]> git.proxmox.com Git - mirror_lxc.git/blob - src/lxc/lsm/apparmor.c
Merge pull request #3235 from xinhua9569/master
[mirror_lxc.git] / src / lxc / lsm / apparmor.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #ifndef _GNU_SOURCE
4 #define _GNU_SOURCE 1
5 #endif
6 #include <errno.h>
7 #include <stdio.h>
8 #include <stdlib.h>
9 #include <sys/mount.h>
10 #include <sys/stat.h>
11 #include <sys/types.h>
12 #include <sys/vfs.h>
13 #include <unistd.h>
14
15 #include "caps.h"
16 #include "conf.h"
17 #include "config.h"
18 #include "initutils.h"
19 #include "log.h"
20 #include "lsm.h"
21 #include "parse.h"
22 #include "raw_syscalls.h"
23 #include "utils.h"
24
25 lxc_log_define(apparmor, lsm);
26
27 /* set by lsm_apparmor_drv_init if true */
28 static int aa_enabled = 0;
29 static bool aa_parser_available = false;
30 static bool aa_supports_unix = false;
31 static bool aa_can_stack = false;
32 static bool aa_is_stacked = false;
33 static bool aa_admin = false;
34
35 static int mount_features_enabled = 0;
36
37 #define AA_DEF_PROFILE "lxc-container-default"
38 #define AA_DEF_PROFILE_CGNS "lxc-container-default-cgns"
39 #define AA_MOUNT_RESTR "/sys/kernel/security/apparmor/features/mount/mask"
40 #define AA_ENABLED_FILE "/sys/module/apparmor/parameters/enabled"
41 #define AA_UNCHANGED "unchanged"
42 #define AA_GENERATED "generated"
43
44 #define AA_CMD_LOAD 'r'
45 #define AA_CMD_UNLOAD 'R'
46 #define AA_CMD_PARSE 'Q'
47
48 static const char AA_PROFILE_BASE[] =
49 " ### Base profile\n"
50 " capability,\n"
51 " dbus,\n"
52 " file,\n"
53 " network,\n"
54 " umount,\n"
55 "\n"
56 " # Allow us to receive signals from anywhere.\n"
57 " signal (receive),\n"
58 "\n"
59 " # Allow us to send signals to ourselves\n"
60 " signal peer=@{profile_name},\n"
61 "\n"
62 " # Allow other processes to read our /proc entries, futexes, perf tracing and\n"
63 " # kcmp for now (they will need 'read' in the first place). Administrators can\n"
64 " # override with:\n"
65 " # deny ptrace (readby) ...\n"
66 " ptrace (readby),\n"
67 "\n"
68 " # Allow other processes to trace us by default (they will need 'trace' in\n"
69 " # the first place). Administrators can override with:\n"
70 " # deny ptrace (tracedby) ...\n"
71 " ptrace (tracedby),\n"
72 "\n"
73 " # Allow us to ptrace ourselves\n"
74 " ptrace peer=@{profile_name},\n"
75 "\n"
76 " # ignore DENIED message on / remount\n"
77 " deny mount options=(ro, remount) -> /,\n"
78 " deny mount options=(ro, remount, silent) -> /,\n"
79 "\n"
80 " # allow tmpfs mounts everywhere\n"
81 " mount fstype=tmpfs,\n"
82 "\n"
83 " # allow hugetlbfs mounts everywhere\n"
84 " mount fstype=hugetlbfs,\n"
85 "\n"
86 " # allow mqueue mounts everywhere\n"
87 " mount fstype=mqueue,\n"
88 "\n"
89 " # allow fuse mounts everywhere\n"
90 " mount fstype=fuse,\n"
91 " mount fstype=fuse.*,\n"
92 "\n"
93 " # deny access under /proc/bus to avoid e.g. messing with pci devices directly\n"
94 " deny @{PROC}/bus/** wklx,\n"
95 "\n"
96 " # deny writes in /proc/sys/fs but allow binfmt_misc to be mounted\n"
97 " mount fstype=binfmt_misc -> /proc/sys/fs/binfmt_misc/,\n"
98 " deny @{PROC}/sys/fs/** wklx,\n"
99 "\n"
100 " # allow efivars to be mounted, writing to it will be blocked though\n"
101 " mount fstype=efivarfs -> /sys/firmware/efi/efivars/,\n"
102 "\n"
103 " # block some other dangerous paths\n"
104 " deny @{PROC}/kcore rwklx,\n"
105 " deny @{PROC}/sysrq-trigger rwklx,\n"
106 " deny @{PROC}/acpi/** rwklx,\n"
107 "\n"
108 " # deny writes in /sys except for /sys/fs/cgroup, also allow\n"
109 " # fusectl, securityfs and debugfs to be mounted there (read-only)\n"
110 " mount fstype=fusectl -> /sys/fs/fuse/connections/,\n"
111 " mount fstype=securityfs -> /sys/kernel/security/,\n"
112 " mount fstype=debugfs -> /sys/kernel/debug/,\n"
113 " deny mount fstype=debugfs -> /var/lib/ureadahead/debugfs/,\n"
114 " mount fstype=proc -> /proc/,\n"
115 " mount fstype=sysfs -> /sys/,\n"
116 " mount options=(rw, nosuid, nodev, noexec, remount) -> /sys/,\n"
117 " deny /sys/firmware/efi/efivars/** rwklx,\n"
118 " # note, /sys/kernel/security/** handled below\n"
119 " mount options=(ro, nosuid, nodev, noexec, remount, strictatime) -> /sys/fs/cgroup/,\n"
120 "\n"
121 " # deny reads from debugfs\n"
122 " deny /sys/kernel/debug/{,**} rwklx,\n"
123 "\n"
124 " # allow paths to be made slave, shared, private or unbindable\n"
125 " # FIXME: This currently doesn't work due to the apparmor parser treating those as allowing all mounts.\n"
126 "# mount options=(rw,make-slave) -> **,\n"
127 "# mount options=(rw,make-rslave) -> **,\n"
128 "# mount options=(rw,make-shared) -> **,\n"
129 "# mount options=(rw,make-rshared) -> **,\n"
130 "# mount options=(rw,make-private) -> **,\n"
131 "# mount options=(rw,make-rprivate) -> **,\n"
132 "# mount options=(rw,make-unbindable) -> **,\n"
133 "# mount options=(rw,make-runbindable) -> **,\n"
134 "\n"
135 " # allow bind-mounts of anything except /proc, /sys and /dev\n"
136 " mount options=(rw,bind) /[^spd]*{,/**},\n"
137 " mount options=(rw,bind) /d[^e]*{,/**},\n"
138 " mount options=(rw,bind) /de[^v]*{,/**},\n"
139 " mount options=(rw,bind) /dev/.[^l]*{,/**},\n"
140 " mount options=(rw,bind) /dev/.l[^x]*{,/**},\n"
141 " mount options=(rw,bind) /dev/.lx[^c]*{,/**},\n"
142 " mount options=(rw,bind) /dev/.lxc?*{,/**},\n"
143 " mount options=(rw,bind) /dev/[^.]*{,/**},\n"
144 " mount options=(rw,bind) /dev?*{,/**},\n"
145 " mount options=(rw,bind) /p[^r]*{,/**},\n"
146 " mount options=(rw,bind) /pr[^o]*{,/**},\n"
147 " mount options=(rw,bind) /pro[^c]*{,/**},\n"
148 " mount options=(rw,bind) /proc?*{,/**},\n"
149 " mount options=(rw,bind) /s[^y]*{,/**},\n"
150 " mount options=(rw,bind) /sy[^s]*{,/**},\n"
151 " mount options=(rw,bind) /sys?*{,/**},\n"
152 "\n"
153 " # allow various ro-bind-*re*-mounts\n"
154 " mount options=(ro,remount,bind),\n"
155 " mount options=(ro,remount,bind,nosuid),\n"
156 " mount options=(ro,remount,bind,noexec),\n"
157 " mount options=(ro,remount,bind,nodev),\n"
158 " mount options=(ro,remount,bind,nosuid,noexec),\n"
159 " mount options=(ro,remount,bind,noexec,nodev),\n"
160 " mount options=(ro,remount,bind,nodev,nosuid),\n"
161 " mount options=(ro,remount,bind,nosuid,noexec,nodev),\n"
162 "\n"
163 " # allow moving mounts except for /proc, /sys and /dev\n"
164 " mount options=(rw,move) /[^spd]*{,/**},\n"
165 " mount options=(rw,move) /d[^e]*{,/**},\n"
166 " mount options=(rw,move) /de[^v]*{,/**},\n"
167 " mount options=(rw,move) /dev/.[^l]*{,/**},\n"
168 " mount options=(rw,move) /dev/.l[^x]*{,/**},\n"
169 " mount options=(rw,move) /dev/.lx[^c]*{,/**},\n"
170 " mount options=(rw,move) /dev/.lxc?*{,/**},\n"
171 " mount options=(rw,move) /dev/[^.]*{,/**},\n"
172 " mount options=(rw,move) /dev?*{,/**},\n"
173 " mount options=(rw,move) /p[^r]*{,/**},\n"
174 " mount options=(rw,move) /pr[^o]*{,/**},\n"
175 " mount options=(rw,move) /pro[^c]*{,/**},\n"
176 " mount options=(rw,move) /proc?*{,/**},\n"
177 " mount options=(rw,move) /s[^y]*{,/**},\n"
178 " mount options=(rw,move) /sy[^s]*{,/**},\n"
179 " mount options=(rw,move) /sys?*{,/**},\n"
180 "\n"
181 " # generated by: lxc-generate-aa-rules.py container-rules.base\n"
182 " deny /proc/sys/[^kn]*{,/**} wklx,\n"
183 " deny /proc/sys/k[^e]*{,/**} wklx,\n"
184 " deny /proc/sys/ke[^r]*{,/**} wklx,\n"
185 " deny /proc/sys/ker[^n]*{,/**} wklx,\n"
186 " deny /proc/sys/kern[^e]*{,/**} wklx,\n"
187 " deny /proc/sys/kerne[^l]*{,/**} wklx,\n"
188 " deny /proc/sys/kernel/[^smhd]*{,/**} wklx,\n"
189 " deny /proc/sys/kernel/d[^o]*{,/**} wklx,\n"
190 " deny /proc/sys/kernel/do[^m]*{,/**} wklx,\n"
191 " deny /proc/sys/kernel/dom[^a]*{,/**} wklx,\n"
192 " deny /proc/sys/kernel/doma[^i]*{,/**} wklx,\n"
193 " deny /proc/sys/kernel/domai[^n]*{,/**} wklx,\n"
194 " deny /proc/sys/kernel/domain[^n]*{,/**} wklx,\n"
195 " deny /proc/sys/kernel/domainn[^a]*{,/**} wklx,\n"
196 " deny /proc/sys/kernel/domainna[^m]*{,/**} wklx,\n"
197 " deny /proc/sys/kernel/domainnam[^e]*{,/**} wklx,\n"
198 " deny /proc/sys/kernel/domainname?*{,/**} wklx,\n"
199 " deny /proc/sys/kernel/h[^o]*{,/**} wklx,\n"
200 " deny /proc/sys/kernel/ho[^s]*{,/**} wklx,\n"
201 " deny /proc/sys/kernel/hos[^t]*{,/**} wklx,\n"
202 " deny /proc/sys/kernel/host[^n]*{,/**} wklx,\n"
203 " deny /proc/sys/kernel/hostn[^a]*{,/**} wklx,\n"
204 " deny /proc/sys/kernel/hostna[^m]*{,/**} wklx,\n"
205 " deny /proc/sys/kernel/hostnam[^e]*{,/**} wklx,\n"
206 " deny /proc/sys/kernel/hostname?*{,/**} wklx,\n"
207 " deny /proc/sys/kernel/m[^s]*{,/**} wklx,\n"
208 " deny /proc/sys/kernel/ms[^g]*{,/**} wklx,\n"
209 " deny /proc/sys/kernel/msg*/** wklx,\n"
210 " deny /proc/sys/kernel/s[^he]*{,/**} wklx,\n"
211 " deny /proc/sys/kernel/se[^m]*{,/**} wklx,\n"
212 " deny /proc/sys/kernel/sem*/** wklx,\n"
213 " deny /proc/sys/kernel/sh[^m]*{,/**} wklx,\n"
214 " deny /proc/sys/kernel/shm*/** wklx,\n"
215 " deny /proc/sys/kernel?*{,/**} wklx,\n"
216 " deny /proc/sys/n[^e]*{,/**} wklx,\n"
217 " deny /proc/sys/ne[^t]*{,/**} wklx,\n"
218 " deny /proc/sys/net?*{,/**} wklx,\n"
219 " deny /sys/[^fdck]*{,/**} wklx,\n"
220 " deny /sys/c[^l]*{,/**} wklx,\n"
221 " deny /sys/cl[^a]*{,/**} wklx,\n"
222 " deny /sys/cla[^s]*{,/**} wklx,\n"
223 " deny /sys/clas[^s]*{,/**} wklx,\n"
224 " deny /sys/class/[^n]*{,/**} wklx,\n"
225 " deny /sys/class/n[^e]*{,/**} wklx,\n"
226 " deny /sys/class/ne[^t]*{,/**} wklx,\n"
227 " deny /sys/class/net?*{,/**} wklx,\n"
228 " deny /sys/class?*{,/**} wklx,\n"
229 " deny /sys/d[^e]*{,/**} wklx,\n"
230 " deny /sys/de[^v]*{,/**} wklx,\n"
231 " deny /sys/dev[^i]*{,/**} wklx,\n"
232 " deny /sys/devi[^c]*{,/**} wklx,\n"
233 " deny /sys/devic[^e]*{,/**} wklx,\n"
234 " deny /sys/device[^s]*{,/**} wklx,\n"
235 " deny /sys/devices/[^v]*{,/**} wklx,\n"
236 " deny /sys/devices/v[^i]*{,/**} wklx,\n"
237 " deny /sys/devices/vi[^r]*{,/**} wklx,\n"
238 " deny /sys/devices/vir[^t]*{,/**} wklx,\n"
239 " deny /sys/devices/virt[^u]*{,/**} wklx,\n"
240 " deny /sys/devices/virtu[^a]*{,/**} wklx,\n"
241 " deny /sys/devices/virtua[^l]*{,/**} wklx,\n"
242 " deny /sys/devices/virtual/[^n]*{,/**} wklx,\n"
243 " deny /sys/devices/virtual/n[^e]*{,/**} wklx,\n"
244 " deny /sys/devices/virtual/ne[^t]*{,/**} wklx,\n"
245 " deny /sys/devices/virtual/net?*{,/**} wklx,\n"
246 " deny /sys/devices/virtual?*{,/**} wklx,\n"
247 " deny /sys/devices?*{,/**} wklx,\n"
248 " deny /sys/f[^s]*{,/**} wklx,\n"
249 " deny /sys/fs/[^c]*{,/**} wklx,\n"
250 " deny /sys/fs/c[^g]*{,/**} wklx,\n"
251 " deny /sys/fs/cg[^r]*{,/**} wklx,\n"
252 " deny /sys/fs/cgr[^o]*{,/**} wklx,\n"
253 " deny /sys/fs/cgro[^u]*{,/**} wklx,\n"
254 " deny /sys/fs/cgrou[^p]*{,/**} wklx,\n"
255 " deny /sys/fs/cgroup?*{,/**} wklx,\n"
256 " deny /sys/fs?*{,/**} wklx,\n"
257 ;
258
259 static const char AA_PROFILE_UNIX_SOCKETS[] =
260 "\n"
261 " ### Feature: unix\n"
262 " # Allow receive via unix sockets from anywhere\n"
263 " unix (receive),\n"
264 "\n"
265 " # Allow all unix sockets in the container\n"
266 " unix peer=(label=@{profile_name}),\n"
267 ;
268
269 static const char AA_PROFILE_CGROUP_NAMESPACES[] =
270 "\n"
271 " ### Feature: cgroup namespace\n"
272 " mount fstype=cgroup -> /sys/fs/cgroup/**,\n"
273 " mount fstype=cgroup2 -> /sys/fs/cgroup/**,\n"
274 ;
275
276 /* '_BASE' because we still need to append generated change_profile rules */
277 static const char AA_PROFILE_STACKING_BASE[] =
278 "\n"
279 " ### Feature: apparmor stacking\n"
280 " ### Configuration: apparmor profile loading (in namespace)\n"
281 " deny /sys/k[^e]*{,/**} wklx,\n"
282 " deny /sys/ke[^r]*{,/**} wklx,\n"
283 " deny /sys/ker[^n]*{,/**} wklx,\n"
284 " deny /sys/kern[^e]*{,/**} wklx,\n"
285 " deny /sys/kerne[^l]*{,/**} wklx,\n"
286 " deny /sys/kernel/[^s]*{,/**} wklx,\n"
287 " deny /sys/kernel/s[^e]*{,/**} wklx,\n"
288 " deny /sys/kernel/se[^c]*{,/**} wklx,\n"
289 " deny /sys/kernel/sec[^u]*{,/**} wklx,\n"
290 " deny /sys/kernel/secu[^r]*{,/**} wklx,\n"
291 " deny /sys/kernel/secur[^i]*{,/**} wklx,\n"
292 " deny /sys/kernel/securi[^t]*{,/**} wklx,\n"
293 " deny /sys/kernel/securit[^y]*{,/**} wklx,\n"
294 " deny /sys/kernel/security/[^a]*{,/**} wklx,\n"
295 " deny /sys/kernel/security/a[^p]*{,/**} wklx,\n"
296 " deny /sys/kernel/security/ap[^p]*{,/**} wklx,\n"
297 " deny /sys/kernel/security/app[^a]*{,/**} wklx,\n"
298 " deny /sys/kernel/security/appa[^r]*{,/**} wklx,\n"
299 " deny /sys/kernel/security/appar[^m]*{,/**} wklx,\n"
300 " deny /sys/kernel/security/apparm[^o]*{,/**} wklx,\n"
301 " deny /sys/kernel/security/apparmo[^r]*{,/**} wklx,\n"
302 " deny /sys/kernel/security/apparmor?*{,/**} wklx,\n"
303 " deny /sys/kernel/security?*{,/**} wklx,\n"
304 " deny /sys/kernel?*{,/**} wklx,\n"
305 ;
306
307 static const char AA_PROFILE_NO_STACKING[] =
308 "\n"
309 " ### Feature: apparmor stacking (not present)\n"
310 " deny /sys/k*{,/**} rwklx,\n"
311 ;
312
313 /* '_BASE' because we need to append change_profile for stacking */
314 static const char AA_PROFILE_NESTING_BASE[] =
315 "\n"
316 " ### Configuration: nesting\n"
317 " pivot_root,\n"
318 " ptrace,\n"
319 " signal,\n"
320 "\n"
321 /* NOTE: See conf.c's "nesting_helpers" for details. */
322 " deny /dev/.lxc/proc/** rw,\n"
323 " deny /dev/.lxc/sys/** rw,\n"
324 "\n"
325 " mount fstype=proc -> /usr/lib/*/lxc/**,\n"
326 " mount fstype=sysfs -> /usr/lib/*/lxc/**,\n"
327 " mount options=(rw,bind),\n"
328 " mount options=(rw,rbind),\n"
329 " mount options=(rw,make-rshared),\n"
330 "\n"
331 /* FIXME: What's the state here on apparmor's side? */
332 " # there doesn't seem to be a way to ask for:\n"
333 " # mount options=(ro,nosuid,nodev,noexec,remount,bind),\n"
334 " # as we always get mount to $cdir/proc/sys with those flags denied\n"
335 " # So allow all mounts until that is straightened out:\n"
336 " mount,\n"
337 ;
338
339 static const char AA_PROFILE_UNPRIVILEGED[] =
340 "\n"
341 " ### Configuration: unprivileged container\n"
342 " pivot_root,\n"
343 "\n"
344 " # Allow modifying mount propagation\n"
345 " mount options=(rw,make-slave) -> **,\n"
346 " mount options=(rw,make-rslave) -> **,\n"
347 " mount options=(rw,make-shared) -> **,\n"
348 " mount options=(rw,make-rshared) -> **,\n"
349 " mount options=(rw,make-private) -> **,\n"
350 " mount options=(rw,make-rprivate) -> **,\n"
351 " mount options=(rw,make-unbindable) -> **,\n"
352 " mount options=(rw,make-runbindable) -> **,\n"
353 "\n"
354 " # Allow all bind-mounts\n"
355 " mount options=(rw,bind),\n"
356 " mount options=(rw,rbind),\n"
357 "\n"
358 " # Allow remounting things read-only\n"
359 " mount options=(ro,remount),\n"
360 ;
361
362 static bool check_mount_feature_enabled(void)
363 {
364 return mount_features_enabled == 1;
365 }
366
367 static void load_mount_features_enabled(void)
368 {
369 struct stat statbuf;
370 int ret;
371
372 ret = stat(AA_MOUNT_RESTR, &statbuf);
373 if (ret == 0)
374 mount_features_enabled = 1;
375 }
376
377 /* aa_getcon is not working right now. Use our hand-rolled version below */
378 static int apparmor_enabled(void)
379 {
380 FILE *fin;
381 char e;
382 int ret;
383
384 fin = fopen_cloexec(AA_ENABLED_FILE, "r");
385 if (!fin)
386 return 0;
387 ret = fscanf(fin, "%c", &e);
388 fclose(fin);
389 if (ret == 1 && e == 'Y') {
390 load_mount_features_enabled();
391 return 1;
392 }
393
394 return 0;
395 }
396
397 static char *apparmor_process_label_get(pid_t pid)
398 {
399 char path[100], *space;
400 int ret;
401 char *buf = NULL, *newbuf;
402 int sz = 0;
403 FILE *f;
404
405 ret = snprintf(path, 100, "/proc/%d/attr/current", pid);
406 if (ret < 0 || ret >= 100) {
407 ERROR("path name too long");
408 return NULL;
409 }
410 again:
411 f = fopen_cloexec(path, "r");
412 if (!f) {
413 SYSERROR("opening %s", path);
414 free(buf);
415 return NULL;
416 }
417 sz += 1024;
418 newbuf = realloc(buf, sz);
419 if (!newbuf) {
420 free(buf);
421 ERROR("out of memory");
422 fclose(f);
423 return NULL;
424 }
425 buf = newbuf;
426 memset(buf, 0, sz);
427 ret = fread(buf, 1, sz - 1, f);
428 fclose(f);
429 if (ret < 0) {
430 ERROR("reading %s", path);
431 free(buf);
432 return NULL;
433 }
434 if (ret >= sz)
435 goto again;
436 space = strchr(buf, '\n');
437 if (space)
438 *space = '\0';
439 space = strchr(buf, ' ');
440 if (space)
441 *space = '\0';
442 return buf;
443 }
444
445 /*
446 * Probably makes sense to reorganize these to only read
447 * the label once
448 */
449 static bool apparmor_am_unconfined(void)
450 {
451 char *p = apparmor_process_label_get(lxc_raw_getpid());
452 bool ret = false;
453 if (!p || strcmp(p, "unconfined") == 0)
454 ret = true;
455 free(p);
456 return ret;
457 }
458
459 static bool aa_needs_transition(char *curlabel)
460 {
461 if (!curlabel)
462 return false;
463 if (strcmp(curlabel, "unconfined") == 0)
464 return false;
465 if (strcmp(curlabel, "/usr/bin/lxc-start") == 0)
466 return false;
467 return true;
468 }
469
470 static inline void uint64hex(char *buf, uint64_t num)
471 {
472 size_t i;
473
474 buf[16] = 0;
475 for (i = 16; i--;) {
476 char c = (char)(num & 0xf);
477 buf[i] = c + (c < 0xa ? '0' : 'a' - 0xa);
478 num >>= 4;
479 }
480 }
481
482 static inline char *shorten_apparmor_name(char *name)
483 {
484 size_t len = strlen(name);
485 if (len + 7 > 253) {
486 uint64_t hash;
487 hash = fnv_64a_buf(name, len, FNV1A_64_INIT);
488 name = must_realloc(name, 16 + 1);
489 uint64hex(name, hash);
490 }
491
492 return name;
493 }
494
495 /* Replace slashes with hyphens */
496 static inline void sanitize_path(char *path)
497 {
498 size_t i;
499
500 for (i = 0; path[i]; i++)
501 if (path[i] == '/')
502 path[i] = '-';
503 }
504
505 static inline char *apparmor_dir(const char *ctname, const char *lxcpath)
506 {
507 return must_make_path(lxcpath, ctname, "apparmor", NULL);
508 }
509
510
511 static inline char *apparmor_profile_full(const char *ctname, const char *lxcpath)
512 {
513 return shorten_apparmor_name(must_concat(NULL, "lxc-", ctname, "_<", lxcpath, ">", NULL));
514 }
515
516 /* Like apparmor_profile_full() but with slashes replaced by hyphens */
517 static inline char *apparmor_namespace(const char *ctname, const char *lxcpath)
518 {
519 char *full;
520
521 full = apparmor_profile_full(ctname, lxcpath);
522 sanitize_path(full);
523
524 return full;
525 }
526
527 /* FIXME: This is currently run only in the context of a constructor (via the
528 * initial lsm_init() called due to its __attribute__((constructor)), so we
529 * do not have ERROR/... macros available, so there are some fprintf(stderr)s
530 * in there.
531 */
532 static bool check_apparmor_parser_version()
533 {
534 struct lxc_popen_FILE *parserpipe;
535 int rc;
536 int major = 0, minor = 0, micro = 0;
537
538 parserpipe = lxc_popen("apparmor_parser --version");
539 if (!parserpipe) {
540 fprintf(stderr, "Failed to run check for apparmor_parser\n");
541 return false;
542 }
543
544 rc = fscanf(parserpipe->f, "AppArmor parser version %d.%d.%d", &major, &minor, &micro);
545 if (rc < 1) {
546 lxc_pclose(parserpipe);
547 /* We stay silent for now as this most likely means the shell
548 * lxc_popen executed failed to find the apparmor_parser binary.
549 * See the FIXME comment above for details.
550 */
551 return false;
552 }
553
554 rc = lxc_pclose(parserpipe);
555 if (rc < 0) {
556 fprintf(stderr, "Error waiting for child process\n");
557 return false;
558 }
559 if (rc != 0) {
560 fprintf(stderr, "'apparmor_parser --version' executed with an error status\n");
561 return false;
562 }
563
564 aa_supports_unix = (major > 2) ||
565 (major == 2 && minor > 10) ||
566 (major == 2 && minor == 10 && micro >= 95);
567
568 return true;
569 }
570
571 static bool file_is_yes(const char *path)
572 {
573 ssize_t rd;
574 int fd;
575 char buf[8]; /* we actually just expect "yes" or "no" */
576
577 fd = open(path, O_RDONLY | O_CLOEXEC);
578 if (fd < 0)
579 return false;
580
581 rd = lxc_read_nointr(fd, buf, sizeof(buf));
582 close(fd);
583
584 return rd >= 4 && strncmp(buf, "yes\n", 4) == 0;
585 }
586
587 static bool apparmor_can_stack()
588 {
589 int major, minor, scanned;
590 FILE *f;
591
592 if (!file_is_yes("/sys/kernel/security/apparmor/features/domain/stack"))
593 return false;
594
595 f = fopen_cloexec("/sys/kernel/security/apparmor/features/domain/version", "r");
596 if (!f)
597 return false;
598
599 scanned = fscanf(f, "%d.%d", &major, &minor);
600 fclose(f);
601 if (scanned != 2)
602 return false;
603
604 return major > 1 || (major == 1 && minor >= 2);
605 }
606
607 static void must_append_sized_full(char **buf, size_t *bufsz, const char *data,
608 size_t size, bool append_newline)
609 {
610 size_t newsize = *bufsz + size;
611
612 if (append_newline)
613 ++newsize;
614
615 *buf = must_realloc(*buf, newsize);
616 memcpy(*buf + *bufsz, data, size);
617
618 if (append_newline)
619 (*buf)[newsize - 1] = '\n';
620
621 *bufsz = newsize;
622 }
623
624 static void must_append_sized(char **buf, size_t *bufsz, const char *data, size_t size)
625 {
626 return must_append_sized_full(buf, bufsz, data, size, false);
627 }
628
629 static bool is_privileged(struct lxc_conf *conf)
630 {
631 return lxc_list_empty(&conf->id_map);
632 }
633
634 static char *get_apparmor_profile_content(struct lxc_conf *conf, const char *lxcpath)
635 {
636 char *profile, *profile_name_full;
637 size_t size;
638 struct lxc_list *it;
639
640 profile_name_full = apparmor_profile_full(conf->name, lxcpath);
641
642 profile = must_concat(NULL,
643 "#include <tunables/global>\n"
644 "profile \"", profile_name_full, "\" flags=(attach_disconnected,mediate_deleted) {\n",
645 NULL);
646 size = strlen(profile);
647
648 must_append_sized(&profile, &size, AA_PROFILE_BASE,
649 STRARRAYLEN(AA_PROFILE_BASE));
650
651 if (aa_supports_unix)
652 must_append_sized(&profile, &size, AA_PROFILE_UNIX_SOCKETS,
653 STRARRAYLEN(AA_PROFILE_UNIX_SOCKETS));
654
655 if (file_exists("/proc/self/ns/cgroup"))
656 must_append_sized(&profile, &size, AA_PROFILE_CGROUP_NAMESPACES,
657 STRARRAYLEN(AA_PROFILE_CGROUP_NAMESPACES));
658
659 if (aa_can_stack && !aa_is_stacked) {
660 char *namespace, *temp;
661
662 must_append_sized(&profile, &size, AA_PROFILE_STACKING_BASE,
663 STRARRAYLEN(AA_PROFILE_STACKING_BASE));
664
665 namespace = apparmor_namespace(conf->name, lxcpath);
666 temp = must_concat(NULL, " change_profile -> \":", namespace, ":*\",\n"
667 " change_profile -> \":", namespace, "://*\",\n",
668 NULL);
669 free(namespace);
670
671 must_append_sized(&profile, &size, temp, strlen(temp));
672 free(temp);
673 } else {
674 must_append_sized(&profile, &size, AA_PROFILE_NO_STACKING,
675 STRARRAYLEN(AA_PROFILE_NO_STACKING));
676 }
677
678 if (conf->lsm_aa_allow_nesting) {
679 must_append_sized(&profile, &size, AA_PROFILE_NESTING_BASE,
680 STRARRAYLEN(AA_PROFILE_NESTING_BASE));
681
682 if (!aa_can_stack || aa_is_stacked) {
683 char *temp;
684
685 temp = must_concat(NULL, " change_profile -> \"",
686 profile_name_full, "\",\n", NULL);
687 must_append_sized(&profile, &size, temp, strlen(temp));
688 free(temp);
689 }
690 }
691
692 if (!is_privileged(conf) || am_host_unpriv())
693 must_append_sized(&profile, &size, AA_PROFILE_UNPRIVILEGED,
694 STRARRAYLEN(AA_PROFILE_UNPRIVILEGED));
695
696 lxc_list_for_each(it, &conf->lsm_aa_raw) {
697 const char *line = it->elem;
698
699 must_append_sized_full(&profile, &size, line, strlen(line), true);
700 }
701
702 /* include terminating \0 byte */
703 must_append_sized(&profile, &size, "}\n", 3);
704
705 free(profile_name_full);
706
707 return profile;
708 }
709
710 /*
711 * apparmor_parser creates a cache file using the parsed file's name as a name.
712 * This means there may be multiple containers with the same name but different
713 * lxcpaths. Therefore we need a sanitized version of the complete profile name
714 * as profile file-name.
715 * We already get this exactly from apparmor_namespace().
716 */
717 static char *make_apparmor_profile_path(const char *ctname, const char *lxcpath)
718 {
719 char *ret, *filename;
720
721 filename = apparmor_namespace(ctname, lxcpath);
722 ret = must_make_path(lxcpath, ctname, "apparmor", filename, NULL);
723 free(filename);
724
725 return ret;
726 }
727
728 static char *make_apparmor_namespace_path(const char *ctname, const char *lxcpath)
729 {
730 char *ret, *namespace;
731
732 namespace = apparmor_namespace(ctname, lxcpath);
733 ret = must_make_path("/sys/kernel/security/apparmor/policy/namespaces", namespace, NULL);
734 free(namespace);
735
736 return ret;
737 }
738
739 static bool make_apparmor_namespace(struct lxc_conf *conf, const char *lxcpath)
740 {
741 char *path;
742
743 if (!aa_can_stack || aa_is_stacked)
744 return true;
745
746 path = make_apparmor_namespace_path(conf->name, lxcpath);
747 errno = 0;
748 if (mkdir(path, 0755) < 0 && errno != EEXIST) {
749 SYSERROR("Error creating AppArmor namespace: %s", path);
750 free(path);
751 return false;
752 }
753 free(path);
754
755 return true;
756 }
757
758 static void remove_apparmor_namespace(struct lxc_conf *conf, const char *lxcpath)
759 {
760 char *path;
761
762 path = make_apparmor_namespace_path(conf->name, lxcpath);
763 if (rmdir(path) != 0)
764 SYSERROR("Error removing AppArmor namespace");
765 free(path);
766 }
767
768 struct apparmor_parser_args {
769 char cmd;
770 char *file;
771 };
772
773 static int apparmor_parser_exec(void *data)
774 {
775 struct apparmor_parser_args *args = data;
776 char cmdbuf[] = { '-', args->cmd, 'W', 'L', 0 };
777
778 execlp("apparmor_parser", "apparmor_parser", cmdbuf, APPARMOR_CACHE_DIR, args->file, NULL);
779
780 return -1;
781 }
782
783 static int run_apparmor_parser(char command,
784 struct lxc_conf *conf,
785 const char *lxcpath)
786 {
787 char output[PATH_MAX];
788 int ret;
789 struct apparmor_parser_args args = {
790 .cmd = command,
791 .file = make_apparmor_profile_path(conf->name, lxcpath),
792 };
793
794 ret = run_command(output, sizeof(output), apparmor_parser_exec, (void*)&args);
795 if (ret < 0) {
796 ERROR("Failed to run apparmor_parser on \"%s\": %s", args.file, output);
797 ret = -1;
798 }
799
800
801 free(args.file);
802 return ret;
803 }
804
805 static void remove_apparmor_profile(struct lxc_conf *conf, const char *lxcpath)
806 {
807 char *path;
808
809 /* It's ok if these deletes fail: if the container was never started,
810 * we'll have never written a profile or cached it.
811 */
812
813 path = make_apparmor_profile_path(conf->name, lxcpath);
814 (void)unlink(path);
815 free(path);
816
817 /* Also remove the apparmor/ subdirectory */
818 path = apparmor_dir(conf->name, lxcpath);
819 (void)rmdir(path);
820 free(path);
821 }
822
823 static int load_apparmor_profile(struct lxc_conf *conf, const char *lxcpath)
824 {
825 struct stat profile_sb;
826 size_t content_len;
827 int ret = -1;
828 size_t old_len = 0;
829 char *profile_path = NULL, *old_content = NULL, *new_content = NULL;
830 int profile_fd = -1;
831
832 if (!make_apparmor_namespace(conf, lxcpath))
833 return -1;
834
835 /* In order to avoid forcing a profile parse (potentially slow) on
836 * every container start, let's use apparmor's binary policy cache,
837 * which checks mtime of the files to figure out if the policy needs to
838 * be regenerated.
839 *
840 * Since it uses mtimes, we shouldn't just always write out our local
841 * apparmor template; instead we should check to see whether the
842 * template is the same as ours. If it isn't we should write our
843 * version out so that the new changes are reflected and we definitely
844 * force a recompile.
845 */
846
847 profile_path = make_apparmor_profile_path(conf->name, lxcpath);
848 profile_fd = open(profile_path, O_RDONLY | O_CLOEXEC);
849 if (profile_fd >= 0) {
850 if (fstat(profile_fd, &profile_sb) < 0) {
851 SYSERROR("Error accessing old profile from %s",
852 profile_path);
853 goto out;
854 }
855 old_len = profile_sb.st_size;
856 old_content = lxc_strmmap(NULL, old_len, PROT_READ,
857 MAP_PRIVATE, profile_fd, 0);
858 if (!old_content) {
859 SYSERROR("Failed to mmap old profile from %s",
860 profile_path);
861 goto out;
862 }
863 } else if (errno != ENOENT) {
864 SYSERROR("Error reading old profile from %s", profile_path);
865 goto out;
866 }
867
868 new_content = get_apparmor_profile_content(conf, lxcpath);
869 if (!new_content)
870 goto out;
871
872 content_len = strlen(new_content);
873
874 if (!old_content || old_len != content_len || memcmp(old_content, new_content, content_len) != 0) {
875 char *path;
876
877 ret = mkdir_p(APPARMOR_CACHE_DIR, 0755);
878 if (ret < 0) {
879 SYSERROR("Error creating AppArmor profile cache directory " APPARMOR_CACHE_DIR);
880 goto out;
881 }
882
883 path = apparmor_dir(conf->name, lxcpath);
884 ret = mkdir_p(path, 0755);
885 if (ret < 0) {
886 SYSERROR("Error creating AppArmor profile directory: %s", path);
887 free(path);
888 goto out;
889 }
890 free(path);
891
892 ret = lxc_write_to_file(profile_path, new_content, content_len, false, 0600);
893 if (ret < 0) {
894 SYSERROR("Error writing profile to %s", profile_path);
895 goto out;
896 }
897 }
898
899 ret = run_apparmor_parser(AA_CMD_LOAD, conf, lxcpath);
900 if (ret != 0)
901 goto out_remove_profile;
902
903 conf->lsm_aa_profile_created = true;
904
905 goto out_ok;
906
907 out_remove_profile:
908 remove_apparmor_profile(conf, lxcpath);
909 out:
910 remove_apparmor_namespace(conf, lxcpath);
911 out_ok:
912 if (profile_fd >= 0) {
913 if (old_content)
914 lxc_strmunmap(old_content, old_len);
915 close(profile_fd);
916 }
917 free(profile_path);
918 free(new_content);
919 return ret;
920 }
921
922 /*
923 * Ensure that the container's policy namespace is unloaded to free kernel
924 * memory. This does not delete the policy from disk or cache.
925 */
926 static void apparmor_cleanup(struct lxc_conf *conf, const char *lxcpath)
927 {
928 if (!aa_admin)
929 return;
930
931 if (!conf->lsm_aa_profile_created)
932 return;
933
934 remove_apparmor_namespace(conf, lxcpath);
935 (void)run_apparmor_parser(AA_CMD_UNLOAD, conf, lxcpath);
936
937 remove_apparmor_profile(conf, lxcpath);
938 }
939
940 static int apparmor_prepare(struct lxc_conf *conf, const char *lxcpath)
941 {
942 int ret = -1;
943 const char *label;
944 char *curlabel = NULL, *genlabel = NULL;
945
946 if (!aa_enabled) {
947 ERROR("AppArmor not enabled");
948 return -1;
949 }
950
951 label = conf->lsm_aa_profile;
952
953 /* user may request that we just ignore apparmor */
954 if (label && strcmp(label, AA_UNCHANGED) == 0) {
955 INFO("AppArmor profile unchanged per user request");
956 conf->lsm_aa_profile_computed = must_copy_string(label);
957 return 0;
958 }
959
960 if (label && strcmp(label, AA_GENERATED) == 0) {
961 if (!aa_parser_available) {
962 ERROR("Cannot use generated profile: apparmor_parser not available");
963 goto out;
964 }
965
966 /* auto-generate profile based on available/requested security features */
967 if (load_apparmor_profile(conf, lxcpath) != 0) {
968 ERROR("Failed to load generated AppArmor profile");
969 goto out;
970 }
971
972 genlabel = apparmor_profile_full(conf->name, lxcpath);
973 if (!genlabel) {
974 ERROR("Failed to build AppArmor profile name");
975 goto out;
976 }
977
978 if (aa_can_stack && !aa_is_stacked) {
979 char *namespace = apparmor_namespace(conf->name, lxcpath);
980 size_t llen = strlen(genlabel);
981 must_append_sized(&genlabel, &llen, "//&:", STRARRAYLEN("//&:"));
982 must_append_sized(&genlabel, &llen, namespace, strlen(namespace));
983 must_append_sized(&genlabel, &llen, ":", STRARRAYLEN(":") + 1); /* with the nul byte */
984 free(namespace);
985 }
986
987 label = genlabel;
988 }
989
990 curlabel = apparmor_process_label_get(lxc_raw_getpid());
991
992 if (!aa_can_stack && aa_needs_transition(curlabel)) {
993 /* we're already confined, and stacking isn't supported */
994
995 if (!label || strcmp(curlabel, label) == 0) {
996 /* no change requested */
997 ret = 0;
998 goto out;
999 }
1000
1001 ERROR("Already AppArmor confined, but new label requested.");
1002 goto out;
1003 }
1004
1005 if (!label) {
1006 if (cgns_supported())
1007 label = AA_DEF_PROFILE_CGNS;
1008 else
1009 label = AA_DEF_PROFILE;
1010 }
1011
1012 if (!check_mount_feature_enabled() && strcmp(label, "unconfined") != 0) {
1013 WARN("Incomplete AppArmor support in your kernel");
1014 if (!conf->lsm_aa_allow_incomplete) {
1015 ERROR("If you really want to start this container, set");
1016 ERROR("lxc.apparmor.allow_incomplete = 1");
1017 ERROR("in your container configuration file");
1018 goto out;
1019 }
1020 }
1021
1022 conf->lsm_aa_profile_computed = must_copy_string(label);
1023 ret = 0;
1024
1025 out:
1026 if (genlabel) {
1027 free(genlabel);
1028 if (ret != 0)
1029 apparmor_cleanup(conf, lxcpath);
1030 }
1031 free(curlabel);
1032 return ret;
1033 }
1034
1035 /*
1036 * apparmor_process_label_set: Set AppArmor process profile
1037 *
1038 * @label : the profile to set
1039 * @conf : the container configuration to use if @label is NULL
1040 * @default : use the default profile if @label is NULL
1041 * @on_exec : this is ignored. Apparmor profile will be changed immediately
1042 *
1043 * Returns 0 on success, < 0 on failure
1044 *
1045 * Notes: This relies on /proc being available.
1046 */
1047 static int apparmor_process_label_set(const char *inlabel, struct lxc_conf *conf,
1048 bool on_exec)
1049 {
1050 int label_fd, ret;
1051 pid_t tid;
1052 const char *label;
1053
1054 if (!aa_enabled) {
1055 ERROR("AppArmor not enabled");
1056 return -1;
1057 }
1058
1059 label = inlabel ? inlabel : conf->lsm_aa_profile_computed;
1060 if (!label) {
1061 ERROR("LSM wasn't prepared");
1062 return -1;
1063 }
1064
1065 /* user may request that we just ignore apparmor */
1066 if (strcmp(label, AA_UNCHANGED) == 0) {
1067 INFO("AppArmor profile unchanged per user request");
1068 return 0;
1069 }
1070
1071 if (strcmp(label, "unconfined") == 0 && apparmor_am_unconfined()) {
1072 INFO("AppArmor profile unchanged");
1073 return 0;
1074 }
1075 tid = lxc_raw_gettid();
1076 label_fd = lsm_process_label_fd_get(tid, on_exec);
1077 if (label_fd < 0) {
1078 SYSERROR("Failed to change AppArmor profile to %s", label);
1079 return -1;
1080 }
1081
1082 ret = lsm_process_label_set_at(label_fd, label, on_exec);
1083 close(label_fd);
1084 if (ret < 0) {
1085 ERROR("Failed to change AppArmor profile to %s", label);
1086 return -1;
1087 }
1088
1089 INFO("Changed AppArmor profile to %s", label);
1090 return 0;
1091 }
1092
1093 static struct lsm_drv apparmor_drv = {
1094 .name = "AppArmor",
1095 .enabled = apparmor_enabled,
1096 .process_label_get = apparmor_process_label_get,
1097 .process_label_set = apparmor_process_label_set,
1098 .prepare = apparmor_prepare,
1099 .cleanup = apparmor_cleanup,
1100 };
1101
1102 struct lsm_drv *lsm_apparmor_drv_init(void)
1103 {
1104 bool have_mac_admin = false;
1105
1106 if (!apparmor_enabled())
1107 return NULL;
1108
1109 /* We only support generated profiles when apparmor_parser is usable */
1110 if (!check_apparmor_parser_version())
1111 goto out;
1112
1113 aa_parser_available = true;
1114
1115 aa_can_stack = apparmor_can_stack();
1116 if (aa_can_stack)
1117 aa_is_stacked = file_is_yes("/sys/kernel/security/apparmor/.ns_stacked");
1118
1119 #if HAVE_LIBCAP
1120 have_mac_admin = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE);
1121 #endif
1122
1123 if (!have_mac_admin)
1124 WARN("Per-container AppArmor profiles are disabled because the mac_admin capability is missing");
1125 else if (am_host_unpriv() && !aa_is_stacked)
1126 WARN("Per-container AppArmor profiles are disabled because LXC is running in an unprivileged container without stacking");
1127 else
1128 aa_admin = true;
1129
1130 out:
1131 aa_enabled = 1;
1132 return &apparmor_drv;
1133 }