md.c revision 248518
1287020Sdelphij/*- 2267460Sedwin * ---------------------------------------------------------------------------- 3287020Sdelphij * "THE BEER-WARE LICENSE" (Revision 42): 42742Swollman * <phk@FreeBSD.ORG> wrote this file. As long as you retain this notice you 5287020Sdelphij * can do whatever you want with this stuff. If we meet some day, and you think 6352354Sphilip * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp 7339630Sphilip * ---------------------------------------------------------------------------- 8352354Sphilip * 9368822Sphilip * $FreeBSD: head/sys/dev/md/md.c 248518 2013-03-19 14:53:23Z kib $ 10368822Sphilip * 11368822Sphilip */ 12368822Sphilip 13287020Sdelphij/*- 14287020Sdelphij * The following functions are based in the vn(4) driver: mdstart_swap(), 15339630Sphilip * mdstart_vnode(), mdcreate_swap(), mdcreate_vnode() and mddestroy(), 16287020Sdelphij * and as such under the following copyright: 17352354Sphilip * 18352354Sphilip * Copyright (c) 1988 University of Utah. 19352354Sphilip * Copyright (c) 1990, 1993 20352354Sphilip * The Regents of the University of California. All rights reserved. 21352354Sphilip * Copyright (c) 2013 The FreeBSD Foundation 22352354Sphilip * All rights reserved. 23287020Sdelphij * 24352354Sphilip * This code is derived from software contributed to Berkeley by 25339630Sphilip * the Systems Programming Group of the University of Utah Computer 26339630Sphilip * Science Department. 27339630Sphilip * 28325159Sphilip * Portions of this software were developed by Konstantin Belousov 29339630Sphilip * under sponsorship from the FreeBSD Foundation. 30342668Sphilip * 31352354Sphilip * Redistribution and use in source and binary forms, with or without 32352354Sphilip * modification, are permitted provided that the following conditions 33352354Sphilip * are met: 34342668Sphilip * 1. Redistributions of source code must retain the above copyright 35342668Sphilip * notice, this list of conditions and the following disclaimer. 362742Swollman * 2. Redistributions in binary form must reproduce the above copyright 37352354Sphilip * notice, this list of conditions and the following disclaimer in the 38352354Sphilip * documentation and/or other materials provided with the distribution. 39352354Sphilip * 4. Neither the name of the University nor the names of its contributors 40352354Sphilip * may be used to endorse or promote products derived from this software 41352354Sphilip * without specific prior written permission. 42352354Sphilip * 432742Swollman * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 442742Swollman * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 452742Swollman * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 462742Swollman * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 472742Swollman * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 482742Swollman * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 492742Swollman * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 502742Swollman * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 512742Swollman * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 522742Swollman * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 532742Swollman * SUCH DAMAGE. 542742Swollman * 552742Swollman * from: Utah Hdr: vn.c 1.13 94/04/02 562742Swollman * 572742Swollman * from: @(#)vn.c 8.6 (Berkeley) 4/1/94 582742Swollman * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03 592742Swollman */ 602742Swollman 612742Swollman#include "opt_geom.h" 6239871Sphk#include "opt_md.h" 6339871Sphk 6439898Sphk#include <sys/param.h> 65147771Sphk#include <sys/systm.h> 66181424Sedwin#include <sys/bio.h> 67233445Sedwin#include <sys/buf.h> 68287020Sdelphij#include <sys/conf.h> 69307361Sbapt#include <sys/devicestat.h> 70267460Sedwin#include <sys/fcntl.h> 71360362Sphilip#include <sys/kernel.h> 72360362Sphilip#include <sys/kthread.h> 73360362Sphilip#include <sys/limits.h> 74360362Sphilip#include <sys/linker.h> 75369144Sgit2svn#include <sys/lock.h> 76360362Sphilip#include <sys/malloc.h> 77339630Sphilip#include <sys/mdioctl.h> 78352354Sphilip#include <sys/mount.h> 79369144Sgit2svn#include <sys/mutex.h> 80339630Sphilip#include <sys/sx.h> 81369144Sgit2svn#include <sys/namei.h> 82369144Sgit2svn#include <sys/proc.h> 83#include <sys/queue.h> 84#include <sys/rwlock.h> 85#include <sys/sbuf.h> 86#include <sys/sched.h> 87#include <sys/sf_buf.h> 88#include <sys/sysctl.h> 89#include <sys/vnode.h> 90 91#include <geom/geom.h> 92 93#include <vm/vm.h> 94#include <vm/vm_param.h> 95#include <vm/vm_object.h> 96#include <vm/vm_page.h> 97#include <vm/vm_pager.h> 98#include <vm/swap_pager.h> 99#include <vm/uma.h> 100 101#define MD_MODVER 1 102 103#define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */ 104#define MD_EXITING 0x20000 /* Worker thread is exiting. */ 105 106#ifndef MD_NSECT 107#define MD_NSECT (10000 * 2) 108#endif 109 110static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk"); 111static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors"); 112 113static int md_debug; 114SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, 115 "Enable md(4) debug messages"); 116static int md_malloc_wait; 117SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0, 118 "Allow malloc to wait for memory allocations"); 119 120#if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE) 121#define MD_ROOT_FSTYPE "ufs" 122#endif 123 124#if defined(MD_ROOT) && defined(MD_ROOT_SIZE) 125/* 126 * Preloaded image gets put here. 127 * Applications that patch the object with the image can determine 128 * the size looking at the start and end markers (strings), 129 * so we want them contiguous. 130 */ 131static struct { 132 u_char start[MD_ROOT_SIZE*1024]; 133 u_char end[128]; 134} mfs_root = { 135 .start = "MFS Filesystem goes here", 136 .end = "MFS Filesystem had better STOP here", 137}; 138#endif 139 140static g_init_t g_md_init; 141static g_fini_t g_md_fini; 142static g_start_t g_md_start; 143static g_access_t g_md_access; 144static void g_md_dumpconf(struct sbuf *sb, const char *indent, 145 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp); 146 147static struct cdev *status_dev = 0; 148static struct sx md_sx; 149static struct unrhdr *md_uh; 150 151static d_ioctl_t mdctlioctl; 152 153static struct cdevsw mdctl_cdevsw = { 154 .d_version = D_VERSION, 155 .d_ioctl = mdctlioctl, 156 .d_name = MD_NAME, 157}; 158 159struct g_class g_md_class = { 160 .name = "MD", 161 .version = G_VERSION, 162 .init = g_md_init, 163 .fini = g_md_fini, 164 .start = g_md_start, 165 .access = g_md_access, 166 .dumpconf = g_md_dumpconf, 167}; 168 169DECLARE_GEOM_CLASS(g_md_class, g_md); 170 171 172static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list); 173 174#define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 175#define NMASK (NINDIR-1) 176static int nshift; 177 178static int md_vnode_pbuf_freecnt; 179 180struct indir { 181 uintptr_t *array; 182 u_int total; 183 u_int used; 184 u_int shift; 185}; 186 187struct md_s { 188 int unit; 189 LIST_ENTRY(md_s) list; 190 struct bio_queue_head bio_queue; 191 struct mtx queue_mtx; 192 struct cdev *dev; 193 enum md_types type; 194 off_t mediasize; 195 unsigned sectorsize; 196 unsigned opencount; 197 unsigned fwheads; 198 unsigned fwsectors; 199 unsigned flags; 200 char name[20]; 201 struct proc *procp; 202 struct g_geom *gp; 203 struct g_provider *pp; 204 int (*start)(struct md_s *sc, struct bio *bp); 205 struct devstat *devstat; 206 207 /* MD_MALLOC related fields */ 208 struct indir *indir; 209 uma_zone_t uma; 210 211 /* MD_PRELOAD related fields */ 212 u_char *pl_ptr; 213 size_t pl_len; 214 215 /* MD_VNODE related fields */ 216 struct vnode *vnode; 217 char file[PATH_MAX]; 218 struct ucred *cred; 219 220 /* MD_SWAP related fields */ 221 vm_object_t object; 222}; 223 224static struct indir * 225new_indir(u_int shift) 226{ 227 struct indir *ip; 228 229 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT) 230 | M_ZERO); 231 if (ip == NULL) 232 return (NULL); 233 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 234 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO); 235 if (ip->array == NULL) { 236 free(ip, M_MD); 237 return (NULL); 238 } 239 ip->total = NINDIR; 240 ip->shift = shift; 241 return (ip); 242} 243 244static void 245del_indir(struct indir *ip) 246{ 247 248 free(ip->array, M_MDSECT); 249 free(ip, M_MD); 250} 251 252static void 253destroy_indir(struct md_s *sc, struct indir *ip) 254{ 255 int i; 256 257 for (i = 0; i < NINDIR; i++) { 258 if (!ip->array[i]) 259 continue; 260 if (ip->shift) 261 destroy_indir(sc, (struct indir*)(ip->array[i])); 262 else if (ip->array[i] > 255) 263 uma_zfree(sc->uma, (void *)(ip->array[i])); 264 } 265 del_indir(ip); 266} 267 268/* 269 * This function does the math and allocates the top level "indir" structure 270 * for a device of "size" sectors. 271 */ 272 273static struct indir * 274dimension(off_t size) 275{ 276 off_t rcnt; 277 struct indir *ip; 278 int layer; 279 280 rcnt = size; 281 layer = 0; 282 while (rcnt > NINDIR) { 283 rcnt /= NINDIR; 284 layer++; 285 } 286 287 /* 288 * XXX: the top layer is probably not fully populated, so we allocate 289 * too much space for ip->array in here. 290 */ 291 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 292 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 293 M_MDSECT, M_WAITOK | M_ZERO); 294 ip->total = NINDIR; 295 ip->shift = layer * nshift; 296 return (ip); 297} 298 299/* 300 * Read a given sector 301 */ 302 303static uintptr_t 304s_read(struct indir *ip, off_t offset) 305{ 306 struct indir *cip; 307 int idx; 308 uintptr_t up; 309 310 if (md_debug > 1) 311 printf("s_read(%jd)\n", (intmax_t)offset); 312 up = 0; 313 for (cip = ip; cip != NULL;) { 314 if (cip->shift) { 315 idx = (offset >> cip->shift) & NMASK; 316 up = cip->array[idx]; 317 cip = (struct indir *)up; 318 continue; 319 } 320 idx = offset & NMASK; 321 return (cip->array[idx]); 322 } 323 return (0); 324} 325 326/* 327 * Write a given sector, prune the tree if the value is 0 328 */ 329 330static int 331s_write(struct indir *ip, off_t offset, uintptr_t ptr) 332{ 333 struct indir *cip, *lip[10]; 334 int idx, li; 335 uintptr_t up; 336 337 if (md_debug > 1) 338 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 339 up = 0; 340 li = 0; 341 cip = ip; 342 for (;;) { 343 lip[li++] = cip; 344 if (cip->shift) { 345 idx = (offset >> cip->shift) & NMASK; 346 up = cip->array[idx]; 347 if (up != 0) { 348 cip = (struct indir *)up; 349 continue; 350 } 351 /* Allocate branch */ 352 cip->array[idx] = 353 (uintptr_t)new_indir(cip->shift - nshift); 354 if (cip->array[idx] == 0) 355 return (ENOSPC); 356 cip->used++; 357 up = cip->array[idx]; 358 cip = (struct indir *)up; 359 continue; 360 } 361 /* leafnode */ 362 idx = offset & NMASK; 363 up = cip->array[idx]; 364 if (up != 0) 365 cip->used--; 366 cip->array[idx] = ptr; 367 if (ptr != 0) 368 cip->used++; 369 break; 370 } 371 if (cip->used != 0 || li == 1) 372 return (0); 373 li--; 374 while (cip->used == 0 && cip != ip) { 375 li--; 376 idx = (offset >> lip[li]->shift) & NMASK; 377 up = lip[li]->array[idx]; 378 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 379 del_indir(cip); 380 lip[li]->array[idx] = 0; 381 lip[li]->used--; 382 cip = lip[li]; 383 } 384 return (0); 385} 386 387 388static int 389g_md_access(struct g_provider *pp, int r, int w, int e) 390{ 391 struct md_s *sc; 392 393 sc = pp->geom->softc; 394 if (sc == NULL) { 395 if (r <= 0 && w <= 0 && e <= 0) 396 return (0); 397 return (ENXIO); 398 } 399 r += pp->acr; 400 w += pp->acw; 401 e += pp->ace; 402 if ((sc->flags & MD_READONLY) != 0 && w > 0) 403 return (EROFS); 404 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 405 sc->opencount = 1; 406 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 407 sc->opencount = 0; 408 } 409 return (0); 410} 411 412static void 413g_md_start(struct bio *bp) 414{ 415 struct md_s *sc; 416 417 sc = bp->bio_to->geom->softc; 418 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) 419 devstat_start_transaction_bio(sc->devstat, bp); 420 mtx_lock(&sc->queue_mtx); 421 bioq_disksort(&sc->bio_queue, bp); 422 mtx_unlock(&sc->queue_mtx); 423 wakeup(sc); 424} 425 426#define MD_MALLOC_MOVE_ZERO 1 427#define MD_MALLOC_MOVE_FILL 2 428#define MD_MALLOC_MOVE_READ 3 429#define MD_MALLOC_MOVE_WRITE 4 430#define MD_MALLOC_MOVE_CMP 5 431 432static int 433md_malloc_move(vm_page_t **mp, int *ma_offs, unsigned sectorsize, 434 void *ptr, u_char fill, int op) 435{ 436 struct sf_buf *sf; 437 vm_page_t m, *mp1; 438 char *p, first; 439 off_t *uc; 440 unsigned n; 441 int error, i, ma_offs1, sz, first_read; 442 443 m = NULL; 444 error = 0; 445 sf = NULL; 446 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */ 447 first = 0; 448 first_read = 0; 449 uc = ptr; 450 mp1 = *mp; 451 ma_offs1 = *ma_offs; 452 /* } */ 453 sched_pin(); 454 for (n = sectorsize; n != 0; n -= sz) { 455 sz = imin(PAGE_SIZE - *ma_offs, n); 456 if (m != **mp) { 457 if (sf != NULL) 458 sf_buf_free(sf); 459 m = **mp; 460 sf = sf_buf_alloc(m, SFB_CPUPRIVATE | 461 (md_malloc_wait ? 0 : SFB_NOWAIT)); 462 if (sf == NULL) { 463 error = ENOMEM; 464 break; 465 } 466 } 467 p = (char *)sf_buf_kva(sf) + *ma_offs; 468 switch (op) { 469 case MD_MALLOC_MOVE_ZERO: 470 bzero(p, sz); 471 break; 472 case MD_MALLOC_MOVE_FILL: 473 memset(p, fill, sz); 474 break; 475 case MD_MALLOC_MOVE_READ: 476 bcopy(ptr, p, sz); 477 cpu_flush_dcache(p, sz); 478 break; 479 case MD_MALLOC_MOVE_WRITE: 480 bcopy(p, ptr, sz); 481 break; 482 case MD_MALLOC_MOVE_CMP: 483 for (i = 0; i < sz; i++, p++) { 484 if (!first_read) { 485 *uc = (u_char)*p; 486 first = *p; 487 first_read = 1; 488 } else if (*p != first) { 489 error = EDOOFUS; 490 break; 491 } 492 } 493 break; 494 default: 495 KASSERT(0, ("md_malloc_move unknown op %d\n", op)); 496 break; 497 } 498 if (error != 0) 499 break; 500 *ma_offs += sz; 501 *ma_offs %= PAGE_SIZE; 502 if (*ma_offs == 0) 503 (*mp)++; 504 ptr = (char *)ptr + sz; 505 } 506 507 if (sf != NULL) 508 sf_buf_free(sf); 509 sched_unpin(); 510 if (op == MD_MALLOC_MOVE_CMP && error != 0) { 511 *mp = mp1; 512 *ma_offs = ma_offs1; 513 } 514 return (error); 515} 516 517static int 518mdstart_malloc(struct md_s *sc, struct bio *bp) 519{ 520 u_char *dst; 521 vm_page_t *m; 522 int i, error, error1, ma_offs, notmapped; 523 off_t secno, nsec, uc; 524 uintptr_t sp, osp; 525 526 switch (bp->bio_cmd) { 527 case BIO_READ: 528 case BIO_WRITE: 529 case BIO_DELETE: 530 break; 531 default: 532 return (EOPNOTSUPP); 533 } 534 535 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0; 536 if (notmapped) { 537 m = bp->bio_ma; 538 ma_offs = bp->bio_ma_offset; 539 dst = NULL; 540 } else { 541 dst = bp->bio_data; 542 } 543 544 nsec = bp->bio_length / sc->sectorsize; 545 secno = bp->bio_offset / sc->sectorsize; 546 error = 0; 547 while (nsec--) { 548 osp = s_read(sc->indir, secno); 549 if (bp->bio_cmd == BIO_DELETE) { 550 if (osp != 0) 551 error = s_write(sc->indir, secno, 0); 552 } else if (bp->bio_cmd == BIO_READ) { 553 if (osp == 0) { 554 if (notmapped) { 555 error = md_malloc_move(&m, &ma_offs, 556 sc->sectorsize, NULL, 0, 557 MD_MALLOC_MOVE_ZERO); 558 } else 559 bzero(dst, sc->sectorsize); 560 } else if (osp <= 255) { 561 if (notmapped) { 562 error = md_malloc_move(&m, &ma_offs, 563 sc->sectorsize, NULL, osp, 564 MD_MALLOC_MOVE_FILL); 565 } else 566 memset(dst, osp, sc->sectorsize); 567 } else { 568 if (notmapped) { 569 error = md_malloc_move(&m, &ma_offs, 570 sc->sectorsize, (void *)osp, 0, 571 MD_MALLOC_MOVE_READ); 572 } else { 573 bcopy((void *)osp, dst, sc->sectorsize); 574 cpu_flush_dcache(dst, sc->sectorsize); 575 } 576 } 577 osp = 0; 578 } else if (bp->bio_cmd == BIO_WRITE) { 579 if (sc->flags & MD_COMPRESS) { 580 if (notmapped) { 581 error1 = md_malloc_move(&m, &ma_offs, 582 sc->sectorsize, &uc, 0, 583 MD_MALLOC_MOVE_CMP); 584 i = error1 == 0 ? sc->sectorsize : 0; 585 } else { 586 uc = dst[0]; 587 for (i = 1; i < sc->sectorsize; i++) { 588 if (dst[i] != uc) 589 break; 590 } 591 } 592 } else { 593 i = 0; 594 uc = 0; 595 } 596 if (i == sc->sectorsize) { 597 if (osp != uc) 598 error = s_write(sc->indir, secno, uc); 599 } else { 600 if (osp <= 255) { 601 sp = (uintptr_t)uma_zalloc(sc->uma, 602 md_malloc_wait ? M_WAITOK : 603 M_NOWAIT); 604 if (sp == 0) { 605 error = ENOSPC; 606 break; 607 } 608 if (notmapped) { 609 error = md_malloc_move(&m, 610 &ma_offs, sc->sectorsize, 611 (void *)sp, 0, 612 MD_MALLOC_MOVE_WRITE); 613 } else { 614 bcopy(dst, (void *)sp, 615 sc->sectorsize); 616 } 617 error = s_write(sc->indir, secno, sp); 618 } else { 619 if (notmapped) { 620 error = md_malloc_move(&m, 621 &ma_offs, sc->sectorsize, 622 (void *)osp, 0, 623 MD_MALLOC_MOVE_WRITE); 624 } else { 625 bcopy(dst, (void *)osp, 626 sc->sectorsize); 627 } 628 osp = 0; 629 } 630 } 631 } else { 632 error = EOPNOTSUPP; 633 } 634 if (osp > 255) 635 uma_zfree(sc->uma, (void*)osp); 636 if (error != 0) 637 break; 638 secno++; 639 if (!notmapped) 640 dst += sc->sectorsize; 641 } 642 bp->bio_resid = 0; 643 return (error); 644} 645 646static int 647mdstart_preload(struct md_s *sc, struct bio *bp) 648{ 649 650 switch (bp->bio_cmd) { 651 case BIO_READ: 652 bcopy(sc->pl_ptr + bp->bio_offset, bp->bio_data, 653 bp->bio_length); 654 cpu_flush_dcache(bp->bio_data, bp->bio_length); 655 break; 656 case BIO_WRITE: 657 bcopy(bp->bio_data, sc->pl_ptr + bp->bio_offset, 658 bp->bio_length); 659 break; 660 } 661 bp->bio_resid = 0; 662 return (0); 663} 664 665static int 666mdstart_vnode(struct md_s *sc, struct bio *bp) 667{ 668 int error; 669 struct uio auio; 670 struct iovec aiov; 671 struct mount *mp; 672 struct vnode *vp; 673 struct buf *pb; 674 struct thread *td; 675 off_t end, zerosize; 676 677 switch (bp->bio_cmd) { 678 case BIO_READ: 679 case BIO_WRITE: 680 case BIO_DELETE: 681 case BIO_FLUSH: 682 break; 683 default: 684 return (EOPNOTSUPP); 685 } 686 687 td = curthread; 688 vp = sc->vnode; 689 690 /* 691 * VNODE I/O 692 * 693 * If an error occurs, we set BIO_ERROR but we do not set 694 * B_INVAL because (for a write anyway), the buffer is 695 * still valid. 696 */ 697 698 if (bp->bio_cmd == BIO_FLUSH) { 699 (void) vn_start_write(vp, &mp, V_WAIT); 700 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 701 error = VOP_FSYNC(vp, MNT_WAIT, td); 702 VOP_UNLOCK(vp, 0); 703 vn_finished_write(mp); 704 return (error); 705 } 706 707 bzero(&auio, sizeof(auio)); 708 709 /* 710 * Special case for BIO_DELETE. On the surface, this is very 711 * similar to BIO_WRITE, except that we write from our own 712 * fixed-length buffer, so we have to loop. The net result is 713 * that the two cases end up having very little in common. 714 */ 715 if (bp->bio_cmd == BIO_DELETE) { 716 zerosize = ZERO_REGION_SIZE - 717 (ZERO_REGION_SIZE % sc->sectorsize); 718 auio.uio_iov = &aiov; 719 auio.uio_iovcnt = 1; 720 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 721 auio.uio_segflg = UIO_SYSSPACE; 722 auio.uio_rw = UIO_WRITE; 723 auio.uio_td = td; 724 end = bp->bio_offset + bp->bio_length; 725 (void) vn_start_write(vp, &mp, V_WAIT); 726 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 727 error = 0; 728 while (auio.uio_offset < end) { 729 aiov.iov_base = __DECONST(void *, zero_region); 730 aiov.iov_len = end - auio.uio_offset; 731 if (aiov.iov_len > zerosize) 732 aiov.iov_len = zerosize; 733 auio.uio_resid = aiov.iov_len; 734 error = VOP_WRITE(vp, &auio, 735 sc->flags & MD_ASYNC ? 0 : IO_SYNC, sc->cred); 736 if (error != 0) 737 break; 738 } 739 VOP_UNLOCK(vp, 0); 740 vn_finished_write(mp); 741 bp->bio_resid = end - auio.uio_offset; 742 return (error); 743 } 744 745 KASSERT(bp->bio_length <= MAXPHYS, ("bio_length %jd", 746 (uintmax_t)bp->bio_length)); 747 if ((bp->bio_flags & BIO_UNMAPPED) == 0) { 748 pb = NULL; 749 aiov.iov_base = bp->bio_data; 750 } else { 751 pb = getpbuf(&md_vnode_pbuf_freecnt); 752 pmap_qenter((vm_offset_t)pb->b_data, bp->bio_ma, bp->bio_ma_n); 753 aiov.iov_base = (void *)((vm_offset_t)pb->b_data + 754 bp->bio_ma_offset); 755 } 756 aiov.iov_len = bp->bio_length; 757 auio.uio_iov = &aiov; 758 auio.uio_iovcnt = 1; 759 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 760 auio.uio_segflg = UIO_SYSSPACE; 761 if (bp->bio_cmd == BIO_READ) 762 auio.uio_rw = UIO_READ; 763 else if (bp->bio_cmd == BIO_WRITE) 764 auio.uio_rw = UIO_WRITE; 765 else 766 panic("wrong BIO_OP in mdstart_vnode"); 767 auio.uio_resid = bp->bio_length; 768 auio.uio_td = td; 769 /* 770 * When reading set IO_DIRECT to try to avoid double-caching 771 * the data. When writing IO_DIRECT is not optimal. 772 */ 773 if (bp->bio_cmd == BIO_READ) { 774 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 775 error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred); 776 VOP_UNLOCK(vp, 0); 777 } else { 778 (void) vn_start_write(vp, &mp, V_WAIT); 779 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 780 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC, 781 sc->cred); 782 VOP_UNLOCK(vp, 0); 783 vn_finished_write(mp); 784 } 785 if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 786 pmap_qremove((vm_offset_t)pb->b_data, bp->bio_ma_n); 787 relpbuf(pb, &md_vnode_pbuf_freecnt); 788 } 789 bp->bio_resid = auio.uio_resid; 790 return (error); 791} 792 793static int 794mdstart_swap(struct md_s *sc, struct bio *bp) 795{ 796 vm_page_t m; 797 u_char *p; 798 vm_pindex_t i, lastp; 799 int rv, ma_offs, offs, len, lastend; 800 801 switch (bp->bio_cmd) { 802 case BIO_READ: 803 case BIO_WRITE: 804 case BIO_DELETE: 805 break; 806 default: 807 return (EOPNOTSUPP); 808 } 809 810 p = bp->bio_data; 811 ma_offs = (bp->bio_flags & BIO_UNMAPPED) == 0 ? 0 : bp->bio_ma_offset; 812 813 /* 814 * offs is the offset at which to start operating on the 815 * next (ie, first) page. lastp is the last page on 816 * which we're going to operate. lastend is the ending 817 * position within that last page (ie, PAGE_SIZE if 818 * we're operating on complete aligned pages). 819 */ 820 offs = bp->bio_offset % PAGE_SIZE; 821 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE; 822 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1; 823 824 rv = VM_PAGER_OK; 825 VM_OBJECT_WLOCK(sc->object); 826 vm_object_pip_add(sc->object, 1); 827 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) { 828 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs; 829 m = vm_page_grab(sc->object, i, VM_ALLOC_NORMAL | 830 VM_ALLOC_RETRY); 831 if (bp->bio_cmd == BIO_READ) { 832 if (m->valid != VM_PAGE_BITS_ALL) 833 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 834 if (rv == VM_PAGER_ERROR) { 835 vm_page_wakeup(m); 836 break; 837 } else if (rv == VM_PAGER_FAIL) { 838 /* 839 * Pager does not have the page. Zero 840 * the allocated page, and mark it as 841 * valid. Do not set dirty, the page 842 * can be recreated if thrown out. 843 */ 844 pmap_zero_page(m); 845 m->valid = VM_PAGE_BITS_ALL; 846 } 847 if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 848 pmap_copy_pages(&m, offs, bp->bio_ma, 849 ma_offs, len); 850 } else { 851 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len); 852 cpu_flush_dcache(p, len); 853 } 854 } else if (bp->bio_cmd == BIO_WRITE) { 855 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 856 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 857 if (rv == VM_PAGER_ERROR) { 858 vm_page_wakeup(m); 859 break; 860 } 861 if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 862 pmap_copy_pages(bp->bio_ma, ma_offs, &m, 863 offs, len); 864 } else { 865 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len); 866 } 867 m->valid = VM_PAGE_BITS_ALL; 868 } else if (bp->bio_cmd == BIO_DELETE) { 869 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 870 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 871 if (rv == VM_PAGER_ERROR) { 872 vm_page_wakeup(m); 873 break; 874 } 875 if (len != PAGE_SIZE) { 876 pmap_zero_page_area(m, offs, len); 877 vm_page_clear_dirty(m, offs, len); 878 m->valid = VM_PAGE_BITS_ALL; 879 } else 880 vm_pager_page_unswapped(m); 881 } 882 vm_page_wakeup(m); 883 vm_page_lock(m); 884 if (bp->bio_cmd == BIO_DELETE && len == PAGE_SIZE) 885 vm_page_free(m); 886 else 887 vm_page_activate(m); 888 vm_page_unlock(m); 889 if (bp->bio_cmd == BIO_WRITE) 890 vm_page_dirty(m); 891 892 /* Actions on further pages start at offset 0 */ 893 p += PAGE_SIZE - offs; 894 offs = 0; 895 ma_offs += len; 896 } 897 vm_object_pip_subtract(sc->object, 1); 898 VM_OBJECT_WUNLOCK(sc->object); 899 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC); 900} 901 902static void 903md_kthread(void *arg) 904{ 905 struct md_s *sc; 906 struct bio *bp; 907 int error; 908 909 sc = arg; 910 thread_lock(curthread); 911 sched_prio(curthread, PRIBIO); 912 thread_unlock(curthread); 913 if (sc->type == MD_VNODE) 914 curthread->td_pflags |= TDP_NORUNNINGBUF; 915 916 for (;;) { 917 mtx_lock(&sc->queue_mtx); 918 if (sc->flags & MD_SHUTDOWN) { 919 sc->flags |= MD_EXITING; 920 mtx_unlock(&sc->queue_mtx); 921 kproc_exit(0); 922 } 923 bp = bioq_takefirst(&sc->bio_queue); 924 if (!bp) { 925 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 926 continue; 927 } 928 mtx_unlock(&sc->queue_mtx); 929 if (bp->bio_cmd == BIO_GETATTR) { 930 if ((sc->fwsectors && sc->fwheads && 931 (g_handleattr_int(bp, "GEOM::fwsectors", 932 sc->fwsectors) || 933 g_handleattr_int(bp, "GEOM::fwheads", 934 sc->fwheads))) || 935 g_handleattr_int(bp, "GEOM::candelete", 1)) 936 error = -1; 937 else 938 error = EOPNOTSUPP; 939 } else { 940 error = sc->start(sc, bp); 941 } 942 943 if (error != -1) { 944 bp->bio_completed = bp->bio_length; 945 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) 946 devstat_end_transaction_bio(sc->devstat, bp); 947 g_io_deliver(bp, error); 948 } 949 } 950} 951 952static struct md_s * 953mdfind(int unit) 954{ 955 struct md_s *sc; 956 957 LIST_FOREACH(sc, &md_softc_list, list) { 958 if (sc->unit == unit) 959 break; 960 } 961 return (sc); 962} 963 964static struct md_s * 965mdnew(int unit, int *errp, enum md_types type) 966{ 967 struct md_s *sc; 968 int error; 969 970 *errp = 0; 971 if (unit == -1) 972 unit = alloc_unr(md_uh); 973 else 974 unit = alloc_unr_specific(md_uh, unit); 975 976 if (unit == -1) { 977 *errp = EBUSY; 978 return (NULL); 979 } 980 981 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 982 sc->type = type; 983 bioq_init(&sc->bio_queue); 984 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 985 sc->unit = unit; 986 sprintf(sc->name, "md%d", unit); 987 LIST_INSERT_HEAD(&md_softc_list, sc, list); 988 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 989 if (error == 0) 990 return (sc); 991 LIST_REMOVE(sc, list); 992 mtx_destroy(&sc->queue_mtx); 993 free_unr(md_uh, sc->unit); 994 free(sc, M_MD); 995 *errp = error; 996 return (NULL); 997} 998 999static void 1000mdinit(struct md_s *sc) 1001{ 1002 struct g_geom *gp; 1003 struct g_provider *pp; 1004 1005 g_topology_lock(); 1006 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 1007 gp->softc = sc; 1008 pp = g_new_providerf(gp, "md%d", sc->unit); 1009 pp->mediasize = sc->mediasize; 1010 pp->sectorsize = sc->sectorsize; 1011 pp->flags |= G_PF_ACCEPT_UNMAPPED; 1012 sc->gp = gp; 1013 sc->pp = pp; 1014 g_error_provider(pp, 0); 1015 g_topology_unlock(); 1016 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize, 1017 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX); 1018} 1019 1020static int 1021mdcreate_malloc(struct md_s *sc, struct md_ioctl *mdio) 1022{ 1023 uintptr_t sp; 1024 int error; 1025 off_t u; 1026 1027 error = 0; 1028 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 1029 return (EINVAL); 1030 if (mdio->md_sectorsize != 0 && !powerof2(mdio->md_sectorsize)) 1031 return (EINVAL); 1032 /* Compression doesn't make sense if we have reserved space */ 1033 if (mdio->md_options & MD_RESERVE) 1034 mdio->md_options &= ~MD_COMPRESS; 1035 if (mdio->md_fwsectors != 0) 1036 sc->fwsectors = mdio->md_fwsectors; 1037 if (mdio->md_fwheads != 0) 1038 sc->fwheads = mdio->md_fwheads; 1039 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 1040 sc->indir = dimension(sc->mediasize / sc->sectorsize); 1041 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL, 1042 0x1ff, 0); 1043 if (mdio->md_options & MD_RESERVE) { 1044 off_t nsectors; 1045 1046 nsectors = sc->mediasize / sc->sectorsize; 1047 for (u = 0; u < nsectors; u++) { 1048 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ? 1049 M_WAITOK : M_NOWAIT) | M_ZERO); 1050 if (sp != 0) 1051 error = s_write(sc->indir, u, sp); 1052 else 1053 error = ENOMEM; 1054 if (error != 0) 1055 break; 1056 } 1057 } 1058 return (error); 1059} 1060 1061 1062static int 1063mdsetcred(struct md_s *sc, struct ucred *cred) 1064{ 1065 char *tmpbuf; 1066 int error = 0; 1067 1068 /* 1069 * Set credits in our softc 1070 */ 1071 1072 if (sc->cred) 1073 crfree(sc->cred); 1074 sc->cred = crhold(cred); 1075 1076 /* 1077 * Horrible kludge to establish credentials for NFS XXX. 1078 */ 1079 1080 if (sc->vnode) { 1081 struct uio auio; 1082 struct iovec aiov; 1083 1084 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK); 1085 bzero(&auio, sizeof(auio)); 1086 1087 aiov.iov_base = tmpbuf; 1088 aiov.iov_len = sc->sectorsize; 1089 auio.uio_iov = &aiov; 1090 auio.uio_iovcnt = 1; 1091 auio.uio_offset = 0; 1092 auio.uio_rw = UIO_READ; 1093 auio.uio_segflg = UIO_SYSSPACE; 1094 auio.uio_resid = aiov.iov_len; 1095 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 1096 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 1097 VOP_UNLOCK(sc->vnode, 0); 1098 free(tmpbuf, M_TEMP); 1099 } 1100 return (error); 1101} 1102 1103static int 1104mdcreate_vnode(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 1105{ 1106 struct vattr vattr; 1107 struct nameidata nd; 1108 char *fname; 1109 int error, flags; 1110 1111 /* 1112 * Kernel-originated requests must have the filename appended 1113 * to the mdio structure to protect against malicious software. 1114 */ 1115 fname = mdio->md_file; 1116 if ((void *)fname != (void *)(mdio + 1)) { 1117 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL); 1118 if (error != 0) 1119 return (error); 1120 } else 1121 strlcpy(sc->file, fname, sizeof(sc->file)); 1122 1123 /* 1124 * If the user specified that this is a read only device, don't 1125 * set the FWRITE mask before trying to open the backing store. 1126 */ 1127 flags = FREAD | ((mdio->md_options & MD_READONLY) ? 0 : FWRITE); 1128 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td); 1129 error = vn_open(&nd, &flags, 0, NULL); 1130 if (error != 0) 1131 return (error); 1132 NDFREE(&nd, NDF_ONLY_PNBUF); 1133 if (nd.ni_vp->v_type != VREG) { 1134 error = EINVAL; 1135 goto bad; 1136 } 1137 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred); 1138 if (error != 0) 1139 goto bad; 1140 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) { 1141 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY); 1142 if (nd.ni_vp->v_iflag & VI_DOOMED) { 1143 /* Forced unmount. */ 1144 error = EBADF; 1145 goto bad; 1146 } 1147 } 1148 nd.ni_vp->v_vflag |= VV_MD; 1149 VOP_UNLOCK(nd.ni_vp, 0); 1150 1151 if (mdio->md_fwsectors != 0) 1152 sc->fwsectors = mdio->md_fwsectors; 1153 if (mdio->md_fwheads != 0) 1154 sc->fwheads = mdio->md_fwheads; 1155 sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC); 1156 if (!(flags & FWRITE)) 1157 sc->flags |= MD_READONLY; 1158 sc->vnode = nd.ni_vp; 1159 1160 error = mdsetcred(sc, td->td_ucred); 1161 if (error != 0) { 1162 sc->vnode = NULL; 1163 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY); 1164 nd.ni_vp->v_vflag &= ~VV_MD; 1165 goto bad; 1166 } 1167 return (0); 1168bad: 1169 VOP_UNLOCK(nd.ni_vp, 0); 1170 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td); 1171 return (error); 1172} 1173 1174static int 1175mddestroy(struct md_s *sc, struct thread *td) 1176{ 1177 1178 if (sc->gp) { 1179 sc->gp->softc = NULL; 1180 g_topology_lock(); 1181 g_wither_geom(sc->gp, ENXIO); 1182 g_topology_unlock(); 1183 sc->gp = NULL; 1184 sc->pp = NULL; 1185 } 1186 if (sc->devstat) { 1187 devstat_remove_entry(sc->devstat); 1188 sc->devstat = NULL; 1189 } 1190 mtx_lock(&sc->queue_mtx); 1191 sc->flags |= MD_SHUTDOWN; 1192 wakeup(sc); 1193 while (!(sc->flags & MD_EXITING)) 1194 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10); 1195 mtx_unlock(&sc->queue_mtx); 1196 mtx_destroy(&sc->queue_mtx); 1197 if (sc->vnode != NULL) { 1198 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 1199 sc->vnode->v_vflag &= ~VV_MD; 1200 VOP_UNLOCK(sc->vnode, 0); 1201 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 1202 FREAD : (FREAD|FWRITE), sc->cred, td); 1203 } 1204 if (sc->cred != NULL) 1205 crfree(sc->cred); 1206 if (sc->object != NULL) 1207 vm_object_deallocate(sc->object); 1208 if (sc->indir) 1209 destroy_indir(sc, sc->indir); 1210 if (sc->uma) 1211 uma_zdestroy(sc->uma); 1212 1213 LIST_REMOVE(sc, list); 1214 free_unr(md_uh, sc->unit); 1215 free(sc, M_MD); 1216 return (0); 1217} 1218 1219static int 1220mdresize(struct md_s *sc, struct md_ioctl *mdio) 1221{ 1222 int error, res; 1223 vm_pindex_t oldpages, newpages; 1224 1225 switch (sc->type) { 1226 case MD_VNODE: 1227 break; 1228 case MD_SWAP: 1229 if (mdio->md_mediasize <= 0 || 1230 (mdio->md_mediasize % PAGE_SIZE) != 0) 1231 return (EDOM); 1232 oldpages = OFF_TO_IDX(round_page(sc->mediasize)); 1233 newpages = OFF_TO_IDX(round_page(mdio->md_mediasize)); 1234 if (newpages < oldpages) { 1235 VM_OBJECT_WLOCK(sc->object); 1236 vm_object_page_remove(sc->object, newpages, 0, 0); 1237 swap_pager_freespace(sc->object, newpages, 1238 oldpages - newpages); 1239 swap_release_by_cred(IDX_TO_OFF(oldpages - 1240 newpages), sc->cred); 1241 sc->object->charge = IDX_TO_OFF(newpages); 1242 sc->object->size = newpages; 1243 VM_OBJECT_WUNLOCK(sc->object); 1244 } else if (newpages > oldpages) { 1245 res = swap_reserve_by_cred(IDX_TO_OFF(newpages - 1246 oldpages), sc->cred); 1247 if (!res) 1248 return (ENOMEM); 1249 if ((mdio->md_options & MD_RESERVE) || 1250 (sc->flags & MD_RESERVE)) { 1251 error = swap_pager_reserve(sc->object, 1252 oldpages, newpages - oldpages); 1253 if (error < 0) { 1254 swap_release_by_cred( 1255 IDX_TO_OFF(newpages - oldpages), 1256 sc->cred); 1257 return (EDOM); 1258 } 1259 } 1260 VM_OBJECT_WLOCK(sc->object); 1261 sc->object->charge = IDX_TO_OFF(newpages); 1262 sc->object->size = newpages; 1263 VM_OBJECT_WUNLOCK(sc->object); 1264 } 1265 break; 1266 default: 1267 return (EOPNOTSUPP); 1268 } 1269 1270 sc->mediasize = mdio->md_mediasize; 1271 g_topology_lock(); 1272 g_resize_provider(sc->pp, sc->mediasize); 1273 g_topology_unlock(); 1274 return (0); 1275} 1276 1277static int 1278mdcreate_swap(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 1279{ 1280 vm_ooffset_t npage; 1281 int error; 1282 1283 /* 1284 * Range check. Disallow negative sizes or any size less then the 1285 * size of a page. Then round to a page. 1286 */ 1287 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0) 1288 return (EDOM); 1289 1290 /* 1291 * Allocate an OBJT_SWAP object. 1292 * 1293 * Note the truncation. 1294 */ 1295 1296 npage = mdio->md_mediasize / PAGE_SIZE; 1297 if (mdio->md_fwsectors != 0) 1298 sc->fwsectors = mdio->md_fwsectors; 1299 if (mdio->md_fwheads != 0) 1300 sc->fwheads = mdio->md_fwheads; 1301 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage, 1302 VM_PROT_DEFAULT, 0, td->td_ucred); 1303 if (sc->object == NULL) 1304 return (ENOMEM); 1305 sc->flags = mdio->md_options & (MD_FORCE | MD_RESERVE); 1306 if (mdio->md_options & MD_RESERVE) { 1307 if (swap_pager_reserve(sc->object, 0, npage) < 0) { 1308 error = EDOM; 1309 goto finish; 1310 } 1311 } 1312 error = mdsetcred(sc, td->td_ucred); 1313 finish: 1314 if (error != 0) { 1315 vm_object_deallocate(sc->object); 1316 sc->object = NULL; 1317 } 1318 return (error); 1319} 1320 1321 1322static int 1323xmdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1324{ 1325 struct md_ioctl *mdio; 1326 struct md_s *sc; 1327 int error, i; 1328 unsigned sectsize; 1329 1330 if (md_debug) 1331 printf("mdctlioctl(%s %lx %p %x %p)\n", 1332 devtoname(dev), cmd, addr, flags, td); 1333 1334 mdio = (struct md_ioctl *)addr; 1335 if (mdio->md_version != MDIOVERSION) 1336 return (EINVAL); 1337 1338 /* 1339 * We assert the version number in the individual ioctl 1340 * handlers instead of out here because (a) it is possible we 1341 * may add another ioctl in the future which doesn't read an 1342 * mdio, and (b) the correct return value for an unknown ioctl 1343 * is ENOIOCTL, not EINVAL. 1344 */ 1345 error = 0; 1346 switch (cmd) { 1347 case MDIOCATTACH: 1348 switch (mdio->md_type) { 1349 case MD_MALLOC: 1350 case MD_PRELOAD: 1351 case MD_VNODE: 1352 case MD_SWAP: 1353 break; 1354 default: 1355 return (EINVAL); 1356 } 1357 if (mdio->md_sectorsize == 0) 1358 sectsize = DEV_BSIZE; 1359 else 1360 sectsize = mdio->md_sectorsize; 1361 if (sectsize > MAXPHYS || mdio->md_mediasize < sectsize) 1362 return (EINVAL); 1363 if (mdio->md_options & MD_AUTOUNIT) 1364 sc = mdnew(-1, &error, mdio->md_type); 1365 else { 1366 if (mdio->md_unit > INT_MAX) 1367 return (EINVAL); 1368 sc = mdnew(mdio->md_unit, &error, mdio->md_type); 1369 } 1370 if (sc == NULL) 1371 return (error); 1372 if (mdio->md_options & MD_AUTOUNIT) 1373 mdio->md_unit = sc->unit; 1374 sc->mediasize = mdio->md_mediasize; 1375 sc->sectorsize = sectsize; 1376 error = EDOOFUS; 1377 switch (sc->type) { 1378 case MD_MALLOC: 1379 sc->start = mdstart_malloc; 1380 error = mdcreate_malloc(sc, mdio); 1381 break; 1382 case MD_PRELOAD: 1383 /* 1384 * We disallow attaching preloaded memory disks via 1385 * ioctl. Preloaded memory disks are automatically 1386 * attached in g_md_init(). 1387 */ 1388 error = EOPNOTSUPP; 1389 break; 1390 case MD_VNODE: 1391 sc->start = mdstart_vnode; 1392 error = mdcreate_vnode(sc, mdio, td); 1393 break; 1394 case MD_SWAP: 1395 sc->start = mdstart_swap; 1396 error = mdcreate_swap(sc, mdio, td); 1397 break; 1398 } 1399 if (error != 0) { 1400 mddestroy(sc, td); 1401 return (error); 1402 } 1403 1404 /* Prune off any residual fractional sector */ 1405 i = sc->mediasize % sc->sectorsize; 1406 sc->mediasize -= i; 1407 1408 mdinit(sc); 1409 return (0); 1410 case MDIOCDETACH: 1411 if (mdio->md_mediasize != 0 || 1412 (mdio->md_options & ~MD_FORCE) != 0) 1413 return (EINVAL); 1414 1415 sc = mdfind(mdio->md_unit); 1416 if (sc == NULL) 1417 return (ENOENT); 1418 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) && 1419 !(mdio->md_options & MD_FORCE)) 1420 return (EBUSY); 1421 return (mddestroy(sc, td)); 1422 case MDIOCRESIZE: 1423 if ((mdio->md_options & ~(MD_FORCE | MD_RESERVE)) != 0) 1424 return (EINVAL); 1425 1426 sc = mdfind(mdio->md_unit); 1427 if (sc == NULL) 1428 return (ENOENT); 1429 if (mdio->md_mediasize < sc->sectorsize) 1430 return (EINVAL); 1431 if (mdio->md_mediasize < sc->mediasize && 1432 !(sc->flags & MD_FORCE) && 1433 !(mdio->md_options & MD_FORCE)) 1434 return (EBUSY); 1435 return (mdresize(sc, mdio)); 1436 case MDIOCQUERY: 1437 sc = mdfind(mdio->md_unit); 1438 if (sc == NULL) 1439 return (ENOENT); 1440 mdio->md_type = sc->type; 1441 mdio->md_options = sc->flags; 1442 mdio->md_mediasize = sc->mediasize; 1443 mdio->md_sectorsize = sc->sectorsize; 1444 if (sc->type == MD_VNODE) 1445 error = copyout(sc->file, mdio->md_file, 1446 strlen(sc->file) + 1); 1447 return (error); 1448 case MDIOCLIST: 1449 i = 1; 1450 LIST_FOREACH(sc, &md_softc_list, list) { 1451 if (i == MDNPAD - 1) 1452 mdio->md_pad[i] = -1; 1453 else 1454 mdio->md_pad[i++] = sc->unit; 1455 } 1456 mdio->md_pad[0] = i - 1; 1457 return (0); 1458 default: 1459 return (ENOIOCTL); 1460 }; 1461} 1462 1463static int 1464mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1465{ 1466 int error; 1467 1468 sx_xlock(&md_sx); 1469 error = xmdctlioctl(dev, cmd, addr, flags, td); 1470 sx_xunlock(&md_sx); 1471 return (error); 1472} 1473 1474static void 1475md_preloaded(u_char *image, size_t length, const char *name) 1476{ 1477 struct md_s *sc; 1478 int error; 1479 1480 sc = mdnew(-1, &error, MD_PRELOAD); 1481 if (sc == NULL) 1482 return; 1483 sc->mediasize = length; 1484 sc->sectorsize = DEV_BSIZE; 1485 sc->pl_ptr = image; 1486 sc->pl_len = length; 1487 sc->start = mdstart_preload; 1488#ifdef MD_ROOT 1489 if (sc->unit == 0) 1490 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0"; 1491#endif 1492 mdinit(sc); 1493 if (name != NULL) { 1494 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n", 1495 MD_NAME, sc->unit, name, length, image); 1496 } 1497} 1498 1499static void 1500g_md_init(struct g_class *mp __unused) 1501{ 1502 caddr_t mod; 1503 u_char *ptr, *name, *type; 1504 unsigned len; 1505 int i; 1506 1507 /* figure out log2(NINDIR) */ 1508 for (i = NINDIR, nshift = -1; i; nshift++) 1509 i >>= 1; 1510 1511 mod = NULL; 1512 sx_init(&md_sx, "MD config lock"); 1513 g_topology_unlock(); 1514 md_uh = new_unrhdr(0, INT_MAX, NULL); 1515#ifdef MD_ROOT_SIZE 1516 sx_xlock(&md_sx); 1517 md_preloaded(mfs_root.start, sizeof(mfs_root.start), NULL); 1518 sx_xunlock(&md_sx); 1519#endif 1520 /* XXX: are preload_* static or do they need Giant ? */ 1521 while ((mod = preload_search_next_name(mod)) != NULL) { 1522 name = (char *)preload_search_info(mod, MODINFO_NAME); 1523 if (name == NULL) 1524 continue; 1525 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1526 if (type == NULL) 1527 continue; 1528 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1529 continue; 1530 ptr = preload_fetch_addr(mod); 1531 len = preload_fetch_size(mod); 1532 if (ptr != NULL && len != 0) { 1533 sx_xlock(&md_sx); 1534 md_preloaded(ptr, len, name); 1535 sx_xunlock(&md_sx); 1536 } 1537 } 1538 md_vnode_pbuf_freecnt = nswbuf / 10; 1539 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL, 1540 0600, MDCTL_NAME); 1541 g_topology_lock(); 1542} 1543 1544static void 1545g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 1546 struct g_consumer *cp __unused, struct g_provider *pp) 1547{ 1548 struct md_s *mp; 1549 char *type; 1550 1551 mp = gp->softc; 1552 if (mp == NULL) 1553 return; 1554 1555 switch (mp->type) { 1556 case MD_MALLOC: 1557 type = "malloc"; 1558 break; 1559 case MD_PRELOAD: 1560 type = "preload"; 1561 break; 1562 case MD_VNODE: 1563 type = "vnode"; 1564 break; 1565 case MD_SWAP: 1566 type = "swap"; 1567 break; 1568 default: 1569 type = "unknown"; 1570 break; 1571 } 1572 1573 if (pp != NULL) { 1574 if (indent == NULL) { 1575 sbuf_printf(sb, " u %d", mp->unit); 1576 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize); 1577 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads); 1578 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors); 1579 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize); 1580 sbuf_printf(sb, " t %s", type); 1581 if (mp->type == MD_VNODE && mp->vnode != NULL) 1582 sbuf_printf(sb, " file %s", mp->file); 1583 } else { 1584 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, 1585 mp->unit); 1586 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n", 1587 indent, (uintmax_t) mp->sectorsize); 1588 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n", 1589 indent, (uintmax_t) mp->fwheads); 1590 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n", 1591 indent, (uintmax_t) mp->fwsectors); 1592 sbuf_printf(sb, "%s<length>%ju</length>\n", 1593 indent, (uintmax_t) mp->mediasize); 1594 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent, 1595 (mp->flags & MD_COMPRESS) == 0 ? "off": "on"); 1596 sbuf_printf(sb, "%s<access>%s</access>\n", indent, 1597 (mp->flags & MD_READONLY) == 0 ? "read-write": 1598 "read-only"); 1599 sbuf_printf(sb, "%s<type>%s</type>\n", indent, 1600 type); 1601 if (mp->type == MD_VNODE && mp->vnode != NULL) 1602 sbuf_printf(sb, "%s<file>%s</file>\n", 1603 indent, mp->file); 1604 } 1605 } 1606} 1607 1608static void 1609g_md_fini(struct g_class *mp __unused) 1610{ 1611 1612 sx_destroy(&md_sx); 1613 if (status_dev != NULL) 1614 destroy_dev(status_dev); 1615 delete_unrhdr(md_uh); 1616} 1617