xref: /linux/kernel/seccomp.c (revision a776c270a0b2fad6715cb714187e4290cadb9237)
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 	u32 flags;
79 
80 	/* Signals when this has entered SECCOMP_NOTIFY_REPLIED */
81 	struct completion ready;
82 
83 	struct list_head list;
84 };
85 
86 /**
87  * struct notification - container for seccomp userspace notifications. Since
88  * most seccomp filters will not have notification listeners attached and this
89  * structure is fairly large, we store the notification-specific stuff in a
90  * separate structure.
91  *
92  * @request: A semaphore that users of this notification can wait on for
93  *           changes. Actual reads and writes are still controlled with
94  *           filter->notify_lock.
95  * @next_id: The id of the next request.
96  * @notifications: A list of struct seccomp_knotif elements.
97  * @wqh: A wait queue for poll.
98  */
99 struct notification {
100 	struct semaphore request;
101 	u64 next_id;
102 	struct list_head notifications;
103 	wait_queue_head_t wqh;
104 };
105 
106 /**
107  * struct seccomp_filter - container for seccomp BPF programs
108  *
109  * @usage: reference count to manage the object lifetime.
110  *         get/put helpers should be used when accessing an instance
111  *         outside of a lifetime-guarded section.  In general, this
112  *         is only needed for handling filters shared across tasks.
113  * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
114  * @prev: points to a previously installed, or inherited, filter
115  * @prog: the BPF program to evaluate
116  * @notif: the struct that holds all notification related information
117  * @notify_lock: A lock for all notification-related accesses.
118  *
119  * seccomp_filter objects are organized in a tree linked via the @prev
120  * pointer.  For any task, it appears to be a singly-linked list starting
121  * with current->seccomp.filter, the most recently attached or inherited filter.
122  * However, multiple filters may share a @prev node, by way of fork(), which
123  * results in a unidirectional tree existing in memory.  This is similar to
124  * how namespaces work.
125  *
126  * seccomp_filter objects should never be modified after being attached
127  * to a task_struct (other than @usage).
128  */
129 struct seccomp_filter {
130 	refcount_t usage;
131 	bool log;
132 	struct seccomp_filter *prev;
133 	struct bpf_prog *prog;
134 	struct notification *notif;
135 	struct mutex notify_lock;
136 };
137 
138 /* Limit any path through the tree to 256KB worth of instructions. */
139 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
140 
141 /*
142  * Endianness is explicitly ignored and left for BPF program authors to manage
143  * as per the specific architecture.
144  */
145 static void populate_seccomp_data(struct seccomp_data *sd)
146 {
147 	struct task_struct *task = current;
148 	struct pt_regs *regs = task_pt_regs(task);
149 	unsigned long args[6];
150 
151 	sd->nr = syscall_get_nr(task, regs);
152 	sd->arch = syscall_get_arch(task);
153 	syscall_get_arguments(task, regs, args);
154 	sd->args[0] = args[0];
155 	sd->args[1] = args[1];
156 	sd->args[2] = args[2];
157 	sd->args[3] = args[3];
158 	sd->args[4] = args[4];
159 	sd->args[5] = args[5];
160 	sd->instruction_pointer = KSTK_EIP(task);
161 }
162 
163 /**
164  *	seccomp_check_filter - verify seccomp filter code
165  *	@filter: filter to verify
166  *	@flen: length of filter
167  *
168  * Takes a previously checked filter (by bpf_check_classic) and
169  * redirects all filter code that loads struct sk_buff data
170  * and related data through seccomp_bpf_load.  It also
171  * enforces length and alignment checking of those loads.
172  *
173  * Returns 0 if the rule set is legal or -EINVAL if not.
174  */
175 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
176 {
177 	int pc;
178 	for (pc = 0; pc < flen; pc++) {
179 		struct sock_filter *ftest = &filter[pc];
180 		u16 code = ftest->code;
181 		u32 k = ftest->k;
182 
183 		switch (code) {
184 		case BPF_LD | BPF_W | BPF_ABS:
185 			ftest->code = BPF_LDX | BPF_W | BPF_ABS;
186 			/* 32-bit aligned and not out of bounds. */
187 			if (k >= sizeof(struct seccomp_data) || k & 3)
188 				return -EINVAL;
189 			continue;
190 		case BPF_LD | BPF_W | BPF_LEN:
191 			ftest->code = BPF_LD | BPF_IMM;
192 			ftest->k = sizeof(struct seccomp_data);
193 			continue;
194 		case BPF_LDX | BPF_W | BPF_LEN:
195 			ftest->code = BPF_LDX | BPF_IMM;
196 			ftest->k = sizeof(struct seccomp_data);
197 			continue;
198 		/* Explicitly include allowed calls. */
199 		case BPF_RET | BPF_K:
200 		case BPF_RET | BPF_A:
201 		case BPF_ALU | BPF_ADD | BPF_K:
202 		case BPF_ALU | BPF_ADD | BPF_X:
203 		case BPF_ALU | BPF_SUB | BPF_K:
204 		case BPF_ALU | BPF_SUB | BPF_X:
205 		case BPF_ALU | BPF_MUL | BPF_K:
206 		case BPF_ALU | BPF_MUL | BPF_X:
207 		case BPF_ALU | BPF_DIV | BPF_K:
208 		case BPF_ALU | BPF_DIV | BPF_X:
209 		case BPF_ALU | BPF_AND | BPF_K:
210 		case BPF_ALU | BPF_AND | BPF_X:
211 		case BPF_ALU | BPF_OR | BPF_K:
212 		case BPF_ALU | BPF_OR | BPF_X:
213 		case BPF_ALU | BPF_XOR | BPF_K:
214 		case BPF_ALU | BPF_XOR | BPF_X:
215 		case BPF_ALU | BPF_LSH | BPF_K:
216 		case BPF_ALU | BPF_LSH | BPF_X:
217 		case BPF_ALU | BPF_RSH | BPF_K:
218 		case BPF_ALU | BPF_RSH | BPF_X:
219 		case BPF_ALU | BPF_NEG:
220 		case BPF_LD | BPF_IMM:
221 		case BPF_LDX | BPF_IMM:
222 		case BPF_MISC | BPF_TAX:
223 		case BPF_MISC | BPF_TXA:
224 		case BPF_LD | BPF_MEM:
225 		case BPF_LDX | BPF_MEM:
226 		case BPF_ST:
227 		case BPF_STX:
228 		case BPF_JMP | BPF_JA:
229 		case BPF_JMP | BPF_JEQ | BPF_K:
230 		case BPF_JMP | BPF_JEQ | BPF_X:
231 		case BPF_JMP | BPF_JGE | BPF_K:
232 		case BPF_JMP | BPF_JGE | BPF_X:
233 		case BPF_JMP | BPF_JGT | BPF_K:
234 		case BPF_JMP | BPF_JGT | BPF_X:
235 		case BPF_JMP | BPF_JSET | BPF_K:
236 		case BPF_JMP | BPF_JSET | BPF_X:
237 			continue;
238 		default:
239 			return -EINVAL;
240 		}
241 	}
242 	return 0;
243 }
244 
245 /**
246  * seccomp_run_filters - evaluates all seccomp filters against @sd
247  * @sd: optional seccomp data to be passed to filters
248  * @match: stores struct seccomp_filter that resulted in the return value,
249  *         unless filter returned SECCOMP_RET_ALLOW, in which case it will
250  *         be unchanged.
251  *
252  * Returns valid seccomp BPF response codes.
253  */
254 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
255 static u32 seccomp_run_filters(const struct seccomp_data *sd,
256 			       struct seccomp_filter **match)
257 {
258 	u32 ret = SECCOMP_RET_ALLOW;
259 	/* Make sure cross-thread synced filter points somewhere sane. */
260 	struct seccomp_filter *f =
261 			READ_ONCE(current->seccomp.filter);
262 
263 	/* Ensure unexpected behavior doesn't result in failing open. */
264 	if (WARN_ON(f == NULL))
265 		return SECCOMP_RET_KILL_PROCESS;
266 
267 	/*
268 	 * All filters in the list are evaluated and the lowest BPF return
269 	 * value always takes priority (ignoring the DATA).
270 	 */
271 	preempt_disable();
272 	for (; f; f = f->prev) {
273 		u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
274 
275 		if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
276 			ret = cur_ret;
277 			*match = f;
278 		}
279 	}
280 	preempt_enable();
281 	return ret;
282 }
283 #endif /* CONFIG_SECCOMP_FILTER */
284 
285 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
286 {
287 	assert_spin_locked(&current->sighand->siglock);
288 
289 	if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
290 		return false;
291 
292 	return true;
293 }
294 
295 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
296 
297 static inline void seccomp_assign_mode(struct task_struct *task,
298 				       unsigned long seccomp_mode,
299 				       unsigned long flags)
300 {
301 	assert_spin_locked(&task->sighand->siglock);
302 
303 	task->seccomp.mode = seccomp_mode;
304 	/*
305 	 * Make sure TIF_SECCOMP cannot be set before the mode (and
306 	 * filter) is set.
307 	 */
308 	smp_mb__before_atomic();
309 	/* Assume default seccomp processes want spec flaw mitigation. */
310 	if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
311 		arch_seccomp_spec_mitigate(task);
312 	set_tsk_thread_flag(task, TIF_SECCOMP);
313 }
314 
315 #ifdef CONFIG_SECCOMP_FILTER
316 /* Returns 1 if the parent is an ancestor of the child. */
317 static int is_ancestor(struct seccomp_filter *parent,
318 		       struct seccomp_filter *child)
319 {
320 	/* NULL is the root ancestor. */
321 	if (parent == NULL)
322 		return 1;
323 	for (; child; child = child->prev)
324 		if (child == parent)
325 			return 1;
326 	return 0;
327 }
328 
329 /**
330  * seccomp_can_sync_threads: checks if all threads can be synchronized
331  *
332  * Expects sighand and cred_guard_mutex locks to be held.
333  *
334  * Returns 0 on success, -ve on error, or the pid of a thread which was
335  * either not in the correct seccomp mode or did not have an ancestral
336  * seccomp filter.
337  */
338 static inline pid_t seccomp_can_sync_threads(void)
339 {
340 	struct task_struct *thread, *caller;
341 
342 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
343 	assert_spin_locked(&current->sighand->siglock);
344 
345 	/* Validate all threads being eligible for synchronization. */
346 	caller = current;
347 	for_each_thread(caller, thread) {
348 		pid_t failed;
349 
350 		/* Skip current, since it is initiating the sync. */
351 		if (thread == caller)
352 			continue;
353 
354 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
355 		    (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
356 		     is_ancestor(thread->seccomp.filter,
357 				 caller->seccomp.filter)))
358 			continue;
359 
360 		/* Return the first thread that cannot be synchronized. */
361 		failed = task_pid_vnr(thread);
362 		/* If the pid cannot be resolved, then return -ESRCH */
363 		if (WARN_ON(failed == 0))
364 			failed = -ESRCH;
365 		return failed;
366 	}
367 
368 	return 0;
369 }
370 
371 /**
372  * seccomp_sync_threads: sets all threads to use current's filter
373  *
374  * Expects sighand and cred_guard_mutex locks to be held, and for
375  * seccomp_can_sync_threads() to have returned success already
376  * without dropping the locks.
377  *
378  */
379 static inline void seccomp_sync_threads(unsigned long flags)
380 {
381 	struct task_struct *thread, *caller;
382 
383 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
384 	assert_spin_locked(&current->sighand->siglock);
385 
386 	/* Synchronize all threads. */
387 	caller = current;
388 	for_each_thread(caller, thread) {
389 		/* Skip current, since it needs no changes. */
390 		if (thread == caller)
391 			continue;
392 
393 		/* Get a task reference for the new leaf node. */
394 		get_seccomp_filter(caller);
395 		/*
396 		 * Drop the task reference to the shared ancestor since
397 		 * current's path will hold a reference.  (This also
398 		 * allows a put before the assignment.)
399 		 */
400 		put_seccomp_filter(thread);
401 		smp_store_release(&thread->seccomp.filter,
402 				  caller->seccomp.filter);
403 
404 		/*
405 		 * Don't let an unprivileged task work around
406 		 * the no_new_privs restriction by creating
407 		 * a thread that sets it up, enters seccomp,
408 		 * then dies.
409 		 */
410 		if (task_no_new_privs(caller))
411 			task_set_no_new_privs(thread);
412 
413 		/*
414 		 * Opt the other thread into seccomp if needed.
415 		 * As threads are considered to be trust-realm
416 		 * equivalent (see ptrace_may_access), it is safe to
417 		 * allow one thread to transition the other.
418 		 */
419 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
420 			seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
421 					    flags);
422 	}
423 }
424 
425 /**
426  * seccomp_prepare_filter: Prepares a seccomp filter for use.
427  * @fprog: BPF program to install
428  *
429  * Returns filter on success or an ERR_PTR on failure.
430  */
431 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
432 {
433 	struct seccomp_filter *sfilter;
434 	int ret;
435 	const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
436 
437 	if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
438 		return ERR_PTR(-EINVAL);
439 
440 	BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
441 
442 	/*
443 	 * Installing a seccomp filter requires that the task has
444 	 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
445 	 * This avoids scenarios where unprivileged tasks can affect the
446 	 * behavior of privileged children.
447 	 */
448 	if (!task_no_new_privs(current) &&
449 	    security_capable(current_cred(), current_user_ns(),
450 				     CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) != 0)
451 		return ERR_PTR(-EACCES);
452 
453 	/* Allocate a new seccomp_filter */
454 	sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
455 	if (!sfilter)
456 		return ERR_PTR(-ENOMEM);
457 
458 	mutex_init(&sfilter->notify_lock);
459 	ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
460 					seccomp_check_filter, save_orig);
461 	if (ret < 0) {
462 		kfree(sfilter);
463 		return ERR_PTR(ret);
464 	}
465 
466 	refcount_set(&sfilter->usage, 1);
467 
468 	return sfilter;
469 }
470 
471 /**
472  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
473  * @user_filter: pointer to the user data containing a sock_fprog.
474  *
475  * Returns 0 on success and non-zero otherwise.
476  */
477 static struct seccomp_filter *
478 seccomp_prepare_user_filter(const char __user *user_filter)
479 {
480 	struct sock_fprog fprog;
481 	struct seccomp_filter *filter = ERR_PTR(-EFAULT);
482 
483 #ifdef CONFIG_COMPAT
484 	if (in_compat_syscall()) {
485 		struct compat_sock_fprog fprog32;
486 		if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
487 			goto out;
488 		fprog.len = fprog32.len;
489 		fprog.filter = compat_ptr(fprog32.filter);
490 	} else /* falls through to the if below. */
491 #endif
492 	if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
493 		goto out;
494 	filter = seccomp_prepare_filter(&fprog);
495 out:
496 	return filter;
497 }
498 
499 /**
500  * seccomp_attach_filter: validate and attach filter
501  * @flags:  flags to change filter behavior
502  * @filter: seccomp filter to add to the current process
503  *
504  * Caller must be holding current->sighand->siglock lock.
505  *
506  * Returns 0 on success, -ve on error, or
507  *   - in TSYNC mode: the pid of a thread which was either not in the correct
508  *     seccomp mode or did not have an ancestral seccomp filter
509  *   - in NEW_LISTENER mode: the fd of the new listener
510  */
511 static long seccomp_attach_filter(unsigned int flags,
512 				  struct seccomp_filter *filter)
513 {
514 	unsigned long total_insns;
515 	struct seccomp_filter *walker;
516 
517 	assert_spin_locked(&current->sighand->siglock);
518 
519 	/* Validate resulting filter length. */
520 	total_insns = filter->prog->len;
521 	for (walker = current->seccomp.filter; walker; walker = walker->prev)
522 		total_insns += walker->prog->len + 4;  /* 4 instr penalty */
523 	if (total_insns > MAX_INSNS_PER_PATH)
524 		return -ENOMEM;
525 
526 	/* If thread sync has been requested, check that it is possible. */
527 	if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
528 		int ret;
529 
530 		ret = seccomp_can_sync_threads();
531 		if (ret) {
532 			if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
533 				return -ESRCH;
534 			else
535 				return ret;
536 		}
537 	}
538 
539 	/* Set log flag, if present. */
540 	if (flags & SECCOMP_FILTER_FLAG_LOG)
541 		filter->log = true;
542 
543 	/*
544 	 * If there is an existing filter, make it the prev and don't drop its
545 	 * task reference.
546 	 */
547 	filter->prev = current->seccomp.filter;
548 	current->seccomp.filter = filter;
549 
550 	/* Now that the new filter is in place, synchronize to all threads. */
551 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
552 		seccomp_sync_threads(flags);
553 
554 	return 0;
555 }
556 
557 static void __get_seccomp_filter(struct seccomp_filter *filter)
558 {
559 	refcount_inc(&filter->usage);
560 }
561 
562 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
563 void get_seccomp_filter(struct task_struct *tsk)
564 {
565 	struct seccomp_filter *orig = tsk->seccomp.filter;
566 	if (!orig)
567 		return;
568 	__get_seccomp_filter(orig);
569 }
570 
571 static inline void seccomp_filter_free(struct seccomp_filter *filter)
572 {
573 	if (filter) {
574 		bpf_prog_destroy(filter->prog);
575 		kfree(filter);
576 	}
577 }
578 
579 static void __put_seccomp_filter(struct seccomp_filter *orig)
580 {
581 	/* Clean up single-reference branches iteratively. */
582 	while (orig && refcount_dec_and_test(&orig->usage)) {
583 		struct seccomp_filter *freeme = orig;
584 		orig = orig->prev;
585 		seccomp_filter_free(freeme);
586 	}
587 }
588 
589 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
590 void put_seccomp_filter(struct task_struct *tsk)
591 {
592 	__put_seccomp_filter(tsk->seccomp.filter);
593 }
594 
595 static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason)
596 {
597 	clear_siginfo(info);
598 	info->si_signo = SIGSYS;
599 	info->si_code = SYS_SECCOMP;
600 	info->si_call_addr = (void __user *)KSTK_EIP(current);
601 	info->si_errno = reason;
602 	info->si_arch = syscall_get_arch(current);
603 	info->si_syscall = syscall;
604 }
605 
606 /**
607  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
608  * @syscall: syscall number to send to userland
609  * @reason: filter-supplied reason code to send to userland (via si_errno)
610  *
611  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
612  */
613 static void seccomp_send_sigsys(int syscall, int reason)
614 {
615 	struct kernel_siginfo info;
616 	seccomp_init_siginfo(&info, syscall, reason);
617 	force_sig_info(&info);
618 }
619 #endif	/* CONFIG_SECCOMP_FILTER */
620 
621 /* For use with seccomp_actions_logged */
622 #define SECCOMP_LOG_KILL_PROCESS	(1 << 0)
623 #define SECCOMP_LOG_KILL_THREAD		(1 << 1)
624 #define SECCOMP_LOG_TRAP		(1 << 2)
625 #define SECCOMP_LOG_ERRNO		(1 << 3)
626 #define SECCOMP_LOG_TRACE		(1 << 4)
627 #define SECCOMP_LOG_LOG			(1 << 5)
628 #define SECCOMP_LOG_ALLOW		(1 << 6)
629 #define SECCOMP_LOG_USER_NOTIF		(1 << 7)
630 
631 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
632 				    SECCOMP_LOG_KILL_THREAD  |
633 				    SECCOMP_LOG_TRAP  |
634 				    SECCOMP_LOG_ERRNO |
635 				    SECCOMP_LOG_USER_NOTIF |
636 				    SECCOMP_LOG_TRACE |
637 				    SECCOMP_LOG_LOG;
638 
639 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
640 			       bool requested)
641 {
642 	bool log = false;
643 
644 	switch (action) {
645 	case SECCOMP_RET_ALLOW:
646 		break;
647 	case SECCOMP_RET_TRAP:
648 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
649 		break;
650 	case SECCOMP_RET_ERRNO:
651 		log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
652 		break;
653 	case SECCOMP_RET_TRACE:
654 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
655 		break;
656 	case SECCOMP_RET_USER_NOTIF:
657 		log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
658 		break;
659 	case SECCOMP_RET_LOG:
660 		log = seccomp_actions_logged & SECCOMP_LOG_LOG;
661 		break;
662 	case SECCOMP_RET_KILL_THREAD:
663 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
664 		break;
665 	case SECCOMP_RET_KILL_PROCESS:
666 	default:
667 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
668 	}
669 
670 	/*
671 	 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
672 	 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
673 	 * any action from being logged by removing the action name from the
674 	 * seccomp_actions_logged sysctl.
675 	 */
676 	if (!log)
677 		return;
678 
679 	audit_seccomp(syscall, signr, action);
680 }
681 
682 /*
683  * Secure computing mode 1 allows only read/write/exit/sigreturn.
684  * To be fully secure this must be combined with rlimit
685  * to limit the stack allocations too.
686  */
687 static const int mode1_syscalls[] = {
688 	__NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
689 	0, /* null terminated */
690 };
691 
692 static void __secure_computing_strict(int this_syscall)
693 {
694 	const int *syscall_whitelist = mode1_syscalls;
695 #ifdef CONFIG_COMPAT
696 	if (in_compat_syscall())
697 		syscall_whitelist = get_compat_mode1_syscalls();
698 #endif
699 	do {
700 		if (*syscall_whitelist == this_syscall)
701 			return;
702 	} while (*++syscall_whitelist);
703 
704 #ifdef SECCOMP_DEBUG
705 	dump_stack();
706 #endif
707 	seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
708 	do_exit(SIGKILL);
709 }
710 
711 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
712 void secure_computing_strict(int this_syscall)
713 {
714 	int mode = current->seccomp.mode;
715 
716 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
717 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
718 		return;
719 
720 	if (mode == SECCOMP_MODE_DISABLED)
721 		return;
722 	else if (mode == SECCOMP_MODE_STRICT)
723 		__secure_computing_strict(this_syscall);
724 	else
725 		BUG();
726 }
727 #else
728 
729 #ifdef CONFIG_SECCOMP_FILTER
730 static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
731 {
732 	/*
733 	 * Note: overflow is ok here, the id just needs to be unique per
734 	 * filter.
735 	 */
736 	lockdep_assert_held(&filter->notify_lock);
737 	return filter->notif->next_id++;
738 }
739 
740 static int seccomp_do_user_notification(int this_syscall,
741 					struct seccomp_filter *match,
742 					const struct seccomp_data *sd)
743 {
744 	int err;
745 	u32 flags = 0;
746 	long ret = 0;
747 	struct seccomp_knotif n = {};
748 
749 	mutex_lock(&match->notify_lock);
750 	err = -ENOSYS;
751 	if (!match->notif)
752 		goto out;
753 
754 	n.task = current;
755 	n.state = SECCOMP_NOTIFY_INIT;
756 	n.data = sd;
757 	n.id = seccomp_next_notify_id(match);
758 	init_completion(&n.ready);
759 	list_add(&n.list, &match->notif->notifications);
760 
761 	up(&match->notif->request);
762 	wake_up_poll(&match->notif->wqh, EPOLLIN | EPOLLRDNORM);
763 	mutex_unlock(&match->notify_lock);
764 
765 	/*
766 	 * This is where we wait for a reply from userspace.
767 	 */
768 	err = wait_for_completion_interruptible(&n.ready);
769 	mutex_lock(&match->notify_lock);
770 	if (err == 0) {
771 		ret = n.val;
772 		err = n.error;
773 		flags = n.flags;
774 	}
775 
776 	/*
777 	 * Note that it's possible the listener died in between the time when
778 	 * we were notified of a respons (or a signal) and when we were able to
779 	 * re-acquire the lock, so only delete from the list if the
780 	 * notification actually exists.
781 	 *
782 	 * Also note that this test is only valid because there's no way to
783 	 * *reattach* to a notifier right now. If one is added, we'll need to
784 	 * keep track of the notif itself and make sure they match here.
785 	 */
786 	if (match->notif)
787 		list_del(&n.list);
788 out:
789 	mutex_unlock(&match->notify_lock);
790 
791 	/* Userspace requests to continue the syscall. */
792 	if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE)
793 		return 0;
794 
795 	syscall_set_return_value(current, task_pt_regs(current),
796 				 err, ret);
797 	return -1;
798 }
799 
800 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
801 			    const bool recheck_after_trace)
802 {
803 	u32 filter_ret, action;
804 	struct seccomp_filter *match = NULL;
805 	int data;
806 	struct seccomp_data sd_local;
807 
808 	/*
809 	 * Make sure that any changes to mode from another thread have
810 	 * been seen after TIF_SECCOMP was seen.
811 	 */
812 	rmb();
813 
814 	if (!sd) {
815 		populate_seccomp_data(&sd_local);
816 		sd = &sd_local;
817 	}
818 
819 	filter_ret = seccomp_run_filters(sd, &match);
820 	data = filter_ret & SECCOMP_RET_DATA;
821 	action = filter_ret & SECCOMP_RET_ACTION_FULL;
822 
823 	switch (action) {
824 	case SECCOMP_RET_ERRNO:
825 		/* Set low-order bits as an errno, capped at MAX_ERRNO. */
826 		if (data > MAX_ERRNO)
827 			data = MAX_ERRNO;
828 		syscall_set_return_value(current, task_pt_regs(current),
829 					 -data, 0);
830 		goto skip;
831 
832 	case SECCOMP_RET_TRAP:
833 		/* Show the handler the original registers. */
834 		syscall_rollback(current, task_pt_regs(current));
835 		/* Let the filter pass back 16 bits of data. */
836 		seccomp_send_sigsys(this_syscall, data);
837 		goto skip;
838 
839 	case SECCOMP_RET_TRACE:
840 		/* We've been put in this state by the ptracer already. */
841 		if (recheck_after_trace)
842 			return 0;
843 
844 		/* ENOSYS these calls if there is no tracer attached. */
845 		if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
846 			syscall_set_return_value(current,
847 						 task_pt_regs(current),
848 						 -ENOSYS, 0);
849 			goto skip;
850 		}
851 
852 		/* Allow the BPF to provide the event message */
853 		ptrace_event(PTRACE_EVENT_SECCOMP, data);
854 		/*
855 		 * The delivery of a fatal signal during event
856 		 * notification may silently skip tracer notification,
857 		 * which could leave us with a potentially unmodified
858 		 * syscall that the tracer would have liked to have
859 		 * changed. Since the process is about to die, we just
860 		 * force the syscall to be skipped and let the signal
861 		 * kill the process and correctly handle any tracer exit
862 		 * notifications.
863 		 */
864 		if (fatal_signal_pending(current))
865 			goto skip;
866 		/* Check if the tracer forced the syscall to be skipped. */
867 		this_syscall = syscall_get_nr(current, task_pt_regs(current));
868 		if (this_syscall < 0)
869 			goto skip;
870 
871 		/*
872 		 * Recheck the syscall, since it may have changed. This
873 		 * intentionally uses a NULL struct seccomp_data to force
874 		 * a reload of all registers. This does not goto skip since
875 		 * a skip would have already been reported.
876 		 */
877 		if (__seccomp_filter(this_syscall, NULL, true))
878 			return -1;
879 
880 		return 0;
881 
882 	case SECCOMP_RET_USER_NOTIF:
883 		if (seccomp_do_user_notification(this_syscall, match, sd))
884 			goto skip;
885 
886 		return 0;
887 
888 	case SECCOMP_RET_LOG:
889 		seccomp_log(this_syscall, 0, action, true);
890 		return 0;
891 
892 	case SECCOMP_RET_ALLOW:
893 		/*
894 		 * Note that the "match" filter will always be NULL for
895 		 * this action since SECCOMP_RET_ALLOW is the starting
896 		 * state in seccomp_run_filters().
897 		 */
898 		return 0;
899 
900 	case SECCOMP_RET_KILL_THREAD:
901 	case SECCOMP_RET_KILL_PROCESS:
902 	default:
903 		seccomp_log(this_syscall, SIGSYS, action, true);
904 		/* Dump core only if this is the last remaining thread. */
905 		if (action == SECCOMP_RET_KILL_PROCESS ||
906 		    get_nr_threads(current) == 1) {
907 			kernel_siginfo_t info;
908 
909 			/* Show the original registers in the dump. */
910 			syscall_rollback(current, task_pt_regs(current));
911 			/* Trigger a manual coredump since do_exit skips it. */
912 			seccomp_init_siginfo(&info, this_syscall, data);
913 			do_coredump(&info);
914 		}
915 		if (action == SECCOMP_RET_KILL_PROCESS)
916 			do_group_exit(SIGSYS);
917 		else
918 			do_exit(SIGSYS);
919 	}
920 
921 	unreachable();
922 
923 skip:
924 	seccomp_log(this_syscall, 0, action, match ? match->log : false);
925 	return -1;
926 }
927 #else
928 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
929 			    const bool recheck_after_trace)
930 {
931 	BUG();
932 }
933 #endif
934 
935 int __secure_computing(const struct seccomp_data *sd)
936 {
937 	int mode = current->seccomp.mode;
938 	int this_syscall;
939 
940 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
941 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
942 		return 0;
943 
944 	this_syscall = sd ? sd->nr :
945 		syscall_get_nr(current, task_pt_regs(current));
946 
947 	switch (mode) {
948 	case SECCOMP_MODE_STRICT:
949 		__secure_computing_strict(this_syscall);  /* may call do_exit */
950 		return 0;
951 	case SECCOMP_MODE_FILTER:
952 		return __seccomp_filter(this_syscall, sd, false);
953 	default:
954 		BUG();
955 	}
956 }
957 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
958 
959 long prctl_get_seccomp(void)
960 {
961 	return current->seccomp.mode;
962 }
963 
964 /**
965  * seccomp_set_mode_strict: internal function for setting strict seccomp
966  *
967  * Once current->seccomp.mode is non-zero, it may not be changed.
968  *
969  * Returns 0 on success or -EINVAL on failure.
970  */
971 static long seccomp_set_mode_strict(void)
972 {
973 	const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
974 	long ret = -EINVAL;
975 
976 	spin_lock_irq(&current->sighand->siglock);
977 
978 	if (!seccomp_may_assign_mode(seccomp_mode))
979 		goto out;
980 
981 #ifdef TIF_NOTSC
982 	disable_TSC();
983 #endif
984 	seccomp_assign_mode(current, seccomp_mode, 0);
985 	ret = 0;
986 
987 out:
988 	spin_unlock_irq(&current->sighand->siglock);
989 
990 	return ret;
991 }
992 
993 #ifdef CONFIG_SECCOMP_FILTER
994 static int seccomp_notify_release(struct inode *inode, struct file *file)
995 {
996 	struct seccomp_filter *filter = file->private_data;
997 	struct seccomp_knotif *knotif;
998 
999 	if (!filter)
1000 		return 0;
1001 
1002 	mutex_lock(&filter->notify_lock);
1003 
1004 	/*
1005 	 * If this file is being closed because e.g. the task who owned it
1006 	 * died, let's wake everyone up who was waiting on us.
1007 	 */
1008 	list_for_each_entry(knotif, &filter->notif->notifications, list) {
1009 		if (knotif->state == SECCOMP_NOTIFY_REPLIED)
1010 			continue;
1011 
1012 		knotif->state = SECCOMP_NOTIFY_REPLIED;
1013 		knotif->error = -ENOSYS;
1014 		knotif->val = 0;
1015 
1016 		complete(&knotif->ready);
1017 	}
1018 
1019 	kfree(filter->notif);
1020 	filter->notif = NULL;
1021 	mutex_unlock(&filter->notify_lock);
1022 	__put_seccomp_filter(filter);
1023 	return 0;
1024 }
1025 
1026 static long seccomp_notify_recv(struct seccomp_filter *filter,
1027 				void __user *buf)
1028 {
1029 	struct seccomp_knotif *knotif = NULL, *cur;
1030 	struct seccomp_notif unotif;
1031 	ssize_t ret;
1032 
1033 	/* Verify that we're not given garbage to keep struct extensible. */
1034 	ret = check_zeroed_user(buf, sizeof(unotif));
1035 	if (ret < 0)
1036 		return ret;
1037 	if (!ret)
1038 		return -EINVAL;
1039 
1040 	memset(&unotif, 0, sizeof(unotif));
1041 
1042 	ret = down_interruptible(&filter->notif->request);
1043 	if (ret < 0)
1044 		return ret;
1045 
1046 	mutex_lock(&filter->notify_lock);
1047 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1048 		if (cur->state == SECCOMP_NOTIFY_INIT) {
1049 			knotif = cur;
1050 			break;
1051 		}
1052 	}
1053 
1054 	/*
1055 	 * If we didn't find a notification, it could be that the task was
1056 	 * interrupted by a fatal signal between the time we were woken and
1057 	 * when we were able to acquire the rw lock.
1058 	 */
1059 	if (!knotif) {
1060 		ret = -ENOENT;
1061 		goto out;
1062 	}
1063 
1064 	unotif.id = knotif->id;
1065 	unotif.pid = task_pid_vnr(knotif->task);
1066 	unotif.data = *(knotif->data);
1067 
1068 	knotif->state = SECCOMP_NOTIFY_SENT;
1069 	wake_up_poll(&filter->notif->wqh, EPOLLOUT | EPOLLWRNORM);
1070 	ret = 0;
1071 out:
1072 	mutex_unlock(&filter->notify_lock);
1073 
1074 	if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1075 		ret = -EFAULT;
1076 
1077 		/*
1078 		 * Userspace screwed up. To make sure that we keep this
1079 		 * notification alive, let's reset it back to INIT. It
1080 		 * may have died when we released the lock, so we need to make
1081 		 * sure it's still around.
1082 		 */
1083 		knotif = NULL;
1084 		mutex_lock(&filter->notify_lock);
1085 		list_for_each_entry(cur, &filter->notif->notifications, list) {
1086 			if (cur->id == unotif.id) {
1087 				knotif = cur;
1088 				break;
1089 			}
1090 		}
1091 
1092 		if (knotif) {
1093 			knotif->state = SECCOMP_NOTIFY_INIT;
1094 			up(&filter->notif->request);
1095 		}
1096 		mutex_unlock(&filter->notify_lock);
1097 	}
1098 
1099 	return ret;
1100 }
1101 
1102 static long seccomp_notify_send(struct seccomp_filter *filter,
1103 				void __user *buf)
1104 {
1105 	struct seccomp_notif_resp resp = {};
1106 	struct seccomp_knotif *knotif = NULL, *cur;
1107 	long ret;
1108 
1109 	if (copy_from_user(&resp, buf, sizeof(resp)))
1110 		return -EFAULT;
1111 
1112 	if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1113 		return -EINVAL;
1114 
1115 	if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) &&
1116 	    (resp.error || resp.val))
1117 		return -EINVAL;
1118 
1119 	ret = mutex_lock_interruptible(&filter->notify_lock);
1120 	if (ret < 0)
1121 		return ret;
1122 
1123 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1124 		if (cur->id == resp.id) {
1125 			knotif = cur;
1126 			break;
1127 		}
1128 	}
1129 
1130 	if (!knotif) {
1131 		ret = -ENOENT;
1132 		goto out;
1133 	}
1134 
1135 	/* Allow exactly one reply. */
1136 	if (knotif->state != SECCOMP_NOTIFY_SENT) {
1137 		ret = -EINPROGRESS;
1138 		goto out;
1139 	}
1140 
1141 	ret = 0;
1142 	knotif->state = SECCOMP_NOTIFY_REPLIED;
1143 	knotif->error = resp.error;
1144 	knotif->val = resp.val;
1145 	knotif->flags = resp.flags;
1146 	complete(&knotif->ready);
1147 out:
1148 	mutex_unlock(&filter->notify_lock);
1149 	return ret;
1150 }
1151 
1152 static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1153 				    void __user *buf)
1154 {
1155 	struct seccomp_knotif *knotif = NULL;
1156 	u64 id;
1157 	long ret;
1158 
1159 	if (copy_from_user(&id, buf, sizeof(id)))
1160 		return -EFAULT;
1161 
1162 	ret = mutex_lock_interruptible(&filter->notify_lock);
1163 	if (ret < 0)
1164 		return ret;
1165 
1166 	ret = -ENOENT;
1167 	list_for_each_entry(knotif, &filter->notif->notifications, list) {
1168 		if (knotif->id == id) {
1169 			if (knotif->state == SECCOMP_NOTIFY_SENT)
1170 				ret = 0;
1171 			goto out;
1172 		}
1173 	}
1174 
1175 out:
1176 	mutex_unlock(&filter->notify_lock);
1177 	return ret;
1178 }
1179 
1180 static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1181 				 unsigned long arg)
1182 {
1183 	struct seccomp_filter *filter = file->private_data;
1184 	void __user *buf = (void __user *)arg;
1185 
1186 	switch (cmd) {
1187 	case SECCOMP_IOCTL_NOTIF_RECV:
1188 		return seccomp_notify_recv(filter, buf);
1189 	case SECCOMP_IOCTL_NOTIF_SEND:
1190 		return seccomp_notify_send(filter, buf);
1191 	case SECCOMP_IOCTL_NOTIF_ID_VALID:
1192 		return seccomp_notify_id_valid(filter, buf);
1193 	default:
1194 		return -EINVAL;
1195 	}
1196 }
1197 
1198 static __poll_t seccomp_notify_poll(struct file *file,
1199 				    struct poll_table_struct *poll_tab)
1200 {
1201 	struct seccomp_filter *filter = file->private_data;
1202 	__poll_t ret = 0;
1203 	struct seccomp_knotif *cur;
1204 
1205 	poll_wait(file, &filter->notif->wqh, poll_tab);
1206 
1207 	if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1208 		return EPOLLERR;
1209 
1210 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1211 		if (cur->state == SECCOMP_NOTIFY_INIT)
1212 			ret |= EPOLLIN | EPOLLRDNORM;
1213 		if (cur->state == SECCOMP_NOTIFY_SENT)
1214 			ret |= EPOLLOUT | EPOLLWRNORM;
1215 		if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1216 			break;
1217 	}
1218 
1219 	mutex_unlock(&filter->notify_lock);
1220 
1221 	return ret;
1222 }
1223 
1224 static const struct file_operations seccomp_notify_ops = {
1225 	.poll = seccomp_notify_poll,
1226 	.release = seccomp_notify_release,
1227 	.unlocked_ioctl = seccomp_notify_ioctl,
1228 	.compat_ioctl = seccomp_notify_ioctl,
1229 };
1230 
1231 static struct file *init_listener(struct seccomp_filter *filter)
1232 {
1233 	struct file *ret = ERR_PTR(-EBUSY);
1234 	struct seccomp_filter *cur;
1235 
1236 	for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1237 		if (cur->notif)
1238 			goto out;
1239 	}
1240 
1241 	ret = ERR_PTR(-ENOMEM);
1242 	filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1243 	if (!filter->notif)
1244 		goto out;
1245 
1246 	sema_init(&filter->notif->request, 0);
1247 	filter->notif->next_id = get_random_u64();
1248 	INIT_LIST_HEAD(&filter->notif->notifications);
1249 	init_waitqueue_head(&filter->notif->wqh);
1250 
1251 	ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1252 				 filter, O_RDWR);
1253 	if (IS_ERR(ret))
1254 		goto out_notif;
1255 
1256 	/* The file has a reference to it now */
1257 	__get_seccomp_filter(filter);
1258 
1259 out_notif:
1260 	if (IS_ERR(ret))
1261 		kfree(filter->notif);
1262 out:
1263 	return ret;
1264 }
1265 
1266 /**
1267  * seccomp_set_mode_filter: internal function for setting seccomp filter
1268  * @flags:  flags to change filter behavior
1269  * @filter: struct sock_fprog containing filter
1270  *
1271  * This function may be called repeatedly to install additional filters.
1272  * Every filter successfully installed will be evaluated (in reverse order)
1273  * for each system call the task makes.
1274  *
1275  * Once current->seccomp.mode is non-zero, it may not be changed.
1276  *
1277  * Returns 0 on success or -EINVAL on failure.
1278  */
1279 static long seccomp_set_mode_filter(unsigned int flags,
1280 				    const char __user *filter)
1281 {
1282 	const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1283 	struct seccomp_filter *prepared = NULL;
1284 	long ret = -EINVAL;
1285 	int listener = -1;
1286 	struct file *listener_f = NULL;
1287 
1288 	/* Validate flags. */
1289 	if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1290 		return -EINVAL;
1291 
1292 	/*
1293 	 * In the successful case, NEW_LISTENER returns the new listener fd.
1294 	 * But in the failure case, TSYNC returns the thread that died. If you
1295 	 * combine these two flags, there's no way to tell whether something
1296 	 * succeeded or failed. So, let's disallow this combination if the user
1297 	 * has not explicitly requested no errors from TSYNC.
1298 	 */
1299 	if ((flags & SECCOMP_FILTER_FLAG_TSYNC) &&
1300 	    (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) &&
1301 	    ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0))
1302 		return -EINVAL;
1303 
1304 	/* Prepare the new filter before holding any locks. */
1305 	prepared = seccomp_prepare_user_filter(filter);
1306 	if (IS_ERR(prepared))
1307 		return PTR_ERR(prepared);
1308 
1309 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1310 		listener = get_unused_fd_flags(O_CLOEXEC);
1311 		if (listener < 0) {
1312 			ret = listener;
1313 			goto out_free;
1314 		}
1315 
1316 		listener_f = init_listener(prepared);
1317 		if (IS_ERR(listener_f)) {
1318 			put_unused_fd(listener);
1319 			ret = PTR_ERR(listener_f);
1320 			goto out_free;
1321 		}
1322 	}
1323 
1324 	/*
1325 	 * Make sure we cannot change seccomp or nnp state via TSYNC
1326 	 * while another thread is in the middle of calling exec.
1327 	 */
1328 	if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1329 	    mutex_lock_killable(&current->signal->cred_guard_mutex))
1330 		goto out_put_fd;
1331 
1332 	spin_lock_irq(&current->sighand->siglock);
1333 
1334 	if (!seccomp_may_assign_mode(seccomp_mode))
1335 		goto out;
1336 
1337 	ret = seccomp_attach_filter(flags, prepared);
1338 	if (ret)
1339 		goto out;
1340 	/* Do not free the successfully attached filter. */
1341 	prepared = NULL;
1342 
1343 	seccomp_assign_mode(current, seccomp_mode, flags);
1344 out:
1345 	spin_unlock_irq(&current->sighand->siglock);
1346 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
1347 		mutex_unlock(&current->signal->cred_guard_mutex);
1348 out_put_fd:
1349 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1350 		if (ret) {
1351 			listener_f->private_data = NULL;
1352 			fput(listener_f);
1353 			put_unused_fd(listener);
1354 		} else {
1355 			fd_install(listener, listener_f);
1356 			ret = listener;
1357 		}
1358 	}
1359 out_free:
1360 	seccomp_filter_free(prepared);
1361 	return ret;
1362 }
1363 #else
1364 static inline long seccomp_set_mode_filter(unsigned int flags,
1365 					   const char __user *filter)
1366 {
1367 	return -EINVAL;
1368 }
1369 #endif
1370 
1371 static long seccomp_get_action_avail(const char __user *uaction)
1372 {
1373 	u32 action;
1374 
1375 	if (copy_from_user(&action, uaction, sizeof(action)))
1376 		return -EFAULT;
1377 
1378 	switch (action) {
1379 	case SECCOMP_RET_KILL_PROCESS:
1380 	case SECCOMP_RET_KILL_THREAD:
1381 	case SECCOMP_RET_TRAP:
1382 	case SECCOMP_RET_ERRNO:
1383 	case SECCOMP_RET_USER_NOTIF:
1384 	case SECCOMP_RET_TRACE:
1385 	case SECCOMP_RET_LOG:
1386 	case SECCOMP_RET_ALLOW:
1387 		break;
1388 	default:
1389 		return -EOPNOTSUPP;
1390 	}
1391 
1392 	return 0;
1393 }
1394 
1395 static long seccomp_get_notif_sizes(void __user *usizes)
1396 {
1397 	struct seccomp_notif_sizes sizes = {
1398 		.seccomp_notif = sizeof(struct seccomp_notif),
1399 		.seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
1400 		.seccomp_data = sizeof(struct seccomp_data),
1401 	};
1402 
1403 	if (copy_to_user(usizes, &sizes, sizeof(sizes)))
1404 		return -EFAULT;
1405 
1406 	return 0;
1407 }
1408 
1409 /* Common entry point for both prctl and syscall. */
1410 static long do_seccomp(unsigned int op, unsigned int flags,
1411 		       void __user *uargs)
1412 {
1413 	switch (op) {
1414 	case SECCOMP_SET_MODE_STRICT:
1415 		if (flags != 0 || uargs != NULL)
1416 			return -EINVAL;
1417 		return seccomp_set_mode_strict();
1418 	case SECCOMP_SET_MODE_FILTER:
1419 		return seccomp_set_mode_filter(flags, uargs);
1420 	case SECCOMP_GET_ACTION_AVAIL:
1421 		if (flags != 0)
1422 			return -EINVAL;
1423 
1424 		return seccomp_get_action_avail(uargs);
1425 	case SECCOMP_GET_NOTIF_SIZES:
1426 		if (flags != 0)
1427 			return -EINVAL;
1428 
1429 		return seccomp_get_notif_sizes(uargs);
1430 	default:
1431 		return -EINVAL;
1432 	}
1433 }
1434 
1435 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
1436 			 void __user *, uargs)
1437 {
1438 	return do_seccomp(op, flags, uargs);
1439 }
1440 
1441 /**
1442  * prctl_set_seccomp: configures current->seccomp.mode
1443  * @seccomp_mode: requested mode to use
1444  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1445  *
1446  * Returns 0 on success or -EINVAL on failure.
1447  */
1448 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
1449 {
1450 	unsigned int op;
1451 	void __user *uargs;
1452 
1453 	switch (seccomp_mode) {
1454 	case SECCOMP_MODE_STRICT:
1455 		op = SECCOMP_SET_MODE_STRICT;
1456 		/*
1457 		 * Setting strict mode through prctl always ignored filter,
1458 		 * so make sure it is always NULL here to pass the internal
1459 		 * check in do_seccomp().
1460 		 */
1461 		uargs = NULL;
1462 		break;
1463 	case SECCOMP_MODE_FILTER:
1464 		op = SECCOMP_SET_MODE_FILTER;
1465 		uargs = filter;
1466 		break;
1467 	default:
1468 		return -EINVAL;
1469 	}
1470 
1471 	/* prctl interface doesn't have flags, so they are always zero. */
1472 	return do_seccomp(op, 0, uargs);
1473 }
1474 
1475 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
1476 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
1477 					     unsigned long filter_off)
1478 {
1479 	struct seccomp_filter *orig, *filter;
1480 	unsigned long count;
1481 
1482 	/*
1483 	 * Note: this is only correct because the caller should be the (ptrace)
1484 	 * tracer of the task, otherwise lock_task_sighand is needed.
1485 	 */
1486 	spin_lock_irq(&task->sighand->siglock);
1487 
1488 	if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
1489 		spin_unlock_irq(&task->sighand->siglock);
1490 		return ERR_PTR(-EINVAL);
1491 	}
1492 
1493 	orig = task->seccomp.filter;
1494 	__get_seccomp_filter(orig);
1495 	spin_unlock_irq(&task->sighand->siglock);
1496 
1497 	count = 0;
1498 	for (filter = orig; filter; filter = filter->prev)
1499 		count++;
1500 
1501 	if (filter_off >= count) {
1502 		filter = ERR_PTR(-ENOENT);
1503 		goto out;
1504 	}
1505 
1506 	count -= filter_off;
1507 	for (filter = orig; filter && count > 1; filter = filter->prev)
1508 		count--;
1509 
1510 	if (WARN_ON(count != 1 || !filter)) {
1511 		filter = ERR_PTR(-ENOENT);
1512 		goto out;
1513 	}
1514 
1515 	__get_seccomp_filter(filter);
1516 
1517 out:
1518 	__put_seccomp_filter(orig);
1519 	return filter;
1520 }
1521 
1522 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1523 			void __user *data)
1524 {
1525 	struct seccomp_filter *filter;
1526 	struct sock_fprog_kern *fprog;
1527 	long ret;
1528 
1529 	if (!capable(CAP_SYS_ADMIN) ||
1530 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1531 		return -EACCES;
1532 	}
1533 
1534 	filter = get_nth_filter(task, filter_off);
1535 	if (IS_ERR(filter))
1536 		return PTR_ERR(filter);
1537 
1538 	fprog = filter->prog->orig_prog;
1539 	if (!fprog) {
1540 		/* This must be a new non-cBPF filter, since we save
1541 		 * every cBPF filter's orig_prog above when
1542 		 * CONFIG_CHECKPOINT_RESTORE is enabled.
1543 		 */
1544 		ret = -EMEDIUMTYPE;
1545 		goto out;
1546 	}
1547 
1548 	ret = fprog->len;
1549 	if (!data)
1550 		goto out;
1551 
1552 	if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1553 		ret = -EFAULT;
1554 
1555 out:
1556 	__put_seccomp_filter(filter);
1557 	return ret;
1558 }
1559 
1560 long seccomp_get_metadata(struct task_struct *task,
1561 			  unsigned long size, void __user *data)
1562 {
1563 	long ret;
1564 	struct seccomp_filter *filter;
1565 	struct seccomp_metadata kmd = {};
1566 
1567 	if (!capable(CAP_SYS_ADMIN) ||
1568 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1569 		return -EACCES;
1570 	}
1571 
1572 	size = min_t(unsigned long, size, sizeof(kmd));
1573 
1574 	if (size < sizeof(kmd.filter_off))
1575 		return -EINVAL;
1576 
1577 	if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
1578 		return -EFAULT;
1579 
1580 	filter = get_nth_filter(task, kmd.filter_off);
1581 	if (IS_ERR(filter))
1582 		return PTR_ERR(filter);
1583 
1584 	if (filter->log)
1585 		kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1586 
1587 	ret = size;
1588 	if (copy_to_user(data, &kmd, size))
1589 		ret = -EFAULT;
1590 
1591 	__put_seccomp_filter(filter);
1592 	return ret;
1593 }
1594 #endif
1595 
1596 #ifdef CONFIG_SYSCTL
1597 
1598 /* Human readable action names for friendly sysctl interaction */
1599 #define SECCOMP_RET_KILL_PROCESS_NAME	"kill_process"
1600 #define SECCOMP_RET_KILL_THREAD_NAME	"kill_thread"
1601 #define SECCOMP_RET_TRAP_NAME		"trap"
1602 #define SECCOMP_RET_ERRNO_NAME		"errno"
1603 #define SECCOMP_RET_USER_NOTIF_NAME	"user_notif"
1604 #define SECCOMP_RET_TRACE_NAME		"trace"
1605 #define SECCOMP_RET_LOG_NAME		"log"
1606 #define SECCOMP_RET_ALLOW_NAME		"allow"
1607 
1608 static const char seccomp_actions_avail[] =
1609 				SECCOMP_RET_KILL_PROCESS_NAME	" "
1610 				SECCOMP_RET_KILL_THREAD_NAME	" "
1611 				SECCOMP_RET_TRAP_NAME		" "
1612 				SECCOMP_RET_ERRNO_NAME		" "
1613 				SECCOMP_RET_USER_NOTIF_NAME     " "
1614 				SECCOMP_RET_TRACE_NAME		" "
1615 				SECCOMP_RET_LOG_NAME		" "
1616 				SECCOMP_RET_ALLOW_NAME;
1617 
1618 struct seccomp_log_name {
1619 	u32		log;
1620 	const char	*name;
1621 };
1622 
1623 static const struct seccomp_log_name seccomp_log_names[] = {
1624 	{ SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1625 	{ SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1626 	{ SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1627 	{ SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1628 	{ SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
1629 	{ SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1630 	{ SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1631 	{ SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1632 	{ }
1633 };
1634 
1635 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1636 					      u32 actions_logged,
1637 					      const char *sep)
1638 {
1639 	const struct seccomp_log_name *cur;
1640 	bool append_sep = false;
1641 
1642 	for (cur = seccomp_log_names; cur->name && size; cur++) {
1643 		ssize_t ret;
1644 
1645 		if (!(actions_logged & cur->log))
1646 			continue;
1647 
1648 		if (append_sep) {
1649 			ret = strscpy(names, sep, size);
1650 			if (ret < 0)
1651 				return false;
1652 
1653 			names += ret;
1654 			size -= ret;
1655 		} else
1656 			append_sep = true;
1657 
1658 		ret = strscpy(names, cur->name, size);
1659 		if (ret < 0)
1660 			return false;
1661 
1662 		names += ret;
1663 		size -= ret;
1664 	}
1665 
1666 	return true;
1667 }
1668 
1669 static bool seccomp_action_logged_from_name(u32 *action_logged,
1670 					    const char *name)
1671 {
1672 	const struct seccomp_log_name *cur;
1673 
1674 	for (cur = seccomp_log_names; cur->name; cur++) {
1675 		if (!strcmp(cur->name, name)) {
1676 			*action_logged = cur->log;
1677 			return true;
1678 		}
1679 	}
1680 
1681 	return false;
1682 }
1683 
1684 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1685 {
1686 	char *name;
1687 
1688 	*actions_logged = 0;
1689 	while ((name = strsep(&names, " ")) && *name) {
1690 		u32 action_logged = 0;
1691 
1692 		if (!seccomp_action_logged_from_name(&action_logged, name))
1693 			return false;
1694 
1695 		*actions_logged |= action_logged;
1696 	}
1697 
1698 	return true;
1699 }
1700 
1701 static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1702 			       size_t *lenp, loff_t *ppos)
1703 {
1704 	char names[sizeof(seccomp_actions_avail)];
1705 	struct ctl_table table;
1706 
1707 	memset(names, 0, sizeof(names));
1708 
1709 	if (!seccomp_names_from_actions_logged(names, sizeof(names),
1710 					       seccomp_actions_logged, " "))
1711 		return -EINVAL;
1712 
1713 	table = *ro_table;
1714 	table.data = names;
1715 	table.maxlen = sizeof(names);
1716 	return proc_dostring(&table, 0, buffer, lenp, ppos);
1717 }
1718 
1719 static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1720 				size_t *lenp, loff_t *ppos, u32 *actions_logged)
1721 {
1722 	char names[sizeof(seccomp_actions_avail)];
1723 	struct ctl_table table;
1724 	int ret;
1725 
1726 	if (!capable(CAP_SYS_ADMIN))
1727 		return -EPERM;
1728 
1729 	memset(names, 0, sizeof(names));
1730 
1731 	table = *ro_table;
1732 	table.data = names;
1733 	table.maxlen = sizeof(names);
1734 	ret = proc_dostring(&table, 1, buffer, lenp, ppos);
1735 	if (ret)
1736 		return ret;
1737 
1738 	if (!seccomp_actions_logged_from_names(actions_logged, table.data))
1739 		return -EINVAL;
1740 
1741 	if (*actions_logged & SECCOMP_LOG_ALLOW)
1742 		return -EINVAL;
1743 
1744 	seccomp_actions_logged = *actions_logged;
1745 	return 0;
1746 }
1747 
1748 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
1749 				 int ret)
1750 {
1751 	char names[sizeof(seccomp_actions_avail)];
1752 	char old_names[sizeof(seccomp_actions_avail)];
1753 	const char *new = names;
1754 	const char *old = old_names;
1755 
1756 	if (!audit_enabled)
1757 		return;
1758 
1759 	memset(names, 0, sizeof(names));
1760 	memset(old_names, 0, sizeof(old_names));
1761 
1762 	if (ret)
1763 		new = "?";
1764 	else if (!actions_logged)
1765 		new = "(none)";
1766 	else if (!seccomp_names_from_actions_logged(names, sizeof(names),
1767 						    actions_logged, ","))
1768 		new = "?";
1769 
1770 	if (!old_actions_logged)
1771 		old = "(none)";
1772 	else if (!seccomp_names_from_actions_logged(old_names,
1773 						    sizeof(old_names),
1774 						    old_actions_logged, ","))
1775 		old = "?";
1776 
1777 	return audit_seccomp_actions_logged(new, old, !ret);
1778 }
1779 
1780 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1781 					  void __user *buffer, size_t *lenp,
1782 					  loff_t *ppos)
1783 {
1784 	int ret;
1785 
1786 	if (write) {
1787 		u32 actions_logged = 0;
1788 		u32 old_actions_logged = seccomp_actions_logged;
1789 
1790 		ret = write_actions_logged(ro_table, buffer, lenp, ppos,
1791 					   &actions_logged);
1792 		audit_actions_logged(actions_logged, old_actions_logged, ret);
1793 	} else
1794 		ret = read_actions_logged(ro_table, buffer, lenp, ppos);
1795 
1796 	return ret;
1797 }
1798 
1799 static struct ctl_path seccomp_sysctl_path[] = {
1800 	{ .procname = "kernel", },
1801 	{ .procname = "seccomp", },
1802 	{ }
1803 };
1804 
1805 static struct ctl_table seccomp_sysctl_table[] = {
1806 	{
1807 		.procname	= "actions_avail",
1808 		.data		= (void *) &seccomp_actions_avail,
1809 		.maxlen		= sizeof(seccomp_actions_avail),
1810 		.mode		= 0444,
1811 		.proc_handler	= proc_dostring,
1812 	},
1813 	{
1814 		.procname	= "actions_logged",
1815 		.mode		= 0644,
1816 		.proc_handler	= seccomp_actions_logged_handler,
1817 	},
1818 	{ }
1819 };
1820 
1821 static int __init seccomp_sysctl_init(void)
1822 {
1823 	struct ctl_table_header *hdr;
1824 
1825 	hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1826 	if (!hdr)
1827 		pr_warn("seccomp: sysctl registration failed\n");
1828 	else
1829 		kmemleak_not_leak(hdr);
1830 
1831 	return 0;
1832 }
1833 
1834 device_initcall(seccomp_sysctl_init)
1835 
1836 #endif /* CONFIG_SYSCTL */
1837