1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/kernel/sys.c 4 * 5 * Copyright (C) 1991, 1992 Linus Torvalds 6 */ 7 8 #include <linux/export.h> 9 #include <linux/mm.h> 10 #include <linux/mm_inline.h> 11 #include <linux/utsname.h> 12 #include <linux/mman.h> 13 #include <linux/reboot.h> 14 #include <linux/prctl.h> 15 #include <linux/highuid.h> 16 #include <linux/fs.h> 17 #include <linux/kmod.h> 18 #include <linux/perf_event.h> 19 #include <linux/resource.h> 20 #include <linux/kernel.h> 21 #include <linux/workqueue.h> 22 #include <linux/capability.h> 23 #include <linux/device.h> 24 #include <linux/key.h> 25 #include <linux/times.h> 26 #include <linux/posix-timers.h> 27 #include <linux/security.h> 28 #include <linux/suspend.h> 29 #include <linux/tty.h> 30 #include <linux/signal.h> 31 #include <linux/cn_proc.h> 32 #include <linux/getcpu.h> 33 #include <linux/task_io_accounting_ops.h> 34 #include <linux/seccomp.h> 35 #include <linux/cpu.h> 36 #include <linux/personality.h> 37 #include <linux/ptrace.h> 38 #include <linux/fs_struct.h> 39 #include <linux/file.h> 40 #include <linux/mount.h> 41 #include <linux/gfp.h> 42 #include <linux/syscore_ops.h> 43 #include <linux/version.h> 44 #include <linux/ctype.h> 45 #include <linux/syscall_user_dispatch.h> 46 47 #include <linux/compat.h> 48 #include <linux/syscalls.h> 49 #include <linux/kprobes.h> 50 #include <linux/user_namespace.h> 51 #include <linux/time_namespace.h> 52 #include <linux/binfmts.h> 53 54 #include <linux/sched.h> 55 #include <linux/sched/autogroup.h> 56 #include <linux/sched/loadavg.h> 57 #include <linux/sched/stat.h> 58 #include <linux/sched/mm.h> 59 #include <linux/sched/coredump.h> 60 #include <linux/sched/task.h> 61 #include <linux/sched/cputime.h> 62 #include <linux/rcupdate.h> 63 #include <linux/uidgid.h> 64 #include <linux/cred.h> 65 66 #include <linux/nospec.h> 67 68 #include <linux/kmsg_dump.h> 69 /* Move somewhere else to avoid recompiling? */ 70 #include <generated/utsrelease.h> 71 72 #include <linux/uaccess.h> 73 #include <asm/io.h> 74 #include <asm/unistd.h> 75 76 #include "uid16.h" 77 78 #ifndef SET_UNALIGN_CTL 79 # define SET_UNALIGN_CTL(a, b) (-EINVAL) 80 #endif 81 #ifndef GET_UNALIGN_CTL 82 # define GET_UNALIGN_CTL(a, b) (-EINVAL) 83 #endif 84 #ifndef SET_FPEMU_CTL 85 # define SET_FPEMU_CTL(a, b) (-EINVAL) 86 #endif 87 #ifndef GET_FPEMU_CTL 88 # define GET_FPEMU_CTL(a, b) (-EINVAL) 89 #endif 90 #ifndef SET_FPEXC_CTL 91 # define SET_FPEXC_CTL(a, b) (-EINVAL) 92 #endif 93 #ifndef GET_FPEXC_CTL 94 # define GET_FPEXC_CTL(a, b) (-EINVAL) 95 #endif 96 #ifndef GET_ENDIAN 97 # define GET_ENDIAN(a, b) (-EINVAL) 98 #endif 99 #ifndef SET_ENDIAN 100 # define SET_ENDIAN(a, b) (-EINVAL) 101 #endif 102 #ifndef GET_TSC_CTL 103 # define GET_TSC_CTL(a) (-EINVAL) 104 #endif 105 #ifndef SET_TSC_CTL 106 # define SET_TSC_CTL(a) (-EINVAL) 107 #endif 108 #ifndef GET_FP_MODE 109 # define GET_FP_MODE(a) (-EINVAL) 110 #endif 111 #ifndef SET_FP_MODE 112 # define SET_FP_MODE(a,b) (-EINVAL) 113 #endif 114 #ifndef SVE_SET_VL 115 # define SVE_SET_VL(a) (-EINVAL) 116 #endif 117 #ifndef SVE_GET_VL 118 # define SVE_GET_VL() (-EINVAL) 119 #endif 120 #ifndef PAC_RESET_KEYS 121 # define PAC_RESET_KEYS(a, b) (-EINVAL) 122 #endif 123 #ifndef PAC_SET_ENABLED_KEYS 124 # define PAC_SET_ENABLED_KEYS(a, b, c) (-EINVAL) 125 #endif 126 #ifndef PAC_GET_ENABLED_KEYS 127 # define PAC_GET_ENABLED_KEYS(a) (-EINVAL) 128 #endif 129 #ifndef SET_TAGGED_ADDR_CTRL 130 # define SET_TAGGED_ADDR_CTRL(a) (-EINVAL) 131 #endif 132 #ifndef GET_TAGGED_ADDR_CTRL 133 # define GET_TAGGED_ADDR_CTRL() (-EINVAL) 134 #endif 135 136 /* 137 * this is where the system-wide overflow UID and GID are defined, for 138 * architectures that now have 32-bit UID/GID but didn't in the past 139 */ 140 141 int overflowuid = DEFAULT_OVERFLOWUID; 142 int overflowgid = DEFAULT_OVERFLOWGID; 143 144 EXPORT_SYMBOL(overflowuid); 145 EXPORT_SYMBOL(overflowgid); 146 147 /* 148 * the same as above, but for filesystems which can only store a 16-bit 149 * UID and GID. as such, this is needed on all architectures 150 */ 151 152 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID; 153 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID; 154 155 EXPORT_SYMBOL(fs_overflowuid); 156 EXPORT_SYMBOL(fs_overflowgid); 157 158 /* 159 * Returns true if current's euid is same as p's uid or euid, 160 * or has CAP_SYS_NICE to p's user_ns. 161 * 162 * Called with rcu_read_lock, creds are safe 163 */ 164 static bool set_one_prio_perm(struct task_struct *p) 165 { 166 const struct cred *cred = current_cred(), *pcred = __task_cred(p); 167 168 if (uid_eq(pcred->uid, cred->euid) || 169 uid_eq(pcred->euid, cred->euid)) 170 return true; 171 if (ns_capable(pcred->user_ns, CAP_SYS_NICE)) 172 return true; 173 return false; 174 } 175 176 /* 177 * set the priority of a task 178 * - the caller must hold the RCU read lock 179 */ 180 static int set_one_prio(struct task_struct *p, int niceval, int error) 181 { 182 int no_nice; 183 184 if (!set_one_prio_perm(p)) { 185 error = -EPERM; 186 goto out; 187 } 188 if (niceval < task_nice(p) && !can_nice(p, niceval)) { 189 error = -EACCES; 190 goto out; 191 } 192 no_nice = security_task_setnice(p, niceval); 193 if (no_nice) { 194 error = no_nice; 195 goto out; 196 } 197 if (error == -ESRCH) 198 error = 0; 199 set_user_nice(p, niceval); 200 out: 201 return error; 202 } 203 204 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval) 205 { 206 struct task_struct *g, *p; 207 struct user_struct *user; 208 const struct cred *cred = current_cred(); 209 int error = -EINVAL; 210 struct pid *pgrp; 211 kuid_t uid; 212 213 if (which > PRIO_USER || which < PRIO_PROCESS) 214 goto out; 215 216 /* normalize: avoid signed division (rounding problems) */ 217 error = -ESRCH; 218 if (niceval < MIN_NICE) 219 niceval = MIN_NICE; 220 if (niceval > MAX_NICE) 221 niceval = MAX_NICE; 222 223 rcu_read_lock(); 224 switch (which) { 225 case PRIO_PROCESS: 226 if (who) 227 p = find_task_by_vpid(who); 228 else 229 p = current; 230 if (p) 231 error = set_one_prio(p, niceval, error); 232 break; 233 case PRIO_PGRP: 234 if (who) 235 pgrp = find_vpid(who); 236 else 237 pgrp = task_pgrp(current); 238 read_lock(&tasklist_lock); 239 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) { 240 error = set_one_prio(p, niceval, error); 241 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p); 242 read_unlock(&tasklist_lock); 243 break; 244 case PRIO_USER: 245 uid = make_kuid(cred->user_ns, who); 246 user = cred->user; 247 if (!who) 248 uid = cred->uid; 249 else if (!uid_eq(uid, cred->uid)) { 250 user = find_user(uid); 251 if (!user) 252 goto out_unlock; /* No processes for this user */ 253 } 254 for_each_process_thread(g, p) { 255 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) 256 error = set_one_prio(p, niceval, error); 257 } 258 if (!uid_eq(uid, cred->uid)) 259 free_uid(user); /* For find_user() */ 260 break; 261 } 262 out_unlock: 263 rcu_read_unlock(); 264 out: 265 return error; 266 } 267 268 /* 269 * Ugh. To avoid negative return values, "getpriority()" will 270 * not return the normal nice-value, but a negated value that 271 * has been offset by 20 (ie it returns 40..1 instead of -20..19) 272 * to stay compatible. 273 */ 274 SYSCALL_DEFINE2(getpriority, int, which, int, who) 275 { 276 struct task_struct *g, *p; 277 struct user_struct *user; 278 const struct cred *cred = current_cred(); 279 long niceval, retval = -ESRCH; 280 struct pid *pgrp; 281 kuid_t uid; 282 283 if (which > PRIO_USER || which < PRIO_PROCESS) 284 return -EINVAL; 285 286 rcu_read_lock(); 287 switch (which) { 288 case PRIO_PROCESS: 289 if (who) 290 p = find_task_by_vpid(who); 291 else 292 p = current; 293 if (p) { 294 niceval = nice_to_rlimit(task_nice(p)); 295 if (niceval > retval) 296 retval = niceval; 297 } 298 break; 299 case PRIO_PGRP: 300 if (who) 301 pgrp = find_vpid(who); 302 else 303 pgrp = task_pgrp(current); 304 read_lock(&tasklist_lock); 305 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) { 306 niceval = nice_to_rlimit(task_nice(p)); 307 if (niceval > retval) 308 retval = niceval; 309 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p); 310 read_unlock(&tasklist_lock); 311 break; 312 case PRIO_USER: 313 uid = make_kuid(cred->user_ns, who); 314 user = cred->user; 315 if (!who) 316 uid = cred->uid; 317 else if (!uid_eq(uid, cred->uid)) { 318 user = find_user(uid); 319 if (!user) 320 goto out_unlock; /* No processes for this user */ 321 } 322 for_each_process_thread(g, p) { 323 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) { 324 niceval = nice_to_rlimit(task_nice(p)); 325 if (niceval > retval) 326 retval = niceval; 327 } 328 } 329 if (!uid_eq(uid, cred->uid)) 330 free_uid(user); /* for find_user() */ 331 break; 332 } 333 out_unlock: 334 rcu_read_unlock(); 335 336 return retval; 337 } 338 339 /* 340 * Unprivileged users may change the real gid to the effective gid 341 * or vice versa. (BSD-style) 342 * 343 * If you set the real gid at all, or set the effective gid to a value not 344 * equal to the real gid, then the saved gid is set to the new effective gid. 345 * 346 * This makes it possible for a setgid program to completely drop its 347 * privileges, which is often a useful assertion to make when you are doing 348 * a security audit over a program. 349 * 350 * The general idea is that a program which uses just setregid() will be 351 * 100% compatible with BSD. A program which uses just setgid() will be 352 * 100% compatible with POSIX with saved IDs. 353 * 354 * SMP: There are not races, the GIDs are checked only by filesystem 355 * operations (as far as semantic preservation is concerned). 356 */ 357 #ifdef CONFIG_MULTIUSER 358 long __sys_setregid(gid_t rgid, gid_t egid) 359 { 360 struct user_namespace *ns = current_user_ns(); 361 const struct cred *old; 362 struct cred *new; 363 int retval; 364 kgid_t krgid, kegid; 365 366 krgid = make_kgid(ns, rgid); 367 kegid = make_kgid(ns, egid); 368 369 if ((rgid != (gid_t) -1) && !gid_valid(krgid)) 370 return -EINVAL; 371 if ((egid != (gid_t) -1) && !gid_valid(kegid)) 372 return -EINVAL; 373 374 new = prepare_creds(); 375 if (!new) 376 return -ENOMEM; 377 old = current_cred(); 378 379 retval = -EPERM; 380 if (rgid != (gid_t) -1) { 381 if (gid_eq(old->gid, krgid) || 382 gid_eq(old->egid, krgid) || 383 ns_capable_setid(old->user_ns, CAP_SETGID)) 384 new->gid = krgid; 385 else 386 goto error; 387 } 388 if (egid != (gid_t) -1) { 389 if (gid_eq(old->gid, kegid) || 390 gid_eq(old->egid, kegid) || 391 gid_eq(old->sgid, kegid) || 392 ns_capable_setid(old->user_ns, CAP_SETGID)) 393 new->egid = kegid; 394 else 395 goto error; 396 } 397 398 if (rgid != (gid_t) -1 || 399 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid))) 400 new->sgid = new->egid; 401 new->fsgid = new->egid; 402 403 retval = security_task_fix_setgid(new, old, LSM_SETID_RE); 404 if (retval < 0) 405 goto error; 406 407 return commit_creds(new); 408 409 error: 410 abort_creds(new); 411 return retval; 412 } 413 414 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid) 415 { 416 return __sys_setregid(rgid, egid); 417 } 418 419 /* 420 * setgid() is implemented like SysV w/ SAVED_IDS 421 * 422 * SMP: Same implicit races as above. 423 */ 424 long __sys_setgid(gid_t gid) 425 { 426 struct user_namespace *ns = current_user_ns(); 427 const struct cred *old; 428 struct cred *new; 429 int retval; 430 kgid_t kgid; 431 432 kgid = make_kgid(ns, gid); 433 if (!gid_valid(kgid)) 434 return -EINVAL; 435 436 new = prepare_creds(); 437 if (!new) 438 return -ENOMEM; 439 old = current_cred(); 440 441 retval = -EPERM; 442 if (ns_capable_setid(old->user_ns, CAP_SETGID)) 443 new->gid = new->egid = new->sgid = new->fsgid = kgid; 444 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid)) 445 new->egid = new->fsgid = kgid; 446 else 447 goto error; 448 449 retval = security_task_fix_setgid(new, old, LSM_SETID_ID); 450 if (retval < 0) 451 goto error; 452 453 return commit_creds(new); 454 455 error: 456 abort_creds(new); 457 return retval; 458 } 459 460 SYSCALL_DEFINE1(setgid, gid_t, gid) 461 { 462 return __sys_setgid(gid); 463 } 464 465 /* 466 * change the user struct in a credentials set to match the new UID 467 */ 468 static int set_user(struct cred *new) 469 { 470 struct user_struct *new_user; 471 472 new_user = alloc_uid(new->uid); 473 if (!new_user) 474 return -EAGAIN; 475 476 free_uid(new->user); 477 new->user = new_user; 478 return 0; 479 } 480 481 static void flag_nproc_exceeded(struct cred *new) 482 { 483 if (new->ucounts == current_ucounts()) 484 return; 485 486 /* 487 * We don't fail in case of NPROC limit excess here because too many 488 * poorly written programs don't check set*uid() return code, assuming 489 * it never fails if called by root. We may still enforce NPROC limit 490 * for programs doing set*uid()+execve() by harmlessly deferring the 491 * failure to the execve() stage. 492 */ 493 if (is_ucounts_overlimit(new->ucounts, UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)) && 494 new->user != INIT_USER) 495 current->flags |= PF_NPROC_EXCEEDED; 496 else 497 current->flags &= ~PF_NPROC_EXCEEDED; 498 } 499 500 /* 501 * Unprivileged users may change the real uid to the effective uid 502 * or vice versa. (BSD-style) 503 * 504 * If you set the real uid at all, or set the effective uid to a value not 505 * equal to the real uid, then the saved uid is set to the new effective uid. 506 * 507 * This makes it possible for a setuid program to completely drop its 508 * privileges, which is often a useful assertion to make when you are doing 509 * a security audit over a program. 510 * 511 * The general idea is that a program which uses just setreuid() will be 512 * 100% compatible with BSD. A program which uses just setuid() will be 513 * 100% compatible with POSIX with saved IDs. 514 */ 515 long __sys_setreuid(uid_t ruid, uid_t euid) 516 { 517 struct user_namespace *ns = current_user_ns(); 518 const struct cred *old; 519 struct cred *new; 520 int retval; 521 kuid_t kruid, keuid; 522 523 kruid = make_kuid(ns, ruid); 524 keuid = make_kuid(ns, euid); 525 526 if ((ruid != (uid_t) -1) && !uid_valid(kruid)) 527 return -EINVAL; 528 if ((euid != (uid_t) -1) && !uid_valid(keuid)) 529 return -EINVAL; 530 531 new = prepare_creds(); 532 if (!new) 533 return -ENOMEM; 534 old = current_cred(); 535 536 retval = -EPERM; 537 if (ruid != (uid_t) -1) { 538 new->uid = kruid; 539 if (!uid_eq(old->uid, kruid) && 540 !uid_eq(old->euid, kruid) && 541 !ns_capable_setid(old->user_ns, CAP_SETUID)) 542 goto error; 543 } 544 545 if (euid != (uid_t) -1) { 546 new->euid = keuid; 547 if (!uid_eq(old->uid, keuid) && 548 !uid_eq(old->euid, keuid) && 549 !uid_eq(old->suid, keuid) && 550 !ns_capable_setid(old->user_ns, CAP_SETUID)) 551 goto error; 552 } 553 554 if (!uid_eq(new->uid, old->uid)) { 555 retval = set_user(new); 556 if (retval < 0) 557 goto error; 558 } 559 if (ruid != (uid_t) -1 || 560 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid))) 561 new->suid = new->euid; 562 new->fsuid = new->euid; 563 564 retval = security_task_fix_setuid(new, old, LSM_SETID_RE); 565 if (retval < 0) 566 goto error; 567 568 retval = set_cred_ucounts(new); 569 if (retval < 0) 570 goto error; 571 572 flag_nproc_exceeded(new); 573 return commit_creds(new); 574 575 error: 576 abort_creds(new); 577 return retval; 578 } 579 580 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid) 581 { 582 return __sys_setreuid(ruid, euid); 583 } 584 585 /* 586 * setuid() is implemented like SysV with SAVED_IDS 587 * 588 * Note that SAVED_ID's is deficient in that a setuid root program 589 * like sendmail, for example, cannot set its uid to be a normal 590 * user and then switch back, because if you're root, setuid() sets 591 * the saved uid too. If you don't like this, blame the bright people 592 * in the POSIX committee and/or USG. Note that the BSD-style setreuid() 593 * will allow a root program to temporarily drop privileges and be able to 594 * regain them by swapping the real and effective uid. 595 */ 596 long __sys_setuid(uid_t uid) 597 { 598 struct user_namespace *ns = current_user_ns(); 599 const struct cred *old; 600 struct cred *new; 601 int retval; 602 kuid_t kuid; 603 604 kuid = make_kuid(ns, uid); 605 if (!uid_valid(kuid)) 606 return -EINVAL; 607 608 new = prepare_creds(); 609 if (!new) 610 return -ENOMEM; 611 old = current_cred(); 612 613 retval = -EPERM; 614 if (ns_capable_setid(old->user_ns, CAP_SETUID)) { 615 new->suid = new->uid = kuid; 616 if (!uid_eq(kuid, old->uid)) { 617 retval = set_user(new); 618 if (retval < 0) 619 goto error; 620 } 621 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) { 622 goto error; 623 } 624 625 new->fsuid = new->euid = kuid; 626 627 retval = security_task_fix_setuid(new, old, LSM_SETID_ID); 628 if (retval < 0) 629 goto error; 630 631 retval = set_cred_ucounts(new); 632 if (retval < 0) 633 goto error; 634 635 flag_nproc_exceeded(new); 636 return commit_creds(new); 637 638 error: 639 abort_creds(new); 640 return retval; 641 } 642 643 SYSCALL_DEFINE1(setuid, uid_t, uid) 644 { 645 return __sys_setuid(uid); 646 } 647 648 649 /* 650 * This function implements a generic ability to update ruid, euid, 651 * and suid. This allows you to implement the 4.4 compatible seteuid(). 652 */ 653 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid) 654 { 655 struct user_namespace *ns = current_user_ns(); 656 const struct cred *old; 657 struct cred *new; 658 int retval; 659 kuid_t kruid, keuid, ksuid; 660 661 kruid = make_kuid(ns, ruid); 662 keuid = make_kuid(ns, euid); 663 ksuid = make_kuid(ns, suid); 664 665 if ((ruid != (uid_t) -1) && !uid_valid(kruid)) 666 return -EINVAL; 667 668 if ((euid != (uid_t) -1) && !uid_valid(keuid)) 669 return -EINVAL; 670 671 if ((suid != (uid_t) -1) && !uid_valid(ksuid)) 672 return -EINVAL; 673 674 new = prepare_creds(); 675 if (!new) 676 return -ENOMEM; 677 678 old = current_cred(); 679 680 retval = -EPERM; 681 if (!ns_capable_setid(old->user_ns, CAP_SETUID)) { 682 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) && 683 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid)) 684 goto error; 685 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) && 686 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid)) 687 goto error; 688 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) && 689 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid)) 690 goto error; 691 } 692 693 if (ruid != (uid_t) -1) { 694 new->uid = kruid; 695 if (!uid_eq(kruid, old->uid)) { 696 retval = set_user(new); 697 if (retval < 0) 698 goto error; 699 } 700 } 701 if (euid != (uid_t) -1) 702 new->euid = keuid; 703 if (suid != (uid_t) -1) 704 new->suid = ksuid; 705 new->fsuid = new->euid; 706 707 retval = security_task_fix_setuid(new, old, LSM_SETID_RES); 708 if (retval < 0) 709 goto error; 710 711 retval = set_cred_ucounts(new); 712 if (retval < 0) 713 goto error; 714 715 flag_nproc_exceeded(new); 716 return commit_creds(new); 717 718 error: 719 abort_creds(new); 720 return retval; 721 } 722 723 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid) 724 { 725 return __sys_setresuid(ruid, euid, suid); 726 } 727 728 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp) 729 { 730 const struct cred *cred = current_cred(); 731 int retval; 732 uid_t ruid, euid, suid; 733 734 ruid = from_kuid_munged(cred->user_ns, cred->uid); 735 euid = from_kuid_munged(cred->user_ns, cred->euid); 736 suid = from_kuid_munged(cred->user_ns, cred->suid); 737 738 retval = put_user(ruid, ruidp); 739 if (!retval) { 740 retval = put_user(euid, euidp); 741 if (!retval) 742 return put_user(suid, suidp); 743 } 744 return retval; 745 } 746 747 /* 748 * Same as above, but for rgid, egid, sgid. 749 */ 750 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid) 751 { 752 struct user_namespace *ns = current_user_ns(); 753 const struct cred *old; 754 struct cred *new; 755 int retval; 756 kgid_t krgid, kegid, ksgid; 757 758 krgid = make_kgid(ns, rgid); 759 kegid = make_kgid(ns, egid); 760 ksgid = make_kgid(ns, sgid); 761 762 if ((rgid != (gid_t) -1) && !gid_valid(krgid)) 763 return -EINVAL; 764 if ((egid != (gid_t) -1) && !gid_valid(kegid)) 765 return -EINVAL; 766 if ((sgid != (gid_t) -1) && !gid_valid(ksgid)) 767 return -EINVAL; 768 769 new = prepare_creds(); 770 if (!new) 771 return -ENOMEM; 772 old = current_cred(); 773 774 retval = -EPERM; 775 if (!ns_capable_setid(old->user_ns, CAP_SETGID)) { 776 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) && 777 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid)) 778 goto error; 779 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) && 780 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid)) 781 goto error; 782 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) && 783 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid)) 784 goto error; 785 } 786 787 if (rgid != (gid_t) -1) 788 new->gid = krgid; 789 if (egid != (gid_t) -1) 790 new->egid = kegid; 791 if (sgid != (gid_t) -1) 792 new->sgid = ksgid; 793 new->fsgid = new->egid; 794 795 retval = security_task_fix_setgid(new, old, LSM_SETID_RES); 796 if (retval < 0) 797 goto error; 798 799 return commit_creds(new); 800 801 error: 802 abort_creds(new); 803 return retval; 804 } 805 806 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid) 807 { 808 return __sys_setresgid(rgid, egid, sgid); 809 } 810 811 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp) 812 { 813 const struct cred *cred = current_cred(); 814 int retval; 815 gid_t rgid, egid, sgid; 816 817 rgid = from_kgid_munged(cred->user_ns, cred->gid); 818 egid = from_kgid_munged(cred->user_ns, cred->egid); 819 sgid = from_kgid_munged(cred->user_ns, cred->sgid); 820 821 retval = put_user(rgid, rgidp); 822 if (!retval) { 823 retval = put_user(egid, egidp); 824 if (!retval) 825 retval = put_user(sgid, sgidp); 826 } 827 828 return retval; 829 } 830 831 832 /* 833 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This 834 * is used for "access()" and for the NFS daemon (letting nfsd stay at 835 * whatever uid it wants to). It normally shadows "euid", except when 836 * explicitly set by setfsuid() or for access.. 837 */ 838 long __sys_setfsuid(uid_t uid) 839 { 840 const struct cred *old; 841 struct cred *new; 842 uid_t old_fsuid; 843 kuid_t kuid; 844 845 old = current_cred(); 846 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid); 847 848 kuid = make_kuid(old->user_ns, uid); 849 if (!uid_valid(kuid)) 850 return old_fsuid; 851 852 new = prepare_creds(); 853 if (!new) 854 return old_fsuid; 855 856 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) || 857 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) || 858 ns_capable_setid(old->user_ns, CAP_SETUID)) { 859 if (!uid_eq(kuid, old->fsuid)) { 860 new->fsuid = kuid; 861 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0) 862 goto change_okay; 863 } 864 } 865 866 abort_creds(new); 867 return old_fsuid; 868 869 change_okay: 870 commit_creds(new); 871 return old_fsuid; 872 } 873 874 SYSCALL_DEFINE1(setfsuid, uid_t, uid) 875 { 876 return __sys_setfsuid(uid); 877 } 878 879 /* 880 * Samma på svenska.. 881 */ 882 long __sys_setfsgid(gid_t gid) 883 { 884 const struct cred *old; 885 struct cred *new; 886 gid_t old_fsgid; 887 kgid_t kgid; 888 889 old = current_cred(); 890 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid); 891 892 kgid = make_kgid(old->user_ns, gid); 893 if (!gid_valid(kgid)) 894 return old_fsgid; 895 896 new = prepare_creds(); 897 if (!new) 898 return old_fsgid; 899 900 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) || 901 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) || 902 ns_capable_setid(old->user_ns, CAP_SETGID)) { 903 if (!gid_eq(kgid, old->fsgid)) { 904 new->fsgid = kgid; 905 if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0) 906 goto change_okay; 907 } 908 } 909 910 abort_creds(new); 911 return old_fsgid; 912 913 change_okay: 914 commit_creds(new); 915 return old_fsgid; 916 } 917 918 SYSCALL_DEFINE1(setfsgid, gid_t, gid) 919 { 920 return __sys_setfsgid(gid); 921 } 922 #endif /* CONFIG_MULTIUSER */ 923 924 /** 925 * sys_getpid - return the thread group id of the current process 926 * 927 * Note, despite the name, this returns the tgid not the pid. The tgid and 928 * the pid are identical unless CLONE_THREAD was specified on clone() in 929 * which case the tgid is the same in all threads of the same group. 930 * 931 * This is SMP safe as current->tgid does not change. 932 */ 933 SYSCALL_DEFINE0(getpid) 934 { 935 return task_tgid_vnr(current); 936 } 937 938 /* Thread ID - the internal kernel "pid" */ 939 SYSCALL_DEFINE0(gettid) 940 { 941 return task_pid_vnr(current); 942 } 943 944 /* 945 * Accessing ->real_parent is not SMP-safe, it could 946 * change from under us. However, we can use a stale 947 * value of ->real_parent under rcu_read_lock(), see 948 * release_task()->call_rcu(delayed_put_task_struct). 949 */ 950 SYSCALL_DEFINE0(getppid) 951 { 952 int pid; 953 954 rcu_read_lock(); 955 pid = task_tgid_vnr(rcu_dereference(current->real_parent)); 956 rcu_read_unlock(); 957 958 return pid; 959 } 960 961 SYSCALL_DEFINE0(getuid) 962 { 963 /* Only we change this so SMP safe */ 964 return from_kuid_munged(current_user_ns(), current_uid()); 965 } 966 967 SYSCALL_DEFINE0(geteuid) 968 { 969 /* Only we change this so SMP safe */ 970 return from_kuid_munged(current_user_ns(), current_euid()); 971 } 972 973 SYSCALL_DEFINE0(getgid) 974 { 975 /* Only we change this so SMP safe */ 976 return from_kgid_munged(current_user_ns(), current_gid()); 977 } 978 979 SYSCALL_DEFINE0(getegid) 980 { 981 /* Only we change this so SMP safe */ 982 return from_kgid_munged(current_user_ns(), current_egid()); 983 } 984 985 static void do_sys_times(struct tms *tms) 986 { 987 u64 tgutime, tgstime, cutime, cstime; 988 989 thread_group_cputime_adjusted(current, &tgutime, &tgstime); 990 cutime = current->signal->cutime; 991 cstime = current->signal->cstime; 992 tms->tms_utime = nsec_to_clock_t(tgutime); 993 tms->tms_stime = nsec_to_clock_t(tgstime); 994 tms->tms_cutime = nsec_to_clock_t(cutime); 995 tms->tms_cstime = nsec_to_clock_t(cstime); 996 } 997 998 SYSCALL_DEFINE1(times, struct tms __user *, tbuf) 999 { 1000 if (tbuf) { 1001 struct tms tmp; 1002 1003 do_sys_times(&tmp); 1004 if (copy_to_user(tbuf, &tmp, sizeof(struct tms))) 1005 return -EFAULT; 1006 } 1007 force_successful_syscall_return(); 1008 return (long) jiffies_64_to_clock_t(get_jiffies_64()); 1009 } 1010 1011 #ifdef CONFIG_COMPAT 1012 static compat_clock_t clock_t_to_compat_clock_t(clock_t x) 1013 { 1014 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x)); 1015 } 1016 1017 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf) 1018 { 1019 if (tbuf) { 1020 struct tms tms; 1021 struct compat_tms tmp; 1022 1023 do_sys_times(&tms); 1024 /* Convert our struct tms to the compat version. */ 1025 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime); 1026 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime); 1027 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime); 1028 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime); 1029 if (copy_to_user(tbuf, &tmp, sizeof(tmp))) 1030 return -EFAULT; 1031 } 1032 force_successful_syscall_return(); 1033 return compat_jiffies_to_clock_t(jiffies); 1034 } 1035 #endif 1036 1037 /* 1038 * This needs some heavy checking ... 1039 * I just haven't the stomach for it. I also don't fully 1040 * understand sessions/pgrp etc. Let somebody who does explain it. 1041 * 1042 * OK, I think I have the protection semantics right.... this is really 1043 * only important on a multi-user system anyway, to make sure one user 1044 * can't send a signal to a process owned by another. -TYT, 12/12/91 1045 * 1046 * !PF_FORKNOEXEC check to conform completely to POSIX. 1047 */ 1048 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid) 1049 { 1050 struct task_struct *p; 1051 struct task_struct *group_leader = current->group_leader; 1052 struct pid *pgrp; 1053 int err; 1054 1055 if (!pid) 1056 pid = task_pid_vnr(group_leader); 1057 if (!pgid) 1058 pgid = pid; 1059 if (pgid < 0) 1060 return -EINVAL; 1061 rcu_read_lock(); 1062 1063 /* From this point forward we keep holding onto the tasklist lock 1064 * so that our parent does not change from under us. -DaveM 1065 */ 1066 write_lock_irq(&tasklist_lock); 1067 1068 err = -ESRCH; 1069 p = find_task_by_vpid(pid); 1070 if (!p) 1071 goto out; 1072 1073 err = -EINVAL; 1074 if (!thread_group_leader(p)) 1075 goto out; 1076 1077 if (same_thread_group(p->real_parent, group_leader)) { 1078 err = -EPERM; 1079 if (task_session(p) != task_session(group_leader)) 1080 goto out; 1081 err = -EACCES; 1082 if (!(p->flags & PF_FORKNOEXEC)) 1083 goto out; 1084 } else { 1085 err = -ESRCH; 1086 if (p != group_leader) 1087 goto out; 1088 } 1089 1090 err = -EPERM; 1091 if (p->signal->leader) 1092 goto out; 1093 1094 pgrp = task_pid(p); 1095 if (pgid != pid) { 1096 struct task_struct *g; 1097 1098 pgrp = find_vpid(pgid); 1099 g = pid_task(pgrp, PIDTYPE_PGID); 1100 if (!g || task_session(g) != task_session(group_leader)) 1101 goto out; 1102 } 1103 1104 err = security_task_setpgid(p, pgid); 1105 if (err) 1106 goto out; 1107 1108 if (task_pgrp(p) != pgrp) 1109 change_pid(p, PIDTYPE_PGID, pgrp); 1110 1111 err = 0; 1112 out: 1113 /* All paths lead to here, thus we are safe. -DaveM */ 1114 write_unlock_irq(&tasklist_lock); 1115 rcu_read_unlock(); 1116 return err; 1117 } 1118 1119 static int do_getpgid(pid_t pid) 1120 { 1121 struct task_struct *p; 1122 struct pid *grp; 1123 int retval; 1124 1125 rcu_read_lock(); 1126 if (!pid) 1127 grp = task_pgrp(current); 1128 else { 1129 retval = -ESRCH; 1130 p = find_task_by_vpid(pid); 1131 if (!p) 1132 goto out; 1133 grp = task_pgrp(p); 1134 if (!grp) 1135 goto out; 1136 1137 retval = security_task_getpgid(p); 1138 if (retval) 1139 goto out; 1140 } 1141 retval = pid_vnr(grp); 1142 out: 1143 rcu_read_unlock(); 1144 return retval; 1145 } 1146 1147 SYSCALL_DEFINE1(getpgid, pid_t, pid) 1148 { 1149 return do_getpgid(pid); 1150 } 1151 1152 #ifdef __ARCH_WANT_SYS_GETPGRP 1153 1154 SYSCALL_DEFINE0(getpgrp) 1155 { 1156 return do_getpgid(0); 1157 } 1158 1159 #endif 1160 1161 SYSCALL_DEFINE1(getsid, pid_t, pid) 1162 { 1163 struct task_struct *p; 1164 struct pid *sid; 1165 int retval; 1166 1167 rcu_read_lock(); 1168 if (!pid) 1169 sid = task_session(current); 1170 else { 1171 retval = -ESRCH; 1172 p = find_task_by_vpid(pid); 1173 if (!p) 1174 goto out; 1175 sid = task_session(p); 1176 if (!sid) 1177 goto out; 1178 1179 retval = security_task_getsid(p); 1180 if (retval) 1181 goto out; 1182 } 1183 retval = pid_vnr(sid); 1184 out: 1185 rcu_read_unlock(); 1186 return retval; 1187 } 1188 1189 static void set_special_pids(struct pid *pid) 1190 { 1191 struct task_struct *curr = current->group_leader; 1192 1193 if (task_session(curr) != pid) 1194 change_pid(curr, PIDTYPE_SID, pid); 1195 1196 if (task_pgrp(curr) != pid) 1197 change_pid(curr, PIDTYPE_PGID, pid); 1198 } 1199 1200 int ksys_setsid(void) 1201 { 1202 struct task_struct *group_leader = current->group_leader; 1203 struct pid *sid = task_pid(group_leader); 1204 pid_t session = pid_vnr(sid); 1205 int err = -EPERM; 1206 1207 write_lock_irq(&tasklist_lock); 1208 /* Fail if I am already a session leader */ 1209 if (group_leader->signal->leader) 1210 goto out; 1211 1212 /* Fail if a process group id already exists that equals the 1213 * proposed session id. 1214 */ 1215 if (pid_task(sid, PIDTYPE_PGID)) 1216 goto out; 1217 1218 group_leader->signal->leader = 1; 1219 set_special_pids(sid); 1220 1221 proc_clear_tty(group_leader); 1222 1223 err = session; 1224 out: 1225 write_unlock_irq(&tasklist_lock); 1226 if (err > 0) { 1227 proc_sid_connector(group_leader); 1228 sched_autogroup_create_attach(group_leader); 1229 } 1230 return err; 1231 } 1232 1233 SYSCALL_DEFINE0(setsid) 1234 { 1235 return ksys_setsid(); 1236 } 1237 1238 DECLARE_RWSEM(uts_sem); 1239 1240 #ifdef COMPAT_UTS_MACHINE 1241 #define override_architecture(name) \ 1242 (personality(current->personality) == PER_LINUX32 && \ 1243 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \ 1244 sizeof(COMPAT_UTS_MACHINE))) 1245 #else 1246 #define override_architecture(name) 0 1247 #endif 1248 1249 /* 1250 * Work around broken programs that cannot handle "Linux 3.0". 1251 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40 1252 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be 1253 * 2.6.60. 1254 */ 1255 static int override_release(char __user *release, size_t len) 1256 { 1257 int ret = 0; 1258 1259 if (current->personality & UNAME26) { 1260 const char *rest = UTS_RELEASE; 1261 char buf[65] = { 0 }; 1262 int ndots = 0; 1263 unsigned v; 1264 size_t copy; 1265 1266 while (*rest) { 1267 if (*rest == '.' && ++ndots >= 3) 1268 break; 1269 if (!isdigit(*rest) && *rest != '.') 1270 break; 1271 rest++; 1272 } 1273 v = LINUX_VERSION_PATCHLEVEL + 60; 1274 copy = clamp_t(size_t, len, 1, sizeof(buf)); 1275 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest); 1276 ret = copy_to_user(release, buf, copy + 1); 1277 } 1278 return ret; 1279 } 1280 1281 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name) 1282 { 1283 struct new_utsname tmp; 1284 1285 down_read(&uts_sem); 1286 memcpy(&tmp, utsname(), sizeof(tmp)); 1287 up_read(&uts_sem); 1288 if (copy_to_user(name, &tmp, sizeof(tmp))) 1289 return -EFAULT; 1290 1291 if (override_release(name->release, sizeof(name->release))) 1292 return -EFAULT; 1293 if (override_architecture(name)) 1294 return -EFAULT; 1295 return 0; 1296 } 1297 1298 #ifdef __ARCH_WANT_SYS_OLD_UNAME 1299 /* 1300 * Old cruft 1301 */ 1302 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name) 1303 { 1304 struct old_utsname tmp; 1305 1306 if (!name) 1307 return -EFAULT; 1308 1309 down_read(&uts_sem); 1310 memcpy(&tmp, utsname(), sizeof(tmp)); 1311 up_read(&uts_sem); 1312 if (copy_to_user(name, &tmp, sizeof(tmp))) 1313 return -EFAULT; 1314 1315 if (override_release(name->release, sizeof(name->release))) 1316 return -EFAULT; 1317 if (override_architecture(name)) 1318 return -EFAULT; 1319 return 0; 1320 } 1321 1322 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name) 1323 { 1324 struct oldold_utsname tmp; 1325 1326 if (!name) 1327 return -EFAULT; 1328 1329 memset(&tmp, 0, sizeof(tmp)); 1330 1331 down_read(&uts_sem); 1332 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN); 1333 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN); 1334 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN); 1335 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN); 1336 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN); 1337 up_read(&uts_sem); 1338 if (copy_to_user(name, &tmp, sizeof(tmp))) 1339 return -EFAULT; 1340 1341 if (override_architecture(name)) 1342 return -EFAULT; 1343 if (override_release(name->release, sizeof(name->release))) 1344 return -EFAULT; 1345 return 0; 1346 } 1347 #endif 1348 1349 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len) 1350 { 1351 int errno; 1352 char tmp[__NEW_UTS_LEN]; 1353 1354 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN)) 1355 return -EPERM; 1356 1357 if (len < 0 || len > __NEW_UTS_LEN) 1358 return -EINVAL; 1359 errno = -EFAULT; 1360 if (!copy_from_user(tmp, name, len)) { 1361 struct new_utsname *u; 1362 1363 down_write(&uts_sem); 1364 u = utsname(); 1365 memcpy(u->nodename, tmp, len); 1366 memset(u->nodename + len, 0, sizeof(u->nodename) - len); 1367 errno = 0; 1368 uts_proc_notify(UTS_PROC_HOSTNAME); 1369 up_write(&uts_sem); 1370 } 1371 return errno; 1372 } 1373 1374 #ifdef __ARCH_WANT_SYS_GETHOSTNAME 1375 1376 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len) 1377 { 1378 int i; 1379 struct new_utsname *u; 1380 char tmp[__NEW_UTS_LEN + 1]; 1381 1382 if (len < 0) 1383 return -EINVAL; 1384 down_read(&uts_sem); 1385 u = utsname(); 1386 i = 1 + strlen(u->nodename); 1387 if (i > len) 1388 i = len; 1389 memcpy(tmp, u->nodename, i); 1390 up_read(&uts_sem); 1391 if (copy_to_user(name, tmp, i)) 1392 return -EFAULT; 1393 return 0; 1394 } 1395 1396 #endif 1397 1398 /* 1399 * Only setdomainname; getdomainname can be implemented by calling 1400 * uname() 1401 */ 1402 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len) 1403 { 1404 int errno; 1405 char tmp[__NEW_UTS_LEN]; 1406 1407 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN)) 1408 return -EPERM; 1409 if (len < 0 || len > __NEW_UTS_LEN) 1410 return -EINVAL; 1411 1412 errno = -EFAULT; 1413 if (!copy_from_user(tmp, name, len)) { 1414 struct new_utsname *u; 1415 1416 down_write(&uts_sem); 1417 u = utsname(); 1418 memcpy(u->domainname, tmp, len); 1419 memset(u->domainname + len, 0, sizeof(u->domainname) - len); 1420 errno = 0; 1421 uts_proc_notify(UTS_PROC_DOMAINNAME); 1422 up_write(&uts_sem); 1423 } 1424 return errno; 1425 } 1426 1427 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim) 1428 { 1429 struct rlimit value; 1430 int ret; 1431 1432 ret = do_prlimit(current, resource, NULL, &value); 1433 if (!ret) 1434 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0; 1435 1436 return ret; 1437 } 1438 1439 #ifdef CONFIG_COMPAT 1440 1441 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource, 1442 struct compat_rlimit __user *, rlim) 1443 { 1444 struct rlimit r; 1445 struct compat_rlimit r32; 1446 1447 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit))) 1448 return -EFAULT; 1449 1450 if (r32.rlim_cur == COMPAT_RLIM_INFINITY) 1451 r.rlim_cur = RLIM_INFINITY; 1452 else 1453 r.rlim_cur = r32.rlim_cur; 1454 if (r32.rlim_max == COMPAT_RLIM_INFINITY) 1455 r.rlim_max = RLIM_INFINITY; 1456 else 1457 r.rlim_max = r32.rlim_max; 1458 return do_prlimit(current, resource, &r, NULL); 1459 } 1460 1461 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource, 1462 struct compat_rlimit __user *, rlim) 1463 { 1464 struct rlimit r; 1465 int ret; 1466 1467 ret = do_prlimit(current, resource, NULL, &r); 1468 if (!ret) { 1469 struct compat_rlimit r32; 1470 if (r.rlim_cur > COMPAT_RLIM_INFINITY) 1471 r32.rlim_cur = COMPAT_RLIM_INFINITY; 1472 else 1473 r32.rlim_cur = r.rlim_cur; 1474 if (r.rlim_max > COMPAT_RLIM_INFINITY) 1475 r32.rlim_max = COMPAT_RLIM_INFINITY; 1476 else 1477 r32.rlim_max = r.rlim_max; 1478 1479 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit))) 1480 return -EFAULT; 1481 } 1482 return ret; 1483 } 1484 1485 #endif 1486 1487 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT 1488 1489 /* 1490 * Back compatibility for getrlimit. Needed for some apps. 1491 */ 1492 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource, 1493 struct rlimit __user *, rlim) 1494 { 1495 struct rlimit x; 1496 if (resource >= RLIM_NLIMITS) 1497 return -EINVAL; 1498 1499 resource = array_index_nospec(resource, RLIM_NLIMITS); 1500 task_lock(current->group_leader); 1501 x = current->signal->rlim[resource]; 1502 task_unlock(current->group_leader); 1503 if (x.rlim_cur > 0x7FFFFFFF) 1504 x.rlim_cur = 0x7FFFFFFF; 1505 if (x.rlim_max > 0x7FFFFFFF) 1506 x.rlim_max = 0x7FFFFFFF; 1507 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0; 1508 } 1509 1510 #ifdef CONFIG_COMPAT 1511 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource, 1512 struct compat_rlimit __user *, rlim) 1513 { 1514 struct rlimit r; 1515 1516 if (resource >= RLIM_NLIMITS) 1517 return -EINVAL; 1518 1519 resource = array_index_nospec(resource, RLIM_NLIMITS); 1520 task_lock(current->group_leader); 1521 r = current->signal->rlim[resource]; 1522 task_unlock(current->group_leader); 1523 if (r.rlim_cur > 0x7FFFFFFF) 1524 r.rlim_cur = 0x7FFFFFFF; 1525 if (r.rlim_max > 0x7FFFFFFF) 1526 r.rlim_max = 0x7FFFFFFF; 1527 1528 if (put_user(r.rlim_cur, &rlim->rlim_cur) || 1529 put_user(r.rlim_max, &rlim->rlim_max)) 1530 return -EFAULT; 1531 return 0; 1532 } 1533 #endif 1534 1535 #endif 1536 1537 static inline bool rlim64_is_infinity(__u64 rlim64) 1538 { 1539 #if BITS_PER_LONG < 64 1540 return rlim64 >= ULONG_MAX; 1541 #else 1542 return rlim64 == RLIM64_INFINITY; 1543 #endif 1544 } 1545 1546 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64) 1547 { 1548 if (rlim->rlim_cur == RLIM_INFINITY) 1549 rlim64->rlim_cur = RLIM64_INFINITY; 1550 else 1551 rlim64->rlim_cur = rlim->rlim_cur; 1552 if (rlim->rlim_max == RLIM_INFINITY) 1553 rlim64->rlim_max = RLIM64_INFINITY; 1554 else 1555 rlim64->rlim_max = rlim->rlim_max; 1556 } 1557 1558 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim) 1559 { 1560 if (rlim64_is_infinity(rlim64->rlim_cur)) 1561 rlim->rlim_cur = RLIM_INFINITY; 1562 else 1563 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur; 1564 if (rlim64_is_infinity(rlim64->rlim_max)) 1565 rlim->rlim_max = RLIM_INFINITY; 1566 else 1567 rlim->rlim_max = (unsigned long)rlim64->rlim_max; 1568 } 1569 1570 /* make sure you are allowed to change @tsk limits before calling this */ 1571 int do_prlimit(struct task_struct *tsk, unsigned int resource, 1572 struct rlimit *new_rlim, struct rlimit *old_rlim) 1573 { 1574 struct rlimit *rlim; 1575 int retval = 0; 1576 1577 if (resource >= RLIM_NLIMITS) 1578 return -EINVAL; 1579 if (new_rlim) { 1580 if (new_rlim->rlim_cur > new_rlim->rlim_max) 1581 return -EINVAL; 1582 if (resource == RLIMIT_NOFILE && 1583 new_rlim->rlim_max > sysctl_nr_open) 1584 return -EPERM; 1585 } 1586 1587 /* protect tsk->signal and tsk->sighand from disappearing */ 1588 read_lock(&tasklist_lock); 1589 if (!tsk->sighand) { 1590 retval = -ESRCH; 1591 goto out; 1592 } 1593 1594 rlim = tsk->signal->rlim + resource; 1595 task_lock(tsk->group_leader); 1596 if (new_rlim) { 1597 /* Keep the capable check against init_user_ns until 1598 cgroups can contain all limits */ 1599 if (new_rlim->rlim_max > rlim->rlim_max && 1600 !capable(CAP_SYS_RESOURCE)) 1601 retval = -EPERM; 1602 if (!retval) 1603 retval = security_task_setrlimit(tsk, resource, new_rlim); 1604 } 1605 if (!retval) { 1606 if (old_rlim) 1607 *old_rlim = *rlim; 1608 if (new_rlim) 1609 *rlim = *new_rlim; 1610 } 1611 task_unlock(tsk->group_leader); 1612 1613 /* 1614 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not 1615 * infinite. In case of RLIM_INFINITY the posix CPU timer code 1616 * ignores the rlimit. 1617 */ 1618 if (!retval && new_rlim && resource == RLIMIT_CPU && 1619 new_rlim->rlim_cur != RLIM_INFINITY && 1620 IS_ENABLED(CONFIG_POSIX_TIMERS)) 1621 update_rlimit_cpu(tsk, new_rlim->rlim_cur); 1622 out: 1623 read_unlock(&tasklist_lock); 1624 return retval; 1625 } 1626 1627 /* rcu lock must be held */ 1628 static int check_prlimit_permission(struct task_struct *task, 1629 unsigned int flags) 1630 { 1631 const struct cred *cred = current_cred(), *tcred; 1632 bool id_match; 1633 1634 if (current == task) 1635 return 0; 1636 1637 tcred = __task_cred(task); 1638 id_match = (uid_eq(cred->uid, tcred->euid) && 1639 uid_eq(cred->uid, tcred->suid) && 1640 uid_eq(cred->uid, tcred->uid) && 1641 gid_eq(cred->gid, tcred->egid) && 1642 gid_eq(cred->gid, tcred->sgid) && 1643 gid_eq(cred->gid, tcred->gid)); 1644 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE)) 1645 return -EPERM; 1646 1647 return security_task_prlimit(cred, tcred, flags); 1648 } 1649 1650 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource, 1651 const struct rlimit64 __user *, new_rlim, 1652 struct rlimit64 __user *, old_rlim) 1653 { 1654 struct rlimit64 old64, new64; 1655 struct rlimit old, new; 1656 struct task_struct *tsk; 1657 unsigned int checkflags = 0; 1658 int ret; 1659 1660 if (old_rlim) 1661 checkflags |= LSM_PRLIMIT_READ; 1662 1663 if (new_rlim) { 1664 if (copy_from_user(&new64, new_rlim, sizeof(new64))) 1665 return -EFAULT; 1666 rlim64_to_rlim(&new64, &new); 1667 checkflags |= LSM_PRLIMIT_WRITE; 1668 } 1669 1670 rcu_read_lock(); 1671 tsk = pid ? find_task_by_vpid(pid) : current; 1672 if (!tsk) { 1673 rcu_read_unlock(); 1674 return -ESRCH; 1675 } 1676 ret = check_prlimit_permission(tsk, checkflags); 1677 if (ret) { 1678 rcu_read_unlock(); 1679 return ret; 1680 } 1681 get_task_struct(tsk); 1682 rcu_read_unlock(); 1683 1684 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL, 1685 old_rlim ? &old : NULL); 1686 1687 if (!ret && old_rlim) { 1688 rlim_to_rlim64(&old, &old64); 1689 if (copy_to_user(old_rlim, &old64, sizeof(old64))) 1690 ret = -EFAULT; 1691 } 1692 1693 put_task_struct(tsk); 1694 return ret; 1695 } 1696 1697 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim) 1698 { 1699 struct rlimit new_rlim; 1700 1701 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim))) 1702 return -EFAULT; 1703 return do_prlimit(current, resource, &new_rlim, NULL); 1704 } 1705 1706 /* 1707 * It would make sense to put struct rusage in the task_struct, 1708 * except that would make the task_struct be *really big*. After 1709 * task_struct gets moved into malloc'ed memory, it would 1710 * make sense to do this. It will make moving the rest of the information 1711 * a lot simpler! (Which we're not doing right now because we're not 1712 * measuring them yet). 1713 * 1714 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have 1715 * races with threads incrementing their own counters. But since word 1716 * reads are atomic, we either get new values or old values and we don't 1717 * care which for the sums. We always take the siglock to protect reading 1718 * the c* fields from p->signal from races with exit.c updating those 1719 * fields when reaping, so a sample either gets all the additions of a 1720 * given child after it's reaped, or none so this sample is before reaping. 1721 * 1722 * Locking: 1723 * We need to take the siglock for CHILDEREN, SELF and BOTH 1724 * for the cases current multithreaded, non-current single threaded 1725 * non-current multithreaded. Thread traversal is now safe with 1726 * the siglock held. 1727 * Strictly speaking, we donot need to take the siglock if we are current and 1728 * single threaded, as no one else can take our signal_struct away, no one 1729 * else can reap the children to update signal->c* counters, and no one else 1730 * can race with the signal-> fields. If we do not take any lock, the 1731 * signal-> fields could be read out of order while another thread was just 1732 * exiting. So we should place a read memory barrier when we avoid the lock. 1733 * On the writer side, write memory barrier is implied in __exit_signal 1734 * as __exit_signal releases the siglock spinlock after updating the signal-> 1735 * fields. But we don't do this yet to keep things simple. 1736 * 1737 */ 1738 1739 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r) 1740 { 1741 r->ru_nvcsw += t->nvcsw; 1742 r->ru_nivcsw += t->nivcsw; 1743 r->ru_minflt += t->min_flt; 1744 r->ru_majflt += t->maj_flt; 1745 r->ru_inblock += task_io_get_inblock(t); 1746 r->ru_oublock += task_io_get_oublock(t); 1747 } 1748 1749 void getrusage(struct task_struct *p, int who, struct rusage *r) 1750 { 1751 struct task_struct *t; 1752 unsigned long flags; 1753 u64 tgutime, tgstime, utime, stime; 1754 unsigned long maxrss = 0; 1755 1756 memset((char *)r, 0, sizeof (*r)); 1757 utime = stime = 0; 1758 1759 if (who == RUSAGE_THREAD) { 1760 task_cputime_adjusted(current, &utime, &stime); 1761 accumulate_thread_rusage(p, r); 1762 maxrss = p->signal->maxrss; 1763 goto out; 1764 } 1765 1766 if (!lock_task_sighand(p, &flags)) 1767 return; 1768 1769 switch (who) { 1770 case RUSAGE_BOTH: 1771 case RUSAGE_CHILDREN: 1772 utime = p->signal->cutime; 1773 stime = p->signal->cstime; 1774 r->ru_nvcsw = p->signal->cnvcsw; 1775 r->ru_nivcsw = p->signal->cnivcsw; 1776 r->ru_minflt = p->signal->cmin_flt; 1777 r->ru_majflt = p->signal->cmaj_flt; 1778 r->ru_inblock = p->signal->cinblock; 1779 r->ru_oublock = p->signal->coublock; 1780 maxrss = p->signal->cmaxrss; 1781 1782 if (who == RUSAGE_CHILDREN) 1783 break; 1784 fallthrough; 1785 1786 case RUSAGE_SELF: 1787 thread_group_cputime_adjusted(p, &tgutime, &tgstime); 1788 utime += tgutime; 1789 stime += tgstime; 1790 r->ru_nvcsw += p->signal->nvcsw; 1791 r->ru_nivcsw += p->signal->nivcsw; 1792 r->ru_minflt += p->signal->min_flt; 1793 r->ru_majflt += p->signal->maj_flt; 1794 r->ru_inblock += p->signal->inblock; 1795 r->ru_oublock += p->signal->oublock; 1796 if (maxrss < p->signal->maxrss) 1797 maxrss = p->signal->maxrss; 1798 t = p; 1799 do { 1800 accumulate_thread_rusage(t, r); 1801 } while_each_thread(p, t); 1802 break; 1803 1804 default: 1805 BUG(); 1806 } 1807 unlock_task_sighand(p, &flags); 1808 1809 out: 1810 r->ru_utime = ns_to_kernel_old_timeval(utime); 1811 r->ru_stime = ns_to_kernel_old_timeval(stime); 1812 1813 if (who != RUSAGE_CHILDREN) { 1814 struct mm_struct *mm = get_task_mm(p); 1815 1816 if (mm) { 1817 setmax_mm_hiwater_rss(&maxrss, mm); 1818 mmput(mm); 1819 } 1820 } 1821 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */ 1822 } 1823 1824 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru) 1825 { 1826 struct rusage r; 1827 1828 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN && 1829 who != RUSAGE_THREAD) 1830 return -EINVAL; 1831 1832 getrusage(current, who, &r); 1833 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0; 1834 } 1835 1836 #ifdef CONFIG_COMPAT 1837 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru) 1838 { 1839 struct rusage r; 1840 1841 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN && 1842 who != RUSAGE_THREAD) 1843 return -EINVAL; 1844 1845 getrusage(current, who, &r); 1846 return put_compat_rusage(&r, ru); 1847 } 1848 #endif 1849 1850 SYSCALL_DEFINE1(umask, int, mask) 1851 { 1852 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO); 1853 return mask; 1854 } 1855 1856 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd) 1857 { 1858 struct fd exe; 1859 struct inode *inode; 1860 int err; 1861 1862 exe = fdget(fd); 1863 if (!exe.file) 1864 return -EBADF; 1865 1866 inode = file_inode(exe.file); 1867 1868 /* 1869 * Because the original mm->exe_file points to executable file, make 1870 * sure that this one is executable as well, to avoid breaking an 1871 * overall picture. 1872 */ 1873 err = -EACCES; 1874 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path)) 1875 goto exit; 1876 1877 err = file_permission(exe.file, MAY_EXEC); 1878 if (err) 1879 goto exit; 1880 1881 err = replace_mm_exe_file(mm, exe.file); 1882 exit: 1883 fdput(exe); 1884 return err; 1885 } 1886 1887 /* 1888 * Check arithmetic relations of passed addresses. 1889 * 1890 * WARNING: we don't require any capability here so be very careful 1891 * in what is allowed for modification from userspace. 1892 */ 1893 static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map) 1894 { 1895 unsigned long mmap_max_addr = TASK_SIZE; 1896 int error = -EINVAL, i; 1897 1898 static const unsigned char offsets[] = { 1899 offsetof(struct prctl_mm_map, start_code), 1900 offsetof(struct prctl_mm_map, end_code), 1901 offsetof(struct prctl_mm_map, start_data), 1902 offsetof(struct prctl_mm_map, end_data), 1903 offsetof(struct prctl_mm_map, start_brk), 1904 offsetof(struct prctl_mm_map, brk), 1905 offsetof(struct prctl_mm_map, start_stack), 1906 offsetof(struct prctl_mm_map, arg_start), 1907 offsetof(struct prctl_mm_map, arg_end), 1908 offsetof(struct prctl_mm_map, env_start), 1909 offsetof(struct prctl_mm_map, env_end), 1910 }; 1911 1912 /* 1913 * Make sure the members are not somewhere outside 1914 * of allowed address space. 1915 */ 1916 for (i = 0; i < ARRAY_SIZE(offsets); i++) { 1917 u64 val = *(u64 *)((char *)prctl_map + offsets[i]); 1918 1919 if ((unsigned long)val >= mmap_max_addr || 1920 (unsigned long)val < mmap_min_addr) 1921 goto out; 1922 } 1923 1924 /* 1925 * Make sure the pairs are ordered. 1926 */ 1927 #define __prctl_check_order(__m1, __op, __m2) \ 1928 ((unsigned long)prctl_map->__m1 __op \ 1929 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL 1930 error = __prctl_check_order(start_code, <, end_code); 1931 error |= __prctl_check_order(start_data,<=, end_data); 1932 error |= __prctl_check_order(start_brk, <=, brk); 1933 error |= __prctl_check_order(arg_start, <=, arg_end); 1934 error |= __prctl_check_order(env_start, <=, env_end); 1935 if (error) 1936 goto out; 1937 #undef __prctl_check_order 1938 1939 error = -EINVAL; 1940 1941 /* 1942 * Neither we should allow to override limits if they set. 1943 */ 1944 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk, 1945 prctl_map->start_brk, prctl_map->end_data, 1946 prctl_map->start_data)) 1947 goto out; 1948 1949 error = 0; 1950 out: 1951 return error; 1952 } 1953 1954 #ifdef CONFIG_CHECKPOINT_RESTORE 1955 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size) 1956 { 1957 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, }; 1958 unsigned long user_auxv[AT_VECTOR_SIZE]; 1959 struct mm_struct *mm = current->mm; 1960 int error; 1961 1962 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv)); 1963 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256); 1964 1965 if (opt == PR_SET_MM_MAP_SIZE) 1966 return put_user((unsigned int)sizeof(prctl_map), 1967 (unsigned int __user *)addr); 1968 1969 if (data_size != sizeof(prctl_map)) 1970 return -EINVAL; 1971 1972 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map))) 1973 return -EFAULT; 1974 1975 error = validate_prctl_map_addr(&prctl_map); 1976 if (error) 1977 return error; 1978 1979 if (prctl_map.auxv_size) { 1980 /* 1981 * Someone is trying to cheat the auxv vector. 1982 */ 1983 if (!prctl_map.auxv || 1984 prctl_map.auxv_size > sizeof(mm->saved_auxv)) 1985 return -EINVAL; 1986 1987 memset(user_auxv, 0, sizeof(user_auxv)); 1988 if (copy_from_user(user_auxv, 1989 (const void __user *)prctl_map.auxv, 1990 prctl_map.auxv_size)) 1991 return -EFAULT; 1992 1993 /* Last entry must be AT_NULL as specification requires */ 1994 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL; 1995 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL; 1996 } 1997 1998 if (prctl_map.exe_fd != (u32)-1) { 1999 /* 2000 * Check if the current user is checkpoint/restore capable. 2001 * At the time of this writing, it checks for CAP_SYS_ADMIN 2002 * or CAP_CHECKPOINT_RESTORE. 2003 * Note that a user with access to ptrace can masquerade an 2004 * arbitrary program as any executable, even setuid ones. 2005 * This may have implications in the tomoyo subsystem. 2006 */ 2007 if (!checkpoint_restore_ns_capable(current_user_ns())) 2008 return -EPERM; 2009 2010 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd); 2011 if (error) 2012 return error; 2013 } 2014 2015 /* 2016 * arg_lock protects concurrent updates but we still need mmap_lock for 2017 * read to exclude races with sys_brk. 2018 */ 2019 mmap_read_lock(mm); 2020 2021 /* 2022 * We don't validate if these members are pointing to 2023 * real present VMAs because application may have correspond 2024 * VMAs already unmapped and kernel uses these members for statistics 2025 * output in procfs mostly, except 2026 * 2027 * - @start_brk/@brk which are used in do_brk_flags but kernel lookups 2028 * for VMAs when updating these members so anything wrong written 2029 * here cause kernel to swear at userspace program but won't lead 2030 * to any problem in kernel itself 2031 */ 2032 2033 spin_lock(&mm->arg_lock); 2034 mm->start_code = prctl_map.start_code; 2035 mm->end_code = prctl_map.end_code; 2036 mm->start_data = prctl_map.start_data; 2037 mm->end_data = prctl_map.end_data; 2038 mm->start_brk = prctl_map.start_brk; 2039 mm->brk = prctl_map.brk; 2040 mm->start_stack = prctl_map.start_stack; 2041 mm->arg_start = prctl_map.arg_start; 2042 mm->arg_end = prctl_map.arg_end; 2043 mm->env_start = prctl_map.env_start; 2044 mm->env_end = prctl_map.env_end; 2045 spin_unlock(&mm->arg_lock); 2046 2047 /* 2048 * Note this update of @saved_auxv is lockless thus 2049 * if someone reads this member in procfs while we're 2050 * updating -- it may get partly updated results. It's 2051 * known and acceptable trade off: we leave it as is to 2052 * not introduce additional locks here making the kernel 2053 * more complex. 2054 */ 2055 if (prctl_map.auxv_size) 2056 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv)); 2057 2058 mmap_read_unlock(mm); 2059 return 0; 2060 } 2061 #endif /* CONFIG_CHECKPOINT_RESTORE */ 2062 2063 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr, 2064 unsigned long len) 2065 { 2066 /* 2067 * This doesn't move the auxiliary vector itself since it's pinned to 2068 * mm_struct, but it permits filling the vector with new values. It's 2069 * up to the caller to provide sane values here, otherwise userspace 2070 * tools which use this vector might be unhappy. 2071 */ 2072 unsigned long user_auxv[AT_VECTOR_SIZE] = {}; 2073 2074 if (len > sizeof(user_auxv)) 2075 return -EINVAL; 2076 2077 if (copy_from_user(user_auxv, (const void __user *)addr, len)) 2078 return -EFAULT; 2079 2080 /* Make sure the last entry is always AT_NULL */ 2081 user_auxv[AT_VECTOR_SIZE - 2] = 0; 2082 user_auxv[AT_VECTOR_SIZE - 1] = 0; 2083 2084 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv)); 2085 2086 task_lock(current); 2087 memcpy(mm->saved_auxv, user_auxv, len); 2088 task_unlock(current); 2089 2090 return 0; 2091 } 2092 2093 static int prctl_set_mm(int opt, unsigned long addr, 2094 unsigned long arg4, unsigned long arg5) 2095 { 2096 struct mm_struct *mm = current->mm; 2097 struct prctl_mm_map prctl_map = { 2098 .auxv = NULL, 2099 .auxv_size = 0, 2100 .exe_fd = -1, 2101 }; 2102 struct vm_area_struct *vma; 2103 int error; 2104 2105 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV && 2106 opt != PR_SET_MM_MAP && 2107 opt != PR_SET_MM_MAP_SIZE))) 2108 return -EINVAL; 2109 2110 #ifdef CONFIG_CHECKPOINT_RESTORE 2111 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE) 2112 return prctl_set_mm_map(opt, (const void __user *)addr, arg4); 2113 #endif 2114 2115 if (!capable(CAP_SYS_RESOURCE)) 2116 return -EPERM; 2117 2118 if (opt == PR_SET_MM_EXE_FILE) 2119 return prctl_set_mm_exe_file(mm, (unsigned int)addr); 2120 2121 if (opt == PR_SET_MM_AUXV) 2122 return prctl_set_auxv(mm, addr, arg4); 2123 2124 if (addr >= TASK_SIZE || addr < mmap_min_addr) 2125 return -EINVAL; 2126 2127 error = -EINVAL; 2128 2129 /* 2130 * arg_lock protects concurrent updates of arg boundaries, we need 2131 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr 2132 * validation. 2133 */ 2134 mmap_read_lock(mm); 2135 vma = find_vma(mm, addr); 2136 2137 spin_lock(&mm->arg_lock); 2138 prctl_map.start_code = mm->start_code; 2139 prctl_map.end_code = mm->end_code; 2140 prctl_map.start_data = mm->start_data; 2141 prctl_map.end_data = mm->end_data; 2142 prctl_map.start_brk = mm->start_brk; 2143 prctl_map.brk = mm->brk; 2144 prctl_map.start_stack = mm->start_stack; 2145 prctl_map.arg_start = mm->arg_start; 2146 prctl_map.arg_end = mm->arg_end; 2147 prctl_map.env_start = mm->env_start; 2148 prctl_map.env_end = mm->env_end; 2149 2150 switch (opt) { 2151 case PR_SET_MM_START_CODE: 2152 prctl_map.start_code = addr; 2153 break; 2154 case PR_SET_MM_END_CODE: 2155 prctl_map.end_code = addr; 2156 break; 2157 case PR_SET_MM_START_DATA: 2158 prctl_map.start_data = addr; 2159 break; 2160 case PR_SET_MM_END_DATA: 2161 prctl_map.end_data = addr; 2162 break; 2163 case PR_SET_MM_START_STACK: 2164 prctl_map.start_stack = addr; 2165 break; 2166 case PR_SET_MM_START_BRK: 2167 prctl_map.start_brk = addr; 2168 break; 2169 case PR_SET_MM_BRK: 2170 prctl_map.brk = addr; 2171 break; 2172 case PR_SET_MM_ARG_START: 2173 prctl_map.arg_start = addr; 2174 break; 2175 case PR_SET_MM_ARG_END: 2176 prctl_map.arg_end = addr; 2177 break; 2178 case PR_SET_MM_ENV_START: 2179 prctl_map.env_start = addr; 2180 break; 2181 case PR_SET_MM_ENV_END: 2182 prctl_map.env_end = addr; 2183 break; 2184 default: 2185 goto out; 2186 } 2187 2188 error = validate_prctl_map_addr(&prctl_map); 2189 if (error) 2190 goto out; 2191 2192 switch (opt) { 2193 /* 2194 * If command line arguments and environment 2195 * are placed somewhere else on stack, we can 2196 * set them up here, ARG_START/END to setup 2197 * command line arguments and ENV_START/END 2198 * for environment. 2199 */ 2200 case PR_SET_MM_START_STACK: 2201 case PR_SET_MM_ARG_START: 2202 case PR_SET_MM_ARG_END: 2203 case PR_SET_MM_ENV_START: 2204 case PR_SET_MM_ENV_END: 2205 if (!vma) { 2206 error = -EFAULT; 2207 goto out; 2208 } 2209 } 2210 2211 mm->start_code = prctl_map.start_code; 2212 mm->end_code = prctl_map.end_code; 2213 mm->start_data = prctl_map.start_data; 2214 mm->end_data = prctl_map.end_data; 2215 mm->start_brk = prctl_map.start_brk; 2216 mm->brk = prctl_map.brk; 2217 mm->start_stack = prctl_map.start_stack; 2218 mm->arg_start = prctl_map.arg_start; 2219 mm->arg_end = prctl_map.arg_end; 2220 mm->env_start = prctl_map.env_start; 2221 mm->env_end = prctl_map.env_end; 2222 2223 error = 0; 2224 out: 2225 spin_unlock(&mm->arg_lock); 2226 mmap_read_unlock(mm); 2227 return error; 2228 } 2229 2230 #ifdef CONFIG_CHECKPOINT_RESTORE 2231 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr) 2232 { 2233 return put_user(me->clear_child_tid, tid_addr); 2234 } 2235 #else 2236 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr) 2237 { 2238 return -EINVAL; 2239 } 2240 #endif 2241 2242 static int propagate_has_child_subreaper(struct task_struct *p, void *data) 2243 { 2244 /* 2245 * If task has has_child_subreaper - all its descendants 2246 * already have these flag too and new descendants will 2247 * inherit it on fork, skip them. 2248 * 2249 * If we've found child_reaper - skip descendants in 2250 * it's subtree as they will never get out pidns. 2251 */ 2252 if (p->signal->has_child_subreaper || 2253 is_child_reaper(task_pid(p))) 2254 return 0; 2255 2256 p->signal->has_child_subreaper = 1; 2257 return 1; 2258 } 2259 2260 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which) 2261 { 2262 return -EINVAL; 2263 } 2264 2265 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which, 2266 unsigned long ctrl) 2267 { 2268 return -EINVAL; 2269 } 2270 2271 #define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE) 2272 2273 #ifdef CONFIG_ANON_VMA_NAME 2274 2275 #define ANON_VMA_NAME_MAX_LEN 80 2276 #define ANON_VMA_NAME_INVALID_CHARS "\\`$[]" 2277 2278 static inline bool is_valid_name_char(char ch) 2279 { 2280 /* printable ascii characters, excluding ANON_VMA_NAME_INVALID_CHARS */ 2281 return ch > 0x1f && ch < 0x7f && 2282 !strchr(ANON_VMA_NAME_INVALID_CHARS, ch); 2283 } 2284 2285 static int prctl_set_vma(unsigned long opt, unsigned long addr, 2286 unsigned long size, unsigned long arg) 2287 { 2288 struct mm_struct *mm = current->mm; 2289 const char __user *uname; 2290 struct anon_vma_name *anon_name = NULL; 2291 int error; 2292 2293 switch (opt) { 2294 case PR_SET_VMA_ANON_NAME: 2295 uname = (const char __user *)arg; 2296 if (uname) { 2297 char *name, *pch; 2298 2299 name = strndup_user(uname, ANON_VMA_NAME_MAX_LEN); 2300 if (IS_ERR(name)) 2301 return PTR_ERR(name); 2302 2303 for (pch = name; *pch != '\0'; pch++) { 2304 if (!is_valid_name_char(*pch)) { 2305 kfree(name); 2306 return -EINVAL; 2307 } 2308 } 2309 /* anon_vma has its own copy */ 2310 anon_name = anon_vma_name_alloc(name); 2311 kfree(name); 2312 if (!anon_name) 2313 return -ENOMEM; 2314 2315 } 2316 2317 mmap_write_lock(mm); 2318 error = madvise_set_anon_name(mm, addr, size, anon_name); 2319 mmap_write_unlock(mm); 2320 anon_vma_name_put(anon_name); 2321 break; 2322 default: 2323 error = -EINVAL; 2324 } 2325 2326 return error; 2327 } 2328 2329 #else /* CONFIG_ANON_VMA_NAME */ 2330 static int prctl_set_vma(unsigned long opt, unsigned long start, 2331 unsigned long size, unsigned long arg) 2332 { 2333 return -EINVAL; 2334 } 2335 #endif /* CONFIG_ANON_VMA_NAME */ 2336 2337 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3, 2338 unsigned long, arg4, unsigned long, arg5) 2339 { 2340 struct task_struct *me = current; 2341 unsigned char comm[sizeof(me->comm)]; 2342 long error; 2343 2344 error = security_task_prctl(option, arg2, arg3, arg4, arg5); 2345 if (error != -ENOSYS) 2346 return error; 2347 2348 error = 0; 2349 switch (option) { 2350 case PR_SET_PDEATHSIG: 2351 if (!valid_signal(arg2)) { 2352 error = -EINVAL; 2353 break; 2354 } 2355 me->pdeath_signal = arg2; 2356 break; 2357 case PR_GET_PDEATHSIG: 2358 error = put_user(me->pdeath_signal, (int __user *)arg2); 2359 break; 2360 case PR_GET_DUMPABLE: 2361 error = get_dumpable(me->mm); 2362 break; 2363 case PR_SET_DUMPABLE: 2364 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) { 2365 error = -EINVAL; 2366 break; 2367 } 2368 set_dumpable(me->mm, arg2); 2369 break; 2370 2371 case PR_SET_UNALIGN: 2372 error = SET_UNALIGN_CTL(me, arg2); 2373 break; 2374 case PR_GET_UNALIGN: 2375 error = GET_UNALIGN_CTL(me, arg2); 2376 break; 2377 case PR_SET_FPEMU: 2378 error = SET_FPEMU_CTL(me, arg2); 2379 break; 2380 case PR_GET_FPEMU: 2381 error = GET_FPEMU_CTL(me, arg2); 2382 break; 2383 case PR_SET_FPEXC: 2384 error = SET_FPEXC_CTL(me, arg2); 2385 break; 2386 case PR_GET_FPEXC: 2387 error = GET_FPEXC_CTL(me, arg2); 2388 break; 2389 case PR_GET_TIMING: 2390 error = PR_TIMING_STATISTICAL; 2391 break; 2392 case PR_SET_TIMING: 2393 if (arg2 != PR_TIMING_STATISTICAL) 2394 error = -EINVAL; 2395 break; 2396 case PR_SET_NAME: 2397 comm[sizeof(me->comm) - 1] = 0; 2398 if (strncpy_from_user(comm, (char __user *)arg2, 2399 sizeof(me->comm) - 1) < 0) 2400 return -EFAULT; 2401 set_task_comm(me, comm); 2402 proc_comm_connector(me); 2403 break; 2404 case PR_GET_NAME: 2405 get_task_comm(comm, me); 2406 if (copy_to_user((char __user *)arg2, comm, sizeof(comm))) 2407 return -EFAULT; 2408 break; 2409 case PR_GET_ENDIAN: 2410 error = GET_ENDIAN(me, arg2); 2411 break; 2412 case PR_SET_ENDIAN: 2413 error = SET_ENDIAN(me, arg2); 2414 break; 2415 case PR_GET_SECCOMP: 2416 error = prctl_get_seccomp(); 2417 break; 2418 case PR_SET_SECCOMP: 2419 error = prctl_set_seccomp(arg2, (char __user *)arg3); 2420 break; 2421 case PR_GET_TSC: 2422 error = GET_TSC_CTL(arg2); 2423 break; 2424 case PR_SET_TSC: 2425 error = SET_TSC_CTL(arg2); 2426 break; 2427 case PR_TASK_PERF_EVENTS_DISABLE: 2428 error = perf_event_task_disable(); 2429 break; 2430 case PR_TASK_PERF_EVENTS_ENABLE: 2431 error = perf_event_task_enable(); 2432 break; 2433 case PR_GET_TIMERSLACK: 2434 if (current->timer_slack_ns > ULONG_MAX) 2435 error = ULONG_MAX; 2436 else 2437 error = current->timer_slack_ns; 2438 break; 2439 case PR_SET_TIMERSLACK: 2440 if (arg2 <= 0) 2441 current->timer_slack_ns = 2442 current->default_timer_slack_ns; 2443 else 2444 current->timer_slack_ns = arg2; 2445 break; 2446 case PR_MCE_KILL: 2447 if (arg4 | arg5) 2448 return -EINVAL; 2449 switch (arg2) { 2450 case PR_MCE_KILL_CLEAR: 2451 if (arg3 != 0) 2452 return -EINVAL; 2453 current->flags &= ~PF_MCE_PROCESS; 2454 break; 2455 case PR_MCE_KILL_SET: 2456 current->flags |= PF_MCE_PROCESS; 2457 if (arg3 == PR_MCE_KILL_EARLY) 2458 current->flags |= PF_MCE_EARLY; 2459 else if (arg3 == PR_MCE_KILL_LATE) 2460 current->flags &= ~PF_MCE_EARLY; 2461 else if (arg3 == PR_MCE_KILL_DEFAULT) 2462 current->flags &= 2463 ~(PF_MCE_EARLY|PF_MCE_PROCESS); 2464 else 2465 return -EINVAL; 2466 break; 2467 default: 2468 return -EINVAL; 2469 } 2470 break; 2471 case PR_MCE_KILL_GET: 2472 if (arg2 | arg3 | arg4 | arg5) 2473 return -EINVAL; 2474 if (current->flags & PF_MCE_PROCESS) 2475 error = (current->flags & PF_MCE_EARLY) ? 2476 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE; 2477 else 2478 error = PR_MCE_KILL_DEFAULT; 2479 break; 2480 case PR_SET_MM: 2481 error = prctl_set_mm(arg2, arg3, arg4, arg5); 2482 break; 2483 case PR_GET_TID_ADDRESS: 2484 error = prctl_get_tid_address(me, (int __user * __user *)arg2); 2485 break; 2486 case PR_SET_CHILD_SUBREAPER: 2487 me->signal->is_child_subreaper = !!arg2; 2488 if (!arg2) 2489 break; 2490 2491 walk_process_tree(me, propagate_has_child_subreaper, NULL); 2492 break; 2493 case PR_GET_CHILD_SUBREAPER: 2494 error = put_user(me->signal->is_child_subreaper, 2495 (int __user *)arg2); 2496 break; 2497 case PR_SET_NO_NEW_PRIVS: 2498 if (arg2 != 1 || arg3 || arg4 || arg5) 2499 return -EINVAL; 2500 2501 task_set_no_new_privs(current); 2502 break; 2503 case PR_GET_NO_NEW_PRIVS: 2504 if (arg2 || arg3 || arg4 || arg5) 2505 return -EINVAL; 2506 return task_no_new_privs(current) ? 1 : 0; 2507 case PR_GET_THP_DISABLE: 2508 if (arg2 || arg3 || arg4 || arg5) 2509 return -EINVAL; 2510 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags); 2511 break; 2512 case PR_SET_THP_DISABLE: 2513 if (arg3 || arg4 || arg5) 2514 return -EINVAL; 2515 if (mmap_write_lock_killable(me->mm)) 2516 return -EINTR; 2517 if (arg2) 2518 set_bit(MMF_DISABLE_THP, &me->mm->flags); 2519 else 2520 clear_bit(MMF_DISABLE_THP, &me->mm->flags); 2521 mmap_write_unlock(me->mm); 2522 break; 2523 case PR_MPX_ENABLE_MANAGEMENT: 2524 case PR_MPX_DISABLE_MANAGEMENT: 2525 /* No longer implemented: */ 2526 return -EINVAL; 2527 case PR_SET_FP_MODE: 2528 error = SET_FP_MODE(me, arg2); 2529 break; 2530 case PR_GET_FP_MODE: 2531 error = GET_FP_MODE(me); 2532 break; 2533 case PR_SVE_SET_VL: 2534 error = SVE_SET_VL(arg2); 2535 break; 2536 case PR_SVE_GET_VL: 2537 error = SVE_GET_VL(); 2538 break; 2539 case PR_GET_SPECULATION_CTRL: 2540 if (arg3 || arg4 || arg5) 2541 return -EINVAL; 2542 error = arch_prctl_spec_ctrl_get(me, arg2); 2543 break; 2544 case PR_SET_SPECULATION_CTRL: 2545 if (arg4 || arg5) 2546 return -EINVAL; 2547 error = arch_prctl_spec_ctrl_set(me, arg2, arg3); 2548 break; 2549 case PR_PAC_RESET_KEYS: 2550 if (arg3 || arg4 || arg5) 2551 return -EINVAL; 2552 error = PAC_RESET_KEYS(me, arg2); 2553 break; 2554 case PR_PAC_SET_ENABLED_KEYS: 2555 if (arg4 || arg5) 2556 return -EINVAL; 2557 error = PAC_SET_ENABLED_KEYS(me, arg2, arg3); 2558 break; 2559 case PR_PAC_GET_ENABLED_KEYS: 2560 if (arg2 || arg3 || arg4 || arg5) 2561 return -EINVAL; 2562 error = PAC_GET_ENABLED_KEYS(me); 2563 break; 2564 case PR_SET_TAGGED_ADDR_CTRL: 2565 if (arg3 || arg4 || arg5) 2566 return -EINVAL; 2567 error = SET_TAGGED_ADDR_CTRL(arg2); 2568 break; 2569 case PR_GET_TAGGED_ADDR_CTRL: 2570 if (arg2 || arg3 || arg4 || arg5) 2571 return -EINVAL; 2572 error = GET_TAGGED_ADDR_CTRL(); 2573 break; 2574 case PR_SET_IO_FLUSHER: 2575 if (!capable(CAP_SYS_RESOURCE)) 2576 return -EPERM; 2577 2578 if (arg3 || arg4 || arg5) 2579 return -EINVAL; 2580 2581 if (arg2 == 1) 2582 current->flags |= PR_IO_FLUSHER; 2583 else if (!arg2) 2584 current->flags &= ~PR_IO_FLUSHER; 2585 else 2586 return -EINVAL; 2587 break; 2588 case PR_GET_IO_FLUSHER: 2589 if (!capable(CAP_SYS_RESOURCE)) 2590 return -EPERM; 2591 2592 if (arg2 || arg3 || arg4 || arg5) 2593 return -EINVAL; 2594 2595 error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER; 2596 break; 2597 case PR_SET_SYSCALL_USER_DISPATCH: 2598 error = set_syscall_user_dispatch(arg2, arg3, arg4, 2599 (char __user *) arg5); 2600 break; 2601 #ifdef CONFIG_SCHED_CORE 2602 case PR_SCHED_CORE: 2603 error = sched_core_share_pid(arg2, arg3, arg4, arg5); 2604 break; 2605 #endif 2606 case PR_SET_VMA: 2607 error = prctl_set_vma(arg2, arg3, arg4, arg5); 2608 break; 2609 default: 2610 error = -EINVAL; 2611 break; 2612 } 2613 return error; 2614 } 2615 2616 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep, 2617 struct getcpu_cache __user *, unused) 2618 { 2619 int err = 0; 2620 int cpu = raw_smp_processor_id(); 2621 2622 if (cpup) 2623 err |= put_user(cpu, cpup); 2624 if (nodep) 2625 err |= put_user(cpu_to_node(cpu), nodep); 2626 return err ? -EFAULT : 0; 2627 } 2628 2629 /** 2630 * do_sysinfo - fill in sysinfo struct 2631 * @info: pointer to buffer to fill 2632 */ 2633 static int do_sysinfo(struct sysinfo *info) 2634 { 2635 unsigned long mem_total, sav_total; 2636 unsigned int mem_unit, bitcount; 2637 struct timespec64 tp; 2638 2639 memset(info, 0, sizeof(struct sysinfo)); 2640 2641 ktime_get_boottime_ts64(&tp); 2642 timens_add_boottime(&tp); 2643 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0); 2644 2645 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT); 2646 2647 info->procs = nr_threads; 2648 2649 si_meminfo(info); 2650 si_swapinfo(info); 2651 2652 /* 2653 * If the sum of all the available memory (i.e. ram + swap) 2654 * is less than can be stored in a 32 bit unsigned long then 2655 * we can be binary compatible with 2.2.x kernels. If not, 2656 * well, in that case 2.2.x was broken anyways... 2657 * 2658 * -Erik Andersen <andersee@debian.org> 2659 */ 2660 2661 mem_total = info->totalram + info->totalswap; 2662 if (mem_total < info->totalram || mem_total < info->totalswap) 2663 goto out; 2664 bitcount = 0; 2665 mem_unit = info->mem_unit; 2666 while (mem_unit > 1) { 2667 bitcount++; 2668 mem_unit >>= 1; 2669 sav_total = mem_total; 2670 mem_total <<= 1; 2671 if (mem_total < sav_total) 2672 goto out; 2673 } 2674 2675 /* 2676 * If mem_total did not overflow, multiply all memory values by 2677 * info->mem_unit and set it to 1. This leaves things compatible 2678 * with 2.2.x, and also retains compatibility with earlier 2.4.x 2679 * kernels... 2680 */ 2681 2682 info->mem_unit = 1; 2683 info->totalram <<= bitcount; 2684 info->freeram <<= bitcount; 2685 info->sharedram <<= bitcount; 2686 info->bufferram <<= bitcount; 2687 info->totalswap <<= bitcount; 2688 info->freeswap <<= bitcount; 2689 info->totalhigh <<= bitcount; 2690 info->freehigh <<= bitcount; 2691 2692 out: 2693 return 0; 2694 } 2695 2696 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info) 2697 { 2698 struct sysinfo val; 2699 2700 do_sysinfo(&val); 2701 2702 if (copy_to_user(info, &val, sizeof(struct sysinfo))) 2703 return -EFAULT; 2704 2705 return 0; 2706 } 2707 2708 #ifdef CONFIG_COMPAT 2709 struct compat_sysinfo { 2710 s32 uptime; 2711 u32 loads[3]; 2712 u32 totalram; 2713 u32 freeram; 2714 u32 sharedram; 2715 u32 bufferram; 2716 u32 totalswap; 2717 u32 freeswap; 2718 u16 procs; 2719 u16 pad; 2720 u32 totalhigh; 2721 u32 freehigh; 2722 u32 mem_unit; 2723 char _f[20-2*sizeof(u32)-sizeof(int)]; 2724 }; 2725 2726 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info) 2727 { 2728 struct sysinfo s; 2729 struct compat_sysinfo s_32; 2730 2731 do_sysinfo(&s); 2732 2733 /* Check to see if any memory value is too large for 32-bit and scale 2734 * down if needed 2735 */ 2736 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) { 2737 int bitcount = 0; 2738 2739 while (s.mem_unit < PAGE_SIZE) { 2740 s.mem_unit <<= 1; 2741 bitcount++; 2742 } 2743 2744 s.totalram >>= bitcount; 2745 s.freeram >>= bitcount; 2746 s.sharedram >>= bitcount; 2747 s.bufferram >>= bitcount; 2748 s.totalswap >>= bitcount; 2749 s.freeswap >>= bitcount; 2750 s.totalhigh >>= bitcount; 2751 s.freehigh >>= bitcount; 2752 } 2753 2754 memset(&s_32, 0, sizeof(s_32)); 2755 s_32.uptime = s.uptime; 2756 s_32.loads[0] = s.loads[0]; 2757 s_32.loads[1] = s.loads[1]; 2758 s_32.loads[2] = s.loads[2]; 2759 s_32.totalram = s.totalram; 2760 s_32.freeram = s.freeram; 2761 s_32.sharedram = s.sharedram; 2762 s_32.bufferram = s.bufferram; 2763 s_32.totalswap = s.totalswap; 2764 s_32.freeswap = s.freeswap; 2765 s_32.procs = s.procs; 2766 s_32.totalhigh = s.totalhigh; 2767 s_32.freehigh = s.freehigh; 2768 s_32.mem_unit = s.mem_unit; 2769 if (copy_to_user(info, &s_32, sizeof(s_32))) 2770 return -EFAULT; 2771 return 0; 2772 } 2773 #endif /* CONFIG_COMPAT */ 2774