1 // SPDX-License-Identifier: GPL-2.0
3 * linux/kernel/seccomp.c
5 * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com>
7 * Copyright (C) 2012 Google, Inc.
8 * Will Drewry <wad@chromium.org>
10 * This defines a simple but solid secure-computing facility.
12 * Mode 1 uses a fixed list of allowed system calls.
13 * Mode 2 allows user-defined system call filters in the form
14 * of Berkeley Packet Filters/Linux Socket Filters.
17 #include <linux/refcount.h>
18 #include <linux/audit.h>
19 #include <linux/compat.h>
20 #include <linux/coredump.h>
21 #include <linux/kmemleak.h>
22 #include <linux/nospec.h>
23 #include <linux/prctl.h>
24 #include <linux/sched.h>
25 #include <linux/sched/task_stack.h>
26 #include <linux/seccomp.h>
27 #include <linux/slab.h>
28 #include <linux/syscalls.h>
29 #include <linux/sysctl.h>
31 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
32 #include <asm/syscall.h>
35 #ifdef CONFIG_SECCOMP_FILTER
36 #include <linux/file.h>
37 #include <linux/filter.h>
38 #include <linux/pid.h>
39 #include <linux/ptrace.h>
40 #include <linux/capability.h>
41 #include <linux/tracehook.h>
42 #include <linux/uaccess.h>
43 #include <linux/anon_inodes.h>
46 * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the
47 * wrong direction flag in the ioctl number. This is the broken one,
48 * which the kernel needs to keep supporting until all userspaces stop
49 * using the wrong command number.
51 #define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR SECCOMP_IOR(2, __u64)
56 SECCOMP_NOTIFY_REPLIED
,
59 struct seccomp_knotif
{
60 /* The struct pid of the task whose filter triggered the notification */
61 struct task_struct
*task
;
63 /* The "cookie" for this request; this is unique for this filter. */
67 * The seccomp data. This pointer is valid the entire time this
68 * notification is active, since it comes from __seccomp_filter which
69 * eclipses the entire lifecycle here.
71 const struct seccomp_data
*data
;
74 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
75 * struct seccomp_knotif is created and starts out in INIT. Once the
76 * handler reads the notification off of an FD, it transitions to SENT.
77 * If a signal is received the state transitions back to INIT and
78 * another message is sent. When the userspace handler replies, state
79 * transitions to REPLIED.
81 enum notify_state state
;
83 /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
88 /* Signals when this has entered SECCOMP_NOTIFY_REPLIED */
89 struct completion ready
;
91 struct list_head list
;
95 * struct notification - container for seccomp userspace notifications. Since
96 * most seccomp filters will not have notification listeners attached and this
97 * structure is fairly large, we store the notification-specific stuff in a
100 * @request: A semaphore that users of this notification can wait on for
101 * changes. Actual reads and writes are still controlled with
102 * filter->notify_lock.
103 * @next_id: The id of the next request.
104 * @notifications: A list of struct seccomp_knotif elements.
105 * @wqh: A wait queue for poll.
107 struct notification
{
108 struct semaphore request
;
110 struct list_head notifications
;
111 wait_queue_head_t wqh
;
115 * struct seccomp_filter - container for seccomp BPF programs
117 * @usage: reference count to manage the object lifetime.
118 * get/put helpers should be used when accessing an instance
119 * outside of a lifetime-guarded section. In general, this
120 * is only needed for handling filters shared across tasks.
121 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
122 * @prev: points to a previously installed, or inherited, filter
123 * @prog: the BPF program to evaluate
124 * @notif: the struct that holds all notification related information
125 * @notify_lock: A lock for all notification-related accesses.
127 * seccomp_filter objects are organized in a tree linked via the @prev
128 * pointer. For any task, it appears to be a singly-linked list starting
129 * with current->seccomp.filter, the most recently attached or inherited filter.
130 * However, multiple filters may share a @prev node, by way of fork(), which
131 * results in a unidirectional tree existing in memory. This is similar to
132 * how namespaces work.
134 * seccomp_filter objects should never be modified after being attached
135 * to a task_struct (other than @usage).
137 struct seccomp_filter
{
140 struct seccomp_filter
*prev
;
141 struct bpf_prog
*prog
;
142 struct notification
*notif
;
143 struct mutex notify_lock
;
146 /* Limit any path through the tree to 256KB worth of instructions. */
147 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
150 * Endianness is explicitly ignored and left for BPF program authors to manage
151 * as per the specific architecture.
153 static void populate_seccomp_data(struct seccomp_data
*sd
)
155 struct task_struct
*task
= current
;
156 struct pt_regs
*regs
= task_pt_regs(task
);
157 unsigned long args
[6];
159 sd
->nr
= syscall_get_nr(task
, regs
);
160 sd
->arch
= syscall_get_arch(task
);
161 syscall_get_arguments(task
, regs
, args
);
162 sd
->args
[0] = args
[0];
163 sd
->args
[1] = args
[1];
164 sd
->args
[2] = args
[2];
165 sd
->args
[3] = args
[3];
166 sd
->args
[4] = args
[4];
167 sd
->args
[5] = args
[5];
168 sd
->instruction_pointer
= KSTK_EIP(task
);
172 * seccomp_check_filter - verify seccomp filter code
173 * @filter: filter to verify
174 * @flen: length of filter
176 * Takes a previously checked filter (by bpf_check_classic) and
177 * redirects all filter code that loads struct sk_buff data
178 * and related data through seccomp_bpf_load. It also
179 * enforces length and alignment checking of those loads.
181 * Returns 0 if the rule set is legal or -EINVAL if not.
183 static int seccomp_check_filter(struct sock_filter
*filter
, unsigned int flen
)
186 for (pc
= 0; pc
< flen
; pc
++) {
187 struct sock_filter
*ftest
= &filter
[pc
];
188 u16 code
= ftest
->code
;
192 case BPF_LD
| BPF_W
| BPF_ABS
:
193 ftest
->code
= BPF_LDX
| BPF_W
| BPF_ABS
;
194 /* 32-bit aligned and not out of bounds. */
195 if (k
>= sizeof(struct seccomp_data
) || k
& 3)
198 case BPF_LD
| BPF_W
| BPF_LEN
:
199 ftest
->code
= BPF_LD
| BPF_IMM
;
200 ftest
->k
= sizeof(struct seccomp_data
);
202 case BPF_LDX
| BPF_W
| BPF_LEN
:
203 ftest
->code
= BPF_LDX
| BPF_IMM
;
204 ftest
->k
= sizeof(struct seccomp_data
);
206 /* Explicitly include allowed calls. */
207 case BPF_RET
| BPF_K
:
208 case BPF_RET
| BPF_A
:
209 case BPF_ALU
| BPF_ADD
| BPF_K
:
210 case BPF_ALU
| BPF_ADD
| BPF_X
:
211 case BPF_ALU
| BPF_SUB
| BPF_K
:
212 case BPF_ALU
| BPF_SUB
| BPF_X
:
213 case BPF_ALU
| BPF_MUL
| BPF_K
:
214 case BPF_ALU
| BPF_MUL
| BPF_X
:
215 case BPF_ALU
| BPF_DIV
| BPF_K
:
216 case BPF_ALU
| BPF_DIV
| BPF_X
:
217 case BPF_ALU
| BPF_AND
| BPF_K
:
218 case BPF_ALU
| BPF_AND
| BPF_X
:
219 case BPF_ALU
| BPF_OR
| BPF_K
:
220 case BPF_ALU
| BPF_OR
| BPF_X
:
221 case BPF_ALU
| BPF_XOR
| BPF_K
:
222 case BPF_ALU
| BPF_XOR
| BPF_X
:
223 case BPF_ALU
| BPF_LSH
| BPF_K
:
224 case BPF_ALU
| BPF_LSH
| BPF_X
:
225 case BPF_ALU
| BPF_RSH
| BPF_K
:
226 case BPF_ALU
| BPF_RSH
| BPF_X
:
227 case BPF_ALU
| BPF_NEG
:
228 case BPF_LD
| BPF_IMM
:
229 case BPF_LDX
| BPF_IMM
:
230 case BPF_MISC
| BPF_TAX
:
231 case BPF_MISC
| BPF_TXA
:
232 case BPF_LD
| BPF_MEM
:
233 case BPF_LDX
| BPF_MEM
:
236 case BPF_JMP
| BPF_JA
:
237 case BPF_JMP
| BPF_JEQ
| BPF_K
:
238 case BPF_JMP
| BPF_JEQ
| BPF_X
:
239 case BPF_JMP
| BPF_JGE
| BPF_K
:
240 case BPF_JMP
| BPF_JGE
| BPF_X
:
241 case BPF_JMP
| BPF_JGT
| BPF_K
:
242 case BPF_JMP
| BPF_JGT
| BPF_X
:
243 case BPF_JMP
| BPF_JSET
| BPF_K
:
244 case BPF_JMP
| BPF_JSET
| BPF_X
:
254 * seccomp_run_filters - evaluates all seccomp filters against @sd
255 * @sd: optional seccomp data to be passed to filters
256 * @match: stores struct seccomp_filter that resulted in the return value,
257 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
260 * Returns valid seccomp BPF response codes.
262 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
263 static u32
seccomp_run_filters(const struct seccomp_data
*sd
,
264 struct seccomp_filter
**match
)
266 u32 ret
= SECCOMP_RET_ALLOW
;
267 /* Make sure cross-thread synced filter points somewhere sane. */
268 struct seccomp_filter
*f
=
269 READ_ONCE(current
->seccomp
.filter
);
271 /* Ensure unexpected behavior doesn't result in failing open. */
272 if (WARN_ON(f
== NULL
))
273 return SECCOMP_RET_KILL_PROCESS
;
276 * All filters in the list are evaluated and the lowest BPF return
277 * value always takes priority (ignoring the DATA).
280 for (; f
; f
= f
->prev
) {
281 u32 cur_ret
= BPF_PROG_RUN(f
->prog
, sd
);
283 if (ACTION_ONLY(cur_ret
) < ACTION_ONLY(ret
)) {
291 #endif /* CONFIG_SECCOMP_FILTER */
293 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode
)
295 assert_spin_locked(¤t
->sighand
->siglock
);
297 if (current
->seccomp
.mode
&& current
->seccomp
.mode
!= seccomp_mode
)
303 void __weak
arch_seccomp_spec_mitigate(struct task_struct
*task
) { }
305 static inline void seccomp_assign_mode(struct task_struct
*task
,
306 unsigned long seccomp_mode
,
309 assert_spin_locked(&task
->sighand
->siglock
);
311 task
->seccomp
.mode
= seccomp_mode
;
313 * Make sure TIF_SECCOMP cannot be set before the mode (and
316 smp_mb__before_atomic();
317 /* Assume default seccomp processes want spec flaw mitigation. */
318 if ((flags
& SECCOMP_FILTER_FLAG_SPEC_ALLOW
) == 0)
319 arch_seccomp_spec_mitigate(task
);
320 set_tsk_thread_flag(task
, TIF_SECCOMP
);
323 #ifdef CONFIG_SECCOMP_FILTER
324 /* Returns 1 if the parent is an ancestor of the child. */
325 static int is_ancestor(struct seccomp_filter
*parent
,
326 struct seccomp_filter
*child
)
328 /* NULL is the root ancestor. */
331 for (; child
; child
= child
->prev
)
338 * seccomp_can_sync_threads: checks if all threads can be synchronized
340 * Expects sighand and cred_guard_mutex locks to be held.
342 * Returns 0 on success, -ve on error, or the pid of a thread which was
343 * either not in the correct seccomp mode or did not have an ancestral
346 static inline pid_t
seccomp_can_sync_threads(void)
348 struct task_struct
*thread
, *caller
;
350 BUG_ON(!mutex_is_locked(¤t
->signal
->cred_guard_mutex
));
351 assert_spin_locked(¤t
->sighand
->siglock
);
353 /* Validate all threads being eligible for synchronization. */
355 for_each_thread(caller
, thread
) {
358 /* Skip current, since it is initiating the sync. */
359 if (thread
== caller
)
362 if (thread
->seccomp
.mode
== SECCOMP_MODE_DISABLED
||
363 (thread
->seccomp
.mode
== SECCOMP_MODE_FILTER
&&
364 is_ancestor(thread
->seccomp
.filter
,
365 caller
->seccomp
.filter
)))
368 /* Return the first thread that cannot be synchronized. */
369 failed
= task_pid_vnr(thread
);
370 /* If the pid cannot be resolved, then return -ESRCH */
371 if (WARN_ON(failed
== 0))
380 * seccomp_sync_threads: sets all threads to use current's filter
382 * Expects sighand and cred_guard_mutex locks to be held, and for
383 * seccomp_can_sync_threads() to have returned success already
384 * without dropping the locks.
387 static inline void seccomp_sync_threads(unsigned long flags
)
389 struct task_struct
*thread
, *caller
;
391 BUG_ON(!mutex_is_locked(¤t
->signal
->cred_guard_mutex
));
392 assert_spin_locked(¤t
->sighand
->siglock
);
394 /* Synchronize all threads. */
396 for_each_thread(caller
, thread
) {
397 /* Skip current, since it needs no changes. */
398 if (thread
== caller
)
401 /* Get a task reference for the new leaf node. */
402 get_seccomp_filter(caller
);
404 * Drop the task reference to the shared ancestor since
405 * current's path will hold a reference. (This also
406 * allows a put before the assignment.)
408 put_seccomp_filter(thread
);
409 smp_store_release(&thread
->seccomp
.filter
,
410 caller
->seccomp
.filter
);
413 * Don't let an unprivileged task work around
414 * the no_new_privs restriction by creating
415 * a thread that sets it up, enters seccomp,
418 if (task_no_new_privs(caller
))
419 task_set_no_new_privs(thread
);
422 * Opt the other thread into seccomp if needed.
423 * As threads are considered to be trust-realm
424 * equivalent (see ptrace_may_access), it is safe to
425 * allow one thread to transition the other.
427 if (thread
->seccomp
.mode
== SECCOMP_MODE_DISABLED
)
428 seccomp_assign_mode(thread
, SECCOMP_MODE_FILTER
,
434 * seccomp_prepare_filter: Prepares a seccomp filter for use.
435 * @fprog: BPF program to install
437 * Returns filter on success or an ERR_PTR on failure.
439 static struct seccomp_filter
*seccomp_prepare_filter(struct sock_fprog
*fprog
)
441 struct seccomp_filter
*sfilter
;
443 const bool save_orig
= IS_ENABLED(CONFIG_CHECKPOINT_RESTORE
);
445 if (fprog
->len
== 0 || fprog
->len
> BPF_MAXINSNS
)
446 return ERR_PTR(-EINVAL
);
448 BUG_ON(INT_MAX
/ fprog
->len
< sizeof(struct sock_filter
));
451 * Installing a seccomp filter requires that the task has
452 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
453 * This avoids scenarios where unprivileged tasks can affect the
454 * behavior of privileged children.
456 if (!task_no_new_privs(current
) &&
457 !ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN
))
458 return ERR_PTR(-EACCES
);
460 /* Allocate a new seccomp_filter */
461 sfilter
= kzalloc(sizeof(*sfilter
), GFP_KERNEL
| __GFP_NOWARN
);
463 return ERR_PTR(-ENOMEM
);
465 mutex_init(&sfilter
->notify_lock
);
466 ret
= bpf_prog_create_from_user(&sfilter
->prog
, fprog
,
467 seccomp_check_filter
, save_orig
);
473 refcount_set(&sfilter
->usage
, 1);
479 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
480 * @user_filter: pointer to the user data containing a sock_fprog.
482 * Returns 0 on success and non-zero otherwise.
484 static struct seccomp_filter
*
485 seccomp_prepare_user_filter(const char __user
*user_filter
)
487 struct sock_fprog fprog
;
488 struct seccomp_filter
*filter
= ERR_PTR(-EFAULT
);
491 if (in_compat_syscall()) {
492 struct compat_sock_fprog fprog32
;
493 if (copy_from_user(&fprog32
, user_filter
, sizeof(fprog32
)))
495 fprog
.len
= fprog32
.len
;
496 fprog
.filter
= compat_ptr(fprog32
.filter
);
497 } else /* falls through to the if below. */
499 if (copy_from_user(&fprog
, user_filter
, sizeof(fprog
)))
501 filter
= seccomp_prepare_filter(&fprog
);
507 * seccomp_attach_filter: validate and attach filter
508 * @flags: flags to change filter behavior
509 * @filter: seccomp filter to add to the current process
511 * Caller must be holding current->sighand->siglock lock.
513 * Returns 0 on success, -ve on error, or
514 * - in TSYNC mode: the pid of a thread which was either not in the correct
515 * seccomp mode or did not have an ancestral seccomp filter
516 * - in NEW_LISTENER mode: the fd of the new listener
518 static long seccomp_attach_filter(unsigned int flags
,
519 struct seccomp_filter
*filter
)
521 unsigned long total_insns
;
522 struct seccomp_filter
*walker
;
524 assert_spin_locked(¤t
->sighand
->siglock
);
526 /* Validate resulting filter length. */
527 total_insns
= filter
->prog
->len
;
528 for (walker
= current
->seccomp
.filter
; walker
; walker
= walker
->prev
)
529 total_insns
+= walker
->prog
->len
+ 4; /* 4 instr penalty */
530 if (total_insns
> MAX_INSNS_PER_PATH
)
533 /* If thread sync has been requested, check that it is possible. */
534 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
) {
537 ret
= seccomp_can_sync_threads();
542 /* Set log flag, if present. */
543 if (flags
& SECCOMP_FILTER_FLAG_LOG
)
547 * If there is an existing filter, make it the prev and don't drop its
550 filter
->prev
= current
->seccomp
.filter
;
551 current
->seccomp
.filter
= filter
;
553 /* Now that the new filter is in place, synchronize to all threads. */
554 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
)
555 seccomp_sync_threads(flags
);
560 static void __get_seccomp_filter(struct seccomp_filter
*filter
)
562 refcount_inc(&filter
->usage
);
565 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
566 void get_seccomp_filter(struct task_struct
*tsk
)
568 struct seccomp_filter
*orig
= tsk
->seccomp
.filter
;
571 __get_seccomp_filter(orig
);
574 static inline void seccomp_filter_free(struct seccomp_filter
*filter
)
577 bpf_prog_destroy(filter
->prog
);
582 static void __put_seccomp_filter(struct seccomp_filter
*orig
)
584 /* Clean up single-reference branches iteratively. */
585 while (orig
&& refcount_dec_and_test(&orig
->usage
)) {
586 struct seccomp_filter
*freeme
= orig
;
588 seccomp_filter_free(freeme
);
592 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
593 void put_seccomp_filter(struct task_struct
*tsk
)
595 __put_seccomp_filter(tsk
->seccomp
.filter
);
598 static void seccomp_init_siginfo(kernel_siginfo_t
*info
, int syscall
, int reason
)
601 info
->si_signo
= SIGSYS
;
602 info
->si_code
= SYS_SECCOMP
;
603 info
->si_call_addr
= (void __user
*)KSTK_EIP(current
);
604 info
->si_errno
= reason
;
605 info
->si_arch
= syscall_get_arch(current
);
606 info
->si_syscall
= syscall
;
610 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
611 * @syscall: syscall number to send to userland
612 * @reason: filter-supplied reason code to send to userland (via si_errno)
614 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
616 static void seccomp_send_sigsys(int syscall
, int reason
)
618 struct kernel_siginfo info
;
619 seccomp_init_siginfo(&info
, syscall
, reason
);
620 force_sig_info(&info
);
622 #endif /* CONFIG_SECCOMP_FILTER */
624 /* For use with seccomp_actions_logged */
625 #define SECCOMP_LOG_KILL_PROCESS (1 << 0)
626 #define SECCOMP_LOG_KILL_THREAD (1 << 1)
627 #define SECCOMP_LOG_TRAP (1 << 2)
628 #define SECCOMP_LOG_ERRNO (1 << 3)
629 #define SECCOMP_LOG_TRACE (1 << 4)
630 #define SECCOMP_LOG_LOG (1 << 5)
631 #define SECCOMP_LOG_ALLOW (1 << 6)
632 #define SECCOMP_LOG_USER_NOTIF (1 << 7)
634 static u32 seccomp_actions_logged
= SECCOMP_LOG_KILL_PROCESS
|
635 SECCOMP_LOG_KILL_THREAD
|
638 SECCOMP_LOG_USER_NOTIF
|
642 static inline void seccomp_log(unsigned long syscall
, long signr
, u32 action
,
648 case SECCOMP_RET_ALLOW
:
650 case SECCOMP_RET_TRAP
:
651 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_TRAP
;
653 case SECCOMP_RET_ERRNO
:
654 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_ERRNO
;
656 case SECCOMP_RET_TRACE
:
657 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_TRACE
;
659 case SECCOMP_RET_USER_NOTIF
:
660 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_USER_NOTIF
;
662 case SECCOMP_RET_LOG
:
663 log
= seccomp_actions_logged
& SECCOMP_LOG_LOG
;
665 case SECCOMP_RET_KILL_THREAD
:
666 log
= seccomp_actions_logged
& SECCOMP_LOG_KILL_THREAD
;
668 case SECCOMP_RET_KILL_PROCESS
:
670 log
= seccomp_actions_logged
& SECCOMP_LOG_KILL_PROCESS
;
674 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
675 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
676 * any action from being logged by removing the action name from the
677 * seccomp_actions_logged sysctl.
682 audit_seccomp(syscall
, signr
, action
);
686 * Secure computing mode 1 allows only read/write/exit/sigreturn.
687 * To be fully secure this must be combined with rlimit
688 * to limit the stack allocations too.
690 static const int mode1_syscalls
[] = {
691 __NR_seccomp_read
, __NR_seccomp_write
, __NR_seccomp_exit
, __NR_seccomp_sigreturn
,
692 0, /* null terminated */
695 static void __secure_computing_strict(int this_syscall
)
697 const int *syscall_whitelist
= mode1_syscalls
;
699 if (in_compat_syscall())
700 syscall_whitelist
= get_compat_mode1_syscalls();
703 if (*syscall_whitelist
== this_syscall
)
705 } while (*++syscall_whitelist
);
710 seccomp_log(this_syscall
, SIGKILL
, SECCOMP_RET_KILL_THREAD
, true);
714 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
715 void secure_computing_strict(int this_syscall
)
717 int mode
= current
->seccomp
.mode
;
719 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE
) &&
720 unlikely(current
->ptrace
& PT_SUSPEND_SECCOMP
))
723 if (mode
== SECCOMP_MODE_DISABLED
)
725 else if (mode
== SECCOMP_MODE_STRICT
)
726 __secure_computing_strict(this_syscall
);
732 #ifdef CONFIG_SECCOMP_FILTER
733 static u64
seccomp_next_notify_id(struct seccomp_filter
*filter
)
736 * Note: overflow is ok here, the id just needs to be unique per
739 lockdep_assert_held(&filter
->notify_lock
);
740 return filter
->notif
->next_id
++;
743 static int seccomp_do_user_notification(int this_syscall
,
744 struct seccomp_filter
*match
,
745 const struct seccomp_data
*sd
)
750 struct seccomp_knotif n
= {};
752 mutex_lock(&match
->notify_lock
);
758 n
.state
= SECCOMP_NOTIFY_INIT
;
760 n
.id
= seccomp_next_notify_id(match
);
761 init_completion(&n
.ready
);
762 list_add(&n
.list
, &match
->notif
->notifications
);
764 up(&match
->notif
->request
);
765 wake_up_poll(&match
->notif
->wqh
, EPOLLIN
| EPOLLRDNORM
);
766 mutex_unlock(&match
->notify_lock
);
769 * This is where we wait for a reply from userspace.
771 err
= wait_for_completion_interruptible(&n
.ready
);
772 mutex_lock(&match
->notify_lock
);
780 * Note that it's possible the listener died in between the time when
781 * we were notified of a respons (or a signal) and when we were able to
782 * re-acquire the lock, so only delete from the list if the
783 * notification actually exists.
785 * Also note that this test is only valid because there's no way to
786 * *reattach* to a notifier right now. If one is added, we'll need to
787 * keep track of the notif itself and make sure they match here.
792 mutex_unlock(&match
->notify_lock
);
794 /* Userspace requests to continue the syscall. */
795 if (flags
& SECCOMP_USER_NOTIF_FLAG_CONTINUE
)
798 syscall_set_return_value(current
, task_pt_regs(current
),
803 static int __seccomp_filter(int this_syscall
, const struct seccomp_data
*sd
,
804 const bool recheck_after_trace
)
806 u32 filter_ret
, action
;
807 struct seccomp_filter
*match
= NULL
;
809 struct seccomp_data sd_local
;
812 * Make sure that any changes to mode from another thread have
813 * been seen after TIF_SECCOMP was seen.
818 populate_seccomp_data(&sd_local
);
822 filter_ret
= seccomp_run_filters(sd
, &match
);
823 data
= filter_ret
& SECCOMP_RET_DATA
;
824 action
= filter_ret
& SECCOMP_RET_ACTION_FULL
;
827 case SECCOMP_RET_ERRNO
:
828 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
829 if (data
> MAX_ERRNO
)
831 syscall_set_return_value(current
, task_pt_regs(current
),
835 case SECCOMP_RET_TRAP
:
836 /* Show the handler the original registers. */
837 syscall_rollback(current
, task_pt_regs(current
));
838 /* Let the filter pass back 16 bits of data. */
839 seccomp_send_sigsys(this_syscall
, data
);
842 case SECCOMP_RET_TRACE
:
843 /* We've been put in this state by the ptracer already. */
844 if (recheck_after_trace
)
847 /* ENOSYS these calls if there is no tracer attached. */
848 if (!ptrace_event_enabled(current
, PTRACE_EVENT_SECCOMP
)) {
849 syscall_set_return_value(current
,
850 task_pt_regs(current
),
855 /* Allow the BPF to provide the event message */
856 ptrace_event(PTRACE_EVENT_SECCOMP
, data
);
858 * The delivery of a fatal signal during event
859 * notification may silently skip tracer notification,
860 * which could leave us with a potentially unmodified
861 * syscall that the tracer would have liked to have
862 * changed. Since the process is about to die, we just
863 * force the syscall to be skipped and let the signal
864 * kill the process and correctly handle any tracer exit
867 if (fatal_signal_pending(current
))
869 /* Check if the tracer forced the syscall to be skipped. */
870 this_syscall
= syscall_get_nr(current
, task_pt_regs(current
));
871 if (this_syscall
< 0)
875 * Recheck the syscall, since it may have changed. This
876 * intentionally uses a NULL struct seccomp_data to force
877 * a reload of all registers. This does not goto skip since
878 * a skip would have already been reported.
880 if (__seccomp_filter(this_syscall
, NULL
, true))
885 case SECCOMP_RET_USER_NOTIF
:
886 if (seccomp_do_user_notification(this_syscall
, match
, sd
))
891 case SECCOMP_RET_LOG
:
892 seccomp_log(this_syscall
, 0, action
, true);
895 case SECCOMP_RET_ALLOW
:
897 * Note that the "match" filter will always be NULL for
898 * this action since SECCOMP_RET_ALLOW is the starting
899 * state in seccomp_run_filters().
903 case SECCOMP_RET_KILL_THREAD
:
904 case SECCOMP_RET_KILL_PROCESS
:
906 seccomp_log(this_syscall
, SIGSYS
, action
, true);
907 /* Dump core only if this is the last remaining thread. */
908 if (action
== SECCOMP_RET_KILL_PROCESS
||
909 get_nr_threads(current
) == 1) {
910 kernel_siginfo_t info
;
912 /* Show the original registers in the dump. */
913 syscall_rollback(current
, task_pt_regs(current
));
914 /* Trigger a manual coredump since do_exit skips it. */
915 seccomp_init_siginfo(&info
, this_syscall
, data
);
918 if (action
== SECCOMP_RET_KILL_PROCESS
)
919 do_group_exit(SIGSYS
);
927 seccomp_log(this_syscall
, 0, action
, match
? match
->log
: false);
931 static int __seccomp_filter(int this_syscall
, const struct seccomp_data
*sd
,
932 const bool recheck_after_trace
)
940 int __secure_computing(const struct seccomp_data
*sd
)
942 int mode
= current
->seccomp
.mode
;
945 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE
) &&
946 unlikely(current
->ptrace
& PT_SUSPEND_SECCOMP
))
949 this_syscall
= sd
? sd
->nr
:
950 syscall_get_nr(current
, task_pt_regs(current
));
953 case SECCOMP_MODE_STRICT
:
954 __secure_computing_strict(this_syscall
); /* may call do_exit */
956 case SECCOMP_MODE_FILTER
:
957 return __seccomp_filter(this_syscall
, sd
, false);
962 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
964 long prctl_get_seccomp(void)
966 return current
->seccomp
.mode
;
970 * seccomp_set_mode_strict: internal function for setting strict seccomp
972 * Once current->seccomp.mode is non-zero, it may not be changed.
974 * Returns 0 on success or -EINVAL on failure.
976 static long seccomp_set_mode_strict(void)
978 const unsigned long seccomp_mode
= SECCOMP_MODE_STRICT
;
981 spin_lock_irq(¤t
->sighand
->siglock
);
983 if (!seccomp_may_assign_mode(seccomp_mode
))
989 seccomp_assign_mode(current
, seccomp_mode
, 0);
993 spin_unlock_irq(¤t
->sighand
->siglock
);
998 #ifdef CONFIG_SECCOMP_FILTER
999 static int seccomp_notify_release(struct inode
*inode
, struct file
*file
)
1001 struct seccomp_filter
*filter
= file
->private_data
;
1002 struct seccomp_knotif
*knotif
;
1007 mutex_lock(&filter
->notify_lock
);
1010 * If this file is being closed because e.g. the task who owned it
1011 * died, let's wake everyone up who was waiting on us.
1013 list_for_each_entry(knotif
, &filter
->notif
->notifications
, list
) {
1014 if (knotif
->state
== SECCOMP_NOTIFY_REPLIED
)
1017 knotif
->state
= SECCOMP_NOTIFY_REPLIED
;
1018 knotif
->error
= -ENOSYS
;
1021 complete(&knotif
->ready
);
1024 kfree(filter
->notif
);
1025 filter
->notif
= NULL
;
1026 mutex_unlock(&filter
->notify_lock
);
1027 __put_seccomp_filter(filter
);
1031 static long seccomp_notify_recv(struct seccomp_filter
*filter
,
1034 struct seccomp_knotif
*knotif
= NULL
, *cur
;
1035 struct seccomp_notif unotif
;
1038 /* Verify that we're not given garbage to keep struct extensible. */
1039 ret
= check_zeroed_user(buf
, sizeof(unotif
));
1045 memset(&unotif
, 0, sizeof(unotif
));
1047 ret
= down_interruptible(&filter
->notif
->request
);
1051 mutex_lock(&filter
->notify_lock
);
1052 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1053 if (cur
->state
== SECCOMP_NOTIFY_INIT
) {
1060 * If we didn't find a notification, it could be that the task was
1061 * interrupted by a fatal signal between the time we were woken and
1062 * when we were able to acquire the rw lock.
1069 unotif
.id
= knotif
->id
;
1070 unotif
.pid
= task_pid_vnr(knotif
->task
);
1071 unotif
.data
= *(knotif
->data
);
1073 knotif
->state
= SECCOMP_NOTIFY_SENT
;
1074 wake_up_poll(&filter
->notif
->wqh
, EPOLLOUT
| EPOLLWRNORM
);
1077 mutex_unlock(&filter
->notify_lock
);
1079 if (ret
== 0 && copy_to_user(buf
, &unotif
, sizeof(unotif
))) {
1083 * Userspace screwed up. To make sure that we keep this
1084 * notification alive, let's reset it back to INIT. It
1085 * may have died when we released the lock, so we need to make
1086 * sure it's still around.
1089 mutex_lock(&filter
->notify_lock
);
1090 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1091 if (cur
->id
== unotif
.id
) {
1098 knotif
->state
= SECCOMP_NOTIFY_INIT
;
1099 up(&filter
->notif
->request
);
1101 mutex_unlock(&filter
->notify_lock
);
1107 static long seccomp_notify_send(struct seccomp_filter
*filter
,
1110 struct seccomp_notif_resp resp
= {};
1111 struct seccomp_knotif
*knotif
= NULL
, *cur
;
1114 if (copy_from_user(&resp
, buf
, sizeof(resp
)))
1117 if (resp
.flags
& ~SECCOMP_USER_NOTIF_FLAG_CONTINUE
)
1120 if ((resp
.flags
& SECCOMP_USER_NOTIF_FLAG_CONTINUE
) &&
1121 (resp
.error
|| resp
.val
))
1124 ret
= mutex_lock_interruptible(&filter
->notify_lock
);
1128 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1129 if (cur
->id
== resp
.id
) {
1140 /* Allow exactly one reply. */
1141 if (knotif
->state
!= SECCOMP_NOTIFY_SENT
) {
1147 knotif
->state
= SECCOMP_NOTIFY_REPLIED
;
1148 knotif
->error
= resp
.error
;
1149 knotif
->val
= resp
.val
;
1150 knotif
->flags
= resp
.flags
;
1151 complete(&knotif
->ready
);
1153 mutex_unlock(&filter
->notify_lock
);
1157 static long seccomp_notify_id_valid(struct seccomp_filter
*filter
,
1160 struct seccomp_knotif
*knotif
= NULL
;
1164 if (copy_from_user(&id
, buf
, sizeof(id
)))
1167 ret
= mutex_lock_interruptible(&filter
->notify_lock
);
1172 list_for_each_entry(knotif
, &filter
->notif
->notifications
, list
) {
1173 if (knotif
->id
== id
) {
1174 if (knotif
->state
== SECCOMP_NOTIFY_SENT
)
1181 mutex_unlock(&filter
->notify_lock
);
1185 static long seccomp_notify_ioctl(struct file
*file
, unsigned int cmd
,
1188 struct seccomp_filter
*filter
= file
->private_data
;
1189 void __user
*buf
= (void __user
*)arg
;
1192 case SECCOMP_IOCTL_NOTIF_RECV
:
1193 return seccomp_notify_recv(filter
, buf
);
1194 case SECCOMP_IOCTL_NOTIF_SEND
:
1195 return seccomp_notify_send(filter
, buf
);
1196 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR
:
1197 case SECCOMP_IOCTL_NOTIF_ID_VALID
:
1198 return seccomp_notify_id_valid(filter
, buf
);
1204 static __poll_t
seccomp_notify_poll(struct file
*file
,
1205 struct poll_table_struct
*poll_tab
)
1207 struct seccomp_filter
*filter
= file
->private_data
;
1209 struct seccomp_knotif
*cur
;
1211 poll_wait(file
, &filter
->notif
->wqh
, poll_tab
);
1213 if (mutex_lock_interruptible(&filter
->notify_lock
) < 0)
1216 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1217 if (cur
->state
== SECCOMP_NOTIFY_INIT
)
1218 ret
|= EPOLLIN
| EPOLLRDNORM
;
1219 if (cur
->state
== SECCOMP_NOTIFY_SENT
)
1220 ret
|= EPOLLOUT
| EPOLLWRNORM
;
1221 if ((ret
& EPOLLIN
) && (ret
& EPOLLOUT
))
1225 mutex_unlock(&filter
->notify_lock
);
1230 static const struct file_operations seccomp_notify_ops
= {
1231 .poll
= seccomp_notify_poll
,
1232 .release
= seccomp_notify_release
,
1233 .unlocked_ioctl
= seccomp_notify_ioctl
,
1234 .compat_ioctl
= seccomp_notify_ioctl
,
1237 static struct file
*init_listener(struct seccomp_filter
*filter
)
1241 ret
= ERR_PTR(-ENOMEM
);
1242 filter
->notif
= kzalloc(sizeof(*(filter
->notif
)), GFP_KERNEL
);
1246 sema_init(&filter
->notif
->request
, 0);
1247 filter
->notif
->next_id
= get_random_u64();
1248 INIT_LIST_HEAD(&filter
->notif
->notifications
);
1249 init_waitqueue_head(&filter
->notif
->wqh
);
1251 ret
= anon_inode_getfile("seccomp notify", &seccomp_notify_ops
,
1256 /* The file has a reference to it now */
1257 __get_seccomp_filter(filter
);
1261 kfree(filter
->notif
);
1267 * Does @new_child have a listener while an ancestor also has a listener?
1268 * If so, we'll want to reject this filter.
1269 * This only has to be tested for the current process, even in the TSYNC case,
1270 * because TSYNC installs @child with the same parent on all threads.
1271 * Note that @new_child is not hooked up to its parent at this point yet, so
1272 * we use current->seccomp.filter.
1274 static bool has_duplicate_listener(struct seccomp_filter
*new_child
)
1276 struct seccomp_filter
*cur
;
1278 /* must be protected against concurrent TSYNC */
1279 lockdep_assert_held(¤t
->sighand
->siglock
);
1281 if (!new_child
->notif
)
1283 for (cur
= current
->seccomp
.filter
; cur
; cur
= cur
->prev
) {
1292 * seccomp_set_mode_filter: internal function for setting seccomp filter
1293 * @flags: flags to change filter behavior
1294 * @filter: struct sock_fprog containing filter
1296 * This function may be called repeatedly to install additional filters.
1297 * Every filter successfully installed will be evaluated (in reverse order)
1298 * for each system call the task makes.
1300 * Once current->seccomp.mode is non-zero, it may not be changed.
1302 * Returns 0 on success or -EINVAL on failure.
1304 static long seccomp_set_mode_filter(unsigned int flags
,
1305 const char __user
*filter
)
1307 const unsigned long seccomp_mode
= SECCOMP_MODE_FILTER
;
1308 struct seccomp_filter
*prepared
= NULL
;
1311 struct file
*listener_f
= NULL
;
1313 /* Validate flags. */
1314 if (flags
& ~SECCOMP_FILTER_FLAG_MASK
)
1318 * In the successful case, NEW_LISTENER returns the new listener fd.
1319 * But in the failure case, TSYNC returns the thread that died. If you
1320 * combine these two flags, there's no way to tell whether something
1321 * succeeded or failed. So, let's disallow this combination.
1323 if ((flags
& SECCOMP_FILTER_FLAG_TSYNC
) &&
1324 (flags
& SECCOMP_FILTER_FLAG_NEW_LISTENER
))
1327 /* Prepare the new filter before holding any locks. */
1328 prepared
= seccomp_prepare_user_filter(filter
);
1329 if (IS_ERR(prepared
))
1330 return PTR_ERR(prepared
);
1332 if (flags
& SECCOMP_FILTER_FLAG_NEW_LISTENER
) {
1333 listener
= get_unused_fd_flags(O_CLOEXEC
);
1339 listener_f
= init_listener(prepared
);
1340 if (IS_ERR(listener_f
)) {
1341 put_unused_fd(listener
);
1342 ret
= PTR_ERR(listener_f
);
1348 * Make sure we cannot change seccomp or nnp state via TSYNC
1349 * while another thread is in the middle of calling exec.
1351 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
&&
1352 mutex_lock_killable(¤t
->signal
->cred_guard_mutex
))
1355 spin_lock_irq(¤t
->sighand
->siglock
);
1357 if (!seccomp_may_assign_mode(seccomp_mode
))
1360 if (has_duplicate_listener(prepared
)) {
1365 ret
= seccomp_attach_filter(flags
, prepared
);
1368 /* Do not free the successfully attached filter. */
1371 seccomp_assign_mode(current
, seccomp_mode
, flags
);
1373 spin_unlock_irq(¤t
->sighand
->siglock
);
1374 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
)
1375 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1377 if (flags
& SECCOMP_FILTER_FLAG_NEW_LISTENER
) {
1379 listener_f
->private_data
= NULL
;
1381 put_unused_fd(listener
);
1383 fd_install(listener
, listener_f
);
1388 seccomp_filter_free(prepared
);
1392 static inline long seccomp_set_mode_filter(unsigned int flags
,
1393 const char __user
*filter
)
1399 static long seccomp_get_action_avail(const char __user
*uaction
)
1403 if (copy_from_user(&action
, uaction
, sizeof(action
)))
1407 case SECCOMP_RET_KILL_PROCESS
:
1408 case SECCOMP_RET_KILL_THREAD
:
1409 case SECCOMP_RET_TRAP
:
1410 case SECCOMP_RET_ERRNO
:
1411 case SECCOMP_RET_USER_NOTIF
:
1412 case SECCOMP_RET_TRACE
:
1413 case SECCOMP_RET_LOG
:
1414 case SECCOMP_RET_ALLOW
:
1423 static long seccomp_get_notif_sizes(void __user
*usizes
)
1425 struct seccomp_notif_sizes sizes
= {
1426 .seccomp_notif
= sizeof(struct seccomp_notif
),
1427 .seccomp_notif_resp
= sizeof(struct seccomp_notif_resp
),
1428 .seccomp_data
= sizeof(struct seccomp_data
),
1431 if (copy_to_user(usizes
, &sizes
, sizeof(sizes
)))
1437 /* Common entry point for both prctl and syscall. */
1438 static long do_seccomp(unsigned int op
, unsigned int flags
,
1442 case SECCOMP_SET_MODE_STRICT
:
1443 if (flags
!= 0 || uargs
!= NULL
)
1445 return seccomp_set_mode_strict();
1446 case SECCOMP_SET_MODE_FILTER
:
1447 return seccomp_set_mode_filter(flags
, uargs
);
1448 case SECCOMP_GET_ACTION_AVAIL
:
1452 return seccomp_get_action_avail(uargs
);
1453 case SECCOMP_GET_NOTIF_SIZES
:
1457 return seccomp_get_notif_sizes(uargs
);
1463 SYSCALL_DEFINE3(seccomp
, unsigned int, op
, unsigned int, flags
,
1464 void __user
*, uargs
)
1466 return do_seccomp(op
, flags
, uargs
);
1470 * prctl_set_seccomp: configures current->seccomp.mode
1471 * @seccomp_mode: requested mode to use
1472 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1474 * Returns 0 on success or -EINVAL on failure.
1476 long prctl_set_seccomp(unsigned long seccomp_mode
, void __user
*filter
)
1481 switch (seccomp_mode
) {
1482 case SECCOMP_MODE_STRICT
:
1483 op
= SECCOMP_SET_MODE_STRICT
;
1485 * Setting strict mode through prctl always ignored filter,
1486 * so make sure it is always NULL here to pass the internal
1487 * check in do_seccomp().
1491 case SECCOMP_MODE_FILTER
:
1492 op
= SECCOMP_SET_MODE_FILTER
;
1499 /* prctl interface doesn't have flags, so they are always zero. */
1500 return do_seccomp(op
, 0, uargs
);
1503 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
1504 static struct seccomp_filter
*get_nth_filter(struct task_struct
*task
,
1505 unsigned long filter_off
)
1507 struct seccomp_filter
*orig
, *filter
;
1508 unsigned long count
;
1511 * Note: this is only correct because the caller should be the (ptrace)
1512 * tracer of the task, otherwise lock_task_sighand is needed.
1514 spin_lock_irq(&task
->sighand
->siglock
);
1516 if (task
->seccomp
.mode
!= SECCOMP_MODE_FILTER
) {
1517 spin_unlock_irq(&task
->sighand
->siglock
);
1518 return ERR_PTR(-EINVAL
);
1521 orig
= task
->seccomp
.filter
;
1522 __get_seccomp_filter(orig
);
1523 spin_unlock_irq(&task
->sighand
->siglock
);
1526 for (filter
= orig
; filter
; filter
= filter
->prev
)
1529 if (filter_off
>= count
) {
1530 filter
= ERR_PTR(-ENOENT
);
1534 count
-= filter_off
;
1535 for (filter
= orig
; filter
&& count
> 1; filter
= filter
->prev
)
1538 if (WARN_ON(count
!= 1 || !filter
)) {
1539 filter
= ERR_PTR(-ENOENT
);
1543 __get_seccomp_filter(filter
);
1546 __put_seccomp_filter(orig
);
1550 long seccomp_get_filter(struct task_struct
*task
, unsigned long filter_off
,
1553 struct seccomp_filter
*filter
;
1554 struct sock_fprog_kern
*fprog
;
1557 if (!capable(CAP_SYS_ADMIN
) ||
1558 current
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
) {
1562 filter
= get_nth_filter(task
, filter_off
);
1564 return PTR_ERR(filter
);
1566 fprog
= filter
->prog
->orig_prog
;
1568 /* This must be a new non-cBPF filter, since we save
1569 * every cBPF filter's orig_prog above when
1570 * CONFIG_CHECKPOINT_RESTORE is enabled.
1580 if (copy_to_user(data
, fprog
->filter
, bpf_classic_proglen(fprog
)))
1584 __put_seccomp_filter(filter
);
1588 long seccomp_get_metadata(struct task_struct
*task
,
1589 unsigned long size
, void __user
*data
)
1592 struct seccomp_filter
*filter
;
1593 struct seccomp_metadata kmd
= {};
1595 if (!capable(CAP_SYS_ADMIN
) ||
1596 current
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
) {
1600 size
= min_t(unsigned long, size
, sizeof(kmd
));
1602 if (size
< sizeof(kmd
.filter_off
))
1605 if (copy_from_user(&kmd
.filter_off
, data
, sizeof(kmd
.filter_off
)))
1608 filter
= get_nth_filter(task
, kmd
.filter_off
);
1610 return PTR_ERR(filter
);
1613 kmd
.flags
|= SECCOMP_FILTER_FLAG_LOG
;
1616 if (copy_to_user(data
, &kmd
, size
))
1619 __put_seccomp_filter(filter
);
1624 #ifdef CONFIG_SYSCTL
1626 /* Human readable action names for friendly sysctl interaction */
1627 #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
1628 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
1629 #define SECCOMP_RET_TRAP_NAME "trap"
1630 #define SECCOMP_RET_ERRNO_NAME "errno"
1631 #define SECCOMP_RET_USER_NOTIF_NAME "user_notif"
1632 #define SECCOMP_RET_TRACE_NAME "trace"
1633 #define SECCOMP_RET_LOG_NAME "log"
1634 #define SECCOMP_RET_ALLOW_NAME "allow"
1636 static const char seccomp_actions_avail
[] =
1637 SECCOMP_RET_KILL_PROCESS_NAME
" "
1638 SECCOMP_RET_KILL_THREAD_NAME
" "
1639 SECCOMP_RET_TRAP_NAME
" "
1640 SECCOMP_RET_ERRNO_NAME
" "
1641 SECCOMP_RET_USER_NOTIF_NAME
" "
1642 SECCOMP_RET_TRACE_NAME
" "
1643 SECCOMP_RET_LOG_NAME
" "
1644 SECCOMP_RET_ALLOW_NAME
;
1646 struct seccomp_log_name
{
1651 static const struct seccomp_log_name seccomp_log_names
[] = {
1652 { SECCOMP_LOG_KILL_PROCESS
, SECCOMP_RET_KILL_PROCESS_NAME
},
1653 { SECCOMP_LOG_KILL_THREAD
, SECCOMP_RET_KILL_THREAD_NAME
},
1654 { SECCOMP_LOG_TRAP
, SECCOMP_RET_TRAP_NAME
},
1655 { SECCOMP_LOG_ERRNO
, SECCOMP_RET_ERRNO_NAME
},
1656 { SECCOMP_LOG_USER_NOTIF
, SECCOMP_RET_USER_NOTIF_NAME
},
1657 { SECCOMP_LOG_TRACE
, SECCOMP_RET_TRACE_NAME
},
1658 { SECCOMP_LOG_LOG
, SECCOMP_RET_LOG_NAME
},
1659 { SECCOMP_LOG_ALLOW
, SECCOMP_RET_ALLOW_NAME
},
1663 static bool seccomp_names_from_actions_logged(char *names
, size_t size
,
1667 const struct seccomp_log_name
*cur
;
1668 bool append_sep
= false;
1670 for (cur
= seccomp_log_names
; cur
->name
&& size
; cur
++) {
1673 if (!(actions_logged
& cur
->log
))
1677 ret
= strscpy(names
, sep
, size
);
1686 ret
= strscpy(names
, cur
->name
, size
);
1697 static bool seccomp_action_logged_from_name(u32
*action_logged
,
1700 const struct seccomp_log_name
*cur
;
1702 for (cur
= seccomp_log_names
; cur
->name
; cur
++) {
1703 if (!strcmp(cur
->name
, name
)) {
1704 *action_logged
= cur
->log
;
1712 static bool seccomp_actions_logged_from_names(u32
*actions_logged
, char *names
)
1716 *actions_logged
= 0;
1717 while ((name
= strsep(&names
, " ")) && *name
) {
1718 u32 action_logged
= 0;
1720 if (!seccomp_action_logged_from_name(&action_logged
, name
))
1723 *actions_logged
|= action_logged
;
1729 static int read_actions_logged(struct ctl_table
*ro_table
, void __user
*buffer
,
1730 size_t *lenp
, loff_t
*ppos
)
1732 char names
[sizeof(seccomp_actions_avail
)];
1733 struct ctl_table table
;
1735 memset(names
, 0, sizeof(names
));
1737 if (!seccomp_names_from_actions_logged(names
, sizeof(names
),
1738 seccomp_actions_logged
, " "))
1743 table
.maxlen
= sizeof(names
);
1744 return proc_dostring(&table
, 0, buffer
, lenp
, ppos
);
1747 static int write_actions_logged(struct ctl_table
*ro_table
, void __user
*buffer
,
1748 size_t *lenp
, loff_t
*ppos
, u32
*actions_logged
)
1750 char names
[sizeof(seccomp_actions_avail
)];
1751 struct ctl_table table
;
1754 if (!capable(CAP_SYS_ADMIN
))
1757 memset(names
, 0, sizeof(names
));
1761 table
.maxlen
= sizeof(names
);
1762 ret
= proc_dostring(&table
, 1, buffer
, lenp
, ppos
);
1766 if (!seccomp_actions_logged_from_names(actions_logged
, table
.data
))
1769 if (*actions_logged
& SECCOMP_LOG_ALLOW
)
1772 seccomp_actions_logged
= *actions_logged
;
1776 static void audit_actions_logged(u32 actions_logged
, u32 old_actions_logged
,
1779 char names
[sizeof(seccomp_actions_avail
)];
1780 char old_names
[sizeof(seccomp_actions_avail
)];
1781 const char *new = names
;
1782 const char *old
= old_names
;
1787 memset(names
, 0, sizeof(names
));
1788 memset(old_names
, 0, sizeof(old_names
));
1792 else if (!actions_logged
)
1794 else if (!seccomp_names_from_actions_logged(names
, sizeof(names
),
1795 actions_logged
, ","))
1798 if (!old_actions_logged
)
1800 else if (!seccomp_names_from_actions_logged(old_names
,
1802 old_actions_logged
, ","))
1805 return audit_seccomp_actions_logged(new, old
, !ret
);
1808 static int seccomp_actions_logged_handler(struct ctl_table
*ro_table
, int write
,
1809 void __user
*buffer
, size_t *lenp
,
1815 u32 actions_logged
= 0;
1816 u32 old_actions_logged
= seccomp_actions_logged
;
1818 ret
= write_actions_logged(ro_table
, buffer
, lenp
, ppos
,
1820 audit_actions_logged(actions_logged
, old_actions_logged
, ret
);
1822 ret
= read_actions_logged(ro_table
, buffer
, lenp
, ppos
);
1827 static struct ctl_path seccomp_sysctl_path
[] = {
1828 { .procname
= "kernel", },
1829 { .procname
= "seccomp", },
1833 static struct ctl_table seccomp_sysctl_table
[] = {
1835 .procname
= "actions_avail",
1836 .data
= (void *) &seccomp_actions_avail
,
1837 .maxlen
= sizeof(seccomp_actions_avail
),
1839 .proc_handler
= proc_dostring
,
1842 .procname
= "actions_logged",
1844 .proc_handler
= seccomp_actions_logged_handler
,
1849 static int __init
seccomp_sysctl_init(void)
1851 struct ctl_table_header
*hdr
;
1853 hdr
= register_sysctl_paths(seccomp_sysctl_path
, seccomp_sysctl_table
);
1855 pr_warn("seccomp: sysctl registration failed\n");
1857 kmemleak_not_leak(hdr
);
1862 device_initcall(seccomp_sysctl_init
)
1864 #endif /* CONFIG_SYSCTL */