seccomp

When adding filters via SECCOMP_SET_MODE_FILTER, args points to a filter program:

struct sock_fprog {
    unsigned short      len;    /* Number of BPF instructions */
    struct sock_filter *filter; /* Pointer to array of
                                   BPF instructions */
};

Each program must contain one or more BPF instructions:

struct sock_filter {            /* Filter block */
    __u16 code;                 /* Actual filter code */
    __u8  jt;                   /* Jump true */
    __u8  jf;                   /* Jump false */
    __u32 k;                    /* Generic multiuse field */
};

When executing the instructions, the BPF program operates on the system call information made available (i.e., use the BPF_ABS addressing mode) as a (read-only) buffer of the following form:

struct seccomp_data {
    int   nr;                   /* System call number */
    __u32 arch;                 /* AUDIT_ARCH_* value
                                   (see <linux/audit.h>) */
    __u64 instruction_pointer;  /* CPU instruction pointer */
    __u64 args[6];              /* Up to 6 system call arguments */
};

Because numbering of system calls varies between architectures and some architectures (e.g., x86-64) allow user-space code to use the calling conventions of multiple architectures (and the convention being used may vary over the life of a process that uses execve(2) to execute binaries that employ the different conventions), it is usually necessary to verify the value of the arch field.

It is strongly recommended to use an allow-list approach whenever possible because such an approach is more robust and simple. A deny-list will have to be updated whenever a potentially dangerous system call is added (or a dangerous flag or option if those are deny-listed), and it is often possible to alter the representation of a value without altering its meaning, leading to a deny-list bypass. See also Caveats below.

The arch field is not unique for all calling conventions. The x86-64 ABI and the x32 ABI both use AUDIT_ARCH_X86_64 as arch, and they run on the same processors. Instead, the mask __X32_SYSCALL_BIT is used on the system call number to tell the two ABIs apart.

This means that in order to create a seccomp-based deny-list for system calls performed through the x86-64 ABI, it is necessary to not only check that arch equals AUDIT_ARCH_X86_64, but also to explicitly reject all system calls that contain __X32_SYSCALL_BIT in nr.

The instruction_pointer field provides the address of the machine-language instruction that performed the system call. This might be useful in conjunction with the use of /proc/[pid]/maps to perform checks based on which region (mapping) of the program made the system call. (Probably, it is wise to lock down the mmap(2) and mprotect(2) system calls to prevent the program from subverting such checks.)

When checking values from args against a deny-list, keep in mind that arguments are often silently truncated before being processed, but after the seccomp check. For example, this happens if the i386 ABI is used on an x86-64 kernel: although the kernel will normally not look beyond the 32 lowest bits of the arguments, the values of the full 64-bit registers will be present in the seccomp data. A less surprising example is that if the x86-64 ABI is used to perform a system call that takes an argument of type int, the more-significant half of the argument register is ignored by the system call, but visible in the seccomp data.

A seccomp filter returns a 32-bit value consisting of two parts: the most significant 16 bits (corresponding to the mask defined by the constant SECCOMP_RET_ACTION_FULL) contain one of the "action" values listed below; the least significant 16-bits (defined by the constant SECCOMP_RET_DATA) are "data" to be associated with this return value.

If multiple filters exist, they are all executed, in reverse order of their addition to the filter tree—that is, the most recently installed filter is executed first. (Note that all filters will be called even if one of the earlier filters returns SECCOMP_RET_KILL. This is done to simplify the kernel code and to provide a tiny speed-up in the execution of sets of filters by avoiding a check for this uncommon case.) The return value for the evaluation of a given system call is the first-seen action value of highest precedence (along with its accompanying data) returned by execution of all of the filters.

In decreasing order of precedence, the action values that may be returned by a seccomp filter are:

SECCOMP_RET_KILL_PROCESS (since Linux 4.14)

This value results in immediate termination of the process, with a core dump. The system call is not executed. By contrast with SECCOMP_RET_KILL_THREAD below, all threads in the thread group are terminated. (For a discussion of thread groups, see the description of the CLONE_THREAD flag in clone(2).)

The process terminates as though killed by a SIGSYS signal. Even if a signal handler has been registered for SIGSYS, the handler will be ignored in this case and the process always terminates. To a parent process that is waiting on this process (using waitpid(2) or similar), the returned wstatus will indicate that its child was terminated as though by a SIGSYS signal.

SECCOMP_RET_KILL_THREAD (or SECCOMP_RET_KILL)

This value results in immediate termination of the thread that made the system call. The system call is not executed. Other threads in the same thread group will continue to execute.

The thread terminates as though killed by a SIGSYS signal. See SECCOMP_RET_KILL_PROCESS above.

Before Linux 4.11, any process terminated in this way would not trigger a coredump (even though SIGSYS is documented in signal(7) as having a default action of termination with a core dump). Since Linux 4.11, a single-threaded process will dump core if terminated in this way.

With the addition of SECCOMP_RET_KILL_PROCESS in Linux 4.14, SECCOMP_RET_KILL_THREAD was added as a synonym for SECCOMP_RET_KILL, in order to more clearly distinguish the two actions.

SECCOMP_RET_TRAP

This value results in the kernel sending a thread-directed SIGSYS signal to the triggering thread. (The system call is not executed.) Various fields will be set in the siginfo_t structure (see sigaction(2)) associated with signal:

• si_signo will contain SIGSYS.
• si_call_addr will show the address of the system call instruction.
• si_syscall and si_arch will indicate which system call was attempted.
• si_code will contain SYS_SECCOMP.
• si_errno will contain the SECCOMP_RET_DATA portion of the filter return value.

The program counter will be as though the system call happened (i.e., the program counter will not point to the system call instruction). The return value register will contain an architecture-dependent value; if resuming execution, set it to something appropriate for the system call. (The architecture dependency is because replacing it with ENOSYS could overwrite some useful information.)

SECCOMP_RET_ERRNO

This value results in the SECCOMP_RET_DATA portion of the filter's return value being passed to user space as the errno value without executing the system call.

SECCOMP_RET_TRACE

When returned, this value will cause the kernel to attempt to notify a ptrace(2)-based tracer prior to executing the system call. If there is no tracer present, the system call is not executed and returns a failure status with errno set to ENOSYS.

A tracer will be notified if it requests PTRACE_O_TRACESECCOMP using ptrace(PTRACE_SETOPTIONS). The tracer will be notified of a PTRACE_EVENT_SECCOMP and the SECCOMP_RET_DATA portion of the filter's return value will be available to the tracer via PTRACE_GETEVENTMSG.

The tracer can skip the system call by changing the system call number to -1. Alternatively, the tracer can change the system call requested by changing the system call to a valid system call number. If the tracer asks to skip the system call, then the system call will appear to return the value that the tracer puts in the return value register.

Before kernel 4.8, the seccomp check will not be run again after the tracer is notified. (This means that, on older kernels, seccomp-based sandboxes must not allow use of ptrace(2)—even of other sandboxed processes—without extreme care; ptracers can use this mechanism to escape from the seccomp sandbox.)

SECCOMP_RET_LOG (since Linux 4.14)

This value results in the system call being executed after the filter return action is logged. An administrator may override the logging of this action via the /proc/sys/kernel/seccomp/actions_logged file.

SECCOMP_RET_ALLOW

This value results in the system call being executed.

If an action value other than one of the above is specified, then the filter action is treated as either SECCOMP_RET_KILL_PROCESS (since Linux 4.14) or SECCOMP_RET_KILL_THREAD (in Linux 4.13 and earlier).

The files in the directory /proc/sys/kernel/seccomp provide additional seccomp information and configuration:

actions_avail (since Linux 4.14)

A read-only ordered list of seccomp filter return actions in string form. The ordering, from left-to-right, is in decreasing order of precedence. The list represents the set of seccomp filter return actions supported by the kernel.

actions_logged (since Linux 4.14)

A read-write ordered list of seccomp filter return actions that are allowed to be logged. Writes to the file do not need to be in ordered form but reads from the file will be ordered in the same way as the actions_avail file.

It is important to note that the value of actions_logged does not prevent certain filter return actions from being logged when the audit subsystem is configured to audit a task. If the action is not found in the actions_logged file, the final decision on whether to audit the action for that task is ultimately left up to the audit subsystem to decide for all filter return actions other than SECCOMP_RET_ALLOW.

The "allow" string is not accepted in the actions_logged file as it is not possible to log SECCOMP_RET_ALLOW actions. Attempting to write "allow" to the file will fail with the error EINVAL.

Since Linux 4.14, the kernel provides the facility to log the actions returned by seccomp filters in the audit log. The kernel makes the decision to log an action based on the action type,  whether or not the action is present in the actions_logged file, and whether kernel auditing is enabled (e.g., via the kernel boot option audit=1). The rules are as follows:

• If the action is SECCOMP_RET_ALLOW, the action is not logged.
• Otherwise, if the action is either SECCOMP_RET_KILL_PROCESS or SECCOMP_RET_KILL_THREAD, and that action appears in the actions_logged file, the action is logged.
• Otherwise, if the filter has requested logging (the SECCOMP_FILTER_FLAG_LOG flag) and the action appears in the actions_logged file, the action is logged.
• Otherwise, if kernel auditing is enabled and the process is being audited (autrace(8)), the action is logged.
• Otherwise, the action is not logged.

Architecture support for seccomp BPF filtering is available on the following architectures:

• x86-64, i386, x32 (since Linux 3.5)
• ARM (since Linux 3.8)
• s390 (since Linux 3.8)
• MIPS (since Linux 3.16)
• ARM-64 (since Linux 3.19)
• PowerPC (since Linux 4.3)
• Tile (since Linux 4.3)
• PA-RISC (since Linux 4.6)

There are various subtleties to consider when applying seccomp filters to a program, including the following:

• Some traditional system calls have user-space implementations in the vdso(7) on many architectures. Notable examples include clock_gettime(2), gettimeofday(2), and time(2). On such architectures, seccomp filtering for these system calls will have no effect. (However, there are cases where the vdso(7) implementations may fall back to invoking the true system call, in which case seccomp filters would see the system call.)

• Seccomp filtering is based on system call numbers. However, applications typically do not directly invoke system calls, but instead call wrapper functions in the C library which in turn invoke the system calls. Consequently, one must be aware of the following:

  • The glibc wrappers for some traditional system calls may actually employ system calls with different names in the kernel. For example, the exit(2) wrapper function actually employs the exit_group(2) system call, and the fork(2) wrapper function actually calls clone(2).
  • The behavior of wrapper functions may vary across architectures, according to the range of system calls provided on those architectures. In other words, the same wrapper function may invoke different system calls on different architectures.
  • Finally, the behavior of wrapper functions can change across glibc versions. For example, in older versions, the glibc wrapper function for open(2) invoked the system call of the same name, but starting in glibc 2.26, the implementation switched to calling openat(2) on all architectures.

The consequence of the above points is that it may be necessary to filter for a system call other than might be expected. Various manual pages in Section 2 provide helpful details about the differences between wrapper functions and the underlying system calls in subsections entitled C library/kernel differences.

Furthermore, note that the application of seccomp filters even risks causing bugs in an application, when the filters cause unexpected failures for legitimate operations that the application might need to perform. Such bugs may not easily be discovered when testing the seccomp filters if the bugs occur in rarely used application code paths.

Note the following BPF details specific to seccomp filters:

• The BPF_H and BPF_B size modifiers are not supported: all operations must load and store (4-byte) words (BPF_W).
• To access the contents of the seccomp_data buffer, use the BPF_ABS addressing mode modifier.
• The BPF_LEN addressing mode modifier yields an immediate mode operand whose value is the size of the seccomp_data buffer.

#include <errno.h>
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <linux/audit.h>
#include <linux/filter.h>
#include <linux/seccomp.h>
#include <sys/prctl.h>

#define X32_SYSCALL_BIT 0x40000000

static int
install_filter(int syscall_nr, int t_arch, int f_errno)
{
    unsigned int upper_nr_limit = 0xffffffff;

    /* Assume that AUDIT_ARCH_X86_64 means the normal x86-64 ABI
       (in the x32 ABI, all system calls have bit 30 set in the
       'nr' field, meaning the numbers are >= X32_SYSCALL_BIT) */
    if (t_arch == AUDIT_ARCH_X86_64)
        upper_nr_limit = X32_SYSCALL_BIT - 1;

    struct sock_filter filter[] = {
        /* [0] Load architecture from 'seccomp_data' buffer into
               accumulator */
        BPF_STMT(BPF_LD | BPF_W | BPF_ABS,
                 (offsetof(struct seccomp_data, arch))),

        /* [1] Jump forward 5 instructions if architecture does not
               match 't_arch' */
        BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, t_arch, 0, 5),

        /* [2] Load system call number from 'seccomp_data' buffer into
               accumulator */
        BPF_STMT(BPF_LD | BPF_W | BPF_ABS,
                 (offsetof(struct seccomp_data, nr))),

        /* [3] Check ABI - only needed for x86-64 in deny-list use
               cases.  Use BPF_JGT instead of checking against the bit
               mask to avoid having to reload the syscall number. */
        BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, upper_nr_limit, 3, 0),

        /* [4] Jump forward 1 instruction if system call number
               does not match 'syscall_nr' */
        BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, syscall_nr, 0, 1),

        /* [5] Matching architecture and system call: don't execute
	       the system call, and return 'f_errno' in 'errno' */
        BPF_STMT(BPF_RET | BPF_K,
                 SECCOMP_RET_ERRNO | (f_errno & SECCOMP_RET_DATA)),

        /* [6] Destination of system call number mismatch: allow other
               system calls */
        BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW),

        /* [7] Destination of architecture mismatch: kill task */
        BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL),
    };

    struct sock_fprog prog = {
        .len = (unsigned short) (sizeof(filter) / sizeof(filter[0])),
        .filter = filter,
    };

    if (seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog)) {
        perror("seccomp");
        return 1;
    }

    return 0;
}

int
main(int argc, char **argv)
{
    if (argc < 5) {
        fprintf(stderr, "Usage: "
                "%s <syscall_nr> <arch> <errno> <prog> [<args>]\n"
                "Hint for <arch>: AUDIT_ARCH_I386: 0x%X\n"
                "                 AUDIT_ARCH_X86_64: 0x%X\n"
                "\n", argv[0], AUDIT_ARCH_I386, AUDIT_ARCH_X86_64);
        exit(EXIT_FAILURE);
    }

    if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
        perror("prctl");
        exit(EXIT_FAILURE);
    }

    if (install_filter(strtol(argv[1], NULL, 0),
                       strtol(argv[2], NULL, 0),
                       strtol(argv[3], NULL, 0)))
        exit(EXIT_FAILURE);

    execv(argv[4], &argv[4]);
    perror("execv");
    exit(EXIT_FAILURE);
}