#include "journal.h"
#include "journal_reclaim.h"
#include "keylist.h"
+#include "recovery.h"
#include "subvolume.h"
#include "replicas.h"
#include <linux/sort.h>
#include <trace/events/bcachefs.h>
+static int __must_check
+bch2_trans_update_by_path(struct btree_trans *, struct btree_path *,
+ struct bkey_i *, enum btree_update_flags);
+
static inline int btree_insert_entry_cmp(const struct btree_insert_entry *l,
const struct btree_insert_entry *r)
{
struct btree_write *w = container_of(pin, struct btree_write, journal);
struct btree *b = container_of(w, struct btree, writes[i]);
- btree_node_lock_type(c, b, SIX_LOCK_read);
+ six_lock_read(&b->c.lock, NULL, NULL);
bch2_btree_node_write_cond(c, b,
(btree_current_write(b) == w && w->journal.seq == seq));
six_unlock_read(&b->c.lock);
int old_live_u64s = b->nr.live_u64s;
int live_u64s_added, u64s_added;
- EBUG_ON(!insert->level &&
- !test_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags));
-
if (unlikely(!bch2_btree_bset_insert_key(trans, insert->path, b,
&insert_l(insert)->iter, insert->k)))
return false;
return ret;
if (!bch2_trans_relock(trans)) {
- trace_trans_restart_journal_preres_get(trans->ip, trace_ip);
+ trace_trans_restart_journal_preres_get(trans->fn, trace_ip);
return -EINTR;
}
l->entry.pad[0] = 0;
l->entry.pad[1] = 0;
l->entry.pad[2] = 0;
- b = snprintf(l->d, buflen, "%ps", (void *) trans->ip);
+ b = min_t(unsigned, strlen(trans->fn), buflen);
+ memcpy(l->d, trans->fn, b);
while (b < buflen)
l->d[b++] = '\0';
}
}
-static noinline void bch2_trans_mark_gc(struct btree_trans *trans)
+/* Triggers: */
+
+static int run_one_mem_trigger(struct btree_trans *trans,
+ struct btree_insert_entry *i,
+ unsigned flags)
+{
+ struct bkey_s_c old = { &i->old_k, i->old_v };
+ struct bkey_i *new = i->k;
+ int ret;
+
+ if (unlikely(flags & BTREE_TRIGGER_NORUN))
+ return 0;
+
+ if (!btree_node_type_needs_gc(i->btree_id))
+ return 0;
+
+ if (old.k->type == new->k.type &&
+ ((1U << old.k->type) & BTREE_TRIGGER_WANTS_OLD_AND_NEW)) {
+ ret = bch2_mark_key(trans, old, bkey_i_to_s_c(new),
+ BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|flags);
+ } else {
+ struct bkey _deleted = KEY(0, 0, 0);
+ struct bkey_s_c deleted = (struct bkey_s_c) { &_deleted, NULL };
+
+ _deleted.p = i->path->pos;
+
+ ret = bch2_mark_key(trans, deleted, bkey_i_to_s_c(new),
+ BTREE_TRIGGER_INSERT|flags) ?:
+ bch2_mark_key(trans, old, deleted,
+ BTREE_TRIGGER_OVERWRITE|flags);
+ }
+
+ return ret;
+}
+
+static int run_one_trans_trigger(struct btree_trans *trans, struct btree_insert_entry *i,
+ bool overwrite)
+{
+ struct bkey_s_c old = { &i->old_k, i->old_v };
+ int ret = 0;
+
+ if ((i->flags & BTREE_TRIGGER_NORUN) ||
+ !(BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->bkey_type)))
+ return 0;
+
+ if (!overwrite) {
+ if (i->insert_trigger_run)
+ return 0;
+
+ BUG_ON(i->overwrite_trigger_run);
+ i->insert_trigger_run = true;
+ } else {
+ if (i->overwrite_trigger_run)
+ return 0;
+
+ BUG_ON(!i->insert_trigger_run);
+ i->overwrite_trigger_run = true;
+ }
+
+ if (overwrite) {
+ ret = bch2_trans_mark_old(trans, old, i->flags);
+ } else if (old.k->type == i->k->k.type &&
+ ((1U << old.k->type) & BTREE_TRIGGER_WANTS_OLD_AND_NEW)) {
+ i->overwrite_trigger_run = true;
+ ret = bch2_trans_mark_key(trans, old, i->k,
+ BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|i->flags);
+ } else {
+ ret = bch2_trans_mark_new(trans, i->k, i->flags);
+ }
+
+ if (ret == -EINTR)
+ trace_trans_restart_mark(trans->fn, _RET_IP_,
+ i->btree_id, &i->path->pos);
+ return ret ?: 1;
+}
+
+static int run_btree_triggers(struct btree_trans *trans, enum btree_id btree_id,
+ struct btree_insert_entry *btree_id_start)
+{
+ struct btree_insert_entry *i;
+ bool trans_trigger_run;
+ int ret, overwrite;
+
+ for (overwrite = 0; overwrite < 2; overwrite++) {
+
+ /*
+ * Running triggers will append more updates to the list of updates as
+ * we're walking it:
+ */
+ do {
+ trans_trigger_run = false;
+
+ for (i = btree_id_start;
+ i < trans->updates + trans->nr_updates && i->btree_id <= btree_id;
+ i++) {
+ if (i->btree_id != btree_id)
+ continue;
+
+ ret = run_one_trans_trigger(trans, i, overwrite);
+ if (ret < 0)
+ return ret;
+ if (ret)
+ trans_trigger_run = true;
+ }
+ } while (trans_trigger_run);
+ }
+
+ return 0;
+}
+
+static int bch2_trans_commit_run_triggers(struct btree_trans *trans)
+{
+ struct btree_insert_entry *i = NULL, *btree_id_start = trans->updates;
+ unsigned btree_id = 0;
+ int ret = 0;
+
+ /*
+ *
+ * For a given btree, this algorithm runs insert triggers before
+ * overwrite triggers: this is so that when extents are being moved
+ * (e.g. by FALLOCATE_FL_INSERT_RANGE), we don't drop references before
+ * they are re-added.
+ */
+ for (btree_id = 0; btree_id < BTREE_ID_NR; btree_id++) {
+ if (btree_id == BTREE_ID_alloc)
+ continue;
+
+ while (btree_id_start < trans->updates + trans->nr_updates &&
+ btree_id_start->btree_id < btree_id)
+ btree_id_start++;
+
+ ret = run_btree_triggers(trans, btree_id, btree_id_start);
+ if (ret)
+ return ret;
+ }
+
+ trans_for_each_update(trans, i) {
+ if (i->btree_id > BTREE_ID_alloc)
+ break;
+ if (i->btree_id == BTREE_ID_alloc) {
+ ret = run_btree_triggers(trans, BTREE_ID_alloc, i);
+ if (ret)
+ return ret;
+ break;
+ }
+ }
+
+ trans_for_each_update(trans, i)
+ BUG_ON(!(i->flags & BTREE_TRIGGER_NORUN) &&
+ (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->bkey_type)) &&
+ (!i->insert_trigger_run || !i->overwrite_trigger_run));
+
+ return 0;
+}
+
+static noinline int bch2_trans_commit_run_gc_triggers(struct btree_trans *trans)
{
struct bch_fs *c = trans->c;
struct btree_insert_entry *i;
+ int ret = 0;
trans_for_each_update(trans, i) {
/*
*/
BUG_ON(i->cached || i->level);
- if (gc_visited(c, gc_pos_btree_node(insert_l(i)->b)))
- bch2_mark_update(trans, i->path, i->k,
- i->flags|BTREE_TRIGGER_GC);
+ if (gc_visited(c, gc_pos_btree_node(insert_l(i)->b))) {
+ ret = run_one_mem_trigger(trans, i, i->flags|BTREE_TRIGGER_GC);
+ if (ret)
+ break;
+ }
}
+
+ return ret;
}
static inline int
int ret;
if (race_fault()) {
- trace_trans_restart_fault_inject(trans->ip, trace_ip);
+ trace_trans_restart_fault_inject(trans->fn, trace_ip);
trans->restarted = true;
return -EINTR;
}
JOURNAL_RES_GET_NONBLOCK);
if (ret)
return ret;
+
+ if (unlikely(trans->journal_transaction_names))
+ journal_transaction_name(trans);
} else {
trans->journal_res.seq = c->journal.replay_journal_seq;
}
- if (unlikely(trans->journal_transaction_names))
- journal_transaction_name(trans);
-
if (unlikely(trans->extra_journal_entry_u64s)) {
memcpy_u64s_small(journal_res_entry(&c->journal, &trans->journal_res),
trans->extra_journal_entries,
return BTREE_INSERT_NEED_MARK_REPLICAS;
trans_for_each_update(trans, i)
- if (BTREE_NODE_TYPE_HAS_MEM_TRIGGERS & (1U << i->bkey_type))
- bch2_mark_update(trans, i->path, i->k, i->flags);
+ if (BTREE_NODE_TYPE_HAS_MEM_TRIGGERS & (1U << i->bkey_type)) {
+ ret = run_one_mem_trigger(trans, i, i->flags);
+ if (ret)
+ return ret;
+ }
- if (unlikely(c->gc_pos.phase))
- bch2_trans_mark_gc(trans);
+ if (unlikely(c->gc_pos.phase)) {
+ ret = bch2_trans_commit_run_gc_triggers(trans);
+ if (ret)
+ return ret;
+ }
trans_for_each_update(trans, i)
do_btree_insert_one(trans, i);
if (have_conflicting_read_lock(trans, i->path))
goto fail;
- __btree_node_lock_type(trans->c, insert_l(i)->b,
- SIX_LOCK_write);
+ btree_node_lock_type(trans, i->path,
+ insert_l(i)->b,
+ i->path->pos, i->level,
+ SIX_LOCK_write, NULL, NULL);
}
bch2_btree_node_prep_for_write(trans, i->path, insert_l(i)->b);
bch2_btree_node_unlock_write_inlined(trans, i->path, insert_l(i)->b);
}
- trace_trans_restart_would_deadlock_write(trans->ip);
+ trace_trans_restart_would_deadlock_write(trans->fn);
return btree_trans_restart(trans);
}
+static noinline void bch2_drop_overwrites_from_journal(struct btree_trans *trans)
+{
+ struct btree_insert_entry *i;
+
+ trans_for_each_update(trans, i)
+ bch2_journal_key_overwritten(trans->c, i->btree_id, i->level, i->k->k.p);
+}
+
/*
* Get journal reservation, take write locks, and attempt to do btree update(s):
*/
{
struct bch_fs *c = trans->c;
struct btree_insert_entry *i;
- struct bkey_s_c old;
int ret, u64s_delta = 0;
trans_for_each_update(trans, i) {
const char *invalid = bch2_bkey_invalid(c,
bkey_i_to_s_c(i->k), i->bkey_type);
if (invalid) {
- char buf[200];
+ struct printbuf buf = PRINTBUF;
- bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(i->k));
- bch_err(c, "invalid bkey %s on insert from %ps -> %ps: %s\n",
- buf, (void *) trans->ip,
- (void *) i->ip_allocated, invalid);
- bch2_fatal_error(c);
+ bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(i->k));
+ bch2_fs_fatal_error(c, "invalid bkey %s on insert from %s -> %ps: %s\n",
+ buf.buf, trans->fn, (void *) i->ip_allocated, invalid);
+ printbuf_exit(&buf);
return -EINVAL;
}
btree_insert_entry_checks(trans, i);
}
trans_for_each_update(trans, i) {
- struct bkey u;
-
- /*
- * peek_slot() doesn't yet work on iterators that point to
- * interior nodes:
- */
- if (i->cached || i->level)
+ if (i->cached)
continue;
- old = bch2_btree_path_peek_slot(i->path, &u);
- ret = bkey_err(old);
- if (unlikely(ret))
- return ret;
-
u64s_delta += !bkey_deleted(&i->k->k) ? i->k->k.u64s : 0;
- u64s_delta -= !bkey_deleted(old.k) ? old.k->u64s : 0;
+ u64s_delta -= i->old_btree_u64s;
if (!same_leaf_as_next(trans, i)) {
if (u64s_delta <= 0) {
ret = bch2_trans_commit_write_locked(trans, stopped_at, trace_ip);
+ if (!ret && unlikely(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)))
+ bch2_drop_overwrites_from_journal(trans);
+
trans_for_each_update(trans, i)
if (!same_leaf_as_prev(trans, i))
bch2_btree_node_unlock_write_inlined(trans, i->path,
return 0;
if (ret == -EINTR)
- trace_trans_restart_btree_node_split(trans->ip, trace_ip,
+ trace_trans_restart_btree_node_split(trans->fn, trace_ip,
i->btree_id, &i->path->pos);
break;
case BTREE_INSERT_NEED_MARK_REPLICAS:
if (bch2_trans_relock(trans))
return 0;
- trace_trans_restart_mark_replicas(trans->ip, trace_ip);
+ trace_trans_restart_mark_replicas(trans->fn, trace_ip);
ret = -EINTR;
break;
case BTREE_INSERT_NEED_JOURNAL_RES:
if (bch2_trans_relock(trans))
return 0;
- trace_trans_restart_journal_res_get(trans->ip, trace_ip);
+ trace_trans_restart_journal_res_get(trans->fn, trace_ip);
ret = -EINTR;
break;
case BTREE_INSERT_NEED_JOURNAL_RECLAIM:
bch2_trans_unlock(trans);
- trace_trans_blocked_journal_reclaim(trans->ip, trace_ip);
+ trace_trans_blocked_journal_reclaim(trans->fn, trace_ip);
wait_event_freezable(c->journal.reclaim_wait,
(ret = journal_reclaim_wait_done(c)));
if (bch2_trans_relock(trans))
return 0;
- trace_trans_restart_journal_reclaim(trans->ip, trace_ip);
+ trace_trans_restart_journal_reclaim(trans->fn, trace_ip);
ret = -EINTR;
break;
default:
}
BUG_ON((ret == EINTR || ret == -EAGAIN) && !trans->restarted);
- BUG_ON(ret == -ENOSPC && (trans->flags & BTREE_INSERT_NOFAIL));
+ BUG_ON(ret == -ENOSPC &&
+ !(trans->flags & BTREE_INSERT_NOWAIT) &&
+ (trans->flags & BTREE_INSERT_NOFAIL));
return ret;
}
struct bch_fs *c = trans->c;
int ret;
- if (likely(!(trans->flags & BTREE_INSERT_LAZY_RW)))
+ if (likely(!(trans->flags & BTREE_INSERT_LAZY_RW)) ||
+ test_bit(BCH_FS_STARTED, &c->flags))
return -EROFS;
bch2_trans_unlock(trans);
return 0;
}
-static int bch2_trans_commit_run_triggers(struct btree_trans *trans)
+/*
+ * This is for updates done in the early part of fsck - btree_gc - before we've
+ * gone RW. we only add the new key to the list of keys for journal replay to
+ * do.
+ */
+static noinline int
+do_bch2_trans_commit_to_journal_replay(struct btree_trans *trans)
{
- struct bkey _deleted = KEY(0, 0, 0);
- struct bkey_s_c deleted = (struct bkey_s_c) { &_deleted, NULL };
- struct bkey_s_c old;
- struct bkey unpacked;
- struct btree_insert_entry *i = NULL, *btree_id_start = trans->updates;
- bool trans_trigger_run;
- unsigned btree_id = 0;
+ struct bch_fs *c = trans->c;
+ struct btree_insert_entry *i;
int ret = 0;
- /*
- *
- * For a given btree, this algorithm runs insert triggers before
- * overwrite triggers: this is so that when extents are being moved
- * (e.g. by FALLOCATE_FL_INSERT_RANGE), we don't drop references before
- * they are re-added.
- */
- for (btree_id = 0; btree_id < BTREE_ID_NR; btree_id++) {
- while (btree_id_start < trans->updates + trans->nr_updates &&
- btree_id_start->btree_id < btree_id)
- btree_id_start++;
-
- /*
- * Running triggers will append more updates to the list of updates as
- * we're walking it:
- */
- do {
- trans_trigger_run = false;
-
- for (i = btree_id_start;
- i < trans->updates + trans->nr_updates && i->btree_id <= btree_id;
- i++) {
- if (i->insert_trigger_run ||
- (i->flags & BTREE_TRIGGER_NORUN) ||
- !(BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->bkey_type)))
- continue;
-
- BUG_ON(i->overwrite_trigger_run);
-
- i->insert_trigger_run = true;
- trans_trigger_run = true;
-
- old = bch2_btree_path_peek_slot(i->path, &unpacked);
- _deleted.p = i->path->pos;
-
- if (old.k->type == i->k->k.type &&
- ((1U << old.k->type) & BTREE_TRIGGER_WANTS_OLD_AND_NEW)) {
- i->overwrite_trigger_run = true;
- ret = bch2_trans_mark_key(trans, old, bkey_i_to_s_c(i->k),
- BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|i->flags);
- } else {
- ret = bch2_trans_mark_key(trans, deleted, bkey_i_to_s_c(i->k),
- BTREE_TRIGGER_INSERT|i->flags);
- }
-
- if (ret == -EINTR)
- trace_trans_restart_mark(trans->ip, _RET_IP_,
- i->btree_id, &i->path->pos);
- if (ret)
- return ret;
- }
- } while (trans_trigger_run);
-
- do {
- trans_trigger_run = false;
-
- for (i = btree_id_start;
- i < trans->updates + trans->nr_updates && i->btree_id <= btree_id;
- i++) {
- if (i->overwrite_trigger_run ||
- (i->flags & BTREE_TRIGGER_NORUN) ||
- !(BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->bkey_type)))
- continue;
-
- BUG_ON(!i->insert_trigger_run);
-
- i->overwrite_trigger_run = true;
- trans_trigger_run = true;
-
- old = bch2_btree_path_peek_slot(i->path, &unpacked);
- _deleted.p = i->path->pos;
-
- ret = bch2_trans_mark_key(trans, old, deleted,
- BTREE_TRIGGER_OVERWRITE|i->flags);
-
- if (ret == -EINTR)
- trace_trans_restart_mark(trans->ip, _RET_IP_,
- i->btree_id, &i->path->pos);
- if (ret)
- return ret;
- }
- } while (trans_trigger_run);
+ trans_for_each_update(trans, i) {
+ ret = bch2_journal_key_insert(c, i->btree_id, i->level, i->k);
+ if (ret)
+ break;
}
- trans_for_each_update(trans, i)
- BUG_ON(!(i->flags & BTREE_TRIGGER_NORUN) &&
- (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->bkey_type)) &&
- (!i->insert_trigger_run || !i->overwrite_trigger_run));
-
- return 0;
+ return ret;
}
int __bch2_trans_commit(struct btree_trans *trans)
if (trans->flags & BTREE_INSERT_GC_LOCK_HELD)
lockdep_assert_held(&c->gc_lock);
- memset(&trans->journal_preres, 0, sizeof(trans->journal_preres));
-
- trans->journal_u64s = trans->extra_journal_entry_u64s;
- trans->journal_preres_u64s = 0;
-
- trans->journal_transaction_names = READ_ONCE(c->opts.journal_transaction_names);
+ ret = bch2_trans_commit_run_triggers(trans);
+ if (ret)
+ goto out_reset;
- if (trans->journal_transaction_names)
- trans->journal_u64s += JSET_ENTRY_LOG_U64s;
+ if (unlikely(!test_bit(BCH_FS_MAY_GO_RW, &c->flags))) {
+ ret = do_bch2_trans_commit_to_journal_replay(trans);
+ goto out_reset;
+ }
if (!(trans->flags & BTREE_INSERT_NOCHECK_RW) &&
unlikely(!percpu_ref_tryget(&c->writes))) {
goto out_reset;
}
-#ifdef CONFIG_BCACHEFS_DEBUG
- /*
- * if BTREE_TRIGGER_NORUN is set, it means we're probably being called
- * from the key cache flush code:
- */
- trans_for_each_update(trans, i)
- if (!i->cached &&
- !(i->flags & BTREE_TRIGGER_NORUN))
- bch2_btree_key_cache_verify_clean(trans,
- i->btree_id, i->k->k.p);
-#endif
+ memset(&trans->journal_preres, 0, sizeof(trans->journal_preres));
- ret = bch2_trans_commit_run_triggers(trans);
- if (ret)
- goto out;
+ trans->journal_u64s = trans->extra_journal_entry_u64s;
+ trans->journal_preres_u64s = 0;
+
+ trans->journal_transaction_names = READ_ONCE(c->opts.journal_transaction_names);
+
+ if (trans->journal_transaction_names)
+ trans->journal_u64s += JSET_ENTRY_LOG_U64s;
trans_for_each_update(trans, i) {
BUG_ON(!i->path->should_be_locked);
if (unlikely(!bch2_btree_path_upgrade(trans, i->path, i->level + 1))) {
- trace_trans_restart_upgrade(trans->ip, _RET_IP_,
+ trace_trans_restart_upgrade(trans->fn, _RET_IP_,
i->btree_id, &i->path->pos);
ret = btree_trans_restart(trans);
goto out;
struct bkey_s_c k;
int ret;
+ if (!btree_type_has_snapshots(id))
+ return 0;
+
if (!snapshot_t(c, pos.snapshot)->children[0])
return 0;
return ret;
}
-static int bch2_trans_update_extent(struct btree_trans *trans,
- struct btree_iter *orig_iter,
- struct bkey_i *insert,
- enum btree_update_flags flags)
+int bch2_trans_update_extent(struct btree_trans *trans,
+ struct btree_iter *orig_iter,
+ struct bkey_i *insert,
+ enum btree_update_flags flags)
{
struct bch_fs *c = trans->c;
struct btree_iter iter, update_iter;
bkey_reassemble(update, k);
bch2_cut_front(insert->k.p, update);
- bch2_trans_copy_iter(&update_iter, &iter);
- update_iter.pos = update->k.p;
- ret = bch2_trans_update(trans, &update_iter, update,
+ ret = bch2_trans_update_by_path(trans, iter.path, update,
BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|
flags);
- bch2_trans_iter_exit(trans, &update_iter);
-
if (ret)
goto err;
goto out;
return ret;
}
-int __must_check bch2_trans_update(struct btree_trans *trans, struct btree_iter *iter,
- struct bkey_i *k, enum btree_update_flags flags)
+static int __must_check
+bch2_trans_update_by_path(struct btree_trans *trans, struct btree_path *path,
+ struct bkey_i *k, enum btree_update_flags flags)
{
+ struct bch_fs *c = trans->c;
struct btree_insert_entry *i, n;
- BUG_ON(!iter->path->should_be_locked);
-
- if (iter->flags & BTREE_ITER_IS_EXTENTS)
- return bch2_trans_update_extent(trans, iter, k, flags);
+ BUG_ON(!path->should_be_locked);
BUG_ON(trans->nr_updates >= BTREE_ITER_MAX);
- BUG_ON(bpos_cmp(k->k.p, iter->path->pos));
+ BUG_ON(bpos_cmp(k->k.p, path->pos));
n = (struct btree_insert_entry) {
.flags = flags,
- .bkey_type = __btree_node_type(iter->path->level, iter->btree_id),
- .btree_id = iter->btree_id,
- .level = iter->path->level,
- .cached = iter->flags & BTREE_ITER_CACHED,
- .path = iter->path,
+ .bkey_type = __btree_node_type(path->level, path->btree_id),
+ .btree_id = path->btree_id,
+ .level = path->level,
+ .cached = path->cached,
+ .path = path,
.k = k,
.ip_allocated = _RET_IP_,
};
btree_insert_entry_cmp(i - 1, i) >= 0);
#endif
- if (bkey_deleted(&n.k->k) &&
- (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)) {
- int ret = need_whiteout_for_snapshot(trans, n.btree_id, n.k->k.p);
- if (unlikely(ret < 0))
- return ret;
-
- if (ret)
- n.k->k.type = KEY_TYPE_whiteout;
- }
-
/*
* Pending updates are kept sorted: first, find position of new update,
* then delete/trim any updates the new update overwrites:
!btree_insert_entry_cmp(&n, i)) {
BUG_ON(i->insert_trigger_run || i->overwrite_trigger_run);
- /*
- * This is a hack to ensure that inode creates update the btree,
- * not the key cache, which helps with cache coherency issues in
- * other areas:
- */
- if (n.cached && !i->cached) {
- i->k = n.k;
- i->flags = n.flags;
- return 0;
- }
-
bch2_path_put(trans, i->path, true);
- *i = n;
- } else
+ i->flags = n.flags;
+ i->cached = n.cached;
+ i->k = n.k;
+ i->path = n.path;
+ i->ip_allocated = n.ip_allocated;
+ } else {
array_insert_item(trans->updates, trans->nr_updates,
i - trans->updates, n);
- __btree_path_get(n.path, true);
+ i->old_v = bch2_btree_path_peek_slot(path, &i->old_k).v;
+ i->old_btree_u64s = !bkey_deleted(&i->old_k) ? i->old_k.u64s : 0;
+ if (unlikely(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))) {
+ struct bkey_i *j_k =
+ bch2_journal_keys_peek(c, n.btree_id, n.level, k->k.p);
+
+ if (j_k && !bpos_cmp(j_k->k.p, i->k->k.p)) {
+ i->old_k = j_k->k;
+ i->old_v = &j_k->v;
+ }
+ }
+ }
+
+ __btree_path_get(n.path, true);
return 0;
}
+int __must_check bch2_trans_update(struct btree_trans *trans, struct btree_iter *iter,
+ struct bkey_i *k, enum btree_update_flags flags)
+{
+ struct btree_path *path = iter->update_path ?: iter->path;
+ struct bkey_cached *ck;
+ int ret;
+
+ if (iter->flags & BTREE_ITER_IS_EXTENTS)
+ return bch2_trans_update_extent(trans, iter, k, flags);
+
+ if (bkey_deleted(&k->k) &&
+ !(flags & BTREE_UPDATE_KEY_CACHE_RECLAIM) &&
+ (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)) {
+ ret = need_whiteout_for_snapshot(trans, iter->btree_id, k->k.p);
+ if (unlikely(ret < 0))
+ return ret;
+
+ if (ret)
+ k->k.type = KEY_TYPE_whiteout;
+ }
+
+ if (!(flags & BTREE_UPDATE_KEY_CACHE_RECLAIM) &&
+ !path->cached &&
+ !path->level &&
+ btree_id_cached(trans->c, path->btree_id)) {
+ if (!iter->key_cache_path ||
+ !iter->key_cache_path->should_be_locked ||
+ bpos_cmp(iter->key_cache_path->pos, k->k.p)) {
+ if (!iter->key_cache_path)
+ iter->key_cache_path =
+ bch2_path_get(trans, path->btree_id, path->pos, 1, 0,
+ BTREE_ITER_INTENT|
+ BTREE_ITER_CACHED, _THIS_IP_);
+
+ iter->key_cache_path =
+ bch2_btree_path_set_pos(trans, iter->key_cache_path, path->pos,
+ iter->flags & BTREE_ITER_INTENT,
+ _THIS_IP_);
+
+ ret = bch2_btree_path_traverse(trans, iter->key_cache_path,
+ BTREE_ITER_CACHED);
+ if (unlikely(ret))
+ return ret;
+
+ ck = (void *) iter->key_cache_path->l[0].b;
+
+ if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
+ trace_trans_restart_key_cache_raced(trans->fn, _RET_IP_);
+ btree_trans_restart(trans);
+ return -EINTR;
+ }
+
+ iter->key_cache_path->should_be_locked = true;
+ }
+
+ path = iter->key_cache_path;
+ }
+
+ return bch2_trans_update_by_path(trans, path, k, flags);
+}
+
void bch2_trans_commit_hook(struct btree_trans *trans,
struct btree_trans_commit_hook *h)
{
projects / bcachefs-tools-debian / blobdiff