1 // SPDX-License-Identifier: GPL-2.0
2
3 #include "bcachefs.h"
4 #include "alloc_foreground.h"
5 #include "bkey_buf.h"
6 #include "btree_update.h"
7 #include "buckets.h"
8 #include "compress.h"
9 #include "data_update.h"
10 #include "disk_groups.h"
11 #include "ec.h"
12 #include "error.h"
13 #include "extents.h"
14 #include "io_write.h"
15 #include "keylist.h"
16 #include "move.h"
17 #include "nocow_locking.h"
18 #include "rebalance.h"
19 #include "snapshot.h"
20 #include "subvolume.h"
21 #include "trace.h"
22
23 static void trace_move_extent_finish2(struct bch_fs *c, struct bkey_s_c k)
24 {
25 if (trace_move_extent_finish_enabled()) {
26 struct printbuf buf = PRINTBUF;
27
28 bch2_bkey_val_to_text(&buf, c, k);
29 trace_move_extent_finish(c, buf.buf);
30 printbuf_exit(&buf);
31 }
32 }
33
34 static void trace_move_extent_fail2(struct data_update *m,
35 struct bkey_s_c new,
36 struct bkey_s_c wrote,
37 struct bkey_i *insert,
38 const char *msg)
39 {
40 struct bch_fs *c = m->op.c;
41 struct bkey_s_c old = bkey_i_to_s_c(m->k.k);
42 const union bch_extent_entry *entry;
43 struct bch_extent_ptr *ptr;
44 struct extent_ptr_decoded p;
45 struct printbuf buf = PRINTBUF;
46 unsigned i, rewrites_found = 0;
47
48 if (!trace_move_extent_fail_enabled())
49 return;
50
51 prt_str(&buf, msg);
52
53 if (insert) {
54 i = 0;
55 bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs_c(old), p, entry) {
56 if (((1U << i) & m->data_opts.rewrite_ptrs) &&
57 (ptr = bch2_extent_has_ptr(old, p, bkey_i_to_s(insert))) &&
58 !ptr->cached)
59 rewrites_found |= 1U << i;
60 i++;
61 }
62 }
63
64 prt_printf(&buf, "\nrewrite ptrs: %u%u%u%u",
65 (m->data_opts.rewrite_ptrs & (1 << 0)) != 0,
66 (m->data_opts.rewrite_ptrs & (1 << 1)) != 0,
67 (m->data_opts.rewrite_ptrs & (1 << 2)) != 0,
68 (m->data_opts.rewrite_ptrs & (1 << 3)) != 0);
69
70 prt_printf(&buf, "\nrewrites found: %u%u%u%u",
71 (rewrites_found & (1 << 0)) != 0,
72 (rewrites_found & (1 << 1)) != 0,
73 (rewrites_found & (1 << 2)) != 0,
74 (rewrites_found & (1 << 3)) != 0);
75
76 prt_str(&buf, "\nold: ");
77 bch2_bkey_val_to_text(&buf, c, old);
78
79 prt_str(&buf, "\nnew: ");
80 bch2_bkey_val_to_text(&buf, c, new);
81
82 prt_str(&buf, "\nwrote: ");
83 bch2_bkey_val_to_text(&buf, c, wrote);
84
85 if (insert) {
86 prt_str(&buf, "\ninsert: ");
87 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(insert));
88 }
89
90 trace_move_extent_fail(c, buf.buf);
91 printbuf_exit(&buf);
92 }
93
94 static int __bch2_data_update_index_update(struct btree_trans *trans,
95 struct bch_write_op *op)
96 {
97 struct bch_fs *c = op->c;
98 struct btree_iter iter;
99 struct data_update *m =
100 container_of(op, struct data_update, op);
101 struct keylist *keys = &op->insert_keys;
102 struct bkey_buf _new, _insert;
103 int ret = 0;
104
105 bch2_bkey_buf_init(&_new);
106 bch2_bkey_buf_init(&_insert);
107 bch2_bkey_buf_realloc(&_insert, c, U8_MAX);
108
109 bch2_trans_iter_init(trans, &iter, m->btree_id,
110 bkey_start_pos(&bch2_keylist_front(keys)->k),
111 BTREE_ITER_slots|BTREE_ITER_intent);
112
113 while (1) {
114 struct bkey_s_c k;
115 struct bkey_s_c old = bkey_i_to_s_c(m->k.k);
116 struct bkey_i *insert = NULL;
117 struct bkey_i_extent *new;
118 const union bch_extent_entry *entry_c;
119 union bch_extent_entry *entry;
120 struct extent_ptr_decoded p;
121 struct bch_extent_ptr *ptr;
122 const struct bch_extent_ptr *ptr_c;
123 struct bpos next_pos;
124 bool should_check_enospc;
125 s64 i_sectors_delta = 0, disk_sectors_delta = 0;
126 unsigned rewrites_found = 0, durability, i;
127
128 bch2_trans_begin(trans);
129
130 k = bch2_btree_iter_peek_slot(&iter);
131 ret = bkey_err(k);
132 if (ret)
133 goto err;
134
135 new = bkey_i_to_extent(bch2_keylist_front(keys));
136
137 if (!bch2_extents_match(k, old)) {
138 trace_move_extent_fail2(m, k, bkey_i_to_s_c(&new->k_i),
139 NULL, "no match:");
140 goto nowork;
141 }
142
143 bkey_reassemble(_insert.k, k);
144 insert = _insert.k;
145
146 bch2_bkey_buf_copy(&_new, c, bch2_keylist_front(keys));
147 new = bkey_i_to_extent(_new.k);
148 bch2_cut_front(iter.pos, &new->k_i);
149
150 bch2_cut_front(iter.pos, insert);
151 bch2_cut_back(new->k.p, insert);
152 bch2_cut_back(insert->k.p, &new->k_i);
153
154 /*
155 * @old: extent that we read from
156 * @insert: key that we're going to update, initialized from
157 * extent currently in btree - same as @old unless we raced with
158 * other updates
159 * @new: extent with new pointers that we'll be adding to @insert
160 *
161 * Fist, drop rewrite_ptrs from @new:
162 */
163 i = 0;
164 bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs_c(old), p, entry_c) {
165 if (((1U << i) & m->data_opts.rewrite_ptrs) &&
166 (ptr = bch2_extent_has_ptr(old, p, bkey_i_to_s(insert))) &&
167 !ptr->cached) {
168 bch2_extent_ptr_set_cached(bkey_i_to_s(insert), ptr);
169 rewrites_found |= 1U << i;
170 }
171 i++;
172 }
173
174 if (m->data_opts.rewrite_ptrs &&
175 !rewrites_found &&
176 bch2_bkey_durability(c, k) >= m->op.opts.data_replicas) {
177 trace_move_extent_fail2(m, k, bkey_i_to_s_c(&new->k_i), insert, "no rewrites found:");
178 goto nowork;
179 }
180
181 /*
182 * A replica that we just wrote might conflict with a replica
183 * that we want to keep, due to racing with another move:
184 */
185 restart_drop_conflicting_replicas:
186 extent_for_each_ptr(extent_i_to_s(new), ptr)
187 if ((ptr_c = bch2_bkey_has_device_c(bkey_i_to_s_c(insert), ptr->dev)) &&
188 !ptr_c->cached) {
189 bch2_bkey_drop_ptr_noerror(bkey_i_to_s(&new->k_i), ptr);
190 goto restart_drop_conflicting_replicas;
191 }
192
193 if (!bkey_val_u64s(&new->k)) {
194 trace_move_extent_fail2(m, k, bkey_i_to_s_c(&new->k_i), insert, "new replicas conflicted:");
195 goto nowork;
196 }
197
198 /* Now, drop pointers that conflict with what we just wrote: */
199 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry)
200 if ((ptr = bch2_bkey_has_device(bkey_i_to_s(insert), p.ptr.dev)))
201 bch2_bkey_drop_ptr_noerror(bkey_i_to_s(insert), ptr);
202
203 durability = bch2_bkey_durability(c, bkey_i_to_s_c(insert)) +
204 bch2_bkey_durability(c, bkey_i_to_s_c(&new->k_i));
205
206 /* Now, drop excess replicas: */
207 rcu_read_lock();
208 restart_drop_extra_replicas:
209 bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs(bkey_i_to_s(insert)), p, entry) {
210 unsigned ptr_durability = bch2_extent_ptr_durability(c, &p);
211
212 if (!p.ptr.cached &&
213 durability - ptr_durability >= m->op.opts.data_replicas) {
214 durability -= ptr_durability;
215
216 bch2_extent_ptr_set_cached(bkey_i_to_s(insert), &entry->ptr);
217 goto restart_drop_extra_replicas;
218 }
219 }
220 rcu_read_unlock();
221
222 /* Finally, add the pointers we just wrote: */
223 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry)
224 bch2_extent_ptr_decoded_append(insert, &p);
225
226 bch2_bkey_narrow_crcs(insert, (struct bch_extent_crc_unpacked) { 0 });
227 bch2_extent_normalize(c, bkey_i_to_s(insert));
228
229 ret = bch2_sum_sector_overwrites(trans, &iter, insert,
230 &should_check_enospc,
231 &i_sectors_delta,
232 &disk_sectors_delta);
233 if (ret)
234 goto err;
235
236 if (disk_sectors_delta > (s64) op->res.sectors) {
237 ret = bch2_disk_reservation_add(c, &op->res,
238 disk_sectors_delta - op->res.sectors,
239 !should_check_enospc
240 ? BCH_DISK_RESERVATION_NOFAIL : 0);
241 if (ret)
242 goto out;
243 }
244
245 next_pos = insert->k.p;
246
247 /*
248 * Check for nonce offset inconsistency:
249 * This is debug code - we've been seeing this bug rarely, and
250 * it's been hard to reproduce, so this should give us some more
251 * information when it does occur:
252 */
253 struct printbuf err = PRINTBUF;
254 int invalid = bch2_bkey_invalid(c, bkey_i_to_s_c(insert), __btree_node_type(0, m->btree_id), 0, &err);
255 printbuf_exit(&err);
256
257 if (invalid) {
258 struct printbuf buf = PRINTBUF;
259
260 prt_str(&buf, "about to insert invalid key in data update path");
261 prt_str(&buf, "\nold: ");
262 bch2_bkey_val_to_text(&buf, c, old);
263 prt_str(&buf, "\nk: ");
264 bch2_bkey_val_to_text(&buf, c, k);
265 prt_str(&buf, "\nnew: ");
266 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(insert));
267
268 bch2_print_string_as_lines(KERN_ERR, buf.buf);
269 printbuf_exit(&buf);
270
271 bch2_fatal_error(c);
272 goto out;
273 }
274
275 if (trace_data_update_enabled()) {
276 struct printbuf buf = PRINTBUF;
277
278 prt_str(&buf, "\nold: ");
279 bch2_bkey_val_to_text(&buf, c, old);
280 prt_str(&buf, "\nk: ");
281 bch2_bkey_val_to_text(&buf, c, k);
282 prt_str(&buf, "\nnew: ");
283 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(insert));
284
285 trace_data_update(c, buf.buf);
286 printbuf_exit(&buf);
287 }
288
289 ret = bch2_insert_snapshot_whiteouts(trans, m->btree_id,
290 k.k->p, bkey_start_pos(&insert->k)) ?:
291 bch2_insert_snapshot_whiteouts(trans, m->btree_id,
292 k.k->p, insert->k.p) ?:
293 bch2_bkey_set_needs_rebalance(c, insert, &op->opts) ?:
294 bch2_trans_update(trans, &iter, insert,
295 BTREE_UPDATE_internal_snapshot_node) ?:
296 bch2_trans_commit(trans, &op->res,
297 NULL,
298 BCH_TRANS_COMMIT_no_check_rw|
299 BCH_TRANS_COMMIT_no_enospc|
300 m->data_opts.btree_insert_flags);
301 if (!ret) {
302 bch2_btree_iter_set_pos(&iter, next_pos);
303
304 this_cpu_add(c->counters[BCH_COUNTER_move_extent_finish], new->k.size);
305 trace_move_extent_finish2(c, bkey_i_to_s_c(&new->k_i));
306 }
307 err:
308 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
309 ret = 0;
310 if (ret)
311 break;
312 next:
313 while (bkey_ge(iter.pos, bch2_keylist_front(keys)->k.p)) {
314 bch2_keylist_pop_front(keys);
315 if (bch2_keylist_empty(keys))
316 goto out;
317 }
318 continue;
319 nowork:
320 if (m->stats) {
321 BUG_ON(k.k->p.offset <= iter.pos.offset);
322 atomic64_inc(&m->stats->keys_raced);
323 atomic64_add(k.k->p.offset - iter.pos.offset,
324 &m->stats->sectors_raced);
325 }
326
327 count_event(c, move_extent_fail);
328
329 bch2_btree_iter_advance(&iter);
330 goto next;
331 }
332 out:
333 bch2_trans_iter_exit(trans, &iter);
334 bch2_bkey_buf_exit(&_insert, c);
335 bch2_bkey_buf_exit(&_new, c);
336 BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart));
337 return ret;
338 }
339
340 int bch2_data_update_index_update(struct bch_write_op *op)
341 {
342 return bch2_trans_run(op->c, __bch2_data_update_index_update(trans, op));
343 }
344
345 void bch2_data_update_read_done(struct data_update *m,
346 struct bch_extent_crc_unpacked crc)
347 {
348 /* write bio must own pages: */
349 BUG_ON(!m->op.wbio.bio.bi_vcnt);
350
351 m->op.crc = crc;
352 m->op.wbio.bio.bi_iter.bi_size = crc.compressed_size << 9;
353
354 closure_call(&m->op.cl, bch2_write, NULL, NULL);
355 }
356
357 void bch2_data_update_exit(struct data_update *update)
358 {
359 struct bch_fs *c = update->op.c;
360 struct bkey_ptrs_c ptrs =
361 bch2_bkey_ptrs_c(bkey_i_to_s_c(update->k.k));
362
363 bkey_for_each_ptr(ptrs, ptr) {
364 struct bch_dev *ca = bch2_dev_have_ref(c, ptr->dev);
365 if (c->opts.nocow_enabled)
366 bch2_bucket_nocow_unlock(&c->nocow_locks,
367 PTR_BUCKET_POS(ca, ptr), 0);
368 bch2_dev_put(ca);
369 }
370
371 bch2_bkey_buf_exit(&update->k, c);
372 bch2_disk_reservation_put(c, &update->op.res);
373 bch2_bio_free_pages_pool(c, &update->op.wbio.bio);
374 }
375
376 static void bch2_update_unwritten_extent(struct btree_trans *trans,
377 struct data_update *update)
378 {
379 struct bch_fs *c = update->op.c;
380 struct bio *bio = &update->op.wbio.bio;
381 struct bkey_i_extent *e;
382 struct write_point *wp;
383 struct closure cl;
384 struct btree_iter iter;
385 struct bkey_s_c k;
386 int ret;
387
388 closure_init_stack(&cl);
389 bch2_keylist_init(&update->op.insert_keys, update->op.inline_keys);
390
391 while (bio_sectors(bio)) {
392 unsigned sectors = bio_sectors(bio);
393
394 bch2_trans_begin(trans);
395
396 bch2_trans_iter_init(trans, &iter, update->btree_id, update->op.pos,
397 BTREE_ITER_slots);
398 ret = lockrestart_do(trans, ({
399 k = bch2_btree_iter_peek_slot(&iter);
400 bkey_err(k);
401 }));
402 bch2_trans_iter_exit(trans, &iter);
403
404 if (ret || !bch2_extents_match(k, bkey_i_to_s_c(update->k.k)))
405 break;
406
407 e = bkey_extent_init(update->op.insert_keys.top);
408 e->k.p = update->op.pos;
409
410 ret = bch2_alloc_sectors_start_trans(trans,
411 update->op.target,
412 false,
413 update->op.write_point,
414 &update->op.devs_have,
415 update->op.nr_replicas,
416 update->op.nr_replicas,
417 update->op.watermark,
418 0, &cl, &wp);
419 if (bch2_err_matches(ret, BCH_ERR_operation_blocked)) {
420 bch2_trans_unlock(trans);
421 closure_sync(&cl);
422 continue;
423 }
424
425 bch_err_fn_ratelimited(c, ret);
426
427 if (ret)
428 return;
429
430 sectors = min(sectors, wp->sectors_free);
431
432 bch2_key_resize(&e->k, sectors);
433
434 bch2_open_bucket_get(c, wp, &update->op.open_buckets);
435 bch2_alloc_sectors_append_ptrs(c, wp, &e->k_i, sectors, false);
436 bch2_alloc_sectors_done(c, wp);
437
438 bio_advance(bio, sectors << 9);
439 update->op.pos.offset += sectors;
440
441 extent_for_each_ptr(extent_i_to_s(e), ptr)
442 ptr->unwritten = true;
443 bch2_keylist_push(&update->op.insert_keys);
444
445 ret = __bch2_data_update_index_update(trans, &update->op);
446
447 bch2_open_buckets_put(c, &update->op.open_buckets);
448
449 if (ret)
450 break;
451 }
452
453 if (closure_nr_remaining(&cl) != 1) {
454 bch2_trans_unlock(trans);
455 closure_sync(&cl);
456 }
457 }
458
459 void bch2_data_update_opts_to_text(struct printbuf *out, struct bch_fs *c,
460 struct bch_io_opts *io_opts,
461 struct data_update_opts *data_opts)
462 {
463 printbuf_tabstop_push(out, 20);
464 prt_str(out, "rewrite ptrs:\t");
465 bch2_prt_u64_base2(out, data_opts->rewrite_ptrs);
466 prt_newline(out);
467
468 prt_str(out, "kill ptrs:\t");
469 bch2_prt_u64_base2(out, data_opts->kill_ptrs);
470 prt_newline(out);
471
472 prt_str(out, "target:\t");
473 bch2_target_to_text(out, c, data_opts->target);
474 prt_newline(out);
475
476 prt_str(out, "compression:\t");
477 bch2_compression_opt_to_text(out, background_compression(*io_opts));
478 prt_newline(out);
479
480 prt_str(out, "extra replicas:\t");
481 prt_u64(out, data_opts->extra_replicas);
482 }
483
484 void bch2_data_update_to_text(struct printbuf *out, struct data_update *m)
485 {
486 bch2_bkey_val_to_text(out, m->op.c, bkey_i_to_s_c(m->k.k));
487 prt_newline(out);
488 bch2_data_update_opts_to_text(out, m->op.c, &m->op.opts, &m->data_opts);
489 }
490
491 int bch2_extent_drop_ptrs(struct btree_trans *trans,
492 struct btree_iter *iter,
493 struct bkey_s_c k,
494 struct data_update_opts data_opts)
495 {
496 struct bch_fs *c = trans->c;
497 struct bkey_i *n;
498 int ret;
499
500 n = bch2_bkey_make_mut_noupdate(trans, k);
501 ret = PTR_ERR_OR_ZERO(n);
502 if (ret)
503 return ret;
504
505 while (data_opts.kill_ptrs) {
506 unsigned i = 0, drop = __fls(data_opts.kill_ptrs);
507
508 bch2_bkey_drop_ptrs(bkey_i_to_s(n), ptr, i++ == drop);
509 data_opts.kill_ptrs ^= 1U << drop;
510 }
511
512 /*
513 * If the new extent no longer has any pointers, bch2_extent_normalize()
514 * will do the appropriate thing with it (turning it into a
515 * KEY_TYPE_error key, or just a discard if it was a cached extent)
516 */
517 bch2_extent_normalize(c, bkey_i_to_s(n));
518
519 /*
520 * Since we're not inserting through an extent iterator
521 * (BTREE_ITER_all_snapshots iterators aren't extent iterators),
522 * we aren't using the extent overwrite path to delete, we're
523 * just using the normal key deletion path:
524 */
525 if (bkey_deleted(&n->k) && !(iter->flags & BTREE_ITER_is_extents))
526 n->k.size = 0;
527
528 return bch2_trans_relock(trans) ?:
529 bch2_trans_update(trans, iter, n, BTREE_UPDATE_internal_snapshot_node) ?:
530 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
531 }
532
533 int bch2_data_update_init(struct btree_trans *trans,
534 struct btree_iter *iter,
535 struct moving_context *ctxt,
536 struct data_update *m,
537 struct write_point_specifier wp,
538 struct bch_io_opts io_opts,
539 struct data_update_opts data_opts,
540 enum btree_id btree_id,
541 struct bkey_s_c k)
542 {
543 struct bch_fs *c = trans->c;
544 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
545 const union bch_extent_entry *entry;
546 struct extent_ptr_decoded p;
547 unsigned i, reserve_sectors = k.k->size * data_opts.extra_replicas;
548 unsigned ptrs_locked = 0;
549 int ret = 0;
550
551 /*
552 * fs is corrupt we have a key for a snapshot node that doesn't exist,
553 * and we have to check for this because we go rw before repairing the
554 * snapshots table - just skip it, we can move it later.
555 */
556 if (unlikely(k.k->p.snapshot && !bch2_snapshot_equiv(c, k.k->p.snapshot)))
557 return -BCH_ERR_data_update_done;
558
559 bch2_bkey_buf_init(&m->k);
560 bch2_bkey_buf_reassemble(&m->k, c, k);
561 m->btree_id = btree_id;
562 m->data_opts = data_opts;
563 m->ctxt = ctxt;
564 m->stats = ctxt ? ctxt->stats : NULL;
565
566 bch2_write_op_init(&m->op, c, io_opts);
567 m->op.pos = bkey_start_pos(k.k);
568 m->op.version = k.k->version;
569 m->op.target = data_opts.target;
570 m->op.write_point = wp;
571 m->op.nr_replicas = 0;
572 m->op.flags |= BCH_WRITE_PAGES_STABLE|
573 BCH_WRITE_PAGES_OWNED|
574 BCH_WRITE_DATA_ENCODED|
575 BCH_WRITE_MOVE|
576 m->data_opts.write_flags;
577 m->op.compression_opt = background_compression(io_opts);
578 m->op.watermark = m->data_opts.btree_insert_flags & BCH_WATERMARK_MASK;
579
580 bkey_for_each_ptr(ptrs, ptr) {
581 if (!bch2_dev_tryget(c, ptr->dev)) {
582 bkey_for_each_ptr(ptrs, ptr2) {
583 if (ptr2 == ptr)
584 break;
585 bch2_dev_put(bch2_dev_have_ref(c, ptr2->dev));
586 }
587 return -BCH_ERR_data_update_done;
588 }
589 }
590
591 unsigned durability_have = 0, durability_removing = 0;
592
593 i = 0;
594 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
595 struct bch_dev *ca = bch2_dev_have_ref(c, p.ptr.dev);
596 struct bpos bucket = PTR_BUCKET_POS(ca, &p.ptr);
597 bool locked;
598
599 rcu_read_lock();
600 if (((1U << i) & m->data_opts.rewrite_ptrs)) {
601 BUG_ON(p.ptr.cached);
602
603 if (crc_is_compressed(p.crc))
604 reserve_sectors += k.k->size;
605
606 m->op.nr_replicas += bch2_extent_ptr_desired_durability(c, &p);
607 durability_removing += bch2_extent_ptr_desired_durability(c, &p);
608 } else if (!p.ptr.cached &&
609 !((1U << i) & m->data_opts.kill_ptrs)) {
610 bch2_dev_list_add_dev(&m->op.devs_have, p.ptr.dev);
611 durability_have += bch2_extent_ptr_durability(c, &p);
612 }
613 rcu_read_unlock();
614
615 /*
616 * op->csum_type is normally initialized from the fs/file's
617 * current options - but if an extent is encrypted, we require
618 * that it stays encrypted:
619 */
620 if (bch2_csum_type_is_encryption(p.crc.csum_type)) {
621 m->op.nonce = p.crc.nonce + p.crc.offset;
622 m->op.csum_type = p.crc.csum_type;
623 }
624
625 if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
626 m->op.incompressible = true;
627
628 if (c->opts.nocow_enabled) {
629 if (ctxt) {
630 move_ctxt_wait_event(ctxt,
631 (locked = bch2_bucket_nocow_trylock(&c->nocow_locks,
632 bucket, 0)) ||
633 list_empty(&ctxt->ios));
634
635 if (!locked)
636 bch2_bucket_nocow_lock(&c->nocow_locks, bucket, 0);
637 } else {
638 if (!bch2_bucket_nocow_trylock(&c->nocow_locks, bucket, 0)) {
639 ret = -BCH_ERR_nocow_lock_blocked;
640 goto err;
641 }
642 }
643 ptrs_locked |= (1U << i);
644 }
645
646 i++;
647 }
648
649 unsigned durability_required = max(0, (int) (io_opts.data_replicas - durability_have));
650
651 /*
652 * If current extent durability is less than io_opts.data_replicas,
653 * we're not trying to rereplicate the extent up to data_replicas here -
654 * unless extra_replicas was specified
655 *
656 * Increasing replication is an explicit operation triggered by
657 * rereplicate, currently, so that users don't get an unexpected -ENOSPC
658 */
659 if (!(m->data_opts.write_flags & BCH_WRITE_CACHED) &&
660 !durability_required) {
661 m->data_opts.kill_ptrs |= m->data_opts.rewrite_ptrs;
662 m->data_opts.rewrite_ptrs = 0;
663 /* if iter == NULL, it's just a promote */
664 if (iter)
665 ret = bch2_extent_drop_ptrs(trans, iter, k, m->data_opts);
666 goto done;
667 }
668
669 m->op.nr_replicas = min(durability_removing, durability_required) +
670 m->data_opts.extra_replicas;
671
672 /*
673 * If device(s) were set to durability=0 after data was written to them
674 * we can end up with a duribilty=0 extent, and the normal algorithm
675 * that tries not to increase durability doesn't work:
676 */
677 if (!(durability_have + durability_removing))
678 m->op.nr_replicas = max((unsigned) m->op.nr_replicas, 1);
679
680 if (!m->op.nr_replicas) {
681 struct printbuf buf = PRINTBUF;
682
683 bch2_data_update_to_text(&buf, m);
684 WARN(1, "trying to move an extent, but nr_replicas=0\n%s", buf.buf);
685 printbuf_exit(&buf);
686 ret = -BCH_ERR_data_update_done;
687 goto done;
688 }
689
690 m->op.nr_replicas_required = m->op.nr_replicas;
691
692 if (reserve_sectors) {
693 ret = bch2_disk_reservation_add(c, &m->op.res, reserve_sectors,
694 m->data_opts.extra_replicas
695 ? 0
696 : BCH_DISK_RESERVATION_NOFAIL);
697 if (ret)
698 goto err;
699 }
700
701 if (bkey_extent_is_unwritten(k)) {
702 bch2_update_unwritten_extent(trans, m);
703 goto done;
704 }
705
706 return 0;
707 err:
708 i = 0;
709 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
710 struct bch_dev *ca = bch2_dev_have_ref(c, p.ptr.dev);
711 struct bpos bucket = PTR_BUCKET_POS(ca, &p.ptr);
712 if ((1U << i) & ptrs_locked)
713 bch2_bucket_nocow_unlock(&c->nocow_locks, bucket, 0);
714 bch2_dev_put(ca);
715 i++;
716 }
717
718 bch2_bkey_buf_exit(&m->k, c);
719 bch2_bio_free_pages_pool(c, &m->op.wbio.bio);
720 return ret;
721 done:
722 bch2_data_update_exit(m);
723 return ret ?: -BCH_ERR_data_update_done;
724 }
725
726 void bch2_data_update_opts_normalize(struct bkey_s_c k, struct data_update_opts *opts)
727 {
728 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
729 unsigned i = 0;
730
731 bkey_for_each_ptr(ptrs, ptr) {
732 if ((opts->rewrite_ptrs & (1U << i)) && ptr->cached) {
733 opts->kill_ptrs |= 1U << i;
734 opts->rewrite_ptrs ^= 1U << i;
735 }
736
737 i++;
738 }
739 }
740
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.