1 // SPDX-License-Identifier: GPL-2.0
3 * linux/kernel/seccomp.c
5 * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com>
7 * Copyright (C) 2012 Google, Inc.
8 * Will Drewry <wad@chromium.org>
10 * This defines a simple but solid secure-computing facility.
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.
17 #include <linux/refcount.h>
18 #include <linux/audit.h>
19 #include <linux/compat.h>
20 #include <linux/coredump.h>
21 #include <linux/kmemleak.h>
22 #include <linux/nospec.h>
23 #include <linux/prctl.h>
24 #include <linux/sched.h>
25 #include <linux/sched/task_stack.h>
26 #include <linux/seccomp.h>
27 #include <linux/slab.h>
28 #include <linux/syscalls.h>
29 #include <linux/sysctl.h>
31 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
32 #include <asm/syscall.h>
35 #ifdef CONFIG_SECCOMP_FILTER
36 #include <linux/filter.h>
37 #include <linux/pid.h>
38 #include <linux/ptrace.h>
39 #include <linux/security.h>
40 #include <linux/tracehook.h>
41 #include <linux/uaccess.h>
44 * struct seccomp_filter - container for seccomp BPF programs
46 * @usage: reference count to manage the object lifetime.
47 * get/put helpers should be used when accessing an instance
48 * outside of a lifetime-guarded section. In general, this
49 * is only needed for handling filters shared across tasks.
50 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
51 * @prev: points to a previously installed, or inherited, filter
52 * @prog: the BPF program to evaluate
54 * seccomp_filter objects are organized in a tree linked via the @prev
55 * pointer. For any task, it appears to be a singly-linked list starting
56 * with current->seccomp.filter, the most recently attached or inherited filter.
57 * However, multiple filters may share a @prev node, by way of fork(), which
58 * results in a unidirectional tree existing in memory. This is similar to
59 * how namespaces work.
61 * seccomp_filter objects should never be modified after being attached
62 * to a task_struct (other than @usage).
64 struct seccomp_filter {
67 struct seccomp_filter *prev;
68 struct bpf_prog *prog;
71 /* Limit any path through the tree to 256KB worth of instructions. */
72 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
75 * Endianness is explicitly ignored and left for BPF program authors to manage
76 * as per the specific architecture.
78 static void populate_seccomp_data(struct seccomp_data *sd)
80 struct task_struct *task = current;
81 struct pt_regs *regs = task_pt_regs(task);
82 unsigned long args[6];
84 sd->nr = syscall_get_nr(task, regs);
85 sd->arch = syscall_get_arch();
86 syscall_get_arguments(task, regs, 0, 6, args);
87 sd->args[0] = args[0];
88 sd->args[1] = args[1];
89 sd->args[2] = args[2];
90 sd->args[3] = args[3];
91 sd->args[4] = args[4];
92 sd->args[5] = args[5];
93 sd->instruction_pointer = KSTK_EIP(task);
97 * seccomp_check_filter - verify seccomp filter code
98 * @filter: filter to verify
99 * @flen: length of filter
101 * Takes a previously checked filter (by bpf_check_classic) and
102 * redirects all filter code that loads struct sk_buff data
103 * and related data through seccomp_bpf_load. It also
104 * enforces length and alignment checking of those loads.
106 * Returns 0 if the rule set is legal or -EINVAL if not.
108 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
111 for (pc = 0; pc < flen; pc++) {
112 struct sock_filter *ftest = &filter[pc];
113 u16 code = ftest->code;
117 case BPF_LD | BPF_W | BPF_ABS:
118 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
119 /* 32-bit aligned and not out of bounds. */
120 if (k >= sizeof(struct seccomp_data) || k & 3)
123 case BPF_LD | BPF_W | BPF_LEN:
124 ftest->code = BPF_LD | BPF_IMM;
125 ftest->k = sizeof(struct seccomp_data);
127 case BPF_LDX | BPF_W | BPF_LEN:
128 ftest->code = BPF_LDX | BPF_IMM;
129 ftest->k = sizeof(struct seccomp_data);
131 /* Explicitly include allowed calls. */
132 case BPF_RET | BPF_K:
133 case BPF_RET | BPF_A:
134 case BPF_ALU | BPF_ADD | BPF_K:
135 case BPF_ALU | BPF_ADD | BPF_X:
136 case BPF_ALU | BPF_SUB | BPF_K:
137 case BPF_ALU | BPF_SUB | BPF_X:
138 case BPF_ALU | BPF_MUL | BPF_K:
139 case BPF_ALU | BPF_MUL | BPF_X:
140 case BPF_ALU | BPF_DIV | BPF_K:
141 case BPF_ALU | BPF_DIV | BPF_X:
142 case BPF_ALU | BPF_AND | BPF_K:
143 case BPF_ALU | BPF_AND | BPF_X:
144 case BPF_ALU | BPF_OR | BPF_K:
145 case BPF_ALU | BPF_OR | BPF_X:
146 case BPF_ALU | BPF_XOR | BPF_K:
147 case BPF_ALU | BPF_XOR | BPF_X:
148 case BPF_ALU | BPF_LSH | BPF_K:
149 case BPF_ALU | BPF_LSH | BPF_X:
150 case BPF_ALU | BPF_RSH | BPF_K:
151 case BPF_ALU | BPF_RSH | BPF_X:
152 case BPF_ALU | BPF_NEG:
153 case BPF_LD | BPF_IMM:
154 case BPF_LDX | BPF_IMM:
155 case BPF_MISC | BPF_TAX:
156 case BPF_MISC | BPF_TXA:
157 case BPF_LD | BPF_MEM:
158 case BPF_LDX | BPF_MEM:
161 case BPF_JMP | BPF_JA:
162 case BPF_JMP | BPF_JEQ | BPF_K:
163 case BPF_JMP | BPF_JEQ | BPF_X:
164 case BPF_JMP | BPF_JGE | BPF_K:
165 case BPF_JMP | BPF_JGE | BPF_X:
166 case BPF_JMP | BPF_JGT | BPF_K:
167 case BPF_JMP | BPF_JGT | BPF_X:
168 case BPF_JMP | BPF_JSET | BPF_K:
169 case BPF_JMP | BPF_JSET | BPF_X:
179 * seccomp_run_filters - evaluates all seccomp filters against @sd
180 * @sd: optional seccomp data to be passed to filters
181 * @match: stores struct seccomp_filter that resulted in the return value,
182 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
185 * Returns valid seccomp BPF response codes.
187 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
188 static u32 seccomp_run_filters(const struct seccomp_data *sd,
189 struct seccomp_filter **match)
191 struct seccomp_data sd_local;
192 u32 ret = SECCOMP_RET_ALLOW;
193 /* Make sure cross-thread synced filter points somewhere sane. */
194 struct seccomp_filter *f =
195 READ_ONCE(current->seccomp.filter);
197 /* Ensure unexpected behavior doesn't result in failing open. */
198 if (unlikely(WARN_ON(f == NULL)))
199 return SECCOMP_RET_KILL_PROCESS;
202 populate_seccomp_data(&sd_local);
207 * All filters in the list are evaluated and the lowest BPF return
208 * value always takes priority (ignoring the DATA).
210 for (; f; f = f->prev) {
211 u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
213 if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
220 #endif /* CONFIG_SECCOMP_FILTER */
222 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
224 assert_spin_locked(¤t->sighand->siglock);
226 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
233 * If a given speculation mitigation is opt-in (prctl()-controlled),
234 * select it, by disabling speculation (enabling mitigation).
236 static inline void spec_mitigate(struct task_struct *task,
239 int state = arch_prctl_spec_ctrl_get(task, which);
241 if (state > 0 && (state & PR_SPEC_PRCTL))
242 arch_prctl_spec_ctrl_set(task, which, PR_SPEC_FORCE_DISABLE);
245 static inline void seccomp_assign_mode(struct task_struct *task,
246 unsigned long seccomp_mode,
249 assert_spin_locked(&task->sighand->siglock);
251 task->seccomp.mode = seccomp_mode;
253 * Make sure TIF_SECCOMP cannot be set before the mode (and
256 smp_mb__before_atomic();
257 /* Assume default seccomp processes want spec flaw mitigation. */
258 if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
259 spec_mitigate(task, PR_SPEC_STORE_BYPASS);
260 set_tsk_thread_flag(task, TIF_SECCOMP);
263 #ifdef CONFIG_SECCOMP_FILTER
264 /* Returns 1 if the parent is an ancestor of the child. */
265 static int is_ancestor(struct seccomp_filter *parent,
266 struct seccomp_filter *child)
268 /* NULL is the root ancestor. */
271 for (; child; child = child->prev)
278 * seccomp_can_sync_threads: checks if all threads can be synchronized
280 * Expects sighand and cred_guard_mutex locks to be held.
282 * Returns 0 on success, -ve on error, or the pid of a thread which was
283 * either not in the correct seccomp mode or it did not have an ancestral
286 static inline pid_t seccomp_can_sync_threads(void)
288 struct task_struct *thread, *caller;
290 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
291 assert_spin_locked(¤t->sighand->siglock);
293 /* Validate all threads being eligible for synchronization. */
295 for_each_thread(caller, thread) {
298 /* Skip current, since it is initiating the sync. */
299 if (thread == caller)
302 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
303 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
304 is_ancestor(thread->seccomp.filter,
305 caller->seccomp.filter)))
308 /* Return the first thread that cannot be synchronized. */
309 failed = task_pid_vnr(thread);
310 /* If the pid cannot be resolved, then return -ESRCH */
311 if (unlikely(WARN_ON(failed == 0)))
320 * seccomp_sync_threads: sets all threads to use current's filter
322 * Expects sighand and cred_guard_mutex locks to be held, and for
323 * seccomp_can_sync_threads() to have returned success already
324 * without dropping the locks.
327 static inline void seccomp_sync_threads(unsigned long flags)
329 struct task_struct *thread, *caller;
331 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
332 assert_spin_locked(¤t->sighand->siglock);
334 /* Synchronize all threads. */
336 for_each_thread(caller, thread) {
337 /* Skip current, since it needs no changes. */
338 if (thread == caller)
341 /* Get a task reference for the new leaf node. */
342 get_seccomp_filter(caller);
344 * Drop the task reference to the shared ancestor since
345 * current's path will hold a reference. (This also
346 * allows a put before the assignment.)
348 put_seccomp_filter(thread);
349 smp_store_release(&thread->seccomp.filter,
350 caller->seccomp.filter);
353 * Don't let an unprivileged task work around
354 * the no_new_privs restriction by creating
355 * a thread that sets it up, enters seccomp,
358 if (task_no_new_privs(caller))
359 task_set_no_new_privs(thread);
362 * Opt the other thread into seccomp if needed.
363 * As threads are considered to be trust-realm
364 * equivalent (see ptrace_may_access), it is safe to
365 * allow one thread to transition the other.
367 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
368 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
374 * seccomp_prepare_filter: Prepares a seccomp filter for use.
375 * @fprog: BPF program to install
377 * Returns filter on success or an ERR_PTR on failure.
379 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
381 struct seccomp_filter *sfilter;
383 const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
385 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
386 return ERR_PTR(-EINVAL);
388 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
391 * Installing a seccomp filter requires that the task has
392 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
393 * This avoids scenarios where unprivileged tasks can affect the
394 * behavior of privileged children.
396 if (!task_no_new_privs(current) &&
397 security_capable_noaudit(current_cred(), current_user_ns(),
399 return ERR_PTR(-EACCES);
401 /* Allocate a new seccomp_filter */
402 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
404 return ERR_PTR(-ENOMEM);
406 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
407 seccomp_check_filter, save_orig);
413 refcount_set(&sfilter->usage, 1);
419 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
420 * @user_filter: pointer to the user data containing a sock_fprog.
422 * Returns 0 on success and non-zero otherwise.
424 static struct seccomp_filter *
425 seccomp_prepare_user_filter(const char __user *user_filter)
427 struct sock_fprog fprog;
428 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
431 if (in_compat_syscall()) {
432 struct compat_sock_fprog fprog32;
433 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
435 fprog.len = fprog32.len;
436 fprog.filter = compat_ptr(fprog32.filter);
437 } else /* falls through to the if below. */
439 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
441 filter = seccomp_prepare_filter(&fprog);
447 * seccomp_attach_filter: validate and attach filter
448 * @flags: flags to change filter behavior
449 * @filter: seccomp filter to add to the current process
451 * Caller must be holding current->sighand->siglock lock.
453 * Returns 0 on success, -ve on error.
455 static long seccomp_attach_filter(unsigned int flags,
456 struct seccomp_filter *filter)
458 unsigned long total_insns;
459 struct seccomp_filter *walker;
461 assert_spin_locked(¤t->sighand->siglock);
463 /* Validate resulting filter length. */
464 total_insns = filter->prog->len;
465 for (walker = current->seccomp.filter; walker; walker = walker->prev)
466 total_insns += walker->prog->len + 4; /* 4 instr penalty */
467 if (total_insns > MAX_INSNS_PER_PATH)
470 /* If thread sync has been requested, check that it is possible. */
471 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
474 ret = seccomp_can_sync_threads();
479 /* Set log flag, if present. */
480 if (flags & SECCOMP_FILTER_FLAG_LOG)
484 * If there is an existing filter, make it the prev and don't drop its
487 filter->prev = current->seccomp.filter;
488 current->seccomp.filter = filter;
490 /* Now that the new filter is in place, synchronize to all threads. */
491 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
492 seccomp_sync_threads(flags);
497 static void __get_seccomp_filter(struct seccomp_filter *filter)
499 /* Reference count is bounded by the number of total processes. */
500 refcount_inc(&filter->usage);
503 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
504 void get_seccomp_filter(struct task_struct *tsk)
506 struct seccomp_filter *orig = tsk->seccomp.filter;
509 __get_seccomp_filter(orig);
512 static inline void seccomp_filter_free(struct seccomp_filter *filter)
515 bpf_prog_destroy(filter->prog);
520 static void __put_seccomp_filter(struct seccomp_filter *orig)
522 /* Clean up single-reference branches iteratively. */
523 while (orig && refcount_dec_and_test(&orig->usage)) {
524 struct seccomp_filter *freeme = orig;
526 seccomp_filter_free(freeme);
530 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
531 void put_seccomp_filter(struct task_struct *tsk)
533 __put_seccomp_filter(tsk->seccomp.filter);
536 static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason)
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;
548 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
549 * @syscall: syscall number to send to userland
550 * @reason: filter-supplied reason code to send to userland (via si_errno)
552 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
554 static void seccomp_send_sigsys(int syscall, int reason)
557 seccomp_init_siginfo(&info, syscall, reason);
558 force_sig_info(SIGSYS, &info, current);
560 #endif /* CONFIG_SECCOMP_FILTER */
562 /* For use with seccomp_actions_logged */
563 #define SECCOMP_LOG_KILL_PROCESS (1 << 0)
564 #define SECCOMP_LOG_KILL_THREAD (1 << 1)
565 #define SECCOMP_LOG_TRAP (1 << 2)
566 #define SECCOMP_LOG_ERRNO (1 << 3)
567 #define SECCOMP_LOG_TRACE (1 << 4)
568 #define SECCOMP_LOG_LOG (1 << 5)
569 #define SECCOMP_LOG_ALLOW (1 << 6)
571 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
572 SECCOMP_LOG_KILL_THREAD |
578 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
584 case SECCOMP_RET_ALLOW:
586 case SECCOMP_RET_TRAP:
587 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
589 case SECCOMP_RET_ERRNO:
590 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
592 case SECCOMP_RET_TRACE:
593 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
595 case SECCOMP_RET_LOG:
596 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
598 case SECCOMP_RET_KILL_THREAD:
599 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
601 case SECCOMP_RET_KILL_PROCESS:
603 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
607 * Force an audit message to be emitted when the action is RET_KILL_*,
608 * RET_LOG, or the FILTER_FLAG_LOG bit was set and the action is
609 * allowed to be logged by the admin.
612 return __audit_seccomp(syscall, signr, action);
615 * Let the audit subsystem decide if the action should be audited based
616 * on whether the current task itself is being audited.
618 return audit_seccomp(syscall, signr, action);
622 * Secure computing mode 1 allows only read/write/exit/sigreturn.
623 * To be fully secure this must be combined with rlimit
624 * to limit the stack allocations too.
626 static const int mode1_syscalls[] = {
627 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
628 0, /* null terminated */
631 static void __secure_computing_strict(int this_syscall)
633 const int *syscall_whitelist = mode1_syscalls;
635 if (in_compat_syscall())
636 syscall_whitelist = get_compat_mode1_syscalls();
639 if (*syscall_whitelist == this_syscall)
641 } while (*++syscall_whitelist);
646 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
650 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
651 void secure_computing_strict(int this_syscall)
653 int mode = current->seccomp.mode;
655 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
656 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
659 if (mode == SECCOMP_MODE_DISABLED)
661 else if (mode == SECCOMP_MODE_STRICT)
662 __secure_computing_strict(this_syscall);
668 #ifdef CONFIG_SECCOMP_FILTER
669 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
670 const bool recheck_after_trace)
672 u32 filter_ret, action;
673 struct seccomp_filter *match = NULL;
677 * Make sure that any changes to mode from another thread have
678 * been seen after TIF_SECCOMP was seen.
682 filter_ret = seccomp_run_filters(sd, &match);
683 data = filter_ret & SECCOMP_RET_DATA;
684 action = filter_ret & SECCOMP_RET_ACTION_FULL;
687 case SECCOMP_RET_ERRNO:
688 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
689 if (data > MAX_ERRNO)
691 syscall_set_return_value(current, task_pt_regs(current),
695 case SECCOMP_RET_TRAP:
696 /* Show the handler the original registers. */
697 syscall_rollback(current, task_pt_regs(current));
698 /* Let the filter pass back 16 bits of data. */
699 seccomp_send_sigsys(this_syscall, data);
702 case SECCOMP_RET_TRACE:
703 /* We've been put in this state by the ptracer already. */
704 if (recheck_after_trace)
707 /* ENOSYS these calls if there is no tracer attached. */
708 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
709 syscall_set_return_value(current,
710 task_pt_regs(current),
715 /* Allow the BPF to provide the event message */
716 ptrace_event(PTRACE_EVENT_SECCOMP, data);
718 * The delivery of a fatal signal during event
719 * notification may silently skip tracer notification,
720 * which could leave us with a potentially unmodified
721 * syscall that the tracer would have liked to have
722 * changed. Since the process is about to die, we just
723 * force the syscall to be skipped and let the signal
724 * kill the process and correctly handle any tracer exit
727 if (fatal_signal_pending(current))
729 /* Check if the tracer forced the syscall to be skipped. */
730 this_syscall = syscall_get_nr(current, task_pt_regs(current));
731 if (this_syscall < 0)
735 * Recheck the syscall, since it may have changed. This
736 * intentionally uses a NULL struct seccomp_data to force
737 * a reload of all registers. This does not goto skip since
738 * a skip would have already been reported.
740 if (__seccomp_filter(this_syscall, NULL, true))
745 case SECCOMP_RET_LOG:
746 seccomp_log(this_syscall, 0, action, true);
749 case SECCOMP_RET_ALLOW:
751 * Note that the "match" filter will always be NULL for
752 * this action since SECCOMP_RET_ALLOW is the starting
753 * state in seccomp_run_filters().
757 case SECCOMP_RET_KILL_THREAD:
758 case SECCOMP_RET_KILL_PROCESS:
760 seccomp_log(this_syscall, SIGSYS, action, true);
761 /* Dump core only if this is the last remaining thread. */
762 if (action == SECCOMP_RET_KILL_PROCESS ||
763 get_nr_threads(current) == 1) {
766 /* Show the original registers in the dump. */
767 syscall_rollback(current, task_pt_regs(current));
768 /* Trigger a manual coredump since do_exit skips it. */
769 seccomp_init_siginfo(&info, this_syscall, data);
772 if (action == SECCOMP_RET_KILL_PROCESS)
773 do_group_exit(SIGSYS);
781 seccomp_log(this_syscall, 0, action, match ? match->log : false);
785 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
786 const bool recheck_after_trace)
792 int __secure_computing(const struct seccomp_data *sd)
794 int mode = current->seccomp.mode;
797 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
798 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
801 this_syscall = sd ? sd->nr :
802 syscall_get_nr(current, task_pt_regs(current));
805 case SECCOMP_MODE_STRICT:
806 __secure_computing_strict(this_syscall); /* may call do_exit */
808 case SECCOMP_MODE_FILTER:
809 return __seccomp_filter(this_syscall, sd, false);
814 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
816 long prctl_get_seccomp(void)
818 return current->seccomp.mode;
822 * seccomp_set_mode_strict: internal function for setting strict seccomp
824 * Once current->seccomp.mode is non-zero, it may not be changed.
826 * Returns 0 on success or -EINVAL on failure.
828 static long seccomp_set_mode_strict(void)
830 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
833 spin_lock_irq(¤t->sighand->siglock);
835 if (!seccomp_may_assign_mode(seccomp_mode))
841 seccomp_assign_mode(current, seccomp_mode, 0);
845 spin_unlock_irq(¤t->sighand->siglock);
850 #ifdef CONFIG_SECCOMP_FILTER
852 * seccomp_set_mode_filter: internal function for setting seccomp filter
853 * @flags: flags to change filter behavior
854 * @filter: struct sock_fprog containing filter
856 * This function may be called repeatedly to install additional filters.
857 * Every filter successfully installed will be evaluated (in reverse order)
858 * for each system call the task makes.
860 * Once current->seccomp.mode is non-zero, it may not be changed.
862 * Returns 0 on success or -EINVAL on failure.
864 static long seccomp_set_mode_filter(unsigned int flags,
865 const char __user *filter)
867 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
868 struct seccomp_filter *prepared = NULL;
871 /* Validate flags. */
872 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
875 /* Prepare the new filter before holding any locks. */
876 prepared = seccomp_prepare_user_filter(filter);
877 if (IS_ERR(prepared))
878 return PTR_ERR(prepared);
881 * Make sure we cannot change seccomp or nnp state via TSYNC
882 * while another thread is in the middle of calling exec.
884 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
885 mutex_lock_killable(¤t->signal->cred_guard_mutex))
888 spin_lock_irq(¤t->sighand->siglock);
890 if (!seccomp_may_assign_mode(seccomp_mode))
893 ret = seccomp_attach_filter(flags, prepared);
896 /* Do not free the successfully attached filter. */
899 seccomp_assign_mode(current, seccomp_mode, flags);
901 spin_unlock_irq(¤t->sighand->siglock);
902 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
903 mutex_unlock(¤t->signal->cred_guard_mutex);
905 seccomp_filter_free(prepared);
909 static inline long seccomp_set_mode_filter(unsigned int flags,
910 const char __user *filter)
916 static long seccomp_get_action_avail(const char __user *uaction)
920 if (copy_from_user(&action, uaction, sizeof(action)))
924 case SECCOMP_RET_KILL_PROCESS:
925 case SECCOMP_RET_KILL_THREAD:
926 case SECCOMP_RET_TRAP:
927 case SECCOMP_RET_ERRNO:
928 case SECCOMP_RET_TRACE:
929 case SECCOMP_RET_LOG:
930 case SECCOMP_RET_ALLOW:
939 /* Common entry point for both prctl and syscall. */
940 static long do_seccomp(unsigned int op, unsigned int flags,
941 const char __user *uargs)
944 case SECCOMP_SET_MODE_STRICT:
945 if (flags != 0 || uargs != NULL)
947 return seccomp_set_mode_strict();
948 case SECCOMP_SET_MODE_FILTER:
949 return seccomp_set_mode_filter(flags, uargs);
950 case SECCOMP_GET_ACTION_AVAIL:
954 return seccomp_get_action_avail(uargs);
960 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
961 const char __user *, uargs)
963 return do_seccomp(op, flags, uargs);
967 * prctl_set_seccomp: configures current->seccomp.mode
968 * @seccomp_mode: requested mode to use
969 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
971 * Returns 0 on success or -EINVAL on failure.
973 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
978 switch (seccomp_mode) {
979 case SECCOMP_MODE_STRICT:
980 op = SECCOMP_SET_MODE_STRICT;
982 * Setting strict mode through prctl always ignored filter,
983 * so make sure it is always NULL here to pass the internal
984 * check in do_seccomp().
988 case SECCOMP_MODE_FILTER:
989 op = SECCOMP_SET_MODE_FILTER;
996 /* prctl interface doesn't have flags, so they are always zero. */
997 return do_seccomp(op, 0, uargs);
1000 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
1001 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
1002 unsigned long filter_off)
1004 struct seccomp_filter *orig, *filter;
1005 unsigned long count;
1008 * Note: this is only correct because the caller should be the (ptrace)
1009 * tracer of the task, otherwise lock_task_sighand is needed.
1011 spin_lock_irq(&task->sighand->siglock);
1013 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
1014 spin_unlock_irq(&task->sighand->siglock);
1015 return ERR_PTR(-EINVAL);
1018 orig = task->seccomp.filter;
1019 __get_seccomp_filter(orig);
1020 spin_unlock_irq(&task->sighand->siglock);
1023 for (filter = orig; filter; filter = filter->prev)
1026 if (filter_off >= count) {
1027 filter = ERR_PTR(-ENOENT);
1031 count -= filter_off;
1032 for (filter = orig; filter && count > 1; filter = filter->prev)
1035 if (WARN_ON(count != 1 || !filter)) {
1036 filter = ERR_PTR(-ENOENT);
1040 __get_seccomp_filter(filter);
1043 __put_seccomp_filter(orig);
1047 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1050 struct seccomp_filter *filter;
1051 struct sock_fprog_kern *fprog;
1054 if (!capable(CAP_SYS_ADMIN) ||
1055 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1059 filter = get_nth_filter(task, filter_off);
1061 return PTR_ERR(filter);
1063 fprog = filter->prog->orig_prog;
1065 /* This must be a new non-cBPF filter, since we save
1066 * every cBPF filter's orig_prog above when
1067 * CONFIG_CHECKPOINT_RESTORE is enabled.
1077 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1081 __put_seccomp_filter(filter);
1085 long seccomp_get_metadata(struct task_struct *task,
1086 unsigned long size, void __user *data)
1089 struct seccomp_filter *filter;
1090 struct seccomp_metadata kmd = {};
1092 if (!capable(CAP_SYS_ADMIN) ||
1093 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1097 size = min_t(unsigned long, size, sizeof(kmd));
1099 if (size < sizeof(kmd.filter_off))
1102 if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
1105 filter = get_nth_filter(task, kmd.filter_off);
1107 return PTR_ERR(filter);
1110 kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1113 if (copy_to_user(data, &kmd, size))
1116 __put_seccomp_filter(filter);
1121 #ifdef CONFIG_SYSCTL
1123 /* Human readable action names for friendly sysctl interaction */
1124 #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
1125 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
1126 #define SECCOMP_RET_TRAP_NAME "trap"
1127 #define SECCOMP_RET_ERRNO_NAME "errno"
1128 #define SECCOMP_RET_TRACE_NAME "trace"
1129 #define SECCOMP_RET_LOG_NAME "log"
1130 #define SECCOMP_RET_ALLOW_NAME "allow"
1132 static const char seccomp_actions_avail[] =
1133 SECCOMP_RET_KILL_PROCESS_NAME " "
1134 SECCOMP_RET_KILL_THREAD_NAME " "
1135 SECCOMP_RET_TRAP_NAME " "
1136 SECCOMP_RET_ERRNO_NAME " "
1137 SECCOMP_RET_TRACE_NAME " "
1138 SECCOMP_RET_LOG_NAME " "
1139 SECCOMP_RET_ALLOW_NAME;
1141 struct seccomp_log_name {
1146 static const struct seccomp_log_name seccomp_log_names[] = {
1147 { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1148 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1149 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1150 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1151 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1152 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1153 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1157 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1160 const struct seccomp_log_name *cur;
1161 bool append_space = false;
1163 for (cur = seccomp_log_names; cur->name && size; cur++) {
1166 if (!(actions_logged & cur->log))
1170 ret = strscpy(names, " ", size);
1177 append_space = true;
1179 ret = strscpy(names, cur->name, size);
1190 static bool seccomp_action_logged_from_name(u32 *action_logged,
1193 const struct seccomp_log_name *cur;
1195 for (cur = seccomp_log_names; cur->name; cur++) {
1196 if (!strcmp(cur->name, name)) {
1197 *action_logged = cur->log;
1205 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1209 *actions_logged = 0;
1210 while ((name = strsep(&names, " ")) && *name) {
1211 u32 action_logged = 0;
1213 if (!seccomp_action_logged_from_name(&action_logged, name))
1216 *actions_logged |= action_logged;
1222 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1223 void __user *buffer, size_t *lenp,
1226 char names[sizeof(seccomp_actions_avail)];
1227 struct ctl_table table;
1230 if (write && !capable(CAP_SYS_ADMIN))
1233 memset(names, 0, sizeof(names));
1236 if (!seccomp_names_from_actions_logged(names, sizeof(names),
1237 seccomp_actions_logged))
1243 table.maxlen = sizeof(names);
1244 ret = proc_dostring(&table, write, buffer, lenp, ppos);
1251 if (!seccomp_actions_logged_from_names(&actions_logged,
1255 if (actions_logged & SECCOMP_LOG_ALLOW)
1258 seccomp_actions_logged = actions_logged;
1264 static struct ctl_path seccomp_sysctl_path[] = {
1265 { .procname = "kernel", },
1266 { .procname = "seccomp", },
1270 static struct ctl_table seccomp_sysctl_table[] = {
1272 .procname = "actions_avail",
1273 .data = (void *) &seccomp_actions_avail,
1274 .maxlen = sizeof(seccomp_actions_avail),
1276 .proc_handler = proc_dostring,
1279 .procname = "actions_logged",
1281 .proc_handler = seccomp_actions_logged_handler,
1286 static int __init seccomp_sysctl_init(void)
1288 struct ctl_table_header *hdr;
1290 hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1292 pr_warn("seccomp: sysctl registration failed\n");
1294 kmemleak_not_leak(hdr);
1299 device_initcall(seccomp_sysctl_init)
1301 #endif /* CONFIG_SYSCTL */