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. */
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;
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);
202 static g_provgone_t g_md_providergone;
204 static struct cdev *status_dev = NULL;
205 static struct sx md_sx;
206 static struct unrhdr *md_uh;
208 static d_ioctl_t mdctlioctl;
210 static struct cdevsw mdctl_cdevsw = {
211 .d_version = D_VERSION,
212 .d_ioctl = mdctlioctl,
216 struct g_class g_md_class = {
218 .version = G_VERSION,
222 .access = g_md_access,
223 .dumpconf = g_md_dumpconf,
224 .providergone = g_md_providergone,
227 DECLARE_GEOM_CLASS(g_md_class, g_md);
230 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
232 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t))
233 #define NMASK (NINDIR-1)
236 static uma_zone_t md_pbuf_zone;
247 LIST_ENTRY(md_s) list;
248 struct bio_queue_head bio_queue;
249 struct mtx queue_mtx;
263 struct g_provider *pp;
264 int (*start)(struct md_s *sc, struct bio *bp);
265 struct devstat *devstat;
267 /* MD_MALLOC related fields */
271 /* MD_PRELOAD related fields */
275 /* MD_VNODE related fields */
278 char label[PATH_MAX];
281 /* MD_SWAP related fields */
285 static struct indir *
286 new_indir(u_int shift)
290 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
294 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
295 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
296 if (ip->array == NULL) {
306 del_indir(struct indir *ip)
309 free(ip->array, M_MDSECT);
314 destroy_indir(struct md_s *sc, struct indir *ip)
318 for (i = 0; i < NINDIR; i++) {
322 destroy_indir(sc, (struct indir*)(ip->array[i]));
323 else if (ip->array[i] > 255)
324 uma_zfree(sc->uma, (void *)(ip->array[i]));
330 * This function does the math and allocates the top level "indir" structure
331 * for a device of "size" sectors.
334 static struct indir *
335 dimension(off_t size)
343 while (rcnt > NINDIR) {
349 * XXX: the top layer is probably not fully populated, so we allocate
350 * too much space for ip->array in here.
352 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
353 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
354 M_MDSECT, M_WAITOK | M_ZERO);
356 ip->shift = layer * nshift;
361 * Read a given sector
365 s_read(struct indir *ip, off_t offset)
372 printf("s_read(%jd)\n", (intmax_t)offset);
374 for (cip = ip; cip != NULL;) {
376 idx = (offset >> cip->shift) & NMASK;
377 up = cip->array[idx];
378 cip = (struct indir *)up;
381 idx = offset & NMASK;
382 return (cip->array[idx]);
388 * Write a given sector, prune the tree if the value is 0
392 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
394 struct indir *cip, *lip[10];
399 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
406 idx = (offset >> cip->shift) & NMASK;
407 up = cip->array[idx];
409 cip = (struct indir *)up;
412 /* Allocate branch */
414 (uintptr_t)new_indir(cip->shift - nshift);
415 if (cip->array[idx] == 0)
418 up = cip->array[idx];
419 cip = (struct indir *)up;
423 idx = offset & NMASK;
424 up = cip->array[idx];
427 cip->array[idx] = ptr;
432 if (cip->used != 0 || li == 1)
435 while (cip->used == 0 && cip != ip) {
437 idx = (offset >> lip[li]->shift) & NMASK;
438 up = lip[li]->array[idx];
439 KASSERT(up == (uintptr_t)cip, ("md screwed up"));
441 lip[li]->array[idx] = 0;
450 g_md_access(struct g_provider *pp, int r, int w, int e)
454 sc = pp->geom->softc;
456 if (r <= 0 && w <= 0 && e <= 0)
463 if ((sc->flags & MD_READONLY) != 0 && w > 0)
465 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
467 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
474 g_md_start(struct bio *bp)
478 sc = bp->bio_to->geom->softc;
479 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
480 mtx_lock(&sc->stat_mtx);
481 devstat_start_transaction_bio(sc->devstat, bp);
482 mtx_unlock(&sc->stat_mtx);
484 mtx_lock(&sc->queue_mtx);
485 bioq_disksort(&sc->bio_queue, bp);
487 mtx_unlock(&sc->queue_mtx);
490 #define MD_MALLOC_MOVE_ZERO 1
491 #define MD_MALLOC_MOVE_FILL 2
492 #define MD_MALLOC_MOVE_READ 3
493 #define MD_MALLOC_MOVE_WRITE 4
494 #define MD_MALLOC_MOVE_CMP 5
497 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize,
498 void *ptr, u_char fill, int op)
505 int error, i, ma_offs1, sz, first_read;
510 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */
518 for (n = sectorsize; n != 0; n -= sz) {
519 sz = imin(PAGE_SIZE - *ma_offs, n);
524 sf = sf_buf_alloc(m, SFB_CPUPRIVATE |
525 (md_malloc_wait ? 0 : SFB_NOWAIT));
531 p = (char *)sf_buf_kva(sf) + *ma_offs;
533 case MD_MALLOC_MOVE_ZERO:
536 case MD_MALLOC_MOVE_FILL:
539 case MD_MALLOC_MOVE_READ:
541 cpu_flush_dcache(p, sz);
543 case MD_MALLOC_MOVE_WRITE:
546 case MD_MALLOC_MOVE_CMP:
547 for (i = 0; i < sz; i++, p++) {
552 } else if (*p != first) {
559 KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op));
565 *ma_offs %= PAGE_SIZE;
568 ptr = (char *)ptr + sz;
574 if (op == MD_MALLOC_MOVE_CMP && error != 0) {
582 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs,
583 unsigned len, void *ptr, u_char fill, int op)
585 bus_dma_segment_t *vlist;
586 uint8_t *p, *end, first;
588 int ma_offs, seg_len;
594 for (; len != 0; len -= seg_len) {
595 seg_len = imin(vlist->ds_len - ma_offs, len);
596 p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs;
598 case MD_MALLOC_MOVE_ZERO:
601 case MD_MALLOC_MOVE_FILL:
602 memset(p, fill, seg_len);
604 case MD_MALLOC_MOVE_READ:
605 bcopy(ptr, p, seg_len);
606 cpu_flush_dcache(p, seg_len);
608 case MD_MALLOC_MOVE_WRITE:
609 bcopy(p, ptr, seg_len);
611 case MD_MALLOC_MOVE_CMP:
614 /* Confirm all following bytes match the first */
621 KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op));
626 if (ma_offs == vlist->ds_len) {
630 ptr = (uint8_t *)ptr + seg_len;
639 mdstart_malloc(struct md_s *sc, struct bio *bp)
643 bus_dma_segment_t *vlist;
644 int i, error, error1, ma_offs, notmapped;
645 off_t secno, nsec, uc;
648 switch (bp->bio_cmd) {
657 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0;
658 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
659 (bus_dma_segment_t *)bp->bio_data : NULL;
662 ma_offs = bp->bio_ma_offset;
664 KASSERT(vlist == NULL, ("vlists cannot be unmapped"));
665 } else if (vlist != NULL) {
666 ma_offs = bp->bio_ma_offset;
672 nsec = bp->bio_length / sc->sectorsize;
673 secno = bp->bio_offset / sc->sectorsize;
676 osp = s_read(sc->indir, secno);
677 if (bp->bio_cmd == BIO_DELETE) {
679 error = s_write(sc->indir, secno, 0);
680 } else if (bp->bio_cmd == BIO_READ) {
683 error = md_malloc_move_ma(&m, &ma_offs,
684 sc->sectorsize, NULL, 0,
685 MD_MALLOC_MOVE_ZERO);
686 } else if (vlist != NULL) {
687 error = md_malloc_move_vlist(&vlist,
688 &ma_offs, sc->sectorsize, NULL, 0,
689 MD_MALLOC_MOVE_ZERO);
691 bzero(dst, sc->sectorsize);
692 } else if (osp <= 255) {
694 error = md_malloc_move_ma(&m, &ma_offs,
695 sc->sectorsize, NULL, osp,
696 MD_MALLOC_MOVE_FILL);
697 } else if (vlist != NULL) {
698 error = md_malloc_move_vlist(&vlist,
699 &ma_offs, sc->sectorsize, NULL, osp,
700 MD_MALLOC_MOVE_FILL);
702 memset(dst, osp, sc->sectorsize);
705 error = md_malloc_move_ma(&m, &ma_offs,
706 sc->sectorsize, (void *)osp, 0,
707 MD_MALLOC_MOVE_READ);
708 } else if (vlist != NULL) {
709 error = md_malloc_move_vlist(&vlist,
710 &ma_offs, sc->sectorsize,
712 MD_MALLOC_MOVE_READ);
714 bcopy((void *)osp, dst, sc->sectorsize);
715 cpu_flush_dcache(dst, sc->sectorsize);
719 } else if (bp->bio_cmd == BIO_WRITE) {
720 if (sc->flags & MD_COMPRESS) {
722 error1 = md_malloc_move_ma(&m, &ma_offs,
723 sc->sectorsize, &uc, 0,
725 i = error1 == 0 ? sc->sectorsize : 0;
726 } else if (vlist != NULL) {
727 error1 = md_malloc_move_vlist(&vlist,
728 &ma_offs, sc->sectorsize, &uc, 0,
730 i = error1 == 0 ? sc->sectorsize : 0;
733 for (i = 1; i < sc->sectorsize; i++) {
742 if (i == sc->sectorsize) {
744 error = s_write(sc->indir, secno, uc);
747 sp = (uintptr_t)uma_zalloc(sc->uma,
748 md_malloc_wait ? M_WAITOK :
755 error = md_malloc_move_ma(&m,
756 &ma_offs, sc->sectorsize,
758 MD_MALLOC_MOVE_WRITE);
759 } else if (vlist != NULL) {
760 error = md_malloc_move_vlist(
762 sc->sectorsize, (void *)sp,
763 0, MD_MALLOC_MOVE_WRITE);
765 bcopy(dst, (void *)sp,
768 error = s_write(sc->indir, secno, sp);
771 error = md_malloc_move_ma(&m,
772 &ma_offs, sc->sectorsize,
774 MD_MALLOC_MOVE_WRITE);
775 } else if (vlist != NULL) {
776 error = md_malloc_move_vlist(
778 sc->sectorsize, (void *)osp,
779 0, MD_MALLOC_MOVE_WRITE);
781 bcopy(dst, (void *)osp,
791 uma_zfree(sc->uma, (void*)osp);
795 if (!notmapped && vlist == NULL)
796 dst += sc->sectorsize;
803 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len)
807 while (offset >= vlist->ds_len) {
808 offset -= vlist->ds_len;
813 seg_len = omin(len, vlist->ds_len - offset);
814 bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset),
817 src = (uint8_t *)src + seg_len;
824 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len)
828 while (offset >= vlist->ds_len) {
829 offset -= vlist->ds_len;
834 seg_len = omin(len, vlist->ds_len - offset);
835 bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst,
838 dst = (uint8_t *)dst + seg_len;
845 mdstart_preload(struct md_s *sc, struct bio *bp)
849 p = sc->pl_ptr + bp->bio_offset;
850 switch (bp->bio_cmd) {
852 if ((bp->bio_flags & BIO_VLIST) != 0) {
853 mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data,
854 bp->bio_ma_offset, bp->bio_length);
856 bcopy(p, bp->bio_data, bp->bio_length);
858 cpu_flush_dcache(bp->bio_data, bp->bio_length);
861 if ((bp->bio_flags & BIO_VLIST) != 0) {
862 mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data,
863 bp->bio_ma_offset, p, bp->bio_length);
865 bcopy(bp->bio_data, p, bp->bio_length);
874 mdstart_vnode(struct md_s *sc, struct bio *bp)
883 bus_dma_segment_t *vlist;
885 off_t iolen, iostart, len, zerosize;
888 switch (bp->bio_cmd) {
890 auio.uio_rw = UIO_READ;
894 auio.uio_rw = UIO_WRITE;
906 ma_offs = bp->bio_ma_offset;
907 len = bp->bio_length;
912 * If an error occurs, we set BIO_ERROR but we do not set
913 * B_INVAL because (for a write anyway), the buffer is
917 if (bp->bio_cmd == BIO_FLUSH) {
918 (void) vn_start_write(vp, &mp, V_WAIT);
919 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
920 error = VOP_FSYNC(vp, MNT_WAIT, td);
922 vn_finished_write(mp);
926 auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
927 auio.uio_resid = bp->bio_length;
928 auio.uio_segflg = UIO_SYSSPACE;
931 if (bp->bio_cmd == BIO_DELETE) {
933 * Emulate BIO_DELETE by writing zeros.
935 zerosize = ZERO_REGION_SIZE -
936 (ZERO_REGION_SIZE % sc->sectorsize);
937 auio.uio_iovcnt = howmany(bp->bio_length, zerosize);
938 piov = malloc(sizeof(*piov) * auio.uio_iovcnt, M_MD, M_WAITOK);
941 piov->iov_base = __DECONST(void *, zero_region);
944 piov->iov_len = zerosize;
945 len -= piov->iov_len;
949 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
950 piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK);
952 vlist = (bus_dma_segment_t *)bp->bio_data;
954 piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr +
956 piov->iov_len = vlist->ds_len - ma_offs;
957 if (piov->iov_len > len)
959 len -= piov->iov_len;
964 auio.uio_iovcnt = piov - auio.uio_iov;
966 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
967 pb = uma_zalloc(md_pbuf_zone, M_WAITOK);
970 npages = atop(min(MAXPHYS, round_page(len + (ma_offs &
972 iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len);
973 KASSERT(iolen > 0, ("zero iolen"));
974 pmap_qenter((vm_offset_t)pb->b_data,
975 &bp->bio_ma[atop(ma_offs)], npages);
976 aiov.iov_base = (void *)((vm_offset_t)pb->b_data +
977 (ma_offs & PAGE_MASK));
978 aiov.iov_len = iolen;
979 auio.uio_iov = &aiov;
981 auio.uio_resid = iolen;
983 aiov.iov_base = bp->bio_data;
984 aiov.iov_len = bp->bio_length;
985 auio.uio_iov = &aiov;
988 iostart = auio.uio_offset;
989 if (auio.uio_rw == UIO_READ) {
990 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
991 error = VOP_READ(vp, &auio, 0, sc->cred);
994 (void) vn_start_write(vp, &mp, V_WAIT);
995 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
996 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
999 vn_finished_write(mp);
1001 sc->flags &= ~MD_VERIFY;
1004 /* When MD_CACHE is set, try to avoid double-caching the data. */
1005 if (error == 0 && (sc->flags & MD_CACHE) == 0)
1006 VOP_ADVISE(vp, iostart, auio.uio_offset - 1,
1007 POSIX_FADV_DONTNEED);
1010 pmap_qremove((vm_offset_t)pb->b_data, npages);
1013 bp->bio_resid -= iolen;
1018 uma_zfree(md_pbuf_zone, pb);
1023 bp->bio_resid = auio.uio_resid;
1028 md_swap_page_free(vm_page_t m)
1038 mdstart_swap(struct md_s *sc, struct bio *bp)
1042 vm_pindex_t i, lastp;
1043 bus_dma_segment_t *vlist;
1044 int rv, ma_offs, offs, len, lastend;
1046 switch (bp->bio_cmd) {
1052 return (EOPNOTSUPP);
1056 ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ?
1057 bp->bio_ma_offset : 0;
1058 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
1059 (bus_dma_segment_t *)bp->bio_data : NULL;
1062 * offs is the offset at which to start operating on the
1063 * next (ie, first) page. lastp is the last page on
1064 * which we're going to operate. lastend is the ending
1065 * position within that last page (ie, PAGE_SIZE if
1066 * we're operating on complete aligned pages).
1068 offs = bp->bio_offset % PAGE_SIZE;
1069 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
1070 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
1073 VM_OBJECT_WLOCK(sc->object);
1074 vm_object_pip_add(sc->object, 1);
1075 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
1076 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
1077 m = vm_page_grab(sc->object, i, VM_ALLOC_SYSTEM);
1078 if (bp->bio_cmd == BIO_READ) {
1079 if (m->valid == VM_PAGE_BITS_ALL)
1082 rv = vm_pager_get_pages(sc->object, &m, 1,
1084 if (rv == VM_PAGER_ERROR) {
1085 md_swap_page_free(m);
1087 } else if (rv == VM_PAGER_FAIL) {
1089 * Pager does not have the page. Zero
1090 * the allocated page, and mark it as
1091 * valid. Do not set dirty, the page
1092 * can be recreated if thrown out.
1095 m->valid = VM_PAGE_BITS_ALL;
1097 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1098 pmap_copy_pages(&m, offs, bp->bio_ma,
1100 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1101 physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs,
1102 vlist, ma_offs, len);
1103 cpu_flush_dcache(p, len);
1105 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1106 cpu_flush_dcache(p, len);
1108 } else if (bp->bio_cmd == BIO_WRITE) {
1109 if (len == PAGE_SIZE || m->valid == VM_PAGE_BITS_ALL)
1112 rv = vm_pager_get_pages(sc->object, &m, 1,
1114 if (rv == VM_PAGER_ERROR) {
1115 md_swap_page_free(m);
1117 } else if (rv == VM_PAGER_FAIL)
1120 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1121 pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1123 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1124 physcopyin_vlist(vlist, ma_offs,
1125 VM_PAGE_TO_PHYS(m) + offs, len);
1127 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1130 m->valid = VM_PAGE_BITS_ALL;
1131 if (m->dirty != VM_PAGE_BITS_ALL) {
1133 vm_pager_page_unswapped(m);
1135 } else if (bp->bio_cmd == BIO_DELETE) {
1136 if (len == PAGE_SIZE || m->valid == VM_PAGE_BITS_ALL)
1139 rv = vm_pager_get_pages(sc->object, &m, 1,
1141 if (rv == VM_PAGER_ERROR) {
1142 md_swap_page_free(m);
1144 } else if (rv == VM_PAGER_FAIL) {
1145 md_swap_page_free(m);
1148 /* Page is valid. */
1149 if (len != PAGE_SIZE) {
1150 pmap_zero_page_area(m, offs, len);
1151 if (m->dirty != VM_PAGE_BITS_ALL) {
1153 vm_pager_page_unswapped(m);
1156 vm_pager_page_unswapped(m);
1157 md_swap_page_free(m);
1165 if (vm_page_active(m))
1166 vm_page_reference(m);
1168 vm_page_activate(m);
1172 /* Actions on further pages start at offset 0 */
1173 p += PAGE_SIZE - offs;
1177 vm_object_pip_wakeup(sc->object);
1178 VM_OBJECT_WUNLOCK(sc->object);
1179 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
1183 mdstart_null(struct md_s *sc, struct bio *bp)
1186 switch (bp->bio_cmd) {
1188 bzero(bp->bio_data, bp->bio_length);
1189 cpu_flush_dcache(bp->bio_data, bp->bio_length);
1199 md_kthread(void *arg)
1206 thread_lock(curthread);
1207 sched_prio(curthread, PRIBIO);
1208 thread_unlock(curthread);
1209 if (sc->type == MD_VNODE)
1210 curthread->td_pflags |= TDP_NORUNNINGBUF;
1213 mtx_lock(&sc->queue_mtx);
1214 if (sc->flags & MD_SHUTDOWN) {
1215 sc->flags |= MD_EXITING;
1216 mtx_unlock(&sc->queue_mtx);
1219 bp = bioq_takefirst(&sc->bio_queue);
1221 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1224 mtx_unlock(&sc->queue_mtx);
1225 if (bp->bio_cmd == BIO_GETATTR) {
1226 int isv = ((sc->flags & MD_VERIFY) != 0);
1228 if ((sc->fwsectors && sc->fwheads &&
1229 (g_handleattr_int(bp, "GEOM::fwsectors",
1231 g_handleattr_int(bp, "GEOM::fwheads",
1233 g_handleattr_int(bp, "GEOM::candelete", 1))
1235 else if (sc->ident[0] != '\0' &&
1236 g_handleattr_str(bp, "GEOM::ident", sc->ident))
1238 else if (g_handleattr_int(bp, "MNT::verified", isv))
1243 error = sc->start(sc, bp);
1246 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
1248 * Devstat uses (bio_bcount, bio_resid) for
1249 * determining the length of the completed part of
1250 * the i/o. g_io_deliver() will translate from
1251 * bio_completed to that, but it also destroys the
1252 * bio so we must do our own translation.
1254 bp->bio_bcount = bp->bio_length;
1255 bp->bio_resid = (error == -1 ? bp->bio_bcount : 0);
1256 devstat_end_transaction_bio(sc->devstat, bp);
1259 bp->bio_completed = bp->bio_length;
1260 g_io_deliver(bp, error);
1265 static struct md_s *
1270 LIST_FOREACH(sc, &md_softc_list, list) {
1271 if (sc->unit == unit)
1277 static struct md_s *
1278 mdnew(int unit, int *errp, enum md_types type)
1285 unit = alloc_unr(md_uh);
1287 unit = alloc_unr_specific(md_uh, unit);
1294 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
1296 bioq_init(&sc->bio_queue);
1297 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1298 mtx_init(&sc->stat_mtx, "md stat", NULL, MTX_DEF);
1300 sprintf(sc->name, "md%d", unit);
1301 LIST_INSERT_HEAD(&md_softc_list, sc, list);
1302 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1305 LIST_REMOVE(sc, list);
1306 mtx_destroy(&sc->stat_mtx);
1307 mtx_destroy(&sc->queue_mtx);
1308 free_unr(md_uh, sc->unit);
1315 mdinit(struct md_s *sc)
1318 struct g_provider *pp;
1321 gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1323 pp = g_new_providerf(gp, "md%d", sc->unit);
1324 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1325 pp->mediasize = sc->mediasize;
1326 pp->sectorsize = sc->sectorsize;
1331 pp->flags |= G_PF_ACCEPT_UNMAPPED;
1339 g_error_provider(pp, 0);
1340 g_topology_unlock();
1341 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1342 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
1346 mdcreate_malloc(struct md_s *sc, struct md_req *mdr)
1353 if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1355 if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize))
1357 /* Compression doesn't make sense if we have reserved space */
1358 if (mdr->md_options & MD_RESERVE)
1359 mdr->md_options &= ~MD_COMPRESS;
1360 if (mdr->md_fwsectors != 0)
1361 sc->fwsectors = mdr->md_fwsectors;
1362 if (mdr->md_fwheads != 0)
1363 sc->fwheads = mdr->md_fwheads;
1364 sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE);
1365 sc->indir = dimension(sc->mediasize / sc->sectorsize);
1366 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
1368 if (mdr->md_options & MD_RESERVE) {
1371 nsectors = sc->mediasize / sc->sectorsize;
1372 for (u = 0; u < nsectors; u++) {
1373 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ?
1374 M_WAITOK : M_NOWAIT) | M_ZERO);
1376 error = s_write(sc->indir, u, sp);
1388 mdsetcred(struct md_s *sc, struct ucred *cred)
1394 * Set credits in our softc
1399 sc->cred = crhold(cred);
1402 * Horrible kludge to establish credentials for NFS XXX.
1409 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1410 bzero(&auio, sizeof(auio));
1412 aiov.iov_base = tmpbuf;
1413 aiov.iov_len = sc->sectorsize;
1414 auio.uio_iov = &aiov;
1415 auio.uio_iovcnt = 1;
1416 auio.uio_offset = 0;
1417 auio.uio_rw = UIO_READ;
1418 auio.uio_segflg = UIO_SYSSPACE;
1419 auio.uio_resid = aiov.iov_len;
1420 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1421 error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
1422 VOP_UNLOCK(sc->vnode, 0);
1423 free(tmpbuf, M_TEMP);
1429 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1432 struct nameidata nd;
1436 fname = mdr->md_file;
1437 if (mdr->md_file_seg == UIO_USERSPACE) {
1438 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
1441 } else if (mdr->md_file_seg == UIO_SYSSPACE)
1442 strlcpy(sc->file, fname, sizeof(sc->file));
1447 * If the user specified that this is a read only device, don't
1448 * set the FWRITE mask before trying to open the backing store.
1450 flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \
1451 | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0);
1452 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td);
1453 error = vn_open(&nd, &flags, 0, NULL);
1456 NDFREE(&nd, NDF_ONLY_PNBUF);
1457 if (nd.ni_vp->v_type != VREG) {
1461 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1464 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1465 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1466 if (nd.ni_vp->v_iflag & VI_DOOMED) {
1467 /* Forced unmount. */
1472 nd.ni_vp->v_vflag |= VV_MD;
1473 VOP_UNLOCK(nd.ni_vp, 0);
1475 if (mdr->md_fwsectors != 0)
1476 sc->fwsectors = mdr->md_fwsectors;
1477 if (mdr->md_fwheads != 0)
1478 sc->fwheads = mdr->md_fwheads;
1479 snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1480 (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1481 sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE |
1483 if (!(flags & FWRITE))
1484 sc->flags |= MD_READONLY;
1485 sc->vnode = nd.ni_vp;
1487 error = mdsetcred(sc, td->td_ucred);
1490 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1491 nd.ni_vp->v_vflag &= ~VV_MD;
1496 VOP_UNLOCK(nd.ni_vp, 0);
1497 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1502 g_md_providergone(struct g_provider *pp)
1504 struct md_s *sc = pp->geom->softc;
1506 mtx_lock(&sc->queue_mtx);
1507 sc->flags |= MD_PROVIDERGONE;
1509 mtx_unlock(&sc->queue_mtx);
1513 mddestroy(struct md_s *sc, struct thread *td)
1518 g_wither_geom(sc->gp, ENXIO);
1519 g_topology_unlock();
1521 mtx_lock(&sc->queue_mtx);
1522 while (!(sc->flags & MD_PROVIDERGONE))
1523 msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0);
1524 mtx_unlock(&sc->queue_mtx);
1527 devstat_remove_entry(sc->devstat);
1530 mtx_lock(&sc->queue_mtx);
1531 sc->flags |= MD_SHUTDOWN;
1533 while (!(sc->flags & MD_EXITING))
1534 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1535 mtx_unlock(&sc->queue_mtx);
1536 mtx_destroy(&sc->stat_mtx);
1537 mtx_destroy(&sc->queue_mtx);
1538 if (sc->vnode != NULL) {
1539 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1540 sc->vnode->v_vflag &= ~VV_MD;
1541 VOP_UNLOCK(sc->vnode, 0);
1542 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1543 FREAD : (FREAD|FWRITE), sc->cred, td);
1545 if (sc->cred != NULL)
1547 if (sc->object != NULL)
1548 vm_object_deallocate(sc->object);
1550 destroy_indir(sc, sc->indir);
1552 uma_zdestroy(sc->uma);
1554 LIST_REMOVE(sc, list);
1555 free_unr(md_uh, sc->unit);
1561 mdresize(struct md_s *sc, struct md_req *mdr)
1564 vm_pindex_t oldpages, newpages;
1571 if (mdr->md_mediasize <= 0 ||
1572 (mdr->md_mediasize % PAGE_SIZE) != 0)
1574 oldpages = OFF_TO_IDX(round_page(sc->mediasize));
1575 newpages = OFF_TO_IDX(round_page(mdr->md_mediasize));
1576 if (newpages < oldpages) {
1577 VM_OBJECT_WLOCK(sc->object);
1578 vm_object_page_remove(sc->object, newpages, 0, 0);
1579 swap_pager_freespace(sc->object, newpages,
1580 oldpages - newpages);
1581 swap_release_by_cred(IDX_TO_OFF(oldpages -
1582 newpages), sc->cred);
1583 sc->object->charge = IDX_TO_OFF(newpages);
1584 sc->object->size = newpages;
1585 VM_OBJECT_WUNLOCK(sc->object);
1586 } else if (newpages > oldpages) {
1587 res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1588 oldpages), sc->cred);
1591 if ((mdr->md_options & MD_RESERVE) ||
1592 (sc->flags & MD_RESERVE)) {
1593 error = swap_pager_reserve(sc->object,
1594 oldpages, newpages - oldpages);
1596 swap_release_by_cred(
1597 IDX_TO_OFF(newpages - oldpages),
1602 VM_OBJECT_WLOCK(sc->object);
1603 sc->object->charge = IDX_TO_OFF(newpages);
1604 sc->object->size = newpages;
1605 VM_OBJECT_WUNLOCK(sc->object);
1609 return (EOPNOTSUPP);
1612 sc->mediasize = mdr->md_mediasize;
1614 g_resize_provider(sc->pp, sc->mediasize);
1615 g_topology_unlock();
1620 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1626 * Range check. Disallow negative sizes and sizes not being
1627 * multiple of page size.
1629 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1633 * Allocate an OBJT_SWAP object.
1635 * Note the truncation.
1638 if ((mdr->md_options & MD_VERIFY) != 0)
1640 npage = mdr->md_mediasize / PAGE_SIZE;
1641 if (mdr->md_fwsectors != 0)
1642 sc->fwsectors = mdr->md_fwsectors;
1643 if (mdr->md_fwheads != 0)
1644 sc->fwheads = mdr->md_fwheads;
1645 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1646 VM_PROT_DEFAULT, 0, td->td_ucred);
1647 if (sc->object == NULL)
1649 sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1650 if (mdr->md_options & MD_RESERVE) {
1651 if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1656 error = mdsetcred(sc, td->td_ucred);
1659 vm_object_deallocate(sc->object);
1666 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1670 * Range check. Disallow negative sizes and sizes not being
1671 * multiple of page size.
1673 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1680 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1686 sx_assert(&md_sx, SA_XLOCKED);
1688 switch (mdr->md_type) {
1698 if (mdr->md_sectorsize == 0)
1699 sectsize = DEV_BSIZE;
1701 sectsize = mdr->md_sectorsize;
1702 if (sectsize > MAXPHYS || mdr->md_mediasize < sectsize)
1704 if (mdr->md_options & MD_AUTOUNIT)
1705 sc = mdnew(-1, &error, mdr->md_type);
1707 if (mdr->md_unit > INT_MAX)
1709 sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1713 if (mdr->md_label != NULL)
1714 error = copyinstr(mdr->md_label, sc->label,
1715 sizeof(sc->label), NULL);
1718 if (mdr->md_options & MD_AUTOUNIT)
1719 mdr->md_unit = sc->unit;
1720 sc->mediasize = mdr->md_mediasize;
1721 sc->sectorsize = sectsize;
1725 sc->start = mdstart_malloc;
1726 error = mdcreate_malloc(sc, mdr);
1730 * We disallow attaching preloaded memory disks via
1731 * ioctl. Preloaded memory disks are automatically
1732 * attached in g_md_init().
1737 sc->start = mdstart_vnode;
1738 error = mdcreate_vnode(sc, mdr, td);
1741 sc->start = mdstart_swap;
1742 error = mdcreate_swap(sc, mdr, td);
1745 sc->start = mdstart_null;
1746 error = mdcreate_null(sc, mdr, td);
1755 /* Prune off any residual fractional sector */
1756 i = sc->mediasize % sc->sectorsize;
1764 kern_mdattach(struct thread *td, struct md_req *mdr)
1769 error = kern_mdattach_locked(td, mdr);
1775 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1779 sx_assert(&md_sx, SA_XLOCKED);
1781 if (mdr->md_mediasize != 0 ||
1782 (mdr->md_options & ~MD_FORCE) != 0)
1785 sc = mdfind(mdr->md_unit);
1788 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1789 !(mdr->md_options & MD_FORCE))
1791 return (mddestroy(sc, td));
1795 kern_mddetach(struct thread *td, struct md_req *mdr)
1800 error = kern_mddetach_locked(td, mdr);
1806 kern_mdresize_locked(struct md_req *mdr)
1810 sx_assert(&md_sx, SA_XLOCKED);
1812 if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1815 sc = mdfind(mdr->md_unit);
1818 if (mdr->md_mediasize < sc->sectorsize)
1820 if (mdr->md_mediasize < sc->mediasize &&
1821 !(sc->flags & MD_FORCE) &&
1822 !(mdr->md_options & MD_FORCE))
1824 return (mdresize(sc, mdr));
1828 kern_mdresize(struct md_req *mdr)
1833 error = kern_mdresize_locked(mdr);
1839 kern_mdquery_locked(struct md_req *mdr)
1844 sx_assert(&md_sx, SA_XLOCKED);
1846 sc = mdfind(mdr->md_unit);
1849 mdr->md_type = sc->type;
1850 mdr->md_options = sc->flags;
1851 mdr->md_mediasize = sc->mediasize;
1852 mdr->md_sectorsize = sc->sectorsize;
1854 if (mdr->md_label != NULL) {
1855 error = copyout(sc->label, mdr->md_label,
1856 strlen(sc->label) + 1);
1860 if (sc->type == MD_VNODE ||
1861 (sc->type == MD_PRELOAD && mdr->md_file != NULL))
1862 error = copyout(sc->file, mdr->md_file,
1863 strlen(sc->file) + 1);
1868 kern_mdquery(struct md_req *mdr)
1873 error = kern_mdquery_locked(mdr);
1878 /* Copy members that are not userspace pointers. */
1879 #define MD_IOCTL2REQ(mdio, mdr) do { \
1880 (mdr)->md_unit = (mdio)->md_unit; \
1881 (mdr)->md_type = (mdio)->md_type; \
1882 (mdr)->md_mediasize = (mdio)->md_mediasize; \
1883 (mdr)->md_sectorsize = (mdio)->md_sectorsize; \
1884 (mdr)->md_options = (mdio)->md_options; \
1885 (mdr)->md_fwheads = (mdio)->md_fwheads; \
1886 (mdr)->md_fwsectors = (mdio)->md_fwsectors; \
1887 (mdr)->md_units = &(mdio)->md_pad[0]; \
1888 (mdr)->md_units_nitems = nitems((mdio)->md_pad); \
1891 /* Copy members that might have been updated */
1892 #define MD_REQ2IOCTL(mdr, mdio) do { \
1893 (mdio)->md_unit = (mdr)->md_unit; \
1894 (mdio)->md_type = (mdr)->md_type; \
1895 (mdio)->md_mediasize = (mdr)->md_mediasize; \
1896 (mdio)->md_sectorsize = (mdr)->md_sectorsize; \
1897 (mdio)->md_options = (mdr)->md_options; \
1898 (mdio)->md_fwheads = (mdr)->md_fwheads; \
1899 (mdio)->md_fwsectors = (mdr)->md_fwsectors; \
1903 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1910 printf("mdctlioctl(%s %lx %p %x %p)\n",
1911 devtoname(dev), cmd, addr, flags, td);
1913 bzero(&mdr, sizeof(mdr));
1919 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1920 if (mdio->md_version != MDIOVERSION)
1922 MD_IOCTL2REQ(mdio, &mdr);
1923 mdr.md_file = mdio->md_file;
1924 mdr.md_file_seg = UIO_USERSPACE;
1925 /* If the file is adjacent to the md_ioctl it's in kernel. */
1926 if ((void *)mdio->md_file == (void *)(mdio + 1))
1927 mdr.md_file_seg = UIO_SYSSPACE;
1928 mdr.md_label = mdio->md_label;
1931 #ifdef COMPAT_FREEBSD32
1932 case MDIOCATTACH_32:
1933 case MDIOCDETACH_32:
1934 case MDIOCRESIZE_32:
1935 case MDIOCQUERY_32: {
1936 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1937 if (mdio->md_version != MDIOVERSION)
1939 MD_IOCTL2REQ(mdio, &mdr);
1940 mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1941 mdr.md_file_seg = UIO_USERSPACE;
1942 mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1947 /* Fall through to handler switch. */
1954 #ifdef COMPAT_FREEBSD32
1955 case MDIOCATTACH_32:
1957 error = kern_mdattach(td, &mdr);
1960 #ifdef COMPAT_FREEBSD32
1961 case MDIOCDETACH_32:
1963 error = kern_mddetach(td, &mdr);
1966 #ifdef COMPAT_FREEBSD32
1967 case MDIOCRESIZE_32:
1969 error = kern_mdresize(&mdr);
1972 #ifdef COMPAT_FREEBSD32
1975 error = kern_mdquery(&mdr);
1984 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1985 MD_REQ2IOCTL(&mdr, mdio);
1988 #ifdef COMPAT_FREEBSD32
1989 case MDIOCATTACH_32:
1990 case MDIOCQUERY_32: {
1991 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1992 MD_REQ2IOCTL(&mdr, mdio);
1997 /* Other commands to not alter mdr. */
2005 md_preloaded(u_char *image, size_t length, const char *name)
2010 sc = mdnew(-1, &error, MD_PRELOAD);
2013 sc->mediasize = length;
2014 sc->sectorsize = DEV_BSIZE;
2016 sc->pl_len = length;
2017 sc->start = mdstart_preload;
2019 strlcpy(sc->file, name, sizeof(sc->file));
2021 if (sc->unit == 0) {
2023 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2025 #ifdef MD_ROOT_READONLY
2026 sc->flags |= MD_READONLY;
2032 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2033 MD_NAME, sc->unit, name, length, image);
2035 printf("%s%d: Embedded image %zd bytes at %p\n",
2036 MD_NAME, sc->unit, length, image);
2041 g_md_init(struct g_class *mp __unused)
2044 u_char *ptr, *name, *type;
2048 /* figure out log2(NINDIR) */
2049 for (i = NINDIR, nshift = -1; i; nshift++)
2053 sx_init(&md_sx, "MD config lock");
2054 g_topology_unlock();
2055 md_uh = new_unrhdr(0, INT_MAX, NULL);
2057 if (mfs_root_size != 0) {
2060 md_preloaded(mfs_root, mfs_root_size, NULL);
2062 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2068 /* XXX: are preload_* static or do they need Giant ? */
2069 while ((mod = preload_search_next_name(mod)) != NULL) {
2070 name = (char *)preload_search_info(mod, MODINFO_NAME);
2073 type = (char *)preload_search_info(mod, MODINFO_TYPE);
2076 if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2078 ptr = preload_fetch_addr(mod);
2079 len = preload_fetch_size(mod);
2080 if (ptr != NULL && len != 0) {
2082 md_preloaded(ptr, len, name);
2086 md_pbuf_zone = pbuf_zsecond_create("mdpbuf", nswbuf / 10);
2087 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2093 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2094 struct g_consumer *cp __unused, struct g_provider *pp)
2125 if (indent == NULL) {
2126 sbuf_printf(sb, " u %d", mp->unit);
2127 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2128 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2129 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2130 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2131 sbuf_printf(sb, " t %s", type);
2132 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2133 (mp->type == MD_PRELOAD && mp->file[0] != '\0'))
2134 sbuf_printf(sb, " file %s", mp->file);
2135 sbuf_printf(sb, " label %s", mp->label);
2137 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2139 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2140 indent, (uintmax_t) mp->sectorsize);
2141 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2142 indent, (uintmax_t) mp->fwheads);
2143 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2144 indent, (uintmax_t) mp->fwsectors);
2145 if (mp->ident[0] != '\0') {
2146 sbuf_printf(sb, "%s<ident>", indent);
2147 g_conf_printf_escaped(sb, "%s", mp->ident);
2148 sbuf_printf(sb, "</ident>\n");
2150 sbuf_printf(sb, "%s<length>%ju</length>\n",
2151 indent, (uintmax_t) mp->mediasize);
2152 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2153 (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2154 sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2155 (mp->flags & MD_READONLY) == 0 ? "read-write":
2157 sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2159 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2160 (mp->type == MD_PRELOAD && mp->file[0] != '\0')) {
2161 sbuf_printf(sb, "%s<file>", indent);
2162 g_conf_printf_escaped(sb, "%s", mp->file);
2163 sbuf_printf(sb, "</file>\n");
2165 if (mp->type == MD_VNODE)
2166 sbuf_printf(sb, "%s<cache>%s</cache>\n", indent,
2167 (mp->flags & MD_CACHE) == 0 ? "off": "on");
2168 sbuf_printf(sb, "%s<label>", indent);
2169 g_conf_printf_escaped(sb, "%s", mp->label);
2170 sbuf_printf(sb, "</label>\n");
2176 g_md_fini(struct g_class *mp __unused)
2180 if (status_dev != NULL)
2181 destroy_dev(status_dev);
2182 uma_zdestroy(md_pbuf_zone);
2183 delete_unrhdr(md_uh);