+/* SPDX-License-Identifier: LGPL-2.1+ */
/***
This file is part of systemd.
#include "string-util.h"
#include "strv.h"
+/* This is the original MAX_HANDLE_SZ definition from the kernel, when the API was introduced. We use that in place of
+ * any more currently defined value to future-proof things: if the size is increased in the API headers, and our code
+ * is recompiled then it would cease working on old kernels, as those refuse any sizes larger than this value with
+ * EINVAL right-away. Hence, let's disconnect ourselves from any such API changes, and stick to the original definition
+ * from when it was introduced. We use it as a start value only anyway (see below), and hence should be able to deal
+ * with large file handles anyway. */
+#define ORIGINAL_MAX_HANDLE_SZ 128
+
+int name_to_handle_at_loop(
+ int fd,
+ const char *path,
+ struct file_handle **ret_handle,
+ int *ret_mnt_id,
+ int flags) {
+
+ _cleanup_free_ struct file_handle *h;
+ size_t n = ORIGINAL_MAX_HANDLE_SZ;
+
+ /* We need to invoke name_to_handle_at() in a loop, given that it might return EOVERFLOW when the specified
+ * buffer is too small. Note that in contrast to what the docs might suggest, MAX_HANDLE_SZ is only good as a
+ * start value, it is not an upper bound on the buffer size required.
+ *
+ * This improves on raw name_to_handle_at() also in one other regard: ret_handle and ret_mnt_id can be passed
+ * as NULL if there's no interest in either. */
+
+ h = malloc0(offsetof(struct file_handle, f_handle) + n);
+ if (!h)
+ return -ENOMEM;
+
+ h->handle_bytes = n;
+
+ for (;;) {
+ int mnt_id = -1;
+
+ if (name_to_handle_at(fd, path, h, &mnt_id, flags) >= 0) {
+
+ if (ret_handle) {
+ *ret_handle = h;
+ h = NULL;
+ }
+
+ if (ret_mnt_id)
+ *ret_mnt_id = mnt_id;
+
+ return 0;
+ }
+ if (errno != EOVERFLOW)
+ return -errno;
+
+ if (!ret_handle && ret_mnt_id && mnt_id >= 0) {
+
+ /* As it appears, name_to_handle_at() fills in mnt_id even when it returns EOVERFLOW when the
+ * buffer is too small, but that's undocumented. Hence, let's make use of this if it appears to
+ * be filled in, and the caller was interested in only the mount ID an nothing else. */
+
+ *ret_mnt_id = mnt_id;
+ return 0;
+ }
+
+ /* If name_to_handle_at() didn't increase the byte size, then this EOVERFLOW is caused by something
+ * else (apparently EOVERFLOW is returned for untriggered nfs4 mounts sometimes), not by the too small
+ * buffer. In that case propagate EOVERFLOW */
+ if (h->handle_bytes <= n)
+ return -EOVERFLOW;
+
+ /* The buffer was too small. Size the new buffer by what name_to_handle_at() returned. */
+ n = h->handle_bytes;
+ if (offsetof(struct file_handle, f_handle) + n < n) /* check for addition overflow */
+ return -EOVERFLOW;
+
+ free(h);
+ h = malloc0(offsetof(struct file_handle, f_handle) + n);
+ if (!h)
+ return -ENOMEM;
+
+ h->handle_bytes = n;
+ }
+}
+
static int fd_fdinfo_mnt_id(int fd, const char *filename, int flags, int *mnt_id) {
char path[strlen("/proc/self/fdinfo/") + DECIMAL_STR_MAX(int)];
_cleanup_free_ char *fdinfo = NULL;
if (r == -ENOENT) /* The fdinfo directory is a relatively new addition */
return -EOPNOTSUPP;
if (r < 0)
- return -errno;
+ return r;
p = startswith(fdinfo, "mnt_id:");
if (!p) {
}
int fd_is_mount_point(int fd, const char *filename, int flags) {
- union file_handle_union h = FILE_HANDLE_INIT, h_parent = FILE_HANDLE_INIT;
+ _cleanup_free_ struct file_handle *h = NULL, *h_parent = NULL;
int mount_id = -1, mount_id_parent = -1;
bool nosupp = false, check_st_dev = true;
struct stat a, b;
* subvolumes have different st_dev, even though they aren't
* real mounts of their own. */
- r = name_to_handle_at(fd, filename, &h.handle, &mount_id, flags);
- if (r < 0) {
- if (IN_SET(errno, ENOSYS, EACCES, EPERM))
- /* This kernel does not support name_to_handle_at() at all, or the syscall was blocked (maybe
- * through seccomp, because we are running inside of a container?): fall back to simpler
- * logic. */
+ r = name_to_handle_at_loop(fd, filename, &h, &mount_id, flags);
+ if (IN_SET(r, -ENOSYS, -EACCES, -EPERM, -EOVERFLOW, -EINVAL))
+ /* This kernel does not support name_to_handle_at() at all (ENOSYS), or the syscall was blocked
+ * (EACCES/EPERM; maybe through seccomp, because we are running inside of a container?), or the mount
+ * point is not triggered yet (EOVERFLOW, think nfs4), or some general name_to_handle_at() flakiness
+ * (EINVAL): fall back to simpler logic. */
+ goto fallback_fdinfo;
+ else if (r == -EOPNOTSUPP)
+ /* This kernel or file system does not support name_to_handle_at(), hence let's see if the upper fs
+ * supports it (in which case it is a mount point), otherwise fallback to the traditional stat()
+ * logic */
+ nosupp = true;
+ else if (r < 0)
+ return r;
+
+ r = name_to_handle_at_loop(fd, "", &h_parent, &mount_id_parent, AT_EMPTY_PATH);
+ if (r == -EOPNOTSUPP) {
+ if (nosupp)
+ /* Neither parent nor child do name_to_handle_at()? We have no choice but to fall back. */
goto fallback_fdinfo;
- else if (errno == EOPNOTSUPP)
- /* This kernel or file system does not support
- * name_to_handle_at(), hence let's see if the
- * upper fs supports it (in which case it is a
- * mount point), otherwise fallback to the
- * traditional stat() logic */
- nosupp = true;
else
- return -errno;
- }
-
- r = name_to_handle_at(fd, "", &h_parent.handle, &mount_id_parent, AT_EMPTY_PATH);
- if (r < 0) {
- if (errno == EOPNOTSUPP) {
- if (nosupp)
- /* Neither parent nor child do name_to_handle_at()?
- We have no choice but to fall back. */
- goto fallback_fdinfo;
- else
- /* The parent can't do name_to_handle_at() but the
- * directory we are interested in can?
- * If so, it must be a mount point. */
- return 1;
- } else
- return -errno;
- }
+ /* The parent can't do name_to_handle_at() but the directory we are interested in can? If so,
+ * it must be a mount point. */
+ return 1;
+ } else if (r < 0)
+ return r;
/* The parent can do name_to_handle_at() but the
* directory we are interested in can't? If so, it
* assume this is the root directory, which is a mount
* point. */
- if (h.handle.handle_bytes == h_parent.handle.handle_bytes &&
- h.handle.handle_type == h_parent.handle.handle_type &&
- memcmp(h.handle.f_handle, h_parent.handle.f_handle, h.handle.handle_bytes) == 0)
+ if (h->handle_bytes == h_parent->handle_bytes &&
+ h->handle_type == h_parent->handle_type &&
+ memcmp(h->f_handle, h_parent->f_handle, h->handle_bytes) == 0)
return 1;
return mount_id != mount_id_parent;
int r;
assert(t);
+ assert((flags & ~AT_SYMLINK_FOLLOW) == 0);
if (path_equal(t, "/"))
return 1;
if (fd < 0)
return -errno;
- return fd_is_mount_point(fd, basename(t), flags);
+ return fd_is_mount_point(fd, last_path_component(t), flags);
+}
+
+int path_get_mnt_id(const char *path, int *ret) {
+ int r;
+
+ r = name_to_handle_at_loop(AT_FDCWD, path, NULL, ret, 0);
+ if (IN_SET(r, -EOPNOTSUPP, -ENOSYS, -EACCES, -EPERM, -EOVERFLOW, -EINVAL)) /* kernel/fs don't support this, or seccomp blocks access, or untriggered mount, or name_to_handle_at() is flaky */
+ return fd_fdinfo_mnt_id(AT_FDCWD, path, 0, ret);
+
+ return r;
}
int umount_recursive(const char *prefix, int flags) {
while ((x = set_steal_first(todo))) {
r = set_consume(done, x);
- if (r == -EEXIST || r == 0)
+ if (IN_SET(r, 0, -EEXIST))
continue;
if (r < 0)
return r;
/* Deal with mount points that are obstructed by a later mount */
r = path_is_mount_point(x, NULL, 0);
- if (r == -ENOENT || r == 0)
+ if (IN_SET(r, 0, -ENOENT))
continue;
if (r < 0)
return r;
}
bool fstype_is_network(const char *fstype) {
- static const char table[] =
- "afs\0"
- "cifs\0"
- "smbfs\0"
- "sshfs\0"
- "ncpfs\0"
- "ncp\0"
- "nfs\0"
- "nfs4\0"
- "gfs\0"
- "gfs2\0"
- "glusterfs\0"
- "pvfs2\0" /* OrangeFS */
- "ocfs2\0"
- "lustre\0"
- ;
-
const char *x;
x = startswith(fstype, "fuse.");
if (x)
fstype = x;
- return nulstr_contains(table, fstype);
+ return STR_IN_SET(fstype,
+ "afs",
+ "cifs",
+ "smbfs",
+ "sshfs",
+ "ncpfs",
+ "ncp",
+ "nfs",
+ "nfs4",
+ "gfs",
+ "gfs2",
+ "glusterfs",
+ "pvfs2", /* OrangeFS */
+ "ocfs2",
+ "lustre");
+}
+
+bool fstype_is_api_vfs(const char *fstype) {
+ return STR_IN_SET(fstype,
+ "autofs",
+ "bpf",
+ "cgroup",
+ "cgroup2",
+ "configfs",
+ "cpuset",
+ "debugfs",
+ "devpts",
+ "devtmpfs",
+ "efivarfs",
+ "fusectl",
+ "hugetlbfs",
+ "mqueue",
+ "proc",
+ "pstore",
+ "ramfs",
+ "securityfs",
+ "sysfs",
+ "tmpfs",
+ "tracefs");
+}
+
+bool fstype_is_ro(const char *fstype) {
+ /* All Linux file systems that are necessarily read-only */
+ return STR_IN_SET(fstype,
+ "DM_verity_hash",
+ "iso9660",
+ "squashfs");
+}
+
+bool fstype_can_discard(const char *fstype) {
+ return STR_IN_SET(fstype,
+ "btrfs",
+ "ext4",
+ "vfat",
+ "xfs");
}
int repeat_unmount(const char *path, int flags) {