3 * - close idle connections
8 * Newsfeeds file entries should look like this:
9 * host.name.of.site[/exclude,exclude,...]\
10 * :pattern,pattern...[/distribution,distribution...]\
14 * sitename[/exclude,exclude,...]\
15 * :pattern,pattern...[/distribution,distribution...]\
21 * or might be blanked out
22 * <spc><spc><spc><spc>....
24 * F site.name main feed file
25 * opened/created, then written, by innd
28 * tokens blanked out by duct when processed
29 * site.name_lock lock preventing multiple ducts
30 * to hold lock must open,F_SETLK[W]
31 * and then stat to check that locked file
32 * still has name site.name_lock
33 * holder of this lock is "duct"
34 * (only) lockholder may remove the lockfile
35 * D site.name_flushing temporary feed file during flush (or crash)
36 * hardlink created by duct
38 * site.name_defer 431'd articles, still being written,
39 * created, written, used by duct
41 * site.name_backlog.<date>.<inum>
42 * 431'd articles, ready for innxmit or duct
43 * created (link/mv) by duct
44 * site.name_backlog<anything-else> (where <anything-else> does not
45 * contain '#' or '~') eg
46 * site.name_backlog.manual
47 * anything the sysadmin likes (eg, feed files
48 * from old feeds to be merged into this one)
49 * created (link/mv) by admin
50 * may be symlinks (in which case links
51 * may be written through, but only links
54 * It is safe to remove backlog files manually,
55 * if it's desired to throw away the backlog.
57 * Backlog files are also processed by innduct. We find the oldest
58 * backlog file which is at least a certain amount old, and feed it
59 * back into our processing. When every article in it has been read
60 * and processed, we unlink it and look for another backlog file.
62 * If we don't have a backlog file that we're reading, we close the
63 * defer file that we're writing and make it into a backlog file at
64 * the first convenient opportunity.
75 | | <----------------<---------------------------------'|
81 | F: innd writing, duct reading |
84 | | duct decides time to flush |
85 | | duct makes hardlink |
87 | V <------------------------'|
89 | F == D: innd writing, duct reading both exist |
92 | | <-----------<-------------<--'|
96 | V <---------------------. |
99 | D: innd writing, duct reading; or ENOENT | |
101 | | duct requests flush of feed | |
102 | | (others can too, harmlessly) | |
106 | D: innd flushing, duct; or ENOENT | |
108 | | inndcomm flush fails | |
109 | |`-------------------------->------------------' |
111 | | inndcomm reports no such site |
112 | |`---------------------------------------------------- | -.
114 | | innd finishes writing D, creates F | |
115 | | inndcomm reports flush successful | |
118 | Separated <----------------' |
119 | F: innd writing F!=D /
120 | D: duct reading; or ENOENT both exist /
122 | | duct gets to the end of D /
123 | | duct opens F too /
126 | F: innd writing, duct reading |
127 | D: duct finishing V
129 | | duct finishes processing D F: ENOENT
130 | V duct unlinks D D: duct reading
132 `--<--' | duct finishes
142 "duct reading" means innduct is reading the file but also
143 overwriting processed tokens.
147 * rune for printing diagrams:
149 perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct.c |a2ps -R -B -ops
154 /*============================== PROGRAM ==============================*/
158 #include "inn/list.h"
165 #include <sys/types.h>
166 #include <sys/wait.h>
167 #include <sys/stat.h>
168 #include <sys/socket.h>
181 #include <oop-read.h>
183 /*----- general definitions, probably best not changed -----*/
185 #define PERIOD_SECONDS 60
187 #define CONNCHILD_ESTATUS_STREAM 4
188 #define CONNCHILD_ESTATUS_NOSTREAM 5
190 #define INNDCOMMCHILD_ESTATUS_FAIL 6
191 #define INNDCOMMCHILD_ESTATUS_NONESUCH 7
193 /*----- doubly linked lists -----*/
195 #define ISNODE(T) struct { T *succ, *pred; } node /* must be at start */
196 #define DEFLIST(T) typedef struct { T *hd, *tl, *tp; int count; } T##List
198 #define NODE(n) (assert((void*)&(n)->node == &(n)), \
199 (struct node*)&(n)->node)
201 #define LIST_CHECKCANHAVENODE(l,n) \
202 ((void)((n) == ((l).hd))) /* just for the type check */
204 #define LIST_ADDSOMEHOW(l,n,list_addsomehow) \
205 ( LIST_CHECKCANHAVENODE(l,n), \
206 list_addsomehow((struct list*)&(l), NODE((n))), \
210 #define LIST_REMSOMEHOW(l,list_remsomehow) \
214 list_remsomehow((struct list*)&(l)) ) \
220 #define LIST_ADDHEAD(l,n) LIST_ADDSOMEHOW((l),(n),list_addhead)
221 #define LIST_ADDTAIL(l,n) LIST_ADDSOMEHOW((l),(n),list_addtail)
222 #define LIST_REMHEAD(l) LIST_REMSOMEHOW((l),list_remhead)
223 #define LIST_REMTAIL(l) LIST_REMSOMEHOW((l),list_remtail)
225 #define LIST_HEAD(l) ((typeof((l).hd))(list_head((struct list*)&(l))))
226 #define LIST_NEXT(n) ((typeof(n))list_succ(NODE((n))))
227 #define LIST_BACK(n) ((typeof(n))list_pred(NODE((n))))
229 #define LIST_REMOVE(l,n) \
230 ( LIST_CHECKCANHAVENODE(l,n), \
231 list_remove(NODE((n))), \
235 #define LIST_INSERT(l,n,pred) \
236 ( LIST_CHECKCANHAVENODE(l,n), \
237 LIST_CHECKCANHAVENODE(l,pred), \
238 list_insert((struct list*)&(l), NODE((n)), NODE((pred))), \
242 /*----- type predeclarations -----*/
244 typedef struct Conn Conn;
245 typedef struct Article Article;
246 typedef struct InputFile InputFile;
247 typedef struct XmitDetails XmitDetails;
248 typedef enum StateMachineState StateMachineState;
253 /*----- function predeclarations -----*/
255 static void conn_maybe_write(Conn *conn);
256 static void conn_make_some_xmits(Conn *conn);
257 static void *conn_write_some_xmits(Conn *conn);
259 static void xmit_free(XmitDetails *d);
261 static int filemon_init(void);
262 static void filemon_setfile(int mainfeed_fd, const char *mainfeed_path);
263 static void filemon_callback(void);
265 static void statemc_setstate(StateMachineState newsms, int periods,
266 const char *forlog, const char *why);
267 static void check_master_queue(void);
268 static void queue_check_input_done(void);
270 static void postfork(const char *what);
271 static void postfork_inputfile(InputFile *ipf);
273 static void open_defer(void);
275 /*----- configuration options -----*/
277 static char *sitename, *feedfile;
278 static const char *remote_host;
279 static int quiet_multiple=0, become_daemon=1;
281 static int max_connections=10, max_queue_per_conn=200;
283 static int connection_setup_timeout=200, port=119, try_stream=1;
284 static int inndcomm_flush_timeout=100;
285 static int reconnect_delay_periods, flushfail_retry_periods, open_wait_periods;
286 static int backlog_retry_minperiods, backlog_spontaneous_rescan_periods;
287 static const char *inndconffile;
289 static double accept_proportion;
290 static double nocheck_thresh_pct= 95.0;
291 static double nocheck_thresh; /* computed in main from _pct */
292 static double nocheck_decay_articles= 100; /* converted to _decay */
293 static double nocheck_decay; /* computed in main from _articles */
294 static int nocheck, nocheck_reported;
297 /*----- statistics -----*/
299 typedef enum { /* in queue in conn->sent */
300 art_Unchecked, /* not checked, not sent checking */
301 art_Wanted, /* checked, wanted sent body as requested */
302 art_Unsolicited, /* - sent body without check */
306 #define RESULT_COUNTS(RCS,RCN) \
314 #define RCI_TRIPLE_FMT_BASE "%d(id%d+bd%d+nc%d)"
315 #define RCI_TRIPLE_VALS_BASE(counts,x) \
316 counts[art_Unchecked] x \
317 + counts[art_Wanted] x \
318 + counts[art_Unsolicited] x, \
319 counts[art_Unchecked] x \
320 , counts[art_Wanted] x \
321 , counts[art_Unsolicited] x
324 #define RC_INDEX(x) RC_##x,
325 RESULT_COUNTS(RC_INDEX, RC_INDEX)
330 /*----- transmission buffers -----*/
335 xk_Malloc, xk_Const, xk_Artdata
347 /*----- core operational data structure types -----*/
350 /* This is also an instance of struct oop_readable */
351 struct oop_readable readable; /* first */
352 oop_readable_call *readable_callback;
353 void *readable_callback_user;
356 struct Filemon_Perfile *filemon;
359 long inprogress; /* no. of articles read but not processed */
362 int counts[art_MaxState][RCI_max];
377 #define SMS_LIST(X) \
385 enum StateMachineState {
386 #define SMS_DEF_ENUM(s) sm_##s,
387 SMS_LIST(SMS_DEF_ENUM)
390 static const char *sms_names[]= {
391 #define SMS_DEF_NAME(s) #s ,
392 SMS_LIST(SMS_DEF_NAME)
398 int fd, max_queue, stream, quitting;
399 ArticleList waiting; /* not yet told peer */
400 ArticleList priority; /* peer says send it now */
401 ArticleList sent; /* offered/transmitted - in xmit or waiting reply */
402 struct iovec xmit[CONNIOVS];
403 XmitDetails xmitd[CONNIOVS];
408 /*----- operational variables -----*/
410 static oop_source *loop;
412 static int until_connect;
413 static ConnList conns;
414 static ArticleList queue;
416 static char *path_lock, *path_flushing, *path_defer;
418 #define SMS(newstate, periods, why) \
419 (statemc_setstate(sm_##newstate,(periods),#newstate,(why)))
421 static StateMachineState sms;
423 static InputFile *main_input_file, *flushing_input_file, *backlog_input_file;
424 static int sm_period_counter;
427 /*========== logging ==========*/
429 static void logcore(int sysloglevel, const char *fmt, ...)
430 __attribute__((__format__(printf,2,3)));
431 static void logcore(int sysloglevel, const char *fmt, ...) {
435 vsyslog(sysloglevel,fmt,al);
437 vfprintf(stderr,fmt,al);
443 static void logv(int sysloglevel, const char *pfx, int errnoval,
444 int exitstatus, const char *fmt, va_list al)
445 __attribute__((__format__(printf,5,0)));
446 static void logv(int sysloglevel, const char *pfx, int errnoval,
447 int exitstatus, const char *fmt, va_list al) {
448 char msgbuf[256]; /* NB do not call xvasprintf here or you'll recurse */
449 vsnprintf(msgbuf,sizeof(msgbuf), fmt,al);
450 msgbuf[sizeof(msgbuf)-1]= 0;
452 if (sysloglevel >= LOG_ERR && (errnoval==EACCES || errnoval==EPERM))
453 sysloglevel= LOG_ERR; /* run by wrong user, probably */
455 logcore(sysloglevel, "<%s>%s: %s%s%s",
456 sitename, pfx, msgbuf,
457 errnoval>=0 ? ": " : "",
458 errnoval>=0 ? strerror(errnoval) : "");
461 #define logwrap(fn, pfx, sysloglevel, err, estatus) \
462 static void fn(const char *fmt, ...) \
463 __attribute__((__format__(printf,1,2))); \
464 static void fn(const char *fmt, ...) { \
467 logv(sysloglevel, pfx, err, estatus, fmt, al); \
470 logwrap(sysdie, " critical", LOG_CRIT, errno, 16);
471 logwrap(die, " critical", LOG_CRIT, -1, 16);
473 logwrap(sysfatal, " fatal", LOG_ERR, errno, 12);
474 logwrap(fatal, " fatal", LOG_ERR, -1, 12);
476 logwrap(syswarn, " warning", LOG_WARNING, errno, 0);
477 logwrap(warn, " warning", LOG_WARNING, -1, 0);
479 logwrap(notice, "", LOG_NOTICE, -1, 0);
480 logwrap(info, " info", LOG_INFO, -1, 0);
481 logwrap(debug, " debug", LOG_DEBUG, -1, 0);
484 /*========== utility functions etc. ==========*/
486 static char *xvasprintf(const char *fmt, va_list al)
487 __attribute__((__format__(printf,1,0)));
488 static char *xvasprintf(const char *fmt, va_list al) {
490 int rc= vasprintf(&str,fmt,al);
491 if (rc<0) sysdie("vasprintf(\"%s\",...) failed", fmt);
494 static char *xasprintf(const char *fmt, ...)
495 __attribute__((__format__(printf,1,2)));
496 static char *xasprintf(const char *fmt, ...) {
499 char *str= xvasprintf(fmt,al);
504 static void perhaps_close(int *fd) { if (*fd) { close(*fd); fd=0; } }
506 static pid_t xfork(const char *what) {
510 if (child==-1) sysdie("cannot fork for %s",what);
511 if (!child) postfork(what);
512 debug("forked %s %ld", what, (unsigned long)child);
516 static void on_fd_read_except(int fd, oop_call_fd callback) {
517 loop->on_fd(loop, fd, OOP_READ, callback, 0);
518 loop->on_fd(loop, fd, OOP_EXCEPTION, callback, 0);
520 static void cancel_fd_read_except(int fd) {
521 loop->cancel_fd(loop, fd, OOP_READ);
522 loop->cancel_fd(loop, fd, OOP_EXCEPTION);
525 static void report_child_status(const char *what, int status) {
526 if (WIFEXITED(status)) {
527 int es= WEXITSTATUS(status);
529 warn("%s: child died with error exit status %d",es);
530 } else if (WIFSIGNALED(status)) {
531 int sig= WTERMSIG(status);
532 const char *sigstr= strsignal(sig);
533 const char *coredump= WCOREDUMP(status) ? " (core dumped)" : "";
535 warn("%s: child died due to fatal signal %s%s", what, sigstr, coredump);
537 warn("%s: child died due to unknown fatal signal %d%s",
538 what, sig, coredump);
540 warn("%s: child died with unknown wait status %d", status);
544 static int xwaitpid(pid_t *pid, const char *what) {
547 int r= kill(*pid, SIGKILL);
548 if (r) sysdie("cannot kill %s child", what);
550 pid_t got= waitpid(*pid, &status, WNOHANG);
551 if (got==-1) sysdie("cannot reap %s child", what);
558 static void xunlink(const char *path, const char *what) {
560 if (r) sysdie("can't unlink %s %s", path, what);
563 static void check_isreg(const struct stat *stab, const char *path,
565 if (!S_ISREG(stab->st_mode))
566 die("%s %s not a plain file (mode 0%lo)",
567 what, path, (unsigned long)stab->st_mode);
570 static void xfstat(int fd, struct stat *stab_r, const char *what) {
571 int r= fstat(fd, stab_r);
572 if (r) sysdie("could not fstat %s", what);
575 static void xfstat_isreg(int fd, struct stat *stab_r,
576 const char *path, const char *what) {
577 xfstat(fd, stab_r, what);
578 check_isreg(stab_r, path, what);
581 static void xlstat_isreg(const char *path, struct stat *stab,
582 int *enoent_r /* 0 means ENOENT is fatal */,
584 int r= lstat(path, stab);
586 if (errno==ENOENT && enoent_r) { *enoent_r=1; return; }
587 sysdie("could not lstat %s %s", what, path);
589 if (enoent_r) *enoent_r= 0;
590 check_isreg(stab, path, what);
593 static void setnonblock(int fd, int nonblocking) {
594 int r= fcntl(fd, F_GETFL); if (r<0) sysdie("setnonblocking fcntl F_GETFL");
595 if (nonblocking) r |= O_NONBLOCK;
596 else r &= ~O_NONBLOCK;
597 r= fcntl(fd, F_SETFL, r); if (r<0) sysdie("setnonblocking fcntl F_SETFL");
600 static int samefile(const struct stat *a, const struct stat *b) {
601 assert(S_ISREG(a->st_mode));
602 assert(S_ISREG(b->st_mode));
603 return (a->st_ino == b->st_ino &&
604 a->st_dev == b->st_dev);
607 static char *sanitise(const char *input) {
608 static char sanibuf[100]; /* returns pointer to this buffer! */
610 const char *p= input;
614 if (q > sanibuf+sizeof(sanibuf)-8) { strcpy(q,"'.."); break; }
616 if (!c) { *q++= '\''; *q=0; break; }
617 if (c>=' ' && c<=126 && c!='\\') { *q++= c; continue; }
618 sprintf(q,"\\x%02x",c);
624 /*========== making new connections ==========*/
626 static void conn_dispose(Conn *conn) {
628 perhaps_close(&conn->fd);
630 until_connect= reconnect_delay_periods;
633 static int connecting_sockets[2]= {-1,-1};
634 static pid_t connecting_child;
636 static void connect_attempt_discard(void) {
637 if (connecting_sockets[0])
638 cancel_fd_read_except(connecting_sockets[0]);
640 perhaps_close(&connecting_sockets[0]);
641 perhaps_close(&connecting_sockets[1]);
643 if (connecting_child) {
644 int r= kill(connecting_child, SIGTERM);
645 if (r) syswarn("failed to kill connecting child");
646 int status= xwaitpid(&connecting_child, "connect");
648 if (!(WIFEXITED(status) ||
649 (WIFSIGNALED(status) && WTERMSIG(status) == SIGKILL)))
650 report_child_status("connect", status);
654 #define PREP_DECL_MSG_CMSG(msg) \
656 memset(&msg,0,sizeof(msg)); \
657 char msg##cbuf[CMSG_SPACE(sizeof(fd))]; \
658 msg.msg_control= msg##cbuf; \
659 msg.msg_controllen= sizeof(msg##cbuf);
661 static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) {
664 conn= xmalloc(sizeof(*conn));
665 memset(conn,0,sizeof(*conn));
667 PREP_DECL_MSG_CMSG(msg);
668 struct cmsghdr *h= 0;
669 ssize_t rs= recvmsg(fd, &msg, MSG_DONTWAIT);
670 if (rs >= 0) h= CMSG_FIRSTHDR(&msg);
673 pid_t got= waitpid(connecting_child, &status, WNOHANG);
675 assert(got==connecting_child);
677 if (WIFEXITED(status) &&
678 (WEXITSTATUS(status) != 0 &&
679 WEXITSTATUS(status) != CONNCHILD_ESTATUS_STREAM &&
680 WEXITSTATUS(status) != CONNCHILD_ESTATUS_NOSTREAM)) {
681 /* child already reported the problem */
682 } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) {
683 warn("connect: connection attempt timed out");
684 } else if (!WIFEXITED(status)) {
685 report_child_status("connect", status);
686 /* that's probably the root cause then */
689 /* child is still running apparently, report the socket problem */
691 syswarn("connect: read from child socket failed");
692 else if (e == OOP_EXCEPTION)
693 warn("connect: unexpected exception on child socket");
695 warn("connect: unexpected EOF on child socket");
700 #define CHK(field, val) \
701 if (h->cmsg_##field != val) { \
702 die("connect: child sent cmsg with cmsg_" #field "=%d, expected %d"); \
705 CHK(level, SOL_SOCKET);
706 CHK(type, SCM_RIGHTS);
707 CHK(len, CMSG_LEN(sizeof(conn->fd)));
710 if (CMSG_NXTHDR(&msg,h)) { die("connect: child sent many cmsgs"); goto x; }
712 memcpy(&conn->fd, CMSG_DATA(h), sizeof(conn->fd));
715 pid_t got= waitpid(connecting_child, &status, 0);
716 if (got==-1) sysdie("connect: real wait for child");
717 assert(got == connecting_child);
720 if (!WIFEXITED(status)) { report_child_status("connect",status); goto x; }
721 int es= WEXITSTATUS(status);
723 case CONNCHILD_ESTATUS_STREAM: conn->stream= 1; break;
724 case CONNCHILD_ESTATUS_NOSTREAM: conn->stream= 0; break;
726 fatal("connect: child gave unexpected exit status %d", es);
730 setnonblock(conn->fd, 1);
731 conn->max_queue= conn->stream ? max_queue_per_conn : 1;
732 LIST_ADDHEAD(conns, conn);
733 notice("#%d connected %s", conn->fd, conn->stream ? "streaming" : "plain");
734 connect_attempt_discard();
735 check_master_queue();
740 connect_attempt_discard();
743 static int allow_connect_start(void) {
744 return conns.count < max_connections
749 static void connect_start(void) {
750 assert(!connecting_sockets[0]);
751 assert(!connecting_sockets[1]);
752 assert(!connecting_child);
754 notice("starting connection attempt");
756 int r= socketpair(AF_UNIX, SOCK_STREAM, 0, connecting_sockets);
757 if (r) { syswarn("connect: cannot create socketpair for child"); goto x; }
759 connecting_child= xfork("connection");
761 if (!connecting_child) {
762 FILE *cn_from, *cn_to;
763 char buf[NNTP_STRLEN+100];
764 int exitstatus= CONNCHILD_ESTATUS_NOSTREAM;
766 r= close(connecting_sockets[0]);
767 if (r) sysdie("connect: close parent socket in child");
769 alarm(connection_setup_timeout);
770 if (NNTPconnect((char*)remote_host, port, &cn_from, &cn_to, buf) < 0) {
771 if (buf[0]) fatal("connect: rejected: %s", sanitise(buf));
772 else sysfatal("connect: connection attempt failed");
774 if (NNTPsendpassword((char*)remote_host, cn_from, cn_to) < 0)
775 sysfatal("connect: authentication failed");
777 if (fputs("MODE STREAM\r\n", cn_to) ||
779 sysfatal("connect: could not send MODE STREAM");
780 buf[sizeof(buf)-1]= 0;
781 if (!fgets(buf, sizeof(buf)-1, cn_from)) {
783 sysfatal("connect: could not read response to MODE STREAM");
785 fatal("connect: connection close in response to MODE STREAM");
790 fatal("connect: response to MODE STREAM is too long: %.100s...",
791 remote_host, sanitise(buf));
792 l--; if (l>0 && buf[l-1]=='\r') l--;
795 int rcode= strtoul(buf,&ep,10);
797 fatal("connect: bad response to MODE STREAM: %.50s", sanitise(buf));
801 exitstatus= CONNCHILD_ESTATUS_STREAM;
807 warn("connect: unexpected response to MODE STREAM: %.50s",
813 int fd= fileno(cn_from);
815 PREP_DECL_MSG_CMSG(msg);
816 struct cmsghdr *cmsg= CMSG_FIRSTHDR(&msg);
817 cmsg->cmsg_level= SOL_SOCKET;
818 cmsg->cmsg_type= SCM_RIGHTS;
819 cmsg->cmsg_len= CMSG_LEN(sizeof(fd));
820 memcpy(CMSG_DATA(cmsg), &fd, sizeof(fd));
822 msg.msg_controllen= cmsg->cmsg_len;
823 r= sendmsg(connecting_sockets[1], &msg, 0);
824 if (r) sysdie("sendmsg failed for new connection");
829 r= close(connecting_sockets[1]); connecting_sockets[1]= 0;
830 if (r) sysdie("connect: close child socket in parent");
832 on_fd_read_except(connecting_sockets[0], connchild_event);
836 connect_attempt_discard();
840 /*========== overall control of article flow ==========*/
842 static void check_master_queue(void) {
850 /* Find a connection to offer this article. We prefer a busy
851 * connection to an idle one, provided it's not full. We take the
852 * first (oldest) and since that's stable, it will mean we fill up
853 * connections in order. That way if we have too many
854 * connections, the spare ones will go away eventually.
856 for (walk=LIST_HEAD(conns); walk; walk=LIST_NEXT(walk)) {
857 int inqueue= walk->sent.count + walk->priority.count
858 + walk->waiting.count;
859 spare= walk->max_queue - inqueue;
860 assert(inqueue <= max_queue_per_conn);
862 if (inqueue==0) /*idle*/ { if (!use) use= walk; }
863 else if (spare>0) /*working*/ { use= walk; break; }
867 Article *art= LIST_REMHEAD(queue);
868 LIST_ADDTAIL(use->waiting, art);
871 conn_maybe_write(use);
872 } else if (allow_connect_start()) {
873 until_connect= reconnect_delay_periods;
882 static void *conn_writeable(oop_source *l, int fd, oop_event ev, void *u) {
887 static void conn_maybe_write(Conn *conn) {
890 conn_make_some_xmits(conn);
892 loop->cancel_fd(loop, conn->fd, OOP_WRITE);
896 void *rp= conn_write_some_xmits(conn);
897 if (rp==OOP_CONTINUE) {
898 loop->on_fd(loop, conn->fd, OOP_WRITE, conn_writeable, conn);
900 } else if (rp==OOP_HALT) {
903 /* transmitted everything */
910 static void vconnfail(Conn *conn, const char *fmt, va_list al)
911 __attribute__((__format__(printf,2,0)));
913 static void vconnfail(Conn *conn, const char *fmt, va_list al) {
914 int requeue[art_MaxState];
917 while ((art= LIST_REMHEAD(conn->priority))) LIST_ADDTAIL(queue, art);
918 while ((art= LIST_REMHEAD(conn->waiting))) LIST_ADDTAIL(queue, art);
919 while ((art= LIST_REMHEAD(conn->sent))) {
920 requeue[art->state]++;
921 if (art->state==art_Unsolicited) art->state= art_Unchecked;
922 LIST_ADDTAIL(queue,art);
927 for (i=0, d=conn->xmitd; i<conn->xmitu; i++, d++)
930 char *m= xvasprintf(fmt,al);
931 warn("#%d connection failed, requeueing " RCI_TRIPLE_FMT_BASE ": %s",
932 conn->fd, RCI_TRIPLE_VALS_BASE(requeue, /*nothing*/), m);
935 LIST_REMOVE(conns,conn);
937 check_master_queue();
940 static void connfail(Conn *conn, const char *fmt, ...)
941 __attribute__((__format__(printf,2,3)));
942 static void connfail(Conn *conn, const char *fmt, ...) {
945 vconnfail(conn,fmt,al);
949 /*========== article transmission ==========*/
951 static XmitDetails *xmit_core(Conn *conn, const char *data, int len,
952 XmitKind kind) { /* caller must then fill in details */
953 struct iovec *v= &conn->xmit[conn->xmitu];
954 XmitDetails *d= &conn->xmitd[conn->xmitu++];
955 v->iov_base= (char*)data;
961 static void xmit_noalloc(Conn *conn, const char *data, int len) {
962 xmit_core(conn,data,len, xk_Const);
964 #define XMIT_LITERAL(lit) (xmit_noalloc(conn, (lit), sizeof(lit)-1))
966 static void xmit_artbody(Conn *conn, ARTHANDLE *ah /* consumed */) {
967 XmitDetails *d= xmit_core(conn, ah->data, ah->len, xk_Artdata);
971 static void xmit_free(XmitDetails *d) {
973 case xk_Malloc: free(d->info.malloc_tofree); break;
974 case xk_Artdata: SMfreearticle(d->info.sm_art); break;
975 case xk_Const: break;
980 static void *conn_write_some_xmits(Conn *conn) {
982 * 0: nothing more to write, no need to call us again
983 * OOP_CONTINUE: more to write but fd not writeable
984 * OOP_HALT: disaster, have destroyed conn
987 int count= conn->xmitu;
988 if (!count) return 0;
990 if (count > IOV_MAX) count= IOV_MAX;
991 ssize_t rs= writev(conn->fd, conn->xmit, count);
993 if (errno == EAGAIN) return OOP_CONTINUE;
994 connfail(conn, "write failed: %s", strerror(errno));
1000 for (done=0; rs && done<conn->xmitu; done++) {
1001 struct iovec *vp= &conn->xmit[done];
1002 XmitDetails *dp= &conn->xmitd[done];
1003 if (rs > vp->iov_len) {
1011 int newu= conn->xmitu - done;
1012 memmove(conn->xmit, conn->xmit + done, newu * sizeof(*conn->xmit));
1013 memmove(conn->xmitd, conn->xmitd + done, newu * sizeof(*conn->xmitd));
1018 static void conn_make_some_xmits(Conn *conn) {
1020 if (conn->xmitu+5 > CONNIOVS)
1023 Article *art= LIST_REMHEAD(conn->priority);
1024 if (!art) art= LIST_REMHEAD(conn->waiting);
1027 if (art->state >= art_Wanted || (conn->stream && nocheck)) {
1028 /* actually send it */
1030 ARTHANDLE *artdata= SMretrieve(art->token, RETR_ALL);
1034 XMIT_LITERAL("TAKETHIS ");
1035 xmit_noalloc(conn, art->messageid, art->midlen);
1036 XMIT_LITERAL("\r\n");
1037 xmit_artbody(conn, artdata);
1040 /* we got 235 from IHAVE */
1042 xmit_artbody(conn, artdata);
1044 XMIT_LITERAL(".\r\n");
1049 art->state == art_Unchecked ? art_Unsolicited :
1050 art->state == art_Wanted ? art_Wanted :
1052 art->ipf->counts[art->state][RC_sent]++;
1053 LIST_ADDTAIL(conn->sent, art);
1059 XMIT_LITERAL("IHAVE ");
1061 XMIT_LITERAL("CHECK ");
1062 xmit_noalloc(conn, art->messageid, art->midlen);
1063 XMIT_LITERAL("\r\n");
1065 assert(art->state == art_Unchecked);
1066 art->ipf->counts[art->state][RC_sent]++;
1067 LIST_ADDTAIL(conn->sent, art);
1073 /*========== handling responses from peer ==========*/
1075 static const oop_rd_style peer_rd_style= {
1076 OOP_RD_DELIM_STRIP, '\n',
1078 OOP_RD_SHORTREC_FORBID
1081 static void *peer_rd_err(oop_source *lp, oop_read *oread, oop_event ev,
1082 const char *errmsg, int errnoval,
1083 const char *data, size_t recsz, void *conn_v) {
1085 connfail(conn, "error receiving from peer: %s", errmsg);
1086 return OOP_CONTINUE;
1089 static Article *article_reply_check(Conn *conn, const char *response,
1090 int code_indicates_streaming,
1092 /* 1:yes, -1:no, 0:dontcare */,
1093 const char *sanitised_response) {
1094 Article *art= LIST_HEAD(conn->sent);
1098 "peer gave unexpected response when no commands outstanding: %s",
1099 sanitised_response);
1103 if (code_indicates_streaming) {
1104 assert(!memchr(response, 0, 4)); /* ensured by peer_rd_ok */
1105 if (!conn->stream) {
1106 connfail(conn, "peer gave streaming response code "
1107 " to IHAVE or subsequent body: %s", sanitised_response);
1110 const char *got_mid= response+4;
1111 int got_midlen= strcspn(got_mid, " \n\r");
1112 if (got_midlen<3 || got_mid[0]!='<' || got_mid[got_midlen-1]!='>') {
1113 connfail(conn, "peer gave streaming response with syntactically invalid"
1114 " messageid: %s", sanitised_response);
1117 if (got_midlen != art->midlen ||
1118 memcmp(got_mid, art->messageid, got_midlen)) {
1119 connfail(conn, "peer gave streaming response code to wrong article -"
1120 " probable synchronisation problem; we offered: %s;"
1122 art->messageid, sanitised_response);
1127 connfail(conn, "peer gave non-streaming response code to"
1128 " CHECK/TAKETHIS: %s", sanitised_response);
1133 if (must_have_sent>0 && art->state < art_Wanted) {
1134 connfail(conn, "peer says article accepted but"
1135 " we had not sent the body: %s", sanitised_response);
1138 if (must_have_sent<0 && art->state >= art_Wanted) {
1139 connfail(conn, "peer says please sent the article but we just did: %s",
1140 sanitised_response);
1144 Article *art_again= LIST_REMHEAD(conn->sent);
1145 assert(art_again == art);
1149 static void update_nocheck(int accepted) {
1150 accept_proportion *= nocheck_decay;
1151 accept_proportion += accepted * (1.0 - nocheck_decay);
1152 int new_nocheck= accept_proportion >= nocheck_thresh;
1153 if (new_nocheck && !nocheck_reported) {
1154 notice("entering nocheck mode for the first time");
1155 nocheck_reported= 1;
1156 } else if (new_nocheck != nocheck) {
1157 debug("nocheck mode %s", new_nocheck ? "start" : "stop");
1159 nocheck= new_nocheck;
1162 static void article_done(Conn *conn, Article *art, int whichcount) {
1163 art->ipf->counts[art->state][whichcount]++;
1164 if (whichcount == RC_accepted) update_nocheck(1);
1165 else if (whichcount == RC_unwanted) update_nocheck(0);
1167 InputFile *ipf= art->ipf;
1168 while (art->blanklen) {
1169 static const char spaces[]=
1174 int w= art->blanklen; if (w >= sizeof(spaces)) w= sizeof(spaces)-1;
1175 int r= pwrite(ipf->fd, spaces, w, art->offset);
1177 if (errno==EINTR) continue;
1178 sysdie("failed to blank entry for %s (length %d at offset %lu) in %s",
1179 art->messageid, art->blanklen, art->offset, ipf->path);
1181 assert(r>=0 && r<=w);
1187 assert(ipf->inprogress >= 0);
1189 if (!ipf->inprogress && ipf != main_input_file)
1190 queue_check_input_done();
1195 static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev,
1196 const char *errmsg, int errnoval,
1197 const char *data, size_t recsz, void *conn_v) {
1200 if (ev == OOP_RD_EOF) {
1201 connfail(conn, "unexpected EOF from peer");
1202 return OOP_CONTINUE;
1204 assert(ev == OOP_RD_OK);
1206 char *sani= sanitise(data);
1209 unsigned long code= strtoul(data, &ep, 10);
1210 if (ep != data+3 || *ep != ' ' || data[0]=='0') {
1211 connfail(conn, "badly formatted response from peer: %s", sani);
1212 return OOP_CONTINUE;
1215 if (conn->quitting) {
1216 if (code!=205 && code!=503) {
1217 connfail(conn, "peer gave unexpected response to QUIT: %s", sani);
1219 notice("#%d idle connection closed\n");
1220 assert(!conn->waiting.count);
1221 assert(!conn->priority.count);
1222 assert(!conn->sent.count);
1223 assert(!conn->xmitu);
1224 LIST_REMOVE(conns,conn);
1227 return OOP_CONTINUE;
1232 #define GET_ARTICLE(musthavesent) \
1233 art= article_reply_check(conn, data, musthavesent, code_streaming, sani); \
1234 if (art) ; else return OOP_CONTINUE /* reply_check has failed the conn */
1236 #define ARTICLE_DEALTWITH(streaming,musthavesent,how) \
1237 code_streaming= (streaming); \
1238 GET_ARTICLE(musthavesent); \
1239 article_done(conn, art, RC_##how); break;
1241 #define PEERBADMSG(m) connfail(conn, m ": %s", sani); return OOP_CONTINUE
1243 int code_streaming= 0;
1247 case 400: PEERBADMSG("peer stopped accepting articles");
1248 case 503: PEERBADMSG("peer timed us out");
1249 default: PEERBADMSG("peer sent unexpected message");
1251 case 435: ARTICLE_DEALTWITH(0,0,unwanted); /* IHAVE says they have it */
1252 case 438: ARTICLE_DEALTWITH(1,0,unwanted); /* CHECK/TAKETHIS: they have it */
1254 case 235: ARTICLE_DEALTWITH(0,1,accepted); /* IHAVE says thanks */
1255 case 239: ARTICLE_DEALTWITH(1,1,accepted); /* TAKETHIS says thanks */
1257 case 437: ARTICLE_DEALTWITH(0,0,rejected); /* IHAVE says rejected */
1258 case 439: ARTICLE_DEALTWITH(1,0,rejected); /* TAKETHIS says rejected */
1260 case 238: /* CHECK says send it */
1262 case 335: /* IHAVE says send it */
1264 assert(art->state == art_Unchecked);
1265 art->ipf->counts[art->state][RC_accepted]++;
1266 art->state= art_Wanted;
1267 LIST_ADDTAIL(conn->priority, art);
1270 case 431: /* CHECK or TAKETHIS says try later */
1272 case 436: /* IHAVE says try later */
1275 if (fprintf(defer, "%s %s\n", TokenToText(art->token), art->messageid) <0
1277 sysfatal("write to defer file %s",path_defer);
1278 article_done(conn, art, RC_deferred);
1283 conn_maybe_write(sendon);
1284 check_master_queue();
1285 return OOP_CONTINUE;
1289 /*========== monitoring of input files ==========*/
1291 static void feedfile_eof(InputFile *ipf) {
1292 assert(ipf != main_input_file); /* promised by tailing_try_read */
1294 inputfile_tailing_stop(ipf);
1295 assert(ipf->fd >= 0);
1296 if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path);
1299 if (ipf == flushing_input_file) {
1300 assert(sms==sm_SEPARATED || sms==sm_DROPPING);
1301 if (main_input_file) inputfile_tailing_start(main_input_file);
1302 statemc_check_flushing_done();
1303 } else if (ipf == backlog_input_file) {
1304 statemc_check_backlog_done();
1306 abort(); /* supposed to wait rather than get EOF on main input file */
1310 static InputFile *open_input_file(const char *path) {
1311 int fd= open(path, O_RDONLY);
1313 if (errno==ENOENT) return 0;
1314 sysfatal("unable to open input file %s", path);
1317 InputFile *ipf= xmalloc(sizeof(*ipf) + strlen(path) + 1);
1318 memset(ipf,0,sizeof(*ipf));
1320 ipf->readable.on_readable= tailing_on_readable;
1321 ipf->readable.on_cancel= tailing_on_cancel;
1322 ipf->readable.try_read= tailing_try_read;
1325 strcpy(ipf->path, path);
1330 static void close_input_file(InputFile *ipf) {
1331 assert(!ipf->readable_callback); /* must have had ->on_cancel */
1332 assert(!ipf->filemon); /* must have had inputfile_tailing_stop */
1333 assert(!ipf->rd); /* must have had inputfile_tailing_stop */
1334 assert(!ipf->inprogress); /* no dangling pointers pointing here */
1337 if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path);
1341 /*---------- dealing with articles read in the input file ----------*/
1343 typedef void *feedfile_got_article(oop_source *lp, oop_read *rd,
1344 oop_rd_event ev, const char *errmsg,
1346 const char *data, size_t recsz,
1348 InputFile *ipf= ipf_v;
1350 char tokentextbuf[sizeof(TOKEN)*2+3];
1352 if (!data) { feedfile_eof(ipf); return OOP_CONTINUE; }
1354 if (data[0] && data[0]!=' ') {
1355 char *space= strchr(data,' ');
1356 int tokenlen= space-data;
1357 int midlen= (int)recsz-tokenlen-1;
1358 if (midlen < 0) goto bad_data;
1360 if (tokenlen != sizeof(TOKEN)*2+2) goto bad_data;
1361 memcpy(tokentextbuf, data, tokenlen);
1362 tokentextbuf[tokenlen]= 0;
1363 if (!IsToken(tokentextbuf)) goto bad_data;
1365 art= xmalloc(sizeof(*art) - 1 + midlen + 1);
1366 art->offset= ipf->offset;
1367 art->blanklen= recsz;
1368 art->midlen= midlen;
1369 art->state= art_Unchecked;
1370 art->ipf= ipf; ipf->inprogress++;
1371 art->token= TextToToken(tokentextbuf);
1372 strcpy(art->messageid, space+1);
1373 LIST_ADDTAIL(queue, art);
1375 ipf->offset += recsz + 1;
1377 if (sms==sm_NORMAL && ipf==main_input_file &&
1378 ipf->offset >= flush_threshold)
1379 statemc_start_flush("feed file size");
1381 check_master_queue();
1384 static void statemc_start_flush(const char *why) { /* Normal => Flushing */
1385 assert(sms == sm_NORMAL);
1387 debug("starting flush (%s) (%lu >= %lu) (%d)",
1389 (unsigned long)ipf->offset, (unsigned long)flush_threshold,
1392 int r= link(feedfile, duct_path);
1393 if (r) sysdie("link feedfile %s to flushing file %s", feedfile,
1397 xunlink(feedfile, "old feedfile link");
1400 spawn_inndcomm_flush(why); /* => Flushing FLUSHING */
1403 /*========== tailing input file ==========*/
1405 static void filemon_start(InputFile *ipf) {
1406 assert(!ipf->filemon);
1408 ipf->filemon= xmalloc(sizeof(*ipf->filemon));
1409 memset(ipf->filemon, 0, sizeof(*ipf->filemon));
1410 filemon_method_startfile(ipf, ipf->filemon);
1413 static void filemon_stop(InputFile *ipf) {
1414 if (!ipf->filemon) return;
1415 filemon_method_stopfile(ipf, ipf->filemon);
1420 static void filemon_callback(InputFile *ipf) {
1421 ipf->readable_callback(ipf->readable_callback_user);
1424 static void *tailing_rable_call_time(oop_source *loop, struct timeval tv,
1426 InputFile *ipf= user;
1427 return ipf->readable_callback(ipf->readable_callback_user);
1430 static void on_cancel(struct oop_readable *rable) {
1431 InputFile *ipf= (void*)rable;
1433 if (ipf->filemon) filemon_stopfile(ipf);
1434 loop->cancel_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf);
1435 ipf->readable_callback= 0;
1438 static void tailing_queue_readable(InputFile *ipf) {
1439 /* lifetime of ipf here is OK because destruction will cause
1440 * on_cancel which will cancel this callback */
1441 loop->on_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf);
1444 static int tailing_on_readable(struct oop_readable *rable,
1445 oop_readable_call *cb, void *user) {
1446 InputFile *ipf= (void*)rable;
1448 tailing_on_cancel(rable);
1449 ipf->readable_callback= cb;
1450 ipf->readable_callback_user= user;
1451 filemon_startfile(ipf);
1453 tailing_queue_readable(ipf);
1457 static ssize_t tailing_try_read(struct oop_readable *rable, void *buffer,
1459 InputFile *ipf= (void*)rable;
1461 ssize_t r= read(ipf->fd, buffer, length);
1463 if (errno==EINTR) continue;
1467 if (ipf==main_input_file) {
1470 } else if (ipf==flushing_input_file) {
1472 assert(sms==sm_SEPARATED || sms==sm_DROPPING);
1473 } else if (ipf==backlog_input_file) {
1483 /*---------- filemon implemented with inotify ----------*/
1485 #if defined(HAVE_INOTIFY) && !defined(HAVE_FILEMON)
1486 #define HAVE_FILEMON
1488 #include <linux/inotify.h>
1490 static int filemon_inotify_fd;
1491 static int filemon_inotify_wdmax;
1492 static InputFile **filemon_inotify_wd2ipf;
1494 typedef struct Filemon_Perfile {
1496 } Filemon_Inotify_Perfile;
1498 static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) {
1499 int wd= inotify_add_watch(filemon_inotify_fd, ipf->path, IN_MODIFY);
1500 if (wd < 0) sysdie("inotify_add_watch %s", ipf->path);
1502 if (wd >= filemon_inotify_wdmax) {
1504 filemon_inotify_wd= xrealloc(filemon_inotify_wd2ipf,
1505 sizeof(*filemon_inotify_wd2ipf) * newmax);
1506 memset(filemon_inotify_wd2ipf + filemon_inotify_wdmax, 0,
1507 sizeof(*filemon_inotify_wd2ipf) * (newmax - filemon_inotify_wdmax));
1508 filemon_inotify_wdmax= newmax;
1511 assert(!filemon_inotify_wd2ipf[wd]);
1512 filemon_inotify_wd2ipf[wd]= ipf;
1514 debug("filemon inotify startfile %p wd=%d wdmax=%d",
1515 ipf, wd, filemon_inotify_wdmax);
1520 static void filemon_method_stopfile(InputFile *ipf, Filemon_Perfile *pf) {
1522 debug("filemon inotify stopfile %p wd=%d", ipf, wd);
1523 int r= inotify_rm_watch(filemon_inotify_fd, filemon_inotify_wd);
1524 if (r) sysdie("inotify_rm_watch");
1525 filemon_inotify_wd2ipf[wd]= 0;
1528 static void *filemon_inotify_readable(oop_source *lp, int fd,
1529 oop_event e, void *u) {
1530 struct inotify_event iev;
1532 int r= read(filemon_inotify_fd, &iev, sizeof(iev));
1534 if (errno==EAGAIN) break;
1535 sysdie("read from inotify master");
1536 } else if (r==sizeof(iev)) {
1537 assert(iev.wd >= 0 && iev.wd < filemon_inotify_wdmax);
1539 die("inotify read %d bytes wanted struct of %d", r, (int)sizeof(iev));
1541 InputFile *ipf= filemon_inotify_wd2ipf[iev.wd];
1542 debug("filemon inotify readable read %p wd=%p", iev.wd, ipf);
1543 filemon_callback(ipf);
1545 return OOP_CONTINUE;
1548 static int filemon_method_init(void) {
1549 filemon_inotify_fd= inotify_init();
1550 if (filemon_inotify_fd<0) {
1551 syswarn("could not initialise inotify: inotify_init failed");
1555 loop->on_fd(loop, filemon_inotify_fd, OOP_READ, filemon_inotify_readable);
1557 debug("filemon inotify init filemon_inotify_fd=%d", filemon_inotify_fd);
1561 #endif /* HAVE_INOTIFY && !HAVE_FILEMON *//
1563 /*---------- filemon dummy implementation ----------*/
1565 #if !defined(HAVE_FILEMON)
1567 typedef struct Filemon_Perfile { int dummy; } Filemon_Dummy_Perfile;
1569 static int filemon_method_init(void) { return 0; }
1570 static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) { }
1571 static void filemon_method_stopfile(InputFile *ipf, Filemon_Perfile *pf) { }
1573 #endif /* !HAVE_FILEMON */
1575 /*---------- interface to start and stop an input file ----------*/
1577 static const oop_rd_style feedfile_rdstyle= {
1578 OOP_RD_DELIM_STRIP, '\n',
1580 OOP_RD_SHORTREC_EOF,
1583 static void inputfile_tailing_start(InputFile *ipf) {
1585 ipf->readable->on_readable= tailing_on_readable;
1586 ipf->readable->on_cancel= tailing_on_cancel;
1587 ipf->readable->try_read= tailing_try_read;
1588 ipf->readable->delete_tidy= 0; /* we never call oop_rd_delete_{tidy,kill} */
1589 ipf->readable->delete_kill= 0;
1591 ipf->readable_callback= 0;
1592 ipf->readable_callback_user= 0;
1594 ipf->rd= oop_rd_new(loop, &ipf->readable, 0,0);
1597 int r= oop_rd_read(ipf->rd, &feedfile_rdstyle, MAX_LINE_FEEDFILE,
1598 feedfile_got_article,ipf, feedfile_problem,ipf);
1599 if (r) sysdie("unable start reading feedfile %s",ipf->path);
1602 static void inputfile_tailing_stop(InputFile *ipf) {
1604 oop_rd_cancel(ipf->rd);
1605 oop_rd_delete(ipf->rd);
1607 assert(!ipf->filemon); /* we shouldn't be monitoring it now */
1611 /*========== interaction with innd - state machine ==========*/
1613 /* See official state diagram at top of file. We implement
1624 |`---------------------------------------------------.
1626 |`---------------- - - - |
1627 D ENOENT | D EXISTS see OVERALL STATES diagram |
1628 | for full startup logic |
1631 | ============ try to |
1637 | | F IS SO BIG WE SHOULD FLUSH, OR TIMEOUT |
1638 ^ | hardlink F to D |
1641 | | our handle onto F is now onto D |
1644 | |<-------------------<---------------------<---------+
1646 | | spawn inndcomm flush |
1648 | ================== |
1649 | FLUSHING[-ABSENT] |
1651 | main D tail/none |
1652 | ================== |
1654 | | INNDCOMM FLUSH FAILS ^
1655 | |`----------------------->----------. |
1657 | | NO SUCH SITE V |
1658 ^ |`--------------->----. ==================== |
1659 | | \ FLUSHFAILED[-ABSENT] |
1661 | | FLUSH OK \ main D tail/none |
1662 | | open F \ ==================== |
1664 | | \ | TIME TO RETRY |
1665 | |`------->----. ,---<---'\ `----------------'
1666 | | D NONE | | D NONE `----.
1668 | ============= V V ============
1669 | SEPARATED-1 | | DROPPING-1
1670 | flsh->fd>=0 | | flsh->fd>=0
1671 | [Separated] | | [Dropping]
1672 | main F idle | | main none
1673 | old D tail | | old D tail
1674 | ============= | | ============
1676 ^ | EOF ON D | | defer | EOF ON D
1678 | =============== | | ===============
1679 | SEPARATED-2 | | DROPPING-2
1680 | flsh->fd==-1 | V flsh->fd==-1
1681 | [Finishing] | | [Dropping]
1682 | main F tail | `. main none
1683 | old D closed | `. old D closed
1684 | =============== V `. ===============
1686 | | ALL D PROCESSED `. | ALL D PROCESSED
1687 | V install defer as backlog `. | install defer
1688 ^ | close D `. | close D
1689 | | unlink D `. | unlink D
1692 `----------' ==============
1712 static void statemc_init(void) {
1713 struct stat stab, stabf;
1715 path_lock= xasprintf("%s_lock", feedfile);
1716 path_flushing= xasprintf("%s_flushing", feedfile);
1717 path_defer= xasprintf("%s_defer", feedfile);
1718 globpat_backlog= xasprintf("%s_backlog*", feedfile);
1721 int lockfd= open(path_lock, O_CREAT|O_RDWR, 0600);
1722 if (lockfd<0) sysfatal("open lockfile %s", path_lock);
1725 memset(&fl,0,sizeof(fl));
1727 fl.l_whence= SEEK_SET;
1728 r= fcntl(lockfd, F_SETLK, &fl);
1730 if (errno==EACCES || errno==EAGAIN) {
1731 if (quiet_multiple) exit(0);
1732 fatal("another duct holds the lockfile");
1734 sysdie("fcntl F_SETLK lockfile %s", path_lock);
1737 xfstat_isreg(lockfd, &stabf, "lockfile");
1739 xlstat_isreg(path_lock, &stab, &lock_noent, "lockfile");
1741 if (!lock_noent && samefile(&stab, &stabf))
1745 sysdie("could not close stale lockfile %s", path_lock);
1747 debug("startup: locked");
1749 search_backlog_file();
1752 xlstat_isreg(path_defer, &stab, &defer_noent, "defer file");
1754 debug("startup: ductdefer ENOENT");
1756 debug("startup: ductdefer nlink=%ld", (long)stab.st_nlink);
1757 switch (stab.st_nlink==1) {
1759 open_defer(); /* so that we will later close it and rename it */
1762 xunlink(path_defer, "stale defer file link"
1763 " (presumably hardlink to backlog file)");
1766 die("defer file %s has unexpected link count %d",
1767 path_defer, stab.st_nlink);
1771 struct stat stab_f, stab_d;
1774 InputFile *file_d= open_input_file(path_flushing);
1775 if (file_d) xfstat_isreg(file_d->fd, &stab_d, "flushing file");
1777 xlstat_isreg(feedfile, &stab_f, &noent_f, "feedfile");
1779 if (!noent_f && file_d && samefile(&stab_f, &stab_d)) {
1780 debug("startup: F==D => Hardlinked");
1781 xunlink(feedfile, "feed file (during startup)"); /* => Moved */
1786 debug("startup: F ENOENT => Moved");
1787 if (file_d) startup_set_input_file(file_d);
1788 spawn_inndcomm_flush("feedfile missing at startup");
1789 /* => Flushing, sms:=FLUSHING */
1792 debug("startup: F!=D => Separated");
1793 startup_set_input_file(file_d);
1794 SMS(SEPARATED, 0, "found both old and current feed files");
1796 debug("startup: F exists, D ENOENT => Normal");
1797 FILE *file_f= open_input_file(feedfile);
1798 if (!file_f) die("feed file vanished during startup");
1799 startup_set_input_file(file_f);
1800 SMS(NORMAL, flushfail_retry_periods, "normal startup");
1805 static void statemc_period_poll(void) {
1806 if (!sm_period_counter) return;
1807 sm_period_counter--;
1808 assert(sm_period_counter>=0);
1810 if (sm_period_counter) return;
1813 statemc_start_flush("periodic"); /* Normal => Flushing; => FLUSHING */
1815 case sm_FLUSHFAILED:
1816 spawn_inndcomm_flush("retry"); /* Moved => Flushing; => FLUSHING */
1823 static void startup_set_input_file(InputFile *f) {
1824 assert(!main_input_file);
1826 inputfile_tailing_start(f);
1829 static int inputfile_is_done(InputFile *ipf) {
1831 if (ipf->inprogress) return 0; /* new article in the meantime */
1832 if (ipf->fd >= 0); return 0; /* not had EOF */
1836 static void notice_processed(InputFile *ipf, const char *what,
1838 #define RCI_NOTHING(x) /* nothing */
1839 #define RCI_TRIPLE_FMT(x) " " #x "=" RCI_TRIPLE_FMT_BASE
1840 #define RCI_TRIPLE_VALS(x) , RCI_TRIPLE_VALS_BASE(ipf->counts, .x)
1842 info("processed %s%s offered=%d(ch%d,nc%d) accepted=%d(ch%d+nc%d)"
1843 RESULT_COUNTS(RCI_NOTHING, RCI_TRIPLE_FMT)
1846 ipf->counts[art_Unchecked].sent + ipf->counts[art_Unsolicited].sent
1847 , ipf->counts[art_Unchecked].sent, ipf->counts[art_Unsolicited].sent,
1848 ipf->counts[art_Wanted].accepted + ipf->counts[art_Unsolicited].accepted
1849 ,ipf->counts[art_Wanted].accepted,ipf->counts[art_Unsolicited].accepted
1850 RESULT_COUNTS(RCI_NOTHING, RCI_TRIPLE_VALS)
1854 static void statemc_check_backlog_done(void) {
1855 InputFile *ipf= backlog_input_file();
1856 if (!inputfile_is_done(ipf)) return;
1858 const char *slash= strrchr(ipf->path, "/");
1859 const char *leaf= slash ? slash+1 : ipf->path;
1860 const char *under= strchr(slash, "_");
1861 const char *rest= under ? under+1 : leaf;
1862 if (!strncmp(rest,"backlog",7)) rest += 7;
1863 notice_processed(ipf,"backlog:",rest);
1865 close_input_file(ipf);
1866 if (unlink(ipf->path)) {
1867 if (errno != ENOENT)
1868 sysdie("could not unlink processed backlog file %s", ipf->path);
1869 warn("backlog file %s vanished while we were reading it"
1870 " so we couldn't remove it (but it's done now, anyway)",
1874 backlog_input_file= 0;
1875 search_backlog_file();
1879 static void statemc_check_flushing_done(void) {
1880 InputFile *ipf= flushing_input_file;
1881 if (!inputfile_is_done(ipf)) return;
1883 assert(sms==sm_SEPARATED || sms==sm_DROPPING);
1885 notice_processed(ipf,"feedfile",0);
1889 xunlink(path_flushing, "old flushing file");
1891 close_input_file(flushing_input_file);
1892 free(flushing_input_file);
1893 flushing_input_file= 0;
1895 if (sms==sm_SEPARATED) {
1896 notice("flush complete");
1897 SMS(NORMAL, 0, "flush complete");
1898 } else if (sms==sm_DROPPING) {
1899 SMS(DROPPED, 0, "old flush complete");
1900 search_backlog_file();
1901 notice("feed dropped, but will continue until backlog is finished");
1905 static void *statemc_check_input_done(oop_source *lp, struct timeval now,
1907 assert(!inputfile_is_done(main_input_file));
1908 statemc_check_flushing_done();
1909 statemc_check_backlog_done();
1910 return OOP_CONTINUE;
1913 static void queue_check_input_done(void) {
1914 loop->on_time(loop, OOP_TIME_NOW, statemc_check_input_done, 0);
1917 static void statemc_setstate(StateMachineState newsms, int periods,
1918 const char *forlog, const char *why) {
1920 sm_period_counter= periods;
1922 const char *xtra= "";
1924 case sm_FLUSHING: sm_FLUSHFAILED:
1925 if (!main_input_file) xtra= "-ABSENT";
1927 case sm_SEPARATED: case sm_DROPPING:
1928 xtra= flushing_input_file->fd >= 0 ? "-1" : "-2";
1934 info("%s%s[%d] %s",forlog,xtra,periods,why);
1936 info("%s%s %s",forlog,xtra,why);
1940 /*---------- defer and backlog files ----------*/
1942 static void open_defer(void) {
1947 defer= fopen(path_defer, "a+");
1948 if (!defer) sysfatal("could not open defer file %s", path_defer);
1950 /* truncate away any half-written records */
1952 xfstat_isreg(fileno(defer), &stab, "newly opened defer file");
1954 if (stab.st_size > LONG_MAX)
1955 die("defer file %s size is far too large", path_defer);
1960 long orgsize= stab.st_size;
1961 long truncto= stab.st_size;
1963 if (!truncto) break; /* was only (if anything) one half-truncated record */
1964 if (fseek(defer, truncto-1, SEEK_SET) < 0)
1965 sysdie("seek in defer file %s while truncating partial", path_defer);
1970 sysdie("failed read from defer file %s", path_defer);
1972 die("defer file %s shrank while we were checking it!", path_defer);
1978 if (stab.st_size != truncto) {
1979 warn("truncating half-record at end of defer file %s -"
1980 " shrinking by %ld bytes from %ld to %ld",
1981 path_defer, orgsize - truncto, orgsize, truncto);
1984 sysfatal("could not flush defer file %s", path_defer);
1985 if (ftruncate(fileno(defer), truncto))
1986 sysdie("could not truncate defer file %s", path_defer);
1989 info("continuing existing defer file %s (%ld bytes)",
1990 path_defer, orgsize);
1992 if (fseek(defer, truncto, SEEK_SET))
1993 sysdie("could not seek to new end of defer file %s", path_defer);
1996 static void close_defer(void) {
2000 xfstat(fileno(defer), &stab, "defer file");
2002 if (fclose(defer)) sysfatal("could not close defer file %s", path_defer);
2005 char *backlog= xasprintf("%s_backlog_%lu.%lu", feedfile,
2006 (unsigned long)now.tv_sec,
2007 (unsigned long)stab.st_ino);
2008 if (link(path_defer, path_backlog))
2009 sysfatal("could not install defer file %s as backlog file %s",
2010 path_defer, backlog);
2011 if (unlink(path_defer))
2012 sysdie("could not unlink old defer link %s to backlog file %s",
2013 path_defer, backlog);
2015 if (until_backlog_nextscan < 0 ||
2016 until_backlog_nextscan > backlog_retry_minperiods + 1)
2017 until_backlog_nextscan= backlog_retry_minperiods + 1;
2020 static void poll_backlog_file(void) {
2021 if (until_backlog_nextscan < 0) return;
2022 if (until_backlog_nextscan-- > 0) return;
2023 search_backlog_file();
2026 static void search_backlog_file(void) {
2027 /* returns non-0 iff there are any backlog files */
2032 const char *oldest_path=0;
2033 time_t oldest_mtime, now;
2035 if (backlog_input_file) return 3;
2039 r= glob(globpat_backlog, GLOB_ERR|GLOB_MARK|GLOB_NOSORT, 0, &gl);
2043 sysdie("failed to expand backlog pattern %s", globpat_backlog);
2045 die("out of memory expanding backlog pattern %s", globpat_backlog);
2047 for (i=0; i<gl.gl_pathc; i++) {
2048 const char *path= gl.gl_pathv[i];
2050 if (strchr(path,'#') || strchr(path,'~')) {
2051 debug("backlog file search skipping %s", path);
2054 r= stat(path, &stab);
2056 syswarn("failed to stat backlog file %s", path);
2059 if (!S_ISREG(stab.st_mode)) {
2060 warn("backlog file %s is not a plain file (or link to one)", path);
2063 if (!oldest_path || stab.st_mtime < oldest_mtime) {
2065 oldest_mtime= stab.st_mtime;
2068 case GLOB_NOMATCH: /* fall through */
2071 sysdie("glob expansion of backlog pattern %s gave unexpected"
2072 " nonzero (error?) return value %d", globpat_backlog, r);
2078 debug("backlog scan: none");
2080 if (sms==sm_DROPPED) {
2081 notice("feed dropped and our work is complete");
2082 xunlink(path_lock, "lockfile for old feed");
2085 until_backlog_nextscan= backlog_spontaneous_rescan_periods;
2089 now= time(); if (now==-1) sysdie("time(2) failed");
2090 double age= difftime(now, oldest_mtime);
2091 long age_deficiency= (backlog_retry_minperiods * PERIOD_SECONDS) - age;
2093 if (age_deficiency <= 0) {
2094 debug("backlog scan: found age=%f deficiency=%ld oldest=%s",
2095 age, age_deficiency, oldest_path);
2097 backlog_input_file= open_input_file(oldest_path);
2098 if (!backlog_input_file) {
2099 warn("backlog file %s vanished as we opened it", backlog_input_file);
2102 inputfile_tailing_start(backlog_input_file);
2103 until_backlog_nextscan= -1;
2107 until_backlog_nextscan= age_deficiency / PERIOD_SECONDS;
2109 if (backlog_spontaneous_rescan_periods >= 0 &&
2110 until_backlog_nextscan > backlog_spontaneous_rescan_periods)
2111 until_backlog_nextscan= backlog_spontaneous_rescan_periods;
2113 debug("backlog scan: young age=%f deficiency=%ld nextscan=%d oldest=%s",
2114 age, age_deficiency, until_backlog_nextscan, oldest_path);
2118 /*========== flushing the feed ==========*/
2120 static pid_t inndcomm_child;
2122 static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) {
2123 assert(inndcomm_child);
2124 int status= xwaitpid(&inndcomm_child, "inndcomm");
2125 loop->cancel_fd(fd);
2128 assert(!flushing_input_file);
2130 if (WIFEXITED(status)) {
2131 switch (WEXITSTATUS(status)) {
2133 case INNDCOMMCHILD_ESTATUS_FAIL:
2136 case INNDCOMMCHILD_ESTATUS_NONESUCH:
2137 warn("feed has been dropped by innd, finishing up");
2138 flushing_input_file= main_input_file;
2139 tailing_queue_readable(flushing_input_file);
2140 /* we probably previously returned EAGAIN from our fake read method
2141 * when in fact we were at EOF, so signal another readable event
2142 * so we actually see the EOF */
2146 if (flushing_input_file) {
2147 SMS(DROPPING, 0, "feed dropped by innd, but must finish last flush");
2150 SMS(DROPPED, 0, "feed dropped by innd");
2151 search_backlog_file();
2153 return OOP_CONTINUE;
2157 flushing_input_file= main_input_file;
2158 tailing_queue_readable(flushing_input_file);
2160 main_input_file= open_input_file(feedfile);
2161 if (!main_input_file)
2162 die("flush succeeded but feedfile %s does not exist!", feedfile);
2164 if (flushing_input_file) {
2165 SMS(SEPARATED, spontaneous_flush_periods, "recovery flush complete");
2168 SMS(NORMAL, spontaneous_flush_periods, "flush complete");
2170 return OOP_CONTINUE;
2173 goto unexpected_exitstatus;
2176 } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) {
2177 warn("flush timed out trying to talk to innd");
2180 unexpected_exitstatus:
2181 report_child_status("inndcomm child", status);
2185 SMS(FLUSHFAILED, flushfail_retry_periods, "flush failed, will retry");
2188 static void inndcommfail(const char *what) {
2189 syswarn("error communicating with innd: %s failed: %s", what, ICCfailure);
2190 exit(INNDCOMMCHILD_ESTATUS_FAIL);
2193 void spawn_inndcomm_flush(const char *why) { /* Moved => Flushing */
2196 notice("flushing %s",why);
2198 assert(sms==sm_NORMAL || sms==sm_FLUSHFAILED);
2199 assert(!inndcomm_child);
2201 if (pipe(pipefds)) sysdie("create pipe for inndcomm child sentinel");
2203 inndcomm_child= xfork();
2205 if (!inndcomm_child) {
2206 static char flushargv[2]= { sitename, 0 };
2211 alarm(inndcomm_flush_timeout);
2212 r= ICCopen(); if (r) inndcommfail("connect");
2213 r= ICCcommand('f',flushargv,&reply); if (r<0) inndcommfail("transmit");
2214 if (!r) exit(0); /* yay! */
2216 if (!strcmp(reply, "1 No such site")) exit(INNDCOMMCHILD_ESTATUS_NONESUCH);
2217 syswarn("innd ctlinnd flush failed: innd said %s", reply);
2218 exit(INNDCOMMCHILD_ESTATUS_FAIL);
2222 int sentinel_fd= pipefds[0];
2223 on_fd_read_except(sentinel_fd, inndcomm_event);
2225 SMS(FLUSHING, 0, why);
2228 /*========== main program ==========*/
2230 static void postfork_inputfile(InputFile *ipf) {
2232 assert(ipf->fd >= 0);
2237 static void postfork_stdio(FILE *f) {
2238 /* we have no stdio streams that are buffered long-term */
2242 static void postfork(const char *what) {
2243 if (signal(SIGPIPE, SIG_DFL) == SIG_ERR)
2244 sysdie("%s child: failed to reset SIGPIPE");
2246 postfork_inputfile(main_input_file);
2247 postfork_inputfile(flushing_input_file);
2250 for (conn=LIST_HEAD(conns); conn; conn=LIST_NEXT(conn))
2253 postfork_stdio(defer);
2256 #define EVERY(what, interval, body) \
2257 static const struct timeval what##_timeout = { 5, 0 }; \
2258 static void what##_schedule(void); \
2259 static void *what##_timedout(oop_source *lp, struct timeval tv, void *u) { \
2261 what##_schedule(); \
2263 static void what##_schedule(void) { \
2264 loop->on_time(loop, what##_timeout, what##_timedout, 0); \
2267 EVERY(filepoll, {5,0}, {
2268 if (main_input_file && main_input_file->readable_callback)
2269 filemon_callback(main_input_file);
2272 #define DEBUGF_IPF(wh) " " #wh "=%p/%s:ip=%ld,off=%ld,fd=%d%s" \
2273 #define DEBUG_IPF(sh) \
2274 wh##_input_file, debug_ipf_path(wh##_input_file), \
2275 wh##_input_file->inprogress, (long)wh##_input_file->offset, \
2276 wh##_input_file->fd, wh##_input_file->rd ? "+" : ""
2277 static const char *debug_ipf_path(InputFile *ipf) {
2278 char *slash= strrchr(ipf->path,'/');
2279 return slash ? slash+1 : ipf->path;
2282 EVERY(period, {PERIOD_SECONDS,0}, {
2284 " sms=%s[%d] conns=%d queue=%d until_connect=%d"
2285 " input_files" DEBUGF_IPF(main) DEBUGF_IPF(old) DEBUGF_FMT(flushing)
2286 " children connecting=%ld inndcomm_child"
2288 sms_names[sms], sm_period_counter,
2289 queue.count, conns.count, until_connect,
2290 DEBUG_IPF(main), DEBUG_IPF(flushing), DEBUG_IPF(flushing),
2291 (long)connecting_child, (long)inndcomm_child
2294 if (until_connect) until_connect--;
2296 poll_backlog_file();
2297 if (!backlog_input_file) close_defer(); /* want to start on a new backlog */
2298 statemc_period_poll();
2299 check_master_queue();
2303 /*========== option parsing ==========*/
2306 of_seconds= 001000u;
2307 of_boolean= 002000u;
2310 typedef struct Option Option;
2311 typedef void OptionParser(const Option*, const char *val);
2321 void op_integer(const Option *o, const char *val) {
2324 unsigned long ul= strtoul(val,&ep,10);
2325 if (*ep || ep==val || errno || ul>INT_MAX)
2326 badusage("bad integer value for %s",o->long);
2327 int *store= o->store;
2331 void op_double(const Option *o, const char *val) {
2332 int *store= o->store;
2335 *store= strtod(val, &ep);
2336 if (*ep || ep==val || errno)
2337 badusage("bad floating point value for %s",o->long);
2340 void op_string(const Option *o, const char *val) {
2341 char **store= o->store;
2346 void op_seconds(const Option *o, const char *val) {
2347 int *store= o->store;
2350 double v= strtod(val,&ep);
2351 if (ep==val) badusage("bad time/duration value for %s",o->long);
2353 if (!*ep || !strcmp(ep,"s")) unit= 1;
2354 else if (!strcmp(ep,"m")) unit= 60;
2355 else if (!strcmp(ep,"h")) unit= 3600;
2356 else if (!strcmp(ep,"d")) unit= 86400;
2357 else badusage("bad units %s for time/duration value for %s",ep,o->long);
2361 if (v > INT_MAX) badusage("time/duration value for %s out of range",o->long);
2365 void op_periods_rndup(const Option *o, const char *val) {
2366 int *store= o->store;
2368 *store += PERIOD_SECONDS-1;
2369 *store /= PERIOD_SECONDS;
2372 void op_periods_booltrue(const Option *o, const char *val) {
2373 int *store= o->store;
2376 void op_periods_boolfalse(const Option *o, const char *val) {
2377 int *store= o->store;
2381 static const Option options[]= {
2382 {'f',"feedfile", &feedfile, op_string },
2383 {'q',"quiet-multiple", &quiet_multiple, op_booltrue, 1 },
2385 { 0, "max-connections", &max_connections op_integer },
2386 { 0, "max-queue-per-conn", &max_queue_per_conn op_integer },
2389 { 0, "streaming", &try_stream, op_booltrue, 1 },
2390 { 0, "no-streaming", &try_stream, op_boolfalse, 1 },
2391 {'P',"port", &port op_integer },
2392 { 0, "inndconf", &inndconffile, op_string },
2393 {'d',"daemon", &become_daemon, op_booltrue, 1 },
2394 { 0, "no-daemon", &become_daemon, op_boolfalse, 1 },
2396 { 0, "no-check-proportion", &nocheck_thresh_pct, op_double },
2397 { 0, "no-check-filter", &nocheck_decay_articles, op_double },
2399 { 0, "reconnect-interval", &reconnect_delay_periods, op_periods_rndup },
2400 { 0, "flush-retry-interval", &flushfail_retry_periods, op_periods_rndup },
2401 { 0, "connection-timeout", &connection_timeout, op_seconds },
2402 { 0, "inndcomm-timeout", &inndcomm_flush_timeout, op_seconds },
2405 int main(int argc, char **argv) {
2411 if (*arg != '-') break;
2412 if (!strcmp(arg,"--")) { arg= *++argv; break; }
2414 while ((a= *++arg)) {
2418 char *equals= strchr(arg,'=');
2419 int len= equals ? (equals - arg) : strlen(arg);
2420 for (o=options; o->long; o++)
2421 if (strlen(o->long) == len && !memcmp(o->long,arg,len))
2423 badusage("unknown long option --%s",arg);
2426 if (equals) badusage("option --%s does not take a value",o->long);
2428 } else if (equals) {
2432 if (!arg) badusage("option --%s needs a value",o->long);
2435 break; /* eaten the whole argument now */
2437 for (o=options; o->long; o++)
2440 badusage("unknown short option -%c",a);
2447 if (!arg) badusage("option -%c needs a value",o->short);
2450 break; /* eaten the whole argument now */
2455 if (!arg) badusage("need site name argument");
2461 if (*++argv) badusage("too many non-option arguments");
2463 if (nocheck_thresh_pct < 0 || nocheck_thresh_pct > 100)
2464 badusage("nocheck threshold percentage must be between 0..100");
2465 nocheck_thresh= nocheck_thresh_pct * 0.01;
2467 if (nocheck_decay_articles < 0.1)
2468 badusage("nocheck decay articles must be at least 0.1");
2469 nocheck_decay= 1 - 1/nocheck_decay_articles;
2472 pathoutgoing= innconf->pathoutgoing;
2473 innconf_read(inndconffile);
2476 feedfile= xasprintf("%s/%s",pathoutgoing,sitename);
2477 else if (!feedfile[0])
2478 badusage("feed filename must be nonempty");
2479 else if (feedfile[strlen(feedfile)-1]=='/')
2480 feedfile= xasprintf("%s%s",feedfile,sitename);
2482 const char *feedfile_forbidden= "?*[~#";
2484 while ((c= *feedfile_forbidden++))
2485 if (strchr(feedfile, c))
2486 badusage("feed filename may not contain metacharacter %c",c);
2488 loop= oop_sys_new();
2489 if (!loop) sysdie("could not create liboop event loop");
2491 if (signal(SIGPIPE, SIG_IGN) == SIG_ERR)
2492 sysdie("could not ignore SIGPIPE");
2494 if (become_daemon) {
2495 for (i=3; i<255; i++)
2496 /* do this now before we open syslog, etc. */
2498 openlog("innduct",LOG_NDELAY|LOG_PID,LOG_NEWS);
2500 int null= open("/dev/null",O_RDWR);
2501 if (null<0) sysdie("failed to open /dev/null");
2507 pid_t child1= xfork("daemonise first fork");
2508 if (child1) _exit(0);
2510 pid_t sid= setsid();
2511 if (sid != child1) sysdie("setsid failed");
2513 pid_t child2= xfork("daemonise second fork");
2514 if (child2) _exit(0);
2519 if (!filemon_init()) {
2520 warn("no file monitoring available, polling");
2521 filepoll_schedule();