1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/kernel/seccomp.c
4 *
5 * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com>
6 *
7 * Copyright (C) 2012 Google, Inc.
8 * Will Drewry <wad@chromium.org>
9 *
10 * This defines a simple but solid secure-computing facility.
11 *
12 * Mode 1 uses a fixed list of allowed system calls.
13 * Mode 2 allows user-defined system call filters in the form
14 * of Berkeley Packet Filters/Linux Socket Filters.
15 */
16 #define pr_fmt(fmt) "seccomp: " fmt
17
18 #include <linux/refcount.h>
19 #include <linux/audit.h>
20 #include <linux/compat.h>
21 #include <linux/coredump.h>
22 #include <linux/kmemleak.h>
23 #include <linux/nospec.h>
24 #include <linux/prctl.h>
25 #include <linux/sched.h>
26 #include <linux/sched/task_stack.h>
27 #include <linux/seccomp.h>
28 #include <linux/slab.h>
29 #include <linux/syscalls.h>
30 #include <linux/sysctl.h>
31
32 /* 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 * @ret: The return value of the installing process. It is set to the fd num
114 * upon success (>= 0).
115 * @completion: Indicates that the installing process has completed fd
116 * installation, or gone away (either due to successful
117 * reply, or signal)
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 * @request: 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 * @next_id: The id of the next request.
145 * @notifications: A list of struct seccomp_knotif elements.
146 */
147 struct notification {
148 struct semaphore request;
149 u64 next_id;
150 struct list_head notifications;
151 };
152
153 #ifdef SECCOMP_ARCH_NATIVE
154 /**
155 * struct action_cache - per-filter cache of seccomp actions per
156 * arch/syscall pair
157 *
158 * @allow_native: A bitmap where each bit represents whether the
159 * filter will always allow the syscall, for the
160 * native architecture.
161 * @allow_compat: A bitmap where each bit represents whether the
162 * filter will always allow the syscall, for the
163 * compat architecture.
164 */
165 struct action_cache {
166 DECLARE_BITMAP(allow_native, SECCOMP_ARCH_NATIVE_NR);
167 #ifdef SECCOMP_ARCH_COMPAT
168 DECLARE_BITMAP(allow_compat, SECCOMP_ARCH_COMPAT_NR);
169 #endif
170 };
171 #else
172 struct action_cache { };
173
seccomp_cache_check_allow(const struct seccomp_filter * sfilter,const struct seccomp_data * sd)174 static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilter,
175 const struct seccomp_data *sd)
176 {
177 return false;
178 }
179
seccomp_cache_prepare(struct seccomp_filter * sfilter)180 static inline void seccomp_cache_prepare(struct seccomp_filter *sfilter)
181 {
182 }
183 #endif /* SECCOMP_ARCH_NATIVE */
184
185 /**
186 * struct seccomp_filter - container for seccomp BPF programs
187 *
188 * @refs: Reference count to manage the object lifetime.
189 * A filter's reference count is incremented for each directly
190 * attached task, once for the dependent filter, and if
191 * requested for the user notifier. When @refs reaches zero,
192 * the filter can be freed.
193 * @users: A filter's @users count is incremented for each directly
194 * attached task (filter installation, fork(), thread_sync),
195 * and once for the dependent filter (tracked in filter->prev).
196 * When it reaches zero it indicates that no direct or indirect
197 * users of that filter exist. No new tasks can get associated with
198 * this filter after reaching 0. The @users count is always smaller
199 * or equal to @refs. Hence, reaching 0 for @users does not mean
200 * the filter can be freed.
201 * @cache: cache of arch/syscall mappings to actions
202 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
203 * @wait_killable_recv: Put notifying process in killable state once the
204 * notification is received by the userspace listener.
205 * @prev: points to a previously installed, or inherited, filter
206 * @prog: the BPF program to evaluate
207 * @notif: the struct that holds all notification related information
208 * @notify_lock: A lock for all notification-related accesses.
209 * @wqh: A wait queue for poll if a notifier is in use.
210 *
211 * seccomp_filter objects are organized in a tree linked via the @prev
212 * pointer. For any task, it appears to be a singly-linked list starting
213 * with current->seccomp.filter, the most recently attached or inherited filter.
214 * However, multiple filters may share a @prev node, by way of fork(), which
215 * results in a unidirectional tree existing in memory. This is similar to
216 * how namespaces work.
217 *
218 * seccomp_filter objects should never be modified after being attached
219 * to a task_struct (other than @refs).
220 */
221 struct seccomp_filter {
222 refcount_t refs;
223 refcount_t users;
224 bool log;
225 bool wait_killable_recv;
226 struct action_cache cache;
227 struct seccomp_filter *prev;
228 struct bpf_prog *prog;
229 struct notification *notif;
230 struct mutex notify_lock;
231 wait_queue_head_t wqh;
232 };
233
234 /* Limit any path through the tree to 256KB worth of instructions. */
235 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
236
237 /*
238 * Endianness is explicitly ignored and left for BPF program authors to manage
239 * as per the specific architecture.
240 */
populate_seccomp_data(struct seccomp_data * sd)241 static void populate_seccomp_data(struct seccomp_data *sd)
242 {
243 /*
244 * Instead of using current_pt_reg(), we're already doing the work
245 * to safely fetch "current", so just use "task" everywhere below.
246 */
247 struct task_struct *task = current;
248 struct pt_regs *regs = task_pt_regs(task);
249 unsigned long args[6];
250
251 sd->nr = syscall_get_nr(task, regs);
252 sd->arch = syscall_get_arch(task);
253 syscall_get_arguments(task, regs, args);
254 sd->args[0] = args[0];
255 sd->args[1] = args[1];
256 sd->args[2] = args[2];
257 sd->args[3] = args[3];
258 sd->args[4] = args[4];
259 sd->args[5] = args[5];
260 sd->instruction_pointer = KSTK_EIP(task);
261 }
262
263 /**
264 * seccomp_check_filter - verify seccomp filter code
265 * @filter: filter to verify
266 * @flen: length of filter
267 *
268 * Takes a previously checked filter (by bpf_check_classic) and
269 * redirects all filter code that loads struct sk_buff data
270 * and related data through seccomp_bpf_load. It also
271 * enforces length and alignment checking of those loads.
272 *
273 * Returns 0 if the rule set is legal or -EINVAL if not.
274 */
seccomp_check_filter(struct sock_filter * filter,unsigned int flen)275 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
276 {
277 int pc;
278 for (pc = 0; pc < flen; pc++) {
279 struct sock_filter *ftest = &filter[pc];
280 u16 code = ftest->code;
281 u32 k = ftest->k;
282
283 switch (code) {
284 case BPF_LD | BPF_W | BPF_ABS:
285 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
286 /* 32-bit aligned and not out of bounds. */
287 if (k >= sizeof(struct seccomp_data) || k & 3)
288 return -EINVAL;
289 continue;
290 case BPF_LD | BPF_W | BPF_LEN:
291 ftest->code = BPF_LD | BPF_IMM;
292 ftest->k = sizeof(struct seccomp_data);
293 continue;
294 case BPF_LDX | BPF_W | BPF_LEN:
295 ftest->code = BPF_LDX | BPF_IMM;
296 ftest->k = sizeof(struct seccomp_data);
297 continue;
298 /* Explicitly include allowed calls. */
299 case BPF_RET | BPF_K:
300 case BPF_RET | BPF_A:
301 case BPF_ALU | BPF_ADD | BPF_K:
302 case BPF_ALU | BPF_ADD | BPF_X:
303 case BPF_ALU | BPF_SUB | BPF_K:
304 case BPF_ALU | BPF_SUB | BPF_X:
305 case BPF_ALU | BPF_MUL | BPF_K:
306 case BPF_ALU | BPF_MUL | BPF_X:
307 case BPF_ALU | BPF_DIV | BPF_K:
308 case BPF_ALU | BPF_DIV | BPF_X:
309 case BPF_ALU | BPF_AND | BPF_K:
310 case BPF_ALU | BPF_AND | BPF_X:
311 case BPF_ALU | BPF_OR | BPF_K:
312 case BPF_ALU | BPF_OR | BPF_X:
313 case BPF_ALU | BPF_XOR | BPF_K:
314 case BPF_ALU | BPF_XOR | BPF_X:
315 case BPF_ALU | BPF_LSH | BPF_K:
316 case BPF_ALU | BPF_LSH | BPF_X:
317 case BPF_ALU | BPF_RSH | BPF_K:
318 case BPF_ALU | BPF_RSH | BPF_X:
319 case BPF_ALU | BPF_NEG:
320 case BPF_LD | BPF_IMM:
321 case BPF_LDX | BPF_IMM:
322 case BPF_MISC | BPF_TAX:
323 case BPF_MISC | BPF_TXA:
324 case BPF_LD | BPF_MEM:
325 case BPF_LDX | BPF_MEM:
326 case BPF_ST:
327 case BPF_STX:
328 case BPF_JMP | BPF_JA:
329 case BPF_JMP | BPF_JEQ | BPF_K:
330 case BPF_JMP | BPF_JEQ | BPF_X:
331 case BPF_JMP | BPF_JGE | BPF_K:
332 case BPF_JMP | BPF_JGE | BPF_X:
333 case BPF_JMP | BPF_JGT | BPF_K:
334 case BPF_JMP | BPF_JGT | BPF_X:
335 case BPF_JMP | BPF_JSET | BPF_K:
336 case BPF_JMP | BPF_JSET | BPF_X:
337 continue;
338 default:
339 return -EINVAL;
340 }
341 }
342 return 0;
343 }
344
345 #ifdef SECCOMP_ARCH_NATIVE
seccomp_cache_check_allow_bitmap(const void * bitmap,size_t bitmap_size,int syscall_nr)346 static inline bool seccomp_cache_check_allow_bitmap(const void *bitmap,
347 size_t bitmap_size,
348 int syscall_nr)
349 {
350 if (unlikely(syscall_nr < 0 || syscall_nr >= bitmap_size))
351 return false;
352 syscall_nr = array_index_nospec(syscall_nr, bitmap_size);
353
354 return test_bit(syscall_nr, bitmap);
355 }
356
357 /**
358 * seccomp_cache_check_allow - lookup seccomp cache
359 * @sfilter: The seccomp filter
360 * @sd: The seccomp data to lookup the cache with
361 *
362 * Returns true if the seccomp_data is cached and allowed.
363 */
seccomp_cache_check_allow(const struct seccomp_filter * sfilter,const struct seccomp_data * sd)364 static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilter,
365 const struct seccomp_data *sd)
366 {
367 int syscall_nr = sd->nr;
368 const struct action_cache *cache = &sfilter->cache;
369
370 #ifndef SECCOMP_ARCH_COMPAT
371 /* A native-only architecture doesn't need to check sd->arch. */
372 return seccomp_cache_check_allow_bitmap(cache->allow_native,
373 SECCOMP_ARCH_NATIVE_NR,
374 syscall_nr);
375 #else
376 if (likely(sd->arch == SECCOMP_ARCH_NATIVE))
377 return seccomp_cache_check_allow_bitmap(cache->allow_native,
378 SECCOMP_ARCH_NATIVE_NR,
379 syscall_nr);
380 if (likely(sd->arch == SECCOMP_ARCH_COMPAT))
381 return seccomp_cache_check_allow_bitmap(cache->allow_compat,
382 SECCOMP_ARCH_COMPAT_NR,
383 syscall_nr);
384 #endif /* SECCOMP_ARCH_COMPAT */
385
386 WARN_ON_ONCE(true);
387 return false;
388 }
389 #endif /* SECCOMP_ARCH_NATIVE */
390
391 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
392 /**
393 * seccomp_run_filters - evaluates all seccomp filters against @sd
394 * @sd: optional seccomp data to be passed to filters
395 * @match: stores struct seccomp_filter that resulted in the return value,
396 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
397 * be unchanged.
398 *
399 * Returns valid seccomp BPF response codes.
400 */
seccomp_run_filters(const struct seccomp_data * sd,struct seccomp_filter ** match)401 static u32 seccomp_run_filters(const struct seccomp_data *sd,
402 struct seccomp_filter **match)
403 {
404 u32 ret = SECCOMP_RET_ALLOW;
405 /* Make sure cross-thread synced filter points somewhere sane. */
406 struct seccomp_filter *f =
407 READ_ONCE(current->seccomp.filter);
408
409 /* Ensure unexpected behavior doesn't result in failing open. */
410 if (WARN_ON(f == NULL))
411 return SECCOMP_RET_KILL_PROCESS;
412
413 if (seccomp_cache_check_allow(f, sd))
414 return SECCOMP_RET_ALLOW;
415
416 /*
417 * All filters in the list are evaluated and the lowest BPF return
418 * value always takes priority (ignoring the DATA).
419 */
420 for (; f; f = f->prev) {
421 u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd);
422
423 if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
424 ret = cur_ret;
425 *match = f;
426 }
427 }
428 return ret;
429 }
430 #endif /* CONFIG_SECCOMP_FILTER */
431
seccomp_may_assign_mode(unsigned long seccomp_mode)432 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
433 {
434 assert_spin_locked(¤t->sighand->siglock);
435
436 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
437 return false;
438
439 return true;
440 }
441
arch_seccomp_spec_mitigate(struct task_struct * task)442 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
443
seccomp_assign_mode(struct task_struct * task,unsigned long seccomp_mode,unsigned long flags)444 static inline void seccomp_assign_mode(struct task_struct *task,
445 unsigned long seccomp_mode,
446 unsigned long flags)
447 {
448 assert_spin_locked(&task->sighand->siglock);
449
450 task->seccomp.mode = seccomp_mode;
451 /*
452 * Make sure SYSCALL_WORK_SECCOMP cannot be set before the mode (and
453 * filter) is set.
454 */
455 smp_mb__before_atomic();
456 /* Assume default seccomp processes want spec flaw mitigation. */
457 if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
458 arch_seccomp_spec_mitigate(task);
459 set_task_syscall_work(task, SECCOMP);
460 }
461
462 #ifdef CONFIG_SECCOMP_FILTER
463 /* Returns 1 if the parent is an ancestor of the child. */
is_ancestor(struct seccomp_filter * parent,struct seccomp_filter * child)464 static int is_ancestor(struct seccomp_filter *parent,
465 struct seccomp_filter *child)
466 {
467 /* NULL is the root ancestor. */
468 if (parent == NULL)
469 return 1;
470 for (; child; child = child->prev)
471 if (child == parent)
472 return 1;
473 return 0;
474 }
475
476 /**
477 * seccomp_can_sync_threads: checks if all threads can be synchronized
478 *
479 * Expects sighand and cred_guard_mutex locks to be held.
480 *
481 * Returns 0 on success, -ve on error, or the pid of a thread which was
482 * either not in the correct seccomp mode or did not have an ancestral
483 * seccomp filter.
484 */
seccomp_can_sync_threads(void)485 static inline pid_t seccomp_can_sync_threads(void)
486 {
487 struct task_struct *thread, *caller;
488
489 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
490 assert_spin_locked(¤t->sighand->siglock);
491
492 /* Validate all threads being eligible for synchronization. */
493 caller = current;
494 for_each_thread(caller, thread) {
495 pid_t failed;
496
497 /* Skip current, since it is initiating the sync. */
498 if (thread == caller)
499 continue;
500
501 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
502 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
503 is_ancestor(thread->seccomp.filter,
504 caller->seccomp.filter)))
505 continue;
506
507 /* Return the first thread that cannot be synchronized. */
508 failed = task_pid_vnr(thread);
509 /* If the pid cannot be resolved, then return -ESRCH */
510 if (WARN_ON(failed == 0))
511 failed = -ESRCH;
512 return failed;
513 }
514
515 return 0;
516 }
517
seccomp_filter_free(struct seccomp_filter * filter)518 static inline void seccomp_filter_free(struct seccomp_filter *filter)
519 {
520 if (filter) {
521 bpf_prog_destroy(filter->prog);
522 kfree(filter);
523 }
524 }
525
__seccomp_filter_orphan(struct seccomp_filter * orig)526 static void __seccomp_filter_orphan(struct seccomp_filter *orig)
527 {
528 while (orig && refcount_dec_and_test(&orig->users)) {
529 if (waitqueue_active(&orig->wqh))
530 wake_up_poll(&orig->wqh, EPOLLHUP);
531 orig = orig->prev;
532 }
533 }
534
__put_seccomp_filter(struct seccomp_filter * orig)535 static void __put_seccomp_filter(struct seccomp_filter *orig)
536 {
537 /* Clean up single-reference branches iteratively. */
538 while (orig && refcount_dec_and_test(&orig->refs)) {
539 struct seccomp_filter *freeme = orig;
540 orig = orig->prev;
541 seccomp_filter_free(freeme);
542 }
543 }
544
__seccomp_filter_release(struct seccomp_filter * orig)545 static void __seccomp_filter_release(struct seccomp_filter *orig)
546 {
547 /* Notify about any unused filters in the task's former filter tree. */
548 __seccomp_filter_orphan(orig);
549 /* Finally drop all references to the task's former tree. */
550 __put_seccomp_filter(orig);
551 }
552
553 /**
554 * seccomp_filter_release - Detach the task from its filter tree,
555 * drop its reference count, and notify
556 * about unused filters
557 *
558 * This function should only be called when the task is exiting as
559 * it detaches it from its filter tree. As such, READ_ONCE() and
560 * barriers are not needed here, as would normally be needed.
561 */
seccomp_filter_release(struct task_struct * tsk)562 void seccomp_filter_release(struct task_struct *tsk)
563 {
564 struct seccomp_filter *orig = tsk->seccomp.filter;
565
566 /* We are effectively holding the siglock by not having any sighand. */
567 WARN_ON(tsk->sighand != NULL);
568
569 /* Detach task from its filter tree. */
570 tsk->seccomp.filter = NULL;
571 __seccomp_filter_release(orig);
572 }
573
574 /**
575 * seccomp_sync_threads: sets all threads to use current's filter
576 *
577 * Expects sighand and cred_guard_mutex locks to be held, and for
578 * seccomp_can_sync_threads() to have returned success already
579 * without dropping the locks.
580 *
581 */
seccomp_sync_threads(unsigned long flags)582 static inline void seccomp_sync_threads(unsigned long flags)
583 {
584 struct task_struct *thread, *caller;
585
586 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
587 assert_spin_locked(¤t->sighand->siglock);
588
589 /* Synchronize all threads. */
590 caller = current;
591 for_each_thread(caller, thread) {
592 /* Skip current, since it needs no changes. */
593 if (thread == caller)
594 continue;
595
596 /* Get a task reference for the new leaf node. */
597 get_seccomp_filter(caller);
598
599 /*
600 * Drop the task reference to the shared ancestor since
601 * current's path will hold a reference. (This also
602 * allows a put before the assignment.)
603 */
604 __seccomp_filter_release(thread->seccomp.filter);
605
606 /* Make our new filter tree visible. */
607 smp_store_release(&thread->seccomp.filter,
608 caller->seccomp.filter);
609 atomic_set(&thread->seccomp.filter_count,
610 atomic_read(&caller->seccomp.filter_count));
611
612 /*
613 * Don't let an unprivileged task work around
614 * the no_new_privs restriction by creating
615 * a thread that sets it up, enters seccomp,
616 * then dies.
617 */
618 if (task_no_new_privs(caller))
619 task_set_no_new_privs(thread);
620
621 /*
622 * Opt the other thread into seccomp if needed.
623 * As threads are considered to be trust-realm
624 * equivalent (see ptrace_may_access), it is safe to
625 * allow one thread to transition the other.
626 */
627 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
628 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
629 flags);
630 }
631 }
632
633 /**
634 * seccomp_prepare_filter: Prepares a seccomp filter for use.
635 * @fprog: BPF program to install
636 *
637 * Returns filter on success or an ERR_PTR on failure.
638 */
seccomp_prepare_filter(struct sock_fprog * fprog)639 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
640 {
641 struct seccomp_filter *sfilter;
642 int ret;
643 const bool save_orig =
644 #if defined(CONFIG_CHECKPOINT_RESTORE) || defined(SECCOMP_ARCH_NATIVE)
645 true;
646 #else
647 false;
648 #endif
649
650 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
651 return ERR_PTR(-EINVAL);
652
653 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
654
655 /*
656 * Installing a seccomp filter requires that the task has
657 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
658 * This avoids scenarios where unprivileged tasks can affect the
659 * behavior of privileged children.
660 */
661 if (!task_no_new_privs(current) &&
662 !ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN))
663 return ERR_PTR(-EACCES);
664
665 /* Allocate a new seccomp_filter */
666 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
667 if (!sfilter)
668 return ERR_PTR(-ENOMEM);
669
670 mutex_init(&sfilter->notify_lock);
671 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
672 seccomp_check_filter, save_orig);
673 if (ret < 0) {
674 kfree(sfilter);
675 return ERR_PTR(ret);
676 }
677
678 refcount_set(&sfilter->refs, 1);
679 refcount_set(&sfilter->users, 1);
680 init_waitqueue_head(&sfilter->wqh);
681
682 return sfilter;
683 }
684
685 /**
686 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
687 * @user_filter: pointer to the user data containing a sock_fprog.
688 *
689 * Returns 0 on success and non-zero otherwise.
690 */
691 static struct seccomp_filter *
seccomp_prepare_user_filter(const char __user * user_filter)692 seccomp_prepare_user_filter(const char __user *user_filter)
693 {
694 struct sock_fprog fprog;
695 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
696
697 #ifdef CONFIG_COMPAT
698 if (in_compat_syscall()) {
699 struct compat_sock_fprog fprog32;
700 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
701 goto out;
702 fprog.len = fprog32.len;
703 fprog.filter = compat_ptr(fprog32.filter);
704 } else /* falls through to the if below. */
705 #endif
706 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
707 goto out;
708 filter = seccomp_prepare_filter(&fprog);
709 out:
710 return filter;
711 }
712
713 #ifdef SECCOMP_ARCH_NATIVE
714 /**
715 * seccomp_is_const_allow - check if filter is constant allow with given data
716 * @fprog: The BPF programs
717 * @sd: The seccomp data to check against, only syscall number and arch
718 * number are considered constant.
719 */
seccomp_is_const_allow(struct sock_fprog_kern * fprog,struct seccomp_data * sd)720 static bool seccomp_is_const_allow(struct sock_fprog_kern *fprog,
721 struct seccomp_data *sd)
722 {
723 unsigned int reg_value = 0;
724 unsigned int pc;
725 bool op_res;
726
727 if (WARN_ON_ONCE(!fprog))
728 return false;
729
730 for (pc = 0; pc < fprog->len; pc++) {
731 struct sock_filter *insn = &fprog->filter[pc];
732 u16 code = insn->code;
733 u32 k = insn->k;
734
735 switch (code) {
736 case BPF_LD | BPF_W | BPF_ABS:
737 switch (k) {
738 case offsetof(struct seccomp_data, nr):
739 reg_value = sd->nr;
740 break;
741 case offsetof(struct seccomp_data, arch):
742 reg_value = sd->arch;
743 break;
744 default:
745 /* can't optimize (non-constant value load) */
746 return false;
747 }
748 break;
749 case BPF_RET | BPF_K:
750 /* reached return with constant values only, check allow */
751 return k == SECCOMP_RET_ALLOW;
752 case BPF_JMP | BPF_JA:
753 pc += insn->k;
754 break;
755 case BPF_JMP | BPF_JEQ | BPF_K:
756 case BPF_JMP | BPF_JGE | BPF_K:
757 case BPF_JMP | BPF_JGT | BPF_K:
758 case BPF_JMP | BPF_JSET | BPF_K:
759 switch (BPF_OP(code)) {
760 case BPF_JEQ:
761 op_res = reg_value == k;
762 break;
763 case BPF_JGE:
764 op_res = reg_value >= k;
765 break;
766 case BPF_JGT:
767 op_res = reg_value > k;
768 break;
769 case BPF_JSET:
770 op_res = !!(reg_value & k);
771 break;
772 default:
773 /* can't optimize (unknown jump) */
774 return false;
775 }
776
777 pc += op_res ? insn->jt : insn->jf;
778 break;
779 case BPF_ALU | BPF_AND | BPF_K:
780 reg_value &= k;
781 break;
782 default:
783 /* can't optimize (unknown insn) */
784 return false;
785 }
786 }
787
788 /* ran off the end of the filter?! */
789 WARN_ON(1);
790 return false;
791 }
792
seccomp_cache_prepare_bitmap(struct seccomp_filter * sfilter,void * bitmap,const void * bitmap_prev,size_t bitmap_size,int arch)793 static void seccomp_cache_prepare_bitmap(struct seccomp_filter *sfilter,
794 void *bitmap, const void *bitmap_prev,
795 size_t bitmap_size, int arch)
796 {
797 struct sock_fprog_kern *fprog = sfilter->prog->orig_prog;
798 struct seccomp_data sd;
799 int nr;
800
801 if (bitmap_prev) {
802 /* The new filter must be as restrictive as the last. */
803 bitmap_copy(bitmap, bitmap_prev, bitmap_size);
804 } else {
805 /* Before any filters, all syscalls are always allowed. */
806 bitmap_fill(bitmap, bitmap_size);
807 }
808
809 for (nr = 0; nr < bitmap_size; nr++) {
810 /* No bitmap change: not a cacheable action. */
811 if (!test_bit(nr, bitmap))
812 continue;
813
814 sd.nr = nr;
815 sd.arch = arch;
816
817 /* No bitmap change: continue to always allow. */
818 if (seccomp_is_const_allow(fprog, &sd))
819 continue;
820
821 /*
822 * Not a cacheable action: always run filters.
823 * atomic clear_bit() not needed, filter not visible yet.
824 */
825 __clear_bit(nr, bitmap);
826 }
827 }
828
829 /**
830 * seccomp_cache_prepare - emulate the filter to find cacheable syscalls
831 * @sfilter: The seccomp filter
832 *
833 * Returns 0 if successful or -errno if error occurred.
834 */
seccomp_cache_prepare(struct seccomp_filter * sfilter)835 static void seccomp_cache_prepare(struct seccomp_filter *sfilter)
836 {
837 struct action_cache *cache = &sfilter->cache;
838 const struct action_cache *cache_prev =
839 sfilter->prev ? &sfilter->prev->cache : NULL;
840
841 seccomp_cache_prepare_bitmap(sfilter, cache->allow_native,
842 cache_prev ? cache_prev->allow_native : NULL,
843 SECCOMP_ARCH_NATIVE_NR,
844 SECCOMP_ARCH_NATIVE);
845
846 #ifdef SECCOMP_ARCH_COMPAT
847 seccomp_cache_prepare_bitmap(sfilter, cache->allow_compat,
848 cache_prev ? cache_prev->allow_compat : NULL,
849 SECCOMP_ARCH_COMPAT_NR,
850 SECCOMP_ARCH_COMPAT);
851 #endif /* SECCOMP_ARCH_COMPAT */
852 }
853 #endif /* SECCOMP_ARCH_NATIVE */
854
855 /**
856 * seccomp_attach_filter: validate and attach filter
857 * @flags: flags to change filter behavior
858 * @filter: seccomp filter to add to the current process
859 *
860 * Caller must be holding current->sighand->siglock lock.
861 *
862 * Returns 0 on success, -ve on error, or
863 * - in TSYNC mode: the pid of a thread which was either not in the correct
864 * seccomp mode or did not have an ancestral seccomp filter
865 * - in NEW_LISTENER mode: the fd of the new listener
866 */
seccomp_attach_filter(unsigned int flags,struct seccomp_filter * filter)867 static long seccomp_attach_filter(unsigned int flags,
868 struct seccomp_filter *filter)
869 {
870 unsigned long total_insns;
871 struct seccomp_filter *walker;
872
873 assert_spin_locked(¤t->sighand->siglock);
874
875 /* Validate resulting filter length. */
876 total_insns = filter->prog->len;
877 for (walker = current->seccomp.filter; walker; walker = walker->prev)
878 total_insns += walker->prog->len + 4; /* 4 instr penalty */
879 if (total_insns > MAX_INSNS_PER_PATH)
880 return -ENOMEM;
881
882 /* If thread sync has been requested, check that it is possible. */
883 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
884 int ret;
885
886 ret = seccomp_can_sync_threads();
887 if (ret) {
888 if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
889 return -ESRCH;
890 else
891 return ret;
892 }
893 }
894
895 /* Set log flag, if present. */
896 if (flags & SECCOMP_FILTER_FLAG_LOG)
897 filter->log = true;
898
899 /* Set wait killable flag, if present. */
900 if (flags & SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV)
901 filter->wait_killable_recv = true;
902
903 /*
904 * If there is an existing filter, make it the prev and don't drop its
905 * task reference.
906 */
907 filter->prev = current->seccomp.filter;
908 seccomp_cache_prepare(filter);
909 current->seccomp.filter = filter;
910 atomic_inc(¤t->seccomp.filter_count);
911
912 /* Now that the new filter is in place, synchronize to all threads. */
913 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
914 seccomp_sync_threads(flags);
915
916 return 0;
917 }
918
__get_seccomp_filter(struct seccomp_filter * filter)919 static void __get_seccomp_filter(struct seccomp_filter *filter)
920 {
921 refcount_inc(&filter->refs);
922 }
923
924 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
get_seccomp_filter(struct task_struct * tsk)925 void get_seccomp_filter(struct task_struct *tsk)
926 {
927 struct seccomp_filter *orig = tsk->seccomp.filter;
928 if (!orig)
929 return;
930 __get_seccomp_filter(orig);
931 refcount_inc(&orig->users);
932 }
933
934 #endif /* CONFIG_SECCOMP_FILTER */
935
936 /* For use with seccomp_actions_logged */
937 #define SECCOMP_LOG_KILL_PROCESS (1 << 0)
938 #define SECCOMP_LOG_KILL_THREAD (1 << 1)
939 #define SECCOMP_LOG_TRAP (1 << 2)
940 #define SECCOMP_LOG_ERRNO (1 << 3)
941 #define SECCOMP_LOG_TRACE (1 << 4)
942 #define SECCOMP_LOG_LOG (1 << 5)
943 #define SECCOMP_LOG_ALLOW (1 << 6)
944 #define SECCOMP_LOG_USER_NOTIF (1 << 7)
945
946 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
947 SECCOMP_LOG_KILL_THREAD |
948 SECCOMP_LOG_TRAP |
949 SECCOMP_LOG_ERRNO |
950 SECCOMP_LOG_USER_NOTIF |
951 SECCOMP_LOG_TRACE |
952 SECCOMP_LOG_LOG;
953
seccomp_log(unsigned long syscall,long signr,u32 action,bool requested)954 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
955 bool requested)
956 {
957 bool log = false;
958
959 switch (action) {
960 case SECCOMP_RET_ALLOW:
961 break;
962 case SECCOMP_RET_TRAP:
963 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
964 break;
965 case SECCOMP_RET_ERRNO:
966 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
967 break;
968 case SECCOMP_RET_TRACE:
969 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
970 break;
971 case SECCOMP_RET_USER_NOTIF:
972 log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
973 break;
974 case SECCOMP_RET_LOG:
975 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
976 break;
977 case SECCOMP_RET_KILL_THREAD:
978 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
979 break;
980 case SECCOMP_RET_KILL_PROCESS:
981 default:
982 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
983 }
984
985 /*
986 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
987 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
988 * any action from being logged by removing the action name from the
989 * seccomp_actions_logged sysctl.
990 */
991 if (!log)
992 return;
993
994 audit_seccomp(syscall, signr, action);
995 }
996
997 /*
998 * Secure computing mode 1 allows only read/write/exit/sigreturn.
999 * To be fully secure this must be combined with rlimit
1000 * to limit the stack allocations too.
1001 */
1002 static const int mode1_syscalls[] = {
1003 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
1004 -1, /* negative terminated */
1005 };
1006
__secure_computing_strict(int this_syscall)1007 static void __secure_computing_strict(int this_syscall)
1008 {
1009 const int *allowed_syscalls = mode1_syscalls;
1010 #ifdef CONFIG_COMPAT
1011 if (in_compat_syscall())
1012 allowed_syscalls = get_compat_mode1_syscalls();
1013 #endif
1014 do {
1015 if (*allowed_syscalls == this_syscall)
1016 return;
1017 } while (*++allowed_syscalls != -1);
1018
1019 #ifdef SECCOMP_DEBUG
1020 dump_stack();
1021 #endif
1022 current->seccomp.mode = SECCOMP_MODE_DEAD;
1023 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
1024 do_exit(SIGKILL);
1025 }
1026
1027 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
secure_computing_strict(int this_syscall)1028 void secure_computing_strict(int this_syscall)
1029 {
1030 int mode = current->seccomp.mode;
1031
1032 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1033 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1034 return;
1035
1036 if (mode == SECCOMP_MODE_DISABLED)
1037 return;
1038 else if (mode == SECCOMP_MODE_STRICT)
1039 __secure_computing_strict(this_syscall);
1040 else
1041 BUG();
1042 }
1043 #else
1044
1045 #ifdef CONFIG_SECCOMP_FILTER
seccomp_next_notify_id(struct seccomp_filter * filter)1046 static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
1047 {
1048 /*
1049 * Note: overflow is ok here, the id just needs to be unique per
1050 * filter.
1051 */
1052 lockdep_assert_held(&filter->notify_lock);
1053 return filter->notif->next_id++;
1054 }
1055
seccomp_handle_addfd(struct seccomp_kaddfd * addfd,struct seccomp_knotif * n)1056 static void seccomp_handle_addfd(struct seccomp_kaddfd *addfd, struct seccomp_knotif *n)
1057 {
1058 int fd;
1059
1060 /*
1061 * Remove the notification, and reset the list pointers, indicating
1062 * that it has been handled.
1063 */
1064 list_del_init(&addfd->list);
1065 if (!addfd->setfd)
1066 fd = receive_fd(addfd->file, addfd->flags);
1067 else
1068 fd = receive_fd_replace(addfd->fd, addfd->file, addfd->flags);
1069 addfd->ret = fd;
1070
1071 if (addfd->ioctl_flags & SECCOMP_ADDFD_FLAG_SEND) {
1072 /* If we fail reset and return an error to the notifier */
1073 if (fd < 0) {
1074 n->state = SECCOMP_NOTIFY_SENT;
1075 } else {
1076 /* Return the FD we just added */
1077 n->flags = 0;
1078 n->error = 0;
1079 n->val = fd;
1080 }
1081 }
1082
1083 /*
1084 * Mark the notification as completed. From this point, addfd mem
1085 * might be invalidated and we can't safely read it anymore.
1086 */
1087 complete(&addfd->completion);
1088 }
1089
should_sleep_killable(struct seccomp_filter * match,struct seccomp_knotif * n)1090 static bool should_sleep_killable(struct seccomp_filter *match,
1091 struct seccomp_knotif *n)
1092 {
1093 return match->wait_killable_recv && n->state == SECCOMP_NOTIFY_SENT;
1094 }
1095
seccomp_do_user_notification(int this_syscall,struct seccomp_filter * match,const struct seccomp_data * sd)1096 static int seccomp_do_user_notification(int this_syscall,
1097 struct seccomp_filter *match,
1098 const struct seccomp_data *sd)
1099 {
1100 int err;
1101 u32 flags = 0;
1102 long ret = 0;
1103 struct seccomp_knotif n = {};
1104 struct seccomp_kaddfd *addfd, *tmp;
1105
1106 mutex_lock(&match->notify_lock);
1107 err = -ENOSYS;
1108 if (!match->notif)
1109 goto out;
1110
1111 n.task = current;
1112 n.state = SECCOMP_NOTIFY_INIT;
1113 n.data = sd;
1114 n.id = seccomp_next_notify_id(match);
1115 init_completion(&n.ready);
1116 list_add_tail(&n.list, &match->notif->notifications);
1117 INIT_LIST_HEAD(&n.addfd);
1118
1119 up(&match->notif->request);
1120 wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM);
1121
1122 /*
1123 * This is where we wait for a reply from userspace.
1124 */
1125 do {
1126 bool wait_killable = should_sleep_killable(match, &n);
1127
1128 mutex_unlock(&match->notify_lock);
1129 if (wait_killable)
1130 err = wait_for_completion_killable(&n.ready);
1131 else
1132 err = wait_for_completion_interruptible(&n.ready);
1133 mutex_lock(&match->notify_lock);
1134
1135 if (err != 0) {
1136 /*
1137 * Check to see if the notifcation got picked up and
1138 * whether we should switch to wait killable.
1139 */
1140 if (!wait_killable && should_sleep_killable(match, &n))
1141 continue;
1142
1143 goto interrupted;
1144 }
1145
1146 addfd = list_first_entry_or_null(&n.addfd,
1147 struct seccomp_kaddfd, list);
1148 /* Check if we were woken up by a addfd message */
1149 if (addfd)
1150 seccomp_handle_addfd(addfd, &n);
1151
1152 } while (n.state != SECCOMP_NOTIFY_REPLIED);
1153
1154 ret = n.val;
1155 err = n.error;
1156 flags = n.flags;
1157
1158 interrupted:
1159 /* If there were any pending addfd calls, clear them out */
1160 list_for_each_entry_safe(addfd, tmp, &n.addfd, list) {
1161 /* The process went away before we got a chance to handle it */
1162 addfd->ret = -ESRCH;
1163 list_del_init(&addfd->list);
1164 complete(&addfd->completion);
1165 }
1166
1167 /*
1168 * Note that it's possible the listener died in between the time when
1169 * we were notified of a response (or a signal) and when we were able to
1170 * re-acquire the lock, so only delete from the list if the
1171 * notification actually exists.
1172 *
1173 * Also note that this test is only valid because there's no way to
1174 * *reattach* to a notifier right now. If one is added, we'll need to
1175 * keep track of the notif itself and make sure they match here.
1176 */
1177 if (match->notif)
1178 list_del(&n.list);
1179 out:
1180 mutex_unlock(&match->notify_lock);
1181
1182 /* Userspace requests to continue the syscall. */
1183 if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1184 return 0;
1185
1186 syscall_set_return_value(current, current_pt_regs(),
1187 err, ret);
1188 return -1;
1189 }
1190
__seccomp_filter(int this_syscall,const struct seccomp_data * sd,const bool recheck_after_trace)1191 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1192 const bool recheck_after_trace)
1193 {
1194 u32 filter_ret, action;
1195 struct seccomp_filter *match = NULL;
1196 int data;
1197 struct seccomp_data sd_local;
1198
1199 /*
1200 * Make sure that any changes to mode from another thread have
1201 * been seen after SYSCALL_WORK_SECCOMP was seen.
1202 */
1203 smp_rmb();
1204
1205 if (!sd) {
1206 populate_seccomp_data(&sd_local);
1207 sd = &sd_local;
1208 }
1209
1210 filter_ret = seccomp_run_filters(sd, &match);
1211 data = filter_ret & SECCOMP_RET_DATA;
1212 action = filter_ret & SECCOMP_RET_ACTION_FULL;
1213
1214 switch (action) {
1215 case SECCOMP_RET_ERRNO:
1216 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
1217 if (data > MAX_ERRNO)
1218 data = MAX_ERRNO;
1219 syscall_set_return_value(current, current_pt_regs(),
1220 -data, 0);
1221 goto skip;
1222
1223 case SECCOMP_RET_TRAP:
1224 /* Show the handler the original registers. */
1225 syscall_rollback(current, current_pt_regs());
1226 /* Let the filter pass back 16 bits of data. */
1227 force_sig_seccomp(this_syscall, data, false);
1228 goto skip;
1229
1230 case SECCOMP_RET_TRACE:
1231 /* We've been put in this state by the ptracer already. */
1232 if (recheck_after_trace)
1233 return 0;
1234
1235 /* ENOSYS these calls if there is no tracer attached. */
1236 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
1237 syscall_set_return_value(current,
1238 current_pt_regs(),
1239 -ENOSYS, 0);
1240 goto skip;
1241 }
1242
1243 /* Allow the BPF to provide the event message */
1244 ptrace_event(PTRACE_EVENT_SECCOMP, data);
1245 /*
1246 * The delivery of a fatal signal during event
1247 * notification may silently skip tracer notification,
1248 * which could leave us with a potentially unmodified
1249 * syscall that the tracer would have liked to have
1250 * changed. Since the process is about to die, we just
1251 * force the syscall to be skipped and let the signal
1252 * kill the process and correctly handle any tracer exit
1253 * notifications.
1254 */
1255 if (fatal_signal_pending(current))
1256 goto skip;
1257 /* Check if the tracer forced the syscall to be skipped. */
1258 this_syscall = syscall_get_nr(current, current_pt_regs());
1259 if (this_syscall < 0)
1260 goto skip;
1261
1262 /*
1263 * Recheck the syscall, since it may have changed. This
1264 * intentionally uses a NULL struct seccomp_data to force
1265 * a reload of all registers. This does not goto skip since
1266 * a skip would have already been reported.
1267 */
1268 if (__seccomp_filter(this_syscall, NULL, true))
1269 return -1;
1270
1271 return 0;
1272
1273 case SECCOMP_RET_USER_NOTIF:
1274 if (seccomp_do_user_notification(this_syscall, match, sd))
1275 goto skip;
1276
1277 return 0;
1278
1279 case SECCOMP_RET_LOG:
1280 seccomp_log(this_syscall, 0, action, true);
1281 return 0;
1282
1283 case SECCOMP_RET_ALLOW:
1284 /*
1285 * Note that the "match" filter will always be NULL for
1286 * this action since SECCOMP_RET_ALLOW is the starting
1287 * state in seccomp_run_filters().
1288 */
1289 return 0;
1290
1291 case SECCOMP_RET_KILL_THREAD:
1292 case SECCOMP_RET_KILL_PROCESS:
1293 default:
1294 current->seccomp.mode = SECCOMP_MODE_DEAD;
1295 seccomp_log(this_syscall, SIGSYS, action, true);
1296 /* Dump core only if this is the last remaining thread. */
1297 if (action != SECCOMP_RET_KILL_THREAD ||
1298 (atomic_read(¤t->signal->live) == 1)) {
1299 /* Show the original registers in the dump. */
1300 syscall_rollback(current, current_pt_regs());
1301 /* Trigger a coredump with SIGSYS */
1302 force_sig_seccomp(this_syscall, data, true);
1303 } else {
1304 do_exit(SIGSYS);
1305 }
1306 return -1; /* skip the syscall go directly to signal handling */
1307 }
1308
1309 unreachable();
1310
1311 skip:
1312 seccomp_log(this_syscall, 0, action, match ? match->log : false);
1313 return -1;
1314 }
1315 #else
__seccomp_filter(int this_syscall,const struct seccomp_data * sd,const bool recheck_after_trace)1316 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1317 const bool recheck_after_trace)
1318 {
1319 BUG();
1320
1321 return -1;
1322 }
1323 #endif
1324
__secure_computing(const struct seccomp_data * sd)1325 int __secure_computing(const struct seccomp_data *sd)
1326 {
1327 int mode = current->seccomp.mode;
1328 int this_syscall;
1329
1330 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1331 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1332 return 0;
1333
1334 this_syscall = sd ? sd->nr :
1335 syscall_get_nr(current, current_pt_regs());
1336
1337 switch (mode) {
1338 case SECCOMP_MODE_STRICT:
1339 __secure_computing_strict(this_syscall); /* may call do_exit */
1340 return 0;
1341 case SECCOMP_MODE_FILTER:
1342 return __seccomp_filter(this_syscall, sd, false);
1343 /* Surviving SECCOMP_RET_KILL_* must be proactively impossible. */
1344 case SECCOMP_MODE_DEAD:
1345 WARN_ON_ONCE(1);
1346 do_exit(SIGKILL);
1347 return -1;
1348 default:
1349 BUG();
1350 }
1351 }
1352 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1353
prctl_get_seccomp(void)1354 long prctl_get_seccomp(void)
1355 {
1356 return current->seccomp.mode;
1357 }
1358
1359 /**
1360 * seccomp_set_mode_strict: internal function for setting strict seccomp
1361 *
1362 * Once current->seccomp.mode is non-zero, it may not be changed.
1363 *
1364 * Returns 0 on success or -EINVAL on failure.
1365 */
seccomp_set_mode_strict(void)1366 static long seccomp_set_mode_strict(void)
1367 {
1368 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
1369 long ret = -EINVAL;
1370
1371 spin_lock_irq(¤t->sighand->siglock);
1372
1373 if (!seccomp_may_assign_mode(seccomp_mode))
1374 goto out;
1375
1376 #ifdef TIF_NOTSC
1377 disable_TSC();
1378 #endif
1379 seccomp_assign_mode(current, seccomp_mode, 0);
1380 ret = 0;
1381
1382 out:
1383 spin_unlock_irq(¤t->sighand->siglock);
1384
1385 return ret;
1386 }
1387
1388 #ifdef CONFIG_SECCOMP_FILTER
seccomp_notify_free(struct seccomp_filter * filter)1389 static void seccomp_notify_free(struct seccomp_filter *filter)
1390 {
1391 kfree(filter->notif);
1392 filter->notif = NULL;
1393 }
1394
seccomp_notify_detach(struct seccomp_filter * filter)1395 static void seccomp_notify_detach(struct seccomp_filter *filter)
1396 {
1397 struct seccomp_knotif *knotif;
1398
1399 if (!filter)
1400 return;
1401
1402 mutex_lock(&filter->notify_lock);
1403
1404 /*
1405 * If this file is being closed because e.g. the task who owned it
1406 * died, let's wake everyone up who was waiting on us.
1407 */
1408 list_for_each_entry(knotif, &filter->notif->notifications, list) {
1409 if (knotif->state == SECCOMP_NOTIFY_REPLIED)
1410 continue;
1411
1412 knotif->state = SECCOMP_NOTIFY_REPLIED;
1413 knotif->error = -ENOSYS;
1414 knotif->val = 0;
1415
1416 /*
1417 * We do not need to wake up any pending addfd messages, as
1418 * the notifier will do that for us, as this just looks
1419 * like a standard reply.
1420 */
1421 complete(&knotif->ready);
1422 }
1423
1424 seccomp_notify_free(filter);
1425 mutex_unlock(&filter->notify_lock);
1426 }
1427
seccomp_notify_release(struct inode * inode,struct file * file)1428 static int seccomp_notify_release(struct inode *inode, struct file *file)
1429 {
1430 struct seccomp_filter *filter = file->private_data;
1431
1432 seccomp_notify_detach(filter);
1433 __put_seccomp_filter(filter);
1434 return 0;
1435 }
1436
1437 /* must be called with notif_lock held */
1438 static inline struct seccomp_knotif *
find_notification(struct seccomp_filter * filter,u64 id)1439 find_notification(struct seccomp_filter *filter, u64 id)
1440 {
1441 struct seccomp_knotif *cur;
1442
1443 lockdep_assert_held(&filter->notify_lock);
1444
1445 list_for_each_entry(cur, &filter->notif->notifications, list) {
1446 if (cur->id == id)
1447 return cur;
1448 }
1449
1450 return NULL;
1451 }
1452
1453
seccomp_notify_recv(struct seccomp_filter * filter,void __user * buf)1454 static long seccomp_notify_recv(struct seccomp_filter *filter,
1455 void __user *buf)
1456 {
1457 struct seccomp_knotif *knotif = NULL, *cur;
1458 struct seccomp_notif unotif;
1459 ssize_t ret;
1460
1461 /* Verify that we're not given garbage to keep struct extensible. */
1462 ret = check_zeroed_user(buf, sizeof(unotif));
1463 if (ret < 0)
1464 return ret;
1465 if (!ret)
1466 return -EINVAL;
1467
1468 memset(&unotif, 0, sizeof(unotif));
1469
1470 ret = down_interruptible(&filter->notif->request);
1471 if (ret < 0)
1472 return ret;
1473
1474 mutex_lock(&filter->notify_lock);
1475 list_for_each_entry(cur, &filter->notif->notifications, list) {
1476 if (cur->state == SECCOMP_NOTIFY_INIT) {
1477 knotif = cur;
1478 break;
1479 }
1480 }
1481
1482 /*
1483 * If we didn't find a notification, it could be that the task was
1484 * interrupted by a fatal signal between the time we were woken and
1485 * when we were able to acquire the rw lock.
1486 */
1487 if (!knotif) {
1488 ret = -ENOENT;
1489 goto out;
1490 }
1491
1492 unotif.id = knotif->id;
1493 unotif.pid = task_pid_vnr(knotif->task);
1494 unotif.data = *(knotif->data);
1495
1496 knotif->state = SECCOMP_NOTIFY_SENT;
1497 wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM);
1498 ret = 0;
1499 out:
1500 mutex_unlock(&filter->notify_lock);
1501
1502 if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1503 ret = -EFAULT;
1504
1505 /*
1506 * Userspace screwed up. To make sure that we keep this
1507 * notification alive, let's reset it back to INIT. It
1508 * may have died when we released the lock, so we need to make
1509 * sure it's still around.
1510 */
1511 mutex_lock(&filter->notify_lock);
1512 knotif = find_notification(filter, unotif.id);
1513 if (knotif) {
1514 /* Reset the process to make sure it's not stuck */
1515 if (should_sleep_killable(filter, knotif))
1516 complete(&knotif->ready);
1517 knotif->state = SECCOMP_NOTIFY_INIT;
1518 up(&filter->notif->request);
1519 }
1520 mutex_unlock(&filter->notify_lock);
1521 }
1522
1523 return ret;
1524 }
1525
seccomp_notify_send(struct seccomp_filter * filter,void __user * buf)1526 static long seccomp_notify_send(struct seccomp_filter *filter,
1527 void __user *buf)
1528 {
1529 struct seccomp_notif_resp resp = {};
1530 struct seccomp_knotif *knotif;
1531 long ret;
1532
1533 if (copy_from_user(&resp, buf, sizeof(resp)))
1534 return -EFAULT;
1535
1536 if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1537 return -EINVAL;
1538
1539 if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) &&
1540 (resp.error || resp.val))
1541 return -EINVAL;
1542
1543 ret = mutex_lock_interruptible(&filter->notify_lock);
1544 if (ret < 0)
1545 return ret;
1546
1547 knotif = find_notification(filter, resp.id);
1548 if (!knotif) {
1549 ret = -ENOENT;
1550 goto out;
1551 }
1552
1553 /* Allow exactly one reply. */
1554 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1555 ret = -EINPROGRESS;
1556 goto out;
1557 }
1558
1559 ret = 0;
1560 knotif->state = SECCOMP_NOTIFY_REPLIED;
1561 knotif->error = resp.error;
1562 knotif->val = resp.val;
1563 knotif->flags = resp.flags;
1564 complete(&knotif->ready);
1565 out:
1566 mutex_unlock(&filter->notify_lock);
1567 return ret;
1568 }
1569
seccomp_notify_id_valid(struct seccomp_filter * filter,void __user * buf)1570 static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1571 void __user *buf)
1572 {
1573 struct seccomp_knotif *knotif;
1574 u64 id;
1575 long ret;
1576
1577 if (copy_from_user(&id, buf, sizeof(id)))
1578 return -EFAULT;
1579
1580 ret = mutex_lock_interruptible(&filter->notify_lock);
1581 if (ret < 0)
1582 return ret;
1583
1584 knotif = find_notification(filter, id);
1585 if (knotif && knotif->state == SECCOMP_NOTIFY_SENT)
1586 ret = 0;
1587 else
1588 ret = -ENOENT;
1589
1590 mutex_unlock(&filter->notify_lock);
1591 return ret;
1592 }
1593
seccomp_notify_addfd(struct seccomp_filter * filter,struct seccomp_notif_addfd __user * uaddfd,unsigned int size)1594 static long seccomp_notify_addfd(struct seccomp_filter *filter,
1595 struct seccomp_notif_addfd __user *uaddfd,
1596 unsigned int size)
1597 {
1598 struct seccomp_notif_addfd addfd;
1599 struct seccomp_knotif *knotif;
1600 struct seccomp_kaddfd kaddfd;
1601 int ret;
1602
1603 BUILD_BUG_ON(sizeof(addfd) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0);
1604 BUILD_BUG_ON(sizeof(addfd) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST);
1605
1606 if (size < SECCOMP_NOTIFY_ADDFD_SIZE_VER0 || size >= PAGE_SIZE)
1607 return -EINVAL;
1608
1609 ret = copy_struct_from_user(&addfd, sizeof(addfd), uaddfd, size);
1610 if (ret)
1611 return ret;
1612
1613 if (addfd.newfd_flags & ~O_CLOEXEC)
1614 return -EINVAL;
1615
1616 if (addfd.flags & ~(SECCOMP_ADDFD_FLAG_SETFD | SECCOMP_ADDFD_FLAG_SEND))
1617 return -EINVAL;
1618
1619 if (addfd.newfd && !(addfd.flags & SECCOMP_ADDFD_FLAG_SETFD))
1620 return -EINVAL;
1621
1622 kaddfd.file = fget(addfd.srcfd);
1623 if (!kaddfd.file)
1624 return -EBADF;
1625
1626 kaddfd.ioctl_flags = addfd.flags;
1627 kaddfd.flags = addfd.newfd_flags;
1628 kaddfd.setfd = addfd.flags & SECCOMP_ADDFD_FLAG_SETFD;
1629 kaddfd.fd = addfd.newfd;
1630 init_completion(&kaddfd.completion);
1631
1632 ret = mutex_lock_interruptible(&filter->notify_lock);
1633 if (ret < 0)
1634 goto out;
1635
1636 knotif = find_notification(filter, addfd.id);
1637 if (!knotif) {
1638 ret = -ENOENT;
1639 goto out_unlock;
1640 }
1641
1642 /*
1643 * We do not want to allow for FD injection to occur before the
1644 * notification has been picked up by a userspace handler, or after
1645 * the notification has been replied to.
1646 */
1647 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1648 ret = -EINPROGRESS;
1649 goto out_unlock;
1650 }
1651
1652 if (addfd.flags & SECCOMP_ADDFD_FLAG_SEND) {
1653 /*
1654 * Disallow queuing an atomic addfd + send reply while there are
1655 * some addfd requests still to process.
1656 *
1657 * There is no clear reason to support it and allows us to keep
1658 * the loop on the other side straight-forward.
1659 */
1660 if (!list_empty(&knotif->addfd)) {
1661 ret = -EBUSY;
1662 goto out_unlock;
1663 }
1664
1665 /* Allow exactly only one reply */
1666 knotif->state = SECCOMP_NOTIFY_REPLIED;
1667 }
1668
1669 list_add(&kaddfd.list, &knotif->addfd);
1670 complete(&knotif->ready);
1671 mutex_unlock(&filter->notify_lock);
1672
1673 /* Now we wait for it to be processed or be interrupted */
1674 ret = wait_for_completion_interruptible(&kaddfd.completion);
1675 if (ret == 0) {
1676 /*
1677 * We had a successful completion. The other side has already
1678 * removed us from the addfd queue, and
1679 * wait_for_completion_interruptible has a memory barrier upon
1680 * success that lets us read this value directly without
1681 * locking.
1682 */
1683 ret = kaddfd.ret;
1684 goto out;
1685 }
1686
1687 mutex_lock(&filter->notify_lock);
1688 /*
1689 * Even though we were woken up by a signal and not a successful
1690 * completion, a completion may have happened in the mean time.
1691 *
1692 * We need to check again if the addfd request has been handled,
1693 * and if not, we will remove it from the queue.
1694 */
1695 if (list_empty(&kaddfd.list))
1696 ret = kaddfd.ret;
1697 else
1698 list_del(&kaddfd.list);
1699
1700 out_unlock:
1701 mutex_unlock(&filter->notify_lock);
1702 out:
1703 fput(kaddfd.file);
1704
1705 return ret;
1706 }
1707
seccomp_notify_ioctl(struct file * file,unsigned int cmd,unsigned long arg)1708 static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1709 unsigned long arg)
1710 {
1711 struct seccomp_filter *filter = file->private_data;
1712 void __user *buf = (void __user *)arg;
1713
1714 /* Fixed-size ioctls */
1715 switch (cmd) {
1716 case SECCOMP_IOCTL_NOTIF_RECV:
1717 return seccomp_notify_recv(filter, buf);
1718 case SECCOMP_IOCTL_NOTIF_SEND:
1719 return seccomp_notify_send(filter, buf);
1720 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR:
1721 case SECCOMP_IOCTL_NOTIF_ID_VALID:
1722 return seccomp_notify_id_valid(filter, buf);
1723 }
1724
1725 /* Extensible Argument ioctls */
1726 #define EA_IOCTL(cmd) ((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1727 switch (EA_IOCTL(cmd)) {
1728 case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD):
1729 return seccomp_notify_addfd(filter, buf, _IOC_SIZE(cmd));
1730 default:
1731 return -EINVAL;
1732 }
1733 }
1734
seccomp_notify_poll(struct file * file,struct poll_table_struct * poll_tab)1735 static __poll_t seccomp_notify_poll(struct file *file,
1736 struct poll_table_struct *poll_tab)
1737 {
1738 struct seccomp_filter *filter = file->private_data;
1739 __poll_t ret = 0;
1740 struct seccomp_knotif *cur;
1741
1742 poll_wait(file, &filter->wqh, poll_tab);
1743
1744 if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1745 return EPOLLERR;
1746
1747 list_for_each_entry(cur, &filter->notif->notifications, list) {
1748 if (cur->state == SECCOMP_NOTIFY_INIT)
1749 ret |= EPOLLIN | EPOLLRDNORM;
1750 if (cur->state == SECCOMP_NOTIFY_SENT)
1751 ret |= EPOLLOUT | EPOLLWRNORM;
1752 if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1753 break;
1754 }
1755
1756 mutex_unlock(&filter->notify_lock);
1757
1758 if (refcount_read(&filter->users) == 0)
1759 ret |= EPOLLHUP;
1760
1761 return ret;
1762 }
1763
1764 static const struct file_operations seccomp_notify_ops = {
1765 .poll = seccomp_notify_poll,
1766 .release = seccomp_notify_release,
1767 .unlocked_ioctl = seccomp_notify_ioctl,
1768 .compat_ioctl = seccomp_notify_ioctl,
1769 };
1770
init_listener(struct seccomp_filter * filter)1771 static struct file *init_listener(struct seccomp_filter *filter)
1772 {
1773 struct file *ret;
1774
1775 ret = ERR_PTR(-ENOMEM);
1776 filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1777 if (!filter->notif)
1778 goto out;
1779
1780 sema_init(&filter->notif->request, 0);
1781 filter->notif->next_id = get_random_u64();
1782 INIT_LIST_HEAD(&filter->notif->notifications);
1783
1784 ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1785 filter, O_RDWR);
1786 if (IS_ERR(ret))
1787 goto out_notif;
1788
1789 /* The file has a reference to it now */
1790 __get_seccomp_filter(filter);
1791
1792 out_notif:
1793 if (IS_ERR(ret))
1794 seccomp_notify_free(filter);
1795 out:
1796 return ret;
1797 }
1798
1799 /*
1800 * Does @new_child have a listener while an ancestor also has a listener?
1801 * If so, we'll want to reject this filter.
1802 * This only has to be tested for the current process, even in the TSYNC case,
1803 * because TSYNC installs @child with the same parent on all threads.
1804 * Note that @new_child is not hooked up to its parent at this point yet, so
1805 * we use current->seccomp.filter.
1806 */
has_duplicate_listener(struct seccomp_filter * new_child)1807 static bool has_duplicate_listener(struct seccomp_filter *new_child)
1808 {
1809 struct seccomp_filter *cur;
1810
1811 /* must be protected against concurrent TSYNC */
1812 lockdep_assert_held(¤t->sighand->siglock);
1813
1814 if (!new_child->notif)
1815 return false;
1816 for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1817 if (cur->notif)
1818 return true;
1819 }
1820
1821 return false;
1822 }
1823
1824 /**
1825 * seccomp_set_mode_filter: internal function for setting seccomp filter
1826 * @flags: flags to change filter behavior
1827 * @filter: struct sock_fprog containing filter
1828 *
1829 * This function may be called repeatedly to install additional filters.
1830 * Every filter successfully installed will be evaluated (in reverse order)
1831 * for each system call the task makes.
1832 *
1833 * Once current->seccomp.mode is non-zero, it may not be changed.
1834 *
1835 * Returns 0 on success or -EINVAL on failure.
1836 */
seccomp_set_mode_filter(unsigned int flags,const char __user * filter)1837 static long seccomp_set_mode_filter(unsigned int flags,
1838 const char __user *filter)
1839 {
1840 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1841 struct seccomp_filter *prepared = NULL;
1842 long ret = -EINVAL;
1843 int listener = -1;
1844 struct file *listener_f = NULL;
1845
1846 /* Validate flags. */
1847 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1848 return -EINVAL;
1849
1850 /*
1851 * In the successful case, NEW_LISTENER returns the new listener fd.
1852 * But in the failure case, TSYNC returns the thread that died. If you
1853 * combine these two flags, there's no way to tell whether something
1854 * succeeded or failed. So, let's disallow this combination if the user
1855 * has not explicitly requested no errors from TSYNC.
1856 */
1857 if ((flags & SECCOMP_FILTER_FLAG_TSYNC) &&
1858 (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) &&
1859 ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0))
1860 return -EINVAL;
1861
1862 /*
1863 * The SECCOMP_FILTER_FLAG_WAIT_KILLABLE_SENT flag doesn't make sense
1864 * without the SECCOMP_FILTER_FLAG_NEW_LISTENER flag.
1865 */
1866 if ((flags & SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV) &&
1867 ((flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) == 0))
1868 return -EINVAL;
1869
1870 /* Prepare the new filter before holding any locks. */
1871 prepared = seccomp_prepare_user_filter(filter);
1872 if (IS_ERR(prepared))
1873 return PTR_ERR(prepared);
1874
1875 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1876 listener = get_unused_fd_flags(O_CLOEXEC);
1877 if (listener < 0) {
1878 ret = listener;
1879 goto out_free;
1880 }
1881
1882 listener_f = init_listener(prepared);
1883 if (IS_ERR(listener_f)) {
1884 put_unused_fd(listener);
1885 ret = PTR_ERR(listener_f);
1886 goto out_free;
1887 }
1888 }
1889
1890 /*
1891 * Make sure we cannot change seccomp or nnp state via TSYNC
1892 * while another thread is in the middle of calling exec.
1893 */
1894 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1895 mutex_lock_killable(¤t->signal->cred_guard_mutex))
1896 goto out_put_fd;
1897
1898 spin_lock_irq(¤t->sighand->siglock);
1899
1900 if (!seccomp_may_assign_mode(seccomp_mode))
1901 goto out;
1902
1903 if (has_duplicate_listener(prepared)) {
1904 ret = -EBUSY;
1905 goto out;
1906 }
1907
1908 ret = seccomp_attach_filter(flags, prepared);
1909 if (ret)
1910 goto out;
1911 /* Do not free the successfully attached filter. */
1912 prepared = NULL;
1913
1914 seccomp_assign_mode(current, seccomp_mode, flags);
1915 out:
1916 spin_unlock_irq(¤t->sighand->siglock);
1917 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
1918 mutex_unlock(¤t->signal->cred_guard_mutex);
1919 out_put_fd:
1920 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1921 if (ret) {
1922 listener_f->private_data = NULL;
1923 fput(listener_f);
1924 put_unused_fd(listener);
1925 seccomp_notify_detach(prepared);
1926 } else {
1927 fd_install(listener, listener_f);
1928 ret = listener;
1929 }
1930 }
1931 out_free:
1932 seccomp_filter_free(prepared);
1933 return ret;
1934 }
1935 #else
seccomp_set_mode_filter(unsigned int flags,const char __user * filter)1936 static inline long seccomp_set_mode_filter(unsigned int flags,
1937 const char __user *filter)
1938 {
1939 return -EINVAL;
1940 }
1941 #endif
1942
seccomp_get_action_avail(const char __user * uaction)1943 static long seccomp_get_action_avail(const char __user *uaction)
1944 {
1945 u32 action;
1946
1947 if (copy_from_user(&action, uaction, sizeof(action)))
1948 return -EFAULT;
1949
1950 switch (action) {
1951 case SECCOMP_RET_KILL_PROCESS:
1952 case SECCOMP_RET_KILL_THREAD:
1953 case SECCOMP_RET_TRAP:
1954 case SECCOMP_RET_ERRNO:
1955 case SECCOMP_RET_USER_NOTIF:
1956 case SECCOMP_RET_TRACE:
1957 case SECCOMP_RET_LOG:
1958 case SECCOMP_RET_ALLOW:
1959 break;
1960 default:
1961 return -EOPNOTSUPP;
1962 }
1963
1964 return 0;
1965 }
1966
seccomp_get_notif_sizes(void __user * usizes)1967 static long seccomp_get_notif_sizes(void __user *usizes)
1968 {
1969 struct seccomp_notif_sizes sizes = {
1970 .seccomp_notif = sizeof(struct seccomp_notif),
1971 .seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
1972 .seccomp_data = sizeof(struct seccomp_data),
1973 };
1974
1975 if (copy_to_user(usizes, &sizes, sizeof(sizes)))
1976 return -EFAULT;
1977
1978 return 0;
1979 }
1980
1981 /* Common entry point for both prctl and syscall. */
do_seccomp(unsigned int op,unsigned int flags,void __user * uargs)1982 static long do_seccomp(unsigned int op, unsigned int flags,
1983 void __user *uargs)
1984 {
1985 switch (op) {
1986 case SECCOMP_SET_MODE_STRICT:
1987 if (flags != 0 || uargs != NULL)
1988 return -EINVAL;
1989 return seccomp_set_mode_strict();
1990 case SECCOMP_SET_MODE_FILTER:
1991 return seccomp_set_mode_filter(flags, uargs);
1992 case SECCOMP_GET_ACTION_AVAIL:
1993 if (flags != 0)
1994 return -EINVAL;
1995
1996 return seccomp_get_action_avail(uargs);
1997 case SECCOMP_GET_NOTIF_SIZES:
1998 if (flags != 0)
1999 return -EINVAL;
2000
2001 return seccomp_get_notif_sizes(uargs);
2002 default:
2003 return -EINVAL;
2004 }
2005 }
2006
SYSCALL_DEFINE3(seccomp,unsigned int,op,unsigned int,flags,void __user *,uargs)2007 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
2008 void __user *, uargs)
2009 {
2010 return do_seccomp(op, flags, uargs);
2011 }
2012
2013 /**
2014 * prctl_set_seccomp: configures current->seccomp.mode
2015 * @seccomp_mode: requested mode to use
2016 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
2017 *
2018 * Returns 0 on success or -EINVAL on failure.
2019 */
prctl_set_seccomp(unsigned long seccomp_mode,void __user * filter)2020 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
2021 {
2022 unsigned int op;
2023 void __user *uargs;
2024
2025 switch (seccomp_mode) {
2026 case SECCOMP_MODE_STRICT:
2027 op = SECCOMP_SET_MODE_STRICT;
2028 /*
2029 * Setting strict mode through prctl always ignored filter,
2030 * so make sure it is always NULL here to pass the internal
2031 * check in do_seccomp().
2032 */
2033 uargs = NULL;
2034 break;
2035 case SECCOMP_MODE_FILTER:
2036 op = SECCOMP_SET_MODE_FILTER;
2037 uargs = filter;
2038 break;
2039 default:
2040 return -EINVAL;
2041 }
2042
2043 /* prctl interface doesn't have flags, so they are always zero. */
2044 return do_seccomp(op, 0, uargs);
2045 }
2046
2047 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
get_nth_filter(struct task_struct * task,unsigned long filter_off)2048 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
2049 unsigned long filter_off)
2050 {
2051 struct seccomp_filter *orig, *filter;
2052 unsigned long count;
2053
2054 /*
2055 * Note: this is only correct because the caller should be the (ptrace)
2056 * tracer of the task, otherwise lock_task_sighand is needed.
2057 */
2058 spin_lock_irq(&task->sighand->siglock);
2059
2060 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
2061 spin_unlock_irq(&task->sighand->siglock);
2062 return ERR_PTR(-EINVAL);
2063 }
2064
2065 orig = task->seccomp.filter;
2066 __get_seccomp_filter(orig);
2067 spin_unlock_irq(&task->sighand->siglock);
2068
2069 count = 0;
2070 for (filter = orig; filter; filter = filter->prev)
2071 count++;
2072
2073 if (filter_off >= count) {
2074 filter = ERR_PTR(-ENOENT);
2075 goto out;
2076 }
2077
2078 count -= filter_off;
2079 for (filter = orig; filter && count > 1; filter = filter->prev)
2080 count--;
2081
2082 if (WARN_ON(count != 1 || !filter)) {
2083 filter = ERR_PTR(-ENOENT);
2084 goto out;
2085 }
2086
2087 __get_seccomp_filter(filter);
2088
2089 out:
2090 __put_seccomp_filter(orig);
2091 return filter;
2092 }
2093
seccomp_get_filter(struct task_struct * task,unsigned long filter_off,void __user * data)2094 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
2095 void __user *data)
2096 {
2097 struct seccomp_filter *filter;
2098 struct sock_fprog_kern *fprog;
2099 long ret;
2100
2101 if (!capable(CAP_SYS_ADMIN) ||
2102 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
2103 return -EACCES;
2104 }
2105
2106 filter = get_nth_filter(task, filter_off);
2107 if (IS_ERR(filter))
2108 return PTR_ERR(filter);
2109
2110 fprog = filter->prog->orig_prog;
2111 if (!fprog) {
2112 /* This must be a new non-cBPF filter, since we save
2113 * every cBPF filter's orig_prog above when
2114 * CONFIG_CHECKPOINT_RESTORE is enabled.
2115 */
2116 ret = -EMEDIUMTYPE;
2117 goto out;
2118 }
2119
2120 ret = fprog->len;
2121 if (!data)
2122 goto out;
2123
2124 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
2125 ret = -EFAULT;
2126
2127 out:
2128 __put_seccomp_filter(filter);
2129 return ret;
2130 }
2131
seccomp_get_metadata(struct task_struct * task,unsigned long size,void __user * data)2132 long seccomp_get_metadata(struct task_struct *task,
2133 unsigned long size, void __user *data)
2134 {
2135 long ret;
2136 struct seccomp_filter *filter;
2137 struct seccomp_metadata kmd = {};
2138
2139 if (!capable(CAP_SYS_ADMIN) ||
2140 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
2141 return -EACCES;
2142 }
2143
2144 size = min_t(unsigned long, size, sizeof(kmd));
2145
2146 if (size < sizeof(kmd.filter_off))
2147 return -EINVAL;
2148
2149 if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
2150 return -EFAULT;
2151
2152 filter = get_nth_filter(task, kmd.filter_off);
2153 if (IS_ERR(filter))
2154 return PTR_ERR(filter);
2155
2156 if (filter->log)
2157 kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
2158
2159 ret = size;
2160 if (copy_to_user(data, &kmd, size))
2161 ret = -EFAULT;
2162
2163 __put_seccomp_filter(filter);
2164 return ret;
2165 }
2166 #endif
2167
2168 #ifdef CONFIG_SYSCTL
2169
2170 /* Human readable action names for friendly sysctl interaction */
2171 #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
2172 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
2173 #define SECCOMP_RET_TRAP_NAME "trap"
2174 #define SECCOMP_RET_ERRNO_NAME "errno"
2175 #define SECCOMP_RET_USER_NOTIF_NAME "user_notif"
2176 #define SECCOMP_RET_TRACE_NAME "trace"
2177 #define SECCOMP_RET_LOG_NAME "log"
2178 #define SECCOMP_RET_ALLOW_NAME "allow"
2179
2180 static const char seccomp_actions_avail[] =
2181 SECCOMP_RET_KILL_PROCESS_NAME " "
2182 SECCOMP_RET_KILL_THREAD_NAME " "
2183 SECCOMP_RET_TRAP_NAME " "
2184 SECCOMP_RET_ERRNO_NAME " "
2185 SECCOMP_RET_USER_NOTIF_NAME " "
2186 SECCOMP_RET_TRACE_NAME " "
2187 SECCOMP_RET_LOG_NAME " "
2188 SECCOMP_RET_ALLOW_NAME;
2189
2190 struct seccomp_log_name {
2191 u32 log;
2192 const char *name;
2193 };
2194
2195 static const struct seccomp_log_name seccomp_log_names[] = {
2196 { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
2197 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
2198 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
2199 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
2200 { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
2201 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
2202 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
2203 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
2204 { }
2205 };
2206
seccomp_names_from_actions_logged(char * names,size_t size,u32 actions_logged,const char * sep)2207 static bool seccomp_names_from_actions_logged(char *names, size_t size,
2208 u32 actions_logged,
2209 const char *sep)
2210 {
2211 const struct seccomp_log_name *cur;
2212 bool append_sep = false;
2213
2214 for (cur = seccomp_log_names; cur->name && size; cur++) {
2215 ssize_t ret;
2216
2217 if (!(actions_logged & cur->log))
2218 continue;
2219
2220 if (append_sep) {
2221 ret = strscpy(names, sep, size);
2222 if (ret < 0)
2223 return false;
2224
2225 names += ret;
2226 size -= ret;
2227 } else
2228 append_sep = true;
2229
2230 ret = strscpy(names, cur->name, size);
2231 if (ret < 0)
2232 return false;
2233
2234 names += ret;
2235 size -= ret;
2236 }
2237
2238 return true;
2239 }
2240
seccomp_action_logged_from_name(u32 * action_logged,const char * name)2241 static bool seccomp_action_logged_from_name(u32 *action_logged,
2242 const char *name)
2243 {
2244 const struct seccomp_log_name *cur;
2245
2246 for (cur = seccomp_log_names; cur->name; cur++) {
2247 if (!strcmp(cur->name, name)) {
2248 *action_logged = cur->log;
2249 return true;
2250 }
2251 }
2252
2253 return false;
2254 }
2255
seccomp_actions_logged_from_names(u32 * actions_logged,char * names)2256 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
2257 {
2258 char *name;
2259
2260 *actions_logged = 0;
2261 while ((name = strsep(&names, " ")) && *name) {
2262 u32 action_logged = 0;
2263
2264 if (!seccomp_action_logged_from_name(&action_logged, name))
2265 return false;
2266
2267 *actions_logged |= action_logged;
2268 }
2269
2270 return true;
2271 }
2272
read_actions_logged(struct ctl_table * ro_table,void * buffer,size_t * lenp,loff_t * ppos)2273 static int read_actions_logged(struct ctl_table *ro_table, void *buffer,
2274 size_t *lenp, loff_t *ppos)
2275 {
2276 char names[sizeof(seccomp_actions_avail)];
2277 struct ctl_table table;
2278
2279 memset(names, 0, sizeof(names));
2280
2281 if (!seccomp_names_from_actions_logged(names, sizeof(names),
2282 seccomp_actions_logged, " "))
2283 return -EINVAL;
2284
2285 table = *ro_table;
2286 table.data = names;
2287 table.maxlen = sizeof(names);
2288 return proc_dostring(&table, 0, buffer, lenp, ppos);
2289 }
2290
write_actions_logged(struct ctl_table * ro_table,void * buffer,size_t * lenp,loff_t * ppos,u32 * actions_logged)2291 static int write_actions_logged(struct ctl_table *ro_table, void *buffer,
2292 size_t *lenp, loff_t *ppos, u32 *actions_logged)
2293 {
2294 char names[sizeof(seccomp_actions_avail)];
2295 struct ctl_table table;
2296 int ret;
2297
2298 if (!capable(CAP_SYS_ADMIN))
2299 return -EPERM;
2300
2301 memset(names, 0, sizeof(names));
2302
2303 table = *ro_table;
2304 table.data = names;
2305 table.maxlen = sizeof(names);
2306 ret = proc_dostring(&table, 1, buffer, lenp, ppos);
2307 if (ret)
2308 return ret;
2309
2310 if (!seccomp_actions_logged_from_names(actions_logged, table.data))
2311 return -EINVAL;
2312
2313 if (*actions_logged & SECCOMP_LOG_ALLOW)
2314 return -EINVAL;
2315
2316 seccomp_actions_logged = *actions_logged;
2317 return 0;
2318 }
2319
audit_actions_logged(u32 actions_logged,u32 old_actions_logged,int ret)2320 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
2321 int ret)
2322 {
2323 char names[sizeof(seccomp_actions_avail)];
2324 char old_names[sizeof(seccomp_actions_avail)];
2325 const char *new = names;
2326 const char *old = old_names;
2327
2328 if (!audit_enabled)
2329 return;
2330
2331 memset(names, 0, sizeof(names));
2332 memset(old_names, 0, sizeof(old_names));
2333
2334 if (ret)
2335 new = "?";
2336 else if (!actions_logged)
2337 new = "(none)";
2338 else if (!seccomp_names_from_actions_logged(names, sizeof(names),
2339 actions_logged, ","))
2340 new = "?";
2341
2342 if (!old_actions_logged)
2343 old = "(none)";
2344 else if (!seccomp_names_from_actions_logged(old_names,
2345 sizeof(old_names),
2346 old_actions_logged, ","))
2347 old = "?";
2348
2349 return audit_seccomp_actions_logged(new, old, !ret);
2350 }
2351
seccomp_actions_logged_handler(struct ctl_table * ro_table,int write,void * buffer,size_t * lenp,loff_t * ppos)2352 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
2353 void *buffer, size_t *lenp,
2354 loff_t *ppos)
2355 {
2356 int ret;
2357
2358 if (write) {
2359 u32 actions_logged = 0;
2360 u32 old_actions_logged = seccomp_actions_logged;
2361
2362 ret = write_actions_logged(ro_table, buffer, lenp, ppos,
2363 &actions_logged);
2364 audit_actions_logged(actions_logged, old_actions_logged, ret);
2365 } else
2366 ret = read_actions_logged(ro_table, buffer, lenp, ppos);
2367
2368 return ret;
2369 }
2370
2371 static struct ctl_path seccomp_sysctl_path[] = {
2372 { .procname = "kernel", },
2373 { .procname = "seccomp", },
2374 { }
2375 };
2376
2377 static struct ctl_table seccomp_sysctl_table[] = {
2378 {
2379 .procname = "actions_avail",
2380 .data = (void *) &seccomp_actions_avail,
2381 .maxlen = sizeof(seccomp_actions_avail),
2382 .mode = 0444,
2383 .proc_handler = proc_dostring,
2384 },
2385 {
2386 .procname = "actions_logged",
2387 .mode = 0644,
2388 .proc_handler = seccomp_actions_logged_handler,
2389 },
2390 { }
2391 };
2392
seccomp_sysctl_init(void)2393 static int __init seccomp_sysctl_init(void)
2394 {
2395 struct ctl_table_header *hdr;
2396
2397 hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
2398 if (!hdr)
2399 pr_warn("sysctl registration failed\n");
2400 else
2401 kmemleak_not_leak(hdr);
2402
2403 return 0;
2404 }
2405
device_initcall(seccomp_sysctl_init)2406 device_initcall(seccomp_sysctl_init)
2407
2408 #endif /* CONFIG_SYSCTL */
2409
2410 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
2411 /* Currently CONFIG_SECCOMP_CACHE_DEBUG implies SECCOMP_ARCH_NATIVE */
2412 static void proc_pid_seccomp_cache_arch(struct seq_file *m, const char *name,
2413 const void *bitmap, size_t bitmap_size)
2414 {
2415 int nr;
2416
2417 for (nr = 0; nr < bitmap_size; nr++) {
2418 bool cached = test_bit(nr, bitmap);
2419 char *status = cached ? "ALLOW" : "FILTER";
2420
2421 seq_printf(m, "%s %d %s\n", name, nr, status);
2422 }
2423 }
2424
proc_pid_seccomp_cache(struct seq_file * m,struct pid_namespace * ns,struct pid * pid,struct task_struct * task)2425 int proc_pid_seccomp_cache(struct seq_file *m, struct pid_namespace *ns,
2426 struct pid *pid, struct task_struct *task)
2427 {
2428 struct seccomp_filter *f;
2429 unsigned long flags;
2430
2431 /*
2432 * We don't want some sandboxed process to know what their seccomp
2433 * filters consist of.
2434 */
2435 if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
2436 return -EACCES;
2437
2438 if (!lock_task_sighand(task, &flags))
2439 return -ESRCH;
2440
2441 f = READ_ONCE(task->seccomp.filter);
2442 if (!f) {
2443 unlock_task_sighand(task, &flags);
2444 return 0;
2445 }
2446
2447 /* prevent filter from being freed while we are printing it */
2448 __get_seccomp_filter(f);
2449 unlock_task_sighand(task, &flags);
2450
2451 proc_pid_seccomp_cache_arch(m, SECCOMP_ARCH_NATIVE_NAME,
2452 f->cache.allow_native,
2453 SECCOMP_ARCH_NATIVE_NR);
2454
2455 #ifdef SECCOMP_ARCH_COMPAT
2456 proc_pid_seccomp_cache_arch(m, SECCOMP_ARCH_COMPAT_NAME,
2457 f->cache.allow_compat,
2458 SECCOMP_ARCH_COMPAT_NR);
2459 #endif /* SECCOMP_ARCH_COMPAT */
2460
2461 __put_seccomp_filter(f);
2462 return 0;
2463 }
2464 #endif /* CONFIG_SECCOMP_CACHE_DEBUG */
2465