1/*
2 * linux/kernel/seccomp.c
3 *
4 * Copyright 2004-2005  Andrea Arcangeli <andrea@cpushare.com>
5 *
6 * Copyright (C) 2012 Google, Inc.
7 * Will Drewry <wad@chromium.org>
8 *
9 * This defines a simple but solid secure-computing facility.
10 *
11 * Mode 1 uses a fixed list of allowed system calls.
12 * Mode 2 allows user-defined system call filters in the form
13 *        of Berkeley Packet Filters/Linux Socket Filters.
14 */
15
16#include <linux/atomic.h>
17#include <linux/audit.h>
18#include <linux/compat.h>
19#include <linux/sched.h>
20#include <linux/seccomp.h>
21#include <linux/slab.h>
22#include <linux/syscalls.h>
23
24#ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
25#include <asm/syscall.h>
26#endif
27
28#ifdef CONFIG_SECCOMP_FILTER
29#include <linux/filter.h>
30#include <linux/pid.h>
31#include <linux/ptrace.h>
32#include <linux/security.h>
33#include <linux/tracehook.h>
34#include <linux/uaccess.h>
35
36/**
37 * struct seccomp_filter - container for seccomp BPF programs
38 *
39 * @usage: reference count to manage the object lifetime.
40 *         get/put helpers should be used when accessing an instance
41 *         outside of a lifetime-guarded section.  In general, this
42 *         is only needed for handling filters shared across tasks.
43 * @prev: points to a previously installed, or inherited, filter
44 * @len: the number of instructions in the program
45 * @insnsi: the BPF program instructions to evaluate
46 *
47 * seccomp_filter objects are organized in a tree linked via the @prev
48 * pointer.  For any task, it appears to be a singly-linked list starting
49 * with current->seccomp.filter, the most recently attached or inherited filter.
50 * However, multiple filters may share a @prev node, by way of fork(), which
51 * results in a unidirectional tree existing in memory.  This is similar to
52 * how namespaces work.
53 *
54 * seccomp_filter objects should never be modified after being attached
55 * to a task_struct (other than @usage).
56 */
57struct seccomp_filter {
58	atomic_t usage;
59	struct seccomp_filter *prev;
60	struct bpf_prog *prog;
61};
62
63/* Limit any path through the tree to 256KB worth of instructions. */
64#define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
65
66/*
67 * Endianness is explicitly ignored and left for BPF program authors to manage
68 * as per the specific architecture.
69 */
70static void populate_seccomp_data(struct seccomp_data *sd)
71{
72	struct task_struct *task = current;
73	struct pt_regs *regs = task_pt_regs(task);
74	unsigned long args[6];
75
76	sd->nr = syscall_get_nr(task, regs);
77	sd->arch = syscall_get_arch();
78	syscall_get_arguments(task, regs, 0, 6, args);
79	sd->args[0] = args[0];
80	sd->args[1] = args[1];
81	sd->args[2] = args[2];
82	sd->args[3] = args[3];
83	sd->args[4] = args[4];
84	sd->args[5] = args[5];
85	sd->instruction_pointer = KSTK_EIP(task);
86}
87
88/**
89 *	seccomp_check_filter - verify seccomp filter code
90 *	@filter: filter to verify
91 *	@flen: length of filter
92 *
93 * Takes a previously checked filter (by bpf_check_classic) and
94 * redirects all filter code that loads struct sk_buff data
95 * and related data through seccomp_bpf_load.  It also
96 * enforces length and alignment checking of those loads.
97 *
98 * Returns 0 if the rule set is legal or -EINVAL if not.
99 */
100static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
101{
102	int pc;
103	for (pc = 0; pc < flen; pc++) {
104		struct sock_filter *ftest = &filter[pc];
105		u16 code = ftest->code;
106		u32 k = ftest->k;
107
108		switch (code) {
109		case BPF_LD | BPF_W | BPF_ABS:
110			ftest->code = BPF_LDX | BPF_W | BPF_ABS;
111			/* 32-bit aligned and not out of bounds. */
112			if (k >= sizeof(struct seccomp_data) || k & 3)
113				return -EINVAL;
114			continue;
115		case BPF_LD | BPF_W | BPF_LEN:
116			ftest->code = BPF_LD | BPF_IMM;
117			ftest->k = sizeof(struct seccomp_data);
118			continue;
119		case BPF_LDX | BPF_W | BPF_LEN:
120			ftest->code = BPF_LDX | BPF_IMM;
121			ftest->k = sizeof(struct seccomp_data);
122			continue;
123		/* Explicitly include allowed calls. */
124		case BPF_RET | BPF_K:
125		case BPF_RET | BPF_A:
126		case BPF_ALU | BPF_ADD | BPF_K:
127		case BPF_ALU | BPF_ADD | BPF_X:
128		case BPF_ALU | BPF_SUB | BPF_K:
129		case BPF_ALU | BPF_SUB | BPF_X:
130		case BPF_ALU | BPF_MUL | BPF_K:
131		case BPF_ALU | BPF_MUL | BPF_X:
132		case BPF_ALU | BPF_DIV | BPF_K:
133		case BPF_ALU | BPF_DIV | BPF_X:
134		case BPF_ALU | BPF_AND | BPF_K:
135		case BPF_ALU | BPF_AND | BPF_X:
136		case BPF_ALU | BPF_OR | BPF_K:
137		case BPF_ALU | BPF_OR | BPF_X:
138		case BPF_ALU | BPF_XOR | BPF_K:
139		case BPF_ALU | BPF_XOR | BPF_X:
140		case BPF_ALU | BPF_LSH | BPF_K:
141		case BPF_ALU | BPF_LSH | BPF_X:
142		case BPF_ALU | BPF_RSH | BPF_K:
143		case BPF_ALU | BPF_RSH | BPF_X:
144		case BPF_ALU | BPF_NEG:
145		case BPF_LD | BPF_IMM:
146		case BPF_LDX | BPF_IMM:
147		case BPF_MISC | BPF_TAX:
148		case BPF_MISC | BPF_TXA:
149		case BPF_LD | BPF_MEM:
150		case BPF_LDX | BPF_MEM:
151		case BPF_ST:
152		case BPF_STX:
153		case BPF_JMP | BPF_JA:
154		case BPF_JMP | BPF_JEQ | BPF_K:
155		case BPF_JMP | BPF_JEQ | BPF_X:
156		case BPF_JMP | BPF_JGE | BPF_K:
157		case BPF_JMP | BPF_JGE | BPF_X:
158		case BPF_JMP | BPF_JGT | BPF_K:
159		case BPF_JMP | BPF_JGT | BPF_X:
160		case BPF_JMP | BPF_JSET | BPF_K:
161		case BPF_JMP | BPF_JSET | BPF_X:
162			continue;
163		default:
164			return -EINVAL;
165		}
166	}
167	return 0;
168}
169
170/**
171 * seccomp_run_filters - evaluates all seccomp filters against @syscall
172 * @syscall: number of the current system call
173 *
174 * Returns valid seccomp BPF response codes.
175 */
176static u32 seccomp_run_filters(struct seccomp_data *sd)
177{
178	struct seccomp_filter *f = ACCESS_ONCE(current->seccomp.filter);
179	struct seccomp_data sd_local;
180	u32 ret = SECCOMP_RET_ALLOW;
181
182	/* Ensure unexpected behavior doesn't result in failing open. */
183	if (unlikely(WARN_ON(f == NULL)))
184		return SECCOMP_RET_KILL;
185
186	/* Make sure cross-thread synced filter points somewhere sane. */
187	smp_read_barrier_depends();
188
189	if (!sd) {
190		populate_seccomp_data(&sd_local);
191		sd = &sd_local;
192	}
193
194	/*
195	 * All filters in the list are evaluated and the lowest BPF return
196	 * value always takes priority (ignoring the DATA).
197	 */
198	for (; f; f = f->prev) {
199		u32 cur_ret = BPF_PROG_RUN(f->prog, (void *)sd);
200
201		if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
202			ret = cur_ret;
203	}
204	return ret;
205}
206#endif /* CONFIG_SECCOMP_FILTER */
207
208static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
209{
210	assert_spin_locked(&current->sighand->siglock);
211
212	if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
213		return false;
214
215	return true;
216}
217
218static inline void seccomp_assign_mode(struct task_struct *task,
219				       unsigned long seccomp_mode)
220{
221	assert_spin_locked(&task->sighand->siglock);
222
223	task->seccomp.mode = seccomp_mode;
224	/*
225	 * Make sure TIF_SECCOMP cannot be set before the mode (and
226	 * filter) is set.
227	 */
228	smp_mb__before_atomic();
229	set_tsk_thread_flag(task, TIF_SECCOMP);
230}
231
232#ifdef CONFIG_SECCOMP_FILTER
233/* Returns 1 if the parent is an ancestor of the child. */
234static int is_ancestor(struct seccomp_filter *parent,
235		       struct seccomp_filter *child)
236{
237	/* NULL is the root ancestor. */
238	if (parent == NULL)
239		return 1;
240	for (; child; child = child->prev)
241		if (child == parent)
242			return 1;
243	return 0;
244}
245
246/**
247 * seccomp_can_sync_threads: checks if all threads can be synchronized
248 *
249 * Expects sighand and cred_guard_mutex locks to be held.
250 *
251 * Returns 0 on success, -ve on error, or the pid of a thread which was
252 * either not in the correct seccomp mode or it did not have an ancestral
253 * seccomp filter.
254 */
255static inline pid_t seccomp_can_sync_threads(void)
256{
257	struct task_struct *thread, *caller;
258
259	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
260	assert_spin_locked(&current->sighand->siglock);
261
262	/* Validate all threads being eligible for synchronization. */
263	caller = current;
264	for_each_thread(caller, thread) {
265		pid_t failed;
266
267		/* Skip current, since it is initiating the sync. */
268		if (thread == caller)
269			continue;
270
271		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
272		    (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
273		     is_ancestor(thread->seccomp.filter,
274				 caller->seccomp.filter)))
275			continue;
276
277		/* Return the first thread that cannot be synchronized. */
278		failed = task_pid_vnr(thread);
279		/* If the pid cannot be resolved, then return -ESRCH */
280		if (unlikely(WARN_ON(failed == 0)))
281			failed = -ESRCH;
282		return failed;
283	}
284
285	return 0;
286}
287
288/**
289 * seccomp_sync_threads: sets all threads to use current's filter
290 *
291 * Expects sighand and cred_guard_mutex locks to be held, and for
292 * seccomp_can_sync_threads() to have returned success already
293 * without dropping the locks.
294 *
295 */
296static inline void seccomp_sync_threads(void)
297{
298	struct task_struct *thread, *caller;
299
300	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
301	assert_spin_locked(&current->sighand->siglock);
302
303	/* Synchronize all threads. */
304	caller = current;
305	for_each_thread(caller, thread) {
306		/* Skip current, since it needs no changes. */
307		if (thread == caller)
308			continue;
309
310		/* Get a task reference for the new leaf node. */
311		get_seccomp_filter(caller);
312		/*
313		 * Drop the task reference to the shared ancestor since
314		 * current's path will hold a reference.  (This also
315		 * allows a put before the assignment.)
316		 */
317		put_seccomp_filter(thread);
318		smp_store_release(&thread->seccomp.filter,
319				  caller->seccomp.filter);
320		/*
321		 * Opt the other thread into seccomp if needed.
322		 * As threads are considered to be trust-realm
323		 * equivalent (see ptrace_may_access), it is safe to
324		 * allow one thread to transition the other.
325		 */
326		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) {
327			/*
328			 * Don't let an unprivileged task work around
329			 * the no_new_privs restriction by creating
330			 * a thread that sets it up, enters seccomp,
331			 * then dies.
332			 */
333			if (task_no_new_privs(caller))
334				task_set_no_new_privs(thread);
335
336			seccomp_assign_mode(thread, SECCOMP_MODE_FILTER);
337		}
338	}
339}
340
341/**
342 * seccomp_prepare_filter: Prepares a seccomp filter for use.
343 * @fprog: BPF program to install
344 *
345 * Returns filter on success or an ERR_PTR on failure.
346 */
347static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
348{
349	struct seccomp_filter *filter;
350	unsigned long fp_size;
351	struct sock_filter *fp;
352	int new_len;
353	long ret;
354
355	if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
356		return ERR_PTR(-EINVAL);
357	BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
358	fp_size = fprog->len * sizeof(struct sock_filter);
359
360	/*
361	 * Installing a seccomp filter requires that the task has
362	 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
363	 * This avoids scenarios where unprivileged tasks can affect the
364	 * behavior of privileged children.
365	 */
366	if (!task_no_new_privs(current) &&
367	    security_capable_noaudit(current_cred(), current_user_ns(),
368				     CAP_SYS_ADMIN) != 0)
369		return ERR_PTR(-EACCES);
370
371	fp = kzalloc(fp_size, GFP_KERNEL|__GFP_NOWARN);
372	if (!fp)
373		return ERR_PTR(-ENOMEM);
374
375	/* Copy the instructions from fprog. */
376	ret = -EFAULT;
377	if (copy_from_user(fp, fprog->filter, fp_size))
378		goto free_prog;
379
380	/* Check and rewrite the fprog via the skb checker */
381	ret = bpf_check_classic(fp, fprog->len);
382	if (ret)
383		goto free_prog;
384
385	/* Check and rewrite the fprog for seccomp use */
386	ret = seccomp_check_filter(fp, fprog->len);
387	if (ret)
388		goto free_prog;
389
390	/* Convert 'sock_filter' insns to 'bpf_insn' insns */
391	ret = bpf_convert_filter(fp, fprog->len, NULL, &new_len);
392	if (ret)
393		goto free_prog;
394
395	/* Allocate a new seccomp_filter */
396	ret = -ENOMEM;
397	filter = kzalloc(sizeof(struct seccomp_filter),
398			 GFP_KERNEL|__GFP_NOWARN);
399	if (!filter)
400		goto free_prog;
401
402	filter->prog = bpf_prog_alloc(bpf_prog_size(new_len), __GFP_NOWARN);
403	if (!filter->prog)
404		goto free_filter;
405
406	ret = bpf_convert_filter(fp, fprog->len, filter->prog->insnsi, &new_len);
407	if (ret)
408		goto free_filter_prog;
409
410	kfree(fp);
411	atomic_set(&filter->usage, 1);
412	filter->prog->len = new_len;
413
414	bpf_prog_select_runtime(filter->prog);
415
416	return filter;
417
418free_filter_prog:
419	__bpf_prog_free(filter->prog);
420free_filter:
421	kfree(filter);
422free_prog:
423	kfree(fp);
424	return ERR_PTR(ret);
425}
426
427/**
428 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
429 * @user_filter: pointer to the user data containing a sock_fprog.
430 *
431 * Returns 0 on success and non-zero otherwise.
432 */
433static struct seccomp_filter *
434seccomp_prepare_user_filter(const char __user *user_filter)
435{
436	struct sock_fprog fprog;
437	struct seccomp_filter *filter = ERR_PTR(-EFAULT);
438
439#ifdef CONFIG_COMPAT
440	if (is_compat_task()) {
441		struct compat_sock_fprog fprog32;
442		if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
443			goto out;
444		fprog.len = fprog32.len;
445		fprog.filter = compat_ptr(fprog32.filter);
446	} else /* falls through to the if below. */
447#endif
448	if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
449		goto out;
450	filter = seccomp_prepare_filter(&fprog);
451out:
452	return filter;
453}
454
455/**
456 * seccomp_attach_filter: validate and attach filter
457 * @flags:  flags to change filter behavior
458 * @filter: seccomp filter to add to the current process
459 *
460 * Caller must be holding current->sighand->siglock lock.
461 *
462 * Returns 0 on success, -ve on error.
463 */
464static long seccomp_attach_filter(unsigned int flags,
465				  struct seccomp_filter *filter)
466{
467	unsigned long total_insns;
468	struct seccomp_filter *walker;
469
470	assert_spin_locked(&current->sighand->siglock);
471
472	/* Validate resulting filter length. */
473	total_insns = filter->prog->len;
474	for (walker = current->seccomp.filter; walker; walker = walker->prev)
475		total_insns += walker->prog->len + 4;  /* 4 instr penalty */
476	if (total_insns > MAX_INSNS_PER_PATH)
477		return -ENOMEM;
478
479	/* If thread sync has been requested, check that it is possible. */
480	if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
481		int ret;
482
483		ret = seccomp_can_sync_threads();
484		if (ret)
485			return ret;
486	}
487
488	/*
489	 * If there is an existing filter, make it the prev and don't drop its
490	 * task reference.
491	 */
492	filter->prev = current->seccomp.filter;
493	current->seccomp.filter = filter;
494
495	/* Now that the new filter is in place, synchronize to all threads. */
496	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
497		seccomp_sync_threads();
498
499	return 0;
500}
501
502/* get_seccomp_filter - increments the reference count of the filter on @tsk */
503void get_seccomp_filter(struct task_struct *tsk)
504{
505	struct seccomp_filter *orig = tsk->seccomp.filter;
506	if (!orig)
507		return;
508	/* Reference count is bounded by the number of total processes. */
509	atomic_inc(&orig->usage);
510}
511
512static inline void seccomp_filter_free(struct seccomp_filter *filter)
513{
514	if (filter) {
515		bpf_prog_free(filter->prog);
516		kfree(filter);
517	}
518}
519
520/* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
521void put_seccomp_filter(struct task_struct *tsk)
522{
523	struct seccomp_filter *orig = tsk->seccomp.filter;
524	/* Clean up single-reference branches iteratively. */
525	while (orig && atomic_dec_and_test(&orig->usage)) {
526		struct seccomp_filter *freeme = orig;
527		orig = orig->prev;
528		seccomp_filter_free(freeme);
529	}
530}
531
532/**
533 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
534 * @syscall: syscall number to send to userland
535 * @reason: filter-supplied reason code to send to userland (via si_errno)
536 *
537 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
538 */
539static void seccomp_send_sigsys(int syscall, int reason)
540{
541	struct siginfo info;
542	memset(&info, 0, sizeof(info));
543	info.si_signo = SIGSYS;
544	info.si_code = SYS_SECCOMP;
545	info.si_call_addr = (void __user *)KSTK_EIP(current);
546	info.si_errno = reason;
547	info.si_arch = syscall_get_arch();
548	info.si_syscall = syscall;
549	force_sig_info(SIGSYS, &info, current);
550}
551#endif	/* CONFIG_SECCOMP_FILTER */
552
553/*
554 * Secure computing mode 1 allows only read/write/exit/sigreturn.
555 * To be fully secure this must be combined with rlimit
556 * to limit the stack allocations too.
557 */
558static int mode1_syscalls[] = {
559	__NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
560	0, /* null terminated */
561};
562
563#ifdef CONFIG_COMPAT
564static int mode1_syscalls_32[] = {
565	__NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
566	0, /* null terminated */
567};
568#endif
569
570static void __secure_computing_strict(int this_syscall)
571{
572	int *syscall_whitelist = mode1_syscalls;
573#ifdef CONFIG_COMPAT
574	if (is_compat_task())
575		syscall_whitelist = mode1_syscalls_32;
576#endif
577	do {
578		if (*syscall_whitelist == this_syscall)
579			return;
580	} while (*++syscall_whitelist);
581
582#ifdef SECCOMP_DEBUG
583	dump_stack();
584#endif
585	audit_seccomp(this_syscall, SIGKILL, SECCOMP_RET_KILL);
586	do_exit(SIGKILL);
587}
588
589#ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
590void secure_computing_strict(int this_syscall)
591{
592	int mode = current->seccomp.mode;
593
594	if (mode == 0)
595		return;
596	else if (mode == SECCOMP_MODE_STRICT)
597		__secure_computing_strict(this_syscall);
598	else
599		BUG();
600}
601#else
602int __secure_computing(void)
603{
604	u32 phase1_result = seccomp_phase1(NULL);
605
606	if (likely(phase1_result == SECCOMP_PHASE1_OK))
607		return 0;
608	else if (likely(phase1_result == SECCOMP_PHASE1_SKIP))
609		return -1;
610	else
611		return seccomp_phase2(phase1_result);
612}
613
614#ifdef CONFIG_SECCOMP_FILTER
615static u32 __seccomp_phase1_filter(int this_syscall, struct seccomp_data *sd)
616{
617	u32 filter_ret, action;
618	int data;
619
620	/*
621	 * Make sure that any changes to mode from another thread have
622	 * been seen after TIF_SECCOMP was seen.
623	 */
624	rmb();
625
626	filter_ret = seccomp_run_filters(sd);
627	data = filter_ret & SECCOMP_RET_DATA;
628	action = filter_ret & SECCOMP_RET_ACTION;
629
630	switch (action) {
631	case SECCOMP_RET_ERRNO:
632		/* Set the low-order 16-bits as a errno. */
633		syscall_set_return_value(current, task_pt_regs(current),
634					 -data, 0);
635		goto skip;
636
637	case SECCOMP_RET_TRAP:
638		/* Show the handler the original registers. */
639		syscall_rollback(current, task_pt_regs(current));
640		/* Let the filter pass back 16 bits of data. */
641		seccomp_send_sigsys(this_syscall, data);
642		goto skip;
643
644	case SECCOMP_RET_TRACE:
645		return filter_ret;  /* Save the rest for phase 2. */
646
647	case SECCOMP_RET_ALLOW:
648		return SECCOMP_PHASE1_OK;
649
650	case SECCOMP_RET_KILL:
651	default:
652		audit_seccomp(this_syscall, SIGSYS, action);
653		do_exit(SIGSYS);
654	}
655
656	unreachable();
657
658skip:
659	audit_seccomp(this_syscall, 0, action);
660	return SECCOMP_PHASE1_SKIP;
661}
662#endif
663
664/**
665 * seccomp_phase1() - run fast path seccomp checks on the current syscall
666 * @arg sd: The seccomp_data or NULL
667 *
668 * This only reads pt_regs via the syscall_xyz helpers.  The only change
669 * it will make to pt_regs is via syscall_set_return_value, and it will
670 * only do that if it returns SECCOMP_PHASE1_SKIP.
671 *
672 * If sd is provided, it will not read pt_regs at all.
673 *
674 * It may also call do_exit or force a signal; these actions must be
675 * safe.
676 *
677 * If it returns SECCOMP_PHASE1_OK, the syscall passes checks and should
678 * be processed normally.
679 *
680 * If it returns SECCOMP_PHASE1_SKIP, then the syscall should not be
681 * invoked.  In this case, seccomp_phase1 will have set the return value
682 * using syscall_set_return_value.
683 *
684 * If it returns anything else, then the return value should be passed
685 * to seccomp_phase2 from a context in which ptrace hooks are safe.
686 */
687u32 seccomp_phase1(struct seccomp_data *sd)
688{
689	int mode = current->seccomp.mode;
690	int this_syscall = sd ? sd->nr :
691		syscall_get_nr(current, task_pt_regs(current));
692
693	switch (mode) {
694	case SECCOMP_MODE_STRICT:
695		__secure_computing_strict(this_syscall);  /* may call do_exit */
696		return SECCOMP_PHASE1_OK;
697#ifdef CONFIG_SECCOMP_FILTER
698	case SECCOMP_MODE_FILTER:
699		return __seccomp_phase1_filter(this_syscall, sd);
700#endif
701	default:
702		BUG();
703	}
704}
705
706/**
707 * seccomp_phase2() - finish slow path seccomp work for the current syscall
708 * @phase1_result: The return value from seccomp_phase1()
709 *
710 * This must be called from a context in which ptrace hooks can be used.
711 *
712 * Returns 0 if the syscall should be processed or -1 to skip the syscall.
713 */
714int seccomp_phase2(u32 phase1_result)
715{
716	struct pt_regs *regs = task_pt_regs(current);
717	u32 action = phase1_result & SECCOMP_RET_ACTION;
718	int data = phase1_result & SECCOMP_RET_DATA;
719
720	BUG_ON(action != SECCOMP_RET_TRACE);
721
722	audit_seccomp(syscall_get_nr(current, regs), 0, action);
723
724	/* Skip these calls if there is no tracer. */
725	if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
726		syscall_set_return_value(current, regs,
727					 -ENOSYS, 0);
728		return -1;
729	}
730
731	/* Allow the BPF to provide the event message */
732	ptrace_event(PTRACE_EVENT_SECCOMP, data);
733	/*
734	 * The delivery of a fatal signal during event
735	 * notification may silently skip tracer notification.
736	 * Terminating the task now avoids executing a system
737	 * call that may not be intended.
738	 */
739	if (fatal_signal_pending(current))
740		do_exit(SIGSYS);
741	if (syscall_get_nr(current, regs) < 0)
742		return -1;  /* Explicit request to skip. */
743
744	return 0;
745}
746#endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
747
748long prctl_get_seccomp(void)
749{
750	return current->seccomp.mode;
751}
752
753/**
754 * seccomp_set_mode_strict: internal function for setting strict seccomp
755 *
756 * Once current->seccomp.mode is non-zero, it may not be changed.
757 *
758 * Returns 0 on success or -EINVAL on failure.
759 */
760static long seccomp_set_mode_strict(void)
761{
762	const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
763	long ret = -EINVAL;
764
765	spin_lock_irq(&current->sighand->siglock);
766
767	if (!seccomp_may_assign_mode(seccomp_mode))
768		goto out;
769
770#ifdef TIF_NOTSC
771	disable_TSC();
772#endif
773	seccomp_assign_mode(current, seccomp_mode);
774	ret = 0;
775
776out:
777	spin_unlock_irq(&current->sighand->siglock);
778
779	return ret;
780}
781
782#ifdef CONFIG_SECCOMP_FILTER
783/**
784 * seccomp_set_mode_filter: internal function for setting seccomp filter
785 * @flags:  flags to change filter behavior
786 * @filter: struct sock_fprog containing filter
787 *
788 * This function may be called repeatedly to install additional filters.
789 * Every filter successfully installed will be evaluated (in reverse order)
790 * for each system call the task makes.
791 *
792 * Once current->seccomp.mode is non-zero, it may not be changed.
793 *
794 * Returns 0 on success or -EINVAL on failure.
795 */
796static long seccomp_set_mode_filter(unsigned int flags,
797				    const char __user *filter)
798{
799	const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
800	struct seccomp_filter *prepared = NULL;
801	long ret = -EINVAL;
802
803	/* Validate flags. */
804	if (flags & ~SECCOMP_FILTER_FLAG_MASK)
805		return -EINVAL;
806
807	/* Prepare the new filter before holding any locks. */
808	prepared = seccomp_prepare_user_filter(filter);
809	if (IS_ERR(prepared))
810		return PTR_ERR(prepared);
811
812	/*
813	 * Make sure we cannot change seccomp or nnp state via TSYNC
814	 * while another thread is in the middle of calling exec.
815	 */
816	if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
817	    mutex_lock_killable(&current->signal->cred_guard_mutex))
818		goto out_free;
819
820	spin_lock_irq(&current->sighand->siglock);
821
822	if (!seccomp_may_assign_mode(seccomp_mode))
823		goto out;
824
825	ret = seccomp_attach_filter(flags, prepared);
826	if (ret)
827		goto out;
828	/* Do not free the successfully attached filter. */
829	prepared = NULL;
830
831	seccomp_assign_mode(current, seccomp_mode);
832out:
833	spin_unlock_irq(&current->sighand->siglock);
834	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
835		mutex_unlock(&current->signal->cred_guard_mutex);
836out_free:
837	seccomp_filter_free(prepared);
838	return ret;
839}
840#else
841static inline long seccomp_set_mode_filter(unsigned int flags,
842					   const char __user *filter)
843{
844	return -EINVAL;
845}
846#endif
847
848/* Common entry point for both prctl and syscall. */
849static long do_seccomp(unsigned int op, unsigned int flags,
850		       const char __user *uargs)
851{
852	switch (op) {
853	case SECCOMP_SET_MODE_STRICT:
854		if (flags != 0 || uargs != NULL)
855			return -EINVAL;
856		return seccomp_set_mode_strict();
857	case SECCOMP_SET_MODE_FILTER:
858		return seccomp_set_mode_filter(flags, uargs);
859	default:
860		return -EINVAL;
861	}
862}
863
864SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
865			 const char __user *, uargs)
866{
867	return do_seccomp(op, flags, uargs);
868}
869
870/**
871 * prctl_set_seccomp: configures current->seccomp.mode
872 * @seccomp_mode: requested mode to use
873 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
874 *
875 * Returns 0 on success or -EINVAL on failure.
876 */
877long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
878{
879	unsigned int op;
880	char __user *uargs;
881
882	switch (seccomp_mode) {
883	case SECCOMP_MODE_STRICT:
884		op = SECCOMP_SET_MODE_STRICT;
885		/*
886		 * Setting strict mode through prctl always ignored filter,
887		 * so make sure it is always NULL here to pass the internal
888		 * check in do_seccomp().
889		 */
890		uargs = NULL;
891		break;
892	case SECCOMP_MODE_FILTER:
893		op = SECCOMP_SET_MODE_FILTER;
894		uargs = filter;
895		break;
896	default:
897		return -EINVAL;
898	}
899
900	/* prctl interface doesn't have flags, so they are always zero. */
901	return do_seccomp(op, 0, uargs);
902}
903