]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - kernel/seccomp.c
arm64: mm: Invalidate both kernel and user ASIDs when performing TLBI
[mirror_ubuntu-artful-kernel.git] / kernel / seccomp.c
1 /*
2 * linux/kernel/seccomp.c
3 *
4 * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com>
5 *
6 * Copyright (C) 2012 Google, Inc.
7 * Will Drewry <wad@chromium.org>
8 *
9 * This defines a simple but solid secure-computing facility.
10 *
11 * Mode 1 uses a fixed list of allowed system calls.
12 * Mode 2 allows user-defined system call filters in the form
13 * of Berkeley Packet Filters/Linux Socket Filters.
14 */
15
16 #include <linux/refcount.h>
17 #include <linux/audit.h>
18 #include <linux/compat.h>
19 #include <linux/coredump.h>
20 #include <linux/kmemleak.h>
21 #include <linux/sched.h>
22 #include <linux/sched/task_stack.h>
23 #include <linux/seccomp.h>
24 #include <linux/slab.h>
25 #include <linux/syscalls.h>
26 #include <linux/sysctl.h>
27
28 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
29 #include <asm/syscall.h>
30 #endif
31
32 #ifdef CONFIG_SECCOMP_FILTER
33 #include <linux/filter.h>
34 #include <linux/pid.h>
35 #include <linux/ptrace.h>
36 #include <linux/security.h>
37 #include <linux/tracehook.h>
38 #include <linux/uaccess.h>
39
40 /**
41 * struct seccomp_filter - container for seccomp BPF programs
42 *
43 * @usage: reference count to manage the object lifetime.
44 * get/put helpers should be used when accessing an instance
45 * outside of a lifetime-guarded section. In general, this
46 * is only needed for handling filters shared across tasks.
47 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
48 * @prev: points to a previously installed, or inherited, filter
49 * @prog: the BPF program to evaluate
50 *
51 * seccomp_filter objects are organized in a tree linked via the @prev
52 * pointer. For any task, it appears to be a singly-linked list starting
53 * with current->seccomp.filter, the most recently attached or inherited filter.
54 * However, multiple filters may share a @prev node, by way of fork(), which
55 * results in a unidirectional tree existing in memory. This is similar to
56 * how namespaces work.
57 *
58 * seccomp_filter objects should never be modified after being attached
59 * to a task_struct (other than @usage).
60 */
61 struct seccomp_filter {
62 refcount_t usage;
63 bool log;
64 struct seccomp_filter *prev;
65 struct bpf_prog *prog;
66 };
67
68 /* Limit any path through the tree to 256KB worth of instructions. */
69 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
70
71 /*
72 * Endianness is explicitly ignored and left for BPF program authors to manage
73 * as per the specific architecture.
74 */
75 static void populate_seccomp_data(struct seccomp_data *sd)
76 {
77 struct task_struct *task = current;
78 struct pt_regs *regs = task_pt_regs(task);
79 unsigned long args[6];
80
81 sd->nr = syscall_get_nr(task, regs);
82 sd->arch = syscall_get_arch();
83 syscall_get_arguments(task, regs, 0, 6, args);
84 sd->args[0] = args[0];
85 sd->args[1] = args[1];
86 sd->args[2] = args[2];
87 sd->args[3] = args[3];
88 sd->args[4] = args[4];
89 sd->args[5] = args[5];
90 sd->instruction_pointer = KSTK_EIP(task);
91 }
92
93 /**
94 * seccomp_check_filter - verify seccomp filter code
95 * @filter: filter to verify
96 * @flen: length of filter
97 *
98 * Takes a previously checked filter (by bpf_check_classic) and
99 * redirects all filter code that loads struct sk_buff data
100 * and related data through seccomp_bpf_load. It also
101 * enforces length and alignment checking of those loads.
102 *
103 * Returns 0 if the rule set is legal or -EINVAL if not.
104 */
105 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
106 {
107 int pc;
108 for (pc = 0; pc < flen; pc++) {
109 struct sock_filter *ftest = &filter[pc];
110 u16 code = ftest->code;
111 u32 k = ftest->k;
112
113 switch (code) {
114 case BPF_LD | BPF_W | BPF_ABS:
115 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
116 /* 32-bit aligned and not out of bounds. */
117 if (k >= sizeof(struct seccomp_data) || k & 3)
118 return -EINVAL;
119 continue;
120 case BPF_LD | BPF_W | BPF_LEN:
121 ftest->code = BPF_LD | BPF_IMM;
122 ftest->k = sizeof(struct seccomp_data);
123 continue;
124 case BPF_LDX | BPF_W | BPF_LEN:
125 ftest->code = BPF_LDX | BPF_IMM;
126 ftest->k = sizeof(struct seccomp_data);
127 continue;
128 /* Explicitly include allowed calls. */
129 case BPF_RET | BPF_K:
130 case BPF_RET | BPF_A:
131 case BPF_ALU | BPF_ADD | BPF_K:
132 case BPF_ALU | BPF_ADD | BPF_X:
133 case BPF_ALU | BPF_SUB | BPF_K:
134 case BPF_ALU | BPF_SUB | BPF_X:
135 case BPF_ALU | BPF_MUL | BPF_K:
136 case BPF_ALU | BPF_MUL | BPF_X:
137 case BPF_ALU | BPF_DIV | BPF_K:
138 case BPF_ALU | BPF_DIV | BPF_X:
139 case BPF_ALU | BPF_AND | BPF_K:
140 case BPF_ALU | BPF_AND | BPF_X:
141 case BPF_ALU | BPF_OR | BPF_K:
142 case BPF_ALU | BPF_OR | BPF_X:
143 case BPF_ALU | BPF_XOR | BPF_K:
144 case BPF_ALU | BPF_XOR | BPF_X:
145 case BPF_ALU | BPF_LSH | BPF_K:
146 case BPF_ALU | BPF_LSH | BPF_X:
147 case BPF_ALU | BPF_RSH | BPF_K:
148 case BPF_ALU | BPF_RSH | BPF_X:
149 case BPF_ALU | BPF_NEG:
150 case BPF_LD | BPF_IMM:
151 case BPF_LDX | BPF_IMM:
152 case BPF_MISC | BPF_TAX:
153 case BPF_MISC | BPF_TXA:
154 case BPF_LD | BPF_MEM:
155 case BPF_LDX | BPF_MEM:
156 case BPF_ST:
157 case BPF_STX:
158 case BPF_JMP | BPF_JA:
159 case BPF_JMP | BPF_JEQ | BPF_K:
160 case BPF_JMP | BPF_JEQ | BPF_X:
161 case BPF_JMP | BPF_JGE | BPF_K:
162 case BPF_JMP | BPF_JGE | BPF_X:
163 case BPF_JMP | BPF_JGT | BPF_K:
164 case BPF_JMP | BPF_JGT | BPF_X:
165 case BPF_JMP | BPF_JSET | BPF_K:
166 case BPF_JMP | BPF_JSET | BPF_X:
167 continue;
168 default:
169 return -EINVAL;
170 }
171 }
172 return 0;
173 }
174
175 /**
176 * seccomp_run_filters - evaluates all seccomp filters against @sd
177 * @sd: optional seccomp data to be passed to filters
178 * @match: stores struct seccomp_filter that resulted in the return value,
179 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
180 * be unchanged.
181 *
182 * Returns valid seccomp BPF response codes.
183 */
184 static u32 seccomp_run_filters(const struct seccomp_data *sd,
185 struct seccomp_filter **match)
186 {
187 struct seccomp_data sd_local;
188 u32 ret = SECCOMP_RET_ALLOW;
189 /* Make sure cross-thread synced filter points somewhere sane. */
190 struct seccomp_filter *f =
191 READ_ONCE(current->seccomp.filter);
192
193 /* Ensure unexpected behavior doesn't result in failing open. */
194 if (unlikely(WARN_ON(f == NULL)))
195 return SECCOMP_RET_KILL;
196
197 if (!sd) {
198 populate_seccomp_data(&sd_local);
199 sd = &sd_local;
200 }
201
202 /*
203 * All filters in the list are evaluated and the lowest BPF return
204 * value always takes priority (ignoring the DATA).
205 */
206 for (; f; f = f->prev) {
207 u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
208
209 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION)) {
210 ret = cur_ret;
211 *match = f;
212 }
213 }
214 return ret;
215 }
216 #endif /* CONFIG_SECCOMP_FILTER */
217
218 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
219 {
220 assert_spin_locked(&current->sighand->siglock);
221
222 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
223 return false;
224
225 return true;
226 }
227
228 static inline void seccomp_assign_mode(struct task_struct *task,
229 unsigned long seccomp_mode)
230 {
231 assert_spin_locked(&task->sighand->siglock);
232
233 task->seccomp.mode = seccomp_mode;
234 /*
235 * Make sure TIF_SECCOMP cannot be set before the mode (and
236 * filter) is set.
237 */
238 smp_mb__before_atomic();
239 set_tsk_thread_flag(task, TIF_SECCOMP);
240 }
241
242 #ifdef CONFIG_SECCOMP_FILTER
243 /* Returns 1 if the parent is an ancestor of the child. */
244 static int is_ancestor(struct seccomp_filter *parent,
245 struct seccomp_filter *child)
246 {
247 /* NULL is the root ancestor. */
248 if (parent == NULL)
249 return 1;
250 for (; child; child = child->prev)
251 if (child == parent)
252 return 1;
253 return 0;
254 }
255
256 /**
257 * seccomp_can_sync_threads: checks if all threads can be synchronized
258 *
259 * Expects sighand and cred_guard_mutex locks to be held.
260 *
261 * Returns 0 on success, -ve on error, or the pid of a thread which was
262 * either not in the correct seccomp mode or it did not have an ancestral
263 * seccomp filter.
264 */
265 static inline pid_t seccomp_can_sync_threads(void)
266 {
267 struct task_struct *thread, *caller;
268
269 BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
270 assert_spin_locked(&current->sighand->siglock);
271
272 /* Validate all threads being eligible for synchronization. */
273 caller = current;
274 for_each_thread(caller, thread) {
275 pid_t failed;
276
277 /* Skip current, since it is initiating the sync. */
278 if (thread == caller)
279 continue;
280
281 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
282 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
283 is_ancestor(thread->seccomp.filter,
284 caller->seccomp.filter)))
285 continue;
286
287 /* Return the first thread that cannot be synchronized. */
288 failed = task_pid_vnr(thread);
289 /* If the pid cannot be resolved, then return -ESRCH */
290 if (unlikely(WARN_ON(failed == 0)))
291 failed = -ESRCH;
292 return failed;
293 }
294
295 return 0;
296 }
297
298 /**
299 * seccomp_sync_threads: sets all threads to use current's filter
300 *
301 * Expects sighand and cred_guard_mutex locks to be held, and for
302 * seccomp_can_sync_threads() to have returned success already
303 * without dropping the locks.
304 *
305 */
306 static inline void seccomp_sync_threads(void)
307 {
308 struct task_struct *thread, *caller;
309
310 BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
311 assert_spin_locked(&current->sighand->siglock);
312
313 /* Synchronize all threads. */
314 caller = current;
315 for_each_thread(caller, thread) {
316 /* Skip current, since it needs no changes. */
317 if (thread == caller)
318 continue;
319
320 /* Get a task reference for the new leaf node. */
321 get_seccomp_filter(caller);
322 /*
323 * Drop the task reference to the shared ancestor since
324 * current's path will hold a reference. (This also
325 * allows a put before the assignment.)
326 */
327 put_seccomp_filter(thread);
328 smp_store_release(&thread->seccomp.filter,
329 caller->seccomp.filter);
330
331 /*
332 * Don't let an unprivileged task work around
333 * the no_new_privs restriction by creating
334 * a thread that sets it up, enters seccomp,
335 * then dies.
336 */
337 if (task_no_new_privs(caller))
338 task_set_no_new_privs(thread);
339
340 /*
341 * Opt the other thread into seccomp if needed.
342 * As threads are considered to be trust-realm
343 * equivalent (see ptrace_may_access), it is safe to
344 * allow one thread to transition the other.
345 */
346 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
347 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER);
348 }
349 }
350
351 /**
352 * seccomp_prepare_filter: Prepares a seccomp filter for use.
353 * @fprog: BPF program to install
354 *
355 * Returns filter on success or an ERR_PTR on failure.
356 */
357 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
358 {
359 struct seccomp_filter *sfilter;
360 int ret;
361 const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
362
363 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
364 return ERR_PTR(-EINVAL);
365
366 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
367
368 /*
369 * Installing a seccomp filter requires that the task has
370 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
371 * This avoids scenarios where unprivileged tasks can affect the
372 * behavior of privileged children.
373 */
374 if (!task_no_new_privs(current) &&
375 security_capable_noaudit(current_cred(), current_user_ns(),
376 CAP_SYS_ADMIN) != 0)
377 return ERR_PTR(-EACCES);
378
379 /* Allocate a new seccomp_filter */
380 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
381 if (!sfilter)
382 return ERR_PTR(-ENOMEM);
383
384 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
385 seccomp_check_filter, save_orig);
386 if (ret < 0) {
387 kfree(sfilter);
388 return ERR_PTR(ret);
389 }
390
391 refcount_set(&sfilter->usage, 1);
392
393 return sfilter;
394 }
395
396 /**
397 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
398 * @user_filter: pointer to the user data containing a sock_fprog.
399 *
400 * Returns 0 on success and non-zero otherwise.
401 */
402 static struct seccomp_filter *
403 seccomp_prepare_user_filter(const char __user *user_filter)
404 {
405 struct sock_fprog fprog;
406 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
407
408 #ifdef CONFIG_COMPAT
409 if (in_compat_syscall()) {
410 struct compat_sock_fprog fprog32;
411 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
412 goto out;
413 fprog.len = fprog32.len;
414 fprog.filter = compat_ptr(fprog32.filter);
415 } else /* falls through to the if below. */
416 #endif
417 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
418 goto out;
419 filter = seccomp_prepare_filter(&fprog);
420 out:
421 return filter;
422 }
423
424 /**
425 * seccomp_attach_filter: validate and attach filter
426 * @flags: flags to change filter behavior
427 * @filter: seccomp filter to add to the current process
428 *
429 * Caller must be holding current->sighand->siglock lock.
430 *
431 * Returns 0 on success, -ve on error.
432 */
433 static long seccomp_attach_filter(unsigned int flags,
434 struct seccomp_filter *filter)
435 {
436 unsigned long total_insns;
437 struct seccomp_filter *walker;
438
439 assert_spin_locked(&current->sighand->siglock);
440
441 /* Validate resulting filter length. */
442 total_insns = filter->prog->len;
443 for (walker = current->seccomp.filter; walker; walker = walker->prev)
444 total_insns += walker->prog->len + 4; /* 4 instr penalty */
445 if (total_insns > MAX_INSNS_PER_PATH)
446 return -ENOMEM;
447
448 /* If thread sync has been requested, check that it is possible. */
449 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
450 int ret;
451
452 ret = seccomp_can_sync_threads();
453 if (ret)
454 return ret;
455 }
456
457 /* Set log flag, if present. */
458 if (flags & SECCOMP_FILTER_FLAG_LOG)
459 filter->log = true;
460
461 /*
462 * If there is an existing filter, make it the prev and don't drop its
463 * task reference.
464 */
465 filter->prev = current->seccomp.filter;
466 current->seccomp.filter = filter;
467
468 /* Now that the new filter is in place, synchronize to all threads. */
469 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
470 seccomp_sync_threads();
471
472 return 0;
473 }
474
475 void __get_seccomp_filter(struct seccomp_filter *filter)
476 {
477 /* Reference count is bounded by the number of total processes. */
478 refcount_inc(&filter->usage);
479 }
480
481 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
482 void get_seccomp_filter(struct task_struct *tsk)
483 {
484 struct seccomp_filter *orig = tsk->seccomp.filter;
485 if (!orig)
486 return;
487 __get_seccomp_filter(orig);
488 }
489
490 static inline void seccomp_filter_free(struct seccomp_filter *filter)
491 {
492 if (filter) {
493 bpf_prog_destroy(filter->prog);
494 kfree(filter);
495 }
496 }
497
498 static void __put_seccomp_filter(struct seccomp_filter *orig)
499 {
500 /* Clean up single-reference branches iteratively. */
501 while (orig && refcount_dec_and_test(&orig->usage)) {
502 struct seccomp_filter *freeme = orig;
503 orig = orig->prev;
504 seccomp_filter_free(freeme);
505 }
506 }
507
508 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
509 void put_seccomp_filter(struct task_struct *tsk)
510 {
511 __put_seccomp_filter(tsk->seccomp.filter);
512 }
513
514 static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason)
515 {
516 memset(info, 0, sizeof(*info));
517 info->si_signo = SIGSYS;
518 info->si_code = SYS_SECCOMP;
519 info->si_call_addr = (void __user *)KSTK_EIP(current);
520 info->si_errno = reason;
521 info->si_arch = syscall_get_arch();
522 info->si_syscall = syscall;
523 }
524
525 /**
526 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
527 * @syscall: syscall number to send to userland
528 * @reason: filter-supplied reason code to send to userland (via si_errno)
529 *
530 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
531 */
532 static void seccomp_send_sigsys(int syscall, int reason)
533 {
534 struct siginfo info;
535 seccomp_init_siginfo(&info, syscall, reason);
536 force_sig_info(SIGSYS, &info, current);
537 }
538 #endif /* CONFIG_SECCOMP_FILTER */
539
540 /* For use with seccomp_actions_logged */
541 #define SECCOMP_LOG_KILL (1 << 0)
542 #define SECCOMP_LOG_TRAP (1 << 2)
543 #define SECCOMP_LOG_ERRNO (1 << 3)
544 #define SECCOMP_LOG_TRACE (1 << 4)
545 #define SECCOMP_LOG_LOG (1 << 5)
546 #define SECCOMP_LOG_ALLOW (1 << 6)
547
548 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL | SECCOMP_LOG_TRAP |
549 SECCOMP_LOG_ERRNO | SECCOMP_LOG_TRACE |
550 SECCOMP_LOG_LOG;
551
552 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
553 bool requested)
554 {
555 bool log = false;
556
557 switch (action) {
558 case SECCOMP_RET_ALLOW:
559 break;
560 case SECCOMP_RET_TRAP:
561 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
562 break;
563 case SECCOMP_RET_ERRNO:
564 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
565 break;
566 case SECCOMP_RET_TRACE:
567 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
568 break;
569 case SECCOMP_RET_LOG:
570 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
571 break;
572 case SECCOMP_RET_KILL:
573 default:
574 log = seccomp_actions_logged & SECCOMP_LOG_KILL;
575 }
576
577 /*
578 * Force an audit message to be emitted when the action is RET_KILL,
579 * RET_LOG, or the FILTER_FLAG_LOG bit was set and the action is
580 * allowed to be logged by the admin.
581 */
582 if (log)
583 return __audit_seccomp(syscall, signr, action);
584
585 /*
586 * Let the audit subsystem decide if the action should be audited based
587 * on whether the current task itself is being audited.
588 */
589 return audit_seccomp(syscall, signr, action);
590 }
591
592 /*
593 * Secure computing mode 1 allows only read/write/exit/sigreturn.
594 * To be fully secure this must be combined with rlimit
595 * to limit the stack allocations too.
596 */
597 static const int mode1_syscalls[] = {
598 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
599 0, /* null terminated */
600 };
601
602 static void __secure_computing_strict(int this_syscall)
603 {
604 const int *syscall_whitelist = mode1_syscalls;
605 #ifdef CONFIG_COMPAT
606 if (in_compat_syscall())
607 syscall_whitelist = get_compat_mode1_syscalls();
608 #endif
609 do {
610 if (*syscall_whitelist == this_syscall)
611 return;
612 } while (*++syscall_whitelist);
613
614 #ifdef SECCOMP_DEBUG
615 dump_stack();
616 #endif
617 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL, true);
618 do_exit(SIGKILL);
619 }
620
621 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
622 void secure_computing_strict(int this_syscall)
623 {
624 int mode = current->seccomp.mode;
625
626 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
627 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
628 return;
629
630 if (mode == SECCOMP_MODE_DISABLED)
631 return;
632 else if (mode == SECCOMP_MODE_STRICT)
633 __secure_computing_strict(this_syscall);
634 else
635 BUG();
636 }
637 #else
638
639 #ifdef CONFIG_SECCOMP_FILTER
640 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
641 const bool recheck_after_trace)
642 {
643 u32 filter_ret, action;
644 struct seccomp_filter *match = NULL;
645 int data;
646
647 /*
648 * Make sure that any changes to mode from another thread have
649 * been seen after TIF_SECCOMP was seen.
650 */
651 rmb();
652
653 filter_ret = seccomp_run_filters(sd, &match);
654 data = filter_ret & SECCOMP_RET_DATA;
655 action = filter_ret & SECCOMP_RET_ACTION;
656
657 switch (action) {
658 case SECCOMP_RET_ERRNO:
659 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
660 if (data > MAX_ERRNO)
661 data = MAX_ERRNO;
662 syscall_set_return_value(current, task_pt_regs(current),
663 -data, 0);
664 goto skip;
665
666 case SECCOMP_RET_TRAP:
667 /* Show the handler the original registers. */
668 syscall_rollback(current, task_pt_regs(current));
669 /* Let the filter pass back 16 bits of data. */
670 seccomp_send_sigsys(this_syscall, data);
671 goto skip;
672
673 case SECCOMP_RET_TRACE:
674 /* We've been put in this state by the ptracer already. */
675 if (recheck_after_trace)
676 return 0;
677
678 /* ENOSYS these calls if there is no tracer attached. */
679 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
680 syscall_set_return_value(current,
681 task_pt_regs(current),
682 -ENOSYS, 0);
683 goto skip;
684 }
685
686 /* Allow the BPF to provide the event message */
687 ptrace_event(PTRACE_EVENT_SECCOMP, data);
688 /*
689 * The delivery of a fatal signal during event
690 * notification may silently skip tracer notification,
691 * which could leave us with a potentially unmodified
692 * syscall that the tracer would have liked to have
693 * changed. Since the process is about to die, we just
694 * force the syscall to be skipped and let the signal
695 * kill the process and correctly handle any tracer exit
696 * notifications.
697 */
698 if (fatal_signal_pending(current))
699 goto skip;
700 /* Check if the tracer forced the syscall to be skipped. */
701 this_syscall = syscall_get_nr(current, task_pt_regs(current));
702 if (this_syscall < 0)
703 goto skip;
704
705 /*
706 * Recheck the syscall, since it may have changed. This
707 * intentionally uses a NULL struct seccomp_data to force
708 * a reload of all registers. This does not goto skip since
709 * a skip would have already been reported.
710 */
711 if (__seccomp_filter(this_syscall, NULL, true))
712 return -1;
713
714 return 0;
715
716 case SECCOMP_RET_LOG:
717 seccomp_log(this_syscall, 0, action, true);
718 return 0;
719
720 case SECCOMP_RET_ALLOW:
721 /*
722 * Note that the "match" filter will always be NULL for
723 * this action since SECCOMP_RET_ALLOW is the starting
724 * state in seccomp_run_filters().
725 */
726 return 0;
727
728 case SECCOMP_RET_KILL:
729 default:
730 seccomp_log(this_syscall, SIGSYS, action, true);
731 /* Dump core only if this is the last remaining thread. */
732 if (get_nr_threads(current) == 1) {
733 siginfo_t info;
734
735 /* Show the original registers in the dump. */
736 syscall_rollback(current, task_pt_regs(current));
737 /* Trigger a manual coredump since do_exit skips it. */
738 seccomp_init_siginfo(&info, this_syscall, data);
739 do_coredump(&info);
740 }
741 do_exit(SIGSYS);
742 }
743
744 unreachable();
745
746 skip:
747 seccomp_log(this_syscall, 0, action, match ? match->log : false);
748 return -1;
749 }
750 #else
751 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
752 const bool recheck_after_trace)
753 {
754 BUG();
755 }
756 #endif
757
758 int __secure_computing(const struct seccomp_data *sd)
759 {
760 int mode = current->seccomp.mode;
761 int this_syscall;
762
763 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
764 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
765 return 0;
766
767 this_syscall = sd ? sd->nr :
768 syscall_get_nr(current, task_pt_regs(current));
769
770 switch (mode) {
771 case SECCOMP_MODE_STRICT:
772 __secure_computing_strict(this_syscall); /* may call do_exit */
773 return 0;
774 case SECCOMP_MODE_FILTER:
775 return __seccomp_filter(this_syscall, sd, false);
776 default:
777 BUG();
778 }
779 }
780 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
781
782 long prctl_get_seccomp(void)
783 {
784 return current->seccomp.mode;
785 }
786
787 /**
788 * seccomp_set_mode_strict: internal function for setting strict seccomp
789 *
790 * Once current->seccomp.mode is non-zero, it may not be changed.
791 *
792 * Returns 0 on success or -EINVAL on failure.
793 */
794 static long seccomp_set_mode_strict(void)
795 {
796 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
797 long ret = -EINVAL;
798
799 spin_lock_irq(&current->sighand->siglock);
800
801 if (!seccomp_may_assign_mode(seccomp_mode))
802 goto out;
803
804 #ifdef TIF_NOTSC
805 disable_TSC();
806 #endif
807 seccomp_assign_mode(current, seccomp_mode);
808 ret = 0;
809
810 out:
811 spin_unlock_irq(&current->sighand->siglock);
812
813 return ret;
814 }
815
816 #ifdef CONFIG_SECCOMP_FILTER
817 /**
818 * seccomp_set_mode_filter: internal function for setting seccomp filter
819 * @flags: flags to change filter behavior
820 * @filter: struct sock_fprog containing filter
821 *
822 * This function may be called repeatedly to install additional filters.
823 * Every filter successfully installed will be evaluated (in reverse order)
824 * for each system call the task makes.
825 *
826 * Once current->seccomp.mode is non-zero, it may not be changed.
827 *
828 * Returns 0 on success or -EINVAL on failure.
829 */
830 static long seccomp_set_mode_filter(unsigned int flags,
831 const char __user *filter)
832 {
833 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
834 struct seccomp_filter *prepared = NULL;
835 long ret = -EINVAL;
836
837 /* Validate flags. */
838 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
839 return -EINVAL;
840
841 /* Prepare the new filter before holding any locks. */
842 prepared = seccomp_prepare_user_filter(filter);
843 if (IS_ERR(prepared))
844 return PTR_ERR(prepared);
845
846 /*
847 * Make sure we cannot change seccomp or nnp state via TSYNC
848 * while another thread is in the middle of calling exec.
849 */
850 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
851 mutex_lock_killable(&current->signal->cred_guard_mutex))
852 goto out_free;
853
854 spin_lock_irq(&current->sighand->siglock);
855
856 if (!seccomp_may_assign_mode(seccomp_mode))
857 goto out;
858
859 ret = seccomp_attach_filter(flags, prepared);
860 if (ret)
861 goto out;
862 /* Do not free the successfully attached filter. */
863 prepared = NULL;
864
865 seccomp_assign_mode(current, seccomp_mode);
866 out:
867 spin_unlock_irq(&current->sighand->siglock);
868 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
869 mutex_unlock(&current->signal->cred_guard_mutex);
870 out_free:
871 seccomp_filter_free(prepared);
872 return ret;
873 }
874 #else
875 static inline long seccomp_set_mode_filter(unsigned int flags,
876 const char __user *filter)
877 {
878 return -EINVAL;
879 }
880 #endif
881
882 static long seccomp_get_action_avail(const char __user *uaction)
883 {
884 u32 action;
885
886 if (copy_from_user(&action, uaction, sizeof(action)))
887 return -EFAULT;
888
889 switch (action) {
890 case SECCOMP_RET_KILL:
891 case SECCOMP_RET_TRAP:
892 case SECCOMP_RET_ERRNO:
893 case SECCOMP_RET_TRACE:
894 case SECCOMP_RET_LOG:
895 case SECCOMP_RET_ALLOW:
896 break;
897 default:
898 return -EOPNOTSUPP;
899 }
900
901 return 0;
902 }
903
904 /* Common entry point for both prctl and syscall. */
905 static long do_seccomp(unsigned int op, unsigned int flags,
906 const char __user *uargs)
907 {
908 switch (op) {
909 case SECCOMP_SET_MODE_STRICT:
910 if (flags != 0 || uargs != NULL)
911 return -EINVAL;
912 return seccomp_set_mode_strict();
913 case SECCOMP_SET_MODE_FILTER:
914 return seccomp_set_mode_filter(flags, uargs);
915 case SECCOMP_GET_ACTION_AVAIL:
916 if (flags != 0)
917 return -EINVAL;
918
919 return seccomp_get_action_avail(uargs);
920 default:
921 return -EINVAL;
922 }
923 }
924
925 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
926 const char __user *, uargs)
927 {
928 return do_seccomp(op, flags, uargs);
929 }
930
931 /**
932 * prctl_set_seccomp: configures current->seccomp.mode
933 * @seccomp_mode: requested mode to use
934 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
935 *
936 * Returns 0 on success or -EINVAL on failure.
937 */
938 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
939 {
940 unsigned int op;
941 char __user *uargs;
942
943 switch (seccomp_mode) {
944 case SECCOMP_MODE_STRICT:
945 op = SECCOMP_SET_MODE_STRICT;
946 /*
947 * Setting strict mode through prctl always ignored filter,
948 * so make sure it is always NULL here to pass the internal
949 * check in do_seccomp().
950 */
951 uargs = NULL;
952 break;
953 case SECCOMP_MODE_FILTER:
954 op = SECCOMP_SET_MODE_FILTER;
955 uargs = filter;
956 break;
957 default:
958 return -EINVAL;
959 }
960
961 /* prctl interface doesn't have flags, so they are always zero. */
962 return do_seccomp(op, 0, uargs);
963 }
964
965 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
966 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
967 void __user *data)
968 {
969 struct seccomp_filter *filter;
970 struct sock_fprog_kern *fprog;
971 long ret;
972 unsigned long count = 0;
973
974 if (!capable(CAP_SYS_ADMIN) ||
975 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
976 return -EACCES;
977 }
978
979 spin_lock_irq(&task->sighand->siglock);
980 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
981 ret = -EINVAL;
982 goto out;
983 }
984
985 filter = task->seccomp.filter;
986 while (filter) {
987 filter = filter->prev;
988 count++;
989 }
990
991 if (filter_off >= count) {
992 ret = -ENOENT;
993 goto out;
994 }
995 count -= filter_off;
996
997 filter = task->seccomp.filter;
998 while (filter && count > 1) {
999 filter = filter->prev;
1000 count--;
1001 }
1002
1003 if (WARN_ON(count != 1 || !filter)) {
1004 /* The filter tree shouldn't shrink while we're using it. */
1005 ret = -ENOENT;
1006 goto out;
1007 }
1008
1009 fprog = filter->prog->orig_prog;
1010 if (!fprog) {
1011 /* This must be a new non-cBPF filter, since we save
1012 * every cBPF filter's orig_prog above when
1013 * CONFIG_CHECKPOINT_RESTORE is enabled.
1014 */
1015 ret = -EMEDIUMTYPE;
1016 goto out;
1017 }
1018
1019 ret = fprog->len;
1020 if (!data)
1021 goto out;
1022
1023 __get_seccomp_filter(filter);
1024 spin_unlock_irq(&task->sighand->siglock);
1025
1026 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1027 ret = -EFAULT;
1028
1029 __put_seccomp_filter(filter);
1030 return ret;
1031
1032 out:
1033 spin_unlock_irq(&task->sighand->siglock);
1034 return ret;
1035 }
1036 #endif
1037
1038 #ifdef CONFIG_SYSCTL
1039
1040 /* Human readable action names for friendly sysctl interaction */
1041 #define SECCOMP_RET_KILL_NAME "kill"
1042 #define SECCOMP_RET_TRAP_NAME "trap"
1043 #define SECCOMP_RET_ERRNO_NAME "errno"
1044 #define SECCOMP_RET_TRACE_NAME "trace"
1045 #define SECCOMP_RET_LOG_NAME "log"
1046 #define SECCOMP_RET_ALLOW_NAME "allow"
1047
1048 static const char seccomp_actions_avail[] = SECCOMP_RET_KILL_NAME " "
1049 SECCOMP_RET_TRAP_NAME " "
1050 SECCOMP_RET_ERRNO_NAME " "
1051 SECCOMP_RET_TRACE_NAME " "
1052 SECCOMP_RET_LOG_NAME " "
1053 SECCOMP_RET_ALLOW_NAME;
1054
1055 struct seccomp_log_name {
1056 u32 log;
1057 const char *name;
1058 };
1059
1060 static const struct seccomp_log_name seccomp_log_names[] = {
1061 { SECCOMP_LOG_KILL, SECCOMP_RET_KILL_NAME },
1062 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1063 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1064 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1065 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1066 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1067 { }
1068 };
1069
1070 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1071 u32 actions_logged)
1072 {
1073 const struct seccomp_log_name *cur;
1074 bool append_space = false;
1075
1076 for (cur = seccomp_log_names; cur->name && size; cur++) {
1077 ssize_t ret;
1078
1079 if (!(actions_logged & cur->log))
1080 continue;
1081
1082 if (append_space) {
1083 ret = strscpy(names, " ", size);
1084 if (ret < 0)
1085 return false;
1086
1087 names += ret;
1088 size -= ret;
1089 } else
1090 append_space = true;
1091
1092 ret = strscpy(names, cur->name, size);
1093 if (ret < 0)
1094 return false;
1095
1096 names += ret;
1097 size -= ret;
1098 }
1099
1100 return true;
1101 }
1102
1103 static bool seccomp_action_logged_from_name(u32 *action_logged,
1104 const char *name)
1105 {
1106 const struct seccomp_log_name *cur;
1107
1108 for (cur = seccomp_log_names; cur->name; cur++) {
1109 if (!strcmp(cur->name, name)) {
1110 *action_logged = cur->log;
1111 return true;
1112 }
1113 }
1114
1115 return false;
1116 }
1117
1118 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1119 {
1120 char *name;
1121
1122 *actions_logged = 0;
1123 while ((name = strsep(&names, " ")) && *name) {
1124 u32 action_logged = 0;
1125
1126 if (!seccomp_action_logged_from_name(&action_logged, name))
1127 return false;
1128
1129 *actions_logged |= action_logged;
1130 }
1131
1132 return true;
1133 }
1134
1135 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1136 void __user *buffer, size_t *lenp,
1137 loff_t *ppos)
1138 {
1139 char names[sizeof(seccomp_actions_avail)];
1140 struct ctl_table table;
1141 int ret;
1142
1143 if (write && !capable(CAP_SYS_ADMIN))
1144 return -EPERM;
1145
1146 memset(names, 0, sizeof(names));
1147
1148 if (!write) {
1149 if (!seccomp_names_from_actions_logged(names, sizeof(names),
1150 seccomp_actions_logged))
1151 return -EINVAL;
1152 }
1153
1154 table = *ro_table;
1155 table.data = names;
1156 table.maxlen = sizeof(names);
1157 ret = proc_dostring(&table, write, buffer, lenp, ppos);
1158 if (ret)
1159 return ret;
1160
1161 if (write) {
1162 u32 actions_logged;
1163
1164 if (!seccomp_actions_logged_from_names(&actions_logged,
1165 table.data))
1166 return -EINVAL;
1167
1168 if (actions_logged & SECCOMP_LOG_ALLOW)
1169 return -EINVAL;
1170
1171 seccomp_actions_logged = actions_logged;
1172 }
1173
1174 return 0;
1175 }
1176
1177 static struct ctl_path seccomp_sysctl_path[] = {
1178 { .procname = "kernel", },
1179 { .procname = "seccomp", },
1180 { }
1181 };
1182
1183 static struct ctl_table seccomp_sysctl_table[] = {
1184 {
1185 .procname = "actions_avail",
1186 .data = (void *) &seccomp_actions_avail,
1187 .maxlen = sizeof(seccomp_actions_avail),
1188 .mode = 0444,
1189 .proc_handler = proc_dostring,
1190 },
1191 {
1192 .procname = "actions_logged",
1193 .mode = 0644,
1194 .proc_handler = seccomp_actions_logged_handler,
1195 },
1196 { }
1197 };
1198
1199 static int __init seccomp_sysctl_init(void)
1200 {
1201 struct ctl_table_header *hdr;
1202
1203 hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1204 if (!hdr)
1205 pr_warn("seccomp: sysctl registration failed\n");
1206 else
1207 kmemleak_not_leak(hdr);
1208
1209 return 0;
1210 }
1211
1212 device_initcall(seccomp_sysctl_init)
1213
1214 #endif /* CONFIG_SYSCTL */