+/*
+ * TODO
+ * - close idle connections
+ * - cope better with garbage in feed file
+ * - cope better with NULs in feed file
+ * - -k kill mode ?
+ */
+
/*
* Newsfeeds file entries should look like this:
* host.name.of.site[/exclude,exclude,...]\
* 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
+
+#include "inn/list.h"
+#include "config.h"
+#include "storage.h"
+#include "nntp.h"
+#include "libinn.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 <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)
+
+/*----- 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);
/*----- configuration options -----*/
static char *sitename, *feedfile;
+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 const char *inndconffile;
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 InputFile {
- /* 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;
- struct Filemon_Perfile *filemon;
+ Filemon_Perfile *filemon;
oop_read *rd;
long inprogress; /* no. of articles read but not processed */
off_t offset;
+ int skippinglong;
- Counts counts;
+ int counts[art_MaxState][RCI_max];
+ int readcount_ok, readcount_blank, readcount_err;
char path[];
-} InputFile;
+};
-typedef enum {
- sm_WAITING,
- sm_NORMAL,
- sm_FLUSHING,
- sm_FLUSHFAIL,
- sm_SEPARATED,
- sm_DROPPING,
-} StateMachineState;
+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;
+ 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;
/*----- 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_lock, *path_flushing, *path_defer;
+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, *backlog_input_file;
+static InputFile *main_input_file, *flushing_input_file, *backlog_input_file;
static int sm_period_counter;
-/*----- function predeclarations -----*/
+/*========== logging ==========*/
-static void conn_check_work(Conn *conn);
+static void logcore(int sysloglevel, const char *fmt, ...)
+ __attribute__((__format__(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);
+ }
+ 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,
+ int exitstatus, const char *fmt, va_list al)
+ __attribute__((__format__(printf,5,0)));
+static void logv(int sysloglevel, const char *pfx, int errnoval,
+ int exitstatus, 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 logwrap(fn, pfx, sysloglevel, err, estatus) \
+ static void fn(const char *fmt, ...) \
+ __attribute__((__format__(printf,1,2))); \
+ static void fn(const char *fmt, ...) { \
+ va_list al; \
+ va_start(al,fmt); \
+ logv(sysloglevel, pfx, err, 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_WARNING, errno, 0);
+logwrap(warn, " warning", LOG_WARNING, -1, 0);
+
+logwrap(notice, "", LOG_NOTICE, -1, 0);
+logwrap(info, " info", LOG_INFO, -1, 0);
+logwrap(debug, " debug", LOG_DEBUG, -1, 0);
-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)
+ __attribute__((__format__(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, ...)
+ __attribute__((__format__(printf,1,2)));
+static char *xasprintf(const char *fmt, ...) {
+ va_list al;
+ va_start(al,fmt);
+ 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) {
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;
}
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);
return status;
}
+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 check_isreg(const struct stat *stab, const char *path,
const char *what) {
if (!S_ISREG(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);
+ int r= fstat(fd, stab_r);
+ if (r) sysdie("could not fstat %s", what);
}
-static void xfstat_isreg(int fd, struct stat *stab_r, const char *what) {
+static void xfstat_isreg(int fd, struct stat *stab_r,
+ const char *path, const char *what) {
xfstat(fd, stab_r, what);
- check_isreg(stab, path, what);
+ check_isreg(stab_r, path, what);
}
static void xlstat_isreg(const char *path, struct stat *stab,
- int *enoent_r /* 0 means ENOENT is fatal */
+ int *enoent_r /* 0 means ENOENT is fatal */,
const char *what) {
int r= lstat(path, stab);
if (r) {
check_isreg(stab, path, what);
}
+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");
+}
+
static int samefile(const struct stat *a, const struct stat *b) {
assert(S_ISREG(a->st_mode));
assert(S_ISREG(b->st_mode));
a->st_dev == b->st_dev);
}
-/*========== logging ==========*/
+static char *sanitise(const char *input) {
+ static char sanibuf[100]; /* returns pointer to this buffer! */
-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);
+ 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 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 */
+/*========== making new connections ==========*/
- logcore(sysloglevel, "<%s>%s: %s%s%s",
- sitename, pfx, msgbuf,
- errnoval>=0 ? ": " : "",
- errnoval>=0 ? strerror(errnoval) : "");
+static void conn_dispose(Conn *conn) {
+ if (!conn) return;
+ perhaps_close(&conn->fd);
+ free(conn);
+ until_connect= reconnect_delay_periods;
}
-#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 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) ||
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);
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);
/* 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");
}
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);
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");
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) ||
}
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;
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;
}
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();
/*========== overall control of article flow ==========*/
static void check_master_queue(void) {
- 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;
+
+ /* 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)) {
+ int 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) {
+ 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);
}
}
+static void vconnfail(Conn *conn, const char *fmt, va_list al)
+ __attribute__((__format__(printf,2,0)));
+
+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, ...)
+ __attribute__((__format__(printf,2,3)));
+static void connfail(Conn *conn, const char *fmt, ...) {
+ va_list al;
+ va_start(al,fmt);
+ vconnfail(conn,fmt,al);
+ va_end(al);
+}
/*========== article transmission ==========*/
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;
#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;
}
}
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) {
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);
}
}
}
- 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);
- art->ipf->counts[art->checked].sent++;
-
} else {
/* check it */
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);
- art->ipf->counts[art->checked].offered++;
}
}
}
return OOP_CONTINUE;
}
-static Article *article_reply_check(Connection *conn, const char *response,
+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,
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);
}
} 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++;
- art->ipf->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);
ipf->inprogress--;
assert(ipf->inprogress >= 0);
- if (!ipf->inprogress)
- loop->on_time(loop, OOP_TIME_NOW, statemc_check_input_done, ipf);
+ if (!ipf->inprogress && ipf != main_input_file)
+ queue_check_input_done();
free(art);
}
}
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\n");
+ 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;
}
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
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))
}
- check_check_work(conn);
+ conn_maybe_write(conn);
+ check_master_queue();
return OOP_CONTINUE;
}
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);
-
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) {
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;
strcpy(ipf->path, path);
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 flushing file %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;
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;
}
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) {
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);
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;
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;
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) { }
#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;
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);
}
/* See official state diagram at top of file. We implement
* this as follows:
-
- ================
- WAITING
- [Nothing/Noduct]
- poll for F
- ================
- |
- | TIMEOUT and no defer, no 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
- | | | 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 startup_set_input_file(InputFile *f) {
+ assert(!main_input_file);
+ main_input_file= f;
+ inputfile_tailing_start(f);
+}
+
static void statemc_init(void) {
struct stat stab, stabf;
- int noent;
path_lock= xasprintf("%s_lock", feedfile);
path_flushing= xasprintf("%s_flushing", feedfile);
memset(&fl,0,sizeof(fl));
fl.l_type= F_WRLCK;
fl.l_whence= SEEK_SET;
- r= fcntl(lockfd, F_SETLK, &fl);
+ int r= fcntl(lockfd, F_SETLK, &fl);
if (r==-1) {
if (errno==EACCES || errno==EAGAIN) {
- if (quiet_if_locked) exit(0);
+ if (quiet_multiple) exit(0);
fatal("another duct holds the lockfile");
}
sysdie("fcntl F_SETLK lockfile %s", path_lock);
}
- xfstat_isreg(lockfd, &stabf, "lockfile");
- xlstat_isreg(path_lock, &stab, &noent, "lockfile");
+ xfstat_isreg(lockfd, &stabf, path_lock, "lockfile");
+ int lock_noent;
+ xlstat_isreg(path_lock, &stab, &lock_noent, "lockfile");
- if (!noent && samefile(&stab, &stabf))
+ if (!lock_noent && samefile(&stab, &stabf))
break;
if (close(lockfd))
search_backlog_file();
- xlstat_isreg(path_defer, &stab, &noent, "defer file");
- if (noent) {
+ int defer_noent;
+ xlstat_isreg(path_defer, &stab, &defer_noent, "defer file");
+ if (defer_noent) {
debug("startup: ductdefer ENOENT");
} else {
debug("startup: ductdefer nlink=%ld", (long)stab.st_nlink);
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",
}
}
- InputFile *file_d= open_input_file(path_flushing);
-
- if (file_d) {
- struct stat stab_f, stab_d;
+ struct stat stab_f, stab_d;
+ int noent_f;
- xlstat_isreg(feedfile, &stab_f, &noent, "feed file");
- if (noent) {
- debug("startup: D exists, F ENOENT => Moved");
- goto found_moved;
- }
+ InputFile *file_d= open_input_file(path_flushing);
+ if (file_d) xfstat_isreg(file_d->fd, &stab_d, path_flushing,"flushing file");
- debug("startup: F and D both exist");
+ xlstat_isreg(feedfile, &stab_f, &noent_f, "feedfile");
- xfstat_isreg(file_d->fd, &stab_d, "flushing file");
+ 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 (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:
- debug(" => 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 {
- debug("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 {
- 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(); 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_input_done(oop_source *lp,
- struct timeval now, void *ipf_v) {
- InputFile *ipf= ipf_v;
- struct stat stab;
-
- 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;
- }
+static void statemc_check_flushing_done(void) {
+ InputFile *ipf= flushing_input_file;
+ if (!inputfile_is_done(ipf)) return;
- assert(ipf == old_input_file);
assert(sms==sm_SEPARATED || sms==sm_DROPPING);
- notice_processed(ipf,"feed file",0);
+ notice_processed(ipf,"feedfile",0);
close_defer();
- if (unlink(path_flushing))
- sysdie("could not unlink old flushing file %s", path_flushing);
+ xunlink(path_flushing, "old flushing file");
- 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);
- }
+ close_input_file(flushing_input_file);
+ free(flushing_input_file);
+ flushing_input_file= 0;
- open_defer();
+ 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");
+ }
+}
- close_input_file(old_input_file);
- old_input_file= 0;
+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;
+}
- notice("flush complete");
- SMS(NORMAL, 0, "flush complete");
+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[%d] %s",periods,forlog,why);
+ info("%s%s[%d] %s",forlog,xtra,periods,why);
} else {
- info("%s %s",forlog,why);
+ info("%s%s %s",forlog,xtra,why);
}
}
/* truncate away any half-written records */
- xfstat_isreg(fileno(defer), &stab, "newly opened defer file");
+ 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 (fseek(defer, truncto-1, SEEK_SET) < 0)
sysdie("seek in defer file %s while truncating partial", path_defer);
- r= getc(defer);
+ int r= getc(defer);
if (r==EOF) {
if (ferror(defer))
sysdie("failed read from defer file %s", path_defer);
if (!defer)
return;
- xfstat(fileno(defer), &stab, "defer file");
+ xfstat(fileno(defer), &stab, path_defer, "defer file");
if (fclose(defer)) sysfatal("could not close defer file %s", path_defer);
defer= 0;
+ time_t now= time(0);
+ if (now==-1) sysdie("time(2) failed");
+
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 (backlog_nextscan_periods < 0 ||
- backlog_nextscan_periods > backlog_retry_minperiods + 1)
- backlog_nextscan_periods= backlog_retry_minperiods + 1;
+ 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 (backlog_nextscan_periods < 0) return;
- if (backlog_nextscan_periods-- > 0) return;
+ if (until_backlog_nextscan < 0) return;
+ if (until_backlog_nextscan-- > 0) return;
search_backlog_file();
}
-static int search_backlog_file(void) {
+static void search_backlog_file(void) {
/* returns non-0 iff there are any backlog files */
glob_t gl;
if (!oldest_path) {
debug("backlog scan: none");
- backlog_nextscan_periods= backlog_spontaneous_rescan_periods;
+
+ 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 0;
}
goto try_again;
}
inputfile_tailing_start(backlog_input_file);
- backlog_nextscan_periods= -1;
+ until_backlog_nextscan= -1;
return 1;
}
- backlog_nextscan_periods= age_deficiency / PERIOD_SECONDS;
+ until_backlog_nextscan= age_deficiency / PERIOD_SECONDS;
if (backlog_spontaneous_rescan_periods >= 0 &&
- backlog_nextscan_periods > backlog_spontaneous_rescan_periods)
- backlog_nextscan_periods= backlog_spontaneous_rescan_periods;
+ 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, backlog_nextscan_periods, oldest_path);
+ age, age_deficiency, until_backlog_nextscan, oldest_path);
return 2;
}
loop->cancel_fd(fd);
close(fd);
- assert(!old_input_file);
+ assert(!flushing_input_file);
if (WIFEXITED(status)) {
switch (WEXITSTATUS(status)) {
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:
}
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) {
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");
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 ==========*/
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);
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);
}
filemon_callback(main_input_file);
});
+#define DEBUGF_IPF(wh) " " #wh "=%p/%s:ip=%ld,off=%ld,fd=%d%s" \
+#define DEBUG_IPF(sh) \
+ 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}, {
- if (connect_delay) connect_delay--;
+ debug("PERIOD"
+ " sms=%s[%d] conns=%d queue=%d until_connect=%d"
+ " input_files" DEBUGF_IPF(main) DEBUGF_IPF(old) DEBUGF_FMT(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_poll();
+ statemc_period_poll();
check_master_queue();
});
}
static const Option options[]= {
+{'f',"feedfile", &feedfile, op_string },
+{'q',"quiet-multiple", &quiet_multiple, op_booltrue, 1 },
+
{ 0, "max-connections", &max_connections op_integer },
+{ 0, "max-queue-per-conn", &max_queue_per_conn 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 },
};
}
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;
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)
else if (feedfile[strlen(feedfile)-1]=='/')
feedfile= xasprintf("%s%s",feedfile,sitename);
- const char *feedfile_forbidden= "?*[";
+ const char *feedfile_forbidden= "?*[~#";
int c;
while ((c= *feedfile_forbidden++))
if (strchr(feedfile, c))
- badusage("feed filename may not contain glob metacharacter %c",c);
+ badusage("feed filename may not contain metacharacter %c",c);
+
+ loop= oop_sys_new();
+ if (!loop) sysdie("could not create liboop event loop");
if (signal(SIGPIPE, SIG_IGN) == SIG_ERR)
sysdie("could not ignore SIGPIPE");
~ian / inn-innduct.git / blobdiff