xref: /openbmc/linux/kernel/seccomp.c (revision 133f9794)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/kernel/seccomp.c
4  *
5  * Copyright 2004-2005  Andrea Arcangeli <andrea@cpushare.com>
6  *
7  * Copyright (C) 2012 Google, Inc.
8  * Will Drewry <wad@chromium.org>
9  *
10  * This defines a simple but solid secure-computing facility.
11  *
12  * Mode 1 uses a fixed list of allowed system calls.
13  * Mode 2 allows user-defined system call filters in the form
14  *        of Berkeley Packet Filters/Linux Socket Filters.
15  */
16 
17 #include <linux/refcount.h>
18 #include <linux/audit.h>
19 #include <linux/compat.h>
20 #include <linux/coredump.h>
21 #include <linux/kmemleak.h>
22 #include <linux/sched.h>
23 #include <linux/sched/task_stack.h>
24 #include <linux/seccomp.h>
25 #include <linux/slab.h>
26 #include <linux/syscalls.h>
27 #include <linux/sysctl.h>
28 
29 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
30 #include <asm/syscall.h>
31 #endif
32 
33 #ifdef CONFIG_SECCOMP_FILTER
34 #include <linux/filter.h>
35 #include <linux/pid.h>
36 #include <linux/ptrace.h>
37 #include <linux/security.h>
38 #include <linux/tracehook.h>
39 #include <linux/uaccess.h>
40 
41 /**
42  * struct seccomp_filter - container for seccomp BPF programs
43  *
44  * @usage: reference count to manage the object lifetime.
45  *         get/put helpers should be used when accessing an instance
46  *         outside of a lifetime-guarded section.  In general, this
47  *         is only needed for handling filters shared across tasks.
48  * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
49  * @prev: points to a previously installed, or inherited, filter
50  * @prog: the BPF program to evaluate
51  *
52  * seccomp_filter objects are organized in a tree linked via the @prev
53  * pointer.  For any task, it appears to be a singly-linked list starting
54  * with current->seccomp.filter, the most recently attached or inherited filter.
55  * However, multiple filters may share a @prev node, by way of fork(), which
56  * results in a unidirectional tree existing in memory.  This is similar to
57  * how namespaces work.
58  *
59  * seccomp_filter objects should never be modified after being attached
60  * to a task_struct (other than @usage).
61  */
62 struct seccomp_filter {
63 	refcount_t usage;
64 	bool log;
65 	struct seccomp_filter *prev;
66 	struct bpf_prog *prog;
67 };
68 
69 /* Limit any path through the tree to 256KB worth of instructions. */
70 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
71 
72 /*
73  * Endianness is explicitly ignored and left for BPF program authors to manage
74  * as per the specific architecture.
75  */
76 static void populate_seccomp_data(struct seccomp_data *sd)
77 {
78 	struct task_struct *task = current;
79 	struct pt_regs *regs = task_pt_regs(task);
80 	unsigned long args[6];
81 
82 	sd->nr = syscall_get_nr(task, regs);
83 	sd->arch = syscall_get_arch();
84 	syscall_get_arguments(task, regs, 0, 6, args);
85 	sd->args[0] = args[0];
86 	sd->args[1] = args[1];
87 	sd->args[2] = args[2];
88 	sd->args[3] = args[3];
89 	sd->args[4] = args[4];
90 	sd->args[5] = args[5];
91 	sd->instruction_pointer = KSTK_EIP(task);
92 }
93 
94 /**
95  *	seccomp_check_filter - verify seccomp filter code
96  *	@filter: filter to verify
97  *	@flen: length of filter
98  *
99  * Takes a previously checked filter (by bpf_check_classic) and
100  * redirects all filter code that loads struct sk_buff data
101  * and related data through seccomp_bpf_load.  It also
102  * enforces length and alignment checking of those loads.
103  *
104  * Returns 0 if the rule set is legal or -EINVAL if not.
105  */
106 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
107 {
108 	int pc;
109 	for (pc = 0; pc < flen; pc++) {
110 		struct sock_filter *ftest = &filter[pc];
111 		u16 code = ftest->code;
112 		u32 k = ftest->k;
113 
114 		switch (code) {
115 		case BPF_LD | BPF_W | BPF_ABS:
116 			ftest->code = BPF_LDX | BPF_W | BPF_ABS;
117 			/* 32-bit aligned and not out of bounds. */
118 			if (k >= sizeof(struct seccomp_data) || k & 3)
119 				return -EINVAL;
120 			continue;
121 		case BPF_LD | BPF_W | BPF_LEN:
122 			ftest->code = BPF_LD | BPF_IMM;
123 			ftest->k = sizeof(struct seccomp_data);
124 			continue;
125 		case BPF_LDX | BPF_W | BPF_LEN:
126 			ftest->code = BPF_LDX | BPF_IMM;
127 			ftest->k = sizeof(struct seccomp_data);
128 			continue;
129 		/* Explicitly include allowed calls. */
130 		case BPF_RET | BPF_K:
131 		case BPF_RET | BPF_A:
132 		case BPF_ALU | BPF_ADD | BPF_K:
133 		case BPF_ALU | BPF_ADD | BPF_X:
134 		case BPF_ALU | BPF_SUB | BPF_K:
135 		case BPF_ALU | BPF_SUB | BPF_X:
136 		case BPF_ALU | BPF_MUL | BPF_K:
137 		case BPF_ALU | BPF_MUL | BPF_X:
138 		case BPF_ALU | BPF_DIV | BPF_K:
139 		case BPF_ALU | BPF_DIV | BPF_X:
140 		case BPF_ALU | BPF_AND | BPF_K:
141 		case BPF_ALU | BPF_AND | BPF_X:
142 		case BPF_ALU | BPF_OR | BPF_K:
143 		case BPF_ALU | BPF_OR | BPF_X:
144 		case BPF_ALU | BPF_XOR | BPF_K:
145 		case BPF_ALU | BPF_XOR | BPF_X:
146 		case BPF_ALU | BPF_LSH | BPF_K:
147 		case BPF_ALU | BPF_LSH | BPF_X:
148 		case BPF_ALU | BPF_RSH | BPF_K:
149 		case BPF_ALU | BPF_RSH | BPF_X:
150 		case BPF_ALU | BPF_NEG:
151 		case BPF_LD | BPF_IMM:
152 		case BPF_LDX | BPF_IMM:
153 		case BPF_MISC | BPF_TAX:
154 		case BPF_MISC | BPF_TXA:
155 		case BPF_LD | BPF_MEM:
156 		case BPF_LDX | BPF_MEM:
157 		case BPF_ST:
158 		case BPF_STX:
159 		case BPF_JMP | BPF_JA:
160 		case BPF_JMP | BPF_JEQ | BPF_K:
161 		case BPF_JMP | BPF_JEQ | BPF_X:
162 		case BPF_JMP | BPF_JGE | BPF_K:
163 		case BPF_JMP | BPF_JGE | BPF_X:
164 		case BPF_JMP | BPF_JGT | BPF_K:
165 		case BPF_JMP | BPF_JGT | BPF_X:
166 		case BPF_JMP | BPF_JSET | BPF_K:
167 		case BPF_JMP | BPF_JSET | BPF_X:
168 			continue;
169 		default:
170 			return -EINVAL;
171 		}
172 	}
173 	return 0;
174 }
175 
176 /**
177  * seccomp_run_filters - evaluates all seccomp filters against @sd
178  * @sd: optional seccomp data to be passed to filters
179  * @match: stores struct seccomp_filter that resulted in the return value,
180  *         unless filter returned SECCOMP_RET_ALLOW, in which case it will
181  *         be unchanged.
182  *
183  * Returns valid seccomp BPF response codes.
184  */
185 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
186 static u32 seccomp_run_filters(const struct seccomp_data *sd,
187 			       struct seccomp_filter **match)
188 {
189 	struct seccomp_data sd_local;
190 	u32 ret = SECCOMP_RET_ALLOW;
191 	/* Make sure cross-thread synced filter points somewhere sane. */
192 	struct seccomp_filter *f =
193 			READ_ONCE(current->seccomp.filter);
194 
195 	/* Ensure unexpected behavior doesn't result in failing open. */
196 	if (unlikely(WARN_ON(f == NULL)))
197 		return SECCOMP_RET_KILL_PROCESS;
198 
199 	if (!sd) {
200 		populate_seccomp_data(&sd_local);
201 		sd = &sd_local;
202 	}
203 
204 	/*
205 	 * All filters in the list are evaluated and the lowest BPF return
206 	 * value always takes priority (ignoring the DATA).
207 	 */
208 	for (; f; f = f->prev) {
209 		u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
210 
211 		if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
212 			ret = cur_ret;
213 			*match = f;
214 		}
215 	}
216 	return ret;
217 }
218 #endif /* CONFIG_SECCOMP_FILTER */
219 
220 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
221 {
222 	assert_spin_locked(&current->sighand->siglock);
223 
224 	if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
225 		return false;
226 
227 	return true;
228 }
229 
230 static inline void seccomp_assign_mode(struct task_struct *task,
231 				       unsigned long seccomp_mode)
232 {
233 	assert_spin_locked(&task->sighand->siglock);
234 
235 	task->seccomp.mode = seccomp_mode;
236 	/*
237 	 * Make sure TIF_SECCOMP cannot be set before the mode (and
238 	 * filter) is set.
239 	 */
240 	smp_mb__before_atomic();
241 	set_tsk_thread_flag(task, TIF_SECCOMP);
242 }
243 
244 #ifdef CONFIG_SECCOMP_FILTER
245 /* Returns 1 if the parent is an ancestor of the child. */
246 static int is_ancestor(struct seccomp_filter *parent,
247 		       struct seccomp_filter *child)
248 {
249 	/* NULL is the root ancestor. */
250 	if (parent == NULL)
251 		return 1;
252 	for (; child; child = child->prev)
253 		if (child == parent)
254 			return 1;
255 	return 0;
256 }
257 
258 /**
259  * seccomp_can_sync_threads: checks if all threads can be synchronized
260  *
261  * Expects sighand and cred_guard_mutex locks to be held.
262  *
263  * Returns 0 on success, -ve on error, or the pid of a thread which was
264  * either not in the correct seccomp mode or it did not have an ancestral
265  * seccomp filter.
266  */
267 static inline pid_t seccomp_can_sync_threads(void)
268 {
269 	struct task_struct *thread, *caller;
270 
271 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
272 	assert_spin_locked(&current->sighand->siglock);
273 
274 	/* Validate all threads being eligible for synchronization. */
275 	caller = current;
276 	for_each_thread(caller, thread) {
277 		pid_t failed;
278 
279 		/* Skip current, since it is initiating the sync. */
280 		if (thread == caller)
281 			continue;
282 
283 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
284 		    (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
285 		     is_ancestor(thread->seccomp.filter,
286 				 caller->seccomp.filter)))
287 			continue;
288 
289 		/* Return the first thread that cannot be synchronized. */
290 		failed = task_pid_vnr(thread);
291 		/* If the pid cannot be resolved, then return -ESRCH */
292 		if (unlikely(WARN_ON(failed == 0)))
293 			failed = -ESRCH;
294 		return failed;
295 	}
296 
297 	return 0;
298 }
299 
300 /**
301  * seccomp_sync_threads: sets all threads to use current's filter
302  *
303  * Expects sighand and cred_guard_mutex locks to be held, and for
304  * seccomp_can_sync_threads() to have returned success already
305  * without dropping the locks.
306  *
307  */
308 static inline void seccomp_sync_threads(void)
309 {
310 	struct task_struct *thread, *caller;
311 
312 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
313 	assert_spin_locked(&current->sighand->siglock);
314 
315 	/* Synchronize all threads. */
316 	caller = current;
317 	for_each_thread(caller, thread) {
318 		/* Skip current, since it needs no changes. */
319 		if (thread == caller)
320 			continue;
321 
322 		/* Get a task reference for the new leaf node. */
323 		get_seccomp_filter(caller);
324 		/*
325 		 * Drop the task reference to the shared ancestor since
326 		 * current's path will hold a reference.  (This also
327 		 * allows a put before the assignment.)
328 		 */
329 		put_seccomp_filter(thread);
330 		smp_store_release(&thread->seccomp.filter,
331 				  caller->seccomp.filter);
332 
333 		/*
334 		 * Don't let an unprivileged task work around
335 		 * the no_new_privs restriction by creating
336 		 * a thread that sets it up, enters seccomp,
337 		 * then dies.
338 		 */
339 		if (task_no_new_privs(caller))
340 			task_set_no_new_privs(thread);
341 
342 		/*
343 		 * Opt the other thread into seccomp if needed.
344 		 * As threads are considered to be trust-realm
345 		 * equivalent (see ptrace_may_access), it is safe to
346 		 * allow one thread to transition the other.
347 		 */
348 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
349 			seccomp_assign_mode(thread, SECCOMP_MODE_FILTER);
350 	}
351 }
352 
353 /**
354  * seccomp_prepare_filter: Prepares a seccomp filter for use.
355  * @fprog: BPF program to install
356  *
357  * Returns filter on success or an ERR_PTR on failure.
358  */
359 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
360 {
361 	struct seccomp_filter *sfilter;
362 	int ret;
363 	const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
364 
365 	if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
366 		return ERR_PTR(-EINVAL);
367 
368 	BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
369 
370 	/*
371 	 * Installing a seccomp filter requires that the task has
372 	 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
373 	 * This avoids scenarios where unprivileged tasks can affect the
374 	 * behavior of privileged children.
375 	 */
376 	if (!task_no_new_privs(current) &&
377 	    security_capable_noaudit(current_cred(), current_user_ns(),
378 				     CAP_SYS_ADMIN) != 0)
379 		return ERR_PTR(-EACCES);
380 
381 	/* Allocate a new seccomp_filter */
382 	sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
383 	if (!sfilter)
384 		return ERR_PTR(-ENOMEM);
385 
386 	ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
387 					seccomp_check_filter, save_orig);
388 	if (ret < 0) {
389 		kfree(sfilter);
390 		return ERR_PTR(ret);
391 	}
392 
393 	refcount_set(&sfilter->usage, 1);
394 
395 	return sfilter;
396 }
397 
398 /**
399  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
400  * @user_filter: pointer to the user data containing a sock_fprog.
401  *
402  * Returns 0 on success and non-zero otherwise.
403  */
404 static struct seccomp_filter *
405 seccomp_prepare_user_filter(const char __user *user_filter)
406 {
407 	struct sock_fprog fprog;
408 	struct seccomp_filter *filter = ERR_PTR(-EFAULT);
409 
410 #ifdef CONFIG_COMPAT
411 	if (in_compat_syscall()) {
412 		struct compat_sock_fprog fprog32;
413 		if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
414 			goto out;
415 		fprog.len = fprog32.len;
416 		fprog.filter = compat_ptr(fprog32.filter);
417 	} else /* falls through to the if below. */
418 #endif
419 	if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
420 		goto out;
421 	filter = seccomp_prepare_filter(&fprog);
422 out:
423 	return filter;
424 }
425 
426 /**
427  * seccomp_attach_filter: validate and attach filter
428  * @flags:  flags to change filter behavior
429  * @filter: seccomp filter to add to the current process
430  *
431  * Caller must be holding current->sighand->siglock lock.
432  *
433  * Returns 0 on success, -ve on error.
434  */
435 static long seccomp_attach_filter(unsigned int flags,
436 				  struct seccomp_filter *filter)
437 {
438 	unsigned long total_insns;
439 	struct seccomp_filter *walker;
440 
441 	assert_spin_locked(&current->sighand->siglock);
442 
443 	/* Validate resulting filter length. */
444 	total_insns = filter->prog->len;
445 	for (walker = current->seccomp.filter; walker; walker = walker->prev)
446 		total_insns += walker->prog->len + 4;  /* 4 instr penalty */
447 	if (total_insns > MAX_INSNS_PER_PATH)
448 		return -ENOMEM;
449 
450 	/* If thread sync has been requested, check that it is possible. */
451 	if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
452 		int ret;
453 
454 		ret = seccomp_can_sync_threads();
455 		if (ret)
456 			return ret;
457 	}
458 
459 	/* Set log flag, if present. */
460 	if (flags & SECCOMP_FILTER_FLAG_LOG)
461 		filter->log = true;
462 
463 	/*
464 	 * If there is an existing filter, make it the prev and don't drop its
465 	 * task reference.
466 	 */
467 	filter->prev = current->seccomp.filter;
468 	current->seccomp.filter = filter;
469 
470 	/* Now that the new filter is in place, synchronize to all threads. */
471 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
472 		seccomp_sync_threads();
473 
474 	return 0;
475 }
476 
477 static void __get_seccomp_filter(struct seccomp_filter *filter)
478 {
479 	/* Reference count is bounded by the number of total processes. */
480 	refcount_inc(&filter->usage);
481 }
482 
483 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
484 void get_seccomp_filter(struct task_struct *tsk)
485 {
486 	struct seccomp_filter *orig = tsk->seccomp.filter;
487 	if (!orig)
488 		return;
489 	__get_seccomp_filter(orig);
490 }
491 
492 static inline void seccomp_filter_free(struct seccomp_filter *filter)
493 {
494 	if (filter) {
495 		bpf_prog_destroy(filter->prog);
496 		kfree(filter);
497 	}
498 }
499 
500 static void __put_seccomp_filter(struct seccomp_filter *orig)
501 {
502 	/* Clean up single-reference branches iteratively. */
503 	while (orig && refcount_dec_and_test(&orig->usage)) {
504 		struct seccomp_filter *freeme = orig;
505 		orig = orig->prev;
506 		seccomp_filter_free(freeme);
507 	}
508 }
509 
510 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
511 void put_seccomp_filter(struct task_struct *tsk)
512 {
513 	__put_seccomp_filter(tsk->seccomp.filter);
514 }
515 
516 static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason)
517 {
518 	clear_siginfo(info);
519 	info->si_signo = SIGSYS;
520 	info->si_code = SYS_SECCOMP;
521 	info->si_call_addr = (void __user *)KSTK_EIP(current);
522 	info->si_errno = reason;
523 	info->si_arch = syscall_get_arch();
524 	info->si_syscall = syscall;
525 }
526 
527 /**
528  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
529  * @syscall: syscall number to send to userland
530  * @reason: filter-supplied reason code to send to userland (via si_errno)
531  *
532  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
533  */
534 static void seccomp_send_sigsys(int syscall, int reason)
535 {
536 	struct siginfo info;
537 	seccomp_init_siginfo(&info, syscall, reason);
538 	force_sig_info(SIGSYS, &info, current);
539 }
540 #endif	/* CONFIG_SECCOMP_FILTER */
541 
542 /* For use with seccomp_actions_logged */
543 #define SECCOMP_LOG_KILL_PROCESS	(1 << 0)
544 #define SECCOMP_LOG_KILL_THREAD		(1 << 1)
545 #define SECCOMP_LOG_TRAP		(1 << 2)
546 #define SECCOMP_LOG_ERRNO		(1 << 3)
547 #define SECCOMP_LOG_TRACE		(1 << 4)
548 #define SECCOMP_LOG_LOG			(1 << 5)
549 #define SECCOMP_LOG_ALLOW		(1 << 6)
550 
551 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
552 				    SECCOMP_LOG_KILL_THREAD  |
553 				    SECCOMP_LOG_TRAP  |
554 				    SECCOMP_LOG_ERRNO |
555 				    SECCOMP_LOG_TRACE |
556 				    SECCOMP_LOG_LOG;
557 
558 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
559 			       bool requested)
560 {
561 	bool log = false;
562 
563 	switch (action) {
564 	case SECCOMP_RET_ALLOW:
565 		break;
566 	case SECCOMP_RET_TRAP:
567 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
568 		break;
569 	case SECCOMP_RET_ERRNO:
570 		log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
571 		break;
572 	case SECCOMP_RET_TRACE:
573 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
574 		break;
575 	case SECCOMP_RET_LOG:
576 		log = seccomp_actions_logged & SECCOMP_LOG_LOG;
577 		break;
578 	case SECCOMP_RET_KILL_THREAD:
579 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
580 		break;
581 	case SECCOMP_RET_KILL_PROCESS:
582 	default:
583 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
584 	}
585 
586 	/*
587 	 * Force an audit message to be emitted when the action is RET_KILL_*,
588 	 * RET_LOG, or the FILTER_FLAG_LOG bit was set and the action is
589 	 * allowed to be logged by the admin.
590 	 */
591 	if (log)
592 		return __audit_seccomp(syscall, signr, action);
593 
594 	/*
595 	 * Let the audit subsystem decide if the action should be audited based
596 	 * on whether the current task itself is being audited.
597 	 */
598 	return audit_seccomp(syscall, signr, action);
599 }
600 
601 /*
602  * Secure computing mode 1 allows only read/write/exit/sigreturn.
603  * To be fully secure this must be combined with rlimit
604  * to limit the stack allocations too.
605  */
606 static const int mode1_syscalls[] = {
607 	__NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
608 	0, /* null terminated */
609 };
610 
611 static void __secure_computing_strict(int this_syscall)
612 {
613 	const int *syscall_whitelist = mode1_syscalls;
614 #ifdef CONFIG_COMPAT
615 	if (in_compat_syscall())
616 		syscall_whitelist = get_compat_mode1_syscalls();
617 #endif
618 	do {
619 		if (*syscall_whitelist == this_syscall)
620 			return;
621 	} while (*++syscall_whitelist);
622 
623 #ifdef SECCOMP_DEBUG
624 	dump_stack();
625 #endif
626 	seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
627 	do_exit(SIGKILL);
628 }
629 
630 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
631 void secure_computing_strict(int this_syscall)
632 {
633 	int mode = current->seccomp.mode;
634 
635 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
636 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
637 		return;
638 
639 	if (mode == SECCOMP_MODE_DISABLED)
640 		return;
641 	else if (mode == SECCOMP_MODE_STRICT)
642 		__secure_computing_strict(this_syscall);
643 	else
644 		BUG();
645 }
646 #else
647 
648 #ifdef CONFIG_SECCOMP_FILTER
649 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
650 			    const bool recheck_after_trace)
651 {
652 	u32 filter_ret, action;
653 	struct seccomp_filter *match = NULL;
654 	int data;
655 
656 	/*
657 	 * Make sure that any changes to mode from another thread have
658 	 * been seen after TIF_SECCOMP was seen.
659 	 */
660 	rmb();
661 
662 	filter_ret = seccomp_run_filters(sd, &match);
663 	data = filter_ret & SECCOMP_RET_DATA;
664 	action = filter_ret & SECCOMP_RET_ACTION_FULL;
665 
666 	switch (action) {
667 	case SECCOMP_RET_ERRNO:
668 		/* Set low-order bits as an errno, capped at MAX_ERRNO. */
669 		if (data > MAX_ERRNO)
670 			data = MAX_ERRNO;
671 		syscall_set_return_value(current, task_pt_regs(current),
672 					 -data, 0);
673 		goto skip;
674 
675 	case SECCOMP_RET_TRAP:
676 		/* Show the handler the original registers. */
677 		syscall_rollback(current, task_pt_regs(current));
678 		/* Let the filter pass back 16 bits of data. */
679 		seccomp_send_sigsys(this_syscall, data);
680 		goto skip;
681 
682 	case SECCOMP_RET_TRACE:
683 		/* We've been put in this state by the ptracer already. */
684 		if (recheck_after_trace)
685 			return 0;
686 
687 		/* ENOSYS these calls if there is no tracer attached. */
688 		if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
689 			syscall_set_return_value(current,
690 						 task_pt_regs(current),
691 						 -ENOSYS, 0);
692 			goto skip;
693 		}
694 
695 		/* Allow the BPF to provide the event message */
696 		ptrace_event(PTRACE_EVENT_SECCOMP, data);
697 		/*
698 		 * The delivery of a fatal signal during event
699 		 * notification may silently skip tracer notification,
700 		 * which could leave us with a potentially unmodified
701 		 * syscall that the tracer would have liked to have
702 		 * changed. Since the process is about to die, we just
703 		 * force the syscall to be skipped and let the signal
704 		 * kill the process and correctly handle any tracer exit
705 		 * notifications.
706 		 */
707 		if (fatal_signal_pending(current))
708 			goto skip;
709 		/* Check if the tracer forced the syscall to be skipped. */
710 		this_syscall = syscall_get_nr(current, task_pt_regs(current));
711 		if (this_syscall < 0)
712 			goto skip;
713 
714 		/*
715 		 * Recheck the syscall, since it may have changed. This
716 		 * intentionally uses a NULL struct seccomp_data to force
717 		 * a reload of all registers. This does not goto skip since
718 		 * a skip would have already been reported.
719 		 */
720 		if (__seccomp_filter(this_syscall, NULL, true))
721 			return -1;
722 
723 		return 0;
724 
725 	case SECCOMP_RET_LOG:
726 		seccomp_log(this_syscall, 0, action, true);
727 		return 0;
728 
729 	case SECCOMP_RET_ALLOW:
730 		/*
731 		 * Note that the "match" filter will always be NULL for
732 		 * this action since SECCOMP_RET_ALLOW is the starting
733 		 * state in seccomp_run_filters().
734 		 */
735 		return 0;
736 
737 	case SECCOMP_RET_KILL_THREAD:
738 	case SECCOMP_RET_KILL_PROCESS:
739 	default:
740 		seccomp_log(this_syscall, SIGSYS, action, true);
741 		/* Dump core only if this is the last remaining thread. */
742 		if (action == SECCOMP_RET_KILL_PROCESS ||
743 		    get_nr_threads(current) == 1) {
744 			siginfo_t info;
745 
746 			/* Show the original registers in the dump. */
747 			syscall_rollback(current, task_pt_regs(current));
748 			/* Trigger a manual coredump since do_exit skips it. */
749 			seccomp_init_siginfo(&info, this_syscall, data);
750 			do_coredump(&info);
751 		}
752 		if (action == SECCOMP_RET_KILL_PROCESS)
753 			do_group_exit(SIGSYS);
754 		else
755 			do_exit(SIGSYS);
756 	}
757 
758 	unreachable();
759 
760 skip:
761 	seccomp_log(this_syscall, 0, action, match ? match->log : false);
762 	return -1;
763 }
764 #else
765 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
766 			    const bool recheck_after_trace)
767 {
768 	BUG();
769 }
770 #endif
771 
772 int __secure_computing(const struct seccomp_data *sd)
773 {
774 	int mode = current->seccomp.mode;
775 	int this_syscall;
776 
777 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
778 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
779 		return 0;
780 
781 	this_syscall = sd ? sd->nr :
782 		syscall_get_nr(current, task_pt_regs(current));
783 
784 	switch (mode) {
785 	case SECCOMP_MODE_STRICT:
786 		__secure_computing_strict(this_syscall);  /* may call do_exit */
787 		return 0;
788 	case SECCOMP_MODE_FILTER:
789 		return __seccomp_filter(this_syscall, sd, false);
790 	default:
791 		BUG();
792 	}
793 }
794 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
795 
796 long prctl_get_seccomp(void)
797 {
798 	return current->seccomp.mode;
799 }
800 
801 /**
802  * seccomp_set_mode_strict: internal function for setting strict seccomp
803  *
804  * Once current->seccomp.mode is non-zero, it may not be changed.
805  *
806  * Returns 0 on success or -EINVAL on failure.
807  */
808 static long seccomp_set_mode_strict(void)
809 {
810 	const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
811 	long ret = -EINVAL;
812 
813 	spin_lock_irq(&current->sighand->siglock);
814 
815 	if (!seccomp_may_assign_mode(seccomp_mode))
816 		goto out;
817 
818 #ifdef TIF_NOTSC
819 	disable_TSC();
820 #endif
821 	seccomp_assign_mode(current, seccomp_mode);
822 	ret = 0;
823 
824 out:
825 	spin_unlock_irq(&current->sighand->siglock);
826 
827 	return ret;
828 }
829 
830 #ifdef CONFIG_SECCOMP_FILTER
831 /**
832  * seccomp_set_mode_filter: internal function for setting seccomp filter
833  * @flags:  flags to change filter behavior
834  * @filter: struct sock_fprog containing filter
835  *
836  * This function may be called repeatedly to install additional filters.
837  * Every filter successfully installed will be evaluated (in reverse order)
838  * for each system call the task makes.
839  *
840  * Once current->seccomp.mode is non-zero, it may not be changed.
841  *
842  * Returns 0 on success or -EINVAL on failure.
843  */
844 static long seccomp_set_mode_filter(unsigned int flags,
845 				    const char __user *filter)
846 {
847 	const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
848 	struct seccomp_filter *prepared = NULL;
849 	long ret = -EINVAL;
850 
851 	/* Validate flags. */
852 	if (flags & ~SECCOMP_FILTER_FLAG_MASK)
853 		return -EINVAL;
854 
855 	/* Prepare the new filter before holding any locks. */
856 	prepared = seccomp_prepare_user_filter(filter);
857 	if (IS_ERR(prepared))
858 		return PTR_ERR(prepared);
859 
860 	/*
861 	 * Make sure we cannot change seccomp or nnp state via TSYNC
862 	 * while another thread is in the middle of calling exec.
863 	 */
864 	if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
865 	    mutex_lock_killable(&current->signal->cred_guard_mutex))
866 		goto out_free;
867 
868 	spin_lock_irq(&current->sighand->siglock);
869 
870 	if (!seccomp_may_assign_mode(seccomp_mode))
871 		goto out;
872 
873 	ret = seccomp_attach_filter(flags, prepared);
874 	if (ret)
875 		goto out;
876 	/* Do not free the successfully attached filter. */
877 	prepared = NULL;
878 
879 	seccomp_assign_mode(current, seccomp_mode);
880 out:
881 	spin_unlock_irq(&current->sighand->siglock);
882 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
883 		mutex_unlock(&current->signal->cred_guard_mutex);
884 out_free:
885 	seccomp_filter_free(prepared);
886 	return ret;
887 }
888 #else
889 static inline long seccomp_set_mode_filter(unsigned int flags,
890 					   const char __user *filter)
891 {
892 	return -EINVAL;
893 }
894 #endif
895 
896 static long seccomp_get_action_avail(const char __user *uaction)
897 {
898 	u32 action;
899 
900 	if (copy_from_user(&action, uaction, sizeof(action)))
901 		return -EFAULT;
902 
903 	switch (action) {
904 	case SECCOMP_RET_KILL_PROCESS:
905 	case SECCOMP_RET_KILL_THREAD:
906 	case SECCOMP_RET_TRAP:
907 	case SECCOMP_RET_ERRNO:
908 	case SECCOMP_RET_TRACE:
909 	case SECCOMP_RET_LOG:
910 	case SECCOMP_RET_ALLOW:
911 		break;
912 	default:
913 		return -EOPNOTSUPP;
914 	}
915 
916 	return 0;
917 }
918 
919 /* Common entry point for both prctl and syscall. */
920 static long do_seccomp(unsigned int op, unsigned int flags,
921 		       const char __user *uargs)
922 {
923 	switch (op) {
924 	case SECCOMP_SET_MODE_STRICT:
925 		if (flags != 0 || uargs != NULL)
926 			return -EINVAL;
927 		return seccomp_set_mode_strict();
928 	case SECCOMP_SET_MODE_FILTER:
929 		return seccomp_set_mode_filter(flags, uargs);
930 	case SECCOMP_GET_ACTION_AVAIL:
931 		if (flags != 0)
932 			return -EINVAL;
933 
934 		return seccomp_get_action_avail(uargs);
935 	default:
936 		return -EINVAL;
937 	}
938 }
939 
940 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
941 			 const char __user *, uargs)
942 {
943 	return do_seccomp(op, flags, uargs);
944 }
945 
946 /**
947  * prctl_set_seccomp: configures current->seccomp.mode
948  * @seccomp_mode: requested mode to use
949  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
950  *
951  * Returns 0 on success or -EINVAL on failure.
952  */
953 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
954 {
955 	unsigned int op;
956 	char __user *uargs;
957 
958 	switch (seccomp_mode) {
959 	case SECCOMP_MODE_STRICT:
960 		op = SECCOMP_SET_MODE_STRICT;
961 		/*
962 		 * Setting strict mode through prctl always ignored filter,
963 		 * so make sure it is always NULL here to pass the internal
964 		 * check in do_seccomp().
965 		 */
966 		uargs = NULL;
967 		break;
968 	case SECCOMP_MODE_FILTER:
969 		op = SECCOMP_SET_MODE_FILTER;
970 		uargs = filter;
971 		break;
972 	default:
973 		return -EINVAL;
974 	}
975 
976 	/* prctl interface doesn't have flags, so they are always zero. */
977 	return do_seccomp(op, 0, uargs);
978 }
979 
980 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
981 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
982 					     unsigned long filter_off)
983 {
984 	struct seccomp_filter *orig, *filter;
985 	unsigned long count;
986 
987 	/*
988 	 * Note: this is only correct because the caller should be the (ptrace)
989 	 * tracer of the task, otherwise lock_task_sighand is needed.
990 	 */
991 	spin_lock_irq(&task->sighand->siglock);
992 
993 	if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
994 		spin_unlock_irq(&task->sighand->siglock);
995 		return ERR_PTR(-EINVAL);
996 	}
997 
998 	orig = task->seccomp.filter;
999 	__get_seccomp_filter(orig);
1000 	spin_unlock_irq(&task->sighand->siglock);
1001 
1002 	count = 0;
1003 	for (filter = orig; filter; filter = filter->prev)
1004 		count++;
1005 
1006 	if (filter_off >= count) {
1007 		filter = ERR_PTR(-ENOENT);
1008 		goto out;
1009 	}
1010 
1011 	count -= filter_off;
1012 	for (filter = orig; filter && count > 1; filter = filter->prev)
1013 		count--;
1014 
1015 	if (WARN_ON(count != 1 || !filter)) {
1016 		filter = ERR_PTR(-ENOENT);
1017 		goto out;
1018 	}
1019 
1020 	__get_seccomp_filter(filter);
1021 
1022 out:
1023 	__put_seccomp_filter(orig);
1024 	return filter;
1025 }
1026 
1027 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1028 			void __user *data)
1029 {
1030 	struct seccomp_filter *filter;
1031 	struct sock_fprog_kern *fprog;
1032 	long ret;
1033 
1034 	if (!capable(CAP_SYS_ADMIN) ||
1035 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1036 		return -EACCES;
1037 	}
1038 
1039 	filter = get_nth_filter(task, filter_off);
1040 	if (IS_ERR(filter))
1041 		return PTR_ERR(filter);
1042 
1043 	fprog = filter->prog->orig_prog;
1044 	if (!fprog) {
1045 		/* This must be a new non-cBPF filter, since we save
1046 		 * every cBPF filter's orig_prog above when
1047 		 * CONFIG_CHECKPOINT_RESTORE is enabled.
1048 		 */
1049 		ret = -EMEDIUMTYPE;
1050 		goto out;
1051 	}
1052 
1053 	ret = fprog->len;
1054 	if (!data)
1055 		goto out;
1056 
1057 	if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1058 		ret = -EFAULT;
1059 
1060 out:
1061 	__put_seccomp_filter(filter);
1062 	return ret;
1063 }
1064 
1065 long seccomp_get_metadata(struct task_struct *task,
1066 			  unsigned long size, void __user *data)
1067 {
1068 	long ret;
1069 	struct seccomp_filter *filter;
1070 	struct seccomp_metadata kmd = {};
1071 
1072 	if (!capable(CAP_SYS_ADMIN) ||
1073 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1074 		return -EACCES;
1075 	}
1076 
1077 	size = min_t(unsigned long, size, sizeof(kmd));
1078 
1079 	if (copy_from_user(&kmd, data, size))
1080 		return -EFAULT;
1081 
1082 	filter = get_nth_filter(task, kmd.filter_off);
1083 	if (IS_ERR(filter))
1084 		return PTR_ERR(filter);
1085 
1086 	memset(&kmd, 0, sizeof(kmd));
1087 	if (filter->log)
1088 		kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1089 
1090 	ret = size;
1091 	if (copy_to_user(data, &kmd, size))
1092 		ret = -EFAULT;
1093 
1094 	__put_seccomp_filter(filter);
1095 	return ret;
1096 }
1097 #endif
1098 
1099 #ifdef CONFIG_SYSCTL
1100 
1101 /* Human readable action names for friendly sysctl interaction */
1102 #define SECCOMP_RET_KILL_PROCESS_NAME	"kill_process"
1103 #define SECCOMP_RET_KILL_THREAD_NAME	"kill_thread"
1104 #define SECCOMP_RET_TRAP_NAME		"trap"
1105 #define SECCOMP_RET_ERRNO_NAME		"errno"
1106 #define SECCOMP_RET_TRACE_NAME		"trace"
1107 #define SECCOMP_RET_LOG_NAME		"log"
1108 #define SECCOMP_RET_ALLOW_NAME		"allow"
1109 
1110 static const char seccomp_actions_avail[] =
1111 				SECCOMP_RET_KILL_PROCESS_NAME	" "
1112 				SECCOMP_RET_KILL_THREAD_NAME	" "
1113 				SECCOMP_RET_TRAP_NAME		" "
1114 				SECCOMP_RET_ERRNO_NAME		" "
1115 				SECCOMP_RET_TRACE_NAME		" "
1116 				SECCOMP_RET_LOG_NAME		" "
1117 				SECCOMP_RET_ALLOW_NAME;
1118 
1119 struct seccomp_log_name {
1120 	u32		log;
1121 	const char	*name;
1122 };
1123 
1124 static const struct seccomp_log_name seccomp_log_names[] = {
1125 	{ SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1126 	{ SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1127 	{ SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1128 	{ SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1129 	{ SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1130 	{ SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1131 	{ SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1132 	{ }
1133 };
1134 
1135 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1136 					      u32 actions_logged)
1137 {
1138 	const struct seccomp_log_name *cur;
1139 	bool append_space = false;
1140 
1141 	for (cur = seccomp_log_names; cur->name && size; cur++) {
1142 		ssize_t ret;
1143 
1144 		if (!(actions_logged & cur->log))
1145 			continue;
1146 
1147 		if (append_space) {
1148 			ret = strscpy(names, " ", size);
1149 			if (ret < 0)
1150 				return false;
1151 
1152 			names += ret;
1153 			size -= ret;
1154 		} else
1155 			append_space = true;
1156 
1157 		ret = strscpy(names, cur->name, size);
1158 		if (ret < 0)
1159 			return false;
1160 
1161 		names += ret;
1162 		size -= ret;
1163 	}
1164 
1165 	return true;
1166 }
1167 
1168 static bool seccomp_action_logged_from_name(u32 *action_logged,
1169 					    const char *name)
1170 {
1171 	const struct seccomp_log_name *cur;
1172 
1173 	for (cur = seccomp_log_names; cur->name; cur++) {
1174 		if (!strcmp(cur->name, name)) {
1175 			*action_logged = cur->log;
1176 			return true;
1177 		}
1178 	}
1179 
1180 	return false;
1181 }
1182 
1183 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1184 {
1185 	char *name;
1186 
1187 	*actions_logged = 0;
1188 	while ((name = strsep(&names, " ")) && *name) {
1189 		u32 action_logged = 0;
1190 
1191 		if (!seccomp_action_logged_from_name(&action_logged, name))
1192 			return false;
1193 
1194 		*actions_logged |= action_logged;
1195 	}
1196 
1197 	return true;
1198 }
1199 
1200 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1201 					  void __user *buffer, size_t *lenp,
1202 					  loff_t *ppos)
1203 {
1204 	char names[sizeof(seccomp_actions_avail)];
1205 	struct ctl_table table;
1206 	int ret;
1207 
1208 	if (write && !capable(CAP_SYS_ADMIN))
1209 		return -EPERM;
1210 
1211 	memset(names, 0, sizeof(names));
1212 
1213 	if (!write) {
1214 		if (!seccomp_names_from_actions_logged(names, sizeof(names),
1215 						       seccomp_actions_logged))
1216 			return -EINVAL;
1217 	}
1218 
1219 	table = *ro_table;
1220 	table.data = names;
1221 	table.maxlen = sizeof(names);
1222 	ret = proc_dostring(&table, write, buffer, lenp, ppos);
1223 	if (ret)
1224 		return ret;
1225 
1226 	if (write) {
1227 		u32 actions_logged;
1228 
1229 		if (!seccomp_actions_logged_from_names(&actions_logged,
1230 						       table.data))
1231 			return -EINVAL;
1232 
1233 		if (actions_logged & SECCOMP_LOG_ALLOW)
1234 			return -EINVAL;
1235 
1236 		seccomp_actions_logged = actions_logged;
1237 	}
1238 
1239 	return 0;
1240 }
1241 
1242 static struct ctl_path seccomp_sysctl_path[] = {
1243 	{ .procname = "kernel", },
1244 	{ .procname = "seccomp", },
1245 	{ }
1246 };
1247 
1248 static struct ctl_table seccomp_sysctl_table[] = {
1249 	{
1250 		.procname	= "actions_avail",
1251 		.data		= (void *) &seccomp_actions_avail,
1252 		.maxlen		= sizeof(seccomp_actions_avail),
1253 		.mode		= 0444,
1254 		.proc_handler	= proc_dostring,
1255 	},
1256 	{
1257 		.procname	= "actions_logged",
1258 		.mode		= 0644,
1259 		.proc_handler	= seccomp_actions_logged_handler,
1260 	},
1261 	{ }
1262 };
1263 
1264 static int __init seccomp_sysctl_init(void)
1265 {
1266 	struct ctl_table_header *hdr;
1267 
1268 	hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1269 	if (!hdr)
1270 		pr_warn("seccomp: sysctl registration failed\n");
1271 	else
1272 		kmemleak_not_leak(hdr);
1273 
1274 	return 0;
1275 }
1276 
1277 device_initcall(seccomp_sysctl_init)
1278 
1279 #endif /* CONFIG_SYSCTL */
1280