seccomp: split filter prep from check and apply
[deliverable/linux.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/atomic.h>
17 #include <linux/audit.h>
18 #include <linux/compat.h>
19 #include <linux/sched.h>
20 #include <linux/seccomp.h>
21 #include <linux/slab.h>
22 #include <linux/syscalls.h>
23
24 /* #define SECCOMP_DEBUG 1 */
25
26 #ifdef CONFIG_SECCOMP_FILTER
27 #include <asm/syscall.h>
28 #include <linux/filter.h>
29 #include <linux/ptrace.h>
30 #include <linux/security.h>
31 #include <linux/tracehook.h>
32 #include <linux/uaccess.h>
33
34 /**
35 * struct seccomp_filter - container for seccomp BPF programs
36 *
37 * @usage: reference count to manage the object lifetime.
38 * get/put helpers should be used when accessing an instance
39 * outside of a lifetime-guarded section. In general, this
40 * is only needed for handling filters shared across tasks.
41 * @prev: points to a previously installed, or inherited, filter
42 * @len: the number of instructions in the program
43 * @insnsi: the BPF program instructions to evaluate
44 *
45 * seccomp_filter objects are organized in a tree linked via the @prev
46 * pointer. For any task, it appears to be a singly-linked list starting
47 * with current->seccomp.filter, the most recently attached or inherited filter.
48 * However, multiple filters may share a @prev node, by way of fork(), which
49 * results in a unidirectional tree existing in memory. This is similar to
50 * how namespaces work.
51 *
52 * seccomp_filter objects should never be modified after being attached
53 * to a task_struct (other than @usage).
54 */
55 struct seccomp_filter {
56 atomic_t usage;
57 struct seccomp_filter *prev;
58 struct sk_filter *prog;
59 };
60
61 /* Limit any path through the tree to 256KB worth of instructions. */
62 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
63
64 /*
65 * Endianness is explicitly ignored and left for BPF program authors to manage
66 * as per the specific architecture.
67 */
68 static void populate_seccomp_data(struct seccomp_data *sd)
69 {
70 struct task_struct *task = current;
71 struct pt_regs *regs = task_pt_regs(task);
72 unsigned long args[6];
73
74 sd->nr = syscall_get_nr(task, regs);
75 sd->arch = syscall_get_arch();
76 syscall_get_arguments(task, regs, 0, 6, args);
77 sd->args[0] = args[0];
78 sd->args[1] = args[1];
79 sd->args[2] = args[2];
80 sd->args[3] = args[3];
81 sd->args[4] = args[4];
82 sd->args[5] = args[5];
83 sd->instruction_pointer = KSTK_EIP(task);
84 }
85
86 /**
87 * seccomp_check_filter - verify seccomp filter code
88 * @filter: filter to verify
89 * @flen: length of filter
90 *
91 * Takes a previously checked filter (by sk_chk_filter) and
92 * redirects all filter code that loads struct sk_buff data
93 * and related data through seccomp_bpf_load. It also
94 * enforces length and alignment checking of those loads.
95 *
96 * Returns 0 if the rule set is legal or -EINVAL if not.
97 */
98 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
99 {
100 int pc;
101 for (pc = 0; pc < flen; pc++) {
102 struct sock_filter *ftest = &filter[pc];
103 u16 code = ftest->code;
104 u32 k = ftest->k;
105
106 switch (code) {
107 case BPF_LD | BPF_W | BPF_ABS:
108 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
109 /* 32-bit aligned and not out of bounds. */
110 if (k >= sizeof(struct seccomp_data) || k & 3)
111 return -EINVAL;
112 continue;
113 case BPF_LD | BPF_W | BPF_LEN:
114 ftest->code = BPF_LD | BPF_IMM;
115 ftest->k = sizeof(struct seccomp_data);
116 continue;
117 case BPF_LDX | BPF_W | BPF_LEN:
118 ftest->code = BPF_LDX | BPF_IMM;
119 ftest->k = sizeof(struct seccomp_data);
120 continue;
121 /* Explicitly include allowed calls. */
122 case BPF_RET | BPF_K:
123 case BPF_RET | BPF_A:
124 case BPF_ALU | BPF_ADD | BPF_K:
125 case BPF_ALU | BPF_ADD | BPF_X:
126 case BPF_ALU | BPF_SUB | BPF_K:
127 case BPF_ALU | BPF_SUB | BPF_X:
128 case BPF_ALU | BPF_MUL | BPF_K:
129 case BPF_ALU | BPF_MUL | BPF_X:
130 case BPF_ALU | BPF_DIV | BPF_K:
131 case BPF_ALU | BPF_DIV | BPF_X:
132 case BPF_ALU | BPF_AND | BPF_K:
133 case BPF_ALU | BPF_AND | BPF_X:
134 case BPF_ALU | BPF_OR | BPF_K:
135 case BPF_ALU | BPF_OR | BPF_X:
136 case BPF_ALU | BPF_XOR | BPF_K:
137 case BPF_ALU | BPF_XOR | BPF_X:
138 case BPF_ALU | BPF_LSH | BPF_K:
139 case BPF_ALU | BPF_LSH | BPF_X:
140 case BPF_ALU | BPF_RSH | BPF_K:
141 case BPF_ALU | BPF_RSH | BPF_X:
142 case BPF_ALU | BPF_NEG:
143 case BPF_LD | BPF_IMM:
144 case BPF_LDX | BPF_IMM:
145 case BPF_MISC | BPF_TAX:
146 case BPF_MISC | BPF_TXA:
147 case BPF_LD | BPF_MEM:
148 case BPF_LDX | BPF_MEM:
149 case BPF_ST:
150 case BPF_STX:
151 case BPF_JMP | BPF_JA:
152 case BPF_JMP | BPF_JEQ | BPF_K:
153 case BPF_JMP | BPF_JEQ | BPF_X:
154 case BPF_JMP | BPF_JGE | BPF_K:
155 case BPF_JMP | BPF_JGE | BPF_X:
156 case BPF_JMP | BPF_JGT | BPF_K:
157 case BPF_JMP | BPF_JGT | BPF_X:
158 case BPF_JMP | BPF_JSET | BPF_K:
159 case BPF_JMP | BPF_JSET | BPF_X:
160 continue;
161 default:
162 return -EINVAL;
163 }
164 }
165 return 0;
166 }
167
168 /**
169 * seccomp_run_filters - evaluates all seccomp filters against @syscall
170 * @syscall: number of the current system call
171 *
172 * Returns valid seccomp BPF response codes.
173 */
174 static u32 seccomp_run_filters(int syscall)
175 {
176 struct seccomp_filter *f;
177 struct seccomp_data sd;
178 u32 ret = SECCOMP_RET_ALLOW;
179
180 /* Ensure unexpected behavior doesn't result in failing open. */
181 if (WARN_ON(current->seccomp.filter == NULL))
182 return SECCOMP_RET_KILL;
183
184 populate_seccomp_data(&sd);
185
186 /*
187 * All filters in the list are evaluated and the lowest BPF return
188 * value always takes priority (ignoring the DATA).
189 */
190 for (f = current->seccomp.filter; f; f = f->prev) {
191 u32 cur_ret = SK_RUN_FILTER(f->prog, (void *)&sd);
192
193 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
194 ret = cur_ret;
195 }
196 return ret;
197 }
198 #endif /* CONFIG_SECCOMP_FILTER */
199
200 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
201 {
202 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
203 return false;
204
205 return true;
206 }
207
208 static inline void seccomp_assign_mode(unsigned long seccomp_mode)
209 {
210 current->seccomp.mode = seccomp_mode;
211 set_tsk_thread_flag(current, TIF_SECCOMP);
212 }
213
214 #ifdef CONFIG_SECCOMP_FILTER
215 /**
216 * seccomp_prepare_filter: Prepares a seccomp filter for use.
217 * @fprog: BPF program to install
218 *
219 * Returns filter on success or an ERR_PTR on failure.
220 */
221 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
222 {
223 struct seccomp_filter *filter;
224 unsigned long fp_size;
225 struct sock_filter *fp;
226 int new_len;
227 long ret;
228
229 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
230 return ERR_PTR(-EINVAL);
231 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
232 fp_size = fprog->len * sizeof(struct sock_filter);
233
234 /*
235 * Installing a seccomp filter requires that the task has
236 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
237 * This avoids scenarios where unprivileged tasks can affect the
238 * behavior of privileged children.
239 */
240 if (!task_no_new_privs(current) &&
241 security_capable_noaudit(current_cred(), current_user_ns(),
242 CAP_SYS_ADMIN) != 0)
243 return ERR_PTR(-EACCES);
244
245 fp = kzalloc(fp_size, GFP_KERNEL|__GFP_NOWARN);
246 if (!fp)
247 return ERR_PTR(-ENOMEM);
248
249 /* Copy the instructions from fprog. */
250 ret = -EFAULT;
251 if (copy_from_user(fp, fprog->filter, fp_size))
252 goto free_prog;
253
254 /* Check and rewrite the fprog via the skb checker */
255 ret = sk_chk_filter(fp, fprog->len);
256 if (ret)
257 goto free_prog;
258
259 /* Check and rewrite the fprog for seccomp use */
260 ret = seccomp_check_filter(fp, fprog->len);
261 if (ret)
262 goto free_prog;
263
264 /* Convert 'sock_filter' insns to 'sock_filter_int' insns */
265 ret = sk_convert_filter(fp, fprog->len, NULL, &new_len);
266 if (ret)
267 goto free_prog;
268
269 /* Allocate a new seccomp_filter */
270 ret = -ENOMEM;
271 filter = kzalloc(sizeof(struct seccomp_filter),
272 GFP_KERNEL|__GFP_NOWARN);
273 if (!filter)
274 goto free_prog;
275
276 filter->prog = kzalloc(sk_filter_size(new_len),
277 GFP_KERNEL|__GFP_NOWARN);
278 if (!filter->prog)
279 goto free_filter;
280
281 ret = sk_convert_filter(fp, fprog->len, filter->prog->insnsi, &new_len);
282 if (ret)
283 goto free_filter_prog;
284 kfree(fp);
285
286 atomic_set(&filter->usage, 1);
287 filter->prog->len = new_len;
288
289 sk_filter_select_runtime(filter->prog);
290
291 return filter;
292
293 free_filter_prog:
294 kfree(filter->prog);
295 free_filter:
296 kfree(filter);
297 free_prog:
298 kfree(fp);
299 return ERR_PTR(ret);
300 }
301
302 /**
303 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
304 * @user_filter: pointer to the user data containing a sock_fprog.
305 *
306 * Returns 0 on success and non-zero otherwise.
307 */
308 static struct seccomp_filter *
309 seccomp_prepare_user_filter(const char __user *user_filter)
310 {
311 struct sock_fprog fprog;
312 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
313
314 #ifdef CONFIG_COMPAT
315 if (is_compat_task()) {
316 struct compat_sock_fprog fprog32;
317 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
318 goto out;
319 fprog.len = fprog32.len;
320 fprog.filter = compat_ptr(fprog32.filter);
321 } else /* falls through to the if below. */
322 #endif
323 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
324 goto out;
325 filter = seccomp_prepare_filter(&fprog);
326 out:
327 return filter;
328 }
329
330 /**
331 * seccomp_attach_filter: validate and attach filter
332 * @flags: flags to change filter behavior
333 * @filter: seccomp filter to add to the current process
334 *
335 * Returns 0 on success, -ve on error.
336 */
337 static long seccomp_attach_filter(unsigned int flags,
338 struct seccomp_filter *filter)
339 {
340 unsigned long total_insns;
341 struct seccomp_filter *walker;
342
343 /* Validate resulting filter length. */
344 total_insns = filter->prog->len;
345 for (walker = current->seccomp.filter; walker; walker = walker->prev)
346 total_insns += walker->prog->len + 4; /* 4 instr penalty */
347 if (total_insns > MAX_INSNS_PER_PATH)
348 return -ENOMEM;
349
350 /*
351 * If there is an existing filter, make it the prev and don't drop its
352 * task reference.
353 */
354 filter->prev = current->seccomp.filter;
355 current->seccomp.filter = filter;
356
357 return 0;
358 }
359
360 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
361 void get_seccomp_filter(struct task_struct *tsk)
362 {
363 struct seccomp_filter *orig = tsk->seccomp.filter;
364 if (!orig)
365 return;
366 /* Reference count is bounded by the number of total processes. */
367 atomic_inc(&orig->usage);
368 }
369
370 static inline void seccomp_filter_free(struct seccomp_filter *filter)
371 {
372 if (filter) {
373 sk_filter_free(filter->prog);
374 kfree(filter);
375 }
376 }
377
378 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
379 void put_seccomp_filter(struct task_struct *tsk)
380 {
381 struct seccomp_filter *orig = tsk->seccomp.filter;
382 /* Clean up single-reference branches iteratively. */
383 while (orig && atomic_dec_and_test(&orig->usage)) {
384 struct seccomp_filter *freeme = orig;
385 orig = orig->prev;
386 seccomp_filter_free(freeme);
387 }
388 }
389
390 /**
391 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
392 * @syscall: syscall number to send to userland
393 * @reason: filter-supplied reason code to send to userland (via si_errno)
394 *
395 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
396 */
397 static void seccomp_send_sigsys(int syscall, int reason)
398 {
399 struct siginfo info;
400 memset(&info, 0, sizeof(info));
401 info.si_signo = SIGSYS;
402 info.si_code = SYS_SECCOMP;
403 info.si_call_addr = (void __user *)KSTK_EIP(current);
404 info.si_errno = reason;
405 info.si_arch = syscall_get_arch();
406 info.si_syscall = syscall;
407 force_sig_info(SIGSYS, &info, current);
408 }
409 #endif /* CONFIG_SECCOMP_FILTER */
410
411 /*
412 * Secure computing mode 1 allows only read/write/exit/sigreturn.
413 * To be fully secure this must be combined with rlimit
414 * to limit the stack allocations too.
415 */
416 static int mode1_syscalls[] = {
417 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
418 0, /* null terminated */
419 };
420
421 #ifdef CONFIG_COMPAT
422 static int mode1_syscalls_32[] = {
423 __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
424 0, /* null terminated */
425 };
426 #endif
427
428 int __secure_computing(int this_syscall)
429 {
430 int mode = current->seccomp.mode;
431 int exit_sig = 0;
432 int *syscall;
433 u32 ret;
434
435 switch (mode) {
436 case SECCOMP_MODE_STRICT:
437 syscall = mode1_syscalls;
438 #ifdef CONFIG_COMPAT
439 if (is_compat_task())
440 syscall = mode1_syscalls_32;
441 #endif
442 do {
443 if (*syscall == this_syscall)
444 return 0;
445 } while (*++syscall);
446 exit_sig = SIGKILL;
447 ret = SECCOMP_RET_KILL;
448 break;
449 #ifdef CONFIG_SECCOMP_FILTER
450 case SECCOMP_MODE_FILTER: {
451 int data;
452 struct pt_regs *regs = task_pt_regs(current);
453 ret = seccomp_run_filters(this_syscall);
454 data = ret & SECCOMP_RET_DATA;
455 ret &= SECCOMP_RET_ACTION;
456 switch (ret) {
457 case SECCOMP_RET_ERRNO:
458 /* Set the low-order 16-bits as a errno. */
459 syscall_set_return_value(current, regs,
460 -data, 0);
461 goto skip;
462 case SECCOMP_RET_TRAP:
463 /* Show the handler the original registers. */
464 syscall_rollback(current, regs);
465 /* Let the filter pass back 16 bits of data. */
466 seccomp_send_sigsys(this_syscall, data);
467 goto skip;
468 case SECCOMP_RET_TRACE:
469 /* Skip these calls if there is no tracer. */
470 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
471 syscall_set_return_value(current, regs,
472 -ENOSYS, 0);
473 goto skip;
474 }
475 /* Allow the BPF to provide the event message */
476 ptrace_event(PTRACE_EVENT_SECCOMP, data);
477 /*
478 * The delivery of a fatal signal during event
479 * notification may silently skip tracer notification.
480 * Terminating the task now avoids executing a system
481 * call that may not be intended.
482 */
483 if (fatal_signal_pending(current))
484 break;
485 if (syscall_get_nr(current, regs) < 0)
486 goto skip; /* Explicit request to skip. */
487
488 return 0;
489 case SECCOMP_RET_ALLOW:
490 return 0;
491 case SECCOMP_RET_KILL:
492 default:
493 break;
494 }
495 exit_sig = SIGSYS;
496 break;
497 }
498 #endif
499 default:
500 BUG();
501 }
502
503 #ifdef SECCOMP_DEBUG
504 dump_stack();
505 #endif
506 audit_seccomp(this_syscall, exit_sig, ret);
507 do_exit(exit_sig);
508 #ifdef CONFIG_SECCOMP_FILTER
509 skip:
510 audit_seccomp(this_syscall, exit_sig, ret);
511 #endif
512 return -1;
513 }
514
515 long prctl_get_seccomp(void)
516 {
517 return current->seccomp.mode;
518 }
519
520 /**
521 * seccomp_set_mode_strict: internal function for setting strict seccomp
522 *
523 * Once current->seccomp.mode is non-zero, it may not be changed.
524 *
525 * Returns 0 on success or -EINVAL on failure.
526 */
527 static long seccomp_set_mode_strict(void)
528 {
529 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
530 long ret = -EINVAL;
531
532 if (!seccomp_may_assign_mode(seccomp_mode))
533 goto out;
534
535 #ifdef TIF_NOTSC
536 disable_TSC();
537 #endif
538 seccomp_assign_mode(seccomp_mode);
539 ret = 0;
540
541 out:
542
543 return ret;
544 }
545
546 #ifdef CONFIG_SECCOMP_FILTER
547 /**
548 * seccomp_set_mode_filter: internal function for setting seccomp filter
549 * @flags: flags to change filter behavior
550 * @filter: struct sock_fprog containing filter
551 *
552 * This function may be called repeatedly to install additional filters.
553 * Every filter successfully installed will be evaluated (in reverse order)
554 * for each system call the task makes.
555 *
556 * Once current->seccomp.mode is non-zero, it may not be changed.
557 *
558 * Returns 0 on success or -EINVAL on failure.
559 */
560 static long seccomp_set_mode_filter(unsigned int flags,
561 const char __user *filter)
562 {
563 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
564 struct seccomp_filter *prepared = NULL;
565 long ret = -EINVAL;
566
567 /* Validate flags. */
568 if (flags != 0)
569 goto out;
570
571 /* Prepare the new filter before holding any locks. */
572 prepared = seccomp_prepare_user_filter(filter);
573 if (IS_ERR(prepared))
574 return PTR_ERR(prepared);
575
576 if (!seccomp_may_assign_mode(seccomp_mode))
577 goto out;
578
579 ret = seccomp_attach_filter(flags, prepared);
580 if (ret)
581 goto out;
582 /* Do not free the successfully attached filter. */
583 prepared = NULL;
584
585 seccomp_assign_mode(seccomp_mode);
586 out:
587 seccomp_filter_free(prepared);
588 return ret;
589 }
590 #else
591 static inline long seccomp_set_mode_filter(unsigned int flags,
592 const char __user *filter)
593 {
594 return -EINVAL;
595 }
596 #endif
597
598 /* Common entry point for both prctl and syscall. */
599 static long do_seccomp(unsigned int op, unsigned int flags,
600 const char __user *uargs)
601 {
602 switch (op) {
603 case SECCOMP_SET_MODE_STRICT:
604 if (flags != 0 || uargs != NULL)
605 return -EINVAL;
606 return seccomp_set_mode_strict();
607 case SECCOMP_SET_MODE_FILTER:
608 return seccomp_set_mode_filter(flags, uargs);
609 default:
610 return -EINVAL;
611 }
612 }
613
614 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
615 const char __user *, uargs)
616 {
617 return do_seccomp(op, flags, uargs);
618 }
619
620 /**
621 * prctl_set_seccomp: configures current->seccomp.mode
622 * @seccomp_mode: requested mode to use
623 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
624 *
625 * Returns 0 on success or -EINVAL on failure.
626 */
627 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
628 {
629 unsigned int op;
630 char __user *uargs;
631
632 switch (seccomp_mode) {
633 case SECCOMP_MODE_STRICT:
634 op = SECCOMP_SET_MODE_STRICT;
635 /*
636 * Setting strict mode through prctl always ignored filter,
637 * so make sure it is always NULL here to pass the internal
638 * check in do_seccomp().
639 */
640 uargs = NULL;
641 break;
642 case SECCOMP_MODE_FILTER:
643 op = SECCOMP_SET_MODE_FILTER;
644 uargs = filter;
645 break;
646 default:
647 return -EINVAL;
648 }
649
650 /* prctl interface doesn't have flags, so they are always zero. */
651 return do_seccomp(op, 0, uargs);
652 }
This page took 0.063572 seconds and 6 git commands to generate.