1 // SPDX-License-Identifier: GPL-2.0
3 * linux/kernel/seccomp.c
5 * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com>
7 * Copyright (C) 2012 Google, Inc.
8 * Will Drewry <wad@chromium.org>
10 * This defines a simple but solid secure-computing facility.
12 * Mode 1 uses a fixed list of allowed system calls.
13 * Mode 2 allows user-defined system call filters in the form
14 * of Berkeley Packet Filters/Linux Socket Filters.
16 #define pr_fmt(fmt) "seccomp: " fmt
18 #include <linux/refcount.h>
19 #include <linux/audit.h>
20 #include <linux/compat.h>
21 #include <linux/coredump.h>
22 #include <linux/kmemleak.h>
23 #include <linux/nospec.h>
24 #include <linux/prctl.h>
25 #include <linux/sched.h>
26 #include <linux/sched/task_stack.h>
27 #include <linux/seccomp.h>
28 #include <linux/slab.h>
29 #include <linux/syscalls.h>
30 #include <linux/sysctl.h>
32 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
33 #include <asm/syscall.h>
36 #ifdef CONFIG_SECCOMP_FILTER
37 #include <linux/file.h>
38 #include <linux/filter.h>
39 #include <linux/pid.h>
40 #include <linux/ptrace.h>
41 #include <linux/capability.h>
42 #include <linux/tracehook.h>
43 #include <linux/uaccess.h>
44 #include <linux/anon_inodes.h>
45 #include <linux/lockdep.h>
48 * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the
49 * wrong direction flag in the ioctl number. This is the broken one,
50 * which the kernel needs to keep supporting until all userspaces stop
51 * using the wrong command number.
53 #define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR SECCOMP_IOR(2, __u64)
58 SECCOMP_NOTIFY_REPLIED
,
61 struct seccomp_knotif
{
62 /* The struct pid of the task whose filter triggered the notification */
63 struct task_struct
*task
;
65 /* The "cookie" for this request; this is unique for this filter. */
69 * The seccomp data. This pointer is valid the entire time this
70 * notification is active, since it comes from __seccomp_filter which
71 * eclipses the entire lifecycle here.
73 const struct seccomp_data
*data
;
76 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
77 * struct seccomp_knotif is created and starts out in INIT. Once the
78 * handler reads the notification off of an FD, it transitions to SENT.
79 * If a signal is received the state transitions back to INIT and
80 * another message is sent. When the userspace handler replies, state
81 * transitions to REPLIED.
83 enum notify_state state
;
85 /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
91 * Signals when this has changed states, such as the listener
92 * dying, a new seccomp addfd message, or changing to REPLIED
94 struct completion ready
;
96 struct list_head list
;
98 /* outstanding addfd requests */
99 struct list_head addfd
;
103 * struct seccomp_kaddfd - container for seccomp_addfd ioctl messages
105 * @file: A reference to the file to install in the other task
106 * @fd: The fd number to install it at. If the fd number is -1, it means the
107 * installing process should allocate the fd as normal.
108 * @flags: The flags for the new file descriptor. At the moment, only O_CLOEXEC
110 * @ret: The return value of the installing process. It is set to the fd num
111 * upon success (>= 0).
112 * @completion: Indicates that the installing process has completed fd
113 * installation, or gone away (either due to successful
117 struct seccomp_kaddfd
{
124 /* To only be set on reply */
127 struct completion completion
;
128 struct list_head list
;
132 * struct notification - container for seccomp userspace notifications. Since
133 * most seccomp filters will not have notification listeners attached and this
134 * structure is fairly large, we store the notification-specific stuff in a
135 * separate structure.
137 * @request: A semaphore that users of this notification can wait on for
138 * changes. Actual reads and writes are still controlled with
139 * filter->notify_lock.
140 * @next_id: The id of the next request.
141 * @notifications: A list of struct seccomp_knotif elements.
143 struct notification
{
144 struct semaphore request
;
146 struct list_head notifications
;
149 #ifdef SECCOMP_ARCH_NATIVE
151 * struct action_cache - per-filter cache of seccomp actions per
154 * @allow_native: A bitmap where each bit represents whether the
155 * filter will always allow the syscall, for the
156 * native architecture.
157 * @allow_compat: A bitmap where each bit represents whether the
158 * filter will always allow the syscall, for the
159 * compat architecture.
161 struct action_cache
{
162 DECLARE_BITMAP(allow_native
, SECCOMP_ARCH_NATIVE_NR
);
163 #ifdef SECCOMP_ARCH_COMPAT
164 DECLARE_BITMAP(allow_compat
, SECCOMP_ARCH_COMPAT_NR
);
168 struct action_cache
{ };
170 static inline bool seccomp_cache_check_allow(const struct seccomp_filter
*sfilter
,
171 const struct seccomp_data
*sd
)
176 static inline void seccomp_cache_prepare(struct seccomp_filter
*sfilter
)
179 #endif /* SECCOMP_ARCH_NATIVE */
182 * struct seccomp_filter - container for seccomp BPF programs
184 * @refs: Reference count to manage the object lifetime.
185 * A filter's reference count is incremented for each directly
186 * attached task, once for the dependent filter, and if
187 * requested for the user notifier. When @refs reaches zero,
188 * the filter can be freed.
189 * @users: A filter's @users count is incremented for each directly
190 * attached task (filter installation, fork(), thread_sync),
191 * and once for the dependent filter (tracked in filter->prev).
192 * When it reaches zero it indicates that no direct or indirect
193 * users of that filter exist. No new tasks can get associated with
194 * this filter after reaching 0. The @users count is always smaller
195 * or equal to @refs. Hence, reaching 0 for @users does not mean
196 * the filter can be freed.
197 * @cache: cache of arch/syscall mappings to actions
198 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
199 * @prev: points to a previously installed, or inherited, filter
200 * @prog: the BPF program to evaluate
201 * @notif: the struct that holds all notification related information
202 * @notify_lock: A lock for all notification-related accesses.
203 * @wqh: A wait queue for poll if a notifier is in use.
205 * seccomp_filter objects are organized in a tree linked via the @prev
206 * pointer. For any task, it appears to be a singly-linked list starting
207 * with current->seccomp.filter, the most recently attached or inherited filter.
208 * However, multiple filters may share a @prev node, by way of fork(), which
209 * results in a unidirectional tree existing in memory. This is similar to
210 * how namespaces work.
212 * seccomp_filter objects should never be modified after being attached
213 * to a task_struct (other than @refs).
215 struct seccomp_filter
{
219 struct action_cache cache
;
220 struct seccomp_filter
*prev
;
221 struct bpf_prog
*prog
;
222 struct notification
*notif
;
223 struct mutex notify_lock
;
224 wait_queue_head_t wqh
;
227 /* Limit any path through the tree to 256KB worth of instructions. */
228 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
231 * Endianness is explicitly ignored and left for BPF program authors to manage
232 * as per the specific architecture.
234 static void populate_seccomp_data(struct seccomp_data
*sd
)
237 * Instead of using current_pt_reg(), we're already doing the work
238 * to safely fetch "current", so just use "task" everywhere below.
240 struct task_struct
*task
= current
;
241 struct pt_regs
*regs
= task_pt_regs(task
);
242 unsigned long args
[6];
244 sd
->nr
= syscall_get_nr(task
, regs
);
245 sd
->arch
= syscall_get_arch(task
);
246 syscall_get_arguments(task
, regs
, args
);
247 sd
->args
[0] = args
[0];
248 sd
->args
[1] = args
[1];
249 sd
->args
[2] = args
[2];
250 sd
->args
[3] = args
[3];
251 sd
->args
[4] = args
[4];
252 sd
->args
[5] = args
[5];
253 sd
->instruction_pointer
= KSTK_EIP(task
);
257 * seccomp_check_filter - verify seccomp filter code
258 * @filter: filter to verify
259 * @flen: length of filter
261 * Takes a previously checked filter (by bpf_check_classic) and
262 * redirects all filter code that loads struct sk_buff data
263 * and related data through seccomp_bpf_load. It also
264 * enforces length and alignment checking of those loads.
266 * Returns 0 if the rule set is legal or -EINVAL if not.
268 static int seccomp_check_filter(struct sock_filter
*filter
, unsigned int flen
)
271 for (pc
= 0; pc
< flen
; pc
++) {
272 struct sock_filter
*ftest
= &filter
[pc
];
273 u16 code
= ftest
->code
;
277 case BPF_LD
| BPF_W
| BPF_ABS
:
278 ftest
->code
= BPF_LDX
| BPF_W
| BPF_ABS
;
279 /* 32-bit aligned and not out of bounds. */
280 if (k
>= sizeof(struct seccomp_data
) || k
& 3)
283 case BPF_LD
| BPF_W
| BPF_LEN
:
284 ftest
->code
= BPF_LD
| BPF_IMM
;
285 ftest
->k
= sizeof(struct seccomp_data
);
287 case BPF_LDX
| BPF_W
| BPF_LEN
:
288 ftest
->code
= BPF_LDX
| BPF_IMM
;
289 ftest
->k
= sizeof(struct seccomp_data
);
291 /* Explicitly include allowed calls. */
292 case BPF_RET
| BPF_K
:
293 case BPF_RET
| BPF_A
:
294 case BPF_ALU
| BPF_ADD
| BPF_K
:
295 case BPF_ALU
| BPF_ADD
| BPF_X
:
296 case BPF_ALU
| BPF_SUB
| BPF_K
:
297 case BPF_ALU
| BPF_SUB
| BPF_X
:
298 case BPF_ALU
| BPF_MUL
| BPF_K
:
299 case BPF_ALU
| BPF_MUL
| BPF_X
:
300 case BPF_ALU
| BPF_DIV
| BPF_K
:
301 case BPF_ALU
| BPF_DIV
| BPF_X
:
302 case BPF_ALU
| BPF_AND
| BPF_K
:
303 case BPF_ALU
| BPF_AND
| BPF_X
:
304 case BPF_ALU
| BPF_OR
| BPF_K
:
305 case BPF_ALU
| BPF_OR
| BPF_X
:
306 case BPF_ALU
| BPF_XOR
| BPF_K
:
307 case BPF_ALU
| BPF_XOR
| BPF_X
:
308 case BPF_ALU
| BPF_LSH
| BPF_K
:
309 case BPF_ALU
| BPF_LSH
| BPF_X
:
310 case BPF_ALU
| BPF_RSH
| BPF_K
:
311 case BPF_ALU
| BPF_RSH
| BPF_X
:
312 case BPF_ALU
| BPF_NEG
:
313 case BPF_LD
| BPF_IMM
:
314 case BPF_LDX
| BPF_IMM
:
315 case BPF_MISC
| BPF_TAX
:
316 case BPF_MISC
| BPF_TXA
:
317 case BPF_LD
| BPF_MEM
:
318 case BPF_LDX
| BPF_MEM
:
321 case BPF_JMP
| BPF_JA
:
322 case BPF_JMP
| BPF_JEQ
| BPF_K
:
323 case BPF_JMP
| BPF_JEQ
| BPF_X
:
324 case BPF_JMP
| BPF_JGE
| BPF_K
:
325 case BPF_JMP
| BPF_JGE
| BPF_X
:
326 case BPF_JMP
| BPF_JGT
| BPF_K
:
327 case BPF_JMP
| BPF_JGT
| BPF_X
:
328 case BPF_JMP
| BPF_JSET
| BPF_K
:
329 case BPF_JMP
| BPF_JSET
| BPF_X
:
338 #ifdef SECCOMP_ARCH_NATIVE
339 static inline bool seccomp_cache_check_allow_bitmap(const void *bitmap
,
343 if (unlikely(syscall_nr
< 0 || syscall_nr
>= bitmap_size
))
345 syscall_nr
= array_index_nospec(syscall_nr
, bitmap_size
);
347 return test_bit(syscall_nr
, bitmap
);
351 * seccomp_cache_check_allow - lookup seccomp cache
352 * @sfilter: The seccomp filter
353 * @sd: The seccomp data to lookup the cache with
355 * Returns true if the seccomp_data is cached and allowed.
357 static inline bool seccomp_cache_check_allow(const struct seccomp_filter
*sfilter
,
358 const struct seccomp_data
*sd
)
360 int syscall_nr
= sd
->nr
;
361 const struct action_cache
*cache
= &sfilter
->cache
;
363 #ifndef SECCOMP_ARCH_COMPAT
364 /* A native-only architecture doesn't need to check sd->arch. */
365 return seccomp_cache_check_allow_bitmap(cache
->allow_native
,
366 SECCOMP_ARCH_NATIVE_NR
,
369 if (likely(sd
->arch
== SECCOMP_ARCH_NATIVE
))
370 return seccomp_cache_check_allow_bitmap(cache
->allow_native
,
371 SECCOMP_ARCH_NATIVE_NR
,
373 if (likely(sd
->arch
== SECCOMP_ARCH_COMPAT
))
374 return seccomp_cache_check_allow_bitmap(cache
->allow_compat
,
375 SECCOMP_ARCH_COMPAT_NR
,
377 #endif /* SECCOMP_ARCH_COMPAT */
382 #endif /* SECCOMP_ARCH_NATIVE */
385 * seccomp_run_filters - evaluates all seccomp filters against @sd
386 * @sd: optional seccomp data to be passed to filters
387 * @match: stores struct seccomp_filter that resulted in the return value,
388 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
391 * Returns valid seccomp BPF response codes.
393 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
394 static u32
seccomp_run_filters(const struct seccomp_data
*sd
,
395 struct seccomp_filter
**match
)
397 u32 ret
= SECCOMP_RET_ALLOW
;
398 /* Make sure cross-thread synced filter points somewhere sane. */
399 struct seccomp_filter
*f
=
400 READ_ONCE(current
->seccomp
.filter
);
402 /* Ensure unexpected behavior doesn't result in failing open. */
403 if (WARN_ON(f
== NULL
))
404 return SECCOMP_RET_KILL_PROCESS
;
406 if (seccomp_cache_check_allow(f
, sd
))
407 return SECCOMP_RET_ALLOW
;
410 * All filters in the list are evaluated and the lowest BPF return
411 * value always takes priority (ignoring the DATA).
413 for (; f
; f
= f
->prev
) {
414 u32 cur_ret
= bpf_prog_run_pin_on_cpu(f
->prog
, sd
);
416 if (ACTION_ONLY(cur_ret
) < ACTION_ONLY(ret
)) {
423 #endif /* CONFIG_SECCOMP_FILTER */
425 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode
)
427 assert_spin_locked(¤t
->sighand
->siglock
);
429 if (current
->seccomp
.mode
&& current
->seccomp
.mode
!= seccomp_mode
)
435 void __weak
arch_seccomp_spec_mitigate(struct task_struct
*task
) { }
437 static inline void seccomp_assign_mode(struct task_struct
*task
,
438 unsigned long seccomp_mode
,
441 assert_spin_locked(&task
->sighand
->siglock
);
443 task
->seccomp
.mode
= seccomp_mode
;
445 * Make sure SYSCALL_WORK_SECCOMP cannot be set before the mode (and
448 smp_mb__before_atomic();
449 /* Assume default seccomp processes want spec flaw mitigation. */
450 if ((flags
& SECCOMP_FILTER_FLAG_SPEC_ALLOW
) == 0)
451 arch_seccomp_spec_mitigate(task
);
452 set_task_syscall_work(task
, SECCOMP
);
455 #ifdef CONFIG_SECCOMP_FILTER
456 /* Returns 1 if the parent is an ancestor of the child. */
457 static int is_ancestor(struct seccomp_filter
*parent
,
458 struct seccomp_filter
*child
)
460 /* NULL is the root ancestor. */
463 for (; child
; child
= child
->prev
)
470 * seccomp_can_sync_threads: checks if all threads can be synchronized
472 * Expects sighand and cred_guard_mutex locks to be held.
474 * Returns 0 on success, -ve on error, or the pid of a thread which was
475 * either not in the correct seccomp mode or did not have an ancestral
478 static inline pid_t
seccomp_can_sync_threads(void)
480 struct task_struct
*thread
, *caller
;
482 BUG_ON(!mutex_is_locked(¤t
->signal
->cred_guard_mutex
));
483 assert_spin_locked(¤t
->sighand
->siglock
);
485 /* Validate all threads being eligible for synchronization. */
487 for_each_thread(caller
, thread
) {
490 /* Skip current, since it is initiating the sync. */
491 if (thread
== caller
)
494 if (thread
->seccomp
.mode
== SECCOMP_MODE_DISABLED
||
495 (thread
->seccomp
.mode
== SECCOMP_MODE_FILTER
&&
496 is_ancestor(thread
->seccomp
.filter
,
497 caller
->seccomp
.filter
)))
500 /* Return the first thread that cannot be synchronized. */
501 failed
= task_pid_vnr(thread
);
502 /* If the pid cannot be resolved, then return -ESRCH */
503 if (WARN_ON(failed
== 0))
511 static inline void seccomp_filter_free(struct seccomp_filter
*filter
)
514 bpf_prog_destroy(filter
->prog
);
519 static void __seccomp_filter_orphan(struct seccomp_filter
*orig
)
521 while (orig
&& refcount_dec_and_test(&orig
->users
)) {
522 if (waitqueue_active(&orig
->wqh
))
523 wake_up_poll(&orig
->wqh
, EPOLLHUP
);
528 static void __put_seccomp_filter(struct seccomp_filter
*orig
)
530 /* Clean up single-reference branches iteratively. */
531 while (orig
&& refcount_dec_and_test(&orig
->refs
)) {
532 struct seccomp_filter
*freeme
= orig
;
534 seccomp_filter_free(freeme
);
538 static void __seccomp_filter_release(struct seccomp_filter
*orig
)
540 /* Notify about any unused filters in the task's former filter tree. */
541 __seccomp_filter_orphan(orig
);
542 /* Finally drop all references to the task's former tree. */
543 __put_seccomp_filter(orig
);
547 * seccomp_filter_release - Detach the task from its filter tree,
548 * drop its reference count, and notify
549 * about unused filters
551 * This function should only be called when the task is exiting as
552 * it detaches it from its filter tree. As such, READ_ONCE() and
553 * barriers are not needed here, as would normally be needed.
555 void seccomp_filter_release(struct task_struct
*tsk
)
557 struct seccomp_filter
*orig
= tsk
->seccomp
.filter
;
559 /* We are effectively holding the siglock by not having any sighand. */
560 WARN_ON(tsk
->sighand
!= NULL
);
562 /* Detach task from its filter tree. */
563 tsk
->seccomp
.filter
= NULL
;
564 __seccomp_filter_release(orig
);
568 * seccomp_sync_threads: sets all threads to use current's filter
570 * Expects sighand and cred_guard_mutex locks to be held, and for
571 * seccomp_can_sync_threads() to have returned success already
572 * without dropping the locks.
575 static inline void seccomp_sync_threads(unsigned long flags
)
577 struct task_struct
*thread
, *caller
;
579 BUG_ON(!mutex_is_locked(¤t
->signal
->cred_guard_mutex
));
580 assert_spin_locked(¤t
->sighand
->siglock
);
582 /* Synchronize all threads. */
584 for_each_thread(caller
, thread
) {
585 /* Skip current, since it needs no changes. */
586 if (thread
== caller
)
589 /* Get a task reference for the new leaf node. */
590 get_seccomp_filter(caller
);
593 * Drop the task reference to the shared ancestor since
594 * current's path will hold a reference. (This also
595 * allows a put before the assignment.)
597 __seccomp_filter_release(thread
->seccomp
.filter
);
599 /* Make our new filter tree visible. */
600 smp_store_release(&thread
->seccomp
.filter
,
601 caller
->seccomp
.filter
);
602 atomic_set(&thread
->seccomp
.filter_count
,
603 atomic_read(&thread
->seccomp
.filter_count
));
606 * Don't let an unprivileged task work around
607 * the no_new_privs restriction by creating
608 * a thread that sets it up, enters seccomp,
611 if (task_no_new_privs(caller
))
612 task_set_no_new_privs(thread
);
615 * Opt the other thread into seccomp if needed.
616 * As threads are considered to be trust-realm
617 * equivalent (see ptrace_may_access), it is safe to
618 * allow one thread to transition the other.
620 if (thread
->seccomp
.mode
== SECCOMP_MODE_DISABLED
)
621 seccomp_assign_mode(thread
, SECCOMP_MODE_FILTER
,
627 * seccomp_prepare_filter: Prepares a seccomp filter for use.
628 * @fprog: BPF program to install
630 * Returns filter on success or an ERR_PTR on failure.
632 static struct seccomp_filter
*seccomp_prepare_filter(struct sock_fprog
*fprog
)
634 struct seccomp_filter
*sfilter
;
636 const bool save_orig
=
637 #if defined(CONFIG_CHECKPOINT_RESTORE) || defined(SECCOMP_ARCH_NATIVE)
643 if (fprog
->len
== 0 || fprog
->len
> BPF_MAXINSNS
)
644 return ERR_PTR(-EINVAL
);
646 BUG_ON(INT_MAX
/ fprog
->len
< sizeof(struct sock_filter
));
649 * Installing a seccomp filter requires that the task has
650 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
651 * This avoids scenarios where unprivileged tasks can affect the
652 * behavior of privileged children.
654 if (!task_no_new_privs(current
) &&
655 !ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN
))
656 return ERR_PTR(-EACCES
);
658 /* Allocate a new seccomp_filter */
659 sfilter
= kzalloc(sizeof(*sfilter
), GFP_KERNEL
| __GFP_NOWARN
);
661 return ERR_PTR(-ENOMEM
);
663 mutex_init(&sfilter
->notify_lock
);
664 ret
= bpf_prog_create_from_user(&sfilter
->prog
, fprog
,
665 seccomp_check_filter
, save_orig
);
671 refcount_set(&sfilter
->refs
, 1);
672 refcount_set(&sfilter
->users
, 1);
673 init_waitqueue_head(&sfilter
->wqh
);
679 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
680 * @user_filter: pointer to the user data containing a sock_fprog.
682 * Returns 0 on success and non-zero otherwise.
684 static struct seccomp_filter
*
685 seccomp_prepare_user_filter(const char __user
*user_filter
)
687 struct sock_fprog fprog
;
688 struct seccomp_filter
*filter
= ERR_PTR(-EFAULT
);
691 if (in_compat_syscall()) {
692 struct compat_sock_fprog fprog32
;
693 if (copy_from_user(&fprog32
, user_filter
, sizeof(fprog32
)))
695 fprog
.len
= fprog32
.len
;
696 fprog
.filter
= compat_ptr(fprog32
.filter
);
697 } else /* falls through to the if below. */
699 if (copy_from_user(&fprog
, user_filter
, sizeof(fprog
)))
701 filter
= seccomp_prepare_filter(&fprog
);
706 #ifdef SECCOMP_ARCH_NATIVE
708 * seccomp_is_const_allow - check if filter is constant allow with given data
709 * @fprog: The BPF programs
710 * @sd: The seccomp data to check against, only syscall number and arch
711 * number are considered constant.
713 static bool seccomp_is_const_allow(struct sock_fprog_kern
*fprog
,
714 struct seccomp_data
*sd
)
716 unsigned int reg_value
= 0;
720 if (WARN_ON_ONCE(!fprog
))
723 for (pc
= 0; pc
< fprog
->len
; pc
++) {
724 struct sock_filter
*insn
= &fprog
->filter
[pc
];
725 u16 code
= insn
->code
;
729 case BPF_LD
| BPF_W
| BPF_ABS
:
731 case offsetof(struct seccomp_data
, nr
):
734 case offsetof(struct seccomp_data
, arch
):
735 reg_value
= sd
->arch
;
738 /* can't optimize (non-constant value load) */
742 case BPF_RET
| BPF_K
:
743 /* reached return with constant values only, check allow */
744 return k
== SECCOMP_RET_ALLOW
;
745 case BPF_JMP
| BPF_JA
:
748 case BPF_JMP
| BPF_JEQ
| BPF_K
:
749 case BPF_JMP
| BPF_JGE
| BPF_K
:
750 case BPF_JMP
| BPF_JGT
| BPF_K
:
751 case BPF_JMP
| BPF_JSET
| BPF_K
:
752 switch (BPF_OP(code
)) {
754 op_res
= reg_value
== k
;
757 op_res
= reg_value
>= k
;
760 op_res
= reg_value
> k
;
763 op_res
= !!(reg_value
& k
);
766 /* can't optimize (unknown jump) */
770 pc
+= op_res
? insn
->jt
: insn
->jf
;
772 case BPF_ALU
| BPF_AND
| BPF_K
:
776 /* can't optimize (unknown insn) */
781 /* ran off the end of the filter?! */
786 static void seccomp_cache_prepare_bitmap(struct seccomp_filter
*sfilter
,
787 void *bitmap
, const void *bitmap_prev
,
788 size_t bitmap_size
, int arch
)
790 struct sock_fprog_kern
*fprog
= sfilter
->prog
->orig_prog
;
791 struct seccomp_data sd
;
795 /* The new filter must be as restrictive as the last. */
796 bitmap_copy(bitmap
, bitmap_prev
, bitmap_size
);
798 /* Before any filters, all syscalls are always allowed. */
799 bitmap_fill(bitmap
, bitmap_size
);
802 for (nr
= 0; nr
< bitmap_size
; nr
++) {
803 /* No bitmap change: not a cacheable action. */
804 if (!test_bit(nr
, bitmap
))
810 /* No bitmap change: continue to always allow. */
811 if (seccomp_is_const_allow(fprog
, &sd
))
815 * Not a cacheable action: always run filters.
816 * atomic clear_bit() not needed, filter not visible yet.
818 __clear_bit(nr
, bitmap
);
823 * seccomp_cache_prepare - emulate the filter to find cacheable syscalls
824 * @sfilter: The seccomp filter
826 * Returns 0 if successful or -errno if error occurred.
828 static void seccomp_cache_prepare(struct seccomp_filter
*sfilter
)
830 struct action_cache
*cache
= &sfilter
->cache
;
831 const struct action_cache
*cache_prev
=
832 sfilter
->prev
? &sfilter
->prev
->cache
: NULL
;
834 seccomp_cache_prepare_bitmap(sfilter
, cache
->allow_native
,
835 cache_prev
? cache_prev
->allow_native
: NULL
,
836 SECCOMP_ARCH_NATIVE_NR
,
837 SECCOMP_ARCH_NATIVE
);
839 #ifdef SECCOMP_ARCH_COMPAT
840 seccomp_cache_prepare_bitmap(sfilter
, cache
->allow_compat
,
841 cache_prev
? cache_prev
->allow_compat
: NULL
,
842 SECCOMP_ARCH_COMPAT_NR
,
843 SECCOMP_ARCH_COMPAT
);
844 #endif /* SECCOMP_ARCH_COMPAT */
846 #endif /* SECCOMP_ARCH_NATIVE */
849 * seccomp_attach_filter: validate and attach filter
850 * @flags: flags to change filter behavior
851 * @filter: seccomp filter to add to the current process
853 * Caller must be holding current->sighand->siglock lock.
855 * Returns 0 on success, -ve on error, or
856 * - in TSYNC mode: the pid of a thread which was either not in the correct
857 * seccomp mode or did not have an ancestral seccomp filter
858 * - in NEW_LISTENER mode: the fd of the new listener
860 static long seccomp_attach_filter(unsigned int flags
,
861 struct seccomp_filter
*filter
)
863 unsigned long total_insns
;
864 struct seccomp_filter
*walker
;
866 assert_spin_locked(¤t
->sighand
->siglock
);
868 /* Validate resulting filter length. */
869 total_insns
= filter
->prog
->len
;
870 for (walker
= current
->seccomp
.filter
; walker
; walker
= walker
->prev
)
871 total_insns
+= walker
->prog
->len
+ 4; /* 4 instr penalty */
872 if (total_insns
> MAX_INSNS_PER_PATH
)
875 /* If thread sync has been requested, check that it is possible. */
876 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
) {
879 ret
= seccomp_can_sync_threads();
881 if (flags
& SECCOMP_FILTER_FLAG_TSYNC_ESRCH
)
888 /* Set log flag, if present. */
889 if (flags
& SECCOMP_FILTER_FLAG_LOG
)
893 * If there is an existing filter, make it the prev and don't drop its
896 filter
->prev
= current
->seccomp
.filter
;
897 seccomp_cache_prepare(filter
);
898 current
->seccomp
.filter
= filter
;
899 atomic_inc(¤t
->seccomp
.filter_count
);
901 /* Now that the new filter is in place, synchronize to all threads. */
902 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
)
903 seccomp_sync_threads(flags
);
908 static void __get_seccomp_filter(struct seccomp_filter
*filter
)
910 refcount_inc(&filter
->refs
);
913 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
914 void get_seccomp_filter(struct task_struct
*tsk
)
916 struct seccomp_filter
*orig
= tsk
->seccomp
.filter
;
919 __get_seccomp_filter(orig
);
920 refcount_inc(&orig
->users
);
923 static void seccomp_init_siginfo(kernel_siginfo_t
*info
, int syscall
, int reason
)
926 info
->si_signo
= SIGSYS
;
927 info
->si_code
= SYS_SECCOMP
;
928 info
->si_call_addr
= (void __user
*)KSTK_EIP(current
);
929 info
->si_errno
= reason
;
930 info
->si_arch
= syscall_get_arch(current
);
931 info
->si_syscall
= syscall
;
935 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
936 * @syscall: syscall number to send to userland
937 * @reason: filter-supplied reason code to send to userland (via si_errno)
939 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
941 static void seccomp_send_sigsys(int syscall
, int reason
)
943 struct kernel_siginfo info
;
944 seccomp_init_siginfo(&info
, syscall
, reason
);
945 force_sig_info(&info
);
947 #endif /* CONFIG_SECCOMP_FILTER */
949 /* For use with seccomp_actions_logged */
950 #define SECCOMP_LOG_KILL_PROCESS (1 << 0)
951 #define SECCOMP_LOG_KILL_THREAD (1 << 1)
952 #define SECCOMP_LOG_TRAP (1 << 2)
953 #define SECCOMP_LOG_ERRNO (1 << 3)
954 #define SECCOMP_LOG_TRACE (1 << 4)
955 #define SECCOMP_LOG_LOG (1 << 5)
956 #define SECCOMP_LOG_ALLOW (1 << 6)
957 #define SECCOMP_LOG_USER_NOTIF (1 << 7)
959 static u32 seccomp_actions_logged
= SECCOMP_LOG_KILL_PROCESS
|
960 SECCOMP_LOG_KILL_THREAD
|
963 SECCOMP_LOG_USER_NOTIF
|
967 static inline void seccomp_log(unsigned long syscall
, long signr
, u32 action
,
973 case SECCOMP_RET_ALLOW
:
975 case SECCOMP_RET_TRAP
:
976 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_TRAP
;
978 case SECCOMP_RET_ERRNO
:
979 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_ERRNO
;
981 case SECCOMP_RET_TRACE
:
982 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_TRACE
;
984 case SECCOMP_RET_USER_NOTIF
:
985 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_USER_NOTIF
;
987 case SECCOMP_RET_LOG
:
988 log
= seccomp_actions_logged
& SECCOMP_LOG_LOG
;
990 case SECCOMP_RET_KILL_THREAD
:
991 log
= seccomp_actions_logged
& SECCOMP_LOG_KILL_THREAD
;
993 case SECCOMP_RET_KILL_PROCESS
:
995 log
= seccomp_actions_logged
& SECCOMP_LOG_KILL_PROCESS
;
999 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
1000 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
1001 * any action from being logged by removing the action name from the
1002 * seccomp_actions_logged sysctl.
1007 audit_seccomp(syscall
, signr
, action
);
1011 * Secure computing mode 1 allows only read/write/exit/sigreturn.
1012 * To be fully secure this must be combined with rlimit
1013 * to limit the stack allocations too.
1015 static const int mode1_syscalls
[] = {
1016 __NR_seccomp_read
, __NR_seccomp_write
, __NR_seccomp_exit
, __NR_seccomp_sigreturn
,
1017 -1, /* negative terminated */
1020 static void __secure_computing_strict(int this_syscall
)
1022 const int *allowed_syscalls
= mode1_syscalls
;
1023 #ifdef CONFIG_COMPAT
1024 if (in_compat_syscall())
1025 allowed_syscalls
= get_compat_mode1_syscalls();
1028 if (*allowed_syscalls
== this_syscall
)
1030 } while (*++allowed_syscalls
!= -1);
1032 #ifdef SECCOMP_DEBUG
1035 seccomp_log(this_syscall
, SIGKILL
, SECCOMP_RET_KILL_THREAD
, true);
1039 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
1040 void secure_computing_strict(int this_syscall
)
1042 int mode
= current
->seccomp
.mode
;
1044 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE
) &&
1045 unlikely(current
->ptrace
& PT_SUSPEND_SECCOMP
))
1048 if (mode
== SECCOMP_MODE_DISABLED
)
1050 else if (mode
== SECCOMP_MODE_STRICT
)
1051 __secure_computing_strict(this_syscall
);
1057 #ifdef CONFIG_SECCOMP_FILTER
1058 static u64
seccomp_next_notify_id(struct seccomp_filter
*filter
)
1061 * Note: overflow is ok here, the id just needs to be unique per
1064 lockdep_assert_held(&filter
->notify_lock
);
1065 return filter
->notif
->next_id
++;
1068 static void seccomp_handle_addfd(struct seccomp_kaddfd
*addfd
)
1071 * Remove the notification, and reset the list pointers, indicating
1072 * that it has been handled.
1074 list_del_init(&addfd
->list
);
1076 addfd
->ret
= receive_fd(addfd
->file
, addfd
->flags
);
1078 addfd
->ret
= receive_fd_replace(addfd
->fd
, addfd
->file
,
1080 complete(&addfd
->completion
);
1083 static int seccomp_do_user_notification(int this_syscall
,
1084 struct seccomp_filter
*match
,
1085 const struct seccomp_data
*sd
)
1090 struct seccomp_knotif n
= {};
1091 struct seccomp_kaddfd
*addfd
, *tmp
;
1093 mutex_lock(&match
->notify_lock
);
1099 n
.state
= SECCOMP_NOTIFY_INIT
;
1101 n
.id
= seccomp_next_notify_id(match
);
1102 init_completion(&n
.ready
);
1103 list_add(&n
.list
, &match
->notif
->notifications
);
1104 INIT_LIST_HEAD(&n
.addfd
);
1106 up(&match
->notif
->request
);
1107 wake_up_poll(&match
->wqh
, EPOLLIN
| EPOLLRDNORM
);
1108 mutex_unlock(&match
->notify_lock
);
1111 * This is where we wait for a reply from userspace.
1114 err
= wait_for_completion_interruptible(&n
.ready
);
1115 mutex_lock(&match
->notify_lock
);
1117 /* Check if we were woken up by a addfd message */
1118 addfd
= list_first_entry_or_null(&n
.addfd
,
1119 struct seccomp_kaddfd
, list
);
1120 if (addfd
&& n
.state
!= SECCOMP_NOTIFY_REPLIED
) {
1121 seccomp_handle_addfd(addfd
);
1122 mutex_unlock(&match
->notify_lock
);
1130 /* If there were any pending addfd calls, clear them out */
1131 list_for_each_entry_safe(addfd
, tmp
, &n
.addfd
, list
) {
1132 /* The process went away before we got a chance to handle it */
1133 addfd
->ret
= -ESRCH
;
1134 list_del_init(&addfd
->list
);
1135 complete(&addfd
->completion
);
1139 * Note that it's possible the listener died in between the time when
1140 * we were notified of a response (or a signal) and when we were able to
1141 * re-acquire the lock, so only delete from the list if the
1142 * notification actually exists.
1144 * Also note that this test is only valid because there's no way to
1145 * *reattach* to a notifier right now. If one is added, we'll need to
1146 * keep track of the notif itself and make sure they match here.
1151 mutex_unlock(&match
->notify_lock
);
1153 /* Userspace requests to continue the syscall. */
1154 if (flags
& SECCOMP_USER_NOTIF_FLAG_CONTINUE
)
1157 syscall_set_return_value(current
, current_pt_regs(),
1162 static int __seccomp_filter(int this_syscall
, const struct seccomp_data
*sd
,
1163 const bool recheck_after_trace
)
1165 u32 filter_ret
, action
;
1166 struct seccomp_filter
*match
= NULL
;
1168 struct seccomp_data sd_local
;
1171 * Make sure that any changes to mode from another thread have
1172 * been seen after SYSCALL_WORK_SECCOMP was seen.
1177 populate_seccomp_data(&sd_local
);
1181 filter_ret
= seccomp_run_filters(sd
, &match
);
1182 data
= filter_ret
& SECCOMP_RET_DATA
;
1183 action
= filter_ret
& SECCOMP_RET_ACTION_FULL
;
1186 case SECCOMP_RET_ERRNO
:
1187 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
1188 if (data
> MAX_ERRNO
)
1190 syscall_set_return_value(current
, current_pt_regs(),
1194 case SECCOMP_RET_TRAP
:
1195 /* Show the handler the original registers. */
1196 syscall_rollback(current
, current_pt_regs());
1197 /* Let the filter pass back 16 bits of data. */
1198 seccomp_send_sigsys(this_syscall
, data
);
1201 case SECCOMP_RET_TRACE
:
1202 /* We've been put in this state by the ptracer already. */
1203 if (recheck_after_trace
)
1206 /* ENOSYS these calls if there is no tracer attached. */
1207 if (!ptrace_event_enabled(current
, PTRACE_EVENT_SECCOMP
)) {
1208 syscall_set_return_value(current
,
1214 /* Allow the BPF to provide the event message */
1215 ptrace_event(PTRACE_EVENT_SECCOMP
, data
);
1217 * The delivery of a fatal signal during event
1218 * notification may silently skip tracer notification,
1219 * which could leave us with a potentially unmodified
1220 * syscall that the tracer would have liked to have
1221 * changed. Since the process is about to die, we just
1222 * force the syscall to be skipped and let the signal
1223 * kill the process and correctly handle any tracer exit
1226 if (fatal_signal_pending(current
))
1228 /* Check if the tracer forced the syscall to be skipped. */
1229 this_syscall
= syscall_get_nr(current
, current_pt_regs());
1230 if (this_syscall
< 0)
1234 * Recheck the syscall, since it may have changed. This
1235 * intentionally uses a NULL struct seccomp_data to force
1236 * a reload of all registers. This does not goto skip since
1237 * a skip would have already been reported.
1239 if (__seccomp_filter(this_syscall
, NULL
, true))
1244 case SECCOMP_RET_USER_NOTIF
:
1245 if (seccomp_do_user_notification(this_syscall
, match
, sd
))
1250 case SECCOMP_RET_LOG
:
1251 seccomp_log(this_syscall
, 0, action
, true);
1254 case SECCOMP_RET_ALLOW
:
1256 * Note that the "match" filter will always be NULL for
1257 * this action since SECCOMP_RET_ALLOW is the starting
1258 * state in seccomp_run_filters().
1262 case SECCOMP_RET_KILL_THREAD
:
1263 case SECCOMP_RET_KILL_PROCESS
:
1265 seccomp_log(this_syscall
, SIGSYS
, action
, true);
1266 /* Dump core only if this is the last remaining thread. */
1267 if (action
!= SECCOMP_RET_KILL_THREAD
||
1268 get_nr_threads(current
) == 1) {
1269 kernel_siginfo_t info
;
1271 /* Show the original registers in the dump. */
1272 syscall_rollback(current
, current_pt_regs());
1273 /* Trigger a manual coredump since do_exit skips it. */
1274 seccomp_init_siginfo(&info
, this_syscall
, data
);
1277 if (action
== SECCOMP_RET_KILL_THREAD
)
1280 do_group_exit(SIGSYS
);
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);
1321 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1323 long prctl_get_seccomp(void)
1325 return current
->seccomp
.mode
;
1329 * seccomp_set_mode_strict: internal function for setting strict seccomp
1331 * Once current->seccomp.mode is non-zero, it may not be changed.
1333 * Returns 0 on success or -EINVAL on failure.
1335 static long seccomp_set_mode_strict(void)
1337 const unsigned long seccomp_mode
= SECCOMP_MODE_STRICT
;
1340 spin_lock_irq(¤t
->sighand
->siglock
);
1342 if (!seccomp_may_assign_mode(seccomp_mode
))
1348 seccomp_assign_mode(current
, seccomp_mode
, 0);
1352 spin_unlock_irq(¤t
->sighand
->siglock
);
1357 #ifdef CONFIG_SECCOMP_FILTER
1358 static void seccomp_notify_free(struct seccomp_filter
*filter
)
1360 kfree(filter
->notif
);
1361 filter
->notif
= NULL
;
1364 static void seccomp_notify_detach(struct seccomp_filter
*filter
)
1366 struct seccomp_knotif
*knotif
;
1371 mutex_lock(&filter
->notify_lock
);
1374 * If this file is being closed because e.g. the task who owned it
1375 * died, let's wake everyone up who was waiting on us.
1377 list_for_each_entry(knotif
, &filter
->notif
->notifications
, list
) {
1378 if (knotif
->state
== SECCOMP_NOTIFY_REPLIED
)
1381 knotif
->state
= SECCOMP_NOTIFY_REPLIED
;
1382 knotif
->error
= -ENOSYS
;
1386 * We do not need to wake up any pending addfd messages, as
1387 * the notifier will do that for us, as this just looks
1388 * like a standard reply.
1390 complete(&knotif
->ready
);
1393 seccomp_notify_free(filter
);
1394 mutex_unlock(&filter
->notify_lock
);
1397 static int seccomp_notify_release(struct inode
*inode
, struct file
*file
)
1399 struct seccomp_filter
*filter
= file
->private_data
;
1401 seccomp_notify_detach(filter
);
1402 __put_seccomp_filter(filter
);
1406 /* must be called with notif_lock held */
1407 static inline struct seccomp_knotif
*
1408 find_notification(struct seccomp_filter
*filter
, u64 id
)
1410 struct seccomp_knotif
*cur
;
1412 lockdep_assert_held(&filter
->notify_lock
);
1414 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1423 static long seccomp_notify_recv(struct seccomp_filter
*filter
,
1426 struct seccomp_knotif
*knotif
= NULL
, *cur
;
1427 struct seccomp_notif unotif
;
1430 /* Verify that we're not given garbage to keep struct extensible. */
1431 ret
= check_zeroed_user(buf
, sizeof(unotif
));
1437 memset(&unotif
, 0, sizeof(unotif
));
1439 ret
= down_interruptible(&filter
->notif
->request
);
1443 mutex_lock(&filter
->notify_lock
);
1444 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1445 if (cur
->state
== SECCOMP_NOTIFY_INIT
) {
1452 * If we didn't find a notification, it could be that the task was
1453 * interrupted by a fatal signal between the time we were woken and
1454 * when we were able to acquire the rw lock.
1461 unotif
.id
= knotif
->id
;
1462 unotif
.pid
= task_pid_vnr(knotif
->task
);
1463 unotif
.data
= *(knotif
->data
);
1465 knotif
->state
= SECCOMP_NOTIFY_SENT
;
1466 wake_up_poll(&filter
->wqh
, EPOLLOUT
| EPOLLWRNORM
);
1469 mutex_unlock(&filter
->notify_lock
);
1471 if (ret
== 0 && copy_to_user(buf
, &unotif
, sizeof(unotif
))) {
1475 * Userspace screwed up. To make sure that we keep this
1476 * notification alive, let's reset it back to INIT. It
1477 * may have died when we released the lock, so we need to make
1478 * sure it's still around.
1480 mutex_lock(&filter
->notify_lock
);
1481 knotif
= find_notification(filter
, unotif
.id
);
1483 knotif
->state
= SECCOMP_NOTIFY_INIT
;
1484 up(&filter
->notif
->request
);
1486 mutex_unlock(&filter
->notify_lock
);
1492 static long seccomp_notify_send(struct seccomp_filter
*filter
,
1495 struct seccomp_notif_resp resp
= {};
1496 struct seccomp_knotif
*knotif
;
1499 if (copy_from_user(&resp
, buf
, sizeof(resp
)))
1502 if (resp
.flags
& ~SECCOMP_USER_NOTIF_FLAG_CONTINUE
)
1505 if ((resp
.flags
& SECCOMP_USER_NOTIF_FLAG_CONTINUE
) &&
1506 (resp
.error
|| resp
.val
))
1509 ret
= mutex_lock_interruptible(&filter
->notify_lock
);
1513 knotif
= find_notification(filter
, resp
.id
);
1519 /* Allow exactly one reply. */
1520 if (knotif
->state
!= SECCOMP_NOTIFY_SENT
) {
1526 knotif
->state
= SECCOMP_NOTIFY_REPLIED
;
1527 knotif
->error
= resp
.error
;
1528 knotif
->val
= resp
.val
;
1529 knotif
->flags
= resp
.flags
;
1530 complete(&knotif
->ready
);
1532 mutex_unlock(&filter
->notify_lock
);
1536 static long seccomp_notify_id_valid(struct seccomp_filter
*filter
,
1539 struct seccomp_knotif
*knotif
;
1543 if (copy_from_user(&id
, buf
, sizeof(id
)))
1546 ret
= mutex_lock_interruptible(&filter
->notify_lock
);
1550 knotif
= find_notification(filter
, id
);
1551 if (knotif
&& knotif
->state
== SECCOMP_NOTIFY_SENT
)
1556 mutex_unlock(&filter
->notify_lock
);
1560 static long seccomp_notify_addfd(struct seccomp_filter
*filter
,
1561 struct seccomp_notif_addfd __user
*uaddfd
,
1564 struct seccomp_notif_addfd addfd
;
1565 struct seccomp_knotif
*knotif
;
1566 struct seccomp_kaddfd kaddfd
;
1569 BUILD_BUG_ON(sizeof(addfd
) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0
);
1570 BUILD_BUG_ON(sizeof(addfd
) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST
);
1572 if (size
< SECCOMP_NOTIFY_ADDFD_SIZE_VER0
|| size
>= PAGE_SIZE
)
1575 ret
= copy_struct_from_user(&addfd
, sizeof(addfd
), uaddfd
, size
);
1579 if (addfd
.newfd_flags
& ~O_CLOEXEC
)
1582 if (addfd
.flags
& ~SECCOMP_ADDFD_FLAG_SETFD
)
1585 if (addfd
.newfd
&& !(addfd
.flags
& SECCOMP_ADDFD_FLAG_SETFD
))
1588 kaddfd
.file
= fget(addfd
.srcfd
);
1592 kaddfd
.flags
= addfd
.newfd_flags
;
1593 kaddfd
.setfd
= addfd
.flags
& SECCOMP_ADDFD_FLAG_SETFD
;
1594 kaddfd
.fd
= addfd
.newfd
;
1595 init_completion(&kaddfd
.completion
);
1597 ret
= mutex_lock_interruptible(&filter
->notify_lock
);
1601 knotif
= find_notification(filter
, addfd
.id
);
1608 * We do not want to allow for FD injection to occur before the
1609 * notification has been picked up by a userspace handler, or after
1610 * the notification has been replied to.
1612 if (knotif
->state
!= SECCOMP_NOTIFY_SENT
) {
1617 list_add(&kaddfd
.list
, &knotif
->addfd
);
1618 complete(&knotif
->ready
);
1619 mutex_unlock(&filter
->notify_lock
);
1621 /* Now we wait for it to be processed or be interrupted */
1622 ret
= wait_for_completion_interruptible(&kaddfd
.completion
);
1625 * We had a successful completion. The other side has already
1626 * removed us from the addfd queue, and
1627 * wait_for_completion_interruptible has a memory barrier upon
1628 * success that lets us read this value directly without
1635 mutex_lock(&filter
->notify_lock
);
1637 * Even though we were woken up by a signal and not a successful
1638 * completion, a completion may have happened in the mean time.
1640 * We need to check again if the addfd request has been handled,
1641 * and if not, we will remove it from the queue.
1643 if (list_empty(&kaddfd
.list
))
1646 list_del(&kaddfd
.list
);
1649 mutex_unlock(&filter
->notify_lock
);
1656 static long seccomp_notify_ioctl(struct file
*file
, unsigned int cmd
,
1659 struct seccomp_filter
*filter
= file
->private_data
;
1660 void __user
*buf
= (void __user
*)arg
;
1662 /* Fixed-size ioctls */
1664 case SECCOMP_IOCTL_NOTIF_RECV
:
1665 return seccomp_notify_recv(filter
, buf
);
1666 case SECCOMP_IOCTL_NOTIF_SEND
:
1667 return seccomp_notify_send(filter
, buf
);
1668 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR
:
1669 case SECCOMP_IOCTL_NOTIF_ID_VALID
:
1670 return seccomp_notify_id_valid(filter
, buf
);
1673 /* Extensible Argument ioctls */
1674 #define EA_IOCTL(cmd) ((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1675 switch (EA_IOCTL(cmd
)) {
1676 case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD
):
1677 return seccomp_notify_addfd(filter
, buf
, _IOC_SIZE(cmd
));
1683 static __poll_t
seccomp_notify_poll(struct file
*file
,
1684 struct poll_table_struct
*poll_tab
)
1686 struct seccomp_filter
*filter
= file
->private_data
;
1688 struct seccomp_knotif
*cur
;
1690 poll_wait(file
, &filter
->wqh
, poll_tab
);
1692 if (mutex_lock_interruptible(&filter
->notify_lock
) < 0)
1695 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1696 if (cur
->state
== SECCOMP_NOTIFY_INIT
)
1697 ret
|= EPOLLIN
| EPOLLRDNORM
;
1698 if (cur
->state
== SECCOMP_NOTIFY_SENT
)
1699 ret
|= EPOLLOUT
| EPOLLWRNORM
;
1700 if ((ret
& EPOLLIN
) && (ret
& EPOLLOUT
))
1704 mutex_unlock(&filter
->notify_lock
);
1706 if (refcount_read(&filter
->users
) == 0)
1712 static const struct file_operations seccomp_notify_ops
= {
1713 .poll
= seccomp_notify_poll
,
1714 .release
= seccomp_notify_release
,
1715 .unlocked_ioctl
= seccomp_notify_ioctl
,
1716 .compat_ioctl
= seccomp_notify_ioctl
,
1719 static struct file
*init_listener(struct seccomp_filter
*filter
)
1723 ret
= ERR_PTR(-ENOMEM
);
1724 filter
->notif
= kzalloc(sizeof(*(filter
->notif
)), GFP_KERNEL
);
1728 sema_init(&filter
->notif
->request
, 0);
1729 filter
->notif
->next_id
= get_random_u64();
1730 INIT_LIST_HEAD(&filter
->notif
->notifications
);
1732 ret
= anon_inode_getfile("seccomp notify", &seccomp_notify_ops
,
1737 /* The file has a reference to it now */
1738 __get_seccomp_filter(filter
);
1742 seccomp_notify_free(filter
);
1748 * Does @new_child have a listener while an ancestor also has a listener?
1749 * If so, we'll want to reject this filter.
1750 * This only has to be tested for the current process, even in the TSYNC case,
1751 * because TSYNC installs @child with the same parent on all threads.
1752 * Note that @new_child is not hooked up to its parent at this point yet, so
1753 * we use current->seccomp.filter.
1755 static bool has_duplicate_listener(struct seccomp_filter
*new_child
)
1757 struct seccomp_filter
*cur
;
1759 /* must be protected against concurrent TSYNC */
1760 lockdep_assert_held(¤t
->sighand
->siglock
);
1762 if (!new_child
->notif
)
1764 for (cur
= current
->seccomp
.filter
; cur
; cur
= cur
->prev
) {
1773 * seccomp_set_mode_filter: internal function for setting seccomp filter
1774 * @flags: flags to change filter behavior
1775 * @filter: struct sock_fprog containing filter
1777 * This function may be called repeatedly to install additional filters.
1778 * Every filter successfully installed will be evaluated (in reverse order)
1779 * for each system call the task makes.
1781 * Once current->seccomp.mode is non-zero, it may not be changed.
1783 * Returns 0 on success or -EINVAL on failure.
1785 static long seccomp_set_mode_filter(unsigned int flags
,
1786 const char __user
*filter
)
1788 const unsigned long seccomp_mode
= SECCOMP_MODE_FILTER
;
1789 struct seccomp_filter
*prepared
= NULL
;
1792 struct file
*listener_f
= NULL
;
1794 /* Validate flags. */
1795 if (flags
& ~SECCOMP_FILTER_FLAG_MASK
)
1799 * In the successful case, NEW_LISTENER returns the new listener fd.
1800 * But in the failure case, TSYNC returns the thread that died. If you
1801 * combine these two flags, there's no way to tell whether something
1802 * succeeded or failed. So, let's disallow this combination if the user
1803 * has not explicitly requested no errors from TSYNC.
1805 if ((flags
& SECCOMP_FILTER_FLAG_TSYNC
) &&
1806 (flags
& SECCOMP_FILTER_FLAG_NEW_LISTENER
) &&
1807 ((flags
& SECCOMP_FILTER_FLAG_TSYNC_ESRCH
) == 0))
1810 /* Prepare the new filter before holding any locks. */
1811 prepared
= seccomp_prepare_user_filter(filter
);
1812 if (IS_ERR(prepared
))
1813 return PTR_ERR(prepared
);
1815 if (flags
& SECCOMP_FILTER_FLAG_NEW_LISTENER
) {
1816 listener
= get_unused_fd_flags(O_CLOEXEC
);
1822 listener_f
= init_listener(prepared
);
1823 if (IS_ERR(listener_f
)) {
1824 put_unused_fd(listener
);
1825 ret
= PTR_ERR(listener_f
);
1831 * Make sure we cannot change seccomp or nnp state via TSYNC
1832 * while another thread is in the middle of calling exec.
1834 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
&&
1835 mutex_lock_killable(¤t
->signal
->cred_guard_mutex
))
1838 spin_lock_irq(¤t
->sighand
->siglock
);
1840 if (!seccomp_may_assign_mode(seccomp_mode
))
1843 if (has_duplicate_listener(prepared
)) {
1848 ret
= seccomp_attach_filter(flags
, prepared
);
1851 /* Do not free the successfully attached filter. */
1854 seccomp_assign_mode(current
, seccomp_mode
, flags
);
1856 spin_unlock_irq(¤t
->sighand
->siglock
);
1857 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
)
1858 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1860 if (flags
& SECCOMP_FILTER_FLAG_NEW_LISTENER
) {
1862 listener_f
->private_data
= NULL
;
1864 put_unused_fd(listener
);
1865 seccomp_notify_detach(prepared
);
1867 fd_install(listener
, listener_f
);
1872 seccomp_filter_free(prepared
);
1876 static inline long seccomp_set_mode_filter(unsigned int flags
,
1877 const char __user
*filter
)
1883 static long seccomp_get_action_avail(const char __user
*uaction
)
1887 if (copy_from_user(&action
, uaction
, sizeof(action
)))
1891 case SECCOMP_RET_KILL_PROCESS
:
1892 case SECCOMP_RET_KILL_THREAD
:
1893 case SECCOMP_RET_TRAP
:
1894 case SECCOMP_RET_ERRNO
:
1895 case SECCOMP_RET_USER_NOTIF
:
1896 case SECCOMP_RET_TRACE
:
1897 case SECCOMP_RET_LOG
:
1898 case SECCOMP_RET_ALLOW
:
1907 static long seccomp_get_notif_sizes(void __user
*usizes
)
1909 struct seccomp_notif_sizes sizes
= {
1910 .seccomp_notif
= sizeof(struct seccomp_notif
),
1911 .seccomp_notif_resp
= sizeof(struct seccomp_notif_resp
),
1912 .seccomp_data
= sizeof(struct seccomp_data
),
1915 if (copy_to_user(usizes
, &sizes
, sizeof(sizes
)))
1921 /* Common entry point for both prctl and syscall. */
1922 static long do_seccomp(unsigned int op
, unsigned int flags
,
1926 case SECCOMP_SET_MODE_STRICT
:
1927 if (flags
!= 0 || uargs
!= NULL
)
1929 return seccomp_set_mode_strict();
1930 case SECCOMP_SET_MODE_FILTER
:
1931 return seccomp_set_mode_filter(flags
, uargs
);
1932 case SECCOMP_GET_ACTION_AVAIL
:
1936 return seccomp_get_action_avail(uargs
);
1937 case SECCOMP_GET_NOTIF_SIZES
:
1941 return seccomp_get_notif_sizes(uargs
);
1947 SYSCALL_DEFINE3(seccomp
, unsigned int, op
, unsigned int, flags
,
1948 void __user
*, uargs
)
1950 return do_seccomp(op
, flags
, uargs
);
1954 * prctl_set_seccomp: configures current->seccomp.mode
1955 * @seccomp_mode: requested mode to use
1956 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1958 * Returns 0 on success or -EINVAL on failure.
1960 long prctl_set_seccomp(unsigned long seccomp_mode
, void __user
*filter
)
1965 switch (seccomp_mode
) {
1966 case SECCOMP_MODE_STRICT
:
1967 op
= SECCOMP_SET_MODE_STRICT
;
1969 * Setting strict mode through prctl always ignored filter,
1970 * so make sure it is always NULL here to pass the internal
1971 * check in do_seccomp().
1975 case SECCOMP_MODE_FILTER
:
1976 op
= SECCOMP_SET_MODE_FILTER
;
1983 /* prctl interface doesn't have flags, so they are always zero. */
1984 return do_seccomp(op
, 0, uargs
);
1987 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
1988 static struct seccomp_filter
*get_nth_filter(struct task_struct
*task
,
1989 unsigned long filter_off
)
1991 struct seccomp_filter
*orig
, *filter
;
1992 unsigned long count
;
1995 * Note: this is only correct because the caller should be the (ptrace)
1996 * tracer of the task, otherwise lock_task_sighand is needed.
1998 spin_lock_irq(&task
->sighand
->siglock
);
2000 if (task
->seccomp
.mode
!= SECCOMP_MODE_FILTER
) {
2001 spin_unlock_irq(&task
->sighand
->siglock
);
2002 return ERR_PTR(-EINVAL
);
2005 orig
= task
->seccomp
.filter
;
2006 __get_seccomp_filter(orig
);
2007 spin_unlock_irq(&task
->sighand
->siglock
);
2010 for (filter
= orig
; filter
; filter
= filter
->prev
)
2013 if (filter_off
>= count
) {
2014 filter
= ERR_PTR(-ENOENT
);
2018 count
-= filter_off
;
2019 for (filter
= orig
; filter
&& count
> 1; filter
= filter
->prev
)
2022 if (WARN_ON(count
!= 1 || !filter
)) {
2023 filter
= ERR_PTR(-ENOENT
);
2027 __get_seccomp_filter(filter
);
2030 __put_seccomp_filter(orig
);
2034 long seccomp_get_filter(struct task_struct
*task
, unsigned long filter_off
,
2037 struct seccomp_filter
*filter
;
2038 struct sock_fprog_kern
*fprog
;
2041 if (!capable(CAP_SYS_ADMIN
) ||
2042 current
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
) {
2046 filter
= get_nth_filter(task
, filter_off
);
2048 return PTR_ERR(filter
);
2050 fprog
= filter
->prog
->orig_prog
;
2052 /* This must be a new non-cBPF filter, since we save
2053 * every cBPF filter's orig_prog above when
2054 * CONFIG_CHECKPOINT_RESTORE is enabled.
2064 if (copy_to_user(data
, fprog
->filter
, bpf_classic_proglen(fprog
)))
2068 __put_seccomp_filter(filter
);
2072 long seccomp_get_metadata(struct task_struct
*task
,
2073 unsigned long size
, void __user
*data
)
2076 struct seccomp_filter
*filter
;
2077 struct seccomp_metadata kmd
= {};
2079 if (!capable(CAP_SYS_ADMIN
) ||
2080 current
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
) {
2084 size
= min_t(unsigned long, size
, sizeof(kmd
));
2086 if (size
< sizeof(kmd
.filter_off
))
2089 if (copy_from_user(&kmd
.filter_off
, data
, sizeof(kmd
.filter_off
)))
2092 filter
= get_nth_filter(task
, kmd
.filter_off
);
2094 return PTR_ERR(filter
);
2097 kmd
.flags
|= SECCOMP_FILTER_FLAG_LOG
;
2100 if (copy_to_user(data
, &kmd
, size
))
2103 __put_seccomp_filter(filter
);
2108 #ifdef CONFIG_SYSCTL
2110 /* Human readable action names for friendly sysctl interaction */
2111 #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
2112 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
2113 #define SECCOMP_RET_TRAP_NAME "trap"
2114 #define SECCOMP_RET_ERRNO_NAME "errno"
2115 #define SECCOMP_RET_USER_NOTIF_NAME "user_notif"
2116 #define SECCOMP_RET_TRACE_NAME "trace"
2117 #define SECCOMP_RET_LOG_NAME "log"
2118 #define SECCOMP_RET_ALLOW_NAME "allow"
2120 static const char seccomp_actions_avail
[] =
2121 SECCOMP_RET_KILL_PROCESS_NAME
" "
2122 SECCOMP_RET_KILL_THREAD_NAME
" "
2123 SECCOMP_RET_TRAP_NAME
" "
2124 SECCOMP_RET_ERRNO_NAME
" "
2125 SECCOMP_RET_USER_NOTIF_NAME
" "
2126 SECCOMP_RET_TRACE_NAME
" "
2127 SECCOMP_RET_LOG_NAME
" "
2128 SECCOMP_RET_ALLOW_NAME
;
2130 struct seccomp_log_name
{
2135 static const struct seccomp_log_name seccomp_log_names
[] = {
2136 { SECCOMP_LOG_KILL_PROCESS
, SECCOMP_RET_KILL_PROCESS_NAME
},
2137 { SECCOMP_LOG_KILL_THREAD
, SECCOMP_RET_KILL_THREAD_NAME
},
2138 { SECCOMP_LOG_TRAP
, SECCOMP_RET_TRAP_NAME
},
2139 { SECCOMP_LOG_ERRNO
, SECCOMP_RET_ERRNO_NAME
},
2140 { SECCOMP_LOG_USER_NOTIF
, SECCOMP_RET_USER_NOTIF_NAME
},
2141 { SECCOMP_LOG_TRACE
, SECCOMP_RET_TRACE_NAME
},
2142 { SECCOMP_LOG_LOG
, SECCOMP_RET_LOG_NAME
},
2143 { SECCOMP_LOG_ALLOW
, SECCOMP_RET_ALLOW_NAME
},
2147 static bool seccomp_names_from_actions_logged(char *names
, size_t size
,
2151 const struct seccomp_log_name
*cur
;
2152 bool append_sep
= false;
2154 for (cur
= seccomp_log_names
; cur
->name
&& size
; cur
++) {
2157 if (!(actions_logged
& cur
->log
))
2161 ret
= strscpy(names
, sep
, size
);
2170 ret
= strscpy(names
, cur
->name
, size
);
2181 static bool seccomp_action_logged_from_name(u32
*action_logged
,
2184 const struct seccomp_log_name
*cur
;
2186 for (cur
= seccomp_log_names
; cur
->name
; cur
++) {
2187 if (!strcmp(cur
->name
, name
)) {
2188 *action_logged
= cur
->log
;
2196 static bool seccomp_actions_logged_from_names(u32
*actions_logged
, char *names
)
2200 *actions_logged
= 0;
2201 while ((name
= strsep(&names
, " ")) && *name
) {
2202 u32 action_logged
= 0;
2204 if (!seccomp_action_logged_from_name(&action_logged
, name
))
2207 *actions_logged
|= action_logged
;
2213 static int read_actions_logged(struct ctl_table
*ro_table
, void *buffer
,
2214 size_t *lenp
, loff_t
*ppos
)
2216 char names
[sizeof(seccomp_actions_avail
)];
2217 struct ctl_table table
;
2219 memset(names
, 0, sizeof(names
));
2221 if (!seccomp_names_from_actions_logged(names
, sizeof(names
),
2222 seccomp_actions_logged
, " "))
2227 table
.maxlen
= sizeof(names
);
2228 return proc_dostring(&table
, 0, buffer
, lenp
, ppos
);
2231 static int write_actions_logged(struct ctl_table
*ro_table
, void *buffer
,
2232 size_t *lenp
, loff_t
*ppos
, u32
*actions_logged
)
2234 char names
[sizeof(seccomp_actions_avail
)];
2235 struct ctl_table table
;
2238 if (!capable(CAP_SYS_ADMIN
))
2241 memset(names
, 0, sizeof(names
));
2245 table
.maxlen
= sizeof(names
);
2246 ret
= proc_dostring(&table
, 1, buffer
, lenp
, ppos
);
2250 if (!seccomp_actions_logged_from_names(actions_logged
, table
.data
))
2253 if (*actions_logged
& SECCOMP_LOG_ALLOW
)
2256 seccomp_actions_logged
= *actions_logged
;
2260 static void audit_actions_logged(u32 actions_logged
, u32 old_actions_logged
,
2263 char names
[sizeof(seccomp_actions_avail
)];
2264 char old_names
[sizeof(seccomp_actions_avail
)];
2265 const char *new = names
;
2266 const char *old
= old_names
;
2271 memset(names
, 0, sizeof(names
));
2272 memset(old_names
, 0, sizeof(old_names
));
2276 else if (!actions_logged
)
2278 else if (!seccomp_names_from_actions_logged(names
, sizeof(names
),
2279 actions_logged
, ","))
2282 if (!old_actions_logged
)
2284 else if (!seccomp_names_from_actions_logged(old_names
,
2286 old_actions_logged
, ","))
2289 return audit_seccomp_actions_logged(new, old
, !ret
);
2292 static int seccomp_actions_logged_handler(struct ctl_table
*ro_table
, int write
,
2293 void *buffer
, size_t *lenp
,
2299 u32 actions_logged
= 0;
2300 u32 old_actions_logged
= seccomp_actions_logged
;
2302 ret
= write_actions_logged(ro_table
, buffer
, lenp
, ppos
,
2304 audit_actions_logged(actions_logged
, old_actions_logged
, ret
);
2306 ret
= read_actions_logged(ro_table
, buffer
, lenp
, ppos
);
2311 static struct ctl_path seccomp_sysctl_path
[] = {
2312 { .procname
= "kernel", },
2313 { .procname
= "seccomp", },
2317 static struct ctl_table seccomp_sysctl_table
[] = {
2319 .procname
= "actions_avail",
2320 .data
= (void *) &seccomp_actions_avail
,
2321 .maxlen
= sizeof(seccomp_actions_avail
),
2323 .proc_handler
= proc_dostring
,
2326 .procname
= "actions_logged",
2328 .proc_handler
= seccomp_actions_logged_handler
,
2333 static int __init
seccomp_sysctl_init(void)
2335 struct ctl_table_header
*hdr
;
2337 hdr
= register_sysctl_paths(seccomp_sysctl_path
, seccomp_sysctl_table
);
2339 pr_warn("sysctl registration failed\n");
2341 kmemleak_not_leak(hdr
);
2346 device_initcall(seccomp_sysctl_init
)
2348 #endif /* CONFIG_SYSCTL */
2350 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
2351 /* Currently CONFIG_SECCOMP_CACHE_DEBUG implies SECCOMP_ARCH_NATIVE */
2352 static void proc_pid_seccomp_cache_arch(struct seq_file
*m
, const char *name
,
2353 const void *bitmap
, size_t bitmap_size
)
2357 for (nr
= 0; nr
< bitmap_size
; nr
++) {
2358 bool cached
= test_bit(nr
, bitmap
);
2359 char *status
= cached
? "ALLOW" : "FILTER";
2361 seq_printf(m
, "%s %d %s\n", name
, nr
, status
);
2365 int proc_pid_seccomp_cache(struct seq_file
*m
, struct pid_namespace
*ns
,
2366 struct pid
*pid
, struct task_struct
*task
)
2368 struct seccomp_filter
*f
;
2369 unsigned long flags
;
2372 * We don't want some sandboxed process to know what their seccomp
2373 * filters consist of.
2375 if (!file_ns_capable(m
->file
, &init_user_ns
, CAP_SYS_ADMIN
))
2378 if (!lock_task_sighand(task
, &flags
))
2381 f
= READ_ONCE(task
->seccomp
.filter
);
2383 unlock_task_sighand(task
, &flags
);
2387 /* prevent filter from being freed while we are printing it */
2388 __get_seccomp_filter(f
);
2389 unlock_task_sighand(task
, &flags
);
2391 proc_pid_seccomp_cache_arch(m
, SECCOMP_ARCH_NATIVE_NAME
,
2392 f
->cache
.allow_native
,
2393 SECCOMP_ARCH_NATIVE_NR
);
2395 #ifdef SECCOMP_ARCH_COMPAT
2396 proc_pid_seccomp_cache_arch(m
, SECCOMP_ARCH_COMPAT_NAME
,
2397 f
->cache
.allow_compat
,
2398 SECCOMP_ARCH_COMPAT_NR
);
2399 #endif /* SECCOMP_ARCH_COMPAT */
2401 __put_seccomp_filter(f
);
2404 #endif /* CONFIG_SECCOMP_CACHE_DEBUG */