src/lxc/lsm/apparmor.c

   1 /* apparmor
   2  *
   3  * Copyright © 2012 Serge Hallyn <serge.hallyn@ubuntu.com>.
   4  * Copyright © 2012 Canonical Ltd.
   5  *
   6  *  This library is free software; you can redistribute it and/or
   7  *  modify it under the terms of the GNU Lesser General Public
   8  *  License as published by the Free Software Foundation; either
   9  *  version 2.1 of the License, or (at your option) any later version.
  10
  11  *  This library is distributed in the hope that it will be useful,
  12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  *  Lesser General Public License for more details.
  15
  16  *  You should have received a copy of the GNU Lesser General Public
  17  *  License along with this library; if not, write to the Free Software
  18  *  Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
  19  */
  20
  21 #ifndef _GNU_SOURCE
  22 #define _GNU_SOURCE 1
  23 #endif
  24 #include <errno.h>
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <sys/mount.h>
  28 #include <sys/stat.h>
  29 #include <sys/types.h>
  30 #include <sys/vfs.h>
  31 #include <unistd.h>
  32
  33 #include "caps.h"
  34 #include "conf.h"
  35 #include "config.h"
  36 #include "initutils.h"
  37 #include "log.h"
  38 #include "lsm.h"
  39 #include "parse.h"
  40 #include "raw_syscalls.h"
  41 #include "utils.h"
  42
  43 lxc_log_define(apparmor, lsm);
  44
  45 /* set by lsm_apparmor_drv_init if true */
  46 static int aa_enabled = 0;
  47 static bool aa_parser_available = false;
  48 static bool aa_supports_unix = false;
  49 static bool aa_can_stack = false;
  50 static bool aa_is_stacked = false;
  51 static bool aa_admin = false;
  52
  53 static int mount_features_enabled = 0;
  54
  55 #define AA_DEF_PROFILE "lxc-container-default"
  56 #define AA_DEF_PROFILE_CGNS "lxc-container-default-cgns"
  57 #define AA_MOUNT_RESTR "/sys/kernel/security/apparmor/features/mount/mask"
  58 #define AA_ENABLED_FILE "/sys/module/apparmor/parameters/enabled"
  59 #define AA_UNCHANGED "unchanged"
  60 #define AA_GENERATED "generated"
  61
  62 #define AA_CMD_LOAD   'r'
  63 #define AA_CMD_UNLOAD 'R'
  64 #define AA_CMD_PARSE  'Q'
  65
  66 static const char AA_PROFILE_BASE[] =
  67 "  ### Base profile\n"
  68 "  capability,\n"
  69 "  dbus,\n"
  70 "  file,\n"
  71 "  network,\n"
  72 "  umount,\n"
  73 "\n"
  74 "  # Allow us to receive signals from anywhere.\n"
  75 "  signal (receive),\n"
  76 "\n"
  77 "  # Allow us to send signals to ourselves\n"
  78 "  signal peer=@{profile_name},\n"
  79 "\n"
  80 "  # Allow other processes to read our /proc entries, futexes, perf tracing and\n"
  81 "  # kcmp for now (they will need 'read' in the first place). Administrators can\n"
  82 "  # override with:\n"
  83 "  #   deny ptrace (readby) ...\n"
  84 "  ptrace (readby),\n"
  85 "\n"
  86 "  # Allow other processes to trace us by default (they will need 'trace' in\n"
  87 "  # the first place). Administrators can override with:\n"
  88 "  #   deny ptrace (tracedby) ...\n"
  89 "  ptrace (tracedby),\n"
  90 "\n"
  91 "  # Allow us to ptrace ourselves\n"
  92 "  ptrace peer=@{profile_name},\n"
  93 "\n"
  94 "  # ignore DENIED message on / remount\n"
  95 "  deny mount options=(ro, remount) -> /,\n"
  96 "  deny mount options=(ro, remount, silent) -> /,\n"
  97 "\n"
  98 "  # allow tmpfs mounts everywhere\n"
  99 "  mount fstype=tmpfs,\n"
 100 "\n"
 101 "  # allow hugetlbfs mounts everywhere\n"
 102 "  mount fstype=hugetlbfs,\n"
 103 "\n"
 104 "  # allow mqueue mounts everywhere\n"
 105 "  mount fstype=mqueue,\n"
 106 "\n"
 107 "  # allow fuse mounts everywhere\n"
 108 "  mount fstype=fuse,\n"
 109 "  mount fstype=fuse.*,\n"
 110 "\n"
 111 "  # deny access under /proc/bus to avoid e.g. messing with pci devices directly\n"
 112 "  deny @{PROC}/bus/** wklx,\n"
 113 "\n"
 114 "  # deny writes in /proc/sys/fs but allow binfmt_misc to be mounted\n"
 115 "  mount fstype=binfmt_misc -> /proc/sys/fs/binfmt_misc/,\n"
 116 "  deny @{PROC}/sys/fs/** wklx,\n"
 117 "\n"
 118 "  # allow efivars to be mounted, writing to it will be blocked though\n"
 119 "  mount fstype=efivarfs -> /sys/firmware/efi/efivars/,\n"
 120 "\n"
 121 "  # block some other dangerous paths\n"
 122 "  deny @{PROC}/kcore rwklx,\n"
 123 "  deny @{PROC}/sysrq-trigger rwklx,\n"
 124 "\n"
 125 "  # deny writes in /sys except for /sys/fs/cgroup, also allow\n"
 126 "  # fusectl, securityfs and debugfs to be mounted there (read-only)\n"
 127 "  mount fstype=fusectl -> /sys/fs/fuse/connections/,\n"
 128 "  mount fstype=securityfs -> /sys/kernel/security/,\n"
 129 "  mount fstype=debugfs -> /sys/kernel/debug/,\n"
 130 "  deny mount fstype=debugfs -> /var/lib/ureadahead/debugfs/,\n"
 131 "  mount fstype=proc -> /proc/,\n"
 132 "  mount fstype=sysfs -> /sys/,\n"
 133 "  mount options=(rw, nosuid, nodev, noexec, remount) -> /sys/,\n"
 134 "  deny /sys/firmware/efi/efivars/** rwklx,\n"
 135 "  # note, /sys/kernel/security/** handled below\n"
 136 "  mount options=(ro, nosuid, nodev, noexec, remount, strictatime) -> /sys/fs/cgroup/,\n"
 137 "\n"
 138 "  # deny reads from debugfs\n"
 139 "  deny /sys/kernel/debug/{,**} rwklx,\n"
 140 "\n"
 141 "  # allow paths to be made slave, shared, private or unbindable\n"
 142 "  # FIXME: This currently doesn't work due to the apparmor parser treating those as allowing all mounts.\n"
 143 "#  mount options=(rw,make-slave) -> **,\n"
 144 "#  mount options=(rw,make-rslave) -> **,\n"
 145 "#  mount options=(rw,make-shared) -> **,\n"
 146 "#  mount options=(rw,make-rshared) -> **,\n"
 147 "#  mount options=(rw,make-private) -> **,\n"
 148 "#  mount options=(rw,make-rprivate) -> **,\n"
 149 "#  mount options=(rw,make-unbindable) -> **,\n"
 150 "#  mount options=(rw,make-runbindable) -> **,\n"
 151 "\n"
 152 "  # allow bind-mounts of anything except /proc, /sys and /dev\n"
 153 "  mount options=(rw,bind) /[^spd]*{,/**},\n"
 154 "  mount options=(rw,bind) /d[^e]*{,/**},\n"
 155 "  mount options=(rw,bind) /de[^v]*{,/**},\n"
 156 "  mount options=(rw,bind) /dev/.[^l]*{,/**},\n"
 157 "  mount options=(rw,bind) /dev/.l[^x]*{,/**},\n"
 158 "  mount options=(rw,bind) /dev/.lx[^c]*{,/**},\n"
 159 "  mount options=(rw,bind) /dev/.lxc?*{,/**},\n"
 160 "  mount options=(rw,bind) /dev/[^.]*{,/**},\n"
 161 "  mount options=(rw,bind) /dev?*{,/**},\n"
 162 "  mount options=(rw,bind) /p[^r]*{,/**},\n"
 163 "  mount options=(rw,bind) /pr[^o]*{,/**},\n"
 164 "  mount options=(rw,bind) /pro[^c]*{,/**},\n"
 165 "  mount options=(rw,bind) /proc?*{,/**},\n"
 166 "  mount options=(rw,bind) /s[^y]*{,/**},\n"
 167 "  mount options=(rw,bind) /sy[^s]*{,/**},\n"
 168 "  mount options=(rw,bind) /sys?*{,/**},\n"
 169 "\n"
 170 "  # allow various ro-bind-*re*-mounts\n"
 171 "  mount options=(ro,remount,bind),\n"
 172 "  mount options=(ro,remount,bind,nosuid),\n"
 173 "  mount options=(ro,remount,bind,noexec),\n"
 174 "  mount options=(ro,remount,bind,nodev),\n"
 175 "  mount options=(ro,remount,bind,nosuid,noexec),\n"
 176 "  mount options=(ro,remount,bind,noexec,nodev),\n"
 177 "  mount options=(ro,remount,bind,nodev,nosuid),\n"
 178 "  mount options=(ro,remount,bind,nosuid,noexec,nodev),\n"
 179 "\n"
 180 "  # allow moving mounts except for /proc, /sys and /dev\n"
 181 "  mount options=(rw,move) /[^spd]*{,/**},\n"
 182 "  mount options=(rw,move) /d[^e]*{,/**},\n"
 183 "  mount options=(rw,move) /de[^v]*{,/**},\n"
 184 "  mount options=(rw,move) /dev/.[^l]*{,/**},\n"
 185 "  mount options=(rw,move) /dev/.l[^x]*{,/**},\n"
 186 "  mount options=(rw,move) /dev/.lx[^c]*{,/**},\n"
 187 "  mount options=(rw,move) /dev/.lxc?*{,/**},\n"
 188 "  mount options=(rw,move) /dev/[^.]*{,/**},\n"
 189 "  mount options=(rw,move) /dev?*{,/**},\n"
 190 "  mount options=(rw,move) /p[^r]*{,/**},\n"
 191 "  mount options=(rw,move) /pr[^o]*{,/**},\n"
 192 "  mount options=(rw,move) /pro[^c]*{,/**},\n"
 193 "  mount options=(rw,move) /proc?*{,/**},\n"
 194 "  mount options=(rw,move) /s[^y]*{,/**},\n"
 195 "  mount options=(rw,move) /sy[^s]*{,/**},\n"
 196 "  mount options=(rw,move) /sys?*{,/**},\n"
 197 "\n"
 198 "  # generated by: lxc-generate-aa-rules.py container-rules.base\n"
 199 "  deny /proc/sys/[^kn]*{,/**} wklx,\n"
 200 "  deny /proc/sys/k[^e]*{,/**} wklx,\n"
 201 "  deny /proc/sys/ke[^r]*{,/**} wklx,\n"
 202 "  deny /proc/sys/ker[^n]*{,/**} wklx,\n"
 203 "  deny /proc/sys/kern[^e]*{,/**} wklx,\n"
 204 "  deny /proc/sys/kerne[^l]*{,/**} wklx,\n"
 205 "  deny /proc/sys/kernel/[^smhd]*{,/**} wklx,\n"
 206 "  deny /proc/sys/kernel/d[^o]*{,/**} wklx,\n"
 207 "  deny /proc/sys/kernel/do[^m]*{,/**} wklx,\n"
 208 "  deny /proc/sys/kernel/dom[^a]*{,/**} wklx,\n"
 209 "  deny /proc/sys/kernel/doma[^i]*{,/**} wklx,\n"
 210 "  deny /proc/sys/kernel/domai[^n]*{,/**} wklx,\n"
 211 "  deny /proc/sys/kernel/domain[^n]*{,/**} wklx,\n"
 212 "  deny /proc/sys/kernel/domainn[^a]*{,/**} wklx,\n"
 213 "  deny /proc/sys/kernel/domainna[^m]*{,/**} wklx,\n"
 214 "  deny /proc/sys/kernel/domainnam[^e]*{,/**} wklx,\n"
 215 "  deny /proc/sys/kernel/domainname?*{,/**} wklx,\n"
 216 "  deny /proc/sys/kernel/h[^o]*{,/**} wklx,\n"
 217 "  deny /proc/sys/kernel/ho[^s]*{,/**} wklx,\n"
 218 "  deny /proc/sys/kernel/hos[^t]*{,/**} wklx,\n"
 219 "  deny /proc/sys/kernel/host[^n]*{,/**} wklx,\n"
 220 "  deny /proc/sys/kernel/hostn[^a]*{,/**} wklx,\n"
 221 "  deny /proc/sys/kernel/hostna[^m]*{,/**} wklx,\n"
 222 "  deny /proc/sys/kernel/hostnam[^e]*{,/**} wklx,\n"
 223 "  deny /proc/sys/kernel/hostname?*{,/**} wklx,\n"
 224 "  deny /proc/sys/kernel/m[^s]*{,/**} wklx,\n"
 225 "  deny /proc/sys/kernel/ms[^g]*{,/**} wklx,\n"
 226 "  deny /proc/sys/kernel/msg*/** wklx,\n"
 227 "  deny /proc/sys/kernel/s[^he]*{,/**} wklx,\n"
 228 "  deny /proc/sys/kernel/se[^m]*{,/**} wklx,\n"
 229 "  deny /proc/sys/kernel/sem*/** wklx,\n"
 230 "  deny /proc/sys/kernel/sh[^m]*{,/**} wklx,\n"
 231 "  deny /proc/sys/kernel/shm*/** wklx,\n"
 232 "  deny /proc/sys/kernel?*{,/**} wklx,\n"
 233 "  deny /proc/sys/n[^e]*{,/**} wklx,\n"
 234 "  deny /proc/sys/ne[^t]*{,/**} wklx,\n"
 235 "  deny /proc/sys/net?*{,/**} wklx,\n"
 236 "  deny /sys/[^fdck]*{,/**} wklx,\n"
 237 "  deny /sys/c[^l]*{,/**} wklx,\n"
 238 "  deny /sys/cl[^a]*{,/**} wklx,\n"
 239 "  deny /sys/cla[^s]*{,/**} wklx,\n"
 240 "  deny /sys/clas[^s]*{,/**} wklx,\n"
 241 "  deny /sys/class/[^n]*{,/**} wklx,\n"
 242 "  deny /sys/class/n[^e]*{,/**} wklx,\n"
 243 "  deny /sys/class/ne[^t]*{,/**} wklx,\n"
 244 "  deny /sys/class/net?*{,/**} wklx,\n"
 245 "  deny /sys/class?*{,/**} wklx,\n"
 246 "  deny /sys/d[^e]*{,/**} wklx,\n"
 247 "  deny /sys/de[^v]*{,/**} wklx,\n"
 248 "  deny /sys/dev[^i]*{,/**} wklx,\n"
 249 "  deny /sys/devi[^c]*{,/**} wklx,\n"
 250 "  deny /sys/devic[^e]*{,/**} wklx,\n"
 251 "  deny /sys/device[^s]*{,/**} wklx,\n"
 252 "  deny /sys/devices/[^v]*{,/**} wklx,\n"
 253 "  deny /sys/devices/v[^i]*{,/**} wklx,\n"
 254 "  deny /sys/devices/vi[^r]*{,/**} wklx,\n"
 255 "  deny /sys/devices/vir[^t]*{,/**} wklx,\n"
 256 "  deny /sys/devices/virt[^u]*{,/**} wklx,\n"
 257 "  deny /sys/devices/virtu[^a]*{,/**} wklx,\n"
 258 "  deny /sys/devices/virtua[^l]*{,/**} wklx,\n"
 259 "  deny /sys/devices/virtual/[^n]*{,/**} wklx,\n"
 260 "  deny /sys/devices/virtual/n[^e]*{,/**} wklx,\n"
 261 "  deny /sys/devices/virtual/ne[^t]*{,/**} wklx,\n"
 262 "  deny /sys/devices/virtual/net?*{,/**} wklx,\n"
 263 "  deny /sys/devices/virtual?*{,/**} wklx,\n"
 264 "  deny /sys/devices?*{,/**} wklx,\n"
 265 "  deny /sys/f[^s]*{,/**} wklx,\n"
 266 "  deny /sys/fs/[^c]*{,/**} wklx,\n"
 267 "  deny /sys/fs/c[^g]*{,/**} wklx,\n"
 268 "  deny /sys/fs/cg[^r]*{,/**} wklx,\n"
 269 "  deny /sys/fs/cgr[^o]*{,/**} wklx,\n"
 270 "  deny /sys/fs/cgro[^u]*{,/**} wklx,\n"
 271 "  deny /sys/fs/cgrou[^p]*{,/**} wklx,\n"
 272 "  deny /sys/fs/cgroup?*{,/**} wklx,\n"
 273 "  deny /sys/fs?*{,/**} wklx,\n"
 274 ;
 275
 276 static const char AA_PROFILE_UNIX_SOCKETS[] =
 277 "\n"
 278 "  ### Feature: unix\n"
 279 "  # Allow receive via unix sockets from anywhere\n"
 280 "  unix (receive),\n"
 281 "\n"
 282 "  # Allow all unix sockets in the container\n"
 283 "  unix peer=(label=@{profile_name}),\n"
 284 ;
 285
 286 static const char AA_PROFILE_CGROUP_NAMESPACES[] =
 287 "\n"
 288 "  ### Feature: cgroup namespace\n"
 289 "  mount fstype=cgroup -> /sys/fs/cgroup/**,\n"
 290 "  mount fstype=cgroup2 -> /sys/fs/cgroup/**,\n"
 291 ;
 292
 293 /* '_BASE' because we still need to append generated change_profile rules */
 294 static const char AA_PROFILE_STACKING_BASE[] =
 295 "\n"
 296 "  ### Feature: apparmor stacking\n"
 297 "  ### Configuration: apparmor profile loading (in namespace)\n"
 298 "  deny /sys/k[^e]*{,/**} wklx,\n"
 299 "  deny /sys/ke[^r]*{,/**} wklx,\n"
 300 "  deny /sys/ker[^n]*{,/**} wklx,\n"
 301 "  deny /sys/kern[^e]*{,/**} wklx,\n"
 302 "  deny /sys/kerne[^l]*{,/**} wklx,\n"
 303 "  deny /sys/kernel/[^s]*{,/**} wklx,\n"
 304 "  deny /sys/kernel/s[^e]*{,/**} wklx,\n"
 305 "  deny /sys/kernel/se[^c]*{,/**} wklx,\n"
 306 "  deny /sys/kernel/sec[^u]*{,/**} wklx,\n"
 307 "  deny /sys/kernel/secu[^r]*{,/**} wklx,\n"
 308 "  deny /sys/kernel/secur[^i]*{,/**} wklx,\n"
 309 "  deny /sys/kernel/securi[^t]*{,/**} wklx,\n"
 310 "  deny /sys/kernel/securit[^y]*{,/**} wklx,\n"
 311 "  deny /sys/kernel/security/[^a]*{,/**} wklx,\n"
 312 "  deny /sys/kernel/security/a[^p]*{,/**} wklx,\n"
 313 "  deny /sys/kernel/security/ap[^p]*{,/**} wklx,\n"
 314 "  deny /sys/kernel/security/app[^a]*{,/**} wklx,\n"
 315 "  deny /sys/kernel/security/appa[^r]*{,/**} wklx,\n"
 316 "  deny /sys/kernel/security/appar[^m]*{,/**} wklx,\n"
 317 "  deny /sys/kernel/security/apparm[^o]*{,/**} wklx,\n"
 318 "  deny /sys/kernel/security/apparmo[^r]*{,/**} wklx,\n"
 319 "  deny /sys/kernel/security/apparmor?*{,/**} wklx,\n"
 320 "  deny /sys/kernel/security?*{,/**} wklx,\n"
 321 "  deny /sys/kernel?*{,/**} wklx,\n"
 322 ;
 323
 324 static const char AA_PROFILE_NO_STACKING[] =
 325 "\n"
 326 "  ### Feature: apparmor stacking (not present)\n"
 327 "  deny /sys/k*{,/**} rwklx,\n"
 328 ;
 329
 330 /* '_BASE' because we need to append change_profile for stacking */
 331 static const char AA_PROFILE_NESTING_BASE[] =
 332 "\n"
 333 "  ### Configuration: nesting\n"
 334 "  pivot_root,\n"
 335 "  ptrace,\n"
 336 "  signal,\n"
 337 "\n"
 338    /* NOTE: See conf.c's "nesting_helpers" for details. */
 339 "  deny /dev/.lxc/proc/** rw,\n"
 340 "  deny /dev/.lxc/sys/** rw,\n"
 341 "\n"
 342 "  mount fstype=proc -> /usr/lib/*/lxc/**,\n"
 343 "  mount fstype=sysfs -> /usr/lib/*/lxc/**,\n"
 344 "  mount options=(rw,bind),\n"
 345 "  mount options=(rw,rbind),\n"
 346 "  mount options=(rw,make-rshared),\n"
 347 "\n"
 348    /* FIXME: What's the state here on apparmor's side? */
 349 "  # there doesn't seem to be a way to ask for:\n"
 350 "  # mount options=(ro,nosuid,nodev,noexec,remount,bind),\n"
 351 "  # as we always get mount to $cdir/proc/sys with those flags denied\n"
 352 "  # So allow all mounts until that is straightened out:\n"
 353 "  mount,\n"
 354 ;
 355
 356 static const char AA_PROFILE_UNPRIVILEGED[] =
 357 "\n"
 358 "  ### Configuration: unprivileged container\n"
 359 "  pivot_root,\n"
 360 "\n"
 361 "  # Allow modifying mount propagation\n"
 362 "  mount options=(rw,make-slave) -> **,\n"
 363 "  mount options=(rw,make-rslave) -> **,\n"
 364 "  mount options=(rw,make-shared) -> **,\n"
 365 "  mount options=(rw,make-rshared) -> **,\n"
 366 "  mount options=(rw,make-private) -> **,\n"
 367 "  mount options=(rw,make-rprivate) -> **,\n"
 368 "  mount options=(rw,make-unbindable) -> **,\n"
 369 "  mount options=(rw,make-runbindable) -> **,\n"
 370 "\n"
 371 "  # Allow all bind-mounts\n"
 372 "  mount options=(rw,bind),\n"
 373 "  mount options=(rw,rbind),\n"
 374 "\n"
 375 "  # Allow remounting things read-only\n"
 376 "  mount options=(ro,remount),\n"
 377 ;
 378
 379 static bool check_mount_feature_enabled(void)
 380 {
 381         return mount_features_enabled == 1;
 382 }
 383
 384 static void load_mount_features_enabled(void)
 385 {
 386         struct stat statbuf;
 387         int ret;
 388
 389         ret = stat(AA_MOUNT_RESTR, &statbuf);
 390         if (ret == 0)
 391                 mount_features_enabled = 1;
 392 }
 393
 394 /* aa_getcon is not working right now.  Use our hand-rolled version below */
 395 static int apparmor_enabled(void)
 396 {
 397         FILE *fin;
 398         char e;
 399         int ret;
 400
 401         fin = fopen_cloexec(AA_ENABLED_FILE, "r");
 402         if (!fin)
 403                 return 0;
 404         ret = fscanf(fin, "%c", &e);
 405         fclose(fin);
 406         if (ret == 1 && e == 'Y') {
 407                 load_mount_features_enabled();
 408                 return 1;
 409         }
 410
 411         return 0;
 412 }
 413
 414 static char *apparmor_process_label_get(pid_t pid)
 415 {
 416         char path[100], *space;
 417         int ret;
 418         char *buf = NULL, *newbuf;
 419         int sz = 0;
 420         FILE *f;
 421
 422         ret = snprintf(path, 100, "/proc/%d/attr/current", pid);
 423         if (ret < 0 || ret >= 100) {
 424                 ERROR("path name too long");
 425                 return NULL;
 426         }
 427 again:
 428         f = fopen_cloexec(path, "r");
 429         if (!f) {
 430                 SYSERROR("opening %s", path);
 431                 free(buf);
 432                 return NULL;
 433         }
 434         sz += 1024;
 435         newbuf = realloc(buf, sz);
 436         if (!newbuf) {
 437                 free(buf);
 438                 ERROR("out of memory");
 439                 fclose(f);
 440                 return NULL;
 441         }
 442         buf = newbuf;
 443         memset(buf, 0, sz);
 444         ret = fread(buf, 1, sz - 1, f);
 445         fclose(f);
 446         if (ret < 0) {
 447                 ERROR("reading %s", path);
 448                 free(buf);
 449                 return NULL;
 450         }
 451         if (ret >= sz)
 452                 goto again;
 453         space = strchr(buf, '\n');
 454         if (space)
 455                 *space = '\0';
 456         space = strchr(buf, ' ');
 457         if (space)
 458                 *space = '\0';
 459         return buf;
 460 }
 461
 462 /*
 463  * Probably makes sense to reorganize these to only read
 464  * the label once
 465  */
 466 static bool apparmor_am_unconfined(void)
 467 {
 468         char *p = apparmor_process_label_get(lxc_raw_getpid());
 469         bool ret = false;
 470         if (!p || strcmp(p, "unconfined") == 0)
 471                 ret = true;
 472         free(p);
 473         return ret;
 474 }
 475
 476 static bool aa_needs_transition(char *curlabel)
 477 {
 478         if (!curlabel)
 479                 return false;
 480         if (strcmp(curlabel, "unconfined") == 0)
 481                 return false;
 482         if (strcmp(curlabel, "/usr/bin/lxc-start") == 0)
 483                 return false;
 484         return true;
 485 }
 486
 487 static inline void uint64hex(char *buf, uint64_t num)
 488 {
 489         size_t i;
 490
 491         buf[16] = 0;
 492         for (i = 16; i--;) {
 493                 char c = (char)(num & 0xf);
 494                 buf[i] = c + (c < 0xa ? '0' : 'a' - 0xa);
 495                 num >>= 4;
 496         }
 497 }
 498
 499 static inline char *shorten_apparmor_name(char *name)
 500 {
 501         size_t len = strlen(name);
 502         if (len + 7 > 253) {
 503                 uint64_t hash;
 504                 hash = fnv_64a_buf(name, len, FNV1A_64_INIT);
 505                 name = must_realloc(name, 16 + 1);
 506                 uint64hex(name, hash);
 507         }
 508
 509         return name;
 510 }
 511
 512 /* Replace slashes with hyphens */
 513 static inline void sanitize_path(char *path)
 514 {
 515         size_t i;
 516
 517         for (i = 0; path[i]; i++)
 518                 if (path[i] == '/')
 519                         path[i] = '-';
 520 }
 521
 522 static inline char *apparmor_dir(const char *ctname, const char *lxcpath)
 523 {
 524         return must_make_path(lxcpath, ctname, "apparmor", NULL);
 525 }
 526
 527
 528 static inline char *apparmor_profile_full(const char *ctname, const char *lxcpath)
 529 {
 530         return shorten_apparmor_name(must_concat("lxc-", ctname, "_<", lxcpath, ">", NULL));
 531 }
 532
 533 /* Like apparmor_profile_full() but with slashes replaced by hyphens */
 534 static inline char *apparmor_namespace(const char *ctname, const char *lxcpath)
 535 {
 536         char *full;
 537
 538         full = apparmor_profile_full(ctname, lxcpath);
 539         sanitize_path(full);
 540
 541         return full;
 542 }
 543
 544 /* FIXME: This is currently run only in the context of a constructor (via the
 545  * initial lsm_init() called due to its __attribute__((constructor)), so we
 546  * do not have ERROR/... macros available, so there are some fprintf(stderr)s
 547  * in there.
 548  */
 549 static bool check_apparmor_parser_version()
 550 {
 551         struct lxc_popen_FILE *parserpipe;
 552         int rc;
 553         int major = 0, minor = 0, micro = 0;
 554
 555         parserpipe = lxc_popen("apparmor_parser --version");
 556         if (!parserpipe) {
 557                 fprintf(stderr, "Failed to run check for apparmor_parser\n");
 558                 return false;
 559         }
 560
 561         rc = fscanf(parserpipe->f, "AppArmor parser version %d.%d.%d", &major, &minor, &micro);
 562         if (rc < 1) {
 563                 lxc_pclose(parserpipe);
 564                 /* We stay silent for now as this most likely means the shell
 565                  * lxc_popen executed failed to find the apparmor_parser binary.
 566                  * See the FIXME comment above for details.
 567                  */
 568                 return false;
 569         }
 570
 571         rc = lxc_pclose(parserpipe);
 572         if (rc < 0) {
 573                 fprintf(stderr, "Error waiting for child process\n");
 574                 return false;
 575         }
 576         if (rc != 0) {
 577                 fprintf(stderr, "'apparmor_parser --version' executed with an error status\n");
 578                 return false;
 579         }
 580
 581         aa_supports_unix = (major > 2) ||
 582                            (major == 2 && minor > 10) ||
 583                            (major == 2 && minor == 10 && micro >= 95);
 584
 585         return true;
 586 }
 587
 588 static bool file_is_yes(const char *path)
 589 {
 590         ssize_t rd;
 591         int fd;
 592         char buf[8]; /* we actually just expect "yes" or "no" */
 593
 594         fd = open(path, O_RDONLY | O_CLOEXEC);
 595         if (fd < 0)
 596                 return false;
 597
 598         rd = lxc_read_nointr(fd, buf, sizeof(buf));
 599         close(fd);
 600
 601         return rd >= 4 && strncmp(buf, "yes\n", 4) == 0;
 602 }
 603
 604 static bool apparmor_can_stack()
 605 {
 606         int major, minor, scanned;
 607         FILE *f;
 608
 609         if (!file_is_yes("/sys/kernel/security/apparmor/features/domain/stack"))
 610                 return false;
 611
 612         f = fopen_cloexec("/sys/kernel/security/apparmor/features/domain/version", "r");
 613         if (!f)
 614                 return false;
 615
 616         scanned = fscanf(f, "%d.%d", &major, &minor);
 617         fclose(f);
 618         if (scanned != 2)
 619                 return false;
 620
 621         return major > 1 || (major == 1 && minor >= 2);
 622 }
 623
 624 static void must_append_sized_full(char **buf, size_t *bufsz, const char *data,
 625                                    size_t size, bool append_newline)
 626 {
 627         size_t newsize = *bufsz + size;
 628
 629         if (append_newline)
 630                 ++newsize;
 631
 632         *buf = must_realloc(*buf, newsize);
 633         memcpy(*buf + *bufsz, data, size);
 634
 635         if (append_newline)
 636                 (*buf)[newsize - 1] = '\n';
 637
 638         *bufsz = newsize;
 639 }
 640
 641 static void must_append_sized(char **buf, size_t *bufsz, const char *data, size_t size)
 642 {
 643         return must_append_sized_full(buf, bufsz, data, size, false);
 644 }
 645
 646 static bool is_privileged(struct lxc_conf *conf)
 647 {
 648         return lxc_list_empty(&conf->id_map);
 649 }
 650
 651 static char *get_apparmor_profile_content(struct lxc_conf *conf, const char *lxcpath)
 652 {
 653         char *profile, *profile_name_full;
 654         size_t size;
 655         struct lxc_list *it;
 656
 657         profile_name_full = apparmor_profile_full(conf->name, lxcpath);
 658
 659         profile = must_concat(
 660 "#include <tunables/global>\n"
 661 "profile \"", profile_name_full, "\" flags=(attach_disconnected,mediate_deleted) {\n",
 662                               NULL);
 663         size = strlen(profile);
 664
 665         must_append_sized(&profile, &size, AA_PROFILE_BASE,
 666                           STRARRAYLEN(AA_PROFILE_BASE));
 667
 668         if (aa_supports_unix)
 669                 must_append_sized(&profile, &size, AA_PROFILE_UNIX_SOCKETS,
 670                                   STRARRAYLEN(AA_PROFILE_UNIX_SOCKETS));
 671
 672         if (file_exists("/proc/self/ns/cgroup"))
 673                 must_append_sized(&profile, &size, AA_PROFILE_CGROUP_NAMESPACES,
 674                                   STRARRAYLEN(AA_PROFILE_CGROUP_NAMESPACES));
 675
 676         if (aa_can_stack && !aa_is_stacked) {
 677                 char *namespace, *temp;
 678
 679                 must_append_sized(&profile, &size, AA_PROFILE_STACKING_BASE,
 680                                   STRARRAYLEN(AA_PROFILE_STACKING_BASE));
 681
 682                 namespace = apparmor_namespace(conf->name, lxcpath);
 683                 temp = must_concat("  change_profile -> \":", namespace, ":*\",\n"
 684                                    "  change_profile -> \":", namespace, "://*\",\n",
 685                                    NULL);
 686                 free(namespace);
 687
 688                 must_append_sized(&profile, &size, temp, strlen(temp));
 689                 free(temp);
 690         } else {
 691                 must_append_sized(&profile, &size, AA_PROFILE_NO_STACKING,
 692                                   STRARRAYLEN(AA_PROFILE_NO_STACKING));
 693         }
 694
 695         if (conf->lsm_aa_allow_nesting) {
 696                 must_append_sized(&profile, &size, AA_PROFILE_NESTING_BASE,
 697                                   STRARRAYLEN(AA_PROFILE_NESTING_BASE));
 698
 699                 if (!aa_can_stack || aa_is_stacked) {
 700                         char *temp;
 701
 702                         temp = must_concat("  change_profile -> \"",
 703                                            profile_name_full, "\",\n", NULL);
 704                         must_append_sized(&profile, &size, temp, strlen(temp));
 705                         free(temp);
 706                 }
 707         }
 708
 709         if (!is_privileged(conf) || am_host_unpriv())
 710                 must_append_sized(&profile, &size, AA_PROFILE_UNPRIVILEGED,
 711                                   STRARRAYLEN(AA_PROFILE_UNPRIVILEGED));
 712
 713         lxc_list_for_each(it, &conf->lsm_aa_raw) {
 714                 const char *line = it->elem;
 715
 716                 must_append_sized_full(&profile, &size, line, strlen(line), true);
 717         }
 718
 719         /* include terminating \0 byte */
 720         must_append_sized(&profile, &size, "}\n", 3);
 721
 722         free(profile_name_full);
 723
 724         return profile;
 725 }
 726
 727 /*
 728  * apparmor_parser creates a cache file using the parsed file's name as a name.
 729  * This means there may be multiple containers with the same name but different
 730  * lxcpaths. Therefore we need a sanitized version of the complete profile name
 731  * as profile file-name.
 732  * We already get this exactly from apparmor_namespace().
 733  */
 734 static char *make_apparmor_profile_path(const char *ctname, const char *lxcpath)
 735 {
 736         char *ret, *filename;
 737
 738         filename = apparmor_namespace(ctname, lxcpath);
 739         ret = must_make_path(lxcpath, ctname, "apparmor", filename, NULL);
 740         free(filename);
 741
 742         return ret;
 743 }
 744
 745 static char *make_apparmor_namespace_path(const char *ctname, const char *lxcpath)
 746 {
 747         char *ret, *namespace;
 748
 749         namespace = apparmor_namespace(ctname, lxcpath);
 750         ret = must_make_path("/sys/kernel/security/apparmor/policy/namespaces", namespace, NULL);
 751         free(namespace);
 752
 753         return ret;
 754 }
 755
 756 static bool make_apparmor_namespace(struct lxc_conf *conf, const char *lxcpath)
 757 {
 758         char *path;
 759
 760         if (!aa_can_stack || aa_is_stacked)
 761                 return true;
 762
 763         path = make_apparmor_namespace_path(conf->name, lxcpath);
 764         errno = 0;
 765         if (mkdir(path, 0755) < 0 && errno != EEXIST) {
 766                 SYSERROR("Error creating AppArmor namespace: %s", path);
 767                 free(path);
 768                 return false;
 769         }
 770         free(path);
 771
 772         return true;
 773 }
 774
 775 static void remove_apparmor_namespace(struct lxc_conf *conf, const char *lxcpath)
 776 {
 777         char *path;
 778
 779         path = make_apparmor_namespace_path(conf->name, lxcpath);
 780         if (rmdir(path) != 0)
 781                 SYSERROR("Error removing AppArmor namespace");
 782         free(path);
 783 }
 784
 785 struct apparmor_parser_args {
 786         char cmd;
 787         char *file;
 788 };
 789
 790 static int apparmor_parser_exec(void *data)
 791 {
 792         struct apparmor_parser_args *args = data;
 793         char cmdbuf[] = { '-', args->cmd, 'W', 'L', 0 };
 794
 795         execlp("apparmor_parser", "apparmor_parser", cmdbuf, APPARMOR_CACHE_DIR, args->file, NULL);
 796
 797         return -1;
 798 }
 799
 800 static int run_apparmor_parser(char command,
 801                                struct lxc_conf *conf,
 802                                const char *lxcpath)
 803 {
 804         char output[PATH_MAX];
 805         int ret;
 806         struct apparmor_parser_args args = {
 807                 .cmd = command,
 808                 .file = make_apparmor_profile_path(conf->name, lxcpath),
 809         };
 810
 811         ret = run_command(output, sizeof(output), apparmor_parser_exec, (void*)&args);
 812         if (ret < 0) {
 813                 ERROR("Failed to run apparmor_parser on \"%s\": %s", args.file, output);
 814                 ret = -1;
 815         }
 816
 817
 818         free(args.file);
 819         return ret;
 820 }
 821
 822 static void remove_apparmor_profile(struct lxc_conf *conf, const char *lxcpath)
 823 {
 824         char *path;
 825
 826         /* It's ok if these deletes fail: if the container was never started,
 827          * we'll have never written a profile or cached it.
 828          */
 829
 830         path = make_apparmor_profile_path(conf->name, lxcpath);
 831         (void)unlink(path);
 832         free(path);
 833
 834         /* Also remove the apparmor/ subdirectory */
 835         path = apparmor_dir(conf->name, lxcpath);
 836         (void)rmdir(path);
 837         free(path);
 838 }
 839
 840 static int load_apparmor_profile(struct lxc_conf *conf, const char *lxcpath)
 841 {
 842         struct stat profile_sb;
 843         size_t content_len;
 844         int ret = -1;
 845         size_t old_len = 0;
 846         char *profile_path = NULL, *old_content = NULL, *new_content = NULL;
 847         int profile_fd = -1;
 848
 849         if (!make_apparmor_namespace(conf, lxcpath))
 850                 return -1;
 851
 852         /* In order to avoid forcing a profile parse (potentially slow) on
 853          * every container start, let's use apparmor's binary policy cache,
 854          * which checks mtime of the files to figure out if the policy needs to
 855          * be regenerated.
 856          *
 857          * Since it uses mtimes, we shouldn't just always write out our local
 858          * apparmor template; instead we should check to see whether the
 859          * template is the same as ours. If it isn't we should write our
 860          * version out so that the new changes are reflected and we definitely
 861          * force a recompile.
 862          */
 863
 864         profile_path = make_apparmor_profile_path(conf->name, lxcpath);
 865         profile_fd = open(profile_path, O_RDONLY | O_CLOEXEC);
 866         if (profile_fd >= 0) {
 867                 if (fstat(profile_fd, &profile_sb) < 0) {
 868                         SYSERROR("Error accessing old profile from %s",
 869                                  profile_path);
 870                         goto out;
 871                 }
 872                 old_len = profile_sb.st_size;
 873                 old_content = lxc_strmmap(NULL, old_len, PROT_READ,
 874                                           MAP_PRIVATE, profile_fd, 0);
 875                 if (!old_content) {
 876                         SYSERROR("Failed to mmap old profile from %s",
 877                                  profile_path);
 878                         goto out;
 879                 }
 880         } else if (errno != ENOENT) {
 881                 SYSERROR("Error reading old profile from %s", profile_path);
 882                 goto out;
 883         }
 884
 885         new_content = get_apparmor_profile_content(conf, lxcpath);
 886         if (!new_content)
 887                 goto out;
 888
 889         content_len = strlen(new_content);
 890
 891         if (!old_content || old_len != content_len || memcmp(old_content, new_content, content_len) != 0) {
 892                 char *path;
 893
 894                 ret = mkdir_p(APPARMOR_CACHE_DIR, 0755);
 895                 if (ret < 0) {
 896                         SYSERROR("Error creating AppArmor profile cache directory " APPARMOR_CACHE_DIR);
 897                         goto out;
 898                 }
 899
 900                 path = apparmor_dir(conf->name, lxcpath);
 901                 ret = mkdir_p(path, 0755);
 902                 if (ret < 0) {
 903                         SYSERROR("Error creating AppArmor profile directory: %s", path);
 904                         free(path);
 905                         goto out;
 906                 }
 907                 free(path);
 908
 909                 ret = lxc_write_to_file(profile_path, new_content, content_len, false, 0600);
 910                 if (ret < 0) {
 911                         SYSERROR("Error writing profile to %s", profile_path);
 912                         goto out;
 913                 }
 914         }
 915
 916         ret = run_apparmor_parser(AA_CMD_LOAD, conf, lxcpath);
 917         if (ret != 0)
 918                 goto out_remove_profile;
 919
 920         conf->lsm_aa_profile_created = true;
 921
 922         goto out_ok;
 923
 924 out_remove_profile:
 925         remove_apparmor_profile(conf, lxcpath);
 926 out:
 927         remove_apparmor_namespace(conf, lxcpath);
 928 out_ok:
 929         if (profile_fd >= 0) {
 930                 if (old_content)
 931                         lxc_strmunmap(old_content, old_len);
 932                 close(profile_fd);
 933         }
 934         free(profile_path);
 935         free(new_content);
 936         return ret;
 937 }
 938
 939 /*
 940  * Ensure that the container's policy namespace is unloaded to free kernel
 941  * memory. This does not delete the policy from disk or cache.
 942  */
 943 static void apparmor_cleanup(struct lxc_conf *conf, const char *lxcpath)
 944 {
 945         if (!aa_admin)
 946                 return;
 947
 948         if (!conf->lsm_aa_profile_created)
 949                 return;
 950
 951         remove_apparmor_namespace(conf, lxcpath);
 952         (void)run_apparmor_parser(AA_CMD_UNLOAD, conf, lxcpath);
 953
 954         remove_apparmor_profile(conf, lxcpath);
 955 }
 956
 957 static int apparmor_prepare(struct lxc_conf *conf, const char *lxcpath)
 958 {
 959         int ret = -1;
 960         const char *label;
 961         char *curlabel = NULL, *genlabel = NULL;
 962
 963         if (!aa_enabled) {
 964                 ERROR("AppArmor not enabled");
 965                 return -1;
 966         }
 967
 968         label = conf->lsm_aa_profile;
 969
 970         /* user may request that we just ignore apparmor */
 971         if (label && strcmp(label, AA_UNCHANGED) == 0) {
 972                 INFO("AppArmor profile unchanged per user request");
 973                 conf->lsm_aa_profile_computed = must_copy_string(label);
 974                 return 0;
 975         }
 976
 977         if (label && strcmp(label, AA_GENERATED) == 0) {
 978                 if (!aa_parser_available) {
 979                         ERROR("Cannot use generated profile: apparmor_parser not available");
 980                         goto out;
 981                 }
 982
 983                 /* auto-generate profile based on available/requested security features */
 984                 if (load_apparmor_profile(conf, lxcpath) != 0) {
 985                         ERROR("Failed to load generated AppArmor profile");
 986                         goto out;
 987                 }
 988
 989                 genlabel = apparmor_profile_full(conf->name, lxcpath);
 990                 if (!genlabel) {
 991                         ERROR("Failed to build AppArmor profile name");
 992                         goto out;
 993                 }
 994
 995                 if (aa_can_stack && !aa_is_stacked) {
 996                         char *namespace = apparmor_namespace(conf->name, lxcpath);
 997                         size_t llen = strlen(genlabel);
 998                         must_append_sized(&genlabel, &llen, "//&:", STRARRAYLEN("//&:"));
 999                         must_append_sized(&genlabel, &llen, namespace, strlen(namespace));
1000                         must_append_sized(&genlabel, &llen, ":", STRARRAYLEN(":") + 1); /* with the nul byte */
1001                         free(namespace);
1002                 }
1003
1004                 label = genlabel;
1005         }
1006
1007         curlabel = apparmor_process_label_get(lxc_raw_getpid());
1008
1009         if (!aa_can_stack && aa_needs_transition(curlabel)) {
1010                 /* we're already confined, and stacking isn't supported */
1011
1012                 if (!label || strcmp(curlabel, label) == 0) {
1013                         /* no change requested */
1014                         ret = 0;
1015                         goto out;
1016                 }
1017
1018                 ERROR("Already AppArmor confined, but new label requested.");
1019                 goto out;
1020         }
1021
1022         if (!label) {
1023                 if (cgns_supported())
1024                         label = AA_DEF_PROFILE_CGNS;
1025                 else
1026                         label = AA_DEF_PROFILE;
1027         }
1028
1029         if (!check_mount_feature_enabled() && strcmp(label, "unconfined") != 0) {
1030                 WARN("Incomplete AppArmor support in your kernel");
1031                 if (!conf->lsm_aa_allow_incomplete) {
1032                         ERROR("If you really want to start this container, set");
1033                         ERROR("lxc.apparmor.allow_incomplete = 1");
1034                         ERROR("in your container configuration file");
1035                         goto out;
1036                 }
1037         }
1038
1039         conf->lsm_aa_profile_computed = must_copy_string(label);
1040         ret = 0;
1041
1042 out:
1043         if (genlabel) {
1044                 free(genlabel);
1045                 if (ret != 0)
1046                         apparmor_cleanup(conf, lxcpath);
1047         }
1048         free(curlabel);
1049         return ret;
1050 }
1051
1052 /*
1053  * apparmor_process_label_set: Set AppArmor process profile
1054  *
1055  * @label   : the profile to set
1056  * @conf    : the container configuration to use if @label is NULL
1057  * @default : use the default profile if @label is NULL
1058  * @on_exec : this is ignored.  Apparmor profile will be changed immediately
1059  *
1060  * Returns 0 on success, < 0 on failure
1061  *
1062  * Notes: This relies on /proc being available.
1063  */
1064 static int apparmor_process_label_set(const char *inlabel, struct lxc_conf *conf,
1065                                       bool on_exec)
1066 {
1067         int label_fd, ret;
1068         pid_t tid;
1069         const char *label;
1070
1071         if (!aa_enabled) {
1072                 ERROR("AppArmor not enabled");
1073                 return -1;
1074         }
1075
1076         label = inlabel ? inlabel : conf->lsm_aa_profile_computed;
1077         if (!label) {
1078                 ERROR("LSM wasn't prepared");
1079                 return -1;
1080         }
1081
1082         /* user may request that we just ignore apparmor */
1083         if (strcmp(label, AA_UNCHANGED) == 0) {
1084                 INFO("AppArmor profile unchanged per user request");
1085                 return 0;
1086         }
1087
1088         if (strcmp(label, "unconfined") == 0 && apparmor_am_unconfined()) {
1089                 INFO("AppArmor profile unchanged");
1090                 return 0;
1091         }
1092         tid = lxc_raw_gettid();
1093         label_fd = lsm_process_label_fd_get(tid, on_exec);
1094         if (label_fd < 0) {
1095                 SYSERROR("Failed to change AppArmor profile to %s", label);
1096                 return -1;
1097         }
1098
1099         ret = lsm_process_label_set_at(label_fd, label, on_exec);
1100         close(label_fd);
1101         if (ret < 0) {
1102                 ERROR("Failed to change AppArmor profile to %s", label);
1103                 return -1;
1104         }
1105
1106         INFO("Changed AppArmor profile to %s", label);
1107         return 0;
1108 }
1109
1110 static struct lsm_drv apparmor_drv = {
1111         .name = "AppArmor",
1112         .enabled           = apparmor_enabled,
1113         .process_label_get = apparmor_process_label_get,
1114         .process_label_set = apparmor_process_label_set,
1115         .prepare           = apparmor_prepare,
1116         .cleanup           = apparmor_cleanup,
1117 };
1118
1119 struct lsm_drv *lsm_apparmor_drv_init(void)
1120 {
1121         bool have_mac_admin = false;
1122
1123         if (!apparmor_enabled())
1124                 return NULL;
1125
1126         /* We only support generated profiles when apparmor_parser is usable */
1127         if (!check_apparmor_parser_version())
1128                 goto out;
1129
1130         aa_parser_available = true;
1131
1132         aa_can_stack = apparmor_can_stack();
1133         if (aa_can_stack)
1134                 aa_is_stacked = file_is_yes("/sys/kernel/security/apparmor/.ns_stacked");
1135
1136         #if HAVE_LIBCAP
1137         have_mac_admin = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE);
1138         #endif
1139
1140         if (!have_mac_admin)
1141                 WARN("Per-container AppArmor profiles are disabled because the mac_admin capability is missing");
1142         else if (am_host_unpriv() && !aa_is_stacked)
1143                 WARN("Per-container AppArmor profiles are disabled because LXC is running in an unprivileged container without stacking");
1144         else
1145                 aa_admin = true;
1146
1147 out:
1148         aa_enabled = 1;
1149         return &apparmor_drv;
1150 }