// SPDX-License-Identifier: GPL-2.0
#include "bcachefs.h"
+#include "bkey_buf.h"
#include "alloc_background.h"
#include "btree_gc.h"
#include "btree_update.h"
#include "journal_io.h"
#include "journal_reclaim.h"
#include "journal_seq_blacklist.h"
+#include "move.h"
#include "quota.h"
#include "recovery.h"
#include "replicas.h"
#define QSTR(n) { { { .len = strlen(n) } }, .name = n }
+/* for -o reconstruct_alloc: */
+static void drop_alloc_keys(struct journal_keys *keys)
+{
+ size_t src, dst;
+
+ for (src = 0, dst = 0; src < keys->nr; src++)
+ if (keys->d[src].btree_id != BTREE_ID_alloc)
+ keys->d[dst++] = keys->d[src];
+
+ keys->nr = dst;
+}
+
/* iterate over keys read from the journal: */
-static struct journal_key *journal_key_search(struct journal_keys *journal_keys,
- enum btree_id id, unsigned level,
- struct bpos pos)
+static int __journal_key_cmp(enum btree_id l_btree_id,
+ unsigned l_level,
+ struct bpos l_pos,
+ 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(struct journal_key *l, struct journal_key *r)
+{
+ return (cmp_int(l->btree_id, r->btree_id) ?:
+ cmp_int(l->level, r->level) ?:
+ bpos_cmp(l->k->k.p, r->k->k.p));
+}
+
+static size_t 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 ((cmp_int(id, journal_keys->d[m].btree_id) ?:
- cmp_int(level, journal_keys->d[m].level) ?:
- bkey_cmp(pos, journal_keys->d[m].k->k.p)) > 0)
+ if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0)
l = m + 1;
else
r = m;
}
BUG_ON(l < journal_keys->nr &&
- (cmp_int(id, journal_keys->d[l].btree_id) ?:
- cmp_int(level, journal_keys->d[l].level) ?:
- bkey_cmp(pos, journal_keys->d[l].k->k.p)) > 0);
+ __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0);
BUG_ON(l &&
- (cmp_int(id, journal_keys->d[l - 1].btree_id) ?:
- cmp_int(level, journal_keys->d[l - 1].level) ?:
- bkey_cmp(pos, journal_keys->d[l - 1].k->k.p)) <= 0);
+ __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(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
+ };
+ struct journal_keys *keys = &c->journal_keys;
+ struct journal_iter *iter;
+ unsigned idx = journal_key_search(keys, id, level, k->k.p);
+
+ 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;
+}
+
+int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
+ unsigned level, struct bpos pos)
+{
+ struct bkey_i *whiteout =
+ kmalloc(sizeof(struct bkey), GFP_KERNEL);
+ int ret;
+
+ if (!whiteout) {
+ bch_err(c, "%s: error allocating new key", __func__);
+ return -ENOMEM;
+ }
+
+ bkey_init(&whiteout->k);
+ whiteout->k.p = pos;
- return l < journal_keys->nr ? journal_keys->d + l : NULL;
+ ret = bch2_journal_key_insert(c, id, level, whiteout);
+ if (ret)
+ kfree(whiteout);
+ return ret;
}
static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
{
- if (iter->k &&
- iter->k < iter->keys->d + iter->keys->nr &&
- iter->k->btree_id == iter->btree_id &&
- iter->k->level == iter->level)
- return iter->k->k;
+ struct journal_key *k = iter->idx - iter->keys->nr
+ ? iter->keys->d + iter->idx : NULL;
- iter->k = NULL;
+ if (k &&
+ k->btree_id == iter->btree_id &&
+ k->level == iter->level)
+ return k->k;
+
+ iter->idx = iter->keys->nr;
return NULL;
}
static void bch2_journal_iter_advance(struct journal_iter *iter)
{
- if (iter->k)
- iter->k++;
+ 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 journal_iter *iter,
- struct journal_keys *journal_keys,
+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 = journal_keys;
- iter->k = journal_key_search(journal_keys, id, level, pos);
+ iter->keys = &c->journal_keys;
+ iter->idx = journal_key_search(&c->journal_keys, id, level, pos);
+ list_add(&iter->list, &c->journal_iters);
}
static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
{
- return iter->btree
- ? bch2_btree_iter_peek(iter->btree)
- : bch2_btree_node_iter_peek_unpack(&iter->node_iter,
- iter->b, &iter->unpacked);
+ 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)
{
- if (iter->btree)
- bch2_btree_iter_next(iter->btree);
- else
- bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
+ bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
}
void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
if (btree_k.k && journal_k.k) {
- int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
+ int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
if (!cmp)
bch2_journal_iter_advance_btree(iter);
ret = iter->last == journal ? journal_k : btree_k;
if (iter->b &&
- bkey_cmp(ret.k->p, iter->b->data->max_key) > 0) {
- iter->journal.k = NULL;
+ 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;
}
return bch2_btree_and_journal_iter_peek(iter);
}
-void bch2_btree_and_journal_iter_init(struct btree_and_journal_iter *iter,
- struct btree_trans *trans,
- struct journal_keys *journal_keys,
- enum btree_id id, struct bpos pos)
+void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
{
- memset(iter, 0, sizeof(*iter));
-
- iter->btree = bch2_trans_get_iter(trans, id, pos, 0);
- bch2_journal_iter_init(&iter->journal, journal_keys, id, 0, pos);
+ bch2_journal_iter_exit(&iter->journal);
}
void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
- struct journal_keys *journal_keys,
+ struct bch_fs *c,
struct btree *b)
{
memset(iter, 0, sizeof(*iter));
iter->b = b;
bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
- bch2_journal_iter_init(&iter->journal, journal_keys,
+ bch2_journal_iter_init(c, &iter->journal,
b->c.btree_id, b->c.level, b->data->min_key);
}
/* Walk btree, overlaying keys from the journal: */
+static void btree_and_journal_iter_prefetch(struct bch_fs *c, struct btree *b,
+ struct btree_and_journal_iter iter)
+{
+ unsigned i = 0, nr = b->c.level > 1 ? 2 : 16;
+ struct bkey_s_c k;
+ struct bkey_buf tmp;
+
+ BUG_ON(!b->c.level);
+
+ bch2_bkey_buf_init(&tmp);
+
+ while (i < nr &&
+ (k = bch2_btree_and_journal_iter_peek(&iter)).k) {
+ bch2_bkey_buf_reassemble(&tmp, c, k);
+
+ bch2_btree_node_prefetch(c, NULL, tmp.k,
+ b->c.btree_id, b->c.level - 1);
+
+ bch2_btree_and_journal_iter_advance(&iter);
+ i++;
+ }
+
+ bch2_bkey_buf_exit(&tmp, c);
+}
+
static int bch2_btree_and_journal_walk_recurse(struct bch_fs *c, struct btree *b,
struct journal_keys *journal_keys,
enum btree_id btree_id,
{
struct btree_and_journal_iter iter;
struct bkey_s_c k;
+ struct bkey_buf tmp;
+ struct btree *child;
int ret = 0;
- bch2_btree_and_journal_iter_init_node_iter(&iter, journal_keys, b);
+ bch2_bkey_buf_init(&tmp);
+ bch2_btree_and_journal_iter_init_node_iter(&iter, c, b);
while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
ret = key_fn(c, btree_id, b->c.level, k);
break;
if (b->c.level) {
- struct btree *child;
- BKEY_PADDED(k) tmp;
-
- bkey_reassemble(&tmp.k, k);
- k = bkey_i_to_s_c(&tmp.k);
+ bch2_bkey_buf_reassemble(&tmp, c, k);
bch2_btree_and_journal_iter_advance(&iter);
- if (b->c.level > 0) {
- child = bch2_btree_node_get_noiter(c, &tmp.k,
- b->c.btree_id, b->c.level - 1);
- ret = PTR_ERR_OR_ZERO(child);
- if (ret)
- break;
+ child = bch2_btree_node_get_noiter(c, tmp.k,
+ b->c.btree_id, b->c.level - 1,
+ false);
- ret = (node_fn ? node_fn(c, b) : 0) ?:
- bch2_btree_and_journal_walk_recurse(c, child,
- journal_keys, btree_id, node_fn, key_fn);
- six_unlock_read(&child->c.lock);
+ ret = PTR_ERR_OR_ZERO(child);
+ if (ret)
+ break;
- if (ret)
- break;
- }
+ btree_and_journal_iter_prefetch(c, b, iter);
+
+ ret = (node_fn ? node_fn(c, b) : 0) ?:
+ bch2_btree_and_journal_walk_recurse(c, child,
+ journal_keys, btree_id, node_fn, key_fn);
+ six_unlock_read(&child->c.lock);
+
+ if (ret)
+ break;
} else {
bch2_btree_and_journal_iter_advance(&iter);
}
}
+ bch2_btree_and_journal_iter_exit(&iter);
+ bch2_bkey_buf_exit(&tmp, c);
return ret;
}
return cmp_int(l->btree_id, r->btree_id) ?:
cmp_int(l->level, r->level) ?:
- bkey_cmp(l->k->k.p, r->k->k.p) ?:
+ bpos_cmp(l->k->k.p, r->k->k.p) ?:
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 *p;
+ struct journal_replay *i;
struct jset_entry *entry;
struct bkey_i *k, *_n;
struct journal_keys keys = { NULL };
if (list_empty(journal_entries))
return keys;
- keys.journal_seq_base =
- le64_to_cpu(list_last_entry(journal_entries,
- struct journal_replay, list)->j.last_seq);
-
- list_for_each_entry(p, journal_entries, list) {
- if (le64_to_cpu(p->j.seq) < keys.journal_seq_base)
+ list_for_each_entry(i, journal_entries, list) {
+ if (i->ignore)
continue;
- for_each_jset_key(k, _n, entry, &p->j)
+ 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]) * nr_keys, GFP_KERNEL);
+ keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
if (!keys.d)
goto err;
- list_for_each_entry(p, journal_entries, list) {
- if (le64_to_cpu(p->j.seq) < keys.journal_seq_base)
+ list_for_each_entry(i, journal_entries, list) {
+ if (i->ignore)
continue;
- for_each_jset_key(k, _n, entry, &p->j)
+ 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(p->j.seq) -
+ .journal_seq = le64_to_cpu(i->j.seq) -
keys.journal_seq_base,
- .journal_offset = k->_data - p->j._data,
+ .journal_offset = k->_data - i->j._data,
};
}
while (src + 1 < keys.d + keys.nr &&
src[0].btree_id == src[1].btree_id &&
src[0].level == src[1].level &&
- !bkey_cmp(src[0].k->k.p, src[1].k->k.p))
+ !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
src++;
*dst++ = *src++;
bch2_journal_pin_put(j, j->replay_journal_seq++);
}
-static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
- struct bkey_i *k)
-{
- struct btree_trans trans;
- struct btree_iter *iter, *split_iter;
- /*
- * We might cause compressed extents to be split, so we need to pass in
- * a disk_reservation:
- */
- struct disk_reservation disk_res =
- bch2_disk_reservation_init(c, 0);
- struct bkey_i *split;
- struct bpos atomic_end;
- /*
- * Some extents aren't equivalent - w.r.t. what the triggers do
- * - if they're split:
- */
- bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
- k->k.type == KEY_TYPE_reflink_p;
- bool remark = false;
- int ret;
-
- bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
-retry:
- bch2_trans_begin(&trans);
-
- iter = bch2_trans_get_iter(&trans, btree_id,
- bkey_start_pos(&k->k),
- BTREE_ITER_INTENT);
-
- do {
- ret = bch2_btree_iter_traverse(iter);
- if (ret)
- goto err;
-
- atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
-
- split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
- ret = PTR_ERR_OR_ZERO(split);
- if (ret)
- goto err;
-
- if (!remark &&
- remark_if_split &&
- bkey_cmp(atomic_end, k->k.p) < 0) {
- ret = bch2_disk_reservation_add(c, &disk_res,
- k->k.size *
- bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
- BCH_DISK_RESERVATION_NOFAIL);
- BUG_ON(ret);
-
- remark = true;
- }
-
- bkey_copy(split, k);
- bch2_cut_front(iter->pos, split);
- bch2_cut_back(atomic_end, split);
-
- split_iter = bch2_trans_copy_iter(&trans, iter);
- ret = PTR_ERR_OR_ZERO(split_iter);
- if (ret)
- goto err;
-
- /*
- * It's important that we don't go through the
- * extent_handle_overwrites() and extent_update_to_keys() path
- * here: journal replay is supposed to treat extents like
- * regular keys
- */
- __bch2_btree_iter_set_pos(split_iter, split->k.p, false);
- bch2_trans_update(&trans, split_iter, split,
- BTREE_TRIGGER_NORUN);
-
- bch2_btree_iter_set_pos(iter, split->k.p);
-
- if (remark) {
- ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(split),
- 0, split->k.size,
- BTREE_TRIGGER_INSERT);
- if (ret)
- goto err;
- }
- } while (bkey_cmp(iter->pos, k->k.p) < 0);
-
- if (remark) {
- ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
- 0, -((s64) k->k.size),
- BTREE_TRIGGER_OVERWRITE);
- if (ret)
- goto err;
- }
-
- ret = bch2_trans_commit(&trans, &disk_res, NULL,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW|
- BTREE_INSERT_JOURNAL_REPLAY);
-err:
- if (ret == -EINTR)
- goto retry;
-
- bch2_disk_reservation_put(c, &disk_res);
-
- return bch2_trans_exit(&trans) ?: ret;
-}
-
static int __bch2_journal_replay_key(struct btree_trans *trans,
enum btree_id id, unsigned level,
struct bkey_i *k)
iter = bch2_trans_get_node_iter(trans, id, k->k.p,
BTREE_MAX_DEPTH, level,
BTREE_ITER_INTENT);
- if (IS_ERR(iter))
- return PTR_ERR(iter);
/*
* iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run
* want that here, journal replay is supposed to treat extents like
* regular keys:
*/
- __bch2_btree_iter_set_pos(iter, k->k.p, false);
+ BUG_ON(iter->flags & BTREE_ITER_IS_EXTENTS);
ret = bch2_btree_iter_traverse(iter) ?:
bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
return ret;
}
-static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
- unsigned level, struct bkey_i *k)
+static int bch2_journal_replay_key(struct bch_fs *c, struct journal_key *k)
{
- return bch2_trans_do(c, NULL, NULL,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW|
- BTREE_INSERT_JOURNAL_REPLAY,
- __bch2_journal_replay_key(&trans, id, level, k));
+ unsigned commit_flags = BTREE_INSERT_NOFAIL|
+ BTREE_INSERT_LAZY_RW;
+
+ if (!k->allocated)
+ commit_flags |= BTREE_INSERT_JOURNAL_REPLAY;
+
+ return bch2_trans_do(c, NULL, NULL, commit_flags,
+ __bch2_journal_replay_key(&trans, k->btree_id, k->level, k->k));
}
static int __bch2_alloc_replay_key(struct btree_trans *trans, struct bkey_i *k)
struct btree_iter *iter;
int ret;
- iter = bch2_trans_get_iter(trans, BTREE_ID_ALLOC, k->k.p,
+ iter = bch2_trans_get_iter(trans, BTREE_ID_alloc, k->k.p,
BTREE_ITER_CACHED|
BTREE_ITER_CACHED_NOFILL|
BTREE_ITER_INTENT);
- ret = PTR_ERR_OR_ZERO(iter) ?:
- bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
+ ret = bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
bch2_trans_iter_put(trans, iter);
return ret;
}
return cmp_int(r->level, l->level) ?:
cmp_int(l->journal_seq, r->journal_seq) ?:
cmp_int(l->btree_id, r->btree_id) ?:
- bkey_cmp(l->k->k.p, r->k->k.p);
+ bpos_cmp(l->k->k.p, r->k->k.p);
}
static int bch2_journal_replay(struct bch_fs *c,
for_each_journal_key(keys, i) {
cond_resched();
- if (!i->level && i->btree_id == BTREE_ID_ALLOC) {
+ if (!i->level && i->btree_id == BTREE_ID_alloc) {
j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
ret = bch2_alloc_replay_key(c, i->k);
if (ret)
if (i->level) {
j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
- ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
+ ret = bch2_journal_replay_key(c, i);
if (ret)
goto err;
}
*/
set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
+ journal_reclaim_kick(j);
j->replay_journal_seq = seq;
for_each_journal_key(keys, i) {
cond_resched();
- if (i->level || i->btree_id == BTREE_ID_ALLOC)
+ if (i->level || i->btree_id == BTREE_ID_alloc)
continue;
replay_now_at(j, keys.journal_seq_base + i->journal_seq);
- ret = i->k->k.size
- ? bch2_extent_replay_key(c, i->btree_id, i->k)
- : bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
+ ret = bch2_journal_replay_key(c, i);
if (ret)
goto err;
}
bch2_journal_flush_all_pins(j);
return bch2_journal_error(j);
err:
- bch_err(c, "journal replay: error %d while replaying key", ret);
- return ret;
-}
-
-static bool journal_empty(struct list_head *journal)
-{
- return list_empty(journal) ||
- journal_entry_empty(&list_last_entry(journal,
- struct journal_replay, list)->j);
-}
-
-static int
-verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
- struct list_head *journal)
-{
- struct journal_replay *i =
- list_last_entry(journal, struct journal_replay, list);
- u64 start_seq = le64_to_cpu(i->j.last_seq);
- u64 end_seq = le64_to_cpu(i->j.seq);
- u64 seq = start_seq;
- int ret = 0;
-
- list_for_each_entry(i, journal, list) {
- if (le64_to_cpu(i->j.seq) < start_seq)
- continue;
-
- fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
- "journal entries %llu-%llu missing! (replaying %llu-%llu)",
- seq, le64_to_cpu(i->j.seq) - 1,
- start_seq, end_seq);
-
- seq = le64_to_cpu(i->j.seq);
-
- fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
- "found blacklisted journal entry %llu", seq);
-
- do {
- seq++;
- } while (bch2_journal_seq_is_blacklisted(c, seq, false));
- }
-fsck_err:
+ bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
+ ret, bch2_btree_ids[i->btree_id], i->level);
return ret;
}
case BCH_JSET_ENTRY_data_usage: {
struct jset_entry_data_usage *u =
container_of(entry, struct jset_entry_data_usage, entry);
+
ret = bch2_replicas_set_usage(c, &u->r,
le64_to_cpu(u->v));
break;
}
+ case BCH_JSET_ENTRY_dev_usage: {
+ struct jset_entry_dev_usage *u =
+ container_of(entry, struct jset_entry_dev_usage, entry);
+ struct bch_dev *ca = bch_dev_bkey_exists(c, u->dev);
+ unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
+ unsigned nr_types = (bytes - sizeof(struct jset_entry_dev_usage)) /
+ sizeof(struct jset_entry_dev_usage_type);
+ unsigned i;
+
+ ca->usage_base->buckets_ec = le64_to_cpu(u->buckets_ec);
+ ca->usage_base->buckets_unavailable = le64_to_cpu(u->buckets_unavailable);
+
+ for (i = 0; i < nr_types; i++) {
+ ca->usage_base->d[i].buckets = le64_to_cpu(u->d[i].buckets);
+ ca->usage_base->d[i].sectors = le64_to_cpu(u->d[i].sectors);
+ ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
+ }
+
+ break;
+ }
case BCH_JSET_ENTRY_blacklist: {
struct jset_entry_blacklist *bl_entry =
container_of(entry, struct jset_entry_blacklist, entry);
le64_to_cpu(bl_entry->end) + 1);
break;
}
+ case BCH_JSET_ENTRY_clock: {
+ struct jset_entry_clock *clock =
+ container_of(entry, struct jset_entry_clock, entry);
+
+ atomic64_set(&c->io_clock[clock->rw].now, clock->time);
+ }
}
return ret;
struct bch_sb_field_clean *clean,
struct list_head *journal)
{
+ struct journal_replay *i;
struct jset_entry *entry;
int ret;
if (clean) {
- c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
- c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
-
for (entry = clean->start;
entry != vstruct_end(&clean->field);
entry = vstruct_next(entry)) {
return ret;
}
} else {
- struct journal_replay *i =
- list_last_entry(journal, struct journal_replay, list);
-
- c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
- c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
+ list_for_each_entry(i, journal, list) {
+ if (i->ignore)
+ continue;
- list_for_each_entry(i, journal, list)
vstruct_for_each(&i->j, entry) {
ret = journal_replay_entry_early(c, entry);
if (ret)
return ret;
}
+ }
}
bch2_fs_usage_initialize(c);
struct bch_sb_field_clean *clean = *cleanp;
int ret = 0;
- if (!c->sb.clean || !j)
- return 0;
-
if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
"superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
le64_to_cpu(clean->journal_seq),
return 0;
}
- mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
- "superblock read clock doesn't match journal after clean shutdown");
- mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
- "superblock read clock doesn't match journal after clean shutdown");
-
for (i = 0; i < BTREE_ID_NR; i++) {
char buf1[200], buf2[200];
struct bkey_i *k1, *k2;
return ERR_PTR(-ENOMEM);
}
- if (le16_to_cpu(c->disk_sb.sb->version) <
- bcachefs_metadata_version_bkey_renumber)
- bch2_sb_clean_renumber(clean, READ);
+ ret = bch2_sb_clean_validate(c, clean, READ);
+ if (ret) {
+ mutex_unlock(&c->sb_lock);
+ return ERR_PTR(ret);
+ }
mutex_unlock(&c->sb_lock);
if (!r->alive)
continue;
- if (i == BTREE_ID_ALLOC &&
+ if (i == BTREE_ID_alloc &&
c->opts.reconstruct_alloc) {
- c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
+ c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
continue;
}
-
if (r->error) {
- __fsck_err(c, i == BTREE_ID_ALLOC
+ __fsck_err(c, i == BTREE_ID_alloc
? FSCK_CAN_IGNORE : 0,
"invalid btree root %s",
bch2_btree_ids[i]);
- if (i == BTREE_ID_ALLOC)
- c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
+ if (i == BTREE_ID_alloc)
+ c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
}
ret = bch2_btree_root_read(c, i, &r->key, r->level);
if (ret) {
- __fsck_err(c, i == BTREE_ID_ALLOC
+ __fsck_err(c, i == BTREE_ID_alloc
? FSCK_CAN_IGNORE : 0,
"error reading btree root %s",
bch2_btree_ids[i]);
- if (i == BTREE_ID_ALLOC)
- c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
+ if (i == BTREE_ID_alloc)
+ c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
}
}
{
const char *err = "cannot allocate memory";
struct bch_sb_field_clean *clean = NULL;
- u64 journal_seq;
- bool wrote = false, write_sb = false;
+ struct jset *last_journal_entry = NULL;
+ u64 blacklist_seq, journal_seq;
+ bool write_sb = false;
int ret;
if (c->sb.clean)
bch_info(c, "recovering from clean shutdown, journal seq %llu",
le64_to_cpu(clean->journal_seq));
+ if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
+ bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
+ ret = -EINVAL;
+ goto err;
+ }
+
+ if (!c->sb.clean &&
+ !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
+ bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
+ ret = -EINVAL;
+ goto err;
+ }
+
+ if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
+ bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
+ ret = -EINVAL;
+ goto err;
+
+ }
+
+ if (!c->sb.clean &&
+ !(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
+ bch_info(c, "BCH_FEATURE_atomic_nlink not set and filesystem dirty, fsck required");
+ c->opts.fsck = true;
+ c->opts.fix_errors = FSCK_OPT_YES;
+ }
+
+ if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
+ bch_info(c, "alloc_v2 feature bit not set, fsck required");
+ c->opts.fsck = true;
+ c->opts.fix_errors = FSCK_OPT_YES;
+ }
+
if (!c->replicas.entries ||
c->opts.rebuild_replicas) {
bch_info(c, "building replicas info");
set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
}
+ ret = bch2_blacklist_table_initialize(c);
+ if (ret) {
+ bch_err(c, "error initializing blacklist table");
+ goto err;
+ }
+
if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
- struct jset *j;
+ struct journal_replay *i;
- ret = bch2_journal_read(c, &c->journal_entries);
+ ret = bch2_journal_read(c, &c->journal_entries,
+ &blacklist_seq, &journal_seq);
if (ret)
goto err;
- if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&c->journal_entries), c,
+ list_for_each_entry_reverse(i, &c->journal_entries, list)
+ if (!i->ignore) {
+ last_journal_entry = &i->j;
+ break;
+ }
+
+ if (mustfix_fsck_err_on(c->sb.clean &&
+ last_journal_entry &&
+ !journal_entry_empty(last_journal_entry), c,
"filesystem marked clean but journal not empty")) {
- c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
+ c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
c->sb.clean = false;
}
- if (!c->sb.clean && list_empty(&c->journal_entries)) {
- bch_err(c, "no journal entries found");
- ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
- goto err;
+ if (!last_journal_entry) {
+ fsck_err_on(!c->sb.clean, c, "no journal entries found");
+ goto use_clean;
}
c->journal_keys = journal_keys_sort(&c->journal_entries);
goto err;
}
- j = &list_last_entry(&c->journal_entries,
- struct journal_replay, list)->j;
-
- ret = verify_superblock_clean(c, &clean, j);
- if (ret)
+ if (c->sb.clean && last_journal_entry) {
+ ret = verify_superblock_clean(c, &clean,
+ last_journal_entry);
+ if (ret)
+ goto err;
+ }
+ } else {
+use_clean:
+ if (!clean) {
+ bch_err(c, "no superblock clean section found");
+ ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
goto err;
- journal_seq = le64_to_cpu(j->seq) + 1;
- } else {
- journal_seq = le64_to_cpu(clean->journal_seq) + 1;
+ }
+ blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
}
- if (!c->sb.clean &&
- !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
- bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
- ret = -EINVAL;
- goto err;
+ if (c->opts.reconstruct_alloc) {
+ c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
+ drop_alloc_keys(&c->journal_keys);
}
ret = journal_replay_early(c, clean, &c->journal_entries);
if (ret)
goto err;
- if (!c->sb.clean) {
+ /*
+ * After an unclean shutdown, skip then next few journal sequence
+ * numbers as they may have been referenced by btree writes that
+ * happened before their corresponding journal writes - those btree
+ * writes need to be ignored, by skipping and blacklisting the next few
+ * journal sequence numbers:
+ */
+ if (!c->sb.clean)
+ journal_seq += 8;
+
+ if (blacklist_seq != journal_seq) {
ret = bch2_journal_seq_blacklist_add(c,
- journal_seq,
- journal_seq + 4);
+ blacklist_seq, journal_seq);
if (ret) {
bch_err(c, "error creating new journal seq blacklist entry");
goto err;
}
-
- journal_seq += 4;
- }
-
- ret = bch2_blacklist_table_initialize(c);
-
- if (!list_empty(&c->journal_entries)) {
- ret = verify_journal_entries_not_blacklisted_or_missing(c,
- &c->journal_entries);
- if (ret)
- goto err;
}
ret = bch2_fs_journal_start(&c->journal, journal_seq,
set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
- if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
- !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
- /*
- * interior btree node updates aren't consistent with the
- * journal; after an unclean shutdown we have to walk all
- * pointers to metadata:
- */
- bch_info(c, "starting metadata mark and sweep");
- err = "error in mark and sweep";
- ret = bch2_gc(c, &c->journal_keys, true, true);
- if (ret)
- goto err;
- bch_verbose(c, "mark and sweep done");
- }
-
if (c->opts.fsck ||
- !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
+ !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
+ !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
+ bool metadata_only = c->opts.norecovery;
+
bch_info(c, "starting mark and sweep");
err = "error in mark and sweep";
- ret = bch2_gc(c, &c->journal_keys, true, false);
+ ret = bch2_gc(c, true, metadata_only);
if (ret)
goto err;
bch_verbose(c, "mark and sweep done");
}
+ bch2_stripes_heap_start(c);
+
clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
goto err;
bch_verbose(c, "journal replay done");
- if (!c->opts.nochanges) {
+ if (test_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags) &&
+ !c->opts.nochanges) {
/*
* note that even when filesystem was clean there might be work
* to do here, if we ran gc (because of fsck) which recalculated
*/
bch_verbose(c, "writing allocation info");
err = "error writing out alloc info";
- ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
- bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
+ ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW) ?:
+ bch2_alloc_write(c, BTREE_INSERT_LAZY_RW);
if (ret) {
bch_err(c, "error writing alloc info");
goto err;
}
bch_verbose(c, "alloc write done");
-
- set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
- }
-
- if (!c->sb.clean) {
- if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
- bch_info(c, "checking inode link counts");
- err = "error in recovery";
- ret = bch2_fsck_inode_nlink(c);
- if (ret)
- goto err;
- bch_verbose(c, "check inodes done");
-
- } else {
- bch_verbose(c, "checking for deleted inodes");
- err = "error in recovery";
- ret = bch2_fsck_walk_inodes_only(c);
- if (ret)
- goto err;
- bch_verbose(c, "check inodes done");
- }
}
if (c->opts.fsck) {
if (ret)
goto err;
bch_verbose(c, "fsck done");
+ } else if (!c->sb.clean) {
+ bch_verbose(c, "checking for deleted inodes");
+ err = "error in recovery";
+ ret = bch2_fsck_walk_inodes_only(c);
+ if (ret)
+ goto err;
+ bch_verbose(c, "check inodes done");
}
if (enabled_qtypes(c)) {
bch_verbose(c, "quotas done");
}
+ if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
+ !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
+ struct bch_move_stats stats = { 0 };
+
+ bch_info(c, "scanning for old btree nodes");
+ ret = bch2_fs_read_write(c);
+ if (ret)
+ goto err;
+
+ ret = bch2_scan_old_btree_nodes(c, &stats);
+ if (ret)
+ goto err;
+ bch_info(c, "scanning for old btree nodes done");
+ }
+
mutex_lock(&c->sb_lock);
if (c->opts.version_upgrade) {
- if (c->sb.version < bcachefs_metadata_version_new_versioning)
- c->disk_sb.sb->version_min =
- le16_to_cpu(bcachefs_metadata_version_min);
c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
write_sb = true;
}
if (!test_bit(BCH_FS_ERROR, &c->flags)) {
- c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
+ c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_alloc_info;
write_sb = true;
}
bch_notice(c, "initializing new filesystem");
mutex_lock(&c->sb_lock);
- for_each_online_member(ca, c, i)
- bch2_mark_dev_superblock(c, ca, 0);
- mutex_unlock(&c->sb_lock);
+ c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_extents_above_btree_updates_done;
+ c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_bformat_overflow_done;
- mutex_lock(&c->sb_lock);
- c->disk_sb.sb->version = c->disk_sb.sb->version_min =
- le16_to_cpu(bcachefs_metadata_version_current);
- c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
- c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
+ if (c->opts.version_upgrade) {
+ c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
+ c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
+ bch2_write_super(c);
+ }
- bch2_write_super(c);
+ for_each_online_member(ca, c, i)
+ bch2_mark_dev_superblock(c, ca, 0);
mutex_unlock(&c->sb_lock);
set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
bch2_fs_journal_start(&c->journal, 1, &journal);
bch2_journal_set_replay_done(&c->journal);
+ err = "error going read-write";
+ ret = bch2_fs_read_write_early(c);
+ if (ret)
+ goto err;
+
+ /*
+ * Write out the superblock and journal buckets, now that we can do
+ * btree updates
+ */
+ err = "error writing alloc info";
+ ret = bch2_alloc_write(c, 0);
+ if (ret)
+ goto err;
+
bch2_inode_init(c, &root_inode, 0, 0,
S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
root_inode.bi_inum = BCACHEFS_ROOT_INO;
- bch2_inode_pack(&packed_inode, &root_inode);
+ bch2_inode_pack(c, &packed_inode, &root_inode);
+ packed_inode.inode.k.p.snapshot = U32_MAX;
err = "error creating root directory";
- ret = bch2_btree_insert(c, BTREE_ID_INODES,
+ ret = bch2_btree_insert(c, BTREE_ID_inodes,
&packed_inode.inode.k_i,
- NULL, NULL, BTREE_INSERT_LAZY_RW);
+ NULL, NULL, 0);
if (ret)
goto err;
&lostfound,
0, 0, S_IFDIR|0700, 0,
NULL, NULL));
- if (ret)
+ if (ret) {
+ bch_err(c, "error creating lost+found");
goto err;
+ }
if (enabled_qtypes(c)) {
ret = bch2_fs_quota_read(c);