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 1.39 2006/04/30 17:22:18 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>
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 struct vnodeopv_entry_desc spec_vnodeop_entries[] = {
81 { &vop_default_desc, vop_defaultop },
82 { &vop_access_desc, vop_ebadf },
83 { &vop_advlock_desc, (vnodeopv_entry_t) spec_advlock },
84 { &vop_bmap_desc, (vnodeopv_entry_t) spec_bmap },
85 { &vop_close_desc, (vnodeopv_entry_t) spec_close },
86 { &vop_old_create_desc, vop_panic },
87 { &vop_freeblks_desc, (vnodeopv_entry_t) spec_freeblks },
88 { &vop_fsync_desc, (vnodeopv_entry_t) spec_fsync },
89 { &vop_getpages_desc, (vnodeopv_entry_t) spec_getpages },
90 { &vop_inactive_desc, (vnodeopv_entry_t) spec_inactive },
91 { &vop_ioctl_desc, (vnodeopv_entry_t) spec_ioctl },
92 { &vop_old_link_desc, vop_panic },
93 { &vop_old_mkdir_desc, vop_panic },
94 { &vop_old_mknod_desc, vop_panic },
95 { &vop_open_desc, (vnodeopv_entry_t) spec_open },
96 { &vop_pathconf_desc, (vnodeopv_entry_t) vop_stdpathconf },
97 { &vop_poll_desc, (vnodeopv_entry_t) spec_poll },
98 { &vop_kqfilter_desc, (vnodeopv_entry_t) spec_kqfilter },
99 { &vop_print_desc, (vnodeopv_entry_t) spec_print },
100 { &vop_read_desc, (vnodeopv_entry_t) spec_read },
101 { &vop_readdir_desc, vop_panic },
102 { &vop_readlink_desc, vop_panic },
103 { &vop_reallocblks_desc, vop_panic },
104 { &vop_reclaim_desc, vop_null },
105 { &vop_old_remove_desc, vop_panic },
106 { &vop_old_rename_desc, vop_panic },
107 { &vop_old_rmdir_desc, vop_panic },
108 { &vop_setattr_desc, vop_ebadf },
109 { &vop_strategy_desc, (vnodeopv_entry_t) spec_strategy },
110 { &vop_old_symlink_desc, vop_panic },
111 { &vop_write_desc, (vnodeopv_entry_t) spec_write },
114 static struct vnodeopv_desc spec_vnodeop_opv_desc =
115 { &spec_vnode_vops, spec_vnodeop_entries, 0 };
116 VNODEOP_SET(spec_vnodeop_opv_desc);
118 extern int dev_ref_debug;
121 * spec_vnoperate(struct vnodeop_desc *a_desc, ...)
124 spec_vnoperate(struct vop_generic_args *ap)
126 return (VOCALL(spec_vnode_vops, ap));
129 static void spec_getpages_iodone (struct bio *bio);
132 * Open a special file.
134 * spec_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
135 * struct file *a_fp, struct thread *a_td)
139 spec_open(struct vop_open_args *ap)
141 struct vnode *vp = ap->a_vp;
144 int isblk = (vp->v_type == VBLK) ? 1 : 0;
148 * Don't allow open if fs is mounted -nodev.
150 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
154 * Resolve the device. If the vnode is already open v_rdev may
155 * already be resolved. However, if the device changes out from
156 * under us we report it (and, for now, we allow it). Since
157 * v_release_rdev() zero's v_opencount, we have to save and restore
158 * it when replacing the rdev reference.
160 if (vp->v_rdev != NULL) {
161 dev = udev2dev(vp->v_udev, isblk);
162 if (dev != vp->v_rdev) {
163 int oc = vp->v_opencount;
165 "Warning: spec_open: dev %s was lost",
166 vp->v_rdev->si_name);
168 error = v_associate_rdev(vp,
169 udev2dev(vp->v_udev, isblk));
171 printf(", reacquisition failed\n");
173 vp->v_opencount = oc;
174 printf(", reacquisition successful\n");
180 error = v_associate_rdev(vp, udev2dev(vp->v_udev, isblk));
186 * 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;
262 * If this is 'disk' or disk-like device, associate a VM object
265 if (vn_isdisk(vp, NULL)) {
266 if (!dev->si_bsize_phys)
267 dev->si_bsize_phys = DEV_BSIZE;
268 vinitvmio(vp, IDX_TO_OFF(INT_MAX));
270 if ((dev_dflags(dev) & D_DISK) == 0) {
273 printf("WARNING: driver %s should register devices with make_dev() (dev_t = \"%s\")\n",
279 * If we were handed a file pointer we may be able to install a
280 * shortcut which issues device read and write operations directly
281 * from the fileops rather then having to go through spec_read()
285 vn_setspecops(ap->a_fp);
288 printf("spec_open: %s %d\n", dev->si_name, vp->v_opencount);
291 if (vp->v_opencount == 0)
302 * spec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
303 * struct ucred *a_cred)
307 spec_read(struct vop_read_args *ap)
320 if (dev == NULL) /* device was revoked */
322 if (uio->uio_resid == 0)
325 VOP_UNLOCK(vp, 0, td);
326 error = dev_dread(dev, uio, ap->a_ioflag);
327 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
334 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
335 * struct ucred *a_cred)
339 spec_write(struct vop_write_args *ap)
352 if (dev == NULL) /* device was revoked */
355 VOP_UNLOCK(vp, 0, td);
356 error = dev_dwrite(dev, uio, ap->a_ioflag);
357 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
362 * Device ioctl operation.
364 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
365 * int a_fflag, struct ucred *a_cred, struct thread *a_td)
369 spec_ioctl(struct vop_ioctl_args *ap)
373 if ((dev = ap->a_vp->v_rdev) == NULL)
374 return (EBADF); /* device was revoked */
376 return (dev_dioctl(dev, ap->a_command, ap->a_data,
377 ap->a_fflag, ap->a_td));
381 * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred,
382 * struct thread *a_td)
386 spec_poll(struct vop_poll_args *ap)
390 if ((dev = ap->a_vp->v_rdev) == NULL)
391 return (EBADF); /* device was revoked */
392 return (dev_dpoll(dev, ap->a_events, ap->a_td));
396 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
400 spec_kqfilter(struct vop_kqfilter_args *ap)
404 if ((dev = ap->a_vp->v_rdev) == NULL)
405 return (EBADF); /* device was revoked */
406 return (dev_dkqfilter(dev, ap->a_kn));
410 * Synch buffers associated with a block device
412 * spec_fsync(struct vnode *a_vp, struct ucred *a_cred,
413 * int a_waitfor, struct thread *a_td)
417 spec_fsync(struct vop_fsync_args *ap)
419 struct vnode *vp = ap->a_vp;
422 if (!vn_isdisk(vp, NULL))
426 * Flush all dirty buffers associated with a block device.
428 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
433 * spec_inactive(struct vnode *a_vp, struct thread *a_td)
436 spec_inactive(struct vop_inactive_args *ap)
442 * Just call the device strategy routine
444 * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
447 spec_strategy(struct vop_strategy_args *ap)
449 struct bio *bio = ap->a_bio;
450 struct buf *bp = bio->bio_buf;
454 if (bp->b_cmd != BUF_CMD_READ &&
455 (LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start) {
456 (*bioops.io_start)(bp);
460 * Collect statistics on synchronous and asynchronous read
461 * and write counts for disks that have associated filesystems.
464 KKASSERT(vp->v_rdev != NULL); /* XXX */
465 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
466 if (bp->b_cmd == BUF_CMD_READ) {
467 if (bp->b_lock.lk_lockholder == LK_KERNTHREAD)
468 mp->mnt_stat.f_asyncreads++;
470 mp->mnt_stat.f_syncreads++;
472 if (bp->b_lock.lk_lockholder == LK_KERNTHREAD)
473 mp->mnt_stat.f_asyncwrites++;
475 mp->mnt_stat.f_syncwrites++;
478 dev_dstrategy_chain(vp->v_rdev, bio);
483 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
486 spec_freeblks(struct vop_freeblks_args *ap)
491 * XXX: This assumes that strategy does the deed right away.
492 * XXX: this may not be TRTTD.
494 KKASSERT(ap->a_vp->v_rdev != NULL);
495 if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0)
497 bp = geteblk(ap->a_length);
498 bp->b_cmd = BUF_CMD_FREEBLKS;
499 bp->b_bio1.bio_offset = ap->a_offset;
500 bp->b_bcount = ap->a_length;
501 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
506 * Implement degenerate case where the block requested is the block
507 * returned, and assume that the entire device is contiguous in regards
508 * to the contiguous block range (runp and runb).
510 * spec_bmap(struct vnode *a_vp, off_t a_loffset, struct vnode **a_vpp,
511 * off_t *a_doffsetp, int *a_runp, int *a_runb)
514 spec_bmap(struct vop_bmap_args *ap)
516 struct vnode *vp = ap->a_vp;
518 if (ap->a_vpp != NULL)
520 if (ap->a_doffsetp != NULL)
521 *ap->a_doffsetp = ap->a_loffset;
522 if (ap->a_runp != NULL)
523 *ap->a_runp = MAXBSIZE;
524 if (ap->a_runb != NULL) {
525 if (ap->a_loffset < MAXBSIZE)
526 *ap->a_runb = (int)ap->a_loffset;
528 *ap->a_runb = MAXBSIZE;
534 * Device close routine
536 * spec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
537 * struct thread *a_td)
539 * NOTE: the vnode may or may not be locked on call.
543 spec_close(struct vop_close_args *ap)
545 struct proc *p = ap->a_td->td_proc;
546 struct vnode *vp = ap->a_vp;
547 dev_t dev = vp->v_rdev;
552 * Hack: a tty device that is a controlling terminal
553 * has a reference from the session structure.
554 * We cannot easily tell that a character device is
555 * a controlling terminal, unless it is the closing
556 * process' controlling terminal. In that case,
557 * if the reference count is 2 (this last descriptor
558 * plus the session), release the reference from the session.
560 * It is possible for v_opencount to be 0 or 1 in this case, 0
561 * because the tty might have been revoked.
565 if (vcount(vp) == 2 && vp->v_opencount <= 1 &&
566 p && vp == p->p_session->s_ttyvp) {
567 p->p_session->s_ttyvp = NULL;
572 * Vnodes can be opened and close multiple times. Do not really
573 * close the device unless (1) it is being closed forcibly,
574 * (2) the device wants to track closes, or (3) this is the last
575 * vnode doing its last close on the device.
577 * XXX the VXLOCK (force close) case can leave vnodes referencing
580 if (dev && ((vp->v_flag & VRECLAIMED) ||
581 (dev_dflags(dev) & D_TRACKCLOSE) ||
582 (vcount(vp) <= 1 && vp->v_opencount == 1))) {
584 if (VOP_ISLOCKED(vp, ap->a_td)) {
586 VOP_UNLOCK(vp, 0, ap->a_td);
588 error = dev_dclose(dev, ap->a_fflag, S_IFCHR, ap->a_td);
590 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
596 * Track the actual opens and closes on the vnode. The last close
597 * disassociates the rdev. If the rdev is already disassociated
598 * the vnode might have been revoked and no further opencount
602 /*KKASSERT(vp->v_opencount > 0);*/
604 printf("spec_close: %s %d\n",
605 dev->si_name, vp->v_opencount - 1);
607 if (vp->v_opencount == 1)
616 * Print out the contents of a special device vnode.
618 * spec_print(struct vnode *a_vp)
621 spec_print(struct vop_print_args *ap)
623 printf("tag VT_NON, dev %s\n", devtoname(ap->a_vp->v_rdev));
628 * Special device advisory byte-level locks.
630 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
631 * struct flock *a_fl, int a_flags)
635 spec_advlock(struct vop_advlock_args *ap)
637 return (ap->a_flags & F_FLOCK ? EOPNOTSUPP : EINVAL);
641 spec_getpages_iodone(struct bio *bio)
643 bio->bio_buf->b_cmd = BUF_CMD_DONE;
644 wakeup(bio->bio_buf);
648 spec_getpages(struct vop_getpages_args *ap)
656 int toff, nextoff, nread;
657 struct vnode *vp = ap->a_vp;
662 pcount = round_page(ap->a_count) / PAGE_SIZE;
665 * Calculate the offset of the transfer and do sanity check.
667 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
670 * Round up physical size for real devices. We cannot round using
671 * v_mount's block size data because v_mount has nothing to do with
672 * the device. i.e. it's usually '/dev'. We need the physical block
673 * size for the device itself.
675 * We can't use v_rdev->si_mountpoint because it only exists when the
676 * block device is mounted. However, we can use v_rdev.
679 if (vn_isdisk(vp, NULL))
680 blksiz = vp->v_rdev->si_bsize_phys;
684 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
687 kva = (vm_offset_t)bp->b_data;
690 * Map the pages to be read into the kva.
692 pmap_qenter(kva, ap->a_m, pcount);
694 /* Build a minimal buffer header. */
695 bp->b_cmd = BUF_CMD_READ;
697 bp->b_bufsize = size;
699 bp->b_runningbufspace = bp->b_bufsize;
700 runningbufspace += bp->b_runningbufspace;
702 bp->b_bio1.bio_offset = offset;
703 bp->b_bio1.bio_done = spec_getpages_iodone;
705 mycpu->gd_cnt.v_vnodein++;
706 mycpu->gd_cnt.v_vnodepgsin += pcount;
709 vn_strategy(ap->a_vp, &bp->b_bio1);
713 /* We definitely need to be at splbio here. */
714 while (bp->b_cmd != BUF_CMD_DONE)
715 tsleep(bp, 0, "spread", 0);
719 if ((bp->b_flags & B_ERROR) != 0) {
726 nread = size - bp->b_resid;
728 if (nread < ap->a_count) {
729 bzero((caddr_t)kva + nread,
730 ap->a_count - nread);
732 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.