src/basic/capability-util.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <errno.h>
   4 #include <stdatomic.h>
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <sys/prctl.h>
   8 #include <unistd.h>
   9
  10 #include "alloc-util.h"
  11 #include "capability-util.h"
  12 #include "cap-list.h"
  13 #include "fileio.h"
  14 #include "log.h"
  15 #include "logarithm.h"
  16 #include "macro.h"
  17 #include "missing_prctl.h"
  18 #include "missing_threads.h"
  19 #include "parse-util.h"
  20 #include "user-util.h"
  21
  22 int have_effective_cap(int value) {
  23         _cleanup_cap_free_ cap_t cap = NULL;
  24         cap_flag_value_t fv = CAP_CLEAR; /* To avoid false-positive use-of-uninitialized-value error reported
  25                                           * by fuzzers. */
  26
  27         cap = cap_get_proc();
  28         if (!cap)
  29                 return -errno;
  30
  31         if (cap_get_flag(cap, value, CAP_EFFECTIVE, &fv) < 0)
  32                 return -errno;
  33
  34         return fv == CAP_SET;
  35 }
  36
  37 unsigned cap_last_cap(void) {
  38         static atomic_int saved = INT_MAX;
  39         int r, c;
  40
  41         c = saved;
  42         if (c != INT_MAX)
  43                 return c;
  44
  45         /* Available since linux-3.2 */
  46         _cleanup_free_ char *content = NULL;
  47         r = read_one_line_file("/proc/sys/kernel/cap_last_cap", &content);
  48         if (r < 0)
  49                 log_debug_errno(r, "Failed to read /proc/sys/kernel/cap_last_cap, ignoring: %m");
  50         else {
  51                 r = safe_atoi(content, &c);
  52                 if (r < 0)
  53                         log_debug_errno(r, "Failed to parse /proc/sys/kernel/cap_last_cap, ignoring: %m");
  54                 else {
  55                         if (c > CAP_LIMIT) /* Safety for the future: if one day the kernel learns more than
  56                                             * 64 caps, then we are in trouble (since we, as much userspace
  57                                             * and kernel space store capability masks in uint64_t types). We
  58                                             * also want to use UINT64_MAX as marker for "unset". Hence let's
  59                                             * hence protect ourselves against that and always cap at 62 for
  60                                             * now. */
  61                                 c = CAP_LIMIT;
  62
  63                         saved = c;
  64                         return c;
  65                 }
  66         }
  67
  68         /* Fall back to syscall-probing for pre linux-3.2, or where /proc/ is not mounted */
  69         unsigned long p = (unsigned long) MIN(CAP_LAST_CAP, CAP_LIMIT);
  70
  71         if (prctl(PR_CAPBSET_READ, p) < 0) {
  72
  73                 /* Hmm, look downwards, until we find one that works */
  74                 for (p--; p > 0; p--)
  75                         if (prctl(PR_CAPBSET_READ, p) >= 0)
  76                                 break;
  77
  78         } else {
  79
  80                 /* Hmm, look upwards, until we find one that doesn't work */
  81                 for (; p < CAP_LIMIT; p++)
  82                         if (prctl(PR_CAPBSET_READ, p+1) < 0)
  83                                 break;
  84         }
  85
  86         c = (int) p;
  87         saved = c;
  88         return c;
  89 }
  90
  91 int capability_update_inherited_set(cap_t caps, uint64_t set) {
  92         /* Add capabilities in the set to the inherited caps, drops capabilities not in the set.
  93          * Do not apply them yet. */
  94
  95         for (unsigned i = 0; i <= cap_last_cap(); i++) {
  96                 cap_flag_value_t flag = set & (UINT64_C(1) << i) ? CAP_SET : CAP_CLEAR;
  97                 cap_value_t v;
  98
  99                 v = (cap_value_t) i;
 100
 101                 if (cap_set_flag(caps, CAP_INHERITABLE, 1, &v, flag) < 0)
 102                         return -errno;
 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         int r;
 111
 112         /* Remove capabilities requested in ambient set, but not in the bounding set */
 113         for (unsigned i = 0; i <= cap_last_cap(); i++) {
 114                 if (set == 0)
 115                         break;
 116
 117                 if (FLAGS_SET(set, (UINT64_C(1) << i)) && prctl(PR_CAPBSET_READ, i) != 1) {
 118                         log_debug("Ambient capability %s requested but missing from bounding set,"
 119                                         " suppressing automatically.", capability_to_name(i));
 120                         set &= ~(UINT64_C(1) << i);
 121                 }
 122         }
 123
 124         /* Add the capabilities to the ambient set (an possibly also the inheritable set) */
 125
 126         /* Check that we can use PR_CAP_AMBIENT or quit early. */
 127         if (!ambient_capabilities_supported())
 128                 return (set & all_capabilities()) == 0 ?
 129                         0 : -EOPNOTSUPP; /* if actually no ambient caps are to be set, be silent,
 130                                           * otherwise fail recognizably */
 131
 132         if (also_inherit) {
 133                 caps = cap_get_proc();
 134                 if (!caps)
 135                         return -errno;
 136
 137                 r = capability_update_inherited_set(caps, set);
 138                 if (r < 0)
 139                         return -errno;
 140
 141                 if (cap_set_proc(caps) < 0)
 142                         return -errno;
 143         }
 144
 145         for (unsigned i = 0; i <= cap_last_cap(); i++) {
 146
 147                 if (set & (UINT64_C(1) << i)) {
 148
 149                         /* Add the capability to the ambient set. */
 150                         if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_RAISE, i, 0, 0) < 0)
 151                                 return -errno;
 152                 } else {
 153
 154                         /* Drop the capability so we don't inherit capabilities we didn't ask for. */
 155                         r = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, i, 0, 0);
 156                         if (r < 0)
 157                                 return -errno;
 158
 159                         if (r)
 160                                 if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_LOWER, i, 0, 0) < 0)
 161                                         return -errno;
 162
 163                 }
 164         }
 165
 166         return 0;
 167 }
 168
 169 int capability_gain_cap_setpcap(cap_t *ret_before_caps) {
 170         _cleanup_cap_free_ cap_t caps = NULL;
 171         cap_flag_value_t fv;
 172         caps = cap_get_proc();
 173         if (!caps)
 174                 return -errno;
 175
 176         if (cap_get_flag(caps, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0)
 177                 return -errno;
 178
 179         if (fv != CAP_SET) {
 180                 _cleanup_cap_free_ cap_t temp_cap = NULL;
 181                 static const cap_value_t v = CAP_SETPCAP;
 182
 183                 temp_cap = cap_dup(caps);
 184                 if (!temp_cap)
 185                         return -errno;
 186
 187                 if (cap_set_flag(temp_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0)
 188                         return -errno;
 189
 190                 if (cap_set_proc(temp_cap) < 0)
 191                         log_debug_errno(errno, "Can't acquire effective CAP_SETPCAP bit, ignoring: %m");
 192
 193                 /* If we didn't manage to acquire the CAP_SETPCAP bit, we continue anyway, after all this just means
 194                  * we'll fail later, when we actually intend to drop some capabilities or try to set securebits. */
 195         }
 196         if (ret_before_caps)
 197                 /* Return the capabilities as they have been before setting CAP_SETPCAP */
 198                 *ret_before_caps = TAKE_PTR(caps);
 199
 200         return 0;
 201 }
 202
 203 int capability_bounding_set_drop(uint64_t keep, bool right_now) {
 204         _cleanup_cap_free_ cap_t before_cap = NULL, after_cap = NULL;
 205         int r;
 206
 207         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 208          * in the effective set (yes, the kernel drops that when
 209          * executing init!), so get it back temporarily so that we can
 210          * call PR_CAPBSET_DROP. */
 211
 212         r = capability_gain_cap_setpcap(&before_cap);
 213         if (r < 0)
 214                 return r;
 215
 216         after_cap = cap_dup(before_cap);
 217         if (!after_cap)
 218                 return -errno;
 219
 220         for (unsigned i = 0; i <= cap_last_cap(); i++) {
 221                 cap_value_t v;
 222
 223                 if ((keep & (UINT64_C(1) << i)))
 224                         continue;
 225
 226                 /* Drop it from the bounding set */
 227                 if (prctl(PR_CAPBSET_DROP, i) < 0) {
 228                         r = -errno;
 229
 230                         /* If dropping the capability failed, let's see if we didn't have it in the first place. If so,
 231                          * continue anyway, as dropping a capability we didn't have in the first place doesn't really
 232                          * matter anyway. */
 233                         if (prctl(PR_CAPBSET_READ, i) != 0)
 234                                 goto finish;
 235                 }
 236                 v = (cap_value_t) i;
 237
 238                 /* Also drop it from the inheritable set, so
 239                  * that anything we exec() loses the
 240                  * capability for good. */
 241                 if (cap_set_flag(after_cap, CAP_INHERITABLE, 1, &v, CAP_CLEAR) < 0) {
 242                         r = -errno;
 243                         goto finish;
 244                 }
 245
 246                 /* If we shall apply this right now drop it
 247                  * also from our own capability sets. */
 248                 if (right_now) {
 249                         if (cap_set_flag(after_cap, CAP_PERMITTED, 1, &v, CAP_CLEAR) < 0 ||
 250                             cap_set_flag(after_cap, CAP_EFFECTIVE, 1, &v, CAP_CLEAR) < 0) {
 251                                 r = -errno;
 252                                 goto finish;
 253                         }
 254                 }
 255         }
 256
 257         r = 0;
 258
 259 finish:
 260         if (cap_set_proc(after_cap) < 0) {
 261                 /* If there are no actual changes anyway then let's ignore this error. */
 262                 if (cap_compare(before_cap, after_cap) != 0)
 263                         r = -errno;
 264         }
 265
 266         return r;
 267 }
 268
 269 static int drop_from_file(const char *fn, uint64_t keep) {
 270         _cleanup_free_ char *p = NULL;
 271         uint64_t current, after;
 272         uint32_t hi, lo;
 273         int r, k;
 274
 275         r = read_one_line_file(fn, &p);
 276         if (r < 0)
 277                 return r;
 278
 279         k = sscanf(p, "%" PRIu32 " %" PRIu32, &lo, &hi);
 280         if (k != 2)
 281                 return -EIO;
 282
 283         current = (uint64_t) lo | ((uint64_t) hi << 32);
 284         after = current & keep;
 285
 286         if (current == after)
 287                 return 0;
 288
 289         lo = after & UINT32_MAX;
 290         hi = (after >> 32) & UINT32_MAX;
 291
 292         return write_string_filef(fn, 0, "%" PRIu32 " %" PRIu32, lo, hi);
 293 }
 294
 295 int capability_bounding_set_drop_usermode(uint64_t keep) {
 296         int r;
 297
 298         r = drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", keep);
 299         if (r < 0)
 300                 return r;
 301
 302         r = drop_from_file("/proc/sys/kernel/usermodehelper/bset", keep);
 303         if (r < 0)
 304                 return r;
 305
 306         return r;
 307 }
 308
 309 int drop_privileges(uid_t uid, gid_t gid, uint64_t keep_capabilities) {
 310         int r;
 311
 312         /* Unfortunately we cannot leave privilege dropping to PID 1 here, since we want to run as user but
 313          * want to keep some capabilities. Since file capabilities have been introduced this cannot be done
 314          * across exec() anymore, unless our binary has the capability configured in the file system, which
 315          * we want to avoid. */
 316
 317         if (setresgid(gid, gid, gid) < 0)
 318                 return log_error_errno(errno, "Failed to change group ID: %m");
 319
 320         r = maybe_setgroups(0, NULL);
 321         if (r < 0)
 322                 return log_error_errno(r, "Failed to drop auxiliary groups list: %m");
 323
 324         /* Ensure we keep the permitted caps across the setresuid(). Note that we do this even if we actually
 325          * don't want to keep any capabilities, since we want to be able to drop them from the bounding set
 326          * too, and we can only do that if we have capabilities. */
 327         if (prctl(PR_SET_KEEPCAPS, 1) < 0)
 328                 return log_error_errno(errno, "Failed to enable keep capabilities flag: %m");
 329
 330         if (setresuid(uid, uid, uid) < 0)
 331                 return log_error_errno(errno, "Failed to change user ID: %m");
 332
 333         if (prctl(PR_SET_KEEPCAPS, 0) < 0)
 334                 return log_error_errno(errno, "Failed to disable keep capabilities flag: %m");
 335
 336         /* Drop all caps from the bounding set (as well as the inheritable/permitted/effective sets), except
 337          * the ones we want to keep */
 338         r = capability_bounding_set_drop(keep_capabilities, true);
 339         if (r < 0)
 340                 return log_error_errno(r, "Failed to drop capabilities: %m");
 341
 342         /* Now upgrade the permitted caps we still kept to effective caps */
 343         if (keep_capabilities != 0) {
 344                 cap_value_t bits[log2u64(keep_capabilities) + 1];
 345                 _cleanup_cap_free_ cap_t d = NULL;
 346                 unsigned i, j = 0;
 347
 348                 d = cap_init();
 349                 if (!d)
 350                         return log_oom();
 351
 352                 for (i = 0; i < ELEMENTSOF(bits); i++)
 353                         if (keep_capabilities & (1ULL << i))
 354                                 bits[j++] = i;
 355
 356                 /* use enough bits */
 357                 assert(i == 64 || (keep_capabilities >> i) == 0);
 358                 /* don't use too many bits */
 359                 assert(keep_capabilities & (UINT64_C(1) << (i - 1)));
 360
 361                 if (cap_set_flag(d, CAP_EFFECTIVE, j, bits, CAP_SET) < 0 ||
 362                     cap_set_flag(d, CAP_PERMITTED, j, bits, CAP_SET) < 0)
 363                         return log_error_errno(errno, "Failed to enable capabilities bits: %m");
 364
 365                 if (cap_set_proc(d) < 0)
 366                         return log_error_errno(errno, "Failed to increase capabilities: %m");
 367         }
 368
 369         return 0;
 370 }
 371
 372 static int change_capability(cap_value_t cv, cap_flag_value_t flag) {
 373         _cleanup_cap_free_ cap_t tmp_cap = NULL;
 374
 375         tmp_cap = cap_get_proc();
 376         if (!tmp_cap)
 377                 return -errno;
 378
 379         if ((cap_set_flag(tmp_cap, CAP_INHERITABLE, 1, &cv, flag) < 0) ||
 380             (cap_set_flag(tmp_cap, CAP_PERMITTED, 1, &cv, flag) < 0) ||
 381             (cap_set_flag(tmp_cap, CAP_EFFECTIVE, 1, &cv, flag) < 0))
 382                 return -errno;
 383
 384         if (cap_set_proc(tmp_cap) < 0)
 385                 return -errno;
 386
 387         return 0;
 388 }
 389
 390 int drop_capability(cap_value_t cv) {
 391         return change_capability(cv, CAP_CLEAR);
 392 }
 393
 394 int keep_capability(cap_value_t cv) {
 395         return change_capability(cv, CAP_SET);
 396 }
 397
 398 bool ambient_capabilities_supported(void) {
 399         static int cache = -1;
 400
 401         if (cache >= 0)
 402                 return cache;
 403
 404         /* If PR_CAP_AMBIENT returns something valid, or an unexpected error code we assume that ambient caps are
 405          * available. */
 406
 407         cache = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, CAP_KILL, 0, 0) >= 0 ||
 408                 !IN_SET(errno, EINVAL, EOPNOTSUPP, ENOSYS);
 409
 410         return cache;
 411 }
 412
 413 bool capability_quintet_mangle(CapabilityQuintet *q) {
 414         uint64_t combined, drop = 0;
 415         bool ambient_supported;
 416
 417         assert(q);
 418
 419         combined = q->effective | q->bounding | q->inheritable | q->permitted;
 420
 421         ambient_supported = q->ambient != CAP_MASK_UNSET;
 422         if (ambient_supported)
 423                 combined |= q->ambient;
 424
 425         for (unsigned i = 0; i <= cap_last_cap(); i++) {
 426                 unsigned long bit = UINT64_C(1) << i;
 427                 if (!FLAGS_SET(combined, bit))
 428                         continue;
 429
 430                 if (prctl(PR_CAPBSET_READ, i) > 0)
 431                         continue;
 432
 433                 drop |= bit;
 434
 435                 log_debug("Not in the current bounding set: %s", capability_to_name(i));
 436         }
 437
 438         q->effective &= ~drop;
 439         q->bounding &= ~drop;
 440         q->inheritable &= ~drop;
 441         q->permitted &= ~drop;
 442
 443         if (ambient_supported)
 444                 q->ambient &= ~drop;
 445
 446         return drop != 0; /* Let the caller know we changed something */
 447 }
 448
 449 int capability_quintet_enforce(const CapabilityQuintet *q) {
 450         _cleanup_cap_free_ cap_t c = NULL, modified = NULL;
 451         int r;
 452
 453         if (q->ambient != CAP_MASK_UNSET) {
 454                 bool changed = false;
 455
 456                 c = cap_get_proc();
 457                 if (!c)
 458                         return -errno;
 459
 460                 /* In order to raise the ambient caps set we first need to raise the matching
 461                  * inheritable + permitted cap */
 462                 for (unsigned i = 0; i <= cap_last_cap(); i++) {
 463                         uint64_t m = UINT64_C(1) << i;
 464                         cap_value_t cv = (cap_value_t) i;
 465                         cap_flag_value_t old_value_inheritable, old_value_permitted;
 466
 467                         if ((q->ambient & m) == 0)
 468                                 continue;
 469
 470                         if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value_inheritable) < 0)
 471                                 return -errno;
 472                         if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value_permitted) < 0)
 473                                 return -errno;
 474
 475                         if (old_value_inheritable == CAP_SET && old_value_permitted == CAP_SET)
 476                                 continue;
 477
 478                         if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, CAP_SET) < 0)
 479                                 return -errno;
 480                         if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
 481                                 return -errno;
 482
 483                         changed = true;
 484                 }
 485
 486                 if (changed)
 487                         if (cap_set_proc(c) < 0)
 488                                 return -errno;
 489
 490                 r = capability_ambient_set_apply(q->ambient, false);
 491                 if (r < 0)
 492                         return r;
 493         }
 494
 495         if (q->inheritable != CAP_MASK_UNSET || q->permitted != CAP_MASK_UNSET || q->effective != CAP_MASK_UNSET) {
 496                 bool changed = false;
 497
 498                 if (!c) {
 499                         c = cap_get_proc();
 500                         if (!c)
 501                                 return -errno;
 502                 }
 503
 504                 for (unsigned i = 0; i <= cap_last_cap(); i++) {
 505                         uint64_t m = UINT64_C(1) << i;
 506                         cap_value_t cv = (cap_value_t) i;
 507
 508                         if (q->inheritable != CAP_MASK_UNSET) {
 509                                 cap_flag_value_t old_value, new_value;
 510
 511                                 if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value) < 0) {
 512                                         if (errno == EINVAL) /* If the kernel knows more caps than this
 513                                                               * version of libcap, then this will return
 514                                                               * EINVAL. In that case, simply ignore it,
 515                                                               * pretend it doesn't exist. */
 516                                                 continue;
 517
 518                                         return -errno;
 519                                 }
 520
 521                                 new_value = (q->inheritable & m) ? CAP_SET : CAP_CLEAR;
 522
 523                                 if (old_value != new_value) {
 524                                         changed = true;
 525
 526                                         if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, new_value) < 0)
 527                                                 return -errno;
 528                                 }
 529                         }
 530
 531                         if (q->permitted != CAP_MASK_UNSET) {
 532                                 cap_flag_value_t old_value, new_value;
 533
 534                                 if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value) < 0) {
 535                                         if (errno == EINVAL)
 536                                                 continue;
 537
 538                                         return -errno;
 539                                 }
 540
 541                                 new_value = (q->permitted & m) ? CAP_SET : CAP_CLEAR;
 542
 543                                 if (old_value != new_value) {
 544                                         changed = true;
 545
 546                                         if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, new_value) < 0)
 547                                                 return -errno;
 548                                 }
 549                         }
 550
 551                         if (q->effective != CAP_MASK_UNSET) {
 552                                 cap_flag_value_t old_value, new_value;
 553
 554                                 if (cap_get_flag(c, cv, CAP_EFFECTIVE, &old_value) < 0) {
 555                                         if (errno == EINVAL)
 556                                                 continue;
 557
 558                                         return -errno;
 559                                 }
 560
 561                                 new_value = (q->effective & m) ? CAP_SET : CAP_CLEAR;
 562
 563                                 if (old_value != new_value) {
 564                                         changed = true;
 565
 566                                         if (cap_set_flag(c, CAP_EFFECTIVE, 1, &cv, new_value) < 0)
 567                                                 return -errno;
 568                                 }
 569                         }
 570                 }
 571
 572                 if (changed) {
 573                         /* In order to change the bounding caps, we need to keep CAP_SETPCAP for a bit
 574                          * longer. Let's add it to our list hence for now. */
 575                         if (q->bounding != CAP_MASK_UNSET) {
 576                                 cap_value_t cv = CAP_SETPCAP;
 577
 578                                 modified = cap_dup(c);
 579                                 if (!modified)
 580                                         return -ENOMEM;
 581
 582                                 if (cap_set_flag(modified, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
 583                                         return -errno;
 584                                 if (cap_set_flag(modified, CAP_EFFECTIVE, 1, &cv, CAP_SET) < 0)
 585                                         return -errno;
 586
 587                                 if (cap_compare(modified, c) == 0) {
 588                                         /* No change? then drop this nonsense again */
 589                                         cap_free(modified);
 590                                         modified = NULL;
 591                                 }
 592                         }
 593
 594                         /* Now, let's enforce the caps for the first time. Note that this is where we acquire
 595                          * caps in any of the sets we currently don't have. We have to do this before
 596                          * dropping the bounding caps below, since at that point we can never acquire new
 597                          * caps in inherited/permitted/effective anymore, but only lose them. */
 598                         if (cap_set_proc(modified ?: c) < 0)
 599                                 return -errno;
 600                 }
 601         }
 602
 603         if (q->bounding != CAP_MASK_UNSET) {
 604                 r = capability_bounding_set_drop(q->bounding, false);
 605                 if (r < 0)
 606                         return r;
 607         }
 608
 609         /* If needed, let's now set the caps again, this time in the final version, which differs from what
 610          * we have already set only in the CAP_SETPCAP bit, which we needed for dropping the bounding
 611          * bits. This call only undoes bits and doesn't acquire any which means the bounding caps don't
 612          * matter. */
 613         if (modified)
 614                 if (cap_set_proc(c) < 0)
 615                         return -errno;
 616
 617         return 0;
 618 }
 619
 620 int capability_get_ambient(uint64_t *ret) {
 621         uint64_t a = 0;
 622         int r;
 623
 624         assert(ret);
 625
 626         if (!ambient_capabilities_supported()) {
 627                 *ret = 0;
 628                 return 0;
 629         }
 630
 631         for (unsigned i = 0; i <= cap_last_cap(); i++) {
 632                 r = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, i, 0, 0);
 633                 if (r < 0)
 634                         return -errno;
 635
 636                 if (r)
 637                         a |= UINT64_C(1) << i;
 638         }
 639
 640
 641         *ret = a;
 642         return 1;
 643 }