HALF ECHO LEARN / REFERENCE COLLECTION

v262.2

src/shared/btrfs-util.c

78,989 bytesGit mode 100644Download original fileLGPL-2.1-or-later

Git blob: f03eab91fd4a4eab4a3201a372335ae0090364e8

Lines 1001–2000 of 2,215

                                /* Subvolume is somewhere further down, hence we need to open the
                                 * containing directory first */
 
                                child_fd = openat(subvol_fd, lookup_args.path, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW);
                                if (child_fd < 0)
                                        return -errno;
 
                                r = subvol_remove_children(child_fd, lookup_args.name, child_subvol_id, flags);
                        }
                        if (r < 0)
                                return r;
                }
 
                if (ioctl_ret >= 0)
                        break;
        }
 
        /* OK, the child subvolumes should all be gone now, let's try
         * again to remove the subvolume */
        if (ioctl(fd, BTRFS_IOC_SNAP_DESTROY, &vol_args) < 0)
                return -errno;
 
finish:
        /* for the leaf subvolumes, the qgroup id is identical to the subvol id */
        (void) btrfs_qgroup_destroy_recursive(fd, subvol_id);
        return 0;
}
 
int btrfs_subvol_remove_at(int dir_fd, const char *path, BtrfsRemoveFlags flags) {
        _cleanup_free_ char *subvolume = NULL;
        _cleanup_close_ int fd = -EBADF;
        int r;
 
        assert(path);
 
        fd = chase_and_openat(XAT_FDROOT, dir_fd, path, CHASE_PARENT|CHASE_EXTRACT_FILENAME, O_CLOEXEC, &subvolume);
        if (fd < 0)
                return fd;
 
        r = btrfs_validate_subvolume_name(subvolume);
        if (r < 0)
                return r;
 
        return subvol_remove_children(fd, subvolume, 0, flags);
}
 
int btrfs_qgroup_copy_limits(int fd, uint64_t old_qgroupid, uint64_t new_qgroupid) {
 
        struct btrfs_ioctl_search_args args = {
                /* Tree of quota items */
                .key.tree_id = BTRFS_QUOTA_TREE_OBJECTID,
 
                /* The object ID is always 0 */
                .key.min_objectid = 0,
                .key.max_objectid = 0,
 
                /* Look precisely for the quota items */
                .key.min_type = BTRFS_QGROUP_LIMIT_KEY,
                .key.max_type = BTRFS_QGROUP_LIMIT_KEY,
 
                /* For our qgroup */
                .key.min_offset = old_qgroupid,
                .key.max_offset = old_qgroupid,
 
                /* No restrictions on the other components */
                .key.min_transid = 0,
                .key.max_transid = UINT64_MAX,
        };
 
        int r;
 
        r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
        if (r < 0)
                return r;
        if (r == 0)
                return -ENOTTY;
 
        while (btrfs_ioctl_search_args_compare(&args) <= 0) {
                struct btrfs_ioctl_search_header sh;
                const void *body;
 
                args.key.nr_items = 256;
                if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args) < 0) {
                        if (errno == ENOENT) /* quota tree missing: quota is not enabled, hence nothing to copy */
                                break;
 
                        return -errno;
                }
 
                if (args.key.nr_items <= 0)
                        break;
 
                FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, args) {
                        struct btrfs_ioctl_qgroup_limit_args qargs;
                        unsigned c;
 
                        /* Make sure we start the next search at least from this entry */
                        btrfs_ioctl_search_args_set(&args, &sh);
 
                        if (sh.objectid != 0)
                                continue;
                        if (sh.type != BTRFS_QGROUP_LIMIT_KEY)
                                continue;
                        if (sh.offset != old_qgroupid)
                                continue;
 
                        /* We found the entry, now copy things over. */
 
                        const struct btrfs_qgroup_limit_item *qli = body;
                        qargs = (struct btrfs_ioctl_qgroup_limit_args) {
                                .qgroupid = new_qgroupid,
 
                                .lim.max_rfer = le64toh(qli->max_rfer),
                                .lim.max_excl = le64toh(qli->max_excl),
                                .lim.rsv_rfer = le64toh(qli->rsv_rfer),
                                .lim.rsv_excl = le64toh(qli->rsv_excl),
 
                                .lim.flags = le64toh(qli->flags) & (BTRFS_QGROUP_LIMIT_MAX_RFER|
                                                                    BTRFS_QGROUP_LIMIT_MAX_EXCL|
                                                                    BTRFS_QGROUP_LIMIT_RSV_RFER|
                                                                    BTRFS_QGROUP_LIMIT_RSV_EXCL),
                        };
 
                        for (c = 0;; c++) {
                                if (ioctl(fd, BTRFS_IOC_QGROUP_LIMIT, &qargs) < 0) {
                                        if (errno == EBUSY && c < 10) {
                                                (void) btrfs_quota_scan_wait(fd);
                                                continue;
                                        }
                                        return -errno;
                                }
 
                                break;
                        }
 
                        return 1;
                }
 
                /* Increase search key by one, to read the next item, if we can. */
                if (!btrfs_ioctl_search_args_inc(&args))
                        break;
        }
 
        return 0;
}
 
static int copy_quota_hierarchy(int fd, uint64_t old_subvol_id, uint64_t new_subvol_id) {
        _cleanup_free_ uint64_t *old_qgroups = NULL, *old_parent_qgroups = NULL;
        bool copy_from_parent = false, insert_intermediary_qgroup = false;
        int n_old_qgroups, n_old_parent_qgroups, r;
        uint64_t old_parent_id;
 
        assert(fd >= 0);
 
        /* Copies a reduced form of quota information from the old to
         * the new subvolume. */
 
        n_old_qgroups = btrfs_qgroup_find_parents(fd, old_subvol_id, &old_qgroups);
        if (n_old_qgroups <= 0) /* Nothing to copy */
                return n_old_qgroups;
 
        assert(old_qgroups); /* Coverity gets confused by the macro iterator allocating this, add a hint */
 
        r = btrfs_subvol_get_parent(fd, old_subvol_id, &old_parent_id);
        if (r == -ENXIO)
                /* We have no parent, hence nothing to copy. */
                n_old_parent_qgroups = 0;
        else if (r < 0)
                return r;
        else {
                n_old_parent_qgroups = btrfs_qgroup_find_parents(fd, old_parent_id, &old_parent_qgroups);
                if (n_old_parent_qgroups < 0)
                        return n_old_parent_qgroups;
        }
 
        for (int i = 0; i < n_old_qgroups; i++) {
                uint64_t id;
 
                r = btrfs_qgroupid_split(old_qgroups[i], NULL, &id);
                if (r < 0)
                        return r;
 
                if (id == old_subvol_id) {
                        /* The old subvolume was member of a qgroup
                         * that had the same id, but a different level
                         * as it self. Let's set up something similar
                         * in the destination. */
                        insert_intermediary_qgroup = true;
                        break;
                }
 
                for (int j = 0; j < n_old_parent_qgroups; j++)
                        if (old_parent_qgroups[j] == old_qgroups[i])
                                /* The old subvolume shared a common
                                 * parent qgroup with its parent
                                 * subvolume. Let's set up something
                                 * similar in the destination. */
                                copy_from_parent = true;
        }
 
        if (!insert_intermediary_qgroup && !copy_from_parent)
                return 0;
 
        return btrfs_subvol_auto_qgroup_fd(fd, new_subvol_id, insert_intermediary_qgroup);
}
 
static int copy_subtree_quota_limits(int fd, uint64_t old_subvol, uint64_t new_subvol) {
        uint64_t old_subtree_qgroup, new_subtree_qgroup;
        bool changed;
        int r;
 
        /* First copy the leaf limits */
        r = btrfs_qgroup_copy_limits(fd, old_subvol, new_subvol);
        if (r < 0)
                return r;
        changed = r > 0;
 
        /* Then, try to copy the subtree limits, if there are any. */
        r = btrfs_subvol_find_subtree_qgroup(fd, old_subvol, &old_subtree_qgroup);
        if (r < 0)
                return r;
        if (r == 0)
                return changed;
 
        r = btrfs_subvol_find_subtree_qgroup(fd, new_subvol, &new_subtree_qgroup);
        if (r < 0)
                return r;
        if (r == 0)
                return changed;
 
        r = btrfs_qgroup_copy_limits(fd, old_subtree_qgroup, new_subtree_qgroup);
        if (r != 0)
                return r;
 
        return changed;
}
 
static int subvol_snapshot_children(
                int old_fd,
                int new_fd,
                const char *subvolume,
                uint64_t old_subvol_id,
                BtrfsSnapshotFlags flags) {
 
        struct btrfs_ioctl_vol_args_v2 vol_args = {
                .flags = flags & BTRFS_SNAPSHOT_READ_ONLY ? BTRFS_SUBVOL_RDONLY : 0,
                .fd = old_fd,
        };
        _cleanup_close_ int subvolume_fd = -EBADF;
        uint64_t new_subvol_id;
        int r;
 
        assert(old_fd >= 0);
        assert(new_fd >= 0);
        assert(subvolume);
 
        strncpy(vol_args.name, subvolume, sizeof(vol_args.name)-1);
 
        if (ioctl(new_fd, BTRFS_IOC_SNAP_CREATE_V2, &vol_args) < 0)
                return -errno;
 
        if (FLAGS_SET(flags, BTRFS_SNAPSHOT_LOCK_BSD)) {
                subvolume_fd = xopenat_lock(new_fd, subvolume,
                                            O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW,
                                            LOCK_BSD,
                                            LOCK_EX);
                if (subvolume_fd < 0)
                        return subvolume_fd;
 
                r = btrfs_is_subvol_fd(subvolume_fd);
                if (r < 0)
                        return r;
                if (r == 0)
                        return -EEXIST;
        }
 
        if (!(flags & BTRFS_SNAPSHOT_RECURSIVE) &&
            !(flags & BTRFS_SNAPSHOT_QUOTA))
                return flags & BTRFS_SNAPSHOT_LOCK_BSD ? TAKE_FD(subvolume_fd) : 0;
 
        if (old_subvol_id == 0) {
                r = btrfs_subvol_get_id_fd(old_fd, &old_subvol_id);
                if (r < 0)
                        return r;
        }
 
        r = btrfs_subvol_get_id(new_fd, vol_args.name, &new_subvol_id);
        if (r < 0)
                return r;
 
        if (flags & BTRFS_SNAPSHOT_QUOTA)
                (void) copy_quota_hierarchy(new_fd, old_subvol_id, new_subvol_id);
 
        if (!(flags & BTRFS_SNAPSHOT_RECURSIVE)) {
 
                if (flags & BTRFS_SNAPSHOT_QUOTA)
                        (void) copy_subtree_quota_limits(new_fd, old_subvol_id, new_subvol_id);
 
                return flags & BTRFS_SNAPSHOT_LOCK_BSD ? TAKE_FD(subvolume_fd) : 0;
        }
 
        /* Enumerate child subvolumes via BTRFS_IOC_GET_SUBVOL_ROOTREF + BTRFS_IOC_INO_LOOKUP_USER
         * (kernel 4.18+), neither of which requires CAP_SYS_ADMIN, unlike BTRFS_IOC_TREE_SEARCH. */
 
        struct btrfs_ioctl_get_subvol_rootref_args rootref_args = {};
 
        for (;;) {
                /* BTRFS_IOC_GET_SUBVOL_ROOTREF returns up to BTRFS_MAX_ROOTREF_BUFFER_NUM entries per
                 * call. If more are available, it returns -EOVERFLOW with num_items filled in and
                 * min_treeid advanced so we can resume on the next iteration. */
                int ioctl_ret = ioctl(old_fd, BTRFS_IOC_GET_SUBVOL_ROOTREF, &rootref_args);
                if (ioctl_ret < 0 && errno != EOVERFLOW)
                        return -errno;
 
                for (uint8_t i = 0; i < rootref_args.num_items; i++) {
                        _cleanup_free_ char *c = NULL, *np = NULL;
                        _cleanup_close_ int old_child_fd = -EBADF, new_child_fd = -EBADF;
                        uint64_t child_subvol_id = rootref_args.rootref[i].treeid;
 
                        /* Avoid running into loops if the new subvolume is below the old one. */
                        if (child_subvol_id == new_subvol_id)
                                continue;
 
                        struct btrfs_ioctl_ino_lookup_user_args lookup_args = {
                                .dirid = rootref_args.rootref[i].dirid,
                                .treeid = child_subvol_id,
                        };
                        if (ioctl(old_fd, BTRFS_IOC_INO_LOOKUP_USER, &lookup_args) < 0)
                                return -errno;
 
                        c = path_join(lookup_args.path, lookup_args.name);
                        if (!c)
                                return -ENOMEM;
 
                        old_child_fd = openat(old_fd, c, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW);
                        if (old_child_fd < 0)
                                return -errno;
 
                        np = path_join(subvolume, lookup_args.path);
                        if (!np)
                                return -ENOMEM;
 
                        new_child_fd = openat(new_fd, np, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW);
                        if (new_child_fd < 0)
                                return -errno;
 
                        if (flags & BTRFS_SNAPSHOT_READ_ONLY) {
                                /* If the snapshot is read-only we need to mark it writable temporarily, to
                                 * put the subsnapshot into place. */
 
                                if (subvolume_fd < 0) {
                                        subvolume_fd = openat(new_fd, subvolume, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW);
                                        if (subvolume_fd < 0)
                                                return -errno;
                                }
 
                                r = btrfs_subvol_set_read_only_fd(subvolume_fd, false);
                                if (r < 0)
                                        return r;
                        }
 
                        /* When btrfs clones the subvolumes, child subvolumes appear as empty
                         * directories. Remove them, so that we can create a new snapshot in their place */
                        if (unlinkat(new_child_fd, lookup_args.name, AT_REMOVEDIR) < 0) {
                                int k = -errno;
 
                                if (flags & BTRFS_SNAPSHOT_READ_ONLY)
                                        (void) btrfs_subvol_set_read_only_fd(subvolume_fd, true);
 
                                return k;
                        }
 
                        r = subvol_snapshot_children(old_child_fd, new_child_fd, lookup_args.name, child_subvol_id,
                                                     flags & ~(BTRFS_SNAPSHOT_FALLBACK_COPY|BTRFS_SNAPSHOT_LOCK_BSD));
 
                        /* Restore the readonly flag */
                        if (flags & BTRFS_SNAPSHOT_READ_ONLY) {
                                int k;
 
                                k = btrfs_subvol_set_read_only_fd(subvolume_fd, true);
                                if (r >= 0 && k < 0)
                                        return k;
                        }
 
                        if (r < 0)
                                return r;
                }
 
                if (ioctl_ret >= 0)
                        break;
        }
 
        if (flags & BTRFS_SNAPSHOT_QUOTA)
                (void) copy_subtree_quota_limits(new_fd, old_subvol_id, new_subvol_id);
 
        return flags & BTRFS_SNAPSHOT_LOCK_BSD ? TAKE_FD(subvolume_fd) : 0;
}
 
int btrfs_subvol_snapshot_at_full(
                int dir_fdf,
                const char *from,
                int dir_fdt,
                const char *to,
                BtrfsSnapshotFlags flags,
                copy_progress_path_t progress_path,
                copy_progress_bytes_t progress_bytes,
                void *userdata) {
 
        _cleanup_free_ char *subvolume = NULL;
        _cleanup_close_ int old_fd = -EBADF, new_fd = -EBADF, subvolume_fd = -EBADF;
        int r;
 
        assert(dir_fdf >= 0 || dir_fdf == AT_FDCWD);
        assert(dir_fdt >= 0 || dir_fdt == AT_FDCWD);
        assert(to);
 
        old_fd = xopenat(dir_fdf, from, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
        if (old_fd < 0)
                return old_fd;
 
        new_fd = chase_and_openat(XAT_FDROOT, dir_fdt, to, CHASE_PARENT|CHASE_EXTRACT_FILENAME, O_CLOEXEC, &subvolume);
        if (new_fd < 0)
                return new_fd;
 
        r = btrfs_validate_subvolume_name(subvolume);
        if (r < 0)
                return r;
 
        r = btrfs_is_subvol_at(dir_fdf, from);
        if (r < 0)
                return r;
        if (r == 0) {
                bool plain_directory = false;
 
                /* If the source isn't a proper subvolume, fail unless fallback is requested */
                if (!(flags & BTRFS_SNAPSHOT_FALLBACK_COPY))
                        return -EISDIR;
 
                r = btrfs_subvol_make(new_fd, subvolume);
                if (r < 0) {
                        if (ERRNO_IS_NOT_SUPPORTED(r) && (flags & BTRFS_SNAPSHOT_FALLBACK_DIRECTORY)) {
                                /* If the destination doesn't support subvolumes, then use a plain directory, if that's requested. */
                                if (mkdirat(new_fd, subvolume, 0755) < 0)
                                        return -errno;
 
                                plain_directory = true;
                        } else
                                return r;
                }
 
                if (FLAGS_SET(flags, BTRFS_SNAPSHOT_LOCK_BSD)) {
                        subvolume_fd = xopenat_lock(new_fd, subvolume,
                                                    O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW,
                                                    LOCK_BSD,
                                                    LOCK_EX);
                        if (subvolume_fd < 0)
                                return subvolume_fd;
 
                        if (!plain_directory) {
                                r = btrfs_is_subvol_fd(subvolume_fd);
                                if (r < 0)
                                        return r;
                                if (r == 0)
                                        return -EEXIST;
                        }
                }
 
                r = copy_directory_at_full(
                                dir_fdf, from,
                                new_fd, subvolume,
                                /* override_uid= */ UID_INVALID,
                                /* override_gid= */ GID_INVALID,
                                COPY_MERGE_EMPTY|
                                COPY_SAME_MOUNT|
                                COPY_HARDLINKS|
                                COPY_ALL_XATTRS|
                                (FLAGS_SET(flags, BTRFS_SNAPSHOT_SIGINT) ? COPY_SIGINT : 0)|
                                (FLAGS_SET(flags, BTRFS_SNAPSHOT_SIGTERM) ? COPY_SIGTERM : 0),
                                progress_path,
                                progress_bytes,
                                userdata);
                if (r < 0)
                        goto fallback_fail;
 
                if (flags & BTRFS_SNAPSHOT_READ_ONLY) {
 
                        if (plain_directory) {
                                /* Plain directories have no recursive read-only flag, but something pretty close to
                                 * it: the IMMUTABLE bit. Let's use this here, if this is requested. */
 
                                if (flags & BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE)
                                        (void) chattr_at(new_fd, subvolume, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL);
                        } else {
                                r = btrfs_subvol_set_read_only_at(new_fd, subvolume, true);
                                if (r < 0)
                                        goto fallback_fail;
                        }
                }
 
                return flags & BTRFS_SNAPSHOT_LOCK_BSD ? TAKE_FD(subvolume_fd) : 0;
 
        fallback_fail:
                (void) rm_rf_at(new_fd, subvolume, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
                return r;
        }
 
        return subvol_snapshot_children(old_fd, new_fd, subvolume, 0, flags);
}
 
int btrfs_qgroup_find_parents(int fd, uint64_t qgroupid, uint64_t **ret) {
 
        struct btrfs_ioctl_search_args args = {
                /* Tree of quota items */
                .key.tree_id = BTRFS_QUOTA_TREE_OBJECTID,
 
                /* Look precisely for the quota relation items */
                .key.min_type = BTRFS_QGROUP_RELATION_KEY,
                .key.max_type = BTRFS_QGROUP_RELATION_KEY,
 
                /* No restrictions on the other components */
                .key.min_offset = 0,
                .key.max_offset = UINT64_MAX,
 
                .key.min_transid = 0,
                .key.max_transid = UINT64_MAX,
        };
 
        _cleanup_free_ uint64_t *items = NULL;
        size_t n_items = 0;
        int r;
 
        assert(fd >= 0);
        assert(ret);
 
        if (qgroupid == 0) {
                r = btrfs_subvol_get_id_fd(fd, &qgroupid);
                if (r < 0)
                        return r;
        } else {
                r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
                if (r < 0)
                        return r;
                if (r == 0)
                        return -ENOTTY;
        }
 
        args.key.min_objectid = args.key.max_objectid = qgroupid;
 
        while (btrfs_ioctl_search_args_compare(&args) <= 0) {
                struct btrfs_ioctl_search_header sh;
                _unused_ const void *body;
 
                args.key.nr_items = 256;
                if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args) < 0) {
                        if (errno == ENOENT) /* quota tree missing: quota is disabled */
                                break;
 
                        return -errno;
                }
 
                if (args.key.nr_items <= 0)
                        break;
 
                FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, args) {
 
                        /* Make sure we start the next search at least from this entry */
                        btrfs_ioctl_search_args_set(&args, &sh);
 
                        if (sh.type != BTRFS_QGROUP_RELATION_KEY)
                                continue;
                        if (sh.offset < sh.objectid)
                                continue;
                        if (sh.objectid != qgroupid)
                                continue;
 
                        if (!GREEDY_REALLOC(items, n_items+1))
                                return -ENOMEM;
 
                        items[n_items++] = sh.offset;
                }
 
                /* Increase search key by one, to read the next item, if we can. */
                if (!btrfs_ioctl_search_args_inc(&args))
                        break;
        }
 
        assert((n_items > 0) == !!items);
        assert(n_items <= INT_MAX);
 
        *ret = TAKE_PTR(items);
        return (int) n_items;
}
 
int btrfs_subvol_auto_qgroup_fd(int fd, uint64_t subvol_id, bool insert_intermediary_qgroup) {
        _cleanup_free_ uint64_t *qgroups = NULL;
        _cleanup_close_ int real_fd = -EBADF;
        uint64_t parent_subvol;
        bool changed = false;
        int n = 0, r;
 
        assert(fd >= 0);
 
        /*
         * Sets up the specified subvolume's qgroup automatically in
         * one of two ways:
         *
         * If insert_intermediary_qgroup is false, the subvolume's
         * leaf qgroup will be assigned to the same parent qgroups as
         * the subvolume's parent subvolume.
         *
         * If insert_intermediary_qgroup is true a new intermediary
         * higher-level qgroup is created, with a higher level number,
         * but reusing the id of the subvolume. The level number is
         * picked as one smaller than the lowest level qgroup the
         * parent subvolume is a member of. If the parent subvolume's
         * leaf qgroup is assigned to no higher-level qgroup a new
         * qgroup of level 255 is created instead. Either way, the new
         * qgroup is then assigned to the parent's higher-level
         * qgroup, and the subvolume itself is assigned to it.
         *
         * If the subvolume is already assigned to a higher level
         * qgroup, no operation is executed.
         *
         * Effectively this means: regardless if
         * insert_intermediary_qgroup is true or not, after this
         * function is invoked the subvolume will be accounted within
         * the same qgroups as the parent. However, if it is true, it
         * will also get its own higher-level qgroup, which may in
         * turn be used by subvolumes created beneath this subvolume
         * later on.
         *
         * This hence defines a simple default qgroup setup for
         * subvolumes, as long as this function is invoked on each
         * created subvolume: each subvolume is always accounting
         * together with its immediate parents. Optionally, if
         * insert_intermediary_qgroup is true, it will also get a
         * qgroup that then includes all its own child subvolumes.
         */
 
        /* Turn this into a proper fd, if it is currently O_PATH */
        fd = fd_reopen_condition(fd, O_RDONLY|O_CLOEXEC, O_PATH, &real_fd);
        if (fd < 0)
                return fd;
 
        if (subvol_id == 0) {
                r = btrfs_is_subvol_fd(fd);
                if (r < 0)
                        return r;
                if (!r)
                        return -ENOTTY;
 
                r = btrfs_subvol_get_id_fd(fd, &subvol_id);
                if (r < 0)
                        return r;
        }
 
        n = btrfs_qgroup_find_parents(fd, subvol_id, &qgroups);
        if (n < 0)
                return n;
        if (n > 0) /* already parent qgroups set up, let's bail */
                return 0;
 
        qgroups = mfree(qgroups);
 
        r = btrfs_subvol_get_parent(fd, subvol_id, &parent_subvol);
        if (r == -ENXIO)
                /* No parent, hence no qgroup memberships */
                n = 0;
        else if (r < 0)
                return r;
        else {
                n = btrfs_qgroup_find_parents(fd, parent_subvol, &qgroups);
                if (n < 0)
                        return n;
        }
 
        if (insert_intermediary_qgroup) {
                uint64_t lowest = 256, new_qgroupid;
                bool created = false;
 
                /* Determine the lowest qgroup that the parent
                 * subvolume is assigned to. */
 
                for (int i = 0; i < n; i++) {
                        uint64_t level;
 
                        r = btrfs_qgroupid_split(qgroups[i], &level, NULL);
                        if (r < 0)
                                return r;
 
                        if (level < lowest)
                                lowest = level;
                }
 
                if (lowest <= 1) /* There are no levels left we could use insert an intermediary qgroup at */
                        return -EBUSY;
 
                r = btrfs_qgroupid_make(lowest - 1, subvol_id, &new_qgroupid);
                if (r < 0)
                        return r;
 
                /* Create the new intermediary group, unless it already exists */
                r = btrfs_qgroup_create(fd, new_qgroupid);
                if (r < 0 && r != -EEXIST)
                        return r;
                if (r >= 0)
                        changed = created = true;
 
                for (int i = 0; i < n; i++) {
                        r = btrfs_qgroup_assign(fd, new_qgroupid, qgroups[i]);
                        if (r < 0 && r != -EEXIST) {
                                if (created)
                                        (void) btrfs_qgroup_destroy_recursive(fd, new_qgroupid);
 
                                return r;
                        }
                        if (r >= 0)
                                changed = true;
                }
 
                r = btrfs_qgroup_assign(fd, subvol_id, new_qgroupid);
                if (r < 0 && r != -EEXIST) {
                        if (created)
                                (void) btrfs_qgroup_destroy_recursive(fd, new_qgroupid);
                        return r;
                }
                if (r >= 0)
                        changed = true;
 
        } else {
                int i;
 
                /* Assign our subvolume to all the same qgroups as the parent */
 
                for (i = 0; i < n; i++) {
                        r = btrfs_qgroup_assign(fd, subvol_id, qgroups[i]);
                        if (r < 0 && r != -EEXIST)
                                return r;
                        if (r >= 0)
                                changed = true;
                }
        }
 
        return changed;
}
 
int btrfs_subvol_auto_qgroup(const char *path, uint64_t subvol_id, bool create_intermediary_qgroup) {
        _cleanup_close_ int fd = -EBADF;
 
        fd = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
        if (fd < 0)
                return -errno;
 
        return btrfs_subvol_auto_qgroup_fd(fd, subvol_id, create_intermediary_qgroup);
}
 
int btrfs_subvol_make_default(const char *path) {
        _cleanup_close_ int fd = -EBADF;
        uint64_t id;
        int r;
 
        assert(path);
 
        fd = open(path, O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
        if (fd < 0)
                return -errno;
 
        r = btrfs_subvol_get_id_fd(fd, &id);
        if (r < 0)
                return r;
 
        return RET_NERRNO(ioctl(fd, BTRFS_IOC_DEFAULT_SUBVOL, &id));
}
 
int btrfs_subvol_get_parent(int fd, uint64_t subvol_id, uint64_t *ret) {
 
        struct btrfs_ioctl_search_args args = {
                /* Tree of tree roots */
                .key.tree_id = BTRFS_ROOT_TREE_OBJECTID,
 
                /* Look precisely for the subvolume items */
                .key.min_type = BTRFS_ROOT_BACKREF_KEY,
                .key.max_type = BTRFS_ROOT_BACKREF_KEY,
 
                /* No restrictions on the other components */
                .key.min_offset = 0,
                .key.max_offset = UINT64_MAX,
 
                .key.min_transid = 0,
                .key.max_transid = UINT64_MAX,
        };
        int r;
 
        assert(fd >= 0);
        assert(ret);
 
        if (subvol_id == 0) {
                r = btrfs_subvol_get_id_fd(fd, &subvol_id);
                if (r < 0)
                        return r;
        } else {
                r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
                if (r < 0)
                        return r;
                if (r == 0)
                        return -ENOTTY;
        }
 
        args.key.min_objectid = args.key.max_objectid = subvol_id;
 
        while (btrfs_ioctl_search_args_compare(&args) <= 0) {
                struct btrfs_ioctl_search_header sh;
                _unused_ const void *body = NULL;
 
                args.key.nr_items = 256;
                if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args) < 0)
                        return negative_errno();
 
                if (args.key.nr_items <= 0)
                        break;
 
                FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, args) {
 
                        if (sh.type != BTRFS_ROOT_BACKREF_KEY)
                                continue;
                        if (sh.objectid != subvol_id)
                                continue;
 
                        *ret = sh.offset;
                        return 0;
                }
        }
 
        return -ENXIO;
}
 
bool btrfs_might_be_subvol(const struct stat *st) {
        if (!st)
                return false;
 
        /* Returns true if this 'struct stat' looks like it could refer to a btrfs subvolume. To make a final
         * decision, needs to be combined with an fstatfs() check to see if this is actually btrfs. */
 
        return S_ISDIR(st->st_mode) && st->st_ino == 256;
}
 
int btrfs_forget_device(const char *path) {
        _cleanup_close_ int control_fd = -EBADF;
        struct btrfs_ioctl_vol_args args = {};
 
        assert(path);
 
        if (strlen(path) > BTRFS_PATH_NAME_MAX)
                return -E2BIG;
 
        strcpy(args.name, path);
 
        control_fd = open("/dev/btrfs-control", O_RDWR|O_CLOEXEC);
        if (control_fd < 0)
                return -errno;
 
        return RET_NERRNO(ioctl(control_fd, BTRFS_IOC_FORGET_DEV, &args));
}
 
typedef struct BtrfsStripe {
        uint64_t devid;
        uint64_t offset;
} BtrfsStripe;
 
typedef struct BtrfsChunk {
        uint64_t offset;
        uint64_t length;
        uint64_t type;
 
        BtrfsStripe *stripes;
        uint16_t n_stripes;
        uint64_t stripe_len;
} BtrfsChunk;
 
typedef struct BtrfsChunkTree {
        BtrfsChunk **chunks;
        size_t n_chunks;
} BtrfsChunkTree;
 
static BtrfsChunk* btrfs_chunk_free(BtrfsChunk *chunk) {
        if (!chunk)
                return NULL;
 
        free(chunk->stripes);
 
        return mfree(chunk);
}
 
DEFINE_TRIVIAL_CLEANUP_FUNC(BtrfsChunk*, btrfs_chunk_free);
 
static void btrfs_chunk_tree_done(BtrfsChunkTree *tree) {
        assert(tree);
 
        FOREACH_ARRAY(i, tree->chunks, tree->n_chunks)
                btrfs_chunk_free(*i);
 
        free(tree->chunks);
}
 
static int btrfs_read_chunk_tree_fd(int fd, BtrfsChunkTree *ret) {
 
        struct btrfs_ioctl_search_args search_args = {
                .key.tree_id = BTRFS_CHUNK_TREE_OBJECTID,
 
                .key.min_type = BTRFS_CHUNK_ITEM_KEY,
                .key.max_type = BTRFS_CHUNK_ITEM_KEY,
 
                .key.min_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID,
                .key.max_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 
                .key.min_offset = 0,
                .key.max_offset = UINT64_MAX,
 
                .key.min_transid = 0,
                .key.max_transid = UINT64_MAX,
        };
 
        _cleanup_(btrfs_chunk_tree_done) BtrfsChunkTree tree = {};
 
        assert(fd >= 0);
        assert(ret);
 
        while (btrfs_ioctl_search_args_compare(&search_args) <= 0) {
                struct btrfs_ioctl_search_header sh;
                const void *body;
 
                search_args.key.nr_items = 256;
 
                if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &search_args) < 0)
                        return -errno;
 
                if (search_args.key.nr_items == 0)
                        break;
 
                FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, search_args) {
                        _cleanup_(btrfs_chunk_freep) BtrfsChunk *chunk = NULL;
 
                        btrfs_ioctl_search_args_set(&search_args, &sh);
 
                        if (sh.objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)
                                continue;
                        if (sh.type != BTRFS_CHUNK_ITEM_KEY)
                                continue;
 
                        if (sh.len < sizeof(struct btrfs_chunk))
                                return -EBADMSG;
 
                        chunk = new(BtrfsChunk, 1);
                        if (!chunk)
                                return -ENOMEM;
 
                        const struct btrfs_chunk *item = body;
 
                        if (le16toh(item->num_stripes) == 0 || le64toh(item->stripe_len) == 0)
                                return -EBADMSG;
 
                        if ((sh.len - sizeof(struct btrfs_chunk)) / sizeof(struct btrfs_stripe) + 1 < le16toh(item->num_stripes))
                                return -EBADMSG;
 
                        *chunk = (BtrfsChunk) {
                                .offset = sh.offset,
                                .length = le64toh(item->length),
                                .type = le64toh(item->type),
                                .n_stripes = le16toh(item->num_stripes),
                                .stripe_len = le64toh(item->stripe_len),
                        };
 
                        chunk->stripes = new(BtrfsStripe, chunk->n_stripes);
                        if (!chunk->stripes)
                                return -ENOMEM;
 
                        for (size_t j = 0; j < chunk->n_stripes; j++) {
                                const struct btrfs_stripe *stripe = &item->stripe + j;
 
                                chunk->stripes[j] = (BtrfsStripe) {
                                        .devid = le64toh(stripe->devid),
                                        .offset = le64toh(stripe->offset),
                                };
                        }
 
                        if (!GREEDY_REALLOC(tree.chunks, tree.n_chunks + 1))
                                return -ENOMEM;
 
                        tree.chunks[tree.n_chunks++] = TAKE_PTR(chunk);
                }
 
                if (!btrfs_ioctl_search_args_inc(&search_args))
                        break;
        }
 
        *ret = TAKE_STRUCT(tree);
        return 0;
}
 
static BtrfsChunk* btrfs_find_chunk_from_logical_address(const BtrfsChunkTree *tree, uint64_t logical) {