1 // SPDX-License-Identifier: GPL-2.0
4 #include "alloc_foreground.h"
6 #include "btree_update.h"
8 #include "data_update.h"
14 #include "subvolume.h"
16 #include <trace/events/bcachefs.h>
18 static int insert_snapshot_whiteouts(struct btree_trans *trans,
23 struct bch_fs *c = trans->c;
24 struct btree_iter iter, update_iter;
29 if (!btree_type_has_snapshots(id))
34 if (bkey_eq(old_pos, new_pos))
37 if (!snapshot_t(c, old_pos.snapshot)->children[0])
40 bch2_trans_iter_init(trans, &iter, id, old_pos,
41 BTREE_ITER_NOT_EXTENTS|
42 BTREE_ITER_ALL_SNAPSHOTS);
44 k = bch2_btree_iter_prev(&iter);
49 if (!bkey_eq(old_pos, k.k->p))
52 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, old_pos.snapshot)) {
53 struct bkey_i *update;
55 if (snapshot_list_has_ancestor(c, &s, k.k->p.snapshot))
58 update = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
60 ret = PTR_ERR_OR_ZERO(update);
64 bkey_init(&update->k);
65 update->k.p = new_pos;
66 update->k.p.snapshot = k.k->p.snapshot;
68 bch2_trans_iter_init(trans, &update_iter, id, update->k.p,
69 BTREE_ITER_NOT_EXTENTS|
70 BTREE_ITER_ALL_SNAPSHOTS|
72 ret = bch2_btree_iter_traverse(&update_iter) ?:
73 bch2_trans_update(trans, &update_iter, update,
74 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
75 bch2_trans_iter_exit(trans, &update_iter);
79 ret = snapshot_list_add(c, &s, k.k->p.snapshot);
84 bch2_trans_iter_exit(trans, &iter);
90 static void bch2_bkey_mark_dev_cached(struct bkey_s k, unsigned dev)
92 struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
93 struct bch_extent_ptr *ptr;
95 bkey_for_each_ptr(ptrs, ptr)
100 static int __bch2_data_update_index_update(struct btree_trans *trans,
101 struct bch_write_op *op)
103 struct bch_fs *c = op->c;
104 struct btree_iter iter;
105 struct data_update *m =
106 container_of(op, struct data_update, op);
107 struct keylist *keys = &op->insert_keys;
108 struct bkey_buf _new, _insert;
111 bch2_bkey_buf_init(&_new);
112 bch2_bkey_buf_init(&_insert);
113 bch2_bkey_buf_realloc(&_insert, c, U8_MAX);
115 bch2_trans_iter_init(trans, &iter, m->btree_id,
116 bkey_start_pos(&bch2_keylist_front(keys)->k),
117 BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
121 struct bkey_s_c old = bkey_i_to_s_c(m->k.k);
122 struct bkey_i *insert;
123 struct bkey_i_extent *new;
124 const union bch_extent_entry *entry;
125 struct extent_ptr_decoded p;
126 struct bpos next_pos;
127 bool did_work = false;
128 bool should_check_enospc;
129 s64 i_sectors_delta = 0, disk_sectors_delta = 0;
132 bch2_trans_begin(trans);
134 k = bch2_btree_iter_peek_slot(&iter);
139 new = bkey_i_to_extent(bch2_keylist_front(keys));
141 if (!bch2_extents_match(k, old))
144 bkey_reassemble(_insert.k, k);
147 bch2_bkey_buf_copy(&_new, c, bch2_keylist_front(keys));
148 new = bkey_i_to_extent(_new.k);
149 bch2_cut_front(iter.pos, &new->k_i);
151 bch2_cut_front(iter.pos, insert);
152 bch2_cut_back(new->k.p, insert);
153 bch2_cut_back(insert->k.p, &new->k_i);
156 * @old: extent that we read from
157 * @insert: key that we're going to update, initialized from
158 * extent currently in btree - same as @old unless we raced with
160 * @new: extent with new pointers that we'll be adding to @insert
162 * Fist, drop rewrite_ptrs from @new:
165 bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs_c(old), p, entry) {
166 if (((1U << i) & m->data_opts.rewrite_ptrs) &&
167 bch2_extent_has_ptr(old, p, bkey_i_to_s_c(insert))) {
169 * If we're going to be adding a pointer to the
170 * same device, we have to drop the old one -
171 * otherwise, we can just mark it cached:
173 if (bch2_bkey_has_device(bkey_i_to_s_c(&new->k_i), p.ptr.dev))
174 bch2_bkey_drop_device_noerror(bkey_i_to_s(insert), p.ptr.dev);
176 bch2_bkey_mark_dev_cached(bkey_i_to_s(insert), p.ptr.dev);
183 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
184 if (bch2_bkey_has_device(bkey_i_to_s_c(insert), p.ptr.dev)) {
186 * raced with another move op? extent already
187 * has a pointer to the device we just wrote
193 bch2_extent_ptr_decoded_append(insert, &p);
200 bch2_bkey_narrow_crcs(insert, (struct bch_extent_crc_unpacked) { 0 });
201 bch2_extent_normalize(c, bkey_i_to_s(insert));
203 ret = bch2_sum_sector_overwrites(trans, &iter, insert,
204 &should_check_enospc,
206 &disk_sectors_delta);
210 if (disk_sectors_delta > (s64) op->res.sectors) {
211 ret = bch2_disk_reservation_add(c, &op->res,
212 disk_sectors_delta - op->res.sectors,
214 ? BCH_DISK_RESERVATION_NOFAIL : 0);
219 next_pos = insert->k.p;
221 ret = insert_snapshot_whiteouts(trans, m->btree_id,
222 k.k->p, insert->k.p) ?:
223 bch2_trans_update(trans, &iter, insert,
224 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
225 bch2_trans_commit(trans, &op->res,
228 m->data_opts.btree_insert_flags);
230 bch2_btree_iter_set_pos(&iter, next_pos);
232 this_cpu_add(c->counters[BCH_COUNTER_move_extent_finish], new->k.size);
233 trace_move_extent_finish(&new->k);
236 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
241 while (bkey_ge(iter.pos, bch2_keylist_front(keys)->k.p)) {
242 bch2_keylist_pop_front(keys);
243 if (bch2_keylist_empty(keys))
249 BUG_ON(k.k->p.offset <= iter.pos.offset);
250 atomic64_inc(&m->ctxt->stats->keys_raced);
251 atomic64_add(k.k->p.offset - iter.pos.offset,
252 &m->ctxt->stats->sectors_raced);
255 this_cpu_add(c->counters[BCH_COUNTER_move_extent_race], new->k.size);
256 trace_move_extent_race(&new->k);
258 bch2_btree_iter_advance(&iter);
262 bch2_trans_iter_exit(trans, &iter);
263 bch2_bkey_buf_exit(&_insert, c);
264 bch2_bkey_buf_exit(&_new, c);
265 BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart));
269 int bch2_data_update_index_update(struct bch_write_op *op)
271 struct bch_fs *c = op->c;
272 struct btree_trans trans;
275 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024);
276 ret = __bch2_data_update_index_update(&trans, op);
277 bch2_trans_exit(&trans);
282 void bch2_data_update_read_done(struct data_update *m,
283 struct bch_extent_crc_unpacked crc)
285 /* write bio must own pages: */
286 BUG_ON(!m->op.wbio.bio.bi_vcnt);
289 m->op.wbio.bio.bi_iter.bi_size = crc.compressed_size << 9;
291 closure_call(&m->op.cl, bch2_write, NULL, NULL);
294 void bch2_data_update_exit(struct data_update *update)
296 struct bch_fs *c = update->op.c;
297 struct bkey_ptrs_c ptrs =
298 bch2_bkey_ptrs_c(bkey_i_to_s_c(update->k.k));
299 const struct bch_extent_ptr *ptr;
301 bkey_for_each_ptr(ptrs, ptr)
302 bch2_bucket_nocow_unlock(&c->nocow_locks,
303 PTR_BUCKET_POS(c, ptr), 0);
305 bch2_bkey_buf_exit(&update->k, c);
306 bch2_disk_reservation_put(c, &update->op.res);
307 bch2_bio_free_pages_pool(c, &update->op.wbio.bio);
310 void bch2_update_unwritten_extent(struct btree_trans *trans,
311 struct data_update *update)
313 struct bch_fs *c = update->op.c;
314 struct bio *bio = &update->op.wbio.bio;
315 struct bkey_i_extent *e;
316 struct write_point *wp;
317 struct bch_extent_ptr *ptr;
319 struct btree_iter iter;
323 closure_init_stack(&cl);
324 bch2_keylist_init(&update->op.insert_keys, update->op.inline_keys);
326 while (bio_sectors(bio)) {
327 unsigned sectors = bio_sectors(bio);
329 bch2_trans_iter_init(trans, &iter, update->btree_id, update->op.pos,
331 ret = lockrestart_do(trans, ({
332 k = bch2_btree_iter_peek_slot(&iter);
335 bch2_trans_iter_exit(trans, &iter);
337 if (ret || !bch2_extents_match(k, bkey_i_to_s_c(update->k.k)))
340 e = bkey_extent_init(update->op.insert_keys.top);
341 e->k.p = update->op.pos;
343 ret = bch2_alloc_sectors_start_trans(trans,
346 update->op.write_point,
347 &update->op.devs_have,
348 update->op.nr_replicas,
349 update->op.nr_replicas,
350 update->op.alloc_reserve,
352 if (ret == -EAGAIN) {
353 bch2_trans_unlock(trans);
361 sectors = min(sectors, wp->sectors_free);
363 bch2_key_resize(&e->k, sectors);
365 bch2_open_bucket_get(c, wp, &update->op.open_buckets);
366 bch2_alloc_sectors_append_ptrs(c, wp, &e->k_i, sectors, false);
367 bch2_alloc_sectors_done(c, wp);
369 bio_advance(bio, sectors << 9);
370 update->op.pos.offset += sectors;
372 extent_for_each_ptr(extent_i_to_s(e), ptr)
373 ptr->unwritten = true;
374 bch2_keylist_push(&update->op.insert_keys);
376 ret = __bch2_data_update_index_update(trans, &update->op);
378 bch2_open_buckets_put(c, &update->op.open_buckets);
384 if ((atomic_read(&cl.remaining) & CLOSURE_REMAINING_MASK) != 1) {
385 bch2_trans_unlock(trans);
390 int bch2_data_update_init(struct bch_fs *c, struct data_update *m,
391 struct write_point_specifier wp,
392 struct bch_io_opts io_opts,
393 struct data_update_opts data_opts,
394 enum btree_id btree_id,
397 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
398 const union bch_extent_entry *entry;
399 struct extent_ptr_decoded p;
400 unsigned i, reserve_sectors = k.k->size * data_opts.extra_replicas;
403 bch2_bkey_buf_init(&m->k);
404 bch2_bkey_buf_reassemble(&m->k, c, k);
405 m->btree_id = btree_id;
406 m->data_opts = data_opts;
408 bch2_write_op_init(&m->op, c, io_opts);
409 m->op.pos = bkey_start_pos(k.k);
410 m->op.version = k.k->version;
411 m->op.target = data_opts.target;
412 m->op.write_point = wp;
413 m->op.flags |= BCH_WRITE_PAGES_STABLE|
414 BCH_WRITE_PAGES_OWNED|
415 BCH_WRITE_DATA_ENCODED|
417 m->data_opts.write_flags;
418 m->op.compression_type =
419 bch2_compression_opt_to_type[io_opts.background_compression ?:
420 io_opts.compression];
421 if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE)
422 m->op.alloc_reserve = RESERVE_movinggc;
425 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
426 if (((1U << i) & m->data_opts.rewrite_ptrs) &&
430 if (!((1U << i) & m->data_opts.rewrite_ptrs))
431 bch2_dev_list_add_dev(&m->op.devs_have, p.ptr.dev);
433 if (((1U << i) & m->data_opts.rewrite_ptrs) &&
434 crc_is_compressed(p.crc))
435 reserve_sectors += k.k->size;
438 * op->csum_type is normally initialized from the fs/file's
439 * current options - but if an extent is encrypted, we require
440 * that it stays encrypted:
442 if (bch2_csum_type_is_encryption(p.crc.csum_type)) {
443 m->op.nonce = p.crc.nonce + p.crc.offset;
444 m->op.csum_type = p.crc.csum_type;
447 if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
448 m->op.incompressible = true;
452 bch2_bucket_nocow_lock(&c->nocow_locks,
453 PTR_BUCKET_POS(c, &p.ptr), 0);
456 if (reserve_sectors) {
457 ret = bch2_disk_reservation_add(c, &m->op.res, reserve_sectors,
458 m->data_opts.extra_replicas
460 : BCH_DISK_RESERVATION_NOFAIL);
465 m->op.nr_replicas = m->op.nr_replicas_required =
466 hweight32(m->data_opts.rewrite_ptrs) + m->data_opts.extra_replicas;
468 BUG_ON(!m->op.nr_replicas);
470 /* Special handling required: */
471 if (bkey_extent_is_unwritten(k))
472 return -BCH_ERR_unwritten_extent_update;
476 void bch2_data_update_opts_normalize(struct bkey_s_c k, struct data_update_opts *opts)
478 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
479 const struct bch_extent_ptr *ptr;
482 bkey_for_each_ptr(ptrs, ptr) {
483 if ((opts->rewrite_ptrs & (1U << i)) && ptr->cached) {
484 opts->kill_ptrs |= 1U << i;
485 opts->rewrite_ptrs ^= 1U << i;