const struct btree_insert_entry *r)
{
return cmp_int(l->btree_id, r->btree_id) ?:
+ cmp_int(l->cached, r->cached) ?:
-cmp_int(l->level, r->level) ?:
bpos_cmp(l->k->k.p, r->k->k.p);
}
struct bch_fs *c = trans->c;
int ret;
- if (trans->flags & BTREE_INSERT_JOURNAL_RESERVED)
- flags |= JOURNAL_RES_GET_RESERVED;
-
ret = bch2_journal_res_get(&c->journal, &trans->journal_res,
- trans->journal_u64s, flags);
+ trans->journal_u64s,
+ flags|
+ (trans->flags & JOURNAL_WATERMARK_MASK));
return ret == -EAGAIN ? BTREE_INSERT_NEED_JOURNAL_RES : ret;
}
#define JSET_ENTRY_LOG_U64s 4
-static noinline void journal_transaction_name(struct btree_trans *trans)
+static void journal_transaction_name(struct btree_trans *trans)
{
struct bch_fs *c = trans->c;
- struct jset_entry *entry = journal_res_entry(&c->journal, &trans->journal_res);
- struct jset_entry_log *l = container_of(entry, struct jset_entry_log, entry);
- unsigned u64s = JSET_ENTRY_LOG_U64s - 1;
- unsigned b, buflen = u64s * sizeof(u64);
-
- l->entry.u64s = cpu_to_le16(u64s);
- l->entry.btree_id = 0;
- l->entry.level = 0;
- l->entry.type = BCH_JSET_ENTRY_log;
- l->entry.pad[0] = 0;
- l->entry.pad[1] = 0;
- l->entry.pad[2] = 0;
- b = min_t(unsigned, strlen(trans->fn), buflen);
- memcpy(l->d, trans->fn, b);
- while (b < buflen)
- l->d[b++] = '\0';
-
- trans->journal_res.offset += JSET_ENTRY_LOG_U64s;
- trans->journal_res.u64s -= JSET_ENTRY_LOG_U64s;
+ struct journal *j = &c->journal;
+ struct jset_entry *entry =
+ bch2_journal_add_entry(j, &trans->journal_res,
+ BCH_JSET_ENTRY_log, 0, 0,
+ JSET_ENTRY_LOG_U64s);
+ struct jset_entry_log *l =
+ container_of(entry, struct jset_entry_log, entry);
+
+ strncpy(l->d, trans->fn, JSET_ENTRY_LOG_U64s * sizeof(u64));
}
static inline enum btree_insert_ret
{
struct bch_fs *c = trans->c;
struct bkey_cached *ck = (void *) path->l[0].b;
- unsigned new_u64s;
+ unsigned old_u64s = ck->u64s, new_u64s;
struct bkey_i *new_k;
EBUG_ON(path->level);
* transaction restart:
*/
trace_trans_restart_key_cache_key_realloced(trans->fn, _RET_IP_,
- path->btree_id, &path->pos);
+ path->btree_id, &path->pos,
+ old_u64s, new_u64s);
/*
* Not using btree_trans_restart() because we can't unlock here, we have
* write locks held:
return -EINTR;
}
-static inline void do_btree_insert_one(struct btree_trans *trans,
- struct btree_insert_entry *i)
-{
- struct bch_fs *c = trans->c;
- struct journal *j = &c->journal;
-
- EBUG_ON(trans->journal_res.ref !=
- !(trans->flags & BTREE_INSERT_JOURNAL_REPLAY));
-
- i->k->k.needs_whiteout = false;
-
- if (!i->cached)
- btree_insert_key_leaf(trans, i);
- else
- bch2_btree_insert_key_cached(trans, i->path, i->k);
-
- if (likely(!(trans->flags & BTREE_INSERT_JOURNAL_REPLAY))) {
- bch2_journal_add_keys(j, &trans->journal_res,
- i->btree_id,
- i->level,
- i->k);
-
- if (trans->journal_seq)
- *trans->journal_seq = trans->journal_res.seq;
- }
-}
-
/* Triggers: */
static int run_one_mem_trigger(struct btree_trans *trans,
}
static int run_one_trans_trigger(struct btree_trans *trans, struct btree_insert_entry *i,
- bool overwrite)
+ bool overwrite)
{
- struct bkey_s_c old = { &i->old_k, i->old_v };
- int ret = 0;
+ /*
+ * Transactional triggers create new btree_insert_entries, so we can't
+ * pass them a pointer to a btree_insert_entry, that memory is going to
+ * move:
+ */
+ struct bkey old_k = i->old_k;
+ struct bkey_s_c old = { &old_k, i->old_v };
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 (bch2_bkey_ops[old.k->type].trans_trigger ==
- bch2_bkey_ops[i->k->k.type].trans_trigger &&
+ if (!i->insert_trigger_run &&
+ !i->overwrite_trigger_run &&
+ bch2_bkey_ops[old.k->type].trans_trigger ==
+ bch2_bkey_ops[i->k->k.type].trans_trigger &&
((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);
+ i->insert_trigger_run = true;
+ return bch2_trans_mark_key(trans, i->btree_id, i->level, old, i->k,
+ BTREE_TRIGGER_INSERT|
+ BTREE_TRIGGER_OVERWRITE|
+ i->flags) ?: 1;
+ } else if (overwrite && !i->overwrite_trigger_run) {
+ i->overwrite_trigger_run = true;
+ return bch2_trans_mark_old(trans, i->btree_id, i->level, old, i->flags) ?: 1;
+ } else if (!overwrite && !i->insert_trigger_run) {
+ i->insert_trigger_run = true;
+ return bch2_trans_mark_new(trans, i->btree_id, i->level, i->k, i->flags) ?: 1;
} else {
- ret = bch2_trans_mark_new(trans, i->k, i->flags);
+ return 0;
}
-
- 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,
bool trans_trigger_run;
int ret, overwrite;
- for (overwrite = 0; overwrite < 2; overwrite++) {
+ for (overwrite = 1; overwrite >= 0; --overwrite) {
/*
* Running triggers will append more updates to the list of updates as
if (unlikely(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))) {
struct bkey_i *j_k =
- bch2_journal_keys_peek(c, i->btree_id, i->level, i->k->k.p);
+ bch2_journal_keys_peek_slot(c, i->btree_id, i->level,
+ i->k->k.p);
- if (j_k && !bpos_cmp(j_k->k.p, i->k->k.p)) {
+ if (j_k) {
i->old_k = j_k->k;
i->old_v = &j_k->v;
}
if (ret)
return ret;
- if (unlikely(trans->journal_transaction_names))
- journal_transaction_name(trans);
+ journal_transaction_name(trans);
} else {
trans->journal_res.seq = c->journal.replay_journal_seq;
}
- if (unlikely(trans->extra_journal_entry_u64s)) {
+ if (unlikely(trans->extra_journal_entries.nr)) {
memcpy_u64s_small(journal_res_entry(&c->journal, &trans->journal_res),
- trans->extra_journal_entries,
- trans->extra_journal_entry_u64s);
+ trans->extra_journal_entries.data,
+ trans->extra_journal_entries.nr);
- trans->journal_res.offset += trans->extra_journal_entry_u64s;
- trans->journal_res.u64s -= trans->extra_journal_entry_u64s;
+ trans->journal_res.offset += trans->extra_journal_entries.nr;
+ trans->journal_res.u64s -= trans->extra_journal_entries.nr;
}
/*
return ret;
}
- trans_for_each_update(trans, i)
- do_btree_insert_one(trans, i);
+ if (likely(!(trans->flags & BTREE_INSERT_JOURNAL_REPLAY))) {
+ trans_for_each_update(trans, i) {
+ struct journal *j = &c->journal;
+ struct jset_entry *entry;
+
+ if (i->key_cache_already_flushed)
+ continue;
+
+ entry = bch2_journal_add_entry(j, &trans->journal_res,
+ BCH_JSET_ENTRY_overwrite,
+ i->btree_id, i->level,
+ i->old_k.u64s);
+ bkey_reassemble(&entry->start[0],
+ (struct bkey_s_c) { &i->old_k, i->old_v });
+
+ entry = bch2_journal_add_entry(j, &trans->journal_res,
+ BCH_JSET_ENTRY_btree_keys,
+ i->btree_id, i->level,
+ i->k->k.u64s);
+ bkey_copy(&entry->start[0], i->k);
+ }
+
+ if (trans->journal_seq)
+ *trans->journal_seq = trans->journal_res.seq;
+ }
+
+ trans_for_each_update(trans, i) {
+ i->k->k.needs_whiteout = false;
+
+ if (!i->cached)
+ btree_insert_key_leaf(trans, i);
+ else if (!i->key_cache_already_flushed)
+ bch2_btree_insert_key_cached(trans, i->path, i->k);
+ else
+ bch2_btree_key_cache_drop(trans, i->path);
+ }
return ret;
}
{
struct bch_fs *c = trans->c;
struct btree_insert_entry *i;
+ struct printbuf buf = PRINTBUF;
int ret, u64s_delta = 0;
+ int rw = (trans->flags & BTREE_INSERT_JOURNAL_REPLAY) ? READ : WRITE;
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) {
- struct printbuf buf = PRINTBUF;
+ if (bch2_bkey_invalid(c, bkey_i_to_s_c(i->k),
+ i->bkey_type, rw, &buf)) {
+ printbuf_reset(&buf);
+ pr_buf(&buf, "invalid bkey on insert from %s -> %ps",
+ trans->fn, (void *) i->ip_allocated);
+ pr_newline(&buf);
+ pr_indent_push(&buf, 2);
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);
+ pr_newline(&buf);
+
+ bch2_bkey_invalid(c, bkey_i_to_s_c(i->k),
+ i->bkey_type, rw, &buf);
+
+ bch2_trans_inconsistent(trans, "%s", buf.buf);
printbuf_exit(&buf);
return -EINVAL;
}
btree_insert_entry_checks(trans, i);
}
+ printbuf_exit(&buf);
+
trans_for_each_update(trans, i) {
if (i->cached)
continue;
ret = bch2_journal_preres_get(&c->journal,
&trans->journal_preres, trans->journal_preres_u64s,
JOURNAL_RES_GET_NONBLOCK|
- ((trans->flags & BTREE_INSERT_JOURNAL_RESERVED)
- ? JOURNAL_RES_GET_RESERVED : 0));
+ (trans->flags & JOURNAL_WATERMARK_MASK));
if (unlikely(ret == -EAGAIN))
ret = bch2_trans_journal_preres_get_cold(trans,
trans->journal_preres_u64s, trace_ip);
bch2_trans_unlock(trans);
if ((trans->flags & BTREE_INSERT_JOURNAL_RECLAIM) &&
- !(trans->flags & BTREE_INSERT_JOURNAL_RESERVED)) {
+ !(trans->flags & JOURNAL_WATERMARK_reserved)) {
trans->restarted = true;
ret = -EAGAIN;
break;
int ret = 0;
if (!trans->nr_updates &&
- !trans->extra_journal_entry_u64s)
+ !trans->extra_journal_entries.nr)
goto out_reset;
if (trans->flags & BTREE_INSERT_GC_LOCK_HELD)
goto out_reset;
}
+ EBUG_ON(test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags));
+
memset(&trans->journal_preres, 0, sizeof(trans->journal_preres));
- trans->journal_u64s = trans->extra_journal_entry_u64s;
+ trans->journal_u64s = trans->extra_journal_entries.nr;
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;
+ /* For journalling transaction name: */
+ trans->journal_u64s += jset_u64s(JSET_ENTRY_LOG_U64s);
trans_for_each_update(trans, i) {
BUG_ON(!i->path->should_be_locked);
BUG_ON(!btree_node_intent_locked(i->path, i->level));
+ if (i->key_cache_already_flushed)
+ continue;
+
+ /* we're going to journal the key being updated: */
u64s = jset_u64s(i->k->k.u64s);
if (i->cached &&
likely(!(trans->flags & BTREE_INSERT_JOURNAL_REPLAY)))
trans->journal_preres_u64s += u64s;
trans->journal_u64s += u64s;
+
+ /* and we're also going to log the overwrite: */
+ trans->journal_u64s += jset_u64s(i->old_k.u64s);
}
if (trans->extra_journal_res) {
if (ret)
goto err;
+
+ trace_transaction_commit(trans->fn, _RET_IP_);
out:
bch2_journal_preres_put(&c->journal, &trans->journal_preres);
if (likely(!(trans->flags & BTREE_INSERT_NOCHECK_RW)))
percpu_ref_put(&c->writes);
out_reset:
- trans_for_each_update(trans, i)
- bch2_path_put(trans, i->path, true);
-
- trans->extra_journal_res = 0;
- trans->nr_updates = 0;
- trans->hooks = NULL;
- trans->extra_journal_entries = NULL;
- trans->extra_journal_entry_u64s = 0;
+ bch2_trans_reset_updates(trans);
if (trans->fs_usage_deltas) {
trans->fs_usage_deltas->used = 0;
- memset(&trans->fs_usage_deltas->memset_start, 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);
}
}
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)
+bch2_trans_update_by_path_trace(struct btree_trans *trans, struct btree_path *path,
+ struct bkey_i *k, enum btree_update_flags flags,
+ unsigned long ip)
{
struct bch_fs *c = trans->c;
struct btree_insert_entry *i, n;
+ int ret = 0;
BUG_ON(!path->should_be_locked);
.cached = path->cached,
.path = path,
.k = k,
- .ip_allocated = _RET_IP_,
+ .ip_allocated = ip,
};
#ifdef CONFIG_BCACHEFS_DEBUG
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);
+ bch2_journal_keys_peek_slot(c, n.btree_id, n.level, k->k.p);
- if (j_k && !bpos_cmp(j_k->k.p, i->k->k.p)) {
+ if (j_k) {
i->old_k = j_k->k;
i->old_v = &j_k->v;
}
}
}
- __btree_path_get(n.path, true);
- return 0;
+ __btree_path_get(i->path, true);
+
+ /*
+ * If a key is present in the key cache, it must also exist in the
+ * btree - this is necessary for cache coherency. When iterating over
+ * a btree that's cached in the key cache, the btree iter code checks
+ * the key cache - but the key has to exist in the btree for that to
+ * work:
+ */
+ if (path->cached &&
+ bkey_deleted(&i->old_k)) {
+ struct btree_path *btree_path;
+
+ i->key_cache_already_flushed = true;
+ i->flags |= BTREE_TRIGGER_NORUN;
+
+ btree_path = bch2_path_get(trans, path->btree_id, path->pos, 1, 0,
+ BTREE_ITER_INTENT, _THIS_IP_);
+
+ ret = bch2_btree_path_traverse(trans, btree_path, 0);
+ if (ret)
+ goto err;
+
+ btree_path->should_be_locked = true;
+ ret = bch2_trans_update_by_path_trace(trans, btree_path, k, flags, ip);
+err:
+ bch2_path_put(trans, btree_path, true);
+ }
+
+ return ret;
+}
+
+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)
+{
+ return bch2_trans_update_by_path_trace(trans, path, k, flags, _RET_IP_);
}
int __must_check bch2_trans_update(struct btree_trans *trans, struct btree_iter *iter,
k->k.type = KEY_TYPE_whiteout;
}
+ /*
+ * Ensure that updates to cached btrees go to the key cache:
+ */
if (!(flags & BTREE_UPDATE_KEY_CACHE_RECLAIM) &&
!path->cached &&
!path->level &&
__bch2_btree_insert(&trans, id, k));
}
-int bch2_btree_delete_at(struct btree_trans *trans,
- struct btree_iter *iter, unsigned update_flags)
+int bch2_btree_delete_extent_at(struct btree_trans *trans, struct btree_iter *iter,
+ unsigned len, unsigned update_flags)
{
struct bkey_i *k;
bkey_init(&k->k);
k->k.p = iter->pos;
+ bch2_key_resize(&k->k, len);
return bch2_trans_update(trans, iter, k, update_flags);
}
+int bch2_btree_delete_at(struct btree_trans *trans,
+ struct btree_iter *iter, unsigned update_flags)
+{
+ return bch2_btree_delete_extent_at(trans, iter, 0, update_flags);
+}
+
int bch2_btree_delete_range_trans(struct btree_trans *trans, enum btree_id id,
struct bpos start, struct bpos end,
unsigned update_flags,
break;
}
- ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
+ ret = bch2_trans_update(trans, &iter, &delete, update_flags) ?:
bch2_trans_commit(trans, &disk_res, journal_seq,
- BTREE_INSERT_NOFAIL|
- update_flags);
+ BTREE_INSERT_NOFAIL);
bch2_disk_reservation_put(trans->c, &disk_res);
if (ret)
break;
bch2_btree_delete_range_trans(&trans, id, start, end,
update_flags, journal_seq));
}
+
+int bch2_trans_log_msg(struct btree_trans *trans, const char *msg)
+{
+ unsigned len = strlen(msg);
+ unsigned u64s = DIV_ROUND_UP(len, sizeof(u64));
+ struct jset_entry_log *l;
+ int ret;
+
+ ret = darray_make_room(&trans->extra_journal_entries, jset_u64s(u64s));
+ if (ret)
+ return ret;
+
+ l = (void *) &darray_top(trans->extra_journal_entries);
+ l->entry.u64s = cpu_to_le16(u64s);
+ l->entry.btree_id = 0;
+ l->entry.level = 1;
+ l->entry.type = BCH_JSET_ENTRY_log;
+ l->entry.pad[0] = 0;
+ l->entry.pad[1] = 0;
+ l->entry.pad[2] = 0;
+ memcpy(l->d, msg, len);
+ while (len & 7)
+ l->d[len++] = '\0';
+
+ trans->extra_journal_entries.nr += jset_u64s(u64s);
+ return 0;
+}
projects / bcachefs-tools-debian / blobdiff