X-Git-Url: http://pileus.org/git/?a=blobdiff_plain;f=uid.c;h=d069d7d0484b5b2bdeb1c01170fd5187762d537b;hb=f16d8d23439b5569f0c2e1af22494708b507f277;hp=000c02330f3e38abe5fcc7fd6aae1c12d8dfc8c6;hpb=4cb267590512956e78cc054f816fc6294088ce2b;p=~andy%2Ffetchmail diff --git a/uid.c b/uid.c index 000c0233..d069d7d0 100644 --- a/uid.c +++ b/uid.c @@ -1,27 +1,12 @@ -/* - * uid.c -- UIDL handling for POP3 servers without LAST +/** + * \file uid.c + * UID list handling (currently, only for POP3) * * For license terms, see the file COPYING in this directory. - */ - -#include "config.h" - -#include -#if defined(STDC_HEADERS) -#include -#include -#endif -#if defined(HAVE_UNISTD_H) -#include -#endif - -#include "fetchmail.h" - -/* - * Machinery for handling UID lists live here. This is mainly to support - * RFC1725-conformant POP3 servers without a LAST command, but may also be - * useful for making the IMAP4 querying logic UID-oriented, if a future - * revision of IMAP forces me to. + * + * Machinery for handling UID lists live here. This is currently used + * by POP3, but may also be useful for making the IMAP4 querying logic + * UID-oriented. * * These functions are also used by the rest of the code to maintain * string lists. @@ -32,21 +17,21 @@ * considered seen in `oldsaved'. These are messages that were left in * the mailbox and *not deleted* on previous queries (we don't need to * remember the UIDs of deleted messages because ... well, they're gone!) - * This list is initially set up by initialized_saved_list() from the + * This list is initially set up by initialize_saved_list() from the * .fetchids file. * - * Early in the query, during the execution of the protocol-specific + * Early in the query, during the execution of the protocol-specific * getrange code, the driver expects that the host's `newsaved' member * will be filled with a list of UIDs and message numbers representing * the mailbox state. If this list is empty, the server did * not respond to the request for a UID listing. * * Each time a message is fetched, we can check its UID against the - * `oldsaved' list to see if it is old. If not, it should be downloaded - * (and possibly deleted). It should be downloaded anyway if --all - * is on. It should not be deleted if --keep is on. + * `oldsaved' list to see if it is old. + * + * Each time a message-id is seen, we mark it with MARK_SEEN. * - * Each time a message is deleted, we mark its id UID_DELETED from the + * Each time a message is deleted, we mark its id UID_DELETED in the * `newsaved' member. When we want to assert that an expunge has been * done on the server, we call expunge_uid() to register that all * deleted messages are gone by marking them UID_EXPUNGED. @@ -54,280 +39,480 @@ * At the end of the query, the `newsaved' member becomes the * `oldsaved' list. The old `oldsaved' list is freed. * - * At the end of the fetchmail run, non-EXPUNGED members of all + * At the end of the fetchmail run, seen and non-EXPUNGED members of all * current `oldsaved' lists are flushed out to the .fetchids file to - * be picked up by the next run. (The UID_EXPUNGED test means that a - * message marked UID_DELETED may still have its ID go to disk if - * there has been no intervening expunge operation. This typically - * comes up if the query was aborted by a line hit before a quit or - * expunge was sent to the server.) If there are no un-expunged + * be picked up by the next run. If there are no un-expunged * messages, the file is deleted. * - * Note: all comparisons are caseblind! + * One disadvantage of UIDL is that all the UIDs have to be downloaded + * before a search for new messages can be done. Typically, new messages + * are appended to mailboxes. Hence, downloading all UIDs just to download + * a few new mails is a waste of bandwidth. If new messages are always at + * the end of the mailbox, fast UIDL will decrease the time required to + * download new mails. + * + * During fast UIDL, the UIDs of all messages are not downloaded! The first + * unseen message is searched for by using a binary search on UIDs. UIDs + * after the first unseen message are downloaded as and when needed. + * + * The advantages of fast UIDL are (this is noticeable only when the + * mailbox has too many mails): + * + * - There is no need to download the UIDs of all mails right at the start. + * - There is no need to save all the UIDs in memory separately in + * `newsaved' list. + * - There is no need to download the UIDs of seen mail (except for the + * first binary search). + * - The first new mail is downloaded considerably faster. + * + * The disadvantages are: + * + * - Since all UIDs are not downloaded, it is not possible to swap old and + * new list. The current state of the mailbox is essentially a merged state + * of old and new mails. + * - If an intermediate mail has been temporarily refused (say, due to 4xx + * code from the smtp server), this mail may not get downloaded. + * - If 'flush' is used, such intermediate mails will also get deleted. + * + * The first two disadvantages can be overcome by doing a linear search + * once in a while (say, every 10th poll). Also, with flush, fast UIDL + * should be disabled. + * + * Note: some comparisons (those used for DNS address lists) are caseblind! */ -/* UIDs associated with un-queried hosts */ -static struct idlist *scratchlist; +#include "config.h" + +#include +#include +#include +#include +#include +#include +#include + +#include "fetchmail.h" +#include "gettext.h" +#include "sdump.h" + +int dofastuidl = 0; #ifdef POP3_ENABLE +/** UIDs associated with un-queried hosts */ +static struct idlist *scratchlist; + +static int dump_saved_uid(struct uid_db_record *rec, void *unused) +{ + char *t; + + (void)unused; + + t = sdump(rec->id, rec->id_len); + report_build(stdout, " %s", t); + free(t); + + return 0; +} + +/** Read saved IDs from \a idfile and attach to each host in \a hostlist. */ void initialize_saved_lists(struct query *hostlist, const char *idfile) -/* read file of saved IDs and attach to each host */ { - int st; + struct stat statbuf; FILE *tmpfp; struct query *ctl; /* make sure lists are initially empty */ - for (ctl = hostlist; ctl; ctl = ctl->next) - ctl->oldsaved = ctl->newsaved = (struct idlist *)NULL; + for (ctl = hostlist; ctl; ctl = ctl->next) { + ctl->skipped = (struct idlist *)NULL; + + init_uid_db(&ctl->oldsaved); + init_uid_db(&ctl->newsaved); + } + + errno = 0; + + /* + * Croak if the uidl directory does not exist. + * This probably means an NFS mount failed and we can't + * see a uidl file that ought to be there. + * Question: is this a portable check? It's not clear + * that all implementations of lstat() will return ENOTDIR + * rather than plain ENOENT in this case... + */ + if (lstat(idfile, &statbuf) < 0) { + if (errno == ENOTDIR) + { + report(stderr, "lstat: %s: %s\n", idfile, strerror(errno)); + exit(PS_IOERR); + } + } /* let's get stored message UIDs from previous queries */ - if ((tmpfp = fopen(idfile, "r")) != (FILE *)NULL) { - char buf[POPBUFSIZE+1],host[HOSTLEN+1],user[USERNAMELEN+1],id[IDLEN+1]; + if ((tmpfp = fopen(idfile, "r")) != (FILE *)NULL) + { + char buf[POPBUFSIZE+1]; + char *host = NULL; /* pacify -Wall */ + char *user; + char *id; + char *atsign; /* temp pointer used in parsing user and host */ + char *delimp1; + char saveddelim1; + char *delimp2; + char saveddelim2 = '\0'; /* pacify -Wall */ while (fgets(buf, POPBUFSIZE, tmpfp) != (char *)NULL) { - /* possible lossage here with very old versions of sscanf(3)... */ - if ((st = sscanf(buf, "%[^@]@%s %s\n", user, host, id)) == 3) + /* + * At this point, we assume the bug has two fields -- a user@host + * part, and an ID part. Either field may contain spurious @ signs. + * The previous version of this code presumed one could split at + * the rightmost '@'. This is not correct, as InterMail puts an + * '@' in the UIDL. + */ + + /* first, skip leading spaces */ + user = buf + strspn(buf, " \t"); + + /* + * First, we split the buf into a userhost part and an id + * part ... but id doesn't necessarily start with a '<', + * espescially if the POP server returns an X-UIDL header + * instead of a Message-ID, as GMX's (www.gmx.net) POP3 + * StreamProxy V1.0 does. + * + * this is one other trick. The userhost part + * may contain ' ' in the user part, at least in + * the lotus notes case. + * So we start looking for the '@' after which the + * host will follow with the ' ' separator with the id. + * + * XXX FIXME: There is a case this code cannot handle: + * the user name cannot have blanks after a '@'. + */ + if ((delimp1 = strchr(user, '@')) != NULL && + (id = strchr(delimp1,' ')) != NULL) { - for (ctl = hostlist; ctl; ctl = ctl->next) - { - if (ctl->server.truename && - strcasecmp(host, ctl->server.truename) == 0 - && strcasecmp(user, ctl->remotename) == 0) - { - save_str(&ctl->oldsaved, UID_KEPT, id); + for (delimp1 = id; delimp1 >= user; delimp1--) + if ((*delimp1 != ' ') && (*delimp1 != '\t')) break; - } + + /* + * It should be safe to assume that id starts after + * the " " - after all, we're writing the " " + * ourselves in write_saved_lists() :-) + */ + id = id + strspn(id, " "); + + delimp1++; /* but what if there is only white space ?!? */ + /* we have at least one @, else we are not in this branch */ + saveddelim1 = *delimp1; /* save char after token */ + *delimp1 = '\0'; /* delimit token with \0 */ + + /* now remove trailing white space chars from id */ + if ((delimp2 = strpbrk(id, " \t\n")) != NULL ) { + saveddelim2 = *delimp2; + *delimp2 = '\0'; } - /* if it's not in a host we're querying, save it anyway */ - if (ctl == (struct query *)NULL) - save_str(&scratchlist, UID_KEPT, buf); + atsign = strrchr(user, '@'); + /* we have at least one @, else we are not in this branch */ + *atsign = '\0'; + host = atsign + 1; + + /* find uidl db and save it */ + for (ctl = hostlist; ctl; ctl = ctl->next) { + if (strcasecmp(host, ctl->server.queryname) == 0 + && strcasecmp(user, ctl->remotename) == 0) { + uid_db_insert(&ctl->oldsaved, id, UID_SEEN); + break; + } + } + /* + * If it's not in a host we're querying, + * save it anyway. Otherwise we'd lose UIDL + * information any time we queried an explicit + * subset of hosts. + */ + if (ctl == (struct query *)NULL) { + /* restore string */ + *delimp1 = saveddelim1; + *atsign = '@'; + if (delimp2 != NULL) { + *delimp2 = saveddelim2; + } + save_str(&scratchlist, buf, UID_SEEN); + } } } - fclose(tmpfp); + fclose(tmpfp); /* not checking should be safe, mode was "r" */ } -} -#endif /* POP3_ENABLE */ -struct idlist *save_str(struct idlist **idl, int num, const char *str) -/* save a number/UID pair on the given UID list */ -{ - struct idlist **end; + if (outlevel >= O_DEBUG) + { + struct idlist *idp; - /* do it nonrecursively so the list is in the right order */ - for (end = idl; *end; end = &(*end)->next) - continue; + for (ctl = hostlist; ctl; ctl = ctl->next) + { + report_build(stdout, GT_("Old UID list from %s:"), + ctl->server.pollname); + + if (!uid_db_n_records(&ctl->oldsaved)) + report_build(stdout, GT_(" ")); + else + traverse_uid_db(&ctl->oldsaved, dump_saved_uid, NULL); - *end = (struct idlist *)xmalloc(sizeof(struct idlist)); - (*end)->val.num = num; - (*end)->id = str ? xstrdup(str) : (char *)NULL; - (*end)->next = NULL; + report_complete(stdout, "\n"); + } - return(*end); + report_build(stdout, GT_("Scratch list of UIDs:")); + if (!scratchlist) + report_build(stdout, GT_(" ")); + else for (idp = scratchlist; idp; idp = idp->next) { + char *t = sdump(idp->id, strlen(idp->id)-1); + report_build(stdout, " %s\n", t); + free(t); + } + report_complete(stdout, "\n"); + } } -void free_str_list(struct idlist **idl) -/* free the given UID list */ +static int mark_as_expunged_if(struct uid_db_record *rec, void *unused) { - if (*idl == (struct idlist *)NULL) - return; + (void)unused; - free_str_list(&(*idl)->next); - free ((*idl)->id); - free(*idl); - *idl = (struct idlist *)NULL; + if (rec->status == UID_DELETED) rec->status = UID_EXPUNGED; + return 0; } -void save_str_pair(struct idlist **idl, const char *str1, const char *str2) -/* save an ID pair on the given list */ +/** Assert that all UIDs marked deleted in query \a ctl have actually been +expunged. */ +void expunge_uids(struct query *ctl) { - struct idlist **end; - - /* do it nonrecursively so the list is in the right order */ - for (end = idl; *end; end = &(*end)->next) - continue; - - *end = (struct idlist *)xmalloc(sizeof(struct idlist)); - (*end)->id = str1 ? xstrdup(str1) : (char *)NULL; - if (str2) - (*end)->val.id2 = xstrdup(str2); - else - (*end)->val.id2 = (char *)NULL; - (*end)->next = (struct idlist *)NULL; + traverse_uid_db(dofastuidl ? &ctl->oldsaved : &ctl->newsaved, + mark_as_expunged_if, NULL); } -#ifdef __UNUSED__ -void free_str_pair_list(struct idlist **idl) -/* free the given ID pair list */ +static const char *str_uidmark(int mark) { - if (*idl == (struct idlist *)NULL) - return; - - free_idpair_list(&(*idl)->next); - free ((*idl)->id); - free ((*idl)->val.id2); - free(*idl); - *idl = (struct idlist *)NULL; + static char buf[20]; + + switch(mark) { + case UID_UNSEEN: + return "UNSEEN"; + case UID_SEEN: + return "SEEN"; + case UID_EXPUNGED: + return "EXPUNGED"; + case UID_DELETED: + return "DELETED"; + default: + if (snprintf(buf, sizeof(buf), "MARK=%d", mark) < 0) + return "ERROR"; + else + return buf; + } } -#endif -int str_in_list(struct idlist **idl, const char *str) -/* is a given ID in the given list? (comparison is caseblind) */ +static int dump_uid_db_record(struct uid_db_record *rec, void *arg) { - if (*idl == (struct idlist *)NULL || str == (char *) NULL) - return(0); - else if (strcasecmp(str, (*idl)->id) == 0) - return(1); - else - return(str_in_list(&(*idl)->next, str)); -} + unsigned *n_recs; + char *t; -int str_nr_in_list( struct idlist **idl, const char *str ) - /* return the position of str in idl */ -{ - int nr; - struct idlist *walk; - if ( !str ) - return -1; - for( walk = *idl, nr = 0; walk; nr ++, walk = walk->next ) - if( strcasecmp( str, walk->id) == 0 ) - return nr; - return -1; -} + n_recs = (unsigned int *)arg; + --*n_recs; -int count_list( struct idlist **idl ) - /* count the number of elements in the list */ -{ - if( !*idl ) - return 0; - return 1 + count_list( &(*idl)->next ); -} + t = sdump(rec->id, rec->id_len); + report_build(stdout, " %s = %s%s", t, str_uidmark(rec->status), *n_recs ? "," : ""); + free(t); -char* str_from_nr_list( struct idlist **idl, int number ) - /* return the number'th string in idl */ -{ - if( !*idl || number < 0) - return 0; - if( number == 0 ) - return (*idl)->id; - return str_from_nr_list(&(*idl)->next, number-1); + return 0; } - -char *str_find(struct idlist **idl, int number) -/* return the id of the given number in the given list. */ +static void dump_uid_db(struct uid_db *db) { - if (*idl == (struct idlist *) 0) - return((char *) 0); - else if (number == (*idl)->val.num) - return((*idl)->id); - else - return(str_find(&(*idl)->next, number)); + unsigned n_recs; + + n_recs = uid_db_n_records(db); + if (!n_recs) { + report_build(stdout, GT_(" ")); + return; + } + + traverse_uid_db(db, dump_uid_db_record, &n_recs); } -char *idpair_find(struct idlist **idl, const char *id) -/* return the id of the given id in the given list (caseblind comparison) */ +/** Finish a successful query */ +void uid_swap_lists(struct query *ctl) { - if (*idl == (struct idlist *) 0) - return((char *) 0); - else if (strcasecmp(id, (*idl)->id) == 0) - return((*idl)->val.id2 ? (*idl)->val.id2 : (*idl)->id); - else - return(idpair_find(&(*idl)->next, id)); + /* debugging code */ + if (outlevel >= O_DEBUG) + { + if (dofastuidl) { + report_build(stdout, GT_("Merged UID list from %s:"), ctl->server.pollname); + dump_uid_db(&ctl->oldsaved); + } else { + report_build(stdout, GT_("New UID list from %s:"), ctl->server.pollname); + dump_uid_db(&ctl->newsaved); + } + report_complete(stdout, "\n"); + } + + /* + * Don't swap UID lists unless we've actually seen UIDLs. + * This is necessary in order to keep UIDL information + * from being heedlessly deleted later on. + * + * Older versions of fetchmail did + * + * free_str_list(&scratchlist); + * + * after swap. This was wrong; we need to preserve the UIDL information + * from unqueried hosts. Unfortunately, not doing this means that + * under some circumstances UIDLs can end up being stored forever -- + * specifically, if a user description is removed from .fetchmailrc + * with UIDLs from that account in .fetchids, there is no way for + * them to ever get garbage-collected. + */ + if (uid_db_n_records(&ctl->newsaved)) + { + swap_uid_db_data(&ctl->newsaved, &ctl->oldsaved); + clear_uid_db(&ctl->newsaved); + } + /* in fast uidl, there is no need to swap lists: the old state of + * mailbox cannot be discarded! */ + else if (outlevel >= O_DEBUG && !dofastuidl) + report(stdout, GT_("not swapping UID lists, no UIDs seen this query\n")); } -int delete_str(struct idlist **idl, int num) -/* delete given message from given list */ +/** Finish a query which had errors */ +void uid_discard_new_list(struct query *ctl) { -#ifdef HARD_DELETE /* not used */ - if (*idl == (struct idlist *)NULL) - return(0); - else if ((*idl)->val.num == num) + /* debugging code */ + if (outlevel >= O_DEBUG) { - struct idlist *next = (*idl)->next; - - free ((*idl)->id); - free(*idl); - *idl = next; - return(1); + /* this is now a merged list! the mails which were seen in this + * poll are marked here. */ + report_build(stdout, GT_("Merged UID list from %s:"), ctl->server.pollname); + dump_uid_db(&ctl->oldsaved); + report_complete(stdout, "\n"); } - else - return(delete_str(&(*idl)->next, num)); -#else - struct idlist *idp; - for (idp = *idl; idp; idp = idp->next) - if (idp->val.num == num) - { - idp->val.num = UID_DELETED; - return(1); - } - return(0); -#endif /* HARD_DELETE */ + if (uid_db_n_records(&ctl->newsaved)) + { + /* new state of mailbox is not reliable */ + if (outlevel >= O_DEBUG) + report(stdout, GT_("discarding new UID list\n")); + clear_uid_db(&ctl->newsaved); + } } -void append_str_list(struct idlist **idl, struct idlist **nidl) -/* append nidl to idl (does not copy *) */ +/** Reset the number associated with each id */ +void uid_reset_num(struct query *ctl) { - if ((*idl) == (struct idlist *)NULL) - *idl = *nidl; - else if ((*idl)->next == (struct idlist *)NULL) - (*idl)->next = *nidl; - else if ((*idl)->next != *nidl) - append_str_list(&(*idl)->next, nidl); + reset_uid_db_nums(&ctl->oldsaved); } -#ifdef POP3_ENABLE -void expunge_uids(struct query *ctl) -/* assert that all UIDs marked deleted have actually been expunged */ +/** Write list of seen messages, at end of run. */ +static int count_seen_deleted(struct uid_db_record *rec, void *arg) { - struct idlist *idl; - - for (idl = ctl->newsaved; idl; idl = idl->next) - if (idl->val.num == UID_DELETED) - idl->val.num = UID_EXPUNGED; + if (rec->status == UID_SEEN || rec->status == UID_DELETED) + ++*(long *)arg; + return 0; } -void update_str_lists(struct query *ctl) -/* perform end-of-query actions on UID lists */ +struct write_saved_info { + struct query *ctl; + FILE *fp; +}; + +static int write_uid_db_record(struct uid_db_record *rec, void *arg) { - free_str_list(&ctl->oldsaved); - ctl->oldsaved = ctl->newsaved; - ctl->newsaved = (struct idlist *) NULL; + struct write_saved_info *info; + int rc; + + if (!(rec->status == UID_SEEN || rec->status == UID_DELETED)) + return 0; + + info = (struct write_saved_info *)arg; + rc = fprintf(info->fp, "%s@%s %s\n", + info->ctl->remotename, info->ctl->server.queryname, + rec->id); + return rc < 0 ? -1 : 0; } +/** Write new list of UIDs (state) to \a idfile. */ void write_saved_lists(struct query *hostlist, const char *idfile) -/* perform end-of-run write of seen-messages list */ { - int idcount; + long idcount; FILE *tmpfp; struct query *ctl; struct idlist *idp; /* if all lists are empty, nuke the file */ idcount = 0; - for (ctl = hostlist; ctl; ctl = ctl->next) { - if (ctl->oldsaved) - idcount++; - } + for (ctl = hostlist; ctl; ctl = ctl->next) + traverse_uid_db(&ctl->oldsaved, count_seen_deleted, &idcount); /* either nuke the file or write updated last-seen IDs */ - if (!idcount) - unlink(idfile); - else - if ((tmpfp = fopen(idfile, "w")) != (FILE *)NULL) { + if (!idcount && !scratchlist) + { + if (outlevel >= O_DEBUG) { + if (access(idfile, F_OK) == 0) + report(stdout, GT_("Deleting fetchids file.\n")); + } + if (unlink(idfile) && errno != ENOENT) + report(stderr, GT_("Error deleting %s: %s\n"), idfile, strerror(errno)); + } else { + char *newnam = (char *)xmalloc(strlen(idfile) + 2); + strcpy(newnam, idfile); + strcat(newnam, "_"); + if (outlevel >= O_DEBUG) + report(stdout, GT_("Writing fetchids file.\n")); + (void)unlink(newnam); /* remove file/link first */ + if ((tmpfp = fopen(newnam, "w")) != (FILE *)NULL) { + struct write_saved_info info; + int errflg = 0; + + info.fp = tmpfp; + for (ctl = hostlist; ctl; ctl = ctl->next) { - for (idp = ctl->oldsaved; idp; idp = idp->next) - if (idp->val.num != UID_EXPUNGED) - fprintf(tmpfp, "%s@%s %s\n", - ctl->remotename, ctl->server.truename, idp->id); + info.ctl = ctl; + + if (traverse_uid_db(&ctl->oldsaved, write_uid_db_record, &info) < 0) { + int e = errno; + report(stderr, GT_("Write error on fetchids file %s: %s\n"), newnam, strerror(e)); + errflg = 1; + goto bailout; + } } + for (idp = scratchlist; idp; idp = idp->next) - fputs(idp->id, tmpfp); - fclose(tmpfp); + if (EOF == fputs(idp->id, tmpfp)) { + int e = errno; + report(stderr, GT_("Write error on fetchids file %s: %s\n"), newnam, strerror(e)); + errflg = 1; + goto bailout; + } + +bailout: + (void)fflush(tmpfp); /* return code ignored, we check ferror instead */ + errflg |= ferror(tmpfp); + errflg |= fclose(tmpfp); + /* if we could write successfully, move into place; + * otherwise, drop */ + if (errflg) { + report(stderr, GT_("Error writing to fetchids file %s, old file left in place.\n"), newnam); + unlink(newnam); + } else { + if (rename(newnam, idfile)) { + report(stderr, GT_("Cannot rename fetchids file %s to %s: %s\n"), newnam, idfile, strerror(errno)); + } + } + } else { + report(stderr, GT_("Cannot open fetchids file %s for writing: %s\n"), newnam, strerror(errno)); } + free(newnam); + } } #endif /* POP3_ENABLE */