md.c revision 126935
1/* 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 * ---------------------------------------------------------------------------- 8 * 9 * $FreeBSD: head/sys/dev/md/md.c 126935 2004-03-14 00:24:30Z alc $ 10 * 11 */ 12 13/* 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: 17 * 18 * Copyright (c) 1988 University of Utah. 19 * Copyright (c) 1990, 1993 20 * The Regents of the University of California. All rights reserved. 21 * 22 * This code is derived from software contributed to Berkeley by 23 * the Systems Programming Group of the University of Utah Computer 24 * Science Department. 25 * 26 * Redistribution and use in source and binary forms, with or without 27 * modification, are permitted provided that the following conditions 28 * are met: 29 * 1. Redistributions of source code must retain the above copyright 30 * notice, this list of conditions and the following disclaimer. 31 * 2. Redistributions in binary form must reproduce the above copyright 32 * notice, this list of conditions and the following disclaimer in the 33 * documentation and/or other materials provided with the distribution. 34 * 3. All advertising materials mentioning features or use of this software 35 * must display the following acknowledgement: 36 * This product includes software developed by the University of 37 * California, Berkeley and its contributors. 38 * 4. Neither the name of the University nor the names of its contributors 39 * may be used to endorse or promote products derived from this software 40 * without specific prior written permission. 41 * 42 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 43 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 44 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 45 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 46 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 47 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 48 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 49 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 50 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 51 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 52 * SUCH DAMAGE. 53 * 54 * from: Utah Hdr: vn.c 1.13 94/04/02 55 * 56 * from: @(#)vn.c 8.6 (Berkeley) 4/1/94 57 * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03 58 */ 59 60#include "opt_geom.h" 61#include "opt_md.h" 62 63#include <sys/param.h> 64#include <sys/systm.h> 65#include <sys/bio.h> 66#include <sys/conf.h> 67#include <sys/fcntl.h> 68#include <sys/kernel.h> 69#include <sys/kthread.h> 70#include <sys/linker.h> 71#include <sys/lock.h> 72#include <sys/malloc.h> 73#include <sys/mdioctl.h> 74#include <sys/mutex.h> 75#include <sys/namei.h> 76#include <sys/proc.h> 77#include <sys/queue.h> 78#include <sys/sysctl.h> 79#include <sys/vnode.h> 80 81#include <geom/geom.h> 82 83#include <vm/vm.h> 84#include <vm/vm_object.h> 85#include <vm/vm_page.h> 86#include <vm/vm_pager.h> 87#include <vm/swap_pager.h> 88#include <vm/uma.h> 89 90#define MD_MODVER 1 91 92#define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */ 93 94#ifndef MD_NSECT 95#define MD_NSECT (10000 * 2) 96#endif 97 98static MALLOC_DEFINE(M_MD, "MD disk", "Memory Disk"); 99static MALLOC_DEFINE(M_MDSECT, "MD sectors", "Memory Disk Sectors"); 100 101static int md_debug; 102SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, ""); 103 104#if defined(MD_ROOT) && defined(MD_ROOT_SIZE) 105/* Image gets put here: */ 106static u_char mfs_root[MD_ROOT_SIZE*1024] = "MFS Filesystem goes here"; 107static u_char end_mfs_root[] __unused = "MFS Filesystem had better STOP here"; 108#endif 109 110static g_init_t md_drvinit; 111 112static int mdrootready; 113static int mdunits; 114static dev_t status_dev = 0; 115 116static d_ioctl_t mdctlioctl; 117 118static struct cdevsw mdctl_cdevsw = { 119 .d_version = D_VERSION, 120 .d_flags = D_NEEDGIANT, 121 .d_ioctl = mdctlioctl, 122 .d_name = MD_NAME, 123}; 124 125 126static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(&md_softc_list); 127 128#define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 129#define NMASK (NINDIR-1) 130static int nshift; 131 132struct indir { 133 uintptr_t *array; 134 u_int total; 135 u_int used; 136 u_int shift; 137}; 138 139struct md_s { 140 int unit; 141 LIST_ENTRY(md_s) list; 142 struct bio_queue_head bio_queue; 143 struct mtx queue_mtx; 144 dev_t dev; 145 enum md_types type; 146 unsigned nsect; 147 unsigned opencount; 148 unsigned secsize; 149 unsigned fwheads; 150 unsigned fwsectors; 151 unsigned flags; 152 char name[20]; 153 struct proc *procp; 154 struct g_geom *gp; 155 struct g_provider *pp; 156 157 /* MD_MALLOC related fields */ 158 struct indir *indir; 159 uma_zone_t uma; 160 161 /* MD_PRELOAD related fields */ 162 u_char *pl_ptr; 163 unsigned pl_len; 164 165 /* MD_VNODE related fields */ 166 struct vnode *vnode; 167 struct ucred *cred; 168 169 /* MD_SWAP related fields */ 170 vm_object_t object; 171 unsigned npage; 172}; 173 174static int mddestroy(struct md_s *sc, struct thread *td); 175 176static struct indir * 177new_indir(u_int shift) 178{ 179 struct indir *ip; 180 181 ip = malloc(sizeof *ip, M_MD, M_NOWAIT | M_ZERO); 182 if (ip == NULL) 183 return (NULL); 184 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 185 M_MDSECT, M_NOWAIT | M_ZERO); 186 if (ip->array == NULL) { 187 free(ip, M_MD); 188 return (NULL); 189 } 190 ip->total = NINDIR; 191 ip->shift = shift; 192 return (ip); 193} 194 195static void 196del_indir(struct indir *ip) 197{ 198 199 free(ip->array, M_MDSECT); 200 free(ip, M_MD); 201} 202 203static void 204destroy_indir(struct md_s *sc, struct indir *ip) 205{ 206 int i; 207 208 for (i = 0; i < NINDIR; i++) { 209 if (!ip->array[i]) 210 continue; 211 if (ip->shift) 212 destroy_indir(sc, (struct indir*)(ip->array[i])); 213 else if (ip->array[i] > 255) 214 uma_zfree(sc->uma, (void *)(ip->array[i])); 215 } 216 del_indir(ip); 217} 218 219/* 220 * This function does the math and alloctes the top level "indir" structure 221 * for a device of "size" sectors. 222 */ 223 224static struct indir * 225dimension(off_t size) 226{ 227 off_t rcnt; 228 struct indir *ip; 229 int i, layer; 230 231 rcnt = size; 232 layer = 0; 233 while (rcnt > NINDIR) { 234 rcnt /= NINDIR; 235 layer++; 236 } 237 /* figure out log2(NINDIR) */ 238 for (i = NINDIR, nshift = -1; i; nshift++) 239 i >>= 1; 240 241 /* 242 * XXX: the top layer is probably not fully populated, so we allocate 243 * too much space for ip->array in here. 244 */ 245 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 246 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 247 M_MDSECT, M_WAITOK | M_ZERO); 248 ip->total = NINDIR; 249 ip->shift = layer * nshift; 250 return (ip); 251} 252 253/* 254 * Read a given sector 255 */ 256 257static uintptr_t 258s_read(struct indir *ip, off_t offset) 259{ 260 struct indir *cip; 261 int idx; 262 uintptr_t up; 263 264 if (md_debug > 1) 265 printf("s_read(%jd)\n", (intmax_t)offset); 266 up = 0; 267 for (cip = ip; cip != NULL;) { 268 if (cip->shift) { 269 idx = (offset >> cip->shift) & NMASK; 270 up = cip->array[idx]; 271 cip = (struct indir *)up; 272 continue; 273 } 274 idx = offset & NMASK; 275 return (cip->array[idx]); 276 } 277 return (0); 278} 279 280/* 281 * Write a given sector, prune the tree if the value is 0 282 */ 283 284static int 285s_write(struct indir *ip, off_t offset, uintptr_t ptr) 286{ 287 struct indir *cip, *lip[10]; 288 int idx, li; 289 uintptr_t up; 290 291 if (md_debug > 1) 292 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 293 up = 0; 294 li = 0; 295 cip = ip; 296 for (;;) { 297 lip[li++] = cip; 298 if (cip->shift) { 299 idx = (offset >> cip->shift) & NMASK; 300 up = cip->array[idx]; 301 if (up != 0) { 302 cip = (struct indir *)up; 303 continue; 304 } 305 /* Allocate branch */ 306 cip->array[idx] = 307 (uintptr_t)new_indir(cip->shift - nshift); 308 if (cip->array[idx] == 0) 309 return (ENOSPC); 310 cip->used++; 311 up = cip->array[idx]; 312 cip = (struct indir *)up; 313 continue; 314 } 315 /* leafnode */ 316 idx = offset & NMASK; 317 up = cip->array[idx]; 318 if (up != 0) 319 cip->used--; 320 cip->array[idx] = ptr; 321 if (ptr != 0) 322 cip->used++; 323 break; 324 } 325 if (cip->used != 0 || li == 1) 326 return (0); 327 li--; 328 while (cip->used == 0 && cip != ip) { 329 li--; 330 idx = (offset >> lip[li]->shift) & NMASK; 331 up = lip[li]->array[idx]; 332 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 333 del_indir(cip); 334 lip[li]->array[idx] = 0; 335 lip[li]->used--; 336 cip = lip[li]; 337 } 338 return (0); 339} 340 341 342struct g_class g_md_class = { 343 .name = "MD", 344 .init = md_drvinit, 345}; 346 347static int 348g_md_access(struct g_provider *pp, int r, int w, int e) 349{ 350 struct md_s *sc; 351 352 sc = pp->geom->softc; 353 if (sc == NULL) 354 return (ENXIO); 355 r += pp->acr; 356 w += pp->acw; 357 e += pp->ace; 358 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 359 sc->opencount = 1; 360 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 361 sc->opencount = 0; 362 } 363 return (0); 364} 365 366static void 367g_md_start(struct bio *bp) 368{ 369 struct md_s *sc; 370 371 sc = bp->bio_to->geom->softc; 372 373 bp->bio_pblkno = bp->bio_offset / sc->secsize; 374 bp->bio_bcount = bp->bio_length; 375 mtx_lock(&sc->queue_mtx); 376 bioq_disksort(&sc->bio_queue, bp); 377 mtx_unlock(&sc->queue_mtx); 378 379 wakeup(sc); 380} 381 382DECLARE_GEOM_CLASS(g_md_class, g_md); 383 384 385static int 386mdstart_malloc(struct md_s *sc, struct bio *bp) 387{ 388 int i, error; 389 u_char *dst; 390 unsigned secno, nsec, uc; 391 uintptr_t sp, osp; 392 393 nsec = bp->bio_bcount / sc->secsize; 394 secno = bp->bio_pblkno; 395 dst = bp->bio_data; 396 error = 0; 397 while (nsec--) { 398 osp = s_read(sc->indir, secno); 399 if (bp->bio_cmd == BIO_DELETE) { 400 if (osp != 0) 401 error = s_write(sc->indir, secno, 0); 402 } else if (bp->bio_cmd == BIO_READ) { 403 if (osp == 0) 404 bzero(dst, sc->secsize); 405 else if (osp <= 255) 406 for (i = 0; i < sc->secsize; i++) 407 dst[i] = osp; 408 else 409 bcopy((void *)osp, dst, sc->secsize); 410 osp = 0; 411 } else if (bp->bio_cmd == BIO_WRITE) { 412 if (sc->flags & MD_COMPRESS) { 413 uc = dst[0]; 414 for (i = 1; i < sc->secsize; i++) 415 if (dst[i] != uc) 416 break; 417 } else { 418 i = 0; 419 uc = 0; 420 } 421 if (i == sc->secsize) { 422 if (osp != uc) 423 error = s_write(sc->indir, secno, uc); 424 } else { 425 if (osp <= 255) { 426 sp = (uintptr_t) uma_zalloc( 427 sc->uma, M_NOWAIT); 428 if (sp == 0) { 429 error = ENOSPC; 430 break; 431 } 432 bcopy(dst, (void *)sp, sc->secsize); 433 error = s_write(sc->indir, secno, sp); 434 } else { 435 bcopy(dst, (void *)osp, sc->secsize); 436 osp = 0; 437 } 438 } 439 } else { 440 error = EOPNOTSUPP; 441 } 442 if (osp > 255) 443 uma_zfree(sc->uma, (void*)osp); 444 if (error) 445 break; 446 secno++; 447 dst += sc->secsize; 448 } 449 bp->bio_resid = 0; 450 return (error); 451} 452 453static int 454mdstart_preload(struct md_s *sc, struct bio *bp) 455{ 456 457 if (bp->bio_cmd == BIO_DELETE) { 458 } else if (bp->bio_cmd == BIO_READ) { 459 bcopy(sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_data, bp->bio_bcount); 460 } else { 461 bcopy(bp->bio_data, sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_bcount); 462 } 463 bp->bio_resid = 0; 464 return (0); 465} 466 467static int 468mdstart_vnode(struct md_s *sc, struct bio *bp) 469{ 470 int error; 471 struct uio auio; 472 struct iovec aiov; 473 struct mount *mp; 474 475 /* 476 * VNODE I/O 477 * 478 * If an error occurs, we set BIO_ERROR but we do not set 479 * B_INVAL because (for a write anyway), the buffer is 480 * still valid. 481 */ 482 483 bzero(&auio, sizeof(auio)); 484 485 aiov.iov_base = bp->bio_data; 486 aiov.iov_len = bp->bio_bcount; 487 auio.uio_iov = &aiov; 488 auio.uio_iovcnt = 1; 489 auio.uio_offset = (vm_ooffset_t)bp->bio_pblkno * sc->secsize; 490 auio.uio_segflg = UIO_SYSSPACE; 491 if(bp->bio_cmd == BIO_READ) 492 auio.uio_rw = UIO_READ; 493 else if(bp->bio_cmd == BIO_WRITE) 494 auio.uio_rw = UIO_WRITE; 495 else 496 panic("wrong BIO_OP in mdstart_vnode"); 497 auio.uio_resid = bp->bio_bcount; 498 auio.uio_td = curthread; 499 /* 500 * When reading set IO_DIRECT to try to avoid double-caching 501 * the data. When writing IO_DIRECT is not optimal. 502 */ 503 if (bp->bio_cmd == BIO_READ) { 504 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 505 error = VOP_READ(sc->vnode, &auio, IO_DIRECT, sc->cred); 506 VOP_UNLOCK(sc->vnode, 0, curthread); 507 } else { 508 (void) vn_start_write(sc->vnode, &mp, V_WAIT); 509 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 510 error = VOP_WRITE(sc->vnode, &auio, 511 sc->flags & MD_ASYNC ? 0 : IO_SYNC, sc->cred); 512 VOP_UNLOCK(sc->vnode, 0, curthread); 513 vn_finished_write(mp); 514 } 515 bp->bio_resid = auio.uio_resid; 516 return (error); 517} 518 519#include <vm/vm_extern.h> 520#include <vm/vm_kern.h> 521 522static int 523mdstart_swap(struct md_s *sc, struct bio *bp) 524{ 525 { 526 int i, rv; 527 int offs, len, lastp, lastend; 528 vm_page_t m; 529 u_char *p; 530 vm_offset_t kva; 531 532 p = bp->bio_data; 533 534 /* 535 * offs is the ofset at whih to start operating on the 536 * next (ie, first) page. lastp is the last page on 537 * which we're going to operate. lastend is the ending 538 * position within that last page (ie, PAGE_SIZE if 539 * we're operating on complete aligned pages). 540 */ 541 offs = bp->bio_offset % PAGE_SIZE; 542 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE; 543 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1; 544 545 kva = kmem_alloc_nofault(kernel_map, PAGE_SIZE); 546 547 VM_OBJECT_LOCK(sc->object); 548 vm_object_pip_add(sc->object, 1); 549 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) { 550 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs; 551 552 m = vm_page_grab(sc->object, i, 553 VM_ALLOC_NORMAL|VM_ALLOC_RETRY); 554 pmap_qenter(kva, &m, 1); 555 if (bp->bio_cmd == BIO_READ) { 556 if (m->valid != VM_PAGE_BITS_ALL) { 557 rv = vm_pager_get_pages(sc->object, 558 &m, 1, 0); 559 } 560 bcopy((void *)(kva + offs), p, len); 561 } else if (bp->bio_cmd == BIO_WRITE) { 562 if (len != PAGE_SIZE && m->valid != 563 VM_PAGE_BITS_ALL) { 564 rv = vm_pager_get_pages(sc->object, 565 &m, 1, 0); 566 } 567 bcopy(p, (void *)(kva + offs), len); 568 m->valid = VM_PAGE_BITS_ALL; 569#if 0 570 } else if (bp->bio_cmd == BIO_DELETE) { 571 if (len != PAGE_SIZE && m->valid != 572 VM_PAGE_BITS_ALL) { 573 rv = vm_pager_get_pages(sc->object, 574 &m, 1, 0); 575 } 576 bzero((void *)(kva + offs), len); 577 vm_page_dirty(m); 578 m->valid = VM_PAGE_BITS_ALL; 579#endif 580 } 581 pmap_qremove(kva, 1); 582 vm_page_lock_queues(); 583 vm_page_wakeup(m); 584 vm_page_activate(m); 585 if (bp->bio_cmd == BIO_WRITE) { 586 vm_page_dirty(m); 587 } 588 vm_page_unlock_queues(); 589 590 /* Actions on further pages start at offset 0 */ 591 p += PAGE_SIZE - offs; 592 offs = 0; 593#if 0 594if (bootverbose || bp->bio_offset / PAGE_SIZE < 17) 595printf("wire_count %d busy %d flags %x hold_count %d act_count %d queue %d valid %d dirty %d @ %d\n", 596 m->wire_count, m->busy, 597 m->flags, m->hold_count, m->act_count, m->queue, m->valid, m->dirty, i); 598#endif 599 } 600 vm_object_pip_subtract(sc->object, 1); 601 vm_object_set_writeable_dirty(sc->object); 602 VM_OBJECT_UNLOCK(sc->object); 603 kmem_free(kernel_map, kva, sc->secsize); 604 return (0); 605 } 606} 607 608static void 609md_kthread(void *arg) 610{ 611 struct md_s *sc; 612 struct bio *bp; 613 int error, hasgiant; 614 615 sc = arg; 616 curthread->td_base_pri = PRIBIO; 617 618 switch (sc->type) { 619 case MD_VNODE: 620 mtx_lock(&Giant); 621 hasgiant = 1; 622 break; 623 case MD_MALLOC: 624 case MD_PRELOAD: 625 case MD_SWAP: 626 default: 627 hasgiant = 0; 628 break; 629 } 630 631 for (;;) { 632 mtx_lock(&sc->queue_mtx); 633 bp = bioq_first(&sc->bio_queue); 634 if (bp) 635 bioq_remove(&sc->bio_queue, bp); 636 if (!bp) { 637 if (sc->flags & MD_SHUTDOWN) { 638 mtx_unlock(&sc->queue_mtx); 639 sc->procp = NULL; 640 wakeup(&sc->procp); 641 if (hasgiant) 642 mtx_unlock(&Giant); 643 kthread_exit(0); 644 } 645 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 646 continue; 647 } 648 mtx_unlock(&sc->queue_mtx); 649 if (bp->bio_cmd == BIO_GETATTR) { 650 if (sc->fwsectors && sc->fwheads && 651 (g_handleattr_int(bp, "GEOM::fwsectors", 652 sc->fwsectors) || 653 g_handleattr_int(bp, "GEOM::fwheads", 654 sc->fwheads))) 655 error = -1; 656 else 657 error = EOPNOTSUPP; 658 } else { 659 switch (sc->type) { 660 case MD_MALLOC: 661 error = mdstart_malloc(sc, bp); 662 break; 663 case MD_PRELOAD: 664 error = mdstart_preload(sc, bp); 665 break; 666 case MD_VNODE: 667 error = mdstart_vnode(sc, bp); 668 break; 669 case MD_SWAP: 670 error = mdstart_swap(sc, bp); 671 break; 672 default: 673 panic("Impossible md(type)"); 674 break; 675 } 676 } 677 678 if (error != -1) { 679 bp->bio_completed = bp->bio_length; 680 g_io_deliver(bp, error); 681 } 682 } 683} 684 685static struct md_s * 686mdfind(int unit) 687{ 688 struct md_s *sc; 689 690 /* XXX: LOCK(unique unit numbers) */ 691 LIST_FOREACH(sc, &md_softc_list, list) { 692 if (sc->unit == unit) 693 break; 694 } 695 /* XXX: UNLOCK(unique unit numbers) */ 696 return (sc); 697} 698 699static struct md_s * 700mdnew(int unit) 701{ 702 struct md_s *sc; 703 int error, max = -1; 704 705 /* XXX: LOCK(unique unit numbers) */ 706 LIST_FOREACH(sc, &md_softc_list, list) { 707 if (sc->unit == unit) { 708 /* XXX: UNLOCK(unique unit numbers) */ 709 return (NULL); 710 } 711 if (sc->unit > max) 712 max = sc->unit; 713 } 714 if (unit == -1) 715 unit = max + 1; 716 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 717 sc->unit = unit; 718 bioq_init(&sc->bio_queue); 719 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 720 sprintf(sc->name, "md%d", unit); 721 error = kthread_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 722 if (error) { 723 free(sc, M_MD); 724 return (NULL); 725 } 726 LIST_INSERT_HEAD(&md_softc_list, sc, list); 727 /* XXX: UNLOCK(unique unit numbers) */ 728 return (sc); 729} 730 731static void 732mdinit(struct md_s *sc) 733{ 734 735 struct g_geom *gp; 736 struct g_provider *pp; 737 738 DROP_GIANT(); 739 g_topology_lock(); 740 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 741 gp->start = g_md_start; 742 gp->access = g_md_access; 743 gp->softc = sc; 744 pp = g_new_providerf(gp, "md%d", sc->unit); 745 pp->mediasize = (off_t)sc->nsect * sc->secsize; 746 pp->sectorsize = sc->secsize; 747 sc->gp = gp; 748 sc->pp = pp; 749 g_error_provider(pp, 0); 750 g_topology_unlock(); 751 PICKUP_GIANT(); 752} 753 754/* 755 * XXX: we should check that the range they feed us is mapped. 756 * XXX: we should implement read-only. 757 */ 758 759static int 760mdcreate_preload(struct md_ioctl *mdio) 761{ 762 struct md_s *sc; 763 764 if (mdio->md_size == 0) 765 return (EINVAL); 766 if (mdio->md_options & ~(MD_AUTOUNIT)) 767 return (EINVAL); 768 if (mdio->md_options & MD_AUTOUNIT) { 769 sc = mdnew(-1); 770 if (sc == NULL) 771 return (ENOMEM); 772 mdio->md_unit = sc->unit; 773 } else { 774 sc = mdnew(mdio->md_unit); 775 if (sc == NULL) 776 return (EBUSY); 777 } 778 sc->type = MD_PRELOAD; 779 sc->secsize = DEV_BSIZE; 780 sc->nsect = mdio->md_size; 781 sc->flags = mdio->md_options & MD_FORCE; 782 /* Cast to pointer size, then to pointer to avoid warning */ 783 sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base; 784 sc->pl_len = (mdio->md_size << DEV_BSHIFT); 785 mdinit(sc); 786 return (0); 787} 788 789 790static int 791mdcreate_malloc(struct md_ioctl *mdio) 792{ 793 struct md_s *sc; 794 off_t u; 795 uintptr_t sp; 796 int error; 797 798 error = 0; 799 if (mdio->md_size == 0) 800 return (EINVAL); 801 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 802 return (EINVAL); 803 if (mdio->md_secsize != 0 && !powerof2(mdio->md_secsize)) 804 return (EINVAL); 805 /* Compression doesn't make sense if we have reserved space */ 806 if (mdio->md_options & MD_RESERVE) 807 mdio->md_options &= ~MD_COMPRESS; 808 if (mdio->md_options & MD_AUTOUNIT) { 809 sc = mdnew(-1); 810 if (sc == NULL) 811 return (ENOMEM); 812 mdio->md_unit = sc->unit; 813 } else { 814 sc = mdnew(mdio->md_unit); 815 if (sc == NULL) 816 return (EBUSY); 817 } 818 sc->type = MD_MALLOC; 819 if (mdio->md_secsize != 0) 820 sc->secsize = mdio->md_secsize; 821 else 822 sc->secsize = DEV_BSIZE; 823 if (mdio->md_fwsectors != 0) 824 sc->fwsectors = mdio->md_fwsectors; 825 if (mdio->md_fwheads != 0) 826 sc->fwheads = mdio->md_fwheads; 827 sc->nsect = mdio->md_size; 828 sc->nsect /= (sc->secsize / DEV_BSIZE); 829 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 830 sc->indir = dimension(sc->nsect); 831 sc->uma = uma_zcreate(sc->name, sc->secsize, 832 NULL, NULL, NULL, NULL, 0x1ff, 0); 833 if (mdio->md_options & MD_RESERVE) { 834 for (u = 0; u < sc->nsect; u++) { 835 sp = (uintptr_t) uma_zalloc(sc->uma, M_NOWAIT | M_ZERO); 836 if (sp != 0) 837 error = s_write(sc->indir, u, sp); 838 else 839 error = ENOMEM; 840 if (error) 841 break; 842 } 843 } 844 if (error) { 845 mddestroy(sc, NULL); 846 return (error); 847 } 848 mdinit(sc); 849 if (!(mdio->md_options & MD_RESERVE)) 850 sc->pp->flags |= G_PF_CANDELETE; 851 return (0); 852} 853 854 855static int 856mdsetcred(struct md_s *sc, struct ucred *cred) 857{ 858 char *tmpbuf; 859 int error = 0; 860 861 /* 862 * Set credits in our softc 863 */ 864 865 if (sc->cred) 866 crfree(sc->cred); 867 sc->cred = crhold(cred); 868 869 /* 870 * Horrible kludge to establish credentials for NFS XXX. 871 */ 872 873 if (sc->vnode) { 874 struct uio auio; 875 struct iovec aiov; 876 877 tmpbuf = malloc(sc->secsize, M_TEMP, M_WAITOK); 878 bzero(&auio, sizeof(auio)); 879 880 aiov.iov_base = tmpbuf; 881 aiov.iov_len = sc->secsize; 882 auio.uio_iov = &aiov; 883 auio.uio_iovcnt = 1; 884 auio.uio_offset = 0; 885 auio.uio_rw = UIO_READ; 886 auio.uio_segflg = UIO_SYSSPACE; 887 auio.uio_resid = aiov.iov_len; 888 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 889 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 890 VOP_UNLOCK(sc->vnode, 0, curthread); 891 free(tmpbuf, M_TEMP); 892 } 893 return (error); 894} 895 896static int 897mdcreate_vnode(struct md_ioctl *mdio, struct thread *td) 898{ 899 struct md_s *sc; 900 struct vattr vattr; 901 struct nameidata nd; 902 int error, flags; 903 904 flags = FREAD|FWRITE; 905 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 906 error = vn_open(&nd, &flags, 0, -1); 907 if (error) { 908 if (error != EACCES && error != EPERM && error != EROFS) 909 return (error); 910 flags &= ~FWRITE; 911 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 912 error = vn_open(&nd, &flags, 0, -1); 913 if (error) 914 return (error); 915 } 916 NDFREE(&nd, NDF_ONLY_PNBUF); 917 if (nd.ni_vp->v_type != VREG || 918 (error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred, td))) { 919 VOP_UNLOCK(nd.ni_vp, 0, td); 920 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 921 return (error ? error : EINVAL); 922 } 923 VOP_UNLOCK(nd.ni_vp, 0, td); 924 925 if (mdio->md_options & MD_AUTOUNIT) { 926 sc = mdnew(-1); 927 mdio->md_unit = sc->unit; 928 } else { 929 sc = mdnew(mdio->md_unit); 930 } 931 if (sc == NULL) { 932 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 933 return (EBUSY); 934 } 935 936 if (mdio->md_fwsectors != 0) 937 sc->fwsectors = mdio->md_fwsectors; 938 if (mdio->md_fwheads != 0) 939 sc->fwheads = mdio->md_fwheads; 940 sc->type = MD_VNODE; 941 sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC); 942 if (!(flags & FWRITE)) 943 sc->flags |= MD_READONLY; 944 sc->secsize = DEV_BSIZE; 945 sc->vnode = nd.ni_vp; 946 947 /* 948 * If the size is specified, override the file attributes. 949 */ 950 if (mdio->md_size) 951 sc->nsect = mdio->md_size; 952 else 953 sc->nsect = vattr.va_size / sc->secsize; /* XXX: round up ? */ 954 if (sc->nsect == 0) { 955 mddestroy(sc, td); 956 return (EINVAL); 957 } 958 error = mdsetcred(sc, td->td_ucred); 959 if (error) { 960 mddestroy(sc, td); 961 return (error); 962 } 963 mdinit(sc); 964 return (0); 965} 966 967static void 968md_zapit(void *p, int cancel) 969{ 970 if (cancel) 971 return; 972 g_wither_geom(p, ENXIO); 973} 974 975static int 976mddestroy(struct md_s *sc, struct thread *td) 977{ 978 979 GIANT_REQUIRED; 980 981 mtx_destroy(&sc->queue_mtx); 982 if (sc->gp) { 983 sc->gp->softc = NULL; 984 g_waitfor_event(md_zapit, sc->gp, M_WAITOK, sc->gp, NULL); 985 sc->gp = NULL; 986 sc->pp = NULL; 987 } 988 sc->flags |= MD_SHUTDOWN; 989 wakeup(sc); 990 while (sc->procp != NULL) 991 tsleep(&sc->procp, PRIBIO, "mddestroy", hz / 10); 992 if (sc->vnode != NULL) 993 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 994 FREAD : (FREAD|FWRITE), sc->cred, td); 995 if (sc->cred != NULL) 996 crfree(sc->cred); 997 if (sc->object != NULL) { 998 vm_object_deallocate(sc->object); 999 } 1000 if (sc->indir) 1001 destroy_indir(sc, sc->indir); 1002 if (sc->uma) 1003 uma_zdestroy(sc->uma); 1004 1005 /* XXX: LOCK(unique unit numbers) */ 1006 LIST_REMOVE(sc, list); 1007 /* XXX: UNLOCK(unique unit numbers) */ 1008 free(sc, M_MD); 1009 return (0); 1010} 1011 1012static int 1013mdcreate_swap(struct md_ioctl *mdio, struct thread *td) 1014{ 1015 int error; 1016 struct md_s *sc; 1017 1018 GIANT_REQUIRED; 1019 1020 if (mdio->md_options & MD_AUTOUNIT) { 1021 sc = mdnew(-1); 1022 mdio->md_unit = sc->unit; 1023 } else { 1024 sc = mdnew(mdio->md_unit); 1025 } 1026 if (sc == NULL) 1027 return (EBUSY); 1028 1029 sc->type = MD_SWAP; 1030 1031 /* 1032 * Range check. Disallow negative sizes or any size less then the 1033 * size of a page. Then round to a page. 1034 */ 1035 1036 if (mdio->md_size == 0) { 1037 mddestroy(sc, td); 1038 return (EDOM); 1039 } 1040 1041 /* 1042 * Allocate an OBJT_SWAP object. 1043 * 1044 * sc_nsect is in units of DEV_BSIZE. 1045 * sc_npage is in units of PAGE_SIZE. 1046 * 1047 * Note the truncation. 1048 */ 1049 1050 sc->secsize = DEV_BSIZE; 1051 sc->npage = mdio->md_size / (PAGE_SIZE / DEV_BSIZE); 1052 sc->nsect = sc->npage * (PAGE_SIZE / DEV_BSIZE); 1053 if (mdio->md_fwsectors != 0) 1054 sc->fwsectors = mdio->md_fwsectors; 1055 if (mdio->md_fwheads != 0) 1056 sc->fwheads = mdio->md_fwheads; 1057 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * 1058 (vm_offset_t)sc->npage, VM_PROT_DEFAULT, 0); 1059 sc->flags = mdio->md_options & MD_FORCE; 1060 if (mdio->md_options & MD_RESERVE) { 1061 if (swap_pager_reserve(sc->object, 0, sc->npage) < 0) { 1062 vm_object_deallocate(sc->object); 1063 sc->object = NULL; 1064 mddestroy(sc, td); 1065 return (EDOM); 1066 } 1067 } 1068 error = mdsetcred(sc, td->td_ucred); 1069 if (error) { 1070 mddestroy(sc, td); 1071 return (error); 1072 } 1073 mdinit(sc); 1074 if (!(mdio->md_options & MD_RESERVE)) 1075 sc->pp->flags |= G_PF_CANDELETE; 1076 return (0); 1077} 1078 1079static int 1080mddetach(int unit, struct thread *td) 1081{ 1082 struct md_s *sc; 1083 1084 sc = mdfind(unit); 1085 if (sc == NULL) 1086 return (ENOENT); 1087 if (sc->opencount != 0 && !(sc->flags & MD_FORCE)) 1088 return (EBUSY); 1089 switch(sc->type) { 1090 case MD_VNODE: 1091 case MD_SWAP: 1092 case MD_MALLOC: 1093 case MD_PRELOAD: 1094 return (mddestroy(sc, td)); 1095 default: 1096 return (EOPNOTSUPP); 1097 } 1098} 1099 1100static int 1101mdctlioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1102{ 1103 struct md_ioctl *mdio; 1104 struct md_s *sc; 1105 int i; 1106 1107 if (md_debug) 1108 printf("mdctlioctl(%s %lx %p %x %p)\n", 1109 devtoname(dev), cmd, addr, flags, td); 1110 1111 /* 1112 * We assert the version number in the individual ioctl 1113 * handlers instead of out here because (a) it is possible we 1114 * may add another ioctl in the future which doesn't read an 1115 * mdio, and (b) the correct return value for an unknown ioctl 1116 * is ENOIOCTL, not EINVAL. 1117 */ 1118 mdio = (struct md_ioctl *)addr; 1119 switch (cmd) { 1120 case MDIOCATTACH: 1121 if (mdio->md_version != MDIOVERSION) 1122 return (EINVAL); 1123 switch (mdio->md_type) { 1124 case MD_MALLOC: 1125 return (mdcreate_malloc(mdio)); 1126 case MD_PRELOAD: 1127 return (mdcreate_preload(mdio)); 1128 case MD_VNODE: 1129 return (mdcreate_vnode(mdio, td)); 1130 case MD_SWAP: 1131 return (mdcreate_swap(mdio, td)); 1132 default: 1133 return (EINVAL); 1134 } 1135 case MDIOCDETACH: 1136 if (mdio->md_version != MDIOVERSION) 1137 return (EINVAL); 1138 if (mdio->md_file != NULL || mdio->md_size != 0 || 1139 mdio->md_options != 0) 1140 return (EINVAL); 1141 return (mddetach(mdio->md_unit, td)); 1142 case MDIOCQUERY: 1143 if (mdio->md_version != MDIOVERSION) 1144 return (EINVAL); 1145 sc = mdfind(mdio->md_unit); 1146 if (sc == NULL) 1147 return (ENOENT); 1148 mdio->md_type = sc->type; 1149 mdio->md_options = sc->flags; 1150 switch (sc->type) { 1151 case MD_MALLOC: 1152 mdio->md_size = sc->nsect; 1153 break; 1154 case MD_PRELOAD: 1155 mdio->md_size = sc->nsect; 1156 mdio->md_base = (uint64_t)(intptr_t)sc->pl_ptr; 1157 break; 1158 case MD_SWAP: 1159 mdio->md_size = sc->nsect; 1160 break; 1161 case MD_VNODE: 1162 mdio->md_size = sc->nsect; 1163 /* XXX fill this in */ 1164 mdio->md_file = NULL; 1165 break; 1166 } 1167 return (0); 1168 case MDIOCLIST: 1169 i = 1; 1170 LIST_FOREACH(sc, &md_softc_list, list) { 1171 if (i == MDNPAD - 1) 1172 mdio->md_pad[i] = -1; 1173 else 1174 mdio->md_pad[i++] = sc->unit; 1175 } 1176 mdio->md_pad[0] = i - 1; 1177 return (0); 1178 default: 1179 return (ENOIOCTL); 1180 }; 1181 return (ENOIOCTL); 1182} 1183 1184static void 1185md_preloaded(u_char *image, unsigned length) 1186{ 1187 struct md_s *sc; 1188 1189 sc = mdnew(-1); 1190 if (sc == NULL) 1191 return; 1192 sc->type = MD_PRELOAD; 1193 sc->secsize = DEV_BSIZE; 1194 sc->nsect = length / DEV_BSIZE; 1195 sc->pl_ptr = image; 1196 sc->pl_len = length; 1197 if (sc->unit == 0) 1198 mdrootready = 1; 1199 mdinit(sc); 1200} 1201 1202static void 1203md_drvinit(struct g_class *mp __unused) 1204{ 1205 1206 caddr_t mod; 1207 caddr_t c; 1208 u_char *ptr, *name, *type; 1209 unsigned len; 1210 1211 mod = NULL; 1212 g_topology_unlock(); 1213#ifdef MD_ROOT_SIZE 1214 md_preloaded(mfs_root, MD_ROOT_SIZE*1024); 1215#endif 1216 while ((mod = preload_search_next_name(mod)) != NULL) { 1217 name = (char *)preload_search_info(mod, MODINFO_NAME); 1218 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1219 if (name == NULL) 1220 continue; 1221 if (type == NULL) 1222 continue; 1223 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1224 continue; 1225 c = preload_search_info(mod, MODINFO_ADDR); 1226 ptr = *(u_char **)c; 1227 c = preload_search_info(mod, MODINFO_SIZE); 1228 len = *(size_t *)c; 1229 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1230 MD_NAME, mdunits, name, len, ptr); 1231 md_preloaded(ptr, len); 1232 } 1233 status_dev = make_dev(&mdctl_cdevsw, 0xffff00ff, UID_ROOT, GID_WHEEL, 1234 0600, MDCTL_NAME); 1235 g_topology_lock(); 1236} 1237 1238static int 1239md_modevent(module_t mod, int type, void *data) 1240{ 1241 int error; 1242 struct md_s *sc; 1243 1244 switch (type) { 1245 case MOD_LOAD: 1246 break; 1247 case MOD_UNLOAD: 1248 LIST_FOREACH(sc, &md_softc_list, list) { 1249 error = mddetach(sc->unit, curthread); 1250 if (error != 0) 1251 return (error); 1252 } 1253 if (status_dev) 1254 destroy_dev(status_dev); 1255 status_dev = 0; 1256 break; 1257 default: 1258 break; 1259 } 1260 return (0); 1261} 1262 1263static moduledata_t md_mod = { 1264 MD_NAME, 1265 md_modevent, 1266 NULL 1267}; 1268DECLARE_MODULE(md, md_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE); 1269MODULE_VERSION(md, MD_MODVER); 1270 1271 1272#ifdef MD_ROOT 1273static void 1274md_takeroot(void *junk) 1275{ 1276 if (mdrootready) 1277 rootdevnames[0] = "ufs:/dev/md0"; 1278} 1279 1280SYSINIT(md_root, SI_SUB_MOUNT_ROOT, SI_ORDER_FIRST, md_takeroot, NULL); 1281#endif /* MD_ROOT */ 1282