2017-03-17 14:18:50 +08:00
|
|
|
// SPDX-License-Identifier: GPL-2.0
|
|
|
|
|
|
|
|
#include "bcachefs.h"
|
|
|
|
#include "bkey_methods.h"
|
2020-12-18 04:08:58 +08:00
|
|
|
#include "bkey_buf.h"
|
2017-03-17 14:18:50 +08:00
|
|
|
#include "btree_cache.h"
|
|
|
|
#include "btree_iter.h"
|
2019-03-08 08:46:10 +08:00
|
|
|
#include "btree_key_cache.h"
|
2017-03-17 14:18:50 +08:00
|
|
|
#include "btree_locking.h"
|
2020-03-06 07:44:59 +08:00
|
|
|
#include "btree_update.h"
|
2017-03-17 14:18:50 +08:00
|
|
|
#include "debug.h"
|
2021-03-20 08:29:11 +08:00
|
|
|
#include "error.h"
|
2017-03-17 14:18:50 +08:00
|
|
|
#include "extents.h"
|
2019-03-08 08:46:10 +08:00
|
|
|
#include "journal.h"
|
2021-04-04 08:29:05 +08:00
|
|
|
#include "replicas.h"
|
2021-03-05 11:29:25 +08:00
|
|
|
#include "subvolume.h"
|
2017-03-17 14:18:50 +08:00
|
|
|
#include "trace.h"
|
|
|
|
|
|
|
|
#include <linux/prefetch.h>
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline void btree_path_list_remove(struct btree_trans *, struct btree_path *);
|
|
|
|
static inline void btree_path_list_add(struct btree_trans *, struct btree_path *,
|
|
|
|
struct btree_path *);
|
|
|
|
|
|
|
|
static struct btree_path *btree_path_alloc(struct btree_trans *, struct btree_path *);
|
|
|
|
|
2021-10-25 04:40:05 +08:00
|
|
|
/*
|
|
|
|
* Unlocks before scheduling
|
|
|
|
* Note: does not revalidate iterator
|
|
|
|
*/
|
|
|
|
static inline int bch2_trans_cond_resched(struct btree_trans *trans)
|
|
|
|
{
|
|
|
|
if (need_resched() || race_fault()) {
|
|
|
|
bch2_trans_unlock(trans);
|
|
|
|
schedule();
|
|
|
|
return bch2_trans_relock(trans) ? 0 : -EINTR;
|
|
|
|
} else {
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline int __btree_path_cmp(const struct btree_path *l,
|
|
|
|
enum btree_id r_btree_id,
|
|
|
|
bool r_cached,
|
|
|
|
struct bpos r_pos,
|
|
|
|
unsigned r_level)
|
2021-06-13 03:45:45 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
return cmp_int(l->btree_id, r_btree_id) ?:
|
2021-11-06 12:03:40 +08:00
|
|
|
cmp_int((int) l->cached, (int) r_cached) ?:
|
2021-08-31 03:18:31 +08:00
|
|
|
bpos_cmp(l->pos, r_pos) ?:
|
|
|
|
-cmp_int(l->level, r_level);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline int btree_path_cmp(const struct btree_path *l,
|
|
|
|
const struct btree_path *r)
|
|
|
|
{
|
|
|
|
return __btree_path_cmp(l, r->btree_id, r->cached, r->pos, r->level);
|
2021-06-13 03:45:45 +08:00
|
|
|
}
|
|
|
|
|
bcachefs: Start using bpos.snapshot field
This patch starts treating the bpos.snapshot field like part of the key
in the btree code:
* bpos_successor() and bpos_predecessor() now include the snapshot field
* Keys in btrees that will be using snapshots (extents, inodes, dirents
and xattrs) now always have their snapshot field set to U32_MAX
The btree iterator code gets a new flag, BTREE_ITER_ALL_SNAPSHOTS, that
determines whether we're iterating over keys in all snapshots or not -
internally, this controlls whether bkey_(successor|predecessor)
increment/decrement the snapshot field, or only the higher bits of the
key.
We add a new member to struct btree_iter, iter->snapshot: when
BTREE_ITER_ALL_SNAPSHOTS is not set, iter->pos.snapshot should always
equal iter->snapshot, which will be 0 for btrees that don't use
snapshots, and alsways U32_MAX for btrees that will use snapshots
(until we enable snapshot creation).
This patch also introduces a new metadata version number, and compat
code for reading from/writing to older versions - this isn't a forced
upgrade (yet).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2021-03-25 06:02:16 +08:00
|
|
|
static inline struct bpos bkey_successor(struct btree_iter *iter, struct bpos p)
|
|
|
|
{
|
|
|
|
/* Are we iterating over keys in all snapshots? */
|
|
|
|
if (iter->flags & BTREE_ITER_ALL_SNAPSHOTS) {
|
|
|
|
p = bpos_successor(p);
|
|
|
|
} else {
|
|
|
|
p = bpos_nosnap_successor(p);
|
|
|
|
p.snapshot = iter->snapshot;
|
|
|
|
}
|
|
|
|
|
|
|
|
return p;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline struct bpos bkey_predecessor(struct btree_iter *iter, struct bpos p)
|
|
|
|
{
|
|
|
|
/* Are we iterating over keys in all snapshots? */
|
|
|
|
if (iter->flags & BTREE_ITER_ALL_SNAPSHOTS) {
|
|
|
|
p = bpos_predecessor(p);
|
|
|
|
} else {
|
|
|
|
p = bpos_nosnap_predecessor(p);
|
|
|
|
p.snapshot = iter->snapshot;
|
|
|
|
}
|
|
|
|
|
|
|
|
return p;
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline bool is_btree_node(struct btree_path *path, unsigned l)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
|
|
|
return l < BTREE_MAX_DEPTH &&
|
2021-08-31 03:18:31 +08:00
|
|
|
(unsigned long) path->l[l].b >= 128;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2020-01-07 11:25:09 +08:00
|
|
|
static inline struct bpos btree_iter_search_key(struct btree_iter *iter)
|
2018-08-22 04:30:14 +08:00
|
|
|
{
|
2020-01-07 11:25:09 +08:00
|
|
|
struct bpos pos = iter->pos;
|
2018-08-22 05:38:41 +08:00
|
|
|
|
2020-01-07 11:25:09 +08:00
|
|
|
if ((iter->flags & BTREE_ITER_IS_EXTENTS) &&
|
|
|
|
bkey_cmp(pos, POS_MAX))
|
bcachefs: Start using bpos.snapshot field
This patch starts treating the bpos.snapshot field like part of the key
in the btree code:
* bpos_successor() and bpos_predecessor() now include the snapshot field
* Keys in btrees that will be using snapshots (extents, inodes, dirents
and xattrs) now always have their snapshot field set to U32_MAX
The btree iterator code gets a new flag, BTREE_ITER_ALL_SNAPSHOTS, that
determines whether we're iterating over keys in all snapshots or not -
internally, this controlls whether bkey_(successor|predecessor)
increment/decrement the snapshot field, or only the higher bits of the
key.
We add a new member to struct btree_iter, iter->snapshot: when
BTREE_ITER_ALL_SNAPSHOTS is not set, iter->pos.snapshot should always
equal iter->snapshot, which will be 0 for btrees that don't use
snapshots, and alsways U32_MAX for btrees that will use snapshots
(until we enable snapshot creation).
This patch also introduces a new metadata version number, and compat
code for reading from/writing to older versions - this isn't a forced
upgrade (yet).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2021-03-25 06:02:16 +08:00
|
|
|
pos = bkey_successor(iter, pos);
|
2020-01-07 11:25:09 +08:00
|
|
|
return pos;
|
2018-08-22 04:30:14 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline bool btree_path_pos_before_node(struct btree_path *path,
|
2020-03-03 02:38:19 +08:00
|
|
|
struct btree *b)
|
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
return bpos_cmp(path->pos, b->data->min_key) < 0;
|
2020-03-03 02:38:19 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline bool btree_path_pos_after_node(struct btree_path *path,
|
2020-03-03 02:38:19 +08:00
|
|
|
struct btree *b)
|
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
return bpos_cmp(b->key.k.p, path->pos) < 0;
|
2020-03-03 02:38:19 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline bool btree_path_pos_in_node(struct btree_path *path,
|
2020-03-03 02:38:19 +08:00
|
|
|
struct btree *b)
|
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
return path->btree_id == b->c.btree_id &&
|
|
|
|
!btree_path_pos_before_node(path, b) &&
|
|
|
|
!btree_path_pos_after_node(path, b);
|
2020-03-03 02:38:19 +08:00
|
|
|
}
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
/* Btree node locking: */
|
|
|
|
|
2021-08-25 09:30:06 +08:00
|
|
|
void bch2_btree_node_unlock_write(struct btree_trans *trans,
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path, struct btree *b)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_node_unlock_write_inlined(trans, path, b);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 02:22:43 +08:00
|
|
|
void __bch2_btree_node_lock_write(struct btree_trans *trans, struct btree *b)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *linked;
|
2017-03-17 14:18:50 +08:00
|
|
|
unsigned readers = 0;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, linked)
|
|
|
|
if (linked->l[b->c.level].b == b &&
|
|
|
|
btree_node_read_locked(linked, b->c.level))
|
2017-03-17 14:18:50 +08:00
|
|
|
readers++;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Must drop our read locks before calling six_lock_write() -
|
|
|
|
* six_unlock() won't do wakeups until the reader count
|
|
|
|
* goes to 0, and it's safe because we have the node intent
|
|
|
|
* locked:
|
|
|
|
*/
|
2021-03-24 11:52:27 +08:00
|
|
|
if (!b->c.lock.readers)
|
|
|
|
atomic64_sub(__SIX_VAL(read_lock, readers),
|
|
|
|
&b->c.lock.state.counter);
|
|
|
|
else
|
|
|
|
this_cpu_sub(*b->c.lock.readers, readers);
|
|
|
|
|
2021-08-25 09:30:06 +08:00
|
|
|
btree_node_lock_type(trans->c, b, SIX_LOCK_write);
|
2021-03-24 11:52:27 +08:00
|
|
|
|
|
|
|
if (!b->c.lock.readers)
|
|
|
|
atomic64_add(__SIX_VAL(read_lock, readers),
|
|
|
|
&b->c.lock.state.counter);
|
|
|
|
else
|
|
|
|
this_cpu_add(*b->c.lock.readers, readers);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 02:22:43 +08:00
|
|
|
bool __bch2_btree_node_relock(struct btree_trans *trans,
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path, unsigned level)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree *b = btree_path_node(path, level);
|
|
|
|
int want = __btree_lock_want(path, level);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!is_btree_node(path, level))
|
2017-03-17 14:18:50 +08:00
|
|
|
return false;
|
|
|
|
|
|
|
|
if (race_fault())
|
|
|
|
return false;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (six_relock_type(&b->c.lock, want, path->l[level].lock_seq) ||
|
|
|
|
(btree_node_lock_seq_matches(path, b, level) &&
|
2021-08-31 02:22:43 +08:00
|
|
|
btree_node_lock_increment(trans, b, level, want))) {
|
2021-09-15 23:15:18 +08:00
|
|
|
mark_btree_node_locked(path, level, want);
|
2019-05-15 02:08:23 +08:00
|
|
|
return true;
|
|
|
|
} else {
|
2017-03-17 14:18:50 +08:00
|
|
|
return false;
|
2019-05-15 02:08:23 +08:00
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-11-04 00:08:02 +08:00
|
|
|
bool bch2_btree_node_upgrade(struct btree_trans *trans,
|
|
|
|
struct btree_path *path, unsigned level)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree *b = path->l[level].b;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!is_btree_node(path, level))
|
2017-03-17 14:18:50 +08:00
|
|
|
return false;
|
|
|
|
|
2021-10-08 02:56:56 +08:00
|
|
|
switch (btree_lock_want(path, level)) {
|
|
|
|
case BTREE_NODE_UNLOCKED:
|
|
|
|
BUG_ON(btree_node_locked(path, level));
|
|
|
|
return true;
|
|
|
|
case BTREE_NODE_READ_LOCKED:
|
|
|
|
BUG_ON(btree_node_intent_locked(path, level));
|
|
|
|
return bch2_btree_node_relock(trans, path, level);
|
|
|
|
case BTREE_NODE_INTENT_LOCKED:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (btree_node_intent_locked(path, level))
|
2017-03-17 14:18:50 +08:00
|
|
|
return true;
|
|
|
|
|
|
|
|
if (race_fault())
|
|
|
|
return false;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (btree_node_locked(path, level)
|
2020-06-07 00:28:01 +08:00
|
|
|
? six_lock_tryupgrade(&b->c.lock)
|
2021-08-31 03:18:31 +08:00
|
|
|
: six_relock_type(&b->c.lock, SIX_LOCK_intent, path->l[level].lock_seq))
|
2017-03-17 14:18:50 +08:00
|
|
|
goto success;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (btree_node_lock_seq_matches(path, b, level) &&
|
2021-08-31 02:22:43 +08:00
|
|
|
btree_node_lock_increment(trans, b, level, BTREE_NODE_INTENT_LOCKED)) {
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_node_unlock(path, level);
|
2017-03-17 14:18:50 +08:00
|
|
|
goto success;
|
|
|
|
}
|
|
|
|
|
|
|
|
return false;
|
|
|
|
success:
|
2021-09-15 23:15:18 +08:00
|
|
|
mark_btree_node_intent_locked(path, level);
|
2017-03-17 14:18:50 +08:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline bool btree_path_get_locks(struct btree_trans *trans,
|
|
|
|
struct btree_path *path,
|
2021-08-25 09:30:06 +08:00
|
|
|
bool upgrade, unsigned long trace_ip)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
unsigned l = path->level;
|
2017-03-17 14:18:50 +08:00
|
|
|
int fail_idx = -1;
|
|
|
|
|
|
|
|
do {
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!btree_path_node(path, l))
|
2017-03-17 14:18:50 +08:00
|
|
|
break;
|
|
|
|
|
|
|
|
if (!(upgrade
|
2021-08-31 03:18:31 +08:00
|
|
|
? bch2_btree_node_upgrade(trans, path, l)
|
2021-09-05 12:05:08 +08:00
|
|
|
: bch2_btree_node_relock(trans, path, l)))
|
2017-03-17 14:18:50 +08:00
|
|
|
fail_idx = l;
|
|
|
|
|
|
|
|
l++;
|
2021-08-31 03:18:31 +08:00
|
|
|
} while (l < path->locks_want);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* When we fail to get a lock, we have to ensure that any child nodes
|
2021-08-31 03:18:31 +08:00
|
|
|
* can't be relocked so bch2_btree_path_traverse has to walk back up to
|
2017-03-17 14:18:50 +08:00
|
|
|
* the node that we failed to relock:
|
|
|
|
*/
|
2021-09-05 09:23:11 +08:00
|
|
|
if (fail_idx >= 0) {
|
|
|
|
__bch2_btree_path_unlock(path);
|
|
|
|
btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
|
|
|
|
|
|
|
|
do {
|
|
|
|
path->l[fail_idx].b = BTREE_ITER_NO_NODE_GET_LOCKS;
|
|
|
|
--fail_idx;
|
|
|
|
} while (fail_idx >= 0);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (path->uptodate == BTREE_ITER_NEED_RELOCK)
|
|
|
|
path->uptodate = BTREE_ITER_UPTODATE;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 02:45:11 +08:00
|
|
|
bch2_trans_verify_locks(trans);
|
2019-03-28 10:03:30 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
return path->uptodate < BTREE_ITER_NEED_RELOCK;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2020-06-16 07:53:46 +08:00
|
|
|
static struct bpos btree_node_pos(struct btree_bkey_cached_common *_b,
|
2021-08-31 03:54:41 +08:00
|
|
|
bool cached)
|
2020-06-16 07:53:46 +08:00
|
|
|
{
|
2021-08-31 03:54:41 +08:00
|
|
|
return !cached
|
2020-06-16 07:53:46 +08:00
|
|
|
? container_of(_b, struct btree, c)->key.k.p
|
|
|
|
: container_of(_b, struct bkey_cached, c)->key.pos;
|
|
|
|
}
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
/* Slowpath: */
|
2021-08-31 02:22:43 +08:00
|
|
|
bool __bch2_btree_node_lock(struct btree_trans *trans,
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path,
|
|
|
|
struct btree *b,
|
|
|
|
struct bpos pos, unsigned level,
|
2020-06-13 10:29:48 +08:00
|
|
|
enum six_lock_type type,
|
2020-10-29 02:17:46 +08:00
|
|
|
six_lock_should_sleep_fn should_sleep_fn, void *p,
|
|
|
|
unsigned long ip)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *linked, *deadlock_path = NULL;
|
2020-06-13 10:29:48 +08:00
|
|
|
u64 start_time = local_clock();
|
2020-10-29 02:17:46 +08:00
|
|
|
unsigned reason = 9;
|
2021-04-01 02:42:36 +08:00
|
|
|
bool ret;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2018-07-25 04:42:27 +08:00
|
|
|
/* Check if it's safe to block: */
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, linked) {
|
2017-03-17 14:18:50 +08:00
|
|
|
if (!linked->nodes_locked)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Can't block taking an intent lock if we have _any_ nodes read
|
|
|
|
* locked:
|
|
|
|
*
|
|
|
|
* - Our read lock blocks another thread with an intent lock on
|
|
|
|
* the same node from getting a write lock, and thus from
|
|
|
|
* dropping its intent lock
|
|
|
|
*
|
|
|
|
* - And the other thread may have multiple nodes intent locked:
|
|
|
|
* both the node we want to intent lock, and the node we
|
|
|
|
* already have read locked - deadlock:
|
|
|
|
*/
|
|
|
|
if (type == SIX_LOCK_intent &&
|
|
|
|
linked->nodes_locked != linked->nodes_intent_locked) {
|
2021-08-31 03:18:31 +08:00
|
|
|
deadlock_path = linked;
|
2021-04-15 01:26:15 +08:00
|
|
|
reason = 1;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (linked->btree_id != path->btree_id) {
|
|
|
|
if (linked->btree_id > path->btree_id) {
|
|
|
|
deadlock_path = linked;
|
2020-11-06 09:49:08 +08:00
|
|
|
reason = 3;
|
|
|
|
}
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2021-08-31 03:18:31 +08:00
|
|
|
* Within the same btree, cached paths come before non
|
|
|
|
* cached paths:
|
2020-11-06 09:49:08 +08:00
|
|
|
*/
|
2021-08-31 03:18:31 +08:00
|
|
|
if (linked->cached != path->cached) {
|
|
|
|
if (path->cached) {
|
|
|
|
deadlock_path = linked;
|
2020-11-06 09:49:08 +08:00
|
|
|
reason = 4;
|
|
|
|
}
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
/*
|
|
|
|
* Interior nodes must be locked before their descendants: if
|
2021-08-31 03:18:31 +08:00
|
|
|
* another path has possible descendants locked of the node
|
2017-03-17 14:18:50 +08:00
|
|
|
* we're about to lock, it must have the ancestors locked too:
|
|
|
|
*/
|
2020-11-06 09:49:08 +08:00
|
|
|
if (level > __fls(linked->nodes_locked)) {
|
2021-08-31 03:18:31 +08:00
|
|
|
deadlock_path = linked;
|
2021-04-15 01:26:15 +08:00
|
|
|
reason = 5;
|
2020-06-13 02:58:07 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Must lock btree nodes in key order: */
|
2020-11-06 09:49:08 +08:00
|
|
|
if (btree_node_locked(linked, level) &&
|
2021-03-05 05:20:16 +08:00
|
|
|
bpos_cmp(pos, btree_node_pos((void *) linked->l[level].b,
|
2021-08-31 03:54:41 +08:00
|
|
|
linked->cached)) <= 0) {
|
2021-08-31 03:18:31 +08:00
|
|
|
deadlock_path = linked;
|
2020-11-06 09:49:08 +08:00
|
|
|
reason = 7;
|
2020-10-29 02:17:46 +08:00
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (unlikely(deadlock_path)) {
|
2021-07-26 05:19:52 +08:00
|
|
|
trace_trans_restart_would_deadlock(trans->ip, ip,
|
2021-04-15 01:26:15 +08:00
|
|
|
trans->in_traverse_all, reason,
|
2021-08-31 03:18:31 +08:00
|
|
|
deadlock_path->btree_id,
|
|
|
|
deadlock_path->cached,
|
|
|
|
&deadlock_path->pos,
|
|
|
|
path->btree_id,
|
|
|
|
path->cached,
|
2021-04-15 01:26:15 +08:00
|
|
|
&pos);
|
2021-07-26 05:19:52 +08:00
|
|
|
btree_trans_restart(trans);
|
2019-04-23 12:10:08 +08:00
|
|
|
return false;
|
|
|
|
}
|
2018-07-13 11:30:45 +08:00
|
|
|
|
2020-06-13 10:29:48 +08:00
|
|
|
if (six_trylock_type(&b->c.lock, type))
|
|
|
|
return true;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans->locking_path_idx = path->idx;
|
2021-04-01 02:42:36 +08:00
|
|
|
trans->locking_pos = pos;
|
2021-08-31 03:18:31 +08:00
|
|
|
trans->locking_btree_id = path->btree_id;
|
2021-04-01 02:42:36 +08:00
|
|
|
trans->locking_level = level;
|
|
|
|
trans->locking = b;
|
2020-06-13 10:29:48 +08:00
|
|
|
|
2021-04-01 02:42:36 +08:00
|
|
|
ret = six_lock_type(&b->c.lock, type, should_sleep_fn, p) == 0;
|
|
|
|
|
|
|
|
trans->locking = NULL;
|
2021-12-17 09:36:26 +08:00
|
|
|
|
2021-04-01 02:42:36 +08:00
|
|
|
if (ret)
|
|
|
|
bch2_time_stats_update(&trans->c->times[lock_to_time_stat(type)],
|
|
|
|
start_time);
|
|
|
|
return ret;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Btree iterator locking: */
|
|
|
|
|
|
|
|
#ifdef CONFIG_BCACHEFS_DEBUG
|
2021-08-31 03:18:31 +08:00
|
|
|
|
|
|
|
static void bch2_btree_path_verify_locks(struct btree_path *path)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
|
|
|
unsigned l;
|
|
|
|
|
2021-09-05 09:23:11 +08:00
|
|
|
if (!path->nodes_locked) {
|
2021-10-08 02:56:56 +08:00
|
|
|
BUG_ON(path->uptodate == BTREE_ITER_UPTODATE &&
|
|
|
|
btree_path_node(path, path->level));
|
2021-09-05 09:23:11 +08:00
|
|
|
return;
|
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-09-05 09:23:11 +08:00
|
|
|
for (l = 0; btree_path_node(path, l); l++)
|
2021-08-31 03:18:31 +08:00
|
|
|
BUG_ON(btree_lock_want(path, l) !=
|
|
|
|
btree_node_locked_type(path, l));
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
2018-11-23 18:19:25 +08:00
|
|
|
|
2021-08-31 02:45:11 +08:00
|
|
|
void bch2_trans_verify_locks(struct btree_trans *trans)
|
2018-11-23 18:19:25 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path;
|
2018-11-23 18:19:25 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, path)
|
|
|
|
bch2_btree_path_verify_locks(path);
|
2018-11-23 18:19:25 +08:00
|
|
|
}
|
2020-03-16 04:15:08 +08:00
|
|
|
#else
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline void bch2_btree_path_verify_locks(struct btree_path *path) {}
|
2017-03-17 14:18:50 +08:00
|
|
|
#endif
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
/* Btree path locking: */
|
|
|
|
|
2021-07-25 05:12:51 +08:00
|
|
|
/*
|
|
|
|
* Only for btree_cache.c - only relocks intent locks
|
|
|
|
*/
|
2021-08-31 03:18:31 +08:00
|
|
|
bool bch2_btree_path_relock_intent(struct btree_trans *trans,
|
|
|
|
struct btree_path *path)
|
2021-07-25 05:12:51 +08:00
|
|
|
{
|
|
|
|
unsigned l;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
for (l = path->level;
|
|
|
|
l < path->locks_want && btree_path_node(path, l);
|
2021-07-25 05:12:51 +08:00
|
|
|
l++) {
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!bch2_btree_node_relock(trans, path, l)) {
|
2021-09-05 09:23:11 +08:00
|
|
|
__bch2_btree_path_unlock(path);
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
|
2021-08-25 09:30:06 +08:00
|
|
|
btree_trans_restart(trans);
|
2021-07-25 05:12:51 +08:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
__flatten
|
2021-08-31 03:18:31 +08:00
|
|
|
static bool bch2_btree_path_relock(struct btree_trans *trans,
|
|
|
|
struct btree_path *path, unsigned long trace_ip)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
bool ret = btree_path_get_locks(trans, path, false, trace_ip);
|
2021-07-26 05:19:52 +08:00
|
|
|
|
|
|
|
if (!ret)
|
2021-08-25 09:30:06 +08:00
|
|
|
btree_trans_restart(trans);
|
2021-07-26 05:19:52 +08:00
|
|
|
return ret;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
bool __bch2_btree_path_upgrade(struct btree_trans *trans,
|
|
|
|
struct btree_path *path,
|
2017-03-17 14:18:50 +08:00
|
|
|
unsigned new_locks_want)
|
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *linked;
|
2021-04-17 02:29:26 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(path->locks_want >= new_locks_want);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
path->locks_want = new_locks_want;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (btree_path_get_locks(trans, path, true, _THIS_IP_))
|
2021-04-17 02:29:26 +08:00
|
|
|
return true;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* XXX: this is ugly - we'd prefer to not be mucking with other
|
|
|
|
* iterators in the btree_trans here.
|
|
|
|
*
|
|
|
|
* On failure to upgrade the iterator, setting iter->locks_want and
|
2021-08-31 03:18:31 +08:00
|
|
|
* calling get_locks() is sufficient to make bch2_btree_path_traverse()
|
2021-04-17 02:29:26 +08:00
|
|
|
* get the locks we want on transaction restart.
|
|
|
|
*
|
|
|
|
* But if this iterator was a clone, on transaction restart what we did
|
|
|
|
* to this iterator isn't going to be preserved.
|
|
|
|
*
|
|
|
|
* Possibly we could add an iterator field for the parent iterator when
|
|
|
|
* an iterator is a copy - for now, we'll just upgrade any other
|
|
|
|
* iterators with the same btree id.
|
|
|
|
*
|
|
|
|
* The code below used to be needed to ensure ancestor nodes get locked
|
|
|
|
* before interior nodes - now that's handled by
|
2021-08-31 03:18:31 +08:00
|
|
|
* bch2_btree_path_traverse_all().
|
2021-04-17 02:29:26 +08:00
|
|
|
*/
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, linked)
|
|
|
|
if (linked != path &&
|
|
|
|
linked->cached == path->cached &&
|
|
|
|
linked->btree_id == path->btree_id &&
|
2021-04-17 02:29:26 +08:00
|
|
|
linked->locks_want < new_locks_want) {
|
|
|
|
linked->locks_want = new_locks_want;
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_path_get_locks(trans, linked, true, _THIS_IP_);
|
2021-04-17 02:29:26 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
return false;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
void __bch2_btree_path_downgrade(struct btree_path *path,
|
2021-04-03 09:29:05 +08:00
|
|
|
unsigned new_locks_want)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-04-03 09:29:05 +08:00
|
|
|
unsigned l;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(path->locks_want < new_locks_want);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
path->locks_want = new_locks_want;
|
2021-04-03 09:29:05 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
while (path->nodes_locked &&
|
|
|
|
(l = __fls(path->nodes_locked)) >= path->locks_want) {
|
|
|
|
if (l > path->level) {
|
|
|
|
btree_node_unlock(path, l);
|
2021-04-03 09:29:05 +08:00
|
|
|
} else {
|
2021-08-31 03:18:31 +08:00
|
|
|
if (btree_node_intent_locked(path, l)) {
|
|
|
|
six_lock_downgrade(&path->l[l].b->c.lock);
|
|
|
|
path->nodes_intent_locked ^= 1 << l;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
2021-04-03 09:29:05 +08:00
|
|
|
break;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
}
|
2018-11-23 18:19:25 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_path_verify_locks(path);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2020-06-09 01:26:48 +08:00
|
|
|
void bch2_trans_downgrade(struct btree_trans *trans)
|
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path;
|
2020-06-09 01:26:48 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, path)
|
|
|
|
bch2_btree_path_downgrade(path);
|
2020-06-09 01:26:48 +08:00
|
|
|
}
|
|
|
|
|
2019-05-15 21:47:40 +08:00
|
|
|
/* Btree transaction locking: */
|
|
|
|
|
|
|
|
bool bch2_trans_relock(struct btree_trans *trans)
|
2019-03-28 10:03:30 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path;
|
2019-03-28 10:03:30 +08:00
|
|
|
|
2021-07-26 05:19:52 +08:00
|
|
|
if (unlikely(trans->restarted))
|
|
|
|
return false;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, path)
|
|
|
|
if (path->should_be_locked &&
|
|
|
|
!bch2_btree_path_relock(trans, path, _RET_IP_)) {
|
2021-06-05 03:18:10 +08:00
|
|
|
trace_trans_restart_relock(trans->ip, _RET_IP_,
|
2021-08-31 03:18:31 +08:00
|
|
|
path->btree_id, &path->pos);
|
2021-07-26 05:19:52 +08:00
|
|
|
BUG_ON(!trans->restarted);
|
2021-03-31 08:35:46 +08:00
|
|
|
return false;
|
2021-04-01 09:44:55 +08:00
|
|
|
}
|
2021-03-31 08:35:46 +08:00
|
|
|
return true;
|
2019-03-28 10:03:30 +08:00
|
|
|
}
|
|
|
|
|
2019-05-15 21:47:40 +08:00
|
|
|
void bch2_trans_unlock(struct btree_trans *trans)
|
2019-03-28 10:03:30 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path;
|
2019-03-28 10:03:30 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, path)
|
|
|
|
__bch2_btree_path_unlock(path);
|
2019-03-28 10:03:30 +08:00
|
|
|
}
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
/* Btree iterator: */
|
|
|
|
|
|
|
|
#ifdef CONFIG_BCACHEFS_DEBUG
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static void bch2_btree_path_verify_cached(struct btree_trans *trans,
|
|
|
|
struct btree_path *path)
|
2020-06-16 07:53:46 +08:00
|
|
|
{
|
|
|
|
struct bkey_cached *ck;
|
2021-08-31 03:18:31 +08:00
|
|
|
bool locked = btree_node_locked(path, 0);
|
2020-06-16 07:53:46 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!bch2_btree_node_relock(trans, path, 0))
|
2020-06-16 07:53:46 +08:00
|
|
|
return;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
ck = (void *) path->l[0].b;
|
|
|
|
BUG_ON(ck->key.btree_id != path->btree_id ||
|
|
|
|
bkey_cmp(ck->key.pos, path->pos));
|
2020-06-16 07:53:46 +08:00
|
|
|
|
|
|
|
if (!locked)
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_node_unlock(path, 0);
|
2020-06-16 07:53:46 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static void bch2_btree_path_verify_level(struct btree_trans *trans,
|
|
|
|
struct btree_path *path, unsigned level)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path_level *l;
|
2021-03-21 10:13:30 +08:00
|
|
|
struct btree_node_iter tmp;
|
|
|
|
bool locked;
|
2020-02-19 05:17:55 +08:00
|
|
|
struct bkey_packed *p, *k;
|
2021-03-05 04:20:22 +08:00
|
|
|
char buf1[100], buf2[100], buf3[100];
|
2020-02-19 05:17:55 +08:00
|
|
|
const char *msg;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2020-11-03 07:20:44 +08:00
|
|
|
if (!bch2_debug_check_iterators)
|
2019-03-28 13:51:47 +08:00
|
|
|
return;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
l = &path->l[level];
|
2021-03-21 10:13:30 +08:00
|
|
|
tmp = l->iter;
|
2021-08-31 03:18:31 +08:00
|
|
|
locked = btree_node_locked(path, level);
|
2021-03-21 10:13:30 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (path->cached) {
|
2020-06-16 07:53:46 +08:00
|
|
|
if (!level)
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_path_verify_cached(trans, path);
|
2020-06-16 07:53:46 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!btree_path_node(path, level))
|
2020-02-19 05:17:55 +08:00
|
|
|
return;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!bch2_btree_node_relock(trans, path, level))
|
2016-07-22 11:05:06 +08:00
|
|
|
return;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
BUG_ON(!btree_path_pos_in_node(path, l->b));
|
2020-02-19 05:17:55 +08:00
|
|
|
|
|
|
|
bch2_btree_node_iter_verify(&l->iter, l->b);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
/*
|
2021-09-05 09:19:48 +08:00
|
|
|
* For interior nodes, the iterator will have skipped past deleted keys:
|
2017-03-17 14:18:50 +08:00
|
|
|
*/
|
2021-09-05 09:19:48 +08:00
|
|
|
p = level
|
2021-02-20 12:41:40 +08:00
|
|
|
? bch2_btree_node_iter_prev(&tmp, l->b)
|
2020-02-19 05:17:55 +08:00
|
|
|
: bch2_btree_node_iter_prev_all(&tmp, l->b);
|
|
|
|
k = bch2_btree_node_iter_peek_all(&l->iter, l->b);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (p && bkey_iter_pos_cmp(l->b, p, &path->pos) >= 0) {
|
2020-02-19 05:17:55 +08:00
|
|
|
msg = "before";
|
|
|
|
goto err;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (k && bkey_iter_pos_cmp(l->b, k, &path->pos) < 0) {
|
2020-02-19 05:17:55 +08:00
|
|
|
msg = "after";
|
|
|
|
goto err;
|
|
|
|
}
|
2021-08-31 03:54:41 +08:00
|
|
|
|
2020-02-19 05:17:55 +08:00
|
|
|
if (!locked)
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_node_unlock(path, level);
|
2020-02-19 05:17:55 +08:00
|
|
|
return;
|
|
|
|
err:
|
|
|
|
strcpy(buf2, "(none)");
|
2021-03-05 04:20:22 +08:00
|
|
|
strcpy(buf3, "(none)");
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_bpos_to_text(&PBUF(buf1), path->pos);
|
2020-02-19 05:17:55 +08:00
|
|
|
|
|
|
|
if (p) {
|
|
|
|
struct bkey uk = bkey_unpack_key(l->b, p);
|
2021-03-05 04:20:22 +08:00
|
|
|
bch2_bkey_to_text(&PBUF(buf2), &uk);
|
2020-02-19 05:17:55 +08:00
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2020-02-19 05:17:55 +08:00
|
|
|
if (k) {
|
|
|
|
struct bkey uk = bkey_unpack_key(l->b, k);
|
2021-03-05 04:20:22 +08:00
|
|
|
bch2_bkey_to_text(&PBUF(buf3), &uk);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
2020-02-19 05:17:55 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
panic("path should be %s key at level %u:\n"
|
|
|
|
"path pos %s\n"
|
2020-02-19 05:17:55 +08:00
|
|
|
"prev key %s\n"
|
|
|
|
"cur key %s\n",
|
2021-03-05 04:20:22 +08:00
|
|
|
msg, level, buf1, buf2, buf3);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static void bch2_btree_path_verify(struct btree_trans *trans,
|
|
|
|
struct btree_path *path)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-07-15 03:13:27 +08:00
|
|
|
struct bch_fs *c = trans->c;
|
2020-02-19 05:17:55 +08:00
|
|
|
unsigned i;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(path->btree_id >= BTREE_ID_NR);
|
|
|
|
|
|
|
|
for (i = 0; i < (!path->cached ? BTREE_MAX_DEPTH : 1); i++) {
|
|
|
|
if (!path->l[i].b) {
|
2021-10-30 17:28:27 +08:00
|
|
|
BUG_ON(!path->cached &&
|
|
|
|
c->btree_roots[path->btree_id].b->c.level > i);
|
2021-08-31 03:18:31 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
bch2_btree_path_verify_level(trans, path, i);
|
|
|
|
}
|
|
|
|
|
|
|
|
bch2_btree_path_verify_locks(path);
|
|
|
|
}
|
|
|
|
|
|
|
|
void bch2_trans_verify_paths(struct btree_trans *trans)
|
|
|
|
{
|
|
|
|
struct btree_path *path;
|
|
|
|
|
|
|
|
trans_for_each_path(trans, path)
|
|
|
|
bch2_btree_path_verify(trans, path);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void bch2_btree_iter_verify(struct btree_iter *iter)
|
|
|
|
{
|
|
|
|
struct btree_trans *trans = iter->trans;
|
|
|
|
|
|
|
|
BUG_ON(iter->btree_id >= BTREE_ID_NR);
|
|
|
|
|
|
|
|
BUG_ON(!!(iter->flags & BTREE_ITER_CACHED) != iter->path->cached);
|
2021-02-12 10:57:32 +08:00
|
|
|
|
bcachefs: Start using bpos.snapshot field
This patch starts treating the bpos.snapshot field like part of the key
in the btree code:
* bpos_successor() and bpos_predecessor() now include the snapshot field
* Keys in btrees that will be using snapshots (extents, inodes, dirents
and xattrs) now always have their snapshot field set to U32_MAX
The btree iterator code gets a new flag, BTREE_ITER_ALL_SNAPSHOTS, that
determines whether we're iterating over keys in all snapshots or not -
internally, this controlls whether bkey_(successor|predecessor)
increment/decrement the snapshot field, or only the higher bits of the
key.
We add a new member to struct btree_iter, iter->snapshot: when
BTREE_ITER_ALL_SNAPSHOTS is not set, iter->pos.snapshot should always
equal iter->snapshot, which will be 0 for btrees that don't use
snapshots, and alsways U32_MAX for btrees that will use snapshots
(until we enable snapshot creation).
This patch also introduces a new metadata version number, and compat
code for reading from/writing to older versions - this isn't a forced
upgrade (yet).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2021-03-25 06:02:16 +08:00
|
|
|
BUG_ON(!(iter->flags & BTREE_ITER_ALL_SNAPSHOTS) &&
|
|
|
|
iter->pos.snapshot != iter->snapshot);
|
|
|
|
|
|
|
|
BUG_ON((iter->flags & BTREE_ITER_IS_EXTENTS) &&
|
|
|
|
(iter->flags & BTREE_ITER_ALL_SNAPSHOTS));
|
|
|
|
|
2021-08-31 03:54:41 +08:00
|
|
|
BUG_ON(!(iter->flags & __BTREE_ITER_ALL_SNAPSHOTS) &&
|
bcachefs: Start using bpos.snapshot field
This patch starts treating the bpos.snapshot field like part of the key
in the btree code:
* bpos_successor() and bpos_predecessor() now include the snapshot field
* Keys in btrees that will be using snapshots (extents, inodes, dirents
and xattrs) now always have their snapshot field set to U32_MAX
The btree iterator code gets a new flag, BTREE_ITER_ALL_SNAPSHOTS, that
determines whether we're iterating over keys in all snapshots or not -
internally, this controlls whether bkey_(successor|predecessor)
increment/decrement the snapshot field, or only the higher bits of the
key.
We add a new member to struct btree_iter, iter->snapshot: when
BTREE_ITER_ALL_SNAPSHOTS is not set, iter->pos.snapshot should always
equal iter->snapshot, which will be 0 for btrees that don't use
snapshots, and alsways U32_MAX for btrees that will use snapshots
(until we enable snapshot creation).
This patch also introduces a new metadata version number, and compat
code for reading from/writing to older versions - this isn't a forced
upgrade (yet).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2021-03-25 06:02:16 +08:00
|
|
|
(iter->flags & BTREE_ITER_ALL_SNAPSHOTS) &&
|
|
|
|
!btree_type_has_snapshots(iter->btree_id));
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_path_verify(trans, iter->path);
|
2020-02-19 05:17:55 +08:00
|
|
|
}
|
|
|
|
|
2021-02-12 10:57:32 +08:00
|
|
|
static void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter)
|
|
|
|
{
|
2021-03-16 10:34:00 +08:00
|
|
|
BUG_ON((iter->flags & BTREE_ITER_FILTER_SNAPSHOTS) &&
|
|
|
|
!iter->pos.snapshot);
|
|
|
|
|
bcachefs: Start using bpos.snapshot field
This patch starts treating the bpos.snapshot field like part of the key
in the btree code:
* bpos_successor() and bpos_predecessor() now include the snapshot field
* Keys in btrees that will be using snapshots (extents, inodes, dirents
and xattrs) now always have their snapshot field set to U32_MAX
The btree iterator code gets a new flag, BTREE_ITER_ALL_SNAPSHOTS, that
determines whether we're iterating over keys in all snapshots or not -
internally, this controlls whether bkey_(successor|predecessor)
increment/decrement the snapshot field, or only the higher bits of the
key.
We add a new member to struct btree_iter, iter->snapshot: when
BTREE_ITER_ALL_SNAPSHOTS is not set, iter->pos.snapshot should always
equal iter->snapshot, which will be 0 for btrees that don't use
snapshots, and alsways U32_MAX for btrees that will use snapshots
(until we enable snapshot creation).
This patch also introduces a new metadata version number, and compat
code for reading from/writing to older versions - this isn't a forced
upgrade (yet).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2021-03-25 06:02:16 +08:00
|
|
|
BUG_ON(!(iter->flags & BTREE_ITER_ALL_SNAPSHOTS) &&
|
|
|
|
iter->pos.snapshot != iter->snapshot);
|
|
|
|
|
2021-08-31 03:54:41 +08:00
|
|
|
BUG_ON(bkey_cmp(iter->pos, bkey_start_pos(&iter->k)) < 0 ||
|
|
|
|
bkey_cmp(iter->pos, iter->k.p) > 0);
|
2021-02-12 10:57:32 +08:00
|
|
|
}
|
|
|
|
|
2021-03-05 11:29:25 +08:00
|
|
|
static int bch2_btree_iter_verify_ret(struct btree_iter *iter, struct bkey_s_c k)
|
|
|
|
{
|
|
|
|
struct btree_trans *trans = iter->trans;
|
|
|
|
struct btree_iter copy;
|
|
|
|
struct bkey_s_c prev;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
if (!bch2_debug_check_iterators)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (!(iter->flags & BTREE_ITER_FILTER_SNAPSHOTS))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (bkey_err(k) || !k.k)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
BUG_ON(!bch2_snapshot_is_ancestor(trans->c,
|
|
|
|
iter->snapshot,
|
|
|
|
k.k->p.snapshot));
|
|
|
|
|
|
|
|
bch2_trans_iter_init(trans, ©, iter->btree_id, iter->pos,
|
2021-12-26 09:13:47 +08:00
|
|
|
BTREE_ITER_NOPRESERVE|
|
2021-03-05 11:29:25 +08:00
|
|
|
BTREE_ITER_ALL_SNAPSHOTS);
|
|
|
|
prev = bch2_btree_iter_prev(©);
|
|
|
|
if (!prev.k)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
ret = bkey_err(prev);
|
|
|
|
if (ret)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
if (!bkey_cmp(prev.k->p, k.k->p) &&
|
|
|
|
bch2_snapshot_is_ancestor(trans->c, iter->snapshot,
|
|
|
|
prev.k->p.snapshot) > 0) {
|
|
|
|
char buf1[100], buf2[200];
|
|
|
|
|
|
|
|
bch2_bkey_to_text(&PBUF(buf1), k.k);
|
|
|
|
bch2_bkey_to_text(&PBUF(buf2), prev.k);
|
|
|
|
|
|
|
|
panic("iter snap %u\n"
|
|
|
|
"k %s\n"
|
|
|
|
"prev %s\n",
|
|
|
|
iter->snapshot,
|
|
|
|
buf1, buf2);
|
|
|
|
}
|
|
|
|
out:
|
|
|
|
bch2_trans_iter_exit(trans, ©);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2021-11-06 12:03:40 +08:00
|
|
|
void bch2_assert_pos_locked(struct btree_trans *trans, enum btree_id id,
|
|
|
|
struct bpos pos, bool key_cache)
|
|
|
|
{
|
|
|
|
struct btree_path *path;
|
|
|
|
unsigned idx;
|
|
|
|
char buf[100];
|
|
|
|
|
|
|
|
trans_for_each_path_inorder(trans, path, idx) {
|
|
|
|
int cmp = cmp_int(path->btree_id, id) ?:
|
|
|
|
cmp_int(path->cached, key_cache);
|
|
|
|
|
|
|
|
if (cmp > 0)
|
|
|
|
break;
|
|
|
|
if (cmp < 0)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (!(path->nodes_locked & 1) ||
|
|
|
|
!path->should_be_locked)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (!key_cache) {
|
|
|
|
if (bkey_cmp(pos, path->l[0].b->data->min_key) >= 0 &&
|
|
|
|
bkey_cmp(pos, path->l[0].b->key.k.p) <= 0)
|
|
|
|
return;
|
|
|
|
} else {
|
|
|
|
if (!bkey_cmp(pos, path->pos))
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
bch2_dump_trans_paths_updates(trans);
|
|
|
|
panic("not locked: %s %s%s\n",
|
|
|
|
bch2_btree_ids[id],
|
|
|
|
(bch2_bpos_to_text(&PBUF(buf), pos), buf),
|
|
|
|
key_cache ? " cached" : "");
|
|
|
|
}
|
|
|
|
|
2016-07-22 11:05:06 +08:00
|
|
|
#else
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline void bch2_btree_path_verify_level(struct btree_trans *trans,
|
|
|
|
struct btree_path *path, unsigned l) {}
|
|
|
|
static inline void bch2_btree_path_verify(struct btree_trans *trans,
|
|
|
|
struct btree_path *path) {}
|
2020-03-16 04:15:08 +08:00
|
|
|
static inline void bch2_btree_iter_verify(struct btree_iter *iter) {}
|
2021-02-12 10:57:32 +08:00
|
|
|
static inline void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter) {}
|
2021-03-05 11:29:25 +08:00
|
|
|
static inline int bch2_btree_iter_verify_ret(struct btree_iter *iter, struct bkey_s_c k) { return 0; }
|
2016-07-22 11:05:06 +08:00
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
#endif
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
/* Btree path: fixups after btree updates */
|
|
|
|
|
2019-08-21 05:43:47 +08:00
|
|
|
static void btree_node_iter_set_set_pos(struct btree_node_iter *iter,
|
|
|
|
struct btree *b,
|
|
|
|
struct bset_tree *t,
|
|
|
|
struct bkey_packed *k)
|
|
|
|
{
|
|
|
|
struct btree_node_iter_set *set;
|
|
|
|
|
|
|
|
btree_node_iter_for_each(iter, set)
|
|
|
|
if (set->end == t->end_offset) {
|
|
|
|
set->k = __btree_node_key_to_offset(b, k);
|
|
|
|
bch2_btree_node_iter_sort(iter, b);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
bch2_btree_node_iter_push(iter, b, k, btree_bkey_last(b, t));
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static void __bch2_btree_path_fix_key_modified(struct btree_path *path,
|
2020-01-07 11:25:09 +08:00
|
|
|
struct btree *b,
|
|
|
|
struct bkey_packed *where)
|
2019-10-02 21:56:39 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path_level *l = &path->l[b->c.level];
|
2019-10-02 21:56:39 +08:00
|
|
|
|
2020-01-07 11:25:09 +08:00
|
|
|
if (where != bch2_btree_node_iter_peek_all(&l->iter, l->b))
|
|
|
|
return;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (bkey_iter_pos_cmp(l->b, where, &path->pos) < 0)
|
2020-01-07 11:25:09 +08:00
|
|
|
bch2_btree_node_iter_advance(&l->iter, l->b);
|
2019-10-02 21:56:39 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
void bch2_btree_path_fix_key_modified(struct btree_trans *trans,
|
2019-10-02 21:56:39 +08:00
|
|
|
struct btree *b,
|
|
|
|
struct bkey_packed *where)
|
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path;
|
2019-10-02 21:56:39 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path_with_node(trans, b, path) {
|
|
|
|
__bch2_btree_path_fix_key_modified(path, b, where);
|
|
|
|
bch2_btree_path_verify_level(trans, path, b->c.level);
|
2019-10-02 21:56:39 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static void __bch2_btree_node_iter_fix(struct btree_path *path,
|
|
|
|
struct btree *b,
|
|
|
|
struct btree_node_iter *node_iter,
|
|
|
|
struct bset_tree *t,
|
|
|
|
struct bkey_packed *where,
|
|
|
|
unsigned clobber_u64s,
|
|
|
|
unsigned new_u64s)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
|
|
|
const struct bkey_packed *end = btree_bkey_last(b, t);
|
|
|
|
struct btree_node_iter_set *set;
|
|
|
|
unsigned offset = __btree_node_key_to_offset(b, where);
|
|
|
|
int shift = new_u64s - clobber_u64s;
|
2016-07-22 11:05:06 +08:00
|
|
|
unsigned old_end = t->end_offset - shift;
|
2019-09-08 07:17:40 +08:00
|
|
|
unsigned orig_iter_pos = node_iter->data[0].k;
|
|
|
|
bool iter_current_key_modified =
|
|
|
|
orig_iter_pos >= offset &&
|
|
|
|
orig_iter_pos <= offset + clobber_u64s;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
btree_node_iter_for_each(node_iter, set)
|
|
|
|
if (set->end == old_end)
|
|
|
|
goto found;
|
|
|
|
|
|
|
|
/* didn't find the bset in the iterator - might have to readd it: */
|
|
|
|
if (new_u64s &&
|
2021-08-31 03:18:31 +08:00
|
|
|
bkey_iter_pos_cmp(b, where, &path->pos) >= 0) {
|
2017-03-17 14:18:50 +08:00
|
|
|
bch2_btree_node_iter_push(node_iter, b, where, end);
|
2019-09-08 07:17:40 +08:00
|
|
|
goto fixup_done;
|
|
|
|
} else {
|
|
|
|
/* Iterator is after key that changed */
|
2019-09-20 04:07:41 +08:00
|
|
|
return;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
found:
|
2016-07-22 11:05:06 +08:00
|
|
|
set->end = t->end_offset;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
/* Iterator hasn't gotten to the key that changed yet: */
|
|
|
|
if (set->k < offset)
|
2019-09-20 04:07:41 +08:00
|
|
|
return;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
if (new_u64s &&
|
2021-08-31 03:18:31 +08:00
|
|
|
bkey_iter_pos_cmp(b, where, &path->pos) >= 0) {
|
2017-03-17 14:18:50 +08:00
|
|
|
set->k = offset;
|
|
|
|
} else if (set->k < offset + clobber_u64s) {
|
|
|
|
set->k = offset + new_u64s;
|
|
|
|
if (set->k == set->end)
|
|
|
|
bch2_btree_node_iter_set_drop(node_iter, set);
|
|
|
|
} else {
|
2019-09-08 07:17:40 +08:00
|
|
|
/* Iterator is after key that changed */
|
2017-03-17 14:18:50 +08:00
|
|
|
set->k = (int) set->k + shift;
|
2019-09-20 04:07:41 +08:00
|
|
|
return;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
bch2_btree_node_iter_sort(node_iter, b);
|
2019-09-08 07:17:40 +08:00
|
|
|
fixup_done:
|
|
|
|
if (node_iter->data[0].k != orig_iter_pos)
|
|
|
|
iter_current_key_modified = true;
|
2018-12-07 01:01:29 +08:00
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
/*
|
2019-08-21 05:43:47 +08:00
|
|
|
* When a new key is added, and the node iterator now points to that
|
|
|
|
* key, the iterator might have skipped past deleted keys that should
|
|
|
|
* come after the key the iterator now points to. We have to rewind to
|
2019-09-08 07:17:40 +08:00
|
|
|
* before those deleted keys - otherwise
|
|
|
|
* bch2_btree_node_iter_prev_all() breaks:
|
2017-03-17 14:18:50 +08:00
|
|
|
*/
|
2019-08-21 05:43:47 +08:00
|
|
|
if (!bch2_btree_node_iter_end(node_iter) &&
|
2019-09-08 07:17:40 +08:00
|
|
|
iter_current_key_modified &&
|
2021-09-05 09:19:48 +08:00
|
|
|
b->c.level) {
|
2019-08-21 05:43:47 +08:00
|
|
|
struct bset_tree *t;
|
|
|
|
struct bkey_packed *k, *k2, *p;
|
|
|
|
|
|
|
|
k = bch2_btree_node_iter_peek_all(node_iter, b);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
for_each_bset(b, t) {
|
2019-08-21 05:43:47 +08:00
|
|
|
bool set_pos = false;
|
|
|
|
|
|
|
|
if (node_iter->data[0].end == t->end_offset)
|
2017-03-17 14:18:50 +08:00
|
|
|
continue;
|
|
|
|
|
2019-08-21 05:43:47 +08:00
|
|
|
k2 = bch2_btree_node_iter_bset_pos(node_iter, b, t);
|
|
|
|
|
|
|
|
while ((p = bch2_bkey_prev_all(b, t, k2)) &&
|
|
|
|
bkey_iter_cmp(b, k, p) < 0) {
|
|
|
|
k2 = p;
|
|
|
|
set_pos = true;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
2019-08-21 05:43:47 +08:00
|
|
|
|
|
|
|
if (set_pos)
|
|
|
|
btree_node_iter_set_set_pos(node_iter,
|
|
|
|
b, t, k2);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-08-25 09:30:06 +08:00
|
|
|
void bch2_btree_node_iter_fix(struct btree_trans *trans,
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path,
|
2018-08-12 07:12:05 +08:00
|
|
|
struct btree *b,
|
|
|
|
struct btree_node_iter *node_iter,
|
|
|
|
struct bkey_packed *where,
|
|
|
|
unsigned clobber_u64s,
|
|
|
|
unsigned new_u64s)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2018-08-12 07:12:05 +08:00
|
|
|
struct bset_tree *t = bch2_bkey_to_bset_inlined(b, where);
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *linked;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (node_iter != &path->l[b->c.level].iter) {
|
|
|
|
__bch2_btree_node_iter_fix(path, b, node_iter, t,
|
2019-09-20 04:07:41 +08:00
|
|
|
where, clobber_u64s, new_u64s);
|
2020-02-19 05:17:55 +08:00
|
|
|
|
2020-11-03 07:20:44 +08:00
|
|
|
if (bch2_debug_check_iterators)
|
2020-02-19 05:17:55 +08:00
|
|
|
bch2_btree_node_iter_verify(node_iter, b);
|
2019-09-20 04:07:41 +08:00
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path_with_node(trans, b, linked) {
|
2017-03-17 14:18:50 +08:00
|
|
|
__bch2_btree_node_iter_fix(linked, b,
|
2019-09-20 04:07:41 +08:00
|
|
|
&linked->l[b->c.level].iter, t,
|
|
|
|
where, clobber_u64s, new_u64s);
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_path_verify_level(trans, linked, b->c.level);
|
2019-09-20 04:07:41 +08:00
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
/* Btree path level: pointer to a particular btree node and node iter */
|
|
|
|
|
|
|
|
static inline struct bkey_s_c __btree_iter_unpack(struct bch_fs *c,
|
|
|
|
struct btree_path_level *l,
|
2017-03-17 14:18:50 +08:00
|
|
|
struct bkey *u,
|
|
|
|
struct bkey_packed *k)
|
|
|
|
{
|
|
|
|
struct bkey_s_c ret;
|
|
|
|
|
|
|
|
if (unlikely(!k)) {
|
|
|
|
/*
|
|
|
|
* signal to bch2_btree_iter_peek_slot() that we're currently at
|
|
|
|
* a hole
|
|
|
|
*/
|
2018-11-02 03:10:01 +08:00
|
|
|
u->type = KEY_TYPE_deleted;
|
2017-03-17 14:18:50 +08:00
|
|
|
return bkey_s_c_null;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = bkey_disassemble(l->b, k, u);
|
|
|
|
|
2021-06-09 04:29:24 +08:00
|
|
|
/*
|
|
|
|
* XXX: bch2_btree_bset_insert_key() generates invalid keys when we
|
|
|
|
* overwrite extents - it sets k->type = KEY_TYPE_deleted on the key
|
|
|
|
* being overwritten but doesn't change k->size. But this is ok, because
|
|
|
|
* those keys are never written out, we just have to avoid a spurious
|
|
|
|
* assertion here:
|
|
|
|
*/
|
|
|
|
if (bch2_debug_check_bkeys && !bkey_deleted(ret.k))
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_bkey_debugcheck(c, l->b, ret);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline struct bkey_s_c btree_path_level_peek_all(struct bch_fs *c,
|
|
|
|
struct btree_path_level *l,
|
|
|
|
struct bkey *u)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
return __btree_iter_unpack(c, l, u,
|
2017-03-17 14:18:50 +08:00
|
|
|
bch2_btree_node_iter_peek_all(&l->iter, l->b));
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline struct bkey_s_c btree_path_level_peek(struct btree_trans *trans,
|
|
|
|
struct btree_path *path,
|
|
|
|
struct btree_path_level *l,
|
|
|
|
struct bkey *u)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct bkey_s_c k = __btree_iter_unpack(trans->c, l, u,
|
2017-03-17 14:18:50 +08:00
|
|
|
bch2_btree_node_iter_peek(&l->iter, l->b));
|
2021-03-22 07:32:01 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
path->pos = k.k ? k.k->p : l->b->key.k.p;
|
|
|
|
trans->paths_sorted = false;
|
2021-03-22 07:32:01 +08:00
|
|
|
return k;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline struct bkey_s_c btree_path_level_prev(struct btree_trans *trans,
|
|
|
|
struct btree_path *path,
|
|
|
|
struct btree_path_level *l,
|
|
|
|
struct bkey *u)
|
2019-09-08 05:17:21 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct bkey_s_c k = __btree_iter_unpack(trans->c, l, u,
|
2019-09-08 05:17:21 +08:00
|
|
|
bch2_btree_node_iter_prev(&l->iter, l->b));
|
2021-03-22 07:32:01 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
path->pos = k.k ? k.k->p : l->b->data->min_key;
|
|
|
|
trans->paths_sorted = false;
|
2021-03-22 07:32:01 +08:00
|
|
|
return k;
|
2019-09-08 05:17:21 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline bool btree_path_advance_to_pos(struct btree_path *path,
|
|
|
|
struct btree_path_level *l,
|
2018-08-22 04:30:14 +08:00
|
|
|
int max_advance)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2018-08-22 04:30:14 +08:00
|
|
|
struct bkey_packed *k;
|
|
|
|
int nr_advanced = 0;
|
|
|
|
|
|
|
|
while ((k = bch2_btree_node_iter_peek_all(&l->iter, l->b)) &&
|
2021-08-31 03:18:31 +08:00
|
|
|
bkey_iter_pos_cmp(l->b, k, &path->pos) < 0) {
|
2018-08-22 04:30:14 +08:00
|
|
|
if (max_advance > 0 && nr_advanced >= max_advance)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
bch2_btree_node_iter_advance(&l->iter, l->b);
|
|
|
|
nr_advanced++;
|
|
|
|
}
|
|
|
|
|
|
|
|
return true;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Verify that iterator for parent node points to child node:
|
|
|
|
*/
|
2021-08-31 03:18:31 +08:00
|
|
|
static void btree_path_verify_new_node(struct btree_trans *trans,
|
|
|
|
struct btree_path *path, struct btree *b)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path_level *l;
|
2017-03-17 14:18:50 +08:00
|
|
|
unsigned plevel;
|
|
|
|
bool parent_locked;
|
|
|
|
struct bkey_packed *k;
|
|
|
|
|
|
|
|
if (!IS_ENABLED(CONFIG_BCACHEFS_DEBUG))
|
|
|
|
return;
|
|
|
|
|
2020-06-07 00:28:01 +08:00
|
|
|
plevel = b->c.level + 1;
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!btree_path_node(path, plevel))
|
2017-03-17 14:18:50 +08:00
|
|
|
return;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
parent_locked = btree_node_locked(path, plevel);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!bch2_btree_node_relock(trans, path, plevel))
|
2017-03-17 14:18:50 +08:00
|
|
|
return;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
l = &path->l[plevel];
|
2017-03-17 14:18:50 +08:00
|
|
|
k = bch2_btree_node_iter_peek_all(&l->iter, l->b);
|
|
|
|
if (!k ||
|
|
|
|
bkey_deleted(k) ||
|
|
|
|
bkey_cmp_left_packed(l->b, k, &b->key.k.p)) {
|
2021-03-05 04:20:22 +08:00
|
|
|
char buf1[100];
|
|
|
|
char buf2[100];
|
|
|
|
char buf3[100];
|
|
|
|
char buf4[100];
|
2017-03-17 14:18:50 +08:00
|
|
|
struct bkey uk = bkey_unpack_key(b, k);
|
|
|
|
|
2021-08-31 02:22:43 +08:00
|
|
|
bch2_dump_btree_node(trans->c, l->b);
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_bpos_to_text(&PBUF(buf1), path->pos);
|
2021-03-05 04:20:22 +08:00
|
|
|
bch2_bkey_to_text(&PBUF(buf2), &uk);
|
|
|
|
bch2_bpos_to_text(&PBUF(buf3), b->data->min_key);
|
|
|
|
bch2_bpos_to_text(&PBUF(buf3), b->data->max_key);
|
2020-12-07 05:30:02 +08:00
|
|
|
panic("parent iter doesn't point to new node:\n"
|
2021-03-05 04:20:22 +08:00
|
|
|
"iter pos %s %s\n"
|
2020-12-07 05:30:02 +08:00
|
|
|
"iter key %s\n"
|
2021-03-05 04:20:22 +08:00
|
|
|
"new node %s-%s\n",
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_ids[path->btree_id], buf1,
|
2021-03-05 04:20:22 +08:00
|
|
|
buf2, buf3, buf4);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
if (!parent_locked)
|
2021-09-05 09:23:11 +08:00
|
|
|
btree_node_unlock(path, plevel);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline void __btree_path_level_init(struct btree_path *path,
|
2021-08-31 02:22:43 +08:00
|
|
|
unsigned level)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path_level *l = &path->l[level];
|
2018-08-22 04:30:14 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_node_iter_init(&l->iter, l->b, &path->pos);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2020-12-12 01:02:48 +08:00
|
|
|
/*
|
|
|
|
* Iterators to interior nodes should always be pointed at the first non
|
|
|
|
* whiteout:
|
|
|
|
*/
|
|
|
|
if (level)
|
|
|
|
bch2_btree_node_iter_peek(&l->iter, l->b);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline void btree_path_level_init(struct btree_trans *trans,
|
|
|
|
struct btree_path *path,
|
2021-08-31 02:22:43 +08:00
|
|
|
struct btree *b)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
BUG_ON(path->cached);
|
2019-03-08 08:46:10 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_path_verify_new_node(trans, path, b);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(!btree_path_pos_in_node(path, b));
|
2020-06-07 00:28:01 +08:00
|
|
|
EBUG_ON(b->c.lock.state.seq & 1);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
path->l[b->c.level].lock_seq = b->c.lock.state.seq;
|
|
|
|
path->l[b->c.level].b = b;
|
|
|
|
__btree_path_level_init(path, b->c.level);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
/* Btree path: fixups after btree node updates: */
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
/*
|
|
|
|
* A btree node is being replaced - update the iterator to point to the new
|
|
|
|
* node:
|
|
|
|
*/
|
2021-08-31 02:36:03 +08:00
|
|
|
void bch2_trans_node_add(struct btree_trans *trans, struct btree *b)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, path)
|
|
|
|
if (!path->cached &&
|
|
|
|
btree_path_pos_in_node(path, b)) {
|
2021-09-05 09:23:11 +08:00
|
|
|
enum btree_node_locked_type t =
|
|
|
|
btree_lock_want(path, b->c.level);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-09-05 09:23:11 +08:00
|
|
|
if (path->nodes_locked &&
|
|
|
|
t != BTREE_NODE_UNLOCKED) {
|
|
|
|
btree_node_unlock(path, b->c.level);
|
2020-06-07 00:28:01 +08:00
|
|
|
six_lock_increment(&b->c.lock, (enum six_lock_type) t);
|
2021-09-15 23:15:18 +08:00
|
|
|
mark_btree_node_locked(path, b->c.level, (enum six_lock_type) t);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_path_level_init(trans, path, b);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* A btree node has been modified in such a way as to invalidate iterators - fix
|
|
|
|
* them:
|
|
|
|
*/
|
2021-08-31 02:36:03 +08:00
|
|
|
void bch2_trans_node_reinit_iter(struct btree_trans *trans, struct btree *b)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path_with_node(trans, b, path)
|
|
|
|
__btree_path_level_init(path, b->c.level);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
/* Btree path: traverse, set_pos: */
|
|
|
|
|
2020-06-13 10:29:48 +08:00
|
|
|
static int lock_root_check_fn(struct six_lock *lock, void *p)
|
|
|
|
{
|
|
|
|
struct btree *b = container_of(lock, struct btree, c.lock);
|
|
|
|
struct btree **rootp = p;
|
|
|
|
|
|
|
|
return b == *rootp ? 0 : -1;
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline int btree_path_lock_root(struct btree_trans *trans,
|
|
|
|
struct btree_path *path,
|
2020-10-29 02:17:46 +08:00
|
|
|
unsigned depth_want,
|
|
|
|
unsigned long trace_ip)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-07-26 05:19:52 +08:00
|
|
|
struct bch_fs *c = trans->c;
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree *b, **rootp = &c->btree_roots[path->btree_id].b;
|
2017-03-17 14:18:50 +08:00
|
|
|
enum six_lock_type lock_type;
|
|
|
|
unsigned i;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(path->nodes_locked);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
while (1) {
|
2020-06-13 10:29:48 +08:00
|
|
|
b = READ_ONCE(*rootp);
|
2021-08-31 03:18:31 +08:00
|
|
|
path->level = READ_ONCE(b->c.level);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (unlikely(path->level < depth_want)) {
|
2017-03-17 14:18:50 +08:00
|
|
|
/*
|
|
|
|
* the root is at a lower depth than the depth we want:
|
|
|
|
* got to the end of the btree, or we're walking nodes
|
|
|
|
* greater than some depth and there are no nodes >=
|
|
|
|
* that depth
|
|
|
|
*/
|
2021-08-31 03:18:31 +08:00
|
|
|
path->level = depth_want;
|
|
|
|
for (i = path->level; i < BTREE_MAX_DEPTH; i++)
|
|
|
|
path->l[i].b = NULL;
|
2018-11-21 15:59:07 +08:00
|
|
|
return 1;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
lock_type = __btree_lock_want(path, path->level);
|
|
|
|
if (unlikely(!btree_node_lock(trans, path, b, SPOS_MAX,
|
|
|
|
path->level, lock_type,
|
2020-10-29 02:17:46 +08:00
|
|
|
lock_root_check_fn, rootp,
|
2021-07-26 05:19:52 +08:00
|
|
|
trace_ip))) {
|
|
|
|
if (trans->restarted)
|
|
|
|
return -EINTR;
|
|
|
|
continue;
|
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2020-06-13 10:29:48 +08:00
|
|
|
if (likely(b == READ_ONCE(*rootp) &&
|
2021-08-31 03:18:31 +08:00
|
|
|
b->c.level == path->level &&
|
2017-03-17 14:18:50 +08:00
|
|
|
!race_fault())) {
|
2021-08-31 03:18:31 +08:00
|
|
|
for (i = 0; i < path->level; i++)
|
|
|
|
path->l[i].b = BTREE_ITER_NO_NODE_LOCK_ROOT;
|
|
|
|
path->l[path->level].b = b;
|
|
|
|
for (i = path->level + 1; i < BTREE_MAX_DEPTH; i++)
|
|
|
|
path->l[i].b = NULL;
|
|
|
|
|
2021-09-15 23:15:18 +08:00
|
|
|
mark_btree_node_locked(path, path->level, lock_type);
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_path_level_init(trans, path, b);
|
2017-03-17 14:18:50 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2020-06-07 00:28:01 +08:00
|
|
|
six_unlock_type(&b->c.lock, lock_type);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
noinline
|
2021-08-31 03:18:31 +08:00
|
|
|
static int btree_path_prefetch(struct btree_trans *trans, struct btree_path *path)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-25 09:30:06 +08:00
|
|
|
struct bch_fs *c = trans->c;
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path_level *l = path_l(path);
|
2017-03-17 14:18:50 +08:00
|
|
|
struct btree_node_iter node_iter = l->iter;
|
|
|
|
struct bkey_packed *k;
|
2020-12-18 04:08:58 +08:00
|
|
|
struct bkey_buf tmp;
|
2019-03-26 05:06:42 +08:00
|
|
|
unsigned nr = test_bit(BCH_FS_STARTED, &c->flags)
|
2021-08-31 03:18:31 +08:00
|
|
|
? (path->level > 1 ? 0 : 2)
|
|
|
|
: (path->level > 1 ? 1 : 16);
|
|
|
|
bool was_locked = btree_node_locked(path, path->level);
|
2021-07-25 07:50:40 +08:00
|
|
|
int ret = 0;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2020-12-18 04:08:58 +08:00
|
|
|
bch2_bkey_buf_init(&tmp);
|
|
|
|
|
2021-07-25 07:50:40 +08:00
|
|
|
while (nr && !ret) {
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!bch2_btree_node_relock(trans, path, path->level))
|
2020-12-18 04:08:58 +08:00
|
|
|
break;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
bch2_btree_node_iter_advance(&node_iter, l->b);
|
|
|
|
k = bch2_btree_node_iter_peek(&node_iter, l->b);
|
|
|
|
if (!k)
|
|
|
|
break;
|
|
|
|
|
2020-12-18 04:08:58 +08:00
|
|
|
bch2_bkey_buf_unpack(&tmp, c, l->b, k);
|
2021-08-31 03:18:31 +08:00
|
|
|
ret = bch2_btree_node_prefetch(c, trans, path, tmp.k, path->btree_id,
|
|
|
|
path->level - 1);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
if (!was_locked)
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_node_unlock(path, path->level);
|
2020-12-18 04:08:58 +08:00
|
|
|
|
|
|
|
bch2_bkey_buf_exit(&tmp, c);
|
2021-07-25 07:50:40 +08:00
|
|
|
return ret;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 02:22:43 +08:00
|
|
|
static noinline void btree_node_mem_ptr_set(struct btree_trans *trans,
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path,
|
2020-02-25 04:25:00 +08:00
|
|
|
unsigned plevel, struct btree *b)
|
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path_level *l = &path->l[plevel];
|
|
|
|
bool locked = btree_node_locked(path, plevel);
|
2020-02-25 04:25:00 +08:00
|
|
|
struct bkey_packed *k;
|
|
|
|
struct bch_btree_ptr_v2 *bp;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!bch2_btree_node_relock(trans, path, plevel))
|
2020-02-25 04:25:00 +08:00
|
|
|
return;
|
|
|
|
|
|
|
|
k = bch2_btree_node_iter_peek_all(&l->iter, l->b);
|
|
|
|
BUG_ON(k->type != KEY_TYPE_btree_ptr_v2);
|
|
|
|
|
|
|
|
bp = (void *) bkeyp_val(&l->b->format, k);
|
|
|
|
bp->mem_ptr = (unsigned long)b;
|
|
|
|
|
|
|
|
if (!locked)
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_node_unlock(path, plevel);
|
2020-02-25 04:25:00 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static __always_inline int btree_path_down(struct btree_trans *trans,
|
|
|
|
struct btree_path *path,
|
|
|
|
unsigned flags,
|
2020-10-29 02:17:46 +08:00
|
|
|
unsigned long trace_ip)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-07-25 05:12:51 +08:00
|
|
|
struct bch_fs *c = trans->c;
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path_level *l = path_l(path);
|
2017-03-17 14:18:50 +08:00
|
|
|
struct btree *b;
|
2021-08-31 03:18:31 +08:00
|
|
|
unsigned level = path->level - 1;
|
|
|
|
enum six_lock_type lock_type = __btree_lock_want(path, level);
|
2020-12-18 04:08:58 +08:00
|
|
|
struct bkey_buf tmp;
|
|
|
|
int ret;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(!btree_node_locked(path, path->level));
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2020-12-18 04:08:58 +08:00
|
|
|
bch2_bkey_buf_init(&tmp);
|
|
|
|
bch2_bkey_buf_unpack(&tmp, c, l->b,
|
2017-03-17 14:18:50 +08:00
|
|
|
bch2_btree_node_iter_peek(&l->iter, l->b));
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
b = bch2_btree_node_get(trans, path, tmp.k, level, lock_type, trace_ip);
|
2020-12-18 04:08:58 +08:00
|
|
|
ret = PTR_ERR_OR_ZERO(b);
|
|
|
|
if (unlikely(ret))
|
|
|
|
goto err;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-09-15 23:15:18 +08:00
|
|
|
mark_btree_node_locked(path, level, lock_type);
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_path_level_init(trans, path, b);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2020-12-18 04:08:58 +08:00
|
|
|
if (tmp.k->k.type == KEY_TYPE_btree_ptr_v2 &&
|
|
|
|
unlikely(b != btree_node_mem_ptr(tmp.k)))
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_node_mem_ptr_set(trans, path, level + 1, b);
|
2020-02-25 04:25:00 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (flags & BTREE_ITER_PREFETCH)
|
|
|
|
ret = btree_path_prefetch(trans, path);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (btree_node_read_locked(path, level + 1))
|
|
|
|
btree_node_unlock(path, level + 1);
|
|
|
|
path->level = level;
|
2021-04-09 10:26:53 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_path_verify_locks(path);
|
2020-12-18 04:08:58 +08:00
|
|
|
err:
|
|
|
|
bch2_bkey_buf_exit(&tmp, c);
|
|
|
|
return ret;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static int btree_path_traverse_one(struct btree_trans *, struct btree_path *,
|
|
|
|
unsigned, unsigned long);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static int __btree_path_traverse_all(struct btree_trans *trans, int ret,
|
2021-06-05 03:18:10 +08:00
|
|
|
unsigned long trace_ip)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2019-03-28 10:54:42 +08:00
|
|
|
struct bch_fs *c = trans->c;
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path, *prev = NULL;
|
2021-06-13 03:45:45 +08:00
|
|
|
int i;
|
2019-03-28 10:54:42 +08:00
|
|
|
|
2019-03-08 08:46:10 +08:00
|
|
|
if (trans->in_traverse_all)
|
|
|
|
return -EINTR;
|
|
|
|
|
|
|
|
trans->in_traverse_all = true;
|
|
|
|
retry_all:
|
2021-07-26 05:19:52 +08:00
|
|
|
trans->restarted = false;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, path)
|
|
|
|
path->should_be_locked = false;
|
2019-03-28 10:54:42 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_trans_sort_paths(trans);
|
2021-04-15 01:26:15 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path_inorder_reverse(trans, path, i) {
|
2021-06-13 03:45:45 +08:00
|
|
|
if (prev) {
|
2021-08-31 03:18:31 +08:00
|
|
|
if (path->btree_id == prev->btree_id &&
|
|
|
|
path->locks_want < prev->locks_want)
|
|
|
|
__bch2_btree_path_upgrade(trans, path, prev->locks_want);
|
|
|
|
else if (!path->locks_want && prev->locks_want)
|
|
|
|
__bch2_btree_path_upgrade(trans, path, 1);
|
2021-06-13 03:45:45 +08:00
|
|
|
}
|
2021-04-15 01:26:15 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
prev = path;
|
2021-04-15 01:26:15 +08:00
|
|
|
}
|
|
|
|
|
2019-05-15 21:47:40 +08:00
|
|
|
bch2_trans_unlock(trans);
|
2020-10-29 02:17:46 +08:00
|
|
|
cond_resched();
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2019-03-28 12:07:24 +08:00
|
|
|
if (unlikely(ret == -ENOMEM)) {
|
2017-03-17 14:18:50 +08:00
|
|
|
struct closure cl;
|
|
|
|
|
|
|
|
closure_init_stack(&cl);
|
|
|
|
|
|
|
|
do {
|
|
|
|
ret = bch2_btree_cache_cannibalize_lock(c, &cl);
|
|
|
|
closure_sync(&cl);
|
|
|
|
} while (ret);
|
|
|
|
}
|
|
|
|
|
2021-10-20 03:08:00 +08:00
|
|
|
if (unlikely(ret == -EIO))
|
2019-03-28 12:07:24 +08:00
|
|
|
goto out;
|
|
|
|
|
|
|
|
BUG_ON(ret && ret != -EINTR);
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
/* Now, redo traversals in correct order: */
|
2021-06-13 03:45:45 +08:00
|
|
|
i = 0;
|
|
|
|
while (i < trans->nr_sorted) {
|
2021-08-31 03:18:31 +08:00
|
|
|
path = trans->paths + trans->sorted[i];
|
2019-03-08 08:46:10 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(!(trans->paths_allocated & (1ULL << path->idx)));
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
ret = btree_path_traverse_one(trans, path, 0, _THIS_IP_);
|
2019-03-28 10:54:42 +08:00
|
|
|
if (ret)
|
|
|
|
goto retry_all;
|
2021-06-13 03:45:45 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(!(trans->paths_allocated & (1ULL << path->idx)));
|
2021-06-13 03:45:45 +08:00
|
|
|
|
2021-10-08 02:56:56 +08:00
|
|
|
if (path->nodes_locked ||
|
|
|
|
!btree_path_node(path, path->level))
|
2021-06-13 03:45:45 +08:00
|
|
|
i++;
|
2019-03-28 10:54:42 +08:00
|
|
|
}
|
2021-06-13 03:45:45 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* BTREE_ITER_NEED_RELOCK is ok here - if we called bch2_trans_unlock()
|
2021-08-31 03:18:31 +08:00
|
|
|
* and relock(), relock() won't relock since path->should_be_locked
|
2021-06-13 03:45:45 +08:00
|
|
|
* isn't set yet, which is all fine
|
|
|
|
*/
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, path)
|
|
|
|
BUG_ON(path->uptodate >= BTREE_ITER_NEED_TRAVERSE);
|
2017-03-17 14:18:50 +08:00
|
|
|
out:
|
|
|
|
bch2_btree_cache_cannibalize_unlock(c);
|
2019-03-08 08:46:10 +08:00
|
|
|
|
2021-06-05 03:18:10 +08:00
|
|
|
trace_trans_traverse_all(trans->ip, trace_ip);
|
2017-03-17 14:18:50 +08:00
|
|
|
return ret;
|
2019-03-28 12:07:24 +08:00
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static int bch2_btree_path_traverse_all(struct btree_trans *trans)
|
2019-03-28 12:07:24 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
return __btree_path_traverse_all(trans, 0, _RET_IP_);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline bool btree_path_good_node(struct btree_trans *trans,
|
|
|
|
struct btree_path *path,
|
2019-09-08 07:19:57 +08:00
|
|
|
unsigned l, int check_pos)
|
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!is_btree_node(path, l) ||
|
|
|
|
!bch2_btree_node_relock(trans, path, l))
|
2019-09-08 07:19:57 +08:00
|
|
|
return false;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (check_pos < 0 && btree_path_pos_before_node(path, path->l[l].b))
|
2019-09-08 07:19:57 +08:00
|
|
|
return false;
|
2021-08-31 03:18:31 +08:00
|
|
|
if (check_pos > 0 && btree_path_pos_after_node(path, path->l[l].b))
|
2019-09-08 07:19:57 +08:00
|
|
|
return false;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline unsigned btree_path_up_until_good_node(struct btree_trans *trans,
|
|
|
|
struct btree_path *path,
|
2019-09-08 07:19:57 +08:00
|
|
|
int check_pos)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-09-08 08:23:30 +08:00
|
|
|
unsigned i, l = path->level;
|
2021-09-05 09:23:11 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
while (btree_path_node(path, l) &&
|
|
|
|
!btree_path_good_node(trans, path, l, check_pos)) {
|
|
|
|
btree_node_unlock(path, l);
|
|
|
|
path->l[l].b = BTREE_ITER_NO_NODE_UP;
|
2017-03-17 14:18:50 +08:00
|
|
|
l++;
|
|
|
|
}
|
|
|
|
|
2021-09-08 08:23:30 +08:00
|
|
|
/* If we need intent locks, take them too: */
|
|
|
|
for (i = l + 1;
|
|
|
|
i < path->locks_want && btree_path_node(path, i);
|
|
|
|
i++)
|
|
|
|
if (!bch2_btree_node_relock(trans, path, i))
|
|
|
|
while (l <= i) {
|
|
|
|
btree_node_unlock(path, l);
|
|
|
|
path->l[l].b = BTREE_ITER_NO_NODE_UP;
|
|
|
|
l++;
|
|
|
|
}
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
return l;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is the main state machine for walking down the btree - walks down to a
|
|
|
|
* specified depth
|
|
|
|
*
|
|
|
|
* Returns 0 on success, -EIO on error (error reading in a btree node).
|
|
|
|
*
|
|
|
|
* On error, caller (peek_node()/peek_key()) must return NULL; the error is
|
2019-05-11 04:09:17 +08:00
|
|
|
* stashed in the iterator and returned from bch2_trans_exit().
|
2017-03-17 14:18:50 +08:00
|
|
|
*/
|
2021-08-31 03:18:31 +08:00
|
|
|
static int btree_path_traverse_one(struct btree_trans *trans,
|
|
|
|
struct btree_path *path,
|
|
|
|
unsigned flags,
|
2020-10-29 02:17:46 +08:00
|
|
|
unsigned long trace_ip)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-09-08 08:23:30 +08:00
|
|
|
unsigned depth_want = path->level;
|
2021-06-05 03:18:10 +08:00
|
|
|
int ret = 0;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-10-25 04:55:17 +08:00
|
|
|
if (unlikely(trans->restarted)) {
|
|
|
|
ret = -EINTR;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2021-07-23 00:39:11 +08:00
|
|
|
/*
|
2021-08-31 03:18:31 +08:00
|
|
|
* Ensure we obey path->should_be_locked: if it's set, we can't unlock
|
|
|
|
* and re-traverse the path without a transaction restart:
|
2021-07-23 00:39:11 +08:00
|
|
|
*/
|
2021-08-31 03:18:31 +08:00
|
|
|
if (path->should_be_locked) {
|
|
|
|
ret = bch2_btree_path_relock(trans, path, trace_ip) ? 0 : -EINTR;
|
2021-07-23 00:39:11 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (path->cached) {
|
|
|
|
ret = bch2_btree_path_traverse_cached(trans, path, flags);
|
2021-06-05 03:18:10 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2019-03-08 08:46:10 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (unlikely(path->level >= BTREE_MAX_DEPTH))
|
2021-06-05 03:18:10 +08:00
|
|
|
goto out;
|
2019-03-08 08:46:10 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
path->level = btree_path_up_until_good_node(trans, path, 0);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
/*
|
2021-08-31 03:18:31 +08:00
|
|
|
* Note: path->nodes[path->level] may be temporarily NULL here - that
|
2017-03-17 14:18:50 +08:00
|
|
|
* would indicate to other code that we got to the end of the btree,
|
|
|
|
* here it indicates that relocking the root failed - it's critical that
|
2021-08-31 03:18:31 +08:00
|
|
|
* btree_path_lock_root() comes next and that it can't fail
|
2017-03-17 14:18:50 +08:00
|
|
|
*/
|
2021-08-31 03:18:31 +08:00
|
|
|
while (path->level > depth_want) {
|
|
|
|
ret = btree_path_node(path, path->level)
|
|
|
|
? btree_path_down(trans, path, flags, trace_ip)
|
|
|
|
: btree_path_lock_root(trans, path, depth_want, trace_ip);
|
2017-03-17 14:18:50 +08:00
|
|
|
if (unlikely(ret)) {
|
2021-06-05 03:18:10 +08:00
|
|
|
if (ret == 1) {
|
|
|
|
/*
|
2021-08-31 03:54:41 +08:00
|
|
|
* No nodes at this level - got to the end of
|
|
|
|
* the btree:
|
2021-06-05 03:18:10 +08:00
|
|
|
*/
|
|
|
|
ret = 0;
|
|
|
|
goto out;
|
|
|
|
}
|
2018-11-21 15:59:07 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
__bch2_btree_path_unlock(path);
|
|
|
|
path->level = depth_want;
|
2019-03-08 08:46:10 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (ret == -EIO)
|
|
|
|
path->l[path->level].b =
|
2019-03-08 08:46:10 +08:00
|
|
|
BTREE_ITER_NO_NODE_ERROR;
|
2021-08-31 03:18:31 +08:00
|
|
|
else
|
|
|
|
path->l[path->level].b =
|
2019-03-08 08:46:10 +08:00
|
|
|
BTREE_ITER_NO_NODE_DOWN;
|
2021-06-05 03:18:10 +08:00
|
|
|
goto out;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
path->uptodate = BTREE_ITER_UPTODATE;
|
2021-06-05 03:18:10 +08:00
|
|
|
out:
|
2021-07-26 05:19:52 +08:00
|
|
|
BUG_ON((ret == -EINTR) != !!trans->restarted);
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_path_verify(trans, path);
|
2021-06-05 03:18:10 +08:00
|
|
|
return ret;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static int __btree_path_traverse_all(struct btree_trans *, int, unsigned long);
|
|
|
|
|
|
|
|
int __must_check bch2_btree_path_traverse(struct btree_trans *trans,
|
|
|
|
struct btree_path *path, unsigned flags)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
if (path->uptodate < BTREE_ITER_NEED_RELOCK)
|
|
|
|
return 0;
|
|
|
|
|
2021-10-22 00:05:21 +08:00
|
|
|
return bch2_trans_cond_resched(trans) ?:
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_path_traverse_one(trans, path, flags, _RET_IP_);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static void btree_path_copy(struct btree_trans *trans, struct btree_path *dst,
|
|
|
|
struct btree_path *src)
|
|
|
|
{
|
|
|
|
unsigned i, offset = offsetof(struct btree_path, pos);
|
|
|
|
|
|
|
|
memcpy((void *) dst + offset,
|
|
|
|
(void *) src + offset,
|
|
|
|
sizeof(struct btree_path) - offset);
|
|
|
|
|
|
|
|
for (i = 0; i < BTREE_MAX_DEPTH; i++)
|
|
|
|
if (btree_node_locked(dst, i))
|
|
|
|
six_lock_increment(&dst->l[i].b->c.lock,
|
|
|
|
__btree_lock_want(dst, i));
|
|
|
|
|
|
|
|
trans->paths_sorted = false;
|
|
|
|
}
|
|
|
|
|
2021-09-08 01:55:33 +08:00
|
|
|
static struct btree_path *btree_path_clone(struct btree_trans *trans, struct btree_path *src,
|
|
|
|
bool intent)
|
2021-08-31 03:18:31 +08:00
|
|
|
{
|
2021-09-08 01:55:33 +08:00
|
|
|
struct btree_path *new = btree_path_alloc(trans, src);
|
2021-08-31 03:18:31 +08:00
|
|
|
|
2021-09-08 01:55:33 +08:00
|
|
|
btree_path_copy(trans, new, src);
|
2021-08-31 03:18:31 +08:00
|
|
|
__btree_path_get(new, intent);
|
2021-09-08 01:55:33 +08:00
|
|
|
return new;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct btree_path * __must_check
|
|
|
|
__bch2_btree_path_make_mut(struct btree_trans *trans,
|
|
|
|
struct btree_path *path, bool intent)
|
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
__btree_path_put(path, intent);
|
2021-09-08 01:55:33 +08:00
|
|
|
path = btree_path_clone(trans, path, intent);
|
2021-08-31 03:18:31 +08:00
|
|
|
path->preserve = false;
|
|
|
|
#ifdef CONFIG_BCACHEFS_DEBUG
|
|
|
|
path->ip_allocated = _RET_IP_;
|
|
|
|
#endif
|
|
|
|
return path;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct btree_path * __must_check
|
|
|
|
__bch2_btree_path_set_pos(struct btree_trans *trans,
|
|
|
|
struct btree_path *path, struct bpos new_pos,
|
|
|
|
bool intent, int cmp)
|
|
|
|
{
|
|
|
|
unsigned l = path->level;
|
|
|
|
|
|
|
|
EBUG_ON(trans->restarted);
|
|
|
|
EBUG_ON(!path->ref);
|
|
|
|
|
|
|
|
path = bch2_btree_path_make_mut(trans, path, intent);
|
|
|
|
|
|
|
|
path->pos = new_pos;
|
|
|
|
path->should_be_locked = false;
|
|
|
|
trans->paths_sorted = false;
|
|
|
|
|
|
|
|
if (unlikely(path->cached)) {
|
|
|
|
btree_node_unlock(path, 0);
|
|
|
|
path->l[0].b = BTREE_ITER_NO_NODE_CACHED;
|
|
|
|
btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
l = btree_path_up_until_good_node(trans, path, cmp);
|
|
|
|
|
|
|
|
if (btree_path_node(path, l)) {
|
|
|
|
/*
|
|
|
|
* We might have to skip over many keys, or just a few: try
|
|
|
|
* advancing the node iterator, and if we have to skip over too
|
|
|
|
* many keys just reinit it (or if we're rewinding, since that
|
|
|
|
* is expensive).
|
|
|
|
*/
|
|
|
|
if (cmp < 0 ||
|
|
|
|
!btree_path_advance_to_pos(path, &path->l[l], 8))
|
|
|
|
__btree_path_level_init(path, l);
|
|
|
|
}
|
|
|
|
|
2021-09-05 09:23:11 +08:00
|
|
|
if (l != path->level) {
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
|
2021-09-05 09:23:11 +08:00
|
|
|
__bch2_btree_path_unlock(path);
|
|
|
|
}
|
2021-08-31 03:18:31 +08:00
|
|
|
out:
|
|
|
|
bch2_btree_path_verify(trans, path);
|
|
|
|
return path;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline struct btree_path * __must_check
|
|
|
|
btree_path_set_pos(struct btree_trans *trans,
|
|
|
|
struct btree_path *path, struct bpos new_pos,
|
|
|
|
bool intent)
|
|
|
|
{
|
|
|
|
int cmp = bpos_cmp(new_pos, path->pos);
|
|
|
|
|
|
|
|
return cmp
|
|
|
|
? __bch2_btree_path_set_pos(trans, path, new_pos, intent, cmp)
|
|
|
|
: path;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Btree path: main interface: */
|
|
|
|
|
|
|
|
static struct btree_path *have_path_at_pos(struct btree_trans *trans, struct btree_path *path)
|
|
|
|
{
|
|
|
|
struct btree_path *next;
|
|
|
|
|
|
|
|
next = prev_btree_path(trans, path);
|
|
|
|
if (next && !btree_path_cmp(next, path))
|
|
|
|
return next;
|
|
|
|
|
|
|
|
next = next_btree_path(trans, path);
|
|
|
|
if (next && !btree_path_cmp(next, path))
|
|
|
|
return next;
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2021-11-07 23:19:37 +08:00
|
|
|
static struct btree_path *have_node_at_pos(struct btree_trans *trans, struct btree_path *path)
|
2021-08-31 03:18:31 +08:00
|
|
|
{
|
|
|
|
struct btree_path *next;
|
|
|
|
|
|
|
|
next = prev_btree_path(trans, path);
|
2021-11-07 23:19:37 +08:00
|
|
|
if (next && next->level == path->level && path_l(next)->b == path_l(path)->b)
|
|
|
|
return next;
|
2021-08-31 03:18:31 +08:00
|
|
|
|
|
|
|
next = next_btree_path(trans, path);
|
2021-11-07 23:19:37 +08:00
|
|
|
if (next && next->level == path->level && path_l(next)->b == path_l(path)->b)
|
|
|
|
return next;
|
2021-08-31 03:18:31 +08:00
|
|
|
|
2021-11-07 23:19:37 +08:00
|
|
|
return NULL;
|
2021-08-31 03:18:31 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void __bch2_path_free(struct btree_trans *trans, struct btree_path *path)
|
2021-03-24 09:22:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
__bch2_btree_path_unlock(path);
|
|
|
|
btree_path_list_remove(trans, path);
|
|
|
|
trans->paths_allocated &= ~(1ULL << path->idx);
|
2021-03-24 09:22:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
void bch2_path_put(struct btree_trans *trans, struct btree_path *path, bool intent)
|
|
|
|
{
|
|
|
|
struct btree_path *dup;
|
|
|
|
|
|
|
|
EBUG_ON(trans->paths + path->idx != path);
|
|
|
|
EBUG_ON(!path->ref);
|
|
|
|
|
|
|
|
if (!__btree_path_put(path, intent))
|
|
|
|
return;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Perhaps instead we should check for duplicate paths in traverse_all:
|
|
|
|
*/
|
|
|
|
if (path->preserve &&
|
|
|
|
(dup = have_path_at_pos(trans, path))) {
|
|
|
|
dup->preserve = true;
|
|
|
|
path->preserve = false;
|
2021-11-07 23:19:37 +08:00
|
|
|
goto free;
|
2021-08-31 03:18:31 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
if (!path->preserve &&
|
2021-11-07 23:19:37 +08:00
|
|
|
(dup = have_node_at_pos(trans, path)))
|
|
|
|
goto free;
|
|
|
|
return;
|
|
|
|
free:
|
|
|
|
if (path->should_be_locked &&
|
|
|
|
!btree_node_locked(dup, path->level))
|
|
|
|
return;
|
|
|
|
|
|
|
|
dup->should_be_locked |= path->should_be_locked;
|
|
|
|
__bch2_path_free(trans, path);
|
2021-08-31 03:18:31 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
noinline __cold
|
|
|
|
void bch2_dump_trans_paths_updates(struct btree_trans *trans)
|
|
|
|
{
|
|
|
|
struct btree_path *path;
|
|
|
|
struct btree_insert_entry *i;
|
|
|
|
unsigned idx;
|
2021-10-14 23:47:52 +08:00
|
|
|
char buf1[300], buf2[300];
|
2021-08-31 03:18:31 +08:00
|
|
|
|
|
|
|
btree_trans_sort_paths(trans);
|
|
|
|
|
|
|
|
trans_for_each_path_inorder(trans, path, idx)
|
2021-11-06 12:03:40 +08:00
|
|
|
printk(KERN_ERR "path: idx %u ref %u:%u%s%s btree %s pos %s locks %u %pS\n",
|
2021-08-31 03:18:31 +08:00
|
|
|
path->idx, path->ref, path->intent_ref,
|
2021-11-06 12:03:40 +08:00
|
|
|
path->should_be_locked ? " S" : "",
|
|
|
|
path->preserve ? " P" : "",
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_ids[path->btree_id],
|
2021-10-14 23:47:52 +08:00
|
|
|
(bch2_bpos_to_text(&PBUF(buf1), path->pos), buf1),
|
2021-11-06 12:03:40 +08:00
|
|
|
path->nodes_locked,
|
2021-08-31 03:18:31 +08:00
|
|
|
#ifdef CONFIG_BCACHEFS_DEBUG
|
|
|
|
(void *) path->ip_allocated
|
|
|
|
#else
|
|
|
|
NULL
|
|
|
|
#endif
|
|
|
|
);
|
|
|
|
|
2021-10-14 23:47:52 +08:00
|
|
|
trans_for_each_update(trans, i) {
|
|
|
|
struct bkey u;
|
|
|
|
struct bkey_s_c old = bch2_btree_path_peek_slot(i->path, &u);
|
|
|
|
|
|
|
|
printk(KERN_ERR "update: btree %s %pS\n old %s\n new %s",
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_ids[i->btree_id],
|
2021-10-14 23:47:52 +08:00
|
|
|
(void *) i->ip_allocated,
|
|
|
|
(bch2_bkey_val_to_text(&PBUF(buf1), trans->c, old), buf1),
|
|
|
|
(bch2_bkey_val_to_text(&PBUF(buf2), trans->c, bkey_i_to_s_c(i->k)), buf2));
|
|
|
|
}
|
2021-08-31 03:18:31 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct btree_path *btree_path_alloc(struct btree_trans *trans,
|
|
|
|
struct btree_path *pos)
|
|
|
|
{
|
|
|
|
struct btree_path *path;
|
|
|
|
unsigned idx;
|
|
|
|
|
|
|
|
if (unlikely(trans->paths_allocated ==
|
|
|
|
~((~0ULL << 1) << (BTREE_ITER_MAX - 1)))) {
|
|
|
|
bch2_dump_trans_paths_updates(trans);
|
|
|
|
panic("trans path oveflow\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
idx = __ffs64(~trans->paths_allocated);
|
|
|
|
trans->paths_allocated |= 1ULL << idx;
|
|
|
|
|
|
|
|
path = &trans->paths[idx];
|
|
|
|
|
|
|
|
path->idx = idx;
|
|
|
|
path->ref = 0;
|
|
|
|
path->intent_ref = 0;
|
|
|
|
path->nodes_locked = 0;
|
|
|
|
path->nodes_intent_locked = 0;
|
|
|
|
|
|
|
|
btree_path_list_add(trans, pos, path);
|
|
|
|
return path;
|
|
|
|
}
|
|
|
|
|
2021-12-22 09:48:26 +08:00
|
|
|
struct btree_path *bch2_path_get(struct btree_trans *trans,
|
2021-08-31 03:18:31 +08:00
|
|
|
enum btree_id btree_id, struct bpos pos,
|
|
|
|
unsigned locks_want, unsigned level,
|
2021-12-22 09:48:26 +08:00
|
|
|
unsigned flags)
|
2021-08-31 03:18:31 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path, *path_pos = NULL;
|
2021-12-22 09:48:26 +08:00
|
|
|
bool cached = flags & BTREE_ITER_CACHED;
|
|
|
|
bool intent = flags & BTREE_ITER_INTENT;
|
2021-08-31 03:18:31 +08:00
|
|
|
int i;
|
|
|
|
|
|
|
|
BUG_ON(trans->restarted);
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_trans_sort_paths(trans);
|
2021-08-31 03:18:31 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path_inorder(trans, path, i) {
|
|
|
|
if (__btree_path_cmp(path,
|
|
|
|
btree_id,
|
|
|
|
cached,
|
|
|
|
pos,
|
|
|
|
level) > 0)
|
|
|
|
break;
|
2021-08-31 03:18:31 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
path_pos = path;
|
2021-08-31 03:18:31 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (path_pos &&
|
|
|
|
path_pos->cached == cached &&
|
|
|
|
path_pos->btree_id == btree_id &&
|
|
|
|
path_pos->level == level) {
|
|
|
|
__btree_path_get(path_pos, intent);
|
|
|
|
path = btree_path_set_pos(trans, path_pos, pos, intent);
|
2021-08-31 03:18:31 +08:00
|
|
|
} else {
|
2021-08-31 03:18:31 +08:00
|
|
|
path = btree_path_alloc(trans, path_pos);
|
|
|
|
path_pos = NULL;
|
2021-08-31 03:18:31 +08:00
|
|
|
|
|
|
|
__btree_path_get(path, intent);
|
|
|
|
path->pos = pos;
|
|
|
|
path->btree_id = btree_id;
|
|
|
|
path->cached = cached;
|
|
|
|
path->uptodate = BTREE_ITER_NEED_TRAVERSE;
|
|
|
|
path->should_be_locked = false;
|
|
|
|
path->level = level;
|
|
|
|
path->locks_want = locks_want;
|
|
|
|
path->nodes_locked = 0;
|
|
|
|
path->nodes_intent_locked = 0;
|
|
|
|
for (i = 0; i < ARRAY_SIZE(path->l); i++)
|
|
|
|
path->l[i].b = BTREE_ITER_NO_NODE_INIT;
|
|
|
|
#ifdef CONFIG_BCACHEFS_DEBUG
|
|
|
|
path->ip_allocated = _RET_IP_;
|
|
|
|
#endif
|
|
|
|
trans->paths_sorted = false;
|
|
|
|
}
|
|
|
|
|
2021-12-22 09:48:26 +08:00
|
|
|
if (!(flags & BTREE_ITER_NOPRESERVE))
|
|
|
|
path->preserve = true;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (path->intent_ref)
|
|
|
|
locks_want = max(locks_want, level + 1);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If the path has locks_want greater than requested, we don't downgrade
|
|
|
|
* it here - on transaction restart because btree node split needs to
|
|
|
|
* upgrade locks, we might be putting/getting the iterator again.
|
|
|
|
* Downgrading iterators only happens via bch2_trans_downgrade(), after
|
|
|
|
* a successful transaction commit.
|
|
|
|
*/
|
|
|
|
|
|
|
|
locks_want = min(locks_want, BTREE_MAX_DEPTH);
|
|
|
|
if (locks_want > path->locks_want) {
|
|
|
|
path->locks_want = locks_want;
|
|
|
|
btree_path_get_locks(trans, path, true, _THIS_IP_);
|
|
|
|
}
|
|
|
|
|
|
|
|
return path;
|
|
|
|
}
|
|
|
|
|
|
|
|
inline struct bkey_s_c bch2_btree_path_peek_slot(struct btree_path *path, struct bkey *u)
|
|
|
|
{
|
|
|
|
|
|
|
|
struct bkey_s_c k;
|
|
|
|
|
|
|
|
BUG_ON(path->uptodate != BTREE_ITER_UPTODATE);
|
|
|
|
|
|
|
|
if (!path->cached) {
|
|
|
|
struct btree_path_level *l = path_l(path);
|
|
|
|
struct bkey_packed *_k =
|
|
|
|
bch2_btree_node_iter_peek_all(&l->iter, l->b);
|
|
|
|
|
|
|
|
k = _k ? bkey_disassemble(l->b, _k, u) : bkey_s_c_null;
|
|
|
|
|
|
|
|
EBUG_ON(k.k && bkey_deleted(k.k) && bpos_cmp(k.k->p, path->pos) == 0);
|
|
|
|
|
|
|
|
if (!k.k || bpos_cmp(path->pos, k.k->p))
|
|
|
|
goto hole;
|
|
|
|
} else {
|
|
|
|
struct bkey_cached *ck = (void *) path->l[0].b;
|
|
|
|
|
|
|
|
EBUG_ON(path->btree_id != ck->key.btree_id ||
|
|
|
|
bkey_cmp(path->pos, ck->key.pos));
|
|
|
|
|
|
|
|
/* BTREE_ITER_CACHED_NOFILL? */
|
|
|
|
if (unlikely(!ck->valid))
|
|
|
|
goto hole;
|
|
|
|
|
|
|
|
k = bkey_i_to_s_c(ck->k);
|
|
|
|
}
|
|
|
|
|
|
|
|
return k;
|
|
|
|
hole:
|
|
|
|
bkey_init(u);
|
|
|
|
u->p = path->pos;
|
|
|
|
return (struct bkey_s_c) { u, NULL };
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Btree iterators: */
|
|
|
|
|
2021-09-08 03:34:16 +08:00
|
|
|
int __must_check
|
|
|
|
__bch2_btree_iter_traverse(struct btree_iter *iter)
|
|
|
|
{
|
|
|
|
return bch2_btree_path_traverse(iter->trans, iter->path, iter->flags);
|
|
|
|
}
|
|
|
|
|
2021-03-24 09:22:50 +08:00
|
|
|
int __must_check
|
|
|
|
bch2_btree_iter_traverse(struct btree_iter *iter)
|
|
|
|
{
|
2021-06-05 05:17:45 +08:00
|
|
|
int ret;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->path = btree_path_set_pos(iter->trans, iter->path,
|
|
|
|
btree_iter_search_key(iter),
|
|
|
|
iter->flags & BTREE_ITER_INTENT);
|
2021-03-24 09:22:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
ret = bch2_btree_path_traverse(iter->trans, iter->path, iter->flags);
|
2021-06-05 05:17:45 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->path->should_be_locked = true;
|
2021-06-05 05:17:45 +08:00
|
|
|
return 0;
|
2021-03-24 09:22:50 +08:00
|
|
|
}
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
/* Iterate across nodes (leaf and interior nodes) */
|
|
|
|
|
|
|
|
struct btree *bch2_btree_iter_peek_node(struct btree_iter *iter)
|
|
|
|
{
|
2021-10-08 02:54:50 +08:00
|
|
|
struct btree_trans *trans = iter->trans;
|
2021-08-31 03:54:41 +08:00
|
|
|
struct btree *b = NULL;
|
2017-03-17 14:18:50 +08:00
|
|
|
int ret;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(iter->path->cached);
|
2021-02-12 10:57:32 +08:00
|
|
|
bch2_btree_iter_verify(iter);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-10-08 02:54:50 +08:00
|
|
|
ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
|
2017-03-17 14:18:50 +08:00
|
|
|
if (ret)
|
2021-10-20 02:20:50 +08:00
|
|
|
goto err;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
b = btree_path_node(iter->path, iter->path->level);
|
2017-03-17 14:18:50 +08:00
|
|
|
if (!b)
|
2021-08-31 03:54:41 +08:00
|
|
|
goto out;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-03-05 05:20:16 +08:00
|
|
|
BUG_ON(bpos_cmp(b->key.k.p, iter->pos) < 0);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:54:41 +08:00
|
|
|
bkey_init(&iter->k);
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->k.p = iter->pos = b->key.k.p;
|
2021-10-08 02:54:50 +08:00
|
|
|
|
|
|
|
iter->path = btree_path_set_pos(trans, iter->path, b->key.k.p,
|
|
|
|
iter->flags & BTREE_ITER_INTENT);
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->path->should_be_locked = true;
|
2021-10-08 02:54:50 +08:00
|
|
|
BUG_ON(iter->path->uptodate);
|
2021-08-31 03:54:41 +08:00
|
|
|
out:
|
|
|
|
bch2_btree_iter_verify_entry_exit(iter);
|
|
|
|
bch2_btree_iter_verify(iter);
|
2020-02-19 05:17:55 +08:00
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
return b;
|
2021-10-20 02:20:50 +08:00
|
|
|
err:
|
|
|
|
b = ERR_PTR(ret);
|
|
|
|
goto out;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2020-02-19 05:17:55 +08:00
|
|
|
struct btree *bch2_btree_iter_next_node(struct btree_iter *iter)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_trans *trans = iter->trans;
|
|
|
|
struct btree_path *path = iter->path;
|
2021-08-31 03:54:41 +08:00
|
|
|
struct btree *b = NULL;
|
2021-10-25 04:55:17 +08:00
|
|
|
unsigned l;
|
2017-03-17 14:18:50 +08:00
|
|
|
int ret;
|
|
|
|
|
2021-10-25 04:55:17 +08:00
|
|
|
BUG_ON(trans->restarted);
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(iter->path->cached);
|
2021-02-12 10:57:32 +08:00
|
|
|
bch2_btree_iter_verify(iter);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-10-25 04:55:17 +08:00
|
|
|
/* already at end? */
|
2021-08-31 03:18:31 +08:00
|
|
|
if (!btree_path_node(path, path->level))
|
2021-10-25 04:55:17 +08:00
|
|
|
return NULL;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-10-25 04:55:17 +08:00
|
|
|
/* got to end? */
|
|
|
|
if (!btree_path_node(path, path->level + 1)) {
|
|
|
|
btree_node_unlock(path, path->level);
|
|
|
|
path->l[path->level].b = BTREE_ITER_NO_NODE_UP;
|
|
|
|
path->level++;
|
|
|
|
return NULL;
|
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-10-25 04:55:17 +08:00
|
|
|
if (!bch2_btree_node_relock(trans, path, path->level + 1)) {
|
|
|
|
__bch2_btree_path_unlock(path);
|
|
|
|
path->l[path->level].b = BTREE_ITER_NO_NODE_GET_LOCKS;
|
|
|
|
path->l[path->level + 1].b = BTREE_ITER_NO_NODE_GET_LOCKS;
|
|
|
|
btree_trans_restart(trans);
|
|
|
|
ret = -EINTR;
|
2021-10-20 02:20:50 +08:00
|
|
|
goto err;
|
2021-10-25 04:55:17 +08:00
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-10-25 04:55:17 +08:00
|
|
|
b = btree_path_node(path, path->level + 1);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-10-25 04:55:17 +08:00
|
|
|
if (!bpos_cmp(iter->pos, b->key.k.p)) {
|
|
|
|
btree_node_unlock(path, path->level);
|
|
|
|
path->l[path->level].b = BTREE_ITER_NO_NODE_UP;
|
|
|
|
path->level++;
|
|
|
|
} else {
|
2017-03-17 14:18:50 +08:00
|
|
|
/*
|
|
|
|
* Haven't gotten to the end of the parent node: go back down to
|
|
|
|
* the next child node
|
|
|
|
*/
|
2021-08-31 03:18:31 +08:00
|
|
|
path = iter->path =
|
|
|
|
btree_path_set_pos(trans, path, bpos_successor(iter->pos),
|
|
|
|
iter->flags & BTREE_ITER_INTENT);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
path->level = iter->min_depth;
|
2021-10-25 04:55:17 +08:00
|
|
|
|
|
|
|
for (l = path->level + 1; l < BTREE_MAX_DEPTH; l++)
|
|
|
|
if (btree_lock_want(path, l) == BTREE_NODE_UNLOCKED)
|
|
|
|
btree_node_unlock(path, l);
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
|
2021-03-22 06:09:02 +08:00
|
|
|
bch2_btree_iter_verify(iter);
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
ret = bch2_btree_path_traverse(trans, path, iter->flags);
|
2021-10-20 02:20:50 +08:00
|
|
|
if (ret)
|
|
|
|
goto err;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
b = path->l[path->level].b;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:54:41 +08:00
|
|
|
bkey_init(&iter->k);
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->k.p = iter->pos = b->key.k.p;
|
2021-10-08 02:54:50 +08:00
|
|
|
|
|
|
|
iter->path = btree_path_set_pos(trans, iter->path, b->key.k.p,
|
|
|
|
iter->flags & BTREE_ITER_INTENT);
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->path->should_be_locked = true;
|
2021-10-08 02:54:50 +08:00
|
|
|
BUG_ON(iter->path->uptodate);
|
2021-08-31 03:54:41 +08:00
|
|
|
out:
|
|
|
|
bch2_btree_iter_verify_entry_exit(iter);
|
|
|
|
bch2_btree_iter_verify(iter);
|
2020-02-19 05:17:55 +08:00
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
return b;
|
2021-10-20 02:20:50 +08:00
|
|
|
err:
|
|
|
|
b = ERR_PTR(ret);
|
|
|
|
goto out;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Iterate across keys (in leaf nodes only) */
|
|
|
|
|
2021-03-22 04:55:25 +08:00
|
|
|
inline bool bch2_btree_iter_advance(struct btree_iter *iter)
|
2021-02-08 10:28:58 +08:00
|
|
|
{
|
2021-02-08 10:11:49 +08:00
|
|
|
struct bpos pos = iter->k.p;
|
2021-10-29 04:34:17 +08:00
|
|
|
bool ret = (iter->flags & BTREE_ITER_ALL_SNAPSHOTS
|
|
|
|
? bpos_cmp(pos, SPOS_MAX)
|
|
|
|
: bkey_cmp(pos, SPOS_MAX)) != 0;
|
2021-02-08 10:11:49 +08:00
|
|
|
|
2021-02-12 10:57:32 +08:00
|
|
|
if (ret && !(iter->flags & BTREE_ITER_IS_EXTENTS))
|
bcachefs: Start using bpos.snapshot field
This patch starts treating the bpos.snapshot field like part of the key
in the btree code:
* bpos_successor() and bpos_predecessor() now include the snapshot field
* Keys in btrees that will be using snapshots (extents, inodes, dirents
and xattrs) now always have their snapshot field set to U32_MAX
The btree iterator code gets a new flag, BTREE_ITER_ALL_SNAPSHOTS, that
determines whether we're iterating over keys in all snapshots or not -
internally, this controlls whether bkey_(successor|predecessor)
increment/decrement the snapshot field, or only the higher bits of the
key.
We add a new member to struct btree_iter, iter->snapshot: when
BTREE_ITER_ALL_SNAPSHOTS is not set, iter->pos.snapshot should always
equal iter->snapshot, which will be 0 for btrees that don't use
snapshots, and alsways U32_MAX for btrees that will use snapshots
(until we enable snapshot creation).
This patch also introduces a new metadata version number, and compat
code for reading from/writing to older versions - this isn't a forced
upgrade (yet).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2021-03-25 06:02:16 +08:00
|
|
|
pos = bkey_successor(iter, pos);
|
2021-02-08 10:11:49 +08:00
|
|
|
bch2_btree_iter_set_pos(iter, pos);
|
2021-02-12 10:57:32 +08:00
|
|
|
return ret;
|
2021-02-08 10:11:49 +08:00
|
|
|
}
|
|
|
|
|
2021-03-22 04:55:25 +08:00
|
|
|
inline bool bch2_btree_iter_rewind(struct btree_iter *iter)
|
2021-02-08 10:11:49 +08:00
|
|
|
{
|
|
|
|
struct bpos pos = bkey_start_pos(&iter->k);
|
2021-07-15 11:35:11 +08:00
|
|
|
bool ret = (iter->flags & BTREE_ITER_ALL_SNAPSHOTS
|
|
|
|
? bpos_cmp(pos, POS_MIN)
|
|
|
|
: bkey_cmp(pos, POS_MIN)) != 0;
|
2021-02-08 10:11:49 +08:00
|
|
|
|
2021-02-12 10:57:32 +08:00
|
|
|
if (ret && !(iter->flags & BTREE_ITER_IS_EXTENTS))
|
bcachefs: Start using bpos.snapshot field
This patch starts treating the bpos.snapshot field like part of the key
in the btree code:
* bpos_successor() and bpos_predecessor() now include the snapshot field
* Keys in btrees that will be using snapshots (extents, inodes, dirents
and xattrs) now always have their snapshot field set to U32_MAX
The btree iterator code gets a new flag, BTREE_ITER_ALL_SNAPSHOTS, that
determines whether we're iterating over keys in all snapshots or not -
internally, this controlls whether bkey_(successor|predecessor)
increment/decrement the snapshot field, or only the higher bits of the
key.
We add a new member to struct btree_iter, iter->snapshot: when
BTREE_ITER_ALL_SNAPSHOTS is not set, iter->pos.snapshot should always
equal iter->snapshot, which will be 0 for btrees that don't use
snapshots, and alsways U32_MAX for btrees that will use snapshots
(until we enable snapshot creation).
This patch also introduces a new metadata version number, and compat
code for reading from/writing to older versions - this isn't a forced
upgrade (yet).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2021-03-25 06:02:16 +08:00
|
|
|
pos = bkey_predecessor(iter, pos);
|
2021-02-08 10:11:49 +08:00
|
|
|
bch2_btree_iter_set_pos(iter, pos);
|
2021-02-12 10:57:32 +08:00
|
|
|
return ret;
|
2021-02-08 10:28:58 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
struct bkey_i *__bch2_btree_trans_peek_updates(struct btree_iter *iter)
|
2021-03-22 07:43:31 +08:00
|
|
|
{
|
|
|
|
struct btree_insert_entry *i;
|
2021-06-04 12:29:49 +08:00
|
|
|
struct bkey_i *ret = NULL;
|
2021-03-22 07:43:31 +08:00
|
|
|
|
2021-06-02 12:18:34 +08:00
|
|
|
trans_for_each_update(iter->trans, i) {
|
2021-06-04 12:29:49 +08:00
|
|
|
if (i->btree_id < iter->btree_id)
|
|
|
|
continue;
|
|
|
|
if (i->btree_id > iter->btree_id)
|
2021-03-22 07:43:31 +08:00
|
|
|
break;
|
2021-08-31 03:18:31 +08:00
|
|
|
if (bpos_cmp(i->k->k.p, iter->path->pos) < 0)
|
2021-06-04 12:29:49 +08:00
|
|
|
continue;
|
|
|
|
if (!ret || bpos_cmp(i->k->k.p, ret->k.p) < 0)
|
|
|
|
ret = i->k;
|
|
|
|
}
|
2021-03-22 07:43:31 +08:00
|
|
|
|
2021-06-04 12:29:49 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* bch2_btree_iter_peek: returns first key greater than or equal to iterator's
|
|
|
|
* current position
|
|
|
|
*/
|
|
|
|
struct bkey_s_c bch2_btree_iter_peek(struct btree_iter *iter)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_trans *trans = iter->trans;
|
2021-03-22 07:43:31 +08:00
|
|
|
struct bpos search_key = btree_iter_search_key(iter);
|
2021-04-30 09:44:05 +08:00
|
|
|
struct bkey_i *next_update;
|
2017-03-17 14:18:50 +08:00
|
|
|
struct bkey_s_c k;
|
2021-08-31 03:18:31 +08:00
|
|
|
int ret, cmp;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(iter->path->cached || iter->path->level);
|
2021-02-12 10:57:32 +08:00
|
|
|
bch2_btree_iter_verify(iter);
|
|
|
|
bch2_btree_iter_verify_entry_exit(iter);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
while (1) {
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->path = btree_path_set_pos(trans, iter->path, search_key,
|
|
|
|
iter->flags & BTREE_ITER_INTENT);
|
2021-08-08 06:19:33 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
|
2021-08-25 04:54:36 +08:00
|
|
|
if (unlikely(ret)) {
|
|
|
|
/* ensure that iter->k is consistent with iter->pos: */
|
|
|
|
bch2_btree_iter_set_pos(iter, iter->pos);
|
|
|
|
k = bkey_s_c_err(ret);
|
|
|
|
goto out;
|
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-08 06:21:35 +08:00
|
|
|
next_update = btree_trans_peek_updates(iter);
|
2021-08-31 03:18:31 +08:00
|
|
|
k = btree_path_level_peek_all(trans->c, &iter->path->l[0], &iter->k);
|
2021-08-25 04:54:36 +08:00
|
|
|
|
|
|
|
/* * In the btree, deleted keys sort before non deleted: */
|
|
|
|
if (k.k && bkey_deleted(k.k) &&
|
|
|
|
(!next_update ||
|
|
|
|
bpos_cmp(k.k->p, next_update->k.p) <= 0)) {
|
|
|
|
search_key = k.k->p;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2021-03-22 07:43:31 +08:00
|
|
|
if (next_update &&
|
2021-08-25 04:54:36 +08:00
|
|
|
bpos_cmp(next_update->k.p,
|
2021-08-31 03:18:31 +08:00
|
|
|
k.k ? k.k->p : iter->path->l[0].b->key.k.p) <= 0) {
|
2021-06-04 12:29:49 +08:00
|
|
|
iter->k = next_update->k;
|
2021-03-22 07:43:31 +08:00
|
|
|
k = bkey_i_to_s_c(next_update);
|
2021-06-04 12:29:49 +08:00
|
|
|
}
|
2021-03-22 07:43:31 +08:00
|
|
|
|
|
|
|
if (likely(k.k)) {
|
2021-03-05 11:29:25 +08:00
|
|
|
/*
|
|
|
|
* We can never have a key in a leaf node at POS_MAX, so
|
|
|
|
* we don't have to check these successor() calls:
|
|
|
|
*/
|
|
|
|
if ((iter->flags & BTREE_ITER_FILTER_SNAPSHOTS) &&
|
|
|
|
!bch2_snapshot_is_ancestor(trans->c,
|
|
|
|
iter->snapshot,
|
|
|
|
k.k->p.snapshot)) {
|
|
|
|
search_key = bpos_successor(k.k->p);
|
|
|
|
continue;
|
|
|
|
}
|
2021-08-25 04:54:36 +08:00
|
|
|
|
2021-03-05 11:29:25 +08:00
|
|
|
if (bkey_whiteout(k.k) &&
|
|
|
|
!(iter->flags & BTREE_ITER_ALL_SNAPSHOTS)) {
|
|
|
|
search_key = bkey_successor(iter, k.k->p);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
break;
|
2021-08-31 03:18:31 +08:00
|
|
|
} else if (likely(bpos_cmp(iter->path->l[0].b->key.k.p, SPOS_MAX))) {
|
2021-08-25 04:54:36 +08:00
|
|
|
/* Advance to next leaf node: */
|
2021-08-31 03:18:31 +08:00
|
|
|
search_key = bpos_successor(iter->path->l[0].b->key.k.p);
|
2021-08-25 04:54:36 +08:00
|
|
|
} else {
|
|
|
|
/* End of btree: */
|
2021-08-08 06:19:33 +08:00
|
|
|
bch2_btree_iter_set_pos(iter, SPOS_MAX);
|
|
|
|
k = bkey_s_c_null;
|
|
|
|
goto out;
|
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2021-03-22 07:43:31 +08:00
|
|
|
* iter->pos should be mononotically increasing, and always be equal to
|
|
|
|
* the key we just returned - except extents can straddle iter->pos:
|
2017-03-17 14:18:50 +08:00
|
|
|
*/
|
2021-04-10 04:52:30 +08:00
|
|
|
if (!(iter->flags & BTREE_ITER_IS_EXTENTS))
|
|
|
|
iter->pos = k.k->p;
|
|
|
|
else if (bkey_cmp(bkey_start_pos(k.k), iter->pos) > 0)
|
2017-03-17 14:18:50 +08:00
|
|
|
iter->pos = bkey_start_pos(k.k);
|
2021-08-31 03:18:31 +08:00
|
|
|
|
2021-03-05 11:29:25 +08:00
|
|
|
if (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)
|
|
|
|
iter->pos.snapshot = iter->snapshot;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
cmp = bpos_cmp(k.k->p, iter->path->pos);
|
|
|
|
if (cmp) {
|
2021-09-08 01:55:33 +08:00
|
|
|
iter->path = bch2_btree_path_make_mut(trans, iter->path,
|
|
|
|
iter->flags & BTREE_ITER_INTENT);
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->path->pos = k.k->p;
|
|
|
|
trans->paths_sorted = false;
|
|
|
|
}
|
2021-08-08 06:19:33 +08:00
|
|
|
out:
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->path->should_be_locked = true;
|
|
|
|
|
2021-02-12 10:57:32 +08:00
|
|
|
bch2_btree_iter_verify_entry_exit(iter);
|
|
|
|
bch2_btree_iter_verify(iter);
|
2021-03-05 11:29:25 +08:00
|
|
|
ret = bch2_btree_iter_verify_ret(iter, k);
|
|
|
|
if (unlikely(ret))
|
|
|
|
return bkey_s_c_err(ret);
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
return k;
|
|
|
|
}
|
|
|
|
|
2019-09-08 07:19:57 +08:00
|
|
|
/**
|
|
|
|
* bch2_btree_iter_next: returns first key greater than iterator's current
|
|
|
|
* position
|
|
|
|
*/
|
2017-03-17 14:18:50 +08:00
|
|
|
struct bkey_s_c bch2_btree_iter_next(struct btree_iter *iter)
|
|
|
|
{
|
2021-03-22 04:55:25 +08:00
|
|
|
if (!bch2_btree_iter_advance(iter))
|
2020-02-19 05:17:55 +08:00
|
|
|
return bkey_s_c_null;
|
2019-09-08 07:19:57 +08:00
|
|
|
|
2020-02-19 05:17:55 +08:00
|
|
|
return bch2_btree_iter_peek(iter);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2019-09-08 05:17:21 +08:00
|
|
|
/**
|
|
|
|
* bch2_btree_iter_peek_prev: returns first key less than or equal to
|
|
|
|
* iterator's current position
|
|
|
|
*/
|
|
|
|
struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *iter)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_trans *trans = iter->trans;
|
2021-08-08 06:19:33 +08:00
|
|
|
struct bpos search_key = iter->pos;
|
2021-03-05 11:29:25 +08:00
|
|
|
struct btree_path *saved_path = NULL;
|
2017-03-17 14:18:50 +08:00
|
|
|
struct bkey_s_c k;
|
2021-03-05 11:29:25 +08:00
|
|
|
struct bkey saved_k;
|
|
|
|
const struct bch_val *saved_v;
|
2017-03-17 14:18:50 +08:00
|
|
|
int ret;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(iter->path->cached || iter->path->level);
|
2021-06-04 12:29:49 +08:00
|
|
|
EBUG_ON(iter->flags & BTREE_ITER_WITH_UPDATES);
|
2021-02-12 10:57:32 +08:00
|
|
|
bch2_btree_iter_verify(iter);
|
|
|
|
bch2_btree_iter_verify_entry_exit(iter);
|
|
|
|
|
2021-03-05 11:29:25 +08:00
|
|
|
if (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)
|
|
|
|
search_key.snapshot = U32_MAX;
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
while (1) {
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->path = btree_path_set_pos(trans, iter->path, search_key,
|
|
|
|
iter->flags & BTREE_ITER_INTENT);
|
2021-08-08 06:19:33 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
|
2021-02-12 10:57:32 +08:00
|
|
|
if (unlikely(ret)) {
|
2021-08-25 04:54:36 +08:00
|
|
|
/* ensure that iter->k is consistent with iter->pos: */
|
|
|
|
bch2_btree_iter_set_pos(iter, iter->pos);
|
2021-02-12 10:57:32 +08:00
|
|
|
k = bkey_s_c_err(ret);
|
2021-08-25 04:54:36 +08:00
|
|
|
goto out;
|
2021-02-12 10:57:32 +08:00
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
k = btree_path_level_peek(trans, iter->path,
|
|
|
|
&iter->path->l[0], &iter->k);
|
2021-02-08 10:11:49 +08:00
|
|
|
if (!k.k ||
|
|
|
|
((iter->flags & BTREE_ITER_IS_EXTENTS)
|
2021-03-05 11:29:25 +08:00
|
|
|
? bpos_cmp(bkey_start_pos(k.k), search_key) >= 0
|
|
|
|
: bpos_cmp(k.k->p, search_key) > 0))
|
2021-08-31 03:18:31 +08:00
|
|
|
k = btree_path_level_prev(trans, iter->path,
|
|
|
|
&iter->path->l[0], &iter->k);
|
2019-09-08 05:17:21 +08:00
|
|
|
|
2021-08-25 04:54:36 +08:00
|
|
|
if (likely(k.k)) {
|
2021-03-05 11:29:25 +08:00
|
|
|
if (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS) {
|
|
|
|
if (k.k->p.snapshot == iter->snapshot)
|
|
|
|
goto got_key;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If we have a saved candidate, and we're no
|
|
|
|
* longer at the same _key_ (not pos), return
|
|
|
|
* that candidate
|
|
|
|
*/
|
|
|
|
if (saved_path && bkey_cmp(k.k->p, saved_k.p)) {
|
|
|
|
bch2_path_put(trans, iter->path,
|
|
|
|
iter->flags & BTREE_ITER_INTENT);
|
|
|
|
iter->path = saved_path;
|
|
|
|
saved_path = NULL;
|
|
|
|
iter->k = saved_k;
|
|
|
|
k.v = saved_v;
|
|
|
|
goto got_key;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (bch2_snapshot_is_ancestor(iter->trans->c,
|
|
|
|
iter->snapshot,
|
|
|
|
k.k->p.snapshot)) {
|
|
|
|
if (saved_path)
|
|
|
|
bch2_path_put(trans, saved_path,
|
|
|
|
iter->flags & BTREE_ITER_INTENT);
|
|
|
|
saved_path = btree_path_clone(trans, iter->path,
|
|
|
|
iter->flags & BTREE_ITER_INTENT);
|
|
|
|
saved_k = *k.k;
|
|
|
|
saved_v = k.v;
|
|
|
|
}
|
|
|
|
|
|
|
|
search_key = bpos_predecessor(k.k->p);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
got_key:
|
|
|
|
if (bkey_whiteout(k.k) &&
|
|
|
|
!(iter->flags & BTREE_ITER_ALL_SNAPSHOTS)) {
|
|
|
|
search_key = bkey_predecessor(iter, k.k->p);
|
|
|
|
if (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)
|
|
|
|
search_key.snapshot = U32_MAX;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
break;
|
2021-08-31 03:18:31 +08:00
|
|
|
} else if (likely(bpos_cmp(iter->path->l[0].b->data->min_key, POS_MIN))) {
|
2021-08-25 04:54:36 +08:00
|
|
|
/* Advance to previous leaf node: */
|
2021-08-31 03:18:31 +08:00
|
|
|
search_key = bpos_predecessor(iter->path->l[0].b->data->min_key);
|
2021-08-25 04:54:36 +08:00
|
|
|
} else {
|
|
|
|
/* Start of btree: */
|
2021-08-08 06:19:33 +08:00
|
|
|
bch2_btree_iter_set_pos(iter, POS_MIN);
|
2021-02-12 10:57:32 +08:00
|
|
|
k = bkey_s_c_null;
|
2021-08-25 04:54:36 +08:00
|
|
|
goto out;
|
2021-02-12 10:57:32 +08:00
|
|
|
}
|
2019-09-08 05:17:21 +08:00
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-03-05 11:40:41 +08:00
|
|
|
EBUG_ON(bkey_cmp(bkey_start_pos(k.k), iter->pos) > 0);
|
2021-02-08 10:11:49 +08:00
|
|
|
|
|
|
|
/* Extents can straddle iter->pos: */
|
2021-03-05 11:40:41 +08:00
|
|
|
if (bkey_cmp(k.k->p, iter->pos) < 0)
|
2021-02-08 10:11:49 +08:00
|
|
|
iter->pos = k.k->p;
|
2021-03-05 11:29:25 +08:00
|
|
|
|
|
|
|
if (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)
|
|
|
|
iter->pos.snapshot = iter->snapshot;
|
2021-02-12 10:57:32 +08:00
|
|
|
out:
|
2021-03-05 11:29:25 +08:00
|
|
|
if (saved_path)
|
|
|
|
bch2_path_put(trans, saved_path, iter->flags & BTREE_ITER_INTENT);
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->path->should_be_locked = true;
|
|
|
|
|
2021-02-12 10:57:32 +08:00
|
|
|
bch2_btree_iter_verify_entry_exit(iter);
|
|
|
|
bch2_btree_iter_verify(iter);
|
2021-08-31 03:18:31 +08:00
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
return k;
|
|
|
|
}
|
|
|
|
|
2019-09-08 05:17:21 +08:00
|
|
|
/**
|
|
|
|
* bch2_btree_iter_prev: returns first key less than iterator's current
|
|
|
|
* position
|
|
|
|
*/
|
|
|
|
struct bkey_s_c bch2_btree_iter_prev(struct btree_iter *iter)
|
|
|
|
{
|
2021-03-22 04:55:25 +08:00
|
|
|
if (!bch2_btree_iter_rewind(iter))
|
2020-02-19 05:17:55 +08:00
|
|
|
return bkey_s_c_null;
|
2019-09-08 05:17:21 +08:00
|
|
|
|
2020-02-19 05:17:55 +08:00
|
|
|
return bch2_btree_iter_peek_prev(iter);
|
2019-09-08 05:17:21 +08:00
|
|
|
}
|
|
|
|
|
2020-03-14 09:41:22 +08:00
|
|
|
struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *iter)
|
2016-07-22 11:05:06 +08:00
|
|
|
{
|
2021-08-25 09:30:06 +08:00
|
|
|
struct btree_trans *trans = iter->trans;
|
2021-06-11 08:15:50 +08:00
|
|
|
struct bpos search_key;
|
2016-07-22 11:05:06 +08:00
|
|
|
struct bkey_s_c k;
|
2020-03-14 09:41:22 +08:00
|
|
|
int ret;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(iter->path->level);
|
2021-02-12 10:57:32 +08:00
|
|
|
bch2_btree_iter_verify(iter);
|
|
|
|
bch2_btree_iter_verify_entry_exit(iter);
|
|
|
|
|
2022-12-31 08:15:53 +08:00
|
|
|
/* extents can't span inode numbers: */
|
|
|
|
if ((iter->flags & BTREE_ITER_IS_EXTENTS) &&
|
2021-06-11 08:15:50 +08:00
|
|
|
unlikely(iter->pos.offset == KEY_OFFSET_MAX)) {
|
2022-12-31 08:15:53 +08:00
|
|
|
if (iter->pos.inode == KEY_INODE_MAX)
|
|
|
|
return bkey_s_c_null;
|
2020-03-14 09:41:22 +08:00
|
|
|
|
2022-12-31 08:15:53 +08:00
|
|
|
bch2_btree_iter_set_pos(iter, bpos_nosnap_successor(iter->pos));
|
|
|
|
}
|
2021-02-11 05:13:57 +08:00
|
|
|
|
2021-06-11 08:15:50 +08:00
|
|
|
search_key = btree_iter_search_key(iter);
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->path = btree_path_set_pos(trans, iter->path, search_key,
|
|
|
|
iter->flags & BTREE_ITER_INTENT);
|
2021-06-11 08:15:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
|
2020-03-14 09:41:22 +08:00
|
|
|
if (unlikely(ret))
|
|
|
|
return bkey_s_c_err(ret);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-03-05 11:29:25 +08:00
|
|
|
if ((iter->flags & BTREE_ITER_CACHED) ||
|
|
|
|
!(iter->flags & (BTREE_ITER_IS_EXTENTS|BTREE_ITER_FILTER_SNAPSHOTS))) {
|
2021-06-11 08:15:50 +08:00
|
|
|
struct bkey_i *next_update;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-08 06:21:35 +08:00
|
|
|
next_update = btree_trans_peek_updates(iter);
|
2022-12-31 08:15:53 +08:00
|
|
|
if (next_update &&
|
2021-08-31 03:18:31 +08:00
|
|
|
!bpos_cmp(next_update->k.p, iter->pos)) {
|
2022-12-31 08:15:53 +08:00
|
|
|
iter->k = next_update->k;
|
|
|
|
k = bkey_i_to_s_c(next_update);
|
2021-08-31 03:18:31 +08:00
|
|
|
} else {
|
|
|
|
k = bch2_btree_path_peek_slot(iter->path, &iter->k);
|
2022-12-31 08:15:53 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
if (!k.k ||
|
|
|
|
((iter->flags & BTREE_ITER_ALL_SNAPSHOTS)
|
|
|
|
? bpos_cmp(iter->pos, k.k->p)
|
|
|
|
: bkey_cmp(iter->pos, k.k->p))) {
|
|
|
|
bkey_init(&iter->k);
|
|
|
|
iter->k.p = iter->pos;
|
|
|
|
k = (struct bkey_s_c) { &iter->k, NULL };
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
struct bpos next;
|
|
|
|
|
2022-12-31 08:15:53 +08:00
|
|
|
if (iter->flags & BTREE_ITER_INTENT) {
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_iter iter2;
|
2022-12-31 08:15:53 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_trans_copy_iter(&iter2, iter);
|
|
|
|
k = bch2_btree_iter_peek(&iter2);
|
2022-12-31 08:15:53 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (k.k && !bkey_err(k)) {
|
|
|
|
iter->k = iter2.k;
|
|
|
|
k.k = &iter->k;
|
|
|
|
}
|
|
|
|
bch2_trans_iter_exit(trans, &iter2);
|
2022-12-31 08:15:53 +08:00
|
|
|
} else {
|
|
|
|
struct bpos pos = iter->pos;
|
|
|
|
|
|
|
|
k = bch2_btree_iter_peek(iter);
|
|
|
|
iter->pos = pos;
|
|
|
|
}
|
2022-12-31 08:15:53 +08:00
|
|
|
|
|
|
|
if (unlikely(bkey_err(k)))
|
|
|
|
return k;
|
|
|
|
|
|
|
|
next = k.k ? bkey_start_pos(k.k) : POS_MAX;
|
|
|
|
|
|
|
|
if (bkey_cmp(iter->pos, next) < 0) {
|
|
|
|
bkey_init(&iter->k);
|
|
|
|
iter->k.p = iter->pos;
|
2021-03-05 11:29:25 +08:00
|
|
|
|
|
|
|
if (iter->flags & BTREE_ITER_IS_EXTENTS) {
|
|
|
|
bch2_key_resize(&iter->k,
|
|
|
|
min_t(u64, KEY_SIZE_MAX,
|
|
|
|
(next.inode == iter->pos.inode
|
|
|
|
? next.offset
|
|
|
|
: KEY_OFFSET_MAX) -
|
|
|
|
iter->pos.offset));
|
|
|
|
EBUG_ON(!iter->k.size);
|
|
|
|
}
|
2022-12-31 08:15:53 +08:00
|
|
|
|
|
|
|
k = (struct bkey_s_c) { &iter->k, NULL };
|
|
|
|
}
|
2016-07-22 11:05:06 +08:00
|
|
|
}
|
2019-08-18 03:17:09 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->path->should_be_locked = true;
|
|
|
|
|
2021-02-12 10:57:32 +08:00
|
|
|
bch2_btree_iter_verify_entry_exit(iter);
|
|
|
|
bch2_btree_iter_verify(iter);
|
2021-03-05 11:29:25 +08:00
|
|
|
ret = bch2_btree_iter_verify_ret(iter, k);
|
|
|
|
if (unlikely(ret))
|
|
|
|
return bkey_s_c_err(ret);
|
2021-06-05 05:17:45 +08:00
|
|
|
|
2019-08-18 03:17:09 +08:00
|
|
|
return k;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
struct bkey_s_c bch2_btree_iter_next_slot(struct btree_iter *iter)
|
|
|
|
{
|
2021-03-22 04:55:25 +08:00
|
|
|
if (!bch2_btree_iter_advance(iter))
|
2020-02-19 05:17:55 +08:00
|
|
|
return bkey_s_c_null;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2020-02-19 05:17:55 +08:00
|
|
|
return bch2_btree_iter_peek_slot(iter);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-03-03 11:45:28 +08:00
|
|
|
struct bkey_s_c bch2_btree_iter_prev_slot(struct btree_iter *iter)
|
|
|
|
{
|
2021-03-22 04:55:25 +08:00
|
|
|
if (!bch2_btree_iter_rewind(iter))
|
2021-03-03 11:45:28 +08:00
|
|
|
return bkey_s_c_null;
|
|
|
|
|
|
|
|
return bch2_btree_iter_peek_slot(iter);
|
|
|
|
}
|
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
/* new transactional stuff: */
|
|
|
|
|
2021-06-13 03:45:45 +08:00
|
|
|
#ifdef CONFIG_BCACHEFS_DEBUG
|
|
|
|
static void btree_trans_verify_sorted_refs(struct btree_trans *trans)
|
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path;
|
2021-06-13 03:45:45 +08:00
|
|
|
unsigned i;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
BUG_ON(trans->nr_sorted != hweight64(trans->paths_allocated));
|
2021-06-13 03:45:45 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, path) {
|
|
|
|
BUG_ON(path->sorted_idx >= trans->nr_sorted);
|
|
|
|
BUG_ON(trans->sorted[path->sorted_idx] != path->idx);
|
2021-06-13 03:45:45 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0; i < trans->nr_sorted; i++) {
|
|
|
|
unsigned idx = trans->sorted[i];
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(!(trans->paths_allocated & (1ULL << idx)));
|
|
|
|
BUG_ON(trans->paths[idx].sorted_idx != i);
|
2021-06-13 03:45:45 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void btree_trans_verify_sorted(struct btree_trans *trans)
|
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path, *prev = NULL;
|
2021-06-13 03:45:45 +08:00
|
|
|
unsigned i;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path_inorder(trans, path, i) {
|
|
|
|
BUG_ON(prev && btree_path_cmp(prev, path) > 0);
|
|
|
|
prev = path;
|
2021-06-13 03:45:45 +08:00
|
|
|
}
|
|
|
|
}
|
2021-09-04 05:18:57 +08:00
|
|
|
#else
|
|
|
|
static inline void btree_trans_verify_sorted_refs(struct btree_trans *trans) {}
|
|
|
|
static inline void btree_trans_verify_sorted(struct btree_trans *trans) {}
|
|
|
|
#endif
|
2021-06-13 03:45:45 +08:00
|
|
|
|
2021-09-04 05:18:57 +08:00
|
|
|
void __bch2_btree_trans_sort_paths(struct btree_trans *trans)
|
2021-06-13 03:45:45 +08:00
|
|
|
{
|
|
|
|
int i, l = 0, r = trans->nr_sorted, inc = 1;
|
|
|
|
bool swapped;
|
|
|
|
|
2021-09-04 05:18:57 +08:00
|
|
|
btree_trans_verify_sorted_refs(trans);
|
|
|
|
|
|
|
|
if (trans->paths_sorted)
|
|
|
|
goto out;
|
|
|
|
|
2021-06-13 03:45:45 +08:00
|
|
|
/*
|
|
|
|
* Cocktail shaker sort: this is efficient because iterators will be
|
|
|
|
* mostly sorteda.
|
|
|
|
*/
|
|
|
|
do {
|
|
|
|
swapped = false;
|
|
|
|
|
|
|
|
for (i = inc > 0 ? l : r - 2;
|
|
|
|
i + 1 < r && i >= l;
|
|
|
|
i += inc) {
|
2021-08-31 03:18:31 +08:00
|
|
|
if (btree_path_cmp(trans->paths + trans->sorted[i],
|
|
|
|
trans->paths + trans->sorted[i + 1]) > 0) {
|
2021-06-13 03:45:45 +08:00
|
|
|
swap(trans->sorted[i], trans->sorted[i + 1]);
|
2021-08-31 03:18:31 +08:00
|
|
|
trans->paths[trans->sorted[i]].sorted_idx = i;
|
|
|
|
trans->paths[trans->sorted[i + 1]].sorted_idx = i + 1;
|
2021-06-13 03:45:45 +08:00
|
|
|
swapped = true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (inc > 0)
|
|
|
|
--r;
|
|
|
|
else
|
|
|
|
l++;
|
|
|
|
inc = -inc;
|
|
|
|
} while (swapped);
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans->paths_sorted = true;
|
2021-09-04 05:18:57 +08:00
|
|
|
out:
|
2021-06-13 03:45:45 +08:00
|
|
|
btree_trans_verify_sorted(trans);
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline void btree_path_list_remove(struct btree_trans *trans,
|
|
|
|
struct btree_path *path)
|
2021-06-13 03:45:45 +08:00
|
|
|
{
|
|
|
|
unsigned i;
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
EBUG_ON(path->sorted_idx >= trans->nr_sorted);
|
2021-06-13 03:45:45 +08:00
|
|
|
#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
|
|
|
|
trans->nr_sorted--;
|
2021-08-31 03:18:31 +08:00
|
|
|
memmove_u64s_down_small(trans->sorted + path->sorted_idx,
|
|
|
|
trans->sorted + path->sorted_idx + 1,
|
|
|
|
DIV_ROUND_UP(trans->nr_sorted - path->sorted_idx, 8));
|
2021-06-13 03:45:45 +08:00
|
|
|
#else
|
2021-08-31 03:18:31 +08:00
|
|
|
array_remove_item(trans->sorted, trans->nr_sorted, path->sorted_idx);
|
2021-06-13 03:45:45 +08:00
|
|
|
#endif
|
2021-08-31 03:18:31 +08:00
|
|
|
for (i = path->sorted_idx; i < trans->nr_sorted; i++)
|
|
|
|
trans->paths[trans->sorted[i]].sorted_idx = i;
|
2021-06-13 03:45:45 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
path->sorted_idx = U8_MAX;
|
2021-06-13 03:45:45 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static inline void btree_path_list_add(struct btree_trans *trans,
|
|
|
|
struct btree_path *pos,
|
|
|
|
struct btree_path *path)
|
2021-06-13 03:45:45 +08:00
|
|
|
{
|
|
|
|
unsigned i;
|
|
|
|
|
2021-09-04 05:18:57 +08:00
|
|
|
path->sorted_idx = pos ? pos->sorted_idx + 1 : trans->nr_sorted;
|
2021-06-13 03:45:45 +08:00
|
|
|
|
|
|
|
#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
|
2021-08-31 03:18:31 +08:00
|
|
|
memmove_u64s_up_small(trans->sorted + path->sorted_idx + 1,
|
|
|
|
trans->sorted + path->sorted_idx,
|
|
|
|
DIV_ROUND_UP(trans->nr_sorted - path->sorted_idx, 8));
|
2021-06-13 03:45:45 +08:00
|
|
|
trans->nr_sorted++;
|
2021-08-31 03:18:31 +08:00
|
|
|
trans->sorted[path->sorted_idx] = path->idx;
|
2021-06-13 03:45:45 +08:00
|
|
|
#else
|
2021-08-31 03:18:31 +08:00
|
|
|
array_insert_item(trans->sorted, trans->nr_sorted, path->sorted_idx, path->idx);
|
2021-06-13 03:45:45 +08:00
|
|
|
#endif
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
for (i = path->sorted_idx; i < trans->nr_sorted; i++)
|
|
|
|
trans->paths[trans->sorted[i]].sorted_idx = i;
|
2019-09-27 10:21:39 +08:00
|
|
|
|
2021-06-13 03:45:45 +08:00
|
|
|
btree_trans_verify_sorted_refs(trans);
|
2019-03-26 10:43:26 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
void bch2_trans_iter_exit(struct btree_trans *trans, struct btree_iter *iter)
|
2019-09-27 10:21:39 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
if (iter->path)
|
|
|
|
bch2_path_put(trans, iter->path,
|
|
|
|
iter->flags & BTREE_ITER_INTENT);
|
|
|
|
iter->path = NULL;
|
2019-09-27 10:21:39 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static void __bch2_trans_iter_init(struct btree_trans *trans,
|
|
|
|
struct btree_iter *iter,
|
|
|
|
enum btree_id btree_id, struct bpos pos,
|
|
|
|
unsigned locks_want,
|
|
|
|
unsigned depth,
|
|
|
|
unsigned flags)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-07-26 05:19:52 +08:00
|
|
|
EBUG_ON(trans->restarted);
|
|
|
|
|
2021-08-31 03:54:41 +08:00
|
|
|
if (!(flags & (BTREE_ITER_ALL_SNAPSHOTS|BTREE_ITER_NOT_EXTENTS)) &&
|
|
|
|
btree_node_type_is_extents(btree_id))
|
2021-04-30 04:56:17 +08:00
|
|
|
flags |= BTREE_ITER_IS_EXTENTS;
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-03-16 10:34:00 +08:00
|
|
|
if (!(flags & __BTREE_ITER_ALL_SNAPSHOTS) &&
|
|
|
|
!btree_type_has_snapshots(btree_id))
|
bcachefs: Start using bpos.snapshot field
This patch starts treating the bpos.snapshot field like part of the key
in the btree code:
* bpos_successor() and bpos_predecessor() now include the snapshot field
* Keys in btrees that will be using snapshots (extents, inodes, dirents
and xattrs) now always have their snapshot field set to U32_MAX
The btree iterator code gets a new flag, BTREE_ITER_ALL_SNAPSHOTS, that
determines whether we're iterating over keys in all snapshots or not -
internally, this controlls whether bkey_(successor|predecessor)
increment/decrement the snapshot field, or only the higher bits of the
key.
We add a new member to struct btree_iter, iter->snapshot: when
BTREE_ITER_ALL_SNAPSHOTS is not set, iter->pos.snapshot should always
equal iter->snapshot, which will be 0 for btrees that don't use
snapshots, and alsways U32_MAX for btrees that will use snapshots
(until we enable snapshot creation).
This patch also introduces a new metadata version number, and compat
code for reading from/writing to older versions - this isn't a forced
upgrade (yet).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2021-03-25 06:02:16 +08:00
|
|
|
flags &= ~BTREE_ITER_ALL_SNAPSHOTS;
|
|
|
|
|
2021-03-16 10:34:00 +08:00
|
|
|
if (!(flags & BTREE_ITER_ALL_SNAPSHOTS) &&
|
|
|
|
btree_type_has_snapshots(btree_id))
|
|
|
|
flags |= BTREE_ITER_FILTER_SNAPSHOTS;
|
bcachefs: Start using bpos.snapshot field
This patch starts treating the bpos.snapshot field like part of the key
in the btree code:
* bpos_successor() and bpos_predecessor() now include the snapshot field
* Keys in btrees that will be using snapshots (extents, inodes, dirents
and xattrs) now always have their snapshot field set to U32_MAX
The btree iterator code gets a new flag, BTREE_ITER_ALL_SNAPSHOTS, that
determines whether we're iterating over keys in all snapshots or not -
internally, this controlls whether bkey_(successor|predecessor)
increment/decrement the snapshot field, or only the higher bits of the
key.
We add a new member to struct btree_iter, iter->snapshot: when
BTREE_ITER_ALL_SNAPSHOTS is not set, iter->pos.snapshot should always
equal iter->snapshot, which will be 0 for btrees that don't use
snapshots, and alsways U32_MAX for btrees that will use snapshots
(until we enable snapshot creation).
This patch also introduces a new metadata version number, and compat
code for reading from/writing to older versions - this isn't a forced
upgrade (yet).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2021-03-25 06:02:16 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->trans = trans;
|
|
|
|
iter->path = NULL;
|
|
|
|
iter->btree_id = btree_id;
|
|
|
|
iter->min_depth = depth;
|
2021-08-31 03:54:41 +08:00
|
|
|
iter->flags = flags;
|
|
|
|
iter->snapshot = pos.snapshot;
|
2021-08-31 03:18:31 +08:00
|
|
|
iter->pos = pos;
|
|
|
|
iter->k.type = KEY_TYPE_deleted;
|
|
|
|
iter->k.p = pos;
|
|
|
|
iter->k.size = 0;
|
bcachefs: Start using bpos.snapshot field
This patch starts treating the bpos.snapshot field like part of the key
in the btree code:
* bpos_successor() and bpos_predecessor() now include the snapshot field
* Keys in btrees that will be using snapshots (extents, inodes, dirents
and xattrs) now always have their snapshot field set to U32_MAX
The btree iterator code gets a new flag, BTREE_ITER_ALL_SNAPSHOTS, that
determines whether we're iterating over keys in all snapshots or not -
internally, this controlls whether bkey_(successor|predecessor)
increment/decrement the snapshot field, or only the higher bits of the
key.
We add a new member to struct btree_iter, iter->snapshot: when
BTREE_ITER_ALL_SNAPSHOTS is not set, iter->pos.snapshot should always
equal iter->snapshot, which will be 0 for btrees that don't use
snapshots, and alsways U32_MAX for btrees that will use snapshots
(until we enable snapshot creation).
This patch also introduces a new metadata version number, and compat
code for reading from/writing to older versions - this isn't a forced
upgrade (yet).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2021-03-25 06:02:16 +08:00
|
|
|
|
2021-12-22 09:48:26 +08:00
|
|
|
iter->path = bch2_path_get(trans, btree_id, iter->pos,
|
|
|
|
locks_want, depth, flags);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
void bch2_trans_iter_init(struct btree_trans *trans,
|
|
|
|
struct btree_iter *iter,
|
|
|
|
unsigned btree_id, struct bpos pos,
|
|
|
|
unsigned flags)
|
2019-03-26 03:10:15 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
__bch2_trans_iter_init(trans, iter, btree_id, pos,
|
|
|
|
0, 0, flags);
|
2019-03-26 03:10:15 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
void bch2_trans_node_iter_init(struct btree_trans *trans,
|
|
|
|
struct btree_iter *iter,
|
|
|
|
enum btree_id btree_id,
|
|
|
|
struct bpos pos,
|
|
|
|
unsigned locks_want,
|
|
|
|
unsigned depth,
|
|
|
|
unsigned flags)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
__bch2_trans_iter_init(trans, iter, btree_id, pos, locks_want, depth,
|
|
|
|
BTREE_ITER_NOT_EXTENTS|
|
|
|
|
__BTREE_ITER_ALL_SNAPSHOTS|
|
|
|
|
BTREE_ITER_ALL_SNAPSHOTS|
|
|
|
|
flags);
|
|
|
|
BUG_ON(iter->path->locks_want < min(locks_want, BTREE_MAX_DEPTH));
|
|
|
|
BUG_ON(iter->path->level != depth);
|
|
|
|
BUG_ON(iter->min_depth != depth);
|
|
|
|
}
|
2019-03-26 10:43:26 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
void bch2_trans_copy_iter(struct btree_iter *dst, struct btree_iter *src)
|
|
|
|
{
|
|
|
|
*dst = *src;
|
|
|
|
if (src->path)
|
|
|
|
__btree_path_get(src->path, src->flags & BTREE_ITER_INTENT);
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-04-16 00:50:09 +08:00
|
|
|
void *bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-04-16 00:50:09 +08:00
|
|
|
size_t new_top = trans->mem_top + size;
|
|
|
|
void *p;
|
|
|
|
|
|
|
|
if (new_top > trans->mem_bytes) {
|
2017-03-17 14:18:50 +08:00
|
|
|
size_t old_bytes = trans->mem_bytes;
|
2021-04-16 00:50:09 +08:00
|
|
|
size_t new_bytes = roundup_pow_of_two(new_top);
|
2021-04-24 12:09:06 +08:00
|
|
|
void *new_mem;
|
|
|
|
|
|
|
|
WARN_ON_ONCE(new_bytes > BTREE_TRANS_MEM_MAX);
|
|
|
|
|
|
|
|
new_mem = krealloc(trans->mem, new_bytes, GFP_NOFS);
|
|
|
|
if (!new_mem && new_bytes <= BTREE_TRANS_MEM_MAX) {
|
|
|
|
new_mem = mempool_alloc(&trans->c->btree_trans_mem_pool, GFP_KERNEL);
|
|
|
|
new_bytes = BTREE_TRANS_MEM_MAX;
|
|
|
|
kfree(trans->mem);
|
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
if (!new_mem)
|
2021-04-16 00:50:09 +08:00
|
|
|
return ERR_PTR(-ENOMEM);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
|
|
|
trans->mem = new_mem;
|
|
|
|
trans->mem_bytes = new_bytes;
|
|
|
|
|
2018-07-13 11:30:45 +08:00
|
|
|
if (old_bytes) {
|
2021-04-16 00:50:09 +08:00
|
|
|
trace_trans_restart_mem_realloced(trans->ip, _RET_IP_, new_bytes);
|
2021-07-26 05:19:52 +08:00
|
|
|
btree_trans_restart(trans);
|
2021-04-16 00:50:09 +08:00
|
|
|
return ERR_PTR(-EINTR);
|
2018-07-13 11:30:45 +08:00
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2019-05-15 22:54:43 +08:00
|
|
|
p = trans->mem + trans->mem_top;
|
2017-03-17 14:18:50 +08:00
|
|
|
trans->mem_top += size;
|
2021-06-08 04:50:30 +08:00
|
|
|
memset(p, 0, size);
|
2019-05-15 22:54:43 +08:00
|
|
|
return p;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-07-08 10:31:36 +08:00
|
|
|
/**
|
2021-07-25 11:57:28 +08:00
|
|
|
* bch2_trans_begin() - reset a transaction after a interrupted attempt
|
2021-07-08 10:31:36 +08:00
|
|
|
* @trans: transaction to reset
|
|
|
|
*
|
|
|
|
* While iterating over nodes or updating nodes a attempt to lock a btree
|
|
|
|
* node may return EINTR when the trylock fails. When this occurs
|
2021-07-25 11:57:28 +08:00
|
|
|
* bch2_trans_begin() should be called and the transaction retried.
|
2021-07-08 10:31:36 +08:00
|
|
|
*/
|
2021-07-25 11:57:28 +08:00
|
|
|
void bch2_trans_begin(struct btree_trans *trans)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_insert_entry *i;
|
|
|
|
struct btree_path *path;
|
2019-03-08 12:13:39 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_update(trans, i)
|
|
|
|
__btree_path_put(i->path, true);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-10-30 06:43:18 +08:00
|
|
|
memset(&trans->journal_res, 0, sizeof(trans->journal_res));
|
2021-06-13 03:45:56 +08:00
|
|
|
trans->extra_journal_res = 0;
|
2019-03-26 03:34:48 +08:00
|
|
|
trans->nr_updates = 0;
|
2020-02-27 04:39:46 +08:00
|
|
|
trans->mem_top = 0;
|
2019-03-28 12:07:24 +08:00
|
|
|
|
2021-02-04 10:51:56 +08:00
|
|
|
trans->hooks = NULL;
|
2020-05-26 07:29:48 +08:00
|
|
|
trans->extra_journal_entries = NULL;
|
|
|
|
trans->extra_journal_entry_u64s = 0;
|
|
|
|
|
2019-12-31 01:43:19 +08:00
|
|
|
if (trans->fs_usage_deltas) {
|
|
|
|
trans->fs_usage_deltas->used = 0;
|
|
|
|
memset((void *) trans->fs_usage_deltas +
|
|
|
|
offsetof(struct replicas_delta_list, memset_start), 0,
|
|
|
|
(void *) &trans->fs_usage_deltas->memset_end -
|
|
|
|
(void *) &trans->fs_usage_deltas->memset_start);
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, path) {
|
|
|
|
/*
|
|
|
|
* XXX: we probably shouldn't be doing this if the transaction
|
|
|
|
* was restarted, but currently we still overflow transaction
|
|
|
|
* iterators if we do that
|
|
|
|
*/
|
|
|
|
if (!path->ref && !path->preserve)
|
|
|
|
__bch2_path_free(trans, path);
|
|
|
|
else
|
|
|
|
path->preserve = path->should_be_locked = false;
|
|
|
|
}
|
|
|
|
|
2021-07-25 11:57:28 +08:00
|
|
|
bch2_trans_cond_resched(trans);
|
2021-03-20 08:29:11 +08:00
|
|
|
|
2021-07-25 11:57:28 +08:00
|
|
|
if (trans->restarted)
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_path_traverse_all(trans);
|
2021-07-26 05:19:52 +08:00
|
|
|
|
|
|
|
trans->restarted = false;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static void bch2_trans_alloc_paths(struct btree_trans *trans, struct bch_fs *c)
|
2020-11-06 09:02:01 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
size_t paths_bytes = sizeof(struct btree_path) * BTREE_ITER_MAX;
|
2020-12-02 12:11:53 +08:00
|
|
|
size_t updates_bytes = sizeof(struct btree_insert_entry) * BTREE_ITER_MAX;
|
2020-11-17 07:20:50 +08:00
|
|
|
void *p = NULL;
|
2020-11-06 09:02:01 +08:00
|
|
|
|
|
|
|
BUG_ON(trans->used_mempool);
|
|
|
|
|
2020-11-17 07:20:50 +08:00
|
|
|
#ifdef __KERNEL__
|
2021-08-31 03:18:31 +08:00
|
|
|
p = this_cpu_xchg(c->btree_paths_bufs->path , NULL);
|
2020-11-17 07:20:50 +08:00
|
|
|
#endif
|
|
|
|
if (!p)
|
2021-08-31 03:18:31 +08:00
|
|
|
p = mempool_alloc(&trans->c->btree_paths_pool, GFP_NOFS);
|
2020-11-06 09:02:01 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans->paths = p; p += paths_bytes;
|
2020-11-06 09:02:01 +08:00
|
|
|
trans->updates = p; p += updates_bytes;
|
|
|
|
}
|
|
|
|
|
2019-05-15 22:54:43 +08:00
|
|
|
void bch2_trans_init(struct btree_trans *trans, struct bch_fs *c,
|
|
|
|
unsigned expected_nr_iters,
|
|
|
|
size_t expected_mem_bytes)
|
2021-05-24 05:04:13 +08:00
|
|
|
__acquires(&c->btree_trans_barrier)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2020-11-03 07:54:33 +08:00
|
|
|
memset(trans, 0, sizeof(*trans));
|
2017-03-17 14:18:50 +08:00
|
|
|
trans->c = c;
|
2019-04-23 12:10:08 +08:00
|
|
|
trans->ip = _RET_IP_;
|
2019-05-15 22:54:43 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_trans_alloc_paths(trans, c);
|
2019-05-15 22:54:43 +08:00
|
|
|
|
2020-11-16 09:52:55 +08:00
|
|
|
if (expected_mem_bytes) {
|
|
|
|
expected_mem_bytes = roundup_pow_of_two(expected_mem_bytes);
|
|
|
|
trans->mem = kmalloc(expected_mem_bytes, GFP_KERNEL);
|
2021-04-24 12:09:06 +08:00
|
|
|
|
|
|
|
if (!unlikely(trans->mem)) {
|
|
|
|
trans->mem = mempool_alloc(&c->btree_trans_mem_pool, GFP_KERNEL);
|
|
|
|
trans->mem_bytes = BTREE_TRANS_MEM_MAX;
|
|
|
|
} else {
|
2020-11-16 09:52:55 +08:00
|
|
|
trans->mem_bytes = expected_mem_bytes;
|
2021-04-24 12:09:06 +08:00
|
|
|
}
|
2020-11-16 09:52:55 +08:00
|
|
|
}
|
2020-06-03 04:36:11 +08:00
|
|
|
|
2020-11-16 05:30:22 +08:00
|
|
|
trans->srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
|
|
|
|
|
2021-12-17 09:36:26 +08:00
|
|
|
if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG_TRANSACTIONS)) {
|
|
|
|
trans->pid = current->pid;
|
|
|
|
mutex_lock(&c->btree_trans_lock);
|
|
|
|
list_add(&trans->list, &c->btree_trans_list);
|
|
|
|
mutex_unlock(&c->btree_trans_lock);
|
|
|
|
}
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
static void check_btree_paths_leaked(struct btree_trans *trans)
|
|
|
|
{
|
|
|
|
#ifdef CONFIG_BCACHEFS_DEBUG
|
|
|
|
struct bch_fs *c = trans->c;
|
|
|
|
struct btree_path *path;
|
|
|
|
|
|
|
|
trans_for_each_path(trans, path)
|
|
|
|
if (path->ref)
|
|
|
|
goto leaked;
|
|
|
|
return;
|
|
|
|
leaked:
|
|
|
|
bch_err(c, "btree paths leaked from %pS!", (void *) trans->ip);
|
|
|
|
trans_for_each_path(trans, path)
|
|
|
|
if (path->ref)
|
|
|
|
printk(KERN_ERR " btree %s %pS\n",
|
|
|
|
bch2_btree_ids[path->btree_id],
|
|
|
|
(void *) path->ip_allocated);
|
|
|
|
/* Be noisy about this: */
|
|
|
|
bch2_fatal_error(c);
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
2021-10-20 03:08:00 +08:00
|
|
|
void bch2_trans_exit(struct btree_trans *trans)
|
2021-05-24 05:04:13 +08:00
|
|
|
__releases(&c->btree_trans_barrier)
|
2017-03-17 14:18:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_insert_entry *i;
|
2020-11-06 09:02:01 +08:00
|
|
|
struct bch_fs *c = trans->c;
|
|
|
|
|
2019-04-05 09:28:16 +08:00
|
|
|
bch2_trans_unlock(trans);
|
2017-03-17 14:18:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_update(trans, i)
|
|
|
|
__btree_path_put(i->path, true);
|
|
|
|
trans->nr_updates = 0;
|
2021-03-21 03:12:05 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
check_btree_paths_leaked(trans);
|
2021-03-20 08:29:11 +08:00
|
|
|
|
2021-12-17 09:36:26 +08:00
|
|
|
if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG_TRANSACTIONS)) {
|
|
|
|
mutex_lock(&c->btree_trans_lock);
|
|
|
|
list_del(&trans->list);
|
|
|
|
mutex_unlock(&c->btree_trans_lock);
|
|
|
|
}
|
2020-06-03 04:36:11 +08:00
|
|
|
|
2020-11-16 05:30:22 +08:00
|
|
|
srcu_read_unlock(&c->btree_trans_barrier, trans->srcu_idx);
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_journal_preres_put(&c->journal, &trans->journal_preres);
|
2019-03-08 08:46:10 +08:00
|
|
|
|
2021-04-24 12:24:25 +08:00
|
|
|
if (trans->fs_usage_deltas) {
|
|
|
|
if (trans->fs_usage_deltas->size + sizeof(trans->fs_usage_deltas) ==
|
|
|
|
REPLICAS_DELTA_LIST_MAX)
|
|
|
|
mempool_free(trans->fs_usage_deltas,
|
2021-08-31 03:18:31 +08:00
|
|
|
&c->replicas_delta_pool);
|
2021-04-24 12:24:25 +08:00
|
|
|
else
|
|
|
|
kfree(trans->fs_usage_deltas);
|
|
|
|
}
|
2021-04-24 12:09:06 +08:00
|
|
|
|
|
|
|
if (trans->mem_bytes == BTREE_TRANS_MEM_MAX)
|
2021-08-31 03:18:31 +08:00
|
|
|
mempool_free(trans->mem, &c->btree_trans_mem_pool);
|
2021-04-24 12:09:06 +08:00
|
|
|
else
|
|
|
|
kfree(trans->mem);
|
2020-11-06 09:02:01 +08:00
|
|
|
|
2020-11-17 07:20:50 +08:00
|
|
|
#ifdef __KERNEL__
|
|
|
|
/*
|
|
|
|
* Userspace doesn't have a real percpu implementation:
|
|
|
|
*/
|
2021-08-31 03:18:31 +08:00
|
|
|
trans->paths = this_cpu_xchg(c->btree_paths_bufs->path, trans->paths);
|
2020-11-17 07:20:50 +08:00
|
|
|
#endif
|
2021-04-24 12:09:06 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
if (trans->paths)
|
|
|
|
mempool_free(trans->paths, &c->btree_paths_pool);
|
2020-11-06 09:02:01 +08:00
|
|
|
|
2017-03-17 14:18:50 +08:00
|
|
|
trans->mem = (void *) 0x1;
|
2021-08-31 03:18:31 +08:00
|
|
|
trans->paths = (void *) 0x1;
|
2017-03-17 14:18:50 +08:00
|
|
|
}
|
2019-09-08 02:16:00 +08:00
|
|
|
|
2020-11-08 01:43:48 +08:00
|
|
|
static void __maybe_unused
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_path_node_to_text(struct printbuf *out,
|
2020-11-08 01:43:48 +08:00
|
|
|
struct btree_bkey_cached_common *_b,
|
2021-08-31 03:54:41 +08:00
|
|
|
bool cached)
|
2020-06-16 07:53:46 +08:00
|
|
|
{
|
2021-04-01 04:43:50 +08:00
|
|
|
pr_buf(out, " l=%u %s:",
|
|
|
|
_b->level, bch2_btree_ids[_b->btree_id]);
|
2021-08-31 03:54:41 +08:00
|
|
|
bch2_bpos_to_text(out, btree_node_pos(_b, cached));
|
2020-06-16 07:53:46 +08:00
|
|
|
}
|
|
|
|
|
2021-12-17 09:36:26 +08:00
|
|
|
#ifdef CONFIG_BCACHEFS_DEBUG_TRANSACTIONS
|
2021-08-31 03:54:41 +08:00
|
|
|
static bool trans_has_locks(struct btree_trans *trans)
|
2021-04-01 04:43:50 +08:00
|
|
|
{
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path;
|
2021-04-01 04:43:50 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, path)
|
|
|
|
if (path->nodes_locked)
|
2021-04-01 04:43:50 +08:00
|
|
|
return true;
|
|
|
|
return false;
|
|
|
|
}
|
2021-04-03 09:29:05 +08:00
|
|
|
#endif
|
2021-04-01 04:43:50 +08:00
|
|
|
|
2020-06-03 04:36:11 +08:00
|
|
|
void bch2_btree_trans_to_text(struct printbuf *out, struct bch_fs *c)
|
|
|
|
{
|
2021-12-17 09:36:26 +08:00
|
|
|
#ifdef CONFIG_BCACHEFS_DEBUG_TRANSACTIONS
|
2020-06-03 04:36:11 +08:00
|
|
|
struct btree_trans *trans;
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_path *path;
|
2020-06-03 04:36:11 +08:00
|
|
|
struct btree *b;
|
|
|
|
unsigned l;
|
|
|
|
|
|
|
|
mutex_lock(&c->btree_trans_lock);
|
|
|
|
list_for_each_entry(trans, &c->btree_trans_list, list) {
|
2021-08-31 03:54:41 +08:00
|
|
|
if (!trans_has_locks(trans))
|
2021-04-01 04:43:50 +08:00
|
|
|
continue;
|
|
|
|
|
|
|
|
pr_buf(out, "%i %ps\n", trans->pid, (void *) trans->ip);
|
2020-06-03 04:36:11 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
trans_for_each_path(trans, path) {
|
|
|
|
if (!path->nodes_locked)
|
2020-06-03 04:36:11 +08:00
|
|
|
continue;
|
|
|
|
|
2021-09-05 09:23:11 +08:00
|
|
|
pr_buf(out, " path %u %c l=%u %s:",
|
2021-08-31 03:18:31 +08:00
|
|
|
path->idx,
|
|
|
|
path->cached ? 'c' : 'b',
|
2021-09-05 09:23:11 +08:00
|
|
|
path->level,
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_ids[path->btree_id]);
|
|
|
|
bch2_bpos_to_text(out, path->pos);
|
2020-06-03 04:36:11 +08:00
|
|
|
pr_buf(out, "\n");
|
|
|
|
|
|
|
|
for (l = 0; l < BTREE_MAX_DEPTH; l++) {
|
2021-08-31 03:18:31 +08:00
|
|
|
if (btree_node_locked(path, l)) {
|
2020-06-16 07:53:46 +08:00
|
|
|
pr_buf(out, " %s l=%u ",
|
2021-08-31 03:18:31 +08:00
|
|
|
btree_node_intent_locked(path, l) ? "i" : "r", l);
|
|
|
|
bch2_btree_path_node_to_text(out,
|
|
|
|
(void *) path->l[l].b,
|
|
|
|
path->cached);
|
2020-06-03 04:36:11 +08:00
|
|
|
pr_buf(out, "\n");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
b = READ_ONCE(trans->locking);
|
|
|
|
if (b) {
|
2021-08-31 03:18:31 +08:00
|
|
|
path = &trans->paths[trans->locking_path_idx];
|
|
|
|
pr_buf(out, " locking path %u %c l=%u %s:",
|
|
|
|
trans->locking_path_idx,
|
|
|
|
path->cached ? 'c' : 'b',
|
2020-06-13 02:58:07 +08:00
|
|
|
trans->locking_level,
|
|
|
|
bch2_btree_ids[trans->locking_btree_id]);
|
|
|
|
bch2_bpos_to_text(out, trans->locking_pos);
|
|
|
|
|
2020-06-16 07:53:46 +08:00
|
|
|
pr_buf(out, " node ");
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_path_node_to_text(out,
|
|
|
|
(void *) b, path->cached);
|
2020-06-03 04:36:11 +08:00
|
|
|
pr_buf(out, "\n");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
mutex_unlock(&c->btree_trans_lock);
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
2019-09-08 02:16:00 +08:00
|
|
|
void bch2_fs_btree_iter_exit(struct bch_fs *c)
|
|
|
|
{
|
2021-12-21 07:18:35 +08:00
|
|
|
if (c->btree_trans_barrier_initialized)
|
|
|
|
cleanup_srcu_struct(&c->btree_trans_barrier);
|
2021-04-24 12:09:06 +08:00
|
|
|
mempool_exit(&c->btree_trans_mem_pool);
|
2021-08-31 03:18:31 +08:00
|
|
|
mempool_exit(&c->btree_paths_pool);
|
2019-09-08 02:16:00 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
int bch2_fs_btree_iter_init(struct bch_fs *c)
|
|
|
|
{
|
|
|
|
unsigned nr = BTREE_ITER_MAX;
|
2021-12-21 07:18:35 +08:00
|
|
|
int ret;
|
2019-09-08 02:16:00 +08:00
|
|
|
|
2020-06-03 04:36:11 +08:00
|
|
|
INIT_LIST_HEAD(&c->btree_trans_list);
|
|
|
|
mutex_init(&c->btree_trans_lock);
|
|
|
|
|
2021-12-21 07:18:35 +08:00
|
|
|
ret = mempool_init_kmalloc_pool(&c->btree_paths_pool, 1,
|
2021-08-31 03:18:31 +08:00
|
|
|
sizeof(struct btree_path) * nr +
|
2021-04-24 12:09:06 +08:00
|
|
|
sizeof(struct btree_insert_entry) * nr) ?:
|
|
|
|
mempool_init_kmalloc_pool(&c->btree_trans_mem_pool, 1,
|
2021-12-21 07:18:35 +08:00
|
|
|
BTREE_TRANS_MEM_MAX) ?:
|
|
|
|
init_srcu_struct(&c->btree_trans_barrier);
|
|
|
|
if (!ret)
|
|
|
|
c->btree_trans_barrier_initialized = true;
|
|
|
|
return ret;
|
2019-09-08 02:16:00 +08:00
|
|
|
}
|