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