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/capability.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)
232 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
234 static inline void seccomp_assign_mode(struct task_struct *task,
235 unsigned long seccomp_mode,
238 assert_spin_locked(&task->sighand->siglock);
240 task->seccomp.mode = seccomp_mode;
242 * Make sure TIF_SECCOMP cannot be set before the mode (and
245 smp_mb__before_atomic();
246 /* Assume default seccomp processes want spec flaw mitigation. */
247 if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
248 arch_seccomp_spec_mitigate(task);
249 set_tsk_thread_flag(task, TIF_SECCOMP);
252 #ifdef CONFIG_SECCOMP_FILTER
253 /* Returns 1 if the parent is an ancestor of the child. */
254 static int is_ancestor(struct seccomp_filter *parent,
255 struct seccomp_filter *child)
257 /* NULL is the root ancestor. */
260 for (; child; child = child->prev)
267 * seccomp_can_sync_threads: checks if all threads can be synchronized
269 * Expects sighand and cred_guard_mutex locks to be held.
271 * Returns 0 on success, -ve on error, or the pid of a thread which was
272 * either not in the correct seccomp mode or it did not have an ancestral
275 static inline pid_t seccomp_can_sync_threads(void)
277 struct task_struct *thread, *caller;
279 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
280 assert_spin_locked(¤t->sighand->siglock);
282 /* Validate all threads being eligible for synchronization. */
284 for_each_thread(caller, thread) {
287 /* Skip current, since it is initiating the sync. */
288 if (thread == caller)
291 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
292 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
293 is_ancestor(thread->seccomp.filter,
294 caller->seccomp.filter)))
297 /* Return the first thread that cannot be synchronized. */
298 failed = task_pid_vnr(thread);
299 /* If the pid cannot be resolved, then return -ESRCH */
300 if (unlikely(WARN_ON(failed == 0)))
309 * seccomp_sync_threads: sets all threads to use current's filter
311 * Expects sighand and cred_guard_mutex locks to be held, and for
312 * seccomp_can_sync_threads() to have returned success already
313 * without dropping the locks.
316 static inline void seccomp_sync_threads(unsigned long flags)
318 struct task_struct *thread, *caller;
320 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
321 assert_spin_locked(¤t->sighand->siglock);
323 /* Synchronize all threads. */
325 for_each_thread(caller, thread) {
326 /* Skip current, since it needs no changes. */
327 if (thread == caller)
330 /* Get a task reference for the new leaf node. */
331 get_seccomp_filter(caller);
333 * Drop the task reference to the shared ancestor since
334 * current's path will hold a reference. (This also
335 * allows a put before the assignment.)
337 put_seccomp_filter(thread);
338 smp_store_release(&thread->seccomp.filter,
339 caller->seccomp.filter);
342 * Don't let an unprivileged task work around
343 * the no_new_privs restriction by creating
344 * a thread that sets it up, enters seccomp,
347 if (task_no_new_privs(caller))
348 task_set_no_new_privs(thread);
351 * Opt the other thread into seccomp if needed.
352 * As threads are considered to be trust-realm
353 * equivalent (see ptrace_may_access), it is safe to
354 * allow one thread to transition the other.
356 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
357 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
363 * seccomp_prepare_filter: Prepares a seccomp filter for use.
364 * @fprog: BPF program to install
366 * Returns filter on success or an ERR_PTR on failure.
368 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
370 struct seccomp_filter *sfilter;
372 const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
374 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
375 return ERR_PTR(-EINVAL);
377 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
380 * Installing a seccomp filter requires that the task has
381 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
382 * This avoids scenarios where unprivileged tasks can affect the
383 * behavior of privileged children.
385 if (!task_no_new_privs(current) &&
386 !ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN))
387 return ERR_PTR(-EACCES);
389 /* Allocate a new seccomp_filter */
390 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
392 return ERR_PTR(-ENOMEM);
394 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
395 seccomp_check_filter, save_orig);
401 refcount_set(&sfilter->usage, 1);
407 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
408 * @user_filter: pointer to the user data containing a sock_fprog.
410 * Returns 0 on success and non-zero otherwise.
412 static struct seccomp_filter *
413 seccomp_prepare_user_filter(const char __user *user_filter)
415 struct sock_fprog fprog;
416 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
419 if (in_compat_syscall()) {
420 struct compat_sock_fprog fprog32;
421 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
423 fprog.len = fprog32.len;
424 fprog.filter = compat_ptr(fprog32.filter);
425 } else /* falls through to the if below. */
427 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
429 filter = seccomp_prepare_filter(&fprog);
435 * seccomp_attach_filter: validate and attach filter
436 * @flags: flags to change filter behavior
437 * @filter: seccomp filter to add to the current process
439 * Caller must be holding current->sighand->siglock lock.
441 * Returns 0 on success, -ve on error.
443 static long seccomp_attach_filter(unsigned int flags,
444 struct seccomp_filter *filter)
446 unsigned long total_insns;
447 struct seccomp_filter *walker;
449 assert_spin_locked(¤t->sighand->siglock);
451 /* Validate resulting filter length. */
452 total_insns = filter->prog->len;
453 for (walker = current->seccomp.filter; walker; walker = walker->prev)
454 total_insns += walker->prog->len + 4; /* 4 instr penalty */
455 if (total_insns > MAX_INSNS_PER_PATH)
458 /* If thread sync has been requested, check that it is possible. */
459 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
462 ret = seccomp_can_sync_threads();
467 /* Set log flag, if present. */
468 if (flags & SECCOMP_FILTER_FLAG_LOG)
472 * If there is an existing filter, make it the prev and don't drop its
475 filter->prev = current->seccomp.filter;
476 current->seccomp.filter = filter;
478 /* Now that the new filter is in place, synchronize to all threads. */
479 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
480 seccomp_sync_threads(flags);
485 static void __get_seccomp_filter(struct seccomp_filter *filter)
487 /* Reference count is bounded by the number of total processes. */
488 refcount_inc(&filter->usage);
491 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
492 void get_seccomp_filter(struct task_struct *tsk)
494 struct seccomp_filter *orig = tsk->seccomp.filter;
497 __get_seccomp_filter(orig);
500 static inline void seccomp_filter_free(struct seccomp_filter *filter)
503 bpf_prog_destroy(filter->prog);
508 static void __put_seccomp_filter(struct seccomp_filter *orig)
510 /* Clean up single-reference branches iteratively. */
511 while (orig && refcount_dec_and_test(&orig->usage)) {
512 struct seccomp_filter *freeme = orig;
514 seccomp_filter_free(freeme);
518 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
519 void put_seccomp_filter(struct task_struct *tsk)
521 __put_seccomp_filter(tsk->seccomp.filter);
524 static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason)
527 info->si_signo = SIGSYS;
528 info->si_code = SYS_SECCOMP;
529 info->si_call_addr = (void __user *)KSTK_EIP(current);
530 info->si_errno = reason;
531 info->si_arch = syscall_get_arch();
532 info->si_syscall = syscall;
536 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
537 * @syscall: syscall number to send to userland
538 * @reason: filter-supplied reason code to send to userland (via si_errno)
540 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
542 static void seccomp_send_sigsys(int syscall, int reason)
545 seccomp_init_siginfo(&info, syscall, reason);
546 force_sig_info(SIGSYS, &info, current);
548 #endif /* CONFIG_SECCOMP_FILTER */
550 /* For use with seccomp_actions_logged */
551 #define SECCOMP_LOG_KILL_PROCESS (1 << 0)
552 #define SECCOMP_LOG_KILL_THREAD (1 << 1)
553 #define SECCOMP_LOG_TRAP (1 << 2)
554 #define SECCOMP_LOG_ERRNO (1 << 3)
555 #define SECCOMP_LOG_TRACE (1 << 4)
556 #define SECCOMP_LOG_LOG (1 << 5)
557 #define SECCOMP_LOG_ALLOW (1 << 6)
559 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
560 SECCOMP_LOG_KILL_THREAD |
566 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
572 case SECCOMP_RET_ALLOW:
574 case SECCOMP_RET_TRAP:
575 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
577 case SECCOMP_RET_ERRNO:
578 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
580 case SECCOMP_RET_TRACE:
581 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
583 case SECCOMP_RET_LOG:
584 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
586 case SECCOMP_RET_KILL_THREAD:
587 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
589 case SECCOMP_RET_KILL_PROCESS:
591 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
595 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
596 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
597 * any action from being logged by removing the action name from the
598 * seccomp_actions_logged sysctl.
603 audit_seccomp(syscall, signr, action);
607 * Secure computing mode 1 allows only read/write/exit/sigreturn.
608 * To be fully secure this must be combined with rlimit
609 * to limit the stack allocations too.
611 static const int mode1_syscalls[] = {
612 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
613 0, /* null terminated */
616 static void __secure_computing_strict(int this_syscall)
618 const int *syscall_whitelist = mode1_syscalls;
620 if (in_compat_syscall())
621 syscall_whitelist = get_compat_mode1_syscalls();
624 if (*syscall_whitelist == this_syscall)
626 } while (*++syscall_whitelist);
631 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
635 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
636 void secure_computing_strict(int this_syscall)
638 int mode = current->seccomp.mode;
640 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
641 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
644 if (mode == SECCOMP_MODE_DISABLED)
646 else if (mode == SECCOMP_MODE_STRICT)
647 __secure_computing_strict(this_syscall);
653 #ifdef CONFIG_SECCOMP_FILTER
654 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
655 const bool recheck_after_trace)
657 u32 filter_ret, action;
658 struct seccomp_filter *match = NULL;
662 * Make sure that any changes to mode from another thread have
663 * been seen after TIF_SECCOMP was seen.
667 filter_ret = seccomp_run_filters(sd, &match);
668 data = filter_ret & SECCOMP_RET_DATA;
669 action = filter_ret & SECCOMP_RET_ACTION_FULL;
672 case SECCOMP_RET_ERRNO:
673 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
674 if (data > MAX_ERRNO)
676 syscall_set_return_value(current, task_pt_regs(current),
680 case SECCOMP_RET_TRAP:
681 /* Show the handler the original registers. */
682 syscall_rollback(current, task_pt_regs(current));
683 /* Let the filter pass back 16 bits of data. */
684 seccomp_send_sigsys(this_syscall, data);
687 case SECCOMP_RET_TRACE:
688 /* We've been put in this state by the ptracer already. */
689 if (recheck_after_trace)
692 /* ENOSYS these calls if there is no tracer attached. */
693 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
694 syscall_set_return_value(current,
695 task_pt_regs(current),
700 /* Allow the BPF to provide the event message */
701 ptrace_event(PTRACE_EVENT_SECCOMP, data);
703 * The delivery of a fatal signal during event
704 * notification may silently skip tracer notification,
705 * which could leave us with a potentially unmodified
706 * syscall that the tracer would have liked to have
707 * changed. Since the process is about to die, we just
708 * force the syscall to be skipped and let the signal
709 * kill the process and correctly handle any tracer exit
712 if (fatal_signal_pending(current))
714 /* Check if the tracer forced the syscall to be skipped. */
715 this_syscall = syscall_get_nr(current, task_pt_regs(current));
716 if (this_syscall < 0)
720 * Recheck the syscall, since it may have changed. This
721 * intentionally uses a NULL struct seccomp_data to force
722 * a reload of all registers. This does not goto skip since
723 * a skip would have already been reported.
725 if (__seccomp_filter(this_syscall, NULL, true))
730 case SECCOMP_RET_LOG:
731 seccomp_log(this_syscall, 0, action, true);
734 case SECCOMP_RET_ALLOW:
736 * Note that the "match" filter will always be NULL for
737 * this action since SECCOMP_RET_ALLOW is the starting
738 * state in seccomp_run_filters().
742 case SECCOMP_RET_KILL_THREAD:
743 case SECCOMP_RET_KILL_PROCESS:
745 seccomp_log(this_syscall, SIGSYS, action, true);
746 /* Dump core only if this is the last remaining thread. */
747 if (action == SECCOMP_RET_KILL_PROCESS ||
748 get_nr_threads(current) == 1) {
751 /* Show the original registers in the dump. */
752 syscall_rollback(current, task_pt_regs(current));
753 /* Trigger a manual coredump since do_exit skips it. */
754 seccomp_init_siginfo(&info, this_syscall, data);
757 if (action == SECCOMP_RET_KILL_PROCESS)
758 do_group_exit(SIGSYS);
766 seccomp_log(this_syscall, 0, action, match ? match->log : false);
770 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
771 const bool recheck_after_trace)
779 int __secure_computing(const struct seccomp_data *sd)
781 int mode = current->seccomp.mode;
784 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
785 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
788 this_syscall = sd ? sd->nr :
789 syscall_get_nr(current, task_pt_regs(current));
792 case SECCOMP_MODE_STRICT:
793 __secure_computing_strict(this_syscall); /* may call do_exit */
795 case SECCOMP_MODE_FILTER:
796 return __seccomp_filter(this_syscall, sd, false);
801 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
803 long prctl_get_seccomp(void)
805 return current->seccomp.mode;
809 * seccomp_set_mode_strict: internal function for setting strict seccomp
811 * Once current->seccomp.mode is non-zero, it may not be changed.
813 * Returns 0 on success or -EINVAL on failure.
815 static long seccomp_set_mode_strict(void)
817 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
820 spin_lock_irq(¤t->sighand->siglock);
822 if (!seccomp_may_assign_mode(seccomp_mode))
828 seccomp_assign_mode(current, seccomp_mode, 0);
832 spin_unlock_irq(¤t->sighand->siglock);
837 #ifdef CONFIG_SECCOMP_FILTER
839 * seccomp_set_mode_filter: internal function for setting seccomp filter
840 * @flags: flags to change filter behavior
841 * @filter: struct sock_fprog containing filter
843 * This function may be called repeatedly to install additional filters.
844 * Every filter successfully installed will be evaluated (in reverse order)
845 * for each system call the task makes.
847 * Once current->seccomp.mode is non-zero, it may not be changed.
849 * Returns 0 on success or -EINVAL on failure.
851 static long seccomp_set_mode_filter(unsigned int flags,
852 const char __user *filter)
854 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
855 struct seccomp_filter *prepared = NULL;
858 /* Validate flags. */
859 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
862 /* Prepare the new filter before holding any locks. */
863 prepared = seccomp_prepare_user_filter(filter);
864 if (IS_ERR(prepared))
865 return PTR_ERR(prepared);
868 * Make sure we cannot change seccomp or nnp state via TSYNC
869 * while another thread is in the middle of calling exec.
871 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
872 mutex_lock_killable(¤t->signal->cred_guard_mutex))
875 spin_lock_irq(¤t->sighand->siglock);
877 if (!seccomp_may_assign_mode(seccomp_mode))
880 ret = seccomp_attach_filter(flags, prepared);
883 /* Do not free the successfully attached filter. */
886 seccomp_assign_mode(current, seccomp_mode, flags);
888 spin_unlock_irq(¤t->sighand->siglock);
889 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
890 mutex_unlock(¤t->signal->cred_guard_mutex);
892 seccomp_filter_free(prepared);
896 static inline long seccomp_set_mode_filter(unsigned int flags,
897 const char __user *filter)
903 static long seccomp_get_action_avail(const char __user *uaction)
907 if (copy_from_user(&action, uaction, sizeof(action)))
911 case SECCOMP_RET_KILL_PROCESS:
912 case SECCOMP_RET_KILL_THREAD:
913 case SECCOMP_RET_TRAP:
914 case SECCOMP_RET_ERRNO:
915 case SECCOMP_RET_TRACE:
916 case SECCOMP_RET_LOG:
917 case SECCOMP_RET_ALLOW:
926 /* Common entry point for both prctl and syscall. */
927 static long do_seccomp(unsigned int op, unsigned int flags,
928 const char __user *uargs)
931 case SECCOMP_SET_MODE_STRICT:
932 if (flags != 0 || uargs != NULL)
934 return seccomp_set_mode_strict();
935 case SECCOMP_SET_MODE_FILTER:
936 return seccomp_set_mode_filter(flags, uargs);
937 case SECCOMP_GET_ACTION_AVAIL:
941 return seccomp_get_action_avail(uargs);
947 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
948 const char __user *, uargs)
950 return do_seccomp(op, flags, uargs);
954 * prctl_set_seccomp: configures current->seccomp.mode
955 * @seccomp_mode: requested mode to use
956 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
958 * Returns 0 on success or -EINVAL on failure.
960 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
965 switch (seccomp_mode) {
966 case SECCOMP_MODE_STRICT:
967 op = SECCOMP_SET_MODE_STRICT;
969 * Setting strict mode through prctl always ignored filter,
970 * so make sure it is always NULL here to pass the internal
971 * check in do_seccomp().
975 case SECCOMP_MODE_FILTER:
976 op = SECCOMP_SET_MODE_FILTER;
983 /* prctl interface doesn't have flags, so they are always zero. */
984 return do_seccomp(op, 0, uargs);
987 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
988 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
989 unsigned long filter_off)
991 struct seccomp_filter *orig, *filter;
995 * Note: this is only correct because the caller should be the (ptrace)
996 * tracer of the task, otherwise lock_task_sighand is needed.
998 spin_lock_irq(&task->sighand->siglock);
1000 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
1001 spin_unlock_irq(&task->sighand->siglock);
1002 return ERR_PTR(-EINVAL);
1005 orig = task->seccomp.filter;
1006 __get_seccomp_filter(orig);
1007 spin_unlock_irq(&task->sighand->siglock);
1010 for (filter = orig; filter; filter = filter->prev)
1013 if (filter_off >= count) {
1014 filter = ERR_PTR(-ENOENT);
1018 count -= filter_off;
1019 for (filter = orig; filter && count > 1; filter = filter->prev)
1022 if (WARN_ON(count != 1 || !filter)) {
1023 filter = ERR_PTR(-ENOENT);
1027 __get_seccomp_filter(filter);
1030 __put_seccomp_filter(orig);
1034 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1037 struct seccomp_filter *filter;
1038 struct sock_fprog_kern *fprog;
1041 if (!capable(CAP_SYS_ADMIN) ||
1042 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1046 filter = get_nth_filter(task, filter_off);
1048 return PTR_ERR(filter);
1050 fprog = filter->prog->orig_prog;
1052 /* This must be a new non-cBPF filter, since we save
1053 * every cBPF filter's orig_prog above when
1054 * CONFIG_CHECKPOINT_RESTORE is enabled.
1064 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1068 __put_seccomp_filter(filter);
1072 long seccomp_get_metadata(struct task_struct *task,
1073 unsigned long size, void __user *data)
1076 struct seccomp_filter *filter;
1077 struct seccomp_metadata kmd = {};
1079 if (!capable(CAP_SYS_ADMIN) ||
1080 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1084 size = min_t(unsigned long, size, sizeof(kmd));
1086 if (size < sizeof(kmd.filter_off))
1089 if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
1092 filter = get_nth_filter(task, kmd.filter_off);
1094 return PTR_ERR(filter);
1097 kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1100 if (copy_to_user(data, &kmd, size))
1103 __put_seccomp_filter(filter);
1108 #ifdef CONFIG_SYSCTL
1110 /* Human readable action names for friendly sysctl interaction */
1111 #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
1112 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
1113 #define SECCOMP_RET_TRAP_NAME "trap"
1114 #define SECCOMP_RET_ERRNO_NAME "errno"
1115 #define SECCOMP_RET_TRACE_NAME "trace"
1116 #define SECCOMP_RET_LOG_NAME "log"
1117 #define SECCOMP_RET_ALLOW_NAME "allow"
1119 static const char seccomp_actions_avail[] =
1120 SECCOMP_RET_KILL_PROCESS_NAME " "
1121 SECCOMP_RET_KILL_THREAD_NAME " "
1122 SECCOMP_RET_TRAP_NAME " "
1123 SECCOMP_RET_ERRNO_NAME " "
1124 SECCOMP_RET_TRACE_NAME " "
1125 SECCOMP_RET_LOG_NAME " "
1126 SECCOMP_RET_ALLOW_NAME;
1128 struct seccomp_log_name {
1133 static const struct seccomp_log_name seccomp_log_names[] = {
1134 { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1135 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1136 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1137 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1138 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1139 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1140 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1144 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1148 const struct seccomp_log_name *cur;
1149 bool append_sep = false;
1151 for (cur = seccomp_log_names; cur->name && size; cur++) {
1154 if (!(actions_logged & cur->log))
1158 ret = strscpy(names, sep, size);
1167 ret = strscpy(names, cur->name, size);
1178 static bool seccomp_action_logged_from_name(u32 *action_logged,
1181 const struct seccomp_log_name *cur;
1183 for (cur = seccomp_log_names; cur->name; cur++) {
1184 if (!strcmp(cur->name, name)) {
1185 *action_logged = cur->log;
1193 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1197 *actions_logged = 0;
1198 while ((name = strsep(&names, " ")) && *name) {
1199 u32 action_logged = 0;
1201 if (!seccomp_action_logged_from_name(&action_logged, name))
1204 *actions_logged |= action_logged;
1210 static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1211 size_t *lenp, loff_t *ppos)
1213 char names[sizeof(seccomp_actions_avail)];
1214 struct ctl_table table;
1216 memset(names, 0, sizeof(names));
1218 if (!seccomp_names_from_actions_logged(names, sizeof(names),
1219 seccomp_actions_logged, " "))
1224 table.maxlen = sizeof(names);
1225 return proc_dostring(&table, 0, buffer, lenp, ppos);
1228 static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1229 size_t *lenp, loff_t *ppos, u32 *actions_logged)
1231 char names[sizeof(seccomp_actions_avail)];
1232 struct ctl_table table;
1235 if (!capable(CAP_SYS_ADMIN))
1238 memset(names, 0, sizeof(names));
1242 table.maxlen = sizeof(names);
1243 ret = proc_dostring(&table, 1, buffer, lenp, ppos);
1247 if (!seccomp_actions_logged_from_names(actions_logged, table.data))
1250 if (*actions_logged & SECCOMP_LOG_ALLOW)
1253 seccomp_actions_logged = *actions_logged;
1257 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
1260 char names[sizeof(seccomp_actions_avail)];
1261 char old_names[sizeof(seccomp_actions_avail)];
1262 const char *new = names;
1263 const char *old = old_names;
1268 memset(names, 0, sizeof(names));
1269 memset(old_names, 0, sizeof(old_names));
1273 else if (!actions_logged)
1275 else if (!seccomp_names_from_actions_logged(names, sizeof(names),
1276 actions_logged, ","))
1279 if (!old_actions_logged)
1281 else if (!seccomp_names_from_actions_logged(old_names,
1283 old_actions_logged, ","))
1286 return audit_seccomp_actions_logged(new, old, !ret);
1289 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1290 void __user *buffer, size_t *lenp,
1296 u32 actions_logged = 0;
1297 u32 old_actions_logged = seccomp_actions_logged;
1299 ret = write_actions_logged(ro_table, buffer, lenp, ppos,
1301 audit_actions_logged(actions_logged, old_actions_logged, ret);
1303 ret = read_actions_logged(ro_table, buffer, lenp, ppos);
1308 static struct ctl_path seccomp_sysctl_path[] = {
1309 { .procname = "kernel", },
1310 { .procname = "seccomp", },
1314 static struct ctl_table seccomp_sysctl_table[] = {
1316 .procname = "actions_avail",
1317 .data = (void *) &seccomp_actions_avail,
1318 .maxlen = sizeof(seccomp_actions_avail),
1320 .proc_handler = proc_dostring,
1323 .procname = "actions_logged",
1325 .proc_handler = seccomp_actions_logged_handler,
1330 static int __init seccomp_sysctl_init(void)
1332 struct ctl_table_header *hdr;
1334 hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1336 pr_warn("seccomp: sysctl registration failed\n");
1338 kmemleak_not_leak(hdr);
1343 device_initcall(seccomp_sysctl_init)
1345 #endif /* CONFIG_SYSCTL */