1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/kernel/sys.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8 #include <linux/export.h>
9 #include <linux/mm.h>
10 #include <linux/mm_inline.h>
11 #include <linux/utsname.h>
12 #include <linux/mman.h>
13 #include <linux/reboot.h>
14 #include <linux/prctl.h>
15 #include <linux/highuid.h>
16 #include <linux/fs.h>
17 #include <linux/kmod.h>
18 #include <linux/ksm.h>
19 #include <linux/perf_event.h>
20 #include <linux/resource.h>
21 #include <linux/kernel.h>
22 #include <linux/workqueue.h>
23 #include <linux/capability.h>
24 #include <linux/device.h>
25 #include <linux/key.h>
26 #include <linux/times.h>
27 #include <linux/posix-timers.h>
28 #include <linux/security.h>
29 #include <linux/random.h>
30 #include <linux/suspend.h>
31 #include <linux/tty.h>
32 #include <linux/signal.h>
33 #include <linux/cn_proc.h>
34 #include <linux/getcpu.h>
35 #include <linux/task_io_accounting_ops.h>
36 #include <linux/seccomp.h>
37 #include <linux/cpu.h>
38 #include <linux/personality.h>
39 #include <linux/ptrace.h>
40 #include <linux/fs_struct.h>
41 #include <linux/file.h>
42 #include <linux/mount.h>
43 #include <linux/gfp.h>
44 #include <linux/syscore_ops.h>
45 #include <linux/version.h>
46 #include <linux/ctype.h>
47 #include <linux/syscall_user_dispatch.h>
48
49 #include <linux/compat.h>
50 #include <linux/syscalls.h>
51 #include <linux/kprobes.h>
52 #include <linux/user_namespace.h>
53 #include <linux/time_namespace.h>
54 #include <linux/binfmts.h>
55
56 #include <linux/sched.h>
57 #include <linux/sched/autogroup.h>
58 #include <linux/sched/loadavg.h>
59 #include <linux/sched/stat.h>
60 #include <linux/sched/mm.h>
61 #include <linux/sched/coredump.h>
62 #include <linux/sched/task.h>
63 #include <linux/sched/cputime.h>
64 #include <linux/rcupdate.h>
65 #include <linux/uidgid.h>
66 #include <linux/cred.h>
67
68 #include <linux/nospec.h>
69
70 #include <linux/kmsg_dump.h>
71 /* Move somewhere else to avoid recompiling? */
72 #include <generated/utsrelease.h>
73
74 #include <linux/uaccess.h>
75 #include <asm/io.h>
76 #include <asm/unistd.h>
77
78 #include <trace/events/task.h>
79
80 #include "uid16.h"
81
82 #ifndef SET_UNALIGN_CTL
83 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
84 #endif
85 #ifndef GET_UNALIGN_CTL
86 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
87 #endif
88 #ifndef SET_FPEMU_CTL
89 # define SET_FPEMU_CTL(a, b) (-EINVAL)
90 #endif
91 #ifndef GET_FPEMU_CTL
92 # define GET_FPEMU_CTL(a, b) (-EINVAL)
93 #endif
94 #ifndef SET_FPEXC_CTL
95 # define SET_FPEXC_CTL(a, b) (-EINVAL)
96 #endif
97 #ifndef GET_FPEXC_CTL
98 # define GET_FPEXC_CTL(a, b) (-EINVAL)
99 #endif
100 #ifndef GET_ENDIAN
101 # define GET_ENDIAN(a, b) (-EINVAL)
102 #endif
103 #ifndef SET_ENDIAN
104 # define SET_ENDIAN(a, b) (-EINVAL)
105 #endif
106 #ifndef GET_TSC_CTL
107 # define GET_TSC_CTL(a) (-EINVAL)
108 #endif
109 #ifndef SET_TSC_CTL
110 # define SET_TSC_CTL(a) (-EINVAL)
111 #endif
112 #ifndef GET_FP_MODE
113 # define GET_FP_MODE(a) (-EINVAL)
114 #endif
115 #ifndef SET_FP_MODE
116 # define SET_FP_MODE(a,b) (-EINVAL)
117 #endif
118 #ifndef SVE_SET_VL
119 # define SVE_SET_VL(a) (-EINVAL)
120 #endif
121 #ifndef SVE_GET_VL
122 # define SVE_GET_VL() (-EINVAL)
123 #endif
124 #ifndef SME_SET_VL
125 # define SME_SET_VL(a) (-EINVAL)
126 #endif
127 #ifndef SME_GET_VL
128 # define SME_GET_VL() (-EINVAL)
129 #endif
130 #ifndef PAC_RESET_KEYS
131 # define PAC_RESET_KEYS(a, b) (-EINVAL)
132 #endif
133 #ifndef PAC_SET_ENABLED_KEYS
134 # define PAC_SET_ENABLED_KEYS(a, b, c) (-EINVAL)
135 #endif
136 #ifndef PAC_GET_ENABLED_KEYS
137 # define PAC_GET_ENABLED_KEYS(a) (-EINVAL)
138 #endif
139 #ifndef SET_TAGGED_ADDR_CTRL
140 # define SET_TAGGED_ADDR_CTRL(a) (-EINVAL)
141 #endif
142 #ifndef GET_TAGGED_ADDR_CTRL
143 # define GET_TAGGED_ADDR_CTRL() (-EINVAL)
144 #endif
145 #ifndef RISCV_V_SET_CONTROL
146 # define RISCV_V_SET_CONTROL(a) (-EINVAL)
147 #endif
148 #ifndef RISCV_V_GET_CONTROL
149 # define RISCV_V_GET_CONTROL() (-EINVAL)
150 #endif
151 #ifndef RISCV_SET_ICACHE_FLUSH_CTX
152 # define RISCV_SET_ICACHE_FLUSH_CTX(a, b) (-EINVAL)
153 #endif
154 #ifndef PPC_GET_DEXCR_ASPECT
155 # define PPC_GET_DEXCR_ASPECT(a, b) (-EINVAL)
156 #endif
157 #ifndef PPC_SET_DEXCR_ASPECT
158 # define PPC_SET_DEXCR_ASPECT(a, b, c) (-EINVAL)
159 #endif
160
161 /*
162 * this is where the system-wide overflow UID and GID are defined, for
163 * architectures that now have 32-bit UID/GID but didn't in the past
164 */
165
166 int overflowuid = DEFAULT_OVERFLOWUID;
167 int overflowgid = DEFAULT_OVERFLOWGID;
168
169 EXPORT_SYMBOL(overflowuid);
170 EXPORT_SYMBOL(overflowgid);
171
172 /*
173 * the same as above, but for filesystems which can only store a 16-bit
174 * UID and GID. as such, this is needed on all architectures
175 */
176
177 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
178 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
179
180 EXPORT_SYMBOL(fs_overflowuid);
181 EXPORT_SYMBOL(fs_overflowgid);
182
183 /*
184 * Returns true if current's euid is same as p's uid or euid,
185 * or has CAP_SYS_NICE to p's user_ns.
186 *
187 * Called with rcu_read_lock, creds are safe
188 */
set_one_prio_perm(struct task_struct * p)189 static bool set_one_prio_perm(struct task_struct *p)
190 {
191 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
192
193 if (uid_eq(pcred->uid, cred->euid) ||
194 uid_eq(pcred->euid, cred->euid))
195 return true;
196 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
197 return true;
198 return false;
199 }
200
201 /*
202 * set the priority of a task
203 * - the caller must hold the RCU read lock
204 */
set_one_prio(struct task_struct * p,int niceval,int error)205 static int set_one_prio(struct task_struct *p, int niceval, int error)
206 {
207 int no_nice;
208
209 if (!set_one_prio_perm(p)) {
210 error = -EPERM;
211 goto out;
212 }
213 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
214 error = -EACCES;
215 goto out;
216 }
217 no_nice = security_task_setnice(p, niceval);
218 if (no_nice) {
219 error = no_nice;
220 goto out;
221 }
222 if (error == -ESRCH)
223 error = 0;
224 set_user_nice(p, niceval);
225 out:
226 return error;
227 }
228
SYSCALL_DEFINE3(setpriority,int,which,int,who,int,niceval)229 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
230 {
231 struct task_struct *g, *p;
232 struct user_struct *user;
233 const struct cred *cred = current_cred();
234 int error = -EINVAL;
235 struct pid *pgrp;
236 kuid_t uid;
237
238 if (which > PRIO_USER || which < PRIO_PROCESS)
239 goto out;
240
241 /* normalize: avoid signed division (rounding problems) */
242 error = -ESRCH;
243 if (niceval < MIN_NICE)
244 niceval = MIN_NICE;
245 if (niceval > MAX_NICE)
246 niceval = MAX_NICE;
247
248 rcu_read_lock();
249 switch (which) {
250 case PRIO_PROCESS:
251 if (who)
252 p = find_task_by_vpid(who);
253 else
254 p = current;
255 if (p)
256 error = set_one_prio(p, niceval, error);
257 break;
258 case PRIO_PGRP:
259 if (who)
260 pgrp = find_vpid(who);
261 else
262 pgrp = task_pgrp(current);
263 read_lock(&tasklist_lock);
264 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
265 error = set_one_prio(p, niceval, error);
266 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
267 read_unlock(&tasklist_lock);
268 break;
269 case PRIO_USER:
270 uid = make_kuid(cred->user_ns, who);
271 user = cred->user;
272 if (!who)
273 uid = cred->uid;
274 else if (!uid_eq(uid, cred->uid)) {
275 user = find_user(uid);
276 if (!user)
277 goto out_unlock; /* No processes for this user */
278 }
279 for_each_process_thread(g, p) {
280 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
281 error = set_one_prio(p, niceval, error);
282 }
283 if (!uid_eq(uid, cred->uid))
284 free_uid(user); /* For find_user() */
285 break;
286 }
287 out_unlock:
288 rcu_read_unlock();
289 out:
290 return error;
291 }
292
293 /*
294 * Ugh. To avoid negative return values, "getpriority()" will
295 * not return the normal nice-value, but a negated value that
296 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
297 * to stay compatible.
298 */
SYSCALL_DEFINE2(getpriority,int,which,int,who)299 SYSCALL_DEFINE2(getpriority, int, which, int, who)
300 {
301 struct task_struct *g, *p;
302 struct user_struct *user;
303 const struct cred *cred = current_cred();
304 long niceval, retval = -ESRCH;
305 struct pid *pgrp;
306 kuid_t uid;
307
308 if (which > PRIO_USER || which < PRIO_PROCESS)
309 return -EINVAL;
310
311 rcu_read_lock();
312 switch (which) {
313 case PRIO_PROCESS:
314 if (who)
315 p = find_task_by_vpid(who);
316 else
317 p = current;
318 if (p) {
319 niceval = nice_to_rlimit(task_nice(p));
320 if (niceval > retval)
321 retval = niceval;
322 }
323 break;
324 case PRIO_PGRP:
325 if (who)
326 pgrp = find_vpid(who);
327 else
328 pgrp = task_pgrp(current);
329 read_lock(&tasklist_lock);
330 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
331 niceval = nice_to_rlimit(task_nice(p));
332 if (niceval > retval)
333 retval = niceval;
334 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
335 read_unlock(&tasklist_lock);
336 break;
337 case PRIO_USER:
338 uid = make_kuid(cred->user_ns, who);
339 user = cred->user;
340 if (!who)
341 uid = cred->uid;
342 else if (!uid_eq(uid, cred->uid)) {
343 user = find_user(uid);
344 if (!user)
345 goto out_unlock; /* No processes for this user */
346 }
347 for_each_process_thread(g, p) {
348 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
349 niceval = nice_to_rlimit(task_nice(p));
350 if (niceval > retval)
351 retval = niceval;
352 }
353 }
354 if (!uid_eq(uid, cred->uid))
355 free_uid(user); /* for find_user() */
356 break;
357 }
358 out_unlock:
359 rcu_read_unlock();
360
361 return retval;
362 }
363
364 /*
365 * Unprivileged users may change the real gid to the effective gid
366 * or vice versa. (BSD-style)
367 *
368 * If you set the real gid at all, or set the effective gid to a value not
369 * equal to the real gid, then the saved gid is set to the new effective gid.
370 *
371 * This makes it possible for a setgid program to completely drop its
372 * privileges, which is often a useful assertion to make when you are doing
373 * a security audit over a program.
374 *
375 * The general idea is that a program which uses just setregid() will be
376 * 100% compatible with BSD. A program which uses just setgid() will be
377 * 100% compatible with POSIX with saved IDs.
378 *
379 * SMP: There are not races, the GIDs are checked only by filesystem
380 * operations (as far as semantic preservation is concerned).
381 */
382 #ifdef CONFIG_MULTIUSER
__sys_setregid(gid_t rgid,gid_t egid)383 long __sys_setregid(gid_t rgid, gid_t egid)
384 {
385 struct user_namespace *ns = current_user_ns();
386 const struct cred *old;
387 struct cred *new;
388 int retval;
389 kgid_t krgid, kegid;
390
391 krgid = make_kgid(ns, rgid);
392 kegid = make_kgid(ns, egid);
393
394 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
395 return -EINVAL;
396 if ((egid != (gid_t) -1) && !gid_valid(kegid))
397 return -EINVAL;
398
399 new = prepare_creds();
400 if (!new)
401 return -ENOMEM;
402 old = current_cred();
403
404 retval = -EPERM;
405 if (rgid != (gid_t) -1) {
406 if (gid_eq(old->gid, krgid) ||
407 gid_eq(old->egid, krgid) ||
408 ns_capable_setid(old->user_ns, CAP_SETGID))
409 new->gid = krgid;
410 else
411 goto error;
412 }
413 if (egid != (gid_t) -1) {
414 if (gid_eq(old->gid, kegid) ||
415 gid_eq(old->egid, kegid) ||
416 gid_eq(old->sgid, kegid) ||
417 ns_capable_setid(old->user_ns, CAP_SETGID))
418 new->egid = kegid;
419 else
420 goto error;
421 }
422
423 if (rgid != (gid_t) -1 ||
424 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
425 new->sgid = new->egid;
426 new->fsgid = new->egid;
427
428 retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
429 if (retval < 0)
430 goto error;
431
432 return commit_creds(new);
433
434 error:
435 abort_creds(new);
436 return retval;
437 }
438
SYSCALL_DEFINE2(setregid,gid_t,rgid,gid_t,egid)439 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
440 {
441 return __sys_setregid(rgid, egid);
442 }
443
444 /*
445 * setgid() is implemented like SysV w/ SAVED_IDS
446 *
447 * SMP: Same implicit races as above.
448 */
__sys_setgid(gid_t gid)449 long __sys_setgid(gid_t gid)
450 {
451 struct user_namespace *ns = current_user_ns();
452 const struct cred *old;
453 struct cred *new;
454 int retval;
455 kgid_t kgid;
456
457 kgid = make_kgid(ns, gid);
458 if (!gid_valid(kgid))
459 return -EINVAL;
460
461 new = prepare_creds();
462 if (!new)
463 return -ENOMEM;
464 old = current_cred();
465
466 retval = -EPERM;
467 if (ns_capable_setid(old->user_ns, CAP_SETGID))
468 new->gid = new->egid = new->sgid = new->fsgid = kgid;
469 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
470 new->egid = new->fsgid = kgid;
471 else
472 goto error;
473
474 retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
475 if (retval < 0)
476 goto error;
477
478 return commit_creds(new);
479
480 error:
481 abort_creds(new);
482 return retval;
483 }
484
SYSCALL_DEFINE1(setgid,gid_t,gid)485 SYSCALL_DEFINE1(setgid, gid_t, gid)
486 {
487 return __sys_setgid(gid);
488 }
489
490 /*
491 * change the user struct in a credentials set to match the new UID
492 */
set_user(struct cred * new)493 static int set_user(struct cred *new)
494 {
495 struct user_struct *new_user;
496
497 new_user = alloc_uid(new->uid);
498 if (!new_user)
499 return -EAGAIN;
500
501 free_uid(new->user);
502 new->user = new_user;
503 return 0;
504 }
505
flag_nproc_exceeded(struct cred * new)506 static void flag_nproc_exceeded(struct cred *new)
507 {
508 if (new->ucounts == current_ucounts())
509 return;
510
511 /*
512 * We don't fail in case of NPROC limit excess here because too many
513 * poorly written programs don't check set*uid() return code, assuming
514 * it never fails if called by root. We may still enforce NPROC limit
515 * for programs doing set*uid()+execve() by harmlessly deferring the
516 * failure to the execve() stage.
517 */
518 if (is_rlimit_overlimit(new->ucounts, UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)) &&
519 new->user != INIT_USER)
520 current->flags |= PF_NPROC_EXCEEDED;
521 else
522 current->flags &= ~PF_NPROC_EXCEEDED;
523 }
524
525 /*
526 * Unprivileged users may change the real uid to the effective uid
527 * or vice versa. (BSD-style)
528 *
529 * If you set the real uid at all, or set the effective uid to a value not
530 * equal to the real uid, then the saved uid is set to the new effective uid.
531 *
532 * This makes it possible for a setuid program to completely drop its
533 * privileges, which is often a useful assertion to make when you are doing
534 * a security audit over a program.
535 *
536 * The general idea is that a program which uses just setreuid() will be
537 * 100% compatible with BSD. A program which uses just setuid() will be
538 * 100% compatible with POSIX with saved IDs.
539 */
__sys_setreuid(uid_t ruid,uid_t euid)540 long __sys_setreuid(uid_t ruid, uid_t euid)
541 {
542 struct user_namespace *ns = current_user_ns();
543 const struct cred *old;
544 struct cred *new;
545 int retval;
546 kuid_t kruid, keuid;
547
548 kruid = make_kuid(ns, ruid);
549 keuid = make_kuid(ns, euid);
550
551 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
552 return -EINVAL;
553 if ((euid != (uid_t) -1) && !uid_valid(keuid))
554 return -EINVAL;
555
556 new = prepare_creds();
557 if (!new)
558 return -ENOMEM;
559 old = current_cred();
560
561 retval = -EPERM;
562 if (ruid != (uid_t) -1) {
563 new->uid = kruid;
564 if (!uid_eq(old->uid, kruid) &&
565 !uid_eq(old->euid, kruid) &&
566 !ns_capable_setid(old->user_ns, CAP_SETUID))
567 goto error;
568 }
569
570 if (euid != (uid_t) -1) {
571 new->euid = keuid;
572 if (!uid_eq(old->uid, keuid) &&
573 !uid_eq(old->euid, keuid) &&
574 !uid_eq(old->suid, keuid) &&
575 !ns_capable_setid(old->user_ns, CAP_SETUID))
576 goto error;
577 }
578
579 if (!uid_eq(new->uid, old->uid)) {
580 retval = set_user(new);
581 if (retval < 0)
582 goto error;
583 }
584 if (ruid != (uid_t) -1 ||
585 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
586 new->suid = new->euid;
587 new->fsuid = new->euid;
588
589 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
590 if (retval < 0)
591 goto error;
592
593 retval = set_cred_ucounts(new);
594 if (retval < 0)
595 goto error;
596
597 flag_nproc_exceeded(new);
598 return commit_creds(new);
599
600 error:
601 abort_creds(new);
602 return retval;
603 }
604
SYSCALL_DEFINE2(setreuid,uid_t,ruid,uid_t,euid)605 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
606 {
607 return __sys_setreuid(ruid, euid);
608 }
609
610 /*
611 * setuid() is implemented like SysV with SAVED_IDS
612 *
613 * Note that SAVED_ID's is deficient in that a setuid root program
614 * like sendmail, for example, cannot set its uid to be a normal
615 * user and then switch back, because if you're root, setuid() sets
616 * the saved uid too. If you don't like this, blame the bright people
617 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
618 * will allow a root program to temporarily drop privileges and be able to
619 * regain them by swapping the real and effective uid.
620 */
__sys_setuid(uid_t uid)621 long __sys_setuid(uid_t uid)
622 {
623 struct user_namespace *ns = current_user_ns();
624 const struct cred *old;
625 struct cred *new;
626 int retval;
627 kuid_t kuid;
628
629 kuid = make_kuid(ns, uid);
630 if (!uid_valid(kuid))
631 return -EINVAL;
632
633 new = prepare_creds();
634 if (!new)
635 return -ENOMEM;
636 old = current_cred();
637
638 retval = -EPERM;
639 if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
640 new->suid = new->uid = kuid;
641 if (!uid_eq(kuid, old->uid)) {
642 retval = set_user(new);
643 if (retval < 0)
644 goto error;
645 }
646 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
647 goto error;
648 }
649
650 new->fsuid = new->euid = kuid;
651
652 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
653 if (retval < 0)
654 goto error;
655
656 retval = set_cred_ucounts(new);
657 if (retval < 0)
658 goto error;
659
660 flag_nproc_exceeded(new);
661 return commit_creds(new);
662
663 error:
664 abort_creds(new);
665 return retval;
666 }
667
SYSCALL_DEFINE1(setuid,uid_t,uid)668 SYSCALL_DEFINE1(setuid, uid_t, uid)
669 {
670 return __sys_setuid(uid);
671 }
672
673
674 /*
675 * This function implements a generic ability to update ruid, euid,
676 * and suid. This allows you to implement the 4.4 compatible seteuid().
677 */
__sys_setresuid(uid_t ruid,uid_t euid,uid_t suid)678 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
679 {
680 struct user_namespace *ns = current_user_ns();
681 const struct cred *old;
682 struct cred *new;
683 int retval;
684 kuid_t kruid, keuid, ksuid;
685 bool ruid_new, euid_new, suid_new;
686
687 kruid = make_kuid(ns, ruid);
688 keuid = make_kuid(ns, euid);
689 ksuid = make_kuid(ns, suid);
690
691 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
692 return -EINVAL;
693
694 if ((euid != (uid_t) -1) && !uid_valid(keuid))
695 return -EINVAL;
696
697 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
698 return -EINVAL;
699
700 old = current_cred();
701
702 /* check for no-op */
703 if ((ruid == (uid_t) -1 || uid_eq(kruid, old->uid)) &&
704 (euid == (uid_t) -1 || (uid_eq(keuid, old->euid) &&
705 uid_eq(keuid, old->fsuid))) &&
706 (suid == (uid_t) -1 || uid_eq(ksuid, old->suid)))
707 return 0;
708
709 ruid_new = ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
710 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid);
711 euid_new = euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
712 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid);
713 suid_new = suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
714 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid);
715 if ((ruid_new || euid_new || suid_new) &&
716 !ns_capable_setid(old->user_ns, CAP_SETUID))
717 return -EPERM;
718
719 new = prepare_creds();
720 if (!new)
721 return -ENOMEM;
722
723 if (ruid != (uid_t) -1) {
724 new->uid = kruid;
725 if (!uid_eq(kruid, old->uid)) {
726 retval = set_user(new);
727 if (retval < 0)
728 goto error;
729 }
730 }
731 if (euid != (uid_t) -1)
732 new->euid = keuid;
733 if (suid != (uid_t) -1)
734 new->suid = ksuid;
735 new->fsuid = new->euid;
736
737 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
738 if (retval < 0)
739 goto error;
740
741 retval = set_cred_ucounts(new);
742 if (retval < 0)
743 goto error;
744
745 flag_nproc_exceeded(new);
746 return commit_creds(new);
747
748 error:
749 abort_creds(new);
750 return retval;
751 }
752
SYSCALL_DEFINE3(setresuid,uid_t,ruid,uid_t,euid,uid_t,suid)753 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
754 {
755 return __sys_setresuid(ruid, euid, suid);
756 }
757
SYSCALL_DEFINE3(getresuid,uid_t __user *,ruidp,uid_t __user *,euidp,uid_t __user *,suidp)758 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
759 {
760 const struct cred *cred = current_cred();
761 int retval;
762 uid_t ruid, euid, suid;
763
764 ruid = from_kuid_munged(cred->user_ns, cred->uid);
765 euid = from_kuid_munged(cred->user_ns, cred->euid);
766 suid = from_kuid_munged(cred->user_ns, cred->suid);
767
768 retval = put_user(ruid, ruidp);
769 if (!retval) {
770 retval = put_user(euid, euidp);
771 if (!retval)
772 return put_user(suid, suidp);
773 }
774 return retval;
775 }
776
777 /*
778 * Same as above, but for rgid, egid, sgid.
779 */
__sys_setresgid(gid_t rgid,gid_t egid,gid_t sgid)780 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
781 {
782 struct user_namespace *ns = current_user_ns();
783 const struct cred *old;
784 struct cred *new;
785 int retval;
786 kgid_t krgid, kegid, ksgid;
787 bool rgid_new, egid_new, sgid_new;
788
789 krgid = make_kgid(ns, rgid);
790 kegid = make_kgid(ns, egid);
791 ksgid = make_kgid(ns, sgid);
792
793 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
794 return -EINVAL;
795 if ((egid != (gid_t) -1) && !gid_valid(kegid))
796 return -EINVAL;
797 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
798 return -EINVAL;
799
800 old = current_cred();
801
802 /* check for no-op */
803 if ((rgid == (gid_t) -1 || gid_eq(krgid, old->gid)) &&
804 (egid == (gid_t) -1 || (gid_eq(kegid, old->egid) &&
805 gid_eq(kegid, old->fsgid))) &&
806 (sgid == (gid_t) -1 || gid_eq(ksgid, old->sgid)))
807 return 0;
808
809 rgid_new = rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
810 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid);
811 egid_new = egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
812 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid);
813 sgid_new = sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
814 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid);
815 if ((rgid_new || egid_new || sgid_new) &&
816 !ns_capable_setid(old->user_ns, CAP_SETGID))
817 return -EPERM;
818
819 new = prepare_creds();
820 if (!new)
821 return -ENOMEM;
822
823 if (rgid != (gid_t) -1)
824 new->gid = krgid;
825 if (egid != (gid_t) -1)
826 new->egid = kegid;
827 if (sgid != (gid_t) -1)
828 new->sgid = ksgid;
829 new->fsgid = new->egid;
830
831 retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
832 if (retval < 0)
833 goto error;
834
835 return commit_creds(new);
836
837 error:
838 abort_creds(new);
839 return retval;
840 }
841
SYSCALL_DEFINE3(setresgid,gid_t,rgid,gid_t,egid,gid_t,sgid)842 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
843 {
844 return __sys_setresgid(rgid, egid, sgid);
845 }
846
SYSCALL_DEFINE3(getresgid,gid_t __user *,rgidp,gid_t __user *,egidp,gid_t __user *,sgidp)847 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
848 {
849 const struct cred *cred = current_cred();
850 int retval;
851 gid_t rgid, egid, sgid;
852
853 rgid = from_kgid_munged(cred->user_ns, cred->gid);
854 egid = from_kgid_munged(cred->user_ns, cred->egid);
855 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
856
857 retval = put_user(rgid, rgidp);
858 if (!retval) {
859 retval = put_user(egid, egidp);
860 if (!retval)
861 retval = put_user(sgid, sgidp);
862 }
863
864 return retval;
865 }
866
867
868 /*
869 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
870 * is used for "access()" and for the NFS daemon (letting nfsd stay at
871 * whatever uid it wants to). It normally shadows "euid", except when
872 * explicitly set by setfsuid() or for access..
873 */
__sys_setfsuid(uid_t uid)874 long __sys_setfsuid(uid_t uid)
875 {
876 const struct cred *old;
877 struct cred *new;
878 uid_t old_fsuid;
879 kuid_t kuid;
880
881 old = current_cred();
882 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
883
884 kuid = make_kuid(old->user_ns, uid);
885 if (!uid_valid(kuid))
886 return old_fsuid;
887
888 new = prepare_creds();
889 if (!new)
890 return old_fsuid;
891
892 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
893 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
894 ns_capable_setid(old->user_ns, CAP_SETUID)) {
895 if (!uid_eq(kuid, old->fsuid)) {
896 new->fsuid = kuid;
897 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
898 goto change_okay;
899 }
900 }
901
902 abort_creds(new);
903 return old_fsuid;
904
905 change_okay:
906 commit_creds(new);
907 return old_fsuid;
908 }
909
SYSCALL_DEFINE1(setfsuid,uid_t,uid)910 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
911 {
912 return __sys_setfsuid(uid);
913 }
914
915 /*
916 * Samma på svenska..
917 */
__sys_setfsgid(gid_t gid)918 long __sys_setfsgid(gid_t gid)
919 {
920 const struct cred *old;
921 struct cred *new;
922 gid_t old_fsgid;
923 kgid_t kgid;
924
925 old = current_cred();
926 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
927
928 kgid = make_kgid(old->user_ns, gid);
929 if (!gid_valid(kgid))
930 return old_fsgid;
931
932 new = prepare_creds();
933 if (!new)
934 return old_fsgid;
935
936 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
937 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
938 ns_capable_setid(old->user_ns, CAP_SETGID)) {
939 if (!gid_eq(kgid, old->fsgid)) {
940 new->fsgid = kgid;
941 if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
942 goto change_okay;
943 }
944 }
945
946 abort_creds(new);
947 return old_fsgid;
948
949 change_okay:
950 commit_creds(new);
951 return old_fsgid;
952 }
953
SYSCALL_DEFINE1(setfsgid,gid_t,gid)954 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
955 {
956 return __sys_setfsgid(gid);
957 }
958 #endif /* CONFIG_MULTIUSER */
959
960 /**
961 * sys_getpid - return the thread group id of the current process
962 *
963 * Note, despite the name, this returns the tgid not the pid. The tgid and
964 * the pid are identical unless CLONE_THREAD was specified on clone() in
965 * which case the tgid is the same in all threads of the same group.
966 *
967 * This is SMP safe as current->tgid does not change.
968 */
SYSCALL_DEFINE0(getpid)969 SYSCALL_DEFINE0(getpid)
970 {
971 return task_tgid_vnr(current);
972 }
973
974 /* Thread ID - the internal kernel "pid" */
SYSCALL_DEFINE0(gettid)975 SYSCALL_DEFINE0(gettid)
976 {
977 return task_pid_vnr(current);
978 }
979
980 /*
981 * Accessing ->real_parent is not SMP-safe, it could
982 * change from under us. However, we can use a stale
983 * value of ->real_parent under rcu_read_lock(), see
984 * release_task()->call_rcu(delayed_put_task_struct).
985 */
SYSCALL_DEFINE0(getppid)986 SYSCALL_DEFINE0(getppid)
987 {
988 int pid;
989
990 rcu_read_lock();
991 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
992 rcu_read_unlock();
993
994 return pid;
995 }
996
SYSCALL_DEFINE0(getuid)997 SYSCALL_DEFINE0(getuid)
998 {
999 /* Only we change this so SMP safe */
1000 return from_kuid_munged(current_user_ns(), current_uid());
1001 }
1002
SYSCALL_DEFINE0(geteuid)1003 SYSCALL_DEFINE0(geteuid)
1004 {
1005 /* Only we change this so SMP safe */
1006 return from_kuid_munged(current_user_ns(), current_euid());
1007 }
1008
SYSCALL_DEFINE0(getgid)1009 SYSCALL_DEFINE0(getgid)
1010 {
1011 /* Only we change this so SMP safe */
1012 return from_kgid_munged(current_user_ns(), current_gid());
1013 }
1014
SYSCALL_DEFINE0(getegid)1015 SYSCALL_DEFINE0(getegid)
1016 {
1017 /* Only we change this so SMP safe */
1018 return from_kgid_munged(current_user_ns(), current_egid());
1019 }
1020
do_sys_times(struct tms * tms)1021 static void do_sys_times(struct tms *tms)
1022 {
1023 u64 tgutime, tgstime, cutime, cstime;
1024
1025 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1026 cutime = current->signal->cutime;
1027 cstime = current->signal->cstime;
1028 tms->tms_utime = nsec_to_clock_t(tgutime);
1029 tms->tms_stime = nsec_to_clock_t(tgstime);
1030 tms->tms_cutime = nsec_to_clock_t(cutime);
1031 tms->tms_cstime = nsec_to_clock_t(cstime);
1032 }
1033
SYSCALL_DEFINE1(times,struct tms __user *,tbuf)1034 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1035 {
1036 if (tbuf) {
1037 struct tms tmp;
1038
1039 do_sys_times(&tmp);
1040 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1041 return -EFAULT;
1042 }
1043 force_successful_syscall_return();
1044 return (long) jiffies_64_to_clock_t(get_jiffies_64());
1045 }
1046
1047 #ifdef CONFIG_COMPAT
clock_t_to_compat_clock_t(clock_t x)1048 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
1049 {
1050 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
1051 }
1052
COMPAT_SYSCALL_DEFINE1(times,struct compat_tms __user *,tbuf)1053 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
1054 {
1055 if (tbuf) {
1056 struct tms tms;
1057 struct compat_tms tmp;
1058
1059 do_sys_times(&tms);
1060 /* Convert our struct tms to the compat version. */
1061 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1062 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1063 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1064 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1065 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1066 return -EFAULT;
1067 }
1068 force_successful_syscall_return();
1069 return compat_jiffies_to_clock_t(jiffies);
1070 }
1071 #endif
1072
1073 /*
1074 * This needs some heavy checking ...
1075 * I just haven't the stomach for it. I also don't fully
1076 * understand sessions/pgrp etc. Let somebody who does explain it.
1077 *
1078 * OK, I think I have the protection semantics right.... this is really
1079 * only important on a multi-user system anyway, to make sure one user
1080 * can't send a signal to a process owned by another. -TYT, 12/12/91
1081 *
1082 * !PF_FORKNOEXEC check to conform completely to POSIX.
1083 */
SYSCALL_DEFINE2(setpgid,pid_t,pid,pid_t,pgid)1084 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1085 {
1086 struct task_struct *p;
1087 struct task_struct *group_leader = current->group_leader;
1088 struct pid *pids[PIDTYPE_MAX] = { 0 };
1089 struct pid *pgrp;
1090 int err;
1091
1092 if (!pid)
1093 pid = task_pid_vnr(group_leader);
1094 if (!pgid)
1095 pgid = pid;
1096 if (pgid < 0)
1097 return -EINVAL;
1098 rcu_read_lock();
1099
1100 /* From this point forward we keep holding onto the tasklist lock
1101 * so that our parent does not change from under us. -DaveM
1102 */
1103 write_lock_irq(&tasklist_lock);
1104
1105 err = -ESRCH;
1106 p = find_task_by_vpid(pid);
1107 if (!p)
1108 goto out;
1109
1110 err = -EINVAL;
1111 if (!thread_group_leader(p))
1112 goto out;
1113
1114 if (same_thread_group(p->real_parent, group_leader)) {
1115 err = -EPERM;
1116 if (task_session(p) != task_session(group_leader))
1117 goto out;
1118 err = -EACCES;
1119 if (!(p->flags & PF_FORKNOEXEC))
1120 goto out;
1121 } else {
1122 err = -ESRCH;
1123 if (p != group_leader)
1124 goto out;
1125 }
1126
1127 err = -EPERM;
1128 if (p->signal->leader)
1129 goto out;
1130
1131 pgrp = task_pid(p);
1132 if (pgid != pid) {
1133 struct task_struct *g;
1134
1135 pgrp = find_vpid(pgid);
1136 g = pid_task(pgrp, PIDTYPE_PGID);
1137 if (!g || task_session(g) != task_session(group_leader))
1138 goto out;
1139 }
1140
1141 err = security_task_setpgid(p, pgid);
1142 if (err)
1143 goto out;
1144
1145 if (task_pgrp(p) != pgrp)
1146 change_pid(pids, p, PIDTYPE_PGID, pgrp);
1147
1148 err = 0;
1149 out:
1150 /* All paths lead to here, thus we are safe. -DaveM */
1151 write_unlock_irq(&tasklist_lock);
1152 rcu_read_unlock();
1153 free_pids(pids);
1154 return err;
1155 }
1156
do_getpgid(pid_t pid)1157 static int do_getpgid(pid_t pid)
1158 {
1159 struct task_struct *p;
1160 struct pid *grp;
1161 int retval;
1162
1163 rcu_read_lock();
1164 if (!pid)
1165 grp = task_pgrp(current);
1166 else {
1167 retval = -ESRCH;
1168 p = find_task_by_vpid(pid);
1169 if (!p)
1170 goto out;
1171 grp = task_pgrp(p);
1172 if (!grp)
1173 goto out;
1174
1175 retval = security_task_getpgid(p);
1176 if (retval)
1177 goto out;
1178 }
1179 retval = pid_vnr(grp);
1180 out:
1181 rcu_read_unlock();
1182 return retval;
1183 }
1184
SYSCALL_DEFINE1(getpgid,pid_t,pid)1185 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1186 {
1187 return do_getpgid(pid);
1188 }
1189
1190 #ifdef __ARCH_WANT_SYS_GETPGRP
1191
SYSCALL_DEFINE0(getpgrp)1192 SYSCALL_DEFINE0(getpgrp)
1193 {
1194 return do_getpgid(0);
1195 }
1196
1197 #endif
1198
SYSCALL_DEFINE1(getsid,pid_t,pid)1199 SYSCALL_DEFINE1(getsid, pid_t, pid)
1200 {
1201 struct task_struct *p;
1202 struct pid *sid;
1203 int retval;
1204
1205 rcu_read_lock();
1206 if (!pid)
1207 sid = task_session(current);
1208 else {
1209 retval = -ESRCH;
1210 p = find_task_by_vpid(pid);
1211 if (!p)
1212 goto out;
1213 sid = task_session(p);
1214 if (!sid)
1215 goto out;
1216
1217 retval = security_task_getsid(p);
1218 if (retval)
1219 goto out;
1220 }
1221 retval = pid_vnr(sid);
1222 out:
1223 rcu_read_unlock();
1224 return retval;
1225 }
1226
set_special_pids(struct pid ** pids,struct pid * pid)1227 static void set_special_pids(struct pid **pids, struct pid *pid)
1228 {
1229 struct task_struct *curr = current->group_leader;
1230
1231 if (task_session(curr) != pid)
1232 change_pid(pids, curr, PIDTYPE_SID, pid);
1233
1234 if (task_pgrp(curr) != pid)
1235 change_pid(pids, curr, PIDTYPE_PGID, pid);
1236 }
1237
ksys_setsid(void)1238 int ksys_setsid(void)
1239 {
1240 struct task_struct *group_leader = current->group_leader;
1241 struct pid *sid = task_pid(group_leader);
1242 struct pid *pids[PIDTYPE_MAX] = { 0 };
1243 pid_t session = pid_vnr(sid);
1244 int err = -EPERM;
1245
1246 write_lock_irq(&tasklist_lock);
1247 /* Fail if I am already a session leader */
1248 if (group_leader->signal->leader)
1249 goto out;
1250
1251 /* Fail if a process group id already exists that equals the
1252 * proposed session id.
1253 */
1254 if (pid_task(sid, PIDTYPE_PGID))
1255 goto out;
1256
1257 group_leader->signal->leader = 1;
1258 set_special_pids(pids, sid);
1259
1260 proc_clear_tty(group_leader);
1261
1262 err = session;
1263 out:
1264 write_unlock_irq(&tasklist_lock);
1265 free_pids(pids);
1266 if (err > 0) {
1267 proc_sid_connector(group_leader);
1268 sched_autogroup_create_attach(group_leader);
1269 }
1270 return err;
1271 }
1272
SYSCALL_DEFINE0(setsid)1273 SYSCALL_DEFINE0(setsid)
1274 {
1275 return ksys_setsid();
1276 }
1277
1278 DECLARE_RWSEM(uts_sem);
1279
1280 #ifdef COMPAT_UTS_MACHINE
1281 #define override_architecture(name) \
1282 (personality(current->personality) == PER_LINUX32 && \
1283 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1284 sizeof(COMPAT_UTS_MACHINE)))
1285 #else
1286 #define override_architecture(name) 0
1287 #endif
1288
1289 /*
1290 * Work around broken programs that cannot handle "Linux 3.0".
1291 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1292 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1293 * 2.6.60.
1294 */
override_release(char __user * release,size_t len)1295 static int override_release(char __user *release, size_t len)
1296 {
1297 int ret = 0;
1298
1299 if (current->personality & UNAME26) {
1300 const char *rest = UTS_RELEASE;
1301 char buf[65] = { 0 };
1302 int ndots = 0;
1303 unsigned v;
1304 size_t copy;
1305
1306 while (*rest) {
1307 if (*rest == '.' && ++ndots >= 3)
1308 break;
1309 if (!isdigit(*rest) && *rest != '.')
1310 break;
1311 rest++;
1312 }
1313 v = LINUX_VERSION_PATCHLEVEL + 60;
1314 copy = clamp_t(size_t, len, 1, sizeof(buf));
1315 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1316 ret = copy_to_user(release, buf, copy + 1);
1317 }
1318 return ret;
1319 }
1320
SYSCALL_DEFINE1(newuname,struct new_utsname __user *,name)1321 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1322 {
1323 struct new_utsname tmp;
1324
1325 down_read(&uts_sem);
1326 memcpy(&tmp, utsname(), sizeof(tmp));
1327 up_read(&uts_sem);
1328 if (copy_to_user(name, &tmp, sizeof(tmp)))
1329 return -EFAULT;
1330
1331 if (override_release(name->release, sizeof(name->release)))
1332 return -EFAULT;
1333 if (override_architecture(name))
1334 return -EFAULT;
1335 return 0;
1336 }
1337
1338 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1339 /*
1340 * Old cruft
1341 */
SYSCALL_DEFINE1(uname,struct old_utsname __user *,name)1342 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1343 {
1344 struct old_utsname tmp;
1345
1346 if (!name)
1347 return -EFAULT;
1348
1349 down_read(&uts_sem);
1350 memcpy(&tmp, utsname(), sizeof(tmp));
1351 up_read(&uts_sem);
1352 if (copy_to_user(name, &tmp, sizeof(tmp)))
1353 return -EFAULT;
1354
1355 if (override_release(name->release, sizeof(name->release)))
1356 return -EFAULT;
1357 if (override_architecture(name))
1358 return -EFAULT;
1359 return 0;
1360 }
1361
SYSCALL_DEFINE1(olduname,struct oldold_utsname __user *,name)1362 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1363 {
1364 struct oldold_utsname tmp;
1365
1366 if (!name)
1367 return -EFAULT;
1368
1369 memset(&tmp, 0, sizeof(tmp));
1370
1371 down_read(&uts_sem);
1372 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1373 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1374 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1375 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1376 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1377 up_read(&uts_sem);
1378 if (copy_to_user(name, &tmp, sizeof(tmp)))
1379 return -EFAULT;
1380
1381 if (override_architecture(name))
1382 return -EFAULT;
1383 if (override_release(name->release, sizeof(name->release)))
1384 return -EFAULT;
1385 return 0;
1386 }
1387 #endif
1388
SYSCALL_DEFINE2(sethostname,char __user *,name,int,len)1389 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1390 {
1391 int errno;
1392 char tmp[__NEW_UTS_LEN];
1393
1394 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1395 return -EPERM;
1396
1397 if (len < 0 || len > __NEW_UTS_LEN)
1398 return -EINVAL;
1399 errno = -EFAULT;
1400 if (!copy_from_user(tmp, name, len)) {
1401 struct new_utsname *u;
1402
1403 add_device_randomness(tmp, len);
1404 down_write(&uts_sem);
1405 u = utsname();
1406 memcpy(u->nodename, tmp, len);
1407 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1408 errno = 0;
1409 uts_proc_notify(UTS_PROC_HOSTNAME);
1410 up_write(&uts_sem);
1411 }
1412 return errno;
1413 }
1414
1415 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1416
SYSCALL_DEFINE2(gethostname,char __user *,name,int,len)1417 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1418 {
1419 int i;
1420 struct new_utsname *u;
1421 char tmp[__NEW_UTS_LEN + 1];
1422
1423 if (len < 0)
1424 return -EINVAL;
1425 down_read(&uts_sem);
1426 u = utsname();
1427 i = 1 + strlen(u->nodename);
1428 if (i > len)
1429 i = len;
1430 memcpy(tmp, u->nodename, i);
1431 up_read(&uts_sem);
1432 if (copy_to_user(name, tmp, i))
1433 return -EFAULT;
1434 return 0;
1435 }
1436
1437 #endif
1438
1439 /*
1440 * Only setdomainname; getdomainname can be implemented by calling
1441 * uname()
1442 */
SYSCALL_DEFINE2(setdomainname,char __user *,name,int,len)1443 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1444 {
1445 int errno;
1446 char tmp[__NEW_UTS_LEN];
1447
1448 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1449 return -EPERM;
1450 if (len < 0 || len > __NEW_UTS_LEN)
1451 return -EINVAL;
1452
1453 errno = -EFAULT;
1454 if (!copy_from_user(tmp, name, len)) {
1455 struct new_utsname *u;
1456
1457 add_device_randomness(tmp, len);
1458 down_write(&uts_sem);
1459 u = utsname();
1460 memcpy(u->domainname, tmp, len);
1461 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1462 errno = 0;
1463 uts_proc_notify(UTS_PROC_DOMAINNAME);
1464 up_write(&uts_sem);
1465 }
1466 return errno;
1467 }
1468
1469 /* make sure you are allowed to change @tsk limits before calling this */
do_prlimit(struct task_struct * tsk,unsigned int resource,struct rlimit * new_rlim,struct rlimit * old_rlim)1470 static int do_prlimit(struct task_struct *tsk, unsigned int resource,
1471 struct rlimit *new_rlim, struct rlimit *old_rlim)
1472 {
1473 struct rlimit *rlim;
1474 int retval = 0;
1475
1476 if (resource >= RLIM_NLIMITS)
1477 return -EINVAL;
1478 resource = array_index_nospec(resource, RLIM_NLIMITS);
1479
1480 if (new_rlim) {
1481 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1482 return -EINVAL;
1483 if (resource == RLIMIT_NOFILE &&
1484 new_rlim->rlim_max > sysctl_nr_open)
1485 return -EPERM;
1486 }
1487
1488 /* Holding a refcount on tsk protects tsk->signal from disappearing. */
1489 rlim = tsk->signal->rlim + resource;
1490 task_lock(tsk->group_leader);
1491 if (new_rlim) {
1492 /*
1493 * Keep the capable check against init_user_ns until cgroups can
1494 * contain all limits.
1495 */
1496 if (new_rlim->rlim_max > rlim->rlim_max &&
1497 !capable(CAP_SYS_RESOURCE))
1498 retval = -EPERM;
1499 if (!retval)
1500 retval = security_task_setrlimit(tsk, resource, new_rlim);
1501 }
1502 if (!retval) {
1503 if (old_rlim)
1504 *old_rlim = *rlim;
1505 if (new_rlim)
1506 *rlim = *new_rlim;
1507 }
1508 task_unlock(tsk->group_leader);
1509
1510 /*
1511 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1512 * infinite. In case of RLIM_INFINITY the posix CPU timer code
1513 * ignores the rlimit.
1514 */
1515 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1516 new_rlim->rlim_cur != RLIM_INFINITY &&
1517 IS_ENABLED(CONFIG_POSIX_TIMERS)) {
1518 /*
1519 * update_rlimit_cpu can fail if the task is exiting, but there
1520 * may be other tasks in the thread group that are not exiting,
1521 * and they need their cpu timers adjusted.
1522 *
1523 * The group_leader is the last task to be released, so if we
1524 * cannot update_rlimit_cpu on it, then the entire process is
1525 * exiting and we do not need to update at all.
1526 */
1527 update_rlimit_cpu(tsk->group_leader, new_rlim->rlim_cur);
1528 }
1529
1530 return retval;
1531 }
1532
SYSCALL_DEFINE2(getrlimit,unsigned int,resource,struct rlimit __user *,rlim)1533 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1534 {
1535 struct rlimit value;
1536 int ret;
1537
1538 ret = do_prlimit(current, resource, NULL, &value);
1539 if (!ret)
1540 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1541
1542 return ret;
1543 }
1544
1545 #ifdef CONFIG_COMPAT
1546
COMPAT_SYSCALL_DEFINE2(setrlimit,unsigned int,resource,struct compat_rlimit __user *,rlim)1547 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1548 struct compat_rlimit __user *, rlim)
1549 {
1550 struct rlimit r;
1551 struct compat_rlimit r32;
1552
1553 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1554 return -EFAULT;
1555
1556 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1557 r.rlim_cur = RLIM_INFINITY;
1558 else
1559 r.rlim_cur = r32.rlim_cur;
1560 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1561 r.rlim_max = RLIM_INFINITY;
1562 else
1563 r.rlim_max = r32.rlim_max;
1564 return do_prlimit(current, resource, &r, NULL);
1565 }
1566
COMPAT_SYSCALL_DEFINE2(getrlimit,unsigned int,resource,struct compat_rlimit __user *,rlim)1567 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1568 struct compat_rlimit __user *, rlim)
1569 {
1570 struct rlimit r;
1571 int ret;
1572
1573 ret = do_prlimit(current, resource, NULL, &r);
1574 if (!ret) {
1575 struct compat_rlimit r32;
1576 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1577 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1578 else
1579 r32.rlim_cur = r.rlim_cur;
1580 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1581 r32.rlim_max = COMPAT_RLIM_INFINITY;
1582 else
1583 r32.rlim_max = r.rlim_max;
1584
1585 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1586 return -EFAULT;
1587 }
1588 return ret;
1589 }
1590
1591 #endif
1592
1593 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1594
1595 /*
1596 * Back compatibility for getrlimit. Needed for some apps.
1597 */
SYSCALL_DEFINE2(old_getrlimit,unsigned int,resource,struct rlimit __user *,rlim)1598 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1599 struct rlimit __user *, rlim)
1600 {
1601 struct rlimit x;
1602 if (resource >= RLIM_NLIMITS)
1603 return -EINVAL;
1604
1605 resource = array_index_nospec(resource, RLIM_NLIMITS);
1606 task_lock(current->group_leader);
1607 x = current->signal->rlim[resource];
1608 task_unlock(current->group_leader);
1609 if (x.rlim_cur > 0x7FFFFFFF)
1610 x.rlim_cur = 0x7FFFFFFF;
1611 if (x.rlim_max > 0x7FFFFFFF)
1612 x.rlim_max = 0x7FFFFFFF;
1613 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1614 }
1615
1616 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(old_getrlimit,unsigned int,resource,struct compat_rlimit __user *,rlim)1617 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1618 struct compat_rlimit __user *, rlim)
1619 {
1620 struct rlimit r;
1621
1622 if (resource >= RLIM_NLIMITS)
1623 return -EINVAL;
1624
1625 resource = array_index_nospec(resource, RLIM_NLIMITS);
1626 task_lock(current->group_leader);
1627 r = current->signal->rlim[resource];
1628 task_unlock(current->group_leader);
1629 if (r.rlim_cur > 0x7FFFFFFF)
1630 r.rlim_cur = 0x7FFFFFFF;
1631 if (r.rlim_max > 0x7FFFFFFF)
1632 r.rlim_max = 0x7FFFFFFF;
1633
1634 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1635 put_user(r.rlim_max, &rlim->rlim_max))
1636 return -EFAULT;
1637 return 0;
1638 }
1639 #endif
1640
1641 #endif
1642
rlim64_is_infinity(__u64 rlim64)1643 static inline bool rlim64_is_infinity(__u64 rlim64)
1644 {
1645 #if BITS_PER_LONG < 64
1646 return rlim64 >= ULONG_MAX;
1647 #else
1648 return rlim64 == RLIM64_INFINITY;
1649 #endif
1650 }
1651
rlim_to_rlim64(const struct rlimit * rlim,struct rlimit64 * rlim64)1652 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1653 {
1654 if (rlim->rlim_cur == RLIM_INFINITY)
1655 rlim64->rlim_cur = RLIM64_INFINITY;
1656 else
1657 rlim64->rlim_cur = rlim->rlim_cur;
1658 if (rlim->rlim_max == RLIM_INFINITY)
1659 rlim64->rlim_max = RLIM64_INFINITY;
1660 else
1661 rlim64->rlim_max = rlim->rlim_max;
1662 }
1663
rlim64_to_rlim(const struct rlimit64 * rlim64,struct rlimit * rlim)1664 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1665 {
1666 if (rlim64_is_infinity(rlim64->rlim_cur))
1667 rlim->rlim_cur = RLIM_INFINITY;
1668 else
1669 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1670 if (rlim64_is_infinity(rlim64->rlim_max))
1671 rlim->rlim_max = RLIM_INFINITY;
1672 else
1673 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1674 }
1675
1676 /* rcu lock must be held */
check_prlimit_permission(struct task_struct * task,unsigned int flags)1677 static int check_prlimit_permission(struct task_struct *task,
1678 unsigned int flags)
1679 {
1680 const struct cred *cred = current_cred(), *tcred;
1681 bool id_match;
1682
1683 if (current == task)
1684 return 0;
1685
1686 tcred = __task_cred(task);
1687 id_match = (uid_eq(cred->uid, tcred->euid) &&
1688 uid_eq(cred->uid, tcred->suid) &&
1689 uid_eq(cred->uid, tcred->uid) &&
1690 gid_eq(cred->gid, tcred->egid) &&
1691 gid_eq(cred->gid, tcred->sgid) &&
1692 gid_eq(cred->gid, tcred->gid));
1693 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1694 return -EPERM;
1695
1696 return security_task_prlimit(cred, tcred, flags);
1697 }
1698
SYSCALL_DEFINE4(prlimit64,pid_t,pid,unsigned int,resource,const struct rlimit64 __user *,new_rlim,struct rlimit64 __user *,old_rlim)1699 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1700 const struct rlimit64 __user *, new_rlim,
1701 struct rlimit64 __user *, old_rlim)
1702 {
1703 struct rlimit64 old64, new64;
1704 struct rlimit old, new;
1705 struct task_struct *tsk;
1706 unsigned int checkflags = 0;
1707 int ret;
1708
1709 if (old_rlim)
1710 checkflags |= LSM_PRLIMIT_READ;
1711
1712 if (new_rlim) {
1713 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1714 return -EFAULT;
1715 rlim64_to_rlim(&new64, &new);
1716 checkflags |= LSM_PRLIMIT_WRITE;
1717 }
1718
1719 rcu_read_lock();
1720 tsk = pid ? find_task_by_vpid(pid) : current;
1721 if (!tsk) {
1722 rcu_read_unlock();
1723 return -ESRCH;
1724 }
1725 ret = check_prlimit_permission(tsk, checkflags);
1726 if (ret) {
1727 rcu_read_unlock();
1728 return ret;
1729 }
1730 get_task_struct(tsk);
1731 rcu_read_unlock();
1732
1733 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1734 old_rlim ? &old : NULL);
1735
1736 if (!ret && old_rlim) {
1737 rlim_to_rlim64(&old, &old64);
1738 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1739 ret = -EFAULT;
1740 }
1741
1742 put_task_struct(tsk);
1743 return ret;
1744 }
1745
SYSCALL_DEFINE2(setrlimit,unsigned int,resource,struct rlimit __user *,rlim)1746 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1747 {
1748 struct rlimit new_rlim;
1749
1750 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1751 return -EFAULT;
1752 return do_prlimit(current, resource, &new_rlim, NULL);
1753 }
1754
1755 /*
1756 * It would make sense to put struct rusage in the task_struct,
1757 * except that would make the task_struct be *really big*. After
1758 * task_struct gets moved into malloc'ed memory, it would
1759 * make sense to do this. It will make moving the rest of the information
1760 * a lot simpler! (Which we're not doing right now because we're not
1761 * measuring them yet).
1762 *
1763 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1764 * races with threads incrementing their own counters. But since word
1765 * reads are atomic, we either get new values or old values and we don't
1766 * care which for the sums. We always take the siglock to protect reading
1767 * the c* fields from p->signal from races with exit.c updating those
1768 * fields when reaping, so a sample either gets all the additions of a
1769 * given child after it's reaped, or none so this sample is before reaping.
1770 *
1771 * Locking:
1772 * We need to take the siglock for CHILDEREN, SELF and BOTH
1773 * for the cases current multithreaded, non-current single threaded
1774 * non-current multithreaded. Thread traversal is now safe with
1775 * the siglock held.
1776 * Strictly speaking, we donot need to take the siglock if we are current and
1777 * single threaded, as no one else can take our signal_struct away, no one
1778 * else can reap the children to update signal->c* counters, and no one else
1779 * can race with the signal-> fields. If we do not take any lock, the
1780 * signal-> fields could be read out of order while another thread was just
1781 * exiting. So we should place a read memory barrier when we avoid the lock.
1782 * On the writer side, write memory barrier is implied in __exit_signal
1783 * as __exit_signal releases the siglock spinlock after updating the signal->
1784 * fields. But we don't do this yet to keep things simple.
1785 *
1786 */
1787
accumulate_thread_rusage(struct task_struct * t,struct rusage * r)1788 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1789 {
1790 r->ru_nvcsw += t->nvcsw;
1791 r->ru_nivcsw += t->nivcsw;
1792 r->ru_minflt += t->min_flt;
1793 r->ru_majflt += t->maj_flt;
1794 r->ru_inblock += task_io_get_inblock(t);
1795 r->ru_oublock += task_io_get_oublock(t);
1796 }
1797
getrusage(struct task_struct * p,int who,struct rusage * r)1798 void getrusage(struct task_struct *p, int who, struct rusage *r)
1799 {
1800 struct task_struct *t;
1801 unsigned long flags;
1802 u64 tgutime, tgstime, utime, stime;
1803 unsigned long maxrss;
1804 struct mm_struct *mm;
1805 struct signal_struct *sig = p->signal;
1806 unsigned int seq = 0;
1807
1808 retry:
1809 memset(r, 0, sizeof(*r));
1810 utime = stime = 0;
1811 maxrss = 0;
1812
1813 if (who == RUSAGE_THREAD) {
1814 task_cputime_adjusted(current, &utime, &stime);
1815 accumulate_thread_rusage(p, r);
1816 maxrss = sig->maxrss;
1817 goto out_thread;
1818 }
1819
1820 flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
1821
1822 switch (who) {
1823 case RUSAGE_BOTH:
1824 case RUSAGE_CHILDREN:
1825 utime = sig->cutime;
1826 stime = sig->cstime;
1827 r->ru_nvcsw = sig->cnvcsw;
1828 r->ru_nivcsw = sig->cnivcsw;
1829 r->ru_minflt = sig->cmin_flt;
1830 r->ru_majflt = sig->cmaj_flt;
1831 r->ru_inblock = sig->cinblock;
1832 r->ru_oublock = sig->coublock;
1833 maxrss = sig->cmaxrss;
1834
1835 if (who == RUSAGE_CHILDREN)
1836 break;
1837 fallthrough;
1838
1839 case RUSAGE_SELF:
1840 r->ru_nvcsw += sig->nvcsw;
1841 r->ru_nivcsw += sig->nivcsw;
1842 r->ru_minflt += sig->min_flt;
1843 r->ru_majflt += sig->maj_flt;
1844 r->ru_inblock += sig->inblock;
1845 r->ru_oublock += sig->oublock;
1846 if (maxrss < sig->maxrss)
1847 maxrss = sig->maxrss;
1848
1849 rcu_read_lock();
1850 __for_each_thread(sig, t)
1851 accumulate_thread_rusage(t, r);
1852 rcu_read_unlock();
1853
1854 break;
1855
1856 default:
1857 BUG();
1858 }
1859
1860 if (need_seqretry(&sig->stats_lock, seq)) {
1861 seq = 1;
1862 goto retry;
1863 }
1864 done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
1865
1866 if (who == RUSAGE_CHILDREN)
1867 goto out_children;
1868
1869 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1870 utime += tgutime;
1871 stime += tgstime;
1872
1873 out_thread:
1874 mm = get_task_mm(p);
1875 if (mm) {
1876 setmax_mm_hiwater_rss(&maxrss, mm);
1877 mmput(mm);
1878 }
1879
1880 out_children:
1881 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1882 r->ru_utime = ns_to_kernel_old_timeval(utime);
1883 r->ru_stime = ns_to_kernel_old_timeval(stime);
1884 }
1885
SYSCALL_DEFINE2(getrusage,int,who,struct rusage __user *,ru)1886 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1887 {
1888 struct rusage r;
1889
1890 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1891 who != RUSAGE_THREAD)
1892 return -EINVAL;
1893
1894 getrusage(current, who, &r);
1895 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1896 }
1897
1898 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(getrusage,int,who,struct compat_rusage __user *,ru)1899 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1900 {
1901 struct rusage r;
1902
1903 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1904 who != RUSAGE_THREAD)
1905 return -EINVAL;
1906
1907 getrusage(current, who, &r);
1908 return put_compat_rusage(&r, ru);
1909 }
1910 #endif
1911
SYSCALL_DEFINE1(umask,int,mask)1912 SYSCALL_DEFINE1(umask, int, mask)
1913 {
1914 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1915 return mask;
1916 }
1917
prctl_set_mm_exe_file(struct mm_struct * mm,unsigned int fd)1918 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1919 {
1920 CLASS(fd, exe)(fd);
1921 struct inode *inode;
1922 int err;
1923
1924 if (fd_empty(exe))
1925 return -EBADF;
1926
1927 inode = file_inode(fd_file(exe));
1928
1929 /*
1930 * Because the original mm->exe_file points to executable file, make
1931 * sure that this one is executable as well, to avoid breaking an
1932 * overall picture.
1933 */
1934 if (!S_ISREG(inode->i_mode) || path_noexec(&fd_file(exe)->f_path))
1935 return -EACCES;
1936
1937 err = file_permission(fd_file(exe), MAY_EXEC);
1938 if (err)
1939 return err;
1940
1941 return replace_mm_exe_file(mm, fd_file(exe));
1942 }
1943
1944 /*
1945 * Check arithmetic relations of passed addresses.
1946 *
1947 * WARNING: we don't require any capability here so be very careful
1948 * in what is allowed for modification from userspace.
1949 */
validate_prctl_map_addr(struct prctl_mm_map * prctl_map)1950 static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1951 {
1952 unsigned long mmap_max_addr = TASK_SIZE;
1953 int error = -EINVAL, i;
1954
1955 static const unsigned char offsets[] = {
1956 offsetof(struct prctl_mm_map, start_code),
1957 offsetof(struct prctl_mm_map, end_code),
1958 offsetof(struct prctl_mm_map, start_data),
1959 offsetof(struct prctl_mm_map, end_data),
1960 offsetof(struct prctl_mm_map, start_brk),
1961 offsetof(struct prctl_mm_map, brk),
1962 offsetof(struct prctl_mm_map, start_stack),
1963 offsetof(struct prctl_mm_map, arg_start),
1964 offsetof(struct prctl_mm_map, arg_end),
1965 offsetof(struct prctl_mm_map, env_start),
1966 offsetof(struct prctl_mm_map, env_end),
1967 };
1968
1969 /*
1970 * Make sure the members are not somewhere outside
1971 * of allowed address space.
1972 */
1973 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1974 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1975
1976 if ((unsigned long)val >= mmap_max_addr ||
1977 (unsigned long)val < mmap_min_addr)
1978 goto out;
1979 }
1980
1981 /*
1982 * Make sure the pairs are ordered.
1983 */
1984 #define __prctl_check_order(__m1, __op, __m2) \
1985 ((unsigned long)prctl_map->__m1 __op \
1986 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1987 error = __prctl_check_order(start_code, <, end_code);
1988 error |= __prctl_check_order(start_data,<=, end_data);
1989 error |= __prctl_check_order(start_brk, <=, brk);
1990 error |= __prctl_check_order(arg_start, <=, arg_end);
1991 error |= __prctl_check_order(env_start, <=, env_end);
1992 if (error)
1993 goto out;
1994 #undef __prctl_check_order
1995
1996 error = -EINVAL;
1997
1998 /*
1999 * Neither we should allow to override limits if they set.
2000 */
2001 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
2002 prctl_map->start_brk, prctl_map->end_data,
2003 prctl_map->start_data))
2004 goto out;
2005
2006 error = 0;
2007 out:
2008 return error;
2009 }
2010
2011 #ifdef CONFIG_CHECKPOINT_RESTORE
prctl_set_mm_map(int opt,const void __user * addr,unsigned long data_size)2012 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
2013 {
2014 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
2015 unsigned long user_auxv[AT_VECTOR_SIZE];
2016 struct mm_struct *mm = current->mm;
2017 int error;
2018
2019 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2020 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
2021
2022 if (opt == PR_SET_MM_MAP_SIZE)
2023 return put_user((unsigned int)sizeof(prctl_map),
2024 (unsigned int __user *)addr);
2025
2026 if (data_size != sizeof(prctl_map))
2027 return -EINVAL;
2028
2029 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
2030 return -EFAULT;
2031
2032 error = validate_prctl_map_addr(&prctl_map);
2033 if (error)
2034 return error;
2035
2036 if (prctl_map.auxv_size) {
2037 /*
2038 * Someone is trying to cheat the auxv vector.
2039 */
2040 if (!prctl_map.auxv ||
2041 prctl_map.auxv_size > sizeof(mm->saved_auxv))
2042 return -EINVAL;
2043
2044 memset(user_auxv, 0, sizeof(user_auxv));
2045 if (copy_from_user(user_auxv,
2046 (const void __user *)prctl_map.auxv,
2047 prctl_map.auxv_size))
2048 return -EFAULT;
2049
2050 /* Last entry must be AT_NULL as specification requires */
2051 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2052 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2053 }
2054
2055 if (prctl_map.exe_fd != (u32)-1) {
2056 /*
2057 * Check if the current user is checkpoint/restore capable.
2058 * At the time of this writing, it checks for CAP_SYS_ADMIN
2059 * or CAP_CHECKPOINT_RESTORE.
2060 * Note that a user with access to ptrace can masquerade an
2061 * arbitrary program as any executable, even setuid ones.
2062 * This may have implications in the tomoyo subsystem.
2063 */
2064 if (!checkpoint_restore_ns_capable(current_user_ns()))
2065 return -EPERM;
2066
2067 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2068 if (error)
2069 return error;
2070 }
2071
2072 /*
2073 * arg_lock protects concurrent updates but we still need mmap_lock for
2074 * read to exclude races with sys_brk.
2075 */
2076 mmap_read_lock(mm);
2077
2078 /*
2079 * We don't validate if these members are pointing to
2080 * real present VMAs because application may have correspond
2081 * VMAs already unmapped and kernel uses these members for statistics
2082 * output in procfs mostly, except
2083 *
2084 * - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2085 * for VMAs when updating these members so anything wrong written
2086 * here cause kernel to swear at userspace program but won't lead
2087 * to any problem in kernel itself
2088 */
2089
2090 spin_lock(&mm->arg_lock);
2091 mm->start_code = prctl_map.start_code;
2092 mm->end_code = prctl_map.end_code;
2093 mm->start_data = prctl_map.start_data;
2094 mm->end_data = prctl_map.end_data;
2095 mm->start_brk = prctl_map.start_brk;
2096 mm->brk = prctl_map.brk;
2097 mm->start_stack = prctl_map.start_stack;
2098 mm->arg_start = prctl_map.arg_start;
2099 mm->arg_end = prctl_map.arg_end;
2100 mm->env_start = prctl_map.env_start;
2101 mm->env_end = prctl_map.env_end;
2102 spin_unlock(&mm->arg_lock);
2103
2104 /*
2105 * Note this update of @saved_auxv is lockless thus
2106 * if someone reads this member in procfs while we're
2107 * updating -- it may get partly updated results. It's
2108 * known and acceptable trade off: we leave it as is to
2109 * not introduce additional locks here making the kernel
2110 * more complex.
2111 */
2112 if (prctl_map.auxv_size)
2113 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2114
2115 mmap_read_unlock(mm);
2116 return 0;
2117 }
2118 #endif /* CONFIG_CHECKPOINT_RESTORE */
2119
prctl_set_auxv(struct mm_struct * mm,unsigned long addr,unsigned long len)2120 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2121 unsigned long len)
2122 {
2123 /*
2124 * This doesn't move the auxiliary vector itself since it's pinned to
2125 * mm_struct, but it permits filling the vector with new values. It's
2126 * up to the caller to provide sane values here, otherwise userspace
2127 * tools which use this vector might be unhappy.
2128 */
2129 unsigned long user_auxv[AT_VECTOR_SIZE] = {};
2130
2131 if (len > sizeof(user_auxv))
2132 return -EINVAL;
2133
2134 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2135 return -EFAULT;
2136
2137 /* Make sure the last entry is always AT_NULL */
2138 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2139 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2140
2141 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2142
2143 task_lock(current);
2144 memcpy(mm->saved_auxv, user_auxv, len);
2145 task_unlock(current);
2146
2147 return 0;
2148 }
2149
prctl_set_mm(int opt,unsigned long addr,unsigned long arg4,unsigned long arg5)2150 static int prctl_set_mm(int opt, unsigned long addr,
2151 unsigned long arg4, unsigned long arg5)
2152 {
2153 struct mm_struct *mm = current->mm;
2154 struct prctl_mm_map prctl_map = {
2155 .auxv = NULL,
2156 .auxv_size = 0,
2157 .exe_fd = -1,
2158 };
2159 struct vm_area_struct *vma;
2160 int error;
2161
2162 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2163 opt != PR_SET_MM_MAP &&
2164 opt != PR_SET_MM_MAP_SIZE)))
2165 return -EINVAL;
2166
2167 #ifdef CONFIG_CHECKPOINT_RESTORE
2168 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2169 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2170 #endif
2171
2172 if (!capable(CAP_SYS_RESOURCE))
2173 return -EPERM;
2174
2175 if (opt == PR_SET_MM_EXE_FILE)
2176 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2177
2178 if (opt == PR_SET_MM_AUXV)
2179 return prctl_set_auxv(mm, addr, arg4);
2180
2181 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2182 return -EINVAL;
2183
2184 error = -EINVAL;
2185
2186 /*
2187 * arg_lock protects concurrent updates of arg boundaries, we need
2188 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2189 * validation.
2190 */
2191 mmap_read_lock(mm);
2192 vma = find_vma(mm, addr);
2193
2194 spin_lock(&mm->arg_lock);
2195 prctl_map.start_code = mm->start_code;
2196 prctl_map.end_code = mm->end_code;
2197 prctl_map.start_data = mm->start_data;
2198 prctl_map.end_data = mm->end_data;
2199 prctl_map.start_brk = mm->start_brk;
2200 prctl_map.brk = mm->brk;
2201 prctl_map.start_stack = mm->start_stack;
2202 prctl_map.arg_start = mm->arg_start;
2203 prctl_map.arg_end = mm->arg_end;
2204 prctl_map.env_start = mm->env_start;
2205 prctl_map.env_end = mm->env_end;
2206
2207 switch (opt) {
2208 case PR_SET_MM_START_CODE:
2209 prctl_map.start_code = addr;
2210 break;
2211 case PR_SET_MM_END_CODE:
2212 prctl_map.end_code = addr;
2213 break;
2214 case PR_SET_MM_START_DATA:
2215 prctl_map.start_data = addr;
2216 break;
2217 case PR_SET_MM_END_DATA:
2218 prctl_map.end_data = addr;
2219 break;
2220 case PR_SET_MM_START_STACK:
2221 prctl_map.start_stack = addr;
2222 break;
2223 case PR_SET_MM_START_BRK:
2224 prctl_map.start_brk = addr;
2225 break;
2226 case PR_SET_MM_BRK:
2227 prctl_map.brk = addr;
2228 break;
2229 case PR_SET_MM_ARG_START:
2230 prctl_map.arg_start = addr;
2231 break;
2232 case PR_SET_MM_ARG_END:
2233 prctl_map.arg_end = addr;
2234 break;
2235 case PR_SET_MM_ENV_START:
2236 prctl_map.env_start = addr;
2237 break;
2238 case PR_SET_MM_ENV_END:
2239 prctl_map.env_end = addr;
2240 break;
2241 default:
2242 goto out;
2243 }
2244
2245 error = validate_prctl_map_addr(&prctl_map);
2246 if (error)
2247 goto out;
2248
2249 switch (opt) {
2250 /*
2251 * If command line arguments and environment
2252 * are placed somewhere else on stack, we can
2253 * set them up here, ARG_START/END to setup
2254 * command line arguments and ENV_START/END
2255 * for environment.
2256 */
2257 case PR_SET_MM_START_STACK:
2258 case PR_SET_MM_ARG_START:
2259 case PR_SET_MM_ARG_END:
2260 case PR_SET_MM_ENV_START:
2261 case PR_SET_MM_ENV_END:
2262 if (!vma) {
2263 error = -EFAULT;
2264 goto out;
2265 }
2266 }
2267
2268 mm->start_code = prctl_map.start_code;
2269 mm->end_code = prctl_map.end_code;
2270 mm->start_data = prctl_map.start_data;
2271 mm->end_data = prctl_map.end_data;
2272 mm->start_brk = prctl_map.start_brk;
2273 mm->brk = prctl_map.brk;
2274 mm->start_stack = prctl_map.start_stack;
2275 mm->arg_start = prctl_map.arg_start;
2276 mm->arg_end = prctl_map.arg_end;
2277 mm->env_start = prctl_map.env_start;
2278 mm->env_end = prctl_map.env_end;
2279
2280 error = 0;
2281 out:
2282 spin_unlock(&mm->arg_lock);
2283 mmap_read_unlock(mm);
2284 return error;
2285 }
2286
2287 #ifdef CONFIG_CHECKPOINT_RESTORE
prctl_get_tid_address(struct task_struct * me,int __user * __user * tid_addr)2288 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2289 {
2290 return put_user(me->clear_child_tid, tid_addr);
2291 }
2292 #else
prctl_get_tid_address(struct task_struct * me,int __user * __user * tid_addr)2293 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2294 {
2295 return -EINVAL;
2296 }
2297 #endif
2298
propagate_has_child_subreaper(struct task_struct * p,void * data)2299 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2300 {
2301 /*
2302 * If task has has_child_subreaper - all its descendants
2303 * already have these flag too and new descendants will
2304 * inherit it on fork, skip them.
2305 *
2306 * If we've found child_reaper - skip descendants in
2307 * it's subtree as they will never get out pidns.
2308 */
2309 if (p->signal->has_child_subreaper ||
2310 is_child_reaper(task_pid(p)))
2311 return 0;
2312
2313 p->signal->has_child_subreaper = 1;
2314 return 1;
2315 }
2316
arch_prctl_spec_ctrl_get(struct task_struct * t,unsigned long which)2317 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2318 {
2319 return -EINVAL;
2320 }
2321
arch_prctl_spec_ctrl_set(struct task_struct * t,unsigned long which,unsigned long ctrl)2322 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2323 unsigned long ctrl)
2324 {
2325 return -EINVAL;
2326 }
2327
arch_get_shadow_stack_status(struct task_struct * t,unsigned long __user * status)2328 int __weak arch_get_shadow_stack_status(struct task_struct *t, unsigned long __user *status)
2329 {
2330 return -EINVAL;
2331 }
2332
arch_set_shadow_stack_status(struct task_struct * t,unsigned long status)2333 int __weak arch_set_shadow_stack_status(struct task_struct *t, unsigned long status)
2334 {
2335 return -EINVAL;
2336 }
2337
arch_lock_shadow_stack_status(struct task_struct * t,unsigned long status)2338 int __weak arch_lock_shadow_stack_status(struct task_struct *t, unsigned long status)
2339 {
2340 return -EINVAL;
2341 }
2342
2343 #define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2344
2345 #ifdef CONFIG_ANON_VMA_NAME
2346
2347 #define ANON_VMA_NAME_MAX_LEN 80
2348 #define ANON_VMA_NAME_INVALID_CHARS "\\`$[]"
2349
is_valid_name_char(char ch)2350 static inline bool is_valid_name_char(char ch)
2351 {
2352 /* printable ascii characters, excluding ANON_VMA_NAME_INVALID_CHARS */
2353 return ch > 0x1f && ch < 0x7f &&
2354 !strchr(ANON_VMA_NAME_INVALID_CHARS, ch);
2355 }
2356
prctl_set_vma(unsigned long opt,unsigned long addr,unsigned long size,unsigned long arg)2357 static int prctl_set_vma(unsigned long opt, unsigned long addr,
2358 unsigned long size, unsigned long arg)
2359 {
2360 struct mm_struct *mm = current->mm;
2361 const char __user *uname;
2362 struct anon_vma_name *anon_name = NULL;
2363 int error;
2364
2365 switch (opt) {
2366 case PR_SET_VMA_ANON_NAME:
2367 uname = (const char __user *)arg;
2368 if (uname) {
2369 char *name, *pch;
2370
2371 name = strndup_user(uname, ANON_VMA_NAME_MAX_LEN);
2372 if (IS_ERR(name))
2373 return PTR_ERR(name);
2374
2375 for (pch = name; *pch != '\0'; pch++) {
2376 if (!is_valid_name_char(*pch)) {
2377 kfree(name);
2378 return -EINVAL;
2379 }
2380 }
2381 /* anon_vma has its own copy */
2382 anon_name = anon_vma_name_alloc(name);
2383 kfree(name);
2384 if (!anon_name)
2385 return -ENOMEM;
2386
2387 }
2388
2389 mmap_write_lock(mm);
2390 error = madvise_set_anon_name(mm, addr, size, anon_name);
2391 mmap_write_unlock(mm);
2392 anon_vma_name_put(anon_name);
2393 break;
2394 default:
2395 error = -EINVAL;
2396 }
2397
2398 return error;
2399 }
2400
2401 #else /* CONFIG_ANON_VMA_NAME */
prctl_set_vma(unsigned long opt,unsigned long start,unsigned long size,unsigned long arg)2402 static int prctl_set_vma(unsigned long opt, unsigned long start,
2403 unsigned long size, unsigned long arg)
2404 {
2405 return -EINVAL;
2406 }
2407 #endif /* CONFIG_ANON_VMA_NAME */
2408
get_current_mdwe(void)2409 static inline unsigned long get_current_mdwe(void)
2410 {
2411 unsigned long ret = 0;
2412
2413 if (test_bit(MMF_HAS_MDWE, ¤t->mm->flags))
2414 ret |= PR_MDWE_REFUSE_EXEC_GAIN;
2415 if (test_bit(MMF_HAS_MDWE_NO_INHERIT, ¤t->mm->flags))
2416 ret |= PR_MDWE_NO_INHERIT;
2417
2418 return ret;
2419 }
2420
prctl_set_mdwe(unsigned long bits,unsigned long arg3,unsigned long arg4,unsigned long arg5)2421 static inline int prctl_set_mdwe(unsigned long bits, unsigned long arg3,
2422 unsigned long arg4, unsigned long arg5)
2423 {
2424 unsigned long current_bits;
2425
2426 if (arg3 || arg4 || arg5)
2427 return -EINVAL;
2428
2429 if (bits & ~(PR_MDWE_REFUSE_EXEC_GAIN | PR_MDWE_NO_INHERIT))
2430 return -EINVAL;
2431
2432 /* NO_INHERIT only makes sense with REFUSE_EXEC_GAIN */
2433 if (bits & PR_MDWE_NO_INHERIT && !(bits & PR_MDWE_REFUSE_EXEC_GAIN))
2434 return -EINVAL;
2435
2436 /*
2437 * EOPNOTSUPP might be more appropriate here in principle, but
2438 * existing userspace depends on EINVAL specifically.
2439 */
2440 if (!arch_memory_deny_write_exec_supported())
2441 return -EINVAL;
2442
2443 current_bits = get_current_mdwe();
2444 if (current_bits && current_bits != bits)
2445 return -EPERM; /* Cannot unset the flags */
2446
2447 if (bits & PR_MDWE_NO_INHERIT)
2448 set_bit(MMF_HAS_MDWE_NO_INHERIT, ¤t->mm->flags);
2449 if (bits & PR_MDWE_REFUSE_EXEC_GAIN)
2450 set_bit(MMF_HAS_MDWE, ¤t->mm->flags);
2451
2452 return 0;
2453 }
2454
prctl_get_mdwe(unsigned long arg2,unsigned long arg3,unsigned long arg4,unsigned long arg5)2455 static inline int prctl_get_mdwe(unsigned long arg2, unsigned long arg3,
2456 unsigned long arg4, unsigned long arg5)
2457 {
2458 if (arg2 || arg3 || arg4 || arg5)
2459 return -EINVAL;
2460 return get_current_mdwe();
2461 }
2462
prctl_get_auxv(void __user * addr,unsigned long len)2463 static int prctl_get_auxv(void __user *addr, unsigned long len)
2464 {
2465 struct mm_struct *mm = current->mm;
2466 unsigned long size = min_t(unsigned long, sizeof(mm->saved_auxv), len);
2467
2468 if (size && copy_to_user(addr, mm->saved_auxv, size))
2469 return -EFAULT;
2470 return sizeof(mm->saved_auxv);
2471 }
2472
SYSCALL_DEFINE5(prctl,int,option,unsigned long,arg2,unsigned long,arg3,unsigned long,arg4,unsigned long,arg5)2473 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2474 unsigned long, arg4, unsigned long, arg5)
2475 {
2476 struct task_struct *me = current;
2477 unsigned char comm[sizeof(me->comm)];
2478 long error;
2479
2480 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2481 if (error != -ENOSYS)
2482 return error;
2483
2484 error = 0;
2485 switch (option) {
2486 case PR_SET_PDEATHSIG:
2487 if (!valid_signal(arg2)) {
2488 error = -EINVAL;
2489 break;
2490 }
2491 me->pdeath_signal = arg2;
2492 break;
2493 case PR_GET_PDEATHSIG:
2494 error = put_user(me->pdeath_signal, (int __user *)arg2);
2495 break;
2496 case PR_GET_DUMPABLE:
2497 error = get_dumpable(me->mm);
2498 break;
2499 case PR_SET_DUMPABLE:
2500 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2501 error = -EINVAL;
2502 break;
2503 }
2504 set_dumpable(me->mm, arg2);
2505 break;
2506
2507 case PR_SET_UNALIGN:
2508 error = SET_UNALIGN_CTL(me, arg2);
2509 break;
2510 case PR_GET_UNALIGN:
2511 error = GET_UNALIGN_CTL(me, arg2);
2512 break;
2513 case PR_SET_FPEMU:
2514 error = SET_FPEMU_CTL(me, arg2);
2515 break;
2516 case PR_GET_FPEMU:
2517 error = GET_FPEMU_CTL(me, arg2);
2518 break;
2519 case PR_SET_FPEXC:
2520 error = SET_FPEXC_CTL(me, arg2);
2521 break;
2522 case PR_GET_FPEXC:
2523 error = GET_FPEXC_CTL(me, arg2);
2524 break;
2525 case PR_GET_TIMING:
2526 error = PR_TIMING_STATISTICAL;
2527 break;
2528 case PR_SET_TIMING:
2529 if (arg2 != PR_TIMING_STATISTICAL)
2530 error = -EINVAL;
2531 break;
2532 case PR_SET_NAME:
2533 comm[sizeof(me->comm) - 1] = 0;
2534 if (strncpy_from_user(comm, (char __user *)arg2,
2535 sizeof(me->comm) - 1) < 0)
2536 return -EFAULT;
2537 set_task_comm(me, comm);
2538 proc_comm_connector(me);
2539 break;
2540 case PR_GET_NAME:
2541 get_task_comm(comm, me);
2542 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2543 return -EFAULT;
2544 break;
2545 case PR_GET_ENDIAN:
2546 error = GET_ENDIAN(me, arg2);
2547 break;
2548 case PR_SET_ENDIAN:
2549 error = SET_ENDIAN(me, arg2);
2550 break;
2551 case PR_GET_SECCOMP:
2552 error = prctl_get_seccomp();
2553 break;
2554 case PR_SET_SECCOMP:
2555 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2556 break;
2557 case PR_GET_TSC:
2558 error = GET_TSC_CTL(arg2);
2559 break;
2560 case PR_SET_TSC:
2561 error = SET_TSC_CTL(arg2);
2562 break;
2563 case PR_TASK_PERF_EVENTS_DISABLE:
2564 error = perf_event_task_disable();
2565 break;
2566 case PR_TASK_PERF_EVENTS_ENABLE:
2567 error = perf_event_task_enable();
2568 break;
2569 case PR_GET_TIMERSLACK:
2570 if (current->timer_slack_ns > ULONG_MAX)
2571 error = ULONG_MAX;
2572 else
2573 error = current->timer_slack_ns;
2574 break;
2575 case PR_SET_TIMERSLACK:
2576 if (rt_or_dl_task_policy(current))
2577 break;
2578 if (arg2 <= 0)
2579 current->timer_slack_ns =
2580 current->default_timer_slack_ns;
2581 else
2582 current->timer_slack_ns = arg2;
2583 break;
2584 case PR_MCE_KILL:
2585 if (arg4 | arg5)
2586 return -EINVAL;
2587 switch (arg2) {
2588 case PR_MCE_KILL_CLEAR:
2589 if (arg3 != 0)
2590 return -EINVAL;
2591 current->flags &= ~PF_MCE_PROCESS;
2592 break;
2593 case PR_MCE_KILL_SET:
2594 current->flags |= PF_MCE_PROCESS;
2595 if (arg3 == PR_MCE_KILL_EARLY)
2596 current->flags |= PF_MCE_EARLY;
2597 else if (arg3 == PR_MCE_KILL_LATE)
2598 current->flags &= ~PF_MCE_EARLY;
2599 else if (arg3 == PR_MCE_KILL_DEFAULT)
2600 current->flags &=
2601 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2602 else
2603 return -EINVAL;
2604 break;
2605 default:
2606 return -EINVAL;
2607 }
2608 break;
2609 case PR_MCE_KILL_GET:
2610 if (arg2 | arg3 | arg4 | arg5)
2611 return -EINVAL;
2612 if (current->flags & PF_MCE_PROCESS)
2613 error = (current->flags & PF_MCE_EARLY) ?
2614 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2615 else
2616 error = PR_MCE_KILL_DEFAULT;
2617 break;
2618 case PR_SET_MM:
2619 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2620 break;
2621 case PR_GET_TID_ADDRESS:
2622 error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2623 break;
2624 case PR_SET_CHILD_SUBREAPER:
2625 me->signal->is_child_subreaper = !!arg2;
2626 if (!arg2)
2627 break;
2628
2629 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2630 break;
2631 case PR_GET_CHILD_SUBREAPER:
2632 error = put_user(me->signal->is_child_subreaper,
2633 (int __user *)arg2);
2634 break;
2635 case PR_SET_NO_NEW_PRIVS:
2636 if (arg2 != 1 || arg3 || arg4 || arg5)
2637 return -EINVAL;
2638
2639 task_set_no_new_privs(current);
2640 break;
2641 case PR_GET_NO_NEW_PRIVS:
2642 if (arg2 || arg3 || arg4 || arg5)
2643 return -EINVAL;
2644 return task_no_new_privs(current) ? 1 : 0;
2645 case PR_GET_THP_DISABLE:
2646 if (arg2 || arg3 || arg4 || arg5)
2647 return -EINVAL;
2648 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2649 break;
2650 case PR_SET_THP_DISABLE:
2651 if (arg3 || arg4 || arg5)
2652 return -EINVAL;
2653 if (mmap_write_lock_killable(me->mm))
2654 return -EINTR;
2655 if (arg2)
2656 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2657 else
2658 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2659 mmap_write_unlock(me->mm);
2660 break;
2661 case PR_MPX_ENABLE_MANAGEMENT:
2662 case PR_MPX_DISABLE_MANAGEMENT:
2663 /* No longer implemented: */
2664 return -EINVAL;
2665 case PR_SET_FP_MODE:
2666 error = SET_FP_MODE(me, arg2);
2667 break;
2668 case PR_GET_FP_MODE:
2669 error = GET_FP_MODE(me);
2670 break;
2671 case PR_SVE_SET_VL:
2672 error = SVE_SET_VL(arg2);
2673 break;
2674 case PR_SVE_GET_VL:
2675 error = SVE_GET_VL();
2676 break;
2677 case PR_SME_SET_VL:
2678 error = SME_SET_VL(arg2);
2679 break;
2680 case PR_SME_GET_VL:
2681 error = SME_GET_VL();
2682 break;
2683 case PR_GET_SPECULATION_CTRL:
2684 if (arg3 || arg4 || arg5)
2685 return -EINVAL;
2686 error = arch_prctl_spec_ctrl_get(me, arg2);
2687 break;
2688 case PR_SET_SPECULATION_CTRL:
2689 if (arg4 || arg5)
2690 return -EINVAL;
2691 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2692 break;
2693 case PR_PAC_RESET_KEYS:
2694 if (arg3 || arg4 || arg5)
2695 return -EINVAL;
2696 error = PAC_RESET_KEYS(me, arg2);
2697 break;
2698 case PR_PAC_SET_ENABLED_KEYS:
2699 if (arg4 || arg5)
2700 return -EINVAL;
2701 error = PAC_SET_ENABLED_KEYS(me, arg2, arg3);
2702 break;
2703 case PR_PAC_GET_ENABLED_KEYS:
2704 if (arg2 || arg3 || arg4 || arg5)
2705 return -EINVAL;
2706 error = PAC_GET_ENABLED_KEYS(me);
2707 break;
2708 case PR_SET_TAGGED_ADDR_CTRL:
2709 if (arg3 || arg4 || arg5)
2710 return -EINVAL;
2711 error = SET_TAGGED_ADDR_CTRL(arg2);
2712 break;
2713 case PR_GET_TAGGED_ADDR_CTRL:
2714 if (arg2 || arg3 || arg4 || arg5)
2715 return -EINVAL;
2716 error = GET_TAGGED_ADDR_CTRL();
2717 break;
2718 case PR_SET_IO_FLUSHER:
2719 if (!capable(CAP_SYS_RESOURCE))
2720 return -EPERM;
2721
2722 if (arg3 || arg4 || arg5)
2723 return -EINVAL;
2724
2725 if (arg2 == 1)
2726 current->flags |= PR_IO_FLUSHER;
2727 else if (!arg2)
2728 current->flags &= ~PR_IO_FLUSHER;
2729 else
2730 return -EINVAL;
2731 break;
2732 case PR_GET_IO_FLUSHER:
2733 if (!capable(CAP_SYS_RESOURCE))
2734 return -EPERM;
2735
2736 if (arg2 || arg3 || arg4 || arg5)
2737 return -EINVAL;
2738
2739 error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2740 break;
2741 case PR_SET_SYSCALL_USER_DISPATCH:
2742 error = set_syscall_user_dispatch(arg2, arg3, arg4,
2743 (char __user *) arg5);
2744 break;
2745 #ifdef CONFIG_SCHED_CORE
2746 case PR_SCHED_CORE:
2747 error = sched_core_share_pid(arg2, arg3, arg4, arg5);
2748 break;
2749 #endif
2750 case PR_SET_MDWE:
2751 error = prctl_set_mdwe(arg2, arg3, arg4, arg5);
2752 break;
2753 case PR_GET_MDWE:
2754 error = prctl_get_mdwe(arg2, arg3, arg4, arg5);
2755 break;
2756 case PR_PPC_GET_DEXCR:
2757 if (arg3 || arg4 || arg5)
2758 return -EINVAL;
2759 error = PPC_GET_DEXCR_ASPECT(me, arg2);
2760 break;
2761 case PR_PPC_SET_DEXCR:
2762 if (arg4 || arg5)
2763 return -EINVAL;
2764 error = PPC_SET_DEXCR_ASPECT(me, arg2, arg3);
2765 break;
2766 case PR_SET_VMA:
2767 error = prctl_set_vma(arg2, arg3, arg4, arg5);
2768 break;
2769 case PR_GET_AUXV:
2770 if (arg4 || arg5)
2771 return -EINVAL;
2772 error = prctl_get_auxv((void __user *)arg2, arg3);
2773 break;
2774 #ifdef CONFIG_KSM
2775 case PR_SET_MEMORY_MERGE:
2776 if (arg3 || arg4 || arg5)
2777 return -EINVAL;
2778 if (mmap_write_lock_killable(me->mm))
2779 return -EINTR;
2780
2781 if (arg2)
2782 error = ksm_enable_merge_any(me->mm);
2783 else
2784 error = ksm_disable_merge_any(me->mm);
2785 mmap_write_unlock(me->mm);
2786 break;
2787 case PR_GET_MEMORY_MERGE:
2788 if (arg2 || arg3 || arg4 || arg5)
2789 return -EINVAL;
2790
2791 error = !!test_bit(MMF_VM_MERGE_ANY, &me->mm->flags);
2792 break;
2793 #endif
2794 case PR_RISCV_V_SET_CONTROL:
2795 error = RISCV_V_SET_CONTROL(arg2);
2796 break;
2797 case PR_RISCV_V_GET_CONTROL:
2798 error = RISCV_V_GET_CONTROL();
2799 break;
2800 case PR_RISCV_SET_ICACHE_FLUSH_CTX:
2801 error = RISCV_SET_ICACHE_FLUSH_CTX(arg2, arg3);
2802 break;
2803 case PR_GET_SHADOW_STACK_STATUS:
2804 if (arg3 || arg4 || arg5)
2805 return -EINVAL;
2806 error = arch_get_shadow_stack_status(me, (unsigned long __user *) arg2);
2807 break;
2808 case PR_SET_SHADOW_STACK_STATUS:
2809 if (arg3 || arg4 || arg5)
2810 return -EINVAL;
2811 error = arch_set_shadow_stack_status(me, arg2);
2812 break;
2813 case PR_LOCK_SHADOW_STACK_STATUS:
2814 if (arg3 || arg4 || arg5)
2815 return -EINVAL;
2816 error = arch_lock_shadow_stack_status(me, arg2);
2817 break;
2818 case PR_TIMER_CREATE_RESTORE_IDS:
2819 if (arg3 || arg4 || arg5)
2820 return -EINVAL;
2821 error = posixtimer_create_prctl(arg2);
2822 break;
2823 default:
2824 trace_task_prctl_unknown(option, arg2, arg3, arg4, arg5);
2825 error = -EINVAL;
2826 break;
2827 }
2828 return error;
2829 }
2830
SYSCALL_DEFINE3(getcpu,unsigned __user *,cpup,unsigned __user *,nodep,struct getcpu_cache __user *,unused)2831 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2832 struct getcpu_cache __user *, unused)
2833 {
2834 int err = 0;
2835 int cpu = raw_smp_processor_id();
2836
2837 if (cpup)
2838 err |= put_user(cpu, cpup);
2839 if (nodep)
2840 err |= put_user(cpu_to_node(cpu), nodep);
2841 return err ? -EFAULT : 0;
2842 }
2843
2844 /**
2845 * do_sysinfo - fill in sysinfo struct
2846 * @info: pointer to buffer to fill
2847 */
do_sysinfo(struct sysinfo * info)2848 static int do_sysinfo(struct sysinfo *info)
2849 {
2850 unsigned long mem_total, sav_total;
2851 unsigned int mem_unit, bitcount;
2852 struct timespec64 tp;
2853
2854 memset(info, 0, sizeof(struct sysinfo));
2855
2856 ktime_get_boottime_ts64(&tp);
2857 timens_add_boottime(&tp);
2858 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2859
2860 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2861
2862 info->procs = nr_threads;
2863
2864 si_meminfo(info);
2865 si_swapinfo(info);
2866
2867 /*
2868 * If the sum of all the available memory (i.e. ram + swap)
2869 * is less than can be stored in a 32 bit unsigned long then
2870 * we can be binary compatible with 2.2.x kernels. If not,
2871 * well, in that case 2.2.x was broken anyways...
2872 *
2873 * -Erik Andersen <andersee@debian.org>
2874 */
2875
2876 mem_total = info->totalram + info->totalswap;
2877 if (mem_total < info->totalram || mem_total < info->totalswap)
2878 goto out;
2879 bitcount = 0;
2880 mem_unit = info->mem_unit;
2881 while (mem_unit > 1) {
2882 bitcount++;
2883 mem_unit >>= 1;
2884 sav_total = mem_total;
2885 mem_total <<= 1;
2886 if (mem_total < sav_total)
2887 goto out;
2888 }
2889
2890 /*
2891 * If mem_total did not overflow, multiply all memory values by
2892 * info->mem_unit and set it to 1. This leaves things compatible
2893 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2894 * kernels...
2895 */
2896
2897 info->mem_unit = 1;
2898 info->totalram <<= bitcount;
2899 info->freeram <<= bitcount;
2900 info->sharedram <<= bitcount;
2901 info->bufferram <<= bitcount;
2902 info->totalswap <<= bitcount;
2903 info->freeswap <<= bitcount;
2904 info->totalhigh <<= bitcount;
2905 info->freehigh <<= bitcount;
2906
2907 out:
2908 return 0;
2909 }
2910
SYSCALL_DEFINE1(sysinfo,struct sysinfo __user *,info)2911 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2912 {
2913 struct sysinfo val;
2914
2915 do_sysinfo(&val);
2916
2917 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2918 return -EFAULT;
2919
2920 return 0;
2921 }
2922
2923 #ifdef CONFIG_COMPAT
2924 struct compat_sysinfo {
2925 s32 uptime;
2926 u32 loads[3];
2927 u32 totalram;
2928 u32 freeram;
2929 u32 sharedram;
2930 u32 bufferram;
2931 u32 totalswap;
2932 u32 freeswap;
2933 u16 procs;
2934 u16 pad;
2935 u32 totalhigh;
2936 u32 freehigh;
2937 u32 mem_unit;
2938 char _f[20-2*sizeof(u32)-sizeof(int)];
2939 };
2940
COMPAT_SYSCALL_DEFINE1(sysinfo,struct compat_sysinfo __user *,info)2941 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2942 {
2943 struct sysinfo s;
2944 struct compat_sysinfo s_32;
2945
2946 do_sysinfo(&s);
2947
2948 /* Check to see if any memory value is too large for 32-bit and scale
2949 * down if needed
2950 */
2951 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2952 int bitcount = 0;
2953
2954 while (s.mem_unit < PAGE_SIZE) {
2955 s.mem_unit <<= 1;
2956 bitcount++;
2957 }
2958
2959 s.totalram >>= bitcount;
2960 s.freeram >>= bitcount;
2961 s.sharedram >>= bitcount;
2962 s.bufferram >>= bitcount;
2963 s.totalswap >>= bitcount;
2964 s.freeswap >>= bitcount;
2965 s.totalhigh >>= bitcount;
2966 s.freehigh >>= bitcount;
2967 }
2968
2969 memset(&s_32, 0, sizeof(s_32));
2970 s_32.uptime = s.uptime;
2971 s_32.loads[0] = s.loads[0];
2972 s_32.loads[1] = s.loads[1];
2973 s_32.loads[2] = s.loads[2];
2974 s_32.totalram = s.totalram;
2975 s_32.freeram = s.freeram;
2976 s_32.sharedram = s.sharedram;
2977 s_32.bufferram = s.bufferram;
2978 s_32.totalswap = s.totalswap;
2979 s_32.freeswap = s.freeswap;
2980 s_32.procs = s.procs;
2981 s_32.totalhigh = s.totalhigh;
2982 s_32.freehigh = s.freehigh;
2983 s_32.mem_unit = s.mem_unit;
2984 if (copy_to_user(info, &s_32, sizeof(s_32)))
2985 return -EFAULT;
2986 return 0;
2987 }
2988 #endif /* CONFIG_COMPAT */
2989