]>
Commit | Line | Data |
---|---|---|
52ad194e | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
f5e65279 MB |
2 | /*** |
3 | This file is part of systemd. | |
4 | ||
5 | Copyright 2016 Daniel Mack | |
6 | ||
7 | systemd is free software; you can redistribute it and/or modify it | |
8 | under the terms of the GNU Lesser General Public License as published by | |
9 | the Free Software Foundation; either version 2.1 of the License, or | |
10 | (at your option) any later version. | |
11 | ||
12 | systemd is distributed in the hope that it will be useful, but | |
13 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 | Lesser General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU Lesser General Public License | |
18 | along with systemd; If not, see <http://www.gnu.org/licenses/>. | |
19 | ***/ | |
20 | ||
21 | #include <arpa/inet.h> | |
22 | #include <assert.h> | |
23 | #include <errno.h> | |
24 | #include <fcntl.h> | |
25 | #include <linux/libbpf.h> | |
26 | #include <net/ethernet.h> | |
27 | #include <net/if.h> | |
28 | #include <netinet/ip.h> | |
29 | #include <netinet/ip6.h> | |
30 | #include <stddef.h> | |
31 | #include <stdio.h> | |
32 | #include <stdlib.h> | |
33 | #include <string.h> | |
34 | #include <unistd.h> | |
35 | ||
36 | #include "alloc-util.h" | |
37 | #include "bpf-firewall.h" | |
38 | #include "bpf-program.h" | |
39 | #include "fd-util.h" | |
40 | #include "ip-address-access.h" | |
41 | #include "unit.h" | |
42 | ||
43 | enum { | |
44 | MAP_KEY_PACKETS, | |
45 | MAP_KEY_BYTES, | |
46 | }; | |
47 | ||
48 | enum { | |
49 | ACCESS_ALLOWED = 1, | |
50 | ACCESS_DENIED = 2, | |
51 | }; | |
52 | ||
53 | /* Compile instructions for one list of addresses, one direction and one specific verdict on matches. */ | |
54 | ||
55 | static int add_lookup_instructions( | |
56 | BPFProgram *p, | |
57 | int map_fd, | |
58 | int protocol, | |
59 | bool is_ingress, | |
60 | int verdict) { | |
61 | ||
62 | int r, addr_offset, addr_size; | |
63 | ||
64 | assert(p); | |
65 | assert(map_fd >= 0); | |
66 | ||
67 | switch (protocol) { | |
68 | ||
69 | case ETH_P_IP: | |
70 | addr_size = sizeof(uint32_t); | |
71 | addr_offset = is_ingress ? | |
72 | offsetof(struct iphdr, saddr) : | |
73 | offsetof(struct iphdr, daddr); | |
74 | break; | |
75 | ||
76 | case ETH_P_IPV6: | |
77 | addr_size = 4 * sizeof(uint32_t); | |
78 | addr_offset = is_ingress ? | |
79 | offsetof(struct ip6_hdr, ip6_src.s6_addr) : | |
80 | offsetof(struct ip6_hdr, ip6_dst.s6_addr); | |
81 | break; | |
82 | ||
83 | default: | |
84 | return -EAFNOSUPPORT; | |
85 | } | |
86 | ||
87 | do { | |
88 | /* Compare IPv4 with one word instruction (32bit) */ | |
89 | struct bpf_insn insn[] = { | |
90 | /* If skb->protocol != ETH_P_IP, skip this whole block. The offset will be set later. */ | |
91 | BPF_JMP_IMM(BPF_JNE, BPF_REG_7, htobe16(protocol), 0), | |
92 | ||
93 | /* | |
94 | * Call into BPF_FUNC_skb_load_bytes to load the dst/src IP address | |
95 | * | |
96 | * R1: Pointer to the skb | |
97 | * R2: Data offset | |
98 | * R3: Destination buffer on the stack (r10 - 4) | |
99 | * R4: Number of bytes to read (4) | |
100 | */ | |
101 | ||
102 | BPF_MOV64_REG(BPF_REG_1, BPF_REG_6), | |
103 | BPF_MOV32_IMM(BPF_REG_2, addr_offset), | |
104 | ||
105 | BPF_MOV64_REG(BPF_REG_3, BPF_REG_10), | |
106 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_3, -addr_size), | |
107 | ||
108 | BPF_MOV32_IMM(BPF_REG_4, addr_size), | |
109 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_skb_load_bytes), | |
110 | ||
111 | /* | |
112 | * Call into BPF_FUNC_map_lookup_elem to see if the address matches any entry in the | |
113 | * LPM trie map. For this to work, the prefixlen field of 'struct bpf_lpm_trie_key' | |
114 | * has to be set to the maximum possible value. | |
115 | * | |
116 | * On success, the looked up value is stored in R0. For this application, the actual | |
117 | * value doesn't matter, however; we just set the bit in @verdict in R8 if we found any | |
118 | * matching value. | |
119 | */ | |
120 | ||
121 | BPF_LD_MAP_FD(BPF_REG_1, map_fd), | |
122 | BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), | |
123 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -addr_size - sizeof(uint32_t)), | |
124 | BPF_ST_MEM(BPF_W, BPF_REG_2, 0, addr_size * 8), | |
125 | ||
126 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
127 | BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 1), | |
128 | BPF_ALU32_IMM(BPF_OR, BPF_REG_8, verdict), | |
129 | }; | |
130 | ||
131 | /* Jump label fixup */ | |
132 | insn[0].off = ELEMENTSOF(insn) - 1; | |
133 | ||
134 | r = bpf_program_add_instructions(p, insn, ELEMENTSOF(insn)); | |
135 | if (r < 0) | |
136 | return r; | |
137 | ||
138 | } while (false); | |
139 | ||
140 | return 0; | |
141 | } | |
142 | ||
143 | static int bpf_firewall_compile_bpf( | |
144 | Unit *u, | |
145 | bool is_ingress, | |
146 | BPFProgram **ret) { | |
147 | ||
148 | struct bpf_insn pre_insn[] = { | |
149 | /* | |
150 | * When the eBPF program is entered, R1 contains the address of the skb. | |
151 | * However, R1-R5 are scratch registers that are not preserved when calling | |
152 | * into kernel functions, so we need to save anything that's supposed to | |
153 | * stay around to R6-R9. Save the skb to R6. | |
154 | */ | |
155 | BPF_MOV64_REG(BPF_REG_6, BPF_REG_1), | |
156 | ||
157 | /* | |
158 | * Although we cannot access the skb data directly from eBPF programs used in this | |
159 | * scenario, the kernel has prepared some fields for us to access through struct __sk_buff. | |
160 | * Load the protocol (IPv4, IPv6) used by the packet in flight once and cache it in R7 | |
161 | * for later use. | |
162 | */ | |
163 | BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_6, offsetof(struct __sk_buff, protocol)), | |
164 | ||
165 | /* | |
166 | * R8 is used to keep track of whether any address check has explicitly allowed or denied the packet | |
167 | * through ACCESS_DENIED or ACCESS_ALLOWED bits. Reset them both to 0 in the beginning. | |
168 | */ | |
169 | BPF_MOV32_IMM(BPF_REG_8, 0), | |
170 | }; | |
171 | ||
172 | /* | |
173 | * The access checkers compiled for the configured allowance and denial lists | |
174 | * write to R8 at runtime. The following code prepares for an early exit that | |
175 | * skip the accounting if the packet is denied. | |
176 | * | |
177 | * R0 = 1 | |
178 | * if (R8 == ACCESS_DENIED) | |
179 | * R0 = 0 | |
180 | * | |
181 | * This means that if both ACCESS_DENIED and ACCESS_ALLOWED are set, the packet | |
182 | * is allowed to pass. | |
183 | */ | |
184 | struct bpf_insn post_insn[] = { | |
185 | BPF_MOV64_IMM(BPF_REG_0, 1), | |
186 | BPF_JMP_IMM(BPF_JNE, BPF_REG_8, ACCESS_DENIED, 1), | |
187 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
188 | }; | |
189 | ||
190 | _cleanup_(bpf_program_unrefp) BPFProgram *p = NULL; | |
191 | int accounting_map_fd, r; | |
192 | bool access_enabled; | |
193 | ||
194 | assert(u); | |
195 | assert(ret); | |
196 | ||
197 | accounting_map_fd = is_ingress ? | |
198 | u->ip_accounting_ingress_map_fd : | |
199 | u->ip_accounting_egress_map_fd; | |
200 | ||
201 | access_enabled = | |
202 | u->ipv4_allow_map_fd >= 0 || | |
203 | u->ipv6_allow_map_fd >= 0 || | |
204 | u->ipv4_deny_map_fd >= 0 || | |
205 | u->ipv6_deny_map_fd >= 0; | |
206 | ||
207 | if (accounting_map_fd < 0 && !access_enabled) { | |
208 | *ret = NULL; | |
209 | return 0; | |
210 | } | |
211 | ||
212 | r = bpf_program_new(BPF_PROG_TYPE_CGROUP_SKB, &p); | |
213 | if (r < 0) | |
214 | return r; | |
215 | ||
216 | r = bpf_program_add_instructions(p, pre_insn, ELEMENTSOF(pre_insn)); | |
217 | if (r < 0) | |
218 | return r; | |
219 | ||
220 | if (access_enabled) { | |
221 | /* | |
222 | * The simple rule this function translates into eBPF instructions is: | |
223 | * | |
224 | * - Access will be granted when an address matches an entry in @list_allow | |
225 | * - Otherwise, access will be denied when an address matches an entry in @list_deny | |
226 | * - Otherwise, access will be granted | |
227 | */ | |
228 | ||
229 | if (u->ipv4_deny_map_fd >= 0) { | |
230 | r = add_lookup_instructions(p, u->ipv4_deny_map_fd, ETH_P_IP, is_ingress, ACCESS_DENIED); | |
231 | if (r < 0) | |
232 | return r; | |
233 | } | |
234 | ||
235 | if (u->ipv6_deny_map_fd >= 0) { | |
236 | r = add_lookup_instructions(p, u->ipv6_deny_map_fd, ETH_P_IPV6, is_ingress, ACCESS_DENIED); | |
237 | if (r < 0) | |
238 | return r; | |
239 | } | |
240 | ||
241 | if (u->ipv4_allow_map_fd >= 0) { | |
242 | r = add_lookup_instructions(p, u->ipv4_allow_map_fd, ETH_P_IP, is_ingress, ACCESS_ALLOWED); | |
243 | if (r < 0) | |
244 | return r; | |
245 | } | |
246 | ||
247 | if (u->ipv6_allow_map_fd >= 0) { | |
248 | r = add_lookup_instructions(p, u->ipv6_allow_map_fd, ETH_P_IPV6, is_ingress, ACCESS_ALLOWED); | |
249 | if (r < 0) | |
250 | return r; | |
251 | } | |
252 | } | |
253 | ||
254 | r = bpf_program_add_instructions(p, post_insn, ELEMENTSOF(post_insn)); | |
255 | if (r < 0) | |
256 | return r; | |
257 | ||
258 | if (accounting_map_fd >= 0) { | |
259 | struct bpf_insn insn[] = { | |
260 | /* | |
261 | * If R0 == 0, the packet will be denied; skip the accounting instructions in this case. | |
262 | * The jump label will be fixed up later. | |
263 | */ | |
264 | BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 0), | |
265 | ||
266 | /* Count packets */ | |
267 | BPF_MOV64_IMM(BPF_REG_0, MAP_KEY_PACKETS), /* r0 = 0 */ | |
268 | BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_0, -4), /* *(u32 *)(fp - 4) = r0 */ | |
269 | BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), | |
270 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), /* r2 = fp - 4 */ | |
271 | BPF_LD_MAP_FD(BPF_REG_1, accounting_map_fd), /* load map fd to r1 */ | |
272 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
273 | BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2), | |
274 | BPF_MOV64_IMM(BPF_REG_1, 1), /* r1 = 1 */ | |
275 | BPF_RAW_INSN(BPF_STX | BPF_XADD | BPF_DW, BPF_REG_0, BPF_REG_1, 0, 0), /* xadd r0 += r1 */ | |
276 | ||
277 | /* Count bytes */ | |
278 | BPF_MOV64_IMM(BPF_REG_0, MAP_KEY_BYTES), /* r0 = 1 */ | |
279 | BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_0, -4), /* *(u32 *)(fp - 4) = r0 */ | |
280 | BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), | |
281 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), /* r2 = fp - 4 */ | |
282 | BPF_LD_MAP_FD(BPF_REG_1, accounting_map_fd), | |
283 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
284 | BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2), | |
285 | BPF_LDX_MEM(BPF_W, BPF_REG_1, BPF_REG_6, offsetof(struct __sk_buff, len)), /* r1 = skb->len */ | |
286 | BPF_RAW_INSN(BPF_STX | BPF_XADD | BPF_DW, BPF_REG_0, BPF_REG_1, 0, 0), /* xadd r0 += r1 */ | |
287 | ||
288 | /* Allow the packet to pass */ | |
289 | BPF_MOV64_IMM(BPF_REG_0, 1), | |
290 | }; | |
291 | ||
292 | /* Jump label fixup */ | |
293 | insn[0].off = ELEMENTSOF(insn) - 1; | |
294 | ||
295 | r = bpf_program_add_instructions(p, insn, ELEMENTSOF(insn)); | |
296 | if (r < 0) | |
297 | return r; | |
298 | } | |
299 | ||
300 | do { | |
301 | /* | |
302 | * Exit from the eBPF program, R0 contains the verdict. | |
303 | * 0 means the packet is denied, 1 means the packet may pass. | |
304 | */ | |
305 | struct bpf_insn insn[] = { | |
306 | BPF_EXIT_INSN() | |
307 | }; | |
308 | ||
309 | r = bpf_program_add_instructions(p, insn, ELEMENTSOF(insn)); | |
310 | if (r < 0) | |
311 | return r; | |
312 | } while (false); | |
313 | ||
314 | *ret = p; | |
315 | p = NULL; | |
316 | ||
317 | return 0; | |
318 | } | |
319 | ||
320 | static int bpf_firewall_count_access_items(IPAddressAccessItem *list, size_t *n_ipv4, size_t *n_ipv6) { | |
321 | IPAddressAccessItem *a; | |
322 | ||
323 | assert(n_ipv4); | |
324 | assert(n_ipv6); | |
325 | ||
326 | LIST_FOREACH(items, a, list) { | |
327 | switch (a->family) { | |
328 | ||
329 | case AF_INET: | |
330 | (*n_ipv4)++; | |
331 | break; | |
332 | ||
333 | case AF_INET6: | |
334 | (*n_ipv6)++; | |
335 | break; | |
336 | ||
337 | default: | |
338 | return -EAFNOSUPPORT; | |
339 | } | |
340 | } | |
341 | ||
342 | return 0; | |
343 | } | |
344 | ||
345 | static int bpf_firewall_add_access_items( | |
346 | IPAddressAccessItem *list, | |
347 | int ipv4_map_fd, | |
348 | int ipv6_map_fd, | |
349 | int verdict) { | |
350 | ||
351 | struct bpf_lpm_trie_key *key_ipv4, *key_ipv6; | |
352 | uint64_t value = verdict; | |
353 | IPAddressAccessItem *a; | |
354 | int r; | |
355 | ||
356 | key_ipv4 = alloca0(offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint32_t)); | |
357 | key_ipv6 = alloca0(offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint32_t) * 4); | |
358 | ||
359 | LIST_FOREACH(items, a, list) { | |
360 | switch (a->family) { | |
361 | ||
362 | case AF_INET: | |
363 | key_ipv4->prefixlen = a->prefixlen; | |
364 | memcpy(key_ipv4->data, &a->address, sizeof(uint32_t)); | |
365 | ||
366 | r = bpf_map_update_element(ipv4_map_fd, key_ipv4, &value); | |
367 | if (r < 0) | |
368 | return r; | |
369 | ||
370 | break; | |
371 | ||
372 | case AF_INET6: | |
373 | key_ipv6->prefixlen = a->prefixlen; | |
374 | memcpy(key_ipv6->data, &a->address, 4 * sizeof(uint32_t)); | |
375 | ||
376 | r = bpf_map_update_element(ipv6_map_fd, key_ipv6, &value); | |
377 | if (r < 0) | |
378 | return r; | |
379 | ||
380 | break; | |
381 | ||
382 | default: | |
383 | return -EAFNOSUPPORT; | |
384 | } | |
385 | } | |
386 | ||
387 | return 0; | |
388 | } | |
389 | ||
390 | static int bpf_firewall_prepare_access_maps( | |
391 | Unit *u, | |
392 | int verdict, | |
393 | int *ret_ipv4_map_fd, | |
394 | int *ret_ipv6_map_fd) { | |
395 | ||
396 | _cleanup_close_ int ipv4_map_fd = -1, ipv6_map_fd = -1; | |
397 | size_t n_ipv4 = 0, n_ipv6 = 0; | |
398 | Unit *p; | |
399 | int r; | |
400 | ||
401 | assert(ret_ipv4_map_fd); | |
402 | assert(ret_ipv6_map_fd); | |
403 | ||
404 | for (p = u; p; p = UNIT_DEREF(p->slice)) { | |
405 | CGroupContext *cc; | |
406 | ||
407 | cc = unit_get_cgroup_context(p); | |
408 | if (!cc) | |
409 | continue; | |
410 | ||
411 | bpf_firewall_count_access_items(verdict == ACCESS_ALLOWED ? cc->ip_address_allow : cc->ip_address_deny, &n_ipv4, &n_ipv6); | |
412 | } | |
413 | ||
414 | if (n_ipv4 > 0) { | |
415 | ipv4_map_fd = bpf_map_new( | |
416 | BPF_MAP_TYPE_LPM_TRIE, | |
417 | offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint32_t), | |
418 | sizeof(uint64_t), | |
419 | n_ipv4, | |
420 | BPF_F_NO_PREALLOC); | |
421 | if (ipv4_map_fd < 0) | |
422 | return ipv4_map_fd; | |
423 | } | |
424 | ||
425 | if (n_ipv6 > 0) { | |
426 | ipv6_map_fd = bpf_map_new( | |
427 | BPF_MAP_TYPE_LPM_TRIE, | |
428 | offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint32_t)*4, | |
429 | sizeof(uint64_t), | |
430 | n_ipv6, | |
431 | BPF_F_NO_PREALLOC); | |
432 | if (ipv6_map_fd < 0) | |
433 | return ipv6_map_fd; | |
434 | } | |
435 | ||
436 | for (p = u; p; p = UNIT_DEREF(p->slice)) { | |
437 | CGroupContext *cc; | |
438 | ||
439 | cc = unit_get_cgroup_context(p); | |
440 | if (!cc) | |
441 | continue; | |
442 | ||
443 | r = bpf_firewall_add_access_items(verdict == ACCESS_ALLOWED ? cc->ip_address_allow : cc->ip_address_deny, | |
444 | ipv4_map_fd, ipv6_map_fd, verdict); | |
445 | if (r < 0) | |
446 | return r; | |
447 | } | |
448 | ||
449 | *ret_ipv4_map_fd = ipv4_map_fd; | |
450 | *ret_ipv6_map_fd = ipv6_map_fd; | |
451 | ||
452 | ipv4_map_fd = ipv6_map_fd = -1; | |
453 | return 0; | |
454 | } | |
455 | ||
456 | static int bpf_firewall_prepare_accounting_maps(bool enabled, int *fd_ingress, int *fd_egress) { | |
457 | int r; | |
458 | ||
459 | assert(fd_ingress); | |
460 | assert(fd_egress); | |
461 | ||
462 | if (enabled) { | |
463 | if (*fd_ingress < 0) { | |
464 | r = bpf_map_new(BPF_MAP_TYPE_ARRAY, sizeof(int), sizeof(uint64_t), 2, 0); | |
465 | if (r < 0) | |
466 | return r; | |
467 | ||
468 | *fd_ingress = r; | |
469 | } | |
470 | ||
471 | if (*fd_egress < 0) { | |
472 | ||
473 | r = bpf_map_new(BPF_MAP_TYPE_ARRAY, sizeof(int), sizeof(uint64_t), 2, 0); | |
474 | if (r < 0) | |
475 | return r; | |
476 | ||
477 | *fd_egress = r; | |
478 | } | |
479 | } else { | |
480 | *fd_ingress = safe_close(*fd_ingress); | |
481 | *fd_egress = safe_close(*fd_egress); | |
482 | } | |
483 | ||
484 | return 0; | |
485 | } | |
486 | ||
487 | int bpf_firewall_compile(Unit *u) { | |
488 | CGroupContext *cc; | |
489 | int r; | |
490 | ||
491 | assert(u); | |
492 | ||
493 | r = bpf_firewall_supported(); | |
494 | if (r < 0) | |
495 | return r; | |
496 | if (r == 0) { | |
497 | log_debug("BPF firewalling not supported on this systemd, proceeding without."); | |
498 | return -EOPNOTSUPP; | |
499 | } | |
500 | ||
501 | /* Note that when we compile a new firewall we first flush out the access maps and the BPF programs themselves, | |
502 | * but we reuse the the accounting maps. That way the firewall in effect always maps to the actual | |
503 | * configuration, but we don't flush out the accounting unnecessarily */ | |
504 | ||
505 | u->ip_bpf_ingress = bpf_program_unref(u->ip_bpf_ingress); | |
506 | u->ip_bpf_egress = bpf_program_unref(u->ip_bpf_egress); | |
507 | ||
508 | u->ipv4_allow_map_fd = safe_close(u->ipv4_allow_map_fd); | |
509 | u->ipv4_deny_map_fd = safe_close(u->ipv4_deny_map_fd); | |
510 | ||
511 | u->ipv6_allow_map_fd = safe_close(u->ipv6_allow_map_fd); | |
512 | u->ipv6_deny_map_fd = safe_close(u->ipv6_deny_map_fd); | |
513 | ||
514 | cc = unit_get_cgroup_context(u); | |
515 | if (!cc) | |
516 | return -EINVAL; | |
517 | ||
518 | r = bpf_firewall_prepare_access_maps(u, ACCESS_ALLOWED, &u->ipv4_allow_map_fd, &u->ipv6_allow_map_fd); | |
519 | if (r < 0) | |
520 | return log_error_errno(r, "Preparation of eBPF allow maps failed: %m"); | |
521 | ||
522 | r = bpf_firewall_prepare_access_maps(u, ACCESS_DENIED, &u->ipv4_deny_map_fd, &u->ipv6_deny_map_fd); | |
523 | if (r < 0) | |
524 | return log_error_errno(r, "Preparation of eBPF deny maps failed: %m"); | |
525 | ||
526 | r = bpf_firewall_prepare_accounting_maps(cc->ip_accounting, &u->ip_accounting_ingress_map_fd, &u->ip_accounting_egress_map_fd); | |
527 | if (r < 0) | |
528 | return log_error_errno(r, "Preparation of eBPF accounting maps failed: %m"); | |
529 | ||
530 | r = bpf_firewall_compile_bpf(u, true, &u->ip_bpf_ingress); | |
531 | if (r < 0) | |
532 | return log_error_errno(r, "Compilation for ingress BPF program failed: %m"); | |
533 | ||
534 | r = bpf_firewall_compile_bpf(u, false, &u->ip_bpf_egress); | |
535 | if (r < 0) | |
536 | return log_error_errno(r, "Compilation for egress BPF program failed: %m"); | |
537 | ||
538 | return 0; | |
539 | } | |
540 | ||
541 | int bpf_firewall_install(Unit *u) { | |
542 | _cleanup_free_ char *path = NULL; | |
543 | CGroupContext *cc; | |
544 | int r; | |
545 | ||
546 | assert(u); | |
547 | ||
548 | if (!u->cgroup_path) | |
549 | return -EINVAL; | |
550 | ||
551 | cc = unit_get_cgroup_context(u); | |
552 | if (!cc) | |
553 | return -EINVAL; | |
554 | ||
555 | r = bpf_firewall_supported(); | |
556 | if (r < 0) | |
557 | return r; | |
558 | if (r == 0) { | |
559 | log_debug("BPF firewalling not supported on this systemd, proceeding without."); | |
560 | return -EOPNOTSUPP; | |
561 | } | |
562 | ||
563 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, NULL, &path); | |
564 | if (r < 0) | |
565 | return log_error_errno(r, "Failed to determine cgroup path: %m"); | |
566 | ||
567 | if (u->ip_bpf_egress) { | |
568 | r = bpf_program_load_kernel(u->ip_bpf_egress, NULL, 0); | |
569 | if (r < 0) | |
570 | return log_error_errno(r, "Kernel upload of egress BPF program failed: %m"); | |
571 | ||
572 | r = bpf_program_cgroup_attach(u->ip_bpf_egress, BPF_CGROUP_INET_EGRESS, path, cc->delegate ? BPF_F_ALLOW_OVERRIDE : 0); | |
573 | if (r < 0) | |
574 | return log_error_errno(r, "Attaching egress BPF program to cgroup %s failed: %m", path); | |
575 | } else { | |
576 | r = bpf_program_cgroup_detach(BPF_CGROUP_INET_EGRESS, path); | |
577 | if (r < 0) | |
578 | return log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_ERR, r, | |
579 | "Detaching egress BPF program from cgroup failed: %m"); | |
580 | } | |
581 | ||
582 | if (u->ip_bpf_ingress) { | |
583 | r = bpf_program_load_kernel(u->ip_bpf_ingress, NULL, 0); | |
584 | if (r < 0) | |
585 | return log_error_errno(r, "Kernel upload of ingress BPF program failed: %m"); | |
586 | ||
587 | r = bpf_program_cgroup_attach(u->ip_bpf_ingress, BPF_CGROUP_INET_INGRESS, path, cc->delegate ? BPF_F_ALLOW_OVERRIDE : 0); | |
588 | if (r < 0) | |
589 | return log_error_errno(r, "Attaching ingress BPF program to cgroup %s failed: %m", path); | |
590 | } else { | |
591 | r = bpf_program_cgroup_detach(BPF_CGROUP_INET_INGRESS, path); | |
592 | if (r < 0) | |
593 | return log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_ERR, r, | |
594 | "Detaching ingress BPF program from cgroup failed: %m"); | |
595 | } | |
596 | ||
597 | return 0; | |
598 | } | |
599 | ||
600 | int bpf_firewall_read_accounting(int map_fd, uint64_t *ret_bytes, uint64_t *ret_packets) { | |
601 | uint64_t key, packets; | |
602 | int r; | |
603 | ||
604 | if (map_fd < 0) | |
605 | return -EBADF; | |
606 | ||
607 | if (ret_packets) { | |
608 | key = MAP_KEY_PACKETS; | |
609 | r = bpf_map_lookup_element(map_fd, &key, &packets); | |
610 | if (r < 0) | |
611 | return r; | |
612 | } | |
613 | ||
614 | if (ret_bytes) { | |
615 | key = MAP_KEY_BYTES; | |
616 | r = bpf_map_lookup_element(map_fd, &key, ret_bytes); | |
617 | if (r < 0) | |
618 | return r; | |
619 | } | |
620 | ||
621 | if (ret_packets) | |
622 | *ret_packets = packets; | |
623 | ||
624 | return 0; | |
625 | } | |
626 | ||
627 | int bpf_firewall_reset_accounting(int map_fd) { | |
628 | uint64_t key, value = 0; | |
629 | int r; | |
630 | ||
631 | if (map_fd < 0) | |
632 | return -EBADF; | |
633 | ||
634 | key = MAP_KEY_PACKETS; | |
635 | r = bpf_map_update_element(map_fd, &key, &value); | |
636 | if (r < 0) | |
637 | return r; | |
638 | ||
639 | key = MAP_KEY_BYTES; | |
640 | return bpf_map_update_element(map_fd, &key, &value); | |
641 | } | |
642 | ||
643 | ||
644 | int bpf_firewall_supported(void) { | |
52ad194e MB |
645 | struct bpf_insn trivial[] = { |
646 | BPF_MOV64_IMM(BPF_REG_0, 1), | |
647 | BPF_EXIT_INSN() | |
648 | }; | |
649 | ||
650 | _cleanup_(bpf_program_unrefp) BPFProgram *program = NULL; | |
f5e65279 | 651 | static int supported = -1; |
52ad194e | 652 | union bpf_attr attr; |
f5e65279 MB |
653 | int fd, r; |
654 | ||
52ad194e | 655 | /* Checks whether BPF firewalling is supported. For this, we check five things: |
f5e65279 MB |
656 | * |
657 | * a) whether we are privileged | |
658 | * b) whether the unified hierarchy is being used | |
659 | * c) the BPF implementation in the kernel supports BPF LPM TRIE maps, which we require | |
52ad194e MB |
660 | * d) the BPF implementation in the kernel supports BPF_PROG_TYPE_CGROUP_SKB programs, which we require |
661 | * e) the BPF implementation in the kernel supports the BPF_PROG_ATTACH call, which we require | |
f5e65279 MB |
662 | * |
663 | */ | |
664 | ||
665 | if (supported >= 0) | |
666 | return supported; | |
667 | ||
52ad194e MB |
668 | if (geteuid() != 0) { |
669 | log_debug("Not enough privileges, BPF firewalling is not supported."); | |
f5e65279 | 670 | return supported = false; |
52ad194e | 671 | } |
f5e65279 MB |
672 | |
673 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
674 | if (r < 0) | |
675 | return log_error_errno(r, "Can't determine whether the unified hierarchy is used: %m"); | |
52ad194e MB |
676 | if (r == 0) { |
677 | log_debug("Not running with unified cgroups, BPF firewalling is not supported."); | |
f5e65279 | 678 | return supported = false; |
52ad194e | 679 | } |
f5e65279 MB |
680 | |
681 | fd = bpf_map_new(BPF_MAP_TYPE_LPM_TRIE, | |
682 | offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint64_t), | |
683 | sizeof(uint64_t), | |
684 | 1, | |
685 | BPF_F_NO_PREALLOC); | |
686 | if (fd < 0) { | |
687 | log_debug_errno(r, "Can't allocate BPF LPM TRIE map, BPF firewalling is not supported: %m"); | |
688 | return supported = false; | |
689 | } | |
690 | ||
691 | safe_close(fd); | |
692 | ||
52ad194e MB |
693 | if (bpf_program_new(BPF_PROG_TYPE_CGROUP_SKB, &program) < 0) { |
694 | log_debug_errno(r, "Can't allocate CGROUP SKB BPF program, BPF firewalling is not supported: %m"); | |
695 | return supported = false; | |
696 | } | |
697 | ||
698 | r = bpf_program_add_instructions(program, trivial, ELEMENTSOF(trivial)); | |
699 | if (r < 0) { | |
700 | log_debug_errno(r, "Can't add trivial instructions to CGROUP SKB BPF program, BPF firewalling is not supported: %m"); | |
701 | return supported = false; | |
702 | } | |
703 | ||
704 | r = bpf_program_load_kernel(program, NULL, 0); | |
705 | if (r < 0) { | |
706 | log_debug_errno(r, "Can't load kernel CGROUP SKB BPF program, BPF firewalling is not supported: %m"); | |
707 | return supported = false; | |
708 | } | |
709 | ||
710 | /* Unfortunately the kernel allows us to create BPF_PROG_TYPE_CGROUP_SKB programs even when CONFIG_CGROUP_BPF | |
711 | * is turned off at kernel compilation time. This sucks of course: why does it allow us to create a cgroup BPF | |
712 | * program if we can't do a thing with it later? | |
713 | * | |
714 | * We detect this case by issuing the BPF_PROG_ATTACH bpf() call with invalid file descriptors: if | |
715 | * CONFIG_CGROUP_BPF is turned off, then the call will fail early with EINVAL. If it is turned on the | |
716 | * parameters are validated however, and that'll fail with EBADF then. */ | |
717 | ||
718 | attr = (union bpf_attr) { | |
719 | .attach_type = BPF_CGROUP_INET_EGRESS, | |
720 | .target_fd = -1, | |
721 | .attach_bpf_fd = -1, | |
722 | }; | |
723 | ||
724 | r = bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); | |
725 | if (r < 0) { | |
726 | if (errno == EBADF) /* YAY! */ | |
727 | return supported = true; | |
728 | ||
729 | log_debug_errno(errno, "Didn't get EBADF from BPF_PROG_ATTACH, BPF firewalling is not supported: %m"); | |
730 | } else | |
731 | log_debug("Wut? kernel accepted our invalid BPF_PROG_ATTACH call? Something is weird, assuming BPF firewalling is broken and hence not supported."); | |
732 | ||
733 | return supported = false; | |
f5e65279 | 734 | } |