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