Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Stream filters related variables and functions. |
| 3 | * |
| 4 | * Copyright (C) 2015 Qualys Inc., Christopher Faulet <cfaulet@qualys.com> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public License |
| 8 | * as published by the Free Software Foundation; either version |
| 9 | * 2 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | */ |
| 12 | |
| 13 | #include <common/buffer.h> |
| 14 | #include <common/cfgparse.h> |
| 15 | #include <common/mini-clist.h> |
| 16 | #include <common/standard.h> |
| 17 | |
| 18 | #include <types/compression.h> |
| 19 | #include <types/filters.h> |
| 20 | #include <types/proto_http.h> |
| 21 | #include <types/proxy.h> |
| 22 | #include <types/sample.h> |
| 23 | |
| 24 | #include <proto/compression.h> |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 25 | #include <proto/filters.h> |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 26 | #include <proto/hdr_idx.h> |
| 27 | #include <proto/proto_http.h> |
| 28 | #include <proto/sample.h> |
| 29 | #include <proto/stream.h> |
| 30 | |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 31 | static const char *http_comp_flt_id = "compression filter"; |
| 32 | |
| 33 | struct flt_ops comp_ops; |
| 34 | |
| 35 | static struct buffer *tmpbuf = &buf_empty; |
| 36 | |
| 37 | struct comp_chunk { |
| 38 | unsigned int start; /* start of the chunk relative to FLT_FWD offset */ |
| 39 | unsigned int end; /* end of the chunk relative to FLT_FWD offset */ |
| 40 | int skip; /* if set to 1, the chunk is skipped. Otherwise it is compressed */ |
| 41 | int is_last; /* if set, this is the last chunk. Data after this |
| 42 | * chunk will be forwarded as it is. */ |
| 43 | struct list list; |
| 44 | }; |
| 45 | |
| 46 | struct comp_state { |
| 47 | struct comp_ctx *comp_ctx; /* compression context */ |
| 48 | struct comp_algo *comp_algo; /* compression algorithm if not NULL */ |
| 49 | struct list comp_chunks; /* data chunks that should be compressed or skipped */ |
| 50 | unsigned int first; /* offset of the first chunk. Data before |
| 51 | * this offset will be forwarded as it |
| 52 | * is. */ |
| 53 | }; |
| 54 | |
| 55 | static int add_comp_chunk(struct comp_state *st, unsigned int start, |
| 56 | unsigned int len, int skip, int is_last); |
| 57 | static int skip_input_data(struct filter *filter, struct http_msg *msg, |
| 58 | unsigned int consumed); |
| 59 | |
| 60 | static int select_compression_request_header(struct comp_state *st, |
| 61 | struct stream *s, |
| 62 | struct http_msg *msg); |
| 63 | static int select_compression_response_header(struct comp_state *st, |
| 64 | struct stream *s, |
| 65 | struct http_msg *msg); |
| 66 | |
| 67 | static int http_compression_buffer_init(struct buffer *in, struct buffer *out); |
| 68 | static int http_compression_buffer_add_data(struct comp_state *st, |
| 69 | struct buffer *in, |
| 70 | struct buffer *out, int sz); |
| 71 | static int http_compression_buffer_end(struct comp_state *st, struct stream *s, |
| 72 | struct buffer **in, struct buffer **out, |
| 73 | unsigned int consumed, int end); |
| 74 | |
| 75 | /***********************************************************************/ |
| 76 | static int |
| 77 | comp_flt_init(struct proxy *px, struct filter *filter) |
| 78 | { |
| 79 | |
| 80 | /* We need a compression buffer in the DATA state to put the output of |
| 81 | * compressed data, and in CRLF state to let the TRAILERS state finish |
| 82 | * the job of removing the trailing CRLF. |
| 83 | */ |
| 84 | if (!tmpbuf->size) { |
| 85 | if (b_alloc(&tmpbuf) == NULL) |
| 86 | return -1; |
| 87 | } |
| 88 | return 0; |
| 89 | } |
| 90 | |
| 91 | static void |
| 92 | comp_flt_deinit(struct proxy *px, struct filter *filter) |
| 93 | { |
| 94 | if (tmpbuf->size) |
| 95 | b_free(&tmpbuf); |
| 96 | } |
| 97 | |
| 98 | static int |
| 99 | comp_start_analyze(struct stream *s, struct filter *filter, struct channel *chn) |
| 100 | { |
| 101 | if (filter->ctx == NULL) { |
| 102 | struct comp_state *st; |
| 103 | |
| 104 | if (!(st = malloc(sizeof(*st)))) |
| 105 | return -1; |
| 106 | |
| 107 | LIST_INIT(&st->comp_chunks); |
| 108 | st->comp_algo = NULL; |
| 109 | st->comp_ctx = NULL; |
| 110 | st->first = 0; |
| 111 | filter->ctx = st; |
| 112 | } |
| 113 | return 1; |
| 114 | } |
| 115 | |
| 116 | static int |
| 117 | comp_analyze(struct stream *s, struct filter *filter, struct channel *chn, |
| 118 | unsigned int an_bit) |
| 119 | { |
| 120 | struct comp_state *st = filter->ctx; |
| 121 | |
| 122 | if (!strm_fe(s)->comp && !s->be->comp) |
| 123 | goto end; |
| 124 | |
| 125 | switch (an_bit) { |
| 126 | case AN_RES_HTTP_PROCESS_BE: |
| 127 | select_compression_response_header(st, s, &s->txn->rsp); |
| 128 | break; |
| 129 | } |
| 130 | end: |
| 131 | return 1; |
| 132 | } |
| 133 | |
| 134 | static int |
| 135 | comp_end_analyze(struct stream *s, struct filter *filter, struct channel *chn) |
| 136 | { |
| 137 | struct comp_state *st = filter->ctx; |
| 138 | struct comp_chunk *cc, *back; |
| 139 | |
| 140 | if (!st || !(chn->flags & CF_ISRESP)) |
| 141 | goto end; |
| 142 | |
| 143 | list_for_each_entry_safe(cc, back, &st->comp_chunks, list) { |
| 144 | LIST_DEL(&cc->list); |
| 145 | free(cc); |
| 146 | } |
| 147 | |
| 148 | if (!st->comp_algo || !s->txn->status) |
| 149 | goto release_ctx; |
| 150 | |
| 151 | if (strm_fe(s)->mode == PR_MODE_HTTP) |
| 152 | strm_fe(s)->fe_counters.p.http.comp_rsp++; |
| 153 | if ((s->flags & SF_BE_ASSIGNED) && (s->be->mode == PR_MODE_HTTP)) |
| 154 | s->be->be_counters.p.http.comp_rsp++; |
| 155 | |
| 156 | /* release any possible compression context */ |
| 157 | st->comp_algo->end(&st->comp_ctx); |
| 158 | |
| 159 | release_ctx: |
| 160 | free(st); |
| 161 | filter->ctx = NULL; |
| 162 | end: |
| 163 | return 1; |
| 164 | } |
| 165 | |
| 166 | static int |
| 167 | comp_http_headers(struct stream *s, struct filter *filter, |
| 168 | struct http_msg *msg) |
| 169 | { |
| 170 | struct comp_state *st = filter->ctx; |
| 171 | |
| 172 | if (strm_fe(s)->comp || s->be->comp) { |
| 173 | if (!(msg->chn->flags & CF_ISRESP)) |
| 174 | select_compression_request_header(st, s, msg); |
| 175 | } |
| 176 | return 1; |
| 177 | } |
| 178 | |
| 179 | static int |
| 180 | comp_skip_http_chunk_envelope(struct stream *s, struct filter *filter, |
| 181 | struct http_msg *msg) |
| 182 | { |
| 183 | struct comp_state *st = filter->ctx; |
| 184 | unsigned int start; |
| 185 | int ret; |
| 186 | |
| 187 | if (!(msg->chn->flags & CF_ISRESP) || !st->comp_algo) { |
| 188 | flt_set_forward_data(filter, msg->chn); |
| 189 | return 1; |
| 190 | } |
| 191 | |
| 192 | start = FLT_NXT(filter, msg->chn) - FLT_FWD(filter, msg->chn); |
| 193 | /* If this is the last chunk, we flag it */ |
| 194 | if (msg->chunk_len == 0 && msg->msg_state == HTTP_MSG_CHUNK_SIZE) |
| 195 | ret = add_comp_chunk(st, start, 0, 1, 1); |
| 196 | else |
| 197 | ret = add_comp_chunk(st, start, msg->sol, 1, 0); |
| 198 | |
| 199 | return !ret ? 1 : -1; |
| 200 | } |
| 201 | |
| 202 | static int |
| 203 | comp_http_data(struct stream *s, struct filter *filter, |
| 204 | struct http_msg *msg) |
| 205 | { |
| 206 | struct comp_state *st = filter->ctx; |
| 207 | unsigned int start; |
| 208 | int is_last, ret; |
| 209 | |
| 210 | ret = MIN(msg->chunk_len + msg->next, msg->chn->buf->i) - FLT_NXT(filter, msg->chn); |
| 211 | if (!(msg->chn->flags & CF_ISRESP) || !st->comp_algo) { |
| 212 | flt_set_forward_data(filter, msg->chn); |
| 213 | goto end; |
| 214 | } |
| 215 | if (!ret) |
| 216 | goto end; |
| 217 | |
| 218 | start = FLT_NXT(filter, msg->chn) - FLT_FWD(filter, msg->chn); |
| 219 | is_last = (!(msg->flags & HTTP_MSGF_TE_CHNK) && |
| 220 | (msg->chunk_len == ret - msg->next + FLT_NXT(filter, msg->chn))); |
| 221 | |
| 222 | if (add_comp_chunk(st, start, ret, 0, is_last) == -1) |
| 223 | ret = -1; |
| 224 | end: |
| 225 | return ret; |
| 226 | } |
| 227 | |
| 228 | static int |
| 229 | comp_http_forward_data(struct stream *s, struct filter *filter, |
| 230 | struct http_msg *msg, unsigned int len) |
| 231 | { |
| 232 | struct comp_state *st = filter->ctx; |
| 233 | struct comp_chunk *cc, *back; |
| 234 | unsigned int sz, consumed = 0, compressed = 0; |
| 235 | int is_last = 0, ret = len; |
| 236 | |
| 237 | if (!(msg->chn->flags & CF_ISRESP) || !st->comp_algo) { |
| 238 | flt_set_forward_data(filter, msg->chn); |
| 239 | goto end; |
| 240 | } |
| 241 | |
| 242 | /* no data to forward or no chunk or the first chunk is too far */ |
| 243 | if (!len || LIST_ISEMPTY(&st->comp_chunks)) |
| 244 | goto end; |
| 245 | if (st->first > len) { |
| 246 | consumed = len; |
| 247 | goto update_chunks; |
| 248 | } |
| 249 | |
| 250 | /* initialize the buffer used to write compressed data */ |
| 251 | b_adv(msg->chn->buf, FLT_FWD(filter, msg->chn) + st->first); |
| 252 | ret = http_compression_buffer_init(msg->chn->buf, tmpbuf); |
| 253 | b_rew(msg->chn->buf, FLT_FWD(filter, msg->chn) + st->first); |
| 254 | if (ret < 0) { |
| 255 | msg->chn->flags |= CF_WAKE_WRITE; |
| 256 | return 0; |
| 257 | } |
| 258 | |
| 259 | /* Loop on all chunks */ |
| 260 | list_for_each_entry_safe(cc, back, &st->comp_chunks, list) { |
| 261 | /* current chunk must not be handled yet */ |
| 262 | if (len <= cc->start) { |
| 263 | consumed = len; |
| 264 | break; |
| 265 | } |
| 266 | |
| 267 | /* Get the number of bytes that must be handled in the current |
| 268 | * chunk */ |
| 269 | sz = MIN(len, cc->end) - cc->start; |
| 270 | |
| 271 | if (cc->skip) { |
| 272 | /* No compression for this chunk, data must be |
| 273 | * skipped. This happens when the HTTP response is |
| 274 | * chunked, the chunk envelope is skipped. */ |
| 275 | ret = sz; |
| 276 | } |
| 277 | else { |
| 278 | /* Compress the chunk */ |
| 279 | b_adv(msg->chn->buf, FLT_FWD(filter, msg->chn) + cc->start); |
| 280 | ret = http_compression_buffer_add_data(st, msg->chn->buf, tmpbuf, sz); |
| 281 | b_rew(msg->chn->buf, FLT_FWD(filter, msg->chn) + cc->start); |
| 282 | if (ret < 0) |
| 283 | goto end; |
| 284 | compressed += ret; |
| 285 | } |
| 286 | |
| 287 | /* Update the chunk by removing consumed bytes. If all bytes are |
| 288 | * consumed, the chunk is removed from the list and we |
| 289 | * loop. Otherwise, we stop here. */ |
| 290 | cc->start += ret; |
| 291 | consumed = cc->start; |
| 292 | if (cc->start != cc->end) |
| 293 | break; |
| 294 | |
| 295 | /* Remember if this is the last chunk */ |
| 296 | is_last = cc->is_last; |
| 297 | LIST_DEL(&cc->list); |
| 298 | free(cc); |
| 299 | } |
| 300 | |
| 301 | if (compressed) { |
| 302 | /* Some data was compressed so we can switch buffers to replace |
| 303 | * uncompressed data by compressed ones. */ |
| 304 | b_adv(msg->chn->buf, FLT_FWD(filter, msg->chn) + st->first); |
| 305 | ret = http_compression_buffer_end(st, s, &msg->chn->buf, &tmpbuf, |
| 306 | consumed - st->first, is_last); |
| 307 | b_rew(msg->chn->buf, FLT_FWD(filter, msg->chn) + st->first); |
| 308 | } |
| 309 | else { |
| 310 | /* Here some data was consumed but no compression was |
| 311 | * preformed. This means that all consumed data must be |
| 312 | * skipped. |
| 313 | */ |
| 314 | ret = skip_input_data(filter, msg, consumed); |
| 315 | } |
| 316 | |
| 317 | if (is_last && !(msg->flags & HTTP_MSGF_TE_CHNK)) { |
| 318 | /* At the end of data, if the original response was not |
| 319 | * chunked-encoded, we must write the empty chunk 0<CRLF>, and |
| 320 | * terminate the (empty) trailers section with a last <CRLF>. If |
| 321 | * we're forwarding a chunked-encoded response, these parts are |
| 322 | * preserved and not rewritten. |
| 323 | */ |
| 324 | char *p = bi_end(msg->chn->buf); |
| 325 | memcpy(p, "0\r\n\r\n", 5); |
| 326 | msg->chn->buf->i += 5; |
| 327 | ret += 5; |
| 328 | } |
| 329 | |
| 330 | /* Then, the last step. We need to update state of other filters. */ |
| 331 | if (ret >= 0) { |
| 332 | flt_change_forward_size(filter, msg->chn, -(consumed - st->first - ret)); |
| 333 | msg->next -= (consumed - st->first - ret); |
| 334 | ret += st->first; |
| 335 | } |
| 336 | |
| 337 | update_chunks: |
| 338 | /* Now, we need to update all remaining chunks to keep them synchronized |
| 339 | * with the next position of buf->p. If the chunk list is empty, we |
| 340 | * forward remaining data, if any. */ |
| 341 | st->first -= MIN(st->first, consumed); |
| 342 | if (LIST_ISEMPTY(&st->comp_chunks)) |
| 343 | ret += len - consumed; |
| 344 | else { |
| 345 | list_for_each_entry(cc, &st->comp_chunks, list) { |
| 346 | cc->start -= consumed; |
| 347 | cc->end -= consumed; |
| 348 | } |
| 349 | } |
| 350 | |
| 351 | end: |
| 352 | return ret; |
| 353 | } |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 354 | |
| 355 | /***********************************************************************/ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 356 | static int |
| 357 | add_comp_chunk(struct comp_state *st, unsigned int start, unsigned int len, |
| 358 | int skip, int is_last) |
| 359 | { |
| 360 | struct comp_chunk *cc; |
| 361 | |
| 362 | if (!(cc = malloc(sizeof(*cc)))) |
| 363 | return -1; |
| 364 | cc->start = start; |
| 365 | cc->end = start + len; |
| 366 | cc->skip = skip; |
| 367 | cc->is_last = is_last; |
| 368 | |
| 369 | if (LIST_ISEMPTY(&st->comp_chunks)) |
| 370 | st->first = cc->start; |
| 371 | |
| 372 | LIST_ADDQ(&st->comp_chunks, &cc->list); |
| 373 | return 0; |
| 374 | } |
| 375 | |
| 376 | /* This function might be moved in a filter function, probably with others to |
| 377 | * add/remove/move/replace buffer data */ |
| 378 | static int |
| 379 | skip_input_data(struct filter *filter, struct http_msg *msg, |
| 380 | unsigned int consumed) |
| 381 | { |
| 382 | struct comp_state *st = filter->ctx; |
| 383 | int block1, block2; |
| 384 | |
| 385 | /* 1. Copy input data, skipping consumed ones. */ |
| 386 | b_adv(msg->chn->buf, FLT_FWD(filter, msg->chn) + st->first + consumed); |
| 387 | block1 = msg->chn->buf->i; |
| 388 | if (block1 > bi_contig_data(msg->chn->buf)) |
| 389 | block1 = bi_contig_data(msg->chn->buf); |
| 390 | block2 = msg->chn->buf->i - block1; |
| 391 | |
| 392 | memcpy(trash.str, bi_ptr(msg->chn->buf), block1); |
| 393 | if (block2 > 0) |
| 394 | memcpy(trash.str + block1, msg->chn->buf->data, block2); |
| 395 | trash.len = block1 + block2; |
| 396 | b_rew(msg->chn->buf, FLT_FWD(filter, msg->chn) + st->first + consumed); |
| 397 | |
| 398 | /* 2. Then write back these data at the right place in the buffer */ |
| 399 | b_adv(msg->chn->buf, FLT_FWD(filter, msg->chn) + st->first); |
| 400 | block1 = trash.len; |
| 401 | if (block1 > bi_contig_data(msg->chn->buf)) |
| 402 | block1 = bi_contig_data(msg->chn->buf); |
| 403 | block2 = trash.len - block1; |
| 404 | |
| 405 | memcpy(bi_ptr(msg->chn->buf), trash.str, block1); |
| 406 | if (block2 > 0) |
| 407 | memcpy(msg->chn->buf->data, trash.str + block1, block2); |
| 408 | b_rew(msg->chn->buf, FLT_FWD(filter, msg->chn) + st->first); |
| 409 | |
| 410 | /* Then adjut the input size */ |
| 411 | msg->chn->buf->i -= consumed; |
| 412 | return 0; |
| 413 | } |
| 414 | |
| 415 | /***********************************************************************/ |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 416 | /* |
| 417 | * Selects a compression algorithm depending on the client request. |
| 418 | */ |
| 419 | int |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 420 | select_compression_request_header(struct comp_state *st, struct stream *s, |
| 421 | struct http_msg *msg) |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 422 | { |
| 423 | struct http_txn *txn = s->txn; |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 424 | struct buffer *req = msg->chn->buf; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 425 | struct hdr_ctx ctx; |
| 426 | struct comp_algo *comp_algo = NULL; |
| 427 | struct comp_algo *comp_algo_back = NULL; |
| 428 | |
| 429 | /* Disable compression for older user agents announcing themselves as "Mozilla/4" |
| 430 | * unless they are known good (MSIE 6 with XP SP2, or MSIE 7 and later). |
| 431 | * See http://zoompf.com/2012/02/lose-the-wait-http-compression for more details. |
| 432 | */ |
| 433 | ctx.idx = 0; |
| 434 | if (http_find_header2("User-Agent", 10, req->p, &txn->hdr_idx, &ctx) && |
| 435 | ctx.vlen >= 9 && |
| 436 | memcmp(ctx.line + ctx.val, "Mozilla/4", 9) == 0 && |
| 437 | (ctx.vlen < 31 || |
| 438 | memcmp(ctx.line + ctx.val + 25, "MSIE ", 5) != 0 || |
| 439 | ctx.line[ctx.val + 30] < '6' || |
| 440 | (ctx.line[ctx.val + 30] == '6' && |
| 441 | (ctx.vlen < 54 || memcmp(ctx.line + 51, "SV1", 3) != 0)))) { |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 442 | st->comp_algo = NULL; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 443 | return 0; |
| 444 | } |
| 445 | |
| 446 | /* search for the algo in the backend in priority or the frontend */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 447 | if ((s->be->comp && (comp_algo_back = s->be->comp->algos)) || |
| 448 | (strm_fe(s)->comp && (comp_algo_back = strm_fe(s)->comp->algos))) { |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 449 | int best_q = 0; |
| 450 | |
| 451 | ctx.idx = 0; |
| 452 | while (http_find_header2("Accept-Encoding", 15, req->p, &txn->hdr_idx, &ctx)) { |
| 453 | const char *qval; |
| 454 | int q; |
| 455 | int toklen; |
| 456 | |
| 457 | /* try to isolate the token from the optional q-value */ |
| 458 | toklen = 0; |
| 459 | while (toklen < ctx.vlen && http_is_token[(unsigned char)*(ctx.line + ctx.val + toklen)]) |
| 460 | toklen++; |
| 461 | |
| 462 | qval = ctx.line + ctx.val + toklen; |
| 463 | while (1) { |
| 464 | while (qval < ctx.line + ctx.val + ctx.vlen && http_is_lws[(unsigned char)*qval]) |
| 465 | qval++; |
| 466 | |
| 467 | if (qval >= ctx.line + ctx.val + ctx.vlen || *qval != ';') { |
| 468 | qval = NULL; |
| 469 | break; |
| 470 | } |
| 471 | qval++; |
| 472 | |
| 473 | while (qval < ctx.line + ctx.val + ctx.vlen && http_is_lws[(unsigned char)*qval]) |
| 474 | qval++; |
| 475 | |
| 476 | if (qval >= ctx.line + ctx.val + ctx.vlen) { |
| 477 | qval = NULL; |
| 478 | break; |
| 479 | } |
| 480 | if (strncmp(qval, "q=", MIN(ctx.line + ctx.val + ctx.vlen - qval, 2)) == 0) |
| 481 | break; |
| 482 | |
| 483 | while (qval < ctx.line + ctx.val + ctx.vlen && *qval != ';') |
| 484 | qval++; |
| 485 | } |
| 486 | |
| 487 | /* here we have qval pointing to the first "q=" attribute or NULL if not found */ |
| 488 | q = qval ? parse_qvalue(qval + 2, NULL) : 1000; |
| 489 | |
| 490 | if (q <= best_q) |
| 491 | continue; |
| 492 | |
| 493 | for (comp_algo = comp_algo_back; comp_algo; comp_algo = comp_algo->next) { |
| 494 | if (*(ctx.line + ctx.val) == '*' || |
| 495 | word_match(ctx.line + ctx.val, toklen, comp_algo->ua_name, comp_algo->ua_name_len)) { |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 496 | st->comp_algo = comp_algo; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 497 | best_q = q; |
| 498 | break; |
| 499 | } |
| 500 | } |
| 501 | } |
| 502 | } |
| 503 | |
| 504 | /* remove all occurrences of the header when "compression offload" is set */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 505 | if (st->comp_algo) { |
| 506 | if ((s->be->comp && s->be->comp->offload) || |
| 507 | (strm_fe(s)->comp && strm_fe(s)->comp->offload)) { |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 508 | http_remove_header2(msg, &txn->hdr_idx, &ctx); |
| 509 | ctx.idx = 0; |
| 510 | while (http_find_header2("Accept-Encoding", 15, req->p, &txn->hdr_idx, &ctx)) { |
| 511 | http_remove_header2(msg, &txn->hdr_idx, &ctx); |
| 512 | } |
| 513 | } |
| 514 | return 1; |
| 515 | } |
| 516 | |
| 517 | /* identity is implicit does not require headers */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 518 | if ((s->be->comp && (comp_algo_back = s->be->comp->algos)) || |
| 519 | (strm_fe(s)->comp && (comp_algo_back = strm_fe(s)->comp->algos))) { |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 520 | for (comp_algo = comp_algo_back; comp_algo; comp_algo = comp_algo->next) { |
| 521 | if (comp_algo->cfg_name_len == 8 && memcmp(comp_algo->cfg_name, "identity", 8) == 0) { |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 522 | st->comp_algo = comp_algo; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 523 | return 1; |
| 524 | } |
| 525 | } |
| 526 | } |
| 527 | |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 528 | st->comp_algo = NULL; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 529 | return 0; |
| 530 | } |
| 531 | |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 532 | |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 533 | /* |
| 534 | * Selects a comression algorithm depending of the server response. |
| 535 | */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 536 | static int |
| 537 | select_compression_response_header(struct comp_state *st, struct stream *s, struct http_msg *msg) |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 538 | { |
| 539 | struct http_txn *txn = s->txn; |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 540 | struct buffer *res = msg->chn->buf; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 541 | struct hdr_ctx ctx; |
| 542 | struct comp_type *comp_type; |
| 543 | |
| 544 | /* no common compression algorithm was found in request header */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 545 | if (st->comp_algo == NULL) |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 546 | goto fail; |
| 547 | |
| 548 | /* HTTP < 1.1 should not be compressed */ |
| 549 | if (!(msg->flags & HTTP_MSGF_VER_11) || !(txn->req.flags & HTTP_MSGF_VER_11)) |
| 550 | goto fail; |
| 551 | |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 552 | if (txn->meth == HTTP_METH_HEAD) |
| 553 | goto fail; |
| 554 | |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 555 | /* compress 200,201,202,203 responses only */ |
| 556 | if ((txn->status != 200) && |
| 557 | (txn->status != 201) && |
| 558 | (txn->status != 202) && |
| 559 | (txn->status != 203)) |
| 560 | goto fail; |
| 561 | |
| 562 | |
| 563 | /* Content-Length is null */ |
| 564 | if (!(msg->flags & HTTP_MSGF_TE_CHNK) && msg->body_len == 0) |
| 565 | goto fail; |
| 566 | |
| 567 | /* content is already compressed */ |
| 568 | ctx.idx = 0; |
| 569 | if (http_find_header2("Content-Encoding", 16, res->p, &txn->hdr_idx, &ctx)) |
| 570 | goto fail; |
| 571 | |
| 572 | /* no compression when Cache-Control: no-transform is present in the message */ |
| 573 | ctx.idx = 0; |
| 574 | while (http_find_header2("Cache-Control", 13, res->p, &txn->hdr_idx, &ctx)) { |
| 575 | if (word_match(ctx.line + ctx.val, ctx.vlen, "no-transform", 12)) |
| 576 | goto fail; |
| 577 | } |
| 578 | |
| 579 | comp_type = NULL; |
| 580 | |
| 581 | /* we don't want to compress multipart content-types, nor content-types that are |
| 582 | * not listed in the "compression type" directive if any. If no content-type was |
| 583 | * found but configuration requires one, we don't compress either. Backend has |
| 584 | * the priority. |
| 585 | */ |
| 586 | ctx.idx = 0; |
| 587 | if (http_find_header2("Content-Type", 12, res->p, &txn->hdr_idx, &ctx)) { |
| 588 | if (ctx.vlen >= 9 && strncasecmp("multipart", ctx.line+ctx.val, 9) == 0) |
| 589 | goto fail; |
| 590 | |
| 591 | if ((s->be->comp && (comp_type = s->be->comp->types)) || |
| 592 | (strm_fe(s)->comp && (comp_type = strm_fe(s)->comp->types))) { |
| 593 | for (; comp_type; comp_type = comp_type->next) { |
| 594 | if (ctx.vlen >= comp_type->name_len && |
| 595 | strncasecmp(ctx.line+ctx.val, comp_type->name, comp_type->name_len) == 0) |
| 596 | /* this Content-Type should be compressed */ |
| 597 | break; |
| 598 | } |
| 599 | /* this Content-Type should not be compressed */ |
| 600 | if (comp_type == NULL) |
| 601 | goto fail; |
| 602 | } |
| 603 | } |
| 604 | else { /* no content-type header */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 605 | if ((s->be->comp && s->be->comp->types) || |
| 606 | (strm_fe(s)->comp && strm_fe(s)->comp->types)) |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 607 | goto fail; /* a content-type was required */ |
| 608 | } |
| 609 | |
| 610 | /* limit compression rate */ |
| 611 | if (global.comp_rate_lim > 0) |
| 612 | if (read_freq_ctr(&global.comp_bps_in) > global.comp_rate_lim) |
| 613 | goto fail; |
| 614 | |
| 615 | /* limit cpu usage */ |
| 616 | if (idle_pct < compress_min_idle) |
| 617 | goto fail; |
| 618 | |
| 619 | /* initialize compression */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 620 | if (st->comp_algo->init(&st->comp_ctx, global.tune.comp_maxlevel) < 0) |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 621 | goto fail; |
| 622 | |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 623 | /* remove Content-Length header */ |
| 624 | ctx.idx = 0; |
| 625 | if ((msg->flags & HTTP_MSGF_CNT_LEN) && http_find_header2("Content-Length", 14, res->p, &txn->hdr_idx, &ctx)) |
| 626 | http_remove_header2(msg, &txn->hdr_idx, &ctx); |
| 627 | |
| 628 | /* add Transfer-Encoding header */ |
| 629 | if (!(msg->flags & HTTP_MSGF_TE_CHNK)) |
| 630 | http_header_add_tail2(&txn->rsp, &txn->hdr_idx, "Transfer-Encoding: chunked", 26); |
| 631 | |
| 632 | /* |
| 633 | * Add Content-Encoding header when it's not identity encoding. |
| 634 | * RFC 2616 : Identity encoding: This content-coding is used only in the |
| 635 | * Accept-Encoding header, and SHOULD NOT be used in the Content-Encoding |
| 636 | * header. |
| 637 | */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 638 | if (st->comp_algo->cfg_name_len != 8 || memcmp(st->comp_algo->cfg_name, "identity", 8) != 0) { |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 639 | trash.len = 18; |
| 640 | memcpy(trash.str, "Content-Encoding: ", trash.len); |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 641 | memcpy(trash.str + trash.len, st->comp_algo->ua_name, st->comp_algo->ua_name_len); |
| 642 | trash.len += st->comp_algo->ua_name_len; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 643 | trash.str[trash.len] = '\0'; |
| 644 | http_header_add_tail2(&txn->rsp, &txn->hdr_idx, trash.str, trash.len); |
| 645 | } |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 646 | msg->flags |= HTTP_MSGF_COMPRESSING; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 647 | return 1; |
| 648 | |
| 649 | fail: |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 650 | st->comp_algo = NULL; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 651 | return 0; |
| 652 | } |
| 653 | |
| 654 | /***********************************************************************/ |
| 655 | /* emit the chunksize followed by a CRLF on the output and return the number of |
| 656 | * bytes written. It goes backwards and starts with the byte before <end>. It |
| 657 | * returns the number of bytes written which will not exceed 10 (8 digits, CR, |
| 658 | * and LF). The caller is responsible for ensuring there is enough room left in |
| 659 | * the output buffer for the string. |
| 660 | */ |
| 661 | static int |
| 662 | http_emit_chunk_size(char *end, unsigned int chksz) |
| 663 | { |
| 664 | char *beg = end; |
| 665 | |
| 666 | *--beg = '\n'; |
| 667 | *--beg = '\r'; |
| 668 | do { |
| 669 | *--beg = hextab[chksz & 0xF]; |
| 670 | } while (chksz >>= 4); |
| 671 | return end - beg; |
| 672 | } |
| 673 | |
| 674 | /* |
| 675 | * Init HTTP compression |
| 676 | */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 677 | static int |
| 678 | http_compression_buffer_init(struct buffer *in, struct buffer *out) |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 679 | { |
| 680 | /* output stream requires at least 10 bytes for the gzip header, plus |
| 681 | * at least 8 bytes for the gzip trailer (crc+len), plus a possible |
| 682 | * plus at most 5 bytes per 32kB block and 2 bytes to close the stream. |
| 683 | */ |
| 684 | if (in->size - buffer_len(in) < 20 + 5 * ((in->i + 32767) >> 15)) |
| 685 | return -1; |
| 686 | |
| 687 | /* prepare an empty output buffer in which we reserve enough room for |
| 688 | * copying the output bytes from <in>, plus 10 extra bytes to write |
| 689 | * the chunk size. We don't copy the bytes yet so that if we have to |
| 690 | * cancel the operation later, it's cheap. |
| 691 | */ |
| 692 | b_reset(out); |
| 693 | out->o = in->o; |
| 694 | out->p += out->o; |
| 695 | out->i = 10; |
| 696 | return 0; |
| 697 | } |
| 698 | |
| 699 | /* |
| 700 | * Add data to compress |
| 701 | */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 702 | static int |
| 703 | http_compression_buffer_add_data(struct comp_state *st, struct buffer *in, |
| 704 | struct buffer *out, int sz) |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 705 | { |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 706 | int consumed_data = 0; |
| 707 | int data_process_len; |
| 708 | int block1, block2; |
| 709 | |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 710 | if (!sz) |
| 711 | return 0; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 712 | |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 713 | /* select the smallest size between the announced chunk size, the input |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 714 | * data, and the available output buffer size. The compressors are |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 715 | * assumed to be able to process all the bytes we pass to them at |
| 716 | * once. */ |
| 717 | data_process_len = sz; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 718 | data_process_len = MIN(out->size - buffer_len(out), data_process_len); |
| 719 | |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 720 | |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 721 | block1 = data_process_len; |
| 722 | if (block1 > bi_contig_data(in)) |
| 723 | block1 = bi_contig_data(in); |
| 724 | block2 = data_process_len - block1; |
| 725 | |
| 726 | /* compressors return < 0 upon error or the amount of bytes read */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 727 | consumed_data = st->comp_algo->add_data(st->comp_ctx, bi_ptr(in), block1, out); |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 728 | if (consumed_data >= 0 && block2 > 0) { |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 729 | consumed_data = st->comp_algo->add_data(st->comp_ctx, in->data, block2, out); |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 730 | if (consumed_data >= 0) |
| 731 | consumed_data += block1; |
| 732 | } |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 733 | return consumed_data; |
| 734 | } |
| 735 | |
| 736 | /* |
| 737 | * Flush data in process, and write the header and footer of the chunk. Upon |
| 738 | * success, in and out buffers are swapped to avoid a copy. |
| 739 | */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 740 | static int |
| 741 | http_compression_buffer_end(struct comp_state *st, struct stream *s, |
| 742 | struct buffer **in, struct buffer **out, |
| 743 | unsigned int consumed, int end) |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 744 | { |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 745 | struct buffer *ib = *in, *ob = *out; |
| 746 | char *tail; |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 747 | int to_forward, left; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 748 | |
| 749 | #if defined(USE_SLZ) || defined(USE_ZLIB) |
| 750 | int ret; |
| 751 | |
| 752 | /* flush data here */ |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 753 | if (end) |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 754 | ret = st->comp_algo->finish(st->comp_ctx, ob); /* end of data */ |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 755 | else |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 756 | ret = st->comp_algo->flush(st->comp_ctx, ob); /* end of buffer */ |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 757 | |
| 758 | if (ret < 0) |
| 759 | return -1; /* flush failed */ |
| 760 | |
| 761 | #endif /* USE_ZLIB */ |
| 762 | |
| 763 | if (ob->i == 10) { |
| 764 | /* No data were appended, let's drop the output buffer and |
| 765 | * keep the input buffer unchanged. |
| 766 | */ |
| 767 | return 0; |
| 768 | } |
| 769 | |
| 770 | /* OK so at this stage, we have an output buffer <ob> looking like this : |
| 771 | * |
| 772 | * <-- o --> <------ i -----> |
| 773 | * +---------+---+------------+-----------+ |
| 774 | * | out | c | comp_in | empty | |
| 775 | * +---------+---+------------+-----------+ |
| 776 | * data p size |
| 777 | * |
| 778 | * <out> is the room reserved to copy ib->o. It starts at ob->data and |
| 779 | * has not yet been filled. <c> is the room reserved to write the chunk |
| 780 | * size (10 bytes). <comp_in> is the compressed equivalent of the data |
| 781 | * part of ib->i. <empty> is the amount of empty bytes at the end of |
| 782 | * the buffer, into which we may have to copy the remaining bytes from |
| 783 | * ib->i after the data (chunk size, trailers, ...). |
| 784 | */ |
| 785 | |
| 786 | /* Write real size at the begining of the chunk, no need of wrapping. |
| 787 | * We write the chunk using a dynamic length and adjust ob->p and ob->i |
| 788 | * accordingly afterwards. That will move <out> away from <data>. |
| 789 | */ |
| 790 | left = 10 - http_emit_chunk_size(ob->p + 10, ob->i - 10); |
| 791 | ob->p += left; |
| 792 | ob->i -= left; |
| 793 | |
| 794 | /* Copy previous data from ib->o into ob->o */ |
| 795 | if (ib->o > 0) { |
| 796 | left = bo_contig_data(ib); |
| 797 | memcpy(ob->p - ob->o, bo_ptr(ib), left); |
| 798 | if (ib->o - left) /* second part of the buffer */ |
| 799 | memcpy(ob->p - ob->o + left, ib->data, ib->o - left); |
| 800 | } |
| 801 | |
| 802 | /* chunked encoding requires CRLF after data */ |
| 803 | tail = ob->p + ob->i; |
| 804 | *tail++ = '\r'; |
| 805 | *tail++ = '\n'; |
| 806 | |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 807 | ob->i = tail - ob->p; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 808 | to_forward = ob->i; |
| 809 | |
| 810 | /* update input rate */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 811 | if (st->comp_ctx && st->comp_ctx->cur_lvl > 0) { |
| 812 | update_freq_ctr(&global.comp_bps_in, consumed); |
| 813 | strm_fe(s)->fe_counters.comp_in += consumed; |
| 814 | s->be->be_counters.comp_in += consumed; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 815 | } else { |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 816 | strm_fe(s)->fe_counters.comp_byp += consumed; |
| 817 | s->be->be_counters.comp_byp += consumed; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 818 | } |
| 819 | |
| 820 | /* copy the remaining data in the tmp buffer. */ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 821 | b_adv(ib, consumed); |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 822 | if (ib->i > 0) { |
| 823 | left = bi_contig_data(ib); |
| 824 | memcpy(ob->p + ob->i, bi_ptr(ib), left); |
| 825 | ob->i += left; |
| 826 | if (ib->i - left) { |
| 827 | memcpy(ob->p + ob->i, ib->data, ib->i - left); |
| 828 | ob->i += ib->i - left; |
| 829 | } |
| 830 | } |
| 831 | |
| 832 | /* swap the buffers */ |
| 833 | *in = ob; |
| 834 | *out = ib; |
| 835 | |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 836 | |
| 837 | if (st->comp_ctx && st->comp_ctx->cur_lvl > 0) { |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 838 | update_freq_ctr(&global.comp_bps_out, to_forward); |
| 839 | strm_fe(s)->fe_counters.comp_out += to_forward; |
| 840 | s->be->be_counters.comp_out += to_forward; |
| 841 | } |
| 842 | |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 843 | return to_forward; |
| 844 | } |
| 845 | |
| 846 | |
| 847 | /***********************************************************************/ |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 848 | struct flt_ops comp_ops = { |
| 849 | .init = comp_flt_init, |
| 850 | .deinit = comp_flt_deinit, |
| 851 | |
| 852 | .channel_start_analyze = comp_start_analyze, |
| 853 | .channel_analyze = comp_analyze, |
| 854 | .channel_end_analyze = comp_end_analyze, |
| 855 | |
| 856 | .http_headers = comp_http_headers, |
| 857 | .http_start_chunk = comp_skip_http_chunk_envelope, |
| 858 | .http_end_chunk = comp_skip_http_chunk_envelope, |
| 859 | .http_last_chunk = comp_skip_http_chunk_envelope, |
| 860 | .http_data = comp_http_data, |
| 861 | .http_forward_data = comp_http_forward_data, |
| 862 | }; |
| 863 | |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 864 | static int |
| 865 | parse_compression_options(char **args, int section, struct proxy *proxy, |
| 866 | struct proxy *defpx, const char *file, int line, |
| 867 | char **err) |
| 868 | { |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 869 | struct comp *comp; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 870 | |
| 871 | if (proxy->comp == NULL) { |
| 872 | comp = calloc(1, sizeof(struct comp)); |
| 873 | proxy->comp = comp; |
| 874 | } |
| 875 | else |
| 876 | comp = proxy->comp; |
| 877 | |
| 878 | if (!strcmp(args[1], "algo")) { |
| 879 | struct comp_ctx *ctx; |
| 880 | int cur_arg = 2; |
| 881 | |
| 882 | if (!*args[cur_arg]) { |
| 883 | memprintf(err, "parsing [%s:%d] : '%s' expects <algorithm>\n", |
| 884 | file, line, args[0]); |
| 885 | return -1; |
| 886 | } |
| 887 | while (*(args[cur_arg])) { |
| 888 | if (comp_append_algo(comp, args[cur_arg]) < 0) { |
| 889 | memprintf(err, "'%s' : '%s' is not a supported algorithm.\n", |
| 890 | args[0], args[cur_arg]); |
| 891 | return -1; |
| 892 | } |
| 893 | if (proxy->comp->algos->init(&ctx, 9) == 0) |
| 894 | proxy->comp->algos->end(&ctx); |
| 895 | else { |
| 896 | memprintf(err, "'%s' : Can't init '%s' algorithm.\n", |
| 897 | args[0], args[cur_arg]); |
| 898 | return -1; |
| 899 | } |
| 900 | cur_arg++; |
| 901 | continue; |
| 902 | } |
| 903 | } |
| 904 | else if (!strcmp(args[1], "offload")) |
| 905 | comp->offload = 1; |
| 906 | else if (!strcmp(args[1], "type")) { |
| 907 | int cur_arg = 2; |
| 908 | |
| 909 | if (!*args[cur_arg]) { |
| 910 | memprintf(err, "'%s' expects <type>\n", args[0]); |
| 911 | return -1; |
| 912 | } |
| 913 | while (*(args[cur_arg])) { |
| 914 | comp_append_type(comp, args[cur_arg]); |
| 915 | cur_arg++; |
| 916 | continue; |
| 917 | } |
| 918 | } |
| 919 | else { |
| 920 | memprintf(err, "'%s' expects 'algo', 'type' or 'offload'\n", |
| 921 | args[0]); |
| 922 | return -1; |
| 923 | } |
| 924 | |
| 925 | return 0; |
| 926 | } |
| 927 | |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 928 | static int |
| 929 | parse_http_comp_flt(char **args, int *cur_arg, struct proxy *px, |
| 930 | struct filter *filter, char **err) |
| 931 | { |
| 932 | struct filter *flt, *back; |
| 933 | |
| 934 | list_for_each_entry_safe(flt, back, &px->filters, list) { |
| 935 | if (flt->id == http_comp_flt_id) { |
| 936 | memprintf(err, "%s: Proxy supports only one compression filter\n", px->id); |
| 937 | return -1; |
| 938 | } |
| 939 | } |
| 940 | |
| 941 | filter->id = http_comp_flt_id; |
| 942 | filter->conf = NULL; |
| 943 | filter->ops = &comp_ops; |
| 944 | (*cur_arg)++; |
| 945 | |
| 946 | return 0; |
| 947 | } |
| 948 | |
| 949 | |
| 950 | int |
| 951 | check_legacy_http_comp_flt(struct proxy *proxy) |
| 952 | { |
| 953 | struct filter *filter; |
| 954 | int err = 0; |
| 955 | |
| 956 | if (proxy->comp == NULL) |
| 957 | goto end; |
| 958 | if (!LIST_ISEMPTY(&proxy->filters)) { |
| 959 | list_for_each_entry(filter, &proxy->filters, list) { |
| 960 | if (filter->id == http_comp_flt_id) |
| 961 | goto end; |
| 962 | } |
| 963 | Alert("config: %s '%s': require an explicit filter declaration to use HTTP compression\n", |
| 964 | proxy_type_str(proxy), proxy->id); |
| 965 | err++; |
| 966 | goto end; |
| 967 | } |
| 968 | |
| 969 | filter = pool_alloc2(pool2_filter); |
| 970 | if (!filter) { |
| 971 | Alert("config: %s '%s': out of memory\n", |
| 972 | proxy_type_str(proxy), proxy->id); |
| 973 | err++; |
| 974 | goto end; |
| 975 | } |
| 976 | memset(filter, 0, sizeof(*filter)); |
| 977 | filter->id = http_comp_flt_id; |
| 978 | filter->conf = NULL; |
| 979 | filter->ops = &comp_ops; |
| 980 | LIST_ADDQ(&proxy->filters, &filter->list); |
| 981 | |
| 982 | end: |
| 983 | return err; |
| 984 | } |
| 985 | |
| 986 | /* |
| 987 | * boolean, returns true if compression is used (either gzip or deflate) in the |
| 988 | * response. |
| 989 | */ |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 990 | static int |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 991 | smp_fetch_res_comp(const struct arg *args, struct sample *smp, const char *kw, |
| 992 | void *private) |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 993 | { |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 994 | struct http_txn *txn = smp->strm->txn; |
| 995 | |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 996 | smp->data.type = SMP_T_BOOL; |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 997 | smp->data.u.sint = (txn && (txn->rsp.flags & HTTP_MSGF_COMPRESSING)); |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 998 | return 1; |
| 999 | } |
| 1000 | |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 1001 | /* |
| 1002 | * string, returns algo |
| 1003 | */ |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 1004 | static int |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 1005 | smp_fetch_res_comp_algo(const struct arg *args, struct sample *smp, |
| 1006 | const char *kw, void *private) |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 1007 | { |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 1008 | struct http_txn *txn = smp->strm->txn; |
| 1009 | struct filter *filter; |
| 1010 | struct comp_state *st; |
| 1011 | |
| 1012 | if (!(txn || !(txn->rsp.flags & HTTP_MSGF_COMPRESSING))) |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 1013 | return 0; |
| 1014 | |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 1015 | list_for_each_entry(filter, &smp->strm->strm_flt.filters, list) { |
| 1016 | if (filter->id != http_comp_flt_id) |
| 1017 | continue; |
| 1018 | |
| 1019 | if (!(st = filter->ctx)) |
| 1020 | break; |
| 1021 | |
| 1022 | smp->data.type = SMP_T_STR; |
| 1023 | smp->flags = SMP_F_CONST; |
| 1024 | smp->data.u.str.str = st->comp_algo->cfg_name; |
| 1025 | smp->data.u.str.len = st->comp_algo->cfg_name_len; |
| 1026 | return 1; |
| 1027 | } |
| 1028 | return 0; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 1029 | } |
| 1030 | |
| 1031 | /* Declare the config parser for "compression" keyword */ |
| 1032 | static struct cfg_kw_list cfg_kws = {ILH, { |
| 1033 | { CFG_LISTEN, "compression", parse_compression_options }, |
| 1034 | { 0, NULL, NULL }, |
| 1035 | } |
| 1036 | }; |
| 1037 | |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 1038 | /* Declare the filter parser for "compression" keyword */ |
| 1039 | static struct flt_kw_list filter_kws = { "COMP", { }, { |
| 1040 | { "compression", parse_http_comp_flt }, |
| 1041 | { NULL, NULL }, |
| 1042 | } |
| 1043 | }; |
| 1044 | |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 1045 | /* Note: must not be declared <const> as its list will be overwritten */ |
| 1046 | static struct sample_fetch_kw_list sample_fetch_keywords = {ILH, { |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 1047 | { "res.comp", smp_fetch_res_comp, 0, NULL, SMP_T_BOOL, SMP_USE_HRSHP }, |
| 1048 | { "res.comp_algo", smp_fetch_res_comp_algo, 0, NULL, SMP_T_STR, SMP_USE_HRSHP }, |
| 1049 | { /* END */ }, |
| 1050 | } |
| 1051 | }; |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 1052 | |
| 1053 | __attribute__((constructor)) |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 1054 | static void |
| 1055 | __flt_http_comp_init(void) |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 1056 | { |
| 1057 | cfg_register_keywords(&cfg_kws); |
Christopher Faulet | 92d3638 | 2015-11-05 13:35:03 +0100 | [diff] [blame] | 1058 | flt_register_keywords(&filter_kws); |
Christopher Faulet | 3d97c90 | 2015-12-09 14:59:38 +0100 | [diff] [blame] | 1059 | sample_register_fetches(&sample_fetch_keywords); |
| 1060 | } |