md.c revision 97784
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 97784 2002-06-03 22:09:04Z iedowse $ 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_md.h" 61 62#include <sys/param.h> 63#include <sys/systm.h> 64#include <sys/bio.h> 65#include <sys/conf.h> 66#include <sys/devicestat.h> 67#include <sys/disk.h> 68#include <sys/fcntl.h> 69#include <sys/kernel.h> 70#include <sys/kthread.h> 71#include <sys/linker.h> 72#include <sys/lock.h> 73#include <sys/malloc.h> 74#include <sys/mdioctl.h> 75#include <sys/mutex.h> 76#include <sys/namei.h> 77#include <sys/proc.h> 78#include <sys/queue.h> 79#include <sys/sysctl.h> 80#include <sys/vnode.h> 81 82#include <vm/vm.h> 83#include <vm/vm_object.h> 84#include <vm/vm_page.h> 85#include <vm/vm_pager.h> 86#include <vm/swap_pager.h> 87#include <vm/uma.h> 88 89#define MD_MODVER 1 90 91#define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */ 92 93#ifndef MD_NSECT 94#define MD_NSECT (10000 * 2) 95#endif 96 97static MALLOC_DEFINE(M_MD, "MD disk", "Memory Disk"); 98static MALLOC_DEFINE(M_MDSECT, "MD sectors", "Memory Disk Sectors"); 99 100static int md_debug; 101SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, ""); 102 103#if defined(MD_ROOT) && defined(MD_ROOT_SIZE) 104/* Image gets put here: */ 105static u_char mfs_root[MD_ROOT_SIZE*1024] = "MFS Filesystem goes here"; 106static u_char end_mfs_root[] __unused = "MFS Filesystem had better STOP here"; 107#endif 108 109static int mdrootready; 110static int mdunits; 111static dev_t status_dev = 0; 112 113#define CDEV_MAJOR 95 114 115static d_strategy_t mdstrategy; 116static d_open_t mdopen; 117static d_close_t mdclose; 118static d_ioctl_t mdioctl, mdctlioctl; 119 120static struct cdevsw md_cdevsw = { 121 /* open */ mdopen, 122 /* close */ mdclose, 123 /* read */ physread, 124 /* write */ physwrite, 125 /* ioctl */ mdioctl, 126 /* poll */ nopoll, 127 /* mmap */ nommap, 128 /* strategy */ mdstrategy, 129 /* name */ MD_NAME, 130 /* maj */ CDEV_MAJOR, 131 /* dump */ nodump, 132 /* psize */ nopsize, 133 /* flags */ D_DISK | D_CANFREE | D_MEMDISK, 134}; 135 136static struct cdevsw mdctl_cdevsw = { 137 /* open */ nullopen, 138 /* close */ nullclose, 139 /* read */ noread, 140 /* write */ nowrite, 141 /* ioctl */ mdctlioctl, 142 /* poll */ nopoll, 143 /* mmap */ nommap, 144 /* strategy */ nostrategy, 145 /* name */ MD_NAME, 146 /* maj */ CDEV_MAJOR 147}; 148 149static struct cdevsw mddisk_cdevsw; 150 151static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(&md_softc_list); 152 153#define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 154#define NMASK (NINDIR-1) 155static int nshift; 156 157struct indir { 158 uintptr_t *array; 159 uint total; 160 uint used; 161 uint shift; 162}; 163 164struct md_s { 165 int unit; 166 LIST_ENTRY(md_s) list; 167 struct devstat stats; 168 struct bio_queue_head bio_queue; 169 struct disk disk; 170 dev_t dev; 171 enum md_types type; 172 unsigned nsect; 173 unsigned opencount; 174 unsigned secsize; 175 unsigned flags; 176 char name[20]; 177 struct proc *procp; 178 179 /* MD_MALLOC related fields */ 180 struct indir *indir; 181 uma_zone_t uma; 182 183 /* MD_PRELOAD related fields */ 184 u_char *pl_ptr; 185 unsigned pl_len; 186 187 /* MD_VNODE related fields */ 188 struct vnode *vnode; 189 struct ucred *cred; 190 191 /* MD_SWAP related fields */ 192 vm_object_t object; 193}; 194 195static int mddestroy(struct md_s *sc, struct thread *td); 196 197static struct indir * 198new_indir(uint shift) 199{ 200 struct indir *ip; 201 202 ip = malloc(sizeof *ip, M_MD, M_NOWAIT | M_ZERO); 203 if (ip == NULL) 204 return(NULL); 205 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 206 M_MDSECT, M_NOWAIT | M_ZERO); 207 if (ip->array == NULL) { 208 free(ip, M_MD); 209 return(NULL); 210 } 211 ip->total = NINDIR; 212 ip->shift = shift; 213 return(ip); 214} 215 216static void 217del_indir(struct indir *ip) 218{ 219 220 free(ip->array, M_MDSECT); 221 free(ip, M_MD); 222} 223 224static void 225destroy_indir(struct md_s *sc, struct indir *ip) 226{ 227 int i; 228 229 for (i = 0; i < NINDIR; i++) { 230 if (!ip->array[i]) 231 continue; 232 if (ip->shift) 233 destroy_indir(sc, (struct indir*)(ip->array[i])); 234 else if (ip->array[i] > 255) 235 uma_zfree(sc->uma, (void *)(ip->array[i])); 236 } 237 del_indir(ip); 238} 239 240 241/* 242 * This function does the math and alloctes the top level "indir" structure 243 * for a device of "size" sectors. 244 */ 245 246static struct indir * 247dimension(off_t size) 248{ 249 off_t rcnt; 250 struct indir *ip; 251 int i, layer; 252 253 rcnt = size; 254 layer = 0; 255 while (rcnt > NINDIR) { 256 rcnt /= NINDIR; 257 layer++; 258 } 259 /* figure out log2(NINDIR) */ 260 for (i = NINDIR, nshift = -1; i; nshift++) 261 i >>= 1; 262 263 /* 264 * XXX: the top layer is probably not fully populated, so we allocate 265 * too much space for ip->array in new_indir() here. 266 */ 267 ip = new_indir(layer * nshift); 268 return (ip); 269} 270 271/* 272 * Read a given sector 273 */ 274 275static uintptr_t 276s_read(struct indir *ip, off_t offset) 277{ 278 struct indir *cip; 279 int idx; 280 uintptr_t up; 281 282 if (md_debug > 1) 283 printf("s_read(%lld)\n", offset); 284 up = 0; 285 for (cip = ip; cip != NULL;) { 286 if (cip->shift) { 287 idx = (offset >> cip->shift) & NMASK; 288 up = cip->array[idx]; 289 cip = (struct indir *)up; 290 continue; 291 } 292 idx = offset & NMASK; 293 return(cip->array[idx]); 294 } 295 return (0); 296} 297 298/* 299 * Write a given sector, prune the tree if the value is 0 300 */ 301 302static int 303s_write(struct indir *ip, off_t offset, uintptr_t ptr) 304{ 305 struct indir *cip, *lip[10]; 306 int idx, li; 307 uintptr_t up; 308 309 if (md_debug > 1) 310 printf("s_write(%lld, %p)\n", offset, (void *)ptr); 311 up = 0; 312 li = 0; 313 cip = ip; 314 for (;;) { 315 lip[li++] = cip; 316 if (cip->shift) { 317 idx = (offset >> cip->shift) & NMASK; 318 up = cip->array[idx]; 319 if (up != 0) { 320 cip = (struct indir *)up; 321 continue; 322 } 323 /* Allocate branch */ 324 cip->array[idx] = 325 (uintptr_t)new_indir(cip->shift - nshift); 326 if (cip->array[idx] == 0) 327 return(ENOMEM); 328 cip->used++; 329 up = cip->array[idx]; 330 cip = (struct indir *)up; 331 continue; 332 } 333 /* leafnode */ 334 idx = offset & NMASK; 335 up = cip->array[idx]; 336 if (up != 0) 337 cip->used--; 338 cip->array[idx] = ptr; 339 if (ptr != 0) 340 cip->used++; 341 break; 342 } 343 if (cip->used != 0 || li == 1) 344 return (0); 345 li--; 346 while (cip->used == 0 && cip != ip) { 347 li--; 348 idx = (offset >> lip[li]->shift) & NMASK; 349 up = lip[li]->array[idx]; 350 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 351 del_indir(cip); 352 lip[li]->array[idx] = NULL; 353 lip[li]->used--; 354 cip = lip[li]; 355 } 356 return (0); 357} 358 359static int 360mdopen(dev_t dev, int flag, int fmt, struct thread *td) 361{ 362 struct md_s *sc; 363 struct disklabel *dl; 364 365 if (md_debug) 366 printf("mdopen(%p %x %x %p)\n", 367 devtoname(dev), flag, fmt, td); 368 369 sc = dev->si_drv1; 370 371 dl = &sc->disk.d_label; 372 bzero(dl, sizeof(*dl)); 373 dl->d_secsize = sc->secsize; 374 dl->d_nsectors = sc->nsect > 63 ? 63 : sc->nsect; 375 dl->d_ntracks = 1; 376 dl->d_secpercyl = dl->d_nsectors * dl->d_ntracks; 377 dl->d_secperunit = sc->nsect; 378 dl->d_ncylinders = dl->d_secperunit / dl->d_secpercyl; 379 sc->opencount++; 380 return (0); 381} 382 383static int 384mdclose(dev_t dev, int flags, int fmt, struct thread *td) 385{ 386 struct md_s *sc = dev->si_drv1; 387 388 sc->opencount--; 389 return (0); 390} 391 392static int 393mdioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 394{ 395 396 if (md_debug) 397 printf("mdioctl(%s %lx %p %x %p)\n", 398 devtoname(dev), cmd, addr, flags, td); 399 400 return (ENOIOCTL); 401} 402 403static int 404mdstart_malloc(struct md_s *sc, struct bio *bp) 405{ 406 int i, error; 407 u_char *dst; 408 unsigned secno, nsec, uc; 409 uintptr_t sp, osp; 410 411 nsec = bp->bio_bcount / sc->secsize; 412 secno = bp->bio_pblkno; 413 dst = bp->bio_data; 414 error = 0; 415 while (nsec--) { 416 osp = s_read(sc->indir, secno); 417 if (bp->bio_cmd == BIO_DELETE) { 418 if (osp != 0) 419 error = s_write(sc->indir, secno, 0); 420 } else if (bp->bio_cmd == BIO_READ) { 421 if (osp == 0) 422 bzero(dst, sc->secsize); 423 else if (osp <= 255) 424 for (i = 0; i < sc->secsize; i++) 425 dst[i] = osp; 426 else 427 bcopy((void *)osp, dst, sc->secsize); 428 osp = 0; 429 } else if (bp->bio_cmd == BIO_WRITE) { 430 if (sc->flags & MD_COMPRESS) { 431 uc = dst[0]; 432 for (i = 1; i < sc->secsize; i++) 433 if (dst[i] != uc) 434 break; 435 } else { 436 i = 0; 437 uc = 0; 438 } 439 if (i == sc->secsize) { 440 if (osp != uc) 441 error = s_write(sc->indir, secno, uc); 442 } else { 443 if (osp <= 255) { 444 sp = (uintptr_t) uma_zalloc( 445 sc->uma, M_NOWAIT); 446 if (sp == 0) { 447 error = ENOSPC; 448 break; 449 } 450 bcopy(dst, (void *)sp, sc->secsize); 451 error = s_write(sc->indir, secno, sp); 452 } else { 453 bcopy(dst, (void *)osp, sc->secsize); 454 osp = 0; 455 } 456 } 457 } else { 458 error = EOPNOTSUPP; 459 } 460 if (osp > 255) 461 uma_zfree(sc->uma, (void*)osp); 462 if (error) 463 break; 464 secno++; 465 dst += sc->secsize; 466 } 467 bp->bio_resid = 0; 468 return (error); 469} 470 471 472static int 473mdstart_preload(struct md_s *sc, struct bio *bp) 474{ 475 476 if (bp->bio_cmd == BIO_DELETE) { 477 } else if (bp->bio_cmd == BIO_READ) { 478 bcopy(sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_data, bp->bio_bcount); 479 } else { 480 bcopy(bp->bio_data, sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_bcount); 481 } 482 bp->bio_resid = 0; 483 return (0); 484} 485 486static int 487mdstart_vnode(struct md_s *sc, struct bio *bp) 488{ 489 int error; 490 struct uio auio; 491 struct iovec aiov; 492 struct mount *mp; 493 494 /* 495 * VNODE I/O 496 * 497 * If an error occurs, we set BIO_ERROR but we do not set 498 * B_INVAL because (for a write anyway), the buffer is 499 * still valid. 500 */ 501 502 bzero(&auio, sizeof(auio)); 503 504 aiov.iov_base = bp->bio_data; 505 aiov.iov_len = bp->bio_bcount; 506 auio.uio_iov = &aiov; 507 auio.uio_iovcnt = 1; 508 auio.uio_offset = (vm_ooffset_t)bp->bio_pblkno * sc->secsize; 509 auio.uio_segflg = UIO_SYSSPACE; 510 if(bp->bio_cmd == BIO_READ) 511 auio.uio_rw = UIO_READ; 512 else 513 auio.uio_rw = UIO_WRITE; 514 auio.uio_resid = bp->bio_bcount; 515 auio.uio_td = curthread; 516 /* 517 * When reading set IO_DIRECT to try to avoid double-caching 518 * the data. When writing IO_DIRECT is not optimal, but we 519 * must set IO_NOWDRAIN to avoid a wdrain deadlock. 520 */ 521 if (bp->bio_cmd == BIO_READ) { 522 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 523 error = VOP_READ(sc->vnode, &auio, IO_DIRECT, sc->cred); 524 } else { 525 (void) vn_start_write(sc->vnode, &mp, V_WAIT); 526 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 527 error = VOP_WRITE(sc->vnode, &auio, IO_NOWDRAIN, sc->cred); 528 vn_finished_write(mp); 529 } 530 VOP_UNLOCK(sc->vnode, 0, curthread); 531 bp->bio_resid = auio.uio_resid; 532 return (error); 533} 534 535static int 536mdstart_swap(struct md_s *sc, struct bio *bp) 537{ 538 539 if ((bp->bio_cmd == BIO_DELETE) && (sc->flags & MD_RESERVE)) 540 biodone(bp); 541 else 542 vm_pager_strategy(sc->object, bp); 543 return (-1); 544} 545 546static void 547mdstrategy(struct bio *bp) 548{ 549 struct md_s *sc; 550 551 if (md_debug > 1) 552 printf("mdstrategy(%p) %s %x, %lld, %ld, %p)\n", 553 (void *)bp, devtoname(bp->bio_dev), bp->bio_flags, 554 (long long)bp->bio_blkno, bp->bio_bcount / DEV_BSIZE, 555 (void *)bp->bio_data); 556 557 sc = bp->bio_dev->si_drv1; 558 559 /* XXX: LOCK(sc->lock) */ 560 bioqdisksort(&sc->bio_queue, bp); 561 /* XXX: UNLOCK(sc->lock) */ 562 563 wakeup(sc); 564} 565 566static void 567md_kthread(void *arg) 568{ 569 struct md_s *sc; 570 struct bio *bp; 571 int error; 572 573 sc = arg; 574 curthread->td_base_pri = PRIBIO; 575 576 mtx_lock(&Giant); 577 for (;;) { 578 /* XXX: LOCK(unique unit numbers) */ 579 bp = bioq_first(&sc->bio_queue); 580 if (bp) 581 bioq_remove(&sc->bio_queue, bp); 582 /* XXX: UNLOCK(unique unit numbers) */ 583 if (!bp) { 584 tsleep(sc, PRIBIO, "mdwait", 0); 585 if (sc->flags & MD_SHUTDOWN) { 586 sc->procp = NULL; 587 wakeup(&sc->procp); 588 kthread_exit(0); 589 } 590 continue; 591 } 592 593 switch (sc->type) { 594 case MD_MALLOC: 595 devstat_start_transaction(&sc->stats); 596 error = mdstart_malloc(sc, bp); 597 break; 598 case MD_PRELOAD: 599 devstat_start_transaction(&sc->stats); 600 error = mdstart_preload(sc, bp); 601 break; 602 case MD_VNODE: 603 devstat_start_transaction(&sc->stats); 604 error = mdstart_vnode(sc, bp); 605 break; 606 case MD_SWAP: 607 error = mdstart_swap(sc, bp); 608 break; 609 default: 610 panic("Impossible md(type)"); 611 break; 612 } 613 614 if (error != -1) 615 biofinish(bp, &sc->stats, error); 616 } 617} 618 619static struct md_s * 620mdfind(int unit) 621{ 622 struct md_s *sc; 623 624 /* XXX: LOCK(unique unit numbers) */ 625 LIST_FOREACH(sc, &md_softc_list, list) { 626 if (sc->unit == unit) 627 break; 628 } 629 /* XXX: UNLOCK(unique unit numbers) */ 630 return (sc); 631} 632 633static struct md_s * 634mdnew(int unit) 635{ 636 struct md_s *sc; 637 int error, max = -1; 638 639 /* XXX: LOCK(unique unit numbers) */ 640 LIST_FOREACH(sc, &md_softc_list, list) { 641 if (sc->unit == unit) { 642 /* XXX: UNLOCK(unique unit numbers) */ 643 return (NULL); 644 } 645 if (sc->unit > max) 646 max = sc->unit; 647 } 648 if (unit == -1) 649 unit = max + 1; 650 if (unit > DKMAXUNIT) 651 return (NULL); 652 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 653 sc->unit = unit; 654 sprintf(sc->name, "md%d", unit); 655 error = kthread_create(md_kthread, sc, &sc->procp, 0, "%s", sc->name); 656 if (error) { 657 free(sc, M_MD); 658 return (NULL); 659 } 660 LIST_INSERT_HEAD(&md_softc_list, sc, list); 661 /* XXX: UNLOCK(unique unit numbers) */ 662 return (sc); 663} 664 665static void 666mdinit(struct md_s *sc) 667{ 668 669 bioq_init(&sc->bio_queue); 670 devstat_add_entry(&sc->stats, MD_NAME, sc->unit, sc->secsize, 671 DEVSTAT_NO_ORDERED_TAGS, 672 DEVSTAT_TYPE_DIRECT | DEVSTAT_TYPE_IF_OTHER, 673 DEVSTAT_PRIORITY_OTHER); 674 sc->dev = disk_create(sc->unit, &sc->disk, 0, &md_cdevsw, &mddisk_cdevsw); 675 sc->dev->si_drv1 = sc; 676} 677 678/* 679 * XXX: we should check that the range they feed us is mapped. 680 * XXX: we should implement read-only. 681 */ 682 683static int 684mdcreate_preload(struct md_ioctl *mdio) 685{ 686 struct md_s *sc; 687 688 if (mdio->md_size == 0) 689 return (EINVAL); 690 if (mdio->md_options & ~(MD_AUTOUNIT)) 691 return (EINVAL); 692 if (mdio->md_options & MD_AUTOUNIT) { 693 sc = mdnew(-1); 694 if (sc == NULL) 695 return (ENOMEM); 696 mdio->md_unit = sc->unit; 697 } else { 698 sc = mdnew(mdio->md_unit); 699 if (sc == NULL) 700 return (EBUSY); 701 } 702 sc->type = MD_PRELOAD; 703 sc->secsize = DEV_BSIZE; 704 sc->nsect = mdio->md_size; 705 sc->flags = mdio->md_options & MD_FORCE; 706 /* Cast to pointer size, then to pointer to avoid warning */ 707 sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base; 708 sc->pl_len = (mdio->md_size << DEV_BSHIFT); 709 mdinit(sc); 710 return (0); 711} 712 713 714static int 715mdcreate_malloc(struct md_ioctl *mdio) 716{ 717 struct md_s *sc; 718 off_t u; 719 uintptr_t sp; 720 int error; 721 722 error = 0; 723 if (mdio->md_size == 0) 724 return (EINVAL); 725 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 726 return (EINVAL); 727 /* Compression doesn't make sense if we have reserved space */ 728 if (mdio->md_options & MD_RESERVE) 729 mdio->md_options &= ~MD_COMPRESS; 730 if (mdio->md_options & MD_AUTOUNIT) { 731 sc = mdnew(-1); 732 if (sc == NULL) 733 return (ENOMEM); 734 mdio->md_unit = sc->unit; 735 } else { 736 sc = mdnew(mdio->md_unit); 737 if (sc == NULL) 738 return (EBUSY); 739 } 740 sc->type = MD_MALLOC; 741 sc->secsize = DEV_BSIZE; 742 sc->nsect = mdio->md_size; 743 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 744 sc->indir = dimension(sc->nsect); 745 sc->uma = uma_zcreate(sc->name, sc->secsize, 746 NULL, NULL, NULL, NULL, 0x1ff, 0); 747 if (mdio->md_options & MD_RESERVE) { 748 for (u = 0; u < sc->nsect; u++) { 749 sp = (uintptr_t) uma_zalloc(sc->uma, M_NOWAIT | M_ZERO); 750 if (sp != 0) 751 error = s_write(sc->indir, u, sp); 752 else 753 error = ENOMEM; 754 if (error) 755 break; 756 } 757 } 758 if (!error) { 759 printf("%s%d: Malloc disk\n", MD_NAME, sc->unit); 760 mdinit(sc); 761 } else 762 mddestroy(sc, NULL); 763 return (error); 764} 765 766 767static int 768mdsetcred(struct md_s *sc, struct ucred *cred) 769{ 770 char *tmpbuf; 771 int error = 0; 772 773 /* 774 * Set credits in our softc 775 */ 776 777 if (sc->cred) 778 crfree(sc->cred); 779 sc->cred = crhold(cred); 780 781 /* 782 * Horrible kludge to establish credentials for NFS XXX. 783 */ 784 785 if (sc->vnode) { 786 struct uio auio; 787 struct iovec aiov; 788 789 tmpbuf = malloc(sc->secsize, M_TEMP, M_WAITOK); 790 bzero(&auio, sizeof(auio)); 791 792 aiov.iov_base = tmpbuf; 793 aiov.iov_len = sc->secsize; 794 auio.uio_iov = &aiov; 795 auio.uio_iovcnt = 1; 796 auio.uio_offset = 0; 797 auio.uio_rw = UIO_READ; 798 auio.uio_segflg = UIO_SYSSPACE; 799 auio.uio_resid = aiov.iov_len; 800 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 801 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 802 VOP_UNLOCK(sc->vnode, 0, curthread); 803 free(tmpbuf, M_TEMP); 804 } 805 return (error); 806} 807 808static int 809mdcreate_vnode(struct md_ioctl *mdio, struct thread *td) 810{ 811 struct md_s *sc; 812 struct vattr vattr; 813 struct nameidata nd; 814 int error, flags; 815 816 flags = FREAD|FWRITE; 817 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 818 error = vn_open(&nd, &flags, 0); 819 if (error) { 820 if (error != EACCES && error != EPERM && error != EROFS) 821 return (error); 822 flags &= ~FWRITE; 823 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 824 error = vn_open(&nd, &flags, 0); 825 if (error) 826 return (error); 827 } 828 NDFREE(&nd, NDF_ONLY_PNBUF); 829 if (nd.ni_vp->v_type != VREG || 830 (error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred, td))) { 831 VOP_UNLOCK(nd.ni_vp, 0, td); 832 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 833 return (error ? error : EINVAL); 834 } 835 VOP_UNLOCK(nd.ni_vp, 0, td); 836 837 if (mdio->md_options & MD_AUTOUNIT) { 838 sc = mdnew(-1); 839 mdio->md_unit = sc->unit; 840 } else { 841 sc = mdnew(mdio->md_unit); 842 } 843 if (sc == NULL) { 844 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 845 return (EBUSY); 846 } 847 848 sc->type = MD_VNODE; 849 sc->flags = mdio->md_options & MD_FORCE; 850 if (!(flags & FWRITE)) 851 sc->flags |= MD_READONLY; 852 sc->secsize = DEV_BSIZE; 853 sc->vnode = nd.ni_vp; 854 855 /* 856 * If the size is specified, override the file attributes. 857 */ 858 if (mdio->md_size) 859 sc->nsect = mdio->md_size; 860 else 861 sc->nsect = vattr.va_size / sc->secsize; /* XXX: round up ? */ 862 if (sc->nsect == 0) { 863 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 864 return (EINVAL); 865 } 866 error = mdsetcred(sc, td->td_ucred); 867 if (error) { 868 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 869 return (error); 870 } 871 mdinit(sc); 872 return (0); 873} 874 875static int 876mddestroy(struct md_s *sc, struct thread *td) 877{ 878 879 GIANT_REQUIRED; 880 881 if (sc->dev != NULL) { 882 devstat_remove_entry(&sc->stats); 883 disk_destroy(sc->dev); 884 } 885 sc->flags |= MD_SHUTDOWN; 886 wakeup(sc); 887 while (sc->procp != NULL) 888 tsleep(&sc->procp, PRIBIO, "mddestroy", hz / 10); 889 if (sc->vnode != NULL) 890 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 891 FREAD : (FREAD|FWRITE), sc->cred, td); 892 if (sc->cred != NULL) 893 crfree(sc->cred); 894 if (sc->object != NULL) { 895 vm_pager_deallocate(sc->object); 896 } 897 if (sc->indir) 898 destroy_indir(sc, sc->indir); 899 if (sc->uma) 900 uma_zdestroy(sc->uma); 901 902 /* XXX: LOCK(unique unit numbers) */ 903 LIST_REMOVE(sc, list); 904 /* XXX: UNLOCK(unique unit numbers) */ 905 free(sc, M_MD); 906 return (0); 907} 908 909static int 910mdcreate_swap(struct md_ioctl *mdio, struct thread *td) 911{ 912 int error; 913 struct md_s *sc; 914 915 GIANT_REQUIRED; 916 917 if (mdio->md_options & MD_AUTOUNIT) { 918 sc = mdnew(-1); 919 mdio->md_unit = sc->unit; 920 } else { 921 sc = mdnew(mdio->md_unit); 922 } 923 if (sc == NULL) 924 return (EBUSY); 925 926 sc->type = MD_SWAP; 927 928 /* 929 * Range check. Disallow negative sizes or any size less then the 930 * size of a page. Then round to a page. 931 */ 932 933 if (mdio->md_size == 0) { 934 mddestroy(sc, td); 935 return (EDOM); 936 } 937 938 /* 939 * Allocate an OBJT_SWAP object. 940 * 941 * sc_secsize is PAGE_SIZE'd 942 * 943 * mdio->size is in DEV_BSIZE'd chunks. 944 * Note the truncation. 945 */ 946 947 sc->secsize = PAGE_SIZE; 948 sc->nsect = mdio->md_size / (PAGE_SIZE / DEV_BSIZE); 949 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, sc->secsize * (vm_offset_t)sc->nsect, VM_PROT_DEFAULT, 0); 950 sc->flags = mdio->md_options & MD_FORCE; 951 if (mdio->md_options & MD_RESERVE) { 952 if (swap_pager_reserve(sc->object, 0, sc->nsect) < 0) { 953 vm_pager_deallocate(sc->object); 954 sc->object = NULL; 955 mddestroy(sc, td); 956 return (EDOM); 957 } 958 } 959 error = mdsetcred(sc, td->td_ucred); 960 if (error) 961 mddestroy(sc, td); 962 else 963 mdinit(sc); 964 return (error); 965} 966 967static int 968mddetach(int unit, struct thread *td) 969{ 970 struct md_s *sc; 971 972 sc = mdfind(unit); 973 if (sc == NULL) 974 return (ENOENT); 975 if (sc->opencount != 0 && !(sc->flags & MD_FORCE)) 976 return (EBUSY); 977 switch(sc->type) { 978 case MD_VNODE: 979 case MD_SWAP: 980 case MD_MALLOC: 981 case MD_PRELOAD: 982 return (mddestroy(sc, td)); 983 default: 984 return (EOPNOTSUPP); 985 } 986} 987 988static int 989mdctlioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 990{ 991 struct md_ioctl *mdio; 992 struct md_s *sc; 993 994 if (md_debug) 995 printf("mdctlioctl(%s %lx %p %x %p)\n", 996 devtoname(dev), cmd, addr, flags, td); 997 998 /* 999 * We assert the version number in the individual ioctl 1000 * handlers instead of out here because (a) it is possible we 1001 * may add another ioctl in the future which doesn't read an 1002 * mdio, and (b) the correct return value for an unknown ioctl 1003 * is ENOIOCTL, not EINVAL. 1004 */ 1005 mdio = (struct md_ioctl *)addr; 1006 switch (cmd) { 1007 case MDIOCATTACH: 1008 if (mdio->md_version != MDIOVERSION) 1009 return (EINVAL); 1010 switch (mdio->md_type) { 1011 case MD_MALLOC: 1012 return (mdcreate_malloc(mdio)); 1013 case MD_PRELOAD: 1014 return (mdcreate_preload(mdio)); 1015 case MD_VNODE: 1016 return (mdcreate_vnode(mdio, td)); 1017 case MD_SWAP: 1018 return (mdcreate_swap(mdio, td)); 1019 default: 1020 return (EINVAL); 1021 } 1022 case MDIOCDETACH: 1023 if (mdio->md_version != MDIOVERSION) 1024 return (EINVAL); 1025 if (mdio->md_file != NULL || mdio->md_size != 0 || 1026 mdio->md_options != 0) 1027 return (EINVAL); 1028 return (mddetach(mdio->md_unit, td)); 1029 case MDIOCQUERY: 1030 if (mdio->md_version != MDIOVERSION) 1031 return (EINVAL); 1032 sc = mdfind(mdio->md_unit); 1033 if (sc == NULL) 1034 return (ENOENT); 1035 mdio->md_type = sc->type; 1036 mdio->md_options = sc->flags; 1037 switch (sc->type) { 1038 case MD_MALLOC: 1039 mdio->md_size = sc->nsect; 1040 break; 1041 case MD_PRELOAD: 1042 mdio->md_size = sc->nsect; 1043 (u_char *)(uintptr_t)mdio->md_base = sc->pl_ptr; 1044 break; 1045 case MD_SWAP: 1046 mdio->md_size = sc->nsect * (PAGE_SIZE / DEV_BSIZE); 1047 break; 1048 case MD_VNODE: 1049 mdio->md_size = sc->nsect; 1050 /* XXX fill this in */ 1051 mdio->md_file = NULL; 1052 break; 1053 } 1054 return (0); 1055 default: 1056 return (ENOIOCTL); 1057 }; 1058 return (ENOIOCTL); 1059} 1060 1061static void 1062md_preloaded(u_char *image, unsigned length) 1063{ 1064 struct md_s *sc; 1065 1066 sc = mdnew(-1); 1067 if (sc == NULL) 1068 return; 1069 sc->type = MD_PRELOAD; 1070 sc->secsize = DEV_BSIZE; 1071 sc->nsect = length / DEV_BSIZE; 1072 sc->pl_ptr = image; 1073 sc->pl_len = length; 1074 if (sc->unit == 0) 1075 mdrootready = 1; 1076 mdinit(sc); 1077} 1078 1079static void 1080md_drvinit(void *unused) 1081{ 1082 1083 caddr_t mod; 1084 caddr_t c; 1085 u_char *ptr, *name, *type; 1086 unsigned len; 1087 1088#ifdef MD_ROOT_SIZE 1089 md_preloaded(mfs_root, MD_ROOT_SIZE*1024); 1090#endif 1091 mod = NULL; 1092 while ((mod = preload_search_next_name(mod)) != NULL) { 1093 name = (char *)preload_search_info(mod, MODINFO_NAME); 1094 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1095 if (name == NULL) 1096 continue; 1097 if (type == NULL) 1098 continue; 1099 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1100 continue; 1101 c = preload_search_info(mod, MODINFO_ADDR); 1102 ptr = *(u_char **)c; 1103 c = preload_search_info(mod, MODINFO_SIZE); 1104 len = *(unsigned *)c; 1105 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1106 MD_NAME, mdunits, name, len, ptr); 1107 md_preloaded(ptr, len); 1108 } 1109 status_dev = make_dev(&mdctl_cdevsw, 0xffff00ff, UID_ROOT, GID_WHEEL, 1110 0600, MDCTL_NAME); 1111} 1112 1113static int 1114md_modevent(module_t mod, int type, void *data) 1115{ 1116 int error; 1117 struct md_s *sc; 1118 1119 switch (type) { 1120 case MOD_LOAD: 1121 md_drvinit(NULL); 1122 break; 1123 case MOD_UNLOAD: 1124 LIST_FOREACH(sc, &md_softc_list, list) { 1125 error = mddetach(sc->unit, curthread); 1126 if (error != 0) 1127 return (error); 1128 } 1129 if (status_dev) 1130 destroy_dev(status_dev); 1131 status_dev = 0; 1132 break; 1133 default: 1134 break; 1135 } 1136 return (0); 1137} 1138 1139static moduledata_t md_mod = { 1140 MD_NAME, 1141 md_modevent, 1142 NULL 1143}; 1144DECLARE_MODULE(md, md_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE+CDEV_MAJOR); 1145MODULE_VERSION(md, MD_MODVER); 1146 1147 1148#ifdef MD_ROOT 1149static void 1150md_takeroot(void *junk) 1151{ 1152 if (mdrootready) 1153 rootdevnames[0] = "ufs:/dev/md0c"; 1154} 1155 1156SYSINIT(md_root, SI_SUB_MOUNT_ROOT, SI_ORDER_FIRST, md_takeroot, NULL); 1157#endif 1158