X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ian/git?p=inn-innduct.git;a=blobdiff_plain;f=backends%2Finnduct.c;h=a5bb448e3e81add5319fb0d917918c85ce4f6b6f;hp=1c41554258d8b1d6e985a274a427262b26ba42c6;hb=8b7074062790fe3362d1fa7ca28bf68d84820e03;hpb=9481af2261f3880d5636da02cd442c40bb022260 diff --git a/backends/innduct.c b/backends/innduct.c index 1c41554..a5bb448 100644 --- a/backends/innduct.c +++ b/backends/innduct.c @@ -15,28 +15,38 @@ * or might be blanked out * .... * - * site.name_duct.lock lock preventing multiple ducts - * holder of this lock is "duct" * F site.name main feed file * opened/created, then written, by innd * read by duct * unlinked by duct * tokens blanked out by duct when processed - * D site.name_duct temporary feed file during flush (or crash) + * site.name_lock lock preventing multiple ducts + * to hold lock must open,F_SETLK[W] + * and then stat to check that locked file + * still has name site.name_lock + * holder of this lock is "duct" + * (only) lockholder may remove the lockfile + * D site.name_flushing temporary feed file during flush (or crash) * hardlink created by duct * unlinked by duct - * site.name_duct.defer 431'd articles, still being written, + * site.name_defer 431'd articles, still being written, * created, written, used by duct - * site.name_backlog.lock lock taken out by innxmit wrapper - * holder and its child are "xmit" - * site.name_backlog_. + * + * site.name_backlog.. * 431'd articles, ready for innxmit or duct * created (link/mv) by duct - * site.name_backlog_ eg - * site.name_backlog_manual + * site.name_backlog (where does not + * contain '#' or '~') eg + * site.name_backlog.manual * anything the sysadmin likes (eg, feed files * from old feeds to be merged into this one) * created (link/mv) by admin + * may be symlinks (in which case links + * may be written through, but only links + * will be removed. + * + * It is safe to remove backlog files manually, + * if it's desired to throw away the backlog. * * Backlog files are also processed by innduct. We find the oldest * backlog file which is at least a certain amount old, and feed it @@ -46,33 +56,20 @@ * 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 | + ,-->--. check F, D + | | | + | | | + | | <----------------<---------------------------------'| + | | F exists | + | | D ENOENT | + | | duct opens F | | V | | Normal | | F: innd writing, duct reading | @@ -86,31 +83,35 @@ | 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 / @@ -135,10 +136,38 @@ "duct reading" means innduct is reading the file but also overwriting processed tokens. + * ->8- -^L- * + * rune for printing diagrams: + +perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct.c |a2ps -R -B -ops + * */ +/*============================== PROGRAM ==============================*/ + +#define _GNU_SOURCE + +#include "config.h" +#include "storage.h" +#include "nntp.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include /*----- general definitions, probably best not changed -----*/ @@ -150,32 +179,12 @@ #define INNDCOMMCHILD_ESTATUS_FAIL 6 #define INNDCOMMCHILD_ESTATUS_NONESUCH 7 - -/*----- configuration options -----*/ - -static char *sitename, *feedfile; -static int max_connections=10, max_queue_per_conn=200; -static int connection_setup_timeout=200, port=119, try_stream=1; -static int inndcomm_flush_timeout=100, quiet_if_locked=0; -static const char *remote_host; -static int reconnect_delay_periods, flushfail_retry_periods, open_wait_periods; -static int backlog_retry_minperiods, backlog_spontaneous_rescan_periods; -static const char *inndconffile; - -static double accept_proportion; -static double nocheck_thresh_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; - - /*----- doubly linked lists -----*/ -#define ISNODE(T) T *next, *back; -#define LIST(T) struct { T *head, *tail, *tailpred; int count; } +#define ISNODE(T) struct { T *succ, *pred; } node +#define DEFLIST(T) typedef struct { T *head, *tail, *tp; int count; } T##List -#define NODE(n) ((struct node*)&(n)->head) +#define NODE(n) ((struct node*)&(n)->node) #define LIST_ADDHEAD(l,n) \ (list_addhead((struct list*)&(l), NODE((n))), (void)(l).count++) @@ -191,34 +200,89 @@ static int nocheck, nocheck_reported; #define LIST_INSERT(l,n,pred) \ (list_insert((struct list*)&(l), NODE((n)), NODE((pred))), (void)(l).count++) +/*----- type predeclarations -----*/ + +typedef struct Conn Conn; +typedef struct Article Article; +typedef enum StateMachineState StateMachineState; + +DEFLIST(Conn); +DEFLIST(Article); + +/*----- function predeclarations -----*/ + +static void conn_check_work(Conn *conn); + +static int filemon_init(void); +static void filemon_setfile(int mainfeed_fd, const char *mainfeed_path); +static void filemon_callback(void); + +static void conn_assign_one_article(ConnList *connlist, Conn **last_assigned); +static void statemc_setstate(StateMachineState newsms, int periods, + const char *forlog, const char *why); +static void check_master_queue(void); + +/*----- configuration options -----*/ + +static char *sitename, *feedfile; +static const char *remote_host; +static int quiet_multiple=0, become_daemon=1; + +static int max_connections=10, max_queue_per_conn=200; + +static int connection_setup_timeout=200, port=119, try_stream=1; +static int inndcomm_flush_timeout=100; +static int reconnect_delay_periods, flushfail_retry_periods, open_wait_periods; +static int backlog_retry_minperiods, backlog_spontaneous_rescan_periods; +static const char *inndconffile; + +static 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; + /*----- 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) RCI_##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 { @@ -232,16 +296,6 @@ typedef struct { /*----- 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 oop_readable readable; /* first */ @@ -249,30 +303,51 @@ typedef struct InputFile { void *readable_callback_user; int fd; - const char *path; /* ptr copy of path_ or feedfile */ struct Filemon_Perfile *filemon; oop_read *rd; long inprogress; /* no. of articles read but not processed */ off_t offset; - Counts counts; + int counts[art_MaxState][RCI_max]; + 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 */ + ArticleList queue; /* not yet told peer, or CHECK said send it */ + ArticleList sent; /* offered/transmitted - in xmit or waiting reply */ struct iovec xmit[CONNIOVS]; XmitDetails xmitd[CONNIOVS]; int xmitu; @@ -281,43 +356,98 @@ struct Conn { /*----- operational variables -----*/ -static int since_connect_attempt; -static int nconns; -static LIST(Conn) idle, working, full; -static LIST(Article) *queue; +static oop_source *loop; + +static int nconns, until_connect; +static ConnList idle, working, full; +static ArticleList queue; -static char *path_ductlock, *path_duct, *path_ductdefer; +static char *path_lock, *path_flushing, *path_defer; #define SMS(newstate, periods, why) \ (statemc_setstate(sm_##newstate,(periods),#newstate,(why))) static StateMachineState sms; static FILE *defer; -static InputFile *main_input_file, *old_input_file, *backlog_input_file; +static InputFile *main_input_file, *flushing_input_file, *backlog_input_file; static int sm_period_counter; -/*----- function predeclarations -----*/ +/*========== logging ==========*/ -static void conn_check_work(Conn *conn); +static void logcore(int sysloglevel, const char *fmt, ...) + __attribute__((__format__(printf,2,3))); +static void logcore(int sysloglevel, const char *fmt, ...) { + va_list al; + va_start(al,fmt); + if (become_daemon) { + vsyslog(sysloglevel,fmt,al); + } else { + vfprintf(stderr,fmt,al); + putc('\n',stderr); + } + va_end(al); +} -static int filemon_init(void); -static void filemon_setfile(int mainfeed_fd, const char *mainfeed_path); -static void filemon_callback(void); +static void logv(int sysloglevel, const char *pfx, int errnoval, + int exitstatus, const char *fmt, va_list al) + __attribute__((__format__(printf,5,0))); +static void logv(int sysloglevel, const char *pfx, int errnoval, + int exitstatus, const char *fmt, va_list al) { + char msgbuf[256]; + vsnprintf(msgbuf,sizeof(msgbuf), fmt,al); + msgbuf[sizeof(msgbuf)-1]= 0; + + if (sysloglevel >= LOG_ERR && (errnoval==EACCES || errnoval==EPERM)) + sysloglevel= LOG_ERR; /* run by wrong user, probably */ + + logcore(sysloglevel, "<%s>%s: %s%s%s", + sitename, pfx, msgbuf, + errnoval>=0 ? ": " : "", + errnoval>=0 ? strerror(errnoval) : ""); +} + +#define logwrap(fn, pfx, sysloglevel, err, estatus) \ + static void fn(const char *fmt, ...) \ + __attribute__((__format__(printf,1,2))); \ + static void fn(const char *fmt, ...) { \ + va_list al; \ + va_start(al,fmt); \ + logv(sysloglevel, pfx, err, estatus, fmt, al); \ + } + +logwrap(sysdie, " critical", LOG_CRIT, errno, 16); +logwrap(die, " critical", LOG_CRIT, -1, 16); + +logwrap(sysfatal, " fatal", LOG_ERR, errno, 12); +logwrap(fatal, " fatal", LOG_ERR, -1, 12); + +logwrap(syswarn, " warning", LOG_WARNING, errno, 0); +logwrap(warn, " warning", LOG_WARNING, -1, 0); + +logwrap(notice, "", LOG_NOTICE, -1, 0); +logwrap(info, " info", LOG_INFO, -1, 0); +logwrap(debug, " debug", LOG_DEBUG, -1, 0); -static void statemc_setstate(StateMachineState newsms, int periods, - const char *forlog, const char *why); /*========== utility functions etc. ==========*/ static void perhaps_close(int *fd) { if (*fd) { close(*fd); fd=0; } } +static void *xmalloc(size_t sz) { + if (!sz) return 0; + void *r= malloc(sz); + if (r) return r; + sysdie("malloc (%ld bytes) failed", (unsigned long)sz); +} + 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); return child; } @@ -352,7 +482,7 @@ static void report_child_status(const char *what, int status) { static int xwaitpid(pid_t *pid, const char *what) { int status; - r= kill(*pid, SIGKILL); + int r= kill(*pid, SIGKILL); if (r) sysdie("cannot kill %s child", what); pid_t got= waitpid(*pid, &status, WNOHANG); @@ -363,6 +493,11 @@ 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 void check_isreg(const struct stat *stab, const char *path, const char *what) { if (!S_ISREG(stab->st_mode)) @@ -371,17 +506,18 @@ 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= fstab(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) { @@ -392,6 +528,13 @@ static void xlstat_isreg(const char *path, struct stat *stab, check_isreg(stab, path, what); } +static void setnonblock(int fd, int nonblocking) { + int r= fcntl(fd, F_GETFL); if (r<0) sysdie("setnonblocking fcntl F_GETFL"); + if (nonblocking) r |= O_NONBLOCK; + else r &= ~O_NONBLOCK; + r= fcntl(fd, F_SETFL, r); if (r<0) sysdie("setnonblocking fcntl F_SETFL"); +} + static int samefile(const struct stat *a, const struct stat *b) { assert(S_ISREG(a->st_mode)); assert(S_ISREG(b->st_mode)); @@ -399,61 +542,6 @@ static int samefile(const struct stat *a, const struct stat *b) { a->st_dev == b->st_dev); } -/*========== logging ==========*/ - -static void logcore(int sysloglevel, const char *fmt, ...) - __attribute__((printf,2,3)) -static void logcore(int sysloglevel, const char *fmt, ...) { - va_list al; - va_start(al,fmt); - if (become_daemon) { - vsyslog(sysloglevel,fmt,al); - } else { - vfprintf(stderr,fmt,al); - putc('\n',stderr); - } -} - -static void logv(int sysloglevel, const char *pfx, int errnoval, - int exitstatus, const char *fmt, va_list al) - __attribute__((printf,4,0)) -static void logv(int sysloglevel, const char *pfx, int errnoval, - int exitstatus, const char *fmt, va_list al) { - char msgbuf[256]; - vsnprintf(msgbuf,sizeof(msgbuf), fmt,al); - msgbuf[sizeof(msgbuf)-1]= 0; - - if (sysloglevel >= LOG_ERR && (errnoval==EACCES || errnoval==EPERM)) - sysloglevel= LOG_ERR; /* run by wrong user, probably */ - - logcore(sysloglevel, "<%s>%s: %s%s%s", - sitename, pfx, msgbuf, - errnoval>=0 ? ": " : "", - errnoval>=0 ? strerror(errnoval) : ""); -} - -#define logwrap(fn, pfx, sysloglevel, errno, estatus) \ - static void fn(const char *fmt, ...) \ - __attribute__((printf,1,2)); \ - static void fn(const char *fmt, ...) { \ - va_list al; \ - va_start(al,fmt); \ - logv(sysloglevel, pfx, errno, estatus, fmt, al); \ - } - -logwrap(sysdie, " critical", LOG_CRIT, errno, 16); -logwrap(die, " critical", LOG_CRIT, -1, 16); - -logwrap(sysfatal, " fatal", LOG_ERR, errno, 12); -logwrap(fatal, " fatal", LOG_ERR, -1, 12); - -logwrap(syswarn, " warning", LOG_WARN, errno, 0); -logwrap(warn, " warning", LOG_WARN, -1, 0); - -logwrap(notice, "", LOG_NOTICE, -1, 0); -logwrap(info, " info", LOG_INFO, -1, 0); - - /*========== making new connections ==========*/ static int connecting_sockets[2]= {-1,-1}; @@ -467,6 +555,8 @@ static void connect_attempt_discard(void) { perhaps_close(&connecting_sockets[1]); if (connecting_child) { + int r= kill(connecting_child, SIGTERM); + if (r) syswarn("failed to kill connecting child"); int status= xwaitpid(&connecting_child, "connect"); if (!(WIFEXITED(status) || @@ -488,7 +578,7 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) { 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); @@ -499,11 +589,11 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) { assert(got==connecting_child); connecting_child= 0; if (WIFEXITED(status) && - (WEXITSTATUS(status) != 0 + (WEXITSTATUS(status) != 0 && WEXITSTATUS(status) != CONNCHILD_ESTATUS_STREAM && WEXITSTATUS(status) != CONNCHILD_ESTATUS_NOSTREAM)) { /* child already reported the problem */ - } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALARM) { + } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) { warn("connect: connection attempt timed out"); } else if (!WIFEXITED(status)) { report_child_status("connect", status); @@ -513,7 +603,7 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) { /* child is still running apparently, report the socket problem */ if (rs < 0) syswarn("connect: read from child socket failed"); - else if (e == OOP_EXCEPTIONN) + else if (e == OOP_EXCEPTION) warn("connect: unexpected exception on child socket"); else warn("connect: unexpected EOF on child socket"); @@ -528,13 +618,14 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) { } CHK(level, SOL_SOCKET); CHK(type, SCM_RIGHTS); - CHK(len, CMSG_LEN(sizeof(conn-b>fd))); + CHK(len, CMSG_LEN(sizeof(conn->fd))); #undef CHK - if (CMSG_NXTHDR,&msg,h) { die("connect: child sent many cmsgs"); goto x; } + if (CMSG_NXTHDR(&msg,h)) { die("connect: child sent many cmsgs"); goto x; } memcpy(&conn->fd, CMSG_DATA(h), sizeof(conn->fd)); + int status; pid_t got= waitpid(connecting_child, &status, 0); if (got==-1) sysdie("connect: real wait for child"); assert(got == connecting_child); @@ -549,11 +640,11 @@ 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; + setnonblocking(conn->fd, 1); /* Phew! */ LIST_ADDHEAD(idle, conn); - notice(CN "connected %s", conn->fd, conn->stream ? "streaming" : "plain"); + notice("#%d connected %s", conn->fd, conn->stream ? "streaming" : "plain"); connect_attempt_discard(); check_master_queue(); return 0; @@ -566,14 +657,14 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) { connect_attempt_discard(); } -static void connect_start() { +static void connect_start(void) { assert(!connecting_sockets[0]); assert(!connecting_sockets[1]); assert(!connecting_child); notice("starting connection attempt"); - r= socketpair(AF_UNIX, SOCK_STREAM, 0, connecting_sockets); + int r= socketpair(AF_UNIX, SOCK_STREAM, 0, connecting_sockets); if (r) { syswarn("connect: cannot create socketpair for child"); goto x; } connecting_child= xfork("connection"); @@ -615,11 +706,11 @@ static void connect_start() { fatal("connect: response to MODE STREAM is too long: %.100s...", remote_host, buf); } - l--; if (l>0 && buf[1-]=='\r') l--; + l--; if (l>0 && buf[l-1]=='\r') l--; buf[l]= 0; char *ep; int rcode= strtoul(buf,&ep,10); - if (ep != buf[3]) { + if (ep != &buf[3]) { sanitise_inplace(buf); fatal("connect: bad response to MODE STREAM: %.50s", buf); } @@ -657,7 +748,7 @@ static void connect_start() { if (r) sysdie("connect: close child socket in parent"); on_fd_read_except(connecting_sockets[0], connchild_event); - return OOP_CONTINUE; + return; x: connect_attempt_discard(); @@ -676,9 +767,9 @@ static void check_master_queue(void) { 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; + } else if (nconns < max_connections && queue.count >= max_queue_per_conn && + !connecting_child && !until_connect) { + until_connect= reconnect_delay_periods; connect_start(); } else { break; @@ -687,8 +778,7 @@ static void check_master_queue(void) { conn_check_work(last_assigned); } -static void conn_assign_one_article(LIST(Conn) *connlist, - Conn **last_assigned) { +static void conn_assign_one_article(ConnList *connlist, Conn **last_assigned) { Conn *conn= connlist->head; LIST_REMOVE(*connlist, conn); @@ -708,7 +798,7 @@ static int conn_total_queued_articles(Conn *conn) { return conn->sent.count + conn->queue.count; } -static LIST(Conn) *conn_determine_right_list(Conn *conn) { +static ConnList *conn_determine_right_list(Conn *conn) { int inqueue= conn_total_queued_articles(conn); assert(inqueue <= max_queue); if (inqueue == 0) return &idle; @@ -744,6 +834,45 @@ static void conn_check_work(Conn *conn) { } } +static void vconnfail(Conn *conn, const char *fmt, va_list al) + __attribute__((printf,2,0)); + +static void vconnfail(Conn *conn, const char *fmt, va_list al) { + int requeue[art_MaxState]; + + Article *art; + while ((art= LIST_REMHEAD(conn->queue))) LIST_ADDTAIL(queue); + while ((art= LIST_REMHEAD(conn->sent))) { + counts[art->state]++; + if (art->state==art_Unsolicited) art->state= art_Unchecked; + LIST_ADDTAIL(queue); + } + + int i; + XmitDetails *xd; + for (i=0, dp=&conn->xmitd; ixmitu; i++, dp++) + xmit_free(dp); + + char *m= xvasprintf(fmt,al); + warn("#%d connection failed, requeueing " RCI_TRIPLE_FMT_BASE ": %s", + conn->fd, RCI_TRIPLE_FMT_VALS(requeue, /*nothing*/), m); + free(m); + + close(conn->fd); + free(conn); + + until_connect= reconnect_delay_periods; + check_master_queue(); +} + +static void connfail(Connection *conn, const char *fmt, ...) + __attribute__((printf,2,3)); +static void connfail(Connection *conn, const char *fmt, ...) { + va_list al; + va_start(al,fmt); + vconnfail(fmt,al); + va_end(al); +} /*========== article transmission ==========*/ @@ -821,7 +950,7 @@ static void conn_make_some_xmits(Conn *conn) { Article *art= LIST_REMHEAD(queue); if (!art) break; - if (art->checked || (conn->stream && nocheck)) { + if (art->state >= art_Wanted || (conn->stream && nocheck)) { /* actually send it */ ARTHANDLE *artdata= SMretrieve(); @@ -842,11 +971,13 @@ 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(); + art->ipf->counts[art->state].sent++; LIST_ADDTAIL(conn->sent, art); - art->ipf->counts[art->checked].sent++; - } else { /* check it */ @@ -857,8 +988,9 @@ static void conn_make_some_xmits(Conn *conn) { xmit_noalloc(art->mid, art->midlen); XMIT_LITERAL("\r\n"); + assert(art->state == art_Unchecked); + art->ipf->counts[art->state].sent++; LIST_ADDTAIL(conn->sent, art); - art->ipf->counts[art->checked].offered++; } } } @@ -882,8 +1014,10 @@ static void *peer_rd_err(oop_source *lp, oop_read *oread, oop_event ev, static Article *article_reply_check(Connection *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= conn->sent.head; if (!art) { connfail(conn, @@ -922,22 +1056,37 @@ static Article *article_reply_check(Connection *conn, const char *response, } } + if (must_have_sent>0 && art->state < art_Wanted) { + connfail("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("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) { + 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]++; + art->ipf->counts[art->state][whichcount]++; if (whichcount == RC_accepted) update_nocheck(1); else if (whichcount == RC_unwanted) update_nocheck(0); @@ -963,8 +1112,8 @@ 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); + if (!ipf->inprogress && ipf != main_input_file) + queue_check_input_done(); free(art); } @@ -1010,13 +1159,13 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev, 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) \ +#define ARTICLE_DEALTWITH(streaming,musthavesent,how) \ code_streaming= (streaming) \ - GET_ARTICLE; \ + GET_ARTICLE(musthavesent); \ article_done(conn, art, RC_##how); break; #define PEERBADMSG(m) connfail(conn, m ": %s", sani); return OOP_CONTINUE @@ -1029,36 +1178,33 @@ 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++; + GET_ARTICLE(-1); + assert(art->state == art_Unchecked); + art->ipf->counts[art->state].accepted++; + art->state= art_Wanted; LIST_ADDTAIL(conn->queue); - if (art->checked) { - connfail("peer gave %d response to article body: %s",code, sani); - return OOP_CONTINUE; - } - art->checked= 1; 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)) - sysfatal("write to defer file %s",path_ductdefer); + sysfatal("write to defer file %s",path_defer); article_done(conn, art, RC_deferred); break; @@ -1074,26 +1220,20 @@ 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; - } - - assert(ipf == old_input_file); - inputfile_tailing_stop(ipf); assert(ipf->fd >= 0); if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path); ipf->fd= -1; - assert(sms==sm_SEPARATED || sms==sm_DROPPING); - - if (main_input_file) - inputfile_tailing_start(main_input_file); + if (ipf == flushing_input_file) { + assert(sms==sm_SEPARATED || sms==sm_DROPPING); + if (main_input_file) inputfile_tailing_start(main_input_file); + statemc_check_flushing_done(); + } else if (ipf == backlog_input_file) { + statemc_check_backlog_done(); + } else { + abort(); /* supposed to wait rather than get EOF on main input file */ + } } static InputFile *open_input_file(const char *path) { @@ -1103,7 +1243,7 @@ static InputFile *open_input_file(const char *path) { sysfatal("unable to open input file %s", path); } - InputFile *ipf= xmalloc(sizeof(InputFile)); + InputFile *ipf= xmalloc(sizeof(*ipf) + strlen(path) + 1); memset(ipf,0,sizeof(*ipf)); ipf->readable.on_readable= tailing_on_readable; @@ -1111,7 +1251,7 @@ static InputFile *open_input_file(const char *path) { ipf->readable.try_read= tailing_try_read; ipf->fd= fd; - ipf->path= path; + strcpy(ipf->path, path); return ipf; } @@ -1124,10 +1264,6 @@ static void close_input_file(InputFile *ipf) { if (ipf->fd >= 0) if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path); - - fixme maybe free ipf->path; - - free(ipf); } @@ -1159,7 +1295,7 @@ typedef void *feedfile_got_article(oop_source *lp, oop_read *rd, art->offset= ipf->offset; art->blanklen= recsz; art->midlen= midlen; - art->checked= art->sentbody= 0; + art->state= art_Unchecked; art->ipf= ipf; ipf->inprogress++; art->token= TextToToken(tokentextbuf); strcpy(art->messageid, space+1); @@ -1167,24 +1303,31 @@ typedef void *feedfile_got_article(oop_source *lp, oop_read *rd, } ipf->offset += recsz + 1; - if (sms==sm_NORMAL && ipf->offset >= flush_threshold) { - notice("starting flush (%lu >= %lu)", - (unsigned long)ipf->offset, (unsigned long)flush_threshold); + if (sms==sm_NORMAL && ipf==main_input_file && + ipf->offset >= flush_threshold) + statemc_start_flush("feed file size"); - int r= link(feedfile, duct_path); - if (r) sysdie("link feedfile %s to ductfile %s", feedfile, dut_path); - /* => Hardlinked */ + check_master_queue(); +} - r= unlink(feedfile); - if (r) sysdie("unlink old feedfile link %s", feedfile); - /* => Moved */ +static void statemc_start_flush(const char *why) { /* Normal => Flushing */ + assert(sms == sm_NORMAL); - spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */ - } + debug("starting flush (%s) (%lu >= %lu) (%d)", + why, + (unsigned long)ipf->offset, (unsigned long)flush_threshold, + sm_period_counter); - check_master_queue(); -} + int r= link(feedfile, duct_path); + if (r) sysdie("link feedfile %s to flushing file %s", feedfile, + path_duct); + /* => Hardlinked */ + + xunlink(feedfile, "old feedfile link"); + /* => Moved */ + spawn_inndcomm_flush(why); /* => Flushing FLUSHING */ +} /*========== tailing input file ==========*/ @@ -1221,6 +1364,12 @@ static void on_cancel(struct oop_readable *rable) { ipf->readable_callback= 0; } +static void tailing_queue_readable(InputFile *ipf) { + /* lifetime of ipf here is OK because destruction will cause + * on_cancel which will cancel this callback */ + 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; @@ -1230,7 +1379,7 @@ static int tailing_on_readable(struct oop_readable *rable, ipf->readable_callback_user= user; filemon_startfile(ipf); - loop->on_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf); + tailing_queue_readable(ipf); return 0; } @@ -1247,7 +1396,7 @@ 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) { + } else if (ipf==flushing_input_file) { assert(ipf->fd>=0); assert(sms==sm_SEPARATED || sms==sm_DROPPING); } else if (ipf==backlog_input_file) { @@ -1291,11 +1440,15 @@ 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; @@ -1315,6 +1468,7 @@ 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; @@ -1329,6 +1483,7 @@ 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; } @@ -1386,99 +1541,114 @@ static void inputfile_tailing_stop(InputFile *ipf) { /* See official state diagram at top of file. We implement * this as follows: - - ================ - WAITING - [Nothing/Noduct] - poll for F - ================ - | - | TIMEOUT - |`--------------------------. - | | install defer as backlog - | OPEN F SUCCEEDS | exit - ,--------->| V - | V ========= - | ======== (ESRCH) - | NORMAL [Dropped] - | [Normal] ========= - | main F tail - | ======== - | | - | | F IS SO BIG WE SHOULD FLUSH - ^ | hardlink F to D - | [Hardlinked] - | | unlink F - | | our handle onto F is now onto D - | [Moved] - | | - | |<---------------------------------------------------. + * -8<- + + .=======. + ||START|| + `=======' + | + | open F + | + | F ENOENT + |`---------------------------------------------------. + F OPEN OK | | + |`---------------- - - - | + D ENOENT | D EXISTS see OVERALL STATES diagram | + | for full startup logic | + ,--------->| | + | V | + | ============ try to | + | NORMAL open D | + | [Normal] | + | main F tail | + | ============ V + | | | + | | F IS SO BIG WE SHOULD FLUSH, OR TIMEOUT | + ^ | hardlink F to D | + | [Hardlinked] | + | | unlink F | + | | our handle onto F is now onto D | + | [Moved] | + | | | + | |<-------------------<---------------------<---------+ | | | | | spawn inndcomm flush | | V | - | ========== | - | FLUSHING | + | ================== | + | FLUSHING[-ABSENT] | | [Flushing] | - | main D tail | - | ========== | + | main D tail/none | + | ================== | | | | | | INNDCOMM FLUSH FAILS ^ | |`----------------------->----------. | | | | | | | NO SUCH SITE V | - ^ |`--------------->----. =========== | - | | \ FLUSHFAIL | - | | \ [Moved] | - | | \ main D tail | - | | \ =========== | + ^ |`--------------->----. ==================== | + | | \ FLUSHFAILED[-ABSENT] | + | | \ [Moved] | + | | FLUSH OK \ main D tail/none | + | | open F \ ==================== | | | \ | | | | \ | TIME TO RETRY | - | | \ `----------------' - | | FLUSH OK \ - | | open F \ - | V V - | ============= ============ - | SEPARATED/ DROPPING/ - | old->fd>=0 old->fd>=0 - | [Separated] [Dropping] - | main F idle main none - | old D tail old D tail - | ============= ============ - | | | - ^ | EOF ON D | EOF ON D - | V V - | =============== =============== - | SEPARATED/ DROPPING/ - | old->fd==-1 old->fd==-1 - | [Finishing] [Dropping] - | main F tail main none - | old D closed old D closed - | =============== =============== - | | | - | | ALL D PROCESSED | ALL D PROCESSED - | V install defer as backlog V install defer as backlog - ^ | close D | close D - | | unlink D | unlink D - | | | exit - `----------' V - ========== - (ESRCH) - [Droppped] - ========== + | |`------->----. ,---<---'\ `----------------' + | | D NONE | | D NONE `----. + | V | | V + | ============= V V ============ + | SEPARATED-1 | | DROPPING-1 + | flsh->fd>=0 | | flsh->fd>=0 + | [Separated] | | [Dropping] + | main F idle | | main none + | old D tail | | old D tail + | ============= | | ============ + | | | | install | + ^ | EOF ON D | | defer | EOF ON D + | V | | V + | =============== | | =============== + | SEPARATED-2 | | DROPPING-2 + | flsh->fd==-1 | V flsh->fd==-1 + | [Finishing] | | [Dropping] + | main F tail | `. main none + | old D closed | `. old D closed + | =============== V `. =============== + | | `. | + | | ALL D PROCESSED `. | ALL D PROCESSED + | V install defer as backlog `. | install defer + ^ | close D `. | close D + | | unlink D `. | unlink D + | | | | + | | V V + `----------' ============== + DROPPED + [Dropped] + main none + old none + some backlog + ============== + | + | ALL BACKLOG DONE + | + | unlink lock + | exit + V + ========== + (ESRCH) + [Droppped] + ========== + * ->8- */ static void statemc_init(void) { struct stat stab, stabf; - int noent; - path_ductlock= xasprintf("%s_duct.lock", feedfile); - path_duct= xasprintf("%s_duct", feedfile); - path_ductdefer= xasprintf("%s_duct.defer", feedfile); - globpat_backlog= xasprintf("%s_backlog_*", feedfile); + path_lock= xasprintf("%s_lock", feedfile); + path_flushing= xasprintf("%s_flushing", feedfile); + path_defer= xasprintf("%s_defer", feedfile); + globpat_backlog= xasprintf("%s_backlog*", feedfile); for (;;) { - int lockfd= open(path_ductlock, O_CREAT|O_RDWR, 0600); - if (lockfd<0) sysfatal("open lockfile %s", path_ductlock); + int lockfd= open(path_lock, O_CREAT|O_RDWR, 0600); + if (lockfd<0) sysfatal("open lockfile %s", path_lock); struct flock fl; memset(&fl,0,sizeof(fl)); @@ -1487,34 +1657,39 @@ static void statemc_init(void) { r= fcntl(lockfd, F_SETLK, &fl); if (r==-1) { if (errno==EACCES || errno==EAGAIN) { - if (quiet_if_locked) exit(0); + if (quiet_multiple) exit(0); fatal("another duct holds the lockfile"); } - sysdie("fcntl F_SETLK lockfile %s", path_ductlock); + sysdie("fcntl F_SETLK lockfile %s", path_lock); } xfstat_isreg(lockfd, &stabf, "lockfile"); - xlstat_isreg(path_ductlock, &stab, &noent, "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_ductlock); + sysdie("could not close stale lockfile %s", path_lock); } debug("startup: locked"); - xlstat_isreg(path_ductdefer, &stab, &noent, "defer file"); - if (noent) { + search_backlog_file(); + + 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); switch (stab.st_nlink==1) { - case 1: /* ok */ break; + case 1: + open_defer(); /* so that we will later close it and rename it */ + break; case 2: - if (unlink(path_defer)) - sysdie("could not unlink stale defer file link %s (presumably" - " 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", @@ -1522,125 +1697,172 @@ static void statemc_init(void) { } } - InputFile *file_d= open_input_file(path_duct); - - if (file_d) { - struct stat stab_f, stab_d; + struct stat stab_f, stab_d; + int noent_f; - xlstat_isreg(feedfile, &stab_f, &noent, "feed file"); - if (noent) { - debug("startup: D exists, F ENOENT => Moved"); - goto found_moved; - } + InputFile *file_d= open_input_file(path_flushing); + if (file_d) xfstat_isreg(file_d->fd, &stab_d, "flushing file"); - debug("startup: F and D both exist"); + xlstat_isreg(feedfile, &stab_f, &noent_f, "feedfile"); - xfstat_isreg(file_d->fd, &stab_d, "ductfile"); + 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_duct); - 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"); + FILE *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"); - 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_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 void startup_set_input_file(InputFile *f) { assert(!main_input_file); main_input_file= f; inputfile_tailing_start(f); } -static void *statemc_check_input_done(oop_source *lp, - struct timeval now, void *ipf_v) { - InputFile *ipf= ipf_v; - struct stat stab; - - if (ipf->inprogress) return; /* new article in the meantime */ - if (ipf->fd >= 0); return; /* not had EOF */ +static int inputfile_is_done(InputFile *ipf) { + if (!ipf) return 0; + if (ipf->inprogress) return 0; /* new article in the meantime */ + if (ipf->fd >= 0); return 0; /* not had EOF */ + return 1; +} - if (ipf == backlog_input_file) { - notice_processed(ipf,"backlog file",ipf->path); - close_input_file(ipf); - if (unlink(ipf->path)) - sysdie("could not unlink done backlog file %s", ipf->path); - backlog_input_file= 0; - search_backlog_file(); - return; +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, .x) + + info("processed %s%s offered=%d(ch%d,nc%d) accepted=%d(ch%d+nc%d)" + RESULT_COUNTS(RCI_NOTHING, RCI_TRIPLE_FMT) + , + what,spec, + ipf->counts[art_Unchecked].sent + ipf->counts[art_Unsolicited].sent + , ipf->counts[art_Unchecked].sent, ipf->counts[art_Unsolicited].sent, + ipf->counts[art_Wanted].accepted + ipf->counts[art_Unsolicited].accepted + ,ipf->counts[art_Wanted].accepted,ipf->counts[art_Unsolicited].accepted + RESULT_COUNTS(RCI_NOTHING, RCI_TRIPLE_VALS) + ); +} + +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_flushing_done(void) { + InputFile *ipf= flushing_input_file; + if (!inputfile_is_done(ipf)) return; - assert(ipf == old_input_file); assert(sms==sm_SEPARATED || sms==sm_DROPPING); - notice_processed(ipf,"feed file",0); + notice_processed(ipf,"feedfile",0); close_defer(); - if (unlink(path_duct)) - sysdie("could not unlink old duct file %s", path_duct); + xunlink(path_flushing, "old flushing file"); - if (sms==sm_DROPPING) { - notice("feed dropped and our work is complete" - " (but check for backlog files)"); - r= unlink(path_ductlock); - if (r) sysdie("unlink lock file for old feed %s", path_ductlock); - exit(0); - } + close_input_file(flushing_input_file); + free(flushing_input_file); + flushing_input_file= 0; - open_defer(); + if (sms==sm_SEPARATED) { + notice("flush complete"); + SMS(NORMAL, 0, "flush complete"); + } else if (sms==sm_DROPPING) { + SMS(DROPPED, 0, "old flush complete"); + search_backlog_file(); + notice("feed dropped, but will continue until backlog is finished"); + } +} - close_input_file(old_input_file); - old_input_file= 0; +static void *statemc_check_input_done(oop_source *lp, struct timeval now, + void *u) { + assert(!inputfile_is_done(main_input_file)); + statemc_check_flushing_done(); + statemc_check_backlog_done(); + return OOP_CONTINUE; +} - notice("flush complete"); - SMS(NORMAL, 0, "flush complete"); +static void queue_check_input_done(void) { + loop->on_time(loop, OOP_TIME_NOW, statemc_check_input_done, 0); } static void statemc_setstate(StateMachineState newsms, int periods, const char *forlog, const char *why) { sms= newsms; sm_period_counter= periods; + + const char *xtra= ""; + switch (sms) { + case sm_FLUSHING: sm_FLUSHFAILED: + if (!main_input_file) xtra= "-ABSENT"; + break; + case sm_SEPARATED: case sm_DROPPING: + xtra= flushing_input_file->fd >= 0 ? "-1" : "-2"; + break; + default:; + } + if (periods) { - info("%s[%d] %s",periods,forlog,why); + info("%s%s[%d] %s",forlog,xtra,periods,why); } else { - info("%s %s",forlog,why); + info("%s%s %s",forlog,xtra,why); } } @@ -1651,15 +1873,15 @@ static void open_defer(void) { if (defer) return; - defer= fopen(path_ductdefer, "a+"); - if (!defer) sysfatal("could not open defer file %s", path_ductdefer); + defer= fopen(path_defer, "a+"); + if (!defer) sysfatal("could not open defer file %s", path_defer); /* truncate away any half-written records */ xfstat_isreg(fileno(defer), &stab, "newly opened defer file"); if (stab.st_size > LONG_MAX) - die("defer file %s size is far too large", path_ductdefer); + die("defer file %s size is far too large", path_defer); if (!stab.st_size) return; @@ -1669,14 +1891,14 @@ static void open_defer(void) { for (;;) { if (!truncto) break; /* was only (if anything) one half-truncated record */ if (fseek(defer, truncto-1, SEEK_SET) < 0) - sysdie("seek in defer file %s while truncating partial", path_ductdefer); + sysdie("seek in defer file %s while truncating partial", path_defer); r= getc(defer); if (r==EOF) { if (ferror(defer)) - sysdie("failed read from defer file %s", path_ductdefer); + sysdie("failed read from defer file %s", path_defer); else - die("defer file %s shrank while we were checking it!", path_ductdefer); + die("defer file %s shrank while we were checking it!", path_defer); } if (r=='\n') break; truncto--; @@ -1685,19 +1907,19 @@ static void open_defer(void) { if (stab.st_size != truncto) { warn("truncating half-record at end of defer file %s -" " shrinking by %ld bytes from %ld to %ld", - path_ductdefer, orgsize - truncto, orgsize, truncto); + path_defer, orgsize - truncto, orgsize, truncto); if (fflush(defer)) - sysfatal("could not flush defer file %s", path_ductdefer); + sysfatal("could not flush defer file %s", path_defer); if (ftruncate(fileno(defer), truncto)) - sysdie("could not truncate defer file %s", path_ductdefer); + sysdie("could not truncate defer file %s", path_defer); } else { info("continuing existing defer file %s (%ld bytes)", - path_ductdefer, orgsize); + path_defer, orgsize); } if (fseek(defer, truncto, SEEK_SET)) - sysdie("could not seek to new end of defer file %s", path_ductdefer); + sysdie("could not seek to new end of defer file %s", path_defer); } static void close_defer(void) { @@ -1719,25 +1941,29 @@ static void close_defer(void) { 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; + 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 void search_backlog_file(void) { + /* returns non-0 iff there are any backlog files */ + glob_t gl; int r; struct stat stab; const char *oldest_path=0; time_t oldest_mtime, now; - assert(!backlog_input_file); + if (backlog_input_file) return 3; + + try_again: r= glob(globpat_backlog, GLOB_ERR|GLOB_MARK|GLOB_NOSORT, 0, &gl); @@ -1749,6 +1975,11 @@ static void search_backlog_file(void) { case 0: for (i=0; i= 0 && - backlog_nextscan_periods > backlog_spontaneous_rescan_periods) - backlog_nextscan_periods= backlog_spontaneous_rescan_periods; + until_backlog_nextscan > backlog_spontaneous_rescan_periods) + until_backlog_nextscan= backlog_spontaneous_rescan_periods; debug("backlog scan: young age=%f deficiency=%ld nextscan=%d oldest=%s", - age, age_deficiency, backlog_nextscan_periods, oldest_path); + age, age_deficiency, until_backlog_nextscan, oldest_path); + return 2; } /*========== flushing the feed ==========*/ @@ -1812,7 +2054,7 @@ static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) { loop->cancel_fd(fd); close(fd); - assert(!old_input_file); + assert(!flushing_input_file); if (WIFEXITED(status)) { switch (WEXITSTATUS(status)) { @@ -1822,17 +2064,38 @@ 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: @@ -1848,7 +2111,7 @@ 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"); } static void inndcommfail(const char *what) { @@ -1856,10 +2119,12 @@ static void inndcommfail(const char *what) { exit(INNDCOMMCHILD_ESTATUS_FAIL); } -void spawn_inndcomm_flush(void) { +void spawn_inndcomm_flush(const char *why) { /* Moved => Flushing */ int pipefds[2]; - assert(sms==sm_NORMAL || sms==sm_FLUSHFAIL); + notice("flushing %s",why); + + assert(sms==sm_NORMAL || sms==sm_FLUSHFAILED); assert(!inndcomm_child); if (pipe(pipefds)) sysdie("create pipe for inndcomm child sentinel"); @@ -1886,7 +2151,7 @@ void spawn_inndcomm_flush(void) { int sentinel_fd= pipefds[0]; on_fd_read_except(sentinel_fd, inndcomm_event); - SMS(FLUSHING, 0, "flush is in progress"); + SMS(FLUSHING, 0, why); } /*========== main program ==========*/ @@ -1915,7 +2180,7 @@ static void postfork(const char *what) { sysdie("%s child: failed to reset SIGPIPE"); postfork_inputfile(main_input_file); - postfork_inputfile(old_input_file); + postfork_inputfile(flushing_input_file); postfork_conns(idle.head); postfork_conns(working.head); postfork_conns(full.head); @@ -1938,11 +2203,34 @@ EVERY(filepoll, {5,0}, { filemon_callback(main_input_file); }); +#define DEBUGF_IPF(wh) " " #wh "=%p/%s:ip=%ld,off=%ld,fd=%d%s" \ +#define DEBUG_IPF(sh) \ + wh##_input_file, debug_ipf_path(wh##_input_file), \ + wh##_input_file->inprogress, (long)wh##_input_file->offset, \ + wh##_input_file->fd, wh##_input_file->rd ? "+" : "" +static const char *debug_ipf_path(InputFile *ipf) { + char *slash= strrchr(ipf->path,'/'); + return slash ? slash+1 : ipf->path; +} + EVERY(period, {PERIOD_SECONDS,0}, { - if (connect_delay) connect_delay--; + debug("PERIOD" + " sms=%s[%d] queue=%d until_connect=%d" + " input_files" DEBUGF_IPF(main) DEBUGF_IPF(old) DEBUGF_FMT(flushing) + " conns idle=%d working=%d full=%d" + " children connecting=%ld inndcomm_child" + , + sms_names[sms], sm_period_counter, queue.count, until_connect, + DEBUG_IPF(main), DEBUG_IPF(flushing), DEBUG_IPF(flushing), + idle.count, working.count, full.count, + (long)connecting_child, (long)inndcomm_child + ); + + if (until_connect) until_connect--; + poll_backlog_file(); if (!backlog_input_file) close_defer(); /* want to start on a new backlog */ - statemc_poll(); + statemc_period_poll(); check_master_queue(); }); @@ -2026,25 +2314,25 @@ void op_periods_boolfalse(const Option *o, const char *val) { } static const Option options[]= { +{'f',"feedfile", &feedfile, op_string }, +{'q',"quiet-multiple", &quiet_multiple, op_booltrue, 1 }, + { 0, "max-connections", &max_connections op_integer }, +{ 0, "max-queue-per-conn", &max_queue_per_conn op_integer }, + + { 0, "streaming", &try_stream, op_booltrue, 1 }, { 0, "no-streaming", &try_stream, op_boolfalse, 1 }, -{'h',"host", &remote_host, op_string }, {'P',"port", &port op_integer }, { 0, "inndconf", &inndconffile, op_string }, -{'f',"feedfile", &feedfile, op_string }, -{'q',"quiet-multiple", &quiet_if_locked, op_booltrue, 1 }, -{ 0, "no-quiet-multiple", &quiet_if_locked, op_boolfalse, 1 }, {'d',"daemon", &become_daemon, op_booltrue, 1 }, { 0, "no-daemon", &become_daemon, op_boolfalse, 1 }, { 0, "no-check-proportion", &nocheck_thresh_pct, op_double }, { 0, "no-check-filter", &nocheck_decay_articles, op_double }, -{ 0, "max-queue-size", &max_queue_per_conn op_integer }, { 0, "reconnect-interval", &reconnect_delay_periods, op_periods_rndup }, { 0, "flush-retry-interval", &flushfail_retry_periods, op_periods_rndup }, -{ 0, "feedfile-open-timeout", &open_wait_periods, op_periods_rndup }, { 0, "connection-timeout", &connection_timeout, op_seconds }, { 0, "inndcomm-timeout", &inndcomm_flush_timeout, op_seconds }, }; @@ -2100,9 +2388,13 @@ int main(int argc, char **argv) { } if (!arg) badusage("need site name argument"); - if (*++argv) badusage("too many non-option arguments"); sitename= arg; + if ((arg= *++argv)) + remote_host= arg; + + if (*++argv) badusage("too many non-option arguments"); + if (nocheck_thresh_pct < 0 || nocheck_thresh_pct > 100) badusage("nocheck threshold percentage must be between 0..100"); nocheck_thresh= nocheck_thresh_pct * 0.01; @@ -2111,6 +2403,8 @@ int main(int argc, char **argv) { badusage("nocheck decay articles must be at least 0.1"); nocheck_decay= 1 - 1/nocheck_decay_articles; + if (!pathoutgoing) + pathoutgoing= innconf->pathoutgoing; innconf_read(inndconffile); if (!feedfile) @@ -2120,11 +2414,14 @@ int main(int argc, char **argv) { else if (feedfile[strlen(feedfile)-1]=='/') feedfile= xasprintf("%s%s",feedfile,sitename); - const char *feedfile_forbidden= "?*["; + const char *feedfile_forbidden= "?*[~#"; int c; while ((c= *feedfile_forbidden++)) if (strchr(feedfile, c)) - badusage("feed filename may not contain glob metacharacter %c",c); + badusage("feed filename may not contain metacharacter %c",c); + + loop= oop_sys_new(); + if (!loop) sysdie("could not create liboop event loop"); if (signal(SIGPIPE, SIG_IGN) == SIG_ERR) sysdie("could not ignore SIGPIPE");