2 * ----------------------------------------------------------------------------
3 * "THE BEER-WARE LICENSE" (Revision 42):
4 * <phk@FreeBSD.ORG> wrote this file. As long as you retain this notice you
5 * can do whatever you want with this stuff. If we meet some day, and you think
6 * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp
7 * ----------------------------------------------------------------------------
14 * The following functions are based in the vn(4) driver: mdstart_swap(),
15 * mdstart_vnode(), mdcreate_swap(), mdcreate_vnode() and mddestroy(),
16 * and as such under the following copyright:
18 * Copyright (c) 1988 University of Utah.
19 * Copyright (c) 1990, 1993
20 * The Regents of the University of California. All rights reserved.
21 * Copyright (c) 2013 The FreeBSD Foundation
22 * All rights reserved.
24 * This code is derived from software contributed to Berkeley by
25 * the Systems Programming Group of the University of Utah Computer
28 * Portions of this software were developed by Konstantin Belousov
29 * under sponsorship from the FreeBSD Foundation.
31 * Redistribution and use in source and binary forms, with or without
32 * modification, are permitted provided that the following conditions
34 * 1. Redistributions of source code must retain the above copyright
35 * notice, this list of conditions and the following disclaimer.
36 * 2. Redistributions in binary form must reproduce the above copyright
37 * notice, this list of conditions and the following disclaimer in the
38 * documentation and/or other materials provided with the distribution.
39 * 4. Neither the name of the University nor the names of its contributors
40 * may be used to endorse or promote products derived from this software
41 * without specific prior written permission.
43 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
44 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
45 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
46 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
47 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
48 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
49 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
50 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
51 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
52 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55 * from: Utah Hdr: vn.c 1.13 94/04/02
57 * from: @(#)vn.c 8.6 (Berkeley) 4/1/94
58 * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03
64 #include <sys/param.h>
65 #include <sys/systm.h>
69 #include <sys/devicestat.h>
70 #include <sys/fcntl.h>
71 #include <sys/kernel.h>
72 #include <sys/kthread.h>
73 #include <sys/limits.h>
74 #include <sys/linker.h>
76 #include <sys/malloc.h>
77 #include <sys/mdioctl.h>
78 #include <sys/mount.h>
79 #include <sys/mutex.h>
81 #include <sys/namei.h>
83 #include <sys/queue.h>
84 #include <sys/rwlock.h>
86 #include <sys/sched.h>
87 #include <sys/sf_buf.h>
88 #include <sys/sysctl.h>
89 #include <sys/vnode.h>
91 #include <geom/geom.h>
92 #include <geom/geom_int.h>
95 #include <vm/vm_param.h>
96 #include <vm/vm_object.h>
97 #include <vm/vm_page.h>
98 #include <vm/vm_pager.h>
99 #include <vm/swap_pager.h>
104 #define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */
105 #define MD_EXITING 0x20000 /* Worker thread is exiting. */
108 #define MD_NSECT (10000 * 2)
111 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk");
112 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors");
115 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0,
116 "Enable md(4) debug messages");
117 static int md_malloc_wait;
118 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0,
119 "Allow malloc to wait for memory allocations");
121 #if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE)
122 #define MD_ROOT_FSTYPE "ufs"
127 * Preloaded image gets put here.
129 #if defined(MD_ROOT_SIZE)
131 * Applications that patch the object with the image can determine
132 * the size looking at the start and end markers (strings),
133 * so we want them contiguous.
136 u_char start[MD_ROOT_SIZE*1024];
139 .start = "MFS Filesystem goes here",
140 .end = "MFS Filesystem had better STOP here",
142 const int mfs_root_size = sizeof(mfs_root.start);
144 extern volatile u_char __weak_symbol mfs_root;
145 extern volatile u_char __weak_symbol mfs_root_end;
147 __GLOBL(mfs_root_end);
148 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root))
152 static g_init_t g_md_init;
153 static g_fini_t g_md_fini;
154 static g_start_t g_md_start;
155 static g_access_t g_md_access;
156 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
157 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
159 static struct cdev *status_dev = 0;
160 static struct sx md_sx;
161 static struct unrhdr *md_uh;
163 static d_ioctl_t mdctlioctl;
165 static struct cdevsw mdctl_cdevsw = {
166 .d_version = D_VERSION,
167 .d_ioctl = mdctlioctl,
171 struct g_class g_md_class = {
173 .version = G_VERSION,
177 .access = g_md_access,
178 .dumpconf = g_md_dumpconf,
181 DECLARE_GEOM_CLASS(g_md_class, g_md);
184 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
186 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t))
187 #define NMASK (NINDIR-1)
190 static int md_vnode_pbuf_freecnt;
201 LIST_ENTRY(md_s) list;
202 struct bio_queue_head bio_queue;
203 struct mtx queue_mtx;
216 struct g_provider *pp;
217 int (*start)(struct md_s *sc, struct bio *bp);
218 struct devstat *devstat;
220 /* MD_MALLOC related fields */
224 /* MD_PRELOAD related fields */
228 /* MD_VNODE related fields */
233 /* MD_SWAP related fields */
237 static struct indir *
238 new_indir(u_int shift)
242 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
246 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
247 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
248 if (ip->array == NULL) {
258 del_indir(struct indir *ip)
261 free(ip->array, M_MDSECT);
266 destroy_indir(struct md_s *sc, struct indir *ip)
270 for (i = 0; i < NINDIR; i++) {
274 destroy_indir(sc, (struct indir*)(ip->array[i]));
275 else if (ip->array[i] > 255)
276 uma_zfree(sc->uma, (void *)(ip->array[i]));
282 * This function does the math and allocates the top level "indir" structure
283 * for a device of "size" sectors.
286 static struct indir *
287 dimension(off_t size)
295 while (rcnt > NINDIR) {
301 * XXX: the top layer is probably not fully populated, so we allocate
302 * too much space for ip->array in here.
304 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
305 ip->array = malloc(sizeof(uintptr_t) * NINDIR,
306 M_MDSECT, M_WAITOK | M_ZERO);
308 ip->shift = layer * nshift;
313 * Read a given sector
317 s_read(struct indir *ip, off_t offset)
324 printf("s_read(%jd)\n", (intmax_t)offset);
326 for (cip = ip; cip != NULL;) {
328 idx = (offset >> cip->shift) & NMASK;
329 up = cip->array[idx];
330 cip = (struct indir *)up;
333 idx = offset & NMASK;
334 return (cip->array[idx]);
340 * Write a given sector, prune the tree if the value is 0
344 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
346 struct indir *cip, *lip[10];
351 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
358 idx = (offset >> cip->shift) & NMASK;
359 up = cip->array[idx];
361 cip = (struct indir *)up;
364 /* Allocate branch */
366 (uintptr_t)new_indir(cip->shift - nshift);
367 if (cip->array[idx] == 0)
370 up = cip->array[idx];
371 cip = (struct indir *)up;
375 idx = offset & NMASK;
376 up = cip->array[idx];
379 cip->array[idx] = ptr;
384 if (cip->used != 0 || li == 1)
387 while (cip->used == 0 && cip != ip) {
389 idx = (offset >> lip[li]->shift) & NMASK;
390 up = lip[li]->array[idx];
391 KASSERT(up == (uintptr_t)cip, ("md screwed up"));
393 lip[li]->array[idx] = 0;
402 g_md_access(struct g_provider *pp, int r, int w, int e)
406 sc = pp->geom->softc;
408 if (r <= 0 && w <= 0 && e <= 0)
415 if ((sc->flags & MD_READONLY) != 0 && w > 0)
417 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
419 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
426 g_md_start(struct bio *bp)
430 sc = bp->bio_to->geom->softc;
431 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
432 mtx_lock(&sc->stat_mtx);
433 devstat_start_transaction_bio(sc->devstat, bp);
434 mtx_unlock(&sc->stat_mtx);
436 mtx_lock(&sc->queue_mtx);
437 bioq_disksort(&sc->bio_queue, bp);
438 mtx_unlock(&sc->queue_mtx);
442 #define MD_MALLOC_MOVE_ZERO 1
443 #define MD_MALLOC_MOVE_FILL 2
444 #define MD_MALLOC_MOVE_READ 3
445 #define MD_MALLOC_MOVE_WRITE 4
446 #define MD_MALLOC_MOVE_CMP 5
449 md_malloc_move(vm_page_t **mp, int *ma_offs, unsigned sectorsize,
450 void *ptr, u_char fill, int op)
457 int error, i, ma_offs1, sz, first_read;
462 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */
470 for (n = sectorsize; n != 0; n -= sz) {
471 sz = imin(PAGE_SIZE - *ma_offs, n);
476 sf = sf_buf_alloc(m, SFB_CPUPRIVATE |
477 (md_malloc_wait ? 0 : SFB_NOWAIT));
483 p = (char *)sf_buf_kva(sf) + *ma_offs;
485 case MD_MALLOC_MOVE_ZERO:
488 case MD_MALLOC_MOVE_FILL:
491 case MD_MALLOC_MOVE_READ:
493 cpu_flush_dcache(p, sz);
495 case MD_MALLOC_MOVE_WRITE:
498 case MD_MALLOC_MOVE_CMP:
499 for (i = 0; i < sz; i++, p++) {
504 } else if (*p != first) {
511 KASSERT(0, ("md_malloc_move unknown op %d\n", op));
517 *ma_offs %= PAGE_SIZE;
520 ptr = (char *)ptr + sz;
526 if (op == MD_MALLOC_MOVE_CMP && error != 0) {
534 mdstart_malloc(struct md_s *sc, struct bio *bp)
538 int i, error, error1, ma_offs, notmapped;
539 off_t secno, nsec, uc;
542 switch (bp->bio_cmd) {
551 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0;
554 ma_offs = bp->bio_ma_offset;
560 nsec = bp->bio_length / sc->sectorsize;
561 secno = bp->bio_offset / sc->sectorsize;
564 osp = s_read(sc->indir, secno);
565 if (bp->bio_cmd == BIO_DELETE) {
567 error = s_write(sc->indir, secno, 0);
568 } else if (bp->bio_cmd == BIO_READ) {
571 error = md_malloc_move(&m, &ma_offs,
572 sc->sectorsize, NULL, 0,
573 MD_MALLOC_MOVE_ZERO);
575 bzero(dst, sc->sectorsize);
576 } else if (osp <= 255) {
578 error = md_malloc_move(&m, &ma_offs,
579 sc->sectorsize, NULL, osp,
580 MD_MALLOC_MOVE_FILL);
582 memset(dst, osp, sc->sectorsize);
585 error = md_malloc_move(&m, &ma_offs,
586 sc->sectorsize, (void *)osp, 0,
587 MD_MALLOC_MOVE_READ);
589 bcopy((void *)osp, dst, sc->sectorsize);
590 cpu_flush_dcache(dst, sc->sectorsize);
594 } else if (bp->bio_cmd == BIO_WRITE) {
595 if (sc->flags & MD_COMPRESS) {
597 error1 = md_malloc_move(&m, &ma_offs,
598 sc->sectorsize, &uc, 0,
600 i = error1 == 0 ? sc->sectorsize : 0;
603 for (i = 1; i < sc->sectorsize; i++) {
612 if (i == sc->sectorsize) {
614 error = s_write(sc->indir, secno, uc);
617 sp = (uintptr_t)uma_zalloc(sc->uma,
618 md_malloc_wait ? M_WAITOK :
625 error = md_malloc_move(&m,
626 &ma_offs, sc->sectorsize,
628 MD_MALLOC_MOVE_WRITE);
630 bcopy(dst, (void *)sp,
633 error = s_write(sc->indir, secno, sp);
636 error = md_malloc_move(&m,
637 &ma_offs, sc->sectorsize,
639 MD_MALLOC_MOVE_WRITE);
641 bcopy(dst, (void *)osp,
651 uma_zfree(sc->uma, (void*)osp);
656 dst += sc->sectorsize;
663 mdstart_preload(struct md_s *sc, struct bio *bp)
666 switch (bp->bio_cmd) {
668 bcopy(sc->pl_ptr + bp->bio_offset, bp->bio_data,
670 cpu_flush_dcache(bp->bio_data, bp->bio_length);
673 bcopy(bp->bio_data, sc->pl_ptr + bp->bio_offset,
682 mdstart_vnode(struct md_s *sc, struct bio *bp)
693 switch (bp->bio_cmd) {
709 * If an error occurs, we set BIO_ERROR but we do not set
710 * B_INVAL because (for a write anyway), the buffer is
714 if (bp->bio_cmd == BIO_FLUSH) {
715 (void) vn_start_write(vp, &mp, V_WAIT);
716 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
717 error = VOP_FSYNC(vp, MNT_WAIT, td);
719 vn_finished_write(mp);
723 bzero(&auio, sizeof(auio));
726 * Special case for BIO_DELETE. On the surface, this is very
727 * similar to BIO_WRITE, except that we write from our own
728 * fixed-length buffer, so we have to loop. The net result is
729 * that the two cases end up having very little in common.
731 if (bp->bio_cmd == BIO_DELETE) {
732 zerosize = ZERO_REGION_SIZE -
733 (ZERO_REGION_SIZE % sc->sectorsize);
734 auio.uio_iov = &aiov;
736 auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
737 auio.uio_segflg = UIO_SYSSPACE;
738 auio.uio_rw = UIO_WRITE;
740 end = bp->bio_offset + bp->bio_length;
741 (void) vn_start_write(vp, &mp, V_WAIT);
742 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
744 while (auio.uio_offset < end) {
745 aiov.iov_base = __DECONST(void *, zero_region);
746 aiov.iov_len = end - auio.uio_offset;
747 if (aiov.iov_len > zerosize)
748 aiov.iov_len = zerosize;
749 auio.uio_resid = aiov.iov_len;
750 error = VOP_WRITE(vp, &auio,
751 sc->flags & MD_ASYNC ? 0 : IO_SYNC, sc->cred);
756 vn_finished_write(mp);
757 bp->bio_resid = end - auio.uio_offset;
761 if ((bp->bio_flags & BIO_UNMAPPED) == 0) {
763 aiov.iov_base = bp->bio_data;
765 KASSERT(bp->bio_length <= MAXPHYS, ("bio_length %jd",
766 (uintmax_t)bp->bio_length));
767 pb = getpbuf(&md_vnode_pbuf_freecnt);
768 pmap_qenter((vm_offset_t)pb->b_data, bp->bio_ma, bp->bio_ma_n);
769 aiov.iov_base = (void *)((vm_offset_t)pb->b_data +
772 aiov.iov_len = bp->bio_length;
773 auio.uio_iov = &aiov;
775 auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
776 auio.uio_segflg = UIO_SYSSPACE;
777 if (bp->bio_cmd == BIO_READ)
778 auio.uio_rw = UIO_READ;
779 else if (bp->bio_cmd == BIO_WRITE)
780 auio.uio_rw = UIO_WRITE;
782 panic("wrong BIO_OP in mdstart_vnode");
783 auio.uio_resid = bp->bio_length;
786 * When reading set IO_DIRECT to try to avoid double-caching
787 * the data. When writing IO_DIRECT is not optimal.
789 if (bp->bio_cmd == BIO_READ) {
790 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
791 error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred);
794 (void) vn_start_write(vp, &mp, V_WAIT);
795 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
796 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
799 vn_finished_write(mp);
801 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
802 pmap_qremove((vm_offset_t)pb->b_data, bp->bio_ma_n);
803 relpbuf(pb, &md_vnode_pbuf_freecnt);
805 bp->bio_resid = auio.uio_resid;
810 mdstart_swap(struct md_s *sc, struct bio *bp)
814 vm_pindex_t i, lastp;
815 int rv, ma_offs, offs, len, lastend;
817 switch (bp->bio_cmd) {
827 ma_offs = (bp->bio_flags & BIO_UNMAPPED) == 0 ? 0 : bp->bio_ma_offset;
830 * offs is the offset at which to start operating on the
831 * next (ie, first) page. lastp is the last page on
832 * which we're going to operate. lastend is the ending
833 * position within that last page (ie, PAGE_SIZE if
834 * we're operating on complete aligned pages).
836 offs = bp->bio_offset % PAGE_SIZE;
837 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
838 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
841 VM_OBJECT_WLOCK(sc->object);
842 vm_object_pip_add(sc->object, 1);
843 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
844 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
845 m = vm_page_grab(sc->object, i, VM_ALLOC_SYSTEM);
846 if (bp->bio_cmd == BIO_READ) {
847 if (m->valid == VM_PAGE_BITS_ALL)
850 rv = vm_pager_get_pages(sc->object, &m, 1, 0);
851 if (rv == VM_PAGER_ERROR) {
854 } else if (rv == VM_PAGER_FAIL) {
856 * Pager does not have the page. Zero
857 * the allocated page, and mark it as
858 * valid. Do not set dirty, the page
859 * can be recreated if thrown out.
862 m->valid = VM_PAGE_BITS_ALL;
864 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
865 pmap_copy_pages(&m, offs, bp->bio_ma,
868 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
869 cpu_flush_dcache(p, len);
871 } else if (bp->bio_cmd == BIO_WRITE) {
872 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL)
873 rv = vm_pager_get_pages(sc->object, &m, 1, 0);
876 if (rv == VM_PAGER_ERROR) {
880 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
881 pmap_copy_pages(bp->bio_ma, ma_offs, &m,
884 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
886 m->valid = VM_PAGE_BITS_ALL;
887 } else if (bp->bio_cmd == BIO_DELETE) {
888 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL)
889 rv = vm_pager_get_pages(sc->object, &m, 1, 0);
892 if (rv == VM_PAGER_ERROR) {
896 if (len != PAGE_SIZE) {
897 pmap_zero_page_area(m, offs, len);
898 vm_page_clear_dirty(m, offs, len);
899 m->valid = VM_PAGE_BITS_ALL;
901 vm_pager_page_unswapped(m);
905 if (bp->bio_cmd == BIO_DELETE && len == PAGE_SIZE)
910 if (bp->bio_cmd == BIO_WRITE) {
912 vm_pager_page_unswapped(m);
915 /* Actions on further pages start at offset 0 */
916 p += PAGE_SIZE - offs;
920 vm_object_pip_wakeup(sc->object);
921 VM_OBJECT_WUNLOCK(sc->object);
922 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
926 mdstart_null(struct md_s *sc, struct bio *bp)
929 switch (bp->bio_cmd) {
931 bzero(bp->bio_data, bp->bio_length);
932 cpu_flush_dcache(bp->bio_data, bp->bio_length);
942 md_kthread(void *arg)
949 thread_lock(curthread);
950 sched_prio(curthread, PRIBIO);
951 thread_unlock(curthread);
952 if (sc->type == MD_VNODE)
953 curthread->td_pflags |= TDP_NORUNNINGBUF;
956 mtx_lock(&sc->queue_mtx);
957 if (sc->flags & MD_SHUTDOWN) {
958 sc->flags |= MD_EXITING;
959 mtx_unlock(&sc->queue_mtx);
962 bp = bioq_takefirst(&sc->bio_queue);
964 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
967 mtx_unlock(&sc->queue_mtx);
968 if (bp->bio_cmd == BIO_GETATTR) {
969 if ((sc->fwsectors && sc->fwheads &&
970 (g_handleattr_int(bp, "GEOM::fwsectors",
972 g_handleattr_int(bp, "GEOM::fwheads",
974 g_handleattr_int(bp, "GEOM::candelete", 1))
979 error = sc->start(sc, bp);
983 bp->bio_completed = bp->bio_length;
984 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE))
985 devstat_end_transaction_bio(sc->devstat, bp);
986 g_io_deliver(bp, error);
996 LIST_FOREACH(sc, &md_softc_list, list) {
997 if (sc->unit == unit)
1003 static struct md_s *
1004 mdnew(int unit, int *errp, enum md_types type)
1011 unit = alloc_unr(md_uh);
1013 unit = alloc_unr_specific(md_uh, unit);
1020 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
1022 bioq_init(&sc->bio_queue);
1023 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1024 mtx_init(&sc->stat_mtx, "md stat", NULL, MTX_DEF);
1026 sprintf(sc->name, "md%d", unit);
1027 LIST_INSERT_HEAD(&md_softc_list, sc, list);
1028 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1031 LIST_REMOVE(sc, list);
1032 mtx_destroy(&sc->stat_mtx);
1033 mtx_destroy(&sc->queue_mtx);
1034 free_unr(md_uh, sc->unit);
1041 mdinit(struct md_s *sc)
1044 struct g_provider *pp;
1047 gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1049 pp = g_new_providerf(gp, "md%d", sc->unit);
1050 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1051 pp->mediasize = sc->mediasize;
1052 pp->sectorsize = sc->sectorsize;
1057 pp->flags |= G_PF_ACCEPT_UNMAPPED;
1065 g_error_provider(pp, 0);
1066 g_topology_unlock();
1067 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1068 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
1072 mdcreate_malloc(struct md_s *sc, struct md_ioctl *mdio)
1079 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1081 if (mdio->md_sectorsize != 0 && !powerof2(mdio->md_sectorsize))
1083 /* Compression doesn't make sense if we have reserved space */
1084 if (mdio->md_options & MD_RESERVE)
1085 mdio->md_options &= ~MD_COMPRESS;
1086 if (mdio->md_fwsectors != 0)
1087 sc->fwsectors = mdio->md_fwsectors;
1088 if (mdio->md_fwheads != 0)
1089 sc->fwheads = mdio->md_fwheads;
1090 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE);
1091 sc->indir = dimension(sc->mediasize / sc->sectorsize);
1092 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
1094 if (mdio->md_options & MD_RESERVE) {
1097 nsectors = sc->mediasize / sc->sectorsize;
1098 for (u = 0; u < nsectors; u++) {
1099 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ?
1100 M_WAITOK : M_NOWAIT) | M_ZERO);
1102 error = s_write(sc->indir, u, sp);
1114 mdsetcred(struct md_s *sc, struct ucred *cred)
1120 * Set credits in our softc
1125 sc->cred = crhold(cred);
1128 * Horrible kludge to establish credentials for NFS XXX.
1135 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1136 bzero(&auio, sizeof(auio));
1138 aiov.iov_base = tmpbuf;
1139 aiov.iov_len = sc->sectorsize;
1140 auio.uio_iov = &aiov;
1141 auio.uio_iovcnt = 1;
1142 auio.uio_offset = 0;
1143 auio.uio_rw = UIO_READ;
1144 auio.uio_segflg = UIO_SYSSPACE;
1145 auio.uio_resid = aiov.iov_len;
1146 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1147 error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
1148 VOP_UNLOCK(sc->vnode, 0);
1149 free(tmpbuf, M_TEMP);
1155 mdcreate_vnode(struct md_s *sc, struct md_ioctl *mdio, struct thread *td)
1158 struct nameidata nd;
1163 * Kernel-originated requests must have the filename appended
1164 * to the mdio structure to protect against malicious software.
1166 fname = mdio->md_file;
1167 if ((void *)fname != (void *)(mdio + 1)) {
1168 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
1172 strlcpy(sc->file, fname, sizeof(sc->file));
1175 * If the user specified that this is a read only device, don't
1176 * set the FWRITE mask before trying to open the backing store.
1178 flags = FREAD | ((mdio->md_options & MD_READONLY) ? 0 : FWRITE);
1179 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td);
1180 error = vn_open(&nd, &flags, 0, NULL);
1183 NDFREE(&nd, NDF_ONLY_PNBUF);
1184 if (nd.ni_vp->v_type != VREG) {
1188 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1191 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1192 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1193 if (nd.ni_vp->v_iflag & VI_DOOMED) {
1194 /* Forced unmount. */
1199 nd.ni_vp->v_vflag |= VV_MD;
1200 VOP_UNLOCK(nd.ni_vp, 0);
1202 if (mdio->md_fwsectors != 0)
1203 sc->fwsectors = mdio->md_fwsectors;
1204 if (mdio->md_fwheads != 0)
1205 sc->fwheads = mdio->md_fwheads;
1206 sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC);
1207 if (!(flags & FWRITE))
1208 sc->flags |= MD_READONLY;
1209 sc->vnode = nd.ni_vp;
1211 error = mdsetcred(sc, td->td_ucred);
1214 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1215 nd.ni_vp->v_vflag &= ~VV_MD;
1220 VOP_UNLOCK(nd.ni_vp, 0);
1221 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1226 mddestroy(struct md_s *sc, struct thread *td)
1230 sc->gp->softc = NULL;
1232 g_wither_geom(sc->gp, ENXIO);
1233 g_topology_unlock();
1238 devstat_remove_entry(sc->devstat);
1241 mtx_lock(&sc->queue_mtx);
1242 sc->flags |= MD_SHUTDOWN;
1244 while (!(sc->flags & MD_EXITING))
1245 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1246 mtx_unlock(&sc->queue_mtx);
1247 mtx_destroy(&sc->stat_mtx);
1248 mtx_destroy(&sc->queue_mtx);
1249 if (sc->vnode != NULL) {
1250 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1251 sc->vnode->v_vflag &= ~VV_MD;
1252 VOP_UNLOCK(sc->vnode, 0);
1253 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1254 FREAD : (FREAD|FWRITE), sc->cred, td);
1256 if (sc->cred != NULL)
1258 if (sc->object != NULL)
1259 vm_object_deallocate(sc->object);
1261 destroy_indir(sc, sc->indir);
1263 uma_zdestroy(sc->uma);
1265 LIST_REMOVE(sc, list);
1266 free_unr(md_uh, sc->unit);
1272 mdresize(struct md_s *sc, struct md_ioctl *mdio)
1275 vm_pindex_t oldpages, newpages;
1282 if (mdio->md_mediasize <= 0 ||
1283 (mdio->md_mediasize % PAGE_SIZE) != 0)
1285 oldpages = OFF_TO_IDX(round_page(sc->mediasize));
1286 newpages = OFF_TO_IDX(round_page(mdio->md_mediasize));
1287 if (newpages < oldpages) {
1288 VM_OBJECT_WLOCK(sc->object);
1289 vm_object_page_remove(sc->object, newpages, 0, 0);
1290 swap_pager_freespace(sc->object, newpages,
1291 oldpages - newpages);
1292 swap_release_by_cred(IDX_TO_OFF(oldpages -
1293 newpages), sc->cred);
1294 sc->object->charge = IDX_TO_OFF(newpages);
1295 sc->object->size = newpages;
1296 VM_OBJECT_WUNLOCK(sc->object);
1297 } else if (newpages > oldpages) {
1298 res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1299 oldpages), sc->cred);
1302 if ((mdio->md_options & MD_RESERVE) ||
1303 (sc->flags & MD_RESERVE)) {
1304 error = swap_pager_reserve(sc->object,
1305 oldpages, newpages - oldpages);
1307 swap_release_by_cred(
1308 IDX_TO_OFF(newpages - oldpages),
1313 VM_OBJECT_WLOCK(sc->object);
1314 sc->object->charge = IDX_TO_OFF(newpages);
1315 sc->object->size = newpages;
1316 VM_OBJECT_WUNLOCK(sc->object);
1320 return (EOPNOTSUPP);
1323 sc->mediasize = mdio->md_mediasize;
1325 g_resize_provider(sc->pp, sc->mediasize);
1326 g_topology_unlock();
1331 mdcreate_swap(struct md_s *sc, struct md_ioctl *mdio, struct thread *td)
1337 * Range check. Disallow negative sizes and sizes not being
1338 * multiple of page size.
1340 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1344 * Allocate an OBJT_SWAP object.
1346 * Note the truncation.
1349 npage = mdio->md_mediasize / PAGE_SIZE;
1350 if (mdio->md_fwsectors != 0)
1351 sc->fwsectors = mdio->md_fwsectors;
1352 if (mdio->md_fwheads != 0)
1353 sc->fwheads = mdio->md_fwheads;
1354 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1355 VM_PROT_DEFAULT, 0, td->td_ucred);
1356 if (sc->object == NULL)
1358 sc->flags = mdio->md_options & (MD_FORCE | MD_RESERVE);
1359 if (mdio->md_options & MD_RESERVE) {
1360 if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1365 error = mdsetcred(sc, td->td_ucred);
1368 vm_object_deallocate(sc->object);
1375 mdcreate_null(struct md_s *sc, struct md_ioctl *mdio, struct thread *td)
1379 * Range check. Disallow negative sizes and sizes not being
1380 * multiple of page size.
1382 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1389 xmdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
1391 struct md_ioctl *mdio;
1397 printf("mdctlioctl(%s %lx %p %x %p)\n",
1398 devtoname(dev), cmd, addr, flags, td);
1400 mdio = (struct md_ioctl *)addr;
1401 if (mdio->md_version != MDIOVERSION)
1405 * We assert the version number in the individual ioctl
1406 * handlers instead of out here because (a) it is possible we
1407 * may add another ioctl in the future which doesn't read an
1408 * mdio, and (b) the correct return value for an unknown ioctl
1409 * is ENOIOCTL, not EINVAL.
1414 switch (mdio->md_type) {
1424 if (mdio->md_sectorsize == 0)
1425 sectsize = DEV_BSIZE;
1427 sectsize = mdio->md_sectorsize;
1428 if (sectsize > MAXPHYS || mdio->md_mediasize < sectsize)
1430 if (mdio->md_options & MD_AUTOUNIT)
1431 sc = mdnew(-1, &error, mdio->md_type);
1433 if (mdio->md_unit > INT_MAX)
1435 sc = mdnew(mdio->md_unit, &error, mdio->md_type);
1439 if (mdio->md_options & MD_AUTOUNIT)
1440 mdio->md_unit = sc->unit;
1441 sc->mediasize = mdio->md_mediasize;
1442 sc->sectorsize = sectsize;
1446 sc->start = mdstart_malloc;
1447 error = mdcreate_malloc(sc, mdio);
1451 * We disallow attaching preloaded memory disks via
1452 * ioctl. Preloaded memory disks are automatically
1453 * attached in g_md_init().
1458 sc->start = mdstart_vnode;
1459 error = mdcreate_vnode(sc, mdio, td);
1462 sc->start = mdstart_swap;
1463 error = mdcreate_swap(sc, mdio, td);
1466 sc->start = mdstart_null;
1467 error = mdcreate_null(sc, mdio, td);
1475 /* Prune off any residual fractional sector */
1476 i = sc->mediasize % sc->sectorsize;
1482 if (mdio->md_mediasize != 0 ||
1483 (mdio->md_options & ~MD_FORCE) != 0)
1486 sc = mdfind(mdio->md_unit);
1489 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1490 !(mdio->md_options & MD_FORCE))
1492 return (mddestroy(sc, td));
1494 if ((mdio->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1497 sc = mdfind(mdio->md_unit);
1500 if (mdio->md_mediasize < sc->sectorsize)
1502 if (mdio->md_mediasize < sc->mediasize &&
1503 !(sc->flags & MD_FORCE) &&
1504 !(mdio->md_options & MD_FORCE))
1506 return (mdresize(sc, mdio));
1508 sc = mdfind(mdio->md_unit);
1511 mdio->md_type = sc->type;
1512 mdio->md_options = sc->flags;
1513 mdio->md_mediasize = sc->mediasize;
1514 mdio->md_sectorsize = sc->sectorsize;
1515 if (sc->type == MD_VNODE)
1516 error = copyout(sc->file, mdio->md_file,
1517 strlen(sc->file) + 1);
1521 LIST_FOREACH(sc, &md_softc_list, list) {
1522 if (i == MDNPAD - 1)
1523 mdio->md_pad[i] = -1;
1525 mdio->md_pad[i++] = sc->unit;
1527 mdio->md_pad[0] = i - 1;
1535 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
1540 error = xmdctlioctl(dev, cmd, addr, flags, td);
1546 md_preloaded(u_char *image, size_t length, const char *name)
1551 sc = mdnew(-1, &error, MD_PRELOAD);
1554 sc->mediasize = length;
1555 sc->sectorsize = DEV_BSIZE;
1557 sc->pl_len = length;
1558 sc->start = mdstart_preload;
1561 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
1565 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
1566 MD_NAME, sc->unit, name, length, image);
1568 printf("%s%d: Embedded image %zd bytes at %p\n",
1569 MD_NAME, sc->unit, length, image);
1574 g_md_init(struct g_class *mp __unused)
1577 u_char *ptr, *name, *type;
1581 /* figure out log2(NINDIR) */
1582 for (i = NINDIR, nshift = -1; i; nshift++)
1586 sx_init(&md_sx, "MD config lock");
1587 g_topology_unlock();
1588 md_uh = new_unrhdr(0, INT_MAX, NULL);
1590 if (mfs_root_size != 0) {
1592 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
1597 /* XXX: are preload_* static or do they need Giant ? */
1598 while ((mod = preload_search_next_name(mod)) != NULL) {
1599 name = (char *)preload_search_info(mod, MODINFO_NAME);
1602 type = (char *)preload_search_info(mod, MODINFO_TYPE);
1605 if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
1607 ptr = preload_fetch_addr(mod);
1608 len = preload_fetch_size(mod);
1609 if (ptr != NULL && len != 0) {
1611 md_preloaded(ptr, len, name);
1615 md_vnode_pbuf_freecnt = nswbuf / 10;
1616 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
1622 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1623 struct g_consumer *cp __unused, struct g_provider *pp)
1654 if (indent == NULL) {
1655 sbuf_printf(sb, " u %d", mp->unit);
1656 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
1657 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
1658 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
1659 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
1660 sbuf_printf(sb, " t %s", type);
1661 if (mp->type == MD_VNODE && mp->vnode != NULL)
1662 sbuf_printf(sb, " file %s", mp->file);
1664 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
1666 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
1667 indent, (uintmax_t) mp->sectorsize);
1668 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
1669 indent, (uintmax_t) mp->fwheads);
1670 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
1671 indent, (uintmax_t) mp->fwsectors);
1672 sbuf_printf(sb, "%s<length>%ju</length>\n",
1673 indent, (uintmax_t) mp->mediasize);
1674 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
1675 (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
1676 sbuf_printf(sb, "%s<access>%s</access>\n", indent,
1677 (mp->flags & MD_READONLY) == 0 ? "read-write":
1679 sbuf_printf(sb, "%s<type>%s</type>\n", indent,
1681 if (mp->type == MD_VNODE && mp->vnode != NULL) {
1682 sbuf_printf(sb, "%s<file>", indent);
1683 g_conf_printf_escaped(sb, "%s", mp->file);
1684 sbuf_printf(sb, "</file>\n");
1691 g_md_fini(struct g_class *mp __unused)
1695 if (status_dev != NULL)
1696 destroy_dev(status_dev);
1697 delete_unrhdr(md_uh);