#include "recovery.h"
#include "replicas.h"
#include "subvolume.h"
+#include "trace.h"
-#include <linux/prandom.h>
+#include <linux/random.h>
#include <linux/prefetch.h>
-#include <trace/events/bcachefs.h>
static inline void btree_path_list_remove(struct btree_trans *, struct btree_path *);
static inline void btree_path_list_add(struct btree_trans *, struct btree_path *,
*/
static inline int bch2_trans_cond_resched(struct btree_trans *trans)
{
- if (need_resched() || race_fault()) {
- bch2_trans_unlock(trans);
- schedule();
- return bch2_trans_relock(trans);
- } else {
+ if (need_resched() || race_fault())
+ return drop_locks_do(trans, (schedule(), 0));
+ else
return 0;
- }
}
static inline int __btree_path_cmp(const struct btree_path *l,
for (i = 0; i < (!path->cached ? BTREE_MAX_DEPTH : 1); i++) {
if (!path->l[i].b) {
BUG_ON(!path->cached &&
- c->btree_roots[path->btree_id].b->c.level > i);
+ bch2_btree_id_root(c, path->btree_id)->b->c.level > i);
break;
}
BUG_ON(path->cached);
EBUG_ON(!btree_path_pos_in_node(path, b));
- EBUG_ON(b->c.lock.state.seq & 1);
- path->l[b->c.level].lock_seq = b->c.lock.state.seq;
+ path->l[b->c.level].lock_seq = six_lock_seq(&b->c.lock);
path->l[b->c.level].b = b;
__btree_path_level_init(path, b->c.level);
}
if (t != BTREE_NODE_UNLOCKED) {
btree_node_unlock(trans, path, b->c.level);
- six_lock_increment(&b->c.lock, t);
- mark_btree_node_locked(trans, path, b->c.level, t);
+ six_lock_increment(&b->c.lock, (enum six_lock_type) t);
+ mark_btree_node_locked(trans, path, b->c.level, (enum six_lock_type) t);
}
bch2_btree_path_level_init(trans, path, b);
unsigned long trace_ip)
{
struct bch_fs *c = trans->c;
- struct btree *b, **rootp = &c->btree_roots[path->btree_id].b;
+ struct btree *b, **rootp = &bch2_btree_id_root(c, path->btree_id)->b;
enum six_lock_type lock_type;
unsigned i;
int ret;
break;
bch2_bkey_buf_unpack(&tmp, c, l->b, k);
- ret = bch2_btree_node_prefetch(c, trans, path, tmp.k, path->btree_id,
+ ret = bch2_btree_node_prefetch(trans, path, tmp.k, path->btree_id,
path->level - 1);
}
break;
bch2_bkey_buf_reassemble(&tmp, c, k);
- ret = bch2_btree_node_prefetch(c, trans, path, tmp.k, path->btree_id,
+ ret = bch2_btree_node_prefetch(trans, path, tmp.k, path->btree_id,
path->level - 1);
}
trans->in_traverse_all = true;
retry_all:
trans->restarted = 0;
+ trans->last_restarted_ip = 0;
trans_for_each_path(trans, path)
path->should_be_locked = false;
__btree_path_put(path, false);
if (bch2_err_matches(ret, BCH_ERR_transaction_restart) ||
- ret == -ENOMEM)
+ bch2_err_matches(ret, ENOMEM))
goto retry_all;
if (ret)
goto err;
path->uptodate = BTREE_ITER_UPTODATE;
out:
- if (bch2_err_matches(ret, BCH_ERR_transaction_restart) != !!trans->restarted) {
- struct printbuf buf = PRINTBUF;
-
- prt_printf(&buf, "ret %s (%i) trans->restarted %s (%i)\n",
- bch2_err_str(ret), ret,
- bch2_err_str(trans->restarted), trans->restarted);
-#ifdef CONFIG_BCACHEFS_DEBUG
- bch2_prt_backtrace(&buf, &trans->last_restarted);
-#endif
- panic("%s", buf.buf);
- }
+ if (bch2_err_matches(ret, BCH_ERR_transaction_restart) != !!trans->restarted)
+ panic("ret %s (%i) trans->restarted %s (%i)\n",
+ bch2_err_str(ret), ret,
+ bch2_err_str(trans->restarted), trans->restarted);
bch2_btree_path_verify(trans, path);
return ret;
}
{
panic("in transaction restart: %s, last restarted by %pS\n",
bch2_err_str(trans->restarted),
- (void *) trans->last_begin_ip);
+ (void *) trans->last_restarted_ip);
}
noinline __cold
prt_newline(out);
}
-noinline __cold
+static noinline __cold
void __bch2_trans_paths_to_text(struct printbuf *out, struct btree_trans *trans,
bool nosort)
{
__bch2_trans_paths_to_text(out, trans, false);
}
-noinline __cold
+static noinline __cold
void __bch2_dump_trans_paths_updates(struct btree_trans *trans, bool nosort)
{
struct printbuf buf = PRINTBUF;
goto out;
}
+struct btree *bch2_btree_iter_peek_node_and_restart(struct btree_iter *iter)
+{
+ struct btree *b;
+
+ while (b = bch2_btree_iter_peek_node(iter),
+ bch2_err_matches(PTR_ERR_OR_ZERO(b), BCH_ERR_transaction_restart))
+ bch2_trans_begin(iter->trans);
+
+ return b;
+}
+
struct btree *bch2_btree_iter_next_node(struct btree_iter *iter)
{
struct btree_trans *trans = iter->trans;
: NULL;
}
-struct bkey_i *bch2_btree_journal_peek(struct btree_trans *trans,
- struct btree_iter *iter,
- struct bpos end_pos)
+static struct bkey_i *bch2_btree_journal_peek(struct btree_trans *trans,
+ struct btree_iter *iter,
+ struct bpos end_pos)
{
struct bkey_i *k;
return bch2_btree_iter_peek_slot(iter);
}
+struct bkey_s_c bch2_btree_iter_peek_and_restart_outlined(struct btree_iter *iter)
+{
+ struct bkey_s_c k;
+
+ while (btree_trans_too_many_iters(iter->trans) ||
+ (k = bch2_btree_iter_peek_type(iter, iter->flags),
+ bch2_err_matches(bkey_err(k), BCH_ERR_transaction_restart)))
+ bch2_trans_begin(iter->trans);
+
+ return k;
+}
+
/* new transactional stuff: */
#ifdef CONFIG_BCACHEFS_DEBUG
void bch2_trans_iter_exit(struct btree_trans *trans, struct btree_iter *iter)
{
- if (iter->path)
- bch2_path_put(trans, iter->path,
- iter->flags & BTREE_ITER_INTENT);
if (iter->update_path)
bch2_path_put_nokeep(trans, iter->update_path,
iter->flags & BTREE_ITER_INTENT);
+ if (iter->path)
+ bch2_path_put(trans, iter->path,
+ iter->flags & BTREE_ITER_INTENT);
if (iter->key_cache_path)
bch2_path_put(trans, iter->key_cache_path,
iter->flags & BTREE_ITER_INTENT);
unsigned new_top = trans->mem_top + size;
size_t old_bytes = trans->mem_bytes;
size_t new_bytes = roundup_pow_of_two(new_top);
+ int ret;
void *new_mem;
void *p;
WARN_ON_ONCE(new_bytes > BTREE_TRANS_MEM_MAX);
- new_mem = krealloc(trans->mem, new_bytes, GFP_NOFS);
- if (!new_mem && new_bytes <= BTREE_TRANS_MEM_MAX) {
- new_mem = mempool_alloc(&trans->c->btree_trans_mem_pool, GFP_KERNEL);
- new_bytes = BTREE_TRANS_MEM_MAX;
- kfree(trans->mem);
- }
+ new_mem = krealloc(trans->mem, new_bytes, GFP_NOWAIT|__GFP_NOWARN);
+ if (unlikely(!new_mem)) {
+ bch2_trans_unlock(trans);
+
+ new_mem = krealloc(trans->mem, new_bytes, GFP_KERNEL);
+ if (!new_mem && new_bytes <= BTREE_TRANS_MEM_MAX) {
+ new_mem = mempool_alloc(&trans->c->btree_trans_mem_pool, GFP_KERNEL);
+ new_bytes = BTREE_TRANS_MEM_MAX;
+ kfree(trans->mem);
+ }
+
+ if (!new_mem)
+ return ERR_PTR(-BCH_ERR_ENOMEM_trans_kmalloc);
- if (!new_mem)
- return ERR_PTR(-ENOMEM);
+ trans->mem = new_mem;
+ trans->mem_bytes = new_bytes;
+
+ ret = bch2_trans_relock(trans);
+ if (ret)
+ return ERR_PTR(ret);
+ }
trans->mem = new_mem;
trans->mem_bytes = new_bytes;
u32 bch2_trans_begin(struct btree_trans *trans)
{
struct btree_path *path;
+ u64 now;
bch2_trans_reset_updates(trans);
path->preserve = false;
}
+ now = local_clock();
if (!trans->restarted &&
(need_resched() ||
- local_clock() - trans->last_begin_time > BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS)) {
- bch2_trans_unlock(trans);
- cond_resched();
- bch2_trans_relock(trans);
+ now - trans->last_begin_time > BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS)) {
+ drop_locks_do(trans, (cond_resched(), 0));
+ now = local_clock();
}
+ trans->last_begin_time = now;
if (unlikely(time_after(jiffies, trans->srcu_lock_time + msecs_to_jiffies(10))))
bch2_trans_reset_srcu_lock(trans);
trans->notrace_relock_fail = false;
}
- trans->last_begin_time = local_clock();
return trans->restart_count;
}
#endif
if (!p)
p = mempool_alloc(&trans->c->btree_paths_pool, GFP_NOFS);
+ /*
+ * paths need to be zeroed, bch2_check_for_deadlock looks at paths in
+ * other threads
+ */
trans->paths = p; p += paths_bytes;
trans->updates = p; p += updates_bytes;
{
struct btree_transaction_stats *s;
- BUG_ON(lock_class_is_held(&bch2_btree_node_lock_key));
+ bch2_assert_btree_nodes_not_locked();
memset(trans, 0, sizeof(*trans));
trans->c = c;
if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG_TRANSACTIONS)) {
struct btree_trans *pos;
- mutex_lock(&c->btree_trans_lock);
+ seqmutex_lock(&c->btree_trans_lock);
list_for_each_entry(pos, &c->btree_trans_list, list) {
+ /*
+ * We'd much prefer to be stricter here and completely
+ * disallow multiple btree_trans in the same thread -
+ * but the data move path calls bch2_write when we
+ * already have a btree_trans initialized.
+ */
+ BUG_ON(trans->locking_wait.task->pid == pos->locking_wait.task->pid &&
+ bch2_trans_locked(pos));
+
if (trans->locking_wait.task->pid < pos->locking_wait.task->pid) {
list_add_tail(&trans->list, &pos->list);
goto list_add_done;
}
list_add_tail(&trans->list, &c->btree_trans_list);
list_add_done:
- mutex_unlock(&c->btree_trans_lock);
+ seqmutex_unlock(&c->btree_trans_lock);
}
}
bch2_trans_unlock(trans);
+ if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG_TRANSACTIONS)) {
+ seqmutex_lock(&c->btree_trans_lock);
+ list_del(&trans->list);
+ seqmutex_unlock(&c->btree_trans_lock);
+ }
+
closure_sync(&trans->ref);
if (s)
check_btree_paths_leaked(trans);
- if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG_TRANSACTIONS)) {
- mutex_lock(&c->btree_trans_lock);
- list_del(&trans->list);
- mutex_unlock(&c->btree_trans_lock);
- }
-
srcu_read_unlock(&c->btree_trans_barrier, trans->srcu_idx);
bch2_journal_preres_put(&c->journal, &trans->journal_preres);
if (trans->paths)
mempool_free(trans->paths, &c->btree_paths_pool);
-#ifdef CONFIG_BCACHEFS_DEBUG
- darray_exit(&trans->last_restarted);
-#endif
-
trans->mem = (void *) 0x1;
trans->paths = (void *) 0x1;
}
struct btree_path *path;
struct btree_bkey_cached_common *b;
static char lock_types[] = { 'r', 'i', 'w' };
- unsigned l;
+ unsigned l, idx;
if (!out->nr_tabstops) {
printbuf_tabstop_push(out, 16);
prt_printf(out, "%i %s\n", trans->locking_wait.task->pid, trans->fn);
- trans_for_each_path(trans, path) {
+ trans_for_each_path_safe(trans, path, idx) {
if (!path->nodes_locked)
continue;
}
INIT_LIST_HEAD(&c->btree_trans_list);
- mutex_init(&c->btree_trans_lock);
+ seqmutex_init(&c->btree_trans_lock);
ret = mempool_init_kmalloc_pool(&c->btree_paths_pool, 1,
sizeof(struct btree_path) * nr +
projects / bcachefs-tools-debian / blobdiff