2 * Copyright (c) 1989, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)spec_vnops.c 8.14 (Berkeley) 5/21/95
34 * $FreeBSD: src/sys/miscfs/specfs/spec_vnops.c,v 1.131.2.4 2001/02/26 04:23:20 jlemon Exp $
35 * $DragonFly: src/sys/vfs/specfs/spec_vnops.c,v 1.50 2006/12/23 00:41:30 swildner Exp $
38 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
44 #include <sys/device.h>
45 #include <sys/mount.h>
46 #include <sys/vnode.h>
48 #include <sys/fcntl.h>
49 #include <sys/vmmeter.h>
53 #include <vm/vm_object.h>
54 #include <vm/vm_page.h>
55 #include <vm/vm_pager.h>
57 #include <machine/limits.h>
61 #include <sys/thread2.h>
63 static int spec_advlock (struct vop_advlock_args *);
64 static int spec_bmap (struct vop_bmap_args *);
65 static int spec_close (struct vop_close_args *);
66 static int spec_freeblks (struct vop_freeblks_args *);
67 static int spec_fsync (struct vop_fsync_args *);
68 static int spec_getpages (struct vop_getpages_args *);
69 static int spec_inactive (struct vop_inactive_args *);
70 static int spec_ioctl (struct vop_ioctl_args *);
71 static int spec_open (struct vop_open_args *);
72 static int spec_poll (struct vop_poll_args *);
73 static int spec_kqfilter (struct vop_kqfilter_args *);
74 static int spec_print (struct vop_print_args *);
75 static int spec_read (struct vop_read_args *);
76 static int spec_strategy (struct vop_strategy_args *);
77 static int spec_write (struct vop_write_args *);
79 struct vop_ops spec_vnode_vops = {
80 .vop_default = vop_defaultop,
81 .vop_access = (void *)vop_ebadf,
82 .vop_advlock = spec_advlock,
83 .vop_bmap = spec_bmap,
84 .vop_close = spec_close,
85 .vop_old_create = (void *)vop_panic,
86 .vop_freeblks = spec_freeblks,
87 .vop_fsync = spec_fsync,
88 .vop_getpages = spec_getpages,
89 .vop_inactive = spec_inactive,
90 .vop_ioctl = spec_ioctl,
91 .vop_old_link = (void *)vop_panic,
92 .vop_old_mkdir = (void *)vop_panic,
93 .vop_old_mknod = (void *)vop_panic,
94 .vop_open = spec_open,
95 .vop_pathconf = vop_stdpathconf,
96 .vop_poll = spec_poll,
97 .vop_kqfilter = spec_kqfilter,
98 .vop_print = spec_print,
99 .vop_read = spec_read,
100 .vop_readdir = (void *)vop_panic,
101 .vop_readlink = (void *)vop_panic,
102 .vop_reallocblks = (void *)vop_panic,
103 .vop_reclaim = (void *)vop_null,
104 .vop_old_remove = (void *)vop_panic,
105 .vop_old_rename = (void *)vop_panic,
106 .vop_old_rmdir = (void *)vop_panic,
107 .vop_setattr = (void *)vop_ebadf,
108 .vop_strategy = spec_strategy,
109 .vop_old_symlink = (void *)vop_panic,
110 .vop_write = spec_write
113 struct vop_ops *spec_vnode_vops_p = &spec_vnode_vops;
115 VNODEOP_SET(spec_vnode_vops);
117 extern int dev_ref_debug;
123 spec_vnoperate(struct vop_generic_args *ap)
125 return (VOCALL(&spec_vnode_vops, ap));
128 static void spec_getpages_iodone (struct bio *bio);
131 * Open a special file.
133 * spec_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
138 spec_open(struct vop_open_args *ap)
140 struct vnode *vp = ap->a_vp;
143 int isblk = (vp->v_type == VBLK) ? 1 : 0;
147 * Don't allow open if fs is mounted -nodev.
149 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
153 * Resolve the device. If the vnode is already open v_rdev may
154 * already be resolved. However, if the device changes out from
155 * under us we report it (and, for now, we allow it). Since
156 * v_release_rdev() zero's v_opencount, we have to save and restore
157 * it when replacing the rdev reference.
159 if (vp->v_rdev != NULL) {
160 dev = udev2dev(vp->v_udev, isblk);
161 if (dev != vp->v_rdev) {
162 int oc = vp->v_opencount;
164 "Warning: spec_open: dev %s was lost",
165 vp->v_rdev->si_name);
167 error = v_associate_rdev(vp,
168 udev2dev(vp->v_udev, isblk));
170 kprintf(", reacquisition failed\n");
172 vp->v_opencount = oc;
173 kprintf(", reacquisition successful\n");
179 error = v_associate_rdev(vp, udev2dev(vp->v_udev, isblk));
185 * Prevent degenerate open/close sequences from nulling out rdev.
188 KKASSERT(dev != NULL);
191 * Make this field valid before any I/O in ->d_open. XXX the
192 * device itself should probably be required to initialize
193 * this field in d_open.
195 if (!dev->si_iosize_max)
196 dev->si_iosize_max = DFLTPHYS;
199 * XXX: Disks get special billing here, but it is mostly wrong.
200 * XXX: diskpartitions can overlap and the real checks should
201 * XXX: take this into account, and consequently they need to
202 * XXX: live in the diskslicing code. Some checks do.
204 if (vn_isdisk(vp, NULL) && ap->a_cred != FSCRED &&
205 (ap->a_mode & FWRITE)) {
207 * Never allow opens for write if the device is mounted R/W
209 if (vp->v_rdev && vp->v_rdev->si_mountpoint &&
210 !(vp->v_rdev->si_mountpoint->mnt_flag & MNT_RDONLY)) {
216 * When running in secure mode, do not allow opens
217 * for writing if the device is mounted
219 if (securelevel >= 1 && vfs_mountedon(vp)) {
225 * When running in very secure mode, do not allow
226 * opens for writing of any devices.
228 if (securelevel >= 2) {
234 /* XXX: Special casing of ttys for deadfs. Probably redundant */
235 if (dev_dflags(dev) & D_TTY)
236 vp->v_flag |= VISTTY;
239 * dev_dopen() is always called for each open. dev_dclose() is
240 * only called for the last close unless D_TRACKCLOSE is set.
243 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred);
244 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
249 if (dev_dflags(dev) & D_TTY) {
254 kprintf("Warning:%s: no t_stop, using nottystop\n", devtoname(dev));
255 tp->t_stop = nottystop;
261 * If this is 'disk' or disk-like device, associate a VM object
264 if (vn_isdisk(vp, NULL)) {
265 if (!dev->si_bsize_phys)
266 dev->si_bsize_phys = DEV_BSIZE;
267 vinitvmio(vp, IDX_TO_OFF(INT_MAX));
269 if ((dev_dflags(dev) & D_DISK) == 0) {
272 kprintf("WARNING: driver %s should register devices with make_dev() (cdev_t = \"%s\")\n",
278 * If we were handed a file pointer we may be able to install a
279 * shortcut which issues device read and write operations directly
280 * from the fileops rather then having to go through spec_read()
284 vn_setspecops(ap->a_fp);
287 kprintf("spec_open: %s %d\n", dev->si_name, vp->v_opencount);
290 if (vp->v_opencount == 0)
301 * spec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
302 * struct ucred *a_cred)
306 spec_read(struct vop_read_args *ap)
319 if (dev == NULL) /* device was revoked */
321 if (uio->uio_resid == 0)
325 error = dev_dread(dev, uio, ap->a_ioflag);
326 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
333 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
334 * struct ucred *a_cred)
338 spec_write(struct vop_write_args *ap)
351 if (dev == NULL) /* device was revoked */
355 error = dev_dwrite(dev, uio, ap->a_ioflag);
356 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
361 * Device ioctl operation.
363 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
364 * int a_fflag, struct ucred *a_cred)
368 spec_ioctl(struct vop_ioctl_args *ap)
372 if ((dev = ap->a_vp->v_rdev) == NULL)
373 return (EBADF); /* device was revoked */
375 return (dev_dioctl(dev, ap->a_command, ap->a_data,
376 ap->a_fflag, ap->a_cred));
380 * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred)
384 spec_poll(struct vop_poll_args *ap)
388 if ((dev = ap->a_vp->v_rdev) == NULL)
389 return (EBADF); /* device was revoked */
390 return (dev_dpoll(dev, ap->a_events));
394 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
398 spec_kqfilter(struct vop_kqfilter_args *ap)
402 if ((dev = ap->a_vp->v_rdev) == NULL)
403 return (EBADF); /* device was revoked */
404 return (dev_dkqfilter(dev, ap->a_kn));
408 * Synch buffers associated with a block device
410 * spec_fsync(struct vnode *a_vp, struct ucred *a_cred,
411 * int a_waitfor, struct thread *a_td)
415 spec_fsync(struct vop_fsync_args *ap)
417 struct vnode *vp = ap->a_vp;
420 if (!vn_isdisk(vp, NULL))
424 * Flush all dirty buffers associated with a block device.
426 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
431 * spec_inactive(struct vnode *a_vp, struct thread *a_td)
434 spec_inactive(struct vop_inactive_args *ap)
440 * Just call the device strategy routine
442 * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
445 spec_strategy(struct vop_strategy_args *ap)
447 struct bio *bio = ap->a_bio;
448 struct buf *bp = bio->bio_buf;
452 if (bp->b_cmd != BUF_CMD_READ &&
453 (LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start) {
454 (*bioops.io_start)(bp);
458 * Collect statistics on synchronous and asynchronous read
459 * and write counts for disks that have associated filesystems.
462 KKASSERT(vp->v_rdev != NULL); /* XXX */
463 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
464 if (bp->b_cmd == BUF_CMD_READ) {
465 if (bp->b_flags & B_ASYNC)
466 mp->mnt_stat.f_asyncreads++;
468 mp->mnt_stat.f_syncreads++;
470 if (bp->b_flags & B_ASYNC)
471 mp->mnt_stat.f_asyncwrites++;
473 mp->mnt_stat.f_syncwrites++;
476 dev_dstrategy_chain(vp->v_rdev, bio);
481 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
484 spec_freeblks(struct vop_freeblks_args *ap)
489 * XXX: This assumes that strategy does the deed right away.
490 * XXX: this may not be TRTTD.
492 KKASSERT(ap->a_vp->v_rdev != NULL);
493 if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0)
495 bp = geteblk(ap->a_length);
496 bp->b_cmd = BUF_CMD_FREEBLKS;
497 bp->b_bio1.bio_offset = ap->a_offset;
498 bp->b_bcount = ap->a_length;
499 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
504 * Implement degenerate case where the block requested is the block
505 * returned, and assume that the entire device is contiguous in regards
506 * to the contiguous block range (runp and runb).
508 * spec_bmap(struct vnode *a_vp, off_t a_loffset, struct vnode **a_vpp,
509 * off_t *a_doffsetp, int *a_runp, int *a_runb)
512 spec_bmap(struct vop_bmap_args *ap)
514 struct vnode *vp = ap->a_vp;
516 if (ap->a_vpp != NULL)
518 if (ap->a_doffsetp != NULL)
519 *ap->a_doffsetp = ap->a_loffset;
520 if (ap->a_runp != NULL)
521 *ap->a_runp = MAXBSIZE;
522 if (ap->a_runb != NULL) {
523 if (ap->a_loffset < MAXBSIZE)
524 *ap->a_runb = (int)ap->a_loffset;
526 *ap->a_runb = MAXBSIZE;
532 * Device close routine
534 * spec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
535 * struct thread *a_td)
537 * NOTE: the vnode may or may not be locked on call.
541 spec_close(struct vop_close_args *ap)
543 struct proc *p = curproc;
544 struct vnode *vp = ap->a_vp;
545 cdev_t dev = vp->v_rdev;
550 * Hack: a tty device that is a controlling terminal
551 * has a reference from the session structure.
552 * We cannot easily tell that a character device is
553 * a controlling terminal, unless it is the closing
554 * process' controlling terminal. In that case,
555 * if the reference count is 2 (this last descriptor
556 * plus the session), release the reference from the session.
558 * It is possible for v_opencount to be 0 or 1 in this case, 0
559 * because the tty might have been revoked.
563 if (vcount(vp) == 2 && vp->v_opencount <= 1 &&
564 p && vp == p->p_session->s_ttyvp) {
565 p->p_session->s_ttyvp = NULL;
570 * Vnodes can be opened and close multiple times. Do not really
571 * close the device unless (1) it is being closed forcibly,
572 * (2) the device wants to track closes, or (3) this is the last
573 * vnode doing its last close on the device.
575 * XXX the VXLOCK (force close) case can leave vnodes referencing
578 if (dev && ((vp->v_flag & VRECLAIMED) ||
579 (dev_dflags(dev) & D_TRACKCLOSE) ||
580 (vcount(vp) <= 1 && vp->v_opencount == 1))) {
582 if (vn_islocked(vp)) {
586 error = dev_dclose(dev, ap->a_fflag, S_IFCHR);
588 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
594 * Track the actual opens and closes on the vnode. The last close
595 * disassociates the rdev. If the rdev is already disassociated
596 * the vnode might have been revoked and no further opencount
600 /*KKASSERT(vp->v_opencount > 0);*/
602 kprintf("spec_close: %s %d\n",
603 dev->si_name, vp->v_opencount - 1);
605 if (vp->v_opencount == 1)
614 * Print out the contents of a special device vnode.
616 * spec_print(struct vnode *a_vp)
619 spec_print(struct vop_print_args *ap)
621 kprintf("tag VT_NON, dev %s\n", devtoname(ap->a_vp->v_rdev));
626 * Special device advisory byte-level locks.
628 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
629 * struct flock *a_fl, int a_flags)
633 spec_advlock(struct vop_advlock_args *ap)
635 return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
639 spec_getpages_iodone(struct bio *bio)
641 bio->bio_buf->b_cmd = BUF_CMD_DONE;
642 wakeup(bio->bio_buf);
646 spec_getpages(struct vop_getpages_args *ap)
654 int toff, nextoff, nread;
655 struct vnode *vp = ap->a_vp;
660 pcount = round_page(ap->a_count) / PAGE_SIZE;
663 * Calculate the offset of the transfer and do sanity check.
665 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
668 * Round up physical size for real devices. We cannot round using
669 * v_mount's block size data because v_mount has nothing to do with
670 * the device. i.e. it's usually '/dev'. We need the physical block
671 * size for the device itself.
673 * We can't use v_rdev->si_mountpoint because it only exists when the
674 * block device is mounted. However, we can use v_rdev.
677 if (vn_isdisk(vp, NULL))
678 blksiz = vp->v_rdev->si_bsize_phys;
682 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
685 kva = (vm_offset_t)bp->b_data;
688 * Map the pages to be read into the kva.
690 pmap_qenter(kva, ap->a_m, pcount);
692 /* Build a minimal buffer header. */
693 bp->b_cmd = BUF_CMD_READ;
696 bp->b_runningbufspace = size;
697 runningbufspace += bp->b_runningbufspace;
699 bp->b_bio1.bio_offset = offset;
700 bp->b_bio1.bio_done = spec_getpages_iodone;
702 mycpu->gd_cnt.v_vnodein++;
703 mycpu->gd_cnt.v_vnodepgsin += pcount;
706 vn_strategy(ap->a_vp, &bp->b_bio1);
710 /* We definitely need to be at splbio here. */
711 while (bp->b_cmd != BUF_CMD_DONE)
712 tsleep(bp, 0, "spread", 0);
716 if (bp->b_flags & B_ERROR) {
724 * If EOF is encountered we must zero-extend the result in order
725 * to ensure that the page does not contain garabge. When no
726 * error occurs, an early EOF is indicated if b_bcount got truncated.
727 * b_resid is relative to b_bcount and should be 0, but some devices
728 * might indicate an EOF with b_resid instead of truncating b_bcount.
730 nread = bp->b_bcount - bp->b_resid;
731 if (nread < ap->a_count)
732 bzero((caddr_t)kva + nread, ap->a_count - nread);
733 pmap_qremove(kva, pcount);
736 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
737 nextoff = toff + PAGE_SIZE;
740 m->flags &= ~PG_ZERO;
742 if (nextoff <= nread) {
743 m->valid = VM_PAGE_BITS_ALL;
745 } else if (toff < nread) {
747 * Since this is a VM request, we have to supply the
748 * unaligned offset to allow vm_page_set_validclean()
749 * to zero sub-DEV_BSIZE'd portions of the page.
751 vm_page_set_validclean(m, 0, nread - toff);
757 if (i != ap->a_reqpage) {
759 * Just in case someone was asking for this page we
760 * now tell them that it is ok to use.
762 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
764 if (m->flags & PG_WANTED) {
767 vm_page_deactivate(m);
776 } else if (m->valid) {
779 * Since this is a VM request, we need to make the
780 * entire page presentable by zeroing invalid sections.
782 if (m->valid != VM_PAGE_BITS_ALL)
783 vm_page_zero_invalid(m, FALSE);
787 m = ap->a_m[ap->a_reqpage];
789 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
790 devtoname(vp->v_rdev), error, bp, bp->b_vp);
792 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
793 size, bp->b_resid, ap->a_count, m->valid);
795 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
796 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
798 * Free the buffer header back to the swap buffer pool.
801 return VM_PAGER_ERROR;
804 * Free the buffer header back to the swap buffer pool.