-d9d1235f3c568a47b3547c0b0adad0d7948f18aa
+4dd9a5a488857137ce6eecadddd9304440fb03e9
struct bch_reflink_p {
struct bch_val v;
__le64 idx;
-
- __le32 reservation_generation;
- __u8 nr_replicas;
- __u8 pad[3];
-};
+ /*
+ * A reflink pointer might point to an indirect extent which is then
+ * later split (by copygc or rebalance). If we only pointed to part of
+ * the original indirect extent, and then one of the fragments is
+ * outside the range we point to, we'd leak a refcount: so when creating
+ * reflink pointers, we need to store pad values to remember the full
+ * range we were taking a reference on.
+ */
+ __le32 front_pad;
+ __le32 back_pad;
+} __attribute__((packed, aligned(8)));
struct bch_reflink_v {
struct bch_val v;
__le64 refcount;
union bch_extent_entry start[0];
__u64 _data[0];
-};
+} __attribute__((packed, aligned(8)));
struct bch_indirect_inline_data {
struct bch_val v;
bcachefs_metadata_version_inode_backpointers = 13,
bcachefs_metadata_version_btree_ptr_sectors_written = 14,
bcachefs_metadata_version_snapshot_2 = 15,
- bcachefs_metadata_version_max = 16,
+ bcachefs_metadata_version_reflink_p_fix = 16,
+ bcachefs_metadata_version_max = 17,
};
#define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1)
struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
struct reflink_gc *ref;
size_t l, r, m;
- u64 idx = le64_to_cpu(p.v->idx);
- unsigned sectors = p.k->size;
+ u64 idx = le64_to_cpu(p.v->idx) - le32_to_cpu(p.v->front_pad);
+ u64 sectors = (u64) le32_to_cpu(p.v->front_pad) +
+ le32_to_cpu(p.v->back_pad) +
+ p.k->size;
s64 ret = 0;
BUG_ON((flags & (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE)) ==
bch2_fs_inconsistent(c,
"%llu:%llu len %u points to nonexistent indirect extent %llu",
p.k->p.inode, p.k->p.offset, p.k->size, idx);
- bch2_inconsistent_error(c);
ret = -EIO;
goto err;
}
- BUG_ON(!*refcount && (flags & BTREE_TRIGGER_OVERWRITE));
+ if (!*refcount && (flags & BTREE_TRIGGER_OVERWRITE)) {
+ bch2_fs_inconsistent(c,
+ "%llu:%llu len %u idx %llu indirect extent refcount underflow",
+ p.k->p.inode, p.k->p.offset, p.k->size, idx);
+ ret = -EIO;
+ goto err;
+ }
+
+ if (flags & BTREE_TRIGGER_INSERT) {
+ struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
+ u64 pad;
+
+ pad = max_t(s64, le32_to_cpu(v->front_pad),
+ le64_to_cpu(v->idx) - bkey_start_offset(k.k));
+ BUG_ON(pad > U32_MAX);
+ v->front_pad = cpu_to_le32(pad);
+
+ pad = max_t(s64, le32_to_cpu(v->back_pad),
+ k.k->p.offset - p.k->size - le64_to_cpu(v->idx));
+ BUG_ON(pad > U32_MAX);
+ v->back_pad = cpu_to_le32(pad);
+ }
+
le64_add_cpu(refcount, add);
if (!*refcount) {
struct bkey_s_c k, unsigned flags)
{
struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
- u64 idx = le64_to_cpu(p.v->idx);
- unsigned sectors = p.k->size;
+ u64 idx, sectors;
s64 ret = 0;
+ if (flags & BTREE_TRIGGER_INSERT) {
+ struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
+
+ v->front_pad = v->back_pad = 0;
+ }
+
+ idx = le64_to_cpu(p.v->idx) - le32_to_cpu(p.v->front_pad);
+ sectors = (u64) le32_to_cpu(p.v->front_pad) +
+ le32_to_cpu(p.v->back_pad) +
+ p.k->size;
+
while (sectors) {
ret = __bch2_trans_mark_reflink_p(trans, p, idx, flags);
if (ret < 0)
return ret;
}
+static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter)
+{
+ struct bkey_s_c k;
+ struct bkey_s_c_reflink_p p;
+ struct bkey_i_reflink_p *u;
+ int ret;
+
+ k = bch2_btree_iter_peek(iter);
+ if (!k.k)
+ return 0;
+
+ ret = bkey_err(k);
+ if (ret)
+ return ret;
+
+ if (k.k->type != KEY_TYPE_reflink_p)
+ return 0;
+
+ p = bkey_s_c_to_reflink_p(k);
+
+ if (!p.v->front_pad && !p.v->back_pad)
+ return 0;
+
+ u = bch2_trans_kmalloc(trans, sizeof(*u));
+ ret = PTR_ERR_OR_ZERO(u);
+ if (ret)
+ return ret;
+
+ bkey_reassemble(&u->k_i, k);
+ u->v.front_pad = 0;
+ u->v.back_pad = 0;
+
+ return bch2_trans_update(trans, iter, &u->k_i, 0);
+}
+
+static int fix_reflink_p(struct bch_fs *c)
+{
+ struct btree_trans trans;
+ struct btree_iter iter;
+ struct bkey_s_c k;
+ int ret;
+
+ if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
+ return 0;
+
+ bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
+
+ for_each_btree_key(&trans, iter, BTREE_ID_extents, POS_MIN,
+ BTREE_ITER_INTENT|
+ BTREE_ITER_PREFETCH|
+ BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
+ if (k.k->type == KEY_TYPE_reflink_p) {
+ ret = __bch2_trans_do(&trans, NULL, NULL,
+ BTREE_INSERT_NOFAIL|
+ BTREE_INSERT_LAZY_RW,
+ fix_reflink_p_key(&trans, &iter));
+ if (ret)
+ break;
+ }
+ }
+ bch2_trans_iter_exit(&trans, &iter);
+
+ bch2_trans_exit(&trans);
+ return ret;
+}
+
/*
* Checks for inconsistencies that shouldn't happen, unless we have a bug.
* Doesn't fix them yet, mainly because they haven't yet been observed:
check_xattrs(c) ?:
check_root(c) ?:
check_directory_structure(c) ?:
- check_nlinks(c);
+ check_nlinks(c) ?:
+ fix_reflink_p(c);
}
int bch2_fsck_walk_inodes_only(struct bch_fs *c)
c->opts.version_upgrade = true;
c->opts.fsck = true;
c->opts.fix_errors = FSCK_OPT_YES;
- } else if (c->sb.version < bcachefs_metadata_version_btree_ptr_sectors_written) {
- bch_info(c, "version prior to btree_ptr_sectors_written, upgrade required");
- c->opts.version_upgrade = true;
- } else if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
- bch_info(c, "filesystem version is prior to snapshots - upgrading");
+ } else if (c->sb.version < bcachefs_metadata_version_reflink_p_fix) {
+ bch_info(c, "filesystem version is prior to reflink_p fix - upgrading");
c->opts.version_upgrade = true;
+ c->opts.fsck = true;
}
ret = bch2_blacklist_table_initialize(c);
if (bkey_val_bytes(p.k) != sizeof(*p.v))
return "incorrect value size";
+ if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix &&
+ le64_to_cpu(p.v->idx) < le32_to_cpu(p.v->front_pad))
+ return "idx < front_pad";
+
return NULL;
}
projects / bcachefs-tools-debian / commitdiff