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 * @ioctl_flags: The flags used for the seccomp_addfd ioctl.
114 * @ret: The return value of the installing process. It is set to the fd num
115 * upon success (>= 0).
116 * @completion: Indicates that the installing process has completed fd
117 * installation, or gone away (either due to successful
121 struct seccomp_kaddfd
{
129 /* To only be set on reply */
132 struct completion completion
;
133 struct list_head list
;
137 * struct notification - container for seccomp userspace notifications. Since
138 * most seccomp filters will not have notification listeners attached and this
139 * structure is fairly large, we store the notification-specific stuff in a
140 * separate structure.
142 * @request: A semaphore that users of this notification can wait on for
143 * changes. Actual reads and writes are still controlled with
144 * filter->notify_lock.
145 * @next_id: The id of the next request.
146 * @notifications: A list of struct seccomp_knotif elements.
148 struct notification
{
149 struct semaphore request
;
151 struct list_head notifications
;
154 #ifdef SECCOMP_ARCH_NATIVE
156 * struct action_cache - per-filter cache of seccomp actions per
159 * @allow_native: A bitmap where each bit represents whether the
160 * filter will always allow the syscall, for the
161 * native architecture.
162 * @allow_compat: A bitmap where each bit represents whether the
163 * filter will always allow the syscall, for the
164 * compat architecture.
166 struct action_cache
{
167 DECLARE_BITMAP(allow_native
, SECCOMP_ARCH_NATIVE_NR
);
168 #ifdef SECCOMP_ARCH_COMPAT
169 DECLARE_BITMAP(allow_compat
, SECCOMP_ARCH_COMPAT_NR
);
173 struct action_cache
{ };
175 static inline bool seccomp_cache_check_allow(const struct seccomp_filter
*sfilter
,
176 const struct seccomp_data
*sd
)
181 static inline void seccomp_cache_prepare(struct seccomp_filter
*sfilter
)
184 #endif /* SECCOMP_ARCH_NATIVE */
187 * struct seccomp_filter - container for seccomp BPF programs
189 * @refs: Reference count to manage the object lifetime.
190 * A filter's reference count is incremented for each directly
191 * attached task, once for the dependent filter, and if
192 * requested for the user notifier. When @refs reaches zero,
193 * the filter can be freed.
194 * @users: A filter's @users count is incremented for each directly
195 * attached task (filter installation, fork(), thread_sync),
196 * and once for the dependent filter (tracked in filter->prev).
197 * When it reaches zero it indicates that no direct or indirect
198 * users of that filter exist. No new tasks can get associated with
199 * this filter after reaching 0. The @users count is always smaller
200 * or equal to @refs. Hence, reaching 0 for @users does not mean
201 * the filter can be freed.
202 * @cache: cache of arch/syscall mappings to actions
203 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
204 * @prev: points to a previously installed, or inherited, filter
205 * @prog: the BPF program to evaluate
206 * @notif: the struct that holds all notification related information
207 * @notify_lock: A lock for all notification-related accesses.
208 * @wqh: A wait queue for poll if a notifier is in use.
210 * seccomp_filter objects are organized in a tree linked via the @prev
211 * pointer. For any task, it appears to be a singly-linked list starting
212 * with current->seccomp.filter, the most recently attached or inherited filter.
213 * However, multiple filters may share a @prev node, by way of fork(), which
214 * results in a unidirectional tree existing in memory. This is similar to
215 * how namespaces work.
217 * seccomp_filter objects should never be modified after being attached
218 * to a task_struct (other than @refs).
220 struct seccomp_filter
{
224 struct action_cache cache
;
225 struct seccomp_filter
*prev
;
226 struct bpf_prog
*prog
;
227 struct notification
*notif
;
228 struct mutex notify_lock
;
229 wait_queue_head_t wqh
;
232 /* Limit any path through the tree to 256KB worth of instructions. */
233 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
236 * Endianness is explicitly ignored and left for BPF program authors to manage
237 * as per the specific architecture.
239 static void populate_seccomp_data(struct seccomp_data
*sd
)
242 * Instead of using current_pt_reg(), we're already doing the work
243 * to safely fetch "current", so just use "task" everywhere below.
245 struct task_struct
*task
= current
;
246 struct pt_regs
*regs
= task_pt_regs(task
);
247 unsigned long args
[6];
249 sd
->nr
= syscall_get_nr(task
, regs
);
250 sd
->arch
= syscall_get_arch(task
);
251 syscall_get_arguments(task
, regs
, args
);
252 sd
->args
[0] = args
[0];
253 sd
->args
[1] = args
[1];
254 sd
->args
[2] = args
[2];
255 sd
->args
[3] = args
[3];
256 sd
->args
[4] = args
[4];
257 sd
->args
[5] = args
[5];
258 sd
->instruction_pointer
= KSTK_EIP(task
);
262 * seccomp_check_filter - verify seccomp filter code
263 * @filter: filter to verify
264 * @flen: length of filter
266 * Takes a previously checked filter (by bpf_check_classic) and
267 * redirects all filter code that loads struct sk_buff data
268 * and related data through seccomp_bpf_load. It also
269 * enforces length and alignment checking of those loads.
271 * Returns 0 if the rule set is legal or -EINVAL if not.
273 static int seccomp_check_filter(struct sock_filter
*filter
, unsigned int flen
)
276 for (pc
= 0; pc
< flen
; pc
++) {
277 struct sock_filter
*ftest
= &filter
[pc
];
278 u16 code
= ftest
->code
;
282 case BPF_LD
| BPF_W
| BPF_ABS
:
283 ftest
->code
= BPF_LDX
| BPF_W
| BPF_ABS
;
284 /* 32-bit aligned and not out of bounds. */
285 if (k
>= sizeof(struct seccomp_data
) || k
& 3)
288 case BPF_LD
| BPF_W
| BPF_LEN
:
289 ftest
->code
= BPF_LD
| BPF_IMM
;
290 ftest
->k
= sizeof(struct seccomp_data
);
292 case BPF_LDX
| BPF_W
| BPF_LEN
:
293 ftest
->code
= BPF_LDX
| BPF_IMM
;
294 ftest
->k
= sizeof(struct seccomp_data
);
296 /* Explicitly include allowed calls. */
297 case BPF_RET
| BPF_K
:
298 case BPF_RET
| BPF_A
:
299 case BPF_ALU
| BPF_ADD
| BPF_K
:
300 case BPF_ALU
| BPF_ADD
| BPF_X
:
301 case BPF_ALU
| BPF_SUB
| BPF_K
:
302 case BPF_ALU
| BPF_SUB
| BPF_X
:
303 case BPF_ALU
| BPF_MUL
| BPF_K
:
304 case BPF_ALU
| BPF_MUL
| BPF_X
:
305 case BPF_ALU
| BPF_DIV
| BPF_K
:
306 case BPF_ALU
| BPF_DIV
| BPF_X
:
307 case BPF_ALU
| BPF_AND
| BPF_K
:
308 case BPF_ALU
| BPF_AND
| BPF_X
:
309 case BPF_ALU
| BPF_OR
| BPF_K
:
310 case BPF_ALU
| BPF_OR
| BPF_X
:
311 case BPF_ALU
| BPF_XOR
| BPF_K
:
312 case BPF_ALU
| BPF_XOR
| BPF_X
:
313 case BPF_ALU
| BPF_LSH
| BPF_K
:
314 case BPF_ALU
| BPF_LSH
| BPF_X
:
315 case BPF_ALU
| BPF_RSH
| BPF_K
:
316 case BPF_ALU
| BPF_RSH
| BPF_X
:
317 case BPF_ALU
| BPF_NEG
:
318 case BPF_LD
| BPF_IMM
:
319 case BPF_LDX
| BPF_IMM
:
320 case BPF_MISC
| BPF_TAX
:
321 case BPF_MISC
| BPF_TXA
:
322 case BPF_LD
| BPF_MEM
:
323 case BPF_LDX
| BPF_MEM
:
326 case BPF_JMP
| BPF_JA
:
327 case BPF_JMP
| BPF_JEQ
| BPF_K
:
328 case BPF_JMP
| BPF_JEQ
| BPF_X
:
329 case BPF_JMP
| BPF_JGE
| BPF_K
:
330 case BPF_JMP
| BPF_JGE
| BPF_X
:
331 case BPF_JMP
| BPF_JGT
| BPF_K
:
332 case BPF_JMP
| BPF_JGT
| BPF_X
:
333 case BPF_JMP
| BPF_JSET
| BPF_K
:
334 case BPF_JMP
| BPF_JSET
| BPF_X
:
343 #ifdef SECCOMP_ARCH_NATIVE
344 static inline bool seccomp_cache_check_allow_bitmap(const void *bitmap
,
348 if (unlikely(syscall_nr
< 0 || syscall_nr
>= bitmap_size
))
350 syscall_nr
= array_index_nospec(syscall_nr
, bitmap_size
);
352 return test_bit(syscall_nr
, bitmap
);
356 * seccomp_cache_check_allow - lookup seccomp cache
357 * @sfilter: The seccomp filter
358 * @sd: The seccomp data to lookup the cache with
360 * Returns true if the seccomp_data is cached and allowed.
362 static inline bool seccomp_cache_check_allow(const struct seccomp_filter
*sfilter
,
363 const struct seccomp_data
*sd
)
365 int syscall_nr
= sd
->nr
;
366 const struct action_cache
*cache
= &sfilter
->cache
;
368 #ifndef SECCOMP_ARCH_COMPAT
369 /* A native-only architecture doesn't need to check sd->arch. */
370 return seccomp_cache_check_allow_bitmap(cache
->allow_native
,
371 SECCOMP_ARCH_NATIVE_NR
,
374 if (likely(sd
->arch
== SECCOMP_ARCH_NATIVE
))
375 return seccomp_cache_check_allow_bitmap(cache
->allow_native
,
376 SECCOMP_ARCH_NATIVE_NR
,
378 if (likely(sd
->arch
== SECCOMP_ARCH_COMPAT
))
379 return seccomp_cache_check_allow_bitmap(cache
->allow_compat
,
380 SECCOMP_ARCH_COMPAT_NR
,
382 #endif /* SECCOMP_ARCH_COMPAT */
387 #endif /* SECCOMP_ARCH_NATIVE */
390 * seccomp_run_filters - evaluates all seccomp filters against @sd
391 * @sd: optional seccomp data to be passed to filters
392 * @match: stores struct seccomp_filter that resulted in the return value,
393 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
396 * Returns valid seccomp BPF response codes.
398 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
399 static u32
seccomp_run_filters(const struct seccomp_data
*sd
,
400 struct seccomp_filter
**match
)
402 u32 ret
= SECCOMP_RET_ALLOW
;
403 /* Make sure cross-thread synced filter points somewhere sane. */
404 struct seccomp_filter
*f
=
405 READ_ONCE(current
->seccomp
.filter
);
407 /* Ensure unexpected behavior doesn't result in failing open. */
408 if (WARN_ON(f
== NULL
))
409 return SECCOMP_RET_KILL_PROCESS
;
411 if (seccomp_cache_check_allow(f
, sd
))
412 return SECCOMP_RET_ALLOW
;
415 * All filters in the list are evaluated and the lowest BPF return
416 * value always takes priority (ignoring the DATA).
418 for (; f
; f
= f
->prev
) {
419 u32 cur_ret
= bpf_prog_run_pin_on_cpu(f
->prog
, sd
);
421 if (ACTION_ONLY(cur_ret
) < ACTION_ONLY(ret
)) {
428 #endif /* CONFIG_SECCOMP_FILTER */
430 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode
)
432 assert_spin_locked(¤t
->sighand
->siglock
);
434 if (current
->seccomp
.mode
&& current
->seccomp
.mode
!= seccomp_mode
)
440 void __weak
arch_seccomp_spec_mitigate(struct task_struct
*task
) { }
442 static inline void seccomp_assign_mode(struct task_struct
*task
,
443 unsigned long seccomp_mode
,
446 assert_spin_locked(&task
->sighand
->siglock
);
448 task
->seccomp
.mode
= seccomp_mode
;
450 * Make sure SYSCALL_WORK_SECCOMP cannot be set before the mode (and
453 smp_mb__before_atomic();
454 /* Assume default seccomp processes want spec flaw mitigation. */
455 if ((flags
& SECCOMP_FILTER_FLAG_SPEC_ALLOW
) == 0)
456 arch_seccomp_spec_mitigate(task
);
457 set_task_syscall_work(task
, SECCOMP
);
460 #ifdef CONFIG_SECCOMP_FILTER
461 /* Returns 1 if the parent is an ancestor of the child. */
462 static int is_ancestor(struct seccomp_filter
*parent
,
463 struct seccomp_filter
*child
)
465 /* NULL is the root ancestor. */
468 for (; child
; child
= child
->prev
)
475 * seccomp_can_sync_threads: checks if all threads can be synchronized
477 * Expects sighand and cred_guard_mutex locks to be held.
479 * Returns 0 on success, -ve on error, or the pid of a thread which was
480 * either not in the correct seccomp mode or did not have an ancestral
483 static inline pid_t
seccomp_can_sync_threads(void)
485 struct task_struct
*thread
, *caller
;
487 BUG_ON(!mutex_is_locked(¤t
->signal
->cred_guard_mutex
));
488 assert_spin_locked(¤t
->sighand
->siglock
);
490 /* Validate all threads being eligible for synchronization. */
492 for_each_thread(caller
, thread
) {
495 /* Skip current, since it is initiating the sync. */
496 if (thread
== caller
)
499 if (thread
->seccomp
.mode
== SECCOMP_MODE_DISABLED
||
500 (thread
->seccomp
.mode
== SECCOMP_MODE_FILTER
&&
501 is_ancestor(thread
->seccomp
.filter
,
502 caller
->seccomp
.filter
)))
505 /* Return the first thread that cannot be synchronized. */
506 failed
= task_pid_vnr(thread
);
507 /* If the pid cannot be resolved, then return -ESRCH */
508 if (WARN_ON(failed
== 0))
516 static inline void seccomp_filter_free(struct seccomp_filter
*filter
)
519 bpf_prog_destroy(filter
->prog
);
524 static void __seccomp_filter_orphan(struct seccomp_filter
*orig
)
526 while (orig
&& refcount_dec_and_test(&orig
->users
)) {
527 if (waitqueue_active(&orig
->wqh
))
528 wake_up_poll(&orig
->wqh
, EPOLLHUP
);
533 static void __put_seccomp_filter(struct seccomp_filter
*orig
)
535 /* Clean up single-reference branches iteratively. */
536 while (orig
&& refcount_dec_and_test(&orig
->refs
)) {
537 struct seccomp_filter
*freeme
= orig
;
539 seccomp_filter_free(freeme
);
543 static void __seccomp_filter_release(struct seccomp_filter
*orig
)
545 /* Notify about any unused filters in the task's former filter tree. */
546 __seccomp_filter_orphan(orig
);
547 /* Finally drop all references to the task's former tree. */
548 __put_seccomp_filter(orig
);
552 * seccomp_filter_release - Detach the task from its filter tree,
553 * drop its reference count, and notify
554 * about unused filters
556 * This function should only be called when the task is exiting as
557 * it detaches it from its filter tree. As such, READ_ONCE() and
558 * barriers are not needed here, as would normally be needed.
560 void seccomp_filter_release(struct task_struct
*tsk
)
562 struct seccomp_filter
*orig
= tsk
->seccomp
.filter
;
564 /* We are effectively holding the siglock by not having any sighand. */
565 WARN_ON(tsk
->sighand
!= NULL
);
567 /* Detach task from its filter tree. */
568 tsk
->seccomp
.filter
= NULL
;
569 __seccomp_filter_release(orig
);
573 * seccomp_sync_threads: sets all threads to use current's filter
575 * Expects sighand and cred_guard_mutex locks to be held, and for
576 * seccomp_can_sync_threads() to have returned success already
577 * without dropping the locks.
580 static inline void seccomp_sync_threads(unsigned long flags
)
582 struct task_struct
*thread
, *caller
;
584 BUG_ON(!mutex_is_locked(¤t
->signal
->cred_guard_mutex
));
585 assert_spin_locked(¤t
->sighand
->siglock
);
587 /* Synchronize all threads. */
589 for_each_thread(caller
, thread
) {
590 /* Skip current, since it needs no changes. */
591 if (thread
== caller
)
594 /* Get a task reference for the new leaf node. */
595 get_seccomp_filter(caller
);
598 * Drop the task reference to the shared ancestor since
599 * current's path will hold a reference. (This also
600 * allows a put before the assignment.)
602 __seccomp_filter_release(thread
->seccomp
.filter
);
604 /* Make our new filter tree visible. */
605 smp_store_release(&thread
->seccomp
.filter
,
606 caller
->seccomp
.filter
);
607 atomic_set(&thread
->seccomp
.filter_count
,
608 atomic_read(&caller
->seccomp
.filter_count
));
611 * Don't let an unprivileged task work around
612 * the no_new_privs restriction by creating
613 * a thread that sets it up, enters seccomp,
616 if (task_no_new_privs(caller
))
617 task_set_no_new_privs(thread
);
620 * Opt the other thread into seccomp if needed.
621 * As threads are considered to be trust-realm
622 * equivalent (see ptrace_may_access), it is safe to
623 * allow one thread to transition the other.
625 if (thread
->seccomp
.mode
== SECCOMP_MODE_DISABLED
)
626 seccomp_assign_mode(thread
, SECCOMP_MODE_FILTER
,
632 * seccomp_prepare_filter: Prepares a seccomp filter for use.
633 * @fprog: BPF program to install
635 * Returns filter on success or an ERR_PTR on failure.
637 static struct seccomp_filter
*seccomp_prepare_filter(struct sock_fprog
*fprog
)
639 struct seccomp_filter
*sfilter
;
641 const bool save_orig
=
642 #if defined(CONFIG_CHECKPOINT_RESTORE) || defined(SECCOMP_ARCH_NATIVE)
648 if (fprog
->len
== 0 || fprog
->len
> BPF_MAXINSNS
)
649 return ERR_PTR(-EINVAL
);
651 BUG_ON(INT_MAX
/ fprog
->len
< sizeof(struct sock_filter
));
654 * Installing a seccomp filter requires that the task has
655 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
656 * This avoids scenarios where unprivileged tasks can affect the
657 * behavior of privileged children.
659 if (!task_no_new_privs(current
) &&
660 !ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN
))
661 return ERR_PTR(-EACCES
);
663 /* Allocate a new seccomp_filter */
664 sfilter
= kzalloc(sizeof(*sfilter
), GFP_KERNEL
| __GFP_NOWARN
);
666 return ERR_PTR(-ENOMEM
);
668 mutex_init(&sfilter
->notify_lock
);
669 ret
= bpf_prog_create_from_user(&sfilter
->prog
, fprog
,
670 seccomp_check_filter
, save_orig
);
676 refcount_set(&sfilter
->refs
, 1);
677 refcount_set(&sfilter
->users
, 1);
678 init_waitqueue_head(&sfilter
->wqh
);
684 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
685 * @user_filter: pointer to the user data containing a sock_fprog.
687 * Returns 0 on success and non-zero otherwise.
689 static struct seccomp_filter
*
690 seccomp_prepare_user_filter(const char __user
*user_filter
)
692 struct sock_fprog fprog
;
693 struct seccomp_filter
*filter
= ERR_PTR(-EFAULT
);
696 if (in_compat_syscall()) {
697 struct compat_sock_fprog fprog32
;
698 if (copy_from_user(&fprog32
, user_filter
, sizeof(fprog32
)))
700 fprog
.len
= fprog32
.len
;
701 fprog
.filter
= compat_ptr(fprog32
.filter
);
702 } else /* falls through to the if below. */
704 if (copy_from_user(&fprog
, user_filter
, sizeof(fprog
)))
706 filter
= seccomp_prepare_filter(&fprog
);
711 #ifdef SECCOMP_ARCH_NATIVE
713 * seccomp_is_const_allow - check if filter is constant allow with given data
714 * @fprog: The BPF programs
715 * @sd: The seccomp data to check against, only syscall number and arch
716 * number are considered constant.
718 static bool seccomp_is_const_allow(struct sock_fprog_kern
*fprog
,
719 struct seccomp_data
*sd
)
721 unsigned int reg_value
= 0;
725 if (WARN_ON_ONCE(!fprog
))
728 for (pc
= 0; pc
< fprog
->len
; pc
++) {
729 struct sock_filter
*insn
= &fprog
->filter
[pc
];
730 u16 code
= insn
->code
;
734 case BPF_LD
| BPF_W
| BPF_ABS
:
736 case offsetof(struct seccomp_data
, nr
):
739 case offsetof(struct seccomp_data
, arch
):
740 reg_value
= sd
->arch
;
743 /* can't optimize (non-constant value load) */
747 case BPF_RET
| BPF_K
:
748 /* reached return with constant values only, check allow */
749 return k
== SECCOMP_RET_ALLOW
;
750 case BPF_JMP
| BPF_JA
:
753 case BPF_JMP
| BPF_JEQ
| BPF_K
:
754 case BPF_JMP
| BPF_JGE
| BPF_K
:
755 case BPF_JMP
| BPF_JGT
| BPF_K
:
756 case BPF_JMP
| BPF_JSET
| BPF_K
:
757 switch (BPF_OP(code
)) {
759 op_res
= reg_value
== k
;
762 op_res
= reg_value
>= k
;
765 op_res
= reg_value
> k
;
768 op_res
= !!(reg_value
& k
);
771 /* can't optimize (unknown jump) */
775 pc
+= op_res
? insn
->jt
: insn
->jf
;
777 case BPF_ALU
| BPF_AND
| BPF_K
:
781 /* can't optimize (unknown insn) */
786 /* ran off the end of the filter?! */
791 static void seccomp_cache_prepare_bitmap(struct seccomp_filter
*sfilter
,
792 void *bitmap
, const void *bitmap_prev
,
793 size_t bitmap_size
, int arch
)
795 struct sock_fprog_kern
*fprog
= sfilter
->prog
->orig_prog
;
796 struct seccomp_data sd
;
800 /* The new filter must be as restrictive as the last. */
801 bitmap_copy(bitmap
, bitmap_prev
, bitmap_size
);
803 /* Before any filters, all syscalls are always allowed. */
804 bitmap_fill(bitmap
, bitmap_size
);
807 for (nr
= 0; nr
< bitmap_size
; nr
++) {
808 /* No bitmap change: not a cacheable action. */
809 if (!test_bit(nr
, bitmap
))
815 /* No bitmap change: continue to always allow. */
816 if (seccomp_is_const_allow(fprog
, &sd
))
820 * Not a cacheable action: always run filters.
821 * atomic clear_bit() not needed, filter not visible yet.
823 __clear_bit(nr
, bitmap
);
828 * seccomp_cache_prepare - emulate the filter to find cacheable syscalls
829 * @sfilter: The seccomp filter
831 * Returns 0 if successful or -errno if error occurred.
833 static void seccomp_cache_prepare(struct seccomp_filter
*sfilter
)
835 struct action_cache
*cache
= &sfilter
->cache
;
836 const struct action_cache
*cache_prev
=
837 sfilter
->prev
? &sfilter
->prev
->cache
: NULL
;
839 seccomp_cache_prepare_bitmap(sfilter
, cache
->allow_native
,
840 cache_prev
? cache_prev
->allow_native
: NULL
,
841 SECCOMP_ARCH_NATIVE_NR
,
842 SECCOMP_ARCH_NATIVE
);
844 #ifdef SECCOMP_ARCH_COMPAT
845 seccomp_cache_prepare_bitmap(sfilter
, cache
->allow_compat
,
846 cache_prev
? cache_prev
->allow_compat
: NULL
,
847 SECCOMP_ARCH_COMPAT_NR
,
848 SECCOMP_ARCH_COMPAT
);
849 #endif /* SECCOMP_ARCH_COMPAT */
851 #endif /* SECCOMP_ARCH_NATIVE */
854 * seccomp_attach_filter: validate and attach filter
855 * @flags: flags to change filter behavior
856 * @filter: seccomp filter to add to the current process
858 * Caller must be holding current->sighand->siglock lock.
860 * Returns 0 on success, -ve on error, or
861 * - in TSYNC mode: the pid of a thread which was either not in the correct
862 * seccomp mode or did not have an ancestral seccomp filter
863 * - in NEW_LISTENER mode: the fd of the new listener
865 static long seccomp_attach_filter(unsigned int flags
,
866 struct seccomp_filter
*filter
)
868 unsigned long total_insns
;
869 struct seccomp_filter
*walker
;
871 assert_spin_locked(¤t
->sighand
->siglock
);
873 /* Validate resulting filter length. */
874 total_insns
= filter
->prog
->len
;
875 for (walker
= current
->seccomp
.filter
; walker
; walker
= walker
->prev
)
876 total_insns
+= walker
->prog
->len
+ 4; /* 4 instr penalty */
877 if (total_insns
> MAX_INSNS_PER_PATH
)
880 /* If thread sync has been requested, check that it is possible. */
881 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
) {
884 ret
= seccomp_can_sync_threads();
886 if (flags
& SECCOMP_FILTER_FLAG_TSYNC_ESRCH
)
893 /* Set log flag, if present. */
894 if (flags
& SECCOMP_FILTER_FLAG_LOG
)
898 * If there is an existing filter, make it the prev and don't drop its
901 filter
->prev
= current
->seccomp
.filter
;
902 seccomp_cache_prepare(filter
);
903 current
->seccomp
.filter
= filter
;
904 atomic_inc(¤t
->seccomp
.filter_count
);
906 /* Now that the new filter is in place, synchronize to all threads. */
907 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
)
908 seccomp_sync_threads(flags
);
913 static void __get_seccomp_filter(struct seccomp_filter
*filter
)
915 refcount_inc(&filter
->refs
);
918 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
919 void get_seccomp_filter(struct task_struct
*tsk
)
921 struct seccomp_filter
*orig
= tsk
->seccomp
.filter
;
924 __get_seccomp_filter(orig
);
925 refcount_inc(&orig
->users
);
928 #endif /* CONFIG_SECCOMP_FILTER */
930 /* For use with seccomp_actions_logged */
931 #define SECCOMP_LOG_KILL_PROCESS (1 << 0)
932 #define SECCOMP_LOG_KILL_THREAD (1 << 1)
933 #define SECCOMP_LOG_TRAP (1 << 2)
934 #define SECCOMP_LOG_ERRNO (1 << 3)
935 #define SECCOMP_LOG_TRACE (1 << 4)
936 #define SECCOMP_LOG_LOG (1 << 5)
937 #define SECCOMP_LOG_ALLOW (1 << 6)
938 #define SECCOMP_LOG_USER_NOTIF (1 << 7)
940 static u32 seccomp_actions_logged
= SECCOMP_LOG_KILL_PROCESS
|
941 SECCOMP_LOG_KILL_THREAD
|
944 SECCOMP_LOG_USER_NOTIF
|
948 static inline void seccomp_log(unsigned long syscall
, long signr
, u32 action
,
954 case SECCOMP_RET_ALLOW
:
956 case SECCOMP_RET_TRAP
:
957 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_TRAP
;
959 case SECCOMP_RET_ERRNO
:
960 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_ERRNO
;
962 case SECCOMP_RET_TRACE
:
963 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_TRACE
;
965 case SECCOMP_RET_USER_NOTIF
:
966 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_USER_NOTIF
;
968 case SECCOMP_RET_LOG
:
969 log
= seccomp_actions_logged
& SECCOMP_LOG_LOG
;
971 case SECCOMP_RET_KILL_THREAD
:
972 log
= seccomp_actions_logged
& SECCOMP_LOG_KILL_THREAD
;
974 case SECCOMP_RET_KILL_PROCESS
:
976 log
= seccomp_actions_logged
& SECCOMP_LOG_KILL_PROCESS
;
980 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
981 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
982 * any action from being logged by removing the action name from the
983 * seccomp_actions_logged sysctl.
988 audit_seccomp(syscall
, signr
, action
);
992 * Secure computing mode 1 allows only read/write/exit/sigreturn.
993 * To be fully secure this must be combined with rlimit
994 * to limit the stack allocations too.
996 static const int mode1_syscalls
[] = {
997 __NR_seccomp_read
, __NR_seccomp_write
, __NR_seccomp_exit
, __NR_seccomp_sigreturn
,
998 -1, /* negative terminated */
1001 static void __secure_computing_strict(int this_syscall
)
1003 const int *allowed_syscalls
= mode1_syscalls
;
1004 #ifdef CONFIG_COMPAT
1005 if (in_compat_syscall())
1006 allowed_syscalls
= get_compat_mode1_syscalls();
1009 if (*allowed_syscalls
== this_syscall
)
1011 } while (*++allowed_syscalls
!= -1);
1013 #ifdef SECCOMP_DEBUG
1016 current
->seccomp
.mode
= SECCOMP_MODE_DEAD
;
1017 seccomp_log(this_syscall
, SIGKILL
, SECCOMP_RET_KILL_THREAD
, true);
1021 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
1022 void secure_computing_strict(int this_syscall
)
1024 int mode
= current
->seccomp
.mode
;
1026 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE
) &&
1027 unlikely(current
->ptrace
& PT_SUSPEND_SECCOMP
))
1030 if (mode
== SECCOMP_MODE_DISABLED
)
1032 else if (mode
== SECCOMP_MODE_STRICT
)
1033 __secure_computing_strict(this_syscall
);
1039 #ifdef CONFIG_SECCOMP_FILTER
1040 static u64
seccomp_next_notify_id(struct seccomp_filter
*filter
)
1043 * Note: overflow is ok here, the id just needs to be unique per
1046 lockdep_assert_held(&filter
->notify_lock
);
1047 return filter
->notif
->next_id
++;
1050 static void seccomp_handle_addfd(struct seccomp_kaddfd
*addfd
, struct seccomp_knotif
*n
)
1055 * Remove the notification, and reset the list pointers, indicating
1056 * that it has been handled.
1058 list_del_init(&addfd
->list
);
1060 fd
= receive_fd(addfd
->file
, addfd
->flags
);
1062 fd
= receive_fd_replace(addfd
->fd
, addfd
->file
, addfd
->flags
);
1065 if (addfd
->ioctl_flags
& SECCOMP_ADDFD_FLAG_SEND
) {
1066 /* If we fail reset and return an error to the notifier */
1068 n
->state
= SECCOMP_NOTIFY_SENT
;
1070 /* Return the FD we just added */
1078 * Mark the notification as completed. From this point, addfd mem
1079 * might be invalidated and we can't safely read it anymore.
1081 complete(&addfd
->completion
);
1084 static int seccomp_do_user_notification(int this_syscall
,
1085 struct seccomp_filter
*match
,
1086 const struct seccomp_data
*sd
)
1091 struct seccomp_knotif n
= {};
1092 struct seccomp_kaddfd
*addfd
, *tmp
;
1094 mutex_lock(&match
->notify_lock
);
1100 n
.state
= SECCOMP_NOTIFY_INIT
;
1102 n
.id
= seccomp_next_notify_id(match
);
1103 init_completion(&n
.ready
);
1104 list_add(&n
.list
, &match
->notif
->notifications
);
1105 INIT_LIST_HEAD(&n
.addfd
);
1107 up(&match
->notif
->request
);
1108 wake_up_poll(&match
->wqh
, EPOLLIN
| EPOLLRDNORM
);
1111 * This is where we wait for a reply from userspace.
1114 mutex_unlock(&match
->notify_lock
);
1115 err
= wait_for_completion_interruptible(&n
.ready
);
1116 mutex_lock(&match
->notify_lock
);
1120 addfd
= list_first_entry_or_null(&n
.addfd
,
1121 struct seccomp_kaddfd
, list
);
1122 /* Check if we were woken up by a addfd message */
1124 seccomp_handle_addfd(addfd
, &n
);
1126 } while (n
.state
!= SECCOMP_NOTIFY_REPLIED
);
1133 /* If there were any pending addfd calls, clear them out */
1134 list_for_each_entry_safe(addfd
, tmp
, &n
.addfd
, list
) {
1135 /* The process went away before we got a chance to handle it */
1136 addfd
->ret
= -ESRCH
;
1137 list_del_init(&addfd
->list
);
1138 complete(&addfd
->completion
);
1142 * Note that it's possible the listener died in between the time when
1143 * we were notified of a response (or a signal) and when we were able to
1144 * re-acquire the lock, so only delete from the list if the
1145 * notification actually exists.
1147 * Also note that this test is only valid because there's no way to
1148 * *reattach* to a notifier right now. If one is added, we'll need to
1149 * keep track of the notif itself and make sure they match here.
1154 mutex_unlock(&match
->notify_lock
);
1156 /* Userspace requests to continue the syscall. */
1157 if (flags
& SECCOMP_USER_NOTIF_FLAG_CONTINUE
)
1160 syscall_set_return_value(current
, current_pt_regs(),
1165 static int __seccomp_filter(int this_syscall
, const struct seccomp_data
*sd
,
1166 const bool recheck_after_trace
)
1168 u32 filter_ret
, action
;
1169 struct seccomp_filter
*match
= NULL
;
1171 struct seccomp_data sd_local
;
1174 * Make sure that any changes to mode from another thread have
1175 * been seen after SYSCALL_WORK_SECCOMP was seen.
1180 populate_seccomp_data(&sd_local
);
1184 filter_ret
= seccomp_run_filters(sd
, &match
);
1185 data
= filter_ret
& SECCOMP_RET_DATA
;
1186 action
= filter_ret
& SECCOMP_RET_ACTION_FULL
;
1189 case SECCOMP_RET_ERRNO
:
1190 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
1191 if (data
> MAX_ERRNO
)
1193 syscall_set_return_value(current
, current_pt_regs(),
1197 case SECCOMP_RET_TRAP
:
1198 /* Show the handler the original registers. */
1199 syscall_rollback(current
, current_pt_regs());
1200 /* Let the filter pass back 16 bits of data. */
1201 force_sig_seccomp(this_syscall
, data
, false);
1204 case SECCOMP_RET_TRACE
:
1205 /* We've been put in this state by the ptracer already. */
1206 if (recheck_after_trace
)
1209 /* ENOSYS these calls if there is no tracer attached. */
1210 if (!ptrace_event_enabled(current
, PTRACE_EVENT_SECCOMP
)) {
1211 syscall_set_return_value(current
,
1217 /* Allow the BPF to provide the event message */
1218 ptrace_event(PTRACE_EVENT_SECCOMP
, data
);
1220 * The delivery of a fatal signal during event
1221 * notification may silently skip tracer notification,
1222 * which could leave us with a potentially unmodified
1223 * syscall that the tracer would have liked to have
1224 * changed. Since the process is about to die, we just
1225 * force the syscall to be skipped and let the signal
1226 * kill the process and correctly handle any tracer exit
1229 if (fatal_signal_pending(current
))
1231 /* Check if the tracer forced the syscall to be skipped. */
1232 this_syscall
= syscall_get_nr(current
, current_pt_regs());
1233 if (this_syscall
< 0)
1237 * Recheck the syscall, since it may have changed. This
1238 * intentionally uses a NULL struct seccomp_data to force
1239 * a reload of all registers. This does not goto skip since
1240 * a skip would have already been reported.
1242 if (__seccomp_filter(this_syscall
, NULL
, true))
1247 case SECCOMP_RET_USER_NOTIF
:
1248 if (seccomp_do_user_notification(this_syscall
, match
, sd
))
1253 case SECCOMP_RET_LOG
:
1254 seccomp_log(this_syscall
, 0, action
, true);
1257 case SECCOMP_RET_ALLOW
:
1259 * Note that the "match" filter will always be NULL for
1260 * this action since SECCOMP_RET_ALLOW is the starting
1261 * state in seccomp_run_filters().
1265 case SECCOMP_RET_KILL_THREAD
:
1266 case SECCOMP_RET_KILL_PROCESS
:
1268 current
->seccomp
.mode
= SECCOMP_MODE_DEAD
;
1269 seccomp_log(this_syscall
, SIGSYS
, action
, true);
1270 /* Dump core only if this is the last remaining thread. */
1271 if (action
!= SECCOMP_RET_KILL_THREAD
||
1272 (atomic_read(¤t
->signal
->live
) == 1)) {
1273 /* Show the original registers in the dump. */
1274 syscall_rollback(current
, current_pt_regs());
1275 /* Trigger a coredump with SIGSYS */
1276 force_sig_seccomp(this_syscall
, data
, true);
1280 return -1; /* skip the syscall go directly to signal handling */
1286 seccomp_log(this_syscall
, 0, action
, match
? match
->log
: false);
1290 static int __seccomp_filter(int this_syscall
, const struct seccomp_data
*sd
,
1291 const bool recheck_after_trace
)
1299 int __secure_computing(const struct seccomp_data
*sd
)
1301 int mode
= current
->seccomp
.mode
;
1304 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE
) &&
1305 unlikely(current
->ptrace
& PT_SUSPEND_SECCOMP
))
1308 this_syscall
= sd
? sd
->nr
:
1309 syscall_get_nr(current
, current_pt_regs());
1312 case SECCOMP_MODE_STRICT
:
1313 __secure_computing_strict(this_syscall
); /* may call do_exit */
1315 case SECCOMP_MODE_FILTER
:
1316 return __seccomp_filter(this_syscall
, sd
, false);
1317 /* Surviving SECCOMP_RET_KILL_* must be proactively impossible. */
1318 case SECCOMP_MODE_DEAD
:
1326 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1328 long prctl_get_seccomp(void)
1330 return current
->seccomp
.mode
;
1334 * seccomp_set_mode_strict: internal function for setting strict seccomp
1336 * Once current->seccomp.mode is non-zero, it may not be changed.
1338 * Returns 0 on success or -EINVAL on failure.
1340 static long seccomp_set_mode_strict(void)
1342 const unsigned long seccomp_mode
= SECCOMP_MODE_STRICT
;
1345 spin_lock_irq(¤t
->sighand
->siglock
);
1347 if (!seccomp_may_assign_mode(seccomp_mode
))
1353 seccomp_assign_mode(current
, seccomp_mode
, 0);
1357 spin_unlock_irq(¤t
->sighand
->siglock
);
1362 #ifdef CONFIG_SECCOMP_FILTER
1363 static void seccomp_notify_free(struct seccomp_filter
*filter
)
1365 kfree(filter
->notif
);
1366 filter
->notif
= NULL
;
1369 static void seccomp_notify_detach(struct seccomp_filter
*filter
)
1371 struct seccomp_knotif
*knotif
;
1376 mutex_lock(&filter
->notify_lock
);
1379 * If this file is being closed because e.g. the task who owned it
1380 * died, let's wake everyone up who was waiting on us.
1382 list_for_each_entry(knotif
, &filter
->notif
->notifications
, list
) {
1383 if (knotif
->state
== SECCOMP_NOTIFY_REPLIED
)
1386 knotif
->state
= SECCOMP_NOTIFY_REPLIED
;
1387 knotif
->error
= -ENOSYS
;
1391 * We do not need to wake up any pending addfd messages, as
1392 * the notifier will do that for us, as this just looks
1393 * like a standard reply.
1395 complete(&knotif
->ready
);
1398 seccomp_notify_free(filter
);
1399 mutex_unlock(&filter
->notify_lock
);
1402 static int seccomp_notify_release(struct inode
*inode
, struct file
*file
)
1404 struct seccomp_filter
*filter
= file
->private_data
;
1406 seccomp_notify_detach(filter
);
1407 __put_seccomp_filter(filter
);
1411 /* must be called with notif_lock held */
1412 static inline struct seccomp_knotif
*
1413 find_notification(struct seccomp_filter
*filter
, u64 id
)
1415 struct seccomp_knotif
*cur
;
1417 lockdep_assert_held(&filter
->notify_lock
);
1419 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1428 static long seccomp_notify_recv(struct seccomp_filter
*filter
,
1431 struct seccomp_knotif
*knotif
= NULL
, *cur
;
1432 struct seccomp_notif unotif
;
1435 /* Verify that we're not given garbage to keep struct extensible. */
1436 ret
= check_zeroed_user(buf
, sizeof(unotif
));
1442 memset(&unotif
, 0, sizeof(unotif
));
1444 ret
= down_interruptible(&filter
->notif
->request
);
1448 mutex_lock(&filter
->notify_lock
);
1449 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1450 if (cur
->state
== SECCOMP_NOTIFY_INIT
) {
1457 * If we didn't find a notification, it could be that the task was
1458 * interrupted by a fatal signal between the time we were woken and
1459 * when we were able to acquire the rw lock.
1466 unotif
.id
= knotif
->id
;
1467 unotif
.pid
= task_pid_vnr(knotif
->task
);
1468 unotif
.data
= *(knotif
->data
);
1470 knotif
->state
= SECCOMP_NOTIFY_SENT
;
1471 wake_up_poll(&filter
->wqh
, EPOLLOUT
| EPOLLWRNORM
);
1474 mutex_unlock(&filter
->notify_lock
);
1476 if (ret
== 0 && copy_to_user(buf
, &unotif
, sizeof(unotif
))) {
1480 * Userspace screwed up. To make sure that we keep this
1481 * notification alive, let's reset it back to INIT. It
1482 * may have died when we released the lock, so we need to make
1483 * sure it's still around.
1485 mutex_lock(&filter
->notify_lock
);
1486 knotif
= find_notification(filter
, unotif
.id
);
1488 knotif
->state
= SECCOMP_NOTIFY_INIT
;
1489 up(&filter
->notif
->request
);
1491 mutex_unlock(&filter
->notify_lock
);
1497 static long seccomp_notify_send(struct seccomp_filter
*filter
,
1500 struct seccomp_notif_resp resp
= {};
1501 struct seccomp_knotif
*knotif
;
1504 if (copy_from_user(&resp
, buf
, sizeof(resp
)))
1507 if (resp
.flags
& ~SECCOMP_USER_NOTIF_FLAG_CONTINUE
)
1510 if ((resp
.flags
& SECCOMP_USER_NOTIF_FLAG_CONTINUE
) &&
1511 (resp
.error
|| resp
.val
))
1514 ret
= mutex_lock_interruptible(&filter
->notify_lock
);
1518 knotif
= find_notification(filter
, resp
.id
);
1524 /* Allow exactly one reply. */
1525 if (knotif
->state
!= SECCOMP_NOTIFY_SENT
) {
1531 knotif
->state
= SECCOMP_NOTIFY_REPLIED
;
1532 knotif
->error
= resp
.error
;
1533 knotif
->val
= resp
.val
;
1534 knotif
->flags
= resp
.flags
;
1535 complete(&knotif
->ready
);
1537 mutex_unlock(&filter
->notify_lock
);
1541 static long seccomp_notify_id_valid(struct seccomp_filter
*filter
,
1544 struct seccomp_knotif
*knotif
;
1548 if (copy_from_user(&id
, buf
, sizeof(id
)))
1551 ret
= mutex_lock_interruptible(&filter
->notify_lock
);
1555 knotif
= find_notification(filter
, id
);
1556 if (knotif
&& knotif
->state
== SECCOMP_NOTIFY_SENT
)
1561 mutex_unlock(&filter
->notify_lock
);
1565 static long seccomp_notify_addfd(struct seccomp_filter
*filter
,
1566 struct seccomp_notif_addfd __user
*uaddfd
,
1569 struct seccomp_notif_addfd addfd
;
1570 struct seccomp_knotif
*knotif
;
1571 struct seccomp_kaddfd kaddfd
;
1574 BUILD_BUG_ON(sizeof(addfd
) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0
);
1575 BUILD_BUG_ON(sizeof(addfd
) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST
);
1577 if (size
< SECCOMP_NOTIFY_ADDFD_SIZE_VER0
|| size
>= PAGE_SIZE
)
1580 ret
= copy_struct_from_user(&addfd
, sizeof(addfd
), uaddfd
, size
);
1584 if (addfd
.newfd_flags
& ~O_CLOEXEC
)
1587 if (addfd
.flags
& ~(SECCOMP_ADDFD_FLAG_SETFD
| SECCOMP_ADDFD_FLAG_SEND
))
1590 if (addfd
.newfd
&& !(addfd
.flags
& SECCOMP_ADDFD_FLAG_SETFD
))
1593 kaddfd
.file
= fget(addfd
.srcfd
);
1597 kaddfd
.ioctl_flags
= addfd
.flags
;
1598 kaddfd
.flags
= addfd
.newfd_flags
;
1599 kaddfd
.setfd
= addfd
.flags
& SECCOMP_ADDFD_FLAG_SETFD
;
1600 kaddfd
.fd
= addfd
.newfd
;
1601 init_completion(&kaddfd
.completion
);
1603 ret
= mutex_lock_interruptible(&filter
->notify_lock
);
1607 knotif
= find_notification(filter
, addfd
.id
);
1614 * We do not want to allow for FD injection to occur before the
1615 * notification has been picked up by a userspace handler, or after
1616 * the notification has been replied to.
1618 if (knotif
->state
!= SECCOMP_NOTIFY_SENT
) {
1623 if (addfd
.flags
& SECCOMP_ADDFD_FLAG_SEND
) {
1625 * Disallow queuing an atomic addfd + send reply while there are
1626 * some addfd requests still to process.
1628 * There is no clear reason to support it and allows us to keep
1629 * the loop on the other side straight-forward.
1631 if (!list_empty(&knotif
->addfd
)) {
1636 /* Allow exactly only one reply */
1637 knotif
->state
= SECCOMP_NOTIFY_REPLIED
;
1640 list_add(&kaddfd
.list
, &knotif
->addfd
);
1641 complete(&knotif
->ready
);
1642 mutex_unlock(&filter
->notify_lock
);
1644 /* Now we wait for it to be processed or be interrupted */
1645 ret
= wait_for_completion_interruptible(&kaddfd
.completion
);
1648 * We had a successful completion. The other side has already
1649 * removed us from the addfd queue, and
1650 * wait_for_completion_interruptible has a memory barrier upon
1651 * success that lets us read this value directly without
1658 mutex_lock(&filter
->notify_lock
);
1660 * Even though we were woken up by a signal and not a successful
1661 * completion, a completion may have happened in the mean time.
1663 * We need to check again if the addfd request has been handled,
1664 * and if not, we will remove it from the queue.
1666 if (list_empty(&kaddfd
.list
))
1669 list_del(&kaddfd
.list
);
1672 mutex_unlock(&filter
->notify_lock
);
1679 static long seccomp_notify_ioctl(struct file
*file
, unsigned int cmd
,
1682 struct seccomp_filter
*filter
= file
->private_data
;
1683 void __user
*buf
= (void __user
*)arg
;
1685 /* Fixed-size ioctls */
1687 case SECCOMP_IOCTL_NOTIF_RECV
:
1688 return seccomp_notify_recv(filter
, buf
);
1689 case SECCOMP_IOCTL_NOTIF_SEND
:
1690 return seccomp_notify_send(filter
, buf
);
1691 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR
:
1692 case SECCOMP_IOCTL_NOTIF_ID_VALID
:
1693 return seccomp_notify_id_valid(filter
, buf
);
1696 /* Extensible Argument ioctls */
1697 #define EA_IOCTL(cmd) ((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1698 switch (EA_IOCTL(cmd
)) {
1699 case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD
):
1700 return seccomp_notify_addfd(filter
, buf
, _IOC_SIZE(cmd
));
1706 static __poll_t
seccomp_notify_poll(struct file
*file
,
1707 struct poll_table_struct
*poll_tab
)
1709 struct seccomp_filter
*filter
= file
->private_data
;
1711 struct seccomp_knotif
*cur
;
1713 poll_wait(file
, &filter
->wqh
, poll_tab
);
1715 if (mutex_lock_interruptible(&filter
->notify_lock
) < 0)
1718 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1719 if (cur
->state
== SECCOMP_NOTIFY_INIT
)
1720 ret
|= EPOLLIN
| EPOLLRDNORM
;
1721 if (cur
->state
== SECCOMP_NOTIFY_SENT
)
1722 ret
|= EPOLLOUT
| EPOLLWRNORM
;
1723 if ((ret
& EPOLLIN
) && (ret
& EPOLLOUT
))
1727 mutex_unlock(&filter
->notify_lock
);
1729 if (refcount_read(&filter
->users
) == 0)
1735 static const struct file_operations seccomp_notify_ops
= {
1736 .poll
= seccomp_notify_poll
,
1737 .release
= seccomp_notify_release
,
1738 .unlocked_ioctl
= seccomp_notify_ioctl
,
1739 .compat_ioctl
= seccomp_notify_ioctl
,
1742 static struct file
*init_listener(struct seccomp_filter
*filter
)
1746 ret
= ERR_PTR(-ENOMEM
);
1747 filter
->notif
= kzalloc(sizeof(*(filter
->notif
)), GFP_KERNEL
);
1751 sema_init(&filter
->notif
->request
, 0);
1752 filter
->notif
->next_id
= get_random_u64();
1753 INIT_LIST_HEAD(&filter
->notif
->notifications
);
1755 ret
= anon_inode_getfile("seccomp notify", &seccomp_notify_ops
,
1760 /* The file has a reference to it now */
1761 __get_seccomp_filter(filter
);
1765 seccomp_notify_free(filter
);
1771 * Does @new_child have a listener while an ancestor also has a listener?
1772 * If so, we'll want to reject this filter.
1773 * This only has to be tested for the current process, even in the TSYNC case,
1774 * because TSYNC installs @child with the same parent on all threads.
1775 * Note that @new_child is not hooked up to its parent at this point yet, so
1776 * we use current->seccomp.filter.
1778 static bool has_duplicate_listener(struct seccomp_filter
*new_child
)
1780 struct seccomp_filter
*cur
;
1782 /* must be protected against concurrent TSYNC */
1783 lockdep_assert_held(¤t
->sighand
->siglock
);
1785 if (!new_child
->notif
)
1787 for (cur
= current
->seccomp
.filter
; cur
; cur
= cur
->prev
) {
1796 * seccomp_set_mode_filter: internal function for setting seccomp filter
1797 * @flags: flags to change filter behavior
1798 * @filter: struct sock_fprog containing filter
1800 * This function may be called repeatedly to install additional filters.
1801 * Every filter successfully installed will be evaluated (in reverse order)
1802 * for each system call the task makes.
1804 * Once current->seccomp.mode is non-zero, it may not be changed.
1806 * Returns 0 on success or -EINVAL on failure.
1808 static long seccomp_set_mode_filter(unsigned int flags
,
1809 const char __user
*filter
)
1811 const unsigned long seccomp_mode
= SECCOMP_MODE_FILTER
;
1812 struct seccomp_filter
*prepared
= NULL
;
1815 struct file
*listener_f
= NULL
;
1817 /* Validate flags. */
1818 if (flags
& ~SECCOMP_FILTER_FLAG_MASK
)
1822 * In the successful case, NEW_LISTENER returns the new listener fd.
1823 * But in the failure case, TSYNC returns the thread that died. If you
1824 * combine these two flags, there's no way to tell whether something
1825 * succeeded or failed. So, let's disallow this combination if the user
1826 * has not explicitly requested no errors from TSYNC.
1828 if ((flags
& SECCOMP_FILTER_FLAG_TSYNC
) &&
1829 (flags
& SECCOMP_FILTER_FLAG_NEW_LISTENER
) &&
1830 ((flags
& SECCOMP_FILTER_FLAG_TSYNC_ESRCH
) == 0))
1833 /* Prepare the new filter before holding any locks. */
1834 prepared
= seccomp_prepare_user_filter(filter
);
1835 if (IS_ERR(prepared
))
1836 return PTR_ERR(prepared
);
1838 if (flags
& SECCOMP_FILTER_FLAG_NEW_LISTENER
) {
1839 listener
= get_unused_fd_flags(O_CLOEXEC
);
1845 listener_f
= init_listener(prepared
);
1846 if (IS_ERR(listener_f
)) {
1847 put_unused_fd(listener
);
1848 ret
= PTR_ERR(listener_f
);
1854 * Make sure we cannot change seccomp or nnp state via TSYNC
1855 * while another thread is in the middle of calling exec.
1857 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
&&
1858 mutex_lock_killable(¤t
->signal
->cred_guard_mutex
))
1861 spin_lock_irq(¤t
->sighand
->siglock
);
1863 if (!seccomp_may_assign_mode(seccomp_mode
))
1866 if (has_duplicate_listener(prepared
)) {
1871 ret
= seccomp_attach_filter(flags
, prepared
);
1874 /* Do not free the successfully attached filter. */
1877 seccomp_assign_mode(current
, seccomp_mode
, flags
);
1879 spin_unlock_irq(¤t
->sighand
->siglock
);
1880 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
)
1881 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1883 if (flags
& SECCOMP_FILTER_FLAG_NEW_LISTENER
) {
1885 listener_f
->private_data
= NULL
;
1887 put_unused_fd(listener
);
1888 seccomp_notify_detach(prepared
);
1890 fd_install(listener
, listener_f
);
1895 seccomp_filter_free(prepared
);
1899 static inline long seccomp_set_mode_filter(unsigned int flags
,
1900 const char __user
*filter
)
1906 static long seccomp_get_action_avail(const char __user
*uaction
)
1910 if (copy_from_user(&action
, uaction
, sizeof(action
)))
1914 case SECCOMP_RET_KILL_PROCESS
:
1915 case SECCOMP_RET_KILL_THREAD
:
1916 case SECCOMP_RET_TRAP
:
1917 case SECCOMP_RET_ERRNO
:
1918 case SECCOMP_RET_USER_NOTIF
:
1919 case SECCOMP_RET_TRACE
:
1920 case SECCOMP_RET_LOG
:
1921 case SECCOMP_RET_ALLOW
:
1930 static long seccomp_get_notif_sizes(void __user
*usizes
)
1932 struct seccomp_notif_sizes sizes
= {
1933 .seccomp_notif
= sizeof(struct seccomp_notif
),
1934 .seccomp_notif_resp
= sizeof(struct seccomp_notif_resp
),
1935 .seccomp_data
= sizeof(struct seccomp_data
),
1938 if (copy_to_user(usizes
, &sizes
, sizeof(sizes
)))
1944 /* Common entry point for both prctl and syscall. */
1945 static long do_seccomp(unsigned int op
, unsigned int flags
,
1949 case SECCOMP_SET_MODE_STRICT
:
1950 if (flags
!= 0 || uargs
!= NULL
)
1952 return seccomp_set_mode_strict();
1953 case SECCOMP_SET_MODE_FILTER
:
1954 return seccomp_set_mode_filter(flags
, uargs
);
1955 case SECCOMP_GET_ACTION_AVAIL
:
1959 return seccomp_get_action_avail(uargs
);
1960 case SECCOMP_GET_NOTIF_SIZES
:
1964 return seccomp_get_notif_sizes(uargs
);
1970 SYSCALL_DEFINE3(seccomp
, unsigned int, op
, unsigned int, flags
,
1971 void __user
*, uargs
)
1973 return do_seccomp(op
, flags
, uargs
);
1977 * prctl_set_seccomp: configures current->seccomp.mode
1978 * @seccomp_mode: requested mode to use
1979 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1981 * Returns 0 on success or -EINVAL on failure.
1983 long prctl_set_seccomp(unsigned long seccomp_mode
, void __user
*filter
)
1988 switch (seccomp_mode
) {
1989 case SECCOMP_MODE_STRICT
:
1990 op
= SECCOMP_SET_MODE_STRICT
;
1992 * Setting strict mode through prctl always ignored filter,
1993 * so make sure it is always NULL here to pass the internal
1994 * check in do_seccomp().
1998 case SECCOMP_MODE_FILTER
:
1999 op
= SECCOMP_SET_MODE_FILTER
;
2006 /* prctl interface doesn't have flags, so they are always zero. */
2007 return do_seccomp(op
, 0, uargs
);
2010 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
2011 static struct seccomp_filter
*get_nth_filter(struct task_struct
*task
,
2012 unsigned long filter_off
)
2014 struct seccomp_filter
*orig
, *filter
;
2015 unsigned long count
;
2018 * Note: this is only correct because the caller should be the (ptrace)
2019 * tracer of the task, otherwise lock_task_sighand is needed.
2021 spin_lock_irq(&task
->sighand
->siglock
);
2023 if (task
->seccomp
.mode
!= SECCOMP_MODE_FILTER
) {
2024 spin_unlock_irq(&task
->sighand
->siglock
);
2025 return ERR_PTR(-EINVAL
);
2028 orig
= task
->seccomp
.filter
;
2029 __get_seccomp_filter(orig
);
2030 spin_unlock_irq(&task
->sighand
->siglock
);
2033 for (filter
= orig
; filter
; filter
= filter
->prev
)
2036 if (filter_off
>= count
) {
2037 filter
= ERR_PTR(-ENOENT
);
2041 count
-= filter_off
;
2042 for (filter
= orig
; filter
&& count
> 1; filter
= filter
->prev
)
2045 if (WARN_ON(count
!= 1 || !filter
)) {
2046 filter
= ERR_PTR(-ENOENT
);
2050 __get_seccomp_filter(filter
);
2053 __put_seccomp_filter(orig
);
2057 long seccomp_get_filter(struct task_struct
*task
, unsigned long filter_off
,
2060 struct seccomp_filter
*filter
;
2061 struct sock_fprog_kern
*fprog
;
2064 if (!capable(CAP_SYS_ADMIN
) ||
2065 current
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
) {
2069 filter
= get_nth_filter(task
, filter_off
);
2071 return PTR_ERR(filter
);
2073 fprog
= filter
->prog
->orig_prog
;
2075 /* This must be a new non-cBPF filter, since we save
2076 * every cBPF filter's orig_prog above when
2077 * CONFIG_CHECKPOINT_RESTORE is enabled.
2087 if (copy_to_user(data
, fprog
->filter
, bpf_classic_proglen(fprog
)))
2091 __put_seccomp_filter(filter
);
2095 long seccomp_get_metadata(struct task_struct
*task
,
2096 unsigned long size
, void __user
*data
)
2099 struct seccomp_filter
*filter
;
2100 struct seccomp_metadata kmd
= {};
2102 if (!capable(CAP_SYS_ADMIN
) ||
2103 current
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
) {
2107 size
= min_t(unsigned long, size
, sizeof(kmd
));
2109 if (size
< sizeof(kmd
.filter_off
))
2112 if (copy_from_user(&kmd
.filter_off
, data
, sizeof(kmd
.filter_off
)))
2115 filter
= get_nth_filter(task
, kmd
.filter_off
);
2117 return PTR_ERR(filter
);
2120 kmd
.flags
|= SECCOMP_FILTER_FLAG_LOG
;
2123 if (copy_to_user(data
, &kmd
, size
))
2126 __put_seccomp_filter(filter
);
2131 #ifdef CONFIG_SYSCTL
2133 /* Human readable action names for friendly sysctl interaction */
2134 #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
2135 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
2136 #define SECCOMP_RET_TRAP_NAME "trap"
2137 #define SECCOMP_RET_ERRNO_NAME "errno"
2138 #define SECCOMP_RET_USER_NOTIF_NAME "user_notif"
2139 #define SECCOMP_RET_TRACE_NAME "trace"
2140 #define SECCOMP_RET_LOG_NAME "log"
2141 #define SECCOMP_RET_ALLOW_NAME "allow"
2143 static const char seccomp_actions_avail
[] =
2144 SECCOMP_RET_KILL_PROCESS_NAME
" "
2145 SECCOMP_RET_KILL_THREAD_NAME
" "
2146 SECCOMP_RET_TRAP_NAME
" "
2147 SECCOMP_RET_ERRNO_NAME
" "
2148 SECCOMP_RET_USER_NOTIF_NAME
" "
2149 SECCOMP_RET_TRACE_NAME
" "
2150 SECCOMP_RET_LOG_NAME
" "
2151 SECCOMP_RET_ALLOW_NAME
;
2153 struct seccomp_log_name
{
2158 static const struct seccomp_log_name seccomp_log_names
[] = {
2159 { SECCOMP_LOG_KILL_PROCESS
, SECCOMP_RET_KILL_PROCESS_NAME
},
2160 { SECCOMP_LOG_KILL_THREAD
, SECCOMP_RET_KILL_THREAD_NAME
},
2161 { SECCOMP_LOG_TRAP
, SECCOMP_RET_TRAP_NAME
},
2162 { SECCOMP_LOG_ERRNO
, SECCOMP_RET_ERRNO_NAME
},
2163 { SECCOMP_LOG_USER_NOTIF
, SECCOMP_RET_USER_NOTIF_NAME
},
2164 { SECCOMP_LOG_TRACE
, SECCOMP_RET_TRACE_NAME
},
2165 { SECCOMP_LOG_LOG
, SECCOMP_RET_LOG_NAME
},
2166 { SECCOMP_LOG_ALLOW
, SECCOMP_RET_ALLOW_NAME
},
2170 static bool seccomp_names_from_actions_logged(char *names
, size_t size
,
2174 const struct seccomp_log_name
*cur
;
2175 bool append_sep
= false;
2177 for (cur
= seccomp_log_names
; cur
->name
&& size
; cur
++) {
2180 if (!(actions_logged
& cur
->log
))
2184 ret
= strscpy(names
, sep
, size
);
2193 ret
= strscpy(names
, cur
->name
, size
);
2204 static bool seccomp_action_logged_from_name(u32
*action_logged
,
2207 const struct seccomp_log_name
*cur
;
2209 for (cur
= seccomp_log_names
; cur
->name
; cur
++) {
2210 if (!strcmp(cur
->name
, name
)) {
2211 *action_logged
= cur
->log
;
2219 static bool seccomp_actions_logged_from_names(u32
*actions_logged
, char *names
)
2223 *actions_logged
= 0;
2224 while ((name
= strsep(&names
, " ")) && *name
) {
2225 u32 action_logged
= 0;
2227 if (!seccomp_action_logged_from_name(&action_logged
, name
))
2230 *actions_logged
|= action_logged
;
2236 static int read_actions_logged(struct ctl_table
*ro_table
, void *buffer
,
2237 size_t *lenp
, loff_t
*ppos
)
2239 char names
[sizeof(seccomp_actions_avail
)];
2240 struct ctl_table table
;
2242 memset(names
, 0, sizeof(names
));
2244 if (!seccomp_names_from_actions_logged(names
, sizeof(names
),
2245 seccomp_actions_logged
, " "))
2250 table
.maxlen
= sizeof(names
);
2251 return proc_dostring(&table
, 0, buffer
, lenp
, ppos
);
2254 static int write_actions_logged(struct ctl_table
*ro_table
, void *buffer
,
2255 size_t *lenp
, loff_t
*ppos
, u32
*actions_logged
)
2257 char names
[sizeof(seccomp_actions_avail
)];
2258 struct ctl_table table
;
2261 if (!capable(CAP_SYS_ADMIN
))
2264 memset(names
, 0, sizeof(names
));
2268 table
.maxlen
= sizeof(names
);
2269 ret
= proc_dostring(&table
, 1, buffer
, lenp
, ppos
);
2273 if (!seccomp_actions_logged_from_names(actions_logged
, table
.data
))
2276 if (*actions_logged
& SECCOMP_LOG_ALLOW
)
2279 seccomp_actions_logged
= *actions_logged
;
2283 static void audit_actions_logged(u32 actions_logged
, u32 old_actions_logged
,
2286 char names
[sizeof(seccomp_actions_avail
)];
2287 char old_names
[sizeof(seccomp_actions_avail
)];
2288 const char *new = names
;
2289 const char *old
= old_names
;
2294 memset(names
, 0, sizeof(names
));
2295 memset(old_names
, 0, sizeof(old_names
));
2299 else if (!actions_logged
)
2301 else if (!seccomp_names_from_actions_logged(names
, sizeof(names
),
2302 actions_logged
, ","))
2305 if (!old_actions_logged
)
2307 else if (!seccomp_names_from_actions_logged(old_names
,
2309 old_actions_logged
, ","))
2312 return audit_seccomp_actions_logged(new, old
, !ret
);
2315 static int seccomp_actions_logged_handler(struct ctl_table
*ro_table
, int write
,
2316 void *buffer
, size_t *lenp
,
2322 u32 actions_logged
= 0;
2323 u32 old_actions_logged
= seccomp_actions_logged
;
2325 ret
= write_actions_logged(ro_table
, buffer
, lenp
, ppos
,
2327 audit_actions_logged(actions_logged
, old_actions_logged
, ret
);
2329 ret
= read_actions_logged(ro_table
, buffer
, lenp
, ppos
);
2334 static struct ctl_path seccomp_sysctl_path
[] = {
2335 { .procname
= "kernel", },
2336 { .procname
= "seccomp", },
2340 static struct ctl_table seccomp_sysctl_table
[] = {
2342 .procname
= "actions_avail",
2343 .data
= (void *) &seccomp_actions_avail
,
2344 .maxlen
= sizeof(seccomp_actions_avail
),
2346 .proc_handler
= proc_dostring
,
2349 .procname
= "actions_logged",
2351 .proc_handler
= seccomp_actions_logged_handler
,
2356 static int __init
seccomp_sysctl_init(void)
2358 struct ctl_table_header
*hdr
;
2360 hdr
= register_sysctl_paths(seccomp_sysctl_path
, seccomp_sysctl_table
);
2362 pr_warn("sysctl registration failed\n");
2364 kmemleak_not_leak(hdr
);
2369 device_initcall(seccomp_sysctl_init
)
2371 #endif /* CONFIG_SYSCTL */
2373 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
2374 /* Currently CONFIG_SECCOMP_CACHE_DEBUG implies SECCOMP_ARCH_NATIVE */
2375 static void proc_pid_seccomp_cache_arch(struct seq_file
*m
, const char *name
,
2376 const void *bitmap
, size_t bitmap_size
)
2380 for (nr
= 0; nr
< bitmap_size
; nr
++) {
2381 bool cached
= test_bit(nr
, bitmap
);
2382 char *status
= cached
? "ALLOW" : "FILTER";
2384 seq_printf(m
, "%s %d %s\n", name
, nr
, status
);
2388 int proc_pid_seccomp_cache(struct seq_file
*m
, struct pid_namespace
*ns
,
2389 struct pid
*pid
, struct task_struct
*task
)
2391 struct seccomp_filter
*f
;
2392 unsigned long flags
;
2395 * We don't want some sandboxed process to know what their seccomp
2396 * filters consist of.
2398 if (!file_ns_capable(m
->file
, &init_user_ns
, CAP_SYS_ADMIN
))
2401 if (!lock_task_sighand(task
, &flags
))
2404 f
= READ_ONCE(task
->seccomp
.filter
);
2406 unlock_task_sighand(task
, &flags
);
2410 /* prevent filter from being freed while we are printing it */
2411 __get_seccomp_filter(f
);
2412 unlock_task_sighand(task
, &flags
);
2414 proc_pid_seccomp_cache_arch(m
, SECCOMP_ARCH_NATIVE_NAME
,
2415 f
->cache
.allow_native
,
2416 SECCOMP_ARCH_NATIVE_NR
);
2418 #ifdef SECCOMP_ARCH_COMPAT
2419 proc_pid_seccomp_cache_arch(m
, SECCOMP_ARCH_COMPAT_NAME
,
2420 f
->cache
.allow_compat
,
2421 SECCOMP_ARCH_COMPAT_NR
);
2422 #endif /* SECCOMP_ARCH_COMPAT */
2424 __put_seccomp_filter(f
);
2427 #endif /* CONFIG_SECCOMP_CACHE_DEBUG */