Avoid declarations after statements.
[pazpar2-moved-to-github.git] / src / pazpar2.c
1 /* $Id: pazpar2.c,v 1.28 2007-01-12 23:37:15 adam Exp $ */
2
3 #include <stdlib.h>
4 #include <stdio.h>
5 #include <string.h>
6 #include <sys/time.h>
7 #include <unistd.h>
8 #include <sys/socket.h>
9 #include <netdb.h>
10 #include <signal.h>
11 #include <ctype.h>
12 #include <assert.h>
13
14 #include <yaz/marcdisp.h>
15 #include <yaz/comstack.h>
16 #include <yaz/tcpip.h>
17 #include <yaz/proto.h>
18 #include <yaz/readconf.h>
19 #include <yaz/pquery.h>
20 #include <yaz/yaz-util.h>
21 #include <yaz/nmem.h>
22
23 #if HAVE_CONFIG_H
24 #include "cconfig.h"
25 #endif
26
27 #define USE_TIMING 0
28 #if USE_TIMING
29 #include <yaz/timing.h>
30 #endif
31
32 #include <netinet/in.h>
33
34 #include "pazpar2.h"
35 #include "eventl.h"
36 #include "http.h"
37 #include "termlists.h"
38 #include "reclists.h"
39 #include "relevance.h"
40 #include "config.h"
41
42 #define MAX_CHUNK 15
43
44 static void client_fatal(struct client *cl);
45 static void connection_destroy(struct connection *co);
46 static int client_prep_connection(struct client *cl);
47 static void ingest_records(struct client *cl, Z_Records *r);
48 static struct conf_retrievalprofile *database_retrieval_profile(struct database *db);
49 void session_alert_watch(struct session *s, int what);
50
51 IOCHAN channel_list = 0;  // Master list of connections we're handling events to
52
53 static struct connection *connection_freelist = 0;
54 static struct client *client_freelist = 0;
55
56 static struct host *hosts = 0;  // The hosts we know about 
57 static struct database *databases = 0; // The databases we know about
58
59 static char *client_states[] = {
60     "Client_Connecting",
61     "Client_Connected",
62     "Client_Idle",
63     "Client_Initializing",
64     "Client_Searching",
65     "Client_Presenting",
66     "Client_Error",
67     "Client_Failed",
68     "Client_Disconnected",
69     "Client_Stopped"
70 };
71
72 // Note: Some things in this structure will eventually move to configuration
73 struct parameters global_parameters = 
74 {
75     "",
76     "",
77     0,
78     0,
79     30,
80     "81",
81     "Index Data PazPar2 (MasterKey)",
82     VERSION,
83     600, // 10 minutes
84     60,
85     100,
86     MAX_CHUNK,
87     0,
88     0,
89     0,
90     0
91 };
92
93 static int send_apdu(struct client *c, Z_APDU *a)
94 {
95     struct connection *co = c->connection;
96     char *buf;
97     int len, r;
98
99     if (!z_APDU(global_parameters.odr_out, &a, 0, 0))
100     {
101         odr_perror(global_parameters.odr_out, "Encoding APDU");
102         abort();
103     }
104     buf = odr_getbuf(global_parameters.odr_out, &len, 0);
105     r = cs_put(co->link, buf, len);
106     if (r < 0)
107     {
108         yaz_log(YLOG_WARN, "cs_put: %s", cs_errmsg(cs_errno(co->link)));
109         return -1;
110     }
111     else if (r == 1)
112     {
113         fprintf(stderr, "cs_put incomplete (ParaZ does not handle that)\n");
114         exit(1);
115     }
116     odr_reset(global_parameters.odr_out); /* release the APDU structure  */
117     co->state = Conn_Waiting;
118     return 0;
119 }
120
121
122 static void send_init(IOCHAN i)
123 {
124     struct connection *co = iochan_getdata(i);
125     struct client *cl = co->client;
126     Z_APDU *a = zget_APDU(global_parameters.odr_out, Z_APDU_initRequest);
127
128     a->u.initRequest->implementationId = global_parameters.implementationId;
129     a->u.initRequest->implementationName = global_parameters.implementationName;
130     a->u.initRequest->implementationVersion =
131         global_parameters.implementationVersion;
132     ODR_MASK_SET(a->u.initRequest->options, Z_Options_search);
133     ODR_MASK_SET(a->u.initRequest->options, Z_Options_present);
134     ODR_MASK_SET(a->u.initRequest->options, Z_Options_namedResultSets);
135
136     ODR_MASK_SET(a->u.initRequest->protocolVersion, Z_ProtocolVersion_1);
137     ODR_MASK_SET(a->u.initRequest->protocolVersion, Z_ProtocolVersion_2);
138     ODR_MASK_SET(a->u.initRequest->protocolVersion, Z_ProtocolVersion_3);
139     if (send_apdu(cl, a) >= 0)
140     {
141         iochan_setflags(i, EVENT_INPUT);
142         cl->state = Client_Initializing;
143     }
144     else
145         cl->state = Client_Error;
146     odr_reset(global_parameters.odr_out);
147 }
148
149 static void send_search(IOCHAN i)
150 {
151     struct connection *co = iochan_getdata(i);
152     struct client *cl = co->client; 
153     struct session *se = cl->session;
154     struct database *db = cl->database;
155     Z_APDU *a = zget_APDU(global_parameters.odr_out, Z_APDU_searchRequest);
156     int ndb, cerror, cpos;
157     char **databaselist;
158     Z_Query *zquery;
159     struct ccl_rpn_node *cn;
160     int ssub = 0, lslb = 100000, mspn = 10;
161
162     yaz_log(YLOG_DEBUG, "Sending search");
163
164     cn = ccl_find_str(global_parameters.ccl_filter, se->query, &cerror, &cpos);
165     if (!cn)
166         return;
167     a->u.searchRequest->query = zquery = odr_malloc(global_parameters.odr_out,
168             sizeof(Z_Query));
169     zquery->which = Z_Query_type_1;
170     zquery->u.type_1 = ccl_rpn_query(global_parameters.odr_out, cn);
171     ccl_rpn_delete(cn);
172
173     for (ndb = 0; db->databases[ndb]; ndb++)
174         ;
175     databaselist = odr_malloc(global_parameters.odr_out, sizeof(char*) * ndb);
176     for (ndb = 0; db->databases[ndb]; ndb++)
177         databaselist[ndb] = db->databases[ndb];
178
179     a->u.presentRequest->preferredRecordSyntax =
180             yaz_oidval_to_z3950oid(global_parameters.odr_out,
181             CLASS_RECSYN, VAL_USMARC);
182     a->u.searchRequest->smallSetUpperBound = &ssub;
183     a->u.searchRequest->largeSetLowerBound = &lslb;
184     a->u.searchRequest->mediumSetPresentNumber = &mspn;
185     a->u.searchRequest->resultSetName = "Default";
186     a->u.searchRequest->databaseNames = databaselist;
187     a->u.searchRequest->num_databaseNames = ndb;
188
189     if (send_apdu(cl, a) >= 0)
190     {
191         iochan_setflags(i, EVENT_INPUT);
192         cl->state = Client_Searching;
193         cl->requestid = se->requestid;
194     }
195     else
196         cl->state = Client_Error;
197
198     odr_reset(global_parameters.odr_out);
199 }
200
201 static void send_present(IOCHAN i)
202 {
203     struct connection *co = iochan_getdata(i);
204     struct client *cl = co->client; 
205     Z_APDU *a = zget_APDU(global_parameters.odr_out, Z_APDU_presentRequest);
206     int toget;
207     int start = cl->records + 1;
208
209     toget = global_parameters.chunk;
210     if (toget > global_parameters.toget - cl->records)
211         toget = global_parameters.toget - cl->records;
212     if (toget > cl->hits - cl->records)
213         toget = cl->hits - cl->records;
214
215     yaz_log(YLOG_DEBUG, "Trying to present %d records\n", toget);
216
217     a->u.presentRequest->resultSetStartPoint = &start;
218     a->u.presentRequest->numberOfRecordsRequested = &toget;
219
220     a->u.presentRequest->resultSetId = "Default";
221
222     a->u.presentRequest->preferredRecordSyntax =
223             yaz_oidval_to_z3950oid(global_parameters.odr_out,
224             CLASS_RECSYN, VAL_USMARC);
225
226     if (send_apdu(cl, a) >= 0)
227     {
228         iochan_setflags(i, EVENT_INPUT);
229         cl->state = Client_Presenting;
230     }
231     else
232         cl->state = Client_Error;
233     odr_reset(global_parameters.odr_out);
234 }
235
236 static void do_initResponse(IOCHAN i, Z_APDU *a)
237 {
238     struct connection *co = iochan_getdata(i);
239     struct client *cl = co->client;
240     Z_InitResponse *r = a->u.initResponse;
241
242     yaz_log(YLOG_DEBUG, "Received init response");
243
244     if (*r->result)
245     {
246         cl->state = Client_Idle;
247     }
248     else
249         cl->state = Client_Failed; // FIXME need to do something to the connection
250 }
251
252 static void do_searchResponse(IOCHAN i, Z_APDU *a)
253 {
254     struct connection *co = iochan_getdata(i);
255     struct client *cl = co->client;
256     struct session *se = cl->session;
257     Z_SearchResponse *r = a->u.searchResponse;
258
259     yaz_log(YLOG_DEBUG, "Searchresponse (status=%d)", *r->searchStatus);
260
261     if (*r->searchStatus)
262     {
263         cl->hits = *r->resultCount;
264         se->total_hits += cl->hits;
265         if (r->presentStatus && !*r->presentStatus && r->records)
266         {
267             yaz_log(YLOG_DEBUG, "Records in search response");
268             cl->records += *r->numberOfRecordsReturned;
269             ingest_records(cl, r->records);
270         }
271         cl->state = Client_Idle;
272     }
273     else
274     {          /*"FAILED"*/
275         cl->hits = 0;
276         cl->state = Client_Error;
277         if (r->records) {
278             Z_Records *recs = r->records;
279             if (recs->which == Z_Records_NSD)
280             {
281                 yaz_log(YLOG_WARN, "Non-surrogate diagnostic");
282                 cl->diagnostic = *recs->u.nonSurrogateDiagnostic->condition;
283                 cl->state = Client_Error;
284             }
285         }
286     }
287 }
288
289 char *normalize_mergekey(char *buf)
290 {
291     char *p = buf, *pout = buf;
292
293     while (*p)
294     {
295         while (*p && !isalnum(*p))
296             p++;
297         while (isalnum(*p))
298             *(pout++) = tolower(*(p++));
299         if (*p)
300             *(pout++) = ' ';
301         while (*p && !isalnum(*p))
302             p++;
303     }
304     if (buf != pout)
305         *pout = '\0';
306
307     return buf;
308 }
309
310
311 #ifdef GAGA
312 // FIXME needs to be generalized. Should flexibly generate X lists per search
313 static void extract_subject(struct session *s, const char *rec)
314 {
315     const char *field, *subfield;
316
317     while ((field = find_field(rec, "650")))
318     {
319         rec = field; 
320         if ((subfield = find_subfield(field, 'a')))
321         {
322             char *e, *ef;
323             char buf[1024];
324             int len;
325
326             ef = index(subfield, '\n');
327             if (!ef)
328                 return;
329             if ((e = index(subfield, '\t')) && e < ef)
330                 ef = e;
331             while (ef > subfield && !isalpha(*(ef - 1)) && *(ef - 1) != ')')
332                 ef--;
333             len = ef - subfield;
334             assert(len < 1023);
335             memcpy(buf, subfield, len);
336             buf[len] = '\0';
337 #ifdef FIXME
338             if (*buf)
339                 termlist_insert(s->termlist, buf);
340 #endif
341         }
342     }
343 }
344 #endif
345
346 static void add_facet(struct session *s, const char *type, const char *value)
347 {
348     int i;
349
350     for (i = 0; i < s->num_termlists; i++)
351         if (!strcmp(s->termlists[i].name, type))
352             break;
353     if (i == s->num_termlists)
354     {
355         if (i == SESSION_MAX_TERMLISTS)
356         {
357             yaz_log(YLOG_FATAL, "Too many termlists");
358             exit(1);
359         }
360         s->termlists[i].name = nmem_strdup(s->nmem, type);
361         s->termlists[i].termlist = termlist_create(s->nmem, s->expected_maxrecs, 15);
362         s->num_termlists = i + 1;
363     }
364     termlist_insert(s->termlists[i].termlist, value);
365 }
366
367 int yaz_marc_write_xml();
368
369 static xmlDoc *normalize_record(struct client *cl, Z_External *rec)
370 {
371     struct conf_retrievalprofile *rprofile = cl->database->rprofile;
372     struct conf_retrievalmap *m;
373     xmlNode *res;
374     xmlDoc *rdoc;
375
376     // First normalize to XML
377     if (rprofile->native_syntax == Nativesyn_iso2709)
378     {
379         char *buf;
380         int len;
381         if (rec->which != Z_External_octet)
382         {
383             yaz_log(YLOG_WARN, "Unexpected external branch, probably BER");
384             return 0;
385         }
386         buf = (char*) rec->u.octet_aligned->buf;
387         len = rec->u.octet_aligned->len;
388         if (yaz_marc_read_iso2709(rprofile->yaz_marc, buf, len) < 0)
389         {
390             yaz_log(YLOG_WARN, "Failed to decode MARC");
391             return 0;
392         }
393         if (yaz_marc_write_xml(rprofile->yaz_marc, &res,
394                     "http://www.loc.gov/MARC21/slim", 0, 0) < 0)
395         {
396             yaz_log(YLOG_WARN, "Failed to encode as XML");
397             return 0;
398         }
399         rdoc = xmlNewDoc("1.0");
400         xmlDocSetRootElement(rdoc, res);
401     }
402     else
403     {
404         yaz_log(YLOG_FATAL, "Unknown native_syntax in normalize_record");
405         exit(1);
406     }
407     for (m = rprofile->maplist; m; m = m->next)
408     {
409         xmlDoc *new;
410         if (m->type != Map_xslt)
411         {
412             yaz_log(YLOG_WARN, "Unknown map type");
413             return 0;
414         }
415         if (!(new = xsltApplyStylesheet(m->stylesheet, rdoc, 0)))
416         {
417             yaz_log(YLOG_WARN, "XSLT transformation failed");
418             return 0;
419         }
420         xmlFreeDoc(rdoc);
421         rdoc = new;
422     }
423     if (global_parameters.dump_records)
424     {
425         fprintf(stderr, "Record:\n----------------\n");
426         xmlDocFormatDump(stderr, rdoc, 1);
427     }
428     return rdoc;
429 }
430
431 // Extract what appears to be years from buf, storing highest and
432 // lowest values.
433 static int extract_years(const char *buf, int *first, int *last)
434 {
435     *first = -1;
436     *last = -1;
437     while (*buf)
438     {
439         const char *e;
440         int len;
441
442         while (*buf && !isdigit(*buf))
443             buf++;
444         len = 0;
445         for (e = buf; *e && isdigit(*e); e++)
446             len++;
447         if (len == 4)
448         {
449             int value = atoi(buf);
450             if (*first < 0 || value < *first)
451                 *first = value;
452             if (*last < 0 || value > *last)
453                 *last = value;
454         }
455         buf = e;
456     }
457     return *first;
458 }
459
460 static struct record *ingest_record(struct client *cl, Z_External *rec)
461 {
462     xmlDoc *xdoc = normalize_record(cl, rec);
463     xmlNode *root, *n;
464     struct record *res;
465     struct record_cluster *cluster;
466     struct session *se = cl->session;
467     xmlChar *mergekey, *mergekey_norm;
468     xmlChar *type;
469     xmlChar *value;
470     struct conf_service *service = global_parameters.server->service;
471
472     if (!xdoc)
473         return 0;
474
475     root = xmlDocGetRootElement(xdoc);
476     if (!(mergekey = xmlGetProp(root, "mergekey")))
477     {
478         yaz_log(YLOG_WARN, "No mergekey found in record");
479         xmlFreeDoc(xdoc);
480         return 0;
481     }
482
483     res = nmem_malloc(se->nmem, sizeof(struct record));
484     res->next = 0;
485     res->metadata = nmem_malloc(se->nmem,
486             sizeof(struct record_metadata*) * service->num_metadata);
487     memset(res->metadata, 0, sizeof(struct record_metadata*) * service->num_metadata);
488
489     mergekey_norm = nmem_strdup(se->nmem, (char*) mergekey);
490     xmlFree(mergekey);
491     normalize_mergekey(mergekey_norm);
492
493     cluster = reclist_insert(se->reclist, res, mergekey_norm, &se->total_merged);
494     if (!cluster)
495     {
496         /* no room for record */
497         xmlFreeDoc(xdoc);
498         return 0;
499     }
500     relevance_newrec(se->relevance, cluster);
501
502     type = value = 0;
503     for (n = root->children; n; n = n->next)
504     {
505         if (type)
506             xmlFree(type);
507         if (value)
508             xmlFree(value);
509         type = value = 0;
510
511         if (n->type != XML_ELEMENT_NODE)
512             continue;
513         if (!strcmp(n->name, "metadata"))
514         {
515             struct conf_metadata *md = 0;
516             struct record_metadata **wheretoput, *newm;
517             int imeta;
518             int first, last;
519
520             type = xmlGetProp(n, "type");
521             value = xmlNodeListGetString(xdoc, n->children, 0);
522             // First, find out what field we're looking at
523             for (imeta = 0; imeta < service->num_metadata; imeta++)
524                 if (!strcmp(type, service->metadata[imeta].name))
525                 {
526                     md = &service->metadata[imeta];
527                     break;
528                 }
529             if (!md)
530             {
531                 yaz_log(YLOG_WARN, "Ignoring unknown metadata element: %s", type);
532                 continue;
533             }
534
535             // Find out where we are putting it
536             if (md->merge == Metadata_merge_no)
537                 wheretoput = &res->metadata[imeta];
538             else
539                 wheretoput = &cluster->metadata[imeta];
540             
541             // Put it there
542             newm = nmem_malloc(se->nmem, sizeof(struct record_metadata));
543             newm->next = 0;
544             if (md->type == Metadata_type_generic)
545             {
546                 newm->data.text = nmem_strdup(se->nmem, value);
547             }
548             else if (md->type == Metadata_type_year)
549             {
550                 if (extract_years(value, &first, &last) < 0)
551                     continue;
552             }
553             else
554             {
555                 yaz_log(YLOG_WARN, "Unknown type in metadata element %s", type);
556                 continue;
557             }
558             if (md->type == Metadata_type_year && md->merge != Metadata_merge_range)
559             {
560                 yaz_log(YLOG_WARN, "Only range merging supported for years");
561                 continue;
562             }
563             if (md->merge == Metadata_merge_unique)
564             {
565                 struct record_metadata *mnode;
566                 for (mnode = *wheretoput; mnode; mnode = mnode->next)
567                     if (!strcmp(mnode->data.text, mnode->data.text))
568                         break;
569                 if (!mnode)
570                 {
571                     newm->next = *wheretoput;
572                     *wheretoput = newm;
573                 }
574             }
575             else if (md->merge == Metadata_merge_longest)
576             {
577                 if (!*wheretoput ||
578                         strlen(newm->data.text) > strlen((*wheretoput)->data.text))
579                 *wheretoput = newm;
580             }
581             else if (md->merge == Metadata_merge_all || md->merge == Metadata_merge_no)
582             {
583                 newm->next = *wheretoput;
584                 *wheretoput = newm;
585             }
586             else if (md->merge == Metadata_merge_range)
587             {
588                 assert(md->type == Metadata_type_year);
589                 if (!*wheretoput)
590                 {
591                     *wheretoput = newm;
592                     (*wheretoput)->data.year.year1 = first;
593                     (*wheretoput)->data.year.year2 = last;
594                 }
595                 else
596                 {
597                     if (first < (*wheretoput)->data.year.year1)
598                         (*wheretoput)->data.year.year1 = first;
599                     if (last > (*wheretoput)->data.year.year2)
600                         (*wheretoput)->data.year.year2 = last;
601                 }
602             }
603             else
604                 yaz_log(YLOG_WARN, "Don't know how to merge on element name %s", md->name);
605
606             if (md->rank)
607                 relevance_countwords(se->relevance, cluster, value, md->rank);
608             if (md->termlist)
609                 add_facet(se, type, value);
610             xmlFree(type);
611             xmlFree(value);
612             type = value = 0;
613         }
614         else
615             yaz_log(YLOG_WARN, "Unexpected element %s in internal record", n->name);
616     }
617
618     xmlFreeDoc(xdoc);
619
620     relevance_donerecord(se->relevance, cluster);
621     se->total_records++;
622
623     return res;
624 }
625
626 static void ingest_records(struct client *cl, Z_Records *r)
627 {
628 #if USE_TIMING
629     yaz_timing_t t = yaz_timing_create();
630 #endif
631     struct record *rec;
632     struct session *s = cl->session;
633     Z_NamePlusRecordList *rlist;
634     int i;
635
636     if (r->which != Z_Records_DBOSD)
637         return;
638     rlist = r->u.databaseOrSurDiagnostics;
639     for (i = 0; i < rlist->num_records; i++)
640     {
641         Z_NamePlusRecord *npr = rlist->records[i];
642
643         if (npr->which != Z_NamePlusRecord_databaseRecord)
644         {
645             yaz_log(YLOG_WARN, "Unexpected record type, probably diagnostic");
646             continue;
647         }
648
649         rec = ingest_record(cl, npr->u.databaseRecord);
650         if (!rec)
651             continue;
652     }
653     if (s->watchlist[SESSION_WATCH_RECORDS].fun && rlist->num_records)
654         session_alert_watch(s, SESSION_WATCH_RECORDS);
655
656 #if USE_TIMING
657     yaz_timing_stop(t);
658     yaz_log(YLOG_LOG, "ingest_records %6.5f %3.2f %3.2f", 
659             yaz_timing_get_real(t), yaz_timing_get_user(t),
660             yaz_timing_get_sys(t));
661     yaz_timing_destroy(&t);
662 #endif
663 }
664
665 static void do_presentResponse(IOCHAN i, Z_APDU *a)
666 {
667     struct connection *co = iochan_getdata(i);
668     struct client *cl = co->client;
669     Z_PresentResponse *r = a->u.presentResponse;
670
671     if (r->records) {
672         Z_Records *recs = r->records;
673         if (recs->which == Z_Records_NSD)
674         {
675             yaz_log(YLOG_WARN, "Non-surrogate diagnostic");
676             cl->diagnostic = *recs->u.nonSurrogateDiagnostic->condition;
677             cl->state = Client_Error;
678         }
679     }
680
681     if (!*r->presentStatus && cl->state != Client_Error)
682     {
683         yaz_log(YLOG_DEBUG, "Good Present response");
684         cl->records += *r->numberOfRecordsReturned;
685         ingest_records(cl, r->records);
686         cl->state = Client_Idle;
687     }
688     else if (*r->presentStatus) 
689     {
690         yaz_log(YLOG_WARN, "Bad Present response");
691         cl->state = Client_Error;
692     }
693 }
694
695 static void handler(IOCHAN i, int event)
696 {
697     struct connection *co = iochan_getdata(i);
698     struct client *cl = co->client;
699     struct session *se = 0;
700
701     if (cl)
702         se = cl->session;
703     else
704     {
705         yaz_log(YLOG_WARN, "Destroying orphan connection");
706         connection_destroy(co);
707         return;
708     }
709
710     if (co->state == Conn_Connecting && event & EVENT_OUTPUT)
711     {
712         int errcode;
713         socklen_t errlen = sizeof(errcode);
714
715         if (getsockopt(cs_fileno(co->link), SOL_SOCKET, SO_ERROR, &errcode,
716             &errlen) < 0 || errcode != 0)
717         {
718             client_fatal(cl);
719             return;
720         }
721         else
722         {
723             yaz_log(YLOG_DEBUG, "Connect OK");
724             co->state = Conn_Open;
725             if (cl)
726                 cl->state = Client_Connected;
727         }
728     }
729
730     else if (event & EVENT_INPUT)
731     {
732         int len = cs_get(co->link, &co->ibuf, &co->ibufsize);
733
734         if (len < 0)
735         {
736             yaz_log(YLOG_WARN|YLOG_ERRNO, "Error reading from Z server");
737             connection_destroy(co);
738             return;
739         }
740         else if (len == 0)
741         {
742             yaz_log(YLOG_WARN, "EOF reading from Z server");
743             connection_destroy(co);
744             return;
745         }
746         else if (len > 1) // We discard input if we have no connection
747         {
748             co->state = Conn_Open;
749
750             if (cl && (cl->requestid == se->requestid || cl->state == Client_Initializing))
751             {
752                 Z_APDU *a;
753
754                 odr_reset(global_parameters.odr_in);
755                 odr_setbuf(global_parameters.odr_in, co->ibuf, len, 0);
756                 if (!z_APDU(global_parameters.odr_in, &a, 0, 0))
757                 {
758                     client_fatal(cl);
759                     return;
760                 }
761                 switch (a->which)
762                 {
763                     case Z_APDU_initResponse:
764                         do_initResponse(i, a);
765                         break;
766                     case Z_APDU_searchResponse:
767                         do_searchResponse(i, a);
768                         break;
769                     case Z_APDU_presentResponse:
770                         do_presentResponse(i, a);
771                         break;
772                     default:
773                         yaz_log(YLOG_WARN, "Unexpected result from server");
774                         client_fatal(cl);
775                         return;
776                 }
777                 // We aren't expecting staggered output from target
778                 // if (cs_more(t->link))
779                 //    iochan_setevent(i, EVENT_INPUT);
780             }
781             else  // we throw away response and go to idle mode
782             {
783                 yaz_log(YLOG_DEBUG, "Ignoring result of expired operation");
784                 cl->state = Client_Idle;
785             }
786         }
787         /* if len==1 we do nothing but wait for more input */
788     }
789
790     if (cl->state == Client_Connected) {
791         send_init(i);
792     }
793
794     if (cl->state == Client_Idle)
795     {
796         if (cl->requestid != se->requestid && *se->query) {
797             send_search(i);
798         }
799         else if (cl->hits > 0 && cl->records < global_parameters.toget &&
800             cl->records < cl->hits) {
801             send_present(i);
802         }
803     }
804 }
805
806 // Disassociate connection from client
807 static void connection_release(struct connection *co)
808 {
809     struct client *cl = co->client;
810
811     yaz_log(YLOG_DEBUG, "Connection release %s", co->host->hostport);
812     if (!cl)
813         return;
814     cl->connection = 0;
815     co->client = 0;
816 }
817
818 // Close connection and recycle structure
819 static void connection_destroy(struct connection *co)
820 {
821     struct host *h = co->host;
822     cs_close(co->link);
823     iochan_destroy(co->iochan);
824
825     yaz_log(YLOG_DEBUG, "Connection destroy %s", co->host->hostport);
826     if (h->connections == co)
827         h->connections = co->next;
828     else
829     {
830         struct connection *pco;
831         for (pco = h->connections; pco && pco->next != co; pco = pco->next)
832             ;
833         if (pco)
834             pco->next = co->next;
835         else
836             abort();
837     }
838     if (co->client)
839     {
840         if (co->client->state != Client_Idle)
841             co->client->state = Client_Disconnected;
842         co->client->connection = 0;
843     }
844     co->next = connection_freelist;
845     connection_freelist = co;
846 }
847
848 // Creates a new connection for client, associated with the host of 
849 // client's database
850 static struct connection *connection_create(struct client *cl)
851 {
852     struct connection *new;
853     COMSTACK link; 
854     int res;
855     void *addr;
856
857     yaz_log(YLOG_DEBUG, "Connection create %s", cl->database->url);
858     if (!(link = cs_create(tcpip_type, 0, PROTO_Z3950)))
859     {
860         yaz_log(YLOG_FATAL|YLOG_ERRNO, "Failed to create comstack");
861         exit(1);
862     }
863
864     if (!(addr = cs_straddr(link, cl->database->host->ipport)))
865     {
866         yaz_log(YLOG_WARN|YLOG_ERRNO, "Lookup of IP address %s failed?", 
867             cl->database->host->ipport);
868         return 0;
869     }
870
871     res = cs_connect(link, addr);
872     if (res < 0)
873     {
874         yaz_log(YLOG_WARN|YLOG_ERRNO, "cs_connect %s", cl->database->url);
875         return 0;
876     }
877
878     if ((new = connection_freelist))
879         connection_freelist = new->next;
880     else
881     {
882         new = xmalloc(sizeof (struct connection));
883         new->ibuf = 0;
884         new->ibufsize = 0;
885     }
886     new->state = Conn_Connecting;
887     new->host = cl->database->host;
888     new->next = new->host->connections;
889     new->host->connections = new;
890     new->client = cl;
891     cl->connection = new;
892     new->link = link;
893
894     new->iochan = iochan_create(cs_fileno(link), handler, 0);
895     iochan_setdata(new->iochan, new);
896     new->iochan->next = channel_list;
897     channel_list = new->iochan;
898     return new;
899 }
900
901 // Close connection and set state to error
902 static void client_fatal(struct client *cl)
903 {
904     yaz_log(YLOG_WARN, "Fatal error from %s", cl->database->url);
905     connection_destroy(cl->connection);
906     cl->state = Client_Error;
907 }
908
909 // Ensure that client has a connection associated
910 static int client_prep_connection(struct client *cl)
911 {
912     struct connection *co;
913     struct session *se = cl->session;
914     struct host *host = cl->database->host;
915
916     co = cl->connection;
917
918     yaz_log(YLOG_DEBUG, "Client prep %s", cl->database->url);
919
920     if (!co)
921     {
922         // See if someone else has an idle connection
923         // We should look at timestamps here to select the longest-idle connection
924         for (co = host->connections; co; co = co->next)
925             if (co->state == Conn_Open && (!co->client || co->client->session != se))
926                 break;
927         if (co)
928         {
929             connection_release(co);
930             cl->connection = co;
931             co->client = cl;
932         }
933         else
934             co = connection_create(cl);
935     }
936     if (co)
937     {
938         if (co->state == Conn_Connecting)
939         {
940             cl->state = Client_Connecting;
941             iochan_setflag(co->iochan, EVENT_OUTPUT);
942         }
943         else if (co->state == Conn_Open)
944         {
945             if (cl->state == Client_Error || cl->state == Client_Disconnected)
946                 cl->state = Client_Idle;
947             iochan_setflag(co->iochan, EVENT_OUTPUT);
948         }
949         return 1;
950     }
951     else
952         return 0;
953 }
954
955 // This function will most likely vanish when a proper target profile mechanism is
956 // introduced.
957 void load_simpletargets(const char *fn)
958 {
959     FILE *f = fopen(fn, "r");
960     char line[256];
961
962     if (!f)
963     {
964         yaz_log(YLOG_WARN|YLOG_ERRNO, "open %s", fn);
965         exit(1);
966     }
967
968     while (fgets(line, 255, f))
969     {
970         char *url, *db;
971         struct host *host;
972         struct database *database;
973
974         if (strncmp(line, "target ", 7))
975             continue;
976         url = line + 7;
977         url[strlen(url) - 1] = '\0';
978         yaz_log(YLOG_DEBUG, "Target: %s", url);
979         if ((db = strchr(url, '/')))
980             *(db++) = '\0';
981         else
982             db = "Default";
983
984         for (host = hosts; host; host = host->next)
985             if (!strcmp(url, host->hostport))
986                 break;
987         if (!host)
988         {
989             struct addrinfo *addrinfo, hints;
990             char *port;
991             char ipport[128];
992             unsigned char addrbuf[4];
993             int res;
994
995             host = xmalloc(sizeof(struct host));
996             host->hostport = xstrdup(url);
997             host->connections = 0;
998
999             if ((port = strchr(url, ':')))
1000                 *(port++) = '\0';
1001             else
1002                 port = "210";
1003
1004             hints.ai_flags = 0;
1005             hints.ai_family = PF_INET;
1006             hints.ai_socktype = SOCK_STREAM;
1007             hints.ai_protocol = IPPROTO_TCP;
1008             hints.ai_addrlen = 0;
1009             hints.ai_addr = 0;
1010             hints.ai_canonname = 0;
1011             hints.ai_next = 0;
1012             // This is not robust code. It assumes that getaddrinfo returns AF_INET
1013             // address.
1014             if ((res = getaddrinfo(url, port, &hints, &addrinfo)))
1015             {
1016                 yaz_log(YLOG_WARN, "Failed to resolve %s: %s", url, gai_strerror(res));
1017                 xfree(host->hostport);
1018                 xfree(host);
1019                 continue;
1020             }
1021             assert(addrinfo->ai_family == PF_INET);
1022             memcpy(addrbuf, &((struct sockaddr_in*)addrinfo->ai_addr)->sin_addr.s_addr, 4);
1023             sprintf(ipport, "%u.%u.%u.%u:%s",
1024                     addrbuf[0], addrbuf[1], addrbuf[2], addrbuf[3], port);
1025             host->ipport = xstrdup(ipport);
1026             freeaddrinfo(addrinfo);
1027             host->next = hosts;
1028             hosts = host;
1029         }
1030         database = xmalloc(sizeof(struct database));
1031         database->host = host;
1032         database->url = xmalloc(strlen(url) + strlen(db) + 2);
1033         strcpy(database->url, url);
1034         strcat(database->url, "/");
1035         strcat(database->url, db);
1036         
1037         database->databases = xmalloc(2 * sizeof(char *));
1038         database->databases[0] = xstrdup(db);
1039         database->databases[1] = 0;
1040         database->errors = 0;
1041         database->qprofile = 0;
1042         database->rprofile = database_retrieval_profile(database);
1043         database->next = databases;
1044         databases = database;
1045
1046     }
1047     fclose(f);
1048 }
1049
1050 static void pull_terms(NMEM nmem, struct ccl_rpn_node *n, char **termlist, int *num)
1051 {
1052     switch (n->kind)
1053     {
1054         case CCL_RPN_AND:
1055         case CCL_RPN_OR:
1056         case CCL_RPN_NOT:
1057         case CCL_RPN_PROX:
1058             pull_terms(nmem, n->u.p[0], termlist, num);
1059             pull_terms(nmem, n->u.p[1], termlist, num);
1060             break;
1061         case CCL_RPN_TERM:
1062             termlist[(*num)++] = nmem_strdup(nmem, n->u.t.term);
1063             break;
1064         default: // NOOP
1065             break;
1066     }
1067 }
1068
1069 // Extract terms from query into null-terminated termlist
1070 static int extract_terms(NMEM nmem, char *query, char **termlist)
1071 {
1072     int error, pos;
1073     struct ccl_rpn_node *n;
1074     int num = 0;
1075
1076     n = ccl_find_str(global_parameters.ccl_filter, query, &error, &pos);
1077     if (!n)
1078         return -1;
1079     pull_terms(nmem, n, termlist, &num);
1080     termlist[num] = 0;
1081     ccl_rpn_delete(n);
1082     return 0;
1083 }
1084
1085 static struct client *client_create(void)
1086 {
1087     struct client *r;
1088     if (client_freelist)
1089     {
1090         r = client_freelist;
1091         client_freelist = client_freelist->next;
1092     }
1093     else
1094         r = xmalloc(sizeof(struct client));
1095     r->database = 0;
1096     r->connection = 0;
1097     r->session = 0;
1098     r->hits = 0;
1099     r->records = 0;
1100     r->setno = 0;
1101     r->requestid = -1;
1102     r->diagnostic = 0;
1103     r->state = Client_Disconnected;
1104     r->next = 0;
1105     return r;
1106 }
1107
1108 void client_destroy(struct client *c)
1109 {
1110     struct session *se = c->session;
1111     if (c == se->clients)
1112         se->clients = c->next;
1113     else
1114     {
1115         struct client *cc;
1116         for (cc = se->clients; cc && cc->next != c; cc = cc->next)
1117             ;
1118         if (cc)
1119             cc->next = c->next;
1120     }
1121     if (c->connection)
1122         connection_release(c->connection);
1123     c->next = client_freelist;
1124     client_freelist = c;
1125 }
1126
1127 void session_set_watch(struct session *s, int what, session_watchfun fun, void *data)
1128 {
1129     s->watchlist[what].fun = fun;
1130     s->watchlist[what].data = data;
1131 }
1132
1133 void session_alert_watch(struct session *s, int what)
1134 {
1135     if (!s->watchlist[what].fun)
1136         return;
1137     (*s->watchlist[what].fun)(s->watchlist[what].data);
1138     s->watchlist[what].fun = 0;
1139     s->watchlist[what].data = 0;
1140 }
1141
1142 // This needs to be extended with selection criteria
1143 static struct conf_retrievalprofile *database_retrieval_profile(struct database *db)
1144 {
1145     if (!config)
1146     {
1147         yaz_log(YLOG_FATAL, "Must load configuration (-f)");
1148         exit(1);
1149     }
1150     if (!config->retrievalprofiles)
1151     {
1152         yaz_log(YLOG_FATAL, "No retrieval profiles defined");
1153     }
1154     return config->retrievalprofiles;
1155 }
1156
1157 // This should be extended with parameters to control selection criteria
1158 // Associates a set of clients with a session;
1159 int select_targets(struct session *se)
1160 {
1161     struct database *db;
1162     int c = 0;
1163
1164     while (se->clients)
1165         client_destroy(se->clients);
1166     for (db = databases; db; db = db->next)
1167     {
1168         struct client *cl = client_create();
1169         cl->database = db;
1170         cl->session = se;
1171         cl->next = se->clients;
1172         se->clients = cl;
1173         c++;
1174     }
1175     return c;
1176 }
1177
1178 int session_active_clients(struct session *s)
1179 {
1180     struct client *c;
1181     int res = 0;
1182
1183     for (c = s->clients; c; c = c->next)
1184         if (c->connection && (c->state == Client_Connecting ||
1185                     c->state == Client_Initializing ||
1186                     c->state == Client_Searching ||
1187                     c->state == Client_Presenting))
1188             res++;
1189
1190     return res;
1191 }
1192
1193 char *search(struct session *se, char *query)
1194 {
1195     int live_channels = 0;
1196     struct client *cl;
1197
1198     yaz_log(YLOG_DEBUG, "Search");
1199
1200     strcpy(se->query, query);
1201     se->requestid++;
1202     nmem_reset(se->nmem);
1203     for (cl = se->clients; cl; cl = cl->next)
1204     {
1205         cl->hits = -1;
1206         cl->records = 0;
1207         cl->diagnostic = 0;
1208
1209         if (client_prep_connection(cl))
1210             live_channels++;
1211     }
1212     if (live_channels)
1213     {
1214         char *p[512];
1215         int maxrecs = live_channels * global_parameters.toget;
1216         se->num_termlists = 0;
1217         se->reclist = reclist_create(se->nmem, maxrecs);
1218         extract_terms(se->nmem, query, p);
1219         se->relevance = relevance_create(se->nmem, (const char **) p, maxrecs);
1220         se->total_records = se->total_hits = se->total_merged = 0;
1221         se->expected_maxrecs = maxrecs;
1222     }
1223     else
1224         return "NOTARGETS";
1225
1226     return 0;
1227 }
1228
1229 void destroy_session(struct session *s)
1230 {
1231     yaz_log(YLOG_LOG, "Destroying session");
1232     while (s->clients)
1233         client_destroy(s->clients);
1234     nmem_destroy(s->nmem);
1235     wrbuf_free(s->wrbuf, 1);
1236 }
1237
1238 struct session *new_session() 
1239 {
1240     int i;
1241     struct session *session = xmalloc(sizeof(*session));
1242
1243     yaz_log(YLOG_DEBUG, "New pazpar2 session");
1244     
1245     session->total_hits = 0;
1246     session->total_records = 0;
1247     session->num_termlists = 0;
1248     session->reclist = 0;
1249     session->requestid = -1;
1250     session->clients = 0;
1251     session->expected_maxrecs = 0;
1252     session->query[0] = '\0';
1253     session->nmem = nmem_create();
1254     session->wrbuf = wrbuf_alloc();
1255     for (i = 0; i <= SESSION_WATCH_MAX; i++)
1256     {
1257         session->watchlist[i].data = 0;
1258         session->watchlist[i].fun = 0;
1259     }
1260
1261     select_targets(session);
1262
1263     return session;
1264 }
1265
1266 struct hitsbytarget *hitsbytarget(struct session *se, int *count)
1267 {
1268     static struct hitsbytarget res[1000]; // FIXME MM
1269     struct client *cl;
1270
1271     *count = 0;
1272     for (cl = se->clients; cl; cl = cl->next)
1273     {
1274         strcpy(res[*count].id, cl->database->host->hostport);
1275         res[*count].hits = cl->hits;
1276         res[*count].records = cl->records;
1277         res[*count].diagnostic = cl->diagnostic;
1278         res[*count].state = client_states[cl->state];
1279         res[*count].connected  = cl->connection ? 1 : 0;
1280         (*count)++;
1281     }
1282
1283     return res;
1284 }
1285
1286 struct termlist_score **termlist(struct session *s, const char *name, int *num)
1287 {
1288     int i;
1289
1290     for (i = 0; i < s->num_termlists; i++)
1291         if (!strcmp(s->termlists[i].name, name))
1292             return termlist_highscore(s->termlists[i].termlist, num);
1293     return 0;
1294 }
1295
1296 #ifdef MISSING_HEADERS
1297 void report_nmem_stats(void)
1298 {
1299     size_t in_use, is_free;
1300
1301     nmem_get_memory_in_use(&in_use);
1302     nmem_get_memory_free(&is_free);
1303
1304     yaz_log(YLOG_LOG, "nmem stat: use=%ld free=%ld", 
1305             (long) in_use, (long) is_free);
1306 }
1307 #endif
1308
1309 struct record_cluster *show_single(struct session *s, int id)
1310 {
1311     struct record_cluster *r;
1312
1313     reclist_rewind(s->reclist);
1314     while ((r = reclist_read_record(s->reclist)))
1315         if (r->recid == id)
1316             return r;
1317     return 0;
1318 }
1319
1320 struct record_cluster **show(struct session *s, int start, int *num, int *total,
1321                      int *sumhits, NMEM nmem_show)
1322 {
1323     struct record_cluster **recs = nmem_malloc(nmem_show, *num 
1324                                        * sizeof(struct record_cluster *));
1325     int i;
1326 #if USE_TIMING    
1327     yaz_timing_t t = yaz_timing_create();
1328 #endif
1329     relevance_prepare_read(s->relevance, s->reclist);
1330
1331     *total = s->reclist->num_records;
1332     *sumhits = s->total_hits;
1333
1334     for (i = 0; i < start; i++)
1335         if (!reclist_read_record(s->reclist))
1336         {
1337             *num = 0;
1338             recs = 0;
1339             break;
1340         }
1341
1342     for (i = 0; i < *num; i++)
1343     {
1344         struct record_cluster *r = reclist_read_record(s->reclist);
1345         if (!r)
1346         {
1347             *num = i;
1348             break;
1349         }
1350         recs[i] = r;
1351     }
1352 #if USE_TIMING
1353     yaz_timing_stop(t);
1354     yaz_log(YLOG_LOG, "show %6.5f %3.2f %3.2f", 
1355             yaz_timing_get_real(t), yaz_timing_get_user(t),
1356             yaz_timing_get_sys(t));
1357     yaz_timing_destroy(&t);
1358 #endif
1359     return recs;
1360 }
1361
1362 void statistics(struct session *se, struct statistics *stat)
1363 {
1364     struct client *cl;
1365     int count = 0;
1366
1367     memset(stat, 0, sizeof(*stat));
1368     for (cl = se->clients; cl; cl = cl->next)
1369     {
1370         if (!cl->connection)
1371             stat->num_no_connection++;
1372         switch (cl->state)
1373         {
1374             case Client_Connecting: stat->num_connecting++; break;
1375             case Client_Initializing: stat->num_initializing++; break;
1376             case Client_Searching: stat->num_searching++; break;
1377             case Client_Presenting: stat->num_presenting++; break;
1378             case Client_Idle: stat->num_idle++; break;
1379             case Client_Failed: stat->num_failed++; break;
1380             case Client_Error: stat->num_error++; break;
1381             default: break;
1382         }
1383         count++;
1384     }
1385     stat->num_hits = se->total_hits;
1386     stat->num_records = se->total_records;
1387
1388     stat->num_clients = count;
1389 }
1390
1391 static CCL_bibset load_cclfile(const char *fn)
1392 {
1393     CCL_bibset res = ccl_qual_mk();
1394     if (ccl_qual_fname(res, fn) < 0)
1395     {
1396         yaz_log(YLOG_FATAL|YLOG_ERRNO, "%s", fn);
1397         exit(1);
1398     }
1399     return res;
1400 }
1401
1402 static void start_http_listener(void)
1403 {
1404     char hp[128] = "";
1405     struct conf_server *ser = global_parameters.server;
1406
1407     if (*global_parameters.listener_override)
1408         strcpy(hp, global_parameters.listener_override);
1409     else
1410     {
1411         strcpy(hp, ser->host ? ser->host : "");
1412         if (ser->port)
1413         {
1414             if (*hp)
1415                 strcat(hp, ":");
1416             sprintf(hp + strlen(hp), "%d", ser->port);
1417         }
1418     }
1419     http_init(hp);
1420 }
1421
1422 static void start_proxy(void)
1423 {
1424     char hp[128] = "";
1425     struct conf_server *ser = global_parameters.server;
1426
1427     if (*global_parameters.proxy_override)
1428         strcpy(hp, global_parameters.proxy_override);
1429     else if (ser->proxy_host || ser->proxy_port)
1430     {
1431         strcpy(hp, ser->proxy_host ? ser->proxy_host : "");
1432         if (ser->proxy_port)
1433         {
1434             if (*hp)
1435                 strcat(hp, ":");
1436             sprintf(hp + strlen(hp), "%d", ser->proxy_port);
1437         }
1438     }
1439     else
1440         return;
1441
1442     http_set_proxyaddr(hp);
1443 }
1444
1445 int main(int argc, char **argv)
1446 {
1447     int ret;
1448     char *arg;
1449
1450     if (signal(SIGPIPE, SIG_IGN) == SIG_ERR)
1451         yaz_log(YLOG_WARN|YLOG_ERRNO, "signal");
1452
1453     yaz_log_init(YLOG_DEFAULT_LEVEL, "pazpar2", 0);
1454
1455     while ((ret = options("f:x:h:p:C:s:d", argv, argc, &arg)) != -2)
1456     {
1457         switch (ret) {
1458             case 'f':
1459                 if (!read_config(arg))
1460                     exit(1);
1461                 break;
1462             case 'h':
1463                 strcpy(global_parameters.listener_override, arg);
1464                 break;
1465             case 'C':
1466                 global_parameters.ccl_filter = load_cclfile(arg);
1467                 break;
1468             case 'p':
1469                 strcpy(global_parameters.proxy_override, arg);
1470                 break;
1471             case 's':
1472                 load_simpletargets(arg);
1473                 break;
1474             case 'd':
1475                 global_parameters.dump_records = 1;
1476                 break;
1477             default:
1478                 fprintf(stderr, "Usage: pazpar2\n"
1479                         "    -f configfile\n"
1480                         "    -h [host:]port          (REST protocol listener)\n"
1481                         "    -C cclconfig\n"
1482                         "    -s simpletargetfile\n"
1483                         "    -p hostname[:portno]    (HTTP proxy)\n");
1484                 exit(1);
1485         }
1486     }
1487
1488     if (!config)
1489     {
1490         yaz_log(YLOG_FATAL, "Load config with -f");
1491         exit(1);
1492     }
1493     global_parameters.server = config->servers;
1494
1495     start_http_listener();
1496     start_proxy();
1497     global_parameters.ccl_filter = load_cclfile("../etc/default.bib");
1498     global_parameters.yaz_marc = yaz_marc_create();
1499     yaz_marc_subfield_str(global_parameters.yaz_marc, "\t");
1500     global_parameters.odr_in = odr_createmem(ODR_DECODE);
1501     global_parameters.odr_out = odr_createmem(ODR_ENCODE);
1502
1503     event_loop(&channel_list);
1504
1505     return 0;
1506 }
1507
1508 /*
1509  * Local variables:
1510  * c-basic-offset: 4
1511  * indent-tabs-mode: nil
1512  * End:
1513  * vim: shiftwidth=4 tabstop=8 expandtab
1514  */