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 vn(4) driver: mdstart_swap(),
15 * mdstart_vnode(), mdcreate_swap(), mdcreate_vnode() and mddestroy(),
16 * 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/param.h>
64 #include <sys/systm.h>
65 #include <sys/bio.h>
66 #include <sys/buf.h>
67 #include <sys/conf.h>
68 #include <sys/devicestat.h>
69 #include <sys/fcntl.h>
70 #include <sys/kernel.h>
71 #include <sys/kthread.h>
72 #include <sys/limits.h>
73 #include <sys/linker.h>
74 #include <sys/lock.h>
75 #include <sys/malloc.h>
76 #include <sys/mdioctl.h>
77 #include <sys/mount.h>
78 #include <sys/mutex.h>
79 #include <sys/sx.h>
80 #include <sys/namei.h>
81 #include <sys/proc.h>
82 #include <sys/queue.h>
83 #include <sys/rwlock.h>
84 #include <sys/sbuf.h>
85 #include <sys/sched.h>
86 #include <sys/sf_buf.h>
87 #include <sys/sysctl.h>
88 #include <sys/uio.h>
89 #include <sys/unistd.h>
90 #include <sys/vnode.h>
91 #include <sys/disk.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 vm_offset_t 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 = (void *)(sc->s_vnode.kva + (ma_offs &
987 PAGE_MASK));
988 aiov.iov_len = iolen;
989 auio.uio_iov = &aiov;
990 auio.uio_iovcnt = 1;
991 auio.uio_resid = iolen;
992 mapped = true;
993 } else {
994 aiov.iov_base = bp->bio_data;
995 aiov.iov_len = bp->bio_length;
996 auio.uio_iov = &aiov;
997 auio.uio_iovcnt = 1;
998 }
999 iostart = auio.uio_offset;
1000 if (bp->bio_cmd == BIO_READ) {
1001 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1002 error = VOP_READ(vp, &auio, 0, sc->cred);
1003 VOP_UNLOCK(vp);
1004 } else {
1005 (void) vn_start_write(vp, &mp, V_WAIT);
1006 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1007 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
1008 sc->cred);
1009 VOP_UNLOCK(vp);
1010 vn_finished_write(mp);
1011 if (error == 0)
1012 sc->flags &= ~MD_VERIFY;
1013 }
1014
1015 /* When MD_CACHE is set, try to avoid double-caching the data. */
1016 if (error == 0 && (sc->flags & MD_CACHE) == 0)
1017 VOP_ADVISE(vp, iostart, auio.uio_offset - 1,
1018 POSIX_FADV_DONTNEED);
1019
1020 if (mapped) {
1021 pmap_qremove(sc->s_vnode.kva, npages);
1022 if (error == 0) {
1023 len -= iolen;
1024 bp->bio_resid -= iolen;
1025 ma_offs += iolen;
1026 if (len > 0)
1027 goto unmapped_step;
1028 }
1029 } else {
1030 bp->bio_resid = auio.uio_resid;
1031 }
1032
1033 free(piov, M_MD);
1034 return (error);
1035 }
1036
1037 static int
mdstart_swap(struct md_s * sc,struct bio * bp)1038 mdstart_swap(struct md_s *sc, struct bio *bp)
1039 {
1040 vm_page_t m;
1041 u_char *p;
1042 vm_pindex_t i, lastp;
1043 bus_dma_segment_t *vlist;
1044 int rv, ma_offs, offs, len, lastend;
1045
1046 switch (bp->bio_cmd) {
1047 case BIO_READ:
1048 case BIO_WRITE:
1049 case BIO_DELETE:
1050 break;
1051 case BIO_FLUSH:
1052 return (0);
1053 default:
1054 return (EOPNOTSUPP);
1055 }
1056
1057 p = bp->bio_data;
1058 ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ?
1059 bp->bio_ma_offset : 0;
1060 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
1061 (bus_dma_segment_t *)bp->bio_data : NULL;
1062
1063 /*
1064 * offs is the offset at which to start operating on the
1065 * next (ie, first) page. lastp is the last page on
1066 * which we're going to operate. lastend is the ending
1067 * position within that last page (ie, PAGE_SIZE if
1068 * we're operating on complete aligned pages).
1069 */
1070 offs = bp->bio_offset % PAGE_SIZE;
1071 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
1072 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
1073
1074 rv = VM_PAGER_OK;
1075 vm_object_pip_add(sc->s_swap.object, 1);
1076 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
1077 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
1078 m = vm_page_grab_unlocked(sc->s_swap.object, i,
1079 VM_ALLOC_SYSTEM);
1080 if (bp->bio_cmd == BIO_READ) {
1081 if (vm_page_all_valid(m))
1082 rv = VM_PAGER_OK;
1083 else
1084 rv = vm_pager_get_pages(sc->s_swap.object,
1085 &m, 1, NULL, NULL);
1086 if (rv == VM_PAGER_ERROR) {
1087 VM_OBJECT_WLOCK(sc->s_swap.object);
1088 vm_page_free(m);
1089 VM_OBJECT_WUNLOCK(sc->s_swap.object);
1090 break;
1091 } else if (rv == VM_PAGER_FAIL) {
1092 /*
1093 * Pager does not have the page. Zero
1094 * the allocated page, and mark it as
1095 * valid. Do not set dirty, the page
1096 * can be recreated if thrown out.
1097 */
1098 pmap_zero_page(m);
1099 vm_page_valid(m);
1100 }
1101 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1102 pmap_copy_pages(&m, offs, bp->bio_ma,
1103 ma_offs, len);
1104 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1105 physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs,
1106 vlist, ma_offs, len);
1107 cpu_flush_dcache(p, len);
1108 } else {
1109 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1110 cpu_flush_dcache(p, len);
1111 }
1112 } else if (bp->bio_cmd == BIO_WRITE) {
1113 if (len == PAGE_SIZE || vm_page_all_valid(m))
1114 rv = VM_PAGER_OK;
1115 else
1116 rv = vm_pager_get_pages(sc->s_swap.object,
1117 &m, 1, NULL, NULL);
1118 if (rv == VM_PAGER_ERROR) {
1119 VM_OBJECT_WLOCK(sc->s_swap.object);
1120 vm_page_free(m);
1121 VM_OBJECT_WUNLOCK(sc->s_swap.object);
1122 break;
1123 } else if (rv == VM_PAGER_FAIL)
1124 pmap_zero_page(m);
1125
1126 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1127 pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1128 offs, len);
1129 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1130 physcopyin_vlist(vlist, ma_offs,
1131 VM_PAGE_TO_PHYS(m) + offs, len);
1132 } else {
1133 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1134 }
1135
1136 vm_page_valid(m);
1137 vm_page_set_dirty(m);
1138 } else if (bp->bio_cmd == BIO_DELETE) {
1139 if (len == PAGE_SIZE || vm_page_all_valid(m))
1140 rv = VM_PAGER_OK;
1141 else
1142 rv = vm_pager_get_pages(sc->s_swap.object,
1143 &m, 1, NULL, NULL);
1144 VM_OBJECT_WLOCK(sc->s_swap.object);
1145 if (rv == VM_PAGER_ERROR) {
1146 vm_page_free(m);
1147 VM_OBJECT_WUNLOCK(sc->s_swap.object);
1148 break;
1149 } else if (rv == VM_PAGER_FAIL) {
1150 vm_page_free(m);
1151 m = NULL;
1152 } else {
1153 /* Page is valid. */
1154 if (len != PAGE_SIZE) {
1155 pmap_zero_page_area(m, offs, len);
1156 vm_page_set_dirty(m);
1157 } else {
1158 vm_pager_page_unswapped(m);
1159 vm_page_free(m);
1160 m = NULL;
1161 }
1162 }
1163 VM_OBJECT_WUNLOCK(sc->s_swap.object);
1164 }
1165 if (m != NULL) {
1166 /*
1167 * The page may be deactivated prior to setting
1168 * PGA_REFERENCED, but in this case it will be
1169 * reactivated by the page daemon.
1170 */
1171 if (vm_page_active(m))
1172 vm_page_reference(m);
1173 else
1174 vm_page_activate(m);
1175 vm_page_xunbusy(m);
1176 }
1177
1178 /* Actions on further pages start at offset 0 */
1179 p += PAGE_SIZE - offs;
1180 offs = 0;
1181 ma_offs += len;
1182 }
1183 vm_object_pip_wakeup(sc->s_swap.object);
1184 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
1185 }
1186
1187 static int
mdstart_null(struct md_s * sc,struct bio * bp)1188 mdstart_null(struct md_s *sc, struct bio *bp)
1189 {
1190
1191 switch (bp->bio_cmd) {
1192 case BIO_READ:
1193 bzero(bp->bio_data, bp->bio_length);
1194 cpu_flush_dcache(bp->bio_data, bp->bio_length);
1195 break;
1196 case BIO_WRITE:
1197 break;
1198 }
1199 bp->bio_resid = 0;
1200 return (0);
1201 }
1202
1203 static void
md_handleattr(struct md_s * sc,struct bio * bp)1204 md_handleattr(struct md_s *sc, struct bio *bp)
1205 {
1206 if (sc->fwsectors && sc->fwheads &&
1207 (g_handleattr_int(bp, "GEOM::fwsectors", sc->fwsectors) != 0 ||
1208 g_handleattr_int(bp, "GEOM::fwheads", sc->fwheads) != 0))
1209 return;
1210 if (g_handleattr_int(bp, "GEOM::candelete", sc->candelete) != 0)
1211 return;
1212 if (sc->ident[0] != '\0' &&
1213 g_handleattr_str(bp, "GEOM::ident", sc->ident) != 0)
1214 return;
1215 if (g_handleattr_int(bp, "MNT::verified", (sc->flags & MD_VERIFY) != 0))
1216 return;
1217 g_io_deliver(bp, EOPNOTSUPP);
1218 }
1219
1220 static void
md_kthread(void * arg)1221 md_kthread(void *arg)
1222 {
1223 struct md_s *sc;
1224 struct bio *bp;
1225 int error;
1226
1227 sc = arg;
1228 thread_lock(curthread);
1229 sched_prio(curthread, PRIBIO);
1230 thread_unlock(curthread);
1231 if (sc->type == MD_VNODE)
1232 curthread->td_pflags |= TDP_NORUNNINGBUF;
1233
1234 for (;;) {
1235 mtx_lock(&sc->queue_mtx);
1236 if (sc->flags & MD_SHUTDOWN) {
1237 sc->flags |= MD_EXITING;
1238 mtx_unlock(&sc->queue_mtx);
1239 kproc_exit(0);
1240 }
1241 bp = bioq_takefirst(&sc->bio_queue);
1242 if (!bp) {
1243 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1244 continue;
1245 }
1246 mtx_unlock(&sc->queue_mtx);
1247 if (bp->bio_cmd == BIO_GETATTR) {
1248 md_handleattr(sc, bp);
1249 } else {
1250 error = sc->start(sc, bp);
1251 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
1252 /*
1253 * Devstat uses (bio_bcount, bio_resid) for
1254 * determining the length of the completed part
1255 * of the i/o. g_io_deliver() will translate
1256 * from bio_completed to that, but it also
1257 * destroys the bio so we must do our own
1258 * translation.
1259 */
1260 bp->bio_bcount = bp->bio_length;
1261 devstat_end_transaction_bio(sc->devstat, bp);
1262 }
1263 bp->bio_completed = bp->bio_length - bp->bio_resid;
1264 g_io_deliver(bp, error);
1265 }
1266 }
1267 }
1268
1269 static struct md_s *
mdfind(int unit)1270 mdfind(int unit)
1271 {
1272 struct md_s *sc;
1273
1274 LIST_FOREACH(sc, &md_softc_list, list) {
1275 if (sc->unit == unit)
1276 break;
1277 }
1278 return (sc);
1279 }
1280
1281 static struct md_s *
mdnew(int unit,int * errp,enum md_types type)1282 mdnew(int unit, int *errp, enum md_types type)
1283 {
1284 struct md_s *sc;
1285 int error;
1286
1287 *errp = 0;
1288 if (unit == -1)
1289 unit = alloc_unr(md_uh);
1290 else
1291 unit = alloc_unr_specific(md_uh, unit);
1292
1293 if (unit == -1) {
1294 *errp = EBUSY;
1295 return (NULL);
1296 }
1297
1298 sc = malloc(sizeof(*sc), M_MD, M_WAITOK | M_ZERO);
1299 sc->type = type;
1300 bioq_init(&sc->bio_queue);
1301 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1302 sc->unit = unit;
1303 sprintf(sc->name, "md%d", unit);
1304 LIST_INSERT_HEAD(&md_softc_list, sc, list);
1305 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1306 if (error == 0)
1307 return (sc);
1308 LIST_REMOVE(sc, list);
1309 mtx_destroy(&sc->queue_mtx);
1310 free_unr(md_uh, sc->unit);
1311 free(sc, M_MD);
1312 *errp = error;
1313 return (NULL);
1314 }
1315
1316 static void
mdinit(struct md_s * sc)1317 mdinit(struct md_s *sc)
1318 {
1319 struct g_geom *gp;
1320 struct g_provider *pp;
1321 unsigned remn;
1322
1323 g_topology_lock();
1324 gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1325 gp->softc = sc;
1326 pp = g_new_providerf(gp, "md%d", sc->unit);
1327 devstat_remove_entry(pp->stat);
1328 pp->stat = NULL;
1329 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1330 /* Prune off any residual fractional sector. */
1331 remn = sc->mediasize % sc->sectorsize;
1332 if (remn != 0) {
1333 printf("md%d: truncating fractional last sector by %u bytes\n",
1334 sc->unit, remn);
1335 sc->mediasize -= remn;
1336 }
1337 pp->mediasize = sc->mediasize;
1338 pp->sectorsize = sc->sectorsize;
1339 switch (sc->type) {
1340 case MD_MALLOC:
1341 case MD_VNODE:
1342 case MD_SWAP:
1343 pp->flags |= G_PF_ACCEPT_UNMAPPED;
1344 break;
1345 case MD_PRELOAD:
1346 case MD_NULL:
1347 break;
1348 }
1349 sc->gp = gp;
1350 sc->pp = pp;
1351 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1352 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
1353 sc->devstat->id = pp;
1354 g_error_provider(pp, 0);
1355 g_topology_unlock();
1356 }
1357
1358 static int
mdcreate_malloc(struct md_s * sc,struct md_req * mdr)1359 mdcreate_malloc(struct md_s *sc, struct md_req *mdr)
1360 {
1361 uintptr_t sp;
1362 int error;
1363 off_t u;
1364
1365 error = 0;
1366 if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1367 return (EINVAL);
1368 if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize))
1369 return (EINVAL);
1370 /* Compression doesn't make sense if we have reserved space */
1371 if (mdr->md_options & MD_RESERVE)
1372 mdr->md_options &= ~MD_COMPRESS;
1373 if (mdr->md_fwsectors != 0)
1374 sc->fwsectors = mdr->md_fwsectors;
1375 if (mdr->md_fwheads != 0)
1376 sc->fwheads = mdr->md_fwheads;
1377 sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE | MD_RESERVE);
1378 sc->s_malloc.indir = dimension(sc->mediasize / sc->sectorsize);
1379 sc->s_malloc.uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL,
1380 NULL, NULL, 0x1ff, 0);
1381 if (mdr->md_options & MD_RESERVE) {
1382 off_t nsectors;
1383
1384 nsectors = sc->mediasize / sc->sectorsize;
1385 for (u = 0; u < nsectors; u++) {
1386 sp = (uintptr_t)uma_zalloc(sc->s_malloc.uma,
1387 (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
1388 if (sp != 0)
1389 error = s_write(sc->s_malloc.indir, u, sp);
1390 else
1391 error = ENOMEM;
1392 if (error != 0)
1393 break;
1394 }
1395 }
1396 return (error);
1397 }
1398
1399 static int
mdsetcred(struct md_s * sc,struct ucred * cred)1400 mdsetcred(struct md_s *sc, struct ucred *cred)
1401 {
1402 char *tmpbuf;
1403 int error = 0;
1404
1405 /*
1406 * Set credits in our softc
1407 */
1408
1409 if (sc->cred)
1410 crfree(sc->cred);
1411 sc->cred = crhold(cred);
1412
1413 /*
1414 * Horrible kludge to establish credentials for NFS XXX.
1415 */
1416
1417 if (sc->type == MD_VNODE && sc->s_vnode.vnode != NULL) {
1418 struct uio auio;
1419 struct iovec aiov;
1420
1421 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1422 bzero(&auio, sizeof(auio));
1423
1424 aiov.iov_base = tmpbuf;
1425 aiov.iov_len = sc->sectorsize;
1426 auio.uio_iov = &aiov;
1427 auio.uio_iovcnt = 1;
1428 auio.uio_offset = 0;
1429 auio.uio_rw = UIO_READ;
1430 auio.uio_segflg = UIO_SYSSPACE;
1431 auio.uio_resid = aiov.iov_len;
1432 vn_lock(sc->s_vnode.vnode, LK_EXCLUSIVE | LK_RETRY);
1433 error = VOP_READ(sc->s_vnode.vnode, &auio, 0, sc->cred);
1434 VOP_UNLOCK(sc->s_vnode.vnode);
1435 free(tmpbuf, M_TEMP);
1436 }
1437 return (error);
1438 }
1439
1440 static int
mdcreate_vnode(struct md_s * sc,struct md_req * mdr,struct thread * td)1441 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1442 {
1443 struct vattr vattr;
1444 struct nameidata nd;
1445 char *fname;
1446 int error, flags;
1447 long v;
1448
1449 fname = mdr->md_file;
1450 if (mdr->md_file_seg == UIO_USERSPACE) {
1451 error = copyinstr(fname, sc->s_vnode.file,
1452 sizeof(sc->s_vnode.file), NULL);
1453 if (error != 0)
1454 return (error);
1455 } else if (mdr->md_file_seg == UIO_SYSSPACE)
1456 strlcpy(sc->s_vnode.file, fname, sizeof(sc->s_vnode.file));
1457 else
1458 return (EDOOFUS);
1459
1460 /*
1461 * If the user specified that this is a read only device, don't
1462 * set the FWRITE mask before trying to open the backing store.
1463 */
1464 flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \
1465 | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0);
1466 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->s_vnode.file);
1467 error = vn_open(&nd, &flags, 0, NULL);
1468 if (error != 0)
1469 return (error);
1470 NDFREE_PNBUF(&nd);
1471 if (nd.ni_vp->v_type != VREG) {
1472 error = EINVAL;
1473 goto bad;
1474 }
1475 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1476 if (error != 0)
1477 goto bad;
1478 if ((mdr->md_options & MD_MUSTDEALLOC) != 0) {
1479 error = VOP_PATHCONF(nd.ni_vp, _PC_DEALLOC_PRESENT, &v);
1480 if (error != 0)
1481 goto bad;
1482 if (v == 0)
1483 sc->candelete = false;
1484 }
1485 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1486 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1487 if (VN_IS_DOOMED(nd.ni_vp)) {
1488 /* Forced unmount. */
1489 error = EBADF;
1490 goto bad;
1491 }
1492 }
1493 nd.ni_vp->v_vflag |= VV_MD;
1494 VOP_UNLOCK(nd.ni_vp);
1495
1496 if (mdr->md_fwsectors != 0)
1497 sc->fwsectors = mdr->md_fwsectors;
1498 if (mdr->md_fwheads != 0)
1499 sc->fwheads = mdr->md_fwheads;
1500 snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1501 (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1502 sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE |
1503 MD_VERIFY | MD_MUSTDEALLOC);
1504 if (!(flags & FWRITE))
1505 sc->flags |= MD_READONLY;
1506 sc->s_vnode.vnode = nd.ni_vp;
1507
1508 error = mdsetcred(sc, td->td_ucred);
1509 if (error != 0) {
1510 sc->s_vnode.vnode = NULL;
1511 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1512 nd.ni_vp->v_vflag &= ~VV_MD;
1513 goto bad;
1514 }
1515
1516 sc->s_vnode.kva = kva_alloc(maxphys + PAGE_SIZE);
1517 return (0);
1518 bad:
1519 VOP_UNLOCK(nd.ni_vp);
1520 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1521 return (error);
1522 }
1523
1524 static void
g_md_providergone(struct g_provider * pp)1525 g_md_providergone(struct g_provider *pp)
1526 {
1527 struct md_s *sc = pp->geom->softc;
1528
1529 mtx_lock(&sc->queue_mtx);
1530 sc->flags |= MD_PROVIDERGONE;
1531 wakeup(&sc->flags);
1532 mtx_unlock(&sc->queue_mtx);
1533 }
1534
1535 static int
mddestroy(struct md_s * sc,struct thread * td)1536 mddestroy(struct md_s *sc, struct thread *td)
1537 {
1538
1539 if (sc->gp) {
1540 g_topology_lock();
1541 g_wither_geom(sc->gp, ENXIO);
1542 g_topology_unlock();
1543
1544 mtx_lock(&sc->queue_mtx);
1545 while (!(sc->flags & MD_PROVIDERGONE))
1546 msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0);
1547 mtx_unlock(&sc->queue_mtx);
1548 }
1549 if (sc->devstat) {
1550 devstat_remove_entry(sc->devstat);
1551 sc->devstat = NULL;
1552 }
1553 mtx_lock(&sc->queue_mtx);
1554 sc->flags |= MD_SHUTDOWN;
1555 wakeup(sc);
1556 while (!(sc->flags & MD_EXITING))
1557 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1558 mtx_unlock(&sc->queue_mtx);
1559 mtx_destroy(&sc->queue_mtx);
1560 switch (sc->type) {
1561 case MD_VNODE:
1562 if (sc->s_vnode.vnode != NULL) {
1563 vn_lock(sc->s_vnode.vnode, LK_EXCLUSIVE | LK_RETRY);
1564 sc->s_vnode.vnode->v_vflag &= ~VV_MD;
1565 VOP_UNLOCK(sc->s_vnode.vnode);
1566 (void)vn_close(sc->s_vnode.vnode,
1567 sc->flags & MD_READONLY ? FREAD : (FREAD|FWRITE),
1568 sc->cred, td);
1569 }
1570 if (sc->s_vnode.kva != 0)
1571 kva_free(sc->s_vnode.kva, maxphys + PAGE_SIZE);
1572 break;
1573 case MD_SWAP:
1574 if (sc->s_swap.object != NULL)
1575 vm_object_deallocate(sc->s_swap.object);
1576 break;
1577 case MD_MALLOC:
1578 if (sc->s_malloc.indir != NULL)
1579 destroy_indir(sc, sc->s_malloc.indir);
1580 if (sc->s_malloc.uma != NULL)
1581 uma_zdestroy(sc->s_malloc.uma);
1582 break;
1583 case MD_PRELOAD:
1584 case MD_NULL:
1585 break;
1586 default:
1587 __assert_unreachable();
1588 }
1589 if (sc->cred != NULL)
1590 crfree(sc->cred);
1591
1592 LIST_REMOVE(sc, list);
1593 free_unr(md_uh, sc->unit);
1594 free(sc, M_MD);
1595 return (0);
1596 }
1597
1598 static int
mdresize(struct md_s * sc,struct md_req * mdr)1599 mdresize(struct md_s *sc, struct md_req *mdr)
1600 {
1601 int error, res;
1602 vm_pindex_t oldpages, newpages;
1603
1604 switch (sc->type) {
1605 case MD_VNODE:
1606 case MD_NULL:
1607 break;
1608 case MD_SWAP:
1609 if (mdr->md_mediasize <= 0 ||
1610 (mdr->md_mediasize % PAGE_SIZE) != 0)
1611 return (EDOM);
1612 oldpages = OFF_TO_IDX(sc->mediasize);
1613 newpages = OFF_TO_IDX(mdr->md_mediasize);
1614 if (newpages < oldpages) {
1615 VM_OBJECT_WLOCK(sc->s_swap.object);
1616 vm_object_page_remove(sc->s_swap.object, newpages,
1617 0, 0);
1618 swap_release_by_cred(IDX_TO_OFF(oldpages -
1619 newpages), sc->cred);
1620 sc->s_swap.object->charge = IDX_TO_OFF(newpages);
1621 sc->s_swap.object->size = newpages;
1622 VM_OBJECT_WUNLOCK(sc->s_swap.object);
1623 } else if (newpages > oldpages) {
1624 res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1625 oldpages), sc->cred);
1626 if (!res)
1627 return (ENOMEM);
1628 if ((mdr->md_options & MD_RESERVE) ||
1629 (sc->flags & MD_RESERVE)) {
1630 error = swap_pager_reserve(sc->s_swap.object,
1631 oldpages, newpages - oldpages);
1632 if (error < 0) {
1633 swap_release_by_cred(
1634 IDX_TO_OFF(newpages - oldpages),
1635 sc->cred);
1636 return (EDOM);
1637 }
1638 }
1639 VM_OBJECT_WLOCK(sc->s_swap.object);
1640 sc->s_swap.object->charge = IDX_TO_OFF(newpages);
1641 sc->s_swap.object->size = newpages;
1642 VM_OBJECT_WUNLOCK(sc->s_swap.object);
1643 }
1644 break;
1645 default:
1646 return (EOPNOTSUPP);
1647 }
1648
1649 sc->mediasize = mdr->md_mediasize;
1650
1651 g_topology_lock();
1652 g_resize_provider(sc->pp, sc->mediasize);
1653 g_topology_unlock();
1654 return (0);
1655 }
1656
1657 static int
mdcreate_swap(struct md_s * sc,struct md_req * mdr,struct thread * td)1658 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1659 {
1660 vm_ooffset_t npage;
1661 int error;
1662
1663 /*
1664 * Range check. Disallow negative sizes and sizes not being
1665 * multiple of page size.
1666 */
1667 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1668 return (EDOM);
1669
1670 /*
1671 * Allocate an OBJT_SWAP object.
1672 *
1673 * Note the truncation.
1674 */
1675
1676 if ((mdr->md_options & MD_VERIFY) != 0)
1677 return (EINVAL);
1678 npage = mdr->md_mediasize / PAGE_SIZE;
1679 if (mdr->md_fwsectors != 0)
1680 sc->fwsectors = mdr->md_fwsectors;
1681 if (mdr->md_fwheads != 0)
1682 sc->fwheads = mdr->md_fwheads;
1683 sc->s_swap.object = vm_pager_allocate(OBJT_SWAP, NULL,
1684 PAGE_SIZE * npage, VM_PROT_DEFAULT, 0, td->td_ucred);
1685 if (sc->s_swap.object == NULL)
1686 return (ENOMEM);
1687 sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1688 if (mdr->md_options & MD_RESERVE) {
1689 if (swap_pager_reserve(sc->s_swap.object, 0, npage) < 0) {
1690 error = EDOM;
1691 goto finish;
1692 }
1693 }
1694 error = mdsetcred(sc, td->td_ucred);
1695 finish:
1696 if (error != 0) {
1697 vm_object_deallocate(sc->s_swap.object);
1698 sc->s_swap.object = NULL;
1699 }
1700 return (error);
1701 }
1702
1703 static int
mdcreate_null(struct md_s * sc,struct md_req * mdr,struct thread * td)1704 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1705 {
1706
1707 /*
1708 * Range check. Disallow negative sizes and sizes not being
1709 * multiple of page size.
1710 */
1711 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1712 return (EDOM);
1713
1714 return (0);
1715 }
1716
1717 static int
kern_mdattach_locked(struct thread * td,struct md_req * mdr)1718 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1719 {
1720 struct md_s *sc;
1721 unsigned sectsize;
1722 int error;
1723
1724 sx_assert(&md_sx, SA_XLOCKED);
1725
1726 switch (mdr->md_type) {
1727 case MD_MALLOC:
1728 case MD_PRELOAD:
1729 case MD_VNODE:
1730 case MD_SWAP:
1731 case MD_NULL:
1732 break;
1733 default:
1734 return (EINVAL);
1735 }
1736 if (mdr->md_sectorsize == 0)
1737 sectsize = DEV_BSIZE;
1738 else
1739 sectsize = mdr->md_sectorsize;
1740 if (sectsize > maxphys || mdr->md_mediasize < sectsize)
1741 return (EINVAL);
1742 if (mdr->md_options & MD_AUTOUNIT)
1743 sc = mdnew(-1, &error, mdr->md_type);
1744 else {
1745 if (mdr->md_unit > INT_MAX)
1746 return (EINVAL);
1747 sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1748 }
1749 if (sc == NULL)
1750 return (error);
1751 if (mdr->md_label != NULL)
1752 error = copyinstr(mdr->md_label, sc->label,
1753 sizeof(sc->label), NULL);
1754 if (error != 0)
1755 goto err_after_new;
1756 if (mdr->md_options & MD_AUTOUNIT)
1757 mdr->md_unit = sc->unit;
1758 sc->mediasize = mdr->md_mediasize;
1759 sc->sectorsize = sectsize;
1760 sc->candelete = true;
1761 error = EDOOFUS;
1762 switch (sc->type) {
1763 case MD_MALLOC:
1764 sc->start = mdstart_malloc;
1765 error = mdcreate_malloc(sc, mdr);
1766 break;
1767 case MD_PRELOAD:
1768 /*
1769 * We disallow attaching preloaded memory disks via
1770 * ioctl. Preloaded memory disks are automatically
1771 * attached in g_md_init().
1772 */
1773 error = EOPNOTSUPP;
1774 break;
1775 case MD_VNODE:
1776 sc->start = mdstart_vnode;
1777 error = mdcreate_vnode(sc, mdr, td);
1778 break;
1779 case MD_SWAP:
1780 sc->start = mdstart_swap;
1781 error = mdcreate_swap(sc, mdr, td);
1782 break;
1783 case MD_NULL:
1784 sc->start = mdstart_null;
1785 error = mdcreate_null(sc, mdr, td);
1786 break;
1787 }
1788 err_after_new:
1789 if (error != 0) {
1790 mddestroy(sc, td);
1791 return (error);
1792 }
1793
1794 mdinit(sc);
1795 return (0);
1796 }
1797
1798 static int
kern_mdattach(struct thread * td,struct md_req * mdr)1799 kern_mdattach(struct thread *td, struct md_req *mdr)
1800 {
1801 int error;
1802
1803 sx_xlock(&md_sx);
1804 error = kern_mdattach_locked(td, mdr);
1805 sx_xunlock(&md_sx);
1806 return (error);
1807 }
1808
1809 static int
kern_mddetach_locked(struct thread * td,struct md_req * mdr)1810 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1811 {
1812 struct md_s *sc;
1813
1814 sx_assert(&md_sx, SA_XLOCKED);
1815
1816 if (mdr->md_mediasize != 0 ||
1817 (mdr->md_options & ~MD_FORCE) != 0)
1818 return (EINVAL);
1819
1820 sc = mdfind(mdr->md_unit);
1821 if (sc == NULL)
1822 return (ENOENT);
1823 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1824 !(mdr->md_options & MD_FORCE))
1825 return (EBUSY);
1826 return (mddestroy(sc, td));
1827 }
1828
1829 static int
kern_mddetach(struct thread * td,struct md_req * mdr)1830 kern_mddetach(struct thread *td, struct md_req *mdr)
1831 {
1832 int error;
1833
1834 sx_xlock(&md_sx);
1835 error = kern_mddetach_locked(td, mdr);
1836 sx_xunlock(&md_sx);
1837 return (error);
1838 }
1839
1840 static int
kern_mdresize_locked(struct md_req * mdr)1841 kern_mdresize_locked(struct md_req *mdr)
1842 {
1843 struct md_s *sc;
1844
1845 sx_assert(&md_sx, SA_XLOCKED);
1846
1847 if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1848 return (EINVAL);
1849
1850 sc = mdfind(mdr->md_unit);
1851 if (sc == NULL)
1852 return (ENOENT);
1853 if (mdr->md_mediasize < sc->sectorsize)
1854 return (EINVAL);
1855 mdr->md_mediasize -= mdr->md_mediasize % sc->sectorsize;
1856 if (mdr->md_mediasize < sc->mediasize &&
1857 !(sc->flags & MD_FORCE) &&
1858 !(mdr->md_options & MD_FORCE))
1859 return (EBUSY);
1860 return (mdresize(sc, mdr));
1861 }
1862
1863 static int
kern_mdresize(struct md_req * mdr)1864 kern_mdresize(struct md_req *mdr)
1865 {
1866 int error;
1867
1868 sx_xlock(&md_sx);
1869 error = kern_mdresize_locked(mdr);
1870 sx_xunlock(&md_sx);
1871 return (error);
1872 }
1873
1874 static int
kern_mdquery_locked(struct md_req * mdr)1875 kern_mdquery_locked(struct md_req *mdr)
1876 {
1877 struct md_s *sc;
1878 int error;
1879
1880 sx_assert(&md_sx, SA_XLOCKED);
1881
1882 sc = mdfind(mdr->md_unit);
1883 if (sc == NULL)
1884 return (ENOENT);
1885 mdr->md_type = sc->type;
1886 mdr->md_options = sc->flags;
1887 mdr->md_mediasize = sc->mediasize;
1888 mdr->md_sectorsize = sc->sectorsize;
1889 error = 0;
1890 if (mdr->md_label != NULL) {
1891 error = copyout(sc->label, mdr->md_label,
1892 strlen(sc->label) + 1);
1893 if (error != 0)
1894 return (error);
1895 }
1896 if (sc->type == MD_VNODE) {
1897 error = copyout(sc->s_vnode.file, mdr->md_file,
1898 strlen(sc->s_vnode.file) + 1);
1899 } else if (sc->type == MD_PRELOAD && mdr->md_file != NULL) {
1900 error = copyout(sc->s_preload.name, mdr->md_file,
1901 strlen(sc->s_preload.name) + 1);
1902 }
1903 return (error);
1904 }
1905
1906 static int
kern_mdquery(struct md_req * mdr)1907 kern_mdquery(struct md_req *mdr)
1908 {
1909 int error;
1910
1911 sx_xlock(&md_sx);
1912 error = kern_mdquery_locked(mdr);
1913 sx_xunlock(&md_sx);
1914 return (error);
1915 }
1916
1917 /* Copy members that are not userspace pointers. */
1918 #define MD_IOCTL2REQ(mdio, mdr) do { \
1919 (mdr)->md_unit = (mdio)->md_unit; \
1920 (mdr)->md_type = (mdio)->md_type; \
1921 (mdr)->md_mediasize = (mdio)->md_mediasize; \
1922 (mdr)->md_sectorsize = (mdio)->md_sectorsize; \
1923 (mdr)->md_options = (mdio)->md_options; \
1924 (mdr)->md_fwheads = (mdio)->md_fwheads; \
1925 (mdr)->md_fwsectors = (mdio)->md_fwsectors; \
1926 (mdr)->md_units = &(mdio)->md_pad[0]; \
1927 (mdr)->md_units_nitems = nitems((mdio)->md_pad); \
1928 } while(0)
1929
1930 /* Copy members that might have been updated */
1931 #define MD_REQ2IOCTL(mdr, mdio) do { \
1932 (mdio)->md_unit = (mdr)->md_unit; \
1933 (mdio)->md_type = (mdr)->md_type; \
1934 (mdio)->md_mediasize = (mdr)->md_mediasize; \
1935 (mdio)->md_sectorsize = (mdr)->md_sectorsize; \
1936 (mdio)->md_options = (mdr)->md_options; \
1937 (mdio)->md_fwheads = (mdr)->md_fwheads; \
1938 (mdio)->md_fwsectors = (mdr)->md_fwsectors; \
1939 } while(0)
1940
1941 static int
mdctlioctl(struct cdev * dev,u_long cmd,caddr_t addr,int flags,struct thread * td)1942 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1943 struct thread *td)
1944 {
1945 struct md_req mdr;
1946 int error;
1947
1948 if (md_debug)
1949 printf("mdctlioctl(%s %lx %p %x %p)\n",
1950 devtoname(dev), cmd, addr, flags, td);
1951
1952 bzero(&mdr, sizeof(mdr));
1953 switch (cmd) {
1954 case MDIOCATTACH:
1955 case MDIOCDETACH:
1956 case MDIOCRESIZE:
1957 case MDIOCQUERY: {
1958 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1959 if (mdio->md_version != MDIOVERSION)
1960 return (EINVAL);
1961 MD_IOCTL2REQ(mdio, &mdr);
1962 mdr.md_file = mdio->md_file;
1963 mdr.md_file_seg = UIO_USERSPACE;
1964 /* If the file is adjacent to the md_ioctl it's in kernel. */
1965 if ((void *)mdio->md_file == (void *)(mdio + 1))
1966 mdr.md_file_seg = UIO_SYSSPACE;
1967 mdr.md_label = mdio->md_label;
1968 break;
1969 }
1970 #ifdef COMPAT_FREEBSD32
1971 case MDIOCATTACH_32:
1972 case MDIOCDETACH_32:
1973 case MDIOCRESIZE_32:
1974 case MDIOCQUERY_32: {
1975 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1976 if (mdio->md_version != MDIOVERSION)
1977 return (EINVAL);
1978 MD_IOCTL2REQ(mdio, &mdr);
1979 mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1980 mdr.md_file_seg = UIO_USERSPACE;
1981 mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1982 break;
1983 }
1984 #endif
1985 default:
1986 /* Fall through to handler switch. */
1987 break;
1988 }
1989
1990 error = 0;
1991 switch (cmd) {
1992 case MDIOCATTACH:
1993 #ifdef COMPAT_FREEBSD32
1994 case MDIOCATTACH_32:
1995 #endif
1996 error = kern_mdattach(td, &mdr);
1997 break;
1998 case MDIOCDETACH:
1999 #ifdef COMPAT_FREEBSD32
2000 case MDIOCDETACH_32:
2001 #endif
2002 error = kern_mddetach(td, &mdr);
2003 break;
2004 case MDIOCRESIZE:
2005 #ifdef COMPAT_FREEBSD32
2006 case MDIOCRESIZE_32:
2007 #endif
2008 error = kern_mdresize(&mdr);
2009 break;
2010 case MDIOCQUERY:
2011 #ifdef COMPAT_FREEBSD32
2012 case MDIOCQUERY_32:
2013 #endif
2014 error = kern_mdquery(&mdr);
2015 break;
2016 default:
2017 error = ENOIOCTL;
2018 }
2019
2020 switch (cmd) {
2021 case MDIOCATTACH:
2022 case MDIOCQUERY: {
2023 struct md_ioctl *mdio = (struct md_ioctl *)addr;
2024 MD_REQ2IOCTL(&mdr, mdio);
2025 break;
2026 }
2027 #ifdef COMPAT_FREEBSD32
2028 case MDIOCATTACH_32:
2029 case MDIOCQUERY_32: {
2030 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
2031 MD_REQ2IOCTL(&mdr, mdio);
2032 break;
2033 }
2034 #endif
2035 default:
2036 /* Other commands to not alter mdr. */
2037 break;
2038 }
2039
2040 return (error);
2041 }
2042
2043 static void
md_preloaded(u_char * image,size_t length,const char * name)2044 md_preloaded(u_char *image, size_t length, const char *name)
2045 {
2046 struct md_s *sc;
2047 int error;
2048
2049 sc = mdnew(-1, &error, MD_PRELOAD);
2050 if (sc == NULL)
2051 return;
2052 sc->mediasize = length;
2053 sc->sectorsize = DEV_BSIZE;
2054 sc->s_preload.pl_ptr = image;
2055 sc->s_preload.pl_len = length;
2056 sc->start = mdstart_preload;
2057 if (name != NULL)
2058 strlcpy(sc->s_preload.name, name,
2059 sizeof(sc->s_preload.name));
2060 #ifdef MD_ROOT
2061 if (sc->unit == 0) {
2062 #ifndef ROOTDEVNAME
2063 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2064 #endif
2065 #ifdef MD_ROOT_READONLY
2066 sc->flags |= MD_READONLY;
2067 #endif
2068 }
2069 #endif
2070 mdinit(sc);
2071 if (name != NULL) {
2072 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2073 MD_NAME, sc->unit, name, length, image);
2074 } else {
2075 printf("%s%d: Embedded image %zd bytes at %p\n",
2076 MD_NAME, sc->unit, length, image);
2077 }
2078 }
2079
2080 static void
g_md_init(struct g_class * mp __unused)2081 g_md_init(struct g_class *mp __unused)
2082 {
2083 caddr_t mod;
2084 u_char *ptr, *name, *type;
2085 unsigned len;
2086 int i;
2087
2088 /* figure out log2(NINDIR) */
2089 for (i = NINDIR, nshift = -1; i; nshift++)
2090 i >>= 1;
2091
2092 mod = NULL;
2093 sx_init(&md_sx, "MD config lock");
2094 g_topology_unlock();
2095 md_uh = new_unrhdr(0, INT_MAX, NULL);
2096 #ifdef MD_ROOT
2097 if (mfs_root_size != 0) {
2098 sx_xlock(&md_sx);
2099 #ifdef MD_ROOT_MEM
2100 md_preloaded(mfs_root, mfs_root_size, NULL);
2101 #else
2102 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2103 NULL);
2104 #endif
2105 sx_xunlock(&md_sx);
2106 }
2107 #endif
2108 /* XXX: are preload_* static or do they need Giant ? */
2109 while ((mod = preload_search_next_name(mod)) != NULL) {
2110 name = (char *)preload_search_info(mod, MODINFO_NAME);
2111 if (name == NULL)
2112 continue;
2113 type = (char *)preload_search_info(mod, MODINFO_TYPE);
2114 if (type == NULL)
2115 continue;
2116 if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2117 continue;
2118 ptr = preload_fetch_addr(mod);
2119 len = preload_fetch_size(mod);
2120 if (ptr != NULL && len != 0) {
2121 sx_xlock(&md_sx);
2122 md_preloaded(ptr, len, name);
2123 sx_xunlock(&md_sx);
2124 }
2125 }
2126 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2127 0600, MDCTL_NAME);
2128 g_topology_lock();
2129 }
2130
2131 static void
g_md_dumpconf(struct sbuf * sb,const char * indent,struct g_geom * gp,struct g_consumer * cp __unused,struct g_provider * pp)2132 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2133 struct g_consumer *cp __unused, struct g_provider *pp)
2134 {
2135 struct md_s *mp;
2136 char *type;
2137
2138 mp = gp->softc;
2139 if (mp == NULL)
2140 return;
2141
2142 switch (mp->type) {
2143 case MD_MALLOC:
2144 type = "malloc";
2145 break;
2146 case MD_PRELOAD:
2147 type = "preload";
2148 break;
2149 case MD_VNODE:
2150 type = "vnode";
2151 break;
2152 case MD_SWAP:
2153 type = "swap";
2154 break;
2155 case MD_NULL:
2156 type = "null";
2157 break;
2158 default:
2159 type = "unknown";
2160 break;
2161 }
2162
2163 if (pp != NULL) {
2164 if (indent == NULL) {
2165 sbuf_printf(sb, " u %d", mp->unit);
2166 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2167 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2168 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2169 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2170 sbuf_printf(sb, " t %s", type);
2171 if (mp->type == MD_VNODE &&
2172 mp->s_vnode.vnode != NULL)
2173 sbuf_printf(sb, " file %s", mp->s_vnode.file);
2174 if (mp->type == MD_PRELOAD &&
2175 mp->s_preload.name[0] != '\0') {
2176 sbuf_printf(sb, " file %s",
2177 mp->s_preload.name);
2178 }
2179 sbuf_printf(sb, " label %s", mp->label);
2180 } else {
2181 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2182 mp->unit);
2183 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2184 indent, (uintmax_t) mp->sectorsize);
2185 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2186 indent, (uintmax_t) mp->fwheads);
2187 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2188 indent, (uintmax_t) mp->fwsectors);
2189 if (mp->ident[0] != '\0') {
2190 sbuf_printf(sb, "%s<ident>", indent);
2191 g_conf_printf_escaped(sb, "%s", mp->ident);
2192 sbuf_printf(sb, "</ident>\n");
2193 }
2194 sbuf_printf(sb, "%s<length>%ju</length>\n",
2195 indent, (uintmax_t) mp->mediasize);
2196 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2197 (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2198 sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2199 (mp->flags & MD_READONLY) == 0 ? "read-write":
2200 "read-only");
2201 sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2202 type);
2203 if (mp->type == MD_VNODE) {
2204 if (mp->s_vnode.vnode != NULL) {
2205 sbuf_printf(sb, "%s<file>", indent);
2206 g_conf_printf_escaped(sb, "%s",
2207 mp->s_vnode.file);
2208 sbuf_printf(sb, "</file>\n");
2209 }
2210 sbuf_printf(sb, "%s<cache>%s</cache>\n", indent,
2211 (mp->flags & MD_CACHE) == 0 ? "off": "on");
2212 }
2213 if (mp->type == MD_PRELOAD &&
2214 mp->s_preload.name[0] != '\0') {
2215 sbuf_printf(sb, "%s<file>", indent);
2216 g_conf_printf_escaped(sb, "%s",
2217 mp->s_preload.name);
2218 sbuf_printf(sb, "</file>\n");
2219 }
2220 sbuf_printf(sb, "%s<label>", indent);
2221 g_conf_printf_escaped(sb, "%s", mp->label);
2222 sbuf_printf(sb, "</label>\n");
2223 }
2224 }
2225 }
2226
2227 static void
g_md_fini(struct g_class * mp __unused)2228 g_md_fini(struct g_class *mp __unused)
2229 {
2230
2231 sx_destroy(&md_sx);
2232 if (status_dev != NULL)
2233 destroy_dev(status_dev);
2234 delete_unrhdr(md_uh);
2235 }
2236