Merge pull request #25723 from keszybz/generators-tmp

Run generators with / ro and /tmp mounted
This commit is contained in:
Yu Watanabe
2022-12-15 12:53:49 +09:00
committed by GitHub
35 changed files with 341 additions and 289 deletions
+1 -1
View File
@@ -687,7 +687,7 @@ static int create_socket(char **ret) {
return r;
sa_len = r;
RUN_WITH_UMASK(0177)
WITH_UMASK(0177)
if (bind(fd, &sa.sa, sa_len) < 0)
return -errno;
+1 -1
View File
@@ -178,7 +178,7 @@ int base_filesystem_create(const char *root, uid_t uid, gid_t gid) {
continue;
}
RUN_WITH_UMASK(0000)
WITH_UMASK(0000)
r = mkdirat(fd, table[i].dir, table[i].mode);
if (r < 0) {
log_full_errno(IN_SET(errno, EEXIST, EROFS) || table[i].ignore_failure ? LOG_DEBUG : LOG_ERR, errno,
+1 -1
View File
@@ -1357,7 +1357,7 @@ int copy_file_full(
if (r < 0)
return r;
RUN_WITH_UMASK(0000) {
WITH_UMASK(0000) {
if (copy_flags & COPY_MAC_CREATE) {
r = mac_selinux_create_file_prepare(to, S_IFREG);
if (r < 0)
+2 -2
View File
@@ -89,7 +89,7 @@ int machine_id_setup(const char *root, bool force_transient, sd_id128_t machine_
etc_machine_id = prefix_roota(root, "/etc/machine-id");
RUN_WITH_UMASK(0000) {
WITH_UMASK(0000) {
/* We create this 0444, to indicate that this isn't really
* something you should ever modify. Of course, since the file
* will be owned by root it doesn't matter much, but maybe
@@ -166,7 +166,7 @@ int machine_id_setup(const char *root, bool force_transient, sd_id128_t machine_
run_machine_id = prefix_roota(root, "/run/machine-id");
RUN_WITH_UMASK(0022)
WITH_UMASK(0022)
r = id128_write(run_machine_id, ID128_FORMAT_PLAIN, machine_id);
if (r < 0) {
(void) unlink(run_machine_id);
+24 -22
View File
@@ -62,55 +62,55 @@ typedef struct MountPoint {
#endif
static const MountPoint mount_table[] = {
{ "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
NULL, MNT_FATAL|MNT_IN_CONTAINER|MNT_FOLLOW_SYMLINK },
{ "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
NULL, MNT_FATAL|MNT_IN_CONTAINER },
{ "devtmpfs", "/dev", "devtmpfs", "mode=755" TMPFS_LIMITS_DEV, MS_NOSUID|MS_STRICTATIME,
{ "devtmpfs", "/dev", "devtmpfs", "mode=0755" TMPFS_LIMITS_DEV, MS_NOSUID|MS_STRICTATIME,
NULL, MNT_FATAL|MNT_IN_CONTAINER },
{ "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
NULL, MNT_NONE },
#if ENABLE_SMACK
{ "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV,
mac_smack_use, MNT_FATAL },
{ "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
{ "tmpfs", "/dev/shm", "tmpfs", "mode=01777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
mac_smack_use, MNT_FATAL },
#endif
{ "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
{ "tmpfs", "/dev/shm", "tmpfs", "mode=01777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
NULL, MNT_FATAL|MNT_IN_CONTAINER },
{ "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
{ "devpts", "/dev/pts", "devpts", "mode=0620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
NULL, MNT_IN_CONTAINER },
#if ENABLE_SMACK
{ "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
{ "tmpfs", "/run", "tmpfs", "mode=0755,smackfsroot=*" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
mac_smack_use, MNT_FATAL },
#endif
{ "tmpfs", "/run", "tmpfs", "mode=755" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
{ "tmpfs", "/run", "tmpfs", "mode=0755" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
NULL, MNT_FATAL|MNT_IN_CONTAINER },
{ "cgroup2", "/sys/fs/cgroup", "cgroup2", "nsdelegate,memory_recursiveprot", MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "cgroup2", "/sys/fs/cgroup", "cgroup2", "nsdelegate,memory_recursiveprot", MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
{ "cgroup2", "/sys/fs/cgroup", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "cgroup2", "/sys/fs/cgroup", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
{ "cgroup2", "/sys/fs/cgroup", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "cgroup2", "/sys/fs/cgroup", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
{ "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755" TMPFS_LIMITS_SYS_FS_CGROUP, MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
{ "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=0755" TMPFS_LIMITS_SYS_FS_CGROUP, MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
{ "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
{ "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
{ "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_legacy_wanted, MNT_IN_CONTAINER },
{ "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
#if ENABLE_PSTORE
{ "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
NULL, MNT_NONE },
#endif
#if ENABLE_EFI
{ "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
is_efi_boot, MNT_NONE },
#endif
{ "bpf", "/sys/fs/bpf", "bpf", "mode=700", MS_NOSUID|MS_NOEXEC|MS_NODEV,
{ "bpf", "/sys/fs/bpf", "bpf", "mode=0700", MS_NOSUID|MS_NOEXEC|MS_NODEV,
NULL, MNT_NONE, },
};
@@ -356,7 +356,9 @@ int mount_cgroup_controllers(void) {
}
/* Now that we mounted everything, let's make the tmpfs the cgroup file systems are mounted into read-only. */
(void) mount_nofollow("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755" TMPFS_LIMITS_SYS_FS_CGROUP);
(void) mount_nofollow("tmpfs", "/sys/fs/cgroup", "tmpfs",
MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY,
"mode=0755" TMPFS_LIMITS_SYS_FS_CGROUP);
return 0;
}
+2 -48
View File
@@ -42,52 +42,6 @@
#include "tmpfile-util.h"
#include "user-util.h"
int mount_fd(const char *source,
int target_fd,
const char *filesystemtype,
unsigned long mountflags,
const void *data) {
if (mount(source, FORMAT_PROC_FD_PATH(target_fd), filesystemtype, mountflags, data) < 0) {
if (errno != ENOENT)
return -errno;
/* ENOENT can mean two things: either that the source is missing, or that /proc/ isn't
* mounted. Check for the latter to generate better error messages. */
if (proc_mounted() == 0)
return -ENOSYS;
return -ENOENT;
}
return 0;
}
int mount_nofollow(
const char *source,
const char *target,
const char *filesystemtype,
unsigned long mountflags,
const void *data) {
_cleanup_close_ int fd = -1;
/* In almost all cases we want to manipulate the mount table without following symlinks, hence
* mount_nofollow() is usually the way to go. The only exceptions are environments where /proc/ is
* not available yet, since we need /proc/self/fd/ for this logic to work. i.e. during the early
* initialization of namespacing/container stuff where /proc is not yet mounted (and maybe even the
* fs to mount) we can only use traditional mount() directly.
*
* Note that this disables following only for the final component of the target, i.e symlinks within
* the path of the target are honoured, as are symlinks in the source path everywhere. */
fd = open(target, O_PATH|O_CLOEXEC|O_NOFOLLOW);
if (fd < 0)
return -errno;
return mount_fd(source, fd, filesystemtype, mountflags, data);
}
int umount_recursive(const char *prefix, int flags) {
int n = 0, r;
bool again;
@@ -809,9 +763,9 @@ int mount_option_mangle(
/* This extracts mount flags from the mount options, and stores
* non-mount-flag options to '*ret_remaining_options'.
* E.g.,
* "rw,nosuid,nodev,relatime,size=1630748k,mode=700,uid=1000,gid=1000"
* "rw,nosuid,nodev,relatime,size=1630748k,mode=0700,uid=1000,gid=1000"
* is split to MS_NOSUID|MS_NODEV|MS_RELATIME and
* "size=1630748k,mode=700,uid=1000,gid=1000".
* "size=1630748k,mode=0700,uid=1000,gid=1000".
* See more examples in test-mount-util.c.
*
* If 'options' does not contain any non-mount-flag options,
-33
View File
@@ -25,39 +25,6 @@ const char* mount_attr_propagation_type_to_string(MountAttrPropagationType t) _c
MountAttrPropagationType mount_attr_propagation_type_from_string(const char *s) _pure_;
unsigned int mount_attr_propagation_type_to_flag(MountAttrPropagationType t);
/* The limit used for /dev itself. 4MB should be enough since device nodes and symlinks don't
* consume any space and udev isn't supposed to create regular file either. There's no limit on the
* max number of inodes since such limit is hard to guess especially on large storage array
* systems. */
#define TMPFS_LIMITS_DEV ",size=4m"
/* The limit used for /dev in private namespaces. 4MB for contents of regular files. The number of
* inodes should be relatively low in private namespaces but for now use a 64k limit. */
#define TMPFS_LIMITS_PRIVATE_DEV ",size=4m,nr_inodes=64k"
/* Very little, if any use expected */
#define TMPFS_LIMITS_EMPTY_OR_ALMOST ",size=4m,nr_inodes=1k"
#define TMPFS_LIMITS_SYS TMPFS_LIMITS_EMPTY_OR_ALMOST
#define TMPFS_LIMITS_SYS_FS_CGROUP TMPFS_LIMITS_EMPTY_OR_ALMOST
/* On an extremely small device with only 256MB of RAM, 20% of RAM should be enough for the re-execution of
* PID1 because 16MB of free space is required. */
#define TMPFS_LIMITS_RUN ",size=20%,nr_inodes=800k"
/* The limit used for various nested tmpfs mounts, in particular for guests started by systemd-nspawn.
* 10% of RAM (using 16GB of RAM as a baseline) translates to 400k inodes (assuming 4k each) and 25%
* translates to 1M inodes.
* (On the host, /tmp is configured through a .mount unit file.) */
#define NESTED_TMPFS_LIMITS ",size=10%,nr_inodes=400k"
/* More space for volatile root and /var */
#define TMPFS_LIMITS_VAR ",size=25%,nr_inodes=1m"
#define TMPFS_LIMITS_ROOTFS TMPFS_LIMITS_VAR
#define TMPFS_LIMITS_VOLATILE_STATE TMPFS_LIMITS_VAR
int mount_fd(const char *source, int target_fd, const char *filesystemtype, unsigned long mountflags, const void *data);
int mount_nofollow(const char *source, const char *target, const char *filesystemtype, unsigned long mountflags, const void *data);
int repeat_unmount(const char *path, int flags);
int umount_recursive(const char *target, int flags);
+1 -1
View File
@@ -42,7 +42,7 @@ int update_reboot_parameter_and_warn(const char *parameter, bool keep) {
return 0;
}
RUN_WITH_UMASK(0022) {
WITH_UMASK(0022) {
r = write_string_file("/run/systemd/reboot-param", parameter,
WRITE_STRING_FILE_CREATE|WRITE_STRING_FILE_ATOMIC);
if (r < 0)
+1 -1
View File
@@ -101,7 +101,7 @@ int socket_address_listen(
(void) mkdir_parents_label(p, directory_mode);
/* Enforce the right access mode for the socket */
RUN_WITH_UMASK(~socket_mode) {
WITH_UMASK(~socket_mode) {
r = mac_selinux_bind(fd, &a->sockaddr.sa, a->size);
if (r == -EADDRINUSE) {
/* Unlink and try again */
+1 -1
View File
@@ -2326,7 +2326,7 @@ int varlink_server_listen_address(VarlinkServer *s, const char *address, mode_t
(void) sockaddr_un_unlink(&sockaddr.un);
RUN_WITH_UMASK(~m & 0777) {
WITH_UMASK(~m & 0777) {
r = mac_selinux_bind(fd, &sockaddr.sa, sockaddr_len);
if (r < 0)
return r;