| /* |
| * HTTP protocol analyzer |
| * |
| * Copyright 2000-2007 Willy Tarreau <w@1wt.eu> |
| * |
| * This program is free software; you can redistribute it and/or |
| * modify it under the terms of the GNU General Public License |
| * as published by the Free Software Foundation; either version |
| * 2 of the License, or (at your option) any later version. |
| * |
| */ |
| |
| #include <ctype.h> |
| #include <errno.h> |
| #include <fcntl.h> |
| #include <stdio.h> |
| #include <stdlib.h> |
| #include <string.h> |
| #include <syslog.h> |
| #include <time.h> |
| |
| #include <sys/socket.h> |
| #include <sys/stat.h> |
| #include <sys/types.h> |
| |
| #include <common/appsession.h> |
| #include <common/compat.h> |
| #include <common/config.h> |
| #include <common/debug.h> |
| #include <common/memory.h> |
| #include <common/mini-clist.h> |
| #include <common/standard.h> |
| #include <common/time.h> |
| #include <common/uri_auth.h> |
| #include <common/version.h> |
| |
| #include <types/acl.h> |
| #include <types/capture.h> |
| #include <types/client.h> |
| #include <types/global.h> |
| #include <types/httperr.h> |
| #include <types/polling.h> |
| #include <types/proxy.h> |
| #include <types/server.h> |
| |
| #include <proto/acl.h> |
| #include <proto/backend.h> |
| #include <proto/buffers.h> |
| #include <proto/fd.h> |
| #include <proto/log.h> |
| #include <proto/hdr_idx.h> |
| #include <proto/proto_http.h> |
| #include <proto/queue.h> |
| #include <proto/session.h> |
| #include <proto/task.h> |
| |
| #ifdef CONFIG_HAP_TCPSPLICE |
| #include <libtcpsplice.h> |
| #endif |
| |
| #define DEBUG_PARSE_NO_SPEEDUP |
| #undef DEBUG_PARSE_NO_SPEEDUP |
| |
| /* This is used to perform a quick jump as an alternative to a break/continue |
| * instruction. The first argument is the label for normal operation, and the |
| * second one is the break/continue instruction in the no_speedup mode. |
| */ |
| |
| #ifdef DEBUG_PARSE_NO_SPEEDUP |
| #define QUICK_JUMP(x,y) y |
| #else |
| #define QUICK_JUMP(x,y) goto x |
| #endif |
| |
| /* This is used by remote monitoring */ |
| const char HTTP_200[] = |
| "HTTP/1.0 200 OK\r\n" |
| "Cache-Control: no-cache\r\n" |
| "Connection: close\r\n" |
| "Content-Type: text/html\r\n" |
| "\r\n" |
| "<html><body><h1>200 OK</h1>\nHAProxy: service ready.\n</body></html>\n"; |
| |
| const struct chunk http_200_chunk = { |
| .str = (char *)&HTTP_200, |
| .len = sizeof(HTTP_200)-1 |
| }; |
| |
| const char *HTTP_302 = |
| "HTTP/1.0 302 Found\r\n" |
| "Cache-Control: no-cache\r\n" |
| "Connection: close\r\n" |
| "Location: "; /* not terminated since it will be concatenated with the URL */ |
| |
| /* same as 302 except that the browser MUST retry with the GET method */ |
| const char *HTTP_303 = |
| "HTTP/1.0 303 See Other\r\n" |
| "Cache-Control: no-cache\r\n" |
| "Connection: close\r\n" |
| "Location: "; /* not terminated since it will be concatenated with the URL */ |
| |
| /* Warning: this one is an sprintf() fmt string, with <realm> as its only argument */ |
| const char *HTTP_401_fmt = |
| "HTTP/1.0 401 Unauthorized\r\n" |
| "Cache-Control: no-cache\r\n" |
| "Connection: close\r\n" |
| "Content-Type: text/html\r\n" |
| "WWW-Authenticate: Basic realm=\"%s\"\r\n" |
| "\r\n" |
| "<html><body><h1>401 Unauthorized</h1>\nYou need a valid user and password to access this content.\n</body></html>\n"; |
| |
| |
| const int http_err_codes[HTTP_ERR_SIZE] = { |
| [HTTP_ERR_400] = 400, |
| [HTTP_ERR_403] = 403, |
| [HTTP_ERR_408] = 408, |
| [HTTP_ERR_500] = 500, |
| [HTTP_ERR_502] = 502, |
| [HTTP_ERR_503] = 503, |
| [HTTP_ERR_504] = 504, |
| }; |
| |
| static const char *http_err_msgs[HTTP_ERR_SIZE] = { |
| [HTTP_ERR_400] = |
| "HTTP/1.0 400 Bad request\r\n" |
| "Cache-Control: no-cache\r\n" |
| "Connection: close\r\n" |
| "Content-Type: text/html\r\n" |
| "\r\n" |
| "<html><body><h1>400 Bad request</h1>\nYour browser sent an invalid request.\n</body></html>\n", |
| |
| [HTTP_ERR_403] = |
| "HTTP/1.0 403 Forbidden\r\n" |
| "Cache-Control: no-cache\r\n" |
| "Connection: close\r\n" |
| "Content-Type: text/html\r\n" |
| "\r\n" |
| "<html><body><h1>403 Forbidden</h1>\nRequest forbidden by administrative rules.\n</body></html>\n", |
| |
| [HTTP_ERR_408] = |
| "HTTP/1.0 408 Request Time-out\r\n" |
| "Cache-Control: no-cache\r\n" |
| "Connection: close\r\n" |
| "Content-Type: text/html\r\n" |
| "\r\n" |
| "<html><body><h1>408 Request Time-out</h1>\nYour browser didn't send a complete request in time.\n</body></html>\n", |
| |
| [HTTP_ERR_500] = |
| "HTTP/1.0 500 Server Error\r\n" |
| "Cache-Control: no-cache\r\n" |
| "Connection: close\r\n" |
| "Content-Type: text/html\r\n" |
| "\r\n" |
| "<html><body><h1>500 Server Error</h1>\nAn internal server error occured.\n</body></html>\n", |
| |
| [HTTP_ERR_502] = |
| "HTTP/1.0 502 Bad Gateway\r\n" |
| "Cache-Control: no-cache\r\n" |
| "Connection: close\r\n" |
| "Content-Type: text/html\r\n" |
| "\r\n" |
| "<html><body><h1>502 Bad Gateway</h1>\nThe server returned an invalid or incomplete response.\n</body></html>\n", |
| |
| [HTTP_ERR_503] = |
| "HTTP/1.0 503 Service Unavailable\r\n" |
| "Cache-Control: no-cache\r\n" |
| "Connection: close\r\n" |
| "Content-Type: text/html\r\n" |
| "\r\n" |
| "<html><body><h1>503 Service Unavailable</h1>\nNo server is available to handle this request.\n</body></html>\n", |
| |
| [HTTP_ERR_504] = |
| "HTTP/1.0 504 Gateway Time-out\r\n" |
| "Cache-Control: no-cache\r\n" |
| "Connection: close\r\n" |
| "Content-Type: text/html\r\n" |
| "\r\n" |
| "<html><body><h1>504 Gateway Time-out</h1>\nThe server didn't respond in time.\n</body></html>\n", |
| |
| }; |
| |
| /* We must put the messages here since GCC cannot initialize consts depending |
| * on strlen(). |
| */ |
| struct chunk http_err_chunks[HTTP_ERR_SIZE]; |
| |
| #define FD_SETS_ARE_BITFIELDS |
| #ifdef FD_SETS_ARE_BITFIELDS |
| /* |
| * This map is used with all the FD_* macros to check whether a particular bit |
| * is set or not. Each bit represents an ACSII code. FD_SET() sets those bytes |
| * which should be encoded. When FD_ISSET() returns non-zero, it means that the |
| * byte should be encoded. Be careful to always pass bytes from 0 to 255 |
| * exclusively to the macros. |
| */ |
| fd_set hdr_encode_map[(sizeof(fd_set) > (256/8)) ? 1 : ((256/8) / sizeof(fd_set))]; |
| fd_set url_encode_map[(sizeof(fd_set) > (256/8)) ? 1 : ((256/8) / sizeof(fd_set))]; |
| |
| #else |
| #error "Check if your OS uses bitfields for fd_sets" |
| #endif |
| |
| void init_proto_http() |
| { |
| int i; |
| char *tmp; |
| int msg; |
| |
| for (msg = 0; msg < HTTP_ERR_SIZE; msg++) { |
| if (!http_err_msgs[msg]) { |
| Alert("Internal error: no message defined for HTTP return code %d. Aborting.\n", msg); |
| abort(); |
| } |
| |
| http_err_chunks[msg].str = (char *)http_err_msgs[msg]; |
| http_err_chunks[msg].len = strlen(http_err_msgs[msg]); |
| } |
| |
| /* initialize the log header encoding map : '{|}"#' should be encoded with |
| * '#' as prefix, as well as non-printable characters ( <32 or >= 127 ). |
| * URL encoding only requires '"', '#' to be encoded as well as non- |
| * printable characters above. |
| */ |
| memset(hdr_encode_map, 0, sizeof(hdr_encode_map)); |
| memset(url_encode_map, 0, sizeof(url_encode_map)); |
| for (i = 0; i < 32; i++) { |
| FD_SET(i, hdr_encode_map); |
| FD_SET(i, url_encode_map); |
| } |
| for (i = 127; i < 256; i++) { |
| FD_SET(i, hdr_encode_map); |
| FD_SET(i, url_encode_map); |
| } |
| |
| tmp = "\"#{|}"; |
| while (*tmp) { |
| FD_SET(*tmp, hdr_encode_map); |
| tmp++; |
| } |
| |
| tmp = "\"#"; |
| while (*tmp) { |
| FD_SET(*tmp, url_encode_map); |
| tmp++; |
| } |
| } |
| |
| /* |
| * We have 26 list of methods (1 per first letter), each of which can have |
| * up to 3 entries (2 valid, 1 null). |
| */ |
| struct http_method_desc { |
| http_meth_t meth; |
| int len; |
| const char text[8]; |
| }; |
| |
| const struct http_method_desc http_methods[26][3] = { |
| ['C' - 'A'] = { |
| [0] = { .meth = HTTP_METH_CONNECT , .len=7, .text="CONNECT" }, |
| }, |
| ['D' - 'A'] = { |
| [0] = { .meth = HTTP_METH_DELETE , .len=6, .text="DELETE" }, |
| }, |
| ['G' - 'A'] = { |
| [0] = { .meth = HTTP_METH_GET , .len=3, .text="GET" }, |
| }, |
| ['H' - 'A'] = { |
| [0] = { .meth = HTTP_METH_HEAD , .len=4, .text="HEAD" }, |
| }, |
| ['P' - 'A'] = { |
| [0] = { .meth = HTTP_METH_POST , .len=4, .text="POST" }, |
| [1] = { .meth = HTTP_METH_PUT , .len=3, .text="PUT" }, |
| }, |
| ['T' - 'A'] = { |
| [0] = { .meth = HTTP_METH_TRACE , .len=5, .text="TRACE" }, |
| }, |
| /* rest is empty like this : |
| * [1] = { .meth = HTTP_METH_NONE , .len=0, .text="" }, |
| */ |
| }; |
| |
| /* It is about twice as fast on recent architectures to lookup a byte in a |
| * table than two perform a boolean AND or OR between two tests. Refer to |
| * RFC2616 for those chars. |
| */ |
| |
| const char http_is_spht[256] = { |
| [' '] = 1, ['\t'] = 1, |
| }; |
| |
| const char http_is_crlf[256] = { |
| ['\r'] = 1, ['\n'] = 1, |
| }; |
| |
| const char http_is_lws[256] = { |
| [' '] = 1, ['\t'] = 1, |
| ['\r'] = 1, ['\n'] = 1, |
| }; |
| |
| const char http_is_sep[256] = { |
| ['('] = 1, [')'] = 1, ['<'] = 1, ['>'] = 1, |
| ['@'] = 1, [','] = 1, [';'] = 1, [':'] = 1, |
| ['"'] = 1, ['/'] = 1, ['['] = 1, [']'] = 1, |
| ['{'] = 1, ['}'] = 1, ['?'] = 1, ['='] = 1, |
| [' '] = 1, ['\t'] = 1, ['\\'] = 1, |
| }; |
| |
| const char http_is_ctl[256] = { |
| [0 ... 31] = 1, |
| [127] = 1, |
| }; |
| |
| /* |
| * A token is any ASCII char that is neither a separator nor a CTL char. |
| * Do not overwrite values in assignment since gcc-2.95 will not handle |
| * them correctly. Instead, define every non-CTL char's status. |
| */ |
| const char http_is_token[256] = { |
| [' '] = 0, ['!'] = 1, ['"'] = 0, ['#'] = 1, |
| ['$'] = 1, ['%'] = 1, ['&'] = 1, ['\''] = 1, |
| ['('] = 0, [')'] = 0, ['*'] = 1, ['+'] = 1, |
| [','] = 0, ['-'] = 1, ['.'] = 1, ['/'] = 0, |
| ['0'] = 1, ['1'] = 1, ['2'] = 1, ['3'] = 1, |
| ['4'] = 1, ['5'] = 1, ['6'] = 1, ['7'] = 1, |
| ['8'] = 1, ['9'] = 1, [':'] = 0, [';'] = 0, |
| ['<'] = 0, ['='] = 0, ['>'] = 0, ['?'] = 0, |
| ['@'] = 0, ['A'] = 1, ['B'] = 1, ['C'] = 1, |
| ['D'] = 1, ['E'] = 1, ['F'] = 1, ['G'] = 1, |
| ['H'] = 1, ['I'] = 1, ['J'] = 1, ['K'] = 1, |
| ['L'] = 1, ['M'] = 1, ['N'] = 1, ['O'] = 1, |
| ['P'] = 1, ['Q'] = 1, ['R'] = 1, ['S'] = 1, |
| ['T'] = 1, ['U'] = 1, ['V'] = 1, ['W'] = 1, |
| ['X'] = 1, ['Y'] = 1, ['Z'] = 1, ['['] = 0, |
| ['\\'] = 0, [']'] = 0, ['^'] = 1, ['_'] = 1, |
| ['`'] = 1, ['a'] = 1, ['b'] = 1, ['c'] = 1, |
| ['d'] = 1, ['e'] = 1, ['f'] = 1, ['g'] = 1, |
| ['h'] = 1, ['i'] = 1, ['j'] = 1, ['k'] = 1, |
| ['l'] = 1, ['m'] = 1, ['n'] = 1, ['o'] = 1, |
| ['p'] = 1, ['q'] = 1, ['r'] = 1, ['s'] = 1, |
| ['t'] = 1, ['u'] = 1, ['v'] = 1, ['w'] = 1, |
| ['x'] = 1, ['y'] = 1, ['z'] = 1, ['{'] = 0, |
| ['|'] = 1, ['}'] = 0, ['~'] = 1, |
| }; |
| |
| |
| /* |
| * An http ver_token is any ASCII which can be found in an HTTP version, |
| * which includes 'H', 'T', 'P', '/', '.' and any digit. |
| */ |
| const char http_is_ver_token[256] = { |
| ['.'] = 1, ['/'] = 1, |
| ['0'] = 1, ['1'] = 1, ['2'] = 1, ['3'] = 1, ['4'] = 1, |
| ['5'] = 1, ['6'] = 1, ['7'] = 1, ['8'] = 1, ['9'] = 1, |
| ['H'] = 1, ['P'] = 1, ['T'] = 1, |
| }; |
| |
| |
| #ifdef DEBUG_FULL |
| static char *cli_stnames[5] = {"HDR", "DAT", "SHR", "SHW", "CLS" }; |
| static char *srv_stnames[7] = {"IDL", "CON", "HDR", "DAT", "SHR", "SHW", "CLS" }; |
| #endif |
| |
| static void http_sess_log(struct session *s); |
| |
| /* |
| * Adds a header and its CRLF at the tail of buffer <b>, just before the last |
| * CRLF. Text length is measured first, so it cannot be NULL. |
| * The header is also automatically added to the index <hdr_idx>, and the end |
| * of headers is automatically adjusted. The number of bytes added is returned |
| * on success, otherwise <0 is returned indicating an error. |
| */ |
| int http_header_add_tail(struct buffer *b, struct http_msg *msg, |
| struct hdr_idx *hdr_idx, const char *text) |
| { |
| int bytes, len; |
| |
| len = strlen(text); |
| bytes = buffer_insert_line2(b, b->data + msg->eoh, text, len); |
| if (!bytes) |
| return -1; |
| msg->eoh += bytes; |
| return hdr_idx_add(len, 1, hdr_idx, hdr_idx->tail); |
| } |
| |
| /* |
| * Adds a header and its CRLF at the tail of buffer <b>, just before the last |
| * CRLF. <len> bytes are copied, not counting the CRLF. If <text> is NULL, then |
| * the buffer is only opened and the space reserved, but nothing is copied. |
| * The header is also automatically added to the index <hdr_idx>, and the end |
| * of headers is automatically adjusted. The number of bytes added is returned |
| * on success, otherwise <0 is returned indicating an error. |
| */ |
| int http_header_add_tail2(struct buffer *b, struct http_msg *msg, |
| struct hdr_idx *hdr_idx, const char *text, int len) |
| { |
| int bytes; |
| |
| bytes = buffer_insert_line2(b, b->data + msg->eoh, text, len); |
| if (!bytes) |
| return -1; |
| msg->eoh += bytes; |
| return hdr_idx_add(len, 1, hdr_idx, hdr_idx->tail); |
| } |
| |
| /* |
| * Checks if <hdr> is exactly <name> for <len> chars, and ends with a colon. |
| * If so, returns the position of the first non-space character relative to |
| * <hdr>, or <end>-<hdr> if not found before. If no value is found, it tries |
| * to return a pointer to the place after the first space. Returns 0 if the |
| * header name does not match. Checks are case-insensitive. |
| */ |
| int http_header_match2(const char *hdr, const char *end, |
| const char *name, int len) |
| { |
| const char *val; |
| |
| if (hdr + len >= end) |
| return 0; |
| if (hdr[len] != ':') |
| return 0; |
| if (strncasecmp(hdr, name, len) != 0) |
| return 0; |
| val = hdr + len + 1; |
| while (val < end && HTTP_IS_SPHT(*val)) |
| val++; |
| if ((val >= end) && (len + 2 <= end - hdr)) |
| return len + 2; /* we may replace starting from second space */ |
| return val - hdr; |
| } |
| |
| /* |
| * returns a message to the client ; the connection is shut down for read, |
| * and the request is cleared so that no server connection can be initiated. |
| * The client must be in a valid state for this (HEADER, DATA ...). |
| * Nothing is performed on the server side. The message is contained in a |
| * "chunk". If it is null, then an empty message is used. |
| * The reply buffer doesn't need to be empty before this. |
| */ |
| void client_retnclose(struct session *s, const struct chunk *msg) |
| { |
| EV_FD_CLR(s->cli_fd, DIR_RD); |
| EV_FD_SET(s->cli_fd, DIR_WR); |
| tv_eternity(&s->req->rex); |
| if (!tv_add_ifset(&s->rep->wex, &now, &s->fe->clitimeout)) |
| tv_eternity(&s->rep->wex); |
| s->cli_state = CL_STSHUTR; |
| buffer_flush(s->rep); |
| if (msg->len) |
| buffer_write(s->rep, msg->str, msg->len); |
| s->req->l = 0; |
| } |
| |
| |
| /* |
| * returns a message into the rep buffer, and flushes the req buffer. |
| * The reply buffer doesn't need to be empty before this. The message |
| * is contained in a "chunk". If it is null, then an empty message is |
| * used. |
| */ |
| void client_return(struct session *s, const struct chunk *msg) |
| { |
| buffer_flush(s->rep); |
| if (msg->len) |
| buffer_write(s->rep, msg->str, msg->len); |
| s->req->l = 0; |
| } |
| |
| |
| /* This function turns the server state into the SV_STCLOSE, and sets |
| * indicators accordingly. Note that if <status> is 0, or if the message |
| * pointer is NULL, then no message is returned. |
| */ |
| void srv_close_with_err(struct session *t, int err, int finst, |
| int status, const struct chunk *msg) |
| { |
| t->srv_state = SV_STCLOSE; |
| if (status > 0 && msg) { |
| t->txn.status = status; |
| if (t->fe->mode == PR_MODE_HTTP) |
| client_return(t, msg); |
| } |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= err; |
| if (!(t->flags & SN_FINST_MASK)) |
| t->flags |= finst; |
| } |
| |
| /* This function returns the appropriate error location for the given session |
| * and message. |
| */ |
| |
| struct chunk *error_message(struct session *s, int msgnum) |
| { |
| if (s->be->errmsg[msgnum].str) |
| return &s->be->errmsg[msgnum]; |
| else if (s->fe->errmsg[msgnum].str) |
| return &s->fe->errmsg[msgnum]; |
| else |
| return &http_err_chunks[msgnum]; |
| } |
| |
| /* |
| * returns HTTP_METH_NONE if there is nothing valid to read (empty or non-text |
| * string), HTTP_METH_OTHER for unknown methods, or the identified method. |
| */ |
| static http_meth_t find_http_meth(const char *str, const int len) |
| { |
| unsigned char m; |
| const struct http_method_desc *h; |
| |
| m = ((unsigned)*str - 'A'); |
| |
| if (m < 26) { |
| for (h = http_methods[m]; h->len > 0; h++) { |
| if (unlikely(h->len != len)) |
| continue; |
| if (likely(memcmp(str, h->text, h->len) == 0)) |
| return h->meth; |
| }; |
| return HTTP_METH_OTHER; |
| } |
| return HTTP_METH_NONE; |
| |
| } |
| |
| /* Processes the client and server jobs of a session task, then |
| * puts it back to the wait queue in a clean state, or |
| * cleans up its resources if it must be deleted. Returns |
| * the time the task accepts to wait, or TIME_ETERNITY for |
| * infinity. |
| */ |
| void process_session(struct task *t, struct timeval *next) |
| { |
| struct session *s = t->context; |
| int fsm_resync = 0; |
| |
| do { |
| fsm_resync = 0; |
| //fprintf(stderr,"before_cli:cli=%d, srv=%d\n", s->cli_state, s->srv_state); |
| fsm_resync |= process_cli(s); |
| //fprintf(stderr,"cli/srv:cli=%d, srv=%d\n", s->cli_state, s->srv_state); |
| fsm_resync |= process_srv(s); |
| //fprintf(stderr,"after_srv:cli=%d, srv=%d\n", s->cli_state, s->srv_state); |
| } while (fsm_resync); |
| |
| if (likely(s->cli_state != CL_STCLOSE || s->srv_state != SV_STCLOSE)) { |
| s->req->flags &= BF_CLEAR_READ & BF_CLEAR_WRITE; |
| s->rep->flags &= BF_CLEAR_READ & BF_CLEAR_WRITE; |
| |
| t->expire = s->req->rex; |
| tv_min(&t->expire, &s->req->rex, &s->req->wex); |
| tv_bound(&t->expire, &s->req->cex); |
| tv_bound(&t->expire, &s->rep->rex); |
| tv_bound(&t->expire, &s->rep->wex); |
| |
| /* restore t to its place in the task list */ |
| task_queue(t); |
| |
| #ifdef DEBUG_FULL |
| /* DEBUG code : this should never ever happen, otherwise it indicates |
| * that a task still has something to do and will provoke a quick loop. |
| */ |
| if (tv_remain2(&now, &t->expire) <= 0) |
| exit(100); |
| #endif |
| *next = t->expire; |
| return; /* nothing more to do */ |
| } |
| |
| s->fe->feconn--; |
| if (s->flags & SN_BE_ASSIGNED) |
| s->be->beconn--; |
| actconn--; |
| |
| if (unlikely((global.mode & MODE_DEBUG) && |
| (!(global.mode & MODE_QUIET) || (global.mode & MODE_VERBOSE)))) { |
| int len; |
| len = sprintf(trash, "%08x:%s.closed[%04x:%04x]\n", |
| s->uniq_id, s->be->id, |
| (unsigned short)s->cli_fd, (unsigned short)s->srv_fd); |
| write(1, trash, len); |
| } |
| |
| s->logs.t_close = tv_ms_elapsed(&s->logs.tv_accept, &now); |
| if (s->req != NULL) |
| s->logs.bytes_in = s->req->total; |
| if (s->rep != NULL) |
| s->logs.bytes_out = s->rep->total; |
| |
| s->fe->bytes_in += s->logs.bytes_in; |
| s->fe->bytes_out += s->logs.bytes_out; |
| if (s->be != s->fe) { |
| s->be->bytes_in += s->logs.bytes_in; |
| s->be->bytes_out += s->logs.bytes_out; |
| } |
| if (s->srv) { |
| s->srv->bytes_in += s->logs.bytes_in; |
| s->srv->bytes_out += s->logs.bytes_out; |
| } |
| |
| /* let's do a final log if we need it */ |
| if (s->logs.logwait && |
| !(s->flags & SN_MONITOR) && |
| (!(s->fe->options & PR_O_NULLNOLOG) || s->req->total)) { |
| if (s->fe->to_log & LW_REQ) |
| http_sess_log(s); |
| else |
| tcp_sess_log(s); |
| } |
| |
| /* the task MUST not be in the run queue anymore */ |
| task_delete(t); |
| session_free(s); |
| task_free(t); |
| tv_eternity(next); |
| } |
| |
| |
| extern const char sess_term_cond[8]; |
| extern const char sess_fin_state[8]; |
| extern const char *monthname[12]; |
| const char sess_cookie[4] = "NIDV"; /* No cookie, Invalid cookie, cookie for a Down server, Valid cookie */ |
| const char sess_set_cookie[8] = "N1I3PD5R"; /* No set-cookie, unknown, Set-Cookie Inserted, unknown, |
| Set-cookie seen and left unchanged (passive), Set-cookie Deleted, |
| unknown, Set-cookie Rewritten */ |
| void **pool_requri = NULL; |
| |
| /* |
| * send a log for the session when we have enough info about it. |
| * Will not log if the frontend has no log defined. |
| */ |
| static void http_sess_log(struct session *s) |
| { |
| char pn[INET6_ADDRSTRLEN + strlen(":65535")]; |
| struct proxy *fe = s->fe; |
| struct proxy *be = s->be; |
| struct proxy *prx_log; |
| struct http_txn *txn = &s->txn; |
| int tolog; |
| char *uri, *h; |
| char *svid; |
| struct tm *tm; |
| static char tmpline[MAX_SYSLOG_LEN]; |
| int hdr; |
| |
| if (fe->logfac1 < 0 && fe->logfac2 < 0) |
| return; |
| prx_log = fe; |
| |
| if (s->cli_addr.ss_family == AF_INET) |
| inet_ntop(AF_INET, |
| (const void *)&((struct sockaddr_in *)&s->cli_addr)->sin_addr, |
| pn, sizeof(pn)); |
| else |
| inet_ntop(AF_INET6, |
| (const void *)&((struct sockaddr_in6 *)(&s->cli_addr))->sin6_addr, |
| pn, sizeof(pn)); |
| |
| tm = localtime((time_t *)&s->logs.tv_accept.tv_sec); |
| |
| |
| /* FIXME: let's limit ourselves to frontend logging for now. */ |
| tolog = fe->to_log; |
| |
| h = tmpline; |
| if (fe->to_log & LW_REQHDR && |
| txn->req.cap && |
| (h < tmpline + sizeof(tmpline) - 10)) { |
| *(h++) = ' '; |
| *(h++) = '{'; |
| for (hdr = 0; hdr < fe->nb_req_cap; hdr++) { |
| if (hdr) |
| *(h++) = '|'; |
| if (txn->req.cap[hdr] != NULL) |
| h = encode_string(h, tmpline + sizeof(tmpline) - 7, |
| '#', hdr_encode_map, txn->req.cap[hdr]); |
| } |
| *(h++) = '}'; |
| } |
| |
| if (fe->to_log & LW_RSPHDR && |
| txn->rsp.cap && |
| (h < tmpline + sizeof(tmpline) - 7)) { |
| *(h++) = ' '; |
| *(h++) = '{'; |
| for (hdr = 0; hdr < fe->nb_rsp_cap; hdr++) { |
| if (hdr) |
| *(h++) = '|'; |
| if (txn->rsp.cap[hdr] != NULL) |
| h = encode_string(h, tmpline + sizeof(tmpline) - 4, |
| '#', hdr_encode_map, txn->rsp.cap[hdr]); |
| } |
| *(h++) = '}'; |
| } |
| |
| if (h < tmpline + sizeof(tmpline) - 4) { |
| *(h++) = ' '; |
| *(h++) = '"'; |
| uri = txn->uri ? txn->uri : "<BADREQ>"; |
| h = encode_string(h, tmpline + sizeof(tmpline) - 1, |
| '#', url_encode_map, uri); |
| *(h++) = '"'; |
| } |
| *h = '\0'; |
| |
| svid = (tolog & LW_SVID) ? |
| (s->data_source != DATA_SRC_STATS) ? |
| (s->srv != NULL) ? s->srv->id : "<NOSRV>" : "<STATS>" : "-"; |
| |
| send_log(prx_log, LOG_INFO, |
| "%s:%d [%02d/%s/%04d:%02d:%02d:%02d.%03d]" |
| " %s %s/%s %d/%d/%d/%d/%s%d %d %s%lld" |
| " %s %s %c%c%c%c %d/%d/%d/%d %d/%d%s\n", |
| pn, |
| (s->cli_addr.ss_family == AF_INET) ? |
| ntohs(((struct sockaddr_in *)&s->cli_addr)->sin_port) : |
| ntohs(((struct sockaddr_in6 *)&s->cli_addr)->sin6_port), |
| tm->tm_mday, monthname[tm->tm_mon], tm->tm_year+1900, |
| tm->tm_hour, tm->tm_min, tm->tm_sec, s->logs.tv_accept.tv_usec/1000, |
| fe->id, be->id, svid, |
| s->logs.t_request, |
| (s->logs.t_queue >= 0) ? s->logs.t_queue - s->logs.t_request : -1, |
| (s->logs.t_connect >= 0) ? s->logs.t_connect - s->logs.t_queue : -1, |
| (s->logs.t_data >= 0) ? s->logs.t_data - s->logs.t_connect : -1, |
| (tolog & LW_BYTES) ? "" : "+", s->logs.t_close, |
| txn->status, |
| (tolog & LW_BYTES) ? "" : "+", s->logs.bytes_in, |
| txn->cli_cookie ? txn->cli_cookie : "-", |
| txn->srv_cookie ? txn->srv_cookie : "-", |
| sess_term_cond[(s->flags & SN_ERR_MASK) >> SN_ERR_SHIFT], |
| sess_fin_state[(s->flags & SN_FINST_MASK) >> SN_FINST_SHIFT], |
| (be->options & PR_O_COOK_ANY) ? sess_cookie[(txn->flags & TX_CK_MASK) >> TX_CK_SHIFT] : '-', |
| (be->options & PR_O_COOK_ANY) ? sess_set_cookie[(txn->flags & TX_SCK_MASK) >> TX_SCK_SHIFT] : '-', |
| actconn, fe->feconn, be->beconn, s->srv ? s->srv->cur_sess : 0, |
| s->logs.srv_queue_size, s->logs.prx_queue_size, tmpline); |
| |
| s->logs.logwait = 0; |
| } |
| |
| |
| /* |
| * Capture headers from message starting at <som> according to header list |
| * <cap_hdr>, and fill the <idx> structure appropriately. |
| */ |
| void capture_headers(char *som, struct hdr_idx *idx, |
| char **cap, struct cap_hdr *cap_hdr) |
| { |
| char *eol, *sol, *col, *sov; |
| int cur_idx; |
| struct cap_hdr *h; |
| int len; |
| |
| sol = som + hdr_idx_first_pos(idx); |
| cur_idx = hdr_idx_first_idx(idx); |
| |
| while (cur_idx) { |
| eol = sol + idx->v[cur_idx].len; |
| |
| col = sol; |
| while (col < eol && *col != ':') |
| col++; |
| |
| sov = col + 1; |
| while (sov < eol && http_is_lws[(unsigned char)*sov]) |
| sov++; |
| |
| for (h = cap_hdr; h; h = h->next) { |
| if ((h->namelen == col - sol) && |
| (strncasecmp(sol, h->name, h->namelen) == 0)) { |
| if (cap[h->index] == NULL) |
| cap[h->index] = |
| pool_alloc_from(h->pool, h->len + 1); |
| |
| if (cap[h->index] == NULL) { |
| Alert("HTTP capture : out of memory.\n"); |
| continue; |
| } |
| |
| len = eol - sov; |
| if (len > h->len) |
| len = h->len; |
| |
| memcpy(cap[h->index], sov, len); |
| cap[h->index][len]=0; |
| } |
| } |
| sol = eol + idx->v[cur_idx].cr + 1; |
| cur_idx = idx->v[cur_idx].next; |
| } |
| } |
| |
| |
| /* either we find an LF at <ptr> or we jump to <bad>. |
| */ |
| #define EXPECT_LF_HERE(ptr, bad) do { if (unlikely(*(ptr) != '\n')) goto bad; } while (0) |
| |
| /* plays with variables <ptr>, <end> and <state>. Jumps to <good> if OK, |
| * otherwise to <http_msg_ood> with <state> set to <st>. |
| */ |
| #define EAT_AND_JUMP_OR_RETURN(good, st) do { \ |
| ptr++; \ |
| if (likely(ptr < end)) \ |
| goto good; \ |
| else { \ |
| state = (st); \ |
| goto http_msg_ood; \ |
| } \ |
| } while (0) |
| |
| |
| /* |
| * This function parses a status line between <ptr> and <end>, starting with |
| * parser state <state>. Only states HTTP_MSG_RPVER, HTTP_MSG_RPVER_SP, |
| * HTTP_MSG_RPCODE, HTTP_MSG_RPCODE_SP and HTTP_MSG_RPREASON are handled. Others |
| * will give undefined results. |
| * Note that it is upon the caller's responsibility to ensure that ptr < end, |
| * and that msg->sol points to the beginning of the response. |
| * If a complete line is found (which implies that at least one CR or LF is |
| * found before <end>, the updated <ptr> is returned, otherwise NULL is |
| * returned indicating an incomplete line (which does not mean that parts have |
| * not been updated). In the incomplete case, if <ret_ptr> or <ret_state> are |
| * non-NULL, they are fed with the new <ptr> and <state> values to be passed |
| * upon next call. |
| * |
| * This function was intentionally designed to be called from |
| * http_msg_analyzer() with the lowest overhead. It should integrate perfectly |
| * within its state machine and use the same macros, hence the need for same |
| * labels and variable names. Note that msg->sol is left unchanged. |
| */ |
| const char *http_parse_stsline(struct http_msg *msg, const char *msg_buf, int state, |
| const char *ptr, const char *end, |
| char **ret_ptr, int *ret_state) |
| { |
| __label__ |
| http_msg_rpver, |
| http_msg_rpver_sp, |
| http_msg_rpcode, |
| http_msg_rpcode_sp, |
| http_msg_rpreason, |
| http_msg_rpline_eol, |
| http_msg_ood, /* out of data */ |
| http_msg_invalid; |
| |
| switch (state) { |
| http_msg_rpver: |
| case HTTP_MSG_RPVER: |
| if (likely(HTTP_IS_VER_TOKEN(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rpver, HTTP_MSG_RPVER); |
| |
| if (likely(HTTP_IS_SPHT(*ptr))) { |
| msg->sl.st.v_l = (ptr - msg_buf) - msg->som; |
| EAT_AND_JUMP_OR_RETURN(http_msg_rpver_sp, HTTP_MSG_RPVER_SP); |
| } |
| goto http_msg_invalid; |
| |
| http_msg_rpver_sp: |
| case HTTP_MSG_RPVER_SP: |
| if (likely(!HTTP_IS_LWS(*ptr))) { |
| msg->sl.st.c = ptr - msg_buf; |
| goto http_msg_rpcode; |
| } |
| if (likely(HTTP_IS_SPHT(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rpver_sp, HTTP_MSG_RPVER_SP); |
| /* so it's a CR/LF, this is invalid */ |
| goto http_msg_invalid; |
| |
| http_msg_rpcode: |
| case HTTP_MSG_RPCODE: |
| if (likely(!HTTP_IS_LWS(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rpcode, HTTP_MSG_RPCODE); |
| |
| if (likely(HTTP_IS_SPHT(*ptr))) { |
| msg->sl.st.c_l = (ptr - msg_buf) - msg->sl.st.c; |
| EAT_AND_JUMP_OR_RETURN(http_msg_rpcode_sp, HTTP_MSG_RPCODE_SP); |
| } |
| |
| /* so it's a CR/LF, so there is no reason phrase */ |
| msg->sl.st.c_l = (ptr - msg_buf) - msg->sl.st.c; |
| http_msg_rsp_reason: |
| /* FIXME: should we support HTTP responses without any reason phrase ? */ |
| msg->sl.st.r = ptr - msg_buf; |
| msg->sl.st.r_l = 0; |
| goto http_msg_rpline_eol; |
| |
| http_msg_rpcode_sp: |
| case HTTP_MSG_RPCODE_SP: |
| if (likely(!HTTP_IS_LWS(*ptr))) { |
| msg->sl.st.r = ptr - msg_buf; |
| goto http_msg_rpreason; |
| } |
| if (likely(HTTP_IS_SPHT(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rpcode_sp, HTTP_MSG_RPCODE_SP); |
| /* so it's a CR/LF, so there is no reason phrase */ |
| goto http_msg_rsp_reason; |
| |
| http_msg_rpreason: |
| case HTTP_MSG_RPREASON: |
| if (likely(!HTTP_IS_CRLF(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rpreason, HTTP_MSG_RPREASON); |
| msg->sl.st.r_l = (ptr - msg_buf) - msg->sl.st.r; |
| http_msg_rpline_eol: |
| /* We have seen the end of line. Note that we do not |
| * necessarily have the \n yet, but at least we know that we |
| * have EITHER \r OR \n, otherwise the response would not be |
| * complete. We can then record the response length and return |
| * to the caller which will be able to register it. |
| */ |
| msg->sl.st.l = ptr - msg->sol; |
| return ptr; |
| |
| #ifdef DEBUG_FULL |
| default: |
| fprintf(stderr, "FIXME !!!! impossible state at %s:%d = %d\n", __FILE__, __LINE__, state); |
| exit(1); |
| #endif |
| } |
| |
| http_msg_ood: |
| /* out of data */ |
| if (ret_state) |
| *ret_state = state; |
| if (ret_ptr) |
| *ret_ptr = (char *)ptr; |
| return NULL; |
| |
| http_msg_invalid: |
| /* invalid message */ |
| if (ret_state) |
| *ret_state = HTTP_MSG_ERROR; |
| return NULL; |
| } |
| |
| |
| /* |
| * This function parses a request line between <ptr> and <end>, starting with |
| * parser state <state>. Only states HTTP_MSG_RQMETH, HTTP_MSG_RQMETH_SP, |
| * HTTP_MSG_RQURI, HTTP_MSG_RQURI_SP and HTTP_MSG_RQVER are handled. Others |
| * will give undefined results. |
| * Note that it is upon the caller's responsibility to ensure that ptr < end, |
| * and that msg->sol points to the beginning of the request. |
| * If a complete line is found (which implies that at least one CR or LF is |
| * found before <end>, the updated <ptr> is returned, otherwise NULL is |
| * returned indicating an incomplete line (which does not mean that parts have |
| * not been updated). In the incomplete case, if <ret_ptr> or <ret_state> are |
| * non-NULL, they are fed with the new <ptr> and <state> values to be passed |
| * upon next call. |
| * |
| * This function was intentionally designed to be called from |
| * http_msg_analyzer() with the lowest overhead. It should integrate perfectly |
| * within its state machine and use the same macros, hence the need for same |
| * labels and variable names. Note that msg->sol is left unchanged. |
| */ |
| const char *http_parse_reqline(struct http_msg *msg, const char *msg_buf, int state, |
| const char *ptr, const char *end, |
| char **ret_ptr, int *ret_state) |
| { |
| __label__ |
| http_msg_rqmeth, |
| http_msg_rqmeth_sp, |
| http_msg_rquri, |
| http_msg_rquri_sp, |
| http_msg_rqver, |
| http_msg_rqline_eol, |
| http_msg_ood, /* out of data */ |
| http_msg_invalid; |
| |
| switch (state) { |
| http_msg_rqmeth: |
| case HTTP_MSG_RQMETH: |
| if (likely(HTTP_IS_TOKEN(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rqmeth, HTTP_MSG_RQMETH); |
| |
| if (likely(HTTP_IS_SPHT(*ptr))) { |
| msg->sl.rq.m_l = (ptr - msg_buf) - msg->som; |
| EAT_AND_JUMP_OR_RETURN(http_msg_rqmeth_sp, HTTP_MSG_RQMETH_SP); |
| } |
| |
| if (likely(HTTP_IS_CRLF(*ptr))) { |
| /* HTTP 0.9 request */ |
| msg->sl.rq.m_l = (ptr - msg_buf) - msg->som; |
| http_msg_req09_uri: |
| msg->sl.rq.u = ptr - msg_buf; |
| http_msg_req09_uri_e: |
| msg->sl.rq.u_l = (ptr - msg_buf) - msg->sl.rq.u; |
| http_msg_req09_ver: |
| msg->sl.rq.v = ptr - msg_buf; |
| msg->sl.rq.v_l = 0; |
| goto http_msg_rqline_eol; |
| } |
| goto http_msg_invalid; |
| |
| http_msg_rqmeth_sp: |
| case HTTP_MSG_RQMETH_SP: |
| if (likely(!HTTP_IS_LWS(*ptr))) { |
| msg->sl.rq.u = ptr - msg_buf; |
| goto http_msg_rquri; |
| } |
| if (likely(HTTP_IS_SPHT(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rqmeth_sp, HTTP_MSG_RQMETH_SP); |
| /* so it's a CR/LF, meaning an HTTP 0.9 request */ |
| goto http_msg_req09_uri; |
| |
| http_msg_rquri: |
| case HTTP_MSG_RQURI: |
| if (likely(!HTTP_IS_LWS(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rquri, HTTP_MSG_RQURI); |
| |
| if (likely(HTTP_IS_SPHT(*ptr))) { |
| msg->sl.rq.u_l = (ptr - msg_buf) - msg->sl.rq.u; |
| EAT_AND_JUMP_OR_RETURN(http_msg_rquri_sp, HTTP_MSG_RQURI_SP); |
| } |
| |
| /* so it's a CR/LF, meaning an HTTP 0.9 request */ |
| goto http_msg_req09_uri_e; |
| |
| http_msg_rquri_sp: |
| case HTTP_MSG_RQURI_SP: |
| if (likely(!HTTP_IS_LWS(*ptr))) { |
| msg->sl.rq.v = ptr - msg_buf; |
| goto http_msg_rqver; |
| } |
| if (likely(HTTP_IS_SPHT(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rquri_sp, HTTP_MSG_RQURI_SP); |
| /* so it's a CR/LF, meaning an HTTP 0.9 request */ |
| goto http_msg_req09_ver; |
| |
| http_msg_rqver: |
| case HTTP_MSG_RQVER: |
| if (likely(HTTP_IS_VER_TOKEN(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rqver, HTTP_MSG_RQVER); |
| |
| if (likely(HTTP_IS_CRLF(*ptr))) { |
| msg->sl.rq.v_l = (ptr - msg_buf) - msg->sl.rq.v; |
| http_msg_rqline_eol: |
| /* We have seen the end of line. Note that we do not |
| * necessarily have the \n yet, but at least we know that we |
| * have EITHER \r OR \n, otherwise the request would not be |
| * complete. We can then record the request length and return |
| * to the caller which will be able to register it. |
| */ |
| msg->sl.rq.l = ptr - msg->sol; |
| return ptr; |
| } |
| |
| /* neither an HTTP_VER token nor a CRLF */ |
| goto http_msg_invalid; |
| |
| #ifdef DEBUG_FULL |
| default: |
| fprintf(stderr, "FIXME !!!! impossible state at %s:%d = %d\n", __FILE__, __LINE__, state); |
| exit(1); |
| #endif |
| } |
| |
| http_msg_ood: |
| /* out of data */ |
| if (ret_state) |
| *ret_state = state; |
| if (ret_ptr) |
| *ret_ptr = (char *)ptr; |
| return NULL; |
| |
| http_msg_invalid: |
| /* invalid message */ |
| if (ret_state) |
| *ret_state = HTTP_MSG_ERROR; |
| return NULL; |
| } |
| |
| |
| /* |
| * This function parses an HTTP message, either a request or a response, |
| * depending on the initial msg->msg_state. It can be preempted everywhere |
| * when data are missing and recalled at the exact same location with no |
| * information loss. The header index is re-initialized when switching from |
| * MSG_R[PQ]BEFORE to MSG_RPVER|MSG_RQMETH. It modifies msg->sol among other |
| * fields. |
| */ |
| void http_msg_analyzer(struct buffer *buf, struct http_msg *msg, struct hdr_idx *idx) |
| { |
| __label__ |
| http_msg_rqbefore, |
| http_msg_rqbefore_cr, |
| http_msg_rqmeth, |
| http_msg_rqline_end, |
| http_msg_hdr_first, |
| http_msg_hdr_name, |
| http_msg_hdr_l1_sp, |
| http_msg_hdr_l1_lf, |
| http_msg_hdr_l1_lws, |
| http_msg_hdr_val, |
| http_msg_hdr_l2_lf, |
| http_msg_hdr_l2_lws, |
| http_msg_complete_header, |
| http_msg_last_lf, |
| http_msg_ood, /* out of data */ |
| http_msg_invalid; |
| |
| int state; /* updated only when leaving the FSM */ |
| register char *ptr, *end; /* request pointers, to avoid dereferences */ |
| |
| state = msg->msg_state; |
| ptr = buf->lr; |
| end = buf->r; |
| |
| if (unlikely(ptr >= end)) |
| goto http_msg_ood; |
| |
| switch (state) { |
| /* |
| * First, states that are specific to the response only. |
| * We check them first so that request and headers are |
| * closer to each other (accessed more often). |
| */ |
| http_msg_rpbefore: |
| case HTTP_MSG_RPBEFORE: |
| if (likely(HTTP_IS_TOKEN(*ptr))) { |
| if (likely(ptr == buf->data)) { |
| msg->sol = ptr; |
| msg->som = 0; |
| } else { |
| #if PARSE_PRESERVE_EMPTY_LINES |
| /* only skip empty leading lines, don't remove them */ |
| msg->sol = ptr; |
| msg->som = ptr - buf->data; |
| #else |
| /* Remove empty leading lines, as recommended by |
| * RFC2616. This takes a lot of time because we |
| * must move all the buffer backwards, but this |
| * is rarely needed. The method above will be |
| * cleaner when we'll be able to start sending |
| * the request from any place in the buffer. |
| */ |
| buf->lr = ptr; |
| buffer_replace2(buf, buf->data, buf->lr, NULL, 0); |
| msg->som = 0; |
| msg->sol = buf->data; |
| ptr = buf->data; |
| end = buf->r; |
| #endif |
| } |
| hdr_idx_init(idx); |
| state = HTTP_MSG_RPVER; |
| goto http_msg_rpver; |
| } |
| |
| if (unlikely(!HTTP_IS_CRLF(*ptr))) |
| goto http_msg_invalid; |
| |
| if (unlikely(*ptr == '\n')) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rpbefore, HTTP_MSG_RPBEFORE); |
| EAT_AND_JUMP_OR_RETURN(http_msg_rpbefore_cr, HTTP_MSG_RPBEFORE_CR); |
| /* stop here */ |
| |
| http_msg_rpbefore_cr: |
| case HTTP_MSG_RPBEFORE_CR: |
| EXPECT_LF_HERE(ptr, http_msg_invalid); |
| EAT_AND_JUMP_OR_RETURN(http_msg_rpbefore, HTTP_MSG_RPBEFORE); |
| /* stop here */ |
| |
| http_msg_rpver: |
| case HTTP_MSG_RPVER: |
| case HTTP_MSG_RPVER_SP: |
| case HTTP_MSG_RPCODE: |
| case HTTP_MSG_RPCODE_SP: |
| case HTTP_MSG_RPREASON: |
| ptr = (char *)http_parse_stsline(msg, buf->data, state, ptr, end, |
| &buf->lr, &msg->msg_state); |
| if (unlikely(!ptr)) |
| return; |
| |
| /* we have a full response and we know that we have either a CR |
| * or an LF at <ptr>. |
| */ |
| //fprintf(stderr,"som=%d rq.l=%d *ptr=0x%02x\n", msg->som, msg->sl.st.l, *ptr); |
| hdr_idx_set_start(idx, msg->sl.st.l, *ptr == '\r'); |
| |
| msg->sol = ptr; |
| if (likely(*ptr == '\r')) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rpline_end, HTTP_MSG_RPLINE_END); |
| goto http_msg_rpline_end; |
| |
| http_msg_rpline_end: |
| case HTTP_MSG_RPLINE_END: |
| /* msg->sol must point to the first of CR or LF. */ |
| EXPECT_LF_HERE(ptr, http_msg_invalid); |
| EAT_AND_JUMP_OR_RETURN(http_msg_hdr_first, HTTP_MSG_HDR_FIRST); |
| /* stop here */ |
| |
| /* |
| * Second, states that are specific to the request only |
| */ |
| http_msg_rqbefore: |
| case HTTP_MSG_RQBEFORE: |
| if (likely(HTTP_IS_TOKEN(*ptr))) { |
| if (likely(ptr == buf->data)) { |
| msg->sol = ptr; |
| msg->som = 0; |
| } else { |
| #if PARSE_PRESERVE_EMPTY_LINES |
| /* only skip empty leading lines, don't remove them */ |
| msg->sol = ptr; |
| msg->som = ptr - buf->data; |
| #else |
| /* Remove empty leading lines, as recommended by |
| * RFC2616. This takes a lot of time because we |
| * must move all the buffer backwards, but this |
| * is rarely needed. The method above will be |
| * cleaner when we'll be able to start sending |
| * the request from any place in the buffer. |
| */ |
| buf->lr = ptr; |
| buffer_replace2(buf, buf->data, buf->lr, NULL, 0); |
| msg->som = 0; |
| msg->sol = buf->data; |
| ptr = buf->data; |
| end = buf->r; |
| #endif |
| } |
| /* we will need this when keep-alive will be supported |
| hdr_idx_init(idx); |
| */ |
| state = HTTP_MSG_RQMETH; |
| goto http_msg_rqmeth; |
| } |
| |
| if (unlikely(!HTTP_IS_CRLF(*ptr))) |
| goto http_msg_invalid; |
| |
| if (unlikely(*ptr == '\n')) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rqbefore, HTTP_MSG_RQBEFORE); |
| EAT_AND_JUMP_OR_RETURN(http_msg_rqbefore_cr, HTTP_MSG_RQBEFORE_CR); |
| /* stop here */ |
| |
| http_msg_rqbefore_cr: |
| case HTTP_MSG_RQBEFORE_CR: |
| EXPECT_LF_HERE(ptr, http_msg_invalid); |
| EAT_AND_JUMP_OR_RETURN(http_msg_rqbefore, HTTP_MSG_RQBEFORE); |
| /* stop here */ |
| |
| http_msg_rqmeth: |
| case HTTP_MSG_RQMETH: |
| case HTTP_MSG_RQMETH_SP: |
| case HTTP_MSG_RQURI: |
| case HTTP_MSG_RQURI_SP: |
| case HTTP_MSG_RQVER: |
| ptr = (char *)http_parse_reqline(msg, buf->data, state, ptr, end, |
| &buf->lr, &msg->msg_state); |
| if (unlikely(!ptr)) |
| return; |
| |
| /* we have a full request and we know that we have either a CR |
| * or an LF at <ptr>. |
| */ |
| //fprintf(stderr,"som=%d rq.l=%d *ptr=0x%02x\n", msg->som, msg->sl.rq.l, *ptr); |
| hdr_idx_set_start(idx, msg->sl.rq.l, *ptr == '\r'); |
| |
| msg->sol = ptr; |
| if (likely(*ptr == '\r')) |
| EAT_AND_JUMP_OR_RETURN(http_msg_rqline_end, HTTP_MSG_RQLINE_END); |
| goto http_msg_rqline_end; |
| |
| http_msg_rqline_end: |
| case HTTP_MSG_RQLINE_END: |
| /* check for HTTP/0.9 request : no version information available. |
| * msg->sol must point to the first of CR or LF. |
| */ |
| if (unlikely(msg->sl.rq.v_l == 0)) |
| goto http_msg_last_lf; |
| |
| EXPECT_LF_HERE(ptr, http_msg_invalid); |
| EAT_AND_JUMP_OR_RETURN(http_msg_hdr_first, HTTP_MSG_HDR_FIRST); |
| /* stop here */ |
| |
| /* |
| * Common states below |
| */ |
| http_msg_hdr_first: |
| case HTTP_MSG_HDR_FIRST: |
| msg->sol = ptr; |
| if (likely(!HTTP_IS_CRLF(*ptr))) { |
| goto http_msg_hdr_name; |
| } |
| |
| if (likely(*ptr == '\r')) |
| EAT_AND_JUMP_OR_RETURN(http_msg_last_lf, HTTP_MSG_LAST_LF); |
| goto http_msg_last_lf; |
| |
| http_msg_hdr_name: |
| case HTTP_MSG_HDR_NAME: |
| /* assumes msg->sol points to the first char */ |
| if (likely(HTTP_IS_TOKEN(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_hdr_name, HTTP_MSG_HDR_NAME); |
| |
| if (likely(*ptr == ':')) { |
| msg->col = ptr - buf->data; |
| EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l1_sp, HTTP_MSG_HDR_L1_SP); |
| } |
| |
| goto http_msg_invalid; |
| |
| http_msg_hdr_l1_sp: |
| case HTTP_MSG_HDR_L1_SP: |
| /* assumes msg->sol points to the first char and msg->col to the colon */ |
| if (likely(HTTP_IS_SPHT(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l1_sp, HTTP_MSG_HDR_L1_SP); |
| |
| /* header value can be basically anything except CR/LF */ |
| msg->sov = ptr - buf->data; |
| |
| if (likely(!HTTP_IS_CRLF(*ptr))) { |
| goto http_msg_hdr_val; |
| } |
| |
| if (likely(*ptr == '\r')) |
| EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l1_lf, HTTP_MSG_HDR_L1_LF); |
| goto http_msg_hdr_l1_lf; |
| |
| http_msg_hdr_l1_lf: |
| case HTTP_MSG_HDR_L1_LF: |
| EXPECT_LF_HERE(ptr, http_msg_invalid); |
| EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l1_lws, HTTP_MSG_HDR_L1_LWS); |
| |
| http_msg_hdr_l1_lws: |
| case HTTP_MSG_HDR_L1_LWS: |
| if (likely(HTTP_IS_SPHT(*ptr))) { |
| /* replace HT,CR,LF with spaces */ |
| for (; buf->data+msg->sov < ptr; msg->sov++) |
| buf->data[msg->sov] = ' '; |
| goto http_msg_hdr_l1_sp; |
| } |
| /* we had a header consisting only in spaces ! */ |
| msg->eol = buf->data + msg->sov; |
| goto http_msg_complete_header; |
| |
| http_msg_hdr_val: |
| case HTTP_MSG_HDR_VAL: |
| /* assumes msg->sol points to the first char, msg->col to the |
| * colon, and msg->sov points to the first character of the |
| * value. |
| */ |
| if (likely(!HTTP_IS_CRLF(*ptr))) |
| EAT_AND_JUMP_OR_RETURN(http_msg_hdr_val, HTTP_MSG_HDR_VAL); |
| |
| msg->eol = ptr; |
| /* Note: we could also copy eol into ->eoh so that we have the |
| * real header end in case it ends with lots of LWS, but is this |
| * really needed ? |
| */ |
| if (likely(*ptr == '\r')) |
| EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l2_lf, HTTP_MSG_HDR_L2_LF); |
| goto http_msg_hdr_l2_lf; |
| |
| http_msg_hdr_l2_lf: |
| case HTTP_MSG_HDR_L2_LF: |
| EXPECT_LF_HERE(ptr, http_msg_invalid); |
| EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l2_lws, HTTP_MSG_HDR_L2_LWS); |
| |
| http_msg_hdr_l2_lws: |
| case HTTP_MSG_HDR_L2_LWS: |
| if (unlikely(HTTP_IS_SPHT(*ptr))) { |
| /* LWS: replace HT,CR,LF with spaces */ |
| for (; msg->eol < ptr; msg->eol++) |
| *msg->eol = ' '; |
| goto http_msg_hdr_val; |
| } |
| http_msg_complete_header: |
| /* |
| * It was a new header, so the last one is finished. |
| * Assumes msg->sol points to the first char, msg->col to the |
| * colon, msg->sov points to the first character of the value |
| * and msg->eol to the first CR or LF so we know how the line |
| * ends. We insert last header into the index. |
| */ |
| /* |
| fprintf(stderr,"registering %-2d bytes : ", msg->eol - msg->sol); |
| write(2, msg->sol, msg->eol-msg->sol); |
| fprintf(stderr,"\n"); |
| */ |
| |
| if (unlikely(hdr_idx_add(msg->eol - msg->sol, *msg->eol == '\r', |
| idx, idx->tail) < 0)) |
| goto http_msg_invalid; |
| |
| msg->sol = ptr; |
| if (likely(!HTTP_IS_CRLF(*ptr))) { |
| goto http_msg_hdr_name; |
| } |
| |
| if (likely(*ptr == '\r')) |
| EAT_AND_JUMP_OR_RETURN(http_msg_last_lf, HTTP_MSG_LAST_LF); |
| goto http_msg_last_lf; |
| |
| http_msg_last_lf: |
| case HTTP_MSG_LAST_LF: |
| /* Assumes msg->sol points to the first of either CR or LF */ |
| EXPECT_LF_HERE(ptr, http_msg_invalid); |
| ptr++; |
| buf->lr = ptr; |
| msg->eoh = msg->sol - buf->data; |
| msg->msg_state = HTTP_MSG_BODY; |
| return; |
| #ifdef DEBUG_FULL |
| default: |
| fprintf(stderr, "FIXME !!!! impossible state at %s:%d = %d\n", __FILE__, __LINE__, state); |
| exit(1); |
| #endif |
| } |
| http_msg_ood: |
| /* out of data */ |
| msg->msg_state = state; |
| buf->lr = ptr; |
| return; |
| |
| http_msg_invalid: |
| /* invalid message */ |
| msg->msg_state = HTTP_MSG_ERROR; |
| return; |
| } |
| |
| /* |
| * manages the client FSM and its socket. BTW, it also tries to handle the |
| * cookie. It returns 1 if a state has changed (and a resync may be needed), |
| * 0 else. |
| */ |
| int process_cli(struct session *t) |
| { |
| int s = t->srv_state; |
| int c = t->cli_state; |
| struct buffer *req = t->req; |
| struct buffer *rep = t->rep; |
| |
| DPRINTF(stderr,"process_cli: c=%s s=%s set(r,w)=%d,%d exp(r,w)=%d.%d,%d.%d\n", |
| cli_stnames[c], srv_stnames[s], |
| EV_FD_ISSET(t->cli_fd, DIR_RD), EV_FD_ISSET(t->cli_fd, DIR_WR), |
| req->rex.tv_sec, req->rex.tv_usec, |
| rep->wex.tv_sec, rep->wex.tv_usec); |
| |
| if (c == CL_STHEADERS) { |
| /* |
| * Now parse the partial (or complete) lines. |
| * We will check the request syntax, and also join multi-line |
| * headers. An index of all the lines will be elaborated while |
| * parsing. |
| * |
| * For the parsing, we use a 28 states FSM. |
| * |
| * Here is the information we currently have : |
| * req->data + req->som = beginning of request |
| * req->data + req->eoh = end of processed headers / start of current one |
| * req->data + req->eol = end of current header or line (LF or CRLF) |
| * req->lr = first non-visited byte |
| * req->r = end of data |
| */ |
| |
| int cur_idx; |
| struct http_txn *txn = &t->txn; |
| struct http_msg *msg = &txn->req; |
| struct proxy *cur_proxy; |
| |
| if (likely(req->lr < req->r)) |
| http_msg_analyzer(req, msg, &txn->hdr_idx); |
| |
| /* 1: we might have to print this header in debug mode */ |
| if (unlikely((global.mode & MODE_DEBUG) && |
| (!(global.mode & MODE_QUIET) || (global.mode & MODE_VERBOSE)) && |
| (msg->msg_state == HTTP_MSG_BODY || msg->msg_state == HTTP_MSG_ERROR))) { |
| char *eol, *sol; |
| |
| sol = req->data + msg->som; |
| eol = sol + msg->sl.rq.l; |
| debug_hdr("clireq", t, sol, eol); |
| |
| sol += hdr_idx_first_pos(&txn->hdr_idx); |
| cur_idx = hdr_idx_first_idx(&txn->hdr_idx); |
| |
| while (cur_idx) { |
| eol = sol + txn->hdr_idx.v[cur_idx].len; |
| debug_hdr("clihdr", t, sol, eol); |
| sol = eol + txn->hdr_idx.v[cur_idx].cr + 1; |
| cur_idx = txn->hdr_idx.v[cur_idx].next; |
| } |
| } |
| |
| |
| /* |
| * Now we quickly check if we have found a full valid request. |
| * If not so, we check the FD and buffer states before leaving. |
| * A full request is indicated by the fact that we have seen |
| * the double LF/CRLF, so the state is HTTP_MSG_BODY. Invalid |
| * requests are checked first. |
| * |
| */ |
| |
| if (unlikely(msg->msg_state != HTTP_MSG_BODY)) { |
| /* |
| * First, let's catch bad requests. |
| */ |
| if (unlikely(msg->msg_state == HTTP_MSG_ERROR)) |
| goto return_bad_req; |
| |
| /* 1: Since we are in header mode, if there's no space |
| * left for headers, we won't be able to free more |
| * later, so the session will never terminate. We |
| * must terminate it now. |
| */ |
| if (unlikely(req->l >= req->rlim - req->data)) { |
| /* FIXME: check if URI is set and return Status |
| * 414 Request URI too long instead. |
| */ |
| goto return_bad_req; |
| } |
| |
| /* 2: have we encountered a read error or a close ? */ |
| else if (unlikely(req->flags & (BF_READ_ERROR | BF_READ_NULL))) { |
| /* read error, or last read : give up. */ |
| tv_eternity(&req->rex); |
| fd_delete(t->cli_fd); |
| t->cli_state = CL_STCLOSE; |
| t->fe->failed_req++; |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_CLICL; |
| if (!(t->flags & SN_FINST_MASK)) |
| t->flags |= SN_FINST_R; |
| return 1; |
| } |
| |
| /* 3: has the read timeout expired ? */ |
| else if (unlikely(tv_isle(&req->rex, &now))) { |
| /* read timeout : give up with an error message. */ |
| txn->status = 408; |
| client_retnclose(t, error_message(t, HTTP_ERR_408)); |
| t->fe->failed_req++; |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_CLITO; |
| if (!(t->flags & SN_FINST_MASK)) |
| t->flags |= SN_FINST_R; |
| return 1; |
| } |
| |
| /* 4: do we need to re-enable the read socket ? */ |
| else if (unlikely(EV_FD_COND_S(t->cli_fd, DIR_RD))) { |
| /* fd in DIR_RD was disabled, perhaps because of a previous buffer |
| * full. We cannot loop here since stream_sock_read will disable it only if |
| * req->l == rlim-data |
| */ |
| if (!tv_add_ifset(&req->rex, &now, &t->fe->clitimeout)) |
| tv_eternity(&req->rex); |
| } |
| return t->cli_state != CL_STHEADERS; |
| } |
| |
| |
| /**************************************************************** |
| * More interesting part now : we know that we have a complete * |
| * request which at least looks like HTTP. We have an indicator * |
| * of each header's length, so we can parse them quickly. * |
| ****************************************************************/ |
| |
| /* ensure we keep this pointer to the beginning of the message */ |
| msg->sol = req->data + msg->som; |
| |
| /* |
| * 1: identify the method |
| */ |
| txn->meth = find_http_meth(&req->data[msg->som], msg->sl.rq.m_l); |
| |
| /* |
| * 2: check if the URI matches the monitor_uri. |
| * We have to do this for every request which gets in, because |
| * the monitor-uri is defined by the frontend. |
| */ |
| if (unlikely((t->fe->monitor_uri_len != 0) && |
| (t->fe->monitor_uri_len == msg->sl.rq.u_l) && |
| !memcmp(&req->data[msg->sl.rq.u], |
| t->fe->monitor_uri, |
| t->fe->monitor_uri_len))) { |
| /* |
| * We have found the monitor URI |
| */ |
| t->flags |= SN_MONITOR; |
| txn->status = 200; |
| client_retnclose(t, &http_200_chunk); |
| goto return_prx_cond; |
| } |
| |
| /* |
| * 3: Maybe we have to copy the original REQURI for the logs ? |
| * Note: we cannot log anymore if the request has been |
| * classified as invalid. |
| */ |
| if (unlikely(t->logs.logwait & LW_REQ)) { |
| /* we have a complete HTTP request that we must log */ |
| if ((txn->uri = pool_alloc(requri)) != NULL) { |
| int urilen = msg->sl.rq.l; |
| |
| if (urilen >= REQURI_LEN) |
| urilen = REQURI_LEN - 1; |
| memcpy(txn->uri, &req->data[msg->som], urilen); |
| txn->uri[urilen] = 0; |
| |
| if (!(t->logs.logwait &= ~LW_REQ)) |
| http_sess_log(t); |
| } else { |
| Alert("HTTP logging : out of memory.\n"); |
| } |
| } |
| |
| |
| /* 4. We may have to convert HTTP/0.9 requests to HTTP/1.0 */ |
| if (unlikely(msg->sl.rq.v_l == 0)) { |
| int delta; |
| char *cur_end; |
| msg->sol = req->data + msg->som; |
| cur_end = msg->sol + msg->sl.rq.l; |
| delta = 0; |
| |
| if (msg->sl.rq.u_l == 0) { |
| /* if no URI was set, add "/" */ |
| delta = buffer_replace2(req, cur_end, cur_end, " /", 2); |
| cur_end += delta; |
| msg->eoh += delta; |
| } |
| /* add HTTP version */ |
| delta = buffer_replace2(req, cur_end, cur_end, " HTTP/1.0\r\n", 11); |
| msg->eoh += delta; |
| cur_end += delta; |
| cur_end = (char *)http_parse_reqline(msg, req->data, |
| HTTP_MSG_RQMETH, |
| msg->sol, cur_end + 1, |
| NULL, NULL); |
| if (unlikely(!cur_end)) |
| goto return_bad_req; |
| |
| /* we have a full HTTP/1.0 request now and we know that |
| * we have either a CR or an LF at <ptr>. |
| */ |
| hdr_idx_set_start(&txn->hdr_idx, msg->sl.rq.l, *cur_end == '\r'); |
| } |
| |
| |
| /* 5: we may need to capture headers */ |
| if (unlikely((t->logs.logwait & LW_REQHDR) && t->fe->req_cap)) |
| capture_headers(req->data + msg->som, &txn->hdr_idx, |
| txn->req.cap, t->fe->req_cap); |
| |
| /* |
| * 6: we will have to evaluate the filters. |
| * As opposed to version 1.2, now they will be evaluated in the |
| * filters order and not in the header order. This means that |
| * each filter has to be validated among all headers. |
| * |
| * We can now check whether we want to switch to another |
| * backend, in which case we will re-check the backend's |
| * filters and various options. In order to support 3-level |
| * switching, here's how we should proceed : |
| * |
| * a) run be. |
| * if (switch) then switch ->be to the new backend. |
| * b) run be if (be != fe). |
| * There cannot be any switch from there, so ->be cannot be |
| * changed anymore. |
| * |
| * => filters always apply to ->be, then ->be may change. |
| * |
| * The response path will be able to apply either ->be, or |
| * ->be then ->fe filters in order to match the reverse of |
| * the forward sequence. |
| */ |
| |
| do { |
| struct acl_cond *cond; |
| struct proxy *rule_set = t->be; |
| cur_proxy = t->be; |
| |
| /* first check whether we have some ACLs set to block this request */ |
| list_for_each_entry(cond, &cur_proxy->block_cond, list) { |
| int ret = acl_exec_cond(cond, cur_proxy, t, txn); |
| if (cond->pol == ACL_COND_UNLESS) |
| ret = !ret; |
| |
| if (ret) { |
| txn->status = 403; |
| /* let's log the request time */ |
| t->logs.t_request = tv_ms_elapsed(&t->logs.tv_accept, &now); |
| client_retnclose(t, error_message(t, HTTP_ERR_403)); |
| goto return_prx_cond; |
| } |
| } |
| |
| /* try headers filters */ |
| if (rule_set->req_exp != NULL) { |
| if (apply_filters_to_request(t, req, rule_set->req_exp) < 0) |
| goto return_bad_req; |
| } |
| |
| if (!(t->flags & SN_BE_ASSIGNED) && (t->be != cur_proxy)) { |
| /* to ensure correct connection accounting on |
| * the backend, we count the connection for the |
| * one managing the queue. |
| */ |
| t->be->beconn++; |
| if (t->be->beconn > t->be->beconn_max) |
| t->be->beconn_max = t->be->beconn; |
| t->be->cum_beconn++; |
| t->flags |= SN_BE_ASSIGNED; |
| } |
| |
| /* has the request been denied ? */ |
| if (txn->flags & TX_CLDENY) { |
| /* no need to go further */ |
| txn->status = 403; |
| /* let's log the request time */ |
| t->logs.t_request = tv_ms_elapsed(&t->logs.tv_accept, &now); |
| client_retnclose(t, error_message(t, HTTP_ERR_403)); |
| goto return_prx_cond; |
| } |
| |
| /* We might have to check for "Connection:" */ |
| if (((t->fe->options | t->be->options) & PR_O_HTTP_CLOSE) && |
| !(t->flags & SN_CONN_CLOSED)) { |
| char *cur_ptr, *cur_end, *cur_next; |
| int cur_idx, old_idx, delta, val; |
| struct hdr_idx_elem *cur_hdr; |
| |
| cur_next = req->data + txn->req.som + hdr_idx_first_pos(&txn->hdr_idx); |
| old_idx = 0; |
| |
| while ((cur_idx = txn->hdr_idx.v[old_idx].next)) { |
| cur_hdr = &txn->hdr_idx.v[cur_idx]; |
| cur_ptr = cur_next; |
| cur_end = cur_ptr + cur_hdr->len; |
| cur_next = cur_end + cur_hdr->cr + 1; |
| |
| val = http_header_match2(cur_ptr, cur_end, "Connection", 10); |
| if (val) { |
| /* 3 possibilities : |
| * - we have already set Connection: close, |
| * so we remove this line. |
| * - we have not yet set Connection: close, |
| * but this line indicates close. We leave |
| * it untouched and set the flag. |
| * - we have not yet set Connection: close, |
| * and this line indicates non-close. We |
| * replace it. |
| */ |
| if (t->flags & SN_CONN_CLOSED) { |
| delta = buffer_replace2(req, cur_ptr, cur_next, NULL, 0); |
| txn->req.eoh += delta; |
| cur_next += delta; |
| txn->hdr_idx.v[old_idx].next = cur_hdr->next; |
| txn->hdr_idx.used--; |
| cur_hdr->len = 0; |
| } else { |
| if (strncasecmp(cur_ptr + val, "close", 5) != 0) { |
| delta = buffer_replace2(req, cur_ptr + val, cur_end, |
| "close", 5); |
| cur_next += delta; |
| cur_hdr->len += delta; |
| txn->req.eoh += delta; |
| } |
| t->flags |= SN_CONN_CLOSED; |
| } |
| } |
| old_idx = cur_idx; |
| } |
| } |
| /* add request headers from the rule sets in the same order */ |
| for (cur_idx = 0; cur_idx < rule_set->nb_reqadd; cur_idx++) { |
| if (unlikely(http_header_add_tail(req, |
| &txn->req, |
| &txn->hdr_idx, |
| rule_set->req_add[cur_idx])) < 0) |
| goto return_bad_req; |
| } |
| |
| /* check if stats URI was requested, and if an auth is needed */ |
| if (rule_set->uri_auth != NULL && |
| (txn->meth == HTTP_METH_GET || txn->meth == HTTP_METH_HEAD)) { |
| /* we have to check the URI and auth for this request */ |
| if (stats_check_uri_auth(t, rule_set)) |
| return 1; |
| } |
| |
| if (!(t->flags & SN_BE_ASSIGNED) && cur_proxy->defbe.be) { |
| /* No backend was set, but there was a default |
| * backend set in the frontend, so we use it and |
| * loop again. |
| */ |
| t->be = cur_proxy->defbe.be; |
| t->be->beconn++; |
| if (t->be->beconn > t->be->beconn_max) |
| t->be->beconn_max = t->be->beconn; |
| t->be->cum_beconn++; |
| t->flags |= SN_BE_ASSIGNED; |
| } |
| } while (t->be != cur_proxy); /* we loop only if t->be has changed */ |
| |
| |
| if (!(t->flags & SN_BE_ASSIGNED)) { |
| /* To ensure correct connection accounting on |
| * the backend, we count the connection for the |
| * one managing the queue. |
| */ |
| t->be->beconn++; |
| if (t->be->beconn > t->be->beconn_max) |
| t->be->beconn_max = t->be->beconn; |
| t->be->cum_beconn++; |
| t->flags |= SN_BE_ASSIGNED; |
| } |
| |
| /* |
| * Right now, we know that we have processed the entire headers |
| * and that unwanted requests have been filtered out. We can do |
| * whatever we want with the remaining request. Also, now we |
| * may have separate values for ->fe, ->be. |
| */ |
| |
| |
| |
| |
| /* |
| * 7: the appsession cookie was looked up very early in 1.2, |
| * so let's do the same now. |
| */ |
| |
| /* It needs to look into the URI */ |
| if (t->be->appsession_name) { |
| get_srv_from_appsession(t, &req->data[msg->som], msg->sl.rq.l); |
| } |
| |
| |
| /* |
| * 8: Now we can work with the cookies. |
| * Note that doing so might move headers in the request, but |
| * the fields will stay coherent and the URI will not move. |
| * This should only be performed in the backend. |
| */ |
| if (!(txn->flags & (TX_CLDENY|TX_CLTARPIT))) |
| manage_client_side_cookies(t, req); |
| |
| |
| /* |
| * 9: add X-Forwarded-For if either the frontend or the backend |
| * asks for it. |
| */ |
| if ((t->fe->options | t->be->options) & PR_O_FWDFOR) { |
| if (t->cli_addr.ss_family == AF_INET) { |
| /* Add an X-Forwarded-For header unless the source IP is |
| * in the 'except' network range. |
| */ |
| if ((!t->fe->except_mask.s_addr || |
| (((struct sockaddr_in *)&t->cli_addr)->sin_addr.s_addr & t->fe->except_mask.s_addr) |
| != t->fe->except_net.s_addr) && |
| (!t->be->except_mask.s_addr || |
| (((struct sockaddr_in *)&t->cli_addr)->sin_addr.s_addr & t->be->except_mask.s_addr) |
| != t->be->except_net.s_addr)) { |
| int len; |
| unsigned char *pn; |
| pn = (unsigned char *)&((struct sockaddr_in *)&t->cli_addr)->sin_addr; |
| |
| len = sprintf(trash, "X-Forwarded-For: %d.%d.%d.%d", |
| pn[0], pn[1], pn[2], pn[3]); |
| |
| if (unlikely(http_header_add_tail2(req, &txn->req, |
| &txn->hdr_idx, trash, len)) < 0) |
| goto return_bad_req; |
| } |
| } |
| else if (t->cli_addr.ss_family == AF_INET6) { |
| /* FIXME: for the sake of completeness, we should also support |
| * 'except' here, although it is mostly useless in this case. |
| */ |
| int len; |
| char pn[INET6_ADDRSTRLEN]; |
| inet_ntop(AF_INET6, |
| (const void *)&((struct sockaddr_in6 *)(&t->cli_addr))->sin6_addr, |
| pn, sizeof(pn)); |
| len = sprintf(trash, "X-Forwarded-For: %s", pn); |
| if (unlikely(http_header_add_tail2(req, &txn->req, |
| &txn->hdr_idx, trash, len)) < 0) |
| goto return_bad_req; |
| } |
| } |
| |
| /* |
| * 10: add "Connection: close" if needed and not yet set. |
| * Note that we do not need to add it in case of HTTP/1.0. |
| */ |
| if (!(t->flags & SN_CONN_CLOSED) && |
| ((t->fe->options | t->be->options) & PR_O_HTTP_CLOSE)) { |
| if ((unlikely(msg->sl.rq.v_l != 8) || |
| unlikely(req->data[msg->som + msg->sl.rq.v + 7] != '0')) && |
| unlikely(http_header_add_tail2(req, &txn->req, &txn->hdr_idx, |
| "Connection: close", 17)) < 0) |
| goto return_bad_req; |
| t->flags |= SN_CONN_CLOSED; |
| } |
| |
| /************************************************************* |
| * OK, that's finished for the headers. We have done what we * |
| * could. Let's switch to the DATA state. * |
| ************************************************************/ |
| |
| t->cli_state = CL_STDATA; |
| req->rlim = req->data + BUFSIZE; /* no more rewrite needed */ |
| |
| t->logs.t_request = tv_ms_elapsed(&t->logs.tv_accept, &now); |
| |
| if (!tv_isset(&t->fe->clitimeout) || |
| (t->srv_state < SV_STDATA && tv_isset(&t->be->srvtimeout))) { |
| /* If the client has no timeout, or if the server is not ready yet, |
| * and we know for sure that it can expire, then it's cleaner to |
| * disable the timeout on the client side so that too low values |
| * cannot make the sessions abort too early. |
| * |
| * FIXME-20050705: the server needs a way to re-enable this time-out |
| * when it switches its state, otherwise a client can stay connected |
| * indefinitely. This now seems to be OK. |
| */ |
| tv_eternity(&req->rex); |
| } |
| |
| /* When a connection is tarpitted, we use the queue timeout for the |
| * tarpit delay, which currently happens to be the server's connect |
| * timeout. If unset, then set it to zero because we really want it |
| * to expire at one moment. |
| */ |
| if (txn->flags & TX_CLTARPIT) { |
| t->req->l = 0; |
| /* flush the request so that we can drop the connection early |
| * if the client closes first. |
| */ |
| if (!tv_add_ifset(&req->cex, &now, &t->be->contimeout)) |
| req->cex = now; |
| } |
| |
| /* OK let's go on with the BODY now */ |
| goto process_data; |
| |
| return_bad_req: /* let's centralize all bad requests */ |
| txn->req.msg_state = HTTP_MSG_ERROR; |
| txn->status = 400; |
| client_retnclose(t, error_message(t, HTTP_ERR_400)); |
| t->fe->failed_req++; |
| return_prx_cond: |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_PRXCOND; |
| if (!(t->flags & SN_FINST_MASK)) |
| t->flags |= SN_FINST_R; |
| return 1; |
| |
| } |
| else if (c == CL_STDATA) { |
| process_data: |
| /* FIXME: this error handling is partly buggy because we always report |
| * a 'DATA' phase while we don't know if the server was in IDLE, CONN |
| * or HEADER phase. BTW, it's not logical to expire the client while |
| * we're waiting for the server to connect. |
| */ |
| /* read or write error */ |
| if (rep->flags & BF_WRITE_ERROR || req->flags & BF_READ_ERROR) { |
| tv_eternity(&req->rex); |
| tv_eternity(&rep->wex); |
| fd_delete(t->cli_fd); |
| t->cli_state = CL_STCLOSE; |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_CLICL; |
| if (!(t->flags & SN_FINST_MASK)) { |
| if (t->pend_pos) |
| t->flags |= SN_FINST_Q; |
| else if (s == SV_STCONN) |
| t->flags |= SN_FINST_C; |
| else |
| t->flags |= SN_FINST_D; |
| } |
| return 1; |
| } |
| /* last read, or end of server write */ |
| else if (req->flags & BF_READ_NULL || s == SV_STSHUTW || s == SV_STCLOSE) { |
| EV_FD_CLR(t->cli_fd, DIR_RD); |
| tv_eternity(&req->rex); |
| t->cli_state = CL_STSHUTR; |
| return 1; |
| } |
| /* last server read and buffer empty */ |
| else if ((s == SV_STSHUTR || s == SV_STCLOSE) && (rep->l == 0)) { |
| EV_FD_CLR(t->cli_fd, DIR_WR); |
| tv_eternity(&rep->wex); |
| shutdown(t->cli_fd, SHUT_WR); |
| /* We must ensure that the read part is still alive when switching |
| * to shutw */ |
| EV_FD_SET(t->cli_fd, DIR_RD); |
| tv_add_ifset(&req->rex, &now, &t->fe->clitimeout); |
| t->cli_state = CL_STSHUTW; |
| //fprintf(stderr,"%p:%s(%d), c=%d, s=%d\n", t, __FUNCTION__, __LINE__, t->cli_state, t->cli_state); |
| return 1; |
| } |
| /* read timeout */ |
| else if (tv_isle(&req->rex, &now)) { |
| EV_FD_CLR(t->cli_fd, DIR_RD); |
| tv_eternity(&req->rex); |
| t->cli_state = CL_STSHUTR; |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_CLITO; |
| if (!(t->flags & SN_FINST_MASK)) { |
| if (t->pend_pos) |
| t->flags |= SN_FINST_Q; |
| else if (s == SV_STCONN) |
| t->flags |= SN_FINST_C; |
| else |
| t->flags |= SN_FINST_D; |
| } |
| return 1; |
| } |
| /* write timeout */ |
| else if (tv_isle(&rep->wex, &now)) { |
| EV_FD_CLR(t->cli_fd, DIR_WR); |
| tv_eternity(&rep->wex); |
| shutdown(t->cli_fd, SHUT_WR); |
| /* We must ensure that the read part is still alive when switching |
| * to shutw */ |
| EV_FD_SET(t->cli_fd, DIR_RD); |
| tv_add_ifset(&req->rex, &now, &t->fe->clitimeout); |
| |
| t->cli_state = CL_STSHUTW; |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_CLITO; |
| if (!(t->flags & SN_FINST_MASK)) { |
| if (t->pend_pos) |
| t->flags |= SN_FINST_Q; |
| else if (s == SV_STCONN) |
| t->flags |= SN_FINST_C; |
| else |
| t->flags |= SN_FINST_D; |
| } |
| return 1; |
| } |
| |
| if (req->l >= req->rlim - req->data) { |
| /* no room to read more data */ |
| if (EV_FD_COND_C(t->cli_fd, DIR_RD)) { |
| /* stop reading until we get some space */ |
| tv_eternity(&req->rex); |
| } |
| } else { |
| /* there's still some space in the buffer */ |
| if (EV_FD_COND_S(t->cli_fd, DIR_RD)) { |
| if (!tv_isset(&t->fe->clitimeout) || |
| (t->srv_state < SV_STDATA && tv_isset(&t->be->srvtimeout))) |
| /* If the client has no timeout, or if the server not ready yet, and we |
| * know for sure that it can expire, then it's cleaner to disable the |
| * timeout on the client side so that too low values cannot make the |
| * sessions abort too early. |
| */ |
| tv_eternity(&req->rex); |
| else |
| tv_add(&req->rex, &now, &t->fe->clitimeout); |
| } |
| } |
| |
| if ((rep->l == 0) || |
| ((s < SV_STDATA) /* FIXME: this may be optimized && (rep->w == rep->h)*/)) { |
| if (EV_FD_COND_C(t->cli_fd, DIR_WR)) { |
| /* stop writing */ |
| tv_eternity(&rep->wex); |
| } |
| } else { |
| /* buffer not empty */ |
| if (EV_FD_COND_S(t->cli_fd, DIR_WR)) { |
| /* restart writing */ |
| if (tv_add_ifset(&rep->wex, &now, &t->fe->clitimeout)) { |
| /* FIXME: to prevent the client from expiring read timeouts during writes, |
| * we refresh it. */ |
| req->rex = rep->wex; |
| } |
| else |
| tv_eternity(&rep->wex); |
| } |
| } |
| return 0; /* other cases change nothing */ |
| } |
| else if (c == CL_STSHUTR) { |
| if (rep->flags & BF_WRITE_ERROR) { |
| tv_eternity(&rep->wex); |
| fd_delete(t->cli_fd); |
| t->cli_state = CL_STCLOSE; |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_CLICL; |
| if (!(t->flags & SN_FINST_MASK)) { |
| if (t->pend_pos) |
| t->flags |= SN_FINST_Q; |
| else if (s == SV_STCONN) |
| t->flags |= SN_FINST_C; |
| else |
| t->flags |= SN_FINST_D; |
| } |
| return 1; |
| } |
| else if ((s == SV_STSHUTR || s == SV_STCLOSE) && (rep->l == 0) |
| && !(t->flags & SN_SELF_GEN)) { |
| tv_eternity(&rep->wex); |
| fd_delete(t->cli_fd); |
| t->cli_state = CL_STCLOSE; |
| return 1; |
| } |
| else if (tv_isle(&rep->wex, &now)) { |
| tv_eternity(&rep->wex); |
| fd_delete(t->cli_fd); |
| t->cli_state = CL_STCLOSE; |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_CLITO; |
| if (!(t->flags & SN_FINST_MASK)) { |
| if (t->pend_pos) |
| t->flags |= SN_FINST_Q; |
| else if (s == SV_STCONN) |
| t->flags |= SN_FINST_C; |
| else |
| t->flags |= SN_FINST_D; |
| } |
| return 1; |
| } |
| |
| if (t->flags & SN_SELF_GEN) { |
| produce_content(t); |
| if (rep->l == 0) { |
| tv_eternity(&rep->wex); |
| fd_delete(t->cli_fd); |
| t->cli_state = CL_STCLOSE; |
| return 1; |
| } |
| } |
| |
| if ((rep->l == 0) |
| || ((s == SV_STHEADERS) /* FIXME: this may be optimized && (rep->w == rep->h)*/)) { |
| if (EV_FD_COND_C(t->cli_fd, DIR_WR)) { |
| /* stop writing */ |
| tv_eternity(&rep->wex); |
| } |
| } else { |
| /* buffer not empty */ |
| if (EV_FD_COND_S(t->cli_fd, DIR_WR)) { |
| /* restart writing */ |
| if (tv_add_ifset(&rep->wex, &now, &t->fe->clitimeout)) { |
| /* FIXME: to prevent the client from expiring read timeouts during writes, |
| * we refresh it. */ |
| req->rex = rep->wex; |
| } |
| else |
| tv_eternity(&rep->wex); |
| } |
| } |
| return 0; |
| } |
| else if (c == CL_STSHUTW) { |
| if (req->flags & BF_READ_ERROR) { |
| tv_eternity(&req->rex); |
| fd_delete(t->cli_fd); |
| t->cli_state = CL_STCLOSE; |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_CLICL; |
| if (!(t->flags & SN_FINST_MASK)) { |
| if (t->pend_pos) |
| t->flags |= SN_FINST_Q; |
| else if (s == SV_STCONN) |
| t->flags |= SN_FINST_C; |
| else |
| t->flags |= SN_FINST_D; |
| } |
| return 1; |
| } |
| else if (req->flags & BF_READ_NULL || s == SV_STSHUTW || s == SV_STCLOSE) { |
| tv_eternity(&req->rex); |
| fd_delete(t->cli_fd); |
| t->cli_state = CL_STCLOSE; |
| return 1; |
| } |
| else if (tv_isle(&req->rex, &now)) { |
| tv_eternity(&req->rex); |
| fd_delete(t->cli_fd); |
| t->cli_state = CL_STCLOSE; |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_CLITO; |
| if (!(t->flags & SN_FINST_MASK)) { |
| if (t->pend_pos) |
| t->flags |= SN_FINST_Q; |
| else if (s == SV_STCONN) |
| t->flags |= SN_FINST_C; |
| else |
| t->flags |= SN_FINST_D; |
| } |
| return 1; |
| } |
| else if (req->l >= req->rlim - req->data) { |
| /* no room to read more data */ |
| |
| /* FIXME-20050705: is it possible for a client to maintain a session |
| * after the timeout by sending more data after it receives a close ? |
| */ |
| |
| if (EV_FD_COND_C(t->cli_fd, DIR_RD)) { |
| /* stop reading until we get some space */ |
| tv_eternity(&req->rex); |
| //fprintf(stderr,"%p:%s(%d), c=%d, s=%d\n", t, __FUNCTION__, __LINE__, t->cli_state, t->cli_state); |
| } |
| } else { |
| /* there's still some space in the buffer */ |
| if (EV_FD_COND_S(t->cli_fd, DIR_RD)) { |
| if (!tv_add_ifset(&req->rex, &now, &t->fe->clitimeout)) |
| tv_eternity(&req->rex); |
| //fprintf(stderr,"%p:%s(%d), c=%d, s=%d\n", t, __FUNCTION__, __LINE__, t->cli_state, t->cli_state); |
| } |
| } |
| return 0; |
| } |
| else { /* CL_STCLOSE: nothing to do */ |
| if ((global.mode & MODE_DEBUG) && (!(global.mode & MODE_QUIET) || (global.mode & MODE_VERBOSE))) { |
| int len; |
| len = sprintf(trash, "%08x:%s.clicls[%04x:%04x]\n", t->uniq_id, t->be->id, (unsigned short)t->cli_fd, (unsigned short)t->srv_fd); |
| write(1, trash, len); |
| } |
| return 0; |
| } |
| return 0; |
| } |
| |
| |
| /* |
| * manages the server FSM and its socket. It returns 1 if a state has changed |
| * (and a resync may be needed), 0 else. |
| */ |
| int process_srv(struct session *t) |
| { |
| int s = t->srv_state; |
| int c = t->cli_state; |
| struct http_txn *txn = &t->txn; |
| struct buffer *req = t->req; |
| struct buffer *rep = t->rep; |
| int conn_err; |
| |
| #ifdef DEBUG_FULL |
| fprintf(stderr,"process_srv: c=%s, s=%s\n", cli_stnames[c], srv_stnames[s]); |
| #endif |
| //fprintf(stderr,"process_srv: c=%d, s=%d, cr=%d, cw=%d, sr=%d, sw=%d\n", c, s, |
| //EV_FD_ISSET(t->cli_fd, DIR_RD), EV_FD_ISSET(t->cli_fd, DIR_WR), |
| //EV_FD_ISSET(t->srv_fd, DIR_RD), EV_FD_ISSET(t->srv_fd, DIR_WR) |
| //); |
| if (s == SV_STIDLE) { |
| if (c == CL_STHEADERS) |
| return 0; /* stay in idle, waiting for data to reach the client side */ |
| else if (c == CL_STCLOSE || c == CL_STSHUTW || |
| (c == CL_STSHUTR && |
| (t->req->l == 0 || t->be->options & PR_O_ABRT_CLOSE))) { /* give up */ |
| tv_eternity(&req->cex); |
| if (t->pend_pos) |
| t->logs.t_queue = tv_ms_elapsed(&t->logs.tv_accept, &now); |
| /* note that this must not return any error because it would be able to |
| * overwrite the client_retnclose() output. |
| */ |
| if (txn->flags & TX_CLTARPIT) |
| srv_close_with_err(t, SN_ERR_CLICL, SN_FINST_T, 0, NULL); |
| else |
| srv_close_with_err(t, SN_ERR_CLICL, t->pend_pos ? SN_FINST_Q : SN_FINST_C, 0, NULL); |
| |
| return 1; |
| } |
| else { |
| if (txn->flags & TX_CLTARPIT) { |
| /* This connection is being tarpitted. The CLIENT side has |
| * already set the connect expiration date to the right |
| * timeout. We just have to check that it has not expired. |
| */ |
| if (!tv_isle(&req->cex, &now)) |
| return 0; |
| |
| /* We will set the queue timer to the time spent, just for |
| * logging purposes. We fake a 500 server error, so that the |
| * attacker will not suspect his connection has been tarpitted. |
| * It will not cause trouble to the logs because we can exclude |
| * the tarpitted connections by filtering on the 'PT' status flags. |
| */ |
| tv_eternity(&req->cex); |
| t->logs.t_queue = tv_ms_elapsed(&t->logs.tv_accept, &now); |
| srv_close_with_err(t, SN_ERR_PRXCOND, SN_FINST_T, |
| 500, error_message(t, HTTP_ERR_500)); |
| return 1; |
| } |
| |
| /* Right now, we will need to create a connection to the server. |
| * We might already have tried, and got a connection pending, in |
| * which case we will not do anything till it's pending. It's up |
| * to any other session to release it and wake us up again. |
| */ |
| if (t->pend_pos) { |
| if (!tv_isle(&req->cex, &now)) |
| return 0; |
| else { |
| /* we've been waiting too long here */ |
| tv_eternity(&req->cex); |
| t->logs.t_queue = tv_ms_elapsed(&t->logs.tv_accept, &now); |
| srv_close_with_err(t, SN_ERR_SRVTO, SN_FINST_Q, |
| 503, error_message(t, HTTP_ERR_503)); |
| if (t->srv) |
| t->srv->failed_conns++; |
| t->fe->failed_conns++; |
| return 1; |
| } |
| } |
| |
| do { |
| /* first, get a connection */ |
| if (srv_redispatch_connect(t)) |
| return t->srv_state != SV_STIDLE; |
| |
| /* try to (re-)connect to the server, and fail if we expire the |
| * number of retries. |
| */ |
| if (srv_retryable_connect(t)) { |
| t->logs.t_queue = tv_ms_elapsed(&t->logs.tv_accept, &now); |
| return t->srv_state != SV_STIDLE; |
| } |
| } while (1); |
| } |
| } |
| else if (s == SV_STCONN) { /* connection in progress */ |
| if (c == CL_STCLOSE || c == CL_STSHUTW || |
| (c == CL_STSHUTR && |
| ((t->req->l == 0 && !(req->flags & BF_WRITE_STATUS)) || |
| t->be->options & PR_O_ABRT_CLOSE))) { /* give up */ |
| tv_eternity(&req->cex); |
| fd_delete(t->srv_fd); |
| if (t->srv) |
| t->srv->cur_sess--; |
| |
| /* note that this must not return any error because it would be able to |
| * overwrite the client_retnclose() output. |
| */ |
| srv_close_with_err(t, SN_ERR_CLICL, SN_FINST_C, 0, NULL); |
| return 1; |
| } |
| if (!(req->flags & BF_WRITE_STATUS) && !tv_isle(&req->cex, &now)) { |
| //fprintf(stderr,"1: c=%d, s=%d, now=%d.%06d, exp=%d.%06d\n", c, s, now.tv_sec, now.tv_usec, req->cex.tv_sec, req->cex.tv_usec); |
| return 0; /* nothing changed */ |
| } |
| else if (!(req->flags & BF_WRITE_STATUS) || (req->flags & BF_WRITE_ERROR)) { |
| /* timeout, asynchronous connect error or first write error */ |
| //fprintf(stderr,"2: c=%d, s=%d\n", c, s); |
| |
| fd_delete(t->srv_fd); |
| if (t->srv) |
| t->srv->cur_sess--; |
| |
| if (!(req->flags & BF_WRITE_STATUS)) |
| conn_err = SN_ERR_SRVTO; // it was a connect timeout. |
| else |
| conn_err = SN_ERR_SRVCL; // it was an asynchronous connect error. |
| |
| /* ensure that we have enough retries left */ |
| if (srv_count_retry_down(t, conn_err)) |
| return 1; |
| |
| if (t->srv && t->conn_retries == 0 && t->be->options & PR_O_REDISP) { |
| /* We're on our last chance, and the REDISP option was specified. |
| * We will ignore cookie and force to balance or use the dispatcher. |
| */ |
| /* let's try to offer this slot to anybody */ |
| if (may_dequeue_tasks(t->srv, t->be)) |
| task_wakeup(t->srv->queue_mgt); |
| |
| if (t->srv) |
| t->srv->failed_conns++; |
| t->be->failed_conns++; |
| |
| t->flags &= ~(SN_DIRECT | SN_ASSIGNED | SN_ADDR_SET); |
| t->srv = NULL; /* it's left to the dispatcher to choose a server */ |
| http_flush_cookie_flags(txn); |
| |
| /* first, get a connection */ |
| if (srv_redispatch_connect(t)) |
| return t->srv_state != SV_STIDLE; |
| } |
| |
| do { |
| /* Now we will try to either reconnect to the same server or |
| * connect to another server. If the connection gets queued |
| * because all servers are saturated, then we will go back to |
| * the SV_STIDLE state. |
| */ |
| if (srv_retryable_connect(t)) { |
| t->logs.t_queue = tv_ms_elapsed(&t->logs.tv_accept, &now); |
| return t->srv_state != SV_STCONN; |
| } |
| |
| /* we need to redispatch the connection to another server */ |
| if (srv_redispatch_connect(t)) |
| return t->srv_state != SV_STCONN; |
| } while (1); |
| } |
| else { /* no error or write 0 */ |
| t->logs.t_connect = tv_ms_elapsed(&t->logs.tv_accept, &now); |
| |
| //fprintf(stderr,"3: c=%d, s=%d\n", c, s); |
| if (req->l == 0) /* nothing to write */ { |
| EV_FD_CLR(t->srv_fd, DIR_WR); |
| tv_eternity(&req->wex); |
| } else /* need the right to write */ { |
| EV_FD_SET(t->srv_fd, DIR_WR); |
| if (tv_add_ifset(&req->wex, &now, &t->be->srvtimeout)) { |
| /* FIXME: to prevent the server from expiring read timeouts during writes, |
| * we refresh it. */ |
| rep->rex = req->wex; |
| } |
| else |
| tv_eternity(&req->wex); |
| } |
| |
| if (t->be->mode == PR_MODE_TCP) { /* let's allow immediate data connection in this case */ |
| EV_FD_SET(t->srv_fd, DIR_RD); |
| if (!tv_add_ifset(&rep->rex, &now, &t->be->srvtimeout)) |
| tv_eternity(&rep->rex); |
| |
| t->srv_state = SV_STDATA; |
| if (t->srv) |
| t->srv->cum_sess++; |
| rep->rlim = rep->data + BUFSIZE; /* no rewrite needed */ |
| |
| /* if the user wants to log as soon as possible, without counting |
| bytes from the server, then this is the right moment. */ |
| if (t->fe->to_log && !(t->logs.logwait & LW_BYTES)) { |
| t->logs.t_close = t->logs.t_connect; /* to get a valid end date */ |
| tcp_sess_log(t); |
| } |
| #ifdef CONFIG_HAP_TCPSPLICE |
| if ((t->fe->options & t->be->options) & PR_O_TCPSPLICE) { |
| /* TCP splicing supported by both FE and BE */ |
| tcp_splice_splicefd(t->cli_fd, t->srv_fd, 0); |
| } |
| #endif |
| } |
| else { |
| t->srv_state = SV_STHEADERS; |
| if (t->srv) |
| t->srv->cum_sess++; |
| rep->rlim = rep->data + BUFSIZE - MAXREWRITE; /* rewrite needed */ |
| t->txn.rsp.msg_state = HTTP_MSG_RPBEFORE; |
| /* reset hdr_idx which was already initialized by the request. |
| * right now, the http parser does it. |
| * hdr_idx_init(&t->txn.hdr_idx); |
| */ |
| } |
| tv_eternity(&req->cex); |
| return 1; |
| } |
| } |
| else if (s == SV_STHEADERS) { /* receiving server headers */ |
| /* |
| * Now parse the partial (or complete) lines. |
| * We will check the response syntax, and also join multi-line |
| * headers. An index of all the lines will be elaborated while |
| * parsing. |
| * |
| * For the parsing, we use a 28 states FSM. |
| * |
| * Here is the information we currently have : |
| * rep->data + req->som = beginning of response |
| * rep->data + req->eoh = end of processed headers / start of current one |
| * rep->data + req->eol = end of current header or line (LF or CRLF) |
| * rep->lr = first non-visited byte |
| * rep->r = end of data |
| */ |
| |
| int cur_idx; |
| struct http_msg *msg = &txn->rsp; |
| struct proxy *cur_proxy; |
| |
| if (likely(rep->lr < rep->r)) |
| http_msg_analyzer(rep, msg, &txn->hdr_idx); |
| |
| /* 1: we might have to print this header in debug mode */ |
| if (unlikely((global.mode & MODE_DEBUG) && |
| (!(global.mode & MODE_QUIET) || (global.mode & MODE_VERBOSE)) && |
| (msg->msg_state == HTTP_MSG_BODY || msg->msg_state == HTTP_MSG_ERROR))) { |
| char *eol, *sol; |
| |
| sol = rep->data + msg->som; |
| eol = sol + msg->sl.rq.l; |
| debug_hdr("srvrep", t, sol, eol); |
| |
| sol += hdr_idx_first_pos(&txn->hdr_idx); |
| cur_idx = hdr_idx_first_idx(&txn->hdr_idx); |
| |
| while (cur_idx) { |
| eol = sol + txn->hdr_idx.v[cur_idx].len; |
| debug_hdr("srvhdr", t, sol, eol); |
| sol = eol + txn->hdr_idx.v[cur_idx].cr + 1; |
| cur_idx = txn->hdr_idx.v[cur_idx].next; |
| } |
| } |
| |
| |
| if ((rep->l < rep->rlim - rep->data) && EV_FD_COND_S(t->srv_fd, DIR_RD)) { |
| /* fd in DIR_RD was disabled, perhaps because of a previous buffer |
| * full. We cannot loop here since stream_sock_read will disable it only if |
| * rep->l == rlim-data |
| */ |
| if (!tv_add_ifset(&rep->rex, &now, &t->be->srvtimeout)) |
| tv_eternity(&rep->rex); |
| } |
| |
| |
| /* |
| * Now we quickly check if we have found a full valid response. |
| * If not so, we check the FD and buffer states before leaving. |
| * A full response is indicated by the fact that we have seen |
| * the double LF/CRLF, so the state is HTTP_MSG_BODY. Invalid |
| * responses are checked first. |
| * |
| * Depending on whether the client is still there or not, we |
| * may send an error response back or not. Note that normally |
| * we should only check for HTTP status there, and check I/O |
| * errors somewhere else. |
| */ |
| |
| if (unlikely(msg->msg_state != HTTP_MSG_BODY)) { |
| |
| /* Invalid response, or read error or write error */ |
| if (unlikely((msg->msg_state == HTTP_MSG_ERROR) || |
| (req->flags & BF_WRITE_ERROR) || |
| (rep->flags & BF_READ_ERROR))) { |
| tv_eternity(&rep->rex); |
| tv_eternity(&req->wex); |
| fd_delete(t->srv_fd); |
| if (t->srv) { |
| t->srv->cur_sess--; |
| t->srv->failed_resp++; |
| } |
| t->be->failed_resp++; |
| t->srv_state = SV_STCLOSE; |
| txn->status = 502; |
| client_return(t, error_message(t, HTTP_ERR_502)); |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_SRVCL; |
| if (!(t->flags & SN_FINST_MASK)) |
| t->flags |= SN_FINST_H; |
| /* We used to have a free connection slot. Since we'll never use it, |
| * we have to inform the server that it may be used by another session. |
| */ |
| if (t->srv && may_dequeue_tasks(t->srv, t->be)) |
| task_wakeup(t->srv->queue_mgt); |
| |
| return 1; |
| } |
| |
| /* end of client write or end of server read. |
| * since we are in header mode, if there's no space left for headers, we |
| * won't be able to free more later, so the session will never terminate. |
| */ |
| else if (unlikely(rep->flags & BF_READ_NULL || |
| c == CL_STSHUTW || c == CL_STCLOSE || |
| rep->l >= rep->rlim - rep->data)) { |
| EV_FD_CLR(t->srv_fd, DIR_RD); |
| tv_eternity(&rep->rex); |
| t->srv_state = SV_STSHUTR; |
| //fprintf(stderr,"%p:%s(%d), c=%d, s=%d\n", t, __FUNCTION__, __LINE__, t->cli_state, t->cli_state); |
| return 1; |
| } |
| |
| /* read timeout : return a 504 to the client. |
| */ |
| else if (unlikely(EV_FD_ISSET(t->srv_fd, DIR_RD) && |
| tv_isle(&rep->rex, &now))) { |
| tv_eternity(&rep->rex); |
| tv_eternity(&req->wex); |
| fd_delete(t->srv_fd); |
| if (t->srv) { |
| t->srv->cur_sess--; |
| t->srv->failed_resp++; |
| } |
| t->be->failed_resp++; |
| t->srv_state = SV_STCLOSE; |
| txn->status = 504; |
| client_return(t, error_message(t, HTTP_ERR_504)); |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_SRVTO; |
| if (!(t->flags & SN_FINST_MASK)) |
| t->flags |= SN_FINST_H; |
| /* We used to have a free connection slot. Since we'll never use it, |
| * we have to inform the server that it may be used by another session. |
| */ |
| if (t->srv && may_dequeue_tasks(t->srv, t->be)) |
| task_wakeup(t->srv->queue_mgt); |
| return 1; |
| } |
| |
| /* last client read and buffer empty */ |
| /* FIXME!!! here, we don't want to switch to SHUTW if the |
| * client shuts read too early, because we may still have |
| * some work to do on the headers. |
| * The side-effect is that if the client completely closes its |
| * connection during SV_STHEADER, the connection to the server |
| * is kept until a response comes back or the timeout is reached. |
| */ |
| else if (unlikely((/*c == CL_STSHUTR ||*/ c == CL_STCLOSE) && |
| (req->l == 0))) { |
| EV_FD_CLR(t->srv_fd, DIR_WR); |
| tv_eternity(&req->wex); |
| |
| /* We must ensure that the read part is still |
| * alive when switching to shutw */ |
| EV_FD_SET(t->srv_fd, DIR_RD); |
| tv_add_ifset(&rep->rex, &now, &t->be->srvtimeout); |
| |
| shutdown(t->srv_fd, SHUT_WR); |
| t->srv_state = SV_STSHUTW; |
| return 1; |
| } |
| |
| /* write timeout */ |
| /* FIXME!!! here, we don't want to switch to SHUTW if the |
| * client shuts read too early, because we may still have |
| * some work to do on the headers. |
| */ |
| else if (unlikely(EV_FD_ISSET(t->srv_fd, DIR_WR) && |
| tv_isle(&req->wex, &now))) { |
| EV_FD_CLR(t->srv_fd, DIR_WR); |
| tv_eternity(&req->wex); |
| shutdown(t->srv_fd, SHUT_WR); |
| /* We must ensure that the read part is still alive |
| * when switching to shutw */ |
| EV_FD_SET(t->srv_fd, DIR_RD); |
| tv_add_ifset(&rep->rex, &now, &t->be->srvtimeout); |
| |
| t->srv_state = SV_STSHUTW; |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_SRVTO; |
| if (!(t->flags & SN_FINST_MASK)) |
| t->flags |= SN_FINST_H; |
| return 1; |
| } |
| |
| /* |
| * And now the non-error cases. |
| */ |
| |
| /* Data remaining in the request buffer. |
| * This happens during the first pass here, and during |
| * long posts. |
| */ |
| else if (likely(req->l)) { |
| if (EV_FD_COND_S(t->srv_fd, DIR_WR)) { |
| /* restart writing */ |
| if (tv_add_ifset(&req->wex, &now, &t->be->srvtimeout)) { |
| /* FIXME: to prevent the server from expiring read timeouts during writes, |
| * we refresh it. */ |
| rep->rex = req->wex; |
| } |
| else |
| tv_eternity(&req->wex); |
| } |
| } |
| |
| /* nothing left in the request buffer */ |
| else { |
| if (EV_FD_COND_C(t->srv_fd, DIR_WR)) { |
| /* stop writing */ |
| tv_eternity(&req->wex); |
| } |
| } |
| |
| return t->srv_state != SV_STHEADERS; |
| } |
| |
| |
| /***************************************************************** |
| * More interesting part now : we know that we have a complete * |
| * response which at least looks like HTTP. We have an indicator * |
| * of each header's length, so we can parse them quickly. * |
| ****************************************************************/ |
| |
| /* ensure we keep this pointer to the beginning of the message */ |
| msg->sol = rep->data + msg->som; |
| |
| /* |
| * 1: get the status code and check for cacheability. |
| */ |
| |
| t->logs.logwait &= ~LW_RESP; |
| txn->status = strl2ui(rep->data + msg->sl.st.c, msg->sl.st.c_l); |
| |
| switch (txn->status) { |
| case 200: |
| case 203: |
| case 206: |
| case 300: |
| case 301: |
| case 410: |
| /* RFC2616 @13.4: |
| * "A response received with a status code of |
| * 200, 203, 206, 300, 301 or 410 MAY be stored |
| * by a cache (...) unless a cache-control |
| * directive prohibits caching." |
| * |
| * RFC2616 @9.5: POST method : |
| * "Responses to this method are not cacheable, |
| * unless the response includes appropriate |
| * Cache-Control or Expires header fields." |
| */ |
| if (likely(txn->meth != HTTP_METH_POST) && |
| unlikely(t->be->options & PR_O_CHK_CACHE)) |
| txn->flags |= TX_CACHEABLE | TX_CACHE_COOK; |
| break; |
| default: |
| break; |
| } |
| |
| /* |
| * 2: we may need to capture headers |
| */ |
| if (unlikely((t->logs.logwait & LW_RSPHDR) && t->fe->rsp_cap)) |
| capture_headers(rep->data + msg->som, &txn->hdr_idx, |
| txn->rsp.cap, t->fe->rsp_cap); |
| |
| /* |
| * 3: we will have to evaluate the filters. |
| * As opposed to version 1.2, now they will be evaluated in the |
| * filters order and not in the header order. This means that |
| * each filter has to be validated among all headers. |
| * |
| * Filters are tried with ->be first, then with ->fe if it is |
| * different from ->be. |
| */ |
| |
| t->flags &= ~SN_CONN_CLOSED; /* prepare for inspection */ |
| |
| cur_proxy = t->be; |
| while (1) { |
| struct proxy *rule_set = cur_proxy; |
| |
| /* try headers filters */ |
| if (rule_set->rsp_exp != NULL) { |
| if (apply_filters_to_response(t, rep, rule_set->rsp_exp) < 0) { |
| return_bad_resp: |
| if (t->srv) { |
| t->srv->cur_sess--; |
| t->srv->failed_resp++; |
| } |
| cur_proxy->failed_resp++; |
| return_srv_prx_502: |
| tv_eternity(&rep->rex); |
| tv_eternity(&req->wex); |
| fd_delete(t->srv_fd); |
| t->srv_state = SV_STCLOSE; |
| txn->status = 502; |
| client_return(t, error_message(t, HTTP_ERR_502)); |
| if (!(t->flags & SN_ERR_MASK)) |
| t->flags |= SN_ERR_PRXCOND; |
| if (!(t->flags & SN_FINST_MASK)) |
| t->flags |= SN_FINST_H; |
| /* We used to have a free connection slot. Since we'll never use it, |
| * we have to inform the server that it may be used by another session. |
| */ |
| if (t->srv && may_dequeue_tasks(t->srv, cur_proxy)) |
| task_wakeup(t->srv->queue_mgt); |
| return 1; |
| } |
| } |
| |
| /* has the response been denied ? */ |
| if (txn->flags & TX_SVDENY) { |
| if (t->srv) { |
| t->srv->cur_sess--; |
| t->srv->failed_secu++; |
| } |
| cur_proxy->denied_resp++; |
| goto return_srv_prx_502; |
| } |
| |
| /* We might have to check for "Connection:" */ |
| if (((t->fe->options | t->be->options) & PR_O_HTTP_CLOSE) && |
| !(t->flags & SN_CONN_CLOSED)) { |
| char *cur_ptr, *cur_end, *cur_next; |
| int cur_idx, old_idx, delta, val; |
| struct hdr_idx_elem *cur_hdr; |
| |
| cur_next = rep->data + txn->rsp.som + hdr_idx_first_pos(&txn->hdr_idx); |
| old_idx = 0; |
| |
| while ((cur_idx = txn->hdr_idx.v[old_idx].next)) { |
| cur_hdr = &txn->hdr_idx.v[cur_idx]; |
| cur_ptr = cur_next; |
| cur_end = cur_ptr + cur_hdr->len; |
| cur_next = cur_end + cur_hdr->cr + 1; |
| |
| val = http_header_match2(cur_ptr, cur_end, "Connection", 10); |
| if (val) { |
| /* 3 possibilities : |
| * - we have already set Connection: close, |
| * so we remove this line. |
| * - we have not yet set Connection: close, |
| * but this line indicates close. We leave |
| * it untouched and set the flag. |
| * - we have not yet set Connection: close, |
| * and this line indicates non-close. We |
| * replace it. |
| */ |
| if (t->flags & SN_CONN_CLOSED) { |
| delta = buffer_replace2(rep, cur_ptr, cur_next, NULL, 0); |
| txn->rsp.eoh += delta; |
| cur_next += delta; |
| txn->hdr_idx.v[old_idx].next = cur_hdr->next; |
| txn->hdr_idx.used--; |
| cur_hdr->len = 0; |
| } else { |
| if (strncasecmp(cur_ptr + val, "close", 5) != 0) { |
| delta = buffer_replace2(rep, cur_ptr + val, cur_end, |
| "close", 5); |
| cur_next += delta; |
| cur_hdr->len += delta; |
| txn->rsp.eoh += delta; |
| } |
| t->flags |= SN_CONN_CLOSED; |
| } |
| } |
| old_idx = cur_idx; |
| } |
| } |
| |
| /* add response headers from the rule sets in the same order */ |
| for (cur_idx = 0; cur_idx < rule_set->nb_rspadd; cur_idx++) { |
| if (unlikely(http_header_add_tail(rep, &txn->rsp, &txn->hdr_idx, |
| rule_set->rsp_add[cur_idx])) < 0) |
| goto return_bad_resp; |
| } |
| |
| /* check whether we're already working on the frontend */ |
| if (cur_proxy == t->fe) |
| break; |
| cur_proxy = t->fe; |
| } |
| |
| /* |
| * 4: check for server cookie. |
| */ |
| manage_server_side_cookies(t, rep); |
| |
| /* |
| * 5: add server cookie in the response if needed |
| */ |
| if ((t->srv) && !(t->flags & SN_DIRECT) && (t->be->options & PR_O_COOK_INS) && |
| (!(t->be->options & PR_O_COOK_POST) || (txn->meth == HTTP_METH_POST))) { |
| int len; |
| |
| /* the server is known, it's not the one the client requested, we have to |
| * insert a set-cookie here, except if we want to insert only on POST |
| * requests and this one isn't. Note that servers which don't have cookies |
| * (eg: some backup servers) will return a full cookie removal request. |
| */ |
| len = sprintf(trash, "Set-Cookie: %s=%s; path=/", |
| t->be->cookie_name, |
| t->srv->cookie ? t->srv->cookie : "; Expires=Thu, 01-Jan-1970 00:00:01 GMT"); |
| |
| if (unlikely(http_header_add_tail2(rep, &txn->rsp, &txn->hdr_idx, |
| trash, len)) < 0) |
| goto return_bad_resp; |
| txn->flags |= TX_SCK_INSERTED; |
| |
| /* Here, we will tell an eventual cache on the client side that we don't |
| * want it to cache this reply because HTTP/1.0 caches also cache cookies ! |
| * Some caches understand the correct form: 'no-cache="set-cookie"', but |
| * others don't (eg: apache <= 1.3.26). So we use 'private' instead. |
| */ |
| if (t->be->options & PR_O_COOK_NOC) { |
| |