X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ian/git?p=inn-innduct.git;a=blobdiff_plain;f=backends%2Finnduct.c;h=0258ad022de0ce6b17341679f0449c7fb43745be;hp=a5bb448e3e81add5319fb0d917918c85ce4f6b6f;hb=c2c88becbf08aa2b088ee42e6ed3331f733cbaaf;hpb=8b7074062790fe3362d1fa7ca28bf68d84820e03 diff --git a/backends/innduct.c b/backends/innduct.c index a5bb448..0258ad0 100644 --- a/backends/innduct.c +++ b/backends/innduct.c @@ -1,3 +1,11 @@ +/* + * TODO + * - close idle connections + * - cope better with garbage in feed file + * - cope better with NULs in feed file + * - -k kill mode ? + */ + /* * Newsfeeds file entries should look like this: * host.name.of.site[/exclude,exclude,...]\ @@ -149,9 +157,11 @@ perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct. #define _GNU_SOURCE +#include "inn/list.h" #include "config.h" #include "storage.h" #include "nntp.h" +#include "libinn.h" #include #include @@ -165,6 +175,9 @@ perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct. #include #include #include +#include +#include +#include #include #include @@ -179,31 +192,64 @@ perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct. #define INNDCOMMCHILD_ESTATUS_FAIL 6 #define INNDCOMMCHILD_ESTATUS_NONESUCH 7 +#define MAX_LINE_FEEDFILE (NNTP_MSGID_MAXLEN + sizeof(TOKEN)*2 + 10) + /*----- doubly linked lists -----*/ -#define ISNODE(T) struct { T *succ, *pred; } node -#define DEFLIST(T) typedef struct { T *head, *tail, *tp; int count; } T##List +#define ISNODE(T) struct { T *succ, *pred; } node /* must be at start */ +#define DEFLIST(T) typedef struct { T *hd, *tl, *tp; int count; } T##List + +#define NODE(n) (assert((void*)&(n)->node == &(n)), \ + (struct node*)&(n)->node) + +#define LIST_CHECKCANHAVENODE(l,n) \ + ((void)((n) == ((l).hd))) /* just for the type check */ -#define NODE(n) ((struct node*)&(n)->node) +#define LIST_ADDSOMEHOW(l,n,list_addsomehow) \ + ( LIST_CHECKCANHAVENODE(l,n), \ + list_addsomehow((struct list*)&(l), NODE((n))), \ + (void)(l).count++ \ + ) -#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_REMSOMEHOW(l,list_remsomehow) \ + ( (typeof((l).hd)) \ + ( (l).count \ + ? ( (l).count--, \ + list_remsomehow((struct list*)&(l)) ) \ + : 0 \ + ) \ + ) + + +#define LIST_ADDHEAD(l,n) LIST_ADDSOMEHOW((l),(n),list_addhead) +#define LIST_ADDTAIL(l,n) LIST_ADDSOMEHOW((l),(n),list_addtail) +#define LIST_REMHEAD(l) LIST_REMSOMEHOW((l),list_remhead) +#define LIST_REMTAIL(l) LIST_REMSOMEHOW((l),list_remtail) + +#define LIST_HEAD(l) ((typeof((l).hd))(list_head((struct list*)&(l)))) +#define LIST_NEXT(n) ((typeof(n))list_succ(NODE((n)))) +#define LIST_BACK(n) ((typeof(n))list_pred(NODE((n)))) -#define LIST_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); @@ -211,16 +257,37 @@ DEFLIST(Article); /*----- function predeclarations -----*/ -static void conn_check_work(Conn *conn); +static void conn_maybe_write(Conn *conn); +static void conn_make_some_xmits(Conn *conn); +static void *conn_write_some_xmits(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 xmit_free(XmitDetails *d); -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 statemc_start_flush(const char *why); /* Normal => Flushing */ +static void spawn_inndcomm_flush(const char *why); /* Moved => Flushing */ + static void check_master_queue(void); +static void queue_check_input_done(void); + +static void statemc_check_flushing_done(void); +static void statemc_check_backlog_done(void); + +static void postfork(const char *what); +static void postfork_inputfile(InputFile *ipf); + +static void open_defer(void); +static void close_defer(void); +static void search_backlog_file(void); + +static void inputfile_tailing_start(InputFile *ipf); +static void inputfile_tailing_stop(InputFile *ipf); + +static int filemon_init(void); +static void filemon_start(InputFile *ipf); +static void filemon_stop(InputFile *ipf); +static void filemon_callback(InputFile *ipf); /*----- configuration options -----*/ @@ -229,6 +296,12 @@ static const char *remote_host; static int quiet_multiple=0, become_daemon=1; static int max_connections=10, max_queue_per_conn=200; +static int target_max_feedfile_size=100000; + +static double max_bad_data_ratio= 0.01; +static int max_bad_data_initial= 30; + /* in one corrupt 4096-byte block the number of newlines has + * mean 16 and standard deviation 3.99. 30 corresponds to z=+3.5 */ static int connection_setup_timeout=200, port=119, try_stream=1; static int inndcomm_flush_timeout=100; @@ -263,7 +336,7 @@ typedef enum { /* in queue in conn->sent */ #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_Unchecked] x \ + counts[art_Wanted] x \ + counts[art_Unsolicited] x, \ counts[art_Unchecked] x \ @@ -271,7 +344,7 @@ typedef enum { /* in queue in conn->sent */ , counts[art_Unsolicited] x typedef enum { -#define RC_INDEX(x) RCI_##x, +#define RC_INDEX(x) RC_##x, RESULT_COUNTS(RC_INDEX, RC_INDEX) RCI_max } ResultCountIndex; @@ -285,33 +358,35 @@ typedef enum { 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 -----*/ -typedef struct InputFile { - /* This is an instance of struct oop_readable */ +struct InputFile { + /* This is also an instance of struct oop_readable */ struct oop_readable readable; /* first */ oop_readable_call *readable_callback; void *readable_callback_user; int fd; - struct Filemon_Perfile *filemon; + Filemon_Perfile *filemon; oop_read *rd; long inprogress; /* no. of articles read but not processed */ off_t offset; + int skippinglong; int counts[art_MaxState][RCI_max]; + int readcount_ok, readcount_blank, readcount_err; char path[]; -} InputFile; +}; struct Article { ISNODE(Article); @@ -345,8 +420,9 @@ static const char *sms_names[]= { struct Conn { ISNODE(Conn); - int fd, max_queue, stream; - ArticleList queue; /* not yet told peer, or CHECK said send it */ + int fd, max_queue, stream, quitting; + ArticleList waiting; /* not yet told peer */ + ArticleList priority; /* peer says send it now */ ArticleList sent; /* offered/transmitted - in xmit or waiting reply */ struct iovec xmit[CONNIOVS]; XmitDetails xmitd[CONNIOVS]; @@ -358,11 +434,11 @@ struct Conn { static oop_source *loop; -static int nconns, until_connect; -static ConnList idle, working, full; +static int until_connect; +static ConnList conns; static ArticleList queue; -static char *path_lock, *path_flushing, *path_defer; +static char *path_lock, *path_flushing, *path_defer, *globpat_backlog; #define SMS(newstate, periods, why) \ (statemc_setstate(sm_##newstate,(periods),#newstate,(why))) @@ -394,7 +470,7 @@ static void logv(int sysloglevel, const char *pfx, int errnoval, __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]; + char msgbuf[256]; /* NB do not call xvasprintf here or you'll recurse */ vsnprintf(msgbuf,sizeof(msgbuf), fmt,al); msgbuf[sizeof(msgbuf)-1]= 0; @@ -432,15 +508,26 @@ logwrap(debug, " debug", LOG_DEBUG, -1, 0); /*========== 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 char *xvasprintf(const char *fmt, va_list al) + __attribute__((__format__(printf,1,0))); +static char *xvasprintf(const char *fmt, va_list al) { + char *str; + int rc= vasprintf(&str,fmt,al); + if (rc<0) sysdie("vasprintf(\"%s\",...) failed", fmt); + return str; +} +static char *xasprintf(const char *fmt, ...) + __attribute__((__format__(printf,1,2))); +static char *xasprintf(const char *fmt, ...) { + va_list al; + va_start(al,fmt); + char *str= xvasprintf(fmt,al); + va_end(al); + return str; } +static void perhaps_close(int *fd) { if (*fd) { close(*fd); fd=0; } } + static pid_t xfork(const char *what) { pid_t child; @@ -506,7 +593,7 @@ 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(fd, stab_r); + int r= fstat(fd, stab_r); if (r) sysdie("could not fstat %s", what); } @@ -542,14 +629,38 @@ static int samefile(const struct stat *a, const struct stat *b) { a->st_dev == b->st_dev); } +static char *sanitise(const char *input) { + static char sanibuf[100]; /* returns pointer to this buffer! */ + + 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; +} + /*========== making new connections ==========*/ +static void conn_dispose(Conn *conn) { + if (!conn) return; + perhaps_close(&conn->fd); + free(conn); + until_connect= reconnect_delay_periods; +} + static int connecting_sockets[2]= {-1,-1}; static pid_t connecting_child; static void connect_attempt_discard(void) { if (connecting_sockets[0]) - cancel_fd(connecting_sockets[0]); + cancel_fd_read_except(connecting_sockets[0]); perhaps_close(&connecting_sockets[0]); perhaps_close(&connecting_sockets[1]); @@ -640,23 +751,26 @@ static void *connchild_event(oop_source *lp, int fd, oop_event e, void *u) { fatal("connect: child gave unexpected exit status %d", es); } - setnonblocking(conn->fd, 1); - /* Phew! */ - LIST_ADDHEAD(idle, conn); - notice("#%d connected %s", conn->fd, conn->stream ? "streaming" : "plain"); + setnonblock(conn->fd, 1); + conn->max_queue= conn->stream ? max_queue_per_conn : 1; + LIST_ADDHEAD(conns, conn); + notice("C%d connected %s", conn->fd, conn->stream ? "streaming" : "plain"); connect_attempt_discard(); check_master_queue(); return 0; x: - if (conn) { - perhaps_close(&conn->fd); - free(conn); - } + conn_dispose(conn); connect_attempt_discard(); } +static int allow_connect_start(void) { + return conns.count < max_connections + && !connecting_child + && !until_connect; +} + static void connect_start(void) { assert(!connecting_sockets[0]); assert(!connecting_sockets[1]); @@ -678,15 +792,11 @@ static void connect_start(void) { if (r) sysdie("connect: close parent socket in child"); alarm(connection_setup_timeout); - if (NNTPconnect(remote_host, port, &cn_from, &cn_to, buf) < 0) { - if (buf[0]) { - sanitise_inplace(buf); - fatal("connect: rejected: %s", buf); - } else { - sysfatal("connect: connection attempt failed"); - } + if (NNTPconnect((char*)remote_host, port, &cn_from, &cn_to, buf) < 0) { + if (buf[0]) fatal("connect: rejected: %s", sanitise(buf)); + else sysfatal("connect: connection attempt failed"); } - if (NNTPsendpassword(remote_host, cn_from, cn_to) < 0) + if (NNTPsendpassword((char*)remote_host, cn_from, cn_to) < 0) sysfatal("connect: authentication failed"); if (try_stream) { if (fputs("MODE STREAM\r\n", cn_to) || @@ -701,19 +811,16 @@ static void connect_start(void) { } 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); - } + remote_host, sanitise(buf)); l--; if (l>0 && buf[l-1]=='\r') l--; buf[l]= 0; char *ep; int rcode= strtoul(buf,&ep,10); - if (ep != &buf[3]) { - sanitise_inplace(buf); - fatal("connect: bad response to MODE STREAM: %.50s", buf); - } + if (ep != &buf[3]) + fatal("connect: bad response to MODE STREAM: %.50s", sanitise(buf)); + switch (rcode) { case 203: exitstatus= CONNCHILD_ESTATUS_STREAM; @@ -722,8 +829,8 @@ static void connect_start(void) { 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; } @@ -758,60 +865,51 @@ static void connect_start(void) { /*========== overall control of article flow ==========*/ static void check_master_queue(void) { - if (!queue.count) - return; - - Conn *last_assigned=0; for (;;) { - if (working.head) { - conn_assign_one_article(&working, &last_assigned); - } else if (idle.head) { - conn_assign_one_article(&idle, &last_assigned); - } else if (nconns < max_connections && queue.count >= max_queue_per_conn && - !connecting_child && !until_connect) { + if (!queue.count) + break; + + Conn *walk, *use=0; + int spare; + + /* Find a connection to offer this article. We prefer a busy + * connection to an idle one, provided it's not full. We take the + * first (oldest) and since that's stable, it will mean we fill up + * connections in order. That way if we have too many + * connections, the spare ones will go away eventually. + */ + for (walk=LIST_HEAD(conns); walk; walk=LIST_NEXT(walk)) { + int inqueue= walk->sent.count + walk->priority.count + + walk->waiting.count; + spare= walk->max_queue - inqueue; + assert(inqueue <= max_queue_per_conn); + assert(spare >= 0); + if (inqueue==0) /*idle*/ { if (!use) use= walk; } + else if (spare>0) /*working*/ { use= walk; break; } + } + if (use) { + while (spare>0) { + Article *art= LIST_REMHEAD(queue); + LIST_ADDTAIL(use->waiting, art); + spare--; + } + conn_maybe_write(use); + } else if (allow_connect_start()) { until_connect= reconnect_delay_periods; connect_start(); + break; } else { break; } } - conn_check_work(last_assigned); -} - -static void conn_assign_one_article(ConnList *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 ConnList *conn_determine_right_list(Conn *conn) { - int inqueue= conn_total_queued_articles(conn); - assert(inqueue <= max_queue); - if (inqueue == 0) return &idle; - if (inqueue == conn->max_queue) return &full; - return &working; -} - -static void *conn_writeable(oop_source *l, int fd, int ev, void *u) { - check_conn_work(u); +static void *conn_writeable(oop_source *l, int fd, oop_event ev, void *u) { + conn_maybe_write(u); return OOP_CONTINUE; } -static void conn_check_work(Conn *conn) { +static void conn_maybe_write(Conn *conn) { void *rp= 0; for (;;) { conn_make_some_xmits(conn); @@ -835,42 +933,41 @@ static void conn_check_work(Conn *conn) { } static void vconnfail(Conn *conn, const char *fmt, va_list al) - __attribute__((printf,2,0)); + __attribute__((__format__(printf,2,0))); static void vconnfail(Conn *conn, const char *fmt, va_list al) { int requeue[art_MaxState]; Article *art; - while ((art= LIST_REMHEAD(conn->queue))) LIST_ADDTAIL(queue); + 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))) { - counts[art->state]++; + requeue[art->state]++; if (art->state==art_Unsolicited) art->state= art_Unchecked; - LIST_ADDTAIL(queue); + LIST_ADDTAIL(queue,art); } int i; - XmitDetails *xd; - for (i=0, dp=&conn->xmitd; ixmitu; i++, dp++) - xmit_free(dp); + XmitDetails *d; + for (i=0, d=conn->xmitd; ixmitu; i++, d++) + xmit_free(d); char *m= xvasprintf(fmt,al); - warn("#%d connection failed, requeueing " RCI_TRIPLE_FMT_BASE ": %s", - conn->fd, RCI_TRIPLE_FMT_VALS(requeue, /*nothing*/), m); + warn("C%d connection failed (requeueing " RCI_TRIPLE_FMT_BASE "): %s", + conn->fd, RCI_TRIPLE_VALS_BASE(requeue, /*nothing*/), m); free(m); - close(conn->fd); - free(conn); - - until_connect= reconnect_delay_periods; + LIST_REMOVE(conns,conn); + conn_dispose(conn); check_master_queue(); } -static void connfail(Connection *conn, const char *fmt, ...) - __attribute__((printf,2,3)); -static void connfail(Connection *conn, const char *fmt, ...) { +static void connfail(Conn *conn, const char *fmt, ...) + __attribute__((__format__(printf,2,3))); +static void connfail(Conn *conn, const char *fmt, ...) { va_list al; va_start(al,fmt); - vconnfail(fmt,al); + vconnfail(conn,fmt,al); va_end(al); } @@ -880,7 +977,7 @@ static XmitDetails *xmit_core(Conn *conn, const char *data, int len, 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; @@ -892,7 +989,7 @@ 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; } @@ -924,7 +1021,8 @@ static void *conn_write_some_xmits(Conn *conn) { } 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) { @@ -947,18 +1045,19 @@ 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->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); } @@ -974,8 +1073,8 @@ static void conn_make_some_xmits(Conn *conn) { art->state= art->state == art_Unchecked ? art_Unsolicited : art->state == art_Wanted ? art_Wanted : - abort(); - art->ipf->counts[art->state].sent++; + (abort(),-1); + art->ipf->counts[art->state][RC_sent]++; LIST_ADDTAIL(conn->sent, art); } else { @@ -985,11 +1084,11 @@ 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].sent++; + art->ipf->counts[art->state][RC_sent]++; LIST_ADDTAIL(conn->sent, art); } } @@ -1012,12 +1111,12 @@ 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= conn->sent.head; + Article *art= LIST_HEAD(conn->sent); if (!art) { connfail(conn, @@ -1029,20 +1128,20 @@ 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); @@ -1050,19 +1149,19 @@ 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("peer says article accepted but we had not sent the body: %s", - sanitised_response); + 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("peer says please sent the article but we just did: %s", + connfail(conn, "peer says please sent the article but we just did: %s", sanitised_response); return 0; } @@ -1073,8 +1172,8 @@ static Article *article_reply_check(Connection *conn, const char *response, } static void update_nocheck(int accepted) { - accept_proportion *= accept_decay; - accept_proportion += accepted; + 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"); @@ -1085,7 +1184,7 @@ static void update_nocheck(int accepted) { nocheck= new_nocheck; } -static void article_done(Connection *conn, Article *art, int 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); @@ -1129,32 +1228,28 @@ 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\n"); + assert(!conn->waiting.count); + assert(!conn->priority.count); + assert(!conn->sent.count); + assert(!conn->xmitu); + LIST_REMOVE(conns,conn); + conn_dispose(conn); } - conn close ok; - return; + return OOP_CONTINUE; } Article *art; @@ -1164,7 +1259,7 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev, if (art) ; else return OOP_CONTINUE /* reply_check has failed the conn */ #define ARTICLE_DEALTWITH(streaming,musthavesent,how) \ - code_streaming= (streaming) \ + code_streaming= (streaming); \ GET_ARTICLE(musthavesent); \ article_done(conn, art, RC_##how); break; @@ -1192,9 +1287,9 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev, case 335: /* IHAVE says send it */ GET_ARTICLE(-1); assert(art->state == art_Unchecked); - art->ipf->counts[art->state].accepted++; + art->ipf->counts[art->state][RC_accepted]++; art->state= art_Wanted; - LIST_ADDTAIL(conn->queue); + LIST_ADDTAIL(conn->priority, art); break; case 431: /* CHECK or TAKETHIS says try later */ @@ -1210,7 +1305,8 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev, } - check_check_work(conn); + conn_maybe_write(conn); + check_master_queue(); return OOP_CONTINUE; } @@ -1246,10 +1342,6 @@ static InputFile *open_input_file(const char *path) { InputFile *ipf= xmalloc(sizeof(*ipf) + strlen(path) + 1); memset(ipf,0,sizeof(*ipf)); - ipf->readable.on_readable= tailing_on_readable; - ipf->readable.on_cancel= tailing_on_cancel; - ipf->readable.try_read= tailing_try_read; - ipf->fd= fd; strcpy(ipf->path, path); @@ -1269,97 +1361,110 @@ static void close_input_file(InputFile *ipf) { /*---------- dealing with articles read in the input file ----------*/ -typedef void *feedfile_got_article(oop_source *lp, oop_read *rd, - oop_rd_event ev, const char *errmsg, - int errnoval, - const char *data, size_t recsz, - void *ipf_v) { +static void *feedfile_got_bad_data(InputFile *ipf, off_t offset, + const char *data, const char *how) { + warn("corrupted file: %s, offset %lu: %s: %s", + ipf->path, (unsigned long)offset, how, sanitise(data)); + ipf->readcount_err++; + if (ipf->readcount_err > max_bad_data_initial + + (ipf->readcount_ok+ipf->readcount_blank) / max_bad_data_ratio) + die("too much garbage in input file! (%d errs, %d ok, %d blank)", + ipf->readcount_err, ipf->readcount_ok, ipf->readcount_blank); + return OOP_CONTINUE; +} + +static void *feedfile_read_err(oop_source *lp, oop_read *rd, + oop_rd_event ev, const char *errmsg, + int errnoval, const char *data, size_t recsz, + void *ipf_v) { + InputFile *ipf= ipf_v; + assert(ev == OOP_RD_SYSTEM); + errno= errnoval; + sysdie("error reading input file: %s, offset %lu", + ipf->path, (unsigned long)ipf->offset); +} + +static void *feedfile_got_article(oop_source *lp, oop_read *rd, + oop_rd_event ev, const char *errmsg, + int errnoval, const char *data, size_t recsz, + void *ipf_v) { InputFile *ipf= ipf_v; Article *art; char tokentextbuf[sizeof(TOKEN)*2+3]; if (!data) { feedfile_eof(ipf); return OOP_CONTINUE; } - if (data[0] && data[0]!=' ') { - char *space= strchr(data,' '); - int tokenlen= space-data; - int midlen= (int)recsz-tokenlen-1; - if (midlen < 0) goto bad_data; - - if (tokenlen != sizeof(TOKEN)*2+2) goto bad_data; - memcpy(tokentextbuf, data, tokenlen); - tokentextbuf[tokenlen]= 0; - if (!IsToken(tokentextbuf)) goto bad_data; - - art= xmalloc(sizeof(*art) - 1 + midlen + 1); - art->offset= ipf->offset; - art->blanklen= recsz; - art->midlen= midlen; - art->state= art_Unchecked; - 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==main_input_file && - ipf->offset >= flush_threshold) - statemc_start_flush("feed file size"); +#define X_BAD_DATA(m) return feedfile_got_bad_data(ipf,old_offset,data,m); - check_master_queue(); -} + if (ev==OOP_RD_PARTREC) + feedfile_got_bad_data(ipf,old_offset,data,"missing final newline"); + /* but process it anyway */ -static void statemc_start_flush(const char *why) { /* Normal => Flushing */ - assert(sms == sm_NORMAL); + if (ipf->skippinglong) { + if (ev==OOP_RD_OK) ipf->skippinglong= 0; /* fine now */ + return; + } + if (ev==OOP_RD_LONG) { + ipf->skippinglong= 1; + X_BAD_DATA("overly long line"); + } - debug("starting flush (%s) (%lu >= %lu) (%d)", - why, - (unsigned long)ipf->offset, (unsigned long)flush_threshold, - sm_period_counter); + if (memchr(data,'\0',recsz)) X_BAD_DATA("nul byte"); + if (!recsz) X_BAD_DATA("empty line"); - int r= link(feedfile, duct_path); - if (r) sysdie("link feedfile %s to flushing file %s", feedfile, - path_duct); - /* => Hardlinked */ + if (data[0]==' ') { + if (strspn(data," ") != recsz) X_BAD_DATA("line partially blanked"); + ipf->readcount_blank++; + return OOP_CONTINUE; + } + + char *space= strchr(data,' '); + int tokenlen= space-data; + int midlen= (int)recsz-tokenlen-1; + if (midlen <= 2) X_BAD_DATA("no room for messageid"); + if (space[1]!='<' || space[midlen]!='>') X_BAD_DATA("invalid messageid"); + + if (tokenlen != sizeof(TOKEN)*2+2) X_BAD_DATA("token wrong length"); + memcpy(tokentextbuf, data, tokenlen); + tokentextbuf[tokenlen]= 0; + if (!IsToken(tokentextbuf)) X_BAD_DATA("token wrong syntax"); + + ipf->readcount_ok++; + + art= xmalloc(sizeof(*art) - 1 + midlen + 1); + art->offset= ipf->offset; + art->blanklen= recsz; + art->midlen= midlen; + art->state= art_Unchecked; + art->ipf= ipf; ipf->inprogress++; + art->token= TextToToken(tokentextbuf); + strcpy(art->messageid, space+1); + LIST_ADDTAIL(queue, art); - xunlink(feedfile, "old feedfile link"); - /* => Moved */ + if (sms==sm_NORMAL && ipf==main_input_file && + ipf->offset >= target_max_feedfile_size) + statemc_start_flush("feed file size"); - spawn_inndcomm_flush(why); /* => Flushing FLUSHING */ + check_master_queue(); + return OOP_CONTINUE; } /*========== tailing input file ==========*/ -static void filemon_start(InputFile *ipf) { - assert(!ipf->filemon); - - ipf->filemon= xmalloc(sizeof(*ipf->filemon)); - memset(ipf->filemon, 0, sizeof(*ipf->filemon)); - filemon_method_startfile(ipf, ipf->filemon); -} - -static void filemon_stop(InputFile *ipf) { - if (!ipf->filemon) return; - filemon_method_stopfile(ipf, ipf->filemon); - free(ipf->filemon); - ipf->filemon= 0; -} - -static void filemon_callback(InputFile *ipf) { - ipf->readable_callback(ipf->readable_callback_user); -} - static void *tailing_rable_call_time(oop_source *loop, struct timeval tv, void *user) { InputFile *ipf= user; - return ipf->readable_callback(ipf->readable_callback_user); + return ipf->readable_callback(loop, &ipf->readable, + ipf->readable_callback_user); } -static void on_cancel(struct oop_readable *rable) { +static void tailing_on_cancel(struct oop_readable *rable) { InputFile *ipf= (void*)rable; - if (ipf->filemon) filemon_stopfile(ipf); + if (ipf->filemon) filemon_stop(ipf); loop->cancel_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf); ipf->readable_callback= 0; } @@ -1377,7 +1482,7 @@ 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); tailing_queue_readable(ipf); return 0; @@ -1420,9 +1525,9 @@ 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); @@ -1487,13 +1592,13 @@ static int filemon_method_init(void) { 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) { } @@ -1501,21 +1606,42 @@ 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; + 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; @@ -1524,7 +1650,7 @@ static void inputfile_tailing_start(InputFile *ipf) { assert(ipf->fd); int r= oop_rd_read(ipf->rd, &feedfile_rdstyle, MAX_LINE_FEEDFILE, - feedfile_got_article,ipf, feedfile_problem,ipf); + feedfile_got_article,ipf, feedfile_read_err, ipf); if (r) sysdie("unable start reading feedfile %s",ipf->path); } @@ -1638,6 +1764,12 @@ static void inputfile_tailing_stop(InputFile *ipf) { * ->8- */ +static void startup_set_input_file(InputFile *f) { + assert(!main_input_file); + main_input_file= f; + inputfile_tailing_start(f); +} + static void statemc_init(void) { struct stat stab, stabf; @@ -1654,7 +1786,7 @@ 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_multiple) exit(0); @@ -1663,7 +1795,7 @@ static void statemc_init(void) { sysdie("fcntl F_SETLK lockfile %s", path_lock); } - xfstat_isreg(lockfd, &stabf, "lockfile"); + xfstat_isreg(lockfd, &stabf, path_lock, "lockfile"); int lock_noent; xlstat_isreg(path_lock, &stab, &lock_noent, "lockfile"); @@ -1701,7 +1833,7 @@ static void statemc_init(void) { int noent_f; InputFile *file_d= open_input_file(path_flushing); - if (file_d) xfstat_isreg(file_d->fd, &stab_d, "flushing file"); + if (file_d) xfstat_isreg(file_d->fd, &stab_d, path_flushing,"flushing file"); xlstat_isreg(feedfile, &stab_f, &noent_f, "feedfile"); @@ -1723,7 +1855,7 @@ static void statemc_init(void) { SMS(SEPARATED, 0, "found both old and current feed files"); } else { debug("startup: F exists, D ENOENT => Normal"); - FILE *file_f= open_input_file(feedfile); + 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"); @@ -1731,6 +1863,26 @@ static void statemc_init(void) { } } +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--; @@ -1749,12 +1901,6 @@ static void statemc_period_poll(void) { } } -static void startup_set_input_file(InputFile *f) { - assert(!main_input_file); - main_input_file= f; - inputfile_tailing_start(f); -} - static int inputfile_is_done(InputFile *ipf) { if (!ipf) return 0; if (ipf->inprogress) return 0; /* new article in the meantime */ @@ -1766,27 +1912,33 @@ 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) +#define RCI_TRIPLE_VALS(x) , RCI_TRIPLE_VALS_BASE(ipf->counts, [RC_##x]) - info("processed %s%s offered=%d(ch%d,nc%d) accepted=%d(ch%d+nc%d)" +#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->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 + 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(); + InputFile *ipf= backlog_input_file; if (!inputfile_is_done(ipf)) return; - const char *slash= strrchr(ipf->path, "/"); + const char *slash= strrchr(ipf->path, '/'); const char *leaf= slash ? slash+1 : ipf->path; - const char *under= strchr(slash, "_"); + const char *under= strchr(slash, '_'); const char *rest= under ? under+1 : leaf; if (!strncmp(rest,"backlog",7)) rest += 7; notice_processed(ipf,"backlog:",rest); @@ -1878,7 +2030,7 @@ 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); @@ -1893,7 +2045,7 @@ 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); @@ -1926,21 +2078,26 @@ static void close_defer(void) { if (!defer) return; - xfstat(fileno(defer), &stab, "defer file"); + xfstat(fileno(defer), &stab, path_defer, "defer file"); if (fclose(defer)) sysfatal("could not close defer file %s", path_defer); defer= 0; + time_t now= time(0); + if (now==-1) sysdie("time(2) failed"); + char *backlog= xasprintf("%s_backlog_%lu.%lu", feedfile, - (unsigned long)now.tv_sec, + (unsigned long)now, (unsigned long)stab.st_ino); - if (link(path_defer, path_backlog)) + if (link(path_defer, backlog)) sysfatal("could not install defer file %s as backlog file %s", path_defer, backlog); if (unlink(path_defer)) sysdie("could not unlink old defer link %s to backlog file %s", path_defer, backlog); + free(backlog); + if (until_backlog_nextscan < 0 || until_backlog_nextscan > backlog_retry_minperiods + 1) until_backlog_nextscan= backlog_retry_minperiods + 1; @@ -2163,13 +2320,6 @@ static void postfork_inputfile(InputFile *ipf) { ipf->fd= -1; } -static void postfork_conns(Connection *conn) { - while (conn) { - close(conn->fd); - conn= conn->next; - } -} - static void postfork_stdio(FILE *f) { /* we have no stdio streams that are buffered long-term */ if (f) fclose(f); @@ -2181,9 +2331,11 @@ static void postfork(const char *what) { postfork_inputfile(main_input_file); postfork_inputfile(flushing_input_file); - postfork_conns(idle.head); - postfork_conns(working.head); - postfork_conns(full.head); + + Conn *conn; + for (conn=LIST_HEAD(conns); conn; conn=LIST_NEXT(conn)) + close(conn->fd); + postfork_stdio(defer); } @@ -2215,14 +2367,13 @@ static const char *debug_ipf_path(InputFile *ipf) { EVERY(period, {PERIOD_SECONDS,0}, { debug("PERIOD" - " sms=%s[%d] queue=%d until_connect=%d" + " sms=%s[%d] conns=%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, + sms_names[sms], sm_period_counter, + queue.count, conns.count, until_connect, DEBUG_IPF(main), DEBUG_IPF(flushing), DEBUG_IPF(flushing), - idle.count, working.count, full.count, (long)connecting_child, (long)inndcomm_child );