]> git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/move.c
Merge pull request #24 from brendon-boldt/new-install-distros
[bcachefs-tools-debian] / libbcachefs / move.c
1 // SPDX-License-Identifier: GPL-2.0
2
3 #include "bcachefs.h"
4 #include "alloc_foreground.h"
5 #include "bkey_on_stack.h"
6 #include "btree_gc.h"
7 #include "btree_update.h"
8 #include "btree_update_interior.h"
9 #include "buckets.h"
10 #include "disk_groups.h"
11 #include "inode.h"
12 #include "io.h"
13 #include "journal_reclaim.h"
14 #include "move.h"
15 #include "replicas.h"
16 #include "super-io.h"
17 #include "keylist.h"
18
19 #include <linux/ioprio.h>
20 #include <linux/kthread.h>
21
22 #include <trace/events/bcachefs.h>
23
24 #define SECTORS_IN_FLIGHT_PER_DEVICE    2048
25
26 struct moving_io {
27         struct list_head        list;
28         struct closure          cl;
29         bool                    read_completed;
30
31         unsigned                read_sectors;
32         unsigned                write_sectors;
33
34         struct bch_read_bio     rbio;
35
36         struct migrate_write    write;
37         /* Must be last since it is variable size */
38         struct bio_vec          bi_inline_vecs[0];
39 };
40
41 struct moving_context {
42         /* Closure for waiting on all reads and writes to complete */
43         struct closure          cl;
44
45         struct bch_move_stats   *stats;
46
47         struct list_head        reads;
48
49         /* in flight sectors: */
50         atomic_t                read_sectors;
51         atomic_t                write_sectors;
52
53         wait_queue_head_t       wait;
54 };
55
56 static int bch2_migrate_index_update(struct bch_write_op *op)
57 {
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         int ret = 0;
65
66         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
67
68         iter = bch2_trans_get_iter(&trans, m->btree_id,
69                                    bkey_start_pos(&bch2_keylist_front(keys)->k),
70                                    BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
71
72         while (1) {
73                 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
74                 struct bkey_i *insert;
75                 struct bkey_i_extent *new =
76                         bkey_i_to_extent(bch2_keylist_front(keys));
77                 BKEY_PADDED(k) _new, _insert;
78                 const union bch_extent_entry *entry;
79                 struct extent_ptr_decoded p;
80                 bool did_work = false;
81                 int nr;
82
83                 ret = bkey_err(k);
84                 if (ret)
85                         break;
86
87                 if (bversion_cmp(k.k->version, new->k.version) ||
88                     !bch2_bkey_matches_ptr(c, k, m->ptr, m->offset))
89                         goto nomatch;
90
91                 if (m->data_cmd == DATA_REWRITE &&
92                     !bch2_bkey_has_device(k, m->data_opts.rewrite_dev))
93                         goto nomatch;
94
95                 bkey_reassemble(&_insert.k, k);
96                 insert = &_insert.k;
97
98                 bkey_copy(&_new.k, bch2_keylist_front(keys));
99                 new = bkey_i_to_extent(&_new.k);
100                 bch2_cut_front(iter->pos, &new->k_i);
101
102                 bch2_cut_front(iter->pos,       insert);
103                 bch2_cut_back(new->k.p,         insert);
104                 bch2_cut_back(insert->k.p,      &new->k_i);
105
106                 if (m->data_cmd == DATA_REWRITE)
107                         bch2_bkey_drop_device(bkey_i_to_s(insert),
108                                               m->data_opts.rewrite_dev);
109
110                 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
111                         if (bch2_bkey_has_device(bkey_i_to_s_c(insert), p.ptr.dev)) {
112                                 /*
113                                  * raced with another move op? extent already
114                                  * has a pointer to the device we just wrote
115                                  * data to
116                                  */
117                                 continue;
118                         }
119
120                         bch2_extent_ptr_decoded_append(insert, &p);
121                         did_work = true;
122                 }
123
124                 if (!did_work)
125                         goto nomatch;
126
127                 bch2_bkey_narrow_crcs(insert,
128                                 (struct bch_extent_crc_unpacked) { 0 });
129                 bch2_extent_normalize(c, bkey_i_to_s(insert));
130                 bch2_bkey_mark_replicas_cached(c, bkey_i_to_s(insert),
131                                                op->opts.background_target,
132                                                op->opts.data_replicas);
133
134                 /*
135                  * If we're not fully overwriting @k, and it's compressed, we
136                  * need a reservation for all the pointers in @insert
137                  */
138                 nr = bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(insert)) -
139                          m->nr_ptrs_reserved;
140
141                 if (insert->k.size < k.k->size &&
142                     bch2_bkey_sectors_compressed(k) &&
143                     nr > 0) {
144                         ret = bch2_disk_reservation_add(c, &op->res,
145                                         keylist_sectors(keys) * nr, 0);
146                         if (ret)
147                                 goto out;
148
149                         m->nr_ptrs_reserved += nr;
150                         goto next;
151                 }
152
153                 bch2_trans_update(&trans, iter, insert);
154
155                 ret = bch2_trans_commit(&trans, &op->res,
156                                 op_journal_seq(op),
157                                 BTREE_INSERT_ATOMIC|
158                                 BTREE_INSERT_NOFAIL|
159                                 BTREE_INSERT_USE_RESERVE|
160                                 m->data_opts.btree_insert_flags);
161                 if (!ret)
162                         atomic_long_inc(&c->extent_migrate_done);
163                 if (ret == -EINTR)
164                         ret = 0;
165                 if (ret)
166                         break;
167 next:
168                 while (bkey_cmp(iter->pos, bch2_keylist_front(keys)->k.p) >= 0) {
169                         bch2_keylist_pop_front(keys);
170                         if (bch2_keylist_empty(keys))
171                                 goto out;
172                 }
173                 continue;
174 nomatch:
175                 if (m->ctxt)
176                         atomic64_add(k.k->p.offset - iter->pos.offset,
177                                      &m->ctxt->stats->sectors_raced);
178                 atomic_long_inc(&c->extent_migrate_raced);
179                 trace_move_race(&new->k);
180                 bch2_btree_iter_next_slot(iter);
181                 goto next;
182         }
183 out:
184         bch2_trans_exit(&trans);
185         BUG_ON(ret == -EINTR);
186         return ret;
187 }
188
189 void bch2_migrate_read_done(struct migrate_write *m, struct bch_read_bio *rbio)
190 {
191         /* write bio must own pages: */
192         BUG_ON(!m->op.wbio.bio.bi_vcnt);
193
194         m->ptr          = rbio->pick.ptr;
195         m->offset       = rbio->pos.offset - rbio->pick.crc.offset;
196         m->op.devs_have = rbio->devs_have;
197         m->op.pos       = rbio->pos;
198         m->op.version   = rbio->version;
199         m->op.crc       = rbio->pick.crc;
200         m->op.wbio.bio.bi_iter.bi_size = m->op.crc.compressed_size << 9;
201
202         if (bch2_csum_type_is_encryption(m->op.crc.csum_type)) {
203                 m->op.nonce     = m->op.crc.nonce + m->op.crc.offset;
204                 m->op.csum_type = m->op.crc.csum_type;
205         }
206
207         if (m->data_cmd == DATA_REWRITE)
208                 bch2_dev_list_drop_dev(&m->op.devs_have, m->data_opts.rewrite_dev);
209 }
210
211 int bch2_migrate_write_init(struct bch_fs *c, struct migrate_write *m,
212                             struct write_point_specifier wp,
213                             struct bch_io_opts io_opts,
214                             enum data_cmd data_cmd,
215                             struct data_opts data_opts,
216                             enum btree_id btree_id,
217                             struct bkey_s_c k)
218 {
219         int ret;
220
221         m->btree_id     = btree_id;
222         m->data_cmd     = data_cmd;
223         m->data_opts    = data_opts;
224         m->nr_ptrs_reserved = 0;
225
226         bch2_write_op_init(&m->op, c, io_opts);
227         m->op.compression_type =
228                 bch2_compression_opt_to_type[io_opts.background_compression ?:
229                                              io_opts.compression];
230         m->op.target    = data_opts.target,
231         m->op.write_point = wp;
232
233         if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE)
234                 m->op.alloc_reserve = RESERVE_MOVINGGC;
235
236         m->op.flags |= BCH_WRITE_ONLY_SPECIFIED_DEVS|
237                 BCH_WRITE_PAGES_STABLE|
238                 BCH_WRITE_PAGES_OWNED|
239                 BCH_WRITE_DATA_ENCODED;
240
241         m->op.nr_replicas       = 1;
242         m->op.nr_replicas_required = 1;
243         m->op.index_update_fn   = bch2_migrate_index_update;
244
245         switch (data_cmd) {
246         case DATA_ADD_REPLICAS: {
247                 /*
248                  * DATA_ADD_REPLICAS is used for moving data to a different
249                  * device in the background, and due to compression the new copy
250                  * might take up more space than the old copy:
251                  */
252 #if 0
253                 int nr = (int) io_opts.data_replicas -
254                         bch2_bkey_nr_ptrs_allocated(k);
255 #endif
256                 int nr = (int) io_opts.data_replicas;
257
258                 if (nr > 0) {
259                         m->op.nr_replicas = m->nr_ptrs_reserved = nr;
260
261                         ret = bch2_disk_reservation_get(c, &m->op.res,
262                                         k.k->size, m->op.nr_replicas, 0);
263                         if (ret)
264                                 return ret;
265                 }
266                 break;
267         }
268         case DATA_REWRITE: {
269                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
270                 const union bch_extent_entry *entry;
271                 struct extent_ptr_decoded p;
272                 unsigned compressed_sectors = 0;
273
274                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
275                         if (!p.ptr.cached &&
276                             p.crc.compression_type != BCH_COMPRESSION_NONE &&
277                             bch2_dev_in_target(c, p.ptr.dev, data_opts.target))
278                                 compressed_sectors += p.crc.compressed_size;
279
280                 if (compressed_sectors) {
281                         ret = bch2_disk_reservation_add(c, &m->op.res,
282                                         compressed_sectors,
283                                         BCH_DISK_RESERVATION_NOFAIL);
284                         if (ret)
285                                 return ret;
286                 }
287                 break;
288         }
289         case DATA_PROMOTE:
290                 m->op.flags     |= BCH_WRITE_ALLOC_NOWAIT;
291                 m->op.flags     |= BCH_WRITE_CACHED;
292                 break;
293         default:
294                 BUG();
295         }
296
297         return 0;
298 }
299
300 static void move_free(struct closure *cl)
301 {
302         struct moving_io *io = container_of(cl, struct moving_io, cl);
303         struct moving_context *ctxt = io->write.ctxt;
304         struct bvec_iter_all iter;
305         struct bio_vec *bv;
306
307         bch2_disk_reservation_put(io->write.op.c, &io->write.op.res);
308
309         bio_for_each_segment_all(bv, &io->write.op.wbio.bio, iter)
310                 if (bv->bv_page)
311                         __free_page(bv->bv_page);
312
313         wake_up(&ctxt->wait);
314
315         kfree(io);
316 }
317
318 static void move_write_done(struct closure *cl)
319 {
320         struct moving_io *io = container_of(cl, struct moving_io, cl);
321
322         atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
323         closure_return_with_destructor(cl, move_free);
324 }
325
326 static void move_write(struct closure *cl)
327 {
328         struct moving_io *io = container_of(cl, struct moving_io, cl);
329
330         if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
331                 closure_return_with_destructor(cl, move_free);
332                 return;
333         }
334
335         bch2_migrate_read_done(&io->write, &io->rbio);
336
337         atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
338         closure_call(&io->write.op.cl, bch2_write, NULL, cl);
339         continue_at(cl, move_write_done, NULL);
340 }
341
342 static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
343 {
344         struct moving_io *io =
345                 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
346
347         return io && io->read_completed ? io : NULL;
348 }
349
350 static void move_read_endio(struct bio *bio)
351 {
352         struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
353         struct moving_context *ctxt = io->write.ctxt;
354
355         atomic_sub(io->read_sectors, &ctxt->read_sectors);
356         io->read_completed = true;
357
358         if (next_pending_write(ctxt))
359                 wake_up(&ctxt->wait);
360
361         closure_put(&ctxt->cl);
362 }
363
364 static void do_pending_writes(struct moving_context *ctxt)
365 {
366         struct moving_io *io;
367
368         while ((io = next_pending_write(ctxt))) {
369                 list_del(&io->list);
370                 closure_call(&io->cl, move_write, NULL, &ctxt->cl);
371         }
372 }
373
374 #define move_ctxt_wait_event(_ctxt, _cond)                      \
375 do {                                                            \
376         do_pending_writes(_ctxt);                               \
377                                                                 \
378         if (_cond)                                              \
379                 break;                                          \
380         __wait_event((_ctxt)->wait,                             \
381                      next_pending_write(_ctxt) || (_cond));     \
382 } while (1)
383
384 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt)
385 {
386         unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
387
388         move_ctxt_wait_event(ctxt,
389                 !atomic_read(&ctxt->write_sectors) ||
390                 atomic_read(&ctxt->write_sectors) != sectors_pending);
391 }
392
393 static int bch2_move_extent(struct bch_fs *c,
394                             struct moving_context *ctxt,
395                             struct write_point_specifier wp,
396                             struct bch_io_opts io_opts,
397                             enum btree_id btree_id,
398                             struct bkey_s_c k,
399                             enum data_cmd data_cmd,
400                             struct data_opts data_opts)
401 {
402         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
403         struct moving_io *io;
404         const union bch_extent_entry *entry;
405         struct extent_ptr_decoded p;
406         unsigned sectors = k.k->size, pages;
407         int ret = -ENOMEM;
408
409         move_ctxt_wait_event(ctxt,
410                 atomic_read(&ctxt->write_sectors) <
411                 SECTORS_IN_FLIGHT_PER_DEVICE);
412
413         move_ctxt_wait_event(ctxt,
414                 atomic_read(&ctxt->read_sectors) <
415                 SECTORS_IN_FLIGHT_PER_DEVICE);
416
417         /* write path might have to decompress data: */
418         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
419                 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
420
421         pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
422         io = kzalloc(sizeof(struct moving_io) +
423                      sizeof(struct bio_vec) * pages, GFP_KERNEL);
424         if (!io)
425                 goto err;
426
427         io->write.ctxt          = ctxt;
428         io->read_sectors        = k.k->size;
429         io->write_sectors       = k.k->size;
430
431         bio_init(&io->write.op.wbio.bio, io->bi_inline_vecs, pages);
432         bio_set_prio(&io->write.op.wbio.bio,
433                      IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
434
435         if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
436                                  GFP_KERNEL))
437                 goto err_free;
438
439         io->rbio.c              = c;
440         io->rbio.opts           = io_opts;
441         bio_init(&io->rbio.bio, io->bi_inline_vecs, pages);
442         io->rbio.bio.bi_vcnt = pages;
443         bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
444         io->rbio.bio.bi_iter.bi_size = sectors << 9;
445
446         bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
447         io->rbio.bio.bi_iter.bi_sector  = bkey_start_offset(k.k);
448         io->rbio.bio.bi_end_io          = move_read_endio;
449
450         ret = bch2_migrate_write_init(c, &io->write, wp, io_opts,
451                                       data_cmd, data_opts, btree_id, k);
452         if (ret)
453                 goto err_free_pages;
454
455         atomic64_inc(&ctxt->stats->keys_moved);
456         atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
457
458         trace_move_extent(k.k);
459
460         atomic_add(io->read_sectors, &ctxt->read_sectors);
461         list_add_tail(&io->list, &ctxt->reads);
462
463         /*
464          * dropped by move_read_endio() - guards against use after free of
465          * ctxt when doing wakeup
466          */
467         closure_get(&ctxt->cl);
468         bch2_read_extent(c, &io->rbio, k, 0,
469                          BCH_READ_NODECODE|
470                          BCH_READ_LAST_FRAGMENT);
471         return 0;
472 err_free_pages:
473         bio_free_pages(&io->write.op.wbio.bio);
474 err_free:
475         kfree(io);
476 err:
477         trace_move_alloc_fail(k.k);
478         return ret;
479 }
480
481 static int __bch2_move_data(struct bch_fs *c,
482                 struct moving_context *ctxt,
483                 struct bch_ratelimit *rate,
484                 struct write_point_specifier wp,
485                 struct bpos start,
486                 struct bpos end,
487                 move_pred_fn pred, void *arg,
488                 struct bch_move_stats *stats,
489                 enum btree_id btree_id)
490 {
491         bool kthread = (current->flags & PF_KTHREAD) != 0;
492         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
493         struct bkey_on_stack sk;
494         struct btree_trans trans;
495         struct btree_iter *iter;
496         struct bkey_s_c k;
497         struct data_opts data_opts;
498         enum data_cmd data_cmd;
499         u64 delay, cur_inum = U64_MAX;
500         int ret = 0, ret2;
501
502         bkey_on_stack_init(&sk);
503         bch2_trans_init(&trans, c, 0, 0);
504
505         stats->data_type = BCH_DATA_USER;
506         stats->btree_id = btree_id;
507         stats->pos      = POS_MIN;
508
509         iter = bch2_trans_get_iter(&trans, btree_id, start,
510                                    BTREE_ITER_PREFETCH);
511
512         if (rate)
513                 bch2_ratelimit_reset(rate);
514
515         while (1) {
516                 do {
517                         delay = rate ? bch2_ratelimit_delay(rate) : 0;
518
519                         if (delay) {
520                                 bch2_trans_unlock(&trans);
521                                 set_current_state(TASK_INTERRUPTIBLE);
522                         }
523
524                         if (kthread && (ret = kthread_should_stop())) {
525                                 __set_current_state(TASK_RUNNING);
526                                 goto out;
527                         }
528
529                         if (delay)
530                                 schedule_timeout(delay);
531
532                         if (unlikely(freezing(current))) {
533                                 bch2_trans_unlock(&trans);
534                                 move_ctxt_wait_event(ctxt, list_empty(&ctxt->reads));
535                                 try_to_freeze();
536                         }
537                 } while (delay);
538 peek:
539                 k = bch2_btree_iter_peek(iter);
540
541                 stats->pos = iter->pos;
542
543                 if (!k.k)
544                         break;
545                 ret = bkey_err(k);
546                 if (ret)
547                         break;
548                 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
549                         break;
550
551                 if (!bkey_extent_is_direct_data(k.k))
552                         goto next_nondata;
553
554                 if (btree_id == BTREE_ID_EXTENTS &&
555                     cur_inum != k.k->p.inode) {
556                         struct bch_inode_unpacked inode;
557
558                         /* don't hold btree locks while looking up inode: */
559                         bch2_trans_unlock(&trans);
560
561                         io_opts = bch2_opts_to_inode_opts(c->opts);
562                         if (!bch2_inode_find_by_inum(c, k.k->p.inode, &inode))
563                                 bch2_io_opts_apply(&io_opts, bch2_inode_opts_get(&inode));
564                         cur_inum = k.k->p.inode;
565                         goto peek;
566                 }
567
568                 switch ((data_cmd = pred(c, arg, k, &io_opts, &data_opts))) {
569                 case DATA_SKIP:
570                         goto next;
571                 case DATA_SCRUB:
572                         BUG();
573                 case DATA_ADD_REPLICAS:
574                 case DATA_REWRITE:
575                 case DATA_PROMOTE:
576                         break;
577                 default:
578                         BUG();
579                 }
580
581                 /* unlock before doing IO: */
582                 bkey_on_stack_reassemble(&sk, c, k);
583                 k = bkey_i_to_s_c(sk.k);
584                 bch2_trans_unlock(&trans);
585
586                 ret2 = bch2_move_extent(c, ctxt, wp, io_opts, btree_id, k,
587                                         data_cmd, data_opts);
588                 if (ret2) {
589                         if (ret2 == -ENOMEM) {
590                                 /* memory allocation failure, wait for some IO to finish */
591                                 bch2_move_ctxt_wait_for_io(ctxt);
592                                 continue;
593                         }
594
595                         /* XXX signal failure */
596                         goto next;
597                 }
598
599                 if (rate)
600                         bch2_ratelimit_increment(rate, k.k->size);
601 next:
602                 atomic64_add(k.k->size * bch2_bkey_nr_ptrs_allocated(k),
603                              &stats->sectors_seen);
604 next_nondata:
605                 bch2_btree_iter_next(iter);
606                 bch2_trans_cond_resched(&trans);
607         }
608 out:
609         ret = bch2_trans_exit(&trans) ?: ret;
610         bkey_on_stack_exit(&sk, c);
611
612         return ret;
613 }
614
615 int bch2_move_data(struct bch_fs *c,
616                    struct bch_ratelimit *rate,
617                    struct write_point_specifier wp,
618                    struct bpos start,
619                    struct bpos end,
620                    move_pred_fn pred, void *arg,
621                    struct bch_move_stats *stats)
622 {
623         struct moving_context ctxt = { .stats = stats };
624         int ret;
625
626         closure_init_stack(&ctxt.cl);
627         INIT_LIST_HEAD(&ctxt.reads);
628         init_waitqueue_head(&ctxt.wait);
629
630         stats->data_type = BCH_DATA_USER;
631
632         ret =   __bch2_move_data(c, &ctxt, rate, wp, start, end,
633                                  pred, arg, stats, BTREE_ID_EXTENTS) ?:
634                 __bch2_move_data(c, &ctxt, rate, wp, start, end,
635                                  pred, arg, stats, BTREE_ID_REFLINK);
636
637         move_ctxt_wait_event(&ctxt, list_empty(&ctxt.reads));
638         closure_sync(&ctxt.cl);
639
640         EBUG_ON(atomic_read(&ctxt.write_sectors));
641
642         trace_move_data(c,
643                         atomic64_read(&stats->sectors_moved),
644                         atomic64_read(&stats->keys_moved));
645
646         return ret;
647 }
648
649 static int bch2_move_btree(struct bch_fs *c,
650                            move_pred_fn pred,
651                            void *arg,
652                            struct bch_move_stats *stats)
653 {
654         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
655         struct btree_trans trans;
656         struct btree_iter *iter;
657         struct btree *b;
658         unsigned id;
659         struct data_opts data_opts;
660         enum data_cmd cmd;
661         int ret = 0;
662
663         bch2_trans_init(&trans, c, 0, 0);
664
665         stats->data_type = BCH_DATA_BTREE;
666
667         for (id = 0; id < BTREE_ID_NR; id++) {
668                 stats->btree_id = id;
669
670                 for_each_btree_node(&trans, iter, id, POS_MIN,
671                                     BTREE_ITER_PREFETCH, b) {
672                         stats->pos = iter->pos;
673
674                         switch ((cmd = pred(c, arg,
675                                             bkey_i_to_s_c(&b->key),
676                                             &io_opts, &data_opts))) {
677                         case DATA_SKIP:
678                                 goto next;
679                         case DATA_SCRUB:
680                                 BUG();
681                         case DATA_ADD_REPLICAS:
682                         case DATA_REWRITE:
683                                 break;
684                         default:
685                                 BUG();
686                         }
687
688                         ret = bch2_btree_node_rewrite(c, iter,
689                                         b->data->keys.seq, 0) ?: ret;
690 next:
691                         bch2_trans_cond_resched(&trans);
692                 }
693
694                 ret = bch2_trans_iter_free(&trans, iter) ?: ret;
695         }
696
697         bch2_trans_exit(&trans);
698
699         return ret;
700 }
701
702 #if 0
703 static enum data_cmd scrub_pred(struct bch_fs *c, void *arg,
704                                 struct bkey_s_c k,
705                                 struct bch_io_opts *io_opts,
706                                 struct data_opts *data_opts)
707 {
708         return DATA_SCRUB;
709 }
710 #endif
711
712 static enum data_cmd rereplicate_pred(struct bch_fs *c, void *arg,
713                                       struct bkey_s_c k,
714                                       struct bch_io_opts *io_opts,
715                                       struct data_opts *data_opts)
716 {
717         unsigned nr_good = bch2_bkey_durability(c, k);
718         unsigned replicas = 0;
719
720         switch (k.k->type) {
721         case KEY_TYPE_btree_ptr:
722                 replicas = c->opts.metadata_replicas;
723                 break;
724         case KEY_TYPE_extent:
725                 replicas = io_opts->data_replicas;
726                 break;
727         }
728
729         if (!nr_good || nr_good >= replicas)
730                 return DATA_SKIP;
731
732         data_opts->target               = 0;
733         data_opts->btree_insert_flags   = 0;
734         return DATA_ADD_REPLICAS;
735 }
736
737 static enum data_cmd migrate_pred(struct bch_fs *c, void *arg,
738                                   struct bkey_s_c k,
739                                   struct bch_io_opts *io_opts,
740                                   struct data_opts *data_opts)
741 {
742         struct bch_ioctl_data *op = arg;
743
744         if (!bch2_bkey_has_device(k, op->migrate.dev))
745                 return DATA_SKIP;
746
747         data_opts->target               = 0;
748         data_opts->btree_insert_flags   = 0;
749         data_opts->rewrite_dev          = op->migrate.dev;
750         return DATA_REWRITE;
751 }
752
753 int bch2_data_job(struct bch_fs *c,
754                   struct bch_move_stats *stats,
755                   struct bch_ioctl_data op)
756 {
757         int ret = 0;
758
759         switch (op.op) {
760         case BCH_DATA_OP_REREPLICATE:
761                 stats->data_type = BCH_DATA_JOURNAL;
762                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
763
764                 ret = bch2_move_btree(c, rereplicate_pred, c, stats) ?: ret;
765
766                 while (1) {
767                         closure_wait_event(&c->btree_interior_update_wait,
768                                            !bch2_btree_interior_updates_nr_pending(c) ||
769                                            c->btree_roots_dirty);
770                         if (!bch2_btree_interior_updates_nr_pending(c))
771                                 break;
772                         bch2_journal_meta(&c->journal);
773                 }
774
775                 ret = bch2_replicas_gc2(c) ?: ret;
776
777                 ret = bch2_move_data(c, NULL,
778                                      writepoint_hashed((unsigned long) current),
779                                      op.start,
780                                      op.end,
781                                      rereplicate_pred, c, stats) ?: ret;
782                 ret = bch2_replicas_gc2(c) ?: ret;
783                 break;
784         case BCH_DATA_OP_MIGRATE:
785                 if (op.migrate.dev >= c->sb.nr_devices)
786                         return -EINVAL;
787
788                 stats->data_type = BCH_DATA_JOURNAL;
789                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
790
791                 ret = bch2_move_btree(c, migrate_pred, &op, stats) ?: ret;
792                 ret = bch2_replicas_gc2(c) ?: ret;
793
794                 ret = bch2_move_data(c, NULL,
795                                      writepoint_hashed((unsigned long) current),
796                                      op.start,
797                                      op.end,
798                                      migrate_pred, &op, stats) ?: ret;
799                 ret = bch2_replicas_gc2(c) ?: ret;
800                 break;
801         default:
802                 ret = -EINVAL;
803         }
804
805         return ret;
806 }