md.c revision 127211
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 127211 2004-03-19 21:19:15Z 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/sf_buf.h> 79#include <sys/sysctl.h> 80#include <sys/vnode.h> 81 82#include <geom/geom.h> 83 84#include <vm/vm.h> 85#include <vm/vm_object.h> 86#include <vm/vm_page.h> 87#include <vm/vm_pager.h> 88#include <vm/swap_pager.h> 89#include <vm/uma.h> 90 91#define MD_MODVER 1 92 93#define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */ 94 95#ifndef MD_NSECT 96#define MD_NSECT (10000 * 2) 97#endif 98 99static MALLOC_DEFINE(M_MD, "MD disk", "Memory Disk"); 100static MALLOC_DEFINE(M_MDSECT, "MD sectors", "Memory Disk Sectors"); 101 102static int md_debug; 103SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, ""); 104 105#if defined(MD_ROOT) && defined(MD_ROOT_SIZE) 106/* Image gets put here: */ 107static u_char mfs_root[MD_ROOT_SIZE*1024] = "MFS Filesystem goes here"; 108static u_char end_mfs_root[] __unused = "MFS Filesystem had better STOP here"; 109#endif 110 111static g_init_t md_drvinit; 112 113static int mdrootready; 114static int mdunits; 115static dev_t status_dev = 0; 116 117static d_ioctl_t mdctlioctl; 118 119static struct cdevsw mdctl_cdevsw = { 120 .d_version = D_VERSION, 121 .d_flags = D_NEEDGIANT, 122 .d_ioctl = mdctlioctl, 123 .d_name = MD_NAME, 124}; 125 126 127static 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) 131static int nshift; 132 133struct indir { 134 uintptr_t *array; 135 u_int total; 136 u_int used; 137 u_int shift; 138}; 139 140struct 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 unsigned npage; 173}; 174 175static int mddestroy(struct md_s *sc, struct thread *td); 176 177static struct indir * 178new_indir(u_int shift) 179{ 180 struct indir *ip; 181 182 ip = malloc(sizeof *ip, M_MD, M_NOWAIT | M_ZERO); 183 if (ip == NULL) 184 return (NULL); 185 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 186 M_MDSECT, M_NOWAIT | M_ZERO); 187 if (ip->array == NULL) { 188 free(ip, M_MD); 189 return (NULL); 190 } 191 ip->total = NINDIR; 192 ip->shift = shift; 193 return (ip); 194} 195 196static void 197del_indir(struct indir *ip) 198{ 199 200 free(ip->array, M_MDSECT); 201 free(ip, M_MD); 202} 203 204static void 205destroy_indir(struct md_s *sc, struct indir *ip) 206{ 207 int i; 208 209 for (i = 0; i < NINDIR; i++) { 210 if (!ip->array[i]) 211 continue; 212 if (ip->shift) 213 destroy_indir(sc, (struct indir*)(ip->array[i])); 214 else if (ip->array[i] > 255) 215 uma_zfree(sc->uma, (void *)(ip->array[i])); 216 } 217 del_indir(ip); 218} 219 220/* 221 * This function does the math and alloctes the top level "indir" structure 222 * for a device of "size" sectors. 223 */ 224 225static struct indir * 226dimension(off_t size) 227{ 228 off_t rcnt; 229 struct indir *ip; 230 int i, layer; 231 232 rcnt = size; 233 layer = 0; 234 while (rcnt > NINDIR) { 235 rcnt /= NINDIR; 236 layer++; 237 } 238 /* figure out log2(NINDIR) */ 239 for (i = NINDIR, nshift = -1; i; nshift++) 240 i >>= 1; 241 242 /* 243 * XXX: the top layer is probably not fully populated, so we allocate 244 * too much space for ip->array in here. 245 */ 246 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 247 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 248 M_MDSECT, M_WAITOK | M_ZERO); 249 ip->total = NINDIR; 250 ip->shift = layer * nshift; 251 return (ip); 252} 253 254/* 255 * Read a given sector 256 */ 257 258static uintptr_t 259s_read(struct indir *ip, off_t offset) 260{ 261 struct indir *cip; 262 int idx; 263 uintptr_t up; 264 265 if (md_debug > 1) 266 printf("s_read(%jd)\n", (intmax_t)offset); 267 up = 0; 268 for (cip = ip; cip != NULL;) { 269 if (cip->shift) { 270 idx = (offset >> cip->shift) & NMASK; 271 up = cip->array[idx]; 272 cip = (struct indir *)up; 273 continue; 274 } 275 idx = offset & NMASK; 276 return (cip->array[idx]); 277 } 278 return (0); 279} 280 281/* 282 * Write a given sector, prune the tree if the value is 0 283 */ 284 285static int 286s_write(struct indir *ip, off_t offset, uintptr_t ptr) 287{ 288 struct indir *cip, *lip[10]; 289 int idx, li; 290 uintptr_t up; 291 292 if (md_debug > 1) 293 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 294 up = 0; 295 li = 0; 296 cip = ip; 297 for (;;) { 298 lip[li++] = cip; 299 if (cip->shift) { 300 idx = (offset >> cip->shift) & NMASK; 301 up = cip->array[idx]; 302 if (up != 0) { 303 cip = (struct indir *)up; 304 continue; 305 } 306 /* Allocate branch */ 307 cip->array[idx] = 308 (uintptr_t)new_indir(cip->shift - nshift); 309 if (cip->array[idx] == 0) 310 return (ENOSPC); 311 cip->used++; 312 up = cip->array[idx]; 313 cip = (struct indir *)up; 314 continue; 315 } 316 /* leafnode */ 317 idx = offset & NMASK; 318 up = cip->array[idx]; 319 if (up != 0) 320 cip->used--; 321 cip->array[idx] = ptr; 322 if (ptr != 0) 323 cip->used++; 324 break; 325 } 326 if (cip->used != 0 || li == 1) 327 return (0); 328 li--; 329 while (cip->used == 0 && cip != ip) { 330 li--; 331 idx = (offset >> lip[li]->shift) & NMASK; 332 up = lip[li]->array[idx]; 333 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 334 del_indir(cip); 335 lip[li]->array[idx] = 0; 336 lip[li]->used--; 337 cip = lip[li]; 338 } 339 return (0); 340} 341 342 343struct g_class g_md_class = { 344 .name = "MD", 345 .init = md_drvinit, 346}; 347 348static int 349g_md_access(struct g_provider *pp, int r, int w, int e) 350{ 351 struct md_s *sc; 352 353 sc = pp->geom->softc; 354 if (sc == NULL) 355 return (ENXIO); 356 r += pp->acr; 357 w += pp->acw; 358 e += pp->ace; 359 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 360 sc->opencount = 1; 361 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 362 sc->opencount = 0; 363 } 364 return (0); 365} 366 367static void 368g_md_start(struct bio *bp) 369{ 370 struct md_s *sc; 371 372 sc = bp->bio_to->geom->softc; 373 374 bp->bio_pblkno = bp->bio_offset / sc->secsize; 375 bp->bio_bcount = bp->bio_length; 376 mtx_lock(&sc->queue_mtx); 377 bioq_disksort(&sc->bio_queue, bp); 378 mtx_unlock(&sc->queue_mtx); 379 380 wakeup(sc); 381} 382 383DECLARE_GEOM_CLASS(g_md_class, g_md); 384 385 386static int 387mdstart_malloc(struct md_s *sc, struct bio *bp) 388{ 389 int i, error; 390 u_char *dst; 391 unsigned secno, nsec, uc; 392 uintptr_t sp, osp; 393 394 nsec = bp->bio_bcount / sc->secsize; 395 secno = bp->bio_pblkno; 396 dst = bp->bio_data; 397 error = 0; 398 while (nsec--) { 399 osp = s_read(sc->indir, secno); 400 if (bp->bio_cmd == BIO_DELETE) { 401 if (osp != 0) 402 error = s_write(sc->indir, secno, 0); 403 } else if (bp->bio_cmd == BIO_READ) { 404 if (osp == 0) 405 bzero(dst, sc->secsize); 406 else if (osp <= 255) 407 for (i = 0; i < sc->secsize; i++) 408 dst[i] = osp; 409 else 410 bcopy((void *)osp, dst, sc->secsize); 411 osp = 0; 412 } else if (bp->bio_cmd == BIO_WRITE) { 413 if (sc->flags & MD_COMPRESS) { 414 uc = dst[0]; 415 for (i = 1; i < sc->secsize; i++) 416 if (dst[i] != uc) 417 break; 418 } else { 419 i = 0; 420 uc = 0; 421 } 422 if (i == sc->secsize) { 423 if (osp != uc) 424 error = s_write(sc->indir, secno, uc); 425 } else { 426 if (osp <= 255) { 427 sp = (uintptr_t) uma_zalloc( 428 sc->uma, M_NOWAIT); 429 if (sp == 0) { 430 error = ENOSPC; 431 break; 432 } 433 bcopy(dst, (void *)sp, sc->secsize); 434 error = s_write(sc->indir, secno, sp); 435 } else { 436 bcopy(dst, (void *)osp, sc->secsize); 437 osp = 0; 438 } 439 } 440 } else { 441 error = EOPNOTSUPP; 442 } 443 if (osp > 255) 444 uma_zfree(sc->uma, (void*)osp); 445 if (error) 446 break; 447 secno++; 448 dst += sc->secsize; 449 } 450 bp->bio_resid = 0; 451 return (error); 452} 453 454static int 455mdstart_preload(struct md_s *sc, struct bio *bp) 456{ 457 458 if (bp->bio_cmd == BIO_DELETE) { 459 } else if (bp->bio_cmd == BIO_READ) { 460 bcopy(sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_data, bp->bio_bcount); 461 } else { 462 bcopy(bp->bio_data, sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_bcount); 463 } 464 bp->bio_resid = 0; 465 return (0); 466} 467 468static int 469mdstart_vnode(struct md_s *sc, struct bio *bp) 470{ 471 int error; 472 struct uio auio; 473 struct iovec aiov; 474 struct mount *mp; 475 476 /* 477 * VNODE I/O 478 * 479 * If an error occurs, we set BIO_ERROR but we do not set 480 * B_INVAL because (for a write anyway), the buffer is 481 * still valid. 482 */ 483 484 bzero(&auio, sizeof(auio)); 485 486 aiov.iov_base = bp->bio_data; 487 aiov.iov_len = bp->bio_bcount; 488 auio.uio_iov = &aiov; 489 auio.uio_iovcnt = 1; 490 auio.uio_offset = (vm_ooffset_t)bp->bio_pblkno * sc->secsize; 491 auio.uio_segflg = UIO_SYSSPACE; 492 if(bp->bio_cmd == BIO_READ) 493 auio.uio_rw = UIO_READ; 494 else if(bp->bio_cmd == BIO_WRITE) 495 auio.uio_rw = UIO_WRITE; 496 else 497 panic("wrong BIO_OP in mdstart_vnode"); 498 auio.uio_resid = bp->bio_bcount; 499 auio.uio_td = curthread; 500 /* 501 * When reading set IO_DIRECT to try to avoid double-caching 502 * the data. When writing IO_DIRECT is not optimal. 503 */ 504 if (bp->bio_cmd == BIO_READ) { 505 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 506 error = VOP_READ(sc->vnode, &auio, IO_DIRECT, sc->cred); 507 VOP_UNLOCK(sc->vnode, 0, curthread); 508 } else { 509 (void) vn_start_write(sc->vnode, &mp, V_WAIT); 510 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 511 error = VOP_WRITE(sc->vnode, &auio, 512 sc->flags & MD_ASYNC ? 0 : IO_SYNC, sc->cred); 513 VOP_UNLOCK(sc->vnode, 0, curthread); 514 vn_finished_write(mp); 515 } 516 bp->bio_resid = auio.uio_resid; 517 return (error); 518} 519 520static int 521mdstart_swap(struct md_s *sc, struct bio *bp) 522{ 523 struct sf_buf *sf; 524 int i, rv; 525 int offs, len, lastp, lastend; 526 vm_page_t m; 527 u_char *p; 528 529 p = bp->bio_data; 530 531 /* 532 * offs is the ofset at whih to start operating on the 533 * next (ie, first) page. lastp is the last page on 534 * which we're going to operate. lastend is the ending 535 * position within that last page (ie, PAGE_SIZE if 536 * we're operating on complete aligned pages). 537 */ 538 offs = bp->bio_offset % PAGE_SIZE; 539 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE; 540 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1; 541 542 VM_OBJECT_LOCK(sc->object); 543 vm_object_pip_add(sc->object, 1); 544 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) { 545 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs; 546 547 m = vm_page_grab(sc->object, i, 548 VM_ALLOC_NORMAL|VM_ALLOC_RETRY); 549 VM_OBJECT_UNLOCK(sc->object); 550 sf = sf_buf_alloc(m); 551 VM_OBJECT_LOCK(sc->object); 552 if (bp->bio_cmd == BIO_READ) { 553 if (m->valid != VM_PAGE_BITS_ALL) 554 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 555 bcopy((void *)(sf_buf_kva(sf) + offs), p, len); 556 } else if (bp->bio_cmd == BIO_WRITE) { 557 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 558 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 559 bcopy(p, (void *)(sf_buf_kva(sf) + offs), len); 560 m->valid = VM_PAGE_BITS_ALL; 561#if 0 562 } else if (bp->bio_cmd == BIO_DELETE) { 563 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 564 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 565 bzero((void *)(sf_buf_kva(sf) + offs), len); 566 vm_page_dirty(m); 567 m->valid = VM_PAGE_BITS_ALL; 568#endif 569 } 570 sf_buf_free(sf); 571 vm_page_lock_queues(); 572 vm_page_wakeup(m); 573 vm_page_activate(m); 574 if (bp->bio_cmd == BIO_WRITE) 575 vm_page_dirty(m); 576 vm_page_unlock_queues(); 577 578 /* Actions on further pages start at offset 0 */ 579 p += PAGE_SIZE - offs; 580 offs = 0; 581#if 0 582if (bootverbose || bp->bio_offset / PAGE_SIZE < 17) 583printf("wire_count %d busy %d flags %x hold_count %d act_count %d queue %d valid %d dirty %d @ %d\n", 584 m->wire_count, m->busy, 585 m->flags, m->hold_count, m->act_count, m->queue, m->valid, m->dirty, i); 586#endif 587 } 588 vm_object_pip_subtract(sc->object, 1); 589 vm_object_set_writeable_dirty(sc->object); 590 VM_OBJECT_UNLOCK(sc->object); 591 return (0); 592} 593 594static void 595md_kthread(void *arg) 596{ 597 struct md_s *sc; 598 struct bio *bp; 599 int error, hasgiant; 600 601 sc = arg; 602 curthread->td_base_pri = PRIBIO; 603 604 switch (sc->type) { 605 case MD_VNODE: 606 mtx_lock(&Giant); 607 hasgiant = 1; 608 break; 609 case MD_MALLOC: 610 case MD_PRELOAD: 611 case MD_SWAP: 612 default: 613 hasgiant = 0; 614 break; 615 } 616 617 for (;;) { 618 mtx_lock(&sc->queue_mtx); 619 bp = bioq_first(&sc->bio_queue); 620 if (bp) 621 bioq_remove(&sc->bio_queue, bp); 622 if (!bp) { 623 if (sc->flags & MD_SHUTDOWN) { 624 mtx_unlock(&sc->queue_mtx); 625 sc->procp = NULL; 626 wakeup(&sc->procp); 627 if (hasgiant) 628 mtx_unlock(&Giant); 629 kthread_exit(0); 630 } 631 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 632 continue; 633 } 634 mtx_unlock(&sc->queue_mtx); 635 if (bp->bio_cmd == BIO_GETATTR) { 636 if (sc->fwsectors && sc->fwheads && 637 (g_handleattr_int(bp, "GEOM::fwsectors", 638 sc->fwsectors) || 639 g_handleattr_int(bp, "GEOM::fwheads", 640 sc->fwheads))) 641 error = -1; 642 else 643 error = EOPNOTSUPP; 644 } else { 645 switch (sc->type) { 646 case MD_MALLOC: 647 error = mdstart_malloc(sc, bp); 648 break; 649 case MD_PRELOAD: 650 error = mdstart_preload(sc, bp); 651 break; 652 case MD_VNODE: 653 error = mdstart_vnode(sc, bp); 654 break; 655 case MD_SWAP: 656 error = mdstart_swap(sc, bp); 657 break; 658 default: 659 panic("Impossible md(type)"); 660 break; 661 } 662 } 663 664 if (error != -1) { 665 bp->bio_completed = bp->bio_length; 666 g_io_deliver(bp, error); 667 } 668 } 669} 670 671static struct md_s * 672mdfind(int unit) 673{ 674 struct md_s *sc; 675 676 /* XXX: LOCK(unique unit numbers) */ 677 LIST_FOREACH(sc, &md_softc_list, list) { 678 if (sc->unit == unit) 679 break; 680 } 681 /* XXX: UNLOCK(unique unit numbers) */ 682 return (sc); 683} 684 685static struct md_s * 686mdnew(int unit) 687{ 688 struct md_s *sc; 689 int error, max = -1; 690 691 /* XXX: LOCK(unique unit numbers) */ 692 LIST_FOREACH(sc, &md_softc_list, list) { 693 if (sc->unit == unit) { 694 /* XXX: UNLOCK(unique unit numbers) */ 695 return (NULL); 696 } 697 if (sc->unit > max) 698 max = sc->unit; 699 } 700 if (unit == -1) 701 unit = max + 1; 702 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 703 sc->unit = unit; 704 bioq_init(&sc->bio_queue); 705 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 706 sprintf(sc->name, "md%d", unit); 707 error = kthread_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 708 if (error) { 709 free(sc, M_MD); 710 return (NULL); 711 } 712 LIST_INSERT_HEAD(&md_softc_list, sc, list); 713 /* XXX: UNLOCK(unique unit numbers) */ 714 return (sc); 715} 716 717static void 718mdinit(struct md_s *sc) 719{ 720 721 struct g_geom *gp; 722 struct g_provider *pp; 723 724 DROP_GIANT(); 725 g_topology_lock(); 726 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 727 gp->start = g_md_start; 728 gp->access = g_md_access; 729 gp->softc = sc; 730 pp = g_new_providerf(gp, "md%d", sc->unit); 731 pp->mediasize = (off_t)sc->nsect * sc->secsize; 732 pp->sectorsize = sc->secsize; 733 sc->gp = gp; 734 sc->pp = pp; 735 g_error_provider(pp, 0); 736 g_topology_unlock(); 737 PICKUP_GIANT(); 738} 739 740/* 741 * XXX: we should check that the range they feed us is mapped. 742 * XXX: we should implement read-only. 743 */ 744 745static int 746mdcreate_preload(struct md_ioctl *mdio) 747{ 748 struct md_s *sc; 749 750 if (mdio->md_size == 0) 751 return (EINVAL); 752 if (mdio->md_options & ~(MD_AUTOUNIT)) 753 return (EINVAL); 754 if (mdio->md_options & MD_AUTOUNIT) { 755 sc = mdnew(-1); 756 if (sc == NULL) 757 return (ENOMEM); 758 mdio->md_unit = sc->unit; 759 } else { 760 sc = mdnew(mdio->md_unit); 761 if (sc == NULL) 762 return (EBUSY); 763 } 764 sc->type = MD_PRELOAD; 765 sc->secsize = DEV_BSIZE; 766 sc->nsect = mdio->md_size; 767 sc->flags = mdio->md_options & MD_FORCE; 768 /* Cast to pointer size, then to pointer to avoid warning */ 769 sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base; 770 sc->pl_len = (mdio->md_size << DEV_BSHIFT); 771 mdinit(sc); 772 return (0); 773} 774 775 776static int 777mdcreate_malloc(struct md_ioctl *mdio) 778{ 779 struct md_s *sc; 780 off_t u; 781 uintptr_t sp; 782 int error; 783 784 error = 0; 785 if (mdio->md_size == 0) 786 return (EINVAL); 787 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 788 return (EINVAL); 789 if (mdio->md_secsize != 0 && !powerof2(mdio->md_secsize)) 790 return (EINVAL); 791 /* Compression doesn't make sense if we have reserved space */ 792 if (mdio->md_options & MD_RESERVE) 793 mdio->md_options &= ~MD_COMPRESS; 794 if (mdio->md_options & MD_AUTOUNIT) { 795 sc = mdnew(-1); 796 if (sc == NULL) 797 return (ENOMEM); 798 mdio->md_unit = sc->unit; 799 } else { 800 sc = mdnew(mdio->md_unit); 801 if (sc == NULL) 802 return (EBUSY); 803 } 804 sc->type = MD_MALLOC; 805 if (mdio->md_secsize != 0) 806 sc->secsize = mdio->md_secsize; 807 else 808 sc->secsize = DEV_BSIZE; 809 if (mdio->md_fwsectors != 0) 810 sc->fwsectors = mdio->md_fwsectors; 811 if (mdio->md_fwheads != 0) 812 sc->fwheads = mdio->md_fwheads; 813 sc->nsect = mdio->md_size; 814 sc->nsect /= (sc->secsize / DEV_BSIZE); 815 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 816 sc->indir = dimension(sc->nsect); 817 sc->uma = uma_zcreate(sc->name, sc->secsize, 818 NULL, NULL, NULL, NULL, 0x1ff, 0); 819 if (mdio->md_options & MD_RESERVE) { 820 for (u = 0; u < sc->nsect; u++) { 821 sp = (uintptr_t) uma_zalloc(sc->uma, M_NOWAIT | M_ZERO); 822 if (sp != 0) 823 error = s_write(sc->indir, u, sp); 824 else 825 error = ENOMEM; 826 if (error) 827 break; 828 } 829 } 830 if (error) { 831 mddestroy(sc, NULL); 832 return (error); 833 } 834 mdinit(sc); 835 if (!(mdio->md_options & MD_RESERVE)) 836 sc->pp->flags |= G_PF_CANDELETE; 837 return (0); 838} 839 840 841static int 842mdsetcred(struct md_s *sc, struct ucred *cred) 843{ 844 char *tmpbuf; 845 int error = 0; 846 847 /* 848 * Set credits in our softc 849 */ 850 851 if (sc->cred) 852 crfree(sc->cred); 853 sc->cred = crhold(cred); 854 855 /* 856 * Horrible kludge to establish credentials for NFS XXX. 857 */ 858 859 if (sc->vnode) { 860 struct uio auio; 861 struct iovec aiov; 862 863 tmpbuf = malloc(sc->secsize, M_TEMP, M_WAITOK); 864 bzero(&auio, sizeof(auio)); 865 866 aiov.iov_base = tmpbuf; 867 aiov.iov_len = sc->secsize; 868 auio.uio_iov = &aiov; 869 auio.uio_iovcnt = 1; 870 auio.uio_offset = 0; 871 auio.uio_rw = UIO_READ; 872 auio.uio_segflg = UIO_SYSSPACE; 873 auio.uio_resid = aiov.iov_len; 874 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 875 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 876 VOP_UNLOCK(sc->vnode, 0, curthread); 877 free(tmpbuf, M_TEMP); 878 } 879 return (error); 880} 881 882static int 883mdcreate_vnode(struct md_ioctl *mdio, struct thread *td) 884{ 885 struct md_s *sc; 886 struct vattr vattr; 887 struct nameidata nd; 888 int error, flags; 889 890 flags = FREAD|FWRITE; 891 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 892 error = vn_open(&nd, &flags, 0, -1); 893 if (error) { 894 if (error != EACCES && error != EPERM && error != EROFS) 895 return (error); 896 flags &= ~FWRITE; 897 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 898 error = vn_open(&nd, &flags, 0, -1); 899 if (error) 900 return (error); 901 } 902 NDFREE(&nd, NDF_ONLY_PNBUF); 903 if (nd.ni_vp->v_type != VREG || 904 (error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred, td))) { 905 VOP_UNLOCK(nd.ni_vp, 0, td); 906 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 907 return (error ? error : EINVAL); 908 } 909 VOP_UNLOCK(nd.ni_vp, 0, td); 910 911 if (mdio->md_options & MD_AUTOUNIT) { 912 sc = mdnew(-1); 913 mdio->md_unit = sc->unit; 914 } else { 915 sc = mdnew(mdio->md_unit); 916 } 917 if (sc == NULL) { 918 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 919 return (EBUSY); 920 } 921 922 if (mdio->md_fwsectors != 0) 923 sc->fwsectors = mdio->md_fwsectors; 924 if (mdio->md_fwheads != 0) 925 sc->fwheads = mdio->md_fwheads; 926 sc->type = MD_VNODE; 927 sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC); 928 if (!(flags & FWRITE)) 929 sc->flags |= MD_READONLY; 930 sc->secsize = DEV_BSIZE; 931 sc->vnode = nd.ni_vp; 932 933 /* 934 * If the size is specified, override the file attributes. 935 */ 936 if (mdio->md_size) 937 sc->nsect = mdio->md_size; 938 else 939 sc->nsect = vattr.va_size / sc->secsize; /* XXX: round up ? */ 940 if (sc->nsect == 0) { 941 mddestroy(sc, td); 942 return (EINVAL); 943 } 944 error = mdsetcred(sc, td->td_ucred); 945 if (error) { 946 mddestroy(sc, td); 947 return (error); 948 } 949 mdinit(sc); 950 return (0); 951} 952 953static void 954md_zapit(void *p, int cancel) 955{ 956 if (cancel) 957 return; 958 g_wither_geom(p, ENXIO); 959} 960 961static int 962mddestroy(struct md_s *sc, struct thread *td) 963{ 964 965 GIANT_REQUIRED; 966 967 mtx_destroy(&sc->queue_mtx); 968 if (sc->gp) { 969 sc->gp->softc = NULL; 970 g_waitfor_event(md_zapit, sc->gp, M_WAITOK, sc->gp, NULL); 971 sc->gp = NULL; 972 sc->pp = NULL; 973 } 974 sc->flags |= MD_SHUTDOWN; 975 wakeup(sc); 976 while (sc->procp != NULL) 977 tsleep(&sc->procp, PRIBIO, "mddestroy", hz / 10); 978 if (sc->vnode != NULL) 979 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 980 FREAD : (FREAD|FWRITE), sc->cred, td); 981 if (sc->cred != NULL) 982 crfree(sc->cred); 983 if (sc->object != NULL) { 984 vm_object_deallocate(sc->object); 985 } 986 if (sc->indir) 987 destroy_indir(sc, sc->indir); 988 if (sc->uma) 989 uma_zdestroy(sc->uma); 990 991 /* XXX: LOCK(unique unit numbers) */ 992 LIST_REMOVE(sc, list); 993 /* XXX: UNLOCK(unique unit numbers) */ 994 free(sc, M_MD); 995 return (0); 996} 997 998static int 999mdcreate_swap(struct md_ioctl *mdio, struct thread *td) 1000{ 1001 int error; 1002 struct md_s *sc; 1003 1004 GIANT_REQUIRED; 1005 1006 if (mdio->md_options & MD_AUTOUNIT) { 1007 sc = mdnew(-1); 1008 mdio->md_unit = sc->unit; 1009 } else { 1010 sc = mdnew(mdio->md_unit); 1011 } 1012 if (sc == NULL) 1013 return (EBUSY); 1014 1015 sc->type = MD_SWAP; 1016 1017 /* 1018 * Range check. Disallow negative sizes or any size less then the 1019 * size of a page. Then round to a page. 1020 */ 1021 1022 if (mdio->md_size == 0) { 1023 mddestroy(sc, td); 1024 return (EDOM); 1025 } 1026 1027 /* 1028 * Allocate an OBJT_SWAP object. 1029 * 1030 * sc_nsect is in units of DEV_BSIZE. 1031 * sc_npage is in units of PAGE_SIZE. 1032 * 1033 * Note the truncation. 1034 */ 1035 1036 sc->secsize = DEV_BSIZE; 1037 sc->npage = mdio->md_size / (PAGE_SIZE / DEV_BSIZE); 1038 sc->nsect = sc->npage * (PAGE_SIZE / DEV_BSIZE); 1039 if (mdio->md_fwsectors != 0) 1040 sc->fwsectors = mdio->md_fwsectors; 1041 if (mdio->md_fwheads != 0) 1042 sc->fwheads = mdio->md_fwheads; 1043 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * 1044 (vm_offset_t)sc->npage, VM_PROT_DEFAULT, 0); 1045 sc->flags = mdio->md_options & MD_FORCE; 1046 if (mdio->md_options & MD_RESERVE) { 1047 if (swap_pager_reserve(sc->object, 0, sc->npage) < 0) { 1048 vm_object_deallocate(sc->object); 1049 sc->object = NULL; 1050 mddestroy(sc, td); 1051 return (EDOM); 1052 } 1053 } 1054 error = mdsetcred(sc, td->td_ucred); 1055 if (error) { 1056 mddestroy(sc, td); 1057 return (error); 1058 } 1059 mdinit(sc); 1060 if (!(mdio->md_options & MD_RESERVE)) 1061 sc->pp->flags |= G_PF_CANDELETE; 1062 return (0); 1063} 1064 1065static int 1066mddetach(int unit, struct thread *td) 1067{ 1068 struct md_s *sc; 1069 1070 sc = mdfind(unit); 1071 if (sc == NULL) 1072 return (ENOENT); 1073 if (sc->opencount != 0 && !(sc->flags & MD_FORCE)) 1074 return (EBUSY); 1075 switch(sc->type) { 1076 case MD_VNODE: 1077 case MD_SWAP: 1078 case MD_MALLOC: 1079 case MD_PRELOAD: 1080 return (mddestroy(sc, td)); 1081 default: 1082 return (EOPNOTSUPP); 1083 } 1084} 1085 1086static int 1087mdctlioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1088{ 1089 struct md_ioctl *mdio; 1090 struct md_s *sc; 1091 int i; 1092 1093 if (md_debug) 1094 printf("mdctlioctl(%s %lx %p %x %p)\n", 1095 devtoname(dev), cmd, addr, flags, td); 1096 1097 /* 1098 * We assert the version number in the individual ioctl 1099 * handlers instead of out here because (a) it is possible we 1100 * may add another ioctl in the future which doesn't read an 1101 * mdio, and (b) the correct return value for an unknown ioctl 1102 * is ENOIOCTL, not EINVAL. 1103 */ 1104 mdio = (struct md_ioctl *)addr; 1105 switch (cmd) { 1106 case MDIOCATTACH: 1107 if (mdio->md_version != MDIOVERSION) 1108 return (EINVAL); 1109 switch (mdio->md_type) { 1110 case MD_MALLOC: 1111 return (mdcreate_malloc(mdio)); 1112 case MD_PRELOAD: 1113 return (mdcreate_preload(mdio)); 1114 case MD_VNODE: 1115 return (mdcreate_vnode(mdio, td)); 1116 case MD_SWAP: 1117 return (mdcreate_swap(mdio, td)); 1118 default: 1119 return (EINVAL); 1120 } 1121 case MDIOCDETACH: 1122 if (mdio->md_version != MDIOVERSION) 1123 return (EINVAL); 1124 if (mdio->md_file != NULL || mdio->md_size != 0 || 1125 mdio->md_options != 0) 1126 return (EINVAL); 1127 return (mddetach(mdio->md_unit, td)); 1128 case MDIOCQUERY: 1129 if (mdio->md_version != MDIOVERSION) 1130 return (EINVAL); 1131 sc = mdfind(mdio->md_unit); 1132 if (sc == NULL) 1133 return (ENOENT); 1134 mdio->md_type = sc->type; 1135 mdio->md_options = sc->flags; 1136 switch (sc->type) { 1137 case MD_MALLOC: 1138 mdio->md_size = sc->nsect; 1139 break; 1140 case MD_PRELOAD: 1141 mdio->md_size = sc->nsect; 1142 mdio->md_base = (uint64_t)(intptr_t)sc->pl_ptr; 1143 break; 1144 case MD_SWAP: 1145 mdio->md_size = sc->nsect; 1146 break; 1147 case MD_VNODE: 1148 mdio->md_size = sc->nsect; 1149 /* XXX fill this in */ 1150 mdio->md_file = NULL; 1151 break; 1152 } 1153 return (0); 1154 case MDIOCLIST: 1155 i = 1; 1156 LIST_FOREACH(sc, &md_softc_list, list) { 1157 if (i == MDNPAD - 1) 1158 mdio->md_pad[i] = -1; 1159 else 1160 mdio->md_pad[i++] = sc->unit; 1161 } 1162 mdio->md_pad[0] = i - 1; 1163 return (0); 1164 default: 1165 return (ENOIOCTL); 1166 }; 1167 return (ENOIOCTL); 1168} 1169 1170static void 1171md_preloaded(u_char *image, unsigned length) 1172{ 1173 struct md_s *sc; 1174 1175 sc = mdnew(-1); 1176 if (sc == NULL) 1177 return; 1178 sc->type = MD_PRELOAD; 1179 sc->secsize = DEV_BSIZE; 1180 sc->nsect = length / DEV_BSIZE; 1181 sc->pl_ptr = image; 1182 sc->pl_len = length; 1183 if (sc->unit == 0) 1184 mdrootready = 1; 1185 mdinit(sc); 1186} 1187 1188static void 1189md_drvinit(struct g_class *mp __unused) 1190{ 1191 1192 caddr_t mod; 1193 caddr_t c; 1194 u_char *ptr, *name, *type; 1195 unsigned len; 1196 1197 mod = NULL; 1198 g_topology_unlock(); 1199#ifdef MD_ROOT_SIZE 1200 md_preloaded(mfs_root, MD_ROOT_SIZE*1024); 1201#endif 1202 while ((mod = preload_search_next_name(mod)) != NULL) { 1203 name = (char *)preload_search_info(mod, MODINFO_NAME); 1204 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1205 if (name == NULL) 1206 continue; 1207 if (type == NULL) 1208 continue; 1209 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1210 continue; 1211 c = preload_search_info(mod, MODINFO_ADDR); 1212 ptr = *(u_char **)c; 1213 c = preload_search_info(mod, MODINFO_SIZE); 1214 len = *(size_t *)c; 1215 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1216 MD_NAME, mdunits, name, len, ptr); 1217 md_preloaded(ptr, len); 1218 } 1219 status_dev = make_dev(&mdctl_cdevsw, 0xffff00ff, UID_ROOT, GID_WHEEL, 1220 0600, MDCTL_NAME); 1221 g_topology_lock(); 1222} 1223 1224static int 1225md_modevent(module_t mod, int type, void *data) 1226{ 1227 int error; 1228 struct md_s *sc; 1229 1230 switch (type) { 1231 case MOD_LOAD: 1232 break; 1233 case MOD_UNLOAD: 1234 LIST_FOREACH(sc, &md_softc_list, list) { 1235 error = mddetach(sc->unit, curthread); 1236 if (error != 0) 1237 return (error); 1238 } 1239 if (status_dev) 1240 destroy_dev(status_dev); 1241 status_dev = 0; 1242 break; 1243 default: 1244 break; 1245 } 1246 return (0); 1247} 1248 1249static moduledata_t md_mod = { 1250 MD_NAME, 1251 md_modevent, 1252 NULL 1253}; 1254DECLARE_MODULE(md, md_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE); 1255MODULE_VERSION(md, MD_MODVER); 1256 1257 1258#ifdef MD_ROOT 1259static void 1260md_takeroot(void *junk) 1261{ 1262 if (mdrootready) 1263 rootdevnames[0] = "ufs:/dev/md0"; 1264} 1265 1266SYSINIT(md_root, SI_SUB_MOUNT_ROOT, SI_ORDER_FIRST, md_takeroot, NULL); 1267#endif /* MD_ROOT */ 1268