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 1.13 2004/03/01 06:33:23 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);
118 struct vop_generic_args /* {
119 struct vnodeop_desc *a_desc;
120 <other random data follows, presumably>
123 return (VOCALL(spec_vnodeop_p, ap->a_desc->vdesc_offset, ap));
126 static void spec_getpages_iodone (struct buf *bp);
129 * Open a special file.
134 struct vop_open_args /* {
137 struct ucred *a_cred;
141 struct vnode *vp = ap->a_vp;
142 dev_t dev = vp->v_rdev;
147 * Don't allow open if fs is mounted -nodev.
149 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
152 if (dev_dport(dev) == NULL)
155 /* Make this field valid before any I/O in ->d_open */
156 if (!dev->si_iosize_max)
157 dev->si_iosize_max = DFLTPHYS;
160 * XXX: Disks get special billing here, but it is mostly wrong.
161 * XXX: diskpartitions can overlap and the real checks should
162 * XXX: take this into account, and consequently they need to
163 * XXX: live in the diskslicing code. Some checks do.
165 if (vn_isdisk(vp, NULL) && ap->a_cred != FSCRED &&
166 (ap->a_mode & FWRITE)) {
168 * Never allow opens for write if the device is mounted R/W
170 if (vp->v_specmountpoint != NULL &&
171 !(vp->v_specmountpoint->mnt_flag & MNT_RDONLY))
175 * When running in secure mode, do not allow opens
176 * for writing if the device is mounted
178 if (securelevel >= 1 && vp->v_specmountpoint != NULL)
182 * When running in very secure mode, do not allow
183 * opens for writing of any devices.
185 if (securelevel >= 2)
189 /* XXX: Special casing of ttys for deadfs. Probably redundant */
190 if (dev_dflags(dev) & D_TTY)
191 vp->v_flag |= VISTTY;
193 VOP_UNLOCK(vp, NULL, 0, ap->a_td);
194 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_td);
195 vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
200 if (dev_dflags(dev) & D_TTY) {
205 printf("Warning:%s: no t_stop, using nottystop\n", devtoname(dev));
206 tp->t_stop = nottystop;
211 if (vn_isdisk(vp, NULL)) {
212 if (!dev->si_bsize_phys)
213 dev->si_bsize_phys = DEV_BSIZE;
215 if ((dev_dflags(dev) & D_DISK) == 0) {
218 printf("WARNING: driver %s should register devices with make_dev() (dev_t = \"%s\")\n",
231 struct vop_read_args /* {
235 struct ucred *a_cred;
249 if (uio->uio_resid == 0)
252 VOP_UNLOCK(vp, NULL, 0, td);
253 error = dev_dread(dev, uio, ap->a_ioflag);
254 vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
264 struct vop_write_args /* {
268 struct ucred *a_cred;
282 VOP_UNLOCK(vp, NULL, 0, td);
283 error = dev_dwrite(dev, uio, ap->a_ioflag);
284 vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
289 * Device ioctl operation.
294 struct vop_ioctl_args /* {
299 struct ucred *a_cred;
305 dev = ap->a_vp->v_rdev;
306 return (dev_dioctl(dev, ap->a_command, ap->a_data,
307 ap->a_fflag, ap->a_td));
313 struct vop_poll_args /* {
316 struct ucred *a_cred;
322 dev = ap->a_vp->v_rdev;
323 return (dev_dpoll(dev, ap->a_events, ap->a_td));
329 struct vop_kqfilter_args /* {
336 dev = ap->a_vp->v_rdev;
337 return (dev_dkqfilter(dev, ap->a_kn));
341 * Synch buffers associated with a block device
346 struct vop_fsync_args /* {
348 struct ucred *a_cred;
353 struct vnode *vp = ap->a_vp;
357 int maxretry = 10000; /* large, arbitrarily chosen */
359 if (!vn_isdisk(vp, NULL))
364 * MARK/SCAN initialization to avoid infinite loops
367 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp;
368 bp = TAILQ_NEXT(bp, b_vnbufs)) {
369 bp->b_flags &= ~B_SCANNED;
374 * Flush all dirty buffers associated with a block device.
378 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
379 nbp = TAILQ_NEXT(bp, b_vnbufs);
380 if ((bp->b_flags & B_SCANNED) != 0)
382 bp->b_flags |= B_SCANNED;
383 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
385 if ((bp->b_flags & B_DELWRI) == 0)
386 panic("spec_fsync: not dirty");
387 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
400 * If synchronous the caller expects us to completely resolve all
401 * dirty buffers in the system. Wait for in-progress I/O to
402 * complete (which could include background bitmap writes), then
403 * retry if dirty blocks still exist.
405 if (ap->a_waitfor == MNT_WAIT) {
406 while (vp->v_numoutput) {
407 vp->v_flag |= VBWAIT;
408 (void) tsleep((caddr_t)&vp->v_numoutput, 0, "spfsyn", 0);
410 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
411 if (--maxretry != 0) {
415 vprint("spec_fsync: giving up on dirty", vp);
424 struct vop_inactive_args /* {
430 VOP_UNLOCK(ap->a_vp, NULL, 0, ap->a_td);
435 * Just call the device strategy routine
439 struct vop_strategy_args /* {
449 if (((bp->b_flags & B_READ) == 0) &&
450 (LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
451 (*bioops.io_start)(bp);
454 * Collect statistics on synchronous and asynchronous read
455 * and write counts for disks that have associated filesystems.
458 if (vn_isdisk(vp, NULL) && (mp = vp->v_specmountpoint) != 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++;
472 KASSERT(devsw(bp->b_dev) != NULL,
473 ("No devsw on dev %s responsible for buffer %p\n",
474 devtoname(bp->b_dev), bp));
475 KASSERT(devsw(bp->b_dev)->d_strategy != NULL,
476 ("No strategy on dev %s responsible for buffer %p\n",
477 devtoname(bp->b_dev), bp));
485 struct vop_freeblks_args /* {
494 * XXX: This assumes that strategy does the deed right away.
495 * XXX: this may not be TRTTD.
497 if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0)
499 bp = geteblk(ap->a_length);
500 bp->b_flags |= B_FREEBUF;
501 bp->b_dev = ap->a_vp->v_rdev;
502 bp->b_blkno = ap->a_addr;
503 bp->b_offset = dbtob(ap->a_addr);
504 bp->b_bcount = ap->a_length;
510 * Implement degenerate case where the block requested is the block
511 * returned, and assume that the entire device is contiguous in regards
512 * to the contiguous block range (runp and runb).
516 struct vop_bmap_args /* {
519 struct vnode **a_vpp;
525 struct vnode *vp = ap->a_vp;
529 if (ap->a_vpp != NULL)
531 if (ap->a_bnp != NULL)
532 *ap->a_bnp = ap->a_bn;
533 if (vp->v_mount != NULL)
534 runp = runb = MAXBSIZE / vp->v_mount->mnt_stat.f_iosize;
535 if (ap->a_runp != NULL)
537 if (ap->a_runb != NULL)
543 * Device close routine
548 struct vop_close_args /* {
551 struct ucred *a_cred;
555 struct proc *p = ap->a_td->td_proc;
556 struct vnode *vp = ap->a_vp;
557 dev_t dev = vp->v_rdev;
560 * Hack: a tty device that is a controlling terminal
561 * has a reference from the session structure.
562 * We cannot easily tell that a character device is
563 * a controlling terminal, unless it is the closing
564 * process' controlling terminal. In that case,
565 * if the reference count is 2 (this last descriptor
566 * plus the session), release the reference from the session.
568 if (vcount(vp) == 2 && p && (vp->v_flag & VXLOCK) == 0 &&
569 vp == p->p_session->s_ttyvp) {
571 p->p_session->s_ttyvp = NULL;
574 * We do not want to really close the device if it
575 * is still in use unless we are trying to close it
576 * forcibly. Since every use (buffer, vnode, swap, cmap)
577 * holds a reference to the vnode, and because we mark
578 * any other vnodes that alias this device, when the
579 * sum of the reference counts on all the aliased
580 * vnodes descends to one, we are on last close.
582 if (vp->v_flag & VXLOCK) {
584 } else if (dev_dflags(dev) & D_TRACKCLOSE) {
585 /* Keep device updated on status */
586 } else if (vcount(vp) > 1) {
589 return (dev_dclose(dev, ap->a_fflag, S_IFCHR, ap->a_td));
593 * Print out the contents of a special device vnode.
597 struct vop_print_args /* {
602 printf("tag VT_NON, dev %s\n", devtoname(ap->a_vp->v_rdev));
607 * Special device advisory byte-level locks.
612 struct vop_advlock_args /* {
621 return (ap->a_flags & F_FLOCK ? EOPNOTSUPP : EINVAL);
625 spec_getpages_iodone(bp)
629 bp->b_flags |= B_DONE;
635 struct vop_getpages_args *ap;
639 int i, pcount, size, s;
644 int toff, nextoff, nread;
645 struct vnode *vp = ap->a_vp;
650 pcount = round_page(ap->a_count) / PAGE_SIZE;
653 * Calculate the offset of the transfer and do sanity check.
654 * FreeBSD currently only supports an 8 TB range due to b_blkno
655 * being in DEV_BSIZE ( usually 512 ) byte chunks on call to
658 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
660 #define DADDR_T_BIT (sizeof(daddr_t)*8)
661 #define OFFSET_MAX ((1LL << (DADDR_T_BIT + DEV_BSHIFT)) - 1)
663 if (offset < 0 || offset > OFFSET_MAX) {
664 /* XXX still no %q in kernel. */
665 printf("spec_getpages: preposterous offset 0x%x%08x\n",
666 (u_int)((u_quad_t)offset >> 32),
667 (u_int)(offset & 0xffffffff));
668 return (VM_PAGER_ERROR);
671 blkno = btodb(offset);
674 * Round up physical size for real devices. We cannot round using
675 * v_mount's block size data because v_mount has nothing to do with
676 * the device. i.e. it's usually '/dev'. We need the physical block
677 * size for the device itself.
679 * We can't use v_specmountpoint because it only exists when the
680 * block device is mounted. However, we can use v_rdev.
683 if (vn_isdisk(vp, NULL))
684 blksiz = vp->v_rdev->si_bsize_phys;
688 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
691 kva = (vm_offset_t)bp->b_data;
694 * Map the pages to be read into the kva.
696 pmap_qenter(kva, ap->a_m, pcount);
698 /* Build a minimal buffer header. */
699 bp->b_flags = B_READ | B_CALL;
700 bp->b_iodone = spec_getpages_iodone;
702 /* B_PHYS is not set, but it is nice to fill this in. */
704 bp->b_lblkno = blkno;
705 pbgetvp(ap->a_vp, bp);
707 bp->b_bufsize = size;
709 bp->b_runningbufspace = bp->b_bufsize;
710 runningbufspace += bp->b_runningbufspace;
712 mycpu->gd_cnt.v_vnodein++;
713 mycpu->gd_cnt.v_vnodepgsin += pcount;
716 VOP_STRATEGY(bp->b_vp, bp);
720 /* We definitely need to be at splbio here. */
721 while ((bp->b_flags & B_DONE) == 0) {
722 tsleep(bp, 0, "spread", 0);
727 if ((bp->b_flags & B_ERROR) != 0) {
734 nread = size - bp->b_resid;
736 if (nread < ap->a_count) {
737 bzero((caddr_t)kva + nread,
738 ap->a_count - nread);
740 pmap_qremove(kva, pcount);
744 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
745 nextoff = toff + PAGE_SIZE;
748 m->flags &= ~PG_ZERO;
750 if (nextoff <= nread) {
751 m->valid = VM_PAGE_BITS_ALL;
753 } else if (toff < nread) {
755 * Since this is a VM request, we have to supply the
756 * unaligned offset to allow vm_page_set_validclean()
757 * to zero sub-DEV_BSIZE'd portions of the page.
759 vm_page_set_validclean(m, 0, nread - toff);
765 if (i != ap->a_reqpage) {
767 * Just in case someone was asking for this page we
768 * now tell them that it is ok to use.
770 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
772 if (m->flags & PG_WANTED) {
775 vm_page_deactivate(m);
784 } else if (m->valid) {
787 * Since this is a VM request, we need to make the
788 * entire page presentable by zeroing invalid sections.
790 if (m->valid != VM_PAGE_BITS_ALL)
791 vm_page_zero_invalid(m, FALSE);
795 m = ap->a_m[ap->a_reqpage];
797 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
798 devtoname(bp->b_dev), error, bp, bp->b_vp);
800 " size: %d, resid: %ld, a_count: %d, valid: 0x%x\n",
801 size, bp->b_resid, ap->a_count, m->valid);
803 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
804 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
806 * Free the buffer header back to the swap buffer pool.
809 return VM_PAGER_ERROR;
812 * Free the buffer header back to the swap buffer pool.