#include "bcachefs.h"
#include "bkey_methods.h"
#include "btree_gc.h"
+#include "btree_io.h"
#include "btree_iter.h"
#include "buckets.h"
#include "checksum.h"
#include "trace.h"
#include "util.h"
-unsigned bch2_bkey_nr_ptrs(struct bkey_s_c k)
-{
- struct bkey_ptrs_c p = bch2_bkey_ptrs_c(k);
- const struct bch_extent_ptr *ptr;
- unsigned nr_ptrs = 0;
-
- bkey_for_each_ptr(p, ptr)
- nr_ptrs++;
-
- return nr_ptrs;
-}
-
-unsigned bch2_bkey_nr_dirty_ptrs(struct bkey_s_c k)
-{
- unsigned nr_ptrs = 0;
-
- switch (k.k->type) {
- case KEY_TYPE_btree_ptr:
- case KEY_TYPE_extent:
- case KEY_TYPE_reflink_v: {
- struct bkey_ptrs_c p = bch2_bkey_ptrs_c(k);
- const struct bch_extent_ptr *ptr;
-
- bkey_for_each_ptr(p, ptr)
- nr_ptrs += !ptr->cached;
- BUG_ON(!nr_ptrs);
- break;
- }
- case KEY_TYPE_reservation:
- nr_ptrs = bkey_s_c_to_reservation(k).v->nr_replicas;
- break;
- }
-
- return nr_ptrs;
-}
-
-static unsigned bch2_extent_ptr_durability(struct bch_fs *c,
- struct extent_ptr_decoded p)
-{
- unsigned durability = 0;
- struct bch_dev *ca;
-
- if (p.ptr.cached)
- return 0;
-
- ca = bch_dev_bkey_exists(c, p.ptr.dev);
-
- if (ca->mi.state != BCH_MEMBER_STATE_FAILED)
- durability = max_t(unsigned, durability, ca->mi.durability);
-
- if (p.has_ec) {
- struct stripe *s =
- genradix_ptr(&c->stripes[0], p.ec.idx);
-
- if (WARN_ON(!s))
- goto out;
-
- durability = max_t(unsigned, durability, s->nr_redundant);
- }
-out:
- return durability;
-}
-
-unsigned bch2_bkey_durability(struct bch_fs *c, struct bkey_s_c k)
-{
- struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
- const union bch_extent_entry *entry;
- struct extent_ptr_decoded p;
- unsigned durability = 0;
-
- bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
- durability += bch2_extent_ptr_durability(c, p);
+static unsigned bch2_crc_field_size_max[] = {
+ [BCH_EXTENT_ENTRY_crc32] = CRC32_SIZE_MAX,
+ [BCH_EXTENT_ENTRY_crc64] = CRC64_SIZE_MAX,
+ [BCH_EXTENT_ENTRY_crc128] = CRC128_SIZE_MAX,
+};
- return durability;
-}
+static void bch2_extent_crc_pack(union bch_extent_crc *,
+ struct bch_extent_crc_unpacked,
+ enum bch_extent_entry_type);
static struct bch_dev_io_failures *dev_io_failures(struct bch_io_failures *f,
unsigned dev)
return bch2_rand_range(l1 + l2) > l1;
}
- if (force_reconstruct_read(c))
+ if (bch2_force_reconstruct_read)
return p1.idx > p2.idx;
return p1.idx < p2.idx;
!bch2_dev_is_readable(ca))
p.idx++;
- if (force_reconstruct_read(c) &&
+ if (bch2_force_reconstruct_read &&
!p.idx && p.has_ec)
p.idx++;
return ret;
}
-void bch2_bkey_append_ptr(struct bkey_i *k,
- struct bch_extent_ptr ptr)
-{
- EBUG_ON(bch2_bkey_has_device(bkey_i_to_s_c(k), ptr.dev));
+/* KEY_TYPE_btree_ptr: */
- switch (k->k.type) {
- case KEY_TYPE_btree_ptr:
- case KEY_TYPE_extent:
- EBUG_ON(bkey_val_u64s(&k->k) >= BKEY_EXTENT_VAL_U64s_MAX);
+const char *bch2_btree_ptr_invalid(const struct bch_fs *c, struct bkey_s_c k)
+{
+ if (bkey_val_u64s(k.k) > BCH_REPLICAS_MAX)
+ return "value too big";
- ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
+ return bch2_bkey_ptrs_invalid(c, k);
+}
- memcpy((void *) &k->v + bkey_val_bytes(&k->k),
- &ptr,
- sizeof(ptr));
- k->u64s++;
- break;
- default:
- BUG();
- }
+void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
+ struct bkey_s_c k)
+{
+ bch2_bkey_ptrs_to_text(out, c, k);
}
-void bch2_bkey_drop_device(struct bkey_s k, unsigned dev)
+const char *bch2_btree_ptr_v2_invalid(const struct bch_fs *c, struct bkey_s_c k)
{
- struct bch_extent_ptr *ptr;
+ struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
- bch2_bkey_drop_ptrs(k, ptr, ptr->dev == dev);
+ if (bkey_val_bytes(k.k) <= sizeof(*bp.v))
+ return "value too small";
+
+ if (bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX)
+ return "value too big";
+
+ if (c->sb.version < bcachefs_metadata_version_snapshot &&
+ bp.v->min_key.snapshot)
+ return "invalid min_key.snapshot";
+
+ return bch2_bkey_ptrs_invalid(c, k);
}
-const struct bch_extent_ptr *
-bch2_bkey_has_device(struct bkey_s_c k, unsigned dev)
+void bch2_btree_ptr_v2_to_text(struct printbuf *out, struct bch_fs *c,
+ struct bkey_s_c k)
{
- struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
- const struct bch_extent_ptr *ptr;
+ struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
- bkey_for_each_ptr(ptrs, ptr)
- if (ptr->dev == dev)
- return ptr;
+ pr_buf(out, "seq %llx written %u min_key ",
+ le64_to_cpu(bp.v->seq),
+ le16_to_cpu(bp.v->sectors_written));
- return NULL;
+ bch2_bpos_to_text(out, bp.v->min_key);
+ pr_buf(out, " ");
+ bch2_bkey_ptrs_to_text(out, c, k);
}
-bool bch2_bkey_has_target(struct bch_fs *c, struct bkey_s_c k, unsigned target)
+void bch2_btree_ptr_v2_compat(enum btree_id btree_id, unsigned version,
+ unsigned big_endian, int write,
+ struct bkey_s k)
{
- struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
- const struct bch_extent_ptr *ptr;
+ struct bkey_s_btree_ptr_v2 bp = bkey_s_to_btree_ptr_v2(k);
- bkey_for_each_ptr(ptrs, ptr)
- if (bch2_dev_in_target(c, ptr->dev, target) &&
- (!ptr->cached ||
- !ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr)))
- return true;
+ compat_bpos(0, btree_id, version, big_endian, write, &bp.v->min_key);
- return false;
+ if (version < bcachefs_metadata_version_inode_btree_change &&
+ btree_node_type_is_extents(btree_id) &&
+ bkey_cmp(bp.v->min_key, POS_MIN))
+ bp.v->min_key = write
+ ? bpos_nosnap_predecessor(bp.v->min_key)
+ : bpos_nosnap_successor(bp.v->min_key);
}
-/* extent specific utility code */
+/* KEY_TYPE_extent: */
-const struct bch_extent_ptr *
-bch2_extent_has_device(struct bkey_s_c_extent e, unsigned dev)
+const char *bch2_extent_invalid(const struct bch_fs *c, struct bkey_s_c k)
{
- const struct bch_extent_ptr *ptr;
-
- extent_for_each_ptr(e, ptr)
- if (ptr->dev == dev)
- return ptr;
-
- return NULL;
+ return bch2_bkey_ptrs_invalid(c, k);
}
-const struct bch_extent_ptr *
-bch2_extent_has_group(struct bch_fs *c, struct bkey_s_c_extent e, unsigned group)
+void bch2_extent_to_text(struct printbuf *out, struct bch_fs *c,
+ struct bkey_s_c k)
{
- const struct bch_extent_ptr *ptr;
+ bch2_bkey_ptrs_to_text(out, c, k);
+}
- extent_for_each_ptr(e, ptr) {
- struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
+enum merge_result bch2_extent_merge(struct bch_fs *c,
+ struct bkey_s _l, struct bkey_s _r)
+{
+ struct bkey_s_extent l = bkey_s_to_extent(_l);
+ struct bkey_s_extent r = bkey_s_to_extent(_r);
+ union bch_extent_entry *en_l = l.v->start;
+ union bch_extent_entry *en_r = r.v->start;
+ struct bch_extent_crc_unpacked crc_l, crc_r;
- if (ca->mi.group &&
- ca->mi.group - 1 == group)
- return ptr;
- }
+ if (bkey_val_u64s(l.k) != bkey_val_u64s(r.k))
+ return BCH_MERGE_NOMERGE;
- return NULL;
-}
+ crc_l = bch2_extent_crc_unpack(l.k, NULL);
-unsigned bch2_extent_is_compressed(struct bkey_s_c k)
-{
- struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
- const union bch_extent_entry *entry;
- struct extent_ptr_decoded p;
- unsigned ret = 0;
+ extent_for_each_entry(l, en_l) {
+ en_r = vstruct_idx(r.v, (u64 *) en_l - l.v->_data);
- bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
- if (!p.ptr.cached &&
- p.crc.compression_type != BCH_COMPRESSION_NONE)
- ret += p.crc.compressed_size;
+ if (extent_entry_type(en_l) != extent_entry_type(en_r))
+ return BCH_MERGE_NOMERGE;
- return ret;
-}
+ switch (extent_entry_type(en_l)) {
+ case BCH_EXTENT_ENTRY_ptr: {
+ const struct bch_extent_ptr *lp = &en_l->ptr;
+ const struct bch_extent_ptr *rp = &en_r->ptr;
+ struct bch_dev *ca;
-bool bch2_bkey_matches_ptr(struct bch_fs *c, struct bkey_s_c k,
- struct bch_extent_ptr m, u64 offset)
-{
- struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
- const union bch_extent_entry *entry;
- struct extent_ptr_decoded p;
+ if (lp->offset + crc_l.compressed_size != rp->offset ||
+ lp->dev != rp->dev ||
+ lp->gen != rp->gen)
+ return BCH_MERGE_NOMERGE;
- bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
- if (p.ptr.dev == m.dev &&
- p.ptr.gen == m.gen &&
- (s64) p.ptr.offset + p.crc.offset - bkey_start_offset(k.k) ==
- (s64) m.offset - offset)
- return true;
+ /* We don't allow extents to straddle buckets: */
+ ca = bch_dev_bkey_exists(c, lp->dev);
- return false;
-}
+ if (PTR_BUCKET_NR(ca, lp) != PTR_BUCKET_NR(ca, rp))
+ return BCH_MERGE_NOMERGE;
-static union bch_extent_entry *extent_entry_prev(struct bkey_ptrs ptrs,
- union bch_extent_entry *entry)
-{
- union bch_extent_entry *i = ptrs.start;
+ break;
+ }
+ case BCH_EXTENT_ENTRY_stripe_ptr:
+ if (en_l->stripe_ptr.block != en_r->stripe_ptr.block ||
+ en_l->stripe_ptr.idx != en_r->stripe_ptr.idx)
+ return BCH_MERGE_NOMERGE;
+ break;
+ case BCH_EXTENT_ENTRY_crc32:
+ case BCH_EXTENT_ENTRY_crc64:
+ case BCH_EXTENT_ENTRY_crc128:
+ crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
+ crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
- if (i == entry)
- return NULL;
+ if (crc_l.csum_type != crc_r.csum_type ||
+ crc_l.compression_type != crc_r.compression_type ||
+ crc_l.nonce != crc_r.nonce)
+ return BCH_MERGE_NOMERGE;
- while (extent_entry_next(i) != entry)
- i = extent_entry_next(i);
- return i;
-}
+ if (crc_l.offset + crc_l.live_size != crc_l.compressed_size ||
+ crc_r.offset)
+ return BCH_MERGE_NOMERGE;
-union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s k,
- struct bch_extent_ptr *ptr)
-{
- struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
- union bch_extent_entry *dst, *src, *prev;
- bool drop_crc = true;
+ if (!bch2_checksum_mergeable(crc_l.csum_type))
+ return BCH_MERGE_NOMERGE;
- EBUG_ON(ptr < &ptrs.start->ptr ||
- ptr >= &ptrs.end->ptr);
- EBUG_ON(ptr->type != 1 << BCH_EXTENT_ENTRY_ptr);
+ if (crc_is_compressed(crc_l))
+ return BCH_MERGE_NOMERGE;
- src = extent_entry_next(to_entry(ptr));
- if (src != ptrs.end &&
- !extent_entry_is_crc(src))
- drop_crc = false;
+ if (crc_l.csum_type &&
+ crc_l.uncompressed_size +
+ crc_r.uncompressed_size > c->sb.encoded_extent_max)
+ return BCH_MERGE_NOMERGE;
- dst = to_entry(ptr);
- while ((prev = extent_entry_prev(ptrs, dst))) {
- if (extent_entry_is_ptr(prev))
- break;
+ if (crc_l.uncompressed_size + crc_r.uncompressed_size >
+ bch2_crc_field_size_max[extent_entry_type(en_l)])
+ return BCH_MERGE_NOMERGE;
- if (extent_entry_is_crc(prev)) {
- if (drop_crc)
- dst = prev;
break;
+ default:
+ return BCH_MERGE_NOMERGE;
}
-
- dst = prev;
}
- memmove_u64s_down(dst, src,
- (u64 *) ptrs.end - (u64 *) src);
- k.k->u64s -= (u64 *) src - (u64 *) dst;
+ extent_for_each_entry(l, en_l) {
+ struct bch_extent_crc_unpacked crc_l, crc_r;
- return dst;
-}
+ en_r = vstruct_idx(r.v, (u64 *) en_l - l.v->_data);
-static inline bool can_narrow_crc(struct bch_extent_crc_unpacked u,
- struct bch_extent_crc_unpacked n)
-{
- return !u.compression_type &&
- u.csum_type &&
- u.uncompressed_size > u.live_size &&
- bch2_csum_type_is_encryption(u.csum_type) ==
+ if (!extent_entry_is_crc(en_l))
+ continue;
+
+ crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
+ crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
+
+ crc_l.csum = bch2_checksum_merge(crc_l.csum_type,
+ crc_l.csum,
+ crc_r.csum,
+ crc_r.uncompressed_size << 9);
+
+ crc_l.uncompressed_size += crc_r.uncompressed_size;
+ crc_l.compressed_size += crc_r.compressed_size;
+
+ bch2_extent_crc_pack(entry_to_crc(en_l), crc_l,
+ extent_entry_type(en_l));
+ }
+
+ bch2_key_resize(l.k, l.k->size + r.k->size);
+
+ return BCH_MERGE_MERGE;
+}
+
+/* KEY_TYPE_reservation: */
+
+const char *bch2_reservation_invalid(const struct bch_fs *c, struct bkey_s_c k)
+{
+ struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
+
+ if (bkey_val_bytes(k.k) != sizeof(struct bch_reservation))
+ return "incorrect value size";
+
+ if (!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX)
+ return "invalid nr_replicas";
+
+ return NULL;
+}
+
+void bch2_reservation_to_text(struct printbuf *out, struct bch_fs *c,
+ struct bkey_s_c k)
+{
+ struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
+
+ pr_buf(out, "generation %u replicas %u",
+ le32_to_cpu(r.v->generation),
+ r.v->nr_replicas);
+}
+
+enum merge_result bch2_reservation_merge(struct bch_fs *c,
+ struct bkey_s _l, struct bkey_s _r)
+{
+ struct bkey_s_reservation l = bkey_s_to_reservation(_l);
+ struct bkey_s_reservation r = bkey_s_to_reservation(_r);
+
+ if (l.v->generation != r.v->generation ||
+ l.v->nr_replicas != r.v->nr_replicas)
+ return BCH_MERGE_NOMERGE;
+
+ if ((u64) l.k->size + r.k->size > KEY_SIZE_MAX) {
+ bch2_key_resize(l.k, KEY_SIZE_MAX);
+ bch2_cut_front_s(l.k->p, r.s);
+ return BCH_MERGE_PARTIAL;
+ }
+
+ bch2_key_resize(l.k, l.k->size + r.k->size);
+
+ return BCH_MERGE_MERGE;
+}
+
+/* Extent checksum entries: */
+
+/* returns true if not equal */
+static inline bool bch2_crc_unpacked_cmp(struct bch_extent_crc_unpacked l,
+ struct bch_extent_crc_unpacked r)
+{
+ return (l.csum_type != r.csum_type ||
+ l.compression_type != r.compression_type ||
+ l.compressed_size != r.compressed_size ||
+ l.uncompressed_size != r.uncompressed_size ||
+ l.offset != r.offset ||
+ l.live_size != r.live_size ||
+ l.nonce != r.nonce ||
+ bch2_crc_cmp(l.csum, r.csum));
+}
+
+static inline bool can_narrow_crc(struct bch_extent_crc_unpacked u,
+ struct bch_extent_crc_unpacked n)
+{
+ return !crc_is_compressed(u) &&
+ u.csum_type &&
+ u.uncompressed_size > u.live_size &&
+ bch2_csum_type_is_encryption(u.csum_type) ==
bch2_csum_type_is_encryption(n.csum_type);
}
/* Find a checksum entry that covers only live data: */
if (!n.csum_type) {
bkey_for_each_crc(&k->k, ptrs, u, i)
- if (!u.compression_type &&
+ if (!crc_is_compressed(u) &&
u.csum_type &&
u.live_size == u.uncompressed_size) {
n = u;
return false;
}
found:
- BUG_ON(n.compression_type);
+ BUG_ON(crc_is_compressed(n));
BUG_ON(n.offset);
BUG_ON(n.live_size != k->k.size);
return ret;
}
-/* returns true if not equal */
-static inline bool bch2_crc_unpacked_cmp(struct bch_extent_crc_unpacked l,
- struct bch_extent_crc_unpacked r)
+static void bch2_extent_crc_pack(union bch_extent_crc *dst,
+ struct bch_extent_crc_unpacked src,
+ enum bch_extent_entry_type type)
{
- return (l.csum_type != r.csum_type ||
- l.compression_type != r.compression_type ||
- l.compressed_size != r.compressed_size ||
- l.uncompressed_size != r.uncompressed_size ||
- l.offset != r.offset ||
- l.live_size != r.live_size ||
- l.nonce != r.nonce ||
- bch2_crc_cmp(l.csum, r.csum));
+#define set_common_fields(_dst, _src) \
+ _dst.type = 1 << type; \
+ _dst.csum_type = _src.csum_type, \
+ _dst.compression_type = _src.compression_type, \
+ _dst._compressed_size = _src.compressed_size - 1, \
+ _dst._uncompressed_size = _src.uncompressed_size - 1, \
+ _dst.offset = _src.offset
+
+ switch (type) {
+ case BCH_EXTENT_ENTRY_crc32:
+ set_common_fields(dst->crc32, src);
+ dst->crc32.csum = *((__le32 *) &src.csum.lo);
+ break;
+ case BCH_EXTENT_ENTRY_crc64:
+ set_common_fields(dst->crc64, src);
+ dst->crc64.nonce = src.nonce;
+ dst->crc64.csum_lo = src.csum.lo;
+ dst->crc64.csum_hi = *((__le16 *) &src.csum.hi);
+ break;
+ case BCH_EXTENT_ENTRY_crc128:
+ set_common_fields(dst->crc128, src);
+ dst->crc128.nonce = src.nonce;
+ dst->crc128.csum = src.csum;
+ break;
+ default:
+ BUG();
+ }
+#undef set_common_fields
}
-void bch2_ptr_swab(const struct bkey_format *f, struct bkey_packed *k)
+void bch2_extent_crc_append(struct bkey_i *k,
+ struct bch_extent_crc_unpacked new)
{
- union bch_extent_entry *entry;
- u64 *d = (u64 *) bkeyp_val(f, k);
- unsigned i;
+ struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
+ union bch_extent_crc *crc = (void *) ptrs.end;
+ enum bch_extent_entry_type type;
- for (i = 0; i < bkeyp_val_u64s(f, k); i++)
- d[i] = swab64(d[i]);
+ if (bch_crc_bytes[new.csum_type] <= 4 &&
+ new.uncompressed_size <= CRC32_SIZE_MAX &&
+ new.nonce <= CRC32_NONCE_MAX)
+ type = BCH_EXTENT_ENTRY_crc32;
+ else if (bch_crc_bytes[new.csum_type] <= 10 &&
+ new.uncompressed_size <= CRC64_SIZE_MAX &&
+ new.nonce <= CRC64_NONCE_MAX)
+ type = BCH_EXTENT_ENTRY_crc64;
+ else if (bch_crc_bytes[new.csum_type] <= 16 &&
+ new.uncompressed_size <= CRC128_SIZE_MAX &&
+ new.nonce <= CRC128_NONCE_MAX)
+ type = BCH_EXTENT_ENTRY_crc128;
+ else
+ BUG();
- for (entry = (union bch_extent_entry *) d;
- entry < (union bch_extent_entry *) (d + bkeyp_val_u64s(f, k));
- entry = extent_entry_next(entry)) {
- switch (extent_entry_type(entry)) {
- case BCH_EXTENT_ENTRY_ptr:
- break;
- case BCH_EXTENT_ENTRY_crc32:
- entry->crc32.csum = swab32(entry->crc32.csum);
- break;
- case BCH_EXTENT_ENTRY_crc64:
- entry->crc64.csum_hi = swab16(entry->crc64.csum_hi);
- entry->crc64.csum_lo = swab64(entry->crc64.csum_lo);
- break;
- case BCH_EXTENT_ENTRY_crc128:
- entry->crc128.csum.hi = (__force __le64)
- swab64((__force u64) entry->crc128.csum.hi);
- entry->crc128.csum.lo = (__force __le64)
- swab64((__force u64) entry->crc128.csum.lo);
- break;
- case BCH_EXTENT_ENTRY_stripe_ptr:
- break;
- }
- }
-}
+ bch2_extent_crc_pack(crc, new, type);
-void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
- struct bkey_s_c k)
-{
- struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
- const union bch_extent_entry *entry;
- struct bch_extent_crc_unpacked crc;
- const struct bch_extent_ptr *ptr;
- const struct bch_extent_stripe_ptr *ec;
- struct bch_dev *ca;
- bool first = true;
+ k->k.u64s += extent_entry_u64s(ptrs.end);
- bkey_extent_entry_for_each(ptrs, entry) {
- if (!first)
- pr_buf(out, " ");
+ EBUG_ON(bkey_val_u64s(&k->k) > BKEY_EXTENT_VAL_U64s_MAX);
+}
- switch (__extent_entry_type(entry)) {
- case BCH_EXTENT_ENTRY_ptr:
- ptr = entry_to_ptr(entry);
- ca = ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
- ? bch_dev_bkey_exists(c, ptr->dev)
- : NULL;
+/* Generic code for keys with pointers: */
- pr_buf(out, "ptr: %u:%llu gen %u%s%s", ptr->dev,
- (u64) ptr->offset, ptr->gen,
- ptr->cached ? " cached" : "",
- ca && ptr_stale(ca, ptr)
- ? " stale" : "");
- break;
- case BCH_EXTENT_ENTRY_crc32:
- case BCH_EXTENT_ENTRY_crc64:
- case BCH_EXTENT_ENTRY_crc128:
- crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
+unsigned bch2_bkey_nr_ptrs(struct bkey_s_c k)
+{
+ return bch2_bkey_devs(k).nr;
+}
- pr_buf(out, "crc: c_size %u size %u offset %u nonce %u csum %u compress %u",
- crc.compressed_size,
- crc.uncompressed_size,
- crc.offset, crc.nonce,
- crc.csum_type,
- crc.compression_type);
- break;
- case BCH_EXTENT_ENTRY_stripe_ptr:
- ec = &entry->stripe_ptr;
+unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c k)
+{
+ return k.k->type == KEY_TYPE_reservation
+ ? bkey_s_c_to_reservation(k).v->nr_replicas
+ : bch2_bkey_dirty_devs(k).nr;
+}
- pr_buf(out, "ec: idx %llu block %u",
- (u64) ec->idx, ec->block);
- break;
- default:
- pr_buf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
- return;
- }
+unsigned bch2_bkey_nr_ptrs_fully_allocated(struct bkey_s_c k)
+{
+ unsigned ret = 0;
- first = false;
+ if (k.k->type == KEY_TYPE_reservation) {
+ ret = bkey_s_c_to_reservation(k).v->nr_replicas;
+ } else {
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ const union bch_extent_entry *entry;
+ struct extent_ptr_decoded p;
+
+ bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
+ ret += !p.ptr.cached && !crc_is_compressed(p.crc);
}
+
+ return ret;
}
-static const char *extent_ptr_invalid(const struct bch_fs *c,
- struct bkey_s_c k,
- const struct bch_extent_ptr *ptr,
- unsigned size_ondisk,
- bool metadata)
+unsigned bch2_bkey_sectors_compressed(struct bkey_s_c k)
{
struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
- const struct bch_extent_ptr *ptr2;
- struct bch_dev *ca;
+ const union bch_extent_entry *entry;
+ struct extent_ptr_decoded p;
+ unsigned ret = 0;
- if (!bch2_dev_exists2(c, ptr->dev))
- return "pointer to invalid device";
+ bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
+ if (!p.ptr.cached && crc_is_compressed(p.crc))
+ ret += p.crc.compressed_size;
- ca = bch_dev_bkey_exists(c, ptr->dev);
- if (!ca)
- return "pointer to invalid device";
+ return ret;
+}
- bkey_for_each_ptr(ptrs, ptr2)
- if (ptr != ptr2 && ptr->dev == ptr2->dev)
- return "multiple pointers to same device";
-
- if (ptr->offset + size_ondisk > bucket_to_sector(ca, ca->mi.nbuckets))
- return "offset past end of device";
-
- if (ptr->offset < bucket_to_sector(ca, ca->mi.first_bucket))
- return "offset before first bucket";
-
- if (bucket_remainder(ca, ptr->offset) +
- size_ondisk > ca->mi.bucket_size)
- return "spans multiple buckets";
-
- return NULL;
-}
-
-const char *bch2_bkey_ptrs_invalid(const struct bch_fs *c, struct bkey_s_c k)
+bool bch2_bkey_is_incompressible(struct bkey_s_c k)
{
struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
const union bch_extent_entry *entry;
struct bch_extent_crc_unpacked crc;
- unsigned size_ondisk = k.k->size;
- const char *reason;
- unsigned nonce = UINT_MAX;
-
- if (k.k->type == KEY_TYPE_btree_ptr)
- size_ondisk = c->opts.btree_node_size;
-
- bkey_extent_entry_for_each(ptrs, entry) {
- if (__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX)
- return "invalid extent entry type";
-
- if (k.k->type == KEY_TYPE_btree_ptr &&
- !extent_entry_is_ptr(entry))
- return "has non ptr field";
-
- switch (extent_entry_type(entry)) {
- case BCH_EXTENT_ENTRY_ptr:
- reason = extent_ptr_invalid(c, k, &entry->ptr,
- size_ondisk, false);
- if (reason)
- return reason;
- break;
- case BCH_EXTENT_ENTRY_crc32:
- case BCH_EXTENT_ENTRY_crc64:
- case BCH_EXTENT_ENTRY_crc128:
- crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
- if (crc.offset + crc.live_size >
- crc.uncompressed_size)
- return "checksum offset + key size > uncompressed size";
+ bkey_for_each_crc(k.k, ptrs, crc, entry)
+ if (crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
+ return true;
+ return false;
+}
- size_ondisk = crc.compressed_size;
+bool bch2_check_range_allocated(struct bch_fs *c, struct bpos pos, u64 size,
+ unsigned nr_replicas, bool compressed)
+{
+ struct btree_trans trans;
+ struct btree_iter *iter;
+ struct bpos end = pos;
+ struct bkey_s_c k;
+ bool ret = true;
+ int err;
- if (!bch2_checksum_type_valid(c, crc.csum_type))
- return "invalid checksum type";
+ end.offset += size;
- if (crc.compression_type >= BCH_COMPRESSION_NR)
- return "invalid compression type";
+ bch2_trans_init(&trans, c, 0, 0);
- if (bch2_csum_type_is_encryption(crc.csum_type)) {
- if (nonce == UINT_MAX)
- nonce = crc.offset + crc.nonce;
- else if (nonce != crc.offset + crc.nonce)
- return "incorrect nonce";
- }
+ for_each_btree_key(&trans, iter, BTREE_ID_extents, pos,
+ BTREE_ITER_SLOTS, k, err) {
+ if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
break;
- case BCH_EXTENT_ENTRY_stripe_ptr:
+
+ if (nr_replicas > bch2_bkey_replicas(c, k) ||
+ (!compressed && bch2_bkey_sectors_compressed(k))) {
+ ret = false;
break;
}
}
+ bch2_trans_iter_put(&trans, iter);
- return NULL;
-}
-
-/* Btree ptrs */
-
-const char *bch2_btree_ptr_invalid(const struct bch_fs *c, struct bkey_s_c k)
-{
- if (bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX)
- return "value too big";
+ bch2_trans_exit(&trans);
- return bch2_bkey_ptrs_invalid(c, k);
+ return ret;
}
-void bch2_btree_ptr_debugcheck(struct bch_fs *c, struct bkey_s_c k)
+unsigned bch2_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
{
struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
- const struct bch_extent_ptr *ptr;
- const char *err;
- char buf[160];
- struct bucket_mark mark;
- struct bch_dev *ca;
-
- bch2_fs_bug_on(!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) &&
- !bch2_bkey_replicas_marked(c, k, false), c,
- "btree key bad (replicas not marked in superblock):\n%s",
- (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf));
-
- if (!test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags))
- return;
+ const union bch_extent_entry *entry;
+ struct extent_ptr_decoded p = { 0 };
+ unsigned replicas = 0;
- bkey_for_each_ptr(ptrs, ptr) {
- ca = bch_dev_bkey_exists(c, ptr->dev);
+ bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
+ if (p.ptr.cached)
+ continue;
- mark = ptr_bucket_mark(ca, ptr);
+ if (p.has_ec)
+ replicas += p.ec.redundancy;
- err = "stale";
- if (gen_after(mark.gen, ptr->gen))
- goto err;
+ replicas++;
- err = "inconsistent";
- if (mark.data_type != BCH_DATA_BTREE ||
- mark.dirty_sectors < c->opts.btree_node_size)
- goto err;
}
- return;
-err:
- bch2_bkey_val_to_text(&PBUF(buf), c, k);
- bch2_fs_bug(c, "%s btree pointer %s: bucket %zi gen %i mark %08x",
- err, buf, PTR_BUCKET_NR(ca, ptr),
- mark.gen, (unsigned) mark.v.counter);
-}
-
-void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
- struct bkey_s_c k)
-{
- bch2_bkey_ptrs_to_text(out, c, k);
+ return replicas;
}
-/* Extents */
-
-int bch2_cut_front_s(struct bpos where, struct bkey_s k)
+static unsigned bch2_extent_ptr_durability(struct bch_fs *c,
+ struct extent_ptr_decoded p)
{
- unsigned new_val_u64s = bkey_val_u64s(k.k);
- int val_u64s_delta;
- u64 sub;
+ unsigned durability = 0;
+ struct bch_dev *ca;
- if (bkey_cmp(where, bkey_start_pos(k.k)) <= 0)
+ if (p.ptr.cached)
return 0;
- EBUG_ON(bkey_cmp(where, k.k->p) > 0);
+ ca = bch_dev_bkey_exists(c, p.ptr.dev);
- sub = where.offset - bkey_start_offset(k.k);
+ if (ca->mi.state != BCH_MEMBER_STATE_failed)
+ durability = max_t(unsigned, durability, ca->mi.durability);
- k.k->size -= sub;
+ if (p.has_ec)
+ durability += p.ec.redundancy;
- if (!k.k->size) {
- k.k->type = KEY_TYPE_deleted;
- new_val_u64s = 0;
- }
+ return durability;
+}
- switch (k.k->type) {
- case KEY_TYPE_deleted:
- case KEY_TYPE_discard:
- case KEY_TYPE_error:
- case KEY_TYPE_cookie:
- break;
- case KEY_TYPE_extent:
- case KEY_TYPE_reflink_v: {
- struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
- union bch_extent_entry *entry;
- bool seen_crc = false;
+unsigned bch2_bkey_durability(struct bch_fs *c, struct bkey_s_c k)
+{
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ const union bch_extent_entry *entry;
+ struct extent_ptr_decoded p;
+ unsigned durability = 0;
- bkey_extent_entry_for_each(ptrs, entry) {
- switch (extent_entry_type(entry)) {
- case BCH_EXTENT_ENTRY_ptr:
- if (!seen_crc)
- entry->ptr.offset += sub;
- break;
- case BCH_EXTENT_ENTRY_crc32:
- entry->crc32.offset += sub;
- break;
- case BCH_EXTENT_ENTRY_crc64:
- entry->crc64.offset += sub;
- break;
- case BCH_EXTENT_ENTRY_crc128:
- entry->crc128.offset += sub;
- break;
- case BCH_EXTENT_ENTRY_stripe_ptr:
- break;
- }
+ bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
+ durability += bch2_extent_ptr_durability(c, p);
- if (extent_entry_is_crc(entry))
- seen_crc = true;
- }
+ return durability;
+}
- break;
- }
- case KEY_TYPE_reflink_p: {
- struct bkey_s_reflink_p p = bkey_s_to_reflink_p(k);
+void bch2_bkey_mark_replicas_cached(struct bch_fs *c, struct bkey_s k,
+ unsigned target,
+ unsigned nr_desired_replicas)
+{
+ struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
+ union bch_extent_entry *entry;
+ struct extent_ptr_decoded p;
+ int extra = bch2_bkey_durability(c, k.s_c) - nr_desired_replicas;
- le64_add_cpu(&p.v->idx, sub);
- break;
- }
- case KEY_TYPE_reservation:
- break;
- default:
- BUG();
- }
+ if (target && extra > 0)
+ bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
+ int n = bch2_extent_ptr_durability(c, p);
- val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
- BUG_ON(val_u64s_delta < 0);
+ if (n && n <= extra &&
+ !bch2_dev_in_target(c, p.ptr.dev, target)) {
+ entry->ptr.cached = true;
+ extra -= n;
+ }
+ }
- set_bkey_val_u64s(k.k, new_val_u64s);
- memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
- return -val_u64s_delta;
+ if (extra > 0)
+ bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
+ int n = bch2_extent_ptr_durability(c, p);
+
+ if (n && n <= extra) {
+ entry->ptr.cached = true;
+ extra -= n;
+ }
+ }
}
-int bch2_cut_back_s(struct bpos where, struct bkey_s k)
+void bch2_bkey_extent_entry_drop(struct bkey_i *k, union bch_extent_entry *entry)
{
- unsigned new_val_u64s = bkey_val_u64s(k.k);
- int val_u64s_delta;
- u64 len = 0;
+ union bch_extent_entry *end = bkey_val_end(bkey_i_to_s(k));
+ union bch_extent_entry *next = extent_entry_next(entry);
- if (bkey_cmp(where, k.k->p) >= 0)
- return 0;
+ memmove_u64s(entry, next, (u64 *) end - (u64 *) next);
+ k->k.u64s -= extent_entry_u64s(entry);
+}
- EBUG_ON(bkey_cmp(where, bkey_start_pos(k.k)) < 0);
+void bch2_bkey_append_ptr(struct bkey_i *k,
+ struct bch_extent_ptr ptr)
+{
+ EBUG_ON(bch2_bkey_has_device(bkey_i_to_s_c(k), ptr.dev));
- len = where.offset - bkey_start_offset(k.k);
+ switch (k->k.type) {
+ case KEY_TYPE_btree_ptr:
+ case KEY_TYPE_btree_ptr_v2:
+ case KEY_TYPE_extent:
+ EBUG_ON(bkey_val_u64s(&k->k) >= BKEY_EXTENT_VAL_U64s_MAX);
- k.k->p = where;
- k.k->size = len;
+ ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
- if (!len) {
- k.k->type = KEY_TYPE_deleted;
- new_val_u64s = 0;
+ memcpy((void *) &k->v + bkey_val_bytes(&k->k),
+ &ptr,
+ sizeof(ptr));
+ k->u64s++;
+ break;
+ default:
+ BUG();
}
-
- val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
- BUG_ON(val_u64s_delta < 0);
-
- set_bkey_val_u64s(k.k, new_val_u64s);
- memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
- return -val_u64s_delta;
}
-const char *bch2_extent_invalid(const struct bch_fs *c, struct bkey_s_c k)
+static inline void __extent_entry_insert(struct bkey_i *k,
+ union bch_extent_entry *dst,
+ union bch_extent_entry *new)
{
- return bch2_bkey_ptrs_invalid(c, k);
-}
+ union bch_extent_entry *end = bkey_val_end(bkey_i_to_s(k));
-void bch2_extent_debugcheck(struct bch_fs *c, struct bkey_s_c k)
-{
- struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
- const union bch_extent_entry *entry;
- struct extent_ptr_decoded p;
- char buf[160];
-
- /*
- * XXX: we should be doing most/all of these checks at startup time,
- * where we check bch2_bkey_invalid() in btree_node_read_done()
- *
- * But note that we can't check for stale pointers or incorrect gc marks
- * until after journal replay is done (it might be an extent that's
- * going to get overwritten during replay)
- */
-
- if (percpu_down_read_trylock(&c->mark_lock)) {
- bch2_fs_bug_on(!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) &&
- !bch2_bkey_replicas_marked_locked(c, e.s_c, false), c,
- "extent key bad (replicas not marked in superblock):\n%s",
- (bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c), buf));
- percpu_up_read(&c->mark_lock);
- }
- /*
- * If journal replay hasn't finished, we might be seeing keys
- * that will be overwritten by the time journal replay is done:
- */
- if (!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
- return;
-
- extent_for_each_ptr_decode(e, p, entry) {
- struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
- struct bucket_mark mark = ptr_bucket_mark(ca, &p.ptr);
- unsigned stale = gen_after(mark.gen, p.ptr.gen);
- unsigned disk_sectors = ptr_disk_sectors(p);
- unsigned mark_sectors = p.ptr.cached
- ? mark.cached_sectors
- : mark.dirty_sectors;
-
- bch2_fs_bug_on(stale && !p.ptr.cached, c,
- "stale dirty pointer (ptr gen %u bucket %u",
- p.ptr.gen, mark.gen);
-
- bch2_fs_bug_on(stale > 96, c, "key too stale: %i", stale);
-
- bch2_fs_bug_on(!stale &&
- (mark.data_type != BCH_DATA_USER ||
- mark_sectors < disk_sectors), c,
- "extent pointer not marked: %s:\n"
- "type %u sectors %u < %u",
- (bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c), buf),
- mark.data_type,
- mark_sectors, disk_sectors);
- }
-}
-
-void bch2_extent_to_text(struct printbuf *out, struct bch_fs *c,
- struct bkey_s_c k)
-{
- bch2_bkey_ptrs_to_text(out, c, k);
-}
-
-static unsigned bch2_crc_field_size_max[] = {
- [BCH_EXTENT_ENTRY_crc32] = CRC32_SIZE_MAX,
- [BCH_EXTENT_ENTRY_crc64] = CRC64_SIZE_MAX,
- [BCH_EXTENT_ENTRY_crc128] = CRC128_SIZE_MAX,
-};
-
-static void bch2_extent_crc_pack(union bch_extent_crc *dst,
- struct bch_extent_crc_unpacked src,
- enum bch_extent_entry_type type)
-{
-#define set_common_fields(_dst, _src) \
- _dst.type = 1 << type; \
- _dst.csum_type = _src.csum_type, \
- _dst.compression_type = _src.compression_type, \
- _dst._compressed_size = _src.compressed_size - 1, \
- _dst._uncompressed_size = _src.uncompressed_size - 1, \
- _dst.offset = _src.offset
-
- switch (type) {
- case BCH_EXTENT_ENTRY_crc32:
- set_common_fields(dst->crc32, src);
- dst->crc32.csum = *((__le32 *) &src.csum.lo);
- break;
- case BCH_EXTENT_ENTRY_crc64:
- set_common_fields(dst->crc64, src);
- dst->crc64.nonce = src.nonce;
- dst->crc64.csum_lo = src.csum.lo;
- dst->crc64.csum_hi = *((__le16 *) &src.csum.hi);
- break;
- case BCH_EXTENT_ENTRY_crc128:
- set_common_fields(dst->crc128, src);
- dst->crc128.nonce = src.nonce;
- dst->crc128.csum = src.csum;
- break;
- default:
- BUG();
- }
-#undef set_common_fields
-}
-
-void bch2_extent_crc_append(struct bkey_i *k,
- struct bch_extent_crc_unpacked new)
-{
- struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
- union bch_extent_crc *crc = (void *) ptrs.end;
- enum bch_extent_entry_type type;
-
- if (bch_crc_bytes[new.csum_type] <= 4 &&
- new.uncompressed_size - 1 <= CRC32_SIZE_MAX &&
- new.nonce <= CRC32_NONCE_MAX)
- type = BCH_EXTENT_ENTRY_crc32;
- else if (bch_crc_bytes[new.csum_type] <= 10 &&
- new.uncompressed_size - 1 <= CRC64_SIZE_MAX &&
- new.nonce <= CRC64_NONCE_MAX)
- type = BCH_EXTENT_ENTRY_crc64;
- else if (bch_crc_bytes[new.csum_type] <= 16 &&
- new.uncompressed_size - 1 <= CRC128_SIZE_MAX &&
- new.nonce <= CRC128_NONCE_MAX)
- type = BCH_EXTENT_ENTRY_crc128;
- else
- BUG();
-
- bch2_extent_crc_pack(crc, new, type);
-
- k->k.u64s += extent_entry_u64s(ptrs.end);
-
- EBUG_ON(bkey_val_u64s(&k->k) > BKEY_EXTENT_VAL_U64s_MAX);
-}
-
-static inline void __extent_entry_insert(struct bkey_i *k,
- union bch_extent_entry *dst,
- union bch_extent_entry *new)
-{
- union bch_extent_entry *end = bkey_val_end(bkey_i_to_s(k));
-
- memmove_u64s_up_small((u64 *) dst + extent_entry_u64s(new),
- dst, (u64 *) end - (u64 *) dst);
- k->k.u64s += extent_entry_u64s(new);
- memcpy_u64s_small(dst, new, extent_entry_u64s(new));
+ memmove_u64s_up_small((u64 *) dst + extent_entry_u64s(new),
+ dst, (u64 *) end - (u64 *) dst);
+ k->k.u64s += extent_entry_u64s(new);
+ memcpy_u64s_small(dst, new, extent_entry_u64s(new));
}
void bch2_extent_ptr_decoded_append(struct bkey_i *k,
}
}
-/*
- * bch_extent_normalize - clean up an extent, dropping stale pointers etc.
- *
- * Returns true if @k should be dropped entirely
- *
- * For existing keys, only called when btree nodes are being rewritten, not when
- * they're merely being compacted/resorted in memory.
- */
-bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
+static union bch_extent_entry *extent_entry_prev(struct bkey_ptrs ptrs,
+ union bch_extent_entry *entry)
{
- struct bch_extent_ptr *ptr;
-
- bch2_bkey_drop_ptrs(k, ptr,
- ptr->cached &&
- ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr));
+ union bch_extent_entry *i = ptrs.start;
- /* will only happen if all pointers were cached: */
- if (!bkey_val_u64s(k.k))
- k.k->type = KEY_TYPE_discard;
+ if (i == entry)
+ return NULL;
- return bkey_whiteout(k.k);
+ while (extent_entry_next(i) != entry)
+ i = extent_entry_next(i);
+ return i;
}
-void bch2_bkey_mark_replicas_cached(struct bch_fs *c, struct bkey_s k,
- unsigned target,
- unsigned nr_desired_replicas)
+union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s k,
+ struct bch_extent_ptr *ptr)
{
struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
- union bch_extent_entry *entry;
- struct extent_ptr_decoded p;
- int extra = bch2_bkey_durability(c, k.s_c) - nr_desired_replicas;
+ union bch_extent_entry *dst, *src, *prev;
+ bool drop_crc = true;
- if (target && extra > 0)
- bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
- int n = bch2_extent_ptr_durability(c, p);
+ EBUG_ON(ptr < &ptrs.start->ptr ||
+ ptr >= &ptrs.end->ptr);
+ EBUG_ON(ptr->type != 1 << BCH_EXTENT_ENTRY_ptr);
- if (n && n <= extra &&
- !bch2_dev_in_target(c, p.ptr.dev, target)) {
- entry->ptr.cached = true;
- extra -= n;
- }
- }
+ src = extent_entry_next(to_entry(ptr));
+ if (src != ptrs.end &&
+ !extent_entry_is_crc(src))
+ drop_crc = false;
- if (extra > 0)
- bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
- int n = bch2_extent_ptr_durability(c, p);
+ dst = to_entry(ptr);
+ while ((prev = extent_entry_prev(ptrs, dst))) {
+ if (extent_entry_is_ptr(prev))
+ break;
- if (n && n <= extra) {
- entry->ptr.cached = true;
- extra -= n;
- }
+ if (extent_entry_is_crc(prev)) {
+ if (drop_crc)
+ dst = prev;
+ break;
}
+
+ dst = prev;
+ }
+
+ memmove_u64s_down(dst, src,
+ (u64 *) ptrs.end - (u64 *) src);
+ k.k->u64s -= (u64 *) src - (u64 *) dst;
+
+ return dst;
}
-enum merge_result bch2_extent_merge(struct bch_fs *c,
- struct bkey_s _l, struct bkey_s _r)
+void bch2_bkey_drop_device(struct bkey_s k, unsigned dev)
{
- struct bkey_s_extent l = bkey_s_to_extent(_l);
- struct bkey_s_extent r = bkey_s_to_extent(_r);
- union bch_extent_entry *en_l = l.v->start;
- union bch_extent_entry *en_r = r.v->start;
- struct bch_extent_crc_unpacked crc_l, crc_r;
+ struct bch_extent_ptr *ptr;
- if (bkey_val_u64s(l.k) != bkey_val_u64s(r.k))
- return BCH_MERGE_NOMERGE;
+ bch2_bkey_drop_ptrs(k, ptr, ptr->dev == dev);
+}
- crc_l = bch2_extent_crc_unpack(l.k, NULL);
+const struct bch_extent_ptr *
+bch2_bkey_has_device(struct bkey_s_c k, unsigned dev)
+{
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ const struct bch_extent_ptr *ptr;
- extent_for_each_entry(l, en_l) {
- en_r = vstruct_idx(r.v, (u64 *) en_l - l.v->_data);
+ bkey_for_each_ptr(ptrs, ptr)
+ if (ptr->dev == dev)
+ return ptr;
- if (extent_entry_type(en_l) != extent_entry_type(en_r))
- return BCH_MERGE_NOMERGE;
+ return NULL;
+}
- switch (extent_entry_type(en_l)) {
- case BCH_EXTENT_ENTRY_ptr: {
- const struct bch_extent_ptr *lp = &en_l->ptr;
- const struct bch_extent_ptr *rp = &en_r->ptr;
- struct bch_dev *ca;
+bool bch2_bkey_has_target(struct bch_fs *c, struct bkey_s_c k, unsigned target)
+{
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ const struct bch_extent_ptr *ptr;
- if (lp->offset + crc_l.compressed_size != rp->offset ||
- lp->dev != rp->dev ||
- lp->gen != rp->gen)
- return BCH_MERGE_NOMERGE;
+ bkey_for_each_ptr(ptrs, ptr)
+ if (bch2_dev_in_target(c, ptr->dev, target) &&
+ (!ptr->cached ||
+ !ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr)))
+ return true;
- /* We don't allow extents to straddle buckets: */
- ca = bch_dev_bkey_exists(c, lp->dev);
+ return false;
+}
- if (PTR_BUCKET_NR(ca, lp) != PTR_BUCKET_NR(ca, rp))
- return BCH_MERGE_NOMERGE;
+bool bch2_bkey_matches_ptr(struct bch_fs *c, struct bkey_s_c k,
+ struct bch_extent_ptr m, u64 offset)
+{
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ const union bch_extent_entry *entry;
+ struct extent_ptr_decoded p;
- break;
- }
- case BCH_EXTENT_ENTRY_stripe_ptr:
- if (en_l->stripe_ptr.block != en_r->stripe_ptr.block ||
- en_l->stripe_ptr.idx != en_r->stripe_ptr.idx)
- return BCH_MERGE_NOMERGE;
- break;
- case BCH_EXTENT_ENTRY_crc32:
- case BCH_EXTENT_ENTRY_crc64:
- case BCH_EXTENT_ENTRY_crc128:
- crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
- crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
+ bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
+ if (p.ptr.dev == m.dev &&
+ p.ptr.gen == m.gen &&
+ (s64) p.ptr.offset + p.crc.offset - bkey_start_offset(k.k) ==
+ (s64) m.offset - offset)
+ return true;
- if (crc_l.csum_type != crc_r.csum_type ||
- crc_l.compression_type != crc_r.compression_type ||
- crc_l.nonce != crc_r.nonce)
- return BCH_MERGE_NOMERGE;
+ return false;
+}
- if (crc_l.offset + crc_l.live_size != crc_l.compressed_size ||
- crc_r.offset)
- return BCH_MERGE_NOMERGE;
+/*
+ * bch_extent_normalize - clean up an extent, dropping stale pointers etc.
+ *
+ * Returns true if @k should be dropped entirely
+ *
+ * For existing keys, only called when btree nodes are being rewritten, not when
+ * they're merely being compacted/resorted in memory.
+ */
+bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
+{
+ struct bch_extent_ptr *ptr;
- if (!bch2_checksum_mergeable(crc_l.csum_type))
- return BCH_MERGE_NOMERGE;
+ bch2_bkey_drop_ptrs(k, ptr,
+ ptr->cached &&
+ ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr));
- if (crc_l.compression_type)
- return BCH_MERGE_NOMERGE;
+ /* will only happen if all pointers were cached: */
+ if (!bch2_bkey_nr_ptrs(k.s_c))
+ k.k->type = KEY_TYPE_deleted;
- if (crc_l.csum_type &&
- crc_l.uncompressed_size +
- crc_r.uncompressed_size > c->sb.encoded_extent_max)
- return BCH_MERGE_NOMERGE;
+ return bkey_deleted(k.k);
+}
- if (crc_l.uncompressed_size + crc_r.uncompressed_size - 1 >
- bch2_crc_field_size_max[extent_entry_type(en_l)])
- return BCH_MERGE_NOMERGE;
+void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
+ struct bkey_s_c k)
+{
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ const union bch_extent_entry *entry;
+ struct bch_extent_crc_unpacked crc;
+ const struct bch_extent_ptr *ptr;
+ const struct bch_extent_stripe_ptr *ec;
+ struct bch_dev *ca;
+ bool first = true;
+
+ bkey_extent_entry_for_each(ptrs, entry) {
+ if (!first)
+ pr_buf(out, " ");
+
+ switch (__extent_entry_type(entry)) {
+ case BCH_EXTENT_ENTRY_ptr:
+ ptr = entry_to_ptr(entry);
+ ca = ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
+ ? bch_dev_bkey_exists(c, ptr->dev)
+ : NULL;
+
+ pr_buf(out, "ptr: %u:%llu gen %u%s%s", ptr->dev,
+ (u64) ptr->offset, ptr->gen,
+ ptr->cached ? " cached" : "",
+ ca && ptr_stale(ca, ptr)
+ ? " stale" : "");
+ break;
+ case BCH_EXTENT_ENTRY_crc32:
+ case BCH_EXTENT_ENTRY_crc64:
+ case BCH_EXTENT_ENTRY_crc128:
+ crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
+
+ pr_buf(out, "crc: c_size %u size %u offset %u nonce %u csum %u compress %u",
+ crc.compressed_size,
+ crc.uncompressed_size,
+ crc.offset, crc.nonce,
+ crc.csum_type,
+ crc.compression_type);
+ break;
+ case BCH_EXTENT_ENTRY_stripe_ptr:
+ ec = &entry->stripe_ptr;
+ pr_buf(out, "ec: idx %llu block %u",
+ (u64) ec->idx, ec->block);
break;
default:
- return BCH_MERGE_NOMERGE;
+ pr_buf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
+ return;
}
- }
- extent_for_each_entry(l, en_l) {
- struct bch_extent_crc_unpacked crc_l, crc_r;
+ first = false;
+ }
+}
- en_r = vstruct_idx(r.v, (u64 *) en_l - l.v->_data);
+static const char *extent_ptr_invalid(const struct bch_fs *c,
+ struct bkey_s_c k,
+ const struct bch_extent_ptr *ptr,
+ unsigned size_ondisk,
+ bool metadata)
+{
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ const struct bch_extent_ptr *ptr2;
+ struct bch_dev *ca;
- if (!extent_entry_is_crc(en_l))
- continue;
+ if (!bch2_dev_exists2(c, ptr->dev))
+ return "pointer to invalid device";
- crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
- crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
+ ca = bch_dev_bkey_exists(c, ptr->dev);
+ if (!ca)
+ return "pointer to invalid device";
- crc_l.csum = bch2_checksum_merge(crc_l.csum_type,
- crc_l.csum,
- crc_r.csum,
- crc_r.uncompressed_size << 9);
+ bkey_for_each_ptr(ptrs, ptr2)
+ if (ptr != ptr2 && ptr->dev == ptr2->dev)
+ return "multiple pointers to same device";
- crc_l.uncompressed_size += crc_r.uncompressed_size;
- crc_l.compressed_size += crc_r.compressed_size;
+ if (ptr->offset + size_ondisk > bucket_to_sector(ca, ca->mi.nbuckets))
+ return "offset past end of device";
- bch2_extent_crc_pack(entry_to_crc(en_l), crc_l,
- extent_entry_type(en_l));
- }
+ if (ptr->offset < bucket_to_sector(ca, ca->mi.first_bucket))
+ return "offset before first bucket";
- bch2_key_resize(l.k, l.k->size + r.k->size);
+ if (bucket_remainder(ca, ptr->offset) +
+ size_ondisk > ca->mi.bucket_size)
+ return "spans multiple buckets";
- return BCH_MERGE_MERGE;
+ return NULL;
}
-bool bch2_check_range_allocated(struct bch_fs *c, struct bpos pos, u64 size,
- unsigned nr_replicas)
+const char *bch2_bkey_ptrs_invalid(const struct bch_fs *c, struct bkey_s_c k)
{
- struct btree_trans trans;
- struct btree_iter *iter;
- struct bpos end = pos;
- struct bkey_s_c k;
- bool ret = true;
- int err;
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ struct bch_devs_list devs;
+ const union bch_extent_entry *entry;
+ struct bch_extent_crc_unpacked crc;
+ unsigned size_ondisk = k.k->size;
+ const char *reason;
+ unsigned nonce = UINT_MAX;
+ unsigned i;
- end.offset += size;
+ if (k.k->type == KEY_TYPE_btree_ptr ||
+ k.k->type == KEY_TYPE_btree_ptr_v2)
+ size_ondisk = c->opts.btree_node_size;
- bch2_trans_init(&trans, c, 0, 0);
+ bkey_extent_entry_for_each(ptrs, entry) {
+ if (__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX)
+ return "invalid extent entry type";
- for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS, pos,
- BTREE_ITER_SLOTS, k, err) {
- if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
+ if (k.k->type == KEY_TYPE_btree_ptr &&
+ !extent_entry_is_ptr(entry))
+ return "has non ptr field";
+
+ switch (extent_entry_type(entry)) {
+ case BCH_EXTENT_ENTRY_ptr:
+ reason = extent_ptr_invalid(c, k, &entry->ptr,
+ size_ondisk, false);
+ if (reason)
+ return reason;
break;
+ case BCH_EXTENT_ENTRY_crc32:
+ case BCH_EXTENT_ENTRY_crc64:
+ case BCH_EXTENT_ENTRY_crc128:
+ crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
- if (nr_replicas > bch2_bkey_nr_ptrs_allocated(k)) {
- ret = false;
+ if (crc.offset + crc.live_size >
+ crc.uncompressed_size)
+ return "checksum offset + key size > uncompressed size";
+
+ size_ondisk = crc.compressed_size;
+
+ if (!bch2_checksum_type_valid(c, crc.csum_type))
+ return "invalid checksum type";
+
+ if (crc.compression_type >= BCH_COMPRESSION_TYPE_NR)
+ return "invalid compression type";
+
+ if (bch2_csum_type_is_encryption(crc.csum_type)) {
+ if (nonce == UINT_MAX)
+ nonce = crc.offset + crc.nonce;
+ else if (nonce != crc.offset + crc.nonce)
+ return "incorrect nonce";
+ }
+ break;
+ case BCH_EXTENT_ENTRY_stripe_ptr:
break;
}
}
- bch2_trans_exit(&trans);
- return ret;
+ devs = bch2_bkey_devs(k);
+ bubble_sort(devs.devs, devs.nr, u8_cmp);
+ for (i = 0; i + 1 < devs.nr; i++)
+ if (devs.devs[i] == devs.devs[i + 1])
+ return "multiple ptrs to same device";
+
+ return NULL;
}
-unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c k)
+void bch2_ptr_swab(struct bkey_s k)
{
- unsigned ret = 0;
+ struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
+ union bch_extent_entry *entry;
+ u64 *d;
+
+ for (d = (u64 *) ptrs.start;
+ d != (u64 *) ptrs.end;
+ d++)
+ *d = swab64(*d);
+
+ for (entry = ptrs.start;
+ entry < ptrs.end;
+ entry = extent_entry_next(entry)) {
+ switch (extent_entry_type(entry)) {
+ case BCH_EXTENT_ENTRY_ptr:
+ break;
+ case BCH_EXTENT_ENTRY_crc32:
+ entry->crc32.csum = swab32(entry->crc32.csum);
+ break;
+ case BCH_EXTENT_ENTRY_crc64:
+ entry->crc64.csum_hi = swab16(entry->crc64.csum_hi);
+ entry->crc64.csum_lo = swab64(entry->crc64.csum_lo);
+ break;
+ case BCH_EXTENT_ENTRY_crc128:
+ entry->crc128.csum.hi = (__force __le64)
+ swab64((__force u64) entry->crc128.csum.hi);
+ entry->crc128.csum.lo = (__force __le64)
+ swab64((__force u64) entry->crc128.csum.lo);
+ break;
+ case BCH_EXTENT_ENTRY_stripe_ptr:
+ break;
+ }
+ }
+}
+
+/* Generic extent code: */
+
+int bch2_cut_front_s(struct bpos where, struct bkey_s k)
+{
+ unsigned new_val_u64s = bkey_val_u64s(k.k);
+ int val_u64s_delta;
+ u64 sub;
+
+ if (bkey_cmp(where, bkey_start_pos(k.k)) <= 0)
+ return 0;
+
+ EBUG_ON(bkey_cmp(where, k.k->p) > 0);
+
+ sub = where.offset - bkey_start_offset(k.k);
+
+ k.k->size -= sub;
+
+ if (!k.k->size) {
+ k.k->type = KEY_TYPE_deleted;
+ new_val_u64s = 0;
+ }
switch (k.k->type) {
- case KEY_TYPE_extent: {
- struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
- const union bch_extent_entry *entry;
- struct extent_ptr_decoded p;
+ case KEY_TYPE_extent:
+ case KEY_TYPE_reflink_v: {
+ struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
+ union bch_extent_entry *entry;
+ bool seen_crc = false;
+
+ bkey_extent_entry_for_each(ptrs, entry) {
+ switch (extent_entry_type(entry)) {
+ case BCH_EXTENT_ENTRY_ptr:
+ if (!seen_crc)
+ entry->ptr.offset += sub;
+ break;
+ case BCH_EXTENT_ENTRY_crc32:
+ entry->crc32.offset += sub;
+ break;
+ case BCH_EXTENT_ENTRY_crc64:
+ entry->crc64.offset += sub;
+ break;
+ case BCH_EXTENT_ENTRY_crc128:
+ entry->crc128.offset += sub;
+ break;
+ case BCH_EXTENT_ENTRY_stripe_ptr:
+ break;
+ }
+
+ if (extent_entry_is_crc(entry))
+ seen_crc = true;
+ }
- extent_for_each_ptr_decode(e, p, entry)
- ret += !p.ptr.cached &&
- p.crc.compression_type == BCH_COMPRESSION_NONE;
break;
}
- case KEY_TYPE_reservation:
- ret = bkey_s_c_to_reservation(k).v->nr_replicas;
+ case KEY_TYPE_reflink_p: {
+ struct bkey_s_reflink_p p = bkey_s_to_reflink_p(k);
+
+ le64_add_cpu(&p.v->idx, sub);
break;
}
+ case KEY_TYPE_inline_data:
+ case KEY_TYPE_indirect_inline_data: {
+ void *p = bkey_inline_data_p(k);
+ unsigned bytes = bkey_inline_data_bytes(k.k);
- return ret;
-}
+ sub = min_t(u64, sub << 9, bytes);
-/* KEY_TYPE_reservation: */
+ memmove(p, p + sub, bytes - sub);
-const char *bch2_reservation_invalid(const struct bch_fs *c, struct bkey_s_c k)
-{
- struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
-
- if (bkey_val_bytes(k.k) != sizeof(struct bch_reservation))
- return "incorrect value size";
+ new_val_u64s -= sub >> 3;
+ break;
+ }
+ }
- if (!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX)
- return "invalid nr_replicas";
+ val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
+ BUG_ON(val_u64s_delta < 0);
- return NULL;
+ set_bkey_val_u64s(k.k, new_val_u64s);
+ memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
+ return -val_u64s_delta;
}
-void bch2_reservation_to_text(struct printbuf *out, struct bch_fs *c,
- struct bkey_s_c k)
+int bch2_cut_back_s(struct bpos where, struct bkey_s k)
{
- struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
+ unsigned new_val_u64s = bkey_val_u64s(k.k);
+ int val_u64s_delta;
+ u64 len = 0;
- pr_buf(out, "generation %u replicas %u",
- le32_to_cpu(r.v->generation),
- r.v->nr_replicas);
-}
+ if (bkey_cmp(where, k.k->p) >= 0)
+ return 0;
-enum merge_result bch2_reservation_merge(struct bch_fs *c,
- struct bkey_s _l, struct bkey_s _r)
-{
- struct bkey_s_reservation l = bkey_s_to_reservation(_l);
- struct bkey_s_reservation r = bkey_s_to_reservation(_r);
+ EBUG_ON(bkey_cmp(where, bkey_start_pos(k.k)) < 0);
- if (l.v->generation != r.v->generation ||
- l.v->nr_replicas != r.v->nr_replicas)
- return BCH_MERGE_NOMERGE;
+ len = where.offset - bkey_start_offset(k.k);
- if ((u64) l.k->size + r.k->size > KEY_SIZE_MAX) {
- bch2_key_resize(l.k, KEY_SIZE_MAX);
- bch2_cut_front_s(l.k->p, r.s);
- return BCH_MERGE_PARTIAL;
+ k.k->p.offset = where.offset;
+ k.k->size = len;
+
+ if (!len) {
+ k.k->type = KEY_TYPE_deleted;
+ new_val_u64s = 0;
}
- bch2_key_resize(l.k, l.k->size + r.k->size);
+ switch (k.k->type) {
+ case KEY_TYPE_inline_data:
+ case KEY_TYPE_indirect_inline_data:
+ new_val_u64s = (bkey_inline_data_offset(k.k) +
+ min(bkey_inline_data_bytes(k.k), k.k->size << 9)) >> 3;
+ break;
+ }
- return BCH_MERGE_MERGE;
+ val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
+ BUG_ON(val_u64s_delta < 0);
+
+ set_bkey_val_u64s(k.k, new_val_u64s);
+ memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
+ return -val_u64s_delta;
}