]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
f9ef2270 | 2 | * Copyright (c) 2009, 2010, 2012, 2013, 2014, 2015 Nicira, Inc. |
064af421 | 3 | * |
a14bc59f BP |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
064af421 | 7 | * |
a14bc59f BP |
8 | * http://www.apache.org/licenses/LICENSE-2.0 |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
064af421 BP |
15 | */ |
16 | ||
17 | #include <config.h> | |
2c78a3e6 | 18 | #include "pcap-file.h" |
064af421 BP |
19 | #include <errno.h> |
20 | #include <inttypes.h> | |
b6cdfbb4 | 21 | #include <stdlib.h> |
064af421 | 22 | #include <string.h> |
b6cdfbb4 | 23 | #include <sys/stat.h> |
4cf5406b | 24 | #include "byte-order.h" |
064af421 | 25 | #include "compiler.h" |
cf62fa4c | 26 | #include "dp-packet.h" |
f3dd1419 | 27 | #include "flow.h" |
ee89ea7b | 28 | #include "openvswitch/hmap.h" |
f3dd1419 | 29 | #include "packets.h" |
a797eab3 | 30 | #include "timeval.h" |
f3dd1419 | 31 | #include "unaligned.h" |
ee89ea7b | 32 | #include "util.h" |
e6211adc | 33 | #include "openvswitch/vlog.h" |
064af421 | 34 | |
d98e6007 | 35 | VLOG_DEFINE_THIS_MODULE(pcap); |
5136ce49 | 36 | |
b6e840ae MM |
37 | enum ts_resolution { |
38 | PCAP_USEC, | |
39 | PCAP_NSEC, | |
40 | }; | |
41 | ||
597177a2 | 42 | enum network_type { |
7f8d02a9 | 43 | PCAP_ETHERNET = 1, |
597177a2 BP |
44 | PCAP_LINUX_SLL = 0x71 |
45 | }; | |
46 | ||
b6e840ae MM |
47 | struct pcap_file { |
48 | FILE *file; | |
49 | enum ts_resolution resolution; | |
597177a2 | 50 | enum network_type network; |
b6e840ae MM |
51 | }; |
52 | ||
064af421 BP |
53 | struct pcap_hdr { |
54 | uint32_t magic_number; /* magic number */ | |
55 | uint16_t version_major; /* major version number */ | |
56 | uint16_t version_minor; /* minor version number */ | |
57 | int32_t thiszone; /* GMT to local correction */ | |
58 | uint32_t sigfigs; /* accuracy of timestamps */ | |
59 | uint32_t snaplen; /* max length of captured packets */ | |
60 | uint32_t network; /* data link type */ | |
381328fe BP |
61 | }; |
62 | BUILD_ASSERT_DECL(sizeof(struct pcap_hdr) == 24); | |
064af421 BP |
63 | |
64 | struct pcaprec_hdr { | |
65 | uint32_t ts_sec; /* timestamp seconds */ | |
b6e840ae | 66 | uint32_t ts_subsec; /* timestamp subseconds */ |
064af421 BP |
67 | uint32_t incl_len; /* number of octets of packet saved in file */ |
68 | uint32_t orig_len; /* actual length of packet */ | |
381328fe BP |
69 | }; |
70 | BUILD_ASSERT_DECL(sizeof(struct pcaprec_hdr) == 16); | |
064af421 | 71 | |
b6e840ae | 72 | struct pcap_file * |
50aa0364 | 73 | ovs_pcap_open(const char *file_name, const char *mode) |
064af421 | 74 | { |
b6cdfbb4 | 75 | struct stat s; |
b6e840ae | 76 | struct pcap_file *p_file; |
b6cdfbb4 | 77 | int error; |
064af421 | 78 | |
b6cdfbb4 BP |
79 | ovs_assert(!strcmp(mode, "rb") || |
80 | !strcmp(mode, "wb") || | |
81 | !strcmp(mode, "ab")); | |
064af421 | 82 | |
b6e840ae MM |
83 | p_file = xmalloc(sizeof *p_file); |
84 | p_file->file = fopen(file_name, mode); | |
85 | p_file->resolution = PCAP_USEC; | |
86 | if (p_file->file == NULL) { | |
b6cdfbb4 BP |
87 | VLOG_WARN("%s: failed to open pcap file for %s (%s)", file_name, |
88 | (mode[0] == 'r' ? "reading" | |
89 | : mode[0] == 'w' ? "writing" | |
90 | : "appending"), | |
91 | ovs_strerror(errno)); | |
064af421 BP |
92 | return NULL; |
93 | } | |
94 | ||
b6cdfbb4 BP |
95 | switch (mode[0]) { |
96 | case 'r': | |
b6e840ae | 97 | error = ovs_pcap_read_header(p_file); |
b6cdfbb4 BP |
98 | if (error) { |
99 | errno = error; | |
b6e840ae | 100 | ovs_pcap_close(p_file); |
064af421 BP |
101 | return NULL; |
102 | } | |
b6cdfbb4 BP |
103 | break; |
104 | ||
105 | case 'w': | |
b6e840ae | 106 | ovs_pcap_write_header(p_file); |
b6cdfbb4 BP |
107 | break; |
108 | ||
109 | case 'a': | |
b6e840ae MM |
110 | if (!fstat(fileno(p_file->file), &s) && !s.st_size) { |
111 | ovs_pcap_write_header(p_file); | |
b6cdfbb4 BP |
112 | } |
113 | break; | |
114 | ||
115 | default: | |
428b2edd | 116 | OVS_NOT_REACHED(); |
064af421 | 117 | } |
b6e840ae MM |
118 | |
119 | return p_file; | |
120 | } | |
121 | ||
122 | struct pcap_file * | |
123 | ovs_pcap_stdout(void) | |
124 | { | |
125 | struct pcap_file *p_file = xmalloc(sizeof *p_file); | |
126 | p_file->file = stdout; | |
127 | return p_file; | |
064af421 BP |
128 | } |
129 | ||
130 | int | |
b6e840ae | 131 | ovs_pcap_read_header(struct pcap_file *p_file) |
064af421 BP |
132 | { |
133 | struct pcap_hdr ph; | |
b6e840ae MM |
134 | if (fread(&ph, sizeof ph, 1, p_file->file) != 1) { |
135 | int error = ferror(p_file->file) ? errno : EOF; | |
c18ea70d | 136 | VLOG_WARN("failed to read pcap header: %s", ovs_retval_to_string(error)); |
064af421 BP |
137 | return error; |
138 | } | |
597177a2 | 139 | bool byte_swap; |
b6e840ae | 140 | if (ph.magic_number == 0xa1b2c3d4 || ph.magic_number == 0xd4c3b2a1) { |
597177a2 | 141 | byte_swap = ph.magic_number == 0xd4c3b2a1; |
b6e840ae MM |
142 | p_file->resolution = PCAP_USEC; |
143 | } else if (ph.magic_number == 0xa1b23c4d || | |
144 | ph.magic_number == 0x4d3cb2a1) { | |
597177a2 | 145 | byte_swap = ph.magic_number == 0x4d3cb2a1; |
b6e840ae MM |
146 | p_file->resolution = PCAP_NSEC; |
147 | } else { | |
064af421 | 148 | VLOG_WARN("bad magic 0x%08"PRIx32" reading pcap file " |
b6e840ae MM |
149 | "(expected 0xa1b2c3d4, 0xa1b23c4d, 0xd4c3b2a1, " |
150 | "or 0x4d3cb2a1)", ph.magic_number); | |
064af421 BP |
151 | return EPROTO; |
152 | } | |
597177a2 BP |
153 | p_file->network = byte_swap ? uint32_byteswap(ph.network) : ph.network; |
154 | if (p_file->network != PCAP_ETHERNET && | |
155 | p_file->network != PCAP_LINUX_SLL) { | |
9a609f63 BP |
156 | VLOG_WARN("unknown network type %u reading pcap file", |
157 | (unsigned int) p_file->network); | |
597177a2 BP |
158 | return EPROTO; |
159 | } | |
064af421 BP |
160 | return 0; |
161 | } | |
162 | ||
163 | void | |
b6e840ae | 164 | ovs_pcap_write_header(struct pcap_file *p_file) |
064af421 BP |
165 | { |
166 | /* The pcap reader is responsible for figuring out endianness based on the | |
167 | * magic number, so the lack of htonX calls here is intentional. */ | |
168 | struct pcap_hdr ph; | |
169 | ph.magic_number = 0xa1b2c3d4; | |
170 | ph.version_major = 2; | |
171 | ph.version_minor = 4; | |
172 | ph.thiszone = 0; | |
173 | ph.sigfigs = 0; | |
174 | ph.snaplen = 1518; | |
175 | ph.network = 1; /* Ethernet */ | |
b6e840ae MM |
176 | ignore(fwrite(&ph, sizeof ph, 1, p_file->file)); |
177 | fflush(p_file->file); | |
064af421 BP |
178 | } |
179 | ||
180 | int | |
b6e840ae MM |
181 | ovs_pcap_read(struct pcap_file *p_file, struct dp_packet **bufp, |
182 | long long int *when) | |
064af421 BP |
183 | { |
184 | struct pcaprec_hdr prh; | |
cf62fa4c | 185 | struct dp_packet *buf; |
064af421 BP |
186 | void *data; |
187 | size_t len; | |
a797eab3 | 188 | bool swap; |
064af421 BP |
189 | |
190 | *bufp = NULL; | |
191 | ||
192 | /* Read header. */ | |
b6e840ae MM |
193 | if (fread(&prh, sizeof prh, 1, p_file->file) != 1) { |
194 | if (ferror(p_file->file)) { | |
f6cdbd3e BP |
195 | int error = errno; |
196 | VLOG_WARN("failed to read pcap record header: %s", | |
197 | ovs_retval_to_string(error)); | |
198 | return error; | |
199 | } else { | |
200 | return EOF; | |
201 | } | |
064af421 BP |
202 | } |
203 | ||
204 | /* Calculate length. */ | |
205 | len = prh.incl_len; | |
a797eab3 BP |
206 | swap = len > 0xffff; |
207 | if (swap) { | |
208 | len = uint32_byteswap(len); | |
209 | if (len > 0xffff) { | |
2f7662a1 | 210 | VLOG_WARN("bad packet length %"PRIuSIZE" or %"PRIu32" " |
064af421 | 211 | "reading pcap file", |
a797eab3 | 212 | len, uint32_byteswap(len)); |
064af421 BP |
213 | return EPROTO; |
214 | } | |
a797eab3 BP |
215 | } |
216 | ||
217 | /* Calculate time. */ | |
218 | if (when) { | |
219 | uint32_t ts_sec = swap ? uint32_byteswap(prh.ts_sec) : prh.ts_sec; | |
b6e840ae MM |
220 | uint32_t ts_subsec = swap ? uint32_byteswap(prh.ts_subsec) |
221 | : prh.ts_subsec; | |
222 | ts_subsec = p_file->resolution == PCAP_USEC ? ts_subsec / 1000 | |
223 | : ts_subsec / 1000000; | |
224 | *when = ts_sec * 1000LL + ts_subsec; | |
064af421 BP |
225 | } |
226 | ||
2482b0b0 | 227 | /* Read packet. Packet type is Ethernet */ |
cf62fa4c PS |
228 | buf = dp_packet_new(len); |
229 | data = dp_packet_put_uninit(buf, len); | |
b6e840ae MM |
230 | if (fread(data, len, 1, p_file->file) != 1) { |
231 | int error = ferror(p_file->file) ? errno : EOF; | |
064af421 | 232 | VLOG_WARN("failed to read pcap packet: %s", |
c18ea70d | 233 | ovs_retval_to_string(error)); |
cf62fa4c | 234 | dp_packet_delete(buf); |
064af421 BP |
235 | return error; |
236 | } | |
597177a2 BP |
237 | |
238 | if (p_file->network == PCAP_LINUX_SLL) { | |
239 | /* This format doesn't include the destination Ethernet address, which | |
240 | * is weird. */ | |
241 | ||
242 | struct sll_header { | |
243 | ovs_be16 packet_type; | |
244 | ovs_be16 arp_hrd; | |
245 | ovs_be16 lla_len; | |
246 | struct eth_addr dl_src; | |
247 | ovs_be16 reserved; | |
248 | ovs_be16 protocol; | |
249 | }; | |
250 | const struct sll_header *sll; | |
251 | if (len < sizeof *sll) { | |
252 | VLOG_WARN("pcap packet too short for SLL header"); | |
253 | dp_packet_delete(buf); | |
254 | return EPROTO; | |
255 | } | |
256 | ||
257 | /* Pull Linux SLL header. */ | |
258 | sll = dp_packet_pull(buf, sizeof *sll); | |
259 | if (sll->lla_len != htons(6)) { | |
260 | ovs_hex_dump(stdout, sll, sizeof *sll, 0, false); | |
261 | VLOG_WARN("bad SLL header"); | |
262 | dp_packet_delete(buf); | |
263 | return EPROTO; | |
264 | } | |
265 | ||
266 | /* Push Ethernet header. */ | |
267 | struct eth_header eth = { | |
268 | /* eth_dst is all zeros because the format doesn't include it. */ | |
269 | .eth_src = sll->dl_src, | |
270 | .eth_type = sll->protocol, | |
271 | }; | |
272 | dp_packet_push(buf, ð, sizeof eth); | |
273 | } | |
274 | ||
064af421 BP |
275 | *bufp = buf; |
276 | return 0; | |
277 | } | |
278 | ||
279 | void | |
b6e840ae | 280 | ovs_pcap_write(struct pcap_file *p_file, struct dp_packet *buf) |
064af421 BP |
281 | { |
282 | struct pcaprec_hdr prh; | |
a797eab3 BP |
283 | struct timeval tv; |
284 | ||
940ac2ce | 285 | ovs_assert(dp_packet_is_eth(buf)); |
2482b0b0 | 286 | |
a797eab3 BP |
287 | xgettimeofday(&tv); |
288 | prh.ts_sec = tv.tv_sec; | |
b6e840ae | 289 | prh.ts_subsec = tv.tv_usec; |
cf62fa4c PS |
290 | prh.incl_len = dp_packet_size(buf); |
291 | prh.orig_len = dp_packet_size(buf); | |
b6e840ae MM |
292 | ignore(fwrite(&prh, sizeof prh, 1, p_file->file)); |
293 | ignore(fwrite(dp_packet_data(buf), dp_packet_size(buf), 1, p_file->file)); | |
294 | fflush(p_file->file); | |
295 | } | |
296 | ||
297 | void | |
298 | ovs_pcap_close(struct pcap_file *p_file) | |
299 | { | |
300 | if (p_file->file != stdout) { | |
301 | fclose(p_file->file); | |
302 | } | |
303 | free(p_file); | |
064af421 | 304 | } |
f3dd1419 BP |
305 | \f |
306 | struct tcp_key { | |
307 | ovs_be32 nw_src, nw_dst; | |
308 | ovs_be16 tp_src, tp_dst; | |
309 | }; | |
310 | ||
311 | struct tcp_stream { | |
312 | struct hmap_node hmap_node; | |
313 | struct tcp_key key; | |
314 | uint32_t seq_no; | |
cf62fa4c | 315 | struct dp_packet payload; |
f3dd1419 BP |
316 | }; |
317 | ||
318 | struct tcp_reader { | |
319 | struct hmap streams; | |
320 | }; | |
321 | ||
322 | static void | |
323 | tcp_stream_destroy(struct tcp_reader *r, struct tcp_stream *stream) | |
324 | { | |
325 | hmap_remove(&r->streams, &stream->hmap_node); | |
cf62fa4c | 326 | dp_packet_uninit(&stream->payload); |
f3dd1419 BP |
327 | free(stream); |
328 | } | |
329 | ||
330 | /* Returns a new data structure for extracting TCP stream data from an | |
331 | * Ethernet packet capture */ | |
332 | struct tcp_reader * | |
333 | tcp_reader_open(void) | |
334 | { | |
335 | struct tcp_reader *r; | |
336 | ||
337 | r = xmalloc(sizeof *r); | |
338 | hmap_init(&r->streams); | |
339 | return r; | |
340 | } | |
341 | ||
342 | /* Closes and frees 'r'. */ | |
343 | void | |
344 | tcp_reader_close(struct tcp_reader *r) | |
345 | { | |
346 | struct tcp_stream *stream, *next_stream; | |
347 | ||
348 | HMAP_FOR_EACH_SAFE (stream, next_stream, hmap_node, &r->streams) { | |
349 | tcp_stream_destroy(r, stream); | |
350 | } | |
351 | hmap_destroy(&r->streams); | |
352 | free(r); | |
353 | } | |
354 | ||
355 | static struct tcp_stream * | |
276e2864 BP |
356 | tcp_stream_lookup(struct tcp_reader *r, |
357 | const struct tcp_key *key, uint32_t hash) | |
f3dd1419 BP |
358 | { |
359 | struct tcp_stream *stream; | |
f3dd1419 BP |
360 | |
361 | HMAP_FOR_EACH_WITH_HASH (stream, hmap_node, hash, &r->streams) { | |
276e2864 | 362 | if (!memcmp(&stream->key, key, sizeof *key)) { |
f3dd1419 BP |
363 | return stream; |
364 | } | |
365 | } | |
276e2864 BP |
366 | return NULL; |
367 | } | |
368 | ||
369 | static struct tcp_stream * | |
370 | tcp_stream_new(struct tcp_reader *r, const struct tcp_key *key, uint32_t hash) | |
371 | { | |
372 | struct tcp_stream *stream; | |
f3dd1419 BP |
373 | |
374 | stream = xmalloc(sizeof *stream); | |
375 | hmap_insert(&r->streams, &stream->hmap_node, hash); | |
276e2864 | 376 | memcpy(&stream->key, key, sizeof *key); |
f3dd1419 | 377 | stream->seq_no = 0; |
cf62fa4c | 378 | dp_packet_init(&stream->payload, 2048); |
f3dd1419 BP |
379 | return stream; |
380 | } | |
381 | ||
382 | /* Processes 'packet' through TCP reader 'r'. The caller must have already | |
383 | * extracted the packet's headers into 'flow', using flow_extract(). | |
384 | * | |
385 | * If 'packet' is a TCP packet, then the reader attempts to reconstruct the | |
cf62fa4c PS |
386 | * data stream. If successful, it returns an dp_packet that represents the data |
387 | * stream so far. The caller may examine the data in the dp_packet and pull off | |
f3dd1419 BP |
388 | * any data that it has fully processed. The remaining data that the caller |
389 | * does not pull off will be presented again in future calls if more data | |
390 | * arrives in the stream. | |
391 | * | |
392 | * Returns null if 'packet' doesn't add new data to a TCP stream. */ | |
cf62fa4c | 393 | struct dp_packet * |
f3dd1419 | 394 | tcp_reader_run(struct tcp_reader *r, const struct flow *flow, |
cf62fa4c | 395 | const struct dp_packet *packet) |
f3dd1419 BP |
396 | { |
397 | struct tcp_stream *stream; | |
398 | struct tcp_header *tcp; | |
cf62fa4c | 399 | struct dp_packet *payload; |
276e2864 BP |
400 | unsigned int l7_length; |
401 | struct tcp_key key; | |
402 | uint32_t hash; | |
f3dd1419 BP |
403 | uint32_t seq; |
404 | uint8_t flags; | |
cf62fa4c | 405 | const char *l7 = dp_packet_get_tcp_payload(packet); |
f3dd1419 BP |
406 | |
407 | if (flow->dl_type != htons(ETH_TYPE_IP) | |
408 | || flow->nw_proto != IPPROTO_TCP | |
5a51b2cd | 409 | || !l7) { |
f3dd1419 BP |
410 | return NULL; |
411 | } | |
cf62fa4c | 412 | tcp = dp_packet_l4(packet); |
f3dd1419 | 413 | flags = TCP_FLAGS(tcp->tcp_ctl); |
cf62fa4c | 414 | l7_length = (char *) dp_packet_tail(packet) - l7; |
f3dd1419 | 415 | seq = ntohl(get_16aligned_be32(&tcp->tcp_seq)); |
276e2864 BP |
416 | |
417 | /* Construct key. */ | |
418 | memset(&key, 0, sizeof key); | |
419 | key.nw_src = flow->nw_src; | |
420 | key.nw_dst = flow->nw_dst; | |
421 | key.tp_src = flow->tp_src; | |
422 | key.tp_dst = flow->tp_dst; | |
423 | hash = hash_bytes(&key, sizeof key, 0); | |
424 | ||
425 | /* Find existing stream or start a new one for a SYN or if there's data. */ | |
426 | stream = tcp_stream_lookup(r, &key, hash); | |
427 | if (!stream) { | |
428 | if (flags & TCP_SYN || l7_length) { | |
429 | stream = tcp_stream_new(r, &key, hash); | |
430 | stream->seq_no = flags & TCP_SYN ? seq + 1 : seq; | |
431 | } else { | |
432 | return NULL; | |
433 | } | |
434 | } | |
435 | ||
436 | payload = &stream->payload; | |
437 | if (flags & TCP_SYN || !stream->seq_no) { | |
cf62fa4c | 438 | dp_packet_clear(payload); |
f3dd1419 BP |
439 | stream->seq_no = seq + 1; |
440 | return NULL; | |
441 | } else if (flags & (TCP_FIN | TCP_RST)) { | |
442 | tcp_stream_destroy(r, stream); | |
443 | return NULL; | |
444 | } else if (seq == stream->seq_no) { | |
f3dd1419 BP |
445 | /* Shift all of the existing payload to the very beginning of the |
446 | * allocated space, so that we reuse allocated space instead of | |
447 | * continually expanding it. */ | |
cf62fa4c | 448 | dp_packet_shift(payload, (char *) dp_packet_base(payload) - (char *) dp_packet_data(payload)); |
f3dd1419 | 449 | |
cf62fa4c | 450 | dp_packet_put(payload, l7, l7_length); |
276e2864 | 451 | stream->seq_no += l7_length; |
f3dd1419 BP |
452 | return payload; |
453 | } else { | |
454 | return NULL; | |
455 | } | |
456 | } |