1 // SPDX-License-Identifier: GPL-2.0
3 * io_misc.c - fallocate, fpunch, truncate:
7 #include "alloc_foreground.h"
9 #include "btree_update.h"
14 #include "extent_update.h"
18 #include "logged_ops.h"
19 #include "rebalance.h"
20 #include "subvolume.h"
22 /* Overwrites whatever was present with zeroes: */
23 int bch2_extent_fallocate(struct btree_trans *trans,
25 struct btree_iter *iter,
27 struct bch_io_opts opts,
29 struct write_point_specifier write_point)
31 struct bch_fs *c = trans->c;
32 struct disk_reservation disk_res = { 0 };
34 struct open_buckets open_buckets = { 0 };
36 struct bkey_buf old, new;
37 unsigned sectors_allocated = 0, new_replicas;
38 bool unwritten = opts.nocow &&
39 c->sb.version >= bcachefs_metadata_version_unwritten_extents;
42 bch2_bkey_buf_init(&old);
43 bch2_bkey_buf_init(&new);
44 closure_init_stack(&cl);
46 k = bch2_btree_iter_peek_slot(iter);
51 sectors = min_t(u64, sectors, k.k->p.offset - iter->pos.offset);
52 new_replicas = max(0, (int) opts.data_replicas -
53 (int) bch2_bkey_nr_ptrs_fully_allocated(k));
56 * Get a disk reservation before (in the nocow case) calling
59 ret = bch2_disk_reservation_get(c, &disk_res, sectors, new_replicas, 0);
63 bch2_bkey_buf_reassemble(&old, c, k);
66 struct bkey_i_reservation *reservation;
68 bch2_bkey_buf_realloc(&new, c, sizeof(*reservation) / sizeof(u64));
69 reservation = bkey_reservation_init(new.k);
70 reservation->k.p = iter->pos;
71 bch2_key_resize(&reservation->k, sectors);
72 reservation->v.nr_replicas = opts.data_replicas;
74 struct bkey_i_extent *e;
75 struct bch_devs_list devs_have;
76 struct write_point *wp;
80 bch2_bkey_buf_realloc(&new, c, BKEY_EXTENT_U64s_MAX);
82 e = bkey_extent_init(new.k);
85 ret = bch2_alloc_sectors_start_trans(trans,
86 opts.foreground_target,
92 BCH_WATERMARK_normal, 0, &cl, &wp);
93 if (bch2_err_matches(ret, BCH_ERR_operation_blocked))
94 ret = -BCH_ERR_transaction_restart_nested;
98 sectors = min_t(u64, sectors, wp->sectors_free);
99 sectors_allocated = sectors;
101 bch2_key_resize(&e->k, sectors);
103 bch2_open_bucket_get(c, wp, &open_buckets);
104 bch2_alloc_sectors_append_ptrs(c, wp, &e->k_i, sectors, false);
105 bch2_alloc_sectors_done(c, wp);
107 extent_for_each_ptr(extent_i_to_s(e), ptr)
108 ptr->unwritten = true;
111 ret = bch2_extent_update(trans, inum, iter, new.k, &disk_res,
112 0, i_sectors_delta, true);
114 if (!ret && sectors_allocated)
115 bch2_increment_clock(c, sectors_allocated, WRITE);
116 if (should_print_err(ret))
117 bch_err_inum_offset_ratelimited(c,
119 iter->pos.offset << 9,
120 "%s(): error: %s", __func__, bch2_err_str(ret));
122 bch2_open_buckets_put(c, &open_buckets);
123 bch2_disk_reservation_put(c, &disk_res);
124 bch2_bkey_buf_exit(&new, c);
125 bch2_bkey_buf_exit(&old, c);
127 if (closure_nr_remaining(&cl) != 1) {
128 bch2_trans_unlock(trans);
136 * Returns -BCH_ERR_transacton_restart if we had to drop locks:
138 int bch2_fpunch_at(struct btree_trans *trans, struct btree_iter *iter,
139 subvol_inum inum, u64 end,
140 s64 *i_sectors_delta)
142 struct bch_fs *c = trans->c;
143 unsigned max_sectors = KEY_SIZE_MAX & (~0 << c->block_bits);
144 struct bpos end_pos = POS(inum.inum, end);
146 int ret = 0, ret2 = 0;
150 bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
151 struct disk_reservation disk_res =
152 bch2_disk_reservation_init(c, 0);
153 struct bkey_i delete;
158 bch2_trans_begin(trans);
160 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
164 bch2_btree_iter_set_snapshot(iter, snapshot);
167 * peek_upto() doesn't have ideal semantics for extents:
169 k = bch2_btree_iter_peek_upto(iter, end_pos);
177 bkey_init(&delete.k);
178 delete.k.p = iter->pos;
180 /* create the biggest key we can */
181 bch2_key_resize(&delete.k, max_sectors);
182 bch2_cut_back(end_pos, &delete);
184 ret = bch2_extent_update(trans, inum, iter, &delete,
185 &disk_res, 0, i_sectors_delta, false);
186 bch2_disk_reservation_put(c, &disk_res);
192 int bch2_fpunch(struct bch_fs *c, subvol_inum inum, u64 start, u64 end,
193 s64 *i_sectors_delta)
195 struct btree_trans *trans = bch2_trans_get(c);
196 struct btree_iter iter;
199 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
200 POS(inum.inum, start),
203 ret = bch2_fpunch_at(trans, &iter, inum, end, i_sectors_delta);
205 bch2_trans_iter_exit(trans, &iter);
206 bch2_trans_put(trans);
208 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
216 void bch2_logged_op_truncate_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
218 struct bkey_s_c_logged_op_truncate op = bkey_s_c_to_logged_op_truncate(k);
220 prt_printf(out, "subvol=%u", le32_to_cpu(op.v->subvol));
221 prt_printf(out, " inum=%llu", le64_to_cpu(op.v->inum));
222 prt_printf(out, " new_i_size=%llu", le64_to_cpu(op.v->new_i_size));
225 static int truncate_set_isize(struct btree_trans *trans,
229 struct btree_iter iter = { NULL };
230 struct bch_inode_unpacked inode_u;
233 ret = bch2_inode_peek(trans, &iter, &inode_u, inum, BTREE_ITER_INTENT) ?:
234 (inode_u.bi_size = new_i_size, 0) ?:
235 bch2_inode_write(trans, &iter, &inode_u);
237 bch2_trans_iter_exit(trans, &iter);
241 static int __bch2_resume_logged_op_truncate(struct btree_trans *trans,
243 u64 *i_sectors_delta)
245 struct bch_fs *c = trans->c;
246 struct btree_iter fpunch_iter;
247 struct bkey_i_logged_op_truncate *op = bkey_i_to_logged_op_truncate(op_k);
248 subvol_inum inum = { le32_to_cpu(op->v.subvol), le64_to_cpu(op->v.inum) };
249 u64 new_i_size = le64_to_cpu(op->v.new_i_size);
252 ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
253 truncate_set_isize(trans, inum, new_i_size));
257 bch2_trans_iter_init(trans, &fpunch_iter, BTREE_ID_extents,
258 POS(inum.inum, round_up(new_i_size, block_bytes(c)) >> 9),
260 ret = bch2_fpunch_at(trans, &fpunch_iter, inum, U64_MAX, i_sectors_delta);
261 bch2_trans_iter_exit(trans, &fpunch_iter);
263 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
266 bch2_logged_op_finish(trans, op_k);
270 int bch2_resume_logged_op_truncate(struct btree_trans *trans, struct bkey_i *op_k)
272 return __bch2_resume_logged_op_truncate(trans, op_k, NULL);
275 int bch2_truncate(struct bch_fs *c, subvol_inum inum, u64 new_i_size, u64 *i_sectors_delta)
277 struct bkey_i_logged_op_truncate op;
279 bkey_logged_op_truncate_init(&op.k_i);
280 op.v.subvol = cpu_to_le32(inum.subvol);
281 op.v.inum = cpu_to_le64(inum.inum);
282 op.v.new_i_size = cpu_to_le64(new_i_size);
285 * Logged ops aren't atomic w.r.t. snapshot creation: creating a
286 * snapshot while they're in progress, then crashing, will result in the
287 * resume only proceeding in one of the snapshots
289 down_read(&c->snapshot_create_lock);
290 int ret = bch2_trans_run(c,
291 bch2_logged_op_start(trans, &op.k_i) ?:
292 __bch2_resume_logged_op_truncate(trans, &op.k_i, i_sectors_delta));
293 up_read(&c->snapshot_create_lock);
298 /* finsert/fcollapse: */
300 void bch2_logged_op_finsert_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
302 struct bkey_s_c_logged_op_finsert op = bkey_s_c_to_logged_op_finsert(k);
304 prt_printf(out, "subvol=%u", le32_to_cpu(op.v->subvol));
305 prt_printf(out, " inum=%llu", le64_to_cpu(op.v->inum));
306 prt_printf(out, " dst_offset=%lli", le64_to_cpu(op.v->dst_offset));
307 prt_printf(out, " src_offset=%llu", le64_to_cpu(op.v->src_offset));
310 static int adjust_i_size(struct btree_trans *trans, subvol_inum inum, u64 offset, s64 len)
312 struct btree_iter iter;
313 struct bch_inode_unpacked inode_u;
319 ret = bch2_inode_peek(trans, &iter, &inode_u, inum, BTREE_ITER_INTENT);
324 if (MAX_LFS_FILESIZE - inode_u.bi_size < len) {
329 if (offset >= inode_u.bi_size) {
335 inode_u.bi_size += len;
336 inode_u.bi_mtime = inode_u.bi_ctime = bch2_current_time(trans->c);
338 ret = bch2_inode_write(trans, &iter, &inode_u);
340 bch2_trans_iter_exit(trans, &iter);
344 static int __bch2_resume_logged_op_finsert(struct btree_trans *trans,
346 u64 *i_sectors_delta)
348 struct bch_fs *c = trans->c;
349 struct btree_iter iter;
350 struct bkey_i_logged_op_finsert *op = bkey_i_to_logged_op_finsert(op_k);
351 subvol_inum inum = { le32_to_cpu(op->v.subvol), le64_to_cpu(op->v.inum) };
352 struct bch_io_opts opts;
353 u64 dst_offset = le64_to_cpu(op->v.dst_offset);
354 u64 src_offset = le64_to_cpu(op->v.src_offset);
355 s64 shift = dst_offset - src_offset;
356 u64 len = abs(shift);
357 u64 pos = le64_to_cpu(op->v.pos);
358 bool insert = shift > 0;
361 ret = bch2_inum_opts_get(trans, inum, &opts);
365 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
369 switch (op->v.state) {
370 case LOGGED_OP_FINSERT_start:
371 op->v.state = LOGGED_OP_FINSERT_shift_extents;
374 ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
375 adjust_i_size(trans, inum, src_offset, len) ?:
376 bch2_logged_op_update(trans, &op->k_i));
380 bch2_btree_iter_set_pos(&iter, POS(inum.inum, src_offset));
382 ret = bch2_fpunch_at(trans, &iter, inum, src_offset + len, i_sectors_delta);
383 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
386 ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
387 bch2_logged_op_update(trans, &op->k_i));
391 case LOGGED_OP_FINSERT_shift_extents:
393 struct disk_reservation disk_res =
394 bch2_disk_reservation_init(c, 0);
395 struct bkey_i delete, *copy;
397 struct bpos src_pos = POS(inum.inum, src_offset);
400 bch2_trans_begin(trans);
402 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
406 bch2_btree_iter_set_snapshot(&iter, snapshot);
407 bch2_btree_iter_set_pos(&iter, SPOS(inum.inum, pos, snapshot));
410 ? bch2_btree_iter_peek_prev(&iter)
411 : bch2_btree_iter_peek_upto(&iter, POS(inum.inum, U64_MAX));
412 if ((ret = bkey_err(k)))
416 k.k->p.inode != inum.inum ||
417 bkey_le(k.k->p, POS(inum.inum, src_offset)))
420 copy = bch2_bkey_make_mut_noupdate(trans, k);
421 if ((ret = PTR_ERR_OR_ZERO(copy)))
425 bkey_lt(bkey_start_pos(k.k), src_pos)) {
426 bch2_cut_front(src_pos, copy);
428 /* Splitting compressed extent? */
429 bch2_disk_reservation_add(c, &disk_res,
431 bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(copy)),
432 BCH_DISK_RESERVATION_NOFAIL);
435 bkey_init(&delete.k);
436 delete.k.p = copy->k.p;
437 delete.k.p.snapshot = snapshot;
438 delete.k.size = copy->k.size;
440 copy->k.p.offset += shift;
441 copy->k.p.snapshot = snapshot;
443 op->v.pos = cpu_to_le64(insert ? bkey_start_offset(&delete.k) : delete.k.p.offset);
445 ret = bch2_bkey_set_needs_rebalance(c, copy, &opts) ?:
446 bch2_btree_insert_trans(trans, BTREE_ID_extents, &delete, 0) ?:
447 bch2_btree_insert_trans(trans, BTREE_ID_extents, copy, 0) ?:
448 bch2_logged_op_update(trans, &op->k_i) ?:
449 bch2_trans_commit(trans, &disk_res, NULL, BCH_TRANS_COMMIT_no_enospc);
451 bch2_disk_reservation_put(c, &disk_res);
453 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
458 pos = le64_to_cpu(op->v.pos);
461 op->v.state = LOGGED_OP_FINSERT_finish;
464 ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
465 adjust_i_size(trans, inum, src_offset, shift) ?:
466 bch2_logged_op_update(trans, &op->k_i));
468 /* We need an inode update to update bi_journal_seq for fsync: */
469 ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
470 adjust_i_size(trans, inum, 0, 0) ?:
471 bch2_logged_op_update(trans, &op->k_i));
475 case LOGGED_OP_FINSERT_finish:
479 bch2_logged_op_finish(trans, op_k);
480 bch2_trans_iter_exit(trans, &iter);
484 int bch2_resume_logged_op_finsert(struct btree_trans *trans, struct bkey_i *op_k)
486 return __bch2_resume_logged_op_finsert(trans, op_k, NULL);
489 int bch2_fcollapse_finsert(struct bch_fs *c, subvol_inum inum,
490 u64 offset, u64 len, bool insert,
491 s64 *i_sectors_delta)
493 struct bkey_i_logged_op_finsert op;
494 s64 shift = insert ? len : -len;
496 bkey_logged_op_finsert_init(&op.k_i);
497 op.v.subvol = cpu_to_le32(inum.subvol);
498 op.v.inum = cpu_to_le64(inum.inum);
499 op.v.dst_offset = cpu_to_le64(offset + shift);
500 op.v.src_offset = cpu_to_le64(offset);
501 op.v.pos = cpu_to_le64(insert ? U64_MAX : offset);
504 * Logged ops aren't atomic w.r.t. snapshot creation: creating a
505 * snapshot while they're in progress, then crashing, will result in the
506 * resume only proceeding in one of the snapshots
508 down_read(&c->snapshot_create_lock);
509 int ret = bch2_trans_run(c,
510 bch2_logged_op_start(trans, &op.k_i) ?:
511 __bch2_resume_logged_op_finsert(trans, &op.k_i, i_sectors_delta));
512 up_read(&c->snapshot_create_lock);