X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ian/git?p=inn-innduct.git;a=blobdiff_plain;f=backends%2Finnduct.c;h=8104fee469826eb8e83f0b0480e3ecdfd35ab660;hp=8e599f56026835595780ad3a28725754699fd06f;hb=d69d12a0e443c2f034a31720e74b40340bc327ff;hpb=fc66b511debe93980fd49331d96939fb876024da diff --git a/backends/innduct.c b/backends/innduct.c index 8e599f5..8104fee 100644 --- a/backends/innduct.c +++ b/backends/innduct.c @@ -1,10 +1,33 @@ /* - * TODO - * - actually implement badusage - * - options for all options - * - manpage - * - pid, sitename, hostname in lockfile - * - -k kill mode ? + * innduct + * tailing reliable realtime streaming feeder for inn + * + * Copyright (C) 2010 Ian Jackson + * + * This program is free software: you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, either version 3 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see . + * + * (I believe that when you compile and link this as part of the inn2 + * build, with the Makefile runes I have provided, all the libraries + * and files which end up included in innduct are licence-compatible + * with GPLv3. If not then please let me know. -Ian Jackson.) + */ + +/* + * todo + * + * don't mind reconnecting if we just disconnected due to idle + * some weird disconnection event still investigating */ /* @@ -172,6 +195,7 @@ perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct. #include #include #include +#include #include #include #include @@ -182,26 +206,38 @@ perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct. #include #include #include +#include #include #include #include +#include #include #include /*----- general definitions, probably best not changed -----*/ -#define CONNCHILD_ESTATUS_STREAM 4 -#define CONNCHILD_ESTATUS_NOSTREAM 5 +#define CONNCHILD_ESTATUS_STREAM 24 +#define CONNCHILD_ESTATUS_NOSTREAM 25 -#define INNDCOMMCHILD_ESTATUS_FAIL 6 -#define INNDCOMMCHILD_ESTATUS_NONESUCH 7 +#define INNDCOMMCHILD_ESTATUS_FAIL 26 +#define INNDCOMMCHILD_ESTATUS_NONESUCH 27 #define MAX_LINE_FEEDFILE (NNTP_MSGID_MAXLEN + sizeof(TOKEN)*2 + 10) +#define MAX_CLI_COMMAND 1000 #define VA va_list al; va_start(al,fmt) #define PRINTF(f,a) __attribute__((__format__(printf,f,a))) #define NORET_PRINTF(f,a) __attribute__((__noreturn__,__format__(printf,f,a))) +#define NORET __attribute__((__noreturn__)) + +#define NEW(ptr) ((ptr)= zxmalloc(sizeof(*(ptr)))) +#define NEW_DECL(type,ptr) type ptr = zxmalloc(sizeof(*(ptr))) + +#define DUMPV(fmt,pfx,v) fprintf(f, " " #v "=" fmt, pfx v); + +#define FOR_CONN(conn) \ + for ((conn)=LIST_HEAD(conns); (conn); (conn)=LIST_NEXT((conn))) /*----- doubly linked lists -----*/ @@ -212,7 +248,7 @@ perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct. int count; \ } T##List -#define NODE(n) (assert((void*)&(n)->list_node == &(n)), &(n)->list_node) +#define NODE(n) (assert((void*)&(n)->list_node == (n)), &(n)->list_node) #define LIST_CHECKCANHAVENODE(l,n) \ ((void)((n) == ((l).u.for_type))) /* just for the type check */ @@ -238,7 +274,7 @@ perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct. #define LIST_REMHEAD(l) LIST_REMSOMEHOW((l),list_remhead) #define LIST_REMTAIL(l) LIST_REMSOMEHOW((l),list_remtail) -#define LIST_INIT(l) (list_new(&(l).u.li)) +#define LIST_INIT(l) ((l).count=0, list_new(&(l).u.li)) #define LIST_HEAD(l) ((typeof((l).u.for_type))(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)))) @@ -264,6 +300,7 @@ typedef struct InputFile InputFile; typedef struct XmitDetails XmitDetails; typedef struct Filemon_Perfile Filemon_Perfile; typedef enum StateMachineState StateMachineState; +typedef struct CliCommand CliCommand; DEFLIST(Conn); DEFLIST(Article); @@ -283,25 +320,37 @@ static void statemc_setstate(StateMachineState newsms, int periods, static void statemc_start_flush(const char *why); /* Normal => Flushing */ static void spawn_inndcomm_flush(const char *why); /* Moved => Flushing */ +static int trigger_flush_ok(const char *why /* 0 means timeout */); + /* => Flushing,FLUSHING, ret 1; or ret 0 */ + +static void article_done(Article *art, int whichcount); static void check_assign_articles(void); static void queue_check_input_done(void); +static void check_reading_pause_resume(InputFile *ipf); static void statemc_check_flushing_done(void); static void statemc_check_backlog_done(void); static void postfork(void); +static void period(void); static void open_defer(void); static void close_defer(void); static void search_backlog_file(void); +static void preterminate(void); +static void raise_default(int signo) NORET; +static char *debug_report_ipf(InputFile *ipf); static void inputfile_reading_start(InputFile *ipf); static void inputfile_reading_stop(InputFile *ipf); +static void inputfile_reading_pause(InputFile *ipf); +static void inputfile_reading_resume(InputFile *ipf); + /* pause and resume are idempotent, and no-op if not done _reading_start */ static void filemon_start(InputFile *ipf); static void filemon_stop(InputFile *ipf); -static void filemon_callback(InputFile *ipf); +static void tailing_make_readable(InputFile *ipf); static void vconnfail(Conn *conn, const char *fmt, va_list al) PRINTF(2,0); static void connfail(Conn *conn, const char *fmt, ...) PRINTF(2,3); @@ -309,12 +358,14 @@ static void connfail(Conn *conn, const char *fmt, ...) PRINTF(2,3); static const oop_rd_style peer_rd_style; static oop_rd_call peer_rd_err, peer_rd_ok; + /*----- configuration options -----*/ +/* when changing defaults, remember to update the manpage */ static const char *sitename, *remote_host; -static const char *feedfile; +static const char *feedfile, *path_run, *path_cli, *path_cli_dir; static int quiet_multiple=0; -static int become_daemon=1; +static int interactive=0, try_filemon=1; static int try_stream=1; static int port=119; static const char *inndconffile; @@ -322,7 +373,9 @@ static const char *inndconffile; static int max_connections=10; static int max_queue_per_conn=200; static int target_max_feedfile_size=100000; -static int period_seconds=60; +static int period_seconds=30; +static int filepoll_seconds=5; +static int max_queue_per_ipf=-1; static int connection_setup_timeout=200; static int inndcomm_flush_timeout=100; @@ -333,10 +386,13 @@ static double nocheck_decay= 100; /* conv'd from articles to lambda by main */ /* all these are initialised to seconds, and converted to periods in main */ static int reconnect_delay_periods=1000; static int flushfail_retry_periods=1000; -static int backlog_retry_minperiods=50; +static int backlog_retry_minperiods=100; static int backlog_spontrescan_periods=300; static int spontaneous_flush_periods=100000; +static int max_separated_periods=2000; static int need_activity_periods=1000; +static int lowvol_thresh=3; +static int lowvol_periods=1000; static double max_bad_data_ratio= 1; /* conv'd from percentage by main */ static int max_bad_data_initial= 30; @@ -350,18 +406,22 @@ 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 + art_MaxState, } ArtState; +static const char *const artstate_names[]= + { "Unchecked", "Wanted", "Unsolicited", 0 }; + #define RESULT_COUNTS(RCS,RCN) \ RCS(sent) \ RCS(accepted) \ RCN(unwanted) \ RCN(rejected) \ RCN(deferred) \ + RCN(missing) \ RCN(connretry) -#define RCI_TRIPLE_FMT_BASE "%d(id%d+bd%d+nc%d)" +#define RCI_TRIPLE_FMT_BASE "%d (id=%d,bod=%d,nc=%d)" #define RCI_TRIPLE_VALS_BASE(counts,x) \ counts[art_Unchecked] x \ + counts[art_Wanted] x \ @@ -382,13 +442,12 @@ typedef enum { #define CONNIOVS 128 typedef enum { - xk_Malloc, xk_Const, xk_Artdata + xk_Const, xk_Artdata } XmitKind; struct XmitDetails { XmitKind kind; union { - char *malloc_tofree; ARTHANDLE *sm_art; } info; }; @@ -406,19 +465,22 @@ struct InputFile { Filemon_Perfile *filemon; oop_read *rd; /* non-0: reading; 0: constructing, or had EOF */ - long inprogress; /* no. of articles read but not processed */ off_t offset; - int skippinglong; + int skippinglong, paused, fake_readable; + + ArticleList queue; + long inprogress; /* includes queue.count and also articles in conns */ + long autodefer; /* -1 means not doing autodefer */ int counts[art_MaxState][RCI_max]; - int readcount_ok, readcount_blank, readcount_err; + int readcount_ok, readcount_blank, readcount_err, count_nooffer_missing; char path[]; }; struct Article { ISNODE(Article); ArtState state; - int midlen; + int midlen, missing; InputFile *ipf; TOKEN token; off_t offset; @@ -449,7 +511,8 @@ struct Conn { ISNODE(Conn); int fd; /* may be 0, meaning closed (during construction/destruction) */ oop_read *rd; /* likewise */ - int max_queue, stream, quitting; + int max_queue, stream; + const char *quitting; int since_activity; /* periods */ ArticleList waiting; /* not yet told peer */ ArticleList priority; /* peer says send it now */ @@ -465,28 +528,36 @@ struct Conn { /* main initialises */ static oop_source *loop; static ConnList conns; -static ArticleList queue; -static char *path_lock, *path_flushing, *path_defer, *globpat_backlog; +static char *path_lock, *path_flushing, *path_defer, *path_dump; +static char *globpat_backlog; +static pid_t self_pid; +static int *lowvol_perperiod; +static int lowvol_circptr; +static int lowvol_total; /* does not include current period */ /* statemc_init initialises */ static StateMachineState sms; -static FILE *defer; +static int until_flush; static InputFile *main_input_file, *flushing_input_file, *backlog_input_file; -static int sm_period_counter; +static FILE *defer; /* initialisation to 0 is good */ static int until_connect, until_backlog_nextscan; static double accept_proportion; -static int nocheck, nocheck_reported; +static int nocheck, nocheck_reported, in_child; + +/* for simulation, debugging, etc. */ +int simulate_flush= -1; /*========== logging ==========*/ static void logcore(int sysloglevel, const char *fmt, ...) PRINTF(2,3); static void logcore(int sysloglevel, const char *fmt, ...) { VA; - if (become_daemon) { + if (interactive < 2) { vsyslog(sysloglevel,fmt,al); } else { + if (self_pid) fprintf(stderr,"[%lu] ",(unsigned long)self_pid); vfprintf(stderr,fmt,al); putc('\n',stderr); } @@ -497,7 +568,7 @@ static void logv(int sysloglevel, const char *pfx, int errnoval, const char *fmt, va_list al) PRINTF(5,0); static void logv(int sysloglevel, const char *pfx, int errnoval, const char *fmt, va_list al) { - char msgbuf[256]; /* NB do not call xvasprintf here or you'll recurse */ + char msgbuf[1024]; /* NB do not call xvasprintf here or you'll recurse */ vsnprintf(msgbuf,sizeof(msgbuf), fmt,al); msgbuf[sizeof(msgbuf)-1]= 0; @@ -513,6 +584,7 @@ static void logv(int sysloglevel, const char *pfx, int errnoval, #define diewrap(fn, pfx, sysloglevel, err, estatus) \ static void fn(const char *fmt, ...) NORET_PRINTF(1,2); \ static void fn(const char *fmt, ...) { \ + preterminate(); \ VA; \ logv(sysloglevel, pfx, err, fmt, al); \ exit(estatus); \ @@ -535,7 +607,7 @@ diewrap(fatal, " fatal", LOG_ERR, -1, 12); logwrap(syswarn, " warning", LOG_WARNING, errno); logwrap(warn, " warning", LOG_WARNING, -1); -logwrap(notice, "", LOG_NOTICE, -1); +logwrap(notice, " notice", LOG_NOTICE, -1); logwrap(info, " info", LOG_INFO, -1); logwrap(debug, " debug", LOG_DEBUG, -1); @@ -558,7 +630,7 @@ static char *xasprintf(const char *fmt, ...) { } static int close_perhaps(int *fd) { - if (!*fd) return 0; + if (*fd <= 0) return 0; int r= close(*fd); *fd=0; return r; @@ -568,7 +640,7 @@ static void xclose(int fd, const char *what, const char *what2) { if (r) sysdie("close %s%s",what,what2?what2:""); } static void xclose_perhaps(int *fd, const char *what, const char *what2) { - if (!*fd) return; + if (*fd <= 0) return; xclose(*fd,what,what2); *fd=0; } @@ -577,7 +649,7 @@ static pid_t xfork(const char *what) { pid_t child; child= fork(); - if (child==-1) sysdie("cannot fork for %s",what); + if (child==-1) sysfatal("cannot fork for %s",what); debug("forked %s %ld", what, (unsigned long)child); if (!child) postfork(); return child; @@ -617,14 +689,21 @@ static int xwaitpid(pid_t *pid, const char *what) { int r= kill(*pid, SIGKILL); if (r) sysdie("cannot kill %s child", what); - pid_t got= waitpid(*pid, &status, WNOHANG); + pid_t got= waitpid(*pid, &status, 0); if (got==-1) sysdie("cannot reap %s child", what); + if (got==0) die("cannot reap %s child", what); *pid= 0; return status; } +static void *zxmalloc(size_t sz) { + void *p= xmalloc(sz); + memset(p,0,sz); + return p; +} + static void xunlink(const char *path, const char *what) { int r= unlink(path); if (r) sysdie("can't unlink %s %s", path, what); @@ -636,6 +715,28 @@ static time_t xtime(void) { return now; } +static void xsigaction(int signo, const struct sigaction *sa) { + int r= sigaction(signo,sa,0); + if (r) sysdie("sigaction failed for \"%s\"", strsignal(signo)); +} + +static void xsigsetdefault(int signo) { + struct sigaction sa; + memset(&sa,0,sizeof(sa)); + sa.sa_handler= SIG_DFL; + xsigaction(signo,&sa); +} + +static void xgettimeofday(struct timeval *tv_r) { + int r= gettimeofday(tv_r,0); + if (r) sysdie("gettimeofday(2) failed"); +} + +static void xsetnonblock(int fd, int nonblocking) { + int errnoval= oop_fd_nonblock(fd, nonblocking); + if (errnoval) { errno= errnoval; sysdie("setnonblocking"); } +} + static void check_isreg(const struct stat *stab, const char *path, const char *what) { if (!S_ISREG(stab->st_mode)) @@ -673,15 +774,16 @@ static int samefile(const struct stat *a, const struct stat *b) { a->st_dev == b->st_dev); } -static char *sanitise(const char *input) { +static char *sanitise(const char *input, int len) { static char sanibuf[100]; /* returns pointer to this buffer! */ const char *p= input; + const char *endp= len>=0 ? input+len : 0; char *q= sanibuf; *q++= '`'; for (;;) { if (q > sanibuf+sizeof(sanibuf)-8) { strcpy(q,"'.."); break; } - int c= *p++; + int c= (!endp || p=' ' && c<=126 && c!='\\') { *q++= c; continue; } sprintf(q,"\\x%02x",c); @@ -694,14 +796,298 @@ static int isewouldblock(int errnoval) { return errnoval==EWOULDBLOCK || errnoval==EAGAIN; } +/*========== command and control (CLI) connections ==========*/ + +static int cli_master; + +typedef struct CliConn CliConn; +struct CliConn { + void (*destroy)(CliConn*); + int fd; + oop_read *rd; + FILE *out; + union { + struct sockaddr sa; + struct sockaddr_un un; + } sa; + socklen_t salen; +}; + +static const oop_rd_style cli_rd_style= { + OOP_RD_DELIM_STRIP, '\n', + OOP_RD_NUL_FORBID, + OOP_RD_SHORTREC_FORBID +}; + +static void cli_destroy(CliConn *cc) { + cc->destroy(cc); +} + +static void cli_checkouterr(CliConn *cc /* may destroy*/) { + if (ferror(cc->out) | fflush(cc->out)) { + info("CTRL%d write error %s", cc->fd, strerror(errno)); + cli_destroy(cc); + } +} + +static void cli_prompt(CliConn *cc /* may destroy*/) { + fprintf(cc->out, "%s| ", sitename); + cli_checkouterr(cc); +} + +struct CliCommand { + const char *cmd; + void (*f)(CliConn *cc, const CliCommand *ccmd, + const char *arg, size_t argsz); + void *xdata; + int xval; +}; + +static const CliCommand cli_commands[]; + +#define CCMD(wh) \ + static void ccmd_##wh(CliConn *cc, const CliCommand *c, \ + const char *arg, size_t argsz) + +CCMD(help) { + fputs("commands:\n", cc->out); + const CliCommand *ccmd; + for (ccmd=cli_commands; ccmd->cmd; ccmd++) + fprintf(cc->out, " %s\n", ccmd->cmd); + fputs("NB: permissible arguments are not shown above." + " Not all commands listed are safe. See innduct(8).\n", cc->out); +} + +CCMD(flush) { + int ok= trigger_flush_ok("manual request"); + if (!ok) fprintf(cc->out,"already flushing (state is %s)\n", sms_names[sms]); +} + +CCMD(stop) { + preterminate(); + notice("terminating (CTRL%d)",cc->fd); + raise_default(SIGTERM); + abort(); +} + +CCMD(dump); + +/* messing with our head: */ +CCMD(period) { period(); } +CCMD(setintarg) { *(int*)c->xdata= atoi(arg); } +CCMD(setint) { *(int*)c->xdata= c->xval; } +CCMD(setint_period) { *(int*)c->xdata= c->xval; period(); } + +static const CliCommand cli_commands[]= { + { "h", ccmd_help }, + { "flush", ccmd_flush }, + { "stop", ccmd_stop }, + { "dump q", ccmd_dump, 0,0 }, + { "dump a", ccmd_dump, 0,1 }, + + { "p", ccmd_period }, + +#define POKES(cmd,func) \ + { cmd "flush", func, &until_flush, 1 }, \ + { cmd "conn", func, &until_connect, 0 }, \ + { cmd "blscan", func, &until_backlog_nextscan, 0 }, +POKES("next ", ccmd_setint) +POKES("prod ", ccmd_setint_period) + + { "pretend flush", ccmd_setintarg, &simulate_flush }, + { "wedge blscan", ccmd_setint, &until_backlog_nextscan, -1 }, + { 0 } +}; + +static void *cli_rd_ok(oop_source *lp, oop_read *oread, oop_rd_event ev, + const char *errmsg, int errnoval, + const char *data, size_t recsz, void *cc_v) { + CliConn *cc= cc_v; + + if (!data) { + info("CTRL%d closed", cc->fd); + cc->destroy(cc); + return OOP_CONTINUE; + } + + if (recsz == 0) goto prompt; + + const CliCommand *ccmd; + for (ccmd=cli_commands; ccmd->cmd; ccmd++) { + int l= strlen(ccmd->cmd); + if (recsz < l) continue; + if (recsz > l && data[l] != ' ') continue; + if (memcmp(data, ccmd->cmd, l)) continue; + + int argl= (int)recsz - (l+1); + ccmd->f(cc, ccmd, argl>=0 ? data+l+1 : 0, argl); + goto prompt; + } + + fputs("unknown command; h for help\n", cc->out); + + prompt: + cli_prompt(cc); + return OOP_CONTINUE; +} + +static void *cli_rd_err(oop_source *lp, oop_read *oread, oop_rd_event ev, + const char *errmsg, int errnoval, + const char *data, size_t recsz, void *cc_v) { + CliConn *cc= cc_v; + + info("CTRL%d read error %s", cc->fd, errmsg); + cc->destroy(cc); + return OOP_CONTINUE; +} + +static int cli_conn_startup(CliConn *cc /* may destroy*/, + const char *how) { + cc->rd= oop_rd_new_fd(loop, cc->fd, 0,0); + if (!cc->rd) { warn("oop_rd_new_fd cli failed"); return -1; } + + int er= oop_rd_read(cc->rd, &cli_rd_style, MAX_CLI_COMMAND, + cli_rd_ok, cc, + cli_rd_err, cc); + if (er) { errno= er; syswarn("oop_rd_read cli failed"); return -1; } + + info("CTRL%d %s ready", cc->fd, how); + cli_prompt(cc); + return 0; +} + +static void cli_stdio_destroy(CliConn *cc) { + if (cc->rd) { + oop_rd_cancel(cc->rd); + errno= oop_rd_delete_tidy(cc->rd); + if (errno) syswarn("oop_rd_delete tidy failed (no-nonblock stdin?)"); + } + free(cc); +} + +static void cli_stdio(void) { + NEW_DECL(CliConn *,cc); + cc->destroy= cli_stdio_destroy; + + cc->fd= 0; + cc->out= stdout; + int r= cli_conn_startup(cc,"stdio"); + if (r) cc->destroy(cc); +} + +static void cli_accepted_destroy(CliConn *cc) { + if (cc->rd) { + oop_rd_cancel(cc->rd); + oop_rd_delete_kill(cc->rd); + } + if (cc->out) { fclose(cc->out); cc->fd=0; } + close_perhaps(&cc->fd); + free(cc); +} + +static void *cli_master_readable(oop_source *lp, int master, + oop_event ev, void *u) { + NEW_DECL(CliConn *,cc); + cc->destroy= cli_accepted_destroy; + + cc->salen= sizeof(cc->sa); + cc->fd= accept(master, &cc->sa.sa, &cc->salen); + if (cc->fd<0) { syswarn("error accepting cli connection"); goto x; } + + cc->out= fdopen(cc->fd, "w"); + if (!cc->out) { syswarn("error fdopening accepted cli connection"); goto x; } + + int r= cli_conn_startup(cc, "accepted"); + if (r) goto x; + + return OOP_CONTINUE; + + x: + cc->destroy(cc); + return OOP_CONTINUE; +} + +#define NOCLI(...) do{ \ + syswarn("no cli listener, because failed to " __VA_ARGS__); \ + goto nocli; \ + }while(0) + +static void cli_init(void) { + union { + struct sockaddr sa; + struct sockaddr_un un; + } sa; + + memset(&sa,0,sizeof(sa)); + int maxlen= sizeof(sa.un.sun_path); + + if (!path_cli) { + info("control command line disabled"); + return; + } + + int pathlen= strlen(path_cli); + if (pathlen > maxlen) { + warn("no cli listener, because cli socket path %s too long (%d>%d)", + path_cli, pathlen, maxlen); + return; + } + + if (path_cli_dir) { + int r= mkdir(path_cli_dir, 0700); + if (r && errno!=EEXIST) + NOCLI("create cli socket directory %s", path_cli_dir); + } + + int r= unlink(path_cli); + if (r && errno!=ENOENT) + NOCLI("remove old cli socket %s", path_cli); + + cli_master= socket(PF_UNIX, SOCK_STREAM, 0); + if (cli_master<0) NOCLI("create new cli master socket"); + + int sl= pathlen + offsetof(struct sockaddr_un, sun_path); + sa.un.sun_family= AF_UNIX; + memcpy(sa.un.sun_path, path_cli, pathlen); + + r= bind(cli_master, &sa.sa, sl); + if (r) NOCLI("bind to cli socket path %s", sa.un.sun_path); + + r= listen(cli_master, 5); + if (r) NOCLI("listen to cli master socket"); + + xsetnonblock(cli_master, 1); + + loop->on_fd(loop, cli_master, OOP_READ, cli_master_readable, 0); + info("cli ready, listening on %s", path_cli); + + return; + + nocli: + xclose_perhaps(&cli_master, "cli master",0); + return; +} + /*========== management of connections ==========*/ +static void reconnect_blocking_event(void) { + until_connect= reconnect_delay_periods; +} + static void conn_closefd(Conn *conn, const char *msgprefix) { int r= close_perhaps(&conn->fd); if (r) info("C%d %serror closing socket: %s", conn->fd, msgprefix, strerror(errno)); } +static int conn_busy(Conn *conn) { + return + conn->waiting.count || + conn->priority.count || + conn->sent.count || + conn->xmitu; +} + static void conn_dispose(Conn *conn) { if (!conn) return; if (conn->rd) { @@ -715,7 +1101,6 @@ static void conn_dispose(Conn *conn) { } conn_closefd(conn,""); free(conn); - until_connect= reconnect_delay_periods; } static void *conn_exception(oop_source *lp, int fd, @@ -733,14 +1118,21 @@ static void *conn_exception(oop_source *lp, int fd, static void vconnfail(Conn *conn, const char *fmt, va_list al) { int requeue[art_MaxState]; + memset(requeue,0,sizeof(requeue)); 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->priority))) + LIST_ADDTAIL(art->ipf->queue, art); + + while ((art= LIST_REMHEAD(conn->waiting))) + LIST_ADDTAIL(art->ipf->queue, art); + while ((art= LIST_REMHEAD(conn->sent))) { requeue[art->state]++; if (art->state==art_Unsolicited) art->state= art_Unchecked; - LIST_ADDTAIL(queue,art); + LIST_ADDTAIL(art->ipf->queue,art); + check_reading_pause_resume(art->ipf); } int i; @@ -748,12 +1140,14 @@ static void vconnfail(Conn *conn, const char *fmt, va_list al) { for (i=0, d=conn->xmitd; ixmitu; i++, d++) xmit_free(d); + LIST_REMOVE(conns,conn); + 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); + warn("C%d (now %d) connection failed requeueing " RCI_TRIPLE_FMT_BASE ": %s", + conn->fd, conns.count, RCI_TRIPLE_VALS_BASE(requeue, /*nothing*/), m); free(m); - LIST_REMOVE(conns,conn); + reconnect_blocking_event(); conn_dispose(conn); check_assign_articles(); } @@ -765,17 +1159,56 @@ static void connfail(Conn *conn, const char *fmt, ...) { va_end(al); } +static void conn_idle_close(Conn *conn, const char *why) { + static const char quitcmd[]= "QUIT\r\n"; + int todo= sizeof(quitcmd)-1; + const char *p= quitcmd; + for (;;) { + int r= write(conn->fd, p, todo); + if (r<0) { + if (isewouldblock(errno)) + connfail(conn, "blocked writing QUIT to idle connection"); + else + connfail(conn, "failed to write QUIT to idle connection: %s", + strerror(errno)); + break; + } + assert(r<=todo); + todo -= r; + if (!todo) { + conn->quitting= why; + conn->since_activity= 0; + debug("C%d is idle (%s), quitting", conn->fd, why); + break; + } + } +} + +/* + * For our last connection, we also shut it down if we have had + * less than K in the last L + */ static void check_idle_conns(void) { Conn *conn; - for (conn=LIST_HEAD(conns); conn; conn=LIST_NEXT(conn)) + + int volthisperiod= lowvol_perperiod[lowvol_circptr]; + lowvol_circptr++; + lowvol_circptr %= lowvol_periods; + lowvol_total += volthisperiod; + lowvol_total -= lowvol_perperiod[lowvol_circptr]; + lowvol_perperiod[lowvol_circptr]= 0; + + FOR_CONN(conn) conn->since_activity++; + search_again: - for (conn=LIST_HEAD(conns); conn; conn=LIST_NEXT(conn)) { + FOR_CONN(conn) { if (conn->since_activity <= need_activity_periods) continue; /* We need to shut this down */ if (conn->quitting) - connfail(conn,"timed out waiting for response to QUIT"); + connfail(conn,"timed out waiting for response to QUIT (%s)", + conn->quitting); else if (conn->sent.count) connfail(conn,"timed out waiting for responses"); else if (conn->waiting.count || conn->priority.count) @@ -783,32 +1216,18 @@ static void check_idle_conns(void) { else if (conn->xmitu) connfail(conn,"peer has been sending responses" " before receiving our commands!"); - else { - static const char quitcmd[]= "QUIT\r\n"; - int todo= sizeof(quitcmd)-1; - const char *p= quitcmd; - for (;;) { - int r= write(conn->fd, p, todo); - if (r<0) { - if (isewouldblock(errno)) - connfail(conn, "blocked writing QUIT to idle connection"); - else - connfail(conn, "failed to write QUIT to idle connection: %s", - strerror(errno)); - break; - } - assert(r<=todo); - todo -= r; - if (!todo) { - conn->quitting= 1; - conn->since_activity= 0; - debug("C%d is idle, quitting", conn->fd); - break; - } - } - } + else + conn_idle_close(conn, "no activity"); + goto search_again; } + + conn= LIST_HEAD(conns); + if (!volthisperiod && + conns.count==1 && + lowvol_total < lowvol_thresh && + !conn_busy(conn)) + conn_idle_close(conn, "low volume"); } /*---------- making new connections ----------*/ @@ -818,10 +1237,7 @@ static int connecting_fdpass_sock; static void connect_attempt_discard(void) { 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) || (WIFSIGNALED(status) && WTERMSIG(status) == SIGKILL))) report_child_status("connect", status); @@ -833,9 +1249,15 @@ static void connect_attempt_discard(void) { } #define PREP_DECL_MSG_CMSG(msg) \ + char msgbyte= 0; \ + struct iovec msgiov; \ + msgiov.iov_base= &msgbyte; \ + msgiov.iov_len= 1; \ struct msghdr msg; \ memset(&msg,0,sizeof(msg)); \ - char msg##cbuf[CMSG_SPACE(sizeof(fd))]; \ + char msg##cbuf[CMSG_SPACE(sizeof(int))]; \ + msg.msg_iov= &msgiov; \ + msg.msg_iovlen= 1; \ msg.msg_control= msg##cbuf; \ msg.msg_controllen= sizeof(msg##cbuf); @@ -844,42 +1266,44 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) { assert(fd == connecting_fdpass_sock); - conn= xmalloc(sizeof(*conn)); - memset(conn,0,sizeof(*conn)); - PREP_DECL_MSG_CMSG(msg); + + ssize_t rs= recvmsg(fd, &msg, 0); + if (rs<0) { + if (isewouldblock(errno)) return OOP_CONTINUE; + syswarn("failed to read socket from connecting child"); + goto x; + } + + NEW(conn); + LIST_INIT(conn->waiting); + LIST_INIT(conn->priority); + LIST_INIT(conn->sent); + struct cmsghdr *h= 0; - ssize_t rs= recvmsg(fd, &msg, MSG_DONTWAIT); if (rs >= 0) h= CMSG_FIRSTHDR(&msg); if (!h) { - int status; - pid_t got= waitpid(connecting_child, &status, WNOHANG); - if (got != -1) { - assert(got==connecting_child); - connecting_child= 0; - if (WIFEXITED(status)) { - if (WEXITSTATUS(status) != 0 && - WEXITSTATUS(status) != CONNCHILD_ESTATUS_STREAM && - WEXITSTATUS(status) != CONNCHILD_ESTATUS_NOSTREAM) - /* child already reported the problem */; - else - warn("connect: connection child exited code %d but no cmsg", + int status= xwaitpid(&connecting_child, "connect child (broken)"); + + if (WIFEXITED(status)) { + if (WEXITSTATUS(status) != 0 && + WEXITSTATUS(status) != CONNCHILD_ESTATUS_STREAM && + WEXITSTATUS(status) != CONNCHILD_ESTATUS_NOSTREAM) + /* child already reported the problem */; + else { + if (e == OOP_EXCEPTION) + warn("connect: connection child exited code %d but" + " unexpected exception on fdpass socket", WEXITSTATUS(status)); - } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) { - warn("connect: connection attempt timed out"); - } else { - report_child_status("connect", status); + else + warn("connect: connection child exited code %d but" + " no cmsg (rs=%d)", + WEXITSTATUS(status), (int)rs); } + } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) { + warn("connect: connection attempt timed out"); } else { - /* child is still running apparently, report the socket problem */ - if (rs < 0) - syswarn("connect: read from fdpass socket failed"); - else if (e == OOP_EXCEPTION) - warn("connect: unexpected exception on fdpass socket"); - else if (!rs) - warn("connect: unexpected EOF on fdpass socket"); - else - fatal("connect: unexpected lack of cmsg from child"); + report_child_status("connect", status); } goto x; } @@ -915,29 +1339,28 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) { } /* Phew! */ - LIST_INIT(conn->waiting); - LIST_INIT(conn->priority); - LIST_INIT(conn->sent); conn->max_queue= conn->stream ? max_queue_per_conn : 1; loop->on_fd(loop, conn->fd, OOP_EXCEPTION, conn_exception, conn); conn->rd= oop_rd_new_fd(loop,conn->fd, 0, 0); /* sets nonblocking, too */ - if (!conn->fd) sysdie("oop_rd_new_fd (fd=%d)",conn->fd); + if (!conn->fd) die("oop_rd_new_fd conn failed (fd=%d)",conn->fd); int r= oop_rd_read(conn->rd, &peer_rd_style, NNTP_STRLEN, &peer_rd_ok, conn, &peer_rd_err, conn); if (r) sysdie("oop_rd_read for peer (fd=%d)",conn->fd); - notice("C%d connected %s", conn->fd, conn->stream ? "streaming" : "plain"); LIST_ADDHEAD(conns, conn); + notice("C%d (now %d) connected %s", + conn->fd, conns.count, conn->stream ? "streaming" : "plain"); connect_attempt_discard(); check_assign_articles(); - return 0; + return OOP_CONTINUE; x: conn_dispose(conn); connect_attempt_discard(); + reconnect_blocking_event(); return OOP_CONTINUE; } @@ -951,7 +1374,9 @@ static void connect_start(void) { assert(!connecting_child); assert(!connecting_fdpass_sock); - notice("starting connection attempt"); + info("starting connection attempt"); + int ok_until_connect= until_connect; + reconnect_blocking_event(); int socks[2]; int r= socketpair(AF_UNIX, SOCK_STREAM, 0, socks); @@ -968,13 +1393,26 @@ static void connect_start(void) { alarm(connection_setup_timeout); 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"); + int l= strlen(buf); + int stripped=0; + while (l>0) { + unsigned char c= buf[l-1]; + if (!isspace(c)) break; + if (c=='\n' || c=='\r') stripped=1; + --l; + } + if (!buf[0]) { + sysfatal("connect: connection attempt failed"); + } else { + buf[l]= 0; + fatal("connect: %s: %s", stripped ? "rejected" : "failed", + sanitise(buf,-1)); + } } 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) || + if (fputs("MODE STREAM\r\n", cn_to)==EOF || fflush(cn_to)) sysfatal("connect: could not send MODE STREAM"); buf[sizeof(buf)-1]= 0; @@ -986,15 +1424,15 @@ static void connect_start(void) { } int l= strlen(buf); assert(l>=1); - if (buf[-1]!='\n') + if (buf[l-1]!='\n') fatal("connect: response to MODE STREAM is too long: %.100s...", - sanitise(buf)); + sanitise(buf,-1)); l--; if (l>0 && buf[l-1]=='\r') l--; buf[l]= 0; char *ep; int rcode= strtoul(buf,&ep,10); if (ep != &buf[3]) - fatal("connect: bad response to MODE STREAM: %.50s", sanitise(buf)); + fatal("connect: bad response to MODE STREAM: %.50s", sanitise(buf,-1)); switch (rcode) { case 203: @@ -1005,7 +1443,7 @@ static void connect_start(void) { break; default: warn("connect: unexpected response to MODE STREAM: %.50s", - sanitise(buf)); + sanitise(buf,-1)); exitstatus= 2; break; } @@ -1021,21 +1459,44 @@ static void connect_start(void) { msg.msg_controllen= cmsg->cmsg_len; r= sendmsg(socks[1], &msg, 0); - if (r) sysdie("sendmsg failed for new connection"); + if (r<0) sysdie("sendmsg failed for new connection"); + if (r!=1) die("sendmsg for new connection gave wrong result %d",r); _exit(exitstatus); } xclose(socks[1], "connecting fdpass child's socket",0); connecting_fdpass_sock= socks[0]; + xsetnonblock(connecting_fdpass_sock, 1); on_fd_read_except(connecting_fdpass_sock, connchild_event); + + if (!conns.count) + until_connect= ok_until_connect; } /*---------- assigning articles to conns, and transmitting ----------*/ +static Article *dequeue_from(int peek, InputFile *ipf) { + if (!ipf) return 0; + if (peek) return LIST_HEAD(ipf->queue); + + Article *art= LIST_REMHEAD(ipf->queue); + if (!art) return 0; + check_reading_pause_resume(ipf); + return art; +} + +static Article *dequeue(int peek) { + Article *art; + art= dequeue_from(peek, flushing_input_file); if (art) return art; + art= dequeue_from(peek, backlog_input_file); if (art) return art; + art= dequeue_from(peek, main_input_file); if (art) return art; + return 0; +} + static void check_assign_articles(void) { for (;;) { - if (!queue.count) + if (!dequeue(1)) break; Conn *walk, *use=0; @@ -1047,7 +1508,7 @@ static void check_assign_articles(void) { * 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)) { + FOR_CONN(walk) { if (walk->quitting) continue; inqueue= walk->sent.count + walk->priority.count + walk->waiting.count; @@ -1060,13 +1521,14 @@ static void check_assign_articles(void) { if (use) { if (!inqueue) use->since_activity= 0; /* reset idle counter */ while (spare>0) { - Article *art= LIST_REMHEAD(queue); + Article *art= dequeue(0); + if (!art) break; LIST_ADDTAIL(use->waiting, art); + lowvol_perperiod[lowvol_circptr]++; spare--; } conn_maybe_write(use); } else if (allow_connect_start()) { - until_connect= reconnect_delay_periods; connect_start(); break; } else { @@ -1102,6 +1564,112 @@ static void conn_maybe_write(Conn *conn) { } } +/*---------- expiry, flow control and deferral ----------*/ + +/* + * flow control notes + * to ensure articles go away eventually + * separate queue for each input file + * queue expiry + * every period, check head of backlog queue for expiry with SMretrieve + * if too old: discard, and check next article + * also check every backlog article as we read it + * flush expiry + * after too long in SEPARATED/DROPPING ie Separated/Finishing/Dropping + * one-off: eat queued articles from flushing and write them to defer + * one-off: connfail all connections which have any articles from flushing + * newly read articles from flushing go straight to defer + * this should take care of it and get us out of this state + * to avoid filling up ram needlessly + * input control + * limit number of queued articles for each ipf + * pause/resume inputfile tailing + */ + +static void check_reading_pause_resume(InputFile *ipf) { + if (ipf->queue.count >= max_queue_per_ipf) + inputfile_reading_pause(ipf); + else + inputfile_reading_resume(ipf); +} + +static void article_defer(Article *art /* not on a queue */, int whichcount) { + open_defer(); + if (fprintf(defer, "%s %s\n", TokenToText(art->token), art->messageid) <0 + || fflush(defer)) + sysfatal("write to defer file %s",path_defer); + article_done(art, whichcount); +} + +static int article_check_expired(Article *art /* must be queued, not conn */) { + ARTHANDLE *artdata= SMretrieve(art->token, RETR_STAT); + if (artdata) { SMfreearticle(artdata); return 0; } + + LIST_REMOVE(art->ipf->queue, art); + art->missing= 1; + art->ipf->count_nooffer_missing++; + article_done(art,-1); + return 1; +} + +static void inputfile_queue_check_expired(InputFile *ipf) { + if (!ipf) return; + + for (;;) { + Article *art= LIST_HEAD(ipf->queue); + int exp= article_check_expired(art); + if (!exp) break; + } + check_reading_pause_resume(ipf); +} + +static void article_autodefer(InputFile *ipf, Article *art) { + ipf->autodefer++; + article_defer(art,-1); +} + +static int has_article_in(const ArticleList *al, InputFile *ipf) { + Article *art; + for (art=LIST_HEAD(*al); art; art=LIST_NEXT(art)) + if (art->ipf == ipf) return 1; + return 0; +} + +static void autodefer_input_file_articles(InputFile *ipf) { + Article *art; + while ((art= LIST_REMHEAD(ipf->queue))) + article_autodefer(ipf, art); +} + +static void autodefer_input_file(InputFile *ipf) { + static const char *const abandon= "stuck"; + ipf->autodefer= 0; + + autodefer_input_file_articles(ipf); + + if (ipf->inprogress) { + Conn *walk; + FOR_CONN(walk) { + if (has_article_in(&walk->waiting, ipf) || + has_article_in(&walk->priority, ipf) || + has_article_in(&walk->sent, ipf)) + walk->quitting= abandon; + } + while (ipf->inprogress) { + FOR_CONN(walk) + if (walk->quitting == abandon) goto found; + abort(); /* where are they ?? */ + + found: + connfail(walk, "connection is stuck or crawling," + " and we need to finish flush"); + autodefer_input_file_articles(ipf); + } + } + + check_reading_pause_resume(ipf); +} + /*========== article transmission ==========*/ static XmitDetails *xmit_core(Conn *conn, const char *data, int len, @@ -1126,7 +1694,6 @@ static void xmit_artbody(Conn *conn, ARTHANDLE *ah /* consumed */) { static void xmit_free(XmitDetails *d) { switch (d->kind) { - case xk_Malloc: free(d->info.malloc_tofree); break; case xk_Artdata: SMfreearticle(d->info.sm_art); break; case xk_Const: break; default: abort(); @@ -1153,15 +1720,18 @@ static void *conn_write_some_xmits(Conn *conn) { assert(rs > 0); int done; - for (done=0; rs && donexmitu; done++) { + for (done=0; rs; ) { + assert(donexmitu); struct iovec *vp= &conn->xmit[done]; XmitDetails *dp= &conn->xmitd[done]; - if (rs > vp->iov_len) { + if (rs >= vp->iov_len) { rs -= vp->iov_len; - xmit_free(dp); + xmit_free(dp); /* vp->iov_len -= vp->iov_len, etc. */ + done++; } else { vp->iov_base= (char*)vp->iov_base + rs; vp->iov_len -= rs; + break; /* rs -= rs */ } } int newu= conn->xmitu - done; @@ -1185,12 +1755,23 @@ static void conn_make_some_xmits(Conn *conn) { ARTHANDLE *artdata= SMretrieve(art->token, RETR_ALL); + art->state= + art->state == art_Unchecked ? art_Unsolicited : + art->state == art_Wanted ? art_Wanted : + (abort(),-1); + + if (!artdata) art->missing= 1; + art->ipf->counts[art->state][ artdata ? RC_sent : RC_missing ]++; + if (conn->stream) { if (artdata) { XMIT_LITERAL("TAKETHIS "); xmit_noalloc(conn, art->messageid, art->midlen); XMIT_LITERAL("\r\n"); xmit_artbody(conn, artdata); + } else { + article_done(art, -1); + continue; } } else { /* we got 235 from IHAVE */ @@ -1201,20 +1782,15 @@ static void conn_make_some_xmits(Conn *conn) { } } - 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); } else { /* check it */ if (conn->stream) - XMIT_LITERAL("IHAVE "); - else XMIT_LITERAL("CHECK "); + else + XMIT_LITERAL("IHAVE "); xmit_noalloc(conn, art->messageid, art->midlen); XMIT_LITERAL("\r\n"); @@ -1225,7 +1801,6 @@ static void conn_make_some_xmits(Conn *conn) { } } - /*========== handling responses from peer ==========*/ static const oop_rd_style peer_rd_style= { @@ -1315,8 +1890,10 @@ static void update_nocheck(int accepted) { nocheck= new_nocheck; } -static void article_done(Conn *conn, Article *art, int whichcount) { - art->ipf->counts[art->state][whichcount]++; +static void article_done(Article *art, int whichcount) { + if (whichcount>=0 && !art->missing) + art->ipf->counts[art->state][whichcount]++; + if (whichcount == RC_accepted) update_nocheck(1); else if (whichcount == RC_unwanted) update_nocheck(0); @@ -1324,6 +1901,11 @@ static void article_done(Conn *conn, Article *art, int whichcount) { while (art->blanklen) { static const char spaces[]= + " " + " " + " " + " " + " " " " " " " " @@ -1360,7 +1942,7 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_rd_event ev, } assert(ev == OOP_RD_OK); - char *sani= sanitise(data); + char *sani= sanitise(data,-1); char *ep; unsigned long code= strtoul(data, &ep, 10); @@ -1369,16 +1951,17 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_rd_event ev, return OOP_CONTINUE; } + int busy= conn_busy(conn); + if (conn->quitting) { - if (code!=205 && code!=503) { - connfail(conn, "peer gave unexpected response to QUIT: %s", sani); + if (code!=205 && code!=400) { + connfail(conn, "peer gave unexpected response to QUIT (%s): %s", + conn->quitting, sani); } else { - notice("C%d idle connection closed", conn->fd); - assert(!conn->waiting.count); - assert(!conn->priority.count); - assert(!conn->sent.count); - assert(!conn->xmitu); LIST_REMOVE(conns,conn); + notice("C%d (now %d) idle connection closed (%s)", + conn->fd, conns.count, conn->quitting); + assert(!busy); conn_dispose(conn); } return OOP_CONTINUE; @@ -1387,25 +1970,38 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_rd_event ev, conn->since_activity= 0; Article *art; -#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 GET_ARTICLE(musthavesent) do{ \ + art= article_reply_check(conn, data, code_streaming, musthavesent, sani); \ + if (!art) return OOP_CONTINUE; /* reply_check has failed the conn */ \ + }while(0) -#define ARTICLE_DEALTWITH(streaming,musthavesent,how) \ - code_streaming= (streaming); \ - GET_ARTICLE(musthavesent); \ - article_done(conn, art, RC_##how); break; +#define ARTICLE_DEALTWITH(streaming,musthavesent,how) do{ \ + code_streaming= (streaming); \ + GET_ARTICLE(musthavesent); \ + article_done(art, RC_##how); \ + goto dealtwith; \ + }while(0) -#define PEERBADMSG(m) connfail(conn, m ": %s", sani); return OOP_CONTINUE +#define PEERBADMSG(m) do { \ + connfail(conn, m ": %s", sani); return OOP_CONTINUE; \ + }while(0) int code_streaming= 0; switch (code) { - case 400: PEERBADMSG("peer stopped accepting articles"); - case 503: PEERBADMSG("peer timed us out"); default: PEERBADMSG("peer sent unexpected message"); + case 400: + if (busy) + PEERBADMSG("peer timed us out or stopped accepting articles"); + + LIST_REMOVE(conns,conn); + notice("C%d (now %d) idle connection closed by peer", + conns.count, conn->fd); + conn_dispose(conn); + return OOP_CONTINUE; + case 435: ARTICLE_DEALTWITH(0,0,unwanted); /* IHAVE says they have it */ case 438: ARTICLE_DEALTWITH(1,0,unwanted); /* CHECK/TAKETHIS: they have it */ @@ -1429,14 +2025,11 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_rd_event ev, code_streaming= 1; case 436: /* IHAVE says try later */ GET_ARTICLE(0); - open_defer(); - if (fprintf(defer, "%s %s\n", TokenToText(art->token), art->messageid) <0 - || fflush(defer)) - sysfatal("write to defer file %s",path_defer); - article_done(conn, art, RC_deferred); + article_defer(art, RC_deferred); break; } +dealtwith: conn_maybe_write(conn); check_assign_articles(); @@ -1462,7 +2055,7 @@ static void feedfile_eof(InputFile *ipf) { } static InputFile *open_input_file(const char *path) { - int fd= open(path, O_RDONLY); + int fd= open(path, O_RDWR); if (fd<0) { if (errno==ENOENT) return 0; sysfatal("unable to open input file %s", path); @@ -1473,6 +2066,8 @@ static InputFile *open_input_file(const char *path) { memset(ipf,0,sizeof(*ipf)); ipf->fd= fd; + ipf->autodefer= -1; + LIST_INIT(ipf->queue); strcpy(ipf->path, path); return ipf; @@ -1491,8 +2086,8 @@ static void close_input_file(InputFile *ipf) { /* does not free */ 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)); + warn("corrupted file: %s, offset %lu: %s: in %s", + ipf->path, (unsigned long)offset, how, sanitise(data,-1)); ipf->readcount_err++; if (ipf->readcount_err > max_bad_data_initial + (ipf->readcount_ok+ipf->readcount_blank) / max_bad_data_ratio) @@ -1523,7 +2118,7 @@ static void *feedfile_got_article(oop_source *lp, oop_read *rd, if (!data) { feedfile_eof(ipf); return OOP_CONTINUE; } off_t old_offset= ipf->offset; - ipf->offset += recsz + 1; + ipf->offset += recsz + !!(ev == OOP_RD_OK); #define X_BAD_DATA(m) return feedfile_got_bad_data(ipf,old_offset,data,m); @@ -1563,20 +2158,27 @@ static void *feedfile_got_article(oop_source *lp, oop_read *rd, ipf->readcount_ok++; art= xmalloc(sizeof(*art) - 1 + midlen + 1); + memset(art,0,sizeof(*art)); art->state= art_Unchecked; art->midlen= midlen; art->ipf= ipf; ipf->inprogress++; art->token= TextToToken(tokentextbuf); - art->offset= ipf->offset; + art->offset= old_offset; art->blanklen= recsz; strcpy(art->messageid, space+1); - LIST_ADDTAIL(queue, art); + LIST_ADDTAIL(ipf->queue, art); + + if (ipf->autodefer >= 0) + article_autodefer(ipf, art); + else if (ipf==backlog_input_file) + article_check_expired(art); if (sms==sm_NORMAL && ipf==main_input_file && ipf->offset >= target_max_feedfile_size) statemc_start_flush("feed file size"); - check_assign_articles(); + check_assign_articles(); /* may destroy conn but that's OK */ + check_reading_pause_resume(ipf); return OOP_CONTINUE; } @@ -1584,7 +2186,16 @@ static void *feedfile_got_article(oop_source *lp, oop_read *rd, static void *tailing_rable_call_time(oop_source *loop, struct timeval tv, void *user) { + /* lifetime of ipf here is OK because destruction will cause + * on_cancel which will cancel this callback */ InputFile *ipf= user; + + if (!ipf->fake_readable) return OOP_CONTINUE; + + /* we just keep calling readable until our caller (oop_rd) + * has called try_read, and try_read has found EOF so given EAGAIN */ + loop->on_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf); + return ipf->readable_callback(loop, &ipf->readable, ipf->readable_callback_user); } @@ -1597,9 +2208,10 @@ static void tailing_on_cancel(struct oop_readable *rable) { 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 */ +static void tailing_make_readable(InputFile *ipf) { + if (!ipf || !ipf->readable_callback) /* so callers can be naive */ + return; + ipf->fake_readable= 1; loop->on_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf); } @@ -1611,8 +2223,7 @@ static int tailing_on_readable(struct oop_readable *rable, ipf->readable_callback= cb; ipf->readable_callback_user= user; filemon_start(ipf); - - tailing_queue_readable(ipf); + tailing_make_readable(ipf); return 0; } @@ -1623,11 +2234,13 @@ static ssize_t tailing_try_read(struct oop_readable *rable, void *buffer, ssize_t r= read(ipf->fd, buffer, length); if (r==-1) { if (errno==EINTR) continue; + ipf->fake_readable= 0; return r; } if (!r) { if (ipf==main_input_file) { errno=EAGAIN; + ipf->fake_readable= 0; return -1; } else if (ipf==flushing_input_file) { assert(ipf->rd); @@ -1644,10 +2257,10 @@ static ssize_t tailing_try_read(struct oop_readable *rable, void *buffer, /*---------- filemon implemented with inotify ----------*/ -#if defined(HAVE_INOTIFY) && !defined(HAVE_FILEMON) +#if defined(HAVE_SYS_INOTIFY_H) && !defined(HAVE_FILEMON) #define HAVE_FILEMON -#include +#include static int filemon_inotify_fd; static int filemon_inotify_wdmax; @@ -1659,11 +2272,11 @@ struct Filemon_Perfile { static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) { int wd= inotify_add_watch(filemon_inotify_fd, ipf->path, IN_MODIFY); - if (wd < 0) sysdie("inotify_add_watch %s", ipf->path); + if (wd < 0) sysfatal("inotify_add_watch %s", ipf->path); if (wd >= filemon_inotify_wdmax) { int newmax= wd+2; - filemon_inotify_wd= xrealloc(filemon_inotify_wd2ipf, + filemon_inotify_wd2ipf= xrealloc(filemon_inotify_wd2ipf, sizeof(*filemon_inotify_wd2ipf) * newmax); memset(filemon_inotify_wd2ipf + filemon_inotify_wdmax, 0, sizeof(*filemon_inotify_wd2ipf) * (newmax - filemon_inotify_wdmax)); @@ -1682,7 +2295,7 @@ static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) { 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); + int r= inotify_rm_watch(filemon_inotify_fd, wd); if (r) sysdie("inotify_rm_watch"); filemon_inotify_wd2ipf[wd]= 0; } @@ -1701,8 +2314,8 @@ static void *filemon_inotify_readable(oop_source *lp, int fd, die("inotify read %d bytes wanted struct of %d", r, (int)sizeof(iev)); } InputFile *ipf= filemon_inotify_wd2ipf[iev.wd]; - debug("filemon inotify readable read %p wd=%p", iev.wd, ipf); - filemon_callback(ipf); + /*debug("filemon inotify readable read %p wd=%d", ipf, iev.wd);*/ + tailing_make_readable(ipf); } return OOP_CONTINUE; } @@ -1710,16 +2323,25 @@ static void *filemon_inotify_readable(oop_source *lp, int fd, static int filemon_method_init(void) { filemon_inotify_fd= inotify_init(); if (filemon_inotify_fd<0) { - syswarn("could not initialise inotify: inotify_init failed"); + syswarn("filemon/inotify: inotify_init failed"); return 0; } - set nonblock; - loop->on_fd(loop, filemon_inotify_fd, OOP_READ, filemon_inotify_readable); + xsetnonblock(filemon_inotify_fd, 1); + loop->on_fd(loop, filemon_inotify_fd, OOP_READ, filemon_inotify_readable, 0); debug("filemon inotify init filemon_inotify_fd=%d", filemon_inotify_fd); return 1; } +static void filemon_method_dump_info(FILE *f) { + int i; + fprintf(f,"inotify"); + DUMPV("%d",,filemon_inotify_fd); + DUMPV("%d",,filemon_inotify_wdmax); + for (i=0; ifilemon); - ipf->filemon= xmalloc(sizeof(*ipf->filemon)); - memset(ipf->filemon, 0, sizeof(*ipf->filemon)); + NEW(ipf->filemon); filemon_method_startfile(ipf, ipf->filemon); } @@ -1751,10 +2376,6 @@ static void filemon_stop(InputFile *ipf) { 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= { @@ -1763,6 +2384,24 @@ static const oop_rd_style feedfile_rdstyle= { OOP_RD_SHORTREC_LONG, }; +static void inputfile_reading_resume(InputFile *ipf) { + if (!ipf->rd) return; + if (!ipf->paused) return; + + int r= oop_rd_read(ipf->rd, &feedfile_rdstyle, MAX_LINE_FEEDFILE, + feedfile_got_article,ipf, feedfile_read_err, ipf); + if (r) sysdie("unable start reading feedfile %s",ipf->path); + + ipf->paused= 0; +} + +static void inputfile_reading_pause(InputFile *ipf) { + if (!ipf->rd) return; + if (ipf->paused) return; + oop_rd_cancel(ipf->rd); + ipf->paused= 1; +} + static void inputfile_reading_start(InputFile *ipf) { assert(!ipf->rd); ipf->readable.on_readable= tailing_on_readable; @@ -1777,14 +2416,13 @@ static void inputfile_reading_start(InputFile *ipf) { ipf->rd= oop_rd_new(loop, &ipf->readable, 0,0); assert(ipf->rd); - int r= oop_rd_read(ipf->rd, &feedfile_rdstyle, MAX_LINE_FEEDFILE, - feedfile_got_article,ipf, feedfile_read_err, ipf); - if (r) sysdie("unable start reading feedfile %s",ipf->path); + ipf->paused= 1; + inputfile_reading_resume(ipf); } static void inputfile_reading_stop(InputFile *ipf) { assert(ipf->rd); - oop_rd_cancel(ipf->rd); + inputfile_reading_pause(ipf); oop_rd_delete(ipf->rd); ipf->rd= 0; assert(!ipf->filemon); /* we shouldn't be monitoring it now */ @@ -1853,7 +2491,7 @@ static void inputfile_reading_stop(InputFile *ipf) { | flsh->rd!=0 | | flsh->rd!=0 | [Separated] | | [Dropping] | main F idle | | main none - | old D tail | | old D tail + | flsh D tail | | flsh D tail | ============= | | ============ | | | | install | ^ | EOF ON D | | defer | EOF ON D @@ -1863,7 +2501,7 @@ static void inputfile_reading_stop(InputFile *ipf) { | flsh->rd==0 | V flsh->rd==0 | [Finishing] | | [Dropping] | main F tail | `. main none - | old D closed | `. old D closed + | flsh D closed | `. flsh D closed | =============== V `. =============== | | `. | | | ALL D PROCESSED `. | ALL D PROCESSED @@ -1876,7 +2514,7 @@ static void inputfile_reading_stop(InputFile *ipf) { DROPPED [Dropped] main none - old none + flsh none some backlog ============== | @@ -1898,16 +2536,12 @@ static void startup_set_input_file(InputFile *f) { inputfile_reading_start(f); } -static void statemc_init(void) { +static void statemc_lock(void) { + int lockfd; struct stat stab, stabf; - - path_lock= xasprintf("%s_lock", feedfile); - path_flushing= xasprintf("%s_flushing", feedfile); - path_defer= xasprintf("%s_defer", feedfile); - globpat_backlog= xasprintf("%s_backlog*", feedfile); - + for (;;) { - int lockfd= open(path_lock, O_CREAT|O_RDWR, 0600); + lockfd= open(path_lock, O_CREAT|O_RDWR, 0600); if (lockfd<0) sysfatal("open lockfile %s", path_lock); struct flock fl; @@ -1920,7 +2554,7 @@ static void statemc_init(void) { if (quiet_multiple) exit(0); fatal("another duct holds the lockfile"); } - sysdie("fcntl F_SETLK lockfile %s", path_lock); + sysfatal("fcntl F_SETLK lockfile %s", path_lock); } xfstat_isreg(lockfd, &stabf, path_lock, "lockfile"); @@ -1932,17 +2566,34 @@ static void statemc_init(void) { xclose(lockfd, "stale lockfile ", path_lock); } + + FILE *lockfile= fdopen(lockfd, "w"); + if (!lockfile) sysdie("fdopen lockfile"); + + int r= ftruncate(lockfd, 0); + if (r) sysdie("truncate lockfile to write new info"); + + if (fprintf(lockfile, "pid %ld\nsite %s\nfeedfile %s\nfqdn %s\n", + (unsigned long)self_pid, + sitename, feedfile, remote_host) == EOF || + fflush(lockfile)) + sysfatal("write info to lockfile %s", path_lock); + debug("startup: locked"); +} + +static void statemc_init(void) { + struct stat stabdefer; search_backlog_file(); int defer_noent; - xlstat_isreg(path_defer, &stab, &defer_noent, "defer file"); + xlstat_isreg(path_defer, &stabdefer, &defer_noent, "defer file"); if (defer_noent) { debug("startup: ductdefer ENOENT"); } else { - debug("startup: ductdefer nlink=%ld", (long)stab.st_nlink); - switch (stab.st_nlink==1) { + debug("startup: ductdefer nlink=%ld", (long)stabdefer.st_nlink); + switch (stabdefer.st_nlink==1) { case 1: open_defer(); /* so that we will later close it and rename it */ break; @@ -1952,7 +2603,7 @@ static void statemc_init(void) { break; default: die("defer file %s has unexpected link count %d", - path_defer, stab.st_nlink); + path_defer, stabdefer.st_nlink); } } @@ -1979,13 +2630,17 @@ static void statemc_init(void) { if (file_d) { debug("startup: F!=D => Separated"); startup_set_input_file(file_d); - SMS(SEPARATED, 0, "found both old and current feed files"); + flushing_input_file= main_input_file; + main_input_file= open_input_file(feedfile); + if (!main_input_file) die("feedfile vanished during startup"); + SMS(SEPARATED, max_separated_periods, + "found both old and current feed files"); } 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"); + SMS(NORMAL, spontaneous_flush_periods, "normal startup"); } } } @@ -1997,11 +2652,11 @@ static void statemc_start_flush(const char *why) { /* Normal => Flushing */ why, (unsigned long)(main_input_file ? main_input_file->offset : 0), (unsigned long)target_max_feedfile_size, - sm_period_counter); + until_flush); int r= link(feedfile, path_flushing); - if (r) sysdie("link feedfile %s to flushing file %s", - feedfile, path_flushing); + if (r) sysfatal("link feedfile %s to flushing file %s", + feedfile, path_flushing); /* => Hardlinked */ xunlink(feedfile, "old feedfile link"); @@ -2010,24 +2665,40 @@ static void statemc_start_flush(const char *why) { /* Normal => Flushing */ 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; +static int trigger_flush_ok(const char *why) { switch (sms) { + case sm_NORMAL: - statemc_start_flush("periodic"); /* Normal => Flushing; => FLUSHING */ - break; + statemc_start_flush(why ? why : "periodic"); + return 1; /* Normal => Flushing; => FLUSHING */ + case sm_FLUSHFAILED: - spawn_inndcomm_flush("retry"); /* Moved => Flushing; => FLUSHING */ - break; + spawn_inndcomm_flush(why ? why : "retry"); + return 1; /* Moved => Flushing; => FLUSHING */ + + case sm_SEPARATED: + case sm_DROPPING: + warn("abandoning old feedfile after flush (%s), autodeferring", + why ? why : "took too long to complete"); + assert(flushing_input_file); + autodefer_input_file(flushing_input_file); + return 1; + default: - abort(); + return 0; } } +static void statemc_period_poll(void) { + if (!until_flush) return; + until_flush--; + assert(until_flush>=0); + + if (until_flush) return; + int ok= trigger_flush_ok(0); + assert(ok); +} + static int inputfile_is_done(InputFile *ipf) { if (!ipf) return 0; if (ipf->inprogress) return 0; /* new article in the meantime */ @@ -2035,20 +2706,30 @@ static int inputfile_is_done(InputFile *ipf) { return 1; } -static void notice_processed(InputFile *ipf, const char *what, - const char *spec) { +static void notice_processed(InputFile *ipf, int completed, + const char *what, const char *spec) { + if (!ipf) return; /* allows preterminate to be lazy */ + #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)" + char *inprog= completed + ? xasprintf("%s","") /* GCC produces a stupid warning for printf("") ! */ + : xasprintf(" inprogress=%ld", ipf->inprogress); + char *autodefer= ipf->autodefer >= 0 + ? xasprintf(" autodeferred=%ld", ipf->autodefer) + : xasprintf("%s",""); + + info("%s %s%s read=%d (+bl=%d,+err=%d)%s%s" + " missing=%d offered=%d (ch=%d,nc=%d) accepted=%d (ch=%d,nc=%d)" RESULT_COUNTS(RCI_NOTHING, RCI_TRIPLE_FMT) , - what, spec, + completed?"completed":"processed", what, spec, ipf->readcount_ok, ipf->readcount_blank, ipf->readcount_err, + inprog, autodefer, ipf->count_nooffer_missing, CNT(Unchecked,sent) + CNT(Unsolicited,sent) , CNT(Unchecked,sent), CNT(Unsolicited,sent), CNT(Wanted,accepted) + CNT(Unsolicited,accepted) @@ -2056,6 +2737,9 @@ static void notice_processed(InputFile *ipf, const char *what, RESULT_COUNTS(RCI_NOTHING, RCI_TRIPLE_VALS) ); + free(inprog); + free(autodefer); + #undef CNT } @@ -2068,7 +2752,7 @@ static void statemc_check_backlog_done(void) { const char *under= strchr(slash, '_'); const char *rest= under ? under+1 : leaf; if (!strncmp(rest,"backlog",7)) rest += 7; - notice_processed(ipf,"backlog:",rest); + notice_processed(ipf,1,"backlog ",rest); close_input_file(ipf); if (unlink(ipf->path)) { @@ -2090,7 +2774,7 @@ static void statemc_check_flushing_done(void) { assert(sms==sm_SEPARATED || sms==sm_DROPPING); - notice_processed(ipf,"feedfile",0); + notice_processed(ipf,1,"feedfile",""); close_defer(); @@ -2102,9 +2786,9 @@ static void statemc_check_flushing_done(void) { if (sms==sm_SEPARATED) { notice("flush complete"); - SMS(NORMAL, 0, "flush complete"); + SMS(NORMAL, spontaneous_flush_periods, "flush complete"); } else if (sms==sm_DROPPING) { - SMS(DROPPED, 0, "old flush complete"); + SMS(DROPPED, max_separated_periods, "old flush complete"); search_backlog_file(); notice("feed dropped, but will continue until backlog is finished"); } @@ -2125,7 +2809,7 @@ static void queue_check_input_done(void) { static void statemc_setstate(StateMachineState newsms, int periods, const char *forlog, const char *why) { sms= newsms; - sm_period_counter= periods; + until_flush= periods; const char *xtra= ""; switch (sms) { @@ -2141,9 +2825,9 @@ static void statemc_setstate(StateMachineState newsms, int periods, } if (periods) { - info("%s%s[%d] %s",forlog,xtra,periods,why); + info("state %s%s[%d] %s",forlog,xtra,periods,why); } else { - info("%s%s %s",forlog,xtra,why); + info("state %s%s %s",forlog,xtra,why); } } @@ -2255,9 +2939,9 @@ static void search_backlog_file(void) { switch (r) { case GLOB_ABORTED: - sysdie("failed to expand backlog pattern %s", globpat_backlog); + sysfatal("failed to expand backlog pattern %s", globpat_backlog); case GLOB_NOSPACE: - die("out of memory expanding backlog pattern %s", globpat_backlog); + fatal("out of memory expanding backlog pattern %s", globpat_backlog); case 0: for (i=0; ipath); +} + +static int signal_self_pipe[2]; +static sig_atomic_t terminate_sig_flag; + +static void raise_default(int signo) { + xsigsetdefault(signo); + raise(signo); + abort(); +} + +static void *sigarrived_event(oop_source *lp, int fd, oop_event e, void *u) { + assert(fd=signal_self_pipe[0]); + char buf[PIPE_BUF]; + int r= read(signal_self_pipe[0], buf, sizeof(buf)); + if (r<0 && !isewouldblock(errno)) sysdie("failed to read signal self pipe"); + if (r==0) die("eof on signal self pipe"); + if (terminate_sig_flag) { + preterminate(); + notice("terminating (%s)", strsignal(terminate_sig_flag)); + raise_default(terminate_sig_flag); + } + return OOP_CONTINUE; +} + +static void sigarrived_handler(int signum) { + static char x; + switch (signum) { + case SIGTERM: + case SIGINT: + if (!terminate_sig_flag) terminate_sig_flag= signum; + break; + default: + abort(); + } + write(signal_self_pipe[1],&x,1); +} + +static void init_signals(void) { + if (signal(SIGPIPE, SIG_IGN) == SIG_ERR) + sysdie("could not ignore SIGPIPE"); + + if (pipe(signal_self_pipe)) sysfatal("create self-pipe for signals"); + + xsetnonblock(signal_self_pipe[0],1); + xsetnonblock(signal_self_pipe[1],1); + + struct sigaction sa; + memset(&sa,0,sizeof(sa)); + sa.sa_handler= sigarrived_handler; + sa.sa_flags= SA_RESTART; + xsigaction(SIGTERM,&sa); + xsigaction(SIGINT,&sa); + + on_fd_read_except(signal_self_pipe[0], sigarrived_event); +} + /*========== flushing the feed ==========*/ static pid_t inndcomm_child; @@ -2341,8 +3097,11 @@ static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) { assert(inndcomm_child); assert(fd == inndcomm_sentinel_fd); int status= xwaitpid(&inndcomm_child, "inndcomm"); + inndcomm_child= 0; + cancel_fd_read_except(fd); xclose_perhaps(&fd, "inndcomm sentinel pipe",0); + inndcomm_sentinel_fd= 0; assert(!flushing_input_file); @@ -2353,9 +3112,9 @@ static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) { goto failed; case INNDCOMMCHILD_ESTATUS_NONESUCH: - warn("feed has been dropped by innd, finishing up"); + notice("feed has been dropped by innd, finishing up"); flushing_input_file= main_input_file; - tailing_queue_readable(flushing_input_file); + tailing_make_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 */ @@ -2363,7 +3122,8 @@ static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) { main_input_file= 0; if (flushing_input_file) { - SMS(DROPPING, 0, "feed dropped by innd, but must finish last flush"); + SMS(DROPPING, max_separated_periods, + "feed dropped by innd, but must finish last flush"); } else { close_defer(); SMS(DROPPED, 0, "feed dropped by innd"); @@ -2374,17 +3134,19 @@ static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) { case 0: /* as above */ flushing_input_file= main_input_file; - tailing_queue_readable(flushing_input_file); + tailing_make_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); + die("flush succeeded but feedfile %s does not exist!" + " (this probably means feedfile does not correspond" + " to site %s in newsfeeds)", feedfile, sitename); if (flushing_input_file) { - SMS(SEPARATED, spontaneous_flush_periods, "recovery flush complete"); + SMS(SEPARATED, max_separated_periods, "flush complete"); } else { close_defer(); - SMS(NORMAL, spontaneous_flush_periods, "flush complete"); + SMS(NORMAL, spontaneous_flush_periods, "recovery flush complete"); } return OOP_CONTINUE; @@ -2419,7 +3181,7 @@ void spawn_inndcomm_flush(const char *why) { /* Moved => Flushing */ assert(!inndcomm_child); assert(!inndcomm_sentinel_fd); - if (pipe(pipefds)) sysdie("create pipe for inndcomm child sentinel"); + if (pipe(pipefds)) sysfatal("create pipe for inndcomm child sentinel"); inndcomm_child= xfork("inndcomm child"); @@ -2431,6 +3193,12 @@ void spawn_inndcomm_flush(const char *why) { /* Moved => Flushing */ xclose(pipefds[0], "(in child) inndcomm sentinel parent's end",0); /* parent spots the autoclose of pipefds[1] when we die or exit */ + if (simulate_flush>=0) { + warn("SIMULATING flush child status %d", simulate_flush); + if (simulate_flush>128) raise(simulate_flush-128); + else exit(simulate_flush); + } + alarm(inndcomm_flush_timeout); r= ICCopen(); if (r) inndcommfail("connect"); r= ICCcommand('f',flushargv,&reply); if (r<0) inndcommfail("transmit"); @@ -2441,6 +3209,8 @@ void spawn_inndcomm_flush(const char *why) { /* Moved => Flushing */ exit(INNDCOMMCHILD_ESTATUS_FAIL); } + simulate_flush= -1; + xclose(pipefds[1], "inndcomm sentinel child's end",0); inndcomm_sentinel_fd= pipefds[0]; assert(inndcomm_sentinel_fd); @@ -2463,62 +3233,90 @@ static void postfork_stdio(FILE *f, const char *what, const char *what2) { } static void postfork(void) { - if (signal(SIGPIPE, SIG_DFL) == SIG_ERR) - sysdie("(in child) failed to reset SIGPIPE"); + in_child= 1; + + xsigsetdefault(SIGTERM); + xsigsetdefault(SIGINT); + xsigsetdefault(SIGPIPE); + if (terminate_sig_flag) raise(terminate_sig_flag); postfork_inputfile(main_input_file); postfork_inputfile(flushing_input_file); Conn *conn; - for (conn=LIST_HEAD(conns); conn; conn=LIST_NEXT(conn)) + FOR_CONN(conn) conn_closefd(conn,"(in child) "); postfork_stdio(defer, "defer file ", path_defer); } -#define EVERY(what, interval_sec, interval_usec, body) \ - static struct timeval what##_timeout = { interval_sec, interval_usec }; \ - static void what##_schedule(void); \ - static void *what##_timedout(oop_source *lp, struct timeval tv, void *u) { \ - body; \ - what##_schedule(); \ - return OOP_CONTINUE; \ - } \ - static void what##_schedule(void) { \ - loop->on_time(loop, what##_timeout, what##_timedout, 0); \ - } +typedef struct Every Every; +struct Every { + struct timeval interval; + int fixed_rate; + void (*f)(void); +}; + +static void every_schedule(Every *e, struct timeval base); + +static void *every_happens(oop_source *lp, struct timeval base, void *e_v) { + Every *e= e_v; + e->f(); + if (!e->fixed_rate) xgettimeofday(&base); + every_schedule(e, base); + return OOP_CONTINUE; +} + +static void every_schedule(Every *e, struct timeval base) { + struct timeval when; + timeradd(&base, &e->interval, &when); + loop->on_time(loop, when, every_happens, e); +} -EVERY(filepoll, 5,0, ({ - if (main_input_file && main_input_file->readable_callback) - filemon_callback(main_input_file); -})); +static void every(int interval, int fixed_rate, void (*f)(void)) { + NEW_DECL(Every *,e); + e->interval.tv_sec= interval; + e->interval.tv_usec= 0; + e->fixed_rate= fixed_rate; + e->f= f; + struct timeval now; + xgettimeofday(&now); + every_schedule(e, now); +} + +static void filepoll(void) { + tailing_make_readable(main_input_file); + tailing_make_readable(flushing_input_file); +} static char *debug_report_ipf(InputFile *ipf) { - if (!ipf) return xasprintf("-"); + if (!ipf) return xasprintf("none"); const char *slash= strrchr(ipf->path,'/'); const char *path= slash ? slash+1 : ipf->path; - return xasprintf("%p/%s:ip=%ld,off=%ld,fd=%d%s", + return xasprintf("%p/%s:queue=%d,ip=%ld,autodef=%ld,off=%ld,fd=%d%s%s%s", ipf, path, - ipf->inprogress, (long)ipf->offset, - ipf->fd, ipf->rd ? "+" : ""); + ipf->queue.count, ipf->inprogress, ipf->autodefer, + (long)ipf->offset, ipf->fd, + ipf->rd ? "" : ",!rd", + ipf->skippinglong ? "*skiplong" : "", + ipf->rd && ipf->paused ? "*paused" : ""); } -EVERY(period, -1,0, ({ +static void period(void) { char *dipf_main= debug_report_ipf(main_input_file); char *dipf_flushing= debug_report_ipf(flushing_input_file); char *dipf_backlog= debug_report_ipf(backlog_input_file); debug("PERIOD" - " sms=%s[%d] conns=%d queue=%d until_connect=%d" - " input_files main:%s old:%s flushing:%s" - " children connecting=%ld inndcomm_child=%ld" + " sms=%s[%d] conns=%d until_connect=%d" + " input_files main:%s flushing:%s backlog:%s[%d]" + " children connecting=%ld inndcomm=%ld lowvol_total=%d" , - sms_names[sms], sm_period_counter, - conns.count, queue.count, until_connect, - dipf_main, dipf_flushing, dipf_backlog, - (long)connecting_child, (long)inndcomm_child + sms_names[sms], until_flush, conns.count, until_connect, + dipf_main, dipf_flushing, dipf_backlog, until_backlog_nextscan, + (long)connecting_child, (long)inndcomm_child, lowvol_total ); free(dipf_main); @@ -2527,13 +3325,153 @@ EVERY(period, -1,0, ({ if (until_connect) until_connect--; + inputfile_queue_check_expired(backlog_input_file); poll_backlog_file(); if (!backlog_input_file) close_defer(); /* want to start on a new backlog */ statemc_period_poll(); check_assign_articles(); check_idle_conns(); -})); +} + + +/*========== dumping state ==========*/ + +static void dump_article_list(FILE *f, const CliCommand *c, + const ArticleList *al) { + fprintf(f, " count=%d\n", al->count); + if (!c->xval) return; + + int i; Article *art; + for (i=0, art=LIST_HEAD(*al); art; i++, art=LIST_NEXT(art)) { + fprintf(f," #%05d %-11s", i, artstate_names[art->state]); + DUMPV("%p", art->,ipf); + DUMPV("%d", art->,missing); + DUMPV("%lu", (unsigned long)art->,offset); + DUMPV("%d", art->,blanklen); + DUMPV("%d", art->,midlen); + fprintf(f, " %s %s\n", TokenToText(art->token), art->messageid); + } +} + +static void dump_input_file(FILE *f, const CliCommand *c, + InputFile *ipf, const char *wh) { + char *dipf= debug_report_ipf(ipf); + fprintf(f,"input %s %s", wh, dipf); + free(dipf); + + if (ipf) { + DUMPV("%d", ipf->,readcount_ok); + DUMPV("%d", ipf->,readcount_blank); + DUMPV("%d", ipf->,readcount_err); + DUMPV("%d", ipf->,count_nooffer_missing); + } + fprintf(f,"\n"); + if (ipf) { + ArtState state; const char *const *statename; + for (state=0, statename=artstate_names; *statename; state++,statename++) { +#define RC_DUMP_FMT(x) " " #x "=%d" +#define RC_DUMP_VAL(x) ,ipf->counts[state][RC_##x] + fprintf(f,"input %s counts %-11s" + RESULT_COUNTS(RC_DUMP_FMT,RC_DUMP_FMT) "\n", + wh, *statename + RESULT_COUNTS(RC_DUMP_VAL,RC_DUMP_VAL)); + } + fprintf(f,"input %s queue", wh); + dump_article_list(f,c,&ipf->queue); + } +} + +CCMD(dump) { + int i; + fprintf(cc->out, "dumping state to %s\n", path_dump); + FILE *f= fopen(path_dump, "w"); + if (!f) { fprintf(cc->out, "failed: open: %s\n", strerror(errno)); return; } + + fprintf(f,"general"); + DUMPV("%s", sms_names,[sms]); + DUMPV("%d", ,until_flush); + DUMPV("%ld", (long),self_pid); + DUMPV("%p", , defer); + DUMPV("%d", , until_connect); + DUMPV("%d", , until_backlog_nextscan); + DUMPV("%d", , simulate_flush); + fprintf(f,"\nnocheck"); + DUMPV("%#.10f", , accept_proportion); + DUMPV("%d", , nocheck); + DUMPV("%d", , nocheck_reported); + fprintf(f,"\n"); + + fprintf(f,"special"); + DUMPV("%ld", (long),connecting_child); + DUMPV("%d", , connecting_fdpass_sock); + DUMPV("%d", , cli_master); + fprintf(f,"\n"); + + fprintf(f,"lowvol"); + DUMPV("%d", , lowvol_circptr); + DUMPV("%d", , lowvol_total); + fprintf(f,":"); + for (i=0; ifd); + DUMPV("%p",conn->,rd); DUMPV("%d",conn->,max_queue); + DUMPV("%d",conn->,stream); DUMPV("\"%s\"",conn->,quitting); + DUMPV("%d",conn->,since_activity); + fprintf(f,"\n"); + + fprintf(f,"C%d waiting", conn->fd); dump_article_list(f,c,&conn->waiting); + fprintf(f,"C%d priority",conn->fd); dump_article_list(f,c,&conn->priority); + fprintf(f,"C%d sent", conn->fd); dump_article_list(f,c,&conn->sent); + + fprintf(f,"C%d xmit xmitu=%d\n", conn->fd, conn->xmitu); + for (i=0; ixmitu; i++) { + const struct iovec *iv= &conn->xmit[i]; + const XmitDetails *xd= &conn->xmitd[i]; + char *dinfo; + switch (xd->kind) { + case xk_Const: dinfo= xasprintf("Const"); break; + case xk_Artdata: dinfo= xasprintf("A%p", xd->info.sm_art); break; + default: + abort(); + } + fprintf(f," #%03d %-11s l=%d %s\n", i, dinfo, iv->iov_len, + sanitise(iv->iov_base, iv->iov_len)); + free(dinfo); + } + } + fprintf(f,"paths"); + DUMPV("%s", , feedfile); + DUMPV("%s", , path_cli); + DUMPV("%s", , path_lock); + DUMPV("%s", , path_flushing); + DUMPV("%s", , path_defer); + DUMPV("%s", , path_dump); + DUMPV("%s", , globpat_backlog); + fprintf(f,"\n"); + + if (!!ferror(f) + !!fclose(f)) { + fprintf(cc->out, "failed: write: %s\n", strerror(errno)); + return; + } +} /*========== option parsing ==========*/ @@ -2543,7 +3481,7 @@ static void vbadusage(const char *fmt, va_list al) { fprintf(stderr, "bad usage: %s\n" "say --help for help, or read the manpage\n", m); - if (become_daemon) + if (interactive < 2) syslog(LOG_CRIT,"innduct: invoked with bad usage: %s",m); exit(8); } @@ -2675,10 +3613,14 @@ static void op_seconds(const Option *o, const char *val) { double v= strtod(val,&ep); if (ep==val) badusage("bad time/duration value for %s",o->lng); - if (!*ep || !strcmp(ep,"s")) unit= 1; - else if (!strcmp(ep,"m")) unit= 60; - else if (!strcmp(ep,"h")) unit= 3600; - else if (!strcmp(ep,"d")) unit= 86400; + if (!*ep || !strcmp(ep,"s") || !strcmp(ep,"sec")) unit= 1; + else if (!strcmp(ep,"m") || !strcmp(ep,"min")) unit= 60; + else if (!strcmp(ep,"h") || !strcmp(ep,"hour")) unit= 3600; + else if (!strcmp(ep,"d") || !strcmp(ep,"day")) unit= 86400; + else if (!strcmp(ep,"das")) unit= 10; + else if (!strcmp(ep,"hs")) unit= 100; + else if (!strcmp(ep,"ks")) unit= 1000; + else if (!strcmp(ep,"Ms")) unit= 1000000; else badusage("bad units %s for time/duration value for %s",ep,o->lng); v *= unit; @@ -2699,19 +3641,25 @@ static void help(const Option *o, const char *val); static const Option innduct_options[]= { {'f',"feedfile", "F", &feedfile, op_string }, {'q',"quiet-multiple", 0, &quiet_multiple, op_setint, 1 }, -{0,"no-daemon", 0, &become_daemon, op_setint, 0 }, +{0,"no-daemon", 0, &interactive, op_setint, 1 }, +{0,"interactive", 0, &interactive, op_setint, 2 }, {0,"no-streaming", 0, &try_stream, op_setint, 0 }, -{0,"inndconf", "F", &inndconffile, op_string }, +{0,"no-filemon", 0, &try_filemon, op_setint, 0 }, +{'C',"inndconf", "F", &inndconffile, op_string }, {'P',"port", "PORT", &port, op_integer }, +{0,"chdir", "DIR", &path_run, op_string }, +{0,"cli", "DIR/|PATH", &path_cli, op_string }, {0,"help", 0, 0, help }, {0,"max-connections", "N", &max_connections, op_integer }, {0,"max-queue-per-conn", "N", &max_queue_per_conn, op_integer }, +{0,"max-queue-per-file", "N", &max_queue_per_ipf, op_integer }, {0,"feedfile-flush-size","BYTES", &target_max_feedfile_size, op_integer }, {0,"period-interval", "TIME", &period_seconds, op_seconds }, -{0,"connection-timeout", "TIME", &connection_setup_timeout, op_seconds }, -{0,"stuck-flush-timeout","TIME", &inndcomm_flush_timeout, op_seconds }, +{0,"connection-timeout", "TIME", &connection_setup_timeout, op_seconds }, +{0,"stuck-flush-timeout", "TIME", &inndcomm_flush_timeout, op_seconds }, +{0,"feedfile-poll", "TIME", &filepoll_seconds, op_seconds }, {0,"no-check-proportion", "PERCENT", &nocheck_thresh, op_double }, {0,"no-check-response-time","ARTICLES", &nocheck_decay, op_double }, @@ -2721,7 +3669,10 @@ static const Option innduct_options[]= { {0,"earliest-deferred-retry","PERIOD", &backlog_retry_minperiods, op_seconds }, {0,"backlog-rescan-interval","PERIOD",&backlog_spontrescan_periods,op_seconds}, {0,"max-flush-interval", "PERIOD", &spontaneous_flush_periods,op_seconds }, +{0,"flush-finish-timeout", "PERIOD", &max_separated_periods, op_seconds }, {0,"idle-timeout", "PERIOD", &need_activity_periods, op_seconds }, +{0,"low-volume-thresh", "PERIOD", &lowvol_thresh, op_integer }, +{0,"low-volume-window", "PERIOD", &lowvol_periods, op_seconds }, {0,"max-bad-input-data-ratio","PERCENT", &max_bad_data_ratio, op_double }, {0,"max-bad-input-data-init", "PERCENT", &max_bad_data_initial, op_integer }, @@ -2749,10 +3700,16 @@ static void convert_to_periods_rndup(int *store) { *store /= period_seconds; } +static int path_ends_slash(const char *specified) { + int l= strlen(specified); + assert(l); + return specified[l-1] == '/'; +} + int main(int argc, char **argv) { if (!argv[1]) { printusage(stderr); - exit(12); + exit(8); } parse_options(innduct_options, &argv); @@ -2761,11 +3718,17 @@ int main(int argc, char **argv) { sitename= *argv++; if (!sitename) badusage("need site name argument"); - remote_host= *argv++; + + if (*argv) remote_host= *argv++; + else remote_host= sitename; + if (*argv) badusage("too many non-option arguments"); /* defaults */ + int r= innconf_read(inndconffile); + if (!r) badusage("could not read inn.conf (more info on stderr)"); + if (!remote_host) remote_host= sitename; if (nocheck_thresh < 0 || nocheck_thresh > 100) @@ -2774,27 +3737,43 @@ int main(int argc, char **argv) { if (nocheck_decay < 0.1) badusage("nocheck decay articles must be at least 0.1"); - nocheck_decay= 1 - 1.0/nocheck_decay; + nocheck_decay= pow(0.5, 1.0/nocheck_decay); convert_to_periods_rndup(&reconnect_delay_periods); convert_to_periods_rndup(&flushfail_retry_periods); convert_to_periods_rndup(&backlog_retry_minperiods); convert_to_periods_rndup(&backlog_spontrescan_periods); convert_to_periods_rndup(&spontaneous_flush_periods); + convert_to_periods_rndup(&max_separated_periods); convert_to_periods_rndup(&need_activity_periods); + convert_to_periods_rndup(&lowvol_periods); if (max_bad_data_ratio < 0 || max_bad_data_ratio > 100) badusage("bad input data ratio must be between 0..100"); max_bad_data_ratio *= 0.01; - if (!feedfile) { - innconf_read(inndconffile); - feedfile= xasprintf("%s/%s",innconf->pathoutgoing,sitename); - } else if (!feedfile[0]) { - badusage("feed filename must be nonempty"); - } else if (feedfile[strlen(feedfile)-1]=='/') { - feedfile= xasprintf("%s%s",feedfile,sitename); + if (!path_run) + path_run= innconf->pathrun; + + if (!feedfile) feedfile= sitename; + if (!feedfile[0]) badusage("feed filename, if specified, must be nonempty"); + if (path_ends_slash(feedfile)) + feedfile= xasprintf("%s%s", feedfile, sitename); + if (feedfile[0] != '/') + feedfile= xasprintf("%s/%s", innconf->pathoutgoing, feedfile); + + if (!path_cli) { + path_cli_dir= "innduct"; + } else if (!path_cli[0] || !strcmp(path_cli,"none")) { + path_cli= 0; /* ok, don't then */ + } else if (path_ends_slash(path_cli)) { + path_cli_dir= xasprintf("%.*s", strlen(path_cli)-1, path_cli); } + if (path_cli_dir) + path_cli= xasprintf("%s/%s", path_cli_dir, sitename); + + if (max_queue_per_ipf<0) + max_queue_per_ipf= max_queue_per_conn * 2; const char *feedfile_forbidden= "?*[~#"; int c; @@ -2802,27 +3781,41 @@ int main(int argc, char **argv) { if (strchr(feedfile, c)) badusage("feed filename may not contain metacharacter %c",c); + int i; + lowvol_perperiod= xcalloc(sizeof(*lowvol_perperiod), lowvol_periods); + for (i=0; i= 2) + cli_stdio(); + + cli_init(); + + int filemon_ok= 0; + if (!try_filemon) { + notice("filemon: suppressed by command line option, polling"); + } else { + filemon_ok= filemon_method_init(); + if (!filemon_ok) + warn("filemon: no file monitoring available, polling"); } + if (!filemon_ok) + every(filepoll_seconds,0,filepoll); - period_timeout.tv_sec= period_seconds; - period_schedule(); + every(period_seconds,1,period); statemc_init(); /* let's go */ - void *r= oop_sys_run(sysloop); - assert(r == OOP_ERROR); + void *run= oop_sys_run(sysloop); + assert(run == OOP_ERROR); sysdie("event loop failed"); }