src/lxc/conf.h

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * (C) Copyright IBM Corp. 2007, 2008
   5  *
   6  * Authors:
   7  * Daniel Lezcano <daniel.lezcano at free.fr>
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23 #ifndef __LXC_CONF_H
  24 #define __LXC_CONF_H
  25
  26 #ifndef _GNU_SOURCE
  27 #define _GNU_SOURCE 1
  28 #endif
  29 #include <linux/magic.h>
  30 #include <net/if.h>
  31 #include <netinet/in.h>
  32 #include <stdbool.h>
  33 #include <stdio.h>
  34 #include <sys/param.h>
  35 #include <sys/types.h>
  36 #include <sys/vfs.h>
  37
  38 #include "compiler.h"
  39 #include "config.h"
  40 #include "list.h"
  41 #include "lxcseccomp.h"
  42 #include "ringbuf.h"
  43 #include "start.h"
  44 #include "terminal.h"
  45
  46 #if HAVE_SYS_RESOURCE_H
  47 #include <sys/resource.h>
  48 #endif
  49
  50 #if HAVE_SCMP_FILTER_CTX
  51 typedef void * scmp_filter_ctx;
  52 #endif
  53
  54 /* worth moving to configure.ac? */
  55 #define subuidfile "/etc/subuid"
  56 #define subgidfile "/etc/subgid"
  57
  58 /*
  59  * Defines a generic struct to configure the control group. It is up to the
  60  * programmer to specify the right subsystem.
  61  * @subsystem : the targeted subsystem
  62  * @value     : the value to set
  63  * @version   : The version of the cgroup filesystem on which the controller
  64  *              resides.
  65  *
  66  * @controllers : The controllers to use for this container.
  67  * @dir         : The name of the directory containing the container's cgroup.
  68  *                Not that this is a per-container setting.
  69  */
  70 struct lxc_cgroup {
  71         union {
  72                 /* information about a specific controller */
  73                 struct /* controller */ {
  74                         int version;
  75                         char *subsystem;
  76                         char *value;
  77                 };
  78
  79                 /* meta information about cgroup configuration */
  80                 struct /* meta */ {
  81                         char *controllers;
  82                         char *dir;
  83                         bool relative;
  84                 };
  85         };
  86 };
  87
  88 #if !HAVE_SYS_RESOURCE_H
  89 #define RLIM_INFINITY ((unsigned long)-1)
  90 struct rlimit {
  91         unsigned long rlim_cur;
  92         unsigned long rlim_max;
  93 };
  94 #endif
  95
  96 /*
  97  * Defines a structure to configure resource limits to set via setrlimit().
  98  * @resource : the resource name in lowercase without the RLIMIT_ prefix
  99  * @limit    : the limit to set
 100  */
 101 struct lxc_limit {
 102         char *resource;
 103         struct rlimit limit;
 104 };
 105
 106 enum idtype {
 107         ID_TYPE_UID,
 108         ID_TYPE_GID
 109 };
 110
 111 /*
 112  * Defines a structure to configure kernel parameters at runtime.
 113  * @key      : the kernel parameters will be configured without the "lxc.sysctl" prefix
 114  * @value    : the value to set
 115  */
 116 struct lxc_sysctl {
 117         char *key;
 118         char *value;
 119 };
 120
 121 /*
 122  * Defines a structure to configure proc filesystem at runtime.
 123  * @filename : the proc filesystem will be configured without the "lxc.proc" prefix
 124  * @value    : the value to set
 125  */
 126 struct lxc_proc {
 127         char *filename;
 128         char *value;
 129 };
 130
 131 /*
 132  * id_map is an id map entry.  Form in confile is:
 133  * lxc.idmap = u 0    9800 100
 134  * lxc.idmap = u 1000 9900 100
 135  * lxc.idmap = g 0    9800 100
 136  * lxc.idmap = g 1000 9900 100
 137  * meaning the container can use uids and gids 0-99 and 1000-1099,
 138  * with [ug]id 0 mapping to [ug]id 9800 on the host, and [ug]id 1000 to
 139  * [ug]id 9900 on the host.
 140  */
 141 struct id_map {
 142         enum idtype idtype;
 143         unsigned long hostid, nsid, range;
 144 };
 145
 146 /* Defines the number of tty configured and contains the
 147  * instantiated ptys
 148  * @max = number of configured ttys
 149  */
 150 struct lxc_tty_info {
 151         size_t max;
 152         char *dir;
 153         char *tty_names;
 154         struct lxc_terminal_info *tty;
 155 };
 156
 157 /* Defines a structure to store the rootfs location, the
 158  * optionals pivot_root, rootfs mount paths
 159  * @path       : the rootfs source (directory or device)
 160  * @mount      : where it is mounted
 161  * @bev_type   : optional backing store type
 162  * @options    : mount options
 163  * @mountflags : the portion of @options that are flags
 164  * @data       : the portion of @options that are not flags
 165  * @managed    : whether it is managed by LXC
 166  */
 167 struct lxc_rootfs {
 168         char *path;
 169         char *mount;
 170         char *bdev_type;
 171         char *options;
 172         unsigned long mountflags;
 173         char *data;
 174         bool managed;
 175 };
 176
 177 /*
 178  * Automatic mounts for LXC to perform inside the container
 179  */
 180 enum {
 181         LXC_AUTO_PROC_RW              = 0x001, /* /proc read-write */
 182         LXC_AUTO_PROC_MIXED           = 0x002, /* /proc/sys and /proc/sysrq-trigger read-only */
 183         LXC_AUTO_PROC_MASK            = 0x003,
 184
 185         LXC_AUTO_SYS_RW               = 0x004, /* /sys */
 186         LXC_AUTO_SYS_RO               = 0x008, /* /sys read-only */
 187         LXC_AUTO_SYS_MIXED            = 0x00C, /* /sys read-only and /sys/class/net read-write */
 188         LXC_AUTO_SYS_MASK             = 0x00C,
 189
 190         LXC_AUTO_CGROUP_RO            = 0x010, /* /sys/fs/cgroup (partial mount, read-only) */
 191         LXC_AUTO_CGROUP_RW            = 0x020, /* /sys/fs/cgroup (partial mount, read-write) */
 192         LXC_AUTO_CGROUP_MIXED         = 0x030, /* /sys/fs/cgroup (partial mount, paths r/o, cgroup r/w) */
 193         LXC_AUTO_CGROUP_FULL_RO       = 0x040, /* /sys/fs/cgroup (full mount, read-only) */
 194         LXC_AUTO_CGROUP_FULL_RW       = 0x050, /* /sys/fs/cgroup (full mount, read-write) */
 195         LXC_AUTO_CGROUP_FULL_MIXED    = 0x060, /* /sys/fs/cgroup (full mount, parent r/o, own r/w) */
 196         /*
 197          * These are defined in such a way as to retain binary compatibility
 198          * with earlier versions of this code. If the previous mask is applied,
 199          * both of these will default back to the _MIXED variants, which is
 200          * safe.
 201          */
 202         LXC_AUTO_CGROUP_NOSPEC        = 0x0B0, /* /sys/fs/cgroup (partial mount, r/w or mixed, depending on caps) */
 203         LXC_AUTO_CGROUP_FULL_NOSPEC   = 0x0E0, /* /sys/fs/cgroup (full mount, r/w or mixed, depending on caps) */
 204         LXC_AUTO_CGROUP_FORCE         = 0x100, /* mount cgroups even when cgroup namespaces are supported */
 205         LXC_AUTO_CGROUP_MASK          = 0x1F0, /* all known cgroup options, doe not contain LXC_AUTO_CGROUP_FORCE */
 206
 207         LXC_AUTO_SHMOUNTS             = 0x200, /* shared mount point */
 208         LXC_AUTO_SHMOUNTS_MASK        = 0x200, /* shared mount point mask */
 209         LXC_AUTO_ALL_MASK             = 0x1FF, /* all known settings */
 210 };
 211
 212 enum lxchooks {
 213         LXCHOOK_PRESTART,
 214         LXCHOOK_PREMOUNT,
 215         LXCHOOK_MOUNT,
 216         LXCHOOK_AUTODEV,
 217         LXCHOOK_START,
 218         LXCHOOK_STOP,
 219         LXCHOOK_POSTSTOP,
 220         LXCHOOK_CLONE,
 221         LXCHOOK_DESTROY,
 222         LXCHOOK_START_HOST,
 223         NUM_LXC_HOOKS
 224 };
 225
 226 extern char *lxchook_names[NUM_LXC_HOOKS];
 227
 228 struct lxc_state_client {
 229         int clientfd;
 230         lxc_state_t states[MAX_STATE];
 231 };
 232
 233 enum {
 234         LXC_BPF_DEVICE_CGROUP_LOCAL_RULE = -1,
 235         LXC_BPF_DEVICE_CGROUP_WHITELIST  =  0,
 236         LXC_BPF_DEVICE_CGROUP_BLACKLIST  =  1,
 237 };
 238
 239 struct device_item {
 240         char type;
 241         int major;
 242         int minor;
 243         char access[4];
 244         int allow;
 245         /*
 246          * LXC_BPF_DEVICE_CGROUP_LOCAL_RULE -> no global rule
 247          * LXC_BPF_DEVICE_CGROUP_WHITELIST  -> whitelist (deny all)
 248          * LXC_BPF_DEVICE_CGROUP_BLACKLIST  -> blacklist (allow all)
 249          */
 250         int global_rule;
 251 };
 252
 253 struct lxc_conf {
 254         /* Pointer to the name of the container. Do not free! */
 255         const char *name;
 256         bool is_execute;
 257         int reboot;
 258         signed long personality;
 259         struct utsname *utsname;
 260
 261         struct {
 262                 struct lxc_list cgroup;
 263                 struct lxc_list cgroup2;
 264                 struct bpf_program *cgroup2_devices;
 265                 /* This should be reimplemented as a hashmap. */
 266                 struct lxc_list devices;
 267         };
 268
 269         struct {
 270                 struct lxc_list id_map;
 271
 272                 /*
 273                  * Pointer to the idmap entry for the container's root uid in
 274                  * the id_map list. Do not free!
 275                  */
 276                 const struct id_map *root_nsuid_map;
 277
 278                 /*
 279                  * Pointer to the idmap entry for the container's root gid in
 280                  * the id_map list. Do not free!
 281                  */
 282                 const struct id_map *root_nsgid_map;
 283         };
 284
 285         struct lxc_list network;
 286
 287         struct {
 288                 char *fstab;
 289                 int auto_mounts;
 290                 struct lxc_list mount_list;
 291         };
 292
 293         struct lxc_list caps;
 294         struct lxc_list keepcaps;
 295
 296         /* /dev/tty<idx> devices */
 297         struct lxc_tty_info ttys;
 298         /* /dev/console device */
 299         struct lxc_terminal console;
 300         /* maximum pty devices allowed by devpts mount */
 301         size_t pty_max;
 302
 303         /* set to true when rootfs has been setup */
 304         bool rootfs_setup;
 305         struct lxc_rootfs rootfs;
 306
 307         bool close_all_fds;
 308
 309         struct {
 310                 unsigned int hooks_version;
 311                 struct lxc_list hooks[NUM_LXC_HOOKS];
 312         };
 313
 314         char *lsm_aa_profile;
 315         char *lsm_aa_profile_computed;
 316         bool lsm_aa_profile_created;
 317         unsigned int lsm_aa_allow_nesting;
 318         unsigned int lsm_aa_allow_incomplete;
 319         struct lxc_list lsm_aa_raw;
 320         char *lsm_se_context;
 321         bool tmp_umount_proc;
 322         struct lxc_seccomp seccomp;
 323         int maincmd_fd;
 324         unsigned int autodev;  /* if 1, mount and fill a /dev at start */
 325         int autodevtmpfssize; /* size of the /dev tmpfs */
 326         int haltsignal; /* signal used to halt container */
 327         int rebootsignal; /* signal used to reboot container */
 328         int stopsignal; /* signal used to hard stop container */
 329         char *rcfile;   /* Copy of the top level rcfile we read */
 330
 331         /* Logfile and loglevel can be set in a container config file. Those
 332          * function as defaults. The defaults can be overridden by command line.
 333          * However we don't want the command line specified values to be saved
 334          * on c->save_config(). So we store the config file specified values
 335          * here. */
 336         char *logfile; /* the logfile as specified in config */
 337         int loglevel; /* loglevel as specified in config (if any) */
 338         int logfd;
 339
 340         unsigned int start_auto;
 341         unsigned int start_delay;
 342         int start_order;
 343         struct lxc_list groups;
 344         int nbd_idx;
 345
 346         /* unshare the mount namespace in the monitor */
 347         unsigned int monitor_unshare;
 348         unsigned int monitor_signal_pdeath;
 349
 350         /* list of included files */
 351         struct lxc_list includes;
 352         /* config entries which are not "lxc.*" are aliens */
 353         struct lxc_list aliens;
 354
 355         /* list of environment variables we'll add to the container when
 356          * started */
 357         struct lxc_list environment;
 358
 359         /* text representation of the config file */
 360         char *unexpanded_config;
 361         size_t unexpanded_len;
 362         size_t unexpanded_alloced;
 363
 364         /* default command for lxc-execute */
 365         char *execute_cmd;
 366
 367         /* init command */
 368         char *init_cmd;
 369
 370         /* if running in a new user namespace, the UID/GID that init and COMMAND
 371          * should run under when using lxc-execute */
 372         uid_t init_uid;
 373         gid_t init_gid;
 374
 375         /* indicator if the container will be destroyed on shutdown */
 376         unsigned int ephemeral;
 377
 378         /* The facility to pass to syslog. Let's users establish as what type of
 379          * program liblxc is supposed to write to the syslog. */
 380         char *syslog;
 381
 382         /* Whether PR_SET_NO_NEW_PRIVS will be set for the container. */
 383         bool no_new_privs;
 384
 385         /* RLIMIT_* limits */
 386         struct lxc_list limits;
 387
 388         /* Contains generic info about the cgroup configuration for this
 389          * container. Note that struct lxc_cgroup contains a union. It is only
 390          * valid to access the members of the anonymous "meta" struct within
 391          * that union.
 392          */
 393         struct lxc_cgroup cgroup_meta;
 394
 395         struct {
 396                 int ns_clone;
 397                 int ns_keep;
 398                 char *ns_share[LXC_NS_MAX];
 399         };
 400
 401         /* init working directory */
 402         char *init_cwd;
 403
 404         /* A list of clients registered to be informed about a container state. */
 405         struct lxc_list state_clients;
 406
 407         /* sysctls */
 408         struct lxc_list sysctls;
 409
 410         /* procs */
 411         struct lxc_list procs;
 412
 413         struct shmount {
 414                 /* Absolute path to the shared mount point on the host */
 415                 char *path_host;
 416                 /* Absolute path (in the container) to the shared mount point */
 417                 char *path_cont;
 418         } shmount;
 419 };
 420
 421 extern int write_id_mapping(enum idtype idtype, pid_t pid, const char *buf,
 422                             size_t buf_size);
 423
 424 #ifdef HAVE_TLS
 425 extern thread_local struct lxc_conf *current_config;
 426 #else
 427 extern struct lxc_conf *current_config;
 428 #endif
 429
 430 extern int run_lxc_hooks(const char *name, char *hook, struct lxc_conf *conf,
 431                          char *argv[]);
 432 extern int detect_shared_rootfs(void);
 433 extern struct lxc_conf *lxc_conf_init(void);
 434 extern void lxc_conf_free(struct lxc_conf *conf);
 435 extern int pin_rootfs(const char *rootfs);
 436 extern int lxc_map_ids(struct lxc_list *idmap, pid_t pid);
 437 extern int lxc_create_tty(const char *name, struct lxc_conf *conf);
 438 extern void lxc_delete_tty(struct lxc_tty_info *ttys);
 439 extern int lxc_clear_config_caps(struct lxc_conf *c);
 440 extern int lxc_clear_config_keepcaps(struct lxc_conf *c);
 441 extern int lxc_clear_cgroups(struct lxc_conf *c, const char *key, int version);
 442 extern int lxc_clear_mount_entries(struct lxc_conf *c);
 443 extern int lxc_clear_automounts(struct lxc_conf *c);
 444 extern int lxc_clear_hooks(struct lxc_conf *c, const char *key);
 445 extern int lxc_clear_idmaps(struct lxc_conf *c);
 446 extern int lxc_clear_groups(struct lxc_conf *c);
 447 extern int lxc_clear_environment(struct lxc_conf *c);
 448 extern int lxc_clear_limits(struct lxc_conf *c, const char *key);
 449 extern int lxc_delete_autodev(struct lxc_handler *handler);
 450 extern int lxc_clear_autodev_tmpfs_size(struct lxc_conf *c);
 451 extern void lxc_clear_includes(struct lxc_conf *conf);
 452 extern int lxc_setup_rootfs_prepare_root(struct lxc_conf *conf,
 453                                          const char *name, const char *lxcpath);
 454 extern int lxc_setup(struct lxc_handler *handler);
 455 extern int lxc_setup_parent(struct lxc_handler *handler);
 456 extern int setup_resource_limits(struct lxc_list *limits, pid_t pid);
 457 extern int find_unmapped_nsid(struct lxc_conf *conf, enum idtype idtype);
 458 extern int mapped_hostid(unsigned id, struct lxc_conf *conf,
 459                          enum idtype idtype);
 460 extern int chown_mapped_root(const char *path, struct lxc_conf *conf);
 461 extern int userns_exec_1(struct lxc_conf *conf, int (*fn)(void *), void *data,
 462                          const char *fn_name);
 463 extern int userns_exec_full(struct lxc_conf *conf, int (*fn)(void *),
 464                             void *data, const char *fn_name);
 465 extern int parse_mntopts(const char *mntopts, unsigned long *mntflags,
 466                          char **mntdata);
 467 extern int parse_propagationopts(const char *mntopts, unsigned long *pflags);
 468 extern void tmp_proc_unmount(struct lxc_conf *lxc_conf);
 469 extern void remount_all_slave(void);
 470 extern void suggest_default_idmap(void);
 471 extern FILE *make_anonymous_mount_file(struct lxc_list *mount,
 472                                        bool include_nesting_helpers);
 473 extern struct lxc_list *sort_cgroup_settings(struct lxc_list *cgroup_settings);
 474 extern unsigned long add_required_remount_flags(const char *s, const char *d,
 475                                                 unsigned long flags);
 476 extern int run_script(const char *name, const char *section, const char *script,
 477                       ...);
 478 extern int run_script_argv(const char *name, unsigned int hook_version,
 479                            const char *section, const char *script,
 480                            const char *hookname, char **argsin);
 481 extern int in_caplist(int cap, struct lxc_list *caps);
 482 extern int setup_sysctl_parameters(struct lxc_list *sysctls);
 483 extern int lxc_clear_sysctls(struct lxc_conf *c, const char *key);
 484 extern int setup_proc_filesystem(struct lxc_list *procs, pid_t pid);
 485 extern int lxc_clear_procs(struct lxc_conf *c, const char *key);
 486 extern int lxc_clear_apparmor_raw(struct lxc_conf *c);
 487 extern int lxc_clear_namespace(struct lxc_conf *c);
 488
 489 #endif /* __LXC_CONF_H */