Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 1 | /* |
| 2 | * Functions to manipulate H1 messages using the internal representation. |
| 3 | * |
| 4 | * Copyright (C) 2019 HAProxy Technologies, Christopher Faulet <cfaulet@haproxy.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 | |
Willy Tarreau | 4c7e4b7 | 2020-05-27 12:58:42 +0200 | [diff] [blame] | 13 | #include <haproxy/api.h> |
Willy Tarreau | 6be7849 | 2020-06-05 00:00:29 +0200 | [diff] [blame] | 14 | #include <haproxy/cfgparse.h> |
Willy Tarreau | dfd3de8 | 2020-06-04 23:46:14 +0200 | [diff] [blame] | 15 | #include <haproxy/global.h> |
Willy Tarreau | 5413a87 | 2020-06-02 19:33:08 +0200 | [diff] [blame] | 16 | #include <haproxy/h1.h> |
Willy Tarreau | c6fe884 | 2020-06-04 09:00:02 +0200 | [diff] [blame] | 17 | #include <haproxy/h1_htx.h> |
Willy Tarreau | cd72d8c | 2020-06-02 19:11:26 +0200 | [diff] [blame] | 18 | #include <haproxy/http.h> |
Willy Tarreau | 16f958c | 2020-06-03 08:44:35 +0200 | [diff] [blame] | 19 | #include <haproxy/htx.h> |
Willy Tarreau | 36979d9 | 2020-06-05 17:27:29 +0200 | [diff] [blame] | 20 | #include <haproxy/tools.h> |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 21 | |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 22 | /* Estimate the size of the HTX headers after the parsing, including the EOH. */ |
| 23 | static size_t h1_eval_htx_hdrs_size(const struct http_hdr *hdrs) |
| 24 | { |
| 25 | size_t sz = 0; |
| 26 | int i; |
| 27 | |
| 28 | for (i = 0; hdrs[i].n.len; i++) |
| 29 | sz += sizeof(struct htx_blk) + hdrs[i].n.len + hdrs[i].v.len; |
| 30 | sz += sizeof(struct htx_blk) + 1; |
| 31 | return sz; |
| 32 | } |
| 33 | |
| 34 | /* Estimate the size of the HTX request after the parsing. */ |
| 35 | static size_t h1_eval_htx_size(const struct ist p1, const struct ist p2, const struct ist p3, |
| 36 | const struct http_hdr *hdrs) |
| 37 | { |
| 38 | size_t sz; |
| 39 | |
| 40 | /* size of the HTX start-line */ |
| 41 | sz = sizeof(struct htx_blk) + sizeof(struct htx_sl) + p1.len + p2.len + p3.len; |
| 42 | sz += h1_eval_htx_hdrs_size(hdrs); |
| 43 | return sz; |
| 44 | } |
| 45 | |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 46 | /* Check the validity of the request version. If the version is valid, it |
| 47 | * returns 1. Otherwise, it returns 0. |
| 48 | */ |
| 49 | static int h1_process_req_vsn(struct h1m *h1m, union h1_sl *sl) |
| 50 | { |
| 51 | /* RFC7230#2.6 has enforced the format of the HTTP version string to be |
| 52 | * exactly one digit "." one digit. This check may be disabled using |
| 53 | * option accept-invalid-http-request. |
| 54 | */ |
| 55 | if (h1m->err_pos == -2) { /* PR_O2_REQBUG_OK not set */ |
| 56 | if (sl->rq.v.len != 8) |
| 57 | return 0; |
| 58 | |
| 59 | if (*(sl->rq.v.ptr + 4) != '/' || |
| 60 | !isdigit((unsigned char)*(sl->rq.v.ptr + 5)) || |
| 61 | *(sl->rq.v.ptr + 6) != '.' || |
| 62 | !isdigit((unsigned char)*(sl->rq.v.ptr + 7))) |
| 63 | return 0; |
| 64 | } |
| 65 | else if (!sl->rq.v.len) { |
| 66 | /* try to convert HTTP/0.9 requests to HTTP/1.0 */ |
| 67 | |
| 68 | /* RFC 1945 allows only GET for HTTP/0.9 requests */ |
| 69 | if (sl->rq.meth != HTTP_METH_GET) |
| 70 | return 0; |
| 71 | |
| 72 | /* HTTP/0.9 requests *must* have a request URI, per RFC 1945 */ |
| 73 | if (!sl->rq.u.len) |
| 74 | return 0; |
| 75 | |
| 76 | /* Add HTTP version */ |
| 77 | sl->rq.v = ist("HTTP/1.0"); |
| 78 | return 1; |
| 79 | } |
| 80 | |
| 81 | if ((sl->rq.v.len == 8) && |
| 82 | ((*(sl->rq.v.ptr + 5) > '1') || |
| 83 | ((*(sl->rq.v.ptr + 5) == '1') && (*(sl->rq.v.ptr + 7) >= '1')))) |
| 84 | h1m->flags |= H1_MF_VER_11; |
| 85 | return 1; |
| 86 | } |
| 87 | |
| 88 | /* Check the validity of the response version. If the version is valid, it |
| 89 | * returns 1. Otherwise, it returns 0. |
| 90 | */ |
| 91 | static int h1_process_res_vsn(struct h1m *h1m, union h1_sl *sl) |
| 92 | { |
| 93 | /* RFC7230#2.6 has enforced the format of the HTTP version string to be |
| 94 | * exactly one digit "." one digit. This check may be disabled using |
| 95 | * option accept-invalid-http-request. |
| 96 | */ |
| 97 | if (h1m->err_pos == -2) { /* PR_O2_REQBUG_OK not set */ |
| 98 | if (sl->st.v.len != 8) |
| 99 | return 0; |
| 100 | |
| 101 | if (*(sl->st.v.ptr + 4) != '/' || |
| 102 | !isdigit((unsigned char)*(sl->st.v.ptr + 5)) || |
| 103 | *(sl->st.v.ptr + 6) != '.' || |
| 104 | !isdigit((unsigned char)*(sl->st.v.ptr + 7))) |
| 105 | return 0; |
| 106 | } |
| 107 | |
| 108 | if ((sl->st.v.len == 8) && |
| 109 | ((*(sl->st.v.ptr + 5) > '1') || |
| 110 | ((*(sl->st.v.ptr + 5) == '1') && (*(sl->st.v.ptr + 7) >= '1')))) |
| 111 | h1m->flags |= H1_MF_VER_11; |
| 112 | |
| 113 | return 1; |
| 114 | } |
| 115 | |
| 116 | /* Convert H1M flags to HTX start-line flags. */ |
| 117 | static unsigned int h1m_htx_sl_flags(struct h1m *h1m) |
| 118 | { |
| 119 | unsigned int flags = HTX_SL_F_NONE; |
| 120 | |
| 121 | if (h1m->flags & H1_MF_RESP) |
| 122 | flags |= HTX_SL_F_IS_RESP; |
| 123 | if (h1m->flags & H1_MF_VER_11) |
| 124 | flags |= HTX_SL_F_VER_11; |
| 125 | if (h1m->flags & H1_MF_XFER_ENC) |
| 126 | flags |= HTX_SL_F_XFER_ENC; |
| 127 | if (h1m->flags & H1_MF_XFER_LEN) { |
| 128 | flags |= HTX_SL_F_XFER_LEN; |
| 129 | if (h1m->flags & H1_MF_CHNK) |
| 130 | flags |= HTX_SL_F_CHNK; |
| 131 | else if (h1m->flags & H1_MF_CLEN) { |
| 132 | flags |= HTX_SL_F_CLEN; |
| 133 | if (h1m->body_len == 0) |
| 134 | flags |= HTX_SL_F_BODYLESS; |
| 135 | } |
| 136 | else |
| 137 | flags |= HTX_SL_F_BODYLESS; |
| 138 | } |
Christopher Faulet | 576c358 | 2021-01-08 15:53:01 +0100 | [diff] [blame] | 139 | if (h1m->flags & H1_MF_CONN_UPG) |
| 140 | flags |= HTX_SL_F_CONN_UPG; |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 141 | return flags; |
| 142 | } |
| 143 | |
| 144 | /* Postprocess the parsed headers for a request and convert them into an htx |
| 145 | * message. It returns the number of bytes parsed if > 0, or 0 if it couldn't |
| 146 | * proceed. Parsing errors are reported by setting the htx flag |
| 147 | * HTX_FL_PARSING_ERROR and filling h1m->err_pos and h1m->err_state fields. |
| 148 | */ |
| 149 | static int h1_postparse_req_hdrs(struct h1m *h1m, union h1_sl *h1sl, struct htx *htx, |
| 150 | struct http_hdr *hdrs, size_t max) |
| 151 | { |
| 152 | struct htx_sl *sl; |
| 153 | struct ist meth, uri, vsn; |
| 154 | unsigned int flags; |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 155 | |
| 156 | /* <h1sl> is always defined for a request */ |
| 157 | meth = h1sl->rq.m; |
| 158 | uri = h1sl->rq.u; |
| 159 | vsn = h1sl->rq.v; |
| 160 | |
| 161 | /* Be sure the message, once converted into HTX, will not exceed the max |
| 162 | * size allowed. |
| 163 | */ |
| 164 | if (h1_eval_htx_size(meth, uri, vsn, hdrs) > max) { |
| 165 | if (htx_is_empty(htx)) |
| 166 | goto error; |
| 167 | h1m_init_res(h1m); |
| 168 | h1m->flags |= (H1_MF_NO_PHDR|H1_MF_CLEAN_CONN_HDR); |
| 169 | return 0; |
| 170 | } |
| 171 | |
| 172 | /* By default, request have always a known length */ |
| 173 | h1m->flags |= H1_MF_XFER_LEN; |
| 174 | |
| 175 | if (h1sl->rq.meth == HTTP_METH_CONNECT) { |
Christopher Faulet | 5be651d | 2021-01-22 15:28:03 +0100 | [diff] [blame] | 176 | h1m->flags &= ~(H1_MF_CLEN|H1_MF_CHNK); |
| 177 | h1m->curr_len = h1m->body_len = 0; |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 178 | } |
| 179 | |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 180 | flags = h1m_htx_sl_flags(h1m); |
| 181 | sl = htx_add_stline(htx, HTX_BLK_REQ_SL, flags, meth, uri, vsn); |
| 182 | if (!sl || !htx_add_all_headers(htx, hdrs)) |
| 183 | goto error; |
| 184 | sl->info.req.meth = h1sl->rq.meth; |
| 185 | |
Christopher Faulet | fe451fb | 2019-10-08 15:01:34 +0200 | [diff] [blame] | 186 | /* Check if the uri contains an authority. Also check if it contains an |
| 187 | * explicit scheme and if it is "http" or "https". */ |
| 188 | if (h1sl->rq.meth == HTTP_METH_CONNECT) |
| 189 | sl->flags |= HTX_SL_F_HAS_AUTHORITY; |
| 190 | else if (uri.len && uri.ptr[0] != '/' && uri.ptr[0] != '*') { |
| 191 | sl->flags |= (HTX_SL_F_HAS_AUTHORITY|HTX_SL_F_HAS_SCHM); |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 192 | if (uri.len > 4 && (uri.ptr[0] | 0x20) == 'h') |
| 193 | sl->flags |= ((uri.ptr[4] == ':') ? HTX_SL_F_SCHM_HTTP : HTX_SL_F_SCHM_HTTPS); |
| 194 | } |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 195 | |
| 196 | /* If body length cannot be determined, set htx->extra to |
| 197 | * ULLONG_MAX. This value is impossible in other cases. |
| 198 | */ |
| 199 | htx->extra = ((h1m->flags & H1_MF_XFER_LEN) ? h1m->curr_len : ULLONG_MAX); |
| 200 | |
| 201 | end: |
| 202 | return 1; |
| 203 | error: |
| 204 | h1m->err_pos = h1m->next; |
| 205 | h1m->err_state = h1m->state; |
| 206 | htx->flags |= HTX_FL_PARSING_ERROR; |
| 207 | return 0; |
| 208 | } |
| 209 | |
| 210 | /* Postprocess the parsed headers for a response and convert them into an htx |
| 211 | * message. It returns the number of bytes parsed if > 0, or 0 if it couldn't |
| 212 | * proceed. Parsing errors are reported by setting the htx flag |
| 213 | * HTX_FL_PARSING_ERROR and filling h1m->err_pos and h1m->err_state fields. |
| 214 | */ |
| 215 | static int h1_postparse_res_hdrs(struct h1m *h1m, union h1_sl *h1sl, struct htx *htx, |
| 216 | struct http_hdr *hdrs, size_t max) |
| 217 | { |
| 218 | struct htx_sl *sl; |
| 219 | struct ist vsn, status, reason; |
| 220 | unsigned int flags; |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 221 | uint16_t code = 0; |
| 222 | |
| 223 | if (h1sl) { |
| 224 | /* For HTTP responses, the start-line was parsed */ |
| 225 | code = h1sl->st.status; |
| 226 | vsn = h1sl->st.v; |
| 227 | status = h1sl->st.c; |
| 228 | reason = h1sl->st.r; |
| 229 | } |
| 230 | else { |
| 231 | /* For FCGI responses, there is no start(-line but the "Status" |
| 232 | * header must be parsed, if found. |
| 233 | */ |
| 234 | int hdr; |
| 235 | |
| 236 | vsn = ((h1m->flags & H1_MF_VER_11) ? ist("HTTP/1.1") : ist("HTTP/1.0")); |
| 237 | for (hdr = 0; hdrs[hdr].n.len; hdr++) { |
| 238 | if (isteqi(hdrs[hdr].n, ist("status"))) { |
| 239 | code = http_parse_status_val(hdrs[hdr].v, &status, &reason); |
| 240 | } |
| 241 | else if (isteqi(hdrs[hdr].n, ist("location"))) { |
| 242 | code = 302; |
| 243 | status = ist("302"); |
| 244 | reason = ist("Moved Temporarily"); |
| 245 | } |
| 246 | } |
| 247 | if (!code) { |
| 248 | code = 200; |
| 249 | status = ist("200"); |
| 250 | reason = ist("OK"); |
| 251 | } |
| 252 | /* FIXME: Check the codes 1xx ? */ |
| 253 | } |
| 254 | |
| 255 | /* Be sure the message, once converted into HTX, will not exceed the max |
| 256 | * size allowed. |
| 257 | */ |
| 258 | if (h1_eval_htx_size(vsn, status, reason, hdrs) > max) { |
| 259 | if (htx_is_empty(htx)) |
| 260 | goto error; |
| 261 | h1m_init_res(h1m); |
| 262 | h1m->flags |= (H1_MF_NO_PHDR|H1_MF_CLEAN_CONN_HDR); |
| 263 | return 0; |
| 264 | } |
| 265 | |
Christopher Faulet | c75668e | 2020-12-07 18:10:32 +0100 | [diff] [blame] | 266 | if (((h1m->flags & H1_MF_METH_CONNECT) && code >= 200 && code < 300) || code == 101) { |
Christopher Faulet | 5be651d | 2021-01-22 15:28:03 +0100 | [diff] [blame] | 267 | h1m->flags &= ~(H1_MF_CLEN|H1_MF_CHNK); |
| 268 | h1m->flags |= H1_MF_XFER_LEN; |
| 269 | h1m->curr_len = h1m->body_len = 0; |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 270 | } |
| 271 | else if ((h1m->flags & H1_MF_METH_HEAD) || (code >= 100 && code < 200) || |
| 272 | (code == 204) || (code == 304)) { |
| 273 | /* Responses known to have no body. */ |
| 274 | h1m->flags &= ~(H1_MF_CLEN|H1_MF_CHNK); |
| 275 | h1m->flags |= H1_MF_XFER_LEN; |
| 276 | h1m->curr_len = h1m->body_len = 0; |
| 277 | } |
| 278 | else if (h1m->flags & (H1_MF_CLEN|H1_MF_CHNK)) { |
| 279 | /* Responses with a known body length. */ |
| 280 | h1m->flags |= H1_MF_XFER_LEN; |
| 281 | } |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 282 | |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 283 | flags = h1m_htx_sl_flags(h1m); |
| 284 | sl = htx_add_stline(htx, HTX_BLK_RES_SL, flags, vsn, status, reason); |
| 285 | if (!sl || !htx_add_all_headers(htx, hdrs)) |
| 286 | goto error; |
| 287 | sl->info.res.status = code; |
| 288 | |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 289 | /* If body length cannot be determined, set htx->extra to |
| 290 | * ULLONG_MAX. This value is impossible in other cases. |
| 291 | */ |
| 292 | htx->extra = ((h1m->flags & H1_MF_XFER_LEN) ? h1m->curr_len : ULLONG_MAX); |
| 293 | |
| 294 | end: |
| 295 | return 1; |
| 296 | error: |
| 297 | h1m->err_pos = h1m->next; |
| 298 | h1m->err_state = h1m->state; |
| 299 | htx->flags |= HTX_FL_PARSING_ERROR; |
| 300 | return 0; |
| 301 | } |
| 302 | |
| 303 | /* Parse HTTP/1 headers. It returns the number of bytes parsed if > 0, or 0 if |
| 304 | * it couldn't proceed. Parsing errors are reported by setting the htx flag |
| 305 | * HTX_FL_PARSING_ERROR and filling h1m->err_pos and h1m->err_state fields. This |
| 306 | * functions is responsible to update the parser state <h1m> and the start-line |
| 307 | * <h1sl> if not NULL. |
| 308 | * For the requests, <h1sl> must always be provided. For responses, <h1sl> may |
| 309 | * be NULL and <h1m> flags HTTP_METH_CONNECT of HTTP_METH_HEAD may be set. |
| 310 | */ |
| 311 | int h1_parse_msg_hdrs(struct h1m *h1m, union h1_sl *h1sl, struct htx *dsthtx, |
| 312 | struct buffer *srcbuf, size_t ofs, size_t max) |
| 313 | { |
| 314 | struct http_hdr hdrs[global.tune.max_http_hdr]; |
| 315 | int ret = 0; |
| 316 | |
| 317 | if (!max || !b_data(srcbuf)) |
| 318 | goto end; |
| 319 | |
| 320 | /* Realing input buffer if necessary */ |
| 321 | if (b_head(srcbuf) + b_data(srcbuf) > b_wrap(srcbuf)) |
| 322 | b_slow_realign(srcbuf, trash.area, 0); |
| 323 | |
| 324 | if (!h1sl) { |
| 325 | /* If there no start-line, be sure to only parse the headers */ |
| 326 | h1m->flags |= H1_MF_HDRS_ONLY; |
| 327 | } |
| 328 | ret = h1_headers_to_hdr_list(b_peek(srcbuf, ofs), b_tail(srcbuf), |
| 329 | hdrs, sizeof(hdrs)/sizeof(hdrs[0]), h1m, h1sl); |
| 330 | if (ret <= 0) { |
| 331 | /* Incomplete or invalid message. If the input buffer only |
| 332 | * contains headers and is full, which is detected by it being |
| 333 | * full and the offset to be zero, it's an error because |
| 334 | * headers are too large to be handled by the parser. */ |
| 335 | if (ret < 0 || (!ret && !ofs && !buf_room_for_htx_data(srcbuf))) |
| 336 | goto error; |
| 337 | goto end; |
| 338 | } |
| 339 | |
| 340 | /* messages headers fully parsed, do some checks to prepare the body |
| 341 | * parsing. |
| 342 | */ |
| 343 | |
| 344 | if (!(h1m->flags & H1_MF_RESP)) { |
| 345 | if (!h1_process_req_vsn(h1m, h1sl)) { |
| 346 | h1m->err_pos = h1sl->rq.v.ptr - b_head(srcbuf); |
| 347 | h1m->err_state = h1m->state; |
| 348 | goto vsn_error; |
| 349 | } |
| 350 | if (!h1_postparse_req_hdrs(h1m, h1sl, dsthtx, hdrs, max)) |
| 351 | ret = 0; |
| 352 | } |
| 353 | else { |
| 354 | if (h1sl && !h1_process_res_vsn(h1m, h1sl)) { |
| 355 | h1m->err_pos = h1sl->st.v.ptr - b_head(srcbuf); |
| 356 | h1m->err_state = h1m->state; |
| 357 | goto vsn_error; |
| 358 | } |
| 359 | if (!h1_postparse_res_hdrs(h1m, h1sl, dsthtx, hdrs, max)) |
| 360 | ret = 0; |
| 361 | } |
| 362 | |
Christopher Faulet | 76014fd | 2019-12-10 11:47:22 +0100 | [diff] [blame] | 363 | /* Switch messages without any payload to DONE state */ |
| 364 | if (((h1m->flags & H1_MF_CLEN) && h1m->body_len == 0) || |
Christopher Faulet | d1ac2b9 | 2020-12-02 19:12:22 +0100 | [diff] [blame] | 365 | ((h1m->flags & (H1_MF_XFER_LEN|H1_MF_CLEN|H1_MF_CHNK)) == H1_MF_XFER_LEN)) { |
Christopher Faulet | 76014fd | 2019-12-10 11:47:22 +0100 | [diff] [blame] | 366 | h1m->state = H1_MSG_DONE; |
Christopher Faulet | d1ac2b9 | 2020-12-02 19:12:22 +0100 | [diff] [blame] | 367 | dsthtx->flags |= HTX_FL_EOM; |
| 368 | } |
Christopher Faulet | 76014fd | 2019-12-10 11:47:22 +0100 | [diff] [blame] | 369 | |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 370 | end: |
| 371 | return ret; |
| 372 | error: |
| 373 | h1m->err_pos = h1m->next; |
| 374 | h1m->err_state = h1m->state; |
| 375 | vsn_error: |
| 376 | dsthtx->flags |= HTX_FL_PARSING_ERROR; |
| 377 | return 0; |
| 378 | |
| 379 | } |
| 380 | |
Christopher Faulet | cc3124c | 2019-08-12 22:42:21 +0200 | [diff] [blame] | 381 | /* Copy data from <srbuf> into an DATA block in <dsthtx>. If possible, a |
| 382 | * zero-copy is performed. It returns the number of bytes copied. |
| 383 | */ |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 384 | static int h1_copy_msg_data(struct htx **dsthtx, struct buffer *srcbuf, size_t ofs, |
Christopher Faulet | cc3124c | 2019-08-12 22:42:21 +0200 | [diff] [blame] | 385 | size_t count, struct buffer *htxbuf) |
| 386 | { |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 387 | struct htx *tmp_htx = *dsthtx; |
| 388 | |
Christopher Faulet | cc3124c | 2019-08-12 22:42:21 +0200 | [diff] [blame] | 389 | /* very often with large files we'll face the following |
| 390 | * situation : |
| 391 | * - htx is empty and points to <htxbuf> |
| 392 | * - ret == srcbuf->data |
| 393 | * - srcbuf->head == sizeof(struct htx) |
| 394 | * => we can swap the buffers and place an htx header into |
| 395 | * the target buffer instead |
| 396 | */ |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 397 | if (unlikely(htx_is_empty(tmp_htx) && count == b_data(srcbuf) && |
Christopher Faulet | cc3124c | 2019-08-12 22:42:21 +0200 | [diff] [blame] | 398 | !ofs && b_head_ofs(srcbuf) == sizeof(struct htx))) { |
| 399 | void *raw_area = srcbuf->area; |
| 400 | void *htx_area = htxbuf->area; |
| 401 | struct htx_blk *blk; |
| 402 | |
| 403 | srcbuf->area = htx_area; |
| 404 | htxbuf->area = raw_area; |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 405 | tmp_htx = (struct htx *)htxbuf->area; |
| 406 | tmp_htx->size = htxbuf->size - sizeof(*tmp_htx); |
| 407 | htx_reset(tmp_htx); |
Christopher Faulet | cc3124c | 2019-08-12 22:42:21 +0200 | [diff] [blame] | 408 | b_set_data(htxbuf, b_size(htxbuf)); |
| 409 | |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 410 | blk = htx_add_blk(tmp_htx, HTX_BLK_DATA, count); |
Christopher Faulet | cc3124c | 2019-08-12 22:42:21 +0200 | [diff] [blame] | 411 | blk->info += count; |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 412 | |
| 413 | *dsthtx = tmp_htx; |
Christopher Faulet | cc3124c | 2019-08-12 22:42:21 +0200 | [diff] [blame] | 414 | /* nothing else to do, the old buffer now contains an |
| 415 | * empty pre-initialized HTX header |
| 416 | */ |
| 417 | return count; |
| 418 | } |
| 419 | |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 420 | return htx_add_data(*dsthtx, ist2(b_peek(srcbuf, ofs), count)); |
Christopher Faulet | cc3124c | 2019-08-12 22:42:21 +0200 | [diff] [blame] | 421 | } |
| 422 | |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 423 | /* Parse HTTP/1 body. It returns the number of bytes parsed if > 0, or 0 if it |
| 424 | * couldn't proceed. Parsing errors are reported by setting the htx flags |
| 425 | * HTX_FL_PARSING_ERROR and filling h1m->err_pos and h1m->err_state fields. This |
| 426 | * functions is responsible to update the parser state <h1m>. |
| 427 | */ |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 428 | int h1_parse_msg_data(struct h1m *h1m, struct htx **dsthtx, |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 429 | struct buffer *srcbuf, size_t ofs, size_t max, |
| 430 | struct buffer *htxbuf) |
| 431 | { |
| 432 | size_t total = 0; |
| 433 | int32_t ret = 0; |
| 434 | |
| 435 | if (h1m->flags & H1_MF_CLEN) { |
| 436 | /* content-length: read only h2m->body_len */ |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 437 | ret = htx_get_max_blksz(*dsthtx, max); |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 438 | if ((uint64_t)ret > h1m->curr_len) |
| 439 | ret = h1m->curr_len; |
| 440 | if (ret > b_contig_data(srcbuf, ofs)) |
| 441 | ret = b_contig_data(srcbuf, ofs); |
| 442 | if (ret) { |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 443 | int32_t try = ret; |
| 444 | |
Christopher Faulet | cc3124c | 2019-08-12 22:42:21 +0200 | [diff] [blame] | 445 | ret = h1_copy_msg_data(dsthtx, srcbuf, ofs, try, htxbuf); |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 446 | h1m->curr_len -= ret; |
| 447 | max -= sizeof(struct htx_blk) + ret; |
| 448 | ofs += ret; |
| 449 | total += ret; |
| 450 | if (ret < try) |
| 451 | goto end; |
| 452 | } |
| 453 | |
Christopher Faulet | d1ac2b9 | 2020-12-02 19:12:22 +0100 | [diff] [blame] | 454 | if (!h1m->curr_len) { |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 455 | h1m->state = H1_MSG_DONE; |
Christopher Faulet | d1ac2b9 | 2020-12-02 19:12:22 +0100 | [diff] [blame] | 456 | (*dsthtx)->flags |= HTX_FL_EOM; |
| 457 | } |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 458 | } |
| 459 | else if (h1m->flags & H1_MF_CHNK) { |
| 460 | /* te:chunked : parse chunks */ |
| 461 | new_chunk: |
| 462 | if (h1m->state == H1_MSG_CHUNK_CRLF) { |
| 463 | ret = h1_skip_chunk_crlf(srcbuf, ofs, b_data(srcbuf)); |
| 464 | if (ret <= 0) |
| 465 | goto end; |
| 466 | h1m->state = H1_MSG_CHUNK_SIZE; |
| 467 | ofs += ret; |
| 468 | total += ret; |
| 469 | } |
| 470 | if (h1m->state == H1_MSG_CHUNK_SIZE) { |
Christopher Faulet | 405f054 | 2021-01-27 15:17:13 +0100 | [diff] [blame] | 471 | uint64_t chksz; |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 472 | |
| 473 | ret = h1_parse_chunk_size(srcbuf, ofs, b_data(srcbuf), &chksz); |
| 474 | if (ret <= 0) |
| 475 | goto end; |
| 476 | h1m->state = ((!chksz) ? H1_MSG_TRAILERS : H1_MSG_DATA); |
| 477 | h1m->curr_len = chksz; |
| 478 | h1m->body_len += chksz; |
| 479 | ofs += ret; |
| 480 | total += ret; |
| 481 | if (!h1m->curr_len) |
| 482 | goto end; |
| 483 | } |
| 484 | if (h1m->state == H1_MSG_DATA) { |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 485 | ret = htx_get_max_blksz(*dsthtx, max); |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 486 | if ((uint64_t)ret > h1m->curr_len) |
| 487 | ret = h1m->curr_len; |
| 488 | if (ret > b_contig_data(srcbuf, ofs)) |
| 489 | ret = b_contig_data(srcbuf, ofs); |
| 490 | if (ret) { |
| 491 | int32_t try = ret; |
| 492 | |
Christopher Faulet | cc3124c | 2019-08-12 22:42:21 +0200 | [diff] [blame] | 493 | ret = h1_copy_msg_data(dsthtx, srcbuf, ofs, try, htxbuf); |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 494 | h1m->curr_len -= ret; |
| 495 | max -= sizeof(struct htx_blk) + ret; |
| 496 | ofs += ret; |
| 497 | total += ret; |
| 498 | if (ret < try) |
| 499 | goto end; |
| 500 | } |
| 501 | if (!h1m->curr_len) { |
| 502 | h1m->state = H1_MSG_CHUNK_CRLF; |
| 503 | goto new_chunk; |
| 504 | } |
| 505 | goto end; |
| 506 | } |
| 507 | } |
| 508 | else if (h1m->flags & H1_MF_XFER_LEN) { |
| 509 | /* XFER_LEN is set but not CLEN nor CHNK, it means there is no |
| 510 | * body. Switch the message in DONE state |
| 511 | */ |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 512 | h1m->state = H1_MSG_DONE; |
Christopher Faulet | d1ac2b9 | 2020-12-02 19:12:22 +0100 | [diff] [blame] | 513 | (*dsthtx)->flags |= HTX_FL_EOM; |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 514 | } |
| 515 | else { |
| 516 | /* no content length, read till SHUTW */ |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 517 | ret = htx_get_max_blksz(*dsthtx, max); |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 518 | if (ret > b_contig_data(srcbuf, ofs)) |
| 519 | ret = b_contig_data(srcbuf, ofs); |
| 520 | if (ret) |
Christopher Faulet | cc3124c | 2019-08-12 22:42:21 +0200 | [diff] [blame] | 521 | total += h1_copy_msg_data(dsthtx, srcbuf, ofs, ret, htxbuf); |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 522 | } |
| 523 | |
| 524 | end: |
| 525 | if (ret < 0) { |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 526 | (*dsthtx)->flags |= HTX_FL_PARSING_ERROR; |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 527 | h1m->err_state = h1m->state; |
| 528 | h1m->err_pos = ofs; |
| 529 | total = 0; |
| 530 | } |
| 531 | |
| 532 | /* update htx->extra, only when the body length is known */ |
| 533 | if (h1m->flags & H1_MF_XFER_LEN) |
Christopher Faulet | af54263 | 2019-10-01 21:52:49 +0200 | [diff] [blame] | 534 | (*dsthtx)->extra = h1m->curr_len; |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 535 | return total; |
| 536 | } |
| 537 | |
| 538 | /* Parse HTTP/1 trailers. It returns the number of bytes parsed if > 0, or 0 if |
| 539 | * it couldn't proceed. Parsing errors are reported by setting the htx flags |
| 540 | * HTX_FL_PARSING_ERROR and filling h1m->err_pos and h1m->err_state fields. This |
| 541 | * functions is responsible to update the parser state <h1m>. |
| 542 | */ |
| 543 | int h1_parse_msg_tlrs(struct h1m *h1m, struct htx *dsthtx, |
| 544 | struct buffer *srcbuf, size_t ofs, size_t max) |
| 545 | { |
| 546 | struct http_hdr hdrs[global.tune.max_http_hdr]; |
| 547 | struct h1m tlr_h1m; |
| 548 | int ret = 0; |
| 549 | |
| 550 | if (!max || !b_data(srcbuf)) |
| 551 | goto end; |
| 552 | |
| 553 | /* Realing input buffer if necessary */ |
| 554 | if (b_peek(srcbuf, ofs) > b_tail(srcbuf)) |
| 555 | b_slow_realign(srcbuf, trash.area, 0); |
| 556 | |
| 557 | tlr_h1m.flags = (H1_MF_NO_PHDR|H1_MF_HDRS_ONLY); |
| 558 | ret = h1_headers_to_hdr_list(b_peek(srcbuf, ofs), b_tail(srcbuf), |
| 559 | hdrs, sizeof(hdrs)/sizeof(hdrs[0]), &tlr_h1m, NULL); |
| 560 | if (ret <= 0) { |
| 561 | /* Incomplete or invalid trailers. If the input buffer only |
| 562 | * contains trailers and is full, which is detected by it being |
| 563 | * full and the offset to be zero, it's an error because |
| 564 | * trailers are too large to be handled by the parser. */ |
| 565 | if (ret < 0 || (!ret && !ofs && !buf_room_for_htx_data(srcbuf))) |
| 566 | goto error; |
| 567 | goto end; |
| 568 | } |
| 569 | |
| 570 | /* messages trailers fully parsed. */ |
| 571 | if (h1_eval_htx_hdrs_size(hdrs) > max) { |
| 572 | if (htx_is_empty(dsthtx)) |
| 573 | goto error; |
| 574 | ret = 0; |
| 575 | goto end; |
| 576 | } |
| 577 | |
| 578 | if (!htx_add_all_trailers(dsthtx, hdrs)) |
| 579 | goto error; |
| 580 | |
Christopher Faulet | 76014fd | 2019-12-10 11:47:22 +0100 | [diff] [blame] | 581 | h1m->state = H1_MSG_DONE; |
Christopher Faulet | d1ac2b9 | 2020-12-02 19:12:22 +0100 | [diff] [blame] | 582 | dsthtx->flags |= HTX_FL_EOM; |
Christopher Faulet | 76014fd | 2019-12-10 11:47:22 +0100 | [diff] [blame] | 583 | |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 584 | end: |
| 585 | return ret; |
| 586 | error: |
| 587 | h1m->err_state = h1m->state; |
| 588 | h1m->err_pos = h1m->next; |
| 589 | dsthtx->flags |= HTX_FL_PARSING_ERROR; |
| 590 | return 0; |
| 591 | } |
| 592 | |
Christopher Faulet | 53a899b | 2019-10-08 16:38:42 +0200 | [diff] [blame] | 593 | /* Appends the H1 representation of the request line <sl> to the chunk <chk>. It |
| 594 | * returns 1 if data are successfully appended, otherwise it returns 0. |
| 595 | */ |
| 596 | int h1_format_htx_reqline(const struct htx_sl *sl, struct buffer *chk) |
| 597 | { |
| 598 | struct ist uri; |
| 599 | size_t sz = chk->data; |
| 600 | |
Christopher Faulet | fb38c91 | 2021-04-26 09:38:55 +0200 | [diff] [blame] | 601 | uri = h1_get_uri(sl); |
Christopher Faulet | 53a899b | 2019-10-08 16:38:42 +0200 | [diff] [blame] | 602 | if (!chunk_memcat(chk, HTX_SL_REQ_MPTR(sl), HTX_SL_REQ_MLEN(sl)) || |
| 603 | !chunk_memcat(chk, " ", 1) || |
| 604 | !chunk_memcat(chk, uri.ptr, uri.len) || |
| 605 | !chunk_memcat(chk, " ", 1)) |
| 606 | goto full; |
| 607 | |
| 608 | if (sl->flags & HTX_SL_F_VER_11) { |
| 609 | if (!chunk_memcat(chk, "HTTP/1.1", 8)) |
| 610 | goto full; |
| 611 | } |
| 612 | else { |
| 613 | if (!chunk_memcat(chk, HTX_SL_REQ_VPTR(sl), HTX_SL_REQ_VLEN(sl))) |
| 614 | goto full; |
| 615 | } |
| 616 | |
| 617 | if (!chunk_memcat(chk, "\r\n", 2)) |
| 618 | goto full; |
| 619 | |
| 620 | return 1; |
| 621 | |
| 622 | full: |
| 623 | chk->data = sz; |
| 624 | return 0; |
| 625 | } |
| 626 | |
| 627 | /* Appends the H1 representation of the status line <sl> to the chunk <chk>. It |
| 628 | * returns 1 if data are successfully appended, otherwise it returns 0. |
| 629 | */ |
| 630 | int h1_format_htx_stline(const struct htx_sl *sl, struct buffer *chk) |
| 631 | { |
| 632 | size_t sz = chk->data; |
| 633 | |
| 634 | if (HTX_SL_LEN(sl) + 4 > b_room(chk)) |
| 635 | return 0; |
| 636 | |
| 637 | if (sl->flags & HTX_SL_F_VER_11) { |
| 638 | if (!chunk_memcat(chk, "HTTP/1.1", 8)) |
| 639 | goto full; |
| 640 | } |
| 641 | else { |
| 642 | if (!chunk_memcat(chk, HTX_SL_RES_VPTR(sl), HTX_SL_RES_VLEN(sl))) |
| 643 | goto full; |
| 644 | } |
| 645 | if (!chunk_memcat(chk, " ", 1) || |
| 646 | !chunk_memcat(chk, HTX_SL_RES_CPTR(sl), HTX_SL_RES_CLEN(sl)) || |
| 647 | !chunk_memcat(chk, " ", 1) || |
| 648 | !chunk_memcat(chk, HTX_SL_RES_RPTR(sl), HTX_SL_RES_RLEN(sl)) || |
| 649 | !chunk_memcat(chk, "\r\n", 2)) |
| 650 | goto full; |
| 651 | |
| 652 | return 1; |
| 653 | |
| 654 | full: |
| 655 | chk->data = sz; |
| 656 | return 0; |
| 657 | } |
| 658 | |
Ilya Shipitsin | 47d1718 | 2020-06-21 21:42:57 +0500 | [diff] [blame] | 659 | /* Appends the H1 representation of the header <n> with the value <v> to the |
Christopher Faulet | 53a899b | 2019-10-08 16:38:42 +0200 | [diff] [blame] | 660 | * chunk <chk>. It returns 1 if data are successfully appended, otherwise it |
| 661 | * returns 0. |
| 662 | */ |
| 663 | int h1_format_htx_hdr(const struct ist n, const struct ist v, struct buffer *chk) |
| 664 | { |
| 665 | size_t sz = chk->data; |
| 666 | |
| 667 | if (n.len + v.len + 4 > b_room(chk)) |
| 668 | return 0; |
| 669 | |
| 670 | if (!chunk_memcat(chk, n.ptr, n.len) || |
| 671 | !chunk_memcat(chk, ": ", 2) || |
| 672 | !chunk_memcat(chk, v.ptr, v.len) || |
| 673 | !chunk_memcat(chk, "\r\n", 2)) |
| 674 | goto full; |
| 675 | |
| 676 | return 1; |
| 677 | |
| 678 | full: |
| 679 | chk->data = sz; |
| 680 | return 0; |
| 681 | } |
| 682 | |
| 683 | /* Appends the H1 representation of the data <data> to the chunk <chk>. If |
| 684 | * <chunked> is non-zero, it emits HTTP/1 chunk-encoded data. It returns 1 if |
| 685 | * data are successfully appended, otherwise it returns 0. |
| 686 | */ |
| 687 | int h1_format_htx_data(const struct ist data, struct buffer *chk, int chunked) |
| 688 | { |
| 689 | size_t sz = chk->data; |
| 690 | |
| 691 | if (chunked) { |
| 692 | uint32_t chksz; |
| 693 | char tmp[10]; |
| 694 | char *beg, *end; |
| 695 | |
| 696 | chksz = data.len; |
| 697 | |
| 698 | beg = end = tmp+10; |
| 699 | *--beg = '\n'; |
| 700 | *--beg = '\r'; |
| 701 | do { |
| 702 | *--beg = hextab[chksz & 0xF]; |
| 703 | } while (chksz >>= 4); |
| 704 | |
| 705 | if (!chunk_memcat(chk, beg, end - beg) || |
| 706 | !chunk_memcat(chk, data.ptr, data.len) || |
| 707 | !chunk_memcat(chk, "\r\n", 2)) |
| 708 | goto full; |
| 709 | } |
| 710 | else { |
| 711 | if (!chunk_memcat(chk, data.ptr, data.len)) |
| 712 | return 0; |
| 713 | } |
| 714 | |
| 715 | return 1; |
| 716 | |
| 717 | full: |
| 718 | chk->data = sz; |
| 719 | return 0; |
| 720 | } |
| 721 | |
Christopher Faulet | 4f0f88a | 2019-08-10 11:17:44 +0200 | [diff] [blame] | 722 | /* |
| 723 | * Local variables: |
| 724 | * c-indent-level: 8 |
| 725 | * c-basic-offset: 8 |
| 726 | * End: |
| 727 | */ |