-/* iterate over keys read from the journal: */
-
-static int __journal_key_cmp(enum btree_id l_btree_id,
- unsigned l_level,
- struct bpos l_pos,
- const struct journal_key *r)
-{
- return (cmp_int(l_btree_id, r->btree_id) ?:
- cmp_int(l_level, r->level) ?:
- bpos_cmp(l_pos, r->k->k.p));
-}
-
-static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
-{
- return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
-}
-
-size_t bch2_journal_key_search(struct journal_keys *journal_keys,
- enum btree_id id, unsigned level,
- struct bpos pos)
-{
- size_t l = 0, r = journal_keys->nr, m;
-
- while (l < r) {
- m = l + ((r - l) >> 1);
- if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0)
- l = m + 1;
- else
- r = m;
- }
-
- BUG_ON(l < journal_keys->nr &&
- __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0);
-
- BUG_ON(l &&
- __journal_key_cmp(id, level, pos, &journal_keys->d[l - 1]) <= 0);
-
- return l;
-}
-
-static void journal_iter_fix(struct bch_fs *c, struct journal_iter *iter, unsigned idx)
-{
- struct bkey_i *n = iter->keys->d[idx].k;
- struct btree_and_journal_iter *biter =
- container_of(iter, struct btree_and_journal_iter, journal);
-
- if (iter->idx > idx ||
- (iter->idx == idx &&
- biter->last &&
- bpos_cmp(n->k.p, biter->unpacked.p) <= 0))
- iter->idx++;
-}
-
-int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
- unsigned level, struct bkey_i *k)
-{
- struct journal_key n = {
- .btree_id = id,
- .level = level,
- .k = k,
- .allocated = true,
- /*
- * Ensure these keys are done last by journal replay, to unblock
- * journal reclaim:
- */
- .journal_seq = U32_MAX,
- };
- struct journal_keys *keys = &c->journal_keys;
- struct journal_iter *iter;
- size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
-
- BUG_ON(test_bit(BCH_FS_RW, &c->flags));
-
- if (idx < keys->nr &&
- journal_key_cmp(&n, &keys->d[idx]) == 0) {
- if (keys->d[idx].allocated)
- kfree(keys->d[idx].k);
- keys->d[idx] = n;
- return 0;
- }
-
- if (keys->nr == keys->size) {
- struct journal_keys new_keys = {
- .nr = keys->nr,
- .size = keys->size * 2,
- .journal_seq_base = keys->journal_seq_base,
- };
-
- new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
- if (!new_keys.d) {
- bch_err(c, "%s: error allocating new key array (size %zu)",
- __func__, new_keys.size);
- return -ENOMEM;
- }
-
- memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
- kvfree(keys->d);
- *keys = new_keys;
- }
-
- array_insert_item(keys->d, keys->nr, idx, n);
-
- list_for_each_entry(iter, &c->journal_iters, list)
- journal_iter_fix(c, iter, idx);
-
- return 0;
-}
-
-/*
- * Can only be used from the recovery thread while we're still RO - can't be
- * used once we've got RW, as journal_keys is at that point used by multiple
- * threads:
- */
-int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
- unsigned level, struct bkey_i *k)
-{
- struct bkey_i *n;
- int ret;
-
- n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
- if (!n)
- return -ENOMEM;
-
- bkey_copy(n, k);
- ret = bch2_journal_key_insert_take(c, id, level, n);
- if (ret)
- kfree(n);
- return ret;
-}
-
-int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
- unsigned level, struct bpos pos)
-{
- struct bkey_i whiteout;
-
- bkey_init(&whiteout.k);
- whiteout.k.p = pos;
-
- return bch2_journal_key_insert(c, id, level, &whiteout);
-}
-
-void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
- unsigned level, struct bpos pos)
-{
- struct journal_keys *keys = &c->journal_keys;
- size_t idx = bch2_journal_key_search(keys, btree, level, pos);
-
- if (idx < keys->nr &&
- keys->d[idx].btree_id == btree &&
- keys->d[idx].level == level &&
- !bpos_cmp(keys->d[idx].k->k.p, pos))
- keys->d[idx].overwritten = true;
-}
-
-static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
-{
- struct journal_key *k = iter->keys->d + iter->idx;
-
- while (k < iter->keys->d + iter->keys->nr &&
- k->btree_id == iter->btree_id &&
- k->level == iter->level) {
- if (!k->overwritten)
- return k->k;
-
- iter->idx++;
- k = iter->keys->d + iter->idx;
- }
-
- return NULL;
-}
-
-static void bch2_journal_iter_advance(struct journal_iter *iter)
-{
- if (iter->idx < iter->keys->nr)
- iter->idx++;
-}
-
-static void bch2_journal_iter_exit(struct journal_iter *iter)
-{
- list_del(&iter->list);
-}
-
-static void bch2_journal_iter_init(struct bch_fs *c,
- struct journal_iter *iter,
- enum btree_id id, unsigned level,
- struct bpos pos)
-{
- iter->btree_id = id;
- iter->level = level;
- iter->keys = &c->journal_keys;
- iter->idx = bch2_journal_key_search(&c->journal_keys, id, level, pos);
-}
-
-static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
-{
- return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
- iter->b, &iter->unpacked);
-}
-
-static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
-{
- bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
-}
-
-void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
-{
- switch (iter->last) {
- case none:
- break;
- case btree:
- bch2_journal_iter_advance_btree(iter);
- break;
- case journal:
- bch2_journal_iter_advance(&iter->journal);
- break;
- }
-
- iter->last = none;
-}
-
-struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
-{
- struct bkey_s_c ret;
-
- while (1) {
- struct bkey_s_c btree_k =
- bch2_journal_iter_peek_btree(iter);
- struct bkey_s_c journal_k =
- bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
-
- if (btree_k.k && journal_k.k) {
- int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
-
- if (!cmp)
- bch2_journal_iter_advance_btree(iter);
-
- iter->last = cmp < 0 ? btree : journal;
- } else if (btree_k.k) {
- iter->last = btree;
- } else if (journal_k.k) {
- iter->last = journal;
- } else {
- iter->last = none;
- return bkey_s_c_null;
- }
-
- ret = iter->last == journal ? journal_k : btree_k;
-
- if (iter->b &&
- bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
- iter->journal.idx = iter->journal.keys->nr;
- iter->last = none;
- return bkey_s_c_null;
- }
-
- if (!bkey_deleted(ret.k))
- break;
-
- bch2_btree_and_journal_iter_advance(iter);
- }
-
- return ret;
-}
-
-struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
-{
- bch2_btree_and_journal_iter_advance(iter);
-
- return bch2_btree_and_journal_iter_peek(iter);
-}
-
-void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
-{
- bch2_journal_iter_exit(&iter->journal);
-}
-
-void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
- struct bch_fs *c,
- struct btree *b,
- struct btree_node_iter node_iter,
- struct bpos pos)
-{
- memset(iter, 0, sizeof(*iter));
-
- iter->b = b;
- iter->node_iter = node_iter;
- bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
- INIT_LIST_HEAD(&iter->journal.list);
-}
-
-/*
- * this version is used by btree_gc before filesystem has gone RW and
- * multithreaded, so uses the journal_iters list:
- */
-void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
- struct bch_fs *c,
- struct btree *b)
-{
- struct btree_node_iter node_iter;
-
- bch2_btree_node_iter_init_from_start(&node_iter, b);
- __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
- list_add(&iter->journal.list, &c->journal_iters);
-}
-
-/* sort and dedup all keys in the journal: */
-
-void bch2_journal_entries_free(struct list_head *list)
-{
-
- while (!list_empty(list)) {
- struct journal_replay *i =
- list_first_entry(list, struct journal_replay, list);
- list_del(&i->list);
- kvpfree(i, offsetof(struct journal_replay, j) +
- vstruct_bytes(&i->j));
- }
-}
-
-/*
- * When keys compare equal, oldest compares first:
- */
-static int journal_sort_key_cmp(const void *_l, const void *_r)
-{
- const struct journal_key *l = _l;
- const struct journal_key *r = _r;
-
- return journal_key_cmp(l, r) ?:
- cmp_int(l->journal_seq, r->journal_seq) ?:
- cmp_int(l->journal_offset, r->journal_offset);
-}
-
-void bch2_journal_keys_free(struct journal_keys *keys)
-{
- struct journal_key *i;
-
- for (i = keys->d; i < keys->d + keys->nr; i++)
- if (i->allocated)
- kfree(i->k);
-
- kvfree(keys->d);
- keys->d = NULL;
- keys->nr = 0;
-}
-
-static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
-{
- struct journal_replay *i;
- struct jset_entry *entry;
- struct bkey_i *k, *_n;
- struct journal_keys keys = { NULL };
- struct journal_key *src, *dst;
- size_t nr_keys = 0;
-
- if (list_empty(journal_entries))
- return keys;
-
- list_for_each_entry(i, journal_entries, list) {
- if (i->ignore)
- continue;
-
- if (!keys.journal_seq_base)
- keys.journal_seq_base = le64_to_cpu(i->j.seq);
-
- for_each_jset_key(k, _n, entry, &i->j)
- nr_keys++;
- }
-
- keys.size = roundup_pow_of_two(nr_keys);
-
- keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
- if (!keys.d)
- goto err;
-
- list_for_each_entry(i, journal_entries, list) {
- if (i->ignore)
- continue;
-
- BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
-
- for_each_jset_key(k, _n, entry, &i->j)
- keys.d[keys.nr++] = (struct journal_key) {
- .btree_id = entry->btree_id,
- .level = entry->level,
- .k = k,
- .journal_seq = le64_to_cpu(i->j.seq) -
- keys.journal_seq_base,
- .journal_offset = k->_data - i->j._data,
- };
- }
-
- sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
-
- src = dst = keys.d;
- while (src < keys.d + keys.nr) {
- while (src + 1 < keys.d + keys.nr &&
- src[0].btree_id == src[1].btree_id &&
- src[0].level == src[1].level &&
- !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
- src++;
-
- *dst++ = *src++;
- }
-
- keys.nr = dst - keys.d;
-err:
- return keys;
-}
-