#include "bkey_methods.h"
#include "btree_cache.h"
#include "btree_gc.h"
+#include "btree_journal_iter.h"
#include "btree_update.h"
#include "btree_update_interior.h"
#include "btree_io.h"
#include "btree_iter.h"
#include "btree_locking.h"
#include "buckets.h"
+#include "clock.h"
#include "error.h"
#include "extents.h"
#include "journal.h"
#include "keylist.h"
#include "replicas.h"
#include "super-io.h"
+#include "trace.h"
#include <linux/random.h>
-#include <trace/events/bcachefs.h>
+
+static int bch2_btree_insert_node(struct btree_update *, struct btree_trans *,
+ btree_path_idx_t, struct btree *,
+ struct keylist *, unsigned);
+static void bch2_btree_update_add_new_node(struct btree_update *, struct btree *);
+
+static btree_path_idx_t get_unlocked_mut_path(struct btree_trans *trans,
+ enum btree_id btree_id,
+ unsigned level,
+ struct bpos pos)
+{
+ btree_path_idx_t path_idx = bch2_path_get(trans, btree_id, pos, level + 1, level,
+ BTREE_ITER_NOPRESERVE|
+ BTREE_ITER_INTENT, _RET_IP_);
+ path_idx = bch2_btree_path_make_mut(trans, path_idx, true, _RET_IP_);
+
+ struct btree_path *path = trans->paths + path_idx;
+ bch2_btree_path_downgrade(trans, path);
+ __bch2_btree_path_unlock(trans, path);
+ return path_idx;
+}
/* Debug code: */
struct bkey_s_c k;
struct bkey_s_c_btree_ptr_v2 bp;
struct bkey unpacked;
- char buf1[100], buf2[100];
+ struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
BUG_ON(!b->c.level);
- if (!test_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags))
+ if (!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
return;
bch2_btree_node_iter_init_from_start(&iter, b);
break;
bp = bkey_s_c_to_btree_ptr_v2(k);
- if (bkey_cmp(next_node, bp.v->min_key)) {
+ if (!bpos_eq(next_node, bp.v->min_key)) {
bch2_dump_btree_node(c, b);
- panic("expected next min_key %s got %s\n",
- (bch2_bpos_to_text(&PBUF(buf1), next_node), buf1),
- (bch2_bpos_to_text(&PBUF(buf2), bp.v->min_key), buf2));
+ bch2_bpos_to_text(&buf1, next_node);
+ bch2_bpos_to_text(&buf2, bp.v->min_key);
+ panic("expected next min_key %s got %s\n", buf1.buf, buf2.buf);
}
bch2_btree_node_iter_advance(&iter, b);
if (bch2_btree_node_iter_end(&iter)) {
- if (bkey_cmp(k.k->p, b->key.k.p)) {
+ if (!bpos_eq(k.k->p, b->key.k.p)) {
bch2_dump_btree_node(c, b);
- panic("expected end %s got %s\n",
- (bch2_bpos_to_text(&PBUF(buf1), b->key.k.p), buf1),
- (bch2_bpos_to_text(&PBUF(buf2), k.k->p), buf2));
+ bch2_bpos_to_text(&buf1, b->key.k.p);
+ bch2_bpos_to_text(&buf2, k.k->p);
+ panic("expected end %s got %s\n", buf1.buf, buf2.buf);
}
break;
}
- next_node = bkey_successor(k.k->p);
+ next_node = bpos_successor(k.k->p);
}
#endif
}
/* Calculate ideal packed bkey format for new btree nodes: */
-void __bch2_btree_calc_format(struct bkey_format_state *s, struct btree *b)
+static void __bch2_btree_calc_format(struct bkey_format_state *s, struct btree *b)
{
struct bkey_packed *k;
struct bset_tree *t;
struct bkey uk;
- bch2_bkey_format_add_pos(s, b->data->min_key);
-
for_each_bset(b, t)
bset_tree_for_each_key(b, t, k)
if (!bkey_deleted(k)) {
struct bkey_format_state s;
bch2_bkey_format_init(&s);
+ bch2_bkey_format_add_pos(&s, b->data->min_key);
+ bch2_bkey_format_add_pos(&s, b->data->max_key);
__bch2_btree_calc_format(&s, b);
return bch2_bkey_format_done(&s);
}
-static size_t btree_node_u64s_with_format(struct btree *b,
+static size_t btree_node_u64s_with_format(struct btree_nr_keys nr,
+ struct bkey_format *old_f,
struct bkey_format *new_f)
{
- struct bkey_format *old_f = &b->format;
-
/* stupid integer promotion rules */
ssize_t delta =
(((int) new_f->key_u64s - old_f->key_u64s) *
- (int) b->nr.packed_keys) +
+ (int) nr.packed_keys) +
(((int) new_f->key_u64s - BKEY_U64s) *
- (int) b->nr.unpacked_keys);
+ (int) nr.unpacked_keys);
- BUG_ON(delta + b->nr.live_u64s < 0);
+ BUG_ON(delta + nr.live_u64s < 0);
- return b->nr.live_u64s + delta;
+ return nr.live_u64s + delta;
}
/**
- * btree_node_format_fits - check if we could rewrite node with a new format
+ * bch2_btree_node_format_fits - check if we could rewrite node with a new format
+ *
+ * @c: filesystem handle
+ * @b: btree node to rewrite
+ * @nr: number of keys for new node (i.e. b->nr)
+ * @new_f: bkey format to translate keys to
+ *
+ * Returns: true if all re-packed keys will be able to fit in a new node.
*
- * This assumes all keys can pack with the new format -- it just checks if
- * the re-packed keys would fit inside the node itself.
+ * Assumes all keys will successfully pack with the new format.
*/
-bool bch2_btree_node_format_fits(struct bch_fs *c, struct btree *b,
+static bool bch2_btree_node_format_fits(struct bch_fs *c, struct btree *b,
+ struct btree_nr_keys nr,
struct bkey_format *new_f)
{
- size_t u64s = btree_node_u64s_with_format(b, new_f);
+ size_t u64s = btree_node_u64s_with_format(nr, &b->format, new_f);
return __vstruct_bytes(struct btree_node, u64s) < btree_bytes(c);
}
/* Btree node freeing/allocation: */
-static void __btree_node_free(struct bch_fs *c, struct btree *b)
+static void __btree_node_free(struct btree_trans *trans, struct btree *b)
{
- trace_btree_node_free(c, b);
+ struct bch_fs *c = trans->c;
+
+ trace_and_count(c, btree_node_free, trans, b);
+ BUG_ON(btree_node_write_blocked(b));
BUG_ON(btree_node_dirty(b));
BUG_ON(btree_node_need_write(b));
BUG_ON(b == btree_node_root(c, b));
clear_btree_node_noevict(b);
- bch2_btree_node_hash_remove(&c->btree_cache, b);
-
mutex_lock(&c->btree_cache.lock);
list_move(&b->list, &c->btree_cache.freeable);
mutex_unlock(&c->btree_cache.lock);
}
-void bch2_btree_node_free_never_inserted(struct bch_fs *c, struct btree *b)
+static void bch2_btree_node_free_inmem(struct btree_trans *trans,
+ struct btree_path *path,
+ struct btree *b)
{
- struct open_buckets ob = b->ob;
-
- b->ob.nr = 0;
-
- clear_btree_node_dirty(c, b);
+ struct bch_fs *c = trans->c;
+ unsigned i, level = b->c.level;
- btree_node_lock_type(c, b, SIX_LOCK_write);
- __btree_node_free(c, b);
+ bch2_btree_node_lock_write_nofail(trans, path, &b->c);
+ bch2_btree_node_hash_remove(&c->btree_cache, b);
+ __btree_node_free(trans, b);
six_unlock_write(&b->c.lock);
+ mark_btree_node_locked_noreset(path, level, BTREE_NODE_INTENT_LOCKED);
- bch2_open_buckets_put(c, &ob);
+ trans_for_each_path(trans, path, i)
+ if (path->l[level].b == b) {
+ btree_node_unlock(trans, path, level);
+ path->l[level].b = ERR_PTR(-BCH_ERR_no_btree_node_init);
+ }
}
-void bch2_btree_node_free_inmem(struct bch_fs *c, struct btree *b,
- struct btree_iter *iter)
+static void bch2_btree_node_free_never_used(struct btree_update *as,
+ struct btree_trans *trans,
+ struct btree *b)
{
- struct btree_iter *linked;
+ struct bch_fs *c = as->c;
+ struct prealloc_nodes *p = &as->prealloc_nodes[b->c.lock.readers != NULL];
+ struct btree_path *path;
+ unsigned i, level = b->c.level;
- trans_for_each_iter(iter->trans, linked)
- BUG_ON(linked->l[b->c.level].b == b);
+ BUG_ON(!list_empty(&b->write_blocked));
+ BUG_ON(b->will_make_reachable != (1UL|(unsigned long) as));
+
+ b->will_make_reachable = 0;
+ closure_put(&as->cl);
+
+ clear_btree_node_will_make_reachable(b);
+ clear_btree_node_accessed(b);
+ clear_btree_node_dirty_acct(c, b);
+ clear_btree_node_need_write(b);
+
+ mutex_lock(&c->btree_cache.lock);
+ list_del_init(&b->list);
+ bch2_btree_node_hash_remove(&c->btree_cache, b);
+ mutex_unlock(&c->btree_cache.lock);
+
+ BUG_ON(p->nr >= ARRAY_SIZE(p->b));
+ p->b[p->nr++] = b;
- six_lock_write(&b->c.lock, NULL, NULL);
- __btree_node_free(c, b);
- six_unlock_write(&b->c.lock);
six_unlock_intent(&b->c.lock);
+
+ trans_for_each_path(trans, path, i)
+ if (path->l[level].b == b) {
+ btree_node_unlock(trans, path, level);
+ path->l[level].b = ERR_PTR(-BCH_ERR_no_btree_node_init);
+ }
}
-static struct btree *__bch2_btree_node_alloc(struct bch_fs *c,
+static struct btree *__bch2_btree_node_alloc(struct btree_trans *trans,
struct disk_reservation *res,
struct closure *cl,
+ bool interior_node,
unsigned flags)
{
+ struct bch_fs *c = trans->c;
struct write_point *wp;
struct btree *b;
- __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp;
- struct open_buckets ob = { .nr = 0 };
+ BKEY_PADDED_ONSTACK(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp;
+ struct open_buckets obs = { .nr = 0 };
struct bch_devs_list devs_have = (struct bch_devs_list) { 0 };
- unsigned nr_reserve;
- enum alloc_reserve alloc_reserve;
-
- if (flags & BTREE_INSERT_USE_RESERVE) {
- nr_reserve = 0;
- alloc_reserve = RESERVE_BTREE_MOVINGGC;
- } else {
- nr_reserve = BTREE_NODE_RESERVE;
- alloc_reserve = RESERVE_BTREE;
- }
+ enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
+ unsigned nr_reserve = watermark > BCH_WATERMARK_reclaim
+ ? BTREE_NODE_RESERVE
+ : 0;
+ int ret;
mutex_lock(&c->btree_reserve_cache_lock);
if (c->btree_reserve_cache_nr > nr_reserve) {
struct btree_alloc *a =
&c->btree_reserve_cache[--c->btree_reserve_cache_nr];
- ob = a->ob;
+ obs = a->ob;
bkey_copy(&tmp.k, &a->k);
mutex_unlock(&c->btree_reserve_cache_lock);
goto mem_alloc;
mutex_unlock(&c->btree_reserve_cache_lock);
retry:
- wp = bch2_alloc_sectors_start(c,
+ ret = bch2_alloc_sectors_start_trans(trans,
c->opts.metadata_target ?:
c->opts.foreground_target,
0,
&devs_have,
res->nr_replicas,
c->opts.metadata_replicas_required,
- alloc_reserve, 0, cl);
- if (IS_ERR(wp))
- return ERR_CAST(wp);
+ watermark, 0, cl, &wp);
+ if (unlikely(ret))
+ return ERR_PTR(ret);
- if (wp->sectors_free < c->opts.btree_node_size) {
+ if (wp->sectors_free < btree_sectors(c)) {
struct open_bucket *ob;
unsigned i;
open_bucket_for_each(c, &wp->ptrs, ob, i)
- if (ob->sectors_free < c->opts.btree_node_size)
+ if (ob->sectors_free < btree_sectors(c))
ob->sectors_free = 0;
bch2_alloc_sectors_done(c, wp);
goto retry;
}
- if (c->sb.features & (1ULL << BCH_FEATURE_btree_ptr_v2))
- bkey_btree_ptr_v2_init(&tmp.k);
- else
- bkey_btree_ptr_init(&tmp.k);
-
- bch2_alloc_sectors_append_ptrs(c, wp, &tmp.k, c->opts.btree_node_size);
+ bkey_btree_ptr_v2_init(&tmp.k);
+ bch2_alloc_sectors_append_ptrs(c, wp, &tmp.k, btree_sectors(c), false);
- bch2_open_bucket_get(c, wp, &ob);
+ bch2_open_bucket_get(c, wp, &obs);
bch2_alloc_sectors_done(c, wp);
mem_alloc:
- b = bch2_btree_node_mem_alloc(c);
+ b = bch2_btree_node_mem_alloc(trans, interior_node);
+ six_unlock_write(&b->c.lock);
+ six_unlock_intent(&b->c.lock);
/* we hold cannibalize_lock: */
BUG_ON(IS_ERR(b));
BUG_ON(b->ob.nr);
bkey_copy(&b->key, &tmp.k);
- b->ob = ob;
+ b->ob = obs;
return b;
}
-static struct btree *bch2_btree_node_alloc(struct btree_update *as, unsigned level)
+static struct btree *bch2_btree_node_alloc(struct btree_update *as,
+ struct btree_trans *trans,
+ unsigned level)
{
struct bch_fs *c = as->c;
struct btree *b;
+ struct prealloc_nodes *p = &as->prealloc_nodes[!!level];
int ret;
BUG_ON(level >= BTREE_MAX_DEPTH);
- BUG_ON(!as->nr_prealloc_nodes);
+ BUG_ON(!p->nr);
- b = as->prealloc_nodes[--as->nr_prealloc_nodes];
+ b = p->b[--p->nr];
+
+ btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
+ btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write);
set_btree_node_accessed(b);
- set_btree_node_dirty(c, b);
+ set_btree_node_dirty_acct(c, b);
set_btree_node_need_write(b);
bch2_bset_init_first(b, &b->data->keys);
memset(&b->nr, 0, sizeof(b->nr));
b->data->magic = cpu_to_le64(bset_magic(c));
+ memset(&b->data->_ptr, 0, sizeof(b->data->_ptr));
b->data->flags = 0;
SET_BTREE_NODE_ID(b->data, as->btree_id);
SET_BTREE_NODE_LEVEL(b->data, level);
- b->data->ptr = bch2_bkey_ptrs_c(bkey_i_to_s_c(&b->key)).start->ptr;
if (b->key.k.type == KEY_TYPE_btree_ptr_v2) {
struct bkey_i_btree_ptr_v2 *bp = bkey_i_to_btree_ptr_v2(&b->key);
ret = bch2_btree_node_hash_insert(&c->btree_cache, b, level, as->btree_id);
BUG_ON(ret);
- trace_btree_node_alloc(c, b);
+ trace_and_count(c, btree_node_alloc, trans, b);
+ bch2_increment_clock(c, btree_sectors(c), WRITE);
return b;
}
b->data->max_key = pos;
}
-struct btree *__bch2_btree_node_alloc_replacement(struct btree_update *as,
- struct btree *b,
- struct bkey_format format)
+static struct btree *bch2_btree_node_alloc_replacement(struct btree_update *as,
+ struct btree_trans *trans,
+ struct btree *b)
{
- struct btree *n;
+ struct btree *n = bch2_btree_node_alloc(as, trans, b->c.level);
+ struct bkey_format format = bch2_btree_calc_format(b);
- n = bch2_btree_node_alloc(as, b->c.level);
+ /*
+ * The keys might expand with the new format - if they wouldn't fit in
+ * the btree node anymore, use the old format for now:
+ */
+ if (!bch2_btree_node_format_fits(as->c, b, b->nr, &format))
+ format = b->format;
SET_BTREE_NODE_SEQ(n->data, BTREE_NODE_SEQ(b->data) + 1);
bch2_btree_sort_into(as->c, n, b);
btree_node_reset_sib_u64s(n);
-
- n->key.k.p = b->key.k.p;
return n;
}
-static struct btree *bch2_btree_node_alloc_replacement(struct btree_update *as,
- struct btree *b)
-{
- struct bkey_format new_f = bch2_btree_calc_format(b);
-
- /*
- * The keys might expand with the new format - if they wouldn't fit in
- * the btree node anymore, use the old format for now:
- */
- if (!bch2_btree_node_format_fits(as->c, b, &new_f))
- new_f = b->format;
-
- return __bch2_btree_node_alloc_replacement(as, b, new_f);
-}
-
-static struct btree *__btree_root_alloc(struct btree_update *as, unsigned level)
+static struct btree *__btree_root_alloc(struct btree_update *as,
+ struct btree_trans *trans, unsigned level)
{
- struct btree *b = bch2_btree_node_alloc(as, level);
+ struct btree *b = bch2_btree_node_alloc(as, trans, level);
btree_set_min(b, POS_MIN);
- btree_set_max(b, POS_MAX);
+ btree_set_max(b, SPOS_MAX);
b->data->format = bch2_btree_calc_format(b);
btree_node_set_format(b, b->data->format);
bch2_btree_build_aux_trees(b);
- bch2_btree_update_add_new_node(as, b);
- six_unlock_write(&b->c.lock);
-
return b;
}
-static void bch2_btree_reserve_put(struct btree_update *as)
+static void bch2_btree_reserve_put(struct btree_update *as, struct btree_trans *trans)
{
struct bch_fs *c = as->c;
+ struct prealloc_nodes *p;
- mutex_lock(&c->btree_reserve_cache_lock);
-
- while (as->nr_prealloc_nodes) {
- struct btree *b = as->prealloc_nodes[--as->nr_prealloc_nodes];
+ for (p = as->prealloc_nodes;
+ p < as->prealloc_nodes + ARRAY_SIZE(as->prealloc_nodes);
+ p++) {
+ while (p->nr) {
+ struct btree *b = p->b[--p->nr];
- six_unlock_write(&b->c.lock);
+ mutex_lock(&c->btree_reserve_cache_lock);
- if (c->btree_reserve_cache_nr <
- ARRAY_SIZE(c->btree_reserve_cache)) {
- struct btree_alloc *a =
- &c->btree_reserve_cache[c->btree_reserve_cache_nr++];
+ if (c->btree_reserve_cache_nr <
+ ARRAY_SIZE(c->btree_reserve_cache)) {
+ struct btree_alloc *a =
+ &c->btree_reserve_cache[c->btree_reserve_cache_nr++];
- a->ob = b->ob;
- b->ob.nr = 0;
- bkey_copy(&a->k, &b->key);
- } else {
- bch2_open_buckets_put(c, &b->ob);
- }
+ a->ob = b->ob;
+ b->ob.nr = 0;
+ bkey_copy(&a->k, &b->key);
+ } else {
+ bch2_open_buckets_put(c, &b->ob);
+ }
- btree_node_lock_type(c, b, SIX_LOCK_write);
- __btree_node_free(c, b);
- six_unlock_write(&b->c.lock);
+ mutex_unlock(&c->btree_reserve_cache_lock);
- six_unlock_intent(&b->c.lock);
+ btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
+ btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write);
+ __btree_node_free(trans, b);
+ six_unlock_write(&b->c.lock);
+ six_unlock_intent(&b->c.lock);
+ }
}
-
- mutex_unlock(&c->btree_reserve_cache_lock);
}
-static int bch2_btree_reserve_get(struct btree_update *as, unsigned nr_nodes,
- unsigned flags, struct closure *cl)
+static int bch2_btree_reserve_get(struct btree_trans *trans,
+ struct btree_update *as,
+ unsigned nr_nodes[2],
+ unsigned flags,
+ struct closure *cl)
{
- struct bch_fs *c = as->c;
struct btree *b;
- int ret;
+ unsigned interior;
+ int ret = 0;
- BUG_ON(nr_nodes > BTREE_RESERVE_MAX);
+ BUG_ON(nr_nodes[0] + nr_nodes[1] > BTREE_RESERVE_MAX);
/*
* Protects reaping from the btree node cache and using the btree node
* open bucket reserve:
*/
- ret = bch2_btree_cache_cannibalize_lock(c, cl);
+ ret = bch2_btree_cache_cannibalize_lock(trans, cl);
if (ret)
return ret;
- while (as->nr_prealloc_nodes < nr_nodes) {
- b = __bch2_btree_node_alloc(c, &as->disk_res,
- flags & BTREE_INSERT_NOWAIT
- ? NULL : cl, flags);
- if (IS_ERR(b)) {
- ret = PTR_ERR(b);
- goto err_free;
- }
+ for (interior = 0; interior < 2; interior++) {
+ struct prealloc_nodes *p = as->prealloc_nodes + interior;
- ret = bch2_mark_bkey_replicas(c, bkey_i_to_s_c(&b->key));
- if (ret)
- goto err_free;
+ while (p->nr < nr_nodes[interior]) {
+ b = __bch2_btree_node_alloc(trans, &as->disk_res, cl,
+ interior, flags);
+ if (IS_ERR(b)) {
+ ret = PTR_ERR(b);
+ goto err;
+ }
- as->prealloc_nodes[as->nr_prealloc_nodes++] = b;
+ p->b[p->nr++] = b;
+ }
}
-
- bch2_btree_cache_cannibalize_unlock(c);
- return 0;
-err_free:
- bch2_btree_cache_cannibalize_unlock(c);
- trace_btree_reserve_get_fail(c, nr_nodes, cl);
+err:
+ bch2_btree_cache_cannibalize_unlock(trans);
return ret;
}
/* Asynchronous interior node update machinery */
-static void bch2_btree_update_free(struct btree_update *as)
+static void bch2_btree_update_free(struct btree_update *as, struct btree_trans *trans)
{
struct bch_fs *c = as->c;
- bch2_journal_preres_put(&c->journal, &as->journal_preres);
+ if (as->took_gc_lock)
+ up_read(&c->gc_lock);
+ as->took_gc_lock = false;
bch2_journal_pin_drop(&c->journal, &as->journal);
bch2_journal_pin_flush(&c->journal, &as->journal);
bch2_disk_reservation_put(c, &as->disk_res);
- bch2_btree_reserve_put(as);
+ bch2_btree_reserve_put(as, trans);
+
+ bch2_time_stats_update(&c->times[BCH_TIME_btree_interior_update_total],
+ as->start_time);
mutex_lock(&c->btree_interior_update_lock);
list_del(&as->unwritten_list);
list_del(&as->list);
- mutex_unlock(&c->btree_interior_update_lock);
closure_debug_destroy(&as->cl);
mempool_free(as, &c->btree_interior_update_pool);
+ /*
+ * Have to do the wakeup with btree_interior_update_lock still held,
+ * since being on btree_interior_update_list is our ref on @c:
+ */
closure_wake_up(&c->btree_interior_update_wait);
+
+ mutex_unlock(&c->btree_interior_update_lock);
}
-static void btree_update_will_delete_key(struct btree_update *as,
- struct bkey_i *k)
+static void btree_update_add_key(struct btree_update *as,
+ struct keylist *keys, struct btree *b)
{
- BUG_ON(bch2_keylist_u64s(&as->old_keys) + k->k.u64s >
+ struct bkey_i *k = &b->key;
+
+ BUG_ON(bch2_keylist_u64s(keys) + k->k.u64s >
ARRAY_SIZE(as->_old_keys));
- bch2_keylist_add(&as->old_keys, k);
-}
-static void btree_update_will_add_key(struct btree_update *as,
- struct bkey_i *k)
-{
- BUG_ON(bch2_keylist_u64s(&as->new_keys) + k->k.u64s >
- ARRAY_SIZE(as->_new_keys));
- bch2_keylist_add(&as->new_keys, k);
+ bkey_copy(keys->top, k);
+ bkey_i_to_btree_ptr_v2(keys->top)->v.mem_ptr = b->c.level + 1;
+
+ bch2_keylist_push(keys);
}
/*
static int btree_update_nodes_written_trans(struct btree_trans *trans,
struct btree_update *as)
{
- struct bkey_i *k;
- int ret;
+ struct jset_entry *e = bch2_trans_jset_entry_alloc(trans, as->journal_u64s);
+ int ret = PTR_ERR_OR_ZERO(e);
+ if (ret)
+ return ret;
+
+ memcpy(e, as->journal_entries, as->journal_u64s * sizeof(u64));
- trans->extra_journal_entries = (void *) &as->journal_entries[0];
- trans->extra_journal_entry_u64s = as->journal_u64s;
trans->journal_pin = &as->journal;
- for_each_keylist_key(&as->new_keys, k) {
- ret = bch2_trans_mark_key(trans,
- bkey_s_c_null,
- bkey_i_to_s_c(k),
- 0, 0, BTREE_TRIGGER_INSERT);
+ for_each_keylist_key(&as->old_keys, k) {
+ unsigned level = bkey_i_to_btree_ptr_v2(k)->v.mem_ptr;
+
+ ret = bch2_key_trigger_old(trans, as->btree_id, level, bkey_i_to_s_c(k),
+ BTREE_TRIGGER_TRANSACTIONAL);
if (ret)
return ret;
}
- for_each_keylist_key(&as->old_keys, k) {
- ret = bch2_trans_mark_key(trans,
- bkey_i_to_s_c(k),
- bkey_s_c_null,
- 0, 0, BTREE_TRIGGER_OVERWRITE);
+ for_each_keylist_key(&as->new_keys, k) {
+ unsigned level = bkey_i_to_btree_ptr_v2(k)->v.mem_ptr;
+
+ ret = bch2_key_trigger_new(trans, as->btree_id, level, bkey_i_to_s(k),
+ BTREE_TRIGGER_TRANSACTIONAL);
if (ret)
return ret;
}
static void btree_update_nodes_written(struct btree_update *as)
{
struct bch_fs *c = as->c;
- struct btree *b = as->b;
- struct btree_trans trans;
+ struct btree *b;
+ struct btree_trans *trans = bch2_trans_get(c);
u64 journal_seq = 0;
unsigned i;
int ret;
if (ret)
goto err;
- BUG_ON(!journal_pin_active(&as->journal));
+ /*
+ * Wait for any in flight writes to finish before we free the old nodes
+ * on disk:
+ */
+ for (i = 0; i < as->nr_old_nodes; i++) {
+ __le64 seq;
+
+ b = as->old_nodes[i];
+
+ btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
+ seq = b->data ? b->data->keys.seq : 0;
+ six_unlock_read(&b->c.lock);
+
+ if (seq == as->old_nodes_seq[i])
+ wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight_inner,
+ TASK_UNINTERRUPTIBLE);
+ }
/*
* We did an update to a parent node where the pointers we added pointed
* journal reclaim does btree updates when flushing bkey_cached entries,
* which may require allocations as well.
*/
- bch2_trans_init(&trans, c, 0, 512);
- ret = __bch2_trans_do(&trans, &as->disk_res, &journal_seq,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_NOCHECK_RW|
- BTREE_INSERT_JOURNAL_RECLAIM|
- BTREE_INSERT_JOURNAL_RESERVED,
- btree_update_nodes_written_trans(&trans, as));
- bch2_trans_exit(&trans);
+ ret = commit_do(trans, &as->disk_res, &journal_seq,
+ BCH_WATERMARK_reclaim|
+ BCH_TRANS_COMMIT_no_enospc|
+ BCH_TRANS_COMMIT_no_check_rw|
+ BCH_TRANS_COMMIT_journal_reclaim,
+ btree_update_nodes_written_trans(trans, as));
+ bch2_trans_unlock(trans);
bch2_fs_fatal_err_on(ret && !bch2_journal_error(&c->journal), c,
- "error %i in btree_update_nodes_written()", ret);
+ "%s(): error %s", __func__, bch2_err_str(ret));
err:
- if (b) {
+ if (as->b) {
+
+ b = as->b;
+ btree_path_idx_t path_idx = get_unlocked_mut_path(trans,
+ as->btree_id, b->c.level, b->key.k.p);
+ struct btree_path *path = trans->paths + path_idx;
/*
* @b is the node we did the final insert into:
*
* we're in journal error state:
*/
- btree_node_lock_type(c, b, SIX_LOCK_intent);
- btree_node_lock_type(c, b, SIX_LOCK_write);
+ /*
+ * Ensure transaction is unlocked before using
+ * btree_node_lock_nopath() (the use of which is always suspect,
+ * we need to work on removing this in the future)
+ *
+ * It should be, but get_unlocked_mut_path() -> bch2_path_get()
+ * calls bch2_path_upgrade(), before we call path_make_mut(), so
+ * we may rarely end up with a locked path besides the one we
+ * have here:
+ */
+ bch2_trans_unlock(trans);
+ btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
+ mark_btree_node_locked(trans, path, b->c.level, BTREE_NODE_INTENT_LOCKED);
+ path->l[b->c.level].lock_seq = six_lock_seq(&b->c.lock);
+ path->l[b->c.level].b = b;
+
+ bch2_btree_node_lock_write_nofail(trans, path, &b->c);
+
mutex_lock(&c->btree_interior_update_lock);
list_del(&as->write_blocked_list);
+ if (list_empty(&b->write_blocked))
+ clear_btree_node_write_blocked(b);
/*
* Node might have been freed, recheck under
* btree_interior_update_lock:
*/
if (as->b == b) {
- struct bset *i = btree_bset_last(b);
-
BUG_ON(!b->c.level);
BUG_ON(!btree_node_dirty(b));
if (!ret) {
- i->journal_seq = cpu_to_le64(
- max(journal_seq,
- le64_to_cpu(i->journal_seq)));
+ struct bset *last = btree_bset_last(b);
+
+ last->journal_seq = cpu_to_le64(
+ max(journal_seq,
+ le64_to_cpu(last->journal_seq)));
bch2_btree_add_journal_pin(c, b, journal_seq);
} else {
}
mutex_unlock(&c->btree_interior_update_lock);
+
+ mark_btree_node_locked_noreset(path, b->c.level, BTREE_NODE_INTENT_LOCKED);
six_unlock_write(&b->c.lock);
btree_node_write_if_need(c, b, SIX_LOCK_intent);
- six_unlock_intent(&b->c.lock);
+ btree_node_unlock(trans, path, b->c.level);
+ bch2_path_put(trans, path_idx, true);
}
bch2_journal_pin_drop(&c->journal, &as->journal);
- bch2_journal_preres_put(&c->journal, &as->journal_preres);
-
mutex_lock(&c->btree_interior_update_lock);
for (i = 0; i < as->nr_new_nodes; i++) {
b = as->new_nodes[i];
BUG_ON(b->will_make_reachable != (unsigned long) as);
b->will_make_reachable = 0;
+ clear_btree_node_will_make_reachable(b);
}
mutex_unlock(&c->btree_interior_update_lock);
for (i = 0; i < as->nr_new_nodes; i++) {
b = as->new_nodes[i];
- btree_node_lock_type(c, b, SIX_LOCK_read);
+ btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
btree_node_write_if_need(c, b, SIX_LOCK_read);
six_unlock_read(&b->c.lock);
}
for (i = 0; i < as->nr_open_buckets; i++)
bch2_open_bucket_put(c, c->open_buckets + as->open_buckets[i]);
- bch2_btree_update_free(as);
+ bch2_btree_update_free(as, trans);
+ bch2_trans_put(trans);
}
static void btree_interior_update_work(struct work_struct *work)
}
}
-static void btree_update_set_nodes_written(struct closure *cl)
+static CLOSURE_CALLBACK(btree_update_set_nodes_written)
{
- struct btree_update *as = container_of(cl, struct btree_update, cl);
+ closure_type(as, struct btree_update, cl);
struct bch_fs *c = as->c;
mutex_lock(&c->btree_interior_update_lock);
BUG_ON(as->mode != BTREE_INTERIOR_NO_UPDATE);
BUG_ON(!btree_node_dirty(b));
+ BUG_ON(!b->c.level);
as->mode = BTREE_INTERIOR_UPDATING_NODE;
as->b = b;
+
+ set_btree_node_write_blocked(b);
list_add(&as->write_blocked_list, &b->write_blocked);
mutex_unlock(&c->btree_interior_update_lock);
}
+static int bch2_update_reparent_journal_pin_flush(struct journal *j,
+ struct journal_entry_pin *_pin, u64 seq)
+{
+ return 0;
+}
+
static void btree_update_reparent(struct btree_update *as,
struct btree_update *child)
{
child->b = NULL;
child->mode = BTREE_INTERIOR_UPDATING_AS;
- bch2_journal_pin_copy(&c->journal, &as->journal, &child->journal, NULL);
+ bch2_journal_pin_copy(&c->journal, &as->journal, &child->journal,
+ bch2_update_reparent_journal_pin_flush);
}
static void btree_update_updated_root(struct btree_update *as, struct btree *b)
* And it adds @b to the list of @as's new nodes, so that we can update sector
* counts in bch2_btree_update_nodes_written:
*/
-void bch2_btree_update_add_new_node(struct btree_update *as, struct btree *b)
+static void bch2_btree_update_add_new_node(struct btree_update *as, struct btree *b)
{
struct bch_fs *c = as->c;
as->new_nodes[as->nr_new_nodes++] = b;
b->will_make_reachable = 1UL|(unsigned long) as;
+ set_btree_node_will_make_reachable(b);
mutex_unlock(&c->btree_interior_update_lock);
- btree_update_will_add_key(as, &b->key);
+ btree_update_add_key(as, &as->new_keys, b);
+
+ if (b->key.k.type == KEY_TYPE_btree_ptr_v2) {
+ unsigned bytes = vstruct_end(&b->data->keys) - (void *) b->data;
+ unsigned sectors = round_up(bytes, block_bytes(c)) >> 9;
+
+ bkey_i_to_btree_ptr_v2(&b->key)->v.sectors_written =
+ cpu_to_le16(sectors);
+ }
}
/*
* xchg() is for synchronization with bch2_btree_complete_write:
*/
v = xchg(&b->will_make_reachable, 0);
+ clear_btree_node_will_make_reachable(b);
as = (struct btree_update *) (v & ~1UL);
if (!as) {
closure_put(&as->cl);
}
-void bch2_btree_update_get_open_buckets(struct btree_update *as, struct btree *b)
+static void bch2_btree_update_get_open_buckets(struct btree_update *as, struct btree *b)
{
while (b->ob.nr)
as->open_buckets[as->nr_open_buckets++] =
b->ob.v[--b->ob.nr];
}
+static int bch2_btree_update_will_free_node_journal_pin_flush(struct journal *j,
+ struct journal_entry_pin *_pin, u64 seq)
+{
+ return 0;
+}
+
/*
* @b is being split/rewritten: it may have pointers to not-yet-written btree
* nodes and thus outstanding btree_updates - redirect @b's
* btree_updates to point to this btree_update:
*/
-void bch2_btree_interior_update_will_free_node(struct btree_update *as,
- struct btree *b)
+static void bch2_btree_interior_update_will_free_node(struct btree_update *as,
+ struct btree *b)
{
struct bch_fs *c = as->c;
struct btree_update *p, *n;
closure_wake_up(&c->btree_interior_update_wait);
}
- clear_btree_node_dirty(c, b);
+ clear_btree_node_dirty_acct(c, b);
clear_btree_node_need_write(b);
+ clear_btree_node_write_blocked(b);
/*
* Does this node have unwritten data that has a pin on the journal?
* when the new nodes are persistent and reachable on disk:
*/
w = btree_current_write(b);
- bch2_journal_pin_copy(&c->journal, &as->journal, &w->journal, NULL);
+ bch2_journal_pin_copy(&c->journal, &as->journal, &w->journal,
+ bch2_btree_update_will_free_node_journal_pin_flush);
bch2_journal_pin_drop(&c->journal, &w->journal);
w = btree_prev_write(b);
- bch2_journal_pin_copy(&c->journal, &as->journal, &w->journal, NULL);
+ bch2_journal_pin_copy(&c->journal, &as->journal, &w->journal,
+ bch2_btree_update_will_free_node_journal_pin_flush);
bch2_journal_pin_drop(&c->journal, &w->journal);
mutex_unlock(&c->btree_interior_update_lock);
*/
btree_update_drop_new_node(c, b);
- btree_update_will_delete_key(as, &b->key);
+ btree_update_add_key(as, &as->old_keys, b);
+
+ as->old_nodes[as->nr_old_nodes] = b;
+ as->old_nodes_seq[as->nr_old_nodes] = b->data->keys.seq;
+ as->nr_old_nodes++;
}
-void bch2_btree_update_done(struct btree_update *as)
+static void bch2_btree_update_done(struct btree_update *as, struct btree_trans *trans)
{
+ struct bch_fs *c = as->c;
+ u64 start_time = as->start_time;
+
BUG_ON(as->mode == BTREE_INTERIOR_NO_UPDATE);
- bch2_btree_reserve_put(as);
+ if (as->took_gc_lock)
+ up_read(&as->c->gc_lock);
+ as->took_gc_lock = false;
- continue_at(&as->cl, btree_update_set_nodes_written, system_freezable_wq);
+ bch2_btree_reserve_put(as, trans);
+
+ continue_at(&as->cl, btree_update_set_nodes_written,
+ as->c->btree_interior_update_worker);
+
+ bch2_time_stats_update(&c->times[BCH_TIME_btree_interior_update_foreground],
+ start_time);
}
-struct btree_update *
-bch2_btree_update_start(struct btree_trans *trans, enum btree_id id,
- unsigned nr_nodes, unsigned flags,
- struct closure *cl)
+static struct btree_update *
+bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
+ unsigned level, bool split, unsigned flags)
{
struct bch_fs *c = trans->c;
struct btree_update *as;
- int disk_res_flags = (flags & BTREE_INSERT_NOFAIL)
+ u64 start_time = local_clock();
+ int disk_res_flags = (flags & BCH_TRANS_COMMIT_no_enospc)
? BCH_DISK_RESERVATION_NOFAIL : 0;
- int journal_flags = (flags & BTREE_INSERT_JOURNAL_RESERVED)
- ? JOURNAL_RES_GET_RECLAIM : 0;
+ unsigned nr_nodes[2] = { 0, 0 };
+ unsigned update_level = level;
+ enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
int ret = 0;
+ u32 restart_count = trans->restart_count;
- /*
- * This check isn't necessary for correctness - it's just to potentially
- * prevent us from doing a lot of work that'll end up being wasted:
- */
- ret = bch2_journal_error(&c->journal);
- if (ret)
- return ERR_PTR(ret);
+ BUG_ON(!path->should_be_locked);
+
+ if (watermark == BCH_WATERMARK_copygc)
+ watermark = BCH_WATERMARK_btree_copygc;
+ if (watermark < BCH_WATERMARK_btree)
+ watermark = BCH_WATERMARK_btree;
+
+ flags &= ~BCH_WATERMARK_MASK;
+ flags |= watermark;
+
+ if (!(flags & BCH_TRANS_COMMIT_journal_reclaim) &&
+ watermark < c->journal.watermark) {
+ struct journal_res res = { 0 };
+
+ ret = drop_locks_do(trans,
+ bch2_journal_res_get(&c->journal, &res, 1,
+ watermark|JOURNAL_RES_GET_CHECK));
+ if (ret)
+ return ERR_PTR(ret);
+ }
+
+ while (1) {
+ nr_nodes[!!update_level] += 1 + split;
+ update_level++;
+
+ ret = bch2_btree_path_upgrade(trans, path, update_level + 1);
+ if (ret)
+ return ERR_PTR(ret);
+
+ if (!btree_path_node(path, update_level)) {
+ /* Allocating new root? */
+ nr_nodes[1] += split;
+ update_level = BTREE_MAX_DEPTH;
+ break;
+ }
+
+ /*
+ * Always check for space for two keys, even if we won't have to
+ * split at prior level - it might have been a merge instead:
+ */
+ if (bch2_btree_node_insert_fits(c, path->l[update_level].b,
+ BKEY_BTREE_PTR_U64s_MAX * 2))
+ break;
+
+ split = path->l[update_level].b->nr.live_u64s > BTREE_SPLIT_THRESHOLD(c);
+ }
+
+ if (!down_read_trylock(&c->gc_lock)) {
+ ret = drop_locks_do(trans, (down_read(&c->gc_lock), 0));
+ if (ret) {
+ up_read(&c->gc_lock);
+ return ERR_PTR(ret);
+ }
+ }
- as = mempool_alloc(&c->btree_interior_update_pool, GFP_NOIO);
+ as = mempool_alloc(&c->btree_interior_update_pool, GFP_NOFS);
memset(as, 0, sizeof(*as));
closure_init(&as->cl, NULL);
as->c = c;
+ as->start_time = start_time;
as->mode = BTREE_INTERIOR_NO_UPDATE;
- as->btree_id = id;
+ as->took_gc_lock = true;
+ as->btree_id = path->btree_id;
+ as->update_level = update_level;
INIT_LIST_HEAD(&as->list);
INIT_LIST_HEAD(&as->unwritten_list);
INIT_LIST_HEAD(&as->write_blocked_list);
bch2_keylist_init(&as->new_keys, as->_new_keys);
bch2_keylist_init(&as->parent_keys, as->inline_keys);
- ret = bch2_journal_preres_get(&c->journal, &as->journal_preres,
- BTREE_UPDATE_JOURNAL_RES,
- journal_flags|JOURNAL_RES_GET_NONBLOCK);
- if (ret == -EAGAIN) {
- if (flags & BTREE_INSERT_NOUNLOCK)
- return ERR_PTR(-EINTR);
+ mutex_lock(&c->btree_interior_update_lock);
+ list_add_tail(&as->list, &c->btree_interior_update_list);
+ mutex_unlock(&c->btree_interior_update_lock);
- bch2_trans_unlock(trans);
+ /*
+ * We don't want to allocate if we're in an error state, that can cause
+ * deadlock on emergency shutdown due to open buckets getting stuck in
+ * the btree_reserve_cache after allocator shutdown has cleared it out.
+ * This check needs to come after adding us to the btree_interior_update
+ * list but before calling bch2_btree_reserve_get, to synchronize with
+ * __bch2_fs_read_only().
+ */
+ ret = bch2_journal_error(&c->journal);
+ if (ret)
+ goto err;
- ret = bch2_journal_preres_get(&c->journal, &as->journal_preres,
- BTREE_UPDATE_JOURNAL_RES,
- journal_flags);
- if (ret)
- return ERR_PTR(ret);
+ ret = bch2_disk_reservation_get(c, &as->disk_res,
+ (nr_nodes[0] + nr_nodes[1]) * btree_sectors(c),
+ c->opts.metadata_replicas,
+ disk_res_flags);
+ if (ret)
+ goto err;
+
+ ret = bch2_btree_reserve_get(trans, as, nr_nodes, flags, NULL);
+ if (bch2_err_matches(ret, ENOSPC) ||
+ bch2_err_matches(ret, ENOMEM)) {
+ struct closure cl;
- if (!bch2_trans_relock(trans)) {
- ret = -EINTR;
+ /*
+ * XXX: this should probably be a separate BTREE_INSERT_NONBLOCK
+ * flag
+ */
+ if (bch2_err_matches(ret, ENOSPC) &&
+ (flags & BCH_TRANS_COMMIT_journal_reclaim) &&
+ watermark != BCH_WATERMARK_reclaim) {
+ ret = -BCH_ERR_journal_reclaim_would_deadlock;
goto err;
}
+
+ closure_init_stack(&cl);
+
+ do {
+ ret = bch2_btree_reserve_get(trans, as, nr_nodes, flags, &cl);
+
+ bch2_trans_unlock(trans);
+ closure_sync(&cl);
+ } while (bch2_err_matches(ret, BCH_ERR_operation_blocked));
}
- ret = bch2_disk_reservation_get(c, &as->disk_res,
- nr_nodes * c->opts.btree_node_size,
- c->opts.metadata_replicas,
- disk_res_flags);
- if (ret)
+ if (ret) {
+ trace_and_count(c, btree_reserve_get_fail, trans->fn,
+ _RET_IP_, nr_nodes[0] + nr_nodes[1], ret);
goto err;
+ }
- ret = bch2_btree_reserve_get(as, nr_nodes, flags, cl);
+ ret = bch2_trans_relock(trans);
if (ret)
goto err;
- bch2_journal_pin_add(&c->journal,
- atomic64_read(&c->journal.seq),
- &as->journal, NULL);
-
- mutex_lock(&c->btree_interior_update_lock);
- list_add_tail(&as->list, &c->btree_interior_update_list);
- mutex_unlock(&c->btree_interior_update_lock);
-
+ bch2_trans_verify_not_restarted(trans, restart_count);
return as;
err:
- bch2_btree_update_free(as);
+ bch2_btree_update_free(as, trans);
+ if (!bch2_err_matches(ret, ENOSPC) &&
+ !bch2_err_matches(ret, EROFS))
+ bch_err_fn_ratelimited(c, ret);
return ERR_PTR(ret);
}
(b->c.level < btree_node_root(c, b)->c.level ||
!btree_node_dying(btree_node_root(c, b))));
- btree_node_root(c, b) = b;
+ bch2_btree_id_root(c, b->c.btree_id)->b = b;
mutex_unlock(&c->btree_root_lock);
bch2_recalc_btree_reserve(c);
}
-/**
- * bch_btree_set_root - update the root in memory and on disk
- *
- * To ensure forward progress, the current task must not be holding any
- * btree node write locks. However, you must hold an intent lock on the
- * old root.
- *
- * Note: This allocates a journal entry but doesn't add any keys to
- * it. All the btree roots are part of every journal write, so there
- * is nothing new to be done. This just guarantees that there is a
- * journal write.
- */
-static void bch2_btree_set_root(struct btree_update *as, struct btree *b,
- struct btree_iter *iter)
+static void bch2_btree_set_root(struct btree_update *as,
+ struct btree_trans *trans,
+ struct btree_path *path,
+ struct btree *b)
{
struct bch_fs *c = as->c;
struct btree *old;
- trace_btree_set_root(c, b);
- BUG_ON(!b->written &&
- !test_bit(BCH_FS_HOLD_BTREE_WRITES, &c->flags));
+ trace_and_count(c, btree_node_set_root, trans, b);
old = btree_node_root(c, b);
* Ensure no one is using the old root while we switch to the
* new root:
*/
- bch2_btree_node_lock_write(old, iter);
+ bch2_btree_node_lock_write_nofail(trans, path, &old->c);
bch2_btree_set_root_inmem(c, b);
* an intent lock on the new root, and any updates that would
* depend on the new root would have to update the new root.
*/
- bch2_btree_node_unlock_write(old, iter);
+ bch2_btree_node_unlock_write(trans, path, old);
}
/* Interior node updates: */
-static void bch2_insert_fixup_btree_ptr(struct btree_update *as, struct btree *b,
- struct btree_iter *iter,
- struct bkey_i *insert,
- struct btree_node_iter *node_iter)
+static void bch2_insert_fixup_btree_ptr(struct btree_update *as,
+ struct btree_trans *trans,
+ struct btree_path *path,
+ struct btree *b,
+ struct btree_node_iter *node_iter,
+ struct bkey_i *insert)
{
struct bch_fs *c = as->c;
struct bkey_packed *k;
- const char *invalid;
-
- invalid = bch2_bkey_invalid(c, bkey_i_to_s_c(insert), btree_node_type(b)) ?:
- bch2_bkey_in_btree_node(b, bkey_i_to_s_c(insert));
- if (invalid) {
- char buf[160];
-
- bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(insert));
- bch2_fs_inconsistent(c, "inserting invalid bkey %s: %s", buf, invalid);
+ struct printbuf buf = PRINTBUF;
+ unsigned long old, new, v;
+
+ BUG_ON(insert->k.type == KEY_TYPE_btree_ptr_v2 &&
+ !btree_ptr_sectors_written(insert));
+
+ if (unlikely(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)))
+ bch2_journal_key_overwritten(c, b->c.btree_id, b->c.level, insert->k.p);
+
+ if (bch2_bkey_invalid(c, bkey_i_to_s_c(insert),
+ btree_node_type(b), WRITE, &buf) ?:
+ bch2_bkey_in_btree_node(c, b, bkey_i_to_s_c(insert), &buf)) {
+ printbuf_reset(&buf);
+ prt_printf(&buf, "inserting invalid bkey\n ");
+ bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(insert));
+ prt_printf(&buf, "\n ");
+ bch2_bkey_invalid(c, bkey_i_to_s_c(insert),
+ btree_node_type(b), WRITE, &buf);
+ bch2_bkey_in_btree_node(c, b, bkey_i_to_s_c(insert), &buf);
+
+ bch2_fs_inconsistent(c, "%s", buf.buf);
dump_stack();
}
bkey_iter_pos_cmp(b, k, &insert->k.p) < 0)
bch2_btree_node_iter_advance(node_iter, b);
- bch2_btree_bset_insert_key(iter, b, node_iter, insert);
- set_btree_node_dirty(c, b);
- set_btree_node_need_write(b);
+ bch2_btree_bset_insert_key(trans, path, b, node_iter, insert);
+ set_btree_node_dirty_acct(c, b);
+
+ v = READ_ONCE(b->flags);
+ do {
+ old = new = v;
+
+ new &= ~BTREE_WRITE_TYPE_MASK;
+ new |= BTREE_WRITE_interior;
+ new |= 1 << BTREE_NODE_need_write;
+ } while ((v = cmpxchg(&b->flags, old, new)) != old);
+
+ printbuf_exit(&buf);
+}
+
+static void
+__bch2_btree_insert_keys_interior(struct btree_update *as,
+ struct btree_trans *trans,
+ struct btree_path *path,
+ struct btree *b,
+ struct btree_node_iter node_iter,
+ struct keylist *keys)
+{
+ struct bkey_i *insert = bch2_keylist_front(keys);
+ struct bkey_packed *k;
+
+ BUG_ON(btree_node_type(b) != BKEY_TYPE_btree);
+
+ while ((k = bch2_btree_node_iter_prev_all(&node_iter, b)) &&
+ (bkey_cmp_left_packed(b, k, &insert->k.p) >= 0))
+ ;
+
+ while (!bch2_keylist_empty(keys)) {
+ insert = bch2_keylist_front(keys);
+
+ if (bpos_gt(insert->k.p, b->key.k.p))
+ break;
+
+ bch2_insert_fixup_btree_ptr(as, trans, path, b, &node_iter, insert);
+ bch2_keylist_pop_front(keys);
+ }
}
/*
* Move keys from n1 (original replacement node, now lower node) to n2 (higher
* node)
*/
-static struct btree *__btree_split_node(struct btree_update *as,
- struct btree *n1,
- struct btree_iter *iter)
+static void __btree_split_node(struct btree_update *as,
+ struct btree_trans *trans,
+ struct btree *b,
+ struct btree *n[2])
{
- size_t nr_packed = 0, nr_unpacked = 0;
- struct btree *n2;
- struct bset *set1, *set2;
- struct bkey_packed *k, *prev = NULL;
-
- n2 = bch2_btree_node_alloc(as, n1->c.level);
- bch2_btree_update_add_new_node(as, n2);
+ struct bkey_packed *k;
+ struct bpos n1_pos = POS_MIN;
+ struct btree_node_iter iter;
+ struct bset *bsets[2];
+ struct bkey_format_state format[2];
+ struct bkey_packed *out[2];
+ struct bkey uk;
+ unsigned u64s, n1_u64s = (b->nr.live_u64s * 3) / 5;
+ struct { unsigned nr_keys, val_u64s; } nr_keys[2];
+ int i;
- n2->data->max_key = n1->data->max_key;
- n2->data->format = n1->format;
- SET_BTREE_NODE_SEQ(n2->data, BTREE_NODE_SEQ(n1->data));
- n2->key.k.p = n1->key.k.p;
+ memset(&nr_keys, 0, sizeof(nr_keys));
- btree_node_set_format(n2, n2->data->format);
+ for (i = 0; i < 2; i++) {
+ BUG_ON(n[i]->nsets != 1);
- set1 = btree_bset_first(n1);
- set2 = btree_bset_first(n2);
+ bsets[i] = btree_bset_first(n[i]);
+ out[i] = bsets[i]->start;
- /*
- * Has to be a linear search because we don't have an auxiliary
- * search tree yet
- */
- k = set1->start;
- while (1) {
- struct bkey_packed *n = bkey_next_skip_noops(k, vstruct_last(set1));
+ SET_BTREE_NODE_SEQ(n[i]->data, BTREE_NODE_SEQ(b->data) + 1);
+ bch2_bkey_format_init(&format[i]);
+ }
- if (n == vstruct_last(set1))
- break;
- if (k->_data - set1->_data >= (le16_to_cpu(set1->u64s) * 3) / 5)
- break;
+ u64s = 0;
+ for_each_btree_node_key(b, k, &iter) {
+ if (bkey_deleted(k))
+ continue;
- if (bkey_packed(k))
- nr_packed++;
- else
- nr_unpacked++;
+ i = u64s >= n1_u64s;
+ u64s += k->u64s;
+ uk = bkey_unpack_key(b, k);
+ if (!i)
+ n1_pos = uk.p;
+ bch2_bkey_format_add_key(&format[i], &uk);
- prev = k;
- k = n;
+ nr_keys[i].nr_keys++;
+ nr_keys[i].val_u64s += bkeyp_val_u64s(&b->format, k);
}
- BUG_ON(!prev);
+ btree_set_min(n[0], b->data->min_key);
+ btree_set_max(n[0], n1_pos);
+ btree_set_min(n[1], bpos_successor(n1_pos));
+ btree_set_max(n[1], b->data->max_key);
- btree_set_max(n1, bkey_unpack_pos(n1, prev));
- btree_set_min(n2, bkey_successor(n1->key.k.p));
+ for (i = 0; i < 2; i++) {
+ bch2_bkey_format_add_pos(&format[i], n[i]->data->min_key);
+ bch2_bkey_format_add_pos(&format[i], n[i]->data->max_key);
- set2->u64s = cpu_to_le16((u64 *) vstruct_end(set1) - (u64 *) k);
- set1->u64s = cpu_to_le16(le16_to_cpu(set1->u64s) - le16_to_cpu(set2->u64s));
+ n[i]->data->format = bch2_bkey_format_done(&format[i]);
- set_btree_bset_end(n1, n1->set);
- set_btree_bset_end(n2, n2->set);
+ unsigned u64s = nr_keys[i].nr_keys * n[i]->data->format.key_u64s +
+ nr_keys[i].val_u64s;
+ if (__vstruct_bytes(struct btree_node, u64s) > btree_bytes(as->c))
+ n[i]->data->format = b->format;
- n2->nr.live_u64s = le16_to_cpu(set2->u64s);
- n2->nr.bset_u64s[0] = le16_to_cpu(set2->u64s);
- n2->nr.packed_keys = n1->nr.packed_keys - nr_packed;
- n2->nr.unpacked_keys = n1->nr.unpacked_keys - nr_unpacked;
+ btree_node_set_format(n[i], n[i]->data->format);
+ }
- n1->nr.live_u64s = le16_to_cpu(set1->u64s);
- n1->nr.bset_u64s[0] = le16_to_cpu(set1->u64s);
- n1->nr.packed_keys = nr_packed;
- n1->nr.unpacked_keys = nr_unpacked;
+ u64s = 0;
+ for_each_btree_node_key(b, k, &iter) {
+ if (bkey_deleted(k))
+ continue;
- BUG_ON(!set1->u64s);
- BUG_ON(!set2->u64s);
+ i = u64s >= n1_u64s;
+ u64s += k->u64s;
- memcpy_u64s(set2->start,
- vstruct_end(set1),
- le16_to_cpu(set2->u64s));
+ if (bch2_bkey_transform(&n[i]->format, out[i], bkey_packed(k)
+ ? &b->format: &bch2_bkey_format_current, k))
+ out[i]->format = KEY_FORMAT_LOCAL_BTREE;
+ else
+ bch2_bkey_unpack(b, (void *) out[i], k);
- btree_node_reset_sib_u64s(n1);
- btree_node_reset_sib_u64s(n2);
+ out[i]->needs_whiteout = false;
- bch2_verify_btree_nr_keys(n1);
- bch2_verify_btree_nr_keys(n2);
-
- if (n1->c.level) {
- btree_node_interior_verify(as->c, n1);
- btree_node_interior_verify(as->c, n2);
+ btree_keys_account_key_add(&n[i]->nr, 0, out[i]);
+ out[i] = bkey_p_next(out[i]);
}
- return n2;
+ for (i = 0; i < 2; i++) {
+ bsets[i]->u64s = cpu_to_le16((u64 *) out[i] - bsets[i]->_data);
+
+ BUG_ON(!bsets[i]->u64s);
+
+ set_btree_bset_end(n[i], n[i]->set);
+
+ btree_node_reset_sib_u64s(n[i]);
+
+ bch2_verify_btree_nr_keys(n[i]);
+
+ if (b->c.level)
+ btree_node_interior_verify(as->c, n[i]);
+ }
}
/*
* nodes that were coalesced, and thus in the middle of a child node post
* coalescing:
*/
-static void btree_split_insert_keys(struct btree_update *as, struct btree *b,
- struct btree_iter *iter,
+static void btree_split_insert_keys(struct btree_update *as,
+ struct btree_trans *trans,
+ btree_path_idx_t path_idx,
+ struct btree *b,
struct keylist *keys)
{
- struct btree_node_iter node_iter;
- struct bkey_i *k = bch2_keylist_front(keys);
- struct bkey_packed *src, *dst, *n;
- struct bset *i;
+ struct btree_path *path = trans->paths + path_idx;
- BUG_ON(btree_node_type(b) != BKEY_TYPE_btree);
+ if (!bch2_keylist_empty(keys) &&
+ bpos_le(bch2_keylist_front(keys)->k.p, b->data->max_key)) {
+ struct btree_node_iter node_iter;
- bch2_btree_node_iter_init(&node_iter, b, &k->k.p);
+ bch2_btree_node_iter_init(&node_iter, b, &bch2_keylist_front(keys)->k.p);
- while (!bch2_keylist_empty(keys)) {
- k = bch2_keylist_front(keys);
-
- bch2_insert_fixup_btree_ptr(as, b, iter, k, &node_iter);
- bch2_keylist_pop_front(keys);
- }
+ __bch2_btree_insert_keys_interior(as, trans, path, b, node_iter, keys);
- /*
- * We can't tolerate whiteouts here - with whiteouts there can be
- * duplicate keys, and it would be rather bad if we picked a duplicate
- * for the pivot:
- */
- i = btree_bset_first(b);
- src = dst = i->start;
- while (src != vstruct_last(i)) {
- n = bkey_next_skip_noops(src, vstruct_last(i));
- if (!bkey_deleted(src)) {
- memmove_u64s_down(dst, src, src->u64s);
- dst = bkey_next(dst);
- }
- src = n;
+ btree_node_interior_verify(as->c, b);
}
-
- /* Also clear out the unwritten whiteouts area: */
- b->whiteout_u64s = 0;
-
- i->u64s = cpu_to_le16((u64 *) dst - i->_data);
- set_btree_bset_end(b, b->set);
-
- BUG_ON(b->nsets != 1 ||
- b->nr.live_u64s != le16_to_cpu(btree_bset_first(b)->u64s));
-
- btree_node_interior_verify(as->c, b);
}
-static void btree_split(struct btree_update *as, struct btree *b,
- struct btree_iter *iter, struct keylist *keys,
- unsigned flags)
+static int btree_split(struct btree_update *as, struct btree_trans *trans,
+ btree_path_idx_t path, struct btree *b,
+ struct keylist *keys, unsigned flags)
{
struct bch_fs *c = as->c;
- struct btree *parent = btree_node_parent(iter, b);
+ struct btree *parent = btree_node_parent(trans->paths + path, b);
struct btree *n1, *n2 = NULL, *n3 = NULL;
+ btree_path_idx_t path1 = 0, path2 = 0;
u64 start_time = local_clock();
+ int ret = 0;
BUG_ON(!parent && (b != btree_node_root(c, b)));
- BUG_ON(!btree_node_intent_locked(iter, btree_node_root(c, b)->c.level));
+ BUG_ON(parent && !btree_node_intent_locked(trans->paths + path, b->c.level + 1));
bch2_btree_interior_update_will_free_node(as, b);
- n1 = bch2_btree_node_alloc_replacement(as, b);
- bch2_btree_update_add_new_node(as, n1);
+ if (b->nr.live_u64s > BTREE_SPLIT_THRESHOLD(c)) {
+ struct btree *n[2];
- if (keys)
- btree_split_insert_keys(as, n1, iter, keys);
+ trace_and_count(c, btree_node_split, trans, b);
- if (bset_u64s(&n1->set[0]) > BTREE_SPLIT_THRESHOLD(c)) {
- trace_btree_split(c, b);
+ n[0] = n1 = bch2_btree_node_alloc(as, trans, b->c.level);
+ n[1] = n2 = bch2_btree_node_alloc(as, trans, b->c.level);
- n2 = __btree_split_node(as, n1, iter);
+ __btree_split_node(as, trans, b, n);
+
+ if (keys) {
+ btree_split_insert_keys(as, trans, path, n1, keys);
+ btree_split_insert_keys(as, trans, path, n2, keys);
+ BUG_ON(!bch2_keylist_empty(keys));
+ }
bch2_btree_build_aux_trees(n2);
bch2_btree_build_aux_trees(n1);
+
+ bch2_btree_update_add_new_node(as, n1);
+ bch2_btree_update_add_new_node(as, n2);
six_unlock_write(&n2->c.lock);
six_unlock_write(&n1->c.lock);
- bch2_btree_node_write(c, n2, SIX_LOCK_intent);
+ path1 = get_unlocked_mut_path(trans, as->btree_id, n1->c.level, n1->key.k.p);
+ six_lock_increment(&n1->c.lock, SIX_LOCK_intent);
+ mark_btree_node_locked(trans, trans->paths + path1, n1->c.level, BTREE_NODE_INTENT_LOCKED);
+ bch2_btree_path_level_init(trans, trans->paths + path1, n1);
+
+ path2 = get_unlocked_mut_path(trans, as->btree_id, n2->c.level, n2->key.k.p);
+ six_lock_increment(&n2->c.lock, SIX_LOCK_intent);
+ mark_btree_node_locked(trans, trans->paths + path2, n2->c.level, BTREE_NODE_INTENT_LOCKED);
+ bch2_btree_path_level_init(trans, trans->paths + path2, n2);
/*
* Note that on recursive parent_keys == keys, so we
if (!parent) {
/* Depth increases, make a new root */
- n3 = __btree_root_alloc(as, b->c.level + 1);
+ n3 = __btree_root_alloc(as, trans, b->c.level + 1);
+
+ bch2_btree_update_add_new_node(as, n3);
+ six_unlock_write(&n3->c.lock);
+
+ trans->paths[path2].locks_want++;
+ BUG_ON(btree_node_locked(trans->paths + path2, n3->c.level));
+ six_lock_increment(&n3->c.lock, SIX_LOCK_intent);
+ mark_btree_node_locked(trans, trans->paths + path2, n3->c.level, BTREE_NODE_INTENT_LOCKED);
+ bch2_btree_path_level_init(trans, trans->paths + path2, n3);
n3->sib_u64s[0] = U16_MAX;
n3->sib_u64s[1] = U16_MAX;
- btree_split_insert_keys(as, n3, iter, &as->parent_keys);
-
- bch2_btree_node_write(c, n3, SIX_LOCK_intent);
+ btree_split_insert_keys(as, trans, path, n3, &as->parent_keys);
}
} else {
- trace_btree_compact(c, b);
+ trace_and_count(c, btree_node_compact, trans, b);
+
+ n1 = bch2_btree_node_alloc_replacement(as, trans, b);
+
+ if (keys) {
+ btree_split_insert_keys(as, trans, path, n1, keys);
+ BUG_ON(!bch2_keylist_empty(keys));
+ }
bch2_btree_build_aux_trees(n1);
+ bch2_btree_update_add_new_node(as, n1);
six_unlock_write(&n1->c.lock);
+ path1 = get_unlocked_mut_path(trans, as->btree_id, n1->c.level, n1->key.k.p);
+ six_lock_increment(&n1->c.lock, SIX_LOCK_intent);
+ mark_btree_node_locked(trans, trans->paths + path1, n1->c.level, BTREE_NODE_INTENT_LOCKED);
+ bch2_btree_path_level_init(trans, trans->paths + path1, n1);
+
if (parent)
bch2_keylist_add(&as->parent_keys, &n1->key);
}
- bch2_btree_node_write(c, n1, SIX_LOCK_intent);
-
/* New nodes all written, now make them visible: */
if (parent) {
/* Split a non root node */
- bch2_btree_insert_node(as, parent, iter, &as->parent_keys, flags);
+ ret = bch2_btree_insert_node(as, trans, path, parent, &as->parent_keys, flags);
+ if (ret)
+ goto err;
} else if (n3) {
- bch2_btree_set_root(as, n3, iter);
+ bch2_btree_set_root(as, trans, trans->paths + path, n3);
} else {
/* Root filled up but didn't need to be split */
- bch2_btree_set_root(as, n1, iter);
+ bch2_btree_set_root(as, trans, trans->paths + path, n1);
}
- bch2_btree_update_get_open_buckets(as, n1);
- if (n2)
- bch2_btree_update_get_open_buckets(as, n2);
- if (n3)
+ if (n3) {
bch2_btree_update_get_open_buckets(as, n3);
-
- /* Successful split, update the iterator to point to the new nodes: */
-
- six_lock_increment(&b->c.lock, SIX_LOCK_intent);
- bch2_btree_iter_node_drop(iter, b);
- if (n3)
- bch2_btree_iter_node_replace(iter, n3);
- if (n2)
- bch2_btree_iter_node_replace(iter, n2);
- bch2_btree_iter_node_replace(iter, n1);
+ bch2_btree_node_write(c, n3, SIX_LOCK_intent, 0);
+ }
+ if (n2) {
+ bch2_btree_update_get_open_buckets(as, n2);
+ bch2_btree_node_write(c, n2, SIX_LOCK_intent, 0);
+ }
+ bch2_btree_update_get_open_buckets(as, n1);
+ bch2_btree_node_write(c, n1, SIX_LOCK_intent, 0);
/*
* The old node must be freed (in memory) _before_ unlocking the new
* node after another thread has locked and updated the new node, thus
* seeing stale data:
*/
- bch2_btree_node_free_inmem(c, b, iter);
+ bch2_btree_node_free_inmem(trans, trans->paths + path, b);
+
+ if (n3)
+ bch2_trans_node_add(trans, trans->paths + path, n3);
+ if (n2)
+ bch2_trans_node_add(trans, trans->paths + path2, n2);
+ bch2_trans_node_add(trans, trans->paths + path1, n1);
if (n3)
six_unlock_intent(&n3->c.lock);
if (n2)
six_unlock_intent(&n2->c.lock);
six_unlock_intent(&n1->c.lock);
+out:
+ if (path2) {
+ __bch2_btree_path_unlock(trans, trans->paths + path2);
+ bch2_path_put(trans, path2, true);
+ }
+ if (path1) {
+ __bch2_btree_path_unlock(trans, trans->paths + path1);
+ bch2_path_put(trans, path1, true);
+ }
- bch2_btree_trans_verify_locks(iter->trans);
+ bch2_trans_verify_locks(trans);
- bch2_time_stats_update(&c->times[BCH_TIME_btree_node_split],
+ bch2_time_stats_update(&c->times[n2
+ ? BCH_TIME_btree_node_split
+ : BCH_TIME_btree_node_compact],
start_time);
+ return ret;
+err:
+ if (n3)
+ bch2_btree_node_free_never_used(as, trans, n3);
+ if (n2)
+ bch2_btree_node_free_never_used(as, trans, n2);
+ bch2_btree_node_free_never_used(as, trans, n1);
+ goto out;
}
static void
-bch2_btree_insert_keys_interior(struct btree_update *as, struct btree *b,
- struct btree_iter *iter, struct keylist *keys)
+bch2_btree_insert_keys_interior(struct btree_update *as,
+ struct btree_trans *trans,
+ struct btree_path *path,
+ struct btree *b,
+ struct keylist *keys)
{
- struct btree_iter *linked;
- struct btree_node_iter node_iter;
- struct bkey_i *insert = bch2_keylist_front(keys);
- struct bkey_packed *k;
-
- /* Don't screw up @iter's position: */
- node_iter = iter->l[b->c.level].iter;
-
- /*
- * btree_split(), btree_gc_coalesce() will insert keys before
- * the iterator's current position - they know the keys go in
- * the node the iterator points to:
- */
- while ((k = bch2_btree_node_iter_prev_all(&node_iter, b)) &&
- (bkey_cmp_left_packed(b, k, &insert->k.p) >= 0))
- ;
+ struct btree_path *linked;
+ unsigned i;
- for_each_keylist_key(keys, insert)
- bch2_insert_fixup_btree_ptr(as, b, iter, insert, &node_iter);
+ __bch2_btree_insert_keys_interior(as, trans, path, b,
+ path->l[b->c.level].iter, keys);
btree_update_updated_node(as, b);
- trans_for_each_iter_with_node(iter->trans, b, linked)
+ trans_for_each_path_with_node(trans, b, linked, i)
bch2_btree_node_iter_peek(&linked->l[b->c.level].iter, b);
- bch2_btree_trans_verify_iters(iter->trans, b);
+ bch2_trans_verify_paths(trans);
}
/**
- * bch_btree_insert_node - insert bkeys into a given btree node
+ * bch2_btree_insert_node - insert bkeys into a given btree node
*
- * @iter: btree iterator
+ * @as: btree_update object
+ * @trans: btree_trans object
+ * @path_idx: path that points to current node
+ * @b: node to insert keys into
* @keys: list of keys to insert
- * @hook: insert callback
- * @persistent: if not null, @persistent will wait on journal write
+ * @flags: transaction commit flags
+ *
+ * Returns: 0 on success, typically transaction restart error on failure
*
* Inserts as many keys as it can into a given btree node, splitting it if full.
* If a split occurred, this function will return early. This can only happen
* for leaf nodes -- inserts into interior nodes have to be atomic.
*/
-void bch2_btree_insert_node(struct btree_update *as, struct btree *b,
- struct btree_iter *iter, struct keylist *keys,
- unsigned flags)
+static int bch2_btree_insert_node(struct btree_update *as, struct btree_trans *trans,
+ btree_path_idx_t path_idx, struct btree *b,
+ struct keylist *keys, unsigned flags)
{
struct bch_fs *c = as->c;
+ struct btree_path *path = trans->paths + path_idx;
int old_u64s = le16_to_cpu(btree_bset_last(b)->u64s);
int old_live_u64s = b->nr.live_u64s;
int live_u64s_added, u64s_added;
+ int ret;
- BUG_ON(!btree_node_intent_locked(iter, btree_node_root(c, b)->c.level));
+ lockdep_assert_held(&c->gc_lock);
+ BUG_ON(!btree_node_intent_locked(path, b->c.level));
BUG_ON(!b->c.level);
BUG_ON(!as || as->b);
bch2_verify_keylist_sorted(keys);
- bch2_btree_node_lock_for_insert(c, b, iter);
+ ret = bch2_btree_node_lock_write(trans, path, &b->c);
+ if (ret)
+ return ret;
+
+ bch2_btree_node_prep_for_write(trans, path, b);
if (!bch2_btree_node_insert_fits(c, b, bch2_keylist_u64s(keys))) {
- bch2_btree_node_unlock_write(b, iter);
+ bch2_btree_node_unlock_write(trans, path, b);
goto split;
}
btree_node_interior_verify(c, b);
- bch2_btree_insert_keys_interior(as, b, iter, keys);
+ bch2_btree_insert_keys_interior(as, trans, path, b, keys);
live_u64s_added = (int) b->nr.live_u64s - old_live_u64s;
u64s_added = (int) le16_to_cpu(btree_bset_last(b)->u64s) - old_u64s;
if (u64s_added > live_u64s_added &&
bch2_maybe_compact_whiteouts(c, b))
- bch2_btree_iter_reinit_node(iter, b);
+ bch2_trans_node_reinit_iter(trans, b);
- bch2_btree_node_unlock_write(b, iter);
+ bch2_btree_node_unlock_write(trans, path, b);
btree_node_interior_verify(c, b);
-
+ return 0;
+split:
/*
- * when called from the btree_split path the new nodes aren't added to
- * the btree iterator yet, so the merge path's unlock/wait/relock dance
- * won't work:
+ * We could attempt to avoid the transaction restart, by calling
+ * bch2_btree_path_upgrade() and allocating more nodes:
*/
- bch2_foreground_maybe_merge(c, iter, b->c.level,
- flags|BTREE_INSERT_NOUNLOCK);
- return;
-split:
- btree_split(as, b, iter, keys, flags);
+ if (b->c.level >= as->update_level) {
+ trace_and_count(c, trans_restart_split_race, trans, _THIS_IP_, b);
+ return btree_trans_restart(trans, BCH_ERR_transaction_restart_split_race);
+ }
+
+ return btree_split(as, trans, path_idx, b, keys, flags);
}
-int bch2_btree_split_leaf(struct bch_fs *c, struct btree_iter *iter,
+int bch2_btree_split_leaf(struct btree_trans *trans,
+ btree_path_idx_t path,
unsigned flags)
{
- struct btree_trans *trans = iter->trans;
- struct btree *b = iter_l(iter)->b;
+ /* btree_split & merge may both cause paths array to be reallocated */
+
+ struct btree *b = path_l(trans->paths + path)->b;
struct btree_update *as;
- struct closure cl;
+ unsigned l;
int ret = 0;
- closure_init_stack(&cl);
-
- /* Hack, because gc and splitting nodes doesn't mix yet: */
- if (!(flags & BTREE_INSERT_GC_LOCK_HELD) &&
- !down_read_trylock(&c->gc_lock)) {
- if (flags & BTREE_INSERT_NOUNLOCK) {
- trace_transaction_restart_ip(trans->ip, _THIS_IP_);
- return -EINTR;
- }
-
- bch2_trans_unlock(trans);
- down_read(&c->gc_lock);
+ as = bch2_btree_update_start(trans, trans->paths + path,
+ trans->paths[path].level,
+ true, flags);
+ if (IS_ERR(as))
+ return PTR_ERR(as);
- if (!bch2_trans_relock(trans))
- ret = -EINTR;
+ ret = btree_split(as, trans, path, b, NULL, flags);
+ if (ret) {
+ bch2_btree_update_free(as, trans);
+ return ret;
}
- /*
- * XXX: figure out how far we might need to split,
- * instead of locking/reserving all the way to the root:
- */
- if (!bch2_btree_iter_upgrade(iter, U8_MAX)) {
- trace_trans_restart_iter_upgrade(trans->ip);
- ret = -EINTR;
- goto out;
- }
+ bch2_btree_update_done(as, trans);
- as = bch2_btree_update_start(trans, iter->btree_id,
- btree_update_reserve_required(c, b), flags,
- !(flags & BTREE_INSERT_NOUNLOCK) ? &cl : NULL);
- if (IS_ERR(as)) {
- ret = PTR_ERR(as);
- if (ret == -EAGAIN) {
- BUG_ON(flags & BTREE_INSERT_NOUNLOCK);
- bch2_trans_unlock(trans);
- ret = -EINTR;
+ for (l = trans->paths[path].level + 1;
+ btree_node_intent_locked(&trans->paths[path], l) && !ret;
+ l++)
+ ret = bch2_foreground_maybe_merge(trans, path, l, flags);
- trace_transaction_restart_ip(trans->ip, _THIS_IP_);
- }
- goto out;
- }
-
- btree_split(as, b, iter, NULL, flags);
- bch2_btree_update_done(as);
-
- /*
- * We haven't successfully inserted yet, so don't downgrade all the way
- * back to read locks;
- */
- __bch2_btree_iter_downgrade(iter, 1);
-out:
- if (!(flags & BTREE_INSERT_GC_LOCK_HELD))
- up_read(&c->gc_lock);
- closure_sync(&cl);
return ret;
}
-void __bch2_foreground_maybe_merge(struct bch_fs *c,
- struct btree_iter *iter,
- unsigned level,
- unsigned flags,
- enum btree_node_sibling sib)
+int __bch2_foreground_maybe_merge(struct btree_trans *trans,
+ btree_path_idx_t path,
+ unsigned level,
+ unsigned flags,
+ enum btree_node_sibling sib)
{
- struct btree_trans *trans = iter->trans;
+ struct bch_fs *c = trans->c;
struct btree_update *as;
struct bkey_format_state new_s;
struct bkey_format new_f;
struct bkey_i delete;
struct btree *b, *m, *n, *prev, *next, *parent;
- struct closure cl;
+ struct bpos sib_pos;
size_t sib_u64s;
+ enum btree_id btree = trans->paths[path].btree_id;
+ btree_path_idx_t sib_path = 0, new_path = 0;
+ u64 start_time = local_clock();
int ret = 0;
- BUG_ON(!btree_node_locked(iter, level));
+ BUG_ON(!trans->paths[path].should_be_locked);
+ BUG_ON(!btree_node_locked(&trans->paths[path], level));
- closure_init_stack(&cl);
-retry:
- BUG_ON(!btree_node_locked(iter, level));
+ b = trans->paths[path].l[level].b;
- b = iter->l[level].b;
+ if ((sib == btree_prev_sib && bpos_eq(b->data->min_key, POS_MIN)) ||
+ (sib == btree_next_sib && bpos_eq(b->data->max_key, SPOS_MAX))) {
+ b->sib_u64s[sib] = U16_MAX;
+ return 0;
+ }
- parent = btree_node_parent(iter, b);
- if (!parent)
- goto out;
+ sib_pos = sib == btree_prev_sib
+ ? bpos_predecessor(b->data->min_key)
+ : bpos_successor(b->data->max_key);
- if (b->sib_u64s[sib] > BTREE_FOREGROUND_MERGE_THRESHOLD(c))
- goto out;
-
- /* XXX: can't be holding read locks */
- m = bch2_btree_node_get_sibling(c, iter, b, sib);
- if (IS_ERR(m)) {
- ret = PTR_ERR(m);
+ sib_path = bch2_path_get(trans, btree, sib_pos,
+ U8_MAX, level, BTREE_ITER_INTENT, _THIS_IP_);
+ ret = bch2_btree_path_traverse(trans, sib_path, false);
+ if (ret)
goto err;
- }
- /* NULL means no sibling: */
- if (!m) {
+ btree_path_set_should_be_locked(trans->paths + sib_path);
+
+ m = trans->paths[sib_path].l[level].b;
+
+ if (btree_node_parent(trans->paths + path, b) !=
+ btree_node_parent(trans->paths + sib_path, m)) {
b->sib_u64s[sib] = U16_MAX;
goto out;
}
next = m;
}
+ if (!bpos_eq(bpos_successor(prev->data->max_key), next->data->min_key)) {
+ struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
+
+ bch2_bpos_to_text(&buf1, prev->data->max_key);
+ bch2_bpos_to_text(&buf2, next->data->min_key);
+ bch_err(c,
+ "%s(): btree topology error:\n"
+ " prev ends at %s\n"
+ " next starts at %s",
+ __func__, buf1.buf, buf2.buf);
+ printbuf_exit(&buf1);
+ printbuf_exit(&buf2);
+ bch2_topology_error(c);
+ ret = -EIO;
+ goto err;
+ }
+
bch2_bkey_format_init(&new_s);
- __bch2_btree_calc_format(&new_s, b);
- __bch2_btree_calc_format(&new_s, m);
+ bch2_bkey_format_add_pos(&new_s, prev->data->min_key);
+ __bch2_btree_calc_format(&new_s, prev);
+ __bch2_btree_calc_format(&new_s, next);
+ bch2_bkey_format_add_pos(&new_s, next->data->max_key);
new_f = bch2_bkey_format_done(&new_s);
- sib_u64s = btree_node_u64s_with_format(b, &new_f) +
- btree_node_u64s_with_format(m, &new_f);
+ sib_u64s = btree_node_u64s_with_format(b->nr, &b->format, &new_f) +
+ btree_node_u64s_with_format(m->nr, &m->format, &new_f);
if (sib_u64s > BTREE_FOREGROUND_MERGE_HYSTERESIS(c)) {
sib_u64s -= BTREE_FOREGROUND_MERGE_HYSTERESIS(c);
}
sib_u64s = min(sib_u64s, btree_max_u64s(c));
+ sib_u64s = min(sib_u64s, (size_t) U16_MAX - 1);
b->sib_u64s[sib] = sib_u64s;
- if (b->sib_u64s[sib] > BTREE_FOREGROUND_MERGE_THRESHOLD(c)) {
- six_unlock_intent(&m->c.lock);
+ if (b->sib_u64s[sib] > c->btree_foreground_merge_threshold)
goto out;
- }
-
- /* We're changing btree topology, doesn't mix with gc: */
- if (!(flags & BTREE_INSERT_GC_LOCK_HELD) &&
- !down_read_trylock(&c->gc_lock))
- goto err_cycle_gc_lock;
- if (!bch2_btree_iter_upgrade(iter, U8_MAX)) {
- ret = -EINTR;
- goto err_unlock;
- }
-
- as = bch2_btree_update_start(trans, iter->btree_id,
- btree_update_reserve_required(c, parent) + 1,
- flags|
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_USE_RESERVE,
- !(flags & BTREE_INSERT_NOUNLOCK) ? &cl : NULL);
- if (IS_ERR(as)) {
- ret = PTR_ERR(as);
- goto err_unlock;
- }
+ parent = btree_node_parent(trans->paths + path, b);
+ as = bch2_btree_update_start(trans, trans->paths + path, level, false,
+ BCH_TRANS_COMMIT_no_enospc|flags);
+ ret = PTR_ERR_OR_ZERO(as);
+ if (ret)
+ goto err;
- trace_btree_merge(c, b);
+ trace_and_count(c, btree_node_merge, trans, b);
bch2_btree_interior_update_will_free_node(as, b);
bch2_btree_interior_update_will_free_node(as, m);
- n = bch2_btree_node_alloc(as, b->c.level);
- bch2_btree_update_add_new_node(as, n);
+ n = bch2_btree_node_alloc(as, trans, b->c.level);
+
+ SET_BTREE_NODE_SEQ(n->data,
+ max(BTREE_NODE_SEQ(b->data),
+ BTREE_NODE_SEQ(m->data)) + 1);
btree_set_min(n, prev->data->min_key);
btree_set_max(n, next->data->max_key);
- n->data->format = new_f;
+ n->data->format = new_f;
btree_node_set_format(n, new_f);
bch2_btree_sort_into(c, n, prev);
bch2_btree_sort_into(c, n, next);
bch2_btree_build_aux_trees(n);
+ bch2_btree_update_add_new_node(as, n);
six_unlock_write(&n->c.lock);
+ new_path = get_unlocked_mut_path(trans, btree, n->c.level, n->key.k.p);
+ six_lock_increment(&n->c.lock, SIX_LOCK_intent);
+ mark_btree_node_locked(trans, trans->paths + new_path, n->c.level, BTREE_NODE_INTENT_LOCKED);
+ bch2_btree_path_level_init(trans, trans->paths + new_path, n);
+
bkey_init(&delete.k);
delete.k.p = prev->key.k.p;
bch2_keylist_add(&as->parent_keys, &delete);
bch2_keylist_add(&as->parent_keys, &n->key);
- bch2_btree_node_write(c, n, SIX_LOCK_intent);
+ bch2_trans_verify_paths(trans);
- bch2_btree_insert_node(as, parent, iter, &as->parent_keys, flags);
+ ret = bch2_btree_insert_node(as, trans, path, parent, &as->parent_keys, flags);
+ if (ret)
+ goto err_free_update;
- bch2_btree_update_get_open_buckets(as, n);
+ bch2_trans_verify_paths(trans);
- six_lock_increment(&b->c.lock, SIX_LOCK_intent);
- bch2_btree_iter_node_drop(iter, b);
- bch2_btree_iter_node_drop(iter, m);
+ bch2_btree_update_get_open_buckets(as, n);
+ bch2_btree_node_write(c, n, SIX_LOCK_intent, 0);
- bch2_btree_iter_node_replace(iter, n);
+ bch2_btree_node_free_inmem(trans, trans->paths + path, b);
+ bch2_btree_node_free_inmem(trans, trans->paths + sib_path, m);
- bch2_btree_trans_verify_iters(trans, n);
+ bch2_trans_node_add(trans, trans->paths + path, n);
- bch2_btree_node_free_inmem(c, b, iter);
- bch2_btree_node_free_inmem(c, m, iter);
+ bch2_trans_verify_paths(trans);
six_unlock_intent(&n->c.lock);
- bch2_btree_update_done(as);
+ bch2_btree_update_done(as, trans);
- if (!(flags & BTREE_INSERT_GC_LOCK_HELD))
- up_read(&c->gc_lock);
+ bch2_time_stats_update(&c->times[BCH_TIME_btree_node_merge], start_time);
out:
- bch2_btree_trans_verify_locks(trans);
-
- /*
- * Don't downgrade locks here: we're called after successful insert,
- * and the caller will downgrade locks after a successful insert
- * anyways (in case e.g. a split was required first)
- *
- * And we're also called when inserting into interior nodes in the
- * split path, and downgrading to read locks in there is potentially
- * confusing:
- */
- closure_sync(&cl);
- return;
-
-err_cycle_gc_lock:
- six_unlock_intent(&m->c.lock);
-
- if (flags & BTREE_INSERT_NOUNLOCK)
- goto out;
-
- bch2_trans_unlock(trans);
-
- down_read(&c->gc_lock);
- up_read(&c->gc_lock);
- ret = -EINTR;
- goto err;
-
-err_unlock:
- six_unlock_intent(&m->c.lock);
- if (!(flags & BTREE_INSERT_GC_LOCK_HELD))
- up_read(&c->gc_lock);
err:
- BUG_ON(ret == -EAGAIN && (flags & BTREE_INSERT_NOUNLOCK));
-
- if ((ret == -EAGAIN || ret == -EINTR) &&
- !(flags & BTREE_INSERT_NOUNLOCK)) {
- bch2_trans_unlock(trans);
- closure_sync(&cl);
- ret = bch2_btree_iter_traverse(iter);
- if (ret)
- goto out;
-
- goto retry;
- }
-
+ if (new_path)
+ bch2_path_put(trans, new_path, true);
+ bch2_path_put(trans, sib_path, true);
+ bch2_trans_verify_locks(trans);
+ return ret;
+err_free_update:
+ bch2_btree_node_free_never_used(as, trans, n);
+ bch2_btree_update_free(as, trans);
goto out;
}
-static int __btree_node_rewrite(struct bch_fs *c, struct btree_iter *iter,
- struct btree *b, unsigned flags,
- struct closure *cl)
+int bch2_btree_node_rewrite(struct btree_trans *trans,
+ struct btree_iter *iter,
+ struct btree *b,
+ unsigned flags)
{
- struct btree *n, *parent = btree_node_parent(iter, b);
+ struct bch_fs *c = trans->c;
+ struct btree *n, *parent;
struct btree_update *as;
+ btree_path_idx_t new_path = 0;
+ int ret;
- as = bch2_btree_update_start(iter->trans, iter->btree_id,
- (parent
- ? btree_update_reserve_required(c, parent)
- : 0) + 1,
- flags, cl);
- if (IS_ERR(as)) {
- trace_btree_gc_rewrite_node_fail(c, b);
- return PTR_ERR(as);
- }
+ flags |= BCH_TRANS_COMMIT_no_enospc;
+
+ struct btree_path *path = btree_iter_path(trans, iter);
+ parent = btree_node_parent(path, b);
+ as = bch2_btree_update_start(trans, path, b->c.level, false, flags);
+ ret = PTR_ERR_OR_ZERO(as);
+ if (ret)
+ goto out;
bch2_btree_interior_update_will_free_node(as, b);
- n = bch2_btree_node_alloc_replacement(as, b);
- bch2_btree_update_add_new_node(as, n);
+ n = bch2_btree_node_alloc_replacement(as, trans, b);
bch2_btree_build_aux_trees(n);
+ bch2_btree_update_add_new_node(as, n);
six_unlock_write(&n->c.lock);
- trace_btree_gc_rewrite_node(c, b);
+ new_path = get_unlocked_mut_path(trans, iter->btree_id, n->c.level, n->key.k.p);
+ six_lock_increment(&n->c.lock, SIX_LOCK_intent);
+ mark_btree_node_locked(trans, trans->paths + new_path, n->c.level, BTREE_NODE_INTENT_LOCKED);
+ bch2_btree_path_level_init(trans, trans->paths + new_path, n);
- bch2_btree_node_write(c, n, SIX_LOCK_intent);
+ trace_and_count(c, btree_node_rewrite, trans, b);
if (parent) {
bch2_keylist_add(&as->parent_keys, &n->key);
- bch2_btree_insert_node(as, parent, iter, &as->parent_keys, flags);
+ ret = bch2_btree_insert_node(as, trans, iter->path,
+ parent, &as->parent_keys, flags);
+ if (ret)
+ goto err;
} else {
- bch2_btree_set_root(as, n, iter);
+ bch2_btree_set_root(as, trans, btree_iter_path(trans, iter), n);
}
bch2_btree_update_get_open_buckets(as, n);
+ bch2_btree_node_write(c, n, SIX_LOCK_intent, 0);
+
+ bch2_btree_node_free_inmem(trans, btree_iter_path(trans, iter), b);
- six_lock_increment(&b->c.lock, SIX_LOCK_intent);
- bch2_btree_iter_node_drop(iter, b);
- bch2_btree_iter_node_replace(iter, n);
- bch2_btree_node_free_inmem(c, b, iter);
+ bch2_trans_node_add(trans, trans->paths + iter->path, n);
six_unlock_intent(&n->c.lock);
- bch2_btree_update_done(as);
- return 0;
+ bch2_btree_update_done(as, trans);
+out:
+ if (new_path)
+ bch2_path_put(trans, new_path, true);
+ bch2_trans_downgrade(trans);
+ return ret;
+err:
+ bch2_btree_node_free_never_used(as, trans, n);
+ bch2_btree_update_free(as, trans);
+ goto out;
}
-/**
- * bch_btree_node_rewrite - Rewrite/move a btree node
- *
- * Returns 0 on success, -EINTR or -EAGAIN on failure (i.e.
- * btree_check_reserve() has to wait)
- */
-int bch2_btree_node_rewrite(struct bch_fs *c, struct btree_iter *iter,
- __le64 seq, unsigned flags)
+struct async_btree_rewrite {
+ struct bch_fs *c;
+ struct work_struct work;
+ struct list_head list;
+ enum btree_id btree_id;
+ unsigned level;
+ struct bpos pos;
+ __le64 seq;
+};
+
+static int async_btree_node_rewrite_trans(struct btree_trans *trans,
+ struct async_btree_rewrite *a)
{
- struct btree_trans *trans = iter->trans;
- struct closure cl;
+ struct bch_fs *c = trans->c;
+ struct btree_iter iter;
struct btree *b;
int ret;
- flags |= BTREE_INSERT_NOFAIL;
+ bch2_trans_node_iter_init(trans, &iter, a->btree_id, a->pos,
+ BTREE_MAX_DEPTH, a->level, 0);
+ b = bch2_btree_iter_peek_node(&iter);
+ ret = PTR_ERR_OR_ZERO(b);
+ if (ret)
+ goto out;
- closure_init_stack(&cl);
+ if (!b || b->data->keys.seq != a->seq) {
+ struct printbuf buf = PRINTBUF;
- bch2_btree_iter_upgrade(iter, U8_MAX);
+ if (b)
+ bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+ else
+ prt_str(&buf, "(null");
+ bch_info(c, "%s: node to rewrite not found:, searching for seq %llu, got\n%s",
+ __func__, a->seq, buf.buf);
+ printbuf_exit(&buf);
+ goto out;
+ }
- if (!(flags & BTREE_INSERT_GC_LOCK_HELD)) {
- if (!down_read_trylock(&c->gc_lock)) {
- bch2_trans_unlock(trans);
- down_read(&c->gc_lock);
+ ret = bch2_btree_node_rewrite(trans, &iter, b, 0);
+out:
+ bch2_trans_iter_exit(trans, &iter);
+
+ return ret;
+}
+
+static void async_btree_node_rewrite_work(struct work_struct *work)
+{
+ struct async_btree_rewrite *a =
+ container_of(work, struct async_btree_rewrite, work);
+ struct bch_fs *c = a->c;
+ int ret;
+
+ ret = bch2_trans_do(c, NULL, NULL, 0,
+ async_btree_node_rewrite_trans(trans, a));
+ bch_err_fn(c, ret);
+ bch2_write_ref_put(c, BCH_WRITE_REF_node_rewrite);
+ kfree(a);
+}
+
+void bch2_btree_node_rewrite_async(struct bch_fs *c, struct btree *b)
+{
+ struct async_btree_rewrite *a;
+ int ret;
+
+ a = kmalloc(sizeof(*a), GFP_NOFS);
+ if (!a) {
+ bch_err(c, "%s: error allocating memory", __func__);
+ return;
+ }
+
+ a->c = c;
+ a->btree_id = b->c.btree_id;
+ a->level = b->c.level;
+ a->pos = b->key.k.p;
+ a->seq = b->data->keys.seq;
+ INIT_WORK(&a->work, async_btree_node_rewrite_work);
+
+ if (unlikely(!test_bit(BCH_FS_may_go_rw, &c->flags))) {
+ mutex_lock(&c->pending_node_rewrites_lock);
+ list_add(&a->list, &c->pending_node_rewrites);
+ mutex_unlock(&c->pending_node_rewrites_lock);
+ return;
+ }
+
+ if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_node_rewrite)) {
+ if (test_bit(BCH_FS_started, &c->flags)) {
+ bch_err(c, "%s: error getting c->writes ref", __func__);
+ kfree(a);
+ return;
}
+
+ ret = bch2_fs_read_write_early(c);
+ bch_err_msg(c, ret, "going read-write");
+ if (ret) {
+ kfree(a);
+ return;
+ }
+
+ bch2_write_ref_get(c, BCH_WRITE_REF_node_rewrite);
}
- while (1) {
- ret = bch2_btree_iter_traverse(iter);
- if (ret)
- break;
+ queue_work(c->btree_interior_update_worker, &a->work);
+}
- b = bch2_btree_iter_peek_node(iter);
- if (!b || b->data->keys.seq != seq)
- break;
+void bch2_do_pending_node_rewrites(struct bch_fs *c)
+{
+ struct async_btree_rewrite *a, *n;
- ret = __btree_node_rewrite(c, iter, b, flags, &cl);
- if (ret != -EAGAIN &&
- ret != -EINTR)
- break;
+ mutex_lock(&c->pending_node_rewrites_lock);
+ list_for_each_entry_safe(a, n, &c->pending_node_rewrites, list) {
+ list_del(&a->list);
- bch2_trans_unlock(trans);
- closure_sync(&cl);
+ bch2_write_ref_get(c, BCH_WRITE_REF_node_rewrite);
+ queue_work(c->btree_interior_update_worker, &a->work);
}
+ mutex_unlock(&c->pending_node_rewrites_lock);
+}
- bch2_btree_iter_downgrade(iter);
+void bch2_free_pending_node_rewrites(struct bch_fs *c)
+{
+ struct async_btree_rewrite *a, *n;
- if (!(flags & BTREE_INSERT_GC_LOCK_HELD))
- up_read(&c->gc_lock);
+ mutex_lock(&c->pending_node_rewrites_lock);
+ list_for_each_entry_safe(a, n, &c->pending_node_rewrites, list) {
+ list_del(&a->list);
- closure_sync(&cl);
- return ret;
+ kfree(a);
+ }
+ mutex_unlock(&c->pending_node_rewrites_lock);
}
-static void __bch2_btree_node_update_key(struct bch_fs *c,
- struct btree_update *as,
- struct btree_iter *iter,
- struct btree *b, struct btree *new_hash,
- struct bkey_i *new_key)
+static int __bch2_btree_node_update_key(struct btree_trans *trans,
+ struct btree_iter *iter,
+ struct btree *b, struct btree *new_hash,
+ struct bkey_i *new_key,
+ unsigned commit_flags,
+ bool skip_triggers)
{
+ struct bch_fs *c = trans->c;
+ struct btree_iter iter2 = { NULL };
struct btree *parent;
int ret;
- btree_update_will_delete_key(as, &b->key);
- btree_update_will_add_key(as, new_key);
+ if (!skip_triggers) {
+ ret = bch2_key_trigger_old(trans, b->c.btree_id, b->c.level + 1,
+ bkey_i_to_s_c(&b->key),
+ BTREE_TRIGGER_TRANSACTIONAL) ?:
+ bch2_key_trigger_new(trans, b->c.btree_id, b->c.level + 1,
+ bkey_i_to_s(new_key),
+ BTREE_TRIGGER_TRANSACTIONAL);
+ if (ret)
+ return ret;
+ }
+
+ if (new_hash) {
+ bkey_copy(&new_hash->key, new_key);
+ ret = bch2_btree_node_hash_insert(&c->btree_cache,
+ new_hash, b->c.level, b->c.btree_id);
+ BUG_ON(ret);
+ }
- parent = btree_node_parent(iter, b);
+ parent = btree_node_parent(btree_iter_path(trans, iter), b);
if (parent) {
- if (new_hash) {
- bkey_copy(&new_hash->key, new_key);
- ret = bch2_btree_node_hash_insert(&c->btree_cache,
- new_hash, b->c.level, b->c.btree_id);
- BUG_ON(ret);
- }
+ bch2_trans_copy_iter(&iter2, iter);
- bch2_keylist_add(&as->parent_keys, new_key);
- bch2_btree_insert_node(as, parent, iter, &as->parent_keys, 0);
+ iter2.path = bch2_btree_path_make_mut(trans, iter2.path,
+ iter2.flags & BTREE_ITER_INTENT,
+ _THIS_IP_);
- if (new_hash) {
- mutex_lock(&c->btree_cache.lock);
- bch2_btree_node_hash_remove(&c->btree_cache, new_hash);
+ struct btree_path *path2 = btree_iter_path(trans, &iter2);
+ BUG_ON(path2->level != b->c.level);
+ BUG_ON(!bpos_eq(path2->pos, new_key->k.p));
- bch2_btree_node_hash_remove(&c->btree_cache, b);
+ btree_path_set_level_up(trans, path2);
- bkey_copy(&b->key, new_key);
- ret = __bch2_btree_node_hash_insert(&c->btree_cache, b);
- BUG_ON(ret);
- mutex_unlock(&c->btree_cache.lock);
- } else {
- bkey_copy(&b->key, new_key);
- }
+ trans->paths_sorted = false;
+
+ ret = bch2_btree_iter_traverse(&iter2) ?:
+ bch2_trans_update(trans, &iter2, new_key, BTREE_TRIGGER_NORUN);
+ if (ret)
+ goto err;
} else {
BUG_ON(btree_node_root(c, b) != b);
- bch2_btree_node_lock_write(b, iter);
- bkey_copy(&b->key, new_key);
+ struct jset_entry *e = bch2_trans_jset_entry_alloc(trans,
+ jset_u64s(new_key->k.u64s));
+ ret = PTR_ERR_OR_ZERO(e);
+ if (ret)
+ return ret;
- if (btree_ptr_hash_val(&b->key) != b->hash_val) {
- mutex_lock(&c->btree_cache.lock);
- bch2_btree_node_hash_remove(&c->btree_cache, b);
+ journal_entry_set(e,
+ BCH_JSET_ENTRY_btree_root,
+ b->c.btree_id, b->c.level,
+ new_key, new_key->k.u64s);
+ }
- ret = __bch2_btree_node_hash_insert(&c->btree_cache, b);
- BUG_ON(ret);
- mutex_unlock(&c->btree_cache.lock);
- }
+ ret = bch2_trans_commit(trans, NULL, NULL, commit_flags);
+ if (ret)
+ goto err;
- btree_update_updated_root(as, b);
- bch2_btree_node_unlock_write(b, iter);
+ bch2_btree_node_lock_write_nofail(trans, btree_iter_path(trans, iter), &b->c);
+
+ if (new_hash) {
+ mutex_lock(&c->btree_cache.lock);
+ bch2_btree_node_hash_remove(&c->btree_cache, new_hash);
+ bch2_btree_node_hash_remove(&c->btree_cache, b);
+
+ bkey_copy(&b->key, new_key);
+ ret = __bch2_btree_node_hash_insert(&c->btree_cache, b);
+ BUG_ON(ret);
+ mutex_unlock(&c->btree_cache.lock);
+ } else {
+ bkey_copy(&b->key, new_key);
}
- bch2_btree_update_done(as);
+ bch2_btree_node_unlock_write(trans, btree_iter_path(trans, iter), b);
+out:
+ bch2_trans_iter_exit(trans, &iter2);
+ return ret;
+err:
+ if (new_hash) {
+ mutex_lock(&c->btree_cache.lock);
+ bch2_btree_node_hash_remove(&c->btree_cache, b);
+ mutex_unlock(&c->btree_cache.lock);
+ }
+ goto out;
}
-int bch2_btree_node_update_key(struct bch_fs *c, struct btree_iter *iter,
- struct btree *b,
- struct bkey_i *new_key)
+int bch2_btree_node_update_key(struct btree_trans *trans, struct btree_iter *iter,
+ struct btree *b, struct bkey_i *new_key,
+ unsigned commit_flags, bool skip_triggers)
{
- struct btree *parent = btree_node_parent(iter, b);
- struct btree_update *as = NULL;
+ struct bch_fs *c = trans->c;
struct btree *new_hash = NULL;
+ struct btree_path *path = btree_iter_path(trans, iter);
struct closure cl;
- int ret;
-
- closure_init_stack(&cl);
-
- if (!bch2_btree_iter_upgrade(iter, U8_MAX))
- return -EINTR;
+ int ret = 0;
- if (!down_read_trylock(&c->gc_lock)) {
- bch2_trans_unlock(iter->trans);
- down_read(&c->gc_lock);
+ ret = bch2_btree_path_upgrade(trans, path, b->c.level + 1);
+ if (ret)
+ return ret;
- if (!bch2_trans_relock(iter->trans)) {
- ret = -EINTR;
- goto err;
- }
- }
+ closure_init_stack(&cl);
/*
* check btree_ptr_hash_val() after @b is locked by
* btree_iter_traverse():
*/
if (btree_ptr_hash_val(new_key) != b->hash_val) {
- /* bch2_btree_reserve_get will unlock */
- ret = bch2_btree_cache_cannibalize_lock(c, &cl);
+ ret = bch2_btree_cache_cannibalize_lock(trans, &cl);
if (ret) {
- bch2_trans_unlock(iter->trans);
- up_read(&c->gc_lock);
- closure_sync(&cl);
- down_read(&c->gc_lock);
-
- if (!bch2_trans_relock(iter->trans)) {
- ret = -EINTR;
- goto err;
- }
+ ret = drop_locks_do(trans, (closure_sync(&cl), 0));
+ if (ret)
+ return ret;
}
- new_hash = bch2_btree_node_mem_alloc(c);
+ new_hash = bch2_btree_node_mem_alloc(trans, false);
}
-retry:
- as = bch2_btree_update_start(iter->trans, iter->btree_id,
- parent ? btree_update_reserve_required(c, parent) : 0,
- BTREE_INSERT_NOFAIL, &cl);
-
- if (IS_ERR(as)) {
- ret = PTR_ERR(as);
- if (ret == -EAGAIN)
- ret = -EINTR;
-
- if (ret == -EINTR) {
- bch2_trans_unlock(iter->trans);
- up_read(&c->gc_lock);
- closure_sync(&cl);
- down_read(&c->gc_lock);
-
- if (bch2_trans_relock(iter->trans))
- goto retry;
- }
-
- goto err;
- }
-
- ret = bch2_mark_bkey_replicas(c, bkey_i_to_s_c(new_key));
- if (ret)
- goto err_free_update;
- __bch2_btree_node_update_key(c, as, iter, b, new_hash, new_key);
+ path->intent_ref++;
+ ret = __bch2_btree_node_update_key(trans, iter, b, new_hash, new_key,
+ commit_flags, skip_triggers);
+ --path->intent_ref;
- bch2_btree_iter_downgrade(iter);
-err:
if (new_hash) {
mutex_lock(&c->btree_cache.lock);
list_move(&new_hash->list, &c->btree_cache.freeable);
six_unlock_write(&new_hash->c.lock);
six_unlock_intent(&new_hash->c.lock);
}
- up_read(&c->gc_lock);
closure_sync(&cl);
+ bch2_btree_cache_cannibalize_unlock(trans);
+ return ret;
+}
+
+int bch2_btree_node_update_key_get_iter(struct btree_trans *trans,
+ struct btree *b, struct bkey_i *new_key,
+ unsigned commit_flags, bool skip_triggers)
+{
+ struct btree_iter iter;
+ int ret;
+
+ bch2_trans_node_iter_init(trans, &iter, b->c.btree_id, b->key.k.p,
+ BTREE_MAX_DEPTH, b->c.level,
+ BTREE_ITER_INTENT);
+ ret = bch2_btree_iter_traverse(&iter);
+ if (ret)
+ goto out;
+
+ /* has node been freed? */
+ if (btree_iter_path(trans, &iter)->l[b->c.level].b != b) {
+ /* node has been freed: */
+ BUG_ON(!btree_node_dying(b));
+ goto out;
+ }
+
+ BUG_ON(!btree_node_hashed(b));
+
+ struct bch_extent_ptr *ptr;
+ bch2_bkey_drop_ptrs(bkey_i_to_s(new_key), ptr,
+ !bch2_bkey_has_device(bkey_i_to_s(&b->key), ptr->dev));
+
+ ret = bch2_btree_node_update_key(trans, &iter, b, new_key,
+ commit_flags, skip_triggers);
+out:
+ bch2_trans_iter_exit(trans, &iter);
return ret;
-err_free_update:
- bch2_btree_update_free(as);
- goto err;
}
/* Init code: */
bch2_btree_set_root_inmem(c, b);
}
-void bch2_btree_root_alloc(struct bch_fs *c, enum btree_id id)
+static int __bch2_btree_root_alloc(struct btree_trans *trans, enum btree_id id)
{
+ struct bch_fs *c = trans->c;
struct closure cl;
struct btree *b;
int ret;
closure_init_stack(&cl);
do {
- ret = bch2_btree_cache_cannibalize_lock(c, &cl);
+ ret = bch2_btree_cache_cannibalize_lock(trans, &cl);
closure_sync(&cl);
} while (ret);
- b = bch2_btree_node_mem_alloc(c);
- bch2_btree_cache_cannibalize_unlock(c);
+ b = bch2_btree_node_mem_alloc(trans, false);
+ bch2_btree_cache_cannibalize_unlock(trans);
set_btree_node_fake(b);
set_btree_node_need_rewrite(b);
b->c.btree_id = id;
bkey_btree_ptr_init(&b->key);
- b->key.k.p = POS_MAX;
+ b->key.k.p = SPOS_MAX;
*((u64 *) bkey_i_to_btree_ptr(&b->key)->v.start) = U64_MAX - id;
bch2_bset_init_first(b, &b->data->keys);
b->data->flags = 0;
btree_set_min(b, POS_MIN);
- btree_set_max(b, POS_MAX);
+ btree_set_max(b, SPOS_MAX);
b->data->format = bch2_btree_calc_format(b);
btree_node_set_format(b, b->data->format);
six_unlock_write(&b->c.lock);
six_unlock_intent(&b->c.lock);
+ return 0;
+}
+
+void bch2_btree_root_alloc(struct bch_fs *c, enum btree_id id)
+{
+ bch2_trans_run(c, __bch2_btree_root_alloc(trans, id));
}
void bch2_btree_updates_to_text(struct printbuf *out, struct bch_fs *c)
mutex_lock(&c->btree_interior_update_lock);
list_for_each_entry(as, &c->btree_interior_update_list, list)
- pr_buf(out, "%p m %u w %u r %u j %llu\n",
+ prt_printf(out, "%p m %u w %u r %u j %llu\n",
as,
as->mode,
as->nodes_written,
- atomic_read(&as->cl.remaining) & CLOSURE_REMAINING_MASK,
+ closure_nr_remaining(&as->cl),
as->journal.seq);
mutex_unlock(&c->btree_interior_update_lock);
}
-size_t bch2_btree_interior_updates_nr_pending(struct bch_fs *c)
+static bool bch2_btree_interior_updates_pending(struct bch_fs *c)
{
- size_t ret = 0;
- struct list_head *i;
+ bool ret;
mutex_lock(&c->btree_interior_update_lock);
- list_for_each(i, &c->btree_interior_update_list)
- ret++;
+ ret = !list_empty(&c->btree_interior_update_list);
mutex_unlock(&c->btree_interior_update_lock);
return ret;
}
-void bch2_journal_entries_to_btree_roots(struct bch_fs *c, struct jset *jset)
+bool bch2_btree_interior_updates_flush(struct bch_fs *c)
{
- struct btree_root *r;
- struct jset_entry *entry;
+ bool ret = bch2_btree_interior_updates_pending(c);
+
+ if (ret)
+ closure_wait_event(&c->btree_interior_update_wait,
+ !bch2_btree_interior_updates_pending(c));
+ return ret;
+}
+
+void bch2_journal_entry_to_btree_root(struct bch_fs *c, struct jset_entry *entry)
+{
+ struct btree_root *r = bch2_btree_id_root(c, entry->btree_id);
mutex_lock(&c->btree_root_lock);
- vstruct_for_each(jset, entry)
- if (entry->type == BCH_JSET_ENTRY_btree_root) {
- r = &c->btree_roots[entry->btree_id];
- r->level = entry->level;
- r->alive = true;
- bkey_copy(&r->key, &entry->start[0]);
- }
+ r->level = entry->level;
+ r->alive = true;
+ bkey_copy(&r->key, (struct bkey_i *) entry->start);
mutex_unlock(&c->btree_root_lock);
}
struct jset_entry *
bch2_btree_roots_to_journal_entries(struct bch_fs *c,
- struct jset_entry *start,
- struct jset_entry *end)
+ struct jset_entry *end,
+ unsigned long skip)
{
- struct jset_entry *entry;
- unsigned long have = 0;
unsigned i;
- for (entry = start; entry < end; entry = vstruct_next(entry))
- if (entry->type == BCH_JSET_ENTRY_btree_root)
- __set_bit(entry->btree_id, &have);
-
mutex_lock(&c->btree_root_lock);
- for (i = 0; i < BTREE_ID_NR; i++)
- if (c->btree_roots[i].alive && !test_bit(i, &have)) {
- journal_entry_set(end,
- BCH_JSET_ENTRY_btree_root,
- i, c->btree_roots[i].level,
- &c->btree_roots[i].key,
- c->btree_roots[i].key.u64s);
+ for (i = 0; i < btree_id_nr_alive(c); i++) {
+ struct btree_root *r = bch2_btree_id_root(c, i);
+
+ if (r->alive && !test_bit(i, &skip)) {
+ journal_entry_set(end, BCH_JSET_ENTRY_btree_root,
+ i, r->level, &r->key, r->key.k.u64s);
end = vstruct_next(end);
}
+ }
mutex_unlock(&c->btree_root_lock);
mempool_exit(&c->btree_interior_update_pool);
}
-int bch2_fs_btree_interior_update_init(struct bch_fs *c)
+void bch2_fs_btree_interior_update_init_early(struct bch_fs *c)
{
mutex_init(&c->btree_reserve_cache_lock);
INIT_LIST_HEAD(&c->btree_interior_update_list);
mutex_init(&c->btree_interior_update_lock);
INIT_WORK(&c->btree_interior_update_work, btree_interior_update_work);
+ INIT_LIST_HEAD(&c->pending_node_rewrites);
+ mutex_init(&c->pending_node_rewrites_lock);
+}
+
+int bch2_fs_btree_interior_update_init(struct bch_fs *c)
+{
c->btree_interior_update_worker =
alloc_workqueue("btree_update", WQ_UNBOUND|WQ_MEM_RECLAIM, 1);
if (!c->btree_interior_update_worker)
- return -ENOMEM;
+ return -BCH_ERR_ENOMEM_btree_interior_update_worker_init;
- return mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
- sizeof(struct btree_update));
+ if (mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
+ sizeof(struct btree_update)))
+ return -BCH_ERR_ENOMEM_btree_interior_update_pool_init;
+
+ return 0;
}