2 * Copyright (c) 1990, 1993, 1995
3 * The Regents of the University of California.
4 * Copyright (c) 2005 Robert N. M. Watson
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)fifo_vnops.c 8.10 (Berkeley) 5/27/95
35 #include <sys/param.h>
36 #include <sys/event.h>
38 #include <sys/filedesc.h>
39 #include <sys/filio.h>
40 #include <sys/fcntl.h>
41 #include <sys/kernel.h>
43 #include <sys/mutex.h>
44 #include <sys/malloc.h>
46 #include <sys/proc.h> /* XXXKSE */
47 #include <sys/signalvar.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
51 #include <sys/systm.h>
53 #include <sys/unistd.h>
54 #include <sys/vnode.h>
55 #include <fs/fifofs/fifo.h>
57 static fo_rdwr_t fifo_read_f;
58 static fo_rdwr_t fifo_write_f;
59 static fo_ioctl_t fifo_ioctl_f;
60 static fo_poll_t fifo_poll_f;
61 static fo_kqfilter_t fifo_kqfilter_f;
62 static fo_stat_t fifo_stat_f;
63 static fo_close_t fifo_close_f;
65 struct fileops fifo_ops_f = {
66 .fo_read = fifo_read_f,
67 .fo_write = fifo_write_f,
68 .fo_ioctl = fifo_ioctl_f,
69 .fo_poll = fifo_poll_f,
70 .fo_kqfilter = fifo_kqfilter_f,
71 .fo_stat = fifo_stat_f,
72 .fo_close = fifo_close_f,
73 .fo_flags = DFLAG_PASSABLE
77 * This structure is associated with the FIFO vnode and stores
78 * the state associated with the FIFO.
81 struct socket *fi_readsock;
82 struct socket *fi_writesock;
87 static vop_print_t fifo_print;
88 static vop_open_t fifo_open;
89 static vop_close_t fifo_close;
90 static vop_ioctl_t fifo_ioctl;
91 static vop_kqfilter_t fifo_kqfilter;
92 static vop_pathconf_t fifo_pathconf;
93 static vop_advlock_t fifo_advlock;
95 static void filt_fifordetach(struct knote *kn);
96 static int filt_fiforead(struct knote *kn, long hint);
97 static void filt_fifowdetach(struct knote *kn);
98 static int filt_fifowrite(struct knote *kn, long hint);
99 static void filt_fifodetach_notsup(struct knote *kn);
100 static int filt_fifo_notsup(struct knote *kn, long hint);
102 static struct filterops fiforead_filtops =
103 { 1, NULL, filt_fifordetach, filt_fiforead };
104 static struct filterops fifowrite_filtops =
105 { 1, NULL, filt_fifowdetach, filt_fifowrite };
106 static struct filterops fifo_notsup_filtops =
107 { 1, NULL, filt_fifodetach_notsup, filt_fifo_notsup };
109 struct vop_vector fifo_specops = {
110 .vop_default = &default_vnodeops,
112 .vop_access = VOP_EBADF,
113 .vop_advlock = fifo_advlock,
114 .vop_close = fifo_close,
115 .vop_create = VOP_PANIC,
116 .vop_getattr = VOP_EBADF,
117 .vop_ioctl = fifo_ioctl,
118 .vop_kqfilter = fifo_kqfilter,
119 .vop_lease = VOP_NULL,
120 .vop_link = VOP_PANIC,
121 .vop_mkdir = VOP_PANIC,
122 .vop_mknod = VOP_PANIC,
123 .vop_open = fifo_open,
124 .vop_pathconf = fifo_pathconf,
125 .vop_print = fifo_print,
126 .vop_read = VOP_PANIC,
127 .vop_readdir = VOP_PANIC,
128 .vop_readlink = VOP_PANIC,
129 .vop_reallocblks = VOP_PANIC,
130 .vop_reclaim = VOP_NULL,
131 .vop_remove = VOP_PANIC,
132 .vop_rename = VOP_PANIC,
133 .vop_rmdir = VOP_PANIC,
134 .vop_setattr = VOP_EBADF,
135 .vop_symlink = VOP_PANIC,
136 .vop_write = VOP_PANIC,
140 MTX_SYSINIT(fifo, &fifo_mtx, "fifo mutex", MTX_DEF);
143 * Dispose of fifo resources.
146 fifo_cleanup(struct vnode *vp)
148 struct fifoinfo *fip = vp->v_fifoinfo;
150 ASSERT_VOP_LOCKED(vp, "fifo_cleanup");
151 if (fip->fi_readers == 0 && fip->fi_writers == 0) {
152 vp->v_fifoinfo = NULL;
153 (void)soclose(fip->fi_readsock);
154 (void)soclose(fip->fi_writesock);
160 * Open called to set up a new instance of a fifo or
161 * to find an active instance of a fifo.
166 struct vop_open_args /* {
169 struct ucred *a_cred;
174 struct vnode *vp = ap->a_vp;
175 struct fifoinfo *fip;
176 struct thread *td = ap->a_td;
177 struct ucred *cred = ap->a_cred;
178 struct file *fp = ap->a_fp;
179 struct socket *rso, *wso;
182 ASSERT_VOP_ELOCKED(vp, "fifo_open");
185 if ((fip = vp->v_fifoinfo) == NULL) {
186 MALLOC(fip, struct fifoinfo *, sizeof(*fip), M_VNODE, M_WAITOK);
187 error = socreate(AF_LOCAL, &rso, SOCK_STREAM, 0, cred, td);
190 fip->fi_readsock = rso;
191 error = socreate(AF_LOCAL, &wso, SOCK_STREAM, 0, cred, td);
194 fip->fi_writesock = wso;
195 error = soconnect2(wso, rso);
204 fip->fi_readers = fip->fi_writers = 0;
205 wso->so_snd.sb_lowat = PIPE_BUF;
206 SOCKBUF_LOCK(&rso->so_rcv);
207 rso->so_rcv.sb_state |= SBS_CANTRCVMORE;
208 SOCKBUF_UNLOCK(&rso->so_rcv);
209 KASSERT(vp->v_fifoinfo == NULL,
210 ("fifo_open: v_fifoinfo race"));
211 vp->v_fifoinfo = fip;
215 * General access to fi_readers and fi_writers is protected using
218 * Protect the increment of fi_readers and fi_writers and the
219 * associated calls to wakeup() with the fifo mutex in addition
220 * to the vnode lock. This allows the vnode lock to be dropped
221 * for the msleep() calls below, and using the fifo mutex with
222 * msleep() prevents the wakeup from being missed.
225 if (ap->a_mode & FREAD) {
227 if (fip->fi_readers == 1) {
228 SOCKBUF_LOCK(&fip->fi_writesock->so_snd);
229 fip->fi_writesock->so_snd.sb_state &= ~SBS_CANTSENDMORE;
230 SOCKBUF_UNLOCK(&fip->fi_writesock->so_snd);
231 if (fip->fi_writers > 0) {
232 wakeup(&fip->fi_writers);
233 sowwakeup(fip->fi_writesock);
237 if (ap->a_mode & FWRITE) {
238 if ((ap->a_mode & O_NONBLOCK) && fip->fi_readers == 0) {
239 mtx_unlock(&fifo_mtx);
243 if (fip->fi_writers == 1) {
244 SOCKBUF_LOCK(&fip->fi_readsock->so_rcv);
245 fip->fi_readsock->so_rcv.sb_state &= ~SBS_CANTRCVMORE;
246 SOCKBUF_UNLOCK(&fip->fi_readsock->so_rcv);
247 if (fip->fi_readers > 0) {
248 wakeup(&fip->fi_readers);
249 sorwakeup(fip->fi_readsock);
253 if ((ap->a_mode & O_NONBLOCK) == 0) {
254 if ((ap->a_mode & FREAD) && fip->fi_writers == 0) {
255 VOP_UNLOCK(vp, 0, td);
256 error = msleep(&fip->fi_readers, &fifo_mtx,
257 PDROP | PCATCH | PSOCK, "fifoor", 0);
258 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
261 if (fip->fi_readers == 0) {
262 socantsendmore(fip->fi_writesock);
269 * We must have got woken up because we had a writer.
270 * That (and not still having one) is the condition
271 * that we must wait for.
274 if ((ap->a_mode & FWRITE) && fip->fi_readers == 0) {
275 VOP_UNLOCK(vp, 0, td);
276 error = msleep(&fip->fi_writers, &fifo_mtx,
277 PDROP | PCATCH | PSOCK, "fifoow", 0);
278 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
281 if (fip->fi_writers == 0) {
282 socantrcvmore(fip->fi_readsock);
288 * We must have got woken up because we had
289 * a reader. That (and not still having one)
290 * is the condition that we must wait for.
295 mtx_unlock(&fifo_mtx);
296 KASSERT(fp != NULL, ("can't fifo/vnode bypass"));
298 KASSERT(fp->f_ops == &badfileops, ("not badfileops in fifo_open"));
300 fp->f_ops = &fifo_ops_f;
306 * Now unused vnode ioctl routine.
311 struct vop_ioctl_args /* {
316 struct ucred *a_cred;
321 printf("WARNING: fifo_ioctl called unexpectedly\n");
326 * Now unused vnode kqfilter routine.
331 struct vop_kqfilter_args /* {
337 printf("WARNING: fifo_kqfilter called unexpectedly\n");
342 filt_fifordetach(struct knote *kn)
344 struct socket *so = (struct socket *)kn->kn_hook;
346 SOCKBUF_LOCK(&so->so_rcv);
347 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
348 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
349 so->so_rcv.sb_flags &= ~SB_KNOTE;
350 SOCKBUF_UNLOCK(&so->so_rcv);
354 filt_fiforead(struct knote *kn, long hint)
356 struct socket *so = (struct socket *)kn->kn_hook;
358 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
359 kn->kn_data = so->so_rcv.sb_cc;
360 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
361 kn->kn_flags |= EV_EOF;
364 kn->kn_flags &= ~EV_EOF;
365 return (kn->kn_data > 0);
370 filt_fifowdetach(struct knote *kn)
372 struct socket *so = (struct socket *)kn->kn_hook;
374 SOCKBUF_LOCK(&so->so_snd);
375 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
376 if (knlist_empty(&so->so_snd.sb_sel.si_note))
377 so->so_snd.sb_flags &= ~SB_KNOTE;
378 SOCKBUF_UNLOCK(&so->so_snd);
382 filt_fifowrite(struct knote *kn, long hint)
384 struct socket *so = (struct socket *)kn->kn_hook;
386 SOCKBUF_LOCK_ASSERT(&so->so_snd);
387 kn->kn_data = sbspace(&so->so_snd);
388 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
389 kn->kn_flags |= EV_EOF;
392 kn->kn_flags &= ~EV_EOF;
393 return (kn->kn_data >= so->so_snd.sb_lowat);
398 filt_fifodetach_notsup(struct knote *kn)
404 filt_fifo_notsup(struct knote *kn, long hint)
411 * Device close routine
416 struct vop_close_args /* {
419 struct ucred *a_cred;
423 struct vnode *vp = ap->a_vp;
424 struct fifoinfo *fip = vp->v_fifoinfo;
426 ASSERT_VOP_LOCKED(vp, "fifo_close");
428 printf("fifo_close: no v_fifoinfo %p\n", vp);
431 if (ap->a_fflag & FREAD) {
433 if (fip->fi_readers == 0)
434 socantsendmore(fip->fi_writesock);
436 if (ap->a_fflag & FWRITE) {
438 if (fip->fi_writers == 0)
439 socantrcvmore(fip->fi_readsock);
446 * Print out internal contents of a fifo vnode.
452 register struct fifoinfo *fip = vp->v_fifoinfo;
455 printf(", NULL v_fifoinfo");
458 printf(", fifo with %ld readers and %ld writers",
459 fip->fi_readers, fip->fi_writers);
464 * Print out the contents of a fifo vnode.
468 struct vop_print_args /* {
472 fifo_printinfo(ap->a_vp);
478 * Return POSIX pathconf information applicable to fifo's.
482 struct vop_pathconf_args /* {
489 switch (ap->a_name) {
491 *ap->a_retval = LINK_MAX;
494 *ap->a_retval = PIPE_BUF;
496 case _PC_CHOWN_RESTRICTED:
506 * Fifo advisory byte-level locks.
511 struct vop_advlock_args /* {
520 return (ap->a_flags & F_FLOCK ? EOPNOTSUPP : EINVAL);
524 fifo_close_f(struct file *fp, struct thread *td)
527 return (vnops.fo_close(fp, td));
531 * The implementation of ioctl() for named fifos is complicated by the fact
532 * that we permit O_RDWR fifo file descriptors, meaning that the actions of
533 * ioctls may have to be applied to both the underlying sockets rather than
534 * just one. The original implementation simply forward the ioctl to one
535 * or both sockets based on fp->f_flag. We now consider each ioctl
536 * separately, as the composition effect requires careful ordering.
538 * We do not blindly pass all ioctls through to the socket in order to avoid
539 * providing unnecessary ioctls that might be improperly depended on by
540 * applications (such as socket-specific, routing, and interface ioctls).
542 * Unlike sys_pipe.c, fifos do not implement the deprecated TIOCSPGRP and
543 * TIOCGPGRP ioctls. Earlier implementations of fifos did forward SIOCSPGRP
544 * and SIOCGPGRP ioctls, so we might need to re-add those here.
547 fifo_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred,
551 struct file filetmp; /* Local, so need not be locked. */
560 * Non-blocking I/O is implemented at the fifo layer using
561 * MSG_NBIO, so does not need to be forwarded down the stack.
569 * These socket ioctls don't have any ordering requirements,
570 * so are called in an arbitrary order, and only on the
571 * sockets indicated by the file descriptor rights.
573 * XXXRW: If O_RDWR and the read socket accepts an ioctl but
574 * the write socket doesn't, the socketpair is left in an
575 * inconsistent state.
577 if (fp->f_flag & FREAD) {
578 filetmp.f_data = fi->fi_readsock;
579 filetmp.f_cred = cred;
580 error = soo_ioctl(&filetmp, com, data, cred, td);
584 if (fp->f_flag & FWRITE) {
585 filetmp.f_data = fi->fi_writesock;
586 filetmp.f_cred = cred;
587 error = soo_ioctl(&filetmp, com, data, cred, td);
593 * FIONREAD will return 0 for non-readable descriptors, and
594 * the results of FIONREAD on the read socket for readable
597 if (!(fp->f_flag & FREAD)) {
601 filetmp.f_data = fi->fi_readsock;
602 filetmp.f_cred = cred;
603 return (soo_ioctl(&filetmp, com, data, cred, td));
611 * Because fifos are now a file descriptor layer object, EVFILT_VNODE is not
612 * implemented. Likely, fifo_kqfilter() should be removed, and
613 * fifo_kqfilter_f() should know how to forward the request to the underling
614 * vnode using f_vnode in the file descriptor here.
617 fifo_kqfilter_f(struct file *fp, struct knote *kn)
626 * If a filter is requested that is not supported by this file
627 * descriptor, don't return an error, but also don't ever generate an
630 if ((kn->kn_filter == EVFILT_READ) && !(fp->f_flag & FREAD)) {
631 kn->kn_fop = &fifo_notsup_filtops;
635 if ((kn->kn_filter == EVFILT_WRITE) && !(fp->f_flag & FWRITE)) {
636 kn->kn_fop = &fifo_notsup_filtops;
640 switch (kn->kn_filter) {
642 kn->kn_fop = &fiforead_filtops;
643 so = fi->fi_readsock;
647 kn->kn_fop = &fifowrite_filtops;
648 so = fi->fi_writesock;
655 kn->kn_hook = (caddr_t)so;
658 knlist_add(&sb->sb_sel.si_note, kn, 1);
659 sb->sb_flags |= SB_KNOTE;
666 fifo_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
668 struct fifoinfo *fip;
670 int levents, revents = 0;
674 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM | POLLRDBAND);
675 if ((fp->f_flag & FREAD) && levents) {
677 * If POLLIN or POLLRDNORM is requested and POLLINIGNEOF is
678 * not, then convert the first two to the last one. This
679 * tells the socket poll function to ignore EOF so that we
680 * block if there is no writer (and no data). Callers can
681 * set POLLINIGNEOF to get non-blocking behavior.
683 if (levents & (POLLIN | POLLRDNORM) &&
684 !(levents & POLLINIGNEOF)) {
685 levents &= ~(POLLIN | POLLRDNORM);
686 levents |= POLLINIGNEOF;
689 filetmp.f_data = fip->fi_readsock;
690 filetmp.f_cred = cred;
691 revents |= soo_poll(&filetmp, levents, cred, td);
693 /* Reverse the above conversion. */
694 if ((revents & POLLINIGNEOF) && !(events & POLLINIGNEOF)) {
695 revents |= (events & (POLLIN | POLLRDNORM));
696 revents &= ~POLLINIGNEOF;
699 levents = events & (POLLOUT | POLLWRNORM | POLLWRBAND);
700 if ((fp->f_flag & FWRITE) && levents) {
701 filetmp.f_data = fip->fi_writesock;
702 filetmp.f_cred = cred;
703 revents |= soo_poll(&filetmp, levents, cred, td);
709 fifo_read_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
711 struct fifoinfo *fip;
715 KASSERT(uio->uio_rw == UIO_READ,("fifo_read mode"));
716 if (uio->uio_resid == 0)
718 sflags = (fp->f_flag & FNONBLOCK) ? MSG_NBIO : 0;
719 return (soreceive(fip->fi_readsock, NULL, uio, NULL, NULL, &sflags));
723 fifo_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
726 return (vnops.fo_stat(fp, sb, cred, td));
730 fifo_write_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
732 struct fifoinfo *fip;
736 KASSERT(uio->uio_rw == UIO_WRITE,("fifo_write mode"));
737 sflags = (fp->f_flag & FNONBLOCK) ? MSG_NBIO : 0;
738 return (sosend(fip->fi_writesock, NULL, uio, 0, NULL, sflags, td));