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