X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ian/git?p=inn-innduct.git;a=blobdiff_plain;f=backends%2Finnduct.c;h=b360d9540d10dc310db146ed370d0b9ce3c44a94;hp=f3be2b7613f07014e5174e4c274b9c70765c0611;hb=84b25112929fba84e4af13907b0d30209deaf404;hpb=e7e211b5ea4799a7b962988c055442e6cec6fdfe diff --git a/backends/innduct.c b/backends/innduct.c index f3be2b7..b360d95 100644 --- a/backends/innduct.c +++ b/backends/innduct.c @@ -15,75 +15,60 @@ * 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_. - * 431'd articles, ready for innxmit + * + * site.name_backlog.. + * 431'd articles, ready for innxmit or duct * created (link/mv) by duct - * read by xmit - * unlinked by xmit - * 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 - * read by xmit - * unlinked by xmit - - - - - BLATHER ABOUT NOT USING INNXMIT - - scan for backlog files - take "old enough" backlog files one at a time and reprocess them - so need another input file - the current backlog file - inhibit scanning for backlog files only when - - last file not "old enough" - - no backlog files found and we haven't made one - - also we have to start a new backlog file every (some interval) - + * 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 + * back into our processing. When every article in it has been read + * and processed, we unlink it and look for another backlog file. + * + * 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. - 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 | @@ -97,31 +82,35 @@ | 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 / @@ -170,6 +159,7 @@ 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; @@ -252,32 +242,42 @@ struct Article { char messageid[1]; }; -typedef struct { +typedef struct InputFile { /* This is an instance of struct oop_readable */ struct oop_readable readable; /* first */ oop_readable_call *readable_callback; 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; + char path[]; } InputFile; +#define SMS_LIST(X) \ + X(WAITING) \ + X(NORMAL) \ + X(FLUSHING) \ + X(FLUSHFAIL) \ + X(SEPARATED) \ + X(DROPPING) + typedef enum { - sm_WAITING, - sm_NORMAL, - sm_FLUSHING, - sm_FLUSHFAIL, - sm_SEPARATED1, - sm_SEPARATED2, /* must follow SEPARATED2 - see feedfile_eof */ - sm_DROPPING1, - sm_DROPPING2, /* must follow DROPPING1 - see feedfile_eof */ +#define SMS_DEF_ENUM(s) sm_##s, + SMS_LIST(SMS_DEF_ENUM) } StateMachineState; +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; @@ -291,19 +291,18 @@ struct Conn { /*----- operational variables -----*/ -static int since_connect_attempt; static int nconns; static LIST(Conn) idle, working, full; static LIST(Article) *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; +static InputFile *main_input_file, *flushing_input_file, *backlog_input_file; static int sm_period_counter; @@ -373,6 +372,47 @@ 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)) + die("%s %s not a plain file (mode 0%lo)", + what, path, (unsigned long)stab->st_mode); +} + +static void xfstat(int fd, struct stat *stab_r, const char *what) { + int r= fstab(path, stab); + if (r) sysdie("could not fstat %s %s", what, path); +} + +static void xfstat_isreg(int fd, struct stat *stab_r, const char *what) { + xfstat(fd, stab_r, what); + check_isreg(stab, path, what); +} + +static void xlstat_isreg(const char *path, struct stat *stab, + int *enoent_r /* 0 means ENOENT is fatal */ + const char *what) { + int r= lstat(path, stab); + if (r) { + if (errno==ENOENT && enoent_r) { *enoent_r=1; return; } + sysdie("could not lstat %s %s", what, path); + } + if (enoent_r) *enoent_r= 0; + check_isreg(stab, path, what); +} + +static int samefile(const struct stat *a, const struct stat *b) { + assert(S_ISREG(a->st_mode)); + assert(S_ISREG(b->st_mode)); + return (a->st_ino == b->st_ino && + a->st_dev == b->st_dev); +} + /*========== logging ==========*/ static void logcore(int sysloglevel, const char *fmt, ...) @@ -426,6 +466,7 @@ logwrap(warn, " warning", LOG_WARN, -1, 0); logwrap(notice, "", LOG_NOTICE, -1, 0); logwrap(info, " info", LOG_INFO, -1, 0); +logwrap(debug, " debug", LOG_DEBUG, -1, 0); /*========== making new connections ==========*/ @@ -641,8 +682,6 @@ static void connect_start() { /*========== overall control of article flow ==========*/ static void check_master_queue(void) { - try reading current feed file; - if (!queue.count) return; @@ -821,7 +860,7 @@ static void conn_make_some_xmits(Conn *conn) { art->sent= 1; LIST_ADDTAIL(conn->sent, art); - counts[art->checked].sent++; + art->ipf->counts[art->checked].sent++; } else { /* check it */ @@ -834,7 +873,7 @@ static void conn_make_some_xmits(Conn *conn) { XMIT_LITERAL("\r\n"); LIST_ADDTAIL(conn->sent, art); - counts[art->checked].offered++; + art->ipf->counts[art->checked].offered++; } } } @@ -848,6 +887,14 @@ static const oop_rd_style peer_rd_style= { OOP_RD_SHORTREC_FORBID }; +static void *peer_rd_err(oop_source *lp, oop_read *oread, oop_event ev, + const char *errmsg, int errnoval, + const char *data, size_t recsz, void *conn_v) { + Conn *conn= conn_v; + connfail(conn, "error receiving from peer: %s", errmsg); + return OOP_CONTINUE; +} + static Article *article_reply_check(Connection *conn, const char *response, int code_indicates_streaming, const char *sanitised_response) { @@ -896,16 +943,19 @@ static Article *article_reply_check(Connection *conn, const char *response, 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 != nockech) { + debug("nocheck mode %s", new_nocheck ? "start" : "stop"); } + nocheck= new_nocheck; } static void article_done(Connection *conn, Article *art, int whichcount) { *count++; - counts.articles[art->checked][whichcount]++; + art->ipf->counts.articles[art->checked][whichcount]++; if (whichcount == RC_accepted) update_nocheck(1); else if (whichcount == RC_unwanted) update_nocheck(0); @@ -932,19 +982,11 @@ static void article_done(Connection *conn, Article *art, int whichcount) { assert(ipf->inprogress >= 0); if (!ipf->inprogress) - loop->on_time(loop, OOP_TIME_NOW, statemc_check_oldinput_done, 0); + loop->on_time(loop, OOP_TIME_NOW, statemc_check_input_done, ipf); free(art); } -static void *peer_rd_err(oop_source *lp, oop_read *oread, oop_event ev, - const char *errmsg, int errnoval, - const char *data, size_t recsz, void *conn_v) { - Conn *conn= conn_v; - connfail(conn, "error receiving from peer: %s", errmsg); - return OOP_CONTINUE; -} - static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev, const char *errmsg, int errnoval, const char *data, size_t recsz, void *conn_v) { @@ -1031,9 +1073,10 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev, code_streaming= 1; case 436: /* IHAVE says try later */ GET_ARTICLE; + 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; @@ -1048,13 +1091,25 @@ 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 */ - assert(ipf == old_input_file); - if (sms==sm_SEPARATED1) SMS(SEPARATED2, 0, "eof on old feed file"); - else if (sms==sm_DROPPING1) SMS(DROPPING2, 0, "eof on dead feed file"); - else abort(); - 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 == flushing_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); } @@ -1066,7 +1121,7 @@ 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; @@ -1074,7 +1129,7 @@ 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; } @@ -1085,7 +1140,9 @@ static void close_input_file(InputFile *ipf) { assert(!ipf->rd); /* must have had inputfile_tailing_stop */ assert(!ipf->inprogress); /* no dangling pointers pointing here */ - if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path); + if (ipf->fd >= 0) + if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path); + free(ipf); } @@ -1126,16 +1183,17 @@ typedef void *feedfile_got_article(oop_source *lp, oop_read *rd, } ipf->offset += recsz + 1; - if (sms==sm_NORMAL && ipf->offset >= flush_threshold) { + if (sms==sm_NORMAL && ipf==main_input_file && + (ipf->offset >= flush_threshold || !until_spontaneous_flush) { + notice("starting flush (%lu >= %lu)", (unsigned long)ipf->offset, (unsigned long)flush_threshold); int r= link(feedfile, duct_path); - if (r) sysdie("link feedfile %s to ductfile %s", feedfile, dut_path); + if (r) sysdie("link feedfile %s to flushing file %s", feedfile, dut_path); /* => Hardlinked */ - r= unlink(feedfile); - if (r) sysdie("unlink old feedfile link %s", feedfile); + xunlink(feedfile, "old feedfile link"); /* => Moved */ spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */ @@ -1203,8 +1261,17 @@ static ssize_t tailing_try_read(struct oop_readable *rable, void *buffer, return r; } if (!r) { - if (ipf==main_input_file) { errno=EAGAIN; return -1; } - assert(sms==sm_SEPARATED1 || sms==sm_DROPPING1); + if (ipf==main_input_file) { + errno=EAGAIN; + return -1; + } else if (ipf==flushing_input_file) { + assert(ipf->fd>=0); + assert(sms==sm_SEPARATED || sms==sm_DROPPING); + } else if (ipf==backlog_input_file) { + assert(ipf->fd>=0); + } else { + abort(); + } } return r; } @@ -1241,11 +1308,15 @@ 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; @@ -1265,6 +1336,7 @@ 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; @@ -1279,6 +1351,7 @@ 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; } @@ -1325,224 +1398,190 @@ static void inputfile_tailing_start(InputFile *ipf) { static void inputfile_tailing_stop(InputFile *ipf) { assert(ipf->fd); + oop_rd_cancel(ipf->rd); oop_rd_delete(ipf->rd); ipf->rd= 0; assert(!ipf->filemon); /* we shouldn't be monitoring it now */ } -/*========== interaction with innd ==========*/ +/*========== interaction with innd - state machine ==========*/ /* 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] - | | - | |<---------------------------------------------------. + .=======. + ||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] | - | main D tail | - | ========== | + | main D tail/none | + | ================ | | | | | | INNDCOMM FLUSH FAILS ^ | |`----------------------->----------. | | | | | | | NO SUCH SITE V | - ^ |`--------------->----. =========== | - | | \ FLUSHFAIL | - | | \ [Moved] | - | | \ main D tail | - | | \ =========== | + ^ |`--------------->----. ================ | + | | \ FLUSHFAIL | + | | \ [Moved] | + | | FLUSH OK \ main D tail/none | + | | open F \ ================ | | | \ | | | | \ | TIME TO RETRY | - | | \ `----------------' - | | FLUSH OK \ - | | open F \ - | V V - | ============= ============ - | SEPARATED1 DROPPING1 - | [Separated] [Dropping] - | main F idle main none - | old D tail old D tail - | ============= ============ - | | | - ^ | EOF ON D | EOF ON D - | V V - | ============= ============ - | SEPARATED2 DROPPING2 - | [Finishing] [Dropping] - | main F tail main none - | old D idle old D idle - | ============= ============ - | | | - | | ALL D PROCESSED | ALL D PROCESSED - | V install defer as backlog V install defer as backlog - ^ | close D | close D + |<--------'| ,---<---'\ `----------------' + | D NONE | | D NONE \ + | | | \ + | | | | + | | V | + | V | V + | ============= | ============ + | SEPARATED/ | DROPPING/ + | flsh->fd>=0 | flsh->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/ + | flsh->fd==-1 V flsh->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 - | | start new defer | exit - | | V - `----------' ========== + | | | unlink lock + | | | exit + `----------' V + ========== (ESRCH) [Droppped] ========== */ -static void open_defer(void) { - struct stat stab; - - assert(!defer); - defer= fopen(path_ductdefer, "a+"); - if (!defer) sysfatal("could not open defer file %s", path_ductdefer); - - /* truncate away any half-written records */ - - r= fstat(fileno(defer), &stab); - if (r) sysdie("could not stat newly opened defer file %s", path_ductdefer); - - if (stab.st_size > LONG_MAX) - die("defer file %s size is far too large", path_ductdefer); +static void statemc_init(void) { + struct stat stab, stabf; + int noent; - if (!stab.st_size) - return; + path_lock= xasprintf("%s_lock", feedfile); + path_flushing= xasprintf("%s_flushing", feedfile); + path_defer= xasprintf("%s_defer", feedfile); + globpat_backlog= xasprintf("%s_backlog*", feedfile); - long orgsize= stab.st_size; - long truncto= stab.st_size; 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); - - r= getc(defer); - if (r==EOF) { - if (ferror(defer)) - sysdie("failed read from defer file %s", path_ductdefer); - else - die("defer file %s shrank while we were checking it!", path_ductdefer); + 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)); + fl.l_type= F_WRLCK; + fl.l_whence= SEEK_SET; + r= fcntl(lockfd, F_SETLK, &fl); + if (r==-1) { + if (errno==EACCES || errno==EAGAIN) { + if (quiet_if_locked) exit(0); + fatal("another duct holds the lockfile"); + } + sysdie("fcntl F_SETLK lockfile %s", path_lock); } - if (r=='\n') break; - truncto--; - } - 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); + xfstat_isreg(lockfd, &stabf, "lockfile"); + xlstat_isreg(path_lock, &stab, &noent, "lockfile"); - if (fflush(defer)) - sysfatal("could not flush defer file %s", path_ductdefer); - if (ftruncate(fileno(defer), truncto)) - sysdie("could not truncate defer file %s", path_ductdefer); + if (!noent && samefile(&stab, &stabf)) + break; - } else { - info("continuing existing defer file %s (%ld bytes)", - path_ductdefer, orgsize); + if (close(lockfd)) + sysdie("could not close stale lockfile %s", path_lock); } - if (fseek(defer, truncto, SEEK_SET)) - sysdie("could not seek to new end of defer file %s", path_ductdefer); -} - -static void statemc_init(void) { - struct stat stab; + debug("startup: locked"); - path_ductlock= xasprintf("%s_duct.lock", feedfile); - path_duct= xasprintf("%s_duct", feedfile); - path_ductdefer= xasprintf("%s_duct.defer", feedfile); + search_backlog_file(); - if (lstat(path_ductdefer, &stab)) { - if (errno!=ENOENT) sysdie("could not check defer file %s", path_defer); + xlstat_isreg(path_defer, &stab, &noent, "defer file"); + if (noent) { + debug("startup: ductdefer ENOENT"); } else { - if (!S_ISREG(stab.st_mode)) - die("defer file %s not a plain file (mode 0%lo)", - path_defer, (unsigned long)stab.st_mode); + 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", path_defer, stab.st_nlink); } } - open_defer(); - int lockfd= open(path_ductlock, O_CREAT|O_RDWR, 0600); - if (lockfd<0) sysfatal("open lockfile %s", path_ductlock); - - struct flock fl; - memset(&fl,0,sizeof(fl)); - fl.l_type= F_WRLCK; - fl.l_whence= SEEK_SET; - r= fcntl(lockfd, F_SETLK, &fl); - if (r==-1) { - if (errno==EACCES || errno==EAGAIN) { - if (quiet_if_locked) exit(0); - fatal("another duct holds the lockfile"); - } - sysdie("fcntl F_SETLK lockfile %s", path_ductlock); - } - - InputFile *file_d= open_input_file(path_duct); + InputFile *file_d= open_input_file(path_flushing); if (file_d) { struct stat stab_f, stab_d; - r= stat(feedfile, &stab_f); - if (r) { - if (errno!=ENOENT) sysdie("check feed file %s", feedfile); - /* D exists, F ENOENT => Moved */ + xlstat_isreg(feedfile, &stab_f, &noent, "feed file"); + if (noent) { + debug("startup: D exists, F ENOENT => Moved"); goto found_moved; } - /* F and D both exist */ + debug("startup: F and D both exist"); - r= fstat(file_d->fd, &stab_d); - if (r) sysdie("check duct file %s", ductfile); + xfstat_isreg(file_d->fd, &stab_d, "flushing file"); - if (stab_d.st_ino == stab_f.st_ino && - stab_d.st_dev == stab_f.st_dev) { - /* F==D => Hardlinked*/ - r= unlink(path_duct); - if (r) sysdie("unlink feed file %s during startup", feedfile); + if (samefile(&stab_d, &stab_f)) { + debug("startup: F==D => Hardlinked"); + xunlink(path_flushing, "feed file (during startup)"); found_moved: - /* => Moved */ + debug(" => Moved"); startup_set_input_file(file_d); spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */ } else { - /* F!=D => Separated */ - SMS(SEPARATED1, 0, "found both old and current feed files"); + debug("F!=D => Separated"); + SMS(SEPARATED, 0, "found both old and current feed files"); startup_set_input_file(file_d); } - } else { /*!file_d*/ + } else { + debug("startup: D ENOENT => Nothing"); SMS(WAITING, open_wait_periods, "no feed file currently exists"); } } @@ -1576,46 +1615,56 @@ static void statemc_waiting_poll(void) { static void startup_set_input_file(InputFile *f) { assert(!main_input_file); main_input_file= f; + until_spontaneous_flush= spontaneous_flush_periods; inputfile_tailing_start(f); } -static void *statemc_check_oldinput_done(oop_source *lp, - struct timeval now, void *u) { +static void *statemc_check_input_done(oop_source *lp, + struct timeval now, void *ipf_v) { + InputFile *ipf= ipf_v; struct stat stab; - int done= (sms==sm_SEPARATED2 || sms==sm_DROPPING2) - && old_input_file->inprogress; - if (!done) return; + if (ipf->inprogress) return; /* new article in the meantime */ + if (ipf->fd >= 0); return; /* not had EOF */ + + if (ipf == backlog_input_file) { + notice_processed(ipf,"backlog file",ipf->path); + close_input_file(ipf); + if (unlink(ipf->path)) { + if (errno != ENOENT) + sysdie("could not unlink processed backlog file %s", ipf->path); + warn("backlog file %s vanished while we were reading it" + " so we couldn't remove it (but it's done now, anyway)", + ipf->path); + } + backlog_input_file= 0; + search_backlog_file(); + return; + } - r= fstat(fileno(defer), &stab); - if (r) sysdie("check defer file %s", path_defer); + assert(ipf == flushing_input_file); + assert(sms==sm_SEPARATED || sms==sm_DROPPING); - if (fclose(defer)) sysfatal("could not close defer file %s", path_defer); - defer= 0; + notice_processed(ipf,"feed file",0); - char *backlog= xasprintf("%s_backlog_%lu.%lu", feedfile, - (unsigned long)now.tv_sec, - (unsigned long)stab.st_ino); - if (link(path_defer, path_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); + 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_DROPPING2) { - notice("feed dropped and our work is complete" - " (but check for backlog files)"); + if (sms==sm_DROPPING) { + if (search_backlog_file()) { + debug("feed dropped but still backlogs to process"); + return; + } + notice("feed dropped and our work is complete"); + xunlink(path_lock, "lockfile for old feed"); exit(0); } open_defer(); - close_input_file(old_input_file); - old_input_file= 0; + close_input_file(flushing_input_file); + flushing_input_file= 0; notice("flush complete"); SMS(NORMAL, 0, "flush complete"); @@ -1632,6 +1681,178 @@ static void statemc_setstate(StateMachineState newsms, int periods, } } +/*---------- defer and backlog files ----------*/ + +static void open_defer(void) { + struct stat stab; + + if (defer) return; + + 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_defer); + + if (!stab.st_size) + return; + + long orgsize= stab.st_size; + long truncto= stab.st_size; + 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_defer); + + r= getc(defer); + if (r==EOF) { + if (ferror(defer)) + sysdie("failed read from defer file %s", path_defer); + else + die("defer file %s shrank while we were checking it!", path_defer); + } + if (r=='\n') break; + truncto--; + } + + if (stab.st_size != truncto) { + warn("truncating half-record at end of defer file %s -" + " shrinking by %ld bytes from %ld to %ld", + path_defer, orgsize - truncto, orgsize, truncto); + + if (fflush(defer)) + sysfatal("could not flush defer file %s", path_defer); + if (ftruncate(fileno(defer), truncto)) + sysdie("could not truncate defer file %s", path_defer); + + } else { + info("continuing existing defer file %s (%ld bytes)", + path_defer, orgsize); + } + if (fseek(defer, truncto, SEEK_SET)) + sysdie("could not seek to new end of defer file %s", path_defer); +} + +static void close_defer(void) { + if (!defer) + return; + + xfstat(fileno(defer), &stab, "defer file"); + + if (fclose(defer)) sysfatal("could not close defer file %s", path_defer); + defer= 0; + + char *backlog= xasprintf("%s_backlog_%lu.%lu", feedfile, + (unsigned long)now.tv_sec, + (unsigned long)stab.st_ino); + if (link(path_defer, path_backlog)) + sysfatal("could not install defer file %s as backlog file %s", + path_defer, backlog); + if (unlink(path_defer)) + sysdie("could not unlink old defer link %s to backlog file %s", + path_defer, backlog); + + if (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 (until_backlog_nextscan < 0) return; + if (until_backlog_nextscan-- > 0) return; + search_backlog_file(); +} + +static int 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; + + if (backlog_input_file) return 3; + + try_again: + + r= glob(globpat_backlog, GLOB_ERR|GLOB_MARK|GLOB_NOSORT, 0, &gl); + + switch (r) { + case GLOB_ABORTED: + sysdie("failed to expand backlog pattern %s", globpat_backlog); + case GLOB_NOSPACE: + die("out of memory expanding backlog pattern %s", globpat_backlog); + case 0: + for (i=0; i= 0 && + 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, until_backlog_nextscan, oldest_path); + return 2; +} + /*========== flushing the feed ==========*/ static pid_t inndcomm_child; @@ -1642,32 +1863,33 @@ 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)) { - + case INNDCOMMCHILD_ESTATUS_FAIL: goto failed; 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; main_input_file= 0; - SMS(DROPPING1, 0, "dropped by innd"); + SMS(DROPPING, 0, "dropped by innd"); return OOP_CONTINUE; case 0: - old_input_file= main_input_file; + flushing_input_file= main_input_file; main_input_file= open_input_file(feedfile); if (!main_input_file) die("flush succeeded but feedfile %s does not exist!", feedfile); - SMS(SEPARATED1, 0, "feed file missing"); + until_spontaneous_flush= spontaneous_flush_periods; + SMS(SEPARATED, 0, "feed file missing"); return OOP_CONTINUE; default: goto unexpected_exitstatus; - + } } else if (WIFSIGNALED(status) && WTERMSIG(status) == SIGALRM) { warn("flush timed out trying to talk to innd"); @@ -1745,14 +1967,13 @@ 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); postfork_stdio(defer); } - #define EVERY(what, interval, body) \ static const struct timeval what##_timeout = { 5, 0 }; \ static void what##_schedule(void); \ @@ -1764,10 +1985,39 @@ static void postfork(const char *what) { loop->on_time(loop, what##_timeout, what##_timedout, 0); \ } -EVERY(filepoll, {5,0}, { check_master_queue(); }) +EVERY(filepoll, {5,0}, { + if (main_input_file && main_input_file->readable_callback) + 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}, { + debug("PERIOD" + " sms=%s queue=%d sm_period_counter=%d" + " connect_delay=%d until_spontaneous_flush=%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], queue.count, sm_period_counter, + connect_delay, until_spontaneous_flush, + DEBUG_IPF(main), DEBUG_IPF(flushing), DEBUG_IPF(flushing), + idle.count, working.count, full.count, + (long)connecting_child, (long)inndcomm_child + ); if (connect_delay) connect_delay--; + if (until_spontaneous_flush) until_spontaneous_flush--; + poll_backlog_file(); + if (!backlog_input_file) close_defer(); /* want to start on a new backlog */ statemc_poll(); check_master_queue(); }); @@ -1819,7 +2069,7 @@ void op_string(const Option *o, const char *val) { void op_seconds(const Option *o, const char *val) { int *store= o->store; char *ep; - + double v= strtod(val,&ep); if (ep==val) badusage("bad time/duration value for %s",o->long); @@ -1877,7 +2127,7 @@ static const Option options[]= { int main(int argc, char **argv) { const char *arg; - + for (;;) { arg= *++argv; if (!arg) break; @@ -1926,9 +2176,15 @@ 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)) { + if (remote_host) badusage("must not specify -h and also host as argument"); + 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; @@ -1936,7 +2192,7 @@ int main(int argc, char **argv) { if (nocheck_decay_articles < 0.1) badusage("nocheck decay articles must be at least 0.1"); nocheck_decay= 1 - 1/nocheck_decay_articles; - + innconf_read(inndconffile); if (!feedfile) @@ -1945,7 +2201,13 @@ int main(int argc, char **argv) { badusage("feed filename must be nonempty"); else if (feedfile[strlen(feedfile)-1]=='/') feedfile= xasprintf("%s%s",feedfile,sitename); - + + const char *feedfile_forbidden= "?*[~#"; + int c; + while ((c= *feedfile_forbidden++)) + if (strchr(feedfile, c)) + badusage("feed filename may not contain metacharacter %c",c); + if (signal(SIGPIPE, SIG_IGN) == SIG_ERR) sysdie("could not ignore SIGPIPE"); @@ -1961,7 +2223,7 @@ int main(int argc, char **argv) { dup2(null,1); dup2(null,2); close(null); - + pid_t child1= xfork("daemonise first fork"); if (child1) _exit(0);