1 // SPDX-License-Identifier: GPL-2.0 2 #include "bcachefs.h" 3 #include "bkey_buf.h" 4 #include "btree_update.h" 5 #include "buckets.h" 6 #include "enumerated_ref.h" 7 #include "error.h" 8 #include "extents.h" 9 #include "inode.h" 10 #include "io_misc.h" 11 #include "io_write.h" 12 #include "rebalance.h" 13 #include "reflink.h" 14 #include "subvolume.h" 15 #include "super-io.h" 16 17 #include <linux/sched/signal.h> 18 19 static inline bool bkey_extent_is_reflink_data(const struct bkey *k) 20 { 21 switch (k->type) { 22 case KEY_TYPE_reflink_v: 23 case KEY_TYPE_indirect_inline_data: 24 return true; 25 default: 26 return false; 27 } 28 } 29 30 static inline unsigned bkey_type_to_indirect(const struct bkey *k) 31 { 32 switch (k->type) { 33 case KEY_TYPE_extent: 34 return KEY_TYPE_reflink_v; 35 case KEY_TYPE_inline_data: 36 return KEY_TYPE_indirect_inline_data; 37 default: 38 return 0; 39 } 40 } 41 42 /* reflink pointers */ 43 44 int bch2_reflink_p_validate(struct bch_fs *c, struct bkey_s_c k, 45 struct bkey_validate_context from) 46 { 47 struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k); 48 int ret = 0; 49 50 bkey_fsck_err_on(REFLINK_P_IDX(p.v) < le32_to_cpu(p.v->front_pad), 51 c, reflink_p_front_pad_bad, 52 "idx < front_pad (%llu < %u)", 53 REFLINK_P_IDX(p.v), le32_to_cpu(p.v->front_pad)); 54 fsck_err: 55 return ret; 56 } 57 58 void bch2_reflink_p_to_text(struct printbuf *out, struct bch_fs *c, 59 struct bkey_s_c k) 60 { 61 struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k); 62 63 prt_printf(out, "idx %llu front_pad %u back_pad %u", 64 REFLINK_P_IDX(p.v), 65 le32_to_cpu(p.v->front_pad), 66 le32_to_cpu(p.v->back_pad)); 67 } 68 69 bool bch2_reflink_p_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r) 70 { 71 struct bkey_s_reflink_p l = bkey_s_to_reflink_p(_l); 72 struct bkey_s_c_reflink_p r = bkey_s_c_to_reflink_p(_r); 73 74 /* 75 * Disabled for now, the triggers code needs to be reworked for merging 76 * of reflink pointers to work: 77 */ 78 return false; 79 80 if (REFLINK_P_IDX(l.v) + l.k->size != REFLINK_P_IDX(r.v)) 81 return false; 82 83 if (REFLINK_P_ERROR(l.v) != REFLINK_P_ERROR(r.v)) 84 return false; 85 86 bch2_key_resize(l.k, l.k->size + r.k->size); 87 return true; 88 } 89 90 /* indirect extents */ 91 92 int bch2_reflink_v_validate(struct bch_fs *c, struct bkey_s_c k, 93 struct bkey_validate_context from) 94 { 95 int ret = 0; 96 97 bkey_fsck_err_on(bkey_gt(k.k->p, POS(0, REFLINK_P_IDX_MAX)), 98 c, reflink_v_pos_bad, 99 "indirect extent above maximum position 0:%llu", 100 REFLINK_P_IDX_MAX); 101 102 ret = bch2_bkey_ptrs_validate(c, k, from); 103 fsck_err: 104 return ret; 105 } 106 107 void bch2_reflink_v_to_text(struct printbuf *out, struct bch_fs *c, 108 struct bkey_s_c k) 109 { 110 struct bkey_s_c_reflink_v r = bkey_s_c_to_reflink_v(k); 111 112 prt_printf(out, "refcount: %llu ", le64_to_cpu(r.v->refcount)); 113 114 bch2_bkey_ptrs_to_text(out, c, k); 115 } 116 117 #if 0 118 Currently disabled, needs to be debugged: 119 120 bool bch2_reflink_v_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r) 121 { 122 struct bkey_s_reflink_v l = bkey_s_to_reflink_v(_l); 123 struct bkey_s_c_reflink_v r = bkey_s_c_to_reflink_v(_r); 124 125 return l.v->refcount == r.v->refcount && bch2_extent_merge(c, _l, _r); 126 } 127 #endif 128 129 /* indirect inline data */ 130 131 int bch2_indirect_inline_data_validate(struct bch_fs *c, struct bkey_s_c k, 132 struct bkey_validate_context from) 133 { 134 return 0; 135 } 136 137 void bch2_indirect_inline_data_to_text(struct printbuf *out, 138 struct bch_fs *c, struct bkey_s_c k) 139 { 140 struct bkey_s_c_indirect_inline_data d = bkey_s_c_to_indirect_inline_data(k); 141 unsigned datalen = bkey_inline_data_bytes(k.k); 142 143 prt_printf(out, "refcount %llu datalen %u: %*phN", 144 le64_to_cpu(d.v->refcount), datalen, 145 min(datalen, 32U), d.v->data); 146 } 147 148 /* lookup */ 149 150 static int bch2_indirect_extent_not_missing(struct btree_trans *trans, struct bkey_s_c_reflink_p p, 151 bool should_commit) 152 { 153 struct bkey_i_reflink_p *new = bch2_bkey_make_mut_noupdate_typed(trans, p.s_c, reflink_p); 154 int ret = PTR_ERR_OR_ZERO(new); 155 if (ret) 156 return ret; 157 158 SET_REFLINK_P_ERROR(&new->v, false); 159 ret = bch2_btree_insert_trans(trans, BTREE_ID_extents, &new->k_i, BTREE_TRIGGER_norun); 160 if (ret) 161 return ret; 162 163 if (!should_commit) 164 return 0; 165 166 return bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?: 167 -BCH_ERR_transaction_restart_nested; 168 } 169 170 static int bch2_indirect_extent_missing_error(struct btree_trans *trans, 171 struct bkey_s_c_reflink_p p, 172 u64 missing_start, u64 missing_end, 173 bool should_commit) 174 { 175 if (REFLINK_P_ERROR(p.v)) 176 return 0; 177 178 struct bch_fs *c = trans->c; 179 u64 live_start = REFLINK_P_IDX(p.v); 180 u64 live_end = REFLINK_P_IDX(p.v) + p.k->size; 181 u64 refd_start = live_start - le32_to_cpu(p.v->front_pad); 182 u64 refd_end = live_end + le32_to_cpu(p.v->back_pad); 183 struct printbuf buf = PRINTBUF; 184 int ret = 0; 185 186 BUG_ON(missing_start < refd_start); 187 BUG_ON(missing_end > refd_end); 188 189 struct bpos missing_pos = bkey_start_pos(p.k); 190 missing_pos.offset += missing_start - live_start; 191 192 prt_printf(&buf, "pointer to missing indirect extent in "); 193 ret = bch2_inum_snap_offset_err_msg_trans(trans, &buf, missing_pos); 194 if (ret) 195 goto err; 196 197 prt_printf(&buf, "-%llu\n", (missing_pos.offset + (missing_end - missing_start)) << 9); 198 bch2_bkey_val_to_text(&buf, c, p.s_c); 199 200 prt_printf(&buf, "\nmissing reflink btree range %llu-%llu", 201 missing_start, missing_end); 202 203 if (fsck_err(trans, reflink_p_to_missing_reflink_v, "%s", buf.buf)) { 204 struct bkey_i_reflink_p *new = bch2_bkey_make_mut_noupdate_typed(trans, p.s_c, reflink_p); 205 ret = PTR_ERR_OR_ZERO(new); 206 if (ret) 207 goto err; 208 209 /* 210 * Is the missing range not actually needed? 211 * 212 * p.v->idx refers to the data that we actually want, but if the 213 * indirect extent we point to was bigger, front_pad and back_pad 214 * indicate the range we took a reference on. 215 */ 216 217 if (missing_end <= live_start) { 218 new->v.front_pad = cpu_to_le32(live_start - missing_end); 219 } else if (missing_start >= live_end) { 220 new->v.back_pad = cpu_to_le32(missing_start - live_end); 221 } else { 222 struct bpos new_start = bkey_start_pos(&new->k); 223 struct bpos new_end = new->k.p; 224 225 if (missing_start > live_start) 226 new_start.offset += missing_start - live_start; 227 if (missing_end < live_end) 228 new_end.offset -= live_end - missing_end; 229 230 bch2_cut_front(new_start, &new->k_i); 231 bch2_cut_back(new_end, &new->k_i); 232 233 SET_REFLINK_P_ERROR(&new->v, true); 234 } 235 236 ret = bch2_btree_insert_trans(trans, BTREE_ID_extents, &new->k_i, BTREE_TRIGGER_norun); 237 if (ret) 238 goto err; 239 240 if (should_commit) 241 ret = bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?: 242 -BCH_ERR_transaction_restart_nested; 243 } 244 err: 245 fsck_err: 246 printbuf_exit(&buf); 247 return ret; 248 } 249 250 /* 251 * This is used from the read path, which doesn't expect to have to do a 252 * transaction commit, and from triggers, which should not be doing a commit: 253 */ 254 struct bkey_s_c bch2_lookup_indirect_extent(struct btree_trans *trans, 255 struct btree_iter *iter, 256 s64 *offset_into_extent, 257 struct bkey_s_c_reflink_p p, 258 bool should_commit, 259 unsigned iter_flags) 260 { 261 BUG_ON(*offset_into_extent < -((s64) le32_to_cpu(p.v->front_pad))); 262 BUG_ON(*offset_into_extent >= p.k->size + le32_to_cpu(p.v->back_pad)); 263 264 u64 reflink_offset = REFLINK_P_IDX(p.v) + *offset_into_extent; 265 266 struct bkey_s_c k = bch2_bkey_get_iter(trans, iter, BTREE_ID_reflink, 267 POS(0, reflink_offset), iter_flags); 268 if (bkey_err(k)) 269 return k; 270 271 if (unlikely(!bkey_extent_is_reflink_data(k.k))) { 272 unsigned size = min((u64) k.k->size, 273 REFLINK_P_IDX(p.v) + p.k->size + le32_to_cpu(p.v->back_pad) - 274 reflink_offset); 275 bch2_key_resize(&iter->k, size); 276 277 int ret = bch2_indirect_extent_missing_error(trans, p, reflink_offset, 278 k.k->p.offset, should_commit); 279 if (ret) { 280 bch2_trans_iter_exit(trans, iter); 281 return bkey_s_c_err(ret); 282 } 283 } else if (unlikely(REFLINK_P_ERROR(p.v))) { 284 int ret = bch2_indirect_extent_not_missing(trans, p, should_commit); 285 if (ret) { 286 bch2_trans_iter_exit(trans, iter); 287 return bkey_s_c_err(ret); 288 } 289 } 290 291 *offset_into_extent = reflink_offset - bkey_start_offset(k.k); 292 return k; 293 } 294 295 /* reflink pointer trigger */ 296 297 static int trans_trigger_reflink_p_segment(struct btree_trans *trans, 298 struct bkey_s_c_reflink_p p, u64 *idx, 299 enum btree_iter_update_trigger_flags flags) 300 { 301 struct bch_fs *c = trans->c; 302 struct printbuf buf = PRINTBUF; 303 304 s64 offset_into_extent = *idx - REFLINK_P_IDX(p.v); 305 struct btree_iter iter; 306 struct bkey_s_c k = bch2_lookup_indirect_extent(trans, &iter, &offset_into_extent, p, false, 307 BTREE_ITER_intent| 308 BTREE_ITER_with_updates); 309 int ret = bkey_err(k); 310 if (ret) 311 return ret; 312 313 if (!bkey_refcount_c(k)) { 314 if (!(flags & BTREE_TRIGGER_overwrite)) 315 ret = bch_err_throw(c, missing_indirect_extent); 316 goto next; 317 } 318 319 struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k); 320 ret = PTR_ERR_OR_ZERO(new); 321 if (ret) 322 goto err; 323 324 __le64 *refcount = bkey_refcount(bkey_i_to_s(new)); 325 if (!*refcount && (flags & BTREE_TRIGGER_overwrite)) { 326 bch2_bkey_val_to_text(&buf, c, p.s_c); 327 prt_newline(&buf); 328 bch2_bkey_val_to_text(&buf, c, k); 329 log_fsck_err(trans, reflink_refcount_underflow, 330 "indirect extent refcount underflow while marking\n%s", 331 buf.buf); 332 goto next; 333 } 334 335 if (flags & BTREE_TRIGGER_insert) { 336 struct bch_reflink_p *v = (struct bch_reflink_p *) p.v; 337 u64 pad; 338 339 pad = max_t(s64, le32_to_cpu(v->front_pad), 340 REFLINK_P_IDX(v) - bkey_start_offset(&new->k)); 341 BUG_ON(pad > U32_MAX); 342 v->front_pad = cpu_to_le32(pad); 343 344 pad = max_t(s64, le32_to_cpu(v->back_pad), 345 new->k.p.offset - p.k->size - REFLINK_P_IDX(v)); 346 BUG_ON(pad > U32_MAX); 347 v->back_pad = cpu_to_le32(pad); 348 } 349 350 le64_add_cpu(refcount, !(flags & BTREE_TRIGGER_overwrite) ? 1 : -1); 351 352 bch2_btree_iter_set_pos_to_extent_start(&iter); 353 ret = bch2_trans_update(trans, &iter, new, 0); 354 if (ret) 355 goto err; 356 next: 357 *idx = k.k->p.offset; 358 err: 359 fsck_err: 360 bch2_trans_iter_exit(trans, &iter); 361 printbuf_exit(&buf); 362 return ret; 363 } 364 365 static s64 gc_trigger_reflink_p_segment(struct btree_trans *trans, 366 struct bkey_s_c_reflink_p p, u64 *idx, 367 enum btree_iter_update_trigger_flags flags, 368 size_t r_idx) 369 { 370 struct bch_fs *c = trans->c; 371 struct reflink_gc *r; 372 int add = !(flags & BTREE_TRIGGER_overwrite) ? 1 : -1; 373 u64 next_idx = REFLINK_P_IDX(p.v) + p.k->size + le32_to_cpu(p.v->back_pad); 374 s64 ret = 0; 375 struct printbuf buf = PRINTBUF; 376 377 if (r_idx >= c->reflink_gc_nr) 378 goto not_found; 379 380 r = genradix_ptr(&c->reflink_gc_table, r_idx); 381 next_idx = min(next_idx, r->offset - r->size); 382 if (*idx < next_idx) 383 goto not_found; 384 385 BUG_ON((s64) r->refcount + add < 0); 386 387 if (flags & BTREE_TRIGGER_gc) 388 r->refcount += add; 389 *idx = r->offset; 390 return 0; 391 not_found: 392 if (flags & BTREE_TRIGGER_check_repair) { 393 ret = bch2_indirect_extent_missing_error(trans, p, *idx, next_idx, false); 394 if (ret) 395 goto err; 396 } 397 398 *idx = next_idx; 399 err: 400 printbuf_exit(&buf); 401 return ret; 402 } 403 404 static int __trigger_reflink_p(struct btree_trans *trans, 405 enum btree_id btree_id, unsigned level, struct bkey_s_c k, 406 enum btree_iter_update_trigger_flags flags) 407 { 408 struct bch_fs *c = trans->c; 409 struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k); 410 int ret = 0; 411 412 u64 idx = REFLINK_P_IDX(p.v) - le32_to_cpu(p.v->front_pad); 413 u64 end = REFLINK_P_IDX(p.v) + p.k->size + le32_to_cpu(p.v->back_pad); 414 415 if (flags & BTREE_TRIGGER_transactional) { 416 while (idx < end && !ret) 417 ret = trans_trigger_reflink_p_segment(trans, p, &idx, flags); 418 } 419 420 if (flags & (BTREE_TRIGGER_check_repair|BTREE_TRIGGER_gc)) { 421 size_t l = 0, r = c->reflink_gc_nr; 422 423 while (l < r) { 424 size_t m = l + (r - l) / 2; 425 struct reflink_gc *ref = genradix_ptr(&c->reflink_gc_table, m); 426 if (ref->offset <= idx) 427 l = m + 1; 428 else 429 r = m; 430 } 431 432 while (idx < end && !ret) 433 ret = gc_trigger_reflink_p_segment(trans, p, &idx, flags, l++); 434 } 435 436 return ret; 437 } 438 439 int bch2_trigger_reflink_p(struct btree_trans *trans, 440 enum btree_id btree_id, unsigned level, 441 struct bkey_s_c old, 442 struct bkey_s new, 443 enum btree_iter_update_trigger_flags flags) 444 { 445 if ((flags & BTREE_TRIGGER_transactional) && 446 (flags & BTREE_TRIGGER_insert)) { 447 struct bch_reflink_p *v = bkey_s_to_reflink_p(new).v; 448 449 v->front_pad = v->back_pad = 0; 450 } 451 452 return trigger_run_overwrite_then_insert(__trigger_reflink_p, trans, btree_id, level, old, new, flags); 453 } 454 455 /* indirect extent trigger */ 456 457 static inline void 458 check_indirect_extent_deleting(struct bkey_s new, 459 enum btree_iter_update_trigger_flags *flags) 460 { 461 if ((*flags & BTREE_TRIGGER_insert) && !*bkey_refcount(new)) { 462 new.k->type = KEY_TYPE_deleted; 463 new.k->size = 0; 464 set_bkey_val_u64s(new.k, 0); 465 *flags &= ~BTREE_TRIGGER_insert; 466 } 467 } 468 469 int bch2_trigger_reflink_v(struct btree_trans *trans, 470 enum btree_id btree_id, unsigned level, 471 struct bkey_s_c old, struct bkey_s new, 472 enum btree_iter_update_trigger_flags flags) 473 { 474 if ((flags & BTREE_TRIGGER_transactional) && 475 (flags & BTREE_TRIGGER_insert)) 476 check_indirect_extent_deleting(new, &flags); 477 478 return bch2_trigger_extent(trans, btree_id, level, old, new, flags); 479 } 480 481 int bch2_trigger_indirect_inline_data(struct btree_trans *trans, 482 enum btree_id btree_id, unsigned level, 483 struct bkey_s_c old, struct bkey_s new, 484 enum btree_iter_update_trigger_flags flags) 485 { 486 check_indirect_extent_deleting(new, &flags); 487 488 return 0; 489 } 490 491 /* create */ 492 493 static int bch2_make_extent_indirect(struct btree_trans *trans, 494 struct btree_iter *extent_iter, 495 struct bkey_i *orig, 496 bool reflink_p_may_update_opts_field) 497 { 498 struct bch_fs *c = trans->c; 499 struct btree_iter reflink_iter = {}; 500 struct bkey_s_c k; 501 struct bkey_i *r_v; 502 struct bkey_i_reflink_p *r_p; 503 __le64 *refcount; 504 int ret; 505 506 if (orig->k.type == KEY_TYPE_inline_data) 507 bch2_check_set_feature(c, BCH_FEATURE_reflink_inline_data); 508 509 bch2_trans_iter_init(trans, &reflink_iter, BTREE_ID_reflink, POS_MAX, 510 BTREE_ITER_intent); 511 k = bch2_btree_iter_peek_prev(trans, &reflink_iter); 512 ret = bkey_err(k); 513 if (ret) 514 goto err; 515 516 /* 517 * XXX: we're assuming that 56 bits will be enough for the life of the 518 * filesystem: we need to implement wraparound, with a cursor in the 519 * logged ops btree: 520 */ 521 if (bkey_ge(reflink_iter.pos, POS(0, REFLINK_P_IDX_MAX - orig->k.size))) 522 return -ENOSPC; 523 524 r_v = bch2_trans_kmalloc(trans, sizeof(__le64) + bkey_bytes(&orig->k)); 525 ret = PTR_ERR_OR_ZERO(r_v); 526 if (ret) 527 goto err; 528 529 bkey_init(&r_v->k); 530 r_v->k.type = bkey_type_to_indirect(&orig->k); 531 r_v->k.p = reflink_iter.pos; 532 bch2_key_resize(&r_v->k, orig->k.size); 533 r_v->k.bversion = orig->k.bversion; 534 535 set_bkey_val_bytes(&r_v->k, sizeof(__le64) + bkey_val_bytes(&orig->k)); 536 537 refcount = bkey_refcount(bkey_i_to_s(r_v)); 538 *refcount = 0; 539 memcpy(refcount + 1, &orig->v, bkey_val_bytes(&orig->k)); 540 541 ret = bch2_trans_update(trans, &reflink_iter, r_v, 0); 542 if (ret) 543 goto err; 544 545 /* 546 * orig is in a bkey_buf which statically allocates 5 64s for the val, 547 * so we know it will be big enough: 548 */ 549 orig->k.type = KEY_TYPE_reflink_p; 550 r_p = bkey_i_to_reflink_p(orig); 551 set_bkey_val_bytes(&r_p->k, sizeof(r_p->v)); 552 553 /* FORTIFY_SOURCE is broken here, and doesn't provide unsafe_memset() */ 554 #if !defined(__NO_FORTIFY) && defined(__OPTIMIZE__) && defined(CONFIG_FORTIFY_SOURCE) 555 __underlying_memset(&r_p->v, 0, sizeof(r_p->v)); 556 #else 557 memset(&r_p->v, 0, sizeof(r_p->v)); 558 #endif 559 560 SET_REFLINK_P_IDX(&r_p->v, bkey_start_offset(&r_v->k)); 561 562 if (reflink_p_may_update_opts_field) 563 SET_REFLINK_P_MAY_UPDATE_OPTIONS(&r_p->v, true); 564 565 ret = bch2_trans_update(trans, extent_iter, &r_p->k_i, 566 BTREE_UPDATE_internal_snapshot_node); 567 err: 568 bch2_trans_iter_exit(trans, &reflink_iter); 569 570 return ret; 571 } 572 573 static struct bkey_s_c get_next_src(struct btree_trans *trans, 574 struct btree_iter *iter, struct bpos end) 575 { 576 struct bkey_s_c k; 577 int ret; 578 579 for_each_btree_key_max_continue_norestart(trans, *iter, end, 0, k, ret) { 580 if (bkey_extent_is_unwritten(k)) 581 continue; 582 583 if (bkey_extent_is_data(k.k)) 584 return k; 585 } 586 587 if (bkey_ge(iter->pos, end)) 588 bch2_btree_iter_set_pos(trans, iter, end); 589 return ret ? bkey_s_c_err(ret) : bkey_s_c_null; 590 } 591 592 s64 bch2_remap_range(struct bch_fs *c, 593 subvol_inum dst_inum, u64 dst_offset, 594 subvol_inum src_inum, u64 src_offset, 595 u64 remap_sectors, 596 u64 new_i_size, s64 *i_sectors_delta, 597 bool may_change_src_io_path_opts) 598 { 599 struct btree_trans *trans; 600 struct btree_iter dst_iter, src_iter; 601 struct bkey_s_c src_k; 602 struct bkey_buf new_dst, new_src; 603 struct bpos dst_start = POS(dst_inum.inum, dst_offset); 604 struct bpos src_start = POS(src_inum.inum, src_offset); 605 struct bpos dst_end = dst_start, src_end = src_start; 606 struct bch_io_opts opts; 607 struct bpos src_want; 608 u64 dst_done = 0; 609 u32 dst_snapshot, src_snapshot; 610 bool reflink_p_may_update_opts_field = 611 !bch2_request_incompat_feature(c, bcachefs_metadata_version_reflink_p_may_update_opts); 612 int ret = 0, ret2 = 0; 613 614 if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_reflink)) 615 return bch_err_throw(c, erofs_no_writes); 616 617 bch2_check_set_feature(c, BCH_FEATURE_reflink); 618 619 dst_end.offset += remap_sectors; 620 src_end.offset += remap_sectors; 621 622 bch2_bkey_buf_init(&new_dst); 623 bch2_bkey_buf_init(&new_src); 624 trans = bch2_trans_get(c); 625 626 ret = bch2_inum_opts_get(trans, src_inum, &opts); 627 if (ret) 628 goto err; 629 630 bch2_trans_iter_init(trans, &src_iter, BTREE_ID_extents, src_start, 631 BTREE_ITER_intent); 632 bch2_trans_iter_init(trans, &dst_iter, BTREE_ID_extents, dst_start, 633 BTREE_ITER_intent); 634 635 while ((ret == 0 || 636 bch2_err_matches(ret, BCH_ERR_transaction_restart)) && 637 bkey_lt(dst_iter.pos, dst_end)) { 638 struct disk_reservation disk_res = { 0 }; 639 640 bch2_trans_begin(trans); 641 642 if (fatal_signal_pending(current)) { 643 ret = -EINTR; 644 break; 645 } 646 647 ret = bch2_subvolume_get_snapshot(trans, src_inum.subvol, 648 &src_snapshot); 649 if (ret) 650 continue; 651 652 bch2_btree_iter_set_snapshot(trans, &src_iter, src_snapshot); 653 654 ret = bch2_subvolume_get_snapshot(trans, dst_inum.subvol, 655 &dst_snapshot); 656 if (ret) 657 continue; 658 659 bch2_btree_iter_set_snapshot(trans, &dst_iter, dst_snapshot); 660 661 if (dst_inum.inum < src_inum.inum) { 662 /* Avoid some lock cycle transaction restarts */ 663 ret = bch2_btree_iter_traverse(trans, &dst_iter); 664 if (ret) 665 continue; 666 } 667 668 dst_done = dst_iter.pos.offset - dst_start.offset; 669 src_want = POS(src_start.inode, src_start.offset + dst_done); 670 bch2_btree_iter_set_pos(trans, &src_iter, src_want); 671 672 src_k = get_next_src(trans, &src_iter, src_end); 673 ret = bkey_err(src_k); 674 if (ret) 675 continue; 676 677 if (bkey_lt(src_want, src_iter.pos)) { 678 ret = bch2_fpunch_at(trans, &dst_iter, dst_inum, 679 min(dst_end.offset, 680 dst_iter.pos.offset + 681 src_iter.pos.offset - src_want.offset), 682 i_sectors_delta); 683 continue; 684 } 685 686 if (src_k.k->type != KEY_TYPE_reflink_p) { 687 bch2_btree_iter_set_pos_to_extent_start(&src_iter); 688 689 bch2_bkey_buf_reassemble(&new_src, c, src_k); 690 src_k = bkey_i_to_s_c(new_src.k); 691 692 ret = bch2_make_extent_indirect(trans, &src_iter, 693 new_src.k, 694 reflink_p_may_update_opts_field); 695 if (ret) 696 continue; 697 698 BUG_ON(src_k.k->type != KEY_TYPE_reflink_p); 699 } 700 701 if (src_k.k->type == KEY_TYPE_reflink_p) { 702 struct bkey_s_c_reflink_p src_p = 703 bkey_s_c_to_reflink_p(src_k); 704 struct bkey_i_reflink_p *dst_p = 705 bkey_reflink_p_init(new_dst.k); 706 707 u64 offset = REFLINK_P_IDX(src_p.v) + 708 (src_want.offset - 709 bkey_start_offset(src_k.k)); 710 711 SET_REFLINK_P_IDX(&dst_p->v, offset); 712 713 if (reflink_p_may_update_opts_field && 714 may_change_src_io_path_opts && 715 REFLINK_P_MAY_UPDATE_OPTIONS(src_p.v)) 716 SET_REFLINK_P_MAY_UPDATE_OPTIONS(&dst_p->v, true); 717 } else { 718 BUG(); 719 } 720 721 new_dst.k->k.p = dst_iter.pos; 722 bch2_key_resize(&new_dst.k->k, 723 min(src_k.k->p.offset - src_want.offset, 724 dst_end.offset - dst_iter.pos.offset)); 725 726 ret = bch2_bkey_set_needs_rebalance(c, &opts, new_dst.k) ?: 727 bch2_extent_update(trans, dst_inum, &dst_iter, 728 new_dst.k, &disk_res, 729 new_i_size, i_sectors_delta, 730 true); 731 bch2_disk_reservation_put(c, &disk_res); 732 } 733 bch2_trans_iter_exit(trans, &dst_iter); 734 bch2_trans_iter_exit(trans, &src_iter); 735 736 BUG_ON(!ret && !bkey_eq(dst_iter.pos, dst_end)); 737 BUG_ON(bkey_gt(dst_iter.pos, dst_end)); 738 739 dst_done = dst_iter.pos.offset - dst_start.offset; 740 new_i_size = min(dst_iter.pos.offset << 9, new_i_size); 741 742 do { 743 struct bch_inode_unpacked inode_u; 744 struct btree_iter inode_iter = {}; 745 746 bch2_trans_begin(trans); 747 748 ret2 = bch2_inode_peek(trans, &inode_iter, &inode_u, 749 dst_inum, BTREE_ITER_intent); 750 751 if (!ret2 && 752 inode_u.bi_size < new_i_size) { 753 inode_u.bi_size = new_i_size; 754 ret2 = bch2_inode_write(trans, &inode_iter, &inode_u) ?: 755 bch2_trans_commit(trans, NULL, NULL, 756 BCH_TRANS_COMMIT_no_enospc); 757 } 758 759 bch2_trans_iter_exit(trans, &inode_iter); 760 } while (bch2_err_matches(ret2, BCH_ERR_transaction_restart)); 761 err: 762 bch2_trans_put(trans); 763 bch2_bkey_buf_exit(&new_src, c); 764 bch2_bkey_buf_exit(&new_dst, c); 765 766 enumerated_ref_put(&c->writes, BCH_WRITE_REF_reflink); 767 768 return dst_done ?: ret ?: ret2; 769 } 770 771 /* fsck */ 772 773 static int bch2_gc_write_reflink_key(struct btree_trans *trans, 774 struct btree_iter *iter, 775 struct bkey_s_c k, 776 size_t *idx) 777 { 778 struct bch_fs *c = trans->c; 779 const __le64 *refcount = bkey_refcount_c(k); 780 struct printbuf buf = PRINTBUF; 781 struct reflink_gc *r; 782 int ret = 0; 783 784 if (!refcount) 785 return 0; 786 787 while ((r = genradix_ptr(&c->reflink_gc_table, *idx)) && 788 r->offset < k.k->p.offset) 789 ++*idx; 790 791 if (!r || 792 r->offset != k.k->p.offset || 793 r->size != k.k->size) { 794 bch_err(c, "unexpected inconsistency walking reflink table at gc finish"); 795 return -EINVAL; 796 } 797 798 if (fsck_err_on(r->refcount != le64_to_cpu(*refcount), 799 trans, reflink_v_refcount_wrong, 800 "reflink key has wrong refcount:\n" 801 "%s\n" 802 "should be %u", 803 (bch2_bkey_val_to_text(&buf, c, k), buf.buf), 804 r->refcount)) { 805 struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k); 806 ret = PTR_ERR_OR_ZERO(new); 807 if (ret) 808 goto out; 809 810 if (!r->refcount) 811 new->k.type = KEY_TYPE_deleted; 812 else 813 *bkey_refcount(bkey_i_to_s(new)) = cpu_to_le64(r->refcount); 814 ret = bch2_trans_update(trans, iter, new, 0); 815 } 816 out: 817 fsck_err: 818 printbuf_exit(&buf); 819 return ret; 820 } 821 822 int bch2_gc_reflink_done(struct bch_fs *c) 823 { 824 size_t idx = 0; 825 826 int ret = bch2_trans_run(c, 827 for_each_btree_key_commit(trans, iter, 828 BTREE_ID_reflink, POS_MIN, 829 BTREE_ITER_prefetch, k, 830 NULL, NULL, BCH_TRANS_COMMIT_no_enospc, 831 bch2_gc_write_reflink_key(trans, &iter, k, &idx))); 832 c->reflink_gc_nr = 0; 833 return ret; 834 } 835 836 int bch2_gc_reflink_start(struct bch_fs *c) 837 { 838 c->reflink_gc_nr = 0; 839 840 int ret = bch2_trans_run(c, 841 for_each_btree_key(trans, iter, BTREE_ID_reflink, POS_MIN, 842 BTREE_ITER_prefetch, k, ({ 843 const __le64 *refcount = bkey_refcount_c(k); 844 845 if (!refcount) 846 continue; 847 848 struct reflink_gc *r = genradix_ptr_alloc(&c->reflink_gc_table, 849 c->reflink_gc_nr++, GFP_KERNEL); 850 if (!r) { 851 ret = bch_err_throw(c, ENOMEM_gc_reflink_start); 852 break; 853 } 854 855 r->offset = k.k->p.offset; 856 r->size = k.k->size; 857 r->refcount = 0; 858 0; 859 }))); 860 861 bch_err_fn(c, ret); 862 return ret; 863 } 864