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 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
33 #include <asm/syscall.h>
36 #ifdef CONFIG_SECCOMP_FILTER
37 #include <linux/file.h>
38 #include <linux/filter.h>
39 #include <linux/pid.h>
40 #include <linux/ptrace.h>
41 #include <linux/security.h>
42 #include <linux/tracehook.h>
43 #include <linux/uaccess.h>
44 #include <linux/anon_inodes.h>
45 #include <linux/lockdep.h>
48 * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the
49 * wrong direction flag in the ioctl number. This is the broken one,
50 * which the kernel needs to keep supporting until all userspaces stop
51 * using the wrong command number.
53 #define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR SECCOMP_IOR(2, __u64)
58 SECCOMP_NOTIFY_REPLIED,
61 struct seccomp_knotif {
62 /* The struct pid of the task whose filter triggered the notification */
63 struct task_struct *task;
65 /* The "cookie" for this request; this is unique for this filter. */
69 * The seccomp data. This pointer is valid the entire time this
70 * notification is active, since it comes from __seccomp_filter which
71 * eclipses the entire lifecycle here.
73 const struct seccomp_data *data;
76 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
77 * struct seccomp_knotif is created and starts out in INIT. Once the
78 * handler reads the notification off of an FD, it transitions to SENT.
79 * If a signal is received the state transitions back to INIT and
80 * another message is sent. When the userspace handler replies, state
81 * transitions to REPLIED.
83 enum notify_state state;
85 /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
90 /* Signals when this has entered SECCOMP_NOTIFY_REPLIED */
91 struct completion ready;
93 struct list_head list;
97 * struct notification - container for seccomp userspace notifications. Since
98 * most seccomp filters will not have notification listeners attached and this
99 * structure is fairly large, we store the notification-specific stuff in a
100 * separate structure.
102 * @request: A semaphore that users of this notification can wait on for
103 * changes. Actual reads and writes are still controlled with
104 * filter->notify_lock.
105 * @next_id: The id of the next request.
106 * @notifications: A list of struct seccomp_knotif elements.
108 struct notification {
109 struct semaphore request;
111 struct list_head notifications;
115 * struct seccomp_filter - container for seccomp BPF programs
117 * @refs: Reference count to manage the object lifetime.
118 * A filter's reference count is incremented for each directly
119 * attached task, once for the dependent filter, and if
120 * requested for the user notifier. When @refs reaches zero,
121 * the filter can be freed.
122 * @users: A filter's @users count is incremented for each directly
123 * attached task (filter installation, fork(), thread_sync),
124 * and once for the dependent filter (tracked in filter->prev).
125 * When it reaches zero it indicates that no direct or indirect
126 * users of that filter exist. No new tasks can get associated with
127 * this filter after reaching 0. The @users count is always smaller
128 * or equal to @refs. Hence, reaching 0 for @users does not mean
129 * the filter can be freed.
130 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
131 * @prev: points to a previously installed, or inherited, filter
132 * @prog: the BPF program to evaluate
133 * @notif: the struct that holds all notification related information
134 * @notify_lock: A lock for all notification-related accesses.
135 * @wqh: A wait queue for poll if a notifier is in use.
137 * seccomp_filter objects are organized in a tree linked via the @prev
138 * pointer. For any task, it appears to be a singly-linked list starting
139 * with current->seccomp.filter, the most recently attached or inherited filter.
140 * However, multiple filters may share a @prev node, by way of fork(), which
141 * results in a unidirectional tree existing in memory. This is similar to
142 * how namespaces work.
144 * seccomp_filter objects should never be modified after being attached
145 * to a task_struct (other than @refs).
147 struct seccomp_filter {
151 struct seccomp_filter *prev;
152 struct bpf_prog *prog;
153 struct notification *notif;
154 struct mutex notify_lock;
155 wait_queue_head_t wqh;
158 /* Limit any path through the tree to 256KB worth of instructions. */
159 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
162 * Endianness is explicitly ignored and left for BPF program authors to manage
163 * as per the specific architecture.
165 static void populate_seccomp_data(struct seccomp_data *sd)
167 struct task_struct *task = current;
168 struct pt_regs *regs = task_pt_regs(task);
169 unsigned long args[6];
171 sd->nr = syscall_get_nr(task, regs);
172 sd->arch = syscall_get_arch(task);
173 syscall_get_arguments(task, regs, args);
174 sd->args[0] = args[0];
175 sd->args[1] = args[1];
176 sd->args[2] = args[2];
177 sd->args[3] = args[3];
178 sd->args[4] = args[4];
179 sd->args[5] = args[5];
180 sd->instruction_pointer = KSTK_EIP(task);
184 * seccomp_check_filter - verify seccomp filter code
185 * @filter: filter to verify
186 * @flen: length of filter
188 * Takes a previously checked filter (by bpf_check_classic) and
189 * redirects all filter code that loads struct sk_buff data
190 * and related data through seccomp_bpf_load. It also
191 * enforces length and alignment checking of those loads.
193 * Returns 0 if the rule set is legal or -EINVAL if not.
195 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
198 for (pc = 0; pc < flen; pc++) {
199 struct sock_filter *ftest = &filter[pc];
200 u16 code = ftest->code;
204 case BPF_LD | BPF_W | BPF_ABS:
205 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
206 /* 32-bit aligned and not out of bounds. */
207 if (k >= sizeof(struct seccomp_data) || k & 3)
210 case BPF_LD | BPF_W | BPF_LEN:
211 ftest->code = BPF_LD | BPF_IMM;
212 ftest->k = sizeof(struct seccomp_data);
214 case BPF_LDX | BPF_W | BPF_LEN:
215 ftest->code = BPF_LDX | BPF_IMM;
216 ftest->k = sizeof(struct seccomp_data);
218 /* Explicitly include allowed calls. */
219 case BPF_RET | BPF_K:
220 case BPF_RET | BPF_A:
221 case BPF_ALU | BPF_ADD | BPF_K:
222 case BPF_ALU | BPF_ADD | BPF_X:
223 case BPF_ALU | BPF_SUB | BPF_K:
224 case BPF_ALU | BPF_SUB | BPF_X:
225 case BPF_ALU | BPF_MUL | BPF_K:
226 case BPF_ALU | BPF_MUL | BPF_X:
227 case BPF_ALU | BPF_DIV | BPF_K:
228 case BPF_ALU | BPF_DIV | BPF_X:
229 case BPF_ALU | BPF_AND | BPF_K:
230 case BPF_ALU | BPF_AND | BPF_X:
231 case BPF_ALU | BPF_OR | BPF_K:
232 case BPF_ALU | BPF_OR | BPF_X:
233 case BPF_ALU | BPF_XOR | BPF_K:
234 case BPF_ALU | BPF_XOR | BPF_X:
235 case BPF_ALU | BPF_LSH | BPF_K:
236 case BPF_ALU | BPF_LSH | BPF_X:
237 case BPF_ALU | BPF_RSH | BPF_K:
238 case BPF_ALU | BPF_RSH | BPF_X:
239 case BPF_ALU | BPF_NEG:
240 case BPF_LD | BPF_IMM:
241 case BPF_LDX | BPF_IMM:
242 case BPF_MISC | BPF_TAX:
243 case BPF_MISC | BPF_TXA:
244 case BPF_LD | BPF_MEM:
245 case BPF_LDX | BPF_MEM:
248 case BPF_JMP | BPF_JA:
249 case BPF_JMP | BPF_JEQ | BPF_K:
250 case BPF_JMP | BPF_JEQ | BPF_X:
251 case BPF_JMP | BPF_JGE | BPF_K:
252 case BPF_JMP | BPF_JGE | BPF_X:
253 case BPF_JMP | BPF_JGT | BPF_K:
254 case BPF_JMP | BPF_JGT | BPF_X:
255 case BPF_JMP | BPF_JSET | BPF_K:
256 case BPF_JMP | BPF_JSET | BPF_X:
266 * seccomp_run_filters - evaluates all seccomp filters against @sd
267 * @sd: optional seccomp data to be passed to filters
268 * @match: stores struct seccomp_filter that resulted in the return value,
269 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
272 * Returns valid seccomp BPF response codes.
274 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
275 static u32 seccomp_run_filters(const struct seccomp_data *sd,
276 struct seccomp_filter **match)
278 u32 ret = SECCOMP_RET_ALLOW;
279 /* Make sure cross-thread synced filter points somewhere sane. */
280 struct seccomp_filter *f =
281 READ_ONCE(current->seccomp.filter);
283 /* Ensure unexpected behavior doesn't result in failing open. */
284 if (WARN_ON(f == NULL))
285 return SECCOMP_RET_KILL_PROCESS;
288 * All filters in the list are evaluated and the lowest BPF return
289 * value always takes priority (ignoring the DATA).
291 for (; f; f = f->prev) {
292 u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd);
294 if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
301 #endif /* CONFIG_SECCOMP_FILTER */
303 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
305 assert_spin_locked(¤t->sighand->siglock);
307 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
313 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
315 static inline void seccomp_assign_mode(struct task_struct *task,
316 unsigned long seccomp_mode,
319 assert_spin_locked(&task->sighand->siglock);
321 task->seccomp.mode = seccomp_mode;
323 * Make sure TIF_SECCOMP cannot be set before the mode (and
326 smp_mb__before_atomic();
327 /* Assume default seccomp processes want spec flaw mitigation. */
328 if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
329 arch_seccomp_spec_mitigate(task);
330 set_tsk_thread_flag(task, TIF_SECCOMP);
333 #ifdef CONFIG_SECCOMP_FILTER
334 /* Returns 1 if the parent is an ancestor of the child. */
335 static int is_ancestor(struct seccomp_filter *parent,
336 struct seccomp_filter *child)
338 /* NULL is the root ancestor. */
341 for (; child; child = child->prev)
348 * seccomp_can_sync_threads: checks if all threads can be synchronized
350 * Expects sighand and cred_guard_mutex locks to be held.
352 * Returns 0 on success, -ve on error, or the pid of a thread which was
353 * either not in the correct seccomp mode or did not have an ancestral
356 static inline pid_t seccomp_can_sync_threads(void)
358 struct task_struct *thread, *caller;
360 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
361 assert_spin_locked(¤t->sighand->siglock);
363 /* Validate all threads being eligible for synchronization. */
365 for_each_thread(caller, thread) {
368 /* Skip current, since it is initiating the sync. */
369 if (thread == caller)
372 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
373 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
374 is_ancestor(thread->seccomp.filter,
375 caller->seccomp.filter)))
378 /* Return the first thread that cannot be synchronized. */
379 failed = task_pid_vnr(thread);
380 /* If the pid cannot be resolved, then return -ESRCH */
381 if (WARN_ON(failed == 0))
389 static inline void seccomp_filter_free(struct seccomp_filter *filter)
392 bpf_prog_destroy(filter->prog);
397 static void __seccomp_filter_orphan(struct seccomp_filter *orig)
399 while (orig && refcount_dec_and_test(&orig->users)) {
400 if (waitqueue_active(&orig->wqh))
401 wake_up_poll(&orig->wqh, EPOLLHUP);
406 static void __put_seccomp_filter(struct seccomp_filter *orig)
408 /* Clean up single-reference branches iteratively. */
409 while (orig && refcount_dec_and_test(&orig->refs)) {
410 struct seccomp_filter *freeme = orig;
412 seccomp_filter_free(freeme);
416 static void __seccomp_filter_release(struct seccomp_filter *orig)
418 /* Notify about any unused filters in the task's former filter tree. */
419 __seccomp_filter_orphan(orig);
420 /* Finally drop all references to the task's former tree. */
421 __put_seccomp_filter(orig);
425 * seccomp_filter_release - Detach the task from its filter tree,
426 * drop its reference count, and notify
427 * about unused filters
429 * This function should only be called when the task is exiting as
430 * it detaches it from its filter tree. As such, READ_ONCE() and
431 * barriers are not needed here, as would normally be needed.
433 void seccomp_filter_release(struct task_struct *tsk)
435 struct seccomp_filter *orig = tsk->seccomp.filter;
437 /* Detach task from its filter tree. */
438 tsk->seccomp.filter = NULL;
439 __seccomp_filter_release(orig);
443 * seccomp_sync_threads: sets all threads to use current's filter
445 * Expects sighand and cred_guard_mutex locks to be held, and for
446 * seccomp_can_sync_threads() to have returned success already
447 * without dropping the locks.
450 static inline void seccomp_sync_threads(unsigned long flags)
452 struct task_struct *thread, *caller;
454 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
455 assert_spin_locked(¤t->sighand->siglock);
457 /* Synchronize all threads. */
459 for_each_thread(caller, thread) {
460 /* Skip current, since it needs no changes. */
461 if (thread == caller)
464 /* Get a task reference for the new leaf node. */
465 get_seccomp_filter(caller);
468 * Drop the task reference to the shared ancestor since
469 * current's path will hold a reference. (This also
470 * allows a put before the assignment.)
472 __seccomp_filter_release(thread->seccomp.filter);
474 /* Make our new filter tree visible. */
475 smp_store_release(&thread->seccomp.filter,
476 caller->seccomp.filter);
477 atomic_set(&thread->seccomp.filter_count,
478 atomic_read(&thread->seccomp.filter_count));
481 * Don't let an unprivileged task work around
482 * the no_new_privs restriction by creating
483 * a thread that sets it up, enters seccomp,
486 if (task_no_new_privs(caller))
487 task_set_no_new_privs(thread);
490 * Opt the other thread into seccomp if needed.
491 * As threads are considered to be trust-realm
492 * equivalent (see ptrace_may_access), it is safe to
493 * allow one thread to transition the other.
495 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
496 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
502 * seccomp_prepare_filter: Prepares a seccomp filter for use.
503 * @fprog: BPF program to install
505 * Returns filter on success or an ERR_PTR on failure.
507 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
509 struct seccomp_filter *sfilter;
511 const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
513 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
514 return ERR_PTR(-EINVAL);
516 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
519 * Installing a seccomp filter requires that the task has
520 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
521 * This avoids scenarios where unprivileged tasks can affect the
522 * behavior of privileged children.
524 if (!task_no_new_privs(current) &&
525 security_capable(current_cred(), current_user_ns(),
526 CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) != 0)
527 return ERR_PTR(-EACCES);
529 /* Allocate a new seccomp_filter */
530 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
532 return ERR_PTR(-ENOMEM);
534 mutex_init(&sfilter->notify_lock);
535 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
536 seccomp_check_filter, save_orig);
542 refcount_set(&sfilter->refs, 1);
543 refcount_set(&sfilter->users, 1);
544 init_waitqueue_head(&sfilter->wqh);
550 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
551 * @user_filter: pointer to the user data containing a sock_fprog.
553 * Returns 0 on success and non-zero otherwise.
555 static struct seccomp_filter *
556 seccomp_prepare_user_filter(const char __user *user_filter)
558 struct sock_fprog fprog;
559 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
562 if (in_compat_syscall()) {
563 struct compat_sock_fprog fprog32;
564 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
566 fprog.len = fprog32.len;
567 fprog.filter = compat_ptr(fprog32.filter);
568 } else /* falls through to the if below. */
570 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
572 filter = seccomp_prepare_filter(&fprog);
578 * seccomp_attach_filter: validate and attach filter
579 * @flags: flags to change filter behavior
580 * @filter: seccomp filter to add to the current process
582 * Caller must be holding current->sighand->siglock lock.
584 * Returns 0 on success, -ve on error, or
585 * - in TSYNC mode: the pid of a thread which was either not in the correct
586 * seccomp mode or did not have an ancestral seccomp filter
587 * - in NEW_LISTENER mode: the fd of the new listener
589 static long seccomp_attach_filter(unsigned int flags,
590 struct seccomp_filter *filter)
592 unsigned long total_insns;
593 struct seccomp_filter *walker;
595 assert_spin_locked(¤t->sighand->siglock);
597 /* Validate resulting filter length. */
598 total_insns = filter->prog->len;
599 for (walker = current->seccomp.filter; walker; walker = walker->prev)
600 total_insns += walker->prog->len + 4; /* 4 instr penalty */
601 if (total_insns > MAX_INSNS_PER_PATH)
604 /* If thread sync has been requested, check that it is possible. */
605 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
608 ret = seccomp_can_sync_threads();
610 if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
617 /* Set log flag, if present. */
618 if (flags & SECCOMP_FILTER_FLAG_LOG)
622 * If there is an existing filter, make it the prev and don't drop its
625 filter->prev = current->seccomp.filter;
626 current->seccomp.filter = filter;
627 atomic_inc(¤t->seccomp.filter_count);
629 /* Now that the new filter is in place, synchronize to all threads. */
630 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
631 seccomp_sync_threads(flags);
636 static void __get_seccomp_filter(struct seccomp_filter *filter)
638 refcount_inc(&filter->refs);
641 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
642 void get_seccomp_filter(struct task_struct *tsk)
644 struct seccomp_filter *orig = tsk->seccomp.filter;
647 __get_seccomp_filter(orig);
648 refcount_inc(&orig->users);
651 static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason)
654 info->si_signo = SIGSYS;
655 info->si_code = SYS_SECCOMP;
656 info->si_call_addr = (void __user *)KSTK_EIP(current);
657 info->si_errno = reason;
658 info->si_arch = syscall_get_arch(current);
659 info->si_syscall = syscall;
663 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
664 * @syscall: syscall number to send to userland
665 * @reason: filter-supplied reason code to send to userland (via si_errno)
667 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
669 static void seccomp_send_sigsys(int syscall, int reason)
671 struct kernel_siginfo info;
672 seccomp_init_siginfo(&info, syscall, reason);
673 force_sig_info(&info);
675 #endif /* CONFIG_SECCOMP_FILTER */
677 /* For use with seccomp_actions_logged */
678 #define SECCOMP_LOG_KILL_PROCESS (1 << 0)
679 #define SECCOMP_LOG_KILL_THREAD (1 << 1)
680 #define SECCOMP_LOG_TRAP (1 << 2)
681 #define SECCOMP_LOG_ERRNO (1 << 3)
682 #define SECCOMP_LOG_TRACE (1 << 4)
683 #define SECCOMP_LOG_LOG (1 << 5)
684 #define SECCOMP_LOG_ALLOW (1 << 6)
685 #define SECCOMP_LOG_USER_NOTIF (1 << 7)
687 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
688 SECCOMP_LOG_KILL_THREAD |
691 SECCOMP_LOG_USER_NOTIF |
695 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
701 case SECCOMP_RET_ALLOW:
703 case SECCOMP_RET_TRAP:
704 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
706 case SECCOMP_RET_ERRNO:
707 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
709 case SECCOMP_RET_TRACE:
710 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
712 case SECCOMP_RET_USER_NOTIF:
713 log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
715 case SECCOMP_RET_LOG:
716 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
718 case SECCOMP_RET_KILL_THREAD:
719 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
721 case SECCOMP_RET_KILL_PROCESS:
723 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
727 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
728 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
729 * any action from being logged by removing the action name from the
730 * seccomp_actions_logged sysctl.
735 audit_seccomp(syscall, signr, action);
739 * Secure computing mode 1 allows only read/write/exit/sigreturn.
740 * To be fully secure this must be combined with rlimit
741 * to limit the stack allocations too.
743 static const int mode1_syscalls[] = {
744 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
745 0, /* null terminated */
748 static void __secure_computing_strict(int this_syscall)
750 const int *syscall_whitelist = mode1_syscalls;
752 if (in_compat_syscall())
753 syscall_whitelist = get_compat_mode1_syscalls();
756 if (*syscall_whitelist == this_syscall)
758 } while (*++syscall_whitelist);
763 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
767 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
768 void secure_computing_strict(int this_syscall)
770 int mode = current->seccomp.mode;
772 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
773 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
776 if (mode == SECCOMP_MODE_DISABLED)
778 else if (mode == SECCOMP_MODE_STRICT)
779 __secure_computing_strict(this_syscall);
785 #ifdef CONFIG_SECCOMP_FILTER
786 static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
789 * Note: overflow is ok here, the id just needs to be unique per
792 lockdep_assert_held(&filter->notify_lock);
793 return filter->notif->next_id++;
796 static int seccomp_do_user_notification(int this_syscall,
797 struct seccomp_filter *match,
798 const struct seccomp_data *sd)
803 struct seccomp_knotif n = {};
805 mutex_lock(&match->notify_lock);
811 n.state = SECCOMP_NOTIFY_INIT;
813 n.id = seccomp_next_notify_id(match);
814 init_completion(&n.ready);
815 list_add(&n.list, &match->notif->notifications);
817 up(&match->notif->request);
818 wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM);
819 mutex_unlock(&match->notify_lock);
822 * This is where we wait for a reply from userspace.
824 err = wait_for_completion_interruptible(&n.ready);
825 mutex_lock(&match->notify_lock);
833 * Note that it's possible the listener died in between the time when
834 * we were notified of a respons (or a signal) and when we were able to
835 * re-acquire the lock, so only delete from the list if the
836 * notification actually exists.
838 * Also note that this test is only valid because there's no way to
839 * *reattach* to a notifier right now. If one is added, we'll need to
840 * keep track of the notif itself and make sure they match here.
845 mutex_unlock(&match->notify_lock);
847 /* Userspace requests to continue the syscall. */
848 if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE)
851 syscall_set_return_value(current, task_pt_regs(current),
856 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
857 const bool recheck_after_trace)
859 u32 filter_ret, action;
860 struct seccomp_filter *match = NULL;
862 struct seccomp_data sd_local;
865 * Make sure that any changes to mode from another thread have
866 * been seen after TIF_SECCOMP was seen.
871 populate_seccomp_data(&sd_local);
875 filter_ret = seccomp_run_filters(sd, &match);
876 data = filter_ret & SECCOMP_RET_DATA;
877 action = filter_ret & SECCOMP_RET_ACTION_FULL;
880 case SECCOMP_RET_ERRNO:
881 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
882 if (data > MAX_ERRNO)
884 syscall_set_return_value(current, task_pt_regs(current),
888 case SECCOMP_RET_TRAP:
889 /* Show the handler the original registers. */
890 syscall_rollback(current, task_pt_regs(current));
891 /* Let the filter pass back 16 bits of data. */
892 seccomp_send_sigsys(this_syscall, data);
895 case SECCOMP_RET_TRACE:
896 /* We've been put in this state by the ptracer already. */
897 if (recheck_after_trace)
900 /* ENOSYS these calls if there is no tracer attached. */
901 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
902 syscall_set_return_value(current,
903 task_pt_regs(current),
908 /* Allow the BPF to provide the event message */
909 ptrace_event(PTRACE_EVENT_SECCOMP, data);
911 * The delivery of a fatal signal during event
912 * notification may silently skip tracer notification,
913 * which could leave us with a potentially unmodified
914 * syscall that the tracer would have liked to have
915 * changed. Since the process is about to die, we just
916 * force the syscall to be skipped and let the signal
917 * kill the process and correctly handle any tracer exit
920 if (fatal_signal_pending(current))
922 /* Check if the tracer forced the syscall to be skipped. */
923 this_syscall = syscall_get_nr(current, task_pt_regs(current));
924 if (this_syscall < 0)
928 * Recheck the syscall, since it may have changed. This
929 * intentionally uses a NULL struct seccomp_data to force
930 * a reload of all registers. This does not goto skip since
931 * a skip would have already been reported.
933 if (__seccomp_filter(this_syscall, NULL, true))
938 case SECCOMP_RET_USER_NOTIF:
939 if (seccomp_do_user_notification(this_syscall, match, sd))
944 case SECCOMP_RET_LOG:
945 seccomp_log(this_syscall, 0, action, true);
948 case SECCOMP_RET_ALLOW:
950 * Note that the "match" filter will always be NULL for
951 * this action since SECCOMP_RET_ALLOW is the starting
952 * state in seccomp_run_filters().
956 case SECCOMP_RET_KILL_THREAD:
957 case SECCOMP_RET_KILL_PROCESS:
959 seccomp_log(this_syscall, SIGSYS, action, true);
960 /* Dump core only if this is the last remaining thread. */
961 if (action == SECCOMP_RET_KILL_PROCESS ||
962 get_nr_threads(current) == 1) {
963 kernel_siginfo_t info;
965 /* Show the original registers in the dump. */
966 syscall_rollback(current, task_pt_regs(current));
967 /* Trigger a manual coredump since do_exit skips it. */
968 seccomp_init_siginfo(&info, this_syscall, data);
971 if (action == SECCOMP_RET_KILL_PROCESS)
972 do_group_exit(SIGSYS);
980 seccomp_log(this_syscall, 0, action, match ? match->log : false);
984 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
985 const bool recheck_after_trace)
991 int __secure_computing(const struct seccomp_data *sd)
993 int mode = current->seccomp.mode;
996 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
997 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1000 this_syscall = sd ? sd->nr :
1001 syscall_get_nr(current, task_pt_regs(current));
1004 case SECCOMP_MODE_STRICT:
1005 __secure_computing_strict(this_syscall); /* may call do_exit */
1007 case SECCOMP_MODE_FILTER:
1008 return __seccomp_filter(this_syscall, sd, false);
1013 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1015 long prctl_get_seccomp(void)
1017 return current->seccomp.mode;
1021 * seccomp_set_mode_strict: internal function for setting strict seccomp
1023 * Once current->seccomp.mode is non-zero, it may not be changed.
1025 * Returns 0 on success or -EINVAL on failure.
1027 static long seccomp_set_mode_strict(void)
1029 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
1032 spin_lock_irq(¤t->sighand->siglock);
1034 if (!seccomp_may_assign_mode(seccomp_mode))
1040 seccomp_assign_mode(current, seccomp_mode, 0);
1044 spin_unlock_irq(¤t->sighand->siglock);
1049 #ifdef CONFIG_SECCOMP_FILTER
1050 static int seccomp_notify_release(struct inode *inode, struct file *file)
1052 struct seccomp_filter *filter = file->private_data;
1053 struct seccomp_knotif *knotif;
1058 mutex_lock(&filter->notify_lock);
1061 * If this file is being closed because e.g. the task who owned it
1062 * died, let's wake everyone up who was waiting on us.
1064 list_for_each_entry(knotif, &filter->notif->notifications, list) {
1065 if (knotif->state == SECCOMP_NOTIFY_REPLIED)
1068 knotif->state = SECCOMP_NOTIFY_REPLIED;
1069 knotif->error = -ENOSYS;
1072 complete(&knotif->ready);
1075 kfree(filter->notif);
1076 filter->notif = NULL;
1077 mutex_unlock(&filter->notify_lock);
1078 __put_seccomp_filter(filter);
1082 /* must be called with notif_lock held */
1083 static inline struct seccomp_knotif *
1084 find_notification(struct seccomp_filter *filter, u64 id)
1086 struct seccomp_knotif *cur;
1088 lockdep_assert_held(&filter->notify_lock);
1090 list_for_each_entry(cur, &filter->notif->notifications, list) {
1099 static long seccomp_notify_recv(struct seccomp_filter *filter,
1102 struct seccomp_knotif *knotif = NULL, *cur;
1103 struct seccomp_notif unotif;
1106 /* Verify that we're not given garbage to keep struct extensible. */
1107 ret = check_zeroed_user(buf, sizeof(unotif));
1113 memset(&unotif, 0, sizeof(unotif));
1115 ret = down_interruptible(&filter->notif->request);
1119 mutex_lock(&filter->notify_lock);
1120 list_for_each_entry(cur, &filter->notif->notifications, list) {
1121 if (cur->state == SECCOMP_NOTIFY_INIT) {
1128 * If we didn't find a notification, it could be that the task was
1129 * interrupted by a fatal signal between the time we were woken and
1130 * when we were able to acquire the rw lock.
1137 unotif.id = knotif->id;
1138 unotif.pid = task_pid_vnr(knotif->task);
1139 unotif.data = *(knotif->data);
1141 knotif->state = SECCOMP_NOTIFY_SENT;
1142 wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM);
1145 mutex_unlock(&filter->notify_lock);
1147 if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1151 * Userspace screwed up. To make sure that we keep this
1152 * notification alive, let's reset it back to INIT. It
1153 * may have died when we released the lock, so we need to make
1154 * sure it's still around.
1156 mutex_lock(&filter->notify_lock);
1157 knotif = find_notification(filter, unotif.id);
1159 knotif->state = SECCOMP_NOTIFY_INIT;
1160 up(&filter->notif->request);
1162 mutex_unlock(&filter->notify_lock);
1168 static long seccomp_notify_send(struct seccomp_filter *filter,
1171 struct seccomp_notif_resp resp = {};
1172 struct seccomp_knotif *knotif;
1175 if (copy_from_user(&resp, buf, sizeof(resp)))
1178 if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1181 if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) &&
1182 (resp.error || resp.val))
1185 ret = mutex_lock_interruptible(&filter->notify_lock);
1189 knotif = find_notification(filter, resp.id);
1195 /* Allow exactly one reply. */
1196 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1202 knotif->state = SECCOMP_NOTIFY_REPLIED;
1203 knotif->error = resp.error;
1204 knotif->val = resp.val;
1205 knotif->flags = resp.flags;
1206 complete(&knotif->ready);
1208 mutex_unlock(&filter->notify_lock);
1212 static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1215 struct seccomp_knotif *knotif;
1219 if (copy_from_user(&id, buf, sizeof(id)))
1222 ret = mutex_lock_interruptible(&filter->notify_lock);
1226 knotif = find_notification(filter, id);
1227 if (knotif && knotif->state == SECCOMP_NOTIFY_SENT)
1232 mutex_unlock(&filter->notify_lock);
1236 static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1239 struct seccomp_filter *filter = file->private_data;
1240 void __user *buf = (void __user *)arg;
1243 case SECCOMP_IOCTL_NOTIF_RECV:
1244 return seccomp_notify_recv(filter, buf);
1245 case SECCOMP_IOCTL_NOTIF_SEND:
1246 return seccomp_notify_send(filter, buf);
1247 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR:
1248 case SECCOMP_IOCTL_NOTIF_ID_VALID:
1249 return seccomp_notify_id_valid(filter, buf);
1255 static __poll_t seccomp_notify_poll(struct file *file,
1256 struct poll_table_struct *poll_tab)
1258 struct seccomp_filter *filter = file->private_data;
1260 struct seccomp_knotif *cur;
1262 poll_wait(file, &filter->wqh, poll_tab);
1264 if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1267 list_for_each_entry(cur, &filter->notif->notifications, list) {
1268 if (cur->state == SECCOMP_NOTIFY_INIT)
1269 ret |= EPOLLIN | EPOLLRDNORM;
1270 if (cur->state == SECCOMP_NOTIFY_SENT)
1271 ret |= EPOLLOUT | EPOLLWRNORM;
1272 if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1276 mutex_unlock(&filter->notify_lock);
1278 if (refcount_read(&filter->users) == 0)
1284 static const struct file_operations seccomp_notify_ops = {
1285 .poll = seccomp_notify_poll,
1286 .release = seccomp_notify_release,
1287 .unlocked_ioctl = seccomp_notify_ioctl,
1288 .compat_ioctl = seccomp_notify_ioctl,
1291 static struct file *init_listener(struct seccomp_filter *filter)
1293 struct file *ret = ERR_PTR(-EBUSY);
1294 struct seccomp_filter *cur;
1296 for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1301 ret = ERR_PTR(-ENOMEM);
1302 filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1306 sema_init(&filter->notif->request, 0);
1307 filter->notif->next_id = get_random_u64();
1308 INIT_LIST_HEAD(&filter->notif->notifications);
1310 ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1315 /* The file has a reference to it now */
1316 __get_seccomp_filter(filter);
1320 kfree(filter->notif);
1326 * seccomp_set_mode_filter: internal function for setting seccomp filter
1327 * @flags: flags to change filter behavior
1328 * @filter: struct sock_fprog containing filter
1330 * This function may be called repeatedly to install additional filters.
1331 * Every filter successfully installed will be evaluated (in reverse order)
1332 * for each system call the task makes.
1334 * Once current->seccomp.mode is non-zero, it may not be changed.
1336 * Returns 0 on success or -EINVAL on failure.
1338 static long seccomp_set_mode_filter(unsigned int flags,
1339 const char __user *filter)
1341 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1342 struct seccomp_filter *prepared = NULL;
1345 struct file *listener_f = NULL;
1347 /* Validate flags. */
1348 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1352 * In the successful case, NEW_LISTENER returns the new listener fd.
1353 * But in the failure case, TSYNC returns the thread that died. If you
1354 * combine these two flags, there's no way to tell whether something
1355 * succeeded or failed. So, let's disallow this combination if the user
1356 * has not explicitly requested no errors from TSYNC.
1358 if ((flags & SECCOMP_FILTER_FLAG_TSYNC) &&
1359 (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) &&
1360 ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0))
1363 /* Prepare the new filter before holding any locks. */
1364 prepared = seccomp_prepare_user_filter(filter);
1365 if (IS_ERR(prepared))
1366 return PTR_ERR(prepared);
1368 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1369 listener = get_unused_fd_flags(O_CLOEXEC);
1375 listener_f = init_listener(prepared);
1376 if (IS_ERR(listener_f)) {
1377 put_unused_fd(listener);
1378 ret = PTR_ERR(listener_f);
1384 * Make sure we cannot change seccomp or nnp state via TSYNC
1385 * while another thread is in the middle of calling exec.
1387 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1388 mutex_lock_killable(¤t->signal->cred_guard_mutex))
1391 spin_lock_irq(¤t->sighand->siglock);
1393 if (!seccomp_may_assign_mode(seccomp_mode))
1396 ret = seccomp_attach_filter(flags, prepared);
1399 /* Do not free the successfully attached filter. */
1402 seccomp_assign_mode(current, seccomp_mode, flags);
1404 spin_unlock_irq(¤t->sighand->siglock);
1405 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
1406 mutex_unlock(¤t->signal->cred_guard_mutex);
1408 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1410 listener_f->private_data = NULL;
1412 put_unused_fd(listener);
1414 fd_install(listener, listener_f);
1419 seccomp_filter_free(prepared);
1423 static inline long seccomp_set_mode_filter(unsigned int flags,
1424 const char __user *filter)
1430 static long seccomp_get_action_avail(const char __user *uaction)
1434 if (copy_from_user(&action, uaction, sizeof(action)))
1438 case SECCOMP_RET_KILL_PROCESS:
1439 case SECCOMP_RET_KILL_THREAD:
1440 case SECCOMP_RET_TRAP:
1441 case SECCOMP_RET_ERRNO:
1442 case SECCOMP_RET_USER_NOTIF:
1443 case SECCOMP_RET_TRACE:
1444 case SECCOMP_RET_LOG:
1445 case SECCOMP_RET_ALLOW:
1454 static long seccomp_get_notif_sizes(void __user *usizes)
1456 struct seccomp_notif_sizes sizes = {
1457 .seccomp_notif = sizeof(struct seccomp_notif),
1458 .seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
1459 .seccomp_data = sizeof(struct seccomp_data),
1462 if (copy_to_user(usizes, &sizes, sizeof(sizes)))
1468 /* Common entry point for both prctl and syscall. */
1469 static long do_seccomp(unsigned int op, unsigned int flags,
1473 case SECCOMP_SET_MODE_STRICT:
1474 if (flags != 0 || uargs != NULL)
1476 return seccomp_set_mode_strict();
1477 case SECCOMP_SET_MODE_FILTER:
1478 return seccomp_set_mode_filter(flags, uargs);
1479 case SECCOMP_GET_ACTION_AVAIL:
1483 return seccomp_get_action_avail(uargs);
1484 case SECCOMP_GET_NOTIF_SIZES:
1488 return seccomp_get_notif_sizes(uargs);
1494 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
1495 void __user *, uargs)
1497 return do_seccomp(op, flags, uargs);
1501 * prctl_set_seccomp: configures current->seccomp.mode
1502 * @seccomp_mode: requested mode to use
1503 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1505 * Returns 0 on success or -EINVAL on failure.
1507 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
1512 switch (seccomp_mode) {
1513 case SECCOMP_MODE_STRICT:
1514 op = SECCOMP_SET_MODE_STRICT;
1516 * Setting strict mode through prctl always ignored filter,
1517 * so make sure it is always NULL here to pass the internal
1518 * check in do_seccomp().
1522 case SECCOMP_MODE_FILTER:
1523 op = SECCOMP_SET_MODE_FILTER;
1530 /* prctl interface doesn't have flags, so they are always zero. */
1531 return do_seccomp(op, 0, uargs);
1534 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
1535 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
1536 unsigned long filter_off)
1538 struct seccomp_filter *orig, *filter;
1539 unsigned long count;
1542 * Note: this is only correct because the caller should be the (ptrace)
1543 * tracer of the task, otherwise lock_task_sighand is needed.
1545 spin_lock_irq(&task->sighand->siglock);
1547 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
1548 spin_unlock_irq(&task->sighand->siglock);
1549 return ERR_PTR(-EINVAL);
1552 orig = task->seccomp.filter;
1553 __get_seccomp_filter(orig);
1554 spin_unlock_irq(&task->sighand->siglock);
1557 for (filter = orig; filter; filter = filter->prev)
1560 if (filter_off >= count) {
1561 filter = ERR_PTR(-ENOENT);
1565 count -= filter_off;
1566 for (filter = orig; filter && count > 1; filter = filter->prev)
1569 if (WARN_ON(count != 1 || !filter)) {
1570 filter = ERR_PTR(-ENOENT);
1574 __get_seccomp_filter(filter);
1577 __put_seccomp_filter(orig);
1581 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1584 struct seccomp_filter *filter;
1585 struct sock_fprog_kern *fprog;
1588 if (!capable(CAP_SYS_ADMIN) ||
1589 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1593 filter = get_nth_filter(task, filter_off);
1595 return PTR_ERR(filter);
1597 fprog = filter->prog->orig_prog;
1599 /* This must be a new non-cBPF filter, since we save
1600 * every cBPF filter's orig_prog above when
1601 * CONFIG_CHECKPOINT_RESTORE is enabled.
1611 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1615 __put_seccomp_filter(filter);
1619 long seccomp_get_metadata(struct task_struct *task,
1620 unsigned long size, void __user *data)
1623 struct seccomp_filter *filter;
1624 struct seccomp_metadata kmd = {};
1626 if (!capable(CAP_SYS_ADMIN) ||
1627 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1631 size = min_t(unsigned long, size, sizeof(kmd));
1633 if (size < sizeof(kmd.filter_off))
1636 if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
1639 filter = get_nth_filter(task, kmd.filter_off);
1641 return PTR_ERR(filter);
1644 kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1647 if (copy_to_user(data, &kmd, size))
1650 __put_seccomp_filter(filter);
1655 #ifdef CONFIG_SYSCTL
1657 /* Human readable action names for friendly sysctl interaction */
1658 #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
1659 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
1660 #define SECCOMP_RET_TRAP_NAME "trap"
1661 #define SECCOMP_RET_ERRNO_NAME "errno"
1662 #define SECCOMP_RET_USER_NOTIF_NAME "user_notif"
1663 #define SECCOMP_RET_TRACE_NAME "trace"
1664 #define SECCOMP_RET_LOG_NAME "log"
1665 #define SECCOMP_RET_ALLOW_NAME "allow"
1667 static const char seccomp_actions_avail[] =
1668 SECCOMP_RET_KILL_PROCESS_NAME " "
1669 SECCOMP_RET_KILL_THREAD_NAME " "
1670 SECCOMP_RET_TRAP_NAME " "
1671 SECCOMP_RET_ERRNO_NAME " "
1672 SECCOMP_RET_USER_NOTIF_NAME " "
1673 SECCOMP_RET_TRACE_NAME " "
1674 SECCOMP_RET_LOG_NAME " "
1675 SECCOMP_RET_ALLOW_NAME;
1677 struct seccomp_log_name {
1682 static const struct seccomp_log_name seccomp_log_names[] = {
1683 { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1684 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1685 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1686 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1687 { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
1688 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1689 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1690 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1694 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1698 const struct seccomp_log_name *cur;
1699 bool append_sep = false;
1701 for (cur = seccomp_log_names; cur->name && size; cur++) {
1704 if (!(actions_logged & cur->log))
1708 ret = strscpy(names, sep, size);
1717 ret = strscpy(names, cur->name, size);
1728 static bool seccomp_action_logged_from_name(u32 *action_logged,
1731 const struct seccomp_log_name *cur;
1733 for (cur = seccomp_log_names; cur->name; cur++) {
1734 if (!strcmp(cur->name, name)) {
1735 *action_logged = cur->log;
1743 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1747 *actions_logged = 0;
1748 while ((name = strsep(&names, " ")) && *name) {
1749 u32 action_logged = 0;
1751 if (!seccomp_action_logged_from_name(&action_logged, name))
1754 *actions_logged |= action_logged;
1760 static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1761 size_t *lenp, loff_t *ppos)
1763 char names[sizeof(seccomp_actions_avail)];
1764 struct ctl_table table;
1766 memset(names, 0, sizeof(names));
1768 if (!seccomp_names_from_actions_logged(names, sizeof(names),
1769 seccomp_actions_logged, " "))
1774 table.maxlen = sizeof(names);
1775 return proc_dostring(&table, 0, buffer, lenp, ppos);
1778 static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1779 size_t *lenp, loff_t *ppos, u32 *actions_logged)
1781 char names[sizeof(seccomp_actions_avail)];
1782 struct ctl_table table;
1785 if (!capable(CAP_SYS_ADMIN))
1788 memset(names, 0, sizeof(names));
1792 table.maxlen = sizeof(names);
1793 ret = proc_dostring(&table, 1, buffer, lenp, ppos);
1797 if (!seccomp_actions_logged_from_names(actions_logged, table.data))
1800 if (*actions_logged & SECCOMP_LOG_ALLOW)
1803 seccomp_actions_logged = *actions_logged;
1807 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
1810 char names[sizeof(seccomp_actions_avail)];
1811 char old_names[sizeof(seccomp_actions_avail)];
1812 const char *new = names;
1813 const char *old = old_names;
1818 memset(names, 0, sizeof(names));
1819 memset(old_names, 0, sizeof(old_names));
1823 else if (!actions_logged)
1825 else if (!seccomp_names_from_actions_logged(names, sizeof(names),
1826 actions_logged, ","))
1829 if (!old_actions_logged)
1831 else if (!seccomp_names_from_actions_logged(old_names,
1833 old_actions_logged, ","))
1836 return audit_seccomp_actions_logged(new, old, !ret);
1839 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1840 void *buffer, size_t *lenp,
1846 u32 actions_logged = 0;
1847 u32 old_actions_logged = seccomp_actions_logged;
1849 ret = write_actions_logged(ro_table, buffer, lenp, ppos,
1851 audit_actions_logged(actions_logged, old_actions_logged, ret);
1853 ret = read_actions_logged(ro_table, buffer, lenp, ppos);
1858 static struct ctl_path seccomp_sysctl_path[] = {
1859 { .procname = "kernel", },
1860 { .procname = "seccomp", },
1864 static struct ctl_table seccomp_sysctl_table[] = {
1866 .procname = "actions_avail",
1867 .data = (void *) &seccomp_actions_avail,
1868 .maxlen = sizeof(seccomp_actions_avail),
1870 .proc_handler = proc_dostring,
1873 .procname = "actions_logged",
1875 .proc_handler = seccomp_actions_logged_handler,
1880 static int __init seccomp_sysctl_init(void)
1882 struct ctl_table_header *hdr;
1884 hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1886 pr_warn("sysctl registration failed\n");
1888 kmemleak_not_leak(hdr);
1893 device_initcall(seccomp_sysctl_init)
1895 #endif /* CONFIG_SYSCTL */