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>
58 #include <sys/kernel.h>
61 #include <sys/nlookup.h>
63 #include <sys/malloc.h>
64 #include <sys/mount.h>
65 #include <sys/vnode.h>
66 #include <sys/fcntl.h>
68 #include <sys/diskslice.h>
71 #include <sys/module.h>
72 #include <sys/vnioctl.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_pager.h>
78 #include <vm/vm_pageout.h>
79 #include <vm/swap_pager.h>
80 #include <vm/vm_extern.h>
81 #include <vm/vm_zone.h>
82 #include <sys/devfs.h>
84 static d_ioctl_t vnioctl;
85 static d_open_t vnopen;
86 static d_close_t vnclose;
87 static d_psize_t vnsize;
88 static d_strategy_t vnstrategy;
89 static d_clone_t vnclone;
91 MALLOC_DEFINE(M_VN, "vn_softc", "vn driver structures");
92 DEVFS_DEFINE_CLONE_BITMAP(vn);
95 #define VN_PREALLOCATED_UNITS 4
97 #define VN_PREALLOCATED_UNITS NVN
100 #define VN_BSIZE_BEST 8192
104 * D_DISK We want to look like a disk
105 * D_CANFREE We support BUF_CMD_FREEBLKS
106 * D_NOEMERGPGR Too complex for emergency pager
109 static struct dev_ops vn_ops = {
110 { "vn", 0, D_DISK | D_CANFREE | D_NOEMERGPGR },
114 .d_write = physwrite,
116 .d_strategy = vnstrategy,
122 int sc_flags; /* flags */
123 u_int64_t sc_size; /* size of vn, sc_secsize scale */
124 int sc_secsize; /* sector size */
126 struct vnode *sc_vp; /* vnode if not NULL */
127 vm_object_t sc_object; /* backing object if not NULL */
128 struct ucred *sc_cred; /* credentials */
129 int sc_maxactive; /* max # of active requests */
130 struct buf sc_tab; /* transfer queue */
131 u_long sc_options; /* options */
132 cdev_t sc_dev; /* devices that refer to this unit */
133 SLIST_ENTRY(vn_softc) sc_list;
136 static SLIST_HEAD(, vn_softc) vn_list;
139 #define VNF_INITED 0x01
140 #define VNF_READONLY 0x02
141 #define VNF_OPENED 0x10
142 #define VNF_DESTROY 0x20
144 static u_long vn_options;
146 #define IFOPT(vn,opt) if (((vn)->sc_options|vn_options) & (opt))
147 #define TESTOPT(vn,opt) (((vn)->sc_options|vn_options) & (opt))
149 static int vnsetcred (struct vn_softc *vn, struct ucred *cred);
150 static void vnclear (struct vn_softc *vn);
151 static int vnget (cdev_t dev, struct vn_softc *vn , struct vn_user *vnu);
152 static int vn_modevent (module_t, int, void *);
153 static int vniocattach_file (struct vn_softc *, struct vn_ioctl *, cdev_t dev, int flag, struct ucred *cred);
154 static int vniocattach_swap (struct vn_softc *, struct vn_ioctl *, cdev_t dev, int flag, struct ucred *cred);
155 static cdev_t vn_create(int unit, struct devfs_bitmap *bitmap, int clone);
158 vnclone(struct dev_clone_args *ap)
162 unit = devfs_clone_bitmap_get(&DEVFS_CLONE_BITMAP(vn), 0);
163 ap->a_dev = vn_create(unit, &DEVFS_CLONE_BITMAP(vn), 1);
169 vnclose(struct dev_close_args *ap)
171 cdev_t dev = ap->a_head.a_dev;
175 KKASSERT(vn != NULL);
177 vn->sc_flags &= ~VNF_OPENED;
179 /* The disk has been detached and can now be safely destroyed */
180 if (vn->sc_flags & VNF_DESTROY) {
181 KKASSERT(disk_getopencount(&vn->sc_disk) == 0);
182 disk_destroy(&vn->sc_disk);
183 devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(vn), dkunit(dev));
184 SLIST_REMOVE(&vn_list, vn, vn_softc, sc_list);
190 static struct vn_softc *
195 vn = kmalloc(sizeof *vn, M_VN, M_WAITOK | M_ZERO);
200 vninitvn(struct vn_softc *vn, cdev_t dev)
204 KKASSERT(vn != NULL);
205 KKASSERT(dev != NULL);
212 SLIST_INSERT_HEAD(&vn_list, vn, sc_list);
216 vnopen(struct dev_open_args *ap)
218 cdev_t dev = ap->a_head.a_dev;
222 * Locate preexisting device
226 KKASSERT(vn != NULL);
229 * Update si_bsize fields for device. This data will be overridden by
230 * the slice/parition code for vn accesses through partitions, and
231 * used directly if you open the 'whole disk' device.
233 * si_bsize_best must be reinitialized in case VN has been
234 * reconfigured, plus make it at least VN_BSIZE_BEST for efficiency.
236 dev->si_bsize_phys = vn->sc_secsize;
237 dev->si_bsize_best = vn->sc_secsize;
238 if (dev->si_bsize_best < VN_BSIZE_BEST)
239 dev->si_bsize_best = VN_BSIZE_BEST;
241 if ((ap->a_oflags & FWRITE) && (vn->sc_flags & VNF_READONLY))
245 kprintf("vnopen(%s, 0x%x, 0x%x)\n",
246 devtoname(dev), ap->a_oflags, ap->a_devtype);
248 vn->sc_flags |= VNF_OPENED;
255 * Run strategy routine for VN device. We use VOP_READ/VOP_WRITE calls
256 * for vnode-backed vn's, and the swap_pager_strategy() call for
257 * vm_object-backed vn's.
260 vnstrategy(struct dev_strategy_args *ap)
262 cdev_t dev = ap->a_head.a_dev;
263 struct bio *bio = ap->a_bio;
272 KKASSERT(vn != NULL);
277 kprintf("vnstrategy(%p): unit %d\n", bp, unit);
279 if ((vn->sc_flags & VNF_INITED) == 0) {
281 bp->b_flags |= B_ERROR;
286 bp->b_resid = bp->b_bcount;
289 * The vnode device is using disk/slice label support.
291 * The dscheck() function is called for validating the
292 * slices that exist ON the vnode device itself, and
293 * translate the "slice-relative" block number, again.
294 * dscheck() will call biodone() and return NULL if
295 * we are at EOF or beyond the device size.
301 * Use the translated nbio from this point on
303 if (vn->sc_vp && bp->b_cmd == BUF_CMD_FREEBLKS) {
305 * Freeblks is not handled for vnode-backed elements yet.
308 /* operation complete */
309 } else if (vn->sc_vp) {
313 * If an error occurs, we set B_ERROR but we do not set
314 * B_INVAL because (for a write anyway), the buffer is
320 bzero(&auio, sizeof(auio));
322 aiov.iov_base = bp->b_data;
323 aiov.iov_len = bp->b_bcount;
324 auio.uio_iov = &aiov;
326 auio.uio_offset = nbio->bio_offset;
327 auio.uio_segflg = UIO_SYSSPACE;
328 if (bp->b_cmd == BUF_CMD_READ)
329 auio.uio_rw = UIO_READ;
331 auio.uio_rw = UIO_WRITE;
332 auio.uio_resid = bp->b_bcount;
333 auio.uio_td = curthread;
336 * Don't use IO_DIRECT here, it really gets in the way
337 * due to typical blocksize differences between the
338 * fs backing the VN device and whatever is running on
343 vn_lock(vn->sc_vp, LK_SHARED | LK_RETRY);
344 error = VOP_READ(vn->sc_vp, &auio, IO_RECURSE,
348 case (BUF_CMD_WRITE):
349 vn_lock(vn->sc_vp, LK_EXCLUSIVE | LK_RETRY);
350 error = VOP_WRITE(vn->sc_vp, &auio, IO_RECURSE,
354 case (BUF_CMD_FLUSH):
356 vn_lock(vn->sc_vp, LK_EXCLUSIVE | LK_RETRY);
357 error = VOP_FSYNC(vn->sc_vp, MNT_WAIT, 0);
364 vn_unlock(vn->sc_vp);
366 bp->b_resid = auio.uio_resid;
369 bp->b_flags |= B_ERROR;
371 /* operation complete */
372 } else if (vn->sc_object) {
374 * OBJT_SWAP I/O (handles read, write, freebuf)
376 * We have nothing to do if freeing blocks on a reserved
377 * swap area, othrewise execute the op.
379 if (bp->b_cmd == BUF_CMD_FREEBLKS && TESTOPT(vn, VN_RESERVE)) {
381 /* operation complete */
383 swap_pager_strategy(vn->sc_object, nbio);
388 bp->b_resid = bp->b_bcount;
389 bp->b_flags |= B_ERROR | B_INVAL;
390 bp->b_error = EINVAL;
391 /* operation complete */
399 vnioctl(struct dev_ioctl_args *ap)
401 cdev_t dev = ap->a_head.a_dev;
403 struct vn_ioctl *vio;
408 IFOPT(vn,VN_FOLLOW) {
409 kprintf("vnioctl(%s, 0x%lx, %p, 0x%x): unit %d\n",
410 devtoname(dev), ap->a_cmd, ap->a_data, ap->a_fflag,
426 if (dkslice(dev) != WHOLE_DISK_SLICE ||
427 dkpart(dev) != WHOLE_SLICE_PART)
433 error = priv_check_cred(ap->a_cred, PRIV_ROOT, 0);
437 vio = (struct vn_ioctl *)ap->a_data;
438 f = (u_long*)ap->a_data;
442 if (vn->sc_flags & VNF_INITED)
445 if (vn->sc_flags & VNF_DESTROY)
448 if (vio->vn_file == NULL)
449 error = vniocattach_swap(vn, vio, dev, ap->a_fflag, ap->a_cred);
451 error = vniocattach_file(vn, vio, dev, ap->a_fflag, ap->a_cred);
455 if ((vn->sc_flags & VNF_INITED) == 0)
458 * XXX handle i/o in progress. Return EBUSY, or wait, or
460 * XXX handle multiple opens of the device. Return EBUSY,
461 * or revoke the fd's.
462 * How are these problems handled for removable and failing
463 * hardware devices? (Hint: They are not)
465 if ((disk_getopencount(&vn->sc_disk)) > 1)
470 kprintf("vnioctl: CLRed\n");
472 if (dkunit(dev) >= VN_PREALLOCATED_UNITS) {
473 vn->sc_flags |= VNF_DESTROY;
479 error = vnget(dev, vn, (struct vn_user *) ap->a_data);
493 vn->sc_options |= *f;
498 vn->sc_options &= ~(*f);
512 * Attach a file to a VN partition. Return the size in the vn_size
517 vniocattach_file(struct vn_softc *vn, struct vn_ioctl *vio, cdev_t dev,
518 int flag, struct ucred *cred)
521 struct nlookupdata nd;
524 struct disk_info info;
526 flags = FREAD|FWRITE;
527 error = nlookup_init(&nd, vio->vn_file,
528 UIO_USERSPACE, NLC_FOLLOW|NLC_LOCKVP);
531 if ((error = vn_open(&nd, NULL, flags, 0)) != 0) {
532 if (error != EACCES && error != EPERM && error != EROFS)
536 error = nlookup_init(&nd, vio->vn_file, UIO_USERSPACE, NLC_FOLLOW|NLC_LOCKVP);
539 if ((error = vn_open(&nd, NULL, flags, 0)) != 0)
543 if (vp->v_type != VREG ||
544 (error = VOP_GETATTR(vp, &vattr))) {
550 vn->sc_secsize = DEV_BSIZE;
552 nd.nl_open_vp = NULL;
555 * If the size is specified, override the file attributes. Note that
556 * the vn_size argument is in PAGE_SIZE sized blocks.
559 vn->sc_size = vio->vn_size * PAGE_SIZE / vn->sc_secsize;
561 vn->sc_size = vattr.va_size / vn->sc_secsize;
562 error = vnsetcred(vn, cred);
565 vn_close(vp, flags, NULL);
568 vn->sc_flags |= VNF_INITED;
570 vn->sc_flags |= VNF_READONLY;
573 * Set the disk info so that probing is triggered
575 bzero(&info, sizeof(struct disk_info));
576 info.d_media_blksize = vn->sc_secsize;
577 info.d_media_blocks = vn->sc_size;
579 * reserve mbr sector for backwards compatibility
580 * when no slices exist.
582 info.d_dsflags = DSO_COMPATMBR | DSO_RAWPSIZE;
583 info.d_secpertrack = 32;
584 info.d_nheads = 64 / (vn->sc_secsize / DEV_BSIZE);
585 info.d_secpercyl = info.d_secpertrack * info.d_nheads;
586 info.d_ncylinders = vn->sc_size / info.d_secpercyl;
587 disk_setdiskinfo_sync(&vn->sc_disk, &info);
589 error = dev_dopen(dev, flag, S_IFCHR, cred, NULL);
594 kprintf("vnioctl: SET vp %p size %llx blks\n",
595 vn->sc_vp, (long long)vn->sc_size);
604 * Attach swap backing store to a VN partition of the size specified
609 vniocattach_swap(struct vn_softc *vn, struct vn_ioctl *vio, cdev_t dev,
610 int flag, struct ucred *cred)
613 struct disk_info info;
616 * Range check. Disallow negative sizes or any size less then the
617 * size of a page. Then round to a page.
620 if (vio->vn_size <= 0)
624 * Allocate an OBJT_SWAP object.
626 * sc_secsize is PAGE_SIZE'd
628 * vio->vn_size is in PAGE_SIZE'd chunks.
629 * sc_size must be in PAGE_SIZE'd chunks.
630 * Note the truncation.
633 vn->sc_secsize = PAGE_SIZE;
634 vn->sc_size = vio->vn_size;
635 vn->sc_object = swap_pager_alloc(NULL,
636 vn->sc_secsize * (off_t)vio->vn_size,
638 vm_object_set_flag(vn->sc_object, OBJ_NOPAGEIN);
639 IFOPT(vn, VN_RESERVE) {
640 if (swap_pager_reserve(vn->sc_object, 0, vn->sc_size) < 0) {
641 vm_pager_deallocate(vn->sc_object);
642 vn->sc_object = NULL;
646 vn->sc_flags |= VNF_INITED;
648 error = vnsetcred(vn, cred);
651 * Set the disk info so that probing is triggered
653 bzero(&info, sizeof(struct disk_info));
654 info.d_media_blksize = vn->sc_secsize;
655 info.d_media_blocks = vn->sc_size;
657 * reserve mbr sector for backwards compatibility
658 * when no slices exist.
660 info.d_dsflags = DSO_COMPATMBR | DSO_RAWPSIZE;
661 info.d_secpertrack = 32;
662 info.d_nheads = 64 / (vn->sc_secsize / DEV_BSIZE);
663 info.d_secpercyl = info.d_secpertrack * info.d_nheads;
664 info.d_ncylinders = vn->sc_size / info.d_secpercyl;
665 disk_setdiskinfo_sync(&vn->sc_disk, &info);
667 error = dev_dopen(dev, flag, S_IFCHR, cred, NULL);
670 IFOPT(vn, VN_FOLLOW) {
671 kprintf("vnioctl: SET vp %p size %llx\n",
672 vn->sc_vp, (long long)vn->sc_size);
681 * Duplicate the current processes' credentials. Since we are called only
682 * as the result of a SET ioctl and only root can do that, any future access
683 * to this "disk" is essentially as root. Note that credentials may change
684 * if some other uid can write directly to the mapped file (NFS).
687 vnsetcred(struct vn_softc *vn, struct ucred *cred)
693 * Set credits in our softc
698 vn->sc_cred = crdup(cred);
701 * Horrible kludge to establish credentials for NFS XXX.
708 tmpbuf = kmalloc(vn->sc_secsize, M_TEMP, M_WAITOK);
709 bzero(&auio, sizeof(auio));
711 aiov.iov_base = tmpbuf;
712 aiov.iov_len = vn->sc_secsize;
713 auio.uio_iov = &aiov;
716 auio.uio_rw = UIO_READ;
717 auio.uio_segflg = UIO_SYSSPACE;
718 auio.uio_resid = aiov.iov_len;
719 vn_lock(vn->sc_vp, LK_EXCLUSIVE | LK_RETRY);
720 error = VOP_READ(vn->sc_vp, &auio, 0, vn->sc_cred);
721 vn_unlock(vn->sc_vp);
722 kfree(tmpbuf, M_TEMP);
728 vnclear(struct vn_softc *vn)
731 kprintf("vnclear(%p): vp=%p\n", vn, vn->sc_vp);
732 vn->sc_flags &= ~VNF_INITED;
733 if (vn->sc_vp != NULL) {
735 (vn->sc_flags & VNF_READONLY) ? FREAD : (FREAD|FWRITE),
739 vn->sc_flags &= ~VNF_READONLY;
744 if (vn->sc_object != NULL) {
745 vm_pager_deallocate(vn->sc_object);
746 vn->sc_object = NULL;
749 disk_unprobe(&vn->sc_disk);
757 * populate a struct vn_user for the VNIOCGET ioctl.
758 * interface conventions defined in sys/sys/vnioctl.h.
762 vnget(cdev_t dev, struct vn_softc *vn, struct vn_user *vnu)
764 int error, found = 0;
765 char *freepath, *fullpath;
768 if (vnu->vnu_unit == -1) {
769 vnu->vnu_unit = dkunit(dev);
771 else if (vnu->vnu_unit < 0)
774 SLIST_FOREACH(vn, &vn_list, sc_list) {
776 if(vn->sc_unit != vnu->vnu_unit)
781 if (vn->sc_flags & VNF_INITED && vn->sc_vp != NULL) {
783 /* note: u_cred checked in vnioctl above */
784 error = VOP_GETATTR(vn->sc_vp, &vattr);
786 kprintf("vnget: VOP_GETATTR for %p failed\n",
791 error = vn_fullpath(curproc, vn->sc_vp,
792 &fullpath, &freepath, 0);
795 kprintf("vnget: unable to resolve vp %p\n",
800 strlcpy(vnu->vnu_file, fullpath,
801 sizeof(vnu->vnu_file));
802 kfree(freepath, M_TEMP);
803 vnu->vnu_dev = vattr.va_fsid;
804 vnu->vnu_ino = vattr.va_fileid;
807 else if (vn->sc_flags & VNF_INITED && vn->sc_object != NULL){
809 strlcpy(vnu->vnu_file, _VN_USER_SWAP,
810 sizeof(vnu->vnu_file));
811 vnu->vnu_size = vn->sc_size;
812 vnu->vnu_secsize = vn->sc_secsize;
816 bzero(vnu->vnu_file, sizeof(vnu->vnu_file));
831 vnsize(struct dev_psize_args *ap)
833 cdev_t dev = ap->a_head.a_dev;
839 if ((vn->sc_flags & VNF_INITED) == 0)
841 ap->a_result = (int64_t)vn->sc_size;
846 vn_create(int unit, struct devfs_bitmap *bitmap, int clone)
849 struct disk_info info;
855 * For clone devices we need to return the top-level cdev,
856 * not the raw dev we'd normally work with.
858 dev = disk_create_clone(unit, &vn->sc_disk, &vn_ops);
859 ret_dev = vn->sc_disk.d_cdev;
861 ret_dev = dev = disk_create(unit, &vn->sc_disk, &vn_ops);
865 bzero(&info, sizeof(struct disk_info));
866 info.d_media_blksize = 512;
867 info.d_media_blocks = 0;
868 info.d_dsflags = DSO_MBRQUIET | DSO_RAWPSIZE;
869 info.d_secpertrack = 32;
871 info.d_secpercyl = info.d_secpertrack * info.d_nheads;
872 info.d_ncylinders = 0;
873 disk_setdiskinfo_sync(&vn->sc_disk, &info);
876 devfs_clone_bitmap_set(bitmap, unit);
882 vn_modevent(module_t mod, int type, void *data)
885 static cdev_t dev = NULL;
890 dev = make_autoclone_dev(&vn_ops, &DEVFS_CLONE_BITMAP(vn), vnclone, UID_ROOT,
891 GID_OPERATOR, 0640, "vn");
893 for (i = 0; i < VN_PREALLOCATED_UNITS; i++) {
894 vn_create(i, &DEVFS_CLONE_BITMAP(vn), 0);
900 while ((vn = SLIST_FIRST(&vn_list)) != NULL) {
902 * XXX: no idea if we can return EBUSY even in the
903 * shutdown case, so err on the side of caution
904 * and just rip stuff out on shutdown.
906 if (type != MOD_SHUTDOWN) {
907 if (vn->sc_flags & VNF_OPENED)
911 disk_destroy(&vn->sc_disk);
913 SLIST_REMOVE_HEAD(&vn_list, sc_list);
915 if (vn->sc_flags & VNF_INITED)
920 destroy_autoclone_dev(dev, &DEVFS_CLONE_BITMAP(vn));
921 dev_ops_remove_all(&vn_ops);
929 DEV_MODULE(vn, vn_modevent, 0);