1 // SPDX-License-Identifier: GPL-2.0
3 * linux/kernel/seccomp.c
5 * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com>
7 * Copyright (C) 2012 Google, Inc.
8 * Will Drewry <wad@chromium.org>
10 * This defines a simple but solid secure-computing facility.
12 * Mode 1 uses a fixed list of allowed system calls.
13 * Mode 2 allows user-defined system call filters in the form
14 * of Berkeley Packet Filters/Linux Socket Filters.
16 #define pr_fmt(fmt) "seccomp: " fmt
18 #include <linux/refcount.h>
19 #include <linux/audit.h>
20 #include <linux/compat.h>
21 #include <linux/coredump.h>
22 #include <linux/kmemleak.h>
23 #include <linux/nospec.h>
24 #include <linux/prctl.h>
25 #include <linux/sched.h>
26 #include <linux/sched/task_stack.h>
27 #include <linux/seccomp.h>
28 #include <linux/slab.h>
29 #include <linux/syscalls.h>
30 #include <linux/sysctl.h>
32 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
33 #include <asm/syscall.h>
36 #ifdef CONFIG_SECCOMP_FILTER
37 #include <linux/file.h>
38 #include <linux/filter.h>
39 #include <linux/pid.h>
40 #include <linux/ptrace.h>
41 #include <linux/security.h>
42 #include <linux/tracehook.h>
43 #include <linux/uaccess.h>
44 #include <linux/anon_inodes.h>
45 #include <linux/lockdep.h>
48 * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the
49 * wrong direction flag in the ioctl number. This is the broken one,
50 * which the kernel needs to keep supporting until all userspaces stop
51 * using the wrong command number.
53 #define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR SECCOMP_IOR(2, __u64)
58 SECCOMP_NOTIFY_REPLIED
,
61 struct seccomp_knotif
{
62 /* The struct pid of the task whose filter triggered the notification */
63 struct task_struct
*task
;
65 /* The "cookie" for this request; this is unique for this filter. */
69 * The seccomp data. This pointer is valid the entire time this
70 * notification is active, since it comes from __seccomp_filter which
71 * eclipses the entire lifecycle here.
73 const struct seccomp_data
*data
;
76 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
77 * struct seccomp_knotif is created and starts out in INIT. Once the
78 * handler reads the notification off of an FD, it transitions to SENT.
79 * If a signal is received the state transitions back to INIT and
80 * another message is sent. When the userspace handler replies, state
81 * transitions to REPLIED.
83 enum notify_state state
;
85 /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
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
{
122 /* To only be set on reply */
124 struct completion completion
;
125 struct list_head list
;
129 * struct notification - container for seccomp userspace notifications. Since
130 * most seccomp filters will not have notification listeners attached and this
131 * structure is fairly large, we store the notification-specific stuff in a
132 * separate structure.
134 * @request: A semaphore that users of this notification can wait on for
135 * changes. Actual reads and writes are still controlled with
136 * filter->notify_lock.
137 * @next_id: The id of the next request.
138 * @notifications: A list of struct seccomp_knotif elements.
140 struct notification
{
141 struct semaphore request
;
143 struct list_head notifications
;
147 * struct seccomp_filter - container for seccomp BPF programs
149 * @refs: Reference count to manage the object lifetime.
150 * A filter's reference count is incremented for each directly
151 * attached task, once for the dependent filter, and if
152 * requested for the user notifier. When @refs reaches zero,
153 * the filter can be freed.
154 * @users: A filter's @users count is incremented for each directly
155 * attached task (filter installation, fork(), thread_sync),
156 * and once for the dependent filter (tracked in filter->prev).
157 * When it reaches zero it indicates that no direct or indirect
158 * users of that filter exist. No new tasks can get associated with
159 * this filter after reaching 0. The @users count is always smaller
160 * or equal to @refs. Hence, reaching 0 for @users does not mean
161 * the filter can be freed.
162 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
163 * @prev: points to a previously installed, or inherited, filter
164 * @prog: the BPF program to evaluate
165 * @notif: the struct that holds all notification related information
166 * @notify_lock: A lock for all notification-related accesses.
167 * @wqh: A wait queue for poll if a notifier is in use.
169 * seccomp_filter objects are organized in a tree linked via the @prev
170 * pointer. For any task, it appears to be a singly-linked list starting
171 * with current->seccomp.filter, the most recently attached or inherited filter.
172 * However, multiple filters may share a @prev node, by way of fork(), which
173 * results in a unidirectional tree existing in memory. This is similar to
174 * how namespaces work.
176 * seccomp_filter objects should never be modified after being attached
177 * to a task_struct (other than @refs).
179 struct seccomp_filter
{
183 struct seccomp_filter
*prev
;
184 struct bpf_prog
*prog
;
185 struct notification
*notif
;
186 struct mutex notify_lock
;
187 wait_queue_head_t wqh
;
190 /* Limit any path through the tree to 256KB worth of instructions. */
191 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
194 * Endianness is explicitly ignored and left for BPF program authors to manage
195 * as per the specific architecture.
197 static void populate_seccomp_data(struct seccomp_data
*sd
)
200 * Instead of using current_pt_reg(), we're already doing the work
201 * to safely fetch "current", so just use "task" everywhere below.
203 struct task_struct
*task
= current
;
204 struct pt_regs
*regs
= task_pt_regs(task
);
205 unsigned long args
[6];
207 sd
->nr
= syscall_get_nr(task
, regs
);
208 sd
->arch
= syscall_get_arch(task
);
209 syscall_get_arguments(task
, regs
, args
);
210 sd
->args
[0] = args
[0];
211 sd
->args
[1] = args
[1];
212 sd
->args
[2] = args
[2];
213 sd
->args
[3] = args
[3];
214 sd
->args
[4] = args
[4];
215 sd
->args
[5] = args
[5];
216 sd
->instruction_pointer
= KSTK_EIP(task
);
220 * seccomp_check_filter - verify seccomp filter code
221 * @filter: filter to verify
222 * @flen: length of filter
224 * Takes a previously checked filter (by bpf_check_classic) and
225 * redirects all filter code that loads struct sk_buff data
226 * and related data through seccomp_bpf_load. It also
227 * enforces length and alignment checking of those loads.
229 * Returns 0 if the rule set is legal or -EINVAL if not.
231 static int seccomp_check_filter(struct sock_filter
*filter
, unsigned int flen
)
234 for (pc
= 0; pc
< flen
; pc
++) {
235 struct sock_filter
*ftest
= &filter
[pc
];
236 u16 code
= ftest
->code
;
240 case BPF_LD
| BPF_W
| BPF_ABS
:
241 ftest
->code
= BPF_LDX
| BPF_W
| BPF_ABS
;
242 /* 32-bit aligned and not out of bounds. */
243 if (k
>= sizeof(struct seccomp_data
) || k
& 3)
246 case BPF_LD
| BPF_W
| BPF_LEN
:
247 ftest
->code
= BPF_LD
| BPF_IMM
;
248 ftest
->k
= sizeof(struct seccomp_data
);
250 case BPF_LDX
| BPF_W
| BPF_LEN
:
251 ftest
->code
= BPF_LDX
| BPF_IMM
;
252 ftest
->k
= sizeof(struct seccomp_data
);
254 /* Explicitly include allowed calls. */
255 case BPF_RET
| BPF_K
:
256 case BPF_RET
| BPF_A
:
257 case BPF_ALU
| BPF_ADD
| BPF_K
:
258 case BPF_ALU
| BPF_ADD
| BPF_X
:
259 case BPF_ALU
| BPF_SUB
| BPF_K
:
260 case BPF_ALU
| BPF_SUB
| BPF_X
:
261 case BPF_ALU
| BPF_MUL
| BPF_K
:
262 case BPF_ALU
| BPF_MUL
| BPF_X
:
263 case BPF_ALU
| BPF_DIV
| BPF_K
:
264 case BPF_ALU
| BPF_DIV
| BPF_X
:
265 case BPF_ALU
| BPF_AND
| BPF_K
:
266 case BPF_ALU
| BPF_AND
| BPF_X
:
267 case BPF_ALU
| BPF_OR
| BPF_K
:
268 case BPF_ALU
| BPF_OR
| BPF_X
:
269 case BPF_ALU
| BPF_XOR
| BPF_K
:
270 case BPF_ALU
| BPF_XOR
| BPF_X
:
271 case BPF_ALU
| BPF_LSH
| BPF_K
:
272 case BPF_ALU
| BPF_LSH
| BPF_X
:
273 case BPF_ALU
| BPF_RSH
| BPF_K
:
274 case BPF_ALU
| BPF_RSH
| BPF_X
:
275 case BPF_ALU
| BPF_NEG
:
276 case BPF_LD
| BPF_IMM
:
277 case BPF_LDX
| BPF_IMM
:
278 case BPF_MISC
| BPF_TAX
:
279 case BPF_MISC
| BPF_TXA
:
280 case BPF_LD
| BPF_MEM
:
281 case BPF_LDX
| BPF_MEM
:
284 case BPF_JMP
| BPF_JA
:
285 case BPF_JMP
| BPF_JEQ
| BPF_K
:
286 case BPF_JMP
| BPF_JEQ
| BPF_X
:
287 case BPF_JMP
| BPF_JGE
| BPF_K
:
288 case BPF_JMP
| BPF_JGE
| BPF_X
:
289 case BPF_JMP
| BPF_JGT
| BPF_K
:
290 case BPF_JMP
| BPF_JGT
| BPF_X
:
291 case BPF_JMP
| BPF_JSET
| BPF_K
:
292 case BPF_JMP
| BPF_JSET
| BPF_X
:
302 * seccomp_run_filters - evaluates all seccomp filters against @sd
303 * @sd: optional seccomp data to be passed to filters
304 * @match: stores struct seccomp_filter that resulted in the return value,
305 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
308 * Returns valid seccomp BPF response codes.
310 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
311 static u32
seccomp_run_filters(const struct seccomp_data
*sd
,
312 struct seccomp_filter
**match
)
314 u32 ret
= SECCOMP_RET_ALLOW
;
315 /* Make sure cross-thread synced filter points somewhere sane. */
316 struct seccomp_filter
*f
=
317 READ_ONCE(current
->seccomp
.filter
);
319 /* Ensure unexpected behavior doesn't result in failing open. */
320 if (WARN_ON(f
== NULL
))
321 return SECCOMP_RET_KILL_PROCESS
;
324 * All filters in the list are evaluated and the lowest BPF return
325 * value always takes priority (ignoring the DATA).
327 for (; f
; f
= f
->prev
) {
328 u32 cur_ret
= bpf_prog_run_pin_on_cpu(f
->prog
, sd
);
330 if (ACTION_ONLY(cur_ret
) < ACTION_ONLY(ret
)) {
337 #endif /* CONFIG_SECCOMP_FILTER */
339 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode
)
341 assert_spin_locked(¤t
->sighand
->siglock
);
343 if (current
->seccomp
.mode
&& current
->seccomp
.mode
!= seccomp_mode
)
349 void __weak
arch_seccomp_spec_mitigate(struct task_struct
*task
) { }
351 static inline void seccomp_assign_mode(struct task_struct
*task
,
352 unsigned long seccomp_mode
,
355 assert_spin_locked(&task
->sighand
->siglock
);
357 task
->seccomp
.mode
= seccomp_mode
;
359 * Make sure TIF_SECCOMP cannot be set before the mode (and
362 smp_mb__before_atomic();
363 /* Assume default seccomp processes want spec flaw mitigation. */
364 if ((flags
& SECCOMP_FILTER_FLAG_SPEC_ALLOW
) == 0)
365 arch_seccomp_spec_mitigate(task
);
366 set_tsk_thread_flag(task
, TIF_SECCOMP
);
369 #ifdef CONFIG_SECCOMP_FILTER
370 /* Returns 1 if the parent is an ancestor of the child. */
371 static int is_ancestor(struct seccomp_filter
*parent
,
372 struct seccomp_filter
*child
)
374 /* NULL is the root ancestor. */
377 for (; child
; child
= child
->prev
)
384 * seccomp_can_sync_threads: checks if all threads can be synchronized
386 * Expects sighand and cred_guard_mutex locks to be held.
388 * Returns 0 on success, -ve on error, or the pid of a thread which was
389 * either not in the correct seccomp mode or did not have an ancestral
392 static inline pid_t
seccomp_can_sync_threads(void)
394 struct task_struct
*thread
, *caller
;
396 BUG_ON(!mutex_is_locked(¤t
->signal
->cred_guard_mutex
));
397 assert_spin_locked(¤t
->sighand
->siglock
);
399 /* Validate all threads being eligible for synchronization. */
401 for_each_thread(caller
, thread
) {
404 /* Skip current, since it is initiating the sync. */
405 if (thread
== caller
)
408 if (thread
->seccomp
.mode
== SECCOMP_MODE_DISABLED
||
409 (thread
->seccomp
.mode
== SECCOMP_MODE_FILTER
&&
410 is_ancestor(thread
->seccomp
.filter
,
411 caller
->seccomp
.filter
)))
414 /* Return the first thread that cannot be synchronized. */
415 failed
= task_pid_vnr(thread
);
416 /* If the pid cannot be resolved, then return -ESRCH */
417 if (WARN_ON(failed
== 0))
425 static inline void seccomp_filter_free(struct seccomp_filter
*filter
)
428 bpf_prog_destroy(filter
->prog
);
433 static void __seccomp_filter_orphan(struct seccomp_filter
*orig
)
435 while (orig
&& refcount_dec_and_test(&orig
->users
)) {
436 if (waitqueue_active(&orig
->wqh
))
437 wake_up_poll(&orig
->wqh
, EPOLLHUP
);
442 static void __put_seccomp_filter(struct seccomp_filter
*orig
)
444 /* Clean up single-reference branches iteratively. */
445 while (orig
&& refcount_dec_and_test(&orig
->refs
)) {
446 struct seccomp_filter
*freeme
= orig
;
448 seccomp_filter_free(freeme
);
452 static void __seccomp_filter_release(struct seccomp_filter
*orig
)
454 /* Notify about any unused filters in the task's former filter tree. */
455 __seccomp_filter_orphan(orig
);
456 /* Finally drop all references to the task's former tree. */
457 __put_seccomp_filter(orig
);
461 * seccomp_filter_release - Detach the task from its filter tree,
462 * drop its reference count, and notify
463 * about unused filters
465 * This function should only be called when the task is exiting as
466 * it detaches it from its filter tree. As such, READ_ONCE() and
467 * barriers are not needed here, as would normally be needed.
469 void seccomp_filter_release(struct task_struct
*tsk
)
471 struct seccomp_filter
*orig
= tsk
->seccomp
.filter
;
473 /* Detach task from its filter tree. */
474 tsk
->seccomp
.filter
= NULL
;
475 __seccomp_filter_release(orig
);
479 * seccomp_sync_threads: sets all threads to use current's filter
481 * Expects sighand and cred_guard_mutex locks to be held, and for
482 * seccomp_can_sync_threads() to have returned success already
483 * without dropping the locks.
486 static inline void seccomp_sync_threads(unsigned long flags
)
488 struct task_struct
*thread
, *caller
;
490 BUG_ON(!mutex_is_locked(¤t
->signal
->cred_guard_mutex
));
491 assert_spin_locked(¤t
->sighand
->siglock
);
493 /* Synchronize all threads. */
495 for_each_thread(caller
, thread
) {
496 /* Skip current, since it needs no changes. */
497 if (thread
== caller
)
500 /* Get a task reference for the new leaf node. */
501 get_seccomp_filter(caller
);
504 * Drop the task reference to the shared ancestor since
505 * current's path will hold a reference. (This also
506 * allows a put before the assignment.)
508 __seccomp_filter_release(thread
->seccomp
.filter
);
510 /* Make our new filter tree visible. */
511 smp_store_release(&thread
->seccomp
.filter
,
512 caller
->seccomp
.filter
);
513 atomic_set(&thread
->seccomp
.filter_count
,
514 atomic_read(&thread
->seccomp
.filter_count
));
517 * Don't let an unprivileged task work around
518 * the no_new_privs restriction by creating
519 * a thread that sets it up, enters seccomp,
522 if (task_no_new_privs(caller
))
523 task_set_no_new_privs(thread
);
526 * Opt the other thread into seccomp if needed.
527 * As threads are considered to be trust-realm
528 * equivalent (see ptrace_may_access), it is safe to
529 * allow one thread to transition the other.
531 if (thread
->seccomp
.mode
== SECCOMP_MODE_DISABLED
)
532 seccomp_assign_mode(thread
, SECCOMP_MODE_FILTER
,
538 * seccomp_prepare_filter: Prepares a seccomp filter for use.
539 * @fprog: BPF program to install
541 * Returns filter on success or an ERR_PTR on failure.
543 static struct seccomp_filter
*seccomp_prepare_filter(struct sock_fprog
*fprog
)
545 struct seccomp_filter
*sfilter
;
547 const bool save_orig
= IS_ENABLED(CONFIG_CHECKPOINT_RESTORE
);
549 if (fprog
->len
== 0 || fprog
->len
> BPF_MAXINSNS
)
550 return ERR_PTR(-EINVAL
);
552 BUG_ON(INT_MAX
/ fprog
->len
< sizeof(struct sock_filter
));
555 * Installing a seccomp filter requires that the task has
556 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
557 * This avoids scenarios where unprivileged tasks can affect the
558 * behavior of privileged children.
560 if (!task_no_new_privs(current
) &&
561 security_capable(current_cred(), current_user_ns(),
562 CAP_SYS_ADMIN
, CAP_OPT_NOAUDIT
) != 0)
563 return ERR_PTR(-EACCES
);
565 /* Allocate a new seccomp_filter */
566 sfilter
= kzalloc(sizeof(*sfilter
), GFP_KERNEL
| __GFP_NOWARN
);
568 return ERR_PTR(-ENOMEM
);
570 mutex_init(&sfilter
->notify_lock
);
571 ret
= bpf_prog_create_from_user(&sfilter
->prog
, fprog
,
572 seccomp_check_filter
, save_orig
);
578 refcount_set(&sfilter
->refs
, 1);
579 refcount_set(&sfilter
->users
, 1);
580 init_waitqueue_head(&sfilter
->wqh
);
586 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
587 * @user_filter: pointer to the user data containing a sock_fprog.
589 * Returns 0 on success and non-zero otherwise.
591 static struct seccomp_filter
*
592 seccomp_prepare_user_filter(const char __user
*user_filter
)
594 struct sock_fprog fprog
;
595 struct seccomp_filter
*filter
= ERR_PTR(-EFAULT
);
598 if (in_compat_syscall()) {
599 struct compat_sock_fprog fprog32
;
600 if (copy_from_user(&fprog32
, user_filter
, sizeof(fprog32
)))
602 fprog
.len
= fprog32
.len
;
603 fprog
.filter
= compat_ptr(fprog32
.filter
);
604 } else /* falls through to the if below. */
606 if (copy_from_user(&fprog
, user_filter
, sizeof(fprog
)))
608 filter
= seccomp_prepare_filter(&fprog
);
614 * seccomp_attach_filter: validate and attach filter
615 * @flags: flags to change filter behavior
616 * @filter: seccomp filter to add to the current process
618 * Caller must be holding current->sighand->siglock lock.
620 * Returns 0 on success, -ve on error, or
621 * - in TSYNC mode: the pid of a thread which was either not in the correct
622 * seccomp mode or did not have an ancestral seccomp filter
623 * - in NEW_LISTENER mode: the fd of the new listener
625 static long seccomp_attach_filter(unsigned int flags
,
626 struct seccomp_filter
*filter
)
628 unsigned long total_insns
;
629 struct seccomp_filter
*walker
;
631 assert_spin_locked(¤t
->sighand
->siglock
);
633 /* Validate resulting filter length. */
634 total_insns
= filter
->prog
->len
;
635 for (walker
= current
->seccomp
.filter
; walker
; walker
= walker
->prev
)
636 total_insns
+= walker
->prog
->len
+ 4; /* 4 instr penalty */
637 if (total_insns
> MAX_INSNS_PER_PATH
)
640 /* If thread sync has been requested, check that it is possible. */
641 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
) {
644 ret
= seccomp_can_sync_threads();
646 if (flags
& SECCOMP_FILTER_FLAG_TSYNC_ESRCH
)
653 /* Set log flag, if present. */
654 if (flags
& SECCOMP_FILTER_FLAG_LOG
)
658 * If there is an existing filter, make it the prev and don't drop its
661 filter
->prev
= current
->seccomp
.filter
;
662 current
->seccomp
.filter
= filter
;
663 atomic_inc(¤t
->seccomp
.filter_count
);
665 /* Now that the new filter is in place, synchronize to all threads. */
666 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
)
667 seccomp_sync_threads(flags
);
672 static void __get_seccomp_filter(struct seccomp_filter
*filter
)
674 refcount_inc(&filter
->refs
);
677 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
678 void get_seccomp_filter(struct task_struct
*tsk
)
680 struct seccomp_filter
*orig
= tsk
->seccomp
.filter
;
683 __get_seccomp_filter(orig
);
684 refcount_inc(&orig
->users
);
687 static void seccomp_init_siginfo(kernel_siginfo_t
*info
, int syscall
, int reason
)
690 info
->si_signo
= SIGSYS
;
691 info
->si_code
= SYS_SECCOMP
;
692 info
->si_call_addr
= (void __user
*)KSTK_EIP(current
);
693 info
->si_errno
= reason
;
694 info
->si_arch
= syscall_get_arch(current
);
695 info
->si_syscall
= syscall
;
699 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
700 * @syscall: syscall number to send to userland
701 * @reason: filter-supplied reason code to send to userland (via si_errno)
703 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
705 static void seccomp_send_sigsys(int syscall
, int reason
)
707 struct kernel_siginfo info
;
708 seccomp_init_siginfo(&info
, syscall
, reason
);
709 force_sig_info(&info
);
711 #endif /* CONFIG_SECCOMP_FILTER */
713 /* For use with seccomp_actions_logged */
714 #define SECCOMP_LOG_KILL_PROCESS (1 << 0)
715 #define SECCOMP_LOG_KILL_THREAD (1 << 1)
716 #define SECCOMP_LOG_TRAP (1 << 2)
717 #define SECCOMP_LOG_ERRNO (1 << 3)
718 #define SECCOMP_LOG_TRACE (1 << 4)
719 #define SECCOMP_LOG_LOG (1 << 5)
720 #define SECCOMP_LOG_ALLOW (1 << 6)
721 #define SECCOMP_LOG_USER_NOTIF (1 << 7)
723 static u32 seccomp_actions_logged
= SECCOMP_LOG_KILL_PROCESS
|
724 SECCOMP_LOG_KILL_THREAD
|
727 SECCOMP_LOG_USER_NOTIF
|
731 static inline void seccomp_log(unsigned long syscall
, long signr
, u32 action
,
737 case SECCOMP_RET_ALLOW
:
739 case SECCOMP_RET_TRAP
:
740 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_TRAP
;
742 case SECCOMP_RET_ERRNO
:
743 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_ERRNO
;
745 case SECCOMP_RET_TRACE
:
746 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_TRACE
;
748 case SECCOMP_RET_USER_NOTIF
:
749 log
= requested
&& seccomp_actions_logged
& SECCOMP_LOG_USER_NOTIF
;
751 case SECCOMP_RET_LOG
:
752 log
= seccomp_actions_logged
& SECCOMP_LOG_LOG
;
754 case SECCOMP_RET_KILL_THREAD
:
755 log
= seccomp_actions_logged
& SECCOMP_LOG_KILL_THREAD
;
757 case SECCOMP_RET_KILL_PROCESS
:
759 log
= seccomp_actions_logged
& SECCOMP_LOG_KILL_PROCESS
;
763 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
764 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
765 * any action from being logged by removing the action name from the
766 * seccomp_actions_logged sysctl.
771 audit_seccomp(syscall
, signr
, action
);
775 * Secure computing mode 1 allows only read/write/exit/sigreturn.
776 * To be fully secure this must be combined with rlimit
777 * to limit the stack allocations too.
779 static const int mode1_syscalls
[] = {
780 __NR_seccomp_read
, __NR_seccomp_write
, __NR_seccomp_exit
, __NR_seccomp_sigreturn
,
781 -1, /* negative terminated */
784 static void __secure_computing_strict(int this_syscall
)
786 const int *allowed_syscalls
= mode1_syscalls
;
788 if (in_compat_syscall())
789 allowed_syscalls
= get_compat_mode1_syscalls();
792 if (*allowed_syscalls
== this_syscall
)
794 } while (*++allowed_syscalls
!= -1);
799 seccomp_log(this_syscall
, SIGKILL
, SECCOMP_RET_KILL_THREAD
, true);
803 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
804 void secure_computing_strict(int this_syscall
)
806 int mode
= current
->seccomp
.mode
;
808 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE
) &&
809 unlikely(current
->ptrace
& PT_SUSPEND_SECCOMP
))
812 if (mode
== SECCOMP_MODE_DISABLED
)
814 else if (mode
== SECCOMP_MODE_STRICT
)
815 __secure_computing_strict(this_syscall
);
821 #ifdef CONFIG_SECCOMP_FILTER
822 static u64
seccomp_next_notify_id(struct seccomp_filter
*filter
)
825 * Note: overflow is ok here, the id just needs to be unique per
828 lockdep_assert_held(&filter
->notify_lock
);
829 return filter
->notif
->next_id
++;
832 static void seccomp_handle_addfd(struct seccomp_kaddfd
*addfd
)
835 * Remove the notification, and reset the list pointers, indicating
836 * that it has been handled.
838 list_del_init(&addfd
->list
);
839 addfd
->ret
= receive_fd_replace(addfd
->fd
, addfd
->file
, addfd
->flags
);
840 complete(&addfd
->completion
);
843 static int seccomp_do_user_notification(int this_syscall
,
844 struct seccomp_filter
*match
,
845 const struct seccomp_data
*sd
)
850 struct seccomp_knotif n
= {};
851 struct seccomp_kaddfd
*addfd
, *tmp
;
853 mutex_lock(&match
->notify_lock
);
859 n
.state
= SECCOMP_NOTIFY_INIT
;
861 n
.id
= seccomp_next_notify_id(match
);
862 init_completion(&n
.ready
);
863 list_add(&n
.list
, &match
->notif
->notifications
);
864 INIT_LIST_HEAD(&n
.addfd
);
866 up(&match
->notif
->request
);
867 wake_up_poll(&match
->wqh
, EPOLLIN
| EPOLLRDNORM
);
868 mutex_unlock(&match
->notify_lock
);
871 * This is where we wait for a reply from userspace.
874 err
= wait_for_completion_interruptible(&n
.ready
);
875 mutex_lock(&match
->notify_lock
);
877 /* Check if we were woken up by a addfd message */
878 addfd
= list_first_entry_or_null(&n
.addfd
,
879 struct seccomp_kaddfd
, list
);
880 if (addfd
&& n
.state
!= SECCOMP_NOTIFY_REPLIED
) {
881 seccomp_handle_addfd(addfd
);
882 mutex_unlock(&match
->notify_lock
);
890 /* If there were any pending addfd calls, clear them out */
891 list_for_each_entry_safe(addfd
, tmp
, &n
.addfd
, list
) {
892 /* The process went away before we got a chance to handle it */
894 list_del_init(&addfd
->list
);
895 complete(&addfd
->completion
);
899 * Note that it's possible the listener died in between the time when
900 * we were notified of a response (or a signal) and when we were able to
901 * re-acquire the lock, so only delete from the list if the
902 * notification actually exists.
904 * Also note that this test is only valid because there's no way to
905 * *reattach* to a notifier right now. If one is added, we'll need to
906 * keep track of the notif itself and make sure they match here.
911 mutex_unlock(&match
->notify_lock
);
913 /* Userspace requests to continue the syscall. */
914 if (flags
& SECCOMP_USER_NOTIF_FLAG_CONTINUE
)
917 syscall_set_return_value(current
, current_pt_regs(),
922 static int __seccomp_filter(int this_syscall
, const struct seccomp_data
*sd
,
923 const bool recheck_after_trace
)
925 u32 filter_ret
, action
;
926 struct seccomp_filter
*match
= NULL
;
928 struct seccomp_data sd_local
;
931 * Make sure that any changes to mode from another thread have
932 * been seen after TIF_SECCOMP was seen.
937 populate_seccomp_data(&sd_local
);
941 filter_ret
= seccomp_run_filters(sd
, &match
);
942 data
= filter_ret
& SECCOMP_RET_DATA
;
943 action
= filter_ret
& SECCOMP_RET_ACTION_FULL
;
946 case SECCOMP_RET_ERRNO
:
947 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
948 if (data
> MAX_ERRNO
)
950 syscall_set_return_value(current
, current_pt_regs(),
954 case SECCOMP_RET_TRAP
:
955 /* Show the handler the original registers. */
956 syscall_rollback(current
, current_pt_regs());
957 /* Let the filter pass back 16 bits of data. */
958 seccomp_send_sigsys(this_syscall
, data
);
961 case SECCOMP_RET_TRACE
:
962 /* We've been put in this state by the ptracer already. */
963 if (recheck_after_trace
)
966 /* ENOSYS these calls if there is no tracer attached. */
967 if (!ptrace_event_enabled(current
, PTRACE_EVENT_SECCOMP
)) {
968 syscall_set_return_value(current
,
974 /* Allow the BPF to provide the event message */
975 ptrace_event(PTRACE_EVENT_SECCOMP
, data
);
977 * The delivery of a fatal signal during event
978 * notification may silently skip tracer notification,
979 * which could leave us with a potentially unmodified
980 * syscall that the tracer would have liked to have
981 * changed. Since the process is about to die, we just
982 * force the syscall to be skipped and let the signal
983 * kill the process and correctly handle any tracer exit
986 if (fatal_signal_pending(current
))
988 /* Check if the tracer forced the syscall to be skipped. */
989 this_syscall
= syscall_get_nr(current
, current_pt_regs());
990 if (this_syscall
< 0)
994 * Recheck the syscall, since it may have changed. This
995 * intentionally uses a NULL struct seccomp_data to force
996 * a reload of all registers. This does not goto skip since
997 * a skip would have already been reported.
999 if (__seccomp_filter(this_syscall
, NULL
, true))
1004 case SECCOMP_RET_USER_NOTIF
:
1005 if (seccomp_do_user_notification(this_syscall
, match
, sd
))
1010 case SECCOMP_RET_LOG
:
1011 seccomp_log(this_syscall
, 0, action
, true);
1014 case SECCOMP_RET_ALLOW
:
1016 * Note that the "match" filter will always be NULL for
1017 * this action since SECCOMP_RET_ALLOW is the starting
1018 * state in seccomp_run_filters().
1022 case SECCOMP_RET_KILL_THREAD
:
1023 case SECCOMP_RET_KILL_PROCESS
:
1025 seccomp_log(this_syscall
, SIGSYS
, action
, true);
1026 /* Dump core only if this is the last remaining thread. */
1027 if (action
!= SECCOMP_RET_KILL_THREAD
||
1028 get_nr_threads(current
) == 1) {
1029 kernel_siginfo_t info
;
1031 /* Show the original registers in the dump. */
1032 syscall_rollback(current
, current_pt_regs());
1033 /* Trigger a manual coredump since do_exit skips it. */
1034 seccomp_init_siginfo(&info
, this_syscall
, data
);
1037 if (action
== SECCOMP_RET_KILL_THREAD
)
1040 do_group_exit(SIGSYS
);
1046 seccomp_log(this_syscall
, 0, action
, match
? match
->log
: false);
1050 static int __seccomp_filter(int this_syscall
, const struct seccomp_data
*sd
,
1051 const bool recheck_after_trace
)
1057 int __secure_computing(const struct seccomp_data
*sd
)
1059 int mode
= current
->seccomp
.mode
;
1062 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE
) &&
1063 unlikely(current
->ptrace
& PT_SUSPEND_SECCOMP
))
1066 this_syscall
= sd
? sd
->nr
:
1067 syscall_get_nr(current
, current_pt_regs());
1070 case SECCOMP_MODE_STRICT
:
1071 __secure_computing_strict(this_syscall
); /* may call do_exit */
1073 case SECCOMP_MODE_FILTER
:
1074 return __seccomp_filter(this_syscall
, sd
, false);
1079 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1081 long prctl_get_seccomp(void)
1083 return current
->seccomp
.mode
;
1087 * seccomp_set_mode_strict: internal function for setting strict seccomp
1089 * Once current->seccomp.mode is non-zero, it may not be changed.
1091 * Returns 0 on success or -EINVAL on failure.
1093 static long seccomp_set_mode_strict(void)
1095 const unsigned long seccomp_mode
= SECCOMP_MODE_STRICT
;
1098 spin_lock_irq(¤t
->sighand
->siglock
);
1100 if (!seccomp_may_assign_mode(seccomp_mode
))
1106 seccomp_assign_mode(current
, seccomp_mode
, 0);
1110 spin_unlock_irq(¤t
->sighand
->siglock
);
1115 #ifdef CONFIG_SECCOMP_FILTER
1116 static void seccomp_notify_free(struct seccomp_filter
*filter
)
1118 kfree(filter
->notif
);
1119 filter
->notif
= NULL
;
1122 static void seccomp_notify_detach(struct seccomp_filter
*filter
)
1124 struct seccomp_knotif
*knotif
;
1129 mutex_lock(&filter
->notify_lock
);
1132 * If this file is being closed because e.g. the task who owned it
1133 * died, let's wake everyone up who was waiting on us.
1135 list_for_each_entry(knotif
, &filter
->notif
->notifications
, list
) {
1136 if (knotif
->state
== SECCOMP_NOTIFY_REPLIED
)
1139 knotif
->state
= SECCOMP_NOTIFY_REPLIED
;
1140 knotif
->error
= -ENOSYS
;
1144 * We do not need to wake up any pending addfd messages, as
1145 * the notifier will do that for us, as this just looks
1146 * like a standard reply.
1148 complete(&knotif
->ready
);
1151 seccomp_notify_free(filter
);
1152 mutex_unlock(&filter
->notify_lock
);
1155 static int seccomp_notify_release(struct inode
*inode
, struct file
*file
)
1157 struct seccomp_filter
*filter
= file
->private_data
;
1159 seccomp_notify_detach(filter
);
1160 __put_seccomp_filter(filter
);
1164 /* must be called with notif_lock held */
1165 static inline struct seccomp_knotif
*
1166 find_notification(struct seccomp_filter
*filter
, u64 id
)
1168 struct seccomp_knotif
*cur
;
1170 lockdep_assert_held(&filter
->notify_lock
);
1172 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1181 static long seccomp_notify_recv(struct seccomp_filter
*filter
,
1184 struct seccomp_knotif
*knotif
= NULL
, *cur
;
1185 struct seccomp_notif unotif
;
1188 /* Verify that we're not given garbage to keep struct extensible. */
1189 ret
= check_zeroed_user(buf
, sizeof(unotif
));
1195 memset(&unotif
, 0, sizeof(unotif
));
1197 ret
= down_interruptible(&filter
->notif
->request
);
1201 mutex_lock(&filter
->notify_lock
);
1202 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1203 if (cur
->state
== SECCOMP_NOTIFY_INIT
) {
1210 * If we didn't find a notification, it could be that the task was
1211 * interrupted by a fatal signal between the time we were woken and
1212 * when we were able to acquire the rw lock.
1219 unotif
.id
= knotif
->id
;
1220 unotif
.pid
= task_pid_vnr(knotif
->task
);
1221 unotif
.data
= *(knotif
->data
);
1223 knotif
->state
= SECCOMP_NOTIFY_SENT
;
1224 wake_up_poll(&filter
->wqh
, EPOLLOUT
| EPOLLWRNORM
);
1227 mutex_unlock(&filter
->notify_lock
);
1229 if (ret
== 0 && copy_to_user(buf
, &unotif
, sizeof(unotif
))) {
1233 * Userspace screwed up. To make sure that we keep this
1234 * notification alive, let's reset it back to INIT. It
1235 * may have died when we released the lock, so we need to make
1236 * sure it's still around.
1238 mutex_lock(&filter
->notify_lock
);
1239 knotif
= find_notification(filter
, unotif
.id
);
1241 knotif
->state
= SECCOMP_NOTIFY_INIT
;
1242 up(&filter
->notif
->request
);
1244 mutex_unlock(&filter
->notify_lock
);
1250 static long seccomp_notify_send(struct seccomp_filter
*filter
,
1253 struct seccomp_notif_resp resp
= {};
1254 struct seccomp_knotif
*knotif
;
1257 if (copy_from_user(&resp
, buf
, sizeof(resp
)))
1260 if (resp
.flags
& ~SECCOMP_USER_NOTIF_FLAG_CONTINUE
)
1263 if ((resp
.flags
& SECCOMP_USER_NOTIF_FLAG_CONTINUE
) &&
1264 (resp
.error
|| resp
.val
))
1267 ret
= mutex_lock_interruptible(&filter
->notify_lock
);
1271 knotif
= find_notification(filter
, resp
.id
);
1277 /* Allow exactly one reply. */
1278 if (knotif
->state
!= SECCOMP_NOTIFY_SENT
) {
1284 knotif
->state
= SECCOMP_NOTIFY_REPLIED
;
1285 knotif
->error
= resp
.error
;
1286 knotif
->val
= resp
.val
;
1287 knotif
->flags
= resp
.flags
;
1288 complete(&knotif
->ready
);
1290 mutex_unlock(&filter
->notify_lock
);
1294 static long seccomp_notify_id_valid(struct seccomp_filter
*filter
,
1297 struct seccomp_knotif
*knotif
;
1301 if (copy_from_user(&id
, buf
, sizeof(id
)))
1304 ret
= mutex_lock_interruptible(&filter
->notify_lock
);
1308 knotif
= find_notification(filter
, id
);
1309 if (knotif
&& knotif
->state
== SECCOMP_NOTIFY_SENT
)
1314 mutex_unlock(&filter
->notify_lock
);
1318 static long seccomp_notify_addfd(struct seccomp_filter
*filter
,
1319 struct seccomp_notif_addfd __user
*uaddfd
,
1322 struct seccomp_notif_addfd addfd
;
1323 struct seccomp_knotif
*knotif
;
1324 struct seccomp_kaddfd kaddfd
;
1327 BUILD_BUG_ON(sizeof(addfd
) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0
);
1328 BUILD_BUG_ON(sizeof(addfd
) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST
);
1330 if (size
< SECCOMP_NOTIFY_ADDFD_SIZE_VER0
|| size
>= PAGE_SIZE
)
1333 ret
= copy_struct_from_user(&addfd
, sizeof(addfd
), uaddfd
, size
);
1337 if (addfd
.newfd_flags
& ~O_CLOEXEC
)
1340 if (addfd
.flags
& ~SECCOMP_ADDFD_FLAG_SETFD
)
1343 if (addfd
.newfd
&& !(addfd
.flags
& SECCOMP_ADDFD_FLAG_SETFD
))
1346 kaddfd
.file
= fget(addfd
.srcfd
);
1350 kaddfd
.flags
= addfd
.newfd_flags
;
1351 kaddfd
.fd
= (addfd
.flags
& SECCOMP_ADDFD_FLAG_SETFD
) ?
1353 init_completion(&kaddfd
.completion
);
1355 ret
= mutex_lock_interruptible(&filter
->notify_lock
);
1359 knotif
= find_notification(filter
, addfd
.id
);
1366 * We do not want to allow for FD injection to occur before the
1367 * notification has been picked up by a userspace handler, or after
1368 * the notification has been replied to.
1370 if (knotif
->state
!= SECCOMP_NOTIFY_SENT
) {
1375 list_add(&kaddfd
.list
, &knotif
->addfd
);
1376 complete(&knotif
->ready
);
1377 mutex_unlock(&filter
->notify_lock
);
1379 /* Now we wait for it to be processed or be interrupted */
1380 ret
= wait_for_completion_interruptible(&kaddfd
.completion
);
1383 * We had a successful completion. The other side has already
1384 * removed us from the addfd queue, and
1385 * wait_for_completion_interruptible has a memory barrier upon
1386 * success that lets us read this value directly without
1393 mutex_lock(&filter
->notify_lock
);
1395 * Even though we were woken up by a signal and not a successful
1396 * completion, a completion may have happened in the mean time.
1398 * We need to check again if the addfd request has been handled,
1399 * and if not, we will remove it from the queue.
1401 if (list_empty(&kaddfd
.list
))
1404 list_del(&kaddfd
.list
);
1407 mutex_unlock(&filter
->notify_lock
);
1414 static long seccomp_notify_ioctl(struct file
*file
, unsigned int cmd
,
1417 struct seccomp_filter
*filter
= file
->private_data
;
1418 void __user
*buf
= (void __user
*)arg
;
1420 /* Fixed-size ioctls */
1422 case SECCOMP_IOCTL_NOTIF_RECV
:
1423 return seccomp_notify_recv(filter
, buf
);
1424 case SECCOMP_IOCTL_NOTIF_SEND
:
1425 return seccomp_notify_send(filter
, buf
);
1426 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR
:
1427 case SECCOMP_IOCTL_NOTIF_ID_VALID
:
1428 return seccomp_notify_id_valid(filter
, buf
);
1431 /* Extensible Argument ioctls */
1432 #define EA_IOCTL(cmd) ((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1433 switch (EA_IOCTL(cmd
)) {
1434 case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD
):
1435 return seccomp_notify_addfd(filter
, buf
, _IOC_SIZE(cmd
));
1441 static __poll_t
seccomp_notify_poll(struct file
*file
,
1442 struct poll_table_struct
*poll_tab
)
1444 struct seccomp_filter
*filter
= file
->private_data
;
1446 struct seccomp_knotif
*cur
;
1448 poll_wait(file
, &filter
->wqh
, poll_tab
);
1450 if (mutex_lock_interruptible(&filter
->notify_lock
) < 0)
1453 list_for_each_entry(cur
, &filter
->notif
->notifications
, list
) {
1454 if (cur
->state
== SECCOMP_NOTIFY_INIT
)
1455 ret
|= EPOLLIN
| EPOLLRDNORM
;
1456 if (cur
->state
== SECCOMP_NOTIFY_SENT
)
1457 ret
|= EPOLLOUT
| EPOLLWRNORM
;
1458 if ((ret
& EPOLLIN
) && (ret
& EPOLLOUT
))
1462 mutex_unlock(&filter
->notify_lock
);
1464 if (refcount_read(&filter
->users
) == 0)
1470 static const struct file_operations seccomp_notify_ops
= {
1471 .poll
= seccomp_notify_poll
,
1472 .release
= seccomp_notify_release
,
1473 .unlocked_ioctl
= seccomp_notify_ioctl
,
1474 .compat_ioctl
= seccomp_notify_ioctl
,
1477 static struct file
*init_listener(struct seccomp_filter
*filter
)
1481 ret
= ERR_PTR(-ENOMEM
);
1482 filter
->notif
= kzalloc(sizeof(*(filter
->notif
)), GFP_KERNEL
);
1486 sema_init(&filter
->notif
->request
, 0);
1487 filter
->notif
->next_id
= get_random_u64();
1488 INIT_LIST_HEAD(&filter
->notif
->notifications
);
1490 ret
= anon_inode_getfile("seccomp notify", &seccomp_notify_ops
,
1495 /* The file has a reference to it now */
1496 __get_seccomp_filter(filter
);
1500 seccomp_notify_free(filter
);
1506 * Does @new_child have a listener while an ancestor also has a listener?
1507 * If so, we'll want to reject this filter.
1508 * This only has to be tested for the current process, even in the TSYNC case,
1509 * because TSYNC installs @child with the same parent on all threads.
1510 * Note that @new_child is not hooked up to its parent at this point yet, so
1511 * we use current->seccomp.filter.
1513 static bool has_duplicate_listener(struct seccomp_filter
*new_child
)
1515 struct seccomp_filter
*cur
;
1517 /* must be protected against concurrent TSYNC */
1518 lockdep_assert_held(¤t
->sighand
->siglock
);
1520 if (!new_child
->notif
)
1522 for (cur
= current
->seccomp
.filter
; cur
; cur
= cur
->prev
) {
1531 * seccomp_set_mode_filter: internal function for setting seccomp filter
1532 * @flags: flags to change filter behavior
1533 * @filter: struct sock_fprog containing filter
1535 * This function may be called repeatedly to install additional filters.
1536 * Every filter successfully installed will be evaluated (in reverse order)
1537 * for each system call the task makes.
1539 * Once current->seccomp.mode is non-zero, it may not be changed.
1541 * Returns 0 on success or -EINVAL on failure.
1543 static long seccomp_set_mode_filter(unsigned int flags
,
1544 const char __user
*filter
)
1546 const unsigned long seccomp_mode
= SECCOMP_MODE_FILTER
;
1547 struct seccomp_filter
*prepared
= NULL
;
1550 struct file
*listener_f
= NULL
;
1552 /* Validate flags. */
1553 if (flags
& ~SECCOMP_FILTER_FLAG_MASK
)
1557 * In the successful case, NEW_LISTENER returns the new listener fd.
1558 * But in the failure case, TSYNC returns the thread that died. If you
1559 * combine these two flags, there's no way to tell whether something
1560 * succeeded or failed. So, let's disallow this combination if the user
1561 * has not explicitly requested no errors from TSYNC.
1563 if ((flags
& SECCOMP_FILTER_FLAG_TSYNC
) &&
1564 (flags
& SECCOMP_FILTER_FLAG_NEW_LISTENER
) &&
1565 ((flags
& SECCOMP_FILTER_FLAG_TSYNC_ESRCH
) == 0))
1568 /* Prepare the new filter before holding any locks. */
1569 prepared
= seccomp_prepare_user_filter(filter
);
1570 if (IS_ERR(prepared
))
1571 return PTR_ERR(prepared
);
1573 if (flags
& SECCOMP_FILTER_FLAG_NEW_LISTENER
) {
1574 listener
= get_unused_fd_flags(O_CLOEXEC
);
1580 listener_f
= init_listener(prepared
);
1581 if (IS_ERR(listener_f
)) {
1582 put_unused_fd(listener
);
1583 ret
= PTR_ERR(listener_f
);
1589 * Make sure we cannot change seccomp or nnp state via TSYNC
1590 * while another thread is in the middle of calling exec.
1592 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
&&
1593 mutex_lock_killable(¤t
->signal
->cred_guard_mutex
))
1596 spin_lock_irq(¤t
->sighand
->siglock
);
1598 if (!seccomp_may_assign_mode(seccomp_mode
))
1601 if (has_duplicate_listener(prepared
)) {
1606 ret
= seccomp_attach_filter(flags
, prepared
);
1609 /* Do not free the successfully attached filter. */
1612 seccomp_assign_mode(current
, seccomp_mode
, flags
);
1614 spin_unlock_irq(¤t
->sighand
->siglock
);
1615 if (flags
& SECCOMP_FILTER_FLAG_TSYNC
)
1616 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1618 if (flags
& SECCOMP_FILTER_FLAG_NEW_LISTENER
) {
1620 listener_f
->private_data
= NULL
;
1622 put_unused_fd(listener
);
1623 seccomp_notify_detach(prepared
);
1625 fd_install(listener
, listener_f
);
1630 seccomp_filter_free(prepared
);
1634 static inline long seccomp_set_mode_filter(unsigned int flags
,
1635 const char __user
*filter
)
1641 static long seccomp_get_action_avail(const char __user
*uaction
)
1645 if (copy_from_user(&action
, uaction
, sizeof(action
)))
1649 case SECCOMP_RET_KILL_PROCESS
:
1650 case SECCOMP_RET_KILL_THREAD
:
1651 case SECCOMP_RET_TRAP
:
1652 case SECCOMP_RET_ERRNO
:
1653 case SECCOMP_RET_USER_NOTIF
:
1654 case SECCOMP_RET_TRACE
:
1655 case SECCOMP_RET_LOG
:
1656 case SECCOMP_RET_ALLOW
:
1665 static long seccomp_get_notif_sizes(void __user
*usizes
)
1667 struct seccomp_notif_sizes sizes
= {
1668 .seccomp_notif
= sizeof(struct seccomp_notif
),
1669 .seccomp_notif_resp
= sizeof(struct seccomp_notif_resp
),
1670 .seccomp_data
= sizeof(struct seccomp_data
),
1673 if (copy_to_user(usizes
, &sizes
, sizeof(sizes
)))
1679 /* Common entry point for both prctl and syscall. */
1680 static long do_seccomp(unsigned int op
, unsigned int flags
,
1684 case SECCOMP_SET_MODE_STRICT
:
1685 if (flags
!= 0 || uargs
!= NULL
)
1687 return seccomp_set_mode_strict();
1688 case SECCOMP_SET_MODE_FILTER
:
1689 return seccomp_set_mode_filter(flags
, uargs
);
1690 case SECCOMP_GET_ACTION_AVAIL
:
1694 return seccomp_get_action_avail(uargs
);
1695 case SECCOMP_GET_NOTIF_SIZES
:
1699 return seccomp_get_notif_sizes(uargs
);
1705 SYSCALL_DEFINE3(seccomp
, unsigned int, op
, unsigned int, flags
,
1706 void __user
*, uargs
)
1708 return do_seccomp(op
, flags
, uargs
);
1712 * prctl_set_seccomp: configures current->seccomp.mode
1713 * @seccomp_mode: requested mode to use
1714 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1716 * Returns 0 on success or -EINVAL on failure.
1718 long prctl_set_seccomp(unsigned long seccomp_mode
, void __user
*filter
)
1723 switch (seccomp_mode
) {
1724 case SECCOMP_MODE_STRICT
:
1725 op
= SECCOMP_SET_MODE_STRICT
;
1727 * Setting strict mode through prctl always ignored filter,
1728 * so make sure it is always NULL here to pass the internal
1729 * check in do_seccomp().
1733 case SECCOMP_MODE_FILTER
:
1734 op
= SECCOMP_SET_MODE_FILTER
;
1741 /* prctl interface doesn't have flags, so they are always zero. */
1742 return do_seccomp(op
, 0, uargs
);
1745 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
1746 static struct seccomp_filter
*get_nth_filter(struct task_struct
*task
,
1747 unsigned long filter_off
)
1749 struct seccomp_filter
*orig
, *filter
;
1750 unsigned long count
;
1753 * Note: this is only correct because the caller should be the (ptrace)
1754 * tracer of the task, otherwise lock_task_sighand is needed.
1756 spin_lock_irq(&task
->sighand
->siglock
);
1758 if (task
->seccomp
.mode
!= SECCOMP_MODE_FILTER
) {
1759 spin_unlock_irq(&task
->sighand
->siglock
);
1760 return ERR_PTR(-EINVAL
);
1763 orig
= task
->seccomp
.filter
;
1764 __get_seccomp_filter(orig
);
1765 spin_unlock_irq(&task
->sighand
->siglock
);
1768 for (filter
= orig
; filter
; filter
= filter
->prev
)
1771 if (filter_off
>= count
) {
1772 filter
= ERR_PTR(-ENOENT
);
1776 count
-= filter_off
;
1777 for (filter
= orig
; filter
&& count
> 1; filter
= filter
->prev
)
1780 if (WARN_ON(count
!= 1 || !filter
)) {
1781 filter
= ERR_PTR(-ENOENT
);
1785 __get_seccomp_filter(filter
);
1788 __put_seccomp_filter(orig
);
1792 long seccomp_get_filter(struct task_struct
*task
, unsigned long filter_off
,
1795 struct seccomp_filter
*filter
;
1796 struct sock_fprog_kern
*fprog
;
1799 if (!capable(CAP_SYS_ADMIN
) ||
1800 current
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
) {
1804 filter
= get_nth_filter(task
, filter_off
);
1806 return PTR_ERR(filter
);
1808 fprog
= filter
->prog
->orig_prog
;
1810 /* This must be a new non-cBPF filter, since we save
1811 * every cBPF filter's orig_prog above when
1812 * CONFIG_CHECKPOINT_RESTORE is enabled.
1822 if (copy_to_user(data
, fprog
->filter
, bpf_classic_proglen(fprog
)))
1826 __put_seccomp_filter(filter
);
1830 long seccomp_get_metadata(struct task_struct
*task
,
1831 unsigned long size
, void __user
*data
)
1834 struct seccomp_filter
*filter
;
1835 struct seccomp_metadata kmd
= {};
1837 if (!capable(CAP_SYS_ADMIN
) ||
1838 current
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
) {
1842 size
= min_t(unsigned long, size
, sizeof(kmd
));
1844 if (size
< sizeof(kmd
.filter_off
))
1847 if (copy_from_user(&kmd
.filter_off
, data
, sizeof(kmd
.filter_off
)))
1850 filter
= get_nth_filter(task
, kmd
.filter_off
);
1852 return PTR_ERR(filter
);
1855 kmd
.flags
|= SECCOMP_FILTER_FLAG_LOG
;
1858 if (copy_to_user(data
, &kmd
, size
))
1861 __put_seccomp_filter(filter
);
1866 #ifdef CONFIG_SYSCTL
1868 /* Human readable action names for friendly sysctl interaction */
1869 #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
1870 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
1871 #define SECCOMP_RET_TRAP_NAME "trap"
1872 #define SECCOMP_RET_ERRNO_NAME "errno"
1873 #define SECCOMP_RET_USER_NOTIF_NAME "user_notif"
1874 #define SECCOMP_RET_TRACE_NAME "trace"
1875 #define SECCOMP_RET_LOG_NAME "log"
1876 #define SECCOMP_RET_ALLOW_NAME "allow"
1878 static const char seccomp_actions_avail
[] =
1879 SECCOMP_RET_KILL_PROCESS_NAME
" "
1880 SECCOMP_RET_KILL_THREAD_NAME
" "
1881 SECCOMP_RET_TRAP_NAME
" "
1882 SECCOMP_RET_ERRNO_NAME
" "
1883 SECCOMP_RET_USER_NOTIF_NAME
" "
1884 SECCOMP_RET_TRACE_NAME
" "
1885 SECCOMP_RET_LOG_NAME
" "
1886 SECCOMP_RET_ALLOW_NAME
;
1888 struct seccomp_log_name
{
1893 static const struct seccomp_log_name seccomp_log_names
[] = {
1894 { SECCOMP_LOG_KILL_PROCESS
, SECCOMP_RET_KILL_PROCESS_NAME
},
1895 { SECCOMP_LOG_KILL_THREAD
, SECCOMP_RET_KILL_THREAD_NAME
},
1896 { SECCOMP_LOG_TRAP
, SECCOMP_RET_TRAP_NAME
},
1897 { SECCOMP_LOG_ERRNO
, SECCOMP_RET_ERRNO_NAME
},
1898 { SECCOMP_LOG_USER_NOTIF
, SECCOMP_RET_USER_NOTIF_NAME
},
1899 { SECCOMP_LOG_TRACE
, SECCOMP_RET_TRACE_NAME
},
1900 { SECCOMP_LOG_LOG
, SECCOMP_RET_LOG_NAME
},
1901 { SECCOMP_LOG_ALLOW
, SECCOMP_RET_ALLOW_NAME
},
1905 static bool seccomp_names_from_actions_logged(char *names
, size_t size
,
1909 const struct seccomp_log_name
*cur
;
1910 bool append_sep
= false;
1912 for (cur
= seccomp_log_names
; cur
->name
&& size
; cur
++) {
1915 if (!(actions_logged
& cur
->log
))
1919 ret
= strscpy(names
, sep
, size
);
1928 ret
= strscpy(names
, cur
->name
, size
);
1939 static bool seccomp_action_logged_from_name(u32
*action_logged
,
1942 const struct seccomp_log_name
*cur
;
1944 for (cur
= seccomp_log_names
; cur
->name
; cur
++) {
1945 if (!strcmp(cur
->name
, name
)) {
1946 *action_logged
= cur
->log
;
1954 static bool seccomp_actions_logged_from_names(u32
*actions_logged
, char *names
)
1958 *actions_logged
= 0;
1959 while ((name
= strsep(&names
, " ")) && *name
) {
1960 u32 action_logged
= 0;
1962 if (!seccomp_action_logged_from_name(&action_logged
, name
))
1965 *actions_logged
|= action_logged
;
1971 static int read_actions_logged(struct ctl_table
*ro_table
, void __user
*buffer
,
1972 size_t *lenp
, loff_t
*ppos
)
1974 char names
[sizeof(seccomp_actions_avail
)];
1975 struct ctl_table table
;
1977 memset(names
, 0, sizeof(names
));
1979 if (!seccomp_names_from_actions_logged(names
, sizeof(names
),
1980 seccomp_actions_logged
, " "))
1985 table
.maxlen
= sizeof(names
);
1986 return proc_dostring(&table
, 0, buffer
, lenp
, ppos
);
1989 static int write_actions_logged(struct ctl_table
*ro_table
, void __user
*buffer
,
1990 size_t *lenp
, loff_t
*ppos
, u32
*actions_logged
)
1992 char names
[sizeof(seccomp_actions_avail
)];
1993 struct ctl_table table
;
1996 if (!capable(CAP_SYS_ADMIN
))
1999 memset(names
, 0, sizeof(names
));
2003 table
.maxlen
= sizeof(names
);
2004 ret
= proc_dostring(&table
, 1, buffer
, lenp
, ppos
);
2008 if (!seccomp_actions_logged_from_names(actions_logged
, table
.data
))
2011 if (*actions_logged
& SECCOMP_LOG_ALLOW
)
2014 seccomp_actions_logged
= *actions_logged
;
2018 static void audit_actions_logged(u32 actions_logged
, u32 old_actions_logged
,
2021 char names
[sizeof(seccomp_actions_avail
)];
2022 char old_names
[sizeof(seccomp_actions_avail
)];
2023 const char *new = names
;
2024 const char *old
= old_names
;
2029 memset(names
, 0, sizeof(names
));
2030 memset(old_names
, 0, sizeof(old_names
));
2034 else if (!actions_logged
)
2036 else if (!seccomp_names_from_actions_logged(names
, sizeof(names
),
2037 actions_logged
, ","))
2040 if (!old_actions_logged
)
2042 else if (!seccomp_names_from_actions_logged(old_names
,
2044 old_actions_logged
, ","))
2047 return audit_seccomp_actions_logged(new, old
, !ret
);
2050 static int seccomp_actions_logged_handler(struct ctl_table
*ro_table
, int write
,
2051 void *buffer
, size_t *lenp
,
2057 u32 actions_logged
= 0;
2058 u32 old_actions_logged
= seccomp_actions_logged
;
2060 ret
= write_actions_logged(ro_table
, buffer
, lenp
, ppos
,
2062 audit_actions_logged(actions_logged
, old_actions_logged
, ret
);
2064 ret
= read_actions_logged(ro_table
, buffer
, lenp
, ppos
);
2069 static struct ctl_path seccomp_sysctl_path
[] = {
2070 { .procname
= "kernel", },
2071 { .procname
= "seccomp", },
2075 static struct ctl_table seccomp_sysctl_table
[] = {
2077 .procname
= "actions_avail",
2078 .data
= (void *) &seccomp_actions_avail
,
2079 .maxlen
= sizeof(seccomp_actions_avail
),
2081 .proc_handler
= proc_dostring
,
2084 .procname
= "actions_logged",
2086 .proc_handler
= seccomp_actions_logged_handler
,
2091 static int __init
seccomp_sysctl_init(void)
2093 struct ctl_table_header
*hdr
;
2095 hdr
= register_sysctl_paths(seccomp_sysctl_path
, seccomp_sysctl_table
);
2097 pr_warn("sysctl registration failed\n");
2099 kmemleak_not_leak(hdr
);
2104 device_initcall(seccomp_sysctl_init
)
2106 #endif /* CONFIG_SYSCTL */