* *** HUGE CHANGES *** *** WARNING: NOT FULLY FUNCTIONAL ***
[citadel.git] / webcit / context_loop.c
index f26d1e6f443233caaf38e38ca8539c5d82ed2b5b..e9ea3772ac210c8bc348c62a564ce6945b1bc2b1 100644 (file)
  * context_loop.c
  *
  * This is the other half of the webserver.  It handles the task of hooking
- * up HTTP requests with the session they belong to, using HTTP cookies to
+ * up HTTP requests with the sessions they belong to, using HTTP cookies to
  * keep track of things.  If the HTTP request doesn't belong to any currently
- * active session, a new session is spawned.
+ * active session, a new session is started.
+ *
+ * $Id$
  */
 
+#include <ctype.h>
 #include <stdlib.h>
+#ifdef HAVE_UNISTD_H
 #include <unistd.h>
+#endif
 #include <stdio.h>
+#ifdef HAVE_FCNTL_H
 #include <fcntl.h>
+#endif
 #include <signal.h>
 #include <sys/types.h>
 #include <sys/wait.h>
 #include <sys/socket.h>
+#ifdef HAVE_SYS_TIME_H
 #include <sys/time.h>
+#endif
+#ifdef HAVE_LIMITS_H
 #include <limits.h>
+#endif
 #include <netinet/in.h>
 #include <netdb.h>
 #include <string.h>
 #include <errno.h>
 #include <stdarg.h>
 #include <pthread.h>
+#include <signal.h>
 #include "webcit.h"
+#include "webserver.h"
 
-/*
- * We keep one of these around for each active session
- */
-struct wc_session {
-       struct wc_session *next;        /* Next session in list */
-       int session_id;                 /* Session ID */
-       int inpipe[2];                  /* Data from webserver to session */
-       int outpipe[2];                 /* Data from session to webserver */
-       pthread_mutex_t critter;        /* Critical section uses pipes */
-       };
+/* Only one thread may manipulate SessionList at a time... */
+pthread_mutex_t SessionListMutex;
 
-struct wc_session *SessionList = NULL;
+struct wcsession *SessionList = NULL;
 
-/* Only one thread may manipulate SessionList at a time... */
-pthread_mutex_t MasterCritter;
+pthread_key_t MyConKey;                         /* TSD key for MySession() */
+
+
+void free_attachments(struct wcsession *sess) {
+       struct wc_attachment *att;
 
-int GenerateSessionID() {
-       return getpid();
+       while (sess->first_attachment != NULL) {
+               att = sess->first_attachment;
+               sess->first_attachment = sess->first_attachment->next;
+               free(att->data);
+               free(att);
        }
+}
 
 
-void gets0(int fd, char buf[]) {
+void do_housekeeping(void)
+{
+       struct wcsession *sptr, *ss, *session_to_kill;
+       int num_sessions = 0;
+       static int num_threads = MIN_WORKER_THREADS;
 
-       buf[0] = 0;
        do {
-               buf[strlen(buf)+1] = 0;
-               read(fd, &buf[strlen(buf)], 1);
-               } while (buf[strlen(buf)-1] >= 32);
-       buf[strlen(buf)-1] = 0;
+               session_to_kill = NULL;
+               pthread_mutex_lock(&SessionListMutex);
+               num_sessions = 0;
+               for (sptr = SessionList; sptr != NULL; sptr = sptr->next) {
+                       ++num_sessions;
+
+                       /* Kill idle sessions */
+                       if ((time(NULL) - (sptr->lastreq)) >
+                          (time_t) WEBCIT_TIMEOUT) {
+                               sptr->killthis = 1;
+                       }
+
+                       /* Remove sessions flagged for kill */
+                       if (sptr->killthis) {
+
+                               lprintf(3, "Destroying session %d\n",
+                                       sptr->wc_session);
+
+                               /* remove session from linked list */
+                               if (sptr == SessionList) {
+                                       SessionList = SessionList->next;
+                               }
+                               else for (ss=SessionList;ss!=NULL;ss=ss->next) {
+                                       if (ss->next == sptr) {
+                                               ss->next = ss->next->next;
+                                       }
+                               }
+
+                               session_to_kill = sptr;
+                               goto BREAKOUT;
+                       }
+               }
+BREAKOUT:      pthread_mutex_unlock(&SessionListMutex);
+
+               if (session_to_kill != NULL) {
+                       pthread_mutex_lock(&session_to_kill->SessionMutex);
+                       close(session_to_kill->serv_sock);
+                       close(session_to_kill->chat_sock);
+                       if (session_to_kill->preferences != NULL) {
+                               free(session_to_kill->preferences);
+                       }
+                       free_attachments(session_to_kill);
+                       pthread_mutex_unlock(&session_to_kill->SessionMutex);
+                       free(session_to_kill);
+               }
+
+       } while (session_to_kill != NULL);
+
+       /*
+        * See if we need more worker threads
+        */
+       while ( (num_sessions > num_threads)
+             && (num_threads <= MAX_WORKER_THREADS) ) {
+               spawn_another_worker_thread();
+               ++num_threads;
+               lprintf(3, "There are %d sessions and %d threads active.\n",
+                       num_sessions, num_threads);
+       }
+}
+
+
+/* 
+ * Wake up occasionally and clean house
+ */
+void housekeeping_loop(void)
+{
+       while (1) {
+               sleeeeeeeeeep(HOUSEKEEPING);
+               do_housekeeping();
        }
+}
+
+
+/*
+ * Generate a unique WebCit session ID (which is not the same thing as the
+ * Citadel session ID).
+ *
+ * FIXME ... ensure that session number is truly unique
+ *
+ */
+int GenerateSessionID(void)
+{
+       static int seq = (-1);
+
+       if (seq < 0) {
+               seq = (int) time(NULL);
+       }
+               
+       return ++seq;
+}
+
 
 /*
  * Collapse multiple cookies on one line
  */
-void req_gets(int sock, char *buf, char *hold) {
+int req_gets(int sock, char *buf, char *hold)
+{
        int a;
 
-       if (strlen(hold)==0) {
-               client_gets(sock, buf);
-               }
-       else {
-               strcpy(buf, hold);
-               }
+       if (strlen(hold) == 0) {
+               strcpy(buf, "");
+               a = client_gets(sock, buf);
+               if (a<1) return(-1);
+       } else {
+               safestrncpy(buf, hold, SIZ);
+       }
        strcpy(hold, "");
 
        if (!strncasecmp(buf, "Cookie: ", 8)) {
-               for (a=0; a<strlen(buf); ++a) if (buf[a]==';') {
-                       sprintf(hold, "Cookie: %s", &buf[a+1]);
-                       buf[a]=0;
-                       while (isspace(hold[8])) strcpy(&hold[8], &hold[9]);
-                       return;
+               for (a = 0; a < strlen(buf); ++a)
+                       if (buf[a] == ';') {
+                               sprintf(hold, "Cookie: %s", &buf[a + 1]);
+                               buf[a] = 0;
+                               while (isspace(hold[8]))
+                                       strcpy(&hold[8], &hold[9]);
+                               return(0);
                        }
-               }
        }
 
+       return(0);
+}
 
 /*
- * This loop gets called once for every HTTP connection made to WebCit.
+ * lingering_close() a`la Apache. see
+ * http://www.apache.org/docs/misc/fin_wait_2.html for rationale
  */
-void *context_loop(int sock) {
-       char req[256][256];
-       char buf[256], hold[256];
-       int num_lines = 0;
-       int a;
-       int f;
-       int desired_session = 0;
-       char str_session[256];
-       struct wc_session *sptr;
-       struct wc_session *TheSession;
-       int ContentLength;
-       int CloseSession = 0;
 
-       printf("Reading request from socket %d\n", sock);
+int lingering_close(int fd)
+{
+       char buf[SIZ];
+       int i;
+       fd_set set;
+       struct timeval tv, start;
 
-       /*
-        * Find out what it is that the web browser is asking for
-        */
-       ContentLength = 0;
+       gettimeofday(&start, NULL);
+       shutdown(fd, 1);
        do {
-               req_gets(sock, buf, hold);
-               if (!strncasecmp(buf, "Cookie: wc_session=", 19)) {
-                       desired_session = atoi(&buf[19]);
+               do {
+                       gettimeofday(&tv, NULL);
+                       tv.tv_sec = SLEEPING - (tv.tv_sec - start.tv_sec);
+                       tv.tv_usec = start.tv_usec - tv.tv_usec;
+                       if (tv.tv_usec < 0) {
+                               tv.tv_sec--;
+                               tv.tv_usec += 1000000;
                        }
-               if (!strncasecmp(buf, "Content-length: ", 16)) {
-                       ContentLength = atoi(&buf[16]);
-                       }
-               strcpy(&req[num_lines++][0], buf);
-               } while(strlen(buf)>0);
+                       FD_ZERO(&set);
+                       FD_SET(fd, &set);
+                       i = select(fd + 1, &set, NULL, NULL, &tv);
+               } while (i == -1 && errno == EINTR);
+
+               if (i <= 0)
+                       break;
+
+               i = read(fd, buf, sizeof buf);
+       } while (i != 0 && (i != -1 || errno == EINTR));
+
+       return close(fd);
+}
+
+
+
+/*
+ * Check for bogus requests coming from (for example) brain-dead
+ * Windoze boxes that are infected with the latest worm-of-the-week.
+ * If we detect one of these, bail out without bothering our Citadel
+ * server.
+ */
+int is_bogus(char *http_cmd) {
+
+       if (!strncasecmp(http_cmd, "GET /scripts/root.exe", 21)) return(1);
+       if (!strncasecmp(http_cmd, "GET /c/winnt", 12)) return(2);
+       if (!strncasecmp(http_cmd, "GET /MSADC/", 11)) return(3);
+
+       return(0);      /* probably ok */
+}
+
+
+
+/*
+ * This loop gets called once for every HTTP connection made to WebCit.  At
+ * this entry point we have an HTTP socket with a browser allegedly on the
+ * other end, but we have not yet bound to a WebCit session.
+ *
+ * The job of this function is to locate the correct session and bind to it,
+ * or create a session if necessary and bind to it, then run the WebCit
+ * transaction loop.  Afterwards, we unbind from the session.  When this
+ * function returns, the worker thread is then free to handle another
+ * transaction.
+ */
+void context_loop(int sock)
+{
+       struct httprequest *req = NULL;
+       struct httprequest *last = NULL;
+       struct httprequest *hptr;
+       char buf[SIZ], hold[SIZ];
+       int desired_session = 0;
+       int got_cookie = 0;
+       struct wcsession *TheSession, *sptr;
 
        /*
-        * See if there's an existing session open with the desired ID
+        * Find out what it is that the web browser is asking for
         */
-       TheSession = NULL;
-       if (desired_session != 0) {
-               pthread_mutex_lock(&MasterCritter);
-               for (sptr=SessionList; sptr!=NULL; sptr=sptr->next) {
-                       if (sptr->session_id == desired_session) {
-                               TheSession = sptr;
-                               }
-                       }
-               pthread_mutex_unlock(&MasterCritter);
+       memset(hold, 0, sizeof(hold));
+       do {
+               if (req_gets(sock, buf, hold) < 0) return;
+
+               if (!strncasecmp(buf, "Cookie: webcit=", 15)) {
+                       cookie_to_stuff(&buf[15], &desired_session,
+                               NULL, NULL, NULL);
+                       got_cookie = 1;
                }
 
+               hptr = (struct httprequest *)
+                       malloc(sizeof(struct httprequest));
+               if (req == NULL)
+                       req = hptr;
+               else
+                       last->next = hptr;
+               hptr->next = NULL;
+               last = hptr;
+
+               safestrncpy(hptr->line, buf, sizeof hptr->line);
+
+       } while (strlen(buf) > 0);
+
+       safestrncpy(buf, req->line, sizeof buf);
+       lprintf(5, "HTTP: %s\n", buf);
+
+       /* Check for bogus requests */
+       if (is_bogus(buf)) goto bail;
+
        /*
-        * Create a new session if we have to
+        * If requesting a non-root page, there should already be a cookie
+        * set.  If there isn't, the client browser has cookies turned off
+        * (or doesn't support them) and we have to barf & bail.
         */
-       if (TheSession == NULL) {
-               printf("Creating a new session\n");
-               pthread_mutex_lock(&MasterCritter);
-               TheSession = (struct wc_session *)
-                       malloc(sizeof(struct wc_session));
-               TheSession->session_id = GenerateSessionID();
-               pipe(TheSession->inpipe);
-               pipe(TheSession->outpipe);
-               pthread_mutex_init(&TheSession->critter, NULL);
-               TheSession->next = SessionList;
-               SessionList = TheSession;
-               sprintf(str_session, "%d", TheSession->session_id);
-               f = fork();
-               fflush(stdout); fflush(stdin);
-               if (f==0) {
-                       dup2(TheSession->inpipe[0], 0);
-                       dup2(TheSession->outpipe[1], 1);
-                       execlp("./webcit", "webcit", str_session, NULL);
-                       printf("HTTP/1.0 404 WebCit Failure\n\n");
-                       printf("Server: %s\n", SERVER);
-                       printf("Content-type: text/html\n");
-                       printf("Content-length: 76\n");
-                       printf("\n");
-                       printf("<HTML><HEAD><TITLE>Error</TITLE></HEAD>\n");
-                       printf("<BODY>execlp() failed</BODY></HTML>\n");
-                       exit(0);
-                       }
-               pthread_mutex_unlock(&MasterCritter);
-               }
+       if (!strncasecmp(buf, "GET ", 4)) strcpy(buf, &buf[4]);
+       else if (!strncasecmp(buf, "HEAD ", 5)) strcpy(buf, &buf[5]);
+       else if (!strncasecmp(buf, "POST ", 5)) strcpy(buf, &buf[5]);
+       if (buf[1]==' ') buf[1]=0;
 
        /*
-        * Grab a lock on the session, so other threads don't try to access
-        * the pipes at the same time.
+        * While we're at it, gracefully handle requests for the
+        * robots.txt and favicon.ico files.
         */
-       printf("Locking session %d...\n", TheSession->session_id);
-       pthread_mutex_lock(&TheSession->critter);
-       printf("   ...got lock\n");
+       if (!strncasecmp(buf, "/robots.txt", 11)) {
+               strcpy(req->line, "GET /static/robots.txt"
+                               "?force_close_session=yes HTTP/1.0");
+       }
+       else if (!strncasecmp(buf, "/favicon.ico", 12)) {
+               strcpy(req->line, "GET /static/favicon.ico");
+       }
 
-       /* 
-        * Send the request to the appropriate session...
+       /* These are the URL's which may be executed without a
+        * session cookie already set.  If it's not one of these,
+        * force the session to close because cookies are
+        * probably disabled on the client browser.
         */
-       printf("   Writing %d lines of command\n", num_lines);
-       printf("%s\n", &req[0][0]);
-       for (a=0; a<num_lines; ++a) {
-               write(TheSession->inpipe[1], &req[a][0], strlen(&req[a][0]));
-               write(TheSession->inpipe[1], "\n", 1);
-               }
-       printf("   Writing %d bytes of content\n", ContentLength);
-       while (ContentLength--) {
-               read(sock, buf, 1);
-               write(TheSession->inpipe[1], buf, 1);
-               }
+       else if ( (strcmp(buf, "/"))
+               && (strncasecmp(buf, "/listsub", 8))
+               && (strncasecmp(buf, "/freebusy", 9))
+               && (strncasecmp(buf, "/do_logout", 10))
+               && (got_cookie == 0)) {
+               strcpy(req->line, "GET /static/nocookies.html"
+                               "?force_close_session=yes HTTP/1.0");
+       }
 
        /*
-        * ...and get the response.
+        * See if there's an existing session open with the desired ID
         */
-       printf("   Reading response\n");
-       ContentLength = 0;
-       do {
-               gets0(TheSession->outpipe[0], buf);
-               write(sock, buf, strlen(buf));
-               write(sock, "\n", 1);
-               if (!strncasecmp(buf, "Content-length: ", 16))
-                       ContentLength = atoi(&buf[16]);
-               if (!strcasecmp(buf, "X-WebCit-Session: close")) {
-                       CloseSession = 1;
+       TheSession = NULL;
+       if (desired_session != 0) {
+               pthread_mutex_lock(&SessionListMutex);
+               for (sptr = SessionList; sptr != NULL; sptr = sptr->next) {
+                       if (sptr->wc_session == desired_session) {
+                               TheSession = sptr;
                        }
-               } while (strlen(buf) > 0);
-
-       printf("   Reading %d bytes of content\n");
-       while(ContentLength--) {
-               read(TheSession->outpipe[0], buf, 1);
-               write(sock, buf, 1);
                }
+               pthread_mutex_unlock(&SessionListMutex);
+       }
 
        /*
-        * Now our HTTP connection is done.  It would be relatively easy
-        * to support HTTP/1.1 "persistent" connections by looping back to
-        * the top of this function.  For now, we'll just close.
+        * Create a new session if we have to
         */
-       printf("   Closing socket\n");
-       close(sock);
+       if (TheSession == NULL) {
+               lprintf(3, "Creating a new session\n");
+               TheSession = (struct wcsession *)
+                       malloc(sizeof(struct wcsession));
+               memset(TheSession, 0, sizeof(struct wcsession));
+               TheSession->serv_sock = (-1);
+               TheSession->chat_sock = (-1);
+               TheSession->wc_session = GenerateSessionID();
+               pthread_mutex_init(&TheSession->SessionMutex, NULL);
+
+               pthread_mutex_lock(&SessionListMutex);
+               TheSession->next = SessionList;
+               SessionList = TheSession;
+               pthread_mutex_unlock(&SessionListMutex);
+       }
 
        /*
-        * Let go of the lock
+        * A future improvement might be to check the session integrity
+        * at this point before continuing.
         */
-       printf("Unlocking.\n");
-       pthread_mutex_unlock(&TheSession->critter);
-
-
 
        /*
-        * If the last response included a "close session" directive,
-        * remove the context now.
+        * Bind to the session and perform the transaction
         */
-       if (CloseSession) {
-               printf("Removing session.\n");
-               pthread_mutex_lock(&MasterCritter);
-
-               if (SessionList==TheSession) {
-                       SessionList = SessionList->next;
-                       }
-               else {
-                       for (sptr=SessionList; sptr!=NULL; sptr=sptr->next) {
-                               if (sptr->next == TheSession) {
-                                       sptr->next = TheSession->next;
-                                       }
-                               }
-                       }
-       
-               free(TheSession);
-       
-               pthread_mutex_unlock(&MasterCritter);
-               }
-
+       pthread_mutex_lock(&TheSession->SessionMutex);          /* bind */
+       pthread_setspecific(MyConKey, (void *)TheSession);
+       TheSession->http_sock = sock;
+       TheSession->lastreq = time(NULL);                       /* log */
+       session_loop(req);                              /* do transaction */
+       pthread_mutex_unlock(&TheSession->SessionMutex);        /* unbind */
 
+       /* Free the request buffer */
+bail:  while (req != NULL) {
+               hptr = req->next;
+               free(req);
+               req = hptr;
+       }
 
-       /*
-        * The thread handling this HTTP connection is now finished.
+       /* Free up any session-local substitution variables which
+        * were set during this transaction
         */
-       pthread_exit(NULL);
-       }
+       clear_local_substs();
+}