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 in 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. */
115 #define MD_NSECT (10000 * 2)
119 unsigned md_unit; /* unit number */
120 enum md_types md_type; /* type of disk */
121 off_t md_mediasize; /* size of disk in bytes */
122 unsigned md_sectorsize; /* sectorsize */
123 unsigned md_options; /* options */
124 int md_fwheads; /* firmware heads */
125 int md_fwsectors; /* firmware sectors */
126 char *md_file; /* pathname of file to mount */
127 enum uio_seg md_file_seg; /* location of md_file */
128 char *md_label; /* label of the device (userspace) */
129 int *md_units; /* pointer to units array (kernel) */
130 size_t md_units_nitems; /* items in md_units array */
133 #ifdef COMPAT_FREEBSD32
137 enum md_types md_type;
140 unsigned md_sectorsize;
147 } __attribute__((__packed__));
148 CTASSERT((sizeof(struct md_ioctl32)) == 436);
150 #define MDIOCATTACH_32 _IOC_NEWTYPE(MDIOCATTACH, struct md_ioctl32)
151 #define MDIOCDETACH_32 _IOC_NEWTYPE(MDIOCDETACH, struct md_ioctl32)
152 #define MDIOCQUERY_32 _IOC_NEWTYPE(MDIOCQUERY, struct md_ioctl32)
153 #define MDIOCLIST_32 _IOC_NEWTYPE(MDIOCLIST, 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);
184 extern volatile u_char __weak_symbol mfs_root;
185 extern volatile u_char __weak_symbol mfs_root_end;
187 __GLOBL(mfs_root_end);
188 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root))
192 static g_init_t g_md_init;
193 static g_fini_t g_md_fini;
194 static g_start_t g_md_start;
195 static g_access_t g_md_access;
196 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
197 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
199 static struct cdev *status_dev = NULL;
200 static struct sx md_sx;
201 static struct unrhdr *md_uh;
203 static d_ioctl_t mdctlioctl;
205 static struct cdevsw mdctl_cdevsw = {
206 .d_version = D_VERSION,
207 .d_ioctl = mdctlioctl,
211 struct g_class g_md_class = {
213 .version = G_VERSION,
217 .access = g_md_access,
218 .dumpconf = g_md_dumpconf,
221 DECLARE_GEOM_CLASS(g_md_class, g_md);
224 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
226 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t))
227 #define NMASK (NINDIR-1)
230 static int md_vnode_pbuf_freecnt;
241 LIST_ENTRY(md_s) list;
242 struct bio_queue_head bio_queue;
243 struct mtx queue_mtx;
257 struct g_provider *pp;
258 int (*start)(struct md_s *sc, struct bio *bp);
259 struct devstat *devstat;
261 /* MD_MALLOC related fields */
265 /* MD_PRELOAD related fields */
269 /* MD_VNODE related fields */
272 char label[PATH_MAX];
275 /* MD_SWAP related fields */
279 static struct indir *
280 new_indir(u_int shift)
284 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
288 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
289 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
290 if (ip->array == NULL) {
300 del_indir(struct indir *ip)
303 free(ip->array, M_MDSECT);
308 destroy_indir(struct md_s *sc, struct indir *ip)
312 for (i = 0; i < NINDIR; i++) {
316 destroy_indir(sc, (struct indir*)(ip->array[i]));
317 else if (ip->array[i] > 255)
318 uma_zfree(sc->uma, (void *)(ip->array[i]));
324 * This function does the math and allocates the top level "indir" structure
325 * for a device of "size" sectors.
328 static struct indir *
329 dimension(off_t size)
337 while (rcnt > NINDIR) {
343 * XXX: the top layer is probably not fully populated, so we allocate
344 * too much space for ip->array in here.
346 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
347 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
348 M_MDSECT, M_WAITOK | M_ZERO);
350 ip->shift = layer * nshift;
355 * Read a given sector
359 s_read(struct indir *ip, off_t offset)
366 printf("s_read(%jd)\n", (intmax_t)offset);
368 for (cip = ip; cip != NULL;) {
370 idx = (offset >> cip->shift) & NMASK;
371 up = cip->array[idx];
372 cip = (struct indir *)up;
375 idx = offset & NMASK;
376 return (cip->array[idx]);
382 * Write a given sector, prune the tree if the value is 0
386 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
388 struct indir *cip, *lip[10];
393 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
400 idx = (offset >> cip->shift) & NMASK;
401 up = cip->array[idx];
403 cip = (struct indir *)up;
406 /* Allocate branch */
408 (uintptr_t)new_indir(cip->shift - nshift);
409 if (cip->array[idx] == 0)
412 up = cip->array[idx];
413 cip = (struct indir *)up;
417 idx = offset & NMASK;
418 up = cip->array[idx];
421 cip->array[idx] = ptr;
426 if (cip->used != 0 || li == 1)
429 while (cip->used == 0 && cip != ip) {
431 idx = (offset >> lip[li]->shift) & NMASK;
432 up = lip[li]->array[idx];
433 KASSERT(up == (uintptr_t)cip, ("md screwed up"));
435 lip[li]->array[idx] = 0;
444 g_md_access(struct g_provider *pp, int r, int w, int e)
448 sc = pp->geom->softc;
450 if (r <= 0 && w <= 0 && e <= 0)
457 if ((sc->flags & MD_READONLY) != 0 && w > 0)
459 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
461 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
468 g_md_start(struct bio *bp)
472 sc = bp->bio_to->geom->softc;
473 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
474 mtx_lock(&sc->stat_mtx);
475 devstat_start_transaction_bio(sc->devstat, bp);
476 mtx_unlock(&sc->stat_mtx);
478 mtx_lock(&sc->queue_mtx);
479 bioq_disksort(&sc->bio_queue, bp);
480 mtx_unlock(&sc->queue_mtx);
484 #define MD_MALLOC_MOVE_ZERO 1
485 #define MD_MALLOC_MOVE_FILL 2
486 #define MD_MALLOC_MOVE_READ 3
487 #define MD_MALLOC_MOVE_WRITE 4
488 #define MD_MALLOC_MOVE_CMP 5
491 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize,
492 void *ptr, u_char fill, int op)
499 int error, i, ma_offs1, sz, first_read;
504 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */
512 for (n = sectorsize; n != 0; n -= sz) {
513 sz = imin(PAGE_SIZE - *ma_offs, n);
518 sf = sf_buf_alloc(m, SFB_CPUPRIVATE |
519 (md_malloc_wait ? 0 : SFB_NOWAIT));
525 p = (char *)sf_buf_kva(sf) + *ma_offs;
527 case MD_MALLOC_MOVE_ZERO:
530 case MD_MALLOC_MOVE_FILL:
533 case MD_MALLOC_MOVE_READ:
535 cpu_flush_dcache(p, sz);
537 case MD_MALLOC_MOVE_WRITE:
540 case MD_MALLOC_MOVE_CMP:
541 for (i = 0; i < sz; i++, p++) {
546 } else if (*p != first) {
553 KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op));
559 *ma_offs %= PAGE_SIZE;
562 ptr = (char *)ptr + sz;
568 if (op == MD_MALLOC_MOVE_CMP && error != 0) {
576 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs,
577 unsigned len, void *ptr, u_char fill, int op)
579 bus_dma_segment_t *vlist;
580 uint8_t *p, *end, first;
582 int ma_offs, seg_len;
588 for (; len != 0; len -= seg_len) {
589 seg_len = imin(vlist->ds_len - ma_offs, len);
590 p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs;
592 case MD_MALLOC_MOVE_ZERO:
595 case MD_MALLOC_MOVE_FILL:
596 memset(p, fill, seg_len);
598 case MD_MALLOC_MOVE_READ:
599 bcopy(ptr, p, seg_len);
600 cpu_flush_dcache(p, seg_len);
602 case MD_MALLOC_MOVE_WRITE:
603 bcopy(p, ptr, seg_len);
605 case MD_MALLOC_MOVE_CMP:
608 /* Confirm all following bytes match the first */
615 KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op));
620 if (ma_offs == vlist->ds_len) {
624 ptr = (uint8_t *)ptr + seg_len;
633 mdstart_malloc(struct md_s *sc, struct bio *bp)
637 bus_dma_segment_t *vlist;
638 int i, error, error1, ma_offs, notmapped;
639 off_t secno, nsec, uc;
642 switch (bp->bio_cmd) {
651 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0;
652 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
653 (bus_dma_segment_t *)bp->bio_data : NULL;
656 ma_offs = bp->bio_ma_offset;
658 KASSERT(vlist == NULL, ("vlists cannot be unmapped"));
659 } else if (vlist != NULL) {
660 ma_offs = bp->bio_ma_offset;
666 nsec = bp->bio_length / sc->sectorsize;
667 secno = bp->bio_offset / sc->sectorsize;
670 osp = s_read(sc->indir, secno);
671 if (bp->bio_cmd == BIO_DELETE) {
673 error = s_write(sc->indir, secno, 0);
674 } else if (bp->bio_cmd == BIO_READ) {
677 error = md_malloc_move_ma(&m, &ma_offs,
678 sc->sectorsize, NULL, 0,
679 MD_MALLOC_MOVE_ZERO);
680 } else if (vlist != NULL) {
681 error = md_malloc_move_vlist(&vlist,
682 &ma_offs, sc->sectorsize, NULL, 0,
683 MD_MALLOC_MOVE_ZERO);
685 bzero(dst, sc->sectorsize);
686 } else if (osp <= 255) {
688 error = md_malloc_move_ma(&m, &ma_offs,
689 sc->sectorsize, NULL, osp,
690 MD_MALLOC_MOVE_FILL);
691 } else if (vlist != NULL) {
692 error = md_malloc_move_vlist(&vlist,
693 &ma_offs, sc->sectorsize, NULL, osp,
694 MD_MALLOC_MOVE_FILL);
696 memset(dst, osp, sc->sectorsize);
699 error = md_malloc_move_ma(&m, &ma_offs,
700 sc->sectorsize, (void *)osp, 0,
701 MD_MALLOC_MOVE_READ);
702 } else if (vlist != NULL) {
703 error = md_malloc_move_vlist(&vlist,
704 &ma_offs, sc->sectorsize,
706 MD_MALLOC_MOVE_READ);
708 bcopy((void *)osp, dst, sc->sectorsize);
709 cpu_flush_dcache(dst, sc->sectorsize);
713 } else if (bp->bio_cmd == BIO_WRITE) {
714 if (sc->flags & MD_COMPRESS) {
716 error1 = md_malloc_move_ma(&m, &ma_offs,
717 sc->sectorsize, &uc, 0,
719 i = error1 == 0 ? sc->sectorsize : 0;
720 } else if (vlist != NULL) {
721 error1 = md_malloc_move_vlist(&vlist,
722 &ma_offs, sc->sectorsize, &uc, 0,
724 i = error1 == 0 ? sc->sectorsize : 0;
727 for (i = 1; i < sc->sectorsize; i++) {
736 if (i == sc->sectorsize) {
738 error = s_write(sc->indir, secno, uc);
741 sp = (uintptr_t)uma_zalloc(sc->uma,
742 md_malloc_wait ? M_WAITOK :
749 error = md_malloc_move_ma(&m,
750 &ma_offs, sc->sectorsize,
752 MD_MALLOC_MOVE_WRITE);
753 } else if (vlist != NULL) {
754 error = md_malloc_move_vlist(
756 sc->sectorsize, (void *)sp,
757 0, MD_MALLOC_MOVE_WRITE);
759 bcopy(dst, (void *)sp,
762 error = s_write(sc->indir, secno, sp);
765 error = md_malloc_move_ma(&m,
766 &ma_offs, sc->sectorsize,
768 MD_MALLOC_MOVE_WRITE);
769 } else if (vlist != NULL) {
770 error = md_malloc_move_vlist(
772 sc->sectorsize, (void *)osp,
773 0, MD_MALLOC_MOVE_WRITE);
775 bcopy(dst, (void *)osp,
785 uma_zfree(sc->uma, (void*)osp);
789 if (!notmapped && vlist == NULL)
790 dst += sc->sectorsize;
797 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len)
801 while (offset >= vlist->ds_len) {
802 offset -= vlist->ds_len;
807 seg_len = omin(len, vlist->ds_len - offset);
808 bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset),
811 src = (uint8_t *)src + seg_len;
818 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len)
822 while (offset >= vlist->ds_len) {
823 offset -= vlist->ds_len;
828 seg_len = omin(len, vlist->ds_len - offset);
829 bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst,
832 dst = (uint8_t *)dst + seg_len;
839 mdstart_preload(struct md_s *sc, struct bio *bp)
843 p = sc->pl_ptr + bp->bio_offset;
844 switch (bp->bio_cmd) {
846 if ((bp->bio_flags & BIO_VLIST) != 0) {
847 mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data,
848 bp->bio_ma_offset, bp->bio_length);
850 bcopy(p, bp->bio_data, bp->bio_length);
852 cpu_flush_dcache(bp->bio_data, bp->bio_length);
855 if ((bp->bio_flags & BIO_VLIST) != 0) {
856 mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data,
857 bp->bio_ma_offset, p, bp->bio_length);
859 bcopy(bp->bio_data, p, bp->bio_length);
868 mdstart_vnode(struct md_s *sc, struct bio *bp)
877 bus_dma_segment_t *vlist;
879 off_t iolen, len, zerosize;
882 switch (bp->bio_cmd) {
884 auio.uio_rw = UIO_READ;
888 auio.uio_rw = UIO_WRITE;
900 ma_offs = bp->bio_ma_offset;
901 len = bp->bio_length;
906 * If an error occurs, we set BIO_ERROR but we do not set
907 * B_INVAL because (for a write anyway), the buffer is
911 if (bp->bio_cmd == BIO_FLUSH) {
912 (void) vn_start_write(vp, &mp, V_WAIT);
913 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
914 error = VOP_FSYNC(vp, MNT_WAIT, td);
916 vn_finished_write(mp);
920 auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
921 auio.uio_resid = bp->bio_length;
922 auio.uio_segflg = UIO_SYSSPACE;
925 if (bp->bio_cmd == BIO_DELETE) {
927 * Emulate BIO_DELETE by writing zeros.
929 zerosize = ZERO_REGION_SIZE -
930 (ZERO_REGION_SIZE % sc->sectorsize);
931 auio.uio_iovcnt = howmany(bp->bio_length, zerosize);
932 piov = malloc(sizeof(*piov) * auio.uio_iovcnt, M_MD, M_WAITOK);
935 piov->iov_base = __DECONST(void *, zero_region);
938 piov->iov_len = zerosize;
939 len -= piov->iov_len;
943 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
944 piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK);
946 vlist = (bus_dma_segment_t *)bp->bio_data;
948 piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr +
950 piov->iov_len = vlist->ds_len - ma_offs;
951 if (piov->iov_len > len)
953 len -= piov->iov_len;
958 auio.uio_iovcnt = piov - auio.uio_iov;
960 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
961 pb = getpbuf(&md_vnode_pbuf_freecnt);
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;
983 * When reading set IO_DIRECT to try to avoid double-caching
984 * the data. When writing IO_DIRECT is not optimal.
986 if (auio.uio_rw == UIO_READ) {
987 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
988 error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred);
991 (void) vn_start_write(vp, &mp, V_WAIT);
992 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
993 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
996 vn_finished_write(mp);
998 sc->flags &= ~MD_VERIFY;
1002 pmap_qremove((vm_offset_t)pb->b_data, npages);
1005 bp->bio_resid -= iolen;
1010 relpbuf(pb, &md_vnode_pbuf_freecnt);
1015 bp->bio_resid = auio.uio_resid;
1020 md_swap_page_free(vm_page_t m)
1030 mdstart_swap(struct md_s *sc, struct bio *bp)
1034 vm_pindex_t i, lastp;
1035 bus_dma_segment_t *vlist;
1036 int rv, ma_offs, offs, len, lastend;
1038 switch (bp->bio_cmd) {
1044 return (EOPNOTSUPP);
1048 ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ?
1049 bp->bio_ma_offset : 0;
1050 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
1051 (bus_dma_segment_t *)bp->bio_data : NULL;
1054 * offs is the offset at which to start operating on the
1055 * next (ie, first) page. lastp is the last page on
1056 * which we're going to operate. lastend is the ending
1057 * position within that last page (ie, PAGE_SIZE if
1058 * we're operating on complete aligned pages).
1060 offs = bp->bio_offset % PAGE_SIZE;
1061 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
1062 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
1065 VM_OBJECT_WLOCK(sc->object);
1066 vm_object_pip_add(sc->object, 1);
1067 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
1068 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
1069 m = vm_page_grab(sc->object, i, VM_ALLOC_SYSTEM);
1070 if (bp->bio_cmd == BIO_READ) {
1071 if (m->valid == VM_PAGE_BITS_ALL)
1074 rv = vm_pager_get_pages(sc->object, &m, 1,
1076 if (rv == VM_PAGER_ERROR) {
1077 md_swap_page_free(m);
1079 } else if (rv == VM_PAGER_FAIL) {
1081 * Pager does not have the page. Zero
1082 * the allocated page, and mark it as
1083 * valid. Do not set dirty, the page
1084 * can be recreated if thrown out.
1087 m->valid = VM_PAGE_BITS_ALL;
1089 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1090 pmap_copy_pages(&m, offs, bp->bio_ma,
1092 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1093 physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs,
1094 vlist, ma_offs, len);
1095 cpu_flush_dcache(p, len);
1097 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1098 cpu_flush_dcache(p, len);
1100 } else if (bp->bio_cmd == BIO_WRITE) {
1101 if (len == PAGE_SIZE || m->valid == VM_PAGE_BITS_ALL)
1104 rv = vm_pager_get_pages(sc->object, &m, 1,
1106 if (rv == VM_PAGER_ERROR) {
1107 md_swap_page_free(m);
1109 } else if (rv == VM_PAGER_FAIL)
1112 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1113 pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1115 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1116 physcopyin_vlist(vlist, ma_offs,
1117 VM_PAGE_TO_PHYS(m) + offs, len);
1119 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1122 m->valid = VM_PAGE_BITS_ALL;
1123 if (m->dirty != VM_PAGE_BITS_ALL) {
1125 vm_pager_page_unswapped(m);
1127 } else if (bp->bio_cmd == BIO_DELETE) {
1128 if (len == PAGE_SIZE || m->valid == VM_PAGE_BITS_ALL)
1131 rv = vm_pager_get_pages(sc->object, &m, 1,
1133 if (rv == VM_PAGER_ERROR) {
1134 md_swap_page_free(m);
1136 } else if (rv == VM_PAGER_FAIL) {
1137 md_swap_page_free(m);
1140 /* Page is valid. */
1141 if (len != PAGE_SIZE) {
1142 pmap_zero_page_area(m, offs, len);
1143 if (m->dirty != VM_PAGE_BITS_ALL) {
1145 vm_pager_page_unswapped(m);
1148 vm_pager_page_unswapped(m);
1149 md_swap_page_free(m);
1157 if (vm_page_active(m))
1158 vm_page_reference(m);
1160 vm_page_activate(m);
1164 /* Actions on further pages start at offset 0 */
1165 p += PAGE_SIZE - offs;
1169 vm_object_pip_wakeup(sc->object);
1170 VM_OBJECT_WUNLOCK(sc->object);
1171 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
1175 mdstart_null(struct md_s *sc, struct bio *bp)
1178 switch (bp->bio_cmd) {
1180 bzero(bp->bio_data, bp->bio_length);
1181 cpu_flush_dcache(bp->bio_data, bp->bio_length);
1191 md_kthread(void *arg)
1198 thread_lock(curthread);
1199 sched_prio(curthread, PRIBIO);
1200 thread_unlock(curthread);
1201 if (sc->type == MD_VNODE)
1202 curthread->td_pflags |= TDP_NORUNNINGBUF;
1205 mtx_lock(&sc->queue_mtx);
1206 if (sc->flags & MD_SHUTDOWN) {
1207 sc->flags |= MD_EXITING;
1208 mtx_unlock(&sc->queue_mtx);
1211 bp = bioq_takefirst(&sc->bio_queue);
1213 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1216 mtx_unlock(&sc->queue_mtx);
1217 if (bp->bio_cmd == BIO_GETATTR) {
1218 int isv = ((sc->flags & MD_VERIFY) != 0);
1220 if ((sc->fwsectors && sc->fwheads &&
1221 (g_handleattr_int(bp, "GEOM::fwsectors",
1223 g_handleattr_int(bp, "GEOM::fwheads",
1225 g_handleattr_int(bp, "GEOM::candelete", 1))
1227 else if (sc->ident[0] != '\0' &&
1228 g_handleattr_str(bp, "GEOM::ident", sc->ident))
1230 else if (g_handleattr_int(bp, "MNT::verified", isv))
1235 error = sc->start(sc, bp);
1239 bp->bio_completed = bp->bio_length;
1240 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE))
1241 devstat_end_transaction_bio(sc->devstat, bp);
1242 g_io_deliver(bp, error);
1247 static struct md_s *
1252 LIST_FOREACH(sc, &md_softc_list, list) {
1253 if (sc->unit == unit)
1259 static struct md_s *
1260 mdnew(int unit, int *errp, enum md_types type)
1267 unit = alloc_unr(md_uh);
1269 unit = alloc_unr_specific(md_uh, unit);
1276 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
1278 bioq_init(&sc->bio_queue);
1279 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1280 mtx_init(&sc->stat_mtx, "md stat", NULL, MTX_DEF);
1282 sprintf(sc->name, "md%d", unit);
1283 LIST_INSERT_HEAD(&md_softc_list, sc, list);
1284 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1287 LIST_REMOVE(sc, list);
1288 mtx_destroy(&sc->stat_mtx);
1289 mtx_destroy(&sc->queue_mtx);
1290 free_unr(md_uh, sc->unit);
1297 mdinit(struct md_s *sc)
1300 struct g_provider *pp;
1303 gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1305 pp = g_new_providerf(gp, "md%d", sc->unit);
1306 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1307 pp->mediasize = sc->mediasize;
1308 pp->sectorsize = sc->sectorsize;
1313 pp->flags |= G_PF_ACCEPT_UNMAPPED;
1321 g_error_provider(pp, 0);
1322 g_topology_unlock();
1323 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1324 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
1328 mdcreate_malloc(struct md_s *sc, struct md_req *mdr)
1335 if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1337 if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize))
1339 /* Compression doesn't make sense if we have reserved space */
1340 if (mdr->md_options & MD_RESERVE)
1341 mdr->md_options &= ~MD_COMPRESS;
1342 if (mdr->md_fwsectors != 0)
1343 sc->fwsectors = mdr->md_fwsectors;
1344 if (mdr->md_fwheads != 0)
1345 sc->fwheads = mdr->md_fwheads;
1346 sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE);
1347 sc->indir = dimension(sc->mediasize / sc->sectorsize);
1348 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
1350 if (mdr->md_options & MD_RESERVE) {
1353 nsectors = sc->mediasize / sc->sectorsize;
1354 for (u = 0; u < nsectors; u++) {
1355 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ?
1356 M_WAITOK : M_NOWAIT) | M_ZERO);
1358 error = s_write(sc->indir, u, sp);
1370 mdsetcred(struct md_s *sc, struct ucred *cred)
1376 * Set credits in our softc
1381 sc->cred = crhold(cred);
1384 * Horrible kludge to establish credentials for NFS XXX.
1391 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1392 bzero(&auio, sizeof(auio));
1394 aiov.iov_base = tmpbuf;
1395 aiov.iov_len = sc->sectorsize;
1396 auio.uio_iov = &aiov;
1397 auio.uio_iovcnt = 1;
1398 auio.uio_offset = 0;
1399 auio.uio_rw = UIO_READ;
1400 auio.uio_segflg = UIO_SYSSPACE;
1401 auio.uio_resid = aiov.iov_len;
1402 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1403 error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
1404 VOP_UNLOCK(sc->vnode, 0);
1405 free(tmpbuf, M_TEMP);
1411 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1414 struct nameidata nd;
1418 fname = mdr->md_file;
1419 if (mdr->md_file_seg == UIO_USERSPACE) {
1420 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
1423 } else if (mdr->md_file_seg == UIO_SYSSPACE)
1424 strlcpy(sc->file, fname, sizeof(sc->file));
1429 * If the user specified that this is a read only device, don't
1430 * set the FWRITE mask before trying to open the backing store.
1432 flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \
1433 | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0);
1434 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td);
1435 error = vn_open(&nd, &flags, 0, NULL);
1438 NDFREE(&nd, NDF_ONLY_PNBUF);
1439 if (nd.ni_vp->v_type != VREG) {
1443 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1446 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1447 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1448 if (nd.ni_vp->v_iflag & VI_DOOMED) {
1449 /* Forced unmount. */
1454 nd.ni_vp->v_vflag |= VV_MD;
1455 VOP_UNLOCK(nd.ni_vp, 0);
1457 if (mdr->md_fwsectors != 0)
1458 sc->fwsectors = mdr->md_fwsectors;
1459 if (mdr->md_fwheads != 0)
1460 sc->fwheads = mdr->md_fwheads;
1461 snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1462 (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1463 sc->flags = mdr->md_options & (MD_FORCE | MD_ASYNC | MD_VERIFY);
1464 if (!(flags & FWRITE))
1465 sc->flags |= MD_READONLY;
1466 sc->vnode = nd.ni_vp;
1468 error = mdsetcred(sc, td->td_ucred);
1471 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1472 nd.ni_vp->v_vflag &= ~VV_MD;
1477 VOP_UNLOCK(nd.ni_vp, 0);
1478 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1483 mddestroy(struct md_s *sc, struct thread *td)
1487 sc->gp->softc = NULL;
1489 g_wither_geom(sc->gp, ENXIO);
1490 g_topology_unlock();
1495 devstat_remove_entry(sc->devstat);
1498 mtx_lock(&sc->queue_mtx);
1499 sc->flags |= MD_SHUTDOWN;
1501 while (!(sc->flags & MD_EXITING))
1502 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1503 mtx_unlock(&sc->queue_mtx);
1504 mtx_destroy(&sc->stat_mtx);
1505 mtx_destroy(&sc->queue_mtx);
1506 if (sc->vnode != NULL) {
1507 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1508 sc->vnode->v_vflag &= ~VV_MD;
1509 VOP_UNLOCK(sc->vnode, 0);
1510 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1511 FREAD : (FREAD|FWRITE), sc->cred, td);
1513 if (sc->cred != NULL)
1515 if (sc->object != NULL)
1516 vm_object_deallocate(sc->object);
1518 destroy_indir(sc, sc->indir);
1520 uma_zdestroy(sc->uma);
1522 LIST_REMOVE(sc, list);
1523 free_unr(md_uh, sc->unit);
1529 mdresize(struct md_s *sc, struct md_req *mdr)
1532 vm_pindex_t oldpages, newpages;
1539 if (mdr->md_mediasize <= 0 ||
1540 (mdr->md_mediasize % PAGE_SIZE) != 0)
1542 oldpages = OFF_TO_IDX(round_page(sc->mediasize));
1543 newpages = OFF_TO_IDX(round_page(mdr->md_mediasize));
1544 if (newpages < oldpages) {
1545 VM_OBJECT_WLOCK(sc->object);
1546 vm_object_page_remove(sc->object, newpages, 0, 0);
1547 swap_pager_freespace(sc->object, newpages,
1548 oldpages - newpages);
1549 swap_release_by_cred(IDX_TO_OFF(oldpages -
1550 newpages), sc->cred);
1551 sc->object->charge = IDX_TO_OFF(newpages);
1552 sc->object->size = newpages;
1553 VM_OBJECT_WUNLOCK(sc->object);
1554 } else if (newpages > oldpages) {
1555 res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1556 oldpages), sc->cred);
1559 if ((mdr->md_options & MD_RESERVE) ||
1560 (sc->flags & MD_RESERVE)) {
1561 error = swap_pager_reserve(sc->object,
1562 oldpages, newpages - oldpages);
1564 swap_release_by_cred(
1565 IDX_TO_OFF(newpages - oldpages),
1570 VM_OBJECT_WLOCK(sc->object);
1571 sc->object->charge = IDX_TO_OFF(newpages);
1572 sc->object->size = newpages;
1573 VM_OBJECT_WUNLOCK(sc->object);
1577 return (EOPNOTSUPP);
1580 sc->mediasize = mdr->md_mediasize;
1582 g_resize_provider(sc->pp, sc->mediasize);
1583 g_topology_unlock();
1588 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1594 * Range check. Disallow negative sizes and sizes not being
1595 * multiple of page size.
1597 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1601 * Allocate an OBJT_SWAP object.
1603 * Note the truncation.
1606 if ((mdr->md_options & MD_VERIFY) != 0)
1608 npage = mdr->md_mediasize / PAGE_SIZE;
1609 if (mdr->md_fwsectors != 0)
1610 sc->fwsectors = mdr->md_fwsectors;
1611 if (mdr->md_fwheads != 0)
1612 sc->fwheads = mdr->md_fwheads;
1613 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1614 VM_PROT_DEFAULT, 0, td->td_ucred);
1615 if (sc->object == NULL)
1617 sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1618 if (mdr->md_options & MD_RESERVE) {
1619 if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1624 error = mdsetcred(sc, td->td_ucred);
1627 vm_object_deallocate(sc->object);
1634 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1638 * Range check. Disallow negative sizes and sizes not being
1639 * multiple of page size.
1641 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1648 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1654 sx_assert(&md_sx, SA_XLOCKED);
1656 switch (mdr->md_type) {
1666 if (mdr->md_sectorsize == 0)
1667 sectsize = DEV_BSIZE;
1669 sectsize = mdr->md_sectorsize;
1670 if (sectsize > MAXPHYS || mdr->md_mediasize < sectsize)
1672 if (mdr->md_options & MD_AUTOUNIT)
1673 sc = mdnew(-1, &error, mdr->md_type);
1675 if (mdr->md_unit > INT_MAX)
1677 sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1681 if (mdr->md_label != NULL)
1682 error = copyinstr(mdr->md_label, sc->label,
1683 sizeof(sc->label), NULL);
1686 if (mdr->md_options & MD_AUTOUNIT)
1687 mdr->md_unit = sc->unit;
1688 sc->mediasize = mdr->md_mediasize;
1689 sc->sectorsize = sectsize;
1693 sc->start = mdstart_malloc;
1694 error = mdcreate_malloc(sc, mdr);
1698 * We disallow attaching preloaded memory disks via
1699 * ioctl. Preloaded memory disks are automatically
1700 * attached in g_md_init().
1705 sc->start = mdstart_vnode;
1706 error = mdcreate_vnode(sc, mdr, td);
1709 sc->start = mdstart_swap;
1710 error = mdcreate_swap(sc, mdr, td);
1713 sc->start = mdstart_null;
1714 error = mdcreate_null(sc, mdr, td);
1723 /* Prune off any residual fractional sector */
1724 i = sc->mediasize % sc->sectorsize;
1732 kern_mdattach(struct thread *td, struct md_req *mdr)
1737 error = kern_mdattach_locked(td, mdr);
1743 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1747 sx_assert(&md_sx, SA_XLOCKED);
1749 if (mdr->md_mediasize != 0 ||
1750 (mdr->md_options & ~MD_FORCE) != 0)
1753 sc = mdfind(mdr->md_unit);
1756 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1757 !(mdr->md_options & MD_FORCE))
1759 return (mddestroy(sc, td));
1763 kern_mddetach(struct thread *td, struct md_req *mdr)
1768 error = kern_mddetach_locked(td, mdr);
1774 kern_mdresize_locked(struct md_req *mdr)
1778 sx_assert(&md_sx, SA_XLOCKED);
1780 if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1783 sc = mdfind(mdr->md_unit);
1786 if (mdr->md_mediasize < sc->sectorsize)
1788 if (mdr->md_mediasize < sc->mediasize &&
1789 !(sc->flags & MD_FORCE) &&
1790 !(mdr->md_options & MD_FORCE))
1792 return (mdresize(sc, mdr));
1796 kern_mdresize(struct md_req *mdr)
1801 error = kern_mdresize_locked(mdr);
1807 kern_mdquery_locked(struct md_req *mdr)
1812 sx_assert(&md_sx, SA_XLOCKED);
1814 sc = mdfind(mdr->md_unit);
1817 mdr->md_type = sc->type;
1818 mdr->md_options = sc->flags;
1819 mdr->md_mediasize = sc->mediasize;
1820 mdr->md_sectorsize = sc->sectorsize;
1822 if (mdr->md_label != NULL) {
1823 error = copyout(sc->label, mdr->md_label,
1824 strlen(sc->label) + 1);
1828 if (sc->type == MD_VNODE ||
1829 (sc->type == MD_PRELOAD && mdr->md_file != NULL))
1830 error = copyout(sc->file, mdr->md_file,
1831 strlen(sc->file) + 1);
1836 kern_mdquery(struct md_req *mdr)
1841 error = kern_mdquery_locked(mdr);
1847 kern_mdlist_locked(struct md_req *mdr)
1852 sx_assert(&md_sx, SA_XLOCKED);
1855 * Write the number of md devices to mdr->md_units[0].
1856 * Write the unit number of the first (mdr->md_units_nitems - 2)
1857 * units to mdr->md_units[1::(mdr->md_units - 2)] and terminate the
1860 * XXX: There is currently no mechanism to retrieve unit
1861 * numbers for more than (MDNPAD - 2) units.
1863 * XXX: Due to the use of LIST_INSERT_HEAD in mdnew(), the
1864 * list of visible unit numbers not stable.
1867 LIST_FOREACH(sc, &md_softc_list, list) {
1868 if (i < mdr->md_units_nitems - 1)
1869 mdr->md_units[i] = sc->unit;
1872 mdr->md_units[MIN(i, mdr->md_units_nitems - 1)] = -1;
1873 mdr->md_units[0] = i - 1;
1878 kern_mdlist(struct md_req *mdr)
1883 error = kern_mdlist_locked(mdr);
1888 /* Copy members that are not userspace pointers. */
1889 #define MD_IOCTL2REQ(mdio, mdr) do { \
1890 (mdr)->md_unit = (mdio)->md_unit; \
1891 (mdr)->md_type = (mdio)->md_type; \
1892 (mdr)->md_mediasize = (mdio)->md_mediasize; \
1893 (mdr)->md_sectorsize = (mdio)->md_sectorsize; \
1894 (mdr)->md_options = (mdio)->md_options; \
1895 (mdr)->md_fwheads = (mdio)->md_fwheads; \
1896 (mdr)->md_fwsectors = (mdio)->md_fwsectors; \
1897 (mdr)->md_units = &(mdio)->md_pad[0]; \
1898 (mdr)->md_units_nitems = nitems((mdio)->md_pad); \
1901 /* Copy members that might have been updated */
1902 #define MD_REQ2IOCTL(mdr, mdio) do { \
1903 (mdio)->md_unit = (mdr)->md_unit; \
1904 (mdio)->md_type = (mdr)->md_type; \
1905 (mdio)->md_mediasize = (mdr)->md_mediasize; \
1906 (mdio)->md_sectorsize = (mdr)->md_sectorsize; \
1907 (mdio)->md_options = (mdr)->md_options; \
1908 (mdio)->md_fwheads = (mdr)->md_fwheads; \
1909 (mdio)->md_fwsectors = (mdr)->md_fwsectors; \
1913 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1920 printf("mdctlioctl(%s %lx %p %x %p)\n",
1921 devtoname(dev), cmd, addr, flags, td);
1923 bzero(&mdr, sizeof(mdr));
1930 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1931 if (mdio->md_version != MDIOVERSION)
1933 MD_IOCTL2REQ(mdio, &mdr);
1934 mdr.md_file = mdio->md_file;
1935 mdr.md_file_seg = UIO_USERSPACE;
1936 /* If the file is adjacent to the md_ioctl it's in kernel. */
1937 if ((void *)mdio->md_file == (void *)(mdio + 1))
1938 mdr.md_file_seg = UIO_SYSSPACE;
1939 mdr.md_label = mdio->md_label;
1942 #ifdef COMPAT_FREEBSD32
1943 case MDIOCATTACH_32:
1944 case MDIOCDETACH_32:
1945 case MDIOCRESIZE_32:
1947 case MDIOCLIST_32: {
1948 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1949 if (mdio->md_version != MDIOVERSION)
1951 MD_IOCTL2REQ(mdio, &mdr);
1952 mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1953 mdr.md_file_seg = UIO_USERSPACE;
1954 mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1959 /* Fall through to handler switch. */
1966 #ifdef COMPAT_FREEBSD32
1967 case MDIOCATTACH_32:
1969 error = kern_mdattach(td, &mdr);
1972 #ifdef COMPAT_FREEBSD32
1973 case MDIOCDETACH_32:
1975 error = kern_mddetach(td, &mdr);
1978 #ifdef COMPAT_FREEBSD32
1979 case MDIOCRESIZE_32:
1981 error = kern_mdresize(&mdr);
1984 #ifdef COMPAT_FREEBSD32
1987 error = kern_mdquery(&mdr);
1990 #ifdef COMPAT_FREEBSD32
1993 error = kern_mdlist(&mdr);
2002 struct md_ioctl *mdio = (struct md_ioctl *)addr;
2003 MD_REQ2IOCTL(&mdr, mdio);
2006 #ifdef COMPAT_FREEBSD32
2007 case MDIOCATTACH_32:
2008 case MDIOCQUERY_32: {
2009 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
2010 MD_REQ2IOCTL(&mdr, mdio);
2015 /* Other commands to not alter mdr. */
2023 md_preloaded(u_char *image, size_t length, const char *name)
2028 sc = mdnew(-1, &error, MD_PRELOAD);
2031 sc->mediasize = length;
2032 sc->sectorsize = DEV_BSIZE;
2034 sc->pl_len = length;
2035 sc->start = mdstart_preload;
2037 strlcpy(sc->file, name, sizeof(sc->file));
2039 if (sc->unit == 0) {
2041 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2043 #ifdef MD_ROOT_READONLY
2044 sc->flags |= MD_READONLY;
2050 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2051 MD_NAME, sc->unit, name, length, image);
2053 printf("%s%d: Embedded image %zd bytes at %p\n",
2054 MD_NAME, sc->unit, length, image);
2059 g_md_init(struct g_class *mp __unused)
2062 u_char *ptr, *name, *type;
2066 /* figure out log2(NINDIR) */
2067 for (i = NINDIR, nshift = -1; i; nshift++)
2071 sx_init(&md_sx, "MD config lock");
2072 g_topology_unlock();
2073 md_uh = new_unrhdr(0, INT_MAX, NULL);
2075 if (mfs_root_size != 0) {
2077 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2082 /* XXX: are preload_* static or do they need Giant ? */
2083 while ((mod = preload_search_next_name(mod)) != NULL) {
2084 name = (char *)preload_search_info(mod, MODINFO_NAME);
2087 type = (char *)preload_search_info(mod, MODINFO_TYPE);
2090 if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2092 ptr = preload_fetch_addr(mod);
2093 len = preload_fetch_size(mod);
2094 if (ptr != NULL && len != 0) {
2096 md_preloaded(ptr, len, name);
2100 md_vnode_pbuf_freecnt = nswbuf / 10;
2101 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2107 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2108 struct g_consumer *cp __unused, struct g_provider *pp)
2139 if (indent == NULL) {
2140 sbuf_printf(sb, " u %d", mp->unit);
2141 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2142 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2143 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2144 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2145 sbuf_printf(sb, " t %s", type);
2146 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2147 (mp->type == MD_PRELOAD && mp->file[0] != '\0'))
2148 sbuf_printf(sb, " file %s", mp->file);
2149 sbuf_printf(sb, " label %s", mp->label);
2151 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2153 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2154 indent, (uintmax_t) mp->sectorsize);
2155 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2156 indent, (uintmax_t) mp->fwheads);
2157 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2158 indent, (uintmax_t) mp->fwsectors);
2159 if (mp->ident[0] != '\0') {
2160 sbuf_printf(sb, "%s<ident>", indent);
2161 g_conf_printf_escaped(sb, "%s", mp->ident);
2162 sbuf_printf(sb, "</ident>\n");
2164 sbuf_printf(sb, "%s<length>%ju</length>\n",
2165 indent, (uintmax_t) mp->mediasize);
2166 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2167 (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2168 sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2169 (mp->flags & MD_READONLY) == 0 ? "read-write":
2171 sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2173 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2174 (mp->type == MD_PRELOAD && mp->file[0] != '\0')) {
2175 sbuf_printf(sb, "%s<file>", indent);
2176 g_conf_printf_escaped(sb, "%s", mp->file);
2177 sbuf_printf(sb, "</file>\n");
2179 sbuf_printf(sb, "%s<label>", indent);
2180 g_conf_printf_escaped(sb, "%s", mp->label);
2181 sbuf_printf(sb, "</label>\n");
2187 g_md_fini(struct g_class *mp __unused)
2191 if (status_dev != NULL)
2192 destroy_dev(status_dev);
2193 delete_unrhdr(md_uh);