1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2010 Kent Overstreet <kent.overstreet@gmail.com>
5 * Code for managing the extent btree and dynamically updating the writeback
10 #include "bkey_methods.h"
13 #include "btree_iter.h"
17 #include "disk_groups.h"
27 #include <trace/events/bcachefs.h>
29 static unsigned bch2_crc_field_size_max[] = {
30 [BCH_EXTENT_ENTRY_crc32] = CRC32_SIZE_MAX,
31 [BCH_EXTENT_ENTRY_crc64] = CRC64_SIZE_MAX,
32 [BCH_EXTENT_ENTRY_crc128] = CRC128_SIZE_MAX,
35 static void bch2_extent_crc_pack(union bch_extent_crc *,
36 struct bch_extent_crc_unpacked,
37 enum bch_extent_entry_type);
39 static struct bch_dev_io_failures *dev_io_failures(struct bch_io_failures *f,
42 struct bch_dev_io_failures *i;
44 for (i = f->devs; i < f->devs + f->nr; i++)
51 void bch2_mark_io_failure(struct bch_io_failures *failed,
52 struct extent_ptr_decoded *p)
54 struct bch_dev_io_failures *f = dev_io_failures(failed, p->ptr.dev);
57 BUG_ON(failed->nr >= ARRAY_SIZE(failed->devs));
59 f = &failed->devs[failed->nr++];
64 } else if (p->idx != f->idx) {
74 * returns true if p1 is better than p2:
76 static inline bool ptr_better(struct bch_fs *c,
77 const struct extent_ptr_decoded p1,
78 const struct extent_ptr_decoded p2)
80 if (likely(!p1.idx && !p2.idx)) {
81 struct bch_dev *dev1 = bch_dev_bkey_exists(c, p1.ptr.dev);
82 struct bch_dev *dev2 = bch_dev_bkey_exists(c, p2.ptr.dev);
84 u64 l1 = atomic64_read(&dev1->cur_latency[READ]);
85 u64 l2 = atomic64_read(&dev2->cur_latency[READ]);
87 /* Pick at random, biased in favor of the faster device: */
89 return bch2_rand_range(l1 + l2) > l1;
92 if (bch2_force_reconstruct_read)
93 return p1.idx > p2.idx;
95 return p1.idx < p2.idx;
99 * This picks a non-stale pointer, preferably from a device other than @avoid.
100 * Avoid can be NULL, meaning pick any. If there are no non-stale pointers to
101 * other devices, it will still pick a pointer from avoid.
103 int bch2_bkey_pick_read_device(struct bch_fs *c, struct bkey_s_c k,
104 struct bch_io_failures *failed,
105 struct extent_ptr_decoded *pick)
107 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
108 const union bch_extent_entry *entry;
109 struct extent_ptr_decoded p;
110 struct bch_dev_io_failures *f;
114 if (k.k->type == KEY_TYPE_error)
117 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
118 ca = bch_dev_bkey_exists(c, p.ptr.dev);
121 * If there are any dirty pointers it's an error if we can't
124 if (!ret && !p.ptr.cached)
127 if (p.ptr.cached && ptr_stale(ca, &p.ptr))
130 f = failed ? dev_io_failures(failed, p.ptr.dev) : NULL;
132 p.idx = f->nr_failed < f->nr_retries
137 !bch2_dev_is_readable(ca))
140 if (bch2_force_reconstruct_read &&
144 if (p.idx >= (unsigned) p.has_ec + 1)
147 if (ret > 0 && !ptr_better(c, p, *pick))
157 /* KEY_TYPE_btree_ptr: */
159 int bch2_btree_ptr_invalid(const struct bch_fs *c, struct bkey_s_c k,
160 int rw, struct printbuf *err)
162 if (bkey_val_u64s(k.k) > BCH_REPLICAS_MAX) {
163 prt_printf(err, "value too big (%zu > %u)",
164 bkey_val_u64s(k.k), BCH_REPLICAS_MAX);
168 return bch2_bkey_ptrs_invalid(c, k, rw, err);
171 void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
174 bch2_bkey_ptrs_to_text(out, c, k);
177 int bch2_btree_ptr_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
178 int rw, struct printbuf *err)
180 struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
182 if (bkey_val_bytes(k.k) <= sizeof(*bp.v)) {
183 prt_printf(err, "value too small (%zu <= %zu)",
184 bkey_val_bytes(k.k), sizeof(*bp.v));
188 if (bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX) {
189 prt_printf(err, "value too big (%zu > %zu)",
190 bkey_val_u64s(k.k), BKEY_BTREE_PTR_VAL_U64s_MAX);
194 if (c->sb.version < bcachefs_metadata_version_snapshot &&
195 bp.v->min_key.snapshot) {
196 prt_printf(err, "invalid min_key.snapshot (%u != 0)",
197 bp.v->min_key.snapshot);
201 return bch2_bkey_ptrs_invalid(c, k, rw, err);
204 void bch2_btree_ptr_v2_to_text(struct printbuf *out, struct bch_fs *c,
207 struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
209 prt_printf(out, "seq %llx written %u min_key %s",
210 le64_to_cpu(bp.v->seq),
211 le16_to_cpu(bp.v->sectors_written),
212 BTREE_PTR_RANGE_UPDATED(bp.v) ? "R " : "");
214 bch2_bpos_to_text(out, bp.v->min_key);
215 prt_printf(out, " ");
216 bch2_bkey_ptrs_to_text(out, c, k);
219 void bch2_btree_ptr_v2_compat(enum btree_id btree_id, unsigned version,
220 unsigned big_endian, int write,
223 struct bkey_s_btree_ptr_v2 bp = bkey_s_to_btree_ptr_v2(k);
225 compat_bpos(0, btree_id, version, big_endian, write, &bp.v->min_key);
227 if (version < bcachefs_metadata_version_inode_btree_change &&
228 btree_node_type_is_extents(btree_id) &&
229 bkey_cmp(bp.v->min_key, POS_MIN))
230 bp.v->min_key = write
231 ? bpos_nosnap_predecessor(bp.v->min_key)
232 : bpos_nosnap_successor(bp.v->min_key);
235 /* KEY_TYPE_extent: */
237 bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
239 struct bkey_ptrs l_ptrs = bch2_bkey_ptrs(l);
240 struct bkey_ptrs_c r_ptrs = bch2_bkey_ptrs_c(r);
241 union bch_extent_entry *en_l;
242 const union bch_extent_entry *en_r;
243 struct extent_ptr_decoded lp, rp;
249 while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
250 if (extent_entry_type(en_l) != extent_entry_type(en_r))
253 en_l = extent_entry_next(en_l);
254 en_r = extent_entry_next(en_r);
257 if (en_l < l_ptrs.end || en_r < r_ptrs.end)
262 lp.crc = bch2_extent_crc_unpack(l.k, NULL);
263 rp.crc = bch2_extent_crc_unpack(r.k, NULL);
265 while (__bkey_ptr_next_decode(l.k, l_ptrs.end, lp, en_l) &&
266 __bkey_ptr_next_decode(r.k, r_ptrs.end, rp, en_r)) {
267 if (lp.ptr.offset + lp.crc.offset + lp.crc.live_size !=
268 rp.ptr.offset + rp.crc.offset ||
269 lp.ptr.dev != rp.ptr.dev ||
270 lp.ptr.gen != rp.ptr.gen ||
271 lp.has_ec != rp.has_ec)
274 /* Extents may not straddle buckets: */
275 ca = bch_dev_bkey_exists(c, lp.ptr.dev);
276 if (PTR_BUCKET_NR(ca, &lp.ptr) != PTR_BUCKET_NR(ca, &rp.ptr))
279 if (lp.has_ec != rp.has_ec ||
281 (lp.ec.block != rp.ec.block ||
282 lp.ec.redundancy != rp.ec.redundancy ||
283 lp.ec.idx != rp.ec.idx)))
286 if (lp.crc.compression_type != rp.crc.compression_type ||
287 lp.crc.nonce != rp.crc.nonce)
290 if (lp.crc.offset + lp.crc.live_size + rp.crc.live_size <=
291 lp.crc.uncompressed_size) {
292 /* can use left extent's crc entry */
293 } else if (lp.crc.live_size <= rp.crc.offset ) {
294 /* can use right extent's crc entry */
296 /* check if checksums can be merged: */
297 if (lp.crc.csum_type != rp.crc.csum_type ||
298 lp.crc.nonce != rp.crc.nonce ||
299 crc_is_compressed(lp.crc) ||
300 !bch2_checksum_mergeable(lp.crc.csum_type))
303 if (lp.crc.offset + lp.crc.live_size != lp.crc.compressed_size ||
307 if (lp.crc.csum_type &&
308 lp.crc.uncompressed_size +
309 rp.crc.uncompressed_size > (c->opts.encoded_extent_max >> 9))
313 en_l = extent_entry_next(en_l);
314 en_r = extent_entry_next(en_r);
319 while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
320 if (extent_entry_is_crc(en_l)) {
321 struct bch_extent_crc_unpacked crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
322 struct bch_extent_crc_unpacked crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
324 if (crc_l.uncompressed_size + crc_r.uncompressed_size >
325 bch2_crc_field_size_max[extent_entry_type(en_l)])
329 en_l = extent_entry_next(en_l);
330 en_r = extent_entry_next(en_r);
333 use_right_ptr = false;
336 while (en_l < l_ptrs.end) {
337 if (extent_entry_type(en_l) == BCH_EXTENT_ENTRY_ptr &&
339 en_l->ptr = en_r->ptr;
341 if (extent_entry_is_crc(en_l)) {
342 struct bch_extent_crc_unpacked crc_l =
343 bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
344 struct bch_extent_crc_unpacked crc_r =
345 bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
347 use_right_ptr = false;
349 if (crc_l.offset + crc_l.live_size + crc_r.live_size <=
350 crc_l.uncompressed_size) {
351 /* can use left extent's crc entry */
352 } else if (crc_l.live_size <= crc_r.offset ) {
353 /* can use right extent's crc entry */
354 crc_r.offset -= crc_l.live_size;
355 bch2_extent_crc_pack(entry_to_crc(en_l), crc_r,
356 extent_entry_type(en_l));
357 use_right_ptr = true;
359 crc_l.csum = bch2_checksum_merge(crc_l.csum_type,
362 crc_r.uncompressed_size << 9);
364 crc_l.uncompressed_size += crc_r.uncompressed_size;
365 crc_l.compressed_size += crc_r.compressed_size;
366 bch2_extent_crc_pack(entry_to_crc(en_l), crc_l,
367 extent_entry_type(en_l));
371 en_l = extent_entry_next(en_l);
372 en_r = extent_entry_next(en_r);
375 bch2_key_resize(l.k, l.k->size + r.k->size);
379 /* KEY_TYPE_reservation: */
381 int bch2_reservation_invalid(const struct bch_fs *c, struct bkey_s_c k,
382 int rw, struct printbuf *err)
384 struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
386 if (bkey_val_bytes(k.k) != sizeof(struct bch_reservation)) {
387 prt_printf(err, "incorrect value size (%zu != %zu)",
388 bkey_val_bytes(k.k), sizeof(*r.v));
392 if (!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX) {
393 prt_printf(err, "invalid nr_replicas (%u)",
401 void bch2_reservation_to_text(struct printbuf *out, struct bch_fs *c,
404 struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
406 prt_printf(out, "generation %u replicas %u",
407 le32_to_cpu(r.v->generation),
411 bool bch2_reservation_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
413 struct bkey_s_reservation l = bkey_s_to_reservation(_l);
414 struct bkey_s_c_reservation r = bkey_s_c_to_reservation(_r);
416 if (l.v->generation != r.v->generation ||
417 l.v->nr_replicas != r.v->nr_replicas)
420 bch2_key_resize(l.k, l.k->size + r.k->size);
424 /* Extent checksum entries: */
426 /* returns true if not equal */
427 static inline bool bch2_crc_unpacked_cmp(struct bch_extent_crc_unpacked l,
428 struct bch_extent_crc_unpacked r)
430 return (l.csum_type != r.csum_type ||
431 l.compression_type != r.compression_type ||
432 l.compressed_size != r.compressed_size ||
433 l.uncompressed_size != r.uncompressed_size ||
434 l.offset != r.offset ||
435 l.live_size != r.live_size ||
436 l.nonce != r.nonce ||
437 bch2_crc_cmp(l.csum, r.csum));
440 static inline bool can_narrow_crc(struct bch_extent_crc_unpacked u,
441 struct bch_extent_crc_unpacked n)
443 return !crc_is_compressed(u) &&
445 u.uncompressed_size > u.live_size &&
446 bch2_csum_type_is_encryption(u.csum_type) ==
447 bch2_csum_type_is_encryption(n.csum_type);
450 bool bch2_can_narrow_extent_crcs(struct bkey_s_c k,
451 struct bch_extent_crc_unpacked n)
453 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
454 struct bch_extent_crc_unpacked crc;
455 const union bch_extent_entry *i;
460 bkey_for_each_crc(k.k, ptrs, crc, i)
461 if (can_narrow_crc(crc, n))
468 * We're writing another replica for this extent, so while we've got the data in
469 * memory we'll be computing a new checksum for the currently live data.
471 * If there are other replicas we aren't moving, and they are checksummed but
472 * not compressed, we can modify them to point to only the data that is
473 * currently live (so that readers won't have to bounce) while we've got the
476 bool bch2_bkey_narrow_crcs(struct bkey_i *k, struct bch_extent_crc_unpacked n)
478 struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
479 struct bch_extent_crc_unpacked u;
480 struct extent_ptr_decoded p;
481 union bch_extent_entry *i;
484 /* Find a checksum entry that covers only live data: */
486 bkey_for_each_crc(&k->k, ptrs, u, i)
487 if (!crc_is_compressed(u) &&
489 u.live_size == u.uncompressed_size) {
496 BUG_ON(crc_is_compressed(n));
498 BUG_ON(n.live_size != k->k.size);
500 restart_narrow_pointers:
501 ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
503 bkey_for_each_ptr_decode(&k->k, ptrs, p, i)
504 if (can_narrow_crc(p.crc, n)) {
505 __bch2_bkey_drop_ptr(bkey_i_to_s(k), &i->ptr);
506 p.ptr.offset += p.crc.offset;
508 bch2_extent_ptr_decoded_append(k, &p);
510 goto restart_narrow_pointers;
516 static void bch2_extent_crc_pack(union bch_extent_crc *dst,
517 struct bch_extent_crc_unpacked src,
518 enum bch_extent_entry_type type)
520 #define set_common_fields(_dst, _src) \
521 _dst.type = 1 << type; \
522 _dst.csum_type = _src.csum_type, \
523 _dst.compression_type = _src.compression_type, \
524 _dst._compressed_size = _src.compressed_size - 1, \
525 _dst._uncompressed_size = _src.uncompressed_size - 1, \
526 _dst.offset = _src.offset
529 case BCH_EXTENT_ENTRY_crc32:
530 set_common_fields(dst->crc32, src);
531 dst->crc32.csum = *((__le32 *) &src.csum.lo);
533 case BCH_EXTENT_ENTRY_crc64:
534 set_common_fields(dst->crc64, src);
535 dst->crc64.nonce = src.nonce;
536 dst->crc64.csum_lo = src.csum.lo;
537 dst->crc64.csum_hi = *((__le16 *) &src.csum.hi);
539 case BCH_EXTENT_ENTRY_crc128:
540 set_common_fields(dst->crc128, src);
541 dst->crc128.nonce = src.nonce;
542 dst->crc128.csum = src.csum;
547 #undef set_common_fields
550 void bch2_extent_crc_append(struct bkey_i *k,
551 struct bch_extent_crc_unpacked new)
553 struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
554 union bch_extent_crc *crc = (void *) ptrs.end;
555 enum bch_extent_entry_type type;
557 if (bch_crc_bytes[new.csum_type] <= 4 &&
558 new.uncompressed_size <= CRC32_SIZE_MAX &&
559 new.nonce <= CRC32_NONCE_MAX)
560 type = BCH_EXTENT_ENTRY_crc32;
561 else if (bch_crc_bytes[new.csum_type] <= 10 &&
562 new.uncompressed_size <= CRC64_SIZE_MAX &&
563 new.nonce <= CRC64_NONCE_MAX)
564 type = BCH_EXTENT_ENTRY_crc64;
565 else if (bch_crc_bytes[new.csum_type] <= 16 &&
566 new.uncompressed_size <= CRC128_SIZE_MAX &&
567 new.nonce <= CRC128_NONCE_MAX)
568 type = BCH_EXTENT_ENTRY_crc128;
572 bch2_extent_crc_pack(crc, new, type);
574 k->k.u64s += extent_entry_u64s(ptrs.end);
576 EBUG_ON(bkey_val_u64s(&k->k) > BKEY_EXTENT_VAL_U64s_MAX);
579 /* Generic code for keys with pointers: */
581 unsigned bch2_bkey_nr_ptrs(struct bkey_s_c k)
583 return bch2_bkey_devs(k).nr;
586 unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c k)
588 return k.k->type == KEY_TYPE_reservation
589 ? bkey_s_c_to_reservation(k).v->nr_replicas
590 : bch2_bkey_dirty_devs(k).nr;
593 unsigned bch2_bkey_nr_ptrs_fully_allocated(struct bkey_s_c k)
597 if (k.k->type == KEY_TYPE_reservation) {
598 ret = bkey_s_c_to_reservation(k).v->nr_replicas;
600 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
601 const union bch_extent_entry *entry;
602 struct extent_ptr_decoded p;
604 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
605 ret += !p.ptr.cached && !crc_is_compressed(p.crc);
611 unsigned bch2_bkey_sectors_compressed(struct bkey_s_c k)
613 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
614 const union bch_extent_entry *entry;
615 struct extent_ptr_decoded p;
618 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
619 if (!p.ptr.cached && crc_is_compressed(p.crc))
620 ret += p.crc.compressed_size;
625 bool bch2_bkey_is_incompressible(struct bkey_s_c k)
627 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
628 const union bch_extent_entry *entry;
629 struct bch_extent_crc_unpacked crc;
631 bkey_for_each_crc(k.k, ptrs, crc, entry)
632 if (crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
637 unsigned bch2_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
639 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
640 const union bch_extent_entry *entry;
641 struct extent_ptr_decoded p = { 0 };
642 unsigned replicas = 0;
644 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
649 replicas += p.ec.redundancy;
658 static unsigned bch2_extent_ptr_durability(struct bch_fs *c,
659 struct extent_ptr_decoded p)
661 unsigned durability = 0;
667 ca = bch_dev_bkey_exists(c, p.ptr.dev);
669 if (ca->mi.state != BCH_MEMBER_STATE_failed)
670 durability = max_t(unsigned, durability, ca->mi.durability);
673 durability += p.ec.redundancy;
678 unsigned bch2_bkey_durability(struct bch_fs *c, struct bkey_s_c k)
680 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
681 const union bch_extent_entry *entry;
682 struct extent_ptr_decoded p;
683 unsigned durability = 0;
685 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
686 durability += bch2_extent_ptr_durability(c, p);
691 void bch2_bkey_mark_replicas_cached(struct bch_fs *c, struct bkey_s k,
693 unsigned nr_desired_replicas)
695 struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
696 union bch_extent_entry *entry;
697 struct extent_ptr_decoded p;
698 int extra = bch2_bkey_durability(c, k.s_c) - nr_desired_replicas;
700 if (target && extra > 0)
701 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
702 int n = bch2_extent_ptr_durability(c, p);
704 if (n && n <= extra &&
705 !bch2_dev_in_target(c, p.ptr.dev, target)) {
706 entry->ptr.cached = true;
712 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
713 int n = bch2_extent_ptr_durability(c, p);
715 if (n && n <= extra) {
716 entry->ptr.cached = true;
722 void bch2_bkey_extent_entry_drop(struct bkey_i *k, union bch_extent_entry *entry)
724 union bch_extent_entry *end = bkey_val_end(bkey_i_to_s(k));
725 union bch_extent_entry *next = extent_entry_next(entry);
727 memmove_u64s(entry, next, (u64 *) end - (u64 *) next);
728 k->k.u64s -= extent_entry_u64s(entry);
731 void bch2_bkey_append_ptr(struct bkey_i *k,
732 struct bch_extent_ptr ptr)
734 EBUG_ON(bch2_bkey_has_device(bkey_i_to_s_c(k), ptr.dev));
737 case KEY_TYPE_btree_ptr:
738 case KEY_TYPE_btree_ptr_v2:
739 case KEY_TYPE_extent:
740 EBUG_ON(bkey_val_u64s(&k->k) >= BKEY_EXTENT_VAL_U64s_MAX);
742 ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
744 memcpy((void *) &k->v + bkey_val_bytes(&k->k),
754 static inline void __extent_entry_insert(struct bkey_i *k,
755 union bch_extent_entry *dst,
756 union bch_extent_entry *new)
758 union bch_extent_entry *end = bkey_val_end(bkey_i_to_s(k));
760 memmove_u64s_up_small((u64 *) dst + extent_entry_u64s(new),
761 dst, (u64 *) end - (u64 *) dst);
762 k->k.u64s += extent_entry_u64s(new);
763 memcpy(dst, new, extent_entry_bytes(new));
766 void bch2_extent_ptr_decoded_append(struct bkey_i *k,
767 struct extent_ptr_decoded *p)
769 struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
770 struct bch_extent_crc_unpacked crc =
771 bch2_extent_crc_unpack(&k->k, NULL);
772 union bch_extent_entry *pos;
774 if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
779 bkey_for_each_crc(&k->k, ptrs, crc, pos)
780 if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
781 pos = extent_entry_next(pos);
785 bch2_extent_crc_append(k, p->crc);
786 pos = bkey_val_end(bkey_i_to_s(k));
788 p->ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
789 __extent_entry_insert(k, pos, to_entry(&p->ptr));
792 p->ec.type = 1 << BCH_EXTENT_ENTRY_stripe_ptr;
793 __extent_entry_insert(k, pos, to_entry(&p->ec));
797 static union bch_extent_entry *extent_entry_prev(struct bkey_ptrs ptrs,
798 union bch_extent_entry *entry)
800 union bch_extent_entry *i = ptrs.start;
805 while (extent_entry_next(i) != entry)
806 i = extent_entry_next(i);
810 static void extent_entry_drop(struct bkey_s k, union bch_extent_entry *entry)
812 union bch_extent_entry *next = extent_entry_next(entry);
814 /* stripes have ptrs, but their layout doesn't work with this code */
815 BUG_ON(k.k->type == KEY_TYPE_stripe);
817 memmove_u64s_down(entry, next,
818 (u64 *) bkey_val_end(k) - (u64 *) next);
819 k.k->u64s -= (u64 *) next - (u64 *) entry;
823 * Returns pointer to the next entry after the one being dropped:
825 union bch_extent_entry *__bch2_bkey_drop_ptr(struct bkey_s k,
826 struct bch_extent_ptr *ptr)
828 struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
829 union bch_extent_entry *entry = to_entry(ptr), *next;
830 union bch_extent_entry *ret = entry;
831 bool drop_crc = true;
833 EBUG_ON(ptr < &ptrs.start->ptr ||
834 ptr >= &ptrs.end->ptr);
835 EBUG_ON(ptr->type != 1 << BCH_EXTENT_ENTRY_ptr);
837 for (next = extent_entry_next(entry);
839 next = extent_entry_next(next)) {
840 if (extent_entry_is_crc(next)) {
842 } else if (extent_entry_is_ptr(next)) {
848 extent_entry_drop(k, entry);
850 while ((entry = extent_entry_prev(ptrs, entry))) {
851 if (extent_entry_is_ptr(entry))
854 if ((extent_entry_is_crc(entry) && drop_crc) ||
855 extent_entry_is_stripe_ptr(entry)) {
856 ret = (void *) ret - extent_entry_bytes(entry);
857 extent_entry_drop(k, entry);
864 union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s k,
865 struct bch_extent_ptr *ptr)
867 bool have_dirty = bch2_bkey_dirty_devs(k.s_c).nr;
868 union bch_extent_entry *ret =
869 __bch2_bkey_drop_ptr(k, ptr);
872 * If we deleted all the dirty pointers and there's still cached
873 * pointers, we could set the cached pointers to dirty if they're not
874 * stale - but to do that correctly we'd need to grab an open_bucket
875 * reference so that we don't race with bucket reuse:
878 !bch2_bkey_dirty_devs(k.s_c).nr) {
879 k.k->type = KEY_TYPE_error;
880 set_bkey_val_u64s(k.k, 0);
882 } else if (!bch2_bkey_nr_ptrs(k.s_c)) {
883 k.k->type = KEY_TYPE_deleted;
884 set_bkey_val_u64s(k.k, 0);
891 void bch2_bkey_drop_device(struct bkey_s k, unsigned dev)
893 struct bch_extent_ptr *ptr;
895 bch2_bkey_drop_ptrs(k, ptr, ptr->dev == dev);
898 const struct bch_extent_ptr *
899 bch2_bkey_has_device(struct bkey_s_c k, unsigned dev)
901 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
902 const struct bch_extent_ptr *ptr;
904 bkey_for_each_ptr(ptrs, ptr)
911 bool bch2_bkey_has_target(struct bch_fs *c, struct bkey_s_c k, unsigned target)
913 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
914 const struct bch_extent_ptr *ptr;
916 bkey_for_each_ptr(ptrs, ptr)
917 if (bch2_dev_in_target(c, ptr->dev, target) &&
919 !ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr)))
925 bool bch2_bkey_matches_ptr(struct bch_fs *c, struct bkey_s_c k,
926 struct bch_extent_ptr m, u64 offset)
928 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
929 const union bch_extent_entry *entry;
930 struct extent_ptr_decoded p;
932 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
933 if (p.ptr.dev == m.dev &&
934 p.ptr.gen == m.gen &&
935 (s64) p.ptr.offset + p.crc.offset - bkey_start_offset(k.k) ==
936 (s64) m.offset - offset)
943 * bch_extent_normalize - clean up an extent, dropping stale pointers etc.
945 * Returns true if @k should be dropped entirely
947 * For existing keys, only called when btree nodes are being rewritten, not when
948 * they're merely being compacted/resorted in memory.
950 bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
952 struct bch_extent_ptr *ptr;
954 bch2_bkey_drop_ptrs(k, ptr,
956 ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr));
958 return bkey_deleted(k.k);
961 void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
964 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
965 const union bch_extent_entry *entry;
966 struct bch_extent_crc_unpacked crc;
967 const struct bch_extent_ptr *ptr;
968 const struct bch_extent_stripe_ptr *ec;
972 bkey_extent_entry_for_each(ptrs, entry) {
974 prt_printf(out, " ");
976 switch (__extent_entry_type(entry)) {
977 case BCH_EXTENT_ENTRY_ptr:
978 ptr = entry_to_ptr(entry);
979 ca = c && ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
980 ? bch_dev_bkey_exists(c, ptr->dev)
984 prt_printf(out, "ptr: %u:%llu gen %u%s", ptr->dev,
985 (u64) ptr->offset, ptr->gen,
986 ptr->cached ? " cached" : "");
989 u64 b = sector_to_bucket_and_offset(ca, ptr->offset, &offset);
991 prt_printf(out, "ptr: %u:%llu:%u gen %u%s", ptr->dev,
993 ptr->cached ? " cached" : "");
995 if (ca && ptr_stale(ca, ptr))
996 prt_printf(out, " stale");
999 case BCH_EXTENT_ENTRY_crc32:
1000 case BCH_EXTENT_ENTRY_crc64:
1001 case BCH_EXTENT_ENTRY_crc128:
1002 crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1004 prt_printf(out, "crc: c_size %u size %u offset %u nonce %u csum %s compress %s",
1005 crc.compressed_size,
1006 crc.uncompressed_size,
1007 crc.offset, crc.nonce,
1008 bch2_csum_types[crc.csum_type],
1009 bch2_compression_types[crc.compression_type]);
1011 case BCH_EXTENT_ENTRY_stripe_ptr:
1012 ec = &entry->stripe_ptr;
1014 prt_printf(out, "ec: idx %llu block %u",
1015 (u64) ec->idx, ec->block);
1018 prt_printf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
1026 static int extent_ptr_invalid(const struct bch_fs *c,
1028 const struct bch_extent_ptr *ptr,
1029 unsigned size_ondisk,
1031 struct printbuf *err)
1033 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1034 const struct bch_extent_ptr *ptr2;
1039 if (!bch2_dev_exists2(c, ptr->dev)) {
1040 prt_printf(err, "pointer to invalid device (%u)", ptr->dev);
1044 ca = bch_dev_bkey_exists(c, ptr->dev);
1045 bkey_for_each_ptr(ptrs, ptr2)
1046 if (ptr != ptr2 && ptr->dev == ptr2->dev) {
1047 prt_printf(err, "multiple pointers to same device (%u)", ptr->dev);
1051 bucket = sector_to_bucket_and_offset(ca, ptr->offset, &bucket_offset);
1053 if (bucket >= ca->mi.nbuckets) {
1054 prt_printf(err, "pointer past last bucket (%llu > %llu)",
1055 bucket, ca->mi.nbuckets);
1059 if (ptr->offset < bucket_to_sector(ca, ca->mi.first_bucket)) {
1060 prt_printf(err, "pointer before first bucket (%llu < %u)",
1061 bucket, ca->mi.first_bucket);
1065 if (bucket_offset + size_ondisk > ca->mi.bucket_size) {
1066 prt_printf(err, "pointer spans multiple buckets (%u + %u > %u)",
1067 bucket_offset, size_ondisk, ca->mi.bucket_size);
1074 int bch2_bkey_ptrs_invalid(const struct bch_fs *c, struct bkey_s_c k,
1075 int rw, struct printbuf *err)
1077 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1078 const union bch_extent_entry *entry;
1079 struct bch_extent_crc_unpacked crc;
1080 unsigned size_ondisk = k.k->size;
1081 unsigned nonce = UINT_MAX;
1084 if (bkey_is_btree_ptr(k.k))
1085 size_ondisk = btree_sectors(c);
1087 bkey_extent_entry_for_each(ptrs, entry) {
1088 if (__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX) {
1089 prt_printf(err, "invalid extent entry type (got %u, max %u)",
1090 __extent_entry_type(entry), BCH_EXTENT_ENTRY_MAX);
1094 if (bkey_is_btree_ptr(k.k) &&
1095 !extent_entry_is_ptr(entry)) {
1096 prt_printf(err, "has non ptr field");
1100 switch (extent_entry_type(entry)) {
1101 case BCH_EXTENT_ENTRY_ptr:
1102 ret = extent_ptr_invalid(c, k, &entry->ptr, size_ondisk,
1107 case BCH_EXTENT_ENTRY_crc32:
1108 case BCH_EXTENT_ENTRY_crc64:
1109 case BCH_EXTENT_ENTRY_crc128:
1110 crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1112 if (crc.offset + crc.live_size >
1113 crc.uncompressed_size) {
1114 prt_printf(err, "checksum offset + key size > uncompressed size");
1118 size_ondisk = crc.compressed_size;
1120 if (!bch2_checksum_type_valid(c, crc.csum_type)) {
1121 prt_printf(err, "invalid checksum type");
1125 if (crc.compression_type >= BCH_COMPRESSION_TYPE_NR) {
1126 prt_printf(err, "invalid compression type");
1130 if (bch2_csum_type_is_encryption(crc.csum_type)) {
1131 if (nonce == UINT_MAX)
1132 nonce = crc.offset + crc.nonce;
1133 else if (nonce != crc.offset + crc.nonce) {
1134 prt_printf(err, "incorrect nonce");
1139 case BCH_EXTENT_ENTRY_stripe_ptr:
1147 void bch2_ptr_swab(struct bkey_s k)
1149 struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1150 union bch_extent_entry *entry;
1153 for (d = (u64 *) ptrs.start;
1154 d != (u64 *) ptrs.end;
1158 for (entry = ptrs.start;
1160 entry = extent_entry_next(entry)) {
1161 switch (extent_entry_type(entry)) {
1162 case BCH_EXTENT_ENTRY_ptr:
1164 case BCH_EXTENT_ENTRY_crc32:
1165 entry->crc32.csum = swab32(entry->crc32.csum);
1167 case BCH_EXTENT_ENTRY_crc64:
1168 entry->crc64.csum_hi = swab16(entry->crc64.csum_hi);
1169 entry->crc64.csum_lo = swab64(entry->crc64.csum_lo);
1171 case BCH_EXTENT_ENTRY_crc128:
1172 entry->crc128.csum.hi = (__force __le64)
1173 swab64((__force u64) entry->crc128.csum.hi);
1174 entry->crc128.csum.lo = (__force __le64)
1175 swab64((__force u64) entry->crc128.csum.lo);
1177 case BCH_EXTENT_ENTRY_stripe_ptr:
1183 /* Generic extent code: */
1185 int bch2_cut_front_s(struct bpos where, struct bkey_s k)
1187 unsigned new_val_u64s = bkey_val_u64s(k.k);
1191 if (bkey_cmp(where, bkey_start_pos(k.k)) <= 0)
1194 EBUG_ON(bkey_cmp(where, k.k->p) > 0);
1196 sub = where.offset - bkey_start_offset(k.k);
1201 k.k->type = KEY_TYPE_deleted;
1205 switch (k.k->type) {
1206 case KEY_TYPE_extent:
1207 case KEY_TYPE_reflink_v: {
1208 struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1209 union bch_extent_entry *entry;
1210 bool seen_crc = false;
1212 bkey_extent_entry_for_each(ptrs, entry) {
1213 switch (extent_entry_type(entry)) {
1214 case BCH_EXTENT_ENTRY_ptr:
1216 entry->ptr.offset += sub;
1218 case BCH_EXTENT_ENTRY_crc32:
1219 entry->crc32.offset += sub;
1221 case BCH_EXTENT_ENTRY_crc64:
1222 entry->crc64.offset += sub;
1224 case BCH_EXTENT_ENTRY_crc128:
1225 entry->crc128.offset += sub;
1227 case BCH_EXTENT_ENTRY_stripe_ptr:
1231 if (extent_entry_is_crc(entry))
1237 case KEY_TYPE_reflink_p: {
1238 struct bkey_s_reflink_p p = bkey_s_to_reflink_p(k);
1240 le64_add_cpu(&p.v->idx, sub);
1243 case KEY_TYPE_inline_data:
1244 case KEY_TYPE_indirect_inline_data: {
1245 void *p = bkey_inline_data_p(k);
1246 unsigned bytes = bkey_inline_data_bytes(k.k);
1248 sub = min_t(u64, sub << 9, bytes);
1250 memmove(p, p + sub, bytes - sub);
1252 new_val_u64s -= sub >> 3;
1257 val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1258 BUG_ON(val_u64s_delta < 0);
1260 set_bkey_val_u64s(k.k, new_val_u64s);
1261 memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1262 return -val_u64s_delta;
1265 int bch2_cut_back_s(struct bpos where, struct bkey_s k)
1267 unsigned new_val_u64s = bkey_val_u64s(k.k);
1271 if (bkey_cmp(where, k.k->p) >= 0)
1274 EBUG_ON(bkey_cmp(where, bkey_start_pos(k.k)) < 0);
1276 len = where.offset - bkey_start_offset(k.k);
1278 k.k->p.offset = where.offset;
1282 k.k->type = KEY_TYPE_deleted;
1286 switch (k.k->type) {
1287 case KEY_TYPE_inline_data:
1288 case KEY_TYPE_indirect_inline_data:
1289 new_val_u64s = (bkey_inline_data_offset(k.k) +
1290 min(bkey_inline_data_bytes(k.k), k.k->size << 9)) >> 3;
1294 val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1295 BUG_ON(val_u64s_delta < 0);
1297 set_bkey_val_u64s(k.k, new_val_u64s);
1298 memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1299 return -val_u64s_delta;