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;
45 struct move_bucket_in_flight *b;
49 unsigned read_sectors;
50 unsigned write_sectors;
52 struct bch_read_bio rbio;
54 struct data_update write;
55 /* Must be last since it is variable size */
56 struct bio_vec bi_inline_vecs[0];
59 static void move_free(struct moving_io *io)
61 struct moving_context *ctxt = io->write.ctxt;
62 struct bch_fs *c = ctxt->c;
65 atomic_dec(&io->b->count);
67 bch2_data_update_exit(&io->write);
69 bch2_write_ref_put(c, BCH_WRITE_REF_move);
73 static void move_write_done(struct bch_write_op *op)
75 struct moving_io *io = container_of(op, struct moving_io, write.op);
76 struct moving_context *ctxt = io->write.ctxt;
78 if (io->write.op.error)
79 ctxt->write_error = true;
81 atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
82 atomic_dec(&io->write.ctxt->write_ios);
84 closure_put(&ctxt->cl);
87 static void move_write(struct moving_io *io)
89 if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
94 closure_get(&io->write.ctxt->cl);
95 atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
96 atomic_inc(&io->write.ctxt->write_ios);
98 bch2_data_update_read_done(&io->write, io->rbio.pick.crc);
101 struct moving_io *bch2_moving_ctxt_next_pending_write(struct moving_context *ctxt)
103 struct moving_io *io =
104 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
106 return io && io->read_completed ? io : NULL;
109 static void move_read_endio(struct bio *bio)
111 struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
112 struct moving_context *ctxt = io->write.ctxt;
114 atomic_sub(io->read_sectors, &ctxt->read_sectors);
115 atomic_dec(&ctxt->read_ios);
116 io->read_completed = true;
118 wake_up(&ctxt->wait);
119 closure_put(&ctxt->cl);
122 void bch2_moving_ctxt_do_pending_writes(struct moving_context *ctxt,
123 struct btree_trans *trans)
125 struct moving_io *io;
128 bch2_trans_unlock(trans);
130 while ((io = bch2_moving_ctxt_next_pending_write(ctxt))) {
136 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt,
137 struct btree_trans *trans)
139 unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
141 move_ctxt_wait_event(ctxt, trans,
142 !atomic_read(&ctxt->write_sectors) ||
143 atomic_read(&ctxt->write_sectors) != sectors_pending);
146 void bch2_moving_ctxt_exit(struct moving_context *ctxt)
148 move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
149 closure_sync(&ctxt->cl);
151 EBUG_ON(atomic_read(&ctxt->write_sectors));
152 EBUG_ON(atomic_read(&ctxt->write_ios));
153 EBUG_ON(atomic_read(&ctxt->read_sectors));
154 EBUG_ON(atomic_read(&ctxt->read_ios));
157 progress_list_del(ctxt->c, ctxt->stats);
159 trace_move_data(ctxt->c,
160 atomic64_read(&ctxt->stats->sectors_moved),
161 atomic64_read(&ctxt->stats->keys_moved));
165 void bch2_moving_ctxt_init(struct moving_context *ctxt,
167 struct bch_ratelimit *rate,
168 struct bch_move_stats *stats,
169 struct write_point_specifier wp,
172 memset(ctxt, 0, sizeof(*ctxt));
178 ctxt->wait_on_copygc = wait_on_copygc;
180 closure_init_stack(&ctxt->cl);
181 INIT_LIST_HEAD(&ctxt->reads);
182 init_waitqueue_head(&ctxt->wait);
185 progress_list_add(c, stats);
186 stats->data_type = BCH_DATA_user;
190 void bch2_move_stats_init(struct bch_move_stats *stats, char *name)
192 memset(stats, 0, sizeof(*stats));
193 scnprintf(stats->name, sizeof(stats->name), "%s", name);
196 static int bch2_extent_drop_ptrs(struct btree_trans *trans,
197 struct btree_iter *iter,
199 struct data_update_opts data_opts)
201 struct bch_fs *c = trans->c;
205 n = bch2_bkey_make_mut(trans, k);
206 ret = PTR_ERR_OR_ZERO(n);
210 while (data_opts.kill_ptrs) {
211 unsigned i = 0, drop = __fls(data_opts.kill_ptrs);
212 struct bch_extent_ptr *ptr;
214 bch2_bkey_drop_ptrs(bkey_i_to_s(n), ptr, i++ == drop);
215 data_opts.kill_ptrs ^= 1U << drop;
219 * If the new extent no longer has any pointers, bch2_extent_normalize()
220 * will do the appropriate thing with it (turning it into a
221 * KEY_TYPE_error key, or just a discard if it was a cached extent)
223 bch2_extent_normalize(c, bkey_i_to_s(n));
226 * Since we're not inserting through an extent iterator
227 * (BTREE_ITER_ALL_SNAPSHOTS iterators aren't extent iterators),
228 * we aren't using the extent overwrite path to delete, we're
229 * just using the normal key deletion path:
231 if (bkey_deleted(&n->k))
234 return bch2_trans_relock(trans) ?:
235 bch2_trans_update(trans, iter, n, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
236 bch2_trans_commit(trans, NULL, NULL, BTREE_INSERT_NOFAIL);
239 static int bch2_move_extent(struct btree_trans *trans,
240 struct btree_iter *iter,
241 struct moving_context *ctxt,
242 struct move_bucket_in_flight *bucket_in_flight,
243 struct bch_io_opts io_opts,
244 enum btree_id btree_id,
246 struct data_update_opts data_opts)
248 struct bch_fs *c = trans->c;
249 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
250 struct moving_io *io;
251 const union bch_extent_entry *entry;
252 struct extent_ptr_decoded p;
253 unsigned sectors = k.k->size, pages;
256 bch2_data_update_opts_normalize(k, &data_opts);
258 if (!data_opts.rewrite_ptrs &&
259 !data_opts.extra_replicas) {
260 if (data_opts.kill_ptrs)
261 return bch2_extent_drop_ptrs(trans, iter, k, data_opts);
265 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_move))
266 return -BCH_ERR_erofs_no_writes;
269 * Before memory allocations & taking nocow locks in
270 * bch2_data_update_init():
272 bch2_trans_unlock(trans);
274 /* write path might have to decompress data: */
275 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
276 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
278 pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
279 io = kzalloc(sizeof(struct moving_io) +
280 sizeof(struct bio_vec) * pages, GFP_KERNEL);
284 io->write.ctxt = ctxt;
285 io->read_sectors = k.k->size;
286 io->write_sectors = k.k->size;
288 bio_init(&io->write.op.wbio.bio, NULL, io->bi_inline_vecs, pages, 0);
289 bio_set_prio(&io->write.op.wbio.bio,
290 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
292 if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
297 io->rbio.opts = io_opts;
298 bio_init(&io->rbio.bio, NULL, io->bi_inline_vecs, pages, 0);
299 io->rbio.bio.bi_vcnt = pages;
300 bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
301 io->rbio.bio.bi_iter.bi_size = sectors << 9;
303 io->rbio.bio.bi_opf = REQ_OP_READ;
304 io->rbio.bio.bi_iter.bi_sector = bkey_start_offset(k.k);
305 io->rbio.bio.bi_end_io = move_read_endio;
307 ret = bch2_data_update_init(trans, ctxt, &io->write, ctxt->wp,
308 io_opts, data_opts, btree_id, k);
309 if (ret && ret != -BCH_ERR_unwritten_extent_update)
312 if (ret == -BCH_ERR_unwritten_extent_update) {
313 bch2_update_unwritten_extent(trans, &io->write);
320 io->write.ctxt = ctxt;
321 io->write.op.end_io = move_write_done;
324 atomic64_inc(&ctxt->stats->keys_moved);
325 atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
328 if (bucket_in_flight) {
329 io->b = bucket_in_flight;
330 atomic_inc(&io->b->count);
333 this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
334 this_cpu_add(c->counters[BCH_COUNTER_move_extent_read], k.k->size);
335 trace_move_extent_read(k.k);
337 atomic_add(io->read_sectors, &ctxt->read_sectors);
338 atomic_inc(&ctxt->read_ios);
339 list_add_tail(&io->list, &ctxt->reads);
342 * dropped by move_read_endio() - guards against use after free of
343 * ctxt when doing wakeup
345 closure_get(&ctxt->cl);
346 bch2_read_extent(trans, &io->rbio,
350 BCH_READ_LAST_FRAGMENT);
353 bio_free_pages(&io->write.op.wbio.bio);
357 bch2_write_ref_put(c, BCH_WRITE_REF_move);
358 trace_and_count(c, move_extent_alloc_mem_fail, k.k);
362 static int lookup_inode(struct btree_trans *trans, struct bpos pos,
363 struct bch_inode_unpacked *inode)
365 struct btree_iter iter;
369 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
370 BTREE_ITER_ALL_SNAPSHOTS);
371 k = bch2_btree_iter_peek(&iter);
376 if (!k.k || !bkey_eq(k.k->p, pos)) {
381 ret = bkey_is_inode(k.k) ? 0 : -EIO;
385 ret = bch2_inode_unpack(k, inode);
389 bch2_trans_iter_exit(trans, &iter);
393 static int move_ratelimit(struct btree_trans *trans,
394 struct moving_context *ctxt)
396 struct bch_fs *c = trans->c;
399 if (ctxt->wait_on_copygc) {
400 bch2_trans_unlock(trans);
401 wait_event_killable(c->copygc_running_wq,
402 !c->copygc_running ||
403 kthread_should_stop());
407 delay = ctxt->rate ? bch2_ratelimit_delay(ctxt->rate) : 0;
410 bch2_trans_unlock(trans);
411 set_current_state(TASK_INTERRUPTIBLE);
414 if ((current->flags & PF_KTHREAD) && kthread_should_stop()) {
415 __set_current_state(TASK_RUNNING);
420 schedule_timeout(delay);
422 if (unlikely(freezing(current))) {
423 move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
429 * XXX: these limits really ought to be per device, SSDs and hard drives
430 * will want different limits
432 move_ctxt_wait_event(ctxt, trans,
433 atomic_read(&ctxt->write_sectors) < c->opts.move_bytes_in_flight >> 9 &&
434 atomic_read(&ctxt->read_sectors) < c->opts.move_bytes_in_flight >> 9 &&
435 atomic_read(&ctxt->write_ios) < c->opts.move_ios_in_flight &&
436 atomic_read(&ctxt->read_ios) < c->opts.move_ios_in_flight);
441 static int move_get_io_opts(struct btree_trans *trans,
442 struct bch_io_opts *io_opts,
443 struct bkey_s_c k, u64 *cur_inum)
445 struct bch_inode_unpacked inode;
448 if (*cur_inum == k.k->p.inode)
451 ret = lookup_inode(trans,
452 SPOS(0, k.k->p.inode, k.k->p.snapshot),
454 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
458 bch2_inode_opts_get(io_opts, trans->c, &inode);
460 *io_opts = bch2_opts_to_inode_opts(trans->c->opts);
461 *cur_inum = k.k->p.inode;
465 static int __bch2_move_data(struct moving_context *ctxt,
468 move_pred_fn pred, void *arg,
469 enum btree_id btree_id)
471 struct bch_fs *c = ctxt->c;
472 struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
474 struct btree_trans trans;
475 struct btree_iter iter;
477 struct data_update_opts data_opts;
478 u64 cur_inum = U64_MAX;
481 bch2_bkey_buf_init(&sk);
482 bch2_trans_init(&trans, c, 0, 0);
485 ctxt->stats->data_type = BCH_DATA_user;
486 ctxt->stats->btree_id = btree_id;
487 ctxt->stats->pos = start;
490 bch2_trans_iter_init(&trans, &iter, btree_id, start,
492 BTREE_ITER_ALL_SNAPSHOTS);
495 bch2_ratelimit_reset(ctxt->rate);
497 while (!move_ratelimit(&trans, ctxt)) {
498 bch2_trans_begin(&trans);
500 k = bch2_btree_iter_peek(&iter);
505 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
510 if (bkey_ge(bkey_start_pos(k.k), end))
514 ctxt->stats->pos = iter.pos;
516 if (!bkey_extent_is_direct_data(k.k))
519 ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
523 memset(&data_opts, 0, sizeof(data_opts));
524 if (!pred(c, arg, k, &io_opts, &data_opts))
528 * The iterator gets unlocked by __bch2_read_extent - need to
529 * save a copy of @k elsewhere:
531 bch2_bkey_buf_reassemble(&sk, c, k);
532 k = bkey_i_to_s_c(sk.k);
533 bch2_trans_unlock(&trans);
535 ret2 = bch2_move_extent(&trans, &iter, ctxt, NULL,
536 io_opts, btree_id, k, data_opts);
538 if (bch2_err_matches(ret2, BCH_ERR_transaction_restart))
541 if (ret2 == -ENOMEM) {
542 /* memory allocation failure, wait for some IO to finish */
543 bch2_move_ctxt_wait_for_io(ctxt, &trans);
547 /* XXX signal failure */
552 bch2_ratelimit_increment(ctxt->rate, k.k->size);
555 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
557 bch2_btree_iter_advance(&iter);
560 bch2_trans_iter_exit(&trans, &iter);
561 bch2_trans_exit(&trans);
562 bch2_bkey_buf_exit(&sk, c);
567 int bch2_move_data(struct bch_fs *c,
568 enum btree_id start_btree_id, struct bpos start_pos,
569 enum btree_id end_btree_id, struct bpos end_pos,
570 struct bch_ratelimit *rate,
571 struct bch_move_stats *stats,
572 struct write_point_specifier wp,
574 move_pred_fn pred, void *arg)
576 struct moving_context ctxt;
580 bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
582 for (id = start_btree_id;
583 id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
585 stats->btree_id = id;
587 if (id != BTREE_ID_extents &&
588 id != BTREE_ID_reflink)
591 ret = __bch2_move_data(&ctxt,
592 id == start_btree_id ? start_pos : POS_MIN,
593 id == end_btree_id ? end_pos : POS_MAX,
599 bch2_moving_ctxt_exit(&ctxt);
604 void bch2_verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
606 struct bch_fs *c = trans->c;
607 struct btree_iter iter;
609 struct printbuf buf = PRINTBUF;
610 struct bch_backpointer bp;
614 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
615 bucket, BTREE_ITER_CACHED);
617 ret = lockrestart_do(trans,
618 bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
620 if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
621 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
623 if (a.v->gen == gen &&
624 a.v->dirty_sectors) {
625 if (a.v->data_type == BCH_DATA_btree) {
626 bch2_trans_unlock(trans);
627 if (bch2_btree_interior_updates_flush(c))
629 goto failed_to_evacuate;
634 bch2_trans_iter_exit(trans, &iter);
637 bch2_trans_iter_exit(trans, &iter);
639 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
642 prt_printf(&buf, bch2_log_msg(c, "failed to evacuate bucket "));
643 bch2_bkey_val_to_text(&buf, c, k);
646 bch2_trans_begin(trans);
648 ret = bch2_get_next_backpointer(trans, bucket, gen,
651 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
655 if (bp_offset == U64_MAX)
658 k = bch2_backpointer_get_key(trans, &iter,
659 bucket, bp_offset, bp);
661 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
668 bch2_bkey_val_to_text(&buf, c, k);
669 bch2_trans_iter_exit(trans, &iter);
672 bch2_print_string_as_lines(KERN_ERR, buf.buf);
676 int __bch2_evacuate_bucket(struct btree_trans *trans,
677 struct moving_context *ctxt,
678 struct move_bucket_in_flight *bucket_in_flight,
679 struct bpos bucket, int gen,
680 struct data_update_opts _data_opts)
682 struct bch_fs *c = ctxt->c;
683 struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
684 struct btree_iter iter;
686 struct bch_backpointer bp;
687 struct bch_alloc_v4 a_convert;
688 const struct bch_alloc_v4 *a;
690 struct data_update_opts data_opts;
691 unsigned dirty_sectors, bucket_size;
693 u64 bp_offset = 0, cur_inum = U64_MAX;
696 bch2_bkey_buf_init(&sk);
698 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
699 bucket, BTREE_ITER_CACHED);
700 ret = lockrestart_do(trans,
701 bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
702 bch2_trans_iter_exit(trans, &iter);
705 bch_err(c, "%s: error looking up alloc key: %s", __func__, bch2_err_str(ret));
709 a = bch2_alloc_to_v4(k, &a_convert);
710 dirty_sectors = a->dirty_sectors;
711 bucket_size = bch_dev_bkey_exists(c, bucket.inode)->mi.bucket_size;
712 fragmentation = a->fragmentation_lru;
714 ret = bch2_btree_write_buffer_flush(trans);
716 bch_err(c, "%s: error flushing btree write buffer: %s", __func__, bch2_err_str(ret));
720 while (!(ret = move_ratelimit(trans, ctxt))) {
721 bch2_trans_begin(trans);
723 ret = bch2_get_next_backpointer(trans, bucket, gen,
726 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
730 if (bp_offset == U64_MAX)
734 const struct bch_extent_ptr *ptr;
738 k = bch2_backpointer_get_key(trans, &iter,
739 bucket, bp_offset, bp);
741 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
748 bch2_bkey_buf_reassemble(&sk, c, k);
749 k = bkey_i_to_s_c(sk.k);
751 ret = move_get_io_opts(trans, &io_opts, k, &cur_inum);
753 bch2_trans_iter_exit(trans, &iter);
757 data_opts = _data_opts;
758 data_opts.target = io_opts.background_target;
759 data_opts.rewrite_ptrs = 0;
761 bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
762 if (ptr->dev == bucket.inode)
763 data_opts.rewrite_ptrs |= 1U << i;
767 ret = bch2_move_extent(trans, &iter, ctxt,
769 io_opts, bp.btree_id, k, data_opts);
770 bch2_trans_iter_exit(trans, &iter);
772 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
774 if (ret == -ENOMEM) {
775 /* memory allocation failure, wait for some IO to finish */
776 bch2_move_ctxt_wait_for_io(ctxt, trans);
783 bch2_ratelimit_increment(ctxt->rate, k.k->size);
785 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
789 b = bch2_backpointer_get_node(trans, &iter,
790 bucket, bp_offset, bp);
791 ret = PTR_ERR_OR_ZERO(b);
792 if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
794 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
801 ret = bch2_btree_node_rewrite(trans, &iter, b, 0);
802 bch2_trans_iter_exit(trans, &iter);
804 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
810 bch2_ratelimit_increment(ctxt->rate,
811 c->opts.btree_node_size >> 9);
813 atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_seen);
814 atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_moved);
821 trace_evacuate_bucket(c, &bucket, dirty_sectors, bucket_size, fragmentation, ret);
823 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
824 bch2_trans_unlock(trans);
825 move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
826 closure_sync(&ctxt->cl);
827 if (!ctxt->write_error)
828 bch2_verify_bucket_evacuated(trans, bucket, gen);
831 bch2_bkey_buf_exit(&sk, c);
835 int bch2_evacuate_bucket(struct bch_fs *c,
836 struct bpos bucket, int gen,
837 struct data_update_opts data_opts,
838 struct bch_ratelimit *rate,
839 struct bch_move_stats *stats,
840 struct write_point_specifier wp,
843 struct btree_trans trans;
844 struct moving_context ctxt;
847 bch2_trans_init(&trans, c, 0, 0);
848 bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
849 ret = __bch2_evacuate_bucket(&trans, &ctxt, NULL, bucket, gen, data_opts);
850 bch2_moving_ctxt_exit(&ctxt);
851 bch2_trans_exit(&trans);
856 typedef bool (*move_btree_pred)(struct bch_fs *, void *,
857 struct btree *, struct bch_io_opts *,
858 struct data_update_opts *);
860 static int bch2_move_btree(struct bch_fs *c,
861 enum btree_id start_btree_id, struct bpos start_pos,
862 enum btree_id end_btree_id, struct bpos end_pos,
863 move_btree_pred pred, void *arg,
864 struct bch_move_stats *stats)
866 bool kthread = (current->flags & PF_KTHREAD) != 0;
867 struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
868 struct btree_trans trans;
869 struct btree_iter iter;
872 struct data_update_opts data_opts;
875 bch2_trans_init(&trans, c, 0, 0);
876 progress_list_add(c, stats);
878 stats->data_type = BCH_DATA_btree;
880 for (id = start_btree_id;
881 id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
883 stats->btree_id = id;
885 bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
886 BTREE_ITER_PREFETCH);
889 while (bch2_trans_begin(&trans),
890 (b = bch2_btree_iter_peek_node(&iter)) &&
891 !(ret = PTR_ERR_OR_ZERO(b))) {
892 if (kthread && kthread_should_stop())
895 if ((cmp_int(id, end_btree_id) ?:
896 bpos_cmp(b->key.k.p, end_pos)) > 0)
899 stats->pos = iter.pos;
901 if (!pred(c, arg, b, &io_opts, &data_opts))
904 ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
905 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
910 bch2_btree_iter_next_node(&iter);
912 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
915 bch2_trans_iter_exit(&trans, &iter);
917 if (kthread && kthread_should_stop())
921 bch2_trans_exit(&trans);
924 bch_err(c, "error in %s(): %s", __func__, bch2_err_str(ret));
926 bch2_btree_interior_updates_flush(c);
928 progress_list_del(c, stats);
932 static bool rereplicate_pred(struct bch_fs *c, void *arg,
934 struct bch_io_opts *io_opts,
935 struct data_update_opts *data_opts)
937 unsigned nr_good = bch2_bkey_durability(c, k);
938 unsigned replicas = bkey_is_btree_ptr(k.k)
939 ? c->opts.metadata_replicas
940 : io_opts->data_replicas;
942 if (!nr_good || nr_good >= replicas)
945 data_opts->target = 0;
946 data_opts->extra_replicas = replicas - nr_good;
947 data_opts->btree_insert_flags = 0;
951 static bool migrate_pred(struct bch_fs *c, void *arg,
953 struct bch_io_opts *io_opts,
954 struct data_update_opts *data_opts)
956 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
957 const struct bch_extent_ptr *ptr;
958 struct bch_ioctl_data *op = arg;
961 data_opts->rewrite_ptrs = 0;
962 data_opts->target = 0;
963 data_opts->extra_replicas = 0;
964 data_opts->btree_insert_flags = 0;
966 bkey_for_each_ptr(ptrs, ptr) {
967 if (ptr->dev == op->migrate.dev)
968 data_opts->rewrite_ptrs |= 1U << i;
972 return data_opts->rewrite_ptrs != 0;
975 static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
977 struct bch_io_opts *io_opts,
978 struct data_update_opts *data_opts)
980 return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
983 static bool migrate_btree_pred(struct bch_fs *c, void *arg,
985 struct bch_io_opts *io_opts,
986 struct data_update_opts *data_opts)
988 return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
991 static bool bformat_needs_redo(struct bkey_format *f)
995 for (i = 0; i < f->nr_fields; i++) {
996 unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
997 u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
998 u64 field_offset = le64_to_cpu(f->field_offset[i]);
1000 if (f->bits_per_field[i] > unpacked_bits)
1003 if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
1006 if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
1015 static bool rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
1017 struct bch_io_opts *io_opts,
1018 struct data_update_opts *data_opts)
1020 if (b->version_ondisk != c->sb.version ||
1021 btree_node_need_rewrite(b) ||
1022 bformat_needs_redo(&b->format)) {
1023 data_opts->target = 0;
1024 data_opts->extra_replicas = 0;
1025 data_opts->btree_insert_flags = 0;
1032 int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
1036 ret = bch2_move_btree(c,
1038 BTREE_ID_NR, SPOS_MAX,
1039 rewrite_old_nodes_pred, c, stats);
1041 mutex_lock(&c->sb_lock);
1042 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1043 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1044 c->disk_sb.sb->version_min = c->disk_sb.sb->version;
1045 bch2_write_super(c);
1046 mutex_unlock(&c->sb_lock);
1052 int bch2_data_job(struct bch_fs *c,
1053 struct bch_move_stats *stats,
1054 struct bch_ioctl_data op)
1059 case BCH_DATA_OP_REREPLICATE:
1060 bch2_move_stats_init(stats, "rereplicate");
1061 stats->data_type = BCH_DATA_journal;
1062 ret = bch2_journal_flush_device_pins(&c->journal, -1);
1064 ret = bch2_move_btree(c,
1065 op.start_btree, op.start_pos,
1066 op.end_btree, op.end_pos,
1067 rereplicate_btree_pred, c, stats) ?: ret;
1068 ret = bch2_replicas_gc2(c) ?: ret;
1070 ret = bch2_move_data(c,
1071 op.start_btree, op.start_pos,
1072 op.end_btree, op.end_pos,
1075 writepoint_hashed((unsigned long) current),
1077 rereplicate_pred, c) ?: ret;
1078 ret = bch2_replicas_gc2(c) ?: ret;
1080 case BCH_DATA_OP_MIGRATE:
1081 if (op.migrate.dev >= c->sb.nr_devices)
1084 bch2_move_stats_init(stats, "migrate");
1085 stats->data_type = BCH_DATA_journal;
1086 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
1088 ret = bch2_move_btree(c,
1089 op.start_btree, op.start_pos,
1090 op.end_btree, op.end_pos,
1091 migrate_btree_pred, &op, stats) ?: ret;
1092 ret = bch2_replicas_gc2(c) ?: ret;
1094 ret = bch2_move_data(c,
1095 op.start_btree, op.start_pos,
1096 op.end_btree, op.end_pos,
1099 writepoint_hashed((unsigned long) current),
1101 migrate_pred, &op) ?: ret;
1102 ret = bch2_replicas_gc2(c) ?: ret;
1104 case BCH_DATA_OP_REWRITE_OLD_NODES:
1105 bch2_move_stats_init(stats, "rewrite_old_nodes");
1106 ret = bch2_scan_old_btree_nodes(c, stats);