seccomp_rule_add - Man Page

#include <seccomp.h>

typedef void * scmp_filter_ctx;

int SCMP_SYS(syscall_name);

struct scmp_arg_cmp SCMP_CMP(unsigned int arg,
                             enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A0(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A1(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A2(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A3(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A4(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A5(enum scmp_compare op, ...);

struct scmp_arg_cmp SCMP_CMP64(unsigned int arg,
                             enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A0_64(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A1_64(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A2_64(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A3_64(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A4_64(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A5_64(enum scmp_compare op, ...);

struct scmp_arg_cmp SCMP_CMP32(unsigned int arg,
                             enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A0_32(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A1_32(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A2_32(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A3_32(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A4_32(enum scmp_compare op, ...);
struct scmp_arg_cmp SCMP_A5_32(enum scmp_compare op, ...);

int seccomp_rule_add(scmp_filter_ctx ctx, uint32_t action,
                     int syscall, unsigned int arg_cnt, ...);
int seccomp_rule_add_exact(scmp_filter_ctx ctx, uint32_t action,
                           int syscall, unsigned int arg_cnt, ...);

int seccomp_rule_add_array(scmp_filter_ctx ctx,
                           uint32_t action, int syscall,
                           unsigned int arg_cnt,
                           const struct scmp_arg_cmp *arg_array);
int seccomp_rule_add_exact_array(scmp_filter_ctx ctx,
                                 uint32_t action, int syscall,
                                 unsigned int arg_cnt,
                                 const struct scmp_arg_cmp *arg_array);

Link with -lseccomp.

The seccomp_rule_add(), seccomp_rule_add_array(), seccomp_rule_add_exact(), and seccomp_rule_add_exact_array() functions all add a new filter rule to the current seccomp filter.  The seccomp_rule_add() and seccomp_rule_add_array() functions will make a "best effort" to add the rule as specified, but may alter the rule slightly due to architecture specifics (e.g. internal rewriting of multiplexed syscalls, like socket and ipc functions on x86).  The seccomp_rule_add_exact() and seccomp_rule_add_exact_array() functions will attempt to add the rule exactly as specified so it may behave differently on different architectures.  While it does not guarantee a exact filter ruleset, seccomp_rule_add() and seccomp_rule_add_array() do guarantee the same behavior regardless of the architecture.

The newly added filter rule does not take effect until the entire filter is loaded into the kernel using seccomp_load(3). When adding rules to a filter, it is important to consider the impact of previously loaded filters; see the seccomp_load(3) documentation for more information.

All of the filter rules supplied by the calling application are combined into a union, with additional logic to eliminate redundant syscall filters.  For example, if a rule is added which allows a given syscall with a specific set of argument values and later a rule is added which allows the same syscall regardless the argument values then the first, more specific rule, is effectively dropped from the filter by the second more generic rule.

The SCMP_CMP(), SCMP_CMP64(), SCMP_A{0-5}(), and SCMP_A{0-5}_64() macros generate a scmp_arg_cmp structure for use with the above functions. The SCMP_CMP() and SCMP_CMP64() macros allows the caller to specify an arbitrary argument along with the comparison operator, 64-bit mask, and 64-bit datum values where the SCMP_A{0-5}() and SCMP_A{0-5}_64() macros are specific to a certain argument.

The SCMP_CMP32() and SCMP_A{0-5}_32() macros are similar to the variants above, but they take 32-bit mask and 32-bit datum values.

It is recommended that whenever possible developers avoid using the SCMP_CMP() and SCMP_A{0-5}() macros and use the variants which are explicitly 32 or 64-bit.  This should help eliminate problems caused by an unwanted sign extension of negative datum values.

If syscall argument comparisons are included in the filter rule, all of the comparisons must be true for the rule to match.

When adding syscall argument comparisons to the filter it is important to remember that while it is possible to have multiple comparisons in a single rule, you can only compare each argument once in a single rule.  In other words, you can not have multiple comparisons of the 3rd syscall argument in a single rule.

In a filter containing multiple architectures, it is an error to add a filter rule for a syscall that does not exist in all of the filter's architectures.

While it is possible to specify the syscall value directly using the standard __NR_syscall values, in order to ensure proper operation across multiple architectures it is highly recommended to use the SCMP_SYS() macro instead.  See the Examples section below.  It is also important to remember that regardless of the architectures present in the filter, the syscall numbers used in filter rules are interpreted in the context of the native architecture.

Starting with Linux v4.8, there may be a need to create a rule with a syscall value of -1 to allow tracing programs to skip a syscall invocation; in order to create a rule with a -1 syscall value it is necessary to first set the SCMP_FLTATR_API_TSKIP attribute.  See seccomp_attr_set(3) for more information.

The filter context ctx is the value returned by the call to seccomp_init(3).

Valid action values are as follows:

SCMP_ACT_KILL

The thread will be killed by the kernel when it calls a syscall that matches the filter rule.

SCMP_ACT_KILL_PROCESS

The process will be killed by the kernel when it calls a syscall that matches the filter rule.

SCMP_ACT_TRAP

The thread will throw a SIGSYS signal when it calls a syscall that matches the filter rule.

SCMP_ACT_ERRNO(uint16_t errno)

The thread will receive a return value of errno when it calls a syscall that matches the filter rule.

SCMP_ACT_TRACE(uint16_t msg_num)

If the thread is being traced and the tracing process specified the PTRACE_O_TRACESECCOMP option in the call to ptrace(2), the tracing process will be notified, via PTRACE_EVENT_SECCOMP , and the value provided in msg_num can be retrieved using the PTRACE_GETEVENTMSG option.

SCMP_ACT_LOG

The seccomp filter will have no effect on the thread calling the syscall if it matches the filter rule but the syscall will be logged.

SCMP_ACT_ALLOW

The seccomp filter will have no effect on the thread calling the syscall if it matches the filter rule.

SCMP_ACT_NOTIFY

A monitoring process will be notified when a process running the seccomp filter calls a syscall that matches the filter rule.  The process that invokes the syscall waits in the kernel until the monitoring process has responded via seccomp_notify_respond (3) .

When a filter utilizing SCMP_ACT_NOTIFY is loaded into the kernel, the kernel generates a notification fd that must be used to communicate between the monitoring process and the process(es) being filtered.  See seccomp_notif_fd (3) for more information.

Valid comparison op values are as follows:

SCMP_CMP_NE

Matches when the argument value is not equal to the datum value, example:

SCMP_CMP( arg , SCMP_CMP_NE , datum )

SCMP_CMP_LT

Matches when the argument value is less than the datum value, example:

SCMP_CMP( arg , SCMP_CMP_LT , datum )

SCMP_CMP_LE

Matches when the argument value is less than or equal to the datum value, example:

SCMP_CMP( arg , SCMP_CMP_LE , datum )

SCMP_CMP_EQ

Matches when the argument value is equal to the datum value, example:

SCMP_CMP( arg , SCMP_CMP_EQ , datum )

SCMP_CMP_GE

Matches when the argument value is greater than or equal to the datum value, example:

SCMP_CMP( arg , SCMP_CMP_GE , datum )

SCMP_CMP_GT

Matches when the argument value is greater than the datum value, example:

SCMP_CMP( arg , SCMP_CMP_GT , datum )

SCMP_CMP_MASKED_EQ

Matches when the masked argument value is equal to the masked datum value, example:

SCMP_CMP( arg , SCMP_CMP_MASKED_EQ , mask , datum )

The SCMP_SYS() macro returns a value suitable for use as the syscall value in the seccomp_rule_add*() functions.  In a similar manner, the SCMP_CMP() and SCMP_A*() macros return values suitable for use as argument comparisons in the seccomp_rule_add() and seccomp_rule_add_exact() functions.

The seccomp_rule_add(), seccomp_rule_add_array(), seccomp_rule_add_exact(), and seccomp_rule_add_exact_array() functions return zero on success or one of the following error codes on failure:

-EDOM

Architecture specific failure.

-EEXIST

The rule already exists.

-EACCCES

The rule conflicts with the filter (for example, the rule action equals the default action of the filter).

-EFAULT

Internal libseccomp failure.

-EINVAL

Invalid input, either the context or architecture token is invalid.

-ENOMEM

The library was unable to allocate enough memory.

-EOPNOTSUPP

The library doesn't support the particular operation.

#include <fcntl.h>
#include <seccomp.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <stddef.h>

#define BUF_SIZE	256

int main(int argc, char *argv[])
{
	int rc = -1;
	scmp_filter_ctx ctx;
	struct scmp_arg_cmp arg_cmp[] = { SCMP_A0(SCMP_CMP_EQ, 2) };
	int fd;
	unsigned char buf[BUF_SIZE];

	ctx = seccomp_init(SCMP_ACT_KILL);
	if (ctx == NULL)
		goto out;

	/* ... */

	fd = open("file.txt", 0);

	/* ... */

	rc = seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(close), 0);
	if (rc < 0)
		goto out;

	rc = seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(exit_group), 0);
	if (rc < 0)
		goto out;

	rc = seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(exit), 0);
	if (rc < 0)
		goto out;

	rc = seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(read), 3,
			      SCMP_A0(SCMP_CMP_EQ, fd),
			      SCMP_A1(SCMP_CMP_EQ, (scmp_datum_t)buf),
			      SCMP_A2(SCMP_CMP_LE, BUF_SIZE));
	if (rc < 0)
		goto out;

	rc = seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(write), 1,
			      SCMP_CMP(0, SCMP_CMP_EQ, fd));
	if (rc < 0)
		goto out;

	rc = seccomp_rule_add_array(ctx, SCMP_ACT_ALLOW, SCMP_SYS(write), 1,
			            arg_cmp);
	if (rc < 0)
		goto out;

	rc = seccomp_load(ctx);
	if (rc < 0)
		goto out;

	/* ... */

out:
	seccomp_release(ctx);
	return -rc;
}

While the seccomp filter can be generated independent of the kernel, kernel support is required to load and enforce the seccomp filter generated by libseccomp.

The libseccomp project site, with more information and the source code repository, can be found at https://github.com/seccomp/libseccomp.  This tool, as well as the libseccomp library, is currently under development, please report any bugs at the project site or directly to the author.

The runtime behavior of seccomp filters is dependent upon the kernel version, the processor architecture, and other libraries including libc. This could affect the return code of a seccomp filter.

• PowerPC glibc will not return a negative number when the getpid() syscall is invoked.  If a seccomp filter has been created where getpid() will return a negative number from the kernel, then PowerPC glibc will return the absolute value of the errno.  In this case, it is very difficult for an application to distinguish between the errno and a valid pid.

Paul Moore <paul@paul-moore.com>

seccomp_syscall_resolve_name_rewrite(3), seccomp_syscall_priority(3), seccomp_load(3), seccomp_attr_set(3)

seccomp(2), seccomp_load(3), seccomp_syscall_priority(3), seccomp_syscall_resolve_name(3).

The man pages seccomp_rule_add_array(3), seccomp_rule_add_exact(3) and seccomp_rule_add_exact_array(3) are aliases of seccomp_rule_add(3).