break;
bp = bkey_s_c_to_btree_ptr_v2(k);
- if (bpos_cmp(next_node, bp.v->min_key)) {
+ if (!bpos_eq(next_node, bp.v->min_key)) {
bch2_dump_btree_node(c, b);
bch2_bpos_to_text(&buf1, next_node);
bch2_bpos_to_text(&buf2, bp.v->min_key);
bch2_btree_node_iter_advance(&iter, b);
if (bch2_btree_node_iter_end(&iter)) {
- if (bpos_cmp(k.k->p, b->key.k.p)) {
+ if (!bpos_eq(k.k->p, b->key.k.p)) {
bch2_dump_btree_node(c, b);
bch2_bpos_to_text(&buf1, b->key.k.p);
bch2_bpos_to_text(&buf2, k.k->p);
{
trace_and_count(c, btree_node_free, c, b);
+ BUG_ON(btree_node_write_blocked(b));
BUG_ON(btree_node_dirty(b));
BUG_ON(btree_node_need_write(b));
BUG_ON(b == btree_node_root(c, b));
struct bch_devs_list devs_have = (struct bch_devs_list) { 0 };
unsigned nr_reserve;
enum alloc_reserve alloc_reserve;
+ int ret;
if (flags & BTREE_INSERT_USE_RESERVE) {
nr_reserve = 0;
mutex_unlock(&c->btree_reserve_cache_lock);
retry:
- wp = bch2_alloc_sectors_start_trans(trans,
+ ret = bch2_alloc_sectors_start_trans(trans,
c->opts.metadata_target ?:
c->opts.foreground_target,
0,
&devs_have,
res->nr_replicas,
c->opts.metadata_replicas_required,
- alloc_reserve, 0, cl);
- if (IS_ERR(wp))
- return ERR_CAST(wp);
+ alloc_reserve, 0, cl, &wp);
+ if (unlikely(ret))
+ return ERR_PTR(ret);
if (wp->sectors_free < btree_sectors(c)) {
struct open_bucket *ob;
b->data->max_key = pos;
}
-struct btree *__bch2_btree_node_alloc_replacement(struct btree_update *as,
- struct btree_trans *trans,
- struct btree *b,
- struct bkey_format format)
+static struct btree *bch2_btree_node_alloc_replacement(struct btree_update *as,
+ struct btree_trans *trans,
+ struct btree *b)
{
- struct btree *n;
+ struct btree *n = bch2_btree_node_alloc(as, trans, b->c.level);
+ struct bkey_format format = bch2_btree_calc_format(b);
- n = bch2_btree_node_alloc(as, trans, b->c.level);
+ /*
+ * The keys might expand with the new format - if they wouldn't fit in
+ * the btree node anymore, use the old format for now:
+ */
+ if (!bch2_btree_node_format_fits(as->c, b, &format))
+ format = b->format;
SET_BTREE_NODE_SEQ(n->data, BTREE_NODE_SEQ(b->data) + 1);
bch2_btree_sort_into(as->c, n, b);
btree_node_reset_sib_u64s(n);
-
- n->key.k.p = b->key.k.p;
return n;
}
-static struct btree *bch2_btree_node_alloc_replacement(struct btree_update *as,
- struct btree_trans *trans,
- struct btree *b)
-{
- struct bkey_format new_f = bch2_btree_calc_format(b);
-
- /*
- * The keys might expand with the new format - if they wouldn't fit in
- * the btree node anymore, use the old format for now:
- */
- if (!bch2_btree_node_format_fits(as->c, b, &new_f))
- new_f = b->format;
-
- return __bch2_btree_node_alloc_replacement(as, trans, b, new_f);
-}
-
static struct btree *__btree_root_alloc(struct btree_update *as,
struct btree_trans *trans, unsigned level)
{
bch2_trans_unlock(&trans);
bch2_fs_fatal_err_on(ret && !bch2_journal_error(&c->journal), c,
- "error %i in btree_update_nodes_written()", ret);
+ "%s(): error %s", __func__, bch2_err_str(ret));
err:
if (as->b) {
struct btree_path *path;
BUG_ON(as->mode != BTREE_INTERIOR_NO_UPDATE);
BUG_ON(!btree_node_dirty(b));
+ BUG_ON(!b->c.level);
as->mode = BTREE_INTERIOR_UPDATING_NODE;
as->b = b;
clear_btree_node_dirty_acct(c, b);
clear_btree_node_need_write(b);
+ clear_btree_node_write_blocked(b);
/*
* Does this node have unwritten data that has a pin on the journal?
bch2_trans_unlock(trans);
closure_sync(&cl);
- } while (ret == -EAGAIN);
+ } while (bch2_err_matches(ret, BCH_ERR_operation_blocked));
}
if (ret) {
- trace_and_count(c, btree_reserve_get_fail, trans->fn, _RET_IP_, nr_nodes[0] + nr_nodes[1]);
+ trace_and_count(c, btree_reserve_get_fail, trans->fn,
+ _RET_IP_, nr_nodes[0] + nr_nodes[1], ret);
goto err;
}
struct bch_fs *c = as->c;
struct bkey_packed *k;
struct printbuf buf = PRINTBUF;
+ unsigned long old, new, v;
BUG_ON(insert->k.type == KEY_TYPE_btree_ptr_v2 &&
!btree_ptr_sectors_written(insert));
bch2_btree_bset_insert_key(trans, path, b, node_iter, insert);
set_btree_node_dirty_acct(c, b);
- set_btree_node_need_write(b);
+
+ v = READ_ONCE(b->flags);
+ do {
+ old = new = v;
+
+ new &= ~BTREE_WRITE_TYPE_MASK;
+ new |= BTREE_WRITE_interior;
+ new |= 1 << BTREE_NODE_need_write;
+ } while ((v = cmpxchg(&b->flags, old, new)) != old);
printbuf_exit(&buf);
}
;
while (!bch2_keylist_empty(keys)) {
- bch2_insert_fixup_btree_ptr(as, trans, path, b,
- &node_iter, bch2_keylist_front(keys));
+ struct bkey_i *k = bch2_keylist_front(keys);
+
+ if (bpos_gt(k->k.p, b->key.k.p))
+ break;
+
+ bch2_insert_fixup_btree_ptr(as, trans, path, b, &node_iter, k);
bch2_keylist_pop_front(keys);
}
}
* Move keys from n1 (original replacement node, now lower node) to n2 (higher
* node)
*/
-static struct btree *__btree_split_node(struct btree_update *as,
- struct btree_trans *trans,
- struct btree *n1)
+static void __btree_split_node(struct btree_update *as,
+ struct btree_trans *trans,
+ struct btree *b,
+ struct btree *n[2])
{
- struct bkey_format_state s;
- size_t nr_packed = 0, nr_unpacked = 0;
- struct btree *n2;
- struct bset *set1, *set2;
- struct bkey_packed *k, *set2_start, *set2_end, *out, *prev = NULL;
- struct bpos n1_pos;
+ struct bkey_packed *k;
+ struct bpos n1_pos = POS_MIN;
+ struct btree_node_iter iter;
+ struct bset *bsets[2];
+ struct bkey_format_state format[2];
+ struct bkey_packed *out[2];
+ struct bkey uk;
+ unsigned u64s, n1_u64s = (b->nr.live_u64s * 3) / 5;
+ int i;
- n2 = bch2_btree_node_alloc(as, trans, n1->c.level);
+ for (i = 0; i < 2; i++) {
+ BUG_ON(n[i]->nsets != 1);
- n2->data->max_key = n1->data->max_key;
- n2->data->format = n1->format;
- SET_BTREE_NODE_SEQ(n2->data, BTREE_NODE_SEQ(n1->data));
- n2->key.k.p = n1->key.k.p;
+ bsets[i] = btree_bset_first(n[i]);
+ out[i] = bsets[i]->start;
- set1 = btree_bset_first(n1);
- set2 = btree_bset_first(n2);
+ SET_BTREE_NODE_SEQ(n[i]->data, BTREE_NODE_SEQ(b->data) + 1);
+ bch2_bkey_format_init(&format[i]);
+ }
- /*
- * Has to be a linear search because we don't have an auxiliary
- * search tree yet
- */
- k = set1->start;
- while (1) {
- struct bkey_packed *n = bkey_next(k);
+ u64s = 0;
+ for_each_btree_node_key(b, k, &iter) {
+ if (bkey_deleted(k))
+ continue;
+
+ i = u64s >= n1_u64s;
+ u64s += k->u64s;
+ uk = bkey_unpack_key(b, k);
+ if (!i)
+ n1_pos = uk.p;
+ bch2_bkey_format_add_key(&format[i], &uk);
+ }
- if (n == vstruct_last(set1))
- break;
- if (k->_data - set1->_data >= (le16_to_cpu(set1->u64s) * 3) / 5)
- break;
+ btree_set_min(n[0], b->data->min_key);
+ btree_set_max(n[0], n1_pos);
+ btree_set_min(n[1], bpos_successor(n1_pos));
+ btree_set_max(n[1], b->data->max_key);
- if (bkey_packed(k))
- nr_packed++;
- else
- nr_unpacked++;
+ for (i = 0; i < 2; i++) {
+ bch2_bkey_format_add_pos(&format[i], n[i]->data->min_key);
+ bch2_bkey_format_add_pos(&format[i], n[i]->data->max_key);
- prev = k;
- k = n;
+ n[i]->data->format = bch2_bkey_format_done(&format[i]);
+ btree_node_set_format(n[i], n[i]->data->format);
}
- BUG_ON(!prev);
- set2_start = k;
- set2_end = vstruct_last(set1);
+ u64s = 0;
+ for_each_btree_node_key(b, k, &iter) {
+ if (bkey_deleted(k))
+ continue;
- set1->u64s = cpu_to_le16((u64 *) set2_start - set1->_data);
- set_btree_bset_end(n1, n1->set);
+ i = u64s >= n1_u64s;
+ u64s += k->u64s;
- n1->nr.live_u64s = le16_to_cpu(set1->u64s);
- n1->nr.bset_u64s[0] = le16_to_cpu(set1->u64s);
- n1->nr.packed_keys = nr_packed;
- n1->nr.unpacked_keys = nr_unpacked;
-
- n1_pos = bkey_unpack_pos(n1, prev);
- if (as->c->sb.version < bcachefs_metadata_version_snapshot)
- n1_pos.snapshot = U32_MAX;
+ if (bch2_bkey_transform(&n[i]->format, out[i], bkey_packed(k)
+ ? &b->format: &bch2_bkey_format_current, k))
+ out[i]->format = KEY_FORMAT_LOCAL_BTREE;
+ else
+ bch2_bkey_unpack(b, (void *) out[i], k);
- btree_set_max(n1, n1_pos);
- btree_set_min(n2, bpos_successor(n1->key.k.p));
+ out[i]->needs_whiteout = false;
- bch2_bkey_format_init(&s);
- bch2_bkey_format_add_pos(&s, n2->data->min_key);
- bch2_bkey_format_add_pos(&s, n2->data->max_key);
-
- for (k = set2_start; k != set2_end; k = bkey_next(k)) {
- struct bkey uk = bkey_unpack_key(n1, k);
- bch2_bkey_format_add_key(&s, &uk);
+ btree_keys_account_key_add(&n[i]->nr, 0, out[i]);
+ out[i] = bkey_next(out[i]);
}
- n2->data->format = bch2_bkey_format_done(&s);
- btree_node_set_format(n2, n2->data->format);
-
- out = set2->start;
- memset(&n2->nr, 0, sizeof(n2->nr));
-
- for (k = set2_start; k != set2_end; k = bkey_next(k)) {
- BUG_ON(!bch2_bkey_transform(&n2->format, out, bkey_packed(k)
- ? &n1->format : &bch2_bkey_format_current, k));
- out->format = KEY_FORMAT_LOCAL_BTREE;
- btree_keys_account_key_add(&n2->nr, 0, out);
- out = bkey_next(out);
- }
+ for (i = 0; i < 2; i++) {
+ bsets[i]->u64s = cpu_to_le16((u64 *) out[i] - bsets[i]->_data);
- set2->u64s = cpu_to_le16((u64 *) out - set2->_data);
- set_btree_bset_end(n2, n2->set);
+ BUG_ON(!bsets[i]->u64s);
- BUG_ON(!set1->u64s);
- BUG_ON(!set2->u64s);
+ set_btree_bset_end(n[i], n[i]->set);
- btree_node_reset_sib_u64s(n1);
- btree_node_reset_sib_u64s(n2);
+ btree_node_reset_sib_u64s(n[i]);
- bch2_verify_btree_nr_keys(n1);
- bch2_verify_btree_nr_keys(n2);
+ bch2_verify_btree_nr_keys(n[i]);
- if (n1->c.level) {
- btree_node_interior_verify(as->c, n1);
- btree_node_interior_verify(as->c, n2);
+ if (b->c.level)
+ btree_node_interior_verify(as->c, n[i]);
}
-
- return n2;
}
/*
struct btree *b,
struct keylist *keys)
{
- struct btree_node_iter node_iter;
- struct bkey_i *k = bch2_keylist_front(keys);
- struct bkey_packed *src, *dst, *n;
- struct bset *i;
+ if (!bch2_keylist_empty(keys) &&
+ bpos_le(bch2_keylist_front(keys)->k.p, b->data->max_key)) {
+ struct btree_node_iter node_iter;
- bch2_btree_node_iter_init(&node_iter, b, &k->k.p);
+ bch2_btree_node_iter_init(&node_iter, b, &bch2_keylist_front(keys)->k.p);
- __bch2_btree_insert_keys_interior(as, trans, path, b, node_iter, keys);
+ __bch2_btree_insert_keys_interior(as, trans, path, b, node_iter, keys);
- /*
- * We can't tolerate whiteouts here - with whiteouts there can be
- * duplicate keys, and it would be rather bad if we picked a duplicate
- * for the pivot:
- */
- i = btree_bset_first(b);
- src = dst = i->start;
- while (src != vstruct_last(i)) {
- n = bkey_next(src);
- if (!bkey_deleted(src)) {
- memmove_u64s_down(dst, src, src->u64s);
- dst = bkey_next(dst);
- }
- src = n;
+ btree_node_interior_verify(as->c, b);
}
-
- /* Also clear out the unwritten whiteouts area: */
- b->whiteout_u64s = 0;
-
- i->u64s = cpu_to_le16((u64 *) dst - i->_data);
- set_btree_bset_end(b, b->set);
-
- BUG_ON(b->nsets != 1 ||
- b->nr.live_u64s != le16_to_cpu(btree_bset_first(b)->u64s));
-
- btree_node_interior_verify(as->c, b);
}
static int btree_split(struct btree_update *as, struct btree_trans *trans,
bch2_btree_interior_update_will_free_node(as, b);
- n1 = bch2_btree_node_alloc_replacement(as, trans, b);
-
- if (keys)
- btree_split_insert_keys(as, trans, path, n1, keys);
+ if (b->nr.live_u64s > BTREE_SPLIT_THRESHOLD(c)) {
+ struct btree *n[2];
- if (bset_u64s(&n1->set[0]) > BTREE_SPLIT_THRESHOLD(c)) {
trace_and_count(c, btree_node_split, c, b);
- n2 = __btree_split_node(as, trans, n1);
+ n[0] = n1 = bch2_btree_node_alloc(as, trans, b->c.level);
+ n[1] = n2 = bch2_btree_node_alloc(as, trans, b->c.level);
+
+ __btree_split_node(as, trans, b, n);
+
+ if (keys) {
+ btree_split_insert_keys(as, trans, path, n1, keys);
+ btree_split_insert_keys(as, trans, path, n2, keys);
+ BUG_ON(!bch2_keylist_empty(keys));
+ }
bch2_btree_build_aux_trees(n2);
bch2_btree_build_aux_trees(n1);
} else {
trace_and_count(c, btree_node_compact, c, b);
+ n1 = bch2_btree_node_alloc_replacement(as, trans, b);
+
+ if (keys) {
+ btree_split_insert_keys(as, trans, path, n1, keys);
+ BUG_ON(!bch2_keylist_empty(keys));
+ }
+
bch2_btree_build_aux_trees(n1);
bch2_btree_update_add_new_node(as, n1);
six_unlock_write(&n1->c.lock);
* We could attempt to avoid the transaction restart, by calling
* bch2_btree_path_upgrade() and allocating more nodes:
*/
- if (b->c.level >= as->update_level)
+ if (b->c.level >= as->update_level) {
+ trace_and_count(c, trans_restart_split_race, trans, _THIS_IP_);
return btree_trans_restart(trans, BCH_ERR_transaction_restart_split_race);
+ }
return btree_split(as, trans, path, b, keys, flags);
}
b = path->l[level].b;
- if ((sib == btree_prev_sib && !bpos_cmp(b->data->min_key, POS_MIN)) ||
- (sib == btree_next_sib && !bpos_cmp(b->data->max_key, SPOS_MAX))) {
+ if ((sib == btree_prev_sib && bpos_eq(b->data->min_key, POS_MIN)) ||
+ (sib == btree_next_sib && bpos_eq(b->data->max_key, SPOS_MAX))) {
b->sib_u64s[sib] = U16_MAX;
return 0;
}
next = m;
}
- if (bkey_cmp(bpos_successor(prev->data->max_key), next->data->min_key)) {
+ if (!bpos_eq(bpos_successor(prev->data->max_key), next->data->min_key)) {
struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
bch2_bpos_to_text(&buf1, prev->data->max_key);
bch2_bpos_to_text(&buf2, next->data->min_key);
bch_err(c,
- "btree topology error in btree merge:\n"
+ "%s(): btree topology error:\n"
" prev ends at %s\n"
" next starts at %s",
- buf1.buf, buf2.buf);
+ __func__, buf1.buf, buf2.buf);
printbuf_exit(&buf1);
printbuf_exit(&buf2);
bch2_topology_error(c);
struct async_btree_rewrite {
struct bch_fs *c;
struct work_struct work;
+ struct list_head list;
enum btree_id btree_id;
unsigned level;
struct bpos pos;
static int async_btree_node_rewrite_trans(struct btree_trans *trans,
struct async_btree_rewrite *a)
{
+ struct bch_fs *c = trans->c;
struct btree_iter iter;
struct btree *b;
int ret;
if (ret)
goto out;
- if (!b || b->data->keys.seq != a->seq)
+ if (!b || b->data->keys.seq != a->seq) {
+ struct printbuf buf = PRINTBUF;
+
+ if (b)
+ bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+ else
+ prt_str(&buf, "(null");
+ bch_info(c, "%s: node to rewrite not found:, searching for seq %llu, got\n%s",
+ __func__, a->seq, buf.buf);
+ printbuf_exit(&buf);
goto out;
+ }
ret = bch2_btree_node_rewrite(trans, &iter, b, 0);
-out :
+out:
bch2_trans_iter_exit(trans, &iter);
return ret;
struct async_btree_rewrite *a =
container_of(work, struct async_btree_rewrite, work);
struct bch_fs *c = a->c;
+ int ret;
- bch2_trans_do(c, NULL, NULL, 0,
+ ret = bch2_trans_do(c, NULL, NULL, 0,
async_btree_node_rewrite_trans(&trans, a));
- percpu_ref_put(&c->writes);
+ if (ret)
+ bch_err(c, "%s: error %s", __func__, bch2_err_str(ret));
+ bch2_write_ref_put(c, BCH_WRITE_REF_node_rewrite);
kfree(a);
}
void bch2_btree_node_rewrite_async(struct bch_fs *c, struct btree *b)
{
struct async_btree_rewrite *a;
-
- if (!percpu_ref_tryget_live(&c->writes))
- return;
+ int ret;
a = kmalloc(sizeof(*a), GFP_NOFS);
if (!a) {
- percpu_ref_put(&c->writes);
+ bch_err(c, "%s: error allocating memory", __func__);
return;
}
a->level = b->c.level;
a->pos = b->key.k.p;
a->seq = b->data->keys.seq;
-
INIT_WORK(&a->work, async_btree_node_rewrite_work);
+
+ if (unlikely(!test_bit(BCH_FS_MAY_GO_RW, &c->flags))) {
+ mutex_lock(&c->pending_node_rewrites_lock);
+ list_add(&a->list, &c->pending_node_rewrites);
+ mutex_unlock(&c->pending_node_rewrites_lock);
+ return;
+ }
+
+ if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_node_rewrite)) {
+ if (test_bit(BCH_FS_STARTED, &c->flags)) {
+ bch_err(c, "%s: error getting c->writes ref", __func__);
+ kfree(a);
+ return;
+ }
+
+ ret = bch2_fs_read_write_early(c);
+ if (ret) {
+ bch_err(c, "%s: error going read-write: %s",
+ __func__, bch2_err_str(ret));
+ kfree(a);
+ return;
+ }
+
+ bch2_write_ref_get(c, BCH_WRITE_REF_node_rewrite);
+ }
+
queue_work(c->btree_interior_update_worker, &a->work);
}
+void bch2_do_pending_node_rewrites(struct bch_fs *c)
+{
+ struct async_btree_rewrite *a, *n;
+
+ mutex_lock(&c->pending_node_rewrites_lock);
+ list_for_each_entry_safe(a, n, &c->pending_node_rewrites, list) {
+ list_del(&a->list);
+
+ bch2_write_ref_get(c, BCH_WRITE_REF_node_rewrite);
+ queue_work(c->btree_interior_update_worker, &a->work);
+ }
+ mutex_unlock(&c->pending_node_rewrites_lock);
+}
+
+void bch2_free_pending_node_rewrites(struct bch_fs *c)
+{
+ struct async_btree_rewrite *a, *n;
+
+ mutex_lock(&c->pending_node_rewrites_lock);
+ list_for_each_entry_safe(a, n, &c->pending_node_rewrites, list) {
+ list_del(&a->list);
+
+ kfree(a);
+ }
+ mutex_unlock(&c->pending_node_rewrites_lock);
+}
+
static int __bch2_btree_node_update_key(struct btree_trans *trans,
struct btree_iter *iter,
struct btree *b, struct btree *new_hash,
_THIS_IP_);
BUG_ON(iter2.path->level != b->c.level);
- BUG_ON(bpos_cmp(iter2.path->pos, new_key->k.p));
+ BUG_ON(!bpos_eq(iter2.path->pos, new_key->k.p));
btree_path_set_level_up(trans, iter2.path);
- bch2_btree_path_check_sort(trans, iter2.path, 0);
+ trans->paths_sorted = false;
ret = bch2_btree_iter_traverse(&iter2) ?:
bch2_trans_update(trans, &iter2, new_key, BTREE_TRIGGER_NORUN);
return ret;
}
-void bch2_journal_entries_to_btree_roots(struct bch_fs *c, struct jset *jset)
+void bch2_journal_entry_to_btree_root(struct bch_fs *c, struct jset_entry *entry)
{
- struct btree_root *r;
- struct jset_entry *entry;
+ struct btree_root *r = &c->btree_roots[entry->btree_id];
mutex_lock(&c->btree_root_lock);
- vstruct_for_each(jset, entry)
- if (entry->type == BCH_JSET_ENTRY_btree_root) {
- r = &c->btree_roots[entry->btree_id];
- r->level = entry->level;
- r->alive = true;
- bkey_copy(&r->key, &entry->start[0]);
- }
+ r->level = entry->level;
+ r->alive = true;
+ bkey_copy(&r->key, &entry->start[0]);
mutex_unlock(&c->btree_root_lock);
}
mutex_init(&c->btree_interior_update_lock);
INIT_WORK(&c->btree_interior_update_work, btree_interior_update_work);
+ INIT_LIST_HEAD(&c->pending_node_rewrites);
+ mutex_init(&c->pending_node_rewrites_lock);
+
c->btree_interior_update_worker =
alloc_workqueue("btree_update", WQ_UNBOUND|WQ_MEM_RECLAIM, 1);
if (!c->btree_interior_update_worker)
projects / bcachefs-tools-debian / blobdiff