Loading...
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#ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
25#include <asm/syscall.h>
26#endif
27
28#ifdef CONFIG_SECCOMP_FILTER
29#include <linux/filter.h>
30#include <linux/pid.h>
31#include <linux/ptrace.h>
32#include <linux/security.h>
33#include <linux/tracehook.h>
34#include <linux/uaccess.h>
35
36/**
37 * struct seccomp_filter - container for seccomp BPF programs
38 *
39 * @usage: reference count to manage the object lifetime.
40 * get/put helpers should be used when accessing an instance
41 * outside of a lifetime-guarded section. In general, this
42 * is only needed for handling filters shared across tasks.
43 * @prev: points to a previously installed, or inherited, filter
44 * @len: the number of instructions in the program
45 * @insnsi: the BPF program instructions to evaluate
46 *
47 * seccomp_filter objects are organized in a tree linked via the @prev
48 * pointer. For any task, it appears to be a singly-linked list starting
49 * with current->seccomp.filter, the most recently attached or inherited filter.
50 * However, multiple filters may share a @prev node, by way of fork(), which
51 * results in a unidirectional tree existing in memory. This is similar to
52 * how namespaces work.
53 *
54 * seccomp_filter objects should never be modified after being attached
55 * to a task_struct (other than @usage).
56 */
57struct seccomp_filter {
58 atomic_t usage;
59 struct seccomp_filter *prev;
60 struct bpf_prog *prog;
61};
62
63/* Limit any path through the tree to 256KB worth of instructions. */
64#define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
65
66/*
67 * Endianness is explicitly ignored and left for BPF program authors to manage
68 * as per the specific architecture.
69 */
70static void populate_seccomp_data(struct seccomp_data *sd)
71{
72 struct task_struct *task = current;
73 struct pt_regs *regs = task_pt_regs(task);
74 unsigned long args[6];
75
76 sd->nr = syscall_get_nr(task, regs);
77 sd->arch = syscall_get_arch();
78 syscall_get_arguments(task, regs, 0, 6, args);
79 sd->args[0] = args[0];
80 sd->args[1] = args[1];
81 sd->args[2] = args[2];
82 sd->args[3] = args[3];
83 sd->args[4] = args[4];
84 sd->args[5] = args[5];
85 sd->instruction_pointer = KSTK_EIP(task);
86}
87
88/**
89 * seccomp_check_filter - verify seccomp filter code
90 * @filter: filter to verify
91 * @flen: length of filter
92 *
93 * Takes a previously checked filter (by bpf_check_classic) and
94 * redirects all filter code that loads struct sk_buff data
95 * and related data through seccomp_bpf_load. It also
96 * enforces length and alignment checking of those loads.
97 *
98 * Returns 0 if the rule set is legal or -EINVAL if not.
99 */
100static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
101{
102 int pc;
103 for (pc = 0; pc < flen; pc++) {
104 struct sock_filter *ftest = &filter[pc];
105 u16 code = ftest->code;
106 u32 k = ftest->k;
107
108 switch (code) {
109 case BPF_LD | BPF_W | BPF_ABS:
110 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
111 /* 32-bit aligned and not out of bounds. */
112 if (k >= sizeof(struct seccomp_data) || k & 3)
113 return -EINVAL;
114 continue;
115 case BPF_LD | BPF_W | BPF_LEN:
116 ftest->code = BPF_LD | BPF_IMM;
117 ftest->k = sizeof(struct seccomp_data);
118 continue;
119 case BPF_LDX | BPF_W | BPF_LEN:
120 ftest->code = BPF_LDX | BPF_IMM;
121 ftest->k = sizeof(struct seccomp_data);
122 continue;
123 /* Explicitly include allowed calls. */
124 case BPF_RET | BPF_K:
125 case BPF_RET | BPF_A:
126 case BPF_ALU | BPF_ADD | BPF_K:
127 case BPF_ALU | BPF_ADD | BPF_X:
128 case BPF_ALU | BPF_SUB | BPF_K:
129 case BPF_ALU | BPF_SUB | BPF_X:
130 case BPF_ALU | BPF_MUL | BPF_K:
131 case BPF_ALU | BPF_MUL | BPF_X:
132 case BPF_ALU | BPF_DIV | BPF_K:
133 case BPF_ALU | BPF_DIV | BPF_X:
134 case BPF_ALU | BPF_AND | BPF_K:
135 case BPF_ALU | BPF_AND | BPF_X:
136 case BPF_ALU | BPF_OR | BPF_K:
137 case BPF_ALU | BPF_OR | BPF_X:
138 case BPF_ALU | BPF_XOR | BPF_K:
139 case BPF_ALU | BPF_XOR | BPF_X:
140 case BPF_ALU | BPF_LSH | BPF_K:
141 case BPF_ALU | BPF_LSH | BPF_X:
142 case BPF_ALU | BPF_RSH | BPF_K:
143 case BPF_ALU | BPF_RSH | BPF_X:
144 case BPF_ALU | BPF_NEG:
145 case BPF_LD | BPF_IMM:
146 case BPF_LDX | BPF_IMM:
147 case BPF_MISC | BPF_TAX:
148 case BPF_MISC | BPF_TXA:
149 case BPF_LD | BPF_MEM:
150 case BPF_LDX | BPF_MEM:
151 case BPF_ST:
152 case BPF_STX:
153 case BPF_JMP | BPF_JA:
154 case BPF_JMP | BPF_JEQ | BPF_K:
155 case BPF_JMP | BPF_JEQ | BPF_X:
156 case BPF_JMP | BPF_JGE | BPF_K:
157 case BPF_JMP | BPF_JGE | BPF_X:
158 case BPF_JMP | BPF_JGT | BPF_K:
159 case BPF_JMP | BPF_JGT | BPF_X:
160 case BPF_JMP | BPF_JSET | BPF_K:
161 case BPF_JMP | BPF_JSET | BPF_X:
162 continue;
163 default:
164 return -EINVAL;
165 }
166 }
167 return 0;
168}
169
170/**
171 * seccomp_run_filters - evaluates all seccomp filters against @syscall
172 * @syscall: number of the current system call
173 *
174 * Returns valid seccomp BPF response codes.
175 */
176static u32 seccomp_run_filters(struct seccomp_data *sd)
177{
178 struct seccomp_data sd_local;
179 u32 ret = SECCOMP_RET_ALLOW;
180 /* Make sure cross-thread synced filter points somewhere sane. */
181 struct seccomp_filter *f =
182 lockless_dereference(current->seccomp.filter);
183
184 /* Ensure unexpected behavior doesn't result in failing open. */
185 if (unlikely(WARN_ON(f == NULL)))
186 return SECCOMP_RET_KILL;
187
188 if (!sd) {
189 populate_seccomp_data(&sd_local);
190 sd = &sd_local;
191 }
192
193 /*
194 * All filters in the list are evaluated and the lowest BPF return
195 * value always takes priority (ignoring the DATA).
196 */
197 for (; f; f = f->prev) {
198 u32 cur_ret = BPF_PROG_RUN(f->prog, (void *)sd);
199
200 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
201 ret = cur_ret;
202 }
203 return ret;
204}
205#endif /* CONFIG_SECCOMP_FILTER */
206
207static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
208{
209 assert_spin_locked(¤t->sighand->siglock);
210
211 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
212 return false;
213
214 return true;
215}
216
217static inline void seccomp_assign_mode(struct task_struct *task,
218 unsigned long seccomp_mode)
219{
220 assert_spin_locked(&task->sighand->siglock);
221
222 task->seccomp.mode = seccomp_mode;
223 /*
224 * Make sure TIF_SECCOMP cannot be set before the mode (and
225 * filter) is set.
226 */
227 smp_mb__before_atomic();
228 set_tsk_thread_flag(task, TIF_SECCOMP);
229}
230
231#ifdef CONFIG_SECCOMP_FILTER
232/* Returns 1 if the parent is an ancestor of the child. */
233static int is_ancestor(struct seccomp_filter *parent,
234 struct seccomp_filter *child)
235{
236 /* NULL is the root ancestor. */
237 if (parent == NULL)
238 return 1;
239 for (; child; child = child->prev)
240 if (child == parent)
241 return 1;
242 return 0;
243}
244
245/**
246 * seccomp_can_sync_threads: checks if all threads can be synchronized
247 *
248 * Expects sighand and cred_guard_mutex locks to be held.
249 *
250 * Returns 0 on success, -ve on error, or the pid of a thread which was
251 * either not in the correct seccomp mode or it did not have an ancestral
252 * seccomp filter.
253 */
254static inline pid_t seccomp_can_sync_threads(void)
255{
256 struct task_struct *thread, *caller;
257
258 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
259 assert_spin_locked(¤t->sighand->siglock);
260
261 /* Validate all threads being eligible for synchronization. */
262 caller = current;
263 for_each_thread(caller, thread) {
264 pid_t failed;
265
266 /* Skip current, since it is initiating the sync. */
267 if (thread == caller)
268 continue;
269
270 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
271 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
272 is_ancestor(thread->seccomp.filter,
273 caller->seccomp.filter)))
274 continue;
275
276 /* Return the first thread that cannot be synchronized. */
277 failed = task_pid_vnr(thread);
278 /* If the pid cannot be resolved, then return -ESRCH */
279 if (unlikely(WARN_ON(failed == 0)))
280 failed = -ESRCH;
281 return failed;
282 }
283
284 return 0;
285}
286
287/**
288 * seccomp_sync_threads: sets all threads to use current's filter
289 *
290 * Expects sighand and cred_guard_mutex locks to be held, and for
291 * seccomp_can_sync_threads() to have returned success already
292 * without dropping the locks.
293 *
294 */
295static inline void seccomp_sync_threads(void)
296{
297 struct task_struct *thread, *caller;
298
299 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
300 assert_spin_locked(¤t->sighand->siglock);
301
302 /* Synchronize all threads. */
303 caller = current;
304 for_each_thread(caller, thread) {
305 /* Skip current, since it needs no changes. */
306 if (thread == caller)
307 continue;
308
309 /* Get a task reference for the new leaf node. */
310 get_seccomp_filter(caller);
311 /*
312 * Drop the task reference to the shared ancestor since
313 * current's path will hold a reference. (This also
314 * allows a put before the assignment.)
315 */
316 put_seccomp_filter(thread);
317 smp_store_release(&thread->seccomp.filter,
318 caller->seccomp.filter);
319
320 /*
321 * Don't let an unprivileged task work around
322 * the no_new_privs restriction by creating
323 * a thread that sets it up, enters seccomp,
324 * then dies.
325 */
326 if (task_no_new_privs(caller))
327 task_set_no_new_privs(thread);
328
329 /*
330 * Opt the other thread into seccomp if needed.
331 * As threads are considered to be trust-realm
332 * equivalent (see ptrace_may_access), it is safe to
333 * allow one thread to transition the other.
334 */
335 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
336 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER);
337 }
338}
339
340/**
341 * seccomp_prepare_filter: Prepares a seccomp filter for use.
342 * @fprog: BPF program to install
343 *
344 * Returns filter on success or an ERR_PTR on failure.
345 */
346static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
347{
348 struct seccomp_filter *sfilter;
349 int ret;
350 const bool save_orig = config_enabled(CONFIG_CHECKPOINT_RESTORE);
351
352 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
353 return ERR_PTR(-EINVAL);
354
355 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
356
357 /*
358 * Installing a seccomp filter requires that the task has
359 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
360 * This avoids scenarios where unprivileged tasks can affect the
361 * behavior of privileged children.
362 */
363 if (!task_no_new_privs(current) &&
364 security_capable_noaudit(current_cred(), current_user_ns(),
365 CAP_SYS_ADMIN) != 0)
366 return ERR_PTR(-EACCES);
367
368 /* Allocate a new seccomp_filter */
369 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
370 if (!sfilter)
371 return ERR_PTR(-ENOMEM);
372
373 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
374 seccomp_check_filter, save_orig);
375 if (ret < 0) {
376 kfree(sfilter);
377 return ERR_PTR(ret);
378 }
379
380 atomic_set(&sfilter->usage, 1);
381
382 return sfilter;
383}
384
385/**
386 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
387 * @user_filter: pointer to the user data containing a sock_fprog.
388 *
389 * Returns 0 on success and non-zero otherwise.
390 */
391static struct seccomp_filter *
392seccomp_prepare_user_filter(const char __user *user_filter)
393{
394 struct sock_fprog fprog;
395 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
396
397#ifdef CONFIG_COMPAT
398 if (in_compat_syscall()) {
399 struct compat_sock_fprog fprog32;
400 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
401 goto out;
402 fprog.len = fprog32.len;
403 fprog.filter = compat_ptr(fprog32.filter);
404 } else /* falls through to the if below. */
405#endif
406 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
407 goto out;
408 filter = seccomp_prepare_filter(&fprog);
409out:
410 return filter;
411}
412
413/**
414 * seccomp_attach_filter: validate and attach filter
415 * @flags: flags to change filter behavior
416 * @filter: seccomp filter to add to the current process
417 *
418 * Caller must be holding current->sighand->siglock lock.
419 *
420 * Returns 0 on success, -ve on error.
421 */
422static long seccomp_attach_filter(unsigned int flags,
423 struct seccomp_filter *filter)
424{
425 unsigned long total_insns;
426 struct seccomp_filter *walker;
427
428 assert_spin_locked(¤t->sighand->siglock);
429
430 /* Validate resulting filter length. */
431 total_insns = filter->prog->len;
432 for (walker = current->seccomp.filter; walker; walker = walker->prev)
433 total_insns += walker->prog->len + 4; /* 4 instr penalty */
434 if (total_insns > MAX_INSNS_PER_PATH)
435 return -ENOMEM;
436
437 /* If thread sync has been requested, check that it is possible. */
438 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
439 int ret;
440
441 ret = seccomp_can_sync_threads();
442 if (ret)
443 return ret;
444 }
445
446 /*
447 * If there is an existing filter, make it the prev and don't drop its
448 * task reference.
449 */
450 filter->prev = current->seccomp.filter;
451 current->seccomp.filter = filter;
452
453 /* Now that the new filter is in place, synchronize to all threads. */
454 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
455 seccomp_sync_threads();
456
457 return 0;
458}
459
460/* get_seccomp_filter - increments the reference count of the filter on @tsk */
461void get_seccomp_filter(struct task_struct *tsk)
462{
463 struct seccomp_filter *orig = tsk->seccomp.filter;
464 if (!orig)
465 return;
466 /* Reference count is bounded by the number of total processes. */
467 atomic_inc(&orig->usage);
468}
469
470static inline void seccomp_filter_free(struct seccomp_filter *filter)
471{
472 if (filter) {
473 bpf_prog_destroy(filter->prog);
474 kfree(filter);
475 }
476}
477
478/* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
479void put_seccomp_filter(struct task_struct *tsk)
480{
481 struct seccomp_filter *orig = tsk->seccomp.filter;
482 /* Clean up single-reference branches iteratively. */
483 while (orig && atomic_dec_and_test(&orig->usage)) {
484 struct seccomp_filter *freeme = orig;
485 orig = orig->prev;
486 seccomp_filter_free(freeme);
487 }
488}
489
490/**
491 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
492 * @syscall: syscall number to send to userland
493 * @reason: filter-supplied reason code to send to userland (via si_errno)
494 *
495 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
496 */
497static void seccomp_send_sigsys(int syscall, int reason)
498{
499 struct siginfo info;
500 memset(&info, 0, sizeof(info));
501 info.si_signo = SIGSYS;
502 info.si_code = SYS_SECCOMP;
503 info.si_call_addr = (void __user *)KSTK_EIP(current);
504 info.si_errno = reason;
505 info.si_arch = syscall_get_arch();
506 info.si_syscall = syscall;
507 force_sig_info(SIGSYS, &info, current);
508}
509#endif /* CONFIG_SECCOMP_FILTER */
510
511/*
512 * Secure computing mode 1 allows only read/write/exit/sigreturn.
513 * To be fully secure this must be combined with rlimit
514 * to limit the stack allocations too.
515 */
516static int mode1_syscalls[] = {
517 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
518 0, /* null terminated */
519};
520
521#ifdef CONFIG_COMPAT
522static int mode1_syscalls_32[] = {
523 __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
524 0, /* null terminated */
525};
526#endif
527
528static void __secure_computing_strict(int this_syscall)
529{
530 int *syscall_whitelist = mode1_syscalls;
531#ifdef CONFIG_COMPAT
532 if (in_compat_syscall())
533 syscall_whitelist = mode1_syscalls_32;
534#endif
535 do {
536 if (*syscall_whitelist == this_syscall)
537 return;
538 } while (*++syscall_whitelist);
539
540#ifdef SECCOMP_DEBUG
541 dump_stack();
542#endif
543 audit_seccomp(this_syscall, SIGKILL, SECCOMP_RET_KILL);
544 do_exit(SIGKILL);
545}
546
547#ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
548void secure_computing_strict(int this_syscall)
549{
550 int mode = current->seccomp.mode;
551
552 if (config_enabled(CONFIG_CHECKPOINT_RESTORE) &&
553 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
554 return;
555
556 if (mode == SECCOMP_MODE_DISABLED)
557 return;
558 else if (mode == SECCOMP_MODE_STRICT)
559 __secure_computing_strict(this_syscall);
560 else
561 BUG();
562}
563#else
564int __secure_computing(void)
565{
566 u32 phase1_result = seccomp_phase1(NULL);
567
568 if (likely(phase1_result == SECCOMP_PHASE1_OK))
569 return 0;
570 else if (likely(phase1_result == SECCOMP_PHASE1_SKIP))
571 return -1;
572 else
573 return seccomp_phase2(phase1_result);
574}
575
576#ifdef CONFIG_SECCOMP_FILTER
577static u32 __seccomp_phase1_filter(int this_syscall, struct seccomp_data *sd)
578{
579 u32 filter_ret, action;
580 int data;
581
582 /*
583 * Make sure that any changes to mode from another thread have
584 * been seen after TIF_SECCOMP was seen.
585 */
586 rmb();
587
588 filter_ret = seccomp_run_filters(sd);
589 data = filter_ret & SECCOMP_RET_DATA;
590 action = filter_ret & SECCOMP_RET_ACTION;
591
592 switch (action) {
593 case SECCOMP_RET_ERRNO:
594 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
595 if (data > MAX_ERRNO)
596 data = MAX_ERRNO;
597 syscall_set_return_value(current, task_pt_regs(current),
598 -data, 0);
599 goto skip;
600
601 case SECCOMP_RET_TRAP:
602 /* Show the handler the original registers. */
603 syscall_rollback(current, task_pt_regs(current));
604 /* Let the filter pass back 16 bits of data. */
605 seccomp_send_sigsys(this_syscall, data);
606 goto skip;
607
608 case SECCOMP_RET_TRACE:
609 return filter_ret; /* Save the rest for phase 2. */
610
611 case SECCOMP_RET_ALLOW:
612 return SECCOMP_PHASE1_OK;
613
614 case SECCOMP_RET_KILL:
615 default:
616 audit_seccomp(this_syscall, SIGSYS, action);
617 do_exit(SIGSYS);
618 }
619
620 unreachable();
621
622skip:
623 audit_seccomp(this_syscall, 0, action);
624 return SECCOMP_PHASE1_SKIP;
625}
626#endif
627
628/**
629 * seccomp_phase1() - run fast path seccomp checks on the current syscall
630 * @arg sd: The seccomp_data or NULL
631 *
632 * This only reads pt_regs via the syscall_xyz helpers. The only change
633 * it will make to pt_regs is via syscall_set_return_value, and it will
634 * only do that if it returns SECCOMP_PHASE1_SKIP.
635 *
636 * If sd is provided, it will not read pt_regs at all.
637 *
638 * It may also call do_exit or force a signal; these actions must be
639 * safe.
640 *
641 * If it returns SECCOMP_PHASE1_OK, the syscall passes checks and should
642 * be processed normally.
643 *
644 * If it returns SECCOMP_PHASE1_SKIP, then the syscall should not be
645 * invoked. In this case, seccomp_phase1 will have set the return value
646 * using syscall_set_return_value.
647 *
648 * If it returns anything else, then the return value should be passed
649 * to seccomp_phase2 from a context in which ptrace hooks are safe.
650 */
651u32 seccomp_phase1(struct seccomp_data *sd)
652{
653 int mode = current->seccomp.mode;
654 int this_syscall = sd ? sd->nr :
655 syscall_get_nr(current, task_pt_regs(current));
656
657 if (config_enabled(CONFIG_CHECKPOINT_RESTORE) &&
658 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
659 return SECCOMP_PHASE1_OK;
660
661 switch (mode) {
662 case SECCOMP_MODE_STRICT:
663 __secure_computing_strict(this_syscall); /* may call do_exit */
664 return SECCOMP_PHASE1_OK;
665#ifdef CONFIG_SECCOMP_FILTER
666 case SECCOMP_MODE_FILTER:
667 return __seccomp_phase1_filter(this_syscall, sd);
668#endif
669 default:
670 BUG();
671 }
672}
673
674/**
675 * seccomp_phase2() - finish slow path seccomp work for the current syscall
676 * @phase1_result: The return value from seccomp_phase1()
677 *
678 * This must be called from a context in which ptrace hooks can be used.
679 *
680 * Returns 0 if the syscall should be processed or -1 to skip the syscall.
681 */
682int seccomp_phase2(u32 phase1_result)
683{
684 struct pt_regs *regs = task_pt_regs(current);
685 u32 action = phase1_result & SECCOMP_RET_ACTION;
686 int data = phase1_result & SECCOMP_RET_DATA;
687
688 BUG_ON(action != SECCOMP_RET_TRACE);
689
690 audit_seccomp(syscall_get_nr(current, regs), 0, action);
691
692 /* Skip these calls if there is no tracer. */
693 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
694 syscall_set_return_value(current, regs,
695 -ENOSYS, 0);
696 return -1;
697 }
698
699 /* Allow the BPF to provide the event message */
700 ptrace_event(PTRACE_EVENT_SECCOMP, data);
701 /*
702 * The delivery of a fatal signal during event
703 * notification may silently skip tracer notification.
704 * Terminating the task now avoids executing a system
705 * call that may not be intended.
706 */
707 if (fatal_signal_pending(current))
708 do_exit(SIGSYS);
709 if (syscall_get_nr(current, regs) < 0)
710 return -1; /* Explicit request to skip. */
711
712 return 0;
713}
714#endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
715
716long prctl_get_seccomp(void)
717{
718 return current->seccomp.mode;
719}
720
721/**
722 * seccomp_set_mode_strict: internal function for setting strict seccomp
723 *
724 * Once current->seccomp.mode is non-zero, it may not be changed.
725 *
726 * Returns 0 on success or -EINVAL on failure.
727 */
728static long seccomp_set_mode_strict(void)
729{
730 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
731 long ret = -EINVAL;
732
733 spin_lock_irq(¤t->sighand->siglock);
734
735 if (!seccomp_may_assign_mode(seccomp_mode))
736 goto out;
737
738#ifdef TIF_NOTSC
739 disable_TSC();
740#endif
741 seccomp_assign_mode(current, seccomp_mode);
742 ret = 0;
743
744out:
745 spin_unlock_irq(¤t->sighand->siglock);
746
747 return ret;
748}
749
750#ifdef CONFIG_SECCOMP_FILTER
751/**
752 * seccomp_set_mode_filter: internal function for setting seccomp filter
753 * @flags: flags to change filter behavior
754 * @filter: struct sock_fprog containing filter
755 *
756 * This function may be called repeatedly to install additional filters.
757 * Every filter successfully installed will be evaluated (in reverse order)
758 * for each system call the task makes.
759 *
760 * Once current->seccomp.mode is non-zero, it may not be changed.
761 *
762 * Returns 0 on success or -EINVAL on failure.
763 */
764static long seccomp_set_mode_filter(unsigned int flags,
765 const char __user *filter)
766{
767 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
768 struct seccomp_filter *prepared = NULL;
769 long ret = -EINVAL;
770
771 /* Validate flags. */
772 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
773 return -EINVAL;
774
775 /* Prepare the new filter before holding any locks. */
776 prepared = seccomp_prepare_user_filter(filter);
777 if (IS_ERR(prepared))
778 return PTR_ERR(prepared);
779
780 /*
781 * Make sure we cannot change seccomp or nnp state via TSYNC
782 * while another thread is in the middle of calling exec.
783 */
784 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
785 mutex_lock_killable(¤t->signal->cred_guard_mutex))
786 goto out_free;
787
788 spin_lock_irq(¤t->sighand->siglock);
789
790 if (!seccomp_may_assign_mode(seccomp_mode))
791 goto out;
792
793 ret = seccomp_attach_filter(flags, prepared);
794 if (ret)
795 goto out;
796 /* Do not free the successfully attached filter. */
797 prepared = NULL;
798
799 seccomp_assign_mode(current, seccomp_mode);
800out:
801 spin_unlock_irq(¤t->sighand->siglock);
802 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
803 mutex_unlock(¤t->signal->cred_guard_mutex);
804out_free:
805 seccomp_filter_free(prepared);
806 return ret;
807}
808#else
809static inline long seccomp_set_mode_filter(unsigned int flags,
810 const char __user *filter)
811{
812 return -EINVAL;
813}
814#endif
815
816/* Common entry point for both prctl and syscall. */
817static long do_seccomp(unsigned int op, unsigned int flags,
818 const char __user *uargs)
819{
820 switch (op) {
821 case SECCOMP_SET_MODE_STRICT:
822 if (flags != 0 || uargs != NULL)
823 return -EINVAL;
824 return seccomp_set_mode_strict();
825 case SECCOMP_SET_MODE_FILTER:
826 return seccomp_set_mode_filter(flags, uargs);
827 default:
828 return -EINVAL;
829 }
830}
831
832SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
833 const char __user *, uargs)
834{
835 return do_seccomp(op, flags, uargs);
836}
837
838/**
839 * prctl_set_seccomp: configures current->seccomp.mode
840 * @seccomp_mode: requested mode to use
841 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
842 *
843 * Returns 0 on success or -EINVAL on failure.
844 */
845long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
846{
847 unsigned int op;
848 char __user *uargs;
849
850 switch (seccomp_mode) {
851 case SECCOMP_MODE_STRICT:
852 op = SECCOMP_SET_MODE_STRICT;
853 /*
854 * Setting strict mode through prctl always ignored filter,
855 * so make sure it is always NULL here to pass the internal
856 * check in do_seccomp().
857 */
858 uargs = NULL;
859 break;
860 case SECCOMP_MODE_FILTER:
861 op = SECCOMP_SET_MODE_FILTER;
862 uargs = filter;
863 break;
864 default:
865 return -EINVAL;
866 }
867
868 /* prctl interface doesn't have flags, so they are always zero. */
869 return do_seccomp(op, 0, uargs);
870}
871
872#if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
873long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
874 void __user *data)
875{
876 struct seccomp_filter *filter;
877 struct sock_fprog_kern *fprog;
878 long ret;
879 unsigned long count = 0;
880
881 if (!capable(CAP_SYS_ADMIN) ||
882 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
883 return -EACCES;
884 }
885
886 spin_lock_irq(&task->sighand->siglock);
887 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
888 ret = -EINVAL;
889 goto out;
890 }
891
892 filter = task->seccomp.filter;
893 while (filter) {
894 filter = filter->prev;
895 count++;
896 }
897
898 if (filter_off >= count) {
899 ret = -ENOENT;
900 goto out;
901 }
902 count -= filter_off;
903
904 filter = task->seccomp.filter;
905 while (filter && count > 1) {
906 filter = filter->prev;
907 count--;
908 }
909
910 if (WARN_ON(count != 1 || !filter)) {
911 /* The filter tree shouldn't shrink while we're using it. */
912 ret = -ENOENT;
913 goto out;
914 }
915
916 fprog = filter->prog->orig_prog;
917 if (!fprog) {
918 /* This must be a new non-cBPF filter, since we save every
919 * every cBPF filter's orig_prog above when
920 * CONFIG_CHECKPOINT_RESTORE is enabled.
921 */
922 ret = -EMEDIUMTYPE;
923 goto out;
924 }
925
926 ret = fprog->len;
927 if (!data)
928 goto out;
929
930 get_seccomp_filter(task);
931 spin_unlock_irq(&task->sighand->siglock);
932
933 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
934 ret = -EFAULT;
935
936 put_seccomp_filter(task);
937 return ret;
938
939out:
940 spin_unlock_irq(&task->sighand->siglock);
941 return ret;
942}
943#endif
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#ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
33#include <asm/syscall.h>
34#endif
35
36#ifdef CONFIG_SECCOMP_FILTER
37#include <linux/file.h>
38#include <linux/filter.h>
39#include <linux/pid.h>
40#include <linux/ptrace.h>
41#include <linux/security.h>
42#include <linux/tracehook.h>
43#include <linux/uaccess.h>
44#include <linux/anon_inodes.h>
45#include <linux/lockdep.h>
46
47/*
48 * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the
49 * wrong direction flag in the ioctl number. This is the broken one,
50 * which the kernel needs to keep supporting until all userspaces stop
51 * using the wrong command number.
52 */
53#define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR SECCOMP_IOR(2, __u64)
54
55enum notify_state {
56 SECCOMP_NOTIFY_INIT,
57 SECCOMP_NOTIFY_SENT,
58 SECCOMP_NOTIFY_REPLIED,
59};
60
61struct seccomp_knotif {
62 /* The struct pid of the task whose filter triggered the notification */
63 struct task_struct *task;
64
65 /* The "cookie" for this request; this is unique for this filter. */
66 u64 id;
67
68 /*
69 * The seccomp data. This pointer is valid the entire time this
70 * notification is active, since it comes from __seccomp_filter which
71 * eclipses the entire lifecycle here.
72 */
73 const struct seccomp_data *data;
74
75 /*
76 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
77 * struct seccomp_knotif is created and starts out in INIT. Once the
78 * handler reads the notification off of an FD, it transitions to SENT.
79 * If a signal is received the state transitions back to INIT and
80 * another message is sent. When the userspace handler replies, state
81 * transitions to REPLIED.
82 */
83 enum notify_state state;
84
85 /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
86 int error;
87 long val;
88 u32 flags;
89
90 /*
91 * Signals when this has changed states, such as the listener
92 * dying, a new seccomp addfd message, or changing to REPLIED
93 */
94 struct completion ready;
95
96 struct list_head list;
97
98 /* outstanding addfd requests */
99 struct list_head addfd;
100};
101
102/**
103 * struct seccomp_kaddfd - container for seccomp_addfd ioctl messages
104 *
105 * @file: A reference to the file to install in the other task
106 * @fd: The fd number to install it at. If the fd number is -1, it means the
107 * installing process should allocate the fd as normal.
108 * @flags: The flags for the new file descriptor. At the moment, only O_CLOEXEC
109 * is allowed.
110 * @ret: The return value of the installing process. It is set to the fd num
111 * upon success (>= 0).
112 * @completion: Indicates that the installing process has completed fd
113 * installation, or gone away (either due to successful
114 * reply, or signal)
115 *
116 */
117struct seccomp_kaddfd {
118 struct file *file;
119 int fd;
120 unsigned int flags;
121
122 /* To only be set on reply */
123 int ret;
124 struct completion completion;
125 struct list_head list;
126};
127
128/**
129 * struct notification - container for seccomp userspace notifications. Since
130 * most seccomp filters will not have notification listeners attached and this
131 * structure is fairly large, we store the notification-specific stuff in a
132 * separate structure.
133 *
134 * @request: A semaphore that users of this notification can wait on for
135 * changes. Actual reads and writes are still controlled with
136 * filter->notify_lock.
137 * @next_id: The id of the next request.
138 * @notifications: A list of struct seccomp_knotif elements.
139 */
140struct notification {
141 struct semaphore request;
142 u64 next_id;
143 struct list_head notifications;
144};
145
146/**
147 * struct seccomp_filter - container for seccomp BPF programs
148 *
149 * @refs: Reference count to manage the object lifetime.
150 * A filter's reference count is incremented for each directly
151 * attached task, once for the dependent filter, and if
152 * requested for the user notifier. When @refs reaches zero,
153 * the filter can be freed.
154 * @users: A filter's @users count is incremented for each directly
155 * attached task (filter installation, fork(), thread_sync),
156 * and once for the dependent filter (tracked in filter->prev).
157 * When it reaches zero it indicates that no direct or indirect
158 * users of that filter exist. No new tasks can get associated with
159 * this filter after reaching 0. The @users count is always smaller
160 * or equal to @refs. Hence, reaching 0 for @users does not mean
161 * the filter can be freed.
162 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
163 * @prev: points to a previously installed, or inherited, filter
164 * @prog: the BPF program to evaluate
165 * @notif: the struct that holds all notification related information
166 * @notify_lock: A lock for all notification-related accesses.
167 * @wqh: A wait queue for poll if a notifier is in use.
168 *
169 * seccomp_filter objects are organized in a tree linked via the @prev
170 * pointer. For any task, it appears to be a singly-linked list starting
171 * with current->seccomp.filter, the most recently attached or inherited filter.
172 * However, multiple filters may share a @prev node, by way of fork(), which
173 * results in a unidirectional tree existing in memory. This is similar to
174 * how namespaces work.
175 *
176 * seccomp_filter objects should never be modified after being attached
177 * to a task_struct (other than @refs).
178 */
179struct seccomp_filter {
180 refcount_t refs;
181 refcount_t users;
182 bool log;
183 struct seccomp_filter *prev;
184 struct bpf_prog *prog;
185 struct notification *notif;
186 struct mutex notify_lock;
187 wait_queue_head_t wqh;
188};
189
190/* Limit any path through the tree to 256KB worth of instructions. */
191#define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
192
193/*
194 * Endianness is explicitly ignored and left for BPF program authors to manage
195 * as per the specific architecture.
196 */
197static void populate_seccomp_data(struct seccomp_data *sd)
198{
199 struct task_struct *task = current;
200 struct pt_regs *regs = task_pt_regs(task);
201 unsigned long args[6];
202
203 sd->nr = syscall_get_nr(task, regs);
204 sd->arch = syscall_get_arch(task);
205 syscall_get_arguments(task, regs, args);
206 sd->args[0] = args[0];
207 sd->args[1] = args[1];
208 sd->args[2] = args[2];
209 sd->args[3] = args[3];
210 sd->args[4] = args[4];
211 sd->args[5] = args[5];
212 sd->instruction_pointer = KSTK_EIP(task);
213}
214
215/**
216 * seccomp_check_filter - verify seccomp filter code
217 * @filter: filter to verify
218 * @flen: length of filter
219 *
220 * Takes a previously checked filter (by bpf_check_classic) and
221 * redirects all filter code that loads struct sk_buff data
222 * and related data through seccomp_bpf_load. It also
223 * enforces length and alignment checking of those loads.
224 *
225 * Returns 0 if the rule set is legal or -EINVAL if not.
226 */
227static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
228{
229 int pc;
230 for (pc = 0; pc < flen; pc++) {
231 struct sock_filter *ftest = &filter[pc];
232 u16 code = ftest->code;
233 u32 k = ftest->k;
234
235 switch (code) {
236 case BPF_LD | BPF_W | BPF_ABS:
237 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
238 /* 32-bit aligned and not out of bounds. */
239 if (k >= sizeof(struct seccomp_data) || k & 3)
240 return -EINVAL;
241 continue;
242 case BPF_LD | BPF_W | BPF_LEN:
243 ftest->code = BPF_LD | BPF_IMM;
244 ftest->k = sizeof(struct seccomp_data);
245 continue;
246 case BPF_LDX | BPF_W | BPF_LEN:
247 ftest->code = BPF_LDX | BPF_IMM;
248 ftest->k = sizeof(struct seccomp_data);
249 continue;
250 /* Explicitly include allowed calls. */
251 case BPF_RET | BPF_K:
252 case BPF_RET | BPF_A:
253 case BPF_ALU | BPF_ADD | BPF_K:
254 case BPF_ALU | BPF_ADD | BPF_X:
255 case BPF_ALU | BPF_SUB | BPF_K:
256 case BPF_ALU | BPF_SUB | BPF_X:
257 case BPF_ALU | BPF_MUL | BPF_K:
258 case BPF_ALU | BPF_MUL | BPF_X:
259 case BPF_ALU | BPF_DIV | BPF_K:
260 case BPF_ALU | BPF_DIV | BPF_X:
261 case BPF_ALU | BPF_AND | BPF_K:
262 case BPF_ALU | BPF_AND | BPF_X:
263 case BPF_ALU | BPF_OR | BPF_K:
264 case BPF_ALU | BPF_OR | BPF_X:
265 case BPF_ALU | BPF_XOR | BPF_K:
266 case BPF_ALU | BPF_XOR | BPF_X:
267 case BPF_ALU | BPF_LSH | BPF_K:
268 case BPF_ALU | BPF_LSH | BPF_X:
269 case BPF_ALU | BPF_RSH | BPF_K:
270 case BPF_ALU | BPF_RSH | BPF_X:
271 case BPF_ALU | BPF_NEG:
272 case BPF_LD | BPF_IMM:
273 case BPF_LDX | BPF_IMM:
274 case BPF_MISC | BPF_TAX:
275 case BPF_MISC | BPF_TXA:
276 case BPF_LD | BPF_MEM:
277 case BPF_LDX | BPF_MEM:
278 case BPF_ST:
279 case BPF_STX:
280 case BPF_JMP | BPF_JA:
281 case BPF_JMP | BPF_JEQ | BPF_K:
282 case BPF_JMP | BPF_JEQ | BPF_X:
283 case BPF_JMP | BPF_JGE | BPF_K:
284 case BPF_JMP | BPF_JGE | BPF_X:
285 case BPF_JMP | BPF_JGT | BPF_K:
286 case BPF_JMP | BPF_JGT | BPF_X:
287 case BPF_JMP | BPF_JSET | BPF_K:
288 case BPF_JMP | BPF_JSET | BPF_X:
289 continue;
290 default:
291 return -EINVAL;
292 }
293 }
294 return 0;
295}
296
297/**
298 * seccomp_run_filters - evaluates all seccomp filters against @sd
299 * @sd: optional seccomp data to be passed to filters
300 * @match: stores struct seccomp_filter that resulted in the return value,
301 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
302 * be unchanged.
303 *
304 * Returns valid seccomp BPF response codes.
305 */
306#define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
307static u32 seccomp_run_filters(const struct seccomp_data *sd,
308 struct seccomp_filter **match)
309{
310 u32 ret = SECCOMP_RET_ALLOW;
311 /* Make sure cross-thread synced filter points somewhere sane. */
312 struct seccomp_filter *f =
313 READ_ONCE(current->seccomp.filter);
314
315 /* Ensure unexpected behavior doesn't result in failing open. */
316 if (WARN_ON(f == NULL))
317 return SECCOMP_RET_KILL_PROCESS;
318
319 /*
320 * All filters in the list are evaluated and the lowest BPF return
321 * value always takes priority (ignoring the DATA).
322 */
323 for (; f; f = f->prev) {
324 u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd);
325
326 if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
327 ret = cur_ret;
328 *match = f;
329 }
330 }
331 return ret;
332}
333#endif /* CONFIG_SECCOMP_FILTER */
334
335static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
336{
337 assert_spin_locked(¤t->sighand->siglock);
338
339 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
340 return false;
341
342 return true;
343}
344
345void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
346
347static inline void seccomp_assign_mode(struct task_struct *task,
348 unsigned long seccomp_mode,
349 unsigned long flags)
350{
351 assert_spin_locked(&task->sighand->siglock);
352
353 task->seccomp.mode = seccomp_mode;
354 /*
355 * Make sure TIF_SECCOMP cannot be set before the mode (and
356 * filter) is set.
357 */
358 smp_mb__before_atomic();
359 /* Assume default seccomp processes want spec flaw mitigation. */
360 if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
361 arch_seccomp_spec_mitigate(task);
362 set_tsk_thread_flag(task, TIF_SECCOMP);
363}
364
365#ifdef CONFIG_SECCOMP_FILTER
366/* Returns 1 if the parent is an ancestor of the child. */
367static int is_ancestor(struct seccomp_filter *parent,
368 struct seccomp_filter *child)
369{
370 /* NULL is the root ancestor. */
371 if (parent == NULL)
372 return 1;
373 for (; child; child = child->prev)
374 if (child == parent)
375 return 1;
376 return 0;
377}
378
379/**
380 * seccomp_can_sync_threads: checks if all threads can be synchronized
381 *
382 * Expects sighand and cred_guard_mutex locks to be held.
383 *
384 * Returns 0 on success, -ve on error, or the pid of a thread which was
385 * either not in the correct seccomp mode or did not have an ancestral
386 * seccomp filter.
387 */
388static inline pid_t seccomp_can_sync_threads(void)
389{
390 struct task_struct *thread, *caller;
391
392 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
393 assert_spin_locked(¤t->sighand->siglock);
394
395 /* Validate all threads being eligible for synchronization. */
396 caller = current;
397 for_each_thread(caller, thread) {
398 pid_t failed;
399
400 /* Skip current, since it is initiating the sync. */
401 if (thread == caller)
402 continue;
403
404 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
405 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
406 is_ancestor(thread->seccomp.filter,
407 caller->seccomp.filter)))
408 continue;
409
410 /* Return the first thread that cannot be synchronized. */
411 failed = task_pid_vnr(thread);
412 /* If the pid cannot be resolved, then return -ESRCH */
413 if (WARN_ON(failed == 0))
414 failed = -ESRCH;
415 return failed;
416 }
417
418 return 0;
419}
420
421static inline void seccomp_filter_free(struct seccomp_filter *filter)
422{
423 if (filter) {
424 bpf_prog_destroy(filter->prog);
425 kfree(filter);
426 }
427}
428
429static void __seccomp_filter_orphan(struct seccomp_filter *orig)
430{
431 while (orig && refcount_dec_and_test(&orig->users)) {
432 if (waitqueue_active(&orig->wqh))
433 wake_up_poll(&orig->wqh, EPOLLHUP);
434 orig = orig->prev;
435 }
436}
437
438static void __put_seccomp_filter(struct seccomp_filter *orig)
439{
440 /* Clean up single-reference branches iteratively. */
441 while (orig && refcount_dec_and_test(&orig->refs)) {
442 struct seccomp_filter *freeme = orig;
443 orig = orig->prev;
444 seccomp_filter_free(freeme);
445 }
446}
447
448static void __seccomp_filter_release(struct seccomp_filter *orig)
449{
450 /* Notify about any unused filters in the task's former filter tree. */
451 __seccomp_filter_orphan(orig);
452 /* Finally drop all references to the task's former tree. */
453 __put_seccomp_filter(orig);
454}
455
456/**
457 * seccomp_filter_release - Detach the task from its filter tree,
458 * drop its reference count, and notify
459 * about unused filters
460 *
461 * This function should only be called when the task is exiting as
462 * it detaches it from its filter tree. As such, READ_ONCE() and
463 * barriers are not needed here, as would normally be needed.
464 */
465void seccomp_filter_release(struct task_struct *tsk)
466{
467 struct seccomp_filter *orig = tsk->seccomp.filter;
468
469 /* Detach task from its filter tree. */
470 tsk->seccomp.filter = NULL;
471 __seccomp_filter_release(orig);
472}
473
474/**
475 * seccomp_sync_threads: sets all threads to use current's filter
476 *
477 * Expects sighand and cred_guard_mutex locks to be held, and for
478 * seccomp_can_sync_threads() to have returned success already
479 * without dropping the locks.
480 *
481 */
482static inline void seccomp_sync_threads(unsigned long flags)
483{
484 struct task_struct *thread, *caller;
485
486 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
487 assert_spin_locked(¤t->sighand->siglock);
488
489 /* Synchronize all threads. */
490 caller = current;
491 for_each_thread(caller, thread) {
492 /* Skip current, since it needs no changes. */
493 if (thread == caller)
494 continue;
495
496 /* Get a task reference for the new leaf node. */
497 get_seccomp_filter(caller);
498
499 /*
500 * Drop the task reference to the shared ancestor since
501 * current's path will hold a reference. (This also
502 * allows a put before the assignment.)
503 */
504 __seccomp_filter_release(thread->seccomp.filter);
505
506 /* Make our new filter tree visible. */
507 smp_store_release(&thread->seccomp.filter,
508 caller->seccomp.filter);
509 atomic_set(&thread->seccomp.filter_count,
510 atomic_read(&thread->seccomp.filter_count));
511
512 /*
513 * Don't let an unprivileged task work around
514 * the no_new_privs restriction by creating
515 * a thread that sets it up, enters seccomp,
516 * then dies.
517 */
518 if (task_no_new_privs(caller))
519 task_set_no_new_privs(thread);
520
521 /*
522 * Opt the other thread into seccomp if needed.
523 * As threads are considered to be trust-realm
524 * equivalent (see ptrace_may_access), it is safe to
525 * allow one thread to transition the other.
526 */
527 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
528 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
529 flags);
530 }
531}
532
533/**
534 * seccomp_prepare_filter: Prepares a seccomp filter for use.
535 * @fprog: BPF program to install
536 *
537 * Returns filter on success or an ERR_PTR on failure.
538 */
539static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
540{
541 struct seccomp_filter *sfilter;
542 int ret;
543 const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
544
545 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
546 return ERR_PTR(-EINVAL);
547
548 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
549
550 /*
551 * Installing a seccomp filter requires that the task has
552 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
553 * This avoids scenarios where unprivileged tasks can affect the
554 * behavior of privileged children.
555 */
556 if (!task_no_new_privs(current) &&
557 security_capable(current_cred(), current_user_ns(),
558 CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) != 0)
559 return ERR_PTR(-EACCES);
560
561 /* Allocate a new seccomp_filter */
562 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
563 if (!sfilter)
564 return ERR_PTR(-ENOMEM);
565
566 mutex_init(&sfilter->notify_lock);
567 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
568 seccomp_check_filter, save_orig);
569 if (ret < 0) {
570 kfree(sfilter);
571 return ERR_PTR(ret);
572 }
573
574 refcount_set(&sfilter->refs, 1);
575 refcount_set(&sfilter->users, 1);
576 init_waitqueue_head(&sfilter->wqh);
577
578 return sfilter;
579}
580
581/**
582 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
583 * @user_filter: pointer to the user data containing a sock_fprog.
584 *
585 * Returns 0 on success and non-zero otherwise.
586 */
587static struct seccomp_filter *
588seccomp_prepare_user_filter(const char __user *user_filter)
589{
590 struct sock_fprog fprog;
591 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
592
593#ifdef CONFIG_COMPAT
594 if (in_compat_syscall()) {
595 struct compat_sock_fprog fprog32;
596 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
597 goto out;
598 fprog.len = fprog32.len;
599 fprog.filter = compat_ptr(fprog32.filter);
600 } else /* falls through to the if below. */
601#endif
602 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
603 goto out;
604 filter = seccomp_prepare_filter(&fprog);
605out:
606 return filter;
607}
608
609/**
610 * seccomp_attach_filter: validate and attach filter
611 * @flags: flags to change filter behavior
612 * @filter: seccomp filter to add to the current process
613 *
614 * Caller must be holding current->sighand->siglock lock.
615 *
616 * Returns 0 on success, -ve on error, or
617 * - in TSYNC mode: the pid of a thread which was either not in the correct
618 * seccomp mode or did not have an ancestral seccomp filter
619 * - in NEW_LISTENER mode: the fd of the new listener
620 */
621static long seccomp_attach_filter(unsigned int flags,
622 struct seccomp_filter *filter)
623{
624 unsigned long total_insns;
625 struct seccomp_filter *walker;
626
627 assert_spin_locked(¤t->sighand->siglock);
628
629 /* Validate resulting filter length. */
630 total_insns = filter->prog->len;
631 for (walker = current->seccomp.filter; walker; walker = walker->prev)
632 total_insns += walker->prog->len + 4; /* 4 instr penalty */
633 if (total_insns > MAX_INSNS_PER_PATH)
634 return -ENOMEM;
635
636 /* If thread sync has been requested, check that it is possible. */
637 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
638 int ret;
639
640 ret = seccomp_can_sync_threads();
641 if (ret) {
642 if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
643 return -ESRCH;
644 else
645 return ret;
646 }
647 }
648
649 /* Set log flag, if present. */
650 if (flags & SECCOMP_FILTER_FLAG_LOG)
651 filter->log = true;
652
653 /*
654 * If there is an existing filter, make it the prev and don't drop its
655 * task reference.
656 */
657 filter->prev = current->seccomp.filter;
658 current->seccomp.filter = filter;
659 atomic_inc(¤t->seccomp.filter_count);
660
661 /* Now that the new filter is in place, synchronize to all threads. */
662 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
663 seccomp_sync_threads(flags);
664
665 return 0;
666}
667
668static void __get_seccomp_filter(struct seccomp_filter *filter)
669{
670 refcount_inc(&filter->refs);
671}
672
673/* get_seccomp_filter - increments the reference count of the filter on @tsk */
674void get_seccomp_filter(struct task_struct *tsk)
675{
676 struct seccomp_filter *orig = tsk->seccomp.filter;
677 if (!orig)
678 return;
679 __get_seccomp_filter(orig);
680 refcount_inc(&orig->users);
681}
682
683static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason)
684{
685 clear_siginfo(info);
686 info->si_signo = SIGSYS;
687 info->si_code = SYS_SECCOMP;
688 info->si_call_addr = (void __user *)KSTK_EIP(current);
689 info->si_errno = reason;
690 info->si_arch = syscall_get_arch(current);
691 info->si_syscall = syscall;
692}
693
694/**
695 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
696 * @syscall: syscall number to send to userland
697 * @reason: filter-supplied reason code to send to userland (via si_errno)
698 *
699 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
700 */
701static void seccomp_send_sigsys(int syscall, int reason)
702{
703 struct kernel_siginfo info;
704 seccomp_init_siginfo(&info, syscall, reason);
705 force_sig_info(&info);
706}
707#endif /* CONFIG_SECCOMP_FILTER */
708
709/* For use with seccomp_actions_logged */
710#define SECCOMP_LOG_KILL_PROCESS (1 << 0)
711#define SECCOMP_LOG_KILL_THREAD (1 << 1)
712#define SECCOMP_LOG_TRAP (1 << 2)
713#define SECCOMP_LOG_ERRNO (1 << 3)
714#define SECCOMP_LOG_TRACE (1 << 4)
715#define SECCOMP_LOG_LOG (1 << 5)
716#define SECCOMP_LOG_ALLOW (1 << 6)
717#define SECCOMP_LOG_USER_NOTIF (1 << 7)
718
719static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
720 SECCOMP_LOG_KILL_THREAD |
721 SECCOMP_LOG_TRAP |
722 SECCOMP_LOG_ERRNO |
723 SECCOMP_LOG_USER_NOTIF |
724 SECCOMP_LOG_TRACE |
725 SECCOMP_LOG_LOG;
726
727static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
728 bool requested)
729{
730 bool log = false;
731
732 switch (action) {
733 case SECCOMP_RET_ALLOW:
734 break;
735 case SECCOMP_RET_TRAP:
736 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
737 break;
738 case SECCOMP_RET_ERRNO:
739 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
740 break;
741 case SECCOMP_RET_TRACE:
742 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
743 break;
744 case SECCOMP_RET_USER_NOTIF:
745 log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
746 break;
747 case SECCOMP_RET_LOG:
748 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
749 break;
750 case SECCOMP_RET_KILL_THREAD:
751 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
752 break;
753 case SECCOMP_RET_KILL_PROCESS:
754 default:
755 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
756 }
757
758 /*
759 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
760 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
761 * any action from being logged by removing the action name from the
762 * seccomp_actions_logged sysctl.
763 */
764 if (!log)
765 return;
766
767 audit_seccomp(syscall, signr, action);
768}
769
770/*
771 * Secure computing mode 1 allows only read/write/exit/sigreturn.
772 * To be fully secure this must be combined with rlimit
773 * to limit the stack allocations too.
774 */
775static const int mode1_syscalls[] = {
776 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
777 -1, /* negative terminated */
778};
779
780static void __secure_computing_strict(int this_syscall)
781{
782 const int *allowed_syscalls = mode1_syscalls;
783#ifdef CONFIG_COMPAT
784 if (in_compat_syscall())
785 allowed_syscalls = get_compat_mode1_syscalls();
786#endif
787 do {
788 if (*allowed_syscalls == this_syscall)
789 return;
790 } while (*++allowed_syscalls != -1);
791
792#ifdef SECCOMP_DEBUG
793 dump_stack();
794#endif
795 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
796 do_exit(SIGKILL);
797}
798
799#ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
800void secure_computing_strict(int this_syscall)
801{
802 int mode = current->seccomp.mode;
803
804 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
805 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
806 return;
807
808 if (mode == SECCOMP_MODE_DISABLED)
809 return;
810 else if (mode == SECCOMP_MODE_STRICT)
811 __secure_computing_strict(this_syscall);
812 else
813 BUG();
814}
815#else
816
817#ifdef CONFIG_SECCOMP_FILTER
818static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
819{
820 /*
821 * Note: overflow is ok here, the id just needs to be unique per
822 * filter.
823 */
824 lockdep_assert_held(&filter->notify_lock);
825 return filter->notif->next_id++;
826}
827
828static void seccomp_handle_addfd(struct seccomp_kaddfd *addfd)
829{
830 /*
831 * Remove the notification, and reset the list pointers, indicating
832 * that it has been handled.
833 */
834 list_del_init(&addfd->list);
835 addfd->ret = receive_fd_replace(addfd->fd, addfd->file, addfd->flags);
836 complete(&addfd->completion);
837}
838
839static int seccomp_do_user_notification(int this_syscall,
840 struct seccomp_filter *match,
841 const struct seccomp_data *sd)
842{
843 int err;
844 u32 flags = 0;
845 long ret = 0;
846 struct seccomp_knotif n = {};
847 struct seccomp_kaddfd *addfd, *tmp;
848
849 mutex_lock(&match->notify_lock);
850 err = -ENOSYS;
851 if (!match->notif)
852 goto out;
853
854 n.task = current;
855 n.state = SECCOMP_NOTIFY_INIT;
856 n.data = sd;
857 n.id = seccomp_next_notify_id(match);
858 init_completion(&n.ready);
859 list_add(&n.list, &match->notif->notifications);
860 INIT_LIST_HEAD(&n.addfd);
861
862 up(&match->notif->request);
863 wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM);
864 mutex_unlock(&match->notify_lock);
865
866 /*
867 * This is where we wait for a reply from userspace.
868 */
869wait:
870 err = wait_for_completion_interruptible(&n.ready);
871 mutex_lock(&match->notify_lock);
872 if (err == 0) {
873 /* Check if we were woken up by a addfd message */
874 addfd = list_first_entry_or_null(&n.addfd,
875 struct seccomp_kaddfd, list);
876 if (addfd && n.state != SECCOMP_NOTIFY_REPLIED) {
877 seccomp_handle_addfd(addfd);
878 mutex_unlock(&match->notify_lock);
879 goto wait;
880 }
881 ret = n.val;
882 err = n.error;
883 flags = n.flags;
884 }
885
886 /* If there were any pending addfd calls, clear them out */
887 list_for_each_entry_safe(addfd, tmp, &n.addfd, list) {
888 /* The process went away before we got a chance to handle it */
889 addfd->ret = -ESRCH;
890 list_del_init(&addfd->list);
891 complete(&addfd->completion);
892 }
893
894 /*
895 * Note that it's possible the listener died in between the time when
896 * we were notified of a response (or a signal) and when we were able to
897 * re-acquire the lock, so only delete from the list if the
898 * notification actually exists.
899 *
900 * Also note that this test is only valid because there's no way to
901 * *reattach* to a notifier right now. If one is added, we'll need to
902 * keep track of the notif itself and make sure they match here.
903 */
904 if (match->notif)
905 list_del(&n.list);
906out:
907 mutex_unlock(&match->notify_lock);
908
909 /* Userspace requests to continue the syscall. */
910 if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE)
911 return 0;
912
913 syscall_set_return_value(current, task_pt_regs(current),
914 err, ret);
915 return -1;
916}
917
918static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
919 const bool recheck_after_trace)
920{
921 u32 filter_ret, action;
922 struct seccomp_filter *match = NULL;
923 int data;
924 struct seccomp_data sd_local;
925
926 /*
927 * Make sure that any changes to mode from another thread have
928 * been seen after TIF_SECCOMP was seen.
929 */
930 rmb();
931
932 if (!sd) {
933 populate_seccomp_data(&sd_local);
934 sd = &sd_local;
935 }
936
937 filter_ret = seccomp_run_filters(sd, &match);
938 data = filter_ret & SECCOMP_RET_DATA;
939 action = filter_ret & SECCOMP_RET_ACTION_FULL;
940
941 switch (action) {
942 case SECCOMP_RET_ERRNO:
943 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
944 if (data > MAX_ERRNO)
945 data = MAX_ERRNO;
946 syscall_set_return_value(current, task_pt_regs(current),
947 -data, 0);
948 goto skip;
949
950 case SECCOMP_RET_TRAP:
951 /* Show the handler the original registers. */
952 syscall_rollback(current, task_pt_regs(current));
953 /* Let the filter pass back 16 bits of data. */
954 seccomp_send_sigsys(this_syscall, data);
955 goto skip;
956
957 case SECCOMP_RET_TRACE:
958 /* We've been put in this state by the ptracer already. */
959 if (recheck_after_trace)
960 return 0;
961
962 /* ENOSYS these calls if there is no tracer attached. */
963 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
964 syscall_set_return_value(current,
965 task_pt_regs(current),
966 -ENOSYS, 0);
967 goto skip;
968 }
969
970 /* Allow the BPF to provide the event message */
971 ptrace_event(PTRACE_EVENT_SECCOMP, data);
972 /*
973 * The delivery of a fatal signal during event
974 * notification may silently skip tracer notification,
975 * which could leave us with a potentially unmodified
976 * syscall that the tracer would have liked to have
977 * changed. Since the process is about to die, we just
978 * force the syscall to be skipped and let the signal
979 * kill the process and correctly handle any tracer exit
980 * notifications.
981 */
982 if (fatal_signal_pending(current))
983 goto skip;
984 /* Check if the tracer forced the syscall to be skipped. */
985 this_syscall = syscall_get_nr(current, task_pt_regs(current));
986 if (this_syscall < 0)
987 goto skip;
988
989 /*
990 * Recheck the syscall, since it may have changed. This
991 * intentionally uses a NULL struct seccomp_data to force
992 * a reload of all registers. This does not goto skip since
993 * a skip would have already been reported.
994 */
995 if (__seccomp_filter(this_syscall, NULL, true))
996 return -1;
997
998 return 0;
999
1000 case SECCOMP_RET_USER_NOTIF:
1001 if (seccomp_do_user_notification(this_syscall, match, sd))
1002 goto skip;
1003
1004 return 0;
1005
1006 case SECCOMP_RET_LOG:
1007 seccomp_log(this_syscall, 0, action, true);
1008 return 0;
1009
1010 case SECCOMP_RET_ALLOW:
1011 /*
1012 * Note that the "match" filter will always be NULL for
1013 * this action since SECCOMP_RET_ALLOW is the starting
1014 * state in seccomp_run_filters().
1015 */
1016 return 0;
1017
1018 case SECCOMP_RET_KILL_THREAD:
1019 case SECCOMP_RET_KILL_PROCESS:
1020 default:
1021 seccomp_log(this_syscall, SIGSYS, action, true);
1022 /* Dump core only if this is the last remaining thread. */
1023 if (action == SECCOMP_RET_KILL_PROCESS ||
1024 get_nr_threads(current) == 1) {
1025 kernel_siginfo_t info;
1026
1027 /* Show the original registers in the dump. */
1028 syscall_rollback(current, task_pt_regs(current));
1029 /* Trigger a manual coredump since do_exit skips it. */
1030 seccomp_init_siginfo(&info, this_syscall, data);
1031 do_coredump(&info);
1032 }
1033 if (action == SECCOMP_RET_KILL_PROCESS)
1034 do_group_exit(SIGSYS);
1035 else
1036 do_exit(SIGSYS);
1037 }
1038
1039 unreachable();
1040
1041skip:
1042 seccomp_log(this_syscall, 0, action, match ? match->log : false);
1043 return -1;
1044}
1045#else
1046static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1047 const bool recheck_after_trace)
1048{
1049 BUG();
1050}
1051#endif
1052
1053int __secure_computing(const struct seccomp_data *sd)
1054{
1055 int mode = current->seccomp.mode;
1056 int this_syscall;
1057
1058 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1059 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1060 return 0;
1061
1062 this_syscall = sd ? sd->nr :
1063 syscall_get_nr(current, task_pt_regs(current));
1064
1065 switch (mode) {
1066 case SECCOMP_MODE_STRICT:
1067 __secure_computing_strict(this_syscall); /* may call do_exit */
1068 return 0;
1069 case SECCOMP_MODE_FILTER:
1070 return __seccomp_filter(this_syscall, sd, false);
1071 default:
1072 BUG();
1073 }
1074}
1075#endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1076
1077long prctl_get_seccomp(void)
1078{
1079 return current->seccomp.mode;
1080}
1081
1082/**
1083 * seccomp_set_mode_strict: internal function for setting strict seccomp
1084 *
1085 * Once current->seccomp.mode is non-zero, it may not be changed.
1086 *
1087 * Returns 0 on success or -EINVAL on failure.
1088 */
1089static long seccomp_set_mode_strict(void)
1090{
1091 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
1092 long ret = -EINVAL;
1093
1094 spin_lock_irq(¤t->sighand->siglock);
1095
1096 if (!seccomp_may_assign_mode(seccomp_mode))
1097 goto out;
1098
1099#ifdef TIF_NOTSC
1100 disable_TSC();
1101#endif
1102 seccomp_assign_mode(current, seccomp_mode, 0);
1103 ret = 0;
1104
1105out:
1106 spin_unlock_irq(¤t->sighand->siglock);
1107
1108 return ret;
1109}
1110
1111#ifdef CONFIG_SECCOMP_FILTER
1112static void seccomp_notify_free(struct seccomp_filter *filter)
1113{
1114 kfree(filter->notif);
1115 filter->notif = NULL;
1116}
1117
1118static void seccomp_notify_detach(struct seccomp_filter *filter)
1119{
1120 struct seccomp_knotif *knotif;
1121
1122 if (!filter)
1123 return;
1124
1125 mutex_lock(&filter->notify_lock);
1126
1127 /*
1128 * If this file is being closed because e.g. the task who owned it
1129 * died, let's wake everyone up who was waiting on us.
1130 */
1131 list_for_each_entry(knotif, &filter->notif->notifications, list) {
1132 if (knotif->state == SECCOMP_NOTIFY_REPLIED)
1133 continue;
1134
1135 knotif->state = SECCOMP_NOTIFY_REPLIED;
1136 knotif->error = -ENOSYS;
1137 knotif->val = 0;
1138
1139 /*
1140 * We do not need to wake up any pending addfd messages, as
1141 * the notifier will do that for us, as this just looks
1142 * like a standard reply.
1143 */
1144 complete(&knotif->ready);
1145 }
1146
1147 seccomp_notify_free(filter);
1148 mutex_unlock(&filter->notify_lock);
1149}
1150
1151static int seccomp_notify_release(struct inode *inode, struct file *file)
1152{
1153 struct seccomp_filter *filter = file->private_data;
1154
1155 seccomp_notify_detach(filter);
1156 __put_seccomp_filter(filter);
1157 return 0;
1158}
1159
1160/* must be called with notif_lock held */
1161static inline struct seccomp_knotif *
1162find_notification(struct seccomp_filter *filter, u64 id)
1163{
1164 struct seccomp_knotif *cur;
1165
1166 lockdep_assert_held(&filter->notify_lock);
1167
1168 list_for_each_entry(cur, &filter->notif->notifications, list) {
1169 if (cur->id == id)
1170 return cur;
1171 }
1172
1173 return NULL;
1174}
1175
1176
1177static long seccomp_notify_recv(struct seccomp_filter *filter,
1178 void __user *buf)
1179{
1180 struct seccomp_knotif *knotif = NULL, *cur;
1181 struct seccomp_notif unotif;
1182 ssize_t ret;
1183
1184 /* Verify that we're not given garbage to keep struct extensible. */
1185 ret = check_zeroed_user(buf, sizeof(unotif));
1186 if (ret < 0)
1187 return ret;
1188 if (!ret)
1189 return -EINVAL;
1190
1191 memset(&unotif, 0, sizeof(unotif));
1192
1193 ret = down_interruptible(&filter->notif->request);
1194 if (ret < 0)
1195 return ret;
1196
1197 mutex_lock(&filter->notify_lock);
1198 list_for_each_entry(cur, &filter->notif->notifications, list) {
1199 if (cur->state == SECCOMP_NOTIFY_INIT) {
1200 knotif = cur;
1201 break;
1202 }
1203 }
1204
1205 /*
1206 * If we didn't find a notification, it could be that the task was
1207 * interrupted by a fatal signal between the time we were woken and
1208 * when we were able to acquire the rw lock.
1209 */
1210 if (!knotif) {
1211 ret = -ENOENT;
1212 goto out;
1213 }
1214
1215 unotif.id = knotif->id;
1216 unotif.pid = task_pid_vnr(knotif->task);
1217 unotif.data = *(knotif->data);
1218
1219 knotif->state = SECCOMP_NOTIFY_SENT;
1220 wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM);
1221 ret = 0;
1222out:
1223 mutex_unlock(&filter->notify_lock);
1224
1225 if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1226 ret = -EFAULT;
1227
1228 /*
1229 * Userspace screwed up. To make sure that we keep this
1230 * notification alive, let's reset it back to INIT. It
1231 * may have died when we released the lock, so we need to make
1232 * sure it's still around.
1233 */
1234 mutex_lock(&filter->notify_lock);
1235 knotif = find_notification(filter, unotif.id);
1236 if (knotif) {
1237 knotif->state = SECCOMP_NOTIFY_INIT;
1238 up(&filter->notif->request);
1239 }
1240 mutex_unlock(&filter->notify_lock);
1241 }
1242
1243 return ret;
1244}
1245
1246static long seccomp_notify_send(struct seccomp_filter *filter,
1247 void __user *buf)
1248{
1249 struct seccomp_notif_resp resp = {};
1250 struct seccomp_knotif *knotif;
1251 long ret;
1252
1253 if (copy_from_user(&resp, buf, sizeof(resp)))
1254 return -EFAULT;
1255
1256 if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1257 return -EINVAL;
1258
1259 if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) &&
1260 (resp.error || resp.val))
1261 return -EINVAL;
1262
1263 ret = mutex_lock_interruptible(&filter->notify_lock);
1264 if (ret < 0)
1265 return ret;
1266
1267 knotif = find_notification(filter, resp.id);
1268 if (!knotif) {
1269 ret = -ENOENT;
1270 goto out;
1271 }
1272
1273 /* Allow exactly one reply. */
1274 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1275 ret = -EINPROGRESS;
1276 goto out;
1277 }
1278
1279 ret = 0;
1280 knotif->state = SECCOMP_NOTIFY_REPLIED;
1281 knotif->error = resp.error;
1282 knotif->val = resp.val;
1283 knotif->flags = resp.flags;
1284 complete(&knotif->ready);
1285out:
1286 mutex_unlock(&filter->notify_lock);
1287 return ret;
1288}
1289
1290static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1291 void __user *buf)
1292{
1293 struct seccomp_knotif *knotif;
1294 u64 id;
1295 long ret;
1296
1297 if (copy_from_user(&id, buf, sizeof(id)))
1298 return -EFAULT;
1299
1300 ret = mutex_lock_interruptible(&filter->notify_lock);
1301 if (ret < 0)
1302 return ret;
1303
1304 knotif = find_notification(filter, id);
1305 if (knotif && knotif->state == SECCOMP_NOTIFY_SENT)
1306 ret = 0;
1307 else
1308 ret = -ENOENT;
1309
1310 mutex_unlock(&filter->notify_lock);
1311 return ret;
1312}
1313
1314static long seccomp_notify_addfd(struct seccomp_filter *filter,
1315 struct seccomp_notif_addfd __user *uaddfd,
1316 unsigned int size)
1317{
1318 struct seccomp_notif_addfd addfd;
1319 struct seccomp_knotif *knotif;
1320 struct seccomp_kaddfd kaddfd;
1321 int ret;
1322
1323 BUILD_BUG_ON(sizeof(addfd) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0);
1324 BUILD_BUG_ON(sizeof(addfd) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST);
1325
1326 if (size < SECCOMP_NOTIFY_ADDFD_SIZE_VER0 || size >= PAGE_SIZE)
1327 return -EINVAL;
1328
1329 ret = copy_struct_from_user(&addfd, sizeof(addfd), uaddfd, size);
1330 if (ret)
1331 return ret;
1332
1333 if (addfd.newfd_flags & ~O_CLOEXEC)
1334 return -EINVAL;
1335
1336 if (addfd.flags & ~SECCOMP_ADDFD_FLAG_SETFD)
1337 return -EINVAL;
1338
1339 if (addfd.newfd && !(addfd.flags & SECCOMP_ADDFD_FLAG_SETFD))
1340 return -EINVAL;
1341
1342 kaddfd.file = fget(addfd.srcfd);
1343 if (!kaddfd.file)
1344 return -EBADF;
1345
1346 kaddfd.flags = addfd.newfd_flags;
1347 kaddfd.fd = (addfd.flags & SECCOMP_ADDFD_FLAG_SETFD) ?
1348 addfd.newfd : -1;
1349 init_completion(&kaddfd.completion);
1350
1351 ret = mutex_lock_interruptible(&filter->notify_lock);
1352 if (ret < 0)
1353 goto out;
1354
1355 knotif = find_notification(filter, addfd.id);
1356 if (!knotif) {
1357 ret = -ENOENT;
1358 goto out_unlock;
1359 }
1360
1361 /*
1362 * We do not want to allow for FD injection to occur before the
1363 * notification has been picked up by a userspace handler, or after
1364 * the notification has been replied to.
1365 */
1366 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1367 ret = -EINPROGRESS;
1368 goto out_unlock;
1369 }
1370
1371 list_add(&kaddfd.list, &knotif->addfd);
1372 complete(&knotif->ready);
1373 mutex_unlock(&filter->notify_lock);
1374
1375 /* Now we wait for it to be processed or be interrupted */
1376 ret = wait_for_completion_interruptible(&kaddfd.completion);
1377 if (ret == 0) {
1378 /*
1379 * We had a successful completion. The other side has already
1380 * removed us from the addfd queue, and
1381 * wait_for_completion_interruptible has a memory barrier upon
1382 * success that lets us read this value directly without
1383 * locking.
1384 */
1385 ret = kaddfd.ret;
1386 goto out;
1387 }
1388
1389 mutex_lock(&filter->notify_lock);
1390 /*
1391 * Even though we were woken up by a signal and not a successful
1392 * completion, a completion may have happened in the mean time.
1393 *
1394 * We need to check again if the addfd request has been handled,
1395 * and if not, we will remove it from the queue.
1396 */
1397 if (list_empty(&kaddfd.list))
1398 ret = kaddfd.ret;
1399 else
1400 list_del(&kaddfd.list);
1401
1402out_unlock:
1403 mutex_unlock(&filter->notify_lock);
1404out:
1405 fput(kaddfd.file);
1406
1407 return ret;
1408}
1409
1410static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1411 unsigned long arg)
1412{
1413 struct seccomp_filter *filter = file->private_data;
1414 void __user *buf = (void __user *)arg;
1415
1416 /* Fixed-size ioctls */
1417 switch (cmd) {
1418 case SECCOMP_IOCTL_NOTIF_RECV:
1419 return seccomp_notify_recv(filter, buf);
1420 case SECCOMP_IOCTL_NOTIF_SEND:
1421 return seccomp_notify_send(filter, buf);
1422 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR:
1423 case SECCOMP_IOCTL_NOTIF_ID_VALID:
1424 return seccomp_notify_id_valid(filter, buf);
1425 }
1426
1427 /* Extensible Argument ioctls */
1428#define EA_IOCTL(cmd) ((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1429 switch (EA_IOCTL(cmd)) {
1430 case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD):
1431 return seccomp_notify_addfd(filter, buf, _IOC_SIZE(cmd));
1432 default:
1433 return -EINVAL;
1434 }
1435}
1436
1437static __poll_t seccomp_notify_poll(struct file *file,
1438 struct poll_table_struct *poll_tab)
1439{
1440 struct seccomp_filter *filter = file->private_data;
1441 __poll_t ret = 0;
1442 struct seccomp_knotif *cur;
1443
1444 poll_wait(file, &filter->wqh, poll_tab);
1445
1446 if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1447 return EPOLLERR;
1448
1449 list_for_each_entry(cur, &filter->notif->notifications, list) {
1450 if (cur->state == SECCOMP_NOTIFY_INIT)
1451 ret |= EPOLLIN | EPOLLRDNORM;
1452 if (cur->state == SECCOMP_NOTIFY_SENT)
1453 ret |= EPOLLOUT | EPOLLWRNORM;
1454 if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1455 break;
1456 }
1457
1458 mutex_unlock(&filter->notify_lock);
1459
1460 if (refcount_read(&filter->users) == 0)
1461 ret |= EPOLLHUP;
1462
1463 return ret;
1464}
1465
1466static const struct file_operations seccomp_notify_ops = {
1467 .poll = seccomp_notify_poll,
1468 .release = seccomp_notify_release,
1469 .unlocked_ioctl = seccomp_notify_ioctl,
1470 .compat_ioctl = seccomp_notify_ioctl,
1471};
1472
1473static struct file *init_listener(struct seccomp_filter *filter)
1474{
1475 struct file *ret = ERR_PTR(-EBUSY);
1476 struct seccomp_filter *cur;
1477
1478 for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1479 if (cur->notif)
1480 goto out;
1481 }
1482
1483 ret = ERR_PTR(-ENOMEM);
1484 filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1485 if (!filter->notif)
1486 goto out;
1487
1488 sema_init(&filter->notif->request, 0);
1489 filter->notif->next_id = get_random_u64();
1490 INIT_LIST_HEAD(&filter->notif->notifications);
1491
1492 ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1493 filter, O_RDWR);
1494 if (IS_ERR(ret))
1495 goto out_notif;
1496
1497 /* The file has a reference to it now */
1498 __get_seccomp_filter(filter);
1499
1500out_notif:
1501 if (IS_ERR(ret))
1502 seccomp_notify_free(filter);
1503out:
1504 return ret;
1505}
1506
1507/**
1508 * seccomp_set_mode_filter: internal function for setting seccomp filter
1509 * @flags: flags to change filter behavior
1510 * @filter: struct sock_fprog containing filter
1511 *
1512 * This function may be called repeatedly to install additional filters.
1513 * Every filter successfully installed will be evaluated (in reverse order)
1514 * for each system call the task makes.
1515 *
1516 * Once current->seccomp.mode is non-zero, it may not be changed.
1517 *
1518 * Returns 0 on success or -EINVAL on failure.
1519 */
1520static long seccomp_set_mode_filter(unsigned int flags,
1521 const char __user *filter)
1522{
1523 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1524 struct seccomp_filter *prepared = NULL;
1525 long ret = -EINVAL;
1526 int listener = -1;
1527 struct file *listener_f = NULL;
1528
1529 /* Validate flags. */
1530 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1531 return -EINVAL;
1532
1533 /*
1534 * In the successful case, NEW_LISTENER returns the new listener fd.
1535 * But in the failure case, TSYNC returns the thread that died. If you
1536 * combine these two flags, there's no way to tell whether something
1537 * succeeded or failed. So, let's disallow this combination if the user
1538 * has not explicitly requested no errors from TSYNC.
1539 */
1540 if ((flags & SECCOMP_FILTER_FLAG_TSYNC) &&
1541 (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) &&
1542 ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0))
1543 return -EINVAL;
1544
1545 /* Prepare the new filter before holding any locks. */
1546 prepared = seccomp_prepare_user_filter(filter);
1547 if (IS_ERR(prepared))
1548 return PTR_ERR(prepared);
1549
1550 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1551 listener = get_unused_fd_flags(O_CLOEXEC);
1552 if (listener < 0) {
1553 ret = listener;
1554 goto out_free;
1555 }
1556
1557 listener_f = init_listener(prepared);
1558 if (IS_ERR(listener_f)) {
1559 put_unused_fd(listener);
1560 ret = PTR_ERR(listener_f);
1561 goto out_free;
1562 }
1563 }
1564
1565 /*
1566 * Make sure we cannot change seccomp or nnp state via TSYNC
1567 * while another thread is in the middle of calling exec.
1568 */
1569 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1570 mutex_lock_killable(¤t->signal->cred_guard_mutex))
1571 goto out_put_fd;
1572
1573 spin_lock_irq(¤t->sighand->siglock);
1574
1575 if (!seccomp_may_assign_mode(seccomp_mode))
1576 goto out;
1577
1578 ret = seccomp_attach_filter(flags, prepared);
1579 if (ret)
1580 goto out;
1581 /* Do not free the successfully attached filter. */
1582 prepared = NULL;
1583
1584 seccomp_assign_mode(current, seccomp_mode, flags);
1585out:
1586 spin_unlock_irq(¤t->sighand->siglock);
1587 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
1588 mutex_unlock(¤t->signal->cred_guard_mutex);
1589out_put_fd:
1590 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1591 if (ret) {
1592 listener_f->private_data = NULL;
1593 fput(listener_f);
1594 put_unused_fd(listener);
1595 seccomp_notify_detach(prepared);
1596 } else {
1597 fd_install(listener, listener_f);
1598 ret = listener;
1599 }
1600 }
1601out_free:
1602 seccomp_filter_free(prepared);
1603 return ret;
1604}
1605#else
1606static inline long seccomp_set_mode_filter(unsigned int flags,
1607 const char __user *filter)
1608{
1609 return -EINVAL;
1610}
1611#endif
1612
1613static long seccomp_get_action_avail(const char __user *uaction)
1614{
1615 u32 action;
1616
1617 if (copy_from_user(&action, uaction, sizeof(action)))
1618 return -EFAULT;
1619
1620 switch (action) {
1621 case SECCOMP_RET_KILL_PROCESS:
1622 case SECCOMP_RET_KILL_THREAD:
1623 case SECCOMP_RET_TRAP:
1624 case SECCOMP_RET_ERRNO:
1625 case SECCOMP_RET_USER_NOTIF:
1626 case SECCOMP_RET_TRACE:
1627 case SECCOMP_RET_LOG:
1628 case SECCOMP_RET_ALLOW:
1629 break;
1630 default:
1631 return -EOPNOTSUPP;
1632 }
1633
1634 return 0;
1635}
1636
1637static long seccomp_get_notif_sizes(void __user *usizes)
1638{
1639 struct seccomp_notif_sizes sizes = {
1640 .seccomp_notif = sizeof(struct seccomp_notif),
1641 .seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
1642 .seccomp_data = sizeof(struct seccomp_data),
1643 };
1644
1645 if (copy_to_user(usizes, &sizes, sizeof(sizes)))
1646 return -EFAULT;
1647
1648 return 0;
1649}
1650
1651/* Common entry point for both prctl and syscall. */
1652static long do_seccomp(unsigned int op, unsigned int flags,
1653 void __user *uargs)
1654{
1655 switch (op) {
1656 case SECCOMP_SET_MODE_STRICT:
1657 if (flags != 0 || uargs != NULL)
1658 return -EINVAL;
1659 return seccomp_set_mode_strict();
1660 case SECCOMP_SET_MODE_FILTER:
1661 return seccomp_set_mode_filter(flags, uargs);
1662 case SECCOMP_GET_ACTION_AVAIL:
1663 if (flags != 0)
1664 return -EINVAL;
1665
1666 return seccomp_get_action_avail(uargs);
1667 case SECCOMP_GET_NOTIF_SIZES:
1668 if (flags != 0)
1669 return -EINVAL;
1670
1671 return seccomp_get_notif_sizes(uargs);
1672 default:
1673 return -EINVAL;
1674 }
1675}
1676
1677SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
1678 void __user *, uargs)
1679{
1680 return do_seccomp(op, flags, uargs);
1681}
1682
1683/**
1684 * prctl_set_seccomp: configures current->seccomp.mode
1685 * @seccomp_mode: requested mode to use
1686 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1687 *
1688 * Returns 0 on success or -EINVAL on failure.
1689 */
1690long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
1691{
1692 unsigned int op;
1693 void __user *uargs;
1694
1695 switch (seccomp_mode) {
1696 case SECCOMP_MODE_STRICT:
1697 op = SECCOMP_SET_MODE_STRICT;
1698 /*
1699 * Setting strict mode through prctl always ignored filter,
1700 * so make sure it is always NULL here to pass the internal
1701 * check in do_seccomp().
1702 */
1703 uargs = NULL;
1704 break;
1705 case SECCOMP_MODE_FILTER:
1706 op = SECCOMP_SET_MODE_FILTER;
1707 uargs = filter;
1708 break;
1709 default:
1710 return -EINVAL;
1711 }
1712
1713 /* prctl interface doesn't have flags, so they are always zero. */
1714 return do_seccomp(op, 0, uargs);
1715}
1716
1717#if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
1718static struct seccomp_filter *get_nth_filter(struct task_struct *task,
1719 unsigned long filter_off)
1720{
1721 struct seccomp_filter *orig, *filter;
1722 unsigned long count;
1723
1724 /*
1725 * Note: this is only correct because the caller should be the (ptrace)
1726 * tracer of the task, otherwise lock_task_sighand is needed.
1727 */
1728 spin_lock_irq(&task->sighand->siglock);
1729
1730 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
1731 spin_unlock_irq(&task->sighand->siglock);
1732 return ERR_PTR(-EINVAL);
1733 }
1734
1735 orig = task->seccomp.filter;
1736 __get_seccomp_filter(orig);
1737 spin_unlock_irq(&task->sighand->siglock);
1738
1739 count = 0;
1740 for (filter = orig; filter; filter = filter->prev)
1741 count++;
1742
1743 if (filter_off >= count) {
1744 filter = ERR_PTR(-ENOENT);
1745 goto out;
1746 }
1747
1748 count -= filter_off;
1749 for (filter = orig; filter && count > 1; filter = filter->prev)
1750 count--;
1751
1752 if (WARN_ON(count != 1 || !filter)) {
1753 filter = ERR_PTR(-ENOENT);
1754 goto out;
1755 }
1756
1757 __get_seccomp_filter(filter);
1758
1759out:
1760 __put_seccomp_filter(orig);
1761 return filter;
1762}
1763
1764long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1765 void __user *data)
1766{
1767 struct seccomp_filter *filter;
1768 struct sock_fprog_kern *fprog;
1769 long ret;
1770
1771 if (!capable(CAP_SYS_ADMIN) ||
1772 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1773 return -EACCES;
1774 }
1775
1776 filter = get_nth_filter(task, filter_off);
1777 if (IS_ERR(filter))
1778 return PTR_ERR(filter);
1779
1780 fprog = filter->prog->orig_prog;
1781 if (!fprog) {
1782 /* This must be a new non-cBPF filter, since we save
1783 * every cBPF filter's orig_prog above when
1784 * CONFIG_CHECKPOINT_RESTORE is enabled.
1785 */
1786 ret = -EMEDIUMTYPE;
1787 goto out;
1788 }
1789
1790 ret = fprog->len;
1791 if (!data)
1792 goto out;
1793
1794 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1795 ret = -EFAULT;
1796
1797out:
1798 __put_seccomp_filter(filter);
1799 return ret;
1800}
1801
1802long seccomp_get_metadata(struct task_struct *task,
1803 unsigned long size, void __user *data)
1804{
1805 long ret;
1806 struct seccomp_filter *filter;
1807 struct seccomp_metadata kmd = {};
1808
1809 if (!capable(CAP_SYS_ADMIN) ||
1810 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1811 return -EACCES;
1812 }
1813
1814 size = min_t(unsigned long, size, sizeof(kmd));
1815
1816 if (size < sizeof(kmd.filter_off))
1817 return -EINVAL;
1818
1819 if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
1820 return -EFAULT;
1821
1822 filter = get_nth_filter(task, kmd.filter_off);
1823 if (IS_ERR(filter))
1824 return PTR_ERR(filter);
1825
1826 if (filter->log)
1827 kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1828
1829 ret = size;
1830 if (copy_to_user(data, &kmd, size))
1831 ret = -EFAULT;
1832
1833 __put_seccomp_filter(filter);
1834 return ret;
1835}
1836#endif
1837
1838#ifdef CONFIG_SYSCTL
1839
1840/* Human readable action names for friendly sysctl interaction */
1841#define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
1842#define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
1843#define SECCOMP_RET_TRAP_NAME "trap"
1844#define SECCOMP_RET_ERRNO_NAME "errno"
1845#define SECCOMP_RET_USER_NOTIF_NAME "user_notif"
1846#define SECCOMP_RET_TRACE_NAME "trace"
1847#define SECCOMP_RET_LOG_NAME "log"
1848#define SECCOMP_RET_ALLOW_NAME "allow"
1849
1850static const char seccomp_actions_avail[] =
1851 SECCOMP_RET_KILL_PROCESS_NAME " "
1852 SECCOMP_RET_KILL_THREAD_NAME " "
1853 SECCOMP_RET_TRAP_NAME " "
1854 SECCOMP_RET_ERRNO_NAME " "
1855 SECCOMP_RET_USER_NOTIF_NAME " "
1856 SECCOMP_RET_TRACE_NAME " "
1857 SECCOMP_RET_LOG_NAME " "
1858 SECCOMP_RET_ALLOW_NAME;
1859
1860struct seccomp_log_name {
1861 u32 log;
1862 const char *name;
1863};
1864
1865static const struct seccomp_log_name seccomp_log_names[] = {
1866 { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1867 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1868 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1869 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1870 { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
1871 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1872 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1873 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1874 { }
1875};
1876
1877static bool seccomp_names_from_actions_logged(char *names, size_t size,
1878 u32 actions_logged,
1879 const char *sep)
1880{
1881 const struct seccomp_log_name *cur;
1882 bool append_sep = false;
1883
1884 for (cur = seccomp_log_names; cur->name && size; cur++) {
1885 ssize_t ret;
1886
1887 if (!(actions_logged & cur->log))
1888 continue;
1889
1890 if (append_sep) {
1891 ret = strscpy(names, sep, size);
1892 if (ret < 0)
1893 return false;
1894
1895 names += ret;
1896 size -= ret;
1897 } else
1898 append_sep = true;
1899
1900 ret = strscpy(names, cur->name, size);
1901 if (ret < 0)
1902 return false;
1903
1904 names += ret;
1905 size -= ret;
1906 }
1907
1908 return true;
1909}
1910
1911static bool seccomp_action_logged_from_name(u32 *action_logged,
1912 const char *name)
1913{
1914 const struct seccomp_log_name *cur;
1915
1916 for (cur = seccomp_log_names; cur->name; cur++) {
1917 if (!strcmp(cur->name, name)) {
1918 *action_logged = cur->log;
1919 return true;
1920 }
1921 }
1922
1923 return false;
1924}
1925
1926static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1927{
1928 char *name;
1929
1930 *actions_logged = 0;
1931 while ((name = strsep(&names, " ")) && *name) {
1932 u32 action_logged = 0;
1933
1934 if (!seccomp_action_logged_from_name(&action_logged, name))
1935 return false;
1936
1937 *actions_logged |= action_logged;
1938 }
1939
1940 return true;
1941}
1942
1943static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1944 size_t *lenp, loff_t *ppos)
1945{
1946 char names[sizeof(seccomp_actions_avail)];
1947 struct ctl_table table;
1948
1949 memset(names, 0, sizeof(names));
1950
1951 if (!seccomp_names_from_actions_logged(names, sizeof(names),
1952 seccomp_actions_logged, " "))
1953 return -EINVAL;
1954
1955 table = *ro_table;
1956 table.data = names;
1957 table.maxlen = sizeof(names);
1958 return proc_dostring(&table, 0, buffer, lenp, ppos);
1959}
1960
1961static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1962 size_t *lenp, loff_t *ppos, u32 *actions_logged)
1963{
1964 char names[sizeof(seccomp_actions_avail)];
1965 struct ctl_table table;
1966 int ret;
1967
1968 if (!capable(CAP_SYS_ADMIN))
1969 return -EPERM;
1970
1971 memset(names, 0, sizeof(names));
1972
1973 table = *ro_table;
1974 table.data = names;
1975 table.maxlen = sizeof(names);
1976 ret = proc_dostring(&table, 1, buffer, lenp, ppos);
1977 if (ret)
1978 return ret;
1979
1980 if (!seccomp_actions_logged_from_names(actions_logged, table.data))
1981 return -EINVAL;
1982
1983 if (*actions_logged & SECCOMP_LOG_ALLOW)
1984 return -EINVAL;
1985
1986 seccomp_actions_logged = *actions_logged;
1987 return 0;
1988}
1989
1990static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
1991 int ret)
1992{
1993 char names[sizeof(seccomp_actions_avail)];
1994 char old_names[sizeof(seccomp_actions_avail)];
1995 const char *new = names;
1996 const char *old = old_names;
1997
1998 if (!audit_enabled)
1999 return;
2000
2001 memset(names, 0, sizeof(names));
2002 memset(old_names, 0, sizeof(old_names));
2003
2004 if (ret)
2005 new = "?";
2006 else if (!actions_logged)
2007 new = "(none)";
2008 else if (!seccomp_names_from_actions_logged(names, sizeof(names),
2009 actions_logged, ","))
2010 new = "?";
2011
2012 if (!old_actions_logged)
2013 old = "(none)";
2014 else if (!seccomp_names_from_actions_logged(old_names,
2015 sizeof(old_names),
2016 old_actions_logged, ","))
2017 old = "?";
2018
2019 return audit_seccomp_actions_logged(new, old, !ret);
2020}
2021
2022static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
2023 void *buffer, size_t *lenp,
2024 loff_t *ppos)
2025{
2026 int ret;
2027
2028 if (write) {
2029 u32 actions_logged = 0;
2030 u32 old_actions_logged = seccomp_actions_logged;
2031
2032 ret = write_actions_logged(ro_table, buffer, lenp, ppos,
2033 &actions_logged);
2034 audit_actions_logged(actions_logged, old_actions_logged, ret);
2035 } else
2036 ret = read_actions_logged(ro_table, buffer, lenp, ppos);
2037
2038 return ret;
2039}
2040
2041static struct ctl_path seccomp_sysctl_path[] = {
2042 { .procname = "kernel", },
2043 { .procname = "seccomp", },
2044 { }
2045};
2046
2047static struct ctl_table seccomp_sysctl_table[] = {
2048 {
2049 .procname = "actions_avail",
2050 .data = (void *) &seccomp_actions_avail,
2051 .maxlen = sizeof(seccomp_actions_avail),
2052 .mode = 0444,
2053 .proc_handler = proc_dostring,
2054 },
2055 {
2056 .procname = "actions_logged",
2057 .mode = 0644,
2058 .proc_handler = seccomp_actions_logged_handler,
2059 },
2060 { }
2061};
2062
2063static int __init seccomp_sysctl_init(void)
2064{
2065 struct ctl_table_header *hdr;
2066
2067 hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
2068 if (!hdr)
2069 pr_warn("sysctl registration failed\n");
2070 else
2071 kmemleak_not_leak(hdr);
2072
2073 return 0;
2074}
2075
2076device_initcall(seccomp_sysctl_init)
2077
2078#endif /* CONFIG_SYSCTL */