1 // This module handles fulltext indexing of the message base.
3 // Copyright (c) 2005-2023 by the citadel.org team
5 // This program is open source software. Use, duplication, or disclosure
6 // is subject to the terms of the GNU General Public License, version 3.
8 #include "../../sysdep.h"
16 #include <sys/types.h>
21 #include <libcitadel.h>
22 #include "../../citadel_defs.h"
23 #include "../../server.h"
24 #include "../../citserver.h"
25 #include "../../support.h"
26 #include "../../config.h"
27 #include "../../room_ops.h"
28 #include "../../database.h"
29 #include "../../msgbase.h"
30 #include "../../control.h"
31 #include "serv_fulltext.h"
32 #include "ft_wordbreaker.h"
33 #include "../../threads.h"
34 #include "../../context.h"
35 #include "../../ctdl_module.h"
38 // These can be global variables because only one indexer runs at a time.
39 Array *messages_to_be_indexed = NULL;
40 long highest_msg_already_indexed = 0;
41 long highest_msg_to_be_indexed = 0;
45 int longcmp(const void *rec1, const void *rec2) {
48 i1 = *(const long *)rec1;
49 i2 = *(const long *)rec2;
51 if (i1 > i2) return(1);
52 if (i1 < i2) return(-1);
57 // Index or de-index a message. (op == 1 to index, 0 to de-index)
58 void ft_index_message(long msgnum, int op) {
60 Array *tokens_in_this_message = NULL;
61 struct cdbdata cdb_bucket;
65 struct CtdlMessage *msg = NULL;
67 if (msgnum == 0) return;
69 msg = CtdlFetchMessage(msgnum, 1);
71 // This is not necessarily an error condition; it could simply mean that the message was
72 // deleted before it could be indexed. This happens often when the load tester is running.
73 syslog(LOG_ERR, "fulltext: ft_index_message() could not load msg %ld", msgnum);
77 if (!CM_IsEmpty(msg, eSuppressIdx)) {
78 syslog(LOG_DEBUG, "fulltext: ft_index_message() excluded msg %ld", msgnum);
83 syslog(LOG_DEBUG, "fulltext: ft_index_message() %s msg %ld", (op ? "adding" : "removing") , msgnum);
85 // Output the message as text before indexing it, so we don't end up indexing a bunch of encoded base64, etc.
86 CC->redirect_buffer = NewStrBufPlain(NULL, SIZ);
87 CtdlOutputPreLoadedMsg(msg, MT_CITADEL, HEADERS_ALL, 0, 1, 0);
89 msgtext = CC->redirect_buffer;
90 CC->redirect_buffer = NULL;
91 if (msgtext != NULL) {
92 syslog(LOG_DEBUG, "fulltext: wordbreaking message %ld (%d bytes)", msgnum, StrLength(msgtext));
94 txt = SmashStrBuf(&msgtext);
95 tokens_in_this_message = wordbreaker(txt);
98 syslog(LOG_DEBUG, "fulltext: %sindexing message %ld [%d tokens]",
101 array_len(tokens_in_this_message)
104 if (array_len(tokens_in_this_message) > 0) {
105 begin_critical_section(S_INDEXER);
106 cdb_begin_transaction();
107 for (i=0; i<array_len(tokens_in_this_message); ++i) {
109 // Identify the bucket which we will be modifying
110 memcpy(&tok, array_get_element_at(tokens_in_this_message, i), sizeof(int));
113 cdb_bucket = cdb_fetch(CDB_FULLTEXT, &tok, sizeof(int));
114 long *newbucket = malloc(cdb_bucket.len + sizeof(long));
115 int nmsgs = cdb_bucket.len / sizeof(long);
117 if (op == 1) { // indexing, add this message to the bucket
118 memcpy(&newbucket[0], cdb_bucket.ptr, cdb_bucket.len);
119 int already_there = 0;
120 for (j=0; j<nmsgs; ++j) {
121 if (newbucket[j] == msgnum) {
125 if (already_there == 0) {
126 memcpy(&newbucket[nmsgs++], &msgnum, sizeof(long));
130 else if (op == 0) { // deindexing, remove this message from the bucket
131 memcpy(newbucket, cdb_bucket.ptr, cdb_bucket.len);
132 for (j=0; j<nmsgs; ++j) {
133 if ((newbucket[j] == msgnum) || (newbucket[j] == 0)) {
134 memcpy(&newbucket[j], &newbucket[j+1], ((nmsgs-j)*sizeof(long)));
141 // Then write it back to disk
142 cdb_store(CDB_FULLTEXT, &tok, sizeof(int), newbucket, (nmsgs*sizeof(long)));
145 if (server_shutting_down) break;
147 cdb_end_transaction();
148 end_critical_section(S_INDEXER);
149 CtdlSetConfigLong("MMfulltext", msgnum);
151 array_free(tokens_in_this_message);
155 // Scan a room for messages to index.
156 void ft_index_room(struct ctdlroom *qrbuf, void *data) {
157 if (server_shutting_down) {
165 // 2023aug30 ajc - old code did another CtdlGetRoom() here. Not only is that redundant,
166 // but for some reason it also made Berkeley DB deadlock after a while. I don't know why.
168 // qrbuf is already populated. fetch the list of messages in this room.
169 num_msgs = CtdlFetchMsgList(qrbuf->QRnumber, &msglist);
171 // Identify messages which have NOT yet been seen by the indexer.
172 if (msglist != NULL) {
173 for (i=0; i<num_msgs; ++i) {
176 && (msglist[i] > highest_msg_already_indexed)
177 && (msglist[i] <= highest_msg_to_be_indexed)
179 array_append(messages_to_be_indexed, &msglist[i]);
187 // Begin the fulltext indexing process.
188 void do_fulltext_indexing(void) {
190 static time_t last_progress = 0L;
191 static int is_running = 0;
192 if (is_running) return; // Concurrency check - only one can run
195 // Don't do this if the site doesn't have it enabled.
196 if (!CtdlGetConfigInt("c_enable_fulltext")) {
199 syslog(LOG_DEBUG, "fulltext: indexing started");
201 // If we've switched wordbreaker modules, burn the index and start over.
202 begin_critical_section(S_CONTROL);
203 if (CtdlGetConfigInt("MM_fulltext_wordbreaker") != FT_WORDBREAKER_ID) {
204 syslog(LOG_DEBUG, "fulltext: wb ver on disk = %d, code ver = %d",
205 CtdlGetConfigInt("MM_fulltext_wordbreaker"), FT_WORDBREAKER_ID
207 syslog(LOG_INFO, "fulltext: (re)initializing index");
208 cdb_trunc(CDB_FULLTEXT);
209 CtdlSetConfigLong("MMfulltext", 0);
211 end_critical_section(S_CONTROL);
213 // Silently return if our fulltext index is up to date with new messages.
214 if ((CtdlGetConfigLong("MMfulltext") >= CtdlGetConfigLong("MMhighest"))) {
215 return; // nothing to do!
218 highest_msg_already_indexed = CtdlGetConfigLong("MMfulltext");
219 highest_msg_to_be_indexed = CtdlGetConfigLong("MMhighest");
220 messages_to_be_indexed = array_new(sizeof(long));
222 // Now go through each room and find messages to index.
223 CtdlForEachRoom(ft_index_room, NULL); // load all msg pointers
224 array_sort(messages_to_be_indexed, longcmp); // sort them
226 // Here it is ... do each message!
228 long prev_msgnum = 0;
229 for (i=0; i<array_len(messages_to_be_indexed); ++i) {
230 memcpy(&msgnum, array_get_element_at(messages_to_be_indexed, i), sizeof(long));
232 if (msgnum != prev_msgnum) { // careful to avoid dupes
233 ft_index_message(msgnum, 1);
236 prev_msgnum = msgnum;
239 array_free(messages_to_be_indexed);
241 CtdlSetConfigInt("MM_fulltext_wordbreaker", FT_WORDBREAKER_ID);
243 syslog(LOG_DEBUG, "fulltext: indexing finished");
249 // API call to perform searches.
250 // (This one does the "all of these words" search.)
251 // Caller is responsible for freeing the message list.
252 Array *ft_search(const char *search_string) {
254 struct cdbdata cdb_bucket;
255 long msgnum, smsgnum;
258 Array *r = array_new(sizeof(long));
259 if (!r) return(NULL);
260 Array *t = wordbreaker(search_string);
262 if ((t != NULL) && (array_len(t) > 0)) {
263 for (i=0; i<array_len(t); ++i) {
264 memcpy(&tok, array_get_element_at(t, i), sizeof(int));
265 cdb_bucket = cdb_fetch(CDB_FULLTEXT, &tok, sizeof(int));
266 if (cdb_bucket.ptr != NULL) {
267 for (j=0; j<(cdb_bucket.len / sizeof(long)); ++j) {
268 memcpy(&msgnum, cdb_bucket.ptr + (j*sizeof(long)), sizeof(long));
269 array_append(r, &msgnum);
275 // We need to return any message containing ALL of the tokens.
276 // If a message number appears in the array `n` times, where `n` is the number of search words,
277 // that means it matched all search words.
278 array_sort(r, longcmp);
279 for (i=0; i<array_len(r); ++i) {
280 memcpy(&msgnum, array_get_element_at(r, i), sizeof(long));
282 for (j=i+1; j<array_len(r); ++j) {
283 memcpy(&smsgnum, array_get_element_at(r, j), sizeof(long));
284 if (msgnum == smsgnum) {
286 array_delete_element_at(r, j);
290 if (count != array_len(t)) {
291 array_delete_element_at(r, i);
301 // This search command is for diagnostic purposes and may be removed or replaced.
302 void cmd_srch(char *argbuf) {
304 char search_string[SIZ];
305 Array *matches = NULL;
308 if (CtdlAccessCheck(ac_logged_in)) return;
310 if (!CtdlGetConfigInt("c_enable_fulltext")) {
311 cprintf("%d Full text index is not enabled on this server.\n", ERROR + CMD_NOT_SUPPORTED);
315 extract_token(search_string, argbuf, 0, '|', sizeof search_string);
316 matches = ft_search(search_string);
318 cprintf("%d %d msgs match all search words:\n", LISTING_FOLLOWS, array_len(matches));
319 if ((matches != NULL) && (array_len(matches) > 0)) {
320 for (i=0; i<array_len(matches); ++i) {
321 memcpy(&msgnum, array_get_element_at(matches, i), sizeof(long));
322 cprintf("%ld\n", msgnum);
325 if (matches != NULL) {
332 void ft_delete_remove(char *room, long msgnum) {
335 // Remove from fulltext index
336 if (CtdlGetConfigInt("c_enable_fulltext")) {
337 ft_index_message(msgnum, 0);
342 // Initialization function, called from modules_init.c
343 char *ctdl_module_init_fulltext(void) {
345 CtdlRegisterProtoHook(cmd_srch, "SRCH", "Full text search");
346 CtdlRegisterDeleteHook(ft_delete_remove);
347 //CtdlRegisterSearchFuncHook(ft_search, "fulltext");
348 CtdlRegisterSessionHook(do_fulltext_indexing, EVT_TIMER, PRIO_CLEANUP + 300);
350 // return our module name for the log