xref: /linux-6.15/kernel/seccomp.c (revision e1cec510)
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 #include <asm/syscall.h>
33 
34 /* Not exposed in headers: strictly internal use only. */
35 #define SECCOMP_MODE_DEAD	(SECCOMP_MODE_FILTER + 1)
36 
37 #ifdef CONFIG_SECCOMP_FILTER
38 #include <linux/file.h>
39 #include <linux/filter.h>
40 #include <linux/pid.h>
41 #include <linux/ptrace.h>
42 #include <linux/capability.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  * @setfd: whether or not SECCOMP_ADDFD_FLAG_SETFD was set during notify_addfd
112  * @ret: The return value of the installing process. It is set to the fd num
113  *       upon success (>= 0).
114  * @completion: Indicates that the installing process has completed fd
115  *              installation, or gone away (either due to successful
116  *              reply, or signal)
117  * @list: list_head for chaining seccomp_kaddfd together.
118  *
119  */
120 struct seccomp_kaddfd {
121 	struct file *file;
122 	int fd;
123 	unsigned int flags;
124 	__u32 ioctl_flags;
125 
126 	union {
127 		bool setfd;
128 		/* To only be set on reply */
129 		int ret;
130 	};
131 	struct completion completion;
132 	struct list_head list;
133 };
134 
135 /**
136  * struct notification - container for seccomp userspace notifications. Since
137  * most seccomp filters will not have notification listeners attached and this
138  * structure is fairly large, we store the notification-specific stuff in a
139  * separate structure.
140  *
141  * @requests: A semaphore that users of this notification can wait on for
142  *            changes. Actual reads and writes are still controlled with
143  *            filter->notify_lock.
144  * @flags: A set of SECCOMP_USER_NOTIF_FD_* flags.
145  * @next_id: The id of the next request.
146  * @notifications: A list of struct seccomp_knotif elements.
147  */
148 
149 struct notification {
150 	atomic_t requests;
151 	u32 flags;
152 	u64 next_id;
153 	struct list_head notifications;
154 };
155 
156 #ifdef SECCOMP_ARCH_NATIVE
157 /**
158  * struct action_cache - per-filter cache of seccomp actions per
159  * arch/syscall pair
160  *
161  * @allow_native: A bitmap where each bit represents whether the
162  *		  filter will always allow the syscall, for the
163  *		  native architecture.
164  * @allow_compat: A bitmap where each bit represents whether the
165  *		  filter will always allow the syscall, for the
166  *		  compat architecture.
167  */
168 struct action_cache {
169 	DECLARE_BITMAP(allow_native, SECCOMP_ARCH_NATIVE_NR);
170 #ifdef SECCOMP_ARCH_COMPAT
171 	DECLARE_BITMAP(allow_compat, SECCOMP_ARCH_COMPAT_NR);
172 #endif
173 };
174 #else
175 struct action_cache { };
176 
177 static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilter,
178 					     const struct seccomp_data *sd)
179 {
180 	return false;
181 }
182 
183 static inline void seccomp_cache_prepare(struct seccomp_filter *sfilter)
184 {
185 }
186 #endif /* SECCOMP_ARCH_NATIVE */
187 
188 /**
189  * struct seccomp_filter - container for seccomp BPF programs
190  *
191  * @refs: Reference count to manage the object lifetime.
192  *	  A filter's reference count is incremented for each directly
193  *	  attached task, once for the dependent filter, and if
194  *	  requested for the user notifier. When @refs reaches zero,
195  *	  the filter can be freed.
196  * @users: A filter's @users count is incremented for each directly
197  *         attached task (filter installation, fork(), thread_sync),
198  *	   and once for the dependent filter (tracked in filter->prev).
199  *	   When it reaches zero it indicates that no direct or indirect
200  *	   users of that filter exist. No new tasks can get associated with
201  *	   this filter after reaching 0. The @users count is always smaller
202  *	   or equal to @refs. Hence, reaching 0 for @users does not mean
203  *	   the filter can be freed.
204  * @cache: cache of arch/syscall mappings to actions
205  * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
206  * @wait_killable_recv: Put notifying process in killable state once the
207  *			notification is received by the userspace listener.
208  * @prev: points to a previously installed, or inherited, filter
209  * @prog: the BPF program to evaluate
210  * @notif: the struct that holds all notification related information
211  * @notify_lock: A lock for all notification-related accesses.
212  * @wqh: A wait queue for poll if a notifier is in use.
213  *
214  * seccomp_filter objects are organized in a tree linked via the @prev
215  * pointer.  For any task, it appears to be a singly-linked list starting
216  * with current->seccomp.filter, the most recently attached or inherited filter.
217  * However, multiple filters may share a @prev node, by way of fork(), which
218  * results in a unidirectional tree existing in memory.  This is similar to
219  * how namespaces work.
220  *
221  * seccomp_filter objects should never be modified after being attached
222  * to a task_struct (other than @refs).
223  */
224 struct seccomp_filter {
225 	refcount_t refs;
226 	refcount_t users;
227 	bool log;
228 	bool wait_killable_recv;
229 	struct action_cache cache;
230 	struct seccomp_filter *prev;
231 	struct bpf_prog *prog;
232 	struct notification *notif;
233 	struct mutex notify_lock;
234 	wait_queue_head_t wqh;
235 };
236 
237 /* Limit any path through the tree to 256KB worth of instructions. */
238 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
239 
240 /*
241  * Endianness is explicitly ignored and left for BPF program authors to manage
242  * as per the specific architecture.
243  */
244 static void populate_seccomp_data(struct seccomp_data *sd)
245 {
246 	/*
247 	 * Instead of using current_pt_reg(), we're already doing the work
248 	 * to safely fetch "current", so just use "task" everywhere below.
249 	 */
250 	struct task_struct *task = current;
251 	struct pt_regs *regs = task_pt_regs(task);
252 	unsigned long args[6];
253 
254 	sd->nr = syscall_get_nr(task, regs);
255 	sd->arch = syscall_get_arch(task);
256 	syscall_get_arguments(task, regs, args);
257 	sd->args[0] = args[0];
258 	sd->args[1] = args[1];
259 	sd->args[2] = args[2];
260 	sd->args[3] = args[3];
261 	sd->args[4] = args[4];
262 	sd->args[5] = args[5];
263 	sd->instruction_pointer = KSTK_EIP(task);
264 }
265 
266 /**
267  *	seccomp_check_filter - verify seccomp filter code
268  *	@filter: filter to verify
269  *	@flen: length of filter
270  *
271  * Takes a previously checked filter (by bpf_check_classic) and
272  * redirects all filter code that loads struct sk_buff data
273  * and related data through seccomp_bpf_load.  It also
274  * enforces length and alignment checking of those loads.
275  *
276  * Returns 0 if the rule set is legal or -EINVAL if not.
277  */
278 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
279 {
280 	int pc;
281 	for (pc = 0; pc < flen; pc++) {
282 		struct sock_filter *ftest = &filter[pc];
283 		u16 code = ftest->code;
284 		u32 k = ftest->k;
285 
286 		switch (code) {
287 		case BPF_LD | BPF_W | BPF_ABS:
288 			ftest->code = BPF_LDX | BPF_W | BPF_ABS;
289 			/* 32-bit aligned and not out of bounds. */
290 			if (k >= sizeof(struct seccomp_data) || k & 3)
291 				return -EINVAL;
292 			continue;
293 		case BPF_LD | BPF_W | BPF_LEN:
294 			ftest->code = BPF_LD | BPF_IMM;
295 			ftest->k = sizeof(struct seccomp_data);
296 			continue;
297 		case BPF_LDX | BPF_W | BPF_LEN:
298 			ftest->code = BPF_LDX | BPF_IMM;
299 			ftest->k = sizeof(struct seccomp_data);
300 			continue;
301 		/* Explicitly include allowed calls. */
302 		case BPF_RET | BPF_K:
303 		case BPF_RET | BPF_A:
304 		case BPF_ALU | BPF_ADD | BPF_K:
305 		case BPF_ALU | BPF_ADD | BPF_X:
306 		case BPF_ALU | BPF_SUB | BPF_K:
307 		case BPF_ALU | BPF_SUB | BPF_X:
308 		case BPF_ALU | BPF_MUL | BPF_K:
309 		case BPF_ALU | BPF_MUL | BPF_X:
310 		case BPF_ALU | BPF_DIV | BPF_K:
311 		case BPF_ALU | BPF_DIV | BPF_X:
312 		case BPF_ALU | BPF_AND | BPF_K:
313 		case BPF_ALU | BPF_AND | BPF_X:
314 		case BPF_ALU | BPF_OR | BPF_K:
315 		case BPF_ALU | BPF_OR | BPF_X:
316 		case BPF_ALU | BPF_XOR | BPF_K:
317 		case BPF_ALU | BPF_XOR | BPF_X:
318 		case BPF_ALU | BPF_LSH | BPF_K:
319 		case BPF_ALU | BPF_LSH | BPF_X:
320 		case BPF_ALU | BPF_RSH | BPF_K:
321 		case BPF_ALU | BPF_RSH | BPF_X:
322 		case BPF_ALU | BPF_NEG:
323 		case BPF_LD | BPF_IMM:
324 		case BPF_LDX | BPF_IMM:
325 		case BPF_MISC | BPF_TAX:
326 		case BPF_MISC | BPF_TXA:
327 		case BPF_LD | BPF_MEM:
328 		case BPF_LDX | BPF_MEM:
329 		case BPF_ST:
330 		case BPF_STX:
331 		case BPF_JMP | BPF_JA:
332 		case BPF_JMP | BPF_JEQ | BPF_K:
333 		case BPF_JMP | BPF_JEQ | BPF_X:
334 		case BPF_JMP | BPF_JGE | BPF_K:
335 		case BPF_JMP | BPF_JGE | BPF_X:
336 		case BPF_JMP | BPF_JGT | BPF_K:
337 		case BPF_JMP | BPF_JGT | BPF_X:
338 		case BPF_JMP | BPF_JSET | BPF_K:
339 		case BPF_JMP | BPF_JSET | BPF_X:
340 			continue;
341 		default:
342 			return -EINVAL;
343 		}
344 	}
345 	return 0;
346 }
347 
348 #ifdef SECCOMP_ARCH_NATIVE
349 static inline bool seccomp_cache_check_allow_bitmap(const void *bitmap,
350 						    size_t bitmap_size,
351 						    int syscall_nr)
352 {
353 	if (unlikely(syscall_nr < 0 || syscall_nr >= bitmap_size))
354 		return false;
355 	syscall_nr = array_index_nospec(syscall_nr, bitmap_size);
356 
357 	return test_bit(syscall_nr, bitmap);
358 }
359 
360 /**
361  * seccomp_cache_check_allow - lookup seccomp cache
362  * @sfilter: The seccomp filter
363  * @sd: The seccomp data to lookup the cache with
364  *
365  * Returns true if the seccomp_data is cached and allowed.
366  */
367 static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilter,
368 					     const struct seccomp_data *sd)
369 {
370 	int syscall_nr = sd->nr;
371 	const struct action_cache *cache = &sfilter->cache;
372 
373 #ifndef SECCOMP_ARCH_COMPAT
374 	/* A native-only architecture doesn't need to check sd->arch. */
375 	return seccomp_cache_check_allow_bitmap(cache->allow_native,
376 						SECCOMP_ARCH_NATIVE_NR,
377 						syscall_nr);
378 #else
379 	if (likely(sd->arch == SECCOMP_ARCH_NATIVE))
380 		return seccomp_cache_check_allow_bitmap(cache->allow_native,
381 							SECCOMP_ARCH_NATIVE_NR,
382 							syscall_nr);
383 	if (likely(sd->arch == SECCOMP_ARCH_COMPAT))
384 		return seccomp_cache_check_allow_bitmap(cache->allow_compat,
385 							SECCOMP_ARCH_COMPAT_NR,
386 							syscall_nr);
387 #endif /* SECCOMP_ARCH_COMPAT */
388 
389 	WARN_ON_ONCE(true);
390 	return false;
391 }
392 #endif /* SECCOMP_ARCH_NATIVE */
393 
394 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
395 /**
396  * seccomp_run_filters - evaluates all seccomp filters against @sd
397  * @sd: optional seccomp data to be passed to filters
398  * @match: stores struct seccomp_filter that resulted in the return value,
399  *         unless filter returned SECCOMP_RET_ALLOW, in which case it will
400  *         be unchanged.
401  *
402  * Returns valid seccomp BPF response codes.
403  */
404 static u32 seccomp_run_filters(const struct seccomp_data *sd,
405 			       struct seccomp_filter **match)
406 {
407 	u32 ret = SECCOMP_RET_ALLOW;
408 	/* Make sure cross-thread synced filter points somewhere sane. */
409 	struct seccomp_filter *f =
410 			READ_ONCE(current->seccomp.filter);
411 
412 	/* Ensure unexpected behavior doesn't result in failing open. */
413 	if (WARN_ON(f == NULL))
414 		return SECCOMP_RET_KILL_PROCESS;
415 
416 	if (seccomp_cache_check_allow(f, sd))
417 		return SECCOMP_RET_ALLOW;
418 
419 	/*
420 	 * All filters in the list are evaluated and the lowest BPF return
421 	 * value always takes priority (ignoring the DATA).
422 	 */
423 	for (; f; f = f->prev) {
424 		u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd);
425 
426 		if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
427 			ret = cur_ret;
428 			*match = f;
429 		}
430 	}
431 	return ret;
432 }
433 #endif /* CONFIG_SECCOMP_FILTER */
434 
435 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
436 {
437 	assert_spin_locked(&current->sighand->siglock);
438 
439 	if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
440 		return false;
441 
442 	return true;
443 }
444 
445 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
446 
447 static inline void seccomp_assign_mode(struct task_struct *task,
448 				       unsigned long seccomp_mode,
449 				       unsigned long flags)
450 {
451 	assert_spin_locked(&task->sighand->siglock);
452 
453 	task->seccomp.mode = seccomp_mode;
454 	/*
455 	 * Make sure SYSCALL_WORK_SECCOMP cannot be set before the mode (and
456 	 * filter) is set.
457 	 */
458 	smp_mb__before_atomic();
459 	/* Assume default seccomp processes want spec flaw mitigation. */
460 	if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
461 		arch_seccomp_spec_mitigate(task);
462 	set_task_syscall_work(task, SECCOMP);
463 }
464 
465 #ifdef CONFIG_SECCOMP_FILTER
466 /* Returns 1 if the parent is an ancestor of the child. */
467 static int is_ancestor(struct seccomp_filter *parent,
468 		       struct seccomp_filter *child)
469 {
470 	/* NULL is the root ancestor. */
471 	if (parent == NULL)
472 		return 1;
473 	for (; child; child = child->prev)
474 		if (child == parent)
475 			return 1;
476 	return 0;
477 }
478 
479 /**
480  * seccomp_can_sync_threads: checks if all threads can be synchronized
481  *
482  * Expects sighand and cred_guard_mutex locks to be held.
483  *
484  * Returns 0 on success, -ve on error, or the pid of a thread which was
485  * either not in the correct seccomp mode or did not have an ancestral
486  * seccomp filter.
487  */
488 static inline pid_t seccomp_can_sync_threads(void)
489 {
490 	struct task_struct *thread, *caller;
491 
492 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
493 	assert_spin_locked(&current->sighand->siglock);
494 
495 	/* Validate all threads being eligible for synchronization. */
496 	caller = current;
497 	for_each_thread(caller, thread) {
498 		pid_t failed;
499 
500 		/* Skip current, since it is initiating the sync. */
501 		if (thread == caller)
502 			continue;
503 		/* Skip exited threads. */
504 		if (thread->flags & PF_EXITING)
505 			continue;
506 
507 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
508 		    (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
509 		     is_ancestor(thread->seccomp.filter,
510 				 caller->seccomp.filter)))
511 			continue;
512 
513 		/* Return the first thread that cannot be synchronized. */
514 		failed = task_pid_vnr(thread);
515 		/* If the pid cannot be resolved, then return -ESRCH */
516 		if (WARN_ON(failed == 0))
517 			failed = -ESRCH;
518 		return failed;
519 	}
520 
521 	return 0;
522 }
523 
524 static inline void seccomp_filter_free(struct seccomp_filter *filter)
525 {
526 	if (filter) {
527 		bpf_prog_destroy(filter->prog);
528 		kfree(filter);
529 	}
530 }
531 
532 static void __seccomp_filter_orphan(struct seccomp_filter *orig)
533 {
534 	while (orig && refcount_dec_and_test(&orig->users)) {
535 		if (waitqueue_active(&orig->wqh))
536 			wake_up_poll(&orig->wqh, EPOLLHUP);
537 		orig = orig->prev;
538 	}
539 }
540 
541 static void __put_seccomp_filter(struct seccomp_filter *orig)
542 {
543 	/* Clean up single-reference branches iteratively. */
544 	while (orig && refcount_dec_and_test(&orig->refs)) {
545 		struct seccomp_filter *freeme = orig;
546 		orig = orig->prev;
547 		seccomp_filter_free(freeme);
548 	}
549 }
550 
551 static void __seccomp_filter_release(struct seccomp_filter *orig)
552 {
553 	/* Notify about any unused filters in the task's former filter tree. */
554 	__seccomp_filter_orphan(orig);
555 	/* Finally drop all references to the task's former tree. */
556 	__put_seccomp_filter(orig);
557 }
558 
559 /**
560  * seccomp_filter_release - Detach the task from its filter tree,
561  *			    drop its reference count, and notify
562  *			    about unused filters
563  *
564  * @tsk: task the filter should be released from.
565  *
566  * This function should only be called when the task is exiting as
567  * it detaches it from its filter tree. PF_EXITING has to be set
568  * for the task.
569  */
570 void seccomp_filter_release(struct task_struct *tsk)
571 {
572 	struct seccomp_filter *orig;
573 
574 	if (WARN_ON((tsk->flags & PF_EXITING) == 0))
575 		return;
576 
577 	spin_lock_irq(&tsk->sighand->siglock);
578 	orig = tsk->seccomp.filter;
579 	/* Detach task from its filter tree. */
580 	tsk->seccomp.filter = NULL;
581 	spin_unlock_irq(&tsk->sighand->siglock);
582 	__seccomp_filter_release(orig);
583 }
584 
585 /**
586  * seccomp_sync_threads: sets all threads to use current's filter
587  *
588  * @flags: SECCOMP_FILTER_FLAG_* flags to set during sync.
589  *
590  * Expects sighand and cred_guard_mutex locks to be held, and for
591  * seccomp_can_sync_threads() to have returned success already
592  * without dropping the locks.
593  *
594  */
595 static inline void seccomp_sync_threads(unsigned long flags)
596 {
597 	struct task_struct *thread, *caller;
598 
599 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
600 	assert_spin_locked(&current->sighand->siglock);
601 
602 	/* Synchronize all threads. */
603 	caller = current;
604 	for_each_thread(caller, thread) {
605 		/* Skip current, since it needs no changes. */
606 		if (thread == caller)
607 			continue;
608 
609 		/*
610 		 * Skip exited threads. seccomp_filter_release could have
611 		 * been already called for this task.
612 		 */
613 		if (thread->flags & PF_EXITING)
614 			continue;
615 
616 		/* Get a task reference for the new leaf node. */
617 		get_seccomp_filter(caller);
618 
619 		/*
620 		 * Drop the task reference to the shared ancestor since
621 		 * current's path will hold a reference.  (This also
622 		 * allows a put before the assignment.)
623 		 */
624 		__seccomp_filter_release(thread->seccomp.filter);
625 
626 		/* Make our new filter tree visible. */
627 		smp_store_release(&thread->seccomp.filter,
628 				  caller->seccomp.filter);
629 		atomic_set(&thread->seccomp.filter_count,
630 			   atomic_read(&caller->seccomp.filter_count));
631 
632 		/*
633 		 * Don't let an unprivileged task work around
634 		 * the no_new_privs restriction by creating
635 		 * a thread that sets it up, enters seccomp,
636 		 * then dies.
637 		 */
638 		if (task_no_new_privs(caller))
639 			task_set_no_new_privs(thread);
640 
641 		/*
642 		 * Opt the other thread into seccomp if needed.
643 		 * As threads are considered to be trust-realm
644 		 * equivalent (see ptrace_may_access), it is safe to
645 		 * allow one thread to transition the other.
646 		 */
647 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
648 			seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
649 					    flags);
650 	}
651 }
652 
653 /**
654  * seccomp_prepare_filter: Prepares a seccomp filter for use.
655  * @fprog: BPF program to install
656  *
657  * Returns filter on success or an ERR_PTR on failure.
658  */
659 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
660 {
661 	struct seccomp_filter *sfilter;
662 	int ret;
663 	const bool save_orig =
664 #if defined(CONFIG_CHECKPOINT_RESTORE) || defined(SECCOMP_ARCH_NATIVE)
665 		true;
666 #else
667 		false;
668 #endif
669 
670 	if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
671 		return ERR_PTR(-EINVAL);
672 
673 	BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
674 
675 	/*
676 	 * Installing a seccomp filter requires that the task has
677 	 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
678 	 * This avoids scenarios where unprivileged tasks can affect the
679 	 * behavior of privileged children.
680 	 */
681 	if (!task_no_new_privs(current) &&
682 			!ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN))
683 		return ERR_PTR(-EACCES);
684 
685 	/* Allocate a new seccomp_filter */
686 	sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
687 	if (!sfilter)
688 		return ERR_PTR(-ENOMEM);
689 
690 	mutex_init(&sfilter->notify_lock);
691 	ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
692 					seccomp_check_filter, save_orig);
693 	if (ret < 0) {
694 		kfree(sfilter);
695 		return ERR_PTR(ret);
696 	}
697 
698 	refcount_set(&sfilter->refs, 1);
699 	refcount_set(&sfilter->users, 1);
700 	init_waitqueue_head(&sfilter->wqh);
701 
702 	return sfilter;
703 }
704 
705 /**
706  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
707  * @user_filter: pointer to the user data containing a sock_fprog.
708  *
709  * Returns 0 on success and non-zero otherwise.
710  */
711 static struct seccomp_filter *
712 seccomp_prepare_user_filter(const char __user *user_filter)
713 {
714 	struct sock_fprog fprog;
715 	struct seccomp_filter *filter = ERR_PTR(-EFAULT);
716 
717 #ifdef CONFIG_COMPAT
718 	if (in_compat_syscall()) {
719 		struct compat_sock_fprog fprog32;
720 		if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
721 			goto out;
722 		fprog.len = fprog32.len;
723 		fprog.filter = compat_ptr(fprog32.filter);
724 	} else /* falls through to the if below. */
725 #endif
726 	if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
727 		goto out;
728 	filter = seccomp_prepare_filter(&fprog);
729 out:
730 	return filter;
731 }
732 
733 #ifdef SECCOMP_ARCH_NATIVE
734 /**
735  * seccomp_is_const_allow - check if filter is constant allow with given data
736  * @fprog: The BPF programs
737  * @sd: The seccomp data to check against, only syscall number and arch
738  *      number are considered constant.
739  */
740 static bool seccomp_is_const_allow(struct sock_fprog_kern *fprog,
741 				   struct seccomp_data *sd)
742 {
743 	unsigned int reg_value = 0;
744 	unsigned int pc;
745 	bool op_res;
746 
747 	if (WARN_ON_ONCE(!fprog))
748 		return false;
749 
750 	/* Our single exception to filtering. */
751 #ifdef __NR_uretprobe
752 #ifdef SECCOMP_ARCH_COMPAT
753 	if (sd->arch == SECCOMP_ARCH_NATIVE)
754 #endif
755 		if (sd->nr == __NR_uretprobe)
756 			return true;
757 #endif
758 
759 	for (pc = 0; pc < fprog->len; pc++) {
760 		struct sock_filter *insn = &fprog->filter[pc];
761 		u16 code = insn->code;
762 		u32 k = insn->k;
763 
764 		switch (code) {
765 		case BPF_LD | BPF_W | BPF_ABS:
766 			switch (k) {
767 			case offsetof(struct seccomp_data, nr):
768 				reg_value = sd->nr;
769 				break;
770 			case offsetof(struct seccomp_data, arch):
771 				reg_value = sd->arch;
772 				break;
773 			default:
774 				/* can't optimize (non-constant value load) */
775 				return false;
776 			}
777 			break;
778 		case BPF_RET | BPF_K:
779 			/* reached return with constant values only, check allow */
780 			return k == SECCOMP_RET_ALLOW;
781 		case BPF_JMP | BPF_JA:
782 			pc += insn->k;
783 			break;
784 		case BPF_JMP | BPF_JEQ | BPF_K:
785 		case BPF_JMP | BPF_JGE | BPF_K:
786 		case BPF_JMP | BPF_JGT | BPF_K:
787 		case BPF_JMP | BPF_JSET | BPF_K:
788 			switch (BPF_OP(code)) {
789 			case BPF_JEQ:
790 				op_res = reg_value == k;
791 				break;
792 			case BPF_JGE:
793 				op_res = reg_value >= k;
794 				break;
795 			case BPF_JGT:
796 				op_res = reg_value > k;
797 				break;
798 			case BPF_JSET:
799 				op_res = !!(reg_value & k);
800 				break;
801 			default:
802 				/* can't optimize (unknown jump) */
803 				return false;
804 			}
805 
806 			pc += op_res ? insn->jt : insn->jf;
807 			break;
808 		case BPF_ALU | BPF_AND | BPF_K:
809 			reg_value &= k;
810 			break;
811 		default:
812 			/* can't optimize (unknown insn) */
813 			return false;
814 		}
815 	}
816 
817 	/* ran off the end of the filter?! */
818 	WARN_ON(1);
819 	return false;
820 }
821 
822 static void seccomp_cache_prepare_bitmap(struct seccomp_filter *sfilter,
823 					 void *bitmap, const void *bitmap_prev,
824 					 size_t bitmap_size, int arch)
825 {
826 	struct sock_fprog_kern *fprog = sfilter->prog->orig_prog;
827 	struct seccomp_data sd;
828 	int nr;
829 
830 	if (bitmap_prev) {
831 		/* The new filter must be as restrictive as the last. */
832 		bitmap_copy(bitmap, bitmap_prev, bitmap_size);
833 	} else {
834 		/* Before any filters, all syscalls are always allowed. */
835 		bitmap_fill(bitmap, bitmap_size);
836 	}
837 
838 	for (nr = 0; nr < bitmap_size; nr++) {
839 		/* No bitmap change: not a cacheable action. */
840 		if (!test_bit(nr, bitmap))
841 			continue;
842 
843 		sd.nr = nr;
844 		sd.arch = arch;
845 
846 		/* No bitmap change: continue to always allow. */
847 		if (seccomp_is_const_allow(fprog, &sd))
848 			continue;
849 
850 		/*
851 		 * Not a cacheable action: always run filters.
852 		 * atomic clear_bit() not needed, filter not visible yet.
853 		 */
854 		__clear_bit(nr, bitmap);
855 	}
856 }
857 
858 /**
859  * seccomp_cache_prepare - emulate the filter to find cacheable syscalls
860  * @sfilter: The seccomp filter
861  *
862  * Returns 0 if successful or -errno if error occurred.
863  */
864 static void seccomp_cache_prepare(struct seccomp_filter *sfilter)
865 {
866 	struct action_cache *cache = &sfilter->cache;
867 	const struct action_cache *cache_prev =
868 		sfilter->prev ? &sfilter->prev->cache : NULL;
869 
870 	seccomp_cache_prepare_bitmap(sfilter, cache->allow_native,
871 				     cache_prev ? cache_prev->allow_native : NULL,
872 				     SECCOMP_ARCH_NATIVE_NR,
873 				     SECCOMP_ARCH_NATIVE);
874 
875 #ifdef SECCOMP_ARCH_COMPAT
876 	seccomp_cache_prepare_bitmap(sfilter, cache->allow_compat,
877 				     cache_prev ? cache_prev->allow_compat : NULL,
878 				     SECCOMP_ARCH_COMPAT_NR,
879 				     SECCOMP_ARCH_COMPAT);
880 #endif /* SECCOMP_ARCH_COMPAT */
881 }
882 #endif /* SECCOMP_ARCH_NATIVE */
883 
884 /**
885  * seccomp_attach_filter: validate and attach filter
886  * @flags:  flags to change filter behavior
887  * @filter: seccomp filter to add to the current process
888  *
889  * Caller must be holding current->sighand->siglock lock.
890  *
891  * Returns 0 on success, -ve on error, or
892  *   - in TSYNC mode: the pid of a thread which was either not in the correct
893  *     seccomp mode or did not have an ancestral seccomp filter
894  *   - in NEW_LISTENER mode: the fd of the new listener
895  */
896 static long seccomp_attach_filter(unsigned int flags,
897 				  struct seccomp_filter *filter)
898 {
899 	unsigned long total_insns;
900 	struct seccomp_filter *walker;
901 
902 	assert_spin_locked(&current->sighand->siglock);
903 
904 	/* Validate resulting filter length. */
905 	total_insns = filter->prog->len;
906 	for (walker = current->seccomp.filter; walker; walker = walker->prev)
907 		total_insns += walker->prog->len + 4;  /* 4 instr penalty */
908 	if (total_insns > MAX_INSNS_PER_PATH)
909 		return -ENOMEM;
910 
911 	/* If thread sync has been requested, check that it is possible. */
912 	if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
913 		int ret;
914 
915 		ret = seccomp_can_sync_threads();
916 		if (ret) {
917 			if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
918 				return -ESRCH;
919 			else
920 				return ret;
921 		}
922 	}
923 
924 	/* Set log flag, if present. */
925 	if (flags & SECCOMP_FILTER_FLAG_LOG)
926 		filter->log = true;
927 
928 	/* Set wait killable flag, if present. */
929 	if (flags & SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV)
930 		filter->wait_killable_recv = true;
931 
932 	/*
933 	 * If there is an existing filter, make it the prev and don't drop its
934 	 * task reference.
935 	 */
936 	filter->prev = current->seccomp.filter;
937 	seccomp_cache_prepare(filter);
938 	current->seccomp.filter = filter;
939 	atomic_inc(&current->seccomp.filter_count);
940 
941 	/* Now that the new filter is in place, synchronize to all threads. */
942 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
943 		seccomp_sync_threads(flags);
944 
945 	return 0;
946 }
947 
948 static void __get_seccomp_filter(struct seccomp_filter *filter)
949 {
950 	refcount_inc(&filter->refs);
951 }
952 
953 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
954 void get_seccomp_filter(struct task_struct *tsk)
955 {
956 	struct seccomp_filter *orig = tsk->seccomp.filter;
957 	if (!orig)
958 		return;
959 	__get_seccomp_filter(orig);
960 	refcount_inc(&orig->users);
961 }
962 
963 #endif	/* CONFIG_SECCOMP_FILTER */
964 
965 /* For use with seccomp_actions_logged */
966 #define SECCOMP_LOG_KILL_PROCESS	(1 << 0)
967 #define SECCOMP_LOG_KILL_THREAD		(1 << 1)
968 #define SECCOMP_LOG_TRAP		(1 << 2)
969 #define SECCOMP_LOG_ERRNO		(1 << 3)
970 #define SECCOMP_LOG_TRACE		(1 << 4)
971 #define SECCOMP_LOG_LOG			(1 << 5)
972 #define SECCOMP_LOG_ALLOW		(1 << 6)
973 #define SECCOMP_LOG_USER_NOTIF		(1 << 7)
974 
975 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
976 				    SECCOMP_LOG_KILL_THREAD  |
977 				    SECCOMP_LOG_TRAP  |
978 				    SECCOMP_LOG_ERRNO |
979 				    SECCOMP_LOG_USER_NOTIF |
980 				    SECCOMP_LOG_TRACE |
981 				    SECCOMP_LOG_LOG;
982 
983 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
984 			       bool requested)
985 {
986 	bool log = false;
987 
988 	switch (action) {
989 	case SECCOMP_RET_ALLOW:
990 		break;
991 	case SECCOMP_RET_TRAP:
992 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
993 		break;
994 	case SECCOMP_RET_ERRNO:
995 		log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
996 		break;
997 	case SECCOMP_RET_TRACE:
998 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
999 		break;
1000 	case SECCOMP_RET_USER_NOTIF:
1001 		log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
1002 		break;
1003 	case SECCOMP_RET_LOG:
1004 		log = seccomp_actions_logged & SECCOMP_LOG_LOG;
1005 		break;
1006 	case SECCOMP_RET_KILL_THREAD:
1007 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
1008 		break;
1009 	case SECCOMP_RET_KILL_PROCESS:
1010 	default:
1011 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
1012 	}
1013 
1014 	/*
1015 	 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
1016 	 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
1017 	 * any action from being logged by removing the action name from the
1018 	 * seccomp_actions_logged sysctl.
1019 	 */
1020 	if (!log)
1021 		return;
1022 
1023 	audit_seccomp(syscall, signr, action);
1024 }
1025 
1026 /*
1027  * Secure computing mode 1 allows only read/write/exit/sigreturn.
1028  * To be fully secure this must be combined with rlimit
1029  * to limit the stack allocations too.
1030  */
1031 static const int mode1_syscalls[] = {
1032 	__NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
1033 #ifdef __NR_uretprobe
1034 	__NR_uretprobe,
1035 #endif
1036 	-1, /* negative terminated */
1037 };
1038 
1039 static void __secure_computing_strict(int this_syscall)
1040 {
1041 	const int *allowed_syscalls = mode1_syscalls;
1042 #ifdef CONFIG_COMPAT
1043 	if (in_compat_syscall())
1044 		allowed_syscalls = get_compat_mode1_syscalls();
1045 #endif
1046 	do {
1047 		if (*allowed_syscalls == this_syscall)
1048 			return;
1049 	} while (*++allowed_syscalls != -1);
1050 
1051 #ifdef SECCOMP_DEBUG
1052 	dump_stack();
1053 #endif
1054 	current->seccomp.mode = SECCOMP_MODE_DEAD;
1055 	seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
1056 	do_exit(SIGKILL);
1057 }
1058 
1059 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
1060 void secure_computing_strict(int this_syscall)
1061 {
1062 	int mode = current->seccomp.mode;
1063 
1064 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1065 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1066 		return;
1067 
1068 	if (mode == SECCOMP_MODE_DISABLED)
1069 		return;
1070 	else if (mode == SECCOMP_MODE_STRICT)
1071 		__secure_computing_strict(this_syscall);
1072 	else
1073 		BUG();
1074 }
1075 int __secure_computing(void)
1076 {
1077 	int this_syscall = syscall_get_nr(current, current_pt_regs());
1078 
1079 	secure_computing_strict(this_syscall);
1080 	return 0;
1081 }
1082 #else
1083 
1084 #ifdef CONFIG_SECCOMP_FILTER
1085 static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
1086 {
1087 	/*
1088 	 * Note: overflow is ok here, the id just needs to be unique per
1089 	 * filter.
1090 	 */
1091 	lockdep_assert_held(&filter->notify_lock);
1092 	return filter->notif->next_id++;
1093 }
1094 
1095 static void seccomp_handle_addfd(struct seccomp_kaddfd *addfd, struct seccomp_knotif *n)
1096 {
1097 	int fd;
1098 
1099 	/*
1100 	 * Remove the notification, and reset the list pointers, indicating
1101 	 * that it has been handled.
1102 	 */
1103 	list_del_init(&addfd->list);
1104 	if (!addfd->setfd)
1105 		fd = receive_fd(addfd->file, NULL, addfd->flags);
1106 	else
1107 		fd = receive_fd_replace(addfd->fd, addfd->file, addfd->flags);
1108 	addfd->ret = fd;
1109 
1110 	if (addfd->ioctl_flags & SECCOMP_ADDFD_FLAG_SEND) {
1111 		/* If we fail reset and return an error to the notifier */
1112 		if (fd < 0) {
1113 			n->state = SECCOMP_NOTIFY_SENT;
1114 		} else {
1115 			/* Return the FD we just added */
1116 			n->flags = 0;
1117 			n->error = 0;
1118 			n->val = fd;
1119 		}
1120 	}
1121 
1122 	/*
1123 	 * Mark the notification as completed. From this point, addfd mem
1124 	 * might be invalidated and we can't safely read it anymore.
1125 	 */
1126 	complete(&addfd->completion);
1127 }
1128 
1129 static bool should_sleep_killable(struct seccomp_filter *match,
1130 				  struct seccomp_knotif *n)
1131 {
1132 	return match->wait_killable_recv && n->state == SECCOMP_NOTIFY_SENT;
1133 }
1134 
1135 static int seccomp_do_user_notification(int this_syscall,
1136 					struct seccomp_filter *match,
1137 					const struct seccomp_data *sd)
1138 {
1139 	int err;
1140 	u32 flags = 0;
1141 	long ret = 0;
1142 	struct seccomp_knotif n = {};
1143 	struct seccomp_kaddfd *addfd, *tmp;
1144 
1145 	mutex_lock(&match->notify_lock);
1146 	err = -ENOSYS;
1147 	if (!match->notif)
1148 		goto out;
1149 
1150 	n.task = current;
1151 	n.state = SECCOMP_NOTIFY_INIT;
1152 	n.data = sd;
1153 	n.id = seccomp_next_notify_id(match);
1154 	init_completion(&n.ready);
1155 	list_add_tail(&n.list, &match->notif->notifications);
1156 	INIT_LIST_HEAD(&n.addfd);
1157 
1158 	atomic_inc(&match->notif->requests);
1159 	if (match->notif->flags & SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP)
1160 		wake_up_poll_on_current_cpu(&match->wqh, EPOLLIN | EPOLLRDNORM);
1161 	else
1162 		wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM);
1163 
1164 	/*
1165 	 * This is where we wait for a reply from userspace.
1166 	 */
1167 	do {
1168 		bool wait_killable = should_sleep_killable(match, &n);
1169 
1170 		mutex_unlock(&match->notify_lock);
1171 		if (wait_killable)
1172 			err = wait_for_completion_killable(&n.ready);
1173 		else
1174 			err = wait_for_completion_interruptible(&n.ready);
1175 		mutex_lock(&match->notify_lock);
1176 
1177 		if (err != 0) {
1178 			/*
1179 			 * Check to see if the notifcation got picked up and
1180 			 * whether we should switch to wait killable.
1181 			 */
1182 			if (!wait_killable && should_sleep_killable(match, &n))
1183 				continue;
1184 
1185 			goto interrupted;
1186 		}
1187 
1188 		addfd = list_first_entry_or_null(&n.addfd,
1189 						 struct seccomp_kaddfd, list);
1190 		/* Check if we were woken up by a addfd message */
1191 		if (addfd)
1192 			seccomp_handle_addfd(addfd, &n);
1193 
1194 	}  while (n.state != SECCOMP_NOTIFY_REPLIED);
1195 
1196 	ret = n.val;
1197 	err = n.error;
1198 	flags = n.flags;
1199 
1200 interrupted:
1201 	/* If there were any pending addfd calls, clear them out */
1202 	list_for_each_entry_safe(addfd, tmp, &n.addfd, list) {
1203 		/* The process went away before we got a chance to handle it */
1204 		addfd->ret = -ESRCH;
1205 		list_del_init(&addfd->list);
1206 		complete(&addfd->completion);
1207 	}
1208 
1209 	/*
1210 	 * Note that it's possible the listener died in between the time when
1211 	 * we were notified of a response (or a signal) and when we were able to
1212 	 * re-acquire the lock, so only delete from the list if the
1213 	 * notification actually exists.
1214 	 *
1215 	 * Also note that this test is only valid because there's no way to
1216 	 * *reattach* to a notifier right now. If one is added, we'll need to
1217 	 * keep track of the notif itself and make sure they match here.
1218 	 */
1219 	if (match->notif)
1220 		list_del(&n.list);
1221 out:
1222 	mutex_unlock(&match->notify_lock);
1223 
1224 	/* Userspace requests to continue the syscall. */
1225 	if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1226 		return 0;
1227 
1228 	syscall_set_return_value(current, current_pt_regs(),
1229 				 err, ret);
1230 	return -1;
1231 }
1232 
1233 static int __seccomp_filter(int this_syscall, const bool recheck_after_trace)
1234 {
1235 	u32 filter_ret, action;
1236 	struct seccomp_data sd;
1237 	struct seccomp_filter *match = NULL;
1238 	int data;
1239 
1240 	/*
1241 	 * Make sure that any changes to mode from another thread have
1242 	 * been seen after SYSCALL_WORK_SECCOMP was seen.
1243 	 */
1244 	smp_rmb();
1245 
1246 	populate_seccomp_data(&sd);
1247 
1248 	filter_ret = seccomp_run_filters(&sd, &match);
1249 	data = filter_ret & SECCOMP_RET_DATA;
1250 	action = filter_ret & SECCOMP_RET_ACTION_FULL;
1251 
1252 	switch (action) {
1253 	case SECCOMP_RET_ERRNO:
1254 		/* Set low-order bits as an errno, capped at MAX_ERRNO. */
1255 		if (data > MAX_ERRNO)
1256 			data = MAX_ERRNO;
1257 		syscall_set_return_value(current, current_pt_regs(),
1258 					 -data, 0);
1259 		goto skip;
1260 
1261 	case SECCOMP_RET_TRAP:
1262 		/* Show the handler the original registers. */
1263 		syscall_rollback(current, current_pt_regs());
1264 		/* Let the filter pass back 16 bits of data. */
1265 		force_sig_seccomp(this_syscall, data, false);
1266 		goto skip;
1267 
1268 	case SECCOMP_RET_TRACE:
1269 		/* We've been put in this state by the ptracer already. */
1270 		if (recheck_after_trace)
1271 			return 0;
1272 
1273 		/* ENOSYS these calls if there is no tracer attached. */
1274 		if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
1275 			syscall_set_return_value(current,
1276 						 current_pt_regs(),
1277 						 -ENOSYS, 0);
1278 			goto skip;
1279 		}
1280 
1281 		/* Allow the BPF to provide the event message */
1282 		ptrace_event(PTRACE_EVENT_SECCOMP, data);
1283 		/*
1284 		 * The delivery of a fatal signal during event
1285 		 * notification may silently skip tracer notification,
1286 		 * which could leave us with a potentially unmodified
1287 		 * syscall that the tracer would have liked to have
1288 		 * changed. Since the process is about to die, we just
1289 		 * force the syscall to be skipped and let the signal
1290 		 * kill the process and correctly handle any tracer exit
1291 		 * notifications.
1292 		 */
1293 		if (fatal_signal_pending(current))
1294 			goto skip;
1295 		/* Check if the tracer forced the syscall to be skipped. */
1296 		this_syscall = syscall_get_nr(current, current_pt_regs());
1297 		if (this_syscall < 0)
1298 			goto skip;
1299 
1300 		/*
1301 		 * Recheck the syscall, since it may have changed. This
1302 		 * intentionally uses a NULL struct seccomp_data to force
1303 		 * a reload of all registers. This does not goto skip since
1304 		 * a skip would have already been reported.
1305 		 */
1306 		if (__seccomp_filter(this_syscall, true))
1307 			return -1;
1308 
1309 		return 0;
1310 
1311 	case SECCOMP_RET_USER_NOTIF:
1312 		if (seccomp_do_user_notification(this_syscall, match, &sd))
1313 			goto skip;
1314 
1315 		return 0;
1316 
1317 	case SECCOMP_RET_LOG:
1318 		seccomp_log(this_syscall, 0, action, true);
1319 		return 0;
1320 
1321 	case SECCOMP_RET_ALLOW:
1322 		/*
1323 		 * Note that the "match" filter will always be NULL for
1324 		 * this action since SECCOMP_RET_ALLOW is the starting
1325 		 * state in seccomp_run_filters().
1326 		 */
1327 		return 0;
1328 
1329 	case SECCOMP_RET_KILL_THREAD:
1330 	case SECCOMP_RET_KILL_PROCESS:
1331 	default:
1332 		current->seccomp.mode = SECCOMP_MODE_DEAD;
1333 		seccomp_log(this_syscall, SIGSYS, action, true);
1334 		/* Dump core only if this is the last remaining thread. */
1335 		if (action != SECCOMP_RET_KILL_THREAD ||
1336 		    (atomic_read(&current->signal->live) == 1)) {
1337 			/* Show the original registers in the dump. */
1338 			syscall_rollback(current, current_pt_regs());
1339 			/* Trigger a coredump with SIGSYS */
1340 			force_sig_seccomp(this_syscall, data, true);
1341 		} else {
1342 			do_exit(SIGSYS);
1343 		}
1344 		return -1; /* skip the syscall go directly to signal handling */
1345 	}
1346 
1347 	unreachable();
1348 
1349 skip:
1350 	seccomp_log(this_syscall, 0, action, match ? match->log : false);
1351 	return -1;
1352 }
1353 #else
1354 static int __seccomp_filter(int this_syscall, const bool recheck_after_trace)
1355 {
1356 	BUG();
1357 
1358 	return -1;
1359 }
1360 #endif
1361 
1362 int __secure_computing(void)
1363 {
1364 	int mode = current->seccomp.mode;
1365 	int this_syscall;
1366 
1367 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1368 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1369 		return 0;
1370 
1371 	this_syscall = syscall_get_nr(current, current_pt_regs());
1372 
1373 	switch (mode) {
1374 	case SECCOMP_MODE_STRICT:
1375 		__secure_computing_strict(this_syscall);  /* may call do_exit */
1376 		return 0;
1377 	case SECCOMP_MODE_FILTER:
1378 		return __seccomp_filter(this_syscall, false);
1379 	/* Surviving SECCOMP_RET_KILL_* must be proactively impossible. */
1380 	case SECCOMP_MODE_DEAD:
1381 		WARN_ON_ONCE(1);
1382 		do_exit(SIGKILL);
1383 		return -1;
1384 	default:
1385 		BUG();
1386 	}
1387 }
1388 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1389 
1390 long prctl_get_seccomp(void)
1391 {
1392 	return current->seccomp.mode;
1393 }
1394 
1395 /**
1396  * seccomp_set_mode_strict: internal function for setting strict seccomp
1397  *
1398  * Once current->seccomp.mode is non-zero, it may not be changed.
1399  *
1400  * Returns 0 on success or -EINVAL on failure.
1401  */
1402 static long seccomp_set_mode_strict(void)
1403 {
1404 	const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
1405 	long ret = -EINVAL;
1406 
1407 	spin_lock_irq(&current->sighand->siglock);
1408 
1409 	if (!seccomp_may_assign_mode(seccomp_mode))
1410 		goto out;
1411 
1412 #ifdef TIF_NOTSC
1413 	disable_TSC();
1414 #endif
1415 	seccomp_assign_mode(current, seccomp_mode, 0);
1416 	ret = 0;
1417 
1418 out:
1419 	spin_unlock_irq(&current->sighand->siglock);
1420 
1421 	return ret;
1422 }
1423 
1424 #ifdef CONFIG_SECCOMP_FILTER
1425 static void seccomp_notify_free(struct seccomp_filter *filter)
1426 {
1427 	kfree(filter->notif);
1428 	filter->notif = NULL;
1429 }
1430 
1431 static void seccomp_notify_detach(struct seccomp_filter *filter)
1432 {
1433 	struct seccomp_knotif *knotif;
1434 
1435 	if (!filter)
1436 		return;
1437 
1438 	mutex_lock(&filter->notify_lock);
1439 
1440 	/*
1441 	 * If this file is being closed because e.g. the task who owned it
1442 	 * died, let's wake everyone up who was waiting on us.
1443 	 */
1444 	list_for_each_entry(knotif, &filter->notif->notifications, list) {
1445 		if (knotif->state == SECCOMP_NOTIFY_REPLIED)
1446 			continue;
1447 
1448 		knotif->state = SECCOMP_NOTIFY_REPLIED;
1449 		knotif->error = -ENOSYS;
1450 		knotif->val = 0;
1451 
1452 		/*
1453 		 * We do not need to wake up any pending addfd messages, as
1454 		 * the notifier will do that for us, as this just looks
1455 		 * like a standard reply.
1456 		 */
1457 		complete(&knotif->ready);
1458 	}
1459 
1460 	seccomp_notify_free(filter);
1461 	mutex_unlock(&filter->notify_lock);
1462 }
1463 
1464 static int seccomp_notify_release(struct inode *inode, struct file *file)
1465 {
1466 	struct seccomp_filter *filter = file->private_data;
1467 
1468 	seccomp_notify_detach(filter);
1469 	__put_seccomp_filter(filter);
1470 	return 0;
1471 }
1472 
1473 /* must be called with notif_lock held */
1474 static inline struct seccomp_knotif *
1475 find_notification(struct seccomp_filter *filter, u64 id)
1476 {
1477 	struct seccomp_knotif *cur;
1478 
1479 	lockdep_assert_held(&filter->notify_lock);
1480 
1481 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1482 		if (cur->id == id)
1483 			return cur;
1484 	}
1485 
1486 	return NULL;
1487 }
1488 
1489 static int recv_wake_function(wait_queue_entry_t *wait, unsigned int mode, int sync,
1490 				  void *key)
1491 {
1492 	/* Avoid a wakeup if event not interesting for us. */
1493 	if (key && !(key_to_poll(key) & (EPOLLIN | EPOLLERR | EPOLLHUP)))
1494 		return 0;
1495 	return autoremove_wake_function(wait, mode, sync, key);
1496 }
1497 
1498 static int recv_wait_event(struct seccomp_filter *filter)
1499 {
1500 	DEFINE_WAIT_FUNC(wait, recv_wake_function);
1501 	int ret;
1502 
1503 	if (refcount_read(&filter->users) == 0)
1504 		return 0;
1505 
1506 	if (atomic_dec_if_positive(&filter->notif->requests) >= 0)
1507 		return 0;
1508 
1509 	for (;;) {
1510 		ret = prepare_to_wait_event(&filter->wqh, &wait, TASK_INTERRUPTIBLE);
1511 
1512 		if (atomic_dec_if_positive(&filter->notif->requests) >= 0)
1513 			break;
1514 		if (refcount_read(&filter->users) == 0)
1515 			break;
1516 
1517 		if (ret)
1518 			return ret;
1519 
1520 		schedule();
1521 	}
1522 	finish_wait(&filter->wqh, &wait);
1523 	return 0;
1524 }
1525 
1526 static long seccomp_notify_recv(struct seccomp_filter *filter,
1527 				void __user *buf)
1528 {
1529 	struct seccomp_knotif *knotif = NULL, *cur;
1530 	struct seccomp_notif unotif;
1531 	ssize_t ret;
1532 
1533 	/* Verify that we're not given garbage to keep struct extensible. */
1534 	ret = check_zeroed_user(buf, sizeof(unotif));
1535 	if (ret < 0)
1536 		return ret;
1537 	if (!ret)
1538 		return -EINVAL;
1539 
1540 	memset(&unotif, 0, sizeof(unotif));
1541 
1542 	ret = recv_wait_event(filter);
1543 	if (ret < 0)
1544 		return ret;
1545 
1546 	mutex_lock(&filter->notify_lock);
1547 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1548 		if (cur->state == SECCOMP_NOTIFY_INIT) {
1549 			knotif = cur;
1550 			break;
1551 		}
1552 	}
1553 
1554 	/*
1555 	 * If we didn't find a notification, it could be that the task was
1556 	 * interrupted by a fatal signal between the time we were woken and
1557 	 * when we were able to acquire the rw lock.
1558 	 */
1559 	if (!knotif) {
1560 		ret = -ENOENT;
1561 		goto out;
1562 	}
1563 
1564 	unotif.id = knotif->id;
1565 	unotif.pid = task_pid_vnr(knotif->task);
1566 	unotif.data = *(knotif->data);
1567 
1568 	knotif->state = SECCOMP_NOTIFY_SENT;
1569 	wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM);
1570 	ret = 0;
1571 out:
1572 	mutex_unlock(&filter->notify_lock);
1573 
1574 	if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1575 		ret = -EFAULT;
1576 
1577 		/*
1578 		 * Userspace screwed up. To make sure that we keep this
1579 		 * notification alive, let's reset it back to INIT. It
1580 		 * may have died when we released the lock, so we need to make
1581 		 * sure it's still around.
1582 		 */
1583 		mutex_lock(&filter->notify_lock);
1584 		knotif = find_notification(filter, unotif.id);
1585 		if (knotif) {
1586 			/* Reset the process to make sure it's not stuck */
1587 			if (should_sleep_killable(filter, knotif))
1588 				complete(&knotif->ready);
1589 			knotif->state = SECCOMP_NOTIFY_INIT;
1590 			atomic_inc(&filter->notif->requests);
1591 			wake_up_poll(&filter->wqh, EPOLLIN | EPOLLRDNORM);
1592 		}
1593 		mutex_unlock(&filter->notify_lock);
1594 	}
1595 
1596 	return ret;
1597 }
1598 
1599 static long seccomp_notify_send(struct seccomp_filter *filter,
1600 				void __user *buf)
1601 {
1602 	struct seccomp_notif_resp resp = {};
1603 	struct seccomp_knotif *knotif;
1604 	long ret;
1605 
1606 	if (copy_from_user(&resp, buf, sizeof(resp)))
1607 		return -EFAULT;
1608 
1609 	if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1610 		return -EINVAL;
1611 
1612 	if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) &&
1613 	    (resp.error || resp.val))
1614 		return -EINVAL;
1615 
1616 	ret = mutex_lock_interruptible(&filter->notify_lock);
1617 	if (ret < 0)
1618 		return ret;
1619 
1620 	knotif = find_notification(filter, resp.id);
1621 	if (!knotif) {
1622 		ret = -ENOENT;
1623 		goto out;
1624 	}
1625 
1626 	/* Allow exactly one reply. */
1627 	if (knotif->state != SECCOMP_NOTIFY_SENT) {
1628 		ret = -EINPROGRESS;
1629 		goto out;
1630 	}
1631 
1632 	ret = 0;
1633 	knotif->state = SECCOMP_NOTIFY_REPLIED;
1634 	knotif->error = resp.error;
1635 	knotif->val = resp.val;
1636 	knotif->flags = resp.flags;
1637 	if (filter->notif->flags & SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP)
1638 		complete_on_current_cpu(&knotif->ready);
1639 	else
1640 		complete(&knotif->ready);
1641 out:
1642 	mutex_unlock(&filter->notify_lock);
1643 	return ret;
1644 }
1645 
1646 static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1647 				    void __user *buf)
1648 {
1649 	struct seccomp_knotif *knotif;
1650 	u64 id;
1651 	long ret;
1652 
1653 	if (copy_from_user(&id, buf, sizeof(id)))
1654 		return -EFAULT;
1655 
1656 	ret = mutex_lock_interruptible(&filter->notify_lock);
1657 	if (ret < 0)
1658 		return ret;
1659 
1660 	knotif = find_notification(filter, id);
1661 	if (knotif && knotif->state == SECCOMP_NOTIFY_SENT)
1662 		ret = 0;
1663 	else
1664 		ret = -ENOENT;
1665 
1666 	mutex_unlock(&filter->notify_lock);
1667 	return ret;
1668 }
1669 
1670 static long seccomp_notify_set_flags(struct seccomp_filter *filter,
1671 				    unsigned long flags)
1672 {
1673 	long ret;
1674 
1675 	if (flags & ~SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP)
1676 		return -EINVAL;
1677 
1678 	ret = mutex_lock_interruptible(&filter->notify_lock);
1679 	if (ret < 0)
1680 		return ret;
1681 	filter->notif->flags = flags;
1682 	mutex_unlock(&filter->notify_lock);
1683 	return 0;
1684 }
1685 
1686 static long seccomp_notify_addfd(struct seccomp_filter *filter,
1687 				 struct seccomp_notif_addfd __user *uaddfd,
1688 				 unsigned int size)
1689 {
1690 	struct seccomp_notif_addfd addfd;
1691 	struct seccomp_knotif *knotif;
1692 	struct seccomp_kaddfd kaddfd;
1693 	int ret;
1694 
1695 	BUILD_BUG_ON(sizeof(addfd) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0);
1696 	BUILD_BUG_ON(sizeof(addfd) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST);
1697 
1698 	if (size < SECCOMP_NOTIFY_ADDFD_SIZE_VER0 || size >= PAGE_SIZE)
1699 		return -EINVAL;
1700 
1701 	ret = copy_struct_from_user(&addfd, sizeof(addfd), uaddfd, size);
1702 	if (ret)
1703 		return ret;
1704 
1705 	if (addfd.newfd_flags & ~O_CLOEXEC)
1706 		return -EINVAL;
1707 
1708 	if (addfd.flags & ~(SECCOMP_ADDFD_FLAG_SETFD | SECCOMP_ADDFD_FLAG_SEND))
1709 		return -EINVAL;
1710 
1711 	if (addfd.newfd && !(addfd.flags & SECCOMP_ADDFD_FLAG_SETFD))
1712 		return -EINVAL;
1713 
1714 	kaddfd.file = fget(addfd.srcfd);
1715 	if (!kaddfd.file)
1716 		return -EBADF;
1717 
1718 	kaddfd.ioctl_flags = addfd.flags;
1719 	kaddfd.flags = addfd.newfd_flags;
1720 	kaddfd.setfd = addfd.flags & SECCOMP_ADDFD_FLAG_SETFD;
1721 	kaddfd.fd = addfd.newfd;
1722 	init_completion(&kaddfd.completion);
1723 
1724 	ret = mutex_lock_interruptible(&filter->notify_lock);
1725 	if (ret < 0)
1726 		goto out;
1727 
1728 	knotif = find_notification(filter, addfd.id);
1729 	if (!knotif) {
1730 		ret = -ENOENT;
1731 		goto out_unlock;
1732 	}
1733 
1734 	/*
1735 	 * We do not want to allow for FD injection to occur before the
1736 	 * notification has been picked up by a userspace handler, or after
1737 	 * the notification has been replied to.
1738 	 */
1739 	if (knotif->state != SECCOMP_NOTIFY_SENT) {
1740 		ret = -EINPROGRESS;
1741 		goto out_unlock;
1742 	}
1743 
1744 	if (addfd.flags & SECCOMP_ADDFD_FLAG_SEND) {
1745 		/*
1746 		 * Disallow queuing an atomic addfd + send reply while there are
1747 		 * some addfd requests still to process.
1748 		 *
1749 		 * There is no clear reason to support it and allows us to keep
1750 		 * the loop on the other side straight-forward.
1751 		 */
1752 		if (!list_empty(&knotif->addfd)) {
1753 			ret = -EBUSY;
1754 			goto out_unlock;
1755 		}
1756 
1757 		/* Allow exactly only one reply */
1758 		knotif->state = SECCOMP_NOTIFY_REPLIED;
1759 	}
1760 
1761 	list_add(&kaddfd.list, &knotif->addfd);
1762 	complete(&knotif->ready);
1763 	mutex_unlock(&filter->notify_lock);
1764 
1765 	/* Now we wait for it to be processed or be interrupted */
1766 	ret = wait_for_completion_interruptible(&kaddfd.completion);
1767 	if (ret == 0) {
1768 		/*
1769 		 * We had a successful completion. The other side has already
1770 		 * removed us from the addfd queue, and
1771 		 * wait_for_completion_interruptible has a memory barrier upon
1772 		 * success that lets us read this value directly without
1773 		 * locking.
1774 		 */
1775 		ret = kaddfd.ret;
1776 		goto out;
1777 	}
1778 
1779 	mutex_lock(&filter->notify_lock);
1780 	/*
1781 	 * Even though we were woken up by a signal and not a successful
1782 	 * completion, a completion may have happened in the mean time.
1783 	 *
1784 	 * We need to check again if the addfd request has been handled,
1785 	 * and if not, we will remove it from the queue.
1786 	 */
1787 	if (list_empty(&kaddfd.list))
1788 		ret = kaddfd.ret;
1789 	else
1790 		list_del(&kaddfd.list);
1791 
1792 out_unlock:
1793 	mutex_unlock(&filter->notify_lock);
1794 out:
1795 	fput(kaddfd.file);
1796 
1797 	return ret;
1798 }
1799 
1800 static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1801 				 unsigned long arg)
1802 {
1803 	struct seccomp_filter *filter = file->private_data;
1804 	void __user *buf = (void __user *)arg;
1805 
1806 	/* Fixed-size ioctls */
1807 	switch (cmd) {
1808 	case SECCOMP_IOCTL_NOTIF_RECV:
1809 		return seccomp_notify_recv(filter, buf);
1810 	case SECCOMP_IOCTL_NOTIF_SEND:
1811 		return seccomp_notify_send(filter, buf);
1812 	case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR:
1813 	case SECCOMP_IOCTL_NOTIF_ID_VALID:
1814 		return seccomp_notify_id_valid(filter, buf);
1815 	case SECCOMP_IOCTL_NOTIF_SET_FLAGS:
1816 		return seccomp_notify_set_flags(filter, arg);
1817 	}
1818 
1819 	/* Extensible Argument ioctls */
1820 #define EA_IOCTL(cmd)	((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1821 	switch (EA_IOCTL(cmd)) {
1822 	case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD):
1823 		return seccomp_notify_addfd(filter, buf, _IOC_SIZE(cmd));
1824 	default:
1825 		return -EINVAL;
1826 	}
1827 }
1828 
1829 static __poll_t seccomp_notify_poll(struct file *file,
1830 				    struct poll_table_struct *poll_tab)
1831 {
1832 	struct seccomp_filter *filter = file->private_data;
1833 	__poll_t ret = 0;
1834 	struct seccomp_knotif *cur;
1835 
1836 	poll_wait(file, &filter->wqh, poll_tab);
1837 
1838 	if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1839 		return EPOLLERR;
1840 
1841 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1842 		if (cur->state == SECCOMP_NOTIFY_INIT)
1843 			ret |= EPOLLIN | EPOLLRDNORM;
1844 		if (cur->state == SECCOMP_NOTIFY_SENT)
1845 			ret |= EPOLLOUT | EPOLLWRNORM;
1846 		if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1847 			break;
1848 	}
1849 
1850 	mutex_unlock(&filter->notify_lock);
1851 
1852 	if (refcount_read(&filter->users) == 0)
1853 		ret |= EPOLLHUP;
1854 
1855 	return ret;
1856 }
1857 
1858 static const struct file_operations seccomp_notify_ops = {
1859 	.poll = seccomp_notify_poll,
1860 	.release = seccomp_notify_release,
1861 	.unlocked_ioctl = seccomp_notify_ioctl,
1862 	.compat_ioctl = seccomp_notify_ioctl,
1863 };
1864 
1865 static struct file *init_listener(struct seccomp_filter *filter)
1866 {
1867 	struct file *ret;
1868 
1869 	ret = ERR_PTR(-ENOMEM);
1870 	filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1871 	if (!filter->notif)
1872 		goto out;
1873 
1874 	filter->notif->next_id = get_random_u64();
1875 	INIT_LIST_HEAD(&filter->notif->notifications);
1876 
1877 	ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1878 				 filter, O_RDWR);
1879 	if (IS_ERR(ret))
1880 		goto out_notif;
1881 
1882 	/* The file has a reference to it now */
1883 	__get_seccomp_filter(filter);
1884 
1885 out_notif:
1886 	if (IS_ERR(ret))
1887 		seccomp_notify_free(filter);
1888 out:
1889 	return ret;
1890 }
1891 
1892 /*
1893  * Does @new_child have a listener while an ancestor also has a listener?
1894  * If so, we'll want to reject this filter.
1895  * This only has to be tested for the current process, even in the TSYNC case,
1896  * because TSYNC installs @child with the same parent on all threads.
1897  * Note that @new_child is not hooked up to its parent at this point yet, so
1898  * we use current->seccomp.filter.
1899  */
1900 static bool has_duplicate_listener(struct seccomp_filter *new_child)
1901 {
1902 	struct seccomp_filter *cur;
1903 
1904 	/* must be protected against concurrent TSYNC */
1905 	lockdep_assert_held(&current->sighand->siglock);
1906 
1907 	if (!new_child->notif)
1908 		return false;
1909 	for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1910 		if (cur->notif)
1911 			return true;
1912 	}
1913 
1914 	return false;
1915 }
1916 
1917 /**
1918  * seccomp_set_mode_filter: internal function for setting seccomp filter
1919  * @flags:  flags to change filter behavior
1920  * @filter: struct sock_fprog containing filter
1921  *
1922  * This function may be called repeatedly to install additional filters.
1923  * Every filter successfully installed will be evaluated (in reverse order)
1924  * for each system call the task makes.
1925  *
1926  * Once current->seccomp.mode is non-zero, it may not be changed.
1927  *
1928  * Returns 0 on success or -EINVAL on failure.
1929  */
1930 static long seccomp_set_mode_filter(unsigned int flags,
1931 				    const char __user *filter)
1932 {
1933 	const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1934 	struct seccomp_filter *prepared = NULL;
1935 	long ret = -EINVAL;
1936 	int listener = -1;
1937 	struct file *listener_f = NULL;
1938 
1939 	/* Validate flags. */
1940 	if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1941 		return -EINVAL;
1942 
1943 	/*
1944 	 * In the successful case, NEW_LISTENER returns the new listener fd.
1945 	 * But in the failure case, TSYNC returns the thread that died. If you
1946 	 * combine these two flags, there's no way to tell whether something
1947 	 * succeeded or failed. So, let's disallow this combination if the user
1948 	 * has not explicitly requested no errors from TSYNC.
1949 	 */
1950 	if ((flags & SECCOMP_FILTER_FLAG_TSYNC) &&
1951 	    (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) &&
1952 	    ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0))
1953 		return -EINVAL;
1954 
1955 	/*
1956 	 * The SECCOMP_FILTER_FLAG_WAIT_KILLABLE_SENT flag doesn't make sense
1957 	 * without the SECCOMP_FILTER_FLAG_NEW_LISTENER flag.
1958 	 */
1959 	if ((flags & SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV) &&
1960 	    ((flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) == 0))
1961 		return -EINVAL;
1962 
1963 	/* Prepare the new filter before holding any locks. */
1964 	prepared = seccomp_prepare_user_filter(filter);
1965 	if (IS_ERR(prepared))
1966 		return PTR_ERR(prepared);
1967 
1968 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1969 		listener = get_unused_fd_flags(O_CLOEXEC);
1970 		if (listener < 0) {
1971 			ret = listener;
1972 			goto out_free;
1973 		}
1974 
1975 		listener_f = init_listener(prepared);
1976 		if (IS_ERR(listener_f)) {
1977 			put_unused_fd(listener);
1978 			ret = PTR_ERR(listener_f);
1979 			goto out_free;
1980 		}
1981 	}
1982 
1983 	/*
1984 	 * Make sure we cannot change seccomp or nnp state via TSYNC
1985 	 * while another thread is in the middle of calling exec.
1986 	 */
1987 	if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1988 	    mutex_lock_killable(&current->signal->cred_guard_mutex))
1989 		goto out_put_fd;
1990 
1991 	spin_lock_irq(&current->sighand->siglock);
1992 
1993 	if (!seccomp_may_assign_mode(seccomp_mode))
1994 		goto out;
1995 
1996 	if (has_duplicate_listener(prepared)) {
1997 		ret = -EBUSY;
1998 		goto out;
1999 	}
2000 
2001 	ret = seccomp_attach_filter(flags, prepared);
2002 	if (ret)
2003 		goto out;
2004 	/* Do not free the successfully attached filter. */
2005 	prepared = NULL;
2006 
2007 	seccomp_assign_mode(current, seccomp_mode, flags);
2008 out:
2009 	spin_unlock_irq(&current->sighand->siglock);
2010 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
2011 		mutex_unlock(&current->signal->cred_guard_mutex);
2012 out_put_fd:
2013 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
2014 		if (ret) {
2015 			listener_f->private_data = NULL;
2016 			fput(listener_f);
2017 			put_unused_fd(listener);
2018 			seccomp_notify_detach(prepared);
2019 		} else {
2020 			fd_install(listener, listener_f);
2021 			ret = listener;
2022 		}
2023 	}
2024 out_free:
2025 	seccomp_filter_free(prepared);
2026 	return ret;
2027 }
2028 #else
2029 static inline long seccomp_set_mode_filter(unsigned int flags,
2030 					   const char __user *filter)
2031 {
2032 	return -EINVAL;
2033 }
2034 #endif
2035 
2036 static long seccomp_get_action_avail(const char __user *uaction)
2037 {
2038 	u32 action;
2039 
2040 	if (copy_from_user(&action, uaction, sizeof(action)))
2041 		return -EFAULT;
2042 
2043 	switch (action) {
2044 	case SECCOMP_RET_KILL_PROCESS:
2045 	case SECCOMP_RET_KILL_THREAD:
2046 	case SECCOMP_RET_TRAP:
2047 	case SECCOMP_RET_ERRNO:
2048 	case SECCOMP_RET_USER_NOTIF:
2049 	case SECCOMP_RET_TRACE:
2050 	case SECCOMP_RET_LOG:
2051 	case SECCOMP_RET_ALLOW:
2052 		break;
2053 	default:
2054 		return -EOPNOTSUPP;
2055 	}
2056 
2057 	return 0;
2058 }
2059 
2060 static long seccomp_get_notif_sizes(void __user *usizes)
2061 {
2062 	struct seccomp_notif_sizes sizes = {
2063 		.seccomp_notif = sizeof(struct seccomp_notif),
2064 		.seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
2065 		.seccomp_data = sizeof(struct seccomp_data),
2066 	};
2067 
2068 	if (copy_to_user(usizes, &sizes, sizeof(sizes)))
2069 		return -EFAULT;
2070 
2071 	return 0;
2072 }
2073 
2074 /* Common entry point for both prctl and syscall. */
2075 static long do_seccomp(unsigned int op, unsigned int flags,
2076 		       void __user *uargs)
2077 {
2078 	switch (op) {
2079 	case SECCOMP_SET_MODE_STRICT:
2080 		if (flags != 0 || uargs != NULL)
2081 			return -EINVAL;
2082 		return seccomp_set_mode_strict();
2083 	case SECCOMP_SET_MODE_FILTER:
2084 		return seccomp_set_mode_filter(flags, uargs);
2085 	case SECCOMP_GET_ACTION_AVAIL:
2086 		if (flags != 0)
2087 			return -EINVAL;
2088 
2089 		return seccomp_get_action_avail(uargs);
2090 	case SECCOMP_GET_NOTIF_SIZES:
2091 		if (flags != 0)
2092 			return -EINVAL;
2093 
2094 		return seccomp_get_notif_sizes(uargs);
2095 	default:
2096 		return -EINVAL;
2097 	}
2098 }
2099 
2100 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
2101 			 void __user *, uargs)
2102 {
2103 	return do_seccomp(op, flags, uargs);
2104 }
2105 
2106 /**
2107  * prctl_set_seccomp: configures current->seccomp.mode
2108  * @seccomp_mode: requested mode to use
2109  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
2110  *
2111  * Returns 0 on success or -EINVAL on failure.
2112  */
2113 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
2114 {
2115 	unsigned int op;
2116 	void __user *uargs;
2117 
2118 	switch (seccomp_mode) {
2119 	case SECCOMP_MODE_STRICT:
2120 		op = SECCOMP_SET_MODE_STRICT;
2121 		/*
2122 		 * Setting strict mode through prctl always ignored filter,
2123 		 * so make sure it is always NULL here to pass the internal
2124 		 * check in do_seccomp().
2125 		 */
2126 		uargs = NULL;
2127 		break;
2128 	case SECCOMP_MODE_FILTER:
2129 		op = SECCOMP_SET_MODE_FILTER;
2130 		uargs = filter;
2131 		break;
2132 	default:
2133 		return -EINVAL;
2134 	}
2135 
2136 	/* prctl interface doesn't have flags, so they are always zero. */
2137 	return do_seccomp(op, 0, uargs);
2138 }
2139 
2140 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
2141 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
2142 					     unsigned long filter_off)
2143 {
2144 	struct seccomp_filter *orig, *filter;
2145 	unsigned long count;
2146 
2147 	/*
2148 	 * Note: this is only correct because the caller should be the (ptrace)
2149 	 * tracer of the task, otherwise lock_task_sighand is needed.
2150 	 */
2151 	spin_lock_irq(&task->sighand->siglock);
2152 
2153 	if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
2154 		spin_unlock_irq(&task->sighand->siglock);
2155 		return ERR_PTR(-EINVAL);
2156 	}
2157 
2158 	orig = task->seccomp.filter;
2159 	__get_seccomp_filter(orig);
2160 	spin_unlock_irq(&task->sighand->siglock);
2161 
2162 	count = 0;
2163 	for (filter = orig; filter; filter = filter->prev)
2164 		count++;
2165 
2166 	if (filter_off >= count) {
2167 		filter = ERR_PTR(-ENOENT);
2168 		goto out;
2169 	}
2170 
2171 	count -= filter_off;
2172 	for (filter = orig; filter && count > 1; filter = filter->prev)
2173 		count--;
2174 
2175 	if (WARN_ON(count != 1 || !filter)) {
2176 		filter = ERR_PTR(-ENOENT);
2177 		goto out;
2178 	}
2179 
2180 	__get_seccomp_filter(filter);
2181 
2182 out:
2183 	__put_seccomp_filter(orig);
2184 	return filter;
2185 }
2186 
2187 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
2188 			void __user *data)
2189 {
2190 	struct seccomp_filter *filter;
2191 	struct sock_fprog_kern *fprog;
2192 	long ret;
2193 
2194 	if (!capable(CAP_SYS_ADMIN) ||
2195 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
2196 		return -EACCES;
2197 	}
2198 
2199 	filter = get_nth_filter(task, filter_off);
2200 	if (IS_ERR(filter))
2201 		return PTR_ERR(filter);
2202 
2203 	fprog = filter->prog->orig_prog;
2204 	if (!fprog) {
2205 		/* This must be a new non-cBPF filter, since we save
2206 		 * every cBPF filter's orig_prog above when
2207 		 * CONFIG_CHECKPOINT_RESTORE is enabled.
2208 		 */
2209 		ret = -EMEDIUMTYPE;
2210 		goto out;
2211 	}
2212 
2213 	ret = fprog->len;
2214 	if (!data)
2215 		goto out;
2216 
2217 	if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
2218 		ret = -EFAULT;
2219 
2220 out:
2221 	__put_seccomp_filter(filter);
2222 	return ret;
2223 }
2224 
2225 long seccomp_get_metadata(struct task_struct *task,
2226 			  unsigned long size, void __user *data)
2227 {
2228 	long ret;
2229 	struct seccomp_filter *filter;
2230 	struct seccomp_metadata kmd = {};
2231 
2232 	if (!capable(CAP_SYS_ADMIN) ||
2233 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
2234 		return -EACCES;
2235 	}
2236 
2237 	size = min_t(unsigned long, size, sizeof(kmd));
2238 
2239 	if (size < sizeof(kmd.filter_off))
2240 		return -EINVAL;
2241 
2242 	if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
2243 		return -EFAULT;
2244 
2245 	filter = get_nth_filter(task, kmd.filter_off);
2246 	if (IS_ERR(filter))
2247 		return PTR_ERR(filter);
2248 
2249 	if (filter->log)
2250 		kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
2251 
2252 	ret = size;
2253 	if (copy_to_user(data, &kmd, size))
2254 		ret = -EFAULT;
2255 
2256 	__put_seccomp_filter(filter);
2257 	return ret;
2258 }
2259 #endif
2260 
2261 #ifdef CONFIG_SYSCTL
2262 
2263 /* Human readable action names for friendly sysctl interaction */
2264 #define SECCOMP_RET_KILL_PROCESS_NAME	"kill_process"
2265 #define SECCOMP_RET_KILL_THREAD_NAME	"kill_thread"
2266 #define SECCOMP_RET_TRAP_NAME		"trap"
2267 #define SECCOMP_RET_ERRNO_NAME		"errno"
2268 #define SECCOMP_RET_USER_NOTIF_NAME	"user_notif"
2269 #define SECCOMP_RET_TRACE_NAME		"trace"
2270 #define SECCOMP_RET_LOG_NAME		"log"
2271 #define SECCOMP_RET_ALLOW_NAME		"allow"
2272 
2273 static const char seccomp_actions_avail[] =
2274 				SECCOMP_RET_KILL_PROCESS_NAME	" "
2275 				SECCOMP_RET_KILL_THREAD_NAME	" "
2276 				SECCOMP_RET_TRAP_NAME		" "
2277 				SECCOMP_RET_ERRNO_NAME		" "
2278 				SECCOMP_RET_USER_NOTIF_NAME     " "
2279 				SECCOMP_RET_TRACE_NAME		" "
2280 				SECCOMP_RET_LOG_NAME		" "
2281 				SECCOMP_RET_ALLOW_NAME;
2282 
2283 struct seccomp_log_name {
2284 	u32		log;
2285 	const char	*name;
2286 };
2287 
2288 static const struct seccomp_log_name seccomp_log_names[] = {
2289 	{ SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
2290 	{ SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
2291 	{ SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
2292 	{ SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
2293 	{ SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
2294 	{ SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
2295 	{ SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
2296 	{ SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
2297 	{ }
2298 };
2299 
2300 static bool seccomp_names_from_actions_logged(char *names, size_t size,
2301 					      u32 actions_logged,
2302 					      const char *sep)
2303 {
2304 	const struct seccomp_log_name *cur;
2305 	bool append_sep = false;
2306 
2307 	for (cur = seccomp_log_names; cur->name && size; cur++) {
2308 		ssize_t ret;
2309 
2310 		if (!(actions_logged & cur->log))
2311 			continue;
2312 
2313 		if (append_sep) {
2314 			ret = strscpy(names, sep, size);
2315 			if (ret < 0)
2316 				return false;
2317 
2318 			names += ret;
2319 			size -= ret;
2320 		} else
2321 			append_sep = true;
2322 
2323 		ret = strscpy(names, cur->name, size);
2324 		if (ret < 0)
2325 			return false;
2326 
2327 		names += ret;
2328 		size -= ret;
2329 	}
2330 
2331 	return true;
2332 }
2333 
2334 static bool seccomp_action_logged_from_name(u32 *action_logged,
2335 					    const char *name)
2336 {
2337 	const struct seccomp_log_name *cur;
2338 
2339 	for (cur = seccomp_log_names; cur->name; cur++) {
2340 		if (!strcmp(cur->name, name)) {
2341 			*action_logged = cur->log;
2342 			return true;
2343 		}
2344 	}
2345 
2346 	return false;
2347 }
2348 
2349 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
2350 {
2351 	char *name;
2352 
2353 	*actions_logged = 0;
2354 	while ((name = strsep(&names, " ")) && *name) {
2355 		u32 action_logged = 0;
2356 
2357 		if (!seccomp_action_logged_from_name(&action_logged, name))
2358 			return false;
2359 
2360 		*actions_logged |= action_logged;
2361 	}
2362 
2363 	return true;
2364 }
2365 
2366 static int read_actions_logged(const struct ctl_table *ro_table, void *buffer,
2367 			       size_t *lenp, loff_t *ppos)
2368 {
2369 	char names[sizeof(seccomp_actions_avail)];
2370 	struct ctl_table table;
2371 
2372 	memset(names, 0, sizeof(names));
2373 
2374 	if (!seccomp_names_from_actions_logged(names, sizeof(names),
2375 					       seccomp_actions_logged, " "))
2376 		return -EINVAL;
2377 
2378 	table = *ro_table;
2379 	table.data = names;
2380 	table.maxlen = sizeof(names);
2381 	return proc_dostring(&table, 0, buffer, lenp, ppos);
2382 }
2383 
2384 static int write_actions_logged(const struct ctl_table *ro_table, void *buffer,
2385 				size_t *lenp, loff_t *ppos, u32 *actions_logged)
2386 {
2387 	char names[sizeof(seccomp_actions_avail)];
2388 	struct ctl_table table;
2389 	int ret;
2390 
2391 	if (!capable(CAP_SYS_ADMIN))
2392 		return -EPERM;
2393 
2394 	memset(names, 0, sizeof(names));
2395 
2396 	table = *ro_table;
2397 	table.data = names;
2398 	table.maxlen = sizeof(names);
2399 	ret = proc_dostring(&table, 1, buffer, lenp, ppos);
2400 	if (ret)
2401 		return ret;
2402 
2403 	if (!seccomp_actions_logged_from_names(actions_logged, table.data))
2404 		return -EINVAL;
2405 
2406 	if (*actions_logged & SECCOMP_LOG_ALLOW)
2407 		return -EINVAL;
2408 
2409 	seccomp_actions_logged = *actions_logged;
2410 	return 0;
2411 }
2412 
2413 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
2414 				 int ret)
2415 {
2416 	char names[sizeof(seccomp_actions_avail)];
2417 	char old_names[sizeof(seccomp_actions_avail)];
2418 	const char *new = names;
2419 	const char *old = old_names;
2420 
2421 	if (!audit_enabled)
2422 		return;
2423 
2424 	memset(names, 0, sizeof(names));
2425 	memset(old_names, 0, sizeof(old_names));
2426 
2427 	if (ret)
2428 		new = "?";
2429 	else if (!actions_logged)
2430 		new = "(none)";
2431 	else if (!seccomp_names_from_actions_logged(names, sizeof(names),
2432 						    actions_logged, ","))
2433 		new = "?";
2434 
2435 	if (!old_actions_logged)
2436 		old = "(none)";
2437 	else if (!seccomp_names_from_actions_logged(old_names,
2438 						    sizeof(old_names),
2439 						    old_actions_logged, ","))
2440 		old = "?";
2441 
2442 	return audit_seccomp_actions_logged(new, old, !ret);
2443 }
2444 
2445 static int seccomp_actions_logged_handler(const struct ctl_table *ro_table, int write,
2446 					  void *buffer, size_t *lenp,
2447 					  loff_t *ppos)
2448 {
2449 	int ret;
2450 
2451 	if (write) {
2452 		u32 actions_logged = 0;
2453 		u32 old_actions_logged = seccomp_actions_logged;
2454 
2455 		ret = write_actions_logged(ro_table, buffer, lenp, ppos,
2456 					   &actions_logged);
2457 		audit_actions_logged(actions_logged, old_actions_logged, ret);
2458 	} else
2459 		ret = read_actions_logged(ro_table, buffer, lenp, ppos);
2460 
2461 	return ret;
2462 }
2463 
2464 static const struct ctl_table seccomp_sysctl_table[] = {
2465 	{
2466 		.procname	= "actions_avail",
2467 		.data		= (void *) &seccomp_actions_avail,
2468 		.maxlen		= sizeof(seccomp_actions_avail),
2469 		.mode		= 0444,
2470 		.proc_handler	= proc_dostring,
2471 	},
2472 	{
2473 		.procname	= "actions_logged",
2474 		.mode		= 0644,
2475 		.proc_handler	= seccomp_actions_logged_handler,
2476 	},
2477 };
2478 
2479 static int __init seccomp_sysctl_init(void)
2480 {
2481 	register_sysctl_init("kernel/seccomp", seccomp_sysctl_table);
2482 	return 0;
2483 }
2484 
2485 device_initcall(seccomp_sysctl_init)
2486 
2487 #endif /* CONFIG_SYSCTL */
2488 
2489 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
2490 /* Currently CONFIG_SECCOMP_CACHE_DEBUG implies SECCOMP_ARCH_NATIVE */
2491 static void proc_pid_seccomp_cache_arch(struct seq_file *m, const char *name,
2492 					const void *bitmap, size_t bitmap_size)
2493 {
2494 	int nr;
2495 
2496 	for (nr = 0; nr < bitmap_size; nr++) {
2497 		bool cached = test_bit(nr, bitmap);
2498 		char *status = cached ? "ALLOW" : "FILTER";
2499 
2500 		seq_printf(m, "%s %d %s\n", name, nr, status);
2501 	}
2502 }
2503 
2504 int proc_pid_seccomp_cache(struct seq_file *m, struct pid_namespace *ns,
2505 			   struct pid *pid, struct task_struct *task)
2506 {
2507 	struct seccomp_filter *f;
2508 	unsigned long flags;
2509 
2510 	/*
2511 	 * We don't want some sandboxed process to know what their seccomp
2512 	 * filters consist of.
2513 	 */
2514 	if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
2515 		return -EACCES;
2516 
2517 	if (!lock_task_sighand(task, &flags))
2518 		return -ESRCH;
2519 
2520 	f = READ_ONCE(task->seccomp.filter);
2521 	if (!f) {
2522 		unlock_task_sighand(task, &flags);
2523 		return 0;
2524 	}
2525 
2526 	/* prevent filter from being freed while we are printing it */
2527 	__get_seccomp_filter(f);
2528 	unlock_task_sighand(task, &flags);
2529 
2530 	proc_pid_seccomp_cache_arch(m, SECCOMP_ARCH_NATIVE_NAME,
2531 				    f->cache.allow_native,
2532 				    SECCOMP_ARCH_NATIVE_NR);
2533 
2534 #ifdef SECCOMP_ARCH_COMPAT
2535 	proc_pid_seccomp_cache_arch(m, SECCOMP_ARCH_COMPAT_NAME,
2536 				    f->cache.allow_compat,
2537 				    SECCOMP_ARCH_COMPAT_NR);
2538 #endif /* SECCOMP_ARCH_COMPAT */
2539 
2540 	__put_seccomp_filter(f);
2541 	return 0;
2542 }
2543 #endif /* CONFIG_SECCOMP_CACHE_DEBUG */
2544