xref: /openbmc/linux/kernel/seccomp.c (revision 2fa49589)
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, 0, 6, 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 	for (; f; f = f->prev) {
271 		u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
272 
273 		if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
274 			ret = cur_ret;
275 			*match = f;
276 		}
277 	}
278 	return ret;
279 }
280 #endif /* CONFIG_SECCOMP_FILTER */
281 
282 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
283 {
284 	assert_spin_locked(&current->sighand->siglock);
285 
286 	if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
287 		return false;
288 
289 	return true;
290 }
291 
292 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
293 
294 static inline void seccomp_assign_mode(struct task_struct *task,
295 				       unsigned long seccomp_mode,
296 				       unsigned long flags)
297 {
298 	assert_spin_locked(&task->sighand->siglock);
299 
300 	task->seccomp.mode = seccomp_mode;
301 	/*
302 	 * Make sure TIF_SECCOMP cannot be set before the mode (and
303 	 * filter) is set.
304 	 */
305 	smp_mb__before_atomic();
306 	/* Assume default seccomp processes want spec flaw mitigation. */
307 	if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
308 		arch_seccomp_spec_mitigate(task);
309 	set_tsk_thread_flag(task, TIF_SECCOMP);
310 }
311 
312 #ifdef CONFIG_SECCOMP_FILTER
313 /* Returns 1 if the parent is an ancestor of the child. */
314 static int is_ancestor(struct seccomp_filter *parent,
315 		       struct seccomp_filter *child)
316 {
317 	/* NULL is the root ancestor. */
318 	if (parent == NULL)
319 		return 1;
320 	for (; child; child = child->prev)
321 		if (child == parent)
322 			return 1;
323 	return 0;
324 }
325 
326 /**
327  * seccomp_can_sync_threads: checks if all threads can be synchronized
328  *
329  * Expects sighand and cred_guard_mutex locks to be held.
330  *
331  * Returns 0 on success, -ve on error, or the pid of a thread which was
332  * either not in the correct seccomp mode or it did not have an ancestral
333  * seccomp filter.
334  */
335 static inline pid_t seccomp_can_sync_threads(void)
336 {
337 	struct task_struct *thread, *caller;
338 
339 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
340 	assert_spin_locked(&current->sighand->siglock);
341 
342 	/* Validate all threads being eligible for synchronization. */
343 	caller = current;
344 	for_each_thread(caller, thread) {
345 		pid_t failed;
346 
347 		/* Skip current, since it is initiating the sync. */
348 		if (thread == caller)
349 			continue;
350 
351 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
352 		    (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
353 		     is_ancestor(thread->seccomp.filter,
354 				 caller->seccomp.filter)))
355 			continue;
356 
357 		/* Return the first thread that cannot be synchronized. */
358 		failed = task_pid_vnr(thread);
359 		/* If the pid cannot be resolved, then return -ESRCH */
360 		if (WARN_ON(failed == 0))
361 			failed = -ESRCH;
362 		return failed;
363 	}
364 
365 	return 0;
366 }
367 
368 /**
369  * seccomp_sync_threads: sets all threads to use current's filter
370  *
371  * Expects sighand and cred_guard_mutex locks to be held, and for
372  * seccomp_can_sync_threads() to have returned success already
373  * without dropping the locks.
374  *
375  */
376 static inline void seccomp_sync_threads(unsigned long flags)
377 {
378 	struct task_struct *thread, *caller;
379 
380 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
381 	assert_spin_locked(&current->sighand->siglock);
382 
383 	/* Synchronize all threads. */
384 	caller = current;
385 	for_each_thread(caller, thread) {
386 		/* Skip current, since it needs no changes. */
387 		if (thread == caller)
388 			continue;
389 
390 		/* Get a task reference for the new leaf node. */
391 		get_seccomp_filter(caller);
392 		/*
393 		 * Drop the task reference to the shared ancestor since
394 		 * current's path will hold a reference.  (This also
395 		 * allows a put before the assignment.)
396 		 */
397 		put_seccomp_filter(thread);
398 		smp_store_release(&thread->seccomp.filter,
399 				  caller->seccomp.filter);
400 
401 		/*
402 		 * Don't let an unprivileged task work around
403 		 * the no_new_privs restriction by creating
404 		 * a thread that sets it up, enters seccomp,
405 		 * then dies.
406 		 */
407 		if (task_no_new_privs(caller))
408 			task_set_no_new_privs(thread);
409 
410 		/*
411 		 * Opt the other thread into seccomp if needed.
412 		 * As threads are considered to be trust-realm
413 		 * equivalent (see ptrace_may_access), it is safe to
414 		 * allow one thread to transition the other.
415 		 */
416 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
417 			seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
418 					    flags);
419 	}
420 }
421 
422 /**
423  * seccomp_prepare_filter: Prepares a seccomp filter for use.
424  * @fprog: BPF program to install
425  *
426  * Returns filter on success or an ERR_PTR on failure.
427  */
428 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
429 {
430 	struct seccomp_filter *sfilter;
431 	int ret;
432 	const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
433 
434 	if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
435 		return ERR_PTR(-EINVAL);
436 
437 	BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
438 
439 	/*
440 	 * Installing a seccomp filter requires that the task has
441 	 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
442 	 * This avoids scenarios where unprivileged tasks can affect the
443 	 * behavior of privileged children.
444 	 */
445 	if (!task_no_new_privs(current) &&
446 	    security_capable_noaudit(current_cred(), current_user_ns(),
447 				     CAP_SYS_ADMIN) != 0)
448 		return ERR_PTR(-EACCES);
449 
450 	/* Allocate a new seccomp_filter */
451 	sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
452 	if (!sfilter)
453 		return ERR_PTR(-ENOMEM);
454 
455 	mutex_init(&sfilter->notify_lock);
456 	ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
457 					seccomp_check_filter, save_orig);
458 	if (ret < 0) {
459 		kfree(sfilter);
460 		return ERR_PTR(ret);
461 	}
462 
463 	refcount_set(&sfilter->usage, 1);
464 
465 	return sfilter;
466 }
467 
468 /**
469  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
470  * @user_filter: pointer to the user data containing a sock_fprog.
471  *
472  * Returns 0 on success and non-zero otherwise.
473  */
474 static struct seccomp_filter *
475 seccomp_prepare_user_filter(const char __user *user_filter)
476 {
477 	struct sock_fprog fprog;
478 	struct seccomp_filter *filter = ERR_PTR(-EFAULT);
479 
480 #ifdef CONFIG_COMPAT
481 	if (in_compat_syscall()) {
482 		struct compat_sock_fprog fprog32;
483 		if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
484 			goto out;
485 		fprog.len = fprog32.len;
486 		fprog.filter = compat_ptr(fprog32.filter);
487 	} else /* falls through to the if below. */
488 #endif
489 	if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
490 		goto out;
491 	filter = seccomp_prepare_filter(&fprog);
492 out:
493 	return filter;
494 }
495 
496 /**
497  * seccomp_attach_filter: validate and attach filter
498  * @flags:  flags to change filter behavior
499  * @filter: seccomp filter to add to the current process
500  *
501  * Caller must be holding current->sighand->siglock lock.
502  *
503  * Returns 0 on success, -ve on error.
504  */
505 static long seccomp_attach_filter(unsigned int flags,
506 				  struct seccomp_filter *filter)
507 {
508 	unsigned long total_insns;
509 	struct seccomp_filter *walker;
510 
511 	assert_spin_locked(&current->sighand->siglock);
512 
513 	/* Validate resulting filter length. */
514 	total_insns = filter->prog->len;
515 	for (walker = current->seccomp.filter; walker; walker = walker->prev)
516 		total_insns += walker->prog->len + 4;  /* 4 instr penalty */
517 	if (total_insns > MAX_INSNS_PER_PATH)
518 		return -ENOMEM;
519 
520 	/* If thread sync has been requested, check that it is possible. */
521 	if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
522 		int ret;
523 
524 		ret = seccomp_can_sync_threads();
525 		if (ret)
526 			return ret;
527 	}
528 
529 	/* Set log flag, if present. */
530 	if (flags & SECCOMP_FILTER_FLAG_LOG)
531 		filter->log = true;
532 
533 	/*
534 	 * If there is an existing filter, make it the prev and don't drop its
535 	 * task reference.
536 	 */
537 	filter->prev = current->seccomp.filter;
538 	current->seccomp.filter = filter;
539 
540 	/* Now that the new filter is in place, synchronize to all threads. */
541 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
542 		seccomp_sync_threads(flags);
543 
544 	return 0;
545 }
546 
547 static void __get_seccomp_filter(struct seccomp_filter *filter)
548 {
549 	refcount_inc(&filter->usage);
550 }
551 
552 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
553 void get_seccomp_filter(struct task_struct *tsk)
554 {
555 	struct seccomp_filter *orig = tsk->seccomp.filter;
556 	if (!orig)
557 		return;
558 	__get_seccomp_filter(orig);
559 }
560 
561 static inline void seccomp_filter_free(struct seccomp_filter *filter)
562 {
563 	if (filter) {
564 		bpf_prog_destroy(filter->prog);
565 		kfree(filter);
566 	}
567 }
568 
569 static void __put_seccomp_filter(struct seccomp_filter *orig)
570 {
571 	/* Clean up single-reference branches iteratively. */
572 	while (orig && refcount_dec_and_test(&orig->usage)) {
573 		struct seccomp_filter *freeme = orig;
574 		orig = orig->prev;
575 		seccomp_filter_free(freeme);
576 	}
577 }
578 
579 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
580 void put_seccomp_filter(struct task_struct *tsk)
581 {
582 	__put_seccomp_filter(tsk->seccomp.filter);
583 }
584 
585 static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason)
586 {
587 	clear_siginfo(info);
588 	info->si_signo = SIGSYS;
589 	info->si_code = SYS_SECCOMP;
590 	info->si_call_addr = (void __user *)KSTK_EIP(current);
591 	info->si_errno = reason;
592 	info->si_arch = syscall_get_arch();
593 	info->si_syscall = syscall;
594 }
595 
596 /**
597  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
598  * @syscall: syscall number to send to userland
599  * @reason: filter-supplied reason code to send to userland (via si_errno)
600  *
601  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
602  */
603 static void seccomp_send_sigsys(int syscall, int reason)
604 {
605 	struct kernel_siginfo info;
606 	seccomp_init_siginfo(&info, syscall, reason);
607 	force_sig_info(SIGSYS, &info, current);
608 }
609 #endif	/* CONFIG_SECCOMP_FILTER */
610 
611 /* For use with seccomp_actions_logged */
612 #define SECCOMP_LOG_KILL_PROCESS	(1 << 0)
613 #define SECCOMP_LOG_KILL_THREAD		(1 << 1)
614 #define SECCOMP_LOG_TRAP		(1 << 2)
615 #define SECCOMP_LOG_ERRNO		(1 << 3)
616 #define SECCOMP_LOG_TRACE		(1 << 4)
617 #define SECCOMP_LOG_LOG			(1 << 5)
618 #define SECCOMP_LOG_ALLOW		(1 << 6)
619 #define SECCOMP_LOG_USER_NOTIF		(1 << 7)
620 
621 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
622 				    SECCOMP_LOG_KILL_THREAD  |
623 				    SECCOMP_LOG_TRAP  |
624 				    SECCOMP_LOG_ERRNO |
625 				    SECCOMP_LOG_USER_NOTIF |
626 				    SECCOMP_LOG_TRACE |
627 				    SECCOMP_LOG_LOG;
628 
629 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
630 			       bool requested)
631 {
632 	bool log = false;
633 
634 	switch (action) {
635 	case SECCOMP_RET_ALLOW:
636 		break;
637 	case SECCOMP_RET_TRAP:
638 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
639 		break;
640 	case SECCOMP_RET_ERRNO:
641 		log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
642 		break;
643 	case SECCOMP_RET_TRACE:
644 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
645 		break;
646 	case SECCOMP_RET_USER_NOTIF:
647 		log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
648 		break;
649 	case SECCOMP_RET_LOG:
650 		log = seccomp_actions_logged & SECCOMP_LOG_LOG;
651 		break;
652 	case SECCOMP_RET_KILL_THREAD:
653 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
654 		break;
655 	case SECCOMP_RET_KILL_PROCESS:
656 	default:
657 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
658 	}
659 
660 	/*
661 	 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
662 	 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
663 	 * any action from being logged by removing the action name from the
664 	 * seccomp_actions_logged sysctl.
665 	 */
666 	if (!log)
667 		return;
668 
669 	audit_seccomp(syscall, signr, action);
670 }
671 
672 /*
673  * Secure computing mode 1 allows only read/write/exit/sigreturn.
674  * To be fully secure this must be combined with rlimit
675  * to limit the stack allocations too.
676  */
677 static const int mode1_syscalls[] = {
678 	__NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
679 	0, /* null terminated */
680 };
681 
682 static void __secure_computing_strict(int this_syscall)
683 {
684 	const int *syscall_whitelist = mode1_syscalls;
685 #ifdef CONFIG_COMPAT
686 	if (in_compat_syscall())
687 		syscall_whitelist = get_compat_mode1_syscalls();
688 #endif
689 	do {
690 		if (*syscall_whitelist == this_syscall)
691 			return;
692 	} while (*++syscall_whitelist);
693 
694 #ifdef SECCOMP_DEBUG
695 	dump_stack();
696 #endif
697 	seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
698 	do_exit(SIGKILL);
699 }
700 
701 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
702 void secure_computing_strict(int this_syscall)
703 {
704 	int mode = current->seccomp.mode;
705 
706 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
707 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
708 		return;
709 
710 	if (mode == SECCOMP_MODE_DISABLED)
711 		return;
712 	else if (mode == SECCOMP_MODE_STRICT)
713 		__secure_computing_strict(this_syscall);
714 	else
715 		BUG();
716 }
717 #else
718 
719 #ifdef CONFIG_SECCOMP_FILTER
720 static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
721 {
722 	/*
723 	 * Note: overflow is ok here, the id just needs to be unique per
724 	 * filter.
725 	 */
726 	lockdep_assert_held(&filter->notify_lock);
727 	return filter->notif->next_id++;
728 }
729 
730 static void seccomp_do_user_notification(int this_syscall,
731 					 struct seccomp_filter *match,
732 					 const struct seccomp_data *sd)
733 {
734 	int err;
735 	long ret = 0;
736 	struct seccomp_knotif n = {};
737 
738 	mutex_lock(&match->notify_lock);
739 	err = -ENOSYS;
740 	if (!match->notif)
741 		goto out;
742 
743 	n.task = current;
744 	n.state = SECCOMP_NOTIFY_INIT;
745 	n.data = sd;
746 	n.id = seccomp_next_notify_id(match);
747 	init_completion(&n.ready);
748 	list_add(&n.list, &match->notif->notifications);
749 
750 	up(&match->notif->request);
751 	wake_up_poll(&match->notif->wqh, EPOLLIN | EPOLLRDNORM);
752 	mutex_unlock(&match->notify_lock);
753 
754 	/*
755 	 * This is where we wait for a reply from userspace.
756 	 */
757 	err = wait_for_completion_interruptible(&n.ready);
758 	mutex_lock(&match->notify_lock);
759 	if (err == 0) {
760 		ret = n.val;
761 		err = n.error;
762 	}
763 
764 	/*
765 	 * Note that it's possible the listener died in between the time when
766 	 * we were notified of a respons (or a signal) and when we were able to
767 	 * re-acquire the lock, so only delete from the list if the
768 	 * notification actually exists.
769 	 *
770 	 * Also note that this test is only valid because there's no way to
771 	 * *reattach* to a notifier right now. If one is added, we'll need to
772 	 * keep track of the notif itself and make sure they match here.
773 	 */
774 	if (match->notif)
775 		list_del(&n.list);
776 out:
777 	mutex_unlock(&match->notify_lock);
778 	syscall_set_return_value(current, task_pt_regs(current),
779 				 err, ret);
780 }
781 
782 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
783 			    const bool recheck_after_trace)
784 {
785 	u32 filter_ret, action;
786 	struct seccomp_filter *match = NULL;
787 	int data;
788 	struct seccomp_data sd_local;
789 
790 	/*
791 	 * Make sure that any changes to mode from another thread have
792 	 * been seen after TIF_SECCOMP was seen.
793 	 */
794 	rmb();
795 
796 	if (!sd) {
797 		populate_seccomp_data(&sd_local);
798 		sd = &sd_local;
799 	}
800 
801 	filter_ret = seccomp_run_filters(sd, &match);
802 	data = filter_ret & SECCOMP_RET_DATA;
803 	action = filter_ret & SECCOMP_RET_ACTION_FULL;
804 
805 	switch (action) {
806 	case SECCOMP_RET_ERRNO:
807 		/* Set low-order bits as an errno, capped at MAX_ERRNO. */
808 		if (data > MAX_ERRNO)
809 			data = MAX_ERRNO;
810 		syscall_set_return_value(current, task_pt_regs(current),
811 					 -data, 0);
812 		goto skip;
813 
814 	case SECCOMP_RET_TRAP:
815 		/* Show the handler the original registers. */
816 		syscall_rollback(current, task_pt_regs(current));
817 		/* Let the filter pass back 16 bits of data. */
818 		seccomp_send_sigsys(this_syscall, data);
819 		goto skip;
820 
821 	case SECCOMP_RET_TRACE:
822 		/* We've been put in this state by the ptracer already. */
823 		if (recheck_after_trace)
824 			return 0;
825 
826 		/* ENOSYS these calls if there is no tracer attached. */
827 		if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
828 			syscall_set_return_value(current,
829 						 task_pt_regs(current),
830 						 -ENOSYS, 0);
831 			goto skip;
832 		}
833 
834 		/* Allow the BPF to provide the event message */
835 		ptrace_event(PTRACE_EVENT_SECCOMP, data);
836 		/*
837 		 * The delivery of a fatal signal during event
838 		 * notification may silently skip tracer notification,
839 		 * which could leave us with a potentially unmodified
840 		 * syscall that the tracer would have liked to have
841 		 * changed. Since the process is about to die, we just
842 		 * force the syscall to be skipped and let the signal
843 		 * kill the process and correctly handle any tracer exit
844 		 * notifications.
845 		 */
846 		if (fatal_signal_pending(current))
847 			goto skip;
848 		/* Check if the tracer forced the syscall to be skipped. */
849 		this_syscall = syscall_get_nr(current, task_pt_regs(current));
850 		if (this_syscall < 0)
851 			goto skip;
852 
853 		/*
854 		 * Recheck the syscall, since it may have changed. This
855 		 * intentionally uses a NULL struct seccomp_data to force
856 		 * a reload of all registers. This does not goto skip since
857 		 * a skip would have already been reported.
858 		 */
859 		if (__seccomp_filter(this_syscall, NULL, true))
860 			return -1;
861 
862 		return 0;
863 
864 	case SECCOMP_RET_USER_NOTIF:
865 		seccomp_do_user_notification(this_syscall, match, sd);
866 		goto skip;
867 
868 	case SECCOMP_RET_LOG:
869 		seccomp_log(this_syscall, 0, action, true);
870 		return 0;
871 
872 	case SECCOMP_RET_ALLOW:
873 		/*
874 		 * Note that the "match" filter will always be NULL for
875 		 * this action since SECCOMP_RET_ALLOW is the starting
876 		 * state in seccomp_run_filters().
877 		 */
878 		return 0;
879 
880 	case SECCOMP_RET_KILL_THREAD:
881 	case SECCOMP_RET_KILL_PROCESS:
882 	default:
883 		seccomp_log(this_syscall, SIGSYS, action, true);
884 		/* Dump core only if this is the last remaining thread. */
885 		if (action == SECCOMP_RET_KILL_PROCESS ||
886 		    get_nr_threads(current) == 1) {
887 			kernel_siginfo_t info;
888 
889 			/* Show the original registers in the dump. */
890 			syscall_rollback(current, task_pt_regs(current));
891 			/* Trigger a manual coredump since do_exit skips it. */
892 			seccomp_init_siginfo(&info, this_syscall, data);
893 			do_coredump(&info);
894 		}
895 		if (action == SECCOMP_RET_KILL_PROCESS)
896 			do_group_exit(SIGSYS);
897 		else
898 			do_exit(SIGSYS);
899 	}
900 
901 	unreachable();
902 
903 skip:
904 	seccomp_log(this_syscall, 0, action, match ? match->log : false);
905 	return -1;
906 }
907 #else
908 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
909 			    const bool recheck_after_trace)
910 {
911 	BUG();
912 }
913 #endif
914 
915 int __secure_computing(const struct seccomp_data *sd)
916 {
917 	int mode = current->seccomp.mode;
918 	int this_syscall;
919 
920 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
921 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
922 		return 0;
923 
924 	this_syscall = sd ? sd->nr :
925 		syscall_get_nr(current, task_pt_regs(current));
926 
927 	switch (mode) {
928 	case SECCOMP_MODE_STRICT:
929 		__secure_computing_strict(this_syscall);  /* may call do_exit */
930 		return 0;
931 	case SECCOMP_MODE_FILTER:
932 		return __seccomp_filter(this_syscall, sd, false);
933 	default:
934 		BUG();
935 	}
936 }
937 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
938 
939 long prctl_get_seccomp(void)
940 {
941 	return current->seccomp.mode;
942 }
943 
944 /**
945  * seccomp_set_mode_strict: internal function for setting strict seccomp
946  *
947  * Once current->seccomp.mode is non-zero, it may not be changed.
948  *
949  * Returns 0 on success or -EINVAL on failure.
950  */
951 static long seccomp_set_mode_strict(void)
952 {
953 	const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
954 	long ret = -EINVAL;
955 
956 	spin_lock_irq(&current->sighand->siglock);
957 
958 	if (!seccomp_may_assign_mode(seccomp_mode))
959 		goto out;
960 
961 #ifdef TIF_NOTSC
962 	disable_TSC();
963 #endif
964 	seccomp_assign_mode(current, seccomp_mode, 0);
965 	ret = 0;
966 
967 out:
968 	spin_unlock_irq(&current->sighand->siglock);
969 
970 	return ret;
971 }
972 
973 #ifdef CONFIG_SECCOMP_FILTER
974 static int seccomp_notify_release(struct inode *inode, struct file *file)
975 {
976 	struct seccomp_filter *filter = file->private_data;
977 	struct seccomp_knotif *knotif;
978 
979 	if (!filter)
980 		return 0;
981 
982 	mutex_lock(&filter->notify_lock);
983 
984 	/*
985 	 * If this file is being closed because e.g. the task who owned it
986 	 * died, let's wake everyone up who was waiting on us.
987 	 */
988 	list_for_each_entry(knotif, &filter->notif->notifications, list) {
989 		if (knotif->state == SECCOMP_NOTIFY_REPLIED)
990 			continue;
991 
992 		knotif->state = SECCOMP_NOTIFY_REPLIED;
993 		knotif->error = -ENOSYS;
994 		knotif->val = 0;
995 
996 		complete(&knotif->ready);
997 	}
998 
999 	kfree(filter->notif);
1000 	filter->notif = NULL;
1001 	mutex_unlock(&filter->notify_lock);
1002 	__put_seccomp_filter(filter);
1003 	return 0;
1004 }
1005 
1006 static long seccomp_notify_recv(struct seccomp_filter *filter,
1007 				void __user *buf)
1008 {
1009 	struct seccomp_knotif *knotif = NULL, *cur;
1010 	struct seccomp_notif unotif;
1011 	ssize_t ret;
1012 
1013 	memset(&unotif, 0, sizeof(unotif));
1014 
1015 	ret = down_interruptible(&filter->notif->request);
1016 	if (ret < 0)
1017 		return ret;
1018 
1019 	mutex_lock(&filter->notify_lock);
1020 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1021 		if (cur->state == SECCOMP_NOTIFY_INIT) {
1022 			knotif = cur;
1023 			break;
1024 		}
1025 	}
1026 
1027 	/*
1028 	 * If we didn't find a notification, it could be that the task was
1029 	 * interrupted by a fatal signal between the time we were woken and
1030 	 * when we were able to acquire the rw lock.
1031 	 */
1032 	if (!knotif) {
1033 		ret = -ENOENT;
1034 		goto out;
1035 	}
1036 
1037 	unotif.id = knotif->id;
1038 	unotif.pid = task_pid_vnr(knotif->task);
1039 	unotif.data = *(knotif->data);
1040 
1041 	knotif->state = SECCOMP_NOTIFY_SENT;
1042 	wake_up_poll(&filter->notif->wqh, EPOLLOUT | EPOLLWRNORM);
1043 	ret = 0;
1044 out:
1045 	mutex_unlock(&filter->notify_lock);
1046 
1047 	if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1048 		ret = -EFAULT;
1049 
1050 		/*
1051 		 * Userspace screwed up. To make sure that we keep this
1052 		 * notification alive, let's reset it back to INIT. It
1053 		 * may have died when we released the lock, so we need to make
1054 		 * sure it's still around.
1055 		 */
1056 		knotif = NULL;
1057 		mutex_lock(&filter->notify_lock);
1058 		list_for_each_entry(cur, &filter->notif->notifications, list) {
1059 			if (cur->id == unotif.id) {
1060 				knotif = cur;
1061 				break;
1062 			}
1063 		}
1064 
1065 		if (knotif) {
1066 			knotif->state = SECCOMP_NOTIFY_INIT;
1067 			up(&filter->notif->request);
1068 		}
1069 		mutex_unlock(&filter->notify_lock);
1070 	}
1071 
1072 	return ret;
1073 }
1074 
1075 static long seccomp_notify_send(struct seccomp_filter *filter,
1076 				void __user *buf)
1077 {
1078 	struct seccomp_notif_resp resp = {};
1079 	struct seccomp_knotif *knotif = NULL, *cur;
1080 	long ret;
1081 
1082 	if (copy_from_user(&resp, buf, sizeof(resp)))
1083 		return -EFAULT;
1084 
1085 	if (resp.flags)
1086 		return -EINVAL;
1087 
1088 	ret = mutex_lock_interruptible(&filter->notify_lock);
1089 	if (ret < 0)
1090 		return ret;
1091 
1092 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1093 		if (cur->id == resp.id) {
1094 			knotif = cur;
1095 			break;
1096 		}
1097 	}
1098 
1099 	if (!knotif) {
1100 		ret = -ENOENT;
1101 		goto out;
1102 	}
1103 
1104 	/* Allow exactly one reply. */
1105 	if (knotif->state != SECCOMP_NOTIFY_SENT) {
1106 		ret = -EINPROGRESS;
1107 		goto out;
1108 	}
1109 
1110 	ret = 0;
1111 	knotif->state = SECCOMP_NOTIFY_REPLIED;
1112 	knotif->error = resp.error;
1113 	knotif->val = resp.val;
1114 	complete(&knotif->ready);
1115 out:
1116 	mutex_unlock(&filter->notify_lock);
1117 	return ret;
1118 }
1119 
1120 static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1121 				    void __user *buf)
1122 {
1123 	struct seccomp_knotif *knotif = NULL;
1124 	u64 id;
1125 	long ret;
1126 
1127 	if (copy_from_user(&id, buf, sizeof(id)))
1128 		return -EFAULT;
1129 
1130 	ret = mutex_lock_interruptible(&filter->notify_lock);
1131 	if (ret < 0)
1132 		return ret;
1133 
1134 	ret = -ENOENT;
1135 	list_for_each_entry(knotif, &filter->notif->notifications, list) {
1136 		if (knotif->id == id) {
1137 			if (knotif->state == SECCOMP_NOTIFY_SENT)
1138 				ret = 0;
1139 			goto out;
1140 		}
1141 	}
1142 
1143 out:
1144 	mutex_unlock(&filter->notify_lock);
1145 	return ret;
1146 }
1147 
1148 static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1149 				 unsigned long arg)
1150 {
1151 	struct seccomp_filter *filter = file->private_data;
1152 	void __user *buf = (void __user *)arg;
1153 
1154 	switch (cmd) {
1155 	case SECCOMP_IOCTL_NOTIF_RECV:
1156 		return seccomp_notify_recv(filter, buf);
1157 	case SECCOMP_IOCTL_NOTIF_SEND:
1158 		return seccomp_notify_send(filter, buf);
1159 	case SECCOMP_IOCTL_NOTIF_ID_VALID:
1160 		return seccomp_notify_id_valid(filter, buf);
1161 	default:
1162 		return -EINVAL;
1163 	}
1164 }
1165 
1166 static __poll_t seccomp_notify_poll(struct file *file,
1167 				    struct poll_table_struct *poll_tab)
1168 {
1169 	struct seccomp_filter *filter = file->private_data;
1170 	__poll_t ret = 0;
1171 	struct seccomp_knotif *cur;
1172 
1173 	poll_wait(file, &filter->notif->wqh, poll_tab);
1174 
1175 	if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1176 		return EPOLLERR;
1177 
1178 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1179 		if (cur->state == SECCOMP_NOTIFY_INIT)
1180 			ret |= EPOLLIN | EPOLLRDNORM;
1181 		if (cur->state == SECCOMP_NOTIFY_SENT)
1182 			ret |= EPOLLOUT | EPOLLWRNORM;
1183 		if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1184 			break;
1185 	}
1186 
1187 	mutex_unlock(&filter->notify_lock);
1188 
1189 	return ret;
1190 }
1191 
1192 static const struct file_operations seccomp_notify_ops = {
1193 	.poll = seccomp_notify_poll,
1194 	.release = seccomp_notify_release,
1195 	.unlocked_ioctl = seccomp_notify_ioctl,
1196 };
1197 
1198 static struct file *init_listener(struct seccomp_filter *filter)
1199 {
1200 	struct file *ret = ERR_PTR(-EBUSY);
1201 	struct seccomp_filter *cur;
1202 
1203 	for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1204 		if (cur->notif)
1205 			goto out;
1206 	}
1207 
1208 	ret = ERR_PTR(-ENOMEM);
1209 	filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1210 	if (!filter->notif)
1211 		goto out;
1212 
1213 	sema_init(&filter->notif->request, 0);
1214 	filter->notif->next_id = get_random_u64();
1215 	INIT_LIST_HEAD(&filter->notif->notifications);
1216 	init_waitqueue_head(&filter->notif->wqh);
1217 
1218 	ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1219 				 filter, O_RDWR);
1220 	if (IS_ERR(ret))
1221 		goto out_notif;
1222 
1223 	/* The file has a reference to it now */
1224 	__get_seccomp_filter(filter);
1225 
1226 out_notif:
1227 	if (IS_ERR(ret))
1228 		kfree(filter->notif);
1229 out:
1230 	return ret;
1231 }
1232 
1233 /**
1234  * seccomp_set_mode_filter: internal function for setting seccomp filter
1235  * @flags:  flags to change filter behavior
1236  * @filter: struct sock_fprog containing filter
1237  *
1238  * This function may be called repeatedly to install additional filters.
1239  * Every filter successfully installed will be evaluated (in reverse order)
1240  * for each system call the task makes.
1241  *
1242  * Once current->seccomp.mode is non-zero, it may not be changed.
1243  *
1244  * Returns 0 on success or -EINVAL on failure.
1245  */
1246 static long seccomp_set_mode_filter(unsigned int flags,
1247 				    const char __user *filter)
1248 {
1249 	const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1250 	struct seccomp_filter *prepared = NULL;
1251 	long ret = -EINVAL;
1252 	int listener = -1;
1253 	struct file *listener_f = NULL;
1254 
1255 	/* Validate flags. */
1256 	if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1257 		return -EINVAL;
1258 
1259 	/* Prepare the new filter before holding any locks. */
1260 	prepared = seccomp_prepare_user_filter(filter);
1261 	if (IS_ERR(prepared))
1262 		return PTR_ERR(prepared);
1263 
1264 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1265 		listener = get_unused_fd_flags(O_CLOEXEC);
1266 		if (listener < 0) {
1267 			ret = listener;
1268 			goto out_free;
1269 		}
1270 
1271 		listener_f = init_listener(prepared);
1272 		if (IS_ERR(listener_f)) {
1273 			put_unused_fd(listener);
1274 			ret = PTR_ERR(listener_f);
1275 			goto out_free;
1276 		}
1277 	}
1278 
1279 	/*
1280 	 * Make sure we cannot change seccomp or nnp state via TSYNC
1281 	 * while another thread is in the middle of calling exec.
1282 	 */
1283 	if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1284 	    mutex_lock_killable(&current->signal->cred_guard_mutex))
1285 		goto out_put_fd;
1286 
1287 	spin_lock_irq(&current->sighand->siglock);
1288 
1289 	if (!seccomp_may_assign_mode(seccomp_mode))
1290 		goto out;
1291 
1292 	ret = seccomp_attach_filter(flags, prepared);
1293 	if (ret)
1294 		goto out;
1295 	/* Do not free the successfully attached filter. */
1296 	prepared = NULL;
1297 
1298 	seccomp_assign_mode(current, seccomp_mode, flags);
1299 out:
1300 	spin_unlock_irq(&current->sighand->siglock);
1301 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
1302 		mutex_unlock(&current->signal->cred_guard_mutex);
1303 out_put_fd:
1304 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1305 		if (ret < 0) {
1306 			listener_f->private_data = NULL;
1307 			fput(listener_f);
1308 			put_unused_fd(listener);
1309 		} else {
1310 			fd_install(listener, listener_f);
1311 			ret = listener;
1312 		}
1313 	}
1314 out_free:
1315 	seccomp_filter_free(prepared);
1316 	return ret;
1317 }
1318 #else
1319 static inline long seccomp_set_mode_filter(unsigned int flags,
1320 					   const char __user *filter)
1321 {
1322 	return -EINVAL;
1323 }
1324 #endif
1325 
1326 static long seccomp_get_action_avail(const char __user *uaction)
1327 {
1328 	u32 action;
1329 
1330 	if (copy_from_user(&action, uaction, sizeof(action)))
1331 		return -EFAULT;
1332 
1333 	switch (action) {
1334 	case SECCOMP_RET_KILL_PROCESS:
1335 	case SECCOMP_RET_KILL_THREAD:
1336 	case SECCOMP_RET_TRAP:
1337 	case SECCOMP_RET_ERRNO:
1338 	case SECCOMP_RET_USER_NOTIF:
1339 	case SECCOMP_RET_TRACE:
1340 	case SECCOMP_RET_LOG:
1341 	case SECCOMP_RET_ALLOW:
1342 		break;
1343 	default:
1344 		return -EOPNOTSUPP;
1345 	}
1346 
1347 	return 0;
1348 }
1349 
1350 static long seccomp_get_notif_sizes(void __user *usizes)
1351 {
1352 	struct seccomp_notif_sizes sizes = {
1353 		.seccomp_notif = sizeof(struct seccomp_notif),
1354 		.seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
1355 		.seccomp_data = sizeof(struct seccomp_data),
1356 	};
1357 
1358 	if (copy_to_user(usizes, &sizes, sizeof(sizes)))
1359 		return -EFAULT;
1360 
1361 	return 0;
1362 }
1363 
1364 /* Common entry point for both prctl and syscall. */
1365 static long do_seccomp(unsigned int op, unsigned int flags,
1366 		       void __user *uargs)
1367 {
1368 	switch (op) {
1369 	case SECCOMP_SET_MODE_STRICT:
1370 		if (flags != 0 || uargs != NULL)
1371 			return -EINVAL;
1372 		return seccomp_set_mode_strict();
1373 	case SECCOMP_SET_MODE_FILTER:
1374 		return seccomp_set_mode_filter(flags, uargs);
1375 	case SECCOMP_GET_ACTION_AVAIL:
1376 		if (flags != 0)
1377 			return -EINVAL;
1378 
1379 		return seccomp_get_action_avail(uargs);
1380 	case SECCOMP_GET_NOTIF_SIZES:
1381 		if (flags != 0)
1382 			return -EINVAL;
1383 
1384 		return seccomp_get_notif_sizes(uargs);
1385 	default:
1386 		return -EINVAL;
1387 	}
1388 }
1389 
1390 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
1391 			 void __user *, uargs)
1392 {
1393 	return do_seccomp(op, flags, uargs);
1394 }
1395 
1396 /**
1397  * prctl_set_seccomp: configures current->seccomp.mode
1398  * @seccomp_mode: requested mode to use
1399  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1400  *
1401  * Returns 0 on success or -EINVAL on failure.
1402  */
1403 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
1404 {
1405 	unsigned int op;
1406 	void __user *uargs;
1407 
1408 	switch (seccomp_mode) {
1409 	case SECCOMP_MODE_STRICT:
1410 		op = SECCOMP_SET_MODE_STRICT;
1411 		/*
1412 		 * Setting strict mode through prctl always ignored filter,
1413 		 * so make sure it is always NULL here to pass the internal
1414 		 * check in do_seccomp().
1415 		 */
1416 		uargs = NULL;
1417 		break;
1418 	case SECCOMP_MODE_FILTER:
1419 		op = SECCOMP_SET_MODE_FILTER;
1420 		uargs = filter;
1421 		break;
1422 	default:
1423 		return -EINVAL;
1424 	}
1425 
1426 	/* prctl interface doesn't have flags, so they are always zero. */
1427 	return do_seccomp(op, 0, uargs);
1428 }
1429 
1430 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
1431 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
1432 					     unsigned long filter_off)
1433 {
1434 	struct seccomp_filter *orig, *filter;
1435 	unsigned long count;
1436 
1437 	/*
1438 	 * Note: this is only correct because the caller should be the (ptrace)
1439 	 * tracer of the task, otherwise lock_task_sighand is needed.
1440 	 */
1441 	spin_lock_irq(&task->sighand->siglock);
1442 
1443 	if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
1444 		spin_unlock_irq(&task->sighand->siglock);
1445 		return ERR_PTR(-EINVAL);
1446 	}
1447 
1448 	orig = task->seccomp.filter;
1449 	__get_seccomp_filter(orig);
1450 	spin_unlock_irq(&task->sighand->siglock);
1451 
1452 	count = 0;
1453 	for (filter = orig; filter; filter = filter->prev)
1454 		count++;
1455 
1456 	if (filter_off >= count) {
1457 		filter = ERR_PTR(-ENOENT);
1458 		goto out;
1459 	}
1460 
1461 	count -= filter_off;
1462 	for (filter = orig; filter && count > 1; filter = filter->prev)
1463 		count--;
1464 
1465 	if (WARN_ON(count != 1 || !filter)) {
1466 		filter = ERR_PTR(-ENOENT);
1467 		goto out;
1468 	}
1469 
1470 	__get_seccomp_filter(filter);
1471 
1472 out:
1473 	__put_seccomp_filter(orig);
1474 	return filter;
1475 }
1476 
1477 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1478 			void __user *data)
1479 {
1480 	struct seccomp_filter *filter;
1481 	struct sock_fprog_kern *fprog;
1482 	long ret;
1483 
1484 	if (!capable(CAP_SYS_ADMIN) ||
1485 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1486 		return -EACCES;
1487 	}
1488 
1489 	filter = get_nth_filter(task, filter_off);
1490 	if (IS_ERR(filter))
1491 		return PTR_ERR(filter);
1492 
1493 	fprog = filter->prog->orig_prog;
1494 	if (!fprog) {
1495 		/* This must be a new non-cBPF filter, since we save
1496 		 * every cBPF filter's orig_prog above when
1497 		 * CONFIG_CHECKPOINT_RESTORE is enabled.
1498 		 */
1499 		ret = -EMEDIUMTYPE;
1500 		goto out;
1501 	}
1502 
1503 	ret = fprog->len;
1504 	if (!data)
1505 		goto out;
1506 
1507 	if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1508 		ret = -EFAULT;
1509 
1510 out:
1511 	__put_seccomp_filter(filter);
1512 	return ret;
1513 }
1514 
1515 long seccomp_get_metadata(struct task_struct *task,
1516 			  unsigned long size, void __user *data)
1517 {
1518 	long ret;
1519 	struct seccomp_filter *filter;
1520 	struct seccomp_metadata kmd = {};
1521 
1522 	if (!capable(CAP_SYS_ADMIN) ||
1523 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1524 		return -EACCES;
1525 	}
1526 
1527 	size = min_t(unsigned long, size, sizeof(kmd));
1528 
1529 	if (size < sizeof(kmd.filter_off))
1530 		return -EINVAL;
1531 
1532 	if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
1533 		return -EFAULT;
1534 
1535 	filter = get_nth_filter(task, kmd.filter_off);
1536 	if (IS_ERR(filter))
1537 		return PTR_ERR(filter);
1538 
1539 	if (filter->log)
1540 		kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1541 
1542 	ret = size;
1543 	if (copy_to_user(data, &kmd, size))
1544 		ret = -EFAULT;
1545 
1546 	__put_seccomp_filter(filter);
1547 	return ret;
1548 }
1549 #endif
1550 
1551 #ifdef CONFIG_SYSCTL
1552 
1553 /* Human readable action names for friendly sysctl interaction */
1554 #define SECCOMP_RET_KILL_PROCESS_NAME	"kill_process"
1555 #define SECCOMP_RET_KILL_THREAD_NAME	"kill_thread"
1556 #define SECCOMP_RET_TRAP_NAME		"trap"
1557 #define SECCOMP_RET_ERRNO_NAME		"errno"
1558 #define SECCOMP_RET_USER_NOTIF_NAME	"user_notif"
1559 #define SECCOMP_RET_TRACE_NAME		"trace"
1560 #define SECCOMP_RET_LOG_NAME		"log"
1561 #define SECCOMP_RET_ALLOW_NAME		"allow"
1562 
1563 static const char seccomp_actions_avail[] =
1564 				SECCOMP_RET_KILL_PROCESS_NAME	" "
1565 				SECCOMP_RET_KILL_THREAD_NAME	" "
1566 				SECCOMP_RET_TRAP_NAME		" "
1567 				SECCOMP_RET_ERRNO_NAME		" "
1568 				SECCOMP_RET_USER_NOTIF_NAME     " "
1569 				SECCOMP_RET_TRACE_NAME		" "
1570 				SECCOMP_RET_LOG_NAME		" "
1571 				SECCOMP_RET_ALLOW_NAME;
1572 
1573 struct seccomp_log_name {
1574 	u32		log;
1575 	const char	*name;
1576 };
1577 
1578 static const struct seccomp_log_name seccomp_log_names[] = {
1579 	{ SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1580 	{ SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1581 	{ SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1582 	{ SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1583 	{ SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
1584 	{ SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1585 	{ SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1586 	{ SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1587 	{ }
1588 };
1589 
1590 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1591 					      u32 actions_logged,
1592 					      const char *sep)
1593 {
1594 	const struct seccomp_log_name *cur;
1595 	bool append_sep = false;
1596 
1597 	for (cur = seccomp_log_names; cur->name && size; cur++) {
1598 		ssize_t ret;
1599 
1600 		if (!(actions_logged & cur->log))
1601 			continue;
1602 
1603 		if (append_sep) {
1604 			ret = strscpy(names, sep, size);
1605 			if (ret < 0)
1606 				return false;
1607 
1608 			names += ret;
1609 			size -= ret;
1610 		} else
1611 			append_sep = true;
1612 
1613 		ret = strscpy(names, cur->name, size);
1614 		if (ret < 0)
1615 			return false;
1616 
1617 		names += ret;
1618 		size -= ret;
1619 	}
1620 
1621 	return true;
1622 }
1623 
1624 static bool seccomp_action_logged_from_name(u32 *action_logged,
1625 					    const char *name)
1626 {
1627 	const struct seccomp_log_name *cur;
1628 
1629 	for (cur = seccomp_log_names; cur->name; cur++) {
1630 		if (!strcmp(cur->name, name)) {
1631 			*action_logged = cur->log;
1632 			return true;
1633 		}
1634 	}
1635 
1636 	return false;
1637 }
1638 
1639 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1640 {
1641 	char *name;
1642 
1643 	*actions_logged = 0;
1644 	while ((name = strsep(&names, " ")) && *name) {
1645 		u32 action_logged = 0;
1646 
1647 		if (!seccomp_action_logged_from_name(&action_logged, name))
1648 			return false;
1649 
1650 		*actions_logged |= action_logged;
1651 	}
1652 
1653 	return true;
1654 }
1655 
1656 static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1657 			       size_t *lenp, loff_t *ppos)
1658 {
1659 	char names[sizeof(seccomp_actions_avail)];
1660 	struct ctl_table table;
1661 
1662 	memset(names, 0, sizeof(names));
1663 
1664 	if (!seccomp_names_from_actions_logged(names, sizeof(names),
1665 					       seccomp_actions_logged, " "))
1666 		return -EINVAL;
1667 
1668 	table = *ro_table;
1669 	table.data = names;
1670 	table.maxlen = sizeof(names);
1671 	return proc_dostring(&table, 0, buffer, lenp, ppos);
1672 }
1673 
1674 static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1675 				size_t *lenp, loff_t *ppos, u32 *actions_logged)
1676 {
1677 	char names[sizeof(seccomp_actions_avail)];
1678 	struct ctl_table table;
1679 	int ret;
1680 
1681 	if (!capable(CAP_SYS_ADMIN))
1682 		return -EPERM;
1683 
1684 	memset(names, 0, sizeof(names));
1685 
1686 	table = *ro_table;
1687 	table.data = names;
1688 	table.maxlen = sizeof(names);
1689 	ret = proc_dostring(&table, 1, buffer, lenp, ppos);
1690 	if (ret)
1691 		return ret;
1692 
1693 	if (!seccomp_actions_logged_from_names(actions_logged, table.data))
1694 		return -EINVAL;
1695 
1696 	if (*actions_logged & SECCOMP_LOG_ALLOW)
1697 		return -EINVAL;
1698 
1699 	seccomp_actions_logged = *actions_logged;
1700 	return 0;
1701 }
1702 
1703 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
1704 				 int ret)
1705 {
1706 	char names[sizeof(seccomp_actions_avail)];
1707 	char old_names[sizeof(seccomp_actions_avail)];
1708 	const char *new = names;
1709 	const char *old = old_names;
1710 
1711 	if (!audit_enabled)
1712 		return;
1713 
1714 	memset(names, 0, sizeof(names));
1715 	memset(old_names, 0, sizeof(old_names));
1716 
1717 	if (ret)
1718 		new = "?";
1719 	else if (!actions_logged)
1720 		new = "(none)";
1721 	else if (!seccomp_names_from_actions_logged(names, sizeof(names),
1722 						    actions_logged, ","))
1723 		new = "?";
1724 
1725 	if (!old_actions_logged)
1726 		old = "(none)";
1727 	else if (!seccomp_names_from_actions_logged(old_names,
1728 						    sizeof(old_names),
1729 						    old_actions_logged, ","))
1730 		old = "?";
1731 
1732 	return audit_seccomp_actions_logged(new, old, !ret);
1733 }
1734 
1735 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1736 					  void __user *buffer, size_t *lenp,
1737 					  loff_t *ppos)
1738 {
1739 	int ret;
1740 
1741 	if (write) {
1742 		u32 actions_logged = 0;
1743 		u32 old_actions_logged = seccomp_actions_logged;
1744 
1745 		ret = write_actions_logged(ro_table, buffer, lenp, ppos,
1746 					   &actions_logged);
1747 		audit_actions_logged(actions_logged, old_actions_logged, ret);
1748 	} else
1749 		ret = read_actions_logged(ro_table, buffer, lenp, ppos);
1750 
1751 	return ret;
1752 }
1753 
1754 static struct ctl_path seccomp_sysctl_path[] = {
1755 	{ .procname = "kernel", },
1756 	{ .procname = "seccomp", },
1757 	{ }
1758 };
1759 
1760 static struct ctl_table seccomp_sysctl_table[] = {
1761 	{
1762 		.procname	= "actions_avail",
1763 		.data		= (void *) &seccomp_actions_avail,
1764 		.maxlen		= sizeof(seccomp_actions_avail),
1765 		.mode		= 0444,
1766 		.proc_handler	= proc_dostring,
1767 	},
1768 	{
1769 		.procname	= "actions_logged",
1770 		.mode		= 0644,
1771 		.proc_handler	= seccomp_actions_logged_handler,
1772 	},
1773 	{ }
1774 };
1775 
1776 static int __init seccomp_sysctl_init(void)
1777 {
1778 	struct ctl_table_header *hdr;
1779 
1780 	hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1781 	if (!hdr)
1782 		pr_warn("seccomp: sysctl registration failed\n");
1783 	else
1784 		kmemleak_not_leak(hdr);
1785 
1786 	return 0;
1787 }
1788 
1789 device_initcall(seccomp_sysctl_init)
1790 
1791 #endif /* CONFIG_SYSCTL */
1792