• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  *  linux/kernel/sys.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 #include <linux/export.h>
8 #include <linux/mm.h>
9 #include <linux/utsname.h>
10 #include <linux/mman.h>
11 #include <linux/reboot.h>
12 #include <linux/prctl.h>
13 #include <linux/highuid.h>
14 #include <linux/fs.h>
15 #include <linux/kmod.h>
16 #include <linux/perf_event.h>
17 #include <linux/resource.h>
18 #include <linux/kernel.h>
19 #include <linux/kexec.h>
20 #include <linux/workqueue.h>
21 #include <linux/capability.h>
22 #include <linux/device.h>
23 #include <linux/key.h>
24 #include <linux/times.h>
25 #include <linux/posix-timers.h>
26 #include <linux/security.h>
27 #include <linux/dcookies.h>
28 #include <linux/suspend.h>
29 #include <linux/tty.h>
30 #include <linux/signal.h>
31 #include <linux/cn_proc.h>
32 #include <linux/getcpu.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/seccomp.h>
35 #include <linux/cpu.h>
36 #include <linux/personality.h>
37 #include <linux/ptrace.h>
38 #include <linux/fs_struct.h>
39 #include <linux/file.h>
40 #include <linux/mount.h>
41 #include <linux/gfp.h>
42 #include <linux/syscore_ops.h>
43 #include <linux/version.h>
44 #include <linux/ctype.h>
45 #include <linux/mm.h>
46 #include <linux/mempolicy.h>
47 #include <linux/sched.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/binfmts.h>
54 
55 #include <linux/sched.h>
56 #include <linux/rcupdate.h>
57 #include <linux/uidgid.h>
58 #include <linux/cred.h>
59 
60 #include <linux/kmsg_dump.h>
61 /* Move somewhere else to avoid recompiling? */
62 #include <generated/utsrelease.h>
63 
64 #include <asm/uaccess.h>
65 #include <asm/io.h>
66 #include <asm/unistd.h>
67 
68 #ifndef SET_UNALIGN_CTL
69 # define SET_UNALIGN_CTL(a,b)	(-EINVAL)
70 #endif
71 #ifndef GET_UNALIGN_CTL
72 # define GET_UNALIGN_CTL(a,b)	(-EINVAL)
73 #endif
74 #ifndef SET_FPEMU_CTL
75 # define SET_FPEMU_CTL(a,b)	(-EINVAL)
76 #endif
77 #ifndef GET_FPEMU_CTL
78 # define GET_FPEMU_CTL(a,b)	(-EINVAL)
79 #endif
80 #ifndef SET_FPEXC_CTL
81 # define SET_FPEXC_CTL(a,b)	(-EINVAL)
82 #endif
83 #ifndef GET_FPEXC_CTL
84 # define GET_FPEXC_CTL(a,b)	(-EINVAL)
85 #endif
86 #ifndef GET_ENDIAN
87 # define GET_ENDIAN(a,b)	(-EINVAL)
88 #endif
89 #ifndef SET_ENDIAN
90 # define SET_ENDIAN(a,b)	(-EINVAL)
91 #endif
92 #ifndef GET_TSC_CTL
93 # define GET_TSC_CTL(a)		(-EINVAL)
94 #endif
95 #ifndef SET_TSC_CTL
96 # define SET_TSC_CTL(a)		(-EINVAL)
97 #endif
98 
99 #ifndef GET_FP_MODE
100 # define GET_FP_MODE(a)         (-EINVAL)
101 #endif
102 #ifndef SET_FP_MODE
103 # define SET_FP_MODE(a,b)       (-EINVAL)
104 #endif
105 
106 /*
107  * this is where the system-wide overflow UID and GID are defined, for
108  * architectures that now have 32-bit UID/GID but didn't in the past
109  */
110 
111 int overflowuid = DEFAULT_OVERFLOWUID;
112 int overflowgid = DEFAULT_OVERFLOWGID;
113 
114 EXPORT_SYMBOL(overflowuid);
115 EXPORT_SYMBOL(overflowgid);
116 
117 /*
118  * the same as above, but for filesystems which can only store a 16-bit
119  * UID and GID. as such, this is needed on all architectures
120  */
121 
122 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
123 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
124 
125 EXPORT_SYMBOL(fs_overflowuid);
126 EXPORT_SYMBOL(fs_overflowgid);
127 
128 /*
129  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
130  */
131 
132 int C_A_D = 1;
133 struct pid *cad_pid;
134 EXPORT_SYMBOL(cad_pid);
135 
136 /*
137  * If set, this is used for preparing the system to power off.
138  */
139 
140 void (*pm_power_off_prepare)(void);
141 
142 /*
143  * Returns true if current's euid is same as p's uid or euid,
144  * or has CAP_SYS_NICE to p's user_ns.
145  *
146  * Called with rcu_read_lock, creds are safe
147  */
set_one_prio_perm(struct task_struct * p)148 static bool set_one_prio_perm(struct task_struct *p)
149 {
150 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
151 
152 	if (uid_eq(pcred->uid,  cred->euid) ||
153 	    uid_eq(pcred->euid, cred->euid))
154 		return true;
155 	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
156 		return true;
157 	return false;
158 }
159 
160 /*
161  * set the priority of a task
162  * - the caller must hold the RCU read lock
163  */
set_one_prio(struct task_struct * p,int niceval,int error)164 static int set_one_prio(struct task_struct *p, int niceval, int error)
165 {
166 	int no_nice;
167 
168 	if (!set_one_prio_perm(p)) {
169 		error = -EPERM;
170 		goto out;
171 	}
172 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
173 		error = -EACCES;
174 		goto out;
175 	}
176 	no_nice = security_task_setnice(p, niceval);
177 	if (no_nice) {
178 		error = no_nice;
179 		goto out;
180 	}
181 	if (error == -ESRCH)
182 		error = 0;
183 	set_user_nice(p, niceval);
184 out:
185 	return error;
186 }
187 
SYSCALL_DEFINE3(setpriority,int,which,int,who,int,niceval)188 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
189 {
190 	struct task_struct *g, *p;
191 	struct user_struct *user;
192 	const struct cred *cred = current_cred();
193 	int error = -EINVAL;
194 	struct pid *pgrp;
195 	kuid_t uid;
196 
197 	if (which > PRIO_USER || which < PRIO_PROCESS)
198 		goto out;
199 
200 	/* normalize: avoid signed division (rounding problems) */
201 	error = -ESRCH;
202 	if (niceval < -20)
203 		niceval = -20;
204 	if (niceval > 19)
205 		niceval = 19;
206 
207 	rcu_read_lock();
208 	read_lock(&tasklist_lock);
209 	switch (which) {
210 		case PRIO_PROCESS:
211 			if (who)
212 				p = find_task_by_vpid(who);
213 			else
214 				p = current;
215 			if (p)
216 				error = set_one_prio(p, niceval, error);
217 			break;
218 		case PRIO_PGRP:
219 			if (who)
220 				pgrp = find_vpid(who);
221 			else
222 				pgrp = task_pgrp(current);
223 			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
224 				error = set_one_prio(p, niceval, error);
225 			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
226 			break;
227 		case PRIO_USER:
228 			uid = make_kuid(cred->user_ns, who);
229 			user = cred->user;
230 			if (!who)
231 				uid = cred->uid;
232 			else if (!uid_eq(uid, cred->uid) &&
233 				 !(user = find_user(uid)))
234 				goto out_unlock;	/* No processes for this user */
235 
236 			do_each_thread(g, p) {
237 				if (uid_eq(task_uid(p), uid))
238 					error = set_one_prio(p, niceval, error);
239 			} while_each_thread(g, p);
240 			if (!uid_eq(uid, cred->uid))
241 				free_uid(user);		/* For find_user() */
242 			break;
243 	}
244 out_unlock:
245 	read_unlock(&tasklist_lock);
246 	rcu_read_unlock();
247 out:
248 	return error;
249 }
250 
251 /*
252  * Ugh. To avoid negative return values, "getpriority()" will
253  * not return the normal nice-value, but a negated value that
254  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
255  * to stay compatible.
256  */
SYSCALL_DEFINE2(getpriority,int,which,int,who)257 SYSCALL_DEFINE2(getpriority, int, which, int, who)
258 {
259 	struct task_struct *g, *p;
260 	struct user_struct *user;
261 	const struct cred *cred = current_cred();
262 	long niceval, retval = -ESRCH;
263 	struct pid *pgrp;
264 	kuid_t uid;
265 
266 	if (which > PRIO_USER || which < PRIO_PROCESS)
267 		return -EINVAL;
268 
269 	rcu_read_lock();
270 	read_lock(&tasklist_lock);
271 	switch (which) {
272 		case PRIO_PROCESS:
273 			if (who)
274 				p = find_task_by_vpid(who);
275 			else
276 				p = current;
277 			if (p) {
278 				niceval = 20 - task_nice(p);
279 				if (niceval > retval)
280 					retval = niceval;
281 			}
282 			break;
283 		case PRIO_PGRP:
284 			if (who)
285 				pgrp = find_vpid(who);
286 			else
287 				pgrp = task_pgrp(current);
288 			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
289 				niceval = 20 - task_nice(p);
290 				if (niceval > retval)
291 					retval = niceval;
292 			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
293 			break;
294 		case PRIO_USER:
295 			uid = make_kuid(cred->user_ns, who);
296 			user = cred->user;
297 			if (!who)
298 				uid = cred->uid;
299 			else if (!uid_eq(uid, cred->uid) &&
300 				 !(user = find_user(uid)))
301 				goto out_unlock;	/* No processes for this user */
302 
303 			do_each_thread(g, p) {
304 				if (uid_eq(task_uid(p), uid)) {
305 					niceval = 20 - task_nice(p);
306 					if (niceval > retval)
307 						retval = niceval;
308 				}
309 			} while_each_thread(g, p);
310 			if (!uid_eq(uid, cred->uid))
311 				free_uid(user);		/* for find_user() */
312 			break;
313 	}
314 out_unlock:
315 	read_unlock(&tasklist_lock);
316 	rcu_read_unlock();
317 
318 	return retval;
319 }
320 
321 /**
322  *	emergency_restart - reboot the system
323  *
324  *	Without shutting down any hardware or taking any locks
325  *	reboot the system.  This is called when we know we are in
326  *	trouble so this is our best effort to reboot.  This is
327  *	safe to call in interrupt context.
328  */
emergency_restart(void)329 void emergency_restart(void)
330 {
331 	kmsg_dump(KMSG_DUMP_EMERG);
332 	machine_emergency_restart();
333 }
334 EXPORT_SYMBOL_GPL(emergency_restart);
335 
kernel_restart_prepare(char * cmd)336 void kernel_restart_prepare(char *cmd)
337 {
338 	blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
339 	system_state = SYSTEM_RESTART;
340 	usermodehelper_disable();
341 	device_shutdown();
342 }
343 
344 /**
345  *	register_reboot_notifier - Register function to be called at reboot time
346  *	@nb: Info about notifier function to be called
347  *
348  *	Registers a function with the list of functions
349  *	to be called at reboot time.
350  *
351  *	Currently always returns zero, as blocking_notifier_chain_register()
352  *	always returns zero.
353  */
register_reboot_notifier(struct notifier_block * nb)354 int register_reboot_notifier(struct notifier_block *nb)
355 {
356 	return blocking_notifier_chain_register(&reboot_notifier_list, nb);
357 }
358 EXPORT_SYMBOL(register_reboot_notifier);
359 
360 /**
361  *	unregister_reboot_notifier - Unregister previously registered reboot notifier
362  *	@nb: Hook to be unregistered
363  *
364  *	Unregisters a previously registered reboot
365  *	notifier function.
366  *
367  *	Returns zero on success, or %-ENOENT on failure.
368  */
unregister_reboot_notifier(struct notifier_block * nb)369 int unregister_reboot_notifier(struct notifier_block *nb)
370 {
371 	return blocking_notifier_chain_unregister(&reboot_notifier_list, nb);
372 }
373 EXPORT_SYMBOL(unregister_reboot_notifier);
374 
375 /* Add backwards compatibility for stable trees. */
376 #ifndef PF_NO_SETAFFINITY
377 #define PF_NO_SETAFFINITY		PF_THREAD_BOUND
378 #endif
379 
migrate_to_reboot_cpu(void)380 static void migrate_to_reboot_cpu(void)
381 {
382 	/* The boot cpu is always logical cpu 0 */
383 	int cpu = 0;
384 
385 	cpu_hotplug_disable();
386 
387 	/* Make certain the cpu I'm about to reboot on is online */
388 	if (!cpu_online(cpu))
389 		cpu = cpumask_first(cpu_online_mask);
390 
391 	/* Prevent races with other tasks migrating this task */
392 	current->flags |= PF_NO_SETAFFINITY;
393 
394 	/* Make certain I only run on the appropriate processor */
395 	set_cpus_allowed_ptr(current, cpumask_of(cpu));
396 }
397 
398 /**
399  *	kernel_restart - reboot the system
400  *	@cmd: pointer to buffer containing command to execute for restart
401  *		or %NULL
402  *
403  *	Shutdown everything and perform a clean reboot.
404  *	This is not safe to call in interrupt context.
405  */
kernel_restart(char * cmd)406 void kernel_restart(char *cmd)
407 {
408 	kernel_restart_prepare(cmd);
409 	migrate_to_reboot_cpu();
410 	syscore_shutdown();
411 	if (!cmd)
412 		printk(KERN_EMERG "Restarting system.\n");
413 	else
414 		printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
415 	kmsg_dump(KMSG_DUMP_RESTART);
416 	machine_restart(cmd);
417 }
418 EXPORT_SYMBOL_GPL(kernel_restart);
419 
kernel_shutdown_prepare(enum system_states state)420 static void kernel_shutdown_prepare(enum system_states state)
421 {
422 	blocking_notifier_call_chain(&reboot_notifier_list,
423 		(state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
424 	system_state = state;
425 	usermodehelper_disable();
426 	device_shutdown();
427 }
428 /**
429  *	kernel_halt - halt the system
430  *
431  *	Shutdown everything and perform a clean system halt.
432  */
kernel_halt(void)433 void kernel_halt(void)
434 {
435 	kernel_shutdown_prepare(SYSTEM_HALT);
436 	migrate_to_reboot_cpu();
437 	syscore_shutdown();
438 	printk(KERN_EMERG "System halted.\n");
439 	kmsg_dump(KMSG_DUMP_HALT);
440 	machine_halt();
441 }
442 
443 EXPORT_SYMBOL_GPL(kernel_halt);
444 
445 /**
446  *	kernel_power_off - power_off the system
447  *
448  *	Shutdown everything and perform a clean system power_off.
449  */
kernel_power_off(void)450 void kernel_power_off(void)
451 {
452 	kernel_shutdown_prepare(SYSTEM_POWER_OFF);
453 	if (pm_power_off_prepare)
454 		pm_power_off_prepare();
455 	migrate_to_reboot_cpu();
456 	syscore_shutdown();
457 	printk(KERN_EMERG "Power down.\n");
458 	kmsg_dump(KMSG_DUMP_POWEROFF);
459 	machine_power_off();
460 }
461 EXPORT_SYMBOL_GPL(kernel_power_off);
462 
463 static DEFINE_MUTEX(reboot_mutex);
464 
465 /*
466  * Reboot system call: for obvious reasons only root may call it,
467  * and even root needs to set up some magic numbers in the registers
468  * so that some mistake won't make this reboot the whole machine.
469  * You can also set the meaning of the ctrl-alt-del-key here.
470  *
471  * reboot doesn't sync: do that yourself before calling this.
472  */
SYSCALL_DEFINE4(reboot,int,magic1,int,magic2,unsigned int,cmd,void __user *,arg)473 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
474 		void __user *, arg)
475 {
476 	struct pid_namespace *pid_ns = task_active_pid_ns(current);
477 	char buffer[256];
478 	int ret = 0;
479 
480 	/* We only trust the superuser with rebooting the system. */
481 	if (!ns_capable(pid_ns->user_ns, CAP_SYS_BOOT))
482 		return -EPERM;
483 
484 	/* For safety, we require "magic" arguments. */
485 	if (magic1 != LINUX_REBOOT_MAGIC1 ||
486 	    (magic2 != LINUX_REBOOT_MAGIC2 &&
487 	                magic2 != LINUX_REBOOT_MAGIC2A &&
488 			magic2 != LINUX_REBOOT_MAGIC2B &&
489 	                magic2 != LINUX_REBOOT_MAGIC2C))
490 		return -EINVAL;
491 
492 	/*
493 	 * If pid namespaces are enabled and the current task is in a child
494 	 * pid_namespace, the command is handled by reboot_pid_ns() which will
495 	 * call do_exit().
496 	 */
497 	ret = reboot_pid_ns(pid_ns, cmd);
498 	if (ret)
499 		return ret;
500 
501 	/* Instead of trying to make the power_off code look like
502 	 * halt when pm_power_off is not set do it the easy way.
503 	 */
504 	if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
505 		cmd = LINUX_REBOOT_CMD_HALT;
506 
507 	mutex_lock(&reboot_mutex);
508 	switch (cmd) {
509 	case LINUX_REBOOT_CMD_RESTART:
510 		kernel_restart(NULL);
511 		break;
512 
513 	case LINUX_REBOOT_CMD_CAD_ON:
514 		C_A_D = 1;
515 		break;
516 
517 	case LINUX_REBOOT_CMD_CAD_OFF:
518 		C_A_D = 0;
519 		break;
520 
521 	case LINUX_REBOOT_CMD_HALT:
522 		kernel_halt();
523 		do_exit(0);
524 		panic("cannot halt");
525 
526 	case LINUX_REBOOT_CMD_POWER_OFF:
527 		kernel_power_off();
528 		do_exit(0);
529 		break;
530 
531 	case LINUX_REBOOT_CMD_RESTART2:
532 		if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
533 			ret = -EFAULT;
534 			break;
535 		}
536 		buffer[sizeof(buffer) - 1] = '\0';
537 
538 		kernel_restart(buffer);
539 		break;
540 
541 #ifdef CONFIG_KEXEC
542 	case LINUX_REBOOT_CMD_KEXEC:
543 		ret = kernel_kexec();
544 		break;
545 #endif
546 
547 #ifdef CONFIG_HIBERNATION
548 	case LINUX_REBOOT_CMD_SW_SUSPEND:
549 		ret = hibernate();
550 		break;
551 #endif
552 
553 	default:
554 		ret = -EINVAL;
555 		break;
556 	}
557 	mutex_unlock(&reboot_mutex);
558 	return ret;
559 }
560 
deferred_cad(struct work_struct * dummy)561 static void deferred_cad(struct work_struct *dummy)
562 {
563 	kernel_restart(NULL);
564 }
565 
566 /*
567  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
568  * As it's called within an interrupt, it may NOT sync: the only choice
569  * is whether to reboot at once, or just ignore the ctrl-alt-del.
570  */
ctrl_alt_del(void)571 void ctrl_alt_del(void)
572 {
573 	static DECLARE_WORK(cad_work, deferred_cad);
574 
575 	if (C_A_D)
576 		schedule_work(&cad_work);
577 	else
578 		kill_cad_pid(SIGINT, 1);
579 }
580 
581 /*
582  * Unprivileged users may change the real gid to the effective gid
583  * or vice versa.  (BSD-style)
584  *
585  * If you set the real gid at all, or set the effective gid to a value not
586  * equal to the real gid, then the saved gid is set to the new effective gid.
587  *
588  * This makes it possible for a setgid program to completely drop its
589  * privileges, which is often a useful assertion to make when you are doing
590  * a security audit over a program.
591  *
592  * The general idea is that a program which uses just setregid() will be
593  * 100% compatible with BSD.  A program which uses just setgid() will be
594  * 100% compatible with POSIX with saved IDs.
595  *
596  * SMP: There are not races, the GIDs are checked only by filesystem
597  *      operations (as far as semantic preservation is concerned).
598  */
SYSCALL_DEFINE2(setregid,gid_t,rgid,gid_t,egid)599 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
600 {
601 	struct user_namespace *ns = current_user_ns();
602 	const struct cred *old;
603 	struct cred *new;
604 	int retval;
605 	kgid_t krgid, kegid;
606 
607 	krgid = make_kgid(ns, rgid);
608 	kegid = make_kgid(ns, egid);
609 
610 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
611 		return -EINVAL;
612 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
613 		return -EINVAL;
614 
615 	new = prepare_creds();
616 	if (!new)
617 		return -ENOMEM;
618 	old = current_cred();
619 
620 	retval = -EPERM;
621 	if (rgid != (gid_t) -1) {
622 		if (gid_eq(old->gid, krgid) ||
623 		    gid_eq(old->egid, krgid) ||
624 		    nsown_capable(CAP_SETGID))
625 			new->gid = krgid;
626 		else
627 			goto error;
628 	}
629 	if (egid != (gid_t) -1) {
630 		if (gid_eq(old->gid, kegid) ||
631 		    gid_eq(old->egid, kegid) ||
632 		    gid_eq(old->sgid, kegid) ||
633 		    nsown_capable(CAP_SETGID))
634 			new->egid = kegid;
635 		else
636 			goto error;
637 	}
638 
639 	if (rgid != (gid_t) -1 ||
640 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
641 		new->sgid = new->egid;
642 	new->fsgid = new->egid;
643 
644 	return commit_creds(new);
645 
646 error:
647 	abort_creds(new);
648 	return retval;
649 }
650 
651 /*
652  * setgid() is implemented like SysV w/ SAVED_IDS
653  *
654  * SMP: Same implicit races as above.
655  */
SYSCALL_DEFINE1(setgid,gid_t,gid)656 SYSCALL_DEFINE1(setgid, gid_t, gid)
657 {
658 	struct user_namespace *ns = current_user_ns();
659 	const struct cred *old;
660 	struct cred *new;
661 	int retval;
662 	kgid_t kgid;
663 
664 	kgid = make_kgid(ns, gid);
665 	if (!gid_valid(kgid))
666 		return -EINVAL;
667 
668 	new = prepare_creds();
669 	if (!new)
670 		return -ENOMEM;
671 	old = current_cred();
672 
673 	retval = -EPERM;
674 	if (nsown_capable(CAP_SETGID))
675 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
676 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
677 		new->egid = new->fsgid = kgid;
678 	else
679 		goto error;
680 
681 	return commit_creds(new);
682 
683 error:
684 	abort_creds(new);
685 	return retval;
686 }
687 
688 /*
689  * change the user struct in a credentials set to match the new UID
690  */
set_user(struct cred * new)691 static int set_user(struct cred *new)
692 {
693 	struct user_struct *new_user;
694 
695 	new_user = alloc_uid(new->uid);
696 	if (!new_user)
697 		return -EAGAIN;
698 
699 	/*
700 	 * We don't fail in case of NPROC limit excess here because too many
701 	 * poorly written programs don't check set*uid() return code, assuming
702 	 * it never fails if called by root.  We may still enforce NPROC limit
703 	 * for programs doing set*uid()+execve() by harmlessly deferring the
704 	 * failure to the execve() stage.
705 	 */
706 	if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
707 			new_user != INIT_USER)
708 		current->flags |= PF_NPROC_EXCEEDED;
709 	else
710 		current->flags &= ~PF_NPROC_EXCEEDED;
711 
712 	free_uid(new->user);
713 	new->user = new_user;
714 	return 0;
715 }
716 
717 /*
718  * Unprivileged users may change the real uid to the effective uid
719  * or vice versa.  (BSD-style)
720  *
721  * If you set the real uid at all, or set the effective uid to a value not
722  * equal to the real uid, then the saved uid is set to the new effective uid.
723  *
724  * This makes it possible for a setuid program to completely drop its
725  * privileges, which is often a useful assertion to make when you are doing
726  * a security audit over a program.
727  *
728  * The general idea is that a program which uses just setreuid() will be
729  * 100% compatible with BSD.  A program which uses just setuid() will be
730  * 100% compatible with POSIX with saved IDs.
731  */
SYSCALL_DEFINE2(setreuid,uid_t,ruid,uid_t,euid)732 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
733 {
734 	struct user_namespace *ns = current_user_ns();
735 	const struct cred *old;
736 	struct cred *new;
737 	int retval;
738 	kuid_t kruid, keuid;
739 
740 	kruid = make_kuid(ns, ruid);
741 	keuid = make_kuid(ns, euid);
742 
743 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
744 		return -EINVAL;
745 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
746 		return -EINVAL;
747 
748 	new = prepare_creds();
749 	if (!new)
750 		return -ENOMEM;
751 	old = current_cred();
752 
753 	retval = -EPERM;
754 	if (ruid != (uid_t) -1) {
755 		new->uid = kruid;
756 		if (!uid_eq(old->uid, kruid) &&
757 		    !uid_eq(old->euid, kruid) &&
758 		    !nsown_capable(CAP_SETUID))
759 			goto error;
760 	}
761 
762 	if (euid != (uid_t) -1) {
763 		new->euid = keuid;
764 		if (!uid_eq(old->uid, keuid) &&
765 		    !uid_eq(old->euid, keuid) &&
766 		    !uid_eq(old->suid, keuid) &&
767 		    !nsown_capable(CAP_SETUID))
768 			goto error;
769 	}
770 
771 	if (!uid_eq(new->uid, old->uid)) {
772 		retval = set_user(new);
773 		if (retval < 0)
774 			goto error;
775 	}
776 	if (ruid != (uid_t) -1 ||
777 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
778 		new->suid = new->euid;
779 	new->fsuid = new->euid;
780 
781 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
782 	if (retval < 0)
783 		goto error;
784 
785 	return commit_creds(new);
786 
787 error:
788 	abort_creds(new);
789 	return retval;
790 }
791 
792 /*
793  * setuid() is implemented like SysV with SAVED_IDS
794  *
795  * Note that SAVED_ID's is deficient in that a setuid root program
796  * like sendmail, for example, cannot set its uid to be a normal
797  * user and then switch back, because if you're root, setuid() sets
798  * the saved uid too.  If you don't like this, blame the bright people
799  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
800  * will allow a root program to temporarily drop privileges and be able to
801  * regain them by swapping the real and effective uid.
802  */
SYSCALL_DEFINE1(setuid,uid_t,uid)803 SYSCALL_DEFINE1(setuid, uid_t, uid)
804 {
805 	struct user_namespace *ns = current_user_ns();
806 	const struct cred *old;
807 	struct cred *new;
808 	int retval;
809 	kuid_t kuid;
810 
811 	kuid = make_kuid(ns, uid);
812 	if (!uid_valid(kuid))
813 		return -EINVAL;
814 
815 	new = prepare_creds();
816 	if (!new)
817 		return -ENOMEM;
818 	old = current_cred();
819 
820 	retval = -EPERM;
821 	if (nsown_capable(CAP_SETUID)) {
822 		new->suid = new->uid = kuid;
823 		if (!uid_eq(kuid, old->uid)) {
824 			retval = set_user(new);
825 			if (retval < 0)
826 				goto error;
827 		}
828 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
829 		goto error;
830 	}
831 
832 	new->fsuid = new->euid = kuid;
833 
834 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
835 	if (retval < 0)
836 		goto error;
837 
838 	return commit_creds(new);
839 
840 error:
841 	abort_creds(new);
842 	return retval;
843 }
844 
845 
846 /*
847  * This function implements a generic ability to update ruid, euid,
848  * and suid.  This allows you to implement the 4.4 compatible seteuid().
849  */
SYSCALL_DEFINE3(setresuid,uid_t,ruid,uid_t,euid,uid_t,suid)850 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
851 {
852 	struct user_namespace *ns = current_user_ns();
853 	const struct cred *old;
854 	struct cred *new;
855 	int retval;
856 	kuid_t kruid, keuid, ksuid;
857 
858 	kruid = make_kuid(ns, ruid);
859 	keuid = make_kuid(ns, euid);
860 	ksuid = make_kuid(ns, suid);
861 
862 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
863 		return -EINVAL;
864 
865 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
866 		return -EINVAL;
867 
868 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
869 		return -EINVAL;
870 
871 	new = prepare_creds();
872 	if (!new)
873 		return -ENOMEM;
874 
875 	old = current_cred();
876 
877 	retval = -EPERM;
878 	if (!nsown_capable(CAP_SETUID)) {
879 		if (ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
880 		    !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
881 			goto error;
882 		if (euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
883 		    !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
884 			goto error;
885 		if (suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
886 		    !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
887 			goto error;
888 	}
889 
890 	if (ruid != (uid_t) -1) {
891 		new->uid = kruid;
892 		if (!uid_eq(kruid, old->uid)) {
893 			retval = set_user(new);
894 			if (retval < 0)
895 				goto error;
896 		}
897 	}
898 	if (euid != (uid_t) -1)
899 		new->euid = keuid;
900 	if (suid != (uid_t) -1)
901 		new->suid = ksuid;
902 	new->fsuid = new->euid;
903 
904 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
905 	if (retval < 0)
906 		goto error;
907 
908 	return commit_creds(new);
909 
910 error:
911 	abort_creds(new);
912 	return retval;
913 }
914 
SYSCALL_DEFINE3(getresuid,uid_t __user *,ruidp,uid_t __user *,euidp,uid_t __user *,suidp)915 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
916 {
917 	const struct cred *cred = current_cred();
918 	int retval;
919 	uid_t ruid, euid, suid;
920 
921 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
922 	euid = from_kuid_munged(cred->user_ns, cred->euid);
923 	suid = from_kuid_munged(cred->user_ns, cred->suid);
924 
925 	if (!(retval   = put_user(ruid, ruidp)) &&
926 	    !(retval   = put_user(euid, euidp)))
927 		retval = put_user(suid, suidp);
928 
929 	return retval;
930 }
931 
932 /*
933  * Same as above, but for rgid, egid, sgid.
934  */
SYSCALL_DEFINE3(setresgid,gid_t,rgid,gid_t,egid,gid_t,sgid)935 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
936 {
937 	struct user_namespace *ns = current_user_ns();
938 	const struct cred *old;
939 	struct cred *new;
940 	int retval;
941 	kgid_t krgid, kegid, ksgid;
942 
943 	krgid = make_kgid(ns, rgid);
944 	kegid = make_kgid(ns, egid);
945 	ksgid = make_kgid(ns, sgid);
946 
947 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
948 		return -EINVAL;
949 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
950 		return -EINVAL;
951 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
952 		return -EINVAL;
953 
954 	new = prepare_creds();
955 	if (!new)
956 		return -ENOMEM;
957 	old = current_cred();
958 
959 	retval = -EPERM;
960 	if (!nsown_capable(CAP_SETGID)) {
961 		if (rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
962 		    !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
963 			goto error;
964 		if (egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
965 		    !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
966 			goto error;
967 		if (sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
968 		    !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
969 			goto error;
970 	}
971 
972 	if (rgid != (gid_t) -1)
973 		new->gid = krgid;
974 	if (egid != (gid_t) -1)
975 		new->egid = kegid;
976 	if (sgid != (gid_t) -1)
977 		new->sgid = ksgid;
978 	new->fsgid = new->egid;
979 
980 	return commit_creds(new);
981 
982 error:
983 	abort_creds(new);
984 	return retval;
985 }
986 
SYSCALL_DEFINE3(getresgid,gid_t __user *,rgidp,gid_t __user *,egidp,gid_t __user *,sgidp)987 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
988 {
989 	const struct cred *cred = current_cred();
990 	int retval;
991 	gid_t rgid, egid, sgid;
992 
993 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
994 	egid = from_kgid_munged(cred->user_ns, cred->egid);
995 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
996 
997 	if (!(retval   = put_user(rgid, rgidp)) &&
998 	    !(retval   = put_user(egid, egidp)))
999 		retval = put_user(sgid, sgidp);
1000 
1001 	return retval;
1002 }
1003 
1004 
1005 /*
1006  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
1007  * is used for "access()" and for the NFS daemon (letting nfsd stay at
1008  * whatever uid it wants to). It normally shadows "euid", except when
1009  * explicitly set by setfsuid() or for access..
1010  */
SYSCALL_DEFINE1(setfsuid,uid_t,uid)1011 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
1012 {
1013 	const struct cred *old;
1014 	struct cred *new;
1015 	uid_t old_fsuid;
1016 	kuid_t kuid;
1017 
1018 	old = current_cred();
1019 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
1020 
1021 	kuid = make_kuid(old->user_ns, uid);
1022 	if (!uid_valid(kuid))
1023 		return old_fsuid;
1024 
1025 	new = prepare_creds();
1026 	if (!new)
1027 		return old_fsuid;
1028 
1029 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
1030 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
1031 	    nsown_capable(CAP_SETUID)) {
1032 		if (!uid_eq(kuid, old->fsuid)) {
1033 			new->fsuid = kuid;
1034 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
1035 				goto change_okay;
1036 		}
1037 	}
1038 
1039 	abort_creds(new);
1040 	return old_fsuid;
1041 
1042 change_okay:
1043 	commit_creds(new);
1044 	return old_fsuid;
1045 }
1046 
1047 /*
1048  * Samma på svenska..
1049  */
SYSCALL_DEFINE1(setfsgid,gid_t,gid)1050 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
1051 {
1052 	const struct cred *old;
1053 	struct cred *new;
1054 	gid_t old_fsgid;
1055 	kgid_t kgid;
1056 
1057 	old = current_cred();
1058 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
1059 
1060 	kgid = make_kgid(old->user_ns, gid);
1061 	if (!gid_valid(kgid))
1062 		return old_fsgid;
1063 
1064 	new = prepare_creds();
1065 	if (!new)
1066 		return old_fsgid;
1067 
1068 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
1069 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
1070 	    nsown_capable(CAP_SETGID)) {
1071 		if (!gid_eq(kgid, old->fsgid)) {
1072 			new->fsgid = kgid;
1073 			goto change_okay;
1074 		}
1075 	}
1076 
1077 	abort_creds(new);
1078 	return old_fsgid;
1079 
1080 change_okay:
1081 	commit_creds(new);
1082 	return old_fsgid;
1083 }
1084 
1085 /**
1086  * sys_getpid - return the thread group id of the current process
1087  *
1088  * Note, despite the name, this returns the tgid not the pid.  The tgid and
1089  * the pid are identical unless CLONE_THREAD was specified on clone() in
1090  * which case the tgid is the same in all threads of the same group.
1091  *
1092  * This is SMP safe as current->tgid does not change.
1093  */
SYSCALL_DEFINE0(getpid)1094 SYSCALL_DEFINE0(getpid)
1095 {
1096 	return task_tgid_vnr(current);
1097 }
1098 
1099 /* Thread ID - the internal kernel "pid" */
SYSCALL_DEFINE0(gettid)1100 SYSCALL_DEFINE0(gettid)
1101 {
1102 	return task_pid_vnr(current);
1103 }
1104 
1105 /*
1106  * Accessing ->real_parent is not SMP-safe, it could
1107  * change from under us. However, we can use a stale
1108  * value of ->real_parent under rcu_read_lock(), see
1109  * release_task()->call_rcu(delayed_put_task_struct).
1110  */
SYSCALL_DEFINE0(getppid)1111 SYSCALL_DEFINE0(getppid)
1112 {
1113 	int pid;
1114 
1115 	rcu_read_lock();
1116 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
1117 	rcu_read_unlock();
1118 
1119 	return pid;
1120 }
1121 
SYSCALL_DEFINE0(getuid)1122 SYSCALL_DEFINE0(getuid)
1123 {
1124 	/* Only we change this so SMP safe */
1125 	return from_kuid_munged(current_user_ns(), current_uid());
1126 }
1127 
SYSCALL_DEFINE0(geteuid)1128 SYSCALL_DEFINE0(geteuid)
1129 {
1130 	/* Only we change this so SMP safe */
1131 	return from_kuid_munged(current_user_ns(), current_euid());
1132 }
1133 
SYSCALL_DEFINE0(getgid)1134 SYSCALL_DEFINE0(getgid)
1135 {
1136 	/* Only we change this so SMP safe */
1137 	return from_kgid_munged(current_user_ns(), current_gid());
1138 }
1139 
SYSCALL_DEFINE0(getegid)1140 SYSCALL_DEFINE0(getegid)
1141 {
1142 	/* Only we change this so SMP safe */
1143 	return from_kgid_munged(current_user_ns(), current_egid());
1144 }
1145 
do_sys_times(struct tms * tms)1146 void do_sys_times(struct tms *tms)
1147 {
1148 	cputime_t tgutime, tgstime, cutime, cstime;
1149 
1150 	spin_lock_irq(&current->sighand->siglock);
1151 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1152 	cutime = current->signal->cutime;
1153 	cstime = current->signal->cstime;
1154 	spin_unlock_irq(&current->sighand->siglock);
1155 	tms->tms_utime = cputime_to_clock_t(tgutime);
1156 	tms->tms_stime = cputime_to_clock_t(tgstime);
1157 	tms->tms_cutime = cputime_to_clock_t(cutime);
1158 	tms->tms_cstime = cputime_to_clock_t(cstime);
1159 }
1160 
SYSCALL_DEFINE1(times,struct tms __user *,tbuf)1161 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1162 {
1163 	if (tbuf) {
1164 		struct tms tmp;
1165 
1166 		do_sys_times(&tmp);
1167 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1168 			return -EFAULT;
1169 	}
1170 	force_successful_syscall_return();
1171 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
1172 }
1173 
1174 /*
1175  * This needs some heavy checking ...
1176  * I just haven't the stomach for it. I also don't fully
1177  * understand sessions/pgrp etc. Let somebody who does explain it.
1178  *
1179  * OK, I think I have the protection semantics right.... this is really
1180  * only important on a multi-user system anyway, to make sure one user
1181  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1182  *
1183  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1184  * LBT 04.03.94
1185  */
SYSCALL_DEFINE2(setpgid,pid_t,pid,pid_t,pgid)1186 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1187 {
1188 	struct task_struct *p;
1189 	struct task_struct *group_leader = current->group_leader;
1190 	struct pid *pgrp;
1191 	int err;
1192 
1193 	if (!pid)
1194 		pid = task_pid_vnr(group_leader);
1195 	if (!pgid)
1196 		pgid = pid;
1197 	if (pgid < 0)
1198 		return -EINVAL;
1199 	rcu_read_lock();
1200 
1201 	/* From this point forward we keep holding onto the tasklist lock
1202 	 * so that our parent does not change from under us. -DaveM
1203 	 */
1204 	write_lock_irq(&tasklist_lock);
1205 
1206 	err = -ESRCH;
1207 	p = find_task_by_vpid(pid);
1208 	if (!p)
1209 		goto out;
1210 
1211 	err = -EINVAL;
1212 	if (!thread_group_leader(p))
1213 		goto out;
1214 
1215 	if (same_thread_group(p->real_parent, group_leader)) {
1216 		err = -EPERM;
1217 		if (task_session(p) != task_session(group_leader))
1218 			goto out;
1219 		err = -EACCES;
1220 		if (p->did_exec)
1221 			goto out;
1222 	} else {
1223 		err = -ESRCH;
1224 		if (p != group_leader)
1225 			goto out;
1226 	}
1227 
1228 	err = -EPERM;
1229 	if (p->signal->leader)
1230 		goto out;
1231 
1232 	pgrp = task_pid(p);
1233 	if (pgid != pid) {
1234 		struct task_struct *g;
1235 
1236 		pgrp = find_vpid(pgid);
1237 		g = pid_task(pgrp, PIDTYPE_PGID);
1238 		if (!g || task_session(g) != task_session(group_leader))
1239 			goto out;
1240 	}
1241 
1242 	err = security_task_setpgid(p, pgid);
1243 	if (err)
1244 		goto out;
1245 
1246 	if (task_pgrp(p) != pgrp)
1247 		change_pid(p, PIDTYPE_PGID, pgrp);
1248 
1249 	err = 0;
1250 out:
1251 	/* All paths lead to here, thus we are safe. -DaveM */
1252 	write_unlock_irq(&tasklist_lock);
1253 	rcu_read_unlock();
1254 	return err;
1255 }
1256 
SYSCALL_DEFINE1(getpgid,pid_t,pid)1257 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1258 {
1259 	struct task_struct *p;
1260 	struct pid *grp;
1261 	int retval;
1262 
1263 	rcu_read_lock();
1264 	if (!pid)
1265 		grp = task_pgrp(current);
1266 	else {
1267 		retval = -ESRCH;
1268 		p = find_task_by_vpid(pid);
1269 		if (!p)
1270 			goto out;
1271 		grp = task_pgrp(p);
1272 		if (!grp)
1273 			goto out;
1274 
1275 		retval = security_task_getpgid(p);
1276 		if (retval)
1277 			goto out;
1278 	}
1279 	retval = pid_vnr(grp);
1280 out:
1281 	rcu_read_unlock();
1282 	return retval;
1283 }
1284 
1285 #ifdef __ARCH_WANT_SYS_GETPGRP
1286 
SYSCALL_DEFINE0(getpgrp)1287 SYSCALL_DEFINE0(getpgrp)
1288 {
1289 	return sys_getpgid(0);
1290 }
1291 
1292 #endif
1293 
SYSCALL_DEFINE1(getsid,pid_t,pid)1294 SYSCALL_DEFINE1(getsid, pid_t, pid)
1295 {
1296 	struct task_struct *p;
1297 	struct pid *sid;
1298 	int retval;
1299 
1300 	rcu_read_lock();
1301 	if (!pid)
1302 		sid = task_session(current);
1303 	else {
1304 		retval = -ESRCH;
1305 		p = find_task_by_vpid(pid);
1306 		if (!p)
1307 			goto out;
1308 		sid = task_session(p);
1309 		if (!sid)
1310 			goto out;
1311 
1312 		retval = security_task_getsid(p);
1313 		if (retval)
1314 			goto out;
1315 	}
1316 	retval = pid_vnr(sid);
1317 out:
1318 	rcu_read_unlock();
1319 	return retval;
1320 }
1321 
SYSCALL_DEFINE0(setsid)1322 SYSCALL_DEFINE0(setsid)
1323 {
1324 	struct task_struct *group_leader = current->group_leader;
1325 	struct pid *sid = task_pid(group_leader);
1326 	pid_t session = pid_vnr(sid);
1327 	int err = -EPERM;
1328 
1329 	write_lock_irq(&tasklist_lock);
1330 	/* Fail if I am already a session leader */
1331 	if (group_leader->signal->leader)
1332 		goto out;
1333 
1334 	/* Fail if a process group id already exists that equals the
1335 	 * proposed session id.
1336 	 */
1337 	if (pid_task(sid, PIDTYPE_PGID))
1338 		goto out;
1339 
1340 	group_leader->signal->leader = 1;
1341 	__set_special_pids(sid);
1342 
1343 	proc_clear_tty(group_leader);
1344 
1345 	err = session;
1346 out:
1347 	write_unlock_irq(&tasklist_lock);
1348 	if (err > 0) {
1349 		proc_sid_connector(group_leader);
1350 		sched_autogroup_create_attach(group_leader);
1351 	}
1352 	return err;
1353 }
1354 
1355 DECLARE_RWSEM(uts_sem);
1356 
1357 #ifdef COMPAT_UTS_MACHINE
1358 #define override_architecture(name) \
1359 	(personality(current->personality) == PER_LINUX32 && \
1360 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1361 		      sizeof(COMPAT_UTS_MACHINE)))
1362 #else
1363 #define override_architecture(name)	0
1364 #endif
1365 
1366 /*
1367  * Work around broken programs that cannot handle "Linux 3.0".
1368  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1369  */
override_release(char __user * release,size_t len)1370 static int override_release(char __user *release, size_t len)
1371 {
1372 	int ret = 0;
1373 
1374 	if (current->personality & UNAME26) {
1375 		const char *rest = UTS_RELEASE;
1376 		char buf[65] = { 0 };
1377 		int ndots = 0;
1378 		unsigned v;
1379 		size_t copy;
1380 
1381 		while (*rest) {
1382 			if (*rest == '.' && ++ndots >= 3)
1383 				break;
1384 			if (!isdigit(*rest) && *rest != '.')
1385 				break;
1386 			rest++;
1387 		}
1388 		v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1389 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1390 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1391 		ret = copy_to_user(release, buf, copy + 1);
1392 	}
1393 	return ret;
1394 }
1395 
SYSCALL_DEFINE1(newuname,struct new_utsname __user *,name)1396 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1397 {
1398 	int errno = 0;
1399 
1400 	down_read(&uts_sem);
1401 	if (copy_to_user(name, utsname(), sizeof *name))
1402 		errno = -EFAULT;
1403 	up_read(&uts_sem);
1404 
1405 	if (!errno && override_release(name->release, sizeof(name->release)))
1406 		errno = -EFAULT;
1407 	if (!errno && override_architecture(name))
1408 		errno = -EFAULT;
1409 	return errno;
1410 }
1411 
1412 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1413 /*
1414  * Old cruft
1415  */
SYSCALL_DEFINE1(uname,struct old_utsname __user *,name)1416 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1417 {
1418 	int error = 0;
1419 
1420 	if (!name)
1421 		return -EFAULT;
1422 
1423 	down_read(&uts_sem);
1424 	if (copy_to_user(name, utsname(), sizeof(*name)))
1425 		error = -EFAULT;
1426 	up_read(&uts_sem);
1427 
1428 	if (!error && override_release(name->release, sizeof(name->release)))
1429 		error = -EFAULT;
1430 	if (!error && override_architecture(name))
1431 		error = -EFAULT;
1432 	return error;
1433 }
1434 
SYSCALL_DEFINE1(olduname,struct oldold_utsname __user *,name)1435 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1436 {
1437 	int error;
1438 
1439 	if (!name)
1440 		return -EFAULT;
1441 	if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1442 		return -EFAULT;
1443 
1444 	down_read(&uts_sem);
1445 	error = __copy_to_user(&name->sysname, &utsname()->sysname,
1446 			       __OLD_UTS_LEN);
1447 	error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1448 	error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1449 				__OLD_UTS_LEN);
1450 	error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1451 	error |= __copy_to_user(&name->release, &utsname()->release,
1452 				__OLD_UTS_LEN);
1453 	error |= __put_user(0, name->release + __OLD_UTS_LEN);
1454 	error |= __copy_to_user(&name->version, &utsname()->version,
1455 				__OLD_UTS_LEN);
1456 	error |= __put_user(0, name->version + __OLD_UTS_LEN);
1457 	error |= __copy_to_user(&name->machine, &utsname()->machine,
1458 				__OLD_UTS_LEN);
1459 	error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1460 	up_read(&uts_sem);
1461 
1462 	if (!error && override_architecture(name))
1463 		error = -EFAULT;
1464 	if (!error && override_release(name->release, sizeof(name->release)))
1465 		error = -EFAULT;
1466 	return error ? -EFAULT : 0;
1467 }
1468 #endif
1469 
SYSCALL_DEFINE2(sethostname,char __user *,name,int,len)1470 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1471 {
1472 	int errno;
1473 	char tmp[__NEW_UTS_LEN];
1474 
1475 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1476 		return -EPERM;
1477 
1478 	if (len < 0 || len > __NEW_UTS_LEN)
1479 		return -EINVAL;
1480 	down_write(&uts_sem);
1481 	errno = -EFAULT;
1482 	if (!copy_from_user(tmp, name, len)) {
1483 		struct new_utsname *u = utsname();
1484 
1485 		memcpy(u->nodename, tmp, len);
1486 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1487 		errno = 0;
1488 		uts_proc_notify(UTS_PROC_HOSTNAME);
1489 	}
1490 	up_write(&uts_sem);
1491 	return errno;
1492 }
1493 
1494 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1495 
SYSCALL_DEFINE2(gethostname,char __user *,name,int,len)1496 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1497 {
1498 	int i, errno;
1499 	struct new_utsname *u;
1500 
1501 	if (len < 0)
1502 		return -EINVAL;
1503 	down_read(&uts_sem);
1504 	u = utsname();
1505 	i = 1 + strlen(u->nodename);
1506 	if (i > len)
1507 		i = len;
1508 	errno = 0;
1509 	if (copy_to_user(name, u->nodename, i))
1510 		errno = -EFAULT;
1511 	up_read(&uts_sem);
1512 	return errno;
1513 }
1514 
1515 #endif
1516 
1517 /*
1518  * Only setdomainname; getdomainname can be implemented by calling
1519  * uname()
1520  */
SYSCALL_DEFINE2(setdomainname,char __user *,name,int,len)1521 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1522 {
1523 	int errno;
1524 	char tmp[__NEW_UTS_LEN];
1525 
1526 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1527 		return -EPERM;
1528 	if (len < 0 || len > __NEW_UTS_LEN)
1529 		return -EINVAL;
1530 
1531 	down_write(&uts_sem);
1532 	errno = -EFAULT;
1533 	if (!copy_from_user(tmp, name, len)) {
1534 		struct new_utsname *u = utsname();
1535 
1536 		memcpy(u->domainname, tmp, len);
1537 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1538 		errno = 0;
1539 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1540 	}
1541 	up_write(&uts_sem);
1542 	return errno;
1543 }
1544 
SYSCALL_DEFINE2(getrlimit,unsigned int,resource,struct rlimit __user *,rlim)1545 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1546 {
1547 	struct rlimit value;
1548 	int ret;
1549 
1550 	ret = do_prlimit(current, resource, NULL, &value);
1551 	if (!ret)
1552 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1553 
1554 	return ret;
1555 }
1556 
1557 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1558 
1559 /*
1560  *	Back compatibility for getrlimit. Needed for some apps.
1561  */
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 	task_lock(current->group_leader);
1571 	x = current->signal->rlim[resource];
1572 	task_unlock(current->group_leader);
1573 	if (x.rlim_cur > 0x7FFFFFFF)
1574 		x.rlim_cur = 0x7FFFFFFF;
1575 	if (x.rlim_max > 0x7FFFFFFF)
1576 		x.rlim_max = 0x7FFFFFFF;
1577 	return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1578 }
1579 
1580 #endif
1581 
rlim64_is_infinity(__u64 rlim64)1582 static inline bool rlim64_is_infinity(__u64 rlim64)
1583 {
1584 #if BITS_PER_LONG < 64
1585 	return rlim64 >= ULONG_MAX;
1586 #else
1587 	return rlim64 == RLIM64_INFINITY;
1588 #endif
1589 }
1590 
rlim_to_rlim64(const struct rlimit * rlim,struct rlimit64 * rlim64)1591 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1592 {
1593 	if (rlim->rlim_cur == RLIM_INFINITY)
1594 		rlim64->rlim_cur = RLIM64_INFINITY;
1595 	else
1596 		rlim64->rlim_cur = rlim->rlim_cur;
1597 	if (rlim->rlim_max == RLIM_INFINITY)
1598 		rlim64->rlim_max = RLIM64_INFINITY;
1599 	else
1600 		rlim64->rlim_max = rlim->rlim_max;
1601 }
1602 
rlim64_to_rlim(const struct rlimit64 * rlim64,struct rlimit * rlim)1603 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1604 {
1605 	if (rlim64_is_infinity(rlim64->rlim_cur))
1606 		rlim->rlim_cur = RLIM_INFINITY;
1607 	else
1608 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1609 	if (rlim64_is_infinity(rlim64->rlim_max))
1610 		rlim->rlim_max = RLIM_INFINITY;
1611 	else
1612 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1613 }
1614 
1615 /* 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)1616 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1617 		struct rlimit *new_rlim, struct rlimit *old_rlim)
1618 {
1619 	struct rlimit *rlim;
1620 	int retval = 0;
1621 
1622 	if (resource >= RLIM_NLIMITS)
1623 		return -EINVAL;
1624 	if (new_rlim) {
1625 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1626 			return -EINVAL;
1627 		if (resource == RLIMIT_NOFILE &&
1628 				new_rlim->rlim_max > sysctl_nr_open)
1629 			return -EPERM;
1630 	}
1631 
1632 	/* protect tsk->signal and tsk->sighand from disappearing */
1633 	read_lock(&tasklist_lock);
1634 	if (!tsk->sighand) {
1635 		retval = -ESRCH;
1636 		goto out;
1637 	}
1638 
1639 	rlim = tsk->signal->rlim + resource;
1640 	task_lock(tsk->group_leader);
1641 	if (new_rlim) {
1642 		/* Keep the capable check against init_user_ns until
1643 		   cgroups can contain all limits */
1644 		if (new_rlim->rlim_max > rlim->rlim_max &&
1645 				!capable(CAP_SYS_RESOURCE))
1646 			retval = -EPERM;
1647 		if (!retval)
1648 			retval = security_task_setrlimit(tsk->group_leader,
1649 					resource, new_rlim);
1650 		if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1651 			/*
1652 			 * The caller is asking for an immediate RLIMIT_CPU
1653 			 * expiry.  But we use the zero value to mean "it was
1654 			 * never set".  So let's cheat and make it one second
1655 			 * instead
1656 			 */
1657 			new_rlim->rlim_cur = 1;
1658 		}
1659 	}
1660 	if (!retval) {
1661 		if (old_rlim)
1662 			*old_rlim = *rlim;
1663 		if (new_rlim)
1664 			*rlim = *new_rlim;
1665 	}
1666 	task_unlock(tsk->group_leader);
1667 
1668 	/*
1669 	 * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1670 	 * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1671 	 * very long-standing error, and fixing it now risks breakage of
1672 	 * applications, so we live with it
1673 	 */
1674 	 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1675 			 new_rlim->rlim_cur != RLIM_INFINITY)
1676 		update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1677 out:
1678 	read_unlock(&tasklist_lock);
1679 	return retval;
1680 }
1681 
1682 /* rcu lock must be held */
check_prlimit_permission(struct task_struct * task)1683 static int check_prlimit_permission(struct task_struct *task)
1684 {
1685 	const struct cred *cred = current_cred(), *tcred;
1686 
1687 	if (current == task)
1688 		return 0;
1689 
1690 	tcred = __task_cred(task);
1691 	if (uid_eq(cred->uid, tcred->euid) &&
1692 	    uid_eq(cred->uid, tcred->suid) &&
1693 	    uid_eq(cred->uid, tcred->uid)  &&
1694 	    gid_eq(cred->gid, tcred->egid) &&
1695 	    gid_eq(cred->gid, tcred->sgid) &&
1696 	    gid_eq(cred->gid, tcred->gid))
1697 		return 0;
1698 	if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1699 		return 0;
1700 
1701 	return -EPERM;
1702 }
1703 
SYSCALL_DEFINE4(prlimit64,pid_t,pid,unsigned int,resource,const struct rlimit64 __user *,new_rlim,struct rlimit64 __user *,old_rlim)1704 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1705 		const struct rlimit64 __user *, new_rlim,
1706 		struct rlimit64 __user *, old_rlim)
1707 {
1708 	struct rlimit64 old64, new64;
1709 	struct rlimit old, new;
1710 	struct task_struct *tsk;
1711 	int ret;
1712 
1713 	if (new_rlim) {
1714 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1715 			return -EFAULT;
1716 		rlim64_to_rlim(&new64, &new);
1717 	}
1718 
1719 	rcu_read_lock();
1720 	tsk = pid ? find_task_by_vpid(pid) : current;
1721 	if (!tsk) {
1722 		rcu_read_unlock();
1723 		return -ESRCH;
1724 	}
1725 	ret = check_prlimit_permission(tsk);
1726 	if (ret) {
1727 		rcu_read_unlock();
1728 		return ret;
1729 	}
1730 	get_task_struct(tsk);
1731 	rcu_read_unlock();
1732 
1733 	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1734 			old_rlim ? &old : NULL);
1735 
1736 	if (!ret && old_rlim) {
1737 		rlim_to_rlim64(&old, &old64);
1738 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1739 			ret = -EFAULT;
1740 	}
1741 
1742 	put_task_struct(tsk);
1743 	return ret;
1744 }
1745 
SYSCALL_DEFINE2(setrlimit,unsigned int,resource,struct rlimit __user *,rlim)1746 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1747 {
1748 	struct rlimit new_rlim;
1749 
1750 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1751 		return -EFAULT;
1752 	return do_prlimit(current, resource, &new_rlim, NULL);
1753 }
1754 
1755 /*
1756  * It would make sense to put struct rusage in the task_struct,
1757  * except that would make the task_struct be *really big*.  After
1758  * task_struct gets moved into malloc'ed memory, it would
1759  * make sense to do this.  It will make moving the rest of the information
1760  * a lot simpler!  (Which we're not doing right now because we're not
1761  * measuring them yet).
1762  *
1763  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1764  * races with threads incrementing their own counters.  But since word
1765  * reads are atomic, we either get new values or old values and we don't
1766  * care which for the sums.  We always take the siglock to protect reading
1767  * the c* fields from p->signal from races with exit.c updating those
1768  * fields when reaping, so a sample either gets all the additions of a
1769  * given child after it's reaped, or none so this sample is before reaping.
1770  *
1771  * Locking:
1772  * We need to take the siglock for CHILDEREN, SELF and BOTH
1773  * for  the cases current multithreaded, non-current single threaded
1774  * non-current multithreaded.  Thread traversal is now safe with
1775  * the siglock held.
1776  * Strictly speaking, we donot need to take the siglock if we are current and
1777  * single threaded,  as no one else can take our signal_struct away, no one
1778  * else can  reap the  children to update signal->c* counters, and no one else
1779  * can race with the signal-> fields. If we do not take any lock, the
1780  * signal-> fields could be read out of order while another thread was just
1781  * exiting. So we should  place a read memory barrier when we avoid the lock.
1782  * On the writer side,  write memory barrier is implied in  __exit_signal
1783  * as __exit_signal releases  the siglock spinlock after updating the signal->
1784  * fields. But we don't do this yet to keep things simple.
1785  *
1786  */
1787 
accumulate_thread_rusage(struct task_struct * t,struct rusage * r)1788 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1789 {
1790 	r->ru_nvcsw += t->nvcsw;
1791 	r->ru_nivcsw += t->nivcsw;
1792 	r->ru_minflt += t->min_flt;
1793 	r->ru_majflt += t->maj_flt;
1794 	r->ru_inblock += task_io_get_inblock(t);
1795 	r->ru_oublock += task_io_get_oublock(t);
1796 }
1797 
k_getrusage(struct task_struct * p,int who,struct rusage * r)1798 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1799 {
1800 	struct task_struct *t;
1801 	unsigned long flags;
1802 	cputime_t tgutime, tgstime, utime, stime;
1803 	unsigned long maxrss = 0;
1804 
1805 	memset((char *) r, 0, sizeof *r);
1806 	utime = stime = 0;
1807 
1808 	if (who == RUSAGE_THREAD) {
1809 		task_cputime_adjusted(current, &utime, &stime);
1810 		accumulate_thread_rusage(p, r);
1811 		maxrss = p->signal->maxrss;
1812 		goto out;
1813 	}
1814 
1815 	if (!lock_task_sighand(p, &flags))
1816 		return;
1817 
1818 	switch (who) {
1819 		case RUSAGE_BOTH:
1820 		case RUSAGE_CHILDREN:
1821 			utime = p->signal->cutime;
1822 			stime = p->signal->cstime;
1823 			r->ru_nvcsw = p->signal->cnvcsw;
1824 			r->ru_nivcsw = p->signal->cnivcsw;
1825 			r->ru_minflt = p->signal->cmin_flt;
1826 			r->ru_majflt = p->signal->cmaj_flt;
1827 			r->ru_inblock = p->signal->cinblock;
1828 			r->ru_oublock = p->signal->coublock;
1829 			maxrss = p->signal->cmaxrss;
1830 
1831 			if (who == RUSAGE_CHILDREN)
1832 				break;
1833 
1834 		case RUSAGE_SELF:
1835 			thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1836 			utime += tgutime;
1837 			stime += tgstime;
1838 			r->ru_nvcsw += p->signal->nvcsw;
1839 			r->ru_nivcsw += p->signal->nivcsw;
1840 			r->ru_minflt += p->signal->min_flt;
1841 			r->ru_majflt += p->signal->maj_flt;
1842 			r->ru_inblock += p->signal->inblock;
1843 			r->ru_oublock += p->signal->oublock;
1844 			if (maxrss < p->signal->maxrss)
1845 				maxrss = p->signal->maxrss;
1846 			t = p;
1847 			do {
1848 				accumulate_thread_rusage(t, r);
1849 				t = next_thread(t);
1850 			} while (t != p);
1851 			break;
1852 
1853 		default:
1854 			BUG();
1855 	}
1856 	unlock_task_sighand(p, &flags);
1857 
1858 out:
1859 	cputime_to_timeval(utime, &r->ru_utime);
1860 	cputime_to_timeval(stime, &r->ru_stime);
1861 
1862 	if (who != RUSAGE_CHILDREN) {
1863 		struct mm_struct *mm = get_task_mm(p);
1864 		if (mm) {
1865 			setmax_mm_hiwater_rss(&maxrss, mm);
1866 			mmput(mm);
1867 		}
1868 	}
1869 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1870 }
1871 
getrusage(struct task_struct * p,int who,struct rusage __user * ru)1872 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1873 {
1874 	struct rusage r;
1875 	k_getrusage(p, who, &r);
1876 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1877 }
1878 
SYSCALL_DEFINE2(getrusage,int,who,struct rusage __user *,ru)1879 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1880 {
1881 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1882 	    who != RUSAGE_THREAD)
1883 		return -EINVAL;
1884 	return getrusage(current, who, ru);
1885 }
1886 
1887 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(getrusage,int,who,struct compat_rusage __user *,ru)1888 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_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 	k_getrusage(current, who, &r);
1897 	return put_compat_rusage(&r, ru);
1898 }
1899 #endif
1900 
SYSCALL_DEFINE1(umask,int,mask)1901 SYSCALL_DEFINE1(umask, int, mask)
1902 {
1903 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1904 	return mask;
1905 }
1906 
prctl_set_mm_exe_file(struct mm_struct * mm,unsigned int fd)1907 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1908 {
1909 	struct fd exe;
1910 	struct inode *inode;
1911 	int err;
1912 
1913 	exe = fdget(fd);
1914 	if (!exe.file)
1915 		return -EBADF;
1916 
1917 	inode = file_inode(exe.file);
1918 
1919 	/*
1920 	 * Because the original mm->exe_file points to executable file, make
1921 	 * sure that this one is executable as well, to avoid breaking an
1922 	 * overall picture.
1923 	 */
1924 	err = -EACCES;
1925 	if (!S_ISREG(inode->i_mode)	||
1926 	    exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1927 		goto exit;
1928 
1929 	err = inode_permission(inode, MAY_EXEC);
1930 	if (err)
1931 		goto exit;
1932 
1933 	down_write(&mm->mmap_sem);
1934 
1935 	/*
1936 	 * Forbid mm->exe_file change if old file still mapped.
1937 	 */
1938 	err = -EBUSY;
1939 	if (mm->exe_file) {
1940 		struct vm_area_struct *vma;
1941 
1942 		for (vma = mm->mmap; vma; vma = vma->vm_next)
1943 			if (vma->vm_file &&
1944 			    path_equal(&vma->vm_file->f_path,
1945 				       &mm->exe_file->f_path))
1946 				goto exit_unlock;
1947 	}
1948 
1949 	/*
1950 	 * The symlink can be changed only once, just to disallow arbitrary
1951 	 * transitions malicious software might bring in. This means one
1952 	 * could make a snapshot over all processes running and monitor
1953 	 * /proc/pid/exe changes to notice unusual activity if needed.
1954 	 */
1955 	err = -EPERM;
1956 	if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1957 		goto exit_unlock;
1958 
1959 	err = 0;
1960 	set_mm_exe_file(mm, exe.file);	/* this grabs a reference to exe.file */
1961 exit_unlock:
1962 	up_write(&mm->mmap_sem);
1963 
1964 exit:
1965 	fdput(exe);
1966 	return err;
1967 }
1968 
prctl_set_mm(int opt,unsigned long addr,unsigned long arg4,unsigned long arg5)1969 static int prctl_set_mm(int opt, unsigned long addr,
1970 			unsigned long arg4, unsigned long arg5)
1971 {
1972 	unsigned long rlim = rlimit(RLIMIT_DATA);
1973 	struct mm_struct *mm = current->mm;
1974 	struct vm_area_struct *vma;
1975 	int error;
1976 
1977 	if (arg5 || (arg4 && opt != PR_SET_MM_AUXV))
1978 		return -EINVAL;
1979 
1980 	if (!capable(CAP_SYS_RESOURCE))
1981 		return -EPERM;
1982 
1983 	if (opt == PR_SET_MM_EXE_FILE)
1984 		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1985 
1986 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
1987 		return -EINVAL;
1988 
1989 	error = -EINVAL;
1990 
1991 	down_read(&mm->mmap_sem);
1992 	vma = find_vma(mm, addr);
1993 
1994 	switch (opt) {
1995 	case PR_SET_MM_START_CODE:
1996 		mm->start_code = addr;
1997 		break;
1998 	case PR_SET_MM_END_CODE:
1999 		mm->end_code = addr;
2000 		break;
2001 	case PR_SET_MM_START_DATA:
2002 		mm->start_data = addr;
2003 		break;
2004 	case PR_SET_MM_END_DATA:
2005 		mm->end_data = addr;
2006 		break;
2007 
2008 	case PR_SET_MM_START_BRK:
2009 		if (addr <= mm->end_data)
2010 			goto out;
2011 
2012 		if (rlim < RLIM_INFINITY &&
2013 		    (mm->brk - addr) +
2014 		    (mm->end_data - mm->start_data) > rlim)
2015 			goto out;
2016 
2017 		mm->start_brk = addr;
2018 		break;
2019 
2020 	case PR_SET_MM_BRK:
2021 		if (addr <= mm->end_data)
2022 			goto out;
2023 
2024 		if (rlim < RLIM_INFINITY &&
2025 		    (addr - mm->start_brk) +
2026 		    (mm->end_data - mm->start_data) > rlim)
2027 			goto out;
2028 
2029 		mm->brk = addr;
2030 		break;
2031 
2032 	/*
2033 	 * If command line arguments and environment
2034 	 * are placed somewhere else on stack, we can
2035 	 * set them up here, ARG_START/END to setup
2036 	 * command line argumets and ENV_START/END
2037 	 * for environment.
2038 	 */
2039 	case PR_SET_MM_START_STACK:
2040 	case PR_SET_MM_ARG_START:
2041 	case PR_SET_MM_ARG_END:
2042 	case PR_SET_MM_ENV_START:
2043 	case PR_SET_MM_ENV_END:
2044 		if (!vma) {
2045 			error = -EFAULT;
2046 			goto out;
2047 		}
2048 		if (opt == PR_SET_MM_START_STACK)
2049 			mm->start_stack = addr;
2050 		else if (opt == PR_SET_MM_ARG_START)
2051 			mm->arg_start = addr;
2052 		else if (opt == PR_SET_MM_ARG_END)
2053 			mm->arg_end = addr;
2054 		else if (opt == PR_SET_MM_ENV_START)
2055 			mm->env_start = addr;
2056 		else if (opt == PR_SET_MM_ENV_END)
2057 			mm->env_end = addr;
2058 		break;
2059 
2060 	/*
2061 	 * This doesn't move auxiliary vector itself
2062 	 * since it's pinned to mm_struct, but allow
2063 	 * to fill vector with new values. It's up
2064 	 * to a caller to provide sane values here
2065 	 * otherwise user space tools which use this
2066 	 * vector might be unhappy.
2067 	 */
2068 	case PR_SET_MM_AUXV: {
2069 		unsigned long user_auxv[AT_VECTOR_SIZE];
2070 
2071 		if (arg4 > sizeof(user_auxv))
2072 			goto out;
2073 		up_read(&mm->mmap_sem);
2074 
2075 		if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
2076 			return -EFAULT;
2077 
2078 		/* Make sure the last entry is always AT_NULL */
2079 		user_auxv[AT_VECTOR_SIZE - 2] = 0;
2080 		user_auxv[AT_VECTOR_SIZE - 1] = 0;
2081 
2082 		BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2083 
2084 		task_lock(current);
2085 		memcpy(mm->saved_auxv, user_auxv, arg4);
2086 		task_unlock(current);
2087 
2088 		return 0;
2089 	}
2090 	default:
2091 		goto out;
2092 	}
2093 
2094 	error = 0;
2095 out:
2096 	up_read(&mm->mmap_sem);
2097 	return error;
2098 }
2099 
2100 #ifdef CONFIG_CHECKPOINT_RESTORE
prctl_get_tid_address(struct task_struct * me,int __user ** tid_addr)2101 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2102 {
2103 	return put_user(me->clear_child_tid, tid_addr);
2104 }
2105 #else
prctl_get_tid_address(struct task_struct * me,int __user ** tid_addr)2106 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2107 {
2108 	return -EINVAL;
2109 }
2110 #endif
2111 
2112 #ifdef CONFIG_MMU
prctl_update_vma_anon_name(struct vm_area_struct * vma,struct vm_area_struct ** prev,unsigned long start,unsigned long end,const char __user * name_addr)2113 static int prctl_update_vma_anon_name(struct vm_area_struct *vma,
2114 		struct vm_area_struct **prev,
2115 		unsigned long start, unsigned long end,
2116 		const char __user *name_addr)
2117 {
2118 	struct mm_struct * mm = vma->vm_mm;
2119 	int error = 0;
2120 	pgoff_t pgoff;
2121 
2122 	if (name_addr == vma_get_anon_name(vma)) {
2123 		*prev = vma;
2124 		goto out;
2125 	}
2126 
2127 	pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
2128 	*prev = vma_merge(mm, *prev, start, end, vma->vm_flags, vma->anon_vma,
2129 				vma->vm_file, pgoff, vma_policy(vma),
2130 				name_addr);
2131 	if (*prev) {
2132 		vma = *prev;
2133 		goto success;
2134 	}
2135 
2136 	*prev = vma;
2137 
2138 	if (start != vma->vm_start) {
2139 		error = split_vma(mm, vma, start, 1);
2140 		if (error)
2141 			goto out;
2142 	}
2143 
2144 	if (end != vma->vm_end) {
2145 		error = split_vma(mm, vma, end, 0);
2146 		if (error)
2147 			goto out;
2148 	}
2149 
2150 success:
2151 	if (!vma->vm_file)
2152 		vma->shared.anon_name = name_addr;
2153 
2154 out:
2155 	if (error == -ENOMEM)
2156 		error = -EAGAIN;
2157 	return error;
2158 }
2159 
prctl_set_vma_anon_name(unsigned long start,unsigned long end,unsigned long arg)2160 static int prctl_set_vma_anon_name(unsigned long start, unsigned long end,
2161 			unsigned long arg)
2162 {
2163 	unsigned long tmp;
2164 	struct vm_area_struct * vma, *prev;
2165 	int unmapped_error = 0;
2166 	int error = -EINVAL;
2167 
2168 	/*
2169 	 * If the interval [start,end) covers some unmapped address
2170 	 * ranges, just ignore them, but return -ENOMEM at the end.
2171 	 * - this matches the handling in madvise.
2172 	 */
2173 	vma = find_vma_prev(current->mm, start, &prev);
2174 	if (vma && start > vma->vm_start)
2175 		prev = vma;
2176 
2177 	for (;;) {
2178 		/* Still start < end. */
2179 		error = -ENOMEM;
2180 		if (!vma)
2181 			return error;
2182 
2183 		/* Here start < (end|vma->vm_end). */
2184 		if (start < vma->vm_start) {
2185 			unmapped_error = -ENOMEM;
2186 			start = vma->vm_start;
2187 			if (start >= end)
2188 				return error;
2189 		}
2190 
2191 		/* Here vma->vm_start <= start < (end|vma->vm_end) */
2192 		tmp = vma->vm_end;
2193 		if (end < tmp)
2194 			tmp = end;
2195 
2196 		/* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
2197 		error = prctl_update_vma_anon_name(vma, &prev, start, tmp,
2198 				(const char __user *)arg);
2199 		if (error)
2200 			return error;
2201 		start = tmp;
2202 		if (prev && start < prev->vm_end)
2203 			start = prev->vm_end;
2204 		error = unmapped_error;
2205 		if (start >= end)
2206 			return error;
2207 		if (prev)
2208 			vma = prev->vm_next;
2209 		else	/* madvise_remove dropped mmap_sem */
2210 			vma = find_vma(current->mm, start);
2211 	}
2212 }
2213 
prctl_set_vma(unsigned long opt,unsigned long start,unsigned long len_in,unsigned long arg)2214 static int prctl_set_vma(unsigned long opt, unsigned long start,
2215 		unsigned long len_in, unsigned long arg)
2216 {
2217 	struct mm_struct *mm = current->mm;
2218 	int error;
2219 	unsigned long len;
2220 	unsigned long end;
2221 
2222 	if (start & ~PAGE_MASK)
2223 		return -EINVAL;
2224 	len = (len_in + ~PAGE_MASK) & PAGE_MASK;
2225 
2226 	/* Check to see whether len was rounded up from small -ve to zero */
2227 	if (len_in && !len)
2228 		return -EINVAL;
2229 
2230 	end = start + len;
2231 	if (end < start)
2232 		return -EINVAL;
2233 
2234 	if (end == start)
2235 		return 0;
2236 
2237 	down_write(&mm->mmap_sem);
2238 
2239 	switch (opt) {
2240 	case PR_SET_VMA_ANON_NAME:
2241 		error = prctl_set_vma_anon_name(start, end, arg);
2242 		break;
2243 	default:
2244 		error = -EINVAL;
2245 	}
2246 
2247 	up_write(&mm->mmap_sem);
2248 
2249 	return error;
2250 }
2251 #else /* CONFIG_MMU */
prctl_set_vma(unsigned long opt,unsigned long start,unsigned long len_in,unsigned long arg)2252 static int prctl_set_vma(unsigned long opt, unsigned long start,
2253 		unsigned long len_in, unsigned long arg)
2254 {
2255 	return -EINVAL;
2256 }
2257 #endif
2258 
SYSCALL_DEFINE5(prctl,int,option,unsigned long,arg2,unsigned long,arg3,unsigned long,arg4,unsigned long,arg5)2259 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2260 		unsigned long, arg4, unsigned long, arg5)
2261 {
2262 	struct task_struct *me = current;
2263 	struct task_struct *tsk;
2264 	unsigned char comm[sizeof(me->comm)];
2265 	long error;
2266 
2267 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2268 	if (error != -ENOSYS)
2269 		return error;
2270 
2271 	error = 0;
2272 	switch (option) {
2273 	case PR_SET_PDEATHSIG:
2274 		if (!valid_signal(arg2)) {
2275 			error = -EINVAL;
2276 			break;
2277 		}
2278 		me->pdeath_signal = arg2;
2279 		break;
2280 	case PR_GET_PDEATHSIG:
2281 		error = put_user(me->pdeath_signal, (int __user *)arg2);
2282 		break;
2283 	case PR_GET_DUMPABLE:
2284 		error = get_dumpable(me->mm);
2285 		break;
2286 	case PR_SET_DUMPABLE:
2287 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2288 			error = -EINVAL;
2289 			break;
2290 		}
2291 		set_dumpable(me->mm, arg2);
2292 		break;
2293 
2294 	case PR_SET_UNALIGN:
2295 		error = SET_UNALIGN_CTL(me, arg2);
2296 		break;
2297 	case PR_GET_UNALIGN:
2298 		error = GET_UNALIGN_CTL(me, arg2);
2299 		break;
2300 	case PR_SET_FPEMU:
2301 		error = SET_FPEMU_CTL(me, arg2);
2302 		break;
2303 	case PR_GET_FPEMU:
2304 		error = GET_FPEMU_CTL(me, arg2);
2305 		break;
2306 	case PR_SET_FPEXC:
2307 		error = SET_FPEXC_CTL(me, arg2);
2308 		break;
2309 	case PR_GET_FPEXC:
2310 		error = GET_FPEXC_CTL(me, arg2);
2311 		break;
2312 	case PR_GET_TIMING:
2313 		error = PR_TIMING_STATISTICAL;
2314 		break;
2315 	case PR_SET_TIMING:
2316 		if (arg2 != PR_TIMING_STATISTICAL)
2317 			error = -EINVAL;
2318 		break;
2319 	case PR_SET_NAME:
2320 		comm[sizeof(me->comm) - 1] = 0;
2321 		if (strncpy_from_user(comm, (char __user *)arg2,
2322 				      sizeof(me->comm) - 1) < 0)
2323 			return -EFAULT;
2324 		set_task_comm(me, comm);
2325 		proc_comm_connector(me);
2326 		break;
2327 	case PR_GET_NAME:
2328 		get_task_comm(comm, me);
2329 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2330 			return -EFAULT;
2331 		break;
2332 	case PR_GET_ENDIAN:
2333 		error = GET_ENDIAN(me, arg2);
2334 		break;
2335 	case PR_SET_ENDIAN:
2336 		error = SET_ENDIAN(me, arg2);
2337 		break;
2338 	case PR_GET_SECCOMP:
2339 		error = prctl_get_seccomp();
2340 		break;
2341 	case PR_SET_SECCOMP:
2342 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2343 		break;
2344 	case PR_GET_TSC:
2345 		error = GET_TSC_CTL(arg2);
2346 		break;
2347 	case PR_SET_TSC:
2348 		error = SET_TSC_CTL(arg2);
2349 		break;
2350 	case PR_TASK_PERF_EVENTS_DISABLE:
2351 		error = perf_event_task_disable();
2352 		break;
2353 	case PR_TASK_PERF_EVENTS_ENABLE:
2354 		error = perf_event_task_enable();
2355 		break;
2356 	case PR_GET_TIMERSLACK:
2357 		error = current->timer_slack_ns;
2358 		break;
2359 	case PR_SET_TIMERSLACK:
2360 		if (arg2 <= 0)
2361 			current->timer_slack_ns =
2362 					current->default_timer_slack_ns;
2363 		else
2364 			current->timer_slack_ns = arg2;
2365 		break;
2366 	case PR_MCE_KILL:
2367 		if (arg4 | arg5)
2368 			return -EINVAL;
2369 		switch (arg2) {
2370 		case PR_MCE_KILL_CLEAR:
2371 			if (arg3 != 0)
2372 				return -EINVAL;
2373 			current->flags &= ~PF_MCE_PROCESS;
2374 			break;
2375 		case PR_MCE_KILL_SET:
2376 			current->flags |= PF_MCE_PROCESS;
2377 			if (arg3 == PR_MCE_KILL_EARLY)
2378 				current->flags |= PF_MCE_EARLY;
2379 			else if (arg3 == PR_MCE_KILL_LATE)
2380 				current->flags &= ~PF_MCE_EARLY;
2381 			else if (arg3 == PR_MCE_KILL_DEFAULT)
2382 				current->flags &=
2383 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2384 			else
2385 				return -EINVAL;
2386 			break;
2387 		default:
2388 			return -EINVAL;
2389 		}
2390 		break;
2391 	case PR_MCE_KILL_GET:
2392 		if (arg2 | arg3 | arg4 | arg5)
2393 			return -EINVAL;
2394 		if (current->flags & PF_MCE_PROCESS)
2395 			error = (current->flags & PF_MCE_EARLY) ?
2396 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2397 		else
2398 			error = PR_MCE_KILL_DEFAULT;
2399 		break;
2400 	case PR_SET_MM:
2401 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2402 		break;
2403 	case PR_GET_TID_ADDRESS:
2404 		error = prctl_get_tid_address(me, (int __user **)arg2);
2405 		break;
2406 	case PR_SET_TIMERSLACK_PID:
2407 		if (task_pid_vnr(current) != (pid_t)arg3 &&
2408 				!capable(CAP_SYS_NICE))
2409 			return -EPERM;
2410 		rcu_read_lock();
2411 		tsk = find_task_by_vpid((pid_t)arg3);
2412 		if (tsk == NULL) {
2413 			rcu_read_unlock();
2414 			return -EINVAL;
2415 		}
2416 		get_task_struct(tsk);
2417 		rcu_read_unlock();
2418 		if (arg2 <= 0)
2419 			tsk->timer_slack_ns =
2420 				tsk->default_timer_slack_ns;
2421 		else
2422 			tsk->timer_slack_ns = arg2;
2423 		put_task_struct(tsk);
2424 		error = 0;
2425 		break;
2426 	case PR_SET_CHILD_SUBREAPER:
2427 		me->signal->is_child_subreaper = !!arg2;
2428 		break;
2429 	case PR_GET_CHILD_SUBREAPER:
2430 		error = put_user(me->signal->is_child_subreaper,
2431 				 (int __user *)arg2);
2432 		break;
2433 	case PR_SET_NO_NEW_PRIVS:
2434 		if (arg2 != 1 || arg3 || arg4 || arg5)
2435 			return -EINVAL;
2436 
2437 		task_set_no_new_privs(current);
2438 		break;
2439 	case PR_GET_NO_NEW_PRIVS:
2440 		if (arg2 || arg3 || arg4 || arg5)
2441 			return -EINVAL;
2442 		return task_no_new_privs(current) ? 1 : 0;
2443 	case PR_SET_FP_MODE:
2444 		error = SET_FP_MODE(me, arg2);
2445 		break;
2446 	case PR_GET_FP_MODE:
2447 		error = GET_FP_MODE(me);
2448 		break;
2449 	case PR_SET_VMA:
2450 		error = prctl_set_vma(arg2, arg3, arg4, arg5);
2451 		break;
2452 	default:
2453 		error = -EINVAL;
2454 		break;
2455 	}
2456 	return error;
2457 }
2458 
SYSCALL_DEFINE3(getcpu,unsigned __user *,cpup,unsigned __user *,nodep,struct getcpu_cache __user *,unused)2459 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2460 		struct getcpu_cache __user *, unused)
2461 {
2462 	int err = 0;
2463 	int cpu = raw_smp_processor_id();
2464 	if (cpup)
2465 		err |= put_user(cpu, cpup);
2466 	if (nodep)
2467 		err |= put_user(cpu_to_node(cpu), nodep);
2468 	return err ? -EFAULT : 0;
2469 }
2470 
2471 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
2472 
__orderly_poweroff(bool force)2473 static int __orderly_poweroff(bool force)
2474 {
2475 	char **argv;
2476 	static char *envp[] = {
2477 		"HOME=/",
2478 		"PATH=/sbin:/bin:/usr/sbin:/usr/bin",
2479 		NULL
2480 	};
2481 	int ret;
2482 
2483 	argv = argv_split(GFP_KERNEL, poweroff_cmd, NULL);
2484 	if (argv) {
2485 		ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
2486 		argv_free(argv);
2487 	} else {
2488 		printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
2489 					 __func__, poweroff_cmd);
2490 		ret = -ENOMEM;
2491 	}
2492 
2493 	if (ret && force) {
2494 		printk(KERN_WARNING "Failed to start orderly shutdown: "
2495 					"forcing the issue\n");
2496 		/*
2497 		 * I guess this should try to kick off some daemon to sync and
2498 		 * poweroff asap.  Or not even bother syncing if we're doing an
2499 		 * emergency shutdown?
2500 		 */
2501 		emergency_sync();
2502 		kernel_power_off();
2503 	}
2504 
2505 	return ret;
2506 }
2507 
2508 static bool poweroff_force;
2509 
poweroff_work_func(struct work_struct * work)2510 static void poweroff_work_func(struct work_struct *work)
2511 {
2512 	__orderly_poweroff(poweroff_force);
2513 }
2514 
2515 static DECLARE_WORK(poweroff_work, poweroff_work_func);
2516 
2517 /**
2518  * orderly_poweroff - Trigger an orderly system poweroff
2519  * @force: force poweroff if command execution fails
2520  *
2521  * This may be called from any context to trigger a system shutdown.
2522  * If the orderly shutdown fails, it will force an immediate shutdown.
2523  */
orderly_poweroff(bool force)2524 int orderly_poweroff(bool force)
2525 {
2526 	if (force) /* do not override the pending "true" */
2527 		poweroff_force = true;
2528 	schedule_work(&poweroff_work);
2529 	return 0;
2530 }
2531 EXPORT_SYMBOL_GPL(orderly_poweroff);
2532 
2533 /**
2534  * do_sysinfo - fill in sysinfo struct
2535  * @info: pointer to buffer to fill
2536  */
do_sysinfo(struct sysinfo * info)2537 static int do_sysinfo(struct sysinfo *info)
2538 {
2539 	unsigned long mem_total, sav_total;
2540 	unsigned int mem_unit, bitcount;
2541 	struct timespec tp;
2542 
2543 	memset(info, 0, sizeof(struct sysinfo));
2544 
2545 	ktime_get_ts(&tp);
2546 	monotonic_to_bootbased(&tp);
2547 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2548 
2549 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2550 
2551 	info->procs = nr_threads;
2552 
2553 	si_meminfo(info);
2554 	si_swapinfo(info);
2555 
2556 	/*
2557 	 * If the sum of all the available memory (i.e. ram + swap)
2558 	 * is less than can be stored in a 32 bit unsigned long then
2559 	 * we can be binary compatible with 2.2.x kernels.  If not,
2560 	 * well, in that case 2.2.x was broken anyways...
2561 	 *
2562 	 *  -Erik Andersen <andersee@debian.org>
2563 	 */
2564 
2565 	mem_total = info->totalram + info->totalswap;
2566 	if (mem_total < info->totalram || mem_total < info->totalswap)
2567 		goto out;
2568 	bitcount = 0;
2569 	mem_unit = info->mem_unit;
2570 	while (mem_unit > 1) {
2571 		bitcount++;
2572 		mem_unit >>= 1;
2573 		sav_total = mem_total;
2574 		mem_total <<= 1;
2575 		if (mem_total < sav_total)
2576 			goto out;
2577 	}
2578 
2579 	/*
2580 	 * If mem_total did not overflow, multiply all memory values by
2581 	 * info->mem_unit and set it to 1.  This leaves things compatible
2582 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2583 	 * kernels...
2584 	 */
2585 
2586 	info->mem_unit = 1;
2587 	info->totalram <<= bitcount;
2588 	info->freeram <<= bitcount;
2589 	info->sharedram <<= bitcount;
2590 	info->bufferram <<= bitcount;
2591 	info->totalswap <<= bitcount;
2592 	info->freeswap <<= bitcount;
2593 	info->totalhigh <<= bitcount;
2594 	info->freehigh <<= bitcount;
2595 
2596 out:
2597 	return 0;
2598 }
2599 
SYSCALL_DEFINE1(sysinfo,struct sysinfo __user *,info)2600 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2601 {
2602 	struct sysinfo val;
2603 
2604 	do_sysinfo(&val);
2605 
2606 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2607 		return -EFAULT;
2608 
2609 	return 0;
2610 }
2611 
2612 #ifdef CONFIG_COMPAT
2613 struct compat_sysinfo {
2614 	s32 uptime;
2615 	u32 loads[3];
2616 	u32 totalram;
2617 	u32 freeram;
2618 	u32 sharedram;
2619 	u32 bufferram;
2620 	u32 totalswap;
2621 	u32 freeswap;
2622 	u16 procs;
2623 	u16 pad;
2624 	u32 totalhigh;
2625 	u32 freehigh;
2626 	u32 mem_unit;
2627 	char _f[20-2*sizeof(u32)-sizeof(int)];
2628 };
2629 
COMPAT_SYSCALL_DEFINE1(sysinfo,struct compat_sysinfo __user *,info)2630 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2631 {
2632 	struct sysinfo s;
2633 
2634 	do_sysinfo(&s);
2635 
2636 	/* Check to see if any memory value is too large for 32-bit and scale
2637 	 *  down if needed
2638 	 */
2639 	if ((s.totalram >> 32) || (s.totalswap >> 32)) {
2640 		int bitcount = 0;
2641 
2642 		while (s.mem_unit < PAGE_SIZE) {
2643 			s.mem_unit <<= 1;
2644 			bitcount++;
2645 		}
2646 
2647 		s.totalram >>= bitcount;
2648 		s.freeram >>= bitcount;
2649 		s.sharedram >>= bitcount;
2650 		s.bufferram >>= bitcount;
2651 		s.totalswap >>= bitcount;
2652 		s.freeswap >>= bitcount;
2653 		s.totalhigh >>= bitcount;
2654 		s.freehigh >>= bitcount;
2655 	}
2656 
2657 	if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2658 	    __put_user(s.uptime, &info->uptime) ||
2659 	    __put_user(s.loads[0], &info->loads[0]) ||
2660 	    __put_user(s.loads[1], &info->loads[1]) ||
2661 	    __put_user(s.loads[2], &info->loads[2]) ||
2662 	    __put_user(s.totalram, &info->totalram) ||
2663 	    __put_user(s.freeram, &info->freeram) ||
2664 	    __put_user(s.sharedram, &info->sharedram) ||
2665 	    __put_user(s.bufferram, &info->bufferram) ||
2666 	    __put_user(s.totalswap, &info->totalswap) ||
2667 	    __put_user(s.freeswap, &info->freeswap) ||
2668 	    __put_user(s.procs, &info->procs) ||
2669 	    __put_user(s.totalhigh, &info->totalhigh) ||
2670 	    __put_user(s.freehigh, &info->freehigh) ||
2671 	    __put_user(s.mem_unit, &info->mem_unit))
2672 		return -EFAULT;
2673 
2674 	return 0;
2675 }
2676 #endif /* CONFIG_COMPAT */
2677