chiark / gitweb /
WIP, before change sm timeout machinery to count down and general counter
[innduct.git] / backends / innduct.c
index ba4c6bd1330fea3b3797ed3d741862acde3005e2..54e61969863931b532961eb1822e5dd7895b1412 100644 (file)
  *
  */
 
+
+/*----- general definitions, probably best not changed -----*/
+
 #define PERIOD_SECONDS 60
 
-static char *feedfile;
+#define CONNCHILD_ESTATUS_STREAM   4
+#define CONNCHILD_ESTATUS_NOSTREAM 5
+
+#define INNDCOMMCHILD_ESTATUS_FAIL     6
+#define INNDCOMMCHILD_ESTATUS_NONESUCH 7
+
+
+/*----- configuration options -----*/
+
+static char *feedname, *feedfile;
 static int max_connections, max_queue_per_conn;
 static int connection_setup_timeout, port, try_stream;
+static int inndcomm_flush_timeout;
 static const char *remote_host;
 
 static double accept_proportion;
@@ -146,6 +159,9 @@ static double nocheck_thresh= 0.95;
 static double nocheck_decay= 1-1/100;
 static int nocheck, nocheck_reported;
 
+
+/*----- doubly linked lists -----*/
+
 #define ISNODE(T)    T *next, *back;
 #define LIST(T)      struct { T *head, *tail, *tailpred; int count; }
 
@@ -166,6 +182,8 @@ static int nocheck, nocheck_reported;
  (list_insert((struct list*)&(l), NODE((n)), NODE((pred))), (void)(l).count++)
 
 
+/*----- statistics -----*/
+
 #define RESULT_COUNTS                          \
   RC(offered)                                  \
   RC(sent)                                     \
@@ -184,21 +202,10 @@ typedef struct {
   int articles[2 /* checked */][RCI_max];
 } Counts;
 
-struct Article {
-  int midlen;
-  int checked, sentbody;
-  InputFile *ipf;
-  TOKEN token;
-  off_t offset;
-  int blanklen;
-  char messageid[1];
-};
 
-#define CONNIOVS 128
-
-#define CN "<%d> "
+/*----- transmission buffers -----*/
 
-typedef struct Conn Conn;
+#define CONNIOVS 128
 
 typedef enum {
   xk_Malloc, xk_Const, xk_Artdata;
@@ -212,6 +219,45 @@ typedef struct {
   } 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 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;
+
+  oop_read *rd;
+  long inprogress; /* no. of articles read but not processed */
+  off_t offset;
+} InputFile;
+
+typedef enum {
+  sm_WAITING,
+  sm_NORMAL,
+  sm_FLUSHING,
+  sm_FLUSHFAIL,
+  sm_SEPARATED1,
+  sm_SEPARATED2, /* must follow SEPARATED2 - see feedfile_eof */
+  sm_DROPPING1,
+  sm_DROPPING2, /* must follow DROPPING1 - see feedfile_eof */
+} StateMachineState;
+
 struct Conn {
   ISNODE(Conn);
   int fd, max_queue, stream;
@@ -222,28 +268,52 @@ struct Conn {
   int xmitu;
 };
 
+
+/*----- operational variables -----*/
+
+static int since_connect_attempt;
+static int nconns;
+static LIST(Conn) idle, working, full;
+static LIST(Article) *queue;
+
+static char *path_ductlock, *path_duct, *path_ductdefer;
+
+static StateMachineState sms;
 static FILE *defer;
+static InputFile *main_input_file, *old_input_file;
+static int waiting_periods_sofar;
+
+
+/*----- function predeclarations -----*/
+
+static void conn_check_work(Conn *conn);
 
 static int filemon_init(void);
 static void filemon_setfile(int mainfeed_fd, const char *mainfeed_path);
 static void filemon_callback(void);
 
 
-#define CHILD_ESTATUS_STREAM   4
-#define CHILD_ESTATUS_NOSTREAM 5
-
-static int since_connect_attempt;
-static int nconns;
-static LIST(Conn) idle, working, full;
-
-static LIST(Article) *queue;
+/*========== utility functions etc. ==========*/
 
 static void perhaps_close(int *fd) { if (*fd) { close(*fd); fd=0; } }
 
-/*========== making new connections ==========*/
+static pid_t xfork(const char *what) {
+  pid_t child;
 
-static int connecting_sockets[2]= {-1,-1};
-static pid_t connecting_child;
+  child= fork();
+  if (child==-1) sysdie("cannot fork for %s",what);
+  if (!child) postfork(what);
+  return child;
+}
+
+static void on_fd_read_except(int fd, oop_call_fd callback) {
+  loop->on_fd(loop, fd, OOP_READ,      callback, 0);
+  loop->on_fd(loop, fd, OOP_EXCEPTION, callback, 0);
+}
+static void cancel_fd_read_except(int fd) {
+  loop->cancel_fd(loop, fd, OOP_READ);
+  loop->cancel_fd(loop, fd, OOP_EXCEPTION);
+}
 
 static void report_child_status(const char *what, int status) {
   if (WIFEXITED(status)) {
@@ -264,27 +334,38 @@ 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);
+  if (r) sysdie("cannot kill %s child", what);
+
+  pid_t got= waitpid(*pid, &status, WNOHANG);
+  if (got==-1) sysdie("cannot reap %s child", what);
+
+  *pid= 0;
+
+  return status;
+}
+
+/*========== making new connections ==========*/
+
+static int connecting_sockets[2]= {-1,-1};
+static pid_t connecting_child;
+
 static void connect_attempt_discard(void) {
-  if (connecting_sockets[0]) {
-    cancel_fd(loop, connecting_sockets[0], OOP_READ);
-    cancel_fd(loop, connecting_sockets[0], OOP_EXCEPTION);
-  }
+  if (connecting_sockets[0])
+    cancel_fd(connecting_sockets[0]);
+
   perhaps_close(&connecting_sockets[0]);
   perhaps_close(&connecting_sockets[1]);
 
   if (connecting_child) {
-    int status;
-    r= kill(connecting_child, SIGKILL);
-    if (r) sysdie("cannot kill connect child");
-
-    pid_t got= waitpid(connecting_child, &status, WNOHANG);
-    if (got==-1) sysdie("cannot reap connect child");
+    int status= xwaitpid(&connecting_child, "connect");
 
     if (!(WIFEXITED(status) ||
-         (WIFSIGNALED(status) && WTERMSIG(status) == SIGKILL))) {
-      report_child_status("connect"
-    }
-    connecting_child= 0;
+         (WIFSIGNALED(status) && WTERMSIG(status) == SIGKILL)))
+      report_child_status("connect", status);
   }
 }
 
@@ -312,8 +393,8 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) {
       connecting_child= 0;
       if (WIFEXITED(status) &&
          (WEXITSTATUS(status) != 0
-          WEXITSTATUS(status) != CHILD_ESTATUS_STREAM &&
-          WEXITSTATUS(status) != CHILD_ESTATUS_NOSTREAM)) {
+          WEXITSTATUS(status) != CONNCHILD_ESTATUS_STREAM &&
+          WEXITSTATUS(status) != CONNCHILD_ESTATUS_NOSTREAM)) {
        /* child already reported the problem */
       } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALARM) {
        warn("connect: connection attempt timed out");
@@ -355,8 +436,8 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) {
   if (!WIFEXITED(status)) { report_child_status("connect",status); goto x; }
   int es= WEXITSTATUS(status);
   switch (es) {
-  case CHILD_ESTATUS_STREAM:    conn->stream= 1;   break;
-  case CHILD_ESTATUS_NOSTREAM:  conn->stream= 0;   break;
+  case CONNCHILD_ESTATUS_STREAM:    conn->stream= 1;   break;
+  case CONNCHILD_ESTATUS_NOSTREAM:  conn->stream= 0;   break;
   default:
     die("connect: child gave unexpected exit status %d", es);
   }
@@ -388,16 +469,12 @@ static void connect_start() {
   r= socketpair(AF_UNIX, SOCK_STREAM, 0, connecting_sockets);
   if (r) { syswarn("connect: cannot create socketpair for child"); goto x; }
 
-  connecting_child= fork();
-  if (connecting_child==-1) { syswarn("connect: cannot fork"); goto x; }
+  connecting_child= xfork("connection");
 
   if (!connecting_child) {
     FILE *cn_from, *cn_to;
     char buf[NNTP_STRLEN+100];
-    int exitstatus= CHILD_ESTATUS_NOSTREAM;
-
-    put sigpipe back;
-    close unwanted fds;
+    int exitstatus= CONNCHILD_ESTATUS_NOSTREAM;
 
     r= close(connecting_sockets[0]);
     if (r) sysdie("connect: close parent socket in child");
@@ -441,7 +518,7 @@ static void connect_start() {
       }
       switch (rcode) {
       case 203:
-       exitstatus= CHILD_ESTATUS_STREAM;
+       exitstatus= CONNCHILD_ESTATUS_STREAM;
        break;
       case 480:
       case 500:
@@ -472,17 +549,15 @@ static void connect_start() {
   r= close(connecting_sockets[1]);  connecting_sockets[1]= 0;
   if (r) sysdie("connect: close child socket in parent");
 
-  loop->on_fd(loop, connecting_sockets[0], OOP_READ,      connchild_event, 0);
-  loop->on_fd(loop, connecting_sockets[0], OOP_EXCEPTION, connchild_event, 0);
+  on_fd_read_except(connecting_sockets[0], connchild_event);
   return OOP_CONTINUE;
 
  x:
   connect_attempt_discard();
 }
 
-/*========== overall control of article flow ==========*/
 
-static void conn_check_work(Conn *conn);
+/*========== overall control of article flow ==========*/
 
 static void check_master_queue(void) {
   try reading current feed file;
@@ -564,8 +639,8 @@ static void conn_check_work(Conn *conn)  {
   }
 }
 
-/*========== article transmission ==========*/
 
+/*========== article transmission ==========*/
 
 static XmitDetails *xmit_core(Conn *conn, const char *data, int len,
                   XmitKind kind) { /* caller must then fill in details */
@@ -610,8 +685,7 @@ static void *conn_write_some_xmits(Conn *conn) {
     ssize_t rs= writev(conn->fd, conn->xmit, count);
     if (rs < 0) {
       if (errno == EAGAIN) return OOP_CONTINUE;
-      syswarn(CN "write failed", conn->fd);
-      conn_failed(conn);
+      connfail(conn, "write failed: %s", strerror(errno));
       return OOP_HALT;
     }
     assert(rs > 0);
@@ -684,7 +758,8 @@ static void conn_make_some_xmits(Conn *conn) {
   }
 }
 
-/*========== responses from peer ==========*/
+
+/*========== handling responses from peer ==========*/
 
 static const oop_rd_style peer_rd_style= {
   OOP_RD_DELIM_STRIP, '\n',
@@ -698,45 +773,43 @@ static Article *article_reply_check(Connection *conn, const char *response,
   Article *art= LIST_REMHEAD(conn->sent);
 
   if (!art) {
-    warn("peer gave unexpected response when no commands outstanding: %s",
-        sanitised_response);
-    goto failed;
+    connfail(conn,
+            "peer gave unexpected response when no commands outstanding: %s",
+            sanitised_response);
+    return 0;
   }
 
   if (code_indicates_streaming) {
     assert(!memchr(response, 0, 4)); /* ensured by peer_rd_ok */
     if (!conn->stream) {
-      warn("peer gave streaming response code "
-          " to IHAVE or subsequent body: %s", sanitised_response);
-      goto failed;
+      connfail("peer gave streaming response code "
+              " to IHAVE or subsequent body: %s", sanitised_response);
+      return 0;
     }
     const char *got_mid= response+4;
     int got_midlen= strcspn(got_mid, " \n\r");
     if (got_midlen<3 || got_mid[0]!='<' || got_mid[got_midlen-1]!='>') {
-      warn("peer gave streaming response with syntactically invalid"
-          " messageid: %s", sanitised_response);
-      goto failed;
+      connfail("peer gave streaming response with syntactically invalid"
+              " messageid: %s", sanitised_response);
+      return 0;
     }
     if (got_midlen != art->midlen ||
        memcmp(got_mid, art->messageid, got_midlen)) {
-      peer("peer gave streaming response code to wrong article -"
-          " probable synchronisation problem; we offered: %s; peer said: %s",
-          art->messageid, sanitised_response);
-      goto failed;
+      connfail("peer gave streaming response code to wrong article -"
+              " probable synchronisation problem; we offered: %s;"
+              " peer said: %s",
+              art->messageid, sanitised_response);
+      return 0;
     }
   } else {
     if (conn->stream) {
-      warn("peer gave non-streaming response code to CHECK/TAKETHIS: %s",
-          sanitised_response);
-      goto failed;
+      connfail("peer gave non-streaming response code to CHECK/TAKETHIS: %s",
+              sanitised_response);
+      return 0;
     }
   }
 
   return art;
-
- failed:
-  conn_failed(conn);
-  return 0;
 }
 
 static void update_nocheck(int accepted) {
@@ -777,18 +850,28 @@ 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);
+
   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) {
   Conn *conn= conn_v;
 
   if (ev == OOP_RD_EOF) {
-    warn("unexpected EOF from peer");
-    conn_failed(conn);
-    return;
+    connfail(conn, "unexpected EOF from peer");
+    return OOP_CONTINUE;
   }
   assert(ev == OOP_RD_OK);
 
@@ -807,16 +890,14 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev,
       sprintf(q,"\\x%02x",c);
       q += 4;
     }
-    warn("badly formatted response from peer: %s", sanibuf);
-    conn_failed(conn);
-    return;
+    connfail(conn, "badly formatted response from peer: %s", sanibuf);
+    return OOP_CONTINUE;
   }
 
   if (conn->quitting) {
     if (code!=205) {
-      warn("peer gave failure response to QUIT: %s", sani);
-      conn_failed(conn);
-      return;
+      connfail(conn, "peer gave failure response to QUIT: %s", sani);
+      return OOP_CONTINUE;
     }
     conn close ok;
     return;
@@ -833,10 +914,16 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev,
   GET_ARTICLE;                                                 \
   article_done(conn, art, RC_##how);  break;
 
+#define PEERBADMSG(m) connfail(conn, m ": %s", sani);  return OOP_CONTINUE
+
   int code_streaming= 0;
 
   switch (code) {
 
+  case 400: PEERBADMSG("peer stopped accepting articles");
+  case 503: PEERBADMSG("peer timed us out");
+  default:  PEERBADMSG("peer sent unexpected message");
+
   case 435: ARTICLE_DEALTWITH(0,unwanted); /* IHAVE says they have it */
   case 438: ARTICLE_DEALTWITH(1,unwanted); /* CHECK/TAKETHIS: they have it */
 
@@ -853,8 +940,8 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev,
     count_checkedwanted++;
     LIST_ADDTAIL(conn->queue);
     if (art->checked) {
-      warn("peer gave %d response to article body",code);
-      goto failed;
+      connfail("peer gave %d response to article body: %s",code, sani);
+      return OOP_CONTINUE;
     }
     art->checked= 1;
     break;
@@ -869,63 +956,23 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev,
     article_done(conn, art, RC_deferred);
     break;
 
-  case 400: warn("peer has stopped accepting articles: %s", sani); goto failed;
-  case 503: warn("peer timed us out: %s", sani);                   goto failed;
-  default:  warn("peer sent unexpected message: %s", sani);        goto failed;
-
-  failed:
-    conn_failed(conn);
-    return OOP_CONTINUE;;
   }
 
   check_check_work(conn);
   return OOP_CONTINUE;
 }
 
-/*========== monitoring of input file ==========*/
+
+/*========== monitoring of input files ==========*/
 
 static void feedfile_eof(InputFile *ipf) {
   assert(ipf != main_input_file); /* promised by tailing_try_read */
   assert(ipf == old_input_file);
-  assert(sms == sm_SEPARATED);
-  sms= sm_FINISHING;
+  assert(sms==sm_SEPARATED1 || sms==sm_DROPPING1);
+  sms++;
   inputfile_tailing_stop(ipf);
-  inputfile_tailing_start(main_input_file);
-}
-
-static void statmc_finishdone(void) {
-  time_t now;
-  struct stat stab;
-
-  assert(sms == sm_FINISHING);
-
-  r= fstat(fileno(defer), &stab);
-  if (r) sysdie("check defer file %s", path_defer);
-
-  if (fclose(defer)) sysdie("could not close defer file %s", path_defer);
-  defer= 0;
-
-  now= time(0);
-  if (now==-1) sysdie("could not get current time for backlog filename");
-
-  char *backlog= xasprintf("%s_backlog_%lu.%lu", feedfile,
-                          (unsigned long)now,
-                          (unsigned long)stab.st_ino);
-  if (link(path_defer, path_backlog))
-    sysdie("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);
-  open_defer();
-
-  close_input_file(old_input_file);
-  old_input_file= 0;
-
-  if (unlink(path_duct))
-    sysdie("could not unlink old duct file %s", path_duct);
-
-  sms= sm_NORMAL;
+  if (main_input_file)
+    inputfile_tailing_start(main_input_file);
 }
 
 static InputFile *open_input_file(const char *path) {
@@ -937,14 +984,14 @@ static InputFile *open_input_file(const char *path) {
 
   InputFile *ipf= xmalloc(sizeof(InputFile));
   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;
-  
+
   return ipf;
 }
 
@@ -958,6 +1005,7 @@ static void close_input_file(InputFile *ipf) {
   free(ipf);
 }
 
+
 /*---------- dealing with articles read in the input file ----------*/
 
 typedef void *feedfile_got_article(oop_source *lp, oop_read *rd,
@@ -995,6 +1043,9 @@ typedef void *feedfile_got_article(oop_source *lp, oop_read *rd,
   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);
+
     int r= link(feedfile, duct_path);
     if (r) sysdie("link feedfile %s to ductfile %s", feedfile, dut_path);
     /* => Hardlinked */
@@ -1009,7 +1060,8 @@ typedef void *feedfile_got_article(oop_source *lp, oop_read *rd,
   check_master_queue();
 }
 
-/*---------- tailing input file ----------*/
+
+/*========== tailing input file ==========*/
 
 static void filemon_start(InputFile *ipf) {
   assert(!ipf->filemon);
@@ -1062,8 +1114,14 @@ static ssize_t tailing_try_read(struct oop_readable *rable, void *buffer,
   InputFile *ipf= (void*)rable;
   for (;;) {
     ssize_t r= read(ipf->fd, buffer, length);
-    if (!r && ipf==main_input_file) { errno=EAGAIN; return -1; }
-    if (r==-1 && errno==EINTR) continue;
+    if (r==-1) {
+      if (errno==EINTR) continue;
+      return r;
+    }
+    if (!r) {
+      if (ipf==main_input_file) { errno=EAGAIN; return -1; }
+      assert(sms==sm_SEPARATED1 || sms==sm_DROPPING1);
+    }
     return r;
   }
 }
@@ -1188,11 +1246,12 @@ static void inputfile_tailing_stop(InputFile *ipf) {
   assert(!ipf->filemon); /* we shouldn't be monitoring it now */
 }
 
+
 /*========== interaction with innd ==========*/
 
 /* See official state diagram at top of file.  We implement
  * this as follows:
- *
+
           ================
            WAITING
           [Nothing/Noduct]
@@ -1208,7 +1267,7 @@ static void inputfile_tailing_stop(InputFile *ipf) {
      |     ========                        (ESRCH)
      |      NORMAL                        [Dropped]
      |     [Normal]                       =========
-     |      read F
+     |      main F tail
      |     ========
      |          |
      |          | F IS SO BIG WE SHOULD FLUSH
@@ -1225,83 +1284,52 @@ static void inputfile_tailing_stop(InputFile *ipf) {
      |     ==========                                                |
      |      FLUSHING                                                 |
      |     [Flushing]                                                |
-     |      read D                                                   |
+     |      main D tail                                              |
      |     ==========                                                |
      |          |                                                    |
      |          |   INNDCOMM FLUSH FAILS                             ^
-     |          |`----------------------->--------.                  |
-     |          |                                 |                  |
-     |          |   NO SUCH SITE                  V                  |
-     ^          |`----------------.            =========             |
-     |          |                 |            FLUSHFAIL             |
-     |          |                 V            [Moved]               |
-     |          |            ==========        read D                |
-     |          |             DROPPING         =========             |
-     |          |            [Dropping]           |                  |
-     |          |             read D              | TIME TO RETRY    |
-     |          |            ==========           `------------------'
-     |          | FLUSH OK        |
-     |          | open F          | AT EOF OF D AND ALL PROCESSED
-     |          V                 | install defer as backlog
-     |     ===========            | unlink D
-     |      SEPARATED             | exit
-     |     [Separated]            V
-     |      read D            ==========
-     |     ===========         (ESRCH)
-     |          |             [Droppped]
-     |          |             ==========
-     |          V
-     |          | AT EOF OF D
-     ^          |
-     |     ===========
-     |      FINISHING
-     |     [Finishing]
-     |      read F
-     |      write D
-     |     ===========
-     |          |
-     |          | ALL D PROCESSED
-     |          | install defer as backlog
-     |          | start new defer
-     ^          V unlink D
-     |          | close D
-     |          |
-     `----------'
-
- *
+     |          |`----------------------->----------.                |
+     |          |                                   |                |
+     |          |   NO SUCH SITE                    V                |
+     ^          |`--------------->----.          ===========         |
+     |          |                      \         FLUSHFAIL           |
+     |          |                       \        [Moved]             |
+     |          |                        \       main D tail         |
+     |          |                         \      ===========         |
+     |          |                          \        |                |
+     |          |                           \       | TIME TO RETRY  |
+     |          |                            \      `----------------'
+     |          | FLUSH OK                    \
+     |          | open F                       \
+     |          V                               V
+     |     =============                     ============
+     |      SEPARATED1                        DROPPING1
+     |     [Separated]                       [Dropping]
+     |      main F idle                       main none
+     |      old  D tail                       old  D tail
+     |     =============                     ============
+     |          |                                 |
+     ^          | EOF ON D                        | EOF ON D
+     |          V                                 V
+     |     =============                     ============
+     |      SEPARATED2                        DROPPING2
+     |     [Finishing]                       [Dropping]
+     |      main F tail                              main none
+     |      old  D idle                       old  D idle
+     |     =============                            ============
+     |          |                               |
+     |          | 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]
+                                             ==========
  */
 
-static char *path_ductlock, *path_duct, *path_ductdefer;
-
-typedef struct {
-  /* This is 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;
-
-  oop_read *rd;
-  long inprogress; /* no. of articles read but not processed */
-  off_t offset;
-} InputFile;
-
-typedef enum {
-  sm_WAITING,
-  sm_NORMAL,
-  sm_FLUSHING,
-  sm_FLUSHFAIL,
-  sm_DROPPING,
-  sm_SEPARATED,
-  sm_FINISHING;
-} StateMachineState;
-
-static InputFile *main_input_file, *old_input_file;
-static StateMachineState sms;
-static int waiting_periods_sofar;
-
 static void open_defer(void) {
   struct stat stab;
 
@@ -1434,8 +1462,7 @@ static void statemc_init(void) {
 }
 
 static void statemc_poll(void) {
-  if (sms == sm_WAITING) statemc_waiting_poll();
-  if (sms == sm_FINISHING && !old_input_file->inprogress) statemc_finishdone();
+  if (sms==sm_WAITING) statemc_waiting_poll();
 }
 
 static void statemc_waiting_poll(void) {
@@ -1455,12 +1482,167 @@ static void startup_set_input_file(InputFile *f) {
   inputfile_tailing_start(f);
 }
 
+static void *statemc_check_oldinput_done(oop_source *lp,
+                                        struct timeval now, void *u) {
+  struct stat stab;
+
+  int done= (sms==sm_SEPARATED2 || sms==sm_DROPPING2)
+         && old_input_file->inprogress;
+  if (!done) return;
+
+  r= fstat(fileno(defer), &stab);
+  if (r) sysdie("check defer file %s", path_defer);
+
+  if (fclose(defer)) sysdie("could not close defer file %s", path_defer);
+  defer= 0;
+
+  char *backlog= xasprintf("%s_backlog_%lu.%lu", feedfile,
+                          (unsigned long)now.tv_sec,
+                          (unsigned long)stab.st_ino);
+  if (link(path_defer, path_backlog))
+    sysdie("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);
+
+  if (sms==sm_DROPPING2) {
+    notice("feed dropped and our work is complete"
+          " (but check for backlog files)");
+    exit(0);
+  }
+
+  open_defer();
+
+  close_input_file(old_input_file);
+  old_input_file= 0;
+
+  sms= sm_NORMAL;
+}
+
 /*========== flushing the feed ==========*/
 
+static pid_t inndcomm_child;
+
+static void inndcommfail(const char *what) {
+  syswarn("error communicating with innd: %s failed: %s", what, ICCfailure);
+  exit(INNDCOMMCHILD_ESTATUS_FAIL);
+}
+
+static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) {
+  assert(inndcomm_child);
+  int status= xwaitpid(&inndcomm_child, "inndcomm");
+  loop->cancel_fd(fd);
+  close(fd);
+
+  assert(!old_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;
+      main_input_file= 0;
+      sms= sm_DROPPING1;
+      return OOP_CONTINUE;
+
+    case 0:
+      old_input_file= main_input_file;
+      main_input_file= open_input_file(feedfile);
+      if (!main_input_file)
+       die("flush succeeded but feedfile %s does not exist!", feedfile);
+      sms= sm_SEPARATED1;
+      return OOP_CONTINUE;
+
+    default:
+      goto unexpected_exitstatus;
+      
+    }
+  } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) {
+    warn("flush of %s timed out trying to talk to innd", feedname);
+    goto failed;
+  } else {
+  unexpected_exitstatus:
+    report_child_status("inndcomm child", status);
+  }
+
+ failed:
+  
+
+void spawn_inndcomm_flush(void) {
+  int pipefds[2];
+
+  assert(sms == sm_NORMAL);
+  assert(!inndcomm_child);
+
+  if (pipe(pipefds)) sysdie("create pipe for inndcomm child sentinel");
+
+  inndcomm_child= xfork();
+
+  if (!inndcomm_child) {
+    static char flushargv[2]= { feedname, 0 };
+    char *reply;
+
+    close(pipefds[0]);
+
+    alarm(inndcomm_flush_timeout);
+    r= ICCopen();                         if (r)   inndcommfail("connect");
+    r= ICCcommand('f',flushargv,&reply);  if (r<0) inndcommfail("transmit");
+    if (!r) exit(0); /* yay! */
+
+    if (!strcmp(reply, "1 No such site")) exit(INNDCOMMCHILD_ESTATUS_NONESUCH);
+    syswarn("innd ctlinnd flush %s failed: innd said %s", feedname, reply);
+    exit(INNDCOMMCHILD_ESTATUS_FAIL);
+  }
+
+  close(pipefds[1]);
+  int sentinel_fd= pipefds[0];
+  on_fd_read_except(sentinel_fd, inndcomm_event);
+
+  sms= sm_FLUSHING;
+}
 
 /*========== main program ==========*/
 
+static void postfork_inputfile(InputFile *ipf) {
+  if (!ipf) return;
+  assert(ipf->fd >= 0);
+  close(ipf->fd);
+  ipf->fd= -1;
+}
+
+static void postfork_conns(Connection *conn) {
+  while (conn) {
+    close(conn->fd);
+    conn= conn->next;
+  }
+}
+
+static void postfork_stdio(FILE *f) {
+  /* we have no stdio streams that are buffered long-term */
+  if (f) fclose(f);
+}
+
+static void postfork(const char *what) {
+  if (signal(SIGPIPE, SIG_DFL) == SIG_ERR)
+    sysdie("%s child: failed to reset SIGPIPE");
+
+  postfork_inputfile(main_input_file);
+  postfork_inputfile(old_input_file);
+  postfork_conns(idle.head);
+  postfork_conns(working.head);
+  postfork_conns(full.head);
+  postfork_stdio(defer);
+}
+
+
 #define EVERY(what, interval, body)                                         \
   static const struct timeval what##_timeout = { 5, 0 };                    \
   static void what##_schedule(void);                                        \