documentation: RCU grace-period memory ordering guarantees
[linux/fpc-iii.git] / kernel / seccomp.c
blobbb3a38005b9cc3f3b9f80d49c40bb9428e12a20d
1 /*
2 * linux/kernel/seccomp.c
4 * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com>
6 * Copyright (C) 2012 Google, Inc.
7 * Will Drewry <wad@chromium.org>
9 * This defines a simple but solid secure-computing facility.
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.
16 #include <linux/refcount.h>
17 #include <linux/audit.h>
18 #include <linux/compat.h>
19 #include <linux/coredump.h>
20 #include <linux/kmemleak.h>
21 #include <linux/sched.h>
22 #include <linux/sched/task_stack.h>
23 #include <linux/seccomp.h>
24 #include <linux/slab.h>
25 #include <linux/syscalls.h>
26 #include <linux/sysctl.h>
28 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
29 #include <asm/syscall.h>
30 #endif
32 #ifdef CONFIG_SECCOMP_FILTER
33 #include <linux/filter.h>
34 #include <linux/pid.h>
35 #include <linux/ptrace.h>
36 #include <linux/security.h>
37 #include <linux/tracehook.h>
38 #include <linux/uaccess.h>
40 /**
41 * struct seccomp_filter - container for seccomp BPF programs
43 * @usage: reference count to manage the object lifetime.
44 * get/put helpers should be used when accessing an instance
45 * outside of a lifetime-guarded section. In general, this
46 * is only needed for handling filters shared across tasks.
47 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
48 * @prev: points to a previously installed, or inherited, filter
49 * @prog: the BPF program to evaluate
51 * seccomp_filter objects are organized in a tree linked via the @prev
52 * pointer. For any task, it appears to be a singly-linked list starting
53 * with current->seccomp.filter, the most recently attached or inherited filter.
54 * However, multiple filters may share a @prev node, by way of fork(), which
55 * results in a unidirectional tree existing in memory. This is similar to
56 * how namespaces work.
58 * seccomp_filter objects should never be modified after being attached
59 * to a task_struct (other than @usage).
61 struct seccomp_filter {
62 refcount_t usage;
63 bool log;
64 struct seccomp_filter *prev;
65 struct bpf_prog *prog;
68 /* Limit any path through the tree to 256KB worth of instructions. */
69 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
72 * Endianness is explicitly ignored and left for BPF program authors to manage
73 * as per the specific architecture.
75 static void populate_seccomp_data(struct seccomp_data *sd)
77 struct task_struct *task = current;
78 struct pt_regs *regs = task_pt_regs(task);
79 unsigned long args[6];
81 sd->nr = syscall_get_nr(task, regs);
82 sd->arch = syscall_get_arch();
83 syscall_get_arguments(task, regs, 0, 6, args);
84 sd->args[0] = args[0];
85 sd->args[1] = args[1];
86 sd->args[2] = args[2];
87 sd->args[3] = args[3];
88 sd->args[4] = args[4];
89 sd->args[5] = args[5];
90 sd->instruction_pointer = KSTK_EIP(task);
93 /**
94 * seccomp_check_filter - verify seccomp filter code
95 * @filter: filter to verify
96 * @flen: length of filter
98 * Takes a previously checked filter (by bpf_check_classic) and
99 * redirects all filter code that loads struct sk_buff data
100 * and related data through seccomp_bpf_load. It also
101 * enforces length and alignment checking of those loads.
103 * Returns 0 if the rule set is legal or -EINVAL if not.
105 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
107 int pc;
108 for (pc = 0; pc < flen; pc++) {
109 struct sock_filter *ftest = &filter[pc];
110 u16 code = ftest->code;
111 u32 k = ftest->k;
113 switch (code) {
114 case BPF_LD | BPF_W | BPF_ABS:
115 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
116 /* 32-bit aligned and not out of bounds. */
117 if (k >= sizeof(struct seccomp_data) || k & 3)
118 return -EINVAL;
119 continue;
120 case BPF_LD | BPF_W | BPF_LEN:
121 ftest->code = BPF_LD | BPF_IMM;
122 ftest->k = sizeof(struct seccomp_data);
123 continue;
124 case BPF_LDX | BPF_W | BPF_LEN:
125 ftest->code = BPF_LDX | BPF_IMM;
126 ftest->k = sizeof(struct seccomp_data);
127 continue;
128 /* Explicitly include allowed calls. */
129 case BPF_RET | BPF_K:
130 case BPF_RET | BPF_A:
131 case BPF_ALU | BPF_ADD | BPF_K:
132 case BPF_ALU | BPF_ADD | BPF_X:
133 case BPF_ALU | BPF_SUB | BPF_K:
134 case BPF_ALU | BPF_SUB | BPF_X:
135 case BPF_ALU | BPF_MUL | BPF_K:
136 case BPF_ALU | BPF_MUL | BPF_X:
137 case BPF_ALU | BPF_DIV | BPF_K:
138 case BPF_ALU | BPF_DIV | BPF_X:
139 case BPF_ALU | BPF_AND | BPF_K:
140 case BPF_ALU | BPF_AND | BPF_X:
141 case BPF_ALU | BPF_OR | BPF_K:
142 case BPF_ALU | BPF_OR | BPF_X:
143 case BPF_ALU | BPF_XOR | BPF_K:
144 case BPF_ALU | BPF_XOR | BPF_X:
145 case BPF_ALU | BPF_LSH | BPF_K:
146 case BPF_ALU | BPF_LSH | BPF_X:
147 case BPF_ALU | BPF_RSH | BPF_K:
148 case BPF_ALU | BPF_RSH | BPF_X:
149 case BPF_ALU | BPF_NEG:
150 case BPF_LD | BPF_IMM:
151 case BPF_LDX | BPF_IMM:
152 case BPF_MISC | BPF_TAX:
153 case BPF_MISC | BPF_TXA:
154 case BPF_LD | BPF_MEM:
155 case BPF_LDX | BPF_MEM:
156 case BPF_ST:
157 case BPF_STX:
158 case BPF_JMP | BPF_JA:
159 case BPF_JMP | BPF_JEQ | BPF_K:
160 case BPF_JMP | BPF_JEQ | BPF_X:
161 case BPF_JMP | BPF_JGE | BPF_K:
162 case BPF_JMP | BPF_JGE | BPF_X:
163 case BPF_JMP | BPF_JGT | BPF_K:
164 case BPF_JMP | BPF_JGT | BPF_X:
165 case BPF_JMP | BPF_JSET | BPF_K:
166 case BPF_JMP | BPF_JSET | BPF_X:
167 continue;
168 default:
169 return -EINVAL;
172 return 0;
176 * seccomp_run_filters - evaluates all seccomp filters against @sd
177 * @sd: optional seccomp data to be passed to filters
178 * @match: stores struct seccomp_filter that resulted in the return value,
179 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
180 * be unchanged.
182 * Returns valid seccomp BPF response codes.
184 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
185 static u32 seccomp_run_filters(const struct seccomp_data *sd,
186 struct seccomp_filter **match)
188 struct seccomp_data sd_local;
189 u32 ret = SECCOMP_RET_ALLOW;
190 /* Make sure cross-thread synced filter points somewhere sane. */
191 struct seccomp_filter *f =
192 lockless_dereference(current->seccomp.filter);
194 /* Ensure unexpected behavior doesn't result in failing open. */
195 if (unlikely(WARN_ON(f == NULL)))
196 return SECCOMP_RET_KILL_PROCESS;
198 if (!sd) {
199 populate_seccomp_data(&sd_local);
200 sd = &sd_local;
204 * All filters in the list are evaluated and the lowest BPF return
205 * value always takes priority (ignoring the DATA).
207 for (; f; f = f->prev) {
208 u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
210 if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
211 ret = cur_ret;
212 *match = f;
215 return ret;
217 #endif /* CONFIG_SECCOMP_FILTER */
219 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
221 assert_spin_locked(&current->sighand->siglock);
223 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
224 return false;
226 return true;
229 static inline void seccomp_assign_mode(struct task_struct *task,
230 unsigned long seccomp_mode)
232 assert_spin_locked(&task->sighand->siglock);
234 task->seccomp.mode = seccomp_mode;
236 * Make sure TIF_SECCOMP cannot be set before the mode (and
237 * filter) is set.
239 smp_mb__before_atomic();
240 set_tsk_thread_flag(task, TIF_SECCOMP);
243 #ifdef CONFIG_SECCOMP_FILTER
244 /* Returns 1 if the parent is an ancestor of the child. */
245 static int is_ancestor(struct seccomp_filter *parent,
246 struct seccomp_filter *child)
248 /* NULL is the root ancestor. */
249 if (parent == NULL)
250 return 1;
251 for (; child; child = child->prev)
252 if (child == parent)
253 return 1;
254 return 0;
258 * seccomp_can_sync_threads: checks if all threads can be synchronized
260 * Expects sighand and cred_guard_mutex locks to be held.
262 * Returns 0 on success, -ve on error, or the pid of a thread which was
263 * either not in the correct seccomp mode or it did not have an ancestral
264 * seccomp filter.
266 static inline pid_t seccomp_can_sync_threads(void)
268 struct task_struct *thread, *caller;
270 BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
271 assert_spin_locked(&current->sighand->siglock);
273 /* Validate all threads being eligible for synchronization. */
274 caller = current;
275 for_each_thread(caller, thread) {
276 pid_t failed;
278 /* Skip current, since it is initiating the sync. */
279 if (thread == caller)
280 continue;
282 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
283 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
284 is_ancestor(thread->seccomp.filter,
285 caller->seccomp.filter)))
286 continue;
288 /* Return the first thread that cannot be synchronized. */
289 failed = task_pid_vnr(thread);
290 /* If the pid cannot be resolved, then return -ESRCH */
291 if (unlikely(WARN_ON(failed == 0)))
292 failed = -ESRCH;
293 return failed;
296 return 0;
300 * seccomp_sync_threads: sets all threads to use current's filter
302 * Expects sighand and cred_guard_mutex locks to be held, and for
303 * seccomp_can_sync_threads() to have returned success already
304 * without dropping the locks.
307 static inline void seccomp_sync_threads(void)
309 struct task_struct *thread, *caller;
311 BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
312 assert_spin_locked(&current->sighand->siglock);
314 /* Synchronize all threads. */
315 caller = current;
316 for_each_thread(caller, thread) {
317 /* Skip current, since it needs no changes. */
318 if (thread == caller)
319 continue;
321 /* Get a task reference for the new leaf node. */
322 get_seccomp_filter(caller);
324 * Drop the task reference to the shared ancestor since
325 * current's path will hold a reference. (This also
326 * allows a put before the assignment.)
328 put_seccomp_filter(thread);
329 smp_store_release(&thread->seccomp.filter,
330 caller->seccomp.filter);
333 * Don't let an unprivileged task work around
334 * the no_new_privs restriction by creating
335 * a thread that sets it up, enters seccomp,
336 * then dies.
338 if (task_no_new_privs(caller))
339 task_set_no_new_privs(thread);
342 * Opt the other thread into seccomp if needed.
343 * As threads are considered to be trust-realm
344 * equivalent (see ptrace_may_access), it is safe to
345 * allow one thread to transition the other.
347 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
348 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER);
353 * seccomp_prepare_filter: Prepares a seccomp filter for use.
354 * @fprog: BPF program to install
356 * Returns filter on success or an ERR_PTR on failure.
358 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
360 struct seccomp_filter *sfilter;
361 int ret;
362 const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
364 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
365 return ERR_PTR(-EINVAL);
367 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
370 * Installing a seccomp filter requires that the task has
371 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
372 * This avoids scenarios where unprivileged tasks can affect the
373 * behavior of privileged children.
375 if (!task_no_new_privs(current) &&
376 security_capable_noaudit(current_cred(), current_user_ns(),
377 CAP_SYS_ADMIN) != 0)
378 return ERR_PTR(-EACCES);
380 /* Allocate a new seccomp_filter */
381 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
382 if (!sfilter)
383 return ERR_PTR(-ENOMEM);
385 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
386 seccomp_check_filter, save_orig);
387 if (ret < 0) {
388 kfree(sfilter);
389 return ERR_PTR(ret);
392 refcount_set(&sfilter->usage, 1);
394 return sfilter;
398 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
399 * @user_filter: pointer to the user data containing a sock_fprog.
401 * Returns 0 on success and non-zero otherwise.
403 static struct seccomp_filter *
404 seccomp_prepare_user_filter(const char __user *user_filter)
406 struct sock_fprog fprog;
407 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
409 #ifdef CONFIG_COMPAT
410 if (in_compat_syscall()) {
411 struct compat_sock_fprog fprog32;
412 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
413 goto out;
414 fprog.len = fprog32.len;
415 fprog.filter = compat_ptr(fprog32.filter);
416 } else /* falls through to the if below. */
417 #endif
418 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
419 goto out;
420 filter = seccomp_prepare_filter(&fprog);
421 out:
422 return filter;
426 * seccomp_attach_filter: validate and attach filter
427 * @flags: flags to change filter behavior
428 * @filter: seccomp filter to add to the current process
430 * Caller must be holding current->sighand->siglock lock.
432 * Returns 0 on success, -ve on error.
434 static long seccomp_attach_filter(unsigned int flags,
435 struct seccomp_filter *filter)
437 unsigned long total_insns;
438 struct seccomp_filter *walker;
440 assert_spin_locked(&current->sighand->siglock);
442 /* Validate resulting filter length. */
443 total_insns = filter->prog->len;
444 for (walker = current->seccomp.filter; walker; walker = walker->prev)
445 total_insns += walker->prog->len + 4; /* 4 instr penalty */
446 if (total_insns > MAX_INSNS_PER_PATH)
447 return -ENOMEM;
449 /* If thread sync has been requested, check that it is possible. */
450 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
451 int ret;
453 ret = seccomp_can_sync_threads();
454 if (ret)
455 return ret;
458 /* Set log flag, if present. */
459 if (flags & SECCOMP_FILTER_FLAG_LOG)
460 filter->log = true;
463 * If there is an existing filter, make it the prev and don't drop its
464 * task reference.
466 filter->prev = current->seccomp.filter;
467 current->seccomp.filter = filter;
469 /* Now that the new filter is in place, synchronize to all threads. */
470 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
471 seccomp_sync_threads();
473 return 0;
476 void __get_seccomp_filter(struct seccomp_filter *filter)
478 /* Reference count is bounded by the number of total processes. */
479 refcount_inc(&filter->usage);
482 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
483 void get_seccomp_filter(struct task_struct *tsk)
485 struct seccomp_filter *orig = tsk->seccomp.filter;
486 if (!orig)
487 return;
488 __get_seccomp_filter(orig);
491 static inline void seccomp_filter_free(struct seccomp_filter *filter)
493 if (filter) {
494 bpf_prog_destroy(filter->prog);
495 kfree(filter);
499 static void __put_seccomp_filter(struct seccomp_filter *orig)
501 /* Clean up single-reference branches iteratively. */
502 while (orig && refcount_dec_and_test(&orig->usage)) {
503 struct seccomp_filter *freeme = orig;
504 orig = orig->prev;
505 seccomp_filter_free(freeme);
509 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
510 void put_seccomp_filter(struct task_struct *tsk)
512 __put_seccomp_filter(tsk->seccomp.filter);
515 static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason)
517 memset(info, 0, sizeof(*info));
518 info->si_signo = SIGSYS;
519 info->si_code = SYS_SECCOMP;
520 info->si_call_addr = (void __user *)KSTK_EIP(current);
521 info->si_errno = reason;
522 info->si_arch = syscall_get_arch();
523 info->si_syscall = syscall;
527 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
528 * @syscall: syscall number to send to userland
529 * @reason: filter-supplied reason code to send to userland (via si_errno)
531 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
533 static void seccomp_send_sigsys(int syscall, int reason)
535 struct siginfo info;
536 seccomp_init_siginfo(&info, syscall, reason);
537 force_sig_info(SIGSYS, &info, current);
539 #endif /* CONFIG_SECCOMP_FILTER */
541 /* For use with seccomp_actions_logged */
542 #define SECCOMP_LOG_KILL_PROCESS (1 << 0)
543 #define SECCOMP_LOG_KILL_THREAD (1 << 1)
544 #define SECCOMP_LOG_TRAP (1 << 2)
545 #define SECCOMP_LOG_ERRNO (1 << 3)
546 #define SECCOMP_LOG_TRACE (1 << 4)
547 #define SECCOMP_LOG_LOG (1 << 5)
548 #define SECCOMP_LOG_ALLOW (1 << 6)
550 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
551 SECCOMP_LOG_KILL_THREAD |
552 SECCOMP_LOG_TRAP |
553 SECCOMP_LOG_ERRNO |
554 SECCOMP_LOG_TRACE |
555 SECCOMP_LOG_LOG;
557 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
558 bool requested)
560 bool log = false;
562 switch (action) {
563 case SECCOMP_RET_ALLOW:
564 break;
565 case SECCOMP_RET_TRAP:
566 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
567 break;
568 case SECCOMP_RET_ERRNO:
569 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
570 break;
571 case SECCOMP_RET_TRACE:
572 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
573 break;
574 case SECCOMP_RET_LOG:
575 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
576 break;
577 case SECCOMP_RET_KILL_THREAD:
578 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
579 break;
580 case SECCOMP_RET_KILL_PROCESS:
581 default:
582 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
586 * Force an audit message to be emitted when the action is RET_KILL_*,
587 * RET_LOG, or the FILTER_FLAG_LOG bit was set and the action is
588 * allowed to be logged by the admin.
590 if (log)
591 return __audit_seccomp(syscall, signr, action);
594 * Let the audit subsystem decide if the action should be audited based
595 * on whether the current task itself is being audited.
597 return audit_seccomp(syscall, signr, action);
601 * Secure computing mode 1 allows only read/write/exit/sigreturn.
602 * To be fully secure this must be combined with rlimit
603 * to limit the stack allocations too.
605 static const int mode1_syscalls[] = {
606 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
607 0, /* null terminated */
610 static void __secure_computing_strict(int this_syscall)
612 const int *syscall_whitelist = mode1_syscalls;
613 #ifdef CONFIG_COMPAT
614 if (in_compat_syscall())
615 syscall_whitelist = get_compat_mode1_syscalls();
616 #endif
617 do {
618 if (*syscall_whitelist == this_syscall)
619 return;
620 } while (*++syscall_whitelist);
622 #ifdef SECCOMP_DEBUG
623 dump_stack();
624 #endif
625 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
626 do_exit(SIGKILL);
629 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
630 void secure_computing_strict(int this_syscall)
632 int mode = current->seccomp.mode;
634 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
635 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
636 return;
638 if (mode == SECCOMP_MODE_DISABLED)
639 return;
640 else if (mode == SECCOMP_MODE_STRICT)
641 __secure_computing_strict(this_syscall);
642 else
643 BUG();
645 #else
647 #ifdef CONFIG_SECCOMP_FILTER
648 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
649 const bool recheck_after_trace)
651 u32 filter_ret, action;
652 struct seccomp_filter *match = NULL;
653 int data;
656 * Make sure that any changes to mode from another thread have
657 * been seen after TIF_SECCOMP was seen.
659 rmb();
661 filter_ret = seccomp_run_filters(sd, &match);
662 data = filter_ret & SECCOMP_RET_DATA;
663 action = filter_ret & SECCOMP_RET_ACTION_FULL;
665 switch (action) {
666 case SECCOMP_RET_ERRNO:
667 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
668 if (data > MAX_ERRNO)
669 data = MAX_ERRNO;
670 syscall_set_return_value(current, task_pt_regs(current),
671 -data, 0);
672 goto skip;
674 case SECCOMP_RET_TRAP:
675 /* Show the handler the original registers. */
676 syscall_rollback(current, task_pt_regs(current));
677 /* Let the filter pass back 16 bits of data. */
678 seccomp_send_sigsys(this_syscall, data);
679 goto skip;
681 case SECCOMP_RET_TRACE:
682 /* We've been put in this state by the ptracer already. */
683 if (recheck_after_trace)
684 return 0;
686 /* ENOSYS these calls if there is no tracer attached. */
687 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
688 syscall_set_return_value(current,
689 task_pt_regs(current),
690 -ENOSYS, 0);
691 goto skip;
694 /* Allow the BPF to provide the event message */
695 ptrace_event(PTRACE_EVENT_SECCOMP, data);
697 * The delivery of a fatal signal during event
698 * notification may silently skip tracer notification,
699 * which could leave us with a potentially unmodified
700 * syscall that the tracer would have liked to have
701 * changed. Since the process is about to die, we just
702 * force the syscall to be skipped and let the signal
703 * kill the process and correctly handle any tracer exit
704 * notifications.
706 if (fatal_signal_pending(current))
707 goto skip;
708 /* Check if the tracer forced the syscall to be skipped. */
709 this_syscall = syscall_get_nr(current, task_pt_regs(current));
710 if (this_syscall < 0)
711 goto skip;
714 * Recheck the syscall, since it may have changed. This
715 * intentionally uses a NULL struct seccomp_data to force
716 * a reload of all registers. This does not goto skip since
717 * a skip would have already been reported.
719 if (__seccomp_filter(this_syscall, NULL, true))
720 return -1;
722 return 0;
724 case SECCOMP_RET_LOG:
725 seccomp_log(this_syscall, 0, action, true);
726 return 0;
728 case SECCOMP_RET_ALLOW:
730 * Note that the "match" filter will always be NULL for
731 * this action since SECCOMP_RET_ALLOW is the starting
732 * state in seccomp_run_filters().
734 return 0;
736 case SECCOMP_RET_KILL_THREAD:
737 case SECCOMP_RET_KILL_PROCESS:
738 default:
739 seccomp_log(this_syscall, SIGSYS, action, true);
740 /* Dump core only if this is the last remaining thread. */
741 if (action == SECCOMP_RET_KILL_PROCESS ||
742 get_nr_threads(current) == 1) {
743 siginfo_t info;
745 /* Show the original registers in the dump. */
746 syscall_rollback(current, task_pt_regs(current));
747 /* Trigger a manual coredump since do_exit skips it. */
748 seccomp_init_siginfo(&info, this_syscall, data);
749 do_coredump(&info);
751 if (action == SECCOMP_RET_KILL_PROCESS)
752 do_group_exit(SIGSYS);
753 else
754 do_exit(SIGSYS);
757 unreachable();
759 skip:
760 seccomp_log(this_syscall, 0, action, match ? match->log : false);
761 return -1;
763 #else
764 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
765 const bool recheck_after_trace)
767 BUG();
769 #endif
771 int __secure_computing(const struct seccomp_data *sd)
773 int mode = current->seccomp.mode;
774 int this_syscall;
776 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
777 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
778 return 0;
780 this_syscall = sd ? sd->nr :
781 syscall_get_nr(current, task_pt_regs(current));
783 switch (mode) {
784 case SECCOMP_MODE_STRICT:
785 __secure_computing_strict(this_syscall); /* may call do_exit */
786 return 0;
787 case SECCOMP_MODE_FILTER:
788 return __seccomp_filter(this_syscall, sd, false);
789 default:
790 BUG();
793 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
795 long prctl_get_seccomp(void)
797 return current->seccomp.mode;
801 * seccomp_set_mode_strict: internal function for setting strict seccomp
803 * Once current->seccomp.mode is non-zero, it may not be changed.
805 * Returns 0 on success or -EINVAL on failure.
807 static long seccomp_set_mode_strict(void)
809 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
810 long ret = -EINVAL;
812 spin_lock_irq(&current->sighand->siglock);
814 if (!seccomp_may_assign_mode(seccomp_mode))
815 goto out;
817 #ifdef TIF_NOTSC
818 disable_TSC();
819 #endif
820 seccomp_assign_mode(current, seccomp_mode);
821 ret = 0;
823 out:
824 spin_unlock_irq(&current->sighand->siglock);
826 return ret;
829 #ifdef CONFIG_SECCOMP_FILTER
831 * seccomp_set_mode_filter: internal function for setting seccomp filter
832 * @flags: flags to change filter behavior
833 * @filter: struct sock_fprog containing filter
835 * This function may be called repeatedly to install additional filters.
836 * Every filter successfully installed will be evaluated (in reverse order)
837 * for each system call the task makes.
839 * Once current->seccomp.mode is non-zero, it may not be changed.
841 * Returns 0 on success or -EINVAL on failure.
843 static long seccomp_set_mode_filter(unsigned int flags,
844 const char __user *filter)
846 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
847 struct seccomp_filter *prepared = NULL;
848 long ret = -EINVAL;
850 /* Validate flags. */
851 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
852 return -EINVAL;
854 /* Prepare the new filter before holding any locks. */
855 prepared = seccomp_prepare_user_filter(filter);
856 if (IS_ERR(prepared))
857 return PTR_ERR(prepared);
860 * Make sure we cannot change seccomp or nnp state via TSYNC
861 * while another thread is in the middle of calling exec.
863 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
864 mutex_lock_killable(&current->signal->cred_guard_mutex))
865 goto out_free;
867 spin_lock_irq(&current->sighand->siglock);
869 if (!seccomp_may_assign_mode(seccomp_mode))
870 goto out;
872 ret = seccomp_attach_filter(flags, prepared);
873 if (ret)
874 goto out;
875 /* Do not free the successfully attached filter. */
876 prepared = NULL;
878 seccomp_assign_mode(current, seccomp_mode);
879 out:
880 spin_unlock_irq(&current->sighand->siglock);
881 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
882 mutex_unlock(&current->signal->cred_guard_mutex);
883 out_free:
884 seccomp_filter_free(prepared);
885 return ret;
887 #else
888 static inline long seccomp_set_mode_filter(unsigned int flags,
889 const char __user *filter)
891 return -EINVAL;
893 #endif
895 static long seccomp_get_action_avail(const char __user *uaction)
897 u32 action;
899 if (copy_from_user(&action, uaction, sizeof(action)))
900 return -EFAULT;
902 switch (action) {
903 case SECCOMP_RET_KILL_PROCESS:
904 case SECCOMP_RET_KILL_THREAD:
905 case SECCOMP_RET_TRAP:
906 case SECCOMP_RET_ERRNO:
907 case SECCOMP_RET_TRACE:
908 case SECCOMP_RET_LOG:
909 case SECCOMP_RET_ALLOW:
910 break;
911 default:
912 return -EOPNOTSUPP;
915 return 0;
918 /* Common entry point for both prctl and syscall. */
919 static long do_seccomp(unsigned int op, unsigned int flags,
920 const char __user *uargs)
922 switch (op) {
923 case SECCOMP_SET_MODE_STRICT:
924 if (flags != 0 || uargs != NULL)
925 return -EINVAL;
926 return seccomp_set_mode_strict();
927 case SECCOMP_SET_MODE_FILTER:
928 return seccomp_set_mode_filter(flags, uargs);
929 case SECCOMP_GET_ACTION_AVAIL:
930 if (flags != 0)
931 return -EINVAL;
933 return seccomp_get_action_avail(uargs);
934 default:
935 return -EINVAL;
939 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
940 const char __user *, uargs)
942 return do_seccomp(op, flags, uargs);
946 * prctl_set_seccomp: configures current->seccomp.mode
947 * @seccomp_mode: requested mode to use
948 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
950 * Returns 0 on success or -EINVAL on failure.
952 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
954 unsigned int op;
955 char __user *uargs;
957 switch (seccomp_mode) {
958 case SECCOMP_MODE_STRICT:
959 op = SECCOMP_SET_MODE_STRICT;
961 * Setting strict mode through prctl always ignored filter,
962 * so make sure it is always NULL here to pass the internal
963 * check in do_seccomp().
965 uargs = NULL;
966 break;
967 case SECCOMP_MODE_FILTER:
968 op = SECCOMP_SET_MODE_FILTER;
969 uargs = filter;
970 break;
971 default:
972 return -EINVAL;
975 /* prctl interface doesn't have flags, so they are always zero. */
976 return do_seccomp(op, 0, uargs);
979 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
980 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
981 void __user *data)
983 struct seccomp_filter *filter;
984 struct sock_fprog_kern *fprog;
985 long ret;
986 unsigned long count = 0;
988 if (!capable(CAP_SYS_ADMIN) ||
989 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
990 return -EACCES;
993 spin_lock_irq(&task->sighand->siglock);
994 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
995 ret = -EINVAL;
996 goto out;
999 filter = task->seccomp.filter;
1000 while (filter) {
1001 filter = filter->prev;
1002 count++;
1005 if (filter_off >= count) {
1006 ret = -ENOENT;
1007 goto out;
1009 count -= filter_off;
1011 filter = task->seccomp.filter;
1012 while (filter && count > 1) {
1013 filter = filter->prev;
1014 count--;
1017 if (WARN_ON(count != 1 || !filter)) {
1018 /* The filter tree shouldn't shrink while we're using it. */
1019 ret = -ENOENT;
1020 goto out;
1023 fprog = filter->prog->orig_prog;
1024 if (!fprog) {
1025 /* This must be a new non-cBPF filter, since we save
1026 * every cBPF filter's orig_prog above when
1027 * CONFIG_CHECKPOINT_RESTORE is enabled.
1029 ret = -EMEDIUMTYPE;
1030 goto out;
1033 ret = fprog->len;
1034 if (!data)
1035 goto out;
1037 __get_seccomp_filter(filter);
1038 spin_unlock_irq(&task->sighand->siglock);
1040 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1041 ret = -EFAULT;
1043 __put_seccomp_filter(filter);
1044 return ret;
1046 out:
1047 spin_unlock_irq(&task->sighand->siglock);
1048 return ret;
1050 #endif
1052 #ifdef CONFIG_SYSCTL
1054 /* Human readable action names for friendly sysctl interaction */
1055 #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
1056 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
1057 #define SECCOMP_RET_TRAP_NAME "trap"
1058 #define SECCOMP_RET_ERRNO_NAME "errno"
1059 #define SECCOMP_RET_TRACE_NAME "trace"
1060 #define SECCOMP_RET_LOG_NAME "log"
1061 #define SECCOMP_RET_ALLOW_NAME "allow"
1063 static const char seccomp_actions_avail[] =
1064 SECCOMP_RET_KILL_PROCESS_NAME " "
1065 SECCOMP_RET_KILL_THREAD_NAME " "
1066 SECCOMP_RET_TRAP_NAME " "
1067 SECCOMP_RET_ERRNO_NAME " "
1068 SECCOMP_RET_TRACE_NAME " "
1069 SECCOMP_RET_LOG_NAME " "
1070 SECCOMP_RET_ALLOW_NAME;
1072 struct seccomp_log_name {
1073 u32 log;
1074 const char *name;
1077 static const struct seccomp_log_name seccomp_log_names[] = {
1078 { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1079 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1080 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1081 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1082 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1083 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1084 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1088 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1089 u32 actions_logged)
1091 const struct seccomp_log_name *cur;
1092 bool append_space = false;
1094 for (cur = seccomp_log_names; cur->name && size; cur++) {
1095 ssize_t ret;
1097 if (!(actions_logged & cur->log))
1098 continue;
1100 if (append_space) {
1101 ret = strscpy(names, " ", size);
1102 if (ret < 0)
1103 return false;
1105 names += ret;
1106 size -= ret;
1107 } else
1108 append_space = true;
1110 ret = strscpy(names, cur->name, size);
1111 if (ret < 0)
1112 return false;
1114 names += ret;
1115 size -= ret;
1118 return true;
1121 static bool seccomp_action_logged_from_name(u32 *action_logged,
1122 const char *name)
1124 const struct seccomp_log_name *cur;
1126 for (cur = seccomp_log_names; cur->name; cur++) {
1127 if (!strcmp(cur->name, name)) {
1128 *action_logged = cur->log;
1129 return true;
1133 return false;
1136 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1138 char *name;
1140 *actions_logged = 0;
1141 while ((name = strsep(&names, " ")) && *name) {
1142 u32 action_logged = 0;
1144 if (!seccomp_action_logged_from_name(&action_logged, name))
1145 return false;
1147 *actions_logged |= action_logged;
1150 return true;
1153 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1154 void __user *buffer, size_t *lenp,
1155 loff_t *ppos)
1157 char names[sizeof(seccomp_actions_avail)];
1158 struct ctl_table table;
1159 int ret;
1161 if (write && !capable(CAP_SYS_ADMIN))
1162 return -EPERM;
1164 memset(names, 0, sizeof(names));
1166 if (!write) {
1167 if (!seccomp_names_from_actions_logged(names, sizeof(names),
1168 seccomp_actions_logged))
1169 return -EINVAL;
1172 table = *ro_table;
1173 table.data = names;
1174 table.maxlen = sizeof(names);
1175 ret = proc_dostring(&table, write, buffer, lenp, ppos);
1176 if (ret)
1177 return ret;
1179 if (write) {
1180 u32 actions_logged;
1182 if (!seccomp_actions_logged_from_names(&actions_logged,
1183 table.data))
1184 return -EINVAL;
1186 if (actions_logged & SECCOMP_LOG_ALLOW)
1187 return -EINVAL;
1189 seccomp_actions_logged = actions_logged;
1192 return 0;
1195 static struct ctl_path seccomp_sysctl_path[] = {
1196 { .procname = "kernel", },
1197 { .procname = "seccomp", },
1201 static struct ctl_table seccomp_sysctl_table[] = {
1203 .procname = "actions_avail",
1204 .data = (void *) &seccomp_actions_avail,
1205 .maxlen = sizeof(seccomp_actions_avail),
1206 .mode = 0444,
1207 .proc_handler = proc_dostring,
1210 .procname = "actions_logged",
1211 .mode = 0644,
1212 .proc_handler = seccomp_actions_logged_handler,
1217 static int __init seccomp_sysctl_init(void)
1219 struct ctl_table_header *hdr;
1221 hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1222 if (!hdr)
1223 pr_warn("seccomp: sysctl registration failed\n");
1224 else
1225 kmemleak_not_leak(hdr);
1227 return 0;
1230 device_initcall(seccomp_sysctl_init)
1232 #endif /* CONFIG_SYSCTL */