-d763e8ab17ff1f5bdd9c5474ac15eb8791d31582
+3cd63315a62c8e2da15aa9fd59fa74ef40c9dc14
x(reflink, 6) \
x(new_siphash, 7) \
x(inline_data, 8) \
- x(new_extent_overwrite, 9)
+ x(new_extent_overwrite, 9) \
+ x(incompressible, 10)
enum bch_sb_feature {
#define x(f, n) BCH_FEATURE_##f,
};
#define BCH_COMPRESSION_TYPES() \
- x(none, 0) \
- x(lz4_old, 1) \
- x(gzip, 2) \
- x(lz4, 3) \
- x(zstd, 4)
+ x(none, 0) \
+ x(lz4_old, 1) \
+ x(gzip, 2) \
+ x(lz4, 3) \
+ x(zstd, 4) \
+ x(incompressible, 5)
enum bch_compression_type {
#define x(t, n) BCH_COMPRESSION_TYPE_##t,
bool filter_whiteouts)
{
struct bkey_packed *prev = NULL, *k_packed;
- struct bkey_s k;
+ struct bkey_on_stack k;
struct btree_nr_keys nr;
- struct bkey unpacked;
memset(&nr, 0, sizeof(nr));
+ bkey_on_stack_init(&k);
while ((k_packed = bch2_btree_node_iter_next_all(iter, src))) {
if (filter_whiteouts && bkey_whiteout(k_packed))
continue;
- k = __bkey_disassemble(src, k_packed, &unpacked);
+ /*
+ * NOTE:
+ * bch2_bkey_normalize may modify the key we pass it (dropping
+ * stale pointers) and we don't have a write lock on the src
+ * node; we have to make a copy of the entire key before calling
+ * normalize
+ */
+ bkey_on_stack_realloc(&k, c, k_packed->u64s + BKEY_U64s);
+ bch2_bkey_unpack(src, k.k, k_packed);
if (filter_whiteouts &&
- bch2_bkey_normalize(c, k))
+ bch2_bkey_normalize(c, bkey_i_to_s(k.k)))
continue;
- extent_sort_append(c, out_f, &nr, vstruct_last(dst), &prev, k);
+ extent_sort_append(c, out_f, &nr, vstruct_last(dst),
+ &prev, bkey_i_to_s(k.k));
}
extent_sort_advance_prev(out_f, &nr, vstruct_last(dst), &prev);
dst->u64s = cpu_to_le16((u64 *) prev - dst->_data);
+ bkey_on_stack_exit(&k, c);
return nr;
}
sort_iter_sort(iter, sort_keys_cmp);
while ((in = sort_iter_next(iter, sort_keys_cmp))) {
+ bool needs_whiteout = false;
+
if (bkey_whiteout(in) &&
(filter_whiteouts || !in->needs_whiteout))
continue;
- if (bkey_whiteout(in) &&
- (next = sort_iter_peek(iter)) &&
- !bkey_cmp_packed(iter->b, in, next)) {
+ while ((next = sort_iter_peek(iter)) &&
+ !bkey_cmp_packed(iter->b, in, next)) {
BUG_ON(in->needs_whiteout &&
next->needs_whiteout);
- /*
- * XXX racy, called with read lock from write path
- *
- * leads to spurious BUG_ON() in bkey_unpack_key() in
- * debug mode
- */
- next->needs_whiteout |= in->needs_whiteout;
- continue;
+ needs_whiteout |= in->needs_whiteout;
+ in = sort_iter_next(iter, sort_keys_cmp);
}
if (bkey_whiteout(in)) {
} else {
bkey_copy(out, in);
}
+ out->needs_whiteout |= needs_whiteout;
out = bkey_next(out);
}
{
if (lossy_packed_search)
while (m != btree_bkey_last(b, t) &&
- bkey_iter_cmp_p_or_unp(b, search, lossy_packed_search,
- m) > 0)
+ bkey_iter_cmp_p_or_unp(b, m,
+ lossy_packed_search, search) < 0)
m = bkey_next_skip_noops(m, btree_bkey_last(b, t));
if (!packed_search)
while (m != btree_bkey_last(b, t) &&
- bkey_iter_pos_cmp(b, search, m) > 0)
+ bkey_iter_pos_cmp(b, m, search) < 0)
m = bkey_next_skip_noops(m, btree_bkey_last(b, t));
if (btree_keys_expensive_checks(b)) {
struct bkey_packed *prev = bch2_bkey_prev_all(b, t, m);
BUG_ON(prev &&
- bkey_iter_cmp_p_or_unp(b, search, packed_search,
- prev) <= 0);
+ bkey_iter_cmp_p_or_unp(b, prev,
+ packed_search, search) >= 0);
}
return m;
if (btree_keys_expensive_checks(b)) {
struct bkey dst2 = __bch2_bkey_unpack_key(&b->format, src);
- /*
- * hack around a harmless race when compacting whiteouts
- * for a write:
- */
- dst2.needs_whiteout = dst->needs_whiteout;
-
BUG_ON(memcmp(dst, &dst2, sizeof(*dst)));
}
}
static inline int bkey_cmp_p_or_unp(const struct btree *b,
const struct bkey_packed *l,
const struct bkey_packed *r_packed,
- struct bpos *r)
+ const struct bpos *r)
{
EBUG_ON(r_packed && !bkey_packed(r_packed));
* XXX: only need to compare pointers for keys that are both within a
* btree_node_iterator - we need to break ties for prev() to work correctly
*/
-static inline int bkey_iter_cmp(struct btree *b,
+static inline int bkey_iter_cmp(const struct btree *b,
const struct bkey_packed *l,
const struct bkey_packed *r)
{
?: cmp_int(l, r);
}
-static inline int btree_node_iter_cmp(struct btree *b,
+static inline int btree_node_iter_cmp(const struct btree *b,
struct btree_node_iter_set l,
struct btree_node_iter_set r)
{
__btree_node_offset_to_key(b, r.k));
}
-/* These assume l (the search key) is not a deleted key: */
-static inline int bkey_iter_pos_cmp(struct btree *b,
- struct bpos *l,
- const struct bkey_packed *r)
+/* These assume r (the search key) is not a deleted key: */
+static inline int bkey_iter_pos_cmp(const struct btree *b,
+ const struct bkey_packed *l,
+ const struct bpos *r)
{
- return -bkey_cmp_left_packed(b, r, l)
- ?: (int) bkey_deleted(r);
+ return bkey_cmp_left_packed(b, l, r)
+ ?: -((int) bkey_deleted(l));
}
-static inline int bkey_iter_cmp_p_or_unp(struct btree *b,
- struct bpos *l,
- const struct bkey_packed *l_packed,
- const struct bkey_packed *r)
+static inline int bkey_iter_cmp_p_or_unp(const struct btree *b,
+ const struct bkey_packed *l,
+ const struct bkey_packed *r_packed,
+ const struct bpos *r)
{
- return -bkey_cmp_p_or_unp(b, r, l_packed, l)
- ?: (int) bkey_deleted(r);
+ return bkey_cmp_p_or_unp(b, l, r_packed, r)
+ ?: -((int) bkey_deleted(l));
}
static inline struct bkey_packed *
return c->opts.btree_node_size >> c->block_bits;
}
-#define BTREE_SPLIT_THRESHOLD(c) (btree_blocks(c) * 3 / 4)
+#define BTREE_SPLIT_THRESHOLD(c) (btree_max_u64s(c) * 3 / 4)
#define BTREE_FOREGROUND_MERGE_THRESHOLD(c) (btree_max_u64s(c) * 1 / 3)
#define BTREE_FOREGROUND_MERGE_HYSTERESIS(c) \
#include <linux/prefetch.h>
#include <trace/events/bcachefs.h>
-static inline struct bkey_s_c __btree_iter_peek_all(struct btree_iter *,
- struct btree_iter_level *,
- struct bkey *);
-
#define BTREE_ITER_NO_NODE_GET_LOCKS ((struct btree *) 1)
#define BTREE_ITER_NO_NODE_DROP ((struct btree *) 2)
#define BTREE_ITER_NO_NODE_LOCK_ROOT ((struct btree *) 3)
(unsigned long) iter->l[l].b >= 128;
}
-/* Returns < 0 if @k is before iter pos, > 0 if @k is after */
-static inline int __btree_iter_pos_cmp(struct btree_iter *iter,
- const struct btree *b,
- const struct bkey_packed *k,
- bool interior_node)
+static inline struct bpos btree_iter_search_key(struct btree_iter *iter)
{
- int cmp = bkey_cmp_left_packed(b, k, &iter->pos);
-
- if (cmp)
- return cmp;
- if (bkey_deleted(k))
- return -1;
-
- /*
- * Normally, for extents we want the first key strictly greater than
- * the iterator position - with the exception that for interior nodes,
- * we don't want to advance past the last key if the iterator position
- * is POS_MAX:
- */
- if (iter->flags & BTREE_ITER_IS_EXTENTS &&
- (!interior_node ||
- bkey_cmp_left_packed_byval(b, k, POS_MAX)))
- return -1;
- return 1;
-}
+ struct bpos pos = iter->pos;
-static inline int btree_iter_pos_cmp(struct btree_iter *iter,
- const struct btree *b,
- const struct bkey_packed *k)
-{
- return __btree_iter_pos_cmp(iter, b, k, b->level != 0);
+ if ((iter->flags & BTREE_ITER_IS_EXTENTS) &&
+ bkey_cmp(pos, POS_MAX))
+ pos = bkey_successor(pos);
+ return pos;
}
/* Btree node locking: */
static void __bch2_btree_iter_verify(struct btree_iter *iter,
struct btree *b)
{
+ struct bpos pos = btree_iter_search_key(iter);
struct btree_iter_level *l = &iter->l[b->level];
struct btree_node_iter tmp = l->iter;
struct bkey_packed *k;
k = b->level || iter->flags & BTREE_ITER_IS_EXTENTS
? bch2_btree_node_iter_prev_filter(&tmp, b, KEY_TYPE_discard)
: bch2_btree_node_iter_prev_all(&tmp, b);
- if (k && btree_iter_pos_cmp(iter, b, k) > 0) {
+ if (k && bkey_iter_pos_cmp(b, k, &pos) >= 0) {
char buf[100];
struct bkey uk = bkey_unpack_key(b, k);
bch2_bkey_to_text(&PBUF(buf), &uk);
- panic("prev key should be before iter pos:\n%s\n%llu:%llu\n",
+ panic("iterator should be before prev key:\n%s\n%llu:%llu\n",
buf, iter->pos.inode, iter->pos.offset);
}
k = bch2_btree_node_iter_peek_all(&l->iter, b);
- if (k && btree_iter_pos_cmp(iter, b, k) < 0) {
+ if (k && bkey_iter_pos_cmp(b, k, &pos) < 0) {
char buf[100];
struct bkey uk = bkey_unpack_key(b, k);
}
static void __bch2_btree_iter_fix_key_modified(struct btree_iter *iter,
- struct btree *b,
- struct bkey_packed *where)
+ struct btree *b,
+ struct bkey_packed *where)
{
- struct btree_node_iter *node_iter = &iter->l[0].iter;
+ struct btree_iter_level *l = &iter->l[b->level];
+ struct bpos pos = btree_iter_search_key(iter);
- if (where == bch2_btree_node_iter_peek_all(node_iter, b)) {
- bkey_disassemble(b, where, &iter->k);
- btree_iter_set_dirty(iter, BTREE_ITER_NEED_PEEK);
- }
+ if (where != bch2_btree_node_iter_peek_all(&l->iter, l->b))
+ return;
+
+ if (bkey_iter_pos_cmp(l->b, where, &pos) < 0)
+ bch2_btree_node_iter_advance(&l->iter, l->b);
+
+ btree_iter_set_dirty(iter, BTREE_ITER_NEED_PEEK);
}
void bch2_btree_iter_fix_key_modified(struct btree_iter *iter,
bool iter_current_key_modified =
orig_iter_pos >= offset &&
orig_iter_pos <= offset + clobber_u64s;
+ struct bpos iter_pos = btree_iter_search_key(iter);
btree_node_iter_for_each(node_iter, set)
if (set->end == old_end)
/* didn't find the bset in the iterator - might have to readd it: */
if (new_u64s &&
- btree_iter_pos_cmp(iter, b, where) > 0) {
+ bkey_iter_pos_cmp(b, where, &iter_pos) >= 0) {
bch2_btree_node_iter_push(node_iter, b, where, end);
goto fixup_done;
} else {
return;
if (new_u64s &&
- btree_iter_pos_cmp(iter, b, where) > 0) {
+ bkey_iter_pos_cmp(b, where, &iter_pos) >= 0) {
set->k = offset;
} else if (set->k < offset + clobber_u64s) {
set->k = offset + new_u64s;
struct btree_iter_level *l,
int max_advance)
{
+ struct bpos pos = btree_iter_search_key(iter);
struct bkey_packed *k;
int nr_advanced = 0;
while ((k = bch2_btree_node_iter_peek_all(&l->iter, l->b)) &&
- btree_iter_pos_cmp(iter, l->b, k) < 0) {
+ bkey_iter_pos_cmp(l->b, k, &pos) < 0) {
if (max_advance > 0 && nr_advanced >= max_advance)
return false;
static inline bool btree_iter_pos_after_node(struct btree_iter *iter,
struct btree *b)
{
- int cmp = bkey_cmp(b->key.k.p, iter->pos);
-
- if (!cmp &&
- (iter->flags & BTREE_ITER_IS_EXTENTS) &&
- bkey_cmp(b->key.k.p, POS_MAX))
- cmp = -1;
- return cmp < 0;
+ return bkey_cmp(b->key.k.p, btree_iter_search_key(iter)) < 0;
}
static inline bool btree_iter_pos_in_node(struct btree_iter *iter,
static inline void __btree_iter_init(struct btree_iter *iter,
unsigned level)
{
+ struct bpos pos = btree_iter_search_key(iter);
struct btree_iter_level *l = &iter->l[level];
- bch2_btree_node_iter_init(&l->iter, l->b, &iter->pos);
-
- if (iter->flags & BTREE_ITER_IS_EXTENTS)
- btree_iter_advance_to_pos(iter, l, -1);
-
- /* Skip to first non whiteout: */
- if (level)
- bch2_btree_node_iter_peek(&l->iter, l->b);
+ bch2_btree_node_iter_init(&l->iter, l->b, &pos);
btree_iter_set_dirty(iter, BTREE_ITER_NEED_PEEK);
}
if (debug_check_iterators(iter->trans->c)) {
struct bkey k = bkey_unpack_key(l->b, _k);
- /*
- * this flag is internal to the btree code,
- * we don't care if it doesn't match - if it's now set
- * it just means the key has been written out to disk:
- */
- k.needs_whiteout = iter->k.needs_whiteout;
BUG_ON(memcmp(&k, &iter->k, sizeof(k)));
}
int ret;
recheck:
- while ((k = __btree_iter_peek_all(iter, l, &iter->k)).k &&
- bkey_cmp(k.k->p, iter->pos) <= 0)
- bch2_btree_node_iter_advance(&l->iter, l->b);
+ btree_iter_advance_to_pos(iter, l, -1);
/*
* iterator is now at the correct position for inserting at iter->pos,
*/
node_iter = l->iter;
- if (k.k && bkey_whiteout(k.k))
- k = __btree_iter_unpack(iter, l, &iter->k,
- bch2_btree_node_iter_peek(&node_iter, l->b));
+ k = __btree_iter_unpack(iter, l, &iter->k,
+ bch2_btree_node_iter_peek(&node_iter, l->b));
+
+ if (k.k && bkey_cmp(bkey_start_pos(k.k), iter->pos) <= 0) {
+ /*
+ * If there wasn't actually a hole, want the iterator to be
+ * pointed at the key we found:
+ *
+ * XXX: actually, we shouldn't be changing the iterator here:
+ * the iterator needs to be correct for inserting at iter->pos,
+ * and there may be whiteouts between iter->pos and what this
+ * iterator points at:
+ */
+ l->iter = node_iter;
+
+ EBUG_ON(bkey_cmp(k.k->p, iter->pos) <= 0);
+ iter->uptodate = BTREE_ITER_UPTODATE;
+
+ __bch2_btree_iter_verify(iter, l->b);
+ return k;
+ }
/*
* If we got to the end of the node, check if we need to traverse to the
goto recheck;
}
- if (k.k &&
- !bkey_whiteout(k.k) &&
- bkey_cmp(bkey_start_pos(k.k), iter->pos) <= 0) {
- /*
- * if we skipped forward to find the first non whiteout and
- * there _wasn't_ actually a hole, we want the iterator to be
- * pointed at the key we found:
- */
- l->iter = node_iter;
-
- EBUG_ON(bkey_cmp(k.k->p, iter->pos) < 0);
- EBUG_ON(bkey_deleted(k.k));
- iter->uptodate = BTREE_ITER_UPTODATE;
-
- __bch2_btree_iter_verify(iter, l->b);
- return k;
- }
-
/* hole */
/* holes can't span inode numbers: */
#define BTREE_ITER_TYPE ((1 << 2) - 1)
+/*
+ * Iterate over all possible positions, synthesizing deleted keys for holes:
+ */
#define BTREE_ITER_SLOTS (1 << 2)
+/*
+ * Indicates that intent locks should be taken on leaf nodes, because we expect
+ * to be doing updates:
+ */
#define BTREE_ITER_INTENT (1 << 3)
+/*
+ * Causes the btree iterator code to prefetch additional btree nodes from disk:
+ */
#define BTREE_ITER_PREFETCH (1 << 4)
+/*
+ * Indicates that this iterator should not be reused until transaction commit,
+ * either because a pending update references it or because the update depends
+ * on that particular key being locked (e.g. by the str_hash code, for hash
+ * table consistency)
+ */
#define BTREE_ITER_KEEP_UNTIL_COMMIT (1 << 5)
/*
* Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
BTREE_TRIGGER_GC);
while ((k = bch2_btree_node_iter_peek_all(node_iter, b)) &&
- bkey_iter_pos_cmp(b, &insert->k.p, k) > 0)
+ bkey_iter_pos_cmp(b, k, &insert->k.p) < 0)
bch2_btree_node_iter_advance(node_iter, b);
/*
if (keys)
btree_split_insert_keys(as, n1, iter, keys);
- if (vstruct_blocks(n1->data, c->block_bits) > BTREE_SPLIT_THRESHOLD(c)) {
+ if (bset_u64s(&n1->set[0]) > BTREE_SPLIT_THRESHOLD(c)) {
trace_btree_split(c, b);
n2 = __btree_split_node(as, n1, iter);
bkey_cmp(insert->k.p, b->data->max_key) > 0);
k = bch2_btree_node_iter_peek_all(node_iter, b);
- if (k && !bkey_cmp_packed(b, k, &insert->k)) {
- BUG_ON(bkey_whiteout(k));
+ if (k && bkey_cmp_packed(b, k, &insert->k))
+ k = NULL;
- if (!bkey_written(b, k) &&
- bkey_val_u64s(&insert->k) == bkeyp_val_u64s(f, k) &&
- !bkey_whiteout(&insert->k)) {
- k->type = insert->k.type;
- memcpy_u64s(bkeyp_val(f, k), &insert->v,
- bkey_val_u64s(&insert->k));
- return true;
- }
+ /* @k is the key being overwritten/deleted, if any: */
- btree_account_key_drop(b, k);
+ EBUG_ON(k && bkey_whiteout(k));
- if (bkey_whiteout(&insert->k)) {
- unsigned clobber_u64s = k->u64s, new_u64s = k->u64s;
+ if (bkey_whiteout(&insert->k)) {
+ /* Deleting: */
- k->type = KEY_TYPE_deleted;
+ /* Not found? Nothing to do: */
+ if (!k)
+ return false;
- if (k->needs_whiteout) {
- push_whiteout(iter->trans->c, b, k);
- k->needs_whiteout = false;
- }
+ btree_account_key_drop(b, k);
+ k->type = KEY_TYPE_deleted;
- if (k >= btree_bset_last(b)->start) {
- bch2_bset_delete(b, k, clobber_u64s);
- new_u64s = 0;
- }
+ if (k->needs_whiteout) {
+ push_whiteout(iter->trans->c, b, k);
+ k->needs_whiteout = false;
+ }
+
+ if (k >= btree_bset_last(b)->start) {
+ clobber_u64s = k->u64s;
+ bch2_bset_delete(b, k, clobber_u64s);
bch2_btree_node_iter_fix(iter, b, node_iter, k,
- clobber_u64s, new_u64s);
- return true;
+ clobber_u64s, 0);
+ } else {
+ bch2_btree_iter_fix_key_modified(iter, b, k);
+ }
+
+ return true;
+ }
+ if (k) {
+ /* Overwriting: */
+ if (!bkey_written(b, k) &&
+ bkey_val_u64s(&insert->k) == bkeyp_val_u64s(f, k)) {
+ k->type = insert->k.type;
+ memcpy_u64s(bkeyp_val(f, k), &insert->v,
+ bkey_val_u64s(&insert->k));
+ return true;
}
+ btree_account_key_drop(b, k);
+ k->type = KEY_TYPE_deleted;
+
insert->k.needs_whiteout = k->needs_whiteout;
k->needs_whiteout = false;
if (k >= btree_bset_last(b)->start) {
clobber_u64s = k->u64s;
goto overwrite;
+ } else {
+ bch2_btree_iter_fix_key_modified(iter, b, k);
}
-
- k->type = KEY_TYPE_deleted;
- /*
- * XXX: we should be able to do this without two calls to
- * bch2_btree_node_iter_fix:
- */
- bch2_btree_node_iter_fix(iter, b, node_iter, k,
- k->u64s, k->u64s);
- } else {
- /*
- * Deleting, but the key to delete wasn't found - nothing to do:
- */
- if (bkey_whiteout(&insert->k))
- return false;
-
- insert->k.needs_whiteout = false;
}
k = bch2_btree_node_iter_bset_pos(node_iter, b, bset_tree_last(b));
trans_trigger_run = false;
trans_for_each_update(trans, i) {
- /* we know trans->nounlock won't be set here: */
- if (unlikely(!(i->iter->locks_want < 1
- ? __bch2_btree_iter_upgrade(i->iter, 1)
- : i->iter->uptodate <= BTREE_ITER_NEED_PEEK))) {
+ if (unlikely(i->iter->uptodate > BTREE_ITER_NEED_PEEK)) {
+ trace_trans_restart_traverse(trans->ip);
+ ret = -EINTR;
+ goto out;
+ }
+
+ /*
+ * We're not using bch2_btree_iter_upgrade here because
+ * we know trans->nounlock can't be set:
+ */
+ if (unlikely(i->iter->locks_want < 1 &&
+ !__bch2_btree_iter_upgrade(i->iter, 1))) {
trace_trans_restart_upgrade(trans->ip);
ret = -EINTR;
goto out;
BUG_ON(len_a + len_b > bio_sectors(bio));
BUG_ON(crc_old.uncompressed_size != bio_sectors(bio));
- BUG_ON(crc_old.compression_type);
+ BUG_ON(crc_is_compressed(crc_old));
BUG_ON(bch2_csum_type_is_encryption(crc_old.csum_type) !=
bch2_csum_type_is_encryption(new_csum_type));
if (i->crc)
*i->crc = (struct bch_extent_crc_unpacked) {
.csum_type = i->csum_type,
+ .compression_type = crc_old.compression_type,
.compressed_size = i->len,
.uncompressed_size = i->len,
.offset = 0,
static inline struct nonce extent_nonce(struct bversion version,
struct bch_extent_crc_unpacked crc)
{
- unsigned size = crc.compression_type ? crc.uncompressed_size : 0;
+ unsigned compression_type = crc_is_compressed(crc)
+ ? crc.compression_type
+ : 0;
+ unsigned size = compression_type ? crc.uncompressed_size : 0;
struct nonce nonce = (struct nonce) {{
[0] = cpu_to_le32(size << 22),
[1] = cpu_to_le32(version.lo),
[2] = cpu_to_le32(version.lo >> 32),
[3] = cpu_to_le32(version.hi|
- (crc.compression_type << 24))^BCH_NONCE_EXTENT,
+ (compression_type << 24))^BCH_NONCE_EXTENT,
}};
return nonce_add(nonce, crc.nonce << 9);
bio_unmap_or_unbounce(c, dst_data);
return compression_type;
err:
- compression_type = 0;
+ compression_type = BCH_COMPRESSION_TYPE_incompressible;
goto out;
}
if (!bch2_checksum_mergeable(crc_l.csum_type))
return BCH_MERGE_NOMERGE;
- if (crc_l.compression_type)
+ if (crc_is_compressed(crc_l))
return BCH_MERGE_NOMERGE;
if (crc_l.csum_type &&
static inline bool can_narrow_crc(struct bch_extent_crc_unpacked u,
struct bch_extent_crc_unpacked n)
{
- return !u.compression_type &&
+ return !crc_is_compressed(u) &&
u.csum_type &&
u.uncompressed_size > u.live_size &&
bch2_csum_type_is_encryption(u.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);
struct extent_ptr_decoded p;
bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
- ret += !p.ptr.cached &&
- p.crc.compression_type == BCH_COMPRESSION_TYPE_none;
+ ret += !p.ptr.cached && !crc_is_compressed(p.crc);
}
return ret;
unsigned ret = 0;
bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
- if (!p.ptr.cached &&
- p.crc.compression_type != BCH_COMPRESSION_TYPE_none)
+ if (!p.ptr.cached && crc_is_compressed(p.crc))
ret += p.crc.compressed_size;
return ret;
}
+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;
+
+ bkey_for_each_crc(k.k, ptrs, crc, entry)
+ if (crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
+ return true;
+ return false;
+}
+
bool bch2_check_range_allocated(struct bch_fs *c, struct bpos pos, u64 size,
unsigned nr_replicas)
{
#undef common_fields
}
+static inline bool crc_is_compressed(struct bch_extent_crc_unpacked crc)
+{
+ return (crc.compression_type != BCH_COMPRESSION_TYPE_none &&
+ crc.compression_type != BCH_COMPRESSION_TYPE_incompressible);
+}
+
/* bkey_ptrs: generically over any key type that has ptrs */
struct bkey_ptrs_c {
unsigned bch2_bkey_nr_ptrs(struct bkey_s_c);
unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c);
unsigned bch2_bkey_nr_ptrs_fully_allocated(struct bkey_s_c);
+bool bch2_bkey_is_incompressible(struct bkey_s_c);
unsigned bch2_bkey_sectors_compressed(struct bkey_s_c);
bool bch2_check_range_allocated(struct bch_fs *, struct bpos, u64, unsigned);
unsigned bch2_bkey_durability(struct bch_fs *, struct bkey_s_c);
* particularly want to plumb io_opts all the way through the btree
* update stack right now
*/
- for_each_keylist_key(keys, k)
+ for_each_keylist_key(keys, k) {
bch2_rebalance_add_key(c, bkey_i_to_s_c(k), &op->opts);
+ if (bch2_bkey_is_incompressible(bkey_i_to_s_c(k)))
+ bch2_check_set_feature(op->c, BCH_FEATURE_incompressible);
+
+ }
+
if (!bch2_keylist_empty(keys)) {
u64 sectors_start = keylist_sectors(keys);
int ret = op->index_update_fn(op);
/* Can we just write the entire extent as is? */
if (op->crc.uncompressed_size == op->crc.live_size &&
op->crc.compressed_size <= wp->sectors_free &&
- op->crc.compression_type == op->compression_type) {
- if (!op->crc.compression_type &&
+ (op->crc.compression_type == op->compression_type ||
+ op->incompressible)) {
+ if (!crc_is_compressed(op->crc) &&
op->csum_type != op->crc.csum_type &&
bch2_write_rechecksum(c, op, op->csum_type))
return PREP_ENCODED_CHECKSUM_ERR;
* If the data is compressed and we couldn't write the entire extent as
* is, we have to decompress it:
*/
- if (op->crc.compression_type) {
+ if (crc_is_compressed(op->crc)) {
struct bch_csum csum;
if (bch2_write_decrypt(op))
ret = -EIO;
goto err;
case PREP_ENCODED_CHECKSUM_ERR:
+ BUG();
goto csum_err;
case PREP_ENCODED_DO_WRITE:
/* XXX look for bug here */
bch2_csum_type_is_encryption(op->crc.csum_type));
BUG_ON(op->compression_type && !bounce);
- crc.compression_type = op->compression_type
- ? bch2_bio_compress(c, dst, &dst_len, src, &src_len,
- op->compression_type)
+ crc.compression_type = op->incompressible
+ ? BCH_COMPRESSION_TYPE_incompressible
+ : op->compression_type
+ ? bch2_bio_compress(c, dst, &dst_len, src, &src_len,
+ op->compression_type)
: 0;
- if (!crc.compression_type) {
+ if (!crc_is_compressed(crc)) {
dst_len = min(dst->bi_iter.bi_size, src->bi_iter.bi_size);
dst_len = min_t(unsigned, dst_len, wp->sectors_free << 9);
}
if ((op->flags & BCH_WRITE_DATA_ENCODED) &&
- !crc.compression_type &&
+ !crc_is_compressed(crc) &&
bch2_csum_type_is_encryption(op->crc.csum_type) ==
bch2_csum_type_is_encryption(op->csum_type)) {
/*
static struct promote_op *__promote_alloc(struct bch_fs *c,
enum btree_id btree_id,
+ struct bkey_s_c k,
struct bpos pos,
struct extent_ptr_decoded *pick,
struct bch_io_opts opts,
(struct data_opts) {
.target = opts.promote_target
},
- btree_id,
- bkey_s_c_null);
+ btree_id, k);
BUG_ON(ret);
return op;
k.k->type == KEY_TYPE_reflink_v
? BTREE_ID_REFLINK
: BTREE_ID_EXTENTS,
- pos, pick, opts, sectors, rbio);
+ k, pos, pick, opts, sectors, rbio);
if (!promote)
return NULL;
u64 data_offset = rbio->pos.offset - rbio->pick.crc.offset;
int ret;
- if (rbio->pick.crc.compression_type)
+ if (crc_is_compressed(rbio->pick.crc))
return;
bkey_on_stack_init(&new);
crc.offset += rbio->offset_into_extent;
crc.live_size = bvec_iter_sectors(rbio->bvec_iter);
- if (crc.compression_type != BCH_COMPRESSION_TYPE_none) {
+ if (crc_is_compressed(crc)) {
bch2_encrypt_bio(c, crc.csum_type, nonce, src);
if (bch2_bio_uncompress(c, src, dst, dst_iter, crc))
goto decompression_err;
}
if (rbio->narrow_crcs ||
- rbio->pick.crc.compression_type ||
+ crc_is_compressed(rbio->pick.crc) ||
bch2_csum_type_is_encryption(rbio->pick.crc.csum_type))
context = RBIO_CONTEXT_UNBOUND, wq = system_unbound_wq;
else if (rbio->pick.crc.csum_type)
EBUG_ON(offset_into_extent + bvec_iter_sectors(iter) > k.k->size);
- if (pick.crc.compression_type != BCH_COMPRESSION_TYPE_none ||
+ if (crc_is_compressed(pick.crc) ||
(pick.crc.csum_type != BCH_CSUM_NONE &&
(bvec_iter_sectors(iter) != pick.crc.uncompressed_size ||
(bch2_csum_type_is_encryption(pick.crc.csum_type) &&
&rbio, &bounce, &read_full);
if (!read_full) {
- EBUG_ON(pick.crc.compression_type);
+ EBUG_ON(crc_is_compressed(pick.crc));
EBUG_ON(pick.crc.csum_type &&
(bvec_iter_sectors(iter) != pick.crc.uncompressed_size ||
bvec_iter_sectors(iter) != pick.crc.live_size ||
unsigned nr_replicas:4;
unsigned nr_replicas_required:4;
unsigned alloc_reserve:4;
+ unsigned incompressible:1;
struct bch_devs_list devs_have;
u16 target;
enum btree_id btree_id,
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;
int ret;
m->btree_id = btree_id;
m->nr_ptrs_reserved = 0;
bch2_write_op_init(&m->op, c, io_opts);
- m->op.compression_type =
- bch2_compression_opt_to_type[io_opts.background_compression ?:
- io_opts.compression];
+
+ if (!bch2_bkey_is_incompressible(k))
+ m->op.compression_type =
+ bch2_compression_opt_to_type[io_opts.background_compression ?:
+ io_opts.compression];
+ else
+ m->op.incompressible = true;
+
m->op.target = data_opts.target,
m->op.write_point = wp;
break;
}
case DATA_REWRITE: {
- struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
- const union bch_extent_entry *entry;
- struct extent_ptr_decoded p;
unsigned compressed_sectors = 0;
bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
if (!p.ptr.cached &&
- p.crc.compression_type != BCH_COMPRESSION_TYPE_none &&
+ crc_is_compressed(p.crc) &&
bch2_dev_in_target(c, p.ptr.dev, data_opts.target))
compressed_sectors += p.crc.compressed_size;
#include <linux/sched/cputime.h>
#include <trace/events/bcachefs.h>
-static inline bool rebalance_ptr_pred(struct bch_fs *c,
- struct extent_ptr_decoded p,
- struct bch_io_opts *io_opts)
+/*
+ * Check if an extent should be moved:
+ * returns -1 if it should not be moved, or
+ * device of pointer that should be moved, if known, or INT_MAX if unknown
+ */
+static int __bch2_rebalance_pred(struct bch_fs *c,
+ struct bkey_s_c k,
+ struct bch_io_opts *io_opts)
{
- if (io_opts->background_target &&
- !bch2_dev_in_target(c, p.ptr.dev, io_opts->background_target) &&
- !p.ptr.cached)
- return true;
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ const union bch_extent_entry *entry;
+ struct extent_ptr_decoded p;
if (io_opts->background_compression &&
- p.crc.compression_type !=
- bch2_compression_opt_to_type[io_opts->background_compression])
- return true;
-
- return false;
+ !bch2_bkey_is_incompressible(k))
+ bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
+ if (!p.ptr.cached &&
+ p.crc.compression_type !=
+ bch2_compression_opt_to_type[io_opts->background_compression])
+ return p.ptr.dev;
+
+ if (io_opts->background_target)
+ bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
+ if (!p.ptr.cached &&
+ !bch2_dev_in_target(c, p.ptr.dev, io_opts->background_target))
+ return p.ptr.dev;
+
+ return -1;
}
void bch2_rebalance_add_key(struct bch_fs *c,
struct bkey_s_c k,
struct bch_io_opts *io_opts)
{
- struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
- const union bch_extent_entry *entry;
- struct extent_ptr_decoded p;
+ atomic64_t *counter;
+ int dev;
- if (!io_opts->background_target &&
- !io_opts->background_compression)
+ dev = __bch2_rebalance_pred(c, k, io_opts);
+ if (dev < 0)
return;
- bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
- if (rebalance_ptr_pred(c, p, io_opts)) {
- struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
+ counter = dev < INT_MAX
+ ? &bch_dev_bkey_exists(c, dev)->rebalance_work
+ : &c->rebalance.work_unknown_dev;
- if (atomic64_add_return(p.crc.compressed_size,
- &ca->rebalance_work) ==
- p.crc.compressed_size)
- rebalance_wakeup(c);
- }
-}
-
-void bch2_rebalance_add_work(struct bch_fs *c, u64 sectors)
-{
- if (atomic64_add_return(sectors, &c->rebalance.work_unknown_dev) ==
- sectors)
+ if (atomic64_add_return(k.k->size, counter) == k.k->size)
rebalance_wakeup(c);
}
struct bch_io_opts *io_opts,
struct data_opts *data_opts)
{
- struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
- const union bch_extent_entry *entry;
- struct extent_ptr_decoded p;
- unsigned nr_replicas = 0;
-
- bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
- nr_replicas += !p.ptr.cached;
-
- if (rebalance_ptr_pred(c, p, io_opts))
- goto found;
+ if (__bch2_rebalance_pred(c, k, io_opts) >= 0) {
+ data_opts->target = io_opts->background_target;
+ data_opts->btree_insert_flags = 0;
+ return DATA_ADD_REPLICAS;
+ } else {
+ return DATA_SKIP;
}
+}
- if (nr_replicas < io_opts->data_replicas)
- goto found;
-
- return DATA_SKIP;
-found:
- data_opts->target = io_opts->background_target;
- data_opts->btree_insert_flags = 0;
- return DATA_ADD_REPLICAS;
+void bch2_rebalance_add_work(struct bch_fs *c, u64 sectors)
+{
+ if (atomic64_add_return(sectors, &c->rebalance.work_unknown_dev) ==
+ sectors)
+ rebalance_wakeup(c);
}
struct rebalance_work {
prev_cputime = curr_cputime();
while (!kthread_wait_freezable(r->enabled)) {
+ cond_resched();
+
start = jiffies;
cputime = curr_cputime();
struct extent_ptr_decoded p;
extent_for_each_ptr_decode(e, p, entry) {
- if (p.crc.compression_type == BCH_COMPRESSION_TYPE_none) {
+ if (!crc_is_compressed(p.crc)) {
nr_uncompressed_extents++;
uncompressed_sectors += e.k->size;
} else {
projects / bcachefs-tools-debian / commitdiff