src/basic/capability-util.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <errno.h>
   4 #include <grp.h>
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <sys/capability.h>
   8 #include <sys/prctl.h>
   9 #include <unistd.h>
  10
  11 #include "alloc-util.h"
  12 #include "capability-util.h"
  13 #include "fileio.h"
  14 #include "log.h"
  15 #include "macro.h"
  16 #include "missing_prctl.h"
  17 #include "parse-util.h"
  18 #include "user-util.h"
  19 #include "util.h"
  20
  21 int have_effective_cap(int value) {
  22         _cleanup_cap_free_ cap_t cap;
  23         cap_flag_value_t fv;
  24
  25         cap = cap_get_proc();
  26         if (!cap)
  27                 return -errno;
  28
  29         if (cap_get_flag(cap, value, CAP_EFFECTIVE, &fv) < 0)
  30                 return -errno;
  31
  32         return fv == CAP_SET;
  33 }
  34
  35 unsigned long cap_last_cap(void) {
  36         static thread_local unsigned long saved;
  37         static thread_local bool valid = false;
  38         _cleanup_free_ char *content = NULL;
  39         unsigned long p = 0;
  40         int r;
  41
  42         if (valid)
  43                 return saved;
  44
  45         /* available since linux-3.2 */
  46         r = read_one_line_file("/proc/sys/kernel/cap_last_cap", &content);
  47         if (r >= 0) {
  48                 r = safe_atolu(content, &p);
  49                 if (r >= 0) {
  50
  51                         if (p > 63) /* Safety for the future: if one day the kernel learns more than 64 caps,
  52                                      * then we are in trouble (since we, as much userspace and kernel space
  53                                      * store capability masks in uint64_t types). Let's hence protect
  54                                      * ourselves against that and always cap at 63 for now. */
  55                                 p = 63;
  56
  57                         saved = p;
  58                         valid = true;
  59                         return p;
  60                 }
  61         }
  62
  63         /* fall back to syscall-probing for pre linux-3.2 */
  64         p = MIN((unsigned long) CAP_LAST_CAP, 63U);
  65
  66         if (prctl(PR_CAPBSET_READ, p) < 0) {
  67
  68                 /* Hmm, look downwards, until we find one that works */
  69                 for (p--; p > 0; p --)
  70                         if (prctl(PR_CAPBSET_READ, p) >= 0)
  71                                 break;
  72
  73         } else {
  74
  75                 /* Hmm, look upwards, until we find one that doesn't work */
  76                 for (; p < 63; p++)
  77                         if (prctl(PR_CAPBSET_READ, p+1) < 0)
  78                                 break;
  79         }
  80
  81         saved = p;
  82         valid = true;
  83
  84         return p;
  85 }
  86
  87 int capability_update_inherited_set(cap_t caps, uint64_t set) {
  88         unsigned long i;
  89
  90         /* Add capabilities in the set to the inherited caps. Do not apply
  91          * them yet. */
  92
  93         for (i = 0; i < cap_last_cap(); i++) {
  94
  95                 if (set & (UINT64_C(1) << i)) {
  96                         cap_value_t v;
  97
  98                         v = (cap_value_t) i;
  99
 100                         /* Make the capability inheritable. */
 101                         if (cap_set_flag(caps, CAP_INHERITABLE, 1, &v, CAP_SET) < 0)
 102                                 return -errno;
 103                 }
 104         }
 105
 106         return 0;
 107 }
 108
 109 int capability_ambient_set_apply(uint64_t set, bool also_inherit) {
 110         _cleanup_cap_free_ cap_t caps = NULL;
 111         unsigned long i;
 112         int r;
 113
 114         /* Add the capabilities to the ambient set. */
 115
 116         if (also_inherit) {
 117                 caps = cap_get_proc();
 118                 if (!caps)
 119                         return -errno;
 120
 121                 r = capability_update_inherited_set(caps, set);
 122                 if (r < 0)
 123                         return -errno;
 124
 125                 if (cap_set_proc(caps) < 0)
 126                         return -errno;
 127         }
 128
 129         for (i = 0; i < cap_last_cap(); i++) {
 130
 131                 if (set & (UINT64_C(1) << i)) {
 132
 133                         /* Add the capability to the ambient set. */
 134                         if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_RAISE, i, 0, 0) < 0)
 135                                 return -errno;
 136                 }
 137         }
 138
 139         return 0;
 140 }
 141
 142 int capability_bounding_set_drop(uint64_t keep, bool right_now) {
 143         _cleanup_cap_free_ cap_t before_cap = NULL, after_cap = NULL;
 144         cap_flag_value_t fv;
 145         unsigned long i;
 146         int r;
 147
 148         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 149          * in the effective set (yes, the kernel drops that when
 150          * executing init!), so get it back temporarily so that we can
 151          * call PR_CAPBSET_DROP. */
 152
 153         before_cap = cap_get_proc();
 154         if (!before_cap)
 155                 return -errno;
 156
 157         if (cap_get_flag(before_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0)
 158                 return -errno;
 159
 160         if (fv != CAP_SET) {
 161                 _cleanup_cap_free_ cap_t temp_cap = NULL;
 162                 static const cap_value_t v = CAP_SETPCAP;
 163
 164                 temp_cap = cap_dup(before_cap);
 165                 if (!temp_cap)
 166                         return -errno;
 167
 168                 if (cap_set_flag(temp_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0)
 169                         return -errno;
 170
 171                 if (cap_set_proc(temp_cap) < 0)
 172                         log_debug_errno(errno, "Can't acquire effective CAP_SETPCAP bit, ignoring: %m");
 173
 174                 /* If we didn't manage to acquire the CAP_SETPCAP bit, we continue anyway, after all this just means
 175                  * we'll fail later, when we actually intend to drop some capabilities. */
 176         }
 177
 178         after_cap = cap_dup(before_cap);
 179         if (!after_cap)
 180                 return -errno;
 181
 182         for (i = 0; i <= cap_last_cap(); i++) {
 183                 cap_value_t v;
 184
 185                 if ((keep & (UINT64_C(1) << i)))
 186                         continue;
 187
 188                 /* Drop it from the bounding set */
 189                 if (prctl(PR_CAPBSET_DROP, i) < 0) {
 190                         r = -errno;
 191
 192                         /* If dropping the capability failed, let's see if we didn't have it in the first place. If so,
 193                          * continue anyway, as dropping a capability we didn't have in the first place doesn't really
 194                          * matter anyway. */
 195                         if (prctl(PR_CAPBSET_READ, i) != 0)
 196                                 goto finish;
 197                 }
 198                 v = (cap_value_t) i;
 199
 200                 /* Also drop it from the inheritable set, so
 201                  * that anything we exec() loses the
 202                  * capability for good. */
 203                 if (cap_set_flag(after_cap, CAP_INHERITABLE, 1, &v, CAP_CLEAR) < 0) {
 204                         r = -errno;
 205                         goto finish;
 206                 }
 207
 208                 /* If we shall apply this right now drop it
 209                  * also from our own capability sets. */
 210                 if (right_now) {
 211                         if (cap_set_flag(after_cap, CAP_PERMITTED, 1, &v, CAP_CLEAR) < 0 ||
 212                             cap_set_flag(after_cap, CAP_EFFECTIVE, 1, &v, CAP_CLEAR) < 0) {
 213                                 r = -errno;
 214                                 goto finish;
 215                         }
 216                 }
 217         }
 218
 219         r = 0;
 220
 221 finish:
 222         if (cap_set_proc(after_cap) < 0) {
 223                 /* If there are no actual changes anyway then let's ignore this error. */
 224                 if (cap_compare(before_cap, after_cap) != 0)
 225                         r = -errno;
 226         }
 227
 228         return r;
 229 }
 230
 231 static int drop_from_file(const char *fn, uint64_t keep) {
 232         _cleanup_free_ char *p = NULL;
 233         uint64_t current, after;
 234         uint32_t hi, lo;
 235         int r, k;
 236
 237         r = read_one_line_file(fn, &p);
 238         if (r < 0)
 239                 return r;
 240
 241         k = sscanf(p, "%" PRIu32 " %" PRIu32, &lo, &hi);
 242         if (k != 2)
 243                 return -EIO;
 244
 245         current = (uint64_t) lo | ((uint64_t) hi << 32);
 246         after = current & keep;
 247
 248         if (current == after)
 249                 return 0;
 250
 251         lo = after & UINT32_C(0xFFFFFFFF);
 252         hi = (after >> 32) & UINT32_C(0xFFFFFFFF);
 253
 254         return write_string_filef(fn, 0, "%" PRIu32 " %" PRIu32, lo, hi);
 255 }
 256
 257 int capability_bounding_set_drop_usermode(uint64_t keep) {
 258         int r;
 259
 260         r = drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", keep);
 261         if (r < 0)
 262                 return r;
 263
 264         r = drop_from_file("/proc/sys/kernel/usermodehelper/bset", keep);
 265         if (r < 0)
 266                 return r;
 267
 268         return r;
 269 }
 270
 271 int drop_privileges(uid_t uid, gid_t gid, uint64_t keep_capabilities) {
 272         int r;
 273
 274         /* Unfortunately we cannot leave privilege dropping to PID 1 here, since we want to run as user but
 275          * want to keep some capabilities. Since file capabilities have been introduced this cannot be done
 276          * across exec() anymore, unless our binary has the capability configured in the file system, which
 277          * we want to avoid. */
 278
 279         if (setresgid(gid, gid, gid) < 0)
 280                 return log_error_errno(errno, "Failed to change group ID: %m");
 281
 282         r = maybe_setgroups(0, NULL);
 283         if (r < 0)
 284                 return log_error_errno(r, "Failed to drop auxiliary groups list: %m");
 285
 286         /* Ensure we keep the permitted caps across the setresuid(). Note that we do this even if we actually
 287          * don't want to keep any capabilities, since we want to be able to drop them from the bounding set
 288          * too, and we can only do that if we have capabilities. */
 289         if (prctl(PR_SET_KEEPCAPS, 1) < 0)
 290                 return log_error_errno(errno, "Failed to enable keep capabilities flag: %m");
 291
 292         if (setresuid(uid, uid, uid) < 0)
 293                 return log_error_errno(errno, "Failed to change user ID: %m");
 294
 295         if (prctl(PR_SET_KEEPCAPS, 0) < 0)
 296                 return log_error_errno(errno, "Failed to disable keep capabilities flag: %m");
 297
 298         /* Drop all caps from the bounding set (as well as the inheritable/permitted/effective sets), except
 299          * the ones we want to keep */
 300         r = capability_bounding_set_drop(keep_capabilities, true);
 301         if (r < 0)
 302                 return log_error_errno(r, "Failed to drop capabilities: %m");
 303
 304         /* Now upgrade the permitted caps we still kept to effective caps */
 305         if (keep_capabilities != 0) {
 306                 cap_value_t bits[u64log2(keep_capabilities) + 1];
 307                 _cleanup_cap_free_ cap_t d = NULL;
 308                 unsigned i, j = 0;
 309
 310                 d = cap_init();
 311                 if (!d)
 312                         return log_oom();
 313
 314                 for (i = 0; i < ELEMENTSOF(bits); i++)
 315                         if (keep_capabilities & (1ULL << i))
 316                                 bits[j++] = i;
 317
 318                 /* use enough bits */
 319                 assert(i == 64 || (keep_capabilities >> i) == 0);
 320                 /* don't use too many bits */
 321                 assert(keep_capabilities & (UINT64_C(1) << (i - 1)));
 322
 323                 if (cap_set_flag(d, CAP_EFFECTIVE, j, bits, CAP_SET) < 0 ||
 324                     cap_set_flag(d, CAP_PERMITTED, j, bits, CAP_SET) < 0)
 325                         return log_error_errno(errno, "Failed to enable capabilities bits: %m");
 326
 327                 if (cap_set_proc(d) < 0)
 328                         return log_error_errno(errno, "Failed to increase capabilities: %m");
 329         }
 330
 331         return 0;
 332 }
 333
 334 int drop_capability(cap_value_t cv) {
 335         _cleanup_cap_free_ cap_t tmp_cap = NULL;
 336
 337         tmp_cap = cap_get_proc();
 338         if (!tmp_cap)
 339                 return -errno;
 340
 341         if ((cap_set_flag(tmp_cap, CAP_INHERITABLE, 1, &cv, CAP_CLEAR) < 0) ||
 342             (cap_set_flag(tmp_cap, CAP_PERMITTED, 1, &cv, CAP_CLEAR) < 0) ||
 343             (cap_set_flag(tmp_cap, CAP_EFFECTIVE, 1, &cv, CAP_CLEAR) < 0))
 344                 return -errno;
 345
 346         if (cap_set_proc(tmp_cap) < 0)
 347                 return -errno;
 348
 349         return 0;
 350 }
 351
 352 bool ambient_capabilities_supported(void) {
 353         static int cache = -1;
 354
 355         if (cache >= 0)
 356                 return cache;
 357
 358         /* If PR_CAP_AMBIENT returns something valid, or an unexpected error code we assume that ambient caps are
 359          * available. */
 360
 361         cache = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, CAP_KILL, 0, 0) >= 0 ||
 362                 !IN_SET(errno, EINVAL, EOPNOTSUPP, ENOSYS);
 363
 364         return cache;
 365 }
 366
 367 int capability_quintet_enforce(const CapabilityQuintet *q) {
 368         _cleanup_cap_free_ cap_t c = NULL, modified = NULL;
 369         int r;
 370
 371         if (q->ambient != (uint64_t) -1) {
 372                 unsigned long i;
 373                 bool changed = false;
 374
 375                 c = cap_get_proc();
 376                 if (!c)
 377                         return -errno;
 378
 379                 /* In order to raise the ambient caps set we first need to raise the matching inheritable + permitted
 380                  * cap */
 381                 for (i = 0; i <= cap_last_cap(); i++) {
 382                         uint64_t m = UINT64_C(1) << i;
 383                         cap_value_t cv = (cap_value_t) i;
 384                         cap_flag_value_t old_value_inheritable, old_value_permitted;
 385
 386                         if ((q->ambient & m) == 0)
 387                                 continue;
 388
 389                         if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value_inheritable) < 0)
 390                                 return -errno;
 391                         if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value_permitted) < 0)
 392                                 return -errno;
 393
 394                         if (old_value_inheritable == CAP_SET && old_value_permitted == CAP_SET)
 395                                 continue;
 396
 397                         if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, CAP_SET) < 0)
 398                                 return -errno;
 399                         if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
 400                                 return -errno;
 401
 402                         changed = true;
 403                 }
 404
 405                 if (changed)
 406                         if (cap_set_proc(c) < 0)
 407                                 return -errno;
 408
 409                 r = capability_ambient_set_apply(q->ambient, false);
 410                 if (r < 0)
 411                         return r;
 412         }
 413
 414         if (q->inheritable != (uint64_t) -1 || q->permitted != (uint64_t) -1 || q->effective != (uint64_t) -1) {
 415                 bool changed = false;
 416                 unsigned long i;
 417
 418                 if (!c) {
 419                         c = cap_get_proc();
 420                         if (!c)
 421                                 return -errno;
 422                 }
 423
 424                 for (i = 0; i <= cap_last_cap(); i++) {
 425                         uint64_t m = UINT64_C(1) << i;
 426                         cap_value_t cv = (cap_value_t) i;
 427
 428                         if (q->inheritable != (uint64_t) -1) {
 429                                 cap_flag_value_t old_value, new_value;
 430
 431                                 if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value) < 0) {
 432                                         if (errno == EINVAL) /* If the kernel knows more caps than this
 433                                                               * version of libcap, then this will return
 434                                                               * EINVAL. In that case, simply ignore it,
 435                                                               * pretend it doesn't exist. */
 436                                                 continue;
 437
 438                                         return -errno;
 439                                 }
 440
 441                                 new_value = (q->inheritable & m) ? CAP_SET : CAP_CLEAR;
 442
 443                                 if (old_value != new_value) {
 444                                         changed = true;
 445
 446                                         if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, new_value) < 0)
 447                                                 return -errno;
 448                                 }
 449                         }
 450
 451                         if (q->permitted != (uint64_t) -1) {
 452                                 cap_flag_value_t old_value, new_value;
 453
 454                                 if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value) < 0) {
 455                                         if (errno == EINVAL)
 456                                                 continue;
 457
 458                                         return -errno;
 459                                 }
 460
 461                                 new_value = (q->permitted & m) ? CAP_SET : CAP_CLEAR;
 462
 463                                 if (old_value != new_value) {
 464                                         changed = true;
 465
 466                                         if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, new_value) < 0)
 467                                                 return -errno;
 468                                 }
 469                         }
 470
 471                         if (q->effective != (uint64_t) -1) {
 472                                 cap_flag_value_t old_value, new_value;
 473
 474                                 if (cap_get_flag(c, cv, CAP_EFFECTIVE, &old_value) < 0) {
 475                                         if (errno == EINVAL)
 476                                                 continue;
 477
 478                                         return -errno;
 479                                 }
 480
 481                                 new_value = (q->effective & m) ? CAP_SET : CAP_CLEAR;
 482
 483                                 if (old_value != new_value) {
 484                                         changed = true;
 485
 486                                         if (cap_set_flag(c, CAP_EFFECTIVE, 1, &cv, new_value) < 0)
 487                                                 return -errno;
 488                                 }
 489                         }
 490                 }
 491
 492                 if (changed) {
 493                         /* In order to change the bounding caps, we need to keep CAP_SETPCAP for a bit
 494                          * longer. Let's add it to our list hence for now. */
 495                         if (q->bounding != (uint64_t) -1) {
 496                                 cap_value_t cv = CAP_SETPCAP;
 497
 498                                 modified = cap_dup(c);
 499                                 if (!modified)
 500                                         return -ENOMEM;
 501
 502                                 if (cap_set_flag(modified, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
 503                                         return -errno;
 504                                 if (cap_set_flag(modified, CAP_EFFECTIVE, 1, &cv, CAP_SET) < 0)
 505                                         return -errno;
 506
 507                                 if (cap_compare(modified, c) == 0) {
 508                                         /* No change? then drop this nonsense again */
 509                                         cap_free(modified);
 510                                         modified = NULL;
 511                                 }
 512                         }
 513
 514                         /* Now, let's enforce the caps for the first time. Note that this is where we acquire
 515                          * caps in any of the sets we currently don't have. We have to do this before
 516                          * dropping the bounding caps below, since at that point we can never acquire new
 517                          * caps in inherited/permitted/effective anymore, but only lose them. */
 518                         if (cap_set_proc(modified ?: c) < 0)
 519                                 return -errno;
 520                 }
 521         }
 522
 523         if (q->bounding != (uint64_t) -1) {
 524                 r = capability_bounding_set_drop(q->bounding, false);
 525                 if (r < 0)
 526                         return r;
 527         }
 528
 529         /* If needed, let's now set the caps again, this time in the final version, which differs from what
 530          * we have already set only in the CAP_SETPCAP bit, which we needed for dropping the bounding
 531          * bits. This call only undoes bits and doesn't acquire any which means the bounding caps don't
 532          * matter. */
 533         if (modified)
 534                 if (cap_set_proc(c) < 0)
 535                         return -errno;
 536
 537         return 0;
 538 }