1 .\" Copyright (C) 2014 Kees Cook <keescook@chromium.org>
2 .\" and Copyright (C) 2012 Will Drewry <wad@chromium.org>
3 .\" and Copyright (C) 2008, 2014,2017 Michael Kerrisk <mtk.manpages@gmail.com>
4 .\" and Copyright (C) 2017 Tyler Hicks <tyhicks@canonical.com>
5 .\" and Copyright (C) 2020 Tycho Andersen <tycho@tycho.ws>
7 .\" %%%LICENSE_START(VERBATIM)
8 .\" Permission is granted to make and distribute verbatim copies of this
9 .\" manual provided the copyright notice and this permission notice are
10 .\" preserved on all copies.
12 .\" Permission is granted to copy and distribute modified versions of this
13 .\" manual under the conditions for verbatim copying, provided that the
14 .\" entire resulting derived work is distributed under the terms of a
15 .\" permission notice identical to this one.
17 .\" Since the Linux kernel and libraries are constantly changing, this
18 .\" manual page may be incorrect or out-of-date. The author(s) assume no
19 .\" responsibility for errors or omissions, or for damages resulting from
20 .\" the use of the information contained herein. The author(s) may not
21 .\" have taken the same level of care in the production of this manual,
22 .\" which is licensed free of charge, as they might when working
25 .\" Formatted or processed versions of this manual, if unaccompanied by
26 .\" the source, must acknowledge the copyright and authors of this work.
29 .TH SECCOMP 2 2021-03-22 "Linux" "Linux Programmer's Manual"
31 seccomp \- operate on Secure Computing state of the process
34 .B #include <linux/seccomp.h>
35 .B #include <linux/filter.h>
36 .B #include <linux/audit.h>
37 .B #include <linux/signal.h>
38 .B #include <sys/ptrace.h>
39 .\" Kees Cook noted: Anything that uses SECCOMP_RET_TRACE returns will
40 .\" need <sys/ptrace.h>
42 .BI "int seccomp(unsigned int " operation ", unsigned int " flags \
47 There is no glibc wrapper for this system call; see NOTES.
51 system call operates on the Secure Computing (seccomp) state of the
54 Currently, Linux supports the following
58 .BR SECCOMP_SET_MODE_STRICT
59 The only system calls that the calling thread is permitted to make are
67 Other system calls result in the delivery of a
70 Strict secure computing mode is useful for number-crunching
71 applications that may need to execute untrusted byte code, perhaps
72 obtained by reading from a pipe or socket.
74 Note that although the calling thread can no longer call
78 to block all signals apart from
84 (for example) is not sufficient for restricting the process's execution time.
85 Instead, to reliably terminate the process,
88 This can be done by using
98 to set the hard limit for
101 This operation is available only if the kernel is configured with
111 This operation is functionally identical to the call:
115 prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT);
119 .BR SECCOMP_SET_MODE_FILTER
120 The system calls allowed are defined by a pointer to a Berkeley Packet
121 Filter (BPF) passed via
123 This argument is a pointer to a
124 .IR "struct\ sock_fprog" ;
125 it can be designed to filter arbitrary system calls and system call
127 If the filter is invalid,
138 is allowed by the filter, any child processes will be constrained to
139 the same system call filters as the parent.
143 the existing filters will be preserved across a call to
147 .BR SECCOMP_SET_MODE_FILTER
148 operation, either the calling thread must have the
150 capability in its user namespace, or the thread must already have the
153 If that bit was not already set by an ancestor of this thread,
154 the thread must make the following call:
158 prctl(PR_SET_NO_NEW_PRIVS, 1);
163 .BR SECCOMP_SET_MODE_FILTER
164 operation fails and returns
168 This requirement ensures that an unprivileged process cannot apply
169 a malicious filter and then invoke a set-user-ID or
170 other privileged program using
172 thus potentially compromising that program.
173 (Such a malicious filter might, for example, cause an attempt to use
175 to set the caller's user IDs to nonzero values to instead
176 return 0 without actually making the system call.
177 Thus, the program might be tricked into retaining superuser privileges
178 in circumstances where it is possible to influence it to do
179 dangerous things because it did not actually drop privileges.)
185 is allowed by the attached filter, further filters may be added.
186 This will increase evaluation time, but allows for further reduction of
187 the attack surface during execution of a thread.
190 .BR SECCOMP_SET_MODE_FILTER
191 operation is available only if the kernel is configured with
192 .BR CONFIG_SECCOMP_FILTER
197 is 0, this operation is functionally identical to the call:
201 prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, args);
210 .BR SECCOMP_FILTER_FLAG_LOG " (since Linux 4.14)"
211 .\" commit e66a39977985b1e69e17c4042cb290768eca9b02
212 All filter return actions except
213 .BR SECCOMP_RET_ALLOW
215 An administrator may override this filter flag by preventing specific
216 actions from being logged via the
217 .IR /proc/sys/kernel/seccomp/actions_logged
220 .BR SECCOMP_FILTER_FLAG_SPEC_ALLOW " (since Linux 4.17)"
221 .\" commit 00a02d0c502a06d15e07b857f8ff921e3e402675
222 Disable Speculative Store Bypass mitigation.
224 .BR SECCOMP_FILTER_FLAG_TSYNC
225 When adding a new filter, synchronize all other threads of the calling
226 process to the same seccomp filter tree.
227 A "filter tree" is the ordered list of filters attached to a thread.
228 (Attaching identical filters in separate
230 calls results in different filters from this perspective.)
232 If any thread cannot synchronize to the same filter tree,
233 the call will not attach the new seccomp filter,
234 and will fail, returning the first thread ID found that cannot synchronize.
235 Synchronization will fail if another thread in the same process is in
236 .BR SECCOMP_MODE_STRICT
237 or if it has attached new seccomp filters to itself,
238 diverging from the calling thread's filter tree.
241 .BR SECCOMP_GET_ACTION_AVAIL " (since Linux 4.14)"
242 .\" commit d612b1fd8010d0d67b5287fe146b8b55bcbb8655
243 Test to see if an action is supported by the kernel.
244 This operation is helpful to confirm that the kernel knows
245 of a more recently added filter return action
246 since the kernel treats all unknown actions as
247 .BR SECCOMP_RET_KILL_PROCESS .
253 must be a pointer to an unsigned 32-bit filter return action.
255 .BR SECCOMP_GET_NOTIF_SIZES " (since Linux 5.0)"
256 .\" commit 6a21cc50f0c7f87dae5259f6cfefe024412313f6
257 Get the sizes of the seccomp user-space notification structures.
258 Since the structures include a
259 .I "struct seccomp_data"
260 which may grow in size, this command can be used to determine how
261 much memory to allocate for sending and receiving notifications.
267 must be a pointer to a
268 .IR "struct seccomp_notif_sizes" ,
269 which has the following form:
273 struct seccomp_notif_sizes
275 __u16 seccomp_notif_resp;
282 .BR seccomp_unotify (2)
286 When adding filters via
287 .BR SECCOMP_SET_MODE_FILTER ,
289 points to a filter program:
294 unsigned short len; /* Number of BPF instructions */
295 struct sock_filter *filter; /* Pointer to array of
301 Each program must contain one or more BPF instructions:
305 struct sock_filter { /* Filter block */
306 __u16 code; /* Actual filter code */
307 __u8 jt; /* Jump true */
308 __u8 jf; /* Jump false */
309 __u32 k; /* Generic multiuse field */
314 When executing the instructions, the BPF program operates on the
315 system call information made available (i.e., use the
317 addressing mode) as a (read-only)
318 .\" Quoting Kees Cook:
319 .\" If BPF even allows changing the data, it's not copied back to
320 .\" the syscall when it runs. Anything wanting to do things like
321 .\" that would need to use ptrace to catch the call and directly
322 .\" modify the registers before continuing with the call.
323 buffer of the following form:
327 struct seccomp_data {
328 int nr; /* System call number */
329 __u32 arch; /* AUDIT_ARCH_* value
330 (see <linux/audit.h>) */
331 __u64 instruction_pointer; /* CPU instruction pointer */
332 __u64 args[6]; /* Up to 6 system call arguments */
337 Because numbering of system calls varies between architectures and
338 some architectures (e.g., x86-64) allow user-space code to use
339 the calling conventions of multiple architectures
340 (and the convention being used may vary over the life of a process that uses
342 to execute binaries that employ the different conventions),
343 it is usually necessary to verify the value of the
347 It is strongly recommended to use an allow-list approach whenever
348 possible because such an approach is more robust and simple.
349 A deny-list will have to be updated whenever a potentially
350 dangerous system call is added (or a dangerous flag or option if those
351 are deny-listed), and it is often possible to alter the
352 representation of a value without altering its meaning, leading to
360 field is not unique for all calling conventions.
361 The x86-64 ABI and the x32 ABI both use
362 .BR AUDIT_ARCH_X86_64
365 and they run on the same processors.
367 .BR __X32_SYSCALL_BIT
368 is used on the system call number to tell the two ABIs apart.
369 .\" As noted by Dave Drysdale in a note at the end of
370 .\" https://lwn.net/Articles/604515/
371 .\" One additional detail to point out for the x32 ABI case:
372 .\" the syscall number gets a high bit set (__X32_SYSCALL_BIT),
373 .\" to mark it as an x32 call.
375 .\" If x32 support is included in the kernel, then __SYSCALL_MASK
376 .\" will have a value that is not all-ones, and this will trigger
377 .\" an extra instruction in system_call to mask off the extra bit,
378 .\" so that the syscall table indexing still works.
380 This means that a policy must either deny all syscalls with
381 .BR __X32_SYSCALL_BIT
382 or it must recognize syscalls with and without
383 .BR __X32_SYSCALL_BIT
385 A list of system calls to be denied based on
387 that does not also contain
390 .BR __X32_SYSCALL_BIT
391 set can be bypassed by a malicious program that sets
392 .BR __X32_SYSCALL_BIT .
394 Additionally, kernels prior to Linux 5.4 incorrectly permitted
396 in the ranges 512-547 as well as the corresponding non-x32 syscalls ORed
398 .BR __X32_SYSCALL_BIT .
404 .BR __X32_SYSCALL_BIT )
405 would result in invocations of
407 with potentially confused x32-vs-x86_64 semantics in the kernel.
408 Policies intended to work on kernels before Linux 5.4 must ensure that they
409 deny or otherwise correctly handle these system calls.
410 On Linux 5.4 and newer,
411 .\" commit 6365b842aae4490ebfafadfc6bb27a6d3cc54757
412 such system calls will fail with the error
414 without doing anything.
417 .I instruction_pointer
418 field provides the address of the machine-language instruction that
419 performed the system call.
420 This might be useful in conjunction with the use of
422 to perform checks based on which region (mapping) of the program
423 made the system call.
424 (Probably, it is wise to lock down the
428 system calls to prevent the program from subverting such checks.)
430 When checking values from
432 keep in mind that arguments are often
433 silently truncated before being processed, but after the seccomp check.
434 For example, this happens if the i386 ABI is used on an
435 x86-64 kernel: although the kernel will normally not look beyond
436 the 32 lowest bits of the arguments, the values of the full
437 64-bit registers will be present in the seccomp data.
438 A less surprising example is that if the x86-64 ABI is used to perform
439 a system call that takes an argument of type
441 the more-significant half of the argument register is ignored by
442 the system call, but visible in the seccomp data.
444 A seccomp filter returns a 32-bit value consisting of two parts:
445 the most significant 16 bits
446 (corresponding to the mask defined by the constant
447 .BR SECCOMP_RET_ACTION_FULL )
448 contain one of the "action" values listed below;
449 the least significant 16-bits (defined by the constant
450 .BR SECCOMP_RET_DATA )
451 are "data" to be associated with this return value.
453 If multiple filters exist, they are \fIall\fP executed,
454 in reverse order of their addition to the filter tree\(emthat is,
455 the most recently installed filter is executed first.
456 (Note that all filters will be called
457 even if one of the earlier filters returns
458 .BR SECCOMP_RET_KILL .
459 This is done to simplify the kernel code and to provide a
460 tiny speed-up in the execution of sets of filters by
461 avoiding a check for this uncommon case.)
462 .\" From an Aug 2015 conversation with Kees Cook where I asked why *all*
463 .\" filters are applied even if one of the early filters returns
464 .\" SECCOMP_RET_KILL:
466 .\" It's just because it would be an optimization that would only speed up
467 .\" the RET_KILL case, but it's the uncommon one and the one that doesn't
468 .\" benefit meaningfully from such a change (you need to kill the process
469 .\" really quickly?). We would speed up killing a program at the (albeit
470 .\" tiny) expense to all other filtered programs. Best to keep the filter
471 .\" execution logic clear, simple, and as fast as possible for all
473 The return value for the evaluation of a given system call is the first-seen
474 action value of highest precedence (along with its accompanying data)
475 returned by execution of all of the filters.
477 In decreasing order of precedence,
478 the action values that may be returned by a seccomp filter are:
480 .BR SECCOMP_RET_KILL_PROCESS " (since Linux 4.14)"
481 .\" commit 4d3b0b05aae9ee9ce0970dc4cc0fb3fad5e85945
482 .\" commit 0466bdb99e8744bc9befa8d62a317f0fd7fd7421
483 This value results in immediate termination of the process,
485 The system call is not executed.
487 .BR SECCOMP_RET_KILL_THREAD
488 below, all threads in the thread group are terminated.
489 (For a discussion of thread groups, see the description of the
494 The process terminates
499 Even if a signal handler has been registered for
501 the handler will be ignored in this case and the process always terminates.
502 To a parent process that is waiting on this process (using
504 or similar), the returned
506 will indicate that its child was terminated as though by a
510 .BR SECCOMP_RET_KILL_THREAD " (or " SECCOMP_RET_KILL )
511 This value results in immediate termination of the thread
512 that made the system call.
513 The system call is not executed.
514 Other threads in the same thread group will continue to execute.
516 The thread terminates
522 .BR SECCOMP_RET_KILL_PROCESS
525 .\" See these commits:
526 .\" seccomp: dump core when using SECCOMP_RET_KILL
527 .\" (b25e67161c295c98acda92123b2dd1e7d8642901)
528 .\" seccomp: Only dump core when single-threaded
529 .\" (d7276e321ff8a53106a59c85ca46d03e34288893)
531 any process terminated in this way would not trigger a coredump
536 as having a default action of termination with a core dump).
538 a single-threaded process will dump core if terminated in this way.
541 .BR SECCOMP_RET_KILL_PROCESS
543 .BR SECCOMP_RET_KILL_THREAD
544 was added as a synonym for
545 .BR SECCOMP_RET_KILL ,
546 in order to more clearly distinguish the two actions.
550 .BR SECCOMP_RET_KILL_THREAD
551 to kill a single thread in a multithreaded process is likely to leave the
552 process in a permanently inconsistent and possibly corrupt state.
555 This value results in the kernel sending a thread-directed
557 signal to the triggering thread.
558 (The system call is not executed.)
559 Various fields will be set in the
563 associated with signal:
571 will show the address of the system call instruction.
576 will indicate which system call was attempted.
585 portion of the filter return value.
588 The program counter will be as though the system call happened
589 (i.e., the program counter will not point to the system call instruction).
590 The return value register will contain an architecture\-dependent value;
591 if resuming execution, set it to something appropriate for the system call.
592 (The architecture dependency is because replacing it with
594 could overwrite some useful information.)
596 .BR SECCOMP_RET_ERRNO
597 This value results in the
599 portion of the filter's return value being passed to user space as the
601 value without executing the system call.
603 .BR SECCOMP_RET_USER_NOTIF " (since Linux 5.0)"
604 .\" commit 6a21cc50f0c7f87dae5259f6cfefe024412313f6
605 Forward the system call to an attached user-space listening
606 process to allow that process to decide what to do with the system call.
607 If there is no attached listener (either
608 because the filter was not installed with the
609 .BR SECCOMP_FILTER_FLAG_NEW_LISTENER
610 flag or because the file descriptor was closed), the filter returns
612 (similar to what happens when a filter returns
613 .BR SECCOMP_RET_TRACE
614 and there is no tracer).
616 .BR seccomp_unotify (2)
619 .BR SECCOMP_RET_TRACE
620 When returned, this value will cause the kernel to attempt to notify a
622 tracer prior to executing the system call.
623 If there is no tracer present,
624 the system call is not executed and returns a failure status with
629 A tracer will be notified if it requests
630 .BR PTRACE_O_TRACESECCOMP
632 .IR ptrace(PTRACE_SETOPTIONS) .
633 The tracer will be notified of a
634 .BR PTRACE_EVENT_SECCOMP
637 portion of the filter's return value will be available to the tracer via
638 .BR PTRACE_GETEVENTMSG .
640 The tracer can skip the system call by changing the system call number
642 Alternatively, the tracer can change the system call
643 requested by changing the system call to a valid system call number.
644 If the tracer asks to skip the system call, then the system call will
645 appear to return the value that the tracer puts in the return value register.
647 .\" This was changed in ce6526e8afa4.
648 .\" A related hole, using PTRACE_SYSCALL instead of SECCOMP_RET_TRACE, was
649 .\" changed in arch-specific commits, e.g. 93e35efb8de4 for X86 and
650 .\" 0f3912fd934c for ARM.
651 Before kernel 4.8, the seccomp check will not be run again after the tracer is
653 (This means that, on older kernels, seccomp-based sandboxes
656 .BR ptrace (2)\(emeven
658 sandboxed processes\(emwithout extreme care;
659 ptracers can use this mechanism to escape from the seccomp sandbox.)
661 Note that a tracer process will not be notified
662 if another filter returns an action value with a precedence greater than
663 .BR SECCOMP_RET_TRACE .
665 .BR SECCOMP_RET_LOG " (since Linux 4.14)"
666 .\" commit 59f5cf44a38284eb9e76270c786fb6cc62ef8ac4
667 This value results in the system call being executed after
668 the filter return action is logged.
669 An administrator may override the logging of this action via
671 .IR /proc/sys/kernel/seccomp/actions_logged
674 .BR SECCOMP_RET_ALLOW
675 This value results in the system call being executed.
677 If an action value other than one of the above is specified,
678 then the filter action is treated as either
679 .BR SECCOMP_RET_KILL_PROCESS
681 .\" commit 4d3b0b05aae9ee9ce0970dc4cc0fb3fad5e85945
683 .BR SECCOMP_RET_KILL_THREAD
684 (in Linux 4.13 and earlier).
687 The files in the directory
688 .IR /proc/sys/kernel/seccomp
689 provide additional seccomp information and configuration:
691 .IR actions_avail " (since Linux 4.14)"
692 .\" commit 8e5f1ad116df6b0de65eac458d5e7c318d1c05af
693 A read-only ordered list of seccomp filter return actions in string form.
694 The ordering, from left-to-right, is in decreasing order of precedence.
695 The list represents the set of seccomp filter return actions
696 supported by the kernel.
698 .IR actions_logged " (since Linux 4.14)"
699 .\" commit 0ddec0fc8900201c0897b87b762b7c420436662f
700 A read-write ordered list of seccomp filter return actions that
701 are allowed to be logged.
702 Writes to the file do not need to be in ordered form but reads from
703 the file will be ordered in the same way as the
707 It is important to note that the value of
709 does not prevent certain filter return actions from being logged when
710 the audit subsystem is configured to audit a task.
711 If the action is not found in the
713 file, the final decision on whether to audit the action for that task is
714 ultimately left up to the audit subsystem to decide for all filter return
716 .BR SECCOMP_RET_ALLOW .
718 The "allow" string is not accepted in the
720 file as it is not possible to log
721 .BR SECCOMP_RET_ALLOW
723 Attempting to write "allow" to the file will fail with the error
726 .SS Audit logging of seccomp actions
727 .\" commit 59f5cf44a38284eb9e76270c786fb6cc62ef8ac4
728 Since Linux 4.14, the kernel provides the facility to log the
729 actions returned by seccomp filters in the audit log.
730 The kernel makes the decision to log an action based on
731 the action type, whether or not the action is present in the
733 file, and whether kernel auditing is enabled
734 (e.g., via the kernel boot option
736 .\" or auditing could be enabled via the netlink API (AUDIT_SET)
737 The rules are as follows:
740 .BR SECCOMP_RET_ALLOW ,
741 the action is not logged.
743 Otherwise, if the action is either
744 .BR SECCOMP_RET_KILL_PROCESS
746 .BR SECCOMP_RET_KILL_THREAD ,
747 and that action appears in the
749 file, the action is logged.
751 Otherwise, if the filter has requested logging (the
752 .BR SECCOMP_FILTER_FLAG_LOG
754 and the action appears in the
756 file, the action is logged.
758 Otherwise, if kernel auditing is enabled and the process is being audited
760 the action is logged.
762 Otherwise, the action is not logged.
768 .BR SECCOMP_FILTER_FLAG_TSYNC
770 the return value is the ID of the thread
771 that caused the synchronization failure.
772 (This ID is a kernel thread ID of the type returned by
776 On other errors, \-1 is returned, and
778 is set to indicate the error.
781 can fail for the following reasons:
784 The caller did not have the
786 capability in its user namespace, or had not set
789 .BR SECCOMP_SET_MODE_FILTER .
793 was not a valid address.
797 is unknown or is not supported by this kernel version or configuration.
802 are invalid for the given
809 but the specified offset was not aligned to a 32-bit boundary or exceeded
810 .IR "sizeof(struct\ seccomp_data)" .
813 .\" See kernel/seccomp.c::seccomp_may_assign_mode() in 3.18 sources
814 A secure computing mode has already been set, and
816 differs from the existing setting.
821 .BR SECCOMP_SET_MODE_FILTER ,
822 but the filter program pointed to by
824 was not valid or the length of the filter program was zero or exceeded
832 .\" ENOMEM in kernel/seccomp.c::seccomp_attach_filter() in 3.18 sources
833 The total length of all filter programs attached
834 to the calling thread would exceed
835 .B MAX_INSNS_PER_PATH
836 (32768) instructions.
837 Note that for the purposes of calculating this limit,
838 each already existing filter program incurs an
839 overhead penalty of 4 instructions.
844 .BR SECCOMP_GET_ACTION_AVAIL ,
845 but the kernel does not support the filter return action specified by
849 Another thread caused a failure during thread sync, but its ID could not
854 system call first appeared in Linux 3.17.
855 .\" FIXME . Add glibc version
859 system call is a nonstandard Linux extension.
861 Rather than hand-coding seccomp filters as shown in the example below,
862 you may prefer to employ the
864 library, which provides a front-end for generating seccomp filters.
869 .IR /proc/[pid]/status
870 file provides a method of viewing the seccomp mode of a process; see
874 provides a superset of the functionality provided by the
877 operation (which does not support
882 .B PTRACE_SECCOMP_GET_FILTER
883 operation can be used to dump a process's seccomp filters.
885 .SS Architecture support for seccomp BPF
886 Architecture support for seccomp BPF filtering
887 .\" Check by grepping for HAVE_ARCH_SECCOMP_FILTER in Kconfig files in
888 .\" kernel source. Last checked in Linux 4.16-rc source.
889 is available on the following architectures:
891 x86-64, i386, x32 (since Linux 3.5)
894 ARM (since Linux 3.8)
896 s390 (since Linux 3.8)
898 MIPS (since Linux 3.16)
900 ARM-64 (since Linux 3.19)
902 PowerPC (since Linux 4.3)
904 Tile (since Linux 4.3)
906 PA-RISC (since Linux 4.6)
907 .\" User mode Linux since Linux 4.6
910 Glibc does not provide a wrapper for this system call; call it using
914 There are various subtleties to consider when applying seccomp filters
915 to a program, including the following:
917 Some traditional system calls have user-space implementations in the
919 on many architectures.
920 Notable examples include
921 .BR clock_gettime (2),
922 .BR gettimeofday (2),
925 On such architectures,
926 seccomp filtering for these system calls will have no effect.
927 (However, there are cases where the
929 implementations may fall back to invoking the true system call,
930 in which case seccomp filters would see the system call.)
932 Seccomp filtering is based on system call numbers.
933 However, applications typically do not directly invoke system calls,
934 but instead call wrapper functions in the C library which
935 in turn invoke the system calls.
936 Consequently, one must be aware of the following:
939 The glibc wrappers for some traditional system calls may actually
940 employ system calls with different names in the kernel.
943 wrapper function actually employs the
947 wrapper function actually calls
950 The behavior of wrapper functions may vary across architectures,
951 according to the range of system calls provided on those architectures.
952 In other words, the same wrapper function may invoke
953 different system calls on different architectures.
955 Finally, the behavior of wrapper functions can change across glibc versions.
956 For example, in older versions, the glibc wrapper function for
958 invoked the system call of the same name,
959 but starting in glibc 2.26, the implementation switched to calling
961 on all architectures.
964 The consequence of the above points is that it may be necessary
965 to filter for a system call other than might be expected.
966 Various manual pages in Section 2 provide helpful details
967 about the differences between wrapper functions and
968 the underlying system calls in subsections entitled
969 .IR "C library/kernel differences" .
971 Furthermore, note that the application of seccomp filters
972 even risks causing bugs in an application,
973 when the filters cause unexpected failures for legitimate operations
974 that the application might need to perform.
975 Such bugs may not easily be discovered when testing the seccomp
976 filters if the bugs occur in rarely used application code paths.
978 .SS Seccomp-specific BPF details
979 Note the following BPF details specific to seccomp filters:
985 size modifiers are not supported: all operations must load and store
989 To access the contents of the
993 addressing mode modifier.
997 addressing mode modifier yields an immediate mode operand
998 whose value is the size of the
1002 The program below accepts four or more arguments.
1003 The first three arguments are a system call number,
1004 a numeric architecture identifier, and an error number.
1005 The program uses these values to construct a BPF filter
1006 that is used at run time to perform the following checks:
1008 If the program is not running on the specified architecture,
1009 the BPF filter causes system calls to fail with the error
1012 If the program attempts to execute the system call with the specified number,
1013 the BPF filter causes the system call to fail, with
1015 being set to the specified error number.
1017 The remaining command-line arguments specify
1018 the pathname and additional arguments of a program
1019 that the example program should attempt to execute using
1021 (a library function that employs the
1024 Some example runs of the program are shown below.
1026 First, we display the architecture that we are running on (x86-64)
1027 and then construct a shell function that looks up system call
1028 numbers on this architecture:
1035 cat /usr/src/linux/arch/x86/syscalls/syscall_64.tbl | \e
1036 awk \(aq$2 != "x32" && $3 == "\(aq$1\(aq" { print $1 }\(aq
1041 When the BPF filter rejects a system call (case [2] above),
1042 it causes the system call to fail with the error number
1043 specified on the command line.
1044 In the experiments shown here, we'll use error number 99:
1049 EADDRNOTAVAIL 99 Cannot assign requested address
1053 In the following example, we attempt to run the command
1055 but the BPF filter rejects the
1057 system call, so that the command is not even executed:
1061 $ \fBsyscall_nr execve\fP
1064 Usage: ./a.out <syscall_nr> <arch> <errno> <prog> [<args>]
1065 Hint for <arch>: AUDIT_ARCH_I386: 0x40000003
1066 AUDIT_ARCH_X86_64: 0xC000003E
1067 $ \fB./a.out 59 0xC000003E 99 /bin/whoami\fP
1068 execv: Cannot assign requested address
1072 In the next example, the BPF filter rejects the
1074 system call, so that, although it is successfully started, the
1076 command is not able to write output:
1080 $ \fBsyscall_nr write\fP
1082 $ \fB./a.out 1 0xC000003E 99 /bin/whoami\fP
1086 In the final example,
1087 the BPF filter rejects a system call that is not used by the
1089 command, so it is able to successfully execute and produce output:
1093 $ \fBsyscall_nr preadv\fP
1095 $ \fB./a.out 295 0xC000003E 99 /bin/whoami\fP
1106 #include <linux/audit.h>
1107 #include <linux/filter.h>
1108 #include <linux/seccomp.h>
1109 #include <sys/prctl.h>
1111 #define X32_SYSCALL_BIT 0x40000000
1112 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
1115 install_filter(int syscall_nr, int t_arch, int f_errno)
1117 unsigned int upper_nr_limit = 0xffffffff;
1119 /* Assume that AUDIT_ARCH_X86_64 means the normal x86\-64 ABI
1120 (in the x32 ABI, all system calls have bit 30 set in the
1121 \(aqnr\(aq field, meaning the numbers are >= X32_SYSCALL_BIT). */
1122 if (t_arch == AUDIT_ARCH_X86_64)
1123 upper_nr_limit = X32_SYSCALL_BIT \- 1;
1125 struct sock_filter filter[] = {
1126 /* [0] Load architecture from \(aqseccomp_data\(aq buffer into
1128 BPF_STMT(BPF_LD | BPF_W | BPF_ABS,
1129 (offsetof(struct seccomp_data, arch))),
1131 /* [1] Jump forward 5 instructions if architecture does not
1132 match \(aqt_arch\(aq. */
1133 BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, t_arch, 0, 5),
1135 /* [2] Load system call number from \(aqseccomp_data\(aq buffer into
1137 BPF_STMT(BPF_LD | BPF_W | BPF_ABS,
1138 (offsetof(struct seccomp_data, nr))),
1140 /* [3] Check ABI \- only needed for x86\-64 in deny\-list use
1141 cases. Use BPF_JGT instead of checking against the bit
1142 mask to avoid having to reload the syscall number. */
1143 BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, upper_nr_limit, 3, 0),
1145 /* [4] Jump forward 1 instruction if system call number
1146 does not match \(aqsyscall_nr\(aq. */
1147 BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, syscall_nr, 0, 1),
1149 /* [5] Matching architecture and system call: don\(aqt execute
1150 the system call, and return \(aqf_errno\(aq in \(aqerrno\(aq. */
1151 BPF_STMT(BPF_RET | BPF_K,
1152 SECCOMP_RET_ERRNO | (f_errno & SECCOMP_RET_DATA)),
1154 /* [6] Destination of system call number mismatch: allow other
1156 BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW),
1158 /* [7] Destination of architecture mismatch: kill process. */
1159 BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL_PROCESS),
1162 struct sock_fprog prog = {
1163 .len = ARRAY_SIZE(filter),
1167 if (seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog)) {
1176 main(int argc, char **argv)
1179 fprintf(stderr, "Usage: "
1180 "%s <syscall_nr> <arch> <errno> <prog> [<args>]\en"
1181 "Hint for <arch>: AUDIT_ARCH_I386: 0x%X\en"
1182 " AUDIT_ARCH_X86_64: 0x%X\en"
1183 "\en", argv[0], AUDIT_ARCH_I386, AUDIT_ARCH_X86_64);
1187 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1192 if (install_filter(strtol(argv[1], NULL, 0),
1193 strtol(argv[2], NULL, 0),
1194 strtol(argv[3], NULL, 0)))
1197 execv(argv[4], &argv[4]);
1213 Various pages from the
1216 .BR scmp_sys_resolver (1),
1217 .BR seccomp_export_bpf (3),
1218 .BR seccomp_init (3),
1219 .BR seccomp_load (3),
1221 .BR seccomp_rule_add (3).
1223 The kernel source files
1224 .IR Documentation/networking/filter.txt
1226 .IR Documentation/userspace\-api/seccomp_filter.rst
1227 .\" commit c061f33f35be0ccc80f4b8e0aea5dfd2ed7e01a3
1229 .IR Documentation/prctl/seccomp_filter.txt
1232 McCanne, S.\& and Jacobson, V.\& (1992)
1233 .IR "The BSD Packet Filter: A New Architecture for User-level Packet Capture" ,
1234 Proceedings of the USENIX Winter 1993 Conference
1235 .UR http://www.tcpdump.org/papers/bpf\-usenix93.pdf