xref: /openbmc/linux/kernel/seccomp.c (revision aa0dc6a7)
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 #define pr_fmt(fmt) "seccomp: " fmt
17 
18 #include <linux/refcount.h>
19 #include <linux/audit.h>
20 #include <linux/compat.h>
21 #include <linux/coredump.h>
22 #include <linux/kmemleak.h>
23 #include <linux/nospec.h>
24 #include <linux/prctl.h>
25 #include <linux/sched.h>
26 #include <linux/sched/task_stack.h>
27 #include <linux/seccomp.h>
28 #include <linux/slab.h>
29 #include <linux/syscalls.h>
30 #include <linux/sysctl.h>
31 
32 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
33 #include <asm/syscall.h>
34 #endif
35 
36 #ifdef CONFIG_SECCOMP_FILTER
37 #include <linux/file.h>
38 #include <linux/filter.h>
39 #include <linux/pid.h>
40 #include <linux/ptrace.h>
41 #include <linux/capability.h>
42 #include <linux/tracehook.h>
43 #include <linux/uaccess.h>
44 #include <linux/anon_inodes.h>
45 #include <linux/lockdep.h>
46 
47 /*
48  * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the
49  * wrong direction flag in the ioctl number. This is the broken one,
50  * which the kernel needs to keep supporting until all userspaces stop
51  * using the wrong command number.
52  */
53 #define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR	SECCOMP_IOR(2, __u64)
54 
55 enum notify_state {
56 	SECCOMP_NOTIFY_INIT,
57 	SECCOMP_NOTIFY_SENT,
58 	SECCOMP_NOTIFY_REPLIED,
59 };
60 
61 struct seccomp_knotif {
62 	/* The struct pid of the task whose filter triggered the notification */
63 	struct task_struct *task;
64 
65 	/* The "cookie" for this request; this is unique for this filter. */
66 	u64 id;
67 
68 	/*
69 	 * The seccomp data. This pointer is valid the entire time this
70 	 * notification is active, since it comes from __seccomp_filter which
71 	 * eclipses the entire lifecycle here.
72 	 */
73 	const struct seccomp_data *data;
74 
75 	/*
76 	 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
77 	 * struct seccomp_knotif is created and starts out in INIT. Once the
78 	 * handler reads the notification off of an FD, it transitions to SENT.
79 	 * If a signal is received the state transitions back to INIT and
80 	 * another message is sent. When the userspace handler replies, state
81 	 * transitions to REPLIED.
82 	 */
83 	enum notify_state state;
84 
85 	/* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
86 	int error;
87 	long val;
88 	u32 flags;
89 
90 	/*
91 	 * Signals when this has changed states, such as the listener
92 	 * dying, a new seccomp addfd message, or changing to REPLIED
93 	 */
94 	struct completion ready;
95 
96 	struct list_head list;
97 
98 	/* outstanding addfd requests */
99 	struct list_head addfd;
100 };
101 
102 /**
103  * struct seccomp_kaddfd - container for seccomp_addfd ioctl messages
104  *
105  * @file: A reference to the file to install in the other task
106  * @fd: The fd number to install it at. If the fd number is -1, it means the
107  *      installing process should allocate the fd as normal.
108  * @flags: The flags for the new file descriptor. At the moment, only O_CLOEXEC
109  *         is allowed.
110  * @ioctl_flags: The flags used for the seccomp_addfd ioctl.
111  * @ret: The return value of the installing process. It is set to the fd num
112  *       upon success (>= 0).
113  * @completion: Indicates that the installing process has completed fd
114  *              installation, or gone away (either due to successful
115  *              reply, or signal)
116  *
117  */
118 struct seccomp_kaddfd {
119 	struct file *file;
120 	int fd;
121 	unsigned int flags;
122 	__u32 ioctl_flags;
123 
124 	union {
125 		bool setfd;
126 		/* To only be set on reply */
127 		int ret;
128 	};
129 	struct completion completion;
130 	struct list_head list;
131 };
132 
133 /**
134  * struct notification - container for seccomp userspace notifications. Since
135  * most seccomp filters will not have notification listeners attached and this
136  * structure is fairly large, we store the notification-specific stuff in a
137  * separate structure.
138  *
139  * @request: A semaphore that users of this notification can wait on for
140  *           changes. Actual reads and writes are still controlled with
141  *           filter->notify_lock.
142  * @next_id: The id of the next request.
143  * @notifications: A list of struct seccomp_knotif elements.
144  */
145 struct notification {
146 	struct semaphore request;
147 	u64 next_id;
148 	struct list_head notifications;
149 };
150 
151 #ifdef SECCOMP_ARCH_NATIVE
152 /**
153  * struct action_cache - per-filter cache of seccomp actions per
154  * arch/syscall pair
155  *
156  * @allow_native: A bitmap where each bit represents whether the
157  *		  filter will always allow the syscall, for the
158  *		  native architecture.
159  * @allow_compat: A bitmap where each bit represents whether the
160  *		  filter will always allow the syscall, for the
161  *		  compat architecture.
162  */
163 struct action_cache {
164 	DECLARE_BITMAP(allow_native, SECCOMP_ARCH_NATIVE_NR);
165 #ifdef SECCOMP_ARCH_COMPAT
166 	DECLARE_BITMAP(allow_compat, SECCOMP_ARCH_COMPAT_NR);
167 #endif
168 };
169 #else
170 struct action_cache { };
171 
172 static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilter,
173 					     const struct seccomp_data *sd)
174 {
175 	return false;
176 }
177 
178 static inline void seccomp_cache_prepare(struct seccomp_filter *sfilter)
179 {
180 }
181 #endif /* SECCOMP_ARCH_NATIVE */
182 
183 /**
184  * struct seccomp_filter - container for seccomp BPF programs
185  *
186  * @refs: Reference count to manage the object lifetime.
187  *	  A filter's reference count is incremented for each directly
188  *	  attached task, once for the dependent filter, and if
189  *	  requested for the user notifier. When @refs reaches zero,
190  *	  the filter can be freed.
191  * @users: A filter's @users count is incremented for each directly
192  *         attached task (filter installation, fork(), thread_sync),
193  *	   and once for the dependent filter (tracked in filter->prev).
194  *	   When it reaches zero it indicates that no direct or indirect
195  *	   users of that filter exist. No new tasks can get associated with
196  *	   this filter after reaching 0. The @users count is always smaller
197  *	   or equal to @refs. Hence, reaching 0 for @users does not mean
198  *	   the filter can be freed.
199  * @cache: cache of arch/syscall mappings to actions
200  * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
201  * @prev: points to a previously installed, or inherited, filter
202  * @prog: the BPF program to evaluate
203  * @notif: the struct that holds all notification related information
204  * @notify_lock: A lock for all notification-related accesses.
205  * @wqh: A wait queue for poll if a notifier is in use.
206  *
207  * seccomp_filter objects are organized in a tree linked via the @prev
208  * pointer.  For any task, it appears to be a singly-linked list starting
209  * with current->seccomp.filter, the most recently attached or inherited filter.
210  * However, multiple filters may share a @prev node, by way of fork(), which
211  * results in a unidirectional tree existing in memory.  This is similar to
212  * how namespaces work.
213  *
214  * seccomp_filter objects should never be modified after being attached
215  * to a task_struct (other than @refs).
216  */
217 struct seccomp_filter {
218 	refcount_t refs;
219 	refcount_t users;
220 	bool log;
221 	struct action_cache cache;
222 	struct seccomp_filter *prev;
223 	struct bpf_prog *prog;
224 	struct notification *notif;
225 	struct mutex notify_lock;
226 	wait_queue_head_t wqh;
227 };
228 
229 /* Limit any path through the tree to 256KB worth of instructions. */
230 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
231 
232 /*
233  * Endianness is explicitly ignored and left for BPF program authors to manage
234  * as per the specific architecture.
235  */
236 static void populate_seccomp_data(struct seccomp_data *sd)
237 {
238 	/*
239 	 * Instead of using current_pt_reg(), we're already doing the work
240 	 * to safely fetch "current", so just use "task" everywhere below.
241 	 */
242 	struct task_struct *task = current;
243 	struct pt_regs *regs = task_pt_regs(task);
244 	unsigned long args[6];
245 
246 	sd->nr = syscall_get_nr(task, regs);
247 	sd->arch = syscall_get_arch(task);
248 	syscall_get_arguments(task, regs, args);
249 	sd->args[0] = args[0];
250 	sd->args[1] = args[1];
251 	sd->args[2] = args[2];
252 	sd->args[3] = args[3];
253 	sd->args[4] = args[4];
254 	sd->args[5] = args[5];
255 	sd->instruction_pointer = KSTK_EIP(task);
256 }
257 
258 /**
259  *	seccomp_check_filter - verify seccomp filter code
260  *	@filter: filter to verify
261  *	@flen: length of filter
262  *
263  * Takes a previously checked filter (by bpf_check_classic) and
264  * redirects all filter code that loads struct sk_buff data
265  * and related data through seccomp_bpf_load.  It also
266  * enforces length and alignment checking of those loads.
267  *
268  * Returns 0 if the rule set is legal or -EINVAL if not.
269  */
270 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
271 {
272 	int pc;
273 	for (pc = 0; pc < flen; pc++) {
274 		struct sock_filter *ftest = &filter[pc];
275 		u16 code = ftest->code;
276 		u32 k = ftest->k;
277 
278 		switch (code) {
279 		case BPF_LD | BPF_W | BPF_ABS:
280 			ftest->code = BPF_LDX | BPF_W | BPF_ABS;
281 			/* 32-bit aligned and not out of bounds. */
282 			if (k >= sizeof(struct seccomp_data) || k & 3)
283 				return -EINVAL;
284 			continue;
285 		case BPF_LD | BPF_W | BPF_LEN:
286 			ftest->code = BPF_LD | BPF_IMM;
287 			ftest->k = sizeof(struct seccomp_data);
288 			continue;
289 		case BPF_LDX | BPF_W | BPF_LEN:
290 			ftest->code = BPF_LDX | BPF_IMM;
291 			ftest->k = sizeof(struct seccomp_data);
292 			continue;
293 		/* Explicitly include allowed calls. */
294 		case BPF_RET | BPF_K:
295 		case BPF_RET | BPF_A:
296 		case BPF_ALU | BPF_ADD | BPF_K:
297 		case BPF_ALU | BPF_ADD | BPF_X:
298 		case BPF_ALU | BPF_SUB | BPF_K:
299 		case BPF_ALU | BPF_SUB | BPF_X:
300 		case BPF_ALU | BPF_MUL | BPF_K:
301 		case BPF_ALU | BPF_MUL | BPF_X:
302 		case BPF_ALU | BPF_DIV | BPF_K:
303 		case BPF_ALU | BPF_DIV | BPF_X:
304 		case BPF_ALU | BPF_AND | BPF_K:
305 		case BPF_ALU | BPF_AND | BPF_X:
306 		case BPF_ALU | BPF_OR | BPF_K:
307 		case BPF_ALU | BPF_OR | BPF_X:
308 		case BPF_ALU | BPF_XOR | BPF_K:
309 		case BPF_ALU | BPF_XOR | BPF_X:
310 		case BPF_ALU | BPF_LSH | BPF_K:
311 		case BPF_ALU | BPF_LSH | BPF_X:
312 		case BPF_ALU | BPF_RSH | BPF_K:
313 		case BPF_ALU | BPF_RSH | BPF_X:
314 		case BPF_ALU | BPF_NEG:
315 		case BPF_LD | BPF_IMM:
316 		case BPF_LDX | BPF_IMM:
317 		case BPF_MISC | BPF_TAX:
318 		case BPF_MISC | BPF_TXA:
319 		case BPF_LD | BPF_MEM:
320 		case BPF_LDX | BPF_MEM:
321 		case BPF_ST:
322 		case BPF_STX:
323 		case BPF_JMP | BPF_JA:
324 		case BPF_JMP | BPF_JEQ | BPF_K:
325 		case BPF_JMP | BPF_JEQ | BPF_X:
326 		case BPF_JMP | BPF_JGE | BPF_K:
327 		case BPF_JMP | BPF_JGE | BPF_X:
328 		case BPF_JMP | BPF_JGT | BPF_K:
329 		case BPF_JMP | BPF_JGT | BPF_X:
330 		case BPF_JMP | BPF_JSET | BPF_K:
331 		case BPF_JMP | BPF_JSET | BPF_X:
332 			continue;
333 		default:
334 			return -EINVAL;
335 		}
336 	}
337 	return 0;
338 }
339 
340 #ifdef SECCOMP_ARCH_NATIVE
341 static inline bool seccomp_cache_check_allow_bitmap(const void *bitmap,
342 						    size_t bitmap_size,
343 						    int syscall_nr)
344 {
345 	if (unlikely(syscall_nr < 0 || syscall_nr >= bitmap_size))
346 		return false;
347 	syscall_nr = array_index_nospec(syscall_nr, bitmap_size);
348 
349 	return test_bit(syscall_nr, bitmap);
350 }
351 
352 /**
353  * seccomp_cache_check_allow - lookup seccomp cache
354  * @sfilter: The seccomp filter
355  * @sd: The seccomp data to lookup the cache with
356  *
357  * Returns true if the seccomp_data is cached and allowed.
358  */
359 static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilter,
360 					     const struct seccomp_data *sd)
361 {
362 	int syscall_nr = sd->nr;
363 	const struct action_cache *cache = &sfilter->cache;
364 
365 #ifndef SECCOMP_ARCH_COMPAT
366 	/* A native-only architecture doesn't need to check sd->arch. */
367 	return seccomp_cache_check_allow_bitmap(cache->allow_native,
368 						SECCOMP_ARCH_NATIVE_NR,
369 						syscall_nr);
370 #else
371 	if (likely(sd->arch == SECCOMP_ARCH_NATIVE))
372 		return seccomp_cache_check_allow_bitmap(cache->allow_native,
373 							SECCOMP_ARCH_NATIVE_NR,
374 							syscall_nr);
375 	if (likely(sd->arch == SECCOMP_ARCH_COMPAT))
376 		return seccomp_cache_check_allow_bitmap(cache->allow_compat,
377 							SECCOMP_ARCH_COMPAT_NR,
378 							syscall_nr);
379 #endif /* SECCOMP_ARCH_COMPAT */
380 
381 	WARN_ON_ONCE(true);
382 	return false;
383 }
384 #endif /* SECCOMP_ARCH_NATIVE */
385 
386 /**
387  * seccomp_run_filters - evaluates all seccomp filters against @sd
388  * @sd: optional seccomp data to be passed to filters
389  * @match: stores struct seccomp_filter that resulted in the return value,
390  *         unless filter returned SECCOMP_RET_ALLOW, in which case it will
391  *         be unchanged.
392  *
393  * Returns valid seccomp BPF response codes.
394  */
395 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
396 static u32 seccomp_run_filters(const struct seccomp_data *sd,
397 			       struct seccomp_filter **match)
398 {
399 	u32 ret = SECCOMP_RET_ALLOW;
400 	/* Make sure cross-thread synced filter points somewhere sane. */
401 	struct seccomp_filter *f =
402 			READ_ONCE(current->seccomp.filter);
403 
404 	/* Ensure unexpected behavior doesn't result in failing open. */
405 	if (WARN_ON(f == NULL))
406 		return SECCOMP_RET_KILL_PROCESS;
407 
408 	if (seccomp_cache_check_allow(f, sd))
409 		return SECCOMP_RET_ALLOW;
410 
411 	/*
412 	 * All filters in the list are evaluated and the lowest BPF return
413 	 * value always takes priority (ignoring the DATA).
414 	 */
415 	for (; f; f = f->prev) {
416 		u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd);
417 
418 		if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
419 			ret = cur_ret;
420 			*match = f;
421 		}
422 	}
423 	return ret;
424 }
425 #endif /* CONFIG_SECCOMP_FILTER */
426 
427 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
428 {
429 	assert_spin_locked(&current->sighand->siglock);
430 
431 	if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
432 		return false;
433 
434 	return true;
435 }
436 
437 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
438 
439 static inline void seccomp_assign_mode(struct task_struct *task,
440 				       unsigned long seccomp_mode,
441 				       unsigned long flags)
442 {
443 	assert_spin_locked(&task->sighand->siglock);
444 
445 	task->seccomp.mode = seccomp_mode;
446 	/*
447 	 * Make sure SYSCALL_WORK_SECCOMP cannot be set before the mode (and
448 	 * filter) is set.
449 	 */
450 	smp_mb__before_atomic();
451 	/* Assume default seccomp processes want spec flaw mitigation. */
452 	if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
453 		arch_seccomp_spec_mitigate(task);
454 	set_task_syscall_work(task, SECCOMP);
455 }
456 
457 #ifdef CONFIG_SECCOMP_FILTER
458 /* Returns 1 if the parent is an ancestor of the child. */
459 static int is_ancestor(struct seccomp_filter *parent,
460 		       struct seccomp_filter *child)
461 {
462 	/* NULL is the root ancestor. */
463 	if (parent == NULL)
464 		return 1;
465 	for (; child; child = child->prev)
466 		if (child == parent)
467 			return 1;
468 	return 0;
469 }
470 
471 /**
472  * seccomp_can_sync_threads: checks if all threads can be synchronized
473  *
474  * Expects sighand and cred_guard_mutex locks to be held.
475  *
476  * Returns 0 on success, -ve on error, or the pid of a thread which was
477  * either not in the correct seccomp mode or did not have an ancestral
478  * seccomp filter.
479  */
480 static inline pid_t seccomp_can_sync_threads(void)
481 {
482 	struct task_struct *thread, *caller;
483 
484 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
485 	assert_spin_locked(&current->sighand->siglock);
486 
487 	/* Validate all threads being eligible for synchronization. */
488 	caller = current;
489 	for_each_thread(caller, thread) {
490 		pid_t failed;
491 
492 		/* Skip current, since it is initiating the sync. */
493 		if (thread == caller)
494 			continue;
495 
496 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
497 		    (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
498 		     is_ancestor(thread->seccomp.filter,
499 				 caller->seccomp.filter)))
500 			continue;
501 
502 		/* Return the first thread that cannot be synchronized. */
503 		failed = task_pid_vnr(thread);
504 		/* If the pid cannot be resolved, then return -ESRCH */
505 		if (WARN_ON(failed == 0))
506 			failed = -ESRCH;
507 		return failed;
508 	}
509 
510 	return 0;
511 }
512 
513 static inline void seccomp_filter_free(struct seccomp_filter *filter)
514 {
515 	if (filter) {
516 		bpf_prog_destroy(filter->prog);
517 		kfree(filter);
518 	}
519 }
520 
521 static void __seccomp_filter_orphan(struct seccomp_filter *orig)
522 {
523 	while (orig && refcount_dec_and_test(&orig->users)) {
524 		if (waitqueue_active(&orig->wqh))
525 			wake_up_poll(&orig->wqh, EPOLLHUP);
526 		orig = orig->prev;
527 	}
528 }
529 
530 static void __put_seccomp_filter(struct seccomp_filter *orig)
531 {
532 	/* Clean up single-reference branches iteratively. */
533 	while (orig && refcount_dec_and_test(&orig->refs)) {
534 		struct seccomp_filter *freeme = orig;
535 		orig = orig->prev;
536 		seccomp_filter_free(freeme);
537 	}
538 }
539 
540 static void __seccomp_filter_release(struct seccomp_filter *orig)
541 {
542 	/* Notify about any unused filters in the task's former filter tree. */
543 	__seccomp_filter_orphan(orig);
544 	/* Finally drop all references to the task's former tree. */
545 	__put_seccomp_filter(orig);
546 }
547 
548 /**
549  * seccomp_filter_release - Detach the task from its filter tree,
550  *			    drop its reference count, and notify
551  *			    about unused filters
552  *
553  * This function should only be called when the task is exiting as
554  * it detaches it from its filter tree. As such, READ_ONCE() and
555  * barriers are not needed here, as would normally be needed.
556  */
557 void seccomp_filter_release(struct task_struct *tsk)
558 {
559 	struct seccomp_filter *orig = tsk->seccomp.filter;
560 
561 	/* We are effectively holding the siglock by not having any sighand. */
562 	WARN_ON(tsk->sighand != NULL);
563 
564 	/* Detach task from its filter tree. */
565 	tsk->seccomp.filter = NULL;
566 	__seccomp_filter_release(orig);
567 }
568 
569 /**
570  * seccomp_sync_threads: sets all threads to use current's filter
571  *
572  * Expects sighand and cred_guard_mutex locks to be held, and for
573  * seccomp_can_sync_threads() to have returned success already
574  * without dropping the locks.
575  *
576  */
577 static inline void seccomp_sync_threads(unsigned long flags)
578 {
579 	struct task_struct *thread, *caller;
580 
581 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
582 	assert_spin_locked(&current->sighand->siglock);
583 
584 	/* Synchronize all threads. */
585 	caller = current;
586 	for_each_thread(caller, thread) {
587 		/* Skip current, since it needs no changes. */
588 		if (thread == caller)
589 			continue;
590 
591 		/* Get a task reference for the new leaf node. */
592 		get_seccomp_filter(caller);
593 
594 		/*
595 		 * Drop the task reference to the shared ancestor since
596 		 * current's path will hold a reference.  (This also
597 		 * allows a put before the assignment.)
598 		 */
599 		__seccomp_filter_release(thread->seccomp.filter);
600 
601 		/* Make our new filter tree visible. */
602 		smp_store_release(&thread->seccomp.filter,
603 				  caller->seccomp.filter);
604 		atomic_set(&thread->seccomp.filter_count,
605 			   atomic_read(&thread->seccomp.filter_count));
606 
607 		/*
608 		 * Don't let an unprivileged task work around
609 		 * the no_new_privs restriction by creating
610 		 * a thread that sets it up, enters seccomp,
611 		 * then dies.
612 		 */
613 		if (task_no_new_privs(caller))
614 			task_set_no_new_privs(thread);
615 
616 		/*
617 		 * Opt the other thread into seccomp if needed.
618 		 * As threads are considered to be trust-realm
619 		 * equivalent (see ptrace_may_access), it is safe to
620 		 * allow one thread to transition the other.
621 		 */
622 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
623 			seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
624 					    flags);
625 	}
626 }
627 
628 /**
629  * seccomp_prepare_filter: Prepares a seccomp filter for use.
630  * @fprog: BPF program to install
631  *
632  * Returns filter on success or an ERR_PTR on failure.
633  */
634 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
635 {
636 	struct seccomp_filter *sfilter;
637 	int ret;
638 	const bool save_orig =
639 #if defined(CONFIG_CHECKPOINT_RESTORE) || defined(SECCOMP_ARCH_NATIVE)
640 		true;
641 #else
642 		false;
643 #endif
644 
645 	if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
646 		return ERR_PTR(-EINVAL);
647 
648 	BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
649 
650 	/*
651 	 * Installing a seccomp filter requires that the task has
652 	 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
653 	 * This avoids scenarios where unprivileged tasks can affect the
654 	 * behavior of privileged children.
655 	 */
656 	if (!task_no_new_privs(current) &&
657 			!ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN))
658 		return ERR_PTR(-EACCES);
659 
660 	/* Allocate a new seccomp_filter */
661 	sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
662 	if (!sfilter)
663 		return ERR_PTR(-ENOMEM);
664 
665 	mutex_init(&sfilter->notify_lock);
666 	ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
667 					seccomp_check_filter, save_orig);
668 	if (ret < 0) {
669 		kfree(sfilter);
670 		return ERR_PTR(ret);
671 	}
672 
673 	refcount_set(&sfilter->refs, 1);
674 	refcount_set(&sfilter->users, 1);
675 	init_waitqueue_head(&sfilter->wqh);
676 
677 	return sfilter;
678 }
679 
680 /**
681  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
682  * @user_filter: pointer to the user data containing a sock_fprog.
683  *
684  * Returns 0 on success and non-zero otherwise.
685  */
686 static struct seccomp_filter *
687 seccomp_prepare_user_filter(const char __user *user_filter)
688 {
689 	struct sock_fprog fprog;
690 	struct seccomp_filter *filter = ERR_PTR(-EFAULT);
691 
692 #ifdef CONFIG_COMPAT
693 	if (in_compat_syscall()) {
694 		struct compat_sock_fprog fprog32;
695 		if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
696 			goto out;
697 		fprog.len = fprog32.len;
698 		fprog.filter = compat_ptr(fprog32.filter);
699 	} else /* falls through to the if below. */
700 #endif
701 	if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
702 		goto out;
703 	filter = seccomp_prepare_filter(&fprog);
704 out:
705 	return filter;
706 }
707 
708 #ifdef SECCOMP_ARCH_NATIVE
709 /**
710  * seccomp_is_const_allow - check if filter is constant allow with given data
711  * @fprog: The BPF programs
712  * @sd: The seccomp data to check against, only syscall number and arch
713  *      number are considered constant.
714  */
715 static bool seccomp_is_const_allow(struct sock_fprog_kern *fprog,
716 				   struct seccomp_data *sd)
717 {
718 	unsigned int reg_value = 0;
719 	unsigned int pc;
720 	bool op_res;
721 
722 	if (WARN_ON_ONCE(!fprog))
723 		return false;
724 
725 	for (pc = 0; pc < fprog->len; pc++) {
726 		struct sock_filter *insn = &fprog->filter[pc];
727 		u16 code = insn->code;
728 		u32 k = insn->k;
729 
730 		switch (code) {
731 		case BPF_LD | BPF_W | BPF_ABS:
732 			switch (k) {
733 			case offsetof(struct seccomp_data, nr):
734 				reg_value = sd->nr;
735 				break;
736 			case offsetof(struct seccomp_data, arch):
737 				reg_value = sd->arch;
738 				break;
739 			default:
740 				/* can't optimize (non-constant value load) */
741 				return false;
742 			}
743 			break;
744 		case BPF_RET | BPF_K:
745 			/* reached return with constant values only, check allow */
746 			return k == SECCOMP_RET_ALLOW;
747 		case BPF_JMP | BPF_JA:
748 			pc += insn->k;
749 			break;
750 		case BPF_JMP | BPF_JEQ | BPF_K:
751 		case BPF_JMP | BPF_JGE | BPF_K:
752 		case BPF_JMP | BPF_JGT | BPF_K:
753 		case BPF_JMP | BPF_JSET | BPF_K:
754 			switch (BPF_OP(code)) {
755 			case BPF_JEQ:
756 				op_res = reg_value == k;
757 				break;
758 			case BPF_JGE:
759 				op_res = reg_value >= k;
760 				break;
761 			case BPF_JGT:
762 				op_res = reg_value > k;
763 				break;
764 			case BPF_JSET:
765 				op_res = !!(reg_value & k);
766 				break;
767 			default:
768 				/* can't optimize (unknown jump) */
769 				return false;
770 			}
771 
772 			pc += op_res ? insn->jt : insn->jf;
773 			break;
774 		case BPF_ALU | BPF_AND | BPF_K:
775 			reg_value &= k;
776 			break;
777 		default:
778 			/* can't optimize (unknown insn) */
779 			return false;
780 		}
781 	}
782 
783 	/* ran off the end of the filter?! */
784 	WARN_ON(1);
785 	return false;
786 }
787 
788 static void seccomp_cache_prepare_bitmap(struct seccomp_filter *sfilter,
789 					 void *bitmap, const void *bitmap_prev,
790 					 size_t bitmap_size, int arch)
791 {
792 	struct sock_fprog_kern *fprog = sfilter->prog->orig_prog;
793 	struct seccomp_data sd;
794 	int nr;
795 
796 	if (bitmap_prev) {
797 		/* The new filter must be as restrictive as the last. */
798 		bitmap_copy(bitmap, bitmap_prev, bitmap_size);
799 	} else {
800 		/* Before any filters, all syscalls are always allowed. */
801 		bitmap_fill(bitmap, bitmap_size);
802 	}
803 
804 	for (nr = 0; nr < bitmap_size; nr++) {
805 		/* No bitmap change: not a cacheable action. */
806 		if (!test_bit(nr, bitmap))
807 			continue;
808 
809 		sd.nr = nr;
810 		sd.arch = arch;
811 
812 		/* No bitmap change: continue to always allow. */
813 		if (seccomp_is_const_allow(fprog, &sd))
814 			continue;
815 
816 		/*
817 		 * Not a cacheable action: always run filters.
818 		 * atomic clear_bit() not needed, filter not visible yet.
819 		 */
820 		__clear_bit(nr, bitmap);
821 	}
822 }
823 
824 /**
825  * seccomp_cache_prepare - emulate the filter to find cacheable syscalls
826  * @sfilter: The seccomp filter
827  *
828  * Returns 0 if successful or -errno if error occurred.
829  */
830 static void seccomp_cache_prepare(struct seccomp_filter *sfilter)
831 {
832 	struct action_cache *cache = &sfilter->cache;
833 	const struct action_cache *cache_prev =
834 		sfilter->prev ? &sfilter->prev->cache : NULL;
835 
836 	seccomp_cache_prepare_bitmap(sfilter, cache->allow_native,
837 				     cache_prev ? cache_prev->allow_native : NULL,
838 				     SECCOMP_ARCH_NATIVE_NR,
839 				     SECCOMP_ARCH_NATIVE);
840 
841 #ifdef SECCOMP_ARCH_COMPAT
842 	seccomp_cache_prepare_bitmap(sfilter, cache->allow_compat,
843 				     cache_prev ? cache_prev->allow_compat : NULL,
844 				     SECCOMP_ARCH_COMPAT_NR,
845 				     SECCOMP_ARCH_COMPAT);
846 #endif /* SECCOMP_ARCH_COMPAT */
847 }
848 #endif /* SECCOMP_ARCH_NATIVE */
849 
850 /**
851  * seccomp_attach_filter: validate and attach filter
852  * @flags:  flags to change filter behavior
853  * @filter: seccomp filter to add to the current process
854  *
855  * Caller must be holding current->sighand->siglock lock.
856  *
857  * Returns 0 on success, -ve on error, or
858  *   - in TSYNC mode: the pid of a thread which was either not in the correct
859  *     seccomp mode or did not have an ancestral seccomp filter
860  *   - in NEW_LISTENER mode: the fd of the new listener
861  */
862 static long seccomp_attach_filter(unsigned int flags,
863 				  struct seccomp_filter *filter)
864 {
865 	unsigned long total_insns;
866 	struct seccomp_filter *walker;
867 
868 	assert_spin_locked(&current->sighand->siglock);
869 
870 	/* Validate resulting filter length. */
871 	total_insns = filter->prog->len;
872 	for (walker = current->seccomp.filter; walker; walker = walker->prev)
873 		total_insns += walker->prog->len + 4;  /* 4 instr penalty */
874 	if (total_insns > MAX_INSNS_PER_PATH)
875 		return -ENOMEM;
876 
877 	/* If thread sync has been requested, check that it is possible. */
878 	if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
879 		int ret;
880 
881 		ret = seccomp_can_sync_threads();
882 		if (ret) {
883 			if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
884 				return -ESRCH;
885 			else
886 				return ret;
887 		}
888 	}
889 
890 	/* Set log flag, if present. */
891 	if (flags & SECCOMP_FILTER_FLAG_LOG)
892 		filter->log = true;
893 
894 	/*
895 	 * If there is an existing filter, make it the prev and don't drop its
896 	 * task reference.
897 	 */
898 	filter->prev = current->seccomp.filter;
899 	seccomp_cache_prepare(filter);
900 	current->seccomp.filter = filter;
901 	atomic_inc(&current->seccomp.filter_count);
902 
903 	/* Now that the new filter is in place, synchronize to all threads. */
904 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
905 		seccomp_sync_threads(flags);
906 
907 	return 0;
908 }
909 
910 static void __get_seccomp_filter(struct seccomp_filter *filter)
911 {
912 	refcount_inc(&filter->refs);
913 }
914 
915 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
916 void get_seccomp_filter(struct task_struct *tsk)
917 {
918 	struct seccomp_filter *orig = tsk->seccomp.filter;
919 	if (!orig)
920 		return;
921 	__get_seccomp_filter(orig);
922 	refcount_inc(&orig->users);
923 }
924 
925 static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason)
926 {
927 	clear_siginfo(info);
928 	info->si_signo = SIGSYS;
929 	info->si_code = SYS_SECCOMP;
930 	info->si_call_addr = (void __user *)KSTK_EIP(current);
931 	info->si_errno = reason;
932 	info->si_arch = syscall_get_arch(current);
933 	info->si_syscall = syscall;
934 }
935 
936 /**
937  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
938  * @syscall: syscall number to send to userland
939  * @reason: filter-supplied reason code to send to userland (via si_errno)
940  *
941  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
942  */
943 static void seccomp_send_sigsys(int syscall, int reason)
944 {
945 	struct kernel_siginfo info;
946 	seccomp_init_siginfo(&info, syscall, reason);
947 	force_sig_info(&info);
948 }
949 #endif	/* CONFIG_SECCOMP_FILTER */
950 
951 /* For use with seccomp_actions_logged */
952 #define SECCOMP_LOG_KILL_PROCESS	(1 << 0)
953 #define SECCOMP_LOG_KILL_THREAD		(1 << 1)
954 #define SECCOMP_LOG_TRAP		(1 << 2)
955 #define SECCOMP_LOG_ERRNO		(1 << 3)
956 #define SECCOMP_LOG_TRACE		(1 << 4)
957 #define SECCOMP_LOG_LOG			(1 << 5)
958 #define SECCOMP_LOG_ALLOW		(1 << 6)
959 #define SECCOMP_LOG_USER_NOTIF		(1 << 7)
960 
961 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
962 				    SECCOMP_LOG_KILL_THREAD  |
963 				    SECCOMP_LOG_TRAP  |
964 				    SECCOMP_LOG_ERRNO |
965 				    SECCOMP_LOG_USER_NOTIF |
966 				    SECCOMP_LOG_TRACE |
967 				    SECCOMP_LOG_LOG;
968 
969 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
970 			       bool requested)
971 {
972 	bool log = false;
973 
974 	switch (action) {
975 	case SECCOMP_RET_ALLOW:
976 		break;
977 	case SECCOMP_RET_TRAP:
978 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
979 		break;
980 	case SECCOMP_RET_ERRNO:
981 		log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
982 		break;
983 	case SECCOMP_RET_TRACE:
984 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
985 		break;
986 	case SECCOMP_RET_USER_NOTIF:
987 		log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
988 		break;
989 	case SECCOMP_RET_LOG:
990 		log = seccomp_actions_logged & SECCOMP_LOG_LOG;
991 		break;
992 	case SECCOMP_RET_KILL_THREAD:
993 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
994 		break;
995 	case SECCOMP_RET_KILL_PROCESS:
996 	default:
997 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
998 	}
999 
1000 	/*
1001 	 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
1002 	 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
1003 	 * any action from being logged by removing the action name from the
1004 	 * seccomp_actions_logged sysctl.
1005 	 */
1006 	if (!log)
1007 		return;
1008 
1009 	audit_seccomp(syscall, signr, action);
1010 }
1011 
1012 /*
1013  * Secure computing mode 1 allows only read/write/exit/sigreturn.
1014  * To be fully secure this must be combined with rlimit
1015  * to limit the stack allocations too.
1016  */
1017 static const int mode1_syscalls[] = {
1018 	__NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
1019 	-1, /* negative terminated */
1020 };
1021 
1022 static void __secure_computing_strict(int this_syscall)
1023 {
1024 	const int *allowed_syscalls = mode1_syscalls;
1025 #ifdef CONFIG_COMPAT
1026 	if (in_compat_syscall())
1027 		allowed_syscalls = get_compat_mode1_syscalls();
1028 #endif
1029 	do {
1030 		if (*allowed_syscalls == this_syscall)
1031 			return;
1032 	} while (*++allowed_syscalls != -1);
1033 
1034 #ifdef SECCOMP_DEBUG
1035 	dump_stack();
1036 #endif
1037 	seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
1038 	do_exit(SIGKILL);
1039 }
1040 
1041 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
1042 void secure_computing_strict(int this_syscall)
1043 {
1044 	int mode = current->seccomp.mode;
1045 
1046 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1047 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1048 		return;
1049 
1050 	if (mode == SECCOMP_MODE_DISABLED)
1051 		return;
1052 	else if (mode == SECCOMP_MODE_STRICT)
1053 		__secure_computing_strict(this_syscall);
1054 	else
1055 		BUG();
1056 }
1057 #else
1058 
1059 #ifdef CONFIG_SECCOMP_FILTER
1060 static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
1061 {
1062 	/*
1063 	 * Note: overflow is ok here, the id just needs to be unique per
1064 	 * filter.
1065 	 */
1066 	lockdep_assert_held(&filter->notify_lock);
1067 	return filter->notif->next_id++;
1068 }
1069 
1070 static void seccomp_handle_addfd(struct seccomp_kaddfd *addfd, struct seccomp_knotif *n)
1071 {
1072 	int fd;
1073 
1074 	/*
1075 	 * Remove the notification, and reset the list pointers, indicating
1076 	 * that it has been handled.
1077 	 */
1078 	list_del_init(&addfd->list);
1079 	if (!addfd->setfd)
1080 		fd = receive_fd(addfd->file, addfd->flags);
1081 	else
1082 		fd = receive_fd_replace(addfd->fd, addfd->file, addfd->flags);
1083 	addfd->ret = fd;
1084 
1085 	if (addfd->ioctl_flags & SECCOMP_ADDFD_FLAG_SEND) {
1086 		/* If we fail reset and return an error to the notifier */
1087 		if (fd < 0) {
1088 			n->state = SECCOMP_NOTIFY_SENT;
1089 		} else {
1090 			/* Return the FD we just added */
1091 			n->flags = 0;
1092 			n->error = 0;
1093 			n->val = fd;
1094 		}
1095 	}
1096 
1097 	/*
1098 	 * Mark the notification as completed. From this point, addfd mem
1099 	 * might be invalidated and we can't safely read it anymore.
1100 	 */
1101 	complete(&addfd->completion);
1102 }
1103 
1104 static int seccomp_do_user_notification(int this_syscall,
1105 					struct seccomp_filter *match,
1106 					const struct seccomp_data *sd)
1107 {
1108 	int err;
1109 	u32 flags = 0;
1110 	long ret = 0;
1111 	struct seccomp_knotif n = {};
1112 	struct seccomp_kaddfd *addfd, *tmp;
1113 
1114 	mutex_lock(&match->notify_lock);
1115 	err = -ENOSYS;
1116 	if (!match->notif)
1117 		goto out;
1118 
1119 	n.task = current;
1120 	n.state = SECCOMP_NOTIFY_INIT;
1121 	n.data = sd;
1122 	n.id = seccomp_next_notify_id(match);
1123 	init_completion(&n.ready);
1124 	list_add(&n.list, &match->notif->notifications);
1125 	INIT_LIST_HEAD(&n.addfd);
1126 
1127 	up(&match->notif->request);
1128 	wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM);
1129 
1130 	/*
1131 	 * This is where we wait for a reply from userspace.
1132 	 */
1133 	do {
1134 		mutex_unlock(&match->notify_lock);
1135 		err = wait_for_completion_interruptible(&n.ready);
1136 		mutex_lock(&match->notify_lock);
1137 		if (err != 0)
1138 			goto interrupted;
1139 
1140 		addfd = list_first_entry_or_null(&n.addfd,
1141 						 struct seccomp_kaddfd, list);
1142 		/* Check if we were woken up by a addfd message */
1143 		if (addfd)
1144 			seccomp_handle_addfd(addfd, &n);
1145 
1146 	}  while (n.state != SECCOMP_NOTIFY_REPLIED);
1147 
1148 	ret = n.val;
1149 	err = n.error;
1150 	flags = n.flags;
1151 
1152 interrupted:
1153 	/* If there were any pending addfd calls, clear them out */
1154 	list_for_each_entry_safe(addfd, tmp, &n.addfd, list) {
1155 		/* The process went away before we got a chance to handle it */
1156 		addfd->ret = -ESRCH;
1157 		list_del_init(&addfd->list);
1158 		complete(&addfd->completion);
1159 	}
1160 
1161 	/*
1162 	 * Note that it's possible the listener died in between the time when
1163 	 * we were notified of a response (or a signal) and when we were able to
1164 	 * re-acquire the lock, so only delete from the list if the
1165 	 * notification actually exists.
1166 	 *
1167 	 * Also note that this test is only valid because there's no way to
1168 	 * *reattach* to a notifier right now. If one is added, we'll need to
1169 	 * keep track of the notif itself and make sure they match here.
1170 	 */
1171 	if (match->notif)
1172 		list_del(&n.list);
1173 out:
1174 	mutex_unlock(&match->notify_lock);
1175 
1176 	/* Userspace requests to continue the syscall. */
1177 	if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1178 		return 0;
1179 
1180 	syscall_set_return_value(current, current_pt_regs(),
1181 				 err, ret);
1182 	return -1;
1183 }
1184 
1185 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1186 			    const bool recheck_after_trace)
1187 {
1188 	u32 filter_ret, action;
1189 	struct seccomp_filter *match = NULL;
1190 	int data;
1191 	struct seccomp_data sd_local;
1192 
1193 	/*
1194 	 * Make sure that any changes to mode from another thread have
1195 	 * been seen after SYSCALL_WORK_SECCOMP was seen.
1196 	 */
1197 	smp_rmb();
1198 
1199 	if (!sd) {
1200 		populate_seccomp_data(&sd_local);
1201 		sd = &sd_local;
1202 	}
1203 
1204 	filter_ret = seccomp_run_filters(sd, &match);
1205 	data = filter_ret & SECCOMP_RET_DATA;
1206 	action = filter_ret & SECCOMP_RET_ACTION_FULL;
1207 
1208 	switch (action) {
1209 	case SECCOMP_RET_ERRNO:
1210 		/* Set low-order bits as an errno, capped at MAX_ERRNO. */
1211 		if (data > MAX_ERRNO)
1212 			data = MAX_ERRNO;
1213 		syscall_set_return_value(current, current_pt_regs(),
1214 					 -data, 0);
1215 		goto skip;
1216 
1217 	case SECCOMP_RET_TRAP:
1218 		/* Show the handler the original registers. */
1219 		syscall_rollback(current, current_pt_regs());
1220 		/* Let the filter pass back 16 bits of data. */
1221 		seccomp_send_sigsys(this_syscall, data);
1222 		goto skip;
1223 
1224 	case SECCOMP_RET_TRACE:
1225 		/* We've been put in this state by the ptracer already. */
1226 		if (recheck_after_trace)
1227 			return 0;
1228 
1229 		/* ENOSYS these calls if there is no tracer attached. */
1230 		if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
1231 			syscall_set_return_value(current,
1232 						 current_pt_regs(),
1233 						 -ENOSYS, 0);
1234 			goto skip;
1235 		}
1236 
1237 		/* Allow the BPF to provide the event message */
1238 		ptrace_event(PTRACE_EVENT_SECCOMP, data);
1239 		/*
1240 		 * The delivery of a fatal signal during event
1241 		 * notification may silently skip tracer notification,
1242 		 * which could leave us with a potentially unmodified
1243 		 * syscall that the tracer would have liked to have
1244 		 * changed. Since the process is about to die, we just
1245 		 * force the syscall to be skipped and let the signal
1246 		 * kill the process and correctly handle any tracer exit
1247 		 * notifications.
1248 		 */
1249 		if (fatal_signal_pending(current))
1250 			goto skip;
1251 		/* Check if the tracer forced the syscall to be skipped. */
1252 		this_syscall = syscall_get_nr(current, current_pt_regs());
1253 		if (this_syscall < 0)
1254 			goto skip;
1255 
1256 		/*
1257 		 * Recheck the syscall, since it may have changed. This
1258 		 * intentionally uses a NULL struct seccomp_data to force
1259 		 * a reload of all registers. This does not goto skip since
1260 		 * a skip would have already been reported.
1261 		 */
1262 		if (__seccomp_filter(this_syscall, NULL, true))
1263 			return -1;
1264 
1265 		return 0;
1266 
1267 	case SECCOMP_RET_USER_NOTIF:
1268 		if (seccomp_do_user_notification(this_syscall, match, sd))
1269 			goto skip;
1270 
1271 		return 0;
1272 
1273 	case SECCOMP_RET_LOG:
1274 		seccomp_log(this_syscall, 0, action, true);
1275 		return 0;
1276 
1277 	case SECCOMP_RET_ALLOW:
1278 		/*
1279 		 * Note that the "match" filter will always be NULL for
1280 		 * this action since SECCOMP_RET_ALLOW is the starting
1281 		 * state in seccomp_run_filters().
1282 		 */
1283 		return 0;
1284 
1285 	case SECCOMP_RET_KILL_THREAD:
1286 	case SECCOMP_RET_KILL_PROCESS:
1287 	default:
1288 		seccomp_log(this_syscall, SIGSYS, action, true);
1289 		/* Dump core only if this is the last remaining thread. */
1290 		if (action != SECCOMP_RET_KILL_THREAD ||
1291 		    get_nr_threads(current) == 1) {
1292 			kernel_siginfo_t info;
1293 
1294 			/* Show the original registers in the dump. */
1295 			syscall_rollback(current, current_pt_regs());
1296 			/* Trigger a manual coredump since do_exit skips it. */
1297 			seccomp_init_siginfo(&info, this_syscall, data);
1298 			do_coredump(&info);
1299 		}
1300 		if (action == SECCOMP_RET_KILL_THREAD)
1301 			do_exit(SIGSYS);
1302 		else
1303 			do_group_exit(SIGSYS);
1304 	}
1305 
1306 	unreachable();
1307 
1308 skip:
1309 	seccomp_log(this_syscall, 0, action, match ? match->log : false);
1310 	return -1;
1311 }
1312 #else
1313 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1314 			    const bool recheck_after_trace)
1315 {
1316 	BUG();
1317 
1318 	return -1;
1319 }
1320 #endif
1321 
1322 int __secure_computing(const struct seccomp_data *sd)
1323 {
1324 	int mode = current->seccomp.mode;
1325 	int this_syscall;
1326 
1327 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1328 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1329 		return 0;
1330 
1331 	this_syscall = sd ? sd->nr :
1332 		syscall_get_nr(current, current_pt_regs());
1333 
1334 	switch (mode) {
1335 	case SECCOMP_MODE_STRICT:
1336 		__secure_computing_strict(this_syscall);  /* may call do_exit */
1337 		return 0;
1338 	case SECCOMP_MODE_FILTER:
1339 		return __seccomp_filter(this_syscall, sd, false);
1340 	default:
1341 		BUG();
1342 	}
1343 }
1344 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1345 
1346 long prctl_get_seccomp(void)
1347 {
1348 	return current->seccomp.mode;
1349 }
1350 
1351 /**
1352  * seccomp_set_mode_strict: internal function for setting strict seccomp
1353  *
1354  * Once current->seccomp.mode is non-zero, it may not be changed.
1355  *
1356  * Returns 0 on success or -EINVAL on failure.
1357  */
1358 static long seccomp_set_mode_strict(void)
1359 {
1360 	const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
1361 	long ret = -EINVAL;
1362 
1363 	spin_lock_irq(&current->sighand->siglock);
1364 
1365 	if (!seccomp_may_assign_mode(seccomp_mode))
1366 		goto out;
1367 
1368 #ifdef TIF_NOTSC
1369 	disable_TSC();
1370 #endif
1371 	seccomp_assign_mode(current, seccomp_mode, 0);
1372 	ret = 0;
1373 
1374 out:
1375 	spin_unlock_irq(&current->sighand->siglock);
1376 
1377 	return ret;
1378 }
1379 
1380 #ifdef CONFIG_SECCOMP_FILTER
1381 static void seccomp_notify_free(struct seccomp_filter *filter)
1382 {
1383 	kfree(filter->notif);
1384 	filter->notif = NULL;
1385 }
1386 
1387 static void seccomp_notify_detach(struct seccomp_filter *filter)
1388 {
1389 	struct seccomp_knotif *knotif;
1390 
1391 	if (!filter)
1392 		return;
1393 
1394 	mutex_lock(&filter->notify_lock);
1395 
1396 	/*
1397 	 * If this file is being closed because e.g. the task who owned it
1398 	 * died, let's wake everyone up who was waiting on us.
1399 	 */
1400 	list_for_each_entry(knotif, &filter->notif->notifications, list) {
1401 		if (knotif->state == SECCOMP_NOTIFY_REPLIED)
1402 			continue;
1403 
1404 		knotif->state = SECCOMP_NOTIFY_REPLIED;
1405 		knotif->error = -ENOSYS;
1406 		knotif->val = 0;
1407 
1408 		/*
1409 		 * We do not need to wake up any pending addfd messages, as
1410 		 * the notifier will do that for us, as this just looks
1411 		 * like a standard reply.
1412 		 */
1413 		complete(&knotif->ready);
1414 	}
1415 
1416 	seccomp_notify_free(filter);
1417 	mutex_unlock(&filter->notify_lock);
1418 }
1419 
1420 static int seccomp_notify_release(struct inode *inode, struct file *file)
1421 {
1422 	struct seccomp_filter *filter = file->private_data;
1423 
1424 	seccomp_notify_detach(filter);
1425 	__put_seccomp_filter(filter);
1426 	return 0;
1427 }
1428 
1429 /* must be called with notif_lock held */
1430 static inline struct seccomp_knotif *
1431 find_notification(struct seccomp_filter *filter, u64 id)
1432 {
1433 	struct seccomp_knotif *cur;
1434 
1435 	lockdep_assert_held(&filter->notify_lock);
1436 
1437 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1438 		if (cur->id == id)
1439 			return cur;
1440 	}
1441 
1442 	return NULL;
1443 }
1444 
1445 
1446 static long seccomp_notify_recv(struct seccomp_filter *filter,
1447 				void __user *buf)
1448 {
1449 	struct seccomp_knotif *knotif = NULL, *cur;
1450 	struct seccomp_notif unotif;
1451 	ssize_t ret;
1452 
1453 	/* Verify that we're not given garbage to keep struct extensible. */
1454 	ret = check_zeroed_user(buf, sizeof(unotif));
1455 	if (ret < 0)
1456 		return ret;
1457 	if (!ret)
1458 		return -EINVAL;
1459 
1460 	memset(&unotif, 0, sizeof(unotif));
1461 
1462 	ret = down_interruptible(&filter->notif->request);
1463 	if (ret < 0)
1464 		return ret;
1465 
1466 	mutex_lock(&filter->notify_lock);
1467 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1468 		if (cur->state == SECCOMP_NOTIFY_INIT) {
1469 			knotif = cur;
1470 			break;
1471 		}
1472 	}
1473 
1474 	/*
1475 	 * If we didn't find a notification, it could be that the task was
1476 	 * interrupted by a fatal signal between the time we were woken and
1477 	 * when we were able to acquire the rw lock.
1478 	 */
1479 	if (!knotif) {
1480 		ret = -ENOENT;
1481 		goto out;
1482 	}
1483 
1484 	unotif.id = knotif->id;
1485 	unotif.pid = task_pid_vnr(knotif->task);
1486 	unotif.data = *(knotif->data);
1487 
1488 	knotif->state = SECCOMP_NOTIFY_SENT;
1489 	wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM);
1490 	ret = 0;
1491 out:
1492 	mutex_unlock(&filter->notify_lock);
1493 
1494 	if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1495 		ret = -EFAULT;
1496 
1497 		/*
1498 		 * Userspace screwed up. To make sure that we keep this
1499 		 * notification alive, let's reset it back to INIT. It
1500 		 * may have died when we released the lock, so we need to make
1501 		 * sure it's still around.
1502 		 */
1503 		mutex_lock(&filter->notify_lock);
1504 		knotif = find_notification(filter, unotif.id);
1505 		if (knotif) {
1506 			knotif->state = SECCOMP_NOTIFY_INIT;
1507 			up(&filter->notif->request);
1508 		}
1509 		mutex_unlock(&filter->notify_lock);
1510 	}
1511 
1512 	return ret;
1513 }
1514 
1515 static long seccomp_notify_send(struct seccomp_filter *filter,
1516 				void __user *buf)
1517 {
1518 	struct seccomp_notif_resp resp = {};
1519 	struct seccomp_knotif *knotif;
1520 	long ret;
1521 
1522 	if (copy_from_user(&resp, buf, sizeof(resp)))
1523 		return -EFAULT;
1524 
1525 	if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1526 		return -EINVAL;
1527 
1528 	if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) &&
1529 	    (resp.error || resp.val))
1530 		return -EINVAL;
1531 
1532 	ret = mutex_lock_interruptible(&filter->notify_lock);
1533 	if (ret < 0)
1534 		return ret;
1535 
1536 	knotif = find_notification(filter, resp.id);
1537 	if (!knotif) {
1538 		ret = -ENOENT;
1539 		goto out;
1540 	}
1541 
1542 	/* Allow exactly one reply. */
1543 	if (knotif->state != SECCOMP_NOTIFY_SENT) {
1544 		ret = -EINPROGRESS;
1545 		goto out;
1546 	}
1547 
1548 	ret = 0;
1549 	knotif->state = SECCOMP_NOTIFY_REPLIED;
1550 	knotif->error = resp.error;
1551 	knotif->val = resp.val;
1552 	knotif->flags = resp.flags;
1553 	complete(&knotif->ready);
1554 out:
1555 	mutex_unlock(&filter->notify_lock);
1556 	return ret;
1557 }
1558 
1559 static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1560 				    void __user *buf)
1561 {
1562 	struct seccomp_knotif *knotif;
1563 	u64 id;
1564 	long ret;
1565 
1566 	if (copy_from_user(&id, buf, sizeof(id)))
1567 		return -EFAULT;
1568 
1569 	ret = mutex_lock_interruptible(&filter->notify_lock);
1570 	if (ret < 0)
1571 		return ret;
1572 
1573 	knotif = find_notification(filter, id);
1574 	if (knotif && knotif->state == SECCOMP_NOTIFY_SENT)
1575 		ret = 0;
1576 	else
1577 		ret = -ENOENT;
1578 
1579 	mutex_unlock(&filter->notify_lock);
1580 	return ret;
1581 }
1582 
1583 static long seccomp_notify_addfd(struct seccomp_filter *filter,
1584 				 struct seccomp_notif_addfd __user *uaddfd,
1585 				 unsigned int size)
1586 {
1587 	struct seccomp_notif_addfd addfd;
1588 	struct seccomp_knotif *knotif;
1589 	struct seccomp_kaddfd kaddfd;
1590 	int ret;
1591 
1592 	BUILD_BUG_ON(sizeof(addfd) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0);
1593 	BUILD_BUG_ON(sizeof(addfd) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST);
1594 
1595 	if (size < SECCOMP_NOTIFY_ADDFD_SIZE_VER0 || size >= PAGE_SIZE)
1596 		return -EINVAL;
1597 
1598 	ret = copy_struct_from_user(&addfd, sizeof(addfd), uaddfd, size);
1599 	if (ret)
1600 		return ret;
1601 
1602 	if (addfd.newfd_flags & ~O_CLOEXEC)
1603 		return -EINVAL;
1604 
1605 	if (addfd.flags & ~(SECCOMP_ADDFD_FLAG_SETFD | SECCOMP_ADDFD_FLAG_SEND))
1606 		return -EINVAL;
1607 
1608 	if (addfd.newfd && !(addfd.flags & SECCOMP_ADDFD_FLAG_SETFD))
1609 		return -EINVAL;
1610 
1611 	kaddfd.file = fget(addfd.srcfd);
1612 	if (!kaddfd.file)
1613 		return -EBADF;
1614 
1615 	kaddfd.ioctl_flags = addfd.flags;
1616 	kaddfd.flags = addfd.newfd_flags;
1617 	kaddfd.setfd = addfd.flags & SECCOMP_ADDFD_FLAG_SETFD;
1618 	kaddfd.fd = addfd.newfd;
1619 	init_completion(&kaddfd.completion);
1620 
1621 	ret = mutex_lock_interruptible(&filter->notify_lock);
1622 	if (ret < 0)
1623 		goto out;
1624 
1625 	knotif = find_notification(filter, addfd.id);
1626 	if (!knotif) {
1627 		ret = -ENOENT;
1628 		goto out_unlock;
1629 	}
1630 
1631 	/*
1632 	 * We do not want to allow for FD injection to occur before the
1633 	 * notification has been picked up by a userspace handler, or after
1634 	 * the notification has been replied to.
1635 	 */
1636 	if (knotif->state != SECCOMP_NOTIFY_SENT) {
1637 		ret = -EINPROGRESS;
1638 		goto out_unlock;
1639 	}
1640 
1641 	if (addfd.flags & SECCOMP_ADDFD_FLAG_SEND) {
1642 		/*
1643 		 * Disallow queuing an atomic addfd + send reply while there are
1644 		 * some addfd requests still to process.
1645 		 *
1646 		 * There is no clear reason to support it and allows us to keep
1647 		 * the loop on the other side straight-forward.
1648 		 */
1649 		if (!list_empty(&knotif->addfd)) {
1650 			ret = -EBUSY;
1651 			goto out_unlock;
1652 		}
1653 
1654 		/* Allow exactly only one reply */
1655 		knotif->state = SECCOMP_NOTIFY_REPLIED;
1656 	}
1657 
1658 	list_add(&kaddfd.list, &knotif->addfd);
1659 	complete(&knotif->ready);
1660 	mutex_unlock(&filter->notify_lock);
1661 
1662 	/* Now we wait for it to be processed or be interrupted */
1663 	ret = wait_for_completion_interruptible(&kaddfd.completion);
1664 	if (ret == 0) {
1665 		/*
1666 		 * We had a successful completion. The other side has already
1667 		 * removed us from the addfd queue, and
1668 		 * wait_for_completion_interruptible has a memory barrier upon
1669 		 * success that lets us read this value directly without
1670 		 * locking.
1671 		 */
1672 		ret = kaddfd.ret;
1673 		goto out;
1674 	}
1675 
1676 	mutex_lock(&filter->notify_lock);
1677 	/*
1678 	 * Even though we were woken up by a signal and not a successful
1679 	 * completion, a completion may have happened in the mean time.
1680 	 *
1681 	 * We need to check again if the addfd request has been handled,
1682 	 * and if not, we will remove it from the queue.
1683 	 */
1684 	if (list_empty(&kaddfd.list))
1685 		ret = kaddfd.ret;
1686 	else
1687 		list_del(&kaddfd.list);
1688 
1689 out_unlock:
1690 	mutex_unlock(&filter->notify_lock);
1691 out:
1692 	fput(kaddfd.file);
1693 
1694 	return ret;
1695 }
1696 
1697 static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1698 				 unsigned long arg)
1699 {
1700 	struct seccomp_filter *filter = file->private_data;
1701 	void __user *buf = (void __user *)arg;
1702 
1703 	/* Fixed-size ioctls */
1704 	switch (cmd) {
1705 	case SECCOMP_IOCTL_NOTIF_RECV:
1706 		return seccomp_notify_recv(filter, buf);
1707 	case SECCOMP_IOCTL_NOTIF_SEND:
1708 		return seccomp_notify_send(filter, buf);
1709 	case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR:
1710 	case SECCOMP_IOCTL_NOTIF_ID_VALID:
1711 		return seccomp_notify_id_valid(filter, buf);
1712 	}
1713 
1714 	/* Extensible Argument ioctls */
1715 #define EA_IOCTL(cmd)	((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1716 	switch (EA_IOCTL(cmd)) {
1717 	case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD):
1718 		return seccomp_notify_addfd(filter, buf, _IOC_SIZE(cmd));
1719 	default:
1720 		return -EINVAL;
1721 	}
1722 }
1723 
1724 static __poll_t seccomp_notify_poll(struct file *file,
1725 				    struct poll_table_struct *poll_tab)
1726 {
1727 	struct seccomp_filter *filter = file->private_data;
1728 	__poll_t ret = 0;
1729 	struct seccomp_knotif *cur;
1730 
1731 	poll_wait(file, &filter->wqh, poll_tab);
1732 
1733 	if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1734 		return EPOLLERR;
1735 
1736 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1737 		if (cur->state == SECCOMP_NOTIFY_INIT)
1738 			ret |= EPOLLIN | EPOLLRDNORM;
1739 		if (cur->state == SECCOMP_NOTIFY_SENT)
1740 			ret |= EPOLLOUT | EPOLLWRNORM;
1741 		if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1742 			break;
1743 	}
1744 
1745 	mutex_unlock(&filter->notify_lock);
1746 
1747 	if (refcount_read(&filter->users) == 0)
1748 		ret |= EPOLLHUP;
1749 
1750 	return ret;
1751 }
1752 
1753 static const struct file_operations seccomp_notify_ops = {
1754 	.poll = seccomp_notify_poll,
1755 	.release = seccomp_notify_release,
1756 	.unlocked_ioctl = seccomp_notify_ioctl,
1757 	.compat_ioctl = seccomp_notify_ioctl,
1758 };
1759 
1760 static struct file *init_listener(struct seccomp_filter *filter)
1761 {
1762 	struct file *ret;
1763 
1764 	ret = ERR_PTR(-ENOMEM);
1765 	filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1766 	if (!filter->notif)
1767 		goto out;
1768 
1769 	sema_init(&filter->notif->request, 0);
1770 	filter->notif->next_id = get_random_u64();
1771 	INIT_LIST_HEAD(&filter->notif->notifications);
1772 
1773 	ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1774 				 filter, O_RDWR);
1775 	if (IS_ERR(ret))
1776 		goto out_notif;
1777 
1778 	/* The file has a reference to it now */
1779 	__get_seccomp_filter(filter);
1780 
1781 out_notif:
1782 	if (IS_ERR(ret))
1783 		seccomp_notify_free(filter);
1784 out:
1785 	return ret;
1786 }
1787 
1788 /*
1789  * Does @new_child have a listener while an ancestor also has a listener?
1790  * If so, we'll want to reject this filter.
1791  * This only has to be tested for the current process, even in the TSYNC case,
1792  * because TSYNC installs @child with the same parent on all threads.
1793  * Note that @new_child is not hooked up to its parent at this point yet, so
1794  * we use current->seccomp.filter.
1795  */
1796 static bool has_duplicate_listener(struct seccomp_filter *new_child)
1797 {
1798 	struct seccomp_filter *cur;
1799 
1800 	/* must be protected against concurrent TSYNC */
1801 	lockdep_assert_held(&current->sighand->siglock);
1802 
1803 	if (!new_child->notif)
1804 		return false;
1805 	for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1806 		if (cur->notif)
1807 			return true;
1808 	}
1809 
1810 	return false;
1811 }
1812 
1813 /**
1814  * seccomp_set_mode_filter: internal function for setting seccomp filter
1815  * @flags:  flags to change filter behavior
1816  * @filter: struct sock_fprog containing filter
1817  *
1818  * This function may be called repeatedly to install additional filters.
1819  * Every filter successfully installed will be evaluated (in reverse order)
1820  * for each system call the task makes.
1821  *
1822  * Once current->seccomp.mode is non-zero, it may not be changed.
1823  *
1824  * Returns 0 on success or -EINVAL on failure.
1825  */
1826 static long seccomp_set_mode_filter(unsigned int flags,
1827 				    const char __user *filter)
1828 {
1829 	const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1830 	struct seccomp_filter *prepared = NULL;
1831 	long ret = -EINVAL;
1832 	int listener = -1;
1833 	struct file *listener_f = NULL;
1834 
1835 	/* Validate flags. */
1836 	if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1837 		return -EINVAL;
1838 
1839 	/*
1840 	 * In the successful case, NEW_LISTENER returns the new listener fd.
1841 	 * But in the failure case, TSYNC returns the thread that died. If you
1842 	 * combine these two flags, there's no way to tell whether something
1843 	 * succeeded or failed. So, let's disallow this combination if the user
1844 	 * has not explicitly requested no errors from TSYNC.
1845 	 */
1846 	if ((flags & SECCOMP_FILTER_FLAG_TSYNC) &&
1847 	    (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) &&
1848 	    ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0))
1849 		return -EINVAL;
1850 
1851 	/* Prepare the new filter before holding any locks. */
1852 	prepared = seccomp_prepare_user_filter(filter);
1853 	if (IS_ERR(prepared))
1854 		return PTR_ERR(prepared);
1855 
1856 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1857 		listener = get_unused_fd_flags(O_CLOEXEC);
1858 		if (listener < 0) {
1859 			ret = listener;
1860 			goto out_free;
1861 		}
1862 
1863 		listener_f = init_listener(prepared);
1864 		if (IS_ERR(listener_f)) {
1865 			put_unused_fd(listener);
1866 			ret = PTR_ERR(listener_f);
1867 			goto out_free;
1868 		}
1869 	}
1870 
1871 	/*
1872 	 * Make sure we cannot change seccomp or nnp state via TSYNC
1873 	 * while another thread is in the middle of calling exec.
1874 	 */
1875 	if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1876 	    mutex_lock_killable(&current->signal->cred_guard_mutex))
1877 		goto out_put_fd;
1878 
1879 	spin_lock_irq(&current->sighand->siglock);
1880 
1881 	if (!seccomp_may_assign_mode(seccomp_mode))
1882 		goto out;
1883 
1884 	if (has_duplicate_listener(prepared)) {
1885 		ret = -EBUSY;
1886 		goto out;
1887 	}
1888 
1889 	ret = seccomp_attach_filter(flags, prepared);
1890 	if (ret)
1891 		goto out;
1892 	/* Do not free the successfully attached filter. */
1893 	prepared = NULL;
1894 
1895 	seccomp_assign_mode(current, seccomp_mode, flags);
1896 out:
1897 	spin_unlock_irq(&current->sighand->siglock);
1898 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
1899 		mutex_unlock(&current->signal->cred_guard_mutex);
1900 out_put_fd:
1901 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1902 		if (ret) {
1903 			listener_f->private_data = NULL;
1904 			fput(listener_f);
1905 			put_unused_fd(listener);
1906 			seccomp_notify_detach(prepared);
1907 		} else {
1908 			fd_install(listener, listener_f);
1909 			ret = listener;
1910 		}
1911 	}
1912 out_free:
1913 	seccomp_filter_free(prepared);
1914 	return ret;
1915 }
1916 #else
1917 static inline long seccomp_set_mode_filter(unsigned int flags,
1918 					   const char __user *filter)
1919 {
1920 	return -EINVAL;
1921 }
1922 #endif
1923 
1924 static long seccomp_get_action_avail(const char __user *uaction)
1925 {
1926 	u32 action;
1927 
1928 	if (copy_from_user(&action, uaction, sizeof(action)))
1929 		return -EFAULT;
1930 
1931 	switch (action) {
1932 	case SECCOMP_RET_KILL_PROCESS:
1933 	case SECCOMP_RET_KILL_THREAD:
1934 	case SECCOMP_RET_TRAP:
1935 	case SECCOMP_RET_ERRNO:
1936 	case SECCOMP_RET_USER_NOTIF:
1937 	case SECCOMP_RET_TRACE:
1938 	case SECCOMP_RET_LOG:
1939 	case SECCOMP_RET_ALLOW:
1940 		break;
1941 	default:
1942 		return -EOPNOTSUPP;
1943 	}
1944 
1945 	return 0;
1946 }
1947 
1948 static long seccomp_get_notif_sizes(void __user *usizes)
1949 {
1950 	struct seccomp_notif_sizes sizes = {
1951 		.seccomp_notif = sizeof(struct seccomp_notif),
1952 		.seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
1953 		.seccomp_data = sizeof(struct seccomp_data),
1954 	};
1955 
1956 	if (copy_to_user(usizes, &sizes, sizeof(sizes)))
1957 		return -EFAULT;
1958 
1959 	return 0;
1960 }
1961 
1962 /* Common entry point for both prctl and syscall. */
1963 static long do_seccomp(unsigned int op, unsigned int flags,
1964 		       void __user *uargs)
1965 {
1966 	switch (op) {
1967 	case SECCOMP_SET_MODE_STRICT:
1968 		if (flags != 0 || uargs != NULL)
1969 			return -EINVAL;
1970 		return seccomp_set_mode_strict();
1971 	case SECCOMP_SET_MODE_FILTER:
1972 		return seccomp_set_mode_filter(flags, uargs);
1973 	case SECCOMP_GET_ACTION_AVAIL:
1974 		if (flags != 0)
1975 			return -EINVAL;
1976 
1977 		return seccomp_get_action_avail(uargs);
1978 	case SECCOMP_GET_NOTIF_SIZES:
1979 		if (flags != 0)
1980 			return -EINVAL;
1981 
1982 		return seccomp_get_notif_sizes(uargs);
1983 	default:
1984 		return -EINVAL;
1985 	}
1986 }
1987 
1988 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
1989 			 void __user *, uargs)
1990 {
1991 	return do_seccomp(op, flags, uargs);
1992 }
1993 
1994 /**
1995  * prctl_set_seccomp: configures current->seccomp.mode
1996  * @seccomp_mode: requested mode to use
1997  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1998  *
1999  * Returns 0 on success or -EINVAL on failure.
2000  */
2001 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
2002 {
2003 	unsigned int op;
2004 	void __user *uargs;
2005 
2006 	switch (seccomp_mode) {
2007 	case SECCOMP_MODE_STRICT:
2008 		op = SECCOMP_SET_MODE_STRICT;
2009 		/*
2010 		 * Setting strict mode through prctl always ignored filter,
2011 		 * so make sure it is always NULL here to pass the internal
2012 		 * check in do_seccomp().
2013 		 */
2014 		uargs = NULL;
2015 		break;
2016 	case SECCOMP_MODE_FILTER:
2017 		op = SECCOMP_SET_MODE_FILTER;
2018 		uargs = filter;
2019 		break;
2020 	default:
2021 		return -EINVAL;
2022 	}
2023 
2024 	/* prctl interface doesn't have flags, so they are always zero. */
2025 	return do_seccomp(op, 0, uargs);
2026 }
2027 
2028 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
2029 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
2030 					     unsigned long filter_off)
2031 {
2032 	struct seccomp_filter *orig, *filter;
2033 	unsigned long count;
2034 
2035 	/*
2036 	 * Note: this is only correct because the caller should be the (ptrace)
2037 	 * tracer of the task, otherwise lock_task_sighand is needed.
2038 	 */
2039 	spin_lock_irq(&task->sighand->siglock);
2040 
2041 	if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
2042 		spin_unlock_irq(&task->sighand->siglock);
2043 		return ERR_PTR(-EINVAL);
2044 	}
2045 
2046 	orig = task->seccomp.filter;
2047 	__get_seccomp_filter(orig);
2048 	spin_unlock_irq(&task->sighand->siglock);
2049 
2050 	count = 0;
2051 	for (filter = orig; filter; filter = filter->prev)
2052 		count++;
2053 
2054 	if (filter_off >= count) {
2055 		filter = ERR_PTR(-ENOENT);
2056 		goto out;
2057 	}
2058 
2059 	count -= filter_off;
2060 	for (filter = orig; filter && count > 1; filter = filter->prev)
2061 		count--;
2062 
2063 	if (WARN_ON(count != 1 || !filter)) {
2064 		filter = ERR_PTR(-ENOENT);
2065 		goto out;
2066 	}
2067 
2068 	__get_seccomp_filter(filter);
2069 
2070 out:
2071 	__put_seccomp_filter(orig);
2072 	return filter;
2073 }
2074 
2075 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
2076 			void __user *data)
2077 {
2078 	struct seccomp_filter *filter;
2079 	struct sock_fprog_kern *fprog;
2080 	long ret;
2081 
2082 	if (!capable(CAP_SYS_ADMIN) ||
2083 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
2084 		return -EACCES;
2085 	}
2086 
2087 	filter = get_nth_filter(task, filter_off);
2088 	if (IS_ERR(filter))
2089 		return PTR_ERR(filter);
2090 
2091 	fprog = filter->prog->orig_prog;
2092 	if (!fprog) {
2093 		/* This must be a new non-cBPF filter, since we save
2094 		 * every cBPF filter's orig_prog above when
2095 		 * CONFIG_CHECKPOINT_RESTORE is enabled.
2096 		 */
2097 		ret = -EMEDIUMTYPE;
2098 		goto out;
2099 	}
2100 
2101 	ret = fprog->len;
2102 	if (!data)
2103 		goto out;
2104 
2105 	if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
2106 		ret = -EFAULT;
2107 
2108 out:
2109 	__put_seccomp_filter(filter);
2110 	return ret;
2111 }
2112 
2113 long seccomp_get_metadata(struct task_struct *task,
2114 			  unsigned long size, void __user *data)
2115 {
2116 	long ret;
2117 	struct seccomp_filter *filter;
2118 	struct seccomp_metadata kmd = {};
2119 
2120 	if (!capable(CAP_SYS_ADMIN) ||
2121 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
2122 		return -EACCES;
2123 	}
2124 
2125 	size = min_t(unsigned long, size, sizeof(kmd));
2126 
2127 	if (size < sizeof(kmd.filter_off))
2128 		return -EINVAL;
2129 
2130 	if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
2131 		return -EFAULT;
2132 
2133 	filter = get_nth_filter(task, kmd.filter_off);
2134 	if (IS_ERR(filter))
2135 		return PTR_ERR(filter);
2136 
2137 	if (filter->log)
2138 		kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
2139 
2140 	ret = size;
2141 	if (copy_to_user(data, &kmd, size))
2142 		ret = -EFAULT;
2143 
2144 	__put_seccomp_filter(filter);
2145 	return ret;
2146 }
2147 #endif
2148 
2149 #ifdef CONFIG_SYSCTL
2150 
2151 /* Human readable action names for friendly sysctl interaction */
2152 #define SECCOMP_RET_KILL_PROCESS_NAME	"kill_process"
2153 #define SECCOMP_RET_KILL_THREAD_NAME	"kill_thread"
2154 #define SECCOMP_RET_TRAP_NAME		"trap"
2155 #define SECCOMP_RET_ERRNO_NAME		"errno"
2156 #define SECCOMP_RET_USER_NOTIF_NAME	"user_notif"
2157 #define SECCOMP_RET_TRACE_NAME		"trace"
2158 #define SECCOMP_RET_LOG_NAME		"log"
2159 #define SECCOMP_RET_ALLOW_NAME		"allow"
2160 
2161 static const char seccomp_actions_avail[] =
2162 				SECCOMP_RET_KILL_PROCESS_NAME	" "
2163 				SECCOMP_RET_KILL_THREAD_NAME	" "
2164 				SECCOMP_RET_TRAP_NAME		" "
2165 				SECCOMP_RET_ERRNO_NAME		" "
2166 				SECCOMP_RET_USER_NOTIF_NAME     " "
2167 				SECCOMP_RET_TRACE_NAME		" "
2168 				SECCOMP_RET_LOG_NAME		" "
2169 				SECCOMP_RET_ALLOW_NAME;
2170 
2171 struct seccomp_log_name {
2172 	u32		log;
2173 	const char	*name;
2174 };
2175 
2176 static const struct seccomp_log_name seccomp_log_names[] = {
2177 	{ SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
2178 	{ SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
2179 	{ SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
2180 	{ SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
2181 	{ SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
2182 	{ SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
2183 	{ SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
2184 	{ SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
2185 	{ }
2186 };
2187 
2188 static bool seccomp_names_from_actions_logged(char *names, size_t size,
2189 					      u32 actions_logged,
2190 					      const char *sep)
2191 {
2192 	const struct seccomp_log_name *cur;
2193 	bool append_sep = false;
2194 
2195 	for (cur = seccomp_log_names; cur->name && size; cur++) {
2196 		ssize_t ret;
2197 
2198 		if (!(actions_logged & cur->log))
2199 			continue;
2200 
2201 		if (append_sep) {
2202 			ret = strscpy(names, sep, size);
2203 			if (ret < 0)
2204 				return false;
2205 
2206 			names += ret;
2207 			size -= ret;
2208 		} else
2209 			append_sep = true;
2210 
2211 		ret = strscpy(names, cur->name, size);
2212 		if (ret < 0)
2213 			return false;
2214 
2215 		names += ret;
2216 		size -= ret;
2217 	}
2218 
2219 	return true;
2220 }
2221 
2222 static bool seccomp_action_logged_from_name(u32 *action_logged,
2223 					    const char *name)
2224 {
2225 	const struct seccomp_log_name *cur;
2226 
2227 	for (cur = seccomp_log_names; cur->name; cur++) {
2228 		if (!strcmp(cur->name, name)) {
2229 			*action_logged = cur->log;
2230 			return true;
2231 		}
2232 	}
2233 
2234 	return false;
2235 }
2236 
2237 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
2238 {
2239 	char *name;
2240 
2241 	*actions_logged = 0;
2242 	while ((name = strsep(&names, " ")) && *name) {
2243 		u32 action_logged = 0;
2244 
2245 		if (!seccomp_action_logged_from_name(&action_logged, name))
2246 			return false;
2247 
2248 		*actions_logged |= action_logged;
2249 	}
2250 
2251 	return true;
2252 }
2253 
2254 static int read_actions_logged(struct ctl_table *ro_table, void *buffer,
2255 			       size_t *lenp, loff_t *ppos)
2256 {
2257 	char names[sizeof(seccomp_actions_avail)];
2258 	struct ctl_table table;
2259 
2260 	memset(names, 0, sizeof(names));
2261 
2262 	if (!seccomp_names_from_actions_logged(names, sizeof(names),
2263 					       seccomp_actions_logged, " "))
2264 		return -EINVAL;
2265 
2266 	table = *ro_table;
2267 	table.data = names;
2268 	table.maxlen = sizeof(names);
2269 	return proc_dostring(&table, 0, buffer, lenp, ppos);
2270 }
2271 
2272 static int write_actions_logged(struct ctl_table *ro_table, void *buffer,
2273 				size_t *lenp, loff_t *ppos, u32 *actions_logged)
2274 {
2275 	char names[sizeof(seccomp_actions_avail)];
2276 	struct ctl_table table;
2277 	int ret;
2278 
2279 	if (!capable(CAP_SYS_ADMIN))
2280 		return -EPERM;
2281 
2282 	memset(names, 0, sizeof(names));
2283 
2284 	table = *ro_table;
2285 	table.data = names;
2286 	table.maxlen = sizeof(names);
2287 	ret = proc_dostring(&table, 1, buffer, lenp, ppos);
2288 	if (ret)
2289 		return ret;
2290 
2291 	if (!seccomp_actions_logged_from_names(actions_logged, table.data))
2292 		return -EINVAL;
2293 
2294 	if (*actions_logged & SECCOMP_LOG_ALLOW)
2295 		return -EINVAL;
2296 
2297 	seccomp_actions_logged = *actions_logged;
2298 	return 0;
2299 }
2300 
2301 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
2302 				 int ret)
2303 {
2304 	char names[sizeof(seccomp_actions_avail)];
2305 	char old_names[sizeof(seccomp_actions_avail)];
2306 	const char *new = names;
2307 	const char *old = old_names;
2308 
2309 	if (!audit_enabled)
2310 		return;
2311 
2312 	memset(names, 0, sizeof(names));
2313 	memset(old_names, 0, sizeof(old_names));
2314 
2315 	if (ret)
2316 		new = "?";
2317 	else if (!actions_logged)
2318 		new = "(none)";
2319 	else if (!seccomp_names_from_actions_logged(names, sizeof(names),
2320 						    actions_logged, ","))
2321 		new = "?";
2322 
2323 	if (!old_actions_logged)
2324 		old = "(none)";
2325 	else if (!seccomp_names_from_actions_logged(old_names,
2326 						    sizeof(old_names),
2327 						    old_actions_logged, ","))
2328 		old = "?";
2329 
2330 	return audit_seccomp_actions_logged(new, old, !ret);
2331 }
2332 
2333 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
2334 					  void *buffer, size_t *lenp,
2335 					  loff_t *ppos)
2336 {
2337 	int ret;
2338 
2339 	if (write) {
2340 		u32 actions_logged = 0;
2341 		u32 old_actions_logged = seccomp_actions_logged;
2342 
2343 		ret = write_actions_logged(ro_table, buffer, lenp, ppos,
2344 					   &actions_logged);
2345 		audit_actions_logged(actions_logged, old_actions_logged, ret);
2346 	} else
2347 		ret = read_actions_logged(ro_table, buffer, lenp, ppos);
2348 
2349 	return ret;
2350 }
2351 
2352 static struct ctl_path seccomp_sysctl_path[] = {
2353 	{ .procname = "kernel", },
2354 	{ .procname = "seccomp", },
2355 	{ }
2356 };
2357 
2358 static struct ctl_table seccomp_sysctl_table[] = {
2359 	{
2360 		.procname	= "actions_avail",
2361 		.data		= (void *) &seccomp_actions_avail,
2362 		.maxlen		= sizeof(seccomp_actions_avail),
2363 		.mode		= 0444,
2364 		.proc_handler	= proc_dostring,
2365 	},
2366 	{
2367 		.procname	= "actions_logged",
2368 		.mode		= 0644,
2369 		.proc_handler	= seccomp_actions_logged_handler,
2370 	},
2371 	{ }
2372 };
2373 
2374 static int __init seccomp_sysctl_init(void)
2375 {
2376 	struct ctl_table_header *hdr;
2377 
2378 	hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
2379 	if (!hdr)
2380 		pr_warn("sysctl registration failed\n");
2381 	else
2382 		kmemleak_not_leak(hdr);
2383 
2384 	return 0;
2385 }
2386 
2387 device_initcall(seccomp_sysctl_init)
2388 
2389 #endif /* CONFIG_SYSCTL */
2390 
2391 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
2392 /* Currently CONFIG_SECCOMP_CACHE_DEBUG implies SECCOMP_ARCH_NATIVE */
2393 static void proc_pid_seccomp_cache_arch(struct seq_file *m, const char *name,
2394 					const void *bitmap, size_t bitmap_size)
2395 {
2396 	int nr;
2397 
2398 	for (nr = 0; nr < bitmap_size; nr++) {
2399 		bool cached = test_bit(nr, bitmap);
2400 		char *status = cached ? "ALLOW" : "FILTER";
2401 
2402 		seq_printf(m, "%s %d %s\n", name, nr, status);
2403 	}
2404 }
2405 
2406 int proc_pid_seccomp_cache(struct seq_file *m, struct pid_namespace *ns,
2407 			   struct pid *pid, struct task_struct *task)
2408 {
2409 	struct seccomp_filter *f;
2410 	unsigned long flags;
2411 
2412 	/*
2413 	 * We don't want some sandboxed process to know what their seccomp
2414 	 * filters consist of.
2415 	 */
2416 	if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
2417 		return -EACCES;
2418 
2419 	if (!lock_task_sighand(task, &flags))
2420 		return -ESRCH;
2421 
2422 	f = READ_ONCE(task->seccomp.filter);
2423 	if (!f) {
2424 		unlock_task_sighand(task, &flags);
2425 		return 0;
2426 	}
2427 
2428 	/* prevent filter from being freed while we are printing it */
2429 	__get_seccomp_filter(f);
2430 	unlock_task_sighand(task, &flags);
2431 
2432 	proc_pid_seccomp_cache_arch(m, SECCOMP_ARCH_NATIVE_NAME,
2433 				    f->cache.allow_native,
2434 				    SECCOMP_ARCH_NATIVE_NR);
2435 
2436 #ifdef SECCOMP_ARCH_COMPAT
2437 	proc_pid_seccomp_cache_arch(m, SECCOMP_ARCH_COMPAT_NAME,
2438 				    f->cache.allow_compat,
2439 				    SECCOMP_ARCH_COMPAT_NR);
2440 #endif /* SECCOMP_ARCH_COMPAT */
2441 
2442 	__put_seccomp_filter(f);
2443 	return 0;
2444 }
2445 #endif /* CONFIG_SECCOMP_CACHE_DEBUG */
2446