]> git.ipfire.org Git - thirdparty/systemd.git/commitdiff
Add firewall eBPF compiler
authorDaniel Mack <daniel@zonque.org>
Fri, 11 Nov 2016 18:41:16 +0000 (19:41 +0100)
committerLennart Poettering <lennart@poettering.net>
Fri, 22 Sep 2017 13:24:55 +0000 (15:24 +0200)
src/core/bpf-firewall.c [new file with mode: 0644]
src/core/bpf-firewall.h [new file with mode: 0644]
src/core/meson.build

diff --git a/src/core/bpf-firewall.c b/src/core/bpf-firewall.c
new file mode 100644 (file)
index 0000000..732c36f
--- /dev/null
@@ -0,0 +1,672 @@
+/***
+  This file is part of systemd.
+
+  Copyright 2016 Daniel Mack
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU Lesser General Public License as published by
+  the Free Software Foundation; either version 2.1 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  Lesser General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <arpa/inet.h>
+#include <assert.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <linux/libbpf.h>
+#include <net/ethernet.h>
+#include <net/if.h>
+#include <netinet/ip.h>
+#include <netinet/ip6.h>
+#include <stddef.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#include "alloc-util.h"
+#include "bpf-firewall.h"
+#include "bpf-program.h"
+#include "fd-util.h"
+#include "ip-address-access.h"
+#include "unit.h"
+
+enum {
+        MAP_KEY_PACKETS,
+        MAP_KEY_BYTES,
+};
+
+enum {
+        ACCESS_ALLOWED = 1,
+        ACCESS_DENIED  = 2,
+};
+
+/* Compile instructions for one list of addresses, one direction and one specific verdict on matches. */
+
+static int add_lookup_instructions(
+                BPFProgram *p,
+                int map_fd,
+                int protocol,
+                bool is_ingress,
+                int verdict) {
+
+        int r, addr_offset, addr_size;
+
+        assert(p);
+        assert(map_fd >= 0);
+
+        switch (protocol) {
+
+        case ETH_P_IP:
+                addr_size = sizeof(uint32_t);
+                addr_offset = is_ingress ?
+                        offsetof(struct iphdr, saddr) :
+                        offsetof(struct iphdr, daddr);
+                break;
+
+        case ETH_P_IPV6:
+                addr_size = 4 * sizeof(uint32_t);
+                addr_offset = is_ingress ?
+                        offsetof(struct ip6_hdr, ip6_src.s6_addr) :
+                        offsetof(struct ip6_hdr, ip6_dst.s6_addr);
+                break;
+
+        default:
+                return -EAFNOSUPPORT;
+        }
+
+        do {
+                /* Compare IPv4 with one word instruction (32bit) */
+                struct bpf_insn insn[] = {
+                        /* If skb->protocol != ETH_P_IP, skip this whole block. The offset will be set later. */
+                        BPF_JMP_IMM(BPF_JNE, BPF_REG_7, htobe16(protocol), 0),
+
+                        /*
+                         * Call into BPF_FUNC_skb_load_bytes to load the dst/src IP address
+                         *
+                         * R1: Pointer to the skb
+                         * R2: Data offset
+                         * R3: Destination buffer on the stack (r10 - 4)
+                         * R4: Number of bytes to read (4)
+                         */
+
+                        BPF_MOV64_REG(BPF_REG_1, BPF_REG_6),
+                        BPF_MOV32_IMM(BPF_REG_2, addr_offset),
+
+                        BPF_MOV64_REG(BPF_REG_3, BPF_REG_10),
+                        BPF_ALU64_IMM(BPF_ADD, BPF_REG_3, -addr_size),
+
+                        BPF_MOV32_IMM(BPF_REG_4, addr_size),
+                        BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_skb_load_bytes),
+
+                        /*
+                         * Call into BPF_FUNC_map_lookup_elem to see if the address matches any entry in the
+                         * LPM trie map. For this to work, the prefixlen field of 'struct bpf_lpm_trie_key'
+                         * has to be set to the maximum possible value.
+                         *
+                         * On success, the looked up value is stored in R0. For this application, the actual
+                         * value doesn't matter, however; we just set the bit in @verdict in R8 if we found any
+                         * matching value.
+                         */
+
+                        BPF_LD_MAP_FD(BPF_REG_1, map_fd),
+                        BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
+                        BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -addr_size - sizeof(uint32_t)),
+                        BPF_ST_MEM(BPF_W, BPF_REG_2, 0, addr_size * 8),
+
+                        BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
+                        BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 1),
+                        BPF_ALU32_IMM(BPF_OR, BPF_REG_8, verdict),
+                };
+
+                /* Jump label fixup */
+                insn[0].off = ELEMENTSOF(insn) - 1;
+
+                r = bpf_program_add_instructions(p, insn, ELEMENTSOF(insn));
+                if (r < 0)
+                        return r;
+
+        } while (false);
+
+        return 0;
+}
+
+static int bpf_firewall_compile_bpf(
+                Unit *u,
+                bool is_ingress,
+                BPFProgram **ret) {
+
+        struct bpf_insn pre_insn[] = {
+                /*
+                 * When the eBPF program is entered, R1 contains the address of the skb.
+                 * However, R1-R5 are scratch registers that are not preserved when calling
+                 * into kernel functions, so we need to save anything that's supposed to
+                 * stay around to R6-R9. Save the skb to R6.
+                 */
+                BPF_MOV64_REG(BPF_REG_6, BPF_REG_1),
+
+                /*
+                 * Although we cannot access the skb data directly from eBPF programs used in this
+                 * scenario, the kernel has prepared some fields for us to access through struct __sk_buff.
+                 * Load the protocol (IPv4, IPv6) used by the packet in flight once and cache it in R7
+                 * for later use.
+                 */
+                BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_6, offsetof(struct __sk_buff, protocol)),
+
+                /*
+                 * R8 is used to keep track of whether any address check has explicitly allowed or denied the packet
+                 * through ACCESS_DENIED or ACCESS_ALLOWED bits. Reset them both to 0 in the beginning.
+                 */
+                BPF_MOV32_IMM(BPF_REG_8, 0),
+        };
+
+        /*
+         * The access checkers compiled for the configured allowance and denial lists
+         * write to R8 at runtime. The following code prepares for an early exit that
+         * skip the accounting if the packet is denied.
+         *
+         * R0 = 1
+         * if (R8 == ACCESS_DENIED)
+         *     R0 = 0
+         *
+         * This means that if both ACCESS_DENIED and ACCESS_ALLOWED are set, the packet
+         * is allowed to pass.
+         */
+        struct bpf_insn post_insn[] = {
+                BPF_MOV64_IMM(BPF_REG_0, 1),
+                BPF_JMP_IMM(BPF_JNE, BPF_REG_8, ACCESS_DENIED, 1),
+                BPF_MOV64_IMM(BPF_REG_0, 0),
+        };
+
+        _cleanup_(bpf_program_unrefp) BPFProgram *p = NULL;
+        int accounting_map_fd, r;
+        bool access_enabled;
+
+        assert(u);
+        assert(ret);
+
+        accounting_map_fd = is_ingress ?
+                u->ip_accounting_ingress_map_fd :
+                u->ip_accounting_egress_map_fd;
+
+        access_enabled =
+                u->ipv4_allow_map_fd >= 0 ||
+                u->ipv6_allow_map_fd >= 0 ||
+                u->ipv4_deny_map_fd >= 0 ||
+                u->ipv6_deny_map_fd >= 0;
+
+        if (accounting_map_fd < 0 && !access_enabled) {
+                *ret = NULL;
+                return 0;
+        }
+
+        r = bpf_program_new(BPF_PROG_TYPE_CGROUP_SKB, &p);
+        if (r < 0)
+                return r;
+
+        r = bpf_program_add_instructions(p, pre_insn, ELEMENTSOF(pre_insn));
+        if (r < 0)
+                return r;
+
+        if (access_enabled) {
+                /*
+                 * The simple rule this function translates into eBPF instructions is:
+                 *
+                 * - Access will be granted when an address matches an entry in @list_allow
+                 * - Otherwise, access will be denied when an address matches an entry in @list_deny
+                 * - Otherwise, access will be granted
+                 */
+
+                if (u->ipv4_deny_map_fd >= 0) {
+                        r = add_lookup_instructions(p, u->ipv4_deny_map_fd, ETH_P_IP, is_ingress, ACCESS_DENIED);
+                        if (r < 0)
+                                return r;
+                }
+
+                if (u->ipv6_deny_map_fd >= 0) {
+                        r = add_lookup_instructions(p, u->ipv6_deny_map_fd, ETH_P_IPV6, is_ingress, ACCESS_DENIED);
+                        if (r < 0)
+                                return r;
+                }
+
+                if (u->ipv4_allow_map_fd >= 0) {
+                        r = add_lookup_instructions(p, u->ipv4_allow_map_fd, ETH_P_IP, is_ingress, ACCESS_ALLOWED);
+                        if (r < 0)
+                                return r;
+                }
+
+                if (u->ipv6_allow_map_fd >= 0) {
+                        r = add_lookup_instructions(p, u->ipv6_allow_map_fd, ETH_P_IPV6, is_ingress, ACCESS_ALLOWED);
+                        if (r < 0)
+                                return r;
+                }
+        }
+
+        r = bpf_program_add_instructions(p, post_insn, ELEMENTSOF(post_insn));
+        if (r < 0)
+                return r;
+
+        if (accounting_map_fd >= 0) {
+                struct bpf_insn insn[] = {
+                        /*
+                         * If R0 == 0, the packet will be denied; skip the accounting instructions in this case.
+                         * The jump label will be fixed up later.
+                         */
+                        BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 0),
+
+                        /* Count packets */
+                        BPF_MOV64_IMM(BPF_REG_0, MAP_KEY_PACKETS), /* r0 = 0 */
+                        BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_0, -4), /* *(u32 *)(fp - 4) = r0 */
+                        BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
+                        BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), /* r2 = fp - 4 */
+                        BPF_LD_MAP_FD(BPF_REG_1, accounting_map_fd), /* load map fd to r1 */
+                        BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
+                        BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2),
+                        BPF_MOV64_IMM(BPF_REG_1, 1), /* r1 = 1 */
+                        BPF_RAW_INSN(BPF_STX | BPF_XADD | BPF_DW, BPF_REG_0, BPF_REG_1, 0, 0), /* xadd r0 += r1 */
+
+                        /* Count bytes */
+                        BPF_MOV64_IMM(BPF_REG_0, MAP_KEY_BYTES), /* r0 = 1 */
+                        BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_0, -4), /* *(u32 *)(fp - 4) = r0 */
+                        BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
+                        BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), /* r2 = fp - 4 */
+                        BPF_LD_MAP_FD(BPF_REG_1, accounting_map_fd),
+                        BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
+                        BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2),
+                        BPF_LDX_MEM(BPF_W, BPF_REG_1, BPF_REG_6, offsetof(struct __sk_buff, len)), /* r1 = skb->len */
+                        BPF_RAW_INSN(BPF_STX | BPF_XADD | BPF_DW, BPF_REG_0, BPF_REG_1, 0, 0), /* xadd r0 += r1 */
+
+                        /* Allow the packet to pass */
+                        BPF_MOV64_IMM(BPF_REG_0, 1),
+                };
+
+                /* Jump label fixup */
+                insn[0].off = ELEMENTSOF(insn) - 1;
+
+                r = bpf_program_add_instructions(p, insn, ELEMENTSOF(insn));
+                if (r < 0)
+                        return r;
+        }
+
+        do {
+                /*
+                 * Exit from the eBPF program, R0 contains the verdict.
+                 * 0 means the packet is denied, 1 means the packet may pass.
+                 */
+                struct bpf_insn insn[] = {
+                        BPF_EXIT_INSN()
+                };
+
+                r = bpf_program_add_instructions(p, insn, ELEMENTSOF(insn));
+                if (r < 0)
+                        return r;
+        } while (false);
+
+        *ret = p;
+        p = NULL;
+
+        return 0;
+}
+
+static int bpf_firewall_count_access_items(IPAddressAccessItem *list, size_t *n_ipv4, size_t *n_ipv6) {
+        IPAddressAccessItem *a;
+
+        assert(n_ipv4);
+        assert(n_ipv6);
+
+        LIST_FOREACH(items, a, list) {
+                switch (a->family) {
+
+                case AF_INET:
+                        (*n_ipv4)++;
+                        break;
+
+                case AF_INET6:
+                        (*n_ipv6)++;
+                        break;
+
+                default:
+                        return -EAFNOSUPPORT;
+                }
+        }
+
+        return 0;
+}
+
+static int bpf_firewall_add_access_items(
+                IPAddressAccessItem *list,
+                int ipv4_map_fd,
+                int ipv6_map_fd,
+                int verdict) {
+
+        struct bpf_lpm_trie_key *key_ipv4, *key_ipv6;
+        uint64_t value = verdict;
+        IPAddressAccessItem *a;
+        int r;
+
+        key_ipv4 = alloca0(offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint32_t));
+        key_ipv6 = alloca0(offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint32_t) * 4);
+
+        LIST_FOREACH(items, a, list) {
+                switch (a->family) {
+
+                case AF_INET:
+                        key_ipv4->prefixlen = a->prefixlen;
+                        memcpy(key_ipv4->data, &a->address, sizeof(uint32_t));
+
+                        r = bpf_map_update_element(ipv4_map_fd, key_ipv4, &value);
+                        if (r < 0)
+                                return r;
+
+                        break;
+
+                case AF_INET6:
+                        key_ipv6->prefixlen = a->prefixlen;
+                        memcpy(key_ipv6->data, &a->address, 4 * sizeof(uint32_t));
+
+                        r = bpf_map_update_element(ipv6_map_fd, key_ipv6, &value);
+                        if (r < 0)
+                                return r;
+
+                        break;
+
+                default:
+                        return -EAFNOSUPPORT;
+                }
+        }
+
+        return 0;
+}
+
+static int bpf_firewall_prepare_access_maps(
+                Unit *u,
+                int verdict,
+                int *ret_ipv4_map_fd,
+                int *ret_ipv6_map_fd) {
+
+        _cleanup_close_ int ipv4_map_fd = -1, ipv6_map_fd = -1;
+        size_t n_ipv4 = 0, n_ipv6 = 0;
+        Unit *p;
+        int r;
+
+        assert(ret_ipv4_map_fd);
+        assert(ret_ipv6_map_fd);
+
+        for (p = u; p; p = UNIT_DEREF(p->slice)) {
+                CGroupContext *cc;
+
+                cc = unit_get_cgroup_context(p);
+                if (!cc)
+                        continue;
+
+                bpf_firewall_count_access_items(verdict == ACCESS_ALLOWED ? cc->ip_address_allow : cc->ip_address_deny, &n_ipv4, &n_ipv6);
+        }
+
+        if (n_ipv4 > 0) {
+                ipv4_map_fd = bpf_map_new(
+                                BPF_MAP_TYPE_LPM_TRIE,
+                                offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint32_t),
+                                sizeof(uint64_t),
+                                n_ipv4,
+                                BPF_F_NO_PREALLOC);
+                if (ipv4_map_fd < 0)
+                        return ipv4_map_fd;
+        }
+
+        if (n_ipv6 > 0) {
+                ipv6_map_fd = bpf_map_new(
+                                BPF_MAP_TYPE_LPM_TRIE,
+                                offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint32_t)*4,
+                                sizeof(uint64_t),
+                                n_ipv6,
+                                BPF_F_NO_PREALLOC);
+                if (ipv6_map_fd < 0)
+                        return ipv6_map_fd;
+        }
+
+        for (p = u; p; p = UNIT_DEREF(p->slice)) {
+                CGroupContext *cc;
+
+                cc = unit_get_cgroup_context(p);
+                if (!cc)
+                        continue;
+
+                r = bpf_firewall_add_access_items(verdict == ACCESS_ALLOWED ? cc->ip_address_allow : cc->ip_address_deny,
+                                                  ipv4_map_fd, ipv6_map_fd, verdict);
+                if (r < 0)
+                        return r;
+        }
+
+        *ret_ipv4_map_fd = ipv4_map_fd;
+        *ret_ipv6_map_fd = ipv6_map_fd;
+
+        ipv4_map_fd = ipv6_map_fd = -1;
+        return 0;
+}
+
+static int bpf_firewall_prepare_accounting_maps(bool enabled, int *fd_ingress, int *fd_egress) {
+        int r;
+
+        assert(fd_ingress);
+        assert(fd_egress);
+
+        if (enabled) {
+                if (*fd_ingress < 0) {
+                        r = bpf_map_new(BPF_MAP_TYPE_ARRAY, sizeof(int), sizeof(uint64_t), 2, 0);
+                        if (r < 0)
+                                return r;
+
+                        *fd_ingress = r;
+                }
+
+                if (*fd_egress < 0) {
+
+                        r = bpf_map_new(BPF_MAP_TYPE_ARRAY, sizeof(int), sizeof(uint64_t), 2, 0);
+                        if (r < 0)
+                                return r;
+
+                        *fd_egress = r;
+                }
+        } else {
+                *fd_ingress = safe_close(*fd_ingress);
+                *fd_egress = safe_close(*fd_egress);
+        }
+
+        return 0;
+}
+
+int bpf_firewall_compile(Unit *u) {
+        CGroupContext *cc;
+        int r;
+
+        assert(u);
+
+        r = bpf_firewall_supported();
+        if (r < 0)
+                return r;
+        if (r == 0) {
+                log_debug("BPF firewalling not supported on this systemd, proceeding without.");
+                return -EOPNOTSUPP;
+        }
+
+        /* Note that when we compile a new firewall we first flush out the access maps and the BPF programs themselves,
+         * but we reuse the the accounting maps. That way the firewall in effect always maps to the actual
+         * configuration, but we don't flush out the accounting unnecessarily */
+
+        u->ip_bpf_ingress = bpf_program_unref(u->ip_bpf_ingress);
+        u->ip_bpf_egress = bpf_program_unref(u->ip_bpf_egress);
+
+        u->ipv4_allow_map_fd = safe_close(u->ipv4_allow_map_fd);
+        u->ipv4_deny_map_fd = safe_close(u->ipv4_deny_map_fd);
+
+        u->ipv6_allow_map_fd = safe_close(u->ipv6_allow_map_fd);
+        u->ipv6_deny_map_fd = safe_close(u->ipv6_deny_map_fd);
+
+        cc = unit_get_cgroup_context(u);
+        if (!cc)
+                return -EINVAL;
+
+        r = bpf_firewall_prepare_access_maps(u, ACCESS_ALLOWED, &u->ipv4_allow_map_fd, &u->ipv6_allow_map_fd);
+        if (r < 0)
+                return log_error_errno(r, "Preparation of eBPF allow maps failed: %m");
+
+        r = bpf_firewall_prepare_access_maps(u, ACCESS_DENIED, &u->ipv4_deny_map_fd, &u->ipv6_deny_map_fd);
+        if (r < 0)
+                return log_error_errno(r, "Preparation of eBPF deny maps failed: %m");
+
+        r = bpf_firewall_prepare_accounting_maps(cc->ip_accounting, &u->ip_accounting_ingress_map_fd, &u->ip_accounting_egress_map_fd);
+        if (r < 0)
+                return log_error_errno(r, "Preparation of eBPF accounting maps failed: %m");
+
+        r = bpf_firewall_compile_bpf(u, true, &u->ip_bpf_ingress);
+        if (r < 0)
+                return log_error_errno(r, "Compilation for ingress BPF program failed: %m");
+
+        r = bpf_firewall_compile_bpf(u, false, &u->ip_bpf_egress);
+        if (r < 0)
+                return log_error_errno(r, "Compilation for egress BPF program failed: %m");
+
+        return 0;
+}
+
+int bpf_firewall_install(Unit *u) {
+        _cleanup_free_ char *path = NULL;
+        int r;
+
+        assert(u);
+
+        r = bpf_firewall_supported();
+        if (r < 0)
+                return r;
+        if (r == 0) {
+                log_debug("BPF firewalling not supported on this systemd, proceeding without.");
+                return -EOPNOTSUPP;
+        }
+
+        r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, NULL, &path);
+        if (r < 0)
+                return log_error_errno(r, "Failed to determine cgroup path: %m");
+
+        if (u->ip_bpf_egress) {
+                r = bpf_program_load_kernel(u->ip_bpf_egress, NULL, 0);
+                if (r < 0)
+                        return log_error_errno(r, "Kernel upload of egress BPF program failed: %m");
+
+                r = bpf_program_cgroup_attach(u->ip_bpf_egress, BPF_CGROUP_INET_EGRESS, path);
+                if (r < 0)
+                        return log_error_errno(r, "Attaching egress BPF program to cgroup %s failed: %m", path);
+        } else {
+                r = bpf_program_cgroup_detach(BPF_CGROUP_INET_EGRESS, path);
+                if (r < 0)
+                        return log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_ERR, r,
+                                              "Detaching egress BPF program from cgroup failed: %m");
+        }
+
+        if (u->ip_bpf_ingress) {
+                r = bpf_program_load_kernel(u->ip_bpf_ingress, NULL, 0);
+                if (r < 0)
+                        return log_error_errno(r, "Kernel upload of ingress BPF program failed: %m");
+
+                r = bpf_program_cgroup_attach(u->ip_bpf_ingress, BPF_CGROUP_INET_INGRESS, path);
+                if (r < 0)
+                        return log_error_errno(r, "Attaching ingress BPF program to cgroup %s failed: %m", path);
+        } else {
+                r = bpf_program_cgroup_detach(BPF_CGROUP_INET_INGRESS, path);
+                if (r < 0)
+                        return log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_ERR, r,
+                                              "Detaching ingress BPF program from cgroup failed: %m");
+        }
+
+        return 0;
+}
+
+int bpf_firewall_read_accounting(int map_fd, uint64_t *ret_bytes, uint64_t *ret_packets) {
+        uint64_t key, packets;
+        int r;
+
+        if (map_fd < 0)
+                return -EBADF;
+
+        if (ret_packets) {
+                key = MAP_KEY_PACKETS;
+                r = bpf_map_lookup_element(map_fd, &key, &packets);
+                if (r < 0)
+                        return r;
+        }
+
+        if (ret_bytes) {
+                key = MAP_KEY_BYTES;
+                r = bpf_map_lookup_element(map_fd, &key, ret_bytes);
+                if (r < 0)
+                        return r;
+        }
+
+        if (ret_packets)
+                *ret_packets = packets;
+
+        return 0;
+}
+
+int bpf_firewall_reset_accounting(int map_fd) {
+        uint64_t key, value = 0;
+        int r;
+
+        if (map_fd < 0)
+                return -EBADF;
+
+        key = MAP_KEY_PACKETS;
+        r = bpf_map_update_element(map_fd, &key, &value);
+        if (r < 0)
+                return r;
+
+        key = MAP_KEY_BYTES;
+        return bpf_map_update_element(map_fd, &key, &value);
+}
+
+
+int bpf_firewall_supported(void) {
+        static int supported = -1;
+        int fd, r;
+
+        /* Checks whether BPF firewalling is supported. For this, we check three things:
+         *
+         * a) whether we are privileged
+         * b) whether the unified hierarchy is being used
+         * c) the BPF implementation in the kernel supports BPF LPM TRIE maps, which we require
+         *
+         */
+
+        if (supported >= 0)
+                return supported;
+
+        if (geteuid() != 0)
+                return supported = false;
+
+        r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
+        if (r < 0)
+                return log_error_errno(r, "Can't determine whether the unified hierarchy is used: %m");
+        if (r == 0)
+                return supported = false;
+
+        fd = bpf_map_new(BPF_MAP_TYPE_LPM_TRIE,
+                         offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint64_t),
+                         sizeof(uint64_t),
+                         1,
+                         BPF_F_NO_PREALLOC);
+        if (fd < 0) {
+                log_debug_errno(r, "Can't allocate BPF LPM TRIE map, BPF firewalling is not supported: %m");
+                return supported = false;
+        }
+
+        safe_close(fd);
+
+        return supported = true;
+}
diff --git a/src/core/bpf-firewall.h b/src/core/bpf-firewall.h
new file mode 100644 (file)
index 0000000..870e314
--- /dev/null
@@ -0,0 +1,32 @@
+#pragma once
+
+/***
+  This file is part of systemd.
+
+  Copyright 2016 Daniel Mack
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU Lesser General Public License as published by
+  the Free Software Foundation; either version 2.1 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  Lesser General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <inttypes.h>
+
+#include "unit.h"
+
+int bpf_firewall_supported(void);
+
+int bpf_firewall_compile(Unit *u);
+int bpf_firewall_install(Unit *u);
+
+int bpf_firewall_read_accounting(int map_fd, uint64_t *ret_bytes, uint64_t *ret_packets);
+int bpf_firewall_reset_accounting(int map_fd);
index bea0d3528d1908e3b5bacedef4ae081be0042e25..ac600be117df9bd6ff54a2264a7249c53676a4f5 100644 (file)
@@ -3,6 +3,8 @@ libcore_la_sources = '''
         audit-fd.h
         automount.c
         automount.h
+        bpf-firewall.c
+        bpf-firewall.h
         cgroup.c
         cgroup.h
         dbus-automount.c