3 * tailing reliable realtime streaming feeder for inn
5 * Copyright (C) 2010 Ian Jackson <ijackson@chiark.greenend.org.uk>
7 * This program is free software: you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation, either version 3 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program. If not, see <http://www.gnu.org/licenses/>.
20 * (I believe that when you compile and link this as part of the inn2
21 * build, with the Makefile runes I have provided, all the libraries
22 * and files which end up included in innduct are licence-compatible
23 * with GPLv3. If not then please let me know. -Ian Jackson.)
27 * Newsfeeds file entries should look like this:
28 * host.name.of.site[/exclude,exclude,...]\
29 * :pattern,pattern...[/distribution,distribution...]\
33 * sitename[/exclude,exclude,...]\
34 * :pattern,pattern...[/distribution,distribution...]\
40 * or might be blanked out
41 * <spc><spc><spc><spc>....
43 * F site.name main feed file
44 * opened/created, then written, by innd
47 * tokens blanked out by duct when processed
48 * site.name_lock lock preventing multiple ducts
49 * to hold lock must open,F_SETLK[W]
50 * and then stat to check that locked file
51 * still has name site.name_lock
52 * holder of this lock is "duct"
53 * (only) lockholder may remove the lockfile
54 * D site.name_flushing temporary feed file during flush (or crash)
55 * hardlink created by duct
57 * site.name_defer 431'd articles, still being written,
58 * created, written, used by duct
60 * site.name_backlog.<date>.<inum>
61 * 431'd articles, ready for innxmit or duct
62 * created (link/mv) by duct
63 * site.name_backlog<anything-else> (where <anything-else> does not
64 * contain '#' or '~') eg
65 * site.name_backlog.manual
66 * anything the sysadmin likes (eg, feed files
67 * from old feeds to be merged into this one)
68 * created (link/mv) by admin
69 * may be symlinks (in which case links
70 * may be written through, but only links
73 * It is safe to remove backlog files manually,
74 * if it's desired to throw away the backlog.
76 * Backlog files are also processed by innduct. We find the oldest
77 * backlog file which is at least a certain amount old, and feed it
78 * back into our processing. When every article in it has been read
79 * and processed, we unlink it and look for another backlog file.
81 * If we don't have a backlog file that we're reading, we close the
82 * defer file that we're writing and make it into a backlog file at
83 * the first convenient opportunity.
94 | | <----------------<---------------------------------'|
100 | F: innd writing, duct reading |
103 | | duct decides time to flush |
104 | | duct makes hardlink |
106 | V <------------------------'|
108 | F == D: innd writing, duct reading both exist |
111 | | <-----------<-------------<--'|
112 | | open D F ENOENT |
115 | V <---------------------. |
118 | D: innd writing, duct reading; or ENOENT | |
120 | | duct requests flush of feed | |
121 | | (others can too, harmlessly) | |
125 | D: innd flushing, duct; or ENOENT | |
127 | | inndcomm flush fails | |
128 | |`-------------------------->------------------' |
130 | | inndcomm reports no such site |
131 | |`---------------------------------------------------- | -.
133 | | innd finishes writing D, creates F | |
134 | | inndcomm reports flush successful | |
137 | Separated <----------------' |
138 | F: innd writing F!=D /
139 | D: duct reading; or ENOENT both exist /
141 | | duct gets to the end of D /
142 | | duct opens F too /
145 | F: innd writing, duct reading |
146 | D: duct finishing V
148 | | duct finishes processing D F: ENOENT
149 | V duct unlinks D D: duct reading
151 `--<--' | duct finishes
161 "duct reading" means innduct is reading the file but also
162 overwriting processed tokens.
166 * rune for printing diagrams:
168 perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct.c |a2ps -R -B -ops
173 /*============================== PROGRAM ==============================*/
175 #define _GNU_SOURCE 1
181 #include "inndcomm.h"
183 #include "inn/list.h"
184 #include "inn/innconf.h"
185 #include "inn/messages.h"
188 #include <sys/types.h>
189 #include <sys/wait.h>
190 #include <sys/stat.h>
191 #include <sys/socket.h>
210 #include <oop-read.h>
212 /*----- general definitions, probably best not changed -----*/
214 #define CONNCHILD_ESTATUS_STREAM 24
215 #define CONNCHILD_ESTATUS_NOSTREAM 25
217 #define INNDCOMMCHILD_ESTATUS_FAIL 26
218 #define INNDCOMMCHILD_ESTATUS_NONESUCH 27
220 #define MAX_LINE_FEEDFILE (NNTP_MSGID_MAXLEN + sizeof(TOKEN)*2 + 10)
221 #define MAX_CLI_COMMAND 1000
223 #define VA va_list al; va_start(al,fmt)
224 #define PRINTF(f,a) __attribute__((__format__(printf,f,a)))
225 #define NORET_PRINTF(f,a) __attribute__((__noreturn__,__format__(printf,f,a)))
226 #define NORET __attribute__((__noreturn__))
228 #define NEW(ptr) ((ptr)= zmmalloc(sizeof(*(ptr))))
229 #define NEW_DECL(type,ptr) type ptr = zmmalloc(sizeof(*(ptr)))
231 #define DUMPV(fmt,pfx,v) fprintf(f, " " #v "=" fmt, pfx v);
233 #define FOR_CONN(conn) \
234 for ((conn)=LIST_HEAD(conns); (conn); (conn)=LIST_NEXT((conn)))
236 /*----- doubly linked lists -----*/
238 #define ISNODE(T) struct node list_node
241 union { struct list li; T *for_type; } u; \
245 #define NODE(n) (assert((void*)&(n)->list_node == (n)), &(n)->list_node)
247 #define LIST_CHECKCANHAVENODE(l,n) \
248 ((void)((n) == ((l).u.for_type))) /* just for the type check */
250 #define LIST_ADDSOMEHOW(l,n,list_addsomehow) \
251 ( LIST_CHECKCANHAVENODE(l,n), \
252 list_addsomehow(&(l).u.li, NODE((n))), \
256 #define LIST_REMSOMEHOW(l,list_remsomehow) \
257 ( (typeof((l).u.for_type)) \
260 list_remsomehow(&(l).u.li) ) \
266 #define LIST_ADDHEAD(l,n) LIST_ADDSOMEHOW((l),(n),list_addhead)
267 #define LIST_ADDTAIL(l,n) LIST_ADDSOMEHOW((l),(n),list_addtail)
268 #define LIST_REMHEAD(l) LIST_REMSOMEHOW((l),list_remhead)
269 #define LIST_REMTAIL(l) LIST_REMSOMEHOW((l),list_remtail)
271 #define LIST_INIT(l) ((l).count=0, list_new(&(l).u.li))
272 #define LIST_HEAD(l) ((typeof((l).u.for_type))(list_head((struct list*)&(l))))
273 #define LIST_NEXT(n) ((typeof(n))list_succ(NODE((n))))
274 #define LIST_BACK(n) ((typeof(n))list_pred(NODE((n))))
276 #define LIST_REMOVE(l,n) \
277 ( LIST_CHECKCANHAVENODE(l,n), \
278 list_remove(NODE((n))), \
282 #define LIST_INSERT(l,n,pred) \
283 ( LIST_CHECKCANHAVENODE(l,n), \
284 LIST_CHECKCANHAVENODE(l,pred), \
285 list_insert((struct list*)&(l), NODE((n)), NODE((pred))), \
289 /*----- type predeclarations -----*/
291 typedef struct Conn Conn;
292 typedef struct Article Article;
293 typedef struct InputFile InputFile;
294 typedef struct XmitDetails XmitDetails;
295 typedef struct Filemon_Perfile Filemon_Perfile;
296 typedef enum StateMachineState StateMachineState;
297 typedef struct CliCommand CliCommand;
302 /*----- function predeclarations -----*/
304 static void conn_maybe_write(Conn *conn);
305 static void conn_make_some_xmits(Conn *conn);
306 static void *conn_write_some_xmits(Conn *conn);
308 static void xmit_free(XmitDetails *d);
310 #define SMS(newstate, periods, why) \
311 (statemc_setstate(sm_##newstate,(periods),#newstate,(why)))
312 static void statemc_setstate(StateMachineState newsms, int periods,
313 const char *forlog, const char *why);
315 static void statemc_start_flush(const char *why); /* Normal => Flushing */
316 static void spawn_inndcomm_flush(const char *why); /* Moved => Flushing */
317 static int trigger_flush_ok(const char *why /* 0 means timeout */);
318 /* => Flushing,FLUSHING, ret 1; or ret 0 */
320 static void article_done(Article *art, int whichcount);
322 static void check_assign_articles(void);
323 static void queue_check_input_done(void);
324 static void check_reading_pause_resume(InputFile *ipf);
326 static void statemc_check_flushing_done(void);
327 static void statemc_check_backlog_done(void);
329 static void postfork(void);
330 static void period(void);
332 static void open_defer(void);
333 static void close_defer(void);
334 static void search_backlog_file(void);
335 static void preterminate(void);
336 static void raise_default(int signo) NORET;
338 static void inputfile_reading_start(InputFile *ipf);
339 static void inputfile_reading_stop(InputFile *ipf);
340 static void inputfile_reading_pause(InputFile *ipf);
341 static void inputfile_reading_resume(InputFile *ipf);
342 /* pause and resume are idempotent, and no-op if not done _reading_start */
344 static void filemon_start(InputFile *ipf);
345 static void filemon_stop(InputFile *ipf);
346 static void tailing_make_readable(InputFile *ipf);
348 static void vconnfail(Conn *conn, const char *fmt, va_list al) PRINTF(2,0);
349 static void connfail(Conn *conn, const char *fmt, ...) PRINTF(2,3);
351 static const oop_rd_style peer_rd_style;
352 static oop_rd_call peer_rd_err, peer_rd_ok;
355 /*----- configuration options -----*/
356 /* when changing defaults, remember to update the manpage */
358 static const char *sitename, *remote_host;
359 static const char *feedfile, *path_run, *path_cli, *path_cli_dir;
360 static int quiet_multiple=0;
361 static int interactive=0, try_filemon=1;
362 static int try_stream=1;
364 static const char *inndconffile;
366 static int max_connections=10;
367 static int max_queue_per_conn=200;
368 static int target_max_feedfile_size=100000;
369 static int period_seconds=30;
370 static int filepoll_seconds=5;
371 static int max_queue_per_ipf=-1;
373 static int connection_setup_timeout=200;
374 static int inndcomm_flush_timeout=100;
376 static double nocheck_thresh= 95.0; /* converted from percentage by main */
377 static double nocheck_decay= 100; /* conv'd from articles to lambda by main */
379 /* all these are initialised to seconds, and converted to periods in main */
380 static int reconnect_delay_periods=1000;
381 static int flushfail_retry_periods=1000;
382 static int backlog_retry_minperiods=100;
383 static int backlog_spontrescan_periods=300;
384 static int spontaneous_flush_periods=100000;
385 static int max_separated_periods=2000;
386 static int need_activity_periods=1000;
387 static int lowvol_thresh=3;
388 static int lowvol_periods=1000;
390 static double max_bad_data_ratio= 1; /* conv'd from percentage by main */
391 static int max_bad_data_initial= 30;
392 /* in one corrupt 4096-byte block the number of newlines has
393 * mean 16 and standard deviation 3.99. 30 corresponds to z=+3.5 */
396 /*----- statistics -----*/
398 typedef enum { /* in queue in conn->sent */
399 art_Unchecked, /* not checked, not sent checking */
400 art_Wanted, /* checked, wanted sent body as requested */
401 art_Unsolicited, /* - sent body without check */
405 static const char *const artstate_names[]=
406 { "Unchecked", "Wanted", "Unsolicited", 0 };
408 #define RESULT_COUNTS(RCS,RCN) \
417 #define RCI_TRIPLE_FMT_BASE "%d (id=%d,bod=%d,nc=%d)"
418 #define RCI_TRIPLE_VALS_BASE(counts,x) \
419 counts[art_Unchecked] x \
420 + counts[art_Wanted] x \
421 + counts[art_Unsolicited] x, \
422 counts[art_Unchecked] x \
423 , counts[art_Wanted] x \
424 , counts[art_Unsolicited] x
427 #define RC_INDEX(x) RC_##x,
428 RESULT_COUNTS(RC_INDEX, RC_INDEX)
433 /*----- transmission buffers -----*/
449 /*----- core operational data structure types -----*/
452 /* This is also an instance of struct oop_readable */
453 struct oop_readable readable; /* first */
454 oop_readable_call *readable_callback;
455 void *readable_callback_user;
458 Filemon_Perfile *filemon;
460 oop_read *rd; /* non-0: reading; 0: constructing, or had EOF */
462 int skippinglong, paused, fake_readable;
465 long inprogress; /* includes queue.count and also articles in conns */
466 long autodefer; /* -1 means not doing autodefer */
468 int counts[art_MaxState][RCI_max];
469 int readcount_ok, readcount_blank, readcount_err, count_nooffer_missing;
484 #define SMS_LIST(X) \
492 enum StateMachineState {
493 #define SMS_DEF_ENUM(s) sm_##s,
494 SMS_LIST(SMS_DEF_ENUM)
497 static const char *sms_names[]= {
498 #define SMS_DEF_NAME(s) #s ,
499 SMS_LIST(SMS_DEF_NAME)
505 int fd; /* may be 0, meaning closed (during construction/destruction) */
506 oop_read *rd; /* likewise */
507 int oopwriting; /* since on_fd is not idempotent */
508 int max_queue, stream;
509 const char *quitting;
510 int since_activity; /* periods */
511 ArticleList waiting; /* not yet told peer */
512 ArticleList priority; /* peer says send it now */
513 ArticleList sent; /* offered/transmitted - in xmit or waiting reply */
514 struct iovec xmit[CONNIOVS];
515 XmitDetails xmitd[CONNIOVS];
520 /*----- general operational variables -----*/
522 /* main initialises */
523 static oop_source *loop;
524 static ConnList conns;
525 static char *path_lock, *path_flushing, *path_defer, *path_dump;
526 static char *globpat_backlog;
527 static pid_t self_pid;
528 static int *lowvol_perperiod;
529 static int lowvol_circptr;
530 static int lowvol_total; /* does not include current period */
532 /* statemc_init initialises */
533 static StateMachineState sms;
534 static int until_flush;
535 static InputFile *main_input_file, *flushing_input_file, *backlog_input_file;
538 /* initialisation to 0 is good */
539 static int until_connect, until_backlog_nextscan;
540 static double accept_proportion;
541 static int nocheck, nocheck_reported, in_child;
543 /* for simulation, debugging, etc. */
544 int simulate_flush= -1;
546 /*========== logging ==========*/
548 static void logcore(int sysloglevel, const char *fmt, ...) PRINTF(2,3);
549 static void logcore(int sysloglevel, const char *fmt, ...) {
551 if (interactive < 2) {
552 vsyslog(sysloglevel,fmt,al);
554 if (self_pid) fprintf(stderr,"[%lu] ",(unsigned long)self_pid);
555 vfprintf(stderr,fmt,al);
561 static void logv(int sysloglevel, const char *pfx, int errnoval,
562 const char *fmt, va_list al) PRINTF(5,0);
563 static void logv(int sysloglevel, const char *pfx, int errnoval,
564 const char *fmt, va_list al) {
565 char msgbuf[1024]; /* NB do not call mvasprintf here or you'll recurse */
566 vsnprintf(msgbuf,sizeof(msgbuf), fmt,al);
567 msgbuf[sizeof(msgbuf)-1]= 0;
569 if (sysloglevel >= LOG_ERR && (errnoval==EACCES || errnoval==EPERM))
570 sysloglevel= LOG_ERR; /* run by wrong user, probably */
572 logcore(sysloglevel, "<%s>%s: %s%s%s",
573 sitename, pfx, msgbuf,
574 errnoval>=0 ? ": " : "",
575 errnoval>=0 ? strerror(errnoval) : "");
578 #define DIEWRAP(fn, pfx, sysloglevel, err, estatus) \
579 static void fn(const char *fmt, ...) NORET_PRINTF(1,2); \
580 static void fn(const char *fmt, ...) { \
583 logv(sysloglevel, pfx, err, fmt, al); \
587 #define LOGWRAP(fn, pfx, sysloglevel, err) \
588 static void fn(const char *fmt, ...) PRINTF(1,2); \
589 static void fn(const char *fmt, ...) { \
591 logv(sysloglevel, pfx, err, fmt, al); \
595 #define INNLOGWRAP_DECLARE(fn, pfx, sysloglevel) \
596 static void duct_log_##fn(int errval, const char *fmt, va_list al, int l) { \
597 logv(sysloglevel, pfx, errval ? errval : -1, fmt, al); \
599 #define INNLOGWRAP_CALL(fn, pfx, sysloglevel) \
600 message_handlers_##fn(1, duct_log_##fn);
603 static int innduct_fatal_cleanup(void) { return 12; } /* used for libinn die */
605 /* We want to extend the set of logging functions from inn, and we
606 * want to prepend the site name to all our messages. */
608 DIEWRAP(syscrash, " critical", LOG_CRIT, errno, 16);
609 DIEWRAP(crash, " critical", LOG_CRIT, -1, 16);
611 DIEWRAP(sysfatal, " fatal", LOG_ERR, errno, 12);
612 DIEWRAP(fatal, " fatal", LOG_ERR, -1, 12);
614 #define INNLOGWRAPS(INNLOGWRAP) \
615 INNLOGWRAP(warn, " warning", LOG_WARNING) \
616 INNLOGWRAP(notice, " notice", LOG_NOTICE)
617 INNLOGWRAPS(INNLOGWRAP_DECLARE)
619 LOGWRAP(info, " info", LOG_INFO, -1)
620 LOGWRAP(dbg, " debug", LOG_DEBUG, -1)
623 /*========== utility functions etc. ==========*/
625 /* error trapping wrappers are called mfoo rather than the more
626 * conventional xfoo because we don't want to clash with the existing
627 * xfoo functions in INN libs which use different error handlers
630 static char *mvasprintf(const char *fmt, va_list al) PRINTF(1,0);
631 static char *mvasprintf(const char *fmt, va_list al) {
633 int rc= vasprintf(&str,fmt,al);
634 if (rc<0) sysfatal("vasprintf(\"%s\",...) failed", fmt);
637 static char *masprintf(const char *fmt, ...) PRINTF(1,2);
638 static char *masprintf(const char *fmt, ...) {
640 char *str= mvasprintf(fmt,al);
645 static int close_perhaps(int *fd) {
646 if (*fd <= 0) return 0;
651 static void mclose(int fd, const char *what, const char *what2) {
653 if (r) syscrash("close %s%s",what,what2?what2:"");
655 static void mclose_perhaps(int *fd, const char *what, const char *what2) {
656 if (*fd <= 0) return;
657 mclose(*fd,what,what2);
661 static pid_t mfork(const char *what) {
665 if (child==-1) sysfatal("cannot fork for %s",what);
666 dbg("forked %s %ld", what, (unsigned long)child);
667 if (!child) postfork();
671 static void on_fd_read_except(int fd, oop_call_fd callback) {
672 loop->on_fd(loop, fd, OOP_READ, callback, 0);
673 loop->on_fd(loop, fd, OOP_EXCEPTION, callback, 0);
675 static void cancel_fd_read_except(int fd) {
676 loop->cancel_fd(loop, fd, OOP_READ);
677 loop->cancel_fd(loop, fd, OOP_EXCEPTION);
680 static void report_child_status(const char *what, int status) {
681 if (WIFEXITED(status)) {
682 int es= WEXITSTATUS(status);
684 warn("%s: child died with error exit status %d", what, es);
685 } else if (WIFSIGNALED(status)) {
686 int sig= WTERMSIG(status);
687 const char *sigstr= strsignal(sig);
688 const char *coredump= WCOREDUMP(status) ? " (core dumped)" : "";
690 warn("%s: child died due to fatal signal %s%s", what, sigstr, coredump);
692 warn("%s: child died due to unknown fatal signal %d%s",
693 what, sig, coredump);
695 warn("%s: child died with unknown wait status %d", what,status);
699 static int mwaitpid(pid_t *pid, const char *what) {
702 int r= kill(*pid, SIGKILL);
703 if (r) syscrash("cannot kill %s child", what);
705 pid_t got= waitpid(*pid, &status, 0);
706 if (got==-1) syscrash("cannot reap %s child", what);
707 if (got==0) crash("cannot reap %s child", what);
714 static void *mmalloc(size_t sz) {
716 if (!p) sysfatal("unable to malloc %lu bytes",(unsigned long)sz);
720 static void *zmmalloc(size_t sz) {
721 void *p= mmalloc(sz);
726 static void munlink(const char *path, const char *what) {
728 if (r) syscrash("can't unlink %s %s", path, what);
731 static time_t mtime(void) {
733 if (now==-1) syscrash("time(2) failed");
737 static void msigaction(int signo, const struct sigaction *sa) {
738 int r= sigaction(signo,sa,0);
739 if (r) syscrash("sigaction failed for \"%s\"", strsignal(signo));
742 static void msigsetdefault(int signo) {
744 memset(&sa,0,sizeof(sa));
745 sa.sa_handler= SIG_DFL;
746 msigaction(signo,&sa);
749 static void mgettimeofday(struct timeval *tv_r) {
750 int r= gettimeofday(tv_r,0);
751 if (r) syscrash("gettimeofday(2) failed");
754 static void msetnonblock(int fd, int nonblocking) {
755 int errnoval= oop_fd_nonblock(fd, nonblocking);
756 if (errnoval) { errno= errnoval; syscrash("setnonblocking"); }
759 static void check_isreg(const struct stat *stab, const char *path,
761 if (!S_ISREG(stab->st_mode))
762 crash("%s %s not a plain file (mode 0%lo)",
763 what, path, (unsigned long)stab->st_mode);
766 static void mfstat(int fd, struct stat *stab_r, const char *what) {
767 int r= fstat(fd, stab_r);
768 if (r) syscrash("could not fstat %s", what);
771 static void mfstat_isreg(int fd, struct stat *stab_r,
772 const char *path, const char *what) {
773 mfstat(fd, stab_r, what);
774 check_isreg(stab_r, path, what);
777 static void mlstat_isreg(const char *path, struct stat *stab,
778 int *enoent_r /* 0 means ENOENT is fatal */,
780 int r= lstat(path, stab);
782 if (errno==ENOENT && enoent_r) { *enoent_r=1; return; }
783 syscrash("could not lstat %s %s", what, path);
785 if (enoent_r) *enoent_r= 0;
786 check_isreg(stab, path, what);
789 static int samefile(const struct stat *a, const struct stat *b) {
790 assert(S_ISREG(a->st_mode));
791 assert(S_ISREG(b->st_mode));
792 return (a->st_ino == b->st_ino &&
793 a->st_dev == b->st_dev);
796 static char *sanitise(const char *input, int len) {
797 static char sanibuf[100]; /* returns pointer to this buffer! */
799 const char *p= input;
800 const char *endp= len>=0 ? input+len : 0;
804 if (q > sanibuf+sizeof(sanibuf)-8) { strcpy(q,"'.."); break; }
805 int c= (!endp || p<endp) ? *p++ : 0;
806 if (!c) { *q++= '\''; *q=0; break; }
807 if (c>=' ' && c<=126 && c!='\\') { *q++= c; continue; }
808 sprintf(q,"\\x%02x",c);
814 static int isewouldblock(int errnoval) {
815 return errnoval==EWOULDBLOCK || errnoval==EAGAIN;
818 /*========== command and control (CLI) connections ==========*/
820 static int cli_master;
822 typedef struct CliConn CliConn;
824 void (*destroy)(CliConn*);
830 struct sockaddr_un un;
835 static const oop_rd_style cli_rd_style= {
836 OOP_RD_DELIM_STRIP, '\n',
838 OOP_RD_SHORTREC_FORBID
841 static void cli_destroy(CliConn *cc) {
845 static void cli_checkouterr(CliConn *cc /* may destroy*/) {
846 if (ferror(cc->out) | fflush(cc->out)) {
847 info("CTRL%d write error %s", cc->fd, strerror(errno));
852 static void cli_prompt(CliConn *cc /* may destroy*/) {
853 fprintf(cc->out, "%s| ", sitename);
859 void (*f)(CliConn *cc, const CliCommand *ccmd,
860 const char *arg, size_t argsz);
865 static const CliCommand cli_commands[];
868 static void ccmd_##wh(CliConn *cc, const CliCommand *c, \
869 const char *arg, size_t argsz)
872 fputs("commands:\n", cc->out);
873 const CliCommand *ccmd;
874 for (ccmd=cli_commands; ccmd->cmd; ccmd++)
875 fprintf(cc->out, " %s\n", ccmd->cmd);
876 fputs("NB: permissible arguments are not shown above."
877 " Not all commands listed are safe. See innduct(8).\n", cc->out);
881 int ok= trigger_flush_ok("manual request");
882 if (!ok) fprintf(cc->out,"already flushing (state is %s)\n", sms_names[sms]);
887 notice("terminating (CTRL%d)",cc->fd);
888 raise_default(SIGTERM);
894 /* messing with our head: */
895 CCMD(period) { period(); }
896 CCMD(setintarg) { *(int*)c->xdata= atoi(arg); }
897 CCMD(setint) { *(int*)c->xdata= c->xval; }
898 CCMD(setint_period) { *(int*)c->xdata= c->xval; period(); }
900 static const CliCommand cli_commands[]= {
902 { "flush", ccmd_flush },
903 { "stop", ccmd_stop },
904 { "dump q", ccmd_dump, 0,0 },
905 { "dump a", ccmd_dump, 0,1 },
907 { "p", ccmd_period },
909 #define POKES(cmd,func) \
910 { cmd "flush", func, &until_flush, 1 }, \
911 { cmd "conn", func, &until_connect, 0 }, \
912 { cmd "blscan", func, &until_backlog_nextscan, 0 },
913 POKES("next ", ccmd_setint)
914 POKES("prod ", ccmd_setint_period)
916 { "pretend flush", ccmd_setintarg, &simulate_flush },
917 { "wedge blscan", ccmd_setint, &until_backlog_nextscan, -1 },
921 static void *cli_rd_ok(oop_source *lp, oop_read *oread, oop_rd_event ev,
922 const char *errmsg, int errnoval,
923 const char *data, size_t recsz, void *cc_v) {
927 info("CTRL%d closed", cc->fd);
932 if (recsz == 0) goto prompt;
934 const CliCommand *ccmd;
935 for (ccmd=cli_commands; ccmd->cmd; ccmd++) {
936 int l= strlen(ccmd->cmd);
937 if (recsz < l) continue;
938 if (recsz > l && data[l] != ' ') continue;
939 if (memcmp(data, ccmd->cmd, l)) continue;
941 int argl= (int)recsz - (l+1);
942 ccmd->f(cc, ccmd, argl>=0 ? data+l+1 : 0, argl);
946 fputs("unknown command; h for help\n", cc->out);
953 static void *cli_rd_err(oop_source *lp, oop_read *oread, oop_rd_event ev,
954 const char *errmsg, int errnoval,
955 const char *data, size_t recsz, void *cc_v) {
958 info("CTRL%d read error %s", cc->fd, errmsg);
963 static int cli_conn_startup(CliConn *cc /* may destroy*/,
965 cc->rd= oop_rd_new_fd(loop, cc->fd, 0,0);
966 if (!cc->rd) { warn("oop_rd_new_fd cli failed"); return -1; }
968 int er= oop_rd_read(cc->rd, &cli_rd_style, MAX_CLI_COMMAND,
971 if (er) { errno= er; syswarn("oop_rd_read cli failed"); return -1; }
973 info("CTRL%d %s ready", cc->fd, how);
978 static void cli_stdio_destroy(CliConn *cc) {
980 oop_rd_cancel(cc->rd);
981 errno= oop_rd_delete_tidy(cc->rd);
982 if (errno) syswarn("oop_rd_delete tidy failed (no-nonblock stdin?)");
987 static void cli_stdio(void) {
988 NEW_DECL(CliConn *,cc);
989 cc->destroy= cli_stdio_destroy;
993 int r= cli_conn_startup(cc,"stdio");
994 if (r) cc->destroy(cc);
997 static void cli_accepted_destroy(CliConn *cc) {
999 oop_rd_cancel(cc->rd);
1000 oop_rd_delete_kill(cc->rd);
1002 if (cc->out) { fclose(cc->out); cc->fd=0; }
1003 close_perhaps(&cc->fd);
1007 static void *cli_master_readable(oop_source *lp, int master,
1008 oop_event ev, void *u) {
1009 NEW_DECL(CliConn *,cc);
1010 cc->destroy= cli_accepted_destroy;
1012 cc->salen= sizeof(cc->sa);
1013 cc->fd= accept(master, &cc->sa.sa, &cc->salen);
1014 if (cc->fd<0) { syswarn("error accepting cli connection"); goto x; }
1016 cc->out= fdopen(cc->fd, "w");
1017 if (!cc->out) { syswarn("error fdopening accepted cli connection"); goto x; }
1019 int r= cli_conn_startup(cc, "accepted");
1022 return OOP_CONTINUE;
1026 return OOP_CONTINUE;
1029 #define NOCLI(...) do{ \
1030 syswarn("no cli listener, because failed to " __VA_ARGS__); \
1034 static void cli_init(void) {
1037 struct sockaddr_un un;
1040 memset(&sa,0,sizeof(sa));
1041 int maxlen= sizeof(sa.un.sun_path);
1044 info("control command line disabled");
1048 int pathlen= strlen(path_cli);
1049 if (pathlen > maxlen) {
1050 warn("no cli listener, because cli socket path %s too long (%d>%d)",
1051 path_cli, pathlen, maxlen);
1056 int r= mkdir(path_cli_dir, 0700);
1057 if (r && errno!=EEXIST)
1058 NOCLI("create cli socket directory %s", path_cli_dir);
1061 int r= unlink(path_cli);
1062 if (r && errno!=ENOENT)
1063 NOCLI("remove old cli socket %s", path_cli);
1065 cli_master= socket(PF_UNIX, SOCK_STREAM, 0);
1066 if (cli_master<0) NOCLI("create new cli master socket");
1068 int sl= pathlen + offsetof(struct sockaddr_un, sun_path);
1069 sa.un.sun_family= AF_UNIX;
1070 memcpy(sa.un.sun_path, path_cli, pathlen);
1072 r= bind(cli_master, &sa.sa, sl);
1073 if (r) NOCLI("bind to cli socket path %s", sa.un.sun_path);
1075 r= listen(cli_master, 5);
1076 if (r) NOCLI("listen to cli master socket");
1078 msetnonblock(cli_master, 1);
1080 loop->on_fd(loop, cli_master, OOP_READ, cli_master_readable, 0);
1081 info("cli ready, listening on %s", path_cli);
1086 mclose_perhaps(&cli_master, "cli master",0);
1090 /*========== management of connections ==========*/
1092 static void reconnect_blocking_event(void) {
1093 until_connect= reconnect_delay_periods;
1096 static void conn_closefd(Conn *conn, const char *msgprefix) {
1097 int r= close_perhaps(&conn->fd);
1098 if (r) info("C%d %serror closing socket: %s",
1099 conn->fd, msgprefix, strerror(errno));
1102 static int conn_busy(Conn *conn) {
1104 conn->waiting.count ||
1105 conn->priority.count ||
1110 static void conn_dispose(Conn *conn) {
1113 oop_rd_cancel(conn->rd);
1114 oop_rd_delete_kill(conn->rd);
1118 loop->cancel_fd(loop, conn->fd, OOP_WRITE);
1119 loop->cancel_fd(loop, conn->fd, OOP_EXCEPTION);
1121 conn_closefd(conn,"");
1125 static void *conn_exception(oop_source *lp, int fd,
1126 oop_event ev, void *conn_v) {
1129 assert(fd == conn->fd);
1130 assert(ev == OOP_EXCEPTION);
1131 int r= read(conn->fd, &ch, 1);
1132 if (r<0) connfail(conn,"read failed: %s",strerror(errno));
1133 else connfail(conn,"exceptional condition on socket (peer sent urgent"
1134 " data? read(,&ch,1)=%d,ch='\\x%02x')",r,ch);
1135 return OOP_CONTINUE;
1138 static void vconnfail(Conn *conn, const char *fmt, va_list al) {
1139 int requeue[art_MaxState];
1140 memset(requeue,0,sizeof(requeue));
1144 while ((art= LIST_REMHEAD(conn->priority)))
1145 LIST_ADDTAIL(art->ipf->queue, art);
1147 while ((art= LIST_REMHEAD(conn->waiting)))
1148 LIST_ADDTAIL(art->ipf->queue, art);
1150 while ((art= LIST_REMHEAD(conn->sent))) {
1151 requeue[art->state]++;
1152 if (art->state==art_Unsolicited) art->state= art_Unchecked;
1153 LIST_ADDTAIL(art->ipf->queue,art);
1154 check_reading_pause_resume(art->ipf);
1159 for (i=0, d=conn->xmitd; i<conn->xmitu; i++, d++)
1162 LIST_REMOVE(conns,conn);
1164 char *m= mvasprintf(fmt,al);
1165 warn("C%d (now %d) connection failed requeueing " RCI_TRIPLE_FMT_BASE ": %s",
1166 conn->fd, conns.count, RCI_TRIPLE_VALS_BASE(requeue, /*nothing*/), m);
1169 reconnect_blocking_event();
1171 check_assign_articles();
1174 static void connfail(Conn *conn, const char *fmt, ...) {
1177 vconnfail(conn,fmt,al);
1181 static void conn_idle_close(Conn *conn, const char *why) {
1182 static const char quitcmd[]= "QUIT\r\n";
1183 int todo= sizeof(quitcmd)-1;
1184 const char *p= quitcmd;
1186 int r= write(conn->fd, p, todo);
1188 if (isewouldblock(errno))
1189 connfail(conn, "blocked writing QUIT to idle connection");
1191 connfail(conn, "failed to write QUIT to idle connection: %s",
1198 conn->quitting= why;
1199 conn->since_activity= 0;
1200 dbg("C%d is idle (%s), quitting", conn->fd, why);
1207 * For our last connection, we also shut it down if we have had
1208 * less than K in the last L
1210 static void check_idle_conns(void) {
1213 int volthisperiod= lowvol_perperiod[lowvol_circptr];
1215 lowvol_circptr %= lowvol_periods;
1216 lowvol_total += volthisperiod;
1217 lowvol_total -= lowvol_perperiod[lowvol_circptr];
1218 lowvol_perperiod[lowvol_circptr]= 0;
1221 conn->since_activity++;
1225 if (conn->since_activity <= need_activity_periods) continue;
1227 /* We need to shut this down */
1229 connfail(conn,"timed out waiting for response to QUIT (%s)",
1231 else if (conn->sent.count)
1232 connfail(conn,"timed out waiting for responses");
1233 else if (conn->waiting.count || conn->priority.count)
1234 connfail(conn,"BUG IN INNDUCT conn has queue but nothing sent");
1235 else if (conn->xmitu)
1236 connfail(conn,"peer has been sending responses"
1237 " before receiving our commands!");
1239 conn_idle_close(conn, "no activity");
1244 conn= LIST_HEAD(conns);
1245 if (!volthisperiod &&
1247 lowvol_total < lowvol_thresh &&
1249 conn_idle_close(conn, "low volume");
1252 /*---------- making new connections ----------*/
1254 static pid_t connecting_child;
1255 static int connecting_fdpass_sock;
1257 static void connect_attempt_discard(void) {
1258 if (connecting_child) {
1259 int status= mwaitpid(&connecting_child, "connect");
1260 if (!(WIFEXITED(status) ||
1261 (WIFSIGNALED(status) && WTERMSIG(status) == SIGKILL)))
1262 report_child_status("connect", status);
1264 if (connecting_fdpass_sock) {
1265 cancel_fd_read_except(connecting_fdpass_sock);
1266 mclose_perhaps(&connecting_fdpass_sock, "connecting fdpass socket",0);
1270 #define PREP_DECL_MSG_CMSG(msg) \
1272 struct iovec msgiov; \
1273 msgiov.iov_base= &msgbyte; \
1274 msgiov.iov_len= 1; \
1275 struct msghdr msg; \
1276 memset(&msg,0,sizeof(msg)); \
1277 char msg##cbuf[CMSG_SPACE(sizeof(int))]; \
1278 msg.msg_iov= &msgiov; \
1279 msg.msg_iovlen= 1; \
1280 msg.msg_control= msg##cbuf; \
1281 msg.msg_controllen= sizeof(msg##cbuf);
1283 static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) {
1286 assert(fd == connecting_fdpass_sock);
1288 PREP_DECL_MSG_CMSG(msg);
1290 ssize_t rs= recvmsg(fd, &msg, 0);
1292 if (isewouldblock(errno)) return OOP_CONTINUE;
1293 syswarn("failed to read socket from connecting child");
1298 LIST_INIT(conn->waiting);
1299 LIST_INIT(conn->priority);
1300 LIST_INIT(conn->sent);
1302 struct cmsghdr *h= 0;
1303 if (rs >= 0) h= CMSG_FIRSTHDR(&msg);
1305 int status= mwaitpid(&connecting_child, "connect child (broken)");
1307 if (WIFEXITED(status)) {
1308 if (WEXITSTATUS(status) != 0 &&
1309 WEXITSTATUS(status) != CONNCHILD_ESTATUS_STREAM &&
1310 WEXITSTATUS(status) != CONNCHILD_ESTATUS_NOSTREAM)
1311 /* child already reported the problem */;
1313 if (e == OOP_EXCEPTION)
1314 warn("connect: connection child exited code %d but"
1315 " unexpected exception on fdpass socket",
1316 WEXITSTATUS(status));
1318 warn("connect: connection child exited code %d but"
1320 WEXITSTATUS(status), (int)rs);
1322 } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) {
1323 warn("connect: connection attempt timed out");
1325 report_child_status("connect", status);
1330 #define CHK(field, val) \
1331 if (h->cmsg_##field != val) { \
1332 crash("connect: child sent cmsg with cmsg_" #field "=%d, expected %d", \
1333 h->cmsg_##field, val); \
1336 CHK(level, SOL_SOCKET);
1337 CHK(type, SCM_RIGHTS);
1338 CHK(len, CMSG_LEN(sizeof(conn->fd)));
1341 if (CMSG_NXTHDR(&msg,h)) crash("connect: child sent many cmsgs");
1343 memcpy(&conn->fd, CMSG_DATA(h), sizeof(conn->fd));
1346 pid_t got= waitpid(connecting_child, &status, 0);
1347 if (got==-1) syscrash("connect: real wait for child");
1348 assert(got == connecting_child);
1349 connecting_child= 0;
1351 if (!WIFEXITED(status)) { report_child_status("connect",status); goto x; }
1352 int es= WEXITSTATUS(status);
1354 case CONNCHILD_ESTATUS_STREAM: conn->stream= 1; break;
1355 case CONNCHILD_ESTATUS_NOSTREAM: conn->stream= 0; break;
1357 fatal("connect: child gave unexpected exit status %d", es);
1361 conn->max_queue= conn->stream ? max_queue_per_conn : 1;
1363 loop->on_fd(loop, conn->fd, OOP_EXCEPTION, conn_exception, conn);
1364 conn->rd= oop_rd_new_fd(loop,conn->fd, 0, 0); /* sets nonblocking, too */
1365 if (!conn->fd) crash("oop_rd_new_fd conn failed (fd=%d)",conn->fd);
1366 int r= oop_rd_read(conn->rd, &peer_rd_style, NNTP_STRLEN,
1368 &peer_rd_err, conn);
1369 if (r) syscrash("oop_rd_read for peer (fd=%d)",conn->fd);
1371 LIST_ADDHEAD(conns, conn);
1372 notice("C%d (now %d) connected %s",
1373 conn->fd, conns.count, conn->stream ? "streaming" : "plain");
1375 connect_attempt_discard();
1376 check_assign_articles();
1377 return OOP_CONTINUE;
1381 connect_attempt_discard();
1382 reconnect_blocking_event();
1383 return OOP_CONTINUE;
1386 static int allow_connect_start(void) {
1387 return conns.count < max_connections
1388 && !connecting_child
1392 static void connect_start(void) {
1393 assert(!connecting_child);
1394 assert(!connecting_fdpass_sock);
1396 info("starting connection attempt");
1397 int ok_until_connect= until_connect;
1398 reconnect_blocking_event();
1401 int r= socketpair(AF_UNIX, SOCK_STREAM, 0, socks);
1402 if (r) { syswarn("connect: cannot create socketpair for child"); return; }
1404 connecting_child= mfork("connection");
1406 if (!connecting_child) {
1407 FILE *cn_from, *cn_to;
1408 char buf[NNTP_STRLEN+100];
1409 int exitstatus= CONNCHILD_ESTATUS_NOSTREAM;
1411 mclose(socks[0], "(in child) parent's connection fdpass socket",0);
1413 alarm(connection_setup_timeout);
1414 if (NNTPconnect((char*)remote_host, port, &cn_from, &cn_to, buf) < 0) {
1418 unsigned char c= buf[l-1];
1419 if (!isspace(c)) break;
1420 if (c=='\n' || c=='\r') stripped=1;
1424 sysfatal("connect: connection attempt failed");
1427 fatal("connect: %s: %s", stripped ? "rejected" : "failed",
1431 if (NNTPsendpassword((char*)remote_host, cn_from, cn_to) < 0)
1432 sysfatal("connect: authentication failed");
1434 if (fputs("MODE STREAM\r\n", cn_to)==EOF ||
1436 sysfatal("connect: could not send MODE STREAM");
1437 buf[sizeof(buf)-1]= 0;
1438 if (!fgets(buf, sizeof(buf)-1, cn_from)) {
1439 if (ferror(cn_from))
1440 sysfatal("connect: could not read response to MODE STREAM");
1442 fatal("connect: connection close in response to MODE STREAM");
1447 fatal("connect: response to MODE STREAM is too long: %.100s...",
1449 l--; if (l>0 && buf[l-1]=='\r') l--;
1452 int rcode= strtoul(buf,&ep,10);
1454 fatal("connect: bad response to MODE STREAM: %.50s", sanitise(buf,-1));
1458 exitstatus= CONNCHILD_ESTATUS_STREAM;
1464 warn("connect: unexpected response to MODE STREAM: %.50s",
1470 int fd= fileno(cn_from);
1472 PREP_DECL_MSG_CMSG(msg);
1473 struct cmsghdr *cmsg= CMSG_FIRSTHDR(&msg);
1474 cmsg->cmsg_level= SOL_SOCKET;
1475 cmsg->cmsg_type= SCM_RIGHTS;
1476 cmsg->cmsg_len= CMSG_LEN(sizeof(fd));
1477 memcpy(CMSG_DATA(cmsg), &fd, sizeof(fd));
1479 msg.msg_controllen= cmsg->cmsg_len;
1480 r= sendmsg(socks[1], &msg, 0);
1481 if (r<0) syscrash("sendmsg failed for new connection");
1482 if (r!=1) crash("sendmsg for new connection gave wrong result %d",r);
1487 mclose(socks[1], "connecting fdpass child's socket",0);
1488 connecting_fdpass_sock= socks[0];
1489 msetnonblock(connecting_fdpass_sock, 1);
1490 on_fd_read_except(connecting_fdpass_sock, connchild_event);
1493 until_connect= ok_until_connect;
1496 /*---------- assigning articles to conns, and transmitting ----------*/
1498 static Article *dequeue_from(int peek, InputFile *ipf) {
1500 if (peek) return LIST_HEAD(ipf->queue);
1502 Article *art= LIST_REMHEAD(ipf->queue);
1504 check_reading_pause_resume(ipf);
1508 static Article *dequeue(int peek) {
1510 art= dequeue_from(peek, flushing_input_file); if (art) return art;
1511 art= dequeue_from(peek, backlog_input_file); if (art) return art;
1512 art= dequeue_from(peek, main_input_file); if (art) return art;
1516 static void check_assign_articles(void) {
1522 int spare=0, inqueue=0;
1524 /* Find a connection to offer this article. We prefer a busy
1525 * connection to an idle one, provided it's not full. We take the
1526 * first (oldest) and since that's stable, it will mean we fill up
1527 * connections in order. That way if we have too many
1528 * connections, the spare ones will go away eventually.
1531 if (walk->quitting) continue;
1532 inqueue= walk->sent.count + walk->priority.count
1533 + walk->waiting.count;
1534 spare= walk->max_queue - inqueue;
1535 assert(inqueue <= max_queue_per_conn);
1537 if (inqueue==0) /*idle*/ { if (!use) use= walk; }
1538 else if (spare>0) /*working*/ { use= walk; break; }
1541 if (!inqueue) use->since_activity= 0; /* reset idle counter */
1543 Article *art= dequeue(0);
1545 LIST_ADDTAIL(use->waiting, art);
1546 lowvol_perperiod[lowvol_circptr]++;
1549 conn_maybe_write(use);
1550 } else if (allow_connect_start()) {
1559 static void *conn_writeable(oop_source *l, int fd, oop_event ev, void *u) {
1560 conn_maybe_write(u);
1561 return OOP_CONTINUE;
1564 static void conn_maybe_write(Conn *conn) {
1566 conn_make_some_xmits(conn);
1568 loop->cancel_fd(loop, conn->fd, OOP_WRITE);
1569 conn->oopwriting= 0;
1573 void *rp= conn_write_some_xmits(conn);
1574 if (rp==OOP_CONTINUE) {
1575 if (!conn->oopwriting) {
1576 loop->on_fd(loop, conn->fd, OOP_WRITE, conn_writeable, conn);
1577 conn->oopwriting= 1;
1580 } else if (rp==OOP_HALT) {
1583 /* transmitted everything */
1590 /*---------- expiry, flow control and deferral ----------*/
1593 * flow control notes
1594 * to ensure articles go away eventually
1595 * separate queue for each input file
1597 * every period, check head of backlog queue for expiry with SMretrieve
1598 * if too old: discard, and check next article
1599 * also check every backlog article as we read it
1601 * after too long in SEPARATED/DROPPING ie Separated/Finishing/Dropping
1602 * one-off: eat queued articles from flushing and write them to defer
1603 * one-off: connfail all connections which have any articles from flushing
1604 * newly read articles from flushing go straight to defer
1605 * this should take care of it and get us out of this state
1606 * to avoid filling up ram needlessly
1608 * limit number of queued articles for each ipf
1609 * pause/resume inputfile tailing
1612 static void check_reading_pause_resume(InputFile *ipf) {
1613 if (ipf->queue.count >= max_queue_per_ipf)
1614 inputfile_reading_pause(ipf);
1616 inputfile_reading_resume(ipf);
1619 static void article_defer(Article *art /* not on a queue */, int whichcount) {
1621 if (fprintf(defer, "%s %s\n", TokenToText(art->token), art->messageid) <0
1623 sysfatal("write to defer file %s",path_defer);
1624 article_done(art, whichcount);
1627 static int article_check_expired(Article *art /* must be queued, not conn */) {
1628 ARTHANDLE *artdata= SMretrieve(art->token, RETR_STAT);
1629 if (artdata) { SMfreearticle(artdata); return 0; }
1631 LIST_REMOVE(art->ipf->queue, art);
1633 art->ipf->count_nooffer_missing++;
1634 article_done(art,-1);
1638 static void inputfile_queue_check_expired(InputFile *ipf) {
1642 Article *art= LIST_HEAD(ipf->queue);
1643 int exp= article_check_expired(art);
1646 check_reading_pause_resume(ipf);
1649 static void article_autodefer(InputFile *ipf, Article *art) {
1651 article_defer(art,-1);
1654 static int has_article_in(const ArticleList *al, InputFile *ipf) {
1656 for (art=LIST_HEAD(*al); art; art=LIST_NEXT(art))
1657 if (art->ipf == ipf) return 1;
1661 static void autodefer_input_file_articles(InputFile *ipf) {
1663 while ((art= LIST_REMHEAD(ipf->queue)))
1664 article_autodefer(ipf, art);
1667 static void autodefer_input_file(InputFile *ipf) {
1668 static const char *const abandon= "stuck";
1671 autodefer_input_file_articles(ipf);
1673 if (ipf->inprogress) {
1676 if (has_article_in(&walk->waiting, ipf) ||
1677 has_article_in(&walk->priority, ipf) ||
1678 has_article_in(&walk->sent, ipf))
1679 walk->quitting= abandon;
1681 while (ipf->inprogress) {
1683 if (walk->quitting == abandon) goto found;
1684 abort(); /* where are they ?? */
1687 connfail(walk, "connection is stuck or crawling,"
1688 " and we need to finish flush");
1689 autodefer_input_file_articles(ipf);
1693 check_reading_pause_resume(ipf);
1696 /*========== article transmission ==========*/
1698 static XmitDetails *xmit_core(Conn *conn, const char *data, int len,
1699 XmitKind kind) { /* caller must then fill in details */
1700 struct iovec *v= &conn->xmit[conn->xmitu];
1701 XmitDetails *d= &conn->xmitd[conn->xmitu++];
1702 v->iov_base= (char*)data;
1708 static void xmit_noalloc(Conn *conn, const char *data, int len) {
1709 xmit_core(conn,data,len, xk_Const);
1711 #define XMIT_LITERAL(lit) (xmit_noalloc(conn, (lit), sizeof(lit)-1))
1713 static void xmit_artbody(Conn *conn, ARTHANDLE *ah /* consumed */) {
1714 XmitDetails *d= xmit_core(conn, ah->data, ah->len, xk_Artdata);
1718 static void xmit_free(XmitDetails *d) {
1720 case xk_Artdata: SMfreearticle(d->info.sm_art); break;
1721 case xk_Const: break;
1726 static void *conn_write_some_xmits(Conn *conn) {
1728 * 0: nothing more to write, no need to call us again
1729 * OOP_CONTINUE: more to write but fd not writeable
1730 * OOP_HALT: disaster, have destroyed conn
1733 int count= conn->xmitu;
1734 if (!count) return 0;
1736 if (count > IOV_MAX) count= IOV_MAX;
1737 ssize_t rs= writev(conn->fd, conn->xmit, count);
1739 if (isewouldblock(errno)) return OOP_CONTINUE;
1740 connfail(conn, "write failed: %s", strerror(errno));
1746 for (done=0; rs; ) {
1747 assert(done<conn->xmitu);
1748 struct iovec *vp= &conn->xmit[done];
1749 XmitDetails *dp= &conn->xmitd[done];
1750 if (rs >= vp->iov_len) {
1752 xmit_free(dp); /* vp->iov_len -= vp->iov_len, etc. */
1755 vp->iov_base= (char*)vp->iov_base + rs;
1757 break; /* rs -= rs */
1760 int newu= conn->xmitu - done;
1761 memmove(conn->xmit, conn->xmit + done, newu * sizeof(*conn->xmit));
1762 memmove(conn->xmitd, conn->xmitd + done, newu * sizeof(*conn->xmitd));
1767 static void conn_make_some_xmits(Conn *conn) {
1769 if (conn->xmitu+5 > CONNIOVS)
1772 Article *art= LIST_REMHEAD(conn->priority);
1773 if (!art) art= LIST_REMHEAD(conn->waiting);
1776 if (art->state >= art_Wanted || (conn->stream && nocheck)) {
1777 /* actually send it */
1779 ARTHANDLE *artdata= SMretrieve(art->token, RETR_ALL);
1782 art->state == art_Unchecked ? art_Unsolicited :
1783 art->state == art_Wanted ? art_Wanted :
1786 if (!artdata) art->missing= 1;
1787 art->ipf->counts[art->state][ artdata ? RC_sent : RC_missing ]++;
1791 XMIT_LITERAL("TAKETHIS ");
1792 xmit_noalloc(conn, art->messageid, art->midlen);
1793 XMIT_LITERAL("\r\n");
1794 xmit_artbody(conn, artdata);
1796 article_done(art, -1);
1800 /* we got 235 from IHAVE */
1802 xmit_artbody(conn, artdata);
1804 XMIT_LITERAL(".\r\n");
1808 LIST_ADDTAIL(conn->sent, art);
1814 XMIT_LITERAL("CHECK ");
1816 XMIT_LITERAL("IHAVE ");
1817 xmit_noalloc(conn, art->messageid, art->midlen);
1818 XMIT_LITERAL("\r\n");
1820 assert(art->state == art_Unchecked);
1821 art->ipf->counts[art->state][RC_sent]++;
1822 LIST_ADDTAIL(conn->sent, art);
1827 /*========== handling responses from peer ==========*/
1829 static const oop_rd_style peer_rd_style= {
1830 OOP_RD_DELIM_STRIP, '\n',
1832 OOP_RD_SHORTREC_FORBID
1835 static void *peer_rd_err(oop_source *lp, oop_read *oread, oop_rd_event ev,
1836 const char *errmsg, int errnoval,
1837 const char *data, size_t recsz, void *conn_v) {
1839 connfail(conn, "error receiving from peer: %s", errmsg);
1840 return OOP_CONTINUE;
1843 static Article *article_reply_check(Conn *conn, const char *response,
1844 int code_indicates_streaming,
1846 /* 1:yes, -1:no, 0:dontcare */,
1847 const char *sanitised_response) {
1848 Article *art= LIST_HEAD(conn->sent);
1852 "peer gave unexpected response when no commands outstanding: %s",
1853 sanitised_response);
1857 if (code_indicates_streaming) {
1858 assert(!memchr(response, 0, 4)); /* ensured by peer_rd_ok */
1859 if (!conn->stream) {
1860 connfail(conn, "peer gave streaming response code "
1861 " to IHAVE or subsequent body: %s", sanitised_response);
1864 const char *got_mid= response+4;
1865 int got_midlen= strcspn(got_mid, " \n\r");
1866 if (got_midlen<3 || got_mid[0]!='<' || got_mid[got_midlen-1]!='>') {
1867 connfail(conn, "peer gave streaming response with syntactically invalid"
1868 " messageid: %s", sanitised_response);
1871 if (got_midlen != art->midlen ||
1872 memcmp(got_mid, art->messageid, got_midlen)) {
1873 connfail(conn, "peer gave streaming response code to wrong article -"
1874 " probable synchronisation problem; we offered: %s;"
1876 art->messageid, sanitised_response);
1881 connfail(conn, "peer gave non-streaming response code to"
1882 " CHECK/TAKETHIS: %s", sanitised_response);
1887 if (must_have_sent>0 && art->state < art_Wanted) {
1888 connfail(conn, "peer says article accepted but"
1889 " we had not sent the body: %s", sanitised_response);
1892 if (must_have_sent<0 && art->state >= art_Wanted) {
1893 connfail(conn, "peer says please sent the article but we just did: %s",
1894 sanitised_response);
1898 Article *art_again= LIST_REMHEAD(conn->sent);
1899 assert(art_again == art);
1903 static void update_nocheck(int accepted) {
1904 accept_proportion *= nocheck_decay;
1905 accept_proportion += accepted * (1.0 - nocheck_decay);
1906 int new_nocheck= accept_proportion >= nocheck_thresh;
1907 if (new_nocheck && !nocheck_reported) {
1908 notice("entering nocheck mode for the first time");
1909 nocheck_reported= 1;
1910 } else if (new_nocheck != nocheck) {
1911 dbg("nocheck mode %s", new_nocheck ? "start" : "stop");
1913 nocheck= new_nocheck;
1916 static void article_done(Article *art, int whichcount) {
1917 if (whichcount>=0 && !art->missing)
1918 art->ipf->counts[art->state][whichcount]++;
1920 if (whichcount == RC_accepted) update_nocheck(1);
1921 else if (whichcount == RC_unwanted) update_nocheck(0);
1923 InputFile *ipf= art->ipf;
1925 while (art->blanklen) {
1926 static const char spaces[]=
1936 int w= art->blanklen; if (w >= sizeof(spaces)) w= sizeof(spaces)-1;
1937 int r= pwrite(ipf->fd, spaces, w, art->offset);
1939 if (errno==EINTR) continue;
1940 syscrash("failed to blank entry for %s (length %d at offset %lu) in %s",
1941 art->messageid, art->blanklen,
1942 (unsigned long)art->offset, ipf->path);
1944 assert(r>=0 && r<=w);
1950 assert(ipf->inprogress >= 0);
1953 if (!ipf->inprogress && ipf != main_input_file)
1954 queue_check_input_done();
1957 static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_rd_event ev,
1958 const char *errmsg, int errnoval,
1959 const char *data, size_t recsz, void *conn_v) {
1962 if (ev == OOP_RD_EOF) {
1963 connfail(conn, "unexpected EOF from peer");
1964 return OOP_CONTINUE;
1966 assert(ev == OOP_RD_OK);
1968 char *sani= sanitise(data,-1);
1971 unsigned long code= strtoul(data, &ep, 10);
1972 if (ep != data+3 || *ep != ' ' || data[0]=='0') {
1973 connfail(conn, "badly formatted response from peer: %s", sani);
1974 return OOP_CONTINUE;
1977 int busy= conn_busy(conn);
1979 if (conn->quitting) {
1980 if (code!=205 && code!=400) {
1981 connfail(conn, "peer gave unexpected response to QUIT (%s): %s",
1982 conn->quitting, sani);
1984 LIST_REMOVE(conns,conn);
1985 notice("C%d (now %d) idle connection closed (%s)",
1986 conn->fd, conns.count, conn->quitting);
1990 return OOP_CONTINUE;
1993 conn->since_activity= 0;
1996 #define GET_ARTICLE(musthavesent) do{ \
1997 art= article_reply_check(conn, data, code_streaming, musthavesent, sani); \
1998 if (!art) return OOP_CONTINUE; /* reply_check has failed the conn */ \
2001 #define ARTICLE_DEALTWITH(streaming,musthavesent,how) do{ \
2002 code_streaming= (streaming); \
2003 GET_ARTICLE(musthavesent); \
2004 article_done(art, RC_##how); \
2008 #define PEERBADMSG(m) do { \
2009 connfail(conn, m ": %s", sani); return OOP_CONTINUE; \
2012 int code_streaming= 0;
2016 default: PEERBADMSG("peer sent unexpected message");
2020 PEERBADMSG("peer timed us out or stopped accepting articles");
2022 LIST_REMOVE(conns,conn);
2023 notice("C%d (now %d) idle connection closed by peer",
2024 conns.count, conn->fd);
2026 return OOP_CONTINUE;
2028 case 435: ARTICLE_DEALTWITH(0,0,unwanted); /* IHAVE says they have it */
2029 case 438: ARTICLE_DEALTWITH(1,0,unwanted); /* CHECK/TAKETHIS: they have it */
2031 case 235: ARTICLE_DEALTWITH(0,1,accepted); /* IHAVE says thanks */
2032 case 239: ARTICLE_DEALTWITH(1,1,accepted); /* TAKETHIS says thanks */
2034 case 437: ARTICLE_DEALTWITH(0,0,rejected); /* IHAVE says rejected */
2035 case 439: ARTICLE_DEALTWITH(1,0,rejected); /* TAKETHIS says rejected */
2037 case 238: /* CHECK says send it */
2039 case 335: /* IHAVE says send it */
2041 assert(art->state == art_Unchecked);
2042 art->ipf->counts[art->state][RC_accepted]++;
2043 art->state= art_Wanted;
2044 LIST_ADDTAIL(conn->priority, art);
2047 case 431: /* CHECK or TAKETHIS says try later */
2049 case 436: /* IHAVE says try later */
2051 article_defer(art, RC_deferred);
2057 conn_maybe_write(conn);
2058 check_assign_articles();
2059 return OOP_CONTINUE;
2063 /*========== monitoring of input files ==========*/
2065 static void feedfile_eof(InputFile *ipf) {
2066 assert(ipf != main_input_file); /* promised by tailing_try_read */
2067 inputfile_reading_stop(ipf);
2069 if (ipf == flushing_input_file) {
2070 assert(sms==sm_SEPARATED || sms==sm_DROPPING);
2071 if (main_input_file) inputfile_reading_start(main_input_file);
2072 statemc_check_flushing_done();
2073 } else if (ipf == backlog_input_file) {
2074 statemc_check_backlog_done();
2076 abort(); /* supposed to wait rather than get EOF on main input file */
2080 static InputFile *open_input_file(const char *path) {
2081 int fd= open(path, O_RDWR);
2083 if (errno==ENOENT) return 0;
2084 sysfatal("unable to open input file %s", path);
2088 InputFile *ipf= mmalloc(sizeof(*ipf) + strlen(path) + 1);
2089 memset(ipf,0,sizeof(*ipf));
2093 LIST_INIT(ipf->queue);
2094 strcpy(ipf->path, path);
2099 static void close_input_file(InputFile *ipf) { /* does not free */
2100 assert(!ipf->readable_callback); /* must have had ->on_cancel */
2101 assert(!ipf->filemon); /* must have had inputfile_reading_stop */
2102 assert(!ipf->rd); /* must have had inputfile_reading_stop */
2103 assert(!ipf->inprogress); /* no dangling pointers pointing here */
2104 mclose_perhaps(&ipf->fd, "input file ", ipf->path);
2108 /*---------- dealing with articles read in the input file ----------*/
2110 static void *feedfile_got_bad_data(InputFile *ipf, off_t offset,
2111 const char *data, const char *how) {
2112 warn("corrupted file: %s, offset %lu: %s: in %s",
2113 ipf->path, (unsigned long)offset, how, sanitise(data,-1));
2114 ipf->readcount_err++;
2115 if (ipf->readcount_err > max_bad_data_initial +
2116 (ipf->readcount_ok+ipf->readcount_blank) / max_bad_data_ratio)
2117 crash("too much garbage in input file! (%d errs, %d ok, %d blank)",
2118 ipf->readcount_err, ipf->readcount_ok, ipf->readcount_blank);
2119 return OOP_CONTINUE;
2122 static void *feedfile_read_err(oop_source *lp, oop_read *rd,
2123 oop_rd_event ev, const char *errmsg,
2124 int errnoval, const char *data, size_t recsz,
2126 InputFile *ipf= ipf_v;
2127 assert(ev == OOP_RD_SYSTEM);
2129 syscrash("error reading input file: %s, offset %lu",
2130 ipf->path, (unsigned long)ipf->offset);
2133 static void *feedfile_got_article(oop_source *lp, oop_read *rd,
2134 oop_rd_event ev, const char *errmsg,
2135 int errnoval, const char *data, size_t recsz,
2137 InputFile *ipf= ipf_v;
2139 char tokentextbuf[sizeof(TOKEN)*2+3];
2141 if (!data) { feedfile_eof(ipf); return OOP_CONTINUE; }
2143 off_t old_offset= ipf->offset;
2144 ipf->offset += recsz + !!(ev == OOP_RD_OK);
2146 #define X_BAD_DATA(m) return feedfile_got_bad_data(ipf,old_offset,data,m);
2148 if (ev==OOP_RD_PARTREC)
2149 feedfile_got_bad_data(ipf,old_offset,data,"missing final newline");
2150 /* but process it anyway */
2152 if (ipf->skippinglong) {
2153 if (ev==OOP_RD_OK) ipf->skippinglong= 0; /* fine now */
2154 return OOP_CONTINUE;
2156 if (ev==OOP_RD_LONG) {
2157 ipf->skippinglong= 1;
2158 X_BAD_DATA("overly long line");
2161 if (memchr(data,'\0',recsz)) X_BAD_DATA("nul byte");
2162 if (!recsz) X_BAD_DATA("empty line");
2165 if (strspn(data," ") != recsz) X_BAD_DATA("line partially blanked");
2166 ipf->readcount_blank++;
2167 return OOP_CONTINUE;
2170 char *space= strchr(data,' ');
2171 int tokenlen= space-data;
2172 int midlen= (int)recsz-tokenlen-1;
2173 if (midlen <= 2) X_BAD_DATA("no room for messageid");
2174 if (space[1]!='<' || space[midlen]!='>') X_BAD_DATA("invalid messageid");
2176 if (tokenlen != sizeof(TOKEN)*2+2) X_BAD_DATA("token wrong length");
2177 memcpy(tokentextbuf, data, tokenlen);
2178 tokentextbuf[tokenlen]= 0;
2179 if (!IsToken(tokentextbuf)) X_BAD_DATA("token wrong syntax");
2181 ipf->readcount_ok++;
2183 art= mmalloc(sizeof(*art) - 1 + midlen + 1);
2184 memset(art,0,sizeof(*art));
2185 art->state= art_Unchecked;
2186 art->midlen= midlen;
2187 art->ipf= ipf; ipf->inprogress++;
2188 art->token= TextToToken(tokentextbuf);
2189 art->offset= old_offset;
2190 art->blanklen= recsz;
2191 strcpy(art->messageid, space+1);
2192 LIST_ADDTAIL(ipf->queue, art);
2194 if (ipf->autodefer >= 0)
2195 article_autodefer(ipf, art);
2196 else if (ipf==backlog_input_file)
2197 article_check_expired(art);
2199 if (sms==sm_NORMAL && ipf==main_input_file &&
2200 ipf->offset >= target_max_feedfile_size)
2201 statemc_start_flush("feed file size");
2203 check_assign_articles(); /* may destroy conn but that's OK */
2204 check_reading_pause_resume(ipf);
2205 return OOP_CONTINUE;
2208 /*========== tailing input file ==========*/
2210 static void *tailing_rable_call_time(oop_source *loop, struct timeval tv,
2212 /* lifetime of ipf here is OK because destruction will cause
2213 * on_cancel which will cancel this callback */
2214 InputFile *ipf= user;
2216 if (!ipf->fake_readable) return OOP_CONTINUE;
2218 /* we just keep calling readable until our caller (oop_rd)
2219 * has called try_read, and try_read has found EOF so given EAGAIN */
2220 loop->on_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf);
2222 return ipf->readable_callback(loop, &ipf->readable,
2223 ipf->readable_callback_user);
2226 static void tailing_on_cancel(struct oop_readable *rable) {
2227 InputFile *ipf= (void*)rable;
2229 if (ipf->filemon) filemon_stop(ipf);
2230 loop->cancel_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf);
2231 ipf->readable_callback= 0;
2234 static void tailing_make_readable(InputFile *ipf) {
2235 if (!ipf || !ipf->readable_callback) /* so callers can be naive */
2237 ipf->fake_readable= 1;
2238 loop->on_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf);
2241 static int tailing_on_readable(struct oop_readable *rable,
2242 oop_readable_call *cb, void *user) {
2243 InputFile *ipf= (void*)rable;
2245 tailing_on_cancel(rable);
2246 ipf->readable_callback= cb;
2247 ipf->readable_callback_user= user;
2249 tailing_make_readable(ipf);
2253 static ssize_t tailing_try_read(struct oop_readable *rable, void *buffer,
2255 InputFile *ipf= (void*)rable;
2257 ssize_t r= read(ipf->fd, buffer, length);
2259 if (errno==EINTR) continue;
2260 ipf->fake_readable= 0;
2264 if (ipf==main_input_file) {
2266 ipf->fake_readable= 0;
2268 } else if (ipf==flushing_input_file) {
2270 assert(sms==sm_SEPARATED || sms==sm_DROPPING);
2271 } else if (ipf==backlog_input_file) {
2281 /*---------- filemon implemented with inotify ----------*/
2283 #if defined(HAVE_SYS_INOTIFY_H) && !defined(HAVE_FILEMON)
2284 #define HAVE_FILEMON
2286 #include <sys/inotify.h>
2288 static int filemon_inotify_fd;
2289 static int filemon_inotify_wdmax;
2290 static InputFile **filemon_inotify_wd2ipf;
2292 struct Filemon_Perfile {
2296 static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) {
2297 int wd= inotify_add_watch(filemon_inotify_fd, ipf->path, IN_MODIFY);
2298 if (wd < 0) sysfatal("inotify_add_watch %s", ipf->path);
2300 if (wd >= filemon_inotify_wdmax) {
2302 filemon_inotify_wd2ipf= xrealloc(filemon_inotify_wd2ipf,
2303 sizeof(*filemon_inotify_wd2ipf) * newmax);
2304 memset(filemon_inotify_wd2ipf + filemon_inotify_wdmax, 0,
2305 sizeof(*filemon_inotify_wd2ipf) * (newmax - filemon_inotify_wdmax));
2306 filemon_inotify_wdmax= newmax;
2309 assert(!filemon_inotify_wd2ipf[wd]);
2310 filemon_inotify_wd2ipf[wd]= ipf;
2312 dbg("filemon inotify startfile %p wd=%d wdmax=%d",
2313 ipf, wd, filemon_inotify_wdmax);
2318 static void filemon_method_stopfile(InputFile *ipf, Filemon_Perfile *pf) {
2320 dbg("filemon inotify stopfile %p wd=%d", ipf, wd);
2321 int r= inotify_rm_watch(filemon_inotify_fd, wd);
2322 if (r) syscrash("inotify_rm_watch");
2323 filemon_inotify_wd2ipf[wd]= 0;
2326 static void *filemon_inotify_readable(oop_source *lp, int fd,
2327 oop_event e, void *u) {
2328 struct inotify_event iev;
2330 int r= read(filemon_inotify_fd, &iev, sizeof(iev));
2332 if (isewouldblock(errno)) break;
2333 syscrash("read from inotify master");
2334 } else if (r==sizeof(iev)) {
2335 assert(iev.wd >= 0 && iev.wd < filemon_inotify_wdmax);
2337 crash("inotify read %d bytes wanted struct of %d", r, (int)sizeof(iev));
2339 InputFile *ipf= filemon_inotify_wd2ipf[iev.wd];
2340 /*dbg("filemon inotify readable read %p wd=%d", ipf, iev.wd);*/
2341 tailing_make_readable(ipf);
2343 return OOP_CONTINUE;
2346 static int filemon_method_init(void) {
2347 filemon_inotify_fd= inotify_init();
2348 if (filemon_inotify_fd<0) {
2349 syswarn("filemon/inotify: inotify_init failed");
2352 msetnonblock(filemon_inotify_fd, 1);
2353 loop->on_fd(loop, filemon_inotify_fd, OOP_READ, filemon_inotify_readable, 0);
2355 dbg("filemon inotify init filemon_inotify_fd=%d", filemon_inotify_fd);
2359 static void filemon_method_dump_info(FILE *f) {
2361 fprintf(f,"inotify");
2362 DUMPV("%d",,filemon_inotify_fd);
2363 DUMPV("%d",,filemon_inotify_wdmax);
2364 for (i=0; i<filemon_inotify_wdmax; i++)
2365 fprintf(f," wd2ipf[%d]=%p\n", i, filemon_inotify_wd2ipf[i]);
2368 #endif /* HAVE_INOTIFY && !HAVE_FILEMON */
2370 /*---------- filemon dummy implementation ----------*/
2372 #if !defined(HAVE_FILEMON)
2374 struct Filemon_Perfile { int dummy; };
2376 static int filemon_method_init(void) {
2377 warn("filemon/dummy: no filemon method compiled in");
2380 static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) { }
2381 static void filemon_method_stopfile(InputFile *ipf, Filemon_Perfile *pf) { }
2382 static void filemon_method_dump_info(FILE *f) { fprintf(f,"dummy\n"); }
2384 #endif /* !HAVE_FILEMON */
2386 /*---------- filemon generic interface ----------*/
2388 static void filemon_start(InputFile *ipf) {
2389 assert(!ipf->filemon);
2392 filemon_method_startfile(ipf, ipf->filemon);
2395 static void filemon_stop(InputFile *ipf) {
2396 if (!ipf->filemon) return;
2397 filemon_method_stopfile(ipf, ipf->filemon);
2402 /*---------- interface to start and stop an input file ----------*/
2404 static const oop_rd_style feedfile_rdstyle= {
2405 OOP_RD_DELIM_STRIP, '\n',
2407 OOP_RD_SHORTREC_LONG,
2410 static void inputfile_reading_resume(InputFile *ipf) {
2411 if (!ipf->rd) return;
2412 if (!ipf->paused) return;
2414 int r= oop_rd_read(ipf->rd, &feedfile_rdstyle, MAX_LINE_FEEDFILE,
2415 feedfile_got_article,ipf, feedfile_read_err, ipf);
2416 if (r) syscrash("unable start reading feedfile %s",ipf->path);
2421 static void inputfile_reading_pause(InputFile *ipf) {
2422 if (!ipf->rd) return;
2423 if (ipf->paused) return;
2424 oop_rd_cancel(ipf->rd);
2428 static void inputfile_reading_start(InputFile *ipf) {
2430 ipf->readable.on_readable= tailing_on_readable;
2431 ipf->readable.on_cancel= tailing_on_cancel;
2432 ipf->readable.try_read= tailing_try_read;
2433 ipf->readable.delete_tidy= 0; /* we never call oop_rd_delete_{tidy,kill} */
2434 ipf->readable.delete_kill= 0;
2436 ipf->readable_callback= 0;
2437 ipf->readable_callback_user= 0;
2439 ipf->rd= oop_rd_new(loop, &ipf->readable, 0,0);
2443 inputfile_reading_resume(ipf);
2446 static void inputfile_reading_stop(InputFile *ipf) {
2448 inputfile_reading_pause(ipf);
2449 oop_rd_delete(ipf->rd);
2451 assert(!ipf->filemon); /* we shouldn't be monitoring it now */
2455 /*========== interaction with innd - state machine ==========*/
2457 /* See official state diagram at top of file. We implement
2468 |`---------------------------------------------------.
2470 |`---------------- - - - |
2471 D ENOENT | D EXISTS see OVERALL STATES diagram |
2472 | for full startup logic |
2475 | ============ try to |
2481 | | F IS SO BIG WE SHOULD FLUSH, OR TIMEOUT |
2482 ^ | hardlink F to D |
2485 | | our handle onto F is now onto D |
2488 | |<-------------------<---------------------<---------+
2490 | | spawn inndcomm flush |
2492 | ================== |
2493 | FLUSHING[-ABSENT] |
2495 | main D tail/none |
2496 | ================== |
2498 | | INNDCOMM FLUSH FAILS ^
2499 | |`----------------------->----------. |
2501 | | NO SUCH SITE V |
2502 ^ |`--------------->----. ==================== |
2503 | | \ FLUSHFAILED[-ABSENT] |
2505 | | FLUSH OK \ main D tail/none |
2506 | | open F \ ==================== |
2508 | | \ | TIME TO RETRY |
2509 | |`------->----. ,---<---'\ `----------------'
2510 | | D NONE | | D NONE `----.
2512 | ============= V V ============
2513 | SEPARATED-1 | | DROPPING-1
2514 | flsh->rd!=0 | | flsh->rd!=0
2515 | [Separated] | | [Dropping]
2516 | main F idle | | main none
2517 | flsh D tail | | flsh D tail
2518 | ============= | | ============
2520 ^ | EOF ON D | | defer | EOF ON D
2522 | =============== | | ===============
2523 | SEPARATED-2 | | DROPPING-2
2524 | flsh->rd==0 | V flsh->rd==0
2525 | [Finishing] | | [Dropping]
2526 | main F tail | `. main none
2527 | flsh D closed | `. flsh D closed
2528 | =============== V `. ===============
2530 | | ALL D PROCESSED `. | ALL D PROCESSED
2531 | V install defer as backlog `. | install defer
2532 ^ | close D `. | close D
2533 | | unlink D `. | unlink D
2536 `----------' ==============
2556 static void startup_set_input_file(InputFile *f) {
2557 assert(!main_input_file);
2559 inputfile_reading_start(f);
2562 static void statemc_lock(void) {
2564 struct stat stab, stabf;
2567 lockfd= open(path_lock, O_CREAT|O_RDWR, 0600);
2568 if (lockfd<0) sysfatal("open lockfile %s", path_lock);
2571 memset(&fl,0,sizeof(fl));
2573 fl.l_whence= SEEK_SET;
2574 int r= fcntl(lockfd, F_SETLK, &fl);
2576 if (errno==EACCES || isewouldblock(errno)) {
2577 if (quiet_multiple) exit(0);
2578 fatal("another duct holds the lockfile");
2580 sysfatal("fcntl F_SETLK lockfile %s", path_lock);
2583 mfstat_isreg(lockfd, &stabf, path_lock, "lockfile");
2585 mlstat_isreg(path_lock, &stab, &lock_noent, "lockfile");
2587 if (!lock_noent && samefile(&stab, &stabf))
2590 mclose(lockfd, "stale lockfile ", path_lock);
2593 FILE *lockfile= fdopen(lockfd, "w");
2594 if (!lockfile) syscrash("fdopen lockfile");
2596 int r= ftruncate(lockfd, 0);
2597 if (r) syscrash("truncate lockfile to write new info");
2599 if (fprintf(lockfile, "pid %ld\nsite %s\nfeedfile %s\nfqdn %s\n",
2600 (unsigned long)self_pid,
2601 sitename, feedfile, remote_host) == EOF ||
2603 sysfatal("write info to lockfile %s", path_lock);
2605 dbg("startup: locked");
2608 static void statemc_init(void) {
2609 struct stat stabdefer;
2611 search_backlog_file();
2614 mlstat_isreg(path_defer, &stabdefer, &defer_noent, "defer file");
2616 dbg("startup: ductdefer ENOENT");
2618 dbg("startup: ductdefer nlink=%ld", (long)stabdefer.st_nlink);
2619 switch (stabdefer.st_nlink==1) {
2621 open_defer(); /* so that we will later close it and rename it */
2624 munlink(path_defer, "stale defer file link"
2625 " (presumably hardlink to backlog file)");
2628 crash("defer file %s has unexpected link count %d",
2629 path_defer, stabdefer.st_nlink);
2633 struct stat stab_f, stab_d;
2636 InputFile *file_d= open_input_file(path_flushing);
2637 if (file_d) mfstat_isreg(file_d->fd, &stab_d, path_flushing,"flushing file");
2639 mlstat_isreg(feedfile, &stab_f, &noent_f, "feedfile");
2641 if (!noent_f && file_d && samefile(&stab_f, &stab_d)) {
2642 dbg("startup: F==D => Hardlinked");
2643 munlink(feedfile, "feed file (during startup)"); /* => Moved */
2648 dbg("startup: F ENOENT => Moved");
2649 if (file_d) startup_set_input_file(file_d);
2650 spawn_inndcomm_flush("feedfile missing at startup");
2651 /* => Flushing, sms:=FLUSHING */
2654 dbg("startup: F!=D => Separated");
2655 startup_set_input_file(file_d);
2656 flushing_input_file= main_input_file;
2657 main_input_file= open_input_file(feedfile);
2658 if (!main_input_file) crash("feedfile vanished during startup");
2659 SMS(SEPARATED, max_separated_periods,
2660 "found both old and current feed files");
2662 dbg("startup: F exists, D ENOENT => Normal");
2663 InputFile *file_f= open_input_file(feedfile);
2664 if (!file_f) crash("feed file vanished during startup");
2665 startup_set_input_file(file_f);
2666 SMS(NORMAL, spontaneous_flush_periods, "normal startup");
2671 static void statemc_start_flush(const char *why) { /* Normal => Flushing */
2672 assert(sms == sm_NORMAL);
2674 dbg("starting flush (%s) (%lu >?= %lu) (%d)",
2676 (unsigned long)(main_input_file ? main_input_file->offset : 0),
2677 (unsigned long)target_max_feedfile_size,
2680 int r= link(feedfile, path_flushing);
2681 if (r) sysfatal("link feedfile %s to flushing file %s",
2682 feedfile, path_flushing);
2685 munlink(feedfile, "old feedfile link");
2688 spawn_inndcomm_flush(why); /* => Flushing FLUSHING */
2691 static int trigger_flush_ok(const char *why) {
2695 statemc_start_flush(why ? why : "periodic");
2696 return 1; /* Normal => Flushing; => FLUSHING */
2698 case sm_FLUSHFAILED:
2699 spawn_inndcomm_flush(why ? why : "retry");
2700 return 1; /* Moved => Flushing; => FLUSHING */
2704 warn("abandoning old feedfile after flush (%s), autodeferring",
2705 why ? why : "took too long to complete");
2706 assert(flushing_input_file);
2707 autodefer_input_file(flushing_input_file);
2715 static void statemc_period_poll(void) {
2716 if (!until_flush) return;
2718 assert(until_flush>=0);
2720 if (until_flush) return;
2721 int ok= trigger_flush_ok(0);
2725 static int inputfile_is_done(InputFile *ipf) {
2727 if (ipf->inprogress) return 0; /* new article in the meantime */
2728 if (ipf->rd) return 0; /* not had EOF */
2732 static void notice_processed(InputFile *ipf, int completed,
2733 const char *what, const char *spec) {
2734 if (!ipf) return; /* allows preterminate to be lazy */
2736 #define RCI_NOTHING(x) /* nothing */
2737 #define RCI_TRIPLE_FMT(x) " " #x "=" RCI_TRIPLE_FMT_BASE
2738 #define RCI_TRIPLE_VALS(x) , RCI_TRIPLE_VALS_BASE(ipf->counts, [RC_##x])
2740 #define CNT(art,rc) (ipf->counts[art_##art][RC_##rc])
2742 char *inprog= completed
2743 ? masprintf("%s","") /* GCC produces a stupid warning for printf("") ! */
2744 : masprintf(" inprogress=%ld", ipf->inprogress);
2745 char *autodefer= ipf->autodefer >= 0
2746 ? masprintf(" autodeferred=%ld", ipf->autodefer)
2747 : masprintf("%s","");
2749 info("%s %s%s read=%d (+bl=%d,+err=%d)%s%s"
2750 " missing=%d offered=%d (ch=%d,nc=%d) accepted=%d (ch=%d,nc=%d)"
2751 RESULT_COUNTS(RCI_NOTHING, RCI_TRIPLE_FMT)
2753 completed?"completed":"processed", what, spec,
2754 ipf->readcount_ok, ipf->readcount_blank, ipf->readcount_err,
2755 inprog, autodefer, ipf->count_nooffer_missing,
2756 CNT(Unchecked,sent) + CNT(Unsolicited,sent)
2757 , CNT(Unchecked,sent), CNT(Unsolicited,sent),
2758 CNT(Wanted,accepted) + CNT(Unsolicited,accepted)
2759 , CNT(Wanted,accepted), CNT(Unsolicited,accepted)
2760 RESULT_COUNTS(RCI_NOTHING, RCI_TRIPLE_VALS)
2769 static void statemc_check_backlog_done(void) {
2770 InputFile *ipf= backlog_input_file;
2771 if (!inputfile_is_done(ipf)) return;
2773 const char *slash= strrchr(ipf->path, '/');
2774 const char *leaf= slash ? slash+1 : ipf->path;
2775 const char *under= strchr(slash, '_');
2776 const char *rest= under ? under+1 : leaf;
2777 if (!strncmp(rest,"backlog",7)) rest += 7;
2778 notice_processed(ipf,1,"backlog ",rest);
2780 close_input_file(ipf);
2781 if (unlink(ipf->path)) {
2782 if (errno != ENOENT)
2783 syscrash("could not unlink processed backlog file %s", ipf->path);
2784 warn("backlog file %s vanished while we were reading it"
2785 " so we couldn't remove it (but it's done now, anyway)",
2789 backlog_input_file= 0;
2790 search_backlog_file();
2794 static void statemc_check_flushing_done(void) {
2795 InputFile *ipf= flushing_input_file;
2796 if (!inputfile_is_done(ipf)) return;
2798 assert(sms==sm_SEPARATED || sms==sm_DROPPING);
2800 notice_processed(ipf,1,"feedfile","");
2804 munlink(path_flushing, "old flushing file");
2806 close_input_file(flushing_input_file);
2807 free(flushing_input_file);
2808 flushing_input_file= 0;
2810 if (sms==sm_SEPARATED) {
2811 notice("flush complete");
2812 SMS(NORMAL, spontaneous_flush_periods, "flush complete");
2813 } else if (sms==sm_DROPPING) {
2814 SMS(DROPPED, max_separated_periods, "old flush complete");
2815 search_backlog_file();
2816 notice("feed dropped, but will continue until backlog is finished");
2820 static void *statemc_check_input_done(oop_source *lp, struct timeval now,
2822 assert(!inputfile_is_done(main_input_file));
2823 statemc_check_flushing_done();
2824 statemc_check_backlog_done();
2825 return OOP_CONTINUE;
2828 static void queue_check_input_done(void) {
2829 loop->on_time(loop, OOP_TIME_NOW, statemc_check_input_done, 0);
2832 static void statemc_setstate(StateMachineState newsms, int periods,
2833 const char *forlog, const char *why) {
2835 until_flush= periods;
2837 const char *xtra= "";
2840 case sm_FLUSHFAILED:
2841 if (!main_input_file) xtra= "-ABSENT";
2845 xtra= flushing_input_file->rd ? "-1" : "-2";
2851 info("state %s%s[%d] %s",forlog,xtra,periods,why);
2853 info("state %s%s %s",forlog,xtra,why);
2857 /*---------- defer and backlog files ----------*/
2859 static void open_defer(void) {
2864 defer= fopen(path_defer, "a+");
2865 if (!defer) sysfatal("could not open defer file %s", path_defer);
2867 /* truncate away any half-written records */
2869 mfstat_isreg(fileno(defer), &stab, path_defer, "newly opened defer file");
2871 if (stab.st_size > LONG_MAX)
2872 crash("defer file %s size is far too large", path_defer);
2877 long orgsize= stab.st_size;
2878 long truncto= stab.st_size;
2880 if (!truncto) break; /* was only (if anything) one half-truncated record */
2881 if (fseek(defer, truncto-1, SEEK_SET) < 0)
2882 syscrash("seek in defer file %s while truncating partial", path_defer);
2887 syscrash("failed read from defer file %s", path_defer);
2889 crash("defer file %s shrank while we were checking it!", path_defer);
2895 if (stab.st_size != truncto) {
2896 warn("truncating half-record at end of defer file %s -"
2897 " shrinking by %ld bytes from %ld to %ld",
2898 path_defer, orgsize - truncto, orgsize, truncto);
2901 sysfatal("could not flush defer file %s", path_defer);
2902 if (ftruncate(fileno(defer), truncto))
2903 syscrash("could not truncate defer file %s", path_defer);
2906 info("continuing existing defer file %s (%ld bytes)",
2907 path_defer, orgsize);
2909 if (fseek(defer, truncto, SEEK_SET))
2910 syscrash("could not seek to new end of defer file %s", path_defer);
2913 static void close_defer(void) {
2918 mfstat_isreg(fileno(defer), &stab, path_defer, "defer file");
2920 if (fclose(defer)) sysfatal("could not close defer file %s", path_defer);
2923 time_t now= mtime();
2925 char *backlog= masprintf("%s_backlog_%lu.%lu", feedfile,
2927 (unsigned long)stab.st_ino);
2928 if (link(path_defer, backlog))
2929 sysfatal("could not install defer file %s as backlog file %s",
2930 path_defer, backlog);
2931 if (unlink(path_defer))
2932 syscrash("could not unlink old defer link %s to backlog file %s",
2933 path_defer, backlog);
2937 if (until_backlog_nextscan < 0 ||
2938 until_backlog_nextscan > backlog_retry_minperiods + 1)
2939 until_backlog_nextscan= backlog_retry_minperiods + 1;
2942 static void poll_backlog_file(void) {
2943 if (until_backlog_nextscan < 0) return;
2944 if (until_backlog_nextscan-- > 0) return;
2945 search_backlog_file();
2948 static void search_backlog_file(void) {
2949 /* returns non-0 iff there are any backlog files */
2954 const char *oldest_path=0;
2955 time_t oldest_mtime=0, now;
2957 if (backlog_input_file) return;
2961 r= glob(globpat_backlog, GLOB_ERR|GLOB_MARK|GLOB_NOSORT, 0, &gl);
2965 sysfatal("failed to expand backlog pattern %s", globpat_backlog);
2967 fatal("out of memory expanding backlog pattern %s", globpat_backlog);
2969 for (i=0; i<gl.gl_pathc; i++) {
2970 const char *path= gl.gl_pathv[i];
2972 if (strchr(path,'#') || strchr(path,'~')) {
2973 dbg("backlog file search skipping %s", path);
2976 r= stat(path, &stab);
2978 syswarn("failed to stat backlog file %s", path);
2981 if (!S_ISREG(stab.st_mode)) {
2982 warn("backlog file %s is not a plain file (or link to one)", path);
2985 if (!oldest_path || stab.st_mtime < oldest_mtime) {
2987 oldest_mtime= stab.st_mtime;
2990 case GLOB_NOMATCH: /* fall through */
2993 syscrash("glob expansion of backlog pattern %s gave unexpected"
2994 " nonzero (error?) return value %d", globpat_backlog, r);
2998 dbg("backlog scan: none");
3000 if (sms==sm_DROPPED) {
3002 notice("feed dropped and our work is complete");
3004 int r= unlink(path_cli);
3005 if (r && errno!=ENOENT)
3006 syswarn("failed to unlink cli socket for old feed");
3008 munlink(path_lock, "lockfile for old feed");
3011 until_backlog_nextscan= backlog_spontrescan_periods;
3016 double age= difftime(now, oldest_mtime);
3017 long age_deficiency= (backlog_retry_minperiods * period_seconds) - age;
3019 if (age_deficiency <= 0) {
3020 dbg("backlog scan: found age=%f deficiency=%ld oldest=%s",
3021 age, age_deficiency, oldest_path);
3023 backlog_input_file= open_input_file(oldest_path);
3024 if (!backlog_input_file) {
3025 warn("backlog file %s vanished as we opened it", oldest_path);
3029 inputfile_reading_start(backlog_input_file);
3030 until_backlog_nextscan= -1;
3034 until_backlog_nextscan= age_deficiency / period_seconds;
3036 if (backlog_spontrescan_periods >= 0 &&
3037 until_backlog_nextscan > backlog_spontrescan_periods)
3038 until_backlog_nextscan= backlog_spontrescan_periods;
3040 dbg("backlog scan: young age=%f deficiency=%ld nextscan=%d oldest=%s",
3041 age, age_deficiency, until_backlog_nextscan, oldest_path);
3048 /*---------- shutdown and signal handling ----------*/
3050 static void preterminate(void) {
3051 if (in_child) return;
3052 notice_processed(main_input_file,0,"feedfile","");
3053 notice_processed(flushing_input_file,0,"flushing","");
3054 if (backlog_input_file)
3055 notice_processed(backlog_input_file,0, "backlog file ",
3056 backlog_input_file->path);
3059 static int signal_self_pipe[2];
3060 static sig_atomic_t terminate_sig_flag;
3062 static void raise_default(int signo) {
3063 msigsetdefault(signo);
3068 static void *sigarrived_event(oop_source *lp, int fd, oop_event e, void *u) {
3069 assert(fd=signal_self_pipe[0]);
3071 int r= read(signal_self_pipe[0], buf, sizeof(buf));
3072 if (r<0 && !isewouldblock(errno))
3073 syscrash("failed to read signal self pipe");
3074 if (r==0) crash("eof on signal self pipe");
3075 if (terminate_sig_flag) {
3077 notice("terminating (%s)", strsignal(terminate_sig_flag));
3078 raise_default(terminate_sig_flag);
3080 return OOP_CONTINUE;
3083 static void sigarrived_handler(int signum) {
3088 if (!terminate_sig_flag) terminate_sig_flag= signum;
3093 write(signal_self_pipe[1],&x,1);
3096 static void init_signals(void) {
3097 if (signal(SIGPIPE, SIG_IGN) == SIG_ERR)
3098 syscrash("could not ignore SIGPIPE");
3100 if (pipe(signal_self_pipe)) sysfatal("create self-pipe for signals");
3102 msetnonblock(signal_self_pipe[0],1);
3103 msetnonblock(signal_self_pipe[1],1);
3105 struct sigaction sa;
3106 memset(&sa,0,sizeof(sa));
3107 sa.sa_handler= sigarrived_handler;
3108 sa.sa_flags= SA_RESTART;
3109 msigaction(SIGTERM,&sa);
3110 msigaction(SIGINT,&sa);
3112 on_fd_read_except(signal_self_pipe[0], sigarrived_event);
3115 /*========== flushing the feed ==========*/
3117 static pid_t inndcomm_child;
3118 static int inndcomm_sentinel_fd;
3120 static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) {
3121 assert(inndcomm_child);
3122 assert(fd == inndcomm_sentinel_fd);
3123 int status= mwaitpid(&inndcomm_child, "inndcomm");
3126 cancel_fd_read_except(fd);
3127 mclose_perhaps(&fd, "inndcomm sentinel pipe",0);
3128 inndcomm_sentinel_fd= 0;
3130 assert(!flushing_input_file);
3132 if (WIFEXITED(status)) {
3133 switch (WEXITSTATUS(status)) {
3135 case INNDCOMMCHILD_ESTATUS_FAIL:
3138 case INNDCOMMCHILD_ESTATUS_NONESUCH:
3139 notice("feed has been dropped by innd, finishing up");
3140 flushing_input_file= main_input_file;
3141 tailing_make_readable(flushing_input_file);
3142 /* we probably previously returned EAGAIN from our fake read method
3143 * when in fact we were at EOF, so signal another readable event
3144 * so we actually see the EOF */
3148 if (flushing_input_file) {
3149 SMS(DROPPING, max_separated_periods,
3150 "feed dropped by innd, but must finish last flush");
3153 SMS(DROPPED, 0, "feed dropped by innd");
3154 search_backlog_file();
3156 return OOP_CONTINUE;
3160 flushing_input_file= main_input_file;
3161 tailing_make_readable(flushing_input_file);
3163 main_input_file= open_input_file(feedfile);
3164 if (!main_input_file)
3165 crash("flush succeeded but feedfile %s does not exist!"
3166 " (this probably means feedfile does not correspond"
3167 " to site %s in newsfeeds)", feedfile, sitename);
3169 if (flushing_input_file) {
3170 SMS(SEPARATED, max_separated_periods, "flush complete");
3173 SMS(NORMAL, spontaneous_flush_periods, "recovery flush complete");
3175 return OOP_CONTINUE;
3178 goto unexpected_exitstatus;
3181 } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) {
3182 warn("flush timed out trying to talk to innd");
3185 unexpected_exitstatus:
3186 report_child_status("inndcomm child", status);
3190 SMS(FLUSHFAILED, flushfail_retry_periods, "flush failed, will retry");
3191 return OOP_CONTINUE;
3194 static void inndcommfail(const char *what) {
3195 syswarn("error communicating with innd: %s failed: %s", what, ICCfailure);
3196 exit(INNDCOMMCHILD_ESTATUS_FAIL);
3199 void spawn_inndcomm_flush(const char *why) { /* Moved => Flushing */
3202 notice("flushing %s",why);
3204 assert(sms==sm_NORMAL || sms==sm_FLUSHFAILED);
3205 assert(!inndcomm_child);
3206 assert(!inndcomm_sentinel_fd);
3208 if (pipe(pipefds)) sysfatal("create pipe for inndcomm child sentinel");
3210 inndcomm_child= mfork("inndcomm child");
3212 if (!inndcomm_child) {
3213 const char *flushargv[2]= { sitename, 0 };
3217 mclose(pipefds[0], "(in child) inndcomm sentinel parent's end",0);
3218 /* parent spots the autoclose of pipefds[1] when we die or exit */
3220 if (simulate_flush>=0) {
3221 warn("SIMULATING flush child status %d", simulate_flush);
3222 if (simulate_flush>128) raise(simulate_flush-128);
3223 else exit(simulate_flush);
3226 alarm(inndcomm_flush_timeout);
3227 r= ICCopen(); if (r) inndcommfail("connect");
3228 r= ICCcommand('f',flushargv,&reply); if (r<0) inndcommfail("transmit");
3229 if (!r) exit(0); /* yay! */
3231 if (!strcmp(reply, "1 No such site")) exit(INNDCOMMCHILD_ESTATUS_NONESUCH);
3232 syswarn("innd ctlinnd flush failed: innd said %s", reply);
3233 exit(INNDCOMMCHILD_ESTATUS_FAIL);
3238 mclose(pipefds[1], "inndcomm sentinel child's end",0);
3239 inndcomm_sentinel_fd= pipefds[0];
3240 assert(inndcomm_sentinel_fd);
3241 on_fd_read_except(inndcomm_sentinel_fd, inndcomm_event);
3243 SMS(FLUSHING, 0, why);
3246 /*========== main program ==========*/
3248 static void postfork_inputfile(InputFile *ipf) {
3250 mclose(ipf->fd, "(in child) input file ", ipf->path);
3253 static void postfork_stdio(FILE *f, const char *what, const char *what2) {
3254 /* we have no stdio streams that are buffered long-term */
3256 if (fclose(f)) syscrash("(in child) close %s%s", what, what2?what2:0);
3259 static void postfork(void) {
3262 msigsetdefault(SIGTERM);
3263 msigsetdefault(SIGINT);
3264 msigsetdefault(SIGPIPE);
3265 if (terminate_sig_flag) raise(terminate_sig_flag);
3267 postfork_inputfile(main_input_file);
3268 postfork_inputfile(flushing_input_file);
3272 conn_closefd(conn,"(in child) ");
3274 postfork_stdio(defer, "defer file ", path_defer);
3277 typedef struct Every Every;
3279 struct timeval interval;
3284 static void every_schedule(Every *e, struct timeval base);
3286 static void *every_happens(oop_source *lp, struct timeval base, void *e_v) {
3289 if (!e->fixed_rate) mgettimeofday(&base);
3290 every_schedule(e, base);
3291 return OOP_CONTINUE;
3294 static void every_schedule(Every *e, struct timeval base) {
3295 struct timeval when;
3296 timeradd(&base, &e->interval, &when);
3297 loop->on_time(loop, when, every_happens, e);
3300 static void every(int interval, int fixed_rate, void (*f)(void)) {
3301 NEW_DECL(Every *,e);
3302 e->interval.tv_sec= interval;
3303 e->interval.tv_usec= 0;
3304 e->fixed_rate= fixed_rate;
3307 mgettimeofday(&now);
3308 every_schedule(e, now);
3311 static void filepoll(void) {
3312 tailing_make_readable(main_input_file);
3313 tailing_make_readable(flushing_input_file);
3316 static char *dbg_report_ipf(InputFile *ipf) {
3317 if (!ipf) return masprintf("none");
3319 const char *slash= strrchr(ipf->path,'/');
3320 const char *path= slash ? slash+1 : ipf->path;
3322 return masprintf("%p/%s:queue=%d,ip=%ld,autodef=%ld,off=%ld,fd=%d%s%s%s",
3324 ipf->queue.count, ipf->inprogress, ipf->autodefer,
3325 (long)ipf->offset, ipf->fd,
3326 ipf->rd ? "" : ",!rd",
3327 ipf->skippinglong ? "*skiplong" : "",
3328 ipf->rd && ipf->paused ? "*paused" : "");
3331 static void period(void) {
3332 char *dipf_main= dbg_report_ipf(main_input_file);
3333 char *dipf_flushing= dbg_report_ipf(flushing_input_file);
3334 char *dipf_backlog= dbg_report_ipf(backlog_input_file);
3337 " sms=%s[%d] conns=%d until_connect=%d"
3338 " input_files main:%s flushing:%s backlog:%s[%d]"
3339 " children connecting=%ld inndcomm=%ld lowvol_total=%d"
3341 sms_names[sms], until_flush, conns.count, until_connect,
3342 dipf_main, dipf_flushing, dipf_backlog, until_backlog_nextscan,
3343 (long)connecting_child, (long)inndcomm_child, lowvol_total
3347 free(dipf_flushing);
3350 if (until_connect) until_connect--;
3352 inputfile_queue_check_expired(backlog_input_file);
3353 poll_backlog_file();
3354 if (!backlog_input_file) close_defer(); /* want to start on a new backlog */
3355 statemc_period_poll();
3356 check_assign_articles();
3361 /*========== dumping state ==========*/
3363 static void dump_article_list(FILE *f, const CliCommand *c,
3364 const ArticleList *al) {
3365 fprintf(f, " count=%d\n", al->count);
3366 if (!c->xval) return;
3368 int i; Article *art;
3369 for (i=0, art=LIST_HEAD(*al); art; i++, art=LIST_NEXT(art)) {
3370 fprintf(f," #%05d %-11s", i, artstate_names[art->state]);
3371 DUMPV("%p", art->,ipf);
3372 DUMPV("%d", art->,missing);
3373 DUMPV("%lu", (unsigned long)art->,offset);
3374 DUMPV("%d", art->,blanklen);
3375 DUMPV("%d", art->,midlen);
3376 fprintf(f, " %s %s\n", TokenToText(art->token), art->messageid);
3380 static void dump_input_file(FILE *f, const CliCommand *c,
3381 InputFile *ipf, const char *wh) {
3382 char *dipf= dbg_report_ipf(ipf);
3383 fprintf(f,"input %s %s", wh, dipf);
3387 DUMPV("%d", ipf->,readcount_ok);
3388 DUMPV("%d", ipf->,readcount_blank);
3389 DUMPV("%d", ipf->,readcount_err);
3390 DUMPV("%d", ipf->,count_nooffer_missing);
3394 ArtState state; const char *const *statename;
3395 for (state=0, statename=artstate_names; *statename; state++,statename++) {
3396 #define RC_DUMP_FMT(x) " " #x "=%d"
3397 #define RC_DUMP_VAL(x) ,ipf->counts[state][RC_##x]
3398 fprintf(f,"input %s counts %-11s"
3399 RESULT_COUNTS(RC_DUMP_FMT,RC_DUMP_FMT) "\n",
3401 RESULT_COUNTS(RC_DUMP_VAL,RC_DUMP_VAL));
3403 fprintf(f,"input %s queue", wh);
3404 dump_article_list(f,c,&ipf->queue);
3410 fprintf(cc->out, "dumping state to %s\n", path_dump);
3411 FILE *f= fopen(path_dump, "w");
3412 if (!f) { fprintf(cc->out, "failed: open: %s\n", strerror(errno)); return; }
3414 fprintf(f,"general");
3415 DUMPV("%s", sms_names,[sms]);
3416 DUMPV("%d", ,until_flush);
3417 DUMPV("%ld", (long),self_pid);
3418 DUMPV("%p", , defer);
3419 DUMPV("%d", , until_connect);
3420 DUMPV("%d", , until_backlog_nextscan);
3421 DUMPV("%d", , simulate_flush);
3422 fprintf(f,"\nnocheck");
3423 DUMPV("%#.10f", , accept_proportion);
3424 DUMPV("%d", , nocheck);
3425 DUMPV("%d", , nocheck_reported);
3428 fprintf(f,"special");
3429 DUMPV("%ld", (long),connecting_child);
3430 DUMPV("%d", , connecting_fdpass_sock);
3431 DUMPV("%d", , cli_master);
3434 fprintf(f,"lowvol");
3435 DUMPV("%d", , lowvol_circptr);
3436 DUMPV("%d", , lowvol_total);
3438 for (i=0; i<lowvol_periods; i++) {
3440 if (i==lowvol_circptr) fprintf(f,"*");
3441 fprintf(f,"%d",lowvol_perperiod[i]);
3445 fprintf(f,"filemon ");
3446 filemon_method_dump_info(f);
3448 dump_input_file(f,c, main_input_file, "main" );
3449 dump_input_file(f,c, flushing_input_file, "flushing");
3450 dump_input_file(f,c, backlog_input_file, "backlog" );
3452 fprintf(f,"conns count=%d\n", conns.count);
3457 fprintf(f,"C%d",conn->fd);
3458 DUMPV("%p",conn->,rd); DUMPV("%d",conn->,max_queue);
3459 DUMPV("%d",conn->,stream); DUMPV("\"%s\"",conn->,quitting);
3460 DUMPV("%d",conn->,since_activity);
3463 fprintf(f,"C%d waiting", conn->fd); dump_article_list(f,c,&conn->waiting);
3464 fprintf(f,"C%d priority",conn->fd); dump_article_list(f,c,&conn->priority);
3465 fprintf(f,"C%d sent", conn->fd); dump_article_list(f,c,&conn->sent);
3467 fprintf(f,"C%d xmit xmitu=%d\n", conn->fd, conn->xmitu);
3468 for (i=0; i<conn->xmitu; i++) {
3469 const struct iovec *iv= &conn->xmit[i];
3470 const XmitDetails *xd= &conn->xmitd[i];
3473 case xk_Const: dinfo= masprintf("Const"); break;
3474 case xk_Artdata: dinfo= masprintf("A%p", xd->info.sm_art); break;
3478 fprintf(f," #%03d %-11s l=%d %s\n", i, dinfo, iv->iov_len,
3479 sanitise(iv->iov_base, iv->iov_len));
3485 DUMPV("%s", , feedfile);
3486 DUMPV("%s", , path_cli);
3487 DUMPV("%s", , path_lock);
3488 DUMPV("%s", , path_flushing);
3489 DUMPV("%s", , path_defer);
3490 DUMPV("%s", , path_dump);
3491 DUMPV("%s", , globpat_backlog);
3494 if (!!ferror(f) + !!fclose(f)) {
3495 fprintf(cc->out, "failed: write: %s\n", strerror(errno));
3500 /*========== option parsing ==========*/
3502 static void vbadusage(const char *fmt, va_list al) NORET_PRINTF(1,0);
3503 static void vbadusage(const char *fmt, va_list al) {
3504 char *m= mvasprintf(fmt,al);
3505 fprintf(stderr, "bad usage: %s\n"
3506 "say --help for help, or read the manpage\n",
3508 if (interactive < 2)
3509 syslog(LOG_ERR,"innduct: invoked with bad usage: %s",m);
3513 /*---------- generic option parser ----------*/
3515 static void badusage(const char *fmt, ...) NORET_PRINTF(1,2);
3516 static void badusage(const char *fmt, ...) {
3523 of_seconds= 001000u,
3524 of_boolean= 002000u,
3527 typedef struct Option Option;
3528 typedef void OptionParser(const Option*, const char *val);
3532 const char *lng, *formarg;
3538 static void parse_options(const Option *options, char ***argvp) {
3539 /* on return *argvp is first non-option arg; argc is not updated */
3542 const char *arg= *++(*argvp);
3544 if (*arg != '-') break;
3545 if (!strcmp(arg,"--")) { arg= *++(*argvp); break; }
3547 while ((a= *++arg)) {
3551 char *equals= strchr(arg,'=');
3552 int len= equals ? (equals - arg) : strlen(arg);
3553 for (o=options; o->shrt || o->lng; o++)
3554 if (strlen(o->lng) == len && !memcmp(o->lng,arg,len))
3556 badusage("unknown long option --%s",arg);
3559 if (equals) badusage("option --%s does not take a value",o->lng);
3561 } else if (equals) {
3565 if (!arg) badusage("option --%s needs a value for %s",
3566 o->lng, o->formarg);
3569 break; /* eaten the whole argument now */
3571 for (o=options; o->shrt || o->lng; o++)
3574 badusage("unknown short option -%c",a);
3581 if (!arg) badusage("option -%c needs a value for %s",
3582 o->shrt, o->formarg);
3585 break; /* eaten the whole argument now */
3591 #define DELIMPERHAPS(delim,str) (str) ? (delim) : "", (str) ? (str) : ""
3593 static void print_options(const Option *options, FILE *f) {
3595 for (o=options; o->shrt || o->lng; o++) {
3596 char shrt[2] = { o->shrt, 0 };
3597 char *optspec= masprintf("%s%s%s%s%s",
3598 o->shrt ? "-" : "", shrt,
3599 o->shrt && o->lng ? "|" : "",
3600 DELIMPERHAPS("--", o->lng));
3601 fprintf(f, " %s%s%s\n", optspec, DELIMPERHAPS(" ", o->formarg));
3606 /*---------- specific option types ----------*/
3608 static void op_integer(const Option *o, const char *val) {
3611 unsigned long ul= strtoul(val,&ep,10);
3612 if (*ep || ep==val || errno || ul>INT_MAX)
3613 badusage("bad integer value for %s",o->lng);
3614 int *store= o->store;
3618 static void op_double(const Option *o, const char *val) {
3619 int *store= o->store;
3622 *store= strtod(val, &ep);
3623 if (*ep || ep==val || errno)
3624 badusage("bad floating point value for %s",o->lng);
3627 static void op_string(const Option *o, const char *val) {
3628 const char **store= o->store;
3632 static void op_seconds(const Option *o, const char *val) {
3633 int *store= o->store;
3637 double v= strtod(val,&ep);
3638 if (ep==val) badusage("bad time/duration value for %s",o->lng);
3640 if (!*ep || !strcmp(ep,"s") || !strcmp(ep,"sec")) unit= 1;
3641 else if (!strcmp(ep,"m") || !strcmp(ep,"min")) unit= 60;
3642 else if (!strcmp(ep,"h") || !strcmp(ep,"hour")) unit= 3600;
3643 else if (!strcmp(ep,"d") || !strcmp(ep,"day")) unit= 86400;
3644 else if (!strcmp(ep,"das")) unit= 10;
3645 else if (!strcmp(ep,"hs")) unit= 100;
3646 else if (!strcmp(ep,"ks")) unit= 1000;
3647 else if (!strcmp(ep,"Ms")) unit= 1000000;
3648 else badusage("bad units %s for time/duration value for %s",ep,o->lng);
3652 if (v > INT_MAX) badusage("time/duration value for %s out of range",o->lng);
3656 static void op_setint(const Option *o, const char *val) {
3657 int *store= o->store;
3661 /*---------- specific options ----------*/
3663 static void help(const Option *o, const char *val);
3665 static const Option innduct_options[]= {
3666 {'f',"feedfile", "F", &feedfile, op_string },
3667 {'q',"quiet-multiple", 0, &quiet_multiple, op_setint, 1 },
3668 {0,"no-daemon", 0, &interactive, op_setint, 1 },
3669 {0,"interactive", 0, &interactive, op_setint, 2 },
3670 {0,"no-streaming", 0, &try_stream, op_setint, 0 },
3671 {0,"no-filemon", 0, &try_filemon, op_setint, 0 },
3672 {'C',"inndconf", "F", &inndconffile, op_string },
3673 {'P',"port", "PORT", &port, op_integer },
3674 {0,"chdir", "DIR", &path_run, op_string },
3675 {0,"cli", "DIR/|PATH", &path_cli, op_string },
3676 {0,"help", 0, 0, help },
3678 {0,"max-connections", "N", &max_connections, op_integer },
3679 {0,"max-queue-per-conn", "N", &max_queue_per_conn, op_integer },
3680 {0,"max-queue-per-file", "N", &max_queue_per_ipf, op_integer },
3681 {0,"feedfile-flush-size","BYTES", &target_max_feedfile_size, op_integer },
3682 {0,"period-interval", "TIME", &period_seconds, op_seconds },
3684 {0,"connection-timeout", "TIME", &connection_setup_timeout, op_seconds },
3685 {0,"stuck-flush-timeout", "TIME", &inndcomm_flush_timeout, op_seconds },
3686 {0,"feedfile-poll", "TIME", &filepoll_seconds, op_seconds },
3688 {0,"no-check-proportion", "PERCENT", &nocheck_thresh, op_double },
3689 {0,"no-check-response-time","ARTICLES", &nocheck_decay, op_double },
3691 {0,"reconnect-interval", "PERIOD", &reconnect_delay_periods, op_seconds },
3692 {0,"flush-retry-interval", "PERIOD", &flushfail_retry_periods, op_seconds },
3693 {0,"earliest-deferred-retry","PERIOD", &backlog_retry_minperiods, op_seconds },
3694 {0,"backlog-rescan-interval","PERIOD",&backlog_spontrescan_periods,op_seconds},
3695 {0,"max-flush-interval", "PERIOD", &spontaneous_flush_periods,op_seconds },
3696 {0,"flush-finish-timeout", "PERIOD", &max_separated_periods, op_seconds },
3697 {0,"idle-timeout", "PERIOD", &need_activity_periods, op_seconds },
3698 {0,"low-volume-thresh", "PERIOD", &lowvol_thresh, op_integer },
3699 {0,"low-volume-window", "PERIOD", &lowvol_periods, op_seconds },
3701 {0,"max-bad-input-data-ratio","PERCENT", &max_bad_data_ratio, op_double },
3702 {0,"max-bad-input-data-init", "PERCENT", &max_bad_data_initial, op_integer },
3707 static void printusage(FILE *f) {
3708 fputs("usage: innduct [options] site [fqdn]\n"
3709 "available options are:\n", f);
3710 print_options(innduct_options, f);
3713 static void help(const Option *o, const char *val) {
3715 if (ferror(stdout) || fflush(stdout)) {
3716 perror("innduct: writing help");
3722 static void convert_to_periods_rndup(int *store) {
3723 *store += period_seconds-1;
3724 *store /= period_seconds;
3727 static int path_ends_slash(const char *specified) {
3728 int l= strlen(specified);
3730 return specified[l-1] == '/';
3733 int main(int argc, char **argv) {
3734 /* set up libinn logging */
3735 message_program_name= "innduct";
3736 message_fatal_cleanup= innduct_fatal_cleanup;
3737 INNLOGWRAPS(INNLOGWRAP_CALL)
3744 parse_options(innduct_options, &argv);
3749 if (!sitename) badusage("need site name argument");
3751 if (*argv) remote_host= *argv++;
3752 else remote_host= sitename;
3754 if (*argv) badusage("too many non-option arguments");
3758 int r= innconf_read(inndconffile);
3759 if (!r) badusage("could not read inn.conf");
3761 if (!remote_host) remote_host= sitename;
3763 if (nocheck_thresh < 0 || nocheck_thresh > 100)
3764 badusage("nocheck threshold percentage must be between 0..100");
3765 nocheck_thresh *= 0.01;
3767 if (nocheck_decay < 0.1)
3768 badusage("nocheck decay articles must be at least 0.1");
3769 nocheck_decay= pow(0.5, 1.0/nocheck_decay);
3771 convert_to_periods_rndup(&reconnect_delay_periods);
3772 convert_to_periods_rndup(&flushfail_retry_periods);
3773 convert_to_periods_rndup(&backlog_retry_minperiods);
3774 convert_to_periods_rndup(&backlog_spontrescan_periods);
3775 convert_to_periods_rndup(&spontaneous_flush_periods);
3776 convert_to_periods_rndup(&max_separated_periods);
3777 convert_to_periods_rndup(&need_activity_periods);
3778 convert_to_periods_rndup(&lowvol_periods);
3780 if (max_bad_data_ratio < 0 || max_bad_data_ratio > 100)
3781 badusage("bad input data ratio must be between 0..100");
3782 max_bad_data_ratio *= 0.01;
3785 path_run= innconf->pathrun;
3787 if (!feedfile) feedfile= sitename;
3788 if (!feedfile[0]) badusage("feed filename, if specified, must be nonempty");
3789 if (path_ends_slash(feedfile))
3790 feedfile= masprintf("%s%s", feedfile, sitename);
3791 if (feedfile[0] != '/')
3792 feedfile= masprintf("%s/%s", innconf->pathoutgoing, feedfile);
3795 path_cli_dir= "innduct";
3796 } else if (!path_cli[0] || !strcmp(path_cli,"none")) {
3797 path_cli= 0; /* ok, don't then */
3798 } else if (path_ends_slash(path_cli)) {
3799 path_cli_dir= masprintf("%.*s", strlen(path_cli)-1, path_cli);
3802 path_cli= masprintf("%s/%s", path_cli_dir, sitename);
3804 if (max_queue_per_ipf<0)
3805 max_queue_per_ipf= max_queue_per_conn * 2;
3807 const char *feedfile_forbidden= "?*[~#";
3809 while ((c= *feedfile_forbidden++))
3810 if (strchr(feedfile, c))
3811 badusage("feed filename may not contain metacharacter %c",c);
3814 lowvol_perperiod= xcalloc(sizeof(*lowvol_perperiod), lowvol_periods);
3815 for (i=0; i<lowvol_periods; i++) {
3816 lowvol_perperiod[i]= lowvol_thresh;
3817 lowvol_total += lowvol_thresh;
3819 lowvol_total -= lowvol_thresh;
3823 path_lock= masprintf("%s_lock", feedfile);
3824 path_flushing= masprintf("%s_flushing", feedfile);
3825 path_defer= masprintf("%s_defer", feedfile);
3826 path_dump= masprintf("%s_dump", feedfile);
3827 globpat_backlog= masprintf("%s_backlog*", feedfile);
3829 oop_source_sys *sysloop= oop_sys_new();
3830 if (!sysloop) syscrash("could not create liboop event loop");
3831 loop= (oop_source*)sysloop;
3835 if (interactive < 1) {
3837 for (i=3; i<255; i++)
3838 /* do this now before we open syslog, etc. */
3842 if (interactive < 2)
3843 openlog("innduct",LOG_NDELAY|LOG_PID,LOG_NEWS);
3845 if (interactive < 1) {
3846 int null= open("/dev/null",O_RDWR);
3847 if (null<0) sysfatal("failed to open /dev/null");
3851 mclose(null, "/dev/null original fd",0);
3853 pid_t child1= mfork("daemonise first fork");
3854 if (child1) _exit(0);
3856 pid_t sid= setsid();
3857 if (sid == -1) sysfatal("setsid failed");
3859 pid_t child2= mfork("daemonise second fork");
3860 if (child2) _exit(0);
3864 if (self_pid==-1) syscrash("getpid");
3867 if (r) sysfatal("could not chdir to pathrun %s", path_run);
3876 r= SMsetup(SM_PREOPEN, &val); if (!r) warn("SMsetup SM_PREOPEN failed");
3877 r= SMinit(); if (!r) fatal("storage manager initialisation (SMinit) failed");
3879 if (interactive >= 2)
3886 notice("filemon: suppressed by command line option, polling");
3888 filemon_ok= filemon_method_init();
3890 warn("filemon: no file monitoring available, polling");
3893 every(filepoll_seconds,0,filepoll);
3895 every(period_seconds,1,period);
3901 void *run= oop_sys_run(sysloop);
3902 assert(run == OOP_ERROR);
3903 syscrash("event loop failed");