#include "error.h"
#include "journal.h"
#include "journal_reclaim.h"
+#include "trace.h"
#include <linux/sched/mm.h>
-#include <trace/events/bcachefs.h>
+#include <linux/seq_buf.h>
static inline bool btree_uses_pcpu_readers(enum btree_id id)
{
const struct bkey_cached *ck = obj;
const struct bkey_cached_key *key = arg->key;
- return cmp_int(ck->key.btree_id, key->btree_id) ?:
- bpos_cmp(ck->key.pos, key->pos);
+ return ck->key.btree_id != key->btree_id ||
+ !bpos_eq(ck->key.pos, key->pos);
}
static const struct rhashtable_params bch2_btree_key_cache_params = {
if (!six_trylock_intent(&ck->c.lock))
return false;
- if (!six_trylock_write(&ck->c.lock)) {
+ if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
six_unlock_intent(&ck->c.lock);
return false;
}
- if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
- six_unlock_write(&ck->c.lock);
+ if (!six_trylock_write(&ck->c.lock)) {
six_unlock_intent(&ck->c.lock);
return false;
}
struct btree_key_cache *bc = &c->btree_key_cache;
struct bkey_cached *ck = NULL;
bool pcpu_readers = btree_uses_pcpu_readers(path->btree_id);
+ int ret;
if (!pcpu_readers) {
#ifdef __KERNEL__
if (ck) {
int ret;
- ret = btree_node_lock_nopath(trans, &ck->c, SIX_LOCK_intent);
+ ret = btree_node_lock_nopath(trans, &ck->c, SIX_LOCK_intent, _THIS_IP_);
if (unlikely(ret)) {
bkey_cached_move_to_freelist(bc, ck);
return ERR_PTR(ret);
}
path->l[0].b = (void *) ck;
- path->l[0].lock_seq = ck->c.lock.state.seq;
+ path->l[0].lock_seq = six_lock_seq(&ck->c.lock);
mark_btree_node_locked(trans, path, 0, SIX_LOCK_intent);
ret = bch2_btree_node_lock_write(trans, path, &ck->c);
return ck;
}
- /* GFP_NOFS because we're holding btree locks: */
- ck = kmem_cache_alloc(bch2_key_cache, GFP_NOFS|__GFP_ZERO);
- if (likely(ck)) {
- INIT_LIST_HEAD(&ck->list);
- __six_lock_init(&ck->c.lock, "b->c.lock", &bch2_btree_node_lock_key);
- if (pcpu_readers)
- six_lock_pcpu_alloc(&ck->c.lock);
-
- ck->c.cached = true;
- BUG_ON(!six_trylock_intent(&ck->c.lock));
- BUG_ON(!six_trylock_write(&ck->c.lock));
- *was_new = true;
- return ck;
+ ck = allocate_dropping_locks(trans, ret,
+ kmem_cache_zalloc(bch2_key_cache, _gfp));
+ if (ret) {
+ kmem_cache_free(bch2_key_cache, ck);
+ return ERR_PTR(ret);
}
- return NULL;
+ if (!ck)
+ return NULL;
+
+ INIT_LIST_HEAD(&ck->list);
+ bch2_btree_lock_init(&ck->c, pcpu_readers ? SIX_LOCK_INIT_PCPU : 0);
+
+ ck->c.cached = true;
+ BUG_ON(!six_trylock_intent(&ck->c.lock));
+ BUG_ON(!six_trylock_write(&ck->c.lock));
+ *was_new = true;
+ return ck;
}
static struct bkey_cached *
if (unlikely(!ck)) {
bch_err(c, "error allocating memory for key cache item, btree %s",
bch2_btree_ids[path->btree_id]);
- return ERR_PTR(-ENOMEM);
+ return ERR_PTR(-BCH_ERR_ENOMEM_btree_key_cache_create);
}
mark_btree_node_locked(trans, path, 0, SIX_LOCK_intent);
- } else {
- if (path->btree_id == BTREE_ID_subvolumes)
- six_lock_pcpu_alloc(&ck->c.lock);
}
ck->c.level = 0;
struct btree_path *ck_path,
struct bkey_cached *ck)
{
- struct btree_path *path;
+ struct btree_iter iter;
struct bkey_s_c k;
unsigned new_u64s = 0;
struct bkey_i *new_k = NULL;
- struct bkey u;
int ret;
- path = bch2_path_get(trans, ck->key.btree_id,
- ck->key.pos, 0, 0, 0, _THIS_IP_);
- ret = bch2_btree_path_traverse(trans, path, 0);
+ k = bch2_bkey_get_iter(trans, &iter, ck->key.btree_id, ck->key.pos,
+ BTREE_ITER_KEY_CACHE_FILL|
+ BTREE_ITER_CACHED_NOFILL);
+ ret = bkey_err(k);
if (ret)
goto err;
- k = bch2_btree_path_peek_slot(path, &u);
-
if (!bch2_btree_node_relock(trans, ck_path, 0)) {
trace_and_count(trans->c, trans_restart_relock_key_cache_fill, trans, _THIS_IP_, ck_path);
- ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_raced);
+ ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_fill);
goto err;
}
if (new_u64s > ck->u64s) {
new_u64s = roundup_pow_of_two(new_u64s);
- new_k = kmalloc(new_u64s * sizeof(u64), GFP_NOFS);
+ new_k = kmalloc(new_u64s * sizeof(u64), GFP_NOWAIT|__GFP_NOWARN);
if (!new_k) {
- bch_err(trans->c, "error allocating memory for key cache key, btree %s u64s %u",
- bch2_btree_ids[ck->key.btree_id], new_u64s);
- ret = -ENOMEM;
- goto err;
+ bch2_trans_unlock(trans);
+
+ new_k = kmalloc(new_u64s * sizeof(u64), GFP_KERNEL);
+ if (!new_k) {
+ bch_err(trans->c, "error allocating memory for key cache key, btree %s u64s %u",
+ bch2_btree_ids[ck->key.btree_id], new_u64s);
+ ret = -BCH_ERR_ENOMEM_btree_key_cache_fill;
+ goto err;
+ }
+
+ if (!bch2_btree_node_relock(trans, ck_path, 0)) {
+ kfree(new_k);
+ trace_and_count(trans->c, trans_restart_relock_key_cache_fill, trans, _THIS_IP_, ck_path);
+ ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_fill);
+ goto err;
+ }
+
+ ret = bch2_trans_relock(trans);
+ if (ret) {
+ kfree(new_k);
+ goto err;
+ }
}
}
bch2_btree_node_unlock_write(trans, ck_path, ck_path->l[0].b);
/* We're not likely to need this iterator again: */
- path->preserve = false;
+ set_btree_iter_dontneed(&iter);
err:
- bch2_path_put(trans, path, 0);
+ bch2_trans_iter_exit(trans, &iter);
return ret;
}
path->l[1].b = NULL;
- if (bch2_btree_node_relock(trans, path, 0)) {
+ if (bch2_btree_node_relock_notrace(trans, path, 0)) {
ck = (void *) path->l[0].b;
goto fill;
}
BUG_ON(ret);
if (ck->key.btree_id != path->btree_id ||
- bpos_cmp(ck->key.pos, path->pos)) {
+ !bpos_eq(ck->key.pos, path->pos)) {
six_unlock_type(&ck->c.lock, lock_want);
goto retry;
}
mark_btree_node_locked(trans, path, 0, lock_want);
}
- path->l[0].lock_seq = ck->c.lock.state.seq;
+ path->l[0].lock_seq = six_lock_seq(&ck->c.lock);
path->l[0].b = (void *) ck;
fill:
- if (!ck->valid) {
+ path->uptodate = BTREE_ITER_UPTODATE;
+
+ if (!ck->valid && !(flags & BTREE_ITER_CACHED_NOFILL)) {
/*
* Using the underscore version because we haven't set
* path->uptodate yet:
ret = btree_key_cache_fill(trans, path, ck);
if (ret)
goto err;
+
+ ret = bch2_btree_path_relock(trans, path, _THIS_IP_);
+ if (ret)
+ goto err;
+
+ path->uptodate = BTREE_ITER_UPTODATE;
}
if (!test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
set_bit(BKEY_CACHED_ACCESSED, &ck->flags);
- path->uptodate = BTREE_ITER_UPTODATE;
- BUG_ON(!ck->valid);
BUG_ON(btree_node_locked_type(path, 0) != btree_lock_want(path, 0));
+ BUG_ON(path->uptodate);
return ret;
err:
+ path->uptodate = BTREE_ITER_NEED_TRAVERSE;
if (!bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
btree_node_unlock(trans, path, 0);
path->l[0].b = ERR_PTR(ret);
path->l[1].b = NULL;
- if (bch2_btree_node_relock(trans, path, 0)) {
+ if (bch2_btree_node_relock_notrace(trans, path, 0)) {
ck = (void *) path->l[0].b;
goto fill;
}
return ret;
if (ck->key.btree_id != path->btree_id ||
- bpos_cmp(ck->key.pos, path->pos)) {
+ !bpos_eq(ck->key.pos, path->pos)) {
six_unlock_type(&ck->c.lock, lock_want);
goto retry;
}
mark_btree_node_locked(trans, path, 0, lock_want);
}
- path->l[0].lock_seq = ck->c.lock.state.seq;
+ path->l[0].lock_seq = six_lock_seq(&ck->c.lock);
path->l[0].b = (void *) ck;
fill:
if (!ck->valid)
bch2_trans_commit(trans, NULL, NULL,
BTREE_INSERT_NOCHECK_RW|
BTREE_INSERT_NOFAIL|
- BTREE_INSERT_USE_RESERVE|
(ck->journal.seq == journal_last_seq(j)
- ? JOURNAL_WATERMARK_reserved
+ ? BCH_WATERMARK_reclaim
: 0)|
commit_flags);
six_unlock_read(&ck->c.lock);
goto unlock;
}
+
+ if (ck->seq != seq) {
+ bch2_journal_pin_update(&c->journal, ck->seq, &ck->journal,
+ bch2_btree_key_cache_journal_flush);
+ six_unlock_read(&ck->c.lock);
+ goto unlock;
+ }
six_unlock_read(&ck->c.lock);
ret = commit_do(&trans, NULL, NULL, 0,
}
bool bch2_btree_insert_key_cached(struct btree_trans *trans,
- struct btree_path *path,
- struct bkey_i *insert)
+ unsigned flags,
+ struct btree_insert_entry *insert_entry)
{
struct bch_fs *c = trans->c;
- struct bkey_cached *ck = (void *) path->l[0].b;
+ struct bkey_cached *ck = (void *) insert_entry->path->l[0].b;
+ struct bkey_i *insert = insert_entry->k;
bool kick_reclaim = false;
- BUG_ON(insert->u64s > ck->u64s);
+ BUG_ON(insert->k.u64s > ck->u64s);
- if (likely(!(trans->flags & BTREE_INSERT_JOURNAL_REPLAY))) {
+ if (likely(!(flags & BTREE_INSERT_JOURNAL_REPLAY))) {
int difference;
- BUG_ON(jset_u64s(insert->u64s) > trans->journal_preres.u64s);
+ BUG_ON(jset_u64s(insert->k.u64s) > trans->journal_preres.u64s);
- difference = jset_u64s(insert->u64s) - ck->res.u64s;
+ difference = jset_u64s(insert->k.u64s) - ck->res.u64s;
if (difference > 0) {
trans->journal_preres.u64s -= difference;
ck->res.u64s += difference;
ck->valid = true;
if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
+ EBUG_ON(test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags));
set_bit(BKEY_CACHED_DIRTY, &ck->flags);
atomic_long_inc(&c->btree_key_cache.nr_dirty);
kick_reclaim = true;
}
- bch2_journal_pin_update(&c->journal, trans->journal_res.seq,
- &ck->journal, bch2_btree_key_cache_journal_flush);
+ /*
+ * To minimize lock contention, we only add the journal pin here and
+ * defer pin updates to the flush callback via ->seq. Be careful not to
+ * update ->seq on nojournal commits because we don't want to update the
+ * pin to a seq that doesn't include journal updates on disk. Otherwise
+ * we risk losing the update after a crash.
+ *
+ * The only exception is if the pin is not active in the first place. We
+ * have to add the pin because journal reclaim drives key cache
+ * flushing. The flush callback will not proceed unless ->seq matches
+ * the latest pin, so make sure it starts with a consistent value.
+ */
+ if (!(insert_entry->flags & BTREE_UPDATE_NOJOURNAL) ||
+ !journal_pin_active(&ck->journal)) {
+ ck->seq = trans->journal_res.seq;
+ }
+ bch2_journal_pin_add(&c->journal, trans->journal_res.seq,
+ &ck->journal, bch2_btree_key_cache_journal_flush);
if (kick_reclaim)
journal_reclaim_kick(&c->journal);
break;
list_del(&ck->list);
- six_lock_pcpu_free(&ck->c.lock);
+ six_lock_exit(&ck->c.lock);
kmem_cache_free(bch2_key_cache, ck);
atomic_long_dec(&bc->nr_freed);
scanned++;
break;
list_del(&ck->list);
- six_lock_pcpu_free(&ck->c.lock);
+ six_lock_exit(&ck->c.lock);
kmem_cache_free(bch2_key_cache, ck);
atomic_long_dec(&bc->nr_freed);
scanned++;
struct bucket_table *tbl;
struct bkey_cached *ck, *n;
struct rhash_head *pos;
+ LIST_HEAD(items);
unsigned i;
#ifdef __KERNEL__
int cpu;
#endif
- if (bc->shrink.list.next)
- unregister_shrinker(&bc->shrink);
+ unregister_shrinker(&bc->shrink);
mutex_lock(&bc->lock);
for (i = 0; i < tbl->size; i++)
rht_for_each_entry_rcu(ck, pos, tbl, i, hash) {
bkey_cached_evict(bc, ck);
- list_add(&ck->list, &bc->freed_nonpcpu);
+ list_add(&ck->list, &items);
}
rcu_read_unlock();
}
for (i = 0; i < f->nr; i++) {
ck = f->objs[i];
- list_add(&ck->list, &bc->freed_nonpcpu);
+ list_add(&ck->list, &items);
}
}
#endif
- list_splice(&bc->freed_pcpu, &bc->freed_nonpcpu);
+ list_splice(&bc->freed_pcpu, &items);
+ list_splice(&bc->freed_nonpcpu, &items);
- list_for_each_entry_safe(ck, n, &bc->freed_nonpcpu, list) {
+ mutex_unlock(&bc->lock);
+
+ list_for_each_entry_safe(ck, n, &items, list) {
cond_resched();
bch2_journal_pin_drop(&c->journal, &ck->journal);
list_del(&ck->list);
kfree(ck->k);
- six_lock_pcpu_free(&ck->c.lock);
+ six_lock_exit(&ck->c.lock);
kmem_cache_free(bch2_key_cache, ck);
}
panic("btree key cache shutdown error: nr_keys nonzero (%li)\n",
atomic_long_read(&bc->nr_keys));
- mutex_unlock(&bc->lock);
-
if (bc->table_init_done)
rhashtable_destroy(&bc->table);
INIT_LIST_HEAD(&c->freed_nonpcpu);
}
-static void bch2_btree_key_cache_shrinker_to_text(struct printbuf *out, struct shrinker *shrink)
+static void bch2_btree_key_cache_shrinker_to_text(struct seq_buf *s, struct shrinker *shrink)
{
struct btree_key_cache *bc =
container_of(shrink, struct btree_key_cache, shrink);
+ char *cbuf;
+ size_t buflen = seq_buf_get_buf(s, &cbuf);
+ struct printbuf out = PRINTBUF_EXTERN(cbuf, buflen);
- bch2_btree_key_cache_to_text(out, bc);
+ bch2_btree_key_cache_to_text(&out, bc);
+ seq_buf_commit(s, out.pos);
}
int bch2_fs_btree_key_cache_init(struct btree_key_cache *bc)
{
struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
- int ret;
#ifdef __KERNEL__
bc->pcpu_freed = alloc_percpu(struct btree_key_cache_freelist);
if (!bc->pcpu_freed)
- return -ENOMEM;
+ return -BCH_ERR_ENOMEM_fs_btree_cache_init;
#endif
- ret = rhashtable_init(&bc->table, &bch2_btree_key_cache_params);
- if (ret)
- return ret;
+ if (rhashtable_init(&bc->table, &bch2_btree_key_cache_params))
+ return -BCH_ERR_ENOMEM_fs_btree_cache_init;
bc->table_init_done = true;
bc->shrink.count_objects = bch2_btree_key_cache_count;
bc->shrink.scan_objects = bch2_btree_key_cache_scan;
bc->shrink.to_text = bch2_btree_key_cache_shrinker_to_text;
- return register_shrinker(&bc->shrink, "%s/btree_key_cache", c->name);
+ if (register_shrinker(&bc->shrink, "%s/btree_key_cache", c->name))
+ return -BCH_ERR_ENOMEM_fs_btree_cache_init;
+ return 0;
}
void bch2_btree_key_cache_to_text(struct printbuf *out, struct btree_key_cache *c)
int __init bch2_btree_key_cache_init(void)
{
- bch2_key_cache = KMEM_CACHE(bkey_cached, 0);
+ bch2_key_cache = KMEM_CACHE(bkey_cached, SLAB_RECLAIM_ACCOUNT);
if (!bch2_key_cache)
return -ENOMEM;