HALF ECHO LEARN / REFERENCE COLLECTION
v262.2src/shared/btrfs-util.c
Git blob: f03eab91fd4a4eab4a3201a372335ae0090364e8
Lines 1–1000 of 2,215
/* SPDX-License-Identifier: LGPL-2.1-or-later */
#include <fcntl.h>
#include <linux/btrfs.h>
#include <linux/btrfs_tree.h>
#include <linux/magic.h>
#include <stdio.h>
#include <sys/file.h>
#include <sys/ioctl.h>
#include <sys/sysmacros.h>
#include <unistd.h>
#include "alloc-util.h"
#include "btrfs-util.h"
#include "chase.h"
#include "chattr-util.h"
#include "copy.h"
#include "errno-util.h"
#include "fd-util.h"
#include "fs-util.h"
#include "log.h"
#include "path-util.h"
#include "rm-rf.h"
#include "sparse-endian.h"
#include "stat-util.h"
#include "stdio-util.h"
#include "string-util.h"
#include "time-util.h"
/* WARNING: Be careful with file system ioctls! When we get an fd, we
* need to make sure it either refers to only a regular file or
* directory, or that it is located on btrfs, before invoking any
* btrfs ioctls. The ioctl numbers are reused by some device drivers
* (such as DRM), and hence might have bad effects when invoked on
* device nodes (that reference drivers) rather than fds to normal
* files or directories. */
int btrfs_is_subvol_at(int dir_fd, const char *path) {
struct stat st;
assert(dir_fd >= 0 || dir_fd == AT_FDCWD);
/* On btrfs subvolumes always have the inode 256 */
if (fstatat(dir_fd, strempty(path), &st, isempty(path) ? AT_EMPTY_PATH : 0) < 0)
return -errno;
if (!btrfs_might_be_subvol(&st))
return 0;
return is_fs_type_at(dir_fd, path, BTRFS_SUPER_MAGIC);
}
int btrfs_subvol_set_read_only_at(int dir_fd, const char *path, bool b) {
_cleanup_close_ int fd = -EBADF;
uint64_t flags, nflags;
struct stat st;
assert(dir_fd >= 0 || dir_fd == AT_FDCWD);
fd = xopenat(dir_fd, path, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
if (fd < 0)
return fd;
if (fstat(fd, &st) < 0)
return -errno;
if (!btrfs_might_be_subvol(&st))
return -EINVAL;
if (ioctl(fd, BTRFS_IOC_SUBVOL_GETFLAGS, &flags) < 0)
return -errno;
nflags = UPDATE_FLAG(flags, BTRFS_SUBVOL_RDONLY, b);
if (flags == nflags)
return 0;
return RET_NERRNO(ioctl(fd, BTRFS_IOC_SUBVOL_SETFLAGS, &nflags));
}
int btrfs_subvol_get_read_only_fd(int fd) {
uint64_t flags;
struct stat st;
assert(fd >= 0);
if (fstat(fd, &st) < 0)
return -errno;
if (!btrfs_might_be_subvol(&st))
return -EINVAL;
if (ioctl(fd, BTRFS_IOC_SUBVOL_GETFLAGS, &flags) < 0)
return -errno;
return !!(flags & BTRFS_SUBVOL_RDONLY);
}
int btrfs_get_block_device_at_full(int dir_fd, const char *path, uint64_t *ret_devid, char **ret_path, dev_t *ret) {
struct btrfs_ioctl_fs_info_args fsi = {};
_cleanup_close_ int fd = -EBADF;
uint64_t id;
int r;
/*
* Returns:
* ret_devid - the device id in the filesystem for the returned block device
* ret_path - the path to the returned block device
* ret - the returned block device
*/
assert(wildcard_fd_is_valid(dir_fd));
fd = xopenat(dir_fd, path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
if (fd < 0)
return fd;
r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
if (r < 0)
return r;
if (r == 0)
return -ENOTTY;
if (ioctl(fd, BTRFS_IOC_FS_INFO, &fsi) < 0)
return -errno;
/* We won't do this for btrfs RAID */
if (fsi.num_devices != 1) {
if (ret_devid)
*ret_devid = 0;
if (ret_path)
*ret_path = NULL;
if (ret)
*ret = 0;
return 0;
}
for (id = 1; id <= fsi.max_id; id++) {
struct btrfs_ioctl_dev_info_args di = {
.devid = id,
};
struct stat st;
_cleanup_free_ char *device_path = NULL;
if (ioctl(fd, BTRFS_IOC_DEV_INFO, &di) < 0) {
if (errno == ENODEV)
continue;
return -errno;
}
/* For the root fs — when no initrd is involved — btrfs returns /dev/root on any kernels from
* the past few years. That sucks, as we have no API to determine the actual root then. let's
* return an recognizable error for this case, so that the caller can maybe print a nice
* message about this.
*
* https://bugzilla.kernel.org/show_bug.cgi?id=89721 */
if (path_equal((char*) di.path, "/dev/root"))
return -EUCLEAN;
if (stat((char*) di.path, &st) < 0)
return -errno;
r = stat_verify_block(&st);
if (r < 0)
return r;
if (major(st.st_rdev) == 0)
return -ENODEV;
device_path = strdup((char*) di.path);
if (!device_path)
return -ENOMEM;
if (ret_path)
*ret_path = TAKE_PTR(device_path);
if (ret)
*ret = st.st_rdev;
if (ret_devid)
*ret_devid = id;
return 1;
}
return -ENODEV;
}
int btrfs_subvol_get_id_fd(int fd, uint64_t *ret) {
struct btrfs_ioctl_ino_lookup_args args = {
.objectid = BTRFS_FIRST_FREE_OBJECTID
};
int r;
assert(fd >= 0);
assert(ret);
r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
if (r < 0)
return r;
if (r == 0)
return -ENOTTY;
if (ioctl(fd, BTRFS_IOC_INO_LOOKUP, &args) < 0)
return -errno;
*ret = args.treeid;
return 0;
}
int btrfs_subvol_get_id(int fd, const char *subvol, uint64_t *ret) {
_cleanup_close_ int subvol_fd = -EBADF;
assert(fd >= 0);
assert(ret);
subvol_fd = openat(fd, subvol, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
if (subvol_fd < 0)
return -errno;
return btrfs_subvol_get_id_fd(subvol_fd, ret);
}
static bool btrfs_ioctl_search_args_inc(struct btrfs_ioctl_search_args *args) {
assert(args);
/* the objectid, type, offset together make up the btrfs key,
* which is considered a single 136byte integer when
* comparing. This call increases the counter by one, dealing
* with the overflow between the overflows */
if (args->key.min_offset < UINT64_MAX) {
args->key.min_offset++;
return true;
}
if (args->key.min_type < UINT8_MAX) {
args->key.min_type++;
args->key.min_offset = 0;
return true;
}
if (args->key.min_objectid < UINT64_MAX) {
args->key.min_objectid++;
args->key.min_offset = 0;
args->key.min_type = 0;
return true;
}
return 0;
}
static void btrfs_ioctl_search_args_set(struct btrfs_ioctl_search_args *args, const struct btrfs_ioctl_search_header *h) {
assert(args);
assert(h);
args->key.min_objectid = h->objectid;
args->key.min_type = h->type;
args->key.min_offset = h->offset;
}
static int btrfs_ioctl_search_args_compare(const struct btrfs_ioctl_search_args *args) {
int r;
assert(args);
/* Compare min and max */
r = CMP(args->key.min_objectid, args->key.max_objectid);
if (r != 0)
return r;
r = CMP(args->key.min_type, args->key.max_type);
if (r != 0)
return r;
return CMP(args->key.min_offset, args->key.max_offset);
}
typedef struct BtrfsForeachIterator {
const struct btrfs_ioctl_search_args *args;
size_t offset;
unsigned index;
struct btrfs_ioctl_search_header *header;
const void **body;
} BtrfsForeachIterator;
static int btrfs_iterate(BtrfsForeachIterator *i) {
assert(i);
assert(i->args);
assert(i->header);
assert(i->body);
if (i->index >= i->args->key.nr_items)
return 0; /* end */
assert_cc(BTRFS_SEARCH_ARGS_BUFSIZE >= sizeof(struct btrfs_ioctl_search_header));
if (i->offset > BTRFS_SEARCH_ARGS_BUFSIZE - sizeof(struct btrfs_ioctl_search_header))
return -EBADMSG;
struct btrfs_ioctl_search_header h;
memcpy(&h, (const uint8_t*) i->args->buf + i->offset, sizeof(struct btrfs_ioctl_search_header));
if (i->offset > BTRFS_SEARCH_ARGS_BUFSIZE - sizeof(struct btrfs_ioctl_search_header) - h.len)
return -EBADMSG;
*i->body = (const uint8_t*) i->args->buf + i->offset + sizeof(struct btrfs_ioctl_search_header);
*i->header = h;
i->offset += sizeof(struct btrfs_ioctl_search_header) + h.len;
i->index++;
return 1;
}
/* Iterates through a series of struct btrfs_file_extent_item elements. They are unfortunately not aligned,
* hence we copy out the header from them */
#define FOREACH_BTRFS_IOCTL_SEARCH_HEADER(_sh, _body, _args) \
for (BtrfsForeachIterator iterator = { \
.args = &(_args), \
.header = &(_sh), \
.body = &(_body), \
}; \
btrfs_iterate(&iterator) > 0; )
int btrfs_subvol_get_info_fd(int fd, uint64_t subvol_id, BtrfsSubvolInfo *ret) {
int r;
assert(fd >= 0);
assert(ret);
/* Make sure this works on O_PATH fds */
_cleanup_close_ int fd_close = -EBADF;
fd = fd_reopen_condition(fd, O_CLOEXEC|O_RDONLY|O_DIRECTORY, O_PATH, &fd_close);
if (fd < 0)
return fd;
r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
if (r < 0)
return r;
if (r == 0)
return -ENOTTY;
struct btrfs_ioctl_get_subvol_info_args info;
if (ioctl(fd, BTRFS_IOC_GET_SUBVOL_INFO, &info) < 0)
return -errno;
*ret = (BtrfsSubvolInfo) {
.subvol_id = info.treeid,
.otime = info.otime.sec * USEC_PER_SEC + (info.otime.nsec / NSEC_PER_USEC),
.ctime = info.ctime.sec * USEC_PER_SEC + (info.ctime.nsec / NSEC_PER_USEC),
.read_only = FLAGS_SET(info.flags, BTRFS_SUBVOL_RDONLY),
};
assert_cc(sizeof(info.uuid) == sizeof(sd_id128_t));
memcpy(&ret->uuid, info.uuid, sizeof(sd_id128_t));
assert_cc(sizeof(info.parent_uuid) == sizeof(sd_id128_t));
memcpy(&ret->parent_uuid, info.parent_uuid, sizeof(sd_id128_t));
return 0;
}
int btrfs_qgroup_get_quota_fd(int fd, uint64_t qgroupid, BtrfsQuotaInfo *ret) {
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_STATUS_KEY,
.key.max_type = BTRFS_QGROUP_LIMIT_KEY,
/* No restrictions on the other components */
.key.min_transid = 0,
.key.max_transid = UINT64_MAX,
};
bool found_info = false, found_limit = false;
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_offset = args.key.max_offset = qgroupid;
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 is missing: quota 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.objectid != 0)
continue;
if (sh.offset != qgroupid)
continue;
if (sh.type == BTRFS_QGROUP_INFO_KEY) {
const struct btrfs_qgroup_info_item *qii = body;
ret->referenced = le64toh(qii->rfer);
ret->exclusive = le64toh(qii->excl);
found_info = true;
} else if (sh.type == BTRFS_QGROUP_LIMIT_KEY) {
const struct btrfs_qgroup_limit_item *qli = body;
if (le64toh(qli->flags) & BTRFS_QGROUP_LIMIT_MAX_RFER)
ret->referenced_max = le64toh(qli->max_rfer);
else
ret->referenced_max = UINT64_MAX;
if (le64toh(qli->flags) & BTRFS_QGROUP_LIMIT_MAX_EXCL)
ret->exclusive_max = le64toh(qli->max_excl);
else
ret->exclusive_max = UINT64_MAX;
found_limit = true;
}
if (found_info && found_limit)
goto finish;
}
/* Increase search key by one, to read the next item, if we can. */
if (!btrfs_ioctl_search_args_inc(&args))
break;
}
finish:
if (!found_limit && !found_info)
return -ENODATA;
if (!found_info) {
ret->referenced = UINT64_MAX;
ret->exclusive = UINT64_MAX;
}
if (!found_limit) {
ret->referenced_max = UINT64_MAX;
ret->exclusive_max = UINT64_MAX;
}
return 0;
}
int btrfs_log_dev_root(int level, int ret, const char *p) {
return log_full_errno(level, ret,
"File system behind %s is reported by btrfs to be backed by pseudo-device /dev/root, which is not a valid userspace accessible device node. "
"Cannot determine correct backing block device.", p);
}
int btrfs_qgroup_get_quota(const char *path, uint64_t qgroupid, BtrfsQuotaInfo *ret) {
_cleanup_close_ int fd = -EBADF;
fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
if (fd < 0)
return -errno;
return btrfs_qgroup_get_quota_fd(fd, qgroupid, ret);
}
int btrfs_subvol_find_subtree_qgroup(int fd, uint64_t subvol_id, uint64_t *ret) {
uint64_t level, lowest = UINT64_MAX, lowest_qgroupid = 0;
_cleanup_free_ uint64_t *qgroups = NULL;
int r, n;
assert(fd >= 0);
assert(ret);
/* This finds the "subtree" qgroup for a specific
* subvolume. This only works for subvolumes that have been
* prepared with btrfs_subvol_auto_qgroup_fd() with
* insert_intermediary_qgroup=true (or equivalent). For others
* it will return the leaf qgroup instead. The two cases may
* be distinguished via the return value, which is 1 in case
* an appropriate "subtree" qgroup was found, and 0
* otherwise. */
if (subvol_id == 0) {
r = btrfs_subvol_get_id_fd(fd, &subvol_id);
if (r < 0)
return r;
}
r = btrfs_qgroupid_split(subvol_id, &level, NULL);
if (r < 0)
return r;
if (level != 0) /* Input must be a leaf qgroup */
return -EINVAL;
n = btrfs_qgroup_find_parents(fd, subvol_id, &qgroups);
if (n < 0)
return n;
for (int i = 0; i < n; i++) {
uint64_t id;
r = btrfs_qgroupid_split(qgroups[i], &level, &id);
if (r < 0)
return r;
if (id != subvol_id)
continue;
if (lowest == UINT64_MAX || level < lowest) {
lowest_qgroupid = qgroups[i];
lowest = level;
}
}
if (lowest == UINT64_MAX) {
/* No suitable higher-level qgroup found, let's return
* the leaf qgroup instead, and indicate that with the
* return value. */
*ret = subvol_id;
return 0;
}
*ret = lowest_qgroupid;
return 1;
}
int btrfs_subvol_get_subtree_quota_fd(int fd, uint64_t subvol_id, BtrfsQuotaInfo *ret) {
uint64_t qgroupid;
int r;
assert(fd >= 0);
assert(ret);
/* This determines the quota data of the qgroup with the
* lowest level, that shares the id part with the specified
* subvolume. This is useful for determining the quota data
* for entire subvolume subtrees, as long as the subtrees have
* been set up with btrfs_qgroup_subvol_auto_fd() or in a
* compatible way */
r = btrfs_subvol_find_subtree_qgroup(fd, subvol_id, &qgroupid);
if (r < 0)
return r;
return btrfs_qgroup_get_quota_fd(fd, qgroupid, ret);
}
int btrfs_subvol_get_subtree_quota(const char *path, uint64_t subvol_id, BtrfsQuotaInfo *ret) {
_cleanup_close_ int fd = -EBADF;
fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
if (fd < 0)
return -errno;
return btrfs_subvol_get_subtree_quota_fd(fd, subvol_id, ret);
}
int btrfs_defrag_fd(int fd) {
int r;
assert(fd >= 0);
r = fd_verify_regular(fd);
if (r < 0)
return r;
return RET_NERRNO(ioctl(fd, BTRFS_IOC_DEFRAG, NULL));
}
int btrfs_defrag(const char *p) {
_cleanup_close_ int fd = -EBADF;
fd = open(p, O_RDWR|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
if (fd < 0)
return -errno;
return btrfs_defrag_fd(fd);
}
int btrfs_quota_enable_fd(int fd, bool b) {
struct btrfs_ioctl_quota_ctl_args args = {
.cmd = b ? BTRFS_QUOTA_CTL_ENABLE : BTRFS_QUOTA_CTL_DISABLE,
};
int r;
assert(fd >= 0);
r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
if (r < 0)
return r;
if (r == 0)
return -ENOTTY;
return RET_NERRNO(ioctl(fd, BTRFS_IOC_QUOTA_CTL, &args));
}
int btrfs_quota_enable(const char *path, bool b) {
_cleanup_close_ int fd = -EBADF;
fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
if (fd < 0)
return -errno;
return btrfs_quota_enable_fd(fd, b);
}
int btrfs_qgroup_set_limit_fd(int fd, uint64_t qgroupid, uint64_t referenced_max) {
struct btrfs_ioctl_qgroup_limit_args args = {
.lim.max_rfer = referenced_max,
.lim.flags = BTRFS_QGROUP_LIMIT_MAX_RFER,
};
int r;
assert(fd >= 0);
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.qgroupid = qgroupid;
for (unsigned c = 0;; c++) {
if (ioctl(fd, BTRFS_IOC_QGROUP_LIMIT, &args) < 0) {
if (errno == EBUSY && c < 10) {
(void) btrfs_quota_scan_wait(fd);
continue;
}
return -errno;
}
break;
}
return 0;
}
int btrfs_qgroup_set_limit(const char *path, uint64_t qgroupid, uint64_t referenced_max) {
_cleanup_close_ int fd = -EBADF;
fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
if (fd < 0)
return -errno;
return btrfs_qgroup_set_limit_fd(fd, qgroupid, referenced_max);
}
int btrfs_subvol_set_subtree_quota_limit_fd(int fd, uint64_t subvol_id, uint64_t referenced_max) {
uint64_t qgroupid;
int r;
assert(fd >= 0);
r = btrfs_subvol_find_subtree_qgroup(fd, subvol_id, &qgroupid);
if (r < 0)
return r;
return btrfs_qgroup_set_limit_fd(fd, qgroupid, referenced_max);
}
int btrfs_subvol_set_subtree_quota_limit(const char *path, uint64_t subvol_id, uint64_t referenced_max) {
_cleanup_close_ int fd = -EBADF;
fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
if (fd < 0)
return -errno;
return btrfs_subvol_set_subtree_quota_limit_fd(fd, subvol_id, referenced_max);
}
int btrfs_qgroupid_make(uint64_t level, uint64_t id, uint64_t *ret) {
assert(ret);
if (level >= (UINT64_C(1) << (64 - BTRFS_QGROUP_LEVEL_SHIFT)))
return -EINVAL;
if (id >= (UINT64_C(1) << BTRFS_QGROUP_LEVEL_SHIFT))
return -EINVAL;
*ret = (level << BTRFS_QGROUP_LEVEL_SHIFT) | id;
return 0;
}
int btrfs_qgroupid_split(uint64_t qgroupid, uint64_t *level, uint64_t *id) {
assert(level || id);
if (level)
*level = qgroupid >> BTRFS_QGROUP_LEVEL_SHIFT;
if (id)
*id = qgroupid & ((UINT64_C(1) << BTRFS_QGROUP_LEVEL_SHIFT) - 1);
return 0;
}
static int qgroup_create_or_destroy(int fd, bool b, uint64_t qgroupid) {
struct btrfs_ioctl_qgroup_create_args args = {
.create = b,
.qgroupid = qgroupid,
};
int r;
r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
if (r < 0)
return r;
if (r == 0)
return -ENOTTY;
for (unsigned c = 0;; c++) {
if (ioctl(fd, BTRFS_IOC_QGROUP_CREATE, &args) < 0) {
/* On old kernels if quota is not enabled, we get EINVAL. On newer kernels we get
* ENOTCONN. Let's always convert this to ENOTCONN to make this recognizable
* everywhere the same way. */
if (IN_SET(errno, EINVAL, ENOTCONN))
return -ENOTCONN;
if (errno == EBUSY && c < 10) {
(void) btrfs_quota_scan_wait(fd);
continue;
}
return -errno;
}
break;
}
return 0;
}
int btrfs_qgroup_create(int fd, uint64_t qgroupid) {
return qgroup_create_or_destroy(fd, true, qgroupid);
}
int btrfs_qgroup_destroy(int fd, uint64_t qgroupid) {
return qgroup_create_or_destroy(fd, false, qgroupid);
}
int btrfs_qgroup_destroy_recursive(int fd, uint64_t qgroupid) {
_cleanup_free_ uint64_t *qgroups = NULL;
uint64_t subvol_id;
int n, r;
/* Destroys the specified qgroup, but unassigns it from all
* its parents first. Also, it recursively destroys all
* qgroups it is assigned to that have the same id part of the
* qgroupid as the specified group. */
r = btrfs_qgroupid_split(qgroupid, NULL, &subvol_id);
if (r < 0)
return r;
n = btrfs_qgroup_find_parents(fd, qgroupid, &qgroups);
if (n < 0)
return n;
for (int i = 0; i < n; i++) {
uint64_t id;
r = btrfs_qgroupid_split(qgroups[i], NULL, &id);
if (r < 0)
return r;
r = btrfs_qgroup_unassign(fd, qgroupid, qgroups[i]);
if (r < 0)
return r;
if (id != subvol_id)
continue;
/* The parent qgroupid shares the same id part with
* us? If so, destroy it too. */
(void) btrfs_qgroup_destroy_recursive(fd, qgroups[i]);
}
return btrfs_qgroup_destroy(fd, qgroupid);
}
int btrfs_quota_scan_start(int fd) {
struct btrfs_ioctl_quota_rescan_args args = {};
assert(fd >= 0);
return RET_NERRNO(ioctl(fd, BTRFS_IOC_QUOTA_RESCAN, &args));
}
int btrfs_quota_scan_wait(int fd) {
assert(fd >= 0);
return RET_NERRNO(ioctl(fd, BTRFS_IOC_QUOTA_RESCAN_WAIT));
}
int btrfs_quota_scan_ongoing(int fd) {
struct btrfs_ioctl_quota_rescan_args args = {};
assert(fd >= 0);
if (ioctl(fd, BTRFS_IOC_QUOTA_RESCAN_STATUS, &args) < 0)
return -errno;
return !!args.flags;
}
static int qgroup_assign_or_unassign(int fd, bool b, uint64_t child, uint64_t parent) {
struct btrfs_ioctl_qgroup_assign_args args = {
.assign = b,
.src = child,
.dst = parent,
};
int r;
r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
if (r < 0)
return r;
if (r == 0)
return -ENOTTY;
for (unsigned c = 0;; c++) {
r = ioctl(fd, BTRFS_IOC_QGROUP_ASSIGN, &args);
if (r < 0) {
if (errno == EBUSY && c < 10) {
(void) btrfs_quota_scan_wait(fd);
continue;
}
return -errno;
}
if (r == 0)
return 0;
/* If the return value is > 0, we need to request a rescan */
(void) btrfs_quota_scan_start(fd);
return 1;
}
}
int btrfs_qgroup_assign(int fd, uint64_t child, uint64_t parent) {
return qgroup_assign_or_unassign(fd, true, child, parent);
}
int btrfs_qgroup_unassign(int fd, uint64_t child, uint64_t parent) {
return qgroup_assign_or_unassign(fd, false, child, parent);
}
static int subvol_remove_children(int fd, const char *subvolume, uint64_t subvol_id, BtrfsRemoveFlags flags) {
struct btrfs_ioctl_vol_args vol_args = {};
_cleanup_close_ int subvol_fd = -EBADF;
bool made_writable = false;
int r;
assert(fd >= 0);
assert(subvolume);
r = fd_verify_directory(fd);
if (r < 0)
return r;
subvol_fd = openat(fd, subvolume, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW);
if (subvol_fd < 0)
return -errno;
/* Let's check if this is actually a subvolume. Note that this is mostly redundant, as BTRFS_IOC_SNAP_DESTROY
* would fail anyway if it is not. However, it's a good thing to check this ahead of time so that we can return
* ENOTTY unconditionally in this case. This is different from the ioctl() which will return EPERM/EACCES if we
* don't have the privileges to remove subvolumes, regardless if the specified directory is actually a
* subvolume or not. In order to make it easy for callers to cover the "this is not a btrfs subvolume" case
* let's prefer ENOTTY over EPERM/EACCES though. */
r = btrfs_is_subvol_fd(subvol_fd);
if (r < 0)
return r;
if (r == 0) /* Not a btrfs subvolume */
return -ENOTTY;
/* Before we try anything, attempt VFS rmdir(). For an empty subvolume btrfs_rmdir dispatches
* to btrfs_delete_subvolume; for the non-empty case it returns ENOTEMPTY and we fall through
* to the destroy ioctl below. */
if (unlinkat(fd, subvolume, AT_REMOVEDIR) >= 0)
goto finish;
if (subvol_id == 0) {
r = btrfs_subvol_get_id_fd(subvol_fd, &subvol_id);
if (r < 0)
return r;
}
/* Try the destroy ioctl. Unlike unlinkat() above, this drops the entire subvolume tree, so
* regular files inside don't matter; it only returns ENOTEMPTY when there are nested
* subvolumes (BTRFS_ROOT_REF_KEY entries), which we then handle below. */
strncpy(vol_args.name, subvolume, sizeof(vol_args.name)-1);
int destroy_errno = 0;
for (;;) {
if (ioctl(fd, BTRFS_IOC_SNAP_DESTROY, &vol_args) >= 0)
goto finish;
destroy_errno = errno;
/* Without CAP_SYS_ADMIN the destroy ioctl is rejected for a read-only subvolume: with
* EROFS when 'user_subvol_rm_allowed' is set (btrfs_permission() fails the inode
* MAY_WRITE check), or with EPERM/EACCES when it is not (the privilege check fires
* first). In either case, if the subvolume is read-only, clear the RDONLY flag (only
* inode ownership is required) and retry. This lets the retry succeed when removal is
* permitted, and otherwise leaves the subvolume writable so the caller (e.g. rm_rf())
* can empty and rmdir() it without privileges. */
if (made_writable || !ERRNO_IS_FS_WRITE_REFUSED(destroy_errno))
break;
r = btrfs_subvol_get_read_only_fd(subvol_fd);
if (r < 0)
return r;
if (r == 0)
break; /* Not read-only; clearing the flag would not help. */
r = btrfs_subvol_set_read_only_fd(subvol_fd, false);
if (r < 0)
return r;
made_writable = true;
}
if (!(flags & BTRFS_REMOVE_RECURSIVE) || destroy_errno != ENOTEMPTY)
return -destroy_errno;
/* OK, there are nested subvolumes — enumerate and recurse into them, then retry.
* BTRFS_IOC_GET_SUBVOL_ROOTREF and BTRFS_IOC_INO_LOOKUP_USER (kernel 4.18+) enumerate child
* subvolumes without requiring CAP_SYS_ADMIN in the initial user namespace, unlike the older
* BTRFS_IOC_TREE_SEARCH ioctl. */
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(subvol_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++) {
uint64_t child_subvol_id = rootref_args.rootref[i].treeid;
struct btrfs_ioctl_ino_lookup_user_args lookup_args = {
.dirid = rootref_args.rootref[i].dirid,
.treeid = child_subvol_id,
};
if (ioctl(subvol_fd, BTRFS_IOC_INO_LOOKUP_USER, &lookup_args) < 0)
return -errno;
if (!made_writable) {
r = btrfs_subvol_set_read_only_fd(subvol_fd, false);
if (r < 0)
return r;
made_writable = true;
}
if (isempty(lookup_args.path))
/* Subvolume is in the top-level directory of the subvolume. */
r = subvol_remove_children(subvol_fd, lookup_args.name, child_subvol_id, flags);
else {
_cleanup_close_ int child_fd = -EBADF;