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"
37 // These can be global variables because only one indexer runs at a time.
38 Array *messages_to_be_indexed = NULL;
39 long highest_msg_already_indexed = 0;
40 long highest_msg_to_be_indexed = 0;
44 int longcmp(const void *rec1, const void *rec2) {
47 i1 = *(const long *)rec1;
48 i2 = *(const long *)rec2;
50 if (i1 > i2) return(1);
51 if (i1 < i2) return(-1);
56 // Index or de-index a message. (op == 1 to index, 0 to de-index)
57 void ft_index_message(long msgnum, int op) {
59 Array *tokens_in_this_message = NULL;
60 struct cdbdata cdb_bucket;
64 struct CtdlMessage *msg = NULL;
66 if (msgnum == 0) return;
68 msg = CtdlFetchMessage(msgnum, 1);
70 // This is not necessarily an error condition; it could simply mean that the message was
71 // deleted before it could be indexed. This happens often when the load tester is running.
72 syslog(LOG_ERR, "fulltext: ft_index_message() could not load msg %ld", msgnum);
76 if (!CM_IsEmpty(msg, eSuppressIdx)) {
77 syslog(LOG_DEBUG, "fulltext: ft_index_message() excluded msg %ld", msgnum);
82 syslog(LOG_DEBUG, "fulltext: ft_index_message() %s msg %ld", (op ? "adding" : "removing") , msgnum);
84 // Output the message as text before indexing it, so we don't end up indexing a bunch of encoded base64, etc.
85 CC->redirect_buffer = NewStrBufPlain(NULL, SIZ);
86 CtdlOutputPreLoadedMsg(msg, MT_CITADEL, HEADERS_ALL, 0, 1, 0);
88 msgtext = CC->redirect_buffer;
89 CC->redirect_buffer = NULL;
90 if (msgtext != NULL) {
91 syslog(LOG_DEBUG, "fulltext: wordbreaking message %ld (%d bytes)", msgnum, StrLength(msgtext));
93 txt = SmashStrBuf(&msgtext);
94 tokens_in_this_message = wordbreaker(txt);
97 syslog(LOG_DEBUG, "fulltext: %sindexing message %ld [%d tokens]",
100 array_len(tokens_in_this_message)
103 if (array_len(tokens_in_this_message) > 0) {
104 begin_critical_section(S_INDEXER);
105 cdb_begin_transaction();
106 for (i=0; i<array_len(tokens_in_this_message); ++i) {
108 // Identify the bucket which we will be modifying
109 memcpy(&tok, array_get_element_at(tokens_in_this_message, i), sizeof(int));
112 cdb_bucket = cdb_fetch(CDB_FULLTEXT, &tok, sizeof(int));
113 long *newbucket = malloc(cdb_bucket.len + sizeof(long));
114 int nmsgs = cdb_bucket.len / sizeof(long);
116 if (op == 1) { // indexing, add this message to the bucket
117 memcpy(&newbucket[0], cdb_bucket.ptr, cdb_bucket.len);
118 int already_there = 0;
119 for (j=0; j<nmsgs; ++j) {
120 if (newbucket[j] == msgnum) {
124 if (already_there == 0) {
125 memcpy(&newbucket[nmsgs++], &msgnum, sizeof(long));
129 else if (op == 0) { // deindexing, remove this message from the bucket
130 memcpy(newbucket, cdb_bucket.ptr, cdb_bucket.len);
131 for (j=0; j<nmsgs; ++j) {
132 if ((newbucket[j] == msgnum) || (newbucket[j] == 0)) {
133 memcpy(&newbucket[j], &newbucket[j+1], ((nmsgs-j)*sizeof(long)));
140 // Then write it back to disk
141 cdb_store(CDB_FULLTEXT, &tok, sizeof(int), newbucket, (nmsgs*sizeof(long)));
144 if (server_shutting_down) break;
146 cdb_end_transaction();
147 end_critical_section(S_INDEXER);
148 CtdlSetConfigLong("MMfulltext", msgnum);
150 array_free(tokens_in_this_message);
154 // Scan a room for messages to index.
155 void ft_index_room(struct ctdlroom *qrbuf, void *data) {
156 if (server_shutting_down) {
164 // 2023aug30 ajc - old code did another CtdlGetRoom() here. Not only is that redundant,
165 // but for some reason it also made Berkeley DB deadlock after a while. I don't know why.
167 // qrbuf is already populated. fetch the list of messages in this room.
168 num_msgs = CtdlFetchMsgList(qrbuf->QRnumber, &msglist);
170 // Identify messages which have NOT yet been seen by the indexer.
171 if (msglist != NULL) {
172 for (i=0; i<num_msgs; ++i) {
175 && (msglist[i] > highest_msg_already_indexed)
176 && (msglist[i] <= highest_msg_to_be_indexed)
178 array_append(messages_to_be_indexed, &msglist[i]);
186 // Begin the fulltext indexing process.
187 void do_fulltext_indexing(void) {
189 static time_t last_progress = 0L;
190 static int is_running = 0;
191 if (is_running) return; // Concurrency check - only one can run
194 // Don't do this if the site doesn't have it enabled.
195 if (!CtdlGetConfigInt("c_enable_fulltext")) {
198 // If we've switched wordbreaker modules, burn the index and start over.
199 begin_critical_section(S_CONTROL);
200 if (CtdlGetConfigInt("MM_fulltext_wordbreaker") != FT_WORDBREAKER_ID) {
201 syslog(LOG_DEBUG, "fulltext: wb ver on disk = %d, code ver = %d",
202 CtdlGetConfigInt("MM_fulltext_wordbreaker"), FT_WORDBREAKER_ID
204 syslog(LOG_INFO, "fulltext: (re)initializing index");
205 cdb_trunc(CDB_FULLTEXT);
206 CtdlSetConfigLong("MMfulltext", 0);
208 end_critical_section(S_CONTROL);
210 // Silently return if our fulltext index is up to date with new messages.
211 if ((CtdlGetConfigLong("MMfulltext") >= CtdlGetConfigLong("MMhighest"))) {
212 return; // nothing to do!
215 highest_msg_already_indexed = CtdlGetConfigLong("MMfulltext");
216 highest_msg_to_be_indexed = CtdlGetConfigLong("MMhighest");
217 syslog(LOG_DEBUG, "fulltext: indexing started. msgs %ld--%ld", highest_msg_already_indexed, highest_msg_to_be_indexed);
219 messages_to_be_indexed = array_new(sizeof(long));
221 // Now go through each room and find messages to index.
222 CtdlForEachRoom(ft_index_room, NULL); // load all msg pointers
223 array_sort(messages_to_be_indexed, longcmp); // sort them
225 // Here it is ... do each message!
227 long prev_msgnum = 0;
228 time_t started_indexing_at = time(NULL);
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;
237 if (time(NULL) - started_indexing_at >= MAXIMUM_INDEXER_RUN_TIME) {
238 syslog(LOG_DEBUG, "fulltext: indexer has run for %ld seconds; yielding the thread", time(NULL) - started_indexing_at);
239 i = array_len(messages_to_be_indexed); // go out of scope to make it stop
243 array_free(messages_to_be_indexed);
245 CtdlSetConfigInt("MM_fulltext_wordbreaker", FT_WORDBREAKER_ID);
247 syslog(LOG_DEBUG, "fulltext: indexing finished");
253 // API call to perform searches.
254 // (This one does the "all of these words" search.)
255 // Caller is responsible for freeing the message list.
256 Array *CtdlFullTextSearch(const char *search_string) {
258 struct cdbdata cdb_bucket;
259 long msgnum, smsgnum;
262 Array *r = array_new(sizeof(long));
263 if (!r) return(NULL);
264 Array *t = wordbreaker(search_string);
266 if ((t != NULL) && (array_len(t) > 0)) {
267 for (i=0; i<array_len(t); ++i) {
268 memcpy(&tok, array_get_element_at(t, i), sizeof(int));
269 cdb_bucket = cdb_fetch(CDB_FULLTEXT, &tok, sizeof(int));
270 if (cdb_bucket.ptr != NULL) {
271 for (j=0; j<(cdb_bucket.len / sizeof(long)); ++j) {
272 memcpy(&msgnum, cdb_bucket.ptr + (j*sizeof(long)), sizeof(long));
273 array_append(r, &msgnum);
279 // We need to return any message containing ALL of the tokens.
280 // If a message number appears in the array `n` times, where `n` is the number of search words,
281 // that means it matched all search words.
282 array_sort(r, longcmp);
283 for (i=0; i<array_len(r); ++i) {
284 memcpy(&msgnum, array_get_element_at(r, i), sizeof(long));
286 for (j=i+1; j<array_len(r); ++j) {
287 memcpy(&smsgnum, array_get_element_at(r, j), sizeof(long));
288 if (msgnum == smsgnum) {
290 array_delete_element_at(r, j);
294 if (count != array_len(t)) {
295 array_delete_element_at(r, i);
305 // This search command is for diagnostic purposes and may be removed or replaced.
306 void cmd_srch(char *argbuf) {
308 char search_string[SIZ];
309 Array *matches = NULL;
312 if (CtdlAccessCheck(ac_logged_in)) return;
314 if (!CtdlGetConfigInt("c_enable_fulltext")) {
315 cprintf("%d Full text index is not enabled on this server.\n", ERROR + CMD_NOT_SUPPORTED);
319 extract_token(search_string, argbuf, 0, '|', sizeof search_string);
320 matches = CtdlFullTextSearch(search_string);
322 cprintf("%d %d msgs match all search words:\n", LISTING_FOLLOWS, array_len(matches));
323 if ((matches != NULL) && (array_len(matches) > 0)) {
324 for (i=0; i<array_len(matches); ++i) {
325 memcpy(&msgnum, array_get_element_at(matches, i), sizeof(long));
326 cprintf("%ld\n", msgnum);
329 if (matches != NULL) {
336 void ft_delete_remove(char *room, long msgnum) {
339 // Remove from fulltext index
340 if (CtdlGetConfigInt("c_enable_fulltext")) {
341 ft_index_message(msgnum, 0);
346 // Initialization function, called from modules_init.c
347 char *ctdl_module_init_fulltext(void) {
349 CtdlRegisterProtoHook(cmd_srch, "SRCH", "Full text search");
350 CtdlRegisterDeleteHook(ft_delete_remove);
351 CtdlRegisterSessionHook(do_fulltext_indexing, EVT_HOUSE, PRIO_CLEANUP + 300);
353 // return our module name for the log