+
+static void *peer_rd_err(oop_source *lp, oop_read *oread, oop_event ev,
+ const char *errmsg, int errnoval,
+ const char *data, size_t recsz, void *conn_v) {
+ Conn *conn= conn_v;
+ connfail(conn, "error receiving from peer: %s", errmsg);
+ return OOP_CONTINUE;
+}
+
+static Article *article_reply_check(Connection *conn, const char *response,
+ int code_indicates_streaming,
+ const char *sanitised_response) {
+ Article *art= LIST_REMHEAD(conn->sent);
+
+ if (!art) {
+ connfail(conn,
+ "peer gave unexpected response when no commands outstanding: %s",
+ sanitised_response);
+ return 0;
+ }
+
+ if (code_indicates_streaming) {
+ assert(!memchr(response, 0, 4)); /* ensured by peer_rd_ok */
+ if (!conn->stream) {
+ connfail("peer gave streaming response code "
+ " to IHAVE or subsequent body: %s", sanitised_response);
+ return 0;
+ }
+ const char *got_mid= response+4;
+ int got_midlen= strcspn(got_mid, " \n\r");
+ if (got_midlen<3 || got_mid[0]!='<' || got_mid[got_midlen-1]!='>') {
+ connfail("peer gave streaming response with syntactically invalid"
+ " messageid: %s", sanitised_response);
+ return 0;
+ }
+ if (got_midlen != art->midlen ||
+ memcmp(got_mid, art->messageid, got_midlen)) {
+ connfail("peer gave streaming response code to wrong article -"
+ " probable synchronisation problem; we offered: %s;"
+ " peer said: %s",
+ art->messageid, sanitised_response);
+ return 0;
+ }
+ } else {
+ if (conn->stream) {
+ connfail("peer gave non-streaming response code to CHECK/TAKETHIS: %s",
+ sanitised_response);
+ return 0;
+ }
+ }
+
+ return art;
+}
+
+static void update_nocheck(int accepted) {
+ accept_proportion *= accept_decay;
+ accept_proportion += accepted;
+ int new_nocheck= accept_proportion >= nocheck_thresh;
+ if (new_nocheck && !nocheck_reported) {
+ notice("entering nocheck mode for the first time");
+ nocheck_reported= 1;
+ } else if (new_nocheck != nockech) {
+ debug("nocheck mode %s", new_nocheck ? "start" : "stop");
+ }
+ nocheck= new_nocheck;
+}
+
+static void article_done(Connection *conn, Article *art, int whichcount) {
+ *count++;
+ art->ipf->counts.articles[art->checked][whichcount]++;
+ if (whichcount == RC_accepted) update_nocheck(1);
+ else if (whichcount == RC_unwanted) update_nocheck(0);
+
+ InputFile *ipf= art->ipf;
+ while (art->blanklen) {
+ static const char spaces[]=
+ " "
+ " "
+ " "
+ " ";
+ int w= art->blanklen; if (w >= sizeof(spaces)) w= sizeof(spaces)-1;
+ int r= pwrite(ipf->fd, spaces, w, art->offset);
+ if (r==-1) {
+ if (errno==EINTR) continue;
+ sysdie("failed to blank entry for %s (length %d at offset %lu) in %s",
+ art->messageid, art->blanklen, art->offset, ipf->path);
+ }
+ assert(r>=0 && r<=w);
+ art->blanklen -= w;
+ art->offset += w;
+ }
+
+ ipf->inprogress--;
+ assert(ipf->inprogress >= 0);
+
+ if (!ipf->inprogress)
+ loop->on_time(loop, OOP_TIME_NOW, statemc_check_input_done, ipf);
+
+ free(art);
+}
+
+static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev,
+ const char *errmsg, int errnoval,
+ const char *data, size_t recsz, void *conn_v) {
+ Conn *conn= conn_v;
+
+ if (ev == OOP_RD_EOF) {
+ connfail(conn, "unexpected EOF from peer");
+ return OOP_CONTINUE;
+ }
+ assert(ev == OOP_RD_OK);
+
+ char *ep;
+ unsigned long code= strtoul(data, &ep, 10);
+ if (ep != data+3 || *ep != ' ' || data[0]=='0') {
+ char sanibuf[100];
+ const char *p= data;
+ char *q= sanibuf;
+ *q++= '`';
+ for (;;) {
+ if (q > sanibuf+sizeof(sanibuf)-8) { strcpy(q,"..."); break; }
+ int c= *p++;
+ if (!c) { *q++= '\''; break; }
+ if (c>=' ' && c<=126 && c!='\\') { *q++= c; continue; }
+ sprintf(q,"\\x%02x",c);
+ q += 4;
+ }
+ connfail(conn, "badly formatted response from peer: %s", sanibuf);
+ return OOP_CONTINUE;
+ }
+
+ if (conn->quitting) {
+ if (code!=205) {
+ connfail(conn, "peer gave failure response to QUIT: %s", sani);
+ return OOP_CONTINUE;
+ }
+ conn close ok;
+ return;
+ }
+
+ Article *art;
+
+#define GET_ARTICLE \
+ art= article_reply_check(conn, data, code_streaming, sani); \
+ if (art) ; else return OOP_CONTINUE /* reply_check has failed the conn */
+
+#define ARTICLE_DEALTWITH(streaming,how) \
+ code_streaming= (streaming) \
+ GET_ARTICLE; \
+ article_done(conn, art, RC_##how); break;
+
+#define PEERBADMSG(m) connfail(conn, m ": %s", sani); return OOP_CONTINUE
+
+ int code_streaming= 0;
+
+ switch (code) {
+
+ case 400: PEERBADMSG("peer stopped accepting articles");
+ case 503: PEERBADMSG("peer timed us out");
+ default: PEERBADMSG("peer sent unexpected message");
+
+ case 435: ARTICLE_DEALTWITH(0,unwanted); /* IHAVE says they have it */
+ case 438: ARTICLE_DEALTWITH(1,unwanted); /* CHECK/TAKETHIS: they have it */
+
+ case 235: ARTICLE_DEALTWITH(0,accepted); /* IHAVE says thanks */
+ case 239: ARTICLE_DEALTWITH(1,accepted); /* TAKETHIS says thanks */
+
+ case 437: ARTICLE_DEALTWITH(0,rejected); /* IHAVE says rejected */
+ case 439: ARTICLE_DEALTWITH(1,rejected); /* TAKETHIS says rejected */
+
+ case 238: /* CHECK says send it */
+ code_streaming= 1;
+ case 335: /* IHAVE says send it */
+ GET_ARTICLE;
+ count_checkedwanted++;
+ LIST_ADDTAIL(conn->queue);
+ if (art->checked) {
+ connfail("peer gave %d response to article body: %s",code, sani);
+ return OOP_CONTINUE;
+ }
+ art->checked= 1;
+ break;
+
+ case 431: /* CHECK or TAKETHIS says try later */
+ code_streaming= 1;
+ case 436: /* IHAVE says try later */
+ GET_ARTICLE;
+ open_defer();
+ if (fprintf(defer, "%s %s\n", TokenToText(art->token), art->messageid) <0
+ || fflush(defer))
+ sysfatal("write to defer file %s",path_defer);
+ article_done(conn, art, RC_deferred);
+ break;
+
+ }
+
+ check_check_work(conn);
+ return OOP_CONTINUE;
+}
+
+
+/*========== monitoring of input files ==========*/
+
+static void feedfile_eof(InputFile *ipf) {
+ assert(ipf != main_input_file); /* promised by tailing_try_read */
+
+ inputfile_tailing_stop(ipf);
+
+ if (ipf == backlog_input_file) {
+ assert(ipf->fd >= 0);
+ if (close(ipf->fd)) sysdie("could not close backlog file %s", ipf->path);
+ ipf->fd= -1;
+ return;
+ }
+
+ assert(ipf == flushing_input_file);
+
+ inputfile_tailing_stop(ipf);
+ assert(ipf->fd >= 0);
+ if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path);
+ ipf->fd= -1;
+
+ assert(sms==sm_SEPARATED || sms==sm_DROPPING);
+
+ if (main_input_file)
+ inputfile_tailing_start(main_input_file);
+}
+
+static InputFile *open_input_file(const char *path) {
+ int fd= open(path, O_RDONLY);
+ if (fd<0) {
+ if (errno==ENOENT) return 0;
+ sysfatal("unable to open input file %s", path);
+ }
+
+ InputFile *ipf= xmalloc(sizeof(*ipf) + strlen(path) + 1);
+ memset(ipf,0,sizeof(*ipf));
+
+ ipf->readable.on_readable= tailing_on_readable;
+ ipf->readable.on_cancel= tailing_on_cancel;
+ ipf->readable.try_read= tailing_try_read;
+
+ ipf->fd= fd;
+ strcpy(ipf->path, path);
+
+ return ipf;
+}
+
+static void close_input_file(InputFile *ipf) {
+ assert(!ipf->readable_callback); /* must have had ->on_cancel */
+ assert(!ipf->filemon); /* must have had inputfile_tailing_stop */
+ assert(!ipf->rd); /* must have had inputfile_tailing_stop */
+ assert(!ipf->inprogress); /* no dangling pointers pointing here */
+
+ if (ipf->fd >= 0)
+ if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path);
+
+ free(ipf);
+}
+
+
+/*---------- dealing with articles read in the input file ----------*/
+
+typedef void *feedfile_got_article(oop_source *lp, oop_read *rd,
+ oop_rd_event ev, const char *errmsg,
+ int errnoval,
+ const char *data, size_t recsz,
+ void *ipf_v) {
+ InputFile *ipf= ipf_v;
+ Article *art;
+ char tokentextbuf[sizeof(TOKEN)*2+3];
+
+ if (!data) { feedfile_eof(ipf); return OOP_CONTINUE; }
+
+ if (data[0] && data[0]!=' ') {
+ char *space= strchr(data,' ');
+ int tokenlen= space-data;
+ int midlen= (int)recsz-tokenlen-1;
+ if (midlen < 0) goto bad_data;
+
+ if (tokenlen != sizeof(TOKEN)*2+2) goto bad_data;
+ memcpy(tokentextbuf, data, tokenlen);
+ tokentextbuf[tokenlen]= 0;
+ if (!IsToken(tokentextbuf)) goto bad_data;
+
+ art= xmalloc(sizeof(*art) - 1 + midlen + 1);
+ art->offset= ipf->offset;
+ art->blanklen= recsz;
+ art->midlen= midlen;
+ art->checked= art->sentbody= 0;
+ art->ipf= ipf; ipf->inprogress++;
+ art->token= TextToToken(tokentextbuf);
+ strcpy(art->messageid, space+1);
+ LIST_ADDTAIL(queue, art);
+ }
+ ipf->offset += recsz + 1;
+
+ if (sms==sm_NORMAL && ipf==main_input_file &&
+ (ipf->offset >= flush_threshold || !until_spontaneous_flush) {
+
+ notice("starting flush (%lu >= %lu)",
+ (unsigned long)ipf->offset, (unsigned long)flush_threshold);
+
+ int r= link(feedfile, duct_path);
+ if (r) sysdie("link feedfile %s to flushing file %s", feedfile, dut_path);
+ /* => Hardlinked */
+
+ xunlink(feedfile, "old feedfile link");
+ /* => Moved */
+
+ spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */
+ }
+
+ check_master_queue();
+}
+
+
+/*========== tailing input file ==========*/
+
+static void filemon_start(InputFile *ipf) {
+ assert(!ipf->filemon);
+
+ ipf->filemon= xmalloc(sizeof(*ipf->filemon));
+ memset(ipf->filemon, 0, sizeof(*ipf->filemon));
+ filemon_method_startfile(ipf, ipf->filemon);
+}
+
+static void filemon_stop(InputFile *ipf) {
+ if (!ipf->filemon) return;
+ filemon_method_stopfile(ipf, ipf->filemon);
+ free(ipf->filemon);
+ ipf->filemon= 0;
+}
+
+static void filemon_callback(InputFile *ipf) {
+ ipf->readable_callback(ipf->readable_callback_user);
+}
+
+static void *tailing_rable_call_time(oop_source *loop, struct timeval tv,
+ void *user) {
+ InputFile *ipf= user;
+ return ipf->readable_callback(ipf->readable_callback_user);
+}
+
+static void on_cancel(struct oop_readable *rable) {
+ InputFile *ipf= (void*)rable;
+
+ if (ipf->filemon) filemon_stopfile(ipf);
+ loop->cancel_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf);
+ ipf->readable_callback= 0;
+}
+
+static int tailing_on_readable(struct oop_readable *rable,
+ oop_readable_call *cb, void *user) {
+ InputFile *ipf= (void*)rable;
+
+ tailing_on_cancel(rable);
+ ipf->readable_callback= cb;
+ ipf->readable_callback_user= user;
+ filemon_startfile(ipf);
+
+ loop->on_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf);
+ return 0;
+}
+
+static ssize_t tailing_try_read(struct oop_readable *rable, void *buffer,
+ size_t length) {
+ InputFile *ipf= (void*)rable;
+ for (;;) {
+ ssize_t r= read(ipf->fd, buffer, length);
+ if (r==-1) {
+ if (errno==EINTR) continue;
+ return r;
+ }
+ if (!r) {
+ if (ipf==main_input_file) {
+ errno=EAGAIN;
+ return -1;
+ } else if (ipf==flushing_input_file) {
+ assert(ipf->fd>=0);
+ assert(sms==sm_SEPARATED || sms==sm_DROPPING);
+ } else if (ipf==backlog_input_file) {
+ assert(ipf->fd>=0);
+ } else {
+ abort();
+ }
+ }
+ return r;
+ }
+}
+
+/*---------- filemon implemented with inotify ----------*/
+
+#if defined(HAVE_INOTIFY) && !defined(HAVE_FILEMON)
+#define HAVE_FILEMON
+
+#include <linux/inotify.h>
+
+static int filemon_inotify_fd;
+static int filemon_inotify_wdmax;
+static InputFile **filemon_inotify_wd2ipf;
+
+typedef struct Filemon_Perfile {
+ int wd;
+} Filemon_Inotify_Perfile;
+
+static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) {
+ int wd= inotify_add_watch(filemon_inotify_fd, ipf->path, IN_MODIFY);
+ if (wd < 0) sysdie("inotify_add_watch %s", ipf->path);
+
+ if (wd >= filemon_inotify_wdmax) {
+ int newmax= wd+2;
+ filemon_inotify_wd= xrealloc(filemon_inotify_wd2ipf,
+ sizeof(*filemon_inotify_wd2ipf) * newmax);
+ memset(filemon_inotify_wd2ipf + filemon_inotify_wdmax, 0,
+ sizeof(*filemon_inotify_wd2ipf) * (newmax - filemon_inotify_wdmax));
+ filemon_inotify_wdmax= newmax;
+ }
+
+ assert(!filemon_inotify_wd2ipf[wd]);
+ filemon_inotify_wd2ipf[wd]= ipf;
+
+ debug("filemon inotify startfile %p wd=%d wdmax=%d",
+ ipf, wd, filemon_inotify_wdmax);
+
+ pf->wd= wd;
+}
+
+static void filemon_method_stopfile(InputFile *ipf, Filemon_Perfile *pf) {
+ int wd= pf->wd;
+ debug("filemon inotify stopfile %p wd=%d", ipf, wd);
+ int r= inotify_rm_watch(filemon_inotify_fd, filemon_inotify_wd);
+ if (r) sysdie("inotify_rm_watch");
+ filemon_inotify_wd2ipf[wd]= 0;
+}
+
+static void *filemon_inotify_readable(oop_source *lp, int fd,
+ oop_event e, void *u) {
+ struct inotify_event iev;
+ for (;;) {
+ int r= read(filemon_inotify_fd, &iev, sizeof(iev));
+ if (r==-1) {
+ if (errno==EAGAIN) break;
+ sysdie("read from inotify master");
+ } else if (r==sizeof(iev)) {
+ assert(iev.wd >= 0 && iev.wd < filemon_inotify_wdmax);
+ } else {
+ die("inotify read %d bytes wanted struct of %d", r, (int)sizeof(iev));
+ }
+ InputFile *ipf= filemon_inotify_wd2ipf[iev.wd];
+ debug("filemon inotify readable read %p wd=%p", iev.wd, ipf);
+ filemon_callback(ipf);
+ }
+ return OOP_CONTINUE;
+}
+
+static int filemon_method_init(void) {
+ filemon_inotify_fd= inotify_init();
+ if (filemon_inotify_fd<0) {
+ syswarn("could not initialise inotify: inotify_init failed");
+ return 0;
+ }
+ set nonblock;
+ loop->on_fd(loop, filemon_inotify_fd, OOP_READ, filemon_inotify_readable);
+
+ debug("filemon inotify init filemon_inotify_fd=%d", filemon_inotify_fd);
+ return 1;
+}
+
+#endif /* HAVE_INOTIFY && !HAVE_FILEMON *//
+
+/*---------- filemon dummy implementation ----------*/
+
+#if !defined(HAVE_FILEMON)
+
+typedef struct Filemon_Perfile { int dummy; } Filemon_Dummy_Perfile;
+
+static int filemon_method_init(void) { return 0; }
+static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) { }
+static void filemon_method_stopfile(InputFile *ipf, Filemon_Perfile *pf) { }
+
+#endif /* !HAVE_FILEMON */
+
+/*---------- interface to start and stop an input file ----------*/
+
+static const oop_rd_style feedfile_rdstyle= {
+ OOP_RD_DELIM_STRIP, '\n',
+ OOP_RD_NUL_FORBID,
+ OOP_RD_SHORTREC_EOF,
+};
+
+static void inputfile_tailing_start(InputFile *ipf) {
+ assert(!ipf->fd);
+ ipf->readable->on_readable= tailing_on_readable;
+ ipf->readable->on_cancel= tailing_on_cancel;
+ ipf->readable->try_read= tailing_try_read;
+ ipf->readable->delete_tidy= 0; /* we never call oop_rd_delete_{tidy,kill} */
+ ipf->readable->delete_kill= 0;
+
+ ipf->readable_callback= 0;
+ ipf->readable_callback_user= 0;
+
+ ipf->rd= oop_rd_new(loop, &ipf->readable, 0,0);
+ assert(ipf->fd);
+
+ int r= oop_rd_read(ipf->rd, &feedfile_rdstyle, MAX_LINE_FEEDFILE,
+ feedfile_got_article,ipf, feedfile_problem,ipf);
+ if (r) sysdie("unable start reading feedfile %s",ipf->path);
+}
+
+static void inputfile_tailing_stop(InputFile *ipf) {
+ assert(ipf->fd);
+ oop_rd_cancel(ipf->rd);
+ oop_rd_delete(ipf->rd);
+ ipf->rd= 0;
+ assert(!ipf->filemon); /* we shouldn't be monitoring it now */
+}
+
+
+/*========== interaction with innd - state machine ==========*/
+
+/* See official state diagram at top of file. We implement
+ * this as follows:
+
+ .=======.
+ ||START||
+ `======='
+ |
+ | open F
+ |
+ | F ENOENT
+ |`---------------------------------------------------.
+ F OPEN OK | |
+ |`---------------- - - - |
+ D ENOENT | D EXISTS see OVERALL STATES diagram |
+ | for full startup logic |
+ ,--------->| |
+ | V |
+ | ============ try to |
+ | NORMAL open D |
+ | [Normal] |
+ | main F tail |
+ | ============ V
+ | | |
+ | | F IS SO BIG WE SHOULD FLUSH, OR TIMEOUT |
+ ^ | hardlink F to D |
+ | [Hardlinked] |
+ | | unlink F |
+ | | our handle onto F is now onto D |
+ | [Moved] |
+ | | |
+ | |<-------------------<---------------------<---------+
+ | | |
+ | | spawn inndcomm flush |
+ | V |
+ | ================ |
+ | FLUSHING |
+ | [Flushing] |
+ | main D tail/none |
+ | ================ |
+ | | |
+ | | INNDCOMM FLUSH FAILS ^
+ | |`----------------------->----------. |
+ | | | |
+ | | NO SUCH SITE V |
+ ^ |`--------------->----. ================ |
+ | | \ FLUSHFAIL |
+ | | \ [Moved] |
+ | | FLUSH OK \ main D tail/none |
+ | | open F \ ================ |
+ | | \ | |
+ | | \ | TIME TO RETRY |
+ |<--------'| ,---<---'\ `----------------'
+ | D NONE | | D NONE \
+ | | | \
+ | | | |
+ | | V |
+ | V | V
+ | ============= | ============
+ | SEPARATED/ | DROPPING/
+ | flsh->fd>=0 | flsh->fd>=0
+ | [Separated] | [Dropping]
+ | main F idle | main none
+ | old D tail | old D tail
+ | ============= | ============
+ | | | |
+ ^ | EOF ON D | | EOF ON D
+ | V | V
+ | =============== | ===============
+ | SEPARATED/ | DROPPING/
+ | flsh->fd==-1 V flsh->fd==-1
+ | [Finishing] | [Dropping]
+ | main F tail `. main none
+ | old D closed `. old D closed
+ | =============== `. ===============
+ | | `. |
+ | | ALL D PROCESSED `. | ALL D PROCESSED
+ | V install defer as backlog `. V install defer as backlog
+ ^ | close D \| close D
+ | | unlink D | unlink D
+ | | | unlink lock
+ | | | exit
+ `----------' V
+ ==========
+ (ESRCH)
+ [Droppped]
+ ==========
+ */
+
+static void statemc_init(void) {
+ struct stat stab, stabf;
+ int noent;
+
+ path_lock= xasprintf("%s_lock", feedfile);
+ path_flushing= xasprintf("%s_flushing", feedfile);
+ path_defer= xasprintf("%s_defer", feedfile);
+ globpat_backlog= xasprintf("%s_backlog*", feedfile);
+
+ for (;;) {
+ int lockfd= open(path_lock, O_CREAT|O_RDWR, 0600);
+ if (lockfd<0) sysfatal("open lockfile %s", path_lock);
+
+ struct flock fl;
+ memset(&fl,0,sizeof(fl));
+ fl.l_type= F_WRLCK;
+ fl.l_whence= SEEK_SET;
+ r= fcntl(lockfd, F_SETLK, &fl);
+ if (r==-1) {
+ if (errno==EACCES || errno==EAGAIN) {
+ if (quiet_if_locked) exit(0);
+ fatal("another duct holds the lockfile");
+ }
+ sysdie("fcntl F_SETLK lockfile %s", path_lock);
+ }
+
+ xfstat_isreg(lockfd, &stabf, "lockfile");
+ xlstat_isreg(path_lock, &stab, &noent, "lockfile");
+
+ if (!noent && samefile(&stab, &stabf))
+ break;
+
+ if (close(lockfd))
+ sysdie("could not close stale lockfile %s", path_lock);
+ }
+ debug("startup: locked");
+
+ search_backlog_file();
+
+ xlstat_isreg(path_defer, &stab, &noent, "defer file");
+ if (noent) {
+ debug("startup: ductdefer ENOENT");
+ } else {
+ debug("startup: ductdefer nlink=%ld", (long)stab.st_nlink);
+ switch (stab.st_nlink==1) {
+ case 1:
+ open_defer(); /* so that we will later close it and rename it */
+ break;
+ case 2:
+ xunlink(path_defer, "stale defer file link"
+ " (presumably hardlink to backlog file)");
+ break;
+ default:
+ die("defer file %s has unexpected link count %d",
+ path_defer, stab.st_nlink);
+ }
+ }
+
+ InputFile *file_d= open_input_file(path_flushing);
+
+ if (file_d) {
+ struct stat stab_f, stab_d;
+
+ xlstat_isreg(feedfile, &stab_f, &noent, "feed file");
+ if (noent) {
+ debug("startup: D exists, F ENOENT => Moved");
+ goto found_moved;
+ }
+
+ debug("startup: F and D both exist");
+
+ xfstat_isreg(file_d->fd, &stab_d, "flushing file");
+
+ if (samefile(&stab_d, &stab_f)) {
+ debug("startup: F==D => Hardlinked");
+ xunlink(path_flushing, "feed file (during startup)");
+ found_moved:
+ debug(" => Moved");
+ startup_set_input_file(file_d);
+ spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */
+ } else {
+ debug("F!=D => Separated");
+ SMS(SEPARATED, 0, "found both old and current feed files");
+ startup_set_input_file(file_d);
+ }
+ } else {
+ debug("startup: D ENOENT => Nothing");
+ SMS(WAITING, open_wait_periods, "no feed file currently exists");
+ }
+}
+
+static void statemc_poll(void) {
+ if (sms==sm_WAITING) { statemc_waiting_poll(); return; }
+
+ if (!sm_period_counter) return;
+ sm_period_counter--;
+ assert(sm_period_counter>=0);
+
+ if (sm_period_counter) return;
+ switch (sms) {
+ case sm_WAITING:
+ fatal("timed out waiting for innd to create feed file %s", feedfile);
+ case sm_FLUSHFAIL:
+ spawn_inndcomm_flush(void);
+ break;
+ default:
+ abort();
+ }
+}
+
+static void statemc_waiting_poll(void) {
+ InputFile *file_f= open_input_file(feedfile);
+ if (!file_f) return;
+ startup_set_input_file(file_d);
+ SMS(NORMAL, 0, "found and opened feed file");
+}
+
+static void startup_set_input_file(InputFile *f) {
+ assert(!main_input_file);
+ main_input_file= f;
+ until_spontaneous_flush= spontaneous_flush_periods;
+ inputfile_tailing_start(f);
+}
+
+static void *statemc_check_input_done(oop_source *lp,
+ struct timeval now, void *ipf_v) {
+ InputFile *ipf= ipf_v;
+ struct stat stab;
+
+ if (ipf->inprogress) return; /* new article in the meantime */
+ if (ipf->fd >= 0); return; /* not had EOF */
+
+ if (ipf == backlog_input_file) {
+ notice_processed(ipf,"backlog file",ipf->path);
+ close_input_file(ipf);
+ if (unlink(ipf->path)) {
+ if (errno != ENOENT)
+ sysdie("could not unlink processed backlog file %s", ipf->path);
+ warn("backlog file %s vanished while we were reading it"
+ " so we couldn't remove it (but it's done now, anyway)",
+ ipf->path);
+ }
+ backlog_input_file= 0;
+ search_backlog_file();
+ return;
+ }
+
+ assert(ipf == flushing_input_file);
+ assert(sms==sm_SEPARATED || sms==sm_DROPPING);
+
+ notice_processed(ipf,"feed file",0);
+
+ close_defer();
+
+ xunlink(path_flushing, "old flushing file");
+
+ if (sms==sm_DROPPING) {
+ if (search_backlog_file()) {
+ debug("feed dropped but still backlogs to process");
+ return;
+ }
+ notice("feed dropped and our work is complete");
+ xunlink(path_lock, "lockfile for old feed");
+ exit(0);
+ }
+
+ open_defer();
+
+ close_input_file(flushing_input_file);
+ flushing_input_file= 0;
+
+ notice("flush complete");
+ SMS(NORMAL, 0, "flush complete");
+}
+
+static void statemc_setstate(StateMachineState newsms, int periods,
+ const char *forlog, const char *why) {
+ sms= newsms;
+ sm_period_counter= periods;
+ if (periods) {
+ info("%s[%d] %s",periods,forlog,why);
+ } else {
+ info("%s %s",forlog,why);
+ }
+}
+
+/*---------- defer and backlog files ----------*/
+
+static void open_defer(void) {
+ struct stat stab;
+
+ if (defer) return;
+
+ defer= fopen(path_defer, "a+");
+ if (!defer) sysfatal("could not open defer file %s", path_defer);
+
+ /* truncate away any half-written records */
+
+ xfstat_isreg(fileno(defer), &stab, "newly opened defer file");
+
+ if (stab.st_size > LONG_MAX)
+ die("defer file %s size is far too large", path_defer);
+
+ if (!stab.st_size)
+ return;
+
+ long orgsize= stab.st_size;
+ long truncto= stab.st_size;
+ for (;;) {
+ if (!truncto) break; /* was only (if anything) one half-truncated record */
+ if (fseek(defer, truncto-1, SEEK_SET) < 0)
+ sysdie("seek in defer file %s while truncating partial", path_defer);
+
+ r= getc(defer);
+ if (r==EOF) {
+ if (ferror(defer))
+ sysdie("failed read from defer file %s", path_defer);
+ else
+ die("defer file %s shrank while we were checking it!", path_defer);
+ }
+ if (r=='\n') break;
+ truncto--;
+ }
+
+ if (stab.st_size != truncto) {
+ warn("truncating half-record at end of defer file %s -"
+ " shrinking by %ld bytes from %ld to %ld",
+ path_defer, orgsize - truncto, orgsize, truncto);
+
+ if (fflush(defer))
+ sysfatal("could not flush defer file %s", path_defer);
+ if (ftruncate(fileno(defer), truncto))
+ sysdie("could not truncate defer file %s", path_defer);
+
+ } else {
+ info("continuing existing defer file %s (%ld bytes)",
+ path_defer, orgsize);
+ }
+ if (fseek(defer, truncto, SEEK_SET))
+ sysdie("could not seek to new end of defer file %s", path_defer);
+}
+
+static void close_defer(void) {
+ if (!defer)
+ return;
+
+ xfstat(fileno(defer), &stab, "defer file");
+
+ if (fclose(defer)) sysfatal("could not close defer file %s", path_defer);
+ defer= 0;
+
+ char *backlog= xasprintf("%s_backlog_%lu.%lu", feedfile,
+ (unsigned long)now.tv_sec,
+ (unsigned long)stab.st_ino);
+ if (link(path_defer, path_backlog))
+ sysfatal("could not install defer file %s as backlog file %s",
+ path_defer, backlog);
+ if (unlink(path_defer))
+ sysdie("could not unlink old defer link %s to backlog file %s",
+ path_defer, backlog);
+
+ if (until_backlog_nextscan < 0 ||
+ until_backlog_nextscan > backlog_retry_minperiods + 1)
+ until_backlog_nextscan= backlog_retry_minperiods + 1;
+}
+
+static void poll_backlog_file(void) {
+ if (until_backlog_nextscan < 0) return;
+ if (until_backlog_nextscan-- > 0) return;
+ search_backlog_file();
+}
+
+static int search_backlog_file(void) {
+ /* returns non-0 iff there are any backlog files */
+
+ glob_t gl;
+ int r;
+ struct stat stab;
+ const char *oldest_path=0;
+ time_t oldest_mtime, now;
+
+ if (backlog_input_file) return 3;
+
+ try_again:
+
+ r= glob(globpat_backlog, GLOB_ERR|GLOB_MARK|GLOB_NOSORT, 0, &gl);
+
+ switch (r) {
+ case GLOB_ABORTED:
+ sysdie("failed to expand backlog pattern %s", globpat_backlog);
+ case GLOB_NOSPACE:
+ die("out of memory expanding backlog pattern %s", globpat_backlog);
+ case 0:
+ for (i=0; i<gl.gl_pathc; i++) {
+ const char *path= gl.gl_pathv[i];
+
+ if (strchr(path,'#') || strchr(path,'~')) {
+ debug("backlog file search skipping %s", path);
+ continue;
+ }
+ r= stat(path, &stab);
+ if (r) {
+ syswarn("failed to stat backlog file %s", path);
+ continue;
+ }
+ if (!S_ISREG(stab.st_mode)) {
+ warn("backlog file %s is not a plain file (or link to one)", path);
+ continue;
+ }
+ if (!oldest_path || stab.st_mtime < oldest_mtime) {
+ oldest_path= path;
+ oldest_mtime= stab.st_mtime;
+ }
+ }
+ case GLOB_NOMATCH: /* fall through */
+ break;
+ default:
+ sysdie("glob expansion of backlog pattern %s gave unexpected"
+ " nonzero (error?) return value %d", globpat_backlog, r);
+ }
+
+ globfree(&gl);
+
+ if (!oldest_path) {
+ debug("backlog scan: none");
+ until_backlog_nextscan= backlog_spontaneous_rescan_periods;
+ return 0;
+ }
+
+ now= time(); if (now==-1) sysdie("time(2) failed");
+ double age= difftime(now, oldest_mtime);
+ long age_deficiency= (backlog_retry_minperiods * PERIOD_SECONDS) - age;
+
+ if (age_deficiency <= 0) {
+ debug("backlog scan: found age=%f deficiency=%ld oldest=%s",
+ age, age_deficiency, oldest_path);
+
+ backlog_input_file= open_input_file(oldest_path);
+ if (!backlog_input_file) {
+ warn("backlog file %s vanished as we opened it", backlog_input_file);
+ goto try_again;
+ }
+ inputfile_tailing_start(backlog_input_file);
+ until_backlog_nextscan= -1;
+ return 1;
+ }
+
+ until_backlog_nextscan= age_deficiency / PERIOD_SECONDS;
+
+ if (backlog_spontaneous_rescan_periods >= 0 &&
+ until_backlog_nextscan > backlog_spontaneous_rescan_periods)
+ until_backlog_nextscan= backlog_spontaneous_rescan_periods;
+
+ debug("backlog scan: young age=%f deficiency=%ld nextscan=%d oldest=%s",
+ age, age_deficiency, until_backlog_nextscan, oldest_path);
+ return 2;
+}
+
+/*========== flushing the feed ==========*/
+
+static pid_t inndcomm_child;
+
+static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) {
+ assert(inndcomm_child);
+ int status= xwaitpid(&inndcomm_child, "inndcomm");
+ loop->cancel_fd(fd);
+ close(fd);
+
+ assert(!flushing_input_file);
+
+ if (WIFEXITED(status)) {
+ switch (WEXITSTATUS(status)) {
+
+ case INNDCOMMCHILD_ESTATUS_FAIL:
+ goto failed;
+
+ case INNDCOMMCHILD_ESTATUS_NONESUCH:
+ warn("feed has been dropped by innd, finishing up");
+ flushing_input_file= main_input_file;
+ main_input_file= 0;
+ SMS(DROPPING, 0, "dropped by innd");
+ return OOP_CONTINUE;
+
+ case 0:
+ flushing_input_file= main_input_file;
+ main_input_file= open_input_file(feedfile);
+ if (!main_input_file)
+ die("flush succeeded but feedfile %s does not exist!", feedfile);
+ until_spontaneous_flush= spontaneous_flush_periods;
+ SMS(SEPARATED, 0, "feed file missing");
+ return OOP_CONTINUE;
+
+ default:
+ goto unexpected_exitstatus;
+
+ }
+ } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) {
+ warn("flush timed out trying to talk to innd");
+ goto failed;
+ } else {
+ unexpected_exitstatus:
+ report_child_status("inndcomm child", status);
+ }
+
+ failed:
+ SMS(FLUSHFAIL, flushfail_retry_periods, "flush failed, will retry");
+}
+
+static void inndcommfail(const char *what) {
+ syswarn("error communicating with innd: %s failed: %s", what, ICCfailure);
+ exit(INNDCOMMCHILD_ESTATUS_FAIL);
+}
+
+void spawn_inndcomm_flush(void) {
+ int pipefds[2];
+
+ assert(sms==sm_NORMAL || sms==sm_FLUSHFAIL);
+ assert(!inndcomm_child);
+
+ if (pipe(pipefds)) sysdie("create pipe for inndcomm child sentinel");
+
+ inndcomm_child= xfork();
+
+ if (!inndcomm_child) {
+ static char flushargv[2]= { sitename, 0 };
+ char *reply;
+
+ close(pipefds[0]);
+
+ alarm(inndcomm_flush_timeout);
+ r= ICCopen(); if (r) inndcommfail("connect");
+ r= ICCcommand('f',flushargv,&reply); if (r<0) inndcommfail("transmit");
+ if (!r) exit(0); /* yay! */
+
+ if (!strcmp(reply, "1 No such site")) exit(INNDCOMMCHILD_ESTATUS_NONESUCH);
+ syswarn("innd ctlinnd flush failed: innd said %s", reply);
+ exit(INNDCOMMCHILD_ESTATUS_FAIL);
+ }
+
+ close(pipefds[1]);
+ int sentinel_fd= pipefds[0];
+ on_fd_read_except(sentinel_fd, inndcomm_event);
+
+ SMS(FLUSHING, 0, "flush is in progress");
+}
+
+/*========== main program ==========*/
+
+static void postfork_inputfile(InputFile *ipf) {
+ if (!ipf) return;
+ assert(ipf->fd >= 0);
+ close(ipf->fd);
+ ipf->fd= -1;
+}
+
+static void postfork_conns(Connection *conn) {
+ while (conn) {
+ close(conn->fd);
+ conn= conn->next;
+ }
+}
+
+static void postfork_stdio(FILE *f) {
+ /* we have no stdio streams that are buffered long-term */
+ if (f) fclose(f);
+}
+
+static void postfork(const char *what) {
+ if (signal(SIGPIPE, SIG_DFL) == SIG_ERR)
+ sysdie("%s child: failed to reset SIGPIPE");
+
+ postfork_inputfile(main_input_file);
+ postfork_inputfile(flushing_input_file);
+ postfork_conns(idle.head);
+ postfork_conns(working.head);
+ postfork_conns(full.head);
+ postfork_stdio(defer);
+}
+
+#define EVERY(what, interval, body) \
+ static const struct timeval what##_timeout = { 5, 0 }; \
+ static void what##_schedule(void); \
+ static void *what##_timedout(oop_source *lp, struct timeval tv, void *u) { \
+ { body } \
+ what##_schedule(); \
+ } \
+ static void what##_schedule(void) { \
+ loop->on_time(loop, what##_timeout, what##_timedout, 0); \
+ }
+
+EVERY(filepoll, {5,0}, {
+ if (main_input_file && main_input_file->readable_callback)
+ filemon_callback(main_input_file);
+});
+
+#define DEBUGF_IPF(wh) " " #wh "=%p/%s:ip=%ld,off=%ld,fd=%d%s" \
+#define DEBUG_IPF(sh) \
+ wh##_input_file, debug_ipf_path(wh##_input_file), \
+ wh##_input_file->inprogress, (long)wh##_input_file->offset, \
+ wh##_input_file->fd, wh##_input_file->rd ? "+" : ""
+static const char *debug_ipf_path(InputFile *ipf) {
+ char *slash= strrchr(ipf->path,'/');
+ return slash ? slash+1 : ipf->path;
+}
+
+EVERY(period, {PERIOD_SECONDS,0}, {
+ debug("PERIOD"
+ " sms=%s queue=%d sm_period_counter=%d"
+ " connect_delay=%d until_spontaneous_flush=%d"
+ " input_files" DEBUGF_IPF(main) DEBUGF_IPF(old) DEBUGF_FMT(flushing)
+ " conns idle=%d working=%d full=%d"
+ " children connecting=%ld inndcomm_child"
+ ,
+ sms_names[sms], queue.count, sm_period_counter,
+ connect_delay, until_spontaneous_flush,
+ DEBUG_IPF(main), DEBUG_IPF(flushing), DEBUG_IPF(flushing),
+ idle.count, working.count, full.count,
+ (long)connecting_child, (long)inndcomm_child
+ );
+ if (connect_delay) connect_delay--;
+ if (until_spontaneous_flush) until_spontaneous_flush--;
+ poll_backlog_file();
+ if (!backlog_input_file) close_defer(); /* want to start on a new backlog */
+ statemc_poll();
+ check_master_queue();
+});
+
+
+/*========== option parsing ==========*/
+
+enum OptFlags {
+ of_seconds= 001000u;
+ of_boolean= 002000u;
+};
+
+typedef struct Option Option;
+typedef void OptionParser(const Option*, const char *val);
+
+struct Option {
+ int short;
+ const char *long;
+ void *store;
+ OptionParser *fn;
+ int noarg;
+};
+
+void op_integer(const Option *o, const char *val) {
+ char *ep;
+ errno= 0;
+ unsigned long ul= strtoul(val,&ep,10);
+ if (*ep || ep==val || errno || ul>INT_MAX)
+ badusage("bad integer value for %s",o->long);
+ int *store= o->store;
+ *store= ul;
+}
+
+void op_double(const Option *o, const char *val) {
+ int *store= o->store;
+ char *ep;
+ errno= 0;
+ *store= strtod(val, &ep);
+ if (*ep || ep==val || errno)
+ badusage("bad floating point value for %s",o->long);
+}
+
+void op_string(const Option *o, const char *val) {
+ char **store= o->store;
+ free(*store);
+ *store= val;
+}
+
+void op_seconds(const Option *o, const char *val) {
+ int *store= o->store;
+ char *ep;
+
+ double v= strtod(val,&ep);
+ if (ep==val) badusage("bad time/duration value for %s",o->long);
+
+ if (!*ep || !strcmp(ep,"s")) unit= 1;
+ else if (!strcmp(ep,"m")) unit= 60;
+ else if (!strcmp(ep,"h")) unit= 3600;
+ else if (!strcmp(ep,"d")) unit= 86400;
+ else badusage("bad units %s for time/duration value for %s",ep,o->long);
+
+ v *= unit;
+ v= ceil(v);
+ if (v > INT_MAX) badusage("time/duration value for %s out of range",o->long);
+ *store= v;
+}
+
+void op_periods_rndup(const Option *o, const char *val) {
+ int *store= o->store;
+ op_seconds(o,val);
+ *store += PERIOD_SECONDS-1;
+ *store /= PERIOD_SECONDS;
+}
+
+void op_periods_booltrue(const Option *o, const char *val) {
+ int *store= o->store;
+ *store= 1;
+}
+void op_periods_boolfalse(const Option *o, const char *val) {
+ int *store= o->store;
+ *store= 0;
+}
+
+static const Option options[]= {
+{ 0, "max-connections", &max_connections op_integer },
+{ 0, "streaming", &try_stream, op_booltrue, 1 },
+{ 0, "no-streaming", &try_stream, op_boolfalse, 1 },
+{'h',"host", &remote_host, op_string },
+{'P',"port", &port op_integer },
+{ 0, "inndconf", &inndconffile, op_string },
+{'f',"feedfile", &feedfile, op_string },
+{'q',"quiet-multiple", &quiet_if_locked, op_booltrue, 1 },
+{ 0, "no-quiet-multiple", &quiet_if_locked, op_boolfalse, 1 },
+{'d',"daemon", &become_daemon, op_booltrue, 1 },
+{ 0, "no-daemon", &become_daemon, op_boolfalse, 1 },
+
+{ 0, "no-check-proportion", &nocheck_thresh_pct, op_double },
+{ 0, "no-check-filter", &nocheck_decay_articles, op_double },
+
+{ 0, "max-queue-size", &max_queue_per_conn op_integer },
+{ 0, "reconnect-interval", &reconnect_delay_periods, op_periods_rndup },
+{ 0, "flush-retry-interval", &flushfail_retry_periods, op_periods_rndup },
+{ 0, "feedfile-open-timeout", &open_wait_periods, op_periods_rndup },
+{ 0, "connection-timeout", &connection_timeout, op_seconds },
+{ 0, "inndcomm-timeout", &inndcomm_flush_timeout, op_seconds },
+};
+
+int main(int argc, char **argv) {
+ const char *arg;
+
+ for (;;) {
+ arg= *++argv;
+ if (!arg) break;
+ if (*arg != '-') break;
+ if (!strcmp(arg,"--")) { arg= *++argv; break; }
+ int a;
+ while ((a= *++arg)) {
+ const Option *o;
+ if (a=='-') {
+ arg++;
+ char *equals= strchr(arg,'=');
+ int len= equals ? (equals - arg) : strlen(arg);
+ for (o=options; o->long; o++)
+ if (strlen(o->long) == len && !memcmp(o->long,arg,len))
+ goto found_long;
+ badusage("unknown long option --%s",arg);
+ found_long:
+ if (o->noarg) {
+ if (equals) badusage("option --%s does not take a value",o->long);
+ arg= 0;
+ } else if (equals) {
+ arg= equals+1;
+ } else {
+ arg= *++argv;
+ if (!arg) badusage("option --%s needs a value",o->long);
+ }
+ o->fn(o, arg);
+ break; /* eaten the whole argument now */
+ }
+ for (o=options; o->long; o++)
+ if (a == o->short)
+ goto found_short;
+ badusage("unknown short option -%c",a);
+ found_short:
+ if (o->noarg) {
+ o->fn(o,0);
+ } else {
+ if (!*++arg) {
+ arg= *++argv;
+ if (!arg) badusage("option -%c needs a value",o->short);
+ }
+ o->fn(o,arg);
+ break; /* eaten the whole argument now */
+ }
+ }
+ }
+
+ if (!arg) badusage("need site name argument");
+ sitename= arg;
+
+ if ((arg= *++argv)) {
+ if (remote_host) badusage("must not specify -h and also host as argument");
+ remote_host= arg;
+ }
+
+ if (*++argv) badusage("too many non-option arguments");
+
+ if (nocheck_thresh_pct < 0 || nocheck_thresh_pct > 100)
+ badusage("nocheck threshold percentage must be between 0..100");
+ nocheck_thresh= nocheck_thresh_pct * 0.01;
+
+ if (nocheck_decay_articles < 0.1)
+ badusage("nocheck decay articles must be at least 0.1");
+ nocheck_decay= 1 - 1/nocheck_decay_articles;
+
+ innconf_read(inndconffile);
+
+ if (!feedfile)
+ feedfile= xasprintf("%s/%s",pathoutgoing,sitename);
+ else if (!feedfile[0])
+ badusage("feed filename must be nonempty");
+ else if (feedfile[strlen(feedfile)-1]=='/')
+ feedfile= xasprintf("%s%s",feedfile,sitename);
+
+ const char *feedfile_forbidden= "?*[~#";
+ int c;
+ while ((c= *feedfile_forbidden++))
+ if (strchr(feedfile, c))
+ badusage("feed filename may not contain metacharacter %c",c);
+
+ if (signal(SIGPIPE, SIG_IGN) == SIG_ERR)
+ sysdie("could not ignore SIGPIPE");
+
+ if (become_daemon) {
+ for (i=3; i<255; i++)
+ /* do this now before we open syslog, etc. */
+ close(i);
+ openlog("innduct",LOG_NDELAY|LOG_PID,LOG_NEWS);
+
+ int null= open("/dev/null",O_RDWR);
+ if (null<0) sysdie("failed to open /dev/null");
+ dup2(null,0);
+ dup2(null,1);
+ dup2(null,2);
+ close(null);
+
+ pid_t child1= xfork("daemonise first fork");
+ if (child1) _exit(0);
+
+ pid_t sid= setsid();
+ if (sid != child1) sysdie("setsid failed");
+
+ pid_t child2= xfork("daemonise second fork");
+ if (child2) _exit(0);
+ }
+
+ notice("starting");
+
+ if (!filemon_init()) {
+ warn("no file monitoring available, polling");
+ filepoll_schedule();
+ }
+
+ period_schedule();
+
+ statemc_init();
+
+ loop->execute.
+}