]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - kernel/sys.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
x86/bugs: Expose x86_spec_ctrl_base directly
[mirror_ubuntu-artful-kernel.git] / kernel / sys.c
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/sched/autogroup.h>
53 #include <linux/sched/loadavg.h>
54 #include <linux/sched/stat.h>
55 #include <linux/sched/mm.h>
56 #include <linux/sched/coredump.h>
57 #include <linux/sched/task.h>
58 #include <linux/sched/cputime.h>
59 #include <linux/rcupdate.h>
60 #include <linux/uidgid.h>
61 #include <linux/cred.h>
62
63 #include <linux/nospec.h>
64
65 #include <linux/kmsg_dump.h>
66 /* Move somewhere else to avoid recompiling? */
67 #include <generated/utsrelease.h>
68
69 #include <linux/uaccess.h>
70 #include <asm/io.h>
71 #include <asm/unistd.h>
72
73 #ifndef SET_UNALIGN_CTL
74 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
75 #endif
76 #ifndef GET_UNALIGN_CTL
77 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
78 #endif
79 #ifndef SET_FPEMU_CTL
80 # define SET_FPEMU_CTL(a, b) (-EINVAL)
81 #endif
82 #ifndef GET_FPEMU_CTL
83 # define GET_FPEMU_CTL(a, b) (-EINVAL)
84 #endif
85 #ifndef SET_FPEXC_CTL
86 # define SET_FPEXC_CTL(a, b) (-EINVAL)
87 #endif
88 #ifndef GET_FPEXC_CTL
89 # define GET_FPEXC_CTL(a, b) (-EINVAL)
90 #endif
91 #ifndef GET_ENDIAN
92 # define GET_ENDIAN(a, b) (-EINVAL)
93 #endif
94 #ifndef SET_ENDIAN
95 # define SET_ENDIAN(a, b) (-EINVAL)
96 #endif
97 #ifndef GET_TSC_CTL
98 # define GET_TSC_CTL(a) (-EINVAL)
99 #endif
100 #ifndef SET_TSC_CTL
101 # define SET_TSC_CTL(a) (-EINVAL)
102 #endif
103 #ifndef MPX_ENABLE_MANAGEMENT
104 # define MPX_ENABLE_MANAGEMENT() (-EINVAL)
105 #endif
106 #ifndef MPX_DISABLE_MANAGEMENT
107 # define MPX_DISABLE_MANAGEMENT() (-EINVAL)
108 #endif
109 #ifndef GET_FP_MODE
110 # define GET_FP_MODE(a) (-EINVAL)
111 #endif
112 #ifndef SET_FP_MODE
113 # define SET_FP_MODE(a,b) (-EINVAL)
114 #endif
115
116 /*
117 * this is where the system-wide overflow UID and GID are defined, for
118 * architectures that now have 32-bit UID/GID but didn't in the past
119 */
120
121 int overflowuid = DEFAULT_OVERFLOWUID;
122 int overflowgid = DEFAULT_OVERFLOWGID;
123
124 EXPORT_SYMBOL(overflowuid);
125 EXPORT_SYMBOL(overflowgid);
126
127 /*
128 * the same as above, but for filesystems which can only store a 16-bit
129 * UID and GID. as such, this is needed on all architectures
130 */
131
132 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
133 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
134
135 EXPORT_SYMBOL(fs_overflowuid);
136 EXPORT_SYMBOL(fs_overflowgid);
137
138 /*
139 * Returns true if current's euid is same as p's uid or euid,
140 * or has CAP_SYS_NICE to p's user_ns.
141 *
142 * Called with rcu_read_lock, creds are safe
143 */
144 static bool set_one_prio_perm(struct task_struct *p)
145 {
146 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
147
148 if (uid_eq(pcred->uid, cred->euid) ||
149 uid_eq(pcred->euid, cred->euid))
150 return true;
151 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
152 return true;
153 return false;
154 }
155
156 /*
157 * set the priority of a task
158 * - the caller must hold the RCU read lock
159 */
160 static int set_one_prio(struct task_struct *p, int niceval, int error)
161 {
162 int no_nice;
163
164 if (!set_one_prio_perm(p)) {
165 error = -EPERM;
166 goto out;
167 }
168 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
169 error = -EACCES;
170 goto out;
171 }
172 no_nice = security_task_setnice(p, niceval);
173 if (no_nice) {
174 error = no_nice;
175 goto out;
176 }
177 if (error == -ESRCH)
178 error = 0;
179 set_user_nice(p, niceval);
180 out:
181 return error;
182 }
183
184 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
185 {
186 struct task_struct *g, *p;
187 struct user_struct *user;
188 const struct cred *cred = current_cred();
189 int error = -EINVAL;
190 struct pid *pgrp;
191 kuid_t uid;
192
193 if (which > PRIO_USER || which < PRIO_PROCESS)
194 goto out;
195
196 /* normalize: avoid signed division (rounding problems) */
197 error = -ESRCH;
198 if (niceval < MIN_NICE)
199 niceval = MIN_NICE;
200 if (niceval > MAX_NICE)
201 niceval = MAX_NICE;
202
203 rcu_read_lock();
204 read_lock(&tasklist_lock);
205 switch (which) {
206 case PRIO_PROCESS:
207 if (who)
208 p = find_task_by_vpid(who);
209 else
210 p = current;
211 if (p)
212 error = set_one_prio(p, niceval, error);
213 break;
214 case PRIO_PGRP:
215 if (who)
216 pgrp = find_vpid(who);
217 else
218 pgrp = task_pgrp(current);
219 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
220 error = set_one_prio(p, niceval, error);
221 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
222 break;
223 case PRIO_USER:
224 uid = make_kuid(cred->user_ns, who);
225 user = cred->user;
226 if (!who)
227 uid = cred->uid;
228 else if (!uid_eq(uid, cred->uid)) {
229 user = find_user(uid);
230 if (!user)
231 goto out_unlock; /* No processes for this user */
232 }
233 do_each_thread(g, p) {
234 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
235 error = set_one_prio(p, niceval, error);
236 } while_each_thread(g, p);
237 if (!uid_eq(uid, cred->uid))
238 free_uid(user); /* For find_user() */
239 break;
240 }
241 out_unlock:
242 read_unlock(&tasklist_lock);
243 rcu_read_unlock();
244 out:
245 return error;
246 }
247
248 /*
249 * Ugh. To avoid negative return values, "getpriority()" will
250 * not return the normal nice-value, but a negated value that
251 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
252 * to stay compatible.
253 */
254 SYSCALL_DEFINE2(getpriority, int, which, int, who)
255 {
256 struct task_struct *g, *p;
257 struct user_struct *user;
258 const struct cred *cred = current_cred();
259 long niceval, retval = -ESRCH;
260 struct pid *pgrp;
261 kuid_t uid;
262
263 if (which > PRIO_USER || which < PRIO_PROCESS)
264 return -EINVAL;
265
266 rcu_read_lock();
267 read_lock(&tasklist_lock);
268 switch (which) {
269 case PRIO_PROCESS:
270 if (who)
271 p = find_task_by_vpid(who);
272 else
273 p = current;
274 if (p) {
275 niceval = nice_to_rlimit(task_nice(p));
276 if (niceval > retval)
277 retval = niceval;
278 }
279 break;
280 case PRIO_PGRP:
281 if (who)
282 pgrp = find_vpid(who);
283 else
284 pgrp = task_pgrp(current);
285 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
286 niceval = nice_to_rlimit(task_nice(p));
287 if (niceval > retval)
288 retval = niceval;
289 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
290 break;
291 case PRIO_USER:
292 uid = make_kuid(cred->user_ns, who);
293 user = cred->user;
294 if (!who)
295 uid = cred->uid;
296 else if (!uid_eq(uid, cred->uid)) {
297 user = find_user(uid);
298 if (!user)
299 goto out_unlock; /* No processes for this user */
300 }
301 do_each_thread(g, p) {
302 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
303 niceval = nice_to_rlimit(task_nice(p));
304 if (niceval > retval)
305 retval = niceval;
306 }
307 } while_each_thread(g, p);
308 if (!uid_eq(uid, cred->uid))
309 free_uid(user); /* for find_user() */
310 break;
311 }
312 out_unlock:
313 read_unlock(&tasklist_lock);
314 rcu_read_unlock();
315
316 return retval;
317 }
318
319 /*
320 * Unprivileged users may change the real gid to the effective gid
321 * or vice versa. (BSD-style)
322 *
323 * If you set the real gid at all, or set the effective gid to a value not
324 * equal to the real gid, then the saved gid is set to the new effective gid.
325 *
326 * This makes it possible for a setgid program to completely drop its
327 * privileges, which is often a useful assertion to make when you are doing
328 * a security audit over a program.
329 *
330 * The general idea is that a program which uses just setregid() will be
331 * 100% compatible with BSD. A program which uses just setgid() will be
332 * 100% compatible with POSIX with saved IDs.
333 *
334 * SMP: There are not races, the GIDs are checked only by filesystem
335 * operations (as far as semantic preservation is concerned).
336 */
337 #ifdef CONFIG_MULTIUSER
338 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
339 {
340 struct user_namespace *ns = current_user_ns();
341 const struct cred *old;
342 struct cred *new;
343 int retval;
344 kgid_t krgid, kegid;
345
346 krgid = make_kgid(ns, rgid);
347 kegid = make_kgid(ns, egid);
348
349 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
350 return -EINVAL;
351 if ((egid != (gid_t) -1) && !gid_valid(kegid))
352 return -EINVAL;
353
354 new = prepare_creds();
355 if (!new)
356 return -ENOMEM;
357 old = current_cred();
358
359 retval = -EPERM;
360 if (rgid != (gid_t) -1) {
361 if (gid_eq(old->gid, krgid) ||
362 gid_eq(old->egid, krgid) ||
363 ns_capable(old->user_ns, CAP_SETGID))
364 new->gid = krgid;
365 else
366 goto error;
367 }
368 if (egid != (gid_t) -1) {
369 if (gid_eq(old->gid, kegid) ||
370 gid_eq(old->egid, kegid) ||
371 gid_eq(old->sgid, kegid) ||
372 ns_capable(old->user_ns, CAP_SETGID))
373 new->egid = kegid;
374 else
375 goto error;
376 }
377
378 if (rgid != (gid_t) -1 ||
379 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
380 new->sgid = new->egid;
381 new->fsgid = new->egid;
382
383 return commit_creds(new);
384
385 error:
386 abort_creds(new);
387 return retval;
388 }
389
390 /*
391 * setgid() is implemented like SysV w/ SAVED_IDS
392 *
393 * SMP: Same implicit races as above.
394 */
395 SYSCALL_DEFINE1(setgid, gid_t, gid)
396 {
397 struct user_namespace *ns = current_user_ns();
398 const struct cred *old;
399 struct cred *new;
400 int retval;
401 kgid_t kgid;
402
403 kgid = make_kgid(ns, gid);
404 if (!gid_valid(kgid))
405 return -EINVAL;
406
407 new = prepare_creds();
408 if (!new)
409 return -ENOMEM;
410 old = current_cred();
411
412 retval = -EPERM;
413 if (ns_capable(old->user_ns, CAP_SETGID))
414 new->gid = new->egid = new->sgid = new->fsgid = kgid;
415 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
416 new->egid = new->fsgid = kgid;
417 else
418 goto error;
419
420 return commit_creds(new);
421
422 error:
423 abort_creds(new);
424 return retval;
425 }
426
427 /*
428 * change the user struct in a credentials set to match the new UID
429 */
430 static int set_user(struct cred *new)
431 {
432 struct user_struct *new_user;
433
434 new_user = alloc_uid(new->uid);
435 if (!new_user)
436 return -EAGAIN;
437
438 /*
439 * We don't fail in case of NPROC limit excess here because too many
440 * poorly written programs don't check set*uid() return code, assuming
441 * it never fails if called by root. We may still enforce NPROC limit
442 * for programs doing set*uid()+execve() by harmlessly deferring the
443 * failure to the execve() stage.
444 */
445 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
446 new_user != INIT_USER)
447 current->flags |= PF_NPROC_EXCEEDED;
448 else
449 current->flags &= ~PF_NPROC_EXCEEDED;
450
451 free_uid(new->user);
452 new->user = new_user;
453 return 0;
454 }
455
456 /*
457 * Unprivileged users may change the real uid to the effective uid
458 * or vice versa. (BSD-style)
459 *
460 * If you set the real uid at all, or set the effective uid to a value not
461 * equal to the real uid, then the saved uid is set to the new effective uid.
462 *
463 * This makes it possible for a setuid program to completely drop its
464 * privileges, which is often a useful assertion to make when you are doing
465 * a security audit over a program.
466 *
467 * The general idea is that a program which uses just setreuid() will be
468 * 100% compatible with BSD. A program which uses just setuid() will be
469 * 100% compatible with POSIX with saved IDs.
470 */
471 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
472 {
473 struct user_namespace *ns = current_user_ns();
474 const struct cred *old;
475 struct cred *new;
476 int retval;
477 kuid_t kruid, keuid;
478
479 kruid = make_kuid(ns, ruid);
480 keuid = make_kuid(ns, euid);
481
482 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
483 return -EINVAL;
484 if ((euid != (uid_t) -1) && !uid_valid(keuid))
485 return -EINVAL;
486
487 new = prepare_creds();
488 if (!new)
489 return -ENOMEM;
490 old = current_cred();
491
492 retval = -EPERM;
493 if (ruid != (uid_t) -1) {
494 new->uid = kruid;
495 if (!uid_eq(old->uid, kruid) &&
496 !uid_eq(old->euid, kruid) &&
497 !ns_capable(old->user_ns, CAP_SETUID))
498 goto error;
499 }
500
501 if (euid != (uid_t) -1) {
502 new->euid = keuid;
503 if (!uid_eq(old->uid, keuid) &&
504 !uid_eq(old->euid, keuid) &&
505 !uid_eq(old->suid, keuid) &&
506 !ns_capable(old->user_ns, CAP_SETUID))
507 goto error;
508 }
509
510 if (!uid_eq(new->uid, old->uid)) {
511 retval = set_user(new);
512 if (retval < 0)
513 goto error;
514 }
515 if (ruid != (uid_t) -1 ||
516 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
517 new->suid = new->euid;
518 new->fsuid = new->euid;
519
520 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
521 if (retval < 0)
522 goto error;
523
524 return commit_creds(new);
525
526 error:
527 abort_creds(new);
528 return retval;
529 }
530
531 /*
532 * setuid() is implemented like SysV with SAVED_IDS
533 *
534 * Note that SAVED_ID's is deficient in that a setuid root program
535 * like sendmail, for example, cannot set its uid to be a normal
536 * user and then switch back, because if you're root, setuid() sets
537 * the saved uid too. If you don't like this, blame the bright people
538 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
539 * will allow a root program to temporarily drop privileges and be able to
540 * regain them by swapping the real and effective uid.
541 */
542 SYSCALL_DEFINE1(setuid, uid_t, uid)
543 {
544 struct user_namespace *ns = current_user_ns();
545 const struct cred *old;
546 struct cred *new;
547 int retval;
548 kuid_t kuid;
549
550 kuid = make_kuid(ns, uid);
551 if (!uid_valid(kuid))
552 return -EINVAL;
553
554 new = prepare_creds();
555 if (!new)
556 return -ENOMEM;
557 old = current_cred();
558
559 retval = -EPERM;
560 if (ns_capable(old->user_ns, CAP_SETUID)) {
561 new->suid = new->uid = kuid;
562 if (!uid_eq(kuid, old->uid)) {
563 retval = set_user(new);
564 if (retval < 0)
565 goto error;
566 }
567 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
568 goto error;
569 }
570
571 new->fsuid = new->euid = kuid;
572
573 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
574 if (retval < 0)
575 goto error;
576
577 return commit_creds(new);
578
579 error:
580 abort_creds(new);
581 return retval;
582 }
583
584
585 /*
586 * This function implements a generic ability to update ruid, euid,
587 * and suid. This allows you to implement the 4.4 compatible seteuid().
588 */
589 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
590 {
591 struct user_namespace *ns = current_user_ns();
592 const struct cred *old;
593 struct cred *new;
594 int retval;
595 kuid_t kruid, keuid, ksuid;
596
597 kruid = make_kuid(ns, ruid);
598 keuid = make_kuid(ns, euid);
599 ksuid = make_kuid(ns, suid);
600
601 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
602 return -EINVAL;
603
604 if ((euid != (uid_t) -1) && !uid_valid(keuid))
605 return -EINVAL;
606
607 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
608 return -EINVAL;
609
610 new = prepare_creds();
611 if (!new)
612 return -ENOMEM;
613
614 old = current_cred();
615
616 retval = -EPERM;
617 if (!ns_capable(old->user_ns, CAP_SETUID)) {
618 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
619 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
620 goto error;
621 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
622 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
623 goto error;
624 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
625 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
626 goto error;
627 }
628
629 if (ruid != (uid_t) -1) {
630 new->uid = kruid;
631 if (!uid_eq(kruid, old->uid)) {
632 retval = set_user(new);
633 if (retval < 0)
634 goto error;
635 }
636 }
637 if (euid != (uid_t) -1)
638 new->euid = keuid;
639 if (suid != (uid_t) -1)
640 new->suid = ksuid;
641 new->fsuid = new->euid;
642
643 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
644 if (retval < 0)
645 goto error;
646
647 return commit_creds(new);
648
649 error:
650 abort_creds(new);
651 return retval;
652 }
653
654 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
655 {
656 const struct cred *cred = current_cred();
657 int retval;
658 uid_t ruid, euid, suid;
659
660 ruid = from_kuid_munged(cred->user_ns, cred->uid);
661 euid = from_kuid_munged(cred->user_ns, cred->euid);
662 suid = from_kuid_munged(cred->user_ns, cred->suid);
663
664 retval = put_user(ruid, ruidp);
665 if (!retval) {
666 retval = put_user(euid, euidp);
667 if (!retval)
668 return put_user(suid, suidp);
669 }
670 return retval;
671 }
672
673 /*
674 * Same as above, but for rgid, egid, sgid.
675 */
676 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
677 {
678 struct user_namespace *ns = current_user_ns();
679 const struct cred *old;
680 struct cred *new;
681 int retval;
682 kgid_t krgid, kegid, ksgid;
683
684 krgid = make_kgid(ns, rgid);
685 kegid = make_kgid(ns, egid);
686 ksgid = make_kgid(ns, sgid);
687
688 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
689 return -EINVAL;
690 if ((egid != (gid_t) -1) && !gid_valid(kegid))
691 return -EINVAL;
692 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
693 return -EINVAL;
694
695 new = prepare_creds();
696 if (!new)
697 return -ENOMEM;
698 old = current_cred();
699
700 retval = -EPERM;
701 if (!ns_capable(old->user_ns, CAP_SETGID)) {
702 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
703 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
704 goto error;
705 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
706 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
707 goto error;
708 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
709 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
710 goto error;
711 }
712
713 if (rgid != (gid_t) -1)
714 new->gid = krgid;
715 if (egid != (gid_t) -1)
716 new->egid = kegid;
717 if (sgid != (gid_t) -1)
718 new->sgid = ksgid;
719 new->fsgid = new->egid;
720
721 return commit_creds(new);
722
723 error:
724 abort_creds(new);
725 return retval;
726 }
727
728 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
729 {
730 const struct cred *cred = current_cred();
731 int retval;
732 gid_t rgid, egid, sgid;
733
734 rgid = from_kgid_munged(cred->user_ns, cred->gid);
735 egid = from_kgid_munged(cred->user_ns, cred->egid);
736 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
737
738 retval = put_user(rgid, rgidp);
739 if (!retval) {
740 retval = put_user(egid, egidp);
741 if (!retval)
742 retval = put_user(sgid, sgidp);
743 }
744
745 return retval;
746 }
747
748
749 /*
750 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
751 * is used for "access()" and for the NFS daemon (letting nfsd stay at
752 * whatever uid it wants to). It normally shadows "euid", except when
753 * explicitly set by setfsuid() or for access..
754 */
755 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
756 {
757 const struct cred *old;
758 struct cred *new;
759 uid_t old_fsuid;
760 kuid_t kuid;
761
762 old = current_cred();
763 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
764
765 kuid = make_kuid(old->user_ns, uid);
766 if (!uid_valid(kuid))
767 return old_fsuid;
768
769 new = prepare_creds();
770 if (!new)
771 return old_fsuid;
772
773 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
774 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
775 ns_capable(old->user_ns, CAP_SETUID)) {
776 if (!uid_eq(kuid, old->fsuid)) {
777 new->fsuid = kuid;
778 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
779 goto change_okay;
780 }
781 }
782
783 abort_creds(new);
784 return old_fsuid;
785
786 change_okay:
787 commit_creds(new);
788 return old_fsuid;
789 }
790
791 /*
792 * Samma på svenska..
793 */
794 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
795 {
796 const struct cred *old;
797 struct cred *new;
798 gid_t old_fsgid;
799 kgid_t kgid;
800
801 old = current_cred();
802 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
803
804 kgid = make_kgid(old->user_ns, gid);
805 if (!gid_valid(kgid))
806 return old_fsgid;
807
808 new = prepare_creds();
809 if (!new)
810 return old_fsgid;
811
812 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
813 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
814 ns_capable(old->user_ns, CAP_SETGID)) {
815 if (!gid_eq(kgid, old->fsgid)) {
816 new->fsgid = kgid;
817 goto change_okay;
818 }
819 }
820
821 abort_creds(new);
822 return old_fsgid;
823
824 change_okay:
825 commit_creds(new);
826 return old_fsgid;
827 }
828 #endif /* CONFIG_MULTIUSER */
829
830 /**
831 * sys_getpid - return the thread group id of the current process
832 *
833 * Note, despite the name, this returns the tgid not the pid. The tgid and
834 * the pid are identical unless CLONE_THREAD was specified on clone() in
835 * which case the tgid is the same in all threads of the same group.
836 *
837 * This is SMP safe as current->tgid does not change.
838 */
839 SYSCALL_DEFINE0(getpid)
840 {
841 return task_tgid_vnr(current);
842 }
843
844 /* Thread ID - the internal kernel "pid" */
845 SYSCALL_DEFINE0(gettid)
846 {
847 return task_pid_vnr(current);
848 }
849
850 /*
851 * Accessing ->real_parent is not SMP-safe, it could
852 * change from under us. However, we can use a stale
853 * value of ->real_parent under rcu_read_lock(), see
854 * release_task()->call_rcu(delayed_put_task_struct).
855 */
856 SYSCALL_DEFINE0(getppid)
857 {
858 int pid;
859
860 rcu_read_lock();
861 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
862 rcu_read_unlock();
863
864 return pid;
865 }
866
867 SYSCALL_DEFINE0(getuid)
868 {
869 /* Only we change this so SMP safe */
870 return from_kuid_munged(current_user_ns(), current_uid());
871 }
872
873 SYSCALL_DEFINE0(geteuid)
874 {
875 /* Only we change this so SMP safe */
876 return from_kuid_munged(current_user_ns(), current_euid());
877 }
878
879 SYSCALL_DEFINE0(getgid)
880 {
881 /* Only we change this so SMP safe */
882 return from_kgid_munged(current_user_ns(), current_gid());
883 }
884
885 SYSCALL_DEFINE0(getegid)
886 {
887 /* Only we change this so SMP safe */
888 return from_kgid_munged(current_user_ns(), current_egid());
889 }
890
891 static void do_sys_times(struct tms *tms)
892 {
893 u64 tgutime, tgstime, cutime, cstime;
894
895 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
896 cutime = current->signal->cutime;
897 cstime = current->signal->cstime;
898 tms->tms_utime = nsec_to_clock_t(tgutime);
899 tms->tms_stime = nsec_to_clock_t(tgstime);
900 tms->tms_cutime = nsec_to_clock_t(cutime);
901 tms->tms_cstime = nsec_to_clock_t(cstime);
902 }
903
904 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
905 {
906 if (tbuf) {
907 struct tms tmp;
908
909 do_sys_times(&tmp);
910 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
911 return -EFAULT;
912 }
913 force_successful_syscall_return();
914 return (long) jiffies_64_to_clock_t(get_jiffies_64());
915 }
916
917 #ifdef CONFIG_COMPAT
918 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
919 {
920 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
921 }
922
923 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
924 {
925 if (tbuf) {
926 struct tms tms;
927 struct compat_tms tmp;
928
929 do_sys_times(&tms);
930 /* Convert our struct tms to the compat version. */
931 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
932 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
933 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
934 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
935 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
936 return -EFAULT;
937 }
938 force_successful_syscall_return();
939 return compat_jiffies_to_clock_t(jiffies);
940 }
941 #endif
942
943 /*
944 * This needs some heavy checking ...
945 * I just haven't the stomach for it. I also don't fully
946 * understand sessions/pgrp etc. Let somebody who does explain it.
947 *
948 * OK, I think I have the protection semantics right.... this is really
949 * only important on a multi-user system anyway, to make sure one user
950 * can't send a signal to a process owned by another. -TYT, 12/12/91
951 *
952 * !PF_FORKNOEXEC check to conform completely to POSIX.
953 */
954 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
955 {
956 struct task_struct *p;
957 struct task_struct *group_leader = current->group_leader;
958 struct pid *pgrp;
959 int err;
960
961 if (!pid)
962 pid = task_pid_vnr(group_leader);
963 if (!pgid)
964 pgid = pid;
965 if (pgid < 0)
966 return -EINVAL;
967 rcu_read_lock();
968
969 /* From this point forward we keep holding onto the tasklist lock
970 * so that our parent does not change from under us. -DaveM
971 */
972 write_lock_irq(&tasklist_lock);
973
974 err = -ESRCH;
975 p = find_task_by_vpid(pid);
976 if (!p)
977 goto out;
978
979 err = -EINVAL;
980 if (!thread_group_leader(p))
981 goto out;
982
983 if (same_thread_group(p->real_parent, group_leader)) {
984 err = -EPERM;
985 if (task_session(p) != task_session(group_leader))
986 goto out;
987 err = -EACCES;
988 if (!(p->flags & PF_FORKNOEXEC))
989 goto out;
990 } else {
991 err = -ESRCH;
992 if (p != group_leader)
993 goto out;
994 }
995
996 err = -EPERM;
997 if (p->signal->leader)
998 goto out;
999
1000 pgrp = task_pid(p);
1001 if (pgid != pid) {
1002 struct task_struct *g;
1003
1004 pgrp = find_vpid(pgid);
1005 g = pid_task(pgrp, PIDTYPE_PGID);
1006 if (!g || task_session(g) != task_session(group_leader))
1007 goto out;
1008 }
1009
1010 err = security_task_setpgid(p, pgid);
1011 if (err)
1012 goto out;
1013
1014 if (task_pgrp(p) != pgrp)
1015 change_pid(p, PIDTYPE_PGID, pgrp);
1016
1017 err = 0;
1018 out:
1019 /* All paths lead to here, thus we are safe. -DaveM */
1020 write_unlock_irq(&tasklist_lock);
1021 rcu_read_unlock();
1022 return err;
1023 }
1024
1025 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1026 {
1027 struct task_struct *p;
1028 struct pid *grp;
1029 int retval;
1030
1031 rcu_read_lock();
1032 if (!pid)
1033 grp = task_pgrp(current);
1034 else {
1035 retval = -ESRCH;
1036 p = find_task_by_vpid(pid);
1037 if (!p)
1038 goto out;
1039 grp = task_pgrp(p);
1040 if (!grp)
1041 goto out;
1042
1043 retval = security_task_getpgid(p);
1044 if (retval)
1045 goto out;
1046 }
1047 retval = pid_vnr(grp);
1048 out:
1049 rcu_read_unlock();
1050 return retval;
1051 }
1052
1053 #ifdef __ARCH_WANT_SYS_GETPGRP
1054
1055 SYSCALL_DEFINE0(getpgrp)
1056 {
1057 return sys_getpgid(0);
1058 }
1059
1060 #endif
1061
1062 SYSCALL_DEFINE1(getsid, pid_t, pid)
1063 {
1064 struct task_struct *p;
1065 struct pid *sid;
1066 int retval;
1067
1068 rcu_read_lock();
1069 if (!pid)
1070 sid = task_session(current);
1071 else {
1072 retval = -ESRCH;
1073 p = find_task_by_vpid(pid);
1074 if (!p)
1075 goto out;
1076 sid = task_session(p);
1077 if (!sid)
1078 goto out;
1079
1080 retval = security_task_getsid(p);
1081 if (retval)
1082 goto out;
1083 }
1084 retval = pid_vnr(sid);
1085 out:
1086 rcu_read_unlock();
1087 return retval;
1088 }
1089
1090 static void set_special_pids(struct pid *pid)
1091 {
1092 struct task_struct *curr = current->group_leader;
1093
1094 if (task_session(curr) != pid)
1095 change_pid(curr, PIDTYPE_SID, pid);
1096
1097 if (task_pgrp(curr) != pid)
1098 change_pid(curr, PIDTYPE_PGID, pid);
1099 }
1100
1101 SYSCALL_DEFINE0(setsid)
1102 {
1103 struct task_struct *group_leader = current->group_leader;
1104 struct pid *sid = task_pid(group_leader);
1105 pid_t session = pid_vnr(sid);
1106 int err = -EPERM;
1107
1108 write_lock_irq(&tasklist_lock);
1109 /* Fail if I am already a session leader */
1110 if (group_leader->signal->leader)
1111 goto out;
1112
1113 /* Fail if a process group id already exists that equals the
1114 * proposed session id.
1115 */
1116 if (pid_task(sid, PIDTYPE_PGID))
1117 goto out;
1118
1119 group_leader->signal->leader = 1;
1120 set_special_pids(sid);
1121
1122 proc_clear_tty(group_leader);
1123
1124 err = session;
1125 out:
1126 write_unlock_irq(&tasklist_lock);
1127 if (err > 0) {
1128 proc_sid_connector(group_leader);
1129 sched_autogroup_create_attach(group_leader);
1130 }
1131 return err;
1132 }
1133
1134 DECLARE_RWSEM(uts_sem);
1135
1136 #ifdef COMPAT_UTS_MACHINE
1137 static char compat_uts_machine[__OLD_UTS_LEN+1] = COMPAT_UTS_MACHINE;
1138
1139 static int __init parse_compat_uts_machine(char *arg)
1140 {
1141 strncpy(compat_uts_machine, arg, __OLD_UTS_LEN);
1142 compat_uts_machine[__OLD_UTS_LEN] = 0;
1143 return 0;
1144 }
1145 early_param("compat_uts_machine", parse_compat_uts_machine);
1146
1147 #undef COMPAT_UTS_MACHINE
1148 #define COMPAT_UTS_MACHINE compat_uts_machine
1149 #endif
1150
1151 #ifdef COMPAT_UTS_MACHINE
1152 #define override_architecture(name) \
1153 (personality(current->personality) == PER_LINUX32 && \
1154 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1155 sizeof(COMPAT_UTS_MACHINE)))
1156 #else
1157 #define override_architecture(name) 0
1158 #endif
1159
1160 /*
1161 * Work around broken programs that cannot handle "Linux 3.0".
1162 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1163 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1164 */
1165 static int override_release(char __user *release, size_t len)
1166 {
1167 int ret = 0;
1168
1169 if (current->personality & UNAME26) {
1170 const char *rest = UTS_RELEASE;
1171 char buf[65] = { 0 };
1172 int ndots = 0;
1173 unsigned v;
1174 size_t copy;
1175
1176 while (*rest) {
1177 if (*rest == '.' && ++ndots >= 3)
1178 break;
1179 if (!isdigit(*rest) && *rest != '.')
1180 break;
1181 rest++;
1182 }
1183 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1184 copy = clamp_t(size_t, len, 1, sizeof(buf));
1185 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1186 ret = copy_to_user(release, buf, copy + 1);
1187 }
1188 return ret;
1189 }
1190
1191 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1192 {
1193 int errno = 0;
1194
1195 down_read(&uts_sem);
1196 if (copy_to_user(name, utsname(), sizeof *name))
1197 errno = -EFAULT;
1198 up_read(&uts_sem);
1199
1200 if (!errno && override_release(name->release, sizeof(name->release)))
1201 errno = -EFAULT;
1202 if (!errno && override_architecture(name))
1203 errno = -EFAULT;
1204 return errno;
1205 }
1206
1207 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1208 /*
1209 * Old cruft
1210 */
1211 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1212 {
1213 int error = 0;
1214
1215 if (!name)
1216 return -EFAULT;
1217
1218 down_read(&uts_sem);
1219 if (copy_to_user(name, utsname(), sizeof(*name)))
1220 error = -EFAULT;
1221 up_read(&uts_sem);
1222
1223 if (!error && override_release(name->release, sizeof(name->release)))
1224 error = -EFAULT;
1225 if (!error && override_architecture(name))
1226 error = -EFAULT;
1227 return error;
1228 }
1229
1230 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1231 {
1232 int error;
1233
1234 if (!name)
1235 return -EFAULT;
1236 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1237 return -EFAULT;
1238
1239 down_read(&uts_sem);
1240 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1241 __OLD_UTS_LEN);
1242 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1243 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1244 __OLD_UTS_LEN);
1245 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1246 error |= __copy_to_user(&name->release, &utsname()->release,
1247 __OLD_UTS_LEN);
1248 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1249 error |= __copy_to_user(&name->version, &utsname()->version,
1250 __OLD_UTS_LEN);
1251 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1252 error |= __copy_to_user(&name->machine, &utsname()->machine,
1253 __OLD_UTS_LEN);
1254 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1255 up_read(&uts_sem);
1256
1257 if (!error && override_architecture(name))
1258 error = -EFAULT;
1259 if (!error && override_release(name->release, sizeof(name->release)))
1260 error = -EFAULT;
1261 return error ? -EFAULT : 0;
1262 }
1263 #endif
1264
1265 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1266 {
1267 int errno;
1268 char tmp[__NEW_UTS_LEN];
1269
1270 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1271 return -EPERM;
1272
1273 if (len < 0 || len > __NEW_UTS_LEN)
1274 return -EINVAL;
1275 down_write(&uts_sem);
1276 errno = -EFAULT;
1277 if (!copy_from_user(tmp, name, len)) {
1278 struct new_utsname *u = utsname();
1279
1280 memcpy(u->nodename, tmp, len);
1281 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1282 errno = 0;
1283 uts_proc_notify(UTS_PROC_HOSTNAME);
1284 }
1285 up_write(&uts_sem);
1286 return errno;
1287 }
1288
1289 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1290
1291 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1292 {
1293 int i, errno;
1294 struct new_utsname *u;
1295
1296 if (len < 0)
1297 return -EINVAL;
1298 down_read(&uts_sem);
1299 u = utsname();
1300 i = 1 + strlen(u->nodename);
1301 if (i > len)
1302 i = len;
1303 errno = 0;
1304 if (copy_to_user(name, u->nodename, i))
1305 errno = -EFAULT;
1306 up_read(&uts_sem);
1307 return errno;
1308 }
1309
1310 #endif
1311
1312 /*
1313 * Only setdomainname; getdomainname can be implemented by calling
1314 * uname()
1315 */
1316 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1317 {
1318 int errno;
1319 char tmp[__NEW_UTS_LEN];
1320
1321 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1322 return -EPERM;
1323 if (len < 0 || len > __NEW_UTS_LEN)
1324 return -EINVAL;
1325
1326 down_write(&uts_sem);
1327 errno = -EFAULT;
1328 if (!copy_from_user(tmp, name, len)) {
1329 struct new_utsname *u = utsname();
1330
1331 memcpy(u->domainname, tmp, len);
1332 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1333 errno = 0;
1334 uts_proc_notify(UTS_PROC_DOMAINNAME);
1335 }
1336 up_write(&uts_sem);
1337 return errno;
1338 }
1339
1340 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1341 {
1342 struct rlimit value;
1343 int ret;
1344
1345 ret = do_prlimit(current, resource, NULL, &value);
1346 if (!ret)
1347 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1348
1349 return ret;
1350 }
1351
1352 #ifdef CONFIG_COMPAT
1353
1354 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1355 struct compat_rlimit __user *, rlim)
1356 {
1357 struct rlimit r;
1358 struct compat_rlimit r32;
1359
1360 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1361 return -EFAULT;
1362
1363 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1364 r.rlim_cur = RLIM_INFINITY;
1365 else
1366 r.rlim_cur = r32.rlim_cur;
1367 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1368 r.rlim_max = RLIM_INFINITY;
1369 else
1370 r.rlim_max = r32.rlim_max;
1371 return do_prlimit(current, resource, &r, NULL);
1372 }
1373
1374 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1375 struct compat_rlimit __user *, rlim)
1376 {
1377 struct rlimit r;
1378 int ret;
1379
1380 ret = do_prlimit(current, resource, NULL, &r);
1381 if (!ret) {
1382 struct compat_rlimit r32;
1383 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1384 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1385 else
1386 r32.rlim_cur = r.rlim_cur;
1387 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1388 r32.rlim_max = COMPAT_RLIM_INFINITY;
1389 else
1390 r32.rlim_max = r.rlim_max;
1391
1392 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1393 return -EFAULT;
1394 }
1395 return ret;
1396 }
1397
1398 #endif
1399
1400 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1401
1402 /*
1403 * Back compatibility for getrlimit. Needed for some apps.
1404 */
1405 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1406 struct rlimit __user *, rlim)
1407 {
1408 struct rlimit x;
1409 if (resource >= RLIM_NLIMITS)
1410 return -EINVAL;
1411
1412 task_lock(current->group_leader);
1413 x = current->signal->rlim[resource];
1414 task_unlock(current->group_leader);
1415 if (x.rlim_cur > 0x7FFFFFFF)
1416 x.rlim_cur = 0x7FFFFFFF;
1417 if (x.rlim_max > 0x7FFFFFFF)
1418 x.rlim_max = 0x7FFFFFFF;
1419 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1420 }
1421
1422 #ifdef CONFIG_COMPAT
1423 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1424 struct compat_rlimit __user *, rlim)
1425 {
1426 struct rlimit r;
1427
1428 if (resource >= RLIM_NLIMITS)
1429 return -EINVAL;
1430
1431 task_lock(current->group_leader);
1432 r = current->signal->rlim[resource];
1433 task_unlock(current->group_leader);
1434 if (r.rlim_cur > 0x7FFFFFFF)
1435 r.rlim_cur = 0x7FFFFFFF;
1436 if (r.rlim_max > 0x7FFFFFFF)
1437 r.rlim_max = 0x7FFFFFFF;
1438
1439 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1440 put_user(r.rlim_max, &rlim->rlim_max))
1441 return -EFAULT;
1442 return 0;
1443 }
1444 #endif
1445
1446 #endif
1447
1448 static inline bool rlim64_is_infinity(__u64 rlim64)
1449 {
1450 #if BITS_PER_LONG < 64
1451 return rlim64 >= ULONG_MAX;
1452 #else
1453 return rlim64 == RLIM64_INFINITY;
1454 #endif
1455 }
1456
1457 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1458 {
1459 if (rlim->rlim_cur == RLIM_INFINITY)
1460 rlim64->rlim_cur = RLIM64_INFINITY;
1461 else
1462 rlim64->rlim_cur = rlim->rlim_cur;
1463 if (rlim->rlim_max == RLIM_INFINITY)
1464 rlim64->rlim_max = RLIM64_INFINITY;
1465 else
1466 rlim64->rlim_max = rlim->rlim_max;
1467 }
1468
1469 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1470 {
1471 if (rlim64_is_infinity(rlim64->rlim_cur))
1472 rlim->rlim_cur = RLIM_INFINITY;
1473 else
1474 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1475 if (rlim64_is_infinity(rlim64->rlim_max))
1476 rlim->rlim_max = RLIM_INFINITY;
1477 else
1478 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1479 }
1480
1481 /* make sure you are allowed to change @tsk limits before calling this */
1482 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1483 struct rlimit *new_rlim, struct rlimit *old_rlim)
1484 {
1485 struct rlimit *rlim;
1486 int retval = 0;
1487
1488 if (resource >= RLIM_NLIMITS)
1489 return -EINVAL;
1490 if (new_rlim) {
1491 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1492 return -EINVAL;
1493 if (resource == RLIMIT_NOFILE &&
1494 new_rlim->rlim_max > sysctl_nr_open)
1495 return -EPERM;
1496 }
1497
1498 /* protect tsk->signal and tsk->sighand from disappearing */
1499 read_lock(&tasklist_lock);
1500 if (!tsk->sighand) {
1501 retval = -ESRCH;
1502 goto out;
1503 }
1504
1505 rlim = tsk->signal->rlim + resource;
1506 task_lock(tsk->group_leader);
1507 if (new_rlim) {
1508 /* Keep the capable check against init_user_ns until
1509 cgroups can contain all limits */
1510 if (new_rlim->rlim_max > rlim->rlim_max &&
1511 !capable(CAP_SYS_RESOURCE))
1512 retval = -EPERM;
1513 if (!retval)
1514 retval = security_task_setrlimit(tsk, resource, new_rlim);
1515 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1516 /*
1517 * The caller is asking for an immediate RLIMIT_CPU
1518 * expiry. But we use the zero value to mean "it was
1519 * never set". So let's cheat and make it one second
1520 * instead
1521 */
1522 new_rlim->rlim_cur = 1;
1523 }
1524 }
1525 if (!retval) {
1526 if (old_rlim)
1527 *old_rlim = *rlim;
1528 if (new_rlim)
1529 *rlim = *new_rlim;
1530 }
1531 task_unlock(tsk->group_leader);
1532
1533 /*
1534 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1535 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1536 * very long-standing error, and fixing it now risks breakage of
1537 * applications, so we live with it
1538 */
1539 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1540 new_rlim->rlim_cur != RLIM_INFINITY &&
1541 IS_ENABLED(CONFIG_POSIX_TIMERS))
1542 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1543 out:
1544 read_unlock(&tasklist_lock);
1545 return retval;
1546 }
1547
1548 /* rcu lock must be held */
1549 static int check_prlimit_permission(struct task_struct *task,
1550 unsigned int flags)
1551 {
1552 const struct cred *cred = current_cred(), *tcred;
1553 bool id_match;
1554
1555 if (current == task)
1556 return 0;
1557
1558 tcred = __task_cred(task);
1559 id_match = (uid_eq(cred->uid, tcred->euid) &&
1560 uid_eq(cred->uid, tcred->suid) &&
1561 uid_eq(cred->uid, tcred->uid) &&
1562 gid_eq(cred->gid, tcred->egid) &&
1563 gid_eq(cred->gid, tcred->sgid) &&
1564 gid_eq(cred->gid, tcred->gid));
1565 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1566 return -EPERM;
1567
1568 return security_task_prlimit(cred, tcred, flags);
1569 }
1570
1571 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1572 const struct rlimit64 __user *, new_rlim,
1573 struct rlimit64 __user *, old_rlim)
1574 {
1575 struct rlimit64 old64, new64;
1576 struct rlimit old, new;
1577 struct task_struct *tsk;
1578 unsigned int checkflags = 0;
1579 int ret;
1580
1581 if (old_rlim)
1582 checkflags |= LSM_PRLIMIT_READ;
1583
1584 if (new_rlim) {
1585 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1586 return -EFAULT;
1587 rlim64_to_rlim(&new64, &new);
1588 checkflags |= LSM_PRLIMIT_WRITE;
1589 }
1590
1591 rcu_read_lock();
1592 tsk = pid ? find_task_by_vpid(pid) : current;
1593 if (!tsk) {
1594 rcu_read_unlock();
1595 return -ESRCH;
1596 }
1597 ret = check_prlimit_permission(tsk, checkflags);
1598 if (ret) {
1599 rcu_read_unlock();
1600 return ret;
1601 }
1602 get_task_struct(tsk);
1603 rcu_read_unlock();
1604
1605 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1606 old_rlim ? &old : NULL);
1607
1608 if (!ret && old_rlim) {
1609 rlim_to_rlim64(&old, &old64);
1610 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1611 ret = -EFAULT;
1612 }
1613
1614 put_task_struct(tsk);
1615 return ret;
1616 }
1617
1618 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1619 {
1620 struct rlimit new_rlim;
1621
1622 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1623 return -EFAULT;
1624 return do_prlimit(current, resource, &new_rlim, NULL);
1625 }
1626
1627 /*
1628 * It would make sense to put struct rusage in the task_struct,
1629 * except that would make the task_struct be *really big*. After
1630 * task_struct gets moved into malloc'ed memory, it would
1631 * make sense to do this. It will make moving the rest of the information
1632 * a lot simpler! (Which we're not doing right now because we're not
1633 * measuring them yet).
1634 *
1635 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1636 * races with threads incrementing their own counters. But since word
1637 * reads are atomic, we either get new values or old values and we don't
1638 * care which for the sums. We always take the siglock to protect reading
1639 * the c* fields from p->signal from races with exit.c updating those
1640 * fields when reaping, so a sample either gets all the additions of a
1641 * given child after it's reaped, or none so this sample is before reaping.
1642 *
1643 * Locking:
1644 * We need to take the siglock for CHILDEREN, SELF and BOTH
1645 * for the cases current multithreaded, non-current single threaded
1646 * non-current multithreaded. Thread traversal is now safe with
1647 * the siglock held.
1648 * Strictly speaking, we donot need to take the siglock if we are current and
1649 * single threaded, as no one else can take our signal_struct away, no one
1650 * else can reap the children to update signal->c* counters, and no one else
1651 * can race with the signal-> fields. If we do not take any lock, the
1652 * signal-> fields could be read out of order while another thread was just
1653 * exiting. So we should place a read memory barrier when we avoid the lock.
1654 * On the writer side, write memory barrier is implied in __exit_signal
1655 * as __exit_signal releases the siglock spinlock after updating the signal->
1656 * fields. But we don't do this yet to keep things simple.
1657 *
1658 */
1659
1660 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1661 {
1662 r->ru_nvcsw += t->nvcsw;
1663 r->ru_nivcsw += t->nivcsw;
1664 r->ru_minflt += t->min_flt;
1665 r->ru_majflt += t->maj_flt;
1666 r->ru_inblock += task_io_get_inblock(t);
1667 r->ru_oublock += task_io_get_oublock(t);
1668 }
1669
1670 void getrusage(struct task_struct *p, int who, struct rusage *r)
1671 {
1672 struct task_struct *t;
1673 unsigned long flags;
1674 u64 tgutime, tgstime, utime, stime;
1675 unsigned long maxrss = 0;
1676
1677 memset((char *)r, 0, sizeof (*r));
1678 utime = stime = 0;
1679
1680 if (who == RUSAGE_THREAD) {
1681 task_cputime_adjusted(current, &utime, &stime);
1682 accumulate_thread_rusage(p, r);
1683 maxrss = p->signal->maxrss;
1684 goto out;
1685 }
1686
1687 if (!lock_task_sighand(p, &flags))
1688 return;
1689
1690 switch (who) {
1691 case RUSAGE_BOTH:
1692 case RUSAGE_CHILDREN:
1693 utime = p->signal->cutime;
1694 stime = p->signal->cstime;
1695 r->ru_nvcsw = p->signal->cnvcsw;
1696 r->ru_nivcsw = p->signal->cnivcsw;
1697 r->ru_minflt = p->signal->cmin_flt;
1698 r->ru_majflt = p->signal->cmaj_flt;
1699 r->ru_inblock = p->signal->cinblock;
1700 r->ru_oublock = p->signal->coublock;
1701 maxrss = p->signal->cmaxrss;
1702
1703 if (who == RUSAGE_CHILDREN)
1704 break;
1705
1706 case RUSAGE_SELF:
1707 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1708 utime += tgutime;
1709 stime += tgstime;
1710 r->ru_nvcsw += p->signal->nvcsw;
1711 r->ru_nivcsw += p->signal->nivcsw;
1712 r->ru_minflt += p->signal->min_flt;
1713 r->ru_majflt += p->signal->maj_flt;
1714 r->ru_inblock += p->signal->inblock;
1715 r->ru_oublock += p->signal->oublock;
1716 if (maxrss < p->signal->maxrss)
1717 maxrss = p->signal->maxrss;
1718 t = p;
1719 do {
1720 accumulate_thread_rusage(t, r);
1721 } while_each_thread(p, t);
1722 break;
1723
1724 default:
1725 BUG();
1726 }
1727 unlock_task_sighand(p, &flags);
1728
1729 out:
1730 r->ru_utime = ns_to_timeval(utime);
1731 r->ru_stime = ns_to_timeval(stime);
1732
1733 if (who != RUSAGE_CHILDREN) {
1734 struct mm_struct *mm = get_task_mm(p);
1735
1736 if (mm) {
1737 setmax_mm_hiwater_rss(&maxrss, mm);
1738 mmput(mm);
1739 }
1740 }
1741 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1742 }
1743
1744 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1745 {
1746 struct rusage r;
1747
1748 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1749 who != RUSAGE_THREAD)
1750 return -EINVAL;
1751
1752 getrusage(current, who, &r);
1753 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1754 }
1755
1756 #ifdef CONFIG_COMPAT
1757 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1758 {
1759 struct rusage r;
1760
1761 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1762 who != RUSAGE_THREAD)
1763 return -EINVAL;
1764
1765 getrusage(current, who, &r);
1766 return put_compat_rusage(&r, ru);
1767 }
1768 #endif
1769
1770 SYSCALL_DEFINE1(umask, int, mask)
1771 {
1772 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1773 return mask;
1774 }
1775
1776 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1777 {
1778 struct fd exe;
1779 struct file *old_exe, *exe_file;
1780 struct inode *inode;
1781 int err;
1782
1783 exe = fdget(fd);
1784 if (!exe.file)
1785 return -EBADF;
1786
1787 inode = file_inode(exe.file);
1788
1789 /*
1790 * Because the original mm->exe_file points to executable file, make
1791 * sure that this one is executable as well, to avoid breaking an
1792 * overall picture.
1793 */
1794 err = -EACCES;
1795 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1796 goto exit;
1797
1798 err = inode_permission(inode, MAY_EXEC);
1799 if (err)
1800 goto exit;
1801
1802 /*
1803 * Forbid mm->exe_file change if old file still mapped.
1804 */
1805 exe_file = get_mm_exe_file(mm);
1806 err = -EBUSY;
1807 if (exe_file) {
1808 struct vm_area_struct *vma;
1809
1810 down_read(&mm->mmap_sem);
1811 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1812 if (!vma->vm_file)
1813 continue;
1814 if (path_equal(&vma->vm_file->f_path,
1815 &exe_file->f_path))
1816 goto exit_err;
1817 }
1818
1819 up_read(&mm->mmap_sem);
1820 fput(exe_file);
1821 }
1822
1823 err = 0;
1824 /* set the new file, lockless */
1825 get_file(exe.file);
1826 old_exe = xchg(&mm->exe_file, exe.file);
1827 if (old_exe)
1828 fput(old_exe);
1829 exit:
1830 fdput(exe);
1831 return err;
1832 exit_err:
1833 up_read(&mm->mmap_sem);
1834 fput(exe_file);
1835 goto exit;
1836 }
1837
1838 /*
1839 * WARNING: we don't require any capability here so be very careful
1840 * in what is allowed for modification from userspace.
1841 */
1842 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1843 {
1844 unsigned long mmap_max_addr = TASK_SIZE;
1845 struct mm_struct *mm = current->mm;
1846 int error = -EINVAL, i;
1847
1848 static const unsigned char offsets[] = {
1849 offsetof(struct prctl_mm_map, start_code),
1850 offsetof(struct prctl_mm_map, end_code),
1851 offsetof(struct prctl_mm_map, start_data),
1852 offsetof(struct prctl_mm_map, end_data),
1853 offsetof(struct prctl_mm_map, start_brk),
1854 offsetof(struct prctl_mm_map, brk),
1855 offsetof(struct prctl_mm_map, start_stack),
1856 offsetof(struct prctl_mm_map, arg_start),
1857 offsetof(struct prctl_mm_map, arg_end),
1858 offsetof(struct prctl_mm_map, env_start),
1859 offsetof(struct prctl_mm_map, env_end),
1860 };
1861
1862 /*
1863 * Make sure the members are not somewhere outside
1864 * of allowed address space.
1865 */
1866 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1867 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1868
1869 if ((unsigned long)val >= mmap_max_addr ||
1870 (unsigned long)val < mmap_min_addr)
1871 goto out;
1872 }
1873
1874 /*
1875 * Make sure the pairs are ordered.
1876 */
1877 #define __prctl_check_order(__m1, __op, __m2) \
1878 ((unsigned long)prctl_map->__m1 __op \
1879 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1880 error = __prctl_check_order(start_code, <, end_code);
1881 error |= __prctl_check_order(start_data, <, end_data);
1882 error |= __prctl_check_order(start_brk, <=, brk);
1883 error |= __prctl_check_order(arg_start, <=, arg_end);
1884 error |= __prctl_check_order(env_start, <=, env_end);
1885 if (error)
1886 goto out;
1887 #undef __prctl_check_order
1888
1889 error = -EINVAL;
1890
1891 /*
1892 * @brk should be after @end_data in traditional maps.
1893 */
1894 if (prctl_map->start_brk <= prctl_map->end_data ||
1895 prctl_map->brk <= prctl_map->end_data)
1896 goto out;
1897
1898 /*
1899 * Neither we should allow to override limits if they set.
1900 */
1901 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1902 prctl_map->start_brk, prctl_map->end_data,
1903 prctl_map->start_data))
1904 goto out;
1905
1906 /*
1907 * Someone is trying to cheat the auxv vector.
1908 */
1909 if (prctl_map->auxv_size) {
1910 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1911 goto out;
1912 }
1913
1914 /*
1915 * Finally, make sure the caller has the rights to
1916 * change /proc/pid/exe link: only local root should
1917 * be allowed to.
1918 */
1919 if (prctl_map->exe_fd != (u32)-1) {
1920 struct user_namespace *ns = current_user_ns();
1921 const struct cred *cred = current_cred();
1922
1923 if (!uid_eq(cred->uid, make_kuid(ns, 0)) ||
1924 !gid_eq(cred->gid, make_kgid(ns, 0)))
1925 goto out;
1926 }
1927
1928 error = 0;
1929 out:
1930 return error;
1931 }
1932
1933 #ifdef CONFIG_CHECKPOINT_RESTORE
1934 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1935 {
1936 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1937 unsigned long user_auxv[AT_VECTOR_SIZE];
1938 struct mm_struct *mm = current->mm;
1939 int error;
1940
1941 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1942 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1943
1944 if (opt == PR_SET_MM_MAP_SIZE)
1945 return put_user((unsigned int)sizeof(prctl_map),
1946 (unsigned int __user *)addr);
1947
1948 if (data_size != sizeof(prctl_map))
1949 return -EINVAL;
1950
1951 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1952 return -EFAULT;
1953
1954 error = validate_prctl_map(&prctl_map);
1955 if (error)
1956 return error;
1957
1958 if (prctl_map.auxv_size) {
1959 memset(user_auxv, 0, sizeof(user_auxv));
1960 if (copy_from_user(user_auxv,
1961 (const void __user *)prctl_map.auxv,
1962 prctl_map.auxv_size))
1963 return -EFAULT;
1964
1965 /* Last entry must be AT_NULL as specification requires */
1966 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1967 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1968 }
1969
1970 if (prctl_map.exe_fd != (u32)-1) {
1971 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1972 if (error)
1973 return error;
1974 }
1975
1976 down_write(&mm->mmap_sem);
1977
1978 /*
1979 * We don't validate if these members are pointing to
1980 * real present VMAs because application may have correspond
1981 * VMAs already unmapped and kernel uses these members for statistics
1982 * output in procfs mostly, except
1983 *
1984 * - @start_brk/@brk which are used in do_brk but kernel lookups
1985 * for VMAs when updating these memvers so anything wrong written
1986 * here cause kernel to swear at userspace program but won't lead
1987 * to any problem in kernel itself
1988 */
1989
1990 mm->start_code = prctl_map.start_code;
1991 mm->end_code = prctl_map.end_code;
1992 mm->start_data = prctl_map.start_data;
1993 mm->end_data = prctl_map.end_data;
1994 mm->start_brk = prctl_map.start_brk;
1995 mm->brk = prctl_map.brk;
1996 mm->start_stack = prctl_map.start_stack;
1997 mm->arg_start = prctl_map.arg_start;
1998 mm->arg_end = prctl_map.arg_end;
1999 mm->env_start = prctl_map.env_start;
2000 mm->env_end = prctl_map.env_end;
2001
2002 /*
2003 * Note this update of @saved_auxv is lockless thus
2004 * if someone reads this member in procfs while we're
2005 * updating -- it may get partly updated results. It's
2006 * known and acceptable trade off: we leave it as is to
2007 * not introduce additional locks here making the kernel
2008 * more complex.
2009 */
2010 if (prctl_map.auxv_size)
2011 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2012
2013 up_write(&mm->mmap_sem);
2014 return 0;
2015 }
2016 #endif /* CONFIG_CHECKPOINT_RESTORE */
2017
2018 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2019 unsigned long len)
2020 {
2021 /*
2022 * This doesn't move the auxiliary vector itself since it's pinned to
2023 * mm_struct, but it permits filling the vector with new values. It's
2024 * up to the caller to provide sane values here, otherwise userspace
2025 * tools which use this vector might be unhappy.
2026 */
2027 unsigned long user_auxv[AT_VECTOR_SIZE];
2028
2029 if (len > sizeof(user_auxv))
2030 return -EINVAL;
2031
2032 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2033 return -EFAULT;
2034
2035 /* Make sure the last entry is always AT_NULL */
2036 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2037 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2038
2039 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2040
2041 task_lock(current);
2042 memcpy(mm->saved_auxv, user_auxv, len);
2043 task_unlock(current);
2044
2045 return 0;
2046 }
2047
2048 static int prctl_set_mm(int opt, unsigned long addr,
2049 unsigned long arg4, unsigned long arg5)
2050 {
2051 struct mm_struct *mm = current->mm;
2052 struct prctl_mm_map prctl_map;
2053 struct vm_area_struct *vma;
2054 int error;
2055
2056 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2057 opt != PR_SET_MM_MAP &&
2058 opt != PR_SET_MM_MAP_SIZE)))
2059 return -EINVAL;
2060
2061 #ifdef CONFIG_CHECKPOINT_RESTORE
2062 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2063 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2064 #endif
2065
2066 if (!capable(CAP_SYS_RESOURCE))
2067 return -EPERM;
2068
2069 if (opt == PR_SET_MM_EXE_FILE)
2070 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2071
2072 if (opt == PR_SET_MM_AUXV)
2073 return prctl_set_auxv(mm, addr, arg4);
2074
2075 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2076 return -EINVAL;
2077
2078 error = -EINVAL;
2079
2080 down_write(&mm->mmap_sem);
2081 vma = find_vma(mm, addr);
2082
2083 prctl_map.start_code = mm->start_code;
2084 prctl_map.end_code = mm->end_code;
2085 prctl_map.start_data = mm->start_data;
2086 prctl_map.end_data = mm->end_data;
2087 prctl_map.start_brk = mm->start_brk;
2088 prctl_map.brk = mm->brk;
2089 prctl_map.start_stack = mm->start_stack;
2090 prctl_map.arg_start = mm->arg_start;
2091 prctl_map.arg_end = mm->arg_end;
2092 prctl_map.env_start = mm->env_start;
2093 prctl_map.env_end = mm->env_end;
2094 prctl_map.auxv = NULL;
2095 prctl_map.auxv_size = 0;
2096 prctl_map.exe_fd = -1;
2097
2098 switch (opt) {
2099 case PR_SET_MM_START_CODE:
2100 prctl_map.start_code = addr;
2101 break;
2102 case PR_SET_MM_END_CODE:
2103 prctl_map.end_code = addr;
2104 break;
2105 case PR_SET_MM_START_DATA:
2106 prctl_map.start_data = addr;
2107 break;
2108 case PR_SET_MM_END_DATA:
2109 prctl_map.end_data = addr;
2110 break;
2111 case PR_SET_MM_START_STACK:
2112 prctl_map.start_stack = addr;
2113 break;
2114 case PR_SET_MM_START_BRK:
2115 prctl_map.start_brk = addr;
2116 break;
2117 case PR_SET_MM_BRK:
2118 prctl_map.brk = addr;
2119 break;
2120 case PR_SET_MM_ARG_START:
2121 prctl_map.arg_start = addr;
2122 break;
2123 case PR_SET_MM_ARG_END:
2124 prctl_map.arg_end = addr;
2125 break;
2126 case PR_SET_MM_ENV_START:
2127 prctl_map.env_start = addr;
2128 break;
2129 case PR_SET_MM_ENV_END:
2130 prctl_map.env_end = addr;
2131 break;
2132 default:
2133 goto out;
2134 }
2135
2136 error = validate_prctl_map(&prctl_map);
2137 if (error)
2138 goto out;
2139
2140 switch (opt) {
2141 /*
2142 * If command line arguments and environment
2143 * are placed somewhere else on stack, we can
2144 * set them up here, ARG_START/END to setup
2145 * command line argumets and ENV_START/END
2146 * for environment.
2147 */
2148 case PR_SET_MM_START_STACK:
2149 case PR_SET_MM_ARG_START:
2150 case PR_SET_MM_ARG_END:
2151 case PR_SET_MM_ENV_START:
2152 case PR_SET_MM_ENV_END:
2153 if (!vma) {
2154 error = -EFAULT;
2155 goto out;
2156 }
2157 }
2158
2159 mm->start_code = prctl_map.start_code;
2160 mm->end_code = prctl_map.end_code;
2161 mm->start_data = prctl_map.start_data;
2162 mm->end_data = prctl_map.end_data;
2163 mm->start_brk = prctl_map.start_brk;
2164 mm->brk = prctl_map.brk;
2165 mm->start_stack = prctl_map.start_stack;
2166 mm->arg_start = prctl_map.arg_start;
2167 mm->arg_end = prctl_map.arg_end;
2168 mm->env_start = prctl_map.env_start;
2169 mm->env_end = prctl_map.env_end;
2170
2171 error = 0;
2172 out:
2173 up_write(&mm->mmap_sem);
2174 return error;
2175 }
2176
2177 #ifdef CONFIG_CHECKPOINT_RESTORE
2178 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2179 {
2180 return put_user(me->clear_child_tid, tid_addr);
2181 }
2182 #else
2183 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2184 {
2185 return -EINVAL;
2186 }
2187 #endif
2188
2189 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2190 {
2191 /*
2192 * If task has has_child_subreaper - all its decendants
2193 * already have these flag too and new decendants will
2194 * inherit it on fork, skip them.
2195 *
2196 * If we've found child_reaper - skip descendants in
2197 * it's subtree as they will never get out pidns.
2198 */
2199 if (p->signal->has_child_subreaper ||
2200 is_child_reaper(task_pid(p)))
2201 return 0;
2202
2203 p->signal->has_child_subreaper = 1;
2204 return 1;
2205 }
2206
2207 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2208 {
2209 return -EINVAL;
2210 }
2211
2212 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2213 unsigned long ctrl)
2214 {
2215 return -EINVAL;
2216 }
2217
2218 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2219 unsigned long, arg4, unsigned long, arg5)
2220 {
2221 struct task_struct *me = current;
2222 unsigned char comm[sizeof(me->comm)];
2223 long error;
2224
2225 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2226 if (error != -ENOSYS)
2227 return error;
2228
2229 error = 0;
2230 switch (option) {
2231 case PR_SET_PDEATHSIG:
2232 if (!valid_signal(arg2)) {
2233 error = -EINVAL;
2234 break;
2235 }
2236 me->pdeath_signal = arg2;
2237 break;
2238 case PR_GET_PDEATHSIG:
2239 error = put_user(me->pdeath_signal, (int __user *)arg2);
2240 break;
2241 case PR_GET_DUMPABLE:
2242 error = get_dumpable(me->mm);
2243 break;
2244 case PR_SET_DUMPABLE:
2245 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2246 error = -EINVAL;
2247 break;
2248 }
2249 set_dumpable(me->mm, arg2);
2250 break;
2251
2252 case PR_SET_UNALIGN:
2253 error = SET_UNALIGN_CTL(me, arg2);
2254 break;
2255 case PR_GET_UNALIGN:
2256 error = GET_UNALIGN_CTL(me, arg2);
2257 break;
2258 case PR_SET_FPEMU:
2259 error = SET_FPEMU_CTL(me, arg2);
2260 break;
2261 case PR_GET_FPEMU:
2262 error = GET_FPEMU_CTL(me, arg2);
2263 break;
2264 case PR_SET_FPEXC:
2265 error = SET_FPEXC_CTL(me, arg2);
2266 break;
2267 case PR_GET_FPEXC:
2268 error = GET_FPEXC_CTL(me, arg2);
2269 break;
2270 case PR_GET_TIMING:
2271 error = PR_TIMING_STATISTICAL;
2272 break;
2273 case PR_SET_TIMING:
2274 if (arg2 != PR_TIMING_STATISTICAL)
2275 error = -EINVAL;
2276 break;
2277 case PR_SET_NAME:
2278 comm[sizeof(me->comm) - 1] = 0;
2279 if (strncpy_from_user(comm, (char __user *)arg2,
2280 sizeof(me->comm) - 1) < 0)
2281 return -EFAULT;
2282 set_task_comm(me, comm);
2283 proc_comm_connector(me);
2284 break;
2285 case PR_GET_NAME:
2286 get_task_comm(comm, me);
2287 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2288 return -EFAULT;
2289 break;
2290 case PR_GET_ENDIAN:
2291 error = GET_ENDIAN(me, arg2);
2292 break;
2293 case PR_SET_ENDIAN:
2294 error = SET_ENDIAN(me, arg2);
2295 break;
2296 case PR_GET_SECCOMP:
2297 error = prctl_get_seccomp();
2298 break;
2299 case PR_SET_SECCOMP:
2300 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2301 break;
2302 case PR_GET_TSC:
2303 error = GET_TSC_CTL(arg2);
2304 break;
2305 case PR_SET_TSC:
2306 error = SET_TSC_CTL(arg2);
2307 break;
2308 case PR_TASK_PERF_EVENTS_DISABLE:
2309 error = perf_event_task_disable();
2310 break;
2311 case PR_TASK_PERF_EVENTS_ENABLE:
2312 error = perf_event_task_enable();
2313 break;
2314 case PR_GET_TIMERSLACK:
2315 if (current->timer_slack_ns > ULONG_MAX)
2316 error = ULONG_MAX;
2317 else
2318 error = current->timer_slack_ns;
2319 break;
2320 case PR_SET_TIMERSLACK:
2321 if (arg2 <= 0)
2322 current->timer_slack_ns =
2323 current->default_timer_slack_ns;
2324 else
2325 current->timer_slack_ns = arg2;
2326 break;
2327 case PR_MCE_KILL:
2328 if (arg4 | arg5)
2329 return -EINVAL;
2330 switch (arg2) {
2331 case PR_MCE_KILL_CLEAR:
2332 if (arg3 != 0)
2333 return -EINVAL;
2334 current->flags &= ~PF_MCE_PROCESS;
2335 break;
2336 case PR_MCE_KILL_SET:
2337 current->flags |= PF_MCE_PROCESS;
2338 if (arg3 == PR_MCE_KILL_EARLY)
2339 current->flags |= PF_MCE_EARLY;
2340 else if (arg3 == PR_MCE_KILL_LATE)
2341 current->flags &= ~PF_MCE_EARLY;
2342 else if (arg3 == PR_MCE_KILL_DEFAULT)
2343 current->flags &=
2344 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2345 else
2346 return -EINVAL;
2347 break;
2348 default:
2349 return -EINVAL;
2350 }
2351 break;
2352 case PR_MCE_KILL_GET:
2353 if (arg2 | arg3 | arg4 | arg5)
2354 return -EINVAL;
2355 if (current->flags & PF_MCE_PROCESS)
2356 error = (current->flags & PF_MCE_EARLY) ?
2357 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2358 else
2359 error = PR_MCE_KILL_DEFAULT;
2360 break;
2361 case PR_SET_MM:
2362 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2363 break;
2364 case PR_GET_TID_ADDRESS:
2365 error = prctl_get_tid_address(me, (int __user **)arg2);
2366 break;
2367 case PR_SET_CHILD_SUBREAPER:
2368 me->signal->is_child_subreaper = !!arg2;
2369 if (!arg2)
2370 break;
2371
2372 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2373 break;
2374 case PR_GET_CHILD_SUBREAPER:
2375 error = put_user(me->signal->is_child_subreaper,
2376 (int __user *)arg2);
2377 break;
2378 case PR_SET_NO_NEW_PRIVS:
2379 if (arg2 != 1 || arg3 || arg4 || arg5)
2380 return -EINVAL;
2381
2382 task_set_no_new_privs(current);
2383 break;
2384 case PR_GET_NO_NEW_PRIVS:
2385 if (arg2 || arg3 || arg4 || arg5)
2386 return -EINVAL;
2387 return task_no_new_privs(current) ? 1 : 0;
2388 case PR_GET_THP_DISABLE:
2389 if (arg2 || arg3 || arg4 || arg5)
2390 return -EINVAL;
2391 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2392 break;
2393 case PR_SET_THP_DISABLE:
2394 if (arg3 || arg4 || arg5)
2395 return -EINVAL;
2396 if (down_write_killable(&me->mm->mmap_sem))
2397 return -EINTR;
2398 if (arg2)
2399 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2400 else
2401 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2402 up_write(&me->mm->mmap_sem);
2403 break;
2404 case PR_MPX_ENABLE_MANAGEMENT:
2405 if (arg2 || arg3 || arg4 || arg5)
2406 return -EINVAL;
2407 error = MPX_ENABLE_MANAGEMENT();
2408 break;
2409 case PR_MPX_DISABLE_MANAGEMENT:
2410 if (arg2 || arg3 || arg4 || arg5)
2411 return -EINVAL;
2412 error = MPX_DISABLE_MANAGEMENT();
2413 break;
2414 case PR_SET_FP_MODE:
2415 error = SET_FP_MODE(me, arg2);
2416 break;
2417 case PR_GET_FP_MODE:
2418 error = GET_FP_MODE(me);
2419 break;
2420 case PR_GET_SPECULATION_CTRL:
2421 if (arg3 || arg4 || arg5)
2422 return -EINVAL;
2423 error = arch_prctl_spec_ctrl_get(me, arg2);
2424 break;
2425 case PR_SET_SPECULATION_CTRL:
2426 if (arg4 || arg5)
2427 return -EINVAL;
2428 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2429 break;
2430 default:
2431 error = -EINVAL;
2432 break;
2433 }
2434 return error;
2435 }
2436
2437 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2438 struct getcpu_cache __user *, unused)
2439 {
2440 int err = 0;
2441 int cpu = raw_smp_processor_id();
2442
2443 if (cpup)
2444 err |= put_user(cpu, cpup);
2445 if (nodep)
2446 err |= put_user(cpu_to_node(cpu), nodep);
2447 return err ? -EFAULT : 0;
2448 }
2449
2450 /**
2451 * do_sysinfo - fill in sysinfo struct
2452 * @info: pointer to buffer to fill
2453 */
2454 static int do_sysinfo(struct sysinfo *info)
2455 {
2456 unsigned long mem_total, sav_total;
2457 unsigned int mem_unit, bitcount;
2458 struct timespec tp;
2459
2460 memset(info, 0, sizeof(struct sysinfo));
2461
2462 get_monotonic_boottime(&tp);
2463 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2464
2465 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2466
2467 info->procs = nr_threads;
2468
2469 si_meminfo(info);
2470 si_swapinfo(info);
2471
2472 /*
2473 * If the sum of all the available memory (i.e. ram + swap)
2474 * is less than can be stored in a 32 bit unsigned long then
2475 * we can be binary compatible with 2.2.x kernels. If not,
2476 * well, in that case 2.2.x was broken anyways...
2477 *
2478 * -Erik Andersen <andersee@debian.org>
2479 */
2480
2481 mem_total = info->totalram + info->totalswap;
2482 if (mem_total < info->totalram || mem_total < info->totalswap)
2483 goto out;
2484 bitcount = 0;
2485 mem_unit = info->mem_unit;
2486 while (mem_unit > 1) {
2487 bitcount++;
2488 mem_unit >>= 1;
2489 sav_total = mem_total;
2490 mem_total <<= 1;
2491 if (mem_total < sav_total)
2492 goto out;
2493 }
2494
2495 /*
2496 * If mem_total did not overflow, multiply all memory values by
2497 * info->mem_unit and set it to 1. This leaves things compatible
2498 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2499 * kernels...
2500 */
2501
2502 info->mem_unit = 1;
2503 info->totalram <<= bitcount;
2504 info->freeram <<= bitcount;
2505 info->sharedram <<= bitcount;
2506 info->bufferram <<= bitcount;
2507 info->totalswap <<= bitcount;
2508 info->freeswap <<= bitcount;
2509 info->totalhigh <<= bitcount;
2510 info->freehigh <<= bitcount;
2511
2512 out:
2513 return 0;
2514 }
2515
2516 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2517 {
2518 struct sysinfo val;
2519
2520 do_sysinfo(&val);
2521
2522 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2523 return -EFAULT;
2524
2525 return 0;
2526 }
2527
2528 #ifdef CONFIG_COMPAT
2529 struct compat_sysinfo {
2530 s32 uptime;
2531 u32 loads[3];
2532 u32 totalram;
2533 u32 freeram;
2534 u32 sharedram;
2535 u32 bufferram;
2536 u32 totalswap;
2537 u32 freeswap;
2538 u16 procs;
2539 u16 pad;
2540 u32 totalhigh;
2541 u32 freehigh;
2542 u32 mem_unit;
2543 char _f[20-2*sizeof(u32)-sizeof(int)];
2544 };
2545
2546 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2547 {
2548 struct sysinfo s;
2549
2550 do_sysinfo(&s);
2551
2552 /* Check to see if any memory value is too large for 32-bit and scale
2553 * down if needed
2554 */
2555 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2556 int bitcount = 0;
2557
2558 while (s.mem_unit < PAGE_SIZE) {
2559 s.mem_unit <<= 1;
2560 bitcount++;
2561 }
2562
2563 s.totalram >>= bitcount;
2564 s.freeram >>= bitcount;
2565 s.sharedram >>= bitcount;
2566 s.bufferram >>= bitcount;
2567 s.totalswap >>= bitcount;
2568 s.freeswap >>= bitcount;
2569 s.totalhigh >>= bitcount;
2570 s.freehigh >>= bitcount;
2571 }
2572
2573 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2574 __put_user(s.uptime, &info->uptime) ||
2575 __put_user(s.loads[0], &info->loads[0]) ||
2576 __put_user(s.loads[1], &info->loads[1]) ||
2577 __put_user(s.loads[2], &info->loads[2]) ||
2578 __put_user(s.totalram, &info->totalram) ||
2579 __put_user(s.freeram, &info->freeram) ||
2580 __put_user(s.sharedram, &info->sharedram) ||
2581 __put_user(s.bufferram, &info->bufferram) ||
2582 __put_user(s.totalswap, &info->totalswap) ||
2583 __put_user(s.freeswap, &info->freeswap) ||
2584 __put_user(s.procs, &info->procs) ||
2585 __put_user(s.totalhigh, &info->totalhigh) ||
2586 __put_user(s.freehigh, &info->freehigh) ||
2587 __put_user(s.mem_unit, &info->mem_unit))
2588 return -EFAULT;
2589
2590 return 0;
2591 }
2592 #endif /* CONFIG_COMPAT */
Ubuntu Artful kernel mirror