lighttpd1.4/src/mod_cgi.c

#include "first.h"

#include "server.h"
#include "stat_cache.h"
#include "keyvalue.h"
#include "log.h"
#include "connections.h"
#include "joblist.h"
#include "response.h"
#include "http_chunk.h"

#include "plugin.h"

#include <sys/types.h>
#include "sys-mmap.h"

#ifdef __WIN32
# include <winsock2.h>
#else
# include <sys/socket.h>
# include <sys/wait.h>
# include <netinet/in.h>
# include <arpa/inet.h>
#endif

#include <unistd.h>
#include <errno.h>
#include <stdlib.h>
#include <string.h>
#include <fdevent.h>
#include <signal.h>
#include <ctype.h>
#include <assert.h>

#include <stdio.h>
#include <fcntl.h>

#ifdef HAVE_PIPE2
#define pipe_cloexec(pipefd) pipe2((pipefd), O_CLOEXEC)
#elif defined FD_CLOEXEC
#define pipe_cloexec(pipefd) \
  (   0 == pipe(pipefd) \
   && 0 == fcntl(pipefd[0], F_SETFD, FD_CLOEXEC) \
   && 0 == fcntl(pipefd[1], F_SETFD, FD_CLOEXEC) \
    ? 0 \
    : -1)
#else
#define pipe_cloexec(pipefd) pipe(pipefd)
#endif

enum {EOL_UNSET, EOL_N, EOL_RN};

typedef struct {
	char **ptr;

	size_t size;
	size_t used;
} char_array;

typedef struct {
	pid_t *ptr;
	size_t used;
	size_t size;
} buffer_pid_t;

typedef struct {
	array *cgi;
	unsigned short execute_x_only;
	unsigned short xsendfile_allow;
	array *xsendfile_docroot;
} plugin_config;

typedef struct {
	PLUGIN_DATA;
	buffer_pid_t cgi_pid;

	buffer *parse_response;

	plugin_config **config_storage;

	plugin_config conf;
} plugin_data;

typedef struct {
	pid_t pid;
	int fd;
	int fdtocgi;
	int fde_ndx; /* index into the fd-event buffer */
	int fde_ndx_tocgi; /* index into the fd-event buffer */

	connection *remote_conn;  /* dumb pointer */
	plugin_data *plugin_data; /* dumb pointer */

	buffer *response;
	buffer *response_header;
	buffer *cgi_handler;      /* dumb pointer */
	plugin_config conf;
} handler_ctx;

static handler_ctx * cgi_handler_ctx_init(void) {
	handler_ctx *hctx = calloc(1, sizeof(*hctx));

	force_assert(hctx);

	hctx->response = buffer_init();
	hctx->response_header = buffer_init();
	hctx->fd = -1;
	hctx->fdtocgi = -1;

	return hctx;
}

static void cgi_handler_ctx_free(handler_ctx *hctx) {
	buffer_free(hctx->response);
	buffer_free(hctx->response_header);

	free(hctx);
}

enum {FDEVENT_HANDLED_UNSET, FDEVENT_HANDLED_FINISHED, FDEVENT_HANDLED_NOT_FINISHED, FDEVENT_HANDLED_COMEBACK, FDEVENT_HANDLED_ERROR};

INIT_FUNC(mod_cgi_init) {
	plugin_data *p;

	p = calloc(1, sizeof(*p));

	force_assert(p);

	p->parse_response = buffer_init();

	return p;
}


FREE_FUNC(mod_cgi_free) {
	plugin_data *p = p_d;
	buffer_pid_t *r = &(p->cgi_pid);

	UNUSED(srv);

	if (p->config_storage) {
		size_t i;
		for (i = 0; i < srv->config_context->used; i++) {
			plugin_config *s = p->config_storage[i];

			if (NULL == s) continue;

			array_free(s->cgi);
			array_free(s->xsendfile_docroot);

			free(s);
		}
		free(p->config_storage);
	}


	if (r->ptr) free(r->ptr);

	buffer_free(p->parse_response);

	free(p);

	return HANDLER_GO_ON;
}

SETDEFAULTS_FUNC(mod_fastcgi_set_defaults) {
	plugin_data *p = p_d;
	size_t i = 0;

	config_values_t cv[] = {
		{ "cgi.assign",                  NULL, T_CONFIG_ARRAY, T_CONFIG_SCOPE_CONNECTION },       /* 0 */
		{ "cgi.execute-x-only",          NULL, T_CONFIG_BOOLEAN, T_CONFIG_SCOPE_CONNECTION },     /* 1 */
		{ "cgi.x-sendfile",              NULL, T_CONFIG_BOOLEAN, T_CONFIG_SCOPE_CONNECTION },     /* 2 */
		{ "cgi.x-sendfile-docroot",      NULL, T_CONFIG_ARRAY,   T_CONFIG_SCOPE_CONNECTION },     /* 3 */
		{ NULL,                          NULL, T_CONFIG_UNSET, T_CONFIG_SCOPE_UNSET}
	};

	if (!p) return HANDLER_ERROR;

	p->config_storage = calloc(1, srv->config_context->used * sizeof(plugin_config *));
	force_assert(p->config_storage);

	for (i = 0; i < srv->config_context->used; i++) {
		data_config const* config = (data_config const*)srv->config_context->data[i];
		plugin_config *s;

		s = calloc(1, sizeof(plugin_config));
		force_assert(s);

		s->cgi    = array_init();
		s->execute_x_only = 0;
		s->xsendfile_allow= 0;
		s->xsendfile_docroot = array_init();

		cv[0].destination = s->cgi;
		cv[1].destination = &(s->execute_x_only);
		cv[2].destination = &(s->xsendfile_allow);
		cv[3].destination = s->xsendfile_docroot;

		p->config_storage[i] = s;

		if (0 != config_insert_values_global(srv, config->value, cv, i == 0 ? T_CONFIG_SCOPE_SERVER : T_CONFIG_SCOPE_CONNECTION)) {
			return HANDLER_ERROR;
		}

		if (s->xsendfile_docroot->used) {
			size_t j;
			for (j = 0; j < s->xsendfile_docroot->used; ++j) {
				data_string *ds = (data_string *)s->xsendfile_docroot->data[j];
				if (ds->type != TYPE_STRING) {
					log_error_write(srv, __FILE__, __LINE__, "s",
						"unexpected type for key cgi.x-sendfile-docroot; expected: cgi.x-sendfile-docroot = ( \"/allowed/path\", ... )");
					return HANDLER_ERROR;
				}
				if (ds->value->ptr[0] != '/') {
					log_error_write(srv, __FILE__, __LINE__, "SBs",
						"cgi.x-sendfile-docroot paths must begin with '/'; invalid: \"", ds->value, "\"");
					return HANDLER_ERROR;
				}
				buffer_path_simplify(ds->value, ds->value);
				buffer_append_slash(ds->value);
			}
		}
	}

	return HANDLER_GO_ON;
}


static int cgi_pid_add(server *srv, plugin_data *p, pid_t pid) {
	int m = -1;
	size_t i;
	buffer_pid_t *r = &(p->cgi_pid);

	UNUSED(srv);

	for (i = 0; i < r->used; i++) {
		if (r->ptr[i] > m) m = r->ptr[i];
	}

	if (r->size == 0) {
		r->size = 16;
		r->ptr = malloc(sizeof(*r->ptr) * r->size);
		force_assert(r->ptr);
	} else if (r->used == r->size) {
		r->size += 16;
		r->ptr = realloc(r->ptr, sizeof(*r->ptr) * r->size);
		force_assert(r->ptr);
	}

	r->ptr[r->used++] = pid;

	return m;
}

static int cgi_pid_del(server *srv, plugin_data *p, pid_t pid) {
	size_t i;
	buffer_pid_t *r = &(p->cgi_pid);

	UNUSED(srv);

	for (i = 0; i < r->used; i++) {
		if (r->ptr[i] == pid) break;
	}

	if (i != r->used) {
		/* found */

		if (i != r->used - 1) {
			r->ptr[i] = r->ptr[r->used - 1];
		}
		r->used--;
	}

	return 0;
}

static int cgi_response_parse(server *srv, connection *con, plugin_data *p, buffer *in) {
	char *ns;
	const char *s;
	int line = 0;

	UNUSED(srv);

	buffer_copy_buffer(p->parse_response, in);

	for (s = p->parse_response->ptr;
	     NULL != (ns = strchr(s, '\n'));
	     s = ns + 1, line++) {
		const char *key, *value;
		int key_len;
		data_string *ds;

		/* strip the \n */
		ns[0] = '\0';

		if (ns > s && ns[-1] == '\r') ns[-1] = '\0';

		if (line == 0 &&
		    0 == strncmp(s, "HTTP/1.", 7)) {
			/* non-parsed header ... we parse them anyway */

			if ((s[7] == '1' ||
			     s[7] == '0') &&
			    s[8] == ' ') {
				int status;
				/* after the space should be a status code for us */

				status = strtol(s+9, NULL, 10);

				if (status >= 100 &&
				    status < 1000) {
					/* we expected 3 digits and didn't got them */
					con->parsed_response |= HTTP_STATUS;
					con->http_status = status;
				}
			}
		} else {
			/* parse the headers */
			key = s;
			if (NULL == (value = strchr(s, ':'))) {
				/* we expect: "<key>: <value>\r\n" */
				continue;
			}

			key_len = value - key;
			value += 1;

			/* skip LWS */
			while (*value == ' ' || *value == '\t') value++;

			if (NULL == (ds = (data_string *)array_get_unused_element(con->response.headers, TYPE_STRING))) {
				ds = data_response_init();
			}
			buffer_copy_string_len(ds->key, key, key_len);
			buffer_copy_string(ds->value, value);

			array_insert_unique(con->response.headers, (data_unset *)ds);

			switch(key_len) {
			case 4:
				if (0 == strncasecmp(key, "Date", key_len)) {
					con->parsed_response |= HTTP_DATE;
				}
				break;
			case 6:
				if (0 == strncasecmp(key, "Status", key_len)) {
					int status = strtol(value, NULL, 10);
					if (status >= 100 && status < 1000) {
						con->http_status = status;
						con->parsed_response |= HTTP_STATUS;
					} else {
						con->http_status = 502;
					}
				}
				break;
			case 8:
				if (0 == strncasecmp(key, "Location", key_len)) {
					con->parsed_response |= HTTP_LOCATION;
				}
				break;
			case 10:
				if (0 == strncasecmp(key, "Connection", key_len)) {
					con->response.keep_alive = (0 == strcasecmp(value, "Keep-Alive")) ? 1 : 0;
					con->parsed_response |= HTTP_CONNECTION;
				}
				break;
			case 14:
				if (0 == strncasecmp(key, "Content-Length", key_len)) {
					con->response.content_length = strtoul(value, NULL, 10);
					con->parsed_response |= HTTP_CONTENT_LENGTH;
				}
				break;
			default:
				break;
			}
		}
	}

	/* CGI/1.1 rev 03 - 7.2.1.2 */
	if ((con->parsed_response & HTTP_LOCATION) &&
	    !(con->parsed_response & HTTP_STATUS)) {
		con->http_status = 302;
	}

	return 0;
}


static int cgi_demux_response(server *srv, handler_ctx *hctx) {
	plugin_data *p    = hctx->plugin_data;
	connection  *con  = hctx->remote_conn;

	while(1) {
		int n;
		int toread;

#if defined(__WIN32)
		buffer_string_prepare_copy(hctx->response, 4 * 1024);
#else
		if (ioctl(con->fd, FIONREAD, &toread) || toread <= 4*1024) {
			buffer_string_prepare_copy(hctx->response, 4 * 1024);
		} else {
			if (toread > MAX_READ_LIMIT) toread = MAX_READ_LIMIT;
			buffer_string_prepare_copy(hctx->response, toread);
		}
#endif

		if (-1 == (n = read(hctx->fd, hctx->response->ptr, hctx->response->size - 1))) {
			if (errno == EAGAIN || errno == EINTR) {
				/* would block, wait for signal */
				fdevent_event_add(srv->ev, &(hctx->fde_ndx), hctx->fd, FDEVENT_IN);
				return FDEVENT_HANDLED_NOT_FINISHED;
			}
			/* error */
			log_error_write(srv, __FILE__, __LINE__, "sdd", strerror(errno), con->fd, hctx->fd);
			return FDEVENT_HANDLED_ERROR;
		}

		if (n == 0) {
			/* read finished */
			return FDEVENT_HANDLED_FINISHED;
		}

		buffer_commit(hctx->response, n);

		/* split header from body */

		if (con->file_started == 0) {
			int is_header = 0;
			int is_header_end = 0;
			size_t last_eol = 0;
			size_t i, header_len;

			buffer_append_string_buffer(hctx->response_header, hctx->response);

			/**
			 * we have to handle a few cases:
			 *
			 * nph:
			 * 
			 *   HTTP/1.0 200 Ok\n
			 *   Header: Value\n
			 *   \n
			 *
			 * CGI:
			 *   Header: Value\n
			 *   Status: 200\n
			 *   \n
			 *
			 * and different mixes of \n and \r\n combinations
			 * 
			 * Some users also forget about CGI and just send a response and hope 
			 * we handle it. No headers, no header-content seperator
			 * 
			 */
			
			/* nph (non-parsed headers) */
			if (0 == strncmp(hctx->response_header->ptr, "HTTP/1.", 7)) is_header = 1;

			header_len = buffer_string_length(hctx->response_header);
			for (i = 0; !is_header_end && i < header_len; i++) {
				char c = hctx->response_header->ptr[i];

				switch (c) {
				case ':':
					/* we found a colon
					 *
					 * looks like we have a normal header 
					 */
					is_header = 1;
					break;
				case '\n':
					/* EOL */
					if (is_header == 0) {
						/* we got a EOL but we don't seem to got a HTTP header */

						is_header_end = 1;

						break;
					}

					/**
					 * check if we saw a \n(\r)?\n sequence 
					 */
					if (last_eol > 0 && 
					    ((i - last_eol == 1) || 
					     (i - last_eol == 2 && hctx->response_header->ptr[i - 1] == '\r'))) {
						is_header_end = 1;
						break;
					}

					last_eol = i;

					break;
				}
			}

			if (is_header_end) {
				if (!is_header) {
					/* no header, but a body */
					if (0 != http_chunk_append_buffer(srv, con, hctx->response_header)) {
						return FDEVENT_HANDLED_ERROR;
					}
				} else {
					const char *bstart;
					size_t blen;

					/* the body starts after the EOL */
					bstart = hctx->response_header->ptr + i;
					blen = header_len - i;

					/**
					 * i still points to the char after the terminating EOL EOL
					 *
					 * put it on the last \n again
					 */
					i--;

					/* string the last \r?\n */
					if (i > 0 && (hctx->response_header->ptr[i - 1] == '\r')) {
						i--;
					}

					buffer_string_set_length(hctx->response_header, i);

					/* parse the response header */
					cgi_response_parse(srv, con, p, hctx->response_header);

					if (con->http_status >= 300 && con->http_status < 400) {
						/*(con->parsed_response & HTTP_LOCATION)*/
						data_string *ds;
						if (NULL != (ds = (data_string *) array_get_element(con->response.headers, "Location"))
						    && ds->value->ptr[0] == '/') {
							if (++con->loops_per_request > 5) {
								log_error_write(srv, __FILE__, __LINE__, "sb", "too many internal loops while processing request:", con->request.orig_uri);
								con->http_status = 500; /* Internal Server Error */
								con->mode = DIRECT;
								return FDEVENT_HANDLED_FINISHED;
							}

							buffer_copy_buffer(con->request.uri, ds->value);

							if (con->request.content_length) {
								if (con->request.content_length != con->request_content_queue->bytes_in) {
									con->keep_alive = 0;
								}
								con->request.content_length = 0;
								chunkqueue_reset(con->request_content_queue);
							}

							if (con->http_status != 307 && con->http_status != 308) {
								/* Note: request body (if any) sent to initial dynamic handler
								 * and is not available to the internal redirect */
								con->request.http_method = HTTP_METHOD_GET;
							}

							connection_response_reset(srv, con); /*(includes con->http_status = 0)*/

							con->mode = DIRECT;
							return FDEVENT_HANDLED_COMEBACK;
						}
					}

					if (hctx->conf.xsendfile_allow) {
						data_string *ds;
						if (NULL != (ds = (data_string *) array_get_element(con->response.headers, "X-Sendfile"))) {
							http_response_xsendfile(srv, con, ds->value, hctx->conf.xsendfile_docroot);
							return FDEVENT_HANDLED_FINISHED;
						}
					}

					if (blen > 0) {
						if (0 != http_chunk_append_mem(srv, con, bstart, blen)) {
							return FDEVENT_HANDLED_ERROR;
						}
					}
				}

				con->file_started = 1;
			} else {
				/*(reuse MAX_HTTP_REQUEST_HEADER as max size for response headers from backends)*/
				if (header_len > MAX_HTTP_REQUEST_HEADER) {
					log_error_write(srv, __FILE__, __LINE__, "sb", "response headers too large for", con->uri.path);
					con->http_status = 502; /* Bad Gateway */
					con->mode = DIRECT;
					return FDEVENT_HANDLED_FINISHED;
				}
			}
		} else {
			if (0 != http_chunk_append_buffer(srv, con, hctx->response)) {
				return FDEVENT_HANDLED_ERROR;
			}
			if ((con->conf.stream_response_body & FDEVENT_STREAM_RESPONSE_BUFMIN)
			    && chunkqueue_length(con->write_queue) > 65536 - 4096) {
				if (!con->is_writable) {
					/*(defer removal of FDEVENT_IN interest since
					 * connection_state_machine() might be able to send data
					 * immediately, unless !con->is_writable, where
					 * connection_state_machine() might not loop back to call
					 * mod_cgi_handle_subrequest())*/
					fdevent_event_clr(srv->ev, &(hctx->fde_ndx), hctx->fd, FDEVENT_IN);
				}
				break;
			}
		}

#if 0
		log_error_write(srv, __FILE__, __LINE__, "ddss", con->fd, hctx->fd, connection_get_state(con->state), b->ptr);
#endif
	}

	return FDEVENT_HANDLED_NOT_FINISHED;
}

static void cgi_connection_close_fdtocgi(server *srv, handler_ctx *hctx) {
	/*(closes only hctx->fdtocgi)*/
	fdevent_event_del(srv->ev, &(hctx->fde_ndx_tocgi), hctx->fdtocgi);
	fdevent_unregister(srv->ev, hctx->fdtocgi);
	fdevent_sched_close(srv->ev, hctx->fdtocgi, 0);
	hctx->fdtocgi = -1;
}

static void cgi_connection_close(server *srv, handler_ctx *hctx) {
	int status;
	pid_t pid;
	plugin_data *p = hctx->plugin_data;
	connection *con = hctx->remote_conn;

#ifndef __WIN32

	/* the connection to the browser went away, but we still have a connection
	 * to the CGI script
	 *
	 * close cgi-connection
	 */

	if (hctx->fd != -1) {
		/* close connection to the cgi-script */
		fdevent_event_del(srv->ev, &(hctx->fde_ndx), hctx->fd);
		fdevent_unregister(srv->ev, hctx->fd);
		fdevent_sched_close(srv->ev, hctx->fd, 0);
	}

	if (hctx->fdtocgi != -1) {
		cgi_connection_close_fdtocgi(srv, hctx); /*(closes only hctx->fdtocgi)*/
	}

	pid = hctx->pid;

	con->plugin_ctx[p->id] = NULL;

	cgi_handler_ctx_free(hctx);

	/* if waitpid hasn't been called by response.c yet, do it here */
	if (pid) {
		/* check if the CGI-script is already gone */
		switch(waitpid(pid, &status, WNOHANG)) {
		case 0:
			/* not finished yet */
#if 0
			log_error_write(srv, __FILE__, __LINE__, "sd", "(debug) child isn't done yet, pid:", pid);
#endif
			break;
		case -1:
			/* */
			if (errno == EINTR) break;

			/*
			 * errno == ECHILD happens if _subrequest catches the process-status before
			 * we have read the response of the cgi process
			 *
			 * -> catch status
			 * -> WAIT_FOR_EVENT
			 * -> read response
			 * -> we get here with waitpid == ECHILD
			 *
			 */
			if (errno != ECHILD) {
				log_error_write(srv, __FILE__, __LINE__, "ss", "waitpid failed: ", strerror(errno));
			}
			/* anyway: don't wait for it anymore */
			pid = 0;
			break;
		default:
			if (WIFEXITED(status)) {
#if 0
				log_error_write(srv, __FILE__, __LINE__, "sd", "(debug) cgi exited fine, pid:", pid);
#endif
			} else {
				log_error_write(srv, __FILE__, __LINE__, "