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 * $FreeBSD$ 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 98 static MALLOC_DEFINE(M_MD, "MD disk", "Memory Disk"); 99 static MALLOC_DEFINE(M_MDSECT, "MD sectors", "Memory Disk Sectors"); 100 101 static int md_debug; 102 SYSCTL_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: */ 106 static u_char mfs_root[MD_ROOT_SIZE*1024] = "MFS Filesystem goes here"; 107 static u_char end_mfs_root[] __unused = "MFS Filesystem had better STOP here"; 108 #endif 109 110 static int mdrootready; 111 static int mdunits; 112 static dev_t status_dev = 0; 113 114 #define CDEV_MAJOR 95 115 116 static d_ioctl_t mdctlioctl; 117 118 static struct cdevsw mdctl_cdevsw = { 119 .d_open = nullopen, 120 .d_close = nullclose, 121 .d_ioctl = mdctlioctl, 122 .d_name = MD_NAME, 123 .d_maj = CDEV_MAJOR 124 }; 125 126 127 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(&md_softc_list); 128 129 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 130 #define NMASK (NINDIR-1) 131 static int nshift; 132 133 struct indir { 134 uintptr_t *array; 135 uint total; 136 uint used; 137 uint shift; 138 }; 139 140 struct md_s { 141 int unit; 142 LIST_ENTRY(md_s) list; 143 struct bio_queue_head bio_queue; 144 struct mtx queue_mtx; 145 dev_t dev; 146 enum md_types type; 147 unsigned nsect; 148 unsigned opencount; 149 unsigned secsize; 150 unsigned fwheads; 151 unsigned fwsectors; 152 unsigned flags; 153 char name[20]; 154 struct proc *procp; 155 struct g_geom *gp; 156 struct g_provider *pp; 157 158 /* MD_MALLOC related fields */ 159 struct indir *indir; 160 uma_zone_t uma; 161 162 /* MD_PRELOAD related fields */ 163 u_char *pl_ptr; 164 unsigned pl_len; 165 166 /* MD_VNODE related fields */ 167 struct vnode *vnode; 168 struct ucred *cred; 169 170 /* MD_SWAP related fields */ 171 vm_object_t object; 172 }; 173 174 static int mddestroy(struct md_s *sc, struct thread *td); 175 176 static struct indir * 177 new_indir(uint 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 } 194 195 static void 196 del_indir(struct indir *ip) 197 { 198 199 free(ip->array, M_MDSECT); 200 free(ip, M_MD); 201 } 202 203 static void 204 destroy_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 */ 223 224 static struct indir * 225 dimension(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 */ 256 257 static uintptr_t 258 s_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 */ 283 284 static int 285 s_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 } 340 341 342 struct g_class g_md_class = { 343 .name = "MD", 344 }; 345 346 static int 347 g_md_access(struct g_provider *pp, int r, int w, int e) 348 { 349 struct md_s *sc; 350 351 sc = pp->geom->softc; 352 if (sc == NULL) 353 return (ENXIO); 354 r += pp->acr; 355 w += pp->acw; 356 e += pp->ace; 357 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 358 sc->opencount = 1; 359 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 360 sc->opencount = 0; 361 } 362 return (0); 363 } 364 365 static void 366 g_md_start(struct bio *bp) 367 { 368 struct md_s *sc; 369 370 sc = bp->bio_to->geom->softc; 371 372 bp->bio_blkno = bp->bio_offset >> DEV_BSHIFT; 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 } 381 382 DECLARE_GEOM_CLASS(g_md_class, g_md); 383 384 385 static int 386 mdstart_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 } 452 453 static int 454 mdstart_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 } 466 467 static int 468 mdstart_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 494 auio.uio_rw = UIO_WRITE; 495 auio.uio_resid = bp->bio_bcount; 496 auio.uio_td = curthread; 497 /* 498 * When reading set IO_DIRECT to try to avoid double-caching 499 * the data. When writing IO_DIRECT is not optimal. 500 */ 501 if (bp->bio_cmd == BIO_READ) { 502 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 503 error = VOP_READ(sc->vnode, &auio, IO_DIRECT, sc->cred); 504 } else { 505 (void) vn_start_write(sc->vnode, &mp, V_WAIT); 506 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 507 error = VOP_WRITE(sc->vnode, &auio, 0, sc->cred); 508 vn_finished_write(mp); 509 } 510 VOP_UNLOCK(sc->vnode, 0, curthread); 511 bp->bio_resid = auio.uio_resid; 512 return (error); 513 } 514 515 static void 516 mddone_swap(struct bio *bp) 517 { 518 519 bp->bio_completed = bp->bio_length - bp->bio_resid; 520 g_std_done(bp); 521 } 522 523 static int 524 mdstart_swap(struct md_s *sc, struct bio *bp) 525 { 526 { 527 struct bio *bp2; 528 529 bp2 = g_clone_bio(bp); 530 bp2->bio_done = mddone_swap; 531 bp2->bio_blkno = bp2->bio_offset >> DEV_BSHIFT; 532 bp2->bio_pblkno = bp2->bio_offset / sc->secsize; 533 bp2->bio_bcount = bp2->bio_length; 534 bp = bp2; 535 } 536 537 bp->bio_resid = 0; 538 if ((bp->bio_cmd == BIO_DELETE) && (sc->flags & MD_RESERVE)) 539 biodone(bp); 540 else 541 vm_pager_strategy(sc->object, bp); 542 return (-1); 543 } 544 545 static void 546 md_kthread(void *arg) 547 { 548 struct md_s *sc; 549 struct bio *bp; 550 int error, hasgiant; 551 552 sc = arg; 553 curthread->td_base_pri = PRIBIO; 554 555 switch (sc->type) { 556 case MD_SWAP: 557 case MD_VNODE: 558 mtx_lock(&Giant); 559 hasgiant = 1; 560 break; 561 case MD_MALLOC: 562 case MD_PRELOAD: 563 default: 564 hasgiant = 0; 565 break; 566 } 567 568 for (;;) { 569 mtx_lock(&sc->queue_mtx); 570 bp = bioq_first(&sc->bio_queue); 571 if (bp) 572 bioq_remove(&sc->bio_queue, bp); 573 if (!bp) { 574 if (sc->flags & MD_SHUTDOWN) { 575 mtx_unlock(&sc->queue_mtx); 576 sc->procp = NULL; 577 wakeup(&sc->procp); 578 if (!hasgiant) 579 mtx_lock(&Giant); 580 kthread_exit(0); 581 } 582 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 583 continue; 584 } 585 mtx_unlock(&sc->queue_mtx); 586 if (bp->bio_cmd == BIO_GETATTR) { 587 if (sc->fwsectors && sc->fwheads && 588 (g_handleattr_int(bp, "GEOM::fwsectors", 589 sc->fwsectors) || 590 g_handleattr_int(bp, "GEOM::fwheads", 591 sc->fwheads))) 592 error = -1; 593 else 594 error = EOPNOTSUPP; 595 } else { 596 switch (sc->type) { 597 case MD_MALLOC: 598 error = mdstart_malloc(sc, bp); 599 break; 600 case MD_PRELOAD: 601 error = mdstart_preload(sc, bp); 602 break; 603 case MD_VNODE: 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 } 614 615 if (error != -1) { 616 bp->bio_completed = bp->bio_length; 617 g_io_deliver(bp, error); 618 } 619 } 620 } 621 622 static struct md_s * 623 mdfind(int unit) 624 { 625 struct md_s *sc; 626 627 /* XXX: LOCK(unique unit numbers) */ 628 LIST_FOREACH(sc, &md_softc_list, list) { 629 if (sc->unit == unit) 630 break; 631 } 632 /* XXX: UNLOCK(unique unit numbers) */ 633 return (sc); 634 } 635 636 static struct md_s * 637 mdnew(int unit) 638 { 639 struct md_s *sc; 640 int error, max = -1; 641 642 /* XXX: LOCK(unique unit numbers) */ 643 LIST_FOREACH(sc, &md_softc_list, list) { 644 if (sc->unit == unit) { 645 /* XXX: UNLOCK(unique unit numbers) */ 646 return (NULL); 647 } 648 if (sc->unit > max) 649 max = sc->unit; 650 } 651 if (unit == -1) 652 unit = max + 1; 653 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 654 sc->unit = unit; 655 bioq_init(&sc->bio_queue); 656 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 657 sprintf(sc->name, "md%d", unit); 658 error = kthread_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 659 if (error) { 660 free(sc, M_MD); 661 return (NULL); 662 } 663 LIST_INSERT_HEAD(&md_softc_list, sc, list); 664 /* XXX: UNLOCK(unique unit numbers) */ 665 return (sc); 666 } 667 668 static void 669 mdinit(struct md_s *sc) 670 { 671 672 struct g_geom *gp; 673 struct g_provider *pp; 674 675 DROP_GIANT(); 676 g_topology_lock(); 677 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 678 gp->start = g_md_start; 679 gp->access = g_md_access; 680 gp->softc = sc; 681 pp = g_new_providerf(gp, "md%d", sc->unit); 682 pp->mediasize = (off_t)sc->nsect * sc->secsize; 683 pp->sectorsize = sc->secsize; 684 sc->gp = gp; 685 sc->pp = pp; 686 g_error_provider(pp, 0); 687 g_topology_unlock(); 688 PICKUP_GIANT(); 689 } 690 691 /* 692 * XXX: we should check that the range they feed us is mapped. 693 * XXX: we should implement read-only. 694 */ 695 696 static int 697 mdcreate_preload(struct md_ioctl *mdio) 698 { 699 struct md_s *sc; 700 701 if (mdio->md_size == 0) 702 return (EINVAL); 703 if (mdio->md_options & ~(MD_AUTOUNIT)) 704 return (EINVAL); 705 if (mdio->md_options & MD_AUTOUNIT) { 706 sc = mdnew(-1); 707 if (sc == NULL) 708 return (ENOMEM); 709 mdio->md_unit = sc->unit; 710 } else { 711 sc = mdnew(mdio->md_unit); 712 if (sc == NULL) 713 return (EBUSY); 714 } 715 sc->type = MD_PRELOAD; 716 sc->secsize = DEV_BSIZE; 717 sc->nsect = mdio->md_size; 718 sc->flags = mdio->md_options & MD_FORCE; 719 /* Cast to pointer size, then to pointer to avoid warning */ 720 sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base; 721 sc->pl_len = (mdio->md_size << DEV_BSHIFT); 722 mdinit(sc); 723 return (0); 724 } 725 726 727 static int 728 mdcreate_malloc(struct md_ioctl *mdio) 729 { 730 struct md_s *sc; 731 off_t u; 732 uintptr_t sp; 733 int error; 734 735 error = 0; 736 if (mdio->md_size == 0) 737 return (EINVAL); 738 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 739 return (EINVAL); 740 if (mdio->md_secsize != 0 && !powerof2(mdio->md_secsize)) 741 return (EINVAL); 742 /* Compression doesn't make sense if we have reserved space */ 743 if (mdio->md_options & MD_RESERVE) 744 mdio->md_options &= ~MD_COMPRESS; 745 if (mdio->md_options & MD_AUTOUNIT) { 746 sc = mdnew(-1); 747 if (sc == NULL) 748 return (ENOMEM); 749 mdio->md_unit = sc->unit; 750 } else { 751 sc = mdnew(mdio->md_unit); 752 if (sc == NULL) 753 return (EBUSY); 754 } 755 sc->type = MD_MALLOC; 756 if (mdio->md_secsize != 0) 757 sc->secsize = mdio->md_secsize; 758 else 759 sc->secsize = DEV_BSIZE; 760 if (mdio->md_fwsectors != 0) 761 sc->fwsectors = mdio->md_fwsectors; 762 if (mdio->md_fwheads != 0) 763 sc->fwheads = mdio->md_fwheads; 764 sc->nsect = mdio->md_size; 765 sc->nsect /= (sc->secsize / DEV_BSIZE); 766 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 767 sc->indir = dimension(sc->nsect); 768 sc->uma = uma_zcreate(sc->name, sc->secsize, 769 NULL, NULL, NULL, NULL, 0x1ff, 0); 770 if (mdio->md_options & MD_RESERVE) { 771 for (u = 0; u < sc->nsect; u++) { 772 sp = (uintptr_t) uma_zalloc(sc->uma, M_NOWAIT | M_ZERO); 773 if (sp != 0) 774 error = s_write(sc->indir, u, sp); 775 else 776 error = ENOMEM; 777 if (error) 778 break; 779 } 780 } 781 if (error) { 782 mddestroy(sc, NULL); 783 return (error); 784 } 785 mdinit(sc); 786 if (!(mdio->md_options & MD_RESERVE)) 787 sc->pp->flags |= G_PF_CANDELETE; 788 return (0); 789 } 790 791 792 static int 793 mdsetcred(struct md_s *sc, struct ucred *cred) 794 { 795 char *tmpbuf; 796 int error = 0; 797 798 /* 799 * Set credits in our softc 800 */ 801 802 if (sc->cred) 803 crfree(sc->cred); 804 sc->cred = crhold(cred); 805 806 /* 807 * Horrible kludge to establish credentials for NFS XXX. 808 */ 809 810 if (sc->vnode) { 811 struct uio auio; 812 struct iovec aiov; 813 814 tmpbuf = malloc(sc->secsize, M_TEMP, M_WAITOK); 815 bzero(&auio, sizeof(auio)); 816 817 aiov.iov_base = tmpbuf; 818 aiov.iov_len = sc->secsize; 819 auio.uio_iov = &aiov; 820 auio.uio_iovcnt = 1; 821 auio.uio_offset = 0; 822 auio.uio_rw = UIO_READ; 823 auio.uio_segflg = UIO_SYSSPACE; 824 auio.uio_resid = aiov.iov_len; 825 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 826 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 827 VOP_UNLOCK(sc->vnode, 0, curthread); 828 free(tmpbuf, M_TEMP); 829 } 830 return (error); 831 } 832 833 static int 834 mdcreate_vnode(struct md_ioctl *mdio, struct thread *td) 835 { 836 struct md_s *sc; 837 struct vattr vattr; 838 struct nameidata nd; 839 int error, flags; 840 841 flags = FREAD|FWRITE; 842 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 843 error = vn_open(&nd, &flags, 0); 844 if (error) { 845 if (error != EACCES && error != EPERM && error != EROFS) 846 return (error); 847 flags &= ~FWRITE; 848 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 849 error = vn_open(&nd, &flags, 0); 850 if (error) 851 return (error); 852 } 853 NDFREE(&nd, NDF_ONLY_PNBUF); 854 if (nd.ni_vp->v_type != VREG || 855 (error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred, td))) { 856 VOP_UNLOCK(nd.ni_vp, 0, td); 857 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 858 return (error ? error : EINVAL); 859 } 860 VOP_UNLOCK(nd.ni_vp, 0, td); 861 862 if (mdio->md_options & MD_AUTOUNIT) { 863 sc = mdnew(-1); 864 mdio->md_unit = sc->unit; 865 } else { 866 sc = mdnew(mdio->md_unit); 867 } 868 if (sc == NULL) { 869 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 870 return (EBUSY); 871 } 872 873 sc->type = MD_VNODE; 874 sc->flags = mdio->md_options & MD_FORCE; 875 if (!(flags & FWRITE)) 876 sc->flags |= MD_READONLY; 877 sc->secsize = DEV_BSIZE; 878 sc->vnode = nd.ni_vp; 879 880 /* 881 * If the size is specified, override the file attributes. 882 */ 883 if (mdio->md_size) 884 sc->nsect = mdio->md_size; 885 else 886 sc->nsect = vattr.va_size / sc->secsize; /* XXX: round up ? */ 887 if (sc->nsect == 0) { 888 mddestroy(sc, td); 889 return (EINVAL); 890 } 891 error = mdsetcred(sc, td->td_ucred); 892 if (error) { 893 mddestroy(sc, td); 894 return (error); 895 } 896 mdinit(sc); 897 return (0); 898 } 899 900 static void 901 md_zapit(void *p, int cancel) 902 { 903 if (cancel) 904 return; 905 g_wither_geom(p, ENXIO); 906 } 907 908 static int 909 mddestroy(struct md_s *sc, struct thread *td) 910 { 911 912 GIANT_REQUIRED; 913 914 mtx_destroy(&sc->queue_mtx); 915 if (sc->gp) { 916 sc->gp->softc = NULL; 917 g_waitfor_event(md_zapit, sc->gp, M_WAITOK, sc->gp, NULL); 918 sc->gp = NULL; 919 sc->pp = NULL; 920 } 921 sc->flags |= MD_SHUTDOWN; 922 wakeup(sc); 923 while (sc->procp != NULL) 924 tsleep(&sc->procp, PRIBIO, "mddestroy", hz / 10); 925 if (sc->vnode != NULL) 926 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 927 FREAD : (FREAD|FWRITE), sc->cred, td); 928 if (sc->cred != NULL) 929 crfree(sc->cred); 930 if (sc->object != NULL) { 931 vm_object_deallocate(sc->object); 932 } 933 if (sc->indir) 934 destroy_indir(sc, sc->indir); 935 if (sc->uma) 936 uma_zdestroy(sc->uma); 937 938 /* XXX: LOCK(unique unit numbers) */ 939 LIST_REMOVE(sc, list); 940 /* XXX: UNLOCK(unique unit numbers) */ 941 free(sc, M_MD); 942 return (0); 943 } 944 945 static int 946 mdcreate_swap(struct md_ioctl *mdio, struct thread *td) 947 { 948 int error; 949 struct md_s *sc; 950 951 GIANT_REQUIRED; 952 953 if (mdio->md_options & MD_AUTOUNIT) { 954 sc = mdnew(-1); 955 mdio->md_unit = sc->unit; 956 } else { 957 sc = mdnew(mdio->md_unit); 958 } 959 if (sc == NULL) 960 return (EBUSY); 961 962 sc->type = MD_SWAP; 963 964 /* 965 * Range check. Disallow negative sizes or any size less then the 966 * size of a page. Then round to a page. 967 */ 968 969 if (mdio->md_size == 0) { 970 mddestroy(sc, td); 971 return (EDOM); 972 } 973 974 /* 975 * Allocate an OBJT_SWAP object. 976 * 977 * sc_secsize is PAGE_SIZE'd 978 * 979 * mdio->size is in DEV_BSIZE'd chunks. 980 * Note the truncation. 981 */ 982 983 sc->secsize = PAGE_SIZE; 984 sc->nsect = mdio->md_size / (PAGE_SIZE / DEV_BSIZE); 985 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, sc->secsize * (vm_offset_t)sc->nsect, VM_PROT_DEFAULT, 0); 986 sc->flags = mdio->md_options & MD_FORCE; 987 if (mdio->md_options & MD_RESERVE) { 988 if (swap_pager_reserve(sc->object, 0, sc->nsect) < 0) { 989 vm_object_deallocate(sc->object); 990 sc->object = NULL; 991 mddestroy(sc, td); 992 return (EDOM); 993 } 994 } 995 error = mdsetcred(sc, td->td_ucred); 996 if (error) { 997 mddestroy(sc, td); 998 return (error); 999 } 1000 mdinit(sc); 1001 if (!(mdio->md_options & MD_RESERVE)) 1002 sc->pp->flags |= G_PF_CANDELETE; 1003 return (0); 1004 } 1005 1006 static int 1007 mddetach(int unit, struct thread *td) 1008 { 1009 struct md_s *sc; 1010 1011 sc = mdfind(unit); 1012 if (sc == NULL) 1013 return (ENOENT); 1014 if (sc->opencount != 0 && !(sc->flags & MD_FORCE)) 1015 return (EBUSY); 1016 switch(sc->type) { 1017 case MD_VNODE: 1018 case MD_SWAP: 1019 case MD_MALLOC: 1020 case MD_PRELOAD: 1021 return (mddestroy(sc, td)); 1022 default: 1023 return (EOPNOTSUPP); 1024 } 1025 } 1026 1027 static int 1028 mdctlioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1029 { 1030 struct md_ioctl *mdio; 1031 struct md_s *sc; 1032 int i; 1033 1034 if (md_debug) 1035 printf("mdctlioctl(%s %lx %p %x %p)\n", 1036 devtoname(dev), cmd, addr, flags, td); 1037 1038 /* 1039 * We assert the version number in the individual ioctl 1040 * handlers instead of out here because (a) it is possible we 1041 * may add another ioctl in the future which doesn't read an 1042 * mdio, and (b) the correct return value for an unknown ioctl 1043 * is ENOIOCTL, not EINVAL. 1044 */ 1045 mdio = (struct md_ioctl *)addr; 1046 switch (cmd) { 1047 case MDIOCATTACH: 1048 if (mdio->md_version != MDIOVERSION) 1049 return (EINVAL); 1050 switch (mdio->md_type) { 1051 case MD_MALLOC: 1052 return (mdcreate_malloc(mdio)); 1053 case MD_PRELOAD: 1054 return (mdcreate_preload(mdio)); 1055 case MD_VNODE: 1056 return (mdcreate_vnode(mdio, td)); 1057 case MD_SWAP: 1058 return (mdcreate_swap(mdio, td)); 1059 default: 1060 return (EINVAL); 1061 } 1062 case MDIOCDETACH: 1063 if (mdio->md_version != MDIOVERSION) 1064 return (EINVAL); 1065 if (mdio->md_file != NULL || mdio->md_size != 0 || 1066 mdio->md_options != 0) 1067 return (EINVAL); 1068 return (mddetach(mdio->md_unit, td)); 1069 case MDIOCQUERY: 1070 if (mdio->md_version != MDIOVERSION) 1071 return (EINVAL); 1072 sc = mdfind(mdio->md_unit); 1073 if (sc == NULL) 1074 return (ENOENT); 1075 mdio->md_type = sc->type; 1076 mdio->md_options = sc->flags; 1077 switch (sc->type) { 1078 case MD_MALLOC: 1079 mdio->md_size = sc->nsect; 1080 break; 1081 case MD_PRELOAD: 1082 mdio->md_size = sc->nsect; 1083 mdio->md_base = (uint64_t)(intptr_t)sc->pl_ptr; 1084 break; 1085 case MD_SWAP: 1086 mdio->md_size = sc->nsect * (PAGE_SIZE / DEV_BSIZE); 1087 break; 1088 case MD_VNODE: 1089 mdio->md_size = sc->nsect; 1090 /* XXX fill this in */ 1091 mdio->md_file = NULL; 1092 break; 1093 } 1094 return (0); 1095 case MDIOCLIST: 1096 i = 1; 1097 LIST_FOREACH(sc, &md_softc_list, list) { 1098 if (i == MDNPAD - 1) 1099 mdio->md_pad[i] = -1; 1100 else 1101 mdio->md_pad[i++] = sc->unit; 1102 } 1103 mdio->md_pad[0] = i - 1; 1104 return (0); 1105 default: 1106 return (ENOIOCTL); 1107 }; 1108 return (ENOIOCTL); 1109 } 1110 1111 static void 1112 md_preloaded(u_char *image, unsigned length) 1113 { 1114 struct md_s *sc; 1115 1116 sc = mdnew(-1); 1117 if (sc == NULL) 1118 return; 1119 sc->type = MD_PRELOAD; 1120 sc->secsize = DEV_BSIZE; 1121 sc->nsect = length / DEV_BSIZE; 1122 sc->pl_ptr = image; 1123 sc->pl_len = length; 1124 if (sc->unit == 0) 1125 mdrootready = 1; 1126 mdinit(sc); 1127 } 1128 1129 static void 1130 md_drvinit(void *unused) 1131 { 1132 1133 caddr_t mod; 1134 caddr_t c; 1135 u_char *ptr, *name, *type; 1136 unsigned len; 1137 1138 #ifdef MD_ROOT_SIZE 1139 md_preloaded(mfs_root, MD_ROOT_SIZE*1024); 1140 #endif 1141 mod = NULL; 1142 while ((mod = preload_search_next_name(mod)) != NULL) { 1143 name = (char *)preload_search_info(mod, MODINFO_NAME); 1144 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1145 if (name == NULL) 1146 continue; 1147 if (type == NULL) 1148 continue; 1149 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1150 continue; 1151 c = preload_search_info(mod, MODINFO_ADDR); 1152 ptr = *(u_char **)c; 1153 c = preload_search_info(mod, MODINFO_SIZE); 1154 len = *(size_t *)c; 1155 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1156 MD_NAME, mdunits, name, len, ptr); 1157 md_preloaded(ptr, len); 1158 } 1159 status_dev = make_dev(&mdctl_cdevsw, 0xffff00ff, UID_ROOT, GID_WHEEL, 1160 0600, MDCTL_NAME); 1161 } 1162 1163 static int 1164 md_modevent(module_t mod, int type, void *data) 1165 { 1166 int error; 1167 struct md_s *sc; 1168 1169 switch (type) { 1170 case MOD_LOAD: 1171 md_drvinit(NULL); 1172 break; 1173 case MOD_UNLOAD: 1174 LIST_FOREACH(sc, &md_softc_list, list) { 1175 error = mddetach(sc->unit, curthread); 1176 if (error != 0) 1177 return (error); 1178 } 1179 if (status_dev) 1180 destroy_dev(status_dev); 1181 status_dev = 0; 1182 break; 1183 default: 1184 break; 1185 } 1186 return (0); 1187 } 1188 1189 static moduledata_t md_mod = { 1190 MD_NAME, 1191 md_modevent, 1192 NULL 1193 }; 1194 DECLARE_MODULE(md, md_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE+CDEV_MAJOR); 1195 MODULE_VERSION(md, MD_MODVER); 1196 1197 1198 #ifdef MD_ROOT 1199 static void 1200 md_takeroot(void *junk) 1201 { 1202 if (mdrootready) 1203 rootdevnames[0] = "ufs:/dev/md0"; 1204 } 1205 1206 SYSINIT(md_root, SI_SUB_MOUNT_ROOT, SI_ORDER_FIRST, md_takeroot, NULL); 1207 #endif /* MD_ROOT */ 1208