]> git.saurik.com Git - apple/xnu.git/blob - bsd/dev/vn/vn.c
projects / apple / xnu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xnu-1228.3.13.tar.gz
[apple/xnu.git] / bsd / dev / vn / vn.c
1 /*
2 * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28
29 /*
30 * Copyright (c) 1988 University of Utah.
31 * Copyright (c) 1990, 1993
32 * The Regents of the University of California. All rights reserved.
33 *
34 * This code is derived from software contributed to Berkeley by
35 * the Systems Programming Group of the University of Utah Computer
36 * Science Department.
37 *
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
40 * are met:
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
53 *
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64 * SUCH DAMAGE.
65 *
66 * from: Utah Hdr: vn.c 1.13 94/04/02
67 *
68 * from: @(#)vn.c 8.6 (Berkeley) 4/1/94
69 * $FreeBSD: src/sys/dev/vn/vn.c,v 1.105.2.4 2001/11/18 07:11:00 dillon Exp $
70 */
71
72 /*
73 * Vnode disk driver.
74 *
75 * Block/character interface to a vnode. Allows one to treat a file
76 * as a disk (e.g. build a filesystem in it, mount it, etc.).
77 *
78 * NOTE 1: This uses the vnop_blockmap/vnop_strategy interface to the vnode
79 * instead of a simple VOP_RDWR. We do this to avoid distorting the
80 * local buffer cache.
81 *
82 * NOTE 2: There is a security issue involved with this driver.
83 * Once mounted all access to the contents of the "mapped" file via
84 * the special file is controlled by the permissions on the special
85 * file, the protection of the mapped file is ignored (effectively,
86 * by using root credentials in all transactions).
87 *
88 * NOTE 3: Doesn't interact with leases, should it?
89 */
90
91 #include "vndevice.h"
92
93 #if NVNDEVICE > 0
94
95 #include <sys/param.h>
96 #include <sys/systm.h>
97 #include <sys/kernel.h>
98 #include <sys/mount.h>
99 #include <sys/namei.h>
100 #include <sys/proc.h>
101 #include <sys/kauth.h>
102 #include <sys/buf.h>
103 #include <sys/malloc.h>
104 #include <sys/vnode_internal.h>
105 #include <sys/fcntl.h>
106 #include <sys/conf.h>
107 #include <sys/disk.h>
108 #include <sys/stat.h>
109 #include <sys/conf.h>
110 #include <sys/uio_internal.h>
111
112 #include <sys/vnioctl.h>
113
114 #include <sys/vm.h>
115
116 #include <vm/vm_pager.h>
117 #include <mach/memory_object_types.h>
118
119 #include <miscfs/devfs/devfs.h>
120
121
122 #include "shadow.h"
123 static void
124 vndevice_do_init(void) __attribute__((section("__TEXT, initcode")));
125
126 static ioctl_fcn_t vnioctl_chr;
127 static ioctl_fcn_t vnioctl_blk;
128 static open_close_fcn_t vnopen;
129 static open_close_fcn_t vnclose;
130 static psize_fcn_t vnsize;
131 static strategy_fcn_t vnstrategy;
132 static read_write_fcn_t vnread;
133 static read_write_fcn_t vnwrite;
134
135 static int vndevice_bdev_major;
136 static int vndevice_cdev_major;
137
138 /*
139 * cdevsw
140 * D_DISK we want to look like a disk
141 * D_CANFREE We support B_FREEBUF
142 */
143
144 static struct bdevsw vn_bdevsw = {
145 /* open */ vnopen,
146 /* close */ vnclose,
147 /* strategy */ vnstrategy,
148 /* ioctl */ vnioctl_blk,
149 /* dump */ eno_dump,
150 /* psize */ vnsize,
151 /* flags */ D_DISK,
152 };
153
154 static struct cdevsw vn_cdevsw = {
155 /* open */ vnopen,
156 /* close */ vnclose,
157 /* read */ vnread,
158 /* write */ vnwrite,
159 /* ioctl */ vnioctl_chr,
160 /* stop */ eno_stop,
161 /* reset */ eno_reset,
162 /* ttys */ NULL,
163 /* select */ eno_select,
164 /* mmap */ eno_mmap,
165 /* strategy */ eno_strat,
166 /* getc */ eno_getc,
167 /* putc */ eno_putc,
168 /* flags */ D_DISK,
169 };
170
171 struct vn_softc {
172 u_int64_t sc_fsize; /* file size in bytes */
173 u_int64_t sc_size; /* size of vn, sc_secsize scale */
174 int sc_flags; /* flags */
175 u_long sc_secsize; /* sector size */
176 struct vnode *sc_vp; /* vnode if not NULL */
177 uint32_t sc_vid;
178 int sc_open_flags;
179 struct vnode *sc_shadow_vp; /* shadow vnode if not NULL */
180 uint32_t sc_shadow_vid;
181 shadow_map_t * sc_shadow_map; /* shadow map if not NULL */
182 kauth_cred_t sc_cred; /* credentials */
183 u_int32_t sc_options; /* options */
184 void * sc_bdev;
185 void * sc_cdev;
186 } vn_table[NVNDEVICE];
187
188 #define ROOT_IMAGE_UNIT 0
189
190 /* sc_flags */
191 #define VNF_INITED 0x01
192 #define VNF_READONLY 0x02
193
194 static u_int32_t vn_options;
195
196 #define IFOPT(vn,opt) if (((vn)->sc_options|vn_options) & (opt))
197 #define TESTOPT(vn,opt) (((vn)->sc_options|vn_options) & (opt))
198
199 static int setcred(struct vnode * vp, kauth_cred_t cred);
200 static void vnclear (struct vn_softc *vn, vfs_context_t ctx);
201 static void vn_ioctl_to_64(struct vn_ioctl *from, struct user_vn_ioctl *to);
202 void vndevice_init(void);
203 int vndevice_root_image(char * path, char devname[], dev_t * dev_p);
204
205 static int
206 vniocattach_file(struct vn_softc *vn,
207 struct user_vn_ioctl *vniop,
208 dev_t dev,
209 int in_kernel,
210 proc_t p);
211 static int
212 vniocattach_shadow(struct vn_softc * vn,
213 struct user_vn_ioctl *vniop,
214 dev_t dev,
215 int in_kernel,
216 proc_t p);
217 static __inline__ int
218 vnunit(dev_t dev)
219 {
220 return (minor(dev));
221 }
222
223 static int
224 vnclose(__unused dev_t dev, __unused int flags,
225 __unused int devtype, __unused proc_t p)
226 {
227 return (0);
228 }
229
230 static int
231 vnopen(dev_t dev, int flags, __unused int devtype, __unused proc_t p)
232 {
233 struct vn_softc *vn;
234 int unit;
235
236 unit = vnunit(dev);
237 if (vnunit(dev) >= NVNDEVICE) {
238 return (ENXIO);
239 }
240 vn = vn_table + unit;
241 if ((flags & FWRITE) && (vn->sc_flags & VNF_READONLY))
242 return (EACCES);
243
244 return(0);
245 }
246
247 static int
248 file_io(struct vnode * vp, vfs_context_t ctx,
249 enum uio_rw op, char * base, off_t offset, user_ssize_t count,
250 user_ssize_t * resid)
251 {
252 uio_t auio;
253 int error;
254 char uio_buf[UIO_SIZEOF(1)];
255
256 auio = uio_createwithbuffer(1, offset, UIO_SYSSPACE, op,
257 &uio_buf[0], sizeof(uio_buf));
258 uio_addiov(auio, CAST_USER_ADDR_T(base), count);
259 if (op == UIO_READ)
260 error = VNOP_READ(vp, auio, IO_SYNC, ctx);
261 else
262 error = VNOP_WRITE(vp, auio, IO_SYNC, ctx);
263
264 if (resid != NULL) {
265 *resid = uio_resid(auio);
266 }
267 return (error);
268 }
269
270 static __inline__ off_t
271 block_round(off_t o, int blocksize)
272 {
273 return ((o + blocksize - 1) / blocksize);
274 }
275
276 static __inline__ off_t
277 block_truncate(off_t o, int blocksize)
278 {
279 return (o / blocksize);
280 }
281
282 static __inline__ int
283 block_remainder(off_t o, int blocksize)
284 {
285 return (o % blocksize);
286 }
287
288 static int
289 vnread_shadow(struct vn_softc * vn, struct uio *uio, int ioflag,
290 vfs_context_t ctx)
291 {
292 u_long blocksize = vn->sc_secsize;
293 int error = 0;
294 off_t offset;
295 user_ssize_t resid;
296 off_t orig_offset;
297 user_ssize_t orig_resid;
298
299 orig_resid = resid = uio_resid(uio);
300 orig_offset = offset = uio_offset(uio);
301
302 while (resid > 0) {
303 u_long remainder;
304 u_long this_block_number;
305 u_long this_block_count;
306 off_t this_offset;
307 user_ssize_t this_resid;
308 struct vnode * vp;
309
310 /* figure out which blocks to read */
311 remainder = block_remainder(offset, blocksize);
312 if (shadow_map_read(vn->sc_shadow_map,
313 block_truncate(offset, blocksize),
314 block_round(resid + remainder, blocksize),
315 &this_block_number, &this_block_count)) {
316 vp = vn->sc_shadow_vp;
317 }
318 else {
319 vp = vn->sc_vp;
320 }
321
322 /* read the blocks (or parts thereof) */
323 this_offset = (off_t)this_block_number * blocksize + remainder;
324 uio_setoffset(uio, this_offset);
325 this_resid = this_block_count * blocksize - remainder;
326 if (this_resid > resid) {
327 this_resid = resid;
328 }
329 uio_setresid(uio, this_resid);
330 error = VNOP_READ(vp, uio, ioflag, ctx);
331 if (error) {
332 break;
333 }
334
335 /* figure out how much we actually read */
336 this_resid -= uio_resid(uio);
337 if (this_resid == 0) {
338 printf("vn device: vnread_shadow zero length read\n");
339 break;
340 }
341 resid -= this_resid;
342 offset += this_resid;
343 }
344 uio_setresid(uio, resid);
345 uio_setoffset(uio, offset);
346 return (error);
347 }
348
349 static int
350 vncopy_block_to_shadow(struct vn_softc * vn, vfs_context_t ctx,
351 u_long file_block, u_long shadow_block)
352 {
353 int error;
354 char * tmpbuf;
355
356 tmpbuf = _MALLOC(vn->sc_secsize, M_TEMP, M_WAITOK);
357 if (tmpbuf == NULL) {
358 return (ENOMEM);
359 }
360 /* read one block from file at file_block offset */
361 error = file_io(vn->sc_vp, ctx, UIO_READ,
362 tmpbuf, (off_t)file_block * vn->sc_secsize,
363 vn->sc_secsize, NULL);
364 if (error) {
365 goto done;
366 }
367 /* write one block to shadow file at shadow_block offset */
368 error = file_io(vn->sc_shadow_vp, ctx, UIO_WRITE,
369 tmpbuf, (off_t)shadow_block * vn->sc_secsize,
370 vn->sc_secsize, NULL);
371 done:
372 FREE(tmpbuf, M_TEMP);
373 return (error);
374 }
375
376 enum {
377 FLAGS_FIRST_BLOCK_PARTIAL = 0x1,
378 FLAGS_LAST_BLOCK_PARTIAL = 0x2
379 };
380
381 static int
382 vnwrite_shadow(struct vn_softc * vn, struct uio *uio, int ioflag,
383 vfs_context_t ctx)
384 {
385 u_long blocksize = vn->sc_secsize;
386 int error = 0;
387 user_ssize_t resid;
388 off_t offset;
389
390 resid = uio_resid(uio);
391 offset = uio_offset(uio);
392
393 while (resid > 0) {
394 int flags = 0;
395 u_long offset_block_number;
396 u_long remainder;
397 u_long resid_block_count;
398 u_long shadow_block_count;
399 u_long shadow_block_number;
400 user_ssize_t this_resid;
401
402 /* figure out which blocks to write */
403 offset_block_number = block_truncate(offset, blocksize);
404 remainder = block_remainder(offset, blocksize);
405 resid_block_count = block_round(resid + remainder, blocksize);
406 /* figure out if the first or last blocks are partial writes */
407 if (remainder > 0
408 && !shadow_map_is_written(vn->sc_shadow_map,
409 offset_block_number)) {
410 /* the first block is a partial write */
411 flags |= FLAGS_FIRST_BLOCK_PARTIAL;
412 }
413 if (resid_block_count > 1
414 && !shadow_map_is_written(vn->sc_shadow_map,
415 offset_block_number
416 + resid_block_count - 1)
417 && block_remainder(offset + resid, blocksize) > 0) {
418 /* the last block is a partial write */
419 flags |= FLAGS_LAST_BLOCK_PARTIAL;
420 }
421 if (shadow_map_write(vn->sc_shadow_map,
422 offset_block_number, resid_block_count,
423 &shadow_block_number,
424 &shadow_block_count)) {
425 /* shadow file is growing */
426 #if 0
427 /* truncate the file to its new length before write */
428 off_t size;
429 size = (off_t)shadow_map_shadow_size(vn->sc_shadow_map)
430 * vn->sc_secsize;
431 vnode_setsize(vn->sc_shadow_vp, size, IO_SYNC, ctx);
432 #endif 0
433 }
434 /* write the blocks (or parts thereof) */
435 uio_setoffset(uio, (off_t)
436 shadow_block_number * blocksize + remainder);
437 this_resid = (off_t)shadow_block_count * blocksize - remainder;
438 if (this_resid >= resid) {
439 this_resid = resid;
440 if ((flags & FLAGS_LAST_BLOCK_PARTIAL) != 0) {
441 /* copy the last block to the shadow */
442 u_long d;
443 u_long s;
444
445 s = offset_block_number
446 + resid_block_count - 1;
447 d = shadow_block_number
448 + shadow_block_count - 1;
449 error = vncopy_block_to_shadow(vn, ctx, s, d);
450 if (error) {
451 printf("vnwrite_shadow: failed to copy"
452 " block %lu to shadow block %lu\n",
453 s, d);
454 break;
455 }
456 }
457 }
458 uio_setresid(uio, this_resid);
459 if ((flags & FLAGS_FIRST_BLOCK_PARTIAL) != 0) {
460 /* copy the first block to the shadow */
461 error = vncopy_block_to_shadow(vn, ctx,
462 offset_block_number,
463 shadow_block_number);
464 if (error) {
465 printf("vnwrite_shadow: failed to"
466 " copy block %lu to shadow block %lu\n",
467 offset_block_number,
468 shadow_block_number);
469 break;
470 }
471 }
472 error = VNOP_WRITE(vn->sc_shadow_vp, uio, ioflag, ctx);
473 if (error) {
474 break;
475 }
476 /* figure out how much we actually wrote */
477 this_resid -= uio_resid(uio);
478 if (this_resid == 0) {
479 printf("vn device: vnwrite_shadow zero length write\n");
480 break;
481 }
482 resid -= this_resid;
483 offset += this_resid;
484 }
485 uio_setresid(uio, resid);
486 uio_setoffset(uio, offset);
487 return (error);
488 }
489
490 static int
491 vnread(dev_t dev, struct uio *uio, int ioflag)
492 {
493 struct vfs_context context;
494 int error = 0;
495 boolean_t funnel_state;
496 off_t offset;
497 proc_t p;
498 user_ssize_t resid;
499 struct vn_softc * vn;
500 int unit;
501
502 unit = vnunit(dev);
503 if (vnunit(dev) >= NVNDEVICE) {
504 return (ENXIO);
505 }
506 p = current_proc();
507 funnel_state = thread_funnel_set(kernel_flock, TRUE);
508 vn = vn_table + unit;
509 if ((vn->sc_flags & VNF_INITED) == 0) {
510 error = ENXIO;
511 goto done;
512 }
513
514 context.vc_thread = current_thread();
515 context.vc_ucred = vn->sc_cred;
516
517 error = vnode_getwithvid(vn->sc_vp, vn->sc_vid);
518 if (error != 0) {
519 /* the vnode is no longer available, abort */
520 error = ENXIO;
521 vnclear(vn, &context);
522 goto done;
523 }
524
525 resid = uio_resid(uio);
526 offset = uio_offset(uio);
527
528 /*
529 * If out of bounds return an error. If at the EOF point,
530 * simply read less.
531 */
532 if (offset >= (off_t)vn->sc_fsize) {
533 if (offset > (off_t)vn->sc_fsize) {
534 error = EINVAL;
535 }
536 goto done;
537 }
538 /*
539 * If the request crosses EOF, truncate the request.
540 */
541 if ((offset + resid) > (off_t)vn->sc_fsize) {
542 resid = vn->sc_fsize - offset;
543 uio_setresid(uio, resid);
544 }
545
546 if (vn->sc_shadow_vp != NULL) {
547 error = vnode_getwithvid(vn->sc_shadow_vp,
548 vn->sc_shadow_vid);
549 if (error != 0) {
550 /* the vnode is no longer available, abort */
551 error = ENXIO;
552 vnode_put(vn->sc_vp);
553 vnclear(vn, &context);
554 goto done;
555 }
556 error = vnread_shadow(vn, uio, ioflag, &context);
557 vnode_put(vn->sc_shadow_vp);
558 } else {
559 error = VNOP_READ(vn->sc_vp, uio, ioflag, &context);
560 }
561 vnode_put(vn->sc_vp);
562 done:
563 (void) thread_funnel_set(kernel_flock, funnel_state);
564 return (error);
565 }
566
567 static int
568 vnwrite(dev_t dev, struct uio *uio, int ioflag)
569 {
570 struct vfs_context context;
571 int error;
572 boolean_t funnel_state;
573 off_t offset;
574 proc_t p;
575 user_ssize_t resid;
576 struct vn_softc * vn;
577 int unit;
578
579 unit = vnunit(dev);
580 if (vnunit(dev) >= NVNDEVICE) {
581 return (ENXIO);
582 }
583 p = current_proc();
584 funnel_state = thread_funnel_set(kernel_flock, TRUE);
585 vn = vn_table + unit;
586 if ((vn->sc_flags & VNF_INITED) == 0) {
587 error = ENXIO;
588 goto done;
589 }
590 if (vn->sc_flags & VNF_READONLY) {
591 error = EROFS;
592 goto done;
593 }
594
595 context.vc_thread = current_thread();
596 context.vc_ucred = vn->sc_cred;
597
598 error = vnode_getwithvid(vn->sc_vp, vn->sc_vid);
599 if (error != 0) {
600 /* the vnode is no longer available, abort */
601 error = ENXIO;
602 vnclear(vn, &context);
603 goto done;
604 }
605 resid = uio_resid(uio);
606 offset = uio_offset(uio);
607
608 /*
609 * If out of bounds return an error. If at the EOF point,
610 * simply write less.
611 */
612 if (offset >= (off_t)vn->sc_fsize) {
613 if (offset > (off_t)vn->sc_fsize) {
614 error = EINVAL;
615 }
616 goto done;
617 }
618 /*
619 * If the request crosses EOF, truncate the request.
620 */
621 if ((offset + resid) > (off_t)vn->sc_fsize) {
622 resid = (off_t)vn->sc_fsize - offset;
623 uio_setresid(uio, resid);
624 }
625
626 if (vn->sc_shadow_vp != NULL) {
627 error = vnode_getwithvid(vn->sc_shadow_vp,
628 vn->sc_shadow_vid);
629 if (error != 0) {
630 /* the vnode is no longer available, abort */
631 error = ENXIO;
632 vnode_put(vn->sc_vp);
633 vnclear(vn, &context);
634 goto done;
635 }
636 error = vnwrite_shadow(vn, uio, ioflag, &context);
637 vnode_put(vn->sc_shadow_vp);
638 } else {
639 error = VNOP_WRITE(vn->sc_vp, uio, ioflag, &context);
640 }
641 vnode_put(vn->sc_vp);
642 done:
643 (void) thread_funnel_set(kernel_flock, funnel_state);
644 return (error);
645 }
646
647 static int
648 shadow_read(struct vn_softc * vn, struct buf * bp, char * base,
649 vfs_context_t ctx)
650 {
651 u_long blocksize = vn->sc_secsize;
652 int error = 0;
653 u_long offset;
654 boolean_t read_shadow;
655 u_long resid;
656 u_long start = 0;
657
658 offset = buf_blkno(bp);
659 resid = buf_resid(bp) / blocksize;
660 while (resid > 0) {
661 user_ssize_t temp_resid;
662 u_long this_offset;
663 u_long this_resid;
664 struct vnode * vp;
665
666 read_shadow = shadow_map_read(vn->sc_shadow_map,
667 offset, resid,
668 &this_offset, &this_resid);
669 if (read_shadow) {
670 vp = vn->sc_shadow_vp;
671 }
672 else {
673 vp = vn->sc_vp;
674 }
675 error = file_io(vp, ctx, UIO_READ, base + start,
676 (off_t)this_offset * blocksize,
677 (user_ssize_t)this_resid * blocksize,
678 &temp_resid);
679 if (error) {
680 break;
681 }
682 this_resid -= (temp_resid / blocksize);
683 if (this_resid == 0) {
684 printf("vn device: shadow_read zero length read\n");
685 break;
686 }
687 resid -= this_resid;
688 offset += this_resid;
689 start += this_resid * blocksize;
690 }
691 buf_setresid(bp, resid * blocksize);
692 return (error);
693 }
694
695 static int
696 shadow_write(struct vn_softc * vn, struct buf * bp, char * base,
697 vfs_context_t ctx)
698 {
699 u_long blocksize = vn->sc_secsize;
700 int error = 0;
701 u_long offset;
702 boolean_t shadow_grew;
703 u_long resid;
704 u_long start = 0;
705
706 offset = buf_blkno(bp);
707 resid = buf_resid(bp) / blocksize;
708 while (resid > 0) {
709 user_ssize_t temp_resid;
710 u_long this_offset;
711 u_long this_resid;
712
713 shadow_grew = shadow_map_write(vn->sc_shadow_map,
714 offset, resid,
715 &this_offset, &this_resid);
716 if (shadow_grew) {
717 #if 0
718 off_t size;
719 /* truncate the file to its new length before write */
720 size = (off_t)shadow_map_shadow_size(vn->sc_shadow_map)
721 * blocksize;
722 vnode_setsize(vn->sc_shadow_vp, size, IO_SYNC, ctx);
723 #endif
724 }
725 error = file_io(vn->sc_shadow_vp, ctx, UIO_WRITE,
726 base + start,
727 (off_t)this_offset * blocksize,
728 (user_ssize_t)this_resid * blocksize,
729 &temp_resid);
730 if (error) {
731 break;
732 }
733 this_resid -= (temp_resid / blocksize);
734 if (this_resid == 0) {
735 printf("vn device: shadow_write zero length write\n");
736 break;
737 }
738 resid -= this_resid;
739 offset += this_resid;
740 start += this_resid * blocksize;
741 }
742 buf_setresid(bp, resid * blocksize);
743 return (error);
744 }
745
746 static int
747 vn_readwrite_io(struct vn_softc * vn, struct buf * bp, vfs_context_t ctx)
748 {
749 int error = 0;
750 char * iov_base;
751 caddr_t vaddr;
752
753 if (buf_map(bp, &vaddr))
754 panic("vn device: buf_map failed");
755 iov_base = (char *)vaddr;
756
757 if (vn->sc_shadow_vp == NULL) {
758 user_ssize_t temp_resid;
759
760 error = file_io(vn->sc_vp, ctx,
761 buf_flags(bp) & B_READ ? UIO_READ : UIO_WRITE,
762 iov_base,
763 (off_t)buf_blkno(bp) * vn->sc_secsize,
764 buf_resid(bp), &temp_resid);
765 buf_setresid(bp, temp_resid);
766 }
767 else {
768 if (buf_flags(bp) & B_READ)
769 error = shadow_read(vn, bp, iov_base, ctx);
770 else
771 error = shadow_write(vn, bp, iov_base, ctx);
772 }
773 buf_unmap(bp);
774
775 return (error);
776 }
777
778 static void
779 vnstrategy(struct buf *bp)
780 {
781 struct vn_softc *vn;
782 int error = 0;
783 long sz; /* in sc_secsize chunks */
784 daddr64_t blk_num;
785 boolean_t funnel_state;
786 struct vnode * shadow_vp = NULL;
787 struct vnode * vp = NULL;
788 struct vfs_context context;
789
790 funnel_state = thread_funnel_set(kernel_flock, TRUE);
791 vn = vn_table + vnunit(buf_device(bp));
792 if ((vn->sc_flags & VNF_INITED) == 0) {
793 error = ENXIO;
794 goto done;
795 }
796
797 context.vc_thread = current_thread();
798 context.vc_ucred = vn->sc_cred;
799
800 buf_setresid(bp, buf_count(bp));
801 /*
802 * Check for required alignment. Transfers must be a valid
803 * multiple of the sector size.
804 */
805 blk_num = buf_blkno(bp);
806 if (buf_count(bp) % vn->sc_secsize != 0) {
807 error = EINVAL;
808 goto done;
809 }
810 sz = howmany(buf_count(bp), vn->sc_secsize);
811
812 /*
813 * If out of bounds return an error. If at the EOF point,
814 * simply read or write less.
815 */
816 if (blk_num >= 0 && (u_int64_t)blk_num >= vn->sc_size) {
817 if (blk_num > 0 && (u_int64_t)blk_num > vn->sc_size) {
818 error = EINVAL;
819 }
820 goto done;
821 }
822 /*
823 * If the request crosses EOF, truncate the request.
824 */
825 if ((blk_num + sz) > 0 && ((u_int64_t)(blk_num + sz)) > vn->sc_size) {
826 buf_setcount(bp, (vn->sc_size - blk_num) * vn->sc_secsize);
827 buf_setresid(bp, buf_count(bp));
828 }
829 vp = vn->sc_vp;
830 if (vp == NULL) {
831 error = ENXIO;
832 goto done;
833 }
834
835 error = vnode_getwithvid(vp, vn->sc_vid);
836 if (error != 0) {
837 /* the vnode is no longer available, abort */
838 error = ENXIO;
839 vnclear(vn, &context);
840 goto done;
841 }
842 shadow_vp = vn->sc_shadow_vp;
843 if (shadow_vp != NULL) {
844 error = vnode_getwithvid(shadow_vp,
845 vn->sc_shadow_vid);
846 if (error != 0) {
847 /* the vnode is no longer available, abort */
848 error = ENXIO;
849 vnode_put(vn->sc_vp);
850 vnclear(vn, &context);
851 goto done;
852 }
853 }
854
855 error = vn_readwrite_io(vn, bp, &context);
856 vnode_put(vp);
857 if (shadow_vp != NULL) {
858 vnode_put(shadow_vp);
859 }
860
861 done:
862 (void) thread_funnel_set(kernel_flock, funnel_state);
863 if (error) {
864 buf_seterror(bp, error);
865 }
866 buf_biodone(bp);
867 return;
868 }
869
870 /* ARGSUSED */
871 static int
872 vnioctl(dev_t dev, u_long cmd, caddr_t data,
873 __unused int flag, proc_t p,
874 int is_char)
875 {
876 struct vn_softc *vn;
877 struct user_vn_ioctl *viop;
878 int error;
879 u_int32_t *f;
880 u_int64_t * o;
881 int unit;
882 struct vfsioattr ioattr;
883 struct user_vn_ioctl user_vnio;
884 boolean_t funnel_state;
885 struct vfs_context context;
886
887 unit = vnunit(dev);
888 if (vnunit(dev) >= NVNDEVICE) {
889 return (ENXIO);
890 }
891
892 funnel_state = thread_funnel_set(kernel_flock, TRUE);
893 vn = vn_table + unit;
894 error = proc_suser(p);
895 if (error) {
896 goto done;
897 }
898
899 context.vc_thread = current_thread();
900 context.vc_ucred = vn->sc_cred;
901
902 viop = (struct user_vn_ioctl *)data;
903 f = (u_int32_t *)data;
904 o = (u_int64_t *)data;
905 switch (cmd) {
906 case VNIOCDETACH:
907 case VNIOCDETACH64:
908 case DKIOCGETBLOCKSIZE:
909 case DKIOCSETBLOCKSIZE:
910 case DKIOCGETMAXBLOCKCOUNTREAD:
911 case DKIOCGETMAXBLOCKCOUNTWRITE:
912 case DKIOCGETMAXSEGMENTCOUNTREAD:
913 case DKIOCGETMAXSEGMENTCOUNTWRITE:
914 case DKIOCGETMAXSEGMENTBYTECOUNTREAD:
915 case DKIOCGETMAXSEGMENTBYTECOUNTWRITE:
916 case DKIOCGETBLOCKCOUNT:
917 case DKIOCGETBLOCKCOUNT32:
918 if ((vn->sc_flags & VNF_INITED) == 0) {
919 error = ENXIO;
920 goto done;
921 }
922 break;
923 default:
924 break;
925 }
926
927 if (vn->sc_vp != NULL)
928 vfs_ioattr(vnode_mount(vn->sc_vp), &ioattr);
929 else
930 bzero(&ioattr, sizeof(ioattr));
931
932 switch (cmd) {
933 case DKIOCISVIRTUAL:
934 *f = 1;
935 break;
936 case DKIOCGETMAXBLOCKCOUNTREAD:
937 *o = ioattr.io_maxreadcnt / vn->sc_secsize;
938 break;
939 case DKIOCGETMAXBLOCKCOUNTWRITE:
940 *o = ioattr.io_maxwritecnt / vn->sc_secsize;
941 break;
942 case DKIOCGETMAXBYTECOUNTREAD:
943 *o = ioattr.io_maxreadcnt;
944 break;
945 case DKIOCGETMAXBYTECOUNTWRITE:
946 *o = ioattr.io_maxwritecnt;
947 break;
948 case DKIOCGETMAXSEGMENTCOUNTREAD:
949 *o = ioattr.io_segreadcnt;
950 break;
951 case DKIOCGETMAXSEGMENTCOUNTWRITE:
952 *o = ioattr.io_segwritecnt;
953 break;
954 case DKIOCGETMAXSEGMENTBYTECOUNTREAD:
955 *o = ioattr.io_maxsegreadsize;
956 break;
957 case DKIOCGETMAXSEGMENTBYTECOUNTWRITE:
958 *o = ioattr.io_maxsegwritesize;
959 break;
960 case DKIOCGETBLOCKSIZE:
961 *f = vn->sc_secsize;
962 break;
963 case DKIOCSETBLOCKSIZE:
964 if (is_char) {
965 /* can only set block size on block device */
966 error = ENODEV;
967 break;
968 }
969 if (*f < DEV_BSIZE) {
970 error = EINVAL;
971 break;
972 }
973 if (vn->sc_shadow_vp != NULL) {
974 if (*f == (unsigned)vn->sc_secsize) {
975 break;
976 }
977 /* can't change the block size if already shadowing */
978 error = EBUSY;
979 break;
980 }
981 vn->sc_secsize = *f;
982 /* recompute the size in terms of the new blocksize */
983 vn->sc_size = vn->sc_fsize / vn->sc_secsize;
984 break;
985 case DKIOCISWRITABLE:
986 *f = 1;
987 break;
988 case DKIOCGETBLOCKCOUNT32:
989 *f = vn->sc_size;
990 break;
991 case DKIOCGETBLOCKCOUNT:
992 *o = vn->sc_size;
993 break;
994 case VNIOCSHADOW:
995 case VNIOCSHADOW64:
996 if (vn->sc_shadow_vp != NULL) {
997 error = EBUSY;
998 break;
999 }
1000 if (vn->sc_vp == NULL) {
1001 /* much be attached before we can shadow */
1002 error = EINVAL;
1003 break;
1004 }
1005 if (!proc_is64bit(p)) {
1006 /* downstream code expects LP64 version of vn_ioctl structure */
1007 vn_ioctl_to_64((struct vn_ioctl *)viop, &user_vnio);
1008 viop = &user_vnio;
1009 }
1010 if (viop->vn_file == USER_ADDR_NULL) {
1011 error = EINVAL;
1012 break;
1013 }
1014 error = vniocattach_shadow(vn, viop, dev, 0, p);
1015 break;
1016
1017 case VNIOCATTACH:
1018 case VNIOCATTACH64:
1019 if (is_char) {
1020 /* attach only on block device */
1021 error = ENODEV;
1022 break;
1023 }
1024 if (vn->sc_flags & VNF_INITED) {
1025 error = EBUSY;
1026 break;
1027 }
1028 if (!proc_is64bit(p)) {
1029 /* downstream code expects LP64 version of vn_ioctl structure */
1030 vn_ioctl_to_64((struct vn_ioctl *)viop, &user_vnio);
1031 viop = &user_vnio;
1032 }
1033 if (viop->vn_file == USER_ADDR_NULL) {
1034 error = EINVAL;
1035 break;
1036 }
1037 error = vniocattach_file(vn, viop, dev, 0, p);
1038 break;
1039
1040 case VNIOCDETACH:
1041 case VNIOCDETACH64:
1042 if (is_char) {
1043 /* detach only on block device */
1044 error = ENODEV;
1045 break;
1046 }
1047 /* Note: spec_open won't open a mounted block device */
1048
1049 /*
1050 * XXX handle i/o in progress. Return EBUSY, or wait, or
1051 * flush the i/o.
1052 * XXX handle multiple opens of the device. Return EBUSY,
1053 * or revoke the fd's.
1054 * How are these problems handled for removable and failing
1055 * hardware devices? (Hint: They are not)
1056 */
1057 vnclear(vn, &context);
1058 break;
1059
1060 case VNIOCGSET:
1061 vn_options |= *f;
1062 *f = vn_options;
1063 break;
1064
1065 case VNIOCGCLEAR:
1066 vn_options &= ~(*f);
1067 *f = vn_options;
1068 break;
1069
1070 case VNIOCUSET:
1071 vn->sc_options |= *f;
1072 *f = vn->sc_options;
1073 break;
1074
1075 case VNIOCUCLEAR:
1076 vn->sc_options &= ~(*f);
1077 *f = vn->sc_options;
1078 break;
1079
1080 default:
1081 error = ENOTTY;
1082 break;
1083 }
1084 done:
1085 (void) thread_funnel_set(kernel_flock, funnel_state);
1086 return(error);
1087 }
1088
1089 static int
1090 vnioctl_chr(dev_t dev, u_long cmd, caddr_t data, int flag, proc_t p)
1091 {
1092 return (vnioctl(dev, cmd, data, flag, p, TRUE));
1093 }
1094
1095 static int
1096 vnioctl_blk(dev_t dev, u_long cmd, caddr_t data, int flag, proc_t p)
1097 {
1098 return (vnioctl(dev, cmd, data, flag, p, FALSE));
1099 }
1100
1101 /*
1102 * vniocattach_file:
1103 *
1104 * Attach a file to a VN partition. Return the size in the vn_size
1105 * field.
1106 */
1107
1108 static int
1109 vniocattach_file(struct vn_softc *vn,
1110 struct user_vn_ioctl *vniop,
1111 dev_t dev,
1112 int in_kernel,
1113 proc_t p)
1114 {
1115 dev_t cdev;
1116 vfs_context_t ctx = vfs_context_current();
1117 kauth_cred_t cred;
1118 struct nameidata nd;
1119 off_t file_size;
1120 int error, flags;
1121
1122 flags = FREAD|FWRITE;
1123 if (in_kernel) {
1124 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE32, vniop->vn_file, ctx);
1125 }
1126 else {
1127 NDINIT(&nd, LOOKUP, FOLLOW,
1128 (IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32),
1129 vniop->vn_file, ctx);
1130 }
1131 /* vn_open gives both long- and short-term references */
1132 error = vn_open(&nd, flags, 0);
1133 if (error) {
1134 if (error != EACCES && error != EPERM && error != EROFS) {
1135 return (error);
1136 }
1137 flags &= ~FWRITE;
1138 if (in_kernel) {
1139 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE32,
1140 vniop->vn_file, ctx);
1141 }
1142 else {
1143 NDINIT(&nd, LOOKUP, FOLLOW,
1144 (IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32),
1145 vniop->vn_file, ctx);
1146 }
1147 error = vn_open(&nd, flags, 0);
1148 if (error) {
1149 return (error);
1150 }
1151 }
1152 if (nd.ni_vp->v_type != VREG) {
1153 error = EINVAL;
1154 }
1155 else {
1156 error = vnode_size(nd.ni_vp, &file_size, ctx);
1157 }
1158 if (error != 0) {
1159 (void) vn_close(nd.ni_vp, flags, ctx);
1160 vnode_put(nd.ni_vp);
1161 return (error);
1162 }
1163 cred = kauth_cred_proc_ref(p);
1164 nd.ni_vp->v_flag |= VNOCACHE_DATA;
1165 error = setcred(nd.ni_vp, cred);
1166 if (error) {
1167 (void)vn_close(nd.ni_vp, flags, ctx);
1168 vnode_put(nd.ni_vp);
1169 kauth_cred_unref(&cred);
1170 return(error);
1171 }
1172 vn->sc_secsize = DEV_BSIZE;
1173 vn->sc_fsize = file_size;
1174 vn->sc_size = file_size / vn->sc_secsize;
1175 vn->sc_vp = nd.ni_vp;
1176 vn->sc_vid = vnode_vid(nd.ni_vp);
1177 vn->sc_open_flags = flags;
1178 vn->sc_cred = cred;
1179 cdev = makedev(vndevice_cdev_major, minor(dev));
1180 vn->sc_cdev = devfs_make_node(cdev, DEVFS_CHAR,
1181 UID_ROOT, GID_OPERATOR,
1182 0600, "rvn%d",
1183 minor(dev));
1184 vn->sc_flags |= VNF_INITED;
1185 if (flags == FREAD)
1186 vn->sc_flags |= VNF_READONLY;
1187 /* lose the short-term reference */
1188 vnode_put(nd.ni_vp);
1189 return(0);
1190 }
1191
1192 static int
1193 vniocattach_shadow(struct vn_softc *vn, struct user_vn_ioctl *vniop,
1194 __unused int dev, int in_kernel, proc_t p)
1195 {
1196 vfs_context_t ctx = vfs_context_current();
1197 struct nameidata nd;
1198 int error, flags;
1199 shadow_map_t * map;
1200 off_t file_size;
1201
1202 flags = FREAD|FWRITE;
1203 if (in_kernel) {
1204 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE32, vniop->vn_file, ctx);
1205 }
1206 else {
1207 NDINIT(&nd, LOOKUP, FOLLOW,
1208 (IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32),
1209 vniop->vn_file, ctx);
1210 }
1211 /* vn_open gives both long- and short-term references */
1212 error = vn_open(&nd, flags, 0);
1213 if (error) {
1214 /* shadow MUST be writable! */
1215 return (error);
1216 }
1217 if (nd.ni_vp->v_type != VREG
1218 || (error = vnode_size(nd.ni_vp, &file_size, ctx))) {
1219 (void)vn_close(nd.ni_vp, flags, ctx);
1220 vnode_put(nd.ni_vp);
1221 return (error ? error : EINVAL);
1222 }
1223 map = shadow_map_create(vn->sc_fsize, file_size,
1224 0, vn->sc_secsize);
1225 if (map == NULL) {
1226 (void)vn_close(nd.ni_vp, flags, ctx);
1227 vnode_put(nd.ni_vp);
1228 vn->sc_shadow_vp = NULL;
1229 return (ENOMEM);
1230 }
1231 vn->sc_shadow_vp = nd.ni_vp;
1232 vn->sc_shadow_vid = vnode_vid(nd.ni_vp);
1233 vn->sc_shadow_vp->v_flag |= VNOCACHE_DATA;
1234 vn->sc_shadow_map = map;
1235 vn->sc_flags &= ~VNF_READONLY; /* we're now read/write */
1236
1237 /* lose the short-term reference */
1238 vnode_put(nd.ni_vp);
1239 return(0);
1240 }
1241
1242 int
1243 vndevice_root_image(char * path, char devname[], dev_t * dev_p)
1244 {
1245 int error = 0;
1246 struct vn_softc * vn;
1247 struct user_vn_ioctl vnio;
1248
1249 vnio.vn_file = CAST_USER_ADDR_T(path);
1250 vnio.vn_size = 0;
1251
1252 vn = vn_table + ROOT_IMAGE_UNIT;
1253 *dev_p = makedev(vndevice_bdev_major,
1254 ROOT_IMAGE_UNIT);
1255 snprintf(devname, 16, "vn%d", ROOT_IMAGE_UNIT);
1256 error = vniocattach_file(vn, &vnio, *dev_p, 1, current_proc());
1257 return (error);
1258 }
1259
1260 /*
1261 * Duplicate the current processes' credentials. Since we are called only
1262 * as the result of a SET ioctl and only root can do that, any future access
1263 * to this "disk" is essentially as root. Note that credentials may change
1264 * if some other uid can write directly to the mapped file (NFS).
1265 */
1266 static int
1267 setcred(struct vnode * vp, kauth_cred_t cred)
1268 {
1269 char *tmpbuf;
1270 int error = 0;
1271 struct vfs_context context;
1272
1273 /*
1274 * Horrible kludge to establish credentials for NFS XXX.
1275 */
1276 context.vc_thread = current_thread();
1277 context.vc_ucred = cred;
1278 tmpbuf = _MALLOC(DEV_BSIZE, M_TEMP, M_WAITOK);
1279 error = file_io(vp, &context, UIO_READ, tmpbuf, 0, DEV_BSIZE, NULL);
1280 FREE(tmpbuf, M_TEMP);
1281 return (error);
1282 }
1283
1284 void
1285 vnclear(struct vn_softc *vn, vfs_context_t ctx)
1286 {
1287 if (vn->sc_vp != NULL) {
1288 /* release long-term reference */
1289 (void)vn_close(vn->sc_vp, vn->sc_open_flags, ctx);
1290 vn->sc_vp = NULL;
1291 }
1292 if (vn->sc_shadow_vp != NULL) {
1293 /* release long-term reference */
1294 (void)vn_close(vn->sc_shadow_vp, FREAD | FWRITE, ctx);
1295 vn->sc_shadow_vp = NULL;
1296 }
1297 if (vn->sc_shadow_map != NULL) {
1298 shadow_map_free(vn->sc_shadow_map);
1299 vn->sc_shadow_map = NULL;
1300 }
1301 vn->sc_flags &= ~(VNF_INITED | VNF_READONLY);
1302 if (vn->sc_cred) {
1303 kauth_cred_unref(&vn->sc_cred);
1304 }
1305 vn->sc_size = 0;
1306 vn->sc_fsize = 0;
1307 if (vn->sc_cdev) {
1308 devfs_remove(vn->sc_cdev);
1309 vn->sc_cdev = NULL;
1310 }
1311 }
1312
1313 static int
1314 vnsize(dev_t dev)
1315 {
1316 int secsize;
1317 struct vn_softc *vn;
1318 int unit;
1319 boolean_t funnel_state;
1320
1321 unit = vnunit(dev);
1322 if (vnunit(dev) >= NVNDEVICE) {
1323 return (-1);
1324 }
1325
1326 funnel_state = thread_funnel_set(kernel_flock, TRUE);
1327 vn = vn_table + unit;
1328 if ((vn->sc_flags & VNF_INITED) == 0)
1329 secsize = -1;
1330 else
1331 secsize = vn->sc_secsize;
1332 (void) thread_funnel_set(kernel_flock, funnel_state);
1333 return (secsize);
1334 }
1335
1336 #define CDEV_MAJOR -1
1337 #define BDEV_MAJOR -1
1338 static int vndevice_inited = 0;
1339
1340 void
1341 vndevice_init(void)
1342 {
1343 if (vndevice_inited)
1344 return;
1345
1346 vndevice_do_init();
1347 }
1348
1349 static void
1350 vndevice_do_init( void )
1351 {
1352 int i;
1353
1354 vndevice_bdev_major = bdevsw_add(BDEV_MAJOR, &vn_bdevsw);
1355
1356 if (vndevice_bdev_major < 0) {
1357 printf("vndevice_init: bdevsw_add() returned %d\n",
1358 vndevice_bdev_major);
1359 return;
1360 }
1361 vndevice_cdev_major = cdevsw_add_with_bdev(CDEV_MAJOR, &vn_cdevsw,
1362 vndevice_bdev_major);
1363 if (vndevice_cdev_major < 0) {
1364 printf("vndevice_init: cdevsw_add() returned %d\n",
1365 vndevice_cdev_major);
1366 return;
1367 }
1368 for (i = 0; i < NVNDEVICE; i++) {
1369 dev_t dev = makedev(vndevice_bdev_major, i);
1370 vn_table[i].sc_bdev = devfs_make_node(dev, DEVFS_BLOCK,
1371 UID_ROOT, GID_OPERATOR,
1372 0600, "vn%d",
1373 i);
1374 if (vn_table[i].sc_bdev == NULL)
1375 printf("vninit: devfs_make_node failed!\n");
1376 }
1377 }
1378
1379 static void
1380 vn_ioctl_to_64(struct vn_ioctl *from, struct user_vn_ioctl *to)
1381 {
1382 to->vn_file = CAST_USER_ADDR_T(from->vn_file);
1383 to->vn_size = from->vn_size;
1384 to->vn_control = from->vn_control;
1385 }
1386
1387 #endif /* NVNDEVICE */
mirror of XNU