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_cmp(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_cmp(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 bch_write_op *op)
102 struct bch_fs *c = op->c;
103 struct btree_trans trans;
104 struct btree_iter iter;
105 struct data_update *m =
106 container_of(op, struct data_update, op);
107 struct open_bucket *ec_ob = ec_open_bucket(c, &op->open_buckets);
108 struct keylist *keys = &op->insert_keys;
109 struct bkey_buf _new, _insert;
112 bch2_bkey_buf_init(&_new);
113 bch2_bkey_buf_init(&_insert);
114 bch2_bkey_buf_realloc(&_insert, c, U8_MAX);
116 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024);
118 bch2_trans_iter_init(&trans, &iter, m->btree_id,
119 bkey_start_pos(&bch2_keylist_front(keys)->k),
120 BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
124 struct bkey_s_c old = bkey_i_to_s_c(m->k.k);
125 struct bkey_i *insert;
126 struct bkey_i_extent *new;
127 const union bch_extent_entry *entry;
128 struct extent_ptr_decoded p;
129 struct bpos next_pos;
130 bool did_work = false;
131 bool should_check_enospc;
132 s64 i_sectors_delta = 0, disk_sectors_delta = 0;
135 bch2_trans_begin(&trans);
137 k = bch2_btree_iter_peek_slot(&iter);
142 new = bkey_i_to_extent(bch2_keylist_front(keys));
144 if (!bch2_extents_match(k, old))
147 bkey_reassemble(_insert.k, k);
150 bch2_bkey_buf_copy(&_new, c, bch2_keylist_front(keys));
151 new = bkey_i_to_extent(_new.k);
152 bch2_cut_front(iter.pos, &new->k_i);
154 bch2_cut_front(iter.pos, insert);
155 bch2_cut_back(new->k.p, insert);
156 bch2_cut_back(insert->k.p, &new->k_i);
159 * @old: extent that we read from
160 * @insert: key that we're going to update, initialized from
161 * extent currently in btree - same as @old unless we raced with
163 * @new: extent with new pointers that we'll be adding to @insert
165 * Fist, drop rewrite_ptrs from @new:
168 bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs_c(old), p, entry) {
169 if (((1U << i) & m->data_opts.rewrite_ptrs) &&
170 bch2_extent_has_ptr(old, p, bkey_i_to_s_c(insert))) {
172 * If we're going to be adding a pointer to the
173 * same device, we have to drop the old one -
174 * otherwise, we can just mark it cached:
176 if (bch2_bkey_has_device(bkey_i_to_s_c(&new->k_i), p.ptr.dev))
177 bch2_bkey_drop_device_noerror(bkey_i_to_s(insert), p.ptr.dev);
179 bch2_bkey_mark_dev_cached(bkey_i_to_s(insert), p.ptr.dev);
186 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
187 if (bch2_bkey_has_device(bkey_i_to_s_c(insert), p.ptr.dev)) {
189 * raced with another move op? extent already
190 * has a pointer to the device we just wrote
196 bch2_extent_ptr_decoded_append(insert, &p);
203 bch2_bkey_narrow_crcs(insert, (struct bch_extent_crc_unpacked) { 0 });
204 bch2_extent_normalize(c, bkey_i_to_s(insert));
206 ret = bch2_sum_sector_overwrites(&trans, &iter, insert,
207 &should_check_enospc,
209 &disk_sectors_delta);
213 if (disk_sectors_delta > (s64) op->res.sectors) {
214 ret = bch2_disk_reservation_add(c, &op->res,
215 disk_sectors_delta - op->res.sectors,
217 ? BCH_DISK_RESERVATION_NOFAIL : 0);
222 next_pos = insert->k.p;
224 ret = insert_snapshot_whiteouts(&trans, m->btree_id,
225 k.k->p, insert->k.p) ?:
226 bch2_trans_update(&trans, &iter, insert,
227 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
228 bch2_trans_commit(&trans, &op->res,
231 m->data_opts.btree_insert_flags);
233 bch2_btree_iter_set_pos(&iter, next_pos);
234 atomic_long_inc(&c->extent_migrate_done);
236 bch2_ob_add_backpointer(c, ec_ob, &insert->k);
239 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
244 while (bkey_cmp(iter.pos, bch2_keylist_front(keys)->k.p) >= 0) {
245 bch2_keylist_pop_front(keys);
246 if (bch2_keylist_empty(keys))
251 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
252 struct printbuf buf = PRINTBUF;
254 bch2_bkey_val_to_text(&buf, c, old);
255 bch_info(c, "no match for %s", buf.buf);
260 BUG_ON(k.k->p.offset <= iter.pos.offset);
261 atomic64_inc(&m->ctxt->stats->keys_raced);
262 atomic64_add(k.k->p.offset - iter.pos.offset,
263 &m->ctxt->stats->sectors_raced);
265 atomic_long_inc(&c->extent_migrate_raced);
266 trace_move_race(&new->k);
267 bch2_btree_iter_advance(&iter);
271 bch2_trans_iter_exit(&trans, &iter);
272 bch2_trans_exit(&trans);
273 bch2_bkey_buf_exit(&_insert, c);
274 bch2_bkey_buf_exit(&_new, c);
275 BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart));
279 void bch2_data_update_read_done(struct data_update *m,
280 struct bch_extent_crc_unpacked crc,
283 /* write bio must own pages: */
284 BUG_ON(!m->op.wbio.bio.bi_vcnt);
287 m->op.wbio.bio.bi_iter.bi_size = crc.compressed_size << 9;
289 closure_call(&m->op.cl, bch2_write, NULL, cl);
292 void bch2_data_update_exit(struct data_update *update)
294 struct bch_fs *c = update->op.c;
296 bch2_bkey_buf_exit(&update->k, c);
297 bch2_disk_reservation_put(c, &update->op.res);
298 bch2_bio_free_pages_pool(c, &update->op.wbio.bio);
301 int bch2_data_update_init(struct bch_fs *c, struct data_update *m,
302 struct write_point_specifier wp,
303 struct bch_io_opts io_opts,
304 struct data_update_opts data_opts,
305 enum btree_id btree_id,
308 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
309 const union bch_extent_entry *entry;
310 struct extent_ptr_decoded p;
311 unsigned i, reserve_sectors = k.k->size * data_opts.extra_replicas;
314 bch2_bkey_buf_init(&m->k);
315 bch2_bkey_buf_reassemble(&m->k, c, k);
316 m->btree_id = btree_id;
317 m->data_opts = data_opts;
319 bch2_write_op_init(&m->op, c, io_opts);
320 m->op.pos = bkey_start_pos(k.k);
321 m->op.version = k.k->version;
322 m->op.target = data_opts.target,
323 m->op.write_point = wp;
324 m->op.flags |= BCH_WRITE_PAGES_STABLE|
325 BCH_WRITE_PAGES_OWNED|
326 BCH_WRITE_DATA_ENCODED|
327 BCH_WRITE_FROM_INTERNAL|
328 m->data_opts.write_flags;
329 m->op.compression_type =
330 bch2_compression_opt_to_type[io_opts.background_compression ?:
331 io_opts.compression];
332 if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE)
333 m->op.alloc_reserve = RESERVE_movinggc;
334 m->op.index_update_fn = bch2_data_update_index_update;
337 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
339 m->data_opts.rewrite_ptrs &= ~(1U << i);
341 if (!((1U << i) & m->data_opts.rewrite_ptrs))
342 bch2_dev_list_add_dev(&m->op.devs_have, p.ptr.dev);
344 if (((1U << i) & m->data_opts.rewrite_ptrs) &&
345 crc_is_compressed(p.crc))
346 reserve_sectors += k.k->size;
349 * op->csum_type is normally initialized from the fs/file's
350 * current options - but if an extent is encrypted, we require
351 * that it stays encrypted:
353 if (bch2_csum_type_is_encryption(p.crc.csum_type)) {
354 m->op.nonce = p.crc.nonce + p.crc.offset;
355 m->op.csum_type = p.crc.csum_type;
358 if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
359 m->op.incompressible = true;
364 if (reserve_sectors) {
365 ret = bch2_disk_reservation_add(c, &m->op.res, reserve_sectors,
366 m->data_opts.extra_replicas
368 : BCH_DISK_RESERVATION_NOFAIL);
373 m->op.nr_replicas = m->op.nr_replicas_required =
374 hweight32(m->data_opts.rewrite_ptrs) + m->data_opts.extra_replicas;