2 * Copyright (c) 1988 University of Utah.
3 * Copyright (c) 1990, 1993
4 * The Regents of the University of California. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
7 * the Systems Programming Group of the University of Utah Computer
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: Utah Hdr: vn.c 1.13 94/04/02
36 * from: @(#)vn.c 8.6 (Berkeley) 4/1/94
37 * $FreeBSD: src/sys/dev/vn/vn.c,v 1.105.2.4 2001/11/18 07:11:00 dillon Exp $
43 * Block/character interface to a vnode. Allows one to treat a file
44 * as a disk (e.g. build a filesystem in it, mount it, etc.).
46 * NOTE 1: There is a security issue involved with this driver.
47 * Once mounted all access to the contents of the "mapped" file via
48 * the special file is controlled by the permissions on the special
49 * file, the protection of the mapped file is ignored (effectively,
50 * by using root credentials in all transactions).
52 * NOTE 2: Doesn't interact with leases, should it?
56 #include <sys/param.h>
57 #include <sys/systm.h>
59 #include <sys/kernel.h>
62 #include <sys/nlookup.h>
64 #include <sys/malloc.h>
65 #include <sys/mount.h>
66 #include <sys/vnode.h>
67 #include <sys/fcntl.h>
69 #include <sys/diskslice.h>
72 #include <sys/module.h>
73 #include <sys/vnioctl.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_pager.h>
79 #include <vm/vm_pageout.h>
80 #include <vm/swap_pager.h>
81 #include <vm/vm_extern.h>
82 #include <vm/vm_zone.h>
83 #include <sys/devfs.h>
85 static d_ioctl_t vnioctl;
86 static d_open_t vnopen;
87 static d_close_t vnclose;
88 static d_psize_t vnsize;
89 static d_strategy_t vnstrategy;
90 static d_clone_t vnclone;
92 MALLOC_DEFINE(M_VN, "vn_softc", "vn driver structures");
93 DEVFS_DEFINE_CLONE_BITMAP(vn);
96 #define VN_PREALLOCATED_UNITS 4
98 #define VN_PREALLOCATED_UNITS NVN
101 #define VN_BSIZE_BEST 8192
105 * D_DISK We want to look like a disk
106 * D_CANFREE We support BUF_CMD_FREEBLKS
107 * D_NOEMERGPGR Too complex for emergency pager
110 static struct dev_ops vn_ops = {
111 { "vn", 0, D_DISK | D_CANFREE | D_NOEMERGPGR },
115 .d_write = physwrite,
117 .d_strategy = vnstrategy,
123 int sc_flags; /* flags */
124 u_int64_t sc_size; /* size of vn, sc_secsize scale */
125 int sc_secsize; /* sector size */
127 struct vnode *sc_vp; /* vnode if not NULL */
128 vm_object_t sc_object; /* backing object if not NULL */
129 struct ucred *sc_cred; /* credentials */
130 int sc_maxactive; /* max # of active requests */
131 struct buf sc_tab; /* transfer queue */
132 u_long sc_options; /* options */
133 cdev_t sc_dev; /* devices that refer to this unit */
134 SLIST_ENTRY(vn_softc) sc_list;
137 static SLIST_HEAD(, vn_softc) vn_list;
140 #define VNF_INITED 0x01
141 #define VNF_READONLY 0x02
142 #define VNF_OPENED 0x10
143 #define VNF_DESTROY 0x20
145 static u_long vn_options;
147 #define IFOPT(vn,opt) if (((vn)->sc_options|vn_options) & (opt))
148 #define TESTOPT(vn,opt) (((vn)->sc_options|vn_options) & (opt))
150 static int vnsetcred (struct vn_softc *vn, struct ucred *cred);
151 static void vnclear (struct vn_softc *vn);
152 static int vnget (cdev_t dev, struct vn_softc *vn , struct vn_user *vnu);
153 static int vn_modevent (module_t, int, void *);
154 static int vniocattach_file (struct vn_softc *, struct vn_ioctl *, cdev_t dev, int flag, struct ucred *cred);
155 static int vniocattach_swap (struct vn_softc *, struct vn_ioctl *, cdev_t dev, int flag, struct ucred *cred);
156 static cdev_t vn_create(int unit, int clone);
159 vnclone(struct dev_clone_args *ap)
163 unit = devfs_clone_bitmap_get(&DEVFS_CLONE_BITMAP(vn), 0);
164 ap->a_dev = vn_create(unit, 1);
170 vnclose(struct dev_close_args *ap)
172 cdev_t dev = ap->a_head.a_dev;
176 KKASSERT(vn != NULL);
178 vn->sc_flags &= ~VNF_OPENED;
180 /* The disk has been detached and can now be safely destroyed */
181 if (vn->sc_flags & VNF_DESTROY) {
182 KKASSERT(disk_getopencount(&vn->sc_disk) == 0);
183 disk_destroy(&vn->sc_disk);
184 devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(vn), dkunit(dev));
185 SLIST_REMOVE(&vn_list, vn, vn_softc, sc_list);
191 static struct vn_softc *
196 vn = kmalloc(sizeof *vn, M_VN, M_WAITOK | M_ZERO);
201 vninitvn(struct vn_softc *vn, cdev_t dev)
205 KKASSERT(vn != NULL);
206 KKASSERT(dev != NULL);
213 SLIST_INSERT_HEAD(&vn_list, vn, sc_list);
217 vnopen(struct dev_open_args *ap)
219 cdev_t dev = ap->a_head.a_dev;
223 * Locate preexisting device
227 KKASSERT(vn != NULL);
230 * Update si_bsize fields for device. This data will be overridden by
231 * the slice/parition code for vn accesses through partitions, and
232 * used directly if you open the 'whole disk' device.
234 * si_bsize_best must be reinitialized in case VN has been
235 * reconfigured, plus make it at least VN_BSIZE_BEST for efficiency.
237 dev->si_bsize_phys = vn->sc_secsize;
238 dev->si_bsize_best = vn->sc_secsize;
239 if (dev->si_bsize_best < VN_BSIZE_BEST)
240 dev->si_bsize_best = VN_BSIZE_BEST;
242 if ((ap->a_oflags & FWRITE) && (vn->sc_flags & VNF_READONLY))
246 kprintf("vnopen(%s, 0x%x, 0x%x)\n",
247 devtoname(dev), ap->a_oflags, ap->a_devtype);
249 vn->sc_flags |= VNF_OPENED;
256 * Run strategy routine for VN device. We use VOP_READ/VOP_WRITE calls
257 * for vnode-backed vn's, and the swap_pager_strategy() call for
258 * vm_object-backed vn's.
261 vnstrategy(struct dev_strategy_args *ap)
263 cdev_t dev = ap->a_head.a_dev;
264 struct bio *bio = ap->a_bio;
273 KKASSERT(vn != NULL);
278 kprintf("vnstrategy(%p): unit %d\n", bp, unit);
280 if ((vn->sc_flags & VNF_INITED) == 0) {
282 bp->b_flags |= B_ERROR;
287 bp->b_resid = bp->b_bcount;
290 * The vnode device is using disk/slice label support.
292 * The dscheck() function is called for validating the
293 * slices that exist ON the vnode device itself, and
294 * translate the "slice-relative" block number, again.
295 * dscheck() will call biodone() and return NULL if
296 * we are at EOF or beyond the device size.
302 * Use the translated nbio from this point on
304 if (vn->sc_vp && bp->b_cmd == BUF_CMD_FREEBLKS) {
306 * Freeblks is not handled for vnode-backed elements yet.
309 /* operation complete */
310 } else if (vn->sc_vp) {
314 * If an error occurs, we set B_ERROR but we do not set
315 * B_INVAL because (for a write anyway), the buffer is
321 bzero(&auio, sizeof(auio));
323 aiov.iov_base = bp->b_data;
324 aiov.iov_len = bp->b_bcount;
325 auio.uio_iov = &aiov;
327 auio.uio_offset = nbio->bio_offset;
328 auio.uio_segflg = UIO_SYSSPACE;
329 if (bp->b_cmd == BUF_CMD_READ)
330 auio.uio_rw = UIO_READ;
332 auio.uio_rw = UIO_WRITE;
333 auio.uio_resid = bp->b_bcount;
334 auio.uio_td = curthread;
337 * Don't use IO_DIRECT here, it really gets in the way
338 * due to typical blocksize differences between the
339 * fs backing the VN device and whatever is running on
344 vn_lock(vn->sc_vp, LK_SHARED | LK_RETRY);
345 error = VOP_READ(vn->sc_vp, &auio, IO_RECURSE,
349 case (BUF_CMD_WRITE):
350 vn_lock(vn->sc_vp, LK_EXCLUSIVE | LK_RETRY);
351 error = VOP_WRITE(vn->sc_vp, &auio, IO_RECURSE,
355 case (BUF_CMD_FLUSH):
357 vn_lock(vn->sc_vp, LK_EXCLUSIVE | LK_RETRY);
358 error = VOP_FSYNC(vn->sc_vp, MNT_WAIT, 0);
365 vn_unlock(vn->sc_vp);
367 bp->b_resid = auio.uio_resid;
370 bp->b_flags |= B_ERROR;
372 /* operation complete */
373 } else if (vn->sc_object) {
375 * OBJT_SWAP I/O (handles read, write, freebuf)
377 * We have nothing to do if freeing blocks on a reserved
378 * swap area, othrewise execute the op.
380 if (bp->b_cmd == BUF_CMD_FREEBLKS && TESTOPT(vn, VN_RESERVE)) {
382 /* operation complete */
384 swap_pager_strategy(vn->sc_object, nbio);
389 bp->b_resid = bp->b_bcount;
390 bp->b_flags |= B_ERROR | B_INVAL;
391 bp->b_error = EINVAL;
392 /* operation complete */
400 vnioctl(struct dev_ioctl_args *ap)
402 cdev_t dev = ap->a_head.a_dev;
404 struct vn_ioctl *vio;
409 IFOPT(vn,VN_FOLLOW) {
410 kprintf("vnioctl(%s, 0x%lx, %p, 0x%x): unit %d\n",
411 devtoname(dev), ap->a_cmd, ap->a_data, ap->a_fflag,
427 if (dkslice(dev) != WHOLE_DISK_SLICE ||
428 dkpart(dev) != WHOLE_SLICE_PART)
434 error = priv_check_cred(ap->a_cred, PRIV_ROOT, 0);
438 vio = (struct vn_ioctl *)ap->a_data;
439 f = (u_long*)ap->a_data;
443 if (vn->sc_flags & VNF_INITED)
446 if (vn->sc_flags & VNF_DESTROY)
449 if (vio->vn_file == NULL)
450 error = vniocattach_swap(vn, vio, dev, ap->a_fflag, ap->a_cred);
452 error = vniocattach_file(vn, vio, dev, ap->a_fflag, ap->a_cred);
456 if ((vn->sc_flags & VNF_INITED) == 0)
459 * XXX handle i/o in progress. Return EBUSY, or wait, or
461 * XXX handle multiple opens of the device. Return EBUSY,
462 * or revoke the fd's.
463 * How are these problems handled for removable and failing
464 * hardware devices? (Hint: They are not)
466 if ((disk_getopencount(&vn->sc_disk)) > 1)
471 kprintf("vnioctl: CLRed\n");
473 if (dkunit(dev) >= VN_PREALLOCATED_UNITS) {
474 vn->sc_flags |= VNF_DESTROY;
480 error = vnget(dev, vn, (struct vn_user *) ap->a_data);
494 vn->sc_options |= *f;
499 vn->sc_options &= ~(*f);
513 * Attach a file to a VN partition. Return the size in the vn_size
518 vniocattach_file(struct vn_softc *vn, struct vn_ioctl *vio, cdev_t dev,
519 int flag, struct ucred *cred)
522 struct nlookupdata nd;
525 struct disk_info info;
527 flags = FREAD|FWRITE;
528 error = nlookup_init(&nd, vio->vn_file,
529 UIO_USERSPACE, NLC_FOLLOW|NLC_LOCKVP);
532 if ((error = vn_open(&nd, NULL, flags, 0)) != 0) {
533 if (error != EACCES && error != EPERM && error != EROFS)
537 error = nlookup_init(&nd, vio->vn_file, UIO_USERSPACE, NLC_FOLLOW|NLC_LOCKVP);
540 if ((error = vn_open(&nd, NULL, flags, 0)) != 0)
544 if (vp->v_type != VREG ||
545 (error = VOP_GETATTR(vp, &vattr))) {
551 vn->sc_secsize = DEV_BSIZE;
553 nd.nl_open_vp = NULL;
556 * If the size is specified, override the file attributes. Note that
557 * the vn_size argument is in PAGE_SIZE sized blocks.
560 vn->sc_size = vio->vn_size * PAGE_SIZE / vn->sc_secsize;
562 vn->sc_size = vattr.va_size / vn->sc_secsize;
563 error = vnsetcred(vn, cred);
566 vn_close(vp, flags, NULL);
569 vn->sc_flags |= VNF_INITED;
571 vn->sc_flags |= VNF_READONLY;
574 * Set the disk info so that probing is triggered
576 bzero(&info, sizeof(struct disk_info));
577 info.d_media_blksize = vn->sc_secsize;
578 info.d_media_blocks = vn->sc_size;
580 * reserve mbr sector for backwards compatibility
581 * when no slices exist.
583 info.d_dsflags = DSO_COMPATMBR | DSO_RAWPSIZE;
584 info.d_secpertrack = 32;
585 info.d_nheads = 64 / (vn->sc_secsize / DEV_BSIZE);
586 info.d_secpercyl = info.d_secpertrack * info.d_nheads;
587 info.d_ncylinders = vn->sc_size / info.d_secpercyl;
588 disk_setdiskinfo_sync(&vn->sc_disk, &info);
590 error = dev_dopen(dev, flag, S_IFCHR, cred, NULL, NULL);
595 kprintf("vnioctl: SET vp %p size %llx blks\n",
596 vn->sc_vp, (long long)vn->sc_size);
605 * Attach swap backing store to a VN partition of the size specified
610 vniocattach_swap(struct vn_softc *vn, struct vn_ioctl *vio, cdev_t dev,
611 int flag, struct ucred *cred)
614 struct disk_info info;
617 * Range check. Disallow negative sizes or any size less then the
618 * size of a page. Then round to a page.
621 if (vio->vn_size <= 0)
625 * Allocate an OBJT_SWAP object.
627 * sc_secsize is PAGE_SIZE'd
629 * vio->vn_size is in PAGE_SIZE'd chunks.
630 * sc_size must be in PAGE_SIZE'd chunks.
631 * Note the truncation.
634 vn->sc_secsize = PAGE_SIZE;
635 vn->sc_size = vio->vn_size;
636 vn->sc_object = swap_pager_alloc(NULL,
637 vn->sc_secsize * (off_t)vio->vn_size,
639 vm_object_set_flag(vn->sc_object, OBJ_NOPAGEIN);
640 IFOPT(vn, VN_RESERVE) {
641 if (swap_pager_reserve(vn->sc_object, 0, vn->sc_size) < 0) {
642 vm_pager_deallocate(vn->sc_object);
643 vn->sc_object = NULL;
647 vn->sc_flags |= VNF_INITED;
649 error = vnsetcred(vn, cred);
652 * Set the disk info so that probing is triggered
654 bzero(&info, sizeof(struct disk_info));
655 info.d_media_blksize = vn->sc_secsize;
656 info.d_media_blocks = vn->sc_size;
658 * reserve mbr sector for backwards compatibility
659 * when no slices exist.
661 info.d_dsflags = DSO_COMPATMBR | DSO_RAWPSIZE;
662 info.d_secpertrack = 32;
663 info.d_nheads = 64 / (vn->sc_secsize / DEV_BSIZE);
664 info.d_secpercyl = info.d_secpertrack * info.d_nheads;
665 info.d_ncylinders = vn->sc_size / info.d_secpercyl;
666 disk_setdiskinfo_sync(&vn->sc_disk, &info);
668 error = dev_dopen(dev, flag, S_IFCHR, cred, NULL, NULL);
671 IFOPT(vn, VN_FOLLOW) {
672 kprintf("vnioctl: SET vp %p size %llx\n",
673 vn->sc_vp, (long long)vn->sc_size);
682 * Duplicate the current processes' credentials. Since we are called only
683 * as the result of a SET ioctl and only root can do that, any future access
684 * to this "disk" is essentially as root. Note that credentials may change
685 * if some other uid can write directly to the mapped file (NFS).
688 vnsetcred(struct vn_softc *vn, struct ucred *cred)
694 * Set credits in our softc
699 vn->sc_cred = crdup(cred);
702 * Horrible kludge to establish credentials for NFS XXX.
709 tmpbuf = kmalloc(vn->sc_secsize, M_TEMP, M_WAITOK);
710 bzero(&auio, sizeof(auio));
712 aiov.iov_base = tmpbuf;
713 aiov.iov_len = vn->sc_secsize;
714 auio.uio_iov = &aiov;
717 auio.uio_rw = UIO_READ;
718 auio.uio_segflg = UIO_SYSSPACE;
719 auio.uio_resid = aiov.iov_len;
720 vn_lock(vn->sc_vp, LK_EXCLUSIVE | LK_RETRY);
721 error = VOP_READ(vn->sc_vp, &auio, 0, vn->sc_cred);
722 vn_unlock(vn->sc_vp);
723 kfree(tmpbuf, M_TEMP);
729 vnclear(struct vn_softc *vn)
732 kprintf("vnclear(%p): vp=%p\n", vn, vn->sc_vp);
733 vn->sc_flags &= ~VNF_INITED;
734 if (vn->sc_vp != NULL) {
736 (vn->sc_flags & VNF_READONLY) ? FREAD : (FREAD|FWRITE),
740 vn->sc_flags &= ~VNF_READONLY;
745 if (vn->sc_object != NULL) {
746 vm_pager_deallocate(vn->sc_object);
747 vn->sc_object = NULL;
750 disk_unprobe(&vn->sc_disk);
758 * populate a struct vn_user for the VNIOCGET ioctl.
759 * interface conventions defined in sys/sys/vnioctl.h.
763 vnget(cdev_t dev, struct vn_softc *vn, struct vn_user *vnu)
765 int error, found = 0;
766 char *freepath, *fullpath;
769 if (vnu->vnu_unit == -1) {
770 vnu->vnu_unit = dkunit(dev);
772 else if (vnu->vnu_unit < 0)
775 SLIST_FOREACH(vn, &vn_list, sc_list) {
777 if(vn->sc_unit != vnu->vnu_unit)
782 if (vn->sc_flags & VNF_INITED && vn->sc_vp != NULL) {
784 /* note: u_cred checked in vnioctl above */
785 error = VOP_GETATTR(vn->sc_vp, &vattr);
787 kprintf("vnget: VOP_GETATTR for %p failed\n",
792 error = vn_fullpath(curproc, vn->sc_vp,
793 &fullpath, &freepath, 0);
796 kprintf("vnget: unable to resolve vp %p\n",
801 strlcpy(vnu->vnu_file, fullpath,
802 sizeof(vnu->vnu_file));
803 kfree(freepath, M_TEMP);
804 vnu->vnu_dev = vattr.va_fsid;
805 vnu->vnu_ino = vattr.va_fileid;
808 else if (vn->sc_flags & VNF_INITED && vn->sc_object != NULL){
810 strlcpy(vnu->vnu_file, _VN_USER_SWAP,
811 sizeof(vnu->vnu_file));
812 vnu->vnu_size = vn->sc_size;
813 vnu->vnu_secsize = vn->sc_secsize;
817 bzero(vnu->vnu_file, sizeof(vnu->vnu_file));
832 vnsize(struct dev_psize_args *ap)
834 cdev_t dev = ap->a_head.a_dev;
840 if ((vn->sc_flags & VNF_INITED) == 0)
842 ap->a_result = (int64_t)vn->sc_size;
847 * Returns NULL only if the specified unit cannot be allocated.
850 vn_create(int unit, int clone)
853 struct disk_info info;
859 * For clone devices we need to return the top-level cdev,
860 * not the raw dev we'd normally work with.
862 dev = disk_create_clone(unit, &vn->sc_disk, &vn_ops);
863 ret_dev = vn->sc_disk.d_cdev;
865 ret_dev = dev = disk_create(unit, &vn->sc_disk, &vn_ops);
869 bzero(&info, sizeof(struct disk_info));
870 info.d_media_blksize = 512;
871 info.d_media_blocks = 0;
872 info.d_dsflags = DSO_MBRQUIET | DSO_RAWPSIZE;
873 info.d_secpertrack = 32;
875 info.d_secpercyl = info.d_secpertrack * info.d_nheads;
876 info.d_ncylinders = 0;
877 disk_setdiskinfo_sync(&vn->sc_disk, &info);
883 vn_modevent(module_t mod, int type, void *data)
886 static cdev_t dev = NULL;
891 dev = make_autoclone_dev(&vn_ops, &DEVFS_CLONE_BITMAP(vn),
892 vnclone, UID_ROOT, GID_OPERATOR,
894 for (i = 0; i < VN_PREALLOCATED_UNITS; i++) {
895 devfs_clone_bitmap_set(&DEVFS_CLONE_BITMAP(vn), i);
902 while ((vn = SLIST_FIRST(&vn_list)) != NULL) {
904 * XXX: no idea if we can return EBUSY even in the
905 * shutdown case, so err on the side of caution
906 * and just rip stuff out on shutdown.
908 if (type != MOD_SHUTDOWN) {
909 if (vn->sc_flags & VNF_OPENED)
913 disk_destroy(&vn->sc_disk);
915 SLIST_REMOVE_HEAD(&vn_list, sc_list);
917 if (vn->sc_flags & VNF_INITED)
922 destroy_autoclone_dev(dev, &DEVFS_CLONE_BITMAP(vn));
923 dev_ops_remove_all(&vn_ops);
931 DEV_MODULE(vn, vn_modevent, 0);