William Lallemand | 82fe75c | 2012-10-23 10:25:10 +0200 | [diff] [blame] | 1 | /* |
| 2 | * HTTP compression. |
| 3 | * |
| 4 | * Copyright 2012 Exceliance, David Du Colombier <dducolombier@exceliance.fr> |
| 5 | * William Lallemand <wlallemand@exceliance.fr> |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or |
| 8 | * modify it under the terms of the GNU General Public License |
| 9 | * as published by the Free Software Foundation; either version |
| 10 | * 2 of the License, or (at your option) any later version. |
| 11 | * |
| 12 | */ |
| 13 | |
| 14 | #include <stdio.h> |
Willy Tarreau | 3476364 | 2012-10-26 15:05:35 +0200 | [diff] [blame^] | 15 | |
| 16 | /* Note: the crappy zlib and openssl libs both define the "free_func" type. |
| 17 | * That's a very clever idea to use such a generic name in general purpose |
| 18 | * libraries, really... The zlib one is easier to redefine than openssl's, |
| 19 | * so let's only fix this one. |
| 20 | */ |
| 21 | #define free_func zlib_free_func |
William Lallemand | 82fe75c | 2012-10-23 10:25:10 +0200 | [diff] [blame] | 22 | #include <zlib.h> |
Willy Tarreau | 3476364 | 2012-10-26 15:05:35 +0200 | [diff] [blame^] | 23 | #undef free_func |
William Lallemand | 82fe75c | 2012-10-23 10:25:10 +0200 | [diff] [blame] | 24 | |
| 25 | #include <common/compat.h> |
| 26 | |
| 27 | #include <types/global.h> |
| 28 | #include <types/compression.h> |
| 29 | |
| 30 | #include <proto/compression.h> |
| 31 | #include <proto/proto_http.h> |
| 32 | |
| 33 | static const struct comp_algo comp_algos[] = |
| 34 | { |
| 35 | { "identity", 8, identity_init, identity_add_data, identity_flush, identity_reset, identity_end }, |
| 36 | #ifdef USE_ZLIB |
| 37 | { "deflate", 7, deflate_init, deflate_add_data, deflate_flush, deflate_reset, deflate_end }, |
| 38 | { "gzip", 4, gzip_init, deflate_add_data, deflate_flush, deflate_reset, deflate_end }, |
| 39 | #endif /* USE_ZLIB */ |
| 40 | { NULL, 0, NULL , NULL, NULL, NULL, NULL } |
| 41 | }; |
| 42 | |
| 43 | /* |
| 44 | * Add a content-type in the configuration |
| 45 | */ |
| 46 | int comp_append_type(struct comp *comp, const char *type) |
| 47 | { |
| 48 | struct comp_type *comp_type; |
| 49 | |
| 50 | comp_type = calloc(1, sizeof(struct comp_type)); |
| 51 | comp_type->name_len = strlen(type); |
| 52 | comp_type->name = strdup(type); |
| 53 | comp_type->next = comp->types; |
| 54 | comp->types = comp_type; |
| 55 | return 0; |
| 56 | } |
| 57 | |
| 58 | /* |
| 59 | * Add an algorithm in the configuration |
| 60 | */ |
| 61 | int comp_append_algo(struct comp *comp, const char *algo) |
| 62 | { |
| 63 | struct comp_algo *comp_algo; |
| 64 | int i; |
| 65 | |
| 66 | for (i = 0; comp_algos[i].name; i++) { |
| 67 | if (!strcmp(algo, comp_algos[i].name)) { |
| 68 | comp_algo = calloc(1, sizeof(struct comp_algo)); |
| 69 | memmove(comp_algo, &comp_algos[i], sizeof(struct comp_algo)); |
| 70 | comp_algo->next = comp->algos; |
| 71 | comp->algos = comp_algo; |
| 72 | return 0; |
| 73 | } |
| 74 | } |
| 75 | return -1; |
| 76 | } |
| 77 | |
| 78 | /* emit the chunksize followed by a CRLF on the output and return the number of |
| 79 | * bytes written. Appends <add_crlf> additional CRLF after the first one. Chunk |
| 80 | * sizes are truncated to 6 hex digits (16 MB) and padded left. The caller is |
| 81 | * responsible for ensuring there is enough room left in the output buffer for |
| 82 | * the string (8 bytes * add_crlf*2). |
| 83 | */ |
| 84 | int http_emit_chunk_size(char *out, unsigned int chksz, int add_crlf) |
| 85 | { |
| 86 | int shift; |
| 87 | int pos = 0; |
| 88 | |
| 89 | for (shift = 20; shift >= 0; shift -= 4) |
| 90 | out[pos++] = hextab[(chksz >> shift) & 0xF]; |
| 91 | |
| 92 | do { |
| 93 | out[pos++] = '\r'; |
| 94 | out[pos++] = '\n'; |
| 95 | } while (--add_crlf >= 0); |
| 96 | |
| 97 | return pos; |
| 98 | } |
| 99 | |
| 100 | /* |
| 101 | * Init HTTP compression |
| 102 | */ |
| 103 | int http_compression_buffer_init(struct session *s, struct buffer *in, struct buffer *out) |
| 104 | { |
| 105 | struct http_msg *msg = &s->txn.rsp; |
| 106 | int left; |
| 107 | |
| 108 | /* not enough space */ |
| 109 | if (in->size - buffer_len(in) < 40) |
| 110 | return -1; |
| 111 | |
| 112 | /* |
| 113 | * Skip data, we don't need them in the new buffer. They are results |
| 114 | * of CHUNK_CRLF and CHUNK_SIZE parsing. |
| 115 | */ |
| 116 | b_adv(in, msg->next); |
| 117 | msg->next = 0; |
| 118 | msg->sov = 0; |
| 119 | msg->sol = 0; |
| 120 | |
| 121 | out->size = global.tune.bufsize; |
| 122 | out->i = 0; |
| 123 | out->o = 0; |
| 124 | out->p = out->data; |
| 125 | /* copy output data */ |
| 126 | if (in->o > 0) { |
| 127 | left = in->o - bo_contig_data(in); |
| 128 | memcpy(out->data, bo_ptr(in), bo_contig_data(in)); |
| 129 | out->p += bo_contig_data(in); |
| 130 | if (left > 0) { /* second part of the buffer */ |
| 131 | memcpy(out->p, in->data, left); |
| 132 | out->p += left; |
| 133 | } |
| 134 | out->o = in->o; |
| 135 | } |
| 136 | out->i += http_emit_chunk_size(out->p, 0, 0); |
| 137 | |
| 138 | return 0; |
| 139 | } |
| 140 | |
| 141 | /* |
| 142 | * Add data to compress |
| 143 | */ |
| 144 | int http_compression_buffer_add_data(struct session *s, struct buffer *in, struct buffer *out) |
| 145 | { |
| 146 | struct http_msg *msg = &s->txn.rsp; |
| 147 | int data_process_len; |
| 148 | int left; |
| 149 | int ret; |
| 150 | |
| 151 | /* |
| 152 | * Skip data, we don't need them in the new buffer. They are results |
| 153 | * of CHUNK_CRLF and CHUNK_SIZE parsing. |
| 154 | */ |
| 155 | b_adv(in, msg->next); |
| 156 | msg->next = 0; |
| 157 | msg->sov = 0; |
| 158 | msg->sol = 0; |
| 159 | |
| 160 | /* |
| 161 | * select the smallest size between the announced chunk size, the input |
| 162 | * data, and the available output buffer size |
| 163 | */ |
| 164 | data_process_len = MIN(in->i, msg->chunk_len); |
| 165 | data_process_len = MIN(out->size - buffer_len(out), data_process_len); |
| 166 | |
| 167 | left = data_process_len - bi_contig_data(in); |
| 168 | if (left <= 0) { |
| 169 | ret = s->comp_algo->add_data(&s->comp_ctx.strm, bi_ptr(in), |
| 170 | data_process_len, bi_end(out), |
| 171 | out->size - buffer_len(out)); |
| 172 | if (ret < 0) |
| 173 | return -1; |
| 174 | out->i += ret; |
| 175 | |
| 176 | } else { |
| 177 | ret = s->comp_algo->add_data(&s->comp_ctx.strm, bi_ptr(in), bi_contig_data(in), bi_end(out), out->size - buffer_len(out)); |
| 178 | if (ret < 0) |
| 179 | return -1; |
| 180 | out->i += ret; |
| 181 | ret = s->comp_algo->add_data(&s->comp_ctx.strm, in->data, left, bi_end(out), out->size - buffer_len(out)); |
| 182 | if (ret < 0) |
| 183 | return -1; |
| 184 | out->i += ret; |
| 185 | } |
| 186 | |
| 187 | b_adv(in, data_process_len); |
| 188 | msg->chunk_len -= data_process_len; |
| 189 | |
| 190 | return 0; |
| 191 | } |
| 192 | |
| 193 | /* |
| 194 | * Flush data in process, and write the header and footer of the chunk. Upon |
| 195 | * success, in and out buffers are swapped to avoid a copy. |
| 196 | */ |
| 197 | int http_compression_buffer_end(struct session *s, struct buffer **in, struct buffer **out, int end) |
| 198 | { |
| 199 | int to_forward; |
| 200 | int left; |
| 201 | struct http_msg *msg = &s->txn.rsp; |
| 202 | struct buffer *ib = *in, *ob = *out; |
| 203 | int ret; |
| 204 | |
| 205 | /* flush data here */ |
| 206 | |
| 207 | if (end) |
| 208 | ret = s->comp_algo->flush(&s->comp_ctx, ob, Z_FINISH); /* end of data */ |
| 209 | else |
| 210 | ret = s->comp_algo->flush(&s->comp_ctx, ob, Z_SYNC_FLUSH); /* end of buffer */ |
| 211 | |
| 212 | if (ret < 0) |
| 213 | return -1; /* flush failed */ |
| 214 | |
| 215 | if (ob->i > 8) { |
| 216 | /* more than a chunk size => some data were emitted */ |
| 217 | char *tail = ob->p + ob->i; |
| 218 | |
| 219 | /* write real size at the begining of the chunk, no need of wrapping */ |
| 220 | http_emit_chunk_size(ob->p, ob->i - 8, 0); |
| 221 | |
| 222 | /* chunked encoding requires CRLF after data */ |
| 223 | *tail++ = '\r'; |
| 224 | *tail++ = '\n'; |
| 225 | |
| 226 | if (!(msg->flags & HTTP_MSGF_TE_CHNK) && msg->chunk_len == 0) { |
| 227 | /* End of data, 0<CRLF><CRLF> is needed but we're not |
| 228 | * in chunked mode on input so we must add it ourselves. |
| 229 | */ |
| 230 | memcpy(tail, "0\r\n\r\n", 5); |
| 231 | tail += 5; |
| 232 | } |
| 233 | ob->i = tail - ob->p; |
| 234 | } else { |
| 235 | /* no data were sent, cancel the chunk size */ |
| 236 | ob->i = 0; |
| 237 | } |
| 238 | |
| 239 | to_forward = ob->i; |
| 240 | |
| 241 | /* copy the remaining data in the tmp buffer. */ |
| 242 | if (ib->i > 0) { |
| 243 | left = ib->i - bi_contig_data(ib); |
| 244 | memcpy(bi_end(ob), bi_ptr(ib), bi_contig_data(ib)); |
| 245 | ob->i += bi_contig_data(ib); |
| 246 | if (left > 0) { |
| 247 | memcpy(bi_end(ob), ib->data, left); |
| 248 | ob->i += left; |
| 249 | } |
| 250 | } |
| 251 | |
| 252 | /* swap the buffers */ |
| 253 | *in = ob; |
| 254 | *out = ib; |
| 255 | |
| 256 | /* forward the new chunk without remaining data */ |
| 257 | b_adv(ob, to_forward); |
| 258 | |
| 259 | /* if there are data between p and next, there are trailers, must forward them */ |
| 260 | b_adv(ob, msg->next); |
| 261 | msg->next = 0; |
| 262 | |
| 263 | return to_forward; |
| 264 | } |
| 265 | |
| 266 | |
| 267 | /**************************** |
| 268 | **** Identity algorithm **** |
| 269 | ****************************/ |
| 270 | |
| 271 | /* |
| 272 | * Init the identity algorithm |
| 273 | */ |
| 274 | int identity_init(void *v, int level) |
| 275 | { |
| 276 | return 0; |
| 277 | } |
| 278 | |
| 279 | /* |
| 280 | * Process data |
| 281 | * Return size of processed data or -1 on error |
| 282 | */ |
| 283 | int identity_add_data(void *comp_ctx, const char *in_data, int in_len, char *out_data, int out_len) |
| 284 | { |
| 285 | if (out_len < in_len) |
| 286 | return -1; |
| 287 | |
| 288 | memcpy(out_data, in_data, in_len); |
| 289 | |
| 290 | return in_len; |
| 291 | } |
| 292 | |
| 293 | int identity_flush(void *comp_ctx, struct buffer *out, int flag) |
| 294 | { |
| 295 | return 0; |
| 296 | } |
| 297 | |
| 298 | |
| 299 | int identity_reset(void *comp_ctx) |
| 300 | { |
| 301 | return 0; |
| 302 | } |
| 303 | |
| 304 | /* |
| 305 | * Deinit the algorithm |
| 306 | */ |
| 307 | int identity_end(void *comp_ctx) |
| 308 | { |
| 309 | return 0; |
| 310 | } |
| 311 | |
| 312 | |
| 313 | #ifdef USE_ZLIB |
| 314 | |
| 315 | /************************** |
| 316 | **** gzip algorithm **** |
| 317 | ***************************/ |
| 318 | int gzip_init(void *v, int level) |
| 319 | { |
| 320 | z_stream *strm; |
| 321 | |
| 322 | strm = v; |
| 323 | |
| 324 | strm->zalloc = Z_NULL; |
| 325 | strm->zfree = Z_NULL; |
| 326 | strm->opaque = Z_NULL; |
| 327 | |
Willy Tarreau | 7e488d7 | 2012-10-26 11:36:40 +0200 | [diff] [blame] | 328 | if (deflateInit2(strm, level, Z_DEFLATED, MAX_WBITS + 16, 8, Z_DEFAULT_STRATEGY) != Z_OK) |
William Lallemand | 82fe75c | 2012-10-23 10:25:10 +0200 | [diff] [blame] | 329 | return -1; |
| 330 | |
| 331 | return 0; |
| 332 | } |
| 333 | /************************** |
| 334 | **** Deflate algorithm **** |
| 335 | ***************************/ |
| 336 | |
| 337 | int deflate_init(void *comp_ctx, int level) |
| 338 | { |
| 339 | z_stream *strm; |
| 340 | |
| 341 | strm = comp_ctx; |
| 342 | |
| 343 | strm->zalloc = Z_NULL; |
| 344 | strm->zfree = Z_NULL; |
| 345 | strm->opaque = Z_NULL; |
| 346 | |
| 347 | if (deflateInit(strm, level) != Z_OK) |
| 348 | return -1; |
| 349 | |
| 350 | return 0; |
| 351 | } |
| 352 | |
| 353 | int deflate_add_data(void *comp_ctx, const char *in_data, int in_len, char *out_data, int out_len) |
| 354 | { |
| 355 | z_stream *strm; |
| 356 | int ret; |
| 357 | |
| 358 | if (in_len <= 0) |
| 359 | return 0; |
| 360 | |
| 361 | |
| 362 | if (out_len <= 0) |
| 363 | return -1; |
| 364 | |
| 365 | strm = comp_ctx; |
| 366 | |
| 367 | strm->next_in = (unsigned char *)in_data; |
| 368 | strm->avail_in = in_len; |
| 369 | strm->next_out = (unsigned char *)out_data; |
| 370 | strm->avail_out = out_len; |
| 371 | |
| 372 | ret = deflate(strm, Z_NO_FLUSH); |
| 373 | if (ret != Z_OK) |
| 374 | return -1; |
| 375 | |
| 376 | /* deflate update the available data out */ |
| 377 | |
| 378 | return out_len - strm->avail_out; |
| 379 | } |
| 380 | |
| 381 | int deflate_flush(void *comp_ctx, struct buffer *out, int flag) |
| 382 | { |
| 383 | int ret; |
| 384 | z_stream *strm; |
| 385 | int out_len = 0; |
| 386 | |
| 387 | strm = comp_ctx; |
| 388 | strm->next_out = (unsigned char *)bi_end(out); |
| 389 | strm->avail_out = out->size - buffer_len(out); |
| 390 | |
| 391 | ret = deflate(strm, flag); |
| 392 | if (ret != Z_OK && ret != Z_STREAM_END) |
| 393 | return -1; |
| 394 | |
| 395 | out_len = (out->size - buffer_len(out)) - strm->avail_out; |
| 396 | out->i += out_len; |
| 397 | |
| 398 | return out_len; |
| 399 | } |
| 400 | |
| 401 | int deflate_reset(void *comp_ctx) |
| 402 | { |
| 403 | z_stream *strm; |
| 404 | |
| 405 | strm = comp_ctx; |
| 406 | if (deflateReset(strm) == Z_OK) |
| 407 | return 0; |
| 408 | return -1; |
| 409 | } |
| 410 | |
| 411 | int deflate_end(void *comp_ctx) |
| 412 | { |
| 413 | z_stream *strm; |
| 414 | |
| 415 | strm = comp_ctx; |
| 416 | if (deflateEnd(strm) == Z_OK) |
| 417 | return 0; |
| 418 | |
| 419 | return -1; |
| 420 | } |
| 421 | |
| 422 | #endif /* USE_ZLIB */ |
| 423 | |