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 1.18 2004/06/15 00:30:55 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 static int spec_advlock (struct vop_advlock_args *);
60 static int spec_bmap (struct vop_bmap_args *);
61 static int spec_close (struct vop_close_args *);
62 static int spec_freeblks (struct vop_freeblks_args *);
63 static int spec_fsync (struct vop_fsync_args *);
64 static int spec_getpages (struct vop_getpages_args *);
65 static int spec_inactive (struct vop_inactive_args *);
66 static int spec_ioctl (struct vop_ioctl_args *);
67 static int spec_open (struct vop_open_args *);
68 static int spec_poll (struct vop_poll_args *);
69 static int spec_kqfilter (struct vop_kqfilter_args *);
70 static int spec_print (struct vop_print_args *);
71 static int spec_read (struct vop_read_args *);
72 static int spec_strategy (struct vop_strategy_args *);
73 static int spec_write (struct vop_write_args *);
75 vop_t **spec_vnodeop_p;
76 static struct vnodeopv_entry_desc spec_vnodeop_entries[] = {
77 { &vop_default_desc, (vop_t *) vop_defaultop },
78 { &vop_access_desc, (vop_t *) vop_ebadf },
79 { &vop_advlock_desc, (vop_t *) spec_advlock },
80 { &vop_bmap_desc, (vop_t *) spec_bmap },
81 { &vop_close_desc, (vop_t *) spec_close },
82 { &vop_create_desc, (vop_t *) vop_panic },
83 { &vop_freeblks_desc, (vop_t *) spec_freeblks },
84 { &vop_fsync_desc, (vop_t *) spec_fsync },
85 { &vop_getpages_desc, (vop_t *) spec_getpages },
86 { &vop_inactive_desc, (vop_t *) spec_inactive },
87 { &vop_ioctl_desc, (vop_t *) spec_ioctl },
88 { &vop_lease_desc, (vop_t *) vop_null },
89 { &vop_link_desc, (vop_t *) vop_panic },
90 { &vop_mkdir_desc, (vop_t *) vop_panic },
91 { &vop_mknod_desc, (vop_t *) vop_panic },
92 { &vop_open_desc, (vop_t *) spec_open },
93 { &vop_pathconf_desc, (vop_t *) vop_stdpathconf },
94 { &vop_poll_desc, (vop_t *) spec_poll },
95 { &vop_kqfilter_desc, (vop_t *) spec_kqfilter },
96 { &vop_print_desc, (vop_t *) spec_print },
97 { &vop_read_desc, (vop_t *) spec_read },
98 { &vop_readdir_desc, (vop_t *) vop_panic },
99 { &vop_readlink_desc, (vop_t *) vop_panic },
100 { &vop_reallocblks_desc, (vop_t *) vop_panic },
101 { &vop_reclaim_desc, (vop_t *) vop_null },
102 { &vop_remove_desc, (vop_t *) vop_panic },
103 { &vop_rename_desc, (vop_t *) vop_panic },
104 { &vop_rmdir_desc, (vop_t *) vop_panic },
105 { &vop_setattr_desc, (vop_t *) vop_ebadf },
106 { &vop_strategy_desc, (vop_t *) spec_strategy },
107 { &vop_symlink_desc, (vop_t *) vop_panic },
108 { &vop_write_desc, (vop_t *) spec_write },
111 static struct vnodeopv_desc spec_vnodeop_opv_desc =
112 { &spec_vnodeop_p, spec_vnodeop_entries };
114 VNODEOP_SET(spec_vnodeop_opv_desc);
116 extern int dev_ref_debug;
119 * spec_vnoperate(struct vnodeop_desc *a_desc, ...)
122 spec_vnoperate(struct vop_generic_args *ap)
124 return (VOCALL(spec_vnodeop_p, ap->a_desc->vdesc_offset, ap));
127 static void spec_getpages_iodone (struct buf *bp);
130 * Open a special file.
132 * spec_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
133 * struct thread *a_td)
137 spec_open(struct vop_open_args *ap)
139 struct vnode *vp = ap->a_vp;
142 int isblk = (vp->v_type == VBLK) ? 1 : 0;
146 * Don't allow open if fs is mounted -nodev.
148 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
152 * Resolve the device. If the vnode is already open v_rdev may
153 * already be resolved. However, if the device changes out from
154 * under us we report it (and, for now, we allow it). Since
155 * v_release_rdev() zero's v_opencount, we have to save and restore
156 * it when replacing the rdev reference.
158 if (vp->v_rdev != NULL) {
159 dev = udev2dev(vp->v_udev, isblk);
160 if (dev != vp->v_rdev) {
161 int oc = vp->v_opencount;
163 "Warning: spec_open: dev %s was lost",
164 vp->v_rdev->si_name);
166 error = v_associate_rdev(vp,
167 udev2dev(vp->v_udev, isblk));
169 printf(", reacquisition failed\n");
171 vp->v_opencount = oc;
172 printf(", reacquisition successful\n");
178 error = v_associate_rdev(vp, udev2dev(vp->v_udev, isblk));
184 * 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.
242 VOP_UNLOCK(vp, NULL, 0, ap->a_td);
243 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_td);
244 vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
249 if (dev_dflags(dev) & D_TTY) {
254 printf("Warning:%s: no t_stop, using nottystop\n", devtoname(dev));
255 tp->t_stop = nottystop;
260 if (vn_isdisk(vp, NULL)) {
261 if (!dev->si_bsize_phys)
262 dev->si_bsize_phys = DEV_BSIZE;
264 if ((dev_dflags(dev) & D_DISK) == 0) {
267 printf("WARNING: driver %s should register devices with make_dev() (dev_t = \"%s\")\n",
272 printf("spec_open: %s %d\n", dev->si_name, vp->v_opencount);
275 KKASSERT(vp->v_opencount > 0);
276 if (--vp->v_opencount == 0)
285 * spec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
286 * struct ucred *a_cred)
290 spec_read(struct vop_read_args *ap)
303 if (dev == NULL) /* device was revoked */
305 if (uio->uio_resid == 0)
308 VOP_UNLOCK(vp, NULL, 0, td);
309 error = dev_dread(dev, uio, ap->a_ioflag);
310 vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
317 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
318 * struct ucred *a_cred)
322 spec_write(struct vop_write_args *ap)
335 if (dev == NULL) /* device was revoked */
338 VOP_UNLOCK(vp, NULL, 0, td);
339 error = dev_dwrite(dev, uio, ap->a_ioflag);
340 vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
345 * Device ioctl operation.
347 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
348 * int a_fflag, struct ucred *a_cred, struct thread *a_td)
352 spec_ioctl(struct vop_ioctl_args *ap)
356 if ((dev = ap->a_vp->v_rdev) == NULL)
357 return (EBADF); /* device was revoked */
359 return (dev_dioctl(dev, ap->a_command, ap->a_data,
360 ap->a_fflag, ap->a_td));
364 * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred,
365 * struct thread *a_td)
369 spec_poll(struct vop_poll_args *ap)
373 if ((dev = ap->a_vp->v_rdev) == NULL)
374 return (EBADF); /* device was revoked */
375 return (dev_dpoll(dev, ap->a_events, ap->a_td));
379 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
383 spec_kqfilter(struct vop_kqfilter_args *ap)
387 if ((dev = ap->a_vp->v_rdev) == NULL)
388 return (EBADF); /* device was revoked */
389 return (dev_dkqfilter(dev, ap->a_kn));
393 * Synch buffers associated with a block device
395 * spec_fsync(struct vnode *a_vp, struct ucred *a_cred,
396 * int a_waitfor, struct thread *a_td)
400 spec_fsync(struct vop_fsync_args *ap)
402 struct vnode *vp = ap->a_vp;
406 int maxretry = 10000; /* large, arbitrarily chosen */
408 if (!vn_isdisk(vp, NULL))
413 * MARK/SCAN initialization to avoid infinite loops
416 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp;
417 bp = TAILQ_NEXT(bp, b_vnbufs)) {
418 bp->b_flags &= ~B_SCANNED;
423 * Flush all dirty buffers associated with a block device.
427 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
428 nbp = TAILQ_NEXT(bp, b_vnbufs);
429 if ((bp->b_flags & B_SCANNED) != 0)
431 bp->b_flags |= B_SCANNED;
432 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
434 if ((bp->b_flags & B_DELWRI) == 0)
435 panic("spec_fsync: not dirty");
436 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
449 * If synchronous the caller expects us to completely resolve all
450 * dirty buffers in the system. Wait for in-progress I/O to
451 * complete (which could include background bitmap writes), then
452 * retry if dirty blocks still exist.
454 if (ap->a_waitfor == MNT_WAIT) {
455 while (vp->v_numoutput) {
456 vp->v_flag |= VBWAIT;
457 (void) tsleep((caddr_t)&vp->v_numoutput, 0, "spfsyn", 0);
459 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
460 if (--maxretry != 0) {
464 vprint("spec_fsync: giving up on dirty", vp);
472 * spec_inactive(struct vnode *a_vp, struct thread *a_td)
475 spec_inactive(struct vop_inactive_args *ap)
477 VOP_UNLOCK(ap->a_vp, NULL, 0, ap->a_td);
482 * Just call the device strategy routine
484 * spec_strategy(struct vnode *a_vp, struct buf *a_bp)
487 spec_strategy(struct vop_strategy_args *ap)
494 if (((bp->b_flags & B_READ) == 0) &&
495 (LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
496 (*bioops.io_start)(bp);
499 * Collect statistics on synchronous and asynchronous read
500 * and write counts for disks that have associated filesystems.
503 KKASSERT(vp->v_rdev != NULL); /* XXX */
504 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
505 if ((bp->b_flags & B_READ) == 0) {
506 if (bp->b_lock.lk_lockholder == LK_KERNTHREAD)
507 mp->mnt_stat.f_asyncwrites++;
509 mp->mnt_stat.f_syncwrites++;
511 if (bp->b_lock.lk_lockholder == LK_KERNTHREAD)
512 mp->mnt_stat.f_asyncreads++;
514 mp->mnt_stat.f_syncreads++;
517 bp->b_dev = vp->v_rdev;
523 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
526 spec_freeblks(struct vop_freeblks_args *ap)
531 * XXX: This assumes that strategy does the deed right away.
532 * XXX: this may not be TRTTD.
534 KKASSERT(ap->a_vp->v_rdev != NULL);
535 if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0)
537 bp = geteblk(ap->a_length);
538 bp->b_flags |= B_FREEBUF;
539 bp->b_dev = ap->a_vp->v_rdev;
540 bp->b_blkno = ap->a_addr;
541 bp->b_offset = dbtob(ap->a_addr);
542 bp->b_bcount = ap->a_length;
548 * Implement degenerate case where the block requested is the block
549 * returned, and assume that the entire device is contiguous in regards
550 * to the contiguous block range (runp and runb).
552 * spec_bmap(struct vnode *a_vp, daddr_t a_bn, struct vnode **a_vpp,
553 * daddr_t *a_bnp, int *a_runp, int *a_runb)
556 spec_bmap(struct vop_bmap_args *ap)
558 struct vnode *vp = ap->a_vp;
562 if (ap->a_vpp != NULL)
564 if (ap->a_bnp != NULL)
565 *ap->a_bnp = ap->a_bn;
566 if (vp->v_mount != NULL)
567 runp = runb = MAXBSIZE / vp->v_mount->mnt_stat.f_iosize;
568 if (ap->a_runp != NULL)
570 if (ap->a_runb != NULL)
576 * Device close routine
578 * spec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
579 * struct thread *a_td)
583 spec_close(struct vop_close_args *ap)
585 struct proc *p = ap->a_td->td_proc;
586 struct vnode *vp = ap->a_vp;
587 dev_t dev = vp->v_rdev;
591 * Hack: a tty device that is a controlling terminal
592 * has a reference from the session structure.
593 * We cannot easily tell that a character device is
594 * a controlling terminal, unless it is the closing
595 * process' controlling terminal. In that case,
596 * if the reference count is 2 (this last descriptor
597 * plus the session), release the reference from the session.
599 * It is possible for v_opencount to be 0 or 1 in this case, 0
600 * because the tty might have been revoked.
604 if (vcount(vp) == 2 && vp->v_opencount <= 1 &&
605 p && (vp->v_flag & VXLOCK) == 0 && vp == p->p_session->s_ttyvp) {
606 p->p_session->s_ttyvp = NULL;
611 * Vnodes can be opened and close multiple times. Do not really
612 * close the device unless (1) it is being closed forcibly,
613 * (2) the device wants to track closes, or (3) this is the last
614 * vnode doing its last close on the device.
616 * XXX the VXLOCK (force close) case can leave vnodes referencing
619 if (dev && ((vp->v_flag & VXLOCK) ||
620 (dev_dflags(dev) & D_TRACKCLOSE) ||
621 (vcount(vp) <= 1 && vp->v_opencount == 1))) {
622 error = dev_dclose(dev, ap->a_fflag, S_IFCHR, ap->a_td);
628 * Track the actual opens and closes on the vnode. The last close
629 * disassociates the rdev. If the rdev is already disassociated
630 * the vnode might have been revoked and no further opencount
634 KKASSERT(vp->v_opencount > 0);
636 printf("spec_close: %s %d\n",
637 dev->si_name, vp->v_opencount - 1);
639 if (--vp->v_opencount == 0)
647 * Print out the contents of a special device vnode.
649 * spec_print(struct vnode *a_vp)
652 spec_print(struct vop_print_args *ap)
654 printf("tag VT_NON, dev %s\n", devtoname(ap->a_vp->v_rdev));
659 * Special device advisory byte-level locks.
661 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
662 * struct flock *a_fl, int a_flags)
666 spec_advlock(struct vop_advlock_args *ap)
668 return (ap->a_flags & F_FLOCK ? EOPNOTSUPP : EINVAL);
672 spec_getpages_iodone(struct buf *bp)
674 bp->b_flags |= B_DONE;
679 spec_getpages(struct vop_getpages_args *ap)
683 int i, pcount, size, s;
688 int toff, nextoff, nread;
689 struct vnode *vp = ap->a_vp;
694 pcount = round_page(ap->a_count) / PAGE_SIZE;
697 * Calculate the offset of the transfer and do sanity check.
698 * FreeBSD currently only supports an 8 TB range due to b_blkno
699 * being in DEV_BSIZE ( usually 512 ) byte chunks on call to
702 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
704 #define DADDR_T_BIT (sizeof(daddr_t)*8)
705 #define OFFSET_MAX ((1LL << (DADDR_T_BIT + DEV_BSHIFT)) - 1)
707 if (offset < 0 || offset > OFFSET_MAX) {
708 /* XXX still no %q in kernel. */
709 printf("spec_getpages: preposterous offset 0x%x%08x\n",
710 (u_int)((u_quad_t)offset >> 32),
711 (u_int)(offset & 0xffffffff));
712 return (VM_PAGER_ERROR);
715 blkno = btodb(offset);
718 * Round up physical size for real devices. We cannot round using
719 * v_mount's block size data because v_mount has nothing to do with
720 * the device. i.e. it's usually '/dev'. We need the physical block
721 * size for the device itself.
723 * We can't use v_rdev->si_mountpoint because it only exists when the
724 * block device is mounted. However, we can use v_rdev.
727 if (vn_isdisk(vp, NULL))
728 blksiz = vp->v_rdev->si_bsize_phys;
732 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
735 kva = (vm_offset_t)bp->b_data;
738 * Map the pages to be read into the kva.
740 pmap_qenter(kva, ap->a_m, pcount);
742 /* Build a minimal buffer header. */
743 bp->b_flags = B_READ | B_CALL;
744 bp->b_iodone = spec_getpages_iodone;
746 /* B_PHYS is not set, but it is nice to fill this in. */
748 bp->b_lblkno = blkno;
749 pbgetvp(ap->a_vp, bp);
751 bp->b_bufsize = size;
753 bp->b_runningbufspace = bp->b_bufsize;
754 runningbufspace += bp->b_runningbufspace;
756 mycpu->gd_cnt.v_vnodein++;
757 mycpu->gd_cnt.v_vnodepgsin += pcount;
760 VOP_STRATEGY(bp->b_vp, bp);
764 /* We definitely need to be at splbio here. */
765 while ((bp->b_flags & B_DONE) == 0) {
766 tsleep(bp, 0, "spread", 0);
771 if ((bp->b_flags & B_ERROR) != 0) {
778 nread = size - bp->b_resid;
780 if (nread < ap->a_count) {
781 bzero((caddr_t)kva + nread,
782 ap->a_count - nread);
784 pmap_qremove(kva, pcount);
788 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
789 nextoff = toff + PAGE_SIZE;
792 m->flags &= ~PG_ZERO;
794 if (nextoff <= nread) {
795 m->valid = VM_PAGE_BITS_ALL;
797 } else if (toff < nread) {
799 * Since this is a VM request, we have to supply the
800 * unaligned offset to allow vm_page_set_validclean()
801 * to zero sub-DEV_BSIZE'd portions of the page.
803 vm_page_set_validclean(m, 0, nread - toff);
809 if (i != ap->a_reqpage) {
811 * Just in case someone was asking for this page we
812 * now tell them that it is ok to use.
814 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
816 if (m->flags & PG_WANTED) {
819 vm_page_deactivate(m);
828 } else if (m->valid) {
831 * Since this is a VM request, we need to make the
832 * entire page presentable by zeroing invalid sections.
834 if (m->valid != VM_PAGE_BITS_ALL)
835 vm_page_zero_invalid(m, FALSE);
839 m = ap->a_m[ap->a_reqpage];
841 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
842 devtoname(bp->b_dev), error, bp, bp->b_vp);
844 " size: %d, resid: %ld, a_count: %d, valid: 0x%x\n",
845 size, bp->b_resid, ap->a_count, m->valid);
847 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
848 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
850 * Free the buffer header back to the swap buffer pool.
853 return VM_PAGER_ERROR;
856 * Free the buffer header back to the swap buffer pool.