xref: /openbmc/linux/kernel/seccomp.c (revision 3557b3fd)
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/nospec.h>
23 #include <linux/prctl.h>
24 #include <linux/sched.h>
25 #include <linux/sched/task_stack.h>
26 #include <linux/seccomp.h>
27 #include <linux/slab.h>
28 #include <linux/syscalls.h>
29 #include <linux/sysctl.h>
30 
31 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
32 #include <asm/syscall.h>
33 #endif
34 
35 #ifdef CONFIG_SECCOMP_FILTER
36 #include <linux/file.h>
37 #include <linux/filter.h>
38 #include <linux/pid.h>
39 #include <linux/ptrace.h>
40 #include <linux/security.h>
41 #include <linux/tracehook.h>
42 #include <linux/uaccess.h>
43 #include <linux/anon_inodes.h>
44 
45 enum notify_state {
46 	SECCOMP_NOTIFY_INIT,
47 	SECCOMP_NOTIFY_SENT,
48 	SECCOMP_NOTIFY_REPLIED,
49 };
50 
51 struct seccomp_knotif {
52 	/* The struct pid of the task whose filter triggered the notification */
53 	struct task_struct *task;
54 
55 	/* The "cookie" for this request; this is unique for this filter. */
56 	u64 id;
57 
58 	/*
59 	 * The seccomp data. This pointer is valid the entire time this
60 	 * notification is active, since it comes from __seccomp_filter which
61 	 * eclipses the entire lifecycle here.
62 	 */
63 	const struct seccomp_data *data;
64 
65 	/*
66 	 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
67 	 * struct seccomp_knotif is created and starts out in INIT. Once the
68 	 * handler reads the notification off of an FD, it transitions to SENT.
69 	 * If a signal is received the state transitions back to INIT and
70 	 * another message is sent. When the userspace handler replies, state
71 	 * transitions to REPLIED.
72 	 */
73 	enum notify_state state;
74 
75 	/* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
76 	int error;
77 	long val;
78 
79 	/* Signals when this has entered SECCOMP_NOTIFY_REPLIED */
80 	struct completion ready;
81 
82 	struct list_head list;
83 };
84 
85 /**
86  * struct notification - container for seccomp userspace notifications. Since
87  * most seccomp filters will not have notification listeners attached and this
88  * structure is fairly large, we store the notification-specific stuff in a
89  * separate structure.
90  *
91  * @request: A semaphore that users of this notification can wait on for
92  *           changes. Actual reads and writes are still controlled with
93  *           filter->notify_lock.
94  * @next_id: The id of the next request.
95  * @notifications: A list of struct seccomp_knotif elements.
96  * @wqh: A wait queue for poll.
97  */
98 struct notification {
99 	struct semaphore request;
100 	u64 next_id;
101 	struct list_head notifications;
102 	wait_queue_head_t wqh;
103 };
104 
105 /**
106  * struct seccomp_filter - container for seccomp BPF programs
107  *
108  * @usage: reference count to manage the object lifetime.
109  *         get/put helpers should be used when accessing an instance
110  *         outside of a lifetime-guarded section.  In general, this
111  *         is only needed for handling filters shared across tasks.
112  * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
113  * @prev: points to a previously installed, or inherited, filter
114  * @prog: the BPF program to evaluate
115  * @notif: the struct that holds all notification related information
116  * @notify_lock: A lock for all notification-related accesses.
117  *
118  * seccomp_filter objects are organized in a tree linked via the @prev
119  * pointer.  For any task, it appears to be a singly-linked list starting
120  * with current->seccomp.filter, the most recently attached or inherited filter.
121  * However, multiple filters may share a @prev node, by way of fork(), which
122  * results in a unidirectional tree existing in memory.  This is similar to
123  * how namespaces work.
124  *
125  * seccomp_filter objects should never be modified after being attached
126  * to a task_struct (other than @usage).
127  */
128 struct seccomp_filter {
129 	refcount_t usage;
130 	bool log;
131 	struct seccomp_filter *prev;
132 	struct bpf_prog *prog;
133 	struct notification *notif;
134 	struct mutex notify_lock;
135 };
136 
137 /* Limit any path through the tree to 256KB worth of instructions. */
138 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
139 
140 /*
141  * Endianness is explicitly ignored and left for BPF program authors to manage
142  * as per the specific architecture.
143  */
144 static void populate_seccomp_data(struct seccomp_data *sd)
145 {
146 	struct task_struct *task = current;
147 	struct pt_regs *regs = task_pt_regs(task);
148 	unsigned long args[6];
149 
150 	sd->nr = syscall_get_nr(task, regs);
151 	sd->arch = syscall_get_arch();
152 	syscall_get_arguments(task, regs, args);
153 	sd->args[0] = args[0];
154 	sd->args[1] = args[1];
155 	sd->args[2] = args[2];
156 	sd->args[3] = args[3];
157 	sd->args[4] = args[4];
158 	sd->args[5] = args[5];
159 	sd->instruction_pointer = KSTK_EIP(task);
160 }
161 
162 /**
163  *	seccomp_check_filter - verify seccomp filter code
164  *	@filter: filter to verify
165  *	@flen: length of filter
166  *
167  * Takes a previously checked filter (by bpf_check_classic) and
168  * redirects all filter code that loads struct sk_buff data
169  * and related data through seccomp_bpf_load.  It also
170  * enforces length and alignment checking of those loads.
171  *
172  * Returns 0 if the rule set is legal or -EINVAL if not.
173  */
174 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
175 {
176 	int pc;
177 	for (pc = 0; pc < flen; pc++) {
178 		struct sock_filter *ftest = &filter[pc];
179 		u16 code = ftest->code;
180 		u32 k = ftest->k;
181 
182 		switch (code) {
183 		case BPF_LD | BPF_W | BPF_ABS:
184 			ftest->code = BPF_LDX | BPF_W | BPF_ABS;
185 			/* 32-bit aligned and not out of bounds. */
186 			if (k >= sizeof(struct seccomp_data) || k & 3)
187 				return -EINVAL;
188 			continue;
189 		case BPF_LD | BPF_W | BPF_LEN:
190 			ftest->code = BPF_LD | BPF_IMM;
191 			ftest->k = sizeof(struct seccomp_data);
192 			continue;
193 		case BPF_LDX | BPF_W | BPF_LEN:
194 			ftest->code = BPF_LDX | BPF_IMM;
195 			ftest->k = sizeof(struct seccomp_data);
196 			continue;
197 		/* Explicitly include allowed calls. */
198 		case BPF_RET | BPF_K:
199 		case BPF_RET | BPF_A:
200 		case BPF_ALU | BPF_ADD | BPF_K:
201 		case BPF_ALU | BPF_ADD | BPF_X:
202 		case BPF_ALU | BPF_SUB | BPF_K:
203 		case BPF_ALU | BPF_SUB | BPF_X:
204 		case BPF_ALU | BPF_MUL | BPF_K:
205 		case BPF_ALU | BPF_MUL | BPF_X:
206 		case BPF_ALU | BPF_DIV | BPF_K:
207 		case BPF_ALU | BPF_DIV | BPF_X:
208 		case BPF_ALU | BPF_AND | BPF_K:
209 		case BPF_ALU | BPF_AND | BPF_X:
210 		case BPF_ALU | BPF_OR | BPF_K:
211 		case BPF_ALU | BPF_OR | BPF_X:
212 		case BPF_ALU | BPF_XOR | BPF_K:
213 		case BPF_ALU | BPF_XOR | BPF_X:
214 		case BPF_ALU | BPF_LSH | BPF_K:
215 		case BPF_ALU | BPF_LSH | BPF_X:
216 		case BPF_ALU | BPF_RSH | BPF_K:
217 		case BPF_ALU | BPF_RSH | BPF_X:
218 		case BPF_ALU | BPF_NEG:
219 		case BPF_LD | BPF_IMM:
220 		case BPF_LDX | BPF_IMM:
221 		case BPF_MISC | BPF_TAX:
222 		case BPF_MISC | BPF_TXA:
223 		case BPF_LD | BPF_MEM:
224 		case BPF_LDX | BPF_MEM:
225 		case BPF_ST:
226 		case BPF_STX:
227 		case BPF_JMP | BPF_JA:
228 		case BPF_JMP | BPF_JEQ | BPF_K:
229 		case BPF_JMP | BPF_JEQ | BPF_X:
230 		case BPF_JMP | BPF_JGE | BPF_K:
231 		case BPF_JMP | BPF_JGE | BPF_X:
232 		case BPF_JMP | BPF_JGT | BPF_K:
233 		case BPF_JMP | BPF_JGT | BPF_X:
234 		case BPF_JMP | BPF_JSET | BPF_K:
235 		case BPF_JMP | BPF_JSET | BPF_X:
236 			continue;
237 		default:
238 			return -EINVAL;
239 		}
240 	}
241 	return 0;
242 }
243 
244 /**
245  * seccomp_run_filters - evaluates all seccomp filters against @sd
246  * @sd: optional seccomp data to be passed to filters
247  * @match: stores struct seccomp_filter that resulted in the return value,
248  *         unless filter returned SECCOMP_RET_ALLOW, in which case it will
249  *         be unchanged.
250  *
251  * Returns valid seccomp BPF response codes.
252  */
253 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
254 static u32 seccomp_run_filters(const struct seccomp_data *sd,
255 			       struct seccomp_filter **match)
256 {
257 	u32 ret = SECCOMP_RET_ALLOW;
258 	/* Make sure cross-thread synced filter points somewhere sane. */
259 	struct seccomp_filter *f =
260 			READ_ONCE(current->seccomp.filter);
261 
262 	/* Ensure unexpected behavior doesn't result in failing open. */
263 	if (WARN_ON(f == NULL))
264 		return SECCOMP_RET_KILL_PROCESS;
265 
266 	/*
267 	 * All filters in the list are evaluated and the lowest BPF return
268 	 * value always takes priority (ignoring the DATA).
269 	 */
270 	preempt_disable();
271 	for (; f; f = f->prev) {
272 		u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
273 
274 		if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
275 			ret = cur_ret;
276 			*match = f;
277 		}
278 	}
279 	preempt_enable();
280 	return ret;
281 }
282 #endif /* CONFIG_SECCOMP_FILTER */
283 
284 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
285 {
286 	assert_spin_locked(&current->sighand->siglock);
287 
288 	if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
289 		return false;
290 
291 	return true;
292 }
293 
294 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
295 
296 static inline void seccomp_assign_mode(struct task_struct *task,
297 				       unsigned long seccomp_mode,
298 				       unsigned long flags)
299 {
300 	assert_spin_locked(&task->sighand->siglock);
301 
302 	task->seccomp.mode = seccomp_mode;
303 	/*
304 	 * Make sure TIF_SECCOMP cannot be set before the mode (and
305 	 * filter) is set.
306 	 */
307 	smp_mb__before_atomic();
308 	/* Assume default seccomp processes want spec flaw mitigation. */
309 	if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
310 		arch_seccomp_spec_mitigate(task);
311 	set_tsk_thread_flag(task, TIF_SECCOMP);
312 }
313 
314 #ifdef CONFIG_SECCOMP_FILTER
315 /* Returns 1 if the parent is an ancestor of the child. */
316 static int is_ancestor(struct seccomp_filter *parent,
317 		       struct seccomp_filter *child)
318 {
319 	/* NULL is the root ancestor. */
320 	if (parent == NULL)
321 		return 1;
322 	for (; child; child = child->prev)
323 		if (child == parent)
324 			return 1;
325 	return 0;
326 }
327 
328 /**
329  * seccomp_can_sync_threads: checks if all threads can be synchronized
330  *
331  * Expects sighand and cred_guard_mutex locks to be held.
332  *
333  * Returns 0 on success, -ve on error, or the pid of a thread which was
334  * either not in the correct seccomp mode or it did not have an ancestral
335  * seccomp filter.
336  */
337 static inline pid_t seccomp_can_sync_threads(void)
338 {
339 	struct task_struct *thread, *caller;
340 
341 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
342 	assert_spin_locked(&current->sighand->siglock);
343 
344 	/* Validate all threads being eligible for synchronization. */
345 	caller = current;
346 	for_each_thread(caller, thread) {
347 		pid_t failed;
348 
349 		/* Skip current, since it is initiating the sync. */
350 		if (thread == caller)
351 			continue;
352 
353 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
354 		    (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
355 		     is_ancestor(thread->seccomp.filter,
356 				 caller->seccomp.filter)))
357 			continue;
358 
359 		/* Return the first thread that cannot be synchronized. */
360 		failed = task_pid_vnr(thread);
361 		/* If the pid cannot be resolved, then return -ESRCH */
362 		if (WARN_ON(failed == 0))
363 			failed = -ESRCH;
364 		return failed;
365 	}
366 
367 	return 0;
368 }
369 
370 /**
371  * seccomp_sync_threads: sets all threads to use current's filter
372  *
373  * Expects sighand and cred_guard_mutex locks to be held, and for
374  * seccomp_can_sync_threads() to have returned success already
375  * without dropping the locks.
376  *
377  */
378 static inline void seccomp_sync_threads(unsigned long flags)
379 {
380 	struct task_struct *thread, *caller;
381 
382 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
383 	assert_spin_locked(&current->sighand->siglock);
384 
385 	/* Synchronize all threads. */
386 	caller = current;
387 	for_each_thread(caller, thread) {
388 		/* Skip current, since it needs no changes. */
389 		if (thread == caller)
390 			continue;
391 
392 		/* Get a task reference for the new leaf node. */
393 		get_seccomp_filter(caller);
394 		/*
395 		 * Drop the task reference to the shared ancestor since
396 		 * current's path will hold a reference.  (This also
397 		 * allows a put before the assignment.)
398 		 */
399 		put_seccomp_filter(thread);
400 		smp_store_release(&thread->seccomp.filter,
401 				  caller->seccomp.filter);
402 
403 		/*
404 		 * Don't let an unprivileged task work around
405 		 * the no_new_privs restriction by creating
406 		 * a thread that sets it up, enters seccomp,
407 		 * then dies.
408 		 */
409 		if (task_no_new_privs(caller))
410 			task_set_no_new_privs(thread);
411 
412 		/*
413 		 * Opt the other thread into seccomp if needed.
414 		 * As threads are considered to be trust-realm
415 		 * equivalent (see ptrace_may_access), it is safe to
416 		 * allow one thread to transition the other.
417 		 */
418 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
419 			seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
420 					    flags);
421 	}
422 }
423 
424 /**
425  * seccomp_prepare_filter: Prepares a seccomp filter for use.
426  * @fprog: BPF program to install
427  *
428  * Returns filter on success or an ERR_PTR on failure.
429  */
430 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
431 {
432 	struct seccomp_filter *sfilter;
433 	int ret;
434 	const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
435 
436 	if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
437 		return ERR_PTR(-EINVAL);
438 
439 	BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
440 
441 	/*
442 	 * Installing a seccomp filter requires that the task has
443 	 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
444 	 * This avoids scenarios where unprivileged tasks can affect the
445 	 * behavior of privileged children.
446 	 */
447 	if (!task_no_new_privs(current) &&
448 	    security_capable(current_cred(), current_user_ns(),
449 				     CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) != 0)
450 		return ERR_PTR(-EACCES);
451 
452 	/* Allocate a new seccomp_filter */
453 	sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
454 	if (!sfilter)
455 		return ERR_PTR(-ENOMEM);
456 
457 	mutex_init(&sfilter->notify_lock);
458 	ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
459 					seccomp_check_filter, save_orig);
460 	if (ret < 0) {
461 		kfree(sfilter);
462 		return ERR_PTR(ret);
463 	}
464 
465 	refcount_set(&sfilter->usage, 1);
466 
467 	return sfilter;
468 }
469 
470 /**
471  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
472  * @user_filter: pointer to the user data containing a sock_fprog.
473  *
474  * Returns 0 on success and non-zero otherwise.
475  */
476 static struct seccomp_filter *
477 seccomp_prepare_user_filter(const char __user *user_filter)
478 {
479 	struct sock_fprog fprog;
480 	struct seccomp_filter *filter = ERR_PTR(-EFAULT);
481 
482 #ifdef CONFIG_COMPAT
483 	if (in_compat_syscall()) {
484 		struct compat_sock_fprog fprog32;
485 		if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
486 			goto out;
487 		fprog.len = fprog32.len;
488 		fprog.filter = compat_ptr(fprog32.filter);
489 	} else /* falls through to the if below. */
490 #endif
491 	if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
492 		goto out;
493 	filter = seccomp_prepare_filter(&fprog);
494 out:
495 	return filter;
496 }
497 
498 /**
499  * seccomp_attach_filter: validate and attach filter
500  * @flags:  flags to change filter behavior
501  * @filter: seccomp filter to add to the current process
502  *
503  * Caller must be holding current->sighand->siglock lock.
504  *
505  * Returns 0 on success, -ve on error.
506  */
507 static long seccomp_attach_filter(unsigned int flags,
508 				  struct seccomp_filter *filter)
509 {
510 	unsigned long total_insns;
511 	struct seccomp_filter *walker;
512 
513 	assert_spin_locked(&current->sighand->siglock);
514 
515 	/* Validate resulting filter length. */
516 	total_insns = filter->prog->len;
517 	for (walker = current->seccomp.filter; walker; walker = walker->prev)
518 		total_insns += walker->prog->len + 4;  /* 4 instr penalty */
519 	if (total_insns > MAX_INSNS_PER_PATH)
520 		return -ENOMEM;
521 
522 	/* If thread sync has been requested, check that it is possible. */
523 	if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
524 		int ret;
525 
526 		ret = seccomp_can_sync_threads();
527 		if (ret)
528 			return ret;
529 	}
530 
531 	/* Set log flag, if present. */
532 	if (flags & SECCOMP_FILTER_FLAG_LOG)
533 		filter->log = true;
534 
535 	/*
536 	 * If there is an existing filter, make it the prev and don't drop its
537 	 * task reference.
538 	 */
539 	filter->prev = current->seccomp.filter;
540 	current->seccomp.filter = filter;
541 
542 	/* Now that the new filter is in place, synchronize to all threads. */
543 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
544 		seccomp_sync_threads(flags);
545 
546 	return 0;
547 }
548 
549 static void __get_seccomp_filter(struct seccomp_filter *filter)
550 {
551 	refcount_inc(&filter->usage);
552 }
553 
554 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
555 void get_seccomp_filter(struct task_struct *tsk)
556 {
557 	struct seccomp_filter *orig = tsk->seccomp.filter;
558 	if (!orig)
559 		return;
560 	__get_seccomp_filter(orig);
561 }
562 
563 static inline void seccomp_filter_free(struct seccomp_filter *filter)
564 {
565 	if (filter) {
566 		bpf_prog_destroy(filter->prog);
567 		kfree(filter);
568 	}
569 }
570 
571 static void __put_seccomp_filter(struct seccomp_filter *orig)
572 {
573 	/* Clean up single-reference branches iteratively. */
574 	while (orig && refcount_dec_and_test(&orig->usage)) {
575 		struct seccomp_filter *freeme = orig;
576 		orig = orig->prev;
577 		seccomp_filter_free(freeme);
578 	}
579 }
580 
581 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
582 void put_seccomp_filter(struct task_struct *tsk)
583 {
584 	__put_seccomp_filter(tsk->seccomp.filter);
585 }
586 
587 static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason)
588 {
589 	clear_siginfo(info);
590 	info->si_signo = SIGSYS;
591 	info->si_code = SYS_SECCOMP;
592 	info->si_call_addr = (void __user *)KSTK_EIP(current);
593 	info->si_errno = reason;
594 	info->si_arch = syscall_get_arch();
595 	info->si_syscall = syscall;
596 }
597 
598 /**
599  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
600  * @syscall: syscall number to send to userland
601  * @reason: filter-supplied reason code to send to userland (via si_errno)
602  *
603  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
604  */
605 static void seccomp_send_sigsys(int syscall, int reason)
606 {
607 	struct kernel_siginfo info;
608 	seccomp_init_siginfo(&info, syscall, reason);
609 	force_sig_info(SIGSYS, &info, current);
610 }
611 #endif	/* CONFIG_SECCOMP_FILTER */
612 
613 /* For use with seccomp_actions_logged */
614 #define SECCOMP_LOG_KILL_PROCESS	(1 << 0)
615 #define SECCOMP_LOG_KILL_THREAD		(1 << 1)
616 #define SECCOMP_LOG_TRAP		(1 << 2)
617 #define SECCOMP_LOG_ERRNO		(1 << 3)
618 #define SECCOMP_LOG_TRACE		(1 << 4)
619 #define SECCOMP_LOG_LOG			(1 << 5)
620 #define SECCOMP_LOG_ALLOW		(1 << 6)
621 #define SECCOMP_LOG_USER_NOTIF		(1 << 7)
622 
623 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
624 				    SECCOMP_LOG_KILL_THREAD  |
625 				    SECCOMP_LOG_TRAP  |
626 				    SECCOMP_LOG_ERRNO |
627 				    SECCOMP_LOG_USER_NOTIF |
628 				    SECCOMP_LOG_TRACE |
629 				    SECCOMP_LOG_LOG;
630 
631 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
632 			       bool requested)
633 {
634 	bool log = false;
635 
636 	switch (action) {
637 	case SECCOMP_RET_ALLOW:
638 		break;
639 	case SECCOMP_RET_TRAP:
640 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
641 		break;
642 	case SECCOMP_RET_ERRNO:
643 		log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
644 		break;
645 	case SECCOMP_RET_TRACE:
646 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
647 		break;
648 	case SECCOMP_RET_USER_NOTIF:
649 		log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
650 		break;
651 	case SECCOMP_RET_LOG:
652 		log = seccomp_actions_logged & SECCOMP_LOG_LOG;
653 		break;
654 	case SECCOMP_RET_KILL_THREAD:
655 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
656 		break;
657 	case SECCOMP_RET_KILL_PROCESS:
658 	default:
659 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
660 	}
661 
662 	/*
663 	 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
664 	 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
665 	 * any action from being logged by removing the action name from the
666 	 * seccomp_actions_logged sysctl.
667 	 */
668 	if (!log)
669 		return;
670 
671 	audit_seccomp(syscall, signr, action);
672 }
673 
674 /*
675  * Secure computing mode 1 allows only read/write/exit/sigreturn.
676  * To be fully secure this must be combined with rlimit
677  * to limit the stack allocations too.
678  */
679 static const int mode1_syscalls[] = {
680 	__NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
681 	0, /* null terminated */
682 };
683 
684 static void __secure_computing_strict(int this_syscall)
685 {
686 	const int *syscall_whitelist = mode1_syscalls;
687 #ifdef CONFIG_COMPAT
688 	if (in_compat_syscall())
689 		syscall_whitelist = get_compat_mode1_syscalls();
690 #endif
691 	do {
692 		if (*syscall_whitelist == this_syscall)
693 			return;
694 	} while (*++syscall_whitelist);
695 
696 #ifdef SECCOMP_DEBUG
697 	dump_stack();
698 #endif
699 	seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
700 	do_exit(SIGKILL);
701 }
702 
703 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
704 void secure_computing_strict(int this_syscall)
705 {
706 	int mode = current->seccomp.mode;
707 
708 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
709 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
710 		return;
711 
712 	if (mode == SECCOMP_MODE_DISABLED)
713 		return;
714 	else if (mode == SECCOMP_MODE_STRICT)
715 		__secure_computing_strict(this_syscall);
716 	else
717 		BUG();
718 }
719 #else
720 
721 #ifdef CONFIG_SECCOMP_FILTER
722 static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
723 {
724 	/*
725 	 * Note: overflow is ok here, the id just needs to be unique per
726 	 * filter.
727 	 */
728 	lockdep_assert_held(&filter->notify_lock);
729 	return filter->notif->next_id++;
730 }
731 
732 static void seccomp_do_user_notification(int this_syscall,
733 					 struct seccomp_filter *match,
734 					 const struct seccomp_data *sd)
735 {
736 	int err;
737 	long ret = 0;
738 	struct seccomp_knotif n = {};
739 
740 	mutex_lock(&match->notify_lock);
741 	err = -ENOSYS;
742 	if (!match->notif)
743 		goto out;
744 
745 	n.task = current;
746 	n.state = SECCOMP_NOTIFY_INIT;
747 	n.data = sd;
748 	n.id = seccomp_next_notify_id(match);
749 	init_completion(&n.ready);
750 	list_add(&n.list, &match->notif->notifications);
751 
752 	up(&match->notif->request);
753 	wake_up_poll(&match->notif->wqh, EPOLLIN | EPOLLRDNORM);
754 	mutex_unlock(&match->notify_lock);
755 
756 	/*
757 	 * This is where we wait for a reply from userspace.
758 	 */
759 	err = wait_for_completion_interruptible(&n.ready);
760 	mutex_lock(&match->notify_lock);
761 	if (err == 0) {
762 		ret = n.val;
763 		err = n.error;
764 	}
765 
766 	/*
767 	 * Note that it's possible the listener died in between the time when
768 	 * we were notified of a respons (or a signal) and when we were able to
769 	 * re-acquire the lock, so only delete from the list if the
770 	 * notification actually exists.
771 	 *
772 	 * Also note that this test is only valid because there's no way to
773 	 * *reattach* to a notifier right now. If one is added, we'll need to
774 	 * keep track of the notif itself and make sure they match here.
775 	 */
776 	if (match->notif)
777 		list_del(&n.list);
778 out:
779 	mutex_unlock(&match->notify_lock);
780 	syscall_set_return_value(current, task_pt_regs(current),
781 				 err, ret);
782 }
783 
784 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
785 			    const bool recheck_after_trace)
786 {
787 	u32 filter_ret, action;
788 	struct seccomp_filter *match = NULL;
789 	int data;
790 	struct seccomp_data sd_local;
791 
792 	/*
793 	 * Make sure that any changes to mode from another thread have
794 	 * been seen after TIF_SECCOMP was seen.
795 	 */
796 	rmb();
797 
798 	if (!sd) {
799 		populate_seccomp_data(&sd_local);
800 		sd = &sd_local;
801 	}
802 
803 	filter_ret = seccomp_run_filters(sd, &match);
804 	data = filter_ret & SECCOMP_RET_DATA;
805 	action = filter_ret & SECCOMP_RET_ACTION_FULL;
806 
807 	switch (action) {
808 	case SECCOMP_RET_ERRNO:
809 		/* Set low-order bits as an errno, capped at MAX_ERRNO. */
810 		if (data > MAX_ERRNO)
811 			data = MAX_ERRNO;
812 		syscall_set_return_value(current, task_pt_regs(current),
813 					 -data, 0);
814 		goto skip;
815 
816 	case SECCOMP_RET_TRAP:
817 		/* Show the handler the original registers. */
818 		syscall_rollback(current, task_pt_regs(current));
819 		/* Let the filter pass back 16 bits of data. */
820 		seccomp_send_sigsys(this_syscall, data);
821 		goto skip;
822 
823 	case SECCOMP_RET_TRACE:
824 		/* We've been put in this state by the ptracer already. */
825 		if (recheck_after_trace)
826 			return 0;
827 
828 		/* ENOSYS these calls if there is no tracer attached. */
829 		if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
830 			syscall_set_return_value(current,
831 						 task_pt_regs(current),
832 						 -ENOSYS, 0);
833 			goto skip;
834 		}
835 
836 		/* Allow the BPF to provide the event message */
837 		ptrace_event(PTRACE_EVENT_SECCOMP, data);
838 		/*
839 		 * The delivery of a fatal signal during event
840 		 * notification may silently skip tracer notification,
841 		 * which could leave us with a potentially unmodified
842 		 * syscall that the tracer would have liked to have
843 		 * changed. Since the process is about to die, we just
844 		 * force the syscall to be skipped and let the signal
845 		 * kill the process and correctly handle any tracer exit
846 		 * notifications.
847 		 */
848 		if (fatal_signal_pending(current))
849 			goto skip;
850 		/* Check if the tracer forced the syscall to be skipped. */
851 		this_syscall = syscall_get_nr(current, task_pt_regs(current));
852 		if (this_syscall < 0)
853 			goto skip;
854 
855 		/*
856 		 * Recheck the syscall, since it may have changed. This
857 		 * intentionally uses a NULL struct seccomp_data to force
858 		 * a reload of all registers. This does not goto skip since
859 		 * a skip would have already been reported.
860 		 */
861 		if (__seccomp_filter(this_syscall, NULL, true))
862 			return -1;
863 
864 		return 0;
865 
866 	case SECCOMP_RET_USER_NOTIF:
867 		seccomp_do_user_notification(this_syscall, match, sd);
868 		goto skip;
869 
870 	case SECCOMP_RET_LOG:
871 		seccomp_log(this_syscall, 0, action, true);
872 		return 0;
873 
874 	case SECCOMP_RET_ALLOW:
875 		/*
876 		 * Note that the "match" filter will always be NULL for
877 		 * this action since SECCOMP_RET_ALLOW is the starting
878 		 * state in seccomp_run_filters().
879 		 */
880 		return 0;
881 
882 	case SECCOMP_RET_KILL_THREAD:
883 	case SECCOMP_RET_KILL_PROCESS:
884 	default:
885 		seccomp_log(this_syscall, SIGSYS, action, true);
886 		/* Dump core only if this is the last remaining thread. */
887 		if (action == SECCOMP_RET_KILL_PROCESS ||
888 		    get_nr_threads(current) == 1) {
889 			kernel_siginfo_t info;
890 
891 			/* Show the original registers in the dump. */
892 			syscall_rollback(current, task_pt_regs(current));
893 			/* Trigger a manual coredump since do_exit skips it. */
894 			seccomp_init_siginfo(&info, this_syscall, data);
895 			do_coredump(&info);
896 		}
897 		if (action == SECCOMP_RET_KILL_PROCESS)
898 			do_group_exit(SIGSYS);
899 		else
900 			do_exit(SIGSYS);
901 	}
902 
903 	unreachable();
904 
905 skip:
906 	seccomp_log(this_syscall, 0, action, match ? match->log : false);
907 	return -1;
908 }
909 #else
910 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
911 			    const bool recheck_after_trace)
912 {
913 	BUG();
914 }
915 #endif
916 
917 int __secure_computing(const struct seccomp_data *sd)
918 {
919 	int mode = current->seccomp.mode;
920 	int this_syscall;
921 
922 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
923 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
924 		return 0;
925 
926 	this_syscall = sd ? sd->nr :
927 		syscall_get_nr(current, task_pt_regs(current));
928 
929 	switch (mode) {
930 	case SECCOMP_MODE_STRICT:
931 		__secure_computing_strict(this_syscall);  /* may call do_exit */
932 		return 0;
933 	case SECCOMP_MODE_FILTER:
934 		return __seccomp_filter(this_syscall, sd, false);
935 	default:
936 		BUG();
937 	}
938 }
939 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
940 
941 long prctl_get_seccomp(void)
942 {
943 	return current->seccomp.mode;
944 }
945 
946 /**
947  * seccomp_set_mode_strict: internal function for setting strict seccomp
948  *
949  * Once current->seccomp.mode is non-zero, it may not be changed.
950  *
951  * Returns 0 on success or -EINVAL on failure.
952  */
953 static long seccomp_set_mode_strict(void)
954 {
955 	const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
956 	long ret = -EINVAL;
957 
958 	spin_lock_irq(&current->sighand->siglock);
959 
960 	if (!seccomp_may_assign_mode(seccomp_mode))
961 		goto out;
962 
963 #ifdef TIF_NOTSC
964 	disable_TSC();
965 #endif
966 	seccomp_assign_mode(current, seccomp_mode, 0);
967 	ret = 0;
968 
969 out:
970 	spin_unlock_irq(&current->sighand->siglock);
971 
972 	return ret;
973 }
974 
975 #ifdef CONFIG_SECCOMP_FILTER
976 static int seccomp_notify_release(struct inode *inode, struct file *file)
977 {
978 	struct seccomp_filter *filter = file->private_data;
979 	struct seccomp_knotif *knotif;
980 
981 	if (!filter)
982 		return 0;
983 
984 	mutex_lock(&filter->notify_lock);
985 
986 	/*
987 	 * If this file is being closed because e.g. the task who owned it
988 	 * died, let's wake everyone up who was waiting on us.
989 	 */
990 	list_for_each_entry(knotif, &filter->notif->notifications, list) {
991 		if (knotif->state == SECCOMP_NOTIFY_REPLIED)
992 			continue;
993 
994 		knotif->state = SECCOMP_NOTIFY_REPLIED;
995 		knotif->error = -ENOSYS;
996 		knotif->val = 0;
997 
998 		complete(&knotif->ready);
999 	}
1000 
1001 	kfree(filter->notif);
1002 	filter->notif = NULL;
1003 	mutex_unlock(&filter->notify_lock);
1004 	__put_seccomp_filter(filter);
1005 	return 0;
1006 }
1007 
1008 static long seccomp_notify_recv(struct seccomp_filter *filter,
1009 				void __user *buf)
1010 {
1011 	struct seccomp_knotif *knotif = NULL, *cur;
1012 	struct seccomp_notif unotif;
1013 	ssize_t ret;
1014 
1015 	memset(&unotif, 0, sizeof(unotif));
1016 
1017 	ret = down_interruptible(&filter->notif->request);
1018 	if (ret < 0)
1019 		return ret;
1020 
1021 	mutex_lock(&filter->notify_lock);
1022 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1023 		if (cur->state == SECCOMP_NOTIFY_INIT) {
1024 			knotif = cur;
1025 			break;
1026 		}
1027 	}
1028 
1029 	/*
1030 	 * If we didn't find a notification, it could be that the task was
1031 	 * interrupted by a fatal signal between the time we were woken and
1032 	 * when we were able to acquire the rw lock.
1033 	 */
1034 	if (!knotif) {
1035 		ret = -ENOENT;
1036 		goto out;
1037 	}
1038 
1039 	unotif.id = knotif->id;
1040 	unotif.pid = task_pid_vnr(knotif->task);
1041 	unotif.data = *(knotif->data);
1042 
1043 	knotif->state = SECCOMP_NOTIFY_SENT;
1044 	wake_up_poll(&filter->notif->wqh, EPOLLOUT | EPOLLWRNORM);
1045 	ret = 0;
1046 out:
1047 	mutex_unlock(&filter->notify_lock);
1048 
1049 	if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1050 		ret = -EFAULT;
1051 
1052 		/*
1053 		 * Userspace screwed up. To make sure that we keep this
1054 		 * notification alive, let's reset it back to INIT. It
1055 		 * may have died when we released the lock, so we need to make
1056 		 * sure it's still around.
1057 		 */
1058 		knotif = NULL;
1059 		mutex_lock(&filter->notify_lock);
1060 		list_for_each_entry(cur, &filter->notif->notifications, list) {
1061 			if (cur->id == unotif.id) {
1062 				knotif = cur;
1063 				break;
1064 			}
1065 		}
1066 
1067 		if (knotif) {
1068 			knotif->state = SECCOMP_NOTIFY_INIT;
1069 			up(&filter->notif->request);
1070 		}
1071 		mutex_unlock(&filter->notify_lock);
1072 	}
1073 
1074 	return ret;
1075 }
1076 
1077 static long seccomp_notify_send(struct seccomp_filter *filter,
1078 				void __user *buf)
1079 {
1080 	struct seccomp_notif_resp resp = {};
1081 	struct seccomp_knotif *knotif = NULL, *cur;
1082 	long ret;
1083 
1084 	if (copy_from_user(&resp, buf, sizeof(resp)))
1085 		return -EFAULT;
1086 
1087 	if (resp.flags)
1088 		return -EINVAL;
1089 
1090 	ret = mutex_lock_interruptible(&filter->notify_lock);
1091 	if (ret < 0)
1092 		return ret;
1093 
1094 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1095 		if (cur->id == resp.id) {
1096 			knotif = cur;
1097 			break;
1098 		}
1099 	}
1100 
1101 	if (!knotif) {
1102 		ret = -ENOENT;
1103 		goto out;
1104 	}
1105 
1106 	/* Allow exactly one reply. */
1107 	if (knotif->state != SECCOMP_NOTIFY_SENT) {
1108 		ret = -EINPROGRESS;
1109 		goto out;
1110 	}
1111 
1112 	ret = 0;
1113 	knotif->state = SECCOMP_NOTIFY_REPLIED;
1114 	knotif->error = resp.error;
1115 	knotif->val = resp.val;
1116 	complete(&knotif->ready);
1117 out:
1118 	mutex_unlock(&filter->notify_lock);
1119 	return ret;
1120 }
1121 
1122 static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1123 				    void __user *buf)
1124 {
1125 	struct seccomp_knotif *knotif = NULL;
1126 	u64 id;
1127 	long ret;
1128 
1129 	if (copy_from_user(&id, buf, sizeof(id)))
1130 		return -EFAULT;
1131 
1132 	ret = mutex_lock_interruptible(&filter->notify_lock);
1133 	if (ret < 0)
1134 		return ret;
1135 
1136 	ret = -ENOENT;
1137 	list_for_each_entry(knotif, &filter->notif->notifications, list) {
1138 		if (knotif->id == id) {
1139 			if (knotif->state == SECCOMP_NOTIFY_SENT)
1140 				ret = 0;
1141 			goto out;
1142 		}
1143 	}
1144 
1145 out:
1146 	mutex_unlock(&filter->notify_lock);
1147 	return ret;
1148 }
1149 
1150 static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1151 				 unsigned long arg)
1152 {
1153 	struct seccomp_filter *filter = file->private_data;
1154 	void __user *buf = (void __user *)arg;
1155 
1156 	switch (cmd) {
1157 	case SECCOMP_IOCTL_NOTIF_RECV:
1158 		return seccomp_notify_recv(filter, buf);
1159 	case SECCOMP_IOCTL_NOTIF_SEND:
1160 		return seccomp_notify_send(filter, buf);
1161 	case SECCOMP_IOCTL_NOTIF_ID_VALID:
1162 		return seccomp_notify_id_valid(filter, buf);
1163 	default:
1164 		return -EINVAL;
1165 	}
1166 }
1167 
1168 static __poll_t seccomp_notify_poll(struct file *file,
1169 				    struct poll_table_struct *poll_tab)
1170 {
1171 	struct seccomp_filter *filter = file->private_data;
1172 	__poll_t ret = 0;
1173 	struct seccomp_knotif *cur;
1174 
1175 	poll_wait(file, &filter->notif->wqh, poll_tab);
1176 
1177 	if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1178 		return EPOLLERR;
1179 
1180 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1181 		if (cur->state == SECCOMP_NOTIFY_INIT)
1182 			ret |= EPOLLIN | EPOLLRDNORM;
1183 		if (cur->state == SECCOMP_NOTIFY_SENT)
1184 			ret |= EPOLLOUT | EPOLLWRNORM;
1185 		if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1186 			break;
1187 	}
1188 
1189 	mutex_unlock(&filter->notify_lock);
1190 
1191 	return ret;
1192 }
1193 
1194 static const struct file_operations seccomp_notify_ops = {
1195 	.poll = seccomp_notify_poll,
1196 	.release = seccomp_notify_release,
1197 	.unlocked_ioctl = seccomp_notify_ioctl,
1198 };
1199 
1200 static struct file *init_listener(struct seccomp_filter *filter)
1201 {
1202 	struct file *ret = ERR_PTR(-EBUSY);
1203 	struct seccomp_filter *cur;
1204 
1205 	for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1206 		if (cur->notif)
1207 			goto out;
1208 	}
1209 
1210 	ret = ERR_PTR(-ENOMEM);
1211 	filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1212 	if (!filter->notif)
1213 		goto out;
1214 
1215 	sema_init(&filter->notif->request, 0);
1216 	filter->notif->next_id = get_random_u64();
1217 	INIT_LIST_HEAD(&filter->notif->notifications);
1218 	init_waitqueue_head(&filter->notif->wqh);
1219 
1220 	ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1221 				 filter, O_RDWR);
1222 	if (IS_ERR(ret))
1223 		goto out_notif;
1224 
1225 	/* The file has a reference to it now */
1226 	__get_seccomp_filter(filter);
1227 
1228 out_notif:
1229 	if (IS_ERR(ret))
1230 		kfree(filter->notif);
1231 out:
1232 	return ret;
1233 }
1234 
1235 /**
1236  * seccomp_set_mode_filter: internal function for setting seccomp filter
1237  * @flags:  flags to change filter behavior
1238  * @filter: struct sock_fprog containing filter
1239  *
1240  * This function may be called repeatedly to install additional filters.
1241  * Every filter successfully installed will be evaluated (in reverse order)
1242  * for each system call the task makes.
1243  *
1244  * Once current->seccomp.mode is non-zero, it may not be changed.
1245  *
1246  * Returns 0 on success or -EINVAL on failure.
1247  */
1248 static long seccomp_set_mode_filter(unsigned int flags,
1249 				    const char __user *filter)
1250 {
1251 	const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1252 	struct seccomp_filter *prepared = NULL;
1253 	long ret = -EINVAL;
1254 	int listener = -1;
1255 	struct file *listener_f = NULL;
1256 
1257 	/* Validate flags. */
1258 	if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1259 		return -EINVAL;
1260 
1261 	/* Prepare the new filter before holding any locks. */
1262 	prepared = seccomp_prepare_user_filter(filter);
1263 	if (IS_ERR(prepared))
1264 		return PTR_ERR(prepared);
1265 
1266 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1267 		listener = get_unused_fd_flags(O_CLOEXEC);
1268 		if (listener < 0) {
1269 			ret = listener;
1270 			goto out_free;
1271 		}
1272 
1273 		listener_f = init_listener(prepared);
1274 		if (IS_ERR(listener_f)) {
1275 			put_unused_fd(listener);
1276 			ret = PTR_ERR(listener_f);
1277 			goto out_free;
1278 		}
1279 	}
1280 
1281 	/*
1282 	 * Make sure we cannot change seccomp or nnp state via TSYNC
1283 	 * while another thread is in the middle of calling exec.
1284 	 */
1285 	if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1286 	    mutex_lock_killable(&current->signal->cred_guard_mutex))
1287 		goto out_put_fd;
1288 
1289 	spin_lock_irq(&current->sighand->siglock);
1290 
1291 	if (!seccomp_may_assign_mode(seccomp_mode))
1292 		goto out;
1293 
1294 	ret = seccomp_attach_filter(flags, prepared);
1295 	if (ret)
1296 		goto out;
1297 	/* Do not free the successfully attached filter. */
1298 	prepared = NULL;
1299 
1300 	seccomp_assign_mode(current, seccomp_mode, flags);
1301 out:
1302 	spin_unlock_irq(&current->sighand->siglock);
1303 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
1304 		mutex_unlock(&current->signal->cred_guard_mutex);
1305 out_put_fd:
1306 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1307 		if (ret < 0) {
1308 			listener_f->private_data = NULL;
1309 			fput(listener_f);
1310 			put_unused_fd(listener);
1311 		} else {
1312 			fd_install(listener, listener_f);
1313 			ret = listener;
1314 		}
1315 	}
1316 out_free:
1317 	seccomp_filter_free(prepared);
1318 	return ret;
1319 }
1320 #else
1321 static inline long seccomp_set_mode_filter(unsigned int flags,
1322 					   const char __user *filter)
1323 {
1324 	return -EINVAL;
1325 }
1326 #endif
1327 
1328 static long seccomp_get_action_avail(const char __user *uaction)
1329 {
1330 	u32 action;
1331 
1332 	if (copy_from_user(&action, uaction, sizeof(action)))
1333 		return -EFAULT;
1334 
1335 	switch (action) {
1336 	case SECCOMP_RET_KILL_PROCESS:
1337 	case SECCOMP_RET_KILL_THREAD:
1338 	case SECCOMP_RET_TRAP:
1339 	case SECCOMP_RET_ERRNO:
1340 	case SECCOMP_RET_USER_NOTIF:
1341 	case SECCOMP_RET_TRACE:
1342 	case SECCOMP_RET_LOG:
1343 	case SECCOMP_RET_ALLOW:
1344 		break;
1345 	default:
1346 		return -EOPNOTSUPP;
1347 	}
1348 
1349 	return 0;
1350 }
1351 
1352 static long seccomp_get_notif_sizes(void __user *usizes)
1353 {
1354 	struct seccomp_notif_sizes sizes = {
1355 		.seccomp_notif = sizeof(struct seccomp_notif),
1356 		.seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
1357 		.seccomp_data = sizeof(struct seccomp_data),
1358 	};
1359 
1360 	if (copy_to_user(usizes, &sizes, sizeof(sizes)))
1361 		return -EFAULT;
1362 
1363 	return 0;
1364 }
1365 
1366 /* Common entry point for both prctl and syscall. */
1367 static long do_seccomp(unsigned int op, unsigned int flags,
1368 		       void __user *uargs)
1369 {
1370 	switch (op) {
1371 	case SECCOMP_SET_MODE_STRICT:
1372 		if (flags != 0 || uargs != NULL)
1373 			return -EINVAL;
1374 		return seccomp_set_mode_strict();
1375 	case SECCOMP_SET_MODE_FILTER:
1376 		return seccomp_set_mode_filter(flags, uargs);
1377 	case SECCOMP_GET_ACTION_AVAIL:
1378 		if (flags != 0)
1379 			return -EINVAL;
1380 
1381 		return seccomp_get_action_avail(uargs);
1382 	case SECCOMP_GET_NOTIF_SIZES:
1383 		if (flags != 0)
1384 			return -EINVAL;
1385 
1386 		return seccomp_get_notif_sizes(uargs);
1387 	default:
1388 		return -EINVAL;
1389 	}
1390 }
1391 
1392 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
1393 			 void __user *, uargs)
1394 {
1395 	return do_seccomp(op, flags, uargs);
1396 }
1397 
1398 /**
1399  * prctl_set_seccomp: configures current->seccomp.mode
1400  * @seccomp_mode: requested mode to use
1401  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1402  *
1403  * Returns 0 on success or -EINVAL on failure.
1404  */
1405 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
1406 {
1407 	unsigned int op;
1408 	void __user *uargs;
1409 
1410 	switch (seccomp_mode) {
1411 	case SECCOMP_MODE_STRICT:
1412 		op = SECCOMP_SET_MODE_STRICT;
1413 		/*
1414 		 * Setting strict mode through prctl always ignored filter,
1415 		 * so make sure it is always NULL here to pass the internal
1416 		 * check in do_seccomp().
1417 		 */
1418 		uargs = NULL;
1419 		break;
1420 	case SECCOMP_MODE_FILTER:
1421 		op = SECCOMP_SET_MODE_FILTER;
1422 		uargs = filter;
1423 		break;
1424 	default:
1425 		return -EINVAL;
1426 	}
1427 
1428 	/* prctl interface doesn't have flags, so they are always zero. */
1429 	return do_seccomp(op, 0, uargs);
1430 }
1431 
1432 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
1433 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
1434 					     unsigned long filter_off)
1435 {
1436 	struct seccomp_filter *orig, *filter;
1437 	unsigned long count;
1438 
1439 	/*
1440 	 * Note: this is only correct because the caller should be the (ptrace)
1441 	 * tracer of the task, otherwise lock_task_sighand is needed.
1442 	 */
1443 	spin_lock_irq(&task->sighand->siglock);
1444 
1445 	if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
1446 		spin_unlock_irq(&task->sighand->siglock);
1447 		return ERR_PTR(-EINVAL);
1448 	}
1449 
1450 	orig = task->seccomp.filter;
1451 	__get_seccomp_filter(orig);
1452 	spin_unlock_irq(&task->sighand->siglock);
1453 
1454 	count = 0;
1455 	for (filter = orig; filter; filter = filter->prev)
1456 		count++;
1457 
1458 	if (filter_off >= count) {
1459 		filter = ERR_PTR(-ENOENT);
1460 		goto out;
1461 	}
1462 
1463 	count -= filter_off;
1464 	for (filter = orig; filter && count > 1; filter = filter->prev)
1465 		count--;
1466 
1467 	if (WARN_ON(count != 1 || !filter)) {
1468 		filter = ERR_PTR(-ENOENT);
1469 		goto out;
1470 	}
1471 
1472 	__get_seccomp_filter(filter);
1473 
1474 out:
1475 	__put_seccomp_filter(orig);
1476 	return filter;
1477 }
1478 
1479 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1480 			void __user *data)
1481 {
1482 	struct seccomp_filter *filter;
1483 	struct sock_fprog_kern *fprog;
1484 	long ret;
1485 
1486 	if (!capable(CAP_SYS_ADMIN) ||
1487 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1488 		return -EACCES;
1489 	}
1490 
1491 	filter = get_nth_filter(task, filter_off);
1492 	if (IS_ERR(filter))
1493 		return PTR_ERR(filter);
1494 
1495 	fprog = filter->prog->orig_prog;
1496 	if (!fprog) {
1497 		/* This must be a new non-cBPF filter, since we save
1498 		 * every cBPF filter's orig_prog above when
1499 		 * CONFIG_CHECKPOINT_RESTORE is enabled.
1500 		 */
1501 		ret = -EMEDIUMTYPE;
1502 		goto out;
1503 	}
1504 
1505 	ret = fprog->len;
1506 	if (!data)
1507 		goto out;
1508 
1509 	if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1510 		ret = -EFAULT;
1511 
1512 out:
1513 	__put_seccomp_filter(filter);
1514 	return ret;
1515 }
1516 
1517 long seccomp_get_metadata(struct task_struct *task,
1518 			  unsigned long size, void __user *data)
1519 {
1520 	long ret;
1521 	struct seccomp_filter *filter;
1522 	struct seccomp_metadata kmd = {};
1523 
1524 	if (!capable(CAP_SYS_ADMIN) ||
1525 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1526 		return -EACCES;
1527 	}
1528 
1529 	size = min_t(unsigned long, size, sizeof(kmd));
1530 
1531 	if (size < sizeof(kmd.filter_off))
1532 		return -EINVAL;
1533 
1534 	if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
1535 		return -EFAULT;
1536 
1537 	filter = get_nth_filter(task, kmd.filter_off);
1538 	if (IS_ERR(filter))
1539 		return PTR_ERR(filter);
1540 
1541 	if (filter->log)
1542 		kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1543 
1544 	ret = size;
1545 	if (copy_to_user(data, &kmd, size))
1546 		ret = -EFAULT;
1547 
1548 	__put_seccomp_filter(filter);
1549 	return ret;
1550 }
1551 #endif
1552 
1553 #ifdef CONFIG_SYSCTL
1554 
1555 /* Human readable action names for friendly sysctl interaction */
1556 #define SECCOMP_RET_KILL_PROCESS_NAME	"kill_process"
1557 #define SECCOMP_RET_KILL_THREAD_NAME	"kill_thread"
1558 #define SECCOMP_RET_TRAP_NAME		"trap"
1559 #define SECCOMP_RET_ERRNO_NAME		"errno"
1560 #define SECCOMP_RET_USER_NOTIF_NAME	"user_notif"
1561 #define SECCOMP_RET_TRACE_NAME		"trace"
1562 #define SECCOMP_RET_LOG_NAME		"log"
1563 #define SECCOMP_RET_ALLOW_NAME		"allow"
1564 
1565 static const char seccomp_actions_avail[] =
1566 				SECCOMP_RET_KILL_PROCESS_NAME	" "
1567 				SECCOMP_RET_KILL_THREAD_NAME	" "
1568 				SECCOMP_RET_TRAP_NAME		" "
1569 				SECCOMP_RET_ERRNO_NAME		" "
1570 				SECCOMP_RET_USER_NOTIF_NAME     " "
1571 				SECCOMP_RET_TRACE_NAME		" "
1572 				SECCOMP_RET_LOG_NAME		" "
1573 				SECCOMP_RET_ALLOW_NAME;
1574 
1575 struct seccomp_log_name {
1576 	u32		log;
1577 	const char	*name;
1578 };
1579 
1580 static const struct seccomp_log_name seccomp_log_names[] = {
1581 	{ SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1582 	{ SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1583 	{ SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1584 	{ SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1585 	{ SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
1586 	{ SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1587 	{ SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1588 	{ SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1589 	{ }
1590 };
1591 
1592 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1593 					      u32 actions_logged,
1594 					      const char *sep)
1595 {
1596 	const struct seccomp_log_name *cur;
1597 	bool append_sep = false;
1598 
1599 	for (cur = seccomp_log_names; cur->name && size; cur++) {
1600 		ssize_t ret;
1601 
1602 		if (!(actions_logged & cur->log))
1603 			continue;
1604 
1605 		if (append_sep) {
1606 			ret = strscpy(names, sep, size);
1607 			if (ret < 0)
1608 				return false;
1609 
1610 			names += ret;
1611 			size -= ret;
1612 		} else
1613 			append_sep = true;
1614 
1615 		ret = strscpy(names, cur->name, size);
1616 		if (ret < 0)
1617 			return false;
1618 
1619 		names += ret;
1620 		size -= ret;
1621 	}
1622 
1623 	return true;
1624 }
1625 
1626 static bool seccomp_action_logged_from_name(u32 *action_logged,
1627 					    const char *name)
1628 {
1629 	const struct seccomp_log_name *cur;
1630 
1631 	for (cur = seccomp_log_names; cur->name; cur++) {
1632 		if (!strcmp(cur->name, name)) {
1633 			*action_logged = cur->log;
1634 			return true;
1635 		}
1636 	}
1637 
1638 	return false;
1639 }
1640 
1641 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1642 {
1643 	char *name;
1644 
1645 	*actions_logged = 0;
1646 	while ((name = strsep(&names, " ")) && *name) {
1647 		u32 action_logged = 0;
1648 
1649 		if (!seccomp_action_logged_from_name(&action_logged, name))
1650 			return false;
1651 
1652 		*actions_logged |= action_logged;
1653 	}
1654 
1655 	return true;
1656 }
1657 
1658 static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1659 			       size_t *lenp, loff_t *ppos)
1660 {
1661 	char names[sizeof(seccomp_actions_avail)];
1662 	struct ctl_table table;
1663 
1664 	memset(names, 0, sizeof(names));
1665 
1666 	if (!seccomp_names_from_actions_logged(names, sizeof(names),
1667 					       seccomp_actions_logged, " "))
1668 		return -EINVAL;
1669 
1670 	table = *ro_table;
1671 	table.data = names;
1672 	table.maxlen = sizeof(names);
1673 	return proc_dostring(&table, 0, buffer, lenp, ppos);
1674 }
1675 
1676 static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1677 				size_t *lenp, loff_t *ppos, u32 *actions_logged)
1678 {
1679 	char names[sizeof(seccomp_actions_avail)];
1680 	struct ctl_table table;
1681 	int ret;
1682 
1683 	if (!capable(CAP_SYS_ADMIN))
1684 		return -EPERM;
1685 
1686 	memset(names, 0, sizeof(names));
1687 
1688 	table = *ro_table;
1689 	table.data = names;
1690 	table.maxlen = sizeof(names);
1691 	ret = proc_dostring(&table, 1, buffer, lenp, ppos);
1692 	if (ret)
1693 		return ret;
1694 
1695 	if (!seccomp_actions_logged_from_names(actions_logged, table.data))
1696 		return -EINVAL;
1697 
1698 	if (*actions_logged & SECCOMP_LOG_ALLOW)
1699 		return -EINVAL;
1700 
1701 	seccomp_actions_logged = *actions_logged;
1702 	return 0;
1703 }
1704 
1705 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
1706 				 int ret)
1707 {
1708 	char names[sizeof(seccomp_actions_avail)];
1709 	char old_names[sizeof(seccomp_actions_avail)];
1710 	const char *new = names;
1711 	const char *old = old_names;
1712 
1713 	if (!audit_enabled)
1714 		return;
1715 
1716 	memset(names, 0, sizeof(names));
1717 	memset(old_names, 0, sizeof(old_names));
1718 
1719 	if (ret)
1720 		new = "?";
1721 	else if (!actions_logged)
1722 		new = "(none)";
1723 	else if (!seccomp_names_from_actions_logged(names, sizeof(names),
1724 						    actions_logged, ","))
1725 		new = "?";
1726 
1727 	if (!old_actions_logged)
1728 		old = "(none)";
1729 	else if (!seccomp_names_from_actions_logged(old_names,
1730 						    sizeof(old_names),
1731 						    old_actions_logged, ","))
1732 		old = "?";
1733 
1734 	return audit_seccomp_actions_logged(new, old, !ret);
1735 }
1736 
1737 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1738 					  void __user *buffer, size_t *lenp,
1739 					  loff_t *ppos)
1740 {
1741 	int ret;
1742 
1743 	if (write) {
1744 		u32 actions_logged = 0;
1745 		u32 old_actions_logged = seccomp_actions_logged;
1746 
1747 		ret = write_actions_logged(ro_table, buffer, lenp, ppos,
1748 					   &actions_logged);
1749 		audit_actions_logged(actions_logged, old_actions_logged, ret);
1750 	} else
1751 		ret = read_actions_logged(ro_table, buffer, lenp, ppos);
1752 
1753 	return ret;
1754 }
1755 
1756 static struct ctl_path seccomp_sysctl_path[] = {
1757 	{ .procname = "kernel", },
1758 	{ .procname = "seccomp", },
1759 	{ }
1760 };
1761 
1762 static struct ctl_table seccomp_sysctl_table[] = {
1763 	{
1764 		.procname	= "actions_avail",
1765 		.data		= (void *) &seccomp_actions_avail,
1766 		.maxlen		= sizeof(seccomp_actions_avail),
1767 		.mode		= 0444,
1768 		.proc_handler	= proc_dostring,
1769 	},
1770 	{
1771 		.procname	= "actions_logged",
1772 		.mode		= 0644,
1773 		.proc_handler	= seccomp_actions_logged_handler,
1774 	},
1775 	{ }
1776 };
1777 
1778 static int __init seccomp_sysctl_init(void)
1779 {
1780 	struct ctl_table_header *hdr;
1781 
1782 	hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1783 	if (!hdr)
1784 		pr_warn("seccomp: sysctl registration failed\n");
1785 	else
1786 		kmemleak_not_leak(hdr);
1787 
1788 	return 0;
1789 }
1790 
1791 device_initcall(seccomp_sysctl_init)
1792 
1793 #endif /* CONFIG_SYSCTL */
1794