1 .\" Copyright (C) 2020 Michael Kerrisk <mtk.manpages@gmail.com>
3 .\" %%%LICENSE_START(VERBATIM)
4 .\" Permission is granted to make and distribute verbatim copies of this
5 .\" manual provided the copyright notice and this permission notice are
6 .\" preserved on all copies.
8 .\" Permission is granted to copy and distribute modified versions of this
9 .\" manual under the conditions for verbatim copying, provided that the
10 .\" entire resulting derived work is distributed under the terms of a
11 .\" permission notice identical to this one.
13 .\" Since the Linux kernel and libraries are constantly changing, this
14 .\" manual page may be incorrect or out-of-date. The author(s) assume no
15 .\" responsibility for errors or omissions, or for damages resulting from
16 .\" the use of the information contained herein. The author(s) may not
17 .\" have taken the same level of care in the production of this manual,
18 .\" which is licensed free of charge, as they might when working
21 .\" Formatted or processed versions of this manual, if unaccompanied by
22 .\" the source, must acknowledge the copyright and authors of this work.
25 .TH SECCOMP_UNOTIFY 2 2021-06-20 "Linux" "Linux Programmer's Manual"
27 seccomp_unotify \- Seccomp user-space notification mechanism
30 .B #include <linux/seccomp.h>
31 .B #include <linux/filter.h>
32 .B #include <linux/audit.h>
34 .BI "int seccomp(unsigned int " operation ", unsigned int " flags \
37 .B #include <sys/ioctl.h>
39 .BI "int ioctl(int " fd ", SECCOMP_IOCTL_NOTIF_RECV,"
40 .BI " struct seccomp_notif *" req );
41 .BI "int ioctl(int " fd ", SECCOMP_IOCTL_NOTIF_SEND,"
42 .BI " struct seccomp_notif_resp *" resp );
43 .BI "int ioctl(int " fd ", SECCOMP_IOCTL_NOTIF_ID_VALID, __u64 *" id );
44 .BI "int ioctl(int " fd ", SECCOMP_IOCTL_NOTIF_ADDFD,"
45 .BI " struct seccomp_notif_addfd *" addfd );
48 This page describes the user-space notification mechanism provided by the
49 Secure Computing (seccomp) facility.
50 As well as the use of the
51 .B SECCOMP_FILTER_FLAG_NEW_LISTENER
53 .BR SECCOMP_RET_USER_NOTIF
55 .B SECCOMP_GET_NOTIF_SIZES
56 operation described in
58 this mechanism involves the use of a number of related
60 operations (described below).
63 In conventional usage of a seccomp filter,
64 the decision about how to treat a system call is made by the filter itself.
65 By contrast, the user-space notification mechanism allows
66 the seccomp filter to delegate
67 the handling of the system call to another user-space process.
68 Note that this mechanism is explicitly
70 intended as a method implementing security policy; see NOTES.
72 In the discussion that follows,
73 the thread(s) on which the seccomp filter is installed is (are)
76 and the process that is notified by the user-space notification
77 mechanism is referred to as the
80 A suitably privileged supervisor can use the user-space notification
81 mechanism to perform actions on behalf of the target.
82 The advantage of the user-space notification mechanism is that
84 usually be able to retrieve information about the target and the
85 performed system call that the seccomp filter itself cannot.
86 (A seccomp filter is limited in the information it can obtain and
87 the actions that it can perform because it
88 is running on a virtual machine inside the kernel.)
90 An overview of the steps performed by the target and the supervisor
92 .\"-------------------------------------
94 The target establishes a seccomp filter in the usual manner,
95 but with two differences:
101 argument includes the flag
102 .BR SECCOMP_FILTER_FLAG_NEW_LISTENER .
103 Consequently, the return value of the (successful)
105 call is a new "listening"
106 file descriptor that can be used to receive notifications.
107 Only one "listening" seccomp filter can be installed for a thread.
109 .\" Is the last sentence above correct?
111 .\" Kees Cook (25 Oct 2020) notes:
113 .\" I like this limitation, but I expect that it'll need to change in the
114 .\" future. Even with LSMs, we see the need for arbitrary stacking, and the
115 .\" idea of there being only 1 supervisor will eventually break down. Right
116 .\" now there is only 1 because only container managers are using this
117 .\" feature. But if some daemon starts using it to isolate some thread,
118 .\" suddenly it might break if a container manager is trying to listen to it
119 .\" too, etc. I expect it won't be needed soon, but I do think it'll change.
122 In cases where it is appropriate, the seccomp filter returns the action value
123 .BR SECCOMP_RET_USER_NOTIF .
124 This return value will trigger a notification event.
126 .\"-------------------------------------
128 In order that the supervisor can obtain notifications
129 using the listening file descriptor,
130 (a duplicate of) that file descriptor must be passed from
131 the target to the supervisor.
132 One way in which this could be done is by passing the file descriptor
133 over a UNIX domain socket connection between the target and the supervisor
136 ancillary message type described in
138 Another way to do this is through the use of
141 .\" Instead of using unix domain sockets to send the fd to the
142 .\" parent, I think you could also use clone3() with
143 .\" flags==CLONE_FILES|SIGCHLD, dup2() the seccomp fd to an fd
144 .\" that was reserved in the parent, call unshare(CLONE_FILES)
145 .\" in the child after setting up the seccomp fd, and wake
146 .\" up the parent with something like pthread_cond_signal()?
147 .\" I'm not sure whether that'd look better or worse in the
148 .\" end though, so maybe just ignore this comment.
149 .\"-------------------------------------
151 The supervisor will receive notification events
152 on the listening file descriptor.
153 These events are returned as structures of type
155 Because this structure and its size may evolve over kernel versions,
156 the supervisor must first determine the size of this structure
159 .B SECCOMP_GET_NOTIF_SIZES
160 operation, which returns a structure of type
161 .IR seccomp_notif_sizes .
162 The supervisor allocates a buffer of size
163 .I seccomp_notif_sizes.seccomp_notif
164 bytes to receive notification events.
165 In addition,the supervisor allocates another buffer of size
166 .I seccomp_notif_sizes.seccomp_notif_resp
167 bytes for the response (a
168 .I struct seccomp_notif_resp
170 that it will provide to the kernel (and thus the target).
171 .\"-------------------------------------
173 The target then performs its workload,
174 which includes system calls that will be controlled by the seccomp filter.
175 Whenever one of these system calls causes the filter to return the
176 .B SECCOMP_RET_USER_NOTIF
177 action value, the kernel does
179 (yet) execute the system call;
180 instead, execution of the target is temporarily blocked inside
181 the kernel (in a sleep state that is interruptible by signals)
182 and a notification event is generated on the listening file descriptor.
183 .\"-------------------------------------
185 The supervisor can now repeatedly monitor the
186 listening file descriptor for
187 .BR SECCOMP_RET_USER_NOTIF -triggered
189 To do this, the supervisor uses the
190 .B SECCOMP_IOCTL_NOTIF_RECV
192 operation to read information about a notification event;
193 this operation blocks until an event is available.
194 The operation returns a
196 structure containing information about the system call
197 that is being attempted by the target.
198 (As described in NOTES,
199 the file descriptor can also be monitored with
205 .\" Christian Brauner:
207 .\" Do we support O_NONBLOCK with SECCOMP_IOCTL_NOTIF_RECV and if
212 .\" A quick test suggests that O_NONBLOCK has no effect on the blocking
213 .\" behavior of SECCOMP_IOCTL_NOTIF_RECV.
215 .\"-------------------------------------
219 structure returned by the
220 .B SECCOMP_IOCTL_NOTIF_RECV
221 operation includes the same information (a
223 structure) that was passed to the seccomp filter.
224 This information allows the supervisor to discover the system call number and
225 the arguments for the target's system call.
226 In addition, the notification event contains the ID of the thread
227 that triggered the notification and a unique cookie value that
228 is used in subsequent
229 .B SECCOMP_IOCTL_NOTIF_ID_VALID
231 .B SECCOMP_IOCTL_NOTIF_SEND
234 The information in the notification can be used to discover the
235 values of pointer arguments for the target's system call.
236 (This is something that can't be done from within a seccomp filter.)
237 One way in which the supervisor can do this is to open the corresponding
241 and read bytes from the location that corresponds to one of
242 the pointer arguments whose value is supplied in the notification event.
243 .\" Tycho Andersen mentioned that there are alternatives to /proc/PID/mem,
244 .\" such as ptrace() and /proc/PID/map_files
245 (The supervisor must be careful to avoid
246 a race condition that can occur when doing this;
247 see the description of the
248 .BR SECCOMP_IOCTL_NOTIF_ID_VALID
252 the supervisor can access other system information that is visible
253 in user space but which is not accessible from a seccomp filter.
254 .\"-------------------------------------
256 Having obtained information as per the previous step,
257 the supervisor may then choose to perform an action in response
258 to the target's system call
259 (which, as noted above, is not executed when the seccomp filter returns the
260 .B SECCOMP_RET_USER_NOTIF
263 One example use case here relates to containers.
264 The target may be located inside a container where
265 it does not have sufficient capabilities to mount a filesystem
266 in the container's mount namespace.
267 However, the supervisor may be a more privileged process that
268 does have sufficient capabilities to perform the mount operation.
269 .\"-------------------------------------
271 The supervisor then sends a response to the notification.
272 The information in this response is used by the kernel to construct
273 a return value for the target's system call and provide
274 a value that will be assigned to the
276 variable of the target.
278 The response is sent using the
279 .B SECCOMP_IOCTL_NOTIF_SEND
281 operation, which is used to transmit a
282 .I seccomp_notif_resp
283 structure to the kernel.
284 This structure includes a cookie value that the supervisor obtained in the
286 structure returned by the
287 .B SECCOMP_IOCTL_NOTIF_RECV
289 This cookie value allows the kernel to associate the response with the
291 This structure must include the cookie value that the supervisor
294 structure returned by the
295 .B SECCOMP_IOCTL_NOTIF_RECV
297 the cookie allows the kernel to associate the response with the target.
298 .\"-------------------------------------
300 Once the notification has been sent,
301 the system call in the target thread unblocks,
302 returning the information that was provided by the supervisor
303 in the notification response.
304 .\"-------------------------------------
306 As a variation on the last two steps,
307 the supervisor can send a response that tells the kernel that it
308 should execute the target thread's system call; see the discussion of
309 .BR SECCOMP_USER_NOTIF_FLAG_CONTINUE ,
315 operations are supported by the seccomp user-space
316 notification file descriptor.
317 For each of these operations, the first (file descriptor) argument of
319 is the listening file descriptor returned by a call to
322 .BR SECCOMP_FILTER_FLAG_NEW_LISTENER
325 .SS SECCOMP_IOCTL_NOTIF_RECV
327 .B SECCOMP_IOCTL_NOTIF_RECV
328 operation (available since Linux 5.0) is used to obtain a user-space
330 If no such event is currently pending,
331 the operation blocks until an event occurs.
334 argument is a pointer to a structure of the following form
335 which contains information about the event.
336 This structure must be zeroed out before the call.
340 struct seccomp_notif {
341 __u64 id; /* Cookie */
342 __u32 pid; /* TID of target thread */
343 __u32 flags; /* Currently unused (0) */
344 struct seccomp_data data; /* See seccomp(2) */
349 The fields in this structure are as follows:
352 This is a cookie for the notification.
353 Each such cookie is guaranteed to be unique for the corresponding
357 The cookie can be used with the
358 .B SECCOMP_IOCTL_NOTIF_ID_VALID
360 operation described below.
362 When returning a notification response to the kernel,
363 the supervisor must include the cookie value in the
364 .IR seccomp_notif_resp
365 structure that is specified as the argument of the
366 .BR SECCOMP_IOCTL_NOTIF_SEND
371 This is the thread ID of the target thread that triggered
372 the notification event.
375 This is a bit mask of flags providing further information on the event.
376 In the current implementation, this field is always zero.
381 structure containing information about the system call that
382 triggered the notification.
383 This is the same structure that is passed to the seccomp filter.
386 for details of this structure.
388 On success, this operation returns 0; on failure, \-1 is returned, and
390 is set to indicate the cause of the error.
391 This operation can fail with the following errors:
393 .BR EINVAL " (since Linux 5.5)"
394 .\" commit 2882d53c9c6f3b8311d225062522f03772cf0179
397 structure that was passed to the call contained nonzero fields.
400 The target thread was killed by a signal as the notification information
402 or the target's (blocked) system call was interrupted by a signal handler.
404 .\" From my experiments,
405 .\" it appears that if a SECCOMP_IOCTL_NOTIF_RECV is done after
406 .\" the target thread terminates, then the ioctl() simply
407 .\" blocks (rather than returning an error to indicate that the
408 .\" target no longer exists).
410 .\" I found that surprising, and it required some contortions in
411 .\" the example program. It was not possible to code my SIGCHLD
412 .\" handler (which reaps the zombie when the worker/target
413 .\" terminates) to simply set a flag checked in the main
414 .\" handleNotifications() loop, since this created an
415 .\" unavoidable race where the child might terminate just after
416 .\" I had checked the flag, but before I blocked (forever!) in the
417 .\" SECCOMP_IOCTL_NOTIF_RECV operation. Instead, I had to code
418 .\" the signal handler to simply call _exit(2) in order to
419 .\" terminate the parent process (the supervisor).
421 .\" Is this expected behavior? It seems to me rather
422 .\" desirable that SECCOMP_IOCTL_NOTIF_RECV should give an error
423 .\" if the target has terminated.
425 .\" Jann posted a patch to rectify this, but there was no response
426 .\" (Lore link: https://bit.ly/3jvUBxk) to his question about fixing
427 .\" this issue. (I've tried building with the patch, but encountered
428 .\" an issue with the target process entering D state after a signal.)
430 .\" For now, this behavior is documented in BUGS.
432 .\" Kees Cook commented: Let's change [this] ASAP!
434 .SS SECCOMP_IOCTL_NOTIF_ID_VALID
436 .B SECCOMP_IOCTL_NOTIF_ID_VALID
437 operation (available since Linux 5.0) is used to check that a notification ID
438 returned by an earlier
439 .B SECCOMP_IOCTL_NOTIF_RECV
440 operation is still valid
441 (i.e., that the target still exists and its system call
442 is still blocked waiting for a response).
446 argument is a pointer to the cookie
449 .B SECCOMP_IOCTL_NOTIF_RECV
452 This operation is necessary to avoid race conditions that can occur when the
455 .B SECCOMP_IOCTL_NOTIF_RECV
456 operation terminates, and that process ID is reused by another process.
457 An example of this kind of race is the following
459 A notification is generated on the listening file descriptor.
462 contains the TID of the target thread (in the
464 field of the structure).
466 The target terminates.
468 Another thread or process is created on the system that by chance reuses the
469 TID that was freed when the target terminated.
475 file for the TID obtained in step 1, with the intention of (say)
476 inspecting the memory location(s) that containing the argument(s) of
477 the system call that triggered the notification in step 1.
479 In the above scenario, the risk is that the supervisor may try
480 to access the memory of a process other than the target.
481 This race can be avoided by following the call to
484 .B SECCOMP_IOCTL_NOTIF_ID_VALID
485 operation to verify that the process that generated the notification
487 (Note that if the target terminates after the latter step,
490 from the file descriptor may return 0, indicating end of file.)
492 .\" the PID can be reused, but the /proc/$pid directory is
493 .\" internally not associated with the numeric PID, but,
494 .\" conceptually speaking, with a specific incarnation of the
495 .\" PID, or something like that. (Actually, it is associated
496 .\" with the "struct pid", which is not reused, instead of the
499 See NOTES for a discussion of other cases where
500 .B SECCOMP_IOCTL_NOTIF_ID_VALID
501 checks must be performed.
503 On success (i.e., the notification ID is still valid),
504 this operation returns 0.
505 On failure (i.e., the notification ID is no longer valid),
511 .SS SECCOMP_IOCTL_NOTIF_SEND
513 .B SECCOMP_IOCTL_NOTIF_SEND
514 operation (available since Linux 5.0)
515 is used to send a notification response back to the kernel.
518 argument of this structure is a pointer to a structure of the following form:
522 struct seccomp_notif_resp {
523 __u64 id; /* Cookie value */
524 __s64 val; /* Success return value */
525 __s32 error; /* 0 (success) or negative error number */
526 __u32 flags; /* See below */
531 The fields of this structure are as follows:
534 This is the cookie value that was obtained using the
535 .B SECCOMP_IOCTL_NOTIF_RECV
537 This cookie value allows the kernel to correctly associate this response
538 with the system call that triggered the user-space notification.
541 This is the value that will be used for a spoofed
542 success return for the target's system call; see below.
545 This is the value that will be used as the error number
547 for a spoofed error return for the target's system call; see below.
550 This is a bit mask that includes zero or more of the following flags:
553 .BR SECCOMP_USER_NOTIF_FLAG_CONTINUE " (since Linux 5.5)"
554 Tell the kernel to execute the target's system call.
555 .\" commit fb3c5386b382d4097476ce9647260fc89b34afdb
558 Two kinds of response are possible:
560 A response to the kernel telling it to execute the
561 target's system call.
565 .B SECCOMP_USER_NOTIF_FLAG_CONTINUE
572 This kind of response can be useful in cases where the supervisor needs
573 to do deeper analysis of the target's system call than is possible
574 from a seccomp filter (e.g., examining the values of pointer arguments),
575 and, having decided that the system call does not require emulation
576 by the supervisor, the supervisor wants the system call to
577 be executed normally in the target.
580 .B SECCOMP_USER_NOTIF_FLAG_CONTINUE
581 flag should be used with caution; see NOTES.
583 A spoofed return value for the target's system call.
584 In this case, the kernel does not execute the target's system call,
585 instead causing the system call to return a spoofed value as specified by
587 .I seccomp_notif_resp
589 The supervisor should set the fields of this structure as follows:
594 .BR SECCOMP_USER_NOTIF_FLAG_CONTINUE .
597 is set either to 0 for a spoofed "success" return or to a negative
598 error number for a spoofed "failure" return.
599 In the former case, the kernel causes the target's system call
600 to return the value specified in the
603 In the latter case, the kernel causes the target's system call
606 is assigned the negated
611 is set to a value that will be used as the return value for a spoofed
612 "success" return for the target's system call.
613 The value in this field is ignored if the
615 field contains a nonzero value.
617 .\" Kees Cook suggested:
619 .\" Strictly speaking, this is architecture specific, but
620 .\" all architectures do it this way. Should seccomp enforce
621 .\" val == 0 when err != 0 ?
623 .\" Christian Brauner
625 .\" Feels like it should, at least for the SEND ioctl where we already
626 .\" verify that val and err are both 0 when CONTINUE is specified (as you
627 .\" pointed out correctly above).
630 On success, this operation returns 0; on failure, \-1 is returned, and
632 is set to indicate the cause of the error.
633 This operation can fail with the following errors:
636 A response to this notification has already been sent.
639 An invalid value was specified in the
647 .BR SECCOMP_USER_NOTIF_FLAG_CONTINUE ,
655 The blocked system call in the target
656 has been interrupted by a signal handler
657 or the target has terminated.
659 .\" you could also get this [ENOENT] if a response has already
660 .\" been sent, instead of EINPROGRESS - the only difference is
661 .\" whether the target thread has picked up the response yet
663 .SS SECCOMP_IOCTL_NOTIF_ADDFD
665 .B SECCOMP_IOCTL_NOTIF_ADDFD
666 operation (available since Linux 5.9)
667 allows the supervisor to install a file descriptor
668 into the target's file descriptor table.
671 messages described in
673 this operation is semantically equivalent to duplicating
674 a file descriptor from the supervisor's file descriptor table
675 into the target's file descriptor table.
678 .BR SECCOMP_IOCTL_NOTIF_ADDFD
679 operation permits the supervisor to emulate a target system call (such as
683 that generates a file descriptor.
684 The supervisor can perform the system call that generates
685 the file descriptor (and associated open file description)
686 and then use this operation to allocate
687 a file descriptor that refers to the same open file description in the target.
688 (For an explanation of open file descriptions, see
691 Once this operation has been performed,
692 the supervisor can close its copy of the file descriptor.
695 the received file descriptor is subject to the same
696 Linux Security Module (LSM) checks as are applied to a file descriptor
697 that is received in an
700 If the file descriptor refers to a socket,
701 it inherits the cgroup version 1 network controller settings
709 argument is a pointer to a structure of the following form:
713 struct seccomp_notif_addfd {
714 __u64 id; /* Cookie value */
715 __u32 flags; /* Flags */
716 __u32 srcfd; /* Local file descriptor number */
717 __u32 newfd; /* 0 or desired file descriptor
719 __u32 newfd_flags; /* Flags to set on target file
725 The fields in this structure are as follows:
728 This field should be set to the notification ID
729 (cookie value) that was obtained via
730 .BR SECCOMP_IOCTL_NOTIF_RECV .
733 This field is a bit mask of flags that modify the behavior of the operation.
734 Currently, only one flag is supported:
737 .BR SECCOMP_ADDFD_FLAG_SETFD
738 When allocating the file descriptor in the target,
739 use the file descriptor number specified in the
743 .BR SECCOMP_ADDFD_FLAG_SEND
744 Available since Linux 5.14, combines the
745 .B SECCOMP_IOCTL_NOTIF_ADDFD
747 .B SECCOMP_IOCTL_NOTIF_SEND
748 into an atomic operation. On successful invocation, the target process's
749 errno will be 0 and the return value will be the file descriptor number that was
750 installed in the target. If allocating the file descriptor in the tatget fails,
751 the target's syscall continues to be blocked until a successful response is
756 This field should be set to the number of the file descriptor
757 in the supervisor that is to be duplicated.
760 This field determines which file descriptor number is allocated in the target.
762 .BR SECCOMP_ADDFD_FLAG_SETFD
764 then this field specifies which file descriptor number should be allocated.
765 If this file descriptor number is already open in the target,
766 it is atomically closed and reused.
767 If the descriptor duplication fails due to an LSM check, or if
769 is not a valid file descriptor,
772 will not be closed in the target process.
775 .BR SECCOMP_ADDFD_FLAG_SETFD
776 flag it not set, then this field must be 0,
777 and the kernel allocates the lowest unused file descriptor number
781 This field is a bit mask specifying flags that should be set on
782 the file descriptor that is received in the target process.
783 Currently, only the following flag is implemented:
787 Set the close-on-exec flag on the received file descriptor.
792 call returns the number of the file descriptor that was allocated
794 Assuming that the emulated system call is one that returns
795 a file descriptor as its function result (e.g.,
797 this value can be used as the return value
799 that is supplied in the response that is subsequently sent with the
800 .BR SECCOMP_IOCTL_NOTIF_SEND
803 On error, \-1 is returned and
805 is set to indicate the cause of the error.
807 This operation can fail with the following errors:
810 Allocating the file descriptor in the target would cause the target's
812 limit to be exceeded (see
817 .B SECCOMP_IOCTL_NOTIF_SEND
818 is used, this means the operation can't proceed until other
819 .B SECCOMP_IOCTL_NOTIF_ADDFD
820 requests are processed.
823 The user-space notification specified in the
825 field exists but has not yet been fetched (by a
826 .BR SECCOMP_IOCTL_NOTIF_RECV )
827 or has already been responded to (by a
828 .BR SECCOMP_IOCTL_NOTIF_SEND ).
831 An invalid flag was specified in the
837 field is nonzero and the
838 .B SECCOMP_ADDFD_FLAG_SETFD
839 flag was not specified in the
844 The file descriptor number specified in
846 exceeds the limit specified in
847 .IR /proc/sys/fs/nr_open .
850 The blocked system call in the target
851 has been interrupted by a signal handler
852 or the target has terminated.
854 Here is some sample code (with error handling omitted) that uses the
855 .B SECCOMP_ADDFD_FLAG_SETFD
856 operation (here, to emulate a call to
863 fd = openat(req->data.args[0], path, req->data.args[2],
866 struct seccomp_notif_addfd addfd;
867 addfd.id = req->id; /* Cookie from SECCOMP_IOCTL_NOTIF_RECV */
871 addfd.newfd_flags = O_CLOEXEC;
873 targetFd = ioctl(notifyFd, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd);
875 close(fd); /* No longer needed in supervisor */
877 struct seccomp_notif_resp *resp;
878 /* Code to allocate 'resp' omitted */
880 resp->error = 0; /* "Success" */
881 resp->val = targetFd;
883 ioctl(notifyFd, SECCOMP_IOCTL_NOTIF_SEND, resp);
887 One example use case for the user-space notification
888 mechanism is to allow a container manager
889 (a process which is typically running with more privilege than
890 the processes inside the container)
891 to mount block devices or create device nodes for the container.
892 The mount use case provides an example of where the
893 .BR SECCOMP_USER_NOTIF_FLAG_CONTINUE
896 Upon receiving a notification for the
898 system call, the container manager (the "supervisor") can distinguish
899 a request to mount a block filesystem
900 (which would not be possible for a "target" process inside the container)
901 and mount that file system.
902 If, on the other hand, the container manager detects that the operation
903 could be performed by the process inside the container
906 filesystem), it can notify the kernel that the target process's
908 system call can continue.
910 .SS select()/poll()/epoll semantics
911 The file descriptor returned when
914 .B SECCOMP_FILTER_FLAG_NEW_LISTENER
915 flag can be monitored using
920 These interfaces indicate that the file descriptor is ready as follows:
922 When a notification is pending,
923 these interfaces indicate that the file descriptor is readable.
924 Following such an indication, a subsequent
925 .B SECCOMP_IOCTL_NOTIF_RECV
927 will not block, returning either information about a notification
928 or else failing with the error
930 if the target has been killed by a signal or its system call
931 has been interrupted by a signal handler.
933 After the notification has been received (i.e., by the
934 .B SECCOMP_IOCTL_NOTIF_RECV
936 operation), these interfaces indicate that the file descriptor is writable,
937 meaning that a notification response can be sent using the
938 .B SECCOMP_IOCTL_NOTIF_SEND
942 After the last thread using the filter has terminated and been reaped using
945 the file descriptor indicates an end-of-file condition (readable in
947 .BR POLLHUP / EPOLLHUP
951 .SS Design goals; use of SECCOMP_USER_NOTIF_FLAG_CONTINUE
952 The intent of the user-space notification feature is
953 to allow system calls to be performed on behalf of the target.
954 The target's system call should either be handled by the supervisor or
955 allowed to continue normally in the kernel (where standard security
956 policies will be applied).
959 this mechanism must not be used to make security policy decisions
960 about the system call,
961 which would be inherently race-prone for reasons described next.
964 .B SECCOMP_USER_NOTIF_FLAG_CONTINUE
965 flag must be used with caution.
966 If set by the supervisor, the target's system call will continue.
967 However, there is a time-of-check, time-of-use race here,
968 since an attacker could exploit the interval of time where the target is
969 blocked waiting on the "continue" response to do things such as
970 rewriting the system call arguments.
972 Note furthermore that a user-space notifier can be bypassed if
973 the existing filters allow the use of
977 to install a filter that returns an action value with a higher precedence than
978 .B SECCOMP_RET_USER_NOTIF
982 It should thus be absolutely clear that the
983 seccomp user-space notification mechanism
985 be used to implement a security policy!
986 It should only ever be used in scenarios where a more privileged process
987 supervises the system calls of a lesser privileged target to
988 get around kernel-enforced security restrictions when
989 the supervisor deems this safe.
991 in order to continue a system call, the supervisor should be sure that
992 another security mechanism or the kernel itself will sufficiently block
993 the system call if its arguments are rewritten to something unsafe.
995 .SS Caveats regarding the use of /proc/[tid]/mem
996 The discussion above noted the need to use the
997 .BR SECCOMP_IOCTL_NOTIF_ID_VALID
1002 to avoid the possibility of accessing the memory of the wrong process
1003 in the event that the target terminates and its ID
1004 is recycled by another (unrelated) thread.
1005 However, the use of this
1007 operation is also necessary in other situations,
1008 as explained in the following paragraphs.
1010 Consider the following scenario, where the supervisor
1011 tries to read the pathname argument of a target's blocked
1015 From one of its functions
1019 which triggers a user-space notification and causes the target to block.
1021 The supervisor receives the notification, opens
1022 .IR /proc/[tid]/mem ,
1023 and (successfully) performs the
1024 .BR SECCOMP_IOCTL_NOTIF_ID_VALID
1027 The target receives a signal, which causes the
1031 The signal handler executes in the target, and returns.
1033 Upon return from the handler, the execution of
1035 resumes, and it returns (and perhaps other functions are called,
1036 overwriting the memory that had been used for the stack frame of
1039 Using the address provided in the notification information,
1040 the supervisor reads from the target's memory location that used to
1041 contain the pathname.
1043 The supervisor now calls
1045 with some arbitrary bytes obtained in the previous step.
1047 The conclusion from the above scenario is this:
1048 since the target's blocked system call may be interrupted by a signal handler,
1049 the supervisor must be written to expect that the
1050 target may abandon its system call at
1053 in such an event, any information that the supervisor obtained from
1054 the target's memory must be considered invalid.
1056 To prevent such scenarios,
1057 every read from the target's memory must be separated from use of
1058 the bytes so obtained by a
1059 .BR SECCOMP_IOCTL_NOTIF_ID_VALID
1061 In the above example, the check would be placed between the two final steps.
1062 An example of such a check is shown in EXAMPLES.
1064 Following on from the above, it should be clear that
1065 a write by the supervisor into the target's memory can
1069 .SS Caveats regarding blocking system calls
1070 Suppose that the target performs a blocking system call (e.g.,
1072 that the supervisor should handle.
1073 The supervisor might then in turn execute the same blocking system call.
1076 it is important to note that if the target's system call is now
1077 interrupted by a signal, the supervisor is
1080 If the supervisor does not take suitable steps to
1081 actively discover that the target's system call has been canceled,
1082 various difficulties can occur.
1083 Taking the example of
1085 the supervisor might remain blocked in its
1087 holding a port number that the target
1088 (which, after the interruption by the signal handler,
1089 perhaps closed its listening socket) might expect to be able to reuse in a
1093 Therefore, when the supervisor wishes to emulate a blocking system call,
1094 it must do so in such a way that it gets informed if the target's
1095 system call is interrupted by a signal handler.
1096 For example, if the supervisor itself executes the same
1097 blocking system call, then it could employ a separate thread
1099 .B SECCOMP_IOCTL_NOTIF_ID_VALID
1100 operation to check if the target is still blocked in its system call.
1101 Alternatively, in the
1103 example, the supervisor might use
1105 to monitor both the notification file descriptor
1106 (so as as to discover when the target's
1108 call has been interrupted) and the listening file descriptor
1109 (so as to know when a connection is available).
1111 If the target's system call is interrupted,
1112 the supervisor must take care to release resources (e.g., file descriptors)
1113 that it acquired on behalf of the target.
1115 .SS Interaction with SA_RESTART signal handlers
1116 Consider the following scenario:
1118 The target process has used
1120 to install a signal handler with the
1124 The target has made a system call that triggered a seccomp
1125 user-space notification and the target is currently blocked
1126 until the supervisor sends a notification response.
1128 A signal is delivered to the target and the signal handler is executed.
1130 When (if) the supervisor attempts to send a notification response, the
1131 .B SECCOMP_IOCTL_NOTIF_SEND
1133 operation will fail with the
1137 In this scenario, the kernel will restart the target's system call.
1138 Consequently, the supervisor will receive another user-space notification.
1139 Thus, depending on how many times the blocked system call
1140 is interrupted by a signal handler,
1141 the supervisor may receive multiple notifications for
1142 the same instance of a system call in the target.
1144 One oddity is that system call restarting as described in this scenario
1145 will occur even for the blocking system calls listed in
1149 normally be restarted by the
1153 Furthermore, if the supervisor response is a file descriptor
1155 .B SECCOMP_IOCTL_NOTIF_ADDFD,
1157 .B SECCOMP_ADDFD_FLAG_SEND
1158 can be used to atomically add the file descriptor and return that value,
1159 making sure no file descriptors are inadvertently leaked into the target.
1161 .\" About the above, Kees Cook commented:
1163 .\" Does this need fixing? I imagine the correct behavior for this case
1164 .\" would be a response to _SEND of EINPROGRESS and the target would see
1167 .\" I mean, it's not like seccomp doesn't already expose weirdness with
1168 .\" syscall restarts. Not even arm64 compat agrees[3] with arm32 in this
1172 .\" Michael Kerrisk:
1173 .\" I wonder about the effect of this oddity for system calls that
1174 .\" are normally nonrestartable because they have timeouts. My
1175 .\" understanding is that the kernel doesn't restart those system
1176 .\" calls because it's impossible for the kernel to restart the call
1177 .\" with the right timeout value. I wonder what happens when those
1178 .\" system calls are restarted in the scenario we're discussing.)
1181 .BR SECCOMP_IOCTL_NOTIF_RECV
1184 .\" or a poll/epoll/select
1185 is performed after the target terminates, then the
1187 call simply blocks (rather than returning an error to indicate that the
1188 target no longer exists).
1190 .\" Comment from Kees Cook:
1192 .\" I want this fixed. It caused me no end of pain when building the
1193 .\" selftests, and ended up spawning my implementing a global test timeout
1194 .\" in kselftest. :P Before the usage counter refactor, there was no sane
1195 .\" way to deal with this, but now I think we're close.
1198 The (somewhat contrived) program shown below demonstrates the use of
1199 the interfaces described in this page.
1200 The program creates a child process that serves as the "target" process.
1201 The child process installs a seccomp filter that returns the
1202 .B SECCOMP_RET_USER_NOTIF
1203 action value if a call is made to
1205 The child process then calls
1207 once for each of the supplied command-line arguments,
1208 and reports the result returned by the call.
1209 After processing all arguments, the child process terminates.
1211 The parent process acts as the supervisor, listening for the notifications
1212 that are generated when the target process calls
1214 When such a notification occurs,
1215 the supervisor examines the memory of the target process (using
1216 .IR /proc/[pid]/mem )
1217 to discover the pathname argument that was supplied to the
1219 call, and performs one of the following actions:
1221 If the pathname begins with the prefix "/tmp/",
1222 then the supervisor attempts to create the specified directory,
1223 and then spoofs a return for the target process based on the return
1224 value of the supervisor's
1227 In the event that that call succeeds,
1228 the spoofed success return value is the length of the pathname.
1230 If the pathname begins with "./" (i.e., it is a relative pathname),
1231 the supervisor sends a
1232 .B SECCOMP_USER_NOTIF_FLAG_CONTINUE
1233 response to the kernel to say that the kernel should execute
1234 the target process's
1238 If the pathname begins with some other prefix,
1239 the supervisor spoofs an error return for the target process,
1240 so that the target process's
1242 call appears to fail with the error
1244 ("Operation not supported").
1245 Additionally, if the specified pathname is exactly "/bye",
1246 then the supervisor terminates.
1248 This program can be used to demonstrate various aspects of the
1249 behavior of the seccomp user-space notification mechanism.
1250 To help aid such demonstrations,
1251 the program logs various messages to show the operation
1252 of the target process (lines prefixed "T:") and the supervisor
1253 (indented lines prefixed "S:").
1255 In the following example, the target attempts to create the directory
1257 Upon receiving the notification, the supervisor creates the directory on the
1259 and spoofs a success return to be received by the target process's
1265 $ \fB./seccomp_unotify /tmp/x\fP
1268 T: about to mkdir("/tmp/x")
1269 S: got notification (ID 0x17445c4a0f4e0e3c) for PID 23168
1270 S: executing: mkdir("/tmp/x", 0700)
1271 S: success! spoofed return = 6
1272 S: sending response (flags = 0; val = 6; error = 0)
1273 T: SUCCESS: mkdir(2) returned 6
1276 S: target has terminated; bye
1280 In the above output, note that the spoofed return value seen by the target
1281 process is 6 (the length of the pathname
1285 call returns 0 on success.
1287 In the next example, the target attempts to create a directory using the
1290 Since this pathname starts with "./",
1291 the supervisor sends a
1292 .B SECCOMP_USER_NOTIF_FLAG_CONTINUE
1293 response to the kernel,
1294 and the kernel then (successfully) executes the target process's
1300 $ \fB./seccomp_unotify ./sub\fP
1303 T: about to mkdir("./sub")
1304 S: got notification (ID 0xddb16abe25b4c12) for PID 23204
1305 S: target can execute system call
1306 S: sending response (flags = 0x1; val = 0; error = 0)
1307 T: SUCCESS: mkdir(2) returned 0
1310 S: target has terminated; bye
1314 If the target process attempts to create a directory with
1315 a pathname that doesn't start with "." and doesn't begin with the prefix
1316 "/tmp/", then the supervisor spoofs an error return
1318 "Operation not supported")
1321 call (which is not executed):
1325 $ \fB./seccomp_unotify /xxx\fP
1328 T: about to mkdir("/xxx")
1329 S: got notification (ID 0xe7dc095d1c524e80) for PID 23178
1330 S: spoofing error response (Operation not supported)
1331 S: sending response (flags = 0; val = 0; error = \-95)
1332 T: ERROR: mkdir(2): Operation not supported
1335 S: target has terminated; bye
1339 In the next example,
1340 the target process attempts to create a directory with the pathname
1341 .BR /tmp/nosuchdir/b .
1342 Upon receiving the notification,
1343 the supervisor attempts to create that directory, but the
1345 call fails because the directory
1348 Consequently, the supervisor spoofs an error return that passes the error
1349 that it received back to the target process's
1355 $ \fB./seccomp_unotify /tmp/nosuchdir/b\fP
1358 T: about to mkdir("/tmp/nosuchdir/b")
1359 S: got notification (ID 0x8744454293506046) for PID 23199
1360 S: executing: mkdir("/tmp/nosuchdir/b", 0700)
1361 S: failure! (errno = 2; No such file or directory)
1362 S: sending response (flags = 0; val = 0; error = \-2)
1363 T: ERROR: mkdir(2): No such file or directory
1366 S: target has terminated; bye
1370 If the supervisor receives a notification and sees that the
1371 argument of the target's
1373 is the string "/bye", then (as well as spoofing an
1375 error), the supervisor terminates.
1376 If the target process subsequently executes another
1378 that triggers its seccomp filter to return the
1379 .B SECCOMP_RET_USER_NOTIF
1380 action value, then the kernel causes the target process's system call to
1383 ("Function not implemented").
1384 This is demonstrated by the following example:
1388 $ \fB./seccomp_unotify /bye /tmp/y\fP
1391 T: about to mkdir("/bye")
1392 S: got notification (ID 0xa81236b1d2f7b0f4) for PID 23185
1393 S: spoofing error response (Operation not supported)
1394 S: sending response (flags = 0; val = 0; error = \-95)
1395 S: terminating **********
1396 T: ERROR: mkdir(2): Operation not supported
1398 T: about to mkdir("/tmp/y")
1399 T: ERROR: mkdir(2): Function not implemented
1411 #include <linux/audit.h>
1412 #include <linux/filter.h>
1413 #include <linux/seccomp.h>
1415 #include <stdbool.h>
1420 #include <sys/socket.h>
1421 #include <sys/ioctl.h>
1422 #include <sys/prctl.h>
1423 #include <sys/stat.h>
1424 #include <sys/types.h>
1426 #include <sys/syscall.h>
1429 #define errExit(msg) do { perror(msg); exit(EXIT_FAILURE); \e
1432 /* Send the file descriptor \(aqfd\(aq over the connected UNIX domain socket
1433 \(aqsockfd\(aq. Returns 0 on success, or \-1 on error. */
1436 sendfd(int sockfd, int fd)
1441 struct cmsghdr *cmsgp;
1443 /* Allocate a char array of suitable size to hold the ancillary data.
1444 However, since this buffer is in reality a \(aqstruct cmsghdr\(aq, use a
1445 union to ensure that it is suitably aligned. */
1447 char buf[CMSG_SPACE(sizeof(int))];
1448 /* Space large enough to hold an \(aqint\(aq */
1449 struct cmsghdr align;
1452 /* The \(aqmsg_name\(aq field can be used to specify the address of the
1453 destination socket when sending a datagram. However, we do not
1454 need to use this field because \(aqsockfd\(aq is a connected socket. */
1456 msgh.msg_name = NULL;
1457 msgh.msg_namelen = 0;
1459 /* On Linux, we must transmit at least one byte of real data in
1460 order to send ancillary data. We transmit an arbitrary integer
1461 whose value is ignored by recvfd(). */
1463 msgh.msg_iov = &iov;
1464 msgh.msg_iovlen = 1;
1465 iov.iov_base = &data;
1466 iov.iov_len = sizeof(int);
1469 /* Set \(aqmsghdr\(aq fields that describe ancillary data */
1471 msgh.msg_control = controlMsg.buf;
1472 msgh.msg_controllen = sizeof(controlMsg.buf);
1474 /* Set up ancillary data describing file descriptor to send */
1476 cmsgp = CMSG_FIRSTHDR(&msgh);
1477 cmsgp\->cmsg_level = SOL_SOCKET;
1478 cmsgp\->cmsg_type = SCM_RIGHTS;
1479 cmsgp\->cmsg_len = CMSG_LEN(sizeof(int));
1480 memcpy(CMSG_DATA(cmsgp), &fd, sizeof(int));
1482 /* Send real plus ancillary data */
1484 if (sendmsg(sockfd, &msgh, 0) == \-1)
1490 /* Receive a file descriptor on a connected UNIX domain socket. Returns
1491 the received file descriptor on success, or \-1 on error. */
1501 /* Allocate a char buffer for the ancillary data. See the comments
1504 char buf[CMSG_SPACE(sizeof(int))];
1505 struct cmsghdr align;
1507 struct cmsghdr *cmsgp;
1509 /* The \(aqmsg_name\(aq field can be used to obtain the address of the
1510 sending socket. However, we do not need this information. */
1512 msgh.msg_name = NULL;
1513 msgh.msg_namelen = 0;
1515 /* Specify buffer for receiving real data */
1517 msgh.msg_iov = &iov;
1518 msgh.msg_iovlen = 1;
1519 iov.iov_base = &data; /* Real data is an \(aqint\(aq */
1520 iov.iov_len = sizeof(int);
1522 /* Set \(aqmsghdr\(aq fields that describe ancillary data */
1524 msgh.msg_control = controlMsg.buf;
1525 msgh.msg_controllen = sizeof(controlMsg.buf);
1527 /* Receive real plus ancillary data; real data is ignored */
1529 nr = recvmsg(sockfd, &msgh, 0);
1533 cmsgp = CMSG_FIRSTHDR(&msgh);
1535 /* Check the validity of the \(aqcmsghdr\(aq */
1537 if (cmsgp == NULL ||
1538 cmsgp\->cmsg_len != CMSG_LEN(sizeof(int)) ||
1539 cmsgp\->cmsg_level != SOL_SOCKET ||
1540 cmsgp\->cmsg_type != SCM_RIGHTS) {
1545 /* Return the received file descriptor to our caller */
1547 memcpy(&fd, CMSG_DATA(cmsgp), sizeof(int));
1552 sigchldHandler(int sig)
1554 char msg[] = "\etS: target has terminated; bye\en";
1556 write(STDOUT_FILENO, msg, sizeof(msg) - 1);
1557 _exit(EXIT_SUCCESS);
1561 seccomp(unsigned int operation, unsigned int flags, void *args)
1563 return syscall(__NR_seccomp, operation, flags, args);
1566 /* The following is the x86\-64\-specific BPF boilerplate code for checking
1567 that the BPF program is running on the right architecture + ABI. At
1568 completion of these instructions, the accumulator contains the system
1571 /* For the x32 ABI, all system call numbers have bit 30 set */
1573 #define X32_SYSCALL_BIT 0x40000000
1575 #define X86_64_CHECK_ARCH_AND_LOAD_SYSCALL_NR \e
1576 BPF_STMT(BPF_LD | BPF_W | BPF_ABS, \e
1577 (offsetof(struct seccomp_data, arch))), \e
1578 BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, AUDIT_ARCH_X86_64, 0, 2), \e
1579 BPF_STMT(BPF_LD | BPF_W | BPF_ABS, \e
1580 (offsetof(struct seccomp_data, nr))), \e
1581 BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, X32_SYSCALL_BIT, 0, 1), \e
1582 BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL_PROCESS)
1584 /* installNotifyFilter() installs a seccomp filter that generates
1585 user\-space notifications (SECCOMP_RET_USER_NOTIF) when the process
1586 calls mkdir(2); the filter allows all other system calls.
1588 The function return value is a file descriptor from which the
1589 user\-space notifications can be fetched. */
1592 installNotifyFilter(void)
1594 struct sock_filter filter[] = {
1595 X86_64_CHECK_ARCH_AND_LOAD_SYSCALL_NR,
1597 /* mkdir() triggers notification to user\-space supervisor */
1599 BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_mkdir, 0, 1),
1600 BPF_STMT(BPF_RET + BPF_K, SECCOMP_RET_USER_NOTIF),
1602 /* Every other system call is allowed */
1604 BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW),
1607 struct sock_fprog prog = {
1608 .len = sizeof(filter) / sizeof(filter[0]),
1612 /* Install the filter with the SECCOMP_FILTER_FLAG_NEW_LISTENER flag;
1613 as a result, seccomp() returns a notification file descriptor. */
1615 int notifyFd = seccomp(SECCOMP_SET_MODE_FILTER,
1616 SECCOMP_FILTER_FLAG_NEW_LISTENER, &prog);
1617 if (notifyFd == \-1)
1618 errExit("seccomp\-install\-notify\-filter");
1623 /* Close a pair of sockets created by socketpair() */
1626 closeSocketPair(int sockPair[2])
1628 if (close(sockPair[0]) == \-1)
1629 errExit("closeSocketPair\-close\-0");
1630 if (close(sockPair[1]) == \-1)
1631 errExit("closeSocketPair\-close\-1");
1634 /* Implementation of the target process; create a child process that:
1636 (1) installs a seccomp filter with the
1637 SECCOMP_FILTER_FLAG_NEW_LISTENER flag;
1638 (2) writes the seccomp notification file descriptor returned from
1639 the previous step onto the UNIX domain socket, \(aqsockPair[0]\(aq;
1640 (3) calls mkdir(2) for each element of \(aqargv\(aq.
1642 The function return value in the parent is the PID of the child
1643 process; the child does not return from this function. */
1646 targetProcess(int sockPair[2], char *argv[])
1648 pid_t targetPid = fork();
1649 if (targetPid == \-1)
1652 if (targetPid > 0) /* In parent, return PID of child */
1655 /* Child falls through to here */
1657 printf("T: PID = %ld\en", (long) getpid());
1659 /* Install seccomp filter(s) */
1661 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0))
1664 int notifyFd = installNotifyFilter();
1666 /* Pass the notification file descriptor to the tracing process over
1667 a UNIX domain socket */
1669 if (sendfd(sockPair[0], notifyFd) == \-1)
1672 /* Notification and socket FDs are no longer needed in target */
1674 if (close(notifyFd) == \-1)
1675 errExit("close\-target\-notify\-fd");
1677 closeSocketPair(sockPair);
1679 /* Perform a mkdir() call for each of the command\-line arguments */
1681 for (char **ap = argv; *ap != NULL; ap++) {
1682 printf("\enT: about to mkdir(\e"%s\e")\en", *ap);
1684 int s = mkdir(*ap, 0700);
1686 perror("T: ERROR: mkdir(2)");
1688 printf("T: SUCCESS: mkdir(2) returned %d\en", s);
1691 printf("\enT: terminating\en");
1695 /* Check that the notification ID provided by a SECCOMP_IOCTL_NOTIF_RECV
1696 operation is still valid. It will no longer be valid if the target
1697 process has terminated or is no longer blocked in the system call that
1698 generated the notification (because it was interrupted by a signal).
1700 This operation can be used when doing such things as accessing
1701 /proc/PID files in the target process in order to avoid TOCTOU race
1702 conditions where the PID that is returned by SECCOMP_IOCTL_NOTIF_RECV
1703 terminates and is reused by another process. */
1706 cookieIsValid(int notifyFd, uint64_t id)
1708 return ioctl(notifyFd, SECCOMP_IOCTL_NOTIF_ID_VALID, &id) == 0;
1711 /* Access the memory of the target process in order to fetch the
1712 pathname referred to by the system call argument \(aqargNum\(aq in
1713 \(aqreq\->data.args[]\(aq. The pathname is returned in \(aqpath\(aq,
1714 a buffer of \(aqlen\(aq bytes allocated by the caller.
1716 Returns true if the pathname is successfully fetched, and false
1717 otherwise. For possible causes of failure, see the comments below. */
1720 getTargetPathname(struct seccomp_notif *req, int notifyFd,
1721 int argNum, char *path, size_t len)
1723 char procMemPath[PATH_MAX];
1725 snprintf(procMemPath, sizeof(procMemPath), "/proc/%d/mem", req\->pid);
1727 int procMemFd = open(procMemPath, O_RDONLY | O_CLOEXEC);
1728 if (procMemFd == \-1)
1731 /* Check that the process whose info we are accessing is still alive
1732 and blocked in the system call that caused the notification.
1733 If the SECCOMP_IOCTL_NOTIF_ID_VALID operation (performed in
1734 cookieIsValid()) succeeded, we know that the /proc/PID/mem file
1735 descriptor that we opened corresponded to the process for which we
1736 received a notification. If that process subsequently terminates,
1737 then read() on that file descriptor will return 0 (EOF). */
1739 if (!cookieIsValid(notifyFd, req\->id)) {
1744 /* Read bytes at the location containing the pathname argument */
1746 ssize_t nread = pread(procMemFd, path, len, req\->data.args[argNum]);
1753 /* Once again check that the notification ID is still valid. The
1754 case we are particularly concerned about here is that just
1755 before we fetched the pathname, the target\(aqs blocked system
1756 call was interrupted by a signal handler, and after the handler
1757 returned, the target carried on execution (past the interrupted
1758 system call). In that case, we have no guarantees about what we
1759 are reading, since the target\(aqs memory may have been arbitrarily
1760 changed by subsequent operations. */
1762 if (!cookieIsValid(notifyFd, req\->id)) {
1763 perror("\etS: notification ID check failed!!!");
1767 /* Even if the target\(aqs system call was not interrupted by a signal,
1768 we have no guarantees about what was in the memory of the target
1769 process. (The memory may have been modified by another thread, or
1770 even by an external attacking process.) We therefore treat the
1771 buffer returned by pread() as untrusted input. The buffer should
1772 contain a terminating null byte; if not, then we will trigger an
1773 error for the target process. */
1775 if (strnlen(path, nread) < nread)
1781 /* Allocate buffers for the seccomp user\-space notification request and
1782 response structures. It is the caller\(aqs responsibility to free the
1783 buffers returned via \(aqreq\(aq and \(aqresp\(aq. */
1786 allocSeccompNotifBuffers(struct seccomp_notif **req,
1787 struct seccomp_notif_resp **resp,
1788 struct seccomp_notif_sizes *sizes)
1790 /* Discover the sizes of the structures that are used to receive
1791 notifications and send notification responses, and allocate
1792 buffers of those sizes. */
1794 if (seccomp(SECCOMP_GET_NOTIF_SIZES, 0, sizes) == \-1)
1795 errExit("seccomp\-SECCOMP_GET_NOTIF_SIZES");
1797 *req = malloc(sizes\->seccomp_notif);
1799 errExit("malloc\-seccomp_notif");
1801 /* When allocating the response buffer, we must allow for the fact
1802 that the user\-space binary may have been built with user\-space
1803 headers where \(aqstruct seccomp_notif_resp\(aq is bigger than the
1804 response buffer expected by the (older) kernel. Therefore, we
1805 allocate a buffer that is the maximum of the two sizes. This
1806 ensures that if the supervisor places bytes into the response
1807 structure that are past the response size that the kernel expects,
1808 then the supervisor is not touching an invalid memory location. */
1810 size_t resp_size = sizes\->seccomp_notif_resp;
1811 if (sizeof(struct seccomp_notif_resp) > resp_size)
1812 resp_size = sizeof(struct seccomp_notif_resp);
1814 *resp = malloc(resp_size);
1816 errExit("malloc\-seccomp_notif_resp");
1820 /* Handle notifications that arrive via the SECCOMP_RET_USER_NOTIF file
1821 descriptor, \(aqnotifyFd\(aq. */
1824 handleNotifications(int notifyFd)
1826 struct seccomp_notif_sizes sizes;
1827 struct seccomp_notif *req;
1828 struct seccomp_notif_resp *resp;
1829 char path[PATH_MAX];
1831 allocSeccompNotifBuffers(&req, &resp, &sizes);
1833 /* Loop handling notifications */
1837 /* Wait for next notification, returning info in \(aq*req\(aq */
1839 memset(req, 0, sizes.seccomp_notif);
1840 if (ioctl(notifyFd, SECCOMP_IOCTL_NOTIF_RECV, req) == \-1) {
1843 errExit("\etS: ioctl\-SECCOMP_IOCTL_NOTIF_RECV");
1846 printf("\etS: got notification (ID %#llx) for PID %d\en",
1847 req\->id, req\->pid);
1849 /* The only system call that can generate a notification event
1850 is mkdir(2). Nevertheless, we check that the notified system
1851 call is indeed mkdir() as kind of future\-proofing of this
1852 code in case the seccomp filter is later modified to
1853 generate notifications for other system calls. */
1855 if (req\->data.nr != __NR_mkdir) {
1856 printf("\etS: notification contained unexpected "
1857 "system call number; bye!!!\en");
1861 bool pathOK = getTargetPathname(req, notifyFd, 0, path,
1864 /* Prepopulate some fields of the response */
1866 resp\->id = req\->id; /* Response includes notification ID */
1870 /* If getTargetPathname() failed, trigger an EINVAL error
1871 response (sending this response may yield an error if the
1872 failure occurred because the notification ID was no longer
1873 valid); if the directory is in /tmp, then create it on behalf
1874 of the supervisor; if the pathname starts with \(aq.\(aq, tell the
1875 kernel to let the target process execute the mkdir();
1876 otherwise, give an error for a directory pathname in any other
1880 resp->error = -EINVAL;
1881 printf("\etS: spoofing error for invalid pathname (%s)\en",
1882 strerror(-resp->error));
1883 } else if (strncmp(path, "/tmp/", strlen("/tmp/")) == 0) {
1884 printf("\etS: executing: mkdir(\e"%s\e", %#llo)\en",
1885 path, req\->data.args[1]);
1887 if (mkdir(path, req\->data.args[1]) == 0) {
1888 resp\->error = 0; /* "Success" */
1889 resp\->val = strlen(path); /* Used as return value of
1890 mkdir() in target */
1891 printf("\etS: success! spoofed return = %lld\en",
1895 /* If mkdir() failed in the supervisor, pass the error
1896 back to the target */
1898 resp\->error = \-errno;
1899 printf("\etS: failure! (errno = %d; %s)\en", errno,
1902 } else if (strncmp(path, "./", strlen("./")) == 0) {
1903 resp\->error = resp\->val = 0;
1904 resp\->flags = SECCOMP_USER_NOTIF_FLAG_CONTINUE;
1905 printf("\etS: target can execute system call\en");
1907 resp\->error = \-EOPNOTSUPP;
1908 printf("\etS: spoofing error response (%s)\en",
1909 strerror(\-resp\->error));
1912 /* Send a response to the notification */
1914 printf("\etS: sending response "
1915 "(flags = %#x; val = %lld; error = %d)\en",
1916 resp\->flags, resp\->val, resp\->error);
1918 if (ioctl(notifyFd, SECCOMP_IOCTL_NOTIF_SEND, resp) == \-1) {
1919 if (errno == ENOENT)
1920 printf("\etS: response failed with ENOENT; "
1921 "perhaps target process\(aqs syscall was "
1922 "interrupted by a signal?\en");
1924 perror("ioctl\-SECCOMP_IOCTL_NOTIF_SEND");
1927 /* If the pathname is just "/bye", then the supervisor breaks out
1928 of the loop and terminates. This allows us to see what happens
1929 if the target process makes further calls to mkdir(2). */
1931 if (strcmp(path, "/bye") == 0)
1937 printf("\etS: terminating **********\en");
1941 /* Implementation of the supervisor process:
1943 (1) obtains the notification file descriptor from \(aqsockPair[1]\(aq
1944 (2) handles notifications that arrive on that file descriptor. */
1947 supervisor(int sockPair[2])
1949 int notifyFd = recvfd(sockPair[1]);
1950 if (notifyFd == \-1)
1953 closeSocketPair(sockPair); /* We no longer need the socket pair */
1955 handleNotifications(notifyFd);
1959 main(int argc, char *argv[])
1963 setbuf(stdout, NULL);
1966 fprintf(stderr, "At least one pathname argument is required\en");
1970 /* Create a UNIX domain socket that is used to pass the seccomp
1971 notification file descriptor from the target process to the
1972 supervisor process. */
1974 if (socketpair(AF_UNIX, SOCK_STREAM, 0, sockPair) == \-1)
1975 errExit("socketpair");
1977 /* Create a child process\-\-the "target"\-\-that installs seccomp
1978 filtering. The target process writes the seccomp notification
1979 file descriptor onto \(aqsockPair[0]\(aq and then calls mkdir(2) for
1980 each directory in the command\-line arguments. */
1982 (void) targetProcess(sockPair, &argv[optind]);
1984 /* Catch SIGCHLD when the target terminates, so that the
1985 supervisor can also terminate. */
1987 struct sigaction sa;
1988 sa.sa_handler = sigchldHandler;
1990 sigemptyset(&sa.sa_mask);
1991 if (sigaction(SIGCHLD, &sa, NULL) == \-1)
1992 errExit("sigaction");
1994 supervisor(sockPair);
2002 .BR pidfd_getfd (2),
2005 A further example program can be found in the kernel source file
2006 .IR samples/seccomp/user-trap.c .