chiark / gitweb /
close idle connections and spot unresponsive ones
[innduct.git] / backends / innduct.c
index 6a6e51a8a2172e1d30a007d2c7c8b1032c234448..47561e397b891ce941cab02f575c3cf3f6cfd5ce 100644 (file)
@@ -1,3 +1,16 @@
+/*
+ * TODO
+ *  - make period size a tuneable
+ *  - check all structs initialised
+ *  - check all fd watches properly undone
+ *  - check all init functions called
+ *  - actually implement badusage
+ *  - options for all options
+ *  - manpage
+ *  - pid, sitename, hostname in lockfile
+ *  - -k kill mode ?
+ */
+
 /*
  * Newsfeeds file entries should look like this:
  *     host.name.of.site[/exclude,exclude,...]\
  * 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)
-
+ *                                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.
+ * -8<-
 
 
    OVERALL STATES:
 
                                                                START
                                                                   |
-                                                             check D, F
-                                                                  |
-                          <--------------------------------------'|
-        Nothing                            F, D both ENOENT       |
-         F: ENOENT                                                       |
-         D: ENOENT                                                       |
-         duct: not not reading anything                           |
-           |                                                     |
-           |`---------------------.                               |
-           |                      | duct times out waiting for F  |
-           V  innd creates F      | duct exits                    |
-           |                      V                               |
-        Noduct                    GO TO Dropped                   |
-         F: innd writing                                         |
-         D: ENOENT                                               |
-         duct: not running or not reading anything                |
-           |                                                     |
-           |                                                     |
-     ,-->--+                   <---------------------------------'|
-     |     |  duct opens F                         F exists       |
-     |     |                                              D ENOENT       |
+     ,-->--.                                                 check F, D
+     |     |                                                      |
+     |     |                                                      |
+     |     |  <----------------<---------------------------------'|
+     |     |                                       F exists       |
+     |     |                                       D ENOENT       |
+     |     |  duct opens F                                        |
      |     V                                                     |
      |  Normal                                                    |
      |   F: innd writing, duct reading                            |
      |   F == D: innd writing, duct reading         both exist    |
      ^     |                                                      |
      |     |  duct unlinks F                                             |
-     |     V                                                     |
-     |  Moved                               <----+------------<--'|
-     |   F: ENOENT                               |  F ENOENT      |
-     |   D: innd writing, duct reading           |  D exists      |
-     |     |                                     |               |
-     |     |  duct requests flush of feed        |               |
-     |     |   (others can too, harmlessly)      |               |
-     |     V                                    |                |
-     |  Flushing                                |                |
-     |   F: ENOENT                              |                |
-     |   D: innd flushing, duct reading                 |                |
-     |     |                                            |                |
-     |     |   inndcomm flush fails                     |                |
-     |     |`-------------------------->---------'               |
-     |     |                                                             |
+     |     |                        <-----------<-------------<--'|
+     |     |                           open D         F ENOENT    |
+     |     |                           if exists                  |
+     |     |                                                      |
+     |     V                        <---------------------.       |
+     |  Moved                                             |       |
+     |   F: ENOENT                                        |       |
+     |   D: innd writing, duct reading; or ENOENT         |       |
+     |     |                                              |      |
+     |     |  duct requests flush of feed                 |      |
+     |     |   (others can too, harmlessly)               |      |
+     |     V                                             |       |
+     |  Flushing                                         |       |
+     |   F: ENOENT                                               |       |
+     |   D: innd flushing, duct; or ENOENT               |       |
+     |     |                                                     |       |
+     |     |   inndcomm flush fails                              |       |
+     |     |`-------------------------->------------------'      |
+     |     |                                                             |
      |     |   inndcomm reports no such site                             |
      |     |`---------------------------------------------------- | -.
-     |     |                                                     |  |
+     |     |                                                     |  |
      |     |  innd finishes writing D, creates F                 |  |
      |     |  inndcomm reports flush successful                          |  |
      |     |                                                             |  |
      |     V                                                     |  |
      |  Separated                                <----------------'  |
      |   F: innd writing                            F!=D             /
-     |   D: duct reading                             both exist     /
+     |   D: duct reading; or ENOENT                  both exist     /
      |     |                                                       /
      |     |  duct gets to the end of D                           /
      |     |  duct opens F too                                          /
    "duct reading" means innduct is reading the file but also
    overwriting processed tokens.
 
+ * ->8- -^L-
  *
+ * rune for printing diagrams:
+
+perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct.c |a2ps -R -B -ops
+
  *
  */
 
+/*============================== PROGRAM ==============================*/
+
+#define _GNU_SOURCE 1
+
+#include "config.h"
+#include "storage.h"
+#include "nntp.h"
+#include "libinn.h"
+#include "inndcomm.h"
+
+#include "inn/list.h"
+#include "inn/innconf.h"
+
+#include <sys/uio.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <sys/stat.h>
+#include <sys/socket.h>
+#include <unistd.h>
+#include <string.h>
+#include <signal.h>
+#include <stdio.h>
+#include <errno.h>
+#include <syslog.h>
+#include <fcntl.h>
+#include <stdarg.h>
+#include <assert.h>
+#include <stdlib.h>
+#include <glob.h>
+#include <time.h>
+#include <math.h>
+
+#include <oop.h>
+#include <oop-read.h>
 
 /*----- general definitions, probably best not changed -----*/
 
 #define INNDCOMMCHILD_ESTATUS_FAIL     6
 #define INNDCOMMCHILD_ESTATUS_NONESUCH 7
 
+#define MAX_LINE_FEEDFILE (NNTP_MSGID_MAXLEN + sizeof(TOKEN)*2 + 10)
+
+#define VA                va_list al;  va_start(al,fmt)
+#define PRINTF(f,a)       __attribute__((__format__(printf,f,a)))
+#define NORET_PRINTF(f,a) __attribute__((__noreturn__,__format__(printf,f,a)))
+
+/*----- doubly linked lists -----*/
+
+#define ISNODE(T)   struct { T *succ, *pred; } node  /* must be at start */
+#define DEFLIST(T)  typedef struct { T *hd, *tl, *tp; int count; } T##List
+
+#define NODE(n) (assert((void*)&(n)->node == &(n)), \
+                (struct node*)&(n)->node)
+
+#define LIST_CHECKCANHAVENODE(l,n) \
+  ((void)((n) == ((l).hd))) /* just for the type check */
+
+#define LIST_ADDSOMEHOW(l,n,list_addsomehow)           \
+ ( LIST_CHECKCANHAVENODE(l,n),                         \
+   list_addsomehow((struct list*)&(l), NODE((n))),     \
+   (void)(l).count++                                   \
+   )
+
+#define LIST_REMSOMEHOW(l,list_remsomehow)     \
+ ( (typeof((l).hd))                            \
+   ( (l).count                                 \
+     ? ( (l).count--,                          \
+        list_remsomehow((struct list*)&(l)) )  \
+     : 0                                       \
+     )                                         \
+   )
+
+
+#define LIST_ADDHEAD(l,n) LIST_ADDSOMEHOW((l),(n),list_addhead)
+#define LIST_ADDTAIL(l,n) LIST_ADDSOMEHOW((l),(n),list_addtail)
+#define LIST_REMHEAD(l) LIST_REMSOMEHOW((l),list_remhead)
+#define LIST_REMTAIL(l) LIST_REMSOMEHOW((l),list_remtail)
+
+#define LIST_HEAD(l) ((typeof((l).hd))(list_head((struct list*)&(l))))
+#define LIST_NEXT(n) ((typeof(n))list_succ(NODE((n))))
+#define LIST_BACK(n) ((typeof(n))list_pred(NODE((n))))
+
+#define LIST_REMOVE(l,n)                       \
+ ( LIST_CHECKCANHAVENODE(l,n),                 \
+   list_remove(NODE((n))),                     \
+   (void)(l).count--                           \
+   )
+
+#define LIST_INSERT(l,n,pred)                                  \
+ ( LIST_CHECKCANHAVENODE(l,n),                                 \
+   LIST_CHECKCANHAVENODE(l,pred),                              \
+   list_insert((struct list*)&(l), NODE((n)), NODE((pred))),   \
+   (void)(l).count++                                           \
+   )
+
+/*----- type predeclarations -----*/
+
+typedef struct Conn Conn;
+typedef struct Article Article;
+typedef struct InputFile InputFile;
+typedef struct XmitDetails XmitDetails;
+typedef struct Filemon_Perfile Filemon_Perfile;
+typedef enum StateMachineState StateMachineState;
+
+DEFLIST(Conn);
+DEFLIST(Article);
+
+/*----- function predeclarations -----*/
+
+static void conn_maybe_write(Conn *conn);
+static void conn_make_some_xmits(Conn *conn);
+static void *conn_write_some_xmits(Conn *conn);
+
+static void xmit_free(XmitDetails *d);
+
+static void statemc_setstate(StateMachineState newsms, int periods,
+                            const char *forlog, const char *why);
+static void statemc_start_flush(const char *why); /* Normal => Flushing */
+static void spawn_inndcomm_flush(const char *why); /* Moved => Flushing */
+
+static void check_master_queue(void);
+static void queue_check_input_done(void);
+
+static void statemc_check_flushing_done(void);
+static void statemc_check_backlog_done(void);
+
+static void postfork(const char *what);
+static void postfork_inputfile(InputFile *ipf);
+
+static void open_defer(void);
+static void close_defer(void);
+static void search_backlog_file(void);
+
+static void inputfile_tailing_start(InputFile *ipf);
+static void inputfile_tailing_stop(InputFile *ipf);
+
+static int filemon_init(void);
+static void filemon_start(InputFile *ipf);
+static void filemon_stop(InputFile *ipf);
+static void filemon_callback(InputFile *ipf);
+
+static void vconnfail(Conn *conn, const char *fmt, va_list al) PRINTF(2,0);
+static void connfail(Conn *conn, const char *fmt, ...)         PRINTF(2,3);
 
 /*----- configuration options -----*/
 
-static char *sitename, *feedfile;
+static const char *sitename, *feedfile, *pathoutgoing;
+static const char *remote_host;
+static int quiet_multiple=0, become_daemon=1;
+
 static int max_connections=10, max_queue_per_conn=200;
+static int target_max_feedfile_size=100000;
+
+static double max_bad_data_ratio= 0.01;
+static int max_bad_data_initial= 30;
+  /* in one corrupt 4096-byte block the number of newlines has
+   * mean 16 and standard deviation 3.99.  30 corresponds to z=+3.5 */
+
 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 inndcomm_flush_timeout=100;
 static int reconnect_delay_periods, flushfail_retry_periods, open_wait_periods;
+static int backlog_retry_minperiods, backlog_spontaneous_rescan_periods;
+static int spontaneous_flush_periods, need_activity_periods;
 static const char *inndconffile;
 
-static double accept_proportion;
 static double nocheck_thresh_pct= 95.0;
 static double nocheck_thresh; /* computed in main from _pct */
 static double nocheck_decay_articles= 100; /* converted to _decay */
 static double nocheck_decay; /* computed in main from _articles */
-static int nocheck, nocheck_reported;
-
-
-/*----- doubly linked lists -----*/
-
-#define ISNODE(T)    T *next, *back;
-#define LIST(T)      struct { T *head, *tail, *tailpred; int count; }
-
-#define NODE(n) ((struct node*)&(n)->head)
-
-#define LIST_ADDHEAD(l,n)                                              \
- (list_addhead((struct list*)&(l), NODE((n))), (void)(l).count++)
-#define LIST_ADDTAIL(l,n)                                              \
- (list_addtail((struct list*)&(l), NODE((n))), (void)(l).count++)
-
-#define LIST_REMHEAD(l)                                                          \
- ((l).count ? ((l).count--, (void*)list_remhead((struct list*)&(l))) : 0)
-#define LIST_REMTAIL(l)                                                          \
- ((l).count ? ((l).count--, (void*)list_remtail((struct list*)&(l))) : 0)
-#define LIST_REMOVE(l,n)                       \
- (list_remove(NODE((n))), (void)(l).count--)
-#define LIST_INSERT(l,n,pred) \
- (list_insert((struct list*)&(l), NODE((n)), NODE((pred))), (void)(l).count++)
 
 
 /*----- statistics -----*/
 
-#define RESULT_COUNTS                          \
-  RC(offered)                                  \
-  RC(sent)                                     \
-  RC(unwanted)                                 \
-  RC(accepted)                                 \
-  RC(rejected)                                 \
-  RC(deferred)
+typedef enum {      /* in queue                 in conn->sent             */
+  art_Unchecked,    /*   not checked, not sent    checking                */
+  art_Wanted,       /*   checked, wanted          sent body as requested  */
+  art_Unsolicited,  /*   -                        sent body without check */
+  art_MaxState
+} ArtState;
+
+#define RESULT_COUNTS(RCS,RCN)                 \
+  RCS(sent)                                    \
+  RCS(accepted)                                        \
+  RCN(unwanted)                                        \
+  RCN(rejected)                                        \
+  RCN(deferred)                                        \
+  RCN(connretry)
+
+#define RCI_TRIPLE_FMT_BASE "%d(id%d+bd%d+nc%d)"
+#define RCI_TRIPLE_VALS_BASE(counts,x)         \
+       counts[art_Unchecked] x                 \
+       + counts[art_Wanted] x                  \
+       + counts[art_Unsolicited] x,            \
+       counts[art_Unchecked] x                 \
+       , counts[art_Wanted] x                  \
+       , counts[art_Unsolicited] x
 
 typedef enum {
-#define RC_INDEX(x) RCI_##x
-  RESULT_COUNTS
+#define RC_INDEX(x) RC_##x,
+  RESULT_COUNTS(RC_INDEX, RC_INDEX)
   RCI_max
 } ResultCountIndex;
 
-typedef struct {
-  int articles[2 /* checked */][RCI_max];
-} Counts;
-
 
 /*----- transmission buffers -----*/
 
 #define CONNIOVS 128
 
 typedef enum {
-  xk_Malloc, xk_Const, xk_Artdata;
+  xk_Malloc, xk_Const, xk_Artdata
 } XmitKind;
 
-typedef struct {
+struct XmitDetails {
   XmitKind kind;
   union {
     char *malloc_tofree;
     ARTHANDLE *sm_art;
   } info;
-} XmitDetails;
+};
 
 
 /*----- core operational data structure types -----*/
 
-struct Article {
-  int midlen;
-  int checked, sentbody;
-  InputFile *ipf;
-  TOKEN token;
-  off_t offset;
-  int blanklen;
-  char messageid[1];
-};
-
-typedef struct {
-  /* This is an instance of struct oop_readable */
+struct InputFile {
+  /* This is also an instance of struct oop_readable */
   struct oop_readable readable; /* first */
   oop_readable_call *readable_callback;
   void *readable_callback_user;
 
   int fd;
-  const char *path; /* ptr copy of path_<foo> or feedfile */
-  struct Filemon_Perfile *filemon;
+  Filemon_Perfile *filemon;
 
   oop_read *rd;
   long inprogress; /* no. of articles read but not processed */
   off_t offset;
-} InputFile;
+  int skippinglong;
 
-typedef enum {
-  sm_WAITING,
-  sm_NORMAL,
-  sm_FLUSHING,
-  sm_FLUSHFAIL,
-  sm_SEPARATED,
-  sm_DROPPING,
-} StateMachineState;
+  int counts[art_MaxState][RCI_max];
+  int readcount_ok, readcount_blank, readcount_err;
+  char path[];
+};
+
+struct Article {
+  ISNODE(Article);
+  ArtState state;
+  int midlen;
+  InputFile *ipf;
+  TOKEN token;
+  off_t offset;
+  int blanklen;
+  char messageid[1];
+};
+
+#define SMS_LIST(X)                            \
+  X(NORMAL)                                    \
+  X(FLUSHING)                                  \
+  X(FLUSHFAILED)                               \
+  X(SEPARATED)                                 \
+  X(DROPPING)                                  \
+  X(DROPPED)
+
+enum StateMachineState {
+#define SMS_DEF_ENUM(s) sm_##s,
+  SMS_LIST(SMS_DEF_ENUM)
+};
+
+static const char *sms_names[]= {
+#define SMS_DEF_NAME(s) #s ,
+  SMS_LIST(SMS_DEF_NAME)
+  0
+};
 
 struct Conn {
   ISNODE(Conn);
-  int fd, max_queue, stream;
-  LIST(Article) queue; /* not yet told peer, or CHECK said send it */
-  LIST(Article) sent; /* offered/transmitted - in xmit or waiting reply */
+  int fd, max_queue, stream, quitting;
+  int since_activity; /* periods */
+  ArticleList waiting; /* not yet told peer */
+  ArticleList priority; /* peer says send it now */
+  ArticleList sent; /* offered/transmitted - in xmit or waiting reply */
   struct iovec xmit[CONNIOVS];
   XmitDetails xmitd[CONNIOVS];
   int xmitu;
@@ -289,35 +450,103 @@ struct Conn {
 
 /*----- operational variables -----*/
 
-static int since_connect_attempt;
-static int nconns;
-static LIST(Conn) idle, working, full;
-static LIST(Article) *queue;
+static oop_source *loop;
+
+static int until_connect;
+static ConnList conns;
+static ArticleList queue;
 
-static char *path_ductlock, *path_duct, *path_ductdefer;
+static char *path_lock, *path_flushing, *path_defer, *globpat_backlog;
 
 #define SMS(newstate, periods, why) \
    (statemc_setstate(sm_##newstate,(periods),#newstate,(why)))
 
 static StateMachineState sms;
 static FILE *defer;
-static InputFile *main_input_file, *old_input_file;
-static int sm_period_counter;
+static InputFile *main_input_file, *flushing_input_file, *backlog_input_file;
+static int sm_period_counter, until_backlog_nextscan;
 
+static double accept_proportion;
+static int nocheck, nocheck_reported;
 
-/*----- function predeclarations -----*/
+/*========== logging ==========*/
 
-static void conn_check_work(Conn *conn);
+static void logcore(int sysloglevel, const char *fmt, ...) PRINTF(2,3);
+static void logcore(int sysloglevel, const char *fmt, ...) {
+  VA;
+  if (become_daemon) {
+    vsyslog(sysloglevel,fmt,al);
+  } else {
+    vfprintf(stderr,fmt,al);
+    putc('\n',stderr);
+  }
+  va_end(al);
+}
 
-static int filemon_init(void);
-static void filemon_setfile(int mainfeed_fd, const char *mainfeed_path);
-static void filemon_callback(void);
+static void logv(int sysloglevel, const char *pfx, int errnoval,
+                const char *fmt, va_list al) PRINTF(5,0);
+static void logv(int sysloglevel, const char *pfx, int errnoval,
+                const char *fmt, va_list al) {
+  char msgbuf[256]; /* NB do not call xvasprintf here or you'll recurse */
+  vsnprintf(msgbuf,sizeof(msgbuf), fmt,al);
+  msgbuf[sizeof(msgbuf)-1]= 0;
+
+  if (sysloglevel >= LOG_ERR && (errnoval==EACCES || errnoval==EPERM))
+    sysloglevel= LOG_ERR; /* run by wrong user, probably */
+
+  logcore(sysloglevel, "<%s>%s: %s%s%s",
+        sitename, pfx, msgbuf,
+        errnoval>=0 ? ": " : "",
+        errnoval>=0 ? strerror(errnoval) : "");
+}
+
+#define diewrap(fn, pfx, sysloglevel, err, estatus)            \
+  static void fn(const char *fmt, ...) NORET_PRINTF(1,2);      \
+  static void fn(const char *fmt, ...) {                       \
+    VA;                                                                \
+    logv(sysloglevel, pfx, err, fmt, al);                      \
+    exit(estatus);                                             \
+  }
+
+#define logwrap(fn, pfx, sysloglevel, err)             \
+  static void fn(const char *fmt, ...) PRINTF(1,2);    \
+  static void fn(const char *fmt, ...) {               \
+    VA;                                                        \
+    logv(sysloglevel, pfx, err, fmt, al);              \
+    va_end(al);                                                \
+  }
+
+diewrap(sysdie,   " critical", LOG_CRIT,    errno, 16);
+diewrap(die,      " critical", LOG_CRIT,    -1,    16);
+
+diewrap(sysfatal, " fatal",    LOG_ERR,     errno, 12);
+diewrap(fatal,    " fatal",    LOG_ERR,     -1,    12);
+
+logwrap(syswarn,  " warning",  LOG_WARNING, errno);
+logwrap(warn,     " warning",  LOG_WARNING, -1);
+
+logwrap(notice,   "",          LOG_NOTICE,  -1);
+logwrap(info,     " info",     LOG_INFO,    -1);
+logwrap(debug,    " debug",    LOG_DEBUG,   -1);
 
-static void statemc_setstate(StateMachineState newsms, int periods,
-                            const char *forlog, const char *why);
 
 /*========== utility functions etc. ==========*/
 
+static char *xvasprintf(const char *fmt, va_list al) PRINTF(1,0);
+static char *xvasprintf(const char *fmt, va_list al) {
+  char *str;
+  int rc= vasprintf(&str,fmt,al);
+  if (rc<0) sysdie("vasprintf(\"%s\",...) failed", fmt);
+  return str;
+}
+static char *xasprintf(const char *fmt, ...) PRINTF(1,2);
+static char *xasprintf(const char *fmt, ...) {
+  VA;
+  char *str= xvasprintf(fmt,al);
+  va_end(al);
+  return str;
+}
+
 static void perhaps_close(int *fd) { if (*fd) { close(*fd); fd=0; } }
 
 static pid_t xfork(const char *what) {
@@ -326,6 +555,7 @@ static pid_t xfork(const char *what) {
   child= fork();
   if (child==-1) sysdie("cannot fork for %s",what);
   if (!child) postfork(what);
+  debug("forked %s %ld", what, (unsigned long)child);
   return child;
 }
 
@@ -360,7 +590,7 @@ static void report_child_status(const char *what, int status) {
 static int xwaitpid(pid_t *pid, const char *what) {
   int status;
 
-  r= kill(*pid, SIGKILL);
+  int r= kill(*pid, SIGKILL);
   if (r) sysdie("cannot kill %s child", what);
 
   pid_t got= waitpid(*pid, &status, WNOHANG);
@@ -371,74 +601,103 @@ static int xwaitpid(pid_t *pid, const char *what) {
   return status;
 }
 
-/*========== logging ==========*/
+static void xunlink(const char *path, const char *what) {
+  int r= unlink(path);
+  if (r) sysdie("can't unlink %s %s", path, what);
+}
 
-static void logcore(int sysloglevel, const char *fmt, ...)
-     __attribute__((printf,2,3))
-static void logcore(int sysloglevel, const char *fmt, ...) {
-  va_list al;
-  va_start(al,fmt);
-  if (become_daemon) {
-    vsyslog(sysloglevel,fmt,al);
-  } else {
-    vfprintf(stderr,fmt,al);
-    putc('\n',stderr);
-  }
+static time_t xtime(void) {
+  time_t now= time(0);
+  if (now==-1) sysdie("time(2) failed");
 }
 
-static void logv(int sysloglevel, const char *pfx, int errnoval,
-                int exitstatus, const char *fmt, va_list al)
-     __attribute__((printf,4,0))
-static void logv(int sysloglevel, const char *pfx, int errnoval,
-                int exitstatus, const char *fmt, va_list al) {
-  char msgbuf[256];
-  vsnprintf(msgbuf,sizeof(msgbuf), fmt,al);
-  msgbuf[sizeof(msgbuf)-1]= 0;
+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);
+}
 
-  if (sysloglevel >= LOG_ERR && (errnoval==EACCES || errnoval==EPERM))
-    sysloglevel= LOG_ERR; /* run by wrong user, probably */
+static void xfstat(int fd, struct stat *stab_r, const char *what) {
+  int r= fstat(fd, stab_r);
+  if (r) sysdie("could not fstat %s", what);
+}
 
-  logcore(sysloglevel, "<%s>%s: %s%s%s",
-        sitename, pfx, msgbuf,
-        errnoval>=0 ? ": " : "",
-        errnoval>=0 ? strerror(errnoval) : "");
+static void xfstat_isreg(int fd, struct stat *stab_r,
+                        const char *path, const char *what) {
+  xfstat(fd, stab_r, what);
+  check_isreg(stab_r, path, what);
 }
 
-#define logwrap(fn, pfx, sysloglevel, errno, estatus)  \
-  static void fn(const char *fmt, ...)                 \
-      __attribute__((printf,1,2));                     \
-  static void fn(const char *fmt, ...) {               \
-    va_list al;                                                \
-    va_start(al,fmt);                                  \
-    logv(sysloglevel, pfx, errno, estatus, fmt, al);   \
+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);
+}
 
-logwrap(sysdie,   " critical", LOG_CRIT,   errno, 16);
-logwrap(die,      " critical", LOG_CRIT,   -1,    16);
+static void setnonblock(int fd, int nonblocking) {
+  int r= fcntl(fd, F_GETFL);  if (r<0) sysdie("setnonblocking fcntl F_GETFL");
+  if (nonblocking) r |= O_NONBLOCK;
+  else r &= ~O_NONBLOCK;
+  r= fcntl(fd, F_SETFL, r);  if (r<0) sysdie("setnonblocking fcntl F_SETFL");
+}
 
-logwrap(sysfatal, " fatal",    LOG_ERR,    errno, 12);
-logwrap(fatal,    " fatal",    LOG_ERR,    -1,    12);
+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);
+}
 
-logwrap(syswarn,  " warning",  LOG_WARN,   errno, 0);
-logwrap(warn,     " warning",  LOG_WARN,   -1,    0);
+static char *sanitise(const char *input) {
+  static char sanibuf[100]; /* returns pointer to this buffer! */
 
-logwrap(notice,   "",          LOG_NOTICE, -1,    0);
-logwrap(info,     " info",     LOG_INFO,   -1,    0);
+  const char *p= input;
+  char *q= sanibuf;
+  *q++= '`';
+  for (;;) {
+    if (q > sanibuf+sizeof(sanibuf)-8) { strcpy(q,"'.."); break; }
+    int c= *p++;
+    if (!c) { *q++= '\''; *q=0; break; }
+    if (c>=' ' && c<=126 && c!='\\') { *q++= c; continue; }
+    sprintf(q,"\\x%02x",c);
+    q += 4;
+  }
+  return sanibuf;
+}
 
+static int isewouldblock(int errnoval) {
+  return errnoval==EWOULDBLOCK || errnoval==EAGAIN;
+}
 
 /*========== making new connections ==========*/
 
+static void conn_dispose(Conn *conn) {
+  if (!conn) return;
+  perhaps_close(&conn->fd);
+  free(conn);
+  until_connect= reconnect_delay_periods;
+}
+
 static int connecting_sockets[2]= {-1,-1};
 static pid_t connecting_child;
 
 static void connect_attempt_discard(void) {
   if (connecting_sockets[0])
-    cancel_fd(connecting_sockets[0]);
+    cancel_fd_read_except(connecting_sockets[0]);
 
   perhaps_close(&connecting_sockets[0]);
   perhaps_close(&connecting_sockets[1]);
 
   if (connecting_child) {
+    int r= kill(connecting_child, SIGTERM);
+    if (r) syswarn("failed to kill connecting child");
     int status= xwaitpid(&connecting_child, "connect");
 
     if (!(WIFEXITED(status) ||
@@ -460,7 +719,7 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) {
   conn= xmalloc(sizeof(*conn));
   memset(conn,0,sizeof(*conn));
 
-  DECL_MSG_CMSG(msg);
+  PREP_DECL_MSG_CMSG(msg);
   struct cmsghdr *h= 0;
   ssize_t rs= recvmsg(fd, &msg, MSG_DONTWAIT);
   if (rs >= 0) h= CMSG_FIRSTHDR(&msg);
@@ -471,11 +730,11 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) {
       assert(got==connecting_child);
       connecting_child= 0;
       if (WIFEXITED(status) &&
-         (WEXITSTATUS(status) != 0
+         (WEXITSTATUS(status) != 0 &&
           WEXITSTATUS(status) != CONNCHILD_ESTATUS_STREAM &&
           WEXITSTATUS(status) != CONNCHILD_ESTATUS_NOSTREAM)) {
        /* child already reported the problem */
-      } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALARM) {
+      } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) {
        warn("connect: connection attempt timed out");
       } else if (!WIFEXITED(status)) {
        report_child_status("connect", status);
@@ -485,7 +744,7 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) {
       /* child is still running apparently, report the socket problem */
       if (rs < 0)
        syswarn("connect: read from child socket failed");
-      else if (e == OOP_EXCEPTIONN)
+      else if (e == OOP_EXCEPTION)
        warn("connect: unexpected exception on child socket");
       else
        warn("connect: unexpected EOF on child socket");
@@ -500,13 +759,14 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) {
   }
   CHK(level, SOL_SOCKET);
   CHK(type,  SCM_RIGHTS);
-  CHK(len,   CMSG_LEN(sizeof(conn-b>fd)));
+  CHK(len,   CMSG_LEN(sizeof(conn->fd)));
 #undef CHK
 
-  if (CMSG_NXTHDR,&msg,h) { die("connect: child sent many cmsgs"); goto x; }
+  if (CMSG_NXTHDR(&msg,h)) { die("connect: child sent many cmsgs"); goto x; }
 
   memcpy(&conn->fd, CMSG_DATA(h), sizeof(conn->fd));
 
+  int status;
   pid_t got= waitpid(connecting_child, &status, 0);
   if (got==-1) sysdie("connect: real wait for child");
   assert(got == connecting_child);
@@ -521,31 +781,34 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) {
     fatal("connect: child gave unexpected exit status %d", es);
   }
 
-  set nonblocking;
-
   /* Phew! */
-  LIST_ADDHEAD(idle, conn);
-  notice(CN "connected %s", conn->fd, conn->stream ? "streaming" : "plain");
+  setnonblock(conn->fd, 1);
+  conn->max_queue= conn->stream ? max_queue_per_conn : 1;
+  LIST_ADDHEAD(conns, conn);
+  notice("C%d connected %s", conn->fd, conn->stream ? "streaming" : "plain");
   connect_attempt_discard();
   check_master_queue();
   return 0;
 
  x:
-  if (conn) {
-    perhaps_close(&conn->fd);
-    free(conn);
-  }
+  conn_dispose(conn);
   connect_attempt_discard();
 }
 
-static void connect_start() {
+static int allow_connect_start(void) {
+  return conns.count < max_connections
+    && !connecting_child
+    && !until_connect;
+}
+
+static void connect_start(void) {
   assert(!connecting_sockets[0]);
   assert(!connecting_sockets[1]);
   assert(!connecting_child);
 
   notice("starting connection attempt");
 
-  r= socketpair(AF_UNIX, SOCK_STREAM, 0, connecting_sockets);
+  int r= socketpair(AF_UNIX, SOCK_STREAM, 0, connecting_sockets);
   if (r) { syswarn("connect: cannot create socketpair for child"); goto x; }
 
   connecting_child= xfork("connection");
@@ -559,15 +822,11 @@ static void connect_start() {
     if (r) sysdie("connect: close parent socket in child");
 
     alarm(connection_setup_timeout);
-    if (NNTPconnect(remote_host, port, &cn_from, &cn_to, buf) < 0) {
-      if (buf[0]) {
-       sanitise_inplace(buf);
-       fatal("connect: rejected: %s", buf);
-      } else {
-       sysfatal("connect: connection attempt failed");
-      }
+    if (NNTPconnect((char*)remote_host, port, &cn_from, &cn_to, buf) < 0) {
+      if (buf[0]) fatal("connect: rejected: %s", sanitise(buf));
+      else sysfatal("connect: connection attempt failed");
     }
-    if (NNTPsendpassword(remote_host, cn_from, cn_to) < 0)
+    if (NNTPsendpassword((char*)remote_host, cn_from, cn_to) < 0)
       sysfatal("connect: authentication failed");
     if (try_stream) {
       if (fputs("MODE STREAM\r\n", cn_to) ||
@@ -582,19 +841,16 @@ static void connect_start() {
       }
       int l= strlen(buf);
       assert(l>=1);
-      if (buf[-1]!='\n') {
-       sanitise_inplace(buf);
+      if (buf[-1]!='\n')
        fatal("connect: response to MODE STREAM is too long: %.100s...",
-           remote_host, buf);
-      }
-      l--;  if (l>0 && buf[1-]=='\r') l--;
+           remote_host, sanitise(buf));
+      l--;  if (l>0 && buf[l-1]=='\r') l--;
       buf[l]= 0;
       char *ep;
       int rcode= strtoul(buf,&ep,10);
-      if (ep != buf[3]) {
-       sanitise_inplace(buf);
-       fatal("connect: bad response to MODE STREAM: %.50s", buf);
-      }
+      if (ep != &buf[3])
+       fatal("connect: bad response to MODE STREAM: %.50s", sanitise(buf));
+
       switch (rcode) {
       case 203:
        exitstatus= CONNCHILD_ESTATUS_STREAM;
@@ -603,8 +859,8 @@ static void connect_start() {
       case 500:
        break;
       default:
-       sanitise_inplace(buf);
-       warn("connect: unexpected response to MODE STREAM: %.50s", buf);
+       warn("connect: unexpected response to MODE STREAM: %.50s",
+            sanitise(buf));
        exitstatus= 2;
        break;
       }
@@ -629,73 +885,108 @@ static void connect_start() {
   if (r) sysdie("connect: close child socket in parent");
 
   on_fd_read_except(connecting_sockets[0], connchild_event);
-  return OOP_CONTINUE;
+  return;
 
  x:
   connect_attempt_discard();
 }
 
+static void check_idle_conns(void) {
+  Conn *conn;
+  for (conn=LIST_HEAD(conns); conn; conn=LIST_NEXT(conn))
+    conn->since_activity++;
+ search_again:
+  for (conn=LIST_HEAD(conns); conn; conn=LIST_NEXT(conn)) {
+    if (conn->since_activity <= need_activity_periods) continue;
+
+    /* We need to shut this down */
+    if (conn->quitting)
+      connfail(conn,"timed out waiting for response to QUIT");
+    else if (conn->sent.count)
+      connfail(conn,"timed out waiting for responses");
+    else if (conn->waiting.count || conn->priority.count)
+      connfail(conn,"BUG IN INNDUCT conn has queue but nothing sent");
+    else if (conn->xmitu)
+      connfail(conn,"peer has been sending responses"
+              " before receiving our commands!");
+    else {
+      static const char quitcmd[]= "QUIT\r\n";
+      int todo= sizeof(quitcmd)-1;
+      const char *p= quitcmd;
+      for (;;) {
+       int r= write(conn->fd, p, todo);
+       if (r<0) {
+         if (isewouldblock(errno))
+           connfail(conn, "blocked writing QUIT to idle connection");
+         else
+           connfail(conn, "failed to write QUIT to idle connection: %s",
+                    strerror(errno));
+         break;
+       }
+       assert(r<=todo);
+       todo -= r;
+       if (!todo) {
+         conn->quitting= 1;
+         conn->since_activity= 0;
+         debug("C%d is idle, quitting", conn->fd);
+         break;
+       }
+      }
+    }
+    goto search_again;
+  }
+}  
 
 /*========== overall control of article flow ==========*/
 
 static void check_master_queue(void) {
-  try reading current feed file;
-
-  if (!queue.count)
-    return;
-
-  Conn *last_assigned=0;
   for (;;) {
-    if (working.head) {
-      conn_assign_one_article(&working, &last_assigned);
-    } else if (idle.head) {
-      conn_assign_one_article(&idle, &last_assigned);
-    } else if (nconns < maxconns && queue.count >= max_queue_per_conn &&
-              !connecting_child && !connect_delay) {
-      connect_delay= reconnect_delay_periods;
+    if (!queue.count)
+      break;
+    
+    Conn *walk, *use=0;
+    int spare, inqueue;
+
+    /* Find a connection to offer this article.  We prefer a busy
+     * connection to an idle one, provided it's not full.  We take the
+     * first (oldest) and since that's stable, it will mean we fill up
+     * connections in order.  That way if we have too many
+     * connections, the spare ones will go away eventually.
+     */
+    for (walk=LIST_HEAD(conns); walk; walk=LIST_NEXT(walk)) {
+      if (walk->quitting) continue;
+      inqueue= walk->sent.count + walk->priority.count
+            + walk->waiting.count;
+      spare= walk->max_queue - inqueue;
+      assert(inqueue <= max_queue_per_conn);
+      assert(spare >= 0);
+      if (inqueue==0) /*idle*/ { if (!use) use= walk; }
+      else if (spare>0) /*working*/ { use= walk; break; }
+    }
+    if (use) {
+      if (!inqueue) use->since_activity= 0; /* reset idle counter */
+      while (spare>0) {
+       Article *art= LIST_REMHEAD(queue);
+       LIST_ADDTAIL(use->waiting, art);
+       spare--;
+      }
+      conn_maybe_write(use);
+    } else if (allow_connect_start()) {
+      until_connect= reconnect_delay_periods;
       connect_start();
+      break;
     } else {
       break;
     }
   }
-  conn_check_work(last_assigned);
-}
-
-static void conn_assign_one_article(LIST(Conn) *connlist,
-                                   Conn **last_assigned) {
-  Conn *conn= connlist->head;
-
-  LIST_REMOVE(*connlist, conn);
-  Article *art= LIST_REMHEAD(queue);
-  LIST_ADDTAIL(conn->queue, art);
-  LIST_ADD(*conn_determine_right_list(conn), conn);
-
-  /* This slightly odd arrangement is so that we call conn_check_work
-   * once after filling the queue for a new connection in
-   * check_master_queue, rather than for each article. */
-  if (conn != *last_assigned && *last_assigned)
-    conn_check_work(*last_assigned);
-  *last_assigned= conn;
-}
-
-static int conn_total_queued_articles(Conn *conn) {
-  return conn->sent.count + conn->queue.count;
-}
-
-static LIST(Conn) *conn_determine_right_list(Conn *conn) {
-  int inqueue= conn_total_queued_articles(conn);
-  assert(inqueue <= max_queue);
-  if (inqueue == 0) return &idle;
-  if (inqueue == conn->max_queue) return &full;
-  return &working;
 }
 
-static void *conn_writeable(oop_source *l, int fd, int ev, void *u) {
-  check_conn_work(u);
+static void *conn_writeable(oop_source *l, int fd, oop_event ev, void *u) {
+  conn_maybe_write(u);
   return OOP_CONTINUE;
 }
 
-static void conn_check_work(Conn *conn)  {
+static void conn_maybe_write(Conn *conn)  {
   void *rp= 0;
   for (;;) {
     conn_make_some_xmits(conn);
@@ -718,6 +1009,39 @@ static void conn_check_work(Conn *conn)  {
   }
 }
 
+static void vconnfail(Conn *conn, const char *fmt, va_list al) {
+  int requeue[art_MaxState];
+
+  Article *art;
+  while ((art= LIST_REMHEAD(conn->priority))) LIST_ADDTAIL(queue, art);
+  while ((art= LIST_REMHEAD(conn->waiting))) LIST_ADDTAIL(queue, art);
+  while ((art= LIST_REMHEAD(conn->sent))) {
+    requeue[art->state]++;
+    if (art->state==art_Unsolicited) art->state= art_Unchecked;
+    LIST_ADDTAIL(queue,art);
+  }
+
+  int i;
+  XmitDetails *d;
+  for (i=0, d=conn->xmitd; i<conn->xmitu; i++, d++)
+    xmit_free(d);
+
+  char *m= xvasprintf(fmt,al);
+  warn("C%d connection failed (requeueing " RCI_TRIPLE_FMT_BASE "): %s",
+       conn->fd, RCI_TRIPLE_VALS_BASE(requeue, /*nothing*/), m);
+  free(m);
+
+  LIST_REMOVE(conns,conn);
+  conn_dispose(conn);
+  check_master_queue();
+}
+
+static void connfail(Conn *conn, const char *fmt, ...) {
+  va_list al;
+  va_start(al,fmt);
+  vconnfail(conn,fmt,al);
+  va_end(al);
+}
 
 /*========== article transmission ==========*/
 
@@ -725,7 +1049,7 @@ static XmitDetails *xmit_core(Conn *conn, const char *data, int len,
                   XmitKind kind) { /* caller must then fill in details */
   struct iovec *v= &conn->xmit[conn->xmitu];
   XmitDetails *d= &conn->xmitd[conn->xmitu++];
-  v->iov_base= data;
+  v->iov_base= (char*)data;
   v->iov_len= len;
   d->kind= kind;
   return d;
@@ -737,7 +1061,7 @@ static void xmit_noalloc(Conn *conn, const char *data, int len) {
 #define XMIT_LITERAL(lit) (xmit_noalloc(conn, (lit), sizeof(lit)-1))
 
 static void xmit_artbody(Conn *conn, ARTHANDLE *ah /* consumed */) {
-  XmitDetails *d= xmit_core(conn, ah->data, ah->len, sk_Artdata);
+  XmitDetails *d= xmit_core(conn, ah->data, ah->len, xk_Artdata);
   d->info.sm_art= ah;
 }
 
@@ -763,13 +1087,14 @@ static void *conn_write_some_xmits(Conn *conn) {
     if (count > IOV_MAX) count= IOV_MAX;
     ssize_t rs= writev(conn->fd, conn->xmit, count);
     if (rs < 0) {
-      if (errno == EAGAIN) return OOP_CONTINUE;
+      if (isewouldblock(errno)) return OOP_CONTINUE;
       connfail(conn, "write failed: %s", strerror(errno));
       return OOP_HALT;
     }
     assert(rs > 0);
 
-    for (done=0; rs && done<xmitu; done++) {
+    int done;
+    for (done=0; rs && done<conn->xmitu; done++) {
       struct iovec *vp= &conn->xmit[done];
       XmitDetails *dp= &conn->xmitd[done];
       if (rs > vp->iov_len) {
@@ -792,18 +1117,19 @@ static void conn_make_some_xmits(Conn *conn) {
     if (conn->xmitu+5 > CONNIOVS)
       break;
 
-    Article *art= LIST_REMHEAD(queue);
+    Article *art= LIST_REMHEAD(conn->priority);
+    if (!art) art= LIST_REMHEAD(conn->waiting);
     if (!art) break;
 
-    if (art->checked || (conn->stream && nocheck)) {
+    if (art->state >= art_Wanted || (conn->stream && nocheck)) {
       /* actually send it */
 
-      ARTHANDLE *artdata= SMretrieve();
+      ARTHANDLE *artdata= SMretrieve(art->token, RETR_ALL);
 
       if (conn->stream) {
        if (artdata) {
          XMIT_LITERAL("TAKETHIS ");
-         xmit_noalloc(conn, art->mid, art->midlen);
+         xmit_noalloc(conn, art->messageid, art->midlen);
          XMIT_LITERAL("\r\n");
          xmit_artbody(conn, artdata);
        }
@@ -816,11 +1142,13 @@ static void conn_make_some_xmits(Conn *conn) {
        }
       }
 
-      art->sent= 1;
+      art->state=
+       art->state == art_Unchecked ? art_Unsolicited :
+       art->state == art_Wanted    ? art_Wanted      :
+       (abort(),-1);
+      art->ipf->counts[art->state][RC_sent]++;
       LIST_ADDTAIL(conn->sent, art);
 
-      counts[art->checked].sent++;
-
     } else {
       /* check it */
 
@@ -828,11 +1156,12 @@ static void conn_make_some_xmits(Conn *conn) {
        XMIT_LITERAL("IHAVE ");
       else
        XMIT_LITERAL("CHECK ");
-      xmit_noalloc(art->mid, art->midlen);
+      xmit_noalloc(conn, art->messageid, art->midlen);
       XMIT_LITERAL("\r\n");
 
+      assert(art->state == art_Unchecked);
+      art->ipf->counts[art->state][RC_sent]++;
       LIST_ADDTAIL(conn->sent, art);
-      counts[art->checked].offered++;
     }
   }
 }
@@ -846,10 +1175,20 @@ static const oop_rd_style peer_rd_style= {
   OOP_RD_SHORTREC_FORBID
 };
 
-static Article *article_reply_check(Connection *conn, const char *response,
+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(Conn *conn, const char *response,
                                    int code_indicates_streaming,
+                                   int must_have_sent
+                                       /* 1:yes, -1:no, 0:dontcare */,
                                    const char *sanitised_response) {
-  Article *art= LIST_REMHEAD(conn->sent);
+  Article *art= LIST_HEAD(conn->sent);
 
   if (!art) {
     connfail(conn,
@@ -861,20 +1200,20 @@ static Article *article_reply_check(Connection *conn, const char *response,
   if (code_indicates_streaming) {
     assert(!memchr(response, 0, 4)); /* ensured by peer_rd_ok */
     if (!conn->stream) {
-      connfail("peer gave streaming response code "
+      connfail(conn, "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"
+      connfail(conn, "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 -"
+      connfail(conn, "peer gave streaming response code to wrong article -"
               " probable synchronisation problem; we offered: %s;"
               " peer said: %s",
               art->messageid, sanitised_response);
@@ -882,28 +1221,43 @@ static Article *article_reply_check(Connection *conn, const char *response,
     }
   } else {
     if (conn->stream) {
-      connfail("peer gave non-streaming response code to CHECK/TAKETHIS: %s",
-              sanitised_response);
+      connfail(conn, "peer gave non-streaming response code to"
+              " CHECK/TAKETHIS: %s", sanitised_response);
       return 0;
     }
   }
 
+  if (must_have_sent>0 && art->state < art_Wanted) {
+    connfail(conn, "peer says article accepted but"
+            " we had not sent the body: %s", sanitised_response);
+    return 0;
+  }
+  if (must_have_sent<0 && art->state >= art_Wanted) {
+    connfail(conn, "peer says please sent the article but we just did: %s",
+            sanitised_response);
+    return 0;
+  }
+
+  Article *art_again= LIST_REMHEAD(conn->sent);
+  assert(art_again == art);
   return art;
 }
 
 static void update_nocheck(int accepted) {
-  accept_proportion *= accept_decay;
-  accept_proportion += accepted;
-  nocheck= accept_proportion >= nocheck_thresh;
-  if (nocheck && !nocheck_reported) {
+  accept_proportion *= nocheck_decay;
+  accept_proportion += accepted * (1.0 - nocheck_decay);
+  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 != nocheck) {
+    debug("nocheck mode %s", new_nocheck ? "start" : "stop");
   }
+  nocheck= new_nocheck;
 }
 
-static void article_done(Connection *conn, Article *art, int whichcount) {
-  *count++;
-  counts.articles[art->checked][whichcount]++;
+static void article_done(Conn *conn, Article *art, int whichcount) {
+  art->ipf->counts[art->state][whichcount]++;
   if (whichcount == RC_accepted) update_nocheck(1);
   else if (whichcount == RC_unwanted) update_nocheck(0);
 
@@ -929,20 +1283,12 @@ static void article_done(Connection *conn, Article *art, int whichcount) {
   ipf->inprogress--;
   assert(ipf->inprogress >= 0);
 
-  if (!ipf->inprogress)
-    loop->on_time(loop, OOP_TIME_NOW, statemc_check_oldinput_done, 0);
+  if (!ipf->inprogress && ipf != main_input_file)
+    queue_check_input_done();
 
   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) {
@@ -954,43 +1300,40 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev,
   }
   assert(ev == OOP_RD_OK);
 
+  char *sani= sanitise(data);
+
   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);
+    connfail(conn, "badly formatted response from peer: %s", sani);
     return OOP_CONTINUE;
   }
 
   if (conn->quitting) {
-    if (code!=205) {
-      connfail(conn, "peer gave failure response to QUIT: %s", sani);
-      return OOP_CONTINUE;
+    if (code!=205 && code!=503) {
+      connfail(conn, "peer gave unexpected response to QUIT: %s", sani);
+    } else {
+      notice("C%d idle connection closed", conn->fd);
+      assert(!conn->waiting.count);
+      assert(!conn->priority.count);
+      assert(!conn->sent.count);
+      assert(!conn->xmitu);
+      LIST_REMOVE(conns,conn);
+      conn_dispose(conn);
     }
-    conn close ok;
-    return;
+    return OOP_CONTINUE;
   }
 
+  conn->since_activity= 0;
   Article *art;
 
-#define GET_ARTICLE                                                        \
-  art= article_reply_check(conn, data, code_streaming, sani);              \
+#define GET_ARTICLE(musthavesent)                                          \
+  art= article_reply_check(conn, data, musthavesent, 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;                                                 \
+#define ARTICLE_DEALTWITH(streaming,musthavesent,how)          \
+  code_streaming= (streaming);                                 \
+  GET_ARTICLE(musthavesent);                                   \
   article_done(conn, art, RC_##how);  break;
 
 #define PEERBADMSG(m) connfail(conn, m ": %s", sani);  return OOP_CONTINUE
@@ -1003,41 +1346,40 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev,
   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 435: ARTICLE_DEALTWITH(0,0,unwanted); /* IHAVE says they have it */
+  case 438: ARTICLE_DEALTWITH(1,0,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 235: ARTICLE_DEALTWITH(0,1,accepted); /* IHAVE says thanks */
+  case 239: ARTICLE_DEALTWITH(1,1,accepted); /* TAKETHIS says thanks */
 
-  case 437: ARTICLE_DEALTWITH(0,rejected); /* IHAVE says rejected */
-  case 439: ARTICLE_DEALTWITH(1,rejected); /* TAKETHIS says rejected */
+  case 437: ARTICLE_DEALTWITH(0,0,rejected); /* IHAVE says rejected */
+  case 439: ARTICLE_DEALTWITH(1,0,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;
+    GET_ARTICLE(-1);
+    assert(art->state == art_Unchecked);
+    art->ipf->counts[art->state][RC_accepted]++;
+    art->state= art_Wanted;
+    LIST_ADDTAIL(conn->priority, art);
     break;
 
   case 431: /* CHECK or TAKETHIS says try later */
     code_streaming= 1;
   case 436: /* IHAVE says try later */
-    GET_ARTICLE;
+    GET_ARTICLE(0);
+    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;
 
   }
 
-  check_check_work(conn);
+  conn_maybe_write(conn);
+  check_master_queue();
   return OOP_CONTINUE;
 }
 
@@ -1046,17 +1388,21 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev,
 
 static void feedfile_eof(InputFile *ipf) {
   assert(ipf != main_input_file); /* promised by tailing_try_read */
-  assert(ipf == old_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);
+  if (ipf == flushing_input_file) {
+    assert(sms==sm_SEPARATED || sms==sm_DROPPING);
+    if (main_input_file) inputfile_tailing_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) {
@@ -1066,15 +1412,11 @@ static InputFile *open_input_file(const char *path) {
     sysfatal("unable to open input file %s", path);
   }
 
-  InputFile *ipf= xmalloc(sizeof(InputFile));
+  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;
-  ipf->path= path;
+  strcpy(ipf->path, path);
 
   return ipf;
 }
@@ -1085,102 +1427,127 @@ static void close_input_file(InputFile *ipf) {
   assert(!ipf->rd); /* must have had inputfile_tailing_stop */
   assert(!ipf->inprogress); /* no dangling pointers pointing here */
 
-  if (ipf->fd>=0)
+  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) {
+static void *feedfile_got_bad_data(InputFile *ipf, off_t offset,
+                                  const char *data, const char *how) {
+  warn("corrupted file: %s, offset %lu: %s: %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; }
 
-  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);
-  }
+  off_t old_offset= ipf->offset;
   ipf->offset += recsz + 1;
 
-  if (sms==sm_NORMAL && ipf->offset >= flush_threshold) {
-    notice("starting flush (%lu >= %lu)",
-          (unsigned long)ipf->offset, (unsigned long)flush_threshold);
+#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 */
 
-    int r= link(feedfile, duct_path);
-    if (r) sysdie("link feedfile %s to ductfile %s", feedfile, dut_path);
-    /* => Hardlinked */
+  if (ipf->skippinglong) {
+    if (ev==OOP_RD_OK) ipf->skippinglong= 0; /* fine now */
+    return;
+  }
+  if (ev==OOP_RD_LONG) {
+    ipf->skippinglong= 1;
+    X_BAD_DATA("overly long line");
+  }
 
-    r= unlink(feedfile);
-    if (r) sysdie("unlink old feedfile link %s", feedfile);
-    /* => Moved */
+  if (memchr(data,'\0',recsz)) X_BAD_DATA("nul byte");
+  if (!recsz) X_BAD_DATA("empty line");
 
-    spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */
+  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->offset= ipf->offset;
+  art->blanklen= recsz;
+  art->midlen= midlen;
+  art->state= art_Unchecked;
+  art->ipf= ipf;  ipf->inprogress++;
+  art->token= TextToToken(tokentextbuf);
+  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_master_queue();
+  return OOP_CONTINUE;
 }
 
-
 /*========== 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);
+  return ipf->readable_callback(loop, &ipf->readable,
+                               ipf->readable_callback_user);
 }
 
-static void on_cancel(struct oop_readable *rable) {
+static void tailing_on_cancel(struct oop_readable *rable) {
   InputFile *ipf= (void*)rable;
 
-  if (ipf->filemon) filemon_stopfile(ipf);
+  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;
@@ -1188,9 +1555,9 @@ static int tailing_on_readable(struct oop_readable *rable,
   tailing_on_cancel(rable);
   ipf->readable_callback= cb;
   ipf->readable_callback_user= user;
-  filemon_startfile(ipf);
+  filemon_start(ipf);
 
-  loop->on_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf);
+  tailing_queue_readable(ipf);
   return 0;
 }
 
@@ -1207,7 +1574,7 @@ static ssize_t tailing_try_read(struct oop_readable *rable, void *buffer,
       if (ipf==main_input_file) {
        errno=EAGAIN;
        return -1;
-      } else if (ipf==old_input_file) {
+      } else if (ipf==flushing_input_file) {
        assert(ipf->fd>=0);
        assert(sms==sm_SEPARATED || sms==sm_DROPPING);
       } else if (ipf==backlog_input_file) {
@@ -1231,9 +1598,9 @@ static int filemon_inotify_fd;
 static int filemon_inotify_wdmax;
 static InputFile **filemon_inotify_wd2ipf;
 
-typedef struct Filemon_Perfile {
+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);
@@ -1251,11 +1618,15 @@ static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) {
   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;
@@ -1267,7 +1638,7 @@ static void *filemon_inotify_readable(oop_source *lp, int fd,
   for (;;) {
     int r= read(filemon_inotify_fd, &iev, sizeof(iev));
     if (r==-1) {
-      if (errno==EAGAIN) break;
+      if (isewouldblock(errno)) break;
       sysdie("read from inotify master");
     } else if (r==sizeof(iev)) {
       assert(iev.wd >= 0 && iev.wd < filemon_inotify_wdmax);
@@ -1275,6 +1646,7 @@ static void *filemon_inotify_readable(oop_source *lp, int fd,
       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;
@@ -1289,16 +1661,17 @@ static int filemon_method_init(void) {
   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 *//
+#endif /* HAVE_INOTIFY && !HAVE_FILEMON */
 
 /*---------- filemon dummy implementation ----------*/
 
 #if !defined(HAVE_FILEMON)
 
-typedef struct Filemon_Perfile { int dummy; } Filemon_Dummy_Perfile;
+struct Filemon_Perfile { int dummy; };
 
 static int filemon_method_init(void) { return 0; }
 static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) { }
@@ -1306,21 +1679,42 @@ 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) {
+  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_FORBID,
-  OOP_RD_SHORTREC_EOF,
+  OOP_RD_NUL_PERMIT,
+  OOP_RD_SHORTREC_LONG,
 };
 
 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.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;
@@ -1329,320 +1723,555 @@ static void inputfile_tailing_start(InputFile *ipf) {
   assert(ipf->fd);
 
   int r= oop_rd_read(ipf->rd, &feedfile_rdstyle, MAX_LINE_FEEDFILE,
-                    feedfile_got_article,ipf, feedfile_problem,ipf);
+                    feedfile_got_article,ipf, feedfile_read_err, 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 ==========*/
+/*========== interaction with innd - state machine ==========*/
 
 /* See official state diagram at top of file.  We implement
  * this as follows:
-
-          ================
-           WAITING
-          [Nothing/Noduct]
-           poll for F
-          ================
-               |
-               |     TIMEOUT
-               |`--------------------------.
-               |                           | install defer as backlog
-                | OPEN F SUCCEEDS           | exit
-     ,--------->|                           V
-     |          V                         =========
-     |     ========                        (ESRCH)
-     |      NORMAL                        [Dropped]
-     |     [Normal]                       =========
-     |      main F tail
-     |     ========
-     |          |
-     |          | F IS SO BIG WE SHOULD FLUSH
-     ^          | hardlink F to D
-     |     [Hardlinked]
-     |          | unlink F
-     |          | our handle onto F is now onto D
-     |     [Moved]
-     |          |
-     |          |<---------------------------------------------------.
+ * -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                                                 |
+     |     ==================                                        |
+     |      FLUSHING[-ABSENT]                                        |
      |     [Flushing]                                                |
-     |      main D tail                                              |
-     |     ==========                                                |
+     |     main D tail/none                                          |
+     |     ==================                                        |
      |          |                                                    |
      |          |   INNDCOMM FLUSH FAILS                             ^
      |          |`----------------------->----------.                |
      |          |                                   |                |
      |          |   NO SUCH SITE                    V                |
-     ^          |`--------------->----.          ===========         |
-     |          |                      \         FLUSHFAIL           |
-     |          |                       \        [Moved]             |
-     |          |                        \       main D tail         |
-     |          |                         \      ===========         |
+     ^          |`--------------->----.         ==================== |
+     |          |                      \        FLUSHFAILED[-ABSENT] |
+     |          |                       \         [Moved]            |
+     |          | FLUSH OK               \       main D tail/none    |
+     |          | open F                  \     ==================== |
      |          |                          \        |                |
      |          |                           \       | TIME TO RETRY  |
-     |          |                            \      `----------------'
-     |          | FLUSH OK                    \
-     |          | open F                       \
-     |          V                               V
-     |     =============                     ============
-     |      SEPARATED/                        DROPPING/
-     |      old->fd>=0                        old->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/
-     |      old->fd==-1                       old->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
-     |          | start new defer                | exit
-     |          |                                V
-     `----------'                            ==========
-                                              (ESRCH)
-                                             [Droppped]
-                                             ==========
+     |          |`------->----.     ,---<---'\      `----------------'
+     |          |    D NONE   |     | D NONE  `----.
+     |          V             |     |              V
+     |     =============      V     V                    ============
+     |      SEPARATED-1       |     |              DROPPING-1
+     |      flsh->fd>=0       |     |              flsh->fd>=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->fd==-1      |     V              flsh->fd==-1
+     |     [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 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 */
+static void startup_set_input_file(InputFile *f) {
+  assert(!main_input_file);
+  main_input_file= f;
+  inputfile_tailing_start(f);
+}
 
-  r= fstat(fileno(defer), &stab);
-  if (r) sysdie("could not stat newly opened defer file %s", path_ductdefer);
+static void statemc_init(void) {
+  struct stat stab, stabf;
 
-  if (stab.st_size > LONG_MAX)
-    die("defer file %s size is far too large", path_ductdefer);
+  path_lock=        xasprintf("%s_lock",      feedfile);
+  path_flushing=    xasprintf("%s_flushing",  feedfile);
+  path_defer=       xasprintf("%s_defer",     feedfile);
+  globpat_backlog=  xasprintf("%s_backlog*",  feedfile);
 
-  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_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;
+    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");
+      }
+      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, path_lock, "lockfile");
+    int lock_noent;
+    xlstat_isreg(path_lock, &stab, &lock_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 (!lock_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);
-}
+  debug("startup: locked");
 
-static void statemc_init(void) {
-  struct stat stab;
+  search_backlog_file();
 
-  path_ductlock=  xasprintf("%s_duct.lock",  feedfile);
-  path_duct=      xasprintf("%s_duct",       feedfile);
-  path_ductdefer= xasprintf("%s_duct.defer", feedfile);
-
-  if (lstat(path_ductdefer, &stab)) {
-    if (errno!=ENOENT) sysdie("could not check defer file %s", path_defer);
+  int defer_noent;
+  xlstat_isreg(path_defer, &stab, &defer_noent, "defer file");
+  if (defer_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"
-              " hardlink to backlog file)", path_defer);
+      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);
     }
   }
-  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);
+  struct stat stab_f, stab_d;
+  int noent_f;
 
-  if (file_d) {
-    struct stat stab_f, stab_d;
+  InputFile *file_d= open_input_file(path_flushing);
+  if (file_d) xfstat_isreg(file_d->fd, &stab_d, path_flushing,"flushing file");
 
-    r= stat(feedfile, &stab_f);
-    if (r) {
-      if (errno!=ENOENT) sysdie("check feed file %s", feedfile);
-      /* D exists, F ENOENT => Moved */
-      goto found_moved;
-    }
-
-    /* F and D both exist */
+  xlstat_isreg(feedfile, &stab_f, &noent_f, "feedfile");
 
-    r= fstat(file_d->fd, &stab_d);
-    if (r) sysdie("check duct file %s", ductfile);
+  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 (stab_d.st_ino == stab_f.st_ino &&
-       stab_d.st_dev == stab_f.st_dev) {
-      /* F==D => Hardlinked*/
-      r= unlink(path_duct);
-      if (r) sysdie("unlink feed file %s during startup", feedfile);
-    found_moved:
-      /* => Moved */
+  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);
-      spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */
-    } else {
-      /* F!=D => Separated */
       SMS(SEPARATED, 0, "found both old and current feed files");
-      startup_set_input_file(file_d);
+    } 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, flushfail_retry_periods, "normal startup");
     }
-  } else { /*!file_d*/
-    SMS(WAITING, open_wait_periods, "no feed file currently exists");
   }
 }
 
-static void statemc_poll(void) {
-  if (sms==sm_WAITING) { statemc_waiting_poll(); return; }
+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) sysdie("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_WAITING:
-    fatal("timed out waiting for innd to create feed file %s", feedfile);
-  case sm_FLUSHFAIL:
-    spawn_inndcomm_flush(void);
+  case sm_NORMAL:
+    statemc_start_flush("periodic"); /* Normal => Flushing; => FLUSHING */
+    break;
+  case sm_FLUSHFAILED:
+    spawn_inndcomm_flush("retry"); /* Moved => Flushing; => FLUSHING */
     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 int inputfile_is_done(InputFile *ipf) {
+  if (!ipf) return 0;
+  if (ipf->inprogress) return 0; /* new article in the meantime */
+  if (ipf->fd >= 0); return 0; /* not had EOF */
+  return 1;
 }
 
-static void startup_set_input_file(InputFile *f) {
-  assert(!main_input_file);
-  main_input_file= f;
-  inputfile_tailing_start(f);
+static void notice_processed(InputFile *ipf, const char *what,
+                            const char *spec) {
+#define RCI_NOTHING(x) /* nothing */
+#define RCI_TRIPLE_FMT(x) " " #x "=" RCI_TRIPLE_FMT_BASE
+#define RCI_TRIPLE_VALS(x) , RCI_TRIPLE_VALS_BASE(ipf->counts, [RC_##x])
+
+#define CNT(art,rc) (ipf->counts[art_##art][RC_##rc])
+
+  info("processed %s%s read=%d(+%dbl,+%derr)"
+       " offered=%d(ch%d,nc%d) accepted=%d(ch%d+nc%d)"
+       RESULT_COUNTS(RCI_NOTHING, RCI_TRIPLE_FMT)
+       ,
+       what, spec,
+       ipf->readcount_ok, ipf->readcount_blank, ipf->readcount_err,
+       CNT(Unchecked,sent) + CNT(Unsolicited,sent)
+       , CNT(Unchecked,sent), CNT(Unsolicited,sent),
+       CNT(Wanted,accepted) + CNT(Unsolicited,accepted)
+       , CNT(Wanted,accepted), CNT(Unsolicited,accepted)
+       RESULT_COUNTS(RCI_NOTHING,  RCI_TRIPLE_VALS)
+       );
+
+#undef CNT
+}
+
+static void statemc_check_backlog_done(void) {
+  InputFile *ipf= backlog_input_file;
+  if (!inputfile_is_done(ipf)) return;
+
+  const char *slash= strrchr(ipf->path, '/');
+  const char *leaf= slash ? slash+1 : ipf->path;
+  const char *under= strchr(slash, '_');
+  const char *rest= under ? under+1 : leaf;
+  if (!strncmp(rest,"backlog",7)) rest += 7;
+  notice_processed(ipf,"backlog:",rest);
+  
+  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);
+  }
+  free(ipf);
+  backlog_input_file= 0;
+  search_backlog_file();
+  return;
 }
 
-static void *statemc_check_oldinput_done(oop_source *lp,
-                                        struct timeval now, void *u) {
+static void statemc_check_flushing_done(void) {
+  InputFile *ipf= flushing_input_file;
+  if (!inputfile_is_done(ipf)) return;
+
+  assert(sms==sm_SEPARATED || sms==sm_DROPPING);
+
+  notice_processed(ipf,"feedfile",0);
+
+  close_defer();
+
+  xunlink(path_flushing, "old flushing file");
+
+  close_input_file(flushing_input_file);
+  free(flushing_input_file);
+  flushing_input_file= 0;
+
+  if (sms==sm_SEPARATED) {
+    notice("flush complete");
+    SMS(NORMAL, 0, "flush complete");
+  } else if (sms==sm_DROPPING) {
+    SMS(DROPPED, 0, "old flush complete");
+    search_backlog_file();
+    notice("feed dropped, but will continue until backlog is finished");
+  }
+}
+
+static void *statemc_check_input_done(oop_source *lp, struct timeval now,
+                                     void *u) {
+  assert(!inputfile_is_done(main_input_file));
+  statemc_check_flushing_done();
+  statemc_check_backlog_done();
+  return OOP_CONTINUE;
+}
+
+static void queue_check_input_done(void) {
+  loop->on_time(loop, OOP_TIME_NOW, statemc_check_input_done, 0);
+}
+
+static void statemc_setstate(StateMachineState newsms, int periods,
+                            const char *forlog, const char *why) {
+  sms= newsms;
+  sm_period_counter= periods;
+
+  const char *xtra= "";
+  switch (sms) {
+  case sm_FLUSHING: sm_FLUSHFAILED:
+    if (!main_input_file) xtra= "-ABSENT";
+    break;
+  case sm_SEPARATED: case sm_DROPPING:
+    xtra= flushing_input_file->fd >= 0 ? "-1" : "-2";
+    break;
+  default:;
+  }
+
+  if (periods) {
+    info("%s%s[%d] %s",forlog,xtra,periods,why);
+  } else {
+    info("%s%s %s",forlog,xtra,why);
+  }
+}
+
+/*---------- defer and backlog files ----------*/
+
+static void open_defer(void) {
   struct stat stab;
 
-  int done= (sms==sm_SEPARATED || sms==sm_DROPPING)
-         && old_input_file->fd==-1
-         && !old_input_file->inprogress;
-  if (!done) return;
+  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, path_defer, "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);
+
+    int 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;
 
-  r= fstat(fileno(defer), &stab);
-  if (r) sysdie("check defer file %s", path_defer);
+  struct stat stab;
+  xfstat_isreg(fileno(defer), &stab, path_defer, "defer file");
 
   if (fclose(defer)) sysfatal("could not close defer file %s", path_defer);
   defer= 0;
 
+  time_t now= xtime();
+
   char *backlog= xasprintf("%s_backlog_%lu.%lu", feedfile,
-                          (unsigned long)now.tv_sec,
+                          (unsigned long)now,
                           (unsigned long)stab.st_ino);
-  if (link(path_defer, path_backlog))
+  if (link(path_defer, 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 (unlink(path_duct))
-    sysdie("could not unlink old duct file %s", path_duct);
+  free(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 void search_backlog_file(void) {
+  /* returns non-0 iff there are any backlog files */
+
+  glob_t gl;
+  int r, i;
+  struct stat stab;
+  const char *oldest_path=0;
+  time_t oldest_mtime, now;
+
+  if (backlog_input_file) return;
 
-  if (sms==sm_DROPPING) {
-    notice("feed dropped and our work is complete"
-          " (but check for backlog files)");
-    exit(0);
+ 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);
   }
 
-  open_defer();
+  globfree(&gl);
 
-  close_input_file(old_input_file);
-  old_input_file= 0;
+  if (!oldest_path) {
+    debug("backlog scan: none");
 
-  notice("flush complete");
-  SMS(NORMAL, 0, "flush complete");
-}
+    if (sms==sm_DROPPED) {
+      notice("feed dropped and our work is complete");
+      xunlink(path_lock, "lockfile for old feed");
+      exit(0);
+    }
+    until_backlog_nextscan= backlog_spontaneous_rescan_periods;
+    return;
+  }
 
-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);
+  now= xtime();
+  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;
   }
+
+  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;
 }
 
 /*========== flushing the feed ==========*/
@@ -1652,35 +2281,56 @@ 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);
+  cancel_fd_read_except(fd);
   close(fd);
 
-  assert(!old_input_file);
+  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");
-      old_input_file= main_input_file;
+      flushing_input_file= main_input_file;
+      tailing_queue_readable(flushing_input_file);
+        /* we probably previously returned EAGAIN from our fake read method
+        * when in fact we were at EOF, so signal another readable event
+        * so we actually see the EOF */
+
       main_input_file= 0;
-      SMS(DROPPING, 0, "dropped by innd");
+
+      if (flushing_input_file) {
+       SMS(DROPPING, 0, "feed dropped by innd, but must finish last flush");
+      } else {
+       close_defer();
+       SMS(DROPPED, 0, "feed dropped by innd");
+       search_backlog_file();
+      }
       return OOP_CONTINUE;
 
     case 0:
-      old_input_file= main_input_file;
+      /* as above */
+      flushing_input_file= main_input_file;
+      tailing_queue_readable(flushing_input_file);
+
       main_input_file= open_input_file(feedfile);
       if (!main_input_file)
        die("flush succeeded but feedfile %s does not exist!", feedfile);
-      SMS(SEPARATED, 0, "feed file missing");
+
+      if (flushing_input_file) {
+       SMS(SEPARATED, spontaneous_flush_periods, "recovery flush complete");
+      } else {
+       close_defer();
+       SMS(NORMAL, spontaneous_flush_periods, "flush complete");
+      }
       return OOP_CONTINUE;
 
     default:
       goto unexpected_exitstatus;
-      
+
     }
   } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) {
     warn("flush timed out trying to talk to innd");
@@ -1691,7 +2341,7 @@ static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) {
   }
 
  failed:
-  SMS(FLUSHFAIL, flushfail_retry_periods, "flush failed, will retry");
+  SMS(FLUSHFAILED, flushfail_retry_periods, "flush failed, will retry");
 }
 
 static void inndcommfail(const char *what) {
@@ -1699,19 +2349,22 @@ static void inndcommfail(const char *what) {
   exit(INNDCOMMCHILD_ESTATUS_FAIL);
 }
 
-void spawn_inndcomm_flush(void) {
+void spawn_inndcomm_flush(const char *why) { /* Moved => Flushing */
   int pipefds[2];
 
-  assert(sms==sm_NORMAL || sms==sm_FLUSHFAIL);
+  notice("flushing %s",why);
+
+  assert(sms==sm_NORMAL || sms==sm_FLUSHFAILED);
   assert(!inndcomm_child);
 
   if (pipe(pipefds)) sysdie("create pipe for inndcomm child sentinel");
 
-  inndcomm_child= xfork();
+  inndcomm_child= xfork("inndcomm child");
 
   if (!inndcomm_child) {
-    static char flushargv[2]= { sitename, 0 };
+    const char *flushargv[2]= { sitename, 0 };
     char *reply;
+    int r;
 
     close(pipefds[0]);
 
@@ -1729,7 +2382,7 @@ void spawn_inndcomm_flush(void) {
   int sentinel_fd= pipefds[0];
   on_fd_read_except(sentinel_fd, inndcomm_event);
 
-  SMS(FLUSHING, 0, "flush is in progress");
+  SMS(FLUSHING, 0, why);
 }
 
 /*========== main program ==========*/
@@ -1741,13 +2394,6 @@ static void postfork_inputfile(InputFile *ipf) {
   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);
@@ -1758,16 +2404,18 @@ static void postfork(const char *what) {
     sysdie("%s child: failed to reset SIGPIPE");
 
   postfork_inputfile(main_input_file);
-  postfork_inputfile(old_input_file);
-  postfork_conns(idle.head);
-  postfork_conns(working.head);
-  postfork_conns(full.head);
+  postfork_inputfile(flushing_input_file);
+
+  Conn *conn;
+  for (conn=LIST_HEAD(conns); conn; conn=LIST_NEXT(conn))
+    close(conn->fd);
+
   postfork_stdio(defer);
 }
 
-
-#define EVERY(what, interval, body)                                         \
-  static const struct timeval what##_timeout = { 5, 0 };                    \
+#define EVERY(what, interval_sec, interval_usec, body)                      \
+  static const struct timeval what##_timeout =                              \
+            { interval_sec, interval_usec };                                \
   static void what##_schedule(void);                                        \
   static void *what##_timedout(oop_source *lp, struct timeval tv, void *u) { \
     { body }                                                                \
@@ -1777,120 +2425,160 @@ 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--;
-  statemc_poll();
+#define DEBUGF_IPF(wh) " " #wh "=%p/%s:ip=%ld,off=%ld,fd=%d%s"
+#define DEBUG_IPF(wh)                                          \
+  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[%d] conns=%d queue=%d until_connect=%d"
+       " input_files" DEBUGF_IPF(main) DEBUGF_IPF(old) DEBUGF_IPF(flushing)
+       " children connecting=%ld inndcomm_child"
+       ,
+       sms_names[sms], sm_period_counter,
+       queue.count, conns.count, until_connect,
+       DEBUG_IPF(main), DEBUG_IPF(flushing), DEBUG_IPF(flushing),
+       (long)connecting_child, (long)inndcomm_child
+       );
+
+  if (until_connect) until_connect--;
+
+  poll_backlog_file();
+  if (!backlog_input_file) close_defer(); /* want to start on a new backlog */
+  statemc_period_poll();
   check_master_queue();
+  check_idle_conns();
 });
 
 
 /*========== option parsing ==========*/
 
+/*---------- generic option parser and logging ----------*/
+
+static void vbadusage(const char *fmt, va_list al) NORET_PRINTF(1,0);
+static void vbadusage(const char *fmt, va_list al) {
+  abort();
+}
+static void badusage(const char *fmt, ...) NORET_PRINTF(1,2);
+static void badusage(const char *fmt, ...) {
+  va_list al;
+  va_start(al,fmt);
+  vbadusage(fmt,al);
+}
+
 enum OptFlags {
-  of_seconds= 001000u;
-  of_boolean= 002000u;
+  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;
+  int shrt;
+  const char *lng;
   void *store;
   OptionParser *fn;
-  int noarg;
+  int noarg, intval;
 };
 
-void op_integer(const Option *o, const char *val) {
+/*---------- specific option types ----------*/
+
+static 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);
+    badusage("bad integer value for %s",o->lng);
   int *store= o->store;
   *store= ul;
 }
 
-void op_double(const Option *o, const char *val) {
+static 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);
+    badusage("bad floating point value for %s",o->lng);
 }
 
-void op_string(const Option *o, const char *val) {
-  char **store= o->store;
-  free(*store);
+static void op_string(const Option *o, const char *val) {
+  const char **store= o->store;
   *store= val;
 }
 
-void op_seconds(const Option *o, const char *val) {
+static void op_seconds(const Option *o, const char *val) {
   int *store= o->store;
   char *ep;
-  
+  int unit;
+
   double v= strtod(val,&ep);
-  if (ep==val) badusage("bad time/duration value for %s",o->long);
+  if (ep==val) badusage("bad time/duration value for %s",o->lng);
 
   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);
+  else badusage("bad units %s for time/duration value for %s",ep,o->lng);
 
   v *= unit;
   v= ceil(v);
-  if (v > INT_MAX) badusage("time/duration value for %s out of range",o->long);
+  if (v > INT_MAX) badusage("time/duration value for %s out of range",o->lng);
   *store= v;
 }
 
-void op_periods_rndup(const Option *o, const char *val) {
+static 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) {
+static void op_setint(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;
+  *store= o->intval;
 }
 
+/*---------- specific options ----------*/
+
 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     },
+{'q',"quiet-multiple",        &quiet_multiple,           op_setint,   1,1    },
+
+{ 0, "max-connections",       &max_connections,          op_integer          },
+{ 0, "max-queue-per-conn",    &max_queue_per_conn,       op_integer          },
+
+
+{ 0, "streaming",             &try_stream,               op_setint,  1,1     },
+{ 0, "no-streaming",          &try_stream,               op_setint,  1,0     },
+{'P',"port",                  &port,                     op_integer          },
+{ 0, "inndconf",              &inndconffile,             op_string           },
+{'d',"daemon",                &become_daemon,            op_setint,  1,1     },
+{ 0, "no-daemon",             &become_daemon,            op_setint,  1,0     },
 
 { 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;
@@ -1903,25 +2591,25 @@ int main(int argc, char **argv) {
        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))
+       for (o=options; o->lng; o++)
+         if (strlen(o->lng) == len && !memcmp(o->lng,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);
+         if (equals) badusage("option --%s does not take a value",o->lng);
          arg= 0;
        } else if (equals) {
          arg= equals+1;
        } else {
          arg= *++argv;
-         if (!arg) badusage("option --%s needs a value",o->long);
+         if (!arg) badusage("option --%s needs a value",o->lng);
        }
        o->fn(o, arg);
        break; /* eaten the whole argument now */
       }
-      for (o=options; o->long; o++)
-       if (a == o->short)
+      for (o=options; o->lng; o++)
+       if (a == o->shrt)
          goto found_short;
       badusage("unknown short option -%c",a);
     found_short:
@@ -1930,7 +2618,7 @@ int main(int argc, char **argv) {
       } else {
        if (!*++arg) {
          arg= *++argv;
-         if (!arg) badusage("option -%c needs a value",o->short);
+         if (!arg) badusage("option -%c needs a value",o->shrt);
        }
        o->fn(o,arg);
        break; /* eaten the whole argument now */
@@ -1939,9 +2627,13 @@ int main(int argc, char **argv) {
   }
 
   if (!arg) badusage("need site name argument");
-  if (*++argv) badusage("too many non-option arguments");
   sitename= arg;
 
+  if ((arg= *++argv))
+    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;
@@ -1949,7 +2641,9 @@ 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;
-  
+
+  if (!pathoutgoing)
+    pathoutgoing= innconf->pathoutgoing;
   innconf_read(inndconffile);
 
   if (!feedfile)
@@ -1958,11 +2652,22 @@ 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 metacharacter %c",c);
+
+  oop_source_sys *sysloop= oop_sys_new();
+  if (!sysloop) sysdie("could not create liboop event loop");
+  loop= (oop_source*)sysloop;
+
   if (signal(SIGPIPE, SIG_IGN) == SIG_ERR)
     sysdie("could not ignore SIGPIPE");
 
   if (become_daemon) {
+    int i;
     for (i=3; i<255; i++)
       /* do this now before we open syslog, etc. */
       close(i);
@@ -1974,7 +2679,7 @@ 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);
 
@@ -1987,7 +2692,7 @@ int main(int argc, char **argv) {
 
   notice("starting");
 
-  if (!filemon_init()) {
+  if (!filemon_method_init()) {
     warn("no file monitoring available, polling");
     filepoll_schedule();
   }
@@ -1996,5 +2701,7 @@ int main(int argc, char **argv) {
 
   statemc_init();
 
-  loop->execute.
+  void *r= oop_sys_run(sysloop);
+  assert(r == OOP_ERROR);
+  sysdie("event loop failed");
 }