FreeBSD/Linux Kernel Cross Reference
sys/dev/md/md.c
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 * $FreeBSD$
12 *
13 */
14
15 /*-
16 * The following functions are based on the vn(4) driver: mdstart_swap(),
17 * mdstart_vnode(), mdcreate_swap(), mdcreate_vnode() and mddestroy(),
18 * and as such under the following copyright:
19 *
20 * Copyright (c) 1988 University of Utah.
21 * Copyright (c) 1990, 1993
22 * The Regents of the University of California. All rights reserved.
23 * Copyright (c) 2013 The FreeBSD Foundation
24 * All rights reserved.
25 *
26 * This code is derived from software contributed to Berkeley by
27 * the Systems Programming Group of the University of Utah Computer
28 * Science Department.
29 *
30 * Portions of this software were developed by Konstantin Belousov
31 * under sponsorship from the FreeBSD Foundation.
32 *
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
35 * are met:
36 * 1. Redistributions of source code must retain the above copyright
37 * notice, this list of conditions and the following disclaimer.
38 * 2. Redistributions in binary form must reproduce the above copyright
39 * notice, this list of conditions and the following disclaimer in the
40 * documentation and/or other materials provided with the distribution.
41 * 3. Neither the name of the University nor the names of its contributors
42 * may be used to endorse or promote products derived from this software
43 * without specific prior written permission.
44 *
45 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55 * SUCH DAMAGE.
56 *
57 * from: Utah Hdr: vn.c 1.13 94/04/02
58 *
59 * from: @(#)vn.c 8.6 (Berkeley) 4/1/94
60 * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03
61 */
62
63 #include "opt_rootdevname.h"
64 #include "opt_geom.h"
65 #include "opt_md.h"
66
67 #include <sys/param.h>
68 #include <sys/systm.h>
69 #include <sys/bio.h>
70 #include <sys/buf.h>
71 #include <sys/conf.h>
72 #include <sys/devicestat.h>
73 #include <sys/fcntl.h>
74 #include <sys/kernel.h>
75 #include <sys/kthread.h>
76 #include <sys/limits.h>
77 #include <sys/linker.h>
78 #include <sys/lock.h>
79 #include <sys/malloc.h>
80 #include <sys/mdioctl.h>
81 #include <sys/mount.h>
82 #include <sys/mutex.h>
83 #include <sys/sx.h>
84 #include <sys/namei.h>
85 #include <sys/proc.h>
86 #include <sys/queue.h>
87 #include <sys/rwlock.h>
88 #include <sys/sbuf.h>
89 #include <sys/sched.h>
90 #include <sys/sf_buf.h>
91 #include <sys/sysctl.h>
92 #include <sys/uio.h>
93 #include <sys/unistd.h>
94 #include <sys/vnode.h>
95 #include <sys/disk.h>
96
97 #include <geom/geom.h>
98 #include <geom/geom_int.h>
99
100 #include <vm/vm.h>
101 #include <vm/vm_param.h>
102 #include <vm/vm_object.h>
103 #include <vm/vm_page.h>
104 #include <vm/vm_pager.h>
105 #include <vm/swap_pager.h>
106 #include <vm/uma.h>
107
108 #include <machine/bus.h>
109
110 #define MD_MODVER 1
111
112 #define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */
113 #define MD_EXITING 0x20000 /* Worker thread is exiting. */
114 #define MD_PROVIDERGONE 0x40000 /* Safe to free the softc */
115
116 #ifndef MD_NSECT
117 #define MD_NSECT (10000 * 2)
118 #endif
119
120 struct md_req {
121 unsigned md_unit; /* unit number */
122 enum md_types md_type; /* type of disk */
123 off_t md_mediasize; /* size of disk in bytes */
124 unsigned md_sectorsize; /* sectorsize */
125 unsigned md_options; /* options */
126 int md_fwheads; /* firmware heads */
127 int md_fwsectors; /* firmware sectors */
128 char *md_file; /* pathname of file to mount */
129 enum uio_seg md_file_seg; /* location of md_file */
130 char *md_label; /* label of the device (userspace) */
131 int *md_units; /* pointer to units array (kernel) */
132 size_t md_units_nitems; /* items in md_units array */
133 };
134
135 #ifdef COMPAT_FREEBSD32
136 struct md_ioctl32 {
137 unsigned md_version;
138 unsigned md_unit;
139 enum md_types md_type;
140 uint32_t md_file;
141 off_t md_mediasize;
142 unsigned md_sectorsize;
143 unsigned md_options;
144 uint64_t md_base;
145 int md_fwheads;
146 int md_fwsectors;
147 uint32_t md_label;
148 int md_pad[MDNPAD];
149 } __attribute__((__packed__));
150 CTASSERT((sizeof(struct md_ioctl32)) == 436);
151
152 #define MDIOCATTACH_32 _IOC_NEWTYPE(MDIOCATTACH, struct md_ioctl32)
153 #define MDIOCDETACH_32 _IOC_NEWTYPE(MDIOCDETACH, struct md_ioctl32)
154 #define MDIOCQUERY_32 _IOC_NEWTYPE(MDIOCQUERY, struct md_ioctl32)
155 #define MDIOCRESIZE_32 _IOC_NEWTYPE(MDIOCRESIZE, struct md_ioctl32)
156 #endif /* COMPAT_FREEBSD32 */
157
158 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk");
159 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors");
160
161 static int md_debug;
162 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0,
163 "Enable md(4) debug messages");
164 static int md_malloc_wait;
165 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0,
166 "Allow malloc to wait for memory allocations");
167
168 #if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE)
169 #define MD_ROOT_FSTYPE "ufs"
170 #endif
171
172 #if defined(MD_ROOT)
173 /*
174 * Preloaded image gets put here.
175 */
176 #if defined(MD_ROOT_SIZE)
177 /*
178 * We put the mfs_root symbol into the oldmfs section of the kernel object file.
179 * Applications that patch the object with the image can determine
180 * the size looking at the oldmfs section size within the kernel.
181 */
182 u_char mfs_root[MD_ROOT_SIZE*1024] __attribute__ ((section ("oldmfs")));
183 const int mfs_root_size = sizeof(mfs_root);
184 #elif defined(MD_ROOT_MEM)
185 /* MD region already mapped in the memory */
186 u_char *mfs_root;
187 int mfs_root_size;
188 #else
189 extern volatile u_char __weak_symbol mfs_root;
190 extern volatile u_char __weak_symbol mfs_root_end;
191 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root))
192 #endif
193 #endif
194
195 static g_init_t g_md_init;
196 static g_fini_t g_md_fini;
197 static g_start_t g_md_start;
198 static g_access_t g_md_access;
199 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
200 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
201 static g_provgone_t g_md_providergone;
202
203 static struct cdev *status_dev = NULL;
204 static struct sx md_sx;
205 static struct unrhdr *md_uh;
206
207 static d_ioctl_t mdctlioctl;
208
209 static struct cdevsw mdctl_cdevsw = {
210 .d_version = D_VERSION,
211 .d_ioctl = mdctlioctl,
212 .d_name = MD_NAME,
213 };
214
215 struct g_class g_md_class = {
216 .name = "MD",
217 .version = G_VERSION,
218 .init = g_md_init,
219 .fini = g_md_fini,
220 .start = g_md_start,
221 .access = g_md_access,
222 .dumpconf = g_md_dumpconf,
223 .providergone = g_md_providergone,
224 };
225
226 DECLARE_GEOM_CLASS(g_md_class, g_md);
227 MODULE_VERSION(geom_md, 0);
228
229 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
230
231 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t))
232 #define NMASK (NINDIR-1)
233 static int nshift;
234
235 static uma_zone_t md_pbuf_zone;
236
237 struct indir {
238 uintptr_t *array;
239 u_int total;
240 u_int used;
241 u_int shift;
242 };
243
244 struct md_s {
245 int unit;
246 LIST_ENTRY(md_s) list;
247 struct bio_queue_head bio_queue;
248 struct mtx queue_mtx;
249 struct cdev *dev;
250 enum md_types type;
251 off_t mediasize;
252 unsigned sectorsize;
253 unsigned opencount;
254 unsigned fwheads;
255 unsigned fwsectors;
256 char ident[32];
257 unsigned flags;
258 char name[20];
259 struct proc *procp;
260 struct g_geom *gp;
261 struct g_provider *pp;
262 int (*start)(struct md_s *sc, struct bio *bp);
263 struct devstat *devstat;
264 bool candelete;
265
266 /* MD_MALLOC related fields */
267 struct indir *indir;
268 uma_zone_t uma;
269
270 /* MD_PRELOAD related fields */
271 u_char *pl_ptr;
272 size_t pl_len;
273
274 /* MD_VNODE related fields */
275 struct vnode *vnode;
276 char file[PATH_MAX];
277 char label[PATH_MAX];
278 struct ucred *cred;
279
280 /* MD_SWAP related fields */
281 vm_object_t object;
282 };
283
284 static struct indir *
285 new_indir(u_int shift)
286 {
287 struct indir *ip;
288
289 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
290 | M_ZERO);
291 if (ip == NULL)
292 return (NULL);
293 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
294 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
295 if (ip->array == NULL) {
296 free(ip, M_MD);
297 return (NULL);
298 }
299 ip->total = NINDIR;
300 ip->shift = shift;
301 return (ip);
302 }
303
304 static void
305 del_indir(struct indir *ip)
306 {
307
308 free(ip->array, M_MDSECT);
309 free(ip, M_MD);
310 }
311
312 static void
313 destroy_indir(struct md_s *sc, struct indir *ip)
314 {
315 int i;
316
317 for (i = 0; i < NINDIR; i++) {
318 if (!ip->array[i])
319 continue;
320 if (ip->shift)
321 destroy_indir(sc, (struct indir*)(ip->array[i]));
322 else if (ip->array[i] > 255)
323 uma_zfree(sc->uma, (void *)(ip->array[i]));
324 }
325 del_indir(ip);
326 }
327
328 /*
329 * This function does the math and allocates the top level "indir" structure
330 * for a device of "size" sectors.
331 */
332
333 static struct indir *
334 dimension(off_t size)
335 {
336 off_t rcnt;
337 struct indir *ip;
338 int layer;
339
340 rcnt = size;
341 layer = 0;
342 while (rcnt > NINDIR) {
343 rcnt /= NINDIR;
344 layer++;
345 }
346
347 /*
348 * XXX: the top layer is probably not fully populated, so we allocate
349 * too much space for ip->array in here.
350 */
351 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
352 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
353 M_MDSECT, M_WAITOK | M_ZERO);
354 ip->total = NINDIR;
355 ip->shift = layer * nshift;
356 return (ip);
357 }
358
359 /*
360 * Read a given sector
361 */
362
363 static uintptr_t
364 s_read(struct indir *ip, off_t offset)
365 {
366 struct indir *cip;
367 int idx;
368 uintptr_t up;
369
370 if (md_debug > 1)
371 printf("s_read(%jd)\n", (intmax_t)offset);
372 up = 0;
373 for (cip = ip; cip != NULL;) {
374 if (cip->shift) {
375 idx = (offset >> cip->shift) & NMASK;
376 up = cip->array[idx];
377 cip = (struct indir *)up;
378 continue;
379 }
380 idx = offset & NMASK;
381 return (cip->array[idx]);
382 }
383 return (0);
384 }
385
386 /*
387 * Write a given sector, prune the tree if the value is 0
388 */
389
390 static int
391 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
392 {
393 struct indir *cip, *lip[10];
394 int idx, li;
395 uintptr_t up;
396
397 if (md_debug > 1)
398 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
399 up = 0;
400 li = 0;
401 cip = ip;
402 for (;;) {
403 lip[li++] = cip;
404 if (cip->shift) {
405 idx = (offset >> cip->shift) & NMASK;
406 up = cip->array[idx];
407 if (up != 0) {
408 cip = (struct indir *)up;
409 continue;
410 }
411 /* Allocate branch */
412 cip->array[idx] =
413 (uintptr_t)new_indir(cip->shift - nshift);
414 if (cip->array[idx] == 0)
415 return (ENOSPC);
416 cip->used++;
417 up = cip->array[idx];
418 cip = (struct indir *)up;
419 continue;
420 }
421 /* leafnode */
422 idx = offset & NMASK;
423 up = cip->array[idx];
424 if (up != 0)
425 cip->used--;
426 cip->array[idx] = ptr;
427 if (ptr != 0)
428 cip->used++;
429 break;
430 }
431 if (cip->used != 0 || li == 1)
432 return (0);
433 li--;
434 while (cip->used == 0 && cip != ip) {
435 li--;
436 idx = (offset >> lip[li]->shift) & NMASK;
437 up = lip[li]->array[idx];
438 KASSERT(up == (uintptr_t)cip, ("md screwed up"));
439 del_indir(cip);
440 lip[li]->array[idx] = 0;
441 lip[li]->used--;
442 cip = lip[li];
443 }
444 return (0);
445 }
446
447 static int
448 g_md_access(struct g_provider *pp, int r, int w, int e)
449 {
450 struct md_s *sc;
451
452 sc = pp->geom->softc;
453 if (sc == NULL) {
454 if (r <= 0 && w <= 0 && e <= 0)
455 return (0);
456 return (ENXIO);
457 }
458 r += pp->acr;
459 w += pp->acw;
460 e += pp->ace;
461 if ((sc->flags & MD_READONLY) != 0 && w > 0)
462 return (EROFS);
463 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
464 sc->opencount = 1;
465 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
466 sc->opencount = 0;
467 }
468 return (0);
469 }
470
471 static void
472 g_md_start(struct bio *bp)
473 {
474 struct md_s *sc;
475
476 sc = bp->bio_to->geom->softc;
477 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
478 devstat_start_transaction_bio(sc->devstat, bp);
479 }
480 mtx_lock(&sc->queue_mtx);
481 bioq_disksort(&sc->bio_queue, bp);
482 wakeup(sc);
483 mtx_unlock(&sc->queue_mtx);
484 }
485
486 #define MD_MALLOC_MOVE_ZERO 1
487 #define MD_MALLOC_MOVE_FILL 2
488 #define MD_MALLOC_MOVE_READ 3
489 #define MD_MALLOC_MOVE_WRITE 4
490 #define MD_MALLOC_MOVE_CMP 5
491
492 static int
493 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize,
494 void *ptr, u_char fill, int op)
495 {
496 struct sf_buf *sf;
497 vm_page_t m, *mp1;
498 char *p, first;
499 off_t *uc;
500 unsigned n;
501 int error, i, ma_offs1, sz, first_read;
502
503 m = NULL;
504 error = 0;
505 sf = NULL;
506 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */
507 first = 0;
508 first_read = 0;
509 uc = ptr;
510 mp1 = *mp;
511 ma_offs1 = *ma_offs;
512 /* } */
513 sched_pin();
514 for (n = sectorsize; n != 0; n -= sz) {
515 sz = imin(PAGE_SIZE - *ma_offs, n);
516 if (m != **mp) {
517 if (sf != NULL)
518 sf_buf_free(sf);
519 m = **mp;
520 sf = sf_buf_alloc(m, SFB_CPUPRIVATE |
521 (md_malloc_wait ? 0 : SFB_NOWAIT));
522 if (sf == NULL) {
523 error = ENOMEM;
524 break;
525 }
526 }
527 p = (char *)sf_buf_kva(sf) + *ma_offs;
528 switch (op) {
529 case MD_MALLOC_MOVE_ZERO:
530 bzero(p, sz);
531 break;
532 case MD_MALLOC_MOVE_FILL:
533 memset(p, fill, sz);
534 break;
535 case MD_MALLOC_MOVE_READ:
536 bcopy(ptr, p, sz);
537 cpu_flush_dcache(p, sz);
538 break;
539 case MD_MALLOC_MOVE_WRITE:
540 bcopy(p, ptr, sz);
541 break;
542 case MD_MALLOC_MOVE_CMP:
543 for (i = 0; i < sz; i++, p++) {
544 if (!first_read) {
545 *uc = (u_char)*p;
546 first = *p;
547 first_read = 1;
548 } else if (*p != first) {
549 error = EDOOFUS;
550 break;
551 }
552 }
553 break;
554 default:
555 KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op));
556 break;
557 }
558 if (error != 0)
559 break;
560 *ma_offs += sz;
561 *ma_offs %= PAGE_SIZE;
562 if (*ma_offs == 0)
563 (*mp)++;
564 ptr = (char *)ptr + sz;
565 }
566
567 if (sf != NULL)
568 sf_buf_free(sf);
569 sched_unpin();
570 if (op == MD_MALLOC_MOVE_CMP && error != 0) {
571 *mp = mp1;
572 *ma_offs = ma_offs1;
573 }
574 return (error);
575 }
576
577 static int
578 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs,
579 unsigned len, void *ptr, u_char fill, int op)
580 {
581 bus_dma_segment_t *vlist;
582 uint8_t *p, *end, first;
583 off_t *uc;
584 int ma_offs, seg_len;
585
586 vlist = *pvlist;
587 ma_offs = *pma_offs;
588 uc = ptr;
589
590 for (; len != 0; len -= seg_len) {
591 seg_len = imin(vlist->ds_len - ma_offs, len);
592 p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs;
593 switch (op) {
594 case MD_MALLOC_MOVE_ZERO:
595 bzero(p, seg_len);
596 break;
597 case MD_MALLOC_MOVE_FILL:
598 memset(p, fill, seg_len);
599 break;
600 case MD_MALLOC_MOVE_READ:
601 bcopy(ptr, p, seg_len);
602 cpu_flush_dcache(p, seg_len);
603 break;
604 case MD_MALLOC_MOVE_WRITE:
605 bcopy(p, ptr, seg_len);
606 break;
607 case MD_MALLOC_MOVE_CMP:
608 end = p + seg_len;
609 first = *uc = *p;
610 /* Confirm all following bytes match the first */
611 while (++p < end) {
612 if (*p != first)
613 return (EDOOFUS);
614 }
615 break;
616 default:
617 KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op));
618 break;
619 }
620
621 ma_offs += seg_len;
622 if (ma_offs == vlist->ds_len) {
623 ma_offs = 0;
624 vlist++;
625 }
626 ptr = (uint8_t *)ptr + seg_len;
627 }
628 *pvlist = vlist;
629 *pma_offs = ma_offs;
630
631 return (0);
632 }
633
634 static int
635 mdstart_malloc(struct md_s *sc, struct bio *bp)
636 {
637 u_char *dst;
638 vm_page_t *m;
639 bus_dma_segment_t *vlist;
640 int i, error, error1, ma_offs, notmapped;
641 off_t secno, nsec, uc;
642 uintptr_t sp, osp;
643
644 switch (bp->bio_cmd) {
645 case BIO_READ:
646 case BIO_WRITE:
647 case BIO_DELETE:
648 break;
649 case BIO_FLUSH:
650 return (0);
651 default:
652 return (EOPNOTSUPP);
653 }
654
655 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0;
656 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
657 (bus_dma_segment_t *)bp->bio_data : NULL;
658 if (notmapped) {
659 m = bp->bio_ma;
660 ma_offs = bp->bio_ma_offset;
661 dst = NULL;
662 KASSERT(vlist == NULL, ("vlists cannot be unmapped"));
663 } else if (vlist != NULL) {
664 ma_offs = bp->bio_ma_offset;
665 dst = NULL;
666 } else {
667 dst = bp->bio_data;
668 }
669
670 nsec = bp->bio_length / sc->sectorsize;
671 secno = bp->bio_offset / sc->sectorsize;
672 error = 0;
673 while (nsec--) {
674 osp = s_read(sc->indir, secno);
675 if (bp->bio_cmd == BIO_DELETE) {
676 if (osp != 0)
677 error = s_write(sc->indir, secno, 0);
678 } else if (bp->bio_cmd == BIO_READ) {
679 if (osp == 0) {
680 if (notmapped) {
681 error = md_malloc_move_ma(&m, &ma_offs,
682 sc->sectorsize, NULL, 0,
683 MD_MALLOC_MOVE_ZERO);
684 } else if (vlist != NULL) {
685 error = md_malloc_move_vlist(&vlist,
686 &ma_offs, sc->sectorsize, NULL, 0,
687 MD_MALLOC_MOVE_ZERO);
688 } else
689 bzero(dst, sc->sectorsize);
690 } else if (osp <= 255) {
691 if (notmapped) {
692 error = md_malloc_move_ma(&m, &ma_offs,
693 sc->sectorsize, NULL, osp,
694 MD_MALLOC_MOVE_FILL);
695 } else if (vlist != NULL) {
696 error = md_malloc_move_vlist(&vlist,
697 &ma_offs, sc->sectorsize, NULL, osp,
698 MD_MALLOC_MOVE_FILL);
699 } else
700 memset(dst, osp, sc->sectorsize);
701 } else {
702 if (notmapped) {
703 error = md_malloc_move_ma(&m, &ma_offs,
704 sc->sectorsize, (void *)osp, 0,
705 MD_MALLOC_MOVE_READ);
706 } else if (vlist != NULL) {
707 error = md_malloc_move_vlist(&vlist,
708 &ma_offs, sc->sectorsize,
709 (void *)osp, 0,
710 MD_MALLOC_MOVE_READ);
711 } else {
712 bcopy((void *)osp, dst, sc->sectorsize);
713 cpu_flush_dcache(dst, sc->sectorsize);
714 }
715 }
716 osp = 0;
717 } else if (bp->bio_cmd == BIO_WRITE) {
718 if (sc->flags & MD_COMPRESS) {
719 if (notmapped) {
720 error1 = md_malloc_move_ma(&m, &ma_offs,
721 sc->sectorsize, &uc, 0,
722 MD_MALLOC_MOVE_CMP);
723 i = error1 == 0 ? sc->sectorsize : 0;
724 } else if (vlist != NULL) {
725 error1 = md_malloc_move_vlist(&vlist,
726 &ma_offs, sc->sectorsize, &uc, 0,
727 MD_MALLOC_MOVE_CMP);
728 i = error1 == 0 ? sc->sectorsize : 0;
729 } else {
730 uc = dst[0];
731 for (i = 1; i < sc->sectorsize; i++) {
732 if (dst[i] != uc)
733 break;
734 }
735 }
736 } else {
737 i = 0;
738 uc = 0;
739 }
740 if (i == sc->sectorsize) {
741 if (osp != uc)
742 error = s_write(sc->indir, secno, uc);
743 } else {
744 if (osp <= 255) {
745 sp = (uintptr_t)uma_zalloc(sc->uma,
746 md_malloc_wait ? M_WAITOK :
747 M_NOWAIT);
748 if (sp == 0) {
749 error = ENOSPC;
750 break;
751 }
752 if (notmapped) {
753 error = md_malloc_move_ma(&m,
754 &ma_offs, sc->sectorsize,
755 (void *)sp, 0,
756 MD_MALLOC_MOVE_WRITE);
757 } else if (vlist != NULL) {
758 error = md_malloc_move_vlist(
759 &vlist, &ma_offs,
760 sc->sectorsize, (void *)sp,
761 0, MD_MALLOC_MOVE_WRITE);
762 } else {
763 bcopy(dst, (void *)sp,
764 sc->sectorsize);
765 }
766 error = s_write(sc->indir, secno, sp);
767 } else {
768 if (notmapped) {
769 error = md_malloc_move_ma(&m,
770 &ma_offs, sc->sectorsize,
771 (void *)osp, 0,
772 MD_MALLOC_MOVE_WRITE);
773 } else if (vlist != NULL) {
774 error = md_malloc_move_vlist(
775 &vlist, &ma_offs,
776 sc->sectorsize, (void *)osp,
777 0, MD_MALLOC_MOVE_WRITE);
778 } else {
779 bcopy(dst, (void *)osp,
780 sc->sectorsize);
781 }
782 osp = 0;
783 }
784 }
785 } else {
786 error = EOPNOTSUPP;
787 }
788 if (osp > 255)
789 uma_zfree(sc->uma, (void*)osp);
790 if (error != 0)
791 break;
792 secno++;
793 if (!notmapped && vlist == NULL)
794 dst += sc->sectorsize;
795 }
796 bp->bio_resid = 0;
797 return (error);
798 }
799
800 static void
801 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len)
802 {
803 off_t seg_len;
804
805 while (offset >= vlist->ds_len) {
806 offset -= vlist->ds_len;
807 vlist++;
808 }
809
810 while (len != 0) {
811 seg_len = omin(len, vlist->ds_len - offset);
812 bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset),
813 seg_len);
814 offset = 0;
815 src = (uint8_t *)src + seg_len;
816 len -= seg_len;
817 vlist++;
818 }
819 }
820
821 static void
822 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len)
823 {
824 off_t seg_len;
825
826 while (offset >= vlist->ds_len) {
827 offset -= vlist->ds_len;
828 vlist++;
829 }
830
831 while (len != 0) {
832 seg_len = omin(len, vlist->ds_len - offset);
833 bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst,
834 seg_len);
835 offset = 0;
836 dst = (uint8_t *)dst + seg_len;
837 len -= seg_len;
838 vlist++;
839 }
840 }
841
842 static int
843 mdstart_preload(struct md_s *sc, struct bio *bp)
844 {
845 uint8_t *p;
846
847 p = sc->pl_ptr + bp->bio_offset;
848 switch (bp->bio_cmd) {
849 case BIO_READ:
850 if ((bp->bio_flags & BIO_VLIST) != 0) {
851 mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data,
852 bp->bio_ma_offset, bp->bio_length);
853 } else {
854 bcopy(p, bp->bio_data, bp->bio_length);
855 }
856 cpu_flush_dcache(bp->bio_data, bp->bio_length);
857 break;
858 case BIO_WRITE:
859 if ((bp->bio_flags & BIO_VLIST) != 0) {
860 mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data,
861 bp->bio_ma_offset, p, bp->bio_length);
862 } else {
863 bcopy(bp->bio_data, p, bp->bio_length);
864 }
865 break;
866 }
867 bp->bio_resid = 0;
868 return (0);
869 }
870
871 static int
872 mdstart_vnode(struct md_s *sc, struct bio *bp)
873 {
874 int error;
875 struct uio auio;
876 struct iovec aiov;
877 struct iovec *piov;
878 struct mount *mp;
879 struct vnode *vp;
880 struct buf *pb;
881 bus_dma_segment_t *vlist;
882 struct thread *td;
883 off_t iolen, iostart, off, len;
884 int ma_offs, npages;
885
886 switch (bp->bio_cmd) {
887 case BIO_READ:
888 auio.uio_rw = UIO_READ;
889 break;
890 case BIO_WRITE:
891 auio.uio_rw = UIO_WRITE;
892 break;
893 case BIO_FLUSH:
894 break;
895 case BIO_DELETE:
896 if (sc->candelete)
897 break;
898 /* FALLTHROUGH */
899 default:
900 return (EOPNOTSUPP);
901 }
902
903 td = curthread;
904 vp = sc->vnode;
905 pb = NULL;
906 piov = NULL;
907 ma_offs = bp->bio_ma_offset;
908 off = bp->bio_offset;
909 len = bp->bio_length;
910
911 /*
912 * VNODE I/O
913 *
914 * If an error occurs, we set BIO_ERROR but we do not set
915 * B_INVAL because (for a write anyway), the buffer is
916 * still valid.
917 */
918
919 if (bp->bio_cmd == BIO_FLUSH) {
920 do {
921 (void)vn_start_write(vp, &mp, V_WAIT);
922 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
923 error = VOP_FSYNC(vp, MNT_WAIT, td);
924 VOP_UNLOCK(vp);
925 vn_finished_write(mp);
926 } while (error == ERELOOKUP);
927 return (error);
928 } else if (bp->bio_cmd == BIO_DELETE) {
929 error = vn_deallocate(vp, &off, &len, 0,
930 sc->flags & MD_ASYNC ? 0 : IO_SYNC, sc->cred, NOCRED);
931 bp->bio_resid = len;
932 return (error);
933 }
934
935 auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
936 auio.uio_resid = bp->bio_length;
937 auio.uio_segflg = UIO_SYSSPACE;
938 auio.uio_td = td;
939
940 if ((bp->bio_flags & BIO_VLIST) != 0) {
941 piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK);
942 auio.uio_iov = piov;
943 vlist = (bus_dma_segment_t *)bp->bio_data;
944 while (len > 0) {
945 piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr +
946 ma_offs);
947 piov->iov_len = vlist->ds_len - ma_offs;
948 if (piov->iov_len > len)
949 piov->iov_len = len;
950 len -= piov->iov_len;
951 ma_offs = 0;
952 vlist++;
953 piov++;
954 }
955 auio.uio_iovcnt = piov - auio.uio_iov;
956 piov = auio.uio_iov;
957 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
958 pb = uma_zalloc(md_pbuf_zone, M_WAITOK);
959 MPASS((pb->b_flags & B_MAXPHYS) != 0);
960 bp->bio_resid = len;
961 unmapped_step:
962 npages = atop(min(maxphys, round_page(len + (ma_offs &
963 PAGE_MASK))));
964 iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len);
965 KASSERT(iolen > 0, ("zero iolen"));
966 pmap_qenter((vm_offset_t)pb->b_data,
967 &bp->bio_ma[atop(ma_offs)], npages);
968 aiov.iov_base = (void *)((vm_offset_t)pb->b_data +
969 (ma_offs & PAGE_MASK));
970 aiov.iov_len = iolen;
971 auio.uio_iov = &aiov;
972 auio.uio_iovcnt = 1;
973 auio.uio_resid = iolen;
974 } else {
975 aiov.iov_base = bp->bio_data;
976 aiov.iov_len = bp->bio_length;
977 auio.uio_iov = &aiov;
978 auio.uio_iovcnt = 1;
979 }
980 iostart = auio.uio_offset;
981 if (auio.uio_rw == UIO_READ) {
982 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
983 error = VOP_READ(vp, &auio, 0, sc->cred);
984 VOP_UNLOCK(vp);
985 } else {
986 (void) vn_start_write(vp, &mp, V_WAIT);
987 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
988 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
989 sc->cred);
990 VOP_UNLOCK(vp);
991 vn_finished_write(mp);
992 if (error == 0)
993 sc->flags &= ~MD_VERIFY;
994 }
995
996 /* When MD_CACHE is set, try to avoid double-caching the data. */
997 if (error == 0 && (sc->flags & MD_CACHE) == 0)
998 VOP_ADVISE(vp, iostart, auio.uio_offset - 1,
999 POSIX_FADV_DONTNEED);
1000
1001 if (pb != NULL) {
1002 pmap_qremove((vm_offset_t)pb->b_data, npages);
1003 if (error == 0) {
1004 len -= iolen;
1005 bp->bio_resid -= iolen;
1006 ma_offs += iolen;
1007 if (len > 0)
1008 goto unmapped_step;
1009 }
1010 uma_zfree(md_pbuf_zone, pb);
1011 } else {
1012 bp->bio_resid = auio.uio_resid;
1013 }
1014
1015 free(piov, M_MD);
1016 return (error);
1017 }
1018
1019 static int
1020 mdstart_swap(struct md_s *sc, struct bio *bp)
1021 {
1022 vm_page_t m;
1023 u_char *p;
1024 vm_pindex_t i, lastp;
1025 bus_dma_segment_t *vlist;
1026 int rv, ma_offs, offs, len, lastend;
1027
1028 switch (bp->bio_cmd) {
1029 case BIO_READ:
1030 case BIO_WRITE:
1031 case BIO_DELETE:
1032 break;
1033 case BIO_FLUSH:
1034 return (0);
1035 default:
1036 return (EOPNOTSUPP);
1037 }
1038
1039 p = bp->bio_data;
1040 ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ?
1041 bp->bio_ma_offset : 0;
1042 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
1043 (bus_dma_segment_t *)bp->bio_data : NULL;
1044
1045 /*
1046 * offs is the offset at which to start operating on the
1047 * next (ie, first) page. lastp is the last page on
1048 * which we're going to operate. lastend is the ending
1049 * position within that last page (ie, PAGE_SIZE if
1050 * we're operating on complete aligned pages).
1051 */
1052 offs = bp->bio_offset % PAGE_SIZE;
1053 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
1054 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
1055
1056 rv = VM_PAGER_OK;
1057 vm_object_pip_add(sc->object, 1);
1058 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
1059 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
1060 m = vm_page_grab_unlocked(sc->object, i, VM_ALLOC_SYSTEM);
1061 if (bp->bio_cmd == BIO_READ) {
1062 if (vm_page_all_valid(m))
1063 rv = VM_PAGER_OK;
1064 else
1065 rv = vm_pager_get_pages(sc->object, &m, 1,
1066 NULL, NULL);
1067 if (rv == VM_PAGER_ERROR) {
1068 VM_OBJECT_WLOCK(sc->object);
1069 vm_page_free(m);
1070 VM_OBJECT_WUNLOCK(sc->object);
1071 break;
1072 } else if (rv == VM_PAGER_FAIL) {
1073 /*
1074 * Pager does not have the page. Zero
1075 * the allocated page, and mark it as
1076 * valid. Do not set dirty, the page
1077 * can be recreated if thrown out.
1078 */
1079 pmap_zero_page(m);
1080 vm_page_valid(m);
1081 }
1082 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1083 pmap_copy_pages(&m, offs, bp->bio_ma,
1084 ma_offs, len);
1085 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1086 physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs,
1087 vlist, ma_offs, len);
1088 cpu_flush_dcache(p, len);
1089 } else {
1090 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1091 cpu_flush_dcache(p, len);
1092 }
1093 } else if (bp->bio_cmd == BIO_WRITE) {
1094 if (len == PAGE_SIZE || vm_page_all_valid(m))
1095 rv = VM_PAGER_OK;
1096 else
1097 rv = vm_pager_get_pages(sc->object, &m, 1,
1098 NULL, NULL);
1099 if (rv == VM_PAGER_ERROR) {
1100 VM_OBJECT_WLOCK(sc->object);
1101 vm_page_free(m);
1102 VM_OBJECT_WUNLOCK(sc->object);
1103 break;
1104 } else if (rv == VM_PAGER_FAIL)
1105 pmap_zero_page(m);
1106
1107 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1108 pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1109 offs, len);
1110 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1111 physcopyin_vlist(vlist, ma_offs,
1112 VM_PAGE_TO_PHYS(m) + offs, len);
1113 } else {
1114 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1115 }
1116
1117 vm_page_valid(m);
1118 vm_page_set_dirty(m);
1119 } else if (bp->bio_cmd == BIO_DELETE) {
1120 if (len == PAGE_SIZE || vm_page_all_valid(m))
1121 rv = VM_PAGER_OK;
1122 else
1123 rv = vm_pager_get_pages(sc->object, &m, 1,
1124 NULL, NULL);
1125 VM_OBJECT_WLOCK(sc->object);
1126 if (rv == VM_PAGER_ERROR) {
1127 vm_page_free(m);
1128 VM_OBJECT_WUNLOCK(sc->object);
1129 break;
1130 } else if (rv == VM_PAGER_FAIL) {
1131 vm_page_free(m);
1132 m = NULL;
1133 } else {
1134 /* Page is valid. */
1135 if (len != PAGE_SIZE) {
1136 pmap_zero_page_area(m, offs, len);
1137 vm_page_set_dirty(m);
1138 } else {
1139 vm_pager_page_unswapped(m);
1140 vm_page_free(m);
1141 m = NULL;
1142 }
1143 }
1144 VM_OBJECT_WUNLOCK(sc->object);
1145 }
1146 if (m != NULL) {
1147 /*
1148 * The page may be deactivated prior to setting
1149 * PGA_REFERENCED, but in this case it will be
1150 * reactivated by the page daemon.
1151 */
1152 if (vm_page_active(m))
1153 vm_page_reference(m);
1154 else
1155 vm_page_activate(m);
1156 vm_page_xunbusy(m);
1157 }
1158
1159 /* Actions on further pages start at offset 0 */
1160 p += PAGE_SIZE - offs;
1161 offs = 0;
1162 ma_offs += len;
1163 }
1164 vm_object_pip_wakeup(sc->object);
1165 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
1166 }
1167
1168 static int
1169 mdstart_null(struct md_s *sc, struct bio *bp)
1170 {
1171
1172 switch (bp->bio_cmd) {
1173 case BIO_READ:
1174 bzero(bp->bio_data, bp->bio_length);
1175 cpu_flush_dcache(bp->bio_data, bp->bio_length);
1176 break;
1177 case BIO_WRITE:
1178 break;
1179 }
1180 bp->bio_resid = 0;
1181 return (0);
1182 }
1183
1184 static void
1185 md_handleattr(struct md_s *sc, struct bio *bp)
1186 {
1187 if (sc->fwsectors && sc->fwheads &&
1188 (g_handleattr_int(bp, "GEOM::fwsectors", sc->fwsectors) != 0 ||
1189 g_handleattr_int(bp, "GEOM::fwheads", sc->fwheads) != 0))
1190 return;
1191 if (g_handleattr_int(bp, "GEOM::candelete", sc->candelete) != 0)
1192 return;
1193 if (sc->ident[0] != '\0' &&
1194 g_handleattr_str(bp, "GEOM::ident", sc->ident) != 0)
1195 return;
1196 if (g_handleattr_int(bp, "MNT::verified", (sc->flags & MD_VERIFY) != 0))
1197 return;
1198 g_io_deliver(bp, EOPNOTSUPP);
1199 }
1200
1201 static void
1202 md_kthread(void *arg)
1203 {
1204 struct md_s *sc;
1205 struct bio *bp;
1206 int error;
1207
1208 sc = arg;
1209 thread_lock(curthread);
1210 sched_prio(curthread, PRIBIO);
1211 thread_unlock(curthread);
1212 if (sc->type == MD_VNODE)
1213 curthread->td_pflags |= TDP_NORUNNINGBUF;
1214
1215 for (;;) {
1216 mtx_lock(&sc->queue_mtx);
1217 if (sc->flags & MD_SHUTDOWN) {
1218 sc->flags |= MD_EXITING;
1219 mtx_unlock(&sc->queue_mtx);
1220 kproc_exit(0);
1221 }
1222 bp = bioq_takefirst(&sc->bio_queue);
1223 if (!bp) {
1224 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1225 continue;
1226 }
1227 mtx_unlock(&sc->queue_mtx);
1228 if (bp->bio_cmd == BIO_GETATTR) {
1229 md_handleattr(sc, bp);
1230 } else {
1231 error = sc->start(sc, bp);
1232 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
1233 /*
1234 * Devstat uses (bio_bcount, bio_resid) for
1235 * determining the length of the completed part
1236 * of the i/o. g_io_deliver() will translate
1237 * from bio_completed to that, but it also
1238 * destroys the bio so we must do our own
1239 * translation.
1240 */
1241 bp->bio_bcount = bp->bio_length;
1242 devstat_end_transaction_bio(sc->devstat, bp);
1243 }
1244 bp->bio_completed = bp->bio_length - bp->bio_resid;
1245 g_io_deliver(bp, error);
1246 }
1247 }
1248 }
1249
1250 static struct md_s *
1251 mdfind(int unit)
1252 {
1253 struct md_s *sc;
1254
1255 LIST_FOREACH(sc, &md_softc_list, list) {
1256 if (sc->unit == unit)
1257 break;
1258 }
1259 return (sc);
1260 }
1261
1262 static struct md_s *
1263 mdnew(int unit, int *errp, enum md_types type)
1264 {
1265 struct md_s *sc;
1266 int error;
1267
1268 *errp = 0;
1269 if (unit == -1)
1270 unit = alloc_unr(md_uh);
1271 else
1272 unit = alloc_unr_specific(md_uh, unit);
1273
1274 if (unit == -1) {
1275 *errp = EBUSY;
1276 return (NULL);
1277 }
1278
1279 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
1280 sc->type = type;
1281 bioq_init(&sc->bio_queue);
1282 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1283 sc->unit = unit;
1284 sprintf(sc->name, "md%d", unit);
1285 LIST_INSERT_HEAD(&md_softc_list, sc, list);
1286 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1287 if (error == 0)
1288 return (sc);
1289 LIST_REMOVE(sc, list);
1290 mtx_destroy(&sc->queue_mtx);
1291 free_unr(md_uh, sc->unit);
1292 free(sc, M_MD);
1293 *errp = error;
1294 return (NULL);
1295 }
1296
1297 static void
1298 mdinit(struct md_s *sc)
1299 {
1300 struct g_geom *gp;
1301 struct g_provider *pp;
1302
1303 g_topology_lock();
1304 gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1305 gp->softc = sc;
1306 pp = g_new_providerf(gp, "md%d", sc->unit);
1307 devstat_remove_entry(pp->stat);
1308 pp->stat = NULL;
1309 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1310 pp->mediasize = sc->mediasize;
1311 pp->sectorsize = sc->sectorsize;
1312 switch (sc->type) {
1313 case MD_MALLOC:
1314 case MD_VNODE:
1315 case MD_SWAP:
1316 pp->flags |= G_PF_ACCEPT_UNMAPPED;
1317 break;
1318 case MD_PRELOAD:
1319 case MD_NULL:
1320 break;
1321 }
1322 sc->gp = gp;
1323 sc->pp = pp;
1324 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1325 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
1326 sc->devstat->id = pp;
1327 g_error_provider(pp, 0);
1328 g_topology_unlock();
1329 }
1330
1331 static int
1332 mdcreate_malloc(struct md_s *sc, struct md_req *mdr)
1333 {
1334 uintptr_t sp;
1335 int error;
1336 off_t u;
1337
1338 error = 0;
1339 if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1340 return (EINVAL);
1341 if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize))
1342 return (EINVAL);
1343 /* Compression doesn't make sense if we have reserved space */
1344 if (mdr->md_options & MD_RESERVE)
1345 mdr->md_options &= ~MD_COMPRESS;
1346 if (mdr->md_fwsectors != 0)
1347 sc->fwsectors = mdr->md_fwsectors;
1348 if (mdr->md_fwheads != 0)
1349 sc->fwheads = mdr->md_fwheads;
1350 sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE);
1351 sc->indir = dimension(sc->mediasize / sc->sectorsize);
1352 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
1353 0x1ff, 0);
1354 if (mdr->md_options & MD_RESERVE) {
1355 off_t nsectors;
1356
1357 nsectors = sc->mediasize / sc->sectorsize;
1358 for (u = 0; u < nsectors; u++) {
1359 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ?
1360 M_WAITOK : M_NOWAIT) | M_ZERO);
1361 if (sp != 0)
1362 error = s_write(sc->indir, u, sp);
1363 else
1364 error = ENOMEM;
1365 if (error != 0)
1366 break;
1367 }
1368 }
1369 return (error);
1370 }
1371
1372 static int
1373 mdsetcred(struct md_s *sc, struct ucred *cred)
1374 {
1375 char *tmpbuf;
1376 int error = 0;
1377
1378 /*
1379 * Set credits in our softc
1380 */
1381
1382 if (sc->cred)
1383 crfree(sc->cred);
1384 sc->cred = crhold(cred);
1385
1386 /*
1387 * Horrible kludge to establish credentials for NFS XXX.
1388 */
1389
1390 if (sc->vnode) {
1391 struct uio auio;
1392 struct iovec aiov;
1393
1394 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1395 bzero(&auio, sizeof(auio));
1396
1397 aiov.iov_base = tmpbuf;
1398 aiov.iov_len = sc->sectorsize;
1399 auio.uio_iov = &aiov;
1400 auio.uio_iovcnt = 1;
1401 auio.uio_offset = 0;
1402 auio.uio_rw = UIO_READ;
1403 auio.uio_segflg = UIO_SYSSPACE;
1404 auio.uio_resid = aiov.iov_len;
1405 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1406 error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
1407 VOP_UNLOCK(sc->vnode);
1408 free(tmpbuf, M_TEMP);
1409 }
1410 return (error);
1411 }
1412
1413 static int
1414 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1415 {
1416 struct vattr vattr;
1417 struct nameidata nd;
1418 char *fname;
1419 int error, flags;
1420 long v;
1421
1422 fname = mdr->md_file;
1423 if (mdr->md_file_seg == UIO_USERSPACE) {
1424 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
1425 if (error != 0)
1426 return (error);
1427 } else if (mdr->md_file_seg == UIO_SYSSPACE)
1428 strlcpy(sc->file, fname, sizeof(sc->file));
1429 else
1430 return (EDOOFUS);
1431
1432 /*
1433 * If the user specified that this is a read only device, don't
1434 * set the FWRITE mask before trying to open the backing store.
1435 */
1436 flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \
1437 | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0);
1438 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file);
1439 error = vn_open(&nd, &flags, 0, NULL);
1440 if (error != 0)
1441 return (error);
1442 NDFREE_PNBUF(&nd);
1443 if (nd.ni_vp->v_type != VREG) {
1444 error = EINVAL;
1445 goto bad;
1446 }
1447 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1448 if (error != 0)
1449 goto bad;
1450 if ((mdr->md_options & MD_MUSTDEALLOC) != 0) {
1451 error = VOP_PATHCONF(nd.ni_vp, _PC_DEALLOC_PRESENT, &v);
1452 if (error != 0)
1453 goto bad;
1454 if (v == 0)
1455 sc->candelete = false;
1456 }
1457 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1458 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1459 if (VN_IS_DOOMED(nd.ni_vp)) {
1460 /* Forced unmount. */
1461 error = EBADF;
1462 goto bad;
1463 }
1464 }
1465 nd.ni_vp->v_vflag |= VV_MD;
1466 VOP_UNLOCK(nd.ni_vp);
1467
1468 if (mdr->md_fwsectors != 0)
1469 sc->fwsectors = mdr->md_fwsectors;
1470 if (mdr->md_fwheads != 0)
1471 sc->fwheads = mdr->md_fwheads;
1472 snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1473 (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1474 sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE |
1475 MD_VERIFY);
1476 if (!(flags & FWRITE))
1477 sc->flags |= MD_READONLY;
1478 sc->vnode = nd.ni_vp;
1479
1480 error = mdsetcred(sc, td->td_ucred);
1481 if (error != 0) {
1482 sc->vnode = NULL;
1483 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1484 nd.ni_vp->v_vflag &= ~VV_MD;
1485 goto bad;
1486 }
1487 return (0);
1488 bad:
1489 VOP_UNLOCK(nd.ni_vp);
1490 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1491 return (error);
1492 }
1493
1494 static void
1495 g_md_providergone(struct g_provider *pp)
1496 {
1497 struct md_s *sc = pp->geom->softc;
1498
1499 mtx_lock(&sc->queue_mtx);
1500 sc->flags |= MD_PROVIDERGONE;
1501 wakeup(&sc->flags);
1502 mtx_unlock(&sc->queue_mtx);
1503 }
1504
1505 static int
1506 mddestroy(struct md_s *sc, struct thread *td)
1507 {
1508
1509 if (sc->gp) {
1510 g_topology_lock();
1511 g_wither_geom(sc->gp, ENXIO);
1512 g_topology_unlock();
1513
1514 mtx_lock(&sc->queue_mtx);
1515 while (!(sc->flags & MD_PROVIDERGONE))
1516 msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0);
1517 mtx_unlock(&sc->queue_mtx);
1518 }
1519 if (sc->devstat) {
1520 devstat_remove_entry(sc->devstat);
1521 sc->devstat = NULL;
1522 }
1523 mtx_lock(&sc->queue_mtx);
1524 sc->flags |= MD_SHUTDOWN;
1525 wakeup(sc);
1526 while (!(sc->flags & MD_EXITING))
1527 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1528 mtx_unlock(&sc->queue_mtx);
1529 mtx_destroy(&sc->queue_mtx);
1530 if (sc->vnode != NULL) {
1531 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1532 sc->vnode->v_vflag &= ~VV_MD;
1533 VOP_UNLOCK(sc->vnode);
1534 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1535 FREAD : (FREAD|FWRITE), sc->cred, td);
1536 }
1537 if (sc->cred != NULL)
1538 crfree(sc->cred);
1539 if (sc->object != NULL)
1540 vm_object_deallocate(sc->object);
1541 if (sc->indir)
1542 destroy_indir(sc, sc->indir);
1543 if (sc->uma)
1544 uma_zdestroy(sc->uma);
1545
1546 LIST_REMOVE(sc, list);
1547 free_unr(md_uh, sc->unit);
1548 free(sc, M_MD);
1549 return (0);
1550 }
1551
1552 static int
1553 mdresize(struct md_s *sc, struct md_req *mdr)
1554 {
1555 int error, res;
1556 vm_pindex_t oldpages, newpages;
1557
1558 switch (sc->type) {
1559 case MD_VNODE:
1560 case MD_NULL:
1561 break;
1562 case MD_SWAP:
1563 if (mdr->md_mediasize <= 0 ||
1564 (mdr->md_mediasize % PAGE_SIZE) != 0)
1565 return (EDOM);
1566 oldpages = OFF_TO_IDX(sc->mediasize);
1567 newpages = OFF_TO_IDX(mdr->md_mediasize);
1568 if (newpages < oldpages) {
1569 VM_OBJECT_WLOCK(sc->object);
1570 vm_object_page_remove(sc->object, newpages, 0, 0);
1571 swap_release_by_cred(IDX_TO_OFF(oldpages -
1572 newpages), sc->cred);
1573 sc->object->charge = IDX_TO_OFF(newpages);
1574 sc->object->size = newpages;
1575 VM_OBJECT_WUNLOCK(sc->object);
1576 } else if (newpages > oldpages) {
1577 res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1578 oldpages), sc->cred);
1579 if (!res)
1580 return (ENOMEM);
1581 if ((mdr->md_options & MD_RESERVE) ||
1582 (sc->flags & MD_RESERVE)) {
1583 error = swap_pager_reserve(sc->object,
1584 oldpages, newpages - oldpages);
1585 if (error < 0) {
1586 swap_release_by_cred(
1587 IDX_TO_OFF(newpages - oldpages),
1588 sc->cred);
1589 return (EDOM);
1590 }
1591 }
1592 VM_OBJECT_WLOCK(sc->object);
1593 sc->object->charge = IDX_TO_OFF(newpages);
1594 sc->object->size = newpages;
1595 VM_OBJECT_WUNLOCK(sc->object);
1596 }
1597 break;
1598 default:
1599 return (EOPNOTSUPP);
1600 }
1601
1602 sc->mediasize = mdr->md_mediasize;
1603
1604 g_topology_lock();
1605 g_resize_provider(sc->pp, sc->mediasize);
1606 g_topology_unlock();
1607 return (0);
1608 }
1609
1610 static int
1611 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1612 {
1613 vm_ooffset_t npage;
1614 int error;
1615
1616 /*
1617 * Range check. Disallow negative sizes and sizes not being
1618 * multiple of page size.
1619 */
1620 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1621 return (EDOM);
1622
1623 /*
1624 * Allocate an OBJT_SWAP object.
1625 *
1626 * Note the truncation.
1627 */
1628
1629 if ((mdr->md_options & MD_VERIFY) != 0)
1630 return (EINVAL);
1631 npage = mdr->md_mediasize / PAGE_SIZE;
1632 if (mdr->md_fwsectors != 0)
1633 sc->fwsectors = mdr->md_fwsectors;
1634 if (mdr->md_fwheads != 0)
1635 sc->fwheads = mdr->md_fwheads;
1636 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1637 VM_PROT_DEFAULT, 0, td->td_ucred);
1638 if (sc->object == NULL)
1639 return (ENOMEM);
1640 sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1641 if (mdr->md_options & MD_RESERVE) {
1642 if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1643 error = EDOM;
1644 goto finish;
1645 }
1646 }
1647 error = mdsetcred(sc, td->td_ucred);
1648 finish:
1649 if (error != 0) {
1650 vm_object_deallocate(sc->object);
1651 sc->object = NULL;
1652 }
1653 return (error);
1654 }
1655
1656 static int
1657 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1658 {
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 return (0);
1668 }
1669
1670 static int
1671 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1672 {
1673 struct md_s *sc;
1674 unsigned sectsize;
1675 int error, i;
1676
1677 sx_assert(&md_sx, SA_XLOCKED);
1678
1679 switch (mdr->md_type) {
1680 case MD_MALLOC:
1681 case MD_PRELOAD:
1682 case MD_VNODE:
1683 case MD_SWAP:
1684 case MD_NULL:
1685 break;
1686 default:
1687 return (EINVAL);
1688 }
1689 if (mdr->md_sectorsize == 0)
1690 sectsize = DEV_BSIZE;
1691 else
1692 sectsize = mdr->md_sectorsize;
1693 if (sectsize > maxphys || mdr->md_mediasize < sectsize)
1694 return (EINVAL);
1695 if (mdr->md_options & MD_AUTOUNIT)
1696 sc = mdnew(-1, &error, mdr->md_type);
1697 else {
1698 if (mdr->md_unit > INT_MAX)
1699 return (EINVAL);
1700 sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1701 }
1702 if (sc == NULL)
1703 return (error);
1704 if (mdr->md_label != NULL)
1705 error = copyinstr(mdr->md_label, sc->label,
1706 sizeof(sc->label), NULL);
1707 if (error != 0)
1708 goto err_after_new;
1709 if (mdr->md_options & MD_AUTOUNIT)
1710 mdr->md_unit = sc->unit;
1711 sc->mediasize = mdr->md_mediasize;
1712 sc->sectorsize = sectsize;
1713 sc->candelete = true;
1714 error = EDOOFUS;
1715 switch (sc->type) {
1716 case MD_MALLOC:
1717 sc->start = mdstart_malloc;
1718 error = mdcreate_malloc(sc, mdr);
1719 break;
1720 case MD_PRELOAD:
1721 /*
1722 * We disallow attaching preloaded memory disks via
1723 * ioctl. Preloaded memory disks are automatically
1724 * attached in g_md_init().
1725 */
1726 error = EOPNOTSUPP;
1727 break;
1728 case MD_VNODE:
1729 sc->start = mdstart_vnode;
1730 error = mdcreate_vnode(sc, mdr, td);
1731 break;
1732 case MD_SWAP:
1733 sc->start = mdstart_swap;
1734 error = mdcreate_swap(sc, mdr, td);
1735 break;
1736 case MD_NULL:
1737 sc->start = mdstart_null;
1738 error = mdcreate_null(sc, mdr, td);
1739 break;
1740 }
1741 err_after_new:
1742 if (error != 0) {
1743 mddestroy(sc, td);
1744 return (error);
1745 }
1746
1747 /* Prune off any residual fractional sector */
1748 i = sc->mediasize % sc->sectorsize;
1749 sc->mediasize -= i;
1750
1751 mdinit(sc);
1752 return (0);
1753 }
1754
1755 static int
1756 kern_mdattach(struct thread *td, struct md_req *mdr)
1757 {
1758 int error;
1759
1760 sx_xlock(&md_sx);
1761 error = kern_mdattach_locked(td, mdr);
1762 sx_xunlock(&md_sx);
1763 return (error);
1764 }
1765
1766 static int
1767 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1768 {
1769 struct md_s *sc;
1770
1771 sx_assert(&md_sx, SA_XLOCKED);
1772
1773 if (mdr->md_mediasize != 0 ||
1774 (mdr->md_options & ~MD_FORCE) != 0)
1775 return (EINVAL);
1776
1777 sc = mdfind(mdr->md_unit);
1778 if (sc == NULL)
1779 return (ENOENT);
1780 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1781 !(mdr->md_options & MD_FORCE))
1782 return (EBUSY);
1783 return (mddestroy(sc, td));
1784 }
1785
1786 static int
1787 kern_mddetach(struct thread *td, struct md_req *mdr)
1788 {
1789 int error;
1790
1791 sx_xlock(&md_sx);
1792 error = kern_mddetach_locked(td, mdr);
1793 sx_xunlock(&md_sx);
1794 return (error);
1795 }
1796
1797 static int
1798 kern_mdresize_locked(struct md_req *mdr)
1799 {
1800 struct md_s *sc;
1801
1802 sx_assert(&md_sx, SA_XLOCKED);
1803
1804 if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1805 return (EINVAL);
1806
1807 sc = mdfind(mdr->md_unit);
1808 if (sc == NULL)
1809 return (ENOENT);
1810 if (mdr->md_mediasize < sc->sectorsize)
1811 return (EINVAL);
1812 mdr->md_mediasize -= mdr->md_mediasize % sc->sectorsize;
1813 if (mdr->md_mediasize < sc->mediasize &&
1814 !(sc->flags & MD_FORCE) &&
1815 !(mdr->md_options & MD_FORCE))
1816 return (EBUSY);
1817 return (mdresize(sc, mdr));
1818 }
1819
1820 static int
1821 kern_mdresize(struct md_req *mdr)
1822 {
1823 int error;
1824
1825 sx_xlock(&md_sx);
1826 error = kern_mdresize_locked(mdr);
1827 sx_xunlock(&md_sx);
1828 return (error);
1829 }
1830
1831 static int
1832 kern_mdquery_locked(struct md_req *mdr)
1833 {
1834 struct md_s *sc;
1835 int error;
1836
1837 sx_assert(&md_sx, SA_XLOCKED);
1838
1839 sc = mdfind(mdr->md_unit);
1840 if (sc == NULL)
1841 return (ENOENT);
1842 mdr->md_type = sc->type;
1843 mdr->md_options = sc->flags;
1844 mdr->md_mediasize = sc->mediasize;
1845 mdr->md_sectorsize = sc->sectorsize;
1846 error = 0;
1847 if (mdr->md_label != NULL) {
1848 error = copyout(sc->label, mdr->md_label,
1849 strlen(sc->label) + 1);
1850 if (error != 0)
1851 return (error);
1852 }
1853 if (sc->type == MD_VNODE ||
1854 (sc->type == MD_PRELOAD && mdr->md_file != NULL))
1855 error = copyout(sc->file, mdr->md_file,
1856 strlen(sc->file) + 1);
1857 return (error);
1858 }
1859
1860 static int
1861 kern_mdquery(struct md_req *mdr)
1862 {
1863 int error;
1864
1865 sx_xlock(&md_sx);
1866 error = kern_mdquery_locked(mdr);
1867 sx_xunlock(&md_sx);
1868 return (error);
1869 }
1870
1871 /* Copy members that are not userspace pointers. */
1872 #define MD_IOCTL2REQ(mdio, mdr) do { \
1873 (mdr)->md_unit = (mdio)->md_unit; \
1874 (mdr)->md_type = (mdio)->md_type; \
1875 (mdr)->md_mediasize = (mdio)->md_mediasize; \
1876 (mdr)->md_sectorsize = (mdio)->md_sectorsize; \
1877 (mdr)->md_options = (mdio)->md_options; \
1878 (mdr)->md_fwheads = (mdio)->md_fwheads; \
1879 (mdr)->md_fwsectors = (mdio)->md_fwsectors; \
1880 (mdr)->md_units = &(mdio)->md_pad[0]; \
1881 (mdr)->md_units_nitems = nitems((mdio)->md_pad); \
1882 } while(0)
1883
1884 /* Copy members that might have been updated */
1885 #define MD_REQ2IOCTL(mdr, mdio) do { \
1886 (mdio)->md_unit = (mdr)->md_unit; \
1887 (mdio)->md_type = (mdr)->md_type; \
1888 (mdio)->md_mediasize = (mdr)->md_mediasize; \
1889 (mdio)->md_sectorsize = (mdr)->md_sectorsize; \
1890 (mdio)->md_options = (mdr)->md_options; \
1891 (mdio)->md_fwheads = (mdr)->md_fwheads; \
1892 (mdio)->md_fwsectors = (mdr)->md_fwsectors; \
1893 } while(0)
1894
1895 static int
1896 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1897 struct thread *td)
1898 {
1899 struct md_req mdr;
1900 int error;
1901
1902 if (md_debug)
1903 printf("mdctlioctl(%s %lx %p %x %p)\n",
1904 devtoname(dev), cmd, addr, flags, td);
1905
1906 bzero(&mdr, sizeof(mdr));
1907 switch (cmd) {
1908 case MDIOCATTACH:
1909 case MDIOCDETACH:
1910 case MDIOCRESIZE:
1911 case MDIOCQUERY: {
1912 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1913 if (mdio->md_version != MDIOVERSION)
1914 return (EINVAL);
1915 MD_IOCTL2REQ(mdio, &mdr);
1916 mdr.md_file = mdio->md_file;
1917 mdr.md_file_seg = UIO_USERSPACE;
1918 /* If the file is adjacent to the md_ioctl it's in kernel. */
1919 if ((void *)mdio->md_file == (void *)(mdio + 1))
1920 mdr.md_file_seg = UIO_SYSSPACE;
1921 mdr.md_label = mdio->md_label;
1922 break;
1923 }
1924 #ifdef COMPAT_FREEBSD32
1925 case MDIOCATTACH_32:
1926 case MDIOCDETACH_32:
1927 case MDIOCRESIZE_32:
1928 case MDIOCQUERY_32: {
1929 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1930 if (mdio->md_version != MDIOVERSION)
1931 return (EINVAL);
1932 MD_IOCTL2REQ(mdio, &mdr);
1933 mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1934 mdr.md_file_seg = UIO_USERSPACE;
1935 mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1936 break;
1937 }
1938 #endif
1939 default:
1940 /* Fall through to handler switch. */
1941 break;
1942 }
1943
1944 error = 0;
1945 switch (cmd) {
1946 case MDIOCATTACH:
1947 #ifdef COMPAT_FREEBSD32
1948 case MDIOCATTACH_32:
1949 #endif
1950 error = kern_mdattach(td, &mdr);
1951 break;
1952 case MDIOCDETACH:
1953 #ifdef COMPAT_FREEBSD32
1954 case MDIOCDETACH_32:
1955 #endif
1956 error = kern_mddetach(td, &mdr);
1957 break;
1958 case MDIOCRESIZE:
1959 #ifdef COMPAT_FREEBSD32
1960 case MDIOCRESIZE_32:
1961 #endif
1962 error = kern_mdresize(&mdr);
1963 break;
1964 case MDIOCQUERY:
1965 #ifdef COMPAT_FREEBSD32
1966 case MDIOCQUERY_32:
1967 #endif
1968 error = kern_mdquery(&mdr);
1969 break;
1970 default:
1971 error = ENOIOCTL;
1972 }
1973
1974 switch (cmd) {
1975 case MDIOCATTACH:
1976 case MDIOCQUERY: {
1977 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1978 MD_REQ2IOCTL(&mdr, mdio);
1979 break;
1980 }
1981 #ifdef COMPAT_FREEBSD32
1982 case MDIOCATTACH_32:
1983 case MDIOCQUERY_32: {
1984 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1985 MD_REQ2IOCTL(&mdr, mdio);
1986 break;
1987 }
1988 #endif
1989 default:
1990 /* Other commands to not alter mdr. */
1991 break;
1992 }
1993
1994 return (error);
1995 }
1996
1997 static void
1998 md_preloaded(u_char *image, size_t length, const char *name)
1999 {
2000 struct md_s *sc;
2001 int error;
2002
2003 sc = mdnew(-1, &error, MD_PRELOAD);
2004 if (sc == NULL)
2005 return;
2006 sc->mediasize = length;
2007 sc->sectorsize = DEV_BSIZE;
2008 sc->pl_ptr = image;
2009 sc->pl_len = length;
2010 sc->start = mdstart_preload;
2011 if (name != NULL)
2012 strlcpy(sc->file, name, sizeof(sc->file));
2013 #ifdef MD_ROOT
2014 if (sc->unit == 0) {
2015 #ifndef ROOTDEVNAME
2016 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2017 #endif
2018 #ifdef MD_ROOT_READONLY
2019 sc->flags |= MD_READONLY;
2020 #endif
2021 }
2022 #endif
2023 mdinit(sc);
2024 if (name != NULL) {
2025 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2026 MD_NAME, sc->unit, name, length, image);
2027 } else {
2028 printf("%s%d: Embedded image %zd bytes at %p\n",
2029 MD_NAME, sc->unit, length, image);
2030 }
2031 }
2032
2033 static void
2034 g_md_init(struct g_class *mp __unused)
2035 {
2036 caddr_t mod;
2037 u_char *ptr, *name, *type;
2038 unsigned len;
2039 int i;
2040
2041 /* figure out log2(NINDIR) */
2042 for (i = NINDIR, nshift = -1; i; nshift++)
2043 i >>= 1;
2044
2045 mod = NULL;
2046 sx_init(&md_sx, "MD config lock");
2047 g_topology_unlock();
2048 md_uh = new_unrhdr(0, INT_MAX, NULL);
2049 #ifdef MD_ROOT
2050 if (mfs_root_size != 0) {
2051 sx_xlock(&md_sx);
2052 #ifdef MD_ROOT_MEM
2053 md_preloaded(mfs_root, mfs_root_size, NULL);
2054 #else
2055 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2056 NULL);
2057 #endif
2058 sx_xunlock(&md_sx);
2059 }
2060 #endif
2061 /* XXX: are preload_* static or do they need Giant ? */
2062 while ((mod = preload_search_next_name(mod)) != NULL) {
2063 name = (char *)preload_search_info(mod, MODINFO_NAME);
2064 if (name == NULL)
2065 continue;
2066 type = (char *)preload_search_info(mod, MODINFO_TYPE);
2067 if (type == NULL)
2068 continue;
2069 if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2070 continue;
2071 ptr = preload_fetch_addr(mod);
2072 len = preload_fetch_size(mod);
2073 if (ptr != NULL && len != 0) {
2074 sx_xlock(&md_sx);
2075 md_preloaded(ptr, len, name);
2076 sx_xunlock(&md_sx);
2077 }
2078 }
2079 md_pbuf_zone = pbuf_zsecond_create("mdpbuf", nswbuf / 10);
2080 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2081 0600, MDCTL_NAME);
2082 g_topology_lock();
2083 }
2084
2085 static void
2086 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2087 struct g_consumer *cp __unused, struct g_provider *pp)
2088 {
2089 struct md_s *mp;
2090 char *type;
2091
2092 mp = gp->softc;
2093 if (mp == NULL)
2094 return;
2095
2096 switch (mp->type) {
2097 case MD_MALLOC:
2098 type = "malloc";
2099 break;
2100 case MD_PRELOAD:
2101 type = "preload";
2102 break;
2103 case MD_VNODE:
2104 type = "vnode";
2105 break;
2106 case MD_SWAP:
2107 type = "swap";
2108 break;
2109 case MD_NULL:
2110 type = "null";
2111 break;
2112 default:
2113 type = "unknown";
2114 break;
2115 }
2116
2117 if (pp != NULL) {
2118 if (indent == NULL) {
2119 sbuf_printf(sb, " u %d", mp->unit);
2120 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2121 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2122 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2123 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2124 sbuf_printf(sb, " t %s", type);
2125 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2126 (mp->type == MD_PRELOAD && mp->file[0] != '\0'))
2127 sbuf_printf(sb, " file %s", mp->file);
2128 sbuf_printf(sb, " label %s", mp->label);
2129 } else {
2130 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2131 mp->unit);
2132 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2133 indent, (uintmax_t) mp->sectorsize);
2134 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2135 indent, (uintmax_t) mp->fwheads);
2136 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2137 indent, (uintmax_t) mp->fwsectors);
2138 if (mp->ident[0] != '\0') {
2139 sbuf_printf(sb, "%s<ident>", indent);
2140 g_conf_printf_escaped(sb, "%s", mp->ident);
2141 sbuf_printf(sb, "</ident>\n");
2142 }
2143 sbuf_printf(sb, "%s<length>%ju</length>\n",
2144 indent, (uintmax_t) mp->mediasize);
2145 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2146 (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2147 sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2148 (mp->flags & MD_READONLY) == 0 ? "read-write":
2149 "read-only");
2150 sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2151 type);
2152 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2153 (mp->type == MD_PRELOAD && mp->file[0] != '\0')) {
2154 sbuf_printf(sb, "%s<file>", indent);
2155 g_conf_printf_escaped(sb, "%s", mp->file);
2156 sbuf_printf(sb, "</file>\n");
2157 }
2158 if (mp->type == MD_VNODE)
2159 sbuf_printf(sb, "%s<cache>%s</cache>\n", indent,
2160 (mp->flags & MD_CACHE) == 0 ? "off": "on");
2161 sbuf_printf(sb, "%s<label>", indent);
2162 g_conf_printf_escaped(sb, "%s", mp->label);
2163 sbuf_printf(sb, "</label>\n");
2164 }
2165 }
2166 }
2167
2168 static void
2169 g_md_fini(struct g_class *mp __unused)
2170 {
2171
2172 sx_destroy(&md_sx);
2173 if (status_dev != NULL)
2174 destroy_dev(status_dev);
2175 uma_zdestroy(md_pbuf_zone);
2176 delete_unrhdr(md_uh);
2177 }
Cache object: 173d8cff8d9f205f7751f5f42896c558
|