]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/shared/mount-util.h
Merge pull request #25509 from enr0n/oomd-unpriv-container-fix
[thirdparty/systemd.git] / src / shared / mount-util.h
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2 #pragma once
3
4 #include <mntent.h>
5 #include <stdio.h>
6 #include <sys/stat.h>
7 #include <unistd.h>
8
9 #include "alloc-util.h"
10 #include "dissect-image.h"
11 #include "errno-util.h"
12 #include "macro.h"
13
14 /* The limit used for /dev itself. 4MB should be enough since device nodes and symlinks don't
15 * consume any space and udev isn't supposed to create regular file either. There's no limit on the
16 * max number of inodes since such limit is hard to guess especially on large storage array
17 * systems. */
18 #define TMPFS_LIMITS_DEV ",size=4m"
19
20 /* The limit used for /dev in private namespaces. 4MB for contents of regular files. The number of
21 * inodes should be relatively low in private namespaces but for now use a 64k limit. */
22 #define TMPFS_LIMITS_PRIVATE_DEV ",size=4m,nr_inodes=64k"
23
24 /* Very little, if any use expected */
25 #define TMPFS_LIMITS_EMPTY_OR_ALMOST ",size=4m,nr_inodes=1k"
26 #define TMPFS_LIMITS_SYS TMPFS_LIMITS_EMPTY_OR_ALMOST
27 #define TMPFS_LIMITS_SYS_FS_CGROUP TMPFS_LIMITS_EMPTY_OR_ALMOST
28
29 /* On an extremely small device with only 256MB of RAM, 20% of RAM should be enough for the re-execution of
30 * PID1 because 16MB of free space is required. */
31 #define TMPFS_LIMITS_RUN ",size=20%,nr_inodes=800k"
32
33 /* The limit used for various nested tmpfs mounts, in particular for guests started by systemd-nspawn.
34 * 10% of RAM (using 16GB of RAM as a baseline) translates to 400k inodes (assuming 4k each) and 25%
35 * translates to 1M inodes.
36 * (On the host, /tmp is configured through a .mount unit file.) */
37 #define NESTED_TMPFS_LIMITS ",size=10%,nr_inodes=400k"
38
39 /* More space for volatile root and /var */
40 #define TMPFS_LIMITS_VAR ",size=25%,nr_inodes=1m"
41 #define TMPFS_LIMITS_ROOTFS TMPFS_LIMITS_VAR
42 #define TMPFS_LIMITS_VOLATILE_STATE TMPFS_LIMITS_VAR
43
44 int mount_fd(const char *source, int target_fd, const char *filesystemtype, unsigned long mountflags, const void *data);
45 int mount_nofollow(const char *source, const char *target, const char *filesystemtype, unsigned long mountflags, const void *data);
46
47 int repeat_unmount(const char *path, int flags);
48 int umount_recursive(const char *target, int flags);
49
50 int bind_remount_recursive_with_mountinfo(const char *prefix, unsigned long new_flags, unsigned long flags_mask, char **deny_list, FILE *proc_self_mountinfo);
51 static inline int bind_remount_recursive(const char *prefix, unsigned long new_flags, unsigned long flags_mask, char **deny_list) {
52 return bind_remount_recursive_with_mountinfo(prefix, new_flags, flags_mask, deny_list, NULL);
53 }
54
55 int bind_remount_one_with_mountinfo(const char *path, unsigned long new_flags, unsigned long flags_mask, FILE *proc_self_mountinfo);
56
57 int mount_move_root(const char *path);
58 int mount_pivot_root(const char *path);
59
60 DEFINE_TRIVIAL_CLEANUP_FUNC_FULL(FILE*, endmntent, NULL);
61 #define _cleanup_endmntent_ _cleanup_(endmntentp)
62
63 int mount_verbose_full(
64 int error_log_level,
65 const char *what,
66 const char *where,
67 const char *type,
68 unsigned long flags,
69 const char *options,
70 bool follow_symlink);
71
72 static inline int mount_follow_verbose(
73 int error_log_level,
74 const char *what,
75 const char *where,
76 const char *type,
77 unsigned long flags,
78 const char *options) {
79 return mount_verbose_full(error_log_level, what, where, type, flags, options, true);
80 }
81
82 static inline int mount_nofollow_verbose(
83 int error_log_level,
84 const char *what,
85 const char *where,
86 const char *type,
87 unsigned long flags,
88 const char *options) {
89 return mount_verbose_full(error_log_level, what, where, type, flags, options, false);
90 }
91
92 int umount_verbose(
93 int error_log_level,
94 const char *where,
95 int flags);
96
97 int mount_option_mangle(
98 const char *options,
99 unsigned long mount_flags,
100 unsigned long *ret_mount_flags,
101 char **ret_remaining_options);
102
103 int mode_to_inaccessible_node(const char *runtime_dir, mode_t mode, char **dest);
104 int mount_flags_to_string(unsigned long flags, char **ret);
105
106 /* Useful for usage with _cleanup_(), unmounts, removes a directory and frees the pointer */
107 static inline char* umount_and_rmdir_and_free(char *p) {
108 PROTECT_ERRNO;
109 if (p) {
110 (void) umount_recursive(p, 0);
111 (void) rmdir(p);
112 }
113 return mfree(p);
114 }
115 DEFINE_TRIVIAL_CLEANUP_FUNC(char*, umount_and_rmdir_and_free);
116
117 int bind_mount_in_namespace(pid_t target, const char *propagate_path, const char *incoming_path, const char *src, const char *dest, bool read_only, bool make_file_or_directory);
118 int mount_image_in_namespace(pid_t target, const char *propagate_path, const char *incoming_path, const char *src, const char *dest, bool read_only, bool make_file_or_directory, const MountOptions *options);
119
120 int make_mount_point(const char *path);
121
122 typedef enum RemountIdmapping {
123 REMOUNT_IDMAPPING_NONE,
124 /* Include a mapping from UID_MAPPED_ROOT (i.e. UID 2^31-2) on the backing fs to UID 0 on the
125 * uidmapped fs. This is useful to ensure that the host root user can safely add inodes to the
126 * uidmapped fs (which otherwise wouldn't work as the host root user is not defined on the uidmapped
127 * mount and any attempts to create inodes will then be refused with EOVERFLOW). The idea is that
128 * these inodes are quickly re-chown()ed to more suitable UIDs/GIDs. Any code that intends to be able
129 * to add inodes to file systems mapped this way should set this flag, but given it comes with
130 * certain security implications defaults to off, and requires explicit opt-in. */
131 REMOUNT_IDMAPPING_HOST_ROOT,
132 /* Define a mapping from root user within the container to the owner of the bind mounted directory.
133 * This ensure no root-owned files will be written in a bind-mounted directory owned by a different
134 * user. No other users are mapped. */
135 REMOUNT_IDMAPPING_HOST_OWNER,
136 _REMOUNT_IDMAPPING_MAX,
137 _REMOUNT_IDMAPPING_INVALID = -EINVAL,
138 } RemountIdmapping;
139
140 int remount_idmap(const char *p, uid_t uid_shift, uid_t uid_range, uid_t owner, RemountIdmapping idmapping);
141
142 /* Creates a mount point (not parents) based on the source path or stat - ie, a file or a directory */
143 int make_mount_point_inode_from_stat(const struct stat *st, const char *dest, mode_t mode);
144 int make_mount_point_inode_from_path(const char *source, const char *dest, mode_t mode);