* 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
+ * may be symlinks (in which case links
+ * may be written through, but only links
+ * will be removed.
+ *
+ * It is safe to remove backlog files manually,
+ * if it's desired to throw away the backlog.
+ *
+ * Backlog files are also processed by innduct. We find the oldest
+ * backlog file which is at least a certain amount old, and feed it
+ * back into our processing. When every article in it has been read
+ * and processed, we unlink it and look for another backlog file.
+ *
+ * If we don't have a backlog file that we're reading, we close the
+ * defer file that we're writing and make it into a backlog file at
+ * the first convenient opportunity.
OVERALL STATES:
/*----- 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 char *sitename, *feedfile;
+static int max_connections=10, max_queue_per_conn=200;
+static int connection_setup_timeout=200, port=119, try_stream=1;
+static int inndcomm_flush_timeout=100, quiet_if_locked=0;
static const char *remote_host;
static int reconnect_delay_periods, flushfail_retry_periods, open_wait_periods;
+static int backlog_retry_minperiods, backlog_spontaneous_rescan_periods;
+static const char *inndconffile;
static double accept_proportion;
-static double nocheck_thresh= 0.95;
-static double nocheck_decay= 1-1/100;
+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;
char messageid[1];
};
-typedef struct {
+typedef struct InputFile {
/* This is an instance of struct oop_readable */
struct oop_readable readable; /* first */
oop_readable_call *readable_callback;
oop_read *rd;
long inprogress; /* no. of articles read but not processed */
off_t offset;
+
+ Counts counts;
} InputFile;
typedef enum {
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 */
+ sm_SEPARATED,
+ sm_DROPPING,
} StateMachineState;
struct Conn {
static LIST(Conn) idle, working, full;
static LIST(Article) *queue;
-static char *path_ductlock, *path_duct, *path_ductdefer;
+static char *path_lock, *path_flushing, *path_defer;
+
+#define SMS(newstate, periods, why) \
+ (statemc_setstate(sm_##newstate,(periods),#newstate,(why)))
static StateMachineState sms;
static FILE *defer;
-static InputFile *main_input_file, *old_input_file;
+static InputFile *main_input_file, *old_input_file, *backlog_input_file;
static int sm_period_counter;
static void filemon_setfile(int mainfeed_fd, const char *mainfeed_path);
static void filemon_callback(void);
+static void statemc_setstate(StateMachineState newsms, int periods,
+ const char *forlog, const char *why);
/*========== utility functions etc. ==========*/
return status;
}
+static void check_isreg(const struct stat *stab, const char *path,
+ const char *what) {
+ if (!S_ISREG(stab->st_mode))
+ die("%s %s not a plain file (mode 0%lo)",
+ what, path, (unsigned long)stab->st_mode);
+}
+
+static void xfstat(int fd, struct stat *stab_r, const char *what) {
+ int r= fstab(path, stab);
+ if (r) sysdie("could not fstat %s %s", what, path);
+}
+
+static void xfstat_isreg(int fd, struct stat *stab_r, const char *what) {
+ xfstat(fd, stab_r, what);
+ check_isreg(stab, path, what);
+}
+
+static void xlstat_isreg(const char *path, struct stat *stab,
+ int *enoent_r /* 0 means ENOENT is fatal */
+ const char *what) {
+ int r= lstat(path, stab);
+ if (r) {
+ if (errno==ENOENT && enoent_r) { *enoent_r=1; return; }
+ sysdie("could not lstat %s %s", what, path);
+ }
+ if (enoent_r) *enoent_r= 0;
+ check_isreg(stab, path, what);
+}
+
+static int samefile(const struct stat *a, const struct stat *b) {
+ assert(S_ISREG(a->st_mode));
+ assert(S_ISREG(b->st_mode));
+ return (a->st_ino == b->st_ino &&
+ a->st_dev == b->st_dev);
+}
+
+/*========== logging ==========*/
+
+static void logcore(int sysloglevel, const char *fmt, ...)
+ __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 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;
+
+ 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 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); \
+ }
+
+logwrap(sysdie, " critical", LOG_CRIT, errno, 16);
+logwrap(die, " critical", LOG_CRIT, -1, 16);
+
+logwrap(sysfatal, " fatal", LOG_ERR, errno, 12);
+logwrap(fatal, " fatal", LOG_ERR, -1, 12);
+
+logwrap(syswarn, " warning", LOG_WARN, errno, 0);
+logwrap(warn, " warning", LOG_WARN, -1, 0);
+
+logwrap(notice, "", LOG_NOTICE, -1, 0);
+logwrap(info, " info", LOG_INFO, -1, 0);
+
+
/*========== making new connections ==========*/
static int connecting_sockets[2]= {-1,-1};
static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) {
Conn *conn= 0;
- conn= xcalloc(sizeof(*conn));
+ conn= xmalloc(sizeof(*conn));
+ memset(conn,0,sizeof(*conn));
DECL_MSG_CMSG(msg);
struct cmsghdr *h= 0;
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);
+ fatal("connect: child gave unexpected exit status %d", es);
}
set nonblocking;
if (NNTPconnect(remote_host, port, &cn_from, &cn_to, buf) < 0) {
if (buf[0]) {
sanitise_inplace(buf);
- die("connect: rejected: %s", buf);
+ fatal("connect: rejected: %s", buf);
} else {
- sysdie("connect: connection attempt failed");
+ sysfatal("connect: connection attempt failed");
}
}
if (NNTPsendpassword(remote_host, cn_from, cn_to) < 0)
- sysdie("connect: authentication failed");
+ sysfatal("connect: authentication failed");
if (try_stream) {
if (fputs("MODE STREAM\r\n", cn_to) ||
fflush(cn_to))
- sysdie("connect: could not send MODE STREAM");
+ sysfatal("connect: could not send MODE STREAM");
buf[sizeof(buf)-1]= 0;
if (!fgets(buf, sizeof(buf)-1, cn_from)) {
if (ferror(cn_from))
- sysdie("connect: could not read response to MODE STREAM");
+ sysfatal("connect: could not read response to MODE STREAM");
else
- die("connect: connection close in response to MODE STREAM");
+ fatal("connect: connection close in response to MODE STREAM");
}
int l= strlen(buf);
assert(l>=1);
if (buf[-1]!='\n') {
sanitise_inplace(buf);
- die("connect: response to MODE STREAM is too long: %.100s...",
+ fatal("connect: response to MODE STREAM is too long: %.100s...",
remote_host, buf);
}
l--; if (l>0 && buf[1-]=='\r') l--;
int rcode= strtoul(buf,&ep,10);
if (ep != buf[3]) {
sanitise_inplace(buf);
- die("connect: bad response to MODE STREAM: %.50s", buf);
+ fatal("connect: bad response to MODE STREAM: %.50s", buf);
}
switch (rcode) {
case 203:
/*========== overall control of article flow ==========*/
static void check_master_queue(void) {
- try reading current feed file;
-
if (!queue.count)
return;
art->sent= 1;
LIST_ADDTAIL(conn->sent, art);
- counts[art->checked].sent++;
+ art->ipf->counts[art->checked].sent++;
} else {
/* check it */
XMIT_LITERAL("\r\n");
LIST_ADDTAIL(conn->sent, art);
- counts[art->checked].offered++;
+ art->ipf->counts[art->checked].offered++;
}
}
}
OOP_RD_SHORTREC_FORBID
};
+static void *peer_rd_err(oop_source *lp, oop_read *oread, oop_event ev,
+ const char *errmsg, int errnoval,
+ const char *data, size_t recsz, void *conn_v) {
+ Conn *conn= conn_v;
+ connfail(conn, "error receiving from peer: %s", errmsg);
+ return OOP_CONTINUE;
+}
+
static Article *article_reply_check(Connection *conn, const char *response,
int code_indicates_streaming,
const char *sanitised_response) {
static void article_done(Connection *conn, Article *art, int whichcount) {
*count++;
- counts.articles[art->checked][whichcount]++;
+ art->ipf->counts.articles[art->checked][whichcount]++;
if (whichcount == RC_accepted) update_nocheck(1);
else if (whichcount == RC_unwanted) update_nocheck(0);
assert(ipf->inprogress >= 0);
if (!ipf->inprogress)
- loop->on_time(loop, OOP_TIME_NOW, statemc_check_oldinput_done, 0);
+ loop->on_time(loop, OOP_TIME_NOW, statemc_check_input_done, ipf);
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) {
code_streaming= 1;
case 436: /* IHAVE says try later */
GET_ARTICLE;
+ open_defer();
if (fprintf(defer, "%s %s\n", TokenToText(art->token), art->messageid) <0
|| fflush(defer))
- sysdie("write to defer file %s",path_ductdefer);
+ sysfatal("write to defer file %s",path_defer);
article_done(conn, art, RC_deferred);
break;
static void feedfile_eof(InputFile *ipf) {
assert(ipf != main_input_file); /* promised by tailing_try_read */
+
+ inputfile_tailing_stop(ipf);
+
+ if (ipf == backlog_input_file) {
+ assert(ipf->fd >= 0);
+ if (close(ipf->fd)) sysdie("could not close backlog file %s", ipf->path);
+ ipf->fd= -1;
+ return;
+ }
+
assert(ipf == old_input_file);
- assert(sms==sm_SEPARATED1 || sms==sm_DROPPING1);
- sms++;
+
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);
}
int fd= open(path, O_RDONLY);
if (fd<0) {
if (errno==ENOENT) return 0;
- sysdie("unable to open input file %s", path);
+ sysfatal("unable to open input file %s", path);
}
InputFile *ipf= xmalloc(sizeof(InputFile));
assert(!ipf->rd); /* must have had inputfile_tailing_stop */
assert(!ipf->inprogress); /* no dangling pointers pointing here */
- if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path);
+ if (ipf->fd >= 0)
+ if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path);
+
+ fixme maybe free ipf->path;
+
free(ipf);
}
(unsigned long)ipf->offset, (unsigned long)flush_threshold);
int r= link(feedfile, duct_path);
- if (r) sysdie("link feedfile %s to ductfile %s", feedfile, dut_path);
+ if (r) sysdie("link feedfile %s to flushing file %s", feedfile, dut_path);
/* => Hardlinked */
r= unlink(feedfile);
return r;
}
if (!r) {
- if (ipf==main_input_file) { errno=EAGAIN; return -1; }
- assert(sms==sm_SEPARATED1 || sms==sm_DROPPING1);
+ if (ipf==main_input_file) {
+ errno=EAGAIN;
+ return -1;
+ } else if (ipf==old_input_file) {
+ assert(ipf->fd>=0);
+ assert(sms==sm_SEPARATED || sms==sm_DROPPING);
+ } else if (ipf==backlog_input_file) {
+ assert(ipf->fd>=0);
+ } else {
+ abort();
+ }
}
return r;
}
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:
poll for F
================
|
- | TIMEOUT
+ | TIMEOUT and no defer, no backlog
|`--------------------------.
- | | install defer as backlog
+ | |
| OPEN F SUCCEEDS | exit
,--------->| V
| V =========
| | open F \
| V V
| ============= ============
- | SEPARATED1 DROPPING1
+ | 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
- | ============= ============
- | SEPARATED2 DROPPING2
+ | =============== ===============
+ | SEPARATED/ DROPPING/
+ | old->fd==-1 old->fd==-1
| [Finishing] [Dropping]
| main F tail main none
- | old D idle old D idle
- | ============= ============
+ | 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
- `----------' ==========
+ | | | exit
+ `----------' V
+ ==========
(ESRCH)
[Droppped]
==========
*/
-static void open_defer(void) {
- struct stat stab;
-
- assert(!defer);
- defer= fopen(path_ductdefer, "a+");
- if (!defer) sysdie("could not open defer file %s", path_ductdefer);
-
- /* truncate away any half-written records */
-
- r= fstat(fileno(defer), &stab);
- if (r) sysdie("could not stat newly opened defer file %s", path_ductdefer);
-
- if (stab.st_size > LONG_MAX)
- die("defer file %s size is far too large", path_ductdefer);
+static void statemc_init(void) {
+ struct stat stab, stabf;
+ int noent;
- if (!stab.st_size)
- return;
+ path_lock= xasprintf("%s_lock", feedfile);
+ path_flushing= xasprintf("%s_flushing", feedfile);
+ path_defer= xasprintf("%s_defer", feedfile);
+ globpat_backlog= xasprintf("%s_backlog*", feedfile);
- long orgsize= stab.st_size;
- long truncto= stab.st_size;
for (;;) {
- if (!truncto) break; /* was only (if anything) one half-truncated record */
- if (fseek(defer, truncto-1, SEEK_SET) < 0)
- sysdie("seek in defer file %s while truncating partial", path_ductdefer);
-
- r= getc(defer);
- if (r==EOF) {
- if (ferror(defer))
- sysdie("failed read from defer file %s", path_ductdefer);
- else
- die("defer file %s shrank while we were checking it!", path_ductdefer);
+ int lockfd= open(path_lock, O_CREAT|O_RDWR, 0600);
+ if (lockfd<0) sysfatal("open lockfile %s", path_lock);
+
+ struct flock fl;
+ memset(&fl,0,sizeof(fl));
+ fl.l_type= F_WRLCK;
+ fl.l_whence= SEEK_SET;
+ r= fcntl(lockfd, F_SETLK, &fl);
+ if (r==-1) {
+ if (errno==EACCES || errno==EAGAIN) {
+ if (quiet_if_locked) exit(0);
+ fatal("another duct holds the lockfile");
+ }
+ sysdie("fcntl F_SETLK lockfile %s", path_lock);
}
- if (r=='\n') break;
- truncto--;
- }
- if (stab.st_size != truncto) {
- warn("truncating half-record at end of defer file %s -"
- " shrinking by %ld bytes from %ld to %ld",
- path_ductdefer, orgsize - truncto, orgsize, truncto);
+ xfstat_isreg(lockfd, &stabf, "lockfile");
+ xlstat_isreg(path_lock, &stab, &noent, "lockfile");
- if (fflush(defer)) sysdie("could not flush defer file %s", path_ductdefer);
- if (ftruncate(fileno(defer), truncto))
- sysdie("could not truncate defer file %s", path_ductdefer);
+ if (!noent && samefile(&stab, &stabf))
+ break;
- } else {
- info("continuing existing defer file %s (%ld bytes)",
- path_ductdefer, orgsize);
+ if (close(lockfd))
+ sysdie("could not close stale lockfile %s", path_lock);
}
- if (fseek(defer, truncto, SEEK_SET))
- sysdie("could not seek to new end of defer file %s", path_ductdefer);
-}
+ 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);
+ xlstat_isreg(path_defer, &stab, &noent, "defer file");
+ if (noent) {
+ debug("startup: ductdefer ENOENT");
} else {
- if (!S_ISREG(stab.st_mode))
- die("defer file %s not a plain file (mode 0%lo)",
- path_defer, (unsigned long)stab.st_mode);
+ debug("startup: ductdefer nlink=%ld", (long)stab.st_nlink);
switch (stab.st_nlink==1) {
- case 1: /* ok */ break;
+ case 1:
+ open_defer(); /* so that we will later close it and rename it */
+ break;
case 2:
if (unlink(path_defer))
sysdie("could not unlink stale defer file link %s (presumably"
path_defer, stab.st_nlink);
}
}
- open_defer();
-
- int lockfd= open(path_ductlock, O_CREAT|O_RDWR, 0600);
- if (lockfd<0) sysdie("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)
- die("another duct holds the lockfile");
- sysdie("fcntl F_SETLK lockfile %s", path_ductlock);
- }
- InputFile *file_d= open_input_file(path_duct);
+ InputFile *file_d= open_input_file(path_flushing);
if (file_d) {
struct stat stab_f, stab_d;
- r= stat(feedfile, &stab_f);
- if (r) {
- if (errno!=ENOENT) sysdie("check feed file %s", feedfile);
- /* D exists, F ENOENT => Moved */
+ xlstat_isreg(feedfile, &stab_f, &noent, "feed file");
+ if (noent) {
+ debug("startup: D exists, F ENOENT => Moved");
goto found_moved;
}
- /* F and D both exist */
+ debug("startup: F and D both exist");
- r= fstat(file_d->fd, &stab_d);
- if (r) sysdie("check duct file %s", ductfile);
+ xfstat_isreg(file_d->fd, &stab_d, "flushing file");
- if (stab_d.st_ino == stab_f.st_ino &&
- stab_d.st_dev == stab_f.st_dev) {
- /* F==D => Hardlinked*/
- r= unlink(path_duct);
+ if (samefile(&stab_d, &stab_f)) {
+ debug("startup: F==D => Hardlinked");
+ r= unlink(path_flushing);
if (r) sysdie("unlink feed file %s during startup", feedfile);
found_moved:
- /* => Moved */
+ debug(" => Moved");
startup_set_input_file(file_d);
spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */
} else {
- /* F!=D => Separated */
- sms= sm_SEPARATED;
+ debug("F!=D => Separated");
+ SMS(SEPARATED, 0, "found both old and current feed files");
startup_set_input_file(file_d);
}
- } else { /*!file_d*/
- sms= sm_WAITING;
- sm_period_counter= open_wait_periods;
+ } else {
+ debug("startup: D ENOENT => Nothing");
+ fixme need to try flushing innd here - needs state diagram changes;
+ SMS(WAITING, open_wait_periods, "no feed file currently exists");
}
}
static void statemc_poll(void) {
- if (sms==sm_WAITING)
- statemc_waiting_poll();
+ if (sms==sm_WAITING) { statemc_waiting_poll(); return; }
if (!sm_period_counter) return;
sm_period_counter--;
if (sm_period_counter) return;
switch (sms) {
case sm_WAITING:
- die("timed out waiting for innd to create feed file %s", feedfile);
+ fatal("timed out waiting for innd to create feed file %s", feedfile);
case sm_FLUSHFAIL:
spawn_inndcomm_flush(void);
break;
InputFile *file_f= open_input_file(feedfile);
if (!file_f) return;
startup_set_input_file(file_d);
- sms= sm_NORMAL;
+ SMS(NORMAL, 0, "found and opened feed file");
}
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) {
+static void *statemc_check_input_done(oop_source *lp,
+ struct timeval now, void *ipf_v) {
+ InputFile *ipf= ipf_v;
struct stat stab;
- int done= (sms==sm_SEPARATED2 || sms==sm_DROPPING2)
- && old_input_file->inprogress;
- if (!done) return;
+ if (ipf->inprogress) return; /* new article in the meantime */
+ if (ipf->fd >= 0); return; /* not had EOF */
+
+ if (ipf == backlog_input_file) {
+ notice_processed(ipf,"backlog file",ipf->path);
+ close_input_file(ipf);
+ if (unlink(ipf->path)) {
+ if (errno != ENOENT)
+ sysdie("could not unlink processed backlog file %s", ipf->path);
+ warn("backlog file %s vanished while we were reading it"
+ " so we couldn't remove it (but it's done now, anyway)",
+ ipf->path);
+ }
+ backlog_input_file= 0;
+ search_backlog_file();
+ return;
+ }
+
+ assert(ipf == old_input_file);
+ assert(sms==sm_SEPARATED || sms==sm_DROPPING);
+
+ notice_processed(ipf,"feed file",0);
+
+ close_defer();
- r= fstat(fileno(defer), &stab);
- if (r) sysdie("check defer file %s", path_defer);
+ if (unlink(path_flushing))
+ sysdie("could not unlink old flushing file %s", path_flushing);
+
+ if (sms==sm_DROPPING) {
+ if (search_backlog_file()) {
+ debug("feed dropped but still backlogs to process");
+ return;
+ }
+ notice("feed dropped and our work is complete");
+ r= unlink(path_lock);
+ if (r) sysdie("unlink lockfile for old feed %s", path_lock);
+ exit(0);
+ }
+
+ open_defer();
+
+ close_input_file(old_input_file);
+ old_input_file= 0;
+
+ notice("flush complete");
+ SMS(NORMAL, 0, "flush complete");
+}
+
+static void statemc_setstate(StateMachineState newsms, int periods,
+ const char *forlog, const char *why) {
+ sms= newsms;
+ sm_period_counter= periods;
+ if (periods) {
+ info("%s[%d] %s",periods,forlog,why);
+ } else {
+ info("%s %s",forlog,why);
+ }
+}
- if (fclose(defer)) sysdie("could not close defer file %s", path_defer);
+/*---------- defer and backlog files ----------*/
+
+static void open_defer(void) {
+ struct stat stab;
+
+ if (defer) return;
+
+ defer= fopen(path_defer, "a+");
+ if (!defer) sysfatal("could not open defer file %s", path_defer);
+
+ /* truncate away any half-written records */
+
+ xfstat_isreg(fileno(defer), &stab, "newly opened defer file");
+
+ if (stab.st_size > LONG_MAX)
+ die("defer file %s size is far too large", path_defer);
+
+ if (!stab.st_size)
+ return;
+
+ long orgsize= stab.st_size;
+ long truncto= stab.st_size;
+ for (;;) {
+ if (!truncto) break; /* was only (if anything) one half-truncated record */
+ if (fseek(defer, truncto-1, SEEK_SET) < 0)
+ sysdie("seek in defer file %s while truncating partial", path_defer);
+
+ r= getc(defer);
+ if (r==EOF) {
+ if (ferror(defer))
+ sysdie("failed read from defer file %s", path_defer);
+ else
+ die("defer file %s shrank while we were checking it!", path_defer);
+ }
+ if (r=='\n') break;
+ truncto--;
+ }
+
+ if (stab.st_size != truncto) {
+ warn("truncating half-record at end of defer file %s -"
+ " shrinking by %ld bytes from %ld to %ld",
+ path_defer, orgsize - truncto, orgsize, truncto);
+
+ if (fflush(defer))
+ sysfatal("could not flush defer file %s", path_defer);
+ if (ftruncate(fileno(defer), truncto))
+ sysdie("could not truncate defer file %s", path_defer);
+
+ } else {
+ info("continuing existing defer file %s (%ld bytes)",
+ path_defer, orgsize);
+ }
+ if (fseek(defer, truncto, SEEK_SET))
+ sysdie("could not seek to new end of defer file %s", path_defer);
+}
+
+static void close_defer(void) {
+ if (!defer)
+ return;
+
+ xfstat(fileno(defer), &stab, "defer file");
+
+ if (fclose(defer)) sysfatal("could not close defer file %s", path_defer);
defer= 0;
char *backlog= xasprintf("%s_backlog_%lu.%lu", feedfile,
(unsigned long)now.tv_sec,
(unsigned long)stab.st_ino);
if (link(path_defer, path_backlog))
- sysdie("could not install defer file %s as backlog file %s",
+ 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);
+ if (backlog_nextscan_periods < 0 ||
+ backlog_nextscan_periods > backlog_retry_minperiods + 1)
+ backlog_nextscan_periods= backlog_retry_minperiods + 1;
+}
- if (sms==sm_DROPPING2) {
- notice("feed dropped and our work is complete"
- " (but check for backlog files)");
- exit(0);
+static void poll_backlog_file(void) {
+ if (backlog_nextscan_periods < 0) return;
+ if (backlog_nextscan_periods-- > 0) return;
+ search_backlog_file();
+}
+
+static int search_backlog_file(void) {
+ /* returns non-0 iff there are any backlog files */
+
+ glob_t gl;
+ int r;
+ struct stat stab;
+ const char *oldest_path=0;
+ time_t oldest_mtime, now;
+
+ if (backlog_input_file) return 3;
+
+ try_again:
+
+ r= glob(globpat_backlog, GLOB_ERR|GLOB_MARK|GLOB_NOSORT, 0, &gl);
+
+ switch (r) {
+ case GLOB_ABORTED:
+ sysdie("failed to expand backlog pattern %s", globpat_backlog);
+ case GLOB_NOSPACE:
+ die("out of memory expanding backlog pattern %s", globpat_backlog);
+ case 0:
+ for (i=0; i<gl.gl_pathc; i++) {
+ const char *path= gl.gl_pathv[i];
+
+ if (strchr(path,'#') || strchr(path,'~')) {
+ debug("backlog file search skipping %s", path);
+ continue;
+ }
+ r= stat(path, &stab);
+ if (r) {
+ syswarn("failed to stat backlog file %s", path);
+ continue;
+ }
+ if (!S_ISREG(stab.st_mode)) {
+ warn("backlog file %s is not a plain file (or link to one)", path);
+ continue;
+ }
+ if (!oldest_path || stab.st_mtime < oldest_mtime) {
+ oldest_path= path;
+ oldest_mtime= stab.st_mtime;
+ }
+ }
+ case GLOB_NOMATCH: /* fall through */
+ break;
+ default:
+ sysdie("glob expansion of backlog pattern %s gave unexpected"
+ " nonzero (error?) return value %d", globpat_backlog, r);
}
- open_defer();
+ globfree(&gl);
- close_input_file(old_input_file);
- old_input_file= 0;
+ if (!oldest_path) {
+ debug("backlog scan: none");
+ backlog_nextscan_periods= backlog_spontaneous_rescan_periods;
+ return 0;
+ }
- notice("flush complete");
+ now= time(); if (now==-1) sysdie("time(2) failed");
+ double age= difftime(now, oldest_mtime);
+ long age_deficiency= (backlog_retry_minperiods * PERIOD_SECONDS) - age;
- sms= sm_NORMAL;
+ if (age_deficiency <= 0) {
+ debug("backlog scan: found age=%f deficiency=%ld oldest=%s",
+ age, age_deficiency, oldest_path);
+
+ backlog_input_file= open_input_file();
+ if (!backlog_input_file) {
+ warn("backlog file %s vanished as we opened it", backlog_input_file);
+ goto try_again;
+ }
+ inputfile_tailing_start(backlog_input_file);
+ backlog_nextscan_periods= -1;
+ return 1;
+ }
+
+ backlog_nextscan_periods= age_deficiency / PERIOD_SECONDS;
+
+ if (backlog_spontaneous_rescan_periods >= 0 &&
+ backlog_nextscan_periods > backlog_spontaneous_rescan_periods)
+ backlog_nextscan_periods= backlog_spontaneous_rescan_periods;
+
+ debug("backlog scan: young age=%f deficiency=%ld nextscan=%d oldest=%s",
+ age, age_deficiency, backlog_nextscan_periods, oldest_path);
+ return 2;
}
/*========== flushing the feed ==========*/
static pid_t inndcomm_child;
-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");
if (WIFEXITED(status)) {
switch (WEXITSTATUS(status)) {
-
+
case INNDCOMMCHILD_ESTATUS_FAIL:
goto failed;
warn("feed has been dropped by innd, finishing up");
old_input_file= main_input_file;
main_input_file= 0;
- sms= sm_DROPPING1;
+ SMS(DROPPING, 0, "dropped by innd");
return OOP_CONTINUE;
case 0:
main_input_file= open_input_file(feedfile);
if (!main_input_file)
die("flush succeeded but feedfile %s does not exist!", feedfile);
- sms= sm_SEPARATED1;
+ SMS(SEPARATED, 0, "feed file missing");
return OOP_CONTINUE;
default:
goto unexpected_exitstatus;
-
+
}
} else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) {
warn("flush timed out trying to talk to innd");
}
failed:
- sm_period_counter= flushfail_retry_periods;
- sms= sm_FLUSHFAIL;;
+ SMS(FLUSHFAIL, flushfail_retry_periods, "flush failed, will retry");
+}
+
+static void inndcommfail(const char *what) {
+ syswarn("error communicating with innd: %s failed: %s", what, ICCfailure);
+ exit(INNDCOMMCHILD_ESTATUS_FAIL);
}
void spawn_inndcomm_flush(void) {
inndcomm_child= xfork();
if (!inndcomm_child) {
- static char flushargv[2]= { feedname, 0 };
+ static char flushargv[2]= { sitename, 0 };
char *reply;
close(pipefds[0]);
int sentinel_fd= pipefds[0];
on_fd_read_except(sentinel_fd, inndcomm_event);
- sms= sm_FLUSHING;
+ SMS(FLUSHING, 0, "flush is in progress");
}
/*========== main program ==========*/
postfork_stdio(defer);
}
-
#define EVERY(what, interval, body) \
static const struct timeval what##_timeout = { 5, 0 }; \
static void what##_schedule(void); \
loop->on_time(loop, what##_timeout, what##_timedout, 0); \
}
-EVERY(filepoll, {5,0}, { check_master_queue(); })
+EVERY(filepoll, {5,0}, {
+ if (main_input_file && main_input_file->readable_callback)
+ filemon_callback(main_input_file);
+});
EVERY(period, {PERIOD_SECONDS,0}, {
if (connect_delay) connect_delay--;
+ poll_backlog_file();
+ if (!backlog_input_file) close_defer(); /* want to start on a new backlog */
statemc_poll();
check_master_queue();
});
-main {
- ignore sigpipe;
- if (!filemon_init())
+
+/*========== option parsing ==========*/
+
+enum OptFlags {
+ of_seconds= 001000u;
+ of_boolean= 002000u;
+};
+
+typedef struct Option Option;
+typedef void OptionParser(const Option*, const char *val);
+
+struct Option {
+ int short;
+ const char *long;
+ void *store;
+ OptionParser *fn;
+ int noarg;
+};
+
+void op_integer(const Option *o, const char *val) {
+ char *ep;
+ errno= 0;
+ unsigned long ul= strtoul(val,&ep,10);
+ if (*ep || ep==val || errno || ul>INT_MAX)
+ badusage("bad integer value for %s",o->long);
+ int *store= o->store;
+ *store= ul;
+}
+
+void op_double(const Option *o, const char *val) {
+ int *store= o->store;
+ char *ep;
+ errno= 0;
+ *store= strtod(val, &ep);
+ if (*ep || ep==val || errno)
+ badusage("bad floating point value for %s",o->long);
+}
+
+void op_string(const Option *o, const char *val) {
+ char **store= o->store;
+ free(*store);
+ *store= val;
+}
+
+void op_seconds(const Option *o, const char *val) {
+ int *store= o->store;
+ char *ep;
+
+ double v= strtod(val,&ep);
+ if (ep==val) badusage("bad time/duration value for %s",o->long);
+
+ if (!*ep || !strcmp(ep,"s")) unit= 1;
+ else if (!strcmp(ep,"m")) unit= 60;
+ else if (!strcmp(ep,"h")) unit= 3600;
+ else if (!strcmp(ep,"d")) unit= 86400;
+ else badusage("bad units %s for time/duration value for %s",ep,o->long);
+
+ v *= unit;
+ v= ceil(v);
+ if (v > INT_MAX) badusage("time/duration value for %s out of range",o->long);
+ *store= v;
+}
+
+void op_periods_rndup(const Option *o, const char *val) {
+ int *store= o->store;
+ op_seconds(o,val);
+ *store += PERIOD_SECONDS-1;
+ *store /= PERIOD_SECONDS;
+}
+
+void op_periods_booltrue(const Option *o, const char *val) {
+ int *store= o->store;
+ *store= 1;
+}
+void op_periods_boolfalse(const Option *o, const char *val) {
+ int *store= o->store;
+ *store= 0;
+}
+
+static const Option options[]= {
+{ 0, "max-connections", &max_connections op_integer },
+{ 0, "streaming", &try_stream, op_booltrue, 1 },
+{ 0, "no-streaming", &try_stream, op_boolfalse, 1 },
+{'h',"host", &remote_host, op_string },
+{'P',"port", &port op_integer },
+{ 0, "inndconf", &inndconffile, op_string },
+{'f',"feedfile", &feedfile, op_string },
+{'q',"quiet-multiple", &quiet_if_locked, op_booltrue, 1 },
+{ 0, "no-quiet-multiple", &quiet_if_locked, op_boolfalse, 1 },
+{'d',"daemon", &become_daemon, op_booltrue, 1 },
+{ 0, "no-daemon", &become_daemon, op_boolfalse, 1 },
+
+{ 0, "no-check-proportion", &nocheck_thresh_pct, op_double },
+{ 0, "no-check-filter", &nocheck_decay_articles, op_double },
+
+{ 0, "max-queue-size", &max_queue_per_conn op_integer },
+{ 0, "reconnect-interval", &reconnect_delay_periods, op_periods_rndup },
+{ 0, "flush-retry-interval", &flushfail_retry_periods, op_periods_rndup },
+{ 0, "feedfile-open-timeout", &open_wait_periods, op_periods_rndup },
+{ 0, "connection-timeout", &connection_timeout, op_seconds },
+{ 0, "inndcomm-timeout", &inndcomm_flush_timeout, op_seconds },
+};
+
+int main(int argc, char **argv) {
+ const char *arg;
+
+ for (;;) {
+ arg= *++argv;
+ if (!arg) break;
+ if (*arg != '-') break;
+ if (!strcmp(arg,"--")) { arg= *++argv; break; }
+ int a;
+ while ((a= *++arg)) {
+ const Option *o;
+ if (a=='-') {
+ arg++;
+ char *equals= strchr(arg,'=');
+ int len= equals ? (equals - arg) : strlen(arg);
+ for (o=options; o->long; o++)
+ if (strlen(o->long) == len && !memcmp(o->long,arg,len))
+ goto found_long;
+ badusage("unknown long option --%s",arg);
+ found_long:
+ if (o->noarg) {
+ if (equals) badusage("option --%s does not take a value",o->long);
+ arg= 0;
+ } else if (equals) {
+ arg= equals+1;
+ } else {
+ arg= *++argv;
+ if (!arg) badusage("option --%s needs a value",o->long);
+ }
+ o->fn(o, arg);
+ break; /* eaten the whole argument now */
+ }
+ for (o=options; o->long; o++)
+ if (a == o->short)
+ goto found_short;
+ badusage("unknown short option -%c",a);
+ found_short:
+ if (o->noarg) {
+ o->fn(o,0);
+ } else {
+ if (!*++arg) {
+ arg= *++argv;
+ if (!arg) badusage("option -%c needs a value",o->short);
+ }
+ o->fn(o,arg);
+ break; /* eaten the whole argument now */
+ }
+ }
+ }
+
+ if (!arg) badusage("need site name argument");
+ if (*++argv) badusage("too many non-option arguments");
+ sitename= arg;
+
+ if (nocheck_thresh_pct < 0 || nocheck_thresh_pct > 100)
+ badusage("nocheck threshold percentage must be between 0..100");
+ nocheck_thresh= nocheck_thresh_pct * 0.01;
+
+ if (nocheck_decay_articles < 0.1)
+ badusage("nocheck decay articles must be at least 0.1");
+ nocheck_decay= 1 - 1/nocheck_decay_articles;
+
+ innconf_read(inndconffile);
+
+ if (!feedfile)
+ feedfile= xasprintf("%s/%s",pathoutgoing,sitename);
+ else if (!feedfile[0])
+ badusage("feed filename must be nonempty");
+ else if (feedfile[strlen(feedfile)-1]=='/')
+ feedfile= xasprintf("%s%s",feedfile,sitename);
+
+ const char *feedfile_forbidden= "?*[";
+ int c;
+ while ((c= *feedfile_forbidden++))
+ if (strchr(feedfile, c))
+ badusage("feed filename may not contain glob metacharacter %c",c);
+
+ if (signal(SIGPIPE, SIG_IGN) == SIG_ERR)
+ sysdie("could not ignore SIGPIPE");
+
+ if (become_daemon) {
+ for (i=3; i<255; i++)
+ /* do this now before we open syslog, etc. */
+ close(i);
+ openlog("innduct",LOG_NDELAY|LOG_PID,LOG_NEWS);
+
+ int null= open("/dev/null",O_RDWR);
+ if (null<0) sysdie("failed to open /dev/null");
+ dup2(null,0);
+ dup2(null,1);
+ dup2(null,2);
+ close(null);
+
+ pid_t child1= xfork("daemonise first fork");
+ if (child1) _exit(0);
+
+ pid_t sid= setsid();
+ if (sid != child1) sysdie("setsid failed");
+
+ pid_t child2= xfork("daemonise second fork");
+ if (child2) _exit(0);
+ }
+
+ notice("starting");
+
+ if (!filemon_init()) {
+ warn("no file monitoring available, polling");
filepoll_schedule();
+ }
+
period_schedule();
-};
+
+ statemc_init();
+
+ loop->execute.
+}
~ian / inn-innduct.git / blobdiff