/* $OpenBSD: file.c,v 1.30 2012/04/11 06:05:43 ratchov Exp $ */ /* * Copyright (c) 2008 Alexandre Ratchov * * Permission to use, copy, modify, and distribute this software for any * purpose with or without fee is hereby granted, provided that the above * copyright notice and this permission notice appear in all copies. * * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */ /* * non-blocking file i/o module: each file can be read or written (or * both). To achieve non-blocking io, we simply use the poll() syscall * in an event loop. If a read() or write() syscall return EAGAIN * (operation will block), then the file is marked as "for polling", else * the file is not polled again. * * the module also provides trivial timeout implementation, * derived from: * * anoncvs@moule.caoua.org:/midish * * midish/timo.c rev 1.18 * midish/mdep.c rev 1.71 * * A timeout is used to schedule the call of a routine (the callback) * there is a global list of timeouts that is processed inside the * event loop. Timeouts work as follows: * * first the timo structure must be initialized with timo_set() * * then the timeout is scheduled (only once) with timo_add() * * if the timeout expires, the call-back is called; then it can * be scheduled again if needed. It's OK to reschedule it again * from the callback * * the timeout can be aborted with timo_del(), it is OK to try to * abort a timout that has expired * */ #include #include #include #include #include #include #include #include #include #include #include "abuf.h" #include "aproc.h" #include "conf.h" #include "file.h" #ifdef DEBUG #include "dbg.h" #endif #define MAXFDS 100 #define TIMER_USEC 10000 struct timespec file_ts; struct filelist file_list; struct timo *timo_queue; unsigned int timo_abstime; int file_slowaccept = 0; #ifdef DEBUG long long file_wtime, file_utime; #endif /* * initialise a timeout structure, arguments are callback and argument * that will be passed to the callback */ void timo_set(struct timo *o, void (*cb)(void *), void *arg) { o->cb = cb; o->arg = arg; o->set = 0; } /* * schedule the callback in 'delta' 24-th of microseconds. The timeout * must not be already scheduled */ void timo_add(struct timo *o, unsigned int delta) { struct timo **i; unsigned int val; int diff; #ifdef DEBUG if (o->set) { dbg_puts("timo_add: already set\n"); dbg_panic(); } if (delta == 0) { dbg_puts("timo_add: zero timeout is evil\n"); dbg_panic(); } #endif val = timo_abstime + delta; for (i = &timo_queue; *i != NULL; i = &(*i)->next) { diff = (*i)->val - val; if (diff > 0) { break; } } o->set = 1; o->val = val; o->next = *i; *i = o; } /* * abort a scheduled timeout */ void timo_del(struct timo *o) { struct timo **i; for (i = &timo_queue; *i != NULL; i = &(*i)->next) { if (*i == o) { *i = o->next; o->set = 0; return; } } #ifdef DEBUG if (debug_level >= 4) dbg_puts("timo_del: not found\n"); #endif } /* * routine to be called by the timer when 'delta' 24-th of microsecond * elapsed. This routine updates time referece used by timeouts and * calls expired timeouts */ void timo_update(unsigned int delta) { struct timo *to; int diff; /* * update time reference */ timo_abstime += delta; /* * remove from the queue and run expired timeouts */ while (timo_queue != NULL) { /* * there is no overflow here because + and - are * modulo 2^32, they are the same for both signed and * unsigned integers */ diff = timo_queue->val - timo_abstime; if (diff > 0) break; to = timo_queue; timo_queue = to->next; to->set = 0; to->cb(to->arg); } } /* * initialize timeout queue */ void timo_init(void) { timo_queue = NULL; timo_abstime = 0; } /* * destroy timeout queue */ void timo_done(void) { #ifdef DEBUG if (timo_queue != NULL) { dbg_puts("timo_done: timo_queue not empty!\n"); dbg_panic(); } #endif timo_queue = (struct timo *)0xdeadbeef; } #ifdef DEBUG void file_dbg(struct file *f) { dbg_puts(f->ops->name); dbg_puts("("); dbg_puts(f->name); dbg_puts("|"); if (f->state & FILE_ROK) dbg_puts("r"); if (f->state & FILE_RINUSE) dbg_puts("R"); if (f->state & FILE_WOK) dbg_puts("w"); if (f->state & FILE_WINUSE) dbg_puts("W"); if (f->state & FILE_EOF) dbg_puts("e"); if (f->state & FILE_HUP) dbg_puts("h"); if (f->state & FILE_ZOMB) dbg_puts("Z"); dbg_puts(")"); } #endif struct file * file_new(struct fileops *ops, char *name, unsigned int nfds) { struct file *f; LIST_FOREACH(f, &file_list, entry) nfds += f->ops->nfds(f); if (nfds > MAXFDS) { #ifdef DEBUG if (debug_level >= 1) { dbg_puts(name); dbg_puts(": too many polled files\n"); } #endif return NULL; } f = malloc(ops->size); if (f == NULL) err(1, "file_new: %s", ops->name); f->ops = ops; f->name = name; f->state = 0; #ifdef DEBUG f->cycles = 0; #endif f->rproc = NULL; f->wproc = NULL; LIST_INSERT_HEAD(&file_list, f, entry); #ifdef DEBUG if (debug_level >= 3) { file_dbg(f); dbg_puts(": created\n"); } #endif return f; } void file_del(struct file *f) { #ifdef DEBUG if (debug_level >= 3) { file_dbg(f); dbg_puts(": terminating...\n"); } #endif if (f->state & (FILE_RINUSE | FILE_WINUSE)) { f->state |= FILE_ZOMB; } else { LIST_REMOVE(f, entry); #ifdef DEBUG if (debug_level >= 3) { file_dbg(f); dbg_puts(": destroyed\n"); } #endif f->ops->close(f); free(f); } } int file_poll(void) { nfds_t nfds, n; short events, revents; struct pollfd pfds[MAXFDS]; struct file *f, *fnext; struct aproc *p; struct timespec ts; #ifdef DEBUG struct timespec sleepts; #endif long long delta_nsec; int res; if (LIST_EMPTY(&file_list) && timo_queue == NULL) { #ifdef DEBUG if (debug_level >= 3) dbg_puts("nothing to do...\n"); #endif return 0; } /* * Fill the pfds[] array with files that are blocked on reading * and/or writing, skipping those that are just waiting. */ #ifdef DEBUG dbg_flush(); if (debug_level >= 4) dbg_puts("poll:"); #endif nfds = 0; LIST_FOREACH(f, &file_list, entry) { events = 0; if (f->rproc && !(f->state & FILE_ROK)) events |= POLLIN; if (f->wproc && !(f->state & FILE_WOK)) events |= POLLOUT; #ifdef DEBUG if (debug_level >= 4) { dbg_puts(" "); file_dbg(f); } #endif n = f->ops->pollfd(f, pfds + nfds, events); if (n == 0) { f->pfd = NULL; continue; } f->pfd = pfds + nfds; nfds += n; } #ifdef DEBUG if (debug_level >= 4) { dbg_puts("\npfds[] ="); for (n = 0; n < nfds; n++) { dbg_puts(" "); dbg_putx(pfds[n].events); } dbg_puts("\n"); } #endif #ifdef DEBUG clock_gettime(CLOCK_MONOTONIC, &sleepts); file_utime += 1000000000LL * (sleepts.tv_sec - file_ts.tv_sec); file_utime += sleepts.tv_nsec - file_ts.tv_nsec; #endif res = poll(pfds, nfds, -1); if (res < 0 && errno != EINTR) err(1, "poll"); clock_gettime(CLOCK_MONOTONIC, &ts); #ifdef DEBUG file_wtime += 1000000000LL * (ts.tv_sec - sleepts.tv_sec); file_wtime += ts.tv_nsec - sleepts.tv_nsec; #endif delta_nsec = 1000000000LL * (ts.tv_sec - file_ts.tv_sec); delta_nsec += ts.tv_nsec - file_ts.tv_nsec; #ifdef DEBUG if (delta_nsec < 0) dbg_puts("file_poll: negative time interval\n"); #endif file_ts = ts; if (delta_nsec >= 0 && delta_nsec < 1000000000LL) timo_update(delta_nsec / 1000); else { #ifdef DEBUG if (debug_level >= 1) dbg_puts("ignored huge clock delta\n"); #endif } if (res <= 0) return 1; f = LIST_FIRST(&file_list); while (f != NULL) { if (f->pfd == NULL) { f = LIST_NEXT(f, entry); continue; } revents = f->ops->revents(f, f->pfd); #ifdef DEBUG if (revents) { f->cycles++; if (f->cycles > FILE_MAXCYCLES) { file_dbg(f); dbg_puts(": busy loop, disconnecting\n"); revents = POLLHUP; } } #endif if (!(f->state & FILE_ZOMB) && (revents & POLLIN)) { revents &= ~POLLIN; #ifdef DEBUG if (debug_level >= 4) { file_dbg(f); dbg_puts(": rok\n"); } #endif f->state |= FILE_ROK; f->state |= FILE_RINUSE; for (;;) { p = f->rproc; if (!p) break; #ifdef DEBUG if (debug_level >= 4) { aproc_dbg(p); dbg_puts(": in\n"); } #endif if (!p->ops->in(p, NULL)) break; } f->state &= ~FILE_RINUSE; } if (!(f->state & FILE_ZOMB) && (revents & POLLOUT)) { revents &= ~POLLOUT; #ifdef DEBUG if (debug_level >= 4) { file_dbg(f); dbg_puts(": wok\n"); } #endif f->state |= FILE_WOK; f->state |= FILE_WINUSE; for (;;) { p = f->wproc; if (!p) break; #ifdef DEBUG if (debug_level >= 4) { aproc_dbg(p); dbg_puts(": out\n"); } #endif if (!p->ops->out(p, NULL)) break; } f->state &= ~FILE_WINUSE; } if (!(f->state & FILE_ZOMB) && (revents & POLLHUP)) { #ifdef DEBUG if (debug_level >= 3) { file_dbg(f); dbg_puts(": disconnected\n"); } #endif f->state |= (FILE_EOF | FILE_HUP); } if (!(f->state & FILE_ZOMB) && (f->state & FILE_EOF)) { #ifdef DEBUG if (debug_level >= 3) { file_dbg(f); dbg_puts(": eof\n"); } #endif p = f->rproc; if (p) { f->state |= FILE_RINUSE; #ifdef DEBUG if (debug_level >= 3) { aproc_dbg(p); dbg_puts(": eof\n"); } #endif p->ops->eof(p, NULL); f->state &= ~FILE_RINUSE; } f->state &= ~FILE_EOF; } if (!(f->state & FILE_ZOMB) && (f->state & FILE_HUP)) { #ifdef DEBUG if (debug_level >= 3) { file_dbg(f); dbg_puts(": hup\n"); } #endif p = f->wproc; if (p) { f->state |= FILE_WINUSE; #ifdef DEBUG if (debug_level >= 3) { aproc_dbg(p); dbg_puts(": hup\n"); } #endif p->ops->hup(p, NULL); f->state &= ~FILE_WINUSE; } f->state &= ~FILE_HUP; } fnext = LIST_NEXT(f, entry); if (f->state & FILE_ZOMB) file_del(f); f = fnext; } if (LIST_EMPTY(&file_list) && timo_queue == NULL) { #ifdef DEBUG if (debug_level >= 3) dbg_puts("no files anymore...\n"); #endif return 0; } return 1; } /* * handler for SIGALRM, invoked periodically */ void file_sigalrm(int i) { /* nothing to do, we only want poll() to return EINTR */ } void filelist_init(void) { static struct sigaction sa; struct itimerval it; sigset_t set; sigemptyset(&set); (void)sigaddset(&set, SIGPIPE); if (sigprocmask(SIG_BLOCK, &set, NULL)) err(1, "sigprocmask"); LIST_INIT(&file_list); if (clock_gettime(CLOCK_MONOTONIC, &file_ts) < 0) { perror("clock_gettime"); exit(1); } sa.sa_flags = SA_RESTART; sa.sa_handler = file_sigalrm; sigfillset(&sa.sa_mask); if (sigaction(SIGALRM, &sa, NULL) < 0) { perror("sigaction"); exit(1); } it.it_interval.tv_sec = 0; it.it_interval.tv_usec = TIMER_USEC; it.it_value.tv_sec = 0; it.it_value.tv_usec = TIMER_USEC; if (setitimer(ITIMER_REAL, &it, NULL) < 0) { perror("setitimer"); exit(1); } timo_init(); #ifdef DEBUG dbg_sync = 0; #endif } void filelist_done(void) { struct itimerval it; #ifdef DEBUG struct file *f; if (!LIST_EMPTY(&file_list)) { LIST_FOREACH(f, &file_list, entry) { file_dbg(f); dbg_puts(" not closed\n"); } dbg_panic(); } dbg_sync = 1; dbg_flush(); #endif timerclear(&it.it_value); timerclear(&it.it_interval); if (setitimer(ITIMER_REAL, &it, NULL) < 0) { perror("setitimer"); exit(1); } timo_done(); } unsigned int file_read(struct file *f, unsigned char *data, unsigned int count) { unsigned int n; #ifdef DEBUG struct timespec ts0, ts1; long us; if (!(f->state & FILE_ROK)) { file_dbg(f); dbg_puts(": read: bad state\n"); dbg_panic(); } clock_gettime(CLOCK_MONOTONIC, &ts0); #endif n = f->ops->read(f, data, count); #ifdef DEBUG if (n > 0) f->cycles = 0; clock_gettime(CLOCK_MONOTONIC, &ts1); us = 1000000L * (ts1.tv_sec - ts0.tv_sec); us += (ts1.tv_nsec - ts0.tv_nsec) / 1000; if (debug_level >= 4 || (debug_level >= 2 && us >= 5000)) { dbg_puts(f->name); dbg_puts(": read "); dbg_putu(n); dbg_puts(" bytes in "); dbg_putu(us); dbg_puts("us\n"); } #endif return n; } unsigned int file_write(struct file *f, unsigned char *data, unsigned int count) { unsigned int n; #ifdef DEBUG struct timespec ts0, ts1; long us; if (!(f->state & FILE_WOK)) { file_dbg(f); dbg_puts(": write: bad state\n"); dbg_panic(); } clock_gettime(CLOCK_MONOTONIC, &ts0); #endif n = f->ops->write(f, data, count); #ifdef DEBUG if (n > 0) f->cycles = 0; clock_gettime(CLOCK_MONOTONIC, &ts1); us = 1000000L * (ts1.tv_sec - ts0.tv_sec); us += (ts1.tv_nsec - ts0.tv_nsec) / 1000; if (debug_level >= 4 || (debug_level >= 2 && us >= 5000)) { dbg_puts(f->name); dbg_puts(": wrote "); dbg_putu(n); dbg_puts(" bytes in "); dbg_putu(us); dbg_puts("us\n"); } #endif return n; } void file_eof(struct file *f) { struct aproc *p; #ifdef DEBUG if (debug_level >= 3) { file_dbg(f); dbg_puts(": eof requested\n"); } #endif if (!(f->state & (FILE_RINUSE | FILE_WINUSE))) { p = f->rproc; if (p) { f->state |= FILE_RINUSE; #ifdef DEBUG if (debug_level >= 3) { aproc_dbg(p); dbg_puts(": eof\n"); } #endif p->ops->eof(p, NULL); f->state &= ~FILE_RINUSE; } if (f->state & FILE_ZOMB) file_del(f); } else { f->state &= ~FILE_ROK; f->state |= FILE_EOF; } } void file_hup(struct file *f) { struct aproc *p; #ifdef DEBUG if (debug_level >= 3) { file_dbg(f); dbg_puts(": hup requested\n"); } #endif if (!(f->state & (FILE_RINUSE | FILE_WINUSE))) { p = f->wproc; if (p) { f->state |= FILE_WINUSE; #ifdef DEBUG if (debug_level >= 3) { aproc_dbg(p); dbg_puts(": hup\n"); } #endif p->ops->hup(p, NULL); f->state &= ~FILE_WINUSE; } if (f->state & FILE_ZOMB) file_del(f); } else { f->state &= ~FILE_WOK; f->state |= FILE_HUP; } } void file_close(struct file *f) { struct aproc *p; #ifdef DEBUG if (debug_level >= 3) { file_dbg(f); dbg_puts(": closing\n"); } #endif if (f->wproc == NULL && f->rproc == NULL) f->state |= FILE_ZOMB; if (!(f->state & (FILE_RINUSE | FILE_WINUSE))) { p = f->rproc; if (p) { f->state |= FILE_RINUSE; #ifdef DEBUG if (debug_level >= 3) { aproc_dbg(p); dbg_puts(": eof\n"); } #endif p->ops->eof(p, NULL); f->state &= ~FILE_RINUSE; } p = f->wproc; if (p) { f->state |= FILE_WINUSE; #ifdef DEBUG if (debug_level >= 3) { aproc_dbg(p); dbg_puts(": hup\n"); } #endif p->ops->hup(p, NULL); f->state &= ~FILE_WINUSE; } if (f->state & FILE_ZOMB) file_del(f); } else { f->state &= ~(FILE_ROK | FILE_WOK); f->state |= (FILE_EOF | FILE_HUP); } }