src/basic/capability-util.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <errno.h>
   4 #include <stdio.h>
   5 #include <stdlib.h>
   6 #include <sys/prctl.h>
   7 #include <unistd.h>
   8
   9 #include "alloc-util.h"
  10 #include "capability-util.h"
  11 #include "cap-list.h"
  12 #include "fileio.h"
  13 #include "log.h"
  14 #include "macro.h"
  15 #include "missing_prctl.h"
  16 #include "parse-util.h"
  17 #include "user-util.h"
  18 #include "util.h"
  19
  20 int have_effective_cap(int value) {
  21         _cleanup_cap_free_ cap_t cap;
  22         cap_flag_value_t fv;
  23
  24         cap = cap_get_proc();
  25         if (!cap)
  26                 return -errno;
  27
  28         if (cap_get_flag(cap, value, CAP_EFFECTIVE, &fv) < 0)
  29                 return -errno;
  30
  31         return fv == CAP_SET;
  32 }
  33
  34 unsigned long cap_last_cap(void) {
  35         static thread_local unsigned long saved;
  36         static thread_local bool valid = false;
  37         _cleanup_free_ char *content = NULL;
  38         unsigned long p = 0;
  39         int r;
  40
  41         if (valid)
  42                 return saved;
  43
  44         /* available since linux-3.2 */
  45         r = read_one_line_file("/proc/sys/kernel/cap_last_cap", &content);
  46         if (r >= 0) {
  47                 r = safe_atolu(content, &p);
  48                 if (r >= 0) {
  49
  50                         if (p > 63) /* Safety for the future: if one day the kernel learns more than 64 caps,
  51                                      * then we are in trouble (since we, as much userspace and kernel space
  52                                      * store capability masks in uint64_t types). Let's hence protect
  53                                      * ourselves against that and always cap at 63 for now. */
  54                                 p = 63;
  55
  56                         saved = p;
  57                         valid = true;
  58                         return p;
  59                 }
  60         }
  61
  62         /* fall back to syscall-probing for pre linux-3.2 */
  63         p = MIN((unsigned long) CAP_LAST_CAP, 63U);
  64
  65         if (prctl(PR_CAPBSET_READ, p) < 0) {
  66
  67                 /* Hmm, look downwards, until we find one that works */
  68                 for (p--; p > 0; p --)
  69                         if (prctl(PR_CAPBSET_READ, p) >= 0)
  70                                 break;
  71
  72         } else {
  73
  74                 /* Hmm, look upwards, until we find one that doesn't work */
  75                 for (; p < 63; p++)
  76                         if (prctl(PR_CAPBSET_READ, p+1) < 0)
  77                                 break;
  78         }
  79
  80         saved = p;
  81         valid = true;
  82
  83         return p;
  84 }
  85
  86 int capability_update_inherited_set(cap_t caps, uint64_t set) {
  87         unsigned long i;
  88
  89         /* Add capabilities in the set to the inherited caps. Do not apply
  90          * them yet. */
  91
  92         for (i = 0; i <= cap_last_cap(); i++) {
  93
  94                 if (set & (UINT64_C(1) << i)) {
  95                         cap_value_t v;
  96
  97                         v = (cap_value_t) i;
  98
  99                         /* Make the capability inheritable. */
 100                         if (cap_set_flag(caps, CAP_INHERITABLE, 1, &v, CAP_SET) < 0)
 101                                 return -errno;
 102                 }
 103         }
 104
 105         return 0;
 106 }
 107
 108 int capability_ambient_set_apply(uint64_t set, bool also_inherit) {
 109         _cleanup_cap_free_ cap_t caps = NULL;
 110         unsigned long i;
 111         int r;
 112
 113         /* Add the capabilities to the ambient set. */
 114
 115         if (also_inherit) {
 116                 caps = cap_get_proc();
 117                 if (!caps)
 118                         return -errno;
 119
 120                 r = capability_update_inherited_set(caps, set);
 121                 if (r < 0)
 122                         return -errno;
 123
 124                 if (cap_set_proc(caps) < 0)
 125                         return -errno;
 126         }
 127
 128         for (i = 0; i <= cap_last_cap(); i++) {
 129
 130                 if (set & (UINT64_C(1) << i)) {
 131
 132                         /* Add the capability to the ambient set. */
 133                         if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_RAISE, i, 0, 0) < 0)
 134                                 return -errno;
 135                 }
 136         }
 137
 138         return 0;
 139 }
 140
 141 int capability_bounding_set_drop(uint64_t keep, bool right_now) {
 142         _cleanup_cap_free_ cap_t before_cap = NULL, after_cap = NULL;
 143         cap_flag_value_t fv;
 144         unsigned long i;
 145         int r;
 146
 147         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 148          * in the effective set (yes, the kernel drops that when
 149          * executing init!), so get it back temporarily so that we can
 150          * call PR_CAPBSET_DROP. */
 151
 152         before_cap = cap_get_proc();
 153         if (!before_cap)
 154                 return -errno;
 155
 156         if (cap_get_flag(before_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0)
 157                 return -errno;
 158
 159         if (fv != CAP_SET) {
 160                 _cleanup_cap_free_ cap_t temp_cap = NULL;
 161                 static const cap_value_t v = CAP_SETPCAP;
 162
 163                 temp_cap = cap_dup(before_cap);
 164                 if (!temp_cap)
 165                         return -errno;
 166
 167                 if (cap_set_flag(temp_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0)
 168                         return -errno;
 169
 170                 if (cap_set_proc(temp_cap) < 0)
 171                         log_debug_errno(errno, "Can't acquire effective CAP_SETPCAP bit, ignoring: %m");
 172
 173                 /* If we didn't manage to acquire the CAP_SETPCAP bit, we continue anyway, after all this just means
 174                  * we'll fail later, when we actually intend to drop some capabilities. */
 175         }
 176
 177         after_cap = cap_dup(before_cap);
 178         if (!after_cap)
 179                 return -errno;
 180
 181         for (i = 0; i <= cap_last_cap(); i++) {
 182                 cap_value_t v;
 183
 184                 if ((keep & (UINT64_C(1) << i)))
 185                         continue;
 186
 187                 /* Drop it from the bounding set */
 188                 if (prctl(PR_CAPBSET_DROP, i) < 0) {
 189                         r = -errno;
 190
 191                         /* If dropping the capability failed, let's see if we didn't have it in the first place. If so,
 192                          * continue anyway, as dropping a capability we didn't have in the first place doesn't really
 193                          * matter anyway. */
 194                         if (prctl(PR_CAPBSET_READ, i) != 0)
 195                                 goto finish;
 196                 }
 197                 v = (cap_value_t) i;
 198
 199                 /* Also drop it from the inheritable set, so
 200                  * that anything we exec() loses the
 201                  * capability for good. */
 202                 if (cap_set_flag(after_cap, CAP_INHERITABLE, 1, &v, CAP_CLEAR) < 0) {
 203                         r = -errno;
 204                         goto finish;
 205                 }
 206
 207                 /* If we shall apply this right now drop it
 208                  * also from our own capability sets. */
 209                 if (right_now) {
 210                         if (cap_set_flag(after_cap, CAP_PERMITTED, 1, &v, CAP_CLEAR) < 0 ||
 211                             cap_set_flag(after_cap, CAP_EFFECTIVE, 1, &v, CAP_CLEAR) < 0) {
 212                                 r = -errno;
 213                                 goto finish;
 214                         }
 215                 }
 216         }
 217
 218         r = 0;
 219
 220 finish:
 221         if (cap_set_proc(after_cap) < 0) {
 222                 /* If there are no actual changes anyway then let's ignore this error. */
 223                 if (cap_compare(before_cap, after_cap) != 0)
 224                         r = -errno;
 225         }
 226
 227         return r;
 228 }
 229
 230 static int drop_from_file(const char *fn, uint64_t keep) {
 231         _cleanup_free_ char *p = NULL;
 232         uint64_t current, after;
 233         uint32_t hi, lo;
 234         int r, k;
 235
 236         r = read_one_line_file(fn, &p);
 237         if (r < 0)
 238                 return r;
 239
 240         k = sscanf(p, "%" PRIu32 " %" PRIu32, &lo, &hi);
 241         if (k != 2)
 242                 return -EIO;
 243
 244         current = (uint64_t) lo | ((uint64_t) hi << 32);
 245         after = current & keep;
 246
 247         if (current == after)
 248                 return 0;
 249
 250         lo = after & UINT32_C(0xFFFFFFFF);
 251         hi = (after >> 32) & UINT32_C(0xFFFFFFFF);
 252
 253         return write_string_filef(fn, 0, "%" PRIu32 " %" PRIu32, lo, hi);
 254 }
 255
 256 int capability_bounding_set_drop_usermode(uint64_t keep) {
 257         int r;
 258
 259         r = drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", keep);
 260         if (r < 0)
 261                 return r;
 262
 263         r = drop_from_file("/proc/sys/kernel/usermodehelper/bset", keep);
 264         if (r < 0)
 265                 return r;
 266
 267         return r;
 268 }
 269
 270 int drop_privileges(uid_t uid, gid_t gid, uint64_t keep_capabilities) {
 271         int r;
 272
 273         /* Unfortunately we cannot leave privilege dropping to PID 1 here, since we want to run as user but
 274          * want to keep some capabilities. Since file capabilities have been introduced this cannot be done
 275          * across exec() anymore, unless our binary has the capability configured in the file system, which
 276          * we want to avoid. */
 277
 278         if (setresgid(gid, gid, gid) < 0)
 279                 return log_error_errno(errno, "Failed to change group ID: %m");
 280
 281         r = maybe_setgroups(0, NULL);
 282         if (r < 0)
 283                 return log_error_errno(r, "Failed to drop auxiliary groups list: %m");
 284
 285         /* Ensure we keep the permitted caps across the setresuid(). Note that we do this even if we actually
 286          * don't want to keep any capabilities, since we want to be able to drop them from the bounding set
 287          * too, and we can only do that if we have capabilities. */
 288         if (prctl(PR_SET_KEEPCAPS, 1) < 0)
 289                 return log_error_errno(errno, "Failed to enable keep capabilities flag: %m");
 290
 291         if (setresuid(uid, uid, uid) < 0)
 292                 return log_error_errno(errno, "Failed to change user ID: %m");
 293
 294         if (prctl(PR_SET_KEEPCAPS, 0) < 0)
 295                 return log_error_errno(errno, "Failed to disable keep capabilities flag: %m");
 296
 297         /* Drop all caps from the bounding set (as well as the inheritable/permitted/effective sets), except
 298          * the ones we want to keep */
 299         r = capability_bounding_set_drop(keep_capabilities, true);
 300         if (r < 0)
 301                 return log_error_errno(r, "Failed to drop capabilities: %m");
 302
 303         /* Now upgrade the permitted caps we still kept to effective caps */
 304         if (keep_capabilities != 0) {
 305                 cap_value_t bits[u64log2(keep_capabilities) + 1];
 306                 _cleanup_cap_free_ cap_t d = NULL;
 307                 unsigned i, j = 0;
 308
 309                 d = cap_init();
 310                 if (!d)
 311                         return log_oom();
 312
 313                 for (i = 0; i < ELEMENTSOF(bits); i++)
 314                         if (keep_capabilities & (1ULL << i))
 315                                 bits[j++] = i;
 316
 317                 /* use enough bits */
 318                 assert(i == 64 || (keep_capabilities >> i) == 0);
 319                 /* don't use too many bits */
 320                 assert(keep_capabilities & (UINT64_C(1) << (i - 1)));
 321
 322                 if (cap_set_flag(d, CAP_EFFECTIVE, j, bits, CAP_SET) < 0 ||
 323                     cap_set_flag(d, CAP_PERMITTED, j, bits, CAP_SET) < 0)
 324                         return log_error_errno(errno, "Failed to enable capabilities bits: %m");
 325
 326                 if (cap_set_proc(d) < 0)
 327                         return log_error_errno(errno, "Failed to increase capabilities: %m");
 328         }
 329
 330         return 0;
 331 }
 332
 333 int drop_capability(cap_value_t cv) {
 334         _cleanup_cap_free_ cap_t tmp_cap = NULL;
 335
 336         tmp_cap = cap_get_proc();
 337         if (!tmp_cap)
 338                 return -errno;
 339
 340         if ((cap_set_flag(tmp_cap, CAP_INHERITABLE, 1, &cv, CAP_CLEAR) < 0) ||
 341             (cap_set_flag(tmp_cap, CAP_PERMITTED, 1, &cv, CAP_CLEAR) < 0) ||
 342             (cap_set_flag(tmp_cap, CAP_EFFECTIVE, 1, &cv, CAP_CLEAR) < 0))
 343                 return -errno;
 344
 345         if (cap_set_proc(tmp_cap) < 0)
 346                 return -errno;
 347
 348         return 0;
 349 }
 350
 351 bool ambient_capabilities_supported(void) {
 352         static int cache = -1;
 353
 354         if (cache >= 0)
 355                 return cache;
 356
 357         /* If PR_CAP_AMBIENT returns something valid, or an unexpected error code we assume that ambient caps are
 358          * available. */
 359
 360         cache = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, CAP_KILL, 0, 0) >= 0 ||
 361                 !IN_SET(errno, EINVAL, EOPNOTSUPP, ENOSYS);
 362
 363         return cache;
 364 }
 365
 366 bool capability_quintet_mangle(CapabilityQuintet *q) {
 367         unsigned long i;
 368         uint64_t combined, drop = 0;
 369         bool ambient_supported;
 370
 371         assert(q);
 372
 373         combined = q->effective | q->bounding | q->inheritable | q->permitted;
 374
 375         ambient_supported = q->ambient != (uint64_t) -1;
 376         if (ambient_supported)
 377                 combined |= q->ambient;
 378
 379         for (i = 0; i <= cap_last_cap(); i++) {
 380                 unsigned long bit = UINT64_C(1) << i;
 381                 if (!FLAGS_SET(combined, bit))
 382                         continue;
 383
 384                 if (prctl(PR_CAPBSET_READ, i) > 0)
 385                         continue;
 386
 387                 drop |= bit;
 388
 389                 log_debug("Not in the current bounding set: %s", capability_to_name(i));
 390         }
 391
 392         q->effective &= ~drop;
 393         q->bounding &= ~drop;
 394         q->inheritable &= ~drop;
 395         q->permitted &= ~drop;
 396
 397         if (ambient_supported)
 398                 q->ambient &= ~drop;
 399
 400         return drop != 0; /* Let the caller know we changed something */
 401 }
 402
 403 int capability_quintet_enforce(const CapabilityQuintet *q) {
 404         _cleanup_cap_free_ cap_t c = NULL, modified = NULL;
 405         int r;
 406
 407         if (q->ambient != (uint64_t) -1) {
 408                 unsigned long i;
 409                 bool changed = false;
 410
 411                 c = cap_get_proc();
 412                 if (!c)
 413                         return -errno;
 414
 415                 /* In order to raise the ambient caps set we first need to raise the matching inheritable + permitted
 416                  * cap */
 417                 for (i = 0; i <= cap_last_cap(); i++) {
 418                         uint64_t m = UINT64_C(1) << i;
 419                         cap_value_t cv = (cap_value_t) i;
 420                         cap_flag_value_t old_value_inheritable, old_value_permitted;
 421
 422                         if ((q->ambient & m) == 0)
 423                                 continue;
 424
 425                         if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value_inheritable) < 0)
 426                                 return -errno;
 427                         if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value_permitted) < 0)
 428                                 return -errno;
 429
 430                         if (old_value_inheritable == CAP_SET && old_value_permitted == CAP_SET)
 431                                 continue;
 432
 433                         if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, CAP_SET) < 0)
 434                                 return -errno;
 435                         if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
 436                                 return -errno;
 437
 438                         changed = true;
 439                 }
 440
 441                 if (changed)
 442                         if (cap_set_proc(c) < 0)
 443                                 return -errno;
 444
 445                 r = capability_ambient_set_apply(q->ambient, false);
 446                 if (r < 0)
 447                         return r;
 448         }
 449
 450         if (q->inheritable != (uint64_t) -1 || q->permitted != (uint64_t) -1 || q->effective != (uint64_t) -1) {
 451                 bool changed = false;
 452                 unsigned long i;
 453
 454                 if (!c) {
 455                         c = cap_get_proc();
 456                         if (!c)
 457                                 return -errno;
 458                 }
 459
 460                 for (i = 0; i <= cap_last_cap(); i++) {
 461                         uint64_t m = UINT64_C(1) << i;
 462                         cap_value_t cv = (cap_value_t) i;
 463
 464                         if (q->inheritable != (uint64_t) -1) {
 465                                 cap_flag_value_t old_value, new_value;
 466
 467                                 if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value) < 0) {
 468                                         if (errno == EINVAL) /* If the kernel knows more caps than this
 469                                                               * version of libcap, then this will return
 470                                                               * EINVAL. In that case, simply ignore it,
 471                                                               * pretend it doesn't exist. */
 472                                                 continue;
 473
 474                                         return -errno;
 475                                 }
 476
 477                                 new_value = (q->inheritable & m) ? CAP_SET : CAP_CLEAR;
 478
 479                                 if (old_value != new_value) {
 480                                         changed = true;
 481
 482                                         if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, new_value) < 0)
 483                                                 return -errno;
 484                                 }
 485                         }
 486
 487                         if (q->permitted != (uint64_t) -1) {
 488                                 cap_flag_value_t old_value, new_value;
 489
 490                                 if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value) < 0) {
 491                                         if (errno == EINVAL)
 492                                                 continue;
 493
 494                                         return -errno;
 495                                 }
 496
 497                                 new_value = (q->permitted & m) ? CAP_SET : CAP_CLEAR;
 498
 499                                 if (old_value != new_value) {
 500                                         changed = true;
 501
 502                                         if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, new_value) < 0)
 503                                                 return -errno;
 504                                 }
 505                         }
 506
 507                         if (q->effective != (uint64_t) -1) {
 508                                 cap_flag_value_t old_value, new_value;
 509
 510                                 if (cap_get_flag(c, cv, CAP_EFFECTIVE, &old_value) < 0) {
 511                                         if (errno == EINVAL)
 512                                                 continue;
 513
 514                                         return -errno;
 515                                 }
 516
 517                                 new_value = (q->effective & m) ? CAP_SET : CAP_CLEAR;
 518
 519                                 if (old_value != new_value) {
 520                                         changed = true;
 521
 522                                         if (cap_set_flag(c, CAP_EFFECTIVE, 1, &cv, new_value) < 0)
 523                                                 return -errno;
 524                                 }
 525                         }
 526                 }
 527
 528                 if (changed) {
 529                         /* In order to change the bounding caps, we need to keep CAP_SETPCAP for a bit
 530                          * longer. Let's add it to our list hence for now. */
 531                         if (q->bounding != (uint64_t) -1) {
 532                                 cap_value_t cv = CAP_SETPCAP;
 533
 534                                 modified = cap_dup(c);
 535                                 if (!modified)
 536                                         return -ENOMEM;
 537
 538                                 if (cap_set_flag(modified, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
 539                                         return -errno;
 540                                 if (cap_set_flag(modified, CAP_EFFECTIVE, 1, &cv, CAP_SET) < 0)
 541                                         return -errno;
 542
 543                                 if (cap_compare(modified, c) == 0) {
 544                                         /* No change? then drop this nonsense again */
 545                                         cap_free(modified);
 546                                         modified = NULL;
 547                                 }
 548                         }
 549
 550                         /* Now, let's enforce the caps for the first time. Note that this is where we acquire
 551                          * caps in any of the sets we currently don't have. We have to do this before
 552                          * dropping the bounding caps below, since at that point we can never acquire new
 553                          * caps in inherited/permitted/effective anymore, but only lose them. */
 554                         if (cap_set_proc(modified ?: c) < 0)
 555                                 return -errno;
 556                 }
 557         }
 558
 559         if (q->bounding != (uint64_t) -1) {
 560                 r = capability_bounding_set_drop(q->bounding, false);
 561                 if (r < 0)
 562                         return r;
 563         }
 564
 565         /* If needed, let's now set the caps again, this time in the final version, which differs from what
 566          * we have already set only in the CAP_SETPCAP bit, which we needed for dropping the bounding
 567          * bits. This call only undoes bits and doesn't acquire any which means the bounding caps don't
 568          * matter. */
 569         if (modified)
 570                 if (cap_set_proc(c) < 0)
 571                         return -errno;
 572
 573         return 0;
 574 }