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