__gc_pos_set(c, new_pos);
}
-/* range_checks - for validating min/max pos of each btree node: */
-
-struct range_checks {
- struct range_level {
- struct bpos min;
- struct bpos max;
- } l[BTREE_MAX_DEPTH];
- unsigned depth;
-};
-
-static void btree_node_range_checks_init(struct range_checks *r, unsigned depth)
-{
- unsigned i;
-
- for (i = 0; i < BTREE_MAX_DEPTH; i++)
- r->l[i].min = r->l[i].max = POS_MIN;
- r->depth = depth;
-}
-
-static void btree_node_range_checks(struct bch_fs *c, struct btree *b,
- struct range_checks *r)
+static int bch2_gc_check_topology(struct bch_fs *c,
+ struct bkey_s_c k,
+ struct bpos *expected_start,
+ struct bpos expected_end,
+ bool is_last)
{
- struct range_level *l = &r->l[b->level];
-
- struct bpos expected_min = bkey_cmp(l->min, l->max)
- ? btree_type_successor(b->btree_id, l->max)
- : l->max;
-
- bch2_fs_inconsistent_on(bkey_cmp(b->data->min_key, expected_min), c,
- "btree node has incorrect min key: %llu:%llu != %llu:%llu",
- b->data->min_key.inode,
- b->data->min_key.offset,
- expected_min.inode,
- expected_min.offset);
-
- l->max = b->data->max_key;
+ int ret = 0;
- if (b->level > r->depth) {
- l = &r->l[b->level - 1];
+ if (k.k->type == KEY_TYPE_btree_ptr_v2) {
+ struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
- bch2_fs_inconsistent_on(bkey_cmp(b->data->min_key, l->min), c,
- "btree node min doesn't match min of child nodes: %llu:%llu != %llu:%llu",
- b->data->min_key.inode,
- b->data->min_key.offset,
- l->min.inode,
- l->min.offset);
+ if (fsck_err_on(bkey_cmp(*expected_start, bp.v->min_key), c,
+ "btree node with incorrect min_key: got %llu:%llu, should be %llu:%llu",
+ bp.v->min_key.inode,
+ bp.v->min_key.offset,
+ expected_start->inode,
+ expected_start->offset)) {
+ BUG();
+ }
+ }
- bch2_fs_inconsistent_on(bkey_cmp(b->data->max_key, l->max), c,
- "btree node max doesn't match max of child nodes: %llu:%llu != %llu:%llu",
- b->data->max_key.inode,
- b->data->max_key.offset,
- l->max.inode,
- l->max.offset);
-
- if (bkey_cmp(b->data->max_key, POS_MAX))
- l->min = l->max =
- btree_type_successor(b->btree_id,
- b->data->max_key);
+ *expected_start = bkey_cmp(k.k->p, POS_MAX)
+ ? bkey_successor(k.k->p)
+ : k.k->p;
+
+ if (fsck_err_on(is_last &&
+ bkey_cmp(k.k->p, expected_end), c,
+ "btree node with incorrect max_key: got %llu:%llu, should be %llu:%llu",
+ k.k->p.inode,
+ k.k->p.offset,
+ expected_end.inode,
+ expected_end.offset)) {
+ BUG();
}
+fsck_err:
+ return ret;
}
/* marking of btree keys/nodes: */
static int btree_gc_mark_node(struct bch_fs *c, struct btree *b, u8 *max_stale,
bool initial)
{
+ struct bpos next_node_start = b->data->min_key;
struct btree_node_iter iter;
struct bkey unpacked;
struct bkey_s_c k;
if (!btree_node_type_needs_gc(btree_node_type(b)))
return 0;
- for_each_btree_node_key_unpack(b, k, &iter,
- &unpacked) {
+ bch2_btree_node_iter_init_from_start(&iter, b);
+
+ while ((k = bch2_btree_node_iter_peek_unpack(&iter, b, &unpacked)).k) {
bch2_bkey_debugcheck(c, b, k);
ret = bch2_gc_mark_key(c, k, max_stale, initial);
if (ret)
break;
+
+ bch2_btree_node_iter_advance(&iter, b);
+
+ if (b->c.level) {
+ ret = bch2_gc_check_topology(c, k,
+ &next_node_start,
+ b->data->max_key,
+ bch2_btree_node_iter_end(&iter));
+ if (ret)
+ break;
+ }
}
return ret;
struct btree_trans trans;
struct btree_iter *iter;
struct btree *b;
- struct range_checks r;
unsigned depth = metadata_only ? 1
: expensive_debug_checks(c) ? 0
: !btree_node_type_needs_gc(btree_id) ? 1
gc_pos_set(c, gc_pos_btree(btree_id, POS_MIN, 0));
- btree_node_range_checks_init(&r, depth);
-
__for_each_btree_node(&trans, iter, btree_id, POS_MIN,
0, depth, BTREE_ITER_PREFETCH, b) {
- btree_node_range_checks(c, b, &r);
-
bch2_verify_btree_nr_keys(b);
gc_pos_set(c, gc_pos_btree_node(b));
if (!btree_node_fake(b))
ret = bch2_gc_mark_key(c, bkey_i_to_s_c(&b->key),
&max_stale, initial);
- gc_pos_set(c, gc_pos_btree_root(b->btree_id));
+ gc_pos_set(c, gc_pos_btree_root(b->c.btree_id));
mutex_unlock(&c->btree_root_lock);
return ret;
}
static int bch2_gc_btree_init_recurse(struct bch_fs *c, struct btree *b,
- struct journal_keys *journal_keys,
- unsigned target_depth)
+ struct journal_keys *journal_keys,
+ unsigned target_depth)
{
struct btree_and_journal_iter iter;
struct bkey_s_c k;
+ struct bpos next_node_start = b->data->min_key;
u8 max_stale = 0;
int ret = 0;
while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
bch2_bkey_debugcheck(c, b, k);
+ BUG_ON(bkey_cmp(k.k->p, b->data->min_key) < 0);
+ BUG_ON(bkey_cmp(k.k->p, b->data->max_key) > 0);
+
ret = bch2_gc_mark_key(c, k, &max_stale, true);
if (ret)
break;
- if (b->level > target_depth) {
+ if (b->c.level) {
struct btree *child;
BKEY_PADDED(k) tmp;
bkey_reassemble(&tmp.k, k);
+ k = bkey_i_to_s_c(&tmp.k);
- child = bch2_btree_node_get_noiter(c, &tmp.k,
- b->btree_id, b->level - 1);
- ret = PTR_ERR_OR_ZERO(child);
+ bch2_btree_and_journal_iter_advance(&iter);
+
+ ret = bch2_gc_check_topology(c, k,
+ &next_node_start,
+ b->data->max_key,
+ !bch2_btree_and_journal_iter_peek(&iter).k);
if (ret)
break;
- bch2_gc_btree_init_recurse(c, child,
- journal_keys, target_depth);
- six_unlock_read(&child->lock);
- }
+ if (b->c.level > target_depth) {
+ child = bch2_btree_node_get_noiter(c, &tmp.k,
+ b->c.btree_id, b->c.level - 1);
+ ret = PTR_ERR_OR_ZERO(child);
+ if (ret)
+ break;
+
+ ret = bch2_gc_btree_init_recurse(c, child,
+ journal_keys, target_depth);
+ six_unlock_read(&child->c.lock);
- bch2_btree_and_journal_iter_advance(&iter);
+ if (ret)
+ break;
+ }
+ } else {
+ bch2_btree_and_journal_iter_advance(&iter);
+ }
}
return ret;
if (btree_node_fake(b))
return 0;
- six_lock_read(&b->lock);
- if (b->level >= target_depth)
+ six_lock_read(&b->c.lock, NULL, NULL);
+ if (fsck_err_on(bkey_cmp(b->data->min_key, POS_MIN), c,
+ "btree root with incorrect min_key: %llu:%llu",
+ b->data->min_key.inode,
+ b->data->min_key.offset)) {
+ BUG();
+ }
+
+ if (fsck_err_on(bkey_cmp(b->data->max_key, POS_MAX), c,
+ "btree root with incorrect min_key: %llu:%llu",
+ b->data->max_key.inode,
+ b->data->max_key.offset)) {
+ BUG();
+ }
+
+ if (b->c.level >= target_depth)
ret = bch2_gc_btree_init_recurse(c, b,
journal_keys, target_depth);
if (!ret)
ret = bch2_gc_mark_key(c, bkey_i_to_s_c(&b->key),
&max_stale, true);
- six_unlock_read(&b->lock);
+fsck_err:
+ six_unlock_read(&b->c.lock);
return ret;
}
mutex_unlock(&c->sb_lock);
}
+#if 0
/* Also see bch2_pending_btree_node_free_insert_done() */
static void bch2_mark_pending_btree_node_frees(struct bch_fs *c)
{
mutex_unlock(&c->btree_interior_update_lock);
}
+#endif
static void bch2_mark_allocator_buckets(struct bch_fs *c)
{
#define copy_bucket_field(_f) \
if (dst->b[b].mark._f != src->b[b].mark._f) { \
if (verify) \
- fsck_err(c, "dev %u bucket %zu has wrong " #_f \
+ fsck_err(c, "bucket %u:%zu gen %u data type %s has wrong " #_f \
": got %u, should be %u", i, b, \
+ dst->b[b].mark.gen, \
+ bch2_data_types[dst->b[b].mark.data_type],\
dst->b[b].mark._f, src->b[b].mark._f); \
dst->b[b]._mark._f = src->b[b].mark._f; \
}
c->usage_gc = __alloc_percpu_gfp(fs_usage_u64s(c) * sizeof(u64),
sizeof(u64), GFP_KERNEL);
- if (!c->usage_gc)
+ if (!c->usage_gc) {
+ bch_err(c, "error allocating c->usage_gc");
return -ENOMEM;
+ }
for_each_member_device(ca, c, i) {
BUG_ON(ca->buckets[1]);
GFP_KERNEL|__GFP_ZERO);
if (!ca->buckets[1]) {
percpu_ref_put(&ca->ref);
+ bch_err(c, "error allocating ca->buckets[gc]");
return -ENOMEM;
}
ca->usage[1] = alloc_percpu(struct bch_dev_usage);
if (!ca->usage[1]) {
+ bch_err(c, "error allocating ca->usage[gc]");
percpu_ref_put(&ca->ref);
return -ENOMEM;
}
}
ret = bch2_ec_mem_alloc(c, true);
- if (ret)
+ if (ret) {
+ bch_err(c, "error allocating ec gc mem");
return ret;
+ }
percpu_down_write(&c->mark_lock);
unsigned i, iter = 0;
int ret;
+ lockdep_assert_held(&c->state_lock);
trace_gc_start(c);
down_write(&c->gc_lock);
+
+ /* flush interior btree updates: */
+ closure_wait_event(&c->btree_interior_update_wait,
+ !bch2_btree_interior_updates_nr_pending(c));
again:
ret = bch2_gc_start(c, metadata_only);
if (ret)
if (ret)
goto out;
+#if 0
bch2_mark_pending_btree_node_frees(c);
+#endif
bch2_mark_allocator_buckets(c);
c->gc_count++;
return ret;
}
+/*
+ * For recalculating oldest gen, we only need to walk keys in leaf nodes; btree
+ * node pointers currently never have cached pointers that can become stale:
+ */
+static int bch2_gc_btree_gens(struct bch_fs *c, enum btree_id id)
+{
+ struct btree_trans trans;
+ struct btree_iter *iter;
+ struct bkey_s_c k;
+ int ret;
+
+ bch2_trans_init(&trans, c, 0, 0);
+
+ for_each_btree_key(&trans, iter, id, POS_MIN, BTREE_ITER_PREFETCH, k, ret) {
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ const struct bch_extent_ptr *ptr;
+
+ bkey_for_each_ptr(ptrs, ptr) {
+ struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
+ struct bucket *g = PTR_BUCKET(ca, ptr, false);
+
+ if (gen_after(g->gc_gen, ptr->gen))
+ g->gc_gen = ptr->gen;
+
+ if (gen_after(g->mark.gen, ptr->gen) > 32) {
+ /* rewrite btree node */
+
+ }
+ }
+ }
+
+ bch2_trans_exit(&trans);
+ return ret;
+}
+
+int bch2_gc_gens(struct bch_fs *c)
+{
+ struct bch_dev *ca;
+ unsigned i;
+ int ret;
+
+ down_read(&c->state_lock);
+
+ for_each_member_device(ca, c, i) {
+ struct bucket_array *buckets = bucket_array(ca);
+ struct bucket *g;
+
+ for_each_bucket(g, buckets)
+ g->gc_gen = g->mark.gen;
+ }
+
+ for (i = 0; i < BTREE_ID_NR; i++)
+ if (btree_node_type_needs_gc(i)) {
+ ret = bch2_gc_btree_gens(c, i);
+ if (ret)
+ goto err;
+ }
+
+ for_each_member_device(ca, c, i) {
+ struct bucket_array *buckets = bucket_array(ca);
+ struct bucket *g;
+
+ for_each_bucket(g, buckets)
+ g->oldest_gen = g->gc_gen;
+ }
+err:
+ up_read(&c->state_lock);
+ return ret;
+}
+
/* Btree coalescing */
static void recalc_packed_keys(struct btree *b)
return;
}
- as = bch2_btree_update_start(c, iter->btree_id,
+ as = bch2_btree_update_start(iter->trans, iter->btree_id,
btree_update_reserve_required(c, parent) + nr_old_nodes,
BTREE_INSERT_NOFAIL|
BTREE_INSERT_USE_RESERVE,
set_btree_bset_end(n1, n1->set);
- six_unlock_write(&n2->lock);
+ six_unlock_write(&n2->c.lock);
bch2_btree_node_free_never_inserted(c, n2);
- six_unlock_intent(&n2->lock);
+ six_unlock_intent(&n2->c.lock);
memmove(new_nodes + i - 1,
new_nodes + i,
n1->key.k.p = n1->data->max_key =
bkey_unpack_pos(n1, last);
- n2->data->min_key =
- btree_type_successor(iter->btree_id,
- n1->data->max_key);
+ n2->data->min_key = bkey_successor(n1->data->max_key);
memcpy_u64s(vstruct_last(s1),
s2->start, u64s);
btree_node_reset_sib_u64s(n);
bch2_btree_build_aux_trees(n);
- six_unlock_write(&n->lock);
+
+ bch2_btree_update_add_new_node(as, n);
+ six_unlock_write(&n->c.lock);
bch2_btree_node_write(c, n, SIX_LOCK_intent);
}
BUG_ON(!bch2_keylist_empty(&keylist));
- BUG_ON(iter->l[old_nodes[0]->level].b != old_nodes[0]);
+ BUG_ON(iter->l[old_nodes[0]->c.level].b != old_nodes[0]);
bch2_btree_iter_node_replace(iter, new_nodes[0]);
for (i = 0; i < nr_new_nodes; i++)
- bch2_open_buckets_put(c, &new_nodes[i]->ob);
+ bch2_btree_update_get_open_buckets(as, new_nodes[i]);
/* Free the old nodes and update our sliding window */
for (i = 0; i < nr_old_nodes; i++) {
}
for (i = 0; i < nr_new_nodes; i++)
- six_unlock_intent(&new_nodes[i]->lock);
+ six_unlock_intent(&new_nodes[i]->c.lock);
bch2_btree_update_done(as);
bch2_keylist_free(&keylist, NULL);
for (i = 1; i < GC_MERGE_NODES; i++) {
if (!merge[i] ||
- !six_relock_intent(&merge[i]->lock, lock_seq[i]))
+ !six_relock_intent(&merge[i]->c.lock, lock_seq[i]))
break;
- if (merge[i]->level != merge[0]->level) {
- six_unlock_intent(&merge[i]->lock);
+ if (merge[i]->c.level != merge[0]->c.level) {
+ six_unlock_intent(&merge[i]->c.lock);
break;
}
}
bch2_coalesce_nodes(c, iter, merge);
for (i = 1; i < GC_MERGE_NODES && merge[i]; i++) {
- lock_seq[i] = merge[i]->lock.state.seq;
- six_unlock_intent(&merge[i]->lock);
+ lock_seq[i] = merge[i]->c.lock.state.seq;
+ six_unlock_intent(&merge[i]->c.lock);
}
- lock_seq[0] = merge[0]->lock.state.seq;
+ lock_seq[0] = merge[0]->c.lock.state.seq;
if (kthread && kthread_should_stop()) {
bch2_trans_exit(&trans);
last = atomic_long_read(&clock->now);
last_kick = atomic_read(&c->kick_gc);
+ /*
+ * Full gc is currently incompatible with btree key cache:
+ */
+#if 0
ret = bch2_gc(c, NULL, false, false);
+#else
+ ret = bch2_gc_gens(c);
+#endif
if (ret)
bch_err(c, "btree gc failed: %i", ret);
projects / bcachefs-tools-debian / blobdiff