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);
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;
191 __GLOBL(mfs_root_end);
192 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root))
196 static g_init_t g_md_init;
197 static g_fini_t g_md_fini;
198 static g_start_t g_md_start;
199 static g_access_t g_md_access;
200 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
201 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
203 static struct cdev *status_dev = NULL;
204 static struct sx md_sx;
205 static struct unrhdr *md_uh;
207 static d_ioctl_t mdctlioctl;
209 static struct cdevsw mdctl_cdevsw = {
210 .d_version = D_VERSION,
211 .d_ioctl = mdctlioctl,
215 struct g_class g_md_class = {
217 .version = G_VERSION,
221 .access = g_md_access,
222 .dumpconf = g_md_dumpconf,
225 DECLARE_GEOM_CLASS(g_md_class, g_md);
228 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
230 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t))
231 #define NMASK (NINDIR-1)
234 static int md_vnode_pbuf_freecnt;
245 LIST_ENTRY(md_s) list;
246 struct bio_queue_head bio_queue;
247 struct mtx queue_mtx;
261 struct g_provider *pp;
262 int (*start)(struct md_s *sc, struct bio *bp);
263 struct devstat *devstat;
265 /* MD_MALLOC related fields */
269 /* MD_PRELOAD related fields */
273 /* MD_VNODE related fields */
276 char label[PATH_MAX];
279 /* MD_SWAP related fields */
283 static struct indir *
284 new_indir(u_int shift)
288 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
292 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
293 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
294 if (ip->array == NULL) {
304 del_indir(struct indir *ip)
307 free(ip->array, M_MDSECT);
312 destroy_indir(struct md_s *sc, struct indir *ip)
316 for (i = 0; i < NINDIR; i++) {
320 destroy_indir(sc, (struct indir*)(ip->array[i]));
321 else if (ip->array[i] > 255)
322 uma_zfree(sc->uma, (void *)(ip->array[i]));
328 * This function does the math and allocates the top level "indir" structure
329 * for a device of "size" sectors.
332 static struct indir *
333 dimension(off_t size)
341 while (rcnt > NINDIR) {
347 * XXX: the top layer is probably not fully populated, so we allocate
348 * too much space for ip->array in here.
350 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
351 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
352 M_MDSECT, M_WAITOK | M_ZERO);
354 ip->shift = layer * nshift;
359 * Read a given sector
363 s_read(struct indir *ip, off_t offset)
370 printf("s_read(%jd)\n", (intmax_t)offset);
372 for (cip = ip; cip != NULL;) {
374 idx = (offset >> cip->shift) & NMASK;
375 up = cip->array[idx];
376 cip = (struct indir *)up;
379 idx = offset & NMASK;
380 return (cip->array[idx]);
386 * Write a given sector, prune the tree if the value is 0
390 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
392 struct indir *cip, *lip[10];
397 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
404 idx = (offset >> cip->shift) & NMASK;
405 up = cip->array[idx];
407 cip = (struct indir *)up;
410 /* Allocate branch */
412 (uintptr_t)new_indir(cip->shift - nshift);
413 if (cip->array[idx] == 0)
416 up = cip->array[idx];
417 cip = (struct indir *)up;
421 idx = offset & NMASK;
422 up = cip->array[idx];
425 cip->array[idx] = ptr;
430 if (cip->used != 0 || li == 1)
433 while (cip->used == 0 && cip != ip) {
435 idx = (offset >> lip[li]->shift) & NMASK;
436 up = lip[li]->array[idx];
437 KASSERT(up == (uintptr_t)cip, ("md screwed up"));
439 lip[li]->array[idx] = 0;
448 g_md_access(struct g_provider *pp, int r, int w, int e)
452 sc = pp->geom->softc;
454 if (r <= 0 && w <= 0 && e <= 0)
461 if ((sc->flags & MD_READONLY) != 0 && w > 0)
463 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
465 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
472 g_md_start(struct bio *bp)
476 sc = bp->bio_to->geom->softc;
477 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
478 mtx_lock(&sc->stat_mtx);
479 devstat_start_transaction_bio(sc->devstat, bp);
480 mtx_unlock(&sc->stat_mtx);
482 mtx_lock(&sc->queue_mtx);
483 bioq_disksort(&sc->bio_queue, bp);
484 mtx_unlock(&sc->queue_mtx);
488 #define MD_MALLOC_MOVE_ZERO 1
489 #define MD_MALLOC_MOVE_FILL 2
490 #define MD_MALLOC_MOVE_READ 3
491 #define MD_MALLOC_MOVE_WRITE 4
492 #define MD_MALLOC_MOVE_CMP 5
495 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize,
496 void *ptr, u_char fill, int op)
503 int error, i, ma_offs1, sz, first_read;
508 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */
516 for (n = sectorsize; n != 0; n -= sz) {
517 sz = imin(PAGE_SIZE - *ma_offs, n);
522 sf = sf_buf_alloc(m, SFB_CPUPRIVATE |
523 (md_malloc_wait ? 0 : SFB_NOWAIT));
529 p = (char *)sf_buf_kva(sf) + *ma_offs;
531 case MD_MALLOC_MOVE_ZERO:
534 case MD_MALLOC_MOVE_FILL:
537 case MD_MALLOC_MOVE_READ:
539 cpu_flush_dcache(p, sz);
541 case MD_MALLOC_MOVE_WRITE:
544 case MD_MALLOC_MOVE_CMP:
545 for (i = 0; i < sz; i++, p++) {
550 } else if (*p != first) {
557 KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op));
563 *ma_offs %= PAGE_SIZE;
566 ptr = (char *)ptr + sz;
572 if (op == MD_MALLOC_MOVE_CMP && error != 0) {
580 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs,
581 unsigned len, void *ptr, u_char fill, int op)
583 bus_dma_segment_t *vlist;
584 uint8_t *p, *end, first;
586 int ma_offs, seg_len;
592 for (; len != 0; len -= seg_len) {
593 seg_len = imin(vlist->ds_len - ma_offs, len);
594 p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs;
596 case MD_MALLOC_MOVE_ZERO:
599 case MD_MALLOC_MOVE_FILL:
600 memset(p, fill, seg_len);
602 case MD_MALLOC_MOVE_READ:
603 bcopy(ptr, p, seg_len);
604 cpu_flush_dcache(p, seg_len);
606 case MD_MALLOC_MOVE_WRITE:
607 bcopy(p, ptr, seg_len);
609 case MD_MALLOC_MOVE_CMP:
612 /* Confirm all following bytes match the first */
619 KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op));
624 if (ma_offs == vlist->ds_len) {
628 ptr = (uint8_t *)ptr + seg_len;
637 mdstart_malloc(struct md_s *sc, struct bio *bp)
641 bus_dma_segment_t *vlist;
642 int i, error, error1, ma_offs, notmapped;
643 off_t secno, nsec, uc;
646 switch (bp->bio_cmd) {
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;
660 ma_offs = bp->bio_ma_offset;
662 KASSERT(vlist == NULL, ("vlists cannot be unmapped"));
663 } else if (vlist != NULL) {
664 ma_offs = bp->bio_ma_offset;
670 nsec = bp->bio_length / sc->sectorsize;
671 secno = bp->bio_offset / sc->sectorsize;
674 osp = s_read(sc->indir, secno);
675 if (bp->bio_cmd == BIO_DELETE) {
677 error = s_write(sc->indir, secno, 0);
678 } else if (bp->bio_cmd == BIO_READ) {
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);
689 bzero(dst, sc->sectorsize);
690 } else if (osp <= 255) {
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);
700 memset(dst, osp, sc->sectorsize);
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,
710 MD_MALLOC_MOVE_READ);
712 bcopy((void *)osp, dst, sc->sectorsize);
713 cpu_flush_dcache(dst, sc->sectorsize);
717 } else if (bp->bio_cmd == BIO_WRITE) {
718 if (sc->flags & MD_COMPRESS) {
720 error1 = md_malloc_move_ma(&m, &ma_offs,
721 sc->sectorsize, &uc, 0,
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,
728 i = error1 == 0 ? sc->sectorsize : 0;
731 for (i = 1; i < sc->sectorsize; i++) {
740 if (i == sc->sectorsize) {
742 error = s_write(sc->indir, secno, uc);
745 sp = (uintptr_t)uma_zalloc(sc->uma,
746 md_malloc_wait ? M_WAITOK :
753 error = md_malloc_move_ma(&m,
754 &ma_offs, sc->sectorsize,
756 MD_MALLOC_MOVE_WRITE);
757 } else if (vlist != NULL) {
758 error = md_malloc_move_vlist(
760 sc->sectorsize, (void *)sp,
761 0, MD_MALLOC_MOVE_WRITE);
763 bcopy(dst, (void *)sp,
766 error = s_write(sc->indir, secno, sp);
769 error = md_malloc_move_ma(&m,
770 &ma_offs, sc->sectorsize,
772 MD_MALLOC_MOVE_WRITE);
773 } else if (vlist != NULL) {
774 error = md_malloc_move_vlist(
776 sc->sectorsize, (void *)osp,
777 0, MD_MALLOC_MOVE_WRITE);
779 bcopy(dst, (void *)osp,
789 uma_zfree(sc->uma, (void*)osp);
793 if (!notmapped && vlist == NULL)
794 dst += sc->sectorsize;
801 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len)
805 while (offset >= vlist->ds_len) {
806 offset -= vlist->ds_len;
811 seg_len = omin(len, vlist->ds_len - offset);
812 bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset),
815 src = (uint8_t *)src + seg_len;
822 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len)
826 while (offset >= vlist->ds_len) {
827 offset -= vlist->ds_len;
832 seg_len = omin(len, vlist->ds_len - offset);
833 bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst,
836 dst = (uint8_t *)dst + seg_len;
843 mdstart_preload(struct md_s *sc, struct bio *bp)
847 p = sc->pl_ptr + bp->bio_offset;
848 switch (bp->bio_cmd) {
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);
854 bcopy(p, bp->bio_data, bp->bio_length);
856 cpu_flush_dcache(bp->bio_data, bp->bio_length);
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);
863 bcopy(bp->bio_data, p, bp->bio_length);
872 mdstart_vnode(struct md_s *sc, struct bio *bp)
881 bus_dma_segment_t *vlist;
883 off_t iolen, iostart, len, zerosize;
886 switch (bp->bio_cmd) {
888 auio.uio_rw = UIO_READ;
892 auio.uio_rw = UIO_WRITE;
904 ma_offs = bp->bio_ma_offset;
905 len = bp->bio_length;
910 * If an error occurs, we set BIO_ERROR but we do not set
911 * B_INVAL because (for a write anyway), the buffer is
915 if (bp->bio_cmd == BIO_FLUSH) {
916 (void) vn_start_write(vp, &mp, V_WAIT);
917 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
918 error = VOP_FSYNC(vp, MNT_WAIT, td);
920 vn_finished_write(mp);
924 auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
925 auio.uio_resid = bp->bio_length;
926 auio.uio_segflg = UIO_SYSSPACE;
929 if (bp->bio_cmd == BIO_DELETE) {
931 * Emulate BIO_DELETE by writing zeros.
933 zerosize = ZERO_REGION_SIZE -
934 (ZERO_REGION_SIZE % sc->sectorsize);
935 auio.uio_iovcnt = howmany(bp->bio_length, zerosize);
936 piov = malloc(sizeof(*piov) * auio.uio_iovcnt, M_MD, M_WAITOK);
939 piov->iov_base = __DECONST(void *, zero_region);
942 piov->iov_len = zerosize;
943 len -= piov->iov_len;
947 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
948 piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK);
950 vlist = (bus_dma_segment_t *)bp->bio_data;
952 piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr +
954 piov->iov_len = vlist->ds_len - ma_offs;
955 if (piov->iov_len > len)
957 len -= piov->iov_len;
962 auio.uio_iovcnt = piov - auio.uio_iov;
964 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
965 pb = getpbuf(&md_vnode_pbuf_freecnt);
968 npages = atop(min(MAXPHYS, round_page(len + (ma_offs &
970 iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len);
971 KASSERT(iolen > 0, ("zero iolen"));
972 pmap_qenter((vm_offset_t)pb->b_data,
973 &bp->bio_ma[atop(ma_offs)], npages);
974 aiov.iov_base = (void *)((vm_offset_t)pb->b_data +
975 (ma_offs & PAGE_MASK));
976 aiov.iov_len = iolen;
977 auio.uio_iov = &aiov;
979 auio.uio_resid = iolen;
981 aiov.iov_base = bp->bio_data;
982 aiov.iov_len = bp->bio_length;
983 auio.uio_iov = &aiov;
986 iostart = auio.uio_offset;
987 if (auio.uio_rw == UIO_READ) {
988 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
989 error = VOP_READ(vp, &auio, 0, sc->cred);
992 (void) vn_start_write(vp, &mp, V_WAIT);
993 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
994 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
997 vn_finished_write(mp);
999 sc->flags &= ~MD_VERIFY;
1002 /* When MD_CACHE is set, try to avoid double-caching the data. */
1003 if (error == 0 && (sc->flags & MD_CACHE) == 0)
1004 VOP_ADVISE(vp, iostart, auio.uio_offset - 1,
1005 POSIX_FADV_DONTNEED);
1008 pmap_qremove((vm_offset_t)pb->b_data, npages);
1011 bp->bio_resid -= iolen;
1016 relpbuf(pb, &md_vnode_pbuf_freecnt);
1021 bp->bio_resid = auio.uio_resid;
1026 md_swap_page_free(vm_page_t m)
1036 mdstart_swap(struct md_s *sc, struct bio *bp)
1040 vm_pindex_t i, lastp;
1041 bus_dma_segment_t *vlist;
1042 int rv, ma_offs, offs, len, lastend;
1044 switch (bp->bio_cmd) {
1050 return (EOPNOTSUPP);
1054 ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ?
1055 bp->bio_ma_offset : 0;
1056 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
1057 (bus_dma_segment_t *)bp->bio_data : NULL;
1060 * offs is the offset at which to start operating on the
1061 * next (ie, first) page. lastp is the last page on
1062 * which we're going to operate. lastend is the ending
1063 * position within that last page (ie, PAGE_SIZE if
1064 * we're operating on complete aligned pages).
1066 offs = bp->bio_offset % PAGE_SIZE;
1067 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
1068 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
1071 VM_OBJECT_WLOCK(sc->object);
1072 vm_object_pip_add(sc->object, 1);
1073 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
1074 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
1075 m = vm_page_grab(sc->object, i, VM_ALLOC_SYSTEM);
1076 if (bp->bio_cmd == BIO_READ) {
1077 if (m->valid == VM_PAGE_BITS_ALL)
1080 rv = vm_pager_get_pages(sc->object, &m, 1,
1082 if (rv == VM_PAGER_ERROR) {
1083 md_swap_page_free(m);
1085 } else if (rv == VM_PAGER_FAIL) {
1087 * Pager does not have the page. Zero
1088 * the allocated page, and mark it as
1089 * valid. Do not set dirty, the page
1090 * can be recreated if thrown out.
1093 m->valid = VM_PAGE_BITS_ALL;
1095 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1096 pmap_copy_pages(&m, offs, bp->bio_ma,
1098 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1099 physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs,
1100 vlist, ma_offs, len);
1101 cpu_flush_dcache(p, len);
1103 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1104 cpu_flush_dcache(p, len);
1106 } else if (bp->bio_cmd == BIO_WRITE) {
1107 if (len == PAGE_SIZE || m->valid == VM_PAGE_BITS_ALL)
1110 rv = vm_pager_get_pages(sc->object, &m, 1,
1112 if (rv == VM_PAGER_ERROR) {
1113 md_swap_page_free(m);
1115 } else if (rv == VM_PAGER_FAIL)
1118 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1119 pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1121 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1122 physcopyin_vlist(vlist, ma_offs,
1123 VM_PAGE_TO_PHYS(m) + offs, len);
1125 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1128 m->valid = VM_PAGE_BITS_ALL;
1129 if (m->dirty != VM_PAGE_BITS_ALL) {
1131 vm_pager_page_unswapped(m);
1133 } else if (bp->bio_cmd == BIO_DELETE) {
1134 if (len == PAGE_SIZE || m->valid == VM_PAGE_BITS_ALL)
1137 rv = vm_pager_get_pages(sc->object, &m, 1,
1139 if (rv == VM_PAGER_ERROR) {
1140 md_swap_page_free(m);
1142 } else if (rv == VM_PAGER_FAIL) {
1143 md_swap_page_free(m);
1146 /* Page is valid. */
1147 if (len != PAGE_SIZE) {
1148 pmap_zero_page_area(m, offs, len);
1149 if (m->dirty != VM_PAGE_BITS_ALL) {
1151 vm_pager_page_unswapped(m);
1154 vm_pager_page_unswapped(m);
1155 md_swap_page_free(m);
1163 if (vm_page_active(m))
1164 vm_page_reference(m);
1166 vm_page_activate(m);
1170 /* Actions on further pages start at offset 0 */
1171 p += PAGE_SIZE - offs;
1175 vm_object_pip_wakeup(sc->object);
1176 VM_OBJECT_WUNLOCK(sc->object);
1177 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
1181 mdstart_null(struct md_s *sc, struct bio *bp)
1184 switch (bp->bio_cmd) {
1186 bzero(bp->bio_data, bp->bio_length);
1187 cpu_flush_dcache(bp->bio_data, bp->bio_length);
1197 md_kthread(void *arg)
1204 thread_lock(curthread);
1205 sched_prio(curthread, PRIBIO);
1206 thread_unlock(curthread);
1207 if (sc->type == MD_VNODE)
1208 curthread->td_pflags |= TDP_NORUNNINGBUF;
1211 mtx_lock(&sc->queue_mtx);
1212 if (sc->flags & MD_SHUTDOWN) {
1213 sc->flags |= MD_EXITING;
1214 mtx_unlock(&sc->queue_mtx);
1217 bp = bioq_takefirst(&sc->bio_queue);
1219 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1222 mtx_unlock(&sc->queue_mtx);
1223 if (bp->bio_cmd == BIO_GETATTR) {
1224 int isv = ((sc->flags & MD_VERIFY) != 0);
1226 if ((sc->fwsectors && sc->fwheads &&
1227 (g_handleattr_int(bp, "GEOM::fwsectors",
1229 g_handleattr_int(bp, "GEOM::fwheads",
1231 g_handleattr_int(bp, "GEOM::candelete", 1))
1233 else if (sc->ident[0] != '\0' &&
1234 g_handleattr_str(bp, "GEOM::ident", sc->ident))
1236 else if (g_handleattr_int(bp, "MNT::verified", isv))
1241 error = sc->start(sc, bp);
1245 bp->bio_completed = bp->bio_length;
1246 g_io_deliver(bp, error);
1248 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE)
1249 devstat_end_transaction_bio(sc->devstat, bp);
1253 static struct md_s *
1258 LIST_FOREACH(sc, &md_softc_list, list) {
1259 if (sc->unit == unit)
1265 static struct md_s *
1266 mdnew(int unit, int *errp, enum md_types type)
1273 unit = alloc_unr(md_uh);
1275 unit = alloc_unr_specific(md_uh, unit);
1282 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
1284 bioq_init(&sc->bio_queue);
1285 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1286 mtx_init(&sc->stat_mtx, "md stat", NULL, MTX_DEF);
1288 sprintf(sc->name, "md%d", unit);
1289 LIST_INSERT_HEAD(&md_softc_list, sc, list);
1290 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1293 LIST_REMOVE(sc, list);
1294 mtx_destroy(&sc->stat_mtx);
1295 mtx_destroy(&sc->queue_mtx);
1296 free_unr(md_uh, sc->unit);
1303 mdinit(struct md_s *sc)
1306 struct g_provider *pp;
1309 gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1311 pp = g_new_providerf(gp, "md%d", sc->unit);
1312 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1313 pp->mediasize = sc->mediasize;
1314 pp->sectorsize = sc->sectorsize;
1319 pp->flags |= G_PF_ACCEPT_UNMAPPED;
1327 g_error_provider(pp, 0);
1328 g_topology_unlock();
1329 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1330 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
1334 mdcreate_malloc(struct md_s *sc, struct md_req *mdr)
1341 if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1343 if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize))
1345 /* Compression doesn't make sense if we have reserved space */
1346 if (mdr->md_options & MD_RESERVE)
1347 mdr->md_options &= ~MD_COMPRESS;
1348 if (mdr->md_fwsectors != 0)
1349 sc->fwsectors = mdr->md_fwsectors;
1350 if (mdr->md_fwheads != 0)
1351 sc->fwheads = mdr->md_fwheads;
1352 sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE);
1353 sc->indir = dimension(sc->mediasize / sc->sectorsize);
1354 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
1356 if (mdr->md_options & MD_RESERVE) {
1359 nsectors = sc->mediasize / sc->sectorsize;
1360 for (u = 0; u < nsectors; u++) {
1361 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ?
1362 M_WAITOK : M_NOWAIT) | M_ZERO);
1364 error = s_write(sc->indir, u, sp);
1376 mdsetcred(struct md_s *sc, struct ucred *cred)
1382 * Set credits in our softc
1387 sc->cred = crhold(cred);
1390 * Horrible kludge to establish credentials for NFS XXX.
1397 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1398 bzero(&auio, sizeof(auio));
1400 aiov.iov_base = tmpbuf;
1401 aiov.iov_len = sc->sectorsize;
1402 auio.uio_iov = &aiov;
1403 auio.uio_iovcnt = 1;
1404 auio.uio_offset = 0;
1405 auio.uio_rw = UIO_READ;
1406 auio.uio_segflg = UIO_SYSSPACE;
1407 auio.uio_resid = aiov.iov_len;
1408 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1409 error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
1410 VOP_UNLOCK(sc->vnode, 0);
1411 free(tmpbuf, M_TEMP);
1417 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1420 struct nameidata nd;
1424 fname = mdr->md_file;
1425 if (mdr->md_file_seg == UIO_USERSPACE) {
1426 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
1429 } else if (mdr->md_file_seg == UIO_SYSSPACE)
1430 strlcpy(sc->file, fname, sizeof(sc->file));
1435 * If the user specified that this is a read only device, don't
1436 * set the FWRITE mask before trying to open the backing store.
1438 flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \
1439 | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0);
1440 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td);
1441 error = vn_open(&nd, &flags, 0, NULL);
1444 NDFREE(&nd, NDF_ONLY_PNBUF);
1445 if (nd.ni_vp->v_type != VREG) {
1449 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1452 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1453 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1454 if (nd.ni_vp->v_iflag & VI_DOOMED) {
1455 /* Forced unmount. */
1460 nd.ni_vp->v_vflag |= VV_MD;
1461 VOP_UNLOCK(nd.ni_vp, 0);
1463 if (mdr->md_fwsectors != 0)
1464 sc->fwsectors = mdr->md_fwsectors;
1465 if (mdr->md_fwheads != 0)
1466 sc->fwheads = mdr->md_fwheads;
1467 snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1468 (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1469 sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE |
1471 if (!(flags & FWRITE))
1472 sc->flags |= MD_READONLY;
1473 sc->vnode = nd.ni_vp;
1475 error = mdsetcred(sc, td->td_ucred);
1478 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1479 nd.ni_vp->v_vflag &= ~VV_MD;
1484 VOP_UNLOCK(nd.ni_vp, 0);
1485 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1490 mddestroy(struct md_s *sc, struct thread *td)
1494 sc->gp->softc = NULL;
1496 g_wither_geom(sc->gp, ENXIO);
1497 g_topology_unlock();
1502 devstat_remove_entry(sc->devstat);
1505 mtx_lock(&sc->queue_mtx);
1506 sc->flags |= MD_SHUTDOWN;
1508 while (!(sc->flags & MD_EXITING))
1509 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1510 mtx_unlock(&sc->queue_mtx);
1511 mtx_destroy(&sc->stat_mtx);
1512 mtx_destroy(&sc->queue_mtx);
1513 if (sc->vnode != NULL) {
1514 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1515 sc->vnode->v_vflag &= ~VV_MD;
1516 VOP_UNLOCK(sc->vnode, 0);
1517 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1518 FREAD : (FREAD|FWRITE), sc->cred, td);
1520 if (sc->cred != NULL)
1522 if (sc->object != NULL)
1523 vm_object_deallocate(sc->object);
1525 destroy_indir(sc, sc->indir);
1527 uma_zdestroy(sc->uma);
1529 LIST_REMOVE(sc, list);
1530 free_unr(md_uh, sc->unit);
1536 mdresize(struct md_s *sc, struct md_req *mdr)
1539 vm_pindex_t oldpages, newpages;
1546 if (mdr->md_mediasize <= 0 ||
1547 (mdr->md_mediasize % PAGE_SIZE) != 0)
1549 oldpages = OFF_TO_IDX(round_page(sc->mediasize));
1550 newpages = OFF_TO_IDX(round_page(mdr->md_mediasize));
1551 if (newpages < oldpages) {
1552 VM_OBJECT_WLOCK(sc->object);
1553 vm_object_page_remove(sc->object, newpages, 0, 0);
1554 swap_pager_freespace(sc->object, newpages,
1555 oldpages - newpages);
1556 swap_release_by_cred(IDX_TO_OFF(oldpages -
1557 newpages), sc->cred);
1558 sc->object->charge = IDX_TO_OFF(newpages);
1559 sc->object->size = newpages;
1560 VM_OBJECT_WUNLOCK(sc->object);
1561 } else if (newpages > oldpages) {
1562 res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1563 oldpages), sc->cred);
1566 if ((mdr->md_options & MD_RESERVE) ||
1567 (sc->flags & MD_RESERVE)) {
1568 error = swap_pager_reserve(sc->object,
1569 oldpages, newpages - oldpages);
1571 swap_release_by_cred(
1572 IDX_TO_OFF(newpages - oldpages),
1577 VM_OBJECT_WLOCK(sc->object);
1578 sc->object->charge = IDX_TO_OFF(newpages);
1579 sc->object->size = newpages;
1580 VM_OBJECT_WUNLOCK(sc->object);
1584 return (EOPNOTSUPP);
1587 sc->mediasize = mdr->md_mediasize;
1589 g_resize_provider(sc->pp, sc->mediasize);
1590 g_topology_unlock();
1595 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1601 * Range check. Disallow negative sizes and sizes not being
1602 * multiple of page size.
1604 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1608 * Allocate an OBJT_SWAP object.
1610 * Note the truncation.
1613 if ((mdr->md_options & MD_VERIFY) != 0)
1615 npage = mdr->md_mediasize / PAGE_SIZE;
1616 if (mdr->md_fwsectors != 0)
1617 sc->fwsectors = mdr->md_fwsectors;
1618 if (mdr->md_fwheads != 0)
1619 sc->fwheads = mdr->md_fwheads;
1620 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1621 VM_PROT_DEFAULT, 0, td->td_ucred);
1622 if (sc->object == NULL)
1624 sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1625 if (mdr->md_options & MD_RESERVE) {
1626 if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1631 error = mdsetcred(sc, td->td_ucred);
1634 vm_object_deallocate(sc->object);
1641 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1645 * Range check. Disallow negative sizes and sizes not being
1646 * multiple of page size.
1648 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1655 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1661 sx_assert(&md_sx, SA_XLOCKED);
1663 switch (mdr->md_type) {
1673 if (mdr->md_sectorsize == 0)
1674 sectsize = DEV_BSIZE;
1676 sectsize = mdr->md_sectorsize;
1677 if (sectsize > MAXPHYS || mdr->md_mediasize < sectsize)
1679 if (mdr->md_options & MD_AUTOUNIT)
1680 sc = mdnew(-1, &error, mdr->md_type);
1682 if (mdr->md_unit > INT_MAX)
1684 sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1688 if (mdr->md_label != NULL)
1689 error = copyinstr(mdr->md_label, sc->label,
1690 sizeof(sc->label), NULL);
1693 if (mdr->md_options & MD_AUTOUNIT)
1694 mdr->md_unit = sc->unit;
1695 sc->mediasize = mdr->md_mediasize;
1696 sc->sectorsize = sectsize;
1700 sc->start = mdstart_malloc;
1701 error = mdcreate_malloc(sc, mdr);
1705 * We disallow attaching preloaded memory disks via
1706 * ioctl. Preloaded memory disks are automatically
1707 * attached in g_md_init().
1712 sc->start = mdstart_vnode;
1713 error = mdcreate_vnode(sc, mdr, td);
1716 sc->start = mdstart_swap;
1717 error = mdcreate_swap(sc, mdr, td);
1720 sc->start = mdstart_null;
1721 error = mdcreate_null(sc, mdr, td);
1730 /* Prune off any residual fractional sector */
1731 i = sc->mediasize % sc->sectorsize;
1739 kern_mdattach(struct thread *td, struct md_req *mdr)
1744 error = kern_mdattach_locked(td, mdr);
1750 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1754 sx_assert(&md_sx, SA_XLOCKED);
1756 if (mdr->md_mediasize != 0 ||
1757 (mdr->md_options & ~MD_FORCE) != 0)
1760 sc = mdfind(mdr->md_unit);
1763 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1764 !(mdr->md_options & MD_FORCE))
1766 return (mddestroy(sc, td));
1770 kern_mddetach(struct thread *td, struct md_req *mdr)
1775 error = kern_mddetach_locked(td, mdr);
1781 kern_mdresize_locked(struct md_req *mdr)
1785 sx_assert(&md_sx, SA_XLOCKED);
1787 if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1790 sc = mdfind(mdr->md_unit);
1793 if (mdr->md_mediasize < sc->sectorsize)
1795 if (mdr->md_mediasize < sc->mediasize &&
1796 !(sc->flags & MD_FORCE) &&
1797 !(mdr->md_options & MD_FORCE))
1799 return (mdresize(sc, mdr));
1803 kern_mdresize(struct md_req *mdr)
1808 error = kern_mdresize_locked(mdr);
1814 kern_mdquery_locked(struct md_req *mdr)
1819 sx_assert(&md_sx, SA_XLOCKED);
1821 sc = mdfind(mdr->md_unit);
1824 mdr->md_type = sc->type;
1825 mdr->md_options = sc->flags;
1826 mdr->md_mediasize = sc->mediasize;
1827 mdr->md_sectorsize = sc->sectorsize;
1829 if (mdr->md_label != NULL) {
1830 error = copyout(sc->label, mdr->md_label,
1831 strlen(sc->label) + 1);
1835 if (sc->type == MD_VNODE ||
1836 (sc->type == MD_PRELOAD && mdr->md_file != NULL))
1837 error = copyout(sc->file, mdr->md_file,
1838 strlen(sc->file) + 1);
1843 kern_mdquery(struct md_req *mdr)
1848 error = kern_mdquery_locked(mdr);
1854 kern_mdlist_locked(struct md_req *mdr)
1859 sx_assert(&md_sx, SA_XLOCKED);
1862 * Write the number of md devices to mdr->md_units[0].
1863 * Write the unit number of the first (mdr->md_units_nitems - 2)
1864 * units to mdr->md_units[1::(mdr->md_units - 2)] and terminate the
1867 * XXX: There is currently no mechanism to retrieve unit
1868 * numbers for more than (MDNPAD - 2) units.
1870 * XXX: Due to the use of LIST_INSERT_HEAD in mdnew(), the
1871 * list of visible unit numbers not stable.
1874 LIST_FOREACH(sc, &md_softc_list, list) {
1875 if (i < mdr->md_units_nitems - 1)
1876 mdr->md_units[i] = sc->unit;
1879 mdr->md_units[MIN(i, mdr->md_units_nitems - 1)] = -1;
1880 mdr->md_units[0] = i - 1;
1885 kern_mdlist(struct md_req *mdr)
1890 error = kern_mdlist_locked(mdr);
1895 /* Copy members that are not userspace pointers. */
1896 #define MD_IOCTL2REQ(mdio, mdr) do { \
1897 (mdr)->md_unit = (mdio)->md_unit; \
1898 (mdr)->md_type = (mdio)->md_type; \
1899 (mdr)->md_mediasize = (mdio)->md_mediasize; \
1900 (mdr)->md_sectorsize = (mdio)->md_sectorsize; \
1901 (mdr)->md_options = (mdio)->md_options; \
1902 (mdr)->md_fwheads = (mdio)->md_fwheads; \
1903 (mdr)->md_fwsectors = (mdio)->md_fwsectors; \
1904 (mdr)->md_units = &(mdio)->md_pad[0]; \
1905 (mdr)->md_units_nitems = nitems((mdio)->md_pad); \
1908 /* Copy members that might have been updated */
1909 #define MD_REQ2IOCTL(mdr, mdio) do { \
1910 (mdio)->md_unit = (mdr)->md_unit; \
1911 (mdio)->md_type = (mdr)->md_type; \
1912 (mdio)->md_mediasize = (mdr)->md_mediasize; \
1913 (mdio)->md_sectorsize = (mdr)->md_sectorsize; \
1914 (mdio)->md_options = (mdr)->md_options; \
1915 (mdio)->md_fwheads = (mdr)->md_fwheads; \
1916 (mdio)->md_fwsectors = (mdr)->md_fwsectors; \
1920 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1927 printf("mdctlioctl(%s %lx %p %x %p)\n",
1928 devtoname(dev), cmd, addr, flags, td);
1930 bzero(&mdr, sizeof(mdr));
1937 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1938 if (mdio->md_version != MDIOVERSION)
1940 MD_IOCTL2REQ(mdio, &mdr);
1941 mdr.md_file = mdio->md_file;
1942 mdr.md_file_seg = UIO_USERSPACE;
1943 /* If the file is adjacent to the md_ioctl it's in kernel. */
1944 if ((void *)mdio->md_file == (void *)(mdio + 1))
1945 mdr.md_file_seg = UIO_SYSSPACE;
1946 mdr.md_label = mdio->md_label;
1949 #ifdef COMPAT_FREEBSD32
1950 case MDIOCATTACH_32:
1951 case MDIOCDETACH_32:
1952 case MDIOCRESIZE_32:
1954 case MDIOCLIST_32: {
1955 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1956 if (mdio->md_version != MDIOVERSION)
1958 MD_IOCTL2REQ(mdio, &mdr);
1959 mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1960 mdr.md_file_seg = UIO_USERSPACE;
1961 mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1966 /* Fall through to handler switch. */
1973 #ifdef COMPAT_FREEBSD32
1974 case MDIOCATTACH_32:
1976 error = kern_mdattach(td, &mdr);
1979 #ifdef COMPAT_FREEBSD32
1980 case MDIOCDETACH_32:
1982 error = kern_mddetach(td, &mdr);
1985 #ifdef COMPAT_FREEBSD32
1986 case MDIOCRESIZE_32:
1988 error = kern_mdresize(&mdr);
1991 #ifdef COMPAT_FREEBSD32
1994 error = kern_mdquery(&mdr);
1997 #ifdef COMPAT_FREEBSD32
2000 error = kern_mdlist(&mdr);
2009 struct md_ioctl *mdio = (struct md_ioctl *)addr;
2010 MD_REQ2IOCTL(&mdr, mdio);
2013 #ifdef COMPAT_FREEBSD32
2014 case MDIOCATTACH_32:
2015 case MDIOCQUERY_32: {
2016 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
2017 MD_REQ2IOCTL(&mdr, mdio);
2022 /* Other commands to not alter mdr. */
2030 md_preloaded(u_char *image, size_t length, const char *name)
2035 sc = mdnew(-1, &error, MD_PRELOAD);
2038 sc->mediasize = length;
2039 sc->sectorsize = DEV_BSIZE;
2041 sc->pl_len = length;
2042 sc->start = mdstart_preload;
2044 strlcpy(sc->file, name, sizeof(sc->file));
2046 if (sc->unit == 0) {
2048 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2050 #ifdef MD_ROOT_READONLY
2051 sc->flags |= MD_READONLY;
2057 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2058 MD_NAME, sc->unit, name, length, image);
2060 printf("%s%d: Embedded image %zd bytes at %p\n",
2061 MD_NAME, sc->unit, length, image);
2066 g_md_init(struct g_class *mp __unused)
2069 u_char *ptr, *name, *type;
2073 /* figure out log2(NINDIR) */
2074 for (i = NINDIR, nshift = -1; i; nshift++)
2078 sx_init(&md_sx, "MD config lock");
2079 g_topology_unlock();
2080 md_uh = new_unrhdr(0, INT_MAX, NULL);
2082 if (mfs_root_size != 0) {
2085 md_preloaded(mfs_root, mfs_root_size, NULL);
2087 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2093 /* XXX: are preload_* static or do they need Giant ? */
2094 while ((mod = preload_search_next_name(mod)) != NULL) {
2095 name = (char *)preload_search_info(mod, MODINFO_NAME);
2098 type = (char *)preload_search_info(mod, MODINFO_TYPE);
2101 if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2103 ptr = preload_fetch_addr(mod);
2104 len = preload_fetch_size(mod);
2105 if (ptr != NULL && len != 0) {
2107 md_preloaded(ptr, len, name);
2111 md_vnode_pbuf_freecnt = nswbuf / 10;
2112 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2118 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2119 struct g_consumer *cp __unused, struct g_provider *pp)
2150 if (indent == NULL) {
2151 sbuf_printf(sb, " u %d", mp->unit);
2152 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2153 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2154 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2155 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2156 sbuf_printf(sb, " t %s", type);
2157 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2158 (mp->type == MD_PRELOAD && mp->file[0] != '\0'))
2159 sbuf_printf(sb, " file %s", mp->file);
2160 sbuf_printf(sb, " label %s", mp->label);
2162 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2164 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2165 indent, (uintmax_t) mp->sectorsize);
2166 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2167 indent, (uintmax_t) mp->fwheads);
2168 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2169 indent, (uintmax_t) mp->fwsectors);
2170 if (mp->ident[0] != '\0') {
2171 sbuf_printf(sb, "%s<ident>", indent);
2172 g_conf_printf_escaped(sb, "%s", mp->ident);
2173 sbuf_printf(sb, "</ident>\n");
2175 sbuf_printf(sb, "%s<length>%ju</length>\n",
2176 indent, (uintmax_t) mp->mediasize);
2177 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2178 (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2179 sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2180 (mp->flags & MD_READONLY) == 0 ? "read-write":
2182 sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2184 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2185 (mp->type == MD_PRELOAD && mp->file[0] != '\0')) {
2186 sbuf_printf(sb, "%s<file>", indent);
2187 g_conf_printf_escaped(sb, "%s", mp->file);
2188 sbuf_printf(sb, "</file>\n");
2190 if (mp->type == MD_VNODE)
2191 sbuf_printf(sb, "%s<cache>%s</cache>\n", indent,
2192 (mp->flags & MD_CACHE) == 0 ? "off": "on");
2193 sbuf_printf(sb, "%s<label>", indent);
2194 g_conf_printf_escaped(sb, "%s", mp->label);
2195 sbuf_printf(sb, "</label>\n");
2201 g_md_fini(struct g_class *mp __unused)
2205 if (status_dev != NULL)
2206 destroy_dev(status_dev);
2207 delete_unrhdr(md_uh);