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, drops capabilities not in the set.
  90          * Do not apply them yet. */
  91
  92         for (i = 0; i <= cap_last_cap(); i++) {
  93                 cap_flag_value_t flag = set & (UINT64_C(1) << i) ? CAP_SET : CAP_CLEAR;
  94                 cap_value_t v;
  95
  96                 v = (cap_value_t) i;
  97
  98                 if (cap_set_flag(caps, CAP_INHERITABLE, 1, &v, flag) < 0)
  99                         return -errno;
 100         }
 101
 102         return 0;
 103 }
 104
 105 int capability_ambient_set_apply(uint64_t set, bool also_inherit) {
 106         _cleanup_cap_free_ cap_t caps = NULL;
 107         unsigned long i;
 108         int r;
 109
 110         /* Remove capabilities requested in ambient set, but not in the bounding set */
 111         for (i = 0; i <= cap_last_cap(); i++) {
 112                 if (set == 0)
 113                         break;
 114
 115                 if (FLAGS_SET(set, (UINT64_C(1) << i)) && prctl(PR_CAPBSET_READ, i) != 1) {
 116                         log_debug("Ambient capability %s requested but missing from bounding set,"
 117                                         " suppressing automatically.", capability_to_name(i));
 118                         set &= ~(UINT64_C(1) << i);
 119                 }
 120         }
 121
 122         /* Add the capabilities to the ambient set (an possibly also the inheritable set) */
 123
 124         /* Check that we can use PR_CAP_AMBIENT or quit early. */
 125         if (!ambient_capabilities_supported())
 126                 return (set & all_capabilities()) == 0 ?
 127                         0 : -EOPNOTSUPP; /* if actually no ambient caps are to be set, be silent,
 128                                           * otherwise fail recognizably */
 129
 130         if (also_inherit) {
 131                 caps = cap_get_proc();
 132                 if (!caps)
 133                         return -errno;
 134
 135                 r = capability_update_inherited_set(caps, set);
 136                 if (r < 0)
 137                         return -errno;
 138
 139                 if (cap_set_proc(caps) < 0)
 140                         return -errno;
 141         }
 142
 143         for (i = 0; i <= cap_last_cap(); i++) {
 144
 145                 if (set & (UINT64_C(1) << i)) {
 146
 147                         /* Add the capability to the ambient set. */
 148                         if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_RAISE, i, 0, 0) < 0)
 149                                 return -errno;
 150                 } else {
 151
 152                         /* Drop the capability so we don't inherit capabilities we didn't ask for. */
 153                         r = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, i, 0, 0);
 154                         if (r < 0)
 155                                 return -errno;
 156
 157                         if (r)
 158                                 if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_LOWER, i, 0, 0) < 0)
 159                                         return -errno;
 160
 161                 }
 162         }
 163
 164         return 0;
 165 }
 166
 167 int capability_bounding_set_drop(uint64_t keep, bool right_now) {
 168         _cleanup_cap_free_ cap_t before_cap = NULL, after_cap = NULL;
 169         cap_flag_value_t fv;
 170         unsigned long i;
 171         int r;
 172
 173         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 174          * in the effective set (yes, the kernel drops that when
 175          * executing init!), so get it back temporarily so that we can
 176          * call PR_CAPBSET_DROP. */
 177
 178         before_cap = cap_get_proc();
 179         if (!before_cap)
 180                 return -errno;
 181
 182         if (cap_get_flag(before_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0)
 183                 return -errno;
 184
 185         if (fv != CAP_SET) {
 186                 _cleanup_cap_free_ cap_t temp_cap = NULL;
 187                 static const cap_value_t v = CAP_SETPCAP;
 188
 189                 temp_cap = cap_dup(before_cap);
 190                 if (!temp_cap)
 191                         return -errno;
 192
 193                 if (cap_set_flag(temp_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0)
 194                         return -errno;
 195
 196                 if (cap_set_proc(temp_cap) < 0)
 197                         log_debug_errno(errno, "Can't acquire effective CAP_SETPCAP bit, ignoring: %m");
 198
 199                 /* If we didn't manage to acquire the CAP_SETPCAP bit, we continue anyway, after all this just means
 200                  * we'll fail later, when we actually intend to drop some capabilities. */
 201         }
 202
 203         after_cap = cap_dup(before_cap);
 204         if (!after_cap)
 205                 return -errno;
 206
 207         for (i = 0; i <= cap_last_cap(); i++) {
 208                 cap_value_t v;
 209
 210                 if ((keep & (UINT64_C(1) << i)))
 211                         continue;
 212
 213                 /* Drop it from the bounding set */
 214                 if (prctl(PR_CAPBSET_DROP, i) < 0) {
 215                         r = -errno;
 216
 217                         /* If dropping the capability failed, let's see if we didn't have it in the first place. If so,
 218                          * continue anyway, as dropping a capability we didn't have in the first place doesn't really
 219                          * matter anyway. */
 220                         if (prctl(PR_CAPBSET_READ, i) != 0)
 221                                 goto finish;
 222                 }
 223                 v = (cap_value_t) i;
 224
 225                 /* Also drop it from the inheritable set, so
 226                  * that anything we exec() loses the
 227                  * capability for good. */
 228                 if (cap_set_flag(after_cap, CAP_INHERITABLE, 1, &v, CAP_CLEAR) < 0) {
 229                         r = -errno;
 230                         goto finish;
 231                 }
 232
 233                 /* If we shall apply this right now drop it
 234                  * also from our own capability sets. */
 235                 if (right_now) {
 236                         if (cap_set_flag(after_cap, CAP_PERMITTED, 1, &v, CAP_CLEAR) < 0 ||
 237                             cap_set_flag(after_cap, CAP_EFFECTIVE, 1, &v, CAP_CLEAR) < 0) {
 238                                 r = -errno;
 239                                 goto finish;
 240                         }
 241                 }
 242         }
 243
 244         r = 0;
 245
 246 finish:
 247         if (cap_set_proc(after_cap) < 0) {
 248                 /* If there are no actual changes anyway then let's ignore this error. */
 249                 if (cap_compare(before_cap, after_cap) != 0)
 250                         r = -errno;
 251         }
 252
 253         return r;
 254 }
 255
 256 static int drop_from_file(const char *fn, uint64_t keep) {
 257         _cleanup_free_ char *p = NULL;
 258         uint64_t current, after;
 259         uint32_t hi, lo;
 260         int r, k;
 261
 262         r = read_one_line_file(fn, &p);
 263         if (r < 0)
 264                 return r;
 265
 266         k = sscanf(p, "%" PRIu32 " %" PRIu32, &lo, &hi);
 267         if (k != 2)
 268                 return -EIO;
 269
 270         current = (uint64_t) lo | ((uint64_t) hi << 32);
 271         after = current & keep;
 272
 273         if (current == after)
 274                 return 0;
 275
 276         lo = after & UINT32_C(0xFFFFFFFF);
 277         hi = (after >> 32) & UINT32_C(0xFFFFFFFF);
 278
 279         return write_string_filef(fn, 0, "%" PRIu32 " %" PRIu32, lo, hi);
 280 }
 281
 282 int capability_bounding_set_drop_usermode(uint64_t keep) {
 283         int r;
 284
 285         r = drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", keep);
 286         if (r < 0)
 287                 return r;
 288
 289         r = drop_from_file("/proc/sys/kernel/usermodehelper/bset", keep);
 290         if (r < 0)
 291                 return r;
 292
 293         return r;
 294 }
 295
 296 int drop_privileges(uid_t uid, gid_t gid, uint64_t keep_capabilities) {
 297         int r;
 298
 299         /* Unfortunately we cannot leave privilege dropping to PID 1 here, since we want to run as user but
 300          * want to keep some capabilities. Since file capabilities have been introduced this cannot be done
 301          * across exec() anymore, unless our binary has the capability configured in the file system, which
 302          * we want to avoid. */
 303
 304         if (setresgid(gid, gid, gid) < 0)
 305                 return log_error_errno(errno, "Failed to change group ID: %m");
 306
 307         r = maybe_setgroups(0, NULL);
 308         if (r < 0)
 309                 return log_error_errno(r, "Failed to drop auxiliary groups list: %m");
 310
 311         /* Ensure we keep the permitted caps across the setresuid(). Note that we do this even if we actually
 312          * don't want to keep any capabilities, since we want to be able to drop them from the bounding set
 313          * too, and we can only do that if we have capabilities. */
 314         if (prctl(PR_SET_KEEPCAPS, 1) < 0)
 315                 return log_error_errno(errno, "Failed to enable keep capabilities flag: %m");
 316
 317         if (setresuid(uid, uid, uid) < 0)
 318                 return log_error_errno(errno, "Failed to change user ID: %m");
 319
 320         if (prctl(PR_SET_KEEPCAPS, 0) < 0)
 321                 return log_error_errno(errno, "Failed to disable keep capabilities flag: %m");
 322
 323         /* Drop all caps from the bounding set (as well as the inheritable/permitted/effective sets), except
 324          * the ones we want to keep */
 325         r = capability_bounding_set_drop(keep_capabilities, true);
 326         if (r < 0)
 327                 return log_error_errno(r, "Failed to drop capabilities: %m");
 328
 329         /* Now upgrade the permitted caps we still kept to effective caps */
 330         if (keep_capabilities != 0) {
 331                 cap_value_t bits[u64log2(keep_capabilities) + 1];
 332                 _cleanup_cap_free_ cap_t d = NULL;
 333                 unsigned i, j = 0;
 334
 335                 d = cap_init();
 336                 if (!d)
 337                         return log_oom();
 338
 339                 for (i = 0; i < ELEMENTSOF(bits); i++)
 340                         if (keep_capabilities & (1ULL << i))
 341                                 bits[j++] = i;
 342
 343                 /* use enough bits */
 344                 assert(i == 64 || (keep_capabilities >> i) == 0);
 345                 /* don't use too many bits */
 346                 assert(keep_capabilities & (UINT64_C(1) << (i - 1)));
 347
 348                 if (cap_set_flag(d, CAP_EFFECTIVE, j, bits, CAP_SET) < 0 ||
 349                     cap_set_flag(d, CAP_PERMITTED, j, bits, CAP_SET) < 0)
 350                         return log_error_errno(errno, "Failed to enable capabilities bits: %m");
 351
 352                 if (cap_set_proc(d) < 0)
 353                         return log_error_errno(errno, "Failed to increase capabilities: %m");
 354         }
 355
 356         return 0;
 357 }
 358
 359 int drop_capability(cap_value_t cv) {
 360         _cleanup_cap_free_ cap_t tmp_cap = NULL;
 361
 362         tmp_cap = cap_get_proc();
 363         if (!tmp_cap)
 364                 return -errno;
 365
 366         if ((cap_set_flag(tmp_cap, CAP_INHERITABLE, 1, &cv, CAP_CLEAR) < 0) ||
 367             (cap_set_flag(tmp_cap, CAP_PERMITTED, 1, &cv, CAP_CLEAR) < 0) ||
 368             (cap_set_flag(tmp_cap, CAP_EFFECTIVE, 1, &cv, CAP_CLEAR) < 0))
 369                 return -errno;
 370
 371         if (cap_set_proc(tmp_cap) < 0)
 372                 return -errno;
 373
 374         return 0;
 375 }
 376
 377 bool ambient_capabilities_supported(void) {
 378         static int cache = -1;
 379
 380         if (cache >= 0)
 381                 return cache;
 382
 383         /* If PR_CAP_AMBIENT returns something valid, or an unexpected error code we assume that ambient caps are
 384          * available. */
 385
 386         cache = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, CAP_KILL, 0, 0) >= 0 ||
 387                 !IN_SET(errno, EINVAL, EOPNOTSUPP, ENOSYS);
 388
 389         return cache;
 390 }
 391
 392 bool capability_quintet_mangle(CapabilityQuintet *q) {
 393         unsigned long i;
 394         uint64_t combined, drop = 0;
 395         bool ambient_supported;
 396
 397         assert(q);
 398
 399         combined = q->effective | q->bounding | q->inheritable | q->permitted;
 400
 401         ambient_supported = q->ambient != (uint64_t) -1;
 402         if (ambient_supported)
 403                 combined |= q->ambient;
 404
 405         for (i = 0; i <= cap_last_cap(); i++) {
 406                 unsigned long bit = UINT64_C(1) << i;
 407                 if (!FLAGS_SET(combined, bit))
 408                         continue;
 409
 410                 if (prctl(PR_CAPBSET_READ, i) > 0)
 411                         continue;
 412
 413                 drop |= bit;
 414
 415                 log_debug("Not in the current bounding set: %s", capability_to_name(i));
 416         }
 417
 418         q->effective &= ~drop;
 419         q->bounding &= ~drop;
 420         q->inheritable &= ~drop;
 421         q->permitted &= ~drop;
 422
 423         if (ambient_supported)
 424                 q->ambient &= ~drop;
 425
 426         return drop != 0; /* Let the caller know we changed something */
 427 }
 428
 429 int capability_quintet_enforce(const CapabilityQuintet *q) {
 430         _cleanup_cap_free_ cap_t c = NULL, modified = NULL;
 431         int r;
 432
 433         if (q->ambient != (uint64_t) -1) {
 434                 unsigned long i;
 435                 bool changed = false;
 436
 437                 c = cap_get_proc();
 438                 if (!c)
 439                         return -errno;
 440
 441                 /* In order to raise the ambient caps set we first need to raise the matching inheritable + permitted
 442                  * cap */
 443                 for (i = 0; i <= cap_last_cap(); i++) {
 444                         uint64_t m = UINT64_C(1) << i;
 445                         cap_value_t cv = (cap_value_t) i;
 446                         cap_flag_value_t old_value_inheritable, old_value_permitted;
 447
 448                         if ((q->ambient & m) == 0)
 449                                 continue;
 450
 451                         if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value_inheritable) < 0)
 452                                 return -errno;
 453                         if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value_permitted) < 0)
 454                                 return -errno;
 455
 456                         if (old_value_inheritable == CAP_SET && old_value_permitted == CAP_SET)
 457                                 continue;
 458
 459                         if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, CAP_SET) < 0)
 460                                 return -errno;
 461                         if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
 462                                 return -errno;
 463
 464                         changed = true;
 465                 }
 466
 467                 if (changed)
 468                         if (cap_set_proc(c) < 0)
 469                                 return -errno;
 470
 471                 r = capability_ambient_set_apply(q->ambient, false);
 472                 if (r < 0)
 473                         return r;
 474         }
 475
 476         if (q->inheritable != (uint64_t) -1 || q->permitted != (uint64_t) -1 || q->effective != (uint64_t) -1) {
 477                 bool changed = false;
 478                 unsigned long i;
 479
 480                 if (!c) {
 481                         c = cap_get_proc();
 482                         if (!c)
 483                                 return -errno;
 484                 }
 485
 486                 for (i = 0; i <= cap_last_cap(); i++) {
 487                         uint64_t m = UINT64_C(1) << i;
 488                         cap_value_t cv = (cap_value_t) i;
 489
 490                         if (q->inheritable != (uint64_t) -1) {
 491                                 cap_flag_value_t old_value, new_value;
 492
 493                                 if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value) < 0) {
 494                                         if (errno == EINVAL) /* If the kernel knows more caps than this
 495                                                               * version of libcap, then this will return
 496                                                               * EINVAL. In that case, simply ignore it,
 497                                                               * pretend it doesn't exist. */
 498                                                 continue;
 499
 500                                         return -errno;
 501                                 }
 502
 503                                 new_value = (q->inheritable & m) ? CAP_SET : CAP_CLEAR;
 504
 505                                 if (old_value != new_value) {
 506                                         changed = true;
 507
 508                                         if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, new_value) < 0)
 509                                                 return -errno;
 510                                 }
 511                         }
 512
 513                         if (q->permitted != (uint64_t) -1) {
 514                                 cap_flag_value_t old_value, new_value;
 515
 516                                 if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value) < 0) {
 517                                         if (errno == EINVAL)
 518                                                 continue;
 519
 520                                         return -errno;
 521                                 }
 522
 523                                 new_value = (q->permitted & m) ? CAP_SET : CAP_CLEAR;
 524
 525                                 if (old_value != new_value) {
 526                                         changed = true;
 527
 528                                         if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, new_value) < 0)
 529                                                 return -errno;
 530                                 }
 531                         }
 532
 533                         if (q->effective != (uint64_t) -1) {
 534                                 cap_flag_value_t old_value, new_value;
 535
 536                                 if (cap_get_flag(c, cv, CAP_EFFECTIVE, &old_value) < 0) {
 537                                         if (errno == EINVAL)
 538                                                 continue;
 539
 540                                         return -errno;
 541                                 }
 542
 543                                 new_value = (q->effective & m) ? CAP_SET : CAP_CLEAR;
 544
 545                                 if (old_value != new_value) {
 546                                         changed = true;
 547
 548                                         if (cap_set_flag(c, CAP_EFFECTIVE, 1, &cv, new_value) < 0)
 549                                                 return -errno;
 550                                 }
 551                         }
 552                 }
 553
 554                 if (changed) {
 555                         /* In order to change the bounding caps, we need to keep CAP_SETPCAP for a bit
 556                          * longer. Let's add it to our list hence for now. */
 557                         if (q->bounding != (uint64_t) -1) {
 558                                 cap_value_t cv = CAP_SETPCAP;
 559
 560                                 modified = cap_dup(c);
 561                                 if (!modified)
 562                                         return -ENOMEM;
 563
 564                                 if (cap_set_flag(modified, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
 565                                         return -errno;
 566                                 if (cap_set_flag(modified, CAP_EFFECTIVE, 1, &cv, CAP_SET) < 0)
 567                                         return -errno;
 568
 569                                 if (cap_compare(modified, c) == 0) {
 570                                         /* No change? then drop this nonsense again */
 571                                         cap_free(modified);
 572                                         modified = NULL;
 573                                 }
 574                         }
 575
 576                         /* Now, let's enforce the caps for the first time. Note that this is where we acquire
 577                          * caps in any of the sets we currently don't have. We have to do this before
 578                          * dropping the bounding caps below, since at that point we can never acquire new
 579                          * caps in inherited/permitted/effective anymore, but only lose them. */
 580                         if (cap_set_proc(modified ?: c) < 0)
 581                                 return -errno;
 582                 }
 583         }
 584
 585         if (q->bounding != (uint64_t) -1) {
 586                 r = capability_bounding_set_drop(q->bounding, false);
 587                 if (r < 0)
 588                         return r;
 589         }
 590
 591         /* If needed, let's now set the caps again, this time in the final version, which differs from what
 592          * we have already set only in the CAP_SETPCAP bit, which we needed for dropping the bounding
 593          * bits. This call only undoes bits and doesn't acquire any which means the bounding caps don't
 594          * matter. */
 595         if (modified)
 596                 if (cap_set_proc(c) < 0)
 597                         return -errno;
 598
 599         return 0;
 600 }