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/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>
24 /* #define SECCOMP_DEBUG 1 */
26 #ifdef CONFIG_SECCOMP_FILTER
27 #include <asm/syscall.h>
28 #include <linux/filter.h>
29 #include <linux/pid.h>
30 #include <linux/ptrace.h>
31 #include <linux/security.h>
32 #include <linux/tracehook.h>
33 #include <linux/uaccess.h>
36 * struct seccomp_filter - container for seccomp BPF programs
38 * @usage: reference count to manage the object lifetime.
39 * get/put helpers should be used when accessing an instance
40 * outside of a lifetime-guarded section. In general, this
41 * is only needed for handling filters shared across tasks.
42 * @prev: points to a previously installed, or inherited, filter
43 * @len: the number of instructions in the program
44 * @insnsi: the BPF program instructions to evaluate
46 * seccomp_filter objects are organized in a tree linked via the @prev
47 * pointer. For any task, it appears to be a singly-linked list starting
48 * with current->seccomp.filter, the most recently attached or inherited filter.
49 * However, multiple filters may share a @prev node, by way of fork(), which
50 * results in a unidirectional tree existing in memory. This is similar to
51 * how namespaces work.
53 * seccomp_filter objects should never be modified after being attached
54 * to a task_struct (other than @usage).
56 struct seccomp_filter {
58 struct seccomp_filter *prev;
59 struct bpf_prog *prog;
62 /* Limit any path through the tree to 256KB worth of instructions. */
63 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
66 * Endianness is explicitly ignored and left for BPF program authors to manage
67 * as per the specific architecture.
69 static void populate_seccomp_data(struct seccomp_data *sd)
71 struct task_struct *task = current;
72 struct pt_regs *regs = task_pt_regs(task);
73 unsigned long args[6];
75 sd->nr = syscall_get_nr(task, regs);
76 sd->arch = syscall_get_arch();
77 syscall_get_arguments(task, regs, 0, 6, args);
78 sd->args[0] = args[0];
79 sd->args[1] = args[1];
80 sd->args[2] = args[2];
81 sd->args[3] = args[3];
82 sd->args[4] = args[4];
83 sd->args[5] = args[5];
84 sd->instruction_pointer = KSTK_EIP(task);
88 * seccomp_check_filter - verify seccomp filter code
89 * @filter: filter to verify
90 * @flen: length of filter
92 * Takes a previously checked filter (by bpf_check_classic) and
93 * redirects all filter code that loads struct sk_buff data
94 * and related data through seccomp_bpf_load. It also
95 * enforces length and alignment checking of those loads.
97 * Returns 0 if the rule set is legal or -EINVAL if not.
99 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
102 for (pc = 0; pc < flen; pc++) {
103 struct sock_filter *ftest = &filter[pc];
104 u16 code = ftest->code;
108 case BPF_LD | BPF_W | BPF_ABS:
109 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
110 /* 32-bit aligned and not out of bounds. */
111 if (k >= sizeof(struct seccomp_data) || k & 3)
114 case BPF_LD | BPF_W | BPF_LEN:
115 ftest->code = BPF_LD | BPF_IMM;
116 ftest->k = sizeof(struct seccomp_data);
118 case BPF_LDX | BPF_W | BPF_LEN:
119 ftest->code = BPF_LDX | BPF_IMM;
120 ftest->k = sizeof(struct seccomp_data);
122 /* Explicitly include allowed calls. */
123 case BPF_RET | BPF_K:
124 case BPF_RET | BPF_A:
125 case BPF_ALU | BPF_ADD | BPF_K:
126 case BPF_ALU | BPF_ADD | BPF_X:
127 case BPF_ALU | BPF_SUB | BPF_K:
128 case BPF_ALU | BPF_SUB | BPF_X:
129 case BPF_ALU | BPF_MUL | BPF_K:
130 case BPF_ALU | BPF_MUL | BPF_X:
131 case BPF_ALU | BPF_DIV | BPF_K:
132 case BPF_ALU | BPF_DIV | BPF_X:
133 case BPF_ALU | BPF_AND | BPF_K:
134 case BPF_ALU | BPF_AND | BPF_X:
135 case BPF_ALU | BPF_OR | BPF_K:
136 case BPF_ALU | BPF_OR | BPF_X:
137 case BPF_ALU | BPF_XOR | BPF_K:
138 case BPF_ALU | BPF_XOR | BPF_X:
139 case BPF_ALU | BPF_LSH | BPF_K:
140 case BPF_ALU | BPF_LSH | BPF_X:
141 case BPF_ALU | BPF_RSH | BPF_K:
142 case BPF_ALU | BPF_RSH | BPF_X:
143 case BPF_ALU | BPF_NEG:
144 case BPF_LD | BPF_IMM:
145 case BPF_LDX | BPF_IMM:
146 case BPF_MISC | BPF_TAX:
147 case BPF_MISC | BPF_TXA:
148 case BPF_LD | BPF_MEM:
149 case BPF_LDX | BPF_MEM:
152 case BPF_JMP | BPF_JA:
153 case BPF_JMP | BPF_JEQ | BPF_K:
154 case BPF_JMP | BPF_JEQ | BPF_X:
155 case BPF_JMP | BPF_JGE | BPF_K:
156 case BPF_JMP | BPF_JGE | BPF_X:
157 case BPF_JMP | BPF_JGT | BPF_K:
158 case BPF_JMP | BPF_JGT | BPF_X:
159 case BPF_JMP | BPF_JSET | BPF_K:
160 case BPF_JMP | BPF_JSET | BPF_X:
170 * seccomp_run_filters - evaluates all seccomp filters against @syscall
171 * @syscall: number of the current system call
173 * Returns valid seccomp BPF response codes.
175 static u32 seccomp_run_filters(int syscall)
177 struct seccomp_filter *f = ACCESS_ONCE(current->seccomp.filter);
178 struct seccomp_data sd;
179 u32 ret = SECCOMP_RET_ALLOW;
181 /* Ensure unexpected behavior doesn't result in failing open. */
182 if (unlikely(WARN_ON(f == NULL)))
183 return SECCOMP_RET_KILL;
185 /* Make sure cross-thread synced filter points somewhere sane. */
186 smp_read_barrier_depends();
188 populate_seccomp_data(&sd);
191 * All filters in the list are evaluated and the lowest BPF return
192 * value always takes priority (ignoring the DATA).
194 for (; f; f = f->prev) {
195 u32 cur_ret = BPF_PROG_RUN(f->prog, (void *)&sd);
197 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
202 #endif /* CONFIG_SECCOMP_FILTER */
204 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
206 assert_spin_locked(¤t->sighand->siglock);
208 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
214 static inline void seccomp_assign_mode(struct task_struct *task,
215 unsigned long seccomp_mode)
217 assert_spin_locked(&task->sighand->siglock);
219 task->seccomp.mode = seccomp_mode;
221 * Make sure TIF_SECCOMP cannot be set before the mode (and
224 smp_mb__before_atomic();
225 set_tsk_thread_flag(task, TIF_SECCOMP);
228 #ifdef CONFIG_SECCOMP_FILTER
229 /* Returns 1 if the parent is an ancestor of the child. */
230 static int is_ancestor(struct seccomp_filter *parent,
231 struct seccomp_filter *child)
233 /* NULL is the root ancestor. */
236 for (; child; child = child->prev)
243 * seccomp_can_sync_threads: checks if all threads can be synchronized
245 * Expects sighand and cred_guard_mutex locks to be held.
247 * Returns 0 on success, -ve on error, or the pid of a thread which was
248 * either not in the correct seccomp mode or it did not have an ancestral
251 static inline pid_t seccomp_can_sync_threads(void)
253 struct task_struct *thread, *caller;
255 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
256 assert_spin_locked(¤t->sighand->siglock);
258 /* Validate all threads being eligible for synchronization. */
260 for_each_thread(caller, thread) {
263 /* Skip current, since it is initiating the sync. */
264 if (thread == caller)
267 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
268 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
269 is_ancestor(thread->seccomp.filter,
270 caller->seccomp.filter)))
273 /* Return the first thread that cannot be synchronized. */
274 failed = task_pid_vnr(thread);
275 /* If the pid cannot be resolved, then return -ESRCH */
276 if (unlikely(WARN_ON(failed == 0)))
285 * seccomp_sync_threads: sets all threads to use current's filter
287 * Expects sighand and cred_guard_mutex locks to be held, and for
288 * seccomp_can_sync_threads() to have returned success already
289 * without dropping the locks.
292 static inline void seccomp_sync_threads(void)
294 struct task_struct *thread, *caller;
296 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
297 assert_spin_locked(¤t->sighand->siglock);
299 /* Synchronize all threads. */
301 for_each_thread(caller, thread) {
302 /* Skip current, since it needs no changes. */
303 if (thread == caller)
306 /* Get a task reference for the new leaf node. */
307 get_seccomp_filter(caller);
309 * Drop the task reference to the shared ancestor since
310 * current's path will hold a reference. (This also
311 * allows a put before the assignment.)
313 put_seccomp_filter(thread);
314 smp_store_release(&thread->seccomp.filter,
315 caller->seccomp.filter);
317 * Opt the other thread into seccomp if needed.
318 * As threads are considered to be trust-realm
319 * equivalent (see ptrace_may_access), it is safe to
320 * allow one thread to transition the other.
322 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) {
324 * Don't let an unprivileged task work around
325 * the no_new_privs restriction by creating
326 * a thread that sets it up, enters seccomp,
329 if (task_no_new_privs(caller))
330 task_set_no_new_privs(thread);
332 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER);
338 * seccomp_prepare_filter: Prepares a seccomp filter for use.
339 * @fprog: BPF program to install
341 * Returns filter on success or an ERR_PTR on failure.
343 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
345 struct seccomp_filter *filter;
346 unsigned long fp_size;
347 struct sock_filter *fp;
351 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
352 return ERR_PTR(-EINVAL);
353 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
354 fp_size = fprog->len * sizeof(struct sock_filter);
357 * Installing a seccomp filter requires that the task has
358 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
359 * This avoids scenarios where unprivileged tasks can affect the
360 * behavior of privileged children.
362 if (!task_no_new_privs(current) &&
363 security_capable_noaudit(current_cred(), current_user_ns(),
365 return ERR_PTR(-EACCES);
367 fp = kzalloc(fp_size, GFP_KERNEL|__GFP_NOWARN);
369 return ERR_PTR(-ENOMEM);
371 /* Copy the instructions from fprog. */
373 if (copy_from_user(fp, fprog->filter, fp_size))
376 /* Check and rewrite the fprog via the skb checker */
377 ret = bpf_check_classic(fp, fprog->len);
381 /* Check and rewrite the fprog for seccomp use */
382 ret = seccomp_check_filter(fp, fprog->len);
386 /* Convert 'sock_filter' insns to 'bpf_insn' insns */
387 ret = bpf_convert_filter(fp, fprog->len, NULL, &new_len);
391 /* Allocate a new seccomp_filter */
393 filter = kzalloc(sizeof(struct seccomp_filter),
394 GFP_KERNEL|__GFP_NOWARN);
398 filter->prog = bpf_prog_alloc(bpf_prog_size(new_len), __GFP_NOWARN);
402 ret = bpf_convert_filter(fp, fprog->len, filter->prog->insnsi, &new_len);
404 goto free_filter_prog;
407 atomic_set(&filter->usage, 1);
408 filter->prog->len = new_len;
410 bpf_prog_select_runtime(filter->prog);
415 __bpf_prog_free(filter->prog);
424 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
425 * @user_filter: pointer to the user data containing a sock_fprog.
427 * Returns 0 on success and non-zero otherwise.
429 static struct seccomp_filter *
430 seccomp_prepare_user_filter(const char __user *user_filter)
432 struct sock_fprog fprog;
433 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
436 if (is_compat_task()) {
437 struct compat_sock_fprog fprog32;
438 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
440 fprog.len = fprog32.len;
441 fprog.filter = compat_ptr(fprog32.filter);
442 } else /* falls through to the if below. */
444 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
446 filter = seccomp_prepare_filter(&fprog);
452 * seccomp_attach_filter: validate and attach filter
453 * @flags: flags to change filter behavior
454 * @filter: seccomp filter to add to the current process
456 * Caller must be holding current->sighand->siglock lock.
458 * Returns 0 on success, -ve on error.
460 static long seccomp_attach_filter(unsigned int flags,
461 struct seccomp_filter *filter)
463 unsigned long total_insns;
464 struct seccomp_filter *walker;
466 assert_spin_locked(¤t->sighand->siglock);
468 /* Validate resulting filter length. */
469 total_insns = filter->prog->len;
470 for (walker = current->seccomp.filter; walker; walker = walker->prev)
471 total_insns += walker->prog->len + 4; /* 4 instr penalty */
472 if (total_insns > MAX_INSNS_PER_PATH)
475 /* If thread sync has been requested, check that it is possible. */
476 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
479 ret = seccomp_can_sync_threads();
485 * If there is an existing filter, make it the prev and don't drop its
488 filter->prev = current->seccomp.filter;
489 current->seccomp.filter = filter;
491 /* Now that the new filter is in place, synchronize to all threads. */
492 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
493 seccomp_sync_threads();
498 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
499 void get_seccomp_filter(struct task_struct *tsk)
501 struct seccomp_filter *orig = tsk->seccomp.filter;
504 /* Reference count is bounded by the number of total processes. */
505 atomic_inc(&orig->usage);
508 static inline void seccomp_filter_free(struct seccomp_filter *filter)
511 bpf_prog_free(filter->prog);
516 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
517 void put_seccomp_filter(struct task_struct *tsk)
519 struct seccomp_filter *orig = tsk->seccomp.filter;
520 /* Clean up single-reference branches iteratively. */
521 while (orig && atomic_dec_and_test(&orig->usage)) {
522 struct seccomp_filter *freeme = orig;
524 seccomp_filter_free(freeme);
529 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
530 * @syscall: syscall number to send to userland
531 * @reason: filter-supplied reason code to send to userland (via si_errno)
533 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
535 static void seccomp_send_sigsys(int syscall, int reason)
538 memset(&info, 0, sizeof(info));
539 info.si_signo = SIGSYS;
540 info.si_code = SYS_SECCOMP;
541 info.si_call_addr = (void __user *)KSTK_EIP(current);
542 info.si_errno = reason;
543 info.si_arch = syscall_get_arch();
544 info.si_syscall = syscall;
545 force_sig_info(SIGSYS, &info, current);
547 #endif /* CONFIG_SECCOMP_FILTER */
550 * Secure computing mode 1 allows only read/write/exit/sigreturn.
551 * To be fully secure this must be combined with rlimit
552 * to limit the stack allocations too.
554 static int mode1_syscalls[] = {
555 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
556 0, /* null terminated */
560 static int mode1_syscalls_32[] = {
561 __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
562 0, /* null terminated */
566 int __secure_computing(int this_syscall)
573 * Make sure that any changes to mode from another thread have
574 * been seen after TIF_SECCOMP was seen.
578 switch (current->seccomp.mode) {
579 case SECCOMP_MODE_STRICT:
580 syscall = mode1_syscalls;
582 if (is_compat_task())
583 syscall = mode1_syscalls_32;
586 if (*syscall == this_syscall)
588 } while (*++syscall);
590 ret = SECCOMP_RET_KILL;
592 #ifdef CONFIG_SECCOMP_FILTER
593 case SECCOMP_MODE_FILTER: {
595 struct pt_regs *regs = task_pt_regs(current);
596 ret = seccomp_run_filters(this_syscall);
597 data = ret & SECCOMP_RET_DATA;
598 ret &= SECCOMP_RET_ACTION;
600 case SECCOMP_RET_ERRNO:
601 /* Set the low-order 16-bits as a errno. */
602 syscall_set_return_value(current, regs,
605 case SECCOMP_RET_TRAP:
606 /* Show the handler the original registers. */
607 syscall_rollback(current, regs);
608 /* Let the filter pass back 16 bits of data. */
609 seccomp_send_sigsys(this_syscall, data);
611 case SECCOMP_RET_TRACE:
612 /* Skip these calls if there is no tracer. */
613 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
614 syscall_set_return_value(current, regs,
618 /* Allow the BPF to provide the event message */
619 ptrace_event(PTRACE_EVENT_SECCOMP, data);
621 * The delivery of a fatal signal during event
622 * notification may silently skip tracer notification.
623 * Terminating the task now avoids executing a system
624 * call that may not be intended.
626 if (fatal_signal_pending(current))
628 if (syscall_get_nr(current, regs) < 0)
629 goto skip; /* Explicit request to skip. */
632 case SECCOMP_RET_ALLOW:
634 case SECCOMP_RET_KILL:
649 audit_seccomp(this_syscall, exit_sig, ret);
651 #ifdef CONFIG_SECCOMP_FILTER
653 audit_seccomp(this_syscall, exit_sig, ret);
658 long prctl_get_seccomp(void)
660 return current->seccomp.mode;
664 * seccomp_set_mode_strict: internal function for setting strict seccomp
666 * Once current->seccomp.mode is non-zero, it may not be changed.
668 * Returns 0 on success or -EINVAL on failure.
670 static long seccomp_set_mode_strict(void)
672 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
675 spin_lock_irq(¤t->sighand->siglock);
677 if (!seccomp_may_assign_mode(seccomp_mode))
683 seccomp_assign_mode(current, seccomp_mode);
687 spin_unlock_irq(¤t->sighand->siglock);
692 #ifdef CONFIG_SECCOMP_FILTER
694 * seccomp_set_mode_filter: internal function for setting seccomp filter
695 * @flags: flags to change filter behavior
696 * @filter: struct sock_fprog containing filter
698 * This function may be called repeatedly to install additional filters.
699 * Every filter successfully installed will be evaluated (in reverse order)
700 * for each system call the task makes.
702 * Once current->seccomp.mode is non-zero, it may not be changed.
704 * Returns 0 on success or -EINVAL on failure.
706 static long seccomp_set_mode_filter(unsigned int flags,
707 const char __user *filter)
709 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
710 struct seccomp_filter *prepared = NULL;
713 /* Validate flags. */
714 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
717 /* Prepare the new filter before holding any locks. */
718 prepared = seccomp_prepare_user_filter(filter);
719 if (IS_ERR(prepared))
720 return PTR_ERR(prepared);
723 * Make sure we cannot change seccomp or nnp state via TSYNC
724 * while another thread is in the middle of calling exec.
726 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
727 mutex_lock_killable(¤t->signal->cred_guard_mutex))
730 spin_lock_irq(¤t->sighand->siglock);
732 if (!seccomp_may_assign_mode(seccomp_mode))
735 ret = seccomp_attach_filter(flags, prepared);
738 /* Do not free the successfully attached filter. */
741 seccomp_assign_mode(current, seccomp_mode);
743 spin_unlock_irq(¤t->sighand->siglock);
744 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
745 mutex_unlock(¤t->signal->cred_guard_mutex);
747 seccomp_filter_free(prepared);
751 static inline long seccomp_set_mode_filter(unsigned int flags,
752 const char __user *filter)
758 /* Common entry point for both prctl and syscall. */
759 static long do_seccomp(unsigned int op, unsigned int flags,
760 const char __user *uargs)
763 case SECCOMP_SET_MODE_STRICT:
764 if (flags != 0 || uargs != NULL)
766 return seccomp_set_mode_strict();
767 case SECCOMP_SET_MODE_FILTER:
768 return seccomp_set_mode_filter(flags, uargs);
774 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
775 const char __user *, uargs)
777 return do_seccomp(op, flags, uargs);
781 * prctl_set_seccomp: configures current->seccomp.mode
782 * @seccomp_mode: requested mode to use
783 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
785 * Returns 0 on success or -EINVAL on failure.
787 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
792 switch (seccomp_mode) {
793 case SECCOMP_MODE_STRICT:
794 op = SECCOMP_SET_MODE_STRICT;
796 * Setting strict mode through prctl always ignored filter,
797 * so make sure it is always NULL here to pass the internal
798 * check in do_seccomp().
802 case SECCOMP_MODE_FILTER:
803 op = SECCOMP_SET_MODE_FILTER;
810 /* prctl interface doesn't have flags, so they are always zero. */
811 return do_seccomp(op, 0, uargs);