Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 1 | /* |
| 2 | * httpread - Manage reading file(s) from HTTP/TCP socket |
| 3 | * Author: Ted Merrill |
| 4 | * Copyright 2008 Atheros Communications |
| 5 | * |
Dmitry Shmidt | c5ec7f5 | 2012-03-06 16:33:24 -0800 | [diff] [blame] | 6 | * This software may be distributed under the terms of the BSD license. |
| 7 | * See README for more details. |
Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 8 | * |
| 9 | * The files are buffered via internal callbacks from eloop, then presented to |
| 10 | * an application callback routine when completely read into memory. May also |
| 11 | * be used if no file is expected but just to get the header, including HTTP |
| 12 | * replies (e.g. HTTP/1.1 200 OK etc.). |
| 13 | * |
| 14 | * This does not attempt to be an optimally efficient implementation, but does |
| 15 | * attempt to be of reasonably small size and memory consumption; assuming that |
| 16 | * only small files are to be read. A maximum file size is provided by |
| 17 | * application and enforced. |
| 18 | * |
| 19 | * It is assumed that the application does not expect any of the following: |
| 20 | * -- transfer encoding other than chunked |
| 21 | * -- trailer fields |
| 22 | * It is assumed that, even if the other side requested that the connection be |
| 23 | * kept open, that we will close it (thus HTTP messages sent by application |
| 24 | * should have the connection closed field); this is allowed by HTTP/1.1 and |
| 25 | * simplifies things for us. |
| 26 | * |
| 27 | * Other limitations: |
| 28 | * -- HTTP header may not exceed a hard-coded size. |
| 29 | * |
| 30 | * Notes: |
| 31 | * This code would be massively simpler without some of the new features of |
| 32 | * HTTP/1.1, especially chunked data. |
| 33 | */ |
| 34 | |
| 35 | #include "includes.h" |
| 36 | |
| 37 | #include "common.h" |
| 38 | #include "eloop.h" |
| 39 | #include "httpread.h" |
| 40 | |
| 41 | |
| 42 | /* Tunable parameters */ |
| 43 | #define HTTPREAD_READBUF_SIZE 1024 /* read in chunks of this size */ |
| 44 | #define HTTPREAD_HEADER_MAX_SIZE 4096 /* max allowed for headers */ |
| 45 | #define HTTPREAD_BODYBUF_DELTA 4096 /* increase allocation by this */ |
| 46 | |
| 47 | #if 0 |
| 48 | /* httpread_debug -- set this global variable > 0 e.g. from debugger |
| 49 | * to enable debugs (larger numbers for more debugs) |
| 50 | * Make this a #define of 0 to eliminate the debugging code. |
| 51 | */ |
| 52 | int httpread_debug = 99; |
| 53 | #else |
| 54 | #define httpread_debug 0 /* eliminates even the debugging code */ |
| 55 | #endif |
| 56 | |
| 57 | |
| 58 | /* control instance -- actual definition (opaque to application) |
| 59 | */ |
| 60 | struct httpread { |
| 61 | /* information from creation */ |
| 62 | int sd; /* descriptor of TCP socket to read from */ |
| 63 | void (*cb)(struct httpread *handle, void *cookie, |
| 64 | enum httpread_event e); /* call on event */ |
| 65 | void *cookie; /* pass to callback */ |
| 66 | int max_bytes; /* maximum file size else abort it */ |
| 67 | int timeout_seconds; /* 0 or total duration timeout period */ |
| 68 | |
| 69 | /* dynamically used information follows */ |
Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 70 | |
| 71 | int got_hdr; /* nonzero when header is finalized */ |
| 72 | char hdr[HTTPREAD_HEADER_MAX_SIZE+1]; /* headers stored here */ |
| 73 | int hdr_nbytes; |
| 74 | |
| 75 | enum httpread_hdr_type hdr_type; |
| 76 | int version; /* 1 if we've seen 1.1 */ |
| 77 | int reply_code; /* for type REPLY, e.g. 200 for HTTP/1.1 200 OK */ |
| 78 | int got_content_length; /* true if we know content length for sure */ |
| 79 | int content_length; /* body length, iff got_content_length */ |
| 80 | int chunked; /* nonzero for chunked data */ |
| 81 | char *uri; |
| 82 | |
| 83 | int got_body; /* nonzero when body is finalized */ |
| 84 | char *body; |
| 85 | int body_nbytes; |
| 86 | int body_alloc_nbytes; /* amount allocated */ |
| 87 | |
| 88 | int got_file; /* here when we are done */ |
| 89 | |
| 90 | /* The following apply if data is chunked: */ |
| 91 | int in_chunk_data; /* 0=in/at header, 1=in the data or tail*/ |
| 92 | int chunk_start; /* offset in body of chunk hdr or data */ |
| 93 | int chunk_size; /* data of chunk (not hdr or ending CRLF)*/ |
| 94 | int in_trailer; /* in header fields after data (chunked only)*/ |
| 95 | enum trailer_state { |
| 96 | trailer_line_begin = 0, |
| 97 | trailer_empty_cr, /* empty line + CR */ |
| 98 | trailer_nonempty, |
| 99 | trailer_nonempty_cr, |
| 100 | } trailer_state; |
| 101 | }; |
| 102 | |
| 103 | |
| 104 | /* Check words for equality, where words consist of graphical characters |
| 105 | * delimited by whitespace |
| 106 | * Returns nonzero if "equal" doing case insensitive comparison. |
| 107 | */ |
| 108 | static int word_eq(char *s1, char *s2) |
| 109 | { |
| 110 | int c1; |
| 111 | int c2; |
| 112 | int end1 = 0; |
| 113 | int end2 = 0; |
| 114 | for (;;) { |
| 115 | c1 = *s1++; |
| 116 | c2 = *s2++; |
| 117 | if (isalpha(c1) && isupper(c1)) |
| 118 | c1 = tolower(c1); |
| 119 | if (isalpha(c2) && isupper(c2)) |
| 120 | c2 = tolower(c2); |
| 121 | end1 = !isgraph(c1); |
| 122 | end2 = !isgraph(c2); |
| 123 | if (end1 || end2 || c1 != c2) |
| 124 | break; |
| 125 | } |
| 126 | return end1 && end2; /* reached end of both words? */ |
| 127 | } |
| 128 | |
| 129 | |
Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 130 | static void httpread_timeout_handler(void *eloop_data, void *user_ctx); |
| 131 | |
| 132 | /* httpread_destroy -- if h is non-NULL, clean up |
| 133 | * This must eventually be called by the application following |
| 134 | * call of the application's callback and may be called |
| 135 | * earlier if desired. |
| 136 | */ |
| 137 | void httpread_destroy(struct httpread *h) |
| 138 | { |
| 139 | if (httpread_debug >= 10) |
| 140 | wpa_printf(MSG_DEBUG, "ENTER httpread_destroy(%p)", h); |
| 141 | if (!h) |
| 142 | return; |
| 143 | |
Dmitry Shmidt | fb79edc | 2014-01-10 10:45:54 -0800 | [diff] [blame] | 144 | eloop_cancel_timeout(httpread_timeout_handler, NULL, h); |
| 145 | eloop_unregister_sock(h->sd, EVENT_TYPE_READ); |
Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 146 | os_free(h->body); |
| 147 | os_free(h->uri); |
| 148 | os_memset(h, 0, sizeof(*h)); /* aid debugging */ |
| 149 | h->sd = -1; /* aid debugging */ |
| 150 | os_free(h); |
| 151 | } |
| 152 | |
| 153 | |
| 154 | /* httpread_timeout_handler -- called on excessive total duration |
| 155 | */ |
| 156 | static void httpread_timeout_handler(void *eloop_data, void *user_ctx) |
| 157 | { |
| 158 | struct httpread *h = user_ctx; |
| 159 | wpa_printf(MSG_DEBUG, "httpread timeout (%p)", h); |
Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 160 | (*h->cb)(h, h->cookie, HTTPREAD_EVENT_TIMEOUT); |
| 161 | } |
| 162 | |
| 163 | |
| 164 | /* Analyze options only so far as is needed to correctly obtain the file. |
| 165 | * The application can look at the raw header to find other options. |
| 166 | */ |
| 167 | static int httpread_hdr_option_analyze( |
| 168 | struct httpread *h, |
| 169 | char *hbp /* pointer to current line in header buffer */ |
| 170 | ) |
| 171 | { |
| 172 | if (word_eq(hbp, "CONTENT-LENGTH:")) { |
| 173 | while (isgraph(*hbp)) |
| 174 | hbp++; |
| 175 | while (*hbp == ' ' || *hbp == '\t') |
| 176 | hbp++; |
| 177 | if (!isdigit(*hbp)) |
| 178 | return -1; |
| 179 | h->content_length = atol(hbp); |
| 180 | h->got_content_length = 1; |
| 181 | return 0; |
| 182 | } |
| 183 | if (word_eq(hbp, "TRANSFER_ENCODING:") || |
| 184 | word_eq(hbp, "TRANSFER-ENCODING:")) { |
| 185 | while (isgraph(*hbp)) |
| 186 | hbp++; |
| 187 | while (*hbp == ' ' || *hbp == '\t') |
| 188 | hbp++; |
| 189 | /* There should (?) be no encodings of interest |
| 190 | * other than chunked... |
| 191 | */ |
| 192 | if (word_eq(hbp, "CHUNKED")) { |
| 193 | h->chunked = 1; |
| 194 | h->in_chunk_data = 0; |
| 195 | /* ignore possible ;<parameters> */ |
| 196 | } |
| 197 | return 0; |
| 198 | } |
| 199 | /* skip anything we don't know, which is a lot */ |
| 200 | return 0; |
| 201 | } |
| 202 | |
| 203 | |
| 204 | static int httpread_hdr_analyze(struct httpread *h) |
| 205 | { |
| 206 | char *hbp = h->hdr; /* pointer into h->hdr */ |
| 207 | int standard_first_line = 1; |
| 208 | |
| 209 | /* First line is special */ |
| 210 | h->hdr_type = HTTPREAD_HDR_TYPE_UNKNOWN; |
| 211 | if (!isgraph(*hbp)) |
| 212 | goto bad; |
| 213 | if (os_strncmp(hbp, "HTTP/", 5) == 0) { |
| 214 | h->hdr_type = HTTPREAD_HDR_TYPE_REPLY; |
| 215 | standard_first_line = 0; |
| 216 | hbp += 5; |
| 217 | if (hbp[0] == '1' && hbp[1] == '.' && |
| 218 | isdigit(hbp[2]) && hbp[2] != '0') |
| 219 | h->version = 1; |
| 220 | while (isgraph(*hbp)) |
| 221 | hbp++; |
| 222 | while (*hbp == ' ' || *hbp == '\t') |
| 223 | hbp++; |
| 224 | if (!isdigit(*hbp)) |
| 225 | goto bad; |
| 226 | h->reply_code = atol(hbp); |
| 227 | } else if (word_eq(hbp, "GET")) |
| 228 | h->hdr_type = HTTPREAD_HDR_TYPE_GET; |
| 229 | else if (word_eq(hbp, "HEAD")) |
| 230 | h->hdr_type = HTTPREAD_HDR_TYPE_HEAD; |
| 231 | else if (word_eq(hbp, "POST")) |
| 232 | h->hdr_type = HTTPREAD_HDR_TYPE_POST; |
| 233 | else if (word_eq(hbp, "PUT")) |
| 234 | h->hdr_type = HTTPREAD_HDR_TYPE_PUT; |
| 235 | else if (word_eq(hbp, "DELETE")) |
| 236 | h->hdr_type = HTTPREAD_HDR_TYPE_DELETE; |
| 237 | else if (word_eq(hbp, "TRACE")) |
| 238 | h->hdr_type = HTTPREAD_HDR_TYPE_TRACE; |
| 239 | else if (word_eq(hbp, "CONNECT")) |
| 240 | h->hdr_type = HTTPREAD_HDR_TYPE_CONNECT; |
| 241 | else if (word_eq(hbp, "NOTIFY")) |
| 242 | h->hdr_type = HTTPREAD_HDR_TYPE_NOTIFY; |
| 243 | else if (word_eq(hbp, "M-SEARCH")) |
| 244 | h->hdr_type = HTTPREAD_HDR_TYPE_M_SEARCH; |
| 245 | else if (word_eq(hbp, "M-POST")) |
| 246 | h->hdr_type = HTTPREAD_HDR_TYPE_M_POST; |
| 247 | else if (word_eq(hbp, "SUBSCRIBE")) |
| 248 | h->hdr_type = HTTPREAD_HDR_TYPE_SUBSCRIBE; |
| 249 | else if (word_eq(hbp, "UNSUBSCRIBE")) |
| 250 | h->hdr_type = HTTPREAD_HDR_TYPE_UNSUBSCRIBE; |
| 251 | else { |
| 252 | } |
| 253 | |
| 254 | if (standard_first_line) { |
| 255 | char *rawuri; |
| 256 | char *uri; |
| 257 | /* skip type */ |
| 258 | while (isgraph(*hbp)) |
| 259 | hbp++; |
| 260 | while (*hbp == ' ' || *hbp == '\t') |
| 261 | hbp++; |
| 262 | /* parse uri. |
| 263 | * Find length, allocate memory for translated |
| 264 | * copy, then translate by changing %<hex><hex> |
| 265 | * into represented value. |
| 266 | */ |
| 267 | rawuri = hbp; |
| 268 | while (isgraph(*hbp)) |
| 269 | hbp++; |
| 270 | h->uri = os_malloc((hbp - rawuri) + 1); |
| 271 | if (h->uri == NULL) |
| 272 | goto bad; |
| 273 | uri = h->uri; |
| 274 | while (rawuri < hbp) { |
| 275 | int c = *rawuri; |
| 276 | if (c == '%' && |
| 277 | isxdigit(rawuri[1]) && isxdigit(rawuri[2])) { |
Dmitry Shmidt | fb79edc | 2014-01-10 10:45:54 -0800 | [diff] [blame] | 278 | *uri++ = hex2byte(rawuri + 1); |
Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 279 | rawuri += 3; |
| 280 | } else { |
| 281 | *uri++ = c; |
| 282 | rawuri++; |
| 283 | } |
| 284 | } |
| 285 | *uri = 0; /* null terminate */ |
| 286 | while (isgraph(*hbp)) |
| 287 | hbp++; |
| 288 | while (*hbp == ' ' || *hbp == '\t') |
| 289 | hbp++; |
| 290 | /* get version */ |
| 291 | if (0 == strncmp(hbp, "HTTP/", 5)) { |
| 292 | hbp += 5; |
| 293 | if (hbp[0] == '1' && hbp[1] == '.' && |
| 294 | isdigit(hbp[2]) && hbp[2] != '0') |
| 295 | h->version = 1; |
| 296 | } |
| 297 | } |
| 298 | /* skip rest of line */ |
| 299 | while (*hbp) |
| 300 | if (*hbp++ == '\n') |
| 301 | break; |
| 302 | |
| 303 | /* Remainder of lines are options, in any order; |
| 304 | * or empty line to terminate |
| 305 | */ |
| 306 | for (;;) { |
| 307 | /* Empty line to terminate */ |
| 308 | if (hbp[0] == '\n' || |
| 309 | (hbp[0] == '\r' && hbp[1] == '\n')) |
| 310 | break; |
| 311 | if (!isgraph(*hbp)) |
| 312 | goto bad; |
| 313 | if (httpread_hdr_option_analyze(h, hbp)) |
| 314 | goto bad; |
| 315 | /* skip line */ |
| 316 | while (*hbp) |
| 317 | if (*hbp++ == '\n') |
| 318 | break; |
| 319 | } |
| 320 | |
| 321 | /* chunked overrides content-length always */ |
| 322 | if (h->chunked) |
| 323 | h->got_content_length = 0; |
| 324 | |
| 325 | /* For some types, we should not try to read a body |
| 326 | * This is in addition to the application determining |
| 327 | * that we should not read a body. |
| 328 | */ |
| 329 | switch (h->hdr_type) { |
| 330 | case HTTPREAD_HDR_TYPE_REPLY: |
| 331 | /* Some codes can have a body and some not. |
| 332 | * For now, just assume that any other than 200 |
| 333 | * do not... |
| 334 | */ |
| 335 | if (h->reply_code != 200) |
| 336 | h->max_bytes = 0; |
| 337 | break; |
| 338 | case HTTPREAD_HDR_TYPE_GET: |
| 339 | case HTTPREAD_HDR_TYPE_HEAD: |
| 340 | /* in practice it appears that it is assumed |
| 341 | * that GETs have a body length of 0... ? |
| 342 | */ |
| 343 | if (h->chunked == 0 && h->got_content_length == 0) |
| 344 | h->max_bytes = 0; |
| 345 | break; |
| 346 | case HTTPREAD_HDR_TYPE_POST: |
| 347 | case HTTPREAD_HDR_TYPE_PUT: |
| 348 | case HTTPREAD_HDR_TYPE_DELETE: |
| 349 | case HTTPREAD_HDR_TYPE_TRACE: |
| 350 | case HTTPREAD_HDR_TYPE_CONNECT: |
| 351 | case HTTPREAD_HDR_TYPE_NOTIFY: |
| 352 | case HTTPREAD_HDR_TYPE_M_SEARCH: |
| 353 | case HTTPREAD_HDR_TYPE_M_POST: |
| 354 | case HTTPREAD_HDR_TYPE_SUBSCRIBE: |
| 355 | case HTTPREAD_HDR_TYPE_UNSUBSCRIBE: |
| 356 | default: |
| 357 | break; |
| 358 | } |
| 359 | |
| 360 | return 0; |
| 361 | |
| 362 | bad: |
| 363 | /* Error */ |
| 364 | return -1; |
| 365 | } |
| 366 | |
| 367 | |
| 368 | /* httpread_read_handler -- called when socket ready to read |
| 369 | * |
| 370 | * Note: any extra data we read past end of transmitted file is ignored; |
| 371 | * if we were to support keeping connections open for multiple files then |
| 372 | * this would have to be addressed. |
| 373 | */ |
| 374 | static void httpread_read_handler(int sd, void *eloop_ctx, void *sock_ctx) |
| 375 | { |
| 376 | struct httpread *h = sock_ctx; |
| 377 | int nread; |
| 378 | char *rbp; /* pointer into read buffer */ |
| 379 | char *hbp; /* pointer into header buffer */ |
| 380 | char *bbp; /* pointer into body buffer */ |
| 381 | char readbuf[HTTPREAD_READBUF_SIZE]; /* temp use to read into */ |
| 382 | |
| 383 | if (httpread_debug >= 20) |
| 384 | wpa_printf(MSG_DEBUG, "ENTER httpread_read_handler(%p)", h); |
| 385 | |
| 386 | /* read some at a time, then search for the interal |
| 387 | * boundaries between header and data and etc. |
| 388 | */ |
| 389 | nread = read(h->sd, readbuf, sizeof(readbuf)); |
| 390 | if (nread < 0) |
| 391 | goto bad; |
| 392 | if (nread == 0) { |
| 393 | /* end of transmission... this may be normal |
| 394 | * or may be an error... in some cases we can't |
| 395 | * tell which so we must assume it is normal then. |
| 396 | */ |
| 397 | if (!h->got_hdr) { |
| 398 | /* Must at least have completed header */ |
| 399 | wpa_printf(MSG_DEBUG, "httpread premature eof(%p)", h); |
| 400 | goto bad; |
| 401 | } |
| 402 | if (h->chunked || h->got_content_length) { |
| 403 | /* Premature EOF; e.g. dropped connection */ |
| 404 | wpa_printf(MSG_DEBUG, |
| 405 | "httpread premature eof(%p) %d/%d", |
| 406 | h, h->body_nbytes, |
| 407 | h->content_length); |
| 408 | goto bad; |
| 409 | } |
| 410 | /* No explicit length, hopefully we have all the data |
| 411 | * although dropped connections can cause false |
| 412 | * end |
| 413 | */ |
| 414 | if (httpread_debug >= 10) |
| 415 | wpa_printf(MSG_DEBUG, "httpread ok eof(%p)", h); |
Dmitry Shmidt | 623d63a | 2014-06-13 11:05:14 -0700 | [diff] [blame] | 416 | h->got_body = 1; |
| 417 | goto got_file; |
Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 418 | } |
| 419 | rbp = readbuf; |
| 420 | |
| 421 | /* Header consists of text lines (terminated by both CR and LF) |
| 422 | * and an empty line (CR LF only). |
| 423 | */ |
| 424 | if (!h->got_hdr) { |
| 425 | hbp = h->hdr + h->hdr_nbytes; |
| 426 | /* add to headers until: |
| 427 | * -- we run out of data in read buffer |
| 428 | * -- or, we run out of header buffer room |
| 429 | * -- or, we get double CRLF in headers |
| 430 | */ |
| 431 | for (;;) { |
| 432 | if (nread == 0) |
| 433 | goto get_more; |
| 434 | if (h->hdr_nbytes == HTTPREAD_HEADER_MAX_SIZE) { |
| 435 | goto bad; |
| 436 | } |
| 437 | *hbp++ = *rbp++; |
| 438 | nread--; |
| 439 | h->hdr_nbytes++; |
| 440 | if (h->hdr_nbytes >= 4 && |
| 441 | hbp[-1] == '\n' && |
| 442 | hbp[-2] == '\r' && |
| 443 | hbp[-3] == '\n' && |
| 444 | hbp[-4] == '\r' ) { |
| 445 | h->got_hdr = 1; |
| 446 | *hbp = 0; /* null terminate */ |
| 447 | break; |
| 448 | } |
| 449 | } |
| 450 | /* here we've just finished reading the header */ |
| 451 | if (httpread_hdr_analyze(h)) { |
| 452 | wpa_printf(MSG_DEBUG, "httpread bad hdr(%p)", h); |
| 453 | goto bad; |
| 454 | } |
| 455 | if (h->max_bytes == 0) { |
| 456 | if (httpread_debug >= 10) |
| 457 | wpa_printf(MSG_DEBUG, |
| 458 | "httpread no body hdr end(%p)", h); |
| 459 | goto got_file; |
| 460 | } |
| 461 | if (h->got_content_length && h->content_length == 0) { |
| 462 | if (httpread_debug >= 10) |
| 463 | wpa_printf(MSG_DEBUG, |
| 464 | "httpread zero content length(%p)", |
| 465 | h); |
| 466 | goto got_file; |
| 467 | } |
| 468 | } |
| 469 | |
| 470 | /* Certain types of requests never have data and so |
| 471 | * must be specially recognized. |
| 472 | */ |
| 473 | if (!os_strncasecmp(h->hdr, "SUBSCRIBE", 9) || |
| 474 | !os_strncasecmp(h->hdr, "UNSUBSCRIBE", 11) || |
| 475 | !os_strncasecmp(h->hdr, "HEAD", 4) || |
| 476 | !os_strncasecmp(h->hdr, "GET", 3)) { |
| 477 | if (!h->got_body) { |
| 478 | if (httpread_debug >= 10) |
| 479 | wpa_printf(MSG_DEBUG, |
| 480 | "httpread NO BODY for sp. type"); |
| 481 | } |
| 482 | h->got_body = 1; |
| 483 | goto got_file; |
| 484 | } |
| 485 | |
| 486 | /* Data can be just plain binary data, or if "chunked" |
| 487 | * consists of chunks each with a header, ending with |
| 488 | * an ending header. |
| 489 | */ |
| 490 | if (nread == 0) |
| 491 | goto get_more; |
| 492 | if (!h->got_body) { |
| 493 | /* Here to get (more of) body */ |
| 494 | /* ensure we have enough room for worst case for body |
| 495 | * plus a null termination character |
| 496 | */ |
| 497 | if (h->body_alloc_nbytes < (h->body_nbytes + nread + 1)) { |
| 498 | char *new_body; |
| 499 | int new_alloc_nbytes; |
| 500 | |
| 501 | if (h->body_nbytes >= h->max_bytes) |
| 502 | goto bad; |
| 503 | new_alloc_nbytes = h->body_alloc_nbytes + |
| 504 | HTTPREAD_BODYBUF_DELTA; |
| 505 | /* For content-length case, the first time |
| 506 | * through we allocate the whole amount |
| 507 | * we need. |
| 508 | */ |
| 509 | if (h->got_content_length && |
| 510 | new_alloc_nbytes < (h->content_length + 1)) |
| 511 | new_alloc_nbytes = h->content_length + 1; |
| 512 | if ((new_body = os_realloc(h->body, new_alloc_nbytes)) |
| 513 | == NULL) |
| 514 | goto bad; |
| 515 | |
| 516 | h->body = new_body; |
| 517 | h->body_alloc_nbytes = new_alloc_nbytes; |
| 518 | } |
| 519 | /* add bytes */ |
| 520 | bbp = h->body + h->body_nbytes; |
| 521 | for (;;) { |
| 522 | int ncopy; |
| 523 | /* See if we need to stop */ |
| 524 | if (h->chunked && h->in_chunk_data == 0) { |
| 525 | /* in chunk header */ |
| 526 | char *cbp = h->body + h->chunk_start; |
| 527 | if (bbp-cbp >= 2 && bbp[-2] == '\r' && |
| 528 | bbp[-1] == '\n') { |
| 529 | /* end of chunk hdr line */ |
| 530 | /* hdr line consists solely |
| 531 | * of a hex numeral and CFLF |
| 532 | */ |
| 533 | if (!isxdigit(*cbp)) |
| 534 | goto bad; |
| 535 | h->chunk_size = strtoul(cbp, NULL, 16); |
| 536 | /* throw away chunk header |
| 537 | * so we have only real data |
| 538 | */ |
| 539 | h->body_nbytes = h->chunk_start; |
| 540 | bbp = cbp; |
| 541 | if (h->chunk_size == 0) { |
| 542 | /* end of chunking */ |
| 543 | /* trailer follows */ |
| 544 | h->in_trailer = 1; |
| 545 | if (httpread_debug >= 20) |
| 546 | wpa_printf( |
| 547 | MSG_DEBUG, |
| 548 | "httpread end chunks(%p)", h); |
| 549 | break; |
| 550 | } |
| 551 | h->in_chunk_data = 1; |
| 552 | /* leave chunk_start alone */ |
| 553 | } |
| 554 | } else if (h->chunked) { |
| 555 | /* in chunk data */ |
| 556 | if ((h->body_nbytes - h->chunk_start) == |
| 557 | (h->chunk_size + 2)) { |
| 558 | /* end of chunk reached, |
| 559 | * new chunk starts |
| 560 | */ |
| 561 | /* check chunk ended w/ CRLF |
| 562 | * which we'll throw away |
| 563 | */ |
| 564 | if (bbp[-1] == '\n' && |
| 565 | bbp[-2] == '\r') { |
| 566 | } else |
| 567 | goto bad; |
| 568 | h->body_nbytes -= 2; |
| 569 | bbp -= 2; |
| 570 | h->chunk_start = h->body_nbytes; |
| 571 | h->in_chunk_data = 0; |
| 572 | h->chunk_size = 0; /* just in case */ |
| 573 | } |
| 574 | } else if (h->got_content_length && |
| 575 | h->body_nbytes >= h->content_length) { |
| 576 | h->got_body = 1; |
| 577 | if (httpread_debug >= 10) |
| 578 | wpa_printf( |
| 579 | MSG_DEBUG, |
| 580 | "httpread got content(%p)", h); |
| 581 | goto got_file; |
| 582 | } |
| 583 | if (nread <= 0) |
| 584 | break; |
| 585 | /* Now transfer. Optimize using memcpy where we can. */ |
| 586 | if (h->chunked && h->in_chunk_data) { |
| 587 | /* copy up to remainder of chunk data |
| 588 | * plus the required CR+LF at end |
| 589 | */ |
| 590 | ncopy = (h->chunk_start + h->chunk_size + 2) - |
| 591 | h->body_nbytes; |
| 592 | } else if (h->chunked) { |
| 593 | /*in chunk header -- don't optimize */ |
| 594 | *bbp++ = *rbp++; |
| 595 | nread--; |
| 596 | h->body_nbytes++; |
| 597 | continue; |
| 598 | } else if (h->got_content_length) { |
| 599 | ncopy = h->content_length - h->body_nbytes; |
| 600 | } else { |
| 601 | ncopy = nread; |
| 602 | } |
| 603 | /* Note: should never be 0 */ |
| 604 | if (ncopy > nread) |
| 605 | ncopy = nread; |
| 606 | os_memcpy(bbp, rbp, ncopy); |
| 607 | bbp += ncopy; |
| 608 | h->body_nbytes += ncopy; |
| 609 | rbp += ncopy; |
| 610 | nread -= ncopy; |
| 611 | } /* body copy loop */ |
| 612 | } /* !got_body */ |
| 613 | if (h->chunked && h->in_trailer) { |
| 614 | /* If "chunked" then there is always a trailer, |
| 615 | * consisting of zero or more non-empty lines |
| 616 | * ending with CR LF and then an empty line w/ CR LF. |
| 617 | * We do NOT support trailers except to skip them -- |
| 618 | * this is supported (generally) by the http spec. |
| 619 | */ |
Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 620 | for (;;) { |
| 621 | int c; |
| 622 | if (nread <= 0) |
| 623 | break; |
| 624 | c = *rbp++; |
| 625 | nread--; |
| 626 | switch (h->trailer_state) { |
| 627 | case trailer_line_begin: |
| 628 | if (c == '\r') |
| 629 | h->trailer_state = trailer_empty_cr; |
| 630 | else |
| 631 | h->trailer_state = trailer_nonempty; |
| 632 | break; |
| 633 | case trailer_empty_cr: |
| 634 | /* end empty line */ |
| 635 | if (c == '\n') { |
| 636 | h->trailer_state = trailer_line_begin; |
| 637 | h->in_trailer = 0; |
| 638 | if (httpread_debug >= 10) |
| 639 | wpa_printf( |
| 640 | MSG_DEBUG, |
| 641 | "httpread got content(%p)", h); |
| 642 | h->got_body = 1; |
| 643 | goto got_file; |
| 644 | } |
| 645 | h->trailer_state = trailer_nonempty; |
| 646 | break; |
| 647 | case trailer_nonempty: |
| 648 | if (c == '\r') |
| 649 | h->trailer_state = trailer_nonempty_cr; |
| 650 | break; |
| 651 | case trailer_nonempty_cr: |
| 652 | if (c == '\n') |
| 653 | h->trailer_state = trailer_line_begin; |
| 654 | else |
| 655 | h->trailer_state = trailer_nonempty; |
| 656 | break; |
| 657 | } |
| 658 | } |
| 659 | } |
| 660 | goto get_more; |
| 661 | |
| 662 | bad: |
| 663 | /* Error */ |
| 664 | wpa_printf(MSG_DEBUG, "httpread read/parse failure (%p)", h); |
| 665 | (*h->cb)(h, h->cookie, HTTPREAD_EVENT_ERROR); |
| 666 | return; |
| 667 | |
| 668 | get_more: |
| 669 | return; |
| 670 | |
| 671 | got_file: |
| 672 | if (httpread_debug >= 10) |
| 673 | wpa_printf(MSG_DEBUG, |
| 674 | "httpread got file %d bytes type %d", |
| 675 | h->body_nbytes, h->hdr_type); |
| 676 | /* Null terminate for convenience of some applications */ |
| 677 | if (h->body) |
| 678 | h->body[h->body_nbytes] = 0; /* null terminate */ |
| 679 | h->got_file = 1; |
| 680 | /* Assume that we do NOT support keeping connection alive, |
| 681 | * and just in case somehow we don't get destroyed right away, |
| 682 | * unregister now. |
| 683 | */ |
Dmitry Shmidt | fb79edc | 2014-01-10 10:45:54 -0800 | [diff] [blame] | 684 | eloop_unregister_sock(h->sd, EVENT_TYPE_READ); |
Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 685 | /* The application can destroy us whenever they feel like... |
| 686 | * cancel timeout. |
| 687 | */ |
Dmitry Shmidt | fb79edc | 2014-01-10 10:45:54 -0800 | [diff] [blame] | 688 | eloop_cancel_timeout(httpread_timeout_handler, NULL, h); |
Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 689 | (*h->cb)(h, h->cookie, HTTPREAD_EVENT_FILE_READY); |
| 690 | } |
| 691 | |
| 692 | |
| 693 | /* httpread_create -- start a new reading session making use of eloop. |
| 694 | * The new instance will use the socket descriptor for reading (until |
| 695 | * it gets a file and not after) but will not close the socket, even |
| 696 | * when the instance is destroyed (the application must do that). |
| 697 | * Return NULL on error. |
| 698 | * |
| 699 | * Provided that httpread_create successfully returns a handle, |
| 700 | * the callback fnc is called to handle httpread_event events. |
| 701 | * The caller should do destroy on any errors or unknown events. |
| 702 | * |
| 703 | * Pass max_bytes == 0 to not read body at all (required for e.g. |
| 704 | * reply to HEAD request). |
| 705 | */ |
| 706 | struct httpread * httpread_create( |
| 707 | int sd, /* descriptor of TCP socket to read from */ |
| 708 | void (*cb)(struct httpread *handle, void *cookie, |
| 709 | enum httpread_event e), /* call on event */ |
| 710 | void *cookie, /* pass to callback */ |
| 711 | int max_bytes, /* maximum body size else abort it */ |
| 712 | int timeout_seconds /* 0; or total duration timeout period */ |
| 713 | ) |
| 714 | { |
| 715 | struct httpread *h = NULL; |
| 716 | |
| 717 | h = os_zalloc(sizeof(*h)); |
| 718 | if (h == NULL) |
| 719 | goto fail; |
| 720 | h->sd = sd; |
| 721 | h->cb = cb; |
| 722 | h->cookie = cookie; |
| 723 | h->max_bytes = max_bytes; |
| 724 | h->timeout_seconds = timeout_seconds; |
| 725 | |
Dmitry Shmidt | fb79edc | 2014-01-10 10:45:54 -0800 | [diff] [blame] | 726 | if (timeout_seconds > 0 && |
| 727 | eloop_register_timeout(timeout_seconds, 0, |
| 728 | httpread_timeout_handler, NULL, h)) { |
| 729 | /* No way to recover (from malloc failure) */ |
| 730 | goto fail; |
Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 731 | } |
| 732 | if (eloop_register_sock(sd, EVENT_TYPE_READ, httpread_read_handler, |
| 733 | NULL, h)) { |
| 734 | /* No way to recover (from malloc failure) */ |
| 735 | goto fail; |
| 736 | } |
Dmitry Shmidt | 8d520ff | 2011-05-09 14:06:53 -0700 | [diff] [blame] | 737 | return h; |
| 738 | |
| 739 | fail: |
| 740 | |
| 741 | /* Error */ |
| 742 | httpread_destroy(h); |
| 743 | return NULL; |
| 744 | } |
| 745 | |
| 746 | |
| 747 | /* httpread_hdr_type_get -- When file is ready, returns header type. */ |
| 748 | enum httpread_hdr_type httpread_hdr_type_get(struct httpread *h) |
| 749 | { |
| 750 | return h->hdr_type; |
| 751 | } |
| 752 | |
| 753 | |
| 754 | /* httpread_uri_get -- When file is ready, uri_get returns (translated) URI |
| 755 | * or possibly NULL (which would be an error). |
| 756 | */ |
| 757 | char * httpread_uri_get(struct httpread *h) |
| 758 | { |
| 759 | return h->uri; |
| 760 | } |
| 761 | |
| 762 | |
| 763 | /* httpread_reply_code_get -- When reply is ready, returns reply code */ |
| 764 | int httpread_reply_code_get(struct httpread *h) |
| 765 | { |
| 766 | return h->reply_code; |
| 767 | } |
| 768 | |
| 769 | |
| 770 | /* httpread_length_get -- When file is ready, returns file length. */ |
| 771 | int httpread_length_get(struct httpread *h) |
| 772 | { |
| 773 | return h->body_nbytes; |
| 774 | } |
| 775 | |
| 776 | |
| 777 | /* httpread_data_get -- When file is ready, returns file content |
| 778 | * with null byte appened. |
| 779 | * Might return NULL in some error condition. |
| 780 | */ |
| 781 | void * httpread_data_get(struct httpread *h) |
| 782 | { |
| 783 | return h->body ? h->body : ""; |
| 784 | } |
| 785 | |
| 786 | |
| 787 | /* httpread_hdr_get -- When file is ready, returns header content |
| 788 | * with null byte appended. |
| 789 | * Might return NULL in some error condition. |
| 790 | */ |
| 791 | char * httpread_hdr_get(struct httpread *h) |
| 792 | { |
| 793 | return h->hdr; |
| 794 | } |
| 795 | |
| 796 | |
| 797 | /* httpread_hdr_line_get -- When file is ready, returns pointer |
| 798 | * to line within header content matching the given tag |
| 799 | * (after the tag itself and any spaces/tabs). |
| 800 | * |
| 801 | * The tag should end with a colon for reliable matching. |
| 802 | * |
| 803 | * If not found, returns NULL; |
| 804 | */ |
| 805 | char * httpread_hdr_line_get(struct httpread *h, const char *tag) |
| 806 | { |
| 807 | int tag_len = os_strlen(tag); |
| 808 | char *hdr = h->hdr; |
| 809 | hdr = os_strchr(hdr, '\n'); |
| 810 | if (hdr == NULL) |
| 811 | return NULL; |
| 812 | hdr++; |
| 813 | for (;;) { |
| 814 | if (!os_strncasecmp(hdr, tag, tag_len)) { |
| 815 | hdr += tag_len; |
| 816 | while (*hdr == ' ' || *hdr == '\t') |
| 817 | hdr++; |
| 818 | return hdr; |
| 819 | } |
| 820 | hdr = os_strchr(hdr, '\n'); |
| 821 | if (hdr == NULL) |
| 822 | return NULL; |
| 823 | hdr++; |
| 824 | } |
| 825 | } |