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 1.28 2005/09/14 01:13:46 dillon 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>
59 #include <sys/thread2.h>
61 static int spec_advlock (struct vop_advlock_args *);
62 static int spec_bmap (struct vop_bmap_args *);
63 static int spec_close (struct vop_close_args *);
64 static int spec_freeblks (struct vop_freeblks_args *);
65 static int spec_fsync (struct vop_fsync_args *);
66 static int spec_getpages (struct vop_getpages_args *);
67 static int spec_inactive (struct vop_inactive_args *);
68 static int spec_ioctl (struct vop_ioctl_args *);
69 static int spec_open (struct vop_open_args *);
70 static int spec_poll (struct vop_poll_args *);
71 static int spec_kqfilter (struct vop_kqfilter_args *);
72 static int spec_print (struct vop_print_args *);
73 static int spec_read (struct vop_read_args *);
74 static int spec_strategy (struct vop_strategy_args *);
75 static int spec_write (struct vop_write_args *);
77 struct vop_ops *spec_vnode_vops;
78 struct vnodeopv_entry_desc spec_vnodeop_entries[] = {
79 { &vop_default_desc, vop_defaultop },
80 { &vop_access_desc, vop_ebadf },
81 { &vop_advlock_desc, (vnodeopv_entry_t) spec_advlock },
82 { &vop_bmap_desc, (vnodeopv_entry_t) spec_bmap },
83 { &vop_close_desc, (vnodeopv_entry_t) spec_close },
84 { &vop_old_create_desc, vop_panic },
85 { &vop_freeblks_desc, (vnodeopv_entry_t) spec_freeblks },
86 { &vop_fsync_desc, (vnodeopv_entry_t) spec_fsync },
87 { &vop_getpages_desc, (vnodeopv_entry_t) spec_getpages },
88 { &vop_inactive_desc, (vnodeopv_entry_t) spec_inactive },
89 { &vop_ioctl_desc, (vnodeopv_entry_t) spec_ioctl },
90 { &vop_lease_desc, vop_null },
91 { &vop_old_link_desc, vop_panic },
92 { &vop_old_mkdir_desc, vop_panic },
93 { &vop_old_mknod_desc, vop_panic },
94 { &vop_open_desc, (vnodeopv_entry_t) spec_open },
95 { &vop_pathconf_desc, (vnodeopv_entry_t) vop_stdpathconf },
96 { &vop_poll_desc, (vnodeopv_entry_t) spec_poll },
97 { &vop_kqfilter_desc, (vnodeopv_entry_t) spec_kqfilter },
98 { &vop_print_desc, (vnodeopv_entry_t) spec_print },
99 { &vop_read_desc, (vnodeopv_entry_t) spec_read },
100 { &vop_readdir_desc, vop_panic },
101 { &vop_readlink_desc, vop_panic },
102 { &vop_reallocblks_desc, vop_panic },
103 { &vop_reclaim_desc, vop_null },
104 { &vop_old_remove_desc, vop_panic },
105 { &vop_old_rename_desc, vop_panic },
106 { &vop_old_rmdir_desc, vop_panic },
107 { &vop_setattr_desc, vop_ebadf },
108 { &vop_strategy_desc, (vnodeopv_entry_t) spec_strategy },
109 { &vop_old_symlink_desc, vop_panic },
110 { &vop_write_desc, (vnodeopv_entry_t) spec_write },
113 static struct vnodeopv_desc spec_vnodeop_opv_desc =
114 { &spec_vnode_vops, spec_vnodeop_entries };
115 VNODEOP_SET(spec_vnodeop_opv_desc);
117 extern int dev_ref_debug;
120 * spec_vnoperate(struct vnodeop_desc *a_desc, ...)
123 spec_vnoperate(struct vop_generic_args *ap)
125 return (VOCALL(spec_vnode_vops, ap));
128 static void spec_getpages_iodone (struct buf *bp);
131 * Open a special file.
133 * spec_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
134 * struct thread *a_td)
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 printf(", reacquisition failed\n");
172 vp->v_opencount = oc;
173 printf(", reacquisition successful\n");
179 error = v_associate_rdev(vp, udev2dev(vp->v_udev, isblk));
185 * Prevent degenerate open/close sequences from nulling out rdev.
189 KKASSERT(dev != NULL);
192 * Make this field valid before any I/O in ->d_open. XXX the
193 * device itself should probably be required to initialize
194 * this field in d_open.
196 if (!dev->si_iosize_max)
197 dev->si_iosize_max = DFLTPHYS;
200 * XXX: Disks get special billing here, but it is mostly wrong.
201 * XXX: diskpartitions can overlap and the real checks should
202 * XXX: take this into account, and consequently they need to
203 * XXX: live in the diskslicing code. Some checks do.
205 if (vn_isdisk(vp, NULL) && ap->a_cred != FSCRED &&
206 (ap->a_mode & FWRITE)) {
208 * Never allow opens for write if the device is mounted R/W
210 if (vp->v_rdev && vp->v_rdev->si_mountpoint &&
211 !(vp->v_rdev->si_mountpoint->mnt_flag & MNT_RDONLY)) {
217 * When running in secure mode, do not allow opens
218 * for writing if the device is mounted
220 if (securelevel >= 1 && vfs_mountedon(vp)) {
226 * When running in very secure mode, do not allow
227 * opens for writing of any devices.
229 if (securelevel >= 2) {
235 /* XXX: Special casing of ttys for deadfs. Probably redundant */
236 if (dev_dflags(dev) & D_TTY)
237 vp->v_flag |= VISTTY;
240 * dev_dopen() is always called for each open. dev_dclose() is
241 * only called for the last close unless D_TRACKCLOSE is set.
243 VOP_UNLOCK(vp, 0, ap->a_td);
244 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_td);
245 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
250 if (dev_dflags(dev) & D_TTY) {
255 printf("Warning:%s: no t_stop, using nottystop\n", devtoname(dev));
256 tp->t_stop = nottystop;
261 if (vn_isdisk(vp, NULL)) {
262 if (!dev->si_bsize_phys)
263 dev->si_bsize_phys = DEV_BSIZE;
265 if ((dev_dflags(dev) & D_DISK) == 0) {
268 printf("WARNING: driver %s should register devices with make_dev() (dev_t = \"%s\")\n",
274 * If we were handed a file pointer we may be able to install a
275 * shortcut which issues device read and write operations directly
276 * from the fileops rather then having to go through spec_read()
280 vn_setspecops(ap->a_fp);
283 printf("spec_open: %s %d\n", dev->si_name, vp->v_opencount);
286 KKASSERT(vp->v_opencount > 0);
287 if (--vp->v_opencount == 0)
296 * spec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
297 * struct ucred *a_cred)
301 spec_read(struct vop_read_args *ap)
314 if (dev == NULL) /* device was revoked */
316 if (uio->uio_resid == 0)
319 VOP_UNLOCK(vp, 0, td);
320 error = dev_dread(dev, uio, ap->a_ioflag);
321 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
328 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
329 * struct ucred *a_cred)
333 spec_write(struct vop_write_args *ap)
346 if (dev == NULL) /* device was revoked */
349 VOP_UNLOCK(vp, 0, td);
350 error = dev_dwrite(dev, uio, ap->a_ioflag);
351 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
356 * Device ioctl operation.
358 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
359 * int a_fflag, struct ucred *a_cred, struct thread *a_td)
363 spec_ioctl(struct vop_ioctl_args *ap)
367 if ((dev = ap->a_vp->v_rdev) == NULL)
368 return (EBADF); /* device was revoked */
370 return (dev_dioctl(dev, ap->a_command, ap->a_data,
371 ap->a_fflag, ap->a_td));
375 * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred,
376 * struct thread *a_td)
380 spec_poll(struct vop_poll_args *ap)
384 if ((dev = ap->a_vp->v_rdev) == NULL)
385 return (EBADF); /* device was revoked */
386 return (dev_dpoll(dev, ap->a_events, ap->a_td));
390 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
394 spec_kqfilter(struct vop_kqfilter_args *ap)
398 if ((dev = ap->a_vp->v_rdev) == NULL)
399 return (EBADF); /* device was revoked */
400 return (dev_dkqfilter(dev, ap->a_kn));
404 * Synch buffers associated with a block device
406 * spec_fsync(struct vnode *a_vp, struct ucred *a_cred,
407 * int a_waitfor, struct thread *a_td)
411 spec_fsync(struct vop_fsync_args *ap)
413 struct vnode *vp = ap->a_vp;
416 if (!vn_isdisk(vp, NULL))
420 * Flush all dirty buffers associated with a block device.
422 error = vfsync(vp, ap->a_waitfor, 10000, (daddr_t)-1, NULL, NULL);
427 * spec_inactive(struct vnode *a_vp, struct thread *a_td)
430 spec_inactive(struct vop_inactive_args *ap)
436 * Just call the device strategy routine
438 * spec_strategy(struct vnode *a_vp, struct buf *a_bp)
441 spec_strategy(struct vop_strategy_args *ap)
448 if (((bp->b_flags & B_READ) == 0) &&
449 (LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
450 (*bioops.io_start)(bp);
453 * Collect statistics on synchronous and asynchronous read
454 * and write counts for disks that have associated filesystems.
457 KKASSERT(vp->v_rdev != NULL); /* XXX */
458 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
459 if ((bp->b_flags & B_READ) == 0) {
460 if (bp->b_lock.lk_lockholder == LK_KERNTHREAD)
461 mp->mnt_stat.f_asyncwrites++;
463 mp->mnt_stat.f_syncwrites++;
465 if (bp->b_lock.lk_lockholder == LK_KERNTHREAD)
466 mp->mnt_stat.f_asyncreads++;
468 mp->mnt_stat.f_syncreads++;
471 bp->b_dev = vp->v_rdev;
477 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
480 spec_freeblks(struct vop_freeblks_args *ap)
485 * XXX: This assumes that strategy does the deed right away.
486 * XXX: this may not be TRTTD.
488 KKASSERT(ap->a_vp->v_rdev != NULL);
489 if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0)
491 bp = geteblk(ap->a_length);
492 bp->b_flags |= B_FREEBUF;
493 bp->b_dev = ap->a_vp->v_rdev;
494 bp->b_blkno = ap->a_addr;
495 bp->b_offset = dbtob(ap->a_addr);
496 bp->b_bcount = ap->a_length;
502 * Implement degenerate case where the block requested is the block
503 * returned, and assume that the entire device is contiguous in regards
504 * to the contiguous block range (runp and runb).
506 * spec_bmap(struct vnode *a_vp, daddr_t a_bn, struct vnode **a_vpp,
507 * daddr_t *a_bnp, int *a_runp, int *a_runb)
510 spec_bmap(struct vop_bmap_args *ap)
512 struct vnode *vp = ap->a_vp;
516 if (ap->a_vpp != NULL)
518 if (ap->a_bnp != NULL)
519 *ap->a_bnp = ap->a_bn;
520 if (vp->v_mount != NULL)
521 runp = runb = MAXBSIZE / vp->v_mount->mnt_stat.f_iosize;
522 if (ap->a_runp != NULL)
524 if (ap->a_runb != NULL)
530 * Device close routine
532 * spec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
533 * struct thread *a_td)
535 * NOTE: the vnode may or may not be locked on call.
539 spec_close(struct vop_close_args *ap)
541 struct proc *p = ap->a_td->td_proc;
542 struct vnode *vp = ap->a_vp;
543 dev_t dev = vp->v_rdev;
548 * Hack: a tty device that is a controlling terminal
549 * has a reference from the session structure.
550 * We cannot easily tell that a character device is
551 * a controlling terminal, unless it is the closing
552 * process' controlling terminal. In that case,
553 * if the reference count is 2 (this last descriptor
554 * plus the session), release the reference from the session.
556 * It is possible for v_opencount to be 0 or 1 in this case, 0
557 * because the tty might have been revoked.
561 if (vcount(vp) == 2 && vp->v_opencount <= 1 &&
562 p && vp == p->p_session->s_ttyvp) {
563 p->p_session->s_ttyvp = NULL;
568 * Vnodes can be opened and close multiple times. Do not really
569 * close the device unless (1) it is being closed forcibly,
570 * (2) the device wants to track closes, or (3) this is the last
571 * vnode doing its last close on the device.
573 * XXX the VXLOCK (force close) case can leave vnodes referencing
576 if (dev && ((vp->v_flag & VRECLAIMED) ||
577 (dev_dflags(dev) & D_TRACKCLOSE) ||
578 (vcount(vp) <= 1 && vp->v_opencount == 1))) {
580 if (VOP_ISLOCKED(vp, ap->a_td)) {
582 VOP_UNLOCK(vp, 0, ap->a_td);
584 error = dev_dclose(dev, ap->a_fflag, S_IFCHR, ap->a_td);
586 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
592 * Track the actual opens and closes on the vnode. The last close
593 * disassociates the rdev. If the rdev is already disassociated
594 * the vnode might have been revoked and no further opencount
598 KKASSERT(vp->v_opencount > 0);
600 printf("spec_close: %s %d\n",
601 dev->si_name, vp->v_opencount - 1);
603 if (--vp->v_opencount == 0)
611 * Print out the contents of a special device vnode.
613 * spec_print(struct vnode *a_vp)
616 spec_print(struct vop_print_args *ap)
618 printf("tag VT_NON, dev %s\n", devtoname(ap->a_vp->v_rdev));
623 * Special device advisory byte-level locks.
625 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
626 * struct flock *a_fl, int a_flags)
630 spec_advlock(struct vop_advlock_args *ap)
632 return (ap->a_flags & F_FLOCK ? EOPNOTSUPP : EINVAL);
636 spec_getpages_iodone(struct buf *bp)
638 bp->b_flags |= B_DONE;
643 spec_getpages(struct vop_getpages_args *ap)
652 int toff, nextoff, nread;
653 struct vnode *vp = ap->a_vp;
658 pcount = round_page(ap->a_count) / PAGE_SIZE;
661 * Calculate the offset of the transfer and do sanity check.
662 * FreeBSD currently only supports an 8 TB range due to b_blkno
663 * being in DEV_BSIZE ( usually 512 ) byte chunks on call to
666 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
668 #define DADDR_T_BIT (sizeof(daddr_t)*8)
669 #define OFFSET_MAX ((1LL << (DADDR_T_BIT + DEV_BSHIFT)) - 1)
671 if (offset < 0 || offset > OFFSET_MAX) {
672 /* XXX still no %q in kernel. */
673 printf("spec_getpages: preposterous offset 0x%x%08x\n",
674 (u_int)((u_quad_t)offset >> 32),
675 (u_int)(offset & 0xffffffff));
676 return (VM_PAGER_ERROR);
679 blkno = btodb(offset);
682 * Round up physical size for real devices. We cannot round using
683 * v_mount's block size data because v_mount has nothing to do with
684 * the device. i.e. it's usually '/dev'. We need the physical block
685 * size for the device itself.
687 * We can't use v_rdev->si_mountpoint because it only exists when the
688 * block device is mounted. However, we can use v_rdev.
691 if (vn_isdisk(vp, NULL))
692 blksiz = vp->v_rdev->si_bsize_phys;
696 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
699 kva = (vm_offset_t)bp->b_data;
702 * Map the pages to be read into the kva.
704 pmap_qenter(kva, ap->a_m, pcount);
706 /* Build a minimal buffer header. */
707 bp->b_flags = B_READ;
708 bp->b_iodone = spec_getpages_iodone;
710 /* B_PHYS is not set, but it is nice to fill this in. */
712 bp->b_lblkno = blkno;
713 pbgetvp(ap->a_vp, bp);
715 bp->b_bufsize = size;
717 bp->b_runningbufspace = bp->b_bufsize;
718 runningbufspace += bp->b_runningbufspace;
720 mycpu->gd_cnt.v_vnodein++;
721 mycpu->gd_cnt.v_vnodepgsin += pcount;
724 VOP_STRATEGY(bp->b_vp, bp);
728 /* We definitely need to be at splbio here. */
729 while ((bp->b_flags & B_DONE) == 0) {
730 tsleep(bp, 0, "spread", 0);
735 if ((bp->b_flags & B_ERROR) != 0) {
742 nread = size - bp->b_resid;
744 if (nread < ap->a_count) {
745 bzero((caddr_t)kva + nread,
746 ap->a_count - nread);
748 pmap_qremove(kva, pcount);
752 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
753 nextoff = toff + PAGE_SIZE;
756 m->flags &= ~PG_ZERO;
758 if (nextoff <= nread) {
759 m->valid = VM_PAGE_BITS_ALL;
761 } else if (toff < nread) {
763 * Since this is a VM request, we have to supply the
764 * unaligned offset to allow vm_page_set_validclean()
765 * to zero sub-DEV_BSIZE'd portions of the page.
767 vm_page_set_validclean(m, 0, nread - toff);
773 if (i != ap->a_reqpage) {
775 * Just in case someone was asking for this page we
776 * now tell them that it is ok to use.
778 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
780 if (m->flags & PG_WANTED) {
783 vm_page_deactivate(m);
792 } else if (m->valid) {
795 * Since this is a VM request, we need to make the
796 * entire page presentable by zeroing invalid sections.
798 if (m->valid != VM_PAGE_BITS_ALL)
799 vm_page_zero_invalid(m, FALSE);
803 m = ap->a_m[ap->a_reqpage];
805 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
806 devtoname(bp->b_dev), error, bp, bp->b_vp);
808 " size: %d, resid: %ld, a_count: %d, valid: 0x%x\n",
809 size, bp->b_resid, ap->a_count, m->valid);
811 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
812 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
814 * Free the buffer header back to the swap buffer pool.
817 return VM_PAGER_ERROR;
820 * Free the buffer header back to the swap buffer pool.