+ }
+
+ conn_maybe_write(conn);
+ check_assign_articles();
+ return OOP_CONTINUE;
+}
+
+
+/*========== monitoring of input files ==========*/
+
+static void feedfile_eof(InputFile *ipf) {
+ assert(ipf != main_input_file); /* promised by tailing_try_read */
+ inputfile_reading_stop(ipf);
+
+ if (ipf == flushing_input_file) {
+ assert(sms==sm_SEPARATED || sms==sm_DROPPING);
+ if (main_input_file) inputfile_reading_start(main_input_file);
+ statemc_check_flushing_done();
+ } else if (ipf == backlog_input_file) {
+ statemc_check_backlog_done();
+ } else {
+ abort(); /* supposed to wait rather than get EOF on main input file */
+ }
+}
+
+static InputFile *open_input_file(const char *path) {
+ int fd= open(path, O_RDWR);
+ if (fd<0) {
+ if (errno==ENOENT) return 0;
+ sysfatal("unable to open input file %s", path);
+ }
+ assert(fd>0);
+
+ InputFile *ipf= xmalloc(sizeof(*ipf) + strlen(path) + 1);
+ memset(ipf,0,sizeof(*ipf));
+
+ ipf->fd= fd;
+ strcpy(ipf->path, path);
+
+ return ipf;
+}
+
+static void close_input_file(InputFile *ipf) { /* does not free */
+ assert(!ipf->readable_callback); /* must have had ->on_cancel */
+ assert(!ipf->filemon); /* must have had inputfile_reading_stop */
+ assert(!ipf->rd); /* must have had inputfile_reading_stop */
+ assert(!ipf->inprogress); /* no dangling pointers pointing here */
+ xclose_perhaps(&ipf->fd, "input file ", ipf->path);
+}
+
+
+/*---------- dealing with articles read in the input file ----------*/
+
+static void *feedfile_got_bad_data(InputFile *ipf, off_t offset,
+ const char *data, const char *how) {
+ warn("corrupted file: %s, offset %lu: %s: in %s",
+ ipf->path, (unsigned long)offset, how, sanitise(data));
+ ipf->readcount_err++;
+ if (ipf->readcount_err > max_bad_data_initial +
+ (ipf->readcount_ok+ipf->readcount_blank) / max_bad_data_ratio)
+ die("too much garbage in input file! (%d errs, %d ok, %d blank)",
+ ipf->readcount_err, ipf->readcount_ok, ipf->readcount_blank);
+ return OOP_CONTINUE;
+}
+
+static void *feedfile_read_err(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;
+ assert(ev == OOP_RD_SYSTEM);
+ errno= errnoval;
+ sysdie("error reading input file: %s, offset %lu",
+ ipf->path, (unsigned long)ipf->offset);
+}
+
+static 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; }
+
+ off_t old_offset= ipf->offset;
+ ipf->offset += recsz + 1;
+
+#define X_BAD_DATA(m) return feedfile_got_bad_data(ipf,old_offset,data,m);
+
+ if (ev==OOP_RD_PARTREC)
+ feedfile_got_bad_data(ipf,old_offset,data,"missing final newline");
+ /* but process it anyway */
+
+ if (ipf->skippinglong) {
+ if (ev==OOP_RD_OK) ipf->skippinglong= 0; /* fine now */
+ return OOP_CONTINUE;
+ }
+ if (ev==OOP_RD_LONG) {
+ ipf->skippinglong= 1;
+ X_BAD_DATA("overly long line");
+ }
+
+ if (memchr(data,'\0',recsz)) X_BAD_DATA("nul byte");
+ if (!recsz) X_BAD_DATA("empty line");
+
+ if (data[0]==' ') {
+ if (strspn(data," ") != recsz) X_BAD_DATA("line partially blanked");
+ ipf->readcount_blank++;
+ return OOP_CONTINUE;
+ }
+
+ char *space= strchr(data,' ');
+ int tokenlen= space-data;
+ int midlen= (int)recsz-tokenlen-1;
+ if (midlen <= 2) X_BAD_DATA("no room for messageid");
+ if (space[1]!='<' || space[midlen]!='>') X_BAD_DATA("invalid messageid");
+
+ if (tokenlen != sizeof(TOKEN)*2+2) X_BAD_DATA("token wrong length");
+ memcpy(tokentextbuf, data, tokenlen);
+ tokentextbuf[tokenlen]= 0;
+ if (!IsToken(tokentextbuf)) X_BAD_DATA("token wrong syntax");
+
+ ipf->readcount_ok++;
+
+ art= xmalloc(sizeof(*art) - 1 + midlen + 1);
+ art->state= art_Unchecked;
+ art->midlen= midlen;
+ art->ipf= ipf; ipf->inprogress++;
+ art->token= TextToToken(tokentextbuf);
+ art->offset= old_offset;
+ art->blanklen= recsz;
+ strcpy(art->messageid, space+1);
+ LIST_ADDTAIL(queue, art);
+
+ if (sms==sm_NORMAL && ipf==main_input_file &&
+ ipf->offset >= target_max_feedfile_size)
+ statemc_start_flush("feed file size");
+
+ check_assign_articles();
+ return OOP_CONTINUE;
+}
+
+/*========== tailing input file ==========*/
+
+static void *tailing_rable_call_time(oop_source *loop, struct timeval tv,
+ void *user) {
+ InputFile *ipf= user;
+ return ipf->readable_callback(loop, &ipf->readable,
+ ipf->readable_callback_user);
+}
+
+static void tailing_on_cancel(struct oop_readable *rable) {
+ InputFile *ipf= (void*)rable;
+
+ if (ipf->filemon) filemon_stop(ipf);
+ loop->cancel_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf);
+ ipf->readable_callback= 0;
+}
+
+static void tailing_queue_readable(InputFile *ipf) {
+ /* lifetime of ipf here is OK because destruction will cause
+ * on_cancel which will cancel this callback */
+ loop->on_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf);
+}
+
+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_start(ipf);
+
+ tailing_queue_readable(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->rd);
+ assert(sms==sm_SEPARATED || sms==sm_DROPPING);
+ } else if (ipf==backlog_input_file) {
+ assert(ipf->rd);
+ } else {
+ abort();
+ }
+ }
+ tailing_queue_readable(ipf);
+ return r;
+ }
+}
+
+/*---------- filemon implemented with inotify ----------*/
+
+#if defined(HAVE_SYS_INOTIFY_H) && !defined(HAVE_FILEMON)
+#define HAVE_FILEMON
+
+#include <sys/inotify.h>
+
+static int filemon_inotify_fd;
+static int filemon_inotify_wdmax;
+static InputFile **filemon_inotify_wd2ipf;
+
+struct Filemon_Perfile {
+ int wd;
+};
+
+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) sysfatal("inotify_add_watch %s", ipf->path);
+
+ if (wd >= filemon_inotify_wdmax) {
+ int newmax= wd+2;
+ filemon_inotify_wd2ipf= 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, 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 (isewouldblock(errno)) 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 %d 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("filemon/inotify: inotify_init failed");
+ return 0;
+ }
+ xsetnonblock(filemon_inotify_fd, 1);
+ loop->on_fd(loop, filemon_inotify_fd, OOP_READ, filemon_inotify_readable, 0);
+
+ 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)
+
+struct Filemon_Perfile { int dummy; };
+
+static int filemon_method_init(void) {
+ warn("filemon/dummy: no filemon method compiled in");
+ 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 */
+
+/*---------- filemon generic interface ----------*/
+
+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) {
+ if (ipf && ipf->readable_callback) /* so filepoll() can be naive */
+ ipf->readable_callback(loop, &ipf->readable, ipf->readable_callback_user);
+}
+
+/*---------- interface to start and stop an input file ----------*/
+
+static const oop_rd_style feedfile_rdstyle= {
+ OOP_RD_DELIM_STRIP, '\n',
+ OOP_RD_NUL_PERMIT,
+ OOP_RD_SHORTREC_LONG,
+};
+
+static void inputfile_reading_start(InputFile *ipf) {
+ assert(!ipf->rd);
+ 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->rd);
+
+ int r= oop_rd_read(ipf->rd, &feedfile_rdstyle, MAX_LINE_FEEDFILE,
+ feedfile_got_article,ipf, feedfile_read_err, ipf);
+ if (r) sysdie("unable start reading feedfile %s",ipf->path);
+}
+
+static void inputfile_reading_stop(InputFile *ipf) {
+ assert(ipf->rd);
+ 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:
+ * -8<-
+
+ .=======.
+ ||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[-ABSENT] |
+ | [Flushing] |
+ | main D tail/none |
+ | ================== |
+ | | |
+ | | INNDCOMM FLUSH FAILS ^
+ | |`----------------------->----------. |
+ | | | |
+ | | NO SUCH SITE V |
+ ^ |`--------------->----. ==================== |
+ | | \ FLUSHFAILED[-ABSENT] |
+ | | \ [Moved] |
+ | | FLUSH OK \ main D tail/none |
+ | | open F \ ==================== |
+ | | \ | |
+ | | \ | TIME TO RETRY |
+ | |`------->----. ,---<---'\ `----------------'
+ | | D NONE | | D NONE `----.
+ | V | | V
+ | ============= V V ============
+ | SEPARATED-1 | | DROPPING-1
+ | flsh->rd!=0 | | flsh->rd!=0
+ | [Separated] | | [Dropping]
+ | main F idle | | main none
+ | old D tail | | old D tail
+ | ============= | | ============
+ | | | | install |
+ ^ | EOF ON D | | defer | EOF ON D
+ | V | | V
+ | =============== | | ===============
+ | SEPARATED-2 | | DROPPING-2
+ | flsh->rd==0 | V flsh->rd==0
+ | [Finishing] | | [Dropping]
+ | main F tail | `. main none
+ | old D closed | `. old D closed
+ | =============== V `. ===============
+ | | `. |
+ | | ALL D PROCESSED `. | ALL D PROCESSED
+ | V install defer as backlog `. | install defer
+ ^ | close D `. | close D
+ | | unlink D `. | unlink D
+ | | | |
+ | | V V
+ `----------' ==============
+ DROPPED
+ [Dropped]
+ main none
+ old none
+ some backlog
+ ==============
+ |
+ | ALL BACKLOG DONE
+ |
+ | unlink lock
+ | exit
+ V
+ ==========
+ (ESRCH)
+ [Droppped]
+ ==========
+ * ->8-
+ */
+
+static void startup_set_input_file(InputFile *f) {
+ assert(!main_input_file);
+ main_input_file= f;
+ inputfile_reading_start(f);
+}
+
+static void statemc_lock(void) {
+ int lockfd;
+ struct stat stab, stabf;
+
+ for (;;) {
+ 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;
+ int r= fcntl(lockfd, F_SETLK, &fl);
+ if (r==-1) {
+ if (errno==EACCES || isewouldblock(errno)) {
+ if (quiet_multiple) exit(0);
+ fatal("another duct holds the lockfile");
+ }
+ sysfatal("fcntl F_SETLK lockfile %s", path_lock);
+ }
+
+ xfstat_isreg(lockfd, &stabf, path_lock, "lockfile");
+ int lock_noent;
+ xlstat_isreg(path_lock, &stab, &lock_noent, "lockfile");
+
+ if (!lock_noent && samefile(&stab, &stabf))
+ break;
+
+ xclose(lockfd, "stale lockfile ", path_lock);
+ }
+
+ FILE *lockfile= fdopen(lockfd, "w");
+ if (!lockfile) sysdie("fdopen lockfile");
+
+ int r= ftruncate(lockfd, 0);
+ if (r) sysdie("truncate lockfile to write new info");
+
+ if (fprintf(lockfile, "pid %ld\nsite %s\nfeedfile %s\nfqdn %s\n",
+ (unsigned long)self_pid,
+ sitename, feedfile, remote_host) == EOF ||
+ fflush(lockfile))
+ sysfatal("write info to lockfile %s", path_lock);
+
+ debug("startup: locked");
+}
+
+static void statemc_init(void) {
+ struct stat stabdefer;
+
+ search_backlog_file();
+
+ int defer_noent;
+ xlstat_isreg(path_defer, &stabdefer, &defer_noent, "defer file");
+ if (defer_noent) {
+ debug("startup: ductdefer ENOENT");
+ } else {
+ debug("startup: ductdefer nlink=%ld", (long)stabdefer.st_nlink);
+ switch (stabdefer.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, stabdefer.st_nlink);
+ }
+ }
+
+ struct stat stab_f, stab_d;
+ int noent_f;
+
+ InputFile *file_d= open_input_file(path_flushing);
+ if (file_d) xfstat_isreg(file_d->fd, &stab_d, path_flushing,"flushing file");
+
+ xlstat_isreg(feedfile, &stab_f, &noent_f, "feedfile");
+
+ if (!noent_f && file_d && samefile(&stab_f, &stab_d)) {
+ debug("startup: F==D => Hardlinked");
+ xunlink(feedfile, "feed file (during startup)"); /* => Moved */
+ noent_f= 1;
+ }
+
+ if (noent_f) {
+ debug("startup: F ENOENT => Moved");
+ if (file_d) startup_set_input_file(file_d);
+ spawn_inndcomm_flush("feedfile missing at startup");
+ /* => Flushing, sms:=FLUSHING */
+ } else {
+ if (file_d) {
+ debug("startup: F!=D => Separated");
+ startup_set_input_file(file_d);
+ SMS(SEPARATED, 0, "found both old and current feed files");
+ } else {
+ debug("startup: F exists, D ENOENT => Normal");
+ InputFile *file_f= open_input_file(feedfile);
+ if (!file_f) die("feed file vanished during startup");
+ startup_set_input_file(file_f);
+ SMS(NORMAL, spontaneous_flush_periods, "normal startup");
+ }
+ }
+}
+
+static void statemc_start_flush(const char *why) { /* Normal => Flushing */
+ assert(sms == sm_NORMAL);
+
+ debug("starting flush (%s) (%lu >?= %lu) (%d)",
+ why,
+ (unsigned long)(main_input_file ? main_input_file->offset : 0),
+ (unsigned long)target_max_feedfile_size,
+ sm_period_counter);
+
+ int r= link(feedfile, path_flushing);
+ if (r) sysfatal("link feedfile %s to flushing file %s",
+ feedfile, path_flushing);
+ /* => Hardlinked */
+
+ xunlink(feedfile, "old feedfile link");
+ /* => Moved */
+
+ spawn_inndcomm_flush(why); /* => Flushing FLUSHING */
+}
+
+static void statemc_period_poll(void) {
+ if (!sm_period_counter) return;
+ sm_period_counter--;
+ assert(sm_period_counter>=0);
+
+ if (sm_period_counter) return;
+ switch (sms) {
+ case sm_NORMAL:
+ statemc_start_flush("periodic"); /* Normal => Flushing; => FLUSHING */
+ break;
+ case sm_FLUSHFAILED:
+ spawn_inndcomm_flush("retry"); /* Moved => Flushing; => FLUSHING */