#include <linux/sched/mm.h>
#include <trace/events/bcachefs.h>
+struct lock_class_key bch2_btree_node_lock_key;
+
void bch2_recalc_btree_reserve(struct bch_fs *c)
{
unsigned i, reserve = 16;
return max_t(int, 0, bc->used - bc->reserve);
}
-static void __btree_node_data_free(struct bch_fs *c, struct btree *b)
+static void btree_node_data_free(struct bch_fs *c, struct btree *b)
{
+ struct btree_cache *bc = &c->btree_cache;
+
EBUG_ON(btree_node_write_in_flight(b));
kvpfree(b->data, btree_bytes(c));
b->data = NULL;
+#ifdef __KERNEL__
vfree(b->aux_data);
+#else
+ munmap(b->aux_data, btree_aux_data_bytes(b));
+#endif
b->aux_data = NULL;
-}
-static void btree_node_data_free(struct bch_fs *c, struct btree *b)
-{
- struct btree_cache *bc = &c->btree_cache;
-
- __btree_node_data_free(c, b);
bc->used--;
list_move(&b->list, &bc->freed);
}
b->data = kvpmalloc(btree_bytes(c), gfp);
if (!b->data)
return -ENOMEM;
-
+#ifdef __KERNEL__
b->aux_data = vmalloc_exec(btree_aux_data_bytes(b), gfp);
+#else
+ b->aux_data = mmap(NULL, btree_aux_data_bytes(b),
+ PROT_READ|PROT_WRITE|PROT_EXEC,
+ MAP_PRIVATE|MAP_ANONYMOUS, 0, 0);
+#endif
if (!b->aux_data) {
kvpfree(b->data, btree_bytes(c));
b->data = NULL;
return NULL;
bkey_btree_ptr_init(&b->key);
- six_lock_init(&b->c.lock);
+ __six_lock_init(&b->c.lock, "b->c.lock", &bch2_btree_node_lock_key);
INIT_LIST_HEAD(&b->list);
INIT_LIST_HEAD(&b->write_blocked);
b->byte_order = ilog2(btree_bytes(c));
return b;
}
-static struct btree *btree_node_mem_alloc(struct bch_fs *c)
+struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *c)
{
struct btree_cache *bc = &c->btree_cache;
struct btree *b = __btree_node_mem_alloc(c);
int ret = 0;
lockdep_assert_held(&bc->lock);
+wait_on_io:
+ if (b->flags & ((1U << BTREE_NODE_dirty)|
+ (1U << BTREE_NODE_read_in_flight)|
+ (1U << BTREE_NODE_write_in_flight))) {
+ if (!flush)
+ return -ENOMEM;
+
+ /* XXX: waiting on IO with btree cache lock held */
+ bch2_btree_node_wait_on_read(b);
+ bch2_btree_node_wait_on_write(b);
+ }
if (!six_trylock_intent(&b->c.lock))
return -ENOMEM;
if (!six_trylock_write(&b->c.lock))
goto out_unlock_intent;
+ /* recheck under lock */
+ if (b->flags & ((1U << BTREE_NODE_read_in_flight)|
+ (1U << BTREE_NODE_write_in_flight))) {
+ if (!flush)
+ goto out_unlock;
+ six_unlock_write(&b->c.lock);
+ six_unlock_intent(&b->c.lock);
+ goto wait_on_io;
+ }
+
if (btree_node_noevict(b))
goto out_unlock;
if (!btree_node_may_write(b))
goto out_unlock;
- if (btree_node_dirty(b) &&
- test_bit(BCH_FS_HOLD_BTREE_WRITES, &c->flags))
- goto out_unlock;
-
- if (btree_node_dirty(b) ||
- btree_node_write_in_flight(b) ||
- btree_node_read_in_flight(b)) {
- if (!flush)
+ if (btree_node_dirty(b)) {
+ if (!flush ||
+ test_bit(BCH_FS_HOLD_BTREE_WRITES, &c->flags))
goto out_unlock;
-
- wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight,
- TASK_UNINTERRUPTIBLE);
-
/*
* Using the underscore version because we don't want to compact
* bsets after the write, since this node is about to be evicted
if (bch2_verify_btree_ondisk)
bch2_btree_node_write(c, b, SIX_LOCK_intent);
else
- __bch2_btree_node_write(c, b, SIX_LOCK_read);
+ __bch2_btree_node_write(c, b, false);
- /* wait for any in flight btree write */
- btree_node_wait_on_io(b);
+ six_unlock_write(&b->c.lock);
+ six_unlock_intent(&b->c.lock);
+ goto wait_on_io;
}
out:
if (b->hash_val && !ret)
flags = memalloc_nofs_save();
mutex_lock(&bc->lock);
-#ifdef CONFIG_BCACHEFS_DEBUG
if (c->verify_data)
list_move(&c->verify_data->list, &bc->live);
kvpfree(c->verify_ondisk, btree_bytes(c));
-#endif
for (i = 0; i < BTREE_ID_NR; i++)
if (c->btree_roots[i].b)
bch2_recalc_btree_reserve(c);
for (i = 0; i < bc->reserve; i++)
- if (!btree_node_mem_alloc(c)) {
+ if (!__bch2_btree_node_mem_alloc(c)) {
ret = -ENOMEM;
goto out;
}
list_splice_init(&bc->live, &bc->freeable);
-#ifdef CONFIG_BCACHEFS_DEBUG
mutex_init(&c->verify_lock);
- c->verify_ondisk = kvpmalloc(btree_bytes(c), GFP_KERNEL);
- if (!c->verify_ondisk) {
- ret = -ENOMEM;
- goto out;
- }
-
- c->verify_data = btree_node_mem_alloc(c);
- if (!c->verify_data) {
- ret = -ENOMEM;
- goto out;
- }
-
- list_del_init(&c->verify_data->list);
-#endif
-
bc->shrink.count_objects = bch2_btree_cache_count;
bc->shrink.scan_objects = bch2_btree_cache_scan;
bc->shrink.seeks = 4;
}
BUG_ON(btree_node_hashed(b));
+ BUG_ON(btree_node_dirty(b));
BUG_ON(btree_node_write_in_flight(b));
out:
b->flags = 0;
b->sib_u64s[1] = 0;
b->whiteout_u64s = 0;
bch2_btree_keys_init(b);
+ set_btree_node_accessed(b);
bch2_time_stats_update(&c->times[BCH_TIME_btree_node_mem_alloc],
start_time);
{
struct btree_cache *bc = &c->btree_cache;
struct btree *b;
+ u32 seq;
BUG_ON(level + 1 >= BTREE_MAX_DEPTH);
/*
return NULL;
}
- /*
- * Unlock before doing IO:
- *
- * XXX: ideally should be dropping all btree node locks here
- */
- if (iter && btree_node_read_locked(iter, level + 1))
- btree_node_unlock(iter, level + 1);
-
- bch2_btree_node_read(c, b, sync);
+ set_btree_node_read_in_flight(b);
six_unlock_write(&b->c.lock);
+ seq = b->c.lock.state.seq;
+ six_unlock_intent(&b->c.lock);
- if (!sync) {
- six_unlock_intent(&b->c.lock);
+ /* Unlock before doing IO: */
+ if (iter && sync)
+ bch2_trans_unlock(iter->trans);
+
+ bch2_btree_node_read(c, b, sync);
+
+ if (!sync)
return NULL;
- }
- if (lock_type == SIX_LOCK_read)
- six_lock_downgrade(&b->c.lock);
+ /*
+ * XXX: this will probably always fail because btree_iter_relock()
+ * currently fails for iterators that aren't pointed at a valid btree
+ * node
+ */
+ if (iter &&
+ (!bch2_trans_relock(iter->trans) ||
+ !bch2_btree_iter_relock(iter, _THIS_IP_)))
+ return ERR_PTR(-EINTR);
+
+ if (!six_relock_type(&b->c.lock, lock_type, seq))
+ return ERR_PTR(-EINTR);
return b;
}
return b->hash_val == btree_ptr_hash_val(k) ? 0 : -1;
}
+static noinline void btree_bad_header(struct bch_fs *c, struct btree *b)
+{
+ char buf1[100], buf2[100], buf3[100], buf4[100];
+
+ if (!test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags))
+ return;
+
+ bch2_bpos_to_text(&PBUF(buf1), b->key.k.type == KEY_TYPE_btree_ptr_v2
+ ? bkey_i_to_btree_ptr_v2(&b->key)->v.min_key
+ : POS_MIN);
+ bch2_bpos_to_text(&PBUF(buf2), b->data->min_key);
+
+ bch2_bpos_to_text(&PBUF(buf3), b->key.k.p);
+ bch2_bpos_to_text(&PBUF(buf4), b->data->max_key);
+ bch2_fs_inconsistent(c, "btree node header doesn't match ptr\n"
+ "btree: ptr %u header %llu\n"
+ "level: ptr %u header %llu\n"
+ "min ptr %s node header %s\n"
+ "max ptr %s node header %s",
+ b->c.btree_id, BTREE_NODE_ID(b->data),
+ b->c.level, BTREE_NODE_LEVEL(b->data),
+ buf1, buf2, buf3, buf4);
+}
+
+static inline void btree_check_header(struct bch_fs *c, struct btree *b)
+{
+ if (b->c.btree_id != BTREE_NODE_ID(b->data) ||
+ b->c.level != BTREE_NODE_LEVEL(b->data) ||
+ bpos_cmp(b->data->max_key, b->key.k.p) ||
+ (b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
+ bpos_cmp(b->data->min_key,
+ bkey_i_to_btree_ptr_v2(&b->key)->v.min_key)))
+ btree_bad_header(c, b);
+}
+
/**
* bch_btree_node_get - find a btree node in the cache and lock it, reading it
* in from disk if necessary.
if (bch2_btree_node_relock(iter, level + 1))
goto retry;
- trace_trans_restart_btree_node_reused(iter->trans->ip);
+ trace_trans_restart_btree_node_reused(iter->trans->ip,
+ trace_ip,
+ iter->btree_id,
+ &iter->real_pos);
return ERR_PTR(-EINTR);
}
}
- /* XXX: waiting on IO with btree locks held: */
- wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight,
- TASK_UNINTERRUPTIBLE);
+ if (unlikely(btree_node_read_in_flight(b))) {
+ u32 seq = b->c.lock.state.seq;
+
+ six_unlock_type(&b->c.lock, lock_type);
+ bch2_trans_unlock(iter->trans);
+
+ bch2_btree_node_wait_on_read(b);
+
+ /*
+ * XXX: check if this always fails - btree_iter_relock()
+ * currently fails for iterators that aren't pointed at a valid
+ * btree node
+ */
+ if (iter &&
+ (!bch2_trans_relock(iter->trans) ||
+ !bch2_btree_iter_relock(iter, _THIS_IP_)))
+ return ERR_PTR(-EINTR);
+
+ if (!six_relock_type(&b->c.lock, lock_type, seq))
+ goto retry;
+ }
prefetch(b->aux_data);
EBUG_ON(b->c.btree_id != iter->btree_id);
EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
- EBUG_ON(bpos_cmp(b->data->max_key, k->k.p));
- EBUG_ON(b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
- bpos_cmp(b->data->min_key,
- bkey_i_to_btree_ptr_v2(&b->key)->v.min_key));
+ btree_check_header(c, b);
return b;
}
}
/* XXX: waiting on IO with btree locks held: */
- wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight,
- TASK_UNINTERRUPTIBLE);
+ __bch2_btree_node_wait_on_read(b);
prefetch(b->aux_data);
EBUG_ON(b->c.btree_id != btree_id);
EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
- EBUG_ON(bpos_cmp(b->data->max_key, k->k.p));
- EBUG_ON(b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
- bpos_cmp(b->data->min_key,
- bkey_i_to_btree_ptr_v2(&b->key)->v.min_key));
+ btree_check_header(c, b);
out:
bch2_btree_cache_cannibalize_unlock(c);
return b;
bch2_btree_node_fill(c, iter, k, btree_id, level, SIX_LOCK_read, false);
}
+void bch2_btree_node_evict(struct bch_fs *c, const struct bkey_i *k)
+{
+ struct btree_cache *bc = &c->btree_cache;
+ struct btree *b;
+
+ b = btree_cache_find(bc, k);
+ if (!b)
+ return;
+wait_on_io:
+ /* not allowed to wait on io with btree locks held: */
+
+ /* XXX we're called from btree_gc which will be holding other btree
+ * nodes locked
+ * */
+ __bch2_btree_node_wait_on_read(b);
+ __bch2_btree_node_wait_on_write(b);
+
+ six_lock_intent(&b->c.lock, NULL, NULL);
+ six_lock_write(&b->c.lock, NULL, NULL);
+
+ if (btree_node_dirty(b)) {
+ __bch2_btree_node_write(c, b, false);
+ six_unlock_write(&b->c.lock);
+ six_unlock_intent(&b->c.lock);
+ goto wait_on_io;
+ }
+
+ BUG_ON(btree_node_dirty(b));
+
+ mutex_lock(&bc->lock);
+ btree_node_data_free(c, b);
+ bch2_btree_node_hash_remove(bc, b);
+ mutex_unlock(&bc->lock);
+
+ six_unlock_write(&b->c.lock);
+ six_unlock_intent(&b->c.lock);
+}
+
void bch2_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
struct btree *b)
{
projects / bcachefs-tools-debian / blobdiff