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 io->write.ctxt = ctxt;
308 io->write.op.end_io = move_write_done;
310 atomic64_inc(&ctxt->stats->keys_moved);
311 atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
313 if (ret == -BCH_ERR_unwritten_extent_update) {
314 bch2_update_unwritten_extent(trans, &io->write);
321 this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
322 this_cpu_add(c->counters[BCH_COUNTER_move_extent_read], k.k->size);
323 trace_move_extent_read(k.k);
325 atomic_add(io->read_sectors, &ctxt->read_sectors);
326 atomic_inc(&ctxt->read_ios);
327 list_add_tail(&io->list, &ctxt->reads);
330 * dropped by move_read_endio() - guards against use after free of
331 * ctxt when doing wakeup
333 closure_get(&ctxt->cl);
334 bch2_read_extent(trans, &io->rbio,
338 BCH_READ_LAST_FRAGMENT);
341 bio_free_pages(&io->write.op.wbio.bio);
345 bch2_write_ref_put(c, BCH_WRITE_REF_move);
346 trace_and_count(c, move_extent_alloc_mem_fail, k.k);
350 static int lookup_inode(struct btree_trans *trans, struct bpos pos,
351 struct bch_inode_unpacked *inode)
353 struct btree_iter iter;
357 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
358 BTREE_ITER_ALL_SNAPSHOTS);
359 k = bch2_btree_iter_peek(&iter);
364 if (!k.k || !bkey_eq(k.k->p, pos)) {
369 ret = bkey_is_inode(k.k) ? 0 : -EIO;
373 ret = bch2_inode_unpack(k, inode);
377 bch2_trans_iter_exit(trans, &iter);
381 static int move_ratelimit(struct btree_trans *trans,
382 struct moving_context *ctxt)
384 struct bch_fs *c = trans->c;
387 if (ctxt->wait_on_copygc) {
388 bch2_trans_unlock(trans);
389 wait_event_killable(c->copygc_running_wq,
390 !c->copygc_running ||
391 kthread_should_stop());
395 delay = ctxt->rate ? bch2_ratelimit_delay(ctxt->rate) : 0;
398 bch2_trans_unlock(trans);
399 set_current_state(TASK_INTERRUPTIBLE);
402 if ((current->flags & PF_KTHREAD) && kthread_should_stop()) {
403 __set_current_state(TASK_RUNNING);
408 schedule_timeout(delay);
410 if (unlikely(freezing(current))) {
411 move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
417 * XXX: these limits really ought to be per device, SSDs and hard drives
418 * will want different limits
420 move_ctxt_wait_event(ctxt, trans,
421 atomic_read(&ctxt->write_sectors) < c->opts.move_bytes_in_flight >> 9 &&
422 atomic_read(&ctxt->read_sectors) < c->opts.move_bytes_in_flight >> 9 &&
423 atomic_read(&ctxt->write_ios) < c->opts.move_ios_in_flight &&
424 atomic_read(&ctxt->read_ios) < c->opts.move_ios_in_flight);
429 static int move_get_io_opts(struct btree_trans *trans,
430 struct bch_io_opts *io_opts,
431 struct bkey_s_c k, u64 *cur_inum)
433 struct bch_inode_unpacked inode;
436 if (*cur_inum == k.k->p.inode)
439 ret = lookup_inode(trans,
440 SPOS(0, k.k->p.inode, k.k->p.snapshot),
442 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
446 bch2_inode_opts_get(io_opts, trans->c, &inode);
448 *io_opts = bch2_opts_to_inode_opts(trans->c->opts);
449 *cur_inum = k.k->p.inode;
453 static int __bch2_move_data(struct moving_context *ctxt,
456 move_pred_fn pred, void *arg,
457 enum btree_id btree_id)
459 struct bch_fs *c = ctxt->c;
460 struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
462 struct btree_trans trans;
463 struct btree_iter iter;
465 struct data_update_opts data_opts;
466 u64 cur_inum = U64_MAX;
469 bch2_bkey_buf_init(&sk);
470 bch2_trans_init(&trans, c, 0, 0);
472 ctxt->stats->data_type = BCH_DATA_user;
473 ctxt->stats->btree_id = btree_id;
474 ctxt->stats->pos = start;
476 bch2_trans_iter_init(&trans, &iter, btree_id, start,
478 BTREE_ITER_ALL_SNAPSHOTS);
481 bch2_ratelimit_reset(ctxt->rate);
483 while (!move_ratelimit(&trans, ctxt)) {
484 bch2_trans_begin(&trans);
486 k = bch2_btree_iter_peek(&iter);
491 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
496 if (bkey_ge(bkey_start_pos(k.k), end))
499 ctxt->stats->pos = iter.pos;
501 if (!bkey_extent_is_direct_data(k.k))
504 ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
508 memset(&data_opts, 0, sizeof(data_opts));
509 if (!pred(c, arg, k, &io_opts, &data_opts))
513 * The iterator gets unlocked by __bch2_read_extent - need to
514 * save a copy of @k elsewhere:
516 bch2_bkey_buf_reassemble(&sk, c, k);
517 k = bkey_i_to_s_c(sk.k);
518 bch2_trans_unlock(&trans);
520 ret2 = bch2_move_extent(&trans, &iter, ctxt, io_opts,
521 btree_id, k, data_opts);
523 if (bch2_err_matches(ret2, BCH_ERR_transaction_restart))
526 if (ret2 == -ENOMEM) {
527 /* memory allocation failure, wait for some IO to finish */
528 bch2_move_ctxt_wait_for_io(ctxt, &trans);
532 /* XXX signal failure */
537 bch2_ratelimit_increment(ctxt->rate, k.k->size);
539 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
541 bch2_btree_iter_advance(&iter);
544 bch2_trans_iter_exit(&trans, &iter);
545 bch2_trans_exit(&trans);
546 bch2_bkey_buf_exit(&sk, c);
551 int bch2_move_data(struct bch_fs *c,
552 enum btree_id start_btree_id, struct bpos start_pos,
553 enum btree_id end_btree_id, struct bpos end_pos,
554 struct bch_ratelimit *rate,
555 struct bch_move_stats *stats,
556 struct write_point_specifier wp,
558 move_pred_fn pred, void *arg)
560 struct moving_context ctxt;
564 bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
566 for (id = start_btree_id;
567 id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
569 stats->btree_id = id;
571 if (id != BTREE_ID_extents &&
572 id != BTREE_ID_reflink)
575 ret = __bch2_move_data(&ctxt,
576 id == start_btree_id ? start_pos : POS_MIN,
577 id == end_btree_id ? end_pos : POS_MAX,
583 bch2_moving_ctxt_exit(&ctxt);
588 static noinline void verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
590 struct bch_fs *c = trans->c;
591 struct btree_iter iter;
593 struct printbuf buf = PRINTBUF;
594 struct bch_backpointer bp;
598 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
599 bucket, BTREE_ITER_CACHED);
601 ret = lockrestart_do(trans,
602 bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
604 if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
605 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
607 if (a.v->gen == gen &&
608 a.v->dirty_sectors) {
609 if (a.v->data_type == BCH_DATA_btree) {
610 bch2_trans_unlock(trans);
611 if (bch2_btree_interior_updates_flush(c))
613 goto failed_to_evacuate;
618 bch2_trans_iter_exit(trans, &iter);
621 bch2_trans_iter_exit(trans, &iter);
623 prt_printf(&buf, bch2_log_msg(c, "failed to evacuate bucket "));
624 bch2_bkey_val_to_text(&buf, c, k);
627 bch2_trans_begin(trans);
629 ret = bch2_get_next_backpointer(trans, bucket, gen,
632 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
636 if (bp_offset == U64_MAX)
639 k = bch2_backpointer_get_key(trans, &iter,
640 bucket, bp_offset, bp);
642 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
649 bch2_bkey_val_to_text(&buf, c, k);
650 bch2_trans_iter_exit(trans, &iter);
653 bch2_print_string_as_lines(KERN_ERR, buf.buf);
657 int __bch2_evacuate_bucket(struct btree_trans *trans,
658 struct moving_context *ctxt,
659 struct bpos bucket, int gen,
660 struct data_update_opts _data_opts)
662 struct bch_fs *c = ctxt->c;
663 struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
664 struct btree_iter iter;
666 struct bch_backpointer bp;
667 struct bch_alloc_v4 a_convert;
668 const struct bch_alloc_v4 *a;
670 struct data_update_opts data_opts;
671 unsigned dirty_sectors, bucket_size;
673 u64 bp_offset = 0, cur_inum = U64_MAX;
676 bch2_bkey_buf_init(&sk);
678 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
679 bucket, BTREE_ITER_CACHED);
680 ret = lockrestart_do(trans,
681 bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
682 bch2_trans_iter_exit(trans, &iter);
685 bch_err(c, "%s: error looking up alloc key: %s", __func__, bch2_err_str(ret));
689 a = bch2_alloc_to_v4(k, &a_convert);
690 dirty_sectors = a->dirty_sectors;
691 bucket_size = bch_dev_bkey_exists(c, bucket.inode)->mi.bucket_size;
692 fragmentation = a->fragmentation_lru;
694 ret = bch2_btree_write_buffer_flush(trans);
696 bch_err(c, "%s: error flushing btree write buffer: %s", __func__, bch2_err_str(ret));
700 while (!(ret = move_ratelimit(trans, ctxt))) {
701 bch2_trans_begin(trans);
703 ret = bch2_get_next_backpointer(trans, bucket, gen,
706 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
710 if (bp_offset == U64_MAX)
714 const struct bch_extent_ptr *ptr;
718 k = bch2_backpointer_get_key(trans, &iter,
719 bucket, bp_offset, bp);
721 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
728 bch2_bkey_buf_reassemble(&sk, c, k);
729 k = bkey_i_to_s_c(sk.k);
731 ret = move_get_io_opts(trans, &io_opts, k, &cur_inum);
733 bch2_trans_iter_exit(trans, &iter);
737 data_opts = _data_opts;
738 data_opts.target = io_opts.background_target;
739 data_opts.rewrite_ptrs = 0;
741 bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
742 if (ptr->dev == bucket.inode)
743 data_opts.rewrite_ptrs |= 1U << i;
747 ret = bch2_move_extent(trans, &iter, ctxt, io_opts,
748 bp.btree_id, k, data_opts);
749 bch2_trans_iter_exit(trans, &iter);
751 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
753 if (ret == -ENOMEM) {
754 /* memory allocation failure, wait for some IO to finish */
755 bch2_move_ctxt_wait_for_io(ctxt, trans);
762 bch2_ratelimit_increment(ctxt->rate, k.k->size);
763 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
767 b = bch2_backpointer_get_node(trans, &iter,
768 bucket, bp_offset, bp);
769 ret = PTR_ERR_OR_ZERO(b);
770 if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
772 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
779 ret = bch2_btree_node_rewrite(trans, &iter, b, 0);
780 bch2_trans_iter_exit(trans, &iter);
782 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
788 bch2_ratelimit_increment(ctxt->rate,
789 c->opts.btree_node_size >> 9);
790 atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_seen);
791 atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_moved);
797 trace_evacuate_bucket(c, &bucket, dirty_sectors, bucket_size, fragmentation, ret);
799 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
800 bch2_trans_unlock(trans);
801 move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
802 closure_sync(&ctxt->cl);
803 if (!ctxt->write_error)
804 verify_bucket_evacuated(trans, bucket, gen);
807 bch2_bkey_buf_exit(&sk, c);
811 int bch2_evacuate_bucket(struct bch_fs *c,
812 struct bpos bucket, int gen,
813 struct data_update_opts data_opts,
814 struct bch_ratelimit *rate,
815 struct bch_move_stats *stats,
816 struct write_point_specifier wp,
819 struct btree_trans trans;
820 struct moving_context ctxt;
823 bch2_trans_init(&trans, c, 0, 0);
824 bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
825 ret = __bch2_evacuate_bucket(&trans, &ctxt, bucket, gen, data_opts);
826 bch2_moving_ctxt_exit(&ctxt);
827 bch2_trans_exit(&trans);
832 typedef bool (*move_btree_pred)(struct bch_fs *, void *,
833 struct btree *, struct bch_io_opts *,
834 struct data_update_opts *);
836 static int bch2_move_btree(struct bch_fs *c,
837 enum btree_id start_btree_id, struct bpos start_pos,
838 enum btree_id end_btree_id, struct bpos end_pos,
839 move_btree_pred pred, void *arg,
840 struct bch_move_stats *stats)
842 bool kthread = (current->flags & PF_KTHREAD) != 0;
843 struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
844 struct btree_trans trans;
845 struct btree_iter iter;
848 struct data_update_opts data_opts;
851 bch2_trans_init(&trans, c, 0, 0);
852 progress_list_add(c, stats);
854 stats->data_type = BCH_DATA_btree;
856 for (id = start_btree_id;
857 id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
859 stats->btree_id = id;
861 bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
862 BTREE_ITER_PREFETCH);
865 while (bch2_trans_begin(&trans),
866 (b = bch2_btree_iter_peek_node(&iter)) &&
867 !(ret = PTR_ERR_OR_ZERO(b))) {
868 if (kthread && kthread_should_stop())
871 if ((cmp_int(id, end_btree_id) ?:
872 bpos_cmp(b->key.k.p, end_pos)) > 0)
875 stats->pos = iter.pos;
877 if (!pred(c, arg, b, &io_opts, &data_opts))
880 ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
881 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
886 bch2_btree_iter_next_node(&iter);
888 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
891 bch2_trans_iter_exit(&trans, &iter);
893 if (kthread && kthread_should_stop())
897 bch2_trans_exit(&trans);
900 bch_err(c, "error in %s(): %s", __func__, bch2_err_str(ret));
902 bch2_btree_interior_updates_flush(c);
904 progress_list_del(c, stats);
908 static bool rereplicate_pred(struct bch_fs *c, void *arg,
910 struct bch_io_opts *io_opts,
911 struct data_update_opts *data_opts)
913 unsigned nr_good = bch2_bkey_durability(c, k);
914 unsigned replicas = bkey_is_btree_ptr(k.k)
915 ? c->opts.metadata_replicas
916 : io_opts->data_replicas;
918 if (!nr_good || nr_good >= replicas)
921 data_opts->target = 0;
922 data_opts->extra_replicas = replicas - nr_good;
923 data_opts->btree_insert_flags = 0;
927 static bool migrate_pred(struct bch_fs *c, void *arg,
929 struct bch_io_opts *io_opts,
930 struct data_update_opts *data_opts)
932 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
933 const struct bch_extent_ptr *ptr;
934 struct bch_ioctl_data *op = arg;
937 data_opts->rewrite_ptrs = 0;
938 data_opts->target = 0;
939 data_opts->extra_replicas = 0;
940 data_opts->btree_insert_flags = 0;
942 bkey_for_each_ptr(ptrs, ptr) {
943 if (ptr->dev == op->migrate.dev)
944 data_opts->rewrite_ptrs |= 1U << i;
948 return data_opts->rewrite_ptrs != 0;
951 static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
953 struct bch_io_opts *io_opts,
954 struct data_update_opts *data_opts)
956 return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
959 static bool migrate_btree_pred(struct bch_fs *c, void *arg,
961 struct bch_io_opts *io_opts,
962 struct data_update_opts *data_opts)
964 return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
967 static bool bformat_needs_redo(struct bkey_format *f)
971 for (i = 0; i < f->nr_fields; i++) {
972 unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
973 u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
974 u64 field_offset = le64_to_cpu(f->field_offset[i]);
976 if (f->bits_per_field[i] > unpacked_bits)
979 if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
982 if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
991 static bool rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
993 struct bch_io_opts *io_opts,
994 struct data_update_opts *data_opts)
996 if (b->version_ondisk != c->sb.version ||
997 btree_node_need_rewrite(b) ||
998 bformat_needs_redo(&b->format)) {
999 data_opts->target = 0;
1000 data_opts->extra_replicas = 0;
1001 data_opts->btree_insert_flags = 0;
1008 int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
1012 ret = bch2_move_btree(c,
1014 BTREE_ID_NR, SPOS_MAX,
1015 rewrite_old_nodes_pred, c, stats);
1017 mutex_lock(&c->sb_lock);
1018 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1019 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1020 c->disk_sb.sb->version_min = c->disk_sb.sb->version;
1021 bch2_write_super(c);
1022 mutex_unlock(&c->sb_lock);
1028 int bch2_data_job(struct bch_fs *c,
1029 struct bch_move_stats *stats,
1030 struct bch_ioctl_data op)
1035 case BCH_DATA_OP_REREPLICATE:
1036 bch2_move_stats_init(stats, "rereplicate");
1037 stats->data_type = BCH_DATA_journal;
1038 ret = bch2_journal_flush_device_pins(&c->journal, -1);
1040 ret = bch2_move_btree(c,
1041 op.start_btree, op.start_pos,
1042 op.end_btree, op.end_pos,
1043 rereplicate_btree_pred, c, stats) ?: ret;
1044 ret = bch2_replicas_gc2(c) ?: ret;
1046 ret = bch2_move_data(c,
1047 op.start_btree, op.start_pos,
1048 op.end_btree, op.end_pos,
1051 writepoint_hashed((unsigned long) current),
1053 rereplicate_pred, c) ?: ret;
1054 ret = bch2_replicas_gc2(c) ?: ret;
1056 case BCH_DATA_OP_MIGRATE:
1057 if (op.migrate.dev >= c->sb.nr_devices)
1060 bch2_move_stats_init(stats, "migrate");
1061 stats->data_type = BCH_DATA_journal;
1062 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
1064 ret = bch2_move_btree(c,
1065 op.start_btree, op.start_pos,
1066 op.end_btree, op.end_pos,
1067 migrate_btree_pred, &op, 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 migrate_pred, &op) ?: ret;
1078 ret = bch2_replicas_gc2(c) ?: ret;
1080 case BCH_DATA_OP_REWRITE_OLD_NODES:
1081 bch2_move_stats_init(stats, "rewrite_old_nodes");
1082 ret = bch2_scan_old_btree_nodes(c, stats);