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_HAVE_ARCH_SECCOMP_FILTER
27 #include <asm/syscall.h>
30 #ifdef CONFIG_SECCOMP_FILTER
31 #include <linux/filter.h>
32 #include <linux/pid.h>
33 #include <linux/ptrace.h>
34 #include <linux/security.h>
35 #include <linux/tracehook.h>
36 #include <linux/uaccess.h>
39 * struct seccomp_filter - container for seccomp BPF programs
41 * @usage: reference count to manage the object lifetime.
42 * get/put helpers should be used when accessing an instance
43 * outside of a lifetime-guarded section. In general, this
44 * is only needed for handling filters shared across tasks.
45 * @prev: points to a previously installed, or inherited, filter
46 * @len: the number of instructions in the program
47 * @insnsi: the BPF program instructions to evaluate
49 * seccomp_filter objects are organized in a tree linked via the @prev
50 * pointer. For any task, it appears to be a singly-linked list starting
51 * with current->seccomp.filter, the most recently attached or inherited filter.
52 * However, multiple filters may share a @prev node, by way of fork(), which
53 * results in a unidirectional tree existing in memory. This is similar to
54 * how namespaces work.
56 * seccomp_filter objects should never be modified after being attached
57 * to a task_struct (other than @usage).
59 struct seccomp_filter {
61 struct seccomp_filter *prev;
62 struct bpf_prog *prog;
65 /* Limit any path through the tree to 256KB worth of instructions. */
66 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
69 * Endianness is explicitly ignored and left for BPF program authors to manage
70 * as per the specific architecture.
72 static void populate_seccomp_data(struct seccomp_data *sd)
74 struct task_struct *task = current;
75 struct pt_regs *regs = task_pt_regs(task);
76 unsigned long args[6];
78 sd->nr = syscall_get_nr(task, regs);
79 sd->arch = syscall_get_arch();
80 syscall_get_arguments(task, regs, 0, 6, args);
81 sd->args[0] = args[0];
82 sd->args[1] = args[1];
83 sd->args[2] = args[2];
84 sd->args[3] = args[3];
85 sd->args[4] = args[4];
86 sd->args[5] = args[5];
87 sd->instruction_pointer = KSTK_EIP(task);
91 * seccomp_check_filter - verify seccomp filter code
92 * @filter: filter to verify
93 * @flen: length of filter
95 * Takes a previously checked filter (by bpf_check_classic) and
96 * redirects all filter code that loads struct sk_buff data
97 * and related data through seccomp_bpf_load. It also
98 * enforces length and alignment checking of those loads.
100 * Returns 0 if the rule set is legal or -EINVAL if not.
102 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
105 for (pc = 0; pc < flen; pc++) {
106 struct sock_filter *ftest = &filter[pc];
107 u16 code = ftest->code;
111 case BPF_LD | BPF_W | BPF_ABS:
112 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
113 /* 32-bit aligned and not out of bounds. */
114 if (k >= sizeof(struct seccomp_data) || k & 3)
117 case BPF_LD | BPF_W | BPF_LEN:
118 ftest->code = BPF_LD | BPF_IMM;
119 ftest->k = sizeof(struct seccomp_data);
121 case BPF_LDX | BPF_W | BPF_LEN:
122 ftest->code = BPF_LDX | BPF_IMM;
123 ftest->k = sizeof(struct seccomp_data);
125 /* Explicitly include allowed calls. */
126 case BPF_RET | BPF_K:
127 case BPF_RET | BPF_A:
128 case BPF_ALU | BPF_ADD | BPF_K:
129 case BPF_ALU | BPF_ADD | BPF_X:
130 case BPF_ALU | BPF_SUB | BPF_K:
131 case BPF_ALU | BPF_SUB | BPF_X:
132 case BPF_ALU | BPF_MUL | BPF_K:
133 case BPF_ALU | BPF_MUL | BPF_X:
134 case BPF_ALU | BPF_DIV | BPF_K:
135 case BPF_ALU | BPF_DIV | BPF_X:
136 case BPF_ALU | BPF_AND | BPF_K:
137 case BPF_ALU | BPF_AND | BPF_X:
138 case BPF_ALU | BPF_OR | BPF_K:
139 case BPF_ALU | BPF_OR | BPF_X:
140 case BPF_ALU | BPF_XOR | BPF_K:
141 case BPF_ALU | BPF_XOR | BPF_X:
142 case BPF_ALU | BPF_LSH | BPF_K:
143 case BPF_ALU | BPF_LSH | BPF_X:
144 case BPF_ALU | BPF_RSH | BPF_K:
145 case BPF_ALU | BPF_RSH | BPF_X:
146 case BPF_ALU | BPF_NEG:
147 case BPF_LD | BPF_IMM:
148 case BPF_LDX | BPF_IMM:
149 case BPF_MISC | BPF_TAX:
150 case BPF_MISC | BPF_TXA:
151 case BPF_LD | BPF_MEM:
152 case BPF_LDX | BPF_MEM:
155 case BPF_JMP | BPF_JA:
156 case BPF_JMP | BPF_JEQ | BPF_K:
157 case BPF_JMP | BPF_JEQ | BPF_X:
158 case BPF_JMP | BPF_JGE | BPF_K:
159 case BPF_JMP | BPF_JGE | BPF_X:
160 case BPF_JMP | BPF_JGT | BPF_K:
161 case BPF_JMP | BPF_JGT | BPF_X:
162 case BPF_JMP | BPF_JSET | BPF_K:
163 case BPF_JMP | BPF_JSET | BPF_X:
173 * seccomp_run_filters - evaluates all seccomp filters against @syscall
174 * @syscall: number of the current system call
176 * Returns valid seccomp BPF response codes.
178 static u32 seccomp_run_filters(void)
180 struct seccomp_filter *f = ACCESS_ONCE(current->seccomp.filter);
181 struct seccomp_data sd;
182 u32 ret = SECCOMP_RET_ALLOW;
184 /* Ensure unexpected behavior doesn't result in failing open. */
185 if (unlikely(WARN_ON(f == NULL)))
186 return SECCOMP_RET_KILL;
188 /* Make sure cross-thread synced filter points somewhere sane. */
189 smp_read_barrier_depends();
191 populate_seccomp_data(&sd);
194 * All filters in the list are evaluated and the lowest BPF return
195 * value always takes priority (ignoring the DATA).
197 for (; f; f = f->prev) {
198 u32 cur_ret = BPF_PROG_RUN(f->prog, (void *)&sd);
200 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
205 #endif /* CONFIG_SECCOMP_FILTER */
207 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
209 assert_spin_locked(¤t->sighand->siglock);
211 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
217 static inline void seccomp_assign_mode(struct task_struct *task,
218 unsigned long seccomp_mode)
220 assert_spin_locked(&task->sighand->siglock);
222 task->seccomp.mode = seccomp_mode;
224 * Make sure TIF_SECCOMP cannot be set before the mode (and
227 smp_mb__before_atomic();
228 set_tsk_thread_flag(task, TIF_SECCOMP);
231 #ifdef CONFIG_SECCOMP_FILTER
232 /* Returns 1 if the parent is an ancestor of the child. */
233 static int is_ancestor(struct seccomp_filter *parent,
234 struct seccomp_filter *child)
236 /* NULL is the root ancestor. */
239 for (; child; child = child->prev)
246 * seccomp_can_sync_threads: checks if all threads can be synchronized
248 * Expects sighand and cred_guard_mutex locks to be held.
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
254 static inline pid_t seccomp_can_sync_threads(void)
256 struct task_struct *thread, *caller;
258 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
259 assert_spin_locked(¤t->sighand->siglock);
261 /* Validate all threads being eligible for synchronization. */
263 for_each_thread(caller, thread) {
266 /* Skip current, since it is initiating the sync. */
267 if (thread == caller)
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)))
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)))
288 * seccomp_sync_threads: sets all threads to use current's filter
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.
295 static inline void seccomp_sync_threads(void)
297 struct task_struct *thread, *caller;
299 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
300 assert_spin_locked(¤t->sighand->siglock);
302 /* Synchronize all threads. */
304 for_each_thread(caller, thread) {
305 /* Skip current, since it needs no changes. */
306 if (thread == caller)
309 /* Get a task reference for the new leaf node. */
310 get_seccomp_filter(caller);
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.)
316 put_seccomp_filter(thread);
317 smp_store_release(&thread->seccomp.filter,
318 caller->seccomp.filter);
320 * Opt the other thread into seccomp if needed.
321 * As threads are considered to be trust-realm
322 * equivalent (see ptrace_may_access), it is safe to
323 * allow one thread to transition the other.
325 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) {
327 * Don't let an unprivileged task work around
328 * the no_new_privs restriction by creating
329 * a thread that sets it up, enters seccomp,
332 if (task_no_new_privs(caller))
333 task_set_no_new_privs(thread);
335 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER);
341 * seccomp_prepare_filter: Prepares a seccomp filter for use.
342 * @fprog: BPF program to install
344 * Returns filter on success or an ERR_PTR on failure.
346 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
348 struct seccomp_filter *filter;
349 unsigned long fp_size;
350 struct sock_filter *fp;
354 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
355 return ERR_PTR(-EINVAL);
356 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
357 fp_size = fprog->len * sizeof(struct sock_filter);
360 * Installing a seccomp filter requires that the task has
361 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
362 * This avoids scenarios where unprivileged tasks can affect the
363 * behavior of privileged children.
365 if (!task_no_new_privs(current) &&
366 security_capable_noaudit(current_cred(), current_user_ns(),
368 return ERR_PTR(-EACCES);
370 fp = kzalloc(fp_size, GFP_KERNEL|__GFP_NOWARN);
372 return ERR_PTR(-ENOMEM);
374 /* Copy the instructions from fprog. */
376 if (copy_from_user(fp, fprog->filter, fp_size))
379 /* Check and rewrite the fprog via the skb checker */
380 ret = bpf_check_classic(fp, fprog->len);
384 /* Check and rewrite the fprog for seccomp use */
385 ret = seccomp_check_filter(fp, fprog->len);
389 /* Convert 'sock_filter' insns to 'bpf_insn' insns */
390 ret = bpf_convert_filter(fp, fprog->len, NULL, &new_len);
394 /* Allocate a new seccomp_filter */
396 filter = kzalloc(sizeof(struct seccomp_filter),
397 GFP_KERNEL|__GFP_NOWARN);
401 filter->prog = kzalloc(bpf_prog_size(new_len),
402 GFP_KERNEL|__GFP_NOWARN);
406 ret = bpf_convert_filter(fp, fprog->len, filter->prog->insnsi, &new_len);
408 goto free_filter_prog;
411 atomic_set(&filter->usage, 1);
412 filter->prog->len = new_len;
414 bpf_prog_select_runtime(filter->prog);
428 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
429 * @user_filter: pointer to the user data containing a sock_fprog.
431 * Returns 0 on success and non-zero otherwise.
433 static struct seccomp_filter *
434 seccomp_prepare_user_filter(const char __user *user_filter)
436 struct sock_fprog fprog;
437 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
440 if (is_compat_task()) {
441 struct compat_sock_fprog fprog32;
442 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
444 fprog.len = fprog32.len;
445 fprog.filter = compat_ptr(fprog32.filter);
446 } else /* falls through to the if below. */
448 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
450 filter = seccomp_prepare_filter(&fprog);
456 * seccomp_attach_filter: validate and attach filter
457 * @flags: flags to change filter behavior
458 * @filter: seccomp filter to add to the current process
460 * Caller must be holding current->sighand->siglock lock.
462 * Returns 0 on success, -ve on error.
464 static long seccomp_attach_filter(unsigned int flags,
465 struct seccomp_filter *filter)
467 unsigned long total_insns;
468 struct seccomp_filter *walker;
470 assert_spin_locked(¤t->sighand->siglock);
472 /* Validate resulting filter length. */
473 total_insns = filter->prog->len;
474 for (walker = current->seccomp.filter; walker; walker = walker->prev)
475 total_insns += walker->prog->len + 4; /* 4 instr penalty */
476 if (total_insns > MAX_INSNS_PER_PATH)
479 /* If thread sync has been requested, check that it is possible. */
480 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
483 ret = seccomp_can_sync_threads();
489 * If there is an existing filter, make it the prev and don't drop its
492 filter->prev = current->seccomp.filter;
493 current->seccomp.filter = filter;
495 /* Now that the new filter is in place, synchronize to all threads. */
496 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
497 seccomp_sync_threads();
502 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
503 void get_seccomp_filter(struct task_struct *tsk)
505 struct seccomp_filter *orig = tsk->seccomp.filter;
508 /* Reference count is bounded by the number of total processes. */
509 atomic_inc(&orig->usage);
512 static inline void seccomp_filter_free(struct seccomp_filter *filter)
515 bpf_prog_free(filter->prog);
520 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
521 void put_seccomp_filter(struct task_struct *tsk)
523 struct seccomp_filter *orig = tsk->seccomp.filter;
524 /* Clean up single-reference branches iteratively. */
525 while (orig && atomic_dec_and_test(&orig->usage)) {
526 struct seccomp_filter *freeme = orig;
528 seccomp_filter_free(freeme);
533 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
534 * @syscall: syscall number to send to userland
535 * @reason: filter-supplied reason code to send to userland (via si_errno)
537 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
539 static void seccomp_send_sigsys(int syscall, int reason)
542 memset(&info, 0, sizeof(info));
543 info.si_signo = SIGSYS;
544 info.si_code = SYS_SECCOMP;
545 info.si_call_addr = (void __user *)KSTK_EIP(current);
546 info.si_errno = reason;
547 info.si_arch = syscall_get_arch();
548 info.si_syscall = syscall;
549 force_sig_info(SIGSYS, &info, current);
551 #endif /* CONFIG_SECCOMP_FILTER */
554 * Secure computing mode 1 allows only read/write/exit/sigreturn.
555 * To be fully secure this must be combined with rlimit
556 * to limit the stack allocations too.
558 static int mode1_syscalls[] = {
559 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
560 0, /* null terminated */
564 static int mode1_syscalls_32[] = {
565 __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
566 0, /* null terminated */
570 static void __secure_computing_strict(int this_syscall)
572 int *syscall_whitelist = mode1_syscalls;
574 if (is_compat_task())
575 syscall_whitelist = mode1_syscalls_32;
578 if (*syscall_whitelist == this_syscall)
580 } while (*++syscall_whitelist);
585 audit_seccomp(this_syscall, SIGKILL, SECCOMP_RET_KILL);
589 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
590 void secure_computing_strict(int this_syscall)
592 int mode = current->seccomp.mode;
596 else if (mode == SECCOMP_MODE_STRICT)
597 __secure_computing_strict(this_syscall);
602 int __secure_computing(void)
604 struct pt_regs *regs = task_pt_regs(current);
605 int this_syscall = syscall_get_nr(current, regs);
610 * Make sure that any changes to mode from another thread have
611 * been seen after TIF_SECCOMP was seen.
615 switch (current->seccomp.mode) {
616 case SECCOMP_MODE_STRICT:
617 __secure_computing_strict(this_syscall);
619 #ifdef CONFIG_SECCOMP_FILTER
620 case SECCOMP_MODE_FILTER: {
622 ret = seccomp_run_filters();
623 data = ret & SECCOMP_RET_DATA;
624 ret &= SECCOMP_RET_ACTION;
626 case SECCOMP_RET_ERRNO:
627 /* Set the low-order 16-bits as a errno. */
628 syscall_set_return_value(current, regs,
631 case SECCOMP_RET_TRAP:
632 /* Show the handler the original registers. */
633 syscall_rollback(current, regs);
634 /* Let the filter pass back 16 bits of data. */
635 seccomp_send_sigsys(this_syscall, data);
637 case SECCOMP_RET_TRACE:
638 /* Skip these calls if there is no tracer. */
639 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
640 syscall_set_return_value(current, regs,
644 /* Allow the BPF to provide the event message */
645 ptrace_event(PTRACE_EVENT_SECCOMP, data);
647 * The delivery of a fatal signal during event
648 * notification may silently skip tracer notification.
649 * Terminating the task now avoids executing a system
650 * call that may not be intended.
652 if (fatal_signal_pending(current))
654 if (syscall_get_nr(current, regs) < 0)
655 goto skip; /* Explicit request to skip. */
658 case SECCOMP_RET_ALLOW:
660 case SECCOMP_RET_KILL:
675 audit_seccomp(this_syscall, exit_sig, ret);
677 #ifdef CONFIG_SECCOMP_FILTER
679 audit_seccomp(this_syscall, exit_sig, ret);
683 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
685 long prctl_get_seccomp(void)
687 return current->seccomp.mode;
691 * seccomp_set_mode_strict: internal function for setting strict seccomp
693 * Once current->seccomp.mode is non-zero, it may not be changed.
695 * Returns 0 on success or -EINVAL on failure.
697 static long seccomp_set_mode_strict(void)
699 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
702 spin_lock_irq(¤t->sighand->siglock);
704 if (!seccomp_may_assign_mode(seccomp_mode))
710 seccomp_assign_mode(current, seccomp_mode);
714 spin_unlock_irq(¤t->sighand->siglock);
719 #ifdef CONFIG_SECCOMP_FILTER
721 * seccomp_set_mode_filter: internal function for setting seccomp filter
722 * @flags: flags to change filter behavior
723 * @filter: struct sock_fprog containing filter
725 * This function may be called repeatedly to install additional filters.
726 * Every filter successfully installed will be evaluated (in reverse order)
727 * for each system call the task makes.
729 * Once current->seccomp.mode is non-zero, it may not be changed.
731 * Returns 0 on success or -EINVAL on failure.
733 static long seccomp_set_mode_filter(unsigned int flags,
734 const char __user *filter)
736 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
737 struct seccomp_filter *prepared = NULL;
740 /* Validate flags. */
741 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
744 /* Prepare the new filter before holding any locks. */
745 prepared = seccomp_prepare_user_filter(filter);
746 if (IS_ERR(prepared))
747 return PTR_ERR(prepared);
750 * Make sure we cannot change seccomp or nnp state via TSYNC
751 * while another thread is in the middle of calling exec.
753 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
754 mutex_lock_killable(¤t->signal->cred_guard_mutex))
757 spin_lock_irq(¤t->sighand->siglock);
759 if (!seccomp_may_assign_mode(seccomp_mode))
762 ret = seccomp_attach_filter(flags, prepared);
765 /* Do not free the successfully attached filter. */
768 seccomp_assign_mode(current, seccomp_mode);
770 spin_unlock_irq(¤t->sighand->siglock);
771 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
772 mutex_unlock(¤t->signal->cred_guard_mutex);
774 seccomp_filter_free(prepared);
778 static inline long seccomp_set_mode_filter(unsigned int flags,
779 const char __user *filter)
785 /* Common entry point for both prctl and syscall. */
786 static long do_seccomp(unsigned int op, unsigned int flags,
787 const char __user *uargs)
790 case SECCOMP_SET_MODE_STRICT:
791 if (flags != 0 || uargs != NULL)
793 return seccomp_set_mode_strict();
794 case SECCOMP_SET_MODE_FILTER:
795 return seccomp_set_mode_filter(flags, uargs);
801 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
802 const char __user *, uargs)
804 return do_seccomp(op, flags, uargs);
808 * prctl_set_seccomp: configures current->seccomp.mode
809 * @seccomp_mode: requested mode to use
810 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
812 * Returns 0 on success or -EINVAL on failure.
814 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
819 switch (seccomp_mode) {
820 case SECCOMP_MODE_STRICT:
821 op = SECCOMP_SET_MODE_STRICT;
823 * Setting strict mode through prctl always ignored filter,
824 * so make sure it is always NULL here to pass the internal
825 * check in do_seccomp().
829 case SECCOMP_MODE_FILTER:
830 op = SECCOMP_SET_MODE_FILTER;
837 /* prctl interface doesn't have flags, so they are always zero. */
838 return do_seccomp(op, 0, uargs);