static void verify_no_dups(struct btree *b,
struct bkey_packed *start,
- struct bkey_packed *end,
- bool extents)
+ struct bkey_packed *end)
{
#ifdef CONFIG_BCACHEFS_DEBUG
struct bkey_packed *k, *p;
if (start == end)
return;
- for (p = start, k = bkey_next_skip_noops(start, end);
+ for (p = start, k = bkey_next(start);
k != end;
- p = k, k = bkey_next_skip_noops(k, end)) {
+ p = k, k = bkey_next(k)) {
struct bkey l = bkey_unpack_key(b, p);
struct bkey r = bkey_unpack_key(b, k);
- BUG_ON(extents
- ? bkey_cmp(l.p, bkey_start_pos(&r)) > 0
- : bkey_cmp(l.p, bkey_start_pos(&r)) >= 0);
- //BUG_ON(bkey_cmp_packed(&b->format, p, k) >= 0);
+ BUG_ON(bpos_cmp(l.p, bkey_start_pos(&r)) >= 0);
}
#endif
}
{
struct bkey_packed *k;
- for (k = i->start;
- k != vstruct_last(i);
- k = bkey_next_skip_noops(k, vstruct_last(i)))
+ for (k = i->start; k != vstruct_last(i); k = bkey_next(k))
k->needs_whiteout = v;
}
break;
for (b = a; c = 2 * b + 1, (d = c + 1) < n;)
- b = bkey_cmp_packed(bt,
+ b = bch2_bkey_cmp_packed(bt,
ptrs[c],
ptrs[d]) >= 0 ? c : d;
if (d == n)
b = c;
while (b != a &&
- bkey_cmp_packed(bt,
+ bch2_bkey_cmp_packed(bt,
ptrs[a],
ptrs[b]) >= 0)
b = (b - 1) / 2;
}
verify_no_dups(b, new_whiteouts,
- (void *) ((u64 *) new_whiteouts + b->whiteout_u64s),
- btree_node_old_extent_overwrite(b));
+ (void *) ((u64 *) new_whiteouts + b->whiteout_u64s));
memcpy_u64s(unwritten_whiteouts_start(c, b),
new_whiteouts, b->whiteout_u64s);
}
}
-static bool bch2_compact_extent_whiteouts(struct bch_fs *c,
- struct btree *b,
- enum compact_mode mode)
-{
- const struct bkey_format *f = &b->format;
- struct bset_tree *t;
- struct bkey_packed *whiteouts = NULL;
- struct bkey_packed *u_start, *u_pos;
- struct sort_iter sort_iter;
- unsigned bytes, whiteout_u64s = 0, u64s;
- bool used_mempool, compacting = false;
-
- BUG_ON(!btree_node_is_extents(b));
-
- for_each_bset(b, t)
- if (should_compact_bset(b, t, whiteout_u64s != 0, mode))
- whiteout_u64s += bset_dead_u64s(b, t);
-
- if (!whiteout_u64s)
- return false;
-
- bch2_sort_whiteouts(c, b);
-
- sort_iter_init(&sort_iter, b);
-
- whiteout_u64s += b->whiteout_u64s;
- bytes = whiteout_u64s * sizeof(u64);
-
- whiteouts = btree_bounce_alloc(c, bytes, &used_mempool);
- u_start = u_pos = whiteouts;
-
- memcpy_u64s(u_pos, unwritten_whiteouts_start(c, b),
- b->whiteout_u64s);
- u_pos = (void *) u_pos + b->whiteout_u64s * sizeof(u64);
-
- sort_iter_add(&sort_iter, u_start, u_pos);
-
- for_each_bset(b, t) {
- struct bset *i = bset(b, t);
- struct bkey_packed *k, *n, *out, *start, *end;
- struct btree_node_entry *src = NULL, *dst = NULL;
-
- if (t != b->set && !bset_written(b, i)) {
- src = container_of(i, struct btree_node_entry, keys);
- dst = max(write_block(b),
- (void *) btree_bkey_last(b, t - 1));
- }
-
- if (src != dst)
- compacting = true;
-
- if (!should_compact_bset(b, t, compacting, mode)) {
- if (src != dst) {
- memmove(dst, src, sizeof(*src) +
- le16_to_cpu(src->keys.u64s) *
- sizeof(u64));
- i = &dst->keys;
- set_btree_bset(b, t, i);
- }
- continue;
- }
-
- compacting = true;
- u_start = u_pos;
- start = i->start;
- end = vstruct_last(i);
-
- if (src != dst) {
- memmove(dst, src, sizeof(*src));
- i = &dst->keys;
- set_btree_bset(b, t, i);
- }
-
- out = i->start;
-
- for (k = start; k != end; k = n) {
- n = bkey_next_skip_noops(k, end);
-
- if (bkey_deleted(k))
- continue;
-
- BUG_ON(bkey_whiteout(k) &&
- k->needs_whiteout &&
- bkey_written(b, k));
-
- if (bkey_whiteout(k) && !k->needs_whiteout)
- continue;
-
- if (bkey_whiteout(k)) {
- memcpy_u64s(u_pos, k, bkeyp_key_u64s(f, k));
- set_bkeyp_val_u64s(f, u_pos, 0);
- u_pos = bkey_next(u_pos);
- } else {
- bkey_copy(out, k);
- out = bkey_next(out);
- }
- }
-
- sort_iter_add(&sort_iter, u_start, u_pos);
-
- i->u64s = cpu_to_le16((u64 *) out - i->_data);
- set_btree_bset_end(b, t);
- bch2_bset_set_no_aux_tree(b, t);
- }
-
- b->whiteout_u64s = (u64 *) u_pos - (u64 *) whiteouts;
-
- BUG_ON((void *) unwritten_whiteouts_start(c, b) <
- (void *) btree_bkey_last(b, bset_tree_last(b)));
-
- u64s = bch2_sort_extent_whiteouts(unwritten_whiteouts_start(c, b),
- &sort_iter);
-
- BUG_ON(u64s > b->whiteout_u64s);
- BUG_ON(u_pos != whiteouts && !u64s);
-
- if (u64s != b->whiteout_u64s) {
- void *src = unwritten_whiteouts_start(c, b);
-
- b->whiteout_u64s = u64s;
- memmove_u64s_up(unwritten_whiteouts_start(c, b), src, u64s);
- }
-
- verify_no_dups(b,
- unwritten_whiteouts_start(c, b),
- unwritten_whiteouts_end(c, b),
- true);
-
- btree_bounce_free(c, bytes, used_mempool, whiteouts);
-
- bch2_btree_build_aux_trees(b);
-
- bch_btree_keys_u64s_remaining(c, b);
- bch2_verify_btree_nr_keys(b);
-
- return true;
-}
-
static bool bch2_drop_whiteouts(struct btree *b, enum compact_mode mode)
{
struct bset_tree *t;
out = i->start;
for (k = start; k != end; k = n) {
- n = bkey_next_skip_noops(k, end);
+ n = bkey_next(k);
- if (!bkey_whiteout(k)) {
+ if (!bkey_deleted(k)) {
bkey_copy(out, k);
out = bkey_next(out);
} else {
bool bch2_compact_whiteouts(struct bch_fs *c, struct btree *b,
enum compact_mode mode)
{
- return !btree_node_old_extent_overwrite(b)
- ? bch2_drop_whiteouts(b, mode)
- : bch2_compact_extent_whiteouts(c, b, mode);
+ return bch2_drop_whiteouts(b, mode);
}
static void btree_node_sort(struct bch_fs *c, struct btree *b,
start_time = local_clock();
- if (btree_node_old_extent_overwrite(b))
- filter_whiteouts = bset_written(b, start_bset);
-
- u64s = (btree_node_old_extent_overwrite(b)
- ? bch2_sort_extents
- : bch2_sort_keys)(out->keys.start,
- &sort_iter,
- filter_whiteouts);
+ u64s = bch2_sort_keys(out->keys.start, &sort_iter, filter_whiteouts);
out->keys.u64s = cpu_to_le16(u64s);
bch2_btree_iter_reinit_node(iter, b);
}
-static void btree_err_msg(struct printbuf *out, struct bch_fs *c,
- struct btree *b, struct bset *i,
- unsigned offset, int write)
+static void btree_pos_to_text(struct printbuf *out, struct bch_fs *c,
+ struct btree *b)
{
- pr_buf(out, "error validating btree node %sat btree %u level %u/%u\n"
- "pos ",
- write ? "before write " : "",
- b->c.btree_id, b->c.level,
+ pr_buf(out, "%s level %u/%u\n ",
+ bch2_btree_ids[b->c.btree_id],
+ b->c.level,
c->btree_roots[b->c.btree_id].level);
bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
+}
- pr_buf(out, " node offset %u", b->written);
+static void btree_err_msg(struct printbuf *out, struct bch_fs *c,
+ struct bch_dev *ca,
+ struct btree *b, struct bset *i,
+ unsigned offset, int write)
+{
+ pr_buf(out, "error validating btree node ");
+ if (write)
+ pr_buf(out, "before write ");
+ if (ca)
+ pr_buf(out, "on %s ", ca->name);
+ pr_buf(out, "at btree ");
+ btree_pos_to_text(out, c, b);
+
+ pr_buf(out, "\n node offset %u", b->written);
if (i)
pr_buf(out, " bset u64s %u", le16_to_cpu(i->u64s));
}
BTREE_RETRY_READ = 64,
};
-#define btree_err(type, c, b, i, msg, ...) \
+#define btree_err(type, c, ca, b, i, msg, ...) \
({ \
__label__ out; \
char _buf[300]; \
+ char *_buf2 = _buf; \
struct printbuf out = PBUF(_buf); \
\
- btree_err_msg(&out, c, b, i, b->written, write); \
+ _buf2 = kmalloc(4096, GFP_ATOMIC); \
+ if (_buf2) \
+ out = _PBUF(_buf2, 4986); \
+ \
+ btree_err_msg(&out, c, ca, b, i, b->written, write); \
pr_buf(&out, ": " msg, ##__VA_ARGS__); \
\
if (type == BTREE_ERR_FIXABLE && \
write == READ && \
!test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags)) { \
- mustfix_fsck_err(c, "%s", _buf); \
+ mustfix_fsck_err(c, "%s", _buf2); \
goto out; \
} \
\
switch (write) { \
case READ: \
- bch_err(c, "%s", _buf); \
+ bch_err(c, "%s", _buf2); \
\
switch (type) { \
case BTREE_ERR_FIXABLE: \
} \
break; \
case WRITE: \
- bch_err(c, "corrupt metadata before write: %s", _buf); \
+ bch_err(c, "corrupt metadata before write: %s", _buf2); \
\
if (bch2_fs_inconsistent(c)) { \
ret = BCH_FSCK_ERRORS_NOT_FIXED; \
break; \
} \
out: \
+ if (_buf2 != _buf) \
+ kfree(_buf2); \
true; \
})
#define btree_err_on(cond, ...) ((cond) ? btree_err(__VA_ARGS__) : false)
-static int validate_bset(struct bch_fs *c, struct btree *b,
- struct bset *i, unsigned sectors,
- int write, bool have_retry)
+static int validate_bset(struct bch_fs *c, struct bch_dev *ca,
+ struct btree *b, struct bset *i,
+ unsigned sectors, int write, bool have_retry)
{
unsigned version = le16_to_cpu(i->version);
const char *err;
+ char buf1[100];
+ char buf2[100];
int ret = 0;
btree_err_on((version != BCH_BSET_VERSION_OLD &&
version < bcachefs_metadata_version_min) ||
version >= bcachefs_metadata_version_max,
- BTREE_ERR_FATAL, c, b, i,
+ BTREE_ERR_FATAL, c, ca, b, i,
"unsupported bset version");
+ if (btree_err_on(version < c->sb.version_min,
+ BTREE_ERR_FIXABLE, c, NULL, b, i,
+ "bset version %u older than superblock version_min %u",
+ version, c->sb.version_min)) {
+ mutex_lock(&c->sb_lock);
+ c->disk_sb.sb->version_min = cpu_to_le16(version);
+ bch2_write_super(c);
+ mutex_unlock(&c->sb_lock);
+ }
+
+ if (btree_err_on(version > c->sb.version,
+ BTREE_ERR_FIXABLE, c, NULL, b, i,
+ "bset version %u newer than superblock version %u",
+ version, c->sb.version)) {
+ mutex_lock(&c->sb_lock);
+ c->disk_sb.sb->version = cpu_to_le16(version);
+ bch2_write_super(c);
+ mutex_unlock(&c->sb_lock);
+ }
+
+ btree_err_on(BSET_SEPARATE_WHITEOUTS(i),
+ BTREE_ERR_FATAL, c, ca, b, i,
+ "BSET_SEPARATE_WHITEOUTS no longer supported");
+
if (btree_err_on(b->written + sectors > c->opts.btree_node_size,
- BTREE_ERR_FIXABLE, c, b, i,
+ BTREE_ERR_FIXABLE, c, ca, b, i,
"bset past end of btree node")) {
i->u64s = 0;
return 0;
}
btree_err_on(b->written && !i->u64s,
- BTREE_ERR_FIXABLE, c, b, i,
+ BTREE_ERR_FIXABLE, c, ca, b, i,
"empty bset");
if (!b->written) {
/* XXX endianness */
btree_err_on(bp->seq != bn->keys.seq,
- BTREE_ERR_MUST_RETRY, c, b, NULL,
+ BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
"incorrect sequence number (wrong btree node)");
}
btree_err_on(BTREE_NODE_ID(bn) != b->c.btree_id,
- BTREE_ERR_MUST_RETRY, c, b, i,
+ BTREE_ERR_MUST_RETRY, c, ca, b, i,
"incorrect btree id");
btree_err_on(BTREE_NODE_LEVEL(bn) != b->c.level,
- BTREE_ERR_MUST_RETRY, c, b, i,
+ BTREE_ERR_MUST_RETRY, c, ca, b, i,
"incorrect level");
- if (BSET_BIG_ENDIAN(i) != CPU_BIG_ENDIAN) {
- u64 *p = (u64 *) &bn->ptr;
-
- *p = swab64(*p);
- }
-
if (!write)
compat_btree_node(b->c.level, b->c.btree_id, version,
BSET_BIG_ENDIAN(i), write, bn);
struct bch_btree_ptr_v2 *bp =
&bkey_i_to_btree_ptr_v2(&b->key)->v;
- btree_err_on(bkey_cmp(b->data->min_key, bp->min_key),
- BTREE_ERR_MUST_RETRY, c, b, NULL,
- "incorrect min_key: got %llu:%llu should be %llu:%llu",
- b->data->min_key.inode,
- b->data->min_key.offset,
- bp->min_key.inode,
- bp->min_key.offset);
+ if (BTREE_PTR_RANGE_UPDATED(bp)) {
+ b->data->min_key = bp->min_key;
+ b->data->max_key = b->key.k.p;
+ }
+
+ btree_err_on(bpos_cmp(b->data->min_key, bp->min_key),
+ BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
+ "incorrect min_key: got %s should be %s",
+ (bch2_bpos_to_text(&PBUF(buf1), bn->min_key), buf1),
+ (bch2_bpos_to_text(&PBUF(buf2), bp->min_key), buf2));
}
- btree_err_on(bkey_cmp(bn->max_key, b->key.k.p),
- BTREE_ERR_MUST_RETRY, c, b, i,
- "incorrect max key %llu:%llu",
- bn->max_key.inode,
- bn->max_key.offset);
+ btree_err_on(bpos_cmp(bn->max_key, b->key.k.p),
+ BTREE_ERR_MUST_RETRY, c, ca, b, i,
+ "incorrect max key %s",
+ (bch2_bpos_to_text(&PBUF(buf1), bn->max_key), buf1));
if (write)
compat_btree_node(b->c.level, b->c.btree_id, version,
BSET_BIG_ENDIAN(i), write, bn);
- /* XXX: ideally we would be validating min_key too */
-#if 0
- /*
- * not correct anymore, due to btree node write error
- * handling
- *
- * need to add bn->seq to btree keys and verify
- * against that
- */
- btree_err_on(!extent_contains_ptr(bkey_i_to_s_c_extent(&b->key),
- bn->ptr),
- BTREE_ERR_FATAL, c, b, i,
- "incorrect backpointer");
-#endif
err = bch2_bkey_format_validate(&bn->format);
btree_err_on(err,
- BTREE_ERR_FATAL, c, b, i,
+ BTREE_ERR_FATAL, c, ca, b, i,
"invalid bkey format: %s", err);
compat_bformat(b->c.level, b->c.btree_id, version,
{
unsigned version = le16_to_cpu(i->version);
struct bkey_packed *k, *prev = NULL;
- bool seen_non_whiteout = false;
int ret = 0;
- if (!BSET_SEPARATE_WHITEOUTS(i)) {
- seen_non_whiteout = true;
- *whiteout_u64s = 0;
- }
-
for (k = i->start;
k != vstruct_last(i);) {
struct bkey_s u;
const char *invalid;
if (btree_err_on(bkey_next(k) > vstruct_last(i),
- BTREE_ERR_FIXABLE, c, b, i,
+ BTREE_ERR_FIXABLE, c, NULL, b, i,
"key extends past end of bset")) {
i->u64s = cpu_to_le16((u64 *) k - i->_data);
break;
}
if (btree_err_on(k->format > KEY_FORMAT_CURRENT,
- BTREE_ERR_FIXABLE, c, b, i,
+ BTREE_ERR_FIXABLE, c, NULL, b, i,
"invalid bkey format %u", k->format)) {
i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - k->u64s);
memmove_u64s_down(k, bkey_next(k),
char buf[160];
bch2_bkey_val_to_text(&PBUF(buf), c, u.s_c);
- btree_err(BTREE_ERR_FIXABLE, c, b, i,
- "invalid bkey:\n%s\n%s", invalid, buf);
+ btree_err(BTREE_ERR_FIXABLE, c, NULL, b, i,
+ "invalid bkey: %s\n%s", invalid, buf);
i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - k->u64s);
memmove_u64s_down(k, bkey_next(k),
BSET_BIG_ENDIAN(i), write,
&b->format, k);
- /*
- * with the separate whiteouts thing (used for extents), the
- * second set of keys actually can have whiteouts too, so we
- * can't solely go off bkey_whiteout()...
- */
-
- if (!seen_non_whiteout &&
- (!bkey_whiteout(k) ||
- (prev && bkey_iter_cmp(b, prev, k) > 0))) {
- *whiteout_u64s = k->_data - i->_data;
- seen_non_whiteout = true;
- } else if (prev && bkey_iter_cmp(b, prev, k) > 0) {
+ if (prev && bkey_iter_cmp(b, prev, k) > 0) {
char buf1[80];
char buf2[80];
struct bkey up = bkey_unpack_key(b, prev);
bch2_bkey_to_text(&PBUF(buf2), u.k);
bch2_dump_bset(c, b, i, 0);
- btree_err(BTREE_ERR_FATAL, c, b, i,
- "keys out of order: %s > %s",
- buf1, buf2);
- /* XXX: repair this */
+
+ if (btree_err(BTREE_ERR_FIXABLE, c, NULL, b, i,
+ "keys out of order: %s > %s",
+ buf1, buf2)) {
+ i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - k->u64s);
+ memmove_u64s_down(k, bkey_next(k),
+ (u64 *) vstruct_end(i) - (u64 *) k);
+ continue;
+ }
}
prev = k;
- k = bkey_next_skip_noops(k, vstruct_last(i));
+ k = bkey_next(k);
}
fsck_err:
return ret;
}
-int bch2_btree_node_read_done(struct bch_fs *c, struct btree *b, bool have_retry)
+int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
+ struct btree *b, bool have_retry)
{
struct btree_node_entry *bne;
struct sort_iter *iter;
unsigned u64s;
int ret, retry_read = 0, write = READ;
+ b->version_ondisk = U16_MAX;
+
iter = mempool_alloc(&c->fill_iter, GFP_NOIO);
sort_iter_init(iter, b);
iter->size = (btree_blocks(c) + 1) * 2;
if (bch2_meta_read_fault("btree"))
- btree_err(BTREE_ERR_MUST_RETRY, c, b, NULL,
+ btree_err(BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
"dynamic fault");
btree_err_on(le64_to_cpu(b->data->magic) != bset_magic(c),
- BTREE_ERR_MUST_RETRY, c, b, NULL,
+ BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
"bad magic");
btree_err_on(!b->data->keys.seq,
- BTREE_ERR_MUST_RETRY, c, b, NULL,
+ BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
"bad btree header");
if (b->key.k.type == KEY_TYPE_btree_ptr_v2) {
&bkey_i_to_btree_ptr_v2(&b->key)->v;
btree_err_on(b->data->keys.seq != bp->seq,
- BTREE_ERR_MUST_RETRY, c, b, NULL,
+ BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
"got wrong btree node (seq %llx want %llx)",
b->data->keys.seq, bp->seq);
}
i = &b->data->keys;
btree_err_on(!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i)),
- BTREE_ERR_WANT_RETRY, c, b, i,
+ BTREE_ERR_WANT_RETRY, c, ca, b, i,
"unknown checksum type %llu",
BSET_CSUM_TYPE(i));
csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, b->data);
btree_err_on(bch2_crc_cmp(csum, b->data->csum),
- BTREE_ERR_WANT_RETRY, c, b, i,
+ BTREE_ERR_WANT_RETRY, c, ca, b, i,
"invalid checksum");
bset_encrypt(c, i, b->written << 9);
- if (btree_node_is_extents(b) &&
- !BTREE_NODE_NEW_EXTENT_OVERWRITE(b->data)) {
- set_btree_node_old_extent_overwrite(b);
- set_btree_node_need_rewrite(b);
- }
+ btree_err_on(btree_node_is_extents(b) &&
+ !BTREE_NODE_NEW_EXTENT_OVERWRITE(b->data),
+ BTREE_ERR_FATAL, c, NULL, b, NULL,
+ "btree node does not have NEW_EXTENT_OVERWRITE set");
sectors = vstruct_sectors(b->data, c->block_bits);
} else {
break;
btree_err_on(!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i)),
- BTREE_ERR_WANT_RETRY, c, b, i,
+ BTREE_ERR_WANT_RETRY, c, ca, b, i,
"unknown checksum type %llu",
BSET_CSUM_TYPE(i));
csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne);
btree_err_on(bch2_crc_cmp(csum, bne->csum),
- BTREE_ERR_WANT_RETRY, c, b, i,
+ BTREE_ERR_WANT_RETRY, c, ca, b, i,
"invalid checksum");
bset_encrypt(c, i, b->written << 9);
sectors = vstruct_sectors(bne, c->block_bits);
}
- ret = validate_bset(c, b, i, sectors,
+ b->version_ondisk = min(b->version_ondisk,
+ le16_to_cpu(i->version));
+
+ ret = validate_bset(c, ca, b, i, sectors,
READ, have_retry);
if (ret)
goto fsck_err;
true);
btree_err_on(blacklisted && first,
- BTREE_ERR_FIXABLE, c, b, i,
+ BTREE_ERR_FIXABLE, c, ca, b, i,
"first btree node bset has blacklisted journal seq");
if (blacklisted && !first)
continue;
bset_byte_offset(b, bne) < btree_bytes(c);
bne = (void *) bne + block_bytes(c))
btree_err_on(bne->keys.seq == b->data->keys.seq,
- BTREE_ERR_WANT_RETRY, c, b, NULL,
+ BTREE_ERR_WANT_RETRY, c, ca, b, NULL,
"found bset signature after last bset");
sorted = btree_bounce_alloc(c, btree_bytes(c), &used_mempool);
set_btree_bset(b, b->set, &b->data->keys);
- b->nr = (btree_node_old_extent_overwrite(b)
- ? bch2_extent_sort_fix_overlapping
- : bch2_key_sort_fix_overlapping)(c, &sorted->keys, iter);
+ b->nr = bch2_key_sort_fix_overlapping(c, &sorted->keys, iter);
u64s = le16_to_cpu(sorted->keys.u64s);
*sorted = *b->data;
const char *invalid = bch2_bkey_val_invalid(c, u.s_c);
if (invalid ||
- (inject_invalid_keys(c) &&
+ (bch2_inject_invalid_keys &&
!bversion_cmp(u.k->version, MAX_VERSION))) {
char buf[160];
bch2_bkey_val_to_text(&PBUF(buf), c, u.s_c);
- btree_err(BTREE_ERR_FIXABLE, c, b, i,
+ btree_err(BTREE_ERR_FIXABLE, c, NULL, b, i,
"invalid bkey %s: %s", buf, invalid);
btree_keys_account_key_drop(&b->nr, 0, k);
bp.v->mem_ptr = 0;
}
- k = bkey_next_skip_noops(k, vstruct_last(i));
+ k = bkey_next(k);
}
bch2_bset_build_aux_tree(b, b->set, false);
bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(&b->key)), ptr) {
struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
- if (ca->mi.state != BCH_MEMBER_STATE_RW)
+ if (ca->mi.state != BCH_MEMBER_STATE_rw)
set_btree_node_need_rewrite(b);
}
out:
struct btree *b = rb->bio.bi_private;
struct bio *bio = &rb->bio;
struct bch_io_failures failed = { .nr = 0 };
+ char buf[200];
+ struct printbuf out;
bool can_retry;
goto start;
bio->bi_status = BLK_STS_REMOVED;
}
start:
- bch2_dev_io_err_on(bio->bi_status, ca, "btree read: %s",
- bch2_blk_status_to_str(bio->bi_status));
+ out = PBUF(buf);
+ btree_pos_to_text(&out, c, b);
+ bch2_dev_io_err_on(bio->bi_status, ca, "btree read error %s for %s",
+ bch2_blk_status_to_str(bio->bi_status), buf);
if (rb->have_ioref)
percpu_ref_put(&ca->io_ref);
rb->have_ioref = false;
&failed, &rb->pick) > 0;
if (!bio->bi_status &&
- !bch2_btree_node_read_done(c, b, can_retry))
+ !bch2_btree_node_read_done(c, ca, b, can_retry))
break;
if (!can_retry) {
struct btree_write_bio *wbio)
{
struct btree *b = wbio->wbio.bio.bi_private;
- __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp;
+ struct bkey_buf k;
struct bch_extent_ptr *ptr;
struct btree_trans trans;
struct btree_iter *iter;
int ret;
+ bch2_bkey_buf_init(&k);
bch2_trans_init(&trans, c, 0, 0);
iter = bch2_trans_get_node_iter(&trans, b->c.btree_id, b->key.k.p,
BUG_ON(!btree_node_hashed(b));
- bkey_copy(&tmp.k, &b->key);
+ bch2_bkey_buf_copy(&k, c, &b->key);
- bch2_bkey_drop_ptrs(bkey_i_to_s(&tmp.k), ptr,
+ bch2_bkey_drop_ptrs(bkey_i_to_s(k.k), ptr,
bch2_dev_list_has_dev(wbio->wbio.failed, ptr->dev));
- if (!bch2_bkey_nr_ptrs(bkey_i_to_s_c(&tmp.k)))
+ if (!bch2_bkey_nr_ptrs(bkey_i_to_s_c(k.k)))
goto err;
- ret = bch2_btree_node_update_key(c, iter, b, &tmp.k);
+ ret = bch2_btree_node_update_key(c, iter, b, k.k);
if (ret == -EINTR)
goto retry;
if (ret)
goto err;
out:
+ bch2_trans_iter_put(&trans, iter);
bch2_trans_exit(&trans);
+ bch2_bkey_buf_exit(&k, c);
bio_put(&wbio->wbio.bio);
btree_node_write_done(c, b);
return;
if (wbio->have_ioref)
bch2_latency_acct(ca, wbio->submit_time, WRITE);
- if (bch2_dev_io_err_on(bio->bi_status, ca, "btree write: %s",
+ if (bch2_dev_io_err_on(bio->bi_status, ca, "btree write error: %s",
bch2_blk_status_to_str(bio->bi_status)) ||
bch2_meta_write_fault("btree")) {
spin_lock_irqsave(&c->btree_write_error_lock, flags);
unsigned whiteout_u64s = 0;
int ret;
- if (bch2_bkey_invalid(c, bkey_i_to_s_c(&b->key), BKEY_TYPE_BTREE))
+ if (bch2_bkey_invalid(c, bkey_i_to_s_c(&b->key), BKEY_TYPE_btree))
return -1;
- ret = validate_bset(c, b, i, sectors, WRITE, false) ?:
- validate_bset_keys(c, b, i, &whiteout_u64s, WRITE, false);
- if (ret)
+ ret = validate_bset_keys(c, b, i, &whiteout_u64s, WRITE, false) ?:
+ validate_bset(c, NULL, b, i, sectors, WRITE, false);
+ if (ret) {
bch2_inconsistent_error(c);
+ dump_stack();
+ }
return ret;
}
struct bset *i;
struct btree_node *bn = NULL;
struct btree_node_entry *bne = NULL;
- BKEY_PADDED(key) k;
+ struct bkey_buf k;
struct bch_extent_ptr *ptr;
struct sort_iter sort_iter;
struct nonce nonce;
bool validate_before_checksum = false;
void *data;
+ bch2_bkey_buf_init(&k);
+
if (test_bit(BCH_FS_HOLD_BTREE_WRITES, &c->flags))
return;
if (!btree_node_may_write(b))
return;
+ if (old & (1 << BTREE_NODE_never_write))
+ return;
+
if (old & (1 << BTREE_NODE_write_in_flight)) {
btree_node_wait_on_io(b);
continue;
new ^= (1 << BTREE_NODE_write_idx);
} while (cmpxchg_acquire(&b->flags, old, new) != old);
+ atomic_dec(&c->btree_cache.dirty);
+
BUG_ON(btree_node_fake(b));
BUG_ON((b->will_make_reachable != 0) != !b->written);
seq = max(seq, le64_to_cpu(i->journal_seq));
}
+ BUG_ON(b->written && !seq);
+
+ /* bch2_varint_decode may read up to 7 bytes past the end of the buffer: */
+ bytes += 8;
+
data = btree_bounce_alloc(c, bytes, &used_mempool);
if (!b->written) {
i->journal_seq = cpu_to_le64(seq);
i->u64s = 0;
- if (!btree_node_old_extent_overwrite(b)) {
- sort_iter_add(&sort_iter,
- unwritten_whiteouts_start(c, b),
- unwritten_whiteouts_end(c, b));
- SET_BSET_SEPARATE_WHITEOUTS(i, false);
- } else {
- memcpy_u64s(i->start,
- unwritten_whiteouts_start(c, b),
- b->whiteout_u64s);
- i->u64s = cpu_to_le16(b->whiteout_u64s);
- SET_BSET_SEPARATE_WHITEOUTS(i, true);
- }
+ sort_iter_add(&sort_iter,
+ unwritten_whiteouts_start(c, b),
+ unwritten_whiteouts_end(c, b));
+ SET_BSET_SEPARATE_WHITEOUTS(i, false);
b->whiteout_u64s = 0;
- u64s = btree_node_old_extent_overwrite(b)
- ? bch2_sort_extents(vstruct_last(i), &sort_iter, false)
- : bch2_sort_keys(i->start, &sort_iter, false);
+ u64s = bch2_sort_keys(i->start, &sort_iter, false);
le16_add_cpu(&i->u64s, u64s);
set_needs_whiteout(i, false);
validate_before_checksum = true;
/* validate_bset will be modifying: */
- if (le16_to_cpu(i->version) < bcachefs_metadata_version_max)
+ if (le16_to_cpu(i->version) < bcachefs_metadata_version_current)
validate_before_checksum = true;
/* if we're going to be encrypting, check metadata validity first: */
* just make all btree node writes FUA to keep things sane.
*/
- bkey_copy(&k.key, &b->key);
+ bch2_bkey_buf_copy(&k, c, &b->key);
- bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(&k.key)), ptr)
+ bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(k.k)), ptr)
ptr->offset += b->written;
b->written += sectors_to_write;
+ atomic64_inc(&c->btree_writes_nr);
+ atomic64_add(sectors_to_write, &c->btree_writes_sectors);
+
/* XXX: submitting IO with btree locks held: */
- bch2_submit_wbio_replicas(&wbio->wbio, c, BCH_DATA_btree, &k.key);
+ bch2_submit_wbio_replicas(&wbio->wbio, c, BCH_DATA_btree, k.k);
+ bch2_bkey_buf_exit(&k, c);
return;
err:
set_btree_node_noevict(b);
__bch2_btree_flush_all(c, BTREE_NODE_write_in_flight);
}
-void bch2_btree_verify_flushed(struct bch_fs *c)
-{
- struct bucket_table *tbl;
- struct rhash_head *pos;
- struct btree *b;
- unsigned i;
-
- rcu_read_lock();
- for_each_cached_btree(b, c, tbl, i, pos) {
- unsigned long flags = READ_ONCE(b->flags);
-
- BUG_ON((flags & (1 << BTREE_NODE_dirty)) ||
- (flags & (1 << BTREE_NODE_write_in_flight)));
- }
- rcu_read_unlock();
-}
-
void bch2_dirty_btree_nodes_to_text(struct printbuf *out, struct bch_fs *c)
{
struct bucket_table *tbl;
projects / bcachefs-tools-debian / blobdiff