// 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"
+#include "subvolume.h"
#include "super-io.h"
#include <linux/sort.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;
+}
+
+/*
+ * Btree node pointers have a field to stack a pointer to the in memory btree
+ * node; we need to zero out this field when reading in btree nodes, or when
+ * reading in keys from the journal:
+ */
+static void zero_out_btree_mem_ptr(struct journal_keys *keys)
+{
+ struct journal_key *i;
+
+ for (i = keys->d; i < keys->d + keys->nr; i++)
+ if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
+ bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
+}
+
/* iterate over keys read from the journal: */
-struct journal_iter bch2_journal_iter_init(struct journal_keys *keys,
- enum btree_id id)
+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 (struct journal_iter) {
- .keys = keys,
- .k = keys->d,
+ 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 (__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(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;
}
-struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
+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;
+
+ 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)
+{
+ struct journal_key *k = iter->idx - iter->keys->nr
+ ? iter->keys->d + iter->idx : 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->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 = 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 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) {
- if (iter->k == iter->keys->d + iter->keys->nr)
+ 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;
+ }
- if (iter->k->btree_id == iter->btree_id)
- return bkey_i_to_s_c(iter->k->k);
+ 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;
+ }
- iter->k++;
+ if (!bkey_deleted(ret.k))
+ break;
+
+ bch2_btree_and_journal_iter_advance(iter);
}
- return bkey_s_c_null;
+ return ret;
}
-struct bkey_s_c bch2_journal_iter_next(struct journal_iter *iter)
+struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
{
- if (iter->k == iter->keys->d + iter->keys->nr)
- return bkey_s_c_null;
+ bch2_btree_and_journal_iter_advance(iter);
- iter->k++;
- return bch2_journal_iter_peek(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)
+{
+ memset(iter, 0, sizeof(*iter));
+
+ iter->b = b;
+ bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
+ 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, 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 btree_trans *trans, struct btree *b,
+ enum btree_id btree_id,
+ btree_walk_key_fn key_fn)
+{
+ struct bch_fs *c = trans->c;
+ struct btree_and_journal_iter iter;
+ struct bkey_s_c k;
+ struct bkey_buf tmp;
+ struct btree *child;
+ int ret = 0;
+
+ 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) {
+ if (b->c.level) {
+ bch2_bkey_buf_reassemble(&tmp, c, k);
+
+ child = bch2_btree_node_get_noiter(c, tmp.k,
+ b->c.btree_id, b->c.level - 1,
+ false);
+
+ ret = PTR_ERR_OR_ZERO(child);
+ if (ret)
+ break;
+
+ btree_and_journal_iter_prefetch(c, b, iter);
+
+ ret = bch2_btree_and_journal_walk_recurse(trans, child,
+ btree_id, key_fn);
+ six_unlock_read(&child->c.lock);
+ } else {
+ ret = key_fn(trans, k);
+ }
+
+ if (ret)
+ break;
+
+ bch2_btree_and_journal_iter_advance(&iter);
+ }
+
+ bch2_btree_and_journal_iter_exit(&iter);
+ bch2_bkey_buf_exit(&tmp, c);
+ return ret;
+}
+
+int bch2_btree_and_journal_walk(struct btree_trans *trans, enum btree_id btree_id,
+ btree_walk_key_fn key_fn)
+{
+ struct bch_fs *c = trans->c;
+ struct btree *b = c->btree_roots[btree_id].b;
+ int ret = 0;
+
+ if (btree_node_fake(b))
+ return 0;
+
+ six_lock_read(&b->c.lock, NULL, NULL);
+ ret = bch2_btree_and_journal_walk_recurse(trans, b, btree_id, key_fn);
+ six_unlock_read(&b->c.lock);
+
+ return ret;
}
/* sort and dedup all keys in the journal: */
-static void journal_entries_free(struct list_head *list)
+void bch2_journal_entries_free(struct list_head *list)
{
while (!list_empty(list)) {
}
}
+/*
+ * 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 cmp_int(l->btree_id, r->btree_id) ?:
- bkey_cmp(l->pos, r->pos) ?:
+ 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) ?:
cmp_int(l->journal_seq, r->journal_seq) ?:
cmp_int(l->journal_offset, r->journal_offset);
}
-static int journal_sort_seq_cmp(const void *_l, const void *_r)
-{
- const struct journal_key *l = _l;
- const struct journal_key *r = _r;
-
- return cmp_int(l->journal_seq, r->journal_seq) ?:
- cmp_int(l->btree_id, r->btree_id) ?:
- bkey_cmp(l->pos, r->pos);
-}
-
-static void journal_keys_sift(struct journal_keys *keys, struct journal_key *i)
-{
- while (i + 1 < keys->d + keys->nr &&
- journal_sort_key_cmp(i, i + 1) > 0) {
- swap(i[0], i[1]);
- i++;
- }
-}
-
-static void journal_keys_free(struct journal_keys *keys)
+void bch2_journal_keys_free(struct journal_keys *keys)
{
struct journal_key *i;
- for_each_journal_key(*keys, 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 }, keys_deduped = { NULL };
- struct journal_key *i;
+ struct journal_keys keys = { NULL };
+ struct journal_key *src, *dst;
size_t nr_keys = 0;
- list_for_each_entry(p, journal_entries, list)
- for_each_jset_key(k, _n, entry, &p->j)
+ 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.journal_seq_base = keys_deduped.journal_seq_base =
- le64_to_cpu(list_first_entry(journal_entries,
- struct journal_replay,
- list)->j.seq);
+ 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;
- keys_deduped.d = kvmalloc(sizeof(keys.d[0]) * nr_keys * 2, GFP_KERNEL);
- if (!keys_deduped.d)
- goto err;
+ list_for_each_entry(i, journal_entries, list) {
+ if (i->ignore)
+ continue;
- list_for_each_entry(p, journal_entries, list)
- 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,
- .pos = bkey_start_pos(&k->k),
+ .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,
};
+ }
- sort(keys.d, nr_keys, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
-
- i = keys.d;
- while (i < keys.d + keys.nr) {
- if (i + 1 < keys.d + keys.nr &&
- i[0].btree_id == i[1].btree_id &&
- !bkey_cmp(i[0].pos, i[1].pos)) {
- if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
- i++;
- } else {
- bch2_cut_front(i[1].k->k.p, i[0].k);
- i[0].pos = i[1].k->k.p;
- journal_keys_sift(&keys, i);
- }
- continue;
- }
+ sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
- if (i + 1 < keys.d + keys.nr &&
- i[0].btree_id == i[1].btree_id &&
- bkey_cmp(i[0].k->k.p, bkey_start_pos(&i[1].k->k)) > 0) {
- if ((cmp_int(i[0].journal_seq, i[1].journal_seq) ?:
- cmp_int(i[0].journal_offset, i[1].journal_offset)) < 0) {
- if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
- bch2_cut_back(bkey_start_pos(&i[1].k->k), i[0].k);
- } else {
- struct bkey_i *split =
- kmalloc(bkey_bytes(i[0].k), GFP_KERNEL);
-
- if (!split)
- goto err;
-
- bkey_copy(split, i[0].k);
- bch2_cut_back(bkey_start_pos(&i[1].k->k), split);
- keys_deduped.d[keys_deduped.nr++] = (struct journal_key) {
- .btree_id = i[0].btree_id,
- .allocated = true,
- .pos = bkey_start_pos(&split->k),
- .k = split,
- .journal_seq = i[0].journal_seq,
- .journal_offset = i[0].journal_offset,
- };
-
- bch2_cut_front(i[1].k->k.p, i[0].k);
- i[0].pos = i[1].k->k.p;
- journal_keys_sift(&keys, i);
- continue;
- }
- } else {
- if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) >= 0) {
- i[1] = i[0];
- i++;
- continue;
- } else {
- bch2_cut_front(i[0].k->k.p, i[1].k);
- i[1].pos = i[0].k->k.p;
- journal_keys_sift(&keys, i + 1);
- continue;
- }
- }
- }
+ 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++;
- keys_deduped.d[keys_deduped.nr++] = *i++;
+ *dst++ = *src++;
}
- kvfree(keys.d);
- return keys_deduped;
+ keys.nr = dst - keys.d;
err:
- journal_keys_free(&keys_deduped);
- kvfree(keys.d);
- return (struct journal_keys) { NULL };
+ return keys;
}
/* journal replay: */
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)
+static int __bch2_journal_replay_key(struct btree_trans *trans,
+ struct journal_key *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;
+ struct btree_iter iter;
+ unsigned iter_flags =
+ BTREE_ITER_INTENT|
+ BTREE_ITER_NOT_EXTENTS;
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_iter = bch2_trans_copy_iter(&trans, iter);
- ret = PTR_ERR_OR_ZERO(split_iter);
- if (ret)
- goto err;
-
- split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
- ret = PTR_ERR_OR_ZERO(split);
- if (ret)
- goto err;
+ if (!k->level && k->btree_id == BTREE_ID_alloc)
+ iter_flags |= BTREE_ITER_CACHED|BTREE_ITER_CACHED_NOFILL;
- 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);
+ bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
+ BTREE_MAX_DEPTH, k->level,
+ iter_flags);
+ ret = bch2_btree_iter_traverse(&iter) ?:
+ bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
+ bch2_trans_iter_exit(trans, &iter);
+ return ret;
+}
- remark = true;
- }
+static int bch2_journal_replay_key(struct bch_fs *c, struct journal_key *k)
+{
+ unsigned commit_flags =
+ BTREE_INSERT_LAZY_RW|
+ BTREE_INSERT_NOFAIL|
+ BTREE_INSERT_JOURNAL_RESERVED;
- bkey_copy(split, k);
- bch2_cut_front(split_iter->pos, split);
- bch2_cut_back(atomic_end, split);
-
- bch2_trans_update(&trans, split_iter, split);
- bch2_btree_iter_set_pos(iter, split->k.p);
- } 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),
- BCH_BUCKET_MARK_OVERWRITE) ?:
- bch2_trans_commit(&trans, &disk_res, NULL,
- BTREE_INSERT_ATOMIC|
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW|
- BTREE_INSERT_NOMARK_OVERWRITES|
- BTREE_INSERT_NO_CLEAR_REPLICAS);
- } else {
- ret = bch2_trans_commit(&trans, &disk_res, NULL,
- BTREE_INSERT_ATOMIC|
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW|
- BTREE_INSERT_JOURNAL_REPLAY|
- BTREE_INSERT_NOMARK);
- }
+ if (!k->allocated)
+ commit_flags |= BTREE_INSERT_JOURNAL_REPLAY;
- if (ret)
- goto err;
-err:
- if (ret == -EINTR)
- goto retry;
+ return bch2_trans_do(c, NULL, NULL, commit_flags,
+ __bch2_journal_replay_key(&trans, k));
+}
- bch2_disk_reservation_put(c, &disk_res);
+static int journal_sort_seq_cmp(const void *_l, const void *_r)
+{
+ const struct journal_key *l = _l;
+ const struct journal_key *r = _r;
- return bch2_trans_exit(&trans) ?: ret;
+ return cmp_int(r->level, l->level) ?:
+ cmp_int(l->journal_seq, r->journal_seq) ?:
+ cmp_int(l->btree_id, r->btree_id) ?:
+ bpos_cmp(l->k->k.p, r->k->k.p);
}
static int bch2_journal_replay(struct bch_fs *c,
{
struct journal *j = &c->journal;
struct journal_key *i;
+ u64 seq;
int ret;
sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
- for_each_journal_key(keys, i) {
- replay_now_at(j, keys.journal_seq_base + i->journal_seq);
+ if (keys.nr)
+ replay_now_at(j, keys.journal_seq_base);
- if (i->btree_id == BTREE_ID_ALLOC)
- ret = bch2_alloc_replay_key(c, i->k);
- else if (btree_node_type_is_extents(i->btree_id))
- ret = bch2_extent_replay_key(c, i->btree_id, i->k);
- else
- ret = bch2_btree_insert(c, i->btree_id, i->k,
- NULL, NULL,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW|
- BTREE_INSERT_JOURNAL_REPLAY|
- BTREE_INSERT_NOMARK);
+ seq = j->replay_journal_seq;
- if (ret) {
- bch_err(c, "journal replay: error %d while replaying key",
- ret);
- return ret;
+ /*
+ * First replay updates to the alloc btree - these will only update the
+ * btree key cache:
+ */
+ for_each_journal_key(keys, i) {
+ cond_resched();
+
+ if (!i->level && i->btree_id == BTREE_ID_alloc) {
+ j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
+ ret = bch2_journal_replay_key(c, i);
+ if (ret)
+ goto err;
}
+ }
+ /*
+ * Next replay updates to interior btree nodes:
+ */
+ for_each_journal_key(keys, i) {
cond_resched();
+
+ if (i->level) {
+ j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
+ ret = bch2_journal_replay_key(c, i);
+ if (ret)
+ goto err;
+ }
}
- replay_now_at(j, j->replay_journal_seq_end);
- j->replay_journal_seq = 0;
+ /*
+ * Now that the btree is in a consistent state, we can start journal
+ * reclaim (which will be flushing entries from the btree key cache back
+ * to the btree:
+ */
+ set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
+ set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
+ journal_reclaim_kick(j);
- bch2_journal_set_replay_done(j);
- bch2_journal_flush_all_pins(j);
- return bch2_journal_error(j);
-}
+ j->replay_journal_seq = seq;
-static bool journal_empty(struct list_head *journal)
-{
- return list_empty(journal) ||
- journal_entry_empty(&list_last_entry(journal,
- struct journal_replay, list)->j);
-}
+ /*
+ * Now replay leaf node updates:
+ */
+ for_each_journal_key(keys, i) {
+ cond_resched();
-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;
+ if (i->level || i->btree_id == BTREE_ID_alloc)
+ continue;
- list_for_each_entry(i, journal, list) {
- 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);
+ replay_now_at(j, keys.journal_seq_base + i->journal_seq);
- seq = le64_to_cpu(i->j.seq);
+ ret = bch2_journal_replay_key(c, i);
+ if (ret)
+ goto err;
+ }
- fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
- "found blacklisted journal entry %llu", seq);
+ replay_now_at(j, j->replay_journal_seq_end);
+ j->replay_journal_seq = 0;
- do {
- seq++;
- } while (bch2_journal_seq_is_blacklisted(c, seq, false));
- }
-fsck_err:
+ bch2_journal_set_replay_done(j);
+ bch2_journal_flush_all_pins(j);
+ return bch2_journal_error(j);
+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, le32_to_cpu(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 < min_t(unsigned, nr_types, BCH_DATA_NR); 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, le64_to_cpu(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;
unsigned l1 = 0, l2 = 0;
k1->k.u64s != k2->k.u64s ||
memcmp(k1, k2, bkey_bytes(k1)) ||
l1 != l2, c,
- "superblock btree root doesn't match journal after clean shutdown");
+ "superblock btree root %u doesn't match journal after clean shutdown\n"
+ "sb: l=%u %s\n"
+ "journal: l=%u %s\n", i,
+ l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
+ l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
}
fsck_err:
return ret;
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);
}
}
return ret;
}
+static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
+{
+ struct bkey_i_snapshot root_snapshot;
+ struct bkey_i_subvolume root_volume;
+ int ret;
+
+ bkey_snapshot_init(&root_snapshot.k_i);
+ root_snapshot.k.p.offset = U32_MAX;
+ root_snapshot.v.flags = 0;
+ root_snapshot.v.parent = 0;
+ root_snapshot.v.subvol = BCACHEFS_ROOT_SUBVOL;
+ root_snapshot.v.pad = 0;
+ SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
+
+ ret = bch2_btree_insert(c, BTREE_ID_snapshots,
+ &root_snapshot.k_i,
+ NULL, NULL, 0);
+ if (ret)
+ return ret;
+
+
+ bkey_subvolume_init(&root_volume.k_i);
+ root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
+ root_volume.v.flags = 0;
+ root_volume.v.snapshot = cpu_to_le32(U32_MAX);
+ root_volume.v.inode = cpu_to_le64(BCACHEFS_ROOT_INO);
+
+ ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
+ &root_volume.k_i,
+ NULL, NULL, 0);
+ if (ret)
+ return ret;
+
+ return 0;
+}
+
+static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
+{
+ struct bch_fs *c = trans->c;
+ struct btree_iter iter;
+ struct bkey_s_c k;
+ struct bch_inode_unpacked inode;
+ int ret;
+
+ bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
+ SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
+ k = bch2_btree_iter_peek_slot(&iter);
+ ret = bkey_err(k);
+ if (ret)
+ goto err;
+
+ if (!bkey_is_inode(k.k)) {
+ bch_err(c, "root inode not found");
+ ret = -ENOENT;
+ goto err;
+ }
+
+ ret = bch2_inode_unpack(k, &inode);
+ BUG_ON(ret);
+
+ inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
+
+ ret = bch2_inode_write(trans, &iter, &inode);
+err:
+ bch2_trans_iter_exit(trans, &iter);
+ return ret;
+}
+
int bch2_fs_recovery(struct bch_fs *c)
{
const char *err = "cannot allocate memory";
struct bch_sb_field_clean *clean = NULL;
- u64 journal_seq;
- LIST_HEAD(journal_entries);
- struct journal_keys journal_keys = { NULL };
- bool wrote = false, write_sb = false;
- int ret;
+ struct jset *last_journal_entry = NULL;
+ u64 blacklist_seq, journal_seq;
+ bool write_sb = false;
+ int ret = 0;
if (c->sb.clean)
clean = read_superblock_clean(c);
if (c->sb.clean)
bch_info(c, "recovering from clean shutdown, journal seq %llu",
le64_to_cpu(clean->journal_seq));
+ else
+ bch_info(c, "recovering from unclean shutdown");
+
+ 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.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) {
+ if (!c->replicas.entries ||
+ c->opts.rebuild_replicas) {
bch_info(c, "building replicas info");
set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
}
- if (!c->sb.clean || c->opts.fsck) {
- struct jset *j;
+ if (!c->opts.nochanges) {
+ if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
+ bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
+ c->opts.version_upgrade = true;
+ c->opts.fsck = true;
+ c->opts.fix_errors = FSCK_OPT_YES;
+ } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
+ bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
+ c->opts.version_upgrade = true;
+ c->opts.fsck = true;
+ } else if (c->sb.version < bcachefs_metadata_version_inode_v2) {
+ bch_info(c, "filesystem version is prior to inode_v2 - upgrading");
+ c->opts.version_upgrade = true;
+ }
+ }
+
+ ret = bch2_blacklist_table_initialize(c);
+ if (ret) {
+ bch_err(c, "error initializing blacklist table");
+ goto err;
+ }
- ret = bch2_journal_read(c, &journal_entries);
+ if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
+ struct journal_replay *i;
+
+ 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(&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(&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;
}
- journal_keys = journal_keys_sort(&journal_entries);
- if (!journal_keys.d) {
+ c->journal_keys = journal_keys_sort(&c->journal_entries);
+ if (!c->journal_keys.d) {
ret = -ENOMEM;
goto err;
}
- j = &list_last_entry(&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->opts.reconstruct_alloc) {
+ c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
+ drop_alloc_keys(&c->journal_keys);
}
- ret = journal_replay_early(c, clean, &journal_entries);
+ zero_out_btree_mem_ptr(&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(&journal_entries)) {
- ret = verify_journal_entries_not_blacklisted_or_missing(c,
- &journal_entries);
- if (ret)
- goto err;
}
ret = bch2_fs_journal_start(&c->journal, journal_seq,
- &journal_entries);
+ &c->journal_entries);
if (ret)
goto err;
bch_verbose(c, "starting alloc read");
err = "error reading allocation information";
- ret = bch2_alloc_read(c, &journal_keys);
+ ret = bch2_alloc_read(c);
if (ret)
goto err;
bch_verbose(c, "alloc read done");
bch_verbose(c, "starting stripes_read");
err = "error reading stripes";
- ret = bch2_stripes_read(c, &journal_keys);
+ ret = bch2_stripes_read(c);
if (ret)
goto err;
bch_verbose(c, "stripes_read done");
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, NULL, 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, &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);
bch_verbose(c, "starting journal replay");
err = "journal replay failed";
- ret = bch2_journal_replay(c, journal_keys);
+ ret = bch2_journal_replay(c, c->journal_keys);
if (ret)
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_alloc_write_all(c, BTREE_INSERT_LAZY_RW);
if (ret) {
bch_err(c, "error writing alloc info");
goto err;
}
bch_verbose(c, "alloc write done");
+ }
+
+ if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
+ bch2_fs_lazy_rw(c);
- set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
+ err = "error creating root snapshot node";
+ ret = bch2_fs_initialize_subvolumes(c);
+ if (ret)
+ goto err;
}
- 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");
+ bch_verbose(c, "reading snapshots table");
+ err = "error reading snapshots table";
+ ret = bch2_fs_snapshots_start(c);
+ if (ret)
+ goto err;
+ bch_verbose(c, "reading snapshots 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->sb.version < bcachefs_metadata_version_snapshot_2) {
+ /* set bi_subvol on root inode */
+ err = "error upgrade root inode for subvolumes";
+ ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
+ bch2_fs_upgrade_for_subvolumes(&trans));
+ if (ret)
+ goto err;
}
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;
+
+ bch_move_stats_init(&stats, "recovery");
+
+ 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->version = cpu_to_le16(bcachefs_metadata_version_current);
+ c->disk_sb.sb->features[0] |= cpu_to_le64(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] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
write_sb = true;
}
if (c->opts.fsck &&
- !test_bit(BCH_FS_ERROR, &c->flags)) {
- c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
+ !test_bit(BCH_FS_ERROR, &c->flags) &&
+ !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
+ SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
write_sb = true;
}
if (c->journal_seq_blacklist_table &&
c->journal_seq_blacklist_table->nr > 128)
queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
-out:
+
ret = 0;
-err:
-fsck_err:
+out:
set_bit(BCH_FS_FSCK_DONE, &c->flags);
bch2_flush_fsck_errs(c);
- journal_keys_free(&journal_keys);
- journal_entries_free(&journal_entries);
+ if (!c->opts.keep_journal) {
+ bch2_journal_keys_free(&c->journal_keys);
+ bch2_journal_entries_free(&c->journal_entries);
+ }
kfree(clean);
if (ret)
bch_err(c, "Error in recovery: %s (%i)", err, ret);
else
bch_verbose(c, "ret %i", ret);
return ret;
+err:
+fsck_err:
+ bch2_fs_emergency_read_only(c);
+ goto out;
}
int bch2_fs_initialize(struct bch_fs *c)
bch_notice(c, "initializing new filesystem");
mutex_lock(&c->sb_lock);
+ c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
+ c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
+
+ if (c->opts.version_upgrade) {
+ c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
+ c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
+ bch2_write_super(c);
+ }
+
for_each_online_member(ca, c, i)
bch2_mark_dev_superblock(c, ca, 0);
mutex_unlock(&c->sb_lock);
for (i = 0; i < BTREE_ID_NR; i++)
bch2_btree_root_alloc(c, i);
+ set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
+ set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
+
err = "unable to allocate journal buckets";
for_each_online_member(ca, c, i) {
ret = bch2_dev_journal_alloc(ca);
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(c, true);
+ 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 marking superblock and journal";
+ for_each_member_device(ca, c, i) {
+ ret = bch2_trans_mark_dev_sb(c, ca);
+ if (ret) {
+ percpu_ref_put(&ca->ref);
+ goto err;
+ }
+ }
+
+ err = "error creating root snapshot node";
+ ret = bch2_fs_initialize_subvolumes(c);
+ if (ret)
+ goto err;
+
+ bch_verbose(c, "reading snapshots table");
+ err = "error reading snapshots table";
+ ret = bch2_fs_snapshots_start(c);
+ if (ret)
+ goto err;
+ bch_verbose(c, "reading snapshots done");
+
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);
+ root_inode.bi_inum = BCACHEFS_ROOT_INO;
+ root_inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
+ 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, 0);
if (ret)
bch2_inode_init_early(c, &lostfound_inode);
err = "error creating lost+found";
- ret = bch2_trans_do(c, NULL, BTREE_INSERT_ATOMIC,
- bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
+ ret = bch2_trans_do(c, NULL, NULL, 0,
+ bch2_create_trans(&trans,
+ BCACHEFS_ROOT_SUBVOL_INUM,
&root_inode, &lostfound_inode,
&lostfound,
0, 0, S_IFDIR|0700, 0,
- NULL, NULL));
- if (ret)
+ NULL, NULL, (subvol_inum) { 0 }, 0));
+ if (ret) {
+ bch_err(c, "error creating lost+found");
goto err;
+ }
if (enabled_qtypes(c)) {
ret = bch2_fs_quota_read(c);
}
err = "error writing first journal entry";
- ret = bch2_journal_meta(&c->journal);
+ ret = bch2_journal_flush(&c->journal);
if (ret)
goto err;
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;
-
SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
SET_BCH_SB_CLEAN(c->disk_sb.sb, false);