src/lxc/conf.h

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #ifndef __LXC_CONF_H
   4 #define __LXC_CONF_H
   5
   6 #ifndef _GNU_SOURCE
   7 #define _GNU_SOURCE 1
   8 #endif
   9 #include <linux/magic.h>
  10 #include <net/if.h>
  11 #include <netinet/in.h>
  12 #include <stdbool.h>
  13 #include <stdio.h>
  14 #include <sys/param.h>
  15 #include <sys/types.h>
  16 #include <sys/vfs.h>
  17
  18 #include "compiler.h"
  19 #include "config.h"
  20 #include "list.h"
  21 #include "lxcseccomp.h"
  22 #include "ringbuf.h"
  23 #include "start.h"
  24 #include "terminal.h"
  25
  26 #if HAVE_SYS_RESOURCE_H
  27 #include <sys/resource.h>
  28 #endif
  29
  30 #if HAVE_SCMP_FILTER_CTX
  31 typedef void * scmp_filter_ctx;
  32 #endif
  33
  34 /* worth moving to configure.ac? */
  35 #define subuidfile "/etc/subuid"
  36 #define subgidfile "/etc/subgid"
  37
  38 /*
  39  * Defines a generic struct to configure the control group. It is up to the
  40  * programmer to specify the right subsystem.
  41  * @subsystem : the targeted subsystem
  42  * @value     : the value to set
  43  * @version   : The version of the cgroup filesystem on which the controller
  44  *              resides.
  45  *
  46  * @controllers : The controllers to use for this container.
  47  * @dir         : The name of the directory containing the container's cgroup.
  48  *                Not that this is a per-container setting.
  49  */
  50 struct lxc_cgroup {
  51         union {
  52                 /* information about a specific controller */
  53                 struct /* controller */ {
  54                         int version;
  55                         char *subsystem;
  56                         char *value;
  57                 };
  58
  59                 /* meta information about cgroup configuration */
  60                 struct /* meta */ {
  61                         char *controllers;
  62                         char *dir;
  63                         char *monitor_dir;
  64                         char *container_dir;
  65                         char *namespace_dir;
  66                         bool relative;
  67                 };
  68         };
  69 };
  70
  71 #if !HAVE_SYS_RESOURCE_H
  72 #define RLIM_INFINITY ((unsigned long)-1)
  73 struct rlimit {
  74         unsigned long rlim_cur;
  75         unsigned long rlim_max;
  76 };
  77 #endif
  78
  79 /*
  80  * Defines a structure to configure resource limits to set via setrlimit().
  81  * @resource : the resource name in lowercase without the RLIMIT_ prefix
  82  * @limit    : the limit to set
  83  */
  84 struct lxc_limit {
  85         char *resource;
  86         struct rlimit limit;
  87 };
  88
  89 enum idtype {
  90         ID_TYPE_UID,
  91         ID_TYPE_GID
  92 };
  93
  94 /*
  95  * Defines a structure to configure kernel parameters at runtime.
  96  * @key      : the kernel parameters will be configured without the "lxc.sysctl" prefix
  97  * @value    : the value to set
  98  */
  99 struct lxc_sysctl {
 100         char *key;
 101         char *value;
 102 };
 103
 104 /*
 105  * Defines a structure to configure proc filesystem at runtime.
 106  * @filename : the proc filesystem will be configured without the "lxc.proc" prefix
 107  * @value    : the value to set
 108  */
 109 struct lxc_proc {
 110         char *filename;
 111         char *value;
 112 };
 113
 114 /*
 115  * id_map is an id map entry.  Form in confile is:
 116  * lxc.idmap = u 0    9800 100
 117  * lxc.idmap = u 1000 9900 100
 118  * lxc.idmap = g 0    9800 100
 119  * lxc.idmap = g 1000 9900 100
 120  * meaning the container can use uids and gids 0-99 and 1000-1099,
 121  * with [ug]id 0 mapping to [ug]id 9800 on the host, and [ug]id 1000 to
 122  * [ug]id 9900 on the host.
 123  */
 124 struct id_map {
 125         enum idtype idtype;
 126         unsigned long hostid, nsid, range;
 127 };
 128
 129 /* Defines the number of tty configured and contains the
 130  * instantiated ptys
 131  * @max = number of configured ttys
 132  */
 133 struct lxc_tty_info {
 134         size_t max;
 135         char *dir;
 136         char *tty_names;
 137         struct lxc_terminal_info *tty;
 138 };
 139
 140 /* Defines a structure to store the rootfs location, the
 141  * optionals pivot_root, rootfs mount paths
 142  * @path       : the rootfs source (directory or device)
 143  * @mount      : where it is mounted
 144  * @bev_type   : optional backing store type
 145  * @options    : mount options
 146  * @mountflags : the portion of @options that are flags
 147  * @data       : the portion of @options that are not flags
 148  * @managed    : whether it is managed by LXC
 149  */
 150 struct lxc_rootfs {
 151         char *path;
 152         char *mount;
 153         char *bdev_type;
 154         char *options;
 155         unsigned long mountflags;
 156         char *data;
 157         bool managed;
 158 };
 159
 160 /*
 161  * Automatic mounts for LXC to perform inside the container
 162  */
 163 enum {
 164         LXC_AUTO_PROC_RW              = 0x001, /* /proc read-write */
 165         LXC_AUTO_PROC_MIXED           = 0x002, /* /proc/sys and /proc/sysrq-trigger read-only */
 166         LXC_AUTO_PROC_MASK            = 0x003,
 167
 168         LXC_AUTO_SYS_RW               = 0x004, /* /sys */
 169         LXC_AUTO_SYS_RO               = 0x008, /* /sys read-only */
 170         LXC_AUTO_SYS_MIXED            = 0x00C, /* /sys read-only and /sys/class/net read-write */
 171         LXC_AUTO_SYS_MASK             = 0x00C,
 172
 173         LXC_AUTO_CGROUP_RO            = 0x010, /* /sys/fs/cgroup (partial mount, read-only) */
 174         LXC_AUTO_CGROUP_RW            = 0x020, /* /sys/fs/cgroup (partial mount, read-write) */
 175         LXC_AUTO_CGROUP_MIXED         = 0x030, /* /sys/fs/cgroup (partial mount, paths r/o, cgroup r/w) */
 176         LXC_AUTO_CGROUP_FULL_RO       = 0x040, /* /sys/fs/cgroup (full mount, read-only) */
 177         LXC_AUTO_CGROUP_FULL_RW       = 0x050, /* /sys/fs/cgroup (full mount, read-write) */
 178         LXC_AUTO_CGROUP_FULL_MIXED    = 0x060, /* /sys/fs/cgroup (full mount, parent r/o, own r/w) */
 179         /*
 180          * These are defined in such a way as to retain binary compatibility
 181          * with earlier versions of this code. If the previous mask is applied,
 182          * both of these will default back to the _MIXED variants, which is
 183          * safe.
 184          */
 185         LXC_AUTO_CGROUP_NOSPEC        = 0x0B0, /* /sys/fs/cgroup (partial mount, r/w or mixed, depending on caps) */
 186         LXC_AUTO_CGROUP_FULL_NOSPEC   = 0x0E0, /* /sys/fs/cgroup (full mount, r/w or mixed, depending on caps) */
 187         LXC_AUTO_CGROUP_FORCE         = 0x100, /* mount cgroups even when cgroup namespaces are supported */
 188         LXC_AUTO_CGROUP_MASK          = 0x1F0, /* all known cgroup options, doe not contain LXC_AUTO_CGROUP_FORCE */
 189
 190         LXC_AUTO_SHMOUNTS             = 0x200, /* shared mount point */
 191         LXC_AUTO_SHMOUNTS_MASK        = 0x200, /* shared mount point mask */
 192         LXC_AUTO_ALL_MASK             = 0x1FF, /* all known settings */
 193 };
 194
 195 enum lxchooks {
 196         LXCHOOK_PRESTART,
 197         LXCHOOK_PREMOUNT,
 198         LXCHOOK_MOUNT,
 199         LXCHOOK_AUTODEV,
 200         LXCHOOK_START,
 201         LXCHOOK_STOP,
 202         LXCHOOK_POSTSTOP,
 203         LXCHOOK_CLONE,
 204         LXCHOOK_DESTROY,
 205         LXCHOOK_START_HOST,
 206         NUM_LXC_HOOKS
 207 };
 208
 209 extern char *lxchook_names[NUM_LXC_HOOKS];
 210
 211 struct lxc_state_client {
 212         int clientfd;
 213         lxc_state_t states[MAX_STATE];
 214 };
 215
 216 enum {
 217         LXC_BPF_DEVICE_CGROUP_LOCAL_RULE = -1,
 218         LXC_BPF_DEVICE_CGROUP_WHITELIST  =  0,
 219         LXC_BPF_DEVICE_CGROUP_BLACKLIST  =  1,
 220 };
 221
 222 struct device_item {
 223         char type;
 224         int major;
 225         int minor;
 226         char access[4];
 227         int allow;
 228         /*
 229          * LXC_BPF_DEVICE_CGROUP_LOCAL_RULE -> no global rule
 230          * LXC_BPF_DEVICE_CGROUP_WHITELIST  -> whitelist (deny all)
 231          * LXC_BPF_DEVICE_CGROUP_BLACKLIST  -> blacklist (allow all)
 232          */
 233         int global_rule;
 234 };
 235
 236 struct lxc_conf {
 237         /* Pointer to the name of the container. Do not free! */
 238         const char *name;
 239         bool is_execute;
 240         int reboot;
 241         signed long personality;
 242         struct utsname *utsname;
 243
 244         struct {
 245                 struct lxc_list cgroup;
 246                 struct lxc_list cgroup2;
 247                 struct bpf_program *cgroup2_devices;
 248                 /* This should be reimplemented as a hashmap. */
 249                 struct lxc_list devices;
 250         };
 251
 252         struct {
 253                 struct lxc_list id_map;
 254
 255                 /*
 256                  * Pointer to the idmap entry for the container's root uid in
 257                  * the id_map list. Do not free!
 258                  */
 259                 const struct id_map *root_nsuid_map;
 260
 261                 /*
 262                  * Pointer to the idmap entry for the container's root gid in
 263                  * the id_map list. Do not free!
 264                  */
 265                 const struct id_map *root_nsgid_map;
 266         };
 267
 268         struct lxc_list network;
 269
 270         struct {
 271                 char *fstab;
 272                 int auto_mounts;
 273                 struct lxc_list mount_list;
 274         };
 275
 276         struct lxc_list caps;
 277         struct lxc_list keepcaps;
 278
 279         /* /dev/tty<idx> devices */
 280         struct lxc_tty_info ttys;
 281         /* /dev/console device */
 282         struct lxc_terminal console;
 283         /* maximum pty devices allowed by devpts mount */
 284         size_t pty_max;
 285
 286         /* set to true when rootfs has been setup */
 287         bool rootfs_setup;
 288         struct lxc_rootfs rootfs;
 289
 290         bool close_all_fds;
 291
 292         struct {
 293                 unsigned int hooks_version;
 294                 struct lxc_list hooks[NUM_LXC_HOOKS];
 295         };
 296
 297         char *lsm_aa_profile;
 298         char *lsm_aa_profile_computed;
 299         bool lsm_aa_profile_created;
 300         unsigned int lsm_aa_allow_nesting;
 301         unsigned int lsm_aa_allow_incomplete;
 302         struct lxc_list lsm_aa_raw;
 303         char *lsm_se_context;
 304         char *lsm_se_keyring_context;
 305         bool keyring_disable_session;
 306         bool tmp_umount_proc;
 307         struct lxc_seccomp seccomp;
 308         int maincmd_fd;
 309         unsigned int autodev;  /* if 1, mount and fill a /dev at start */
 310         int autodevtmpfssize; /* size of the /dev tmpfs */
 311         int haltsignal; /* signal used to halt container */
 312         int rebootsignal; /* signal used to reboot container */
 313         int stopsignal; /* signal used to hard stop container */
 314         char *rcfile;   /* Copy of the top level rcfile we read */
 315
 316         /* Logfile and loglevel can be set in a container config file. Those
 317          * function as defaults. The defaults can be overridden by command line.
 318          * However we don't want the command line specified values to be saved
 319          * on c->save_config(). So we store the config file specified values
 320          * here. */
 321         char *logfile; /* the logfile as specified in config */
 322         int loglevel; /* loglevel as specified in config (if any) */
 323         int logfd;
 324
 325         unsigned int start_auto;
 326         unsigned int start_delay;
 327         int start_order;
 328         struct lxc_list groups;
 329         int nbd_idx;
 330
 331         /* unshare the mount namespace in the monitor */
 332         unsigned int monitor_unshare;
 333         unsigned int monitor_signal_pdeath;
 334
 335         /* list of included files */
 336         struct lxc_list includes;
 337         /* config entries which are not "lxc.*" are aliens */
 338         struct lxc_list aliens;
 339
 340         /* list of environment variables we'll add to the container when
 341          * started */
 342         struct lxc_list environment;
 343
 344         /* text representation of the config file */
 345         char *unexpanded_config;
 346         size_t unexpanded_len;
 347         size_t unexpanded_alloced;
 348
 349         /* default command for lxc-execute */
 350         char *execute_cmd;
 351
 352         /* init command */
 353         char *init_cmd;
 354
 355         /* if running in a new user namespace, the UID/GID that init and COMMAND
 356          * should run under when using lxc-execute */
 357         uid_t init_uid;
 358         gid_t init_gid;
 359
 360         /* indicator if the container will be destroyed on shutdown */
 361         unsigned int ephemeral;
 362
 363         /* The facility to pass to syslog. Let's users establish as what type of
 364          * program liblxc is supposed to write to the syslog. */
 365         char *syslog;
 366
 367         /* Whether PR_SET_NO_NEW_PRIVS will be set for the container. */
 368         bool no_new_privs;
 369
 370         /* RLIMIT_* limits */
 371         struct lxc_list limits;
 372
 373         /* Contains generic info about the cgroup configuration for this
 374          * container. Note that struct lxc_cgroup contains a union. It is only
 375          * valid to access the members of the anonymous "meta" struct within
 376          * that union.
 377          */
 378         struct lxc_cgroup cgroup_meta;
 379
 380         struct {
 381                 int ns_clone;
 382                 int ns_keep;
 383                 char *ns_share[LXC_NS_MAX];
 384         };
 385
 386         /* init working directory */
 387         char *init_cwd;
 388
 389         /* A list of clients registered to be informed about a container state. */
 390         struct lxc_list state_clients;
 391
 392         /* sysctls */
 393         struct lxc_list sysctls;
 394
 395         /* procs */
 396         struct lxc_list procs;
 397
 398         struct shmount {
 399                 /* Absolute path to the shared mount point on the host */
 400                 char *path_host;
 401                 /* Absolute path (in the container) to the shared mount point */
 402                 char *path_cont;
 403         } shmount;
 404 };
 405
 406 extern int write_id_mapping(enum idtype idtype, pid_t pid, const char *buf,
 407                             size_t buf_size)
 408 __access_r(3, 4);
 409
 410 #ifdef HAVE_TLS
 411 extern thread_local struct lxc_conf *current_config;
 412 #else
 413 extern struct lxc_conf *current_config;
 414 #endif
 415
 416 extern int run_lxc_hooks(const char *name, char *hook, struct lxc_conf *conf,
 417                          char *argv[]);
 418 extern int detect_shared_rootfs(void);
 419 extern struct lxc_conf *lxc_conf_init(void);
 420 extern void lxc_conf_free(struct lxc_conf *conf);
 421 extern int pin_rootfs(const char *rootfs);
 422 extern int lxc_map_ids(struct lxc_list *idmap, pid_t pid);
 423 extern int lxc_create_tty(const char *name, struct lxc_conf *conf);
 424 extern void lxc_delete_tty(struct lxc_tty_info *ttys);
 425 extern int lxc_clear_config_caps(struct lxc_conf *c);
 426 extern int lxc_clear_config_keepcaps(struct lxc_conf *c);
 427 extern int lxc_clear_cgroups(struct lxc_conf *c, const char *key, int version);
 428 extern int lxc_clear_mount_entries(struct lxc_conf *c);
 429 extern int lxc_clear_automounts(struct lxc_conf *c);
 430 extern int lxc_clear_hooks(struct lxc_conf *c, const char *key);
 431 extern int lxc_clear_idmaps(struct lxc_conf *c);
 432 extern int lxc_clear_groups(struct lxc_conf *c);
 433 extern int lxc_clear_environment(struct lxc_conf *c);
 434 extern int lxc_clear_limits(struct lxc_conf *c, const char *key);
 435 extern int lxc_delete_autodev(struct lxc_handler *handler);
 436 extern int lxc_clear_autodev_tmpfs_size(struct lxc_conf *c);
 437 extern void lxc_clear_includes(struct lxc_conf *conf);
 438 extern int lxc_setup_rootfs_prepare_root(struct lxc_conf *conf,
 439                                          const char *name, const char *lxcpath);
 440 extern int lxc_setup(struct lxc_handler *handler);
 441 extern int lxc_setup_parent(struct lxc_handler *handler);
 442 extern int setup_resource_limits(struct lxc_list *limits, pid_t pid);
 443 extern int find_unmapped_nsid(const struct lxc_conf *conf, enum idtype idtype);
 444 extern int mapped_hostid(unsigned id, const struct lxc_conf *conf,
 445                          enum idtype idtype);
 446 extern int userns_exec_1(const struct lxc_conf *conf, int (*fn)(void *),
 447                          void *data, const char *fn_name);
 448 extern int userns_exec_full(struct lxc_conf *conf, int (*fn)(void *),
 449                             void *data, const char *fn_name);
 450 extern int parse_mntopts(const char *mntopts, unsigned long *mntflags,
 451                          char **mntdata);
 452 extern int parse_propagationopts(const char *mntopts, unsigned long *pflags);
 453 extern void tmp_proc_unmount(struct lxc_conf *lxc_conf);
 454 extern void turn_into_dependent_mounts(void);
 455 extern void suggest_default_idmap(void);
 456 extern FILE *make_anonymous_mount_file(struct lxc_list *mount,
 457                                        bool include_nesting_helpers);
 458 extern struct lxc_list *sort_cgroup_settings(struct lxc_list *cgroup_settings);
 459 extern unsigned long add_required_remount_flags(const char *s, const char *d,
 460                                                 unsigned long flags);
 461 extern int run_script(const char *name, const char *section, const char *script,
 462                       ...);
 463 extern int run_script_argv(const char *name, unsigned int hook_version,
 464                            const char *section, const char *script,
 465                            const char *hookname, char **argsin);
 466 extern int in_caplist(int cap, struct lxc_list *caps);
 467 extern int setup_sysctl_parameters(struct lxc_list *sysctls);
 468 extern int lxc_clear_sysctls(struct lxc_conf *c, const char *key);
 469 extern int setup_proc_filesystem(struct lxc_list *procs, pid_t pid);
 470 extern int lxc_clear_procs(struct lxc_conf *c, const char *key);
 471 extern int lxc_clear_apparmor_raw(struct lxc_conf *c);
 472 extern int lxc_clear_namespace(struct lxc_conf *c);
 473 extern int userns_exec_minimal(const struct lxc_conf *conf,
 474                                int (*fn_parent)(void *), void *fn_parent_data,
 475                                int (*fn_child)(void *), void *fn_child_data);
 476 extern int userns_exec_mapped_root(const char *path, int path_fd,
 477                                    const struct lxc_conf *conf);
 478 static inline int chown_mapped_root(const char *path, const struct lxc_conf *conf)
 479 {
 480         return userns_exec_mapped_root(path, -EBADF, conf);
 481 }
 482
 483 #endif /* __LXC_CONF_H */