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_compat.h"
64 #include "opt_rootdevname.h"
68 #include <sys/param.h>
69 #include <sys/systm.h>
73 #include <sys/devicestat.h>
74 #include <sys/fcntl.h>
75 #include <sys/kernel.h>
76 #include <sys/kthread.h>
77 #include <sys/limits.h>
78 #include <sys/linker.h>
80 #include <sys/malloc.h>
81 #include <sys/mdioctl.h>
82 #include <sys/mount.h>
83 #include <sys/mutex.h>
85 #include <sys/namei.h>
87 #include <sys/queue.h>
88 #include <sys/rwlock.h>
90 #include <sys/sched.h>
91 #include <sys/sf_buf.h>
92 #include <sys/sysctl.h>
94 #include <sys/vnode.h>
97 #include <geom/geom.h>
98 #include <geom/geom_int.h>
101 #include <vm/vm_param.h>
102 #include <vm/vm_object.h>
103 #include <vm/vm_page.h>
104 #include <vm/vm_pager.h>
105 #include <vm/swap_pager.h>
108 #include <machine/bus.h>
112 #define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */
113 #define MD_EXITING 0x20000 /* Worker thread is exiting. */
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 MDIOCLIST_32 _IOC_NEWTYPE(MDIOCLIST, struct md_ioctl32)
155 #define MDIOCRESIZE_32 _IOC_NEWTYPE(MDIOCRESIZE, struct md_ioctl32)
156 #endif /* COMPAT_FREEBSD32 */
158 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk");
159 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors");
162 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0,
163 "Enable md(4) debug messages");
164 static int md_malloc_wait;
165 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0,
166 "Allow malloc to wait for memory allocations");
168 #if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE)
169 #define MD_ROOT_FSTYPE "ufs"
174 * Preloaded image gets put here.
176 #if defined(MD_ROOT_SIZE)
178 * We put the mfs_root symbol into the oldmfs section of the kernel object file.
179 * Applications that patch the object with the image can determine
180 * the size looking at the oldmfs section size within the kernel.
182 u_char mfs_root[MD_ROOT_SIZE*1024] __attribute__ ((section ("oldmfs")));
183 const int mfs_root_size = sizeof(mfs_root);
185 extern volatile u_char __weak_symbol mfs_root;
186 extern volatile u_char __weak_symbol mfs_root_end;
188 __GLOBL(mfs_root_end);
189 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root))
193 static g_init_t g_md_init;
194 static g_fini_t g_md_fini;
195 static g_start_t g_md_start;
196 static g_access_t g_md_access;
197 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
198 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
200 static struct cdev *status_dev = NULL;
201 static struct sx md_sx;
202 static struct unrhdr *md_uh;
204 static d_ioctl_t mdctlioctl;
206 static struct cdevsw mdctl_cdevsw = {
207 .d_version = D_VERSION,
208 .d_ioctl = mdctlioctl,
212 struct g_class g_md_class = {
214 .version = G_VERSION,
218 .access = g_md_access,
219 .dumpconf = g_md_dumpconf,
222 DECLARE_GEOM_CLASS(g_md_class, g_md);
225 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
227 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t))
228 #define NMASK (NINDIR-1)
231 static int md_vnode_pbuf_freecnt;
242 LIST_ENTRY(md_s) list;
243 struct bio_queue_head bio_queue;
244 struct mtx queue_mtx;
258 struct g_provider *pp;
259 int (*start)(struct md_s *sc, struct bio *bp);
260 struct devstat *devstat;
262 /* MD_MALLOC related fields */
266 /* MD_PRELOAD related fields */
270 /* MD_VNODE related fields */
273 char label[PATH_MAX];
276 /* MD_SWAP related fields */
280 static struct indir *
281 new_indir(u_int shift)
285 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
289 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
290 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
291 if (ip->array == NULL) {
301 del_indir(struct indir *ip)
304 free(ip->array, M_MDSECT);
309 destroy_indir(struct md_s *sc, struct indir *ip)
313 for (i = 0; i < NINDIR; i++) {
317 destroy_indir(sc, (struct indir*)(ip->array[i]));
318 else if (ip->array[i] > 255)
319 uma_zfree(sc->uma, (void *)(ip->array[i]));
325 * This function does the math and allocates the top level "indir" structure
326 * for a device of "size" sectors.
329 static struct indir *
330 dimension(off_t size)
338 while (rcnt > NINDIR) {
344 * XXX: the top layer is probably not fully populated, so we allocate
345 * too much space for ip->array in here.
347 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
348 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
349 M_MDSECT, M_WAITOK | M_ZERO);
351 ip->shift = layer * nshift;
356 * Read a given sector
360 s_read(struct indir *ip, off_t offset)
367 printf("s_read(%jd)\n", (intmax_t)offset);
369 for (cip = ip; cip != NULL;) {
371 idx = (offset >> cip->shift) & NMASK;
372 up = cip->array[idx];
373 cip = (struct indir *)up;
376 idx = offset & NMASK;
377 return (cip->array[idx]);
383 * Write a given sector, prune the tree if the value is 0
387 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
389 struct indir *cip, *lip[10];
394 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
401 idx = (offset >> cip->shift) & NMASK;
402 up = cip->array[idx];
404 cip = (struct indir *)up;
407 /* Allocate branch */
409 (uintptr_t)new_indir(cip->shift - nshift);
410 if (cip->array[idx] == 0)
413 up = cip->array[idx];
414 cip = (struct indir *)up;
418 idx = offset & NMASK;
419 up = cip->array[idx];
422 cip->array[idx] = ptr;
427 if (cip->used != 0 || li == 1)
430 while (cip->used == 0 && cip != ip) {
432 idx = (offset >> lip[li]->shift) & NMASK;
433 up = lip[li]->array[idx];
434 KASSERT(up == (uintptr_t)cip, ("md screwed up"));
436 lip[li]->array[idx] = 0;
445 g_md_access(struct g_provider *pp, int r, int w, int e)
449 sc = pp->geom->softc;
451 if (r <= 0 && w <= 0 && e <= 0)
458 if ((sc->flags & MD_READONLY) != 0 && w > 0)
460 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
462 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
469 g_md_start(struct bio *bp)
473 sc = bp->bio_to->geom->softc;
474 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
475 mtx_lock(&sc->stat_mtx);
476 devstat_start_transaction_bio(sc->devstat, bp);
477 mtx_unlock(&sc->stat_mtx);
479 mtx_lock(&sc->queue_mtx);
480 bioq_disksort(&sc->bio_queue, bp);
481 mtx_unlock(&sc->queue_mtx);
485 #define MD_MALLOC_MOVE_ZERO 1
486 #define MD_MALLOC_MOVE_FILL 2
487 #define MD_MALLOC_MOVE_READ 3
488 #define MD_MALLOC_MOVE_WRITE 4
489 #define MD_MALLOC_MOVE_CMP 5
492 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize,
493 void *ptr, u_char fill, int op)
500 int error, i, ma_offs1, sz, first_read;
505 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */
513 for (n = sectorsize; n != 0; n -= sz) {
514 sz = imin(PAGE_SIZE - *ma_offs, n);
519 sf = sf_buf_alloc(m, SFB_CPUPRIVATE |
520 (md_malloc_wait ? 0 : SFB_NOWAIT));
526 p = (char *)sf_buf_kva(sf) + *ma_offs;
528 case MD_MALLOC_MOVE_ZERO:
531 case MD_MALLOC_MOVE_FILL:
534 case MD_MALLOC_MOVE_READ:
536 cpu_flush_dcache(p, sz);
538 case MD_MALLOC_MOVE_WRITE:
541 case MD_MALLOC_MOVE_CMP:
542 for (i = 0; i < sz; i++, p++) {
547 } else if (*p != first) {
554 KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op));
560 *ma_offs %= PAGE_SIZE;
563 ptr = (char *)ptr + sz;
569 if (op == MD_MALLOC_MOVE_CMP && error != 0) {
577 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs,
578 unsigned len, void *ptr, u_char fill, int op)
580 bus_dma_segment_t *vlist;
581 uint8_t *p, *end, first;
583 int ma_offs, seg_len;
589 for (; len != 0; len -= seg_len) {
590 seg_len = imin(vlist->ds_len - ma_offs, len);
591 p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs;
593 case MD_MALLOC_MOVE_ZERO:
596 case MD_MALLOC_MOVE_FILL:
597 memset(p, fill, seg_len);
599 case MD_MALLOC_MOVE_READ:
600 bcopy(ptr, p, seg_len);
601 cpu_flush_dcache(p, seg_len);
603 case MD_MALLOC_MOVE_WRITE:
604 bcopy(p, ptr, seg_len);
606 case MD_MALLOC_MOVE_CMP:
609 /* Confirm all following bytes match the first */
616 KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op));
621 if (ma_offs == vlist->ds_len) {
625 ptr = (uint8_t *)ptr + seg_len;
634 mdstart_malloc(struct md_s *sc, struct bio *bp)
638 bus_dma_segment_t *vlist;
639 int i, error, error1, ma_offs, notmapped;
640 off_t secno, nsec, uc;
643 switch (bp->bio_cmd) {
652 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0;
653 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
654 (bus_dma_segment_t *)bp->bio_data : NULL;
657 ma_offs = bp->bio_ma_offset;
659 KASSERT(vlist == NULL, ("vlists cannot be unmapped"));
660 } else if (vlist != NULL) {
661 ma_offs = bp->bio_ma_offset;
667 nsec = bp->bio_length / sc->sectorsize;
668 secno = bp->bio_offset / sc->sectorsize;
671 osp = s_read(sc->indir, secno);
672 if (bp->bio_cmd == BIO_DELETE) {
674 error = s_write(sc->indir, secno, 0);
675 } else if (bp->bio_cmd == BIO_READ) {
678 error = md_malloc_move_ma(&m, &ma_offs,
679 sc->sectorsize, NULL, 0,
680 MD_MALLOC_MOVE_ZERO);
681 } else if (vlist != NULL) {
682 error = md_malloc_move_vlist(&vlist,
683 &ma_offs, sc->sectorsize, NULL, 0,
684 MD_MALLOC_MOVE_ZERO);
686 bzero(dst, sc->sectorsize);
687 } else if (osp <= 255) {
689 error = md_malloc_move_ma(&m, &ma_offs,
690 sc->sectorsize, NULL, osp,
691 MD_MALLOC_MOVE_FILL);
692 } else if (vlist != NULL) {
693 error = md_malloc_move_vlist(&vlist,
694 &ma_offs, sc->sectorsize, NULL, osp,
695 MD_MALLOC_MOVE_FILL);
697 memset(dst, osp, sc->sectorsize);
700 error = md_malloc_move_ma(&m, &ma_offs,
701 sc->sectorsize, (void *)osp, 0,
702 MD_MALLOC_MOVE_READ);
703 } else if (vlist != NULL) {
704 error = md_malloc_move_vlist(&vlist,
705 &ma_offs, sc->sectorsize,
707 MD_MALLOC_MOVE_READ);
709 bcopy((void *)osp, dst, sc->sectorsize);
710 cpu_flush_dcache(dst, sc->sectorsize);
714 } else if (bp->bio_cmd == BIO_WRITE) {
715 if (sc->flags & MD_COMPRESS) {
717 error1 = md_malloc_move_ma(&m, &ma_offs,
718 sc->sectorsize, &uc, 0,
720 i = error1 == 0 ? sc->sectorsize : 0;
721 } else if (vlist != NULL) {
722 error1 = md_malloc_move_vlist(&vlist,
723 &ma_offs, sc->sectorsize, &uc, 0,
725 i = error1 == 0 ? sc->sectorsize : 0;
728 for (i = 1; i < sc->sectorsize; i++) {
737 if (i == sc->sectorsize) {
739 error = s_write(sc->indir, secno, uc);
742 sp = (uintptr_t)uma_zalloc(sc->uma,
743 md_malloc_wait ? M_WAITOK :
750 error = md_malloc_move_ma(&m,
751 &ma_offs, sc->sectorsize,
753 MD_MALLOC_MOVE_WRITE);
754 } else if (vlist != NULL) {
755 error = md_malloc_move_vlist(
757 sc->sectorsize, (void *)sp,
758 0, MD_MALLOC_MOVE_WRITE);
760 bcopy(dst, (void *)sp,
763 error = s_write(sc->indir, secno, sp);
766 error = md_malloc_move_ma(&m,
767 &ma_offs, sc->sectorsize,
769 MD_MALLOC_MOVE_WRITE);
770 } else if (vlist != NULL) {
771 error = md_malloc_move_vlist(
773 sc->sectorsize, (void *)osp,
774 0, MD_MALLOC_MOVE_WRITE);
776 bcopy(dst, (void *)osp,
786 uma_zfree(sc->uma, (void*)osp);
790 if (!notmapped && vlist == NULL)
791 dst += sc->sectorsize;
798 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len)
802 while (offset >= vlist->ds_len) {
803 offset -= vlist->ds_len;
808 seg_len = omin(len, vlist->ds_len - offset);
809 bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset),
812 src = (uint8_t *)src + seg_len;
819 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len)
823 while (offset >= vlist->ds_len) {
824 offset -= vlist->ds_len;
829 seg_len = omin(len, vlist->ds_len - offset);
830 bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst,
833 dst = (uint8_t *)dst + seg_len;
840 mdstart_preload(struct md_s *sc, struct bio *bp)
844 p = sc->pl_ptr + bp->bio_offset;
845 switch (bp->bio_cmd) {
847 if ((bp->bio_flags & BIO_VLIST) != 0) {
848 mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data,
849 bp->bio_ma_offset, bp->bio_length);
851 bcopy(p, bp->bio_data, bp->bio_length);
853 cpu_flush_dcache(bp->bio_data, bp->bio_length);
856 if ((bp->bio_flags & BIO_VLIST) != 0) {
857 mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data,
858 bp->bio_ma_offset, p, bp->bio_length);
860 bcopy(bp->bio_data, p, bp->bio_length);
869 mdstart_vnode(struct md_s *sc, struct bio *bp)
878 bus_dma_segment_t *vlist;
880 off_t iolen, len, zerosize;
883 switch (bp->bio_cmd) {
885 auio.uio_rw = UIO_READ;
889 auio.uio_rw = UIO_WRITE;
901 ma_offs = bp->bio_ma_offset;
902 len = bp->bio_length;
907 * If an error occurs, we set BIO_ERROR but we do not set
908 * B_INVAL because (for a write anyway), the buffer is
912 if (bp->bio_cmd == BIO_FLUSH) {
913 (void) vn_start_write(vp, &mp, V_WAIT);
914 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
915 error = VOP_FSYNC(vp, MNT_WAIT, td);
917 vn_finished_write(mp);
921 auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
922 auio.uio_resid = bp->bio_length;
923 auio.uio_segflg = UIO_SYSSPACE;
926 if (bp->bio_cmd == BIO_DELETE) {
928 * Emulate BIO_DELETE by writing zeros.
930 zerosize = ZERO_REGION_SIZE -
931 (ZERO_REGION_SIZE % sc->sectorsize);
932 auio.uio_iovcnt = howmany(bp->bio_length, zerosize);
933 piov = malloc(sizeof(*piov) * auio.uio_iovcnt, M_MD, M_WAITOK);
936 piov->iov_base = __DECONST(void *, zero_region);
939 piov->iov_len = zerosize;
940 len -= piov->iov_len;
944 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
945 piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK);
947 vlist = (bus_dma_segment_t *)bp->bio_data;
949 piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr +
951 piov->iov_len = vlist->ds_len - ma_offs;
952 if (piov->iov_len > len)
954 len -= piov->iov_len;
959 auio.uio_iovcnt = piov - auio.uio_iov;
961 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
962 pb = getpbuf(&md_vnode_pbuf_freecnt);
965 npages = atop(min(MAXPHYS, round_page(len + (ma_offs &
967 iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len);
968 KASSERT(iolen > 0, ("zero iolen"));
969 pmap_qenter((vm_offset_t)pb->b_data,
970 &bp->bio_ma[atop(ma_offs)], npages);
971 aiov.iov_base = (void *)((vm_offset_t)pb->b_data +
972 (ma_offs & PAGE_MASK));
973 aiov.iov_len = iolen;
974 auio.uio_iov = &aiov;
976 auio.uio_resid = iolen;
978 aiov.iov_base = bp->bio_data;
979 aiov.iov_len = bp->bio_length;
980 auio.uio_iov = &aiov;
984 * When reading set IO_DIRECT to try to avoid double-caching
985 * the data. When writing IO_DIRECT is not optimal.
987 if (auio.uio_rw == UIO_READ) {
988 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
989 error = VOP_READ(vp, &auio, IO_DIRECT, 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;
1003 pmap_qremove((vm_offset_t)pb->b_data, npages);
1006 bp->bio_resid -= iolen;
1011 relpbuf(pb, &md_vnode_pbuf_freecnt);
1016 bp->bio_resid = auio.uio_resid;
1021 md_swap_page_free(vm_page_t m)
1031 mdstart_swap(struct md_s *sc, struct bio *bp)
1035 vm_pindex_t i, lastp;
1036 bus_dma_segment_t *vlist;
1037 int rv, ma_offs, offs, len, lastend;
1039 switch (bp->bio_cmd) {
1045 return (EOPNOTSUPP);
1049 ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ?
1050 bp->bio_ma_offset : 0;
1051 vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
1052 (bus_dma_segment_t *)bp->bio_data : NULL;
1055 * offs is the offset at which to start operating on the
1056 * next (ie, first) page. lastp is the last page on
1057 * which we're going to operate. lastend is the ending
1058 * position within that last page (ie, PAGE_SIZE if
1059 * we're operating on complete aligned pages).
1061 offs = bp->bio_offset % PAGE_SIZE;
1062 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
1063 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
1066 VM_OBJECT_WLOCK(sc->object);
1067 vm_object_pip_add(sc->object, 1);
1068 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
1069 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
1070 m = vm_page_grab(sc->object, i, VM_ALLOC_SYSTEM);
1071 if (bp->bio_cmd == BIO_READ) {
1072 if (m->valid == VM_PAGE_BITS_ALL)
1075 rv = vm_pager_get_pages(sc->object, &m, 1,
1077 if (rv == VM_PAGER_ERROR) {
1078 md_swap_page_free(m);
1080 } else if (rv == VM_PAGER_FAIL) {
1082 * Pager does not have the page. Zero
1083 * the allocated page, and mark it as
1084 * valid. Do not set dirty, the page
1085 * can be recreated if thrown out.
1088 m->valid = VM_PAGE_BITS_ALL;
1090 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1091 pmap_copy_pages(&m, offs, bp->bio_ma,
1093 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1094 physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs,
1095 vlist, ma_offs, len);
1096 cpu_flush_dcache(p, len);
1098 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1099 cpu_flush_dcache(p, len);
1101 } else if (bp->bio_cmd == BIO_WRITE) {
1102 if (len == PAGE_SIZE || m->valid == VM_PAGE_BITS_ALL)
1105 rv = vm_pager_get_pages(sc->object, &m, 1,
1107 if (rv == VM_PAGER_ERROR) {
1108 md_swap_page_free(m);
1110 } else if (rv == VM_PAGER_FAIL)
1113 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1114 pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1116 } else if ((bp->bio_flags & BIO_VLIST) != 0) {
1117 physcopyin_vlist(vlist, ma_offs,
1118 VM_PAGE_TO_PHYS(m) + offs, len);
1120 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1123 m->valid = VM_PAGE_BITS_ALL;
1124 if (m->dirty != VM_PAGE_BITS_ALL) {
1126 vm_pager_page_unswapped(m);
1128 } else if (bp->bio_cmd == BIO_DELETE) {
1129 if (len == PAGE_SIZE || m->valid == VM_PAGE_BITS_ALL)
1132 rv = vm_pager_get_pages(sc->object, &m, 1,
1134 if (rv == VM_PAGER_ERROR) {
1135 md_swap_page_free(m);
1137 } else if (rv == VM_PAGER_FAIL) {
1138 md_swap_page_free(m);
1141 /* Page is valid. */
1142 if (len != PAGE_SIZE) {
1143 pmap_zero_page_area(m, offs, len);
1144 if (m->dirty != VM_PAGE_BITS_ALL) {
1146 vm_pager_page_unswapped(m);
1149 vm_pager_page_unswapped(m);
1150 md_swap_page_free(m);
1158 if (vm_page_active(m))
1159 vm_page_reference(m);
1161 vm_page_activate(m);
1165 /* Actions on further pages start at offset 0 */
1166 p += PAGE_SIZE - offs;
1170 vm_object_pip_wakeup(sc->object);
1171 VM_OBJECT_WUNLOCK(sc->object);
1172 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
1176 mdstart_null(struct md_s *sc, struct bio *bp)
1179 switch (bp->bio_cmd) {
1181 bzero(bp->bio_data, bp->bio_length);
1182 cpu_flush_dcache(bp->bio_data, bp->bio_length);
1192 md_kthread(void *arg)
1199 thread_lock(curthread);
1200 sched_prio(curthread, PRIBIO);
1201 thread_unlock(curthread);
1202 if (sc->type == MD_VNODE)
1203 curthread->td_pflags |= TDP_NORUNNINGBUF;
1206 mtx_lock(&sc->queue_mtx);
1207 if (sc->flags & MD_SHUTDOWN) {
1208 sc->flags |= MD_EXITING;
1209 mtx_unlock(&sc->queue_mtx);
1212 bp = bioq_takefirst(&sc->bio_queue);
1214 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1217 mtx_unlock(&sc->queue_mtx);
1218 if (bp->bio_cmd == BIO_GETATTR) {
1219 int isv = ((sc->flags & MD_VERIFY) != 0);
1221 if ((sc->fwsectors && sc->fwheads &&
1222 (g_handleattr_int(bp, "GEOM::fwsectors",
1224 g_handleattr_int(bp, "GEOM::fwheads",
1226 g_handleattr_int(bp, "GEOM::candelete", 1))
1228 else if (sc->ident[0] != '\0' &&
1229 g_handleattr_str(bp, "GEOM::ident", sc->ident))
1231 else if (g_handleattr_int(bp, "MNT::verified", isv))
1236 error = sc->start(sc, bp);
1240 bp->bio_completed = bp->bio_length;
1241 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE))
1242 devstat_end_transaction_bio(sc->devstat, bp);
1243 g_io_deliver(bp, error);
1248 static struct md_s *
1253 LIST_FOREACH(sc, &md_softc_list, list) {
1254 if (sc->unit == unit)
1260 static struct md_s *
1261 mdnew(int unit, int *errp, enum md_types type)
1268 unit = alloc_unr(md_uh);
1270 unit = alloc_unr_specific(md_uh, unit);
1277 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
1279 bioq_init(&sc->bio_queue);
1280 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1281 mtx_init(&sc->stat_mtx, "md stat", NULL, MTX_DEF);
1283 sprintf(sc->name, "md%d", unit);
1284 LIST_INSERT_HEAD(&md_softc_list, sc, list);
1285 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1288 LIST_REMOVE(sc, list);
1289 mtx_destroy(&sc->stat_mtx);
1290 mtx_destroy(&sc->queue_mtx);
1291 free_unr(md_uh, sc->unit);
1298 mdinit(struct md_s *sc)
1301 struct g_provider *pp;
1304 gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1306 pp = g_new_providerf(gp, "md%d", sc->unit);
1307 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1308 pp->mediasize = sc->mediasize;
1309 pp->sectorsize = sc->sectorsize;
1314 pp->flags |= G_PF_ACCEPT_UNMAPPED;
1322 g_error_provider(pp, 0);
1323 g_topology_unlock();
1324 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1325 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
1329 mdcreate_malloc(struct md_s *sc, struct md_req *mdr)
1336 if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1338 if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize))
1340 /* Compression doesn't make sense if we have reserved space */
1341 if (mdr->md_options & MD_RESERVE)
1342 mdr->md_options &= ~MD_COMPRESS;
1343 if (mdr->md_fwsectors != 0)
1344 sc->fwsectors = mdr->md_fwsectors;
1345 if (mdr->md_fwheads != 0)
1346 sc->fwheads = mdr->md_fwheads;
1347 sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE);
1348 sc->indir = dimension(sc->mediasize / sc->sectorsize);
1349 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
1351 if (mdr->md_options & MD_RESERVE) {
1354 nsectors = sc->mediasize / sc->sectorsize;
1355 for (u = 0; u < nsectors; u++) {
1356 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ?
1357 M_WAITOK : M_NOWAIT) | M_ZERO);
1359 error = s_write(sc->indir, u, sp);
1371 mdsetcred(struct md_s *sc, struct ucred *cred)
1377 * Set credits in our softc
1382 sc->cred = crhold(cred);
1385 * Horrible kludge to establish credentials for NFS XXX.
1392 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1393 bzero(&auio, sizeof(auio));
1395 aiov.iov_base = tmpbuf;
1396 aiov.iov_len = sc->sectorsize;
1397 auio.uio_iov = &aiov;
1398 auio.uio_iovcnt = 1;
1399 auio.uio_offset = 0;
1400 auio.uio_rw = UIO_READ;
1401 auio.uio_segflg = UIO_SYSSPACE;
1402 auio.uio_resid = aiov.iov_len;
1403 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1404 error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
1405 VOP_UNLOCK(sc->vnode, 0);
1406 free(tmpbuf, M_TEMP);
1412 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1415 struct nameidata nd;
1419 fname = mdr->md_file;
1420 if (mdr->md_file_seg == UIO_USERSPACE) {
1421 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
1424 } else if (mdr->md_file_seg == UIO_SYSSPACE)
1425 strlcpy(sc->file, fname, sizeof(sc->file));
1430 * If the user specified that this is a read only device, don't
1431 * set the FWRITE mask before trying to open the backing store.
1433 flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \
1434 | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0);
1435 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td);
1436 error = vn_open(&nd, &flags, 0, NULL);
1439 NDFREE(&nd, NDF_ONLY_PNBUF);
1440 if (nd.ni_vp->v_type != VREG) {
1444 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1447 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1448 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1449 if (nd.ni_vp->v_iflag & VI_DOOMED) {
1450 /* Forced unmount. */
1455 nd.ni_vp->v_vflag |= VV_MD;
1456 VOP_UNLOCK(nd.ni_vp, 0);
1458 if (mdr->md_fwsectors != 0)
1459 sc->fwsectors = mdr->md_fwsectors;
1460 if (mdr->md_fwheads != 0)
1461 sc->fwheads = mdr->md_fwheads;
1462 snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1463 (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1464 sc->flags = mdr->md_options & (MD_FORCE | MD_ASYNC | MD_VERIFY);
1465 if (!(flags & FWRITE))
1466 sc->flags |= MD_READONLY;
1467 sc->vnode = nd.ni_vp;
1469 error = mdsetcred(sc, td->td_ucred);
1472 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1473 nd.ni_vp->v_vflag &= ~VV_MD;
1478 VOP_UNLOCK(nd.ni_vp, 0);
1479 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1484 mddestroy(struct md_s *sc, struct thread *td)
1488 sc->gp->softc = NULL;
1490 g_wither_geom(sc->gp, ENXIO);
1491 g_topology_unlock();
1496 devstat_remove_entry(sc->devstat);
1499 mtx_lock(&sc->queue_mtx);
1500 sc->flags |= MD_SHUTDOWN;
1502 while (!(sc->flags & MD_EXITING))
1503 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1504 mtx_unlock(&sc->queue_mtx);
1505 mtx_destroy(&sc->stat_mtx);
1506 mtx_destroy(&sc->queue_mtx);
1507 if (sc->vnode != NULL) {
1508 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1509 sc->vnode->v_vflag &= ~VV_MD;
1510 VOP_UNLOCK(sc->vnode, 0);
1511 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1512 FREAD : (FREAD|FWRITE), sc->cred, td);
1514 if (sc->cred != NULL)
1516 if (sc->object != NULL)
1517 vm_object_deallocate(sc->object);
1519 destroy_indir(sc, sc->indir);
1521 uma_zdestroy(sc->uma);
1523 LIST_REMOVE(sc, list);
1524 free_unr(md_uh, sc->unit);
1530 mdresize(struct md_s *sc, struct md_req *mdr)
1533 vm_pindex_t oldpages, newpages;
1540 if (mdr->md_mediasize <= 0 ||
1541 (mdr->md_mediasize % PAGE_SIZE) != 0)
1543 oldpages = OFF_TO_IDX(round_page(sc->mediasize));
1544 newpages = OFF_TO_IDX(round_page(mdr->md_mediasize));
1545 if (newpages < oldpages) {
1546 VM_OBJECT_WLOCK(sc->object);
1547 vm_object_page_remove(sc->object, newpages, 0, 0);
1548 swap_pager_freespace(sc->object, newpages,
1549 oldpages - newpages);
1550 swap_release_by_cred(IDX_TO_OFF(oldpages -
1551 newpages), sc->cred);
1552 sc->object->charge = IDX_TO_OFF(newpages);
1553 sc->object->size = newpages;
1554 VM_OBJECT_WUNLOCK(sc->object);
1555 } else if (newpages > oldpages) {
1556 res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1557 oldpages), sc->cred);
1560 if ((mdr->md_options & MD_RESERVE) ||
1561 (sc->flags & MD_RESERVE)) {
1562 error = swap_pager_reserve(sc->object,
1563 oldpages, newpages - oldpages);
1565 swap_release_by_cred(
1566 IDX_TO_OFF(newpages - oldpages),
1571 VM_OBJECT_WLOCK(sc->object);
1572 sc->object->charge = IDX_TO_OFF(newpages);
1573 sc->object->size = newpages;
1574 VM_OBJECT_WUNLOCK(sc->object);
1578 return (EOPNOTSUPP);
1581 sc->mediasize = mdr->md_mediasize;
1583 g_resize_provider(sc->pp, sc->mediasize);
1584 g_topology_unlock();
1589 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1595 * Range check. Disallow negative sizes and sizes not being
1596 * multiple of page size.
1598 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1602 * Allocate an OBJT_SWAP object.
1604 * Note the truncation.
1607 if ((mdr->md_options & MD_VERIFY) != 0)
1609 npage = mdr->md_mediasize / PAGE_SIZE;
1610 if (mdr->md_fwsectors != 0)
1611 sc->fwsectors = mdr->md_fwsectors;
1612 if (mdr->md_fwheads != 0)
1613 sc->fwheads = mdr->md_fwheads;
1614 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1615 VM_PROT_DEFAULT, 0, td->td_ucred);
1616 if (sc->object == NULL)
1618 sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1619 if (mdr->md_options & MD_RESERVE) {
1620 if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1625 error = mdsetcred(sc, td->td_ucred);
1628 vm_object_deallocate(sc->object);
1635 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1639 * Range check. Disallow negative sizes and sizes not being
1640 * multiple of page size.
1642 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1649 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1655 sx_assert(&md_sx, SA_XLOCKED);
1657 switch (mdr->md_type) {
1667 if (mdr->md_sectorsize == 0)
1668 sectsize = DEV_BSIZE;
1670 sectsize = mdr->md_sectorsize;
1671 if (sectsize > MAXPHYS || mdr->md_mediasize < sectsize)
1673 if (mdr->md_options & MD_AUTOUNIT)
1674 sc = mdnew(-1, &error, mdr->md_type);
1676 if (mdr->md_unit > INT_MAX)
1678 sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1682 if (mdr->md_label != NULL)
1683 error = copyinstr(mdr->md_label, sc->label,
1684 sizeof(sc->label), NULL);
1687 if (mdr->md_options & MD_AUTOUNIT)
1688 mdr->md_unit = sc->unit;
1689 sc->mediasize = mdr->md_mediasize;
1690 sc->sectorsize = sectsize;
1694 sc->start = mdstart_malloc;
1695 error = mdcreate_malloc(sc, mdr);
1699 * We disallow attaching preloaded memory disks via
1700 * ioctl. Preloaded memory disks are automatically
1701 * attached in g_md_init().
1706 sc->start = mdstart_vnode;
1707 error = mdcreate_vnode(sc, mdr, td);
1710 sc->start = mdstart_swap;
1711 error = mdcreate_swap(sc, mdr, td);
1714 sc->start = mdstart_null;
1715 error = mdcreate_null(sc, mdr, td);
1724 /* Prune off any residual fractional sector */
1725 i = sc->mediasize % sc->sectorsize;
1733 kern_mdattach(struct thread *td, struct md_req *mdr)
1738 error = kern_mdattach_locked(td, mdr);
1744 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1748 sx_assert(&md_sx, SA_XLOCKED);
1750 if (mdr->md_mediasize != 0 ||
1751 (mdr->md_options & ~MD_FORCE) != 0)
1754 sc = mdfind(mdr->md_unit);
1757 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1758 !(mdr->md_options & MD_FORCE))
1760 return (mddestroy(sc, td));
1764 kern_mddetach(struct thread *td, struct md_req *mdr)
1769 error = kern_mddetach_locked(td, mdr);
1775 kern_mdresize_locked(struct md_req *mdr)
1779 sx_assert(&md_sx, SA_XLOCKED);
1781 if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1784 sc = mdfind(mdr->md_unit);
1787 if (mdr->md_mediasize < sc->sectorsize)
1789 if (mdr->md_mediasize < sc->mediasize &&
1790 !(sc->flags & MD_FORCE) &&
1791 !(mdr->md_options & MD_FORCE))
1793 return (mdresize(sc, mdr));
1797 kern_mdresize(struct md_req *mdr)
1802 error = kern_mdresize_locked(mdr);
1808 kern_mdquery_locked(struct md_req *mdr)
1813 sx_assert(&md_sx, SA_XLOCKED);
1815 sc = mdfind(mdr->md_unit);
1818 mdr->md_type = sc->type;
1819 mdr->md_options = sc->flags;
1820 mdr->md_mediasize = sc->mediasize;
1821 mdr->md_sectorsize = sc->sectorsize;
1823 if (mdr->md_label != NULL) {
1824 error = copyout(sc->label, mdr->md_label,
1825 strlen(sc->label) + 1);
1829 if (sc->type == MD_VNODE ||
1830 (sc->type == MD_PRELOAD && mdr->md_file != NULL))
1831 error = copyout(sc->file, mdr->md_file,
1832 strlen(sc->file) + 1);
1837 kern_mdquery(struct md_req *mdr)
1842 error = kern_mdquery_locked(mdr);
1848 kern_mdlist_locked(struct md_req *mdr)
1853 sx_assert(&md_sx, SA_XLOCKED);
1856 * Write the number of md devices to mdr->md_units[0].
1857 * Write the unit number of the first (mdr->md_units_nitems - 2)
1858 * units to mdr->md_units[1::(mdr->md_units - 2)] and terminate the
1861 * XXX: There is currently no mechanism to retrieve unit
1862 * numbers for more than (MDNPAD - 2) units.
1864 * XXX: Due to the use of LIST_INSERT_HEAD in mdnew(), the
1865 * list of visible unit numbers not stable.
1868 LIST_FOREACH(sc, &md_softc_list, list) {
1869 if (i < mdr->md_units_nitems - 1)
1870 mdr->md_units[i] = sc->unit;
1873 mdr->md_units[MIN(i, mdr->md_units_nitems - 1)] = -1;
1874 mdr->md_units[0] = i - 1;
1879 kern_mdlist(struct md_req *mdr)
1884 error = kern_mdlist_locked(mdr);
1889 /* Copy members that are not userspace pointers. */
1890 #define MD_IOCTL2REQ(mdio, mdr) do { \
1891 (mdr)->md_unit = (mdio)->md_unit; \
1892 (mdr)->md_type = (mdio)->md_type; \
1893 (mdr)->md_mediasize = (mdio)->md_mediasize; \
1894 (mdr)->md_sectorsize = (mdio)->md_sectorsize; \
1895 (mdr)->md_options = (mdio)->md_options; \
1896 (mdr)->md_fwheads = (mdio)->md_fwheads; \
1897 (mdr)->md_fwsectors = (mdio)->md_fwsectors; \
1898 (mdr)->md_units = &(mdio)->md_pad[0]; \
1899 (mdr)->md_units_nitems = nitems((mdio)->md_pad); \
1902 /* Copy members that might have been updated */
1903 #define MD_REQ2IOCTL(mdr, mdio) do { \
1904 (mdio)->md_unit = (mdr)->md_unit; \
1905 (mdio)->md_type = (mdr)->md_type; \
1906 (mdio)->md_mediasize = (mdr)->md_mediasize; \
1907 (mdio)->md_sectorsize = (mdr)->md_sectorsize; \
1908 (mdio)->md_options = (mdr)->md_options; \
1909 (mdio)->md_fwheads = (mdr)->md_fwheads; \
1910 (mdio)->md_fwsectors = (mdr)->md_fwsectors; \
1914 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1921 printf("mdctlioctl(%s %lx %p %x %p)\n",
1922 devtoname(dev), cmd, addr, flags, td);
1924 bzero(&mdr, sizeof(mdr));
1931 struct md_ioctl *mdio = (struct md_ioctl *)addr;
1932 if (mdio->md_version != MDIOVERSION)
1934 MD_IOCTL2REQ(mdio, &mdr);
1935 mdr.md_file = mdio->md_file;
1936 mdr.md_file_seg = UIO_USERSPACE;
1937 /* If the file is adjacent to the md_ioctl it's in kernel. */
1938 if ((void *)mdio->md_file == (void *)(mdio + 1))
1939 mdr.md_file_seg = UIO_SYSSPACE;
1940 mdr.md_label = mdio->md_label;
1943 #ifdef COMPAT_FREEBSD32
1944 case MDIOCATTACH_32:
1945 case MDIOCDETACH_32:
1946 case MDIOCRESIZE_32:
1948 case MDIOCLIST_32: {
1949 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1950 if (mdio->md_version != MDIOVERSION)
1952 MD_IOCTL2REQ(mdio, &mdr);
1953 mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1954 mdr.md_file_seg = UIO_USERSPACE;
1955 mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1960 /* Fall through to handler switch. */
1967 #ifdef COMPAT_FREEBSD32
1968 case MDIOCATTACH_32:
1970 error = kern_mdattach(td, &mdr);
1973 #ifdef COMPAT_FREEBSD32
1974 case MDIOCDETACH_32:
1976 error = kern_mddetach(td, &mdr);
1979 #ifdef COMPAT_FREEBSD32
1980 case MDIOCRESIZE_32:
1982 error = kern_mdresize(&mdr);
1985 #ifdef COMPAT_FREEBSD32
1988 error = kern_mdquery(&mdr);
1991 #ifdef COMPAT_FREEBSD32
1994 error = kern_mdlist(&mdr);
2003 struct md_ioctl *mdio = (struct md_ioctl *)addr;
2004 MD_REQ2IOCTL(&mdr, mdio);
2007 #ifdef COMPAT_FREEBSD32
2008 case MDIOCATTACH_32:
2009 case MDIOCQUERY_32: {
2010 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
2011 MD_REQ2IOCTL(&mdr, mdio);
2016 /* Other commands to not alter mdr. */
2024 md_preloaded(u_char *image, size_t length, const char *name)
2029 sc = mdnew(-1, &error, MD_PRELOAD);
2032 sc->mediasize = length;
2033 sc->sectorsize = DEV_BSIZE;
2035 sc->pl_len = length;
2036 sc->start = mdstart_preload;
2038 strlcpy(sc->file, name, sizeof(sc->file));
2040 if (sc->unit == 0) {
2042 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2044 #ifdef MD_ROOT_READONLY
2045 sc->flags |= MD_READONLY;
2051 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2052 MD_NAME, sc->unit, name, length, image);
2054 printf("%s%d: Embedded image %zd bytes at %p\n",
2055 MD_NAME, sc->unit, length, image);
2060 g_md_init(struct g_class *mp __unused)
2063 u_char *ptr, *name, *type;
2067 /* figure out log2(NINDIR) */
2068 for (i = NINDIR, nshift = -1; i; nshift++)
2072 sx_init(&md_sx, "MD config lock");
2073 g_topology_unlock();
2074 md_uh = new_unrhdr(0, INT_MAX, NULL);
2076 if (mfs_root_size != 0) {
2078 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2083 /* XXX: are preload_* static or do they need Giant ? */
2084 while ((mod = preload_search_next_name(mod)) != NULL) {
2085 name = (char *)preload_search_info(mod, MODINFO_NAME);
2088 type = (char *)preload_search_info(mod, MODINFO_TYPE);
2091 if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2093 ptr = preload_fetch_addr(mod);
2094 len = preload_fetch_size(mod);
2095 if (ptr != NULL && len != 0) {
2097 md_preloaded(ptr, len, name);
2101 md_vnode_pbuf_freecnt = nswbuf / 10;
2102 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2108 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2109 struct g_consumer *cp __unused, struct g_provider *pp)
2140 if (indent == NULL) {
2141 sbuf_printf(sb, " u %d", mp->unit);
2142 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2143 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2144 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2145 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2146 sbuf_printf(sb, " t %s", type);
2147 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2148 (mp->type == MD_PRELOAD && mp->file[0] != '\0'))
2149 sbuf_printf(sb, " file %s", mp->file);
2150 sbuf_printf(sb, " label %s", mp->label);
2152 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2154 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2155 indent, (uintmax_t) mp->sectorsize);
2156 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2157 indent, (uintmax_t) mp->fwheads);
2158 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2159 indent, (uintmax_t) mp->fwsectors);
2160 if (mp->ident[0] != '\0') {
2161 sbuf_printf(sb, "%s<ident>", indent);
2162 g_conf_printf_escaped(sb, "%s", mp->ident);
2163 sbuf_printf(sb, "</ident>\n");
2165 sbuf_printf(sb, "%s<length>%ju</length>\n",
2166 indent, (uintmax_t) mp->mediasize);
2167 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2168 (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2169 sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2170 (mp->flags & MD_READONLY) == 0 ? "read-write":
2172 sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2174 if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2175 (mp->type == MD_PRELOAD && mp->file[0] != '\0')) {
2176 sbuf_printf(sb, "%s<file>", indent);
2177 g_conf_printf_escaped(sb, "%s", mp->file);
2178 sbuf_printf(sb, "</file>\n");
2180 sbuf_printf(sb, "%s<label>", indent);
2181 g_conf_printf_escaped(sb, "%s", mp->label);
2182 sbuf_printf(sb, "</label>\n");
2188 g_md_fini(struct g_class *mp __unused)
2192 if (status_dev != NULL)
2193 destroy_dev(status_dev);
2194 delete_unrhdr(md_uh);