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