src/basic/capability-util.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <errno.h>
   4 #include <grp.h>
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <sys/capability.h>
   8 #include <sys/prctl.h>
   9 #include <unistd.h>
  10
  11 #include "alloc-util.h"
  12 #include "capability-util.h"
  13 #include "cap-list.h"
  14 #include "fileio.h"
  15 #include "log.h"
  16 #include "macro.h"
  17 #include "missing_prctl.h"
  18 #include "parse-util.h"
  19 #include "user-util.h"
  20 #include "util.h"
  21
  22 int have_effective_cap(int value) {
  23         _cleanup_cap_free_ cap_t cap;
  24         cap_flag_value_t fv;
  25
  26         cap = cap_get_proc();
  27         if (!cap)
  28                 return -errno;
  29
  30         if (cap_get_flag(cap, value, CAP_EFFECTIVE, &fv) < 0)
  31                 return -errno;
  32
  33         return fv == CAP_SET;
  34 }
  35
  36 unsigned long cap_last_cap(void) {
  37         static thread_local unsigned long saved;
  38         static thread_local bool valid = false;
  39         _cleanup_free_ char *content = NULL;
  40         unsigned long p = 0;
  41         int r;
  42
  43         if (valid)
  44                 return saved;
  45
  46         /* available since linux-3.2 */
  47         r = read_one_line_file("/proc/sys/kernel/cap_last_cap", &content);
  48         if (r >= 0) {
  49                 r = safe_atolu(content, &p);
  50                 if (r >= 0) {
  51
  52                         if (p > 63) /* Safety for the future: if one day the kernel learns more than 64 caps,
  53                                      * then we are in trouble (since we, as much userspace and kernel space
  54                                      * store capability masks in uint64_t types). Let's hence protect
  55                                      * ourselves against that and always cap at 63 for now. */
  56                                 p = 63;
  57
  58                         saved = p;
  59                         valid = true;
  60                         return p;
  61                 }
  62         }
  63
  64         /* fall back to syscall-probing for pre linux-3.2 */
  65         p = MIN((unsigned long) CAP_LAST_CAP, 63U);
  66
  67         if (prctl(PR_CAPBSET_READ, p) < 0) {
  68
  69                 /* Hmm, look downwards, until we find one that works */
  70                 for (p--; p > 0; p --)
  71                         if (prctl(PR_CAPBSET_READ, p) >= 0)
  72                                 break;
  73
  74         } else {
  75
  76                 /* Hmm, look upwards, until we find one that doesn't work */
  77                 for (; p < 63; p++)
  78                         if (prctl(PR_CAPBSET_READ, p+1) < 0)
  79                                 break;
  80         }
  81
  82         saved = p;
  83         valid = true;
  84
  85         return p;
  86 }
  87
  88 int capability_update_inherited_set(cap_t caps, uint64_t set) {
  89         unsigned long i;
  90
  91         /* Add capabilities in the set to the inherited caps. Do not apply
  92          * them yet. */
  93
  94         for (i = 0; i <= cap_last_cap(); i++) {
  95
  96                 if (set & (UINT64_C(1) << i)) {
  97                         cap_value_t v;
  98
  99                         v = (cap_value_t) i;
 100
 101                         /* Make the capability inheritable. */
 102                         if (cap_set_flag(caps, CAP_INHERITABLE, 1, &v, CAP_SET) < 0)
 103                                 return -errno;
 104                 }
 105         }
 106
 107         return 0;
 108 }
 109
 110 int capability_ambient_set_apply(uint64_t set, bool also_inherit) {
 111         _cleanup_cap_free_ cap_t caps = NULL;
 112         unsigned long i;
 113         int r;
 114
 115         /* Add the capabilities to the ambient set. */
 116
 117         if (also_inherit) {
 118                 caps = cap_get_proc();
 119                 if (!caps)
 120                         return -errno;
 121
 122                 r = capability_update_inherited_set(caps, set);
 123                 if (r < 0)
 124                         return -errno;
 125
 126                 if (cap_set_proc(caps) < 0)
 127                         return -errno;
 128         }
 129
 130         for (i = 0; i <= cap_last_cap(); i++) {
 131
 132                 if (set & (UINT64_C(1) << i)) {
 133
 134                         /* Add the capability to the ambient set. */
 135                         if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_RAISE, i, 0, 0) < 0)
 136                                 return -errno;
 137                 }
 138         }
 139
 140         return 0;
 141 }
 142
 143 int capability_bounding_set_drop(uint64_t keep, bool right_now) {
 144         _cleanup_cap_free_ cap_t before_cap = NULL, after_cap = NULL;
 145         cap_flag_value_t fv;
 146         unsigned long i;
 147         int r;
 148
 149         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 150          * in the effective set (yes, the kernel drops that when
 151          * executing init!), so get it back temporarily so that we can
 152          * call PR_CAPBSET_DROP. */
 153
 154         before_cap = cap_get_proc();
 155         if (!before_cap)
 156                 return -errno;
 157
 158         if (cap_get_flag(before_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0)
 159                 return -errno;
 160
 161         if (fv != CAP_SET) {
 162                 _cleanup_cap_free_ cap_t temp_cap = NULL;
 163                 static const cap_value_t v = CAP_SETPCAP;
 164
 165                 temp_cap = cap_dup(before_cap);
 166                 if (!temp_cap)
 167                         return -errno;
 168
 169                 if (cap_set_flag(temp_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0)
 170                         return -errno;
 171
 172                 if (cap_set_proc(temp_cap) < 0)
 173                         log_debug_errno(errno, "Can't acquire effective CAP_SETPCAP bit, ignoring: %m");
 174
 175                 /* If we didn't manage to acquire the CAP_SETPCAP bit, we continue anyway, after all this just means
 176                  * we'll fail later, when we actually intend to drop some capabilities. */
 177         }
 178
 179         after_cap = cap_dup(before_cap);
 180         if (!after_cap)
 181                 return -errno;
 182
 183         for (i = 0; i <= cap_last_cap(); i++) {
 184                 cap_value_t v;
 185
 186                 if ((keep & (UINT64_C(1) << i)))
 187                         continue;
 188
 189                 /* Drop it from the bounding set */
 190                 if (prctl(PR_CAPBSET_DROP, i) < 0) {
 191                         r = -errno;
 192
 193                         /* If dropping the capability failed, let's see if we didn't have it in the first place. If so,
 194                          * continue anyway, as dropping a capability we didn't have in the first place doesn't really
 195                          * matter anyway. */
 196                         if (prctl(PR_CAPBSET_READ, i) != 0)
 197                                 goto finish;
 198                 }
 199                 v = (cap_value_t) i;
 200
 201                 /* Also drop it from the inheritable set, so
 202                  * that anything we exec() loses the
 203                  * capability for good. */
 204                 if (cap_set_flag(after_cap, CAP_INHERITABLE, 1, &v, CAP_CLEAR) < 0) {
 205                         r = -errno;
 206                         goto finish;
 207                 }
 208
 209                 /* If we shall apply this right now drop it
 210                  * also from our own capability sets. */
 211                 if (right_now) {
 212                         if (cap_set_flag(after_cap, CAP_PERMITTED, 1, &v, CAP_CLEAR) < 0 ||
 213                             cap_set_flag(after_cap, CAP_EFFECTIVE, 1, &v, CAP_CLEAR) < 0) {
 214                                 r = -errno;
 215                                 goto finish;
 216                         }
 217                 }
 218         }
 219
 220         r = 0;
 221
 222 finish:
 223         if (cap_set_proc(after_cap) < 0) {
 224                 /* If there are no actual changes anyway then let's ignore this error. */
 225                 if (cap_compare(before_cap, after_cap) != 0)
 226                         r = -errno;
 227         }
 228
 229         return r;
 230 }
 231
 232 static int drop_from_file(const char *fn, uint64_t keep) {
 233         _cleanup_free_ char *p = NULL;
 234         uint64_t current, after;
 235         uint32_t hi, lo;
 236         int r, k;
 237
 238         r = read_one_line_file(fn, &p);
 239         if (r < 0)
 240                 return r;
 241
 242         k = sscanf(p, "%" PRIu32 " %" PRIu32, &lo, &hi);
 243         if (k != 2)
 244                 return -EIO;
 245
 246         current = (uint64_t) lo | ((uint64_t) hi << 32);
 247         after = current & keep;
 248
 249         if (current == after)
 250                 return 0;
 251
 252         lo = after & UINT32_C(0xFFFFFFFF);
 253         hi = (after >> 32) & UINT32_C(0xFFFFFFFF);
 254
 255         return write_string_filef(fn, 0, "%" PRIu32 " %" PRIu32, lo, hi);
 256 }
 257
 258 int capability_bounding_set_drop_usermode(uint64_t keep) {
 259         int r;
 260
 261         r = drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", keep);
 262         if (r < 0)
 263                 return r;
 264
 265         r = drop_from_file("/proc/sys/kernel/usermodehelper/bset", keep);
 266         if (r < 0)
 267                 return r;
 268
 269         return r;
 270 }
 271
 272 int drop_privileges(uid_t uid, gid_t gid, uint64_t keep_capabilities) {
 273         int r;
 274
 275         /* Unfortunately we cannot leave privilege dropping to PID 1 here, since we want to run as user but
 276          * want to keep some capabilities. Since file capabilities have been introduced this cannot be done
 277          * across exec() anymore, unless our binary has the capability configured in the file system, which
 278          * we want to avoid. */
 279
 280         if (setresgid(gid, gid, gid) < 0)
 281                 return log_error_errno(errno, "Failed to change group ID: %m");
 282
 283         r = maybe_setgroups(0, NULL);
 284         if (r < 0)
 285                 return log_error_errno(r, "Failed to drop auxiliary groups list: %m");
 286
 287         /* Ensure we keep the permitted caps across the setresuid(). Note that we do this even if we actually
 288          * don't want to keep any capabilities, since we want to be able to drop them from the bounding set
 289          * too, and we can only do that if we have capabilities. */
 290         if (prctl(PR_SET_KEEPCAPS, 1) < 0)
 291                 return log_error_errno(errno, "Failed to enable keep capabilities flag: %m");
 292
 293         if (setresuid(uid, uid, uid) < 0)
 294                 return log_error_errno(errno, "Failed to change user ID: %m");
 295
 296         if (prctl(PR_SET_KEEPCAPS, 0) < 0)
 297                 return log_error_errno(errno, "Failed to disable keep capabilities flag: %m");
 298
 299         /* Drop all caps from the bounding set (as well as the inheritable/permitted/effective sets), except
 300          * the ones we want to keep */
 301         r = capability_bounding_set_drop(keep_capabilities, true);
 302         if (r < 0)
 303                 return log_error_errno(r, "Failed to drop capabilities: %m");
 304
 305         /* Now upgrade the permitted caps we still kept to effective caps */
 306         if (keep_capabilities != 0) {
 307                 cap_value_t bits[u64log2(keep_capabilities) + 1];
 308                 _cleanup_cap_free_ cap_t d = NULL;
 309                 unsigned i, j = 0;
 310
 311                 d = cap_init();
 312                 if (!d)
 313                         return log_oom();
 314
 315                 for (i = 0; i < ELEMENTSOF(bits); i++)
 316                         if (keep_capabilities & (1ULL << i))
 317                                 bits[j++] = i;
 318
 319                 /* use enough bits */
 320                 assert(i == 64 || (keep_capabilities >> i) == 0);
 321                 /* don't use too many bits */
 322                 assert(keep_capabilities & (UINT64_C(1) << (i - 1)));
 323
 324                 if (cap_set_flag(d, CAP_EFFECTIVE, j, bits, CAP_SET) < 0 ||
 325                     cap_set_flag(d, CAP_PERMITTED, j, bits, CAP_SET) < 0)
 326                         return log_error_errno(errno, "Failed to enable capabilities bits: %m");
 327
 328                 if (cap_set_proc(d) < 0)
 329                         return log_error_errno(errno, "Failed to increase capabilities: %m");
 330         }
 331
 332         return 0;
 333 }
 334
 335 int drop_capability(cap_value_t cv) {
 336         _cleanup_cap_free_ cap_t tmp_cap = NULL;
 337
 338         tmp_cap = cap_get_proc();
 339         if (!tmp_cap)
 340                 return -errno;
 341
 342         if ((cap_set_flag(tmp_cap, CAP_INHERITABLE, 1, &cv, CAP_CLEAR) < 0) ||
 343             (cap_set_flag(tmp_cap, CAP_PERMITTED, 1, &cv, CAP_CLEAR) < 0) ||
 344             (cap_set_flag(tmp_cap, CAP_EFFECTIVE, 1, &cv, CAP_CLEAR) < 0))
 345                 return -errno;
 346
 347         if (cap_set_proc(tmp_cap) < 0)
 348                 return -errno;
 349
 350         return 0;
 351 }
 352
 353 bool ambient_capabilities_supported(void) {
 354         static int cache = -1;
 355
 356         if (cache >= 0)
 357                 return cache;
 358
 359         /* If PR_CAP_AMBIENT returns something valid, or an unexpected error code we assume that ambient caps are
 360          * available. */
 361
 362         cache = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, CAP_KILL, 0, 0) >= 0 ||
 363                 !IN_SET(errno, EINVAL, EOPNOTSUPP, ENOSYS);
 364
 365         return cache;
 366 }
 367
 368 bool capability_quintet_mangle(CapabilityQuintet *q) {
 369         unsigned long i;
 370         uint64_t combined, drop = 0;
 371         bool ambient_supported;
 372
 373         assert(q);
 374
 375         combined = q->effective | q->bounding | q->inheritable | q->permitted;
 376
 377         ambient_supported = q->ambient != (uint64_t) -1;
 378         if (ambient_supported)
 379                 combined |= q->ambient;
 380
 381         for (i = 0; i <= cap_last_cap(); i++) {
 382                 unsigned long bit = UINT64_C(1) << i;
 383                 if (!FLAGS_SET(combined, bit))
 384                         continue;
 385
 386                 if (prctl(PR_CAPBSET_READ, i) > 0)
 387                         continue;
 388
 389                 drop |= bit;
 390
 391                 log_debug("Not in the current bounding set: %s", capability_to_name(i));
 392         }
 393
 394         q->effective &= ~drop;
 395         q->bounding &= ~drop;
 396         q->inheritable &= ~drop;
 397         q->permitted &= ~drop;
 398
 399         if (ambient_supported)
 400                 q->ambient &= ~drop;
 401
 402         return drop != 0; /* Let the caller know we changed something */
 403 }
 404
 405 int capability_quintet_enforce(const CapabilityQuintet *q) {
 406         _cleanup_cap_free_ cap_t c = NULL, modified = NULL;
 407         int r;
 408
 409         if (q->ambient != (uint64_t) -1) {
 410                 unsigned long i;
 411                 bool changed = false;
 412
 413                 c = cap_get_proc();
 414                 if (!c)
 415                         return -errno;
 416
 417                 /* In order to raise the ambient caps set we first need to raise the matching inheritable + permitted
 418                  * cap */
 419                 for (i = 0; i <= cap_last_cap(); i++) {
 420                         uint64_t m = UINT64_C(1) << i;
 421                         cap_value_t cv = (cap_value_t) i;
 422                         cap_flag_value_t old_value_inheritable, old_value_permitted;
 423
 424                         if ((q->ambient & m) == 0)
 425                                 continue;
 426
 427                         if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value_inheritable) < 0)
 428                                 return -errno;
 429                         if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value_permitted) < 0)
 430                                 return -errno;
 431
 432                         if (old_value_inheritable == CAP_SET && old_value_permitted == CAP_SET)
 433                                 continue;
 434
 435                         if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, CAP_SET) < 0)
 436                                 return -errno;
 437                         if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
 438                                 return -errno;
 439
 440                         changed = true;
 441                 }
 442
 443                 if (changed)
 444                         if (cap_set_proc(c) < 0)
 445                                 return -errno;
 446
 447                 r = capability_ambient_set_apply(q->ambient, false);
 448                 if (r < 0)
 449                         return r;
 450         }
 451
 452         if (q->inheritable != (uint64_t) -1 || q->permitted != (uint64_t) -1 || q->effective != (uint64_t) -1) {
 453                 bool changed = false;
 454                 unsigned long i;
 455
 456                 if (!c) {
 457                         c = cap_get_proc();
 458                         if (!c)
 459                                 return -errno;
 460                 }
 461
 462                 for (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
 466                         if (q->inheritable != (uint64_t) -1) {
 467                                 cap_flag_value_t old_value, new_value;
 468
 469                                 if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value) < 0) {
 470                                         if (errno == EINVAL) /* If the kernel knows more caps than this
 471                                                               * version of libcap, then this will return
 472                                                               * EINVAL. In that case, simply ignore it,
 473                                                               * pretend it doesn't exist. */
 474                                                 continue;
 475
 476                                         return -errno;
 477                                 }
 478
 479                                 new_value = (q->inheritable & m) ? CAP_SET : CAP_CLEAR;
 480
 481                                 if (old_value != new_value) {
 482                                         changed = true;
 483
 484                                         if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, new_value) < 0)
 485                                                 return -errno;
 486                                 }
 487                         }
 488
 489                         if (q->permitted != (uint64_t) -1) {
 490                                 cap_flag_value_t old_value, new_value;
 491
 492                                 if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value) < 0) {
 493                                         if (errno == EINVAL)
 494                                                 continue;
 495
 496                                         return -errno;
 497                                 }
 498
 499                                 new_value = (q->permitted & m) ? CAP_SET : CAP_CLEAR;
 500
 501                                 if (old_value != new_value) {
 502                                         changed = true;
 503
 504                                         if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, new_value) < 0)
 505                                                 return -errno;
 506                                 }
 507                         }
 508
 509                         if (q->effective != (uint64_t) -1) {
 510                                 cap_flag_value_t old_value, new_value;
 511
 512                                 if (cap_get_flag(c, cv, CAP_EFFECTIVE, &old_value) < 0) {
 513                                         if (errno == EINVAL)
 514                                                 continue;
 515
 516                                         return -errno;
 517                                 }
 518
 519                                 new_value = (q->effective & m) ? CAP_SET : CAP_CLEAR;
 520
 521                                 if (old_value != new_value) {
 522                                         changed = true;
 523
 524                                         if (cap_set_flag(c, CAP_EFFECTIVE, 1, &cv, new_value) < 0)
 525                                                 return -errno;
 526                                 }
 527                         }
 528                 }
 529
 530                 if (changed) {
 531                         /* In order to change the bounding caps, we need to keep CAP_SETPCAP for a bit
 532                          * longer. Let's add it to our list hence for now. */
 533                         if (q->bounding != (uint64_t) -1) {
 534                                 cap_value_t cv = CAP_SETPCAP;
 535
 536                                 modified = cap_dup(c);
 537                                 if (!modified)
 538                                         return -ENOMEM;
 539
 540                                 if (cap_set_flag(modified, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
 541                                         return -errno;
 542                                 if (cap_set_flag(modified, CAP_EFFECTIVE, 1, &cv, CAP_SET) < 0)
 543                                         return -errno;
 544
 545                                 if (cap_compare(modified, c) == 0) {
 546                                         /* No change? then drop this nonsense again */
 547                                         cap_free(modified);
 548                                         modified = NULL;
 549                                 }
 550                         }
 551
 552                         /* Now, let's enforce the caps for the first time. Note that this is where we acquire
 553                          * caps in any of the sets we currently don't have. We have to do this before
 554                          * dropping the bounding caps below, since at that point we can never acquire new
 555                          * caps in inherited/permitted/effective anymore, but only lose them. */
 556                         if (cap_set_proc(modified ?: c) < 0)
 557                                 return -errno;
 558                 }
 559         }
 560
 561         if (q->bounding != (uint64_t) -1) {
 562                 r = capability_bounding_set_drop(q->bounding, false);
 563                 if (r < 0)
 564                         return r;
 565         }
 566
 567         /* If needed, let's now set the caps again, this time in the final version, which differs from what
 568          * we have already set only in the CAP_SETPCAP bit, which we needed for dropping the bounding
 569          * bits. This call only undoes bits and doesn't acquire any which means the bounding caps don't
 570          * matter. */
 571         if (modified)
 572                 if (cap_set_proc(c) < 0)
 573                         return -errno;
 574
 575         return 0;
 576 }