2 * uid.c -- UIDL handling for POP3 servers without LAST
4 * For license terms, see the file COPYING in this directory.
13 #if defined(STDC_HEADERS)
17 #if defined(HAVE_UNISTD_H)
21 #include "fetchmail.h"
25 * Machinery for handling UID lists live here. This is mainly to support
26 * RFC1725/RFC1939-conformant POP3 servers without a LAST command, but may also
27 * be useful for making the IMAP4 querying logic UID-oriented, if a future
28 * revision of IMAP forces me to.
30 * These functions are also used by the rest of the code to maintain
35 * At start of a query, we have a (possibly empty) list of UIDs to be
36 * considered seen in `oldsaved'. These are messages that were left in
37 * the mailbox and *not deleted* on previous queries (we don't need to
38 * remember the UIDs of deleted messages because ... well, they're gone!)
39 * This list is initially set up by initialize_saved_list() from the
42 * Early in the query, during the execution of the protocol-specific
43 * getrange code, the driver expects that the host's `newsaved' member
44 * will be filled with a list of UIDs and message numbers representing
45 * the mailbox state. If this list is empty, the server did
46 * not respond to the request for a UID listing.
48 * Each time a message is fetched, we can check its UID against the
49 * `oldsaved' list to see if it is old.
51 * Each time a message-id is seen, we mark it with MARK_SEEN.
53 * Each time a message is deleted, we mark its id UID_DELETED in the
54 * `newsaved' member. When we want to assert that an expunge has been
55 * done on the server, we call expunge_uid() to register that all
56 * deleted messages are gone by marking them UID_EXPUNGED.
58 * At the end of the query, the `newsaved' member becomes the
59 * `oldsaved' list. The old `oldsaved' list is freed.
61 * At the end of the fetchmail run, seen and non-EXPUNGED members of all
62 * current `oldsaved' lists are flushed out to the .fetchids file to
63 * be picked up by the next run. If there are no un-expunged
64 * messages, the file is deleted.
66 * One disadvantage of UIDL is that all the UIDs have to be downloaded
67 * before a search for new messages can be done. Typically, new messages
68 * are appended to mailboxes. Hence, downloading all UIDs just to download
69 * a few new mails is a waste of bandwidth. If new messages are always at
70 * the end of the mailbox, fast UIDL will decrease the time required to
73 * During fast UIDL, the UIDs of all messages are not downloaded! The first
74 * unseen message is searched for by using a binary search on UIDs. UIDs
75 * after the first unseen message are downloaded as and when needed.
77 * The advantages of fast UIDL are (this is noticeable only when the
78 * mailbox has too many mails):
80 * - There is no need to download the UIDs of all mails right at the start.
81 * - There is no need to save all the UIDs in memory separately in
83 * - There is no need to download the UIDs of seen mail (except for the
84 * first binary search).
85 * - The first new mail is downloaded considerably faster.
87 * The disadvantages are:
89 * - Since all UIDs are not downloaded, it is not possible to swap old and
90 * new list. The current state of the mailbox is essentially a merged state
91 * of old and new mails.
92 * - If an intermediate mail has been temporarily refused (say, due to 4xx
93 * code from the smtp server), this mail may not get downloaded.
94 * - If 'flush' is used, such intermediate mails will also get deleted.
96 * The first two disadvantages can be overcome by doing a linear search
97 * once in a while (say, every 10th poll). Also, with flush, fast UIDL
100 * Note: some comparisons (those used for DNS address lists) are caseblind!
105 /* UIDs associated with un-queried hosts */
106 static struct idlist *scratchlist;
109 void initialize_saved_lists(struct query *hostlist, const char *idfile)
110 /* read file of saved IDs and attach to each host */
116 /* make sure lists are initially empty */
117 for (ctl = hostlist; ctl; ctl = ctl->next) {
118 ctl->skipped = (struct idlist *)NULL;
119 ctl->oldsaved = (struct idlist *)NULL;
120 ctl->newsaved = (struct idlist *)NULL;
121 ctl->oldsavedend = &ctl->oldsaved;
127 * Croak if the uidl directory does not exist.
128 * This probably means an NFS mount failed and we can't
129 * see a uidl file that ought to be there.
130 * Question: is this a portable check? It's not clear
131 * that all implementations of lstat() will return ENOTDIR
132 * rather than plain ENOENT in this case...
134 if (lstat(idfile, &statbuf) < 0) {
135 if (errno == ENOTDIR)
137 report(stderr, "lstat: %s: %s\n", idfile, strerror(errno));
142 /* let's get stored message UIDs from previous queries */
143 if ((tmpfp = fopen(idfile, "r")) != (FILE *)NULL)
145 char buf[POPBUFSIZE+1];
146 char *host = NULL; /* pacify -Wall */
149 char *atsign; /* temp pointer used in parsing user and host */
153 char saveddelim2 = '\0'; /* pacify -Wall */
155 while (fgets(buf, POPBUFSIZE, tmpfp) != (char *)NULL)
158 * At this point, we assume the bug has two fields -- a user@host
159 * part, and an ID part. Either field may contain spurious @ signs.
160 * The previous version of this code presumed one could split at
161 * the rightmost '@'. This is not correct, as InterMail puts an
165 /* first, skip leading spaces */
166 user = buf + strspn(buf, " \t");
169 * First, we split the buf into a userhost part and an id
170 * part ... but id doesn't necessarily start with a '<',
171 * espescially if the POP server returns an X-UIDL header
172 * instead of a Message-ID, as GMX's (www.gmx.net) POP3
173 * StreamProxy V1.0 does.
175 * this is one other trick. The userhost part
176 * may contain ' ' in the user part, at least in
177 * the lotus notes case.
178 * So we start looking for the '@' after which the
179 * host will follow with the ' ' seperator finaly id.
181 if ((delimp1 = strchr(user, '@')) != NULL &&
182 (id = strchr(delimp1,' ')) != NULL)
184 for (delimp1 = id; delimp1 >= user; delimp1--)
185 if ((*delimp1 != ' ') && (*delimp1 != '\t'))
189 * It should be safe to assume that id starts after
190 * the " " - after all, we're writing the " "
191 * ourselves in write_saved_lists() :-)
193 id = id + strspn(id, " ");
195 delimp1++; /* but what if there is only white space ?!? */
196 saveddelim1 = *delimp1; /* save char after token */
197 *delimp1 = '\0'; /* delimit token with \0 */
200 /* now remove trailing white space chars from id */
201 if ((delimp2 = strpbrk(id, " \t\n")) != NULL ) {
202 saveddelim2 = *delimp2;
205 atsign = strrchr(user, '@');
211 for (ctl = hostlist; ctl; ctl = ctl->next) {
212 if (strcasecmp(host, ctl->server.queryname) == 0
213 && strcasecmp(user, ctl->remotename) == 0) {
215 save_str(&ctl->oldsaved, id, UID_SEEN);
220 * If it's not in a host we're querying,
221 * save it anyway. Otherwise we'd lose UIDL
222 * information any time we queried an explicit
225 if (ctl == (struct query *)NULL) {
227 *delimp1 = saveddelim1;
229 if (delimp2 != NULL) {
230 *delimp2 = saveddelim2;
232 save_str(&scratchlist, buf, UID_SEEN);
237 fclose(tmpfp); /* not checking should be safe, mode was "r" */
240 if (outlevel >= O_DEBUG)
245 for (ctl = hostlist; ctl; ctl = ctl->next)
246 if (ctl->server.uidl)
248 report_build(stdout, GT_("Old UID list from %s:"),
249 ctl->server.pollname);
250 for (idp = ctl->oldsaved; idp; idp = idp->next)
251 report_build(stdout, " %s", idp->id);
253 report_build(stdout, GT_(" <empty>"));
254 report_complete(stdout, "\n");
260 report_build(stdout, GT_("Scratch list of UIDs:"));
261 for (idp = scratchlist; idp; idp = idp->next)
262 report_build(stdout, " %s", idp->id);
264 report_build(stdout, GT_(" <empty>"));
265 report_complete(stdout, "\n");
269 #endif /* POP3_ENABLE */
271 /* return a pointer to the last element of the list to help the quick,
272 * constant-time addition to the list, NOTE: this function does not dup
273 * the string, the caller must do that. */
274 /*@shared@*/ static struct idlist **save_str_quick(/*@shared@*/ struct idlist **idl,
275 /*@only@*/ char *str, flag status)
276 /* save a number/UID pair on the given UID list */
280 /* do it nonrecursively so the list is in the right order */
281 for (end = idl; *end; end = &(*end)->next)
284 *end = (struct idlist *)xmalloc(sizeof(struct idlist));
286 (*end)->val.status.mark = status;
287 (*end)->val.status.num = 0;
293 /* return the end list element for direct modification */
294 struct idlist *save_str(struct idlist **idl, const char *str, flag st)
296 return *save_str_quick(idl, str ? xstrdup(str) : NULL, st);
299 void free_str_list(struct idlist **idl)
300 /* free the given UID list */
302 struct idlist *i = *idl;
305 struct idlist *t = i->next;
313 void save_str_pair(struct idlist **idl, const char *str1, const char *str2)
314 /* save an ID pair on the given list */
318 /* do it nonrecursively so the list is in the right order */
319 for (end = idl; *end; end = &(*end)->next)
322 *end = (struct idlist *)xmalloc(sizeof(struct idlist));
323 (*end)->id = str1 ? xstrdup(str1) : (char *)NULL;
325 (*end)->val.id2 = xstrdup(str2);
327 (*end)->val.id2 = (char *)NULL;
328 (*end)->next = (struct idlist *)NULL;
332 void free_str_pair_list(struct idlist **idl)
333 /* free the given ID pair list */
335 if (*idl == (struct idlist *)NULL)
338 free_idpair_list(&(*idl)->next);
340 free ((*idl)->val.id2);
342 *idl = (struct idlist *)NULL;
346 struct idlist *str_in_list(struct idlist **idl, const char *str, const flag caseblind)
347 /* is a given ID in the given list? (comparison may be caseblind) */
351 for( walk = *idl; walk; walk = walk->next )
352 if( strcasecmp( str, walk->id) == 0 )
355 for( walk = *idl; walk; walk = walk->next )
356 if( strcmp( str, walk->id) == 0 )
362 int str_nr_in_list( struct idlist **idl, const char *str )
363 /* return the position of str in idl */
369 for( walk = *idl, nr = 0; walk; nr ++, walk = walk->next )
370 if( strcmp( str, walk->id) == 0 )
375 int str_nr_last_in_list( struct idlist **idl, const char *str)
376 /* return the last position of str in idl */
382 for( walk = *idl, nr = 0; walk; nr ++, walk = walk->next )
383 if( strcmp( str, walk->id) == 0 )
388 void str_set_mark( struct idlist **idl, const char *str, const flag val)
389 /* update the mark on an of an id to given value */
395 for(walk = *idl, nr = 0; walk; nr ++, walk = walk->next)
396 if (strcmp(str, walk->id) == 0)
397 walk->val.status.mark = val;
400 int count_list( struct idlist **idl)
401 /* count the number of elements in the list */
405 return 1 + count_list( &(*idl)->next );
408 /*@null@*/ char *str_from_nr_list(struct idlist **idl, long number)
409 /* return the number'th string in idl */
411 if( !*idl || number < 0)
415 return str_from_nr_list(&(*idl)->next, number-1);
419 char *str_find(struct idlist **idl, long number)
420 /* return the id of the given number in the given list. */
422 if (*idl == (struct idlist *) 0)
424 else if (number == (*idl)->val.status.num)
427 return(str_find(&(*idl)->next, number));
430 struct idlist *id_find(struct idlist **idl, long number)
431 /* return the id of the given number in the given list. */
434 for (idp = *idl; idp; idp = idp->next)
435 if (idp->val.status.num == number)
440 char *idpair_find(struct idlist **idl, const char *id)
441 /* return the id of the given id in the given list (caseblind comparison) */
443 if (*idl == (struct idlist *) 0)
445 else if (strcasecmp(id, (*idl)->id) == 0)
446 return((*idl)->val.id2 ? (*idl)->val.id2 : (*idl)->id);
448 return(idpair_find(&(*idl)->next, id));
451 int delete_str(struct idlist **idl, long num)
452 /* delete given message from given list */
456 for (idp = *idl; idp; idp = idp->next)
457 if (idp->val.status.num == num)
459 idp->val.status.mark = UID_DELETED;
465 struct idlist *copy_str_list(struct idlist *idl)
466 /* copy the given UID list */
468 struct idlist *newnode ;
470 if (idl == (struct idlist *)NULL)
474 newnode = (struct idlist *)xmalloc(sizeof(struct idlist));
475 memcpy(newnode, idl, sizeof(struct idlist));
476 newnode->next = copy_str_list(idl->next);
481 void append_str_list(struct idlist **idl, struct idlist **nidl)
482 /* append nidl to idl (does not copy *) */
484 if ((*nidl) == (struct idlist *)NULL || *nidl == *idl)
486 else if ((*idl) == (struct idlist *)NULL)
488 else if ((*idl)->next == (struct idlist *)NULL)
489 (*idl)->next = *nidl;
490 else if ((*idl)->next != *nidl)
491 append_str_list(&(*idl)->next, nidl);
495 void expunge_uids(struct query *ctl)
496 /* assert that all UIDs marked deleted have actually been expunged */
500 for (idl = dofastuidl ? ctl->oldsaved : ctl->newsaved; idl; idl = idl->next)
501 if (idl->val.status.mark == UID_DELETED)
502 idl->val.status.mark = UID_EXPUNGED;
505 void uid_swap_lists(struct query *ctl)
509 if (ctl->server.uidl && outlevel >= O_DEBUG)
514 report_build(stdout, GT_("Merged UID list from %s:"), ctl->server.pollname);
516 report_build(stdout, GT_("New UID list from %s:"), ctl->server.pollname);
517 for (idp = dofastuidl ? ctl->oldsaved : ctl->newsaved; idp; idp = idp->next)
518 report_build(stdout, " %s = %d", idp->id, idp->val.status.mark);
520 report_build(stdout, GT_(" <empty>"));
521 report_complete(stdout, "\n");
525 * Don't swap UID lists unless we've actually seen UIDLs.
526 * This is necessary in order to keep UIDL information
527 * from being heedlessly deleted later on.
529 * Older versions of fetchmail did
531 * free_str_list(&scratchlist);
533 * after swap. This was wrong; we need to preserve the UIDL information
534 * from unqueried hosts. Unfortunately, not doing this means that
535 * under some circumstances UIDLs can end up being stored forever --
536 * specifically, if a user description is removed from .fetchmailrc
537 * with UIDLs from that account in .fetchids, there is no way for
538 * them to ever get garbage-collected.
542 /* old state of mailbox may now be irrelevant */
543 struct idlist *temp = ctl->oldsaved;
544 if (outlevel >= O_DEBUG)
545 report(stdout, GT_("swapping UID lists\n"));
546 ctl->oldsaved = ctl->newsaved;
547 ctl->newsaved = (struct idlist *) NULL;
548 free_str_list(&temp);
550 /* in fast uidl, there is no need to swap lists: the old state of
551 * mailbox cannot be discarded! */
552 else if (outlevel >= O_DEBUG && !dofastuidl)
553 report(stdout, GT_("not swapping UID lists, no UIDs seen this query\n"));
556 void uid_discard_new_list(struct query *ctl)
557 /* finish a query which had errors */
560 if (ctl->server.uidl && outlevel >= O_DEBUG)
564 /* this is now a merged list! the mails which were seen in this
565 * poll are marked here. */
566 report_build(stdout, GT_("Merged UID list from %s:"), ctl->server.pollname);
567 for (idp = ctl->oldsaved; idp; idp = idp->next)
568 report_build(stdout, " %s = %d", idp->id, idp->val.status.mark);
570 report_build(stdout, GT_(" <empty>"));
571 report_complete(stdout, "\n");
576 /* new state of mailbox is not reliable */
577 if (outlevel >= O_DEBUG)
578 report(stdout, GT_("discarding new UID list\n"));
579 free_str_list(&ctl->newsaved);
580 ctl->newsaved = (struct idlist *) NULL;
584 void uid_reset_num(struct query *ctl)
585 /* reset the number associated with each id */
588 for (idp = ctl->oldsaved; idp; idp = idp->next)
589 idp->val.status.num = 0;
592 void write_saved_lists(struct query *hostlist, const char *idfile)
593 /* perform end-of-run write of seen-messages list */
600 /* if all lists are empty, nuke the file */
602 for (ctl = hostlist; ctl; ctl = ctl->next) {
603 for (idp = ctl->oldsaved; idp; idp = idp->next)
604 if (idp->val.status.mark == UID_SEEN
605 || idp->val.status.mark == UID_DELETED)
609 /* either nuke the file or write updated last-seen IDs */
610 if (!idcount && !scratchlist)
612 if (outlevel >= O_DEBUG)
613 report(stdout, GT_("Deleting fetchids file.\n"));
614 if (unlink(idfile) && errno != ENOENT)
615 report(stderr, GT_("Error deleting %s: %s\n"), idfile, strerror(errno));
617 char *newnam = xmalloc(strlen(idfile) + 2);
618 strcpy(newnam, idfile);
620 if (outlevel >= O_DEBUG)
621 report(stdout, GT_("Writing fetchids file.\n"));
622 (void)unlink(newnam); /* remove file/link first */
623 if ((tmpfp = fopen(newnam, "w")) != (FILE *)NULL) {
625 for (ctl = hostlist; ctl; ctl = ctl->next) {
626 for (idp = ctl->oldsaved; idp; idp = idp->next)
627 if (idp->val.status.mark == UID_SEEN
628 || idp->val.status.mark == UID_DELETED)
629 fprintf(tmpfp, "%s@%s %s\n",
630 ctl->remotename, ctl->server.queryname, idp->id);
632 for (idp = scratchlist; idp; idp = idp->next)
633 fputs(idp->id, tmpfp);
635 errflg = ferror(tmpfp);
637 /* if we could write successfully, move into place;
640 report(stderr, GT_("Error writing to fetchids file %s, old file left in place.\n"), newnam);
643 if (rename(newnam, idfile)) {
644 report(stderr, GT_("Cannot rename fetchids file %s to %s: %s\n"), newnam, idfile, strerror(errno));
648 report(stderr, GT_("Cannot open fetchids file %s for writing: %s\n"), newnam, strerror(errno));
653 #endif /* POP3_ENABLE */
655 /* uid.c ends here */