// SPDX-License-Identifier: GPL-2.0
#include "bcachefs.h"
+#include "bkey_buf.h"
#include "btree_update.h"
#include "dirent.h"
#include "error.h"
-#include "fs.h"
+#include "fs-common.h"
#include "fsck.h"
#include "inode.h"
#include "keylist.h"
u64 sectors = 0;
int ret;
- for_each_btree_key(trans, iter, BTREE_ID_EXTENTS,
+ for_each_btree_key(trans, iter, BTREE_ID_extents,
POS(inum, 0), 0, k, ret) {
if (k.k->p.inode != inum)
break;
return ret ?: sectors;
}
-static int remove_dirent(struct btree_trans *trans,
- struct bkey_s_c_dirent dirent)
+static int __remove_dirent(struct btree_trans *trans,
+ struct bkey_s_c_dirent dirent)
{
struct bch_fs *c = trans->c;
struct qstr name;
char *buf;
name.len = bch2_dirent_name_bytes(dirent);
- buf = kmalloc(name.len + 1, GFP_KERNEL);
- if (!buf)
- return -ENOMEM;
+ buf = bch2_trans_kmalloc(trans, name.len + 1);
+ if (IS_ERR(buf))
+ return PTR_ERR(buf);
memcpy(buf, dirent.v->d_name, name.len);
buf[name.len] = '\0';
name.name = buf;
- /* Unlock so we don't deadlock, after copying name: */
- bch2_trans_unlock(trans);
-
- ret = bch2_inode_find_by_inum(c, dir_inum, &dir_inode);
- if (ret) {
+ ret = __bch2_inode_find_by_inum_trans(trans, dir_inum, &dir_inode, 0);
+ if (ret && ret != -EINTR)
bch_err(c, "remove_dirent: err %i looking up directory inode", ret);
- goto err;
- }
+ if (ret)
+ return ret;
dir_hash_info = bch2_hash_info_init(c, &dir_inode);
- ret = bch2_dirent_delete(c, dir_inum, &dir_hash_info, &name, NULL);
- if (ret)
+ ret = bch2_hash_delete(trans, bch2_dirent_hash_desc,
+ &dir_hash_info, dir_inum, &name);
+ if (ret && ret != -EINTR)
bch_err(c, "remove_dirent: err %i deleting dirent", ret);
-err:
- kfree(buf);
- return ret;
+ if (ret)
+ return ret;
+
+ return 0;
+}
+
+static int remove_dirent(struct btree_trans *trans,
+ struct bkey_s_c_dirent dirent)
+{
+ return __bch2_trans_do(trans, NULL, NULL,
+ BTREE_INSERT_NOFAIL|
+ BTREE_INSERT_LAZY_RW,
+ __remove_dirent(trans, dirent));
}
static int reattach_inode(struct bch_fs *c,
struct bch_inode_unpacked *lostfound_inode,
u64 inum)
{
- struct bch_hash_info lostfound_hash_info =
- bch2_hash_info_init(c, lostfound_inode);
- struct bkey_inode_buf packed;
+ struct bch_inode_unpacked dir_u, inode_u;
char name_buf[20];
struct qstr name;
int ret;
snprintf(name_buf, sizeof(name_buf), "%llu", inum);
name = (struct qstr) QSTR(name_buf);
- lostfound_inode->bi_nlink++;
-
- bch2_inode_pack(&packed, lostfound_inode);
-
- ret = bch2_btree_insert(c, BTREE_ID_INODES, &packed.inode.k_i,
- NULL, NULL,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW);
- if (ret) {
- bch_err(c, "error %i reattaching inode %llu while updating lost+found",
- ret, inum);
- return ret;
- }
+ ret = bch2_trans_do(c, NULL, NULL,
+ BTREE_INSERT_LAZY_RW,
+ bch2_link_trans(&trans, lostfound_inode->bi_inum,
+ inum, &dir_u, &inode_u, &name));
+ if (ret)
+ bch_err(c, "error %i reattaching inode %llu", ret, inum);
- ret = bch2_dirent_create(c, lostfound_inode->bi_inum,
- &lostfound_hash_info,
- DT_DIR, &name, inum, NULL,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW);
- if (ret) {
- bch_err(c, "error %i reattaching inode %llu while creating new dirent",
- ret, inum);
- return ret;
- }
return ret;
}
struct inode_walker *w, u64 inum)
{
if (inum != w->cur_inum) {
- int ret = bch2_inode_find_by_inum_trans(trans, inum,
- &w->inode);
+ int ret = __bch2_inode_find_by_inum_trans(trans, inum,
+ &w->inode, 0);
if (ret && ret != -ENOENT)
return ret;
static void hash_check_init(struct hash_check *h)
{
h->chain = NULL;
+ h->chain_end = 0;
}
static void hash_stop_chain(struct btree_trans *trans,
struct btree_iter *k_iter, struct bkey_s_c k,
u64 hashed)
{
+ struct bkey_i delete;
struct bkey_i *tmp;
- int ret = 0;
- tmp = kmalloc(bkey_bytes(k.k), GFP_KERNEL);
- if (!tmp)
- return -ENOMEM;
+ tmp = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
+ if (IS_ERR(tmp))
+ return PTR_ERR(tmp);
bkey_reassemble(tmp, k);
- ret = bch2_btree_delete_at(trans, k_iter, 0);
- if (ret)
- goto err;
+ bkey_init(&delete.k);
+ delete.k.p = k_iter->pos;
+ bch2_trans_update(trans, k_iter, &delete, 0);
- bch2_hash_set(trans, desc, &h->info, k_iter->pos.inode,
- tmp, BCH_HASH_SET_MUST_CREATE);
- ret = bch2_trans_commit(trans, NULL, NULL,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW);
-err:
- kfree(tmp);
- return ret;
+ return bch2_hash_set(trans, desc, &h->info, k_iter->pos.inode,
+ tmp, 0);
}
static int fsck_hash_delete_at(struct btree_trans *trans,
retry:
ret = bch2_hash_delete_at(trans, desc, info, iter) ?:
bch2_trans_commit(trans, NULL, NULL,
- BTREE_INSERT_ATOMIC|
BTREE_INSERT_NOFAIL|
BTREE_INSERT_LAZY_RW);
if (ret == -EINTR) {
return 0;
iter = bch2_trans_copy_iter(trans, h->chain);
- BUG_ON(IS_ERR(iter));
- for_each_btree_key_continue(iter, 0, k2) {
+ for_each_btree_key_continue(iter, 0, k2, ret) {
if (bkey_cmp(k2.k->p, k.k->p) >= 0)
break;
struct hash_check *h,
struct btree_iter *k_iter, struct bkey_s_c k)
{
- bool hole = (k.k->type != KEY_TYPE_whiteout &&
+ bool hole = (k.k->type != KEY_TYPE_hash_whiteout &&
k.k->type != desc.key_type);
if (hole || k.k->p.offset > h->chain_end + 1)
hash_stop_chain(trans, h);
if (!hole) {
- if (!h->chain) {
+ if (!h->chain)
h->chain = bch2_trans_copy_iter(trans, k_iter);
- BUG_ON(IS_ERR(h->chain));
- }
h->chain_end = k.k->p.offset;
}
"hashed to %llu chain starts at %llu\n%s",
desc.btree_id, k.k->p.offset,
hashed, h->chain->pos.offset,
- (bch2_bkey_val_to_text(&PBUF(buf), c,
- k), buf))) {
- ret = hash_redo_key(desc, trans, h, k_iter, k, hashed);
+ (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf))) {
+ ret = __bch2_trans_do(trans, NULL, NULL,
+ BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
+ hash_redo_key(desc, trans, h, k_iter, k, hashed));
if (ret) {
bch_err(c, "hash_redo_key err %i", ret);
return ret;
}
- return 1;
+ return -EINTR;
}
ret = hash_check_duplicates(trans, desc, h, k_iter, k);
if (fsck_err(c, "dirent with junk at end, was %s (%zu) now %s (%u)",
buf, strlen(buf), d->v.d_name, len)) {
- bch2_trans_update(trans, BTREE_INSERT_ENTRY(iter, &d->k_i));
-
- ret = bch2_trans_commit(trans, NULL, NULL,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW);
+ ret = __bch2_trans_do(trans, NULL, NULL,
+ BTREE_INSERT_NOFAIL|
+ BTREE_INSERT_LAZY_RW,
+ (bch2_trans_update(trans, iter, &d->k_i, 0), 0));
if (ret)
goto err;
if (fsck_err(c, "cannot fix dirent by removing trailing garbage %s (%zu)\n"
"hash table key at wrong offset: btree %u, offset %llu, "
"hashed to %llu chain starts at %llu\n%s",
- buf, strlen(buf), BTREE_ID_DIRENTS,
+ buf, strlen(buf), BTREE_ID_dirents,
k->k->p.offset, hash, h->chain->pos.offset,
(bch2_bkey_val_to_text(&PBUF(buf), c,
*k), buf))) {
- ret = hash_redo_key(bch2_dirent_hash_desc, trans,
- h, iter, *k, hash);
+ ret = __bch2_trans_do(trans, NULL, NULL,
+ BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
+ hash_redo_key(bch2_dirent_hash_desc, trans,
+ h, iter, *k, hash));
if (ret)
bch_err(c, "hash_redo_key err %i", ret);
else
goto err;
}
-static int bch2_inode_truncate(struct bch_fs *c, u64 inode_nr, u64 new_size)
+static int fix_overlapping_extent(struct btree_trans *trans,
+ struct bkey_s_c k, struct bpos cut_at)
{
- return bch2_btree_delete_range(c, BTREE_ID_EXTENTS,
- POS(inode_nr, round_up(new_size, block_bytes(c)) >> 9),
- POS(inode_nr + 1, 0), NULL);
+ struct btree_iter *iter;
+ struct bkey_i *u;
+ int ret;
+
+ u = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
+ ret = PTR_ERR_OR_ZERO(u);
+ if (ret)
+ return ret;
+
+ bkey_reassemble(u, k);
+ bch2_cut_front(cut_at, u);
+
+
+ /*
+ * We don't want to go through the extent_handle_overwrites path:
+ *
+ * XXX: this is going to screw up disk accounting, extent triggers
+ * assume things about extent overwrites - we should be running the
+ * triggers manually here
+ */
+ iter = bch2_trans_get_iter(trans, BTREE_ID_extents, u->k.p,
+ BTREE_ITER_INTENT|BTREE_ITER_NOT_EXTENTS);
+
+ BUG_ON(iter->flags & BTREE_ITER_IS_EXTENTS);
+ bch2_trans_update(trans, iter, u, BTREE_TRIGGER_NORUN);
+ bch2_trans_iter_put(trans, iter);
+
+ return bch2_trans_commit(trans, NULL, NULL,
+ BTREE_INSERT_NOFAIL|
+ BTREE_INSERT_LAZY_RW);
}
/*
struct btree_trans trans;
struct btree_iter *iter;
struct bkey_s_c k;
- u64 i_sectors;
+ struct bkey_buf prev;
+ u64 i_sectors = 0;
int ret = 0;
+ bch2_bkey_buf_init(&prev);
+ prev.k->k = KEY(0, 0, 0);
bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
bch_verbose(c, "checking extents");
- iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
- POS(BCACHEFS_ROOT_INO, 0), 0);
+ iter = bch2_trans_get_iter(&trans, BTREE_ID_extents,
+ POS(BCACHEFS_ROOT_INO, 0),
+ BTREE_ITER_INTENT);
retry:
- for_each_btree_key_continue(iter, 0, k) {
- ret = walk_inode(&trans, &w, k.k->p.inode);
- if (ret)
- break;
+ while ((k = bch2_btree_iter_peek(iter)).k &&
+ !(ret = bkey_err(k))) {
+ if (w.have_inode &&
+ w.cur_inum != k.k->p.inode &&
+ !(w.inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY) &&
+ fsck_err_on(w.inode.bi_sectors != i_sectors, c,
+ "inode %llu has incorrect i_sectors: got %llu, should be %llu",
+ w.inode.bi_inum,
+ w.inode.bi_sectors, i_sectors)) {
+ struct btree_iter *inode_iter =
+ bch2_trans_get_iter(&trans, BTREE_ID_inodes,
+ POS(0, w.cur_inum),
+ BTREE_ITER_INTENT);
- if (fsck_err_on(!w.have_inode, c,
- "extent type %u for missing inode %llu",
- k.k->type, k.k->p.inode) ||
- fsck_err_on(w.have_inode &&
- !S_ISREG(w.inode.bi_mode) && !S_ISLNK(w.inode.bi_mode), c,
- "extent type %u for non regular file, inode %llu mode %o",
- k.k->type, k.k->p.inode, w.inode.bi_mode)) {
- bch2_trans_unlock(&trans);
+ w.inode.bi_sectors = i_sectors;
- ret = bch2_inode_truncate(c, k.k->p.inode, 0);
+ ret = __bch2_trans_do(&trans, NULL, NULL,
+ BTREE_INSERT_NOFAIL|
+ BTREE_INSERT_LAZY_RW,
+ bch2_inode_write(&trans, inode_iter, &w.inode));
+ bch2_trans_iter_put(&trans, inode_iter);
if (ret)
- goto err;
- continue;
+ break;
}
- if (fsck_err_on(w.first_this_inode &&
- w.have_inode &&
- !(w.inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY) &&
- w.inode.bi_sectors !=
- (i_sectors = bch2_count_inode_sectors(&trans, w.cur_inum)),
- c, "i_sectors wrong: got %llu, should be %llu",
- w.inode.bi_sectors, i_sectors)) {
- struct bkey_inode_buf p;
+ if (bkey_cmp(prev.k->k.p, bkey_start_pos(k.k)) > 0) {
+ char buf1[200];
+ char buf2[200];
- w.inode.bi_sectors = i_sectors;
+ bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(prev.k));
+ bch2_bkey_val_to_text(&PBUF(buf2), c, k);
- bch2_trans_unlock(&trans);
+ if (fsck_err(c, "overlapping extents:\n%s\n%s", buf1, buf2))
+ return fix_overlapping_extent(&trans, k, prev.k->k.p) ?: -EINTR;
+ }
- bch2_inode_pack(&p, &w.inode);
+ ret = walk_inode(&trans, &w, k.k->p.inode);
+ if (ret)
+ break;
- ret = bch2_btree_insert(c, BTREE_ID_INODES,
- &p.inode.k_i, NULL, NULL,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW);
- if (ret) {
- bch_err(c, "error in fsck: error %i updating inode", ret);
- goto err;
- }
+ if (w.first_this_inode)
+ i_sectors = 0;
- /* revalidate iterator: */
- k = bch2_btree_iter_peek(iter);
+ if (fsck_err_on(!w.have_inode, c,
+ "extent type %u for missing inode %llu",
+ k.k->type, k.k->p.inode) ||
+ fsck_err_on(w.have_inode &&
+ !S_ISREG(w.inode.bi_mode) && !S_ISLNK(w.inode.bi_mode), c,
+ "extent type %u for non regular file, inode %llu mode %o",
+ k.k->type, k.k->p.inode, w.inode.bi_mode)) {
+ bch2_fs_lazy_rw(c);
+ return bch2_btree_delete_range_trans(&trans, BTREE_ID_extents,
+ POS(k.k->p.inode, 0),
+ POS(k.k->p.inode, U64_MAX),
+ NULL) ?: -EINTR;
}
if (fsck_err_on(w.have_inode &&
- !(w.inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
- k.k->type != KEY_TYPE_reservation &&
- k.k->p.offset > round_up(w.inode.bi_size, PAGE_SIZE) >> 9, c,
- "extent type %u offset %llu past end of inode %llu, i_size %llu",
- k.k->type, k.k->p.offset, k.k->p.inode, w.inode.bi_size)) {
- bch2_trans_unlock(&trans);
-
- ret = bch2_inode_truncate(c, k.k->p.inode,
- w.inode.bi_size);
- if (ret)
- goto err;
- continue;
+ !(w.inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
+ k.k->type != KEY_TYPE_reservation &&
+ k.k->p.offset > round_up(w.inode.bi_size, block_bytes(c)) >> 9, c,
+ "extent type %u offset %llu past end of inode %llu, i_size %llu",
+ k.k->type, k.k->p.offset, k.k->p.inode, w.inode.bi_size)) {
+ bch2_fs_lazy_rw(c);
+ return bch2_btree_delete_range_trans(&trans, BTREE_ID_extents,
+ POS(k.k->p.inode, round_up(w.inode.bi_size, block_bytes(c))),
+ POS(k.k->p.inode, U64_MAX),
+ NULL) ?: -EINTR;
}
+
+ if (bkey_extent_is_allocation(k.k))
+ i_sectors += k.k->size;
+ bch2_bkey_buf_reassemble(&prev, c, k);
+
+ bch2_btree_iter_advance(iter);
}
-err:
fsck_err:
if (ret == -EINTR)
goto retry;
+ bch2_trans_iter_put(&trans, iter);
+ bch2_bkey_buf_exit(&prev, c);
return bch2_trans_exit(&trans) ?: ret;
}
hash_check_init(&h);
- iter = bch2_trans_get_iter(&trans, BTREE_ID_DIRENTS,
+ iter = bch2_trans_get_iter(&trans, BTREE_ID_dirents,
POS(BCACHEFS_ROOT_INO, 0), 0);
retry:
- for_each_btree_key_continue(iter, 0, k) {
+ while ((k = bch2_btree_iter_peek(iter)).k &&
+ !(ret = bkey_err(k))) {
struct bkey_s_c_dirent d;
struct bch_inode_unpacked target;
bool have_target;
continue;
}
- ret = bch2_inode_find_by_inum_trans(&trans, d_inum, &target);
+ ret = __bch2_inode_find_by_inum_trans(&trans, d_inum, &target, 0);
if (ret && ret != -ENOENT)
break;
continue;
}
+ if (!target.bi_nlink &&
+ !(target.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) &&
+ (target.bi_dir != k.k->p.inode ||
+ target.bi_dir_offset != k.k->p.offset) &&
+ (fsck_err_on(c->sb.version >= bcachefs_metadata_version_inode_backpointers, c,
+ "inode %llu has wrong backpointer:\n"
+ "got %llu:%llu\n"
+ "should be %llu:%llu",
+ d_inum,
+ target.bi_dir,
+ target.bi_dir_offset,
+ k.k->p.inode,
+ k.k->p.offset) ||
+ c->opts.version_upgrade)) {
+ struct bkey_inode_buf p;
+
+ target.bi_dir = k.k->p.inode;
+ target.bi_dir_offset = k.k->p.offset;
+ bch2_trans_unlock(&trans);
+
+ bch2_inode_pack(c, &p, &target);
+
+ ret = bch2_btree_insert(c, BTREE_ID_inodes,
+ &p.inode.k_i, NULL, NULL,
+ BTREE_INSERT_NOFAIL|
+ BTREE_INSERT_LAZY_RW);
+ if (ret) {
+ bch_err(c, "error in fsck: error %i updating inode", ret);
+ goto err;
+ }
+ continue;
+ }
+
if (fsck_err_on(have_target &&
d.v->d_type !=
mode_to_type(target.bi_mode), c,
bkey_reassemble(&n->k_i, d.s_c);
n->v.d_type = mode_to_type(target.bi_mode);
- bch2_trans_update(&trans,
- BTREE_INSERT_ENTRY(iter, &n->k_i));
-
- ret = bch2_trans_commit(&trans, NULL, NULL,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW);
+ ret = __bch2_trans_do(&trans, NULL, NULL,
+ BTREE_INSERT_NOFAIL|
+ BTREE_INSERT_LAZY_RW,
+ (bch2_trans_update(&trans, iter, &n->k_i, 0), 0));
kfree(n);
if (ret)
goto err;
}
+
+ bch2_btree_iter_advance(iter);
}
hash_stop_chain(&trans, &h);
if (ret == -EINTR)
goto retry;
+ bch2_trans_iter_put(&trans, h.chain);
+ bch2_trans_iter_put(&trans, iter);
return bch2_trans_exit(&trans) ?: ret;
}
bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
- iter = bch2_trans_get_iter(&trans, BTREE_ID_XATTRS,
+ iter = bch2_trans_get_iter(&trans, BTREE_ID_xattrs,
POS(BCACHEFS_ROOT_INO, 0), 0);
retry:
- for_each_btree_key_continue(iter, 0, k) {
+ while ((k = bch2_btree_iter_peek(iter)).k &&
+ !(ret = bkey_err(k))) {
ret = walk_inode(&trans, &w, k.k->p.inode);
if (ret)
break;
k.k->p.inode)) {
ret = bch2_btree_delete_at(&trans, iter, 0);
if (ret)
- goto err;
+ break;
continue;
}
ret = hash_check_key(&trans, bch2_xattr_hash_desc,
&h, iter, k);
if (ret)
- goto fsck_err;
+ break;
+
+ bch2_btree_iter_advance(iter);
}
-err:
fsck_err:
if (ret == -EINTR)
goto retry;
+
+ bch2_trans_iter_put(&trans, h.chain);
+ bch2_trans_iter_put(&trans, iter);
return bch2_trans_exit(&trans) ?: ret;
}
bch_verbose(c, "checking root directory");
- ret = bch2_inode_find_by_inum(c, BCACHEFS_ROOT_INO, root_inode);
+ ret = bch2_trans_do(c, NULL, NULL, 0,
+ __bch2_inode_find_by_inum_trans(&trans, BCACHEFS_ROOT_INO,
+ root_inode, 0));
if (ret && ret != -ENOENT)
return ret;
fsck_err:
return ret;
create_root:
- bch2_inode_init(c, root_inode, 0, 0, S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO,
+ bch2_inode_init(c, root_inode, 0, 0, S_IFDIR|0755,
0, NULL);
root_inode->bi_inum = BCACHEFS_ROOT_INO;
- bch2_inode_pack(&packed, root_inode);
+ bch2_inode_pack(c, &packed, root_inode);
- return bch2_btree_insert(c, BTREE_ID_INODES, &packed.inode.k_i,
+ return bch2_btree_insert(c, BTREE_ID_inodes, &packed.inode.k_i,
NULL, NULL,
BTREE_INSERT_NOFAIL|
BTREE_INSERT_LAZY_RW);
struct qstr lostfound = QSTR("lost+found");
struct bch_hash_info root_hash_info =
bch2_hash_info_init(c, root_inode);
- struct bkey_inode_buf packed;
u64 inum;
int ret;
goto create_lostfound;
}
- ret = bch2_inode_find_by_inum(c, inum, lostfound_inode);
+ ret = bch2_trans_do(c, NULL, NULL, 0,
+ __bch2_inode_find_by_inum_trans(&trans, inum, lostfound_inode, 0));
if (ret && ret != -ENOENT)
return ret;
fsck_err:
return ret;
create_lostfound:
- root_inode->bi_nlink++;
-
- bch2_inode_pack(&packed, root_inode);
-
- ret = bch2_btree_insert(c, BTREE_ID_INODES, &packed.inode.k_i,
- NULL, NULL,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW);
- if (ret)
- return ret;
-
- bch2_inode_init(c, lostfound_inode, 0, 0, S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO,
- 0, root_inode);
-
- ret = bch2_inode_create(c, lostfound_inode, BLOCKDEV_INODE_MAX, 0,
- &c->unused_inode_hint);
- if (ret)
- return ret;
-
- ret = bch2_dirent_create(c, BCACHEFS_ROOT_INO, &root_hash_info, DT_DIR,
- &lostfound, lostfound_inode->bi_inum, NULL,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW);
+ bch2_inode_init_early(c, lostfound_inode);
+
+ ret = bch2_trans_do(c, NULL, NULL,
+ BTREE_INSERT_NOFAIL|
+ BTREE_INSERT_LAZY_RW,
+ bch2_create_trans(&trans,
+ BCACHEFS_ROOT_INO, root_inode,
+ lostfound_inode, &lostfound,
+ 0, 0, S_IFDIR|0700, 0, NULL, NULL));
if (ret)
- return ret;
+ bch_err(c, "error creating lost+found: %i", ret);
- return 0;
+ return ret;
}
-struct inode_bitmap {
- unsigned long *bits;
- size_t size;
-};
+typedef GENRADIX(unsigned long) inode_bitmap;
-static inline bool inode_bitmap_test(struct inode_bitmap *b, size_t nr)
+static inline bool inode_bitmap_test(inode_bitmap *b, size_t nr)
{
- return nr < b->size ? test_bit(nr, b->bits) : false;
+ unsigned long *w = genradix_ptr(b, nr / BITS_PER_LONG);
+ return w ? test_bit(nr & (BITS_PER_LONG - 1), w) : false;
}
-static inline int inode_bitmap_set(struct inode_bitmap *b, size_t nr)
+static inline int inode_bitmap_set(inode_bitmap *b, size_t nr)
{
- if (nr >= b->size) {
- size_t new_size = max_t(size_t, max_t(size_t,
- PAGE_SIZE * 8,
- b->size * 2),
- nr + 1);
- void *n;
-
- new_size = roundup_pow_of_two(new_size);
- n = krealloc(b->bits, new_size / 8, GFP_KERNEL|__GFP_ZERO);
- if (!n) {
- return -ENOMEM;
- }
+ unsigned long *w = genradix_ptr_alloc(b, nr / BITS_PER_LONG, GFP_KERNEL);
- b->bits = n;
- b->size = new_size;
- }
+ if (!w)
+ return -ENOMEM;
- __set_bit(nr, b->bits);
+ *w |= 1UL << (nr & (BITS_PER_LONG - 1));
return 0;
}
static int check_directory_structure(struct bch_fs *c,
struct bch_inode_unpacked *lostfound_inode)
{
- struct inode_bitmap dirs_done = { NULL, 0 };
+ inode_bitmap dirs_done;
struct pathbuf path = { 0, 0, NULL };
struct pathbuf_entry *e;
struct btree_trans trans;
/* DFS: */
restart_dfs:
+ genradix_init(&dirs_done);
had_unreachable = false;
ret = inode_bitmap_set(&dirs_done, BCACHEFS_ROOT_INO);
if (e->offset == U64_MAX)
goto up;
- for_each_btree_key(&trans, iter, BTREE_ID_DIRENTS,
+ for_each_btree_key(&trans, iter, BTREE_ID_dirents,
POS(e->inum, e->offset + 1), 0, k, ret) {
if (k.k->p.inode != e->inum)
break;
path.nr--;
}
- iter = bch2_trans_get_iter(&trans, BTREE_ID_INODES, POS_MIN, 0);
+ iter = bch2_trans_get_iter(&trans, BTREE_ID_inodes, POS_MIN, 0);
retry:
- for_each_btree_key_continue(iter, 0, k) {
+ for_each_btree_key_continue(iter, 0, k, ret) {
if (k.k->type != KEY_TYPE_inode)
continue;
if (!ret)
continue;
- if (fsck_err_on(!inode_bitmap_test(&dirs_done, k.k->p.inode), c,
+ if (fsck_err_on(!inode_bitmap_test(&dirs_done, k.k->p.offset), c,
"unreachable directory found (inum %llu)",
- k.k->p.inode)) {
+ k.k->p.offset)) {
bch2_trans_unlock(&trans);
- ret = reattach_inode(c, lostfound_inode, k.k->p.inode);
+ ret = reattach_inode(c, lostfound_inode, k.k->p.offset);
if (ret) {
goto err;
}
had_unreachable = true;
}
}
- ret = bch2_trans_iter_free(&trans, iter);
+ bch2_trans_iter_free(&trans, iter);
if (ret)
goto err;
if (had_unreachable) {
bch_info(c, "reattached unreachable directories, restarting pass to check for loops");
- kfree(dirs_done.bits);
+ genradix_free(&dirs_done);
kfree(path.entries);
memset(&dirs_done, 0, sizeof(dirs_done));
memset(&path, 0, sizeof(path));
err:
fsck_err:
ret = bch2_trans_exit(&trans) ?: ret;
- kfree(dirs_done.bits);
+ genradix_free(&dirs_done);
kfree(path.entries);
return ret;
}
if (inum < range_start || inum >= *range_end)
return;
+ if (inum - range_start >= SIZE_MAX / sizeof(struct nlink)) {
+ *range_end = inum;
+ return;
+ }
+
link = genradix_ptr_alloc(links, inum - range_start, GFP_KERNEL);
if (!link) {
bch_verbose(c, "allocation failed during fsck - will need another pass");
inc_link(c, links, range_start, range_end, BCACHEFS_ROOT_INO, false);
- for_each_btree_key(&trans, iter, BTREE_ID_DIRENTS, POS_MIN, 0, k, ret) {
+ for_each_btree_key(&trans, iter, BTREE_ID_dirents, POS_MIN, 0, k, ret) {
switch (k.k->type) {
case KEY_TYPE_dirent:
d = bkey_s_c_to_dirent(k);
bch2_trans_cond_resched(&trans);
}
+ bch2_trans_iter_put(&trans, iter);
+
ret = bch2_trans_exit(&trans) ?: ret;
if (ret)
bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
struct nlink *link,
bool *do_update)
{
- u32 i_nlink = u->bi_flags & BCH_INODE_UNLINKED
- ? 0
- : u->bi_nlink + nlink_bias(u->bi_mode);
+ u32 i_nlink = bch2_inode_nlink_get(u);
u32 real_i_nlink =
link->count * nlink_bias(u->bi_mode) +
link->dir_count;
}
if (!S_ISDIR(u->bi_mode) && link->dir_count) {
- need_fsck_err(c, "non directory with subdirectories",
+ need_fsck_err(c, "non directory with subdirectories (inum %llu)",
u->bi_inum);
return 0;
}
if (!link->count &&
!(u->bi_flags & BCH_INODE_UNLINKED) &&
- (c->sb.features & (1 << BCH_FEATURE_ATOMIC_NLINK))) {
+ (c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
if (fsck_err(c, "unreachable inode %llu not marked as unlinked (type %u)",
u->bi_inum, mode_to_type(u->bi_mode)) ==
FSCK_ERR_IGNORE)
}
if (i_nlink != real_i_nlink &&
- (c->sb.features & (1 << BCH_FEATURE_ATOMIC_NLINK))) {
+ (c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
if (fsck_err(c, "inode %llu has wrong i_nlink "
"(type %u i_nlink %u, should be %u)",
u->bi_inum, mode_to_type(u->bi_mode),
u->bi_inum, i_nlink, real_i_nlink);
set_i_nlink:
if (i_nlink != real_i_nlink) {
- if (real_i_nlink) {
- u->bi_nlink = real_i_nlink - nlink_bias(u->bi_mode);
- u->bi_flags &= ~BCH_INODE_UNLINKED;
- } else {
- u->bi_nlink = 0;
- u->bi_flags |= BCH_INODE_UNLINKED;
- }
-
+ bch2_inode_nlink_set(u, real_i_nlink);
*do_update = true;
}
fsck_err:
u.bi_inum))) {
bch_verbose(c, "deleting inode %llu", u.bi_inum);
- ret = bch2_inode_rm(c, u.bi_inum);
+ bch2_fs_lazy_rw(c);
+
+ ret = bch2_inode_rm(c, u.bi_inum, false);
if (ret)
bch_err(c, "error in fsck: error %i while deleting inode", ret);
return ret;
u.bi_inum))) {
bch_verbose(c, "truncating inode %llu", u.bi_inum);
+ bch2_fs_lazy_rw(c);
+
/*
* XXX: need to truncate partial blocks too here - or ideally
* just switch units to bytes and that issue goes away
*/
-
- ret = bch2_inode_truncate(c, u.bi_inum, u.bi_size);
+ ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
+ POS(u.bi_inum, round_up(u.bi_size, block_bytes(c))),
+ POS(u.bi_inum, U64_MAX),
+ NULL);
if (ret) {
bch_err(c, "error in fsck: error %i truncating inode", ret);
return ret;
do_update = true;
}
+ if (!S_ISDIR(u.bi_mode) &&
+ u.bi_nlink &&
+ !(u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) &&
+ (fsck_err_on(c->sb.version >= bcachefs_metadata_version_inode_backpointers, c,
+ "inode missing BCH_INODE_BACKPTR_UNTRUSTED flags") ||
+ c->opts.version_upgrade)) {
+ u.bi_flags |= BCH_INODE_BACKPTR_UNTRUSTED;
+ do_update = true;
+ }
+
if (do_update) {
struct bkey_inode_buf p;
- bch2_inode_pack(&p, &u);
- bch2_trans_update(trans, BTREE_INSERT_ENTRY(iter, &p.inode.k_i));
+ bch2_inode_pack(c, &p, &u);
- ret = bch2_trans_commit(trans, NULL, NULL,
- BTREE_INSERT_NOFAIL|
- BTREE_INSERT_LAZY_RW);
- if (ret && ret != -EINTR)
+ ret = __bch2_trans_do(trans, NULL, NULL,
+ BTREE_INSERT_NOFAIL|
+ BTREE_INSERT_LAZY_RW,
+ (bch2_trans_update(trans, iter, &p.inode.k_i, 0), 0));
+ if (ret)
bch_err(c, "error in fsck: error %i "
"updating inode", ret);
}
bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
- iter = bch2_trans_get_iter(&trans, BTREE_ID_INODES,
- POS(range_start, 0), 0);
+ iter = bch2_trans_get_iter(&trans, BTREE_ID_inodes,
+ POS(0, range_start), 0);
nlinks_iter = genradix_iter_init(links, 0);
while ((k = bch2_btree_iter_peek(iter)).k &&
- !(ret2 = bkey_err(k))) {
+ !(ret2 = bkey_err(k)) &&
+ iter->pos.offset < range_end) {
peek_nlinks: link = genradix_iter_peek(&nlinks_iter, links);
- if (!link && (!k.k || iter->pos.inode >= range_end))
+ if (!link && (!k.k || iter->pos.offset >= range_end))
break;
nlinks_pos = range_start + nlinks_iter.pos;
- if (iter->pos.inode > nlinks_pos) {
+
+ if (link && nlinks_pos < iter->pos.offset) {
/* Should have been caught by dirents pass: */
- need_fsck_err_on(link && link->count, c,
+ need_fsck_err_on(link->count, c,
"missing inode %llu (nlink %u)",
nlinks_pos, link->count);
genradix_iter_advance(&nlinks_iter, links);
goto peek_nlinks;
}
- if (iter->pos.inode < nlinks_pos || !link)
+ if (!link || nlinks_pos > iter->pos.offset)
link = &zero_links;
if (k.k && k.k->type == KEY_TYPE_inode) {
nlinks_pos, link->count);
}
- if (nlinks_pos == iter->pos.inode)
+ if (nlinks_pos == iter->pos.offset)
genradix_iter_advance(&nlinks_iter, links);
- bch2_btree_iter_next(iter);
+ bch2_btree_iter_advance(iter);
bch2_trans_cond_resched(&trans);
}
fsck_err:
+ bch2_trans_iter_put(&trans, iter);
bch2_trans_exit(&trans);
if (ret2)
bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
- for_each_btree_key(&trans, iter, BTREE_ID_INODES, POS_MIN, 0, k, ret) {
+ for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN, 0, k, ret) {
if (k.k->type != KEY_TYPE_inode)
continue;
BCH_INODE_I_SECTORS_DIRTY|
BCH_INODE_UNLINKED)) {
ret = check_inode(&trans, NULL, iter, inode, NULL);
- BUG_ON(ret == -EINTR);
if (ret)
break;
}
}
+ bch2_trans_iter_put(&trans, iter);
+
BUG_ON(ret == -EINTR);
return bch2_trans_exit(&trans) ?: ret;
projects / bcachefs-tools-debian / blobdiff