2 * SPDX-License-Identifier: (Beerware AND BSD-3-Clause)
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 * ----------------------------------------------------------------------------
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:
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.
26 * This code is derived from software contributed to Berkeley by
27 * the Systems Programming Group of the University of Utah Computer
30 * Portions of this software were developed by Konstantin Belousov
31 * under sponsorship from the FreeBSD Foundation.
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
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.
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
57 * from: Utah Hdr: vn.c 1.13 94/04/02
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
63 #include "opt_rootdevname.h"
67 #include <sys/param.h>
68 #include <sys/systm.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>
79 #include <sys/malloc.h>
80 #include <sys/mdioctl.h>
81 #include <sys/mount.h>
82 #include <sys/mutex.h>
84 #include <sys/namei.h>
86 #include <sys/queue.h>
87 #include <sys/rwlock.h>
89 #include <sys/sched.h>
90 #include <sys/sf_buf.h>
91 #include <sys/sysctl.h>
93 #include <sys/vnode.h>
96 #include <geom/geom.h>
97 #include <geom/geom_int.h>
100 #include <vm/vm_param.h>
101 #include <vm/vm_object.h>
102 #include <vm/vm_page.h>
103 #include <vm/vm_pager.h>
104 #include <vm/swap_pager.h>
107 #include <machine/bus.h>
111 #define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */
112 #define MD_EXITING 0x20000 /* Worker thread is exiting. */
113 #define MD_PROVIDERGONE 0x40000 /* Safe to free the softc */
116 #define MD_NSECT (10000 * 2)
120 unsigned md_unit; /* unit number */
121 enum md_types md_type; /* type of disk */
122 off_t md_mediasize; /* size of disk in bytes */
123 unsigned md_sectorsize; /* sectorsize */
124 unsigned md_options; /* options */
125 int md_fwheads; /* firmware heads */
126 int md_fwsectors; /* firmware sectors */
127 char *md_file; /* pathname of file to mount */
128 enum uio_seg md_file_seg; /* location of md_file */
129 char *md_label; /* label of the device (userspace) */
130 int *md_units; /* pointer to units array (kernel) */
131 size_t md_units_nitems; /* items in md_units array */
134 #ifdef COMPAT_FREEBSD32
138 enum md_types md_type;
141 unsigned md_sectorsize;
148 } __attribute__((__packed__));
149 CTASSERT((sizeof(struct md_ioctl32)) == 436);
151 #define MDIOCATTACH_32 _IOC_NEWTYPE(MDIOCATTACH, struct md_ioctl32)
152 #define MDIOCDETACH_32 _IOC_NEWTYPE(MDIOCDETACH, struct md_ioctl32)
153 #define MDIOCQUERY_32 _IOC_NEWTYPE(MDIOCQUERY, struct md_ioctl32)
154 #define MDIOCRESIZE_32 _IOC_NEWTYPE(MDIOCRESIZE, struct md_ioctl32)
155 #endif /* COMPAT_FREEBSD32 */
157 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk");
158 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors");
161 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0,
162 "Enable md(4) debug messages");
163 static int md_malloc_wait;
164 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0,
165 "Allow malloc to wait for memory allocations");
167 #if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE)
168 #define MD_ROOT_FSTYPE "ufs"
173 * Preloaded image gets put here.
175 #if defined(MD_ROOT_SIZE)
177 * We put the mfs_root symbol into the oldmfs section of the kernel object file.
178 * Applications that patch the object with the image can determine
179 * the size looking at the oldmfs section size within the kernel.
181 u_char mfs_root[MD_ROOT_SIZE*1024] __attribute__ ((section ("oldmfs")));
182 const int mfs_root_size = sizeof(mfs_root);
183 #elif defined(MD_ROOT_MEM)
184 /* MD region already mapped in the memory */
188 extern volatile u_char __weak_symbol mfs_root;
189 extern volatile u_char __weak_symbol mfs_root_end;
190 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root))
194 static g_init_t g_md_init;
195 static g_fini_t g_md_fini;
196 static g_start_t g_md_start;
197 static g_access_t g_md_access;
198 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
199 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
200 static g_provgone_t g_md_providergone;
202 static struct cdev *status_dev = NULL;
203 static struct sx md_sx;
204 static struct unrhdr *md_uh;
206 static d_ioctl_t mdctlioctl;
208 static struct cdevsw mdctl_cdevsw = {
209 .d_version = D_VERSION,
210 .d_ioctl = mdctlioctl,
214 struct g_class g_md_class = {
216 .version = G_VERSION,
220 .access = g_md_access,
221 .dumpconf = g_md_dumpconf,
222 .providergone = g_md_providergone,
225 DECLARE_GEOM_CLASS(g_md_class, g_md);
227 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
229 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t))
230 #define NMASK (NINDIR-1)
233 static uma_zone_t md_pbuf_zone;
244 LIST_ENTRY(md_s) list;
245 struct bio_queue_head bio_queue;
246 struct mtx queue_mtx;
259 struct g_provider *pp;
260 int (*start)(struct md_s *sc, struct bio *bp);
261 struct devstat *devstat;
263 /* MD_MALLOC related fields */
267 /* MD_PRELOAD related fields */
271 /* MD_VNODE related fields */
274 char label[PATH_MAX];
277 /* MD_SWAP related fields */
281 static struct indir *
282 new_indir(u_int shift)
286 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
290 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
291 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
292 if (ip->array == NULL) {
302 del_indir(struct indir *ip)
305 free(ip->array, M_MDSECT);
310 destroy_indir(struct md_s *sc, struct indir *ip)
314 for (i = 0; i < NINDIR; i++) {
318 destroy_indir(sc, (struct indir*)(ip->array[i]));
319 else if (ip->array[i] > 255)
320 uma_zfree(sc->uma, (void *)(ip->array[i]));
326 * This function does the math and allocates the top level "indir" structure
327 * for a device of "size" sectors.
330 static struct indir *
331 dimension(off_t size)
339 while (rcnt > NINDIR) {
345 * XXX: the top layer is probably not fully populated, so we allocate
346 * too much space for ip->array in here.
348 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
349 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
350 M_MDSECT, M_WAITOK | M_ZERO);
352 ip->shift = layer * nshift;
357 * Read a given sector
361 s_read(struct indir *ip, off_t offset)
368 printf("s_read(%jd)\n", (intmax_t)offset);
370 for (cip = ip; cip != NULL;) {
372 idx = (offset >> cip->shift) & NMASK;
373 up = cip->array[idx];
374 cip = (struct indir *)up;
377 idx = offset & NMASK;
378 return (cip->array[idx]);
384 * Write a given sector, prune the tree if the value is 0
388 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
390 struct indir *cip, *lip[10];
395 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
402 idx = (offset >> cip->shift) & NMASK;
403 up = cip->array[idx];
405 cip = (struct indir *)up;
408 /* Allocate branch */
410 (uintptr_t)new_indir(cip->shift - nshift);
411 if (cip->array[idx] == 0)
414 up = cip->array[idx];
415 cip = (struct indir *)up;
419 idx = offset & NMASK;
420 up = cip->array[idx];
423 cip->array[idx] = ptr;
428 if (cip->used != 0 || li == 1)
431 while (cip->used == 0 && cip != ip) {
433 idx = (offset >> lip[li]->shift) & NMASK;
434 up = lip[li]->array[idx];
435 KASSERT(up == (uintptr_t)cip, ("md screwed up"));
437 lip[li]->array[idx] = 0;
445 g_md_access(struct g_provider *pp, int r, int w, int e)
449 sc = pp->geom->softc;
451 if (r <= 0 && w <= 0 && e <= 0)
458 if ((sc->flags & MD_READONLY) != 0 && w > 0)
460 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
462 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
469 g_md_start(struct bio *bp)
473 sc = bp->bio_to->geom->softc;
474 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
475 devstat_start_transaction_bio(sc->devstat, bp);
477 mtx_lock(&sc->queue_mtx);
478 bioq_disksort(&sc->bio_queue, bp);
480 mtx_unlock(&sc->queue_mtx);
483 #define MD_MALLOC_MOVE_ZERO 1
484 #define MD_MALLOC_MOVE_FILL 2
485 #define MD_MALLOC_MOVE_READ 3
486 #define MD_MALLOC_MOVE_WRITE 4
487 #define MD_MALLOC_MOVE_CMP 5
490 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize,
491 void *ptr, u_char fill, int op)
498 int error, i, ma_offs1, sz, first_read;
503 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */
511 for (n = sectorsize; n != 0; n -= sz) {
512 sz = imin(PAGE_SIZE - *ma_offs, n);
517 sf = sf_buf_alloc(m, SFB_CPUPRIVATE |
518 (md_malloc_wait ? 0 : SFB_NOWAIT));
524 p = (char *)sf_buf_kva(sf) + *ma_offs;
526 case MD_MALLOC_MOVE_ZERO:
529 case MD_MALLOC_MOVE_FILL:
532 case MD_MALLOC_MOVE_READ:
534 cpu_flush_dcache(p, sz);
536 case MD_MALLOC_MOVE_WRITE:
539 case MD_MALLOC_MOVE_CMP:
540 for (i = 0; i < sz; i++, p++) {
545 } else if (*p != first) {
552 KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op));
558 *ma_offs %= PAGE_SIZE;
561 ptr = (char *)ptr + sz;
567 if (op == MD_MALLOC_MOVE_CMP && error != 0) {
575 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs,
576 unsigned len, void *ptr, u_char fill, int op)
578 bus_dma_segment_t *vlist;
579 uint8_t *p, *end, first;
581 int ma_offs, seg_len;
587 for (; len != 0; len -= seg_len) {
588 seg_len = imin(vlist->ds_len - ma_offs, len);
589 p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs;
591 case MD_MALLOC_MOVE_ZERO:
594 case MD_MALLOC_MOVE_FILL:
595 memset(p, fill, seg_len);
597 case MD_MALLOC_MOVE_READ:
598 bcopy(ptr, p, seg_len);
599 cpu_flush_dcache(p, seg_len);
601 case MD_MALLOC_MOVE_WRITE:
602 bcopy(p, ptr, seg_len);
604 case MD_MALLOC_MOVE_CMP:
607 /* Confirm all following bytes match the first */
614 KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op));
619 if (ma_offs == vlist->ds_len) {
623 ptr = (uint8_t *)ptr + seg_len;
632 mdstart_malloc(struct md_s *sc, struct bio *bp)
636 bus_dma_segment_t *vlist;
637 int i, error, error1, ma_offs, notmapped;
638 off_t secno, nsec, uc;
641 switch (bp->bio_cmd) {
650 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0;
651 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
652 (bus_dma_segment_t *)bp->bio_data : NULL;
655 ma_offs = bp->bio_ma_offset;
657 KASSERT(vlist == NULL, ("vlists cannot be unmapped"));
658 } else if (vlist != NULL) {
659 ma_offs = bp->bio_ma_offset;
665 nsec = bp->bio_length / sc->sectorsize;
666 secno = bp->bio_offset / sc->sectorsize;
669 osp = s_read(sc->indir, secno);
670 if (bp->bio_cmd == BIO_DELETE) {
672 error = s_write(sc->indir, secno, 0);
673 } else if (bp->bio_cmd == BIO_READ) {
676 error = md_malloc_move_ma(&m, &ma_offs,
677 sc->sectorsize, NULL, 0,
678 MD_MALLOC_MOVE_ZERO);
679 } else if (vlist != NULL) {
680 error = md_malloc_move_vlist(&vlist,
681 &ma_offs, sc->sectorsize, NULL, 0,
682 MD_MALLOC_MOVE_ZERO);
684 bzero(dst, sc->sectorsize);
685 } else if (osp <= 255) {
687 error = md_malloc_move_ma(&m, &ma_offs,
688 sc->sectorsize, NULL, osp,
689 MD_MALLOC_MOVE_FILL);
690 } else if (vlist != NULL) {
691 error = md_malloc_move_vlist(&vlist,
692 &ma_offs, sc->sectorsize, NULL, osp,
693 MD_MALLOC_MOVE_FILL);
695 memset(dst, osp, sc->sectorsize);
698 error = md_malloc_move_ma(&m, &ma_offs,
699 sc->sectorsize, (void *)osp, 0,
700 MD_MALLOC_MOVE_READ);
701 } else if (vlist != NULL) {
702 error = md_malloc_move_vlist(&vlist,
703 &ma_offs, sc->sectorsize,
705 MD_MALLOC_MOVE_READ);
707 bcopy((void *)osp, dst, sc->sectorsize);
708 cpu_flush_dcache(dst, sc->sectorsize);
712 } else if (bp->bio_cmd == BIO_WRITE) {
713 if (sc->flags & MD_COMPRESS) {
715 error1 = md_malloc_move_ma(&m, &ma_offs,
716 sc->sectorsize, &uc, 0,
718 i = error1 == 0 ? sc->sectorsize : 0;
719 } else if (vlist != NULL) {
720 error1 = md_malloc_move_vlist(&vlist,
721 &ma_offs, sc->sectorsize, &uc, 0,
723 i = error1 == 0 ? sc->sectorsize : 0;
726 for (i = 1; i < sc->sectorsize; i++) {
735 if (i == sc->sectorsize) {
737 error = s_write(sc->indir, secno, uc);
740 sp = (uintptr_t)uma_zalloc(sc->uma,
741 md_malloc_wait ? M_WAITOK :
748 error = md_malloc_move_ma(&m,
749 &ma_offs, sc->sectorsize,
751 MD_MALLOC_MOVE_WRITE);
752 } else if (vlist != NULL) {
753 error = md_malloc_move_vlist(
755 sc->sectorsize, (void *)sp,
756 0, MD_MALLOC_MOVE_WRITE);
758 bcopy(dst, (void *)sp,
761 error = s_write(sc->indir, secno, sp);
764 error = md_malloc_move_ma(&m,
765 &ma_offs, sc->sectorsize,
767 MD_MALLOC_MOVE_WRITE);
768 } else if (vlist != NULL) {
769 error = md_malloc_move_vlist(
771 sc->sectorsize, (void *)osp,
772 0, MD_MALLOC_MOVE_WRITE);
774 bcopy(dst, (void *)osp,
784 uma_zfree(sc->uma, (void*)osp);
788 if (!notmapped && vlist == NULL)
789 dst += sc->sectorsize;
796 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len)
800 while (offset >= vlist->ds_len) {
801 offset -= vlist->ds_len;
806 seg_len = omin(len, vlist->ds_len - offset);
807 bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset),
810 src = (uint8_t *)src + seg_len;
817 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len)
821 while (offset >= vlist->ds_len) {
822 offset -= vlist->ds_len;
827 seg_len = omin(len, vlist->ds_len - offset);
828 bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst,
831 dst = (uint8_t *)dst + seg_len;
838 mdstart_preload(struct md_s *sc, struct bio *bp)
842 p = sc->pl_ptr + bp->bio_offset;
843 switch (bp->bio_cmd) {
845 if ((bp->bio_flags & BIO_VLIST) != 0) {
846 mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data,
847 bp->bio_ma_offset, bp->bio_length);
849 bcopy(p, bp->bio_data, bp->bio_length);
851 cpu_flush_dcache(bp->bio_data, bp->bio_length);
854 if ((bp->bio_flags & BIO_VLIST) != 0) {
855 mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data,
856 bp->bio_ma_offset, p, bp->bio_length);
858 bcopy(bp->bio_data, p, bp->bio_length);
867 mdstart_vnode(struct md_s *sc, struct bio *bp)
876 bus_dma_segment_t *vlist;
878 off_t iolen, iostart, len, zerosize;
881 switch (bp->bio_cmd) {
883 auio.uio_rw = UIO_READ;
887 auio.uio_rw = UIO_WRITE;
899 ma_offs = bp->bio_ma_offset;
900 len = bp->bio_length;
905 * If an error occurs, we set BIO_ERROR but we do not set
906 * B_INVAL because (for a write anyway), the buffer is
910 if (bp->bio_cmd == BIO_FLUSH) {
911 (void) vn_start_write(vp, &mp, V_WAIT);
912 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
913 error = VOP_FSYNC(vp, MNT_WAIT, td);
915 vn_finished_write(mp);
919 auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
920 auio.uio_resid = bp->bio_length;
921 auio.uio_segflg = UIO_SYSSPACE;
924 if (bp->bio_cmd == BIO_DELETE) {
926 * Emulate BIO_DELETE by writing zeros.
928 zerosize = ZERO_REGION_SIZE -
929 (ZERO_REGION_SIZE % sc->sectorsize);
930 auio.uio_iovcnt = howmany(bp->bio_length, zerosize);
931 piov = malloc(sizeof(*piov) * auio.uio_iovcnt, M_MD, M_WAITOK);
934 piov->iov_base = __DECONST(void *, zero_region);
937 piov->iov_len = zerosize;
938 len -= piov->iov_len;
942 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
943 piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK);
945 vlist = (bus_dma_segment_t *)bp->bio_data;
947 piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr +
949 piov->iov_len = vlist->ds_len - ma_offs;
950 if (piov->iov_len > len)
952 len -= piov->iov_len;
957 auio.uio_iovcnt = piov - auio.uio_iov;
959 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
960 pb = uma_zalloc(md_pbuf_zone, M_WAITOK);
961 MPASS((pb->b_flags & B_MAXPHYS) != 0);
964 npages = atop(min(maxphys, round_page(len + (ma_offs &
966 iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len);
967 KASSERT(iolen > 0, ("zero iolen"));
968 pmap_qenter((vm_offset_t)pb->b_data,
969 &bp->bio_ma[atop(ma_offs)], npages);
970 aiov.iov_base = (void *)((vm_offset_t)pb->b_data +
971 (ma_offs & PAGE_MASK));
972 aiov.iov_len = iolen;
973 auio.uio_iov = &aiov;
975 auio.uio_resid = iolen;
977 aiov.iov_base = bp->bio_data;
978 aiov.iov_len = bp->bio_length;
979 auio.uio_iov = &aiov;
982 iostart = auio.uio_offset;
983 if (auio.uio_rw == UIO_READ) {
984 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
985 error = VOP_READ(vp, &auio, 0, sc->cred);
988 (void) vn_start_write(vp, &mp, V_WAIT);
989 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
990 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
993 vn_finished_write(mp);
995 sc->flags &= ~MD_VERIFY;
998 /* When MD_CACHE is set, try to avoid double-caching the data. */
999 if (error == 0 && (sc->flags & MD_CACHE) == 0)
1000 VOP_ADVISE(vp, iostart, auio.uio_offset - 1,
1001 POSIX_FADV_DONTNEED);
1004 pmap_qremove((vm_offset_t)pb->b_data, npages);
1007 bp->bio_resid -= iolen;
1012 uma_zfree(md_pbuf_zone, pb);
1014 bp->bio_resid = auio.uio_resid;
1022 mdstart_swap(struct md_s *sc, struct bio *bp)
1026 vm_pindex_t i, lastp;
1027 bus_dma_segment_t *vlist;
1028 int rv, ma_offs, offs, len, lastend;
1030 switch (bp->bio_cmd) {
1036 return (EOPNOTSUPP);
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;
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).
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;
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))
1065 rv = vm_pager_get_pages(sc->object, &m, 1,
1067 if (rv == VM_PAGER_ERROR) {
1068 VM_OBJECT_WLOCK(sc->object);
1070 VM_OBJECT_WUNLOCK(sc->object);
1072 } else if (rv == VM_PAGER_FAIL) {
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.
1082 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1083 pmap_copy_pages(&m, offs, bp->bio_ma,
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);
1090 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1091 cpu_flush_dcache(p, len);
1093 } else if (bp->bio_cmd == BIO_WRITE) {
1094 if (len == PAGE_SIZE || vm_page_all_valid(m))
1097 rv = vm_pager_get_pages(sc->object, &m, 1,
1099 if (rv == VM_PAGER_ERROR) {
1100 VM_OBJECT_WLOCK(sc->object);
1102 VM_OBJECT_WUNLOCK(sc->object);
1104 } else if (rv == VM_PAGER_FAIL)
1107 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1108 pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1110 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1111 physcopyin_vlist(vlist, ma_offs,
1112 VM_PAGE_TO_PHYS(m) + offs, len);
1114 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1118 vm_page_set_dirty(m);
1119 } else if (bp->bio_cmd == BIO_DELETE) {
1120 if (len == PAGE_SIZE || vm_page_all_valid(m))
1123 rv = vm_pager_get_pages(sc->object, &m, 1,
1125 VM_OBJECT_WLOCK(sc->object);
1126 if (rv == VM_PAGER_ERROR) {
1128 VM_OBJECT_WUNLOCK(sc->object);
1130 } else if (rv == VM_PAGER_FAIL) {
1134 /* Page is valid. */
1135 if (len != PAGE_SIZE) {
1136 pmap_zero_page_area(m, offs, len);
1137 vm_page_set_dirty(m);
1139 vm_pager_page_unswapped(m);
1144 VM_OBJECT_WUNLOCK(sc->object);
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.
1152 if (vm_page_active(m))
1153 vm_page_reference(m);
1155 vm_page_activate(m);
1159 /* Actions on further pages start at offset 0 */
1160 p += PAGE_SIZE - offs;
1164 vm_object_pip_wakeup(sc->object);
1165 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
1169 mdstart_null(struct md_s *sc, struct bio *bp)
1172 switch (bp->bio_cmd) {
1174 bzero(bp->bio_data, bp->bio_length);
1175 cpu_flush_dcache(bp->bio_data, bp->bio_length);
1185 md_handleattr(struct md_s *sc, struct bio *bp)
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))
1191 if (g_handleattr_int(bp, "GEOM::candelete", 1) != 0)
1193 if (sc->ident[0] != '\0' &&
1194 g_handleattr_str(bp, "GEOM::ident", sc->ident) != 0)
1196 if (g_handleattr_int(bp, "MNT::verified", (sc->flags & MD_VERIFY) != 0))
1198 g_io_deliver(bp, EOPNOTSUPP);
1202 md_kthread(void *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;
1216 mtx_lock(&sc->queue_mtx);
1217 if (sc->flags & MD_SHUTDOWN) {
1218 sc->flags |= MD_EXITING;
1219 mtx_unlock(&sc->queue_mtx);
1222 bp = bioq_takefirst(&sc->bio_queue);
1224 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1227 mtx_unlock(&sc->queue_mtx);
1228 if (bp->bio_cmd == BIO_GETATTR) {
1229 md_handleattr(sc, bp);
1231 error = sc->start(sc, bp);
1232 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
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
1241 bp->bio_bcount = bp->bio_length;
1242 devstat_end_transaction_bio(sc->devstat, bp);
1244 bp->bio_completed = bp->bio_length - bp->bio_resid;
1245 g_io_deliver(bp, error);
1250 static struct md_s *
1255 LIST_FOREACH(sc, &md_softc_list, list) {
1256 if (sc->unit == unit)
1262 static struct md_s *
1263 mdnew(int unit, int *errp, enum md_types type)
1270 unit = alloc_unr(md_uh);
1272 unit = alloc_unr_specific(md_uh, unit);
1279 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
1281 bioq_init(&sc->bio_queue);
1282 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
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);
1289 LIST_REMOVE(sc, list);
1290 mtx_destroy(&sc->queue_mtx);
1291 free_unr(md_uh, sc->unit);
1298 mdinit(struct md_s *sc)
1301 struct g_provider *pp;
1304 gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1306 pp = g_new_providerf(gp, "md%d", sc->unit);
1307 devstat_remove_entry(pp->stat);
1309 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1310 pp->mediasize = sc->mediasize;
1311 pp->sectorsize = sc->sectorsize;
1316 pp->flags |= G_PF_ACCEPT_UNMAPPED;
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();
1332 mdcreate_malloc(struct md_s *sc, struct md_req *mdr)
1339 if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1341 if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize))
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,
1354 if (mdr->md_options & MD_RESERVE) {
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);
1362 error = s_write(sc->indir, u, sp);
1373 mdsetcred(struct md_s *sc, struct ucred *cred)
1379 * Set credits in our softc
1384 sc->cred = crhold(cred);
1387 * Horrible kludge to establish credentials for NFS XXX.
1394 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1395 bzero(&auio, sizeof(auio));
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);
1414 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1417 struct nameidata nd;
1421 fname = mdr->md_file;
1422 if (mdr->md_file_seg == UIO_USERSPACE) {
1423 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
1426 } else if (mdr->md_file_seg == UIO_SYSSPACE)
1427 strlcpy(sc->file, fname, sizeof(sc->file));
1432 * If the user specified that this is a read only device, don't
1433 * set the FWRITE mask before trying to open the backing store.
1435 flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \
1436 | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0);
1437 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td);
1438 error = vn_open(&nd, &flags, 0, NULL);
1441 NDFREE(&nd, NDF_ONLY_PNBUF);
1442 if (nd.ni_vp->v_type != VREG) {
1446 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1449 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1450 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1451 if (VN_IS_DOOMED(nd.ni_vp)) {
1452 /* Forced unmount. */
1457 nd.ni_vp->v_vflag |= VV_MD;
1458 VOP_UNLOCK(nd.ni_vp);
1460 if (mdr->md_fwsectors != 0)
1461 sc->fwsectors = mdr->md_fwsectors;
1462 if (mdr->md_fwheads != 0)
1463 sc->fwheads = mdr->md_fwheads;
1464 snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1465 (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1466 sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE |
1468 if (!(flags & FWRITE))
1469 sc->flags |= MD_READONLY;
1470 sc->vnode = nd.ni_vp;
1472 error = mdsetcred(sc, td->td_ucred);
1475 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1476 nd.ni_vp->v_vflag &= ~VV_MD;
1481 VOP_UNLOCK(nd.ni_vp);
1482 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1487 g_md_providergone(struct g_provider *pp)
1489 struct md_s *sc = pp->geom->softc;
1491 mtx_lock(&sc->queue_mtx);
1492 sc->flags |= MD_PROVIDERGONE;
1494 mtx_unlock(&sc->queue_mtx);
1498 mddestroy(struct md_s *sc, struct thread *td)
1503 g_wither_geom(sc->gp, ENXIO);
1504 g_topology_unlock();
1506 mtx_lock(&sc->queue_mtx);
1507 while (!(sc->flags & MD_PROVIDERGONE))
1508 msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0);
1509 mtx_unlock(&sc->queue_mtx);
1512 devstat_remove_entry(sc->devstat);
1515 mtx_lock(&sc->queue_mtx);
1516 sc->flags |= MD_SHUTDOWN;
1518 while (!(sc->flags & MD_EXITING))
1519 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1520 mtx_unlock(&sc->queue_mtx);
1521 mtx_destroy(&sc->queue_mtx);
1522 if (sc->vnode != NULL) {
1523 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1524 sc->vnode->v_vflag &= ~VV_MD;
1525 VOP_UNLOCK(sc->vnode);
1526 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1527 FREAD : (FREAD|FWRITE), sc->cred, td);
1529 if (sc->cred != NULL)
1531 if (sc->object != NULL)
1532 vm_object_deallocate(sc->object);
1534 destroy_indir(sc, sc->indir);
1536 uma_zdestroy(sc->uma);
1538 LIST_REMOVE(sc, list);
1539 free_unr(md_uh, sc->unit);
1545 mdresize(struct md_s *sc, struct md_req *mdr)
1548 vm_pindex_t oldpages, newpages;
1555 if (mdr->md_mediasize <= 0 ||
1556 (mdr->md_mediasize % PAGE_SIZE) != 0)
1558 oldpages = OFF_TO_IDX(sc->mediasize);
1559 newpages = OFF_TO_IDX(mdr->md_mediasize);
1560 if (newpages < oldpages) {
1561 VM_OBJECT_WLOCK(sc->object);
1562 vm_object_page_remove(sc->object, newpages, 0, 0);
1563 swap_release_by_cred(IDX_TO_OFF(oldpages -
1564 newpages), sc->cred);
1565 sc->object->charge = IDX_TO_OFF(newpages);
1566 sc->object->size = newpages;
1567 VM_OBJECT_WUNLOCK(sc->object);
1568 } else if (newpages > oldpages) {
1569 res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1570 oldpages), sc->cred);
1573 if ((mdr->md_options & MD_RESERVE) ||
1574 (sc->flags & MD_RESERVE)) {
1575 error = swap_pager_reserve(sc->object,
1576 oldpages, newpages - oldpages);
1578 swap_release_by_cred(
1579 IDX_TO_OFF(newpages - oldpages),
1584 VM_OBJECT_WLOCK(sc->object);
1585 sc->object->charge = IDX_TO_OFF(newpages);
1586 sc->object->size = newpages;
1587 VM_OBJECT_WUNLOCK(sc->object);
1591 return (EOPNOTSUPP);
1594 sc->mediasize = mdr->md_mediasize;
1596 g_resize_provider(sc->pp, sc->mediasize);
1597 g_topology_unlock();
1602 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1608 * Range check. Disallow negative sizes and sizes not being
1609 * multiple of page size.
1611 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1615 * Allocate an OBJT_SWAP object.
1617 * Note the truncation.
1620 if ((mdr->md_options & MD_VERIFY) != 0)
1622 npage = mdr->md_mediasize / PAGE_SIZE;
1623 if (mdr->md_fwsectors != 0)
1624 sc->fwsectors = mdr->md_fwsectors;
1625 if (mdr->md_fwheads != 0)
1626 sc->fwheads = mdr->md_fwheads;
1627 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1628 VM_PROT_DEFAULT, 0, td->td_ucred);
1629 if (sc->object == NULL)
1631 sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1632 if (mdr->md_options & MD_RESERVE) {
1633 if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1638 error = mdsetcred(sc, td->td_ucred);
1641 vm_object_deallocate(sc->object);
1648 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1652 * Range check. Disallow negative sizes and sizes not being
1653 * multiple of page size.
1655 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1662 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1668 sx_assert(&md_sx, SA_XLOCKED);
1670 switch (mdr->md_type) {
1680 if (mdr->md_sectorsize == 0)
1681 sectsize = DEV_BSIZE;
1683 sectsize = mdr->md_sectorsize;
1684 if (sectsize > maxphys || mdr->md_mediasize < sectsize)
1686 if (mdr->md_options & MD_AUTOUNIT)
1687 sc = mdnew(-1, &error, mdr->md_type);
1689 if (mdr->md_unit > INT_MAX)
1691 sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1695 if (mdr->md_label != NULL)
1696 error = copyinstr(mdr->md_label, sc->label,
1697 sizeof(sc->label), NULL);
1700 if (mdr->md_options & MD_AUTOUNIT)
1701 mdr->md_unit = sc->unit;
1702 sc->mediasize = mdr->md_mediasize;
1703 sc->sectorsize = sectsize;
1707 sc->start = mdstart_malloc;
1708 error = mdcreate_malloc(sc, mdr);
1712 * We disallow attaching preloaded memory disks via
1713 * ioctl. Preloaded memory disks are automatically
1714 * attached in g_md_init().
1719 sc->start = mdstart_vnode;
1720 error = mdcreate_vnode(sc, mdr, td);
1723 sc->start = mdstart_swap;
1724 error = mdcreate_swap(sc, mdr, td);
1727 sc->start = mdstart_null;
1728 error = mdcreate_null(sc, mdr, td);
1737 /* Prune off any residual fractional sector */
1738 i = sc->mediasize % sc->sectorsize;
1746 kern_mdattach(struct thread *td, struct md_req *mdr)
1751 error = kern_mdattach_locked(td, mdr);
1757 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1761 sx_assert(&md_sx, SA_XLOCKED);
1763 if (mdr->md_mediasize != 0 ||
1764 (mdr->md_options & ~MD_FORCE) != 0)
1767 sc = mdfind(mdr->md_unit);
1770 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1771 !(mdr->md_options & MD_FORCE))
1773 return (mddestroy(sc, td));
1777 kern_mddetach(struct thread *td, struct md_req *mdr)
1782 error = kern_mddetach_locked(td, mdr);
1788 kern_mdresize_locked(struct md_req *mdr)
1792 sx_assert(&md_sx, SA_XLOCKED);
1794 if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1797 sc = mdfind(mdr->md_unit);
1800 if (mdr->md_mediasize < sc->sectorsize)
1802 if (mdr->md_mediasize < sc->mediasize &&
1803 !(sc->flags & MD_FORCE) &&
1804 !(mdr->md_options & MD_FORCE))
1806 return (mdresize(sc, mdr));
1810 kern_mdresize(struct md_req *mdr)
1815 error = kern_mdresize_locked(mdr);
1821 kern_mdquery_locked(struct md_req *mdr)
1826 sx_assert(&md_sx, SA_XLOCKED);
1828 sc = mdfind(mdr->md_unit);
1831 mdr->md_type = sc->type;
1832 mdr->md_options = sc->flags;
1833 mdr->md_mediasize = sc->mediasize;
1834 mdr->md_sectorsize = sc->sectorsize;
1836 if (mdr->md_label != NULL) {
1837 error = copyout(sc->label, mdr->md_label,
1838 strlen(sc->label) + 1);
1842 if (sc->type == MD_VNODE ||
1843 (sc->type == MD_PRELOAD && mdr->md_file != NULL))
1844 error = copyout(sc->file, mdr->md_file,
1845 strlen(sc->file) + 1);
1850 kern_mdquery(struct md_req *mdr)
1855 error = kern_mdquery_locked(mdr);
1860 /* Copy members that are not userspace pointers. */
1861 #define MD_IOCTL2REQ(mdio, mdr) do { \
1862 (mdr)->md_unit = (mdio)->md_unit; \
1863 (mdr)->md_type = (mdio)->md_type; \
1864 (mdr)->md_mediasize = (mdio)->md_mediasize; \
1865 (mdr)->md_sectorsize = (mdio)->md_sectorsize; \
1866 (mdr)->md_options = (mdio)->md_options; \
1867 (mdr)->md_fwheads = (mdio)->md_fwheads; \
1868 (mdr)->md_fwsectors = (mdio)->md_fwsectors; \
1869 (mdr)->md_units = &(mdio)->md_pad[0]; \
1870 (mdr)->md_units_nitems = nitems((mdio)->md_pad); \
1873 /* Copy members that might have been updated */
1874 #define MD_REQ2IOCTL(mdr, mdio) do { \
1875 (mdio)->md_unit = (mdr)->md_unit; \
1876 (mdio)->md_type = (mdr)->md_type; \
1877 (mdio)->md_mediasize = (mdr)->md_mediasize; \
1878 (mdio)->md_sectorsize = (mdr)->md_sectorsize; \
1879 (mdio)->md_options = (mdr)->md_options; \
1880 (mdio)->md_fwheads = (mdr)->md_fwheads; \
1881 (mdio)->md_fwsectors = (mdr)->md_fwsectors; \
1885 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1892 printf("mdctlioctl(%s %lx %p %x %p)\n",
1893 devtoname(dev), cmd, addr, flags, td);
1895 bzero(&mdr, sizeof(mdr));
1901 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1902 if (mdio->md_version != MDIOVERSION)
1904 MD_IOCTL2REQ(mdio, &mdr);
1905 mdr.md_file = mdio->md_file;
1906 mdr.md_file_seg = UIO_USERSPACE;
1907 /* If the file is adjacent to the md_ioctl it's in kernel. */
1908 if ((void *)mdio->md_file == (void *)(mdio + 1))
1909 mdr.md_file_seg = UIO_SYSSPACE;
1910 mdr.md_label = mdio->md_label;
1913 #ifdef COMPAT_FREEBSD32
1914 case MDIOCATTACH_32:
1915 case MDIOCDETACH_32:
1916 case MDIOCRESIZE_32:
1917 case MDIOCQUERY_32: {
1918 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1919 if (mdio->md_version != MDIOVERSION)
1921 MD_IOCTL2REQ(mdio, &mdr);
1922 mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1923 mdr.md_file_seg = UIO_USERSPACE;
1924 mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1929 /* Fall through to handler switch. */
1936 #ifdef COMPAT_FREEBSD32
1937 case MDIOCATTACH_32:
1939 error = kern_mdattach(td, &mdr);
1942 #ifdef COMPAT_FREEBSD32
1943 case MDIOCDETACH_32:
1945 error = kern_mddetach(td, &mdr);
1948 #ifdef COMPAT_FREEBSD32
1949 case MDIOCRESIZE_32:
1951 error = kern_mdresize(&mdr);
1954 #ifdef COMPAT_FREEBSD32
1957 error = kern_mdquery(&mdr);
1966 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1967 MD_REQ2IOCTL(&mdr, mdio);
1970 #ifdef COMPAT_FREEBSD32
1971 case MDIOCATTACH_32:
1972 case MDIOCQUERY_32: {
1973 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1974 MD_REQ2IOCTL(&mdr, mdio);
1979 /* Other commands to not alter mdr. */
1987 md_preloaded(u_char *image, size_t length, const char *name)
1992 sc = mdnew(-1, &error, MD_PRELOAD);
1995 sc->mediasize = length;
1996 sc->sectorsize = DEV_BSIZE;
1998 sc->pl_len = length;
1999 sc->start = mdstart_preload;
2001 strlcpy(sc->file, name, sizeof(sc->file));
2003 if (sc->unit == 0) {
2005 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2007 #ifdef MD_ROOT_READONLY
2008 sc->flags |= MD_READONLY;
2014 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2015 MD_NAME, sc->unit, name, length, image);
2017 printf("%s%d: Embedded image %zd bytes at %p\n",
2018 MD_NAME, sc->unit, length, image);
2023 g_md_init(struct g_class *mp __unused)
2026 u_char *ptr, *name, *type;
2030 /* figure out log2(NINDIR) */
2031 for (i = NINDIR, nshift = -1; i; nshift++)
2035 sx_init(&md_sx, "MD config lock");
2036 g_topology_unlock();
2037 md_uh = new_unrhdr(0, INT_MAX, NULL);
2039 if (mfs_root_size != 0) {
2042 md_preloaded(mfs_root, mfs_root_size, NULL);
2044 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2050 /* XXX: are preload_* static or do they need Giant ? */
2051 while ((mod = preload_search_next_name(mod)) != NULL) {
2052 name = (char *)preload_search_info(mod, MODINFO_NAME);
2055 type = (char *)preload_search_info(mod, MODINFO_TYPE);
2058 if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2060 ptr = preload_fetch_addr(mod);
2061 len = preload_fetch_size(mod);
2062 if (ptr != NULL && len != 0) {
2064 md_preloaded(ptr, len, name);
2068 md_pbuf_zone = pbuf_zsecond_create("mdpbuf", nswbuf / 10);
2069 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2075 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2076 struct g_consumer *cp __unused, struct g_provider *pp)
2107 if (indent == NULL) {
2108 sbuf_printf(sb, " u %d", mp->unit);
2109 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2110 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2111 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2112 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2113 sbuf_printf(sb, " t %s", type);
2114 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2115 (mp->type == MD_PRELOAD && mp->file[0] != '\0'))
2116 sbuf_printf(sb, " file %s", mp->file);
2117 sbuf_printf(sb, " label %s", mp->label);
2119 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2121 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2122 indent, (uintmax_t) mp->sectorsize);
2123 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2124 indent, (uintmax_t) mp->fwheads);
2125 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2126 indent, (uintmax_t) mp->fwsectors);
2127 if (mp->ident[0] != '\0') {
2128 sbuf_printf(sb, "%s<ident>", indent);
2129 g_conf_printf_escaped(sb, "%s", mp->ident);
2130 sbuf_printf(sb, "</ident>\n");
2132 sbuf_printf(sb, "%s<length>%ju</length>\n",
2133 indent, (uintmax_t) mp->mediasize);
2134 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2135 (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2136 sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2137 (mp->flags & MD_READONLY) == 0 ? "read-write":
2139 sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2141 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2142 (mp->type == MD_PRELOAD && mp->file[0] != '\0')) {
2143 sbuf_printf(sb, "%s<file>", indent);
2144 g_conf_printf_escaped(sb, "%s", mp->file);
2145 sbuf_printf(sb, "</file>\n");
2147 if (mp->type == MD_VNODE)
2148 sbuf_printf(sb, "%s<cache>%s</cache>\n", indent,
2149 (mp->flags & MD_CACHE) == 0 ? "off": "on");
2150 sbuf_printf(sb, "%s<label>", indent);
2151 g_conf_printf_escaped(sb, "%s", mp->label);
2152 sbuf_printf(sb, "</label>\n");
2158 g_md_fini(struct g_class *mp __unused)
2162 if (status_dev != NULL)
2163 destroy_dev(status_dev);
2164 uma_zdestroy(md_pbuf_zone);
2165 delete_unrhdr(md_uh);