Use some macros to increase readabiltiy of BPF rules

This commit is contained in:
Albert S. 2021-12-27 12:30:27 +01:00
parent 77adf09d34
commit ca0f82790c

120
exile.h
View File

@ -68,13 +68,6 @@
#error Seccomp support has not been tested for exile.h for this platform yet #error Seccomp support has not been tested for exile.h for this platform yet
#endif #endif
#define SYSCALL(nr, jt) \
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, (nr), 0, 1), jt
#define EXILE_BPF_NOP \
BPF_STMT(BPF_JMP+BPF_JA,0)
#define EXILE_UNSHARE_NETWORK 1<<1 #define EXILE_UNSHARE_NETWORK 1<<1
#define EXILE_UNSHARE_USER 1<<2 #define EXILE_UNSHARE_USER 1<<2
#define EXILE_UNSHARE_MOUNT 1<<3 #define EXILE_UNSHARE_MOUNT 1<<3
@ -263,6 +256,34 @@ struct exile_path_policy
#define EXILE_SYSCALL_DENY_KILL_PROCESS 2 #define EXILE_SYSCALL_DENY_KILL_PROCESS 2
#define EXILE_SYSCALL_DENY_RET_ERROR 3 #define EXILE_SYSCALL_DENY_RET_ERROR 3
#define EXILE_BPF_NOP \
BPF_STMT(BPF_JMP+BPF_JA,0)
#define EXILE_BPF_LOAD_SECCOMP_ARG(nr) \
BPF_STMT(BPF_LD+BPF_W+BPF_ABS, (offsetof(struct seccomp_data, args[nr])))
#define EXILE_BPF_CMP_EQ(val,t,f) \
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, val, t, f)
#define EXILE_BPF_CMP_SET(val,t,f) \
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, val, t, f)
/* Immediatly go to the syscall action, do not process any other arg filters */
#define EXILE_BPF_MATCH(argval) \
EXILE_BPF_CMP_EQ(argval, EXILE_SYSCALL_EXIT_BPF_RETURN, 0)
#define EXILE_BPF_MATCH_SET(argval) \
EXILE_BPF_CMP_SET(argval, EXILE_SYSCALL_EXIT_BPF_RETURN, 0)
/* Immediatly go beyond the syscall action, do not process any other arg filters. What to do with this syscall
is thus up to the default policy */
#define EXILE_BPF_NO_MATCH(argval) \
EXILE_BPF_CMP_EQ(argval, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0)
#define EXILE_BPF_NO_MATCH_SET(argval) \
EXILE_BPF_CMP_SET(argval, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0)
/* Pledge definitions */ /* Pledge definitions */
#define EXILE_SYSCALL_PLEDGE_CHOWN ((uint64_t)1<<1) #define EXILE_SYSCALL_PLEDGE_CHOWN ((uint64_t)1<<1)
#define EXILE_SYSCALL_PLEDGE_CLONE ((uint64_t)1<<2) #define EXILE_SYSCALL_PLEDGE_CLONE ((uint64_t)1<<2)
@ -645,64 +666,71 @@ int exile_append_syscall_default_policy(struct exile_policy *exile_policy, unsig
static int get_pledge_argfilter(long syscall, uint64_t pledge_promises, struct sock_filter *filter) static int get_pledge_argfilter(long syscall, uint64_t pledge_promises, struct sock_filter *filter)
{ {
/* How to read this:
* Keep in mind our default action is do deny, unless it's a syscall from an pledge promise. Then it will be
* accepted if the argument values are good (if we care about them at all).
* EXILE_BPF_MATCH() means the argument value is good, and the syscall can be accepted
* EXILE_BPF_NO_MATCH() means the syscall won't be allowed because the value is illegal
*/
struct sock_filter mmap_no_exec[] = { struct sock_filter mmap_no_exec[] = {
BPF_STMT(BPF_LD+BPF_W+BPF_ABS, (offsetof(struct seccomp_data, args[2]))), EXILE_BPF_LOAD_SECCOMP_ARG(2),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, PROT_EXEC, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0)}; EXILE_BPF_NO_MATCH(PROT_EXEC)
};
struct sock_filter ioctl_default[] = { struct sock_filter ioctl_default[] = {
BPF_STMT(BPF_LD+BPF_W+BPF_ABS, (offsetof(struct seccomp_data, args[1]))), EXILE_BPF_LOAD_SECCOMP_ARG(1),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, FIONREAD, EXILE_SYSCALL_EXIT_BPF_RETURN, 0), EXILE_BPF_MATCH(FIONREAD),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, FIONBIO, EXILE_SYSCALL_EXIT_BPF_RETURN, 0), EXILE_BPF_MATCH(FIONBIO),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, FIOCLEX, EXILE_SYSCALL_EXIT_BPF_RETURN, 0), EXILE_BPF_MATCH(FIOCLEX),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, FIONCLEX, EXILE_SYSCALL_EXIT_BPF_RETURN, EXILE_SYSCALL_EXIT_BPF_NO_MATCH) EXILE_BPF_CMP_EQ(FIONCLEX, EXILE_SYSCALL_EXIT_BPF_RETURN, EXILE_SYSCALL_EXIT_BPF_NO_MATCH)
}; };
/* open() and friends with read-only flags */ /* open() and friends with read-only flags */
struct sock_filter open_rdonly[] = { struct sock_filter open_rdonly[] = {
BPF_STMT(BPF_LD+BPF_W+BPF_ABS, (offsetof(struct seccomp_data, args[1]))), EXILE_BPF_LOAD_SECCOMP_ARG(1),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, O_CREAT, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0), EXILE_BPF_NO_MATCH_SET(O_CREAT),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, O_TMPFILE, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0), EXILE_BPF_NO_MATCH_SET(O_TMPFILE),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, O_WRONLY, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0), EXILE_BPF_NO_MATCH_SET(O_WRONLY),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, O_RDWR, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0), EXILE_BPF_NO_MATCH_SET(O_RDWR),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, O_APPEND, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0), EXILE_BPF_NO_MATCH_SET(O_APPEND),
}; };
struct sock_filter socket_filter[4] = { struct sock_filter socket_filter[4] = {
BPF_STMT(BPF_LD+BPF_W+BPF_ABS, (offsetof(struct seccomp_data, args[0]))), EXILE_BPF_LOAD_SECCOMP_ARG(0),
EXILE_BPF_NOP, EXILE_BPF_NOP,
EXILE_BPF_NOP, EXILE_BPF_NOP,
EXILE_BPF_NOP EXILE_BPF_NOP
}; };
struct sock_filter setsockopt_filter[] = { struct sock_filter setsockopt_filter[] = {
BPF_STMT(BPF_LD+BPF_W+BPF_ABS, (offsetof(struct seccomp_data, args[2]))), EXILE_BPF_LOAD_SECCOMP_ARG(2),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, SO_DEBUG, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0), EXILE_BPF_NO_MATCH(SO_DEBUG),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, SO_SNDBUFFORCE, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0) EXILE_BPF_NO_MATCH(SO_SNDBUFFORCE)
}; };
struct sock_filter clone_filter[] = { struct sock_filter clone_filter[] = {
/* It's the first argument for x86_64 */ /* It's the first argument for x86_64 */
BPF_STMT(BPF_LD+BPF_W+BPF_ABS, (offsetof(struct seccomp_data, args[0]))), EXILE_BPF_LOAD_SECCOMP_ARG(0),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, CLONE_VM, 0, EXILE_SYSCALL_EXIT_BPF_NO_MATCH), EXILE_BPF_NO_MATCH_SET(CLONE_VM),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, CLONE_THREAD, 0, EXILE_SYSCALL_EXIT_BPF_NO_MATCH), EXILE_BPF_NO_MATCH_SET(CLONE_THREAD),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, CLONE_NEWCGROUP, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0), EXILE_BPF_NO_MATCH_SET(CLONE_NEWCGROUP),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, CLONE_NEWIPC, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0), EXILE_BPF_NO_MATCH_SET(CLONE_NEWIPC),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, CLONE_NEWNET, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0), EXILE_BPF_NO_MATCH_SET(CLONE_NEWNET),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, CLONE_NEWNS, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0), EXILE_BPF_NO_MATCH_SET(CLONE_NEWNS),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, CLONE_NEWPID, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0), EXILE_BPF_NO_MATCH_SET(CLONE_NEWPID),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, CLONE_NEWUSER, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0), EXILE_BPF_NO_MATCH_SET(CLONE_NEWUSER),
BPF_JUMP(BPF_JMP+BPF_JSET+BPF_K, CLONE_NEWUTS, EXILE_SYSCALL_EXIT_BPF_NO_MATCH, 0) EXILE_BPF_NO_MATCH_SET(CLONE_NEWUTS),
}; };
struct sock_filter prctl_default[] ={ struct sock_filter prctl_default[] ={
BPF_STMT(BPF_LD+BPF_W+BPF_ABS, (offsetof(struct seccomp_data, args[0]))), EXILE_BPF_LOAD_SECCOMP_ARG(0),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, PR_SET_NO_NEW_PRIVS, EXILE_SYSCALL_EXIT_BPF_RETURN, 0), EXILE_BPF_MATCH(PR_SET_NO_NEW_PRIVS),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, PR_GET_NO_NEW_PRIVS, EXILE_SYSCALL_EXIT_BPF_RETURN, 0), EXILE_BPF_MATCH(PR_GET_NO_NEW_PRIVS),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, PR_GET_NAME, EXILE_SYSCALL_EXIT_BPF_RETURN, 0), EXILE_BPF_MATCH(PR_GET_NAME),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, PR_SET_NAME, EXILE_SYSCALL_EXIT_BPF_RETURN, 0), EXILE_BPF_MATCH(PR_SET_NAME),
EXILE_BPF_NOP, EXILE_BPF_CMP_EQ(PR_CAPBSET_READ, EXILE_SYSCALL_EXIT_BPF_RETURN, EXILE_SYSCALL_EXIT_BPF_NO_MATCH),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, PR_CAPBSET_READ, EXILE_SYSCALL_EXIT_BPF_RETURN, EXILE_SYSCALL_EXIT_BPF_NO_MATCH),
}; };
int result = 0; int result = 0;
@ -769,14 +797,14 @@ static int get_pledge_argfilter(long syscall, uint64_t pledge_promises, struct s
case EXILE_SYS(socket): case EXILE_SYS(socket):
if(pledge_promises & EXILE_SYSCALL_PLEDGE_UNIX) if(pledge_promises & EXILE_SYSCALL_PLEDGE_UNIX)
{ {
socket_filter[current_filter_index] = (struct sock_filter) BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, AF_UNIX, EXILE_SYSCALL_EXIT_BPF_RETURN, 0); socket_filter[current_filter_index] = (struct sock_filter) EXILE_BPF_MATCH(AF_UNIX);
++current_filter_index; ++current_filter_index;
} }
if(pledge_promises & EXILE_SYSCALL_PLEDGE_INET) if(pledge_promises & EXILE_SYSCALL_PLEDGE_INET)
{ {
socket_filter[current_filter_index] = (struct sock_filter) BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, AF_INET, EXILE_SYSCALL_EXIT_BPF_RETURN, 0); socket_filter[current_filter_index] = (struct sock_filter) EXILE_BPF_MATCH(AF_INET);
++current_filter_index; ++current_filter_index;
socket_filter[current_filter_index] = (struct sock_filter) BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, AF_INET6, EXILE_SYSCALL_EXIT_BPF_RETURN, 0); socket_filter[current_filter_index] = (struct sock_filter) EXILE_BPF_MATCH(AF_INET6);
++current_filter_index; ++current_filter_index;
} }
socket_filter[current_filter_index-1].jf = EXILE_SYSCALL_EXIT_BPF_NO_MATCH; socket_filter[current_filter_index-1].jf = EXILE_SYSCALL_EXIT_BPF_NO_MATCH;
@ -804,7 +832,7 @@ static int get_pledge_argfilter(long syscall, uint64_t pledge_promises, struct s
} }
if(pledge_promises & EXILE_SYSCALL_PLEDGE_SECCOMP_INSTALL) if(pledge_promises & EXILE_SYSCALL_PLEDGE_SECCOMP_INSTALL)
{ {
prctl_default[3] = (struct sock_filter) BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, PR_SET_SECCOMP, EXILE_SYSCALL_EXIT_BPF_RETURN, 0); prctl_default[3] = (struct sock_filter) EXILE_BPF_MATCH(PR_SET_SECCOMP);
} }
result = sizeof(prctl_default)/sizeof(prctl_default[0]); result = sizeof(prctl_default)/sizeof(prctl_default[0]);
memcpy(filter, prctl_default, sizeof(prctl_default)); memcpy(filter, prctl_default, sizeof(prctl_default));
@ -1278,7 +1306,7 @@ static void append_syscall_to_bpf(struct exile_syscall_policy *syscallpolicy, st
/* TODO: Return error */ /* TODO: Return error */
return; return;
} }
struct sock_filter syscall_check = BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, (unsigned int) syscall, 0, next_syscall_pc); struct sock_filter syscall_check = EXILE_BPF_CMP_EQ((unsigned int) syscall, 0, next_syscall_pc);
filter[(*start_index)++] = syscall_check; filter[(*start_index)++] = syscall_check;
--next_syscall_pc; --next_syscall_pc;