X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ian/git?p=inn-innduct.git;a=blobdiff_plain;f=backends%2Finnduct.c;h=8e599f56026835595780ad3a28725754699fd06f;hp=f0cbf3496c830ad177967cbfb83fa2252359dfb0;hb=fc66b511debe93980fd49331d96939fb876024da;hpb=6b1d9ed0da80f22b6ff1c3c678fff5a2ce492fa2;ds=sidebyside diff --git a/backends/innduct.c b/backends/innduct.c index f0cbf34..8e599f5 100644 --- a/backends/innduct.c +++ b/backends/innduct.c @@ -1,3 +1,12 @@ +/* + * TODO + * - actually implement badusage + * - options for all options + * - manpage + * - pid, sitename, hostname in lockfile + * - -k kill mode ? + */ + /* * Newsfeeds file entries should look like this: * host.name.of.site[/exclude,exclude,...]\ @@ -56,33 +65,20 @@ * 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 | + START + | + ,-->--. check F, D + | | | + | | | + | | <----------------<---------------------------------'| + | | F exists | + | | D ENOENT | + | | duct opens F | | V | | Normal | | F: innd writing, duct reading | @@ -96,31 +92,35 @@ | 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 / @@ -133,201 +133,453 @@ | V duct unlinks D D: duct reading | | | `--<--' | duct finishes - | processing D - | duct unlinks D - | duct exits - V - Dropped - F: ENOENT - D: ENOENT - duct not running + | processing D + | duct unlinks D + | duct exits + V + Dropped + F: ENOENT + D: ENOENT + duct not running "duct reading" means innduct is reading the file but also overwriting processed tokens. + * ->8- -^L- * + * rune for printing diagrams: + +perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct.c |a2ps -R -B -ops + * */ +/*============================== PROGRAM ==============================*/ + +#define _GNU_SOURCE 1 + +#include "config.h" +#include "storage.h" +#include "nntp.h" +#include "libinn.h" +#include "inndcomm.h" + +#include "inn/list.h" +#include "inn/innconf.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include /*----- general definitions, probably best not changed -----*/ -#define PERIOD_SECONDS 60 - #define CONNCHILD_ESTATUS_STREAM 4 #define CONNCHILD_ESTATUS_NOSTREAM 5 #define INNDCOMMCHILD_ESTATUS_FAIL 6 #define INNDCOMMCHILD_ESTATUS_NONESUCH 7 +#define MAX_LINE_FEEDFILE (NNTP_MSGID_MAXLEN + sizeof(TOKEN)*2 + 10) -/*----- configuration options -----*/ +#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))) -static char *sitename, *feedfile; -static int max_connections=10, max_queue_per_conn=200; -static int connection_setup_timeout=200, port=119, try_stream=1; -static int inndcomm_flush_timeout=100, quiet_if_locked=0; -static const char *remote_host; -static int reconnect_delay_periods, flushfail_retry_periods, open_wait_periods; -static int backlog_retry_minperiods, backlog_spontaneous_rescan_periods; -static const char *inndconffile; +/*----- doubly linked lists -----*/ -static double accept_proportion; -static double nocheck_thresh_pct= 95.0; -static double nocheck_thresh; /* computed in main from _pct */ -static double nocheck_decay_articles= 100; /* converted to _decay */ -static double nocheck_decay; /* computed in main from _articles */ -static int nocheck, nocheck_reported; +#define ISNODE(T) struct node list_node +#define DEFLIST(T) \ + typedef struct { \ + union { struct list li; T *for_type; } u; \ + int count; \ + } T##List +#define NODE(n) (assert((void*)&(n)->list_node == &(n)), &(n)->list_node) -/*----- doubly linked lists -----*/ +#define LIST_CHECKCANHAVENODE(l,n) \ + ((void)((n) == ((l).u.for_type))) /* just for the type check */ + +#define LIST_ADDSOMEHOW(l,n,list_addsomehow) \ + ( LIST_CHECKCANHAVENODE(l,n), \ + list_addsomehow(&(l).u.li, NODE((n))), \ + (void)(l).count++ \ + ) -#define ISNODE(T) T *next, *back; -#define LIST(T) struct { T *head, *tail, *tailpred; int count; } +#define LIST_REMSOMEHOW(l,list_remsomehow) \ + ( (typeof((l).u.for_type)) \ + ( (l).count \ + ? ( (l).count--, \ + list_remsomehow(&(l).u.li) ) \ + : 0 \ + ) \ + ) -#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_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_INIT(l) (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)))) -#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++) + ( 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); + +#define SMS(newstate, periods, why) \ + (statemc_setstate(sm_##newstate,(periods),#newstate,(why))) +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_assign_articles(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(void); + +static void open_defer(void); +static void close_defer(void); +static void search_backlog_file(void); + +static void inputfile_reading_start(InputFile *ipf); +static void inputfile_reading_stop(InputFile *ipf); + +static void filemon_start(InputFile *ipf); +static void filemon_stop(InputFile *ipf); +static void filemon_callback(InputFile *ipf); + +static void vconnfail(Conn *conn, const char *fmt, va_list al) PRINTF(2,0); +static void connfail(Conn *conn, const char *fmt, ...) PRINTF(2,3); + +static const oop_rd_style peer_rd_style; +static oop_rd_call peer_rd_err, peer_rd_ok; + +/*----- configuration options -----*/ + +static const char *sitename, *remote_host; +static const char *feedfile; +static int quiet_multiple=0; +static int become_daemon=1; +static int try_stream=1; +static int port=119; +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 connection_setup_timeout=200; +static int inndcomm_flush_timeout=100; + +static double nocheck_thresh= 95.0; /* converted from percentage by main */ +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_spontrescan_periods=300; +static int spontaneous_flush_periods=100000; +static int need_activity_periods=1000; + +static double max_bad_data_ratio= 1; /* conv'd from percentage by main */ +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 */ /*----- 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; + 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; - 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; /* may be 0, meaning closed (during construction/destruction) */ + oop_read *rd; /* likewise */ + int max_queue, stream, quitting; + int since_activity; /* periods */ + ArticleList waiting; /* not yet told peer */ + ArticleList priority; /* peer says send it now */ + ArticleList sent; /* offered/transmitted - in xmit or waiting reply */ struct iovec xmit[CONNIOVS]; XmitDetails xmitd[CONNIOVS]; int xmitu; }; -/*----- operational variables -----*/ +/*----- general operational variables -----*/ -static int since_connect_attempt; -static int nconns; -static LIST(Conn) idle, working, full; -static LIST(Article) *queue; - -static char *path_lock, *path_flushing, *path_defer; - -#define SMS(newstate, periods, why) \ - (statemc_setstate(sm_##newstate,(periods),#newstate,(why))) +/* main initialises */ +static oop_source *loop; +static ConnList conns; +static ArticleList queue; +static char *path_lock, *path_flushing, *path_defer, *globpat_backlog; +/* statemc_init initialises */ 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; +/* initialisation to 0 is good */ +static int until_connect, until_backlog_nextscan; +static double accept_proportion; +static int nocheck, nocheck_reported; -/*----- function predeclarations -----*/ +/*========== logging ==========*/ -static void conn_check_work(Conn *conn); +static void logcore(int sysloglevel, const char *fmt, ...) PRINTF(2,3); +static void logcore(int sysloglevel, const char *fmt, ...) { + VA; + if (become_daemon) { + vsyslog(sysloglevel,fmt,al); + } else { + vfprintf(stderr,fmt,al); + putc('\n',stderr); + } + va_end(al); +} -static int filemon_init(void); -static void filemon_setfile(int mainfeed_fd, const char *mainfeed_path); -static void filemon_callback(void); +static void logv(int sysloglevel, const char *pfx, int errnoval, + const char *fmt, va_list al) PRINTF(5,0); +static void logv(int sysloglevel, const char *pfx, int errnoval, + const char *fmt, va_list al) { + char msgbuf[256]; /* NB do not call xvasprintf here or you'll recurse */ + vsnprintf(msgbuf,sizeof(msgbuf), fmt,al); + msgbuf[sizeof(msgbuf)-1]= 0; + + if (sysloglevel >= LOG_ERR && (errnoval==EACCES || errnoval==EPERM)) + sysloglevel= LOG_ERR; /* run by wrong user, probably */ + + logcore(sysloglevel, "<%s>%s: %s%s%s", + sitename, pfx, msgbuf, + errnoval>=0 ? ": " : "", + errnoval>=0 ? strerror(errnoval) : ""); +} + +#define diewrap(fn, pfx, sysloglevel, err, estatus) \ + static void fn(const char *fmt, ...) NORET_PRINTF(1,2); \ + static void fn(const char *fmt, ...) { \ + VA; \ + logv(sysloglevel, pfx, err, fmt, al); \ + exit(estatus); \ + } + +#define logwrap(fn, pfx, sysloglevel, err) \ + static void fn(const char *fmt, ...) PRINTF(1,2); \ + static void fn(const char *fmt, ...) { \ + VA; \ + logv(sysloglevel, pfx, err, fmt, al); \ + va_end(al); \ + } + +diewrap(sysdie, " critical", LOG_CRIT, errno, 16); +diewrap(die, " critical", LOG_CRIT, -1, 16); + +diewrap(sysfatal, " fatal", LOG_ERR, errno, 12); +diewrap(fatal, " fatal", LOG_ERR, -1, 12); + +logwrap(syswarn, " warning", LOG_WARNING, errno); +logwrap(warn, " warning", LOG_WARNING, -1); + +logwrap(notice, "", LOG_NOTICE, -1); +logwrap(info, " info", LOG_INFO, -1); +logwrap(debug, " debug", LOG_DEBUG, -1); -static void statemc_setstate(StateMachineState newsms, int periods, - const char *forlog, const char *why); /*========== utility functions etc. ==========*/ -static void perhaps_close(int *fd) { if (*fd) { close(*fd); fd=0; } } +static char *xvasprintf(const char *fmt, va_list al) PRINTF(1,0); +static char *xvasprintf(const char *fmt, va_list al) { + char *str; + int rc= vasprintf(&str,fmt,al); + if (rc<0) sysdie("vasprintf(\"%s\",...) failed", fmt); + return str; +} +static char *xasprintf(const char *fmt, ...) PRINTF(1,2); +static char *xasprintf(const char *fmt, ...) { + VA; + char *str= xvasprintf(fmt,al); + va_end(al); + return str; +} + +static int close_perhaps(int *fd) { + if (!*fd) return 0; + int r= close(*fd); + *fd=0; + return r; +} +static void xclose(int fd, const char *what, const char *what2) { + int r= close(fd); + if (r) sysdie("close %s%s",what,what2?what2:""); +} +static void xclose_perhaps(int *fd, const char *what, const char *what2) { + if (!*fd) return; + xclose(*fd,what,what2); + *fd=0; +} static pid_t xfork(const char *what) { pid_t child; child= fork(); if (child==-1) sysdie("cannot fork for %s",what); - if (!child) postfork(what); + debug("forked %s %ld", what, (unsigned long)child); + if (!child) postfork(); return child; } @@ -344,7 +596,7 @@ static void report_child_status(const char *what, int status) { if (WIFEXITED(status)) { int es= WEXITSTATUS(status); if (es) - warn("%s: child died with error exit status %d",es); + warn("%s: child died with error exit status %d", what, es); } else if (WIFSIGNALED(status)) { int sig= WTERMSIG(status); const char *sigstr= strsignal(sig); @@ -355,14 +607,14 @@ static void report_child_status(const char *what, int status) { warn("%s: child died due to unknown fatal signal %d%s", what, sig, coredump); } else { - warn("%s: child died with unknown wait status %d", status); + warn("%s: child died with unknown wait status %d", what,status); } } static int xwaitpid(pid_t *pid, const char *what) { int status; - r= kill(*pid, SIGKILL); + int r= kill(*pid, SIGKILL); if (r) sysdie("cannot kill %s child", what); pid_t got= waitpid(*pid, &status, WNOHANG); @@ -373,6 +625,17 @@ static int xwaitpid(pid_t *pid, const char *what) { 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 time_t xtime(void) { + time_t now= time(0); + if (now==-1) sysdie("time(2) failed"); + return now; +} + static void check_isreg(const struct stat *stab, const char *path, const char *what) { if (!S_ISREG(stab->st_mode)) @@ -381,17 +644,18 @@ static void check_isreg(const struct stat *stab, const char *path, } 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) { @@ -409,80 +673,163 @@ static int samefile(const struct stat *a, const struct stat *b) { 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; +static int isewouldblock(int errnoval) { + return errnoval==EWOULDBLOCK || errnoval==EAGAIN; +} - if (sysloglevel >= LOG_ERR && (errnoval==EACCES || errnoval==EPERM)) - sysloglevel= LOG_ERR; /* run by wrong user, probably */ +/*========== management of connections ==========*/ - logcore(sysloglevel, "<%s>%s: %s%s%s", - sitename, pfx, msgbuf, - errnoval>=0 ? ": " : "", - errnoval>=0 ? strerror(errnoval) : ""); +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)); } -#define logwrap(fn, pfx, sysloglevel, errno, estatus) \ - static void fn(const char *fmt, ...) \ - __attribute__((printf,1,2)); \ - static void fn(const char *fmt, ...) { \ - va_list al; \ - va_start(al,fmt); \ - logv(sysloglevel, pfx, errno, estatus, fmt, al); \ +static void conn_dispose(Conn *conn) { + if (!conn) return; + if (conn->rd) { + oop_rd_cancel(conn->rd); + oop_rd_delete_kill(conn->rd); + conn->rd= 0; } + if (conn->fd) { + loop->cancel_fd(loop, conn->fd, OOP_WRITE); + loop->cancel_fd(loop, conn->fd, OOP_EXCEPTION); + } + conn_closefd(conn,""); + free(conn); + until_connect= reconnect_delay_periods; +} -logwrap(sysdie, " critical", LOG_CRIT, errno, 16); -logwrap(die, " critical", LOG_CRIT, -1, 16); +static void *conn_exception(oop_source *lp, int fd, + oop_event ev, void *conn_v) { + Conn *conn= conn_v; + unsigned char ch; + assert(fd == conn->fd); + assert(ev == OOP_EXCEPTION); + int r= read(conn->fd, &ch, 1); + if (r<0) connfail(conn,"read failed: %s",strerror(errno)); + else connfail(conn,"exceptional condition on socket (peer sent urgent" + " data? read(,&ch,1)=%d,ch='\\x%02x')",r,ch); + return OOP_CONTINUE; +} -logwrap(sysfatal, " fatal", LOG_ERR, errno, 12); -logwrap(fatal, " fatal", LOG_ERR, -1, 12); +static void vconnfail(Conn *conn, const char *fmt, va_list al) { + int requeue[art_MaxState]; -logwrap(syswarn, " warning", LOG_WARN, errno, 0); -logwrap(warn, " warning", LOG_WARN, -1, 0); + 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); + } -logwrap(notice, "", LOG_NOTICE, -1, 0); -logwrap(info, " info", LOG_INFO, -1, 0); + int i; + XmitDetails *d; + for (i=0, d=conn->xmitd; ixmitu; 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); -/*========== making new connections ==========*/ + LIST_REMOVE(conns,conn); + conn_dispose(conn); + check_assign_articles(); +} -static int connecting_sockets[2]= {-1,-1}; -static pid_t connecting_child; +static void connfail(Conn *conn, const char *fmt, ...) { + va_list al; + va_start(al,fmt); + vconnfail(conn,fmt,al); + va_end(al); +} + +static void check_idle_conns(void) { + Conn *conn; + for (conn=LIST_HEAD(conns); conn; conn=LIST_NEXT(conn)) + conn->since_activity++; + search_again: + for (conn=LIST_HEAD(conns); conn; conn=LIST_NEXT(conn)) { + if (conn->since_activity <= need_activity_periods) continue; + + /* We need to shut this down */ + if (conn->quitting) + connfail(conn,"timed out waiting for response to QUIT"); + else if (conn->sent.count) + connfail(conn,"timed out waiting for responses"); + else if (conn->waiting.count || conn->priority.count) + connfail(conn,"BUG IN INNDUCT conn has queue but nothing sent"); + else if (conn->xmitu) + connfail(conn,"peer has been sending responses" + " before receiving our commands!"); + else { + static const char quitcmd[]= "QUIT\r\n"; + int todo= sizeof(quitcmd)-1; + const char *p= quitcmd; + for (;;) { + int r= write(conn->fd, p, todo); + if (r<0) { + if (isewouldblock(errno)) + connfail(conn, "blocked writing QUIT to idle connection"); + else + connfail(conn, "failed to write QUIT to idle connection: %s", + strerror(errno)); + break; + } + assert(r<=todo); + todo -= r; + if (!todo) { + conn->quitting= 1; + conn->since_activity= 0; + debug("C%d is idle, quitting", conn->fd); + break; + } + } + } + goto search_again; + } +} -static void connect_attempt_discard(void) { - if (connecting_sockets[0]) - cancel_fd(connecting_sockets[0]); +/*---------- making new connections ----------*/ - perhaps_close(&connecting_sockets[0]); - perhaps_close(&connecting_sockets[1]); +static pid_t connecting_child; +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); } + if (connecting_fdpass_sock) { + cancel_fd_read_except(connecting_fdpass_sock); + xclose_perhaps(&connecting_fdpass_sock, "connecting fdpass socket",0); + } } #define PREP_DECL_MSG_CMSG(msg) \ @@ -495,10 +842,12 @@ static void connect_attempt_discard(void) { static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) { Conn *conn= 0; + assert(fd == connecting_fdpass_sock); + 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); @@ -508,43 +857,49 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) { if (got != -1) { assert(got==connecting_child); connecting_child= 0; - if (WIFEXITED(status) && - (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) { + 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", + WEXITSTATUS(status)); + } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) { warn("connect: connection attempt timed out"); - } else if (!WIFEXITED(status)) { + } else { report_child_status("connect", status); - /* that's probably the root cause then */ } } else { /* child is still running apparently, report the socket problem */ if (rs < 0) - syswarn("connect: read from child socket failed"); - else if (e == OOP_EXCEPTIONN) - warn("connect: unexpected exception on child socket"); + 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 - warn("connect: unexpected EOF on child socket"); + fatal("connect: unexpected lack of cmsg from child"); } goto x; } -#define CHK(field, val) \ - if (h->cmsg_##field != val) { \ - die("connect: child sent cmsg with cmsg_" #field "=%d, expected %d"); \ - goto x; \ +#define CHK(field, val) \ + if (h->cmsg_##field != val) { \ + die("connect: child sent cmsg with cmsg_" #field "=%d, expected %d", \ + h->cmsg_##field, val); \ + goto x; \ } 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"); 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); @@ -559,32 +914,48 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) { fatal("connect: child gave unexpected exit status %d", es); } - set nonblocking; - /* Phew! */ - LIST_ADDHEAD(idle, conn); - notice(CN "connected %s", conn->fd, conn->stream ? "streaming" : "plain"); + 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); + 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); + connect_attempt_discard(); - check_master_queue(); + check_assign_articles(); return 0; x: - if (conn) { - perhaps_close(&conn->fd); - free(conn); - } + conn_dispose(conn); connect_attempt_discard(); + return OOP_CONTINUE; +} + +static int allow_connect_start(void) { + return conns.count < max_connections + && !connecting_child + && !until_connect; } -static void connect_start() { - assert(!connecting_sockets[0]); - assert(!connecting_sockets[1]); +static void connect_start(void) { assert(!connecting_child); + assert(!connecting_fdpass_sock); notice("starting connection attempt"); - r= socketpair(AF_UNIX, SOCK_STREAM, 0, connecting_sockets); - if (r) { syswarn("connect: cannot create socketpair for child"); goto x; } + int socks[2]; + int r= socketpair(AF_UNIX, SOCK_STREAM, 0, socks); + if (r) { syswarn("connect: cannot create socketpair for child"); return; } connecting_child= xfork("connection"); @@ -593,19 +964,14 @@ static void connect_start() { char buf[NNTP_STRLEN+100]; int exitstatus= CONNCHILD_ESTATUS_NOSTREAM; - r= close(connecting_sockets[0]); - if (r) sysdie("connect: close parent socket in child"); + xclose(socks[0], "(in child) parent's connection fdpass socket",0); 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) || @@ -620,19 +986,16 @@ static void connect_start() { } int l= strlen(buf); assert(l>=1); - if (buf[-1]!='\n') { - sanitise_inplace(buf); + if (buf[-1]!='\n') fatal("connect: response to MODE STREAM is too long: %.100s...", - remote_host, buf); - } - l--; if (l>0 && buf[1-]=='\r') l--; + 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; @@ -641,8 +1004,8 @@ static void connect_start() { case 500: break; default: - sanitise_inplace(buf); - warn("connect: unexpected response to MODE STREAM: %.50s", buf); + warn("connect: unexpected response to MODE STREAM: %.50s", + sanitise(buf)); exitstatus= 2; break; } @@ -657,82 +1020,67 @@ static void connect_start() { memcpy(CMSG_DATA(cmsg), &fd, sizeof(fd)); msg.msg_controllen= cmsg->cmsg_len; - r= sendmsg(connecting_sockets[1], &msg, 0); + r= sendmsg(socks[1], &msg, 0); if (r) sysdie("sendmsg failed for new connection"); _exit(exitstatus); } - r= close(connecting_sockets[1]); connecting_sockets[1]= 0; - if (r) sysdie("connect: close child socket in parent"); - - on_fd_read_except(connecting_sockets[0], connchild_event); - return OOP_CONTINUE; - - x: - connect_attempt_discard(); + xclose(socks[1], "connecting fdpass child's socket",0); + connecting_fdpass_sock= socks[0]; + on_fd_read_except(connecting_fdpass_sock, connchild_event); } +/*---------- assigning articles to conns, and transmitting ----------*/ -/*========== overall control of article flow ==========*/ - -static void check_master_queue(void) { - if (!queue.count) - return; - - Conn *last_assigned=0; +static void check_assign_articles(void) { 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=0, inqueue=0; + + /* Find a connection to offer this article. We prefer a busy + * connection to an idle one, provided it's not full. We take the + * first (oldest) and since that's stable, it will mean we fill up + * connections in order. That way if we have too many + * connections, the spare ones will go away eventually. + */ + for (walk=LIST_HEAD(conns); walk; walk=LIST_NEXT(walk)) { + if (walk->quitting) continue; + inqueue= walk->sent.count + walk->priority.count + + walk->waiting.count; + spare= walk->max_queue - inqueue; + assert(inqueue <= max_queue_per_conn); + assert(spare >= 0); + if (inqueue==0) /*idle*/ { if (!use) use= walk; } + else if (spare>0) /*working*/ { use= walk; break; } + } + if (use) { + if (!inqueue) use->since_activity= 0; /* reset idle counter */ + while (spare>0) { + Article *art= LIST_REMHEAD(queue); + LIST_ADDTAIL(use->waiting, art); + spare--; + } + conn_maybe_write(use); + } else if (allow_connect_start()) { + until_connect= reconnect_delay_periods; connect_start(); + break; } else { break; } } - conn_check_work(last_assigned); -} - -static void conn_assign_one_article(LIST(Conn) *connlist, - Conn **last_assigned) { - Conn *conn= connlist->head; - - LIST_REMOVE(*connlist, conn); - Article *art= LIST_REMHEAD(queue); - LIST_ADDTAIL(conn->queue, art); - LIST_ADD(*conn_determine_right_list(conn), conn); - - /* This slightly odd arrangement is so that we call conn_check_work - * once after filling the queue for a new connection in - * check_master_queue, rather than for each article. */ - if (conn != *last_assigned && *last_assigned) - conn_check_work(*last_assigned); - *last_assigned= conn; -} - -static int conn_total_queued_articles(Conn *conn) { - return conn->sent.count + conn->queue.count; } -static LIST(Conn) *conn_determine_right_list(Conn *conn) { - int inqueue= conn_total_queued_articles(conn); - assert(inqueue <= max_queue); - if (inqueue == 0) return &idle; - if (inqueue == conn->max_queue) return &full; - return &working; -} - -static void *conn_writeable(oop_source *l, int fd, int ev, void *u) { - check_conn_work(u); +static void *conn_writeable(oop_source *l, int fd, oop_event ev, void *u) { + conn_maybe_write(u); return OOP_CONTINUE; } -static void conn_check_work(Conn *conn) { - void *rp= 0; +static void conn_maybe_write(Conn *conn) { for (;;) { conn_make_some_xmits(conn); if (!conn->xmitu) { @@ -754,14 +1102,13 @@ static void conn_check_work(Conn *conn) { } } - /*========== article transmission ==========*/ static XmitDetails *xmit_core(Conn *conn, const char *data, int len, - XmitKind kind) { /* caller must then fill in details */ + 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; @@ -773,7 +1120,7 @@ static void xmit_noalloc(Conn *conn, const char *data, int len) { #define XMIT_LITERAL(lit) (xmit_noalloc(conn, (lit), sizeof(lit)-1)) static void xmit_artbody(Conn *conn, ARTHANDLE *ah /* consumed */) { - XmitDetails *d= xmit_core(conn, ah->data, ah->len, sk_Artdata); + XmitDetails *d= xmit_core(conn, ah->data, ah->len, xk_Artdata); d->info.sm_art= ah; } @@ -799,20 +1146,21 @@ static void *conn_write_some_xmits(Conn *conn) { if (count > IOV_MAX) count= IOV_MAX; ssize_t rs= writev(conn->fd, conn->xmit, count); if (rs < 0) { - if (errno == EAGAIN) return OOP_CONTINUE; + if (isewouldblock(errno)) return OOP_CONTINUE; connfail(conn, "write failed: %s", strerror(errno)); return OOP_HALT; } assert(rs > 0); - for (done=0; rs && donexmitu; done++) { struct iovec *vp= &conn->xmit[done]; XmitDetails *dp= &conn->xmitd[done]; if (rs > vp->iov_len) { rs -= vp->iov_len; xmit_free(dp); } else { - vp->iov_base += rs; + vp->iov_base= (char*)vp->iov_base + rs; vp->iov_len -= rs; } } @@ -828,18 +1176,19 @@ static void conn_make_some_xmits(Conn *conn) { if (conn->xmitu+5 > CONNIOVS) break; - Article *art= LIST_REMHEAD(queue); + Article *art= LIST_REMHEAD(conn->priority); + if (!art) art= LIST_REMHEAD(conn->waiting); if (!art) break; - if (art->checked || (conn->stream && nocheck)) { + if (art->state >= art_Wanted || (conn->stream && nocheck)) { /* actually send it */ - ARTHANDLE *artdata= SMretrieve(); + ARTHANDLE *artdata= SMretrieve(art->token, RETR_ALL); if (conn->stream) { if (artdata) { XMIT_LITERAL("TAKETHIS "); - xmit_noalloc(conn, art->mid, art->midlen); + xmit_noalloc(conn, art->messageid, art->midlen); XMIT_LITERAL("\r\n"); xmit_artbody(conn, artdata); } @@ -852,11 +1201,13 @@ static void conn_make_some_xmits(Conn *conn) { } } - art->sent= 1; + art->state= + art->state == art_Unchecked ? art_Unsolicited : + art->state == art_Wanted ? art_Wanted : + (abort(),-1); + art->ipf->counts[art->state][RC_sent]++; LIST_ADDTAIL(conn->sent, art); - art->ipf->counts[art->checked].sent++; - } else { /* check it */ @@ -864,11 +1215,12 @@ static void conn_make_some_xmits(Conn *conn) { XMIT_LITERAL("IHAVE "); else XMIT_LITERAL("CHECK "); - xmit_noalloc(art->mid, art->midlen); + xmit_noalloc(conn, art->messageid, art->midlen); XMIT_LITERAL("\r\n"); + assert(art->state == art_Unchecked); + art->ipf->counts[art->state][RC_sent]++; LIST_ADDTAIL(conn->sent, art); - art->ipf->counts[art->checked].offered++; } } } @@ -882,7 +1234,7 @@ static const oop_rd_style peer_rd_style= { OOP_RD_SHORTREC_FORBID }; -static void *peer_rd_err(oop_source *lp, oop_read *oread, oop_event ev, +static void *peer_rd_err(oop_source *lp, oop_read *oread, oop_rd_event ev, const char *errmsg, int errnoval, const char *data, size_t recsz, void *conn_v) { Conn *conn= conn_v; @@ -890,10 +1242,12 @@ static void *peer_rd_err(oop_source *lp, oop_read *oread, oop_event ev, 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, @@ -905,20 +1259,20 @@ static Article *article_reply_check(Connection *conn, const char *response, if (code_indicates_streaming) { assert(!memchr(response, 0, 4)); /* ensured by peer_rd_ok */ if (!conn->stream) { - connfail("peer gave streaming response code " + connfail(conn, "peer gave streaming response code " " to IHAVE or subsequent body: %s", sanitised_response); return 0; } const char *got_mid= response+4; int got_midlen= strcspn(got_mid, " \n\r"); if (got_midlen<3 || got_mid[0]!='<' || got_mid[got_midlen-1]!='>') { - connfail("peer gave streaming response with syntactically invalid" + connfail(conn, "peer gave streaming response with syntactically invalid" " messageid: %s", sanitised_response); return 0; } if (got_midlen != art->midlen || memcmp(got_mid, art->messageid, got_midlen)) { - connfail("peer gave streaming response code to wrong article -" + connfail(conn, "peer gave streaming response code to wrong article -" " probable synchronisation problem; we offered: %s;" " peer said: %s", art->messageid, sanitised_response); @@ -926,32 +1280,48 @@ static Article *article_reply_check(Connection *conn, const char *response, } } else { if (conn->stream) { - connfail("peer gave non-streaming response code to CHECK/TAKETHIS: %s", - sanitised_response); + connfail(conn, "peer gave non-streaming response code to" + " CHECK/TAKETHIS: %s", sanitised_response); return 0; } } + if (must_have_sent>0 && art->state < art_Wanted) { + connfail(conn, "peer says article accepted but" + " we had not sent the body: %s", sanitised_response); + return 0; + } + if (must_have_sent<0 && art->state >= art_Wanted) { + connfail(conn, "peer says please sent the article but we just did: %s", + sanitised_response); + return 0; + } + + Article *art_again= LIST_REMHEAD(conn->sent); + assert(art_again == art); return art; } static void update_nocheck(int accepted) { - accept_proportion *= accept_decay; - accept_proportion += accepted; - nocheck= accept_proportion >= nocheck_thresh; - if (nocheck && !nocheck_reported) { + accept_proportion *= nocheck_decay; + accept_proportion += accepted * (1.0 - nocheck_decay); + int new_nocheck= accept_proportion >= nocheck_thresh; + if (new_nocheck && !nocheck_reported) { notice("entering nocheck mode for the first time"); nocheck_reported= 1; + } else if (new_nocheck != nocheck) { + debug("nocheck mode %s", new_nocheck ? "start" : "stop"); } + nocheck= new_nocheck; } -static void article_done(Connection *conn, Article *art, int whichcount) { - *count++; - 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); InputFile *ipf= art->ipf; + while (art->blanklen) { static const char spaces[]= " " @@ -963,7 +1333,8 @@ static void article_done(Connection *conn, Article *art, int whichcount) { if (r==-1) { if (errno==EINTR) continue; sysdie("failed to blank entry for %s (length %d at offset %lu) in %s", - art->messageid, art->blanklen, art->offset, ipf->path); + art->messageid, art->blanklen, + (unsigned long)art->offset, ipf->path); } assert(r>=0 && r<=w); art->blanklen -= w; @@ -972,14 +1343,13 @@ static void article_done(Connection *conn, Article *art, int whichcount) { ipf->inprogress--; assert(ipf->inprogress >= 0); - - if (!ipf->inprogress) - loop->on_time(loop, OOP_TIME_NOW, statemc_check_input_done, ipf); - free(art); + + if (!ipf->inprogress && ipf != main_input_file) + queue_check_input_done(); } -static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev, +static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_rd_event ev, const char *errmsg, int errnoval, const char *data, size_t recsz, void *conn_v) { Conn *conn= conn_v; @@ -990,43 +1360,40 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev, } assert(ev == OOP_RD_OK); + char *sani= sanitise(data); + char *ep; unsigned long code= strtoul(data, &ep, 10); if (ep != data+3 || *ep != ' ' || data[0]=='0') { - char sanibuf[100]; - const char *p= data; - char *q= sanibuf; - *q++= '`'; - for (;;) { - if (q > sanibuf+sizeof(sanibuf)-8) { strcpy(q,"..."); break; } - int c= *p++; - if (!c) { *q++= '\''; break; } - if (c>=' ' && c<=126 && c!='\\') { *q++= c; continue; } - sprintf(q,"\\x%02x",c); - q += 4; - } - connfail(conn, "badly formatted response from peer: %s", sanibuf); + connfail(conn, "badly formatted response from peer: %s", sani); return OOP_CONTINUE; } if (conn->quitting) { - if (code!=205) { - connfail(conn, "peer gave failure response to QUIT: %s", sani); - return OOP_CONTINUE; + if (code!=205 && code!=503) { + connfail(conn, "peer gave unexpected response to QUIT: %s", sani); + } else { + notice("C%d idle connection closed", conn->fd); + assert(!conn->waiting.count); + assert(!conn->priority.count); + assert(!conn->sent.count); + assert(!conn->xmitu); + LIST_REMOVE(conns,conn); + conn_dispose(conn); } - conn close ok; - return; + return OOP_CONTINUE; } + conn->since_activity= 0; Article *art; -#define GET_ARTICLE \ - art= article_reply_check(conn, data, code_streaming, sani); \ +#define GET_ARTICLE(musthavesent) \ + art= article_reply_check(conn, data, musthavesent, code_streaming, sani); \ if (art) ; else return OOP_CONTINUE /* reply_check has failed the conn */ -#define ARTICLE_DEALTWITH(streaming,how) \ - code_streaming= (streaming) \ - GET_ARTICLE; \ +#define ARTICLE_DEALTWITH(streaming,musthavesent,how) \ + code_streaming= (streaming); \ + GET_ARTICLE(musthavesent); \ article_done(conn, art, RC_##how); break; #define PEERBADMSG(m) connfail(conn, m ": %s", sani); return OOP_CONTINUE @@ -1039,32 +1406,29 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev, case 503: PEERBADMSG("peer timed us out"); default: PEERBADMSG("peer sent unexpected message"); - case 435: ARTICLE_DEALTWITH(0,unwanted); /* IHAVE says they have it */ - case 438: ARTICLE_DEALTWITH(1,unwanted); /* CHECK/TAKETHIS: they have it */ + case 435: ARTICLE_DEALTWITH(0,0,unwanted); /* IHAVE says they have it */ + case 438: ARTICLE_DEALTWITH(1,0,unwanted); /* CHECK/TAKETHIS: they have it */ - case 235: ARTICLE_DEALTWITH(0,accepted); /* IHAVE says thanks */ - case 239: ARTICLE_DEALTWITH(1,accepted); /* TAKETHIS says thanks */ + case 235: ARTICLE_DEALTWITH(0,1,accepted); /* IHAVE says thanks */ + case 239: ARTICLE_DEALTWITH(1,1,accepted); /* TAKETHIS says thanks */ - case 437: ARTICLE_DEALTWITH(0,rejected); /* IHAVE says rejected */ - case 439: ARTICLE_DEALTWITH(1,rejected); /* TAKETHIS says rejected */ + case 437: ARTICLE_DEALTWITH(0,0,rejected); /* IHAVE says rejected */ + case 439: ARTICLE_DEALTWITH(1,0,rejected); /* TAKETHIS says rejected */ case 238: /* CHECK says send it */ code_streaming= 1; case 335: /* IHAVE says send it */ - GET_ARTICLE; - count_checkedwanted++; - LIST_ADDTAIL(conn->queue); - if (art->checked) { - connfail("peer gave %d response to article body: %s",code, sani); - return OOP_CONTINUE; - } - art->checked= 1; + GET_ARTICLE(-1); + assert(art->state == art_Unchecked); + art->ipf->counts[art->state][RC_accepted]++; + art->state= art_Wanted; + LIST_ADDTAIL(conn->priority, art); break; case 431: /* CHECK or TAKETHIS says try later */ code_streaming= 1; case 436: /* IHAVE says try later */ - GET_ARTICLE; + GET_ARTICLE(0); open_defer(); if (fprintf(defer, "%s %s\n", TokenToText(art->token), art->messageid) <0 || fflush(defer)) @@ -1074,7 +1438,8 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev, } - check_check_work(conn); + conn_maybe_write(conn); + check_assign_articles(); return OOP_CONTINUE; } @@ -1083,27 +1448,17 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev, static void feedfile_eof(InputFile *ipf) { assert(ipf != main_input_file); /* promised by tailing_try_read */ - - 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; + inputfile_reading_stop(ipf); + + if (ipf == flushing_input_file) { + assert(sms==sm_SEPARATED || sms==sm_DROPPING); + if (main_input_file) inputfile_reading_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 */ } - - 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); } static InputFile *open_input_file(const char *path) { @@ -1112,123 +1467,142 @@ static InputFile *open_input_file(const char *path) { if (errno==ENOENT) return 0; sysfatal("unable to open input file %s", path); } + assert(fd>0); 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); return ipf; } -static void close_input_file(InputFile *ipf) { +static void close_input_file(InputFile *ipf) { /* does not free */ assert(!ipf->readable_callback); /* must have had ->on_cancel */ - assert(!ipf->filemon); /* must have had inputfile_tailing_stop */ - assert(!ipf->rd); /* must have had inputfile_tailing_stop */ + assert(!ipf->filemon); /* must have had inputfile_reading_stop */ + assert(!ipf->rd); /* must have had inputfile_reading_stop */ assert(!ipf->inprogress); /* no dangling pointers pointing here */ - - if (ipf->fd >= 0) - if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path); - - free(ipf); + xclose_perhaps(&ipf->fd, "input file ", ipf->path); } /*---------- 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); - - int r= link(feedfile, duct_path); - if (r) sysdie("link feedfile %s to flushing file %s", feedfile, dut_path); - /* => Hardlinked */ +#define X_BAD_DATA(m) return feedfile_got_bad_data(ipf,old_offset,data,m); - r= unlink(feedfile); - if (r) sysdie("unlink old feedfile link %s", feedfile); - /* => Moved */ + if (ev==OOP_RD_PARTREC) + feedfile_got_bad_data(ipf,old_offset,data,"missing final newline"); + /* but process it anyway */ - spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */ + if (ipf->skippinglong) { + if (ev==OOP_RD_OK) ipf->skippinglong= 0; /* fine now */ + return OOP_CONTINUE; + } + if (ev==OOP_RD_LONG) { + ipf->skippinglong= 1; + X_BAD_DATA("overly long line"); } - check_master_queue(); -} - - -/*========== tailing input file ==========*/ - -static void filemon_start(InputFile *ipf) { - assert(!ipf->filemon); + if (memchr(data,'\0',recsz)) X_BAD_DATA("nul byte"); + if (!recsz) X_BAD_DATA("empty line"); - ipf->filemon= xmalloc(sizeof(*ipf->filemon)); - memset(ipf->filemon, 0, sizeof(*ipf->filemon)); - filemon_method_startfile(ipf, ipf->filemon); -} + if (data[0]==' ') { + if (strspn(data," ") != recsz) X_BAD_DATA("line partially blanked"); + ipf->readcount_blank++; + return OOP_CONTINUE; + } -static void filemon_stop(InputFile *ipf) { - if (!ipf->filemon) return; - filemon_method_stopfile(ipf, ipf->filemon); - free(ipf->filemon); - ipf->filemon= 0; + 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->state= art_Unchecked; + art->midlen= midlen; + art->ipf= ipf; ipf->inprogress++; + art->token= TextToToken(tokentextbuf); + art->offset= ipf->offset; + art->blanklen= recsz; + 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_assign_articles(); + return OOP_CONTINUE; } -static void filemon_callback(InputFile *ipf) { - ipf->readable_callback(ipf->readable_callback_user); -} +/*========== tailing input file ==========*/ 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; @@ -1236,9 +1610,9 @@ static int tailing_on_readable(struct oop_readable *rable, tailing_on_cancel(rable); ipf->readable_callback= cb; ipf->readable_callback_user= user; - filemon_startfile(ipf); + filemon_start(ipf); - loop->on_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf); + tailing_queue_readable(ipf); return 0; } @@ -1255,11 +1629,11 @@ static ssize_t tailing_try_read(struct oop_readable *rable, void *buffer, if (ipf==main_input_file) { errno=EAGAIN; return -1; - } else if (ipf==old_input_file) { - assert(ipf->fd>=0); + } else if (ipf==flushing_input_file) { + assert(ipf->rd); assert(sms==sm_SEPARATED || sms==sm_DROPPING); } else if (ipf==backlog_input_file) { - assert(ipf->fd>=0); + assert(ipf->rd); } else { abort(); } @@ -1279,9 +1653,9 @@ static int filemon_inotify_fd; static int filemon_inotify_wdmax; static InputFile **filemon_inotify_wd2ipf; -typedef struct Filemon_Perfile { +struct Filemon_Perfile { int wd; -} Filemon_Inotify_Perfile; +}; static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) { int wd= inotify_add_watch(filemon_inotify_fd, ipf->path, IN_MODIFY); @@ -1299,11 +1673,15 @@ static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) { assert(!filemon_inotify_wd2ipf[wd]); filemon_inotify_wd2ipf[wd]= ipf; + debug("filemon inotify startfile %p wd=%d wdmax=%d", + ipf, wd, filemon_inotify_wdmax); + pf->wd= wd; } static void filemon_method_stopfile(InputFile *ipf, Filemon_Perfile *pf) { int wd= pf->wd; + debug("filemon inotify stopfile %p wd=%d", ipf, wd); int r= inotify_rm_watch(filemon_inotify_fd, filemon_inotify_wd); if (r) sysdie("inotify_rm_watch"); filemon_inotify_wd2ipf[wd]= 0; @@ -1315,7 +1693,7 @@ static void *filemon_inotify_readable(oop_source *lp, int fd, for (;;) { int r= read(filemon_inotify_fd, &iev, sizeof(iev)); if (r==-1) { - if (errno==EAGAIN) break; + if (isewouldblock(errno)) break; sysdie("read from inotify master"); } else if (r==sizeof(iev)) { assert(iev.wd >= 0 && iev.wd < filemon_inotify_wdmax); @@ -1323,6 +1701,7 @@ static void *filemon_inotify_readable(oop_source *lp, int fd, die("inotify read %d bytes wanted struct of %d", r, (int)sizeof(iev)); } InputFile *ipf= filemon_inotify_wd2ipf[iev.wd]; + debug("filemon inotify readable read %p wd=%p", iev.wd, ipf); filemon_callback(ipf); } return OOP_CONTINUE; @@ -1337,16 +1716,17 @@ static int filemon_method_init(void) { set nonblock; loop->on_fd(loop, filemon_inotify_fd, OOP_READ, filemon_inotify_readable); + debug("filemon inotify init filemon_inotify_fd=%d", filemon_inotify_fd); return 1; } -#endif /* HAVE_INOTIFY && !HAVE_FILEMON *// +#endif /* HAVE_INOTIFY && !HAVE_FILEMON */ /*---------- filemon dummy implementation ----------*/ #if !defined(HAVE_FILEMON) -typedef struct Filemon_Perfile { int dummy; } Filemon_Dummy_Perfile; +struct Filemon_Perfile { int dummy; }; static int filemon_method_init(void) { return 0; } static void filemon_method_startfile(InputFile *ipf, Filemon_Perfile *pf) { } @@ -1354,35 +1734,56 @@ static void filemon_method_stopfile(InputFile *ipf, Filemon_Perfile *pf) { } #endif /* !HAVE_FILEMON */ +/*---------- filemon generic interface ----------*/ + +static void filemon_start(InputFile *ipf) { + assert(!ipf->filemon); + + ipf->filemon= xmalloc(sizeof(*ipf->filemon)); + memset(ipf->filemon, 0, sizeof(*ipf->filemon)); + filemon_method_startfile(ipf, ipf->filemon); +} + +static void filemon_stop(InputFile *ipf) { + if (!ipf->filemon) return; + filemon_method_stopfile(ipf, ipf->filemon); + free(ipf->filemon); + ipf->filemon= 0; +} + +static void filemon_callback(InputFile *ipf) { + ipf->readable_callback(loop, &ipf->readable, ipf->readable_callback_user); +} + /*---------- interface to start and stop an input file ----------*/ static const oop_rd_style feedfile_rdstyle= { OOP_RD_DELIM_STRIP, '\n', - OOP_RD_NUL_FORBID, - OOP_RD_SHORTREC_EOF, + OOP_RD_NUL_PERMIT, + OOP_RD_SHORTREC_LONG, }; -static void inputfile_tailing_start(InputFile *ipf) { - assert(!ipf->fd); - ipf->readable->on_readable= tailing_on_readable; - ipf->readable->on_cancel= tailing_on_cancel; - ipf->readable->try_read= tailing_try_read; - ipf->readable->delete_tidy= 0; /* we never call oop_rd_delete_{tidy,kill} */ - ipf->readable->delete_kill= 0; +static void inputfile_reading_start(InputFile *ipf) { + assert(!ipf->rd); + 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; ipf->rd= oop_rd_new(loop, &ipf->readable, 0,0); - assert(ipf->fd); + assert(ipf->rd); int r= oop_rd_read(ipf->rd, &feedfile_rdstyle, MAX_LINE_FEEDFILE, - feedfile_got_article,ipf, feedfile_problem,ipf); + feedfile_got_article,ipf, feedfile_read_err, ipf); if (r) sysdie("unable start reading feedfile %s",ipf->path); } -static void inputfile_tailing_stop(InputFile *ipf) { - assert(ipf->fd); +static void inputfile_reading_stop(InputFile *ipf) { + assert(ipf->rd); oop_rd_cancel(ipf->rd); oop_rd_delete(ipf->rd); ipf->rd= 0; @@ -1394,90 +1795,111 @@ static void inputfile_tailing_stop(InputFile *ipf) { /* See official state diagram at top of file. We implement * this as follows: + * -8<- - ================ - WAITING - [Nothing/Noduct] - poll for F - ================ + .=======. + ||START|| + `=======' | - | 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] - | | - | |<---------------------------------------------------. + | 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->rd!=0 | | flsh->rd!=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->rd==0 | V flsh->rd==0 + | [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_reading_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); @@ -1492,30 +1914,31 @@ static void statemc_init(void) { 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 (errno==EACCES || isewouldblock(errno)) { + 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)) - sysdie("could not close stale lockfile %s", path_lock); + xclose(lockfd, "stale lockfile ", path_lock); } debug("startup: locked"); 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); @@ -1524,9 +1947,8 @@ static void statemc_init(void) { 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", @@ -1534,134 +1956,194 @@ static void statemc_init(void) { } } - InputFile *file_d= open_input_file(path_flushing); + struct stat stab_f, stab_d; + int noent_f; - if (file_d) { - struct stat stab_f, stab_d; - - 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->rd) 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; +static void statemc_check_flushing_done(void) { + InputFile *ipf= flushing_input_file; + if (!inputfile_is_done(ipf)) return; - if (ipf->inprogress) return; /* new article in the meantime */ - if (ipf->fd >= 0); return; /* not had EOF */ - - if (ipf == backlog_input_file) { - notice_processed(ipf,"backlog file",ipf->path); - close_input_file(ipf); - if (unlink(ipf->path)) { - if (errno != ENOENT) - sysdie("could not unlink processed backlog file %s", ipf->path); - warn("backlog file %s vanished while we were reading it" - " so we couldn't remove it (but it's done now, anyway)", - ipf->path); - } - backlog_input_file= 0; - search_backlog_file(); - return; - } - - assert(ipf == old_input_file); assert(sms==sm_SEPARATED || sms==sm_DROPPING); - notice_processed(ipf,"feed file",0); + 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: + case sm_FLUSHFAILED: + if (!main_input_file) xtra= "-ABSENT"; + break; + case sm_SEPARATED: + case sm_DROPPING: + xtra= flushing_input_file->rd ? "-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); } } @@ -1677,7 +2159,7 @@ static void open_defer(void) { /* 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); @@ -1692,7 +2174,7 @@ static void open_defer(void) { 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); @@ -1725,42 +2207,47 @@ static void close_defer(void) { if (!defer) return; - xfstat(fileno(defer), &stab, "defer file"); + struct stat stab; + xfstat_isreg(fileno(defer), &stab, path_defer, "defer file"); if (fclose(defer)) sysfatal("could not close defer file %s", path_defer); defer= 0; + time_t now= xtime(); + char *backlog= xasprintf("%s_backlog_%lu.%lu", feedfile, - (unsigned long)now.tv_sec, + (unsigned long)now, (unsigned long)stab.st_ino); - if (link(path_defer, path_backlog)) + if (link(path_defer, backlog)) sysfatal("could not install defer file %s as backlog file %s", path_defer, backlog); if (unlink(path_defer)) sysdie("could not unlink old defer link %s to backlog file %s", path_defer, backlog); - if (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; - int r; + int r, i; struct stat stab; const char *oldest_path=0; - time_t oldest_mtime, now; + time_t oldest_mtime=0, now; - if (backlog_input_file) return 3; + if (backlog_input_file) return; try_again: @@ -1800,17 +2287,21 @@ static int search_backlog_file(void) { " nonzero (error?) return value %d", globpat_backlog, r); } - globfree(&gl); - if (!oldest_path) { debug("backlog scan: none"); - backlog_nextscan_periods= backlog_spontaneous_rescan_periods; - return 0; + + 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_spontrescan_periods; + goto xfree; } - now= time(); if (now==-1) sysdie("time(2) failed"); + now= xtime(); double age= difftime(now, oldest_mtime); - long age_deficiency= (backlog_retry_minperiods * PERIOD_SECONDS) - age; + long age_deficiency= (backlog_retry_minperiods * period_seconds) - age; if (age_deficiency <= 0) { debug("backlog scan: found age=%f deficiency=%ld oldest=%s", @@ -1818,36 +2309,42 @@ static int search_backlog_file(void) { backlog_input_file= open_input_file(oldest_path); if (!backlog_input_file) { - warn("backlog file %s vanished as we opened it", backlog_input_file); + warn("backlog file %s vanished as we opened it", oldest_path); + globfree(&gl); goto try_again; } - inputfile_tailing_start(backlog_input_file); - backlog_nextscan_periods= -1; - return 1; + inputfile_reading_start(backlog_input_file); + until_backlog_nextscan= -1; + goto xfree; } - 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; + if (backlog_spontrescan_periods >= 0 && + until_backlog_nextscan > backlog_spontrescan_periods) + until_backlog_nextscan= backlog_spontrescan_periods; debug("backlog scan: young age=%f deficiency=%ld nextscan=%d oldest=%s", - age, age_deficiency, backlog_nextscan_periods, oldest_path); - return 2; + age, age_deficiency, until_backlog_nextscan, oldest_path); + + xfree: + globfree(&gl); + return; } /*========== flushing the feed ==========*/ static pid_t inndcomm_child; +static int inndcomm_sentinel_fd; 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"); - loop->cancel_fd(fd); - close(fd); + cancel_fd_read_except(fd); + xclose_perhaps(&fd, "inndcomm sentinel pipe",0); - assert(!old_input_file); + assert(!flushing_input_file); if (WIFEXITED(status)) { switch (WEXITSTATUS(status)) { @@ -1857,17 +2354,38 @@ static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) { 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: @@ -1883,7 +2401,8 @@ static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) { } failed: - SMS(FLUSHFAIL, flushfail_retry_periods, "flush failed, will retry"); + SMS(FLUSHFAILED, flushfail_retry_periods, "flush failed, will retry"); + return OOP_CONTINUE; } static void inndcommfail(const char *what) { @@ -1891,21 +2410,26 @@ 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); + assert(!inndcomm_sentinel_fd); if (pipe(pipefds)) sysdie("create pipe for inndcomm child sentinel"); - inndcomm_child= xfork(); + inndcomm_child= xfork("inndcomm child"); if (!inndcomm_child) { - static char flushargv[2]= { sitename, 0 }; + const char *flushargv[2]= { sitename, 0 }; char *reply; + int r; - close(pipefds[0]); + xclose(pipefds[0], "(in child) inndcomm sentinel parent's end",0); + /* parent spots the autoclose of pipefds[1] when we die or exit */ alarm(inndcomm_flush_timeout); r= ICCopen(); if (r) inndcommfail("connect"); @@ -1917,254 +2441,381 @@ void spawn_inndcomm_flush(void) { exit(INNDCOMMCHILD_ESTATUS_FAIL); } - close(pipefds[1]); - int sentinel_fd= pipefds[0]; - on_fd_read_except(sentinel_fd, inndcomm_event); + xclose(pipefds[1], "inndcomm sentinel child's end",0); + inndcomm_sentinel_fd= pipefds[0]; + assert(inndcomm_sentinel_fd); + on_fd_read_except(inndcomm_sentinel_fd, inndcomm_event); - SMS(FLUSHING, 0, "flush is in progress"); + SMS(FLUSHING, 0, why); } /*========== main program ==========*/ static void postfork_inputfile(InputFile *ipf) { if (!ipf) return; - assert(ipf->fd >= 0); - close(ipf->fd); - ipf->fd= -1; -} - -static void postfork_conns(Connection *conn) { - while (conn) { - close(conn->fd); - conn= conn->next; - } + xclose(ipf->fd, "(in child) input file ", ipf->path); } -static void postfork_stdio(FILE *f) { +static void postfork_stdio(FILE *f, const char *what, const char *what2) { /* we have no stdio streams that are buffered long-term */ - if (f) fclose(f); + if (!f) return; + if (fclose(f)) sysdie("(in child) close %s%s", what, what2?what2:0); } -static void postfork(const char *what) { +static void postfork(void) { if (signal(SIGPIPE, SIG_DFL) == SIG_ERR) - sysdie("%s child: failed to reset SIGPIPE"); + sysdie("(in child) failed to reset SIGPIPE"); postfork_inputfile(main_input_file); - postfork_inputfile(old_input_file); - postfork_conns(idle.head); - postfork_conns(working.head); - postfork_conns(full.head); - postfork_stdio(defer); + postfork_inputfile(flushing_input_file); + + Conn *conn; + for (conn=LIST_HEAD(conns); conn; conn=LIST_NEXT(conn)) + conn_closefd(conn,"(in child) "); + + postfork_stdio(defer, "defer file ", path_defer); } -#define EVERY(what, interval, body) \ - static const struct timeval what##_timeout = { 5, 0 }; \ +#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 } \ + body; \ what##_schedule(); \ + return OOP_CONTINUE; \ } \ static void what##_schedule(void) { \ loop->on_time(loop, what##_timeout, what##_timedout, 0); \ } -EVERY(filepoll, {5,0}, { +EVERY(filepoll, 5,0, ({ if (main_input_file && main_input_file->readable_callback) filemon_callback(main_input_file); -}); +})); + +static char *debug_report_ipf(InputFile *ipf) { + if (!ipf) return xasprintf("-"); + + 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", + ipf, path, + ipf->inprogress, (long)ipf->offset, + ipf->fd, ipf->rd ? "+" : ""); +} + +EVERY(period, -1,0, ({ + 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_names[sms], sm_period_counter, + conns.count, queue.count, until_connect, + dipf_main, dipf_flushing, dipf_backlog, + (long)connecting_child, (long)inndcomm_child + ); + + free(dipf_main); + free(dipf_flushing); + free(dipf_backlog); + + if (until_connect) until_connect--; -EVERY(period, {PERIOD_SECONDS,0}, { - if (connect_delay) connect_delay--; poll_backlog_file(); if (!backlog_input_file) close_defer(); /* want to start on a new backlog */ - statemc_poll(); - check_master_queue(); -}); + statemc_period_poll(); + check_assign_articles(); + check_idle_conns(); +})); /*========== option parsing ==========*/ +static void vbadusage(const char *fmt, va_list al) NORET_PRINTF(1,0); +static void vbadusage(const char *fmt, va_list al) { + char *m= xvasprintf(fmt,al); + fprintf(stderr, "bad usage: %s\n" + "say --help for help, or read the manpage\n", + m); + if (become_daemon) + syslog(LOG_CRIT,"innduct: invoked with bad usage: %s",m); + exit(8); +} + +/*---------- generic option parser ----------*/ + +static void badusage(const char *fmt, ...) NORET_PRINTF(1,2); +static void badusage(const char *fmt, ...) { + va_list al; + va_start(al,fmt); + vbadusage(fmt,al); +} + enum OptFlags { - of_seconds= 001000u; - of_boolean= 002000u; + of_seconds= 001000u, + of_boolean= 002000u, }; typedef struct Option Option; typedef void OptionParser(const Option*, const char *val); struct Option { - int short; - const char *long; + int shrt; + const char *lng, *formarg; void *store; OptionParser *fn; - int noarg; + int intval; }; -void op_integer(const Option *o, const char *val) { +static void parse_options(const Option *options, char ***argvp) { + /* on return *argvp is first non-option arg; argc is not updated */ + + for (;;) { + const char *arg= *++(*argvp); + if (!arg) break; + if (*arg != '-') break; + if (!strcmp(arg,"--")) { arg= *++(*argvp); break; } + int a; + while ((a= *++arg)) { + const Option *o; + if (a=='-') { + arg++; + char *equals= strchr(arg,'='); + int len= equals ? (equals - arg) : strlen(arg); + for (o=options; o->shrt || o->lng; o++) + if (strlen(o->lng) == len && !memcmp(o->lng,arg,len)) + goto found_long; + badusage("unknown long option --%s",arg); + found_long: + if (!o->formarg) { + if (equals) badusage("option --%s does not take a value",o->lng); + arg= 0; + } else if (equals) { + arg= equals+1; + } else { + arg= *++(*argvp); + if (!arg) badusage("option --%s needs a value for %s", + o->lng, o->formarg); + } + o->fn(o, arg); + break; /* eaten the whole argument now */ + } + for (o=options; o->shrt || o->lng; o++) + if (a == o->shrt) + goto found_short; + badusage("unknown short option -%c",a); + found_short: + if (!o->formarg) { + o->fn(o,0); + } else { + if (!*++arg) { + arg= *++(*argvp); + if (!arg) badusage("option -%c needs a value for %s", + o->shrt, o->formarg); + } + o->fn(o,arg); + break; /* eaten the whole argument now */ + } + } + } +} + +#define DELIMPERHAPS(delim,str) (str) ? (delim) : "", (str) ? (str) : "" + +static void print_options(const Option *options, FILE *f) { + const Option *o; + for (o=options; o->shrt || o->lng; o++) { + char shrt[2] = { o->shrt, 0 }; + char *optspec= xasprintf("%s%s%s%s%s", + o->shrt ? "-" : "", shrt, + o->shrt && o->lng ? "|" : "", + DELIMPERHAPS("--", o->lng)); + fprintf(f, " %s%s%s\n", optspec, DELIMPERHAPS(" ", o->formarg)); + free(optspec); + } +} + +/*---------- specific option types ----------*/ + +static void op_integer(const Option *o, const char *val) { char *ep; errno= 0; unsigned long ul= strtoul(val,&ep,10); if (*ep || ep==val || errno || ul>INT_MAX) - badusage("bad integer value for %s",o->long); + badusage("bad integer value for %s",o->lng); int *store= o->store; *store= ul; } -void op_double(const Option *o, const char *val) { +static void op_double(const Option *o, const char *val) { int *store= o->store; char *ep; errno= 0; *store= strtod(val, &ep); if (*ep || ep==val || errno) - badusage("bad floating point value for %s",o->long); + badusage("bad floating point value for %s",o->lng); } -void op_string(const Option *o, const char *val) { - char **store= o->store; - free(*store); +static void op_string(const Option *o, const char *val) { + const char **store= o->store; *store= val; } -void op_seconds(const Option *o, const char *val) { +static void op_seconds(const Option *o, const char *val) { int *store= o->store; char *ep; + int unit; double v= strtod(val,&ep); - if (ep==val) badusage("bad time/duration value for %s",o->long); + if (ep==val) badusage("bad time/duration value for %s",o->lng); if (!*ep || !strcmp(ep,"s")) unit= 1; else if (!strcmp(ep,"m")) unit= 60; else if (!strcmp(ep,"h")) unit= 3600; else if (!strcmp(ep,"d")) unit= 86400; - else badusage("bad units %s for time/duration value for %s",ep,o->long); + else badusage("bad units %s for time/duration value for %s",ep,o->lng); v *= unit; v= ceil(v); - if (v > INT_MAX) badusage("time/duration value for %s out of range",o->long); + if (v > INT_MAX) badusage("time/duration value for %s out of range",o->lng); *store= v; } -void op_periods_rndup(const Option *o, const char *val) { +static void op_setint(const Option *o, const char *val) { int *store= o->store; - op_seconds(o,val); - *store += PERIOD_SECONDS-1; - *store /= PERIOD_SECONDS; + *store= o->intval; } -void op_periods_booltrue(const Option *o, const char *val) { - int *store= o->store; - *store= 1; -} -void op_periods_boolfalse(const Option *o, const char *val) { - int *store= o->store; - *store= 0; -} - -static const Option options[]= { -{ 0, "max-connections", &max_connections op_integer }, -{ 0, "streaming", &try_stream, op_booltrue, 1 }, -{ 0, "no-streaming", &try_stream, op_boolfalse, 1 }, -{'h',"host", &remote_host, op_string }, -{'P',"port", &port op_integer }, -{ 0, "inndconf", &inndconffile, op_string }, -{'f',"feedfile", &feedfile, op_string }, -{'q',"quiet-multiple", &quiet_if_locked, op_booltrue, 1 }, -{ 0, "no-quiet-multiple", &quiet_if_locked, op_boolfalse, 1 }, -{'d',"daemon", &become_daemon, op_booltrue, 1 }, -{ 0, "no-daemon", &become_daemon, op_boolfalse, 1 }, - -{ 0, "no-check-proportion", &nocheck_thresh_pct, op_double }, -{ 0, "no-check-filter", &nocheck_decay_articles, op_double }, - -{ 0, "max-queue-size", &max_queue_per_conn op_integer }, -{ 0, "reconnect-interval", &reconnect_delay_periods, op_periods_rndup }, -{ 0, "flush-retry-interval", &flushfail_retry_periods, op_periods_rndup }, -{ 0, "feedfile-open-timeout", &open_wait_periods, op_periods_rndup }, -{ 0, "connection-timeout", &connection_timeout, op_seconds }, -{ 0, "inndcomm-timeout", &inndcomm_flush_timeout, op_seconds }, +/*---------- specific options ----------*/ + +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-streaming", 0, &try_stream, op_setint, 0 }, +{0,"inndconf", "F", &inndconffile, op_string }, +{'P',"port", "PORT", &port, op_integer }, +{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,"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,"no-check-proportion", "PERCENT", &nocheck_thresh, op_double }, +{0,"no-check-response-time","ARTICLES", &nocheck_decay, op_double }, + +{0,"reconnect-interval", "PERIOD", &reconnect_delay_periods, op_seconds }, +{0,"flush-retry-interval", "PERIOD", &flushfail_retry_periods, op_seconds }, +{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,"idle-timeout", "PERIOD", &need_activity_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 }, + +{0,0} }; -int main(int argc, char **argv) { - const char *arg; +static void printusage(FILE *f) { + fputs("usage: innduct [options] site [fqdn]\n" + "available options are:\n", f); + print_options(innduct_options, f); +} - for (;;) { - arg= *++argv; - if (!arg) break; - if (*arg != '-') break; - if (!strcmp(arg,"--")) { arg= *++argv; break; } - int a; - while ((a= *++arg)) { - const Option *o; - if (a=='-') { - arg++; - char *equals= strchr(arg,'='); - int len= equals ? (equals - arg) : strlen(arg); - for (o=options; o->long; o++) - if (strlen(o->long) == len && !memcmp(o->long,arg,len)) - goto found_long; - badusage("unknown long option --%s",arg); - found_long: - if (o->noarg) { - if (equals) badusage("option --%s does not take a value",o->long); - arg= 0; - } else if (equals) { - arg= equals+1; - } else { - arg= *++argv; - if (!arg) badusage("option --%s needs a value",o->long); - } - o->fn(o, arg); - break; /* eaten the whole argument now */ - } - for (o=options; o->long; o++) - if (a == o->short) - goto found_short; - badusage("unknown short option -%c",a); - found_short: - if (o->noarg) { - o->fn(o,0); - } else { - if (!*++arg) { - arg= *++argv; - if (!arg) badusage("option -%c needs a value",o->short); - } - o->fn(o,arg); - break; /* eaten the whole argument now */ - } - } +static void help(const Option *o, const char *val) { + printusage(stdout); + if (ferror(stdout) || fflush(stdout)) { + perror("innduct: writing help"); + exit(12); + } + exit(0); +} + +static void convert_to_periods_rndup(int *store) { + *store += period_seconds-1; + *store /= period_seconds; +} + +int main(int argc, char **argv) { + if (!argv[1]) { + printusage(stderr); + exit(12); } - if (!arg) badusage("need site name argument"); - if (*++argv) badusage("too many non-option arguments"); - sitename= arg; + parse_options(innduct_options, &argv); - 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; + /* arguments */ - if (nocheck_decay_articles < 0.1) - badusage("nocheck decay articles must be at least 0.1"); - nocheck_decay= 1 - 1/nocheck_decay_articles; + sitename= *argv++; + if (!sitename) badusage("need site name argument"); + remote_host= *argv++; + if (*argv) badusage("too many non-option arguments"); - innconf_read(inndconffile); + /* defaults */ - if (!feedfile) - feedfile= xasprintf("%s/%s",pathoutgoing,sitename); - else if (!feedfile[0]) + if (!remote_host) remote_host= sitename; + + if (nocheck_thresh < 0 || nocheck_thresh > 100) + badusage("nocheck threshold percentage must be between 0..100"); + nocheck_thresh *= 0.01; + + if (nocheck_decay < 0.1) + badusage("nocheck decay articles must be at least 0.1"); + nocheck_decay= 1 - 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(&need_activity_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]=='/') + } 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); + + /* set things up */ + + oop_source_sys *sysloop= oop_sys_new(); + if (!sysloop) sysdie("could not create liboop event loop"); + loop= (oop_source*)sysloop; if (signal(SIGPIPE, SIG_IGN) == SIG_ERR) sysdie("could not ignore SIGPIPE"); + LIST_INIT(conns); + LIST_INIT(queue); + if (become_daemon) { + int i; for (i=3; i<255; i++) /* do this now before we open syslog, etc. */ close(i); @@ -2175,7 +2826,7 @@ int main(int argc, char **argv) { dup2(null,0); dup2(null,1); dup2(null,2); - close(null); + xclose(null, "/dev/null original fd",0); pid_t child1= xfork("daemonise first fork"); if (child1) _exit(0); @@ -2189,14 +2840,19 @@ int main(int argc, char **argv) { notice("starting"); - if (!filemon_init()) { + if (!filemon_method_init()) { warn("no file monitoring available, polling"); filepoll_schedule(); } + period_timeout.tv_sec= period_seconds; period_schedule(); statemc_init(); - loop->execute. + /* let's go */ + + void *r= oop_sys_run(sysloop); + assert(r == OOP_ERROR); + sysdie("event loop failed"); }