void bch2_btree_node_io_lock(struct btree *b)
{
- BUG_ON(lock_class_is_held(&bch2_btree_node_lock_key));
+ bch2_assert_btree_nodes_not_locked();
wait_on_bit_lock_io(&b->flags, BTREE_NODE_write_in_flight,
TASK_UNINTERRUPTIBLE);
void bch2_btree_node_wait_on_read(struct btree *b)
{
- BUG_ON(lock_class_is_held(&bch2_btree_node_lock_key));
+ bch2_assert_btree_nodes_not_locked();
wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight,
TASK_UNINTERRUPTIBLE);
void bch2_btree_node_wait_on_write(struct btree *b)
{
- BUG_ON(lock_class_is_held(&bch2_btree_node_lock_key));
+ bch2_assert_btree_nodes_not_locked();
wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight,
TASK_UNINTERRUPTIBLE);
if (start == end)
return;
- for (p = start, k = bkey_next(start);
+ for (p = start, k = bkey_p_next(start);
k != end;
- p = k, k = bkey_next(k)) {
+ p = k, k = bkey_p_next(k)) {
struct bkey l = bkey_unpack_key(b, p);
struct bkey r = bkey_unpack_key(b, k);
- BUG_ON(bpos_cmp(l.p, bkey_start_pos(&r)) >= 0);
+ BUG_ON(bpos_ge(l.p, bkey_start_pos(&r)));
}
#endif
}
{
struct bkey_packed *k;
- for (k = i->start; k != vstruct_last(i); k = bkey_next(k))
+ for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k))
k->needs_whiteout = v;
}
for (k = unwritten_whiteouts_start(c, b);
k != unwritten_whiteouts_end(c, b);
- k = bkey_next(k))
+ k = bkey_p_next(k))
*--ptrs = k;
sort_bkey_ptrs(b, ptrs, ptrs_end - ptrs);
while (ptrs != ptrs_end) {
bkey_copy(k, *ptrs);
- k = bkey_next(k);
+ k = bkey_p_next(k);
ptrs++;
}
out = i->start;
for (k = start; k != end; k = n) {
- n = bkey_next(k);
+ n = bkey_p_next(k);
if (!bkey_deleted(k)) {
bkey_copy(out, k);
- out = bkey_next(out);
+ out = bkey_p_next(out);
} else {
BUG_ON(k->needs_whiteout);
}
t == bset_tree_last(b));
}
+/*
+ * If we have MAX_BSETS (3) bsets, should we sort them all down to just one?
+ *
+ * The first bset is going to be of similar order to the size of the node, the
+ * last bset is bounded by btree_write_set_buffer(), which is set to keep the
+ * memmove on insert from being too expensive: the middle bset should, ideally,
+ * be the geometric mean of the first and the last.
+ *
+ * Returns true if the middle bset is greater than that geometric mean:
+ */
+static inline bool should_compact_all(struct bch_fs *c, struct btree *b)
+{
+ unsigned mid_u64s_bits =
+ (ilog2(btree_max_u64s(c)) + BTREE_WRITE_SET_U64s_BITS) / 2;
+
+ return bset_u64s(&b->set[1]) > 1U << mid_u64s_bits;
+}
+
/*
* @bch_btree_init_next - initialize a new (unwritten) bset that can then be
* inserted into
EBUG_ON(!(b->c.lock.state.seq & 1));
BUG_ON(bset_written(b, bset(b, &b->set[1])));
+ BUG_ON(btree_node_just_written(b));
if (b->nsets == MAX_BSETS &&
- !btree_node_write_in_flight(b)) {
- unsigned log_u64s[] = {
- ilog2(bset_u64s(&b->set[0])),
- ilog2(bset_u64s(&b->set[1])),
- ilog2(bset_u64s(&b->set[2])),
- };
-
- if (log_u64s[1] >= (log_u64s[0] + log_u64s[2]) / 2) {
- bch2_btree_node_write(c, b, SIX_LOCK_write, 0);
- reinit_iter = true;
- }
+ !btree_node_write_in_flight(b) &&
+ should_compact_all(c, b)) {
+ bch2_btree_node_write(c, b, SIX_LOCK_write,
+ BTREE_WRITE_init_next_bset);
+ reinit_iter = true;
}
if (b->nsets == MAX_BSETS &&
static void btree_pos_to_text(struct printbuf *out, struct bch_fs *c,
struct btree *b)
{
- pr_buf(out, "%s level %u/%u\n ",
+ prt_printf(out, "%s level %u/%u\n ",
bch2_btree_ids[b->c.btree_id],
b->c.level,
c->btree_roots[b->c.btree_id].level);
struct btree *b, struct bset *i,
unsigned offset, int write)
{
- pr_buf(out, "error validating btree node ");
- if (write)
- pr_buf(out, "before write ");
+ prt_printf(out, bch2_log_msg(c, "%s"),
+ write == READ
+ ? "error validating btree node "
+ : "corrupt btree node before write ");
if (ca)
- pr_buf(out, "on %s ", ca->name);
- pr_buf(out, "at btree ");
+ prt_printf(out, "on %s ", ca->name);
+ prt_printf(out, "at btree ");
btree_pos_to_text(out, c, b);
- pr_buf(out, "\n node offset %u", b->written);
+ prt_printf(out, "\n node offset %u", b->written);
if (i)
- pr_buf(out, " bset u64s %u", le16_to_cpu(i->u64s));
+ prt_printf(out, " bset u64s %u", le16_to_cpu(i->u64s));
+ prt_str(out, ": ");
}
enum btree_err_type {
+ /*
+ * We can repair this locally, and we're after the checksum check so
+ * there's no need to try another replica:
+ */
BTREE_ERR_FIXABLE,
+ /*
+ * We can repair this if we have to, but we should try reading another
+ * replica if we can:
+ */
BTREE_ERR_WANT_RETRY,
+ /*
+ * Read another replica if we have one, otherwise consider the whole
+ * node bad:
+ */
BTREE_ERR_MUST_RETRY,
- BTREE_ERR_FATAL,
+ BTREE_ERR_BAD_NODE,
+ BTREE_ERR_INCOMPATIBLE,
};
enum btree_validate_ret {
BTREE_RETRY_READ = 64,
};
+static int __btree_err(enum btree_err_type type,
+ struct bch_fs *c,
+ struct bch_dev *ca,
+ struct btree *b,
+ struct bset *i,
+ int write,
+ bool have_retry,
+ const char *fmt, ...)
+{
+ struct printbuf out = PRINTBUF;
+ va_list args;
+ int ret = -BCH_ERR_fsck_fix;
+
+ btree_err_msg(&out, c, ca, b, i, b->written, write);
+
+ va_start(args, fmt);
+ prt_vprintf(&out, fmt, args);
+ va_end(args);
+
+ if (write == WRITE) {
+ bch2_print_string_as_lines(KERN_ERR, out.buf);
+ ret = c->opts.errors == BCH_ON_ERROR_continue
+ ? 0
+ : -BCH_ERR_fsck_errors_not_fixed;
+ goto out;
+ }
+
+ if (!have_retry && type == BTREE_ERR_WANT_RETRY)
+ type = BTREE_ERR_FIXABLE;
+ if (!have_retry && type == BTREE_ERR_MUST_RETRY)
+ type = BTREE_ERR_BAD_NODE;
+
+ switch (type) {
+ case BTREE_ERR_FIXABLE:
+ mustfix_fsck_err(c, "%s", out.buf);
+ ret = -BCH_ERR_fsck_fix;
+ break;
+ case BTREE_ERR_WANT_RETRY:
+ case BTREE_ERR_MUST_RETRY:
+ bch2_print_string_as_lines(KERN_ERR, out.buf);
+ ret = BTREE_RETRY_READ;
+ break;
+ case BTREE_ERR_BAD_NODE:
+ bch2_print_string_as_lines(KERN_ERR, out.buf);
+ bch2_topology_error(c);
+ ret = -BCH_ERR_need_topology_repair;
+ break;
+ case BTREE_ERR_INCOMPATIBLE:
+ bch2_print_string_as_lines(KERN_ERR, out.buf);
+ ret = -BCH_ERR_fsck_errors_not_fixed;
+ break;
+ default:
+ BUG();
+ }
+out:
+fsck_err:
+ printbuf_exit(&out);
+ return ret;
+}
+
#define btree_err(type, c, ca, b, i, msg, ...) \
({ \
- __label__ out; \
- struct printbuf out = PRINTBUF; \
- \
- 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", out.buf); \
- goto out; \
- } \
- \
- switch (write) { \
- case READ: \
- bch_err(c, "%s", out.buf); \
+ int _ret = __btree_err(type, c, ca, b, i, write, have_retry, msg, ##__VA_ARGS__);\
\
- switch (type) { \
- case BTREE_ERR_FIXABLE: \
- ret = BCH_FSCK_ERRORS_NOT_FIXED; \
- goto fsck_err; \
- case BTREE_ERR_WANT_RETRY: \
- if (have_retry) { \
- ret = BTREE_RETRY_READ; \
- goto fsck_err; \
- } \
- break; \
- case BTREE_ERR_MUST_RETRY: \
- ret = BTREE_RETRY_READ; \
- goto fsck_err; \
- case BTREE_ERR_FATAL: \
- ret = BCH_FSCK_ERRORS_NOT_FIXED; \
- goto fsck_err; \
- } \
- break; \
- case WRITE: \
- bch_err(c, "corrupt metadata before write: %s", out.buf);\
- \
- if (bch2_fs_inconsistent(c)) { \
- ret = BCH_FSCK_ERRORS_NOT_FIXED; \
- goto fsck_err; \
- } \
- break; \
- } \
-out: \
- printbuf_exit(&out); \
- true; \
+ if (_ret != -BCH_ERR_fsck_fix) \
+ goto fsck_err; \
+ *saw_error = true; \
})
#define btree_err_on(cond, ...) ((cond) ? btree_err(__VA_ARGS__) : false)
* When btree topology repair changes the start or end of a node, that might
* mean we have to drop keys that are no longer inside the node:
*/
+__cold
void bch2_btree_node_drop_keys_outside_node(struct btree *b)
{
struct bset_tree *t;
struct bset *i = bset(b, t);
struct bkey_packed *k;
- for (k = i->start; k != vstruct_last(i); k = bkey_next(k))
+ for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k))
if (bkey_cmp_left_packed(b, k, &b->data->min_key) >= 0)
break;
(u64 *) vstruct_end(i) - (u64 *) k);
i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - shift);
set_btree_bset_end(b, t);
- bch2_bset_set_no_aux_tree(b, t);
}
- for (k = i->start; k != vstruct_last(i); k = bkey_next(k))
+ for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k))
if (bkey_cmp_left_packed(b, k, &b->data->max_key) > 0)
break;
if (k != vstruct_last(i)) {
i->u64s = cpu_to_le16((u64 *) k - (u64 *) i->start);
set_btree_bset_end(b, t);
- bch2_bset_set_no_aux_tree(b, t);
}
}
+ /*
+ * Always rebuild search trees: eytzinger search tree nodes directly
+ * depend on the values of min/max key:
+ */
+ bch2_bset_set_no_aux_tree(b, b->set);
bch2_btree_build_aux_trees(b);
for_each_btree_node_key_unpack(b, k, &iter, &unpacked) {
- BUG_ON(bpos_cmp(k.k->p, b->data->min_key) < 0);
- BUG_ON(bpos_cmp(k.k->p, b->data->max_key) > 0);
+ BUG_ON(bpos_lt(k.k->p, b->data->min_key));
+ BUG_ON(bpos_gt(k.k->p, b->data->max_key));
}
}
static int validate_bset(struct bch_fs *c, struct bch_dev *ca,
struct btree *b, struct bset *i,
unsigned offset, unsigned sectors,
- int write, bool have_retry)
+ int write, bool have_retry, bool *saw_error)
{
unsigned version = le16_to_cpu(i->version);
const char *err;
btree_err_on((version != BCH_BSET_VERSION_OLD &&
version < bcachefs_metadata_version_min) ||
version >= bcachefs_metadata_version_max,
- BTREE_ERR_FATAL, c, ca, b, i,
+ BTREE_ERR_INCOMPATIBLE, c, ca, b, i,
"unsupported bset version");
if (btree_err_on(version < c->sb.version_min,
}
btree_err_on(BSET_SEPARATE_WHITEOUTS(i),
- BTREE_ERR_FATAL, c, ca, b, i,
+ BTREE_ERR_INCOMPATIBLE, c, ca, b, i,
"BSET_SEPARATE_WHITEOUTS no longer supported");
if (btree_err_on(offset + sectors > btree_sectors(c),
b->data->max_key = b->key.k.p;
}
- btree_err_on(bpos_cmp(b->data->min_key, bp->min_key),
+ btree_err_on(!bpos_eq(b->data->min_key, bp->min_key),
BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
"incorrect min_key: got %s should be %s",
(printbuf_reset(&buf1),
bch2_bpos_to_text(&buf2, bp->min_key), buf2.buf));
}
- btree_err_on(bpos_cmp(bn->max_key, b->key.k.p),
+ btree_err_on(!bpos_eq(bn->max_key, b->key.k.p),
BTREE_ERR_MUST_RETRY, c, ca, b, i,
"incorrect max key %s",
(printbuf_reset(&buf1),
err = bch2_bkey_format_validate(&bn->format);
btree_err_on(err,
- BTREE_ERR_FATAL, c, ca, b, i,
+ BTREE_ERR_BAD_NODE, c, ca, b, i,
"invalid bkey format: %s", err);
compat_bformat(b->c.level, b->c.btree_id, version,
}
static int validate_bset_keys(struct bch_fs *c, struct btree *b,
- struct bset *i, unsigned *whiteout_u64s,
- int write, bool have_retry)
+ struct bset *i, int write,
+ bool have_retry, bool *saw_error)
{
unsigned version = le16_to_cpu(i->version);
struct bkey_packed *k, *prev = NULL;
struct bkey_s u;
struct bkey tmp;
- if (btree_err_on(bkey_next(k) > vstruct_last(i),
+ if (btree_err_on(bkey_p_next(k) > vstruct_last(i),
BTREE_ERR_FIXABLE, c, NULL, b, i,
"key extends past end of bset")) {
i->u64s = cpu_to_le16((u64 *) k - i->_data);
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),
+ memmove_u64s_down(k, bkey_p_next(k),
(u64 *) vstruct_end(i) - (u64 *) k);
continue;
}
printbuf_reset(&buf);
if (bset_key_invalid(c, b, u.s_c, updated_range, write, &buf)) {
printbuf_reset(&buf);
- pr_buf(&buf, "invalid bkey: ");
+ prt_printf(&buf, "invalid bkey: ");
bset_key_invalid(c, b, u.s_c, updated_range, write, &buf);
- pr_buf(&buf, "\n ");
+ prt_printf(&buf, "\n ");
bch2_bkey_val_to_text(&buf, c, u.s_c);
btree_err(BTREE_ERR_FIXABLE, c, NULL, b, i, "%s", buf.buf);
i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - k->u64s);
- memmove_u64s_down(k, bkey_next(k),
+ memmove_u64s_down(k, bkey_p_next(k),
(u64 *) vstruct_end(i) - (u64 *) k);
continue;
}
struct bkey up = bkey_unpack_key(b, prev);
printbuf_reset(&buf);
- pr_buf(&buf, "keys out of order: ");
+ prt_printf(&buf, "keys out of order: ");
bch2_bkey_to_text(&buf, &up);
- pr_buf(&buf, " > ");
+ prt_printf(&buf, " > ");
bch2_bkey_to_text(&buf, u.k);
bch2_dump_bset(c, b, i, 0);
if (btree_err(BTREE_ERR_FIXABLE, c, NULL, b, i, "%s", buf.buf)) {
i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - k->u64s);
- memmove_u64s_down(k, bkey_next(k),
+ memmove_u64s_down(k, bkey_p_next(k),
(u64 *) vstruct_end(i) - (u64 *) k);
continue;
}
}
prev = k;
- k = bkey_next(k);
+ k = bkey_p_next(k);
}
fsck_err:
printbuf_exit(&buf);
}
int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
- struct btree *b, bool have_retry)
+ struct btree *b, bool have_retry, bool *saw_error)
{
struct btree_node_entry *bne;
struct sort_iter *iter;
unsigned blacklisted_written, nonblacklisted_written = 0;
unsigned ptr_written = btree_ptr_sectors_written(&b->key);
struct printbuf buf = PRINTBUF;
- int ret, retry_read = 0, write = READ;
+ int ret = 0, retry_read = 0, write = READ;
b->version_ondisk = U16_MAX;
+ /* We might get called multiple times on read retry: */
+ b->written = 0;
iter = mempool_alloc(&c->fill_iter, GFP_NOIO);
sort_iter_init(iter, b);
}
while (b->written < (ptr_written ?: btree_sectors(c))) {
- unsigned sectors, whiteout_u64s = 0;
+ unsigned sectors;
struct nonce nonce;
struct bch_csum csum;
bool first = !b->written;
btree_err_on(btree_node_type_is_extents(btree_node_type(b)) &&
!BTREE_NODE_NEW_EXTENT_OVERWRITE(b->data),
- BTREE_ERR_FATAL, c, NULL, b, NULL,
+ BTREE_ERR_INCOMPATIBLE, c, NULL, b, NULL,
"btree node does not have NEW_EXTENT_OVERWRITE set");
sectors = vstruct_sectors(b->data, c->block_bits);
le16_to_cpu(i->version));
ret = validate_bset(c, ca, b, i, b->written, sectors,
- READ, have_retry);
+ READ, have_retry, saw_error);
if (ret)
goto fsck_err;
if (!b->written)
btree_node_set_format(b, b->data->format);
- ret = validate_bset_keys(c, b, i, &whiteout_u64s,
- READ, have_retry);
+ ret = validate_bset_keys(c, b, i, READ, have_retry, saw_error);
if (ret)
goto fsck_err;
if (blacklisted && !first)
continue;
- sort_iter_add(iter, i->start,
- vstruct_idx(i, whiteout_u64s));
-
sort_iter_add(iter,
- vstruct_idx(i, whiteout_u64s),
+ vstruct_idx(i, 0),
vstruct_last(i));
nonblacklisted_written = b->written;
!bversion_cmp(u.k->version, MAX_VERSION))) {
printbuf_reset(&buf);
- pr_buf(&buf, "invalid bkey: ");
+ prt_printf(&buf, "invalid bkey: ");
bch2_bkey_val_invalid(c, u.s_c, READ, &buf);
- pr_buf(&buf, "\n ");
+ prt_printf(&buf, "\n ");
bch2_bkey_val_to_text(&buf, c, u.s_c);
btree_err(BTREE_ERR_FIXABLE, c, NULL, b, i, "%s", buf.buf);
btree_keys_account_key_drop(&b->nr, 0, k);
i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - k->u64s);
- memmove_u64s_down(k, bkey_next(k),
+ memmove_u64s_down(k, bkey_p_next(k),
(u64 *) vstruct_end(i) - (u64 *) k);
set_btree_bset_end(b, b->set);
continue;
bp.v->mem_ptr = 0;
}
- k = bkey_next(k);
+ k = bkey_p_next(k);
}
bch2_bset_build_aux_tree(b, b->set, false);
printbuf_exit(&buf);
return retry_read;
fsck_err:
- if (ret == BTREE_RETRY_READ) {
+ if (ret == BTREE_RETRY_READ)
retry_read = 1;
- } else {
- bch2_inconsistent_error(c);
+ else
set_btree_node_read_error(b);
- }
goto out;
}
bch_info(c, "retrying read");
ca = bch_dev_bkey_exists(c, rb->pick.ptr.dev);
rb->have_ioref = bch2_dev_get_ioref(ca, READ);
- bio_reset(bio);
- bio->bi_opf = REQ_OP_READ|REQ_SYNC|REQ_META;
+ bio_reset(bio, NULL, REQ_OP_READ|REQ_SYNC|REQ_META);
bio->bi_iter.bi_sector = rb->pick.ptr.offset;
bio->bi_iter.bi_size = btree_bytes(c);
&failed, &rb->pick) > 0;
if (!bio->bi_status &&
- !bch2_btree_node_read_done(c, ca, b, can_retry)) {
+ !bch2_btree_node_read_done(c, ca, b, can_retry, &saw_error)) {
if (retry)
bch_info(c, "retry success");
break;
bio_put(&rb->bio);
printbuf_exit(&buf);
- if (saw_error && !btree_node_read_error(b))
+ if (saw_error && !btree_node_read_error(b)) {
+ struct printbuf buf = PRINTBUF;
+
+ bch2_bpos_to_text(&buf, b->key.k.p);
+ bch_info(c, "%s: rewriting btree node at btree=%s level=%u %s due to error",
+ __func__, bch2_btree_ids[b->c.btree_id], b->c.level, buf.buf);
+ printbuf_exit(&buf);
+
bch2_btree_node_rewrite_async(c, b);
+ }
clear_btree_node_read_in_flight(b);
wake_up_bit(&b->flags, BTREE_NODE_read_in_flight);
if (rb->have_ioref) {
struct bch_dev *ca = bch_dev_bkey_exists(c, rb->pick.ptr.dev);
+
bch2_latency_acct(ca, rb->start_time, READ);
}
unsigned i, written = 0, written2 = 0;
__le64 seq = b->key.k.type == KEY_TYPE_btree_ptr_v2
? bkey_i_to_btree_ptr_v2(&b->key)->v.seq : 0;
+ bool _saw_error = false, *saw_error = &_saw_error;
for (i = 0; i < ra->nr; i++) {
struct btree_node *bn = ra->buf[i];
sectors = vstruct_sectors(bne, c->block_bits);
}
- pr_buf(&buf, " %u-%u", offset, offset + sectors);
+ prt_printf(&buf, " %u-%u", offset, offset + sectors);
if (bne && bch2_journal_seq_is_blacklisted(c,
le64_to_cpu(bne->keys.journal_seq), false))
- pr_buf(&buf, "*");
+ prt_printf(&buf, "*");
offset += sectors;
}
bne = ra->buf[i] + (offset << 9);
if (bne->keys.seq == bn->keys.seq) {
if (!gap)
- pr_buf(&buf, " GAP");
+ prt_printf(&buf, " GAP");
gap = true;
sectors = vstruct_sectors(bne, c->block_bits);
- pr_buf(&buf, " %u-%u", offset, offset + sectors);
+ prt_printf(&buf, " %u-%u", offset, offset + sectors);
if (bch2_journal_seq_is_blacklisted(c,
le64_to_cpu(bne->keys.journal_seq), false))
- pr_buf(&buf, "*");
+ prt_printf(&buf, "*");
}
offset++;
}
if (best >= 0) {
memcpy(b->data, ra->buf[best], btree_bytes(c));
- ret = bch2_btree_node_read_done(c, NULL, b, false);
+ ret = bch2_btree_node_read_done(c, NULL, b, false, saw_error);
} else {
ret = -1;
}
if (ret)
set_btree_node_read_error(b);
+ else if (*saw_error)
+ bch2_btree_node_rewrite_async(c, b);
for (i = 0; i < ra->nr; i++) {
mempool_free(ra->buf[i], &c->btree_bounce_pool);
if (rb->have_ioref) {
struct bch_dev *ca = bch_dev_bkey_exists(c, rb->pick.ptr.dev);
+
bch2_latency_acct(ca, rb->start_time, READ);
}
for (i = 0; i < ra->nr; i++) {
ra->buf[i] = mempool_alloc(&c->btree_bounce_pool, GFP_NOFS);
- ra->bio[i] = bio_alloc_bioset(GFP_NOFS, buf_pages(ra->buf[i],
- btree_bytes(c)),
+ ra->bio[i] = bio_alloc_bioset(NULL,
+ buf_pages(ra->buf[i], btree_bytes(c)),
+ REQ_OP_READ|REQ_SYNC|REQ_META,
+ GFP_NOFS,
&c->btree_bio);
}
rb->have_ioref = bch2_dev_get_ioref(ca, READ);
rb->idx = i;
rb->pick = pick;
- rb->bio.bi_opf = REQ_OP_READ|REQ_SYNC|REQ_META;
rb->bio.bi_iter.bi_sector = pick.ptr.offset;
rb->bio.bi_end_io = btree_node_read_all_replicas_endio;
bch2_bio_map(&rb->bio, ra->buf[i], btree_bytes(c));
struct btree_read_bio *rb;
struct bch_dev *ca;
struct bio *bio;
- struct printbuf buf = PRINTBUF;
int ret;
- btree_pos_to_text(&buf, c, b);
- trace_btree_read(c, b);
+ trace_and_count(c, btree_node_read, c, b);
if (bch2_verify_all_btree_replicas &&
!btree_node_read_all_replicas(c, b, sync))
- goto out;
+ return;
ret = bch2_bkey_pick_read_device(c, bkey_i_to_s_c(&b->key),
NULL, &pick);
- if (bch2_fs_fatal_err_on(ret <= 0, c,
- "btree node read error: no device to read from\n"
- " at %s", buf.buf)) {
+
+ if (ret <= 0) {
+ struct printbuf buf = PRINTBUF;
+
+ prt_str(&buf, "btree node read error: no device to read from\n at ");
+ btree_pos_to_text(&buf, c, b);
+ bch_err(c, "%s", buf.buf);
+
+ if (test_bit(BCH_FS_TOPOLOGY_REPAIR_DONE, &c->flags))
+ bch2_fatal_error(c);
+
set_btree_node_read_error(b);
- goto out;
+ clear_btree_node_read_in_flight(b);
+ wake_up_bit(&b->flags, BTREE_NODE_read_in_flight);
+ printbuf_exit(&buf);
+ return;
}
ca = bch_dev_bkey_exists(c, pick.ptr.dev);
- bio = bio_alloc_bioset(GFP_NOIO, buf_pages(b->data,
- btree_bytes(c)),
+ bio = bio_alloc_bioset(NULL,
+ buf_pages(b->data, btree_bytes(c)),
+ REQ_OP_READ|REQ_SYNC|REQ_META,
+ GFP_NOIO,
&c->btree_bio);
rb = container_of(bio, struct btree_read_bio, bio);
rb->c = c;
rb->have_ioref = bch2_dev_get_ioref(ca, READ);
rb->pick = pick;
INIT_WORK(&rb->work, btree_node_read_work);
- bio->bi_opf = REQ_OP_READ|REQ_SYNC|REQ_META;
bio->bi_iter.bi_sector = pick.ptr.offset;
bio->bi_end_io = btree_node_read_endio;
bch2_bio_map(bio, b->data, btree_bytes(c));
else
queue_work(c->io_complete_wq, &rb->work);
}
-out:
- printbuf_exit(&buf);
}
-int bch2_btree_root_read(struct bch_fs *c, enum btree_id id,
- const struct bkey_i *k, unsigned level)
+static int __bch2_btree_root_read(struct btree_trans *trans, enum btree_id id,
+ const struct bkey_i *k, unsigned level)
{
+ struct bch_fs *c = trans->c;
struct closure cl;
struct btree *b;
int ret;
closure_sync(&cl);
} while (ret);
- b = bch2_btree_node_mem_alloc(c, level != 0);
+ b = bch2_btree_node_mem_alloc(trans, level != 0);
bch2_btree_cache_cannibalize_unlock(c);
BUG_ON(IS_ERR(b));
return ret;
}
+int bch2_btree_root_read(struct bch_fs *c, enum btree_id id,
+ const struct bkey_i *k, unsigned level)
+{
+ return bch2_trans_run(c, __bch2_btree_root_read(&trans, id, k, level));
+
+}
+
void bch2_btree_complete_write(struct bch_fs *c, struct btree *b,
struct btree_write *w)
{
{
struct btree_write *w = btree_prev_write(b);
unsigned long old, new, v;
+ unsigned type = 0;
bch2_btree_complete_write(c, b, w);
new |= (1U << BTREE_NODE_write_in_flight_inner);
new |= (1U << BTREE_NODE_just_written);
new ^= (1U << BTREE_NODE_write_idx);
+
+ type = new & BTREE_WRITE_TYPE_MASK;
+ new &= ~BTREE_WRITE_TYPE_MASK;
} else {
new &= ~(1U << BTREE_NODE_write_in_flight);
new &= ~(1U << BTREE_NODE_write_in_flight_inner);
} while ((v = cmpxchg(&b->flags, old, new)) != old);
if (new & (1U << BTREE_NODE_write_in_flight))
- __bch2_btree_node_write(c, b, BTREE_WRITE_ALREADY_STARTED);
+ __bch2_btree_node_write(c, b, BTREE_WRITE_ALREADY_STARTED|type);
else
wake_up_bit(&b->flags, BTREE_NODE_write_in_flight);
}
static void btree_node_write_done(struct bch_fs *c, struct btree *b)
{
- six_lock_read(&b->c.lock, NULL, NULL);
+ struct btree_trans trans;
+
+ bch2_trans_init(&trans, c, 0, 0);
+
+ btree_node_lock_nopath_nofail(&trans, &b->c, SIX_LOCK_read);
__btree_node_write_done(c, b);
six_unlock_read(&b->c.lock);
+
+ bch2_trans_exit(&trans);
}
static void btree_node_write_work(struct work_struct *work)
static int validate_bset_for_write(struct bch_fs *c, struct btree *b,
struct bset *i, unsigned sectors)
{
- unsigned whiteout_u64s = 0;
struct printbuf buf = PRINTBUF;
+ bool saw_error;
int ret;
ret = bch2_bkey_invalid(c, bkey_i_to_s_c(&b->key),
if (ret)
return ret;
- ret = validate_bset_keys(c, b, i, &whiteout_u64s, WRITE, false) ?:
- validate_bset(c, NULL, b, i, b->written, sectors, WRITE, false);
+ ret = validate_bset_keys(c, b, i, WRITE, false, &saw_error) ?:
+ validate_bset(c, NULL, b, i, b->written, sectors, WRITE, false, &saw_error);
if (ret) {
bch2_inconsistent_error(c);
dump_stack();
{
struct btree_write_bio *wbio = container_of(work, struct btree_write_bio, work);
struct bch_extent_ptr *ptr;
- __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp;
+ BKEY_PADDED_ONSTACK(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp;
bkey_copy(&tmp.k, &wbio->key);
bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(&tmp.k)), ptr)
ptr->offset += wbio->sector_offset;
- bch2_submit_wbio_replicas(&wbio->wbio, wbio->wbio.c, BCH_DATA_btree, &tmp.k);
+ bch2_submit_wbio_replicas(&wbio->wbio, wbio->wbio.c, BCH_DATA_btree,
+ &tmp.k, false);
}
void __bch2_btree_node_write(struct bch_fs *c, struct btree *b, unsigned flags)
bool used_mempool;
unsigned long old, new;
bool validate_before_checksum = false;
+ enum btree_write_type type = flags & BTREE_WRITE_TYPE_MASK;
void *data;
int ret;
if (old & (1 << BTREE_NODE_write_in_flight))
return;
+ if (flags & BTREE_WRITE_ONLY_IF_NEED)
+ type = new & BTREE_WRITE_TYPE_MASK;
+ new &= ~BTREE_WRITE_TYPE_MASK;
+
new &= ~(1 << BTREE_NODE_dirty);
new &= ~(1 << BTREE_NODE_need_write);
new |= (1 << BTREE_NODE_write_in_flight);
if (new & (1U << BTREE_NODE_need_write))
return;
do_write:
+ BUG_ON((type == BTREE_WRITE_initial) != (b->written == 0));
+
atomic_dec(&c->btree_cache.dirty);
BUG_ON(btree_node_fake(b));
u64s = bch2_sort_keys(i->start, &sort_iter, false);
le16_add_cpu(&i->u64s, u64s);
+ BUG_ON(!b->written && i->u64s != b->data->keys.u64s);
+
set_needs_whiteout(i, false);
/* do we have data to write? */
bytes_to_write = vstruct_end(i) - data;
sectors_to_write = round_up(bytes_to_write, block_bytes(c)) >> 9;
+ if (!b->written &&
+ b->key.k.type == KEY_TYPE_btree_ptr_v2)
+ BUG_ON(btree_ptr_sectors_written(&b->key) != sectors_to_write);
+
memset(data + bytes_to_write, 0,
(sectors_to_write << 9) - bytes_to_write);
c->opts.nochanges)
goto err;
- trace_btree_write(b, bytes_to_write, sectors_to_write);
+ trace_and_count(c, btree_node_write, b, bytes_to_write, sectors_to_write);
- wbio = container_of(bio_alloc_bioset(GFP_NOIO,
+ wbio = container_of(bio_alloc_bioset(NULL,
buf_pages(data, sectors_to_write << 9),
+ REQ_OP_WRITE|REQ_META,
+ GFP_NOIO,
&c->btree_bio),
struct btree_write_bio, wbio.bio);
wbio_init(&wbio->wbio.bio);
wbio->wbio.c = c;
wbio->wbio.used_mempool = used_mempool;
wbio->wbio.first_btree_write = !b->written;
- wbio->wbio.bio.bi_opf = REQ_OP_WRITE|REQ_META;
wbio->wbio.bio.bi_end_io = btree_node_write_endio;
wbio->wbio.bio.bi_private = b;
b->written += sectors_to_write;
- if (wbio->wbio.first_btree_write &&
- b->key.k.type == KEY_TYPE_btree_ptr_v2)
- bkey_i_to_btree_ptr_v2(&b->key)->v.sectors_written =
- cpu_to_le16(b->written);
-
if (wbio->key.k.type == KEY_TYPE_btree_ptr_v2)
bkey_i_to_btree_ptr_v2(&wbio->key)->v.sectors_written =
cpu_to_le16(b->written);
- atomic64_inc(&c->btree_writes_nr);
- atomic64_add(sectors_to_write, &c->btree_writes_sectors);
+ atomic64_inc(&c->btree_write_stats[type].nr);
+ atomic64_add(bytes_to_write, &c->btree_write_stats[type].bytes);
INIT_WORK(&wbio->work, btree_write_submit);
queue_work(c->io_complete_wq, &wbio->work);
return;
err:
set_btree_node_noevict(b);
- if (!b->written &&
- b->key.k.type == KEY_TYPE_btree_ptr_v2)
- bkey_i_to_btree_ptr_v2(&b->key)->v.sectors_written =
- cpu_to_le16(sectors_to_write);
b->written += sectors_to_write;
nowrite:
btree_bounce_free(c, bytes, used_mempool, data);
{
return __bch2_btree_flush_all(c, BTREE_NODE_write_in_flight);
}
+
+const char * const bch2_btree_write_types[] = {
+#define x(t, n) [n] = #t,
+ BCH_BTREE_WRITE_TYPES()
+ NULL
+};
+
+void bch2_btree_write_stats_to_text(struct printbuf *out, struct bch_fs *c)
+{
+ printbuf_tabstop_push(out, 20);
+ printbuf_tabstop_push(out, 10);
+
+ prt_tab(out);
+ prt_str(out, "nr");
+ prt_tab(out);
+ prt_str(out, "size");
+ prt_newline(out);
+
+ for (unsigned i = 0; i < BTREE_WRITE_TYPE_NR; i++) {
+ u64 nr = atomic64_read(&c->btree_write_stats[i].nr);
+ u64 bytes = atomic64_read(&c->btree_write_stats[i].bytes);
+
+ prt_printf(out, "%s:", bch2_btree_write_types[i]);
+ prt_tab(out);
+ prt_u64(out, nr);
+ prt_tab(out);
+ prt_human_readable_u64(out, nr ? div64_u64(bytes, nr) : 0);
+ prt_newline(out);
+ }
+}