md.c revision 109623
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 109623 2003-01-21 08:56:16Z alfred $ 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/devicestat.h> 68#include <sys/disk.h> 69#include <sys/fcntl.h> 70#include <sys/kernel.h> 71#include <sys/kthread.h> 72#include <sys/linker.h> 73#include <sys/lock.h> 74#include <sys/malloc.h> 75#include <sys/mdioctl.h> 76#include <sys/mutex.h> 77#include <sys/namei.h> 78#include <sys/proc.h> 79#include <sys/queue.h> 80#include <sys/stdint.h> 81#include <sys/sysctl.h> 82#include <sys/vnode.h> 83 84#ifndef NO_GEOM 85#include <geom/geom.h> 86#endif 87 88#include <vm/vm.h> 89#include <vm/vm_object.h> 90#include <vm/vm_page.h> 91#include <vm/vm_pager.h> 92#include <vm/swap_pager.h> 93#include <vm/uma.h> 94 95#define MD_MODVER 1 96 97#define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */ 98 99#ifndef MD_NSECT 100#define MD_NSECT (10000 * 2) 101#endif 102 103static MALLOC_DEFINE(M_MD, "MD disk", "Memory Disk"); 104static MALLOC_DEFINE(M_MDSECT, "MD sectors", "Memory Disk Sectors"); 105 106static int md_debug; 107SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, ""); 108 109#if defined(MD_ROOT) && defined(MD_ROOT_SIZE) 110/* Image gets put here: */ 111static u_char mfs_root[MD_ROOT_SIZE*1024] = "MFS Filesystem goes here"; 112static u_char end_mfs_root[] __unused = "MFS Filesystem had better STOP here"; 113#endif 114 115static int mdrootready; 116static int mdunits; 117static dev_t status_dev = 0; 118 119#define CDEV_MAJOR 95 120 121static d_ioctl_t mdctlioctl; 122 123static struct cdevsw mdctl_cdevsw = { 124 /* open */ nullopen, 125 /* close */ nullclose, 126 /* read */ noread, 127 /* write */ nowrite, 128 /* ioctl */ mdctlioctl, 129 /* poll */ nopoll, 130 /* mmap */ nommap, 131 /* strategy */ nostrategy, 132 /* name */ MD_NAME, 133 /* maj */ CDEV_MAJOR 134}; 135 136 137static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(&md_softc_list); 138 139#define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 140#define NMASK (NINDIR-1) 141static int nshift; 142 143struct indir { 144 uintptr_t *array; 145 uint total; 146 uint used; 147 uint shift; 148}; 149 150struct md_s { 151 int unit; 152 LIST_ENTRY(md_s) list; 153 struct devstat stats; 154 struct bio_queue_head bio_queue; 155 struct mtx queue_mtx; 156 struct disk disk; 157 dev_t dev; 158 enum md_types type; 159 unsigned nsect; 160 unsigned opencount; 161 unsigned secsize; 162 unsigned flags; 163 char name[20]; 164 struct proc *procp; 165#ifndef NO_GEOM 166 struct g_geom *gp; 167 struct g_provider *pp; 168#endif 169 170 /* MD_MALLOC related fields */ 171 struct indir *indir; 172 uma_zone_t uma; 173 174 /* MD_PRELOAD related fields */ 175 u_char *pl_ptr; 176 unsigned pl_len; 177 178 /* MD_VNODE related fields */ 179 struct vnode *vnode; 180 struct ucred *cred; 181 182 /* MD_SWAP related fields */ 183 vm_object_t object; 184}; 185 186static int mddestroy(struct md_s *sc, struct thread *td); 187 188static struct indir * 189new_indir(uint shift) 190{ 191 struct indir *ip; 192 193 ip = malloc(sizeof *ip, M_MD, M_NOWAIT | M_ZERO); 194 if (ip == NULL) 195 return (NULL); 196 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 197 M_MDSECT, M_NOWAIT | M_ZERO); 198 if (ip->array == NULL) { 199 free(ip, M_MD); 200 return (NULL); 201 } 202 ip->total = NINDIR; 203 ip->shift = shift; 204 return (ip); 205} 206 207static void 208del_indir(struct indir *ip) 209{ 210 211 free(ip->array, M_MDSECT); 212 free(ip, M_MD); 213} 214 215static void 216destroy_indir(struct md_s *sc, struct indir *ip) 217{ 218 int i; 219 220 for (i = 0; i < NINDIR; i++) { 221 if (!ip->array[i]) 222 continue; 223 if (ip->shift) 224 destroy_indir(sc, (struct indir*)(ip->array[i])); 225 else if (ip->array[i] > 255) 226 uma_zfree(sc->uma, (void *)(ip->array[i])); 227 } 228 del_indir(ip); 229} 230 231/* 232 * This function does the math and alloctes the top level "indir" structure 233 * for a device of "size" sectors. 234 */ 235 236static struct indir * 237dimension(off_t size) 238{ 239 off_t rcnt; 240 struct indir *ip; 241 int i, layer; 242 243 rcnt = size; 244 layer = 0; 245 while (rcnt > NINDIR) { 246 rcnt /= NINDIR; 247 layer++; 248 } 249 /* figure out log2(NINDIR) */ 250 for (i = NINDIR, nshift = -1; i; nshift++) 251 i >>= 1; 252 253 /* 254 * XXX: the top layer is probably not fully populated, so we allocate 255 * too much space for ip->array in new_indir() here. 256 */ 257 ip = new_indir(layer * nshift); 258 return (ip); 259} 260 261/* 262 * Read a given sector 263 */ 264 265static uintptr_t 266s_read(struct indir *ip, off_t offset) 267{ 268 struct indir *cip; 269 int idx; 270 uintptr_t up; 271 272 if (md_debug > 1) 273 printf("s_read(%jd)\n", (intmax_t)offset); 274 up = 0; 275 for (cip = ip; cip != NULL;) { 276 if (cip->shift) { 277 idx = (offset >> cip->shift) & NMASK; 278 up = cip->array[idx]; 279 cip = (struct indir *)up; 280 continue; 281 } 282 idx = offset & NMASK; 283 return (cip->array[idx]); 284 } 285 return (0); 286} 287 288/* 289 * Write a given sector, prune the tree if the value is 0 290 */ 291 292static int 293s_write(struct indir *ip, off_t offset, uintptr_t ptr) 294{ 295 struct indir *cip, *lip[10]; 296 int idx, li; 297 uintptr_t up; 298 299 if (md_debug > 1) 300 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 301 up = 0; 302 li = 0; 303 cip = ip; 304 for (;;) { 305 lip[li++] = cip; 306 if (cip->shift) { 307 idx = (offset >> cip->shift) & NMASK; 308 up = cip->array[idx]; 309 if (up != 0) { 310 cip = (struct indir *)up; 311 continue; 312 } 313 /* Allocate branch */ 314 cip->array[idx] = 315 (uintptr_t)new_indir(cip->shift - nshift); 316 if (cip->array[idx] == 0) 317 return (ENOSPC); 318 cip->used++; 319 up = cip->array[idx]; 320 cip = (struct indir *)up; 321 continue; 322 } 323 /* leafnode */ 324 idx = offset & NMASK; 325 up = cip->array[idx]; 326 if (up != 0) 327 cip->used--; 328 cip->array[idx] = ptr; 329 if (ptr != 0) 330 cip->used++; 331 break; 332 } 333 if (cip->used != 0 || li == 1) 334 return (0); 335 li--; 336 while (cip->used == 0 && cip != ip) { 337 li--; 338 idx = (offset >> lip[li]->shift) & NMASK; 339 up = lip[li]->array[idx]; 340 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 341 del_indir(cip); 342 lip[li]->array[idx] = 0; 343 lip[li]->used--; 344 cip = lip[li]; 345 } 346 return (0); 347} 348 349#ifndef NO_GEOM 350 351struct g_class g_md_class = { 352 "MD", 353 NULL, 354 NULL, 355 G_CLASS_INITIALIZER 356 357}; 358 359static int 360g_md_access(struct g_provider *pp, int r, int w, int e) 361{ 362 struct md_s *sc; 363 364 sc = pp->geom->softc; 365 r += pp->acr; 366 w += pp->acw; 367 e += pp->ace; 368 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 369 sc->opencount = 1; 370 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 371 sc->opencount = 0; 372 } 373 return (0); 374} 375 376static void 377g_md_start(struct bio *bp) 378{ 379 struct md_s *sc; 380 381 sc = bp->bio_to->geom->softc; 382 383 switch(bp->bio_cmd) { 384 case BIO_GETATTR: 385 case BIO_SETATTR: 386 g_io_deliver(bp, EOPNOTSUPP); 387 return; 388 } 389 bp->bio_blkno = bp->bio_offset >> DEV_BSHIFT; 390 bp->bio_pblkno = bp->bio_offset / sc->secsize; 391 bp->bio_bcount = bp->bio_length; 392 mtx_lock(&sc->queue_mtx); 393 bioqdisksort(&sc->bio_queue, bp); 394 mtx_unlock(&sc->queue_mtx); 395 396 wakeup(sc); 397} 398 399DECLARE_GEOM_CLASS(g_md_class, g_md); 400#endif 401 402#ifdef NO_GEOM 403 404static d_strategy_t mdstrategy; 405static d_open_t mdopen; 406static d_close_t mdclose; 407static d_ioctl_t mdioctl; 408 409static struct cdevsw md_cdevsw = { 410 /* open */ mdopen, 411 /* close */ mdclose, 412 /* read */ physread, 413 /* write */ physwrite, 414 /* ioctl */ mdioctl, 415 /* poll */ nopoll, 416 /* mmap */ nommap, 417 /* strategy */ mdstrategy, 418 /* name */ MD_NAME, 419 /* maj */ CDEV_MAJOR, 420 /* dump */ nodump, 421 /* psize */ nopsize, 422 /* flags */ D_DISK | D_CANFREE | D_MEMDISK, 423}; 424 425static struct cdevsw mddisk_cdevsw; 426 427static int 428mdopen(dev_t dev, int flag, int fmt, struct thread *td) 429{ 430 struct md_s *sc; 431 432 if (md_debug) 433 printf("mdopen(%s %x %x %p)\n", 434 devtoname(dev), flag, fmt, td); 435 436 sc = dev->si_drv1; 437 438 sc->disk.d_sectorsize = sc->secsize; 439 sc->disk.d_mediasize = (off_t)sc->nsect * sc->secsize; 440 sc->disk.d_fwsectors = sc->nsect > 63 ? 63 : sc->nsect; 441 sc->disk.d_fwheads = 1; 442 sc->opencount++; 443 return (0); 444} 445 446static int 447mdclose(dev_t dev, int flags, int fmt, struct thread *td) 448{ 449 struct md_s *sc = dev->si_drv1; 450 451 sc->opencount--; 452 return (0); 453} 454 455static int 456mdioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 457{ 458 459 if (md_debug) 460 printf("mdioctl(%s %lx %p %x %p)\n", 461 devtoname(dev), cmd, addr, flags, td); 462 463 return (ENOIOCTL); 464} 465 466static void 467mdstrategy(struct bio *bp) 468{ 469 struct md_s *sc; 470 471 if (md_debug > 1) 472 printf("mdstrategy(%p) %s %x, %jd, %jd %ld, %p)\n", 473 (void *)bp, devtoname(bp->bio_dev), bp->bio_flags, 474 (intmax_t)bp->bio_blkno, 475 (intmax_t)bp->bio_pblkno, 476 bp->bio_bcount / DEV_BSIZE, 477 (void *)bp->bio_data); 478 479 sc = bp->bio_dev->si_drv1; 480 481 mtx_lock(&sc->queue_mtx); 482 bioqdisksort(&sc->bio_queue, bp); 483 mtx_unlock(&sc->queue_mtx); 484 485 wakeup(sc); 486} 487 488#endif /* NO_GEOM */ 489 490static int 491mdstart_malloc(struct md_s *sc, struct bio *bp) 492{ 493 int i, error; 494 u_char *dst; 495 unsigned secno, nsec, uc; 496 uintptr_t sp, osp; 497 498 nsec = bp->bio_bcount / sc->secsize; 499 secno = bp->bio_pblkno; 500 dst = bp->bio_data; 501 error = 0; 502 while (nsec--) { 503 osp = s_read(sc->indir, secno); 504 if (bp->bio_cmd == BIO_DELETE) { 505 if (osp != 0) 506 error = s_write(sc->indir, secno, 0); 507 } else if (bp->bio_cmd == BIO_READ) { 508 if (osp == 0) 509 bzero(dst, sc->secsize); 510 else if (osp <= 255) 511 for (i = 0; i < sc->secsize; i++) 512 dst[i] = osp; 513 else 514 bcopy((void *)osp, dst, sc->secsize); 515 osp = 0; 516 } else if (bp->bio_cmd == BIO_WRITE) { 517 if (sc->flags & MD_COMPRESS) { 518 uc = dst[0]; 519 for (i = 1; i < sc->secsize; i++) 520 if (dst[i] != uc) 521 break; 522 } else { 523 i = 0; 524 uc = 0; 525 } 526 if (i == sc->secsize) { 527 if (osp != uc) 528 error = s_write(sc->indir, secno, uc); 529 } else { 530 if (osp <= 255) { 531 sp = (uintptr_t) uma_zalloc( 532 sc->uma, M_NOWAIT); 533 if (sp == 0) { 534 error = ENOSPC; 535 break; 536 } 537 bcopy(dst, (void *)sp, sc->secsize); 538 error = s_write(sc->indir, secno, sp); 539 } else { 540 bcopy(dst, (void *)osp, sc->secsize); 541 osp = 0; 542 } 543 } 544 } else { 545 error = EOPNOTSUPP; 546 } 547 if (osp > 255) 548 uma_zfree(sc->uma, (void*)osp); 549 if (error) 550 break; 551 secno++; 552 dst += sc->secsize; 553 } 554 bp->bio_resid = 0; 555 return (error); 556} 557 558static int 559mdstart_preload(struct md_s *sc, struct bio *bp) 560{ 561 562 if (bp->bio_cmd == BIO_DELETE) { 563 } else if (bp->bio_cmd == BIO_READ) { 564 bcopy(sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_data, bp->bio_bcount); 565 } else { 566 bcopy(bp->bio_data, sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_bcount); 567 } 568 bp->bio_resid = 0; 569 return (0); 570} 571 572static int 573mdstart_vnode(struct md_s *sc, struct bio *bp) 574{ 575 int error; 576 struct uio auio; 577 struct iovec aiov; 578 struct mount *mp; 579 580 /* 581 * VNODE I/O 582 * 583 * If an error occurs, we set BIO_ERROR but we do not set 584 * B_INVAL because (for a write anyway), the buffer is 585 * still valid. 586 */ 587 588 bzero(&auio, sizeof(auio)); 589 590 aiov.iov_base = bp->bio_data; 591 aiov.iov_len = bp->bio_bcount; 592 auio.uio_iov = &aiov; 593 auio.uio_iovcnt = 1; 594 auio.uio_offset = (vm_ooffset_t)bp->bio_pblkno * sc->secsize; 595 auio.uio_segflg = UIO_SYSSPACE; 596 if(bp->bio_cmd == BIO_READ) 597 auio.uio_rw = UIO_READ; 598 else 599 auio.uio_rw = UIO_WRITE; 600 auio.uio_resid = bp->bio_bcount; 601 auio.uio_td = curthread; 602 /* 603 * When reading set IO_DIRECT to try to avoid double-caching 604 * the data. When writing IO_DIRECT is not optimal, but we 605 * must set IO_NOWDRAIN to avoid a wdrain deadlock. 606 */ 607 if (bp->bio_cmd == BIO_READ) { 608 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 609 error = VOP_READ(sc->vnode, &auio, IO_DIRECT, sc->cred); 610 } else { 611 (void) vn_start_write(sc->vnode, &mp, V_WAIT); 612 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 613 error = VOP_WRITE(sc->vnode, &auio, IO_NOWDRAIN, sc->cred); 614 vn_finished_write(mp); 615 } 616 VOP_UNLOCK(sc->vnode, 0, curthread); 617 bp->bio_resid = auio.uio_resid; 618 return (error); 619} 620 621#ifndef NO_GEOM 622static void 623mddone_swap(struct bio *bp) 624{ 625 626 bp->bio_completed = bp->bio_length - bp->bio_resid; 627 g_std_done(bp); 628} 629#endif 630 631static int 632mdstart_swap(struct md_s *sc, struct bio *bp) 633{ 634#ifndef NO_GEOM 635 { 636 struct bio *bp2; 637 638 bp2 = g_clone_bio(bp); 639 bp2->bio_done = mddone_swap; 640 bp2->bio_blkno = bp2->bio_offset >> DEV_BSHIFT; 641 bp2->bio_pblkno = bp2->bio_offset / sc->secsize; 642 bp2->bio_bcount = bp2->bio_length; 643 bp = bp2; 644 } 645#endif 646 647 bp->bio_resid = 0; 648 if ((bp->bio_cmd == BIO_DELETE) && (sc->flags & MD_RESERVE)) 649 biodone(bp); 650 else 651 vm_pager_strategy(sc->object, bp); 652 return (-1); 653} 654 655static void 656md_kthread(void *arg) 657{ 658 struct md_s *sc; 659 struct bio *bp; 660 int error, hasgiant; 661 662 sc = arg; 663 curthread->td_base_pri = PRIBIO; 664 665 switch (sc->type) { 666 case MD_SWAP: 667 case MD_VNODE: 668 mtx_lock(&Giant); 669 hasgiant = 1; 670 break; 671 case MD_MALLOC: 672 case MD_PRELOAD: 673 default: 674 hasgiant = 0; 675 break; 676 } 677 678 for (;;) { 679 mtx_lock(&sc->queue_mtx); 680 bp = bioq_first(&sc->bio_queue); 681 if (bp) 682 bioq_remove(&sc->bio_queue, bp); 683 if (!bp) { 684 if (sc->flags & MD_SHUTDOWN) { 685 mtx_unlock(&sc->queue_mtx); 686 sc->procp = NULL; 687 wakeup(&sc->procp); 688 if (!hasgiant) 689 mtx_lock(&Giant); 690 kthread_exit(0); 691 } 692 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 693 continue; 694 } 695 mtx_unlock(&sc->queue_mtx); 696 697 switch (sc->type) { 698 case MD_MALLOC: 699 devstat_start_transaction(&sc->stats); 700 error = mdstart_malloc(sc, bp); 701 break; 702 case MD_PRELOAD: 703 devstat_start_transaction(&sc->stats); 704 error = mdstart_preload(sc, bp); 705 break; 706 case MD_VNODE: 707 devstat_start_transaction(&sc->stats); 708 error = mdstart_vnode(sc, bp); 709 break; 710 case MD_SWAP: 711 error = mdstart_swap(sc, bp); 712 break; 713 default: 714 panic("Impossible md(type)"); 715 break; 716 } 717 718 if (error != -1) { 719#ifdef NO_GEOM 720 biofinish(bp, &sc->stats, error); 721#else /* !NO_GEOM */ 722 bp->bio_completed = bp->bio_length; 723 g_io_deliver(bp, error); 724#endif 725 } 726 } 727} 728 729static struct md_s * 730mdfind(int unit) 731{ 732 struct md_s *sc; 733 734 /* XXX: LOCK(unique unit numbers) */ 735 LIST_FOREACH(sc, &md_softc_list, list) { 736 if (sc->unit == unit) 737 break; 738 } 739 /* XXX: UNLOCK(unique unit numbers) */ 740 return (sc); 741} 742 743static struct md_s * 744mdnew(int unit) 745{ 746 struct md_s *sc; 747 int error, max = -1; 748 749 /* XXX: LOCK(unique unit numbers) */ 750 LIST_FOREACH(sc, &md_softc_list, list) { 751 if (sc->unit == unit) { 752 /* XXX: UNLOCK(unique unit numbers) */ 753 return (NULL); 754 } 755 if (sc->unit > max) 756 max = sc->unit; 757 } 758 if (unit == -1) 759 unit = max + 1; 760 if (unit > 255) 761 return (NULL); 762 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_ZERO); 763 sc->unit = unit; 764 bioq_init(&sc->bio_queue); 765 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 766 sprintf(sc->name, "md%d", unit); 767 error = kthread_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 768 if (error) { 769 free(sc, M_MD); 770 return (NULL); 771 } 772 LIST_INSERT_HEAD(&md_softc_list, sc, list); 773 /* XXX: UNLOCK(unique unit numbers) */ 774 return (sc); 775} 776 777static void 778mdinit(struct md_s *sc) 779{ 780 781 devstat_add_entry(&sc->stats, MD_NAME, sc->unit, sc->secsize, 782 DEVSTAT_NO_ORDERED_TAGS, 783 DEVSTAT_TYPE_DIRECT | DEVSTAT_TYPE_IF_OTHER, 784 DEVSTAT_PRIORITY_OTHER); 785#ifdef NO_GEOM 786 sc->dev = disk_create(sc->unit, &sc->disk, 0, &md_cdevsw, &mddisk_cdevsw); 787 sc->dev->si_drv1 = sc; 788#else /* !NO_GEOM */ 789 { 790 struct g_geom *gp; 791 struct g_provider *pp; 792 793 DROP_GIANT(); 794 g_topology_lock(); 795 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 796 gp->start = g_md_start; 797 gp->access = g_md_access; 798 gp->softc = sc; 799 pp = g_new_providerf(gp, "md%d", sc->unit); 800 pp->mediasize = (off_t)sc->nsect * sc->secsize; 801 pp->sectorsize = sc->secsize; 802 sc->gp = gp; 803 sc->pp = pp; 804 g_error_provider(pp, 0); 805 g_topology_unlock(); 806 PICKUP_GIANT(); 807 } 808#endif /* NO_GEOM */ 809} 810 811/* 812 * XXX: we should check that the range they feed us is mapped. 813 * XXX: we should implement read-only. 814 */ 815 816static int 817mdcreate_preload(struct md_ioctl *mdio) 818{ 819 struct md_s *sc; 820 821 if (mdio->md_size == 0) 822 return (EINVAL); 823 if (mdio->md_options & ~(MD_AUTOUNIT)) 824 return (EINVAL); 825 if (mdio->md_options & MD_AUTOUNIT) { 826 sc = mdnew(-1); 827 if (sc == NULL) 828 return (ENOMEM); 829 mdio->md_unit = sc->unit; 830 } else { 831 sc = mdnew(mdio->md_unit); 832 if (sc == NULL) 833 return (EBUSY); 834 } 835 sc->type = MD_PRELOAD; 836 sc->secsize = DEV_BSIZE; 837 sc->nsect = mdio->md_size; 838 sc->flags = mdio->md_options & MD_FORCE; 839 /* Cast to pointer size, then to pointer to avoid warning */ 840 sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base; 841 sc->pl_len = (mdio->md_size << DEV_BSHIFT); 842 mdinit(sc); 843 return (0); 844} 845 846 847static int 848mdcreate_malloc(struct md_ioctl *mdio) 849{ 850 struct md_s *sc; 851 off_t u; 852 uintptr_t sp; 853 int error; 854 855 error = 0; 856 if (mdio->md_size == 0) 857 return (EINVAL); 858 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 859 return (EINVAL); 860 /* Compression doesn't make sense if we have reserved space */ 861 if (mdio->md_options & MD_RESERVE) 862 mdio->md_options &= ~MD_COMPRESS; 863 if (mdio->md_options & MD_AUTOUNIT) { 864 sc = mdnew(-1); 865 if (sc == NULL) 866 return (ENOMEM); 867 mdio->md_unit = sc->unit; 868 } else { 869 sc = mdnew(mdio->md_unit); 870 if (sc == NULL) 871 return (EBUSY); 872 } 873 sc->type = MD_MALLOC; 874 sc->secsize = DEV_BSIZE; 875 sc->nsect = mdio->md_size; 876 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 877 sc->indir = dimension(sc->nsect); 878 sc->uma = uma_zcreate(sc->name, sc->secsize, 879 NULL, NULL, NULL, NULL, 0x1ff, 0); 880 if (mdio->md_options & MD_RESERVE) { 881 for (u = 0; u < sc->nsect; u++) { 882 sp = (uintptr_t) uma_zalloc(sc->uma, M_NOWAIT | M_ZERO); 883 if (sp != 0) 884 error = s_write(sc->indir, u, sp); 885 else 886 error = ENOMEM; 887 if (error) 888 break; 889 } 890 } 891 if (!error) { 892 mdinit(sc); 893 } else 894 mddestroy(sc, NULL); 895 return (error); 896} 897 898 899static int 900mdsetcred(struct md_s *sc, struct ucred *cred) 901{ 902 char *tmpbuf; 903 int error = 0; 904 905 /* 906 * Set credits in our softc 907 */ 908 909 if (sc->cred) 910 crfree(sc->cred); 911 sc->cred = crhold(cred); 912 913 /* 914 * Horrible kludge to establish credentials for NFS XXX. 915 */ 916 917 if (sc->vnode) { 918 struct uio auio; 919 struct iovec aiov; 920 921 tmpbuf = malloc(sc->secsize, M_TEMP, 0); 922 bzero(&auio, sizeof(auio)); 923 924 aiov.iov_base = tmpbuf; 925 aiov.iov_len = sc->secsize; 926 auio.uio_iov = &aiov; 927 auio.uio_iovcnt = 1; 928 auio.uio_offset = 0; 929 auio.uio_rw = UIO_READ; 930 auio.uio_segflg = UIO_SYSSPACE; 931 auio.uio_resid = aiov.iov_len; 932 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 933 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 934 VOP_UNLOCK(sc->vnode, 0, curthread); 935 free(tmpbuf, M_TEMP); 936 } 937 return (error); 938} 939 940static int 941mdcreate_vnode(struct md_ioctl *mdio, struct thread *td) 942{ 943 struct md_s *sc; 944 struct vattr vattr; 945 struct nameidata nd; 946 int error, flags; 947 948 flags = FREAD|FWRITE; 949 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 950 error = vn_open(&nd, &flags, 0); 951 if (error) { 952 if (error != EACCES && error != EPERM && error != EROFS) 953 return (error); 954 flags &= ~FWRITE; 955 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 956 error = vn_open(&nd, &flags, 0); 957 if (error) 958 return (error); 959 } 960 NDFREE(&nd, NDF_ONLY_PNBUF); 961 if (nd.ni_vp->v_type != VREG || 962 (error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred, td))) { 963 VOP_UNLOCK(nd.ni_vp, 0, td); 964 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 965 return (error ? error : EINVAL); 966 } 967 VOP_UNLOCK(nd.ni_vp, 0, td); 968 969 if (mdio->md_options & MD_AUTOUNIT) { 970 sc = mdnew(-1); 971 mdio->md_unit = sc->unit; 972 } else { 973 sc = mdnew(mdio->md_unit); 974 } 975 if (sc == NULL) { 976 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 977 return (EBUSY); 978 } 979 980 sc->type = MD_VNODE; 981 sc->flags = mdio->md_options & MD_FORCE; 982 if (!(flags & FWRITE)) 983 sc->flags |= MD_READONLY; 984 sc->secsize = DEV_BSIZE; 985 sc->vnode = nd.ni_vp; 986 987 /* 988 * If the size is specified, override the file attributes. 989 */ 990 if (mdio->md_size) 991 sc->nsect = mdio->md_size; 992 else 993 sc->nsect = vattr.va_size / sc->secsize; /* XXX: round up ? */ 994 if (sc->nsect == 0) { 995 mddestroy(sc, td); 996 return (EINVAL); 997 } 998 error = mdsetcred(sc, td->td_ucred); 999 if (error) { 1000 mddestroy(sc, td); 1001 return (error); 1002 } 1003 mdinit(sc); 1004 return (0); 1005} 1006 1007static int 1008mddestroy(struct md_s *sc, struct thread *td) 1009{ 1010 1011 GIANT_REQUIRED; 1012 1013 mtx_destroy(&sc->queue_mtx); 1014 devstat_remove_entry(&sc->stats); 1015#ifdef NO_GEOM 1016 if (sc->dev != NULL) 1017 disk_destroy(sc->dev); 1018#else /* !NO_GEOM */ 1019 { 1020 if (sc->gp) { 1021 sc->gp->flags |= G_GEOM_WITHER; 1022 sc->gp->softc = NULL; 1023 } 1024 if (sc->pp) 1025 g_orphan_provider(sc->pp, ENXIO); 1026 } 1027#endif 1028 sc->flags |= MD_SHUTDOWN; 1029 wakeup(sc); 1030 while (sc->procp != NULL) 1031 tsleep(&sc->procp, PRIBIO, "mddestroy", hz / 10); 1032 if (sc->vnode != NULL) 1033 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 1034 FREAD : (FREAD|FWRITE), sc->cred, td); 1035 if (sc->cred != NULL) 1036 crfree(sc->cred); 1037 if (sc->object != NULL) { 1038 vm_pager_deallocate(sc->object); 1039 } 1040 if (sc->indir) 1041 destroy_indir(sc, sc->indir); 1042 if (sc->uma) 1043 uma_zdestroy(sc->uma); 1044 1045 /* XXX: LOCK(unique unit numbers) */ 1046 LIST_REMOVE(sc, list); 1047 /* XXX: UNLOCK(unique unit numbers) */ 1048 free(sc, M_MD); 1049 return (0); 1050} 1051 1052static int 1053mdcreate_swap(struct md_ioctl *mdio, struct thread *td) 1054{ 1055 int error; 1056 struct md_s *sc; 1057 1058 GIANT_REQUIRED; 1059 1060 if (mdio->md_options & MD_AUTOUNIT) { 1061 sc = mdnew(-1); 1062 mdio->md_unit = sc->unit; 1063 } else { 1064 sc = mdnew(mdio->md_unit); 1065 } 1066 if (sc == NULL) 1067 return (EBUSY); 1068 1069 sc->type = MD_SWAP; 1070 1071 /* 1072 * Range check. Disallow negative sizes or any size less then the 1073 * size of a page. Then round to a page. 1074 */ 1075 1076 if (mdio->md_size == 0) { 1077 mddestroy(sc, td); 1078 return (EDOM); 1079 } 1080 1081 /* 1082 * Allocate an OBJT_SWAP object. 1083 * 1084 * sc_secsize is PAGE_SIZE'd 1085 * 1086 * mdio->size is in DEV_BSIZE'd chunks. 1087 * Note the truncation. 1088 */ 1089 1090 sc->secsize = PAGE_SIZE; 1091 sc->nsect = mdio->md_size / (PAGE_SIZE / DEV_BSIZE); 1092 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, sc->secsize * (vm_offset_t)sc->nsect, VM_PROT_DEFAULT, 0); 1093 sc->flags = mdio->md_options & MD_FORCE; 1094 if (mdio->md_options & MD_RESERVE) { 1095 if (swap_pager_reserve(sc->object, 0, sc->nsect) < 0) { 1096 vm_pager_deallocate(sc->object); 1097 sc->object = NULL; 1098 mddestroy(sc, td); 1099 return (EDOM); 1100 } 1101 } 1102 error = mdsetcred(sc, td->td_ucred); 1103 if (error) 1104 mddestroy(sc, td); 1105 else 1106 mdinit(sc); 1107 return (error); 1108} 1109 1110static int 1111mddetach(int unit, struct thread *td) 1112{ 1113 struct md_s *sc; 1114 1115 sc = mdfind(unit); 1116 if (sc == NULL) 1117 return (ENOENT); 1118 if (sc->opencount != 0 && !(sc->flags & MD_FORCE)) 1119 return (EBUSY); 1120 switch(sc->type) { 1121 case MD_VNODE: 1122 case MD_SWAP: 1123 case MD_MALLOC: 1124 case MD_PRELOAD: 1125 return (mddestroy(sc, td)); 1126 default: 1127 return (EOPNOTSUPP); 1128 } 1129} 1130 1131static int 1132mdctlioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1133{ 1134 struct md_ioctl *mdio; 1135 struct md_s *sc; 1136 1137 if (md_debug) 1138 printf("mdctlioctl(%s %lx %p %x %p)\n", 1139 devtoname(dev), cmd, addr, flags, td); 1140 1141 /* 1142 * We assert the version number in the individual ioctl 1143 * handlers instead of out here because (a) it is possible we 1144 * may add another ioctl in the future which doesn't read an 1145 * mdio, and (b) the correct return value for an unknown ioctl 1146 * is ENOIOCTL, not EINVAL. 1147 */ 1148 mdio = (struct md_ioctl *)addr; 1149 switch (cmd) { 1150 case MDIOCATTACH: 1151 if (mdio->md_version != MDIOVERSION) 1152 return (EINVAL); 1153 switch (mdio->md_type) { 1154 case MD_MALLOC: 1155 return (mdcreate_malloc(mdio)); 1156 case MD_PRELOAD: 1157 return (mdcreate_preload(mdio)); 1158 case MD_VNODE: 1159 return (mdcreate_vnode(mdio, td)); 1160 case MD_SWAP: 1161 return (mdcreate_swap(mdio, td)); 1162 default: 1163 return (EINVAL); 1164 } 1165 case MDIOCDETACH: 1166 if (mdio->md_version != MDIOVERSION) 1167 return (EINVAL); 1168 if (mdio->md_file != NULL || mdio->md_size != 0 || 1169 mdio->md_options != 0) 1170 return (EINVAL); 1171 return (mddetach(mdio->md_unit, td)); 1172 case MDIOCQUERY: 1173 if (mdio->md_version != MDIOVERSION) 1174 return (EINVAL); 1175 sc = mdfind(mdio->md_unit); 1176 if (sc == NULL) 1177 return (ENOENT); 1178 mdio->md_type = sc->type; 1179 mdio->md_options = sc->flags; 1180 switch (sc->type) { 1181 case MD_MALLOC: 1182 mdio->md_size = sc->nsect; 1183 break; 1184 case MD_PRELOAD: 1185 mdio->md_size = sc->nsect; 1186 mdio->md_base = (uint64_t)(intptr_t)sc->pl_ptr; 1187 break; 1188 case MD_SWAP: 1189 mdio->md_size = sc->nsect * (PAGE_SIZE / DEV_BSIZE); 1190 break; 1191 case MD_VNODE: 1192 mdio->md_size = sc->nsect; 1193 /* XXX fill this in */ 1194 mdio->md_file = NULL; 1195 break; 1196 } 1197 return (0); 1198 default: 1199 return (ENOIOCTL); 1200 }; 1201 return (ENOIOCTL); 1202} 1203 1204static void 1205md_preloaded(u_char *image, unsigned length) 1206{ 1207 struct md_s *sc; 1208 1209 sc = mdnew(-1); 1210 if (sc == NULL) 1211 return; 1212 sc->type = MD_PRELOAD; 1213 sc->secsize = DEV_BSIZE; 1214 sc->nsect = length / DEV_BSIZE; 1215 sc->pl_ptr = image; 1216 sc->pl_len = length; 1217 if (sc->unit == 0) 1218 mdrootready = 1; 1219 mdinit(sc); 1220} 1221 1222static void 1223md_drvinit(void *unused) 1224{ 1225 1226 caddr_t mod; 1227 caddr_t c; 1228 u_char *ptr, *name, *type; 1229 unsigned len; 1230 1231#ifdef MD_ROOT_SIZE 1232 md_preloaded(mfs_root, MD_ROOT_SIZE*1024); 1233#endif 1234 mod = NULL; 1235 while ((mod = preload_search_next_name(mod)) != NULL) { 1236 name = (char *)preload_search_info(mod, MODINFO_NAME); 1237 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1238 if (name == NULL) 1239 continue; 1240 if (type == NULL) 1241 continue; 1242 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1243 continue; 1244 c = preload_search_info(mod, MODINFO_ADDR); 1245 ptr = *(u_char **)c; 1246 c = preload_search_info(mod, MODINFO_SIZE); 1247 len = *(size_t *)c; 1248 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1249 MD_NAME, mdunits, name, len, ptr); 1250 md_preloaded(ptr, len); 1251 } 1252 status_dev = make_dev(&mdctl_cdevsw, 0xffff00ff, UID_ROOT, GID_WHEEL, 1253 0600, MDCTL_NAME); 1254} 1255 1256static int 1257md_modevent(module_t mod, int type, void *data) 1258{ 1259 int error; 1260 struct md_s *sc; 1261 1262 switch (type) { 1263 case MOD_LOAD: 1264 md_drvinit(NULL); 1265 break; 1266 case MOD_UNLOAD: 1267 LIST_FOREACH(sc, &md_softc_list, list) { 1268 error = mddetach(sc->unit, curthread); 1269 if (error != 0) 1270 return (error); 1271 } 1272 if (status_dev) 1273 destroy_dev(status_dev); 1274 status_dev = 0; 1275 break; 1276 default: 1277 break; 1278 } 1279 return (0); 1280} 1281 1282static moduledata_t md_mod = { 1283 MD_NAME, 1284 md_modevent, 1285 NULL 1286}; 1287DECLARE_MODULE(md, md_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE+CDEV_MAJOR); 1288MODULE_VERSION(md, MD_MODVER); 1289 1290 1291#ifdef MD_ROOT 1292static void 1293md_takeroot(void *junk) 1294{ 1295 if (mdrootready) 1296 rootdevnames[0] = "ufs:/dev/md0"; 1297} 1298 1299SYSINIT(md_root, SI_SUB_MOUNT_ROOT, SI_ORDER_FIRST, md_takeroot, NULL); 1300#endif /* MD_ROOT */ 1301