]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
1988a9d1 DM |
2 | /*** |
3 | This file is part of systemd. | |
4 | ||
5 | Copyright 2016 Daniel Mack | |
1988a9d1 DM |
6 | ***/ |
7 | ||
8 | #include <arpa/inet.h> | |
9 | #include <assert.h> | |
10 | #include <errno.h> | |
11 | #include <fcntl.h> | |
12 | #include <linux/libbpf.h> | |
13 | #include <net/ethernet.h> | |
14 | #include <net/if.h> | |
15 | #include <netinet/ip.h> | |
16 | #include <netinet/ip6.h> | |
17 | #include <stddef.h> | |
18 | #include <stdio.h> | |
19 | #include <stdlib.h> | |
20 | #include <string.h> | |
21 | #include <unistd.h> | |
22 | ||
23 | #include "alloc-util.h" | |
24 | #include "bpf-firewall.h" | |
25 | #include "bpf-program.h" | |
26 | #include "fd-util.h" | |
27 | #include "ip-address-access.h" | |
28 | #include "unit.h" | |
29 | ||
30 | enum { | |
31 | MAP_KEY_PACKETS, | |
32 | MAP_KEY_BYTES, | |
33 | }; | |
34 | ||
35 | enum { | |
36 | ACCESS_ALLOWED = 1, | |
37 | ACCESS_DENIED = 2, | |
38 | }; | |
39 | ||
40 | /* Compile instructions for one list of addresses, one direction and one specific verdict on matches. */ | |
41 | ||
42 | static int add_lookup_instructions( | |
43 | BPFProgram *p, | |
44 | int map_fd, | |
45 | int protocol, | |
46 | bool is_ingress, | |
47 | int verdict) { | |
48 | ||
49 | int r, addr_offset, addr_size; | |
50 | ||
51 | assert(p); | |
52 | assert(map_fd >= 0); | |
53 | ||
54 | switch (protocol) { | |
55 | ||
56 | case ETH_P_IP: | |
57 | addr_size = sizeof(uint32_t); | |
58 | addr_offset = is_ingress ? | |
59 | offsetof(struct iphdr, saddr) : | |
60 | offsetof(struct iphdr, daddr); | |
61 | break; | |
62 | ||
63 | case ETH_P_IPV6: | |
64 | addr_size = 4 * sizeof(uint32_t); | |
65 | addr_offset = is_ingress ? | |
66 | offsetof(struct ip6_hdr, ip6_src.s6_addr) : | |
67 | offsetof(struct ip6_hdr, ip6_dst.s6_addr); | |
68 | break; | |
69 | ||
70 | default: | |
71 | return -EAFNOSUPPORT; | |
72 | } | |
73 | ||
74 | do { | |
75 | /* Compare IPv4 with one word instruction (32bit) */ | |
76 | struct bpf_insn insn[] = { | |
77 | /* If skb->protocol != ETH_P_IP, skip this whole block. The offset will be set later. */ | |
78 | BPF_JMP_IMM(BPF_JNE, BPF_REG_7, htobe16(protocol), 0), | |
79 | ||
80 | /* | |
81 | * Call into BPF_FUNC_skb_load_bytes to load the dst/src IP address | |
82 | * | |
83 | * R1: Pointer to the skb | |
84 | * R2: Data offset | |
85 | * R3: Destination buffer on the stack (r10 - 4) | |
86 | * R4: Number of bytes to read (4) | |
87 | */ | |
88 | ||
89 | BPF_MOV64_REG(BPF_REG_1, BPF_REG_6), | |
90 | BPF_MOV32_IMM(BPF_REG_2, addr_offset), | |
91 | ||
92 | BPF_MOV64_REG(BPF_REG_3, BPF_REG_10), | |
93 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_3, -addr_size), | |
94 | ||
95 | BPF_MOV32_IMM(BPF_REG_4, addr_size), | |
96 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_skb_load_bytes), | |
97 | ||
98 | /* | |
99 | * Call into BPF_FUNC_map_lookup_elem to see if the address matches any entry in the | |
100 | * LPM trie map. For this to work, the prefixlen field of 'struct bpf_lpm_trie_key' | |
101 | * has to be set to the maximum possible value. | |
102 | * | |
103 | * On success, the looked up value is stored in R0. For this application, the actual | |
104 | * value doesn't matter, however; we just set the bit in @verdict in R8 if we found any | |
105 | * matching value. | |
106 | */ | |
107 | ||
108 | BPF_LD_MAP_FD(BPF_REG_1, map_fd), | |
109 | BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), | |
110 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -addr_size - sizeof(uint32_t)), | |
111 | BPF_ST_MEM(BPF_W, BPF_REG_2, 0, addr_size * 8), | |
112 | ||
113 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
114 | BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 1), | |
115 | BPF_ALU32_IMM(BPF_OR, BPF_REG_8, verdict), | |
116 | }; | |
117 | ||
118 | /* Jump label fixup */ | |
119 | insn[0].off = ELEMENTSOF(insn) - 1; | |
120 | ||
121 | r = bpf_program_add_instructions(p, insn, ELEMENTSOF(insn)); | |
122 | if (r < 0) | |
123 | return r; | |
124 | ||
125 | } while (false); | |
126 | ||
127 | return 0; | |
128 | } | |
129 | ||
130 | static int bpf_firewall_compile_bpf( | |
131 | Unit *u, | |
132 | bool is_ingress, | |
133 | BPFProgram **ret) { | |
134 | ||
135 | struct bpf_insn pre_insn[] = { | |
136 | /* | |
137 | * When the eBPF program is entered, R1 contains the address of the skb. | |
138 | * However, R1-R5 are scratch registers that are not preserved when calling | |
139 | * into kernel functions, so we need to save anything that's supposed to | |
140 | * stay around to R6-R9. Save the skb to R6. | |
141 | */ | |
142 | BPF_MOV64_REG(BPF_REG_6, BPF_REG_1), | |
143 | ||
144 | /* | |
145 | * Although we cannot access the skb data directly from eBPF programs used in this | |
146 | * scenario, the kernel has prepared some fields for us to access through struct __sk_buff. | |
147 | * Load the protocol (IPv4, IPv6) used by the packet in flight once and cache it in R7 | |
148 | * for later use. | |
149 | */ | |
150 | BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_6, offsetof(struct __sk_buff, protocol)), | |
151 | ||
152 | /* | |
153 | * R8 is used to keep track of whether any address check has explicitly allowed or denied the packet | |
154 | * through ACCESS_DENIED or ACCESS_ALLOWED bits. Reset them both to 0 in the beginning. | |
155 | */ | |
156 | BPF_MOV32_IMM(BPF_REG_8, 0), | |
157 | }; | |
158 | ||
159 | /* | |
160 | * The access checkers compiled for the configured allowance and denial lists | |
161 | * write to R8 at runtime. The following code prepares for an early exit that | |
162 | * skip the accounting if the packet is denied. | |
163 | * | |
164 | * R0 = 1 | |
165 | * if (R8 == ACCESS_DENIED) | |
166 | * R0 = 0 | |
167 | * | |
168 | * This means that if both ACCESS_DENIED and ACCESS_ALLOWED are set, the packet | |
169 | * is allowed to pass. | |
170 | */ | |
171 | struct bpf_insn post_insn[] = { | |
172 | BPF_MOV64_IMM(BPF_REG_0, 1), | |
173 | BPF_JMP_IMM(BPF_JNE, BPF_REG_8, ACCESS_DENIED, 1), | |
174 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
175 | }; | |
176 | ||
177 | _cleanup_(bpf_program_unrefp) BPFProgram *p = NULL; | |
178 | int accounting_map_fd, r; | |
179 | bool access_enabled; | |
180 | ||
181 | assert(u); | |
182 | assert(ret); | |
183 | ||
184 | accounting_map_fd = is_ingress ? | |
185 | u->ip_accounting_ingress_map_fd : | |
186 | u->ip_accounting_egress_map_fd; | |
187 | ||
188 | access_enabled = | |
189 | u->ipv4_allow_map_fd >= 0 || | |
190 | u->ipv6_allow_map_fd >= 0 || | |
191 | u->ipv4_deny_map_fd >= 0 || | |
192 | u->ipv6_deny_map_fd >= 0; | |
193 | ||
194 | if (accounting_map_fd < 0 && !access_enabled) { | |
195 | *ret = NULL; | |
196 | return 0; | |
197 | } | |
198 | ||
199 | r = bpf_program_new(BPF_PROG_TYPE_CGROUP_SKB, &p); | |
200 | if (r < 0) | |
201 | return r; | |
202 | ||
203 | r = bpf_program_add_instructions(p, pre_insn, ELEMENTSOF(pre_insn)); | |
204 | if (r < 0) | |
205 | return r; | |
206 | ||
207 | if (access_enabled) { | |
208 | /* | |
209 | * The simple rule this function translates into eBPF instructions is: | |
210 | * | |
211 | * - Access will be granted when an address matches an entry in @list_allow | |
212 | * - Otherwise, access will be denied when an address matches an entry in @list_deny | |
213 | * - Otherwise, access will be granted | |
214 | */ | |
215 | ||
216 | if (u->ipv4_deny_map_fd >= 0) { | |
217 | r = add_lookup_instructions(p, u->ipv4_deny_map_fd, ETH_P_IP, is_ingress, ACCESS_DENIED); | |
218 | if (r < 0) | |
219 | return r; | |
220 | } | |
221 | ||
222 | if (u->ipv6_deny_map_fd >= 0) { | |
223 | r = add_lookup_instructions(p, u->ipv6_deny_map_fd, ETH_P_IPV6, is_ingress, ACCESS_DENIED); | |
224 | if (r < 0) | |
225 | return r; | |
226 | } | |
227 | ||
228 | if (u->ipv4_allow_map_fd >= 0) { | |
229 | r = add_lookup_instructions(p, u->ipv4_allow_map_fd, ETH_P_IP, is_ingress, ACCESS_ALLOWED); | |
230 | if (r < 0) | |
231 | return r; | |
232 | } | |
233 | ||
234 | if (u->ipv6_allow_map_fd >= 0) { | |
235 | r = add_lookup_instructions(p, u->ipv6_allow_map_fd, ETH_P_IPV6, is_ingress, ACCESS_ALLOWED); | |
236 | if (r < 0) | |
237 | return r; | |
238 | } | |
239 | } | |
240 | ||
241 | r = bpf_program_add_instructions(p, post_insn, ELEMENTSOF(post_insn)); | |
242 | if (r < 0) | |
243 | return r; | |
244 | ||
245 | if (accounting_map_fd >= 0) { | |
246 | struct bpf_insn insn[] = { | |
247 | /* | |
248 | * If R0 == 0, the packet will be denied; skip the accounting instructions in this case. | |
249 | * The jump label will be fixed up later. | |
250 | */ | |
251 | BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 0), | |
252 | ||
253 | /* Count packets */ | |
254 | BPF_MOV64_IMM(BPF_REG_0, MAP_KEY_PACKETS), /* r0 = 0 */ | |
255 | BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_0, -4), /* *(u32 *)(fp - 4) = r0 */ | |
256 | BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), | |
257 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), /* r2 = fp - 4 */ | |
258 | BPF_LD_MAP_FD(BPF_REG_1, accounting_map_fd), /* load map fd to r1 */ | |
259 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
260 | BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2), | |
261 | BPF_MOV64_IMM(BPF_REG_1, 1), /* r1 = 1 */ | |
262 | BPF_RAW_INSN(BPF_STX | BPF_XADD | BPF_DW, BPF_REG_0, BPF_REG_1, 0, 0), /* xadd r0 += r1 */ | |
263 | ||
264 | /* Count bytes */ | |
265 | BPF_MOV64_IMM(BPF_REG_0, MAP_KEY_BYTES), /* r0 = 1 */ | |
266 | BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_0, -4), /* *(u32 *)(fp - 4) = r0 */ | |
267 | BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), | |
268 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), /* r2 = fp - 4 */ | |
269 | BPF_LD_MAP_FD(BPF_REG_1, accounting_map_fd), | |
270 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
271 | BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2), | |
272 | BPF_LDX_MEM(BPF_W, BPF_REG_1, BPF_REG_6, offsetof(struct __sk_buff, len)), /* r1 = skb->len */ | |
273 | BPF_RAW_INSN(BPF_STX | BPF_XADD | BPF_DW, BPF_REG_0, BPF_REG_1, 0, 0), /* xadd r0 += r1 */ | |
274 | ||
275 | /* Allow the packet to pass */ | |
276 | BPF_MOV64_IMM(BPF_REG_0, 1), | |
277 | }; | |
278 | ||
279 | /* Jump label fixup */ | |
280 | insn[0].off = ELEMENTSOF(insn) - 1; | |
281 | ||
282 | r = bpf_program_add_instructions(p, insn, ELEMENTSOF(insn)); | |
283 | if (r < 0) | |
284 | return r; | |
285 | } | |
286 | ||
287 | do { | |
288 | /* | |
289 | * Exit from the eBPF program, R0 contains the verdict. | |
290 | * 0 means the packet is denied, 1 means the packet may pass. | |
291 | */ | |
292 | struct bpf_insn insn[] = { | |
293 | BPF_EXIT_INSN() | |
294 | }; | |
295 | ||
296 | r = bpf_program_add_instructions(p, insn, ELEMENTSOF(insn)); | |
297 | if (r < 0) | |
298 | return r; | |
299 | } while (false); | |
300 | ||
1cc6c93a | 301 | *ret = TAKE_PTR(p); |
1988a9d1 DM |
302 | |
303 | return 0; | |
304 | } | |
305 | ||
306 | static int bpf_firewall_count_access_items(IPAddressAccessItem *list, size_t *n_ipv4, size_t *n_ipv6) { | |
307 | IPAddressAccessItem *a; | |
308 | ||
309 | assert(n_ipv4); | |
310 | assert(n_ipv6); | |
311 | ||
312 | LIST_FOREACH(items, a, list) { | |
313 | switch (a->family) { | |
314 | ||
315 | case AF_INET: | |
316 | (*n_ipv4)++; | |
317 | break; | |
318 | ||
319 | case AF_INET6: | |
320 | (*n_ipv6)++; | |
321 | break; | |
322 | ||
323 | default: | |
324 | return -EAFNOSUPPORT; | |
325 | } | |
326 | } | |
327 | ||
328 | return 0; | |
329 | } | |
330 | ||
331 | static int bpf_firewall_add_access_items( | |
332 | IPAddressAccessItem *list, | |
333 | int ipv4_map_fd, | |
334 | int ipv6_map_fd, | |
335 | int verdict) { | |
336 | ||
337 | struct bpf_lpm_trie_key *key_ipv4, *key_ipv6; | |
338 | uint64_t value = verdict; | |
339 | IPAddressAccessItem *a; | |
340 | int r; | |
341 | ||
342 | key_ipv4 = alloca0(offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint32_t)); | |
343 | key_ipv6 = alloca0(offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint32_t) * 4); | |
344 | ||
345 | LIST_FOREACH(items, a, list) { | |
346 | switch (a->family) { | |
347 | ||
348 | case AF_INET: | |
349 | key_ipv4->prefixlen = a->prefixlen; | |
350 | memcpy(key_ipv4->data, &a->address, sizeof(uint32_t)); | |
351 | ||
352 | r = bpf_map_update_element(ipv4_map_fd, key_ipv4, &value); | |
353 | if (r < 0) | |
354 | return r; | |
355 | ||
356 | break; | |
357 | ||
358 | case AF_INET6: | |
359 | key_ipv6->prefixlen = a->prefixlen; | |
360 | memcpy(key_ipv6->data, &a->address, 4 * sizeof(uint32_t)); | |
361 | ||
362 | r = bpf_map_update_element(ipv6_map_fd, key_ipv6, &value); | |
363 | if (r < 0) | |
364 | return r; | |
365 | ||
366 | break; | |
367 | ||
368 | default: | |
369 | return -EAFNOSUPPORT; | |
370 | } | |
371 | } | |
372 | ||
373 | return 0; | |
374 | } | |
375 | ||
376 | static int bpf_firewall_prepare_access_maps( | |
377 | Unit *u, | |
378 | int verdict, | |
379 | int *ret_ipv4_map_fd, | |
380 | int *ret_ipv6_map_fd) { | |
381 | ||
382 | _cleanup_close_ int ipv4_map_fd = -1, ipv6_map_fd = -1; | |
383 | size_t n_ipv4 = 0, n_ipv6 = 0; | |
384 | Unit *p; | |
385 | int r; | |
386 | ||
387 | assert(ret_ipv4_map_fd); | |
388 | assert(ret_ipv6_map_fd); | |
389 | ||
390 | for (p = u; p; p = UNIT_DEREF(p->slice)) { | |
391 | CGroupContext *cc; | |
392 | ||
393 | cc = unit_get_cgroup_context(p); | |
394 | if (!cc) | |
395 | continue; | |
396 | ||
397 | bpf_firewall_count_access_items(verdict == ACCESS_ALLOWED ? cc->ip_address_allow : cc->ip_address_deny, &n_ipv4, &n_ipv6); | |
398 | } | |
399 | ||
400 | if (n_ipv4 > 0) { | |
401 | ipv4_map_fd = bpf_map_new( | |
402 | BPF_MAP_TYPE_LPM_TRIE, | |
403 | offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint32_t), | |
404 | sizeof(uint64_t), | |
405 | n_ipv4, | |
406 | BPF_F_NO_PREALLOC); | |
407 | if (ipv4_map_fd < 0) | |
408 | return ipv4_map_fd; | |
409 | } | |
410 | ||
411 | if (n_ipv6 > 0) { | |
412 | ipv6_map_fd = bpf_map_new( | |
413 | BPF_MAP_TYPE_LPM_TRIE, | |
414 | offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint32_t)*4, | |
415 | sizeof(uint64_t), | |
416 | n_ipv6, | |
417 | BPF_F_NO_PREALLOC); | |
418 | if (ipv6_map_fd < 0) | |
419 | return ipv6_map_fd; | |
420 | } | |
421 | ||
422 | for (p = u; p; p = UNIT_DEREF(p->slice)) { | |
423 | CGroupContext *cc; | |
424 | ||
425 | cc = unit_get_cgroup_context(p); | |
426 | if (!cc) | |
427 | continue; | |
428 | ||
429 | r = bpf_firewall_add_access_items(verdict == ACCESS_ALLOWED ? cc->ip_address_allow : cc->ip_address_deny, | |
430 | ipv4_map_fd, ipv6_map_fd, verdict); | |
431 | if (r < 0) | |
432 | return r; | |
433 | } | |
434 | ||
435 | *ret_ipv4_map_fd = ipv4_map_fd; | |
436 | *ret_ipv6_map_fd = ipv6_map_fd; | |
437 | ||
438 | ipv4_map_fd = ipv6_map_fd = -1; | |
439 | return 0; | |
440 | } | |
441 | ||
51283461 | 442 | static int bpf_firewall_prepare_accounting_maps(Unit *u, bool enabled, int *fd_ingress, int *fd_egress) { |
1988a9d1 DM |
443 | int r; |
444 | ||
51283461 | 445 | assert(u); |
1988a9d1 DM |
446 | assert(fd_ingress); |
447 | assert(fd_egress); | |
448 | ||
449 | if (enabled) { | |
450 | if (*fd_ingress < 0) { | |
451 | r = bpf_map_new(BPF_MAP_TYPE_ARRAY, sizeof(int), sizeof(uint64_t), 2, 0); | |
452 | if (r < 0) | |
453 | return r; | |
454 | ||
455 | *fd_ingress = r; | |
456 | } | |
457 | ||
458 | if (*fd_egress < 0) { | |
459 | ||
460 | r = bpf_map_new(BPF_MAP_TYPE_ARRAY, sizeof(int), sizeof(uint64_t), 2, 0); | |
461 | if (r < 0) | |
462 | return r; | |
463 | ||
464 | *fd_egress = r; | |
465 | } | |
51283461 | 466 | |
1988a9d1 DM |
467 | } else { |
468 | *fd_ingress = safe_close(*fd_ingress); | |
469 | *fd_egress = safe_close(*fd_egress); | |
51283461 LP |
470 | |
471 | zero(u->ip_accounting_extra); | |
1988a9d1 DM |
472 | } |
473 | ||
474 | return 0; | |
475 | } | |
476 | ||
477 | int bpf_firewall_compile(Unit *u) { | |
478 | CGroupContext *cc; | |
acf7f253 | 479 | int r, supported; |
1988a9d1 DM |
480 | |
481 | assert(u); | |
482 | ||
51283461 LP |
483 | cc = unit_get_cgroup_context(u); |
484 | if (!cc) | |
485 | return -EINVAL; | |
486 | ||
acf7f253 LP |
487 | supported = bpf_firewall_supported(); |
488 | if (supported < 0) | |
489 | return supported; | |
490 | if (supported == BPF_FIREWALL_UNSUPPORTED) { | |
418cdd69 | 491 | log_debug("BPF firewalling not supported on this manager, proceeding without."); |
1988a9d1 DM |
492 | return -EOPNOTSUPP; |
493 | } | |
acf7f253 LP |
494 | if (supported != BPF_FIREWALL_SUPPORTED_WITH_MULTI && u->type == UNIT_SLICE) { |
495 | /* If BPF_F_ALLOW_MULTI is not supported we don't support any BPF magic on inner nodes (i.e. on slice | |
496 | * units), since that would mean leaf nodes couldn't do any BPF anymore at all. Under the assumption | |
497 | * that BPF is more interesting on leaf nodes we hence avoid it on inner nodes in that case. This is | |
498 | * consistent with old systemd behaviour from before v238, where BPF wasn't supported in inner nodes at | |
499 | * all, either. */ | |
500 | log_debug("BPF_F_ALLOW_MULTI is not supported on this manager, not doing BPF firewall on slice units."); | |
501 | return -EOPNOTSUPP; | |
502 | } | |
1988a9d1 DM |
503 | |
504 | /* Note that when we compile a new firewall we first flush out the access maps and the BPF programs themselves, | |
505 | * but we reuse the the accounting maps. That way the firewall in effect always maps to the actual | |
506 | * configuration, but we don't flush out the accounting unnecessarily */ | |
507 | ||
508 | u->ip_bpf_ingress = bpf_program_unref(u->ip_bpf_ingress); | |
509 | u->ip_bpf_egress = bpf_program_unref(u->ip_bpf_egress); | |
510 | ||
511 | u->ipv4_allow_map_fd = safe_close(u->ipv4_allow_map_fd); | |
512 | u->ipv4_deny_map_fd = safe_close(u->ipv4_deny_map_fd); | |
513 | ||
514 | u->ipv6_allow_map_fd = safe_close(u->ipv6_allow_map_fd); | |
515 | u->ipv6_deny_map_fd = safe_close(u->ipv6_deny_map_fd); | |
516 | ||
acf7f253 LP |
517 | if (u->type != UNIT_SLICE) { |
518 | /* In inner nodes we only do accounting, we do not actually bother with access control. However, leaf | |
519 | * nodes will incorporate all IP access rules set on all their parent nodes. This has the benefit that | |
520 | * they can optionally cancel out system-wide rules. Since inner nodes can't contain processes this | |
521 | * means that all configure IP access rules *will* take effect on processes, even though we never | |
522 | * compile them for inner nodes. */ | |
1988a9d1 | 523 | |
acf7f253 LP |
524 | r = bpf_firewall_prepare_access_maps(u, ACCESS_ALLOWED, &u->ipv4_allow_map_fd, &u->ipv6_allow_map_fd); |
525 | if (r < 0) | |
526 | return log_error_errno(r, "Preparation of eBPF allow maps failed: %m"); | |
527 | ||
528 | r = bpf_firewall_prepare_access_maps(u, ACCESS_DENIED, &u->ipv4_deny_map_fd, &u->ipv6_deny_map_fd); | |
529 | if (r < 0) | |
530 | return log_error_errno(r, "Preparation of eBPF deny maps failed: %m"); | |
531 | } | |
1988a9d1 | 532 | |
51283461 | 533 | r = bpf_firewall_prepare_accounting_maps(u, cc->ip_accounting, &u->ip_accounting_ingress_map_fd, &u->ip_accounting_egress_map_fd); |
1988a9d1 DM |
534 | if (r < 0) |
535 | return log_error_errno(r, "Preparation of eBPF accounting maps failed: %m"); | |
536 | ||
537 | r = bpf_firewall_compile_bpf(u, true, &u->ip_bpf_ingress); | |
538 | if (r < 0) | |
539 | return log_error_errno(r, "Compilation for ingress BPF program failed: %m"); | |
540 | ||
541 | r = bpf_firewall_compile_bpf(u, false, &u->ip_bpf_egress); | |
542 | if (r < 0) | |
543 | return log_error_errno(r, "Compilation for egress BPF program failed: %m"); | |
544 | ||
545 | return 0; | |
546 | } | |
547 | ||
548 | int bpf_firewall_install(Unit *u) { | |
549 | _cleanup_free_ char *path = NULL; | |
9f2e6892 | 550 | CGroupContext *cc; |
acf7f253 | 551 | int r, supported; |
aa2b6f1d | 552 | uint32_t flags; |
1988a9d1 DM |
553 | |
554 | assert(u); | |
555 | ||
9f2e6892 LP |
556 | cc = unit_get_cgroup_context(u); |
557 | if (!cc) | |
558 | return -EINVAL; | |
aa2b6f1d LP |
559 | if (!u->cgroup_path) |
560 | return -EINVAL; | |
561 | if (!u->cgroup_realized) | |
562 | return -EINVAL; | |
9f2e6892 | 563 | |
acf7f253 LP |
564 | supported = bpf_firewall_supported(); |
565 | if (supported < 0) | |
566 | return supported; | |
567 | if (supported == BPF_FIREWALL_UNSUPPORTED) { | |
418cdd69 | 568 | log_debug("BPF firewalling not supported on this manager, proceeding without."); |
1988a9d1 DM |
569 | return -EOPNOTSUPP; |
570 | } | |
acf7f253 LP |
571 | if (supported != BPF_FIREWALL_SUPPORTED_WITH_MULTI && u->type == UNIT_SLICE) { |
572 | log_debug("BPF_F_ALLOW_MULTI is not supported on this manager, not doing BPF firewall on slice units."); | |
573 | return -EOPNOTSUPP; | |
574 | } | |
1988a9d1 DM |
575 | |
576 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, NULL, &path); | |
577 | if (r < 0) | |
578 | return log_error_errno(r, "Failed to determine cgroup path: %m"); | |
579 | ||
acf7f253 LP |
580 | flags = (supported == BPF_FIREWALL_SUPPORTED_WITH_MULTI && |
581 | (u->type == UNIT_SLICE || unit_cgroup_delegate(u))) ? BPF_F_ALLOW_MULTI : 0; | |
582 | ||
aa2b6f1d LP |
583 | /* Unref the old BPF program (which will implicitly detach it) right before attaching the new program, to |
584 | * minimize the time window when we don't account for IP traffic. */ | |
585 | u->ip_bpf_egress_installed = bpf_program_unref(u->ip_bpf_egress_installed); | |
586 | u->ip_bpf_ingress_installed = bpf_program_unref(u->ip_bpf_ingress_installed); | |
1988a9d1 | 587 | |
aa2b6f1d | 588 | if (u->ip_bpf_egress) { |
acf7f253 | 589 | r = bpf_program_cgroup_attach(u->ip_bpf_egress, BPF_CGROUP_INET_EGRESS, path, flags); |
1988a9d1 DM |
590 | if (r < 0) |
591 | return log_error_errno(r, "Attaching egress BPF program to cgroup %s failed: %m", path); | |
aa2b6f1d LP |
592 | |
593 | /* Remember that this BPF program is installed now. */ | |
594 | u->ip_bpf_egress_installed = bpf_program_ref(u->ip_bpf_egress); | |
1988a9d1 DM |
595 | } |
596 | ||
597 | if (u->ip_bpf_ingress) { | |
acf7f253 | 598 | r = bpf_program_cgroup_attach(u->ip_bpf_ingress, BPF_CGROUP_INET_INGRESS, path, flags); |
1988a9d1 DM |
599 | if (r < 0) |
600 | return log_error_errno(r, "Attaching ingress BPF program to cgroup %s failed: %m", path); | |
aa2b6f1d LP |
601 | |
602 | u->ip_bpf_ingress_installed = bpf_program_ref(u->ip_bpf_ingress); | |
1988a9d1 DM |
603 | } |
604 | ||
605 | return 0; | |
606 | } | |
607 | ||
608 | int bpf_firewall_read_accounting(int map_fd, uint64_t *ret_bytes, uint64_t *ret_packets) { | |
609 | uint64_t key, packets; | |
610 | int r; | |
611 | ||
612 | if (map_fd < 0) | |
613 | return -EBADF; | |
614 | ||
615 | if (ret_packets) { | |
616 | key = MAP_KEY_PACKETS; | |
617 | r = bpf_map_lookup_element(map_fd, &key, &packets); | |
618 | if (r < 0) | |
619 | return r; | |
620 | } | |
621 | ||
622 | if (ret_bytes) { | |
623 | key = MAP_KEY_BYTES; | |
624 | r = bpf_map_lookup_element(map_fd, &key, ret_bytes); | |
625 | if (r < 0) | |
626 | return r; | |
627 | } | |
628 | ||
629 | if (ret_packets) | |
630 | *ret_packets = packets; | |
631 | ||
632 | return 0; | |
633 | } | |
634 | ||
635 | int bpf_firewall_reset_accounting(int map_fd) { | |
636 | uint64_t key, value = 0; | |
637 | int r; | |
638 | ||
639 | if (map_fd < 0) | |
640 | return -EBADF; | |
641 | ||
642 | key = MAP_KEY_PACKETS; | |
643 | r = bpf_map_update_element(map_fd, &key, &value); | |
644 | if (r < 0) | |
645 | return r; | |
646 | ||
647 | key = MAP_KEY_BYTES; | |
648 | return bpf_map_update_element(map_fd, &key, &value); | |
649 | } | |
650 | ||
1988a9d1 | 651 | int bpf_firewall_supported(void) { |
93e93da5 LP |
652 | struct bpf_insn trivial[] = { |
653 | BPF_MOV64_IMM(BPF_REG_0, 1), | |
654 | BPF_EXIT_INSN() | |
655 | }; | |
656 | ||
657 | _cleanup_(bpf_program_unrefp) BPFProgram *program = NULL; | |
1988a9d1 | 658 | static int supported = -1; |
e583759b | 659 | union bpf_attr attr; |
1988a9d1 DM |
660 | int fd, r; |
661 | ||
e583759b | 662 | /* Checks whether BPF firewalling is supported. For this, we check five things: |
1988a9d1 DM |
663 | * |
664 | * a) whether we are privileged | |
665 | * b) whether the unified hierarchy is being used | |
666 | * c) the BPF implementation in the kernel supports BPF LPM TRIE maps, which we require | |
e583759b LP |
667 | * d) the BPF implementation in the kernel supports BPF_PROG_TYPE_CGROUP_SKB programs, which we require |
668 | * e) the BPF implementation in the kernel supports the BPF_PROG_ATTACH call, which we require | |
1988a9d1 DM |
669 | * |
670 | */ | |
671 | ||
672 | if (supported >= 0) | |
673 | return supported; | |
674 | ||
93e93da5 LP |
675 | if (geteuid() != 0) { |
676 | log_debug("Not enough privileges, BPF firewalling is not supported."); | |
2ae7ee58 | 677 | return supported = BPF_FIREWALL_UNSUPPORTED; |
93e93da5 | 678 | } |
1988a9d1 DM |
679 | |
680 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
681 | if (r < 0) | |
682 | return log_error_errno(r, "Can't determine whether the unified hierarchy is used: %m"); | |
e583759b LP |
683 | if (r == 0) { |
684 | log_debug("Not running with unified cgroups, BPF firewalling is not supported."); | |
2ae7ee58 | 685 | return supported = BPF_FIREWALL_UNSUPPORTED; |
e583759b | 686 | } |
1988a9d1 DM |
687 | |
688 | fd = bpf_map_new(BPF_MAP_TYPE_LPM_TRIE, | |
689 | offsetof(struct bpf_lpm_trie_key, data) + sizeof(uint64_t), | |
690 | sizeof(uint64_t), | |
691 | 1, | |
692 | BPF_F_NO_PREALLOC); | |
693 | if (fd < 0) { | |
694 | log_debug_errno(r, "Can't allocate BPF LPM TRIE map, BPF firewalling is not supported: %m"); | |
2ae7ee58 | 695 | return supported = BPF_FIREWALL_UNSUPPORTED; |
1988a9d1 DM |
696 | } |
697 | ||
698 | safe_close(fd); | |
699 | ||
93e93da5 LP |
700 | if (bpf_program_new(BPF_PROG_TYPE_CGROUP_SKB, &program) < 0) { |
701 | log_debug_errno(r, "Can't allocate CGROUP SKB BPF program, BPF firewalling is not supported: %m"); | |
2ae7ee58 | 702 | return supported = BPF_FIREWALL_UNSUPPORTED; |
93e93da5 LP |
703 | } |
704 | ||
705 | r = bpf_program_add_instructions(program, trivial, ELEMENTSOF(trivial)); | |
706 | if (r < 0) { | |
707 | log_debug_errno(r, "Can't add trivial instructions to CGROUP SKB BPF program, BPF firewalling is not supported: %m"); | |
2ae7ee58 | 708 | return supported = BPF_FIREWALL_UNSUPPORTED; |
93e93da5 LP |
709 | } |
710 | ||
711 | r = bpf_program_load_kernel(program, NULL, 0); | |
712 | if (r < 0) { | |
713 | log_debug_errno(r, "Can't load kernel CGROUP SKB BPF program, BPF firewalling is not supported: %m"); | |
2ae7ee58 | 714 | return supported = BPF_FIREWALL_UNSUPPORTED; |
93e93da5 LP |
715 | } |
716 | ||
e583759b LP |
717 | /* Unfortunately the kernel allows us to create BPF_PROG_TYPE_CGROUP_SKB programs even when CONFIG_CGROUP_BPF |
718 | * is turned off at kernel compilation time. This sucks of course: why does it allow us to create a cgroup BPF | |
719 | * program if we can't do a thing with it later? | |
720 | * | |
721 | * We detect this case by issuing the BPF_PROG_ATTACH bpf() call with invalid file descriptors: if | |
722 | * CONFIG_CGROUP_BPF is turned off, then the call will fail early with EINVAL. If it is turned on the | |
723 | * parameters are validated however, and that'll fail with EBADF then. */ | |
724 | ||
725 | attr = (union bpf_attr) { | |
726 | .attach_type = BPF_CGROUP_INET_EGRESS, | |
727 | .target_fd = -1, | |
728 | .attach_bpf_fd = -1, | |
729 | }; | |
730 | ||
731 | r = bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); | |
732 | if (r < 0) { | |
2ae7ee58 LP |
733 | if (errno != EBADF) { |
734 | log_debug_errno(errno, "Didn't get EBADF from BPF_PROG_ATTACH, BPF firewalling is not supported: %m"); | |
735 | return supported = BPF_FIREWALL_UNSUPPORTED; | |
736 | } | |
737 | ||
738 | /* YAY! */ | |
739 | } else { | |
740 | log_debug("Wut? Kernel accepted our invalid BPF_PROG_ATTACH call? Something is weird, assuming BPF firewalling is broken and hence not supported."); | |
741 | return supported = BPF_FIREWALL_UNSUPPORTED; | |
742 | } | |
e583759b | 743 | |
2ae7ee58 LP |
744 | /* So now we know that the BPF program is generally available, let's see if BPF_F_ALLOW_MULTI is also supported |
745 | * (which was added in kernel 4.15). We use a similar logic as before, but this time we use | |
746 | * BPF_F_ALLOW_MULTI. Since the flags are checked early in the system call we'll get EINVAL if it's not | |
747 | * supported, and EBADF as before if it is available. */ | |
e583759b | 748 | |
2ae7ee58 LP |
749 | attr = (union bpf_attr) { |
750 | .attach_type = BPF_CGROUP_INET_EGRESS, | |
751 | .target_fd = -1, | |
752 | .attach_bpf_fd = -1, | |
753 | .attach_flags = BPF_F_ALLOW_MULTI, | |
754 | }; | |
755 | ||
756 | r = bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); | |
757 | if (r < 0) { | |
758 | if (errno == EBADF) { | |
759 | log_debug_errno(errno, "Got EBADF when using BPF_F_ALLOW_MULTI, which indicates it is supported. Yay!"); | |
760 | return supported = BPF_FIREWALL_SUPPORTED_WITH_MULTI; | |
761 | } | |
762 | ||
763 | if (errno == EINVAL) | |
764 | log_debug_errno(errno, "Got EINVAL error when using BPF_F_ALLOW_MULTI, which indicates it's not supported."); | |
765 | else | |
766 | log_debug_errno(errno, "Got unexpected error when using BPF_F_ALLOW_MULTI, assuming it's not supported: %m"); | |
767 | ||
768 | return supported = BPF_FIREWALL_SUPPORTED; | |
769 | } else { | |
770 | log_debug("Wut? Kernel accepted our invalid BPF_PROG_ATTACH+BPF_F_ALLOW_MULTI call? Something is weird, assuming BPF firewalling is broken and hence not supported."); | |
771 | return supported = BPF_FIREWALL_UNSUPPORTED; | |
772 | } | |
1988a9d1 | 773 | } |