xref: /linux-6.15/kernel/seccomp.c (revision 95036a79)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/kernel/seccomp.c
4  *
5  * Copyright 2004-2005  Andrea Arcangeli <[email protected]>
6  *
7  * Copyright (C) 2012 Google, Inc.
8  * Will Drewry <[email protected]>
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 /* Not exposed in headers: strictly internal use only. */
33 #define SECCOMP_MODE_DEAD	(SECCOMP_MODE_FILTER + 1)
34 
35 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
36 #include <asm/syscall.h>
37 #endif
38 
39 #ifdef CONFIG_SECCOMP_FILTER
40 #include <linux/file.h>
41 #include <linux/filter.h>
42 #include <linux/pid.h>
43 #include <linux/ptrace.h>
44 #include <linux/capability.h>
45 #include <linux/uaccess.h>
46 #include <linux/anon_inodes.h>
47 #include <linux/lockdep.h>
48 
49 /*
50  * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the
51  * wrong direction flag in the ioctl number. This is the broken one,
52  * which the kernel needs to keep supporting until all userspaces stop
53  * using the wrong command number.
54  */
55 #define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR	SECCOMP_IOR(2, __u64)
56 
57 enum notify_state {
58 	SECCOMP_NOTIFY_INIT,
59 	SECCOMP_NOTIFY_SENT,
60 	SECCOMP_NOTIFY_REPLIED,
61 };
62 
63 struct seccomp_knotif {
64 	/* The struct pid of the task whose filter triggered the notification */
65 	struct task_struct *task;
66 
67 	/* The "cookie" for this request; this is unique for this filter. */
68 	u64 id;
69 
70 	/*
71 	 * The seccomp data. This pointer is valid the entire time this
72 	 * notification is active, since it comes from __seccomp_filter which
73 	 * eclipses the entire lifecycle here.
74 	 */
75 	const struct seccomp_data *data;
76 
77 	/*
78 	 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
79 	 * struct seccomp_knotif is created and starts out in INIT. Once the
80 	 * handler reads the notification off of an FD, it transitions to SENT.
81 	 * If a signal is received the state transitions back to INIT and
82 	 * another message is sent. When the userspace handler replies, state
83 	 * transitions to REPLIED.
84 	 */
85 	enum notify_state state;
86 
87 	/* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
88 	int error;
89 	long val;
90 	u32 flags;
91 
92 	/*
93 	 * Signals when this has changed states, such as the listener
94 	 * dying, a new seccomp addfd message, or changing to REPLIED
95 	 */
96 	struct completion ready;
97 
98 	struct list_head list;
99 
100 	/* outstanding addfd requests */
101 	struct list_head addfd;
102 };
103 
104 /**
105  * struct seccomp_kaddfd - container for seccomp_addfd ioctl messages
106  *
107  * @file: A reference to the file to install in the other task
108  * @fd: The fd number to install it at. If the fd number is -1, it means the
109  *      installing process should allocate the fd as normal.
110  * @flags: The flags for the new file descriptor. At the moment, only O_CLOEXEC
111  *         is allowed.
112  * @ioctl_flags: The flags used for the seccomp_addfd ioctl.
113  * @setfd: whether or not SECCOMP_ADDFD_FLAG_SETFD was set during notify_addfd
114  * @ret: The return value of the installing process. It is set to the fd num
115  *       upon success (>= 0).
116  * @completion: Indicates that the installing process has completed fd
117  *              installation, or gone away (either due to successful
118  *              reply, or signal)
119  * @list: list_head for chaining seccomp_kaddfd together.
120  *
121  */
122 struct seccomp_kaddfd {
123 	struct file *file;
124 	int fd;
125 	unsigned int flags;
126 	__u32 ioctl_flags;
127 
128 	union {
129 		bool setfd;
130 		/* To only be set on reply */
131 		int ret;
132 	};
133 	struct completion completion;
134 	struct list_head list;
135 };
136 
137 /**
138  * struct notification - container for seccomp userspace notifications. Since
139  * most seccomp filters will not have notification listeners attached and this
140  * structure is fairly large, we store the notification-specific stuff in a
141  * separate structure.
142  *
143  * @requests: A semaphore that users of this notification can wait on for
144  *            changes. Actual reads and writes are still controlled with
145  *            filter->notify_lock.
146  * @flags: A set of SECCOMP_USER_NOTIF_FD_* flags.
147  * @next_id: The id of the next request.
148  * @notifications: A list of struct seccomp_knotif elements.
149  */
150 
151 struct notification {
152 	atomic_t requests;
153 	u32 flags;
154 	u64 next_id;
155 	struct list_head notifications;
156 };
157 
158 #ifdef SECCOMP_ARCH_NATIVE
159 /**
160  * struct action_cache - per-filter cache of seccomp actions per
161  * arch/syscall pair
162  *
163  * @allow_native: A bitmap where each bit represents whether the
164  *		  filter will always allow the syscall, for the
165  *		  native architecture.
166  * @allow_compat: A bitmap where each bit represents whether the
167  *		  filter will always allow the syscall, for the
168  *		  compat architecture.
169  */
170 struct action_cache {
171 	DECLARE_BITMAP(allow_native, SECCOMP_ARCH_NATIVE_NR);
172 #ifdef SECCOMP_ARCH_COMPAT
173 	DECLARE_BITMAP(allow_compat, SECCOMP_ARCH_COMPAT_NR);
174 #endif
175 };
176 #else
177 struct action_cache { };
178 
179 static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilter,
180 					     const struct seccomp_data *sd)
181 {
182 	return false;
183 }
184 
185 static inline void seccomp_cache_prepare(struct seccomp_filter *sfilter)
186 {
187 }
188 #endif /* SECCOMP_ARCH_NATIVE */
189 
190 /**
191  * struct seccomp_filter - container for seccomp BPF programs
192  *
193  * @refs: Reference count to manage the object lifetime.
194  *	  A filter's reference count is incremented for each directly
195  *	  attached task, once for the dependent filter, and if
196  *	  requested for the user notifier. When @refs reaches zero,
197  *	  the filter can be freed.
198  * @users: A filter's @users count is incremented for each directly
199  *         attached task (filter installation, fork(), thread_sync),
200  *	   and once for the dependent filter (tracked in filter->prev).
201  *	   When it reaches zero it indicates that no direct or indirect
202  *	   users of that filter exist. No new tasks can get associated with
203  *	   this filter after reaching 0. The @users count is always smaller
204  *	   or equal to @refs. Hence, reaching 0 for @users does not mean
205  *	   the filter can be freed.
206  * @cache: cache of arch/syscall mappings to actions
207  * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
208  * @wait_killable_recv: Put notifying process in killable state once the
209  *			notification is received by the userspace listener.
210  * @prev: points to a previously installed, or inherited, filter
211  * @prog: the BPF program to evaluate
212  * @notif: the struct that holds all notification related information
213  * @notify_lock: A lock for all notification-related accesses.
214  * @wqh: A wait queue for poll if a notifier is in use.
215  *
216  * seccomp_filter objects are organized in a tree linked via the @prev
217  * pointer.  For any task, it appears to be a singly-linked list starting
218  * with current->seccomp.filter, the most recently attached or inherited filter.
219  * However, multiple filters may share a @prev node, by way of fork(), which
220  * results in a unidirectional tree existing in memory.  This is similar to
221  * how namespaces work.
222  *
223  * seccomp_filter objects should never be modified after being attached
224  * to a task_struct (other than @refs).
225  */
226 struct seccomp_filter {
227 	refcount_t refs;
228 	refcount_t users;
229 	bool log;
230 	bool wait_killable_recv;
231 	struct action_cache cache;
232 	struct seccomp_filter *prev;
233 	struct bpf_prog *prog;
234 	struct notification *notif;
235 	struct mutex notify_lock;
236 	wait_queue_head_t wqh;
237 };
238 
239 /* Limit any path through the tree to 256KB worth of instructions. */
240 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
241 
242 /*
243  * Endianness is explicitly ignored and left for BPF program authors to manage
244  * as per the specific architecture.
245  */
246 static void populate_seccomp_data(struct seccomp_data *sd)
247 {
248 	/*
249 	 * Instead of using current_pt_reg(), we're already doing the work
250 	 * to safely fetch "current", so just use "task" everywhere below.
251 	 */
252 	struct task_struct *task = current;
253 	struct pt_regs *regs = task_pt_regs(task);
254 	unsigned long args[6];
255 
256 	sd->nr = syscall_get_nr(task, regs);
257 	sd->arch = syscall_get_arch(task);
258 	syscall_get_arguments(task, regs, args);
259 	sd->args[0] = args[0];
260 	sd->args[1] = args[1];
261 	sd->args[2] = args[2];
262 	sd->args[3] = args[3];
263 	sd->args[4] = args[4];
264 	sd->args[5] = args[5];
265 	sd->instruction_pointer = KSTK_EIP(task);
266 }
267 
268 /**
269  *	seccomp_check_filter - verify seccomp filter code
270  *	@filter: filter to verify
271  *	@flen: length of filter
272  *
273  * Takes a previously checked filter (by bpf_check_classic) and
274  * redirects all filter code that loads struct sk_buff data
275  * and related data through seccomp_bpf_load.  It also
276  * enforces length and alignment checking of those loads.
277  *
278  * Returns 0 if the rule set is legal or -EINVAL if not.
279  */
280 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
281 {
282 	int pc;
283 	for (pc = 0; pc < flen; pc++) {
284 		struct sock_filter *ftest = &filter[pc];
285 		u16 code = ftest->code;
286 		u32 k = ftest->k;
287 
288 		switch (code) {
289 		case BPF_LD | BPF_W | BPF_ABS:
290 			ftest->code = BPF_LDX | BPF_W | BPF_ABS;
291 			/* 32-bit aligned and not out of bounds. */
292 			if (k >= sizeof(struct seccomp_data) || k & 3)
293 				return -EINVAL;
294 			continue;
295 		case BPF_LD | BPF_W | BPF_LEN:
296 			ftest->code = BPF_LD | BPF_IMM;
297 			ftest->k = sizeof(struct seccomp_data);
298 			continue;
299 		case BPF_LDX | BPF_W | BPF_LEN:
300 			ftest->code = BPF_LDX | BPF_IMM;
301 			ftest->k = sizeof(struct seccomp_data);
302 			continue;
303 		/* Explicitly include allowed calls. */
304 		case BPF_RET | BPF_K:
305 		case BPF_RET | BPF_A:
306 		case BPF_ALU | BPF_ADD | BPF_K:
307 		case BPF_ALU | BPF_ADD | BPF_X:
308 		case BPF_ALU | BPF_SUB | BPF_K:
309 		case BPF_ALU | BPF_SUB | BPF_X:
310 		case BPF_ALU | BPF_MUL | BPF_K:
311 		case BPF_ALU | BPF_MUL | BPF_X:
312 		case BPF_ALU | BPF_DIV | BPF_K:
313 		case BPF_ALU | BPF_DIV | BPF_X:
314 		case BPF_ALU | BPF_AND | BPF_K:
315 		case BPF_ALU | BPF_AND | BPF_X:
316 		case BPF_ALU | BPF_OR | BPF_K:
317 		case BPF_ALU | BPF_OR | BPF_X:
318 		case BPF_ALU | BPF_XOR | BPF_K:
319 		case BPF_ALU | BPF_XOR | BPF_X:
320 		case BPF_ALU | BPF_LSH | BPF_K:
321 		case BPF_ALU | BPF_LSH | BPF_X:
322 		case BPF_ALU | BPF_RSH | BPF_K:
323 		case BPF_ALU | BPF_RSH | BPF_X:
324 		case BPF_ALU | BPF_NEG:
325 		case BPF_LD | BPF_IMM:
326 		case BPF_LDX | BPF_IMM:
327 		case BPF_MISC | BPF_TAX:
328 		case BPF_MISC | BPF_TXA:
329 		case BPF_LD | BPF_MEM:
330 		case BPF_LDX | BPF_MEM:
331 		case BPF_ST:
332 		case BPF_STX:
333 		case BPF_JMP | BPF_JA:
334 		case BPF_JMP | BPF_JEQ | BPF_K:
335 		case BPF_JMP | BPF_JEQ | BPF_X:
336 		case BPF_JMP | BPF_JGE | BPF_K:
337 		case BPF_JMP | BPF_JGE | BPF_X:
338 		case BPF_JMP | BPF_JGT | BPF_K:
339 		case BPF_JMP | BPF_JGT | BPF_X:
340 		case BPF_JMP | BPF_JSET | BPF_K:
341 		case BPF_JMP | BPF_JSET | BPF_X:
342 			continue;
343 		default:
344 			return -EINVAL;
345 		}
346 	}
347 	return 0;
348 }
349 
350 #ifdef SECCOMP_ARCH_NATIVE
351 static inline bool seccomp_cache_check_allow_bitmap(const void *bitmap,
352 						    size_t bitmap_size,
353 						    int syscall_nr)
354 {
355 	if (unlikely(syscall_nr < 0 || syscall_nr >= bitmap_size))
356 		return false;
357 	syscall_nr = array_index_nospec(syscall_nr, bitmap_size);
358 
359 	return test_bit(syscall_nr, bitmap);
360 }
361 
362 /**
363  * seccomp_cache_check_allow - lookup seccomp cache
364  * @sfilter: The seccomp filter
365  * @sd: The seccomp data to lookup the cache with
366  *
367  * Returns true if the seccomp_data is cached and allowed.
368  */
369 static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilter,
370 					     const struct seccomp_data *sd)
371 {
372 	int syscall_nr = sd->nr;
373 	const struct action_cache *cache = &sfilter->cache;
374 
375 #ifndef SECCOMP_ARCH_COMPAT
376 	/* A native-only architecture doesn't need to check sd->arch. */
377 	return seccomp_cache_check_allow_bitmap(cache->allow_native,
378 						SECCOMP_ARCH_NATIVE_NR,
379 						syscall_nr);
380 #else
381 	if (likely(sd->arch == SECCOMP_ARCH_NATIVE))
382 		return seccomp_cache_check_allow_bitmap(cache->allow_native,
383 							SECCOMP_ARCH_NATIVE_NR,
384 							syscall_nr);
385 	if (likely(sd->arch == SECCOMP_ARCH_COMPAT))
386 		return seccomp_cache_check_allow_bitmap(cache->allow_compat,
387 							SECCOMP_ARCH_COMPAT_NR,
388 							syscall_nr);
389 #endif /* SECCOMP_ARCH_COMPAT */
390 
391 	WARN_ON_ONCE(true);
392 	return false;
393 }
394 #endif /* SECCOMP_ARCH_NATIVE */
395 
396 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
397 /**
398  * seccomp_run_filters - evaluates all seccomp filters against @sd
399  * @sd: optional seccomp data to be passed to filters
400  * @match: stores struct seccomp_filter that resulted in the return value,
401  *         unless filter returned SECCOMP_RET_ALLOW, in which case it will
402  *         be unchanged.
403  *
404  * Returns valid seccomp BPF response codes.
405  */
406 static u32 seccomp_run_filters(const struct seccomp_data *sd,
407 			       struct seccomp_filter **match)
408 {
409 	u32 ret = SECCOMP_RET_ALLOW;
410 	/* Make sure cross-thread synced filter points somewhere sane. */
411 	struct seccomp_filter *f =
412 			READ_ONCE(current->seccomp.filter);
413 
414 	/* Ensure unexpected behavior doesn't result in failing open. */
415 	if (WARN_ON(f == NULL))
416 		return SECCOMP_RET_KILL_PROCESS;
417 
418 	if (seccomp_cache_check_allow(f, sd))
419 		return SECCOMP_RET_ALLOW;
420 
421 	/*
422 	 * All filters in the list are evaluated and the lowest BPF return
423 	 * value always takes priority (ignoring the DATA).
424 	 */
425 	for (; f; f = f->prev) {
426 		u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd);
427 
428 		if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
429 			ret = cur_ret;
430 			*match = f;
431 		}
432 	}
433 	return ret;
434 }
435 #endif /* CONFIG_SECCOMP_FILTER */
436 
437 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
438 {
439 	assert_spin_locked(&current->sighand->siglock);
440 
441 	if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
442 		return false;
443 
444 	return true;
445 }
446 
447 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
448 
449 static inline void seccomp_assign_mode(struct task_struct *task,
450 				       unsigned long seccomp_mode,
451 				       unsigned long flags)
452 {
453 	assert_spin_locked(&task->sighand->siglock);
454 
455 	task->seccomp.mode = seccomp_mode;
456 	/*
457 	 * Make sure SYSCALL_WORK_SECCOMP cannot be set before the mode (and
458 	 * filter) is set.
459 	 */
460 	smp_mb__before_atomic();
461 	/* Assume default seccomp processes want spec flaw mitigation. */
462 	if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
463 		arch_seccomp_spec_mitigate(task);
464 	set_task_syscall_work(task, SECCOMP);
465 }
466 
467 #ifdef CONFIG_SECCOMP_FILTER
468 /* Returns 1 if the parent is an ancestor of the child. */
469 static int is_ancestor(struct seccomp_filter *parent,
470 		       struct seccomp_filter *child)
471 {
472 	/* NULL is the root ancestor. */
473 	if (parent == NULL)
474 		return 1;
475 	for (; child; child = child->prev)
476 		if (child == parent)
477 			return 1;
478 	return 0;
479 }
480 
481 /**
482  * seccomp_can_sync_threads: checks if all threads can be synchronized
483  *
484  * Expects sighand and cred_guard_mutex locks to be held.
485  *
486  * Returns 0 on success, -ve on error, or the pid of a thread which was
487  * either not in the correct seccomp mode or did not have an ancestral
488  * seccomp filter.
489  */
490 static inline pid_t seccomp_can_sync_threads(void)
491 {
492 	struct task_struct *thread, *caller;
493 
494 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
495 	assert_spin_locked(&current->sighand->siglock);
496 
497 	/* Validate all threads being eligible for synchronization. */
498 	caller = current;
499 	for_each_thread(caller, thread) {
500 		pid_t failed;
501 
502 		/* Skip current, since it is initiating the sync. */
503 		if (thread == caller)
504 			continue;
505 
506 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
507 		    (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
508 		     is_ancestor(thread->seccomp.filter,
509 				 caller->seccomp.filter)))
510 			continue;
511 
512 		/* Return the first thread that cannot be synchronized. */
513 		failed = task_pid_vnr(thread);
514 		/* If the pid cannot be resolved, then return -ESRCH */
515 		if (WARN_ON(failed == 0))
516 			failed = -ESRCH;
517 		return failed;
518 	}
519 
520 	return 0;
521 }
522 
523 static inline void seccomp_filter_free(struct seccomp_filter *filter)
524 {
525 	if (filter) {
526 		bpf_prog_destroy(filter->prog);
527 		kfree(filter);
528 	}
529 }
530 
531 static void __seccomp_filter_orphan(struct seccomp_filter *orig)
532 {
533 	while (orig && refcount_dec_and_test(&orig->users)) {
534 		if (waitqueue_active(&orig->wqh))
535 			wake_up_poll(&orig->wqh, EPOLLHUP);
536 		orig = orig->prev;
537 	}
538 }
539 
540 static void __put_seccomp_filter(struct seccomp_filter *orig)
541 {
542 	/* Clean up single-reference branches iteratively. */
543 	while (orig && refcount_dec_and_test(&orig->refs)) {
544 		struct seccomp_filter *freeme = orig;
545 		orig = orig->prev;
546 		seccomp_filter_free(freeme);
547 	}
548 }
549 
550 static void __seccomp_filter_release(struct seccomp_filter *orig)
551 {
552 	/* Notify about any unused filters in the task's former filter tree. */
553 	__seccomp_filter_orphan(orig);
554 	/* Finally drop all references to the task's former tree. */
555 	__put_seccomp_filter(orig);
556 }
557 
558 /**
559  * seccomp_filter_release - Detach the task from its filter tree,
560  *			    drop its reference count, and notify
561  *			    about unused filters
562  *
563  * @tsk: task the filter should be released from.
564  *
565  * This function should only be called when the task is exiting as
566  * it detaches it from its filter tree. As such, READ_ONCE() and
567  * barriers are not needed here, as would normally be needed.
568  */
569 void seccomp_filter_release(struct task_struct *tsk)
570 {
571 	struct seccomp_filter *orig = tsk->seccomp.filter;
572 
573 	/* We are effectively holding the siglock by not having any sighand. */
574 	WARN_ON(tsk->sighand != NULL);
575 
576 	/* Detach task from its filter tree. */
577 	tsk->seccomp.filter = NULL;
578 	__seccomp_filter_release(orig);
579 }
580 
581 /**
582  * seccomp_sync_threads: sets all threads to use current's filter
583  *
584  * @flags: SECCOMP_FILTER_FLAG_* flags to set during sync.
585  *
586  * Expects sighand and cred_guard_mutex locks to be held, and for
587  * seccomp_can_sync_threads() to have returned success already
588  * without dropping the locks.
589  *
590  */
591 static inline void seccomp_sync_threads(unsigned long flags)
592 {
593 	struct task_struct *thread, *caller;
594 
595 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
596 	assert_spin_locked(&current->sighand->siglock);
597 
598 	/* Synchronize all threads. */
599 	caller = current;
600 	for_each_thread(caller, thread) {
601 		/* Skip current, since it needs no changes. */
602 		if (thread == caller)
603 			continue;
604 
605 		/* Get a task reference for the new leaf node. */
606 		get_seccomp_filter(caller);
607 
608 		/*
609 		 * Drop the task reference to the shared ancestor since
610 		 * current's path will hold a reference.  (This also
611 		 * allows a put before the assignment.)
612 		 */
613 		__seccomp_filter_release(thread->seccomp.filter);
614 
615 		/* Make our new filter tree visible. */
616 		smp_store_release(&thread->seccomp.filter,
617 				  caller->seccomp.filter);
618 		atomic_set(&thread->seccomp.filter_count,
619 			   atomic_read(&caller->seccomp.filter_count));
620 
621 		/*
622 		 * Don't let an unprivileged task work around
623 		 * the no_new_privs restriction by creating
624 		 * a thread that sets it up, enters seccomp,
625 		 * then dies.
626 		 */
627 		if (task_no_new_privs(caller))
628 			task_set_no_new_privs(thread);
629 
630 		/*
631 		 * Opt the other thread into seccomp if needed.
632 		 * As threads are considered to be trust-realm
633 		 * equivalent (see ptrace_may_access), it is safe to
634 		 * allow one thread to transition the other.
635 		 */
636 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
637 			seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
638 					    flags);
639 	}
640 }
641 
642 /**
643  * seccomp_prepare_filter: Prepares a seccomp filter for use.
644  * @fprog: BPF program to install
645  *
646  * Returns filter on success or an ERR_PTR on failure.
647  */
648 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
649 {
650 	struct seccomp_filter *sfilter;
651 	int ret;
652 	const bool save_orig =
653 #if defined(CONFIG_CHECKPOINT_RESTORE) || defined(SECCOMP_ARCH_NATIVE)
654 		true;
655 #else
656 		false;
657 #endif
658 
659 	if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
660 		return ERR_PTR(-EINVAL);
661 
662 	BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
663 
664 	/*
665 	 * Installing a seccomp filter requires that the task has
666 	 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
667 	 * This avoids scenarios where unprivileged tasks can affect the
668 	 * behavior of privileged children.
669 	 */
670 	if (!task_no_new_privs(current) &&
671 			!ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN))
672 		return ERR_PTR(-EACCES);
673 
674 	/* Allocate a new seccomp_filter */
675 	sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
676 	if (!sfilter)
677 		return ERR_PTR(-ENOMEM);
678 
679 	mutex_init(&sfilter->notify_lock);
680 	ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
681 					seccomp_check_filter, save_orig);
682 	if (ret < 0) {
683 		kfree(sfilter);
684 		return ERR_PTR(ret);
685 	}
686 
687 	refcount_set(&sfilter->refs, 1);
688 	refcount_set(&sfilter->users, 1);
689 	init_waitqueue_head(&sfilter->wqh);
690 
691 	return sfilter;
692 }
693 
694 /**
695  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
696  * @user_filter: pointer to the user data containing a sock_fprog.
697  *
698  * Returns 0 on success and non-zero otherwise.
699  */
700 static struct seccomp_filter *
701 seccomp_prepare_user_filter(const char __user *user_filter)
702 {
703 	struct sock_fprog fprog;
704 	struct seccomp_filter *filter = ERR_PTR(-EFAULT);
705 
706 #ifdef CONFIG_COMPAT
707 	if (in_compat_syscall()) {
708 		struct compat_sock_fprog fprog32;
709 		if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
710 			goto out;
711 		fprog.len = fprog32.len;
712 		fprog.filter = compat_ptr(fprog32.filter);
713 	} else /* falls through to the if below. */
714 #endif
715 	if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
716 		goto out;
717 	filter = seccomp_prepare_filter(&fprog);
718 out:
719 	return filter;
720 }
721 
722 #ifdef SECCOMP_ARCH_NATIVE
723 /**
724  * seccomp_is_const_allow - check if filter is constant allow with given data
725  * @fprog: The BPF programs
726  * @sd: The seccomp data to check against, only syscall number and arch
727  *      number are considered constant.
728  */
729 static bool seccomp_is_const_allow(struct sock_fprog_kern *fprog,
730 				   struct seccomp_data *sd)
731 {
732 	unsigned int reg_value = 0;
733 	unsigned int pc;
734 	bool op_res;
735 
736 	if (WARN_ON_ONCE(!fprog))
737 		return false;
738 
739 	for (pc = 0; pc < fprog->len; pc++) {
740 		struct sock_filter *insn = &fprog->filter[pc];
741 		u16 code = insn->code;
742 		u32 k = insn->k;
743 
744 		switch (code) {
745 		case BPF_LD | BPF_W | BPF_ABS:
746 			switch (k) {
747 			case offsetof(struct seccomp_data, nr):
748 				reg_value = sd->nr;
749 				break;
750 			case offsetof(struct seccomp_data, arch):
751 				reg_value = sd->arch;
752 				break;
753 			default:
754 				/* can't optimize (non-constant value load) */
755 				return false;
756 			}
757 			break;
758 		case BPF_RET | BPF_K:
759 			/* reached return with constant values only, check allow */
760 			return k == SECCOMP_RET_ALLOW;
761 		case BPF_JMP | BPF_JA:
762 			pc += insn->k;
763 			break;
764 		case BPF_JMP | BPF_JEQ | BPF_K:
765 		case BPF_JMP | BPF_JGE | BPF_K:
766 		case BPF_JMP | BPF_JGT | BPF_K:
767 		case BPF_JMP | BPF_JSET | BPF_K:
768 			switch (BPF_OP(code)) {
769 			case BPF_JEQ:
770 				op_res = reg_value == k;
771 				break;
772 			case BPF_JGE:
773 				op_res = reg_value >= k;
774 				break;
775 			case BPF_JGT:
776 				op_res = reg_value > k;
777 				break;
778 			case BPF_JSET:
779 				op_res = !!(reg_value & k);
780 				break;
781 			default:
782 				/* can't optimize (unknown jump) */
783 				return false;
784 			}
785 
786 			pc += op_res ? insn->jt : insn->jf;
787 			break;
788 		case BPF_ALU | BPF_AND | BPF_K:
789 			reg_value &= k;
790 			break;
791 		default:
792 			/* can't optimize (unknown insn) */
793 			return false;
794 		}
795 	}
796 
797 	/* ran off the end of the filter?! */
798 	WARN_ON(1);
799 	return false;
800 }
801 
802 static void seccomp_cache_prepare_bitmap(struct seccomp_filter *sfilter,
803 					 void *bitmap, const void *bitmap_prev,
804 					 size_t bitmap_size, int arch)
805 {
806 	struct sock_fprog_kern *fprog = sfilter->prog->orig_prog;
807 	struct seccomp_data sd;
808 	int nr;
809 
810 	if (bitmap_prev) {
811 		/* The new filter must be as restrictive as the last. */
812 		bitmap_copy(bitmap, bitmap_prev, bitmap_size);
813 	} else {
814 		/* Before any filters, all syscalls are always allowed. */
815 		bitmap_fill(bitmap, bitmap_size);
816 	}
817 
818 	for (nr = 0; nr < bitmap_size; nr++) {
819 		/* No bitmap change: not a cacheable action. */
820 		if (!test_bit(nr, bitmap))
821 			continue;
822 
823 		sd.nr = nr;
824 		sd.arch = arch;
825 
826 		/* No bitmap change: continue to always allow. */
827 		if (seccomp_is_const_allow(fprog, &sd))
828 			continue;
829 
830 		/*
831 		 * Not a cacheable action: always run filters.
832 		 * atomic clear_bit() not needed, filter not visible yet.
833 		 */
834 		__clear_bit(nr, bitmap);
835 	}
836 }
837 
838 /**
839  * seccomp_cache_prepare - emulate the filter to find cacheable syscalls
840  * @sfilter: The seccomp filter
841  *
842  * Returns 0 if successful or -errno if error occurred.
843  */
844 static void seccomp_cache_prepare(struct seccomp_filter *sfilter)
845 {
846 	struct action_cache *cache = &sfilter->cache;
847 	const struct action_cache *cache_prev =
848 		sfilter->prev ? &sfilter->prev->cache : NULL;
849 
850 	seccomp_cache_prepare_bitmap(sfilter, cache->allow_native,
851 				     cache_prev ? cache_prev->allow_native : NULL,
852 				     SECCOMP_ARCH_NATIVE_NR,
853 				     SECCOMP_ARCH_NATIVE);
854 
855 #ifdef SECCOMP_ARCH_COMPAT
856 	seccomp_cache_prepare_bitmap(sfilter, cache->allow_compat,
857 				     cache_prev ? cache_prev->allow_compat : NULL,
858 				     SECCOMP_ARCH_COMPAT_NR,
859 				     SECCOMP_ARCH_COMPAT);
860 #endif /* SECCOMP_ARCH_COMPAT */
861 }
862 #endif /* SECCOMP_ARCH_NATIVE */
863 
864 /**
865  * seccomp_attach_filter: validate and attach filter
866  * @flags:  flags to change filter behavior
867  * @filter: seccomp filter to add to the current process
868  *
869  * Caller must be holding current->sighand->siglock lock.
870  *
871  * Returns 0 on success, -ve on error, or
872  *   - in TSYNC mode: the pid of a thread which was either not in the correct
873  *     seccomp mode or did not have an ancestral seccomp filter
874  *   - in NEW_LISTENER mode: the fd of the new listener
875  */
876 static long seccomp_attach_filter(unsigned int flags,
877 				  struct seccomp_filter *filter)
878 {
879 	unsigned long total_insns;
880 	struct seccomp_filter *walker;
881 
882 	assert_spin_locked(&current->sighand->siglock);
883 
884 	/* Validate resulting filter length. */
885 	total_insns = filter->prog->len;
886 	for (walker = current->seccomp.filter; walker; walker = walker->prev)
887 		total_insns += walker->prog->len + 4;  /* 4 instr penalty */
888 	if (total_insns > MAX_INSNS_PER_PATH)
889 		return -ENOMEM;
890 
891 	/* If thread sync has been requested, check that it is possible. */
892 	if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
893 		int ret;
894 
895 		ret = seccomp_can_sync_threads();
896 		if (ret) {
897 			if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
898 				return -ESRCH;
899 			else
900 				return ret;
901 		}
902 	}
903 
904 	/* Set log flag, if present. */
905 	if (flags & SECCOMP_FILTER_FLAG_LOG)
906 		filter->log = true;
907 
908 	/* Set wait killable flag, if present. */
909 	if (flags & SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV)
910 		filter->wait_killable_recv = true;
911 
912 	/*
913 	 * If there is an existing filter, make it the prev and don't drop its
914 	 * task reference.
915 	 */
916 	filter->prev = current->seccomp.filter;
917 	seccomp_cache_prepare(filter);
918 	current->seccomp.filter = filter;
919 	atomic_inc(&current->seccomp.filter_count);
920 
921 	/* Now that the new filter is in place, synchronize to all threads. */
922 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
923 		seccomp_sync_threads(flags);
924 
925 	return 0;
926 }
927 
928 static void __get_seccomp_filter(struct seccomp_filter *filter)
929 {
930 	refcount_inc(&filter->refs);
931 }
932 
933 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
934 void get_seccomp_filter(struct task_struct *tsk)
935 {
936 	struct seccomp_filter *orig = tsk->seccomp.filter;
937 	if (!orig)
938 		return;
939 	__get_seccomp_filter(orig);
940 	refcount_inc(&orig->users);
941 }
942 
943 #endif	/* CONFIG_SECCOMP_FILTER */
944 
945 /* For use with seccomp_actions_logged */
946 #define SECCOMP_LOG_KILL_PROCESS	(1 << 0)
947 #define SECCOMP_LOG_KILL_THREAD		(1 << 1)
948 #define SECCOMP_LOG_TRAP		(1 << 2)
949 #define SECCOMP_LOG_ERRNO		(1 << 3)
950 #define SECCOMP_LOG_TRACE		(1 << 4)
951 #define SECCOMP_LOG_LOG			(1 << 5)
952 #define SECCOMP_LOG_ALLOW		(1 << 6)
953 #define SECCOMP_LOG_USER_NOTIF		(1 << 7)
954 
955 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
956 				    SECCOMP_LOG_KILL_THREAD  |
957 				    SECCOMP_LOG_TRAP  |
958 				    SECCOMP_LOG_ERRNO |
959 				    SECCOMP_LOG_USER_NOTIF |
960 				    SECCOMP_LOG_TRACE |
961 				    SECCOMP_LOG_LOG;
962 
963 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
964 			       bool requested)
965 {
966 	bool log = false;
967 
968 	switch (action) {
969 	case SECCOMP_RET_ALLOW:
970 		break;
971 	case SECCOMP_RET_TRAP:
972 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
973 		break;
974 	case SECCOMP_RET_ERRNO:
975 		log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
976 		break;
977 	case SECCOMP_RET_TRACE:
978 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
979 		break;
980 	case SECCOMP_RET_USER_NOTIF:
981 		log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
982 		break;
983 	case SECCOMP_RET_LOG:
984 		log = seccomp_actions_logged & SECCOMP_LOG_LOG;
985 		break;
986 	case SECCOMP_RET_KILL_THREAD:
987 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
988 		break;
989 	case SECCOMP_RET_KILL_PROCESS:
990 	default:
991 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
992 	}
993 
994 	/*
995 	 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
996 	 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
997 	 * any action from being logged by removing the action name from the
998 	 * seccomp_actions_logged sysctl.
999 	 */
1000 	if (!log)
1001 		return;
1002 
1003 	audit_seccomp(syscall, signr, action);
1004 }
1005 
1006 /*
1007  * Secure computing mode 1 allows only read/write/exit/sigreturn.
1008  * To be fully secure this must be combined with rlimit
1009  * to limit the stack allocations too.
1010  */
1011 static const int mode1_syscalls[] = {
1012 	__NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
1013 	-1, /* negative terminated */
1014 };
1015 
1016 static void __secure_computing_strict(int this_syscall)
1017 {
1018 	const int *allowed_syscalls = mode1_syscalls;
1019 #ifdef CONFIG_COMPAT
1020 	if (in_compat_syscall())
1021 		allowed_syscalls = get_compat_mode1_syscalls();
1022 #endif
1023 	do {
1024 		if (*allowed_syscalls == this_syscall)
1025 			return;
1026 	} while (*++allowed_syscalls != -1);
1027 
1028 #ifdef SECCOMP_DEBUG
1029 	dump_stack();
1030 #endif
1031 	current->seccomp.mode = SECCOMP_MODE_DEAD;
1032 	seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
1033 	do_exit(SIGKILL);
1034 }
1035 
1036 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
1037 void secure_computing_strict(int this_syscall)
1038 {
1039 	int mode = current->seccomp.mode;
1040 
1041 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1042 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1043 		return;
1044 
1045 	if (mode == SECCOMP_MODE_DISABLED)
1046 		return;
1047 	else if (mode == SECCOMP_MODE_STRICT)
1048 		__secure_computing_strict(this_syscall);
1049 	else
1050 		BUG();
1051 }
1052 #else
1053 
1054 #ifdef CONFIG_SECCOMP_FILTER
1055 static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
1056 {
1057 	/*
1058 	 * Note: overflow is ok here, the id just needs to be unique per
1059 	 * filter.
1060 	 */
1061 	lockdep_assert_held(&filter->notify_lock);
1062 	return filter->notif->next_id++;
1063 }
1064 
1065 static void seccomp_handle_addfd(struct seccomp_kaddfd *addfd, struct seccomp_knotif *n)
1066 {
1067 	int fd;
1068 
1069 	/*
1070 	 * Remove the notification, and reset the list pointers, indicating
1071 	 * that it has been handled.
1072 	 */
1073 	list_del_init(&addfd->list);
1074 	if (!addfd->setfd)
1075 		fd = receive_fd(addfd->file, NULL, addfd->flags);
1076 	else
1077 		fd = receive_fd_replace(addfd->fd, addfd->file, addfd->flags);
1078 	addfd->ret = fd;
1079 
1080 	if (addfd->ioctl_flags & SECCOMP_ADDFD_FLAG_SEND) {
1081 		/* If we fail reset and return an error to the notifier */
1082 		if (fd < 0) {
1083 			n->state = SECCOMP_NOTIFY_SENT;
1084 		} else {
1085 			/* Return the FD we just added */
1086 			n->flags = 0;
1087 			n->error = 0;
1088 			n->val = fd;
1089 		}
1090 	}
1091 
1092 	/*
1093 	 * Mark the notification as completed. From this point, addfd mem
1094 	 * might be invalidated and we can't safely read it anymore.
1095 	 */
1096 	complete(&addfd->completion);
1097 }
1098 
1099 static bool should_sleep_killable(struct seccomp_filter *match,
1100 				  struct seccomp_knotif *n)
1101 {
1102 	return match->wait_killable_recv && n->state == SECCOMP_NOTIFY_SENT;
1103 }
1104 
1105 static int seccomp_do_user_notification(int this_syscall,
1106 					struct seccomp_filter *match,
1107 					const struct seccomp_data *sd)
1108 {
1109 	int err;
1110 	u32 flags = 0;
1111 	long ret = 0;
1112 	struct seccomp_knotif n = {};
1113 	struct seccomp_kaddfd *addfd, *tmp;
1114 
1115 	mutex_lock(&match->notify_lock);
1116 	err = -ENOSYS;
1117 	if (!match->notif)
1118 		goto out;
1119 
1120 	n.task = current;
1121 	n.state = SECCOMP_NOTIFY_INIT;
1122 	n.data = sd;
1123 	n.id = seccomp_next_notify_id(match);
1124 	init_completion(&n.ready);
1125 	list_add_tail(&n.list, &match->notif->notifications);
1126 	INIT_LIST_HEAD(&n.addfd);
1127 
1128 	atomic_inc(&match->notif->requests);
1129 	if (match->notif->flags & SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP)
1130 		wake_up_poll_on_current_cpu(&match->wqh, EPOLLIN | EPOLLRDNORM);
1131 	else
1132 		wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM);
1133 
1134 	/*
1135 	 * This is where we wait for a reply from userspace.
1136 	 */
1137 	do {
1138 		bool wait_killable = should_sleep_killable(match, &n);
1139 
1140 		mutex_unlock(&match->notify_lock);
1141 		if (wait_killable)
1142 			err = wait_for_completion_killable(&n.ready);
1143 		else
1144 			err = wait_for_completion_interruptible(&n.ready);
1145 		mutex_lock(&match->notify_lock);
1146 
1147 		if (err != 0) {
1148 			/*
1149 			 * Check to see if the notifcation got picked up and
1150 			 * whether we should switch to wait killable.
1151 			 */
1152 			if (!wait_killable && should_sleep_killable(match, &n))
1153 				continue;
1154 
1155 			goto interrupted;
1156 		}
1157 
1158 		addfd = list_first_entry_or_null(&n.addfd,
1159 						 struct seccomp_kaddfd, list);
1160 		/* Check if we were woken up by a addfd message */
1161 		if (addfd)
1162 			seccomp_handle_addfd(addfd, &n);
1163 
1164 	}  while (n.state != SECCOMP_NOTIFY_REPLIED);
1165 
1166 	ret = n.val;
1167 	err = n.error;
1168 	flags = n.flags;
1169 
1170 interrupted:
1171 	/* If there were any pending addfd calls, clear them out */
1172 	list_for_each_entry_safe(addfd, tmp, &n.addfd, list) {
1173 		/* The process went away before we got a chance to handle it */
1174 		addfd->ret = -ESRCH;
1175 		list_del_init(&addfd->list);
1176 		complete(&addfd->completion);
1177 	}
1178 
1179 	/*
1180 	 * Note that it's possible the listener died in between the time when
1181 	 * we were notified of a response (or a signal) and when we were able to
1182 	 * re-acquire the lock, so only delete from the list if the
1183 	 * notification actually exists.
1184 	 *
1185 	 * Also note that this test is only valid because there's no way to
1186 	 * *reattach* to a notifier right now. If one is added, we'll need to
1187 	 * keep track of the notif itself and make sure they match here.
1188 	 */
1189 	if (match->notif)
1190 		list_del(&n.list);
1191 out:
1192 	mutex_unlock(&match->notify_lock);
1193 
1194 	/* Userspace requests to continue the syscall. */
1195 	if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1196 		return 0;
1197 
1198 	syscall_set_return_value(current, current_pt_regs(),
1199 				 err, ret);
1200 	return -1;
1201 }
1202 
1203 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1204 			    const bool recheck_after_trace)
1205 {
1206 	u32 filter_ret, action;
1207 	struct seccomp_filter *match = NULL;
1208 	int data;
1209 	struct seccomp_data sd_local;
1210 
1211 	/*
1212 	 * Make sure that any changes to mode from another thread have
1213 	 * been seen after SYSCALL_WORK_SECCOMP was seen.
1214 	 */
1215 	smp_rmb();
1216 
1217 	if (!sd) {
1218 		populate_seccomp_data(&sd_local);
1219 		sd = &sd_local;
1220 	}
1221 
1222 	filter_ret = seccomp_run_filters(sd, &match);
1223 	data = filter_ret & SECCOMP_RET_DATA;
1224 	action = filter_ret & SECCOMP_RET_ACTION_FULL;
1225 
1226 	switch (action) {
1227 	case SECCOMP_RET_ERRNO:
1228 		/* Set low-order bits as an errno, capped at MAX_ERRNO. */
1229 		if (data > MAX_ERRNO)
1230 			data = MAX_ERRNO;
1231 		syscall_set_return_value(current, current_pt_regs(),
1232 					 -data, 0);
1233 		goto skip;
1234 
1235 	case SECCOMP_RET_TRAP:
1236 		/* Show the handler the original registers. */
1237 		syscall_rollback(current, current_pt_regs());
1238 		/* Let the filter pass back 16 bits of data. */
1239 		force_sig_seccomp(this_syscall, data, false);
1240 		goto skip;
1241 
1242 	case SECCOMP_RET_TRACE:
1243 		/* We've been put in this state by the ptracer already. */
1244 		if (recheck_after_trace)
1245 			return 0;
1246 
1247 		/* ENOSYS these calls if there is no tracer attached. */
1248 		if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
1249 			syscall_set_return_value(current,
1250 						 current_pt_regs(),
1251 						 -ENOSYS, 0);
1252 			goto skip;
1253 		}
1254 
1255 		/* Allow the BPF to provide the event message */
1256 		ptrace_event(PTRACE_EVENT_SECCOMP, data);
1257 		/*
1258 		 * The delivery of a fatal signal during event
1259 		 * notification may silently skip tracer notification,
1260 		 * which could leave us with a potentially unmodified
1261 		 * syscall that the tracer would have liked to have
1262 		 * changed. Since the process is about to die, we just
1263 		 * force the syscall to be skipped and let the signal
1264 		 * kill the process and correctly handle any tracer exit
1265 		 * notifications.
1266 		 */
1267 		if (fatal_signal_pending(current))
1268 			goto skip;
1269 		/* Check if the tracer forced the syscall to be skipped. */
1270 		this_syscall = syscall_get_nr(current, current_pt_regs());
1271 		if (this_syscall < 0)
1272 			goto skip;
1273 
1274 		/*
1275 		 * Recheck the syscall, since it may have changed. This
1276 		 * intentionally uses a NULL struct seccomp_data to force
1277 		 * a reload of all registers. This does not goto skip since
1278 		 * a skip would have already been reported.
1279 		 */
1280 		if (__seccomp_filter(this_syscall, NULL, true))
1281 			return -1;
1282 
1283 		return 0;
1284 
1285 	case SECCOMP_RET_USER_NOTIF:
1286 		if (seccomp_do_user_notification(this_syscall, match, sd))
1287 			goto skip;
1288 
1289 		return 0;
1290 
1291 	case SECCOMP_RET_LOG:
1292 		seccomp_log(this_syscall, 0, action, true);
1293 		return 0;
1294 
1295 	case SECCOMP_RET_ALLOW:
1296 		/*
1297 		 * Note that the "match" filter will always be NULL for
1298 		 * this action since SECCOMP_RET_ALLOW is the starting
1299 		 * state in seccomp_run_filters().
1300 		 */
1301 		return 0;
1302 
1303 	case SECCOMP_RET_KILL_THREAD:
1304 	case SECCOMP_RET_KILL_PROCESS:
1305 	default:
1306 		current->seccomp.mode = SECCOMP_MODE_DEAD;
1307 		seccomp_log(this_syscall, SIGSYS, action, true);
1308 		/* Dump core only if this is the last remaining thread. */
1309 		if (action != SECCOMP_RET_KILL_THREAD ||
1310 		    (atomic_read(&current->signal->live) == 1)) {
1311 			/* Show the original registers in the dump. */
1312 			syscall_rollback(current, current_pt_regs());
1313 			/* Trigger a coredump with SIGSYS */
1314 			force_sig_seccomp(this_syscall, data, true);
1315 		} else {
1316 			do_exit(SIGSYS);
1317 		}
1318 		return -1; /* skip the syscall go directly to signal handling */
1319 	}
1320 
1321 	unreachable();
1322 
1323 skip:
1324 	seccomp_log(this_syscall, 0, action, match ? match->log : false);
1325 	return -1;
1326 }
1327 #else
1328 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1329 			    const bool recheck_after_trace)
1330 {
1331 	BUG();
1332 
1333 	return -1;
1334 }
1335 #endif
1336 
1337 int __secure_computing(const struct seccomp_data *sd)
1338 {
1339 	int mode = current->seccomp.mode;
1340 	int this_syscall;
1341 
1342 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1343 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1344 		return 0;
1345 
1346 	this_syscall = sd ? sd->nr :
1347 		syscall_get_nr(current, current_pt_regs());
1348 
1349 	switch (mode) {
1350 	case SECCOMP_MODE_STRICT:
1351 		__secure_computing_strict(this_syscall);  /* may call do_exit */
1352 		return 0;
1353 	case SECCOMP_MODE_FILTER:
1354 		return __seccomp_filter(this_syscall, sd, false);
1355 	/* Surviving SECCOMP_RET_KILL_* must be proactively impossible. */
1356 	case SECCOMP_MODE_DEAD:
1357 		WARN_ON_ONCE(1);
1358 		do_exit(SIGKILL);
1359 		return -1;
1360 	default:
1361 		BUG();
1362 	}
1363 }
1364 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1365 
1366 long prctl_get_seccomp(void)
1367 {
1368 	return current->seccomp.mode;
1369 }
1370 
1371 /**
1372  * seccomp_set_mode_strict: internal function for setting strict seccomp
1373  *
1374  * Once current->seccomp.mode is non-zero, it may not be changed.
1375  *
1376  * Returns 0 on success or -EINVAL on failure.
1377  */
1378 static long seccomp_set_mode_strict(void)
1379 {
1380 	const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
1381 	long ret = -EINVAL;
1382 
1383 	spin_lock_irq(&current->sighand->siglock);
1384 
1385 	if (!seccomp_may_assign_mode(seccomp_mode))
1386 		goto out;
1387 
1388 #ifdef TIF_NOTSC
1389 	disable_TSC();
1390 #endif
1391 	seccomp_assign_mode(current, seccomp_mode, 0);
1392 	ret = 0;
1393 
1394 out:
1395 	spin_unlock_irq(&current->sighand->siglock);
1396 
1397 	return ret;
1398 }
1399 
1400 #ifdef CONFIG_SECCOMP_FILTER
1401 static void seccomp_notify_free(struct seccomp_filter *filter)
1402 {
1403 	kfree(filter->notif);
1404 	filter->notif = NULL;
1405 }
1406 
1407 static void seccomp_notify_detach(struct seccomp_filter *filter)
1408 {
1409 	struct seccomp_knotif *knotif;
1410 
1411 	if (!filter)
1412 		return;
1413 
1414 	mutex_lock(&filter->notify_lock);
1415 
1416 	/*
1417 	 * If this file is being closed because e.g. the task who owned it
1418 	 * died, let's wake everyone up who was waiting on us.
1419 	 */
1420 	list_for_each_entry(knotif, &filter->notif->notifications, list) {
1421 		if (knotif->state == SECCOMP_NOTIFY_REPLIED)
1422 			continue;
1423 
1424 		knotif->state = SECCOMP_NOTIFY_REPLIED;
1425 		knotif->error = -ENOSYS;
1426 		knotif->val = 0;
1427 
1428 		/*
1429 		 * We do not need to wake up any pending addfd messages, as
1430 		 * the notifier will do that for us, as this just looks
1431 		 * like a standard reply.
1432 		 */
1433 		complete(&knotif->ready);
1434 	}
1435 
1436 	seccomp_notify_free(filter);
1437 	mutex_unlock(&filter->notify_lock);
1438 }
1439 
1440 static int seccomp_notify_release(struct inode *inode, struct file *file)
1441 {
1442 	struct seccomp_filter *filter = file->private_data;
1443 
1444 	seccomp_notify_detach(filter);
1445 	__put_seccomp_filter(filter);
1446 	return 0;
1447 }
1448 
1449 /* must be called with notif_lock held */
1450 static inline struct seccomp_knotif *
1451 find_notification(struct seccomp_filter *filter, u64 id)
1452 {
1453 	struct seccomp_knotif *cur;
1454 
1455 	lockdep_assert_held(&filter->notify_lock);
1456 
1457 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1458 		if (cur->id == id)
1459 			return cur;
1460 	}
1461 
1462 	return NULL;
1463 }
1464 
1465 static int recv_wake_function(wait_queue_entry_t *wait, unsigned int mode, int sync,
1466 				  void *key)
1467 {
1468 	/* Avoid a wakeup if event not interesting for us. */
1469 	if (key && !(key_to_poll(key) & (EPOLLIN | EPOLLERR | EPOLLHUP)))
1470 		return 0;
1471 	return autoremove_wake_function(wait, mode, sync, key);
1472 }
1473 
1474 static int recv_wait_event(struct seccomp_filter *filter)
1475 {
1476 	DEFINE_WAIT_FUNC(wait, recv_wake_function);
1477 	int ret;
1478 
1479 	if (refcount_read(&filter->users) == 0)
1480 		return 0;
1481 
1482 	if (atomic_dec_if_positive(&filter->notif->requests) >= 0)
1483 		return 0;
1484 
1485 	for (;;) {
1486 		ret = prepare_to_wait_event(&filter->wqh, &wait, TASK_INTERRUPTIBLE);
1487 
1488 		if (atomic_dec_if_positive(&filter->notif->requests) >= 0)
1489 			break;
1490 		if (refcount_read(&filter->users) == 0)
1491 			break;
1492 
1493 		if (ret)
1494 			return ret;
1495 
1496 		schedule();
1497 	}
1498 	finish_wait(&filter->wqh, &wait);
1499 	return 0;
1500 }
1501 
1502 static long seccomp_notify_recv(struct seccomp_filter *filter,
1503 				void __user *buf)
1504 {
1505 	struct seccomp_knotif *knotif = NULL, *cur;
1506 	struct seccomp_notif unotif;
1507 	ssize_t ret;
1508 
1509 	/* Verify that we're not given garbage to keep struct extensible. */
1510 	ret = check_zeroed_user(buf, sizeof(unotif));
1511 	if (ret < 0)
1512 		return ret;
1513 	if (!ret)
1514 		return -EINVAL;
1515 
1516 	memset(&unotif, 0, sizeof(unotif));
1517 
1518 	ret = recv_wait_event(filter);
1519 	if (ret < 0)
1520 		return ret;
1521 
1522 	mutex_lock(&filter->notify_lock);
1523 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1524 		if (cur->state == SECCOMP_NOTIFY_INIT) {
1525 			knotif = cur;
1526 			break;
1527 		}
1528 	}
1529 
1530 	/*
1531 	 * If we didn't find a notification, it could be that the task was
1532 	 * interrupted by a fatal signal between the time we were woken and
1533 	 * when we were able to acquire the rw lock.
1534 	 */
1535 	if (!knotif) {
1536 		ret = -ENOENT;
1537 		goto out;
1538 	}
1539 
1540 	unotif.id = knotif->id;
1541 	unotif.pid = task_pid_vnr(knotif->task);
1542 	unotif.data = *(knotif->data);
1543 
1544 	knotif->state = SECCOMP_NOTIFY_SENT;
1545 	wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM);
1546 	ret = 0;
1547 out:
1548 	mutex_unlock(&filter->notify_lock);
1549 
1550 	if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1551 		ret = -EFAULT;
1552 
1553 		/*
1554 		 * Userspace screwed up. To make sure that we keep this
1555 		 * notification alive, let's reset it back to INIT. It
1556 		 * may have died when we released the lock, so we need to make
1557 		 * sure it's still around.
1558 		 */
1559 		mutex_lock(&filter->notify_lock);
1560 		knotif = find_notification(filter, unotif.id);
1561 		if (knotif) {
1562 			/* Reset the process to make sure it's not stuck */
1563 			if (should_sleep_killable(filter, knotif))
1564 				complete(&knotif->ready);
1565 			knotif->state = SECCOMP_NOTIFY_INIT;
1566 			atomic_inc(&filter->notif->requests);
1567 			wake_up_poll(&filter->wqh, EPOLLIN | EPOLLRDNORM);
1568 		}
1569 		mutex_unlock(&filter->notify_lock);
1570 	}
1571 
1572 	return ret;
1573 }
1574 
1575 static long seccomp_notify_send(struct seccomp_filter *filter,
1576 				void __user *buf)
1577 {
1578 	struct seccomp_notif_resp resp = {};
1579 	struct seccomp_knotif *knotif;
1580 	long ret;
1581 
1582 	if (copy_from_user(&resp, buf, sizeof(resp)))
1583 		return -EFAULT;
1584 
1585 	if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1586 		return -EINVAL;
1587 
1588 	if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) &&
1589 	    (resp.error || resp.val))
1590 		return -EINVAL;
1591 
1592 	ret = mutex_lock_interruptible(&filter->notify_lock);
1593 	if (ret < 0)
1594 		return ret;
1595 
1596 	knotif = find_notification(filter, resp.id);
1597 	if (!knotif) {
1598 		ret = -ENOENT;
1599 		goto out;
1600 	}
1601 
1602 	/* Allow exactly one reply. */
1603 	if (knotif->state != SECCOMP_NOTIFY_SENT) {
1604 		ret = -EINPROGRESS;
1605 		goto out;
1606 	}
1607 
1608 	ret = 0;
1609 	knotif->state = SECCOMP_NOTIFY_REPLIED;
1610 	knotif->error = resp.error;
1611 	knotif->val = resp.val;
1612 	knotif->flags = resp.flags;
1613 	if (filter->notif->flags & SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP)
1614 		complete_on_current_cpu(&knotif->ready);
1615 	else
1616 		complete(&knotif->ready);
1617 out:
1618 	mutex_unlock(&filter->notify_lock);
1619 	return ret;
1620 }
1621 
1622 static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1623 				    void __user *buf)
1624 {
1625 	struct seccomp_knotif *knotif;
1626 	u64 id;
1627 	long ret;
1628 
1629 	if (copy_from_user(&id, buf, sizeof(id)))
1630 		return -EFAULT;
1631 
1632 	ret = mutex_lock_interruptible(&filter->notify_lock);
1633 	if (ret < 0)
1634 		return ret;
1635 
1636 	knotif = find_notification(filter, id);
1637 	if (knotif && knotif->state == SECCOMP_NOTIFY_SENT)
1638 		ret = 0;
1639 	else
1640 		ret = -ENOENT;
1641 
1642 	mutex_unlock(&filter->notify_lock);
1643 	return ret;
1644 }
1645 
1646 static long seccomp_notify_set_flags(struct seccomp_filter *filter,
1647 				    unsigned long flags)
1648 {
1649 	long ret;
1650 
1651 	if (flags & ~SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP)
1652 		return -EINVAL;
1653 
1654 	ret = mutex_lock_interruptible(&filter->notify_lock);
1655 	if (ret < 0)
1656 		return ret;
1657 	filter->notif->flags = flags;
1658 	mutex_unlock(&filter->notify_lock);
1659 	return 0;
1660 }
1661 
1662 static long seccomp_notify_addfd(struct seccomp_filter *filter,
1663 				 struct seccomp_notif_addfd __user *uaddfd,
1664 				 unsigned int size)
1665 {
1666 	struct seccomp_notif_addfd addfd;
1667 	struct seccomp_knotif *knotif;
1668 	struct seccomp_kaddfd kaddfd;
1669 	int ret;
1670 
1671 	BUILD_BUG_ON(sizeof(addfd) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0);
1672 	BUILD_BUG_ON(sizeof(addfd) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST);
1673 
1674 	if (size < SECCOMP_NOTIFY_ADDFD_SIZE_VER0 || size >= PAGE_SIZE)
1675 		return -EINVAL;
1676 
1677 	ret = copy_struct_from_user(&addfd, sizeof(addfd), uaddfd, size);
1678 	if (ret)
1679 		return ret;
1680 
1681 	if (addfd.newfd_flags & ~O_CLOEXEC)
1682 		return -EINVAL;
1683 
1684 	if (addfd.flags & ~(SECCOMP_ADDFD_FLAG_SETFD | SECCOMP_ADDFD_FLAG_SEND))
1685 		return -EINVAL;
1686 
1687 	if (addfd.newfd && !(addfd.flags & SECCOMP_ADDFD_FLAG_SETFD))
1688 		return -EINVAL;
1689 
1690 	kaddfd.file = fget(addfd.srcfd);
1691 	if (!kaddfd.file)
1692 		return -EBADF;
1693 
1694 	kaddfd.ioctl_flags = addfd.flags;
1695 	kaddfd.flags = addfd.newfd_flags;
1696 	kaddfd.setfd = addfd.flags & SECCOMP_ADDFD_FLAG_SETFD;
1697 	kaddfd.fd = addfd.newfd;
1698 	init_completion(&kaddfd.completion);
1699 
1700 	ret = mutex_lock_interruptible(&filter->notify_lock);
1701 	if (ret < 0)
1702 		goto out;
1703 
1704 	knotif = find_notification(filter, addfd.id);
1705 	if (!knotif) {
1706 		ret = -ENOENT;
1707 		goto out_unlock;
1708 	}
1709 
1710 	/*
1711 	 * We do not want to allow for FD injection to occur before the
1712 	 * notification has been picked up by a userspace handler, or after
1713 	 * the notification has been replied to.
1714 	 */
1715 	if (knotif->state != SECCOMP_NOTIFY_SENT) {
1716 		ret = -EINPROGRESS;
1717 		goto out_unlock;
1718 	}
1719 
1720 	if (addfd.flags & SECCOMP_ADDFD_FLAG_SEND) {
1721 		/*
1722 		 * Disallow queuing an atomic addfd + send reply while there are
1723 		 * some addfd requests still to process.
1724 		 *
1725 		 * There is no clear reason to support it and allows us to keep
1726 		 * the loop on the other side straight-forward.
1727 		 */
1728 		if (!list_empty(&knotif->addfd)) {
1729 			ret = -EBUSY;
1730 			goto out_unlock;
1731 		}
1732 
1733 		/* Allow exactly only one reply */
1734 		knotif->state = SECCOMP_NOTIFY_REPLIED;
1735 	}
1736 
1737 	list_add(&kaddfd.list, &knotif->addfd);
1738 	complete(&knotif->ready);
1739 	mutex_unlock(&filter->notify_lock);
1740 
1741 	/* Now we wait for it to be processed or be interrupted */
1742 	ret = wait_for_completion_interruptible(&kaddfd.completion);
1743 	if (ret == 0) {
1744 		/*
1745 		 * We had a successful completion. The other side has already
1746 		 * removed us from the addfd queue, and
1747 		 * wait_for_completion_interruptible has a memory barrier upon
1748 		 * success that lets us read this value directly without
1749 		 * locking.
1750 		 */
1751 		ret = kaddfd.ret;
1752 		goto out;
1753 	}
1754 
1755 	mutex_lock(&filter->notify_lock);
1756 	/*
1757 	 * Even though we were woken up by a signal and not a successful
1758 	 * completion, a completion may have happened in the mean time.
1759 	 *
1760 	 * We need to check again if the addfd request has been handled,
1761 	 * and if not, we will remove it from the queue.
1762 	 */
1763 	if (list_empty(&kaddfd.list))
1764 		ret = kaddfd.ret;
1765 	else
1766 		list_del(&kaddfd.list);
1767 
1768 out_unlock:
1769 	mutex_unlock(&filter->notify_lock);
1770 out:
1771 	fput(kaddfd.file);
1772 
1773 	return ret;
1774 }
1775 
1776 static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1777 				 unsigned long arg)
1778 {
1779 	struct seccomp_filter *filter = file->private_data;
1780 	void __user *buf = (void __user *)arg;
1781 
1782 	/* Fixed-size ioctls */
1783 	switch (cmd) {
1784 	case SECCOMP_IOCTL_NOTIF_RECV:
1785 		return seccomp_notify_recv(filter, buf);
1786 	case SECCOMP_IOCTL_NOTIF_SEND:
1787 		return seccomp_notify_send(filter, buf);
1788 	case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR:
1789 	case SECCOMP_IOCTL_NOTIF_ID_VALID:
1790 		return seccomp_notify_id_valid(filter, buf);
1791 	case SECCOMP_IOCTL_NOTIF_SET_FLAGS:
1792 		return seccomp_notify_set_flags(filter, arg);
1793 	}
1794 
1795 	/* Extensible Argument ioctls */
1796 #define EA_IOCTL(cmd)	((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1797 	switch (EA_IOCTL(cmd)) {
1798 	case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD):
1799 		return seccomp_notify_addfd(filter, buf, _IOC_SIZE(cmd));
1800 	default:
1801 		return -EINVAL;
1802 	}
1803 }
1804 
1805 static __poll_t seccomp_notify_poll(struct file *file,
1806 				    struct poll_table_struct *poll_tab)
1807 {
1808 	struct seccomp_filter *filter = file->private_data;
1809 	__poll_t ret = 0;
1810 	struct seccomp_knotif *cur;
1811 
1812 	poll_wait(file, &filter->wqh, poll_tab);
1813 
1814 	if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1815 		return EPOLLERR;
1816 
1817 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1818 		if (cur->state == SECCOMP_NOTIFY_INIT)
1819 			ret |= EPOLLIN | EPOLLRDNORM;
1820 		if (cur->state == SECCOMP_NOTIFY_SENT)
1821 			ret |= EPOLLOUT | EPOLLWRNORM;
1822 		if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1823 			break;
1824 	}
1825 
1826 	mutex_unlock(&filter->notify_lock);
1827 
1828 	if (refcount_read(&filter->users) == 0)
1829 		ret |= EPOLLHUP;
1830 
1831 	return ret;
1832 }
1833 
1834 static const struct file_operations seccomp_notify_ops = {
1835 	.poll = seccomp_notify_poll,
1836 	.release = seccomp_notify_release,
1837 	.unlocked_ioctl = seccomp_notify_ioctl,
1838 	.compat_ioctl = seccomp_notify_ioctl,
1839 };
1840 
1841 static struct file *init_listener(struct seccomp_filter *filter)
1842 {
1843 	struct file *ret;
1844 
1845 	ret = ERR_PTR(-ENOMEM);
1846 	filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1847 	if (!filter->notif)
1848 		goto out;
1849 
1850 	filter->notif->next_id = get_random_u64();
1851 	INIT_LIST_HEAD(&filter->notif->notifications);
1852 
1853 	ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1854 				 filter, O_RDWR);
1855 	if (IS_ERR(ret))
1856 		goto out_notif;
1857 
1858 	/* The file has a reference to it now */
1859 	__get_seccomp_filter(filter);
1860 
1861 out_notif:
1862 	if (IS_ERR(ret))
1863 		seccomp_notify_free(filter);
1864 out:
1865 	return ret;
1866 }
1867 
1868 /*
1869  * Does @new_child have a listener while an ancestor also has a listener?
1870  * If so, we'll want to reject this filter.
1871  * This only has to be tested for the current process, even in the TSYNC case,
1872  * because TSYNC installs @child with the same parent on all threads.
1873  * Note that @new_child is not hooked up to its parent at this point yet, so
1874  * we use current->seccomp.filter.
1875  */
1876 static bool has_duplicate_listener(struct seccomp_filter *new_child)
1877 {
1878 	struct seccomp_filter *cur;
1879 
1880 	/* must be protected against concurrent TSYNC */
1881 	lockdep_assert_held(&current->sighand->siglock);
1882 
1883 	if (!new_child->notif)
1884 		return false;
1885 	for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1886 		if (cur->notif)
1887 			return true;
1888 	}
1889 
1890 	return false;
1891 }
1892 
1893 /**
1894  * seccomp_set_mode_filter: internal function for setting seccomp filter
1895  * @flags:  flags to change filter behavior
1896  * @filter: struct sock_fprog containing filter
1897  *
1898  * This function may be called repeatedly to install additional filters.
1899  * Every filter successfully installed will be evaluated (in reverse order)
1900  * for each system call the task makes.
1901  *
1902  * Once current->seccomp.mode is non-zero, it may not be changed.
1903  *
1904  * Returns 0 on success or -EINVAL on failure.
1905  */
1906 static long seccomp_set_mode_filter(unsigned int flags,
1907 				    const char __user *filter)
1908 {
1909 	const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1910 	struct seccomp_filter *prepared = NULL;
1911 	long ret = -EINVAL;
1912 	int listener = -1;
1913 	struct file *listener_f = NULL;
1914 
1915 	/* Validate flags. */
1916 	if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1917 		return -EINVAL;
1918 
1919 	/*
1920 	 * In the successful case, NEW_LISTENER returns the new listener fd.
1921 	 * But in the failure case, TSYNC returns the thread that died. If you
1922 	 * combine these two flags, there's no way to tell whether something
1923 	 * succeeded or failed. So, let's disallow this combination if the user
1924 	 * has not explicitly requested no errors from TSYNC.
1925 	 */
1926 	if ((flags & SECCOMP_FILTER_FLAG_TSYNC) &&
1927 	    (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) &&
1928 	    ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0))
1929 		return -EINVAL;
1930 
1931 	/*
1932 	 * The SECCOMP_FILTER_FLAG_WAIT_KILLABLE_SENT flag doesn't make sense
1933 	 * without the SECCOMP_FILTER_FLAG_NEW_LISTENER flag.
1934 	 */
1935 	if ((flags & SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV) &&
1936 	    ((flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) == 0))
1937 		return -EINVAL;
1938 
1939 	/* Prepare the new filter before holding any locks. */
1940 	prepared = seccomp_prepare_user_filter(filter);
1941 	if (IS_ERR(prepared))
1942 		return PTR_ERR(prepared);
1943 
1944 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1945 		listener = get_unused_fd_flags(O_CLOEXEC);
1946 		if (listener < 0) {
1947 			ret = listener;
1948 			goto out_free;
1949 		}
1950 
1951 		listener_f = init_listener(prepared);
1952 		if (IS_ERR(listener_f)) {
1953 			put_unused_fd(listener);
1954 			ret = PTR_ERR(listener_f);
1955 			goto out_free;
1956 		}
1957 	}
1958 
1959 	/*
1960 	 * Make sure we cannot change seccomp or nnp state via TSYNC
1961 	 * while another thread is in the middle of calling exec.
1962 	 */
1963 	if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1964 	    mutex_lock_killable(&current->signal->cred_guard_mutex))
1965 		goto out_put_fd;
1966 
1967 	spin_lock_irq(&current->sighand->siglock);
1968 
1969 	if (!seccomp_may_assign_mode(seccomp_mode))
1970 		goto out;
1971 
1972 	if (has_duplicate_listener(prepared)) {
1973 		ret = -EBUSY;
1974 		goto out;
1975 	}
1976 
1977 	ret = seccomp_attach_filter(flags, prepared);
1978 	if (ret)
1979 		goto out;
1980 	/* Do not free the successfully attached filter. */
1981 	prepared = NULL;
1982 
1983 	seccomp_assign_mode(current, seccomp_mode, flags);
1984 out:
1985 	spin_unlock_irq(&current->sighand->siglock);
1986 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
1987 		mutex_unlock(&current->signal->cred_guard_mutex);
1988 out_put_fd:
1989 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1990 		if (ret) {
1991 			listener_f->private_data = NULL;
1992 			fput(listener_f);
1993 			put_unused_fd(listener);
1994 			seccomp_notify_detach(prepared);
1995 		} else {
1996 			fd_install(listener, listener_f);
1997 			ret = listener;
1998 		}
1999 	}
2000 out_free:
2001 	seccomp_filter_free(prepared);
2002 	return ret;
2003 }
2004 #else
2005 static inline long seccomp_set_mode_filter(unsigned int flags,
2006 					   const char __user *filter)
2007 {
2008 	return -EINVAL;
2009 }
2010 #endif
2011 
2012 static long seccomp_get_action_avail(const char __user *uaction)
2013 {
2014 	u32 action;
2015 
2016 	if (copy_from_user(&action, uaction, sizeof(action)))
2017 		return -EFAULT;
2018 
2019 	switch (action) {
2020 	case SECCOMP_RET_KILL_PROCESS:
2021 	case SECCOMP_RET_KILL_THREAD:
2022 	case SECCOMP_RET_TRAP:
2023 	case SECCOMP_RET_ERRNO:
2024 	case SECCOMP_RET_USER_NOTIF:
2025 	case SECCOMP_RET_TRACE:
2026 	case SECCOMP_RET_LOG:
2027 	case SECCOMP_RET_ALLOW:
2028 		break;
2029 	default:
2030 		return -EOPNOTSUPP;
2031 	}
2032 
2033 	return 0;
2034 }
2035 
2036 static long seccomp_get_notif_sizes(void __user *usizes)
2037 {
2038 	struct seccomp_notif_sizes sizes = {
2039 		.seccomp_notif = sizeof(struct seccomp_notif),
2040 		.seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
2041 		.seccomp_data = sizeof(struct seccomp_data),
2042 	};
2043 
2044 	if (copy_to_user(usizes, &sizes, sizeof(sizes)))
2045 		return -EFAULT;
2046 
2047 	return 0;
2048 }
2049 
2050 /* Common entry point for both prctl and syscall. */
2051 static long do_seccomp(unsigned int op, unsigned int flags,
2052 		       void __user *uargs)
2053 {
2054 	switch (op) {
2055 	case SECCOMP_SET_MODE_STRICT:
2056 		if (flags != 0 || uargs != NULL)
2057 			return -EINVAL;
2058 		return seccomp_set_mode_strict();
2059 	case SECCOMP_SET_MODE_FILTER:
2060 		return seccomp_set_mode_filter(flags, uargs);
2061 	case SECCOMP_GET_ACTION_AVAIL:
2062 		if (flags != 0)
2063 			return -EINVAL;
2064 
2065 		return seccomp_get_action_avail(uargs);
2066 	case SECCOMP_GET_NOTIF_SIZES:
2067 		if (flags != 0)
2068 			return -EINVAL;
2069 
2070 		return seccomp_get_notif_sizes(uargs);
2071 	default:
2072 		return -EINVAL;
2073 	}
2074 }
2075 
2076 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
2077 			 void __user *, uargs)
2078 {
2079 	return do_seccomp(op, flags, uargs);
2080 }
2081 
2082 /**
2083  * prctl_set_seccomp: configures current->seccomp.mode
2084  * @seccomp_mode: requested mode to use
2085  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
2086  *
2087  * Returns 0 on success or -EINVAL on failure.
2088  */
2089 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
2090 {
2091 	unsigned int op;
2092 	void __user *uargs;
2093 
2094 	switch (seccomp_mode) {
2095 	case SECCOMP_MODE_STRICT:
2096 		op = SECCOMP_SET_MODE_STRICT;
2097 		/*
2098 		 * Setting strict mode through prctl always ignored filter,
2099 		 * so make sure it is always NULL here to pass the internal
2100 		 * check in do_seccomp().
2101 		 */
2102 		uargs = NULL;
2103 		break;
2104 	case SECCOMP_MODE_FILTER:
2105 		op = SECCOMP_SET_MODE_FILTER;
2106 		uargs = filter;
2107 		break;
2108 	default:
2109 		return -EINVAL;
2110 	}
2111 
2112 	/* prctl interface doesn't have flags, so they are always zero. */
2113 	return do_seccomp(op, 0, uargs);
2114 }
2115 
2116 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
2117 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
2118 					     unsigned long filter_off)
2119 {
2120 	struct seccomp_filter *orig, *filter;
2121 	unsigned long count;
2122 
2123 	/*
2124 	 * Note: this is only correct because the caller should be the (ptrace)
2125 	 * tracer of the task, otherwise lock_task_sighand is needed.
2126 	 */
2127 	spin_lock_irq(&task->sighand->siglock);
2128 
2129 	if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
2130 		spin_unlock_irq(&task->sighand->siglock);
2131 		return ERR_PTR(-EINVAL);
2132 	}
2133 
2134 	orig = task->seccomp.filter;
2135 	__get_seccomp_filter(orig);
2136 	spin_unlock_irq(&task->sighand->siglock);
2137 
2138 	count = 0;
2139 	for (filter = orig; filter; filter = filter->prev)
2140 		count++;
2141 
2142 	if (filter_off >= count) {
2143 		filter = ERR_PTR(-ENOENT);
2144 		goto out;
2145 	}
2146 
2147 	count -= filter_off;
2148 	for (filter = orig; filter && count > 1; filter = filter->prev)
2149 		count--;
2150 
2151 	if (WARN_ON(count != 1 || !filter)) {
2152 		filter = ERR_PTR(-ENOENT);
2153 		goto out;
2154 	}
2155 
2156 	__get_seccomp_filter(filter);
2157 
2158 out:
2159 	__put_seccomp_filter(orig);
2160 	return filter;
2161 }
2162 
2163 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
2164 			void __user *data)
2165 {
2166 	struct seccomp_filter *filter;
2167 	struct sock_fprog_kern *fprog;
2168 	long ret;
2169 
2170 	if (!capable(CAP_SYS_ADMIN) ||
2171 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
2172 		return -EACCES;
2173 	}
2174 
2175 	filter = get_nth_filter(task, filter_off);
2176 	if (IS_ERR(filter))
2177 		return PTR_ERR(filter);
2178 
2179 	fprog = filter->prog->orig_prog;
2180 	if (!fprog) {
2181 		/* This must be a new non-cBPF filter, since we save
2182 		 * every cBPF filter's orig_prog above when
2183 		 * CONFIG_CHECKPOINT_RESTORE is enabled.
2184 		 */
2185 		ret = -EMEDIUMTYPE;
2186 		goto out;
2187 	}
2188 
2189 	ret = fprog->len;
2190 	if (!data)
2191 		goto out;
2192 
2193 	if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
2194 		ret = -EFAULT;
2195 
2196 out:
2197 	__put_seccomp_filter(filter);
2198 	return ret;
2199 }
2200 
2201 long seccomp_get_metadata(struct task_struct *task,
2202 			  unsigned long size, void __user *data)
2203 {
2204 	long ret;
2205 	struct seccomp_filter *filter;
2206 	struct seccomp_metadata kmd = {};
2207 
2208 	if (!capable(CAP_SYS_ADMIN) ||
2209 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
2210 		return -EACCES;
2211 	}
2212 
2213 	size = min_t(unsigned long, size, sizeof(kmd));
2214 
2215 	if (size < sizeof(kmd.filter_off))
2216 		return -EINVAL;
2217 
2218 	if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
2219 		return -EFAULT;
2220 
2221 	filter = get_nth_filter(task, kmd.filter_off);
2222 	if (IS_ERR(filter))
2223 		return PTR_ERR(filter);
2224 
2225 	if (filter->log)
2226 		kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
2227 
2228 	ret = size;
2229 	if (copy_to_user(data, &kmd, size))
2230 		ret = -EFAULT;
2231 
2232 	__put_seccomp_filter(filter);
2233 	return ret;
2234 }
2235 #endif
2236 
2237 #ifdef CONFIG_SYSCTL
2238 
2239 /* Human readable action names for friendly sysctl interaction */
2240 #define SECCOMP_RET_KILL_PROCESS_NAME	"kill_process"
2241 #define SECCOMP_RET_KILL_THREAD_NAME	"kill_thread"
2242 #define SECCOMP_RET_TRAP_NAME		"trap"
2243 #define SECCOMP_RET_ERRNO_NAME		"errno"
2244 #define SECCOMP_RET_USER_NOTIF_NAME	"user_notif"
2245 #define SECCOMP_RET_TRACE_NAME		"trace"
2246 #define SECCOMP_RET_LOG_NAME		"log"
2247 #define SECCOMP_RET_ALLOW_NAME		"allow"
2248 
2249 static const char seccomp_actions_avail[] =
2250 				SECCOMP_RET_KILL_PROCESS_NAME	" "
2251 				SECCOMP_RET_KILL_THREAD_NAME	" "
2252 				SECCOMP_RET_TRAP_NAME		" "
2253 				SECCOMP_RET_ERRNO_NAME		" "
2254 				SECCOMP_RET_USER_NOTIF_NAME     " "
2255 				SECCOMP_RET_TRACE_NAME		" "
2256 				SECCOMP_RET_LOG_NAME		" "
2257 				SECCOMP_RET_ALLOW_NAME;
2258 
2259 struct seccomp_log_name {
2260 	u32		log;
2261 	const char	*name;
2262 };
2263 
2264 static const struct seccomp_log_name seccomp_log_names[] = {
2265 	{ SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
2266 	{ SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
2267 	{ SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
2268 	{ SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
2269 	{ SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
2270 	{ SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
2271 	{ SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
2272 	{ SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
2273 	{ }
2274 };
2275 
2276 static bool seccomp_names_from_actions_logged(char *names, size_t size,
2277 					      u32 actions_logged,
2278 					      const char *sep)
2279 {
2280 	const struct seccomp_log_name *cur;
2281 	bool append_sep = false;
2282 
2283 	for (cur = seccomp_log_names; cur->name && size; cur++) {
2284 		ssize_t ret;
2285 
2286 		if (!(actions_logged & cur->log))
2287 			continue;
2288 
2289 		if (append_sep) {
2290 			ret = strscpy(names, sep, size);
2291 			if (ret < 0)
2292 				return false;
2293 
2294 			names += ret;
2295 			size -= ret;
2296 		} else
2297 			append_sep = true;
2298 
2299 		ret = strscpy(names, cur->name, size);
2300 		if (ret < 0)
2301 			return false;
2302 
2303 		names += ret;
2304 		size -= ret;
2305 	}
2306 
2307 	return true;
2308 }
2309 
2310 static bool seccomp_action_logged_from_name(u32 *action_logged,
2311 					    const char *name)
2312 {
2313 	const struct seccomp_log_name *cur;
2314 
2315 	for (cur = seccomp_log_names; cur->name; cur++) {
2316 		if (!strcmp(cur->name, name)) {
2317 			*action_logged = cur->log;
2318 			return true;
2319 		}
2320 	}
2321 
2322 	return false;
2323 }
2324 
2325 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
2326 {
2327 	char *name;
2328 
2329 	*actions_logged = 0;
2330 	while ((name = strsep(&names, " ")) && *name) {
2331 		u32 action_logged = 0;
2332 
2333 		if (!seccomp_action_logged_from_name(&action_logged, name))
2334 			return false;
2335 
2336 		*actions_logged |= action_logged;
2337 	}
2338 
2339 	return true;
2340 }
2341 
2342 static int read_actions_logged(const struct ctl_table *ro_table, void *buffer,
2343 			       size_t *lenp, loff_t *ppos)
2344 {
2345 	char names[sizeof(seccomp_actions_avail)];
2346 	struct ctl_table table;
2347 
2348 	memset(names, 0, sizeof(names));
2349 
2350 	if (!seccomp_names_from_actions_logged(names, sizeof(names),
2351 					       seccomp_actions_logged, " "))
2352 		return -EINVAL;
2353 
2354 	table = *ro_table;
2355 	table.data = names;
2356 	table.maxlen = sizeof(names);
2357 	return proc_dostring(&table, 0, buffer, lenp, ppos);
2358 }
2359 
2360 static int write_actions_logged(const struct ctl_table *ro_table, void *buffer,
2361 				size_t *lenp, loff_t *ppos, u32 *actions_logged)
2362 {
2363 	char names[sizeof(seccomp_actions_avail)];
2364 	struct ctl_table table;
2365 	int ret;
2366 
2367 	if (!capable(CAP_SYS_ADMIN))
2368 		return -EPERM;
2369 
2370 	memset(names, 0, sizeof(names));
2371 
2372 	table = *ro_table;
2373 	table.data = names;
2374 	table.maxlen = sizeof(names);
2375 	ret = proc_dostring(&table, 1, buffer, lenp, ppos);
2376 	if (ret)
2377 		return ret;
2378 
2379 	if (!seccomp_actions_logged_from_names(actions_logged, table.data))
2380 		return -EINVAL;
2381 
2382 	if (*actions_logged & SECCOMP_LOG_ALLOW)
2383 		return -EINVAL;
2384 
2385 	seccomp_actions_logged = *actions_logged;
2386 	return 0;
2387 }
2388 
2389 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
2390 				 int ret)
2391 {
2392 	char names[sizeof(seccomp_actions_avail)];
2393 	char old_names[sizeof(seccomp_actions_avail)];
2394 	const char *new = names;
2395 	const char *old = old_names;
2396 
2397 	if (!audit_enabled)
2398 		return;
2399 
2400 	memset(names, 0, sizeof(names));
2401 	memset(old_names, 0, sizeof(old_names));
2402 
2403 	if (ret)
2404 		new = "?";
2405 	else if (!actions_logged)
2406 		new = "(none)";
2407 	else if (!seccomp_names_from_actions_logged(names, sizeof(names),
2408 						    actions_logged, ","))
2409 		new = "?";
2410 
2411 	if (!old_actions_logged)
2412 		old = "(none)";
2413 	else if (!seccomp_names_from_actions_logged(old_names,
2414 						    sizeof(old_names),
2415 						    old_actions_logged, ","))
2416 		old = "?";
2417 
2418 	return audit_seccomp_actions_logged(new, old, !ret);
2419 }
2420 
2421 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
2422 					  void *buffer, size_t *lenp,
2423 					  loff_t *ppos)
2424 {
2425 	int ret;
2426 
2427 	if (write) {
2428 		u32 actions_logged = 0;
2429 		u32 old_actions_logged = seccomp_actions_logged;
2430 
2431 		ret = write_actions_logged(ro_table, buffer, lenp, ppos,
2432 					   &actions_logged);
2433 		audit_actions_logged(actions_logged, old_actions_logged, ret);
2434 	} else
2435 		ret = read_actions_logged(ro_table, buffer, lenp, ppos);
2436 
2437 	return ret;
2438 }
2439 
2440 static struct ctl_table seccomp_sysctl_table[] = {
2441 	{
2442 		.procname	= "actions_avail",
2443 		.data		= (void *) &seccomp_actions_avail,
2444 		.maxlen		= sizeof(seccomp_actions_avail),
2445 		.mode		= 0444,
2446 		.proc_handler	= proc_dostring,
2447 	},
2448 	{
2449 		.procname	= "actions_logged",
2450 		.mode		= 0644,
2451 		.proc_handler	= seccomp_actions_logged_handler,
2452 	},
2453 };
2454 
2455 static int __init seccomp_sysctl_init(void)
2456 {
2457 	register_sysctl_init("kernel/seccomp", seccomp_sysctl_table);
2458 	return 0;
2459 }
2460 
2461 device_initcall(seccomp_sysctl_init)
2462 
2463 #endif /* CONFIG_SYSCTL */
2464 
2465 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
2466 /* Currently CONFIG_SECCOMP_CACHE_DEBUG implies SECCOMP_ARCH_NATIVE */
2467 static void proc_pid_seccomp_cache_arch(struct seq_file *m, const char *name,
2468 					const void *bitmap, size_t bitmap_size)
2469 {
2470 	int nr;
2471 
2472 	for (nr = 0; nr < bitmap_size; nr++) {
2473 		bool cached = test_bit(nr, bitmap);
2474 		char *status = cached ? "ALLOW" : "FILTER";
2475 
2476 		seq_printf(m, "%s %d %s\n", name, nr, status);
2477 	}
2478 }
2479 
2480 int proc_pid_seccomp_cache(struct seq_file *m, struct pid_namespace *ns,
2481 			   struct pid *pid, struct task_struct *task)
2482 {
2483 	struct seccomp_filter *f;
2484 	unsigned long flags;
2485 
2486 	/*
2487 	 * We don't want some sandboxed process to know what their seccomp
2488 	 * filters consist of.
2489 	 */
2490 	if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
2491 		return -EACCES;
2492 
2493 	if (!lock_task_sighand(task, &flags))
2494 		return -ESRCH;
2495 
2496 	f = READ_ONCE(task->seccomp.filter);
2497 	if (!f) {
2498 		unlock_task_sighand(task, &flags);
2499 		return 0;
2500 	}
2501 
2502 	/* prevent filter from being freed while we are printing it */
2503 	__get_seccomp_filter(f);
2504 	unlock_task_sighand(task, &flags);
2505 
2506 	proc_pid_seccomp_cache_arch(m, SECCOMP_ARCH_NATIVE_NAME,
2507 				    f->cache.allow_native,
2508 				    SECCOMP_ARCH_NATIVE_NR);
2509 
2510 #ifdef SECCOMP_ARCH_COMPAT
2511 	proc_pid_seccomp_cache_arch(m, SECCOMP_ARCH_COMPAT_NAME,
2512 				    f->cache.allow_compat,
2513 				    SECCOMP_ARCH_COMPAT_NR);
2514 #endif /* SECCOMP_ARCH_COMPAT */
2515 
2516 	__put_seccomp_filter(f);
2517 	return 0;
2518 }
2519 #endif /* CONFIG_SECCOMP_CACHE_DEBUG */
2520