X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ian/git?p=inn-innduct.git;a=blobdiff_plain;f=backends%2Finnduct.c;h=2a22ba20a937b0acdd5d845c43e4bc15f3ff66c3;hp=0d4d05b0ad5dea68d39c8b7460e9d03f98c47b43;hb=8c547d5d95d290391bc645a17d195412d20aea9a;hpb=76eef73f36c7ab8c3cbc7f41ca58310994f7cb96

diff --git a/backends/innduct.c b/backends/innduct.c
index 0d4d05b..2a22ba2 100644
--- a/backends/innduct.c
+++ b/backends/innduct.c
@@ -15,54 +15,49 @@
  * or might be blanked out
  *    <spc><spc><spc><spc>....
  *
- *   site.name_duct.lock       lock preventing multiple ducts
- *                                holder of this lock is "duct"
  * F site.name                 main feed file
  *                                opened/created, then written, by innd
  *                                read by duct
  *                                unlinked by duct
  *                                tokens blanked out by duct when processed
- * D site.name_duct            temporary feed file during flush (or crash)
+ *   site.name_lock            lock preventing multiple ducts
+ *                                to hold lock must open,F_SETLK[W]
+ *                                  and then stat to check that locked file
+ *                                  still has name site.name_lock
+ *                                holder of this lock is "duct"
+ *                                (only) lockholder may remove the lockfile
+ * D site.name_flushing        temporary feed file during flush (or crash)
  *                                hardlink created by duct
  *                                unlinked by duct
- *   site.name_duct.defer      431'd articles, still being written,
+ *   site.name_defer           431'd articles, still being written,
  *                                created, written, used by duct
- *   site.name_backlog.lock    lock taken out by innxmit wrapper
- *                                holder and its child are "xmit"
- *   site.name_backlog_<date>.<inum>
- *                             431'd articles, ready for innxmit
+ *
+ *   site.name_backlog.<date>.<inum>
+ *                             431'd articles, ready for innxmit or duct
  *                                created (link/mv) by duct
- *                                read by xmit
- *                                unlinked by xmit
- *   site.name_backlog_<letters> eg
- *   site.name_backlog_manual
+ *   site.name_backlog<anything-else>  (where <anything-else> does not
+ *                                      contain '#' or '~') eg
+ *   site.name_backlog.manual
  *                             anything the sysadmin likes (eg, feed files
  *                             from old feeds to be merged into this one)
  *                                created (link/mv) by admin
- *                                read by xmit
- *                                unlinked by xmit
-
-
-
-
- BLATHER ABOUT NOT USING INNXMIT
-
- scan for backlog files
- take "old enough" backlog files one at a time and reprocess them
- so need another input file - the current backlog file
- inhibit scanning for backlog files only when
-   - last file not "old enough"
-   - no backlog files found and we haven't made one
-
- also we have to start a new backlog file every (some interval)
-
- TODO for backlog file inputs
- - all code to search for and open these files
- - write proper algorithm comment
-
+ *                                may be symlinks (in which case links
+ *                                may be written through, but only links
+ *                                will be removed.
+ *
+ *                             It is safe to remove backlog files manually,
+ *                             if it's desired to throw away the backlog.
+ *
+ * Backlog files are also processed by innduct.  We find the oldest
+ * backlog file which is at least a certain amount old, and feed it
+ * back into our processing.  When every article in it has been read
+ * and processed, we unlink it and look for another backlog file.
+ *
+ * If we don't have a backlog file that we're reading, we close the
+ * defer file that we're writing and make it into a backlog file at
+ * the first convenient opportunity.
 
 
- 
    OVERALL STATES:
 
   	    					                START
@@ -174,6 +169,7 @@ static int connection_setup_timeout=200, port=119, try_stream=1;
 static int inndcomm_flush_timeout=100, quiet_if_locked=0;
 static const char *remote_host;
 static int reconnect_delay_periods, flushfail_retry_periods, open_wait_periods;
+static int backlog_retry_minperiods, backlog_spontaneous_rescan_periods;
 static const char *inndconffile;
 
 static double accept_proportion;
@@ -300,7 +296,7 @@ static int nconns;
 static LIST(Conn) idle, working, full;
 static LIST(Article) *queue;
 
-static char *path_ductlock, *path_duct, *path_ductdefer;
+static char *path_lock, *path_flushing, *path_defer;
 
 #define SMS(newstate, periods, why) \
    (statemc_setstate(sm_##newstate,(periods),#newstate,(why)))
@@ -377,6 +373,42 @@ static int xwaitpid(pid_t *pid, const char *what) {
   return status;
 }
 
+static void check_isreg(const struct stat *stab, const char *path,
+			const char *what) {
+  if (!S_ISREG(stab->st_mode))
+    die("%s %s not a plain file (mode 0%lo)",
+	what, path, (unsigned long)stab->st_mode);
+}
+
+static void xfstat(int fd, struct stat *stab_r, const char *what) {
+  int r= fstab(path, stab);
+  if (r) sysdie("could not fstat %s %s", what, path);
+}
+
+static void xfstat_isreg(int fd, struct stat *stab_r, const char *what) {
+  xfstat(fd, stab_r, what);
+  check_isreg(stab, path, what);
+}
+
+static void xlstat_isreg(const char *path, struct stat *stab,
+			 int *enoent_r /* 0 means ENOENT is fatal */
+			 const char *what) {
+  int r= lstat(path, stab);
+  if (r) {
+    if (errno==ENOENT && enoent_r) { *enoent_r=1; return; }
+    sysdie("could not lstat %s %s", what, path);
+  }
+  if (enoent_r) *enoent_r= 0;
+  check_isreg(stab, path, what);
+}
+
+static int samefile(const struct stat *a, const struct stat *b) {
+  assert(S_ISREG(a->st_mode));
+  assert(S_ISREG(b->st_mode));
+  return (a->st_ino == b->st_ino &&
+	  a->st_dev == b->st_dev);
+}
+
 /*========== logging ==========*/
 
 static void logcore(int sysloglevel, const char *fmt, ...)
@@ -645,8 +677,6 @@ static void connect_start() {
 /*========== overall control of article flow ==========*/
 
 static void check_master_queue(void) {
-  try reading current feed file;
-
   if (!queue.count)
     return;
 
@@ -852,6 +882,14 @@ static const oop_rd_style peer_rd_style= {
   OOP_RD_SHORTREC_FORBID
 };
 
+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) {
@@ -941,14 +979,6 @@ static void article_done(Connection *conn, Article *art, int whichcount) {
   free(art);
 }
 
-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 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) {
@@ -1035,9 +1065,10 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev,
     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_ductdefer);
+      sysfatal("write to defer file %s",path_defer);
     article_done(conn, art, RC_deferred);
     break;
 
@@ -1070,7 +1101,7 @@ static void feedfile_eof(InputFile *ipf) {
   ipf->fd= -1;
 
   assert(sms==sm_SEPARATED || sms==sm_DROPPING);
-  
+
   if (main_input_file)
     inputfile_tailing_start(main_input_file);
 }
@@ -1151,7 +1182,7 @@ typedef void *feedfile_got_article(oop_source *lp, oop_read *rd,
 	   (unsigned long)ipf->offset, (unsigned long)flush_threshold);
 
     int r= link(feedfile, duct_path);
-    if (r) sysdie("link feedfile %s to ductfile %s", feedfile, dut_path);
+    if (r) sysdie("link feedfile %s to flushing file %s", feedfile, dut_path);
     /* => Hardlinked */
 
     r= unlink(feedfile);
@@ -1354,13 +1385,14 @@ static void inputfile_tailing_start(InputFile *ipf) {
 
 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 ==========*/
+/*========== interaction with innd - state machine ==========*/
 
 /* See official state diagram at top of file.  We implement
  * this as follows:
@@ -1371,9 +1403,9 @@ static void inputfile_tailing_stop(InputFile *ipf) {
 	    poll for F
 	   ================
 		|
-		|     TIMEOUT
+		|     TIMEOUT and no defer, no backlog
 		|`--------------------------.
-	   	|                           | install defer as backlog
+	   	|                           |
                 | OPEN F SUCCEEDS           | exit
      ,--------->|                           V
      |          V                         =========
@@ -1437,83 +1469,62 @@ static void inputfile_tailing_stop(InputFile *ipf) {
      |          V install defer as backlog       V install defer as backlog
      ^          | close D                        | close D
      |          | unlink D                       | unlink D
-     |          | start new defer                | exit
-     |          |                                V
-     `----------'                            ==========
+     |          |                                | exit
+     `----------'                                V
+                                             ==========
                                               (ESRCH)
                                              [Droppped]
                                              ==========
  */
 
-static void open_defer(void) {
-  struct stat stab;
-
-  assert(!defer);
-  defer= fopen(path_ductdefer, "a+");
-  if (!defer) sysfatal("could not open defer file %s", path_ductdefer);
-
-  /* truncate away any half-written records */
-
-  r= fstat(fileno(defer), &stab);
-  if (r) sysdie("could not stat newly opened defer file %s", path_ductdefer);
-
-  if (stab.st_size > LONG_MAX)
-    die("defer file %s size is far too large", path_ductdefer);
+static void statemc_init(void) {
+  struct stat stab, stabf;
+  int noent;
 
-  if (!stab.st_size)
-    return;
+  path_lock=        xasprintf("%s_lock",      feedfile);
+  path_flushing=    xasprintf("%s_flushing",  feedfile);
+  path_defer=       xasprintf("%s_defer",     feedfile);
+  globpat_backlog=  xasprintf("%s_backlog*",  feedfile);
 
-  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_ductdefer);
-
-    r= getc(defer);
-    if (r==EOF) {
-      if (ferror(defer))
-	sysdie("failed read from defer file %s", path_ductdefer);
-      else
-	die("defer file %s shrank while we were checking it!", path_ductdefer);
+    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);
     }
-    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_ductdefer, orgsize - truncto, orgsize, truncto);
+    xfstat_isreg(lockfd, &stabf, "lockfile");
+    xlstat_isreg(path_lock, &stab, &noent, "lockfile");
 
-    if (fflush(defer))
-      sysfatal("could not flush defer file %s", path_ductdefer);
-    if (ftruncate(fileno(defer), truncto))
-      sysdie("could not truncate defer file %s", path_ductdefer);
+    if (!noent && samefile(&stab, &stabf))
+      break;
 
-  } else {
-    info("continuing existing defer file %s (%ld bytes)",
-	 path_ductdefer, orgsize);
+    if (close(lockfd))
+      sysdie("could not close stale lockfile %s", path_lock);
   }
-  if (fseek(defer, truncto, SEEK_SET))
-    sysdie("could not seek to new end of defer file %s", path_ductdefer);
-}
-
-static void statemc_init(void) {
-  struct stat stab;
+  debug("startup: locked");
 
-  path_ductlock=  xasprintf("%s_duct.lock",  feedfile);
-  path_duct=      xasprintf("%s_duct",       feedfile);
-  path_ductdefer= xasprintf("%s_duct.defer", feedfile);
+  search_backlog_file();
 
-  if (lstat(path_ductdefer, &stab)) {
-    if (errno!=ENOENT) sysdie("could not check defer file %s", path_defer);
+  xlstat_isreg(path_defer, &stab, &noent, "defer file");
+  if (noent) {
+    debug("startup: ductdefer ENOENT");
   } else {
-    if (!S_ISREG(stab.st_mode))
-      die("defer file %s not a plain file (mode 0%lo)",
-	  path_defer, (unsigned long)stab.st_mode);
+    debug("startup: ductdefer nlink=%ld", (long)stab.st_nlink);
     switch (stab.st_nlink==1) {
-    case 1: /* ok */ break;
+    case 1:
+      open_defer(); /* so that we will later close it and rename it */
+      break;
     case 2:
       if (unlink(path_defer))
 	sysdie("could not unlink stale defer file link %s (presumably"
@@ -1524,56 +1535,38 @@ static void statemc_init(void) {
 	  path_defer, stab.st_nlink);
     }
   }
-  open_defer();
 
-  int lockfd= open(path_ductlock, O_CREAT|O_RDWR, 0600);
-  if (lockfd<0) sysfatal("open lockfile %s", path_ductlock);
-
-  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_ductlock);
-  }
-
-  InputFile *file_d= open_input_file(path_duct);
+  InputFile *file_d= open_input_file(path_flushing);
 
   if (file_d) {
     struct stat stab_f, stab_d;
 
-    r= stat(feedfile, &stab_f);
-    if (r) {
-      if (errno!=ENOENT) sysdie("check feed file %s", feedfile);
-      /* D exists, F ENOENT => Moved */
+    xlstat_isreg(feedfile, &stab_f, &noent, "feed file");
+    if (noent) {
+      debug("startup: D exists, F ENOENT => Moved");
       goto found_moved;
     }
 
-    /* F and D both exist */
+    debug("startup: F and D both exist");
 
-    r= fstat(file_d->fd, &stab_d);
-    if (r) sysdie("check duct file %s", ductfile);
+    xfstat_isreg(file_d->fd, &stab_d, "flushing file");
 
-    if (stab_d.st_ino == stab_f.st_ino &&
-	stab_d.st_dev == stab_f.st_dev) {
-      /* F==D => Hardlinked*/
-      r= unlink(path_duct);
+    if (samefile(&stab_d, &stab_f)) {
+      debug("startup: F==D => Hardlinked");
+      r= unlink(path_flushing);
       if (r) sysdie("unlink feed file %s during startup", feedfile);
     found_moved:
-      /* => Moved */
+      debug(" => Moved");
       startup_set_input_file(file_d);
       spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */
     } else {
-      /* F!=D => Separated */
+      debug("F!=D => Separated");
       SMS(SEPARATED, 0, "found both old and current feed files");
       startup_set_input_file(file_d);
     }
-  } else { /*!file_d*/
+  } else {
+    debug("startup: D ENOENT => Nothing");
+    fixme need to try flushing innd here - needs state diagram changes;
     SMS(WAITING, open_wait_periods, "no feed file currently exists");
   }
 }
@@ -1620,10 +1613,17 @@ static void *statemc_check_input_done(oop_source *lp,
 
   if (ipf == backlog_input_file) {
     notice_processed(ipf,"backlog file",ipf->path);
-    if (unlink(ipf->path))
-      sysdie("could not unlink done backlog file %s", ipf->path);
     close_input_file(ipf);
-    fixme trigger search for new backlog file;
+    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 == old_input_file);
@@ -1631,28 +1631,19 @@ static void *statemc_check_input_done(oop_source *lp,
 
   notice_processed(ipf,"feed file",0);
 
-  r= fstat(fileno(defer), &stab);
-  if (r) sysdie("check defer file %s", path_defer);
-
-  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);
+  close_defer();
 
-  if (unlink(path_duct))
-    sysdie("could not unlink old duct file %s", path_duct);
+  if (unlink(path_flushing))
+    sysdie("could not unlink old flushing file %s", path_flushing);
 
   if (sms==sm_DROPPING) {
-    notice("feed dropped and our work is complete"
-	   " (but check for backlog files)");
+    if (search_backlog_file()) {
+      debug("feed dropped but still backlogs to process");
+      return;
+    }
+    notice("feed dropped and our work is complete");
+    r= unlink(path_lock);
+    if (r) sysdie("unlink lockfile for old feed %s", path_lock);
     exit(0);
   }
 
@@ -1676,6 +1667,178 @@ static void statemc_setstate(StateMachineState newsms, int periods,
   }
 }
 
+/*---------- 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 (backlog_nextscan_periods < 0 ||
+      backlog_nextscan_periods > backlog_retry_minperiods + 1)
+    backlog_nextscan_periods= backlog_retry_minperiods + 1;
+}
+
+static void poll_backlog_file(void) {
+  if (backlog_nextscan_periods < 0) return;
+  if (backlog_nextscan_periods-- > 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");
+    backlog_nextscan_periods= 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();
+    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);
+    backlog_nextscan_periods= -1;
+    return 1;
+  }
+
+  backlog_nextscan_periods= age_deficiency / PERIOD_SECONDS;
+
+  if (backlog_spontaneous_rescan_periods >= 0 &&
+      backlog_nextscan_periods > backlog_spontaneous_rescan_periods)
+    backlog_nextscan_periods= backlog_spontaneous_rescan_periods;
+
+  debug("backlog scan: young age=%f deficiency=%ld nextscan=%d oldest=%s",
+	age, age_deficiency, backlog_nextscan_periods, oldest_path);
+  return 2;
+}
+
 /*========== flushing the feed ==========*/
 
 static pid_t inndcomm_child;
@@ -1690,7 +1853,7 @@ static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) {
 
   if (WIFEXITED(status)) {
     switch (WEXITSTATUS(status)) {
-      
+
     case INNDCOMMCHILD_ESTATUS_FAIL:
       goto failed;
 
@@ -1711,7 +1874,7 @@ static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) {
 
     default:
       goto unexpected_exitstatus;
-      
+
     }
   } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) {
     warn("flush timed out trying to talk to innd");
@@ -1796,7 +1959,6 @@ static void postfork(const char *what) {
   postfork_stdio(defer);
 }
 
-
 #define EVERY(what, interval, body)					     \
   static const struct timeval what##_timeout = { 5, 0 };		     \
   static void what##_schedule(void);					     \
@@ -1808,10 +1970,15 @@ static void postfork(const char *what) {
     loop->on_time(loop, what##_timeout, what##_timedout, 0);		     \
   }
 
-EVERY(filepoll, {5,0}, { check_master_queue(); })
+EVERY(filepoll, {5,0}, {
+  if (main_input_file && main_input_file->readable_callback)
+    filemon_callback(main_input_file);
+});
 
 EVERY(period, {PERIOD_SECONDS,0}, {
   if (connect_delay) connect_delay--;
+  poll_backlog_file();
+  if (!backlog_input_file) close_defer(); /* want to start on a new backlog */
   statemc_poll();
   check_master_queue();
 });
@@ -1863,7 +2030,7 @@ void op_string(const Option *o, const char *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);
 
@@ -1921,7 +2088,7 @@ static const Option options[]= {
 
 int main(int argc, char **argv) {
   const char *arg;
-  
+
   for (;;) {
     arg= *++argv;
     if (!arg) break;
@@ -1980,7 +2147,7 @@ int main(int argc, char **argv) {
   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)
@@ -1989,7 +2156,13 @@ int main(int argc, char **argv) {
     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 glob metacharacter %c",c);
+
   if (signal(SIGPIPE, SIG_IGN) == SIG_ERR)
     sysdie("could not ignore SIGPIPE");
 
@@ -2005,7 +2178,7 @@ int main(int argc, char **argv) {
     dup2(null,1);
     dup2(null,2);
     close(null);
-    
+
     pid_t child1= xfork("daemonise first fork");
     if (child1) _exit(0);