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