write_buffer_flush(trans);
705 bch_err(c, "%s: error flushing btree write buffer: %s", __func__, bch2_err_str(ret));
709 while (!(ret = move_ratelimit(trans, ctxt))) {
710 bch2_trans_begin(trans);
712 ret = bch2_get_next_backpointer(trans, bucket, gen,
715 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
719 if (bp_offset == U64_MAX)
723 const struct bch_extent_ptr *ptr;
727 k = bch2_backpointer_get_key(trans, &iter,
728 bucket, bp_offset, bp);
730 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
737 bch2_bkey_buf_reassemble(&sk, c, k);
738 k = bkey_i_to_s_c(sk.k);
740 ret = move_get_io_opts(trans, &io_opts, k, &cur_inum);
742 bch2_trans_iter_exit(trans, &iter);
746 data_opts = _data_opts;
747 data_opts.target = io_opts.background_target;
748 data_opts.rewrite_ptrs = 0;
750 bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
751 if (ptr->dev == bucket.inode)
752 data_opts.rewrite_ptrs |= 1U << i;
756 ret = bch2_move_extent(trans, &iter, ctxt, io_opts,
757 bp.btree_id, k, data_opts);
758 bch2_trans_iter_exit(trans, &iter);
760 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
762 if (ret == -ENOMEM) {
763 /* memory allocation failure, wait for some IO to finish */
764 bch2_move_ctxt_wait_for_io(ctxt, trans);
771 bch2_ratelimit_increment(ctxt->rate, k.k->size);
773 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
777 b = bch2_backpointer_get_node(trans, &iter,
778 bucket, bp_offset, bp);
779 ret = PTR_ERR_OR_ZERO(b);
780 if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
782 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
789 ret = bch2_btree_node_rewrite(trans, &iter, b, 0);
790 bch2_trans_iter_exit(trans, &iter);
792 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
798 bch2_ratelimit_increment(ctxt->rate,
799 c->opts.btree_node_size >> 9);
801 atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_seen);
802 atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_moved);
809 trace_evacuate_bucket(c, &bucket, dirty_sectors, bucket_size, fragmentation, ret);
811 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
812 bch2_trans_unlock(trans);
813 move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
814 closure_sync(&ctxt->cl);
815 if (!ctxt->write_error)
816 bch2_verify_bucket_evacuated(trans, bucket, gen);
819 bch2_bkey_buf_exit(&sk, c);
823 int bch2_evacuate_bucket(struct bch_fs *c,
824 struct bpos bucket, int gen,
825 struct data_update_opts data_opts,
826 struct bch_ratelimit *rate,
827 struct bch_move_stats *stats,
828 struct write_point_specifier wp,
831 struct btree_trans trans;
832 struct moving_context ctxt;
835 bch2_trans_init(&trans, c, 0, 0);
836 bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
837 ret = __bch2_evacuate_bucket(&trans, &ctxt, bucket, gen, data_opts);
838 bch2_moving_ctxt_exit(&ctxt);
839 bch2_trans_exit(&trans);
844 typedef bool (*move_btree_pred)(struct bch_fs *, void *,
845 struct btree *, struct bch_io_opts *,
846 struct data_update_opts *);
848 static int bch2_move_btree(struct bch_fs *c,
849 enum btree_id start_btree_id, struct bpos start_pos,
850 enum btree_id end_btree_id, struct bpos end_pos,
851 move_btree_pred pred, void *arg,
852 struct bch_move_stats *stats)
854 bool kthread = (current->flags & PF_KTHREAD) != 0;
855 struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
856 struct btree_trans trans;
857 struct btree_iter iter;
860 struct data_update_opts data_opts;
863 bch2_trans_init(&trans, c, 0, 0);
864 progress_list_add(c, stats);
866 stats->data_type = BCH_DATA_btree;
868 for (id = start_btree_id;
869 id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
871 stats->btree_id = id;
873 bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
874 BTREE_ITER_PREFETCH);
877 while (bch2_trans_begin(&trans),
878 (b = bch2_btree_iter_peek_node(&iter)) &&
879 !(ret = PTR_ERR_OR_ZERO(b))) {
880 if (kthread && kthread_should_stop())
883 if ((cmp_int(id, end_btree_id) ?:
884 bpos_cmp(b->key.k.p, end_pos)) > 0)
887 stats->pos = iter.pos;
889 if (!pred(c, arg, b, &io_opts, &data_opts))
892 ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
893 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
898 bch2_btree_iter_next_node(&iter);
900 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
903 bch2_trans_iter_exit(&trans, &iter);
905 if (kthread && kthread_should_stop())
909 bch2_trans_exit(&trans);
912 bch_err(c, "error in %s(): %s", __func__, bch2_err_str(ret));
914 bch2_btree_interior_updates_flush(c);
916 progress_list_del(c, stats);
920 static bool rereplicate_pred(struct bch_fs *c, void *arg,
922 struct bch_io_opts *io_opts,
923 struct data_update_opts *data_opts)
925 unsigned nr_good = bch2_bkey_durability(c, k);
926 unsigned replicas = bkey_is_btree_ptr(k.k)
927 ? c->opts.metadata_replicas
928 : io_opts->data_replicas;
930 if (!nr_good || nr_good >= replicas)
933 data_opts->target = 0;
934 data_opts->extra_replicas = replicas - nr_good;
935 data_opts->btree_insert_flags = 0;
939 static bool migrate_pred(struct bch_fs *c, void *arg,
941 struct bch_io_opts *io_opts,
942 struct data_update_opts *data_opts)
944 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
945 const struct bch_extent_ptr *ptr;
946 struct bch_ioctl_data *op = arg;
949 data_opts->rewrite_ptrs = 0;
950 data_opts->target = 0;
951 data_opts->extra_replicas = 0;
952 data_opts->btree_insert_flags = 0;
954 bkey_for_each_ptr(ptrs, ptr) {
955 if (ptr->dev == op->migrate.dev)
956 data_opts->rewrite_ptrs |= 1U << i;
960 return data_opts->rewrite_ptrs != 0;
963 static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
965 struct bch_io_opts *io_opts,
966 struct data_update_opts *data_opts)
968 return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
971 static bool migrate_btree_pred(struct bch_fs *c, void *arg,
973 struct bch_io_opts *io_opts,
974 struct data_update_opts *data_opts)
976 return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
979 static bool bformat_needs_redo(struct bkey_format *f)
983 for (i = 0; i < f->nr_fields; i++) {
984 unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
985 u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
986 u64 field_offset = le64_to_cpu(f->field_offset[i]);
988 if (f->bits_per_field[i] > unpacked_bits)
991 if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
994 if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
1003 static bool rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
1005 struct bch_io_opts *io_opts,
1006 struct data_update_opts *data_opts)
1008 if (b->version_ondisk != c->sb.version ||
1009 btree_node_need_rewrite(b) ||
1010 bformat_needs_redo(&b->format)) {
1011 data_opts->target = 0;
1012 data_opts->extra_replicas = 0;
1013 data_opts->btree_insert_flags = 0;
1020 int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
1024 ret = bch2_move_btree(c,
1026 BTREE_ID_NR, SPOS_MAX,
1027 rewrite_old_nodes_pred, c, stats);
1029 mutex_lock(&c->sb_lock);
1030 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1031 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1032 c->disk_sb.sb->version_min = c->disk_sb.sb->version;
1033 bch2_write_super(c);
1034 mutex_unlock(&c->sb_lock);
1040 int bch2_data_job(struct bch_fs *c,
1041 struct bch_move_stats *stats,
1042 struct bch_ioctl_data op)
1047 case BCH_DATA_OP_REREPLICATE:
1048 bch2_move_stats_init(stats, "rereplicate");
1049 stats->data_type = BCH_DATA_journal;
1050 ret = bch2_journal_flush_device_pins(&c->journal, -1);
1052 ret = bch2_move_btree(c,
1053 op.start_btree, op.start_pos,
1054 op.end_btree, op.end_pos,
1055 rereplicate_btree_pred, c, stats) ?: ret;
1056 ret = bch2_replicas_gc2(c) ?: ret;
1058 ret = bch2_move_data(c,
1059 op.start_btree, op.start_pos,
1060 op.end_btree, op.end_pos,
1063 writepoint_hashed((unsigned long) current),
1065 rereplicate_pred, c) ?: ret;
1066 ret = bch2_replicas_gc2(c) ?: ret;
1068 case BCH_DATA_OP_MIGRATE:
1069 if (op.migrate.dev >= c->sb.nr_devices)
1072 bch2_move_stats_init(stats, "migrate");
1073 stats->data_type = BCH_DATA_journal;
1074 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
1076 ret = bch2_move_btree(c,
1077 op.start_btree, op.start_pos,
1078 op.end_btree, op.end_pos,
1079 migrate_btree_pred, &op, stats) ?: ret;
1080 ret = bch2_replicas_gc2(c) ?: ret;
1082 ret = bch2_move_data(c,
1083 op.start_btree, op.start_pos,
1084 op.end_btree, op.end_pos,
1087 writepoint_hashed((unsigned long) current),
1089 migrate_pred, &op) ?: ret;
1090 ret = bch2_replicas_gc2(c) ?: ret;
1092 case BCH_DATA_OP_REWRITE_OLD_NODES:
1093 bch2_move_stats_init(stats, "rewrite_old_nodes");
1094 ret = bch2_scan_old_btree_nodes(c, stats);