1 /*-
2 * SPDX-License-Identifier: (Beerware AND BSD-3-Clause)
3 *
4 * ----------------------------------------------------------------------------
5 * "THE BEER-WARE LICENSE" (Revision 42):
6 * <[email protected]> wrote this file. As long as you retain this notice you
7 * can do whatever you want with this stuff. If we meet some day, and you think
8 * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp
9 * ----------------------------------------------------------------------------
10 *
11 * $FreeBSD$
12 *
13 */
14
15 /*-
16 * The following functions are based on the vn(4) driver: mdstart_swap(),
17 * mdstart_vnode(), mdcreate_swap(), mdcreate_vnode() and mddestroy(),
18 * and as such under the following copyright:
19 *
20 * Copyright (c) 1988 University of Utah.
21 * Copyright (c) 1990, 1993
22 * The Regents of the University of California. All rights reserved.
23 * Copyright (c) 2013 The FreeBSD Foundation
24 * All rights reserved.
25 *
26 * This code is derived from software contributed to Berkeley by
27 * the Systems Programming Group of the University of Utah Computer
28 * Science Department.
29 *
30 * Portions of this software were developed by Konstantin Belousov
31 * under sponsorship from the FreeBSD Foundation.
32 *
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
35 * are met:
36 * 1. Redistributions of source code must retain the above copyright
37 * notice, this list of conditions and the following disclaimer.
38 * 2. Redistributions in binary form must reproduce the above copyright
39 * notice, this list of conditions and the following disclaimer in the
40 * documentation and/or other materials provided with the distribution.
41 * 3. Neither the name of the University nor the names of its contributors
42 * may be used to endorse or promote products derived from this software
43 * without specific prior written permission.
44 *
45 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55 * SUCH DAMAGE.
56 *
57 * from: Utah Hdr: vn.c 1.13 94/04/02
58 *
59 * from: @(#)vn.c 8.6 (Berkeley) 4/1/94
60 * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03
61 */
62
63 #include "opt_rootdevname.h"
64 #include "opt_geom.h"
65 #include "opt_md.h"
66
67 #include <sys/param.h>
68 #include <sys/systm.h>
69 #include <sys/bio.h>
70 #include <sys/buf.h>
71 #include <sys/conf.h>
72 #include <sys/devicestat.h>
73 #include <sys/fcntl.h>
74 #include <sys/kernel.h>
75 #include <sys/kthread.h>
76 #include <sys/limits.h>
77 #include <sys/linker.h>
78 #include <sys/lock.h>
79 #include <sys/malloc.h>
80 #include <sys/mdioctl.h>
81 #include <sys/mount.h>
82 #include <sys/mutex.h>
83 #include <sys/sx.h>
84 #include <sys/namei.h>
85 #include <sys/proc.h>
86 #include <sys/queue.h>
87 #include <sys/rwlock.h>
88 #include <sys/sbuf.h>
89 #include <sys/sched.h>
90 #include <sys/sf_buf.h>
91 #include <sys/sysctl.h>
92 #include <sys/uio.h>
93 #include <sys/vnode.h>
94 #include <sys/disk.h>
95
96 #include <geom/geom.h>
97 #include <geom/geom_int.h>
98
99 #include <vm/vm.h>
100 #include <vm/vm_param.h>
101 #include <vm/vm_object.h>
102 #include <vm/vm_page.h>
103 #include <vm/vm_pager.h>
104 #include <vm/swap_pager.h>
105 #include <vm/uma.h>
106
107 #include <machine/bus.h>
108
109 #define MD_MODVER 1
110
111 #define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */
112 #define MD_EXITING 0x20000 /* Worker thread is exiting. */
113 #define MD_PROVIDERGONE 0x40000 /* Safe to free the softc */
114
115 #ifndef MD_NSECT
116 #define MD_NSECT (10000 * 2)
117 #endif
118
119 struct md_req {
120 unsigned md_unit; /* unit number */
121 enum md_types md_type; /* type of disk */
122 off_t md_mediasize; /* size of disk in bytes */
123 unsigned md_sectorsize; /* sectorsize */
124 unsigned md_options; /* options */
125 int md_fwheads; /* firmware heads */
126 int md_fwsectors; /* firmware sectors */
127 char *md_file; /* pathname of file to mount */
128 enum uio_seg md_file_seg; /* location of md_file */
129 char *md_label; /* label of the device (userspace) */
130 int *md_units; /* pointer to units array (kernel) */
131 size_t md_units_nitems; /* items in md_units array */
132 };
133
134 #ifdef COMPAT_FREEBSD32
135 struct md_ioctl32 {
136 unsigned md_version;
137 unsigned md_unit;
138 enum md_types md_type;
139 uint32_t md_file;
140 off_t md_mediasize;
141 unsigned md_sectorsize;
142 unsigned md_options;
143 uint64_t md_base;
144 int md_fwheads;
145 int md_fwsectors;
146 uint32_t md_label;
147 int md_pad[MDNPAD];
148 } __attribute__((__packed__));
149 CTASSERT((sizeof(struct md_ioctl32)) == 436);
150
151 #define MDIOCATTACH_32 _IOC_NEWTYPE(MDIOCATTACH, struct md_ioctl32)
152 #define MDIOCDETACH_32 _IOC_NEWTYPE(MDIOCDETACH, struct md_ioctl32)
153 #define MDIOCQUERY_32 _IOC_NEWTYPE(MDIOCQUERY, struct md_ioctl32)
154 #define MDIOCRESIZE_32 _IOC_NEWTYPE(MDIOCRESIZE, struct md_ioctl32)
155 #endif /* COMPAT_FREEBSD32 */
156
157 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk");
158 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors");
159
160 static int md_debug;
161 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0,
162 "Enable md(4) debug messages");
163 static int md_malloc_wait;
164 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0,
165 "Allow malloc to wait for memory allocations");
166
167 #if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE)
168 #define MD_ROOT_FSTYPE "ufs"
169 #endif
170
171 #if defined(MD_ROOT)
172 /*
173 * Preloaded image gets put here.
174 */
175 #if defined(MD_ROOT_SIZE)
176 /*
177 * We put the mfs_root symbol into the oldmfs section of the kernel object file.
178 * Applications that patch the object with the image can determine
179 * the size looking at the oldmfs section size within the kernel.
180 */
181 u_char mfs_root[MD_ROOT_SIZE*1024] __attribute__ ((section ("oldmfs")));
182 const int mfs_root_size = sizeof(mfs_root);
183 #elif defined(MD_ROOT_MEM)
184 /* MD region already mapped in the memory */
185 u_char *mfs_root;
186 int mfs_root_size;
187 #else
188 extern volatile u_char __weak_symbol mfs_root;
189 extern volatile u_char __weak_symbol mfs_root_end;
190 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root))
191 #endif
192 #endif
193
194 static g_init_t g_md_init;
195 static g_fini_t g_md_fini;
196 static g_start_t g_md_start;
197 static g_access_t g_md_access;
198 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
199 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
200 static g_provgone_t g_md_providergone;
201
202 static struct cdev *status_dev = NULL;
203 static struct sx md_sx;
204 static struct unrhdr *md_uh;
205
206 static d_ioctl_t mdctlioctl;
207
208 static struct cdevsw mdctl_cdevsw = {
209 .d_version = D_VERSION,
210 .d_ioctl = mdctlioctl,
211 .d_name = MD_NAME,
212 };
213
214 struct g_class g_md_class = {
215 .name = "MD",
216 .version = G_VERSION,
217 .init = g_md_init,
218 .fini = g_md_fini,
219 .start = g_md_start,
220 .access = g_md_access,
221 .dumpconf = g_md_dumpconf,
222 .providergone = g_md_providergone,
223 };
224
225 DECLARE_GEOM_CLASS(g_md_class, g_md);
226 MODULE_VERSION(geom_md, 0);
227
228 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
229
230 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t))
231 #define NMASK (NINDIR-1)
232 static int nshift;
233
234 static uma_zone_t md_pbuf_zone;
235
236 struct indir {
237 uintptr_t *array;
238 u_int total;
239 u_int used;
240 u_int shift;
241 };
242
243 struct md_s {
244 int unit;
245 LIST_ENTRY(md_s) list;
246 struct bio_queue_head bio_queue;
247 struct mtx queue_mtx;
248 struct cdev *dev;
249 enum md_types type;
250 off_t mediasize;
251 unsigned sectorsize;
252 unsigned opencount;
253 unsigned fwheads;
254 unsigned fwsectors;
255 char ident[32];
256 unsigned flags;
257 char name[20];
258 struct proc *procp;
259 struct g_geom *gp;
260 struct g_provider *pp;
261 int (*start)(struct md_s *sc, struct bio *bp);
262 struct devstat *devstat;
263
264 /* MD_MALLOC related fields */
265 struct indir *indir;
266 uma_zone_t uma;
267
268 /* MD_PRELOAD related fields */
269 u_char *pl_ptr;
270 size_t pl_len;
271
272 /* MD_VNODE related fields */
273 struct vnode *vnode;
274 char file[PATH_MAX];
275 char label[PATH_MAX];
276 struct ucred *cred;
277
278 /* MD_SWAP related fields */
279 vm_object_t object;
280 };
281
282 static struct indir *
new_indir(u_int shift)283 new_indir(u_int shift)
284 {
285 struct indir *ip;
286
287 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
288 | M_ZERO);
289 if (ip == NULL)
290 return (NULL);
291 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
292 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
293 if (ip->array == NULL) {
294 free(ip, M_MD);
295 return (NULL);
296 }
297 ip->total = NINDIR;
298 ip->shift = shift;
299 return (ip);
300 }
301
302 static void
del_indir(struct indir * ip)303 del_indir(struct indir *ip)
304 {
305
306 free(ip->array, M_MDSECT);
307 free(ip, M_MD);
308 }
309
310 static void
destroy_indir(struct md_s * sc,struct indir * ip)311 destroy_indir(struct md_s *sc, struct indir *ip)
312 {
313 int i;
314
315 for (i = 0; i < NINDIR; i++) {
316 if (!ip->array[i])
317 continue;
318 if (ip->shift)
319 destroy_indir(sc, (struct indir*)(ip->array[i]));
320 else if (ip->array[i] > 255)
321 uma_zfree(sc->uma, (void *)(ip->array[i]));
322 }
323 del_indir(ip);
324 }
325
326 /*
327 * This function does the math and allocates the top level "indir" structure
328 * for a device of "size" sectors.
329 */
330
331 static struct indir *
dimension(off_t size)332 dimension(off_t size)
333 {
334 off_t rcnt;
335 struct indir *ip;
336 int layer;
337
338 rcnt = size;
339 layer = 0;
340 while (rcnt > NINDIR) {
341 rcnt /= NINDIR;
342 layer++;
343 }
344
345 /*
346 * XXX: the top layer is probably not fully populated, so we allocate
347 * too much space for ip->array in here.
348 */
349 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
350 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
351 M_MDSECT, M_WAITOK | M_ZERO);
352 ip->total = NINDIR;
353 ip->shift = layer * nshift;
354 return (ip);
355 }
356
357 /*
358 * Read a given sector
359 */
360
361 static uintptr_t
s_read(struct indir * ip,off_t offset)362 s_read(struct indir *ip, off_t offset)
363 {
364 struct indir *cip;
365 int idx;
366 uintptr_t up;
367
368 if (md_debug > 1)
369 printf("s_read(%jd)\n", (intmax_t)offset);
370 up = 0;
371 for (cip = ip; cip != NULL;) {
372 if (cip->shift) {
373 idx = (offset >> cip->shift) & NMASK;
374 up = cip->array[idx];
375 cip = (struct indir *)up;
376 continue;
377 }
378 idx = offset & NMASK;
379 return (cip->array[idx]);
380 }
381 return (0);
382 }
383
384 /*
385 * Write a given sector, prune the tree if the value is 0
386 */
387
388 static int
s_write(struct indir * ip,off_t offset,uintptr_t ptr)389 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
390 {
391 struct indir *cip, *lip[10];
392 int idx, li;
393 uintptr_t up;
394
395 if (md_debug > 1)
396 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
397 up = 0;
398 li = 0;
399 cip = ip;
400 for (;;) {
401 lip[li++] = cip;
402 if (cip->shift) {
403 idx = (offset >> cip->shift) & NMASK;
404 up = cip->array[idx];
405 if (up != 0) {
406 cip = (struct indir *)up;
407 continue;
408 }
409 /* Allocate branch */
410 cip->array[idx] =
411 (uintptr_t)new_indir(cip->shift - nshift);
412 if (cip->array[idx] == 0)
413 return (ENOSPC);
414 cip->used++;
415 up = cip->array[idx];
416 cip = (struct indir *)up;
417 continue;
418 }
419 /* leafnode */
420 idx = offset & NMASK;
421 up = cip->array[idx];
422 if (up != 0)
423 cip->used--;
424 cip->array[idx] = ptr;
425 if (ptr != 0)
426 cip->used++;
427 break;
428 }
429 if (cip->used != 0 || li == 1)
430 return (0);
431 li--;
432 while (cip->used == 0 && cip != ip) {
433 li--;
434 idx = (offset >> lip[li]->shift) & NMASK;
435 up = lip[li]->array[idx];
436 KASSERT(up == (uintptr_t)cip, ("md screwed up"));
437 del_indir(cip);
438 lip[li]->array[idx] = 0;
439 lip[li]->used--;
440 cip = lip[li];
441 }
442 return (0);
443 }
444
445 static int
g_md_access(struct g_provider * pp,int r,int w,int e)446 g_md_access(struct g_provider *pp, int r, int w, int e)
447 {
448 struct md_s *sc;
449
450 sc = pp->geom->softc;
451 if (sc == NULL) {
452 if (r <= 0 && w <= 0 && e <= 0)
453 return (0);
454 return (ENXIO);
455 }
456 r += pp->acr;
457 w += pp->acw;
458 e += pp->ace;
459 if ((sc->flags & MD_READONLY) != 0 && w > 0)
460 return (EROFS);
461 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
462 sc->opencount = 1;
463 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
464 sc->opencount = 0;
465 }
466 return (0);
467 }
468
469 static void
g_md_start(struct bio * bp)470 g_md_start(struct bio *bp)
471 {
472 struct md_s *sc;
473
474 sc = bp->bio_to->geom->softc;
475 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
476 devstat_start_transaction_bio(sc->devstat, bp);
477 }
478 mtx_lock(&sc->queue_mtx);
479 bioq_disksort(&sc->bio_queue, bp);
480 wakeup(sc);
481 mtx_unlock(&sc->queue_mtx);
482 }
483
484 #define MD_MALLOC_MOVE_ZERO 1
485 #define MD_MALLOC_MOVE_FILL 2
486 #define MD_MALLOC_MOVE_READ 3
487 #define MD_MALLOC_MOVE_WRITE 4
488 #define MD_MALLOC_MOVE_CMP 5
489
490 static int
md_malloc_move_ma(vm_page_t ** mp,int * ma_offs,unsigned sectorsize,void * ptr,u_char fill,int op)491 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize,
492 void *ptr, u_char fill, int op)
493 {
494 struct sf_buf *sf;
495 vm_page_t m, *mp1;
496 char *p, first;
497 off_t *uc;
498 unsigned n;
499 int error, i, ma_offs1, sz, first_read;
500
501 m = NULL;
502 error = 0;
503 sf = NULL;
504 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */
505 first = 0;
506 first_read = 0;
507 uc = ptr;
508 mp1 = *mp;
509 ma_offs1 = *ma_offs;
510 /* } */
511 sched_pin();
512 for (n = sectorsize; n != 0; n -= sz) {
513 sz = imin(PAGE_SIZE - *ma_offs, n);
514 if (m != **mp) {
515 if (sf != NULL)
516 sf_buf_free(sf);
517 m = **mp;
518 sf = sf_buf_alloc(m, SFB_CPUPRIVATE |
519 (md_malloc_wait ? 0 : SFB_NOWAIT));
520 if (sf == NULL) {
521 error = ENOMEM;
522 break;
523 }
524 }
525 p = (char *)sf_buf_kva(sf) + *ma_offs;
526 switch (op) {
527 case MD_MALLOC_MOVE_ZERO:
528 bzero(p, sz);
529 break;
530 case MD_MALLOC_MOVE_FILL:
531 memset(p, fill, sz);
532 break;
533 case MD_MALLOC_MOVE_READ:
534 bcopy(ptr, p, sz);
535 cpu_flush_dcache(p, sz);
536 break;
537 case MD_MALLOC_MOVE_WRITE:
538 bcopy(p, ptr, sz);
539 break;
540 case MD_MALLOC_MOVE_CMP:
541 for (i = 0; i < sz; i++, p++) {
542 if (!first_read) {
543 *uc = (u_char)*p;
544 first = *p;
545 first_read = 1;
546 } else if (*p != first) {
547 error = EDOOFUS;
548 break;
549 }
550 }
551 break;
552 default:
553 KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op));
554 break;
555 }
556 if (error != 0)
557 break;
558 *ma_offs += sz;
559 *ma_offs %= PAGE_SIZE;
560 if (*ma_offs == 0)
561 (*mp)++;
562 ptr = (char *)ptr + sz;
563 }
564
565 if (sf != NULL)
566 sf_buf_free(sf);
567 sched_unpin();
568 if (op == MD_MALLOC_MOVE_CMP && error != 0) {
569 *mp = mp1;
570 *ma_offs = ma_offs1;
571 }
572 return (error);
573 }
574
575 static int
md_malloc_move_vlist(bus_dma_segment_t ** pvlist,int * pma_offs,unsigned len,void * ptr,u_char fill,int op)576 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs,
577 unsigned len, void *ptr, u_char fill, int op)
578 {
579 bus_dma_segment_t *vlist;
580 uint8_t *p, *end, first;
581 off_t *uc;
582 int ma_offs, seg_len;
583
584 vlist = *pvlist;
585 ma_offs = *pma_offs;
586 uc = ptr;
587
588 for (; len != 0; len -= seg_len) {
589 seg_len = imin(vlist->ds_len - ma_offs, len);
590 p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs;
591 switch (op) {
592 case MD_MALLOC_MOVE_ZERO:
593 bzero(p, seg_len);
594 break;
595 case MD_MALLOC_MOVE_FILL:
596 memset(p, fill, seg_len);
597 break;
598 case MD_MALLOC_MOVE_READ:
599 bcopy(ptr, p, seg_len);
600 cpu_flush_dcache(p, seg_len);
601 break;
602 case MD_MALLOC_MOVE_WRITE:
603 bcopy(p, ptr, seg_len);
604 break;
605 case MD_MALLOC_MOVE_CMP:
606 end = p + seg_len;
607 first = *uc = *p;
608 /* Confirm all following bytes match the first */
609 while (++p < end) {
610 if (*p != first)
611 return (EDOOFUS);
612 }
613 break;
614 default:
615 KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op));
616 break;
617 }
618
619 ma_offs += seg_len;
620 if (ma_offs == vlist->ds_len) {
621 ma_offs = 0;
622 vlist++;
623 }
624 ptr = (uint8_t *)ptr + seg_len;
625 }
626 *pvlist = vlist;
627 *pma_offs = ma_offs;
628
629 return (0);
630 }
631
632 static int
mdstart_malloc(struct md_s * sc,struct bio * bp)633 mdstart_malloc(struct md_s *sc, struct bio *bp)
634 {
635 u_char *dst;
636 vm_page_t *m;
637 bus_dma_segment_t *vlist;
638 int i, error, error1, ma_offs, notmapped;
639 off_t secno, nsec, uc;
640 uintptr_t sp, osp;
641
642 switch (bp->bio_cmd) {
643 case BIO_READ:
644 case BIO_WRITE:
645 case BIO_DELETE:
646 break;
647 case BIO_FLUSH:
648 return (0);
649 default:
650 return (EOPNOTSUPP);
651 }
652
653 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0;
654 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
655 (bus_dma_segment_t *)bp->bio_data : NULL;
656 if (notmapped) {
657 m = bp->bio_ma;
658 ma_offs = bp->bio_ma_offset;
659 dst = NULL;
660 KASSERT(vlist == NULL, ("vlists cannot be unmapped"));
661 } else if (vlist != NULL) {
662 ma_offs = bp->bio_ma_offset;
663 dst = NULL;
664 } else {
665 dst = bp->bio_data;
666 }
667
668 nsec = bp->bio_length / sc->sectorsize;
669 secno = bp->bio_offset / sc->sectorsize;
670 error = 0;
671 while (nsec--) {
672 osp = s_read(sc->indir, secno);
673 if (bp->bio_cmd == BIO_DELETE) {
674 if (osp != 0)
675 error = s_write(sc->indir, secno, 0);
676 } else if (bp->bio_cmd == BIO_READ) {
677 if (osp == 0) {
678 if (notmapped) {
679 error = md_malloc_move_ma(&m, &ma_offs,
680 sc->sectorsize, NULL, 0,
681 MD_MALLOC_MOVE_ZERO);
682 } else if (vlist != NULL) {
683 error = md_malloc_move_vlist(&vlist,
684 &ma_offs, sc->sectorsize, NULL, 0,
685 MD_MALLOC_MOVE_ZERO);
686 } else
687 bzero(dst, sc->sectorsize);
688 } else if (osp <= 255) {
689 if (notmapped) {
690 error = md_malloc_move_ma(&m, &ma_offs,
691 sc->sectorsize, NULL, osp,
692 MD_MALLOC_MOVE_FILL);
693 } else if (vlist != NULL) {
694 error = md_malloc_move_vlist(&vlist,
695 &ma_offs, sc->sectorsize, NULL, osp,
696 MD_MALLOC_MOVE_FILL);
697 } else
698 memset(dst, osp, sc->sectorsize);
699 } else {
700 if (notmapped) {
701 error = md_malloc_move_ma(&m, &ma_offs,
702 sc->sectorsize, (void *)osp, 0,
703 MD_MALLOC_MOVE_READ);
704 } else if (vlist != NULL) {
705 error = md_malloc_move_vlist(&vlist,
706 &ma_offs, sc->sectorsize,
707 (void *)osp, 0,
708 MD_MALLOC_MOVE_READ);
709 } else {
710 bcopy((void *)osp, dst, sc->sectorsize);
711 cpu_flush_dcache(dst, sc->sectorsize);
712 }
713 }
714 osp = 0;
715 } else if (bp->bio_cmd == BIO_WRITE) {
716 if (sc->flags & MD_COMPRESS) {
717 if (notmapped) {
718 error1 = md_malloc_move_ma(&m, &ma_offs,
719 sc->sectorsize, &uc, 0,
720 MD_MALLOC_MOVE_CMP);
721 i = error1 == 0 ? sc->sectorsize : 0;
722 } else if (vlist != NULL) {
723 error1 = md_malloc_move_vlist(&vlist,
724 &ma_offs, sc->sectorsize, &uc, 0,
725 MD_MALLOC_MOVE_CMP);
726 i = error1 == 0 ? sc->sectorsize : 0;
727 } else {
728 uc = dst[0];
729 for (i = 1; i < sc->sectorsize; i++) {
730 if (dst[i] != uc)
731 break;
732 }
733 }
734 } else {
735 i = 0;
736 uc = 0;
737 }
738 if (i == sc->sectorsize) {
739 if (osp != uc)
740 error = s_write(sc->indir, secno, uc);
741 } else {
742 if (osp <= 255) {
743 sp = (uintptr_t)uma_zalloc(sc->uma,
744 md_malloc_wait ? M_WAITOK :
745 M_NOWAIT);
746 if (sp == 0) {
747 error = ENOSPC;
748 break;
749 }
750 if (notmapped) {
751 error = md_malloc_move_ma(&m,
752 &ma_offs, sc->sectorsize,
753 (void *)sp, 0,
754 MD_MALLOC_MOVE_WRITE);
755 } else if (vlist != NULL) {
756 error = md_malloc_move_vlist(
757 &vlist, &ma_offs,
758 sc->sectorsize, (void *)sp,
759 0, MD_MALLOC_MOVE_WRITE);
760 } else {
761 bcopy(dst, (void *)sp,
762 sc->sectorsize);
763 }
764 error = s_write(sc->indir, secno, sp);
765 } else {
766 if (notmapped) {
767 error = md_malloc_move_ma(&m,
768 &ma_offs, sc->sectorsize,
769 (void *)osp, 0,
770 MD_MALLOC_MOVE_WRITE);
771 } else if (vlist != NULL) {
772 error = md_malloc_move_vlist(
773 &vlist, &ma_offs,
774 sc->sectorsize, (void *)osp,
775 0, MD_MALLOC_MOVE_WRITE);
776 } else {
777 bcopy(dst, (void *)osp,
778 sc->sectorsize);
779 }
780 osp = 0;
781 }
782 }
783 } else {
784 error = EOPNOTSUPP;
785 }
786 if (osp > 255)
787 uma_zfree(sc->uma, (void*)osp);
788 if (error != 0)
789 break;
790 secno++;
791 if (!notmapped && vlist == NULL)
792 dst += sc->sectorsize;
793 }
794 bp->bio_resid = 0;
795 return (error);
796 }
797
798 static void
mdcopyto_vlist(void * src,bus_dma_segment_t * vlist,off_t offset,off_t len)799 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len)
800 {
801 off_t seg_len;
802
803 while (offset >= vlist->ds_len) {
804 offset -= vlist->ds_len;
805 vlist++;
806 }
807
808 while (len != 0) {
809 seg_len = omin(len, vlist->ds_len - offset);
810 bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset),
811 seg_len);
812 offset = 0;
813 src = (uint8_t *)src + seg_len;
814 len -= seg_len;
815 vlist++;
816 }
817 }
818
819 static void
mdcopyfrom_vlist(bus_dma_segment_t * vlist,off_t offset,void * dst,off_t len)820 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len)
821 {
822 off_t seg_len;
823
824 while (offset >= vlist->ds_len) {
825 offset -= vlist->ds_len;
826 vlist++;
827 }
828
829 while (len != 0) {
830 seg_len = omin(len, vlist->ds_len - offset);
831 bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst,
832 seg_len);
833 offset = 0;
834 dst = (uint8_t *)dst + seg_len;
835 len -= seg_len;
836 vlist++;
837 }
838 }
839
840 static int
mdstart_preload(struct md_s * sc,struct bio * bp)841 mdstart_preload(struct md_s *sc, struct bio *bp)
842 {
843 uint8_t *p;
844
845 p = sc->pl_ptr + bp->bio_offset;
846 switch (bp->bio_cmd) {
847 case BIO_READ:
848 if ((bp->bio_flags & BIO_VLIST) != 0) {
849 mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data,
850 bp->bio_ma_offset, bp->bio_length);
851 } else {
852 bcopy(p, bp->bio_data, bp->bio_length);
853 }
854 cpu_flush_dcache(bp->bio_data, bp->bio_length);
855 break;
856 case BIO_WRITE:
857 if ((bp->bio_flags & BIO_VLIST) != 0) {
858 mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data,
859 bp->bio_ma_offset, p, bp->bio_length);
860 } else {
861 bcopy(bp->bio_data, p, bp->bio_length);
862 }
863 break;
864 }
865 bp->bio_resid = 0;
866 return (0);
867 }
868
869 static int
mdstart_vnode(struct md_s * sc,struct bio * bp)870 mdstart_vnode(struct md_s *sc, struct bio *bp)
871 {
872 int error;
873 struct uio auio;
874 struct iovec aiov;
875 struct iovec *piov;
876 struct mount *mp;
877 struct vnode *vp;
878 struct buf *pb;
879 bus_dma_segment_t *vlist;
880 struct thread *td;
881 off_t iolen, iostart, len, zerosize;
882 int ma_offs, npages;
883
884 switch (bp->bio_cmd) {
885 case BIO_READ:
886 auio.uio_rw = UIO_READ;
887 break;
888 case BIO_WRITE:
889 case BIO_DELETE:
890 auio.uio_rw = UIO_WRITE;
891 break;
892 case BIO_FLUSH:
893 break;
894 default:
895 return (EOPNOTSUPP);
896 }
897
898 td = curthread;
899 vp = sc->vnode;
900 pb = NULL;
901 piov = NULL;
902 ma_offs = bp->bio_ma_offset;
903 len = bp->bio_length;
904
905 /*
906 * VNODE I/O
907 *
908 * If an error occurs, we set BIO_ERROR but we do not set
909 * B_INVAL because (for a write anyway), the buffer is
910 * still valid.
911 */
912
913 if (bp->bio_cmd == BIO_FLUSH) {
914 (void) vn_start_write(vp, &mp, V_WAIT);
915 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
916 error = VOP_FSYNC(vp, MNT_WAIT, td);
917 VOP_UNLOCK(vp);
918 vn_finished_write(mp);
919 return (error);
920 }
921
922 auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
923 auio.uio_resid = bp->bio_length;
924 auio.uio_segflg = UIO_SYSSPACE;
925 auio.uio_td = td;
926
927 if (bp->bio_cmd == BIO_DELETE) {
928 /*
929 * Emulate BIO_DELETE by writing zeros.
930 */
931 zerosize = ZERO_REGION_SIZE -
932 (ZERO_REGION_SIZE % sc->sectorsize);
933 auio.uio_iovcnt = howmany(bp->bio_length, zerosize);
934 piov = malloc(sizeof(*piov) * auio.uio_iovcnt, M_MD, M_WAITOK);
935 auio.uio_iov = piov;
936 while (len > 0) {
937 piov->iov_base = __DECONST(void *, zero_region);
938 piov->iov_len = len;
939 if (len > zerosize)
940 piov->iov_len = zerosize;
941 len -= piov->iov_len;
942 piov++;
943 }
944 piov = auio.uio_iov;
945 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
946 piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK);
947 auio.uio_iov = piov;
948 vlist = (bus_dma_segment_t *)bp->bio_data;
949 while (len > 0) {
950 piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr +
951 ma_offs);
952 piov->iov_len = vlist->ds_len - ma_offs;
953 if (piov->iov_len > len)
954 piov->iov_len = len;
955 len -= piov->iov_len;
956 ma_offs = 0;
957 vlist++;
958 piov++;
959 }
960 auio.uio_iovcnt = piov - auio.uio_iov;
961 piov = auio.uio_iov;
962 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
963 pb = uma_zalloc(md_pbuf_zone, M_WAITOK);
964 MPASS((pb->b_flags & B_MAXPHYS) != 0);
965 bp->bio_resid = len;
966 unmapped_step:
967 npages = atop(min(maxphys, round_page(len + (ma_offs &
968 PAGE_MASK))));
969 iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len);
970 KASSERT(iolen > 0, ("zero iolen"));
971 pmap_qenter((vm_offset_t)pb->b_data,
972 &bp->bio_ma[atop(ma_offs)], npages);
973 aiov.iov_base = (void *)((vm_offset_t)pb->b_data +
974 (ma_offs & PAGE_MASK));
975 aiov.iov_len = iolen;
976 auio.uio_iov = &aiov;
977 auio.uio_iovcnt = 1;
978 auio.uio_resid = iolen;
979 } else {
980 aiov.iov_base = bp->bio_data;
981 aiov.iov_len = bp->bio_length;
982 auio.uio_iov = &aiov;
983 auio.uio_iovcnt = 1;
984 }
985 iostart = auio.uio_offset;
986 if (auio.uio_rw == UIO_READ) {
987 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
988 error = VOP_READ(vp, &auio, 0, sc->cred);
989 VOP_UNLOCK(vp);
990 } else {
991 (void) vn_start_write(vp, &mp, V_WAIT);
992 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
993 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
994 sc->cred);
995 VOP_UNLOCK(vp);
996 vn_finished_write(mp);
997 if (error == 0)
998 sc->flags &= ~MD_VERIFY;
999 }
1000
1001 /* When MD_CACHE is set, try to avoid double-caching the data. */
1002 if (error == 0 && (sc->flags & MD_CACHE) == 0)
1003 VOP_ADVISE(vp, iostart, auio.uio_offset - 1,
1004 POSIX_FADV_DONTNEED);
1005
1006 if (pb != NULL) {
1007 pmap_qremove((vm_offset_t)pb->b_data, npages);
1008 if (error == 0) {
1009 len -= iolen;
1010 bp->bio_resid -= iolen;
1011 ma_offs += iolen;
1012 if (len > 0)
1013 goto unmapped_step;
1014 }
1015 uma_zfree(md_pbuf_zone, pb);
1016 } else {
1017 bp->bio_resid = auio.uio_resid;
1018 }
1019
1020 free(piov, M_MD);
1021 return (error);
1022 }
1023
1024 static int
mdstart_swap(struct md_s * sc,struct bio * bp)1025 mdstart_swap(struct md_s *sc, struct bio *bp)
1026 {
1027 vm_page_t m;
1028 u_char *p;
1029 vm_pindex_t i, lastp;
1030 bus_dma_segment_t *vlist;
1031 int rv, ma_offs, offs, len, lastend;
1032
1033 switch (bp->bio_cmd) {
1034 case BIO_READ:
1035 case BIO_WRITE:
1036 case BIO_DELETE:
1037 break;
1038 case BIO_FLUSH:
1039 return (0);
1040 default:
1041 return (EOPNOTSUPP);
1042 }
1043
1044 p = bp->bio_data;
1045 ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ?
1046 bp->bio_ma_offset : 0;
1047 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
1048 (bus_dma_segment_t *)bp->bio_data : NULL;
1049
1050 /*
1051 * offs is the offset at which to start operating on the
1052 * next (ie, first) page. lastp is the last page on
1053 * which we're going to operate. lastend is the ending
1054 * position within that last page (ie, PAGE_SIZE if
1055 * we're operating on complete aligned pages).
1056 */
1057 offs = bp->bio_offset % PAGE_SIZE;
1058 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
1059 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
1060
1061 rv = VM_PAGER_OK;
1062 vm_object_pip_add(sc->object, 1);
1063 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
1064 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
1065 m = vm_page_grab_unlocked(sc->object, i, VM_ALLOC_SYSTEM);
1066 if (bp->bio_cmd == BIO_READ) {
1067 if (vm_page_all_valid(m))
1068 rv = VM_PAGER_OK;
1069 else
1070 rv = vm_pager_get_pages(sc->object, &m, 1,
1071 NULL, NULL);
1072 if (rv == VM_PAGER_ERROR) {
1073 VM_OBJECT_WLOCK(sc->object);
1074 vm_page_free(m);
1075 VM_OBJECT_WUNLOCK(sc->object);
1076 break;
1077 } else if (rv == VM_PAGER_FAIL) {
1078 /*
1079 * Pager does not have the page. Zero
1080 * the allocated page, and mark it as
1081 * valid. Do not set dirty, the page
1082 * can be recreated if thrown out.
1083 */
1084 pmap_zero_page(m);
1085 vm_page_valid(m);
1086 }
1087 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1088 pmap_copy_pages(&m, offs, bp->bio_ma,
1089 ma_offs, len);
1090 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1091 physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs,
1092 vlist, ma_offs, len);
1093 cpu_flush_dcache(p, len);
1094 } else {
1095 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1096 cpu_flush_dcache(p, len);
1097 }
1098 } else if (bp->bio_cmd == BIO_WRITE) {
1099 if (len == PAGE_SIZE || vm_page_all_valid(m))
1100 rv = VM_PAGER_OK;
1101 else
1102 rv = vm_pager_get_pages(sc->object, &m, 1,
1103 NULL, NULL);
1104 if (rv == VM_PAGER_ERROR) {
1105 VM_OBJECT_WLOCK(sc->object);
1106 vm_page_free(m);
1107 VM_OBJECT_WUNLOCK(sc->object);
1108 break;
1109 } else if (rv == VM_PAGER_FAIL)
1110 pmap_zero_page(m);
1111
1112 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1113 pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1114 offs, len);
1115 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1116 physcopyin_vlist(vlist, ma_offs,
1117 VM_PAGE_TO_PHYS(m) + offs, len);
1118 } else {
1119 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1120 }
1121
1122 vm_page_valid(m);
1123 vm_page_set_dirty(m);
1124 } else if (bp->bio_cmd == BIO_DELETE) {
1125 if (len == PAGE_SIZE || vm_page_all_valid(m))
1126 rv = VM_PAGER_OK;
1127 else
1128 rv = vm_pager_get_pages(sc->object, &m, 1,
1129 NULL, NULL);
1130 VM_OBJECT_WLOCK(sc->object);
1131 if (rv == VM_PAGER_ERROR) {
1132 vm_page_free(m);
1133 VM_OBJECT_WUNLOCK(sc->object);
1134 break;
1135 } else if (rv == VM_PAGER_FAIL) {
1136 vm_page_free(m);
1137 m = NULL;
1138 } else {
1139 /* Page is valid. */
1140 if (len != PAGE_SIZE) {
1141 pmap_zero_page_area(m, offs, len);
1142 vm_page_set_dirty(m);
1143 } else {
1144 vm_pager_page_unswapped(m);
1145 vm_page_free(m);
1146 m = NULL;
1147 }
1148 }
1149 VM_OBJECT_WUNLOCK(sc->object);
1150 }
1151 if (m != NULL) {
1152 /*
1153 * The page may be deactivated prior to setting
1154 * PGA_REFERENCED, but in this case it will be
1155 * reactivated by the page daemon.
1156 */
1157 if (vm_page_active(m))
1158 vm_page_reference(m);
1159 else
1160 vm_page_activate(m);
1161 vm_page_xunbusy(m);
1162 }
1163
1164 /* Actions on further pages start at offset 0 */
1165 p += PAGE_SIZE - offs;
1166 offs = 0;
1167 ma_offs += len;
1168 }
1169 vm_object_pip_wakeup(sc->object);
1170 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
1171 }
1172
1173 static int
mdstart_null(struct md_s * sc,struct bio * bp)1174 mdstart_null(struct md_s *sc, struct bio *bp)
1175 {
1176
1177 switch (bp->bio_cmd) {
1178 case BIO_READ:
1179 bzero(bp->bio_data, bp->bio_length);
1180 cpu_flush_dcache(bp->bio_data, bp->bio_length);
1181 break;
1182 case BIO_WRITE:
1183 break;
1184 }
1185 bp->bio_resid = 0;
1186 return (0);
1187 }
1188
1189 static void
md_handleattr(struct md_s * sc,struct bio * bp)1190 md_handleattr(struct md_s *sc, struct bio *bp)
1191 {
1192 if (sc->fwsectors && sc->fwheads &&
1193 (g_handleattr_int(bp, "GEOM::fwsectors", sc->fwsectors) != 0 ||
1194 g_handleattr_int(bp, "GEOM::fwheads", sc->fwheads) != 0))
1195 return;
1196 if (g_handleattr_int(bp, "GEOM::candelete", 1) != 0)
1197 return;
1198 if (sc->ident[0] != '\0' &&
1199 g_handleattr_str(bp, "GEOM::ident", sc->ident) != 0)
1200 return;
1201 if (g_handleattr_int(bp, "MNT::verified", (sc->flags & MD_VERIFY) != 0))
1202 return;
1203 g_io_deliver(bp, EOPNOTSUPP);
1204 }
1205
1206 static void
md_kthread(void * arg)1207 md_kthread(void *arg)
1208 {
1209 struct md_s *sc;
1210 struct bio *bp;
1211 int error;
1212
1213 sc = arg;
1214 thread_lock(curthread);
1215 sched_prio(curthread, PRIBIO);
1216 thread_unlock(curthread);
1217 if (sc->type == MD_VNODE)
1218 curthread->td_pflags |= TDP_NORUNNINGBUF;
1219
1220 for (;;) {
1221 mtx_lock(&sc->queue_mtx);
1222 if (sc->flags & MD_SHUTDOWN) {
1223 sc->flags |= MD_EXITING;
1224 mtx_unlock(&sc->queue_mtx);
1225 kproc_exit(0);
1226 }
1227 bp = bioq_takefirst(&sc->bio_queue);
1228 if (!bp) {
1229 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1230 continue;
1231 }
1232 mtx_unlock(&sc->queue_mtx);
1233 if (bp->bio_cmd == BIO_GETATTR) {
1234 md_handleattr(sc, bp);
1235 } else {
1236 error = sc->start(sc, bp);
1237 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
1238 /*
1239 * Devstat uses (bio_bcount, bio_resid) for
1240 * determining the length of the completed part
1241 * of the i/o. g_io_deliver() will translate
1242 * from bio_completed to that, but it also
1243 * destroys the bio so we must do our own
1244 * translation.
1245 */
1246 bp->bio_bcount = bp->bio_length;
1247 devstat_end_transaction_bio(sc->devstat, bp);
1248 }
1249 bp->bio_completed = bp->bio_length - bp->bio_resid;
1250 g_io_deliver(bp, error);
1251 }
1252 }
1253 }
1254
1255 static struct md_s *
mdfind(int unit)1256 mdfind(int unit)
1257 {
1258 struct md_s *sc;
1259
1260 LIST_FOREACH(sc, &md_softc_list, list) {
1261 if (sc->unit == unit)
1262 break;
1263 }
1264 return (sc);
1265 }
1266
1267 static struct md_s *
mdnew(int unit,int * errp,enum md_types type)1268 mdnew(int unit, int *errp, enum md_types type)
1269 {
1270 struct md_s *sc;
1271 int error;
1272
1273 *errp = 0;
1274 if (unit == -1)
1275 unit = alloc_unr(md_uh);
1276 else
1277 unit = alloc_unr_specific(md_uh, unit);
1278
1279 if (unit == -1) {
1280 *errp = EBUSY;
1281 return (NULL);
1282 }
1283
1284 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
1285 sc->type = type;
1286 bioq_init(&sc->bio_queue);
1287 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1288 sc->unit = unit;
1289 sprintf(sc->name, "md%d", unit);
1290 LIST_INSERT_HEAD(&md_softc_list, sc, list);
1291 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1292 if (error == 0)
1293 return (sc);
1294 LIST_REMOVE(sc, list);
1295 mtx_destroy(&sc->queue_mtx);
1296 free_unr(md_uh, sc->unit);
1297 free(sc, M_MD);
1298 *errp = error;
1299 return (NULL);
1300 }
1301
1302 static void
mdinit(struct md_s * sc)1303 mdinit(struct md_s *sc)
1304 {
1305 struct g_geom *gp;
1306 struct g_provider *pp;
1307
1308 g_topology_lock();
1309 gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1310 gp->softc = sc;
1311 pp = g_new_providerf(gp, "md%d", sc->unit);
1312 devstat_remove_entry(pp->stat);
1313 pp->stat = NULL;
1314 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1315 pp->mediasize = sc->mediasize;
1316 pp->sectorsize = sc->sectorsize;
1317 switch (sc->type) {
1318 case MD_MALLOC:
1319 case MD_VNODE:
1320 case MD_SWAP:
1321 pp->flags |= G_PF_ACCEPT_UNMAPPED;
1322 break;
1323 case MD_PRELOAD:
1324 case MD_NULL:
1325 break;
1326 }
1327 sc->gp = gp;
1328 sc->pp = pp;
1329 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1330 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
1331 sc->devstat->id = pp;
1332 g_error_provider(pp, 0);
1333 g_topology_unlock();
1334 }
1335
1336 static int
mdcreate_malloc(struct md_s * sc,struct md_req * mdr)1337 mdcreate_malloc(struct md_s *sc, struct md_req *mdr)
1338 {
1339 uintptr_t sp;
1340 int error;
1341 off_t u;
1342
1343 error = 0;
1344 if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1345 return (EINVAL);
1346 if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize))
1347 return (EINVAL);
1348 /* Compression doesn't make sense if we have reserved space */
1349 if (mdr->md_options & MD_RESERVE)
1350 mdr->md_options &= ~MD_COMPRESS;
1351 if (mdr->md_fwsectors != 0)
1352 sc->fwsectors = mdr->md_fwsectors;
1353 if (mdr->md_fwheads != 0)
1354 sc->fwheads = mdr->md_fwheads;
1355 sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE);
1356 sc->indir = dimension(sc->mediasize / sc->sectorsize);
1357 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
1358 0x1ff, 0);
1359 if (mdr->md_options & MD_RESERVE) {
1360 off_t nsectors;
1361
1362 nsectors = sc->mediasize / sc->sectorsize;
1363 for (u = 0; u < nsectors; u++) {
1364 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ?
1365 M_WAITOK : M_NOWAIT) | M_ZERO);
1366 if (sp != 0)
1367 error = s_write(sc->indir, u, sp);
1368 else
1369 error = ENOMEM;
1370 if (error != 0)
1371 break;
1372 }
1373 }
1374 return (error);
1375 }
1376
1377 static int
mdsetcred(struct md_s * sc,struct ucred * cred)1378 mdsetcred(struct md_s *sc, struct ucred *cred)
1379 {
1380 char *tmpbuf;
1381 int error = 0;
1382
1383 /*
1384 * Set credits in our softc
1385 */
1386
1387 if (sc->cred)
1388 crfree(sc->cred);
1389 sc->cred = crhold(cred);
1390
1391 /*
1392 * Horrible kludge to establish credentials for NFS XXX.
1393 */
1394
1395 if (sc->vnode) {
1396 struct uio auio;
1397 struct iovec aiov;
1398
1399 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1400 bzero(&auio, sizeof(auio));
1401
1402 aiov.iov_base = tmpbuf;
1403 aiov.iov_len = sc->sectorsize;
1404 auio.uio_iov = &aiov;
1405 auio.uio_iovcnt = 1;
1406 auio.uio_offset = 0;
1407 auio.uio_rw = UIO_READ;
1408 auio.uio_segflg = UIO_SYSSPACE;
1409 auio.uio_resid = aiov.iov_len;
1410 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1411 error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
1412 VOP_UNLOCK(sc->vnode);
1413 free(tmpbuf, M_TEMP);
1414 }
1415 return (error);
1416 }
1417
1418 static int
mdcreate_vnode(struct md_s * sc,struct md_req * mdr,struct thread * td)1419 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1420 {
1421 struct vattr vattr;
1422 struct nameidata nd;
1423 char *fname;
1424 int error, flags;
1425
1426 fname = mdr->md_file;
1427 if (mdr->md_file_seg == UIO_USERSPACE) {
1428 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
1429 if (error != 0)
1430 return (error);
1431 } else if (mdr->md_file_seg == UIO_SYSSPACE)
1432 strlcpy(sc->file, fname, sizeof(sc->file));
1433 else
1434 return (EDOOFUS);
1435
1436 /*
1437 * If the user specified that this is a read only device, don't
1438 * set the FWRITE mask before trying to open the backing store.
1439 */
1440 flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \
1441 | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0);
1442 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td);
1443 error = vn_open(&nd, &flags, 0, NULL);
1444 if (error != 0)
1445 return (error);
1446 NDFREE(&nd, NDF_ONLY_PNBUF);
1447 if (nd.ni_vp->v_type != VREG) {
1448 error = EINVAL;
1449 goto bad;
1450 }
1451 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1452 if (error != 0)
1453 goto bad;
1454 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1455 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1456 if (VN_IS_DOOMED(nd.ni_vp)) {
1457 /* Forced unmount. */
1458 error = EBADF;
1459 goto bad;
1460 }
1461 }
1462 nd.ni_vp->v_vflag |= VV_MD;
1463 VOP_UNLOCK(nd.ni_vp);
1464
1465 if (mdr->md_fwsectors != 0)
1466 sc->fwsectors = mdr->md_fwsectors;
1467 if (mdr->md_fwheads != 0)
1468 sc->fwheads = mdr->md_fwheads;
1469 snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1470 (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1471 sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE |
1472 MD_VERIFY);
1473 if (!(flags & FWRITE))
1474 sc->flags |= MD_READONLY;
1475 sc->vnode = nd.ni_vp;
1476
1477 error = mdsetcred(sc, td->td_ucred);
1478 if (error != 0) {
1479 sc->vnode = NULL;
1480 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1481 nd.ni_vp->v_vflag &= ~VV_MD;
1482 goto bad;
1483 }
1484 return (0);
1485 bad:
1486 VOP_UNLOCK(nd.ni_vp);
1487 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1488 return (error);
1489 }
1490
1491 static void
g_md_providergone(struct g_provider * pp)1492 g_md_providergone(struct g_provider *pp)
1493 {
1494 struct md_s *sc = pp->geom->softc;
1495
1496 mtx_lock(&sc->queue_mtx);
1497 sc->flags |= MD_PROVIDERGONE;
1498 wakeup(&sc->flags);
1499 mtx_unlock(&sc->queue_mtx);
1500 }
1501
1502 static int
mddestroy(struct md_s * sc,struct thread * td)1503 mddestroy(struct md_s *sc, struct thread *td)
1504 {
1505
1506 if (sc->gp) {
1507 g_topology_lock();
1508 g_wither_geom(sc->gp, ENXIO);
1509 g_topology_unlock();
1510
1511 mtx_lock(&sc->queue_mtx);
1512 while (!(sc->flags & MD_PROVIDERGONE))
1513 msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0);
1514 mtx_unlock(&sc->queue_mtx);
1515 }
1516 if (sc->devstat) {
1517 devstat_remove_entry(sc->devstat);
1518 sc->devstat = NULL;
1519 }
1520 mtx_lock(&sc->queue_mtx);
1521 sc->flags |= MD_SHUTDOWN;
1522 wakeup(sc);
1523 while (!(sc->flags & MD_EXITING))
1524 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1525 mtx_unlock(&sc->queue_mtx);
1526 mtx_destroy(&sc->queue_mtx);
1527 if (sc->vnode != NULL) {
1528 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1529 sc->vnode->v_vflag &= ~VV_MD;
1530 VOP_UNLOCK(sc->vnode);
1531 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1532 FREAD : (FREAD|FWRITE), sc->cred, td);
1533 }
1534 if (sc->cred != NULL)
1535 crfree(sc->cred);
1536 if (sc->object != NULL)
1537 vm_object_deallocate(sc->object);
1538 if (sc->indir)
1539 destroy_indir(sc, sc->indir);
1540 if (sc->uma)
1541 uma_zdestroy(sc->uma);
1542
1543 LIST_REMOVE(sc, list);
1544 free_unr(md_uh, sc->unit);
1545 free(sc, M_MD);
1546 return (0);
1547 }
1548
1549 static int
mdresize(struct md_s * sc,struct md_req * mdr)1550 mdresize(struct md_s *sc, struct md_req *mdr)
1551 {
1552 int error, res;
1553 vm_pindex_t oldpages, newpages;
1554
1555 switch (sc->type) {
1556 case MD_VNODE:
1557 case MD_NULL:
1558 break;
1559 case MD_SWAP:
1560 if (mdr->md_mediasize <= 0 ||
1561 (mdr->md_mediasize % PAGE_SIZE) != 0)
1562 return (EDOM);
1563 oldpages = OFF_TO_IDX(sc->mediasize);
1564 newpages = OFF_TO_IDX(mdr->md_mediasize);
1565 if (newpages < oldpages) {
1566 VM_OBJECT_WLOCK(sc->object);
1567 vm_object_page_remove(sc->object, newpages, 0, 0);
1568 swap_release_by_cred(IDX_TO_OFF(oldpages -
1569 newpages), sc->cred);
1570 sc->object->charge = IDX_TO_OFF(newpages);
1571 sc->object->size = newpages;
1572 VM_OBJECT_WUNLOCK(sc->object);
1573 } else if (newpages > oldpages) {
1574 res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1575 oldpages), sc->cred);
1576 if (!res)
1577 return (ENOMEM);
1578 if ((mdr->md_options & MD_RESERVE) ||
1579 (sc->flags & MD_RESERVE)) {
1580 error = swap_pager_reserve(sc->object,
1581 oldpages, newpages - oldpages);
1582 if (error < 0) {
1583 swap_release_by_cred(
1584 IDX_TO_OFF(newpages - oldpages),
1585 sc->cred);
1586 return (EDOM);
1587 }
1588 }
1589 VM_OBJECT_WLOCK(sc->object);
1590 sc->object->charge = IDX_TO_OFF(newpages);
1591 sc->object->size = newpages;
1592 VM_OBJECT_WUNLOCK(sc->object);
1593 }
1594 break;
1595 default:
1596 return (EOPNOTSUPP);
1597 }
1598
1599 sc->mediasize = mdr->md_mediasize;
1600
1601 g_topology_lock();
1602 g_resize_provider(sc->pp, sc->mediasize);
1603 g_topology_unlock();
1604 return (0);
1605 }
1606
1607 static int
mdcreate_swap(struct md_s * sc,struct md_req * mdr,struct thread * td)1608 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1609 {
1610 vm_ooffset_t npage;
1611 int error;
1612
1613 /*
1614 * Range check. Disallow negative sizes and sizes not being
1615 * multiple of page size.
1616 */
1617 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1618 return (EDOM);
1619
1620 /*
1621 * Allocate an OBJT_SWAP object.
1622 *
1623 * Note the truncation.
1624 */
1625
1626 if ((mdr->md_options & MD_VERIFY) != 0)
1627 return (EINVAL);
1628 npage = mdr->md_mediasize / PAGE_SIZE;
1629 if (mdr->md_fwsectors != 0)
1630 sc->fwsectors = mdr->md_fwsectors;
1631 if (mdr->md_fwheads != 0)
1632 sc->fwheads = mdr->md_fwheads;
1633 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1634 VM_PROT_DEFAULT, 0, td->td_ucred);
1635 if (sc->object == NULL)
1636 return (ENOMEM);
1637 sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1638 if (mdr->md_options & MD_RESERVE) {
1639 if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1640 error = EDOM;
1641 goto finish;
1642 }
1643 }
1644 error = mdsetcred(sc, td->td_ucred);
1645 finish:
1646 if (error != 0) {
1647 vm_object_deallocate(sc->object);
1648 sc->object = NULL;
1649 }
1650 return (error);
1651 }
1652
1653 static int
mdcreate_null(struct md_s * sc,struct md_req * mdr,struct thread * td)1654 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1655 {
1656
1657 /*
1658 * Range check. Disallow negative sizes and sizes not being
1659 * multiple of page size.
1660 */
1661 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1662 return (EDOM);
1663
1664 return (0);
1665 }
1666
1667 static int
kern_mdattach_locked(struct thread * td,struct md_req * mdr)1668 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1669 {
1670 struct md_s *sc;
1671 unsigned sectsize;
1672 int error, i;
1673
1674 sx_assert(&md_sx, SA_XLOCKED);
1675
1676 switch (mdr->md_type) {
1677 case MD_MALLOC:
1678 case MD_PRELOAD:
1679 case MD_VNODE:
1680 case MD_SWAP:
1681 case MD_NULL:
1682 break;
1683 default:
1684 return (EINVAL);
1685 }
1686 if (mdr->md_sectorsize == 0)
1687 sectsize = DEV_BSIZE;
1688 else
1689 sectsize = mdr->md_sectorsize;
1690 if (sectsize > maxphys || mdr->md_mediasize < sectsize)
1691 return (EINVAL);
1692 if (mdr->md_options & MD_AUTOUNIT)
1693 sc = mdnew(-1, &error, mdr->md_type);
1694 else {
1695 if (mdr->md_unit > INT_MAX)
1696 return (EINVAL);
1697 sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1698 }
1699 if (sc == NULL)
1700 return (error);
1701 if (mdr->md_label != NULL)
1702 error = copyinstr(mdr->md_label, sc->label,
1703 sizeof(sc->label), NULL);
1704 if (error != 0)
1705 goto err_after_new;
1706 if (mdr->md_options & MD_AUTOUNIT)
1707 mdr->md_unit = sc->unit;
1708 sc->mediasize = mdr->md_mediasize;
1709 sc->sectorsize = sectsize;
1710 error = EDOOFUS;
1711 switch (sc->type) {
1712 case MD_MALLOC:
1713 sc->start = mdstart_malloc;
1714 error = mdcreate_malloc(sc, mdr);
1715 break;
1716 case MD_PRELOAD:
1717 /*
1718 * We disallow attaching preloaded memory disks via
1719 * ioctl. Preloaded memory disks are automatically
1720 * attached in g_md_init().
1721 */
1722 error = EOPNOTSUPP;
1723 break;
1724 case MD_VNODE:
1725 sc->start = mdstart_vnode;
1726 error = mdcreate_vnode(sc, mdr, td);
1727 break;
1728 case MD_SWAP:
1729 sc->start = mdstart_swap;
1730 error = mdcreate_swap(sc, mdr, td);
1731 break;
1732 case MD_NULL:
1733 sc->start = mdstart_null;
1734 error = mdcreate_null(sc, mdr, td);
1735 break;
1736 }
1737 err_after_new:
1738 if (error != 0) {
1739 mddestroy(sc, td);
1740 return (error);
1741 }
1742
1743 /* Prune off any residual fractional sector */
1744 i = sc->mediasize % sc->sectorsize;
1745 sc->mediasize -= i;
1746
1747 mdinit(sc);
1748 return (0);
1749 }
1750
1751 static int
kern_mdattach(struct thread * td,struct md_req * mdr)1752 kern_mdattach(struct thread *td, struct md_req *mdr)
1753 {
1754 int error;
1755
1756 sx_xlock(&md_sx);
1757 error = kern_mdattach_locked(td, mdr);
1758 sx_xunlock(&md_sx);
1759 return (error);
1760 }
1761
1762 static int
kern_mddetach_locked(struct thread * td,struct md_req * mdr)1763 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1764 {
1765 struct md_s *sc;
1766
1767 sx_assert(&md_sx, SA_XLOCKED);
1768
1769 if (mdr->md_mediasize != 0 ||
1770 (mdr->md_options & ~MD_FORCE) != 0)
1771 return (EINVAL);
1772
1773 sc = mdfind(mdr->md_unit);
1774 if (sc == NULL)
1775 return (ENOENT);
1776 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1777 !(mdr->md_options & MD_FORCE))
1778 return (EBUSY);
1779 return (mddestroy(sc, td));
1780 }
1781
1782 static int
kern_mddetach(struct thread * td,struct md_req * mdr)1783 kern_mddetach(struct thread *td, struct md_req *mdr)
1784 {
1785 int error;
1786
1787 sx_xlock(&md_sx);
1788 error = kern_mddetach_locked(td, mdr);
1789 sx_xunlock(&md_sx);
1790 return (error);
1791 }
1792
1793 static int
kern_mdresize_locked(struct md_req * mdr)1794 kern_mdresize_locked(struct md_req *mdr)
1795 {
1796 struct md_s *sc;
1797
1798 sx_assert(&md_sx, SA_XLOCKED);
1799
1800 if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1801 return (EINVAL);
1802
1803 sc = mdfind(mdr->md_unit);
1804 if (sc == NULL)
1805 return (ENOENT);
1806 if (mdr->md_mediasize < sc->sectorsize)
1807 return (EINVAL);
1808 mdr->md_mediasize -= mdr->md_mediasize % sc->sectorsize;
1809 if (mdr->md_mediasize < sc->mediasize &&
1810 !(sc->flags & MD_FORCE) &&
1811 !(mdr->md_options & MD_FORCE))
1812 return (EBUSY);
1813 return (mdresize(sc, mdr));
1814 }
1815
1816 static int
kern_mdresize(struct md_req * mdr)1817 kern_mdresize(struct md_req *mdr)
1818 {
1819 int error;
1820
1821 sx_xlock(&md_sx);
1822 error = kern_mdresize_locked(mdr);
1823 sx_xunlock(&md_sx);
1824 return (error);
1825 }
1826
1827 static int
kern_mdquery_locked(struct md_req * mdr)1828 kern_mdquery_locked(struct md_req *mdr)
1829 {
1830 struct md_s *sc;
1831 int error;
1832
1833 sx_assert(&md_sx, SA_XLOCKED);
1834
1835 sc = mdfind(mdr->md_unit);
1836 if (sc == NULL)
1837 return (ENOENT);
1838 mdr->md_type = sc->type;
1839 mdr->md_options = sc->flags;
1840 mdr->md_mediasize = sc->mediasize;
1841 mdr->md_sectorsize = sc->sectorsize;
1842 error = 0;
1843 if (mdr->md_label != NULL) {
1844 error = copyout(sc->label, mdr->md_label,
1845 strlen(sc->label) + 1);
1846 if (error != 0)
1847 return (error);
1848 }
1849 if (sc->type == MD_VNODE ||
1850 (sc->type == MD_PRELOAD && mdr->md_file != NULL))
1851 error = copyout(sc->file, mdr->md_file,
1852 strlen(sc->file) + 1);
1853 return (error);
1854 }
1855
1856 static int
kern_mdquery(struct md_req * mdr)1857 kern_mdquery(struct md_req *mdr)
1858 {
1859 int error;
1860
1861 sx_xlock(&md_sx);
1862 error = kern_mdquery_locked(mdr);
1863 sx_xunlock(&md_sx);
1864 return (error);
1865 }
1866
1867 /* Copy members that are not userspace pointers. */
1868 #define MD_IOCTL2REQ(mdio, mdr) do { \
1869 (mdr)->md_unit = (mdio)->md_unit; \
1870 (mdr)->md_type = (mdio)->md_type; \
1871 (mdr)->md_mediasize = (mdio)->md_mediasize; \
1872 (mdr)->md_sectorsize = (mdio)->md_sectorsize; \
1873 (mdr)->md_options = (mdio)->md_options; \
1874 (mdr)->md_fwheads = (mdio)->md_fwheads; \
1875 (mdr)->md_fwsectors = (mdio)->md_fwsectors; \
1876 (mdr)->md_units = &(mdio)->md_pad[0]; \
1877 (mdr)->md_units_nitems = nitems((mdio)->md_pad); \
1878 } while(0)
1879
1880 /* Copy members that might have been updated */
1881 #define MD_REQ2IOCTL(mdr, mdio) do { \
1882 (mdio)->md_unit = (mdr)->md_unit; \
1883 (mdio)->md_type = (mdr)->md_type; \
1884 (mdio)->md_mediasize = (mdr)->md_mediasize; \
1885 (mdio)->md_sectorsize = (mdr)->md_sectorsize; \
1886 (mdio)->md_options = (mdr)->md_options; \
1887 (mdio)->md_fwheads = (mdr)->md_fwheads; \
1888 (mdio)->md_fwsectors = (mdr)->md_fwsectors; \
1889 } while(0)
1890
1891 static int
mdctlioctl(struct cdev * dev,u_long cmd,caddr_t addr,int flags,struct thread * td)1892 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1893 struct thread *td)
1894 {
1895 struct md_req mdr;
1896 int error;
1897
1898 if (md_debug)
1899 printf("mdctlioctl(%s %lx %p %x %p)\n",
1900 devtoname(dev), cmd, addr, flags, td);
1901
1902 bzero(&mdr, sizeof(mdr));
1903 switch (cmd) {
1904 case MDIOCATTACH:
1905 case MDIOCDETACH:
1906 case MDIOCRESIZE:
1907 case MDIOCQUERY: {
1908 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1909 if (mdio->md_version != MDIOVERSION)
1910 return (EINVAL);
1911 MD_IOCTL2REQ(mdio, &mdr);
1912 mdr.md_file = mdio->md_file;
1913 mdr.md_file_seg = UIO_USERSPACE;
1914 /* If the file is adjacent to the md_ioctl it's in kernel. */
1915 if ((void *)mdio->md_file == (void *)(mdio + 1))
1916 mdr.md_file_seg = UIO_SYSSPACE;
1917 mdr.md_label = mdio->md_label;
1918 break;
1919 }
1920 #ifdef COMPAT_FREEBSD32
1921 case MDIOCATTACH_32:
1922 case MDIOCDETACH_32:
1923 case MDIOCRESIZE_32:
1924 case MDIOCQUERY_32: {
1925 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1926 if (mdio->md_version != MDIOVERSION)
1927 return (EINVAL);
1928 MD_IOCTL2REQ(mdio, &mdr);
1929 mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1930 mdr.md_file_seg = UIO_USERSPACE;
1931 mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1932 break;
1933 }
1934 #endif
1935 default:
1936 /* Fall through to handler switch. */
1937 break;
1938 }
1939
1940 error = 0;
1941 switch (cmd) {
1942 case MDIOCATTACH:
1943 #ifdef COMPAT_FREEBSD32
1944 case MDIOCATTACH_32:
1945 #endif
1946 error = kern_mdattach(td, &mdr);
1947 break;
1948 case MDIOCDETACH:
1949 #ifdef COMPAT_FREEBSD32
1950 case MDIOCDETACH_32:
1951 #endif
1952 error = kern_mddetach(td, &mdr);
1953 break;
1954 case MDIOCRESIZE:
1955 #ifdef COMPAT_FREEBSD32
1956 case MDIOCRESIZE_32:
1957 #endif
1958 error = kern_mdresize(&mdr);
1959 break;
1960 case MDIOCQUERY:
1961 #ifdef COMPAT_FREEBSD32
1962 case MDIOCQUERY_32:
1963 #endif
1964 error = kern_mdquery(&mdr);
1965 break;
1966 default:
1967 error = ENOIOCTL;
1968 }
1969
1970 switch (cmd) {
1971 case MDIOCATTACH:
1972 case MDIOCQUERY: {
1973 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1974 MD_REQ2IOCTL(&mdr, mdio);
1975 break;
1976 }
1977 #ifdef COMPAT_FREEBSD32
1978 case MDIOCATTACH_32:
1979 case MDIOCQUERY_32: {
1980 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1981 MD_REQ2IOCTL(&mdr, mdio);
1982 break;
1983 }
1984 #endif
1985 default:
1986 /* Other commands to not alter mdr. */
1987 break;
1988 }
1989
1990 return (error);
1991 }
1992
1993 static void
md_preloaded(u_char * image,size_t length,const char * name)1994 md_preloaded(u_char *image, size_t length, const char *name)
1995 {
1996 struct md_s *sc;
1997 int error;
1998
1999 sc = mdnew(-1, &error, MD_PRELOAD);
2000 if (sc == NULL)
2001 return;
2002 sc->mediasize = length;
2003 sc->sectorsize = DEV_BSIZE;
2004 sc->pl_ptr = image;
2005 sc->pl_len = length;
2006 sc->start = mdstart_preload;
2007 if (name != NULL)
2008 strlcpy(sc->file, name, sizeof(sc->file));
2009 #ifdef MD_ROOT
2010 if (sc->unit == 0) {
2011 #ifndef ROOTDEVNAME
2012 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2013 #endif
2014 #ifdef MD_ROOT_READONLY
2015 sc->flags |= MD_READONLY;
2016 #endif
2017 }
2018 #endif
2019 mdinit(sc);
2020 if (name != NULL) {
2021 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2022 MD_NAME, sc->unit, name, length, image);
2023 } else {
2024 printf("%s%d: Embedded image %zd bytes at %p\n",
2025 MD_NAME, sc->unit, length, image);
2026 }
2027 }
2028
2029 static void
g_md_init(struct g_class * mp __unused)2030 g_md_init(struct g_class *mp __unused)
2031 {
2032 caddr_t mod;
2033 u_char *ptr, *name, *type;
2034 unsigned len;
2035 int i;
2036
2037 /* figure out log2(NINDIR) */
2038 for (i = NINDIR, nshift = -1; i; nshift++)
2039 i >>= 1;
2040
2041 mod = NULL;
2042 sx_init(&md_sx, "MD config lock");
2043 g_topology_unlock();
2044 md_uh = new_unrhdr(0, INT_MAX, NULL);
2045 #ifdef MD_ROOT
2046 if (mfs_root_size != 0) {
2047 sx_xlock(&md_sx);
2048 #ifdef MD_ROOT_MEM
2049 md_preloaded(mfs_root, mfs_root_size, NULL);
2050 #else
2051 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2052 NULL);
2053 #endif
2054 sx_xunlock(&md_sx);
2055 }
2056 #endif
2057 /* XXX: are preload_* static or do they need Giant ? */
2058 while ((mod = preload_search_next_name(mod)) != NULL) {
2059 name = (char *)preload_search_info(mod, MODINFO_NAME);
2060 if (name == NULL)
2061 continue;
2062 type = (char *)preload_search_info(mod, MODINFO_TYPE);
2063 if (type == NULL)
2064 continue;
2065 if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2066 continue;
2067 ptr = preload_fetch_addr(mod);
2068 len = preload_fetch_size(mod);
2069 if (ptr != NULL && len != 0) {
2070 sx_xlock(&md_sx);
2071 md_preloaded(ptr, len, name);
2072 sx_xunlock(&md_sx);
2073 }
2074 }
2075 md_pbuf_zone = pbuf_zsecond_create("mdpbuf", nswbuf / 10);
2076 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2077 0600, MDCTL_NAME);
2078 g_topology_lock();
2079 }
2080
2081 static void
g_md_dumpconf(struct sbuf * sb,const char * indent,struct g_geom * gp,struct g_consumer * cp __unused,struct g_provider * pp)2082 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2083 struct g_consumer *cp __unused, struct g_provider *pp)
2084 {
2085 struct md_s *mp;
2086 char *type;
2087
2088 mp = gp->softc;
2089 if (mp == NULL)
2090 return;
2091
2092 switch (mp->type) {
2093 case MD_MALLOC:
2094 type = "malloc";
2095 break;
2096 case MD_PRELOAD:
2097 type = "preload";
2098 break;
2099 case MD_VNODE:
2100 type = "vnode";
2101 break;
2102 case MD_SWAP:
2103 type = "swap";
2104 break;
2105 case MD_NULL:
2106 type = "null";
2107 break;
2108 default:
2109 type = "unknown";
2110 break;
2111 }
2112
2113 if (pp != NULL) {
2114 if (indent == NULL) {
2115 sbuf_printf(sb, " u %d", mp->unit);
2116 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2117 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2118 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2119 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2120 sbuf_printf(sb, " t %s", type);
2121 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2122 (mp->type == MD_PRELOAD && mp->file[0] != '\0'))
2123 sbuf_printf(sb, " file %s", mp->file);
2124 sbuf_printf(sb, " label %s", mp->label);
2125 } else {
2126 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2127 mp->unit);
2128 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2129 indent, (uintmax_t) mp->sectorsize);
2130 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2131 indent, (uintmax_t) mp->fwheads);
2132 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2133 indent, (uintmax_t) mp->fwsectors);
2134 if (mp->ident[0] != '\0') {
2135 sbuf_printf(sb, "%s<ident>", indent);
2136 g_conf_printf_escaped(sb, "%s", mp->ident);
2137 sbuf_printf(sb, "</ident>\n");
2138 }
2139 sbuf_printf(sb, "%s<length>%ju</length>\n",
2140 indent, (uintmax_t) mp->mediasize);
2141 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2142 (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2143 sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2144 (mp->flags & MD_READONLY) == 0 ? "read-write":
2145 "read-only");
2146 sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2147 type);
2148 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2149 (mp->type == MD_PRELOAD && mp->file[0] != '\0')) {
2150 sbuf_printf(sb, "%s<file>", indent);
2151 g_conf_printf_escaped(sb, "%s", mp->file);
2152 sbuf_printf(sb, "</file>\n");
2153 }
2154 if (mp->type == MD_VNODE)
2155 sbuf_printf(sb, "%s<cache>%s</cache>\n", indent,
2156 (mp->flags & MD_CACHE) == 0 ? "off": "on");
2157 sbuf_printf(sb, "%s<label>", indent);
2158 g_conf_printf_escaped(sb, "%s", mp->label);
2159 sbuf_printf(sb, "</label>\n");
2160 }
2161 }
2162 }
2163
2164 static void
g_md_fini(struct g_class * mp __unused)2165 g_md_fini(struct g_class *mp __unused)
2166 {
2167
2168 sx_destroy(&md_sx);
2169 if (status_dev != NULL)
2170 destroy_dev(status_dev);
2171 uma_zdestroy(md_pbuf_zone);
2172 delete_unrhdr(md_uh);
2173 }
2174