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 cap_last_cap(void) {
  35         static thread_local unsigned 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         /* Add capabilities in the set to the inherited caps, drops capabilities not in the set.
  88          * Do not apply them yet. */
  89
  90         for (unsigned i = 0; i <= cap_last_cap(); i++) {
  91                 cap_flag_value_t flag = set & (UINT64_C(1) << i) ? CAP_SET : CAP_CLEAR;
  92                 cap_value_t v;
  93
  94                 v = (cap_value_t) i;
  95
  96                 if (cap_set_flag(caps, CAP_INHERITABLE, 1, &v, flag) < 0)
  97                         return -errno;
  98         }
  99
 100         return 0;
 101 }
 102
 103 int capability_ambient_set_apply(uint64_t set, bool also_inherit) {
 104         _cleanup_cap_free_ cap_t caps = NULL;
 105         int r;
 106
 107         /* Remove capabilities requested in ambient set, but not in the bounding set */
 108         for (unsigned i = 0; i <= cap_last_cap(); i++) {
 109                 if (set == 0)
 110                         break;
 111
 112                 if (FLAGS_SET(set, (UINT64_C(1) << i)) && prctl(PR_CAPBSET_READ, i) != 1) {
 113                         log_debug("Ambient capability %s requested but missing from bounding set,"
 114                                         " suppressing automatically.", capability_to_name(i));
 115                         set &= ~(UINT64_C(1) << i);
 116                 }
 117         }
 118
 119         /* Add the capabilities to the ambient set (an possibly also the inheritable set) */
 120
 121         /* Check that we can use PR_CAP_AMBIENT or quit early. */
 122         if (!ambient_capabilities_supported())
 123                 return (set & all_capabilities()) == 0 ?
 124                         0 : -EOPNOTSUPP; /* if actually no ambient caps are to be set, be silent,
 125                                           * otherwise fail recognizably */
 126
 127         if (also_inherit) {
 128                 caps = cap_get_proc();
 129                 if (!caps)
 130                         return -errno;
 131
 132                 r = capability_update_inherited_set(caps, set);
 133                 if (r < 0)
 134                         return -errno;
 135
 136                 if (cap_set_proc(caps) < 0)
 137                         return -errno;
 138         }
 139
 140         for (unsigned i = 0; i <= cap_last_cap(); i++) {
 141
 142                 if (set & (UINT64_C(1) << i)) {
 143
 144                         /* Add the capability to the ambient set. */
 145                         if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_RAISE, i, 0, 0) < 0)
 146                                 return -errno;
 147                 } else {
 148
 149                         /* Drop the capability so we don't inherit capabilities we didn't ask for. */
 150                         r = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, i, 0, 0);
 151                         if (r < 0)
 152                                 return -errno;
 153
 154                         if (r)
 155                                 if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_LOWER, i, 0, 0) < 0)
 156                                         return -errno;
 157
 158                 }
 159         }
 160
 161         return 0;
 162 }
 163
 164 int capability_bounding_set_drop(uint64_t keep, bool right_now) {
 165         _cleanup_cap_free_ cap_t before_cap = NULL, after_cap = NULL;
 166         cap_flag_value_t fv;
 167         int r;
 168
 169         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 170          * in the effective set (yes, the kernel drops that when
 171          * executing init!), so get it back temporarily so that we can
 172          * call PR_CAPBSET_DROP. */
 173
 174         before_cap = cap_get_proc();
 175         if (!before_cap)
 176                 return -errno;
 177
 178         if (cap_get_flag(before_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0)
 179                 return -errno;
 180
 181         if (fv != CAP_SET) {
 182                 _cleanup_cap_free_ cap_t temp_cap = NULL;
 183                 static const cap_value_t v = CAP_SETPCAP;
 184
 185                 temp_cap = cap_dup(before_cap);
 186                 if (!temp_cap)
 187                         return -errno;
 188
 189                 if (cap_set_flag(temp_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0)
 190                         return -errno;
 191
 192                 if (cap_set_proc(temp_cap) < 0)
 193                         log_debug_errno(errno, "Can't acquire effective CAP_SETPCAP bit, ignoring: %m");
 194
 195                 /* If we didn't manage to acquire the CAP_SETPCAP bit, we continue anyway, after all this just means
 196                  * we'll fail later, when we actually intend to drop some capabilities. */
 197         }
 198
 199         after_cap = cap_dup(before_cap);
 200         if (!after_cap)
 201                 return -errno;
 202
 203         for (unsigned i = 0; i <= cap_last_cap(); i++) {
 204                 cap_value_t v;
 205
 206                 if ((keep & (UINT64_C(1) << i)))
 207                         continue;
 208
 209                 /* Drop it from the bounding set */
 210                 if (prctl(PR_CAPBSET_DROP, i) < 0) {
 211                         r = -errno;
 212
 213                         /* If dropping the capability failed, let's see if we didn't have it in the first place. If so,
 214                          * continue anyway, as dropping a capability we didn't have in the first place doesn't really
 215                          * matter anyway. */
 216                         if (prctl(PR_CAPBSET_READ, i) != 0)
 217                                 goto finish;
 218                 }
 219                 v = (cap_value_t) i;
 220
 221                 /* Also drop it from the inheritable set, so
 222                  * that anything we exec() loses the
 223                  * capability for good. */
 224                 if (cap_set_flag(after_cap, CAP_INHERITABLE, 1, &v, CAP_CLEAR) < 0) {
 225                         r = -errno;
 226                         goto finish;
 227                 }
 228
 229                 /* If we shall apply this right now drop it
 230                  * also from our own capability sets. */
 231                 if (right_now) {
 232                         if (cap_set_flag(after_cap, CAP_PERMITTED, 1, &v, CAP_CLEAR) < 0 ||
 233                             cap_set_flag(after_cap, CAP_EFFECTIVE, 1, &v, CAP_CLEAR) < 0) {
 234                                 r = -errno;
 235                                 goto finish;
 236                         }
 237                 }
 238         }
 239
 240         r = 0;
 241
 242 finish:
 243         if (cap_set_proc(after_cap) < 0) {
 244                 /* If there are no actual changes anyway then let's ignore this error. */
 245                 if (cap_compare(before_cap, after_cap) != 0)
 246                         r = -errno;
 247         }
 248
 249         return r;
 250 }
 251
 252 static int drop_from_file(const char *fn, uint64_t keep) {
 253         _cleanup_free_ char *p = NULL;
 254         uint64_t current, after;
 255         uint32_t hi, lo;
 256         int r, k;
 257
 258         r = read_one_line_file(fn, &p);
 259         if (r < 0)
 260                 return r;
 261
 262         k = sscanf(p, "%" PRIu32 " %" PRIu32, &lo, &hi);
 263         if (k != 2)
 264                 return -EIO;
 265
 266         current = (uint64_t) lo | ((uint64_t) hi << 32);
 267         after = current & keep;
 268
 269         if (current == after)
 270                 return 0;
 271
 272         lo = after & UINT32_C(0xFFFFFFFF);
 273         hi = (after >> 32) & UINT32_C(0xFFFFFFFF);
 274
 275         return write_string_filef(fn, 0, "%" PRIu32 " %" PRIu32, lo, hi);
 276 }
 277
 278 int capability_bounding_set_drop_usermode(uint64_t keep) {
 279         int r;
 280
 281         r = drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", keep);
 282         if (r < 0)
 283                 return r;
 284
 285         r = drop_from_file("/proc/sys/kernel/usermodehelper/bset", keep);
 286         if (r < 0)
 287                 return r;
 288
 289         return r;
 290 }
 291
 292 int drop_privileges(uid_t uid, gid_t gid, uint64_t keep_capabilities) {
 293         int r;
 294
 295         /* Unfortunately we cannot leave privilege dropping to PID 1 here, since we want to run as user but
 296          * want to keep some capabilities. Since file capabilities have been introduced this cannot be done
 297          * across exec() anymore, unless our binary has the capability configured in the file system, which
 298          * we want to avoid. */
 299
 300         if (setresgid(gid, gid, gid) < 0)
 301                 return log_error_errno(errno, "Failed to change group ID: %m");
 302
 303         r = maybe_setgroups(0, NULL);
 304         if (r < 0)
 305                 return log_error_errno(r, "Failed to drop auxiliary groups list: %m");
 306
 307         /* Ensure we keep the permitted caps across the setresuid(). Note that we do this even if we actually
 308          * don't want to keep any capabilities, since we want to be able to drop them from the bounding set
 309          * too, and we can only do that if we have capabilities. */
 310         if (prctl(PR_SET_KEEPCAPS, 1) < 0)
 311                 return log_error_errno(errno, "Failed to enable keep capabilities flag: %m");
 312
 313         if (setresuid(uid, uid, uid) < 0)
 314                 return log_error_errno(errno, "Failed to change user ID: %m");
 315
 316         if (prctl(PR_SET_KEEPCAPS, 0) < 0)
 317                 return log_error_errno(errno, "Failed to disable keep capabilities flag: %m");
 318
 319         /* Drop all caps from the bounding set (as well as the inheritable/permitted/effective sets), except
 320          * the ones we want to keep */
 321         r = capability_bounding_set_drop(keep_capabilities, true);
 322         if (r < 0)
 323                 return log_error_errno(r, "Failed to drop capabilities: %m");
 324
 325         /* Now upgrade the permitted caps we still kept to effective caps */
 326         if (keep_capabilities != 0) {
 327                 cap_value_t bits[u64log2(keep_capabilities) + 1];
 328                 _cleanup_cap_free_ cap_t d = NULL;
 329                 unsigned i, j = 0;
 330
 331                 d = cap_init();
 332                 if (!d)
 333                         return log_oom();
 334
 335                 for (i = 0; i < ELEMENTSOF(bits); i++)
 336                         if (keep_capabilities & (1ULL << i))
 337                                 bits[j++] = i;
 338
 339                 /* use enough bits */
 340                 assert(i == 64 || (keep_capabilities >> i) == 0);
 341                 /* don't use too many bits */
 342                 assert(keep_capabilities & (UINT64_C(1) << (i - 1)));
 343
 344                 if (cap_set_flag(d, CAP_EFFECTIVE, j, bits, CAP_SET) < 0 ||
 345                     cap_set_flag(d, CAP_PERMITTED, j, bits, CAP_SET) < 0)
 346                         return log_error_errno(errno, "Failed to enable capabilities bits: %m");
 347
 348                 if (cap_set_proc(d) < 0)
 349                         return log_error_errno(errno, "Failed to increase capabilities: %m");
 350         }
 351
 352         return 0;
 353 }
 354
 355 int drop_capability(cap_value_t cv) {
 356         _cleanup_cap_free_ cap_t tmp_cap = NULL;
 357
 358         tmp_cap = cap_get_proc();
 359         if (!tmp_cap)
 360                 return -errno;
 361
 362         if ((cap_set_flag(tmp_cap, CAP_INHERITABLE, 1, &cv, CAP_CLEAR) < 0) ||
 363             (cap_set_flag(tmp_cap, CAP_PERMITTED, 1, &cv, CAP_CLEAR) < 0) ||
 364             (cap_set_flag(tmp_cap, CAP_EFFECTIVE, 1, &cv, CAP_CLEAR) < 0))
 365                 return -errno;
 366
 367         if (cap_set_proc(tmp_cap) < 0)
 368                 return -errno;
 369
 370         return 0;
 371 }
 372
 373 bool ambient_capabilities_supported(void) {
 374         static int cache = -1;
 375
 376         if (cache >= 0)
 377                 return cache;
 378
 379         /* If PR_CAP_AMBIENT returns something valid, or an unexpected error code we assume that ambient caps are
 380          * available. */
 381
 382         cache = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, CAP_KILL, 0, 0) >= 0 ||
 383                 !IN_SET(errno, EINVAL, EOPNOTSUPP, ENOSYS);
 384
 385         return cache;
 386 }
 387
 388 bool capability_quintet_mangle(CapabilityQuintet *q) {
 389         uint64_t combined, drop = 0;
 390         bool ambient_supported;
 391
 392         assert(q);
 393
 394         combined = q->effective | q->bounding | q->inheritable | q->permitted;
 395
 396         ambient_supported = q->ambient != (uint64_t) -1;
 397         if (ambient_supported)
 398                 combined |= q->ambient;
 399
 400         for (unsigned i = 0; i <= cap_last_cap(); i++) {
 401                 unsigned long bit = UINT64_C(1) << i;
 402                 if (!FLAGS_SET(combined, bit))
 403                         continue;
 404
 405                 if (prctl(PR_CAPBSET_READ, i) > 0)
 406                         continue;
 407
 408                 drop |= bit;
 409
 410                 log_debug("Not in the current bounding set: %s", capability_to_name(i));
 411         }
 412
 413         q->effective &= ~drop;
 414         q->bounding &= ~drop;
 415         q->inheritable &= ~drop;
 416         q->permitted &= ~drop;
 417
 418         if (ambient_supported)
 419                 q->ambient &= ~drop;
 420
 421         return drop != 0; /* Let the caller know we changed something */
 422 }
 423
 424 int capability_quintet_enforce(const CapabilityQuintet *q) {
 425         _cleanup_cap_free_ cap_t c = NULL, modified = NULL;
 426         int r;
 427
 428         if (q->ambient != (uint64_t) -1) {
 429                 bool changed = false;
 430
 431                 c = cap_get_proc();
 432                 if (!c)
 433                         return -errno;
 434
 435                 /* In order to raise the ambient caps set we first need to raise the matching
 436                  * inheritable + permitted cap */
 437                 for (unsigned i = 0; i <= cap_last_cap(); i++) {
 438                         uint64_t m = UINT64_C(1) << i;
 439                         cap_value_t cv = (cap_value_t) i;
 440                         cap_flag_value_t old_value_inheritable, old_value_permitted;
 441
 442                         if ((q->ambient & m) == 0)
 443                                 continue;
 444
 445                         if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value_inheritable) < 0)
 446                                 return -errno;
 447                         if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value_permitted) < 0)
 448                                 return -errno;
 449
 450                         if (old_value_inheritable == CAP_SET && old_value_permitted == CAP_SET)
 451                                 continue;
 452
 453                         if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, CAP_SET) < 0)
 454                                 return -errno;
 455                         if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
 456                                 return -errno;
 457
 458                         changed = true;
 459                 }
 460
 461                 if (changed)
 462                         if (cap_set_proc(c) < 0)
 463                                 return -errno;
 464
 465                 r = capability_ambient_set_apply(q->ambient, false);
 466                 if (r < 0)
 467                         return r;
 468         }
 469
 470         if (q->inheritable != (uint64_t) -1 || q->permitted != (uint64_t) -1 || q->effective != (uint64_t) -1) {
 471                 bool changed = false;
 472
 473                 if (!c) {
 474                         c = cap_get_proc();
 475                         if (!c)
 476                                 return -errno;
 477                 }
 478
 479                 for (unsigned i = 0; i <= cap_last_cap(); i++) {
 480                         uint64_t m = UINT64_C(1) << i;
 481                         cap_value_t cv = (cap_value_t) i;
 482
 483                         if (q->inheritable != (uint64_t) -1) {
 484                                 cap_flag_value_t old_value, new_value;
 485
 486                                 if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value) < 0) {
 487                                         if (errno == EINVAL) /* If the kernel knows more caps than this
 488                                                               * version of libcap, then this will return
 489                                                               * EINVAL. In that case, simply ignore it,
 490                                                               * pretend it doesn't exist. */
 491                                                 continue;
 492
 493                                         return -errno;
 494                                 }
 495
 496                                 new_value = (q->inheritable & m) ? CAP_SET : CAP_CLEAR;
 497
 498                                 if (old_value != new_value) {
 499                                         changed = true;
 500
 501                                         if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, new_value) < 0)
 502                                                 return -errno;
 503                                 }
 504                         }
 505
 506                         if (q->permitted != (uint64_t) -1) {
 507                                 cap_flag_value_t old_value, new_value;
 508
 509                                 if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value) < 0) {
 510                                         if (errno == EINVAL)
 511                                                 continue;
 512
 513                                         return -errno;
 514                                 }
 515
 516                                 new_value = (q->permitted & m) ? CAP_SET : CAP_CLEAR;
 517
 518                                 if (old_value != new_value) {
 519                                         changed = true;
 520
 521                                         if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, new_value) < 0)
 522                                                 return -errno;
 523                                 }
 524                         }
 525
 526                         if (q->effective != (uint64_t) -1) {
 527                                 cap_flag_value_t old_value, new_value;
 528
 529                                 if (cap_get_flag(c, cv, CAP_EFFECTIVE, &old_value) < 0) {
 530                                         if (errno == EINVAL)
 531                                                 continue;
 532
 533                                         return -errno;
 534                                 }
 535
 536                                 new_value = (q->effective & m) ? CAP_SET : CAP_CLEAR;
 537
 538                                 if (old_value != new_value) {
 539                                         changed = true;
 540
 541                                         if (cap_set_flag(c, CAP_EFFECTIVE, 1, &cv, new_value) < 0)
 542                                                 return -errno;
 543                                 }
 544                         }
 545                 }
 546
 547                 if (changed) {
 548                         /* In order to change the bounding caps, we need to keep CAP_SETPCAP for a bit
 549                          * longer. Let's add it to our list hence for now. */
 550                         if (q->bounding != (uint64_t) -1) {
 551                                 cap_value_t cv = CAP_SETPCAP;
 552
 553                                 modified = cap_dup(c);
 554                                 if (!modified)
 555                                         return -ENOMEM;
 556
 557                                 if (cap_set_flag(modified, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
 558                                         return -errno;
 559                                 if (cap_set_flag(modified, CAP_EFFECTIVE, 1, &cv, CAP_SET) < 0)
 560                                         return -errno;
 561
 562                                 if (cap_compare(modified, c) == 0) {
 563                                         /* No change? then drop this nonsense again */
 564                                         cap_free(modified);
 565                                         modified = NULL;
 566                                 }
 567                         }
 568
 569                         /* Now, let's enforce the caps for the first time. Note that this is where we acquire
 570                          * caps in any of the sets we currently don't have. We have to do this before
 571                          * dropping the bounding caps below, since at that point we can never acquire new
 572                          * caps in inherited/permitted/effective anymore, but only lose them. */
 573                         if (cap_set_proc(modified ?: c) < 0)
 574                                 return -errno;
 575                 }
 576         }
 577
 578         if (q->bounding != (uint64_t) -1) {
 579                 r = capability_bounding_set_drop(q->bounding, false);
 580                 if (r < 0)
 581                         return r;
 582         }
 583
 584         /* If needed, let's now set the caps again, this time in the final version, which differs from what
 585          * we have already set only in the CAP_SETPCAP bit, which we needed for dropping the bounding
 586          * bits. This call only undoes bits and doesn't acquire any which means the bounding caps don't
 587          * matter. */
 588         if (modified)
 589                 if (cap_set_proc(c) < 0)
 590                         return -errno;
 591
 592         return 0;
 593 }