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.
16 #define pr_fmt(fmt) "seccomp: " fmt
18 #include <linux/refcount.h>
19 #include <linux/audit.h>
20 #include <linux/compat.h>
21 #include <linux/coredump.h>
22 #include <linux/kmemleak.h>
23 #include <linux/nospec.h>
24 #include <linux/prctl.h>
25 #include <linux/sched.h>
26 #include <linux/sched/task_stack.h>
27 #include <linux/seccomp.h>
28 #include <linux/slab.h>
29 #include <linux/syscalls.h>
30 #include <linux/sysctl.h>
32 /* Not exposed in headers: strictly internal use only. */
33 #define SECCOMP_MODE_DEAD (SECCOMP_MODE_FILTER + 1)
35 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
36 #include <asm/syscall.h>
39 #ifdef CONFIG_SECCOMP_FILTER
40 #include <linux/file.h>
41 #include <linux/filter.h>
42 #include <linux/pid.h>
43 #include <linux/ptrace.h>
44 #include <linux/capability.h>
45 #include <linux/tracehook.h>
46 #include <linux/uaccess.h>
47 #include <linux/anon_inodes.h>
48 #include <linux/lockdep.h>
51 * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the
52 * wrong direction flag in the ioctl number. This is the broken one,
53 * which the kernel needs to keep supporting until all userspaces stop
54 * using the wrong command number.
56 #define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR SECCOMP_IOR(2, __u64)
61 SECCOMP_NOTIFY_REPLIED,
64 struct seccomp_knotif {
65 /* The struct pid of the task whose filter triggered the notification */
66 struct task_struct *task;
68 /* The "cookie" for this request; this is unique for this filter. */
72 * The seccomp data. This pointer is valid the entire time this
73 * notification is active, since it comes from __seccomp_filter which
74 * eclipses the entire lifecycle here.
76 const struct seccomp_data *data;
79 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
80 * struct seccomp_knotif is created and starts out in INIT. Once the
81 * handler reads the notification off of an FD, it transitions to SENT.
82 * If a signal is received the state transitions back to INIT and
83 * another message is sent. When the userspace handler replies, state
84 * transitions to REPLIED.
86 enum notify_state state;
88 /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
94 * Signals when this has changed states, such as the listener
95 * dying, a new seccomp addfd message, or changing to REPLIED
97 struct completion ready;
99 struct list_head list;
101 /* outstanding addfd requests */
102 struct list_head addfd;
106 * struct seccomp_kaddfd - container for seccomp_addfd ioctl messages
108 * @file: A reference to the file to install in the other task
109 * @fd: The fd number to install it at. If the fd number is -1, it means the
110 * installing process should allocate the fd as normal.
111 * @flags: The flags for the new file descriptor. At the moment, only O_CLOEXEC
113 * @ret: The return value of the installing process. It is set to the fd num
114 * upon success (>= 0).
115 * @completion: Indicates that the installing process has completed fd
116 * installation, or gone away (either due to successful
120 struct seccomp_kaddfd {
125 /* To only be set on reply */
127 struct completion completion;
128 struct list_head list;
132 * struct notification - container for seccomp userspace notifications. Since
133 * most seccomp filters will not have notification listeners attached and this
134 * structure is fairly large, we store the notification-specific stuff in a
135 * separate structure.
137 * @request: A semaphore that users of this notification can wait on for
138 * changes. Actual reads and writes are still controlled with
139 * filter->notify_lock.
140 * @next_id: The id of the next request.
141 * @notifications: A list of struct seccomp_knotif elements.
143 struct notification {
144 struct semaphore request;
146 struct list_head notifications;
150 * struct seccomp_filter - container for seccomp BPF programs
152 * @refs: Reference count to manage the object lifetime.
153 * A filter's reference count is incremented for each directly
154 * attached task, once for the dependent filter, and if
155 * requested for the user notifier. When @refs reaches zero,
156 * the filter can be freed.
157 * @users: A filter's @users count is incremented for each directly
158 * attached task (filter installation, fork(), thread_sync),
159 * and once for the dependent filter (tracked in filter->prev).
160 * When it reaches zero it indicates that no direct or indirect
161 * users of that filter exist. No new tasks can get associated with
162 * this filter after reaching 0. The @users count is always smaller
163 * or equal to @refs. Hence, reaching 0 for @users does not mean
164 * the filter can be freed.
165 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
166 * @prev: points to a previously installed, or inherited, filter
167 * @prog: the BPF program to evaluate
168 * @notif: the struct that holds all notification related information
169 * @notify_lock: A lock for all notification-related accesses.
170 * @wqh: A wait queue for poll if a notifier is in use.
172 * seccomp_filter objects are organized in a tree linked via the @prev
173 * pointer. For any task, it appears to be a singly-linked list starting
174 * with current->seccomp.filter, the most recently attached or inherited filter.
175 * However, multiple filters may share a @prev node, by way of fork(), which
176 * results in a unidirectional tree existing in memory. This is similar to
177 * how namespaces work.
179 * seccomp_filter objects should never be modified after being attached
180 * to a task_struct (other than @refs).
182 struct seccomp_filter {
186 struct seccomp_filter *prev;
187 struct bpf_prog *prog;
188 struct notification *notif;
189 struct mutex notify_lock;
190 wait_queue_head_t wqh;
193 /* Limit any path through the tree to 256KB worth of instructions. */
194 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
197 * Endianness is explicitly ignored and left for BPF program authors to manage
198 * as per the specific architecture.
200 static void populate_seccomp_data(struct seccomp_data *sd)
203 * Instead of using current_pt_reg(), we're already doing the work
204 * to safely fetch "current", so just use "task" everywhere below.
206 struct task_struct *task = current;
207 struct pt_regs *regs = task_pt_regs(task);
208 unsigned long args[6];
210 sd->nr = syscall_get_nr(task, regs);
211 sd->arch = syscall_get_arch(task);
212 syscall_get_arguments(task, regs, args);
213 sd->args[0] = args[0];
214 sd->args[1] = args[1];
215 sd->args[2] = args[2];
216 sd->args[3] = args[3];
217 sd->args[4] = args[4];
218 sd->args[5] = args[5];
219 sd->instruction_pointer = KSTK_EIP(task);
223 * seccomp_check_filter - verify seccomp filter code
224 * @filter: filter to verify
225 * @flen: length of filter
227 * Takes a previously checked filter (by bpf_check_classic) and
228 * redirects all filter code that loads struct sk_buff data
229 * and related data through seccomp_bpf_load. It also
230 * enforces length and alignment checking of those loads.
232 * Returns 0 if the rule set is legal or -EINVAL if not.
234 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
237 for (pc = 0; pc < flen; pc++) {
238 struct sock_filter *ftest = &filter[pc];
239 u16 code = ftest->code;
243 case BPF_LD | BPF_W | BPF_ABS:
244 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
245 /* 32-bit aligned and not out of bounds. */
246 if (k >= sizeof(struct seccomp_data) || k & 3)
249 case BPF_LD | BPF_W | BPF_LEN:
250 ftest->code = BPF_LD | BPF_IMM;
251 ftest->k = sizeof(struct seccomp_data);
253 case BPF_LDX | BPF_W | BPF_LEN:
254 ftest->code = BPF_LDX | BPF_IMM;
255 ftest->k = sizeof(struct seccomp_data);
257 /* Explicitly include allowed calls. */
258 case BPF_RET | BPF_K:
259 case BPF_RET | BPF_A:
260 case BPF_ALU | BPF_ADD | BPF_K:
261 case BPF_ALU | BPF_ADD | BPF_X:
262 case BPF_ALU | BPF_SUB | BPF_K:
263 case BPF_ALU | BPF_SUB | BPF_X:
264 case BPF_ALU | BPF_MUL | BPF_K:
265 case BPF_ALU | BPF_MUL | BPF_X:
266 case BPF_ALU | BPF_DIV | BPF_K:
267 case BPF_ALU | BPF_DIV | BPF_X:
268 case BPF_ALU | BPF_AND | BPF_K:
269 case BPF_ALU | BPF_AND | BPF_X:
270 case BPF_ALU | BPF_OR | BPF_K:
271 case BPF_ALU | BPF_OR | BPF_X:
272 case BPF_ALU | BPF_XOR | BPF_K:
273 case BPF_ALU | BPF_XOR | BPF_X:
274 case BPF_ALU | BPF_LSH | BPF_K:
275 case BPF_ALU | BPF_LSH | BPF_X:
276 case BPF_ALU | BPF_RSH | BPF_K:
277 case BPF_ALU | BPF_RSH | BPF_X:
278 case BPF_ALU | BPF_NEG:
279 case BPF_LD | BPF_IMM:
280 case BPF_LDX | BPF_IMM:
281 case BPF_MISC | BPF_TAX:
282 case BPF_MISC | BPF_TXA:
283 case BPF_LD | BPF_MEM:
284 case BPF_LDX | BPF_MEM:
287 case BPF_JMP | BPF_JA:
288 case BPF_JMP | BPF_JEQ | BPF_K:
289 case BPF_JMP | BPF_JEQ | BPF_X:
290 case BPF_JMP | BPF_JGE | BPF_K:
291 case BPF_JMP | BPF_JGE | BPF_X:
292 case BPF_JMP | BPF_JGT | BPF_K:
293 case BPF_JMP | BPF_JGT | BPF_X:
294 case BPF_JMP | BPF_JSET | BPF_K:
295 case BPF_JMP | BPF_JSET | BPF_X:
305 * seccomp_run_filters - evaluates all seccomp filters against @sd
306 * @sd: optional seccomp data to be passed to filters
307 * @match: stores struct seccomp_filter that resulted in the return value,
308 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
311 * Returns valid seccomp BPF response codes.
313 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
314 static u32 seccomp_run_filters(const struct seccomp_data *sd,
315 struct seccomp_filter **match)
317 u32 ret = SECCOMP_RET_ALLOW;
318 /* Make sure cross-thread synced filter points somewhere sane. */
319 struct seccomp_filter *f =
320 READ_ONCE(current->seccomp.filter);
322 /* Ensure unexpected behavior doesn't result in failing open. */
323 if (WARN_ON(f == NULL))
324 return SECCOMP_RET_KILL_PROCESS;
327 * All filters in the list are evaluated and the lowest BPF return
328 * value always takes priority (ignoring the DATA).
330 for (; f; f = f->prev) {
331 u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd);
333 if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
340 #endif /* CONFIG_SECCOMP_FILTER */
342 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
344 assert_spin_locked(¤t->sighand->siglock);
346 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
352 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
354 static inline void seccomp_assign_mode(struct task_struct *task,
355 unsigned long seccomp_mode,
358 assert_spin_locked(&task->sighand->siglock);
360 task->seccomp.mode = seccomp_mode;
362 * Make sure TIF_SECCOMP cannot be set before the mode (and
365 smp_mb__before_atomic();
366 /* Assume default seccomp processes want spec flaw mitigation. */
367 if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
368 arch_seccomp_spec_mitigate(task);
369 set_tsk_thread_flag(task, TIF_SECCOMP);
372 #ifdef CONFIG_SECCOMP_FILTER
373 /* Returns 1 if the parent is an ancestor of the child. */
374 static int is_ancestor(struct seccomp_filter *parent,
375 struct seccomp_filter *child)
377 /* NULL is the root ancestor. */
380 for (; child; child = child->prev)
387 * seccomp_can_sync_threads: checks if all threads can be synchronized
389 * Expects sighand and cred_guard_mutex locks to be held.
391 * Returns 0 on success, -ve on error, or the pid of a thread which was
392 * either not in the correct seccomp mode or did not have an ancestral
395 static inline pid_t seccomp_can_sync_threads(void)
397 struct task_struct *thread, *caller;
399 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
400 assert_spin_locked(¤t->sighand->siglock);
402 /* Validate all threads being eligible for synchronization. */
404 for_each_thread(caller, thread) {
407 /* Skip current, since it is initiating the sync. */
408 if (thread == caller)
411 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
412 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
413 is_ancestor(thread->seccomp.filter,
414 caller->seccomp.filter)))
417 /* Return the first thread that cannot be synchronized. */
418 failed = task_pid_vnr(thread);
419 /* If the pid cannot be resolved, then return -ESRCH */
420 if (WARN_ON(failed == 0))
428 static inline void seccomp_filter_free(struct seccomp_filter *filter)
431 bpf_prog_destroy(filter->prog);
436 static void __seccomp_filter_orphan(struct seccomp_filter *orig)
438 while (orig && refcount_dec_and_test(&orig->users)) {
439 if (waitqueue_active(&orig->wqh))
440 wake_up_poll(&orig->wqh, EPOLLHUP);
445 static void __put_seccomp_filter(struct seccomp_filter *orig)
447 /* Clean up single-reference branches iteratively. */
448 while (orig && refcount_dec_and_test(&orig->refs)) {
449 struct seccomp_filter *freeme = orig;
451 seccomp_filter_free(freeme);
455 static void __seccomp_filter_release(struct seccomp_filter *orig)
457 /* Notify about any unused filters in the task's former filter tree. */
458 __seccomp_filter_orphan(orig);
459 /* Finally drop all references to the task's former tree. */
460 __put_seccomp_filter(orig);
464 * seccomp_filter_release - Detach the task from its filter tree,
465 * drop its reference count, and notify
466 * about unused filters
468 * This function should only be called when the task is exiting as
469 * it detaches it from its filter tree. As such, READ_ONCE() and
470 * barriers are not needed here, as would normally be needed.
472 void seccomp_filter_release(struct task_struct *tsk)
474 struct seccomp_filter *orig = tsk->seccomp.filter;
476 /* Detach task from its filter tree. */
477 tsk->seccomp.filter = NULL;
478 __seccomp_filter_release(orig);
482 * seccomp_sync_threads: sets all threads to use current's filter
484 * Expects sighand and cred_guard_mutex locks to be held, and for
485 * seccomp_can_sync_threads() to have returned success already
486 * without dropping the locks.
489 static inline void seccomp_sync_threads(unsigned long flags)
491 struct task_struct *thread, *caller;
493 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
494 assert_spin_locked(¤t->sighand->siglock);
496 /* Synchronize all threads. */
498 for_each_thread(caller, thread) {
499 /* Skip current, since it needs no changes. */
500 if (thread == caller)
503 /* Get a task reference for the new leaf node. */
504 get_seccomp_filter(caller);
507 * Drop the task reference to the shared ancestor since
508 * current's path will hold a reference. (This also
509 * allows a put before the assignment.)
511 __seccomp_filter_release(thread->seccomp.filter);
513 /* Make our new filter tree visible. */
514 smp_store_release(&thread->seccomp.filter,
515 caller->seccomp.filter);
516 atomic_set(&thread->seccomp.filter_count,
517 atomic_read(&caller->seccomp.filter_count));
520 * Don't let an unprivileged task work around
521 * the no_new_privs restriction by creating
522 * a thread that sets it up, enters seccomp,
525 if (task_no_new_privs(caller))
526 task_set_no_new_privs(thread);
529 * Opt the other thread into seccomp if needed.
530 * As threads are considered to be trust-realm
531 * equivalent (see ptrace_may_access), it is safe to
532 * allow one thread to transition the other.
534 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
535 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
541 * seccomp_prepare_filter: Prepares a seccomp filter for use.
542 * @fprog: BPF program to install
544 * Returns filter on success or an ERR_PTR on failure.
546 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
548 struct seccomp_filter *sfilter;
550 const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
552 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
553 return ERR_PTR(-EINVAL);
555 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
558 * Installing a seccomp filter requires that the task has
559 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
560 * This avoids scenarios where unprivileged tasks can affect the
561 * behavior of privileged children.
563 if (!task_no_new_privs(current) &&
564 !ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN))
565 return ERR_PTR(-EACCES);
567 /* Allocate a new seccomp_filter */
568 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
570 return ERR_PTR(-ENOMEM);
572 mutex_init(&sfilter->notify_lock);
573 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
574 seccomp_check_filter, save_orig);
580 refcount_set(&sfilter->refs, 1);
581 refcount_set(&sfilter->users, 1);
582 init_waitqueue_head(&sfilter->wqh);
588 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
589 * @user_filter: pointer to the user data containing a sock_fprog.
591 * Returns 0 on success and non-zero otherwise.
593 static struct seccomp_filter *
594 seccomp_prepare_user_filter(const char __user *user_filter)
596 struct sock_fprog fprog;
597 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
600 if (in_compat_syscall()) {
601 struct compat_sock_fprog fprog32;
602 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
604 fprog.len = fprog32.len;
605 fprog.filter = compat_ptr(fprog32.filter);
606 } else /* falls through to the if below. */
608 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
610 filter = seccomp_prepare_filter(&fprog);
616 * seccomp_attach_filter: validate and attach filter
617 * @flags: flags to change filter behavior
618 * @filter: seccomp filter to add to the current process
620 * Caller must be holding current->sighand->siglock lock.
622 * Returns 0 on success, -ve on error, or
623 * - in TSYNC mode: the pid of a thread which was either not in the correct
624 * seccomp mode or did not have an ancestral seccomp filter
625 * - in NEW_LISTENER mode: the fd of the new listener
627 static long seccomp_attach_filter(unsigned int flags,
628 struct seccomp_filter *filter)
630 unsigned long total_insns;
631 struct seccomp_filter *walker;
633 assert_spin_locked(¤t->sighand->siglock);
635 /* Validate resulting filter length. */
636 total_insns = filter->prog->len;
637 for (walker = current->seccomp.filter; walker; walker = walker->prev)
638 total_insns += walker->prog->len + 4; /* 4 instr penalty */
639 if (total_insns > MAX_INSNS_PER_PATH)
642 /* If thread sync has been requested, check that it is possible. */
643 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
646 ret = seccomp_can_sync_threads();
648 if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
655 /* Set log flag, if present. */
656 if (flags & SECCOMP_FILTER_FLAG_LOG)
660 * If there is an existing filter, make it the prev and don't drop its
663 filter->prev = current->seccomp.filter;
664 current->seccomp.filter = filter;
665 atomic_inc(¤t->seccomp.filter_count);
667 /* Now that the new filter is in place, synchronize to all threads. */
668 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
669 seccomp_sync_threads(flags);
674 static void __get_seccomp_filter(struct seccomp_filter *filter)
676 refcount_inc(&filter->refs);
679 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
680 void get_seccomp_filter(struct task_struct *tsk)
682 struct seccomp_filter *orig = tsk->seccomp.filter;
685 __get_seccomp_filter(orig);
686 refcount_inc(&orig->users);
689 static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason)
692 info->si_signo = SIGSYS;
693 info->si_code = SYS_SECCOMP;
694 info->si_call_addr = (void __user *)KSTK_EIP(current);
695 info->si_errno = reason;
696 info->si_arch = syscall_get_arch(current);
697 info->si_syscall = syscall;
701 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
702 * @syscall: syscall number to send to userland
703 * @reason: filter-supplied reason code to send to userland (via si_errno)
705 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
707 static void seccomp_send_sigsys(int syscall, int reason)
709 struct kernel_siginfo info;
710 seccomp_init_siginfo(&info, syscall, reason);
711 force_sig_info(&info);
713 #endif /* CONFIG_SECCOMP_FILTER */
715 /* For use with seccomp_actions_logged */
716 #define SECCOMP_LOG_KILL_PROCESS (1 << 0)
717 #define SECCOMP_LOG_KILL_THREAD (1 << 1)
718 #define SECCOMP_LOG_TRAP (1 << 2)
719 #define SECCOMP_LOG_ERRNO (1 << 3)
720 #define SECCOMP_LOG_TRACE (1 << 4)
721 #define SECCOMP_LOG_LOG (1 << 5)
722 #define SECCOMP_LOG_ALLOW (1 << 6)
723 #define SECCOMP_LOG_USER_NOTIF (1 << 7)
725 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
726 SECCOMP_LOG_KILL_THREAD |
729 SECCOMP_LOG_USER_NOTIF |
733 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
739 case SECCOMP_RET_ALLOW:
741 case SECCOMP_RET_TRAP:
742 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
744 case SECCOMP_RET_ERRNO:
745 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
747 case SECCOMP_RET_TRACE:
748 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
750 case SECCOMP_RET_USER_NOTIF:
751 log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
753 case SECCOMP_RET_LOG:
754 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
756 case SECCOMP_RET_KILL_THREAD:
757 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
759 case SECCOMP_RET_KILL_PROCESS:
761 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
765 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
766 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
767 * any action from being logged by removing the action name from the
768 * seccomp_actions_logged sysctl.
773 audit_seccomp(syscall, signr, action);
777 * Secure computing mode 1 allows only read/write/exit/sigreturn.
778 * To be fully secure this must be combined with rlimit
779 * to limit the stack allocations too.
781 static const int mode1_syscalls[] = {
782 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
783 -1, /* negative terminated */
786 static void __secure_computing_strict(int this_syscall)
788 const int *allowed_syscalls = mode1_syscalls;
790 if (in_compat_syscall())
791 allowed_syscalls = get_compat_mode1_syscalls();
794 if (*allowed_syscalls == this_syscall)
796 } while (*++allowed_syscalls != -1);
801 current->seccomp.mode = SECCOMP_MODE_DEAD;
802 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
806 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
807 void secure_computing_strict(int this_syscall)
809 int mode = current->seccomp.mode;
811 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
812 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
815 if (mode == SECCOMP_MODE_DISABLED)
817 else if (mode == SECCOMP_MODE_STRICT)
818 __secure_computing_strict(this_syscall);
824 #ifdef CONFIG_SECCOMP_FILTER
825 static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
828 * Note: overflow is ok here, the id just needs to be unique per
831 lockdep_assert_held(&filter->notify_lock);
832 return filter->notif->next_id++;
835 static void seccomp_handle_addfd(struct seccomp_kaddfd *addfd)
838 * Remove the notification, and reset the list pointers, indicating
839 * that it has been handled.
841 list_del_init(&addfd->list);
842 addfd->ret = receive_fd_replace(addfd->fd, addfd->file, addfd->flags);
843 complete(&addfd->completion);
846 static int seccomp_do_user_notification(int this_syscall,
847 struct seccomp_filter *match,
848 const struct seccomp_data *sd)
853 struct seccomp_knotif n = {};
854 struct seccomp_kaddfd *addfd, *tmp;
856 mutex_lock(&match->notify_lock);
862 n.state = SECCOMP_NOTIFY_INIT;
864 n.id = seccomp_next_notify_id(match);
865 init_completion(&n.ready);
866 list_add(&n.list, &match->notif->notifications);
867 INIT_LIST_HEAD(&n.addfd);
869 up(&match->notif->request);
870 wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM);
873 * This is where we wait for a reply from userspace.
876 mutex_unlock(&match->notify_lock);
877 err = wait_for_completion_interruptible(&n.ready);
878 mutex_lock(&match->notify_lock);
882 addfd = list_first_entry_or_null(&n.addfd,
883 struct seccomp_kaddfd, list);
884 /* Check if we were woken up by a addfd message */
886 seccomp_handle_addfd(addfd);
888 } while (n.state != SECCOMP_NOTIFY_REPLIED);
895 /* If there were any pending addfd calls, clear them out */
896 list_for_each_entry_safe(addfd, tmp, &n.addfd, list) {
897 /* The process went away before we got a chance to handle it */
899 list_del_init(&addfd->list);
900 complete(&addfd->completion);
904 * Note that it's possible the listener died in between the time when
905 * we were notified of a response (or a signal) and when we were able to
906 * re-acquire the lock, so only delete from the list if the
907 * notification actually exists.
909 * Also note that this test is only valid because there's no way to
910 * *reattach* to a notifier right now. If one is added, we'll need to
911 * keep track of the notif itself and make sure they match here.
916 mutex_unlock(&match->notify_lock);
918 /* Userspace requests to continue the syscall. */
919 if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE)
922 syscall_set_return_value(current, current_pt_regs(),
927 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
928 const bool recheck_after_trace)
930 u32 filter_ret, action;
931 struct seccomp_filter *match = NULL;
933 struct seccomp_data sd_local;
936 * Make sure that any changes to mode from another thread have
937 * been seen after TIF_SECCOMP was seen.
942 populate_seccomp_data(&sd_local);
946 filter_ret = seccomp_run_filters(sd, &match);
947 data = filter_ret & SECCOMP_RET_DATA;
948 action = filter_ret & SECCOMP_RET_ACTION_FULL;
951 case SECCOMP_RET_ERRNO:
952 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
953 if (data > MAX_ERRNO)
955 syscall_set_return_value(current, current_pt_regs(),
959 case SECCOMP_RET_TRAP:
960 /* Show the handler the original registers. */
961 syscall_rollback(current, current_pt_regs());
962 /* Let the filter pass back 16 bits of data. */
963 seccomp_send_sigsys(this_syscall, data);
966 case SECCOMP_RET_TRACE:
967 /* We've been put in this state by the ptracer already. */
968 if (recheck_after_trace)
971 /* ENOSYS these calls if there is no tracer attached. */
972 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
973 syscall_set_return_value(current,
979 /* Allow the BPF to provide the event message */
980 ptrace_event(PTRACE_EVENT_SECCOMP, data);
982 * The delivery of a fatal signal during event
983 * notification may silently skip tracer notification,
984 * which could leave us with a potentially unmodified
985 * syscall that the tracer would have liked to have
986 * changed. Since the process is about to die, we just
987 * force the syscall to be skipped and let the signal
988 * kill the process and correctly handle any tracer exit
991 if (fatal_signal_pending(current))
993 /* Check if the tracer forced the syscall to be skipped. */
994 this_syscall = syscall_get_nr(current, current_pt_regs());
995 if (this_syscall < 0)
999 * Recheck the syscall, since it may have changed. This
1000 * intentionally uses a NULL struct seccomp_data to force
1001 * a reload of all registers. This does not goto skip since
1002 * a skip would have already been reported.
1004 if (__seccomp_filter(this_syscall, NULL, true))
1009 case SECCOMP_RET_USER_NOTIF:
1010 if (seccomp_do_user_notification(this_syscall, match, sd))
1015 case SECCOMP_RET_LOG:
1016 seccomp_log(this_syscall, 0, action, true);
1019 case SECCOMP_RET_ALLOW:
1021 * Note that the "match" filter will always be NULL for
1022 * this action since SECCOMP_RET_ALLOW is the starting
1023 * state in seccomp_run_filters().
1027 case SECCOMP_RET_KILL_THREAD:
1028 case SECCOMP_RET_KILL_PROCESS:
1030 current->seccomp.mode = SECCOMP_MODE_DEAD;
1031 seccomp_log(this_syscall, SIGSYS, action, true);
1032 /* Dump core only if this is the last remaining thread. */
1033 if (action != SECCOMP_RET_KILL_THREAD ||
1034 get_nr_threads(current) == 1) {
1035 kernel_siginfo_t info;
1037 /* Show the original registers in the dump. */
1038 syscall_rollback(current, current_pt_regs());
1039 /* Trigger a manual coredump since do_exit skips it. */
1040 seccomp_init_siginfo(&info, this_syscall, data);
1043 if (action == SECCOMP_RET_KILL_THREAD)
1046 do_group_exit(SIGSYS);
1052 seccomp_log(this_syscall, 0, action, match ? match->log : false);
1056 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1057 const bool recheck_after_trace)
1065 int __secure_computing(const struct seccomp_data *sd)
1067 int mode = current->seccomp.mode;
1070 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1071 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1074 this_syscall = sd ? sd->nr :
1075 syscall_get_nr(current, current_pt_regs());
1078 case SECCOMP_MODE_STRICT:
1079 __secure_computing_strict(this_syscall); /* may call do_exit */
1081 case SECCOMP_MODE_FILTER:
1082 return __seccomp_filter(this_syscall, sd, false);
1083 /* Surviving SECCOMP_RET_KILL_* must be proactively impossible. */
1084 case SECCOMP_MODE_DEAD:
1092 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1094 long prctl_get_seccomp(void)
1096 return current->seccomp.mode;
1100 * seccomp_set_mode_strict: internal function for setting strict seccomp
1102 * Once current->seccomp.mode is non-zero, it may not be changed.
1104 * Returns 0 on success or -EINVAL on failure.
1106 static long seccomp_set_mode_strict(void)
1108 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
1111 spin_lock_irq(¤t->sighand->siglock);
1113 if (!seccomp_may_assign_mode(seccomp_mode))
1119 seccomp_assign_mode(current, seccomp_mode, 0);
1123 spin_unlock_irq(¤t->sighand->siglock);
1128 #ifdef CONFIG_SECCOMP_FILTER
1129 static void seccomp_notify_free(struct seccomp_filter *filter)
1131 kfree(filter->notif);
1132 filter->notif = NULL;
1135 static void seccomp_notify_detach(struct seccomp_filter *filter)
1137 struct seccomp_knotif *knotif;
1142 mutex_lock(&filter->notify_lock);
1145 * If this file is being closed because e.g. the task who owned it
1146 * died, let's wake everyone up who was waiting on us.
1148 list_for_each_entry(knotif, &filter->notif->notifications, list) {
1149 if (knotif->state == SECCOMP_NOTIFY_REPLIED)
1152 knotif->state = SECCOMP_NOTIFY_REPLIED;
1153 knotif->error = -ENOSYS;
1157 * We do not need to wake up any pending addfd messages, as
1158 * the notifier will do that for us, as this just looks
1159 * like a standard reply.
1161 complete(&knotif->ready);
1164 seccomp_notify_free(filter);
1165 mutex_unlock(&filter->notify_lock);
1168 static int seccomp_notify_release(struct inode *inode, struct file *file)
1170 struct seccomp_filter *filter = file->private_data;
1172 seccomp_notify_detach(filter);
1173 __put_seccomp_filter(filter);
1177 /* must be called with notif_lock held */
1178 static inline struct seccomp_knotif *
1179 find_notification(struct seccomp_filter *filter, u64 id)
1181 struct seccomp_knotif *cur;
1183 lockdep_assert_held(&filter->notify_lock);
1185 list_for_each_entry(cur, &filter->notif->notifications, list) {
1194 static long seccomp_notify_recv(struct seccomp_filter *filter,
1197 struct seccomp_knotif *knotif = NULL, *cur;
1198 struct seccomp_notif unotif;
1201 /* Verify that we're not given garbage to keep struct extensible. */
1202 ret = check_zeroed_user(buf, sizeof(unotif));
1208 memset(&unotif, 0, sizeof(unotif));
1210 ret = down_interruptible(&filter->notif->request);
1214 mutex_lock(&filter->notify_lock);
1215 list_for_each_entry(cur, &filter->notif->notifications, list) {
1216 if (cur->state == SECCOMP_NOTIFY_INIT) {
1223 * If we didn't find a notification, it could be that the task was
1224 * interrupted by a fatal signal between the time we were woken and
1225 * when we were able to acquire the rw lock.
1232 unotif.id = knotif->id;
1233 unotif.pid = task_pid_vnr(knotif->task);
1234 unotif.data = *(knotif->data);
1236 knotif->state = SECCOMP_NOTIFY_SENT;
1237 wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM);
1240 mutex_unlock(&filter->notify_lock);
1242 if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1246 * Userspace screwed up. To make sure that we keep this
1247 * notification alive, let's reset it back to INIT. It
1248 * may have died when we released the lock, so we need to make
1249 * sure it's still around.
1251 mutex_lock(&filter->notify_lock);
1252 knotif = find_notification(filter, unotif.id);
1254 knotif->state = SECCOMP_NOTIFY_INIT;
1255 up(&filter->notif->request);
1257 mutex_unlock(&filter->notify_lock);
1263 static long seccomp_notify_send(struct seccomp_filter *filter,
1266 struct seccomp_notif_resp resp = {};
1267 struct seccomp_knotif *knotif;
1270 if (copy_from_user(&resp, buf, sizeof(resp)))
1273 if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1276 if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) &&
1277 (resp.error || resp.val))
1280 ret = mutex_lock_interruptible(&filter->notify_lock);
1284 knotif = find_notification(filter, resp.id);
1290 /* Allow exactly one reply. */
1291 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1297 knotif->state = SECCOMP_NOTIFY_REPLIED;
1298 knotif->error = resp.error;
1299 knotif->val = resp.val;
1300 knotif->flags = resp.flags;
1301 complete(&knotif->ready);
1303 mutex_unlock(&filter->notify_lock);
1307 static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1310 struct seccomp_knotif *knotif;
1314 if (copy_from_user(&id, buf, sizeof(id)))
1317 ret = mutex_lock_interruptible(&filter->notify_lock);
1321 knotif = find_notification(filter, id);
1322 if (knotif && knotif->state == SECCOMP_NOTIFY_SENT)
1327 mutex_unlock(&filter->notify_lock);
1331 static long seccomp_notify_addfd(struct seccomp_filter *filter,
1332 struct seccomp_notif_addfd __user *uaddfd,
1335 struct seccomp_notif_addfd addfd;
1336 struct seccomp_knotif *knotif;
1337 struct seccomp_kaddfd kaddfd;
1340 BUILD_BUG_ON(sizeof(addfd) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0);
1341 BUILD_BUG_ON(sizeof(addfd) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST);
1343 if (size < SECCOMP_NOTIFY_ADDFD_SIZE_VER0 || size >= PAGE_SIZE)
1346 ret = copy_struct_from_user(&addfd, sizeof(addfd), uaddfd, size);
1350 if (addfd.newfd_flags & ~O_CLOEXEC)
1353 if (addfd.flags & ~SECCOMP_ADDFD_FLAG_SETFD)
1356 if (addfd.newfd && !(addfd.flags & SECCOMP_ADDFD_FLAG_SETFD))
1359 kaddfd.file = fget(addfd.srcfd);
1363 kaddfd.flags = addfd.newfd_flags;
1364 kaddfd.fd = (addfd.flags & SECCOMP_ADDFD_FLAG_SETFD) ?
1366 init_completion(&kaddfd.completion);
1368 ret = mutex_lock_interruptible(&filter->notify_lock);
1372 knotif = find_notification(filter, addfd.id);
1379 * We do not want to allow for FD injection to occur before the
1380 * notification has been picked up by a userspace handler, or after
1381 * the notification has been replied to.
1383 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1388 list_add(&kaddfd.list, &knotif->addfd);
1389 complete(&knotif->ready);
1390 mutex_unlock(&filter->notify_lock);
1392 /* Now we wait for it to be processed or be interrupted */
1393 ret = wait_for_completion_interruptible(&kaddfd.completion);
1396 * We had a successful completion. The other side has already
1397 * removed us from the addfd queue, and
1398 * wait_for_completion_interruptible has a memory barrier upon
1399 * success that lets us read this value directly without
1406 mutex_lock(&filter->notify_lock);
1408 * Even though we were woken up by a signal and not a successful
1409 * completion, a completion may have happened in the mean time.
1411 * We need to check again if the addfd request has been handled,
1412 * and if not, we will remove it from the queue.
1414 if (list_empty(&kaddfd.list))
1417 list_del(&kaddfd.list);
1420 mutex_unlock(&filter->notify_lock);
1427 static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1430 struct seccomp_filter *filter = file->private_data;
1431 void __user *buf = (void __user *)arg;
1433 /* Fixed-size ioctls */
1435 case SECCOMP_IOCTL_NOTIF_RECV:
1436 return seccomp_notify_recv(filter, buf);
1437 case SECCOMP_IOCTL_NOTIF_SEND:
1438 return seccomp_notify_send(filter, buf);
1439 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR:
1440 case SECCOMP_IOCTL_NOTIF_ID_VALID:
1441 return seccomp_notify_id_valid(filter, buf);
1444 /* Extensible Argument ioctls */
1445 #define EA_IOCTL(cmd) ((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1446 switch (EA_IOCTL(cmd)) {
1447 case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD):
1448 return seccomp_notify_addfd(filter, buf, _IOC_SIZE(cmd));
1454 static __poll_t seccomp_notify_poll(struct file *file,
1455 struct poll_table_struct *poll_tab)
1457 struct seccomp_filter *filter = file->private_data;
1459 struct seccomp_knotif *cur;
1461 poll_wait(file, &filter->wqh, poll_tab);
1463 if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1466 list_for_each_entry(cur, &filter->notif->notifications, list) {
1467 if (cur->state == SECCOMP_NOTIFY_INIT)
1468 ret |= EPOLLIN | EPOLLRDNORM;
1469 if (cur->state == SECCOMP_NOTIFY_SENT)
1470 ret |= EPOLLOUT | EPOLLWRNORM;
1471 if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1475 mutex_unlock(&filter->notify_lock);
1477 if (refcount_read(&filter->users) == 0)
1483 static const struct file_operations seccomp_notify_ops = {
1484 .poll = seccomp_notify_poll,
1485 .release = seccomp_notify_release,
1486 .unlocked_ioctl = seccomp_notify_ioctl,
1487 .compat_ioctl = seccomp_notify_ioctl,
1490 static struct file *init_listener(struct seccomp_filter *filter)
1494 ret = ERR_PTR(-ENOMEM);
1495 filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1499 sema_init(&filter->notif->request, 0);
1500 filter->notif->next_id = get_random_u64();
1501 INIT_LIST_HEAD(&filter->notif->notifications);
1503 ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1508 /* The file has a reference to it now */
1509 __get_seccomp_filter(filter);
1513 seccomp_notify_free(filter);
1519 * Does @new_child have a listener while an ancestor also has a listener?
1520 * If so, we'll want to reject this filter.
1521 * This only has to be tested for the current process, even in the TSYNC case,
1522 * because TSYNC installs @child with the same parent on all threads.
1523 * Note that @new_child is not hooked up to its parent at this point yet, so
1524 * we use current->seccomp.filter.
1526 static bool has_duplicate_listener(struct seccomp_filter *new_child)
1528 struct seccomp_filter *cur;
1530 /* must be protected against concurrent TSYNC */
1531 lockdep_assert_held(¤t->sighand->siglock);
1533 if (!new_child->notif)
1535 for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1544 * seccomp_set_mode_filter: internal function for setting seccomp filter
1545 * @flags: flags to change filter behavior
1546 * @filter: struct sock_fprog containing filter
1548 * This function may be called repeatedly to install additional filters.
1549 * Every filter successfully installed will be evaluated (in reverse order)
1550 * for each system call the task makes.
1552 * Once current->seccomp.mode is non-zero, it may not be changed.
1554 * Returns 0 on success or -EINVAL on failure.
1556 static long seccomp_set_mode_filter(unsigned int flags,
1557 const char __user *filter)
1559 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1560 struct seccomp_filter *prepared = NULL;
1563 struct file *listener_f = NULL;
1565 /* Validate flags. */
1566 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1570 * In the successful case, NEW_LISTENER returns the new listener fd.
1571 * But in the failure case, TSYNC returns the thread that died. If you
1572 * combine these two flags, there's no way to tell whether something
1573 * succeeded or failed. So, let's disallow this combination if the user
1574 * has not explicitly requested no errors from TSYNC.
1576 if ((flags & SECCOMP_FILTER_FLAG_TSYNC) &&
1577 (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) &&
1578 ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0))
1581 /* Prepare the new filter before holding any locks. */
1582 prepared = seccomp_prepare_user_filter(filter);
1583 if (IS_ERR(prepared))
1584 return PTR_ERR(prepared);
1586 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1587 listener = get_unused_fd_flags(O_CLOEXEC);
1593 listener_f = init_listener(prepared);
1594 if (IS_ERR(listener_f)) {
1595 put_unused_fd(listener);
1596 ret = PTR_ERR(listener_f);
1602 * Make sure we cannot change seccomp or nnp state via TSYNC
1603 * while another thread is in the middle of calling exec.
1605 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1606 mutex_lock_killable(¤t->signal->cred_guard_mutex))
1609 spin_lock_irq(¤t->sighand->siglock);
1611 if (!seccomp_may_assign_mode(seccomp_mode))
1614 if (has_duplicate_listener(prepared)) {
1619 ret = seccomp_attach_filter(flags, prepared);
1622 /* Do not free the successfully attached filter. */
1625 seccomp_assign_mode(current, seccomp_mode, flags);
1627 spin_unlock_irq(¤t->sighand->siglock);
1628 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
1629 mutex_unlock(¤t->signal->cred_guard_mutex);
1631 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1633 listener_f->private_data = NULL;
1635 put_unused_fd(listener);
1636 seccomp_notify_detach(prepared);
1638 fd_install(listener, listener_f);
1643 seccomp_filter_free(prepared);
1647 static inline long seccomp_set_mode_filter(unsigned int flags,
1648 const char __user *filter)
1654 static long seccomp_get_action_avail(const char __user *uaction)
1658 if (copy_from_user(&action, uaction, sizeof(action)))
1662 case SECCOMP_RET_KILL_PROCESS:
1663 case SECCOMP_RET_KILL_THREAD:
1664 case SECCOMP_RET_TRAP:
1665 case SECCOMP_RET_ERRNO:
1666 case SECCOMP_RET_USER_NOTIF:
1667 case SECCOMP_RET_TRACE:
1668 case SECCOMP_RET_LOG:
1669 case SECCOMP_RET_ALLOW:
1678 static long seccomp_get_notif_sizes(void __user *usizes)
1680 struct seccomp_notif_sizes sizes = {
1681 .seccomp_notif = sizeof(struct seccomp_notif),
1682 .seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
1683 .seccomp_data = sizeof(struct seccomp_data),
1686 if (copy_to_user(usizes, &sizes, sizeof(sizes)))
1692 /* Common entry point for both prctl and syscall. */
1693 static long do_seccomp(unsigned int op, unsigned int flags,
1697 case SECCOMP_SET_MODE_STRICT:
1698 if (flags != 0 || uargs != NULL)
1700 return seccomp_set_mode_strict();
1701 case SECCOMP_SET_MODE_FILTER:
1702 return seccomp_set_mode_filter(flags, uargs);
1703 case SECCOMP_GET_ACTION_AVAIL:
1707 return seccomp_get_action_avail(uargs);
1708 case SECCOMP_GET_NOTIF_SIZES:
1712 return seccomp_get_notif_sizes(uargs);
1718 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
1719 void __user *, uargs)
1721 return do_seccomp(op, flags, uargs);
1725 * prctl_set_seccomp: configures current->seccomp.mode
1726 * @seccomp_mode: requested mode to use
1727 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1729 * Returns 0 on success or -EINVAL on failure.
1731 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
1736 switch (seccomp_mode) {
1737 case SECCOMP_MODE_STRICT:
1738 op = SECCOMP_SET_MODE_STRICT;
1740 * Setting strict mode through prctl always ignored filter,
1741 * so make sure it is always NULL here to pass the internal
1742 * check in do_seccomp().
1746 case SECCOMP_MODE_FILTER:
1747 op = SECCOMP_SET_MODE_FILTER;
1754 /* prctl interface doesn't have flags, so they are always zero. */
1755 return do_seccomp(op, 0, uargs);
1758 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
1759 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
1760 unsigned long filter_off)
1762 struct seccomp_filter *orig, *filter;
1763 unsigned long count;
1766 * Note: this is only correct because the caller should be the (ptrace)
1767 * tracer of the task, otherwise lock_task_sighand is needed.
1769 spin_lock_irq(&task->sighand->siglock);
1771 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
1772 spin_unlock_irq(&task->sighand->siglock);
1773 return ERR_PTR(-EINVAL);
1776 orig = task->seccomp.filter;
1777 __get_seccomp_filter(orig);
1778 spin_unlock_irq(&task->sighand->siglock);
1781 for (filter = orig; filter; filter = filter->prev)
1784 if (filter_off >= count) {
1785 filter = ERR_PTR(-ENOENT);
1789 count -= filter_off;
1790 for (filter = orig; filter && count > 1; filter = filter->prev)
1793 if (WARN_ON(count != 1 || !filter)) {
1794 filter = ERR_PTR(-ENOENT);
1798 __get_seccomp_filter(filter);
1801 __put_seccomp_filter(orig);
1805 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1808 struct seccomp_filter *filter;
1809 struct sock_fprog_kern *fprog;
1812 if (!capable(CAP_SYS_ADMIN) ||
1813 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1817 filter = get_nth_filter(task, filter_off);
1819 return PTR_ERR(filter);
1821 fprog = filter->prog->orig_prog;
1823 /* This must be a new non-cBPF filter, since we save
1824 * every cBPF filter's orig_prog above when
1825 * CONFIG_CHECKPOINT_RESTORE is enabled.
1835 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1839 __put_seccomp_filter(filter);
1843 long seccomp_get_metadata(struct task_struct *task,
1844 unsigned long size, void __user *data)
1847 struct seccomp_filter *filter;
1848 struct seccomp_metadata kmd = {};
1850 if (!capable(CAP_SYS_ADMIN) ||
1851 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1855 size = min_t(unsigned long, size, sizeof(kmd));
1857 if (size < sizeof(kmd.filter_off))
1860 if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
1863 filter = get_nth_filter(task, kmd.filter_off);
1865 return PTR_ERR(filter);
1868 kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1871 if (copy_to_user(data, &kmd, size))
1874 __put_seccomp_filter(filter);
1879 #ifdef CONFIG_SYSCTL
1881 /* Human readable action names for friendly sysctl interaction */
1882 #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
1883 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
1884 #define SECCOMP_RET_TRAP_NAME "trap"
1885 #define SECCOMP_RET_ERRNO_NAME "errno"
1886 #define SECCOMP_RET_USER_NOTIF_NAME "user_notif"
1887 #define SECCOMP_RET_TRACE_NAME "trace"
1888 #define SECCOMP_RET_LOG_NAME "log"
1889 #define SECCOMP_RET_ALLOW_NAME "allow"
1891 static const char seccomp_actions_avail[] =
1892 SECCOMP_RET_KILL_PROCESS_NAME " "
1893 SECCOMP_RET_KILL_THREAD_NAME " "
1894 SECCOMP_RET_TRAP_NAME " "
1895 SECCOMP_RET_ERRNO_NAME " "
1896 SECCOMP_RET_USER_NOTIF_NAME " "
1897 SECCOMP_RET_TRACE_NAME " "
1898 SECCOMP_RET_LOG_NAME " "
1899 SECCOMP_RET_ALLOW_NAME;
1901 struct seccomp_log_name {
1906 static const struct seccomp_log_name seccomp_log_names[] = {
1907 { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1908 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1909 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1910 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1911 { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
1912 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1913 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1914 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1918 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1922 const struct seccomp_log_name *cur;
1923 bool append_sep = false;
1925 for (cur = seccomp_log_names; cur->name && size; cur++) {
1928 if (!(actions_logged & cur->log))
1932 ret = strscpy(names, sep, size);
1941 ret = strscpy(names, cur->name, size);
1952 static bool seccomp_action_logged_from_name(u32 *action_logged,
1955 const struct seccomp_log_name *cur;
1957 for (cur = seccomp_log_names; cur->name; cur++) {
1958 if (!strcmp(cur->name, name)) {
1959 *action_logged = cur->log;
1967 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1971 *actions_logged = 0;
1972 while ((name = strsep(&names, " ")) && *name) {
1973 u32 action_logged = 0;
1975 if (!seccomp_action_logged_from_name(&action_logged, name))
1978 *actions_logged |= action_logged;
1984 static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1985 size_t *lenp, loff_t *ppos)
1987 char names[sizeof(seccomp_actions_avail)];
1988 struct ctl_table table;
1990 memset(names, 0, sizeof(names));
1992 if (!seccomp_names_from_actions_logged(names, sizeof(names),
1993 seccomp_actions_logged, " "))
1998 table.maxlen = sizeof(names);
1999 return proc_dostring(&table, 0, buffer, lenp, ppos);
2002 static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer,
2003 size_t *lenp, loff_t *ppos, u32 *actions_logged)
2005 char names[sizeof(seccomp_actions_avail)];
2006 struct ctl_table table;
2009 if (!capable(CAP_SYS_ADMIN))
2012 memset(names, 0, sizeof(names));
2016 table.maxlen = sizeof(names);
2017 ret = proc_dostring(&table, 1, buffer, lenp, ppos);
2021 if (!seccomp_actions_logged_from_names(actions_logged, table.data))
2024 if (*actions_logged & SECCOMP_LOG_ALLOW)
2027 seccomp_actions_logged = *actions_logged;
2031 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
2034 char names[sizeof(seccomp_actions_avail)];
2035 char old_names[sizeof(seccomp_actions_avail)];
2036 const char *new = names;
2037 const char *old = old_names;
2042 memset(names, 0, sizeof(names));
2043 memset(old_names, 0, sizeof(old_names));
2047 else if (!actions_logged)
2049 else if (!seccomp_names_from_actions_logged(names, sizeof(names),
2050 actions_logged, ","))
2053 if (!old_actions_logged)
2055 else if (!seccomp_names_from_actions_logged(old_names,
2057 old_actions_logged, ","))
2060 return audit_seccomp_actions_logged(new, old, !ret);
2063 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
2064 void *buffer, size_t *lenp,
2070 u32 actions_logged = 0;
2071 u32 old_actions_logged = seccomp_actions_logged;
2073 ret = write_actions_logged(ro_table, buffer, lenp, ppos,
2075 audit_actions_logged(actions_logged, old_actions_logged, ret);
2077 ret = read_actions_logged(ro_table, buffer, lenp, ppos);
2082 static struct ctl_path seccomp_sysctl_path[] = {
2083 { .procname = "kernel", },
2084 { .procname = "seccomp", },
2088 static struct ctl_table seccomp_sysctl_table[] = {
2090 .procname = "actions_avail",
2091 .data = (void *) &seccomp_actions_avail,
2092 .maxlen = sizeof(seccomp_actions_avail),
2094 .proc_handler = proc_dostring,
2097 .procname = "actions_logged",
2099 .proc_handler = seccomp_actions_logged_handler,
2104 static int __init seccomp_sysctl_init(void)
2106 struct ctl_table_header *hdr;
2108 hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
2110 pr_warn("sysctl registration failed\n");
2112 kmemleak_not_leak(hdr);
2117 device_initcall(seccomp_sysctl_init)
2119 #endif /* CONFIG_SYSCTL */