{
unsigned i, reserve = 16;
- if (!c->btree_roots[0].b)
+ if (!c->btree_roots_known[0].b)
reserve += 8;
- for (i = 0; i < BTREE_ID_NR; i++)
- if (c->btree_roots[i].b)
- reserve += min_t(unsigned, 1,
- c->btree_roots[i].b->c.level) * 8;
+ for (i = 0; i < btree_id_nr_alive(c); i++) {
+ struct btree_root *r = bch2_btree_id_root(c, i);
+
+ if (r->b)
+ reserve += min_t(unsigned, 1, r->b->c.level) * 8;
+ }
c->btree_cache.reserve = reserve;
}
EBUG_ON(btree_node_write_in_flight(b));
+ clear_btree_node_just_written(b);
+
kvpfree(b->data, btree_bytes(c));
b->data = NULL;
#ifdef __KERNEL__
- vfree(b->aux_data);
+ kvfree(b->aux_data);
#else
munmap(b->aux_data, btree_aux_data_bytes(b));
#endif
if (!b->data)
return -BCH_ERR_ENOMEM_btree_node_mem_alloc;
#ifdef __KERNEL__
- b->aux_data = vmalloc_exec(btree_aux_data_bytes(b), gfp);
+ b->aux_data = kvmalloc(btree_aux_data_bytes(b), gfp);
#else
b->aux_data = mmap(NULL, btree_aux_data_bytes(b),
PROT_READ|PROT_WRITE|PROT_EXEC,
return NULL;
bkey_btree_ptr_init(&b->key);
- bch2_btree_lock_init(&b->c);
-#ifdef CONFIG_DEBUG_LOCK_ALLOC
- lockdep_set_no_check_recursion(&b->c.lock.dep_map);
-#endif
INIT_LIST_HEAD(&b->list);
INIT_LIST_HEAD(&b->write_blocked);
b->byte_order = ilog2(btree_bytes(c));
return NULL;
}
+ bch2_btree_lock_init(&b->c, 0);
+
bc->used++;
list_add(&b->list, &bc->freeable);
return b;
struct btree *b;
unsigned i, flags;
- if (bc->shrink.list.next)
- unregister_shrinker(&bc->shrink);
+ unregister_shrinker(&bc->shrink);
/* vfree() can allocate memory: */
flags = memalloc_nofs_save();
kvpfree(c->verify_ondisk, btree_bytes(c));
- for (i = 0; i < BTREE_ID_NR; i++)
- if (c->btree_roots[i].b)
- list_add(&c->btree_roots[i].b->list, &bc->live);
+ for (i = 0; i < btree_id_nr_alive(c); i++) {
+ struct btree_root *r = bch2_btree_id_root(c, i);
+
+ if (r->b)
+ list_add(&r->b->list, &bc->live);
+ }
list_splice(&bc->freeable, &bc->live);
while (!list_empty(&bc->freed_nonpcpu)) {
b = list_first_entry(&bc->freed_nonpcpu, struct btree, list);
list_del(&b->list);
- six_lock_pcpu_free(&b->c.lock);
+ six_lock_exit(&b->c.lock);
kfree(b);
}
unsigned i;
int ret = 0;
- pr_verbose_init(c->opts, "");
-
ret = rhashtable_init(&bc->table, &bch_btree_cache_params);
if (ret)
- goto out;
+ goto err;
bc->table_init_done = true;
bch2_recalc_btree_reserve(c);
for (i = 0; i < bc->reserve; i++)
- if (!__bch2_btree_node_mem_alloc(c)) {
- ret = -BCH_ERR_ENOMEM_fs_btree_cache_init;
- goto out;
- }
+ if (!__bch2_btree_node_mem_alloc(c))
+ goto err;
list_splice_init(&bc->live, &bc->freeable);
bc->shrink.scan_objects = bch2_btree_cache_scan;
bc->shrink.to_text = bch2_btree_cache_shrinker_to_text;
bc->shrink.seeks = 4;
- ret = register_shrinker(&bc->shrink, "%s/btree_cache", c->name);
-out:
- pr_verbose_init(c->opts, "ret %i", ret);
- return ret;
+ ret = register_shrinker(&bc->shrink, "%s-btree_cache", c->name);
+ if (ret)
+ goto err;
+
+ return 0;
+err:
+ return -BCH_ERR_ENOMEM_fs_btree_cache_init;
}
void bch2_fs_btree_cache_init_early(struct btree_cache *bc)
goto got_node;
}
- b = __btree_node_mem_alloc(c, __GFP_NOWARN);
+ b = __btree_node_mem_alloc(c, GFP_NOWAIT|__GFP_NOWARN);
if (!b) {
mutex_unlock(&bc->lock);
+ bch2_trans_unlock(trans);
b = __btree_node_mem_alloc(c, GFP_KERNEL);
if (!b)
goto err;
mutex_lock(&bc->lock);
}
- if (pcpu_read_locks)
- six_lock_pcpu_alloc(&b->c.lock);
+ bch2_btree_lock_init(&b->c, pcpu_read_locks ? SIX_LOCK_INIT_PCPU : 0);
BUG_ON(!six_trylock_intent(&b->c.lock));
BUG_ON(!six_trylock_write(&b->c.lock));
mutex_unlock(&bc->lock);
- if (btree_node_data_alloc(c, b, __GFP_NOWARN|GFP_KERNEL))
- goto err;
+ if (btree_node_data_alloc(c, b, GFP_NOWAIT|__GFP_NOWARN)) {
+ bch2_trans_unlock(trans);
+ if (btree_node_data_alloc(c, b, GFP_KERNEL|__GFP_NOWARN))
+ goto err;
+ }
mutex_lock(&bc->lock);
bc->used++;
/* Try to cannibalize another cached btree node: */
if (bc->alloc_lock == current) {
b2 = btree_node_cannibalize(c);
+ clear_btree_node_just_written(b2);
bch2_btree_node_hash_remove(bc, b2);
if (b) {
set_btree_node_read_in_flight(b);
six_unlock_write(&b->c.lock);
- seq = b->c.lock.state.seq;
+ seq = six_lock_seq(&b->c.lock);
six_unlock_intent(&b->c.lock);
/* Unlock before doing IO: */
- if (trans && sync)
- bch2_trans_unlock(trans);
+ if (path && sync)
+ bch2_trans_unlock_noassert(trans);
bch2_btree_node_read(c, b, sync);
{
struct printbuf buf = PRINTBUF;
- if (!test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags))
+ if (c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_allocations)
return;
prt_printf(&buf,
"btree node header doesn't match ptr\n"
"btree %s level %u\n"
"ptr: ",
- bch2_btree_ids[b->c.btree_id], b->c.level);
+ bch2_btree_id_str(b->c.btree_id), b->c.level);
bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
prt_printf(&buf, "\nheader: btree %s level %llu\n"
"min ",
- bch2_btree_ids[BTREE_NODE_ID(b->data)],
+ bch2_btree_id_str(BTREE_NODE_ID(b->data)),
BTREE_NODE_LEVEL(b->data));
bch2_bpos_to_text(&buf, b->data->min_key);
struct btree_cache *bc = &c->btree_cache;
struct btree *b;
struct bset_tree *t;
+ bool need_relock = false;
int ret;
EBUG_ON(level >= BTREE_MAX_DEPTH);
*/
b = bch2_btree_node_fill(trans, path, k, path->btree_id,
level, lock_type, true);
+ need_relock = true;
/* We raced and found the btree node in the cache */
if (!b)
}
if (unlikely(btree_node_read_in_flight(b))) {
- u32 seq = b->c.lock.state.seq;
+ u32 seq = six_lock_seq(&b->c.lock);
six_unlock_type(&b->c.lock, lock_type);
bch2_trans_unlock(trans);
+ need_relock = true;
bch2_btree_node_wait_on_read(b);
* should_be_locked is not set on this path yet, so we need to
* relock it specifically:
*/
- if (trans) {
- int ret = bch2_trans_relock(trans) ?:
- bch2_btree_path_relock_intent(trans, path);
- if (ret) {
- BUG_ON(!trans->restarted);
- return ERR_PTR(ret);
- }
- }
-
if (!six_relock_type(&b->c.lock, lock_type, seq))
goto retry;
}
+ if (unlikely(need_relock)) {
+ ret = bch2_trans_relock(trans) ?:
+ bch2_btree_path_relock_intent(trans, path);
+ if (ret) {
+ six_unlock_type(&b->c.lock, lock_type);
+ return ERR_PTR(ret);
+ }
+ }
+
prefetch(b->aux_data);
for_each_bset(b, t) {
}
/**
- * bch_btree_node_get - find a btree node in the cache and lock it, reading it
+ * bch2_btree_node_get - find a btree node in the cache and lock it, reading it
* in from disk if necessary.
*
+ * @trans: btree transaction object
+ * @path: btree_path being traversed
+ * @k: pointer to btree node (generally KEY_TYPE_btree_ptr_v2)
+ * @level: level of btree node being looked up (0 == leaf node)
+ * @lock_type: SIX_LOCK_read or SIX_LOCK_intent
+ * @trace_ip: ip of caller of btree iterator code (i.e. caller of bch2_btree_iter_peek())
+ *
* The btree node will have either a read or a write lock held, depending on
* the @write parameter.
+ *
+ * Returns: btree node or ERR_PTR()
*/
struct btree *bch2_btree_node_get(struct btree_trans *trans, struct btree_path *path,
const struct bkey_i *k, unsigned level,
}
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(trans);
-
- bch2_btree_node_wait_on_read(b);
-
- /*
- * should_be_locked is not set on this path yet, so we need to
- * relock it specifically:
- */
- if (trans) {
- int ret = bch2_trans_relock(trans) ?:
- bch2_btree_path_relock_intent(trans, path);
- if (ret) {
- BUG_ON(!trans->restarted);
- return ERR_PTR(ret);
- }
- }
-
- if (!six_relock_type(&b->c.lock, lock_type, seq))
- return __bch2_btree_node_get(trans, path, k, level, lock_type, trace_ip);
+ return __bch2_btree_node_get(trans, path, k, level, lock_type, trace_ip);
}
prefetch(b->aux_data);
six_unlock_intent(&b->c.lock);
}
-void bch2_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
- const struct btree *b)
+const char *bch2_btree_id_str(enum btree_id btree)
+{
+ return btree < BTREE_ID_NR ? __bch2_btree_ids[btree] : "(unknown)";
+}
+
+void bch2_btree_pos_to_text(struct printbuf *out, struct bch_fs *c, const struct btree *b)
+{
+ prt_printf(out, "%s level %u/%u\n ",
+ bch2_btree_id_str(b->c.btree_id),
+ b->c.level,
+ bch2_btree_id_root(c, b->c.btree_id)->level);
+ bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
+}
+
+void bch2_btree_node_to_text(struct printbuf *out, struct bch_fs *c, const struct btree *b)
{
- const struct bkey_format *f = &b->format;
struct bset_stats stats;
memset(&stats, 0, sizeof(stats));
prt_printf(out, ":\n"
" ptrs: ");
bch2_val_to_text(out, c, bkey_i_to_s_c(&b->key));
+ prt_newline(out);
+
+ prt_printf(out,
+ " format: ");
+ bch2_bkey_format_to_text(out, &b->format);
- prt_printf(out, "\n"
- " format: u64s %u fields %u %u %u %u %u\n"
+ prt_printf(out,
" unpack fn len: %u\n"
" bytes used %zu/%zu (%zu%% full)\n"
" sib u64s: %u, %u (merge threshold %u)\n"
" nr unpacked keys %u\n"
" floats %zu\n"
" failed unpacked %zu\n",
- f->key_u64s,
- f->bits_per_field[0],
- f->bits_per_field[1],
- f->bits_per_field[2],
- f->bits_per_field[3],
- f->bits_per_field[4],
b->unpack_fn_len,
b->nr.live_u64s * sizeof(u64),
btree_bytes(c) - sizeof(struct btree_node),
projects / bcachefs-tools-debian / blobdiff