1 /*-
2 * SPDX-License-Identifier: (Beerware AND BSD-3-Clause)
3 *
4 * ----------------------------------------------------------------------------
5 * "THE BEER-WARE LICENSE" (Revision 42):
6 * <phk@FreeBSD.ORG> 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 */
12
13 /*-
14 * The following functions are based on the historical vn(4) driver:
15 * mdstart_swap(), mdstart_vnode(), mdcreate_swap(), mdcreate_vnode()
16 * and mddestroy(), and as such under the following copyright:
17 *
18 * Copyright (c) 1988 University of Utah.
19 * Copyright (c) 1990, 1993
20 * The Regents of the University of California. All rights reserved.
21 * Copyright (c) 2013 The FreeBSD Foundation
22 * All rights reserved.
23 *
24 * This code is derived from software contributed to Berkeley by
25 * the Systems Programming Group of the University of Utah Computer
26 * Science Department.
27 *
28 * Portions of this software were developed by Konstantin Belousov
29 * under sponsorship from the FreeBSD Foundation.
30 *
31 * Redistribution and use in source and binary forms, with or without
32 * modification, are permitted provided that the following conditions
33 * are met:
34 * 1. Redistributions of source code must retain the above copyright
35 * notice, this list of conditions and the following disclaimer.
36 * 2. Redistributions in binary form must reproduce the above copyright
37 * notice, this list of conditions and the following disclaimer in the
38 * documentation and/or other materials provided with the distribution.
39 * 3. Neither the name of the University nor the names of its contributors
40 * may be used to endorse or promote products derived from this software
41 * without specific prior written permission.
42 *
43 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
44 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
45 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
46 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
47 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
48 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
49 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
50 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
51 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
52 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
53 * SUCH DAMAGE.
54 *
55 * from: Utah Hdr: vn.c 1.13 94/04/02
56 * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03
57 */
58
59 #include "opt_rootdevname.h"
60 #include "opt_geom.h"
61 #include "opt_md.h"
62
63 #include <sys/systm.h>
64 #include <sys/bio.h>
65 #include <sys/buf.h>
66 #include <sys/bus.h>
67 #include <sys/conf.h>
68 #include <sys/devicestat.h>
69 #include <sys/disk.h>
70 #include <sys/fcntl.h>
71 #include <sys/kernel.h>
72 #include <sys/kthread.h>
73 #include <sys/limits.h>
74 #include <sys/linker.h>
75 #include <sys/lock.h>
76 #include <sys/malloc.h>
77 #include <sys/mdioctl.h>
78 #include <sys/mount.h>
79 #include <sys/mutex.h>
80 #include <sys/namei.h>
81 #include <sys/proc.h>
82 #include <sys/queue.h>
83 #include <sys/rwlock.h>
84 #include <sys/sx.h>
85 #include <sys/sbuf.h>
86 #include <sys/sched.h>
87 #include <sys/sf_buf.h>
88 #include <sys/sysctl.h>
89 #include <sys/uio.h>
90 #include <sys/unistd.h>
91 #include <sys/vnode.h>
92
93 #include <geom/geom.h>
94 #include <geom/geom_int.h>
95
96 #include <vm/vm.h>
97 #include <vm/vm_extern.h>
98 #include <vm/vm_param.h>
99 #include <vm/vm_object.h>
100 #include <vm/vm_page.h>
101 #include <vm/vm_pager.h>
102 #include <vm/swap_pager.h>
103 #include <vm/uma.h>
104
105 #include <machine/bus.h>
106
107 #define MD_MODVER 1
108
109 #define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */
110 #define MD_EXITING 0x20000 /* Worker thread is exiting. */
111 #define MD_PROVIDERGONE 0x40000 /* Safe to free the softc */
112
113 #ifndef MD_NSECT
114 #define MD_NSECT (10000 * 2)
115 #endif
116
117 struct md_req {
118 unsigned md_unit; /* unit number */
119 enum md_types md_type; /* type of disk */
120 off_t md_mediasize; /* size of disk in bytes */
121 unsigned md_sectorsize; /* sectorsize */
122 unsigned md_options; /* options */
123 int md_fwheads; /* firmware heads */
124 int md_fwsectors; /* firmware sectors */
125 char *md_file; /* pathname of file to mount */
126 enum uio_seg md_file_seg; /* location of md_file */
127 char *md_label; /* label of the device (userspace) */
128 int *md_units; /* pointer to units array (kernel) */
129 size_t md_units_nitems; /* items in md_units array */
130 };
131
132 #ifdef COMPAT_FREEBSD32
133 struct md_ioctl32 {
134 unsigned md_version;
135 unsigned md_unit;
136 enum md_types md_type;
137 uint32_t md_file;
138 off_t md_mediasize;
139 unsigned md_sectorsize;
140 unsigned md_options;
141 uint64_t md_base;
142 int md_fwheads;
143 int md_fwsectors;
144 uint32_t md_label;
145 int md_pad[MDNPAD];
146 }
147 #ifdef __amd64__
148 __attribute__((__packed__))
149 #endif
150 ;
151 #ifndef __amd64__
152 CTASSERT((sizeof(struct md_ioctl32)) == 440);
153 #else
154 CTASSERT((sizeof(struct md_ioctl32)) == 436);
155 #endif
156
157 #define MDIOCATTACH_32 _IOC_NEWTYPE(MDIOCATTACH, struct md_ioctl32)
158 #define MDIOCDETACH_32 _IOC_NEWTYPE(MDIOCDETACH, struct md_ioctl32)
159 #define MDIOCQUERY_32 _IOC_NEWTYPE(MDIOCQUERY, struct md_ioctl32)
160 #define MDIOCRESIZE_32 _IOC_NEWTYPE(MDIOCRESIZE, struct md_ioctl32)
161 #endif /* COMPAT_FREEBSD32 */
162
163 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk");
164 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors");
165
166 static int md_debug;
167 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0,
168 "Enable md(4) debug messages");
169 static int md_malloc_wait;
170 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0,
171 "Allow malloc to wait for memory allocations");
172
173 #if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE)
174 #define MD_ROOT_FSTYPE "ufs"
175 #endif
176
177 #if defined(MD_ROOT)
178 /*
179 * Preloaded image gets put here.
180 */
181 #if defined(MD_ROOT_SIZE)
182 /*
183 * We put the mfs_root symbol into the oldmfs section of the kernel object file.
184 * Applications that patch the object with the image can determine
185 * the size looking at the oldmfs section size within the kernel.
186 */
187 u_char mfs_root[MD_ROOT_SIZE*1024] __attribute__ ((section ("oldmfs")));
188 const int mfs_root_size = sizeof(mfs_root);
189 #elif defined(MD_ROOT_MEM)
190 /* MD region already mapped in the memory */
191 u_char *mfs_root;
192 int mfs_root_size;
193 #else
194 extern volatile u_char __weak_symbol mfs_root;
195 extern volatile u_char __weak_symbol mfs_root_end;
196 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root))
197 #endif
198 #endif
199
200 static g_init_t g_md_init;
201 static g_fini_t g_md_fini;
202 static g_start_t g_md_start;
203 static g_access_t g_md_access;
204 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
205 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
206 static g_provgone_t g_md_providergone;
207
208 static struct cdev *status_dev = NULL;
209 static struct sx md_sx;
210 static struct unrhdr *md_uh;
211
212 static d_ioctl_t mdctlioctl;
213
214 static struct cdevsw mdctl_cdevsw = {
215 .d_version = D_VERSION,
216 .d_ioctl = mdctlioctl,
217 .d_name = MD_NAME,
218 };
219
220 struct g_class g_md_class = {
221 .name = "MD",
222 .version = G_VERSION,
223 .init = g_md_init,
224 .fini = g_md_fini,
225 .start = g_md_start,
226 .access = g_md_access,
227 .dumpconf = g_md_dumpconf,
228 .providergone = g_md_providergone,
229 };
230
231 DECLARE_GEOM_CLASS(g_md_class, g_md);
232 MODULE_VERSION(geom_md, 0);
233
234 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
235
236 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t))
237 #define NMASK (NINDIR-1)
238 static int nshift;
239
240 struct indir {
241 uintptr_t *array;
242 u_int total;
243 u_int used;
244 u_int shift;
245 };
246
247 struct md_s {
248 int unit;
249 LIST_ENTRY(md_s) list;
250 struct bio_queue_head bio_queue;
251 struct mtx queue_mtx;
252 struct cdev *dev;
253 enum md_types type;
254 off_t mediasize;
255 unsigned sectorsize;
256 unsigned opencount;
257 unsigned fwheads;
258 unsigned fwsectors;
259 char ident[DISK_IDENT_SIZE];
260 unsigned flags;
261 char name[20];
262 struct proc *procp;
263 struct g_geom *gp;
264 struct g_provider *pp;
265 int (*start)(struct md_s *sc, struct bio *bp);
266 struct devstat *devstat;
267 struct ucred *cred;
268 char label[PATH_MAX];
269 bool candelete;
270
271 union {
272 /* MD_MALLOC related fields */
273 struct {
274 struct indir *indir;
275 uma_zone_t uma;
276 } s_malloc;
277
278 /* MD_PRELOAD related fields */
279 struct {
280 u_char *pl_ptr;
281 size_t pl_len;
282 char name[PATH_MAX];
283 } s_preload;
284
285 /* MD_VNODE related fields */
286 struct {
287 struct vnode *vnode;
288 char file[PATH_MAX];
289 char *kva;
290 } s_vnode;
291
292 /* MD_SWAP related fields */
293 struct {
294 vm_object_t object;
295 } s_swap;
296
297 /* MD_NULL */
298 struct {
299 } s_null;
300 };
301 };
302
303 static struct indir *
new_indir(u_int shift)304 new_indir(u_int shift)
305 {
306 struct indir *ip;
307
308 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
309 | M_ZERO);
310 if (ip == NULL)
311 return (NULL);
312 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
313 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
314 if (ip->array == NULL) {
315 free(ip, M_MD);
316 return (NULL);
317 }
318 ip->total = NINDIR;
319 ip->shift = shift;
320 return (ip);
321 }
322
323 static void
del_indir(struct indir * ip)324 del_indir(struct indir *ip)
325 {
326
327 free(ip->array, M_MDSECT);
328 free(ip, M_MD);
329 }
330
331 static void
destroy_indir(struct md_s * sc,struct indir * ip)332 destroy_indir(struct md_s *sc, struct indir *ip)
333 {
334 int i;
335
336 for (i = 0; i < NINDIR; i++) {
337 if (!ip->array[i])
338 continue;
339 if (ip->shift)
340 destroy_indir(sc, (struct indir*)(ip->array[i]));
341 else if (ip->array[i] > 255)
342 uma_zfree(sc->s_malloc.uma, (void *)(ip->array[i]));
343 }
344 del_indir(ip);
345 }
346
347 /*
348 * This function does the math and allocates the top level "indir" structure
349 * for a device of "size" sectors.
350 */
351
352 static struct indir *
dimension(off_t size)353 dimension(off_t size)
354 {
355 off_t rcnt;
356 struct indir *ip;
357 int layer;
358
359 rcnt = size;
360 layer = 0;
361 while (rcnt > NINDIR) {
362 rcnt /= NINDIR;
363 layer++;
364 }
365
366 /*
367 * XXX: the top layer is probably not fully populated, so we allocate
368 * too much space for ip->array in here.
369 */
370 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
371 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
372 M_MDSECT, M_WAITOK | M_ZERO);
373 ip->total = NINDIR;
374 ip->shift = layer * nshift;
375 return (ip);
376 }
377
378 /*
379 * Read a given sector
380 */
381
382 static uintptr_t
s_read(struct indir * ip,off_t offset)383 s_read(struct indir *ip, off_t offset)
384 {
385 struct indir *cip;
386 int idx;
387 uintptr_t up;
388
389 if (md_debug > 1)
390 printf("s_read(%jd)\n", (intmax_t)offset);
391 up = 0;
392 for (cip = ip; cip != NULL;) {
393 if (cip->shift) {
394 idx = (offset >> cip->shift) & NMASK;
395 up = cip->array[idx];
396 cip = (struct indir *)up;
397 continue;
398 }
399 idx = offset & NMASK;
400 return (cip->array[idx]);
401 }
402 return (0);
403 }
404
405 /*
406 * Write a given sector, prune the tree if the value is 0
407 */
408
409 static int
s_write(struct indir * ip,off_t offset,uintptr_t ptr)410 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
411 {
412 struct indir *cip, *lip[10];
413 int idx, li;
414 uintptr_t up;
415
416 if (md_debug > 1)
417 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
418 up = 0;
419 li = 0;
420 cip = ip;
421 for (;;) {
422 lip[li++] = cip;
423 if (cip->shift) {
424 idx = (offset >> cip->shift) & NMASK;
425 up = cip->array[idx];
426 if (up != 0) {
427 cip = (struct indir *)up;
428 continue;
429 }
430 /* Allocate branch */
431 cip->array[idx] =
432 (uintptr_t)new_indir(cip->shift - nshift);
433 if (cip->array[idx] == 0)
434 return (ENOSPC);
435 cip->used++;
436 up = cip->array[idx];
437 cip = (struct indir *)up;
438 continue;
439 }
440 /* leafnode */
441 idx = offset & NMASK;
442 up = cip->array[idx];
443 if (up != 0)
444 cip->used--;
445 cip->array[idx] = ptr;
446 if (ptr != 0)
447 cip->used++;
448 break;
449 }
450 if (cip->used != 0 || li == 1)
451 return (0);
452 li--;
453 while (cip->used == 0 && cip != ip) {
454 li--;
455 idx = (offset >> lip[li]->shift) & NMASK;
456 up = lip[li]->array[idx];
457 KASSERT(up == (uintptr_t)cip, ("md screwed up"));
458 del_indir(cip);
459 lip[li]->array[idx] = 0;
460 lip[li]->used--;
461 cip = lip[li];
462 }
463 return (0);
464 }
465
466 static int
g_md_access(struct g_provider * pp,int r,int w,int e)467 g_md_access(struct g_provider *pp, int r, int w, int e)
468 {
469 struct md_s *sc;
470
471 sc = pp->geom->softc;
472 if (sc == NULL) {
473 if (r <= 0 && w <= 0 && e <= 0)
474 return (0);
475 return (ENXIO);
476 }
477 r += pp->acr;
478 w += pp->acw;
479 e += pp->ace;
480 if ((sc->flags & MD_READONLY) != 0 && w > 0)
481 return (EROFS);
482 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
483 sc->opencount = 1;
484 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
485 sc->opencount = 0;
486 }
487 return (0);
488 }
489
490 static void
g_md_start(struct bio * bp)491 g_md_start(struct bio *bp)
492 {
493 struct md_s *sc;
494
495 sc = bp->bio_to->geom->softc;
496 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
497 devstat_start_transaction_bio(sc->devstat, bp);
498 }
499 mtx_lock(&sc->queue_mtx);
500 bioq_disksort(&sc->bio_queue, bp);
501 wakeup(sc);
502 mtx_unlock(&sc->queue_mtx);
503 }
504
505 #define MD_MALLOC_MOVE_ZERO 1
506 #define MD_MALLOC_MOVE_FILL 2
507 #define MD_MALLOC_MOVE_READ 3
508 #define MD_MALLOC_MOVE_WRITE 4
509 #define MD_MALLOC_MOVE_CMP 5
510
511 static int
md_malloc_move_ma(vm_page_t ** mp,int * ma_offs,unsigned sectorsize,void * ptr,u_char fill,int op)512 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize,
513 void *ptr, u_char fill, int op)
514 {
515 struct sf_buf *sf;
516 vm_page_t m, *mp1;
517 char *p, first;
518 off_t *uc;
519 unsigned n;
520 int error, i, ma_offs1, sz, first_read;
521
522 m = NULL;
523 error = 0;
524 sf = NULL;
525 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */
526 first = 0;
527 first_read = 0;
528 uc = ptr;
529 mp1 = *mp;
530 ma_offs1 = *ma_offs;
531 /* } */
532 sched_pin();
533 for (n = sectorsize; n != 0; n -= sz) {
534 sz = imin(PAGE_SIZE - *ma_offs, n);
535 if (m != **mp) {
536 if (sf != NULL)
537 sf_buf_free(sf);
538 m = **mp;
539 sf = sf_buf_alloc(m, SFB_CPUPRIVATE |
540 (md_malloc_wait ? 0 : SFB_NOWAIT));
541 if (sf == NULL) {
542 error = ENOMEM;
543 break;
544 }
545 }
546 p = (char *)sf_buf_kva(sf) + *ma_offs;
547 switch (op) {
548 case MD_MALLOC_MOVE_ZERO:
549 bzero(p, sz);
550 break;
551 case MD_MALLOC_MOVE_FILL:
552 memset(p, fill, sz);
553 break;
554 case MD_MALLOC_MOVE_READ:
555 bcopy(ptr, p, sz);
556 cpu_flush_dcache(p, sz);
557 break;
558 case MD_MALLOC_MOVE_WRITE:
559 bcopy(p, ptr, sz);
560 break;
561 case MD_MALLOC_MOVE_CMP:
562 for (i = 0; i < sz; i++, p++) {
563 if (!first_read) {
564 *uc = (u_char)*p;
565 first = *p;
566 first_read = 1;
567 } else if (*p != first) {
568 error = EDOOFUS;
569 break;
570 }
571 }
572 break;
573 default:
574 KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op));
575 break;
576 }
577 if (error != 0)
578 break;
579 *ma_offs += sz;
580 *ma_offs %= PAGE_SIZE;
581 if (*ma_offs == 0)
582 (*mp)++;
583 ptr = (char *)ptr + sz;
584 }
585
586 if (sf != NULL)
587 sf_buf_free(sf);
588 sched_unpin();
589 if (op == MD_MALLOC_MOVE_CMP && error != 0) {
590 *mp = mp1;
591 *ma_offs = ma_offs1;
592 }
593 return (error);
594 }
595
596 static int
md_malloc_move_vlist(bus_dma_segment_t ** pvlist,int * pma_offs,unsigned len,void * ptr,u_char fill,int op)597 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs,
598 unsigned len, void *ptr, u_char fill, int op)
599 {
600 bus_dma_segment_t *vlist;
601 uint8_t *p, *end, first;
602 off_t *uc;
603 int ma_offs, seg_len;
604
605 vlist = *pvlist;
606 ma_offs = *pma_offs;
607 uc = ptr;
608
609 for (; len != 0; len -= seg_len) {
610 seg_len = imin(vlist->ds_len - ma_offs, len);
611 p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs;
612 switch (op) {
613 case MD_MALLOC_MOVE_ZERO:
614 bzero(p, seg_len);
615 break;
616 case MD_MALLOC_MOVE_FILL:
617 memset(p, fill, seg_len);
618 break;
619 case MD_MALLOC_MOVE_READ:
620 bcopy(ptr, p, seg_len);
621 cpu_flush_dcache(p, seg_len);
622 break;
623 case MD_MALLOC_MOVE_WRITE:
624 bcopy(p, ptr, seg_len);
625 break;
626 case MD_MALLOC_MOVE_CMP:
627 end = p + seg_len;
628 first = *uc = *p;
629 /* Confirm all following bytes match the first */
630 while (++p < end) {
631 if (*p != first)
632 return (EDOOFUS);
633 }
634 break;
635 default:
636 KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op));
637 break;
638 }
639
640 ma_offs += seg_len;
641 if (ma_offs == vlist->ds_len) {
642 ma_offs = 0;
643 vlist++;
644 }
645 ptr = (uint8_t *)ptr + seg_len;
646 }
647 *pvlist = vlist;
648 *pma_offs = ma_offs;
649
650 return (0);
651 }
652
653 static int
mdstart_malloc(struct md_s * sc,struct bio * bp)654 mdstart_malloc(struct md_s *sc, struct bio *bp)
655 {
656 u_char *dst;
657 vm_page_t *m;
658 bus_dma_segment_t *vlist;
659 int i, error, error1, ma_offs, notmapped;
660 off_t secno, nsec, uc;
661 uintptr_t sp, osp;
662
663 switch (bp->bio_cmd) {
664 case BIO_READ:
665 case BIO_WRITE:
666 case BIO_DELETE:
667 break;
668 case BIO_FLUSH:
669 return (0);
670 default:
671 return (EOPNOTSUPP);
672 }
673
674 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0;
675 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
676 (bus_dma_segment_t *)bp->bio_data : NULL;
677 if (notmapped) {
678 m = bp->bio_ma;
679 ma_offs = bp->bio_ma_offset;
680 dst = NULL;
681 KASSERT(vlist == NULL, ("vlists cannot be unmapped"));
682 } else if (vlist != NULL) {
683 ma_offs = bp->bio_ma_offset;
684 dst = NULL;
685 } else {
686 dst = bp->bio_data;
687 }
688
689 nsec = bp->bio_length / sc->sectorsize;
690 secno = bp->bio_offset / sc->sectorsize;
691 error = 0;
692 while (nsec--) {
693 osp = s_read(sc->s_malloc.indir, secno);
694 if (bp->bio_cmd == BIO_DELETE) {
695 if (osp != 0)
696 error = s_write(sc->s_malloc.indir, secno, 0);
697 } else if (bp->bio_cmd == BIO_READ) {
698 if (osp == 0) {
699 if (notmapped) {
700 error = md_malloc_move_ma(&m, &ma_offs,
701 sc->sectorsize, NULL, 0,
702 MD_MALLOC_MOVE_ZERO);
703 } else if (vlist != NULL) {
704 error = md_malloc_move_vlist(&vlist,
705 &ma_offs, sc->sectorsize, NULL, 0,
706 MD_MALLOC_MOVE_ZERO);
707 } else
708 bzero(dst, sc->sectorsize);
709 } else if (osp <= 255) {
710 if (notmapped) {
711 error = md_malloc_move_ma(&m, &ma_offs,
712 sc->sectorsize, NULL, osp,
713 MD_MALLOC_MOVE_FILL);
714 } else if (vlist != NULL) {
715 error = md_malloc_move_vlist(&vlist,
716 &ma_offs, sc->sectorsize, NULL, osp,
717 MD_MALLOC_MOVE_FILL);
718 } else
719 memset(dst, osp, sc->sectorsize);
720 } else {
721 if (notmapped) {
722 error = md_malloc_move_ma(&m, &ma_offs,
723 sc->sectorsize, (void *)osp, 0,
724 MD_MALLOC_MOVE_READ);
725 } else if (vlist != NULL) {
726 error = md_malloc_move_vlist(&vlist,
727 &ma_offs, sc->sectorsize,
728 (void *)osp, 0,
729 MD_MALLOC_MOVE_READ);
730 } else {
731 bcopy((void *)osp, dst, sc->sectorsize);
732 cpu_flush_dcache(dst, sc->sectorsize);
733 }
734 }
735 osp = 0;
736 } else if (bp->bio_cmd == BIO_WRITE) {
737 if (sc->flags & MD_COMPRESS) {
738 if (notmapped) {
739 error1 = md_malloc_move_ma(&m, &ma_offs,
740 sc->sectorsize, &uc, 0,
741 MD_MALLOC_MOVE_CMP);
742 i = error1 == 0 ? sc->sectorsize : 0;
743 } else if (vlist != NULL) {
744 error1 = md_malloc_move_vlist(&vlist,
745 &ma_offs, sc->sectorsize, &uc, 0,
746 MD_MALLOC_MOVE_CMP);
747 i = error1 == 0 ? sc->sectorsize : 0;
748 } else {
749 uc = dst[0];
750 for (i = 1; i < sc->sectorsize; i++) {
751 if (dst[i] != uc)
752 break;
753 }
754 }
755 } else {
756 i = 0;
757 uc = 0;
758 }
759 if (i == sc->sectorsize) {
760 if (osp != uc)
761 error = s_write(sc->s_malloc.indir,
762 secno, uc);
763 } else {
764 if (osp <= 255) {
765 sp = (uintptr_t)uma_zalloc(
766 sc->s_malloc.uma,
767 md_malloc_wait ? M_WAITOK :
768 M_NOWAIT);
769 if (sp == 0) {
770 error = ENOSPC;
771 break;
772 }
773 if (notmapped) {
774 error = md_malloc_move_ma(&m,
775 &ma_offs, sc->sectorsize,
776 (void *)sp, 0,
777 MD_MALLOC_MOVE_WRITE);
778 } else if (vlist != NULL) {
779 error = md_malloc_move_vlist(
780 &vlist, &ma_offs,
781 sc->sectorsize, (void *)sp,
782 0, MD_MALLOC_MOVE_WRITE);
783 } else {
784 bcopy(dst, (void *)sp,
785 sc->sectorsize);
786 }
787 error = s_write(sc->s_malloc.indir,
788 secno, sp);
789 } else {
790 if (notmapped) {
791 error = md_malloc_move_ma(&m,
792 &ma_offs, sc->sectorsize,
793 (void *)osp, 0,
794 MD_MALLOC_MOVE_WRITE);
795 } else if (vlist != NULL) {
796 error = md_malloc_move_vlist(
797 &vlist, &ma_offs,
798 sc->sectorsize, (void *)osp,
799 0, MD_MALLOC_MOVE_WRITE);
800 } else {
801 bcopy(dst, (void *)osp,
802 sc->sectorsize);
803 }
804 osp = 0;
805 }
806 }
807 } else {
808 error = EOPNOTSUPP;
809 }
810 if (osp > 255)
811 uma_zfree(sc->s_malloc.uma, (void*)osp);
812 if (error != 0)
813 break;
814 secno++;
815 if (!notmapped && vlist == NULL)
816 dst += sc->sectorsize;
817 }
818 bp->bio_resid = 0;
819 return (error);
820 }
821
822 static void
mdcopyto_vlist(void * src,bus_dma_segment_t * vlist,off_t offset,off_t len)823 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len)
824 {
825 off_t seg_len;
826
827 while (offset >= vlist->ds_len) {
828 offset -= vlist->ds_len;
829 vlist++;
830 }
831
832 while (len != 0) {
833 seg_len = omin(len, vlist->ds_len - offset);
834 bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset),
835 seg_len);
836 offset = 0;
837 src = (uint8_t *)src + seg_len;
838 len -= seg_len;
839 vlist++;
840 }
841 }
842
843 static void
mdcopyfrom_vlist(bus_dma_segment_t * vlist,off_t offset,void * dst,off_t len)844 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len)
845 {
846 off_t seg_len;
847
848 while (offset >= vlist->ds_len) {
849 offset -= vlist->ds_len;
850 vlist++;
851 }
852
853 while (len != 0) {
854 seg_len = omin(len, vlist->ds_len - offset);
855 bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst,
856 seg_len);
857 offset = 0;
858 dst = (uint8_t *)dst + seg_len;
859 len -= seg_len;
860 vlist++;
861 }
862 }
863
864 static int
mdstart_preload(struct md_s * sc,struct bio * bp)865 mdstart_preload(struct md_s *sc, struct bio *bp)
866 {
867 uint8_t *p;
868
869 p = sc->s_preload.pl_ptr + bp->bio_offset;
870 switch (bp->bio_cmd) {
871 case BIO_READ:
872 if ((bp->bio_flags & BIO_VLIST) != 0) {
873 mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data,
874 bp->bio_ma_offset, bp->bio_length);
875 } else {
876 bcopy(p, bp->bio_data, bp->bio_length);
877 }
878 cpu_flush_dcache(bp->bio_data, bp->bio_length);
879 break;
880 case BIO_WRITE:
881 if ((bp->bio_flags & BIO_VLIST) != 0) {
882 mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data,
883 bp->bio_ma_offset, p, bp->bio_length);
884 } else {
885 bcopy(bp->bio_data, p, bp->bio_length);
886 }
887 break;
888 }
889 bp->bio_resid = 0;
890 return (0);
891 }
892
893 static int
mdstart_vnode(struct md_s * sc,struct bio * bp)894 mdstart_vnode(struct md_s *sc, struct bio *bp)
895 {
896 int error;
897 struct uio auio;
898 struct iovec aiov;
899 struct iovec *piov;
900 struct mount *mp;
901 struct vnode *vp;
902 bus_dma_segment_t *vlist;
903 struct thread *td;
904 off_t iolen, iostart, off, len;
905 int ma_offs, npages;
906 bool mapped;
907
908 td = curthread;
909 vp = sc->s_vnode.vnode;
910 piov = NULL;
911 ma_offs = bp->bio_ma_offset;
912 off = bp->bio_offset;
913 len = bp->bio_length;
914 mapped = false;
915
916 /*
917 * VNODE I/O
918 *
919 * If an error occurs, we set BIO_ERROR but we do not set
920 * B_INVAL because (for a write anyway), the buffer is
921 * still valid.
922 */
923
924 switch (bp->bio_cmd) {
925 case BIO_READ:
926 auio.uio_rw = UIO_READ;
927 break;
928 case BIO_WRITE:
929 auio.uio_rw = UIO_WRITE;
930 break;
931 case BIO_FLUSH:
932 do {
933 (void)vn_start_write(vp, &mp, V_WAIT);
934 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
935 error = VOP_FSYNC(vp, MNT_WAIT, td);
936 VOP_UNLOCK(vp);
937 vn_finished_write(mp);
938 } while (error == ERELOOKUP);
939 return (error);
940 case BIO_DELETE:
941 if (sc->candelete) {
942 error = vn_deallocate(vp, &off, &len, 0,
943 sc->flags & MD_ASYNC ? 0 : IO_SYNC,
944 sc->cred, NOCRED);
945 bp->bio_resid = len;
946 return (error);
947 }
948 /* FALLTHROUGH */
949 default:
950 return (EOPNOTSUPP);
951 }
952
953 auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
954 auio.uio_resid = bp->bio_length;
955 auio.uio_segflg = UIO_SYSSPACE;
956 auio.uio_td = td;
957
958 if ((bp->bio_flags & BIO_VLIST) != 0) {
959 piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK);
960 auio.uio_iov = piov;
961 vlist = (bus_dma_segment_t *)bp->bio_data;
962 while (len > 0) {
963 piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr +
964 ma_offs);
965 piov->iov_len = vlist->ds_len - ma_offs;
966 if (piov->iov_len > len)
967 piov->iov_len = len;
968 len -= piov->iov_len;
969 ma_offs = 0;
970 vlist++;
971 piov++;
972 }
973 auio.uio_iovcnt = piov - auio.uio_iov;
974 piov = auio.uio_iov;
975 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
976 bp->bio_resid = len;
977 unmapped_step:
978 npages = atop(min(maxphys, round_page(len + (ma_offs &
979 PAGE_MASK))));
980 iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len);
981 KASSERT(iolen > 0, ("zero iolen"));
982 KASSERT(npages <= atop(maxphys + PAGE_SIZE),
983 ("npages %d too large", npages));
984 pmap_qenter(sc->s_vnode.kva, &bp->bio_ma[atop(ma_offs)],
985 npages);
986 aiov.iov_base = sc->s_vnode.kva + (ma_offs & PAGE_MASK);
987 aiov.iov_len = iolen;
988 auio.uio_iov = &aiov;
989 auio.uio_iovcnt = 1;
990 auio.uio_resid = iolen;
991 mapped = true;
992 } else {
993 aiov.iov_base = bp->bio_data;
994 aiov.iov_len = bp->bio_length;
995 auio.uio_iov = &aiov;
996 auio.uio_iovcnt = 1;
997 }
998 iostart = auio.uio_offset;
999 if (bp->bio_cmd == BIO_READ) {
1000 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1001 error = VOP_READ(vp, &auio, 0, sc->cred);
1002 VOP_UNLOCK(vp);
1003 } else {
1004 (void) vn_start_write(vp, &mp, V_WAIT);
1005 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1006 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
1007 sc->cred);
1008 VOP_UNLOCK(vp);
1009 vn_finished_write(mp);
1010 if (error == 0)
1011 sc->flags &= ~MD_VERIFY;
1012 }
1013
1014 /* When MD_CACHE is set, try to avoid double-caching the data. */
1015 if (error == 0 && (sc->flags & MD_CACHE) == 0)
1016 VOP_ADVISE(vp, iostart, auio.uio_offset - 1,
1017 POSIX_FADV_DONTNEED);
1018
1019 if (mapped) {
1020 pmap_qremove(sc->s_vnode.kva, npages);
1021 if (error == 0) {
1022 len -= iolen;
1023 bp->bio_resid -= iolen;
1024 ma_offs += iolen;
1025 if (len > 0)
1026 goto unmapped_step;
1027 }
1028 } else {
1029 bp->bio_resid = auio.uio_resid;
1030 }
1031
1032 free(piov, M_MD);
1033 return (error);
1034 }
1035
1036 static int
mdstart_swap(struct md_s * sc,struct bio * bp)1037 mdstart_swap(struct md_s *sc, struct bio *bp)
1038 {
1039 vm_page_t m;
1040 u_char *p;
1041 vm_pindex_t i, lastp;
1042 bus_dma_segment_t *vlist;
1043 int rv, ma_offs, offs, len, lastend;
1044
1045 switch (bp->bio_cmd) {
1046 case BIO_READ:
1047 case BIO_WRITE:
1048 case BIO_DELETE:
1049 break;
1050 case BIO_FLUSH:
1051 return (0);
1052 default:
1053 return (EOPNOTSUPP);
1054 }
1055
1056 p = bp->bio_data;
1057 ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ?
1058 bp->bio_ma_offset : 0;
1059 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
1060 (bus_dma_segment_t *)bp->bio_data : NULL;
1061
1062 /*
1063 * offs is the offset at which to start operating on the
1064 * next (ie, first) page. lastp is the last page on
1065 * which we're going to operate. lastend is the ending
1066 * position within that last page (ie, PAGE_SIZE if
1067 * we're operating on complete aligned pages).
1068 */
1069 offs = bp->bio_offset % PAGE_SIZE;
1070 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
1071 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
1072
1073 rv = VM_PAGER_OK;
1074 vm_object_pip_add(sc->s_swap.object, 1);
1075 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
1076 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
1077 m = vm_page_grab_unlocked(sc->s_swap.object, i,
1078 VM_ALLOC_SYSTEM);
1079 if (bp->bio_cmd == BIO_READ) {
1080 if (vm_page_all_valid(m))
1081 rv = VM_PAGER_OK;
1082 else
1083 rv = vm_pager_get_pages(sc->s_swap.object,
1084 &m, 1, NULL, NULL);
1085 if (rv == VM_PAGER_ERROR) {
1086 VM_OBJECT_WLOCK(sc->s_swap.object);
1087 vm_page_free(m);
1088 VM_OBJECT_WUNLOCK(sc->s_swap.object);
1089 break;
1090 } else if (rv == VM_PAGER_FAIL) {
1091 /*
1092 * Pager does not have the page. Zero
1093 * the allocated page, and mark it as
1094 * valid. Do not set dirty, the page
1095 * can be recreated if thrown out.
1096 */
1097 pmap_zero_page(m);
1098 vm_page_valid(m);
1099 }
1100 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1101 pmap_copy_pages(&m, offs, bp->bio_ma,
1102 ma_offs, len);
1103 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1104 physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs,
1105 vlist, ma_offs, len);
1106 cpu_flush_dcache(p, len);
1107 } else {
1108 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1109 cpu_flush_dcache(p, len);
1110 }
1111 } else if (bp->bio_cmd == BIO_WRITE) {
1112 if (len == PAGE_SIZE || vm_page_all_valid(m))
1113 rv = VM_PAGER_OK;
1114 else
1115 rv = vm_pager_get_pages(sc->s_swap.object,
1116 &m, 1, NULL, NULL);
1117 if (rv == VM_PAGER_ERROR) {
1118 VM_OBJECT_WLOCK(sc->s_swap.object);
1119 vm_page_free(m);
1120 VM_OBJECT_WUNLOCK(sc->s_swap.object);
1121 break;
1122 } else if (rv == VM_PAGER_FAIL)
1123 pmap_zero_page(m);
1124
1125 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1126 pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1127 offs, len);
1128 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1129 physcopyin_vlist(vlist, ma_offs,
1130 VM_PAGE_TO_PHYS(m) + offs, len);
1131 } else {
1132 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1133 }
1134
1135 vm_page_valid(m);
1136 vm_page_set_dirty(m);
1137 } else if (bp->bio_cmd == BIO_DELETE) {
1138 if (len == PAGE_SIZE || vm_page_all_valid(m))
1139 rv = VM_PAGER_OK;
1140 else
1141 rv = vm_pager_get_pages(sc->s_swap.object,
1142 &m, 1, NULL, NULL);
1143 VM_OBJECT_WLOCK(sc->s_swap.object);
1144 if (rv == VM_PAGER_ERROR) {
1145 vm_page_free(m);
1146 VM_OBJECT_WUNLOCK(sc->s_swap.object);
1147 break;
1148 } else if (rv == VM_PAGER_FAIL) {
1149 vm_page_free(m);
1150 m = NULL;
1151 } else {
1152 /* Page is valid. */
1153 if (len != PAGE_SIZE) {
1154 pmap_zero_page_area(m, offs, len);
1155 vm_page_set_dirty(m);
1156 } else {
1157 vm_pager_page_unswapped(m);
1158 vm_page_free(m);
1159 m = NULL;
1160 }
1161 }
1162 VM_OBJECT_WUNLOCK(sc->s_swap.object);
1163 }
1164 if (m != NULL) {
1165 /*
1166 * The page may be deactivated prior to setting
1167 * PGA_REFERENCED, but in this case it will be
1168 * reactivated by the page daemon.
1169 */
1170 if (vm_page_active(m))
1171 vm_page_reference(m);
1172 else
1173 vm_page_activate(m);
1174 vm_page_xunbusy(m);
1175 }
1176
1177 /* Actions on further pages start at offset 0 */
1178 p += PAGE_SIZE - offs;
1179 offs = 0;
1180 ma_offs += len;
1181 }
1182 vm_object_pip_wakeup(sc->s_swap.object);
1183 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
1184 }
1185
1186 static int
mdstart_null(struct md_s * sc,struct bio * bp)1187 mdstart_null(struct md_s *sc, struct bio *bp)
1188 {
1189
1190 switch (bp->bio_cmd) {
1191 case BIO_READ:
1192 bzero(bp->bio_data, bp->bio_length);
1193 cpu_flush_dcache(bp->bio_data, bp->bio_length);
1194 break;
1195 case BIO_WRITE:
1196 break;
1197 }
1198 bp->bio_resid = 0;
1199 return (0);
1200 }
1201
1202 static void
md_handleattr(struct md_s * sc,struct bio * bp)1203 md_handleattr(struct md_s *sc, struct bio *bp)
1204 {
1205 if (sc->fwsectors && sc->fwheads &&
1206 (g_handleattr_int(bp, "GEOM::fwsectors", sc->fwsectors) != 0 ||
1207 g_handleattr_int(bp, "GEOM::fwheads", sc->fwheads) != 0))
1208 return;
1209 if (g_handleattr_int(bp, "GEOM::candelete", sc->candelete) != 0)
1210 return;
1211 if (sc->ident[0] != '\0' &&
1212 g_handleattr_str(bp, "GEOM::ident", sc->ident) != 0)
1213 return;
1214 if (g_handleattr_int(bp, "MNT::verified", (sc->flags & MD_VERIFY) != 0))
1215 return;
1216 g_io_deliver(bp, EOPNOTSUPP);
1217 }
1218
1219 static void
md_kthread(void * arg)1220 md_kthread(void *arg)
1221 {
1222 struct md_s *sc;
1223 struct bio *bp;
1224 int error;
1225
1226 sc = arg;
1227 thread_lock(curthread);
1228 sched_prio(curthread, PRIBIO);
1229 thread_unlock(curthread);
1230 if (sc->type == MD_VNODE)
1231 curthread->td_pflags |= TDP_NORUNNINGBUF;
1232
1233 for (;;) {
1234 mtx_lock(&sc->queue_mtx);
1235 if (sc->flags & MD_SHUTDOWN) {
1236 sc->flags |= MD_EXITING;
1237 mtx_unlock(&sc->queue_mtx);
1238 kproc_exit(0);
1239 }
1240 bp = bioq_takefirst(&sc->bio_queue);
1241 if (!bp) {
1242 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1243 continue;
1244 }
1245 mtx_unlock(&sc->queue_mtx);
1246 if (bp->bio_cmd == BIO_GETATTR) {
1247 md_handleattr(sc, bp);
1248 } else {
1249 error = sc->start(sc, bp);
1250 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
1251 /*
1252 * Devstat uses (bio_bcount, bio_resid) for
1253 * determining the length of the completed part
1254 * of the i/o. g_io_deliver() will translate
1255 * from bio_completed to that, but it also
1256 * destroys the bio so we must do our own
1257 * translation.
1258 */
1259 bp->bio_bcount = bp->bio_length;
1260 devstat_end_transaction_bio(sc->devstat, bp);
1261 }
1262 bp->bio_completed = bp->bio_length - bp->bio_resid;
1263 g_io_deliver(bp, error);
1264 }
1265 }
1266 }
1267
1268 static struct md_s *
mdfind(int unit)1269 mdfind(int unit)
1270 {
1271 struct md_s *sc;
1272
1273 LIST_FOREACH(sc, &md_softc_list, list) {
1274 if (sc->unit == unit)
1275 break;
1276 }
1277 return (sc);
1278 }
1279
1280 static struct md_s *
mdnew(int unit,int * errp,enum md_types type)1281 mdnew(int unit, int *errp, enum md_types type)
1282 {
1283 struct md_s *sc;
1284 int error;
1285
1286 *errp = 0;
1287 if (unit == -1)
1288 unit = alloc_unr(md_uh);
1289 else
1290 unit = alloc_unr_specific(md_uh, unit);
1291
1292 if (unit == -1) {
1293 *errp = EBUSY;
1294 return (NULL);
1295 }
1296
1297 sc = malloc(sizeof(*sc), M_MD, M_WAITOK | M_ZERO);
1298 sc->type = type;
1299 bioq_init(&sc->bio_queue);
1300 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1301 sc->unit = unit;
1302 sprintf(sc->name, "md%d", unit);
1303 LIST_INSERT_HEAD(&md_softc_list, sc, list);
1304 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1305 if (error == 0)
1306 return (sc);
1307 LIST_REMOVE(sc, list);
1308 mtx_destroy(&sc->queue_mtx);
1309 free_unr(md_uh, sc->unit);
1310 free(sc, M_MD);
1311 *errp = error;
1312 return (NULL);
1313 }
1314
1315 static void
mdinit(struct md_s * sc)1316 mdinit(struct md_s *sc)
1317 {
1318 struct g_geom *gp;
1319 struct g_provider *pp;
1320 unsigned remn;
1321
1322 g_topology_lock();
1323 gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1324 gp->softc = sc;
1325 pp = g_new_providerf(gp, "md%d", sc->unit);
1326 devstat_remove_entry(pp->stat);
1327 pp->stat = NULL;
1328 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1329 /* Prune off any residual fractional sector. */
1330 remn = sc->mediasize % sc->sectorsize;
1331 if (remn != 0) {
1332 printf("md%d: truncating fractional last sector by %u bytes\n",
1333 sc->unit, remn);
1334 sc->mediasize -= remn;
1335 }
1336 pp->mediasize = sc->mediasize;
1337 pp->sectorsize = sc->sectorsize;
1338 switch (sc->type) {
1339 case MD_MALLOC:
1340 case MD_VNODE:
1341 case MD_SWAP:
1342 pp->flags |= G_PF_ACCEPT_UNMAPPED;
1343 break;
1344 case MD_PRELOAD:
1345 case MD_NULL:
1346 break;
1347 }
1348 sc->gp = gp;
1349 sc->pp = pp;
1350 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1351 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
1352 sc->devstat->id = pp;
1353 g_error_provider(pp, 0);
1354 g_topology_unlock();
1355 }
1356
1357 static int
mdcreate_malloc(struct md_s * sc,struct md_req * mdr)1358 mdcreate_malloc(struct md_s *sc, struct md_req *mdr)
1359 {
1360 uintptr_t sp;
1361 int error;
1362 off_t u;
1363
1364 error = 0;
1365 if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1366 return (EINVAL);
1367 if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize))
1368 return (EINVAL);
1369 /* Compression doesn't make sense if we have reserved space */
1370 if (mdr->md_options & MD_RESERVE)
1371 mdr->md_options &= ~MD_COMPRESS;
1372 if (mdr->md_fwsectors != 0)
1373 sc->fwsectors = mdr->md_fwsectors;
1374 if (mdr->md_fwheads != 0)
1375 sc->fwheads = mdr->md_fwheads;
1376 sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE | MD_RESERVE);
1377 sc->s_malloc.indir = dimension(sc->mediasize / sc->sectorsize);
1378 sc->s_malloc.uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL,
1379 NULL, NULL, 0x1ff, 0);
1380 if (mdr->md_options & MD_RESERVE) {
1381 off_t nsectors;
1382
1383 nsectors = sc->mediasize / sc->sectorsize;
1384 for (u = 0; u < nsectors; u++) {
1385 sp = (uintptr_t)uma_zalloc(sc->s_malloc.uma,
1386 (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
1387 if (sp != 0)
1388 error = s_write(sc->s_malloc.indir, u, sp);
1389 else
1390 error = ENOMEM;
1391 if (error != 0)
1392 break;
1393 }
1394 }
1395 return (error);
1396 }
1397
1398 static int
mdsetcred(struct md_s * sc,struct ucred * cred)1399 mdsetcred(struct md_s *sc, struct ucred *cred)
1400 {
1401 char *tmpbuf;
1402 int error = 0;
1403
1404 /*
1405 * Set credits in our softc
1406 */
1407
1408 if (sc->cred)
1409 crfree(sc->cred);
1410 sc->cred = crhold(cred);
1411
1412 /*
1413 * Horrible kludge to establish credentials for NFS XXX.
1414 */
1415
1416 if (sc->type == MD_VNODE && sc->s_vnode.vnode != NULL) {
1417 struct uio auio;
1418 struct iovec aiov;
1419
1420 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1421 bzero(&auio, sizeof(auio));
1422
1423 aiov.iov_base = tmpbuf;
1424 aiov.iov_len = sc->sectorsize;
1425 auio.uio_iov = &aiov;
1426 auio.uio_iovcnt = 1;
1427 auio.uio_offset = 0;
1428 auio.uio_rw = UIO_READ;
1429 auio.uio_segflg = UIO_SYSSPACE;
1430 auio.uio_resid = aiov.iov_len;
1431 vn_lock(sc->s_vnode.vnode, LK_EXCLUSIVE | LK_RETRY);
1432 error = VOP_READ(sc->s_vnode.vnode, &auio, 0, sc->cred);
1433 VOP_UNLOCK(sc->s_vnode.vnode);
1434 free(tmpbuf, M_TEMP);
1435 }
1436 return (error);
1437 }
1438
1439 static int
mdcreate_vnode(struct md_s * sc,struct md_req * mdr,struct thread * td)1440 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1441 {
1442 struct vattr vattr;
1443 struct nameidata nd;
1444 char *fname;
1445 int error, flags;
1446 long v;
1447
1448 fname = mdr->md_file;
1449 if (mdr->md_file_seg == UIO_USERSPACE) {
1450 error = copyinstr(fname, sc->s_vnode.file,
1451 sizeof(sc->s_vnode.file), NULL);
1452 if (error != 0)
1453 return (error);
1454 } else if (mdr->md_file_seg == UIO_SYSSPACE)
1455 strlcpy(sc->s_vnode.file, fname, sizeof(sc->s_vnode.file));
1456 else
1457 return (EDOOFUS);
1458
1459 /*
1460 * If the user specified that this is a read only device, don't
1461 * set the FWRITE mask before trying to open the backing store.
1462 */
1463 flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \
1464 | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0);
1465 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->s_vnode.file);
1466 error = vn_open(&nd, &flags, 0, NULL);
1467 if (error != 0)
1468 return (error);
1469 NDFREE_PNBUF(&nd);
1470 if (nd.ni_vp->v_type != VREG) {
1471 error = EINVAL;
1472 goto bad;
1473 }
1474 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1475 if (error != 0)
1476 goto bad;
1477 if ((mdr->md_options & MD_MUSTDEALLOC) != 0) {
1478 error = VOP_PATHCONF(nd.ni_vp, _PC_DEALLOC_PRESENT, &v);
1479 if (error != 0)
1480 goto bad;
1481 if (v == 0)
1482 sc->candelete = false;
1483 }
1484 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1485 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1486 if (VN_IS_DOOMED(nd.ni_vp)) {
1487 /* Forced unmount. */
1488 error = EBADF;
1489 goto bad;
1490 }
1491 }
1492 nd.ni_vp->v_vflag |= VV_MD;
1493 VOP_UNLOCK(nd.ni_vp);
1494
1495 if (mdr->md_fwsectors != 0)
1496 sc->fwsectors = mdr->md_fwsectors;
1497 if (mdr->md_fwheads != 0)
1498 sc->fwheads = mdr->md_fwheads;
1499 snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1500 (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1501 sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE |
1502 MD_VERIFY | MD_MUSTDEALLOC);
1503 if (!(flags & FWRITE))
1504 sc->flags |= MD_READONLY;
1505 sc->s_vnode.vnode = nd.ni_vp;
1506
1507 error = mdsetcred(sc, td->td_ucred);
1508 if (error != 0) {
1509 sc->s_vnode.vnode = NULL;
1510 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1511 nd.ni_vp->v_vflag &= ~VV_MD;
1512 goto bad;
1513 }
1514
1515 sc->s_vnode.kva = kva_alloc(maxphys + PAGE_SIZE);
1516 return (0);
1517 bad:
1518 VOP_UNLOCK(nd.ni_vp);
1519 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1520 return (error);
1521 }
1522
1523 static void
g_md_providergone(struct g_provider * pp)1524 g_md_providergone(struct g_provider *pp)
1525 {
1526 struct md_s *sc = pp->geom->softc;
1527
1528 mtx_lock(&sc->queue_mtx);
1529 sc->flags |= MD_PROVIDERGONE;
1530 wakeup(&sc->flags);
1531 mtx_unlock(&sc->queue_mtx);
1532 }
1533
1534 static int
mddestroy(struct md_s * sc,struct thread * td)1535 mddestroy(struct md_s *sc, struct thread *td)
1536 {
1537
1538 if (sc->gp) {
1539 g_topology_lock();
1540 g_wither_geom(sc->gp, ENXIO);
1541 g_topology_unlock();
1542
1543 mtx_lock(&sc->queue_mtx);
1544 while (!(sc->flags & MD_PROVIDERGONE))
1545 msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0);
1546 mtx_unlock(&sc->queue_mtx);
1547 }
1548 if (sc->devstat) {
1549 devstat_remove_entry(sc->devstat);
1550 sc->devstat = NULL;
1551 }
1552 mtx_lock(&sc->queue_mtx);
1553 sc->flags |= MD_SHUTDOWN;
1554 wakeup(sc);
1555 while (!(sc->flags & MD_EXITING))
1556 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1557 mtx_unlock(&sc->queue_mtx);
1558 mtx_destroy(&sc->queue_mtx);
1559 switch (sc->type) {
1560 case MD_VNODE:
1561 if (sc->s_vnode.vnode != NULL) {
1562 vn_lock(sc->s_vnode.vnode, LK_EXCLUSIVE | LK_RETRY);
1563 sc->s_vnode.vnode->v_vflag &= ~VV_MD;
1564 VOP_UNLOCK(sc->s_vnode.vnode);
1565 (void)vn_close(sc->s_vnode.vnode,
1566 sc->flags & MD_READONLY ? FREAD : (FREAD|FWRITE),
1567 sc->cred, td);
1568 }
1569 if (sc->s_vnode.kva != NULL)
1570 kva_free(sc->s_vnode.kva, maxphys + PAGE_SIZE);
1571 break;
1572 case MD_SWAP:
1573 if (sc->s_swap.object != NULL)
1574 vm_object_deallocate(sc->s_swap.object);
1575 break;
1576 case MD_MALLOC:
1577 if (sc->s_malloc.indir != NULL)
1578 destroy_indir(sc, sc->s_malloc.indir);
1579 if (sc->s_malloc.uma != NULL)
1580 uma_zdestroy(sc->s_malloc.uma);
1581 break;
1582 case MD_PRELOAD:
1583 case MD_NULL:
1584 break;
1585 default:
1586 __assert_unreachable();
1587 }
1588 if (sc->cred != NULL)
1589 crfree(sc->cred);
1590
1591 LIST_REMOVE(sc, list);
1592 free_unr(md_uh, sc->unit);
1593 free(sc, M_MD);
1594 return (0);
1595 }
1596
1597 static int
mdresize(struct md_s * sc,struct md_req * mdr)1598 mdresize(struct md_s *sc, struct md_req *mdr)
1599 {
1600 int error, res;
1601 vm_pindex_t oldpages, newpages;
1602
1603 switch (sc->type) {
1604 case MD_VNODE:
1605 case MD_NULL:
1606 break;
1607 case MD_SWAP:
1608 if (mdr->md_mediasize <= 0 ||
1609 (mdr->md_mediasize % PAGE_SIZE) != 0)
1610 return (EDOM);
1611 oldpages = OFF_TO_IDX(sc->mediasize);
1612 newpages = OFF_TO_IDX(mdr->md_mediasize);
1613 if (newpages < oldpages) {
1614 VM_OBJECT_WLOCK(sc->s_swap.object);
1615 vm_object_page_remove(sc->s_swap.object, newpages,
1616 0, 0);
1617 swap_release_by_cred(IDX_TO_OFF(oldpages -
1618 newpages), sc->cred);
1619 sc->s_swap.object->size = newpages;
1620 VM_OBJECT_WUNLOCK(sc->s_swap.object);
1621 } else if (newpages > oldpages) {
1622 res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1623 oldpages), sc->cred);
1624 if (!res)
1625 return (ENOMEM);
1626 if ((mdr->md_options & MD_RESERVE) ||
1627 (sc->flags & MD_RESERVE)) {
1628 error = swap_pager_reserve(sc->s_swap.object,
1629 oldpages, newpages - oldpages);
1630 if (error < 0) {
1631 swap_release_by_cred(
1632 IDX_TO_OFF(newpages - oldpages),
1633 sc->cred);
1634 return (EDOM);
1635 }
1636 }
1637 VM_OBJECT_WLOCK(sc->s_swap.object);
1638 sc->s_swap.object->size = newpages;
1639 VM_OBJECT_WUNLOCK(sc->s_swap.object);
1640 }
1641 break;
1642 default:
1643 return (EOPNOTSUPP);
1644 }
1645
1646 sc->mediasize = mdr->md_mediasize;
1647
1648 g_topology_lock();
1649 g_resize_provider(sc->pp, sc->mediasize);
1650 g_topology_unlock();
1651 return (0);
1652 }
1653
1654 static int
mdcreate_swap(struct md_s * sc,struct md_req * mdr,struct thread * td)1655 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1656 {
1657 vm_ooffset_t npage;
1658 int error;
1659
1660 /*
1661 * Range check. Disallow negative sizes and sizes not being
1662 * multiple of page size.
1663 */
1664 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1665 return (EDOM);
1666
1667 /*
1668 * Allocate an OBJT_SWAP object.
1669 *
1670 * Note the truncation.
1671 */
1672
1673 if ((mdr->md_options & MD_VERIFY) != 0)
1674 return (EINVAL);
1675 npage = mdr->md_mediasize / PAGE_SIZE;
1676 if (mdr->md_fwsectors != 0)
1677 sc->fwsectors = mdr->md_fwsectors;
1678 if (mdr->md_fwheads != 0)
1679 sc->fwheads = mdr->md_fwheads;
1680 sc->s_swap.object = vm_pager_allocate(OBJT_SWAP, NULL,
1681 PAGE_SIZE * npage, VM_PROT_DEFAULT, 0, td->td_ucred);
1682 if (sc->s_swap.object == NULL)
1683 return (ENOMEM);
1684 sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1685 if (mdr->md_options & MD_RESERVE) {
1686 if (swap_pager_reserve(sc->s_swap.object, 0, npage) < 0) {
1687 error = EDOM;
1688 goto finish;
1689 }
1690 }
1691 error = mdsetcred(sc, td->td_ucred);
1692 finish:
1693 if (error != 0) {
1694 vm_object_deallocate(sc->s_swap.object);
1695 sc->s_swap.object = NULL;
1696 }
1697 return (error);
1698 }
1699
1700 static int
mdcreate_null(struct md_s * sc,struct md_req * mdr,struct thread * td)1701 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1702 {
1703
1704 /*
1705 * Range check. Disallow negative sizes and sizes not being
1706 * multiple of page size.
1707 */
1708 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1709 return (EDOM);
1710
1711 return (0);
1712 }
1713
1714 static int
kern_mdattach_locked(struct thread * td,struct md_req * mdr)1715 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1716 {
1717 struct md_s *sc;
1718 unsigned sectsize;
1719 int error;
1720
1721 sx_assert(&md_sx, SA_XLOCKED);
1722
1723 switch (mdr->md_type) {
1724 case MD_MALLOC:
1725 case MD_PRELOAD:
1726 case MD_VNODE:
1727 case MD_SWAP:
1728 case MD_NULL:
1729 break;
1730 default:
1731 return (EINVAL);
1732 }
1733 if (mdr->md_sectorsize == 0)
1734 sectsize = DEV_BSIZE;
1735 else
1736 sectsize = mdr->md_sectorsize;
1737 if (sectsize > maxphys || mdr->md_mediasize < sectsize)
1738 return (EINVAL);
1739 if (mdr->md_options & MD_AUTOUNIT)
1740 sc = mdnew(-1, &error, mdr->md_type);
1741 else {
1742 if (mdr->md_unit > INT_MAX)
1743 return (EINVAL);
1744 sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1745 }
1746 if (sc == NULL)
1747 return (error);
1748 if (mdr->md_label != NULL)
1749 error = copyinstr(mdr->md_label, sc->label,
1750 sizeof(sc->label), NULL);
1751 if (error != 0)
1752 goto err_after_new;
1753 if (mdr->md_options & MD_AUTOUNIT)
1754 mdr->md_unit = sc->unit;
1755 sc->mediasize = mdr->md_mediasize;
1756 sc->sectorsize = sectsize;
1757 sc->candelete = true;
1758 error = EDOOFUS;
1759 switch (sc->type) {
1760 case MD_MALLOC:
1761 sc->start = mdstart_malloc;
1762 error = mdcreate_malloc(sc, mdr);
1763 break;
1764 case MD_PRELOAD:
1765 /*
1766 * We disallow attaching preloaded memory disks via
1767 * ioctl. Preloaded memory disks are automatically
1768 * attached in g_md_init().
1769 */
1770 error = EOPNOTSUPP;
1771 break;
1772 case MD_VNODE:
1773 sc->start = mdstart_vnode;
1774 error = mdcreate_vnode(sc, mdr, td);
1775 break;
1776 case MD_SWAP:
1777 sc->start = mdstart_swap;
1778 error = mdcreate_swap(sc, mdr, td);
1779 break;
1780 case MD_NULL:
1781 sc->start = mdstart_null;
1782 error = mdcreate_null(sc, mdr, td);
1783 break;
1784 }
1785 err_after_new:
1786 if (error != 0) {
1787 mddestroy(sc, td);
1788 return (error);
1789 }
1790
1791 mdinit(sc);
1792 return (0);
1793 }
1794
1795 static int
kern_mdattach(struct thread * td,struct md_req * mdr)1796 kern_mdattach(struct thread *td, struct md_req *mdr)
1797 {
1798 int error;
1799
1800 sx_xlock(&md_sx);
1801 error = kern_mdattach_locked(td, mdr);
1802 sx_xunlock(&md_sx);
1803 return (error);
1804 }
1805
1806 static int
kern_mddetach_locked(struct thread * td,struct md_req * mdr)1807 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1808 {
1809 struct md_s *sc;
1810
1811 sx_assert(&md_sx, SA_XLOCKED);
1812
1813 if (mdr->md_mediasize != 0 ||
1814 (mdr->md_options & ~MD_FORCE) != 0)
1815 return (EINVAL);
1816
1817 sc = mdfind(mdr->md_unit);
1818 if (sc == NULL)
1819 return (ENOENT);
1820 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1821 !(mdr->md_options & MD_FORCE))
1822 return (EBUSY);
1823 return (mddestroy(sc, td));
1824 }
1825
1826 static int
kern_mddetach(struct thread * td,struct md_req * mdr)1827 kern_mddetach(struct thread *td, struct md_req *mdr)
1828 {
1829 int error;
1830
1831 sx_xlock(&md_sx);
1832 error = kern_mddetach_locked(td, mdr);
1833 sx_xunlock(&md_sx);
1834 return (error);
1835 }
1836
1837 static int
kern_mdresize_locked(struct md_req * mdr)1838 kern_mdresize_locked(struct md_req *mdr)
1839 {
1840 struct md_s *sc;
1841
1842 sx_assert(&md_sx, SA_XLOCKED);
1843
1844 if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1845 return (EINVAL);
1846
1847 sc = mdfind(mdr->md_unit);
1848 if (sc == NULL)
1849 return (ENOENT);
1850 if (mdr->md_mediasize < sc->sectorsize)
1851 return (EINVAL);
1852 mdr->md_mediasize -= mdr->md_mediasize % sc->sectorsize;
1853 if (mdr->md_mediasize < sc->mediasize &&
1854 !(sc->flags & MD_FORCE) &&
1855 !(mdr->md_options & MD_FORCE))
1856 return (EBUSY);
1857 return (mdresize(sc, mdr));
1858 }
1859
1860 static int
kern_mdresize(struct md_req * mdr)1861 kern_mdresize(struct md_req *mdr)
1862 {
1863 int error;
1864
1865 sx_xlock(&md_sx);
1866 error = kern_mdresize_locked(mdr);
1867 sx_xunlock(&md_sx);
1868 return (error);
1869 }
1870
1871 static int
kern_mdquery_locked(struct md_req * mdr)1872 kern_mdquery_locked(struct md_req *mdr)
1873 {
1874 struct md_s *sc;
1875 int error;
1876
1877 sx_assert(&md_sx, SA_XLOCKED);
1878
1879 sc = mdfind(mdr->md_unit);
1880 if (sc == NULL)
1881 return (ENOENT);
1882 mdr->md_type = sc->type;
1883 mdr->md_options = sc->flags;
1884 mdr->md_mediasize = sc->mediasize;
1885 mdr->md_sectorsize = sc->sectorsize;
1886 error = 0;
1887 if (mdr->md_label != NULL) {
1888 error = copyout(sc->label, mdr->md_label,
1889 strlen(sc->label) + 1);
1890 if (error != 0)
1891 return (error);
1892 }
1893 if (sc->type == MD_VNODE) {
1894 error = copyout(sc->s_vnode.file, mdr->md_file,
1895 strlen(sc->s_vnode.file) + 1);
1896 } else if (sc->type == MD_PRELOAD && mdr->md_file != NULL) {
1897 error = copyout(sc->s_preload.name, mdr->md_file,
1898 strlen(sc->s_preload.name) + 1);
1899 }
1900 return (error);
1901 }
1902
1903 static int
kern_mdquery(struct md_req * mdr)1904 kern_mdquery(struct md_req *mdr)
1905 {
1906 int error;
1907
1908 sx_xlock(&md_sx);
1909 error = kern_mdquery_locked(mdr);
1910 sx_xunlock(&md_sx);
1911 return (error);
1912 }
1913
1914 /* Copy members that are not userspace pointers. */
1915 #define MD_IOCTL2REQ(mdio, mdr) do { \
1916 (mdr)->md_unit = (mdio)->md_unit; \
1917 (mdr)->md_type = (mdio)->md_type; \
1918 (mdr)->md_mediasize = (mdio)->md_mediasize; \
1919 (mdr)->md_sectorsize = (mdio)->md_sectorsize; \
1920 (mdr)->md_options = (mdio)->md_options; \
1921 (mdr)->md_fwheads = (mdio)->md_fwheads; \
1922 (mdr)->md_fwsectors = (mdio)->md_fwsectors; \
1923 (mdr)->md_units = &(mdio)->md_pad[0]; \
1924 (mdr)->md_units_nitems = nitems((mdio)->md_pad); \
1925 } while(0)
1926
1927 /* Copy members that might have been updated */
1928 #define MD_REQ2IOCTL(mdr, mdio) do { \
1929 (mdio)->md_unit = (mdr)->md_unit; \
1930 (mdio)->md_type = (mdr)->md_type; \
1931 (mdio)->md_mediasize = (mdr)->md_mediasize; \
1932 (mdio)->md_sectorsize = (mdr)->md_sectorsize; \
1933 (mdio)->md_options = (mdr)->md_options; \
1934 (mdio)->md_fwheads = (mdr)->md_fwheads; \
1935 (mdio)->md_fwsectors = (mdr)->md_fwsectors; \
1936 } while(0)
1937
1938 static int
mdctlioctl(struct cdev * dev,u_long cmd,caddr_t addr,int flags,struct thread * td)1939 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1940 struct thread *td)
1941 {
1942 struct md_req mdr;
1943 int error;
1944
1945 if (md_debug)
1946 printf("mdctlioctl(%s %lx %p %x %p)\n",
1947 devtoname(dev), cmd, addr, flags, td);
1948
1949 bzero(&mdr, sizeof(mdr));
1950 switch (cmd) {
1951 case MDIOCATTACH:
1952 case MDIOCDETACH:
1953 case MDIOCRESIZE:
1954 case MDIOCQUERY: {
1955 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1956 if (mdio->md_version != MDIOVERSION)
1957 return (EINVAL);
1958 MD_IOCTL2REQ(mdio, &mdr);
1959 mdr.md_file = mdio->md_file;
1960 mdr.md_file_seg = UIO_USERSPACE;
1961 /* If the file is adjacent to the md_ioctl it's in kernel. */
1962 if ((void *)mdio->md_file == (void *)(mdio + 1))
1963 mdr.md_file_seg = UIO_SYSSPACE;
1964 mdr.md_label = mdio->md_label;
1965 break;
1966 }
1967 #ifdef COMPAT_FREEBSD32
1968 case MDIOCATTACH_32:
1969 case MDIOCDETACH_32:
1970 case MDIOCRESIZE_32:
1971 case MDIOCQUERY_32: {
1972 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1973 if (mdio->md_version != MDIOVERSION)
1974 return (EINVAL);
1975 MD_IOCTL2REQ(mdio, &mdr);
1976 mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1977 mdr.md_file_seg = UIO_USERSPACE;
1978 mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1979 break;
1980 }
1981 #endif
1982 default:
1983 /* Fall through to handler switch. */
1984 break;
1985 }
1986
1987 error = 0;
1988 switch (cmd) {
1989 case MDIOCATTACH:
1990 #ifdef COMPAT_FREEBSD32
1991 case MDIOCATTACH_32:
1992 #endif
1993 error = kern_mdattach(td, &mdr);
1994 break;
1995 case MDIOCDETACH:
1996 #ifdef COMPAT_FREEBSD32
1997 case MDIOCDETACH_32:
1998 #endif
1999 error = kern_mddetach(td, &mdr);
2000 break;
2001 case MDIOCRESIZE:
2002 #ifdef COMPAT_FREEBSD32
2003 case MDIOCRESIZE_32:
2004 #endif
2005 error = kern_mdresize(&mdr);
2006 break;
2007 case MDIOCQUERY:
2008 #ifdef COMPAT_FREEBSD32
2009 case MDIOCQUERY_32:
2010 #endif
2011 error = kern_mdquery(&mdr);
2012 break;
2013 default:
2014 error = ENOIOCTL;
2015 }
2016
2017 switch (cmd) {
2018 case MDIOCATTACH:
2019 case MDIOCQUERY: {
2020 struct md_ioctl *mdio = (struct md_ioctl *)addr;
2021 MD_REQ2IOCTL(&mdr, mdio);
2022 break;
2023 }
2024 #ifdef COMPAT_FREEBSD32
2025 case MDIOCATTACH_32:
2026 case MDIOCQUERY_32: {
2027 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
2028 MD_REQ2IOCTL(&mdr, mdio);
2029 break;
2030 }
2031 #endif
2032 default:
2033 /* Other commands to not alter mdr. */
2034 break;
2035 }
2036
2037 return (error);
2038 }
2039
2040 static void
md_preloaded(u_char * image,size_t length,const char * name)2041 md_preloaded(u_char *image, size_t length, const char *name)
2042 {
2043 struct md_s *sc;
2044 int error;
2045
2046 sc = mdnew(-1, &error, MD_PRELOAD);
2047 if (sc == NULL)
2048 return;
2049 sc->mediasize = length;
2050 sc->sectorsize = DEV_BSIZE;
2051 sc->s_preload.pl_ptr = image;
2052 sc->s_preload.pl_len = length;
2053 sc->start = mdstart_preload;
2054 if (name != NULL)
2055 strlcpy(sc->s_preload.name, name,
2056 sizeof(sc->s_preload.name));
2057 #ifdef MD_ROOT
2058 if (sc->unit == 0) {
2059 #ifndef ROOTDEVNAME
2060 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2061 #endif
2062 #ifdef MD_ROOT_READONLY
2063 sc->flags |= MD_READONLY;
2064 #endif
2065 }
2066 #endif
2067 mdinit(sc);
2068 if (name != NULL) {
2069 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2070 MD_NAME, sc->unit, name, length, image);
2071 } else {
2072 printf("%s%d: Embedded image %zd bytes at %p\n",
2073 MD_NAME, sc->unit, length, image);
2074 }
2075 }
2076
2077 static void
g_md_init(struct g_class * mp __unused)2078 g_md_init(struct g_class *mp __unused)
2079 {
2080 caddr_t mod;
2081 u_char *ptr, *name, *type;
2082 u_char scratch[40];
2083 unsigned len;
2084 int i;
2085 vm_offset_t paddr;
2086
2087 /* figure out log2(NINDIR) */
2088 for (i = NINDIR, nshift = -1; i; nshift++)
2089 i >>= 1;
2090
2091 mod = NULL;
2092 sx_init(&md_sx, "MD config lock");
2093 g_topology_unlock();
2094 md_uh = new_unrhdr(0, INT_MAX, NULL);
2095 #ifdef MD_ROOT
2096 if (mfs_root_size != 0) {
2097 sx_xlock(&md_sx);
2098 #ifdef MD_ROOT_MEM
2099 md_preloaded(mfs_root, mfs_root_size, NULL);
2100 #else
2101 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2102 NULL);
2103 #endif
2104 sx_xunlock(&md_sx);
2105 }
2106 #endif
2107 /* XXX: are preload_* static or do they need Giant ? */
2108 while ((mod = preload_search_next_name(mod)) != NULL) {
2109 name = (char *)preload_search_info(mod, MODINFO_NAME);
2110 if (name == NULL)
2111 continue;
2112 type = (char *)preload_search_info(mod, MODINFO_TYPE);
2113 if (type == NULL)
2114 continue;
2115 if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2116 continue;
2117 ptr = preload_fetch_addr(mod);
2118 len = preload_fetch_size(mod);
2119 if (ptr != NULL && len != 0) {
2120 sx_xlock(&md_sx);
2121 md_preloaded(ptr, len, name);
2122 sx_xunlock(&md_sx);
2123 }
2124 }
2125
2126 /*
2127 * Load up to 32 pre-loaded disks
2128 */
2129 for (int i = 0; i < 32; i++) {
2130 if (resource_long_value("md", i, "physaddr",
2131 (long *) &paddr) != 0 ||
2132 resource_int_value("md", i, "len", &len) != 0)
2133 break;
2134 ptr = pmap_map(NULL, paddr, paddr + len, VM_PROT_READ);
2135 if (ptr != NULL && len != 0) {
2136 sprintf(scratch, "preload%d 0x%016jx", i,
2137 (uintmax_t)paddr);
2138 sx_xlock(&md_sx);
2139 md_preloaded(ptr, len, scratch);
2140 sx_xunlock(&md_sx);
2141 }
2142 }
2143
2144 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2145 0600, MDCTL_NAME);
2146 g_topology_lock();
2147 }
2148
2149 static void
g_md_dumpconf(struct sbuf * sb,const char * indent,struct g_geom * gp,struct g_consumer * cp __unused,struct g_provider * pp)2150 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2151 struct g_consumer *cp __unused, struct g_provider *pp)
2152 {
2153 struct md_s *mp;
2154 char *type;
2155
2156 mp = gp->softc;
2157 if (mp == NULL)
2158 return;
2159
2160 switch (mp->type) {
2161 case MD_MALLOC:
2162 type = "malloc";
2163 break;
2164 case MD_PRELOAD:
2165 type = "preload";
2166 break;
2167 case MD_VNODE:
2168 type = "vnode";
2169 break;
2170 case MD_SWAP:
2171 type = "swap";
2172 break;
2173 case MD_NULL:
2174 type = "null";
2175 break;
2176 default:
2177 type = "unknown";
2178 break;
2179 }
2180
2181 if (pp != NULL) {
2182 if (indent == NULL) {
2183 sbuf_printf(sb, " u %d", mp->unit);
2184 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2185 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2186 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2187 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2188 sbuf_printf(sb, " t %s", type);
2189 if (mp->type == MD_VNODE &&
2190 mp->s_vnode.vnode != NULL)
2191 sbuf_printf(sb, " file %s", mp->s_vnode.file);
2192 if (mp->type == MD_PRELOAD &&
2193 mp->s_preload.name[0] != '\0') {
2194 sbuf_printf(sb, " file %s",
2195 mp->s_preload.name);
2196 }
2197 sbuf_printf(sb, " label %s", mp->label);
2198 } else {
2199 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2200 mp->unit);
2201 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2202 indent, (uintmax_t) mp->sectorsize);
2203 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2204 indent, (uintmax_t) mp->fwheads);
2205 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2206 indent, (uintmax_t) mp->fwsectors);
2207 if (mp->ident[0] != '\0') {
2208 sbuf_printf(sb, "%s<ident>", indent);
2209 g_conf_printf_escaped(sb, "%s", mp->ident);
2210 sbuf_printf(sb, "</ident>\n");
2211 }
2212 sbuf_printf(sb, "%s<length>%ju</length>\n",
2213 indent, (uintmax_t) mp->mediasize);
2214 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2215 (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2216 sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2217 (mp->flags & MD_READONLY) == 0 ? "read-write":
2218 "read-only");
2219 sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2220 type);
2221 if (mp->type == MD_VNODE) {
2222 if (mp->s_vnode.vnode != NULL) {
2223 sbuf_printf(sb, "%s<file>", indent);
2224 g_conf_printf_escaped(sb, "%s",
2225 mp->s_vnode.file);
2226 sbuf_printf(sb, "</file>\n");
2227 }
2228 sbuf_printf(sb, "%s<cache>%s</cache>\n", indent,
2229 (mp->flags & MD_CACHE) == 0 ? "off": "on");
2230 }
2231 if (mp->type == MD_PRELOAD &&
2232 mp->s_preload.name[0] != '\0') {
2233 sbuf_printf(sb, "%s<file>", indent);
2234 g_conf_printf_escaped(sb, "%s",
2235 mp->s_preload.name);
2236 sbuf_printf(sb, "</file>\n");
2237 }
2238 sbuf_printf(sb, "%s<label>", indent);
2239 g_conf_printf_escaped(sb, "%s", mp->label);
2240 sbuf_printf(sb, "</label>\n");
2241 }
2242 }
2243 }
2244
2245 static void
g_md_fini(struct g_class * mp __unused)2246 g_md_fini(struct g_class *mp __unused)
2247 {
2248
2249 sx_destroy(&md_sx);
2250 if (status_dev != NULL)
2251 destroy_dev(status_dev);
2252 delete_unrhdr(md_uh);
2253 }
2254