md.c revision 286727
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 286727 2015-08-13 15:16:34Z marcel $ 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 * Copyright (c) 2013 The FreeBSD Foundation 22 * All rights reserved. 23 * 24 * This code is derived from software contributed to Berkeley by 25 * the Systems Programming Group of the University of Utah Computer 26 * Science Department. 27 * 28 * Portions of this software were developed by Konstantin Belousov 29 * under sponsorship from the FreeBSD Foundation. 30 * 31 * Redistribution and use in source and binary forms, with or without 32 * modification, are permitted provided that the following conditions 33 * are met: 34 * 1. Redistributions of source code must retain the above copyright 35 * notice, this list of conditions and the following disclaimer. 36 * 2. Redistributions in binary form must reproduce the above copyright 37 * notice, this list of conditions and the following disclaimer in the 38 * documentation and/or other materials provided with the distribution. 39 * 4. Neither the name of the University nor the names of its contributors 40 * may be used to endorse or promote products derived from this software 41 * without specific prior written permission. 42 * 43 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 44 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 45 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 46 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 47 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 48 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 49 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 50 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 51 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 52 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 53 * SUCH DAMAGE. 54 * 55 * from: Utah Hdr: vn.c 1.13 94/04/02 56 * 57 * from: @(#)vn.c 8.6 (Berkeley) 4/1/94 58 * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03 59 */ 60 61#include "opt_geom.h" 62#include "opt_md.h" 63 64#include <sys/param.h> 65#include <sys/systm.h> 66#include <sys/bio.h> 67#include <sys/buf.h> 68#include <sys/conf.h> 69#include <sys/devicestat.h> 70#include <sys/fcntl.h> 71#include <sys/kernel.h> 72#include <sys/kthread.h> 73#include <sys/limits.h> 74#include <sys/linker.h> 75#include <sys/lock.h> 76#include <sys/malloc.h> 77#include <sys/mdioctl.h> 78#include <sys/mount.h> 79#include <sys/mutex.h> 80#include <sys/sx.h> 81#include <sys/namei.h> 82#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#include <geom/geom_int.h> 93 94#include <vm/vm.h> 95#include <vm/vm_param.h> 96#include <vm/vm_object.h> 97#include <vm/vm_page.h> 98#include <vm/vm_pager.h> 99#include <vm/swap_pager.h> 100#include <vm/uma.h> 101 102#define MD_MODVER 1 103 104#define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */ 105#define MD_EXITING 0x20000 /* Worker thread is exiting. */ 106 107#ifndef MD_NSECT 108#define MD_NSECT (10000 * 2) 109#endif 110 111static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk"); 112static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors"); 113 114static int md_debug; 115SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, 116 "Enable md(4) debug messages"); 117static int md_malloc_wait; 118SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0, 119 "Allow malloc to wait for memory allocations"); 120 121#if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE) 122#define MD_ROOT_FSTYPE "ufs" 123#endif 124 125#if defined(MD_ROOT) 126/* 127 * Preloaded image gets put here. 128 */ 129#if defined(MD_ROOT_SIZE) 130/* 131 * Applications that patch the object with the image can determine 132 * the size looking at the start and end markers (strings), 133 * so we want them contiguous. 134 */ 135static struct { 136 u_char start[MD_ROOT_SIZE*1024]; 137 u_char end[128]; 138} mfs_root = { 139 .start = "MFS Filesystem goes here", 140 .end = "MFS Filesystem had better STOP here", 141}; 142const int mfs_root_size = sizeof(mfs_root.start); 143#else 144extern volatile u_char __weak_symbol mfs_root; 145extern volatile u_char __weak_symbol mfs_root_end; 146__GLOBL(mfs_root); 147__GLOBL(mfs_root_end); 148#define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root)) 149#endif 150#endif 151 152static g_init_t g_md_init; 153static g_fini_t g_md_fini; 154static g_start_t g_md_start; 155static g_access_t g_md_access; 156static void g_md_dumpconf(struct sbuf *sb, const char *indent, 157 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp); 158 159static struct cdev *status_dev = 0; 160static struct sx md_sx; 161static struct unrhdr *md_uh; 162 163static d_ioctl_t mdctlioctl; 164 165static struct cdevsw mdctl_cdevsw = { 166 .d_version = D_VERSION, 167 .d_ioctl = mdctlioctl, 168 .d_name = MD_NAME, 169}; 170 171struct g_class g_md_class = { 172 .name = "MD", 173 .version = G_VERSION, 174 .init = g_md_init, 175 .fini = g_md_fini, 176 .start = g_md_start, 177 .access = g_md_access, 178 .dumpconf = g_md_dumpconf, 179}; 180 181DECLARE_GEOM_CLASS(g_md_class, g_md); 182 183 184static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list); 185 186#define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 187#define NMASK (NINDIR-1) 188static int nshift; 189 190static int md_vnode_pbuf_freecnt; 191 192struct indir { 193 uintptr_t *array; 194 u_int total; 195 u_int used; 196 u_int shift; 197}; 198 199struct md_s { 200 int unit; 201 LIST_ENTRY(md_s) list; 202 struct bio_queue_head bio_queue; 203 struct mtx queue_mtx; 204 struct mtx stat_mtx; 205 struct cdev *dev; 206 enum md_types type; 207 off_t mediasize; 208 unsigned sectorsize; 209 unsigned opencount; 210 unsigned fwheads; 211 unsigned fwsectors; 212 unsigned flags; 213 char name[20]; 214 struct proc *procp; 215 struct g_geom *gp; 216 struct g_provider *pp; 217 int (*start)(struct md_s *sc, struct bio *bp); 218 struct devstat *devstat; 219 220 /* MD_MALLOC related fields */ 221 struct indir *indir; 222 uma_zone_t uma; 223 224 /* MD_PRELOAD related fields */ 225 u_char *pl_ptr; 226 size_t pl_len; 227 228 /* MD_VNODE related fields */ 229 struct vnode *vnode; 230 char file[PATH_MAX]; 231 struct ucred *cred; 232 233 /* MD_SWAP related fields */ 234 vm_object_t object; 235}; 236 237static struct indir * 238new_indir(u_int shift) 239{ 240 struct indir *ip; 241 242 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT) 243 | M_ZERO); 244 if (ip == NULL) 245 return (NULL); 246 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 247 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO); 248 if (ip->array == NULL) { 249 free(ip, M_MD); 250 return (NULL); 251 } 252 ip->total = NINDIR; 253 ip->shift = shift; 254 return (ip); 255} 256 257static void 258del_indir(struct indir *ip) 259{ 260 261 free(ip->array, M_MDSECT); 262 free(ip, M_MD); 263} 264 265static void 266destroy_indir(struct md_s *sc, struct indir *ip) 267{ 268 int i; 269 270 for (i = 0; i < NINDIR; i++) { 271 if (!ip->array[i]) 272 continue; 273 if (ip->shift) 274 destroy_indir(sc, (struct indir*)(ip->array[i])); 275 else if (ip->array[i] > 255) 276 uma_zfree(sc->uma, (void *)(ip->array[i])); 277 } 278 del_indir(ip); 279} 280 281/* 282 * This function does the math and allocates the top level "indir" structure 283 * for a device of "size" sectors. 284 */ 285 286static struct indir * 287dimension(off_t size) 288{ 289 off_t rcnt; 290 struct indir *ip; 291 int layer; 292 293 rcnt = size; 294 layer = 0; 295 while (rcnt > NINDIR) { 296 rcnt /= NINDIR; 297 layer++; 298 } 299 300 /* 301 * XXX: the top layer is probably not fully populated, so we allocate 302 * too much space for ip->array in here. 303 */ 304 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 305 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 306 M_MDSECT, M_WAITOK | M_ZERO); 307 ip->total = NINDIR; 308 ip->shift = layer * nshift; 309 return (ip); 310} 311 312/* 313 * Read a given sector 314 */ 315 316static uintptr_t 317s_read(struct indir *ip, off_t offset) 318{ 319 struct indir *cip; 320 int idx; 321 uintptr_t up; 322 323 if (md_debug > 1) 324 printf("s_read(%jd)\n", (intmax_t)offset); 325 up = 0; 326 for (cip = ip; cip != NULL;) { 327 if (cip->shift) { 328 idx = (offset >> cip->shift) & NMASK; 329 up = cip->array[idx]; 330 cip = (struct indir *)up; 331 continue; 332 } 333 idx = offset & NMASK; 334 return (cip->array[idx]); 335 } 336 return (0); 337} 338 339/* 340 * Write a given sector, prune the tree if the value is 0 341 */ 342 343static int 344s_write(struct indir *ip, off_t offset, uintptr_t ptr) 345{ 346 struct indir *cip, *lip[10]; 347 int idx, li; 348 uintptr_t up; 349 350 if (md_debug > 1) 351 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 352 up = 0; 353 li = 0; 354 cip = ip; 355 for (;;) { 356 lip[li++] = cip; 357 if (cip->shift) { 358 idx = (offset >> cip->shift) & NMASK; 359 up = cip->array[idx]; 360 if (up != 0) { 361 cip = (struct indir *)up; 362 continue; 363 } 364 /* Allocate branch */ 365 cip->array[idx] = 366 (uintptr_t)new_indir(cip->shift - nshift); 367 if (cip->array[idx] == 0) 368 return (ENOSPC); 369 cip->used++; 370 up = cip->array[idx]; 371 cip = (struct indir *)up; 372 continue; 373 } 374 /* leafnode */ 375 idx = offset & NMASK; 376 up = cip->array[idx]; 377 if (up != 0) 378 cip->used--; 379 cip->array[idx] = ptr; 380 if (ptr != 0) 381 cip->used++; 382 break; 383 } 384 if (cip->used != 0 || li == 1) 385 return (0); 386 li--; 387 while (cip->used == 0 && cip != ip) { 388 li--; 389 idx = (offset >> lip[li]->shift) & NMASK; 390 up = lip[li]->array[idx]; 391 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 392 del_indir(cip); 393 lip[li]->array[idx] = 0; 394 lip[li]->used--; 395 cip = lip[li]; 396 } 397 return (0); 398} 399 400 401static int 402g_md_access(struct g_provider *pp, int r, int w, int e) 403{ 404 struct md_s *sc; 405 406 sc = pp->geom->softc; 407 if (sc == NULL) { 408 if (r <= 0 && w <= 0 && e <= 0) 409 return (0); 410 return (ENXIO); 411 } 412 r += pp->acr; 413 w += pp->acw; 414 e += pp->ace; 415 if ((sc->flags & MD_READONLY) != 0 && w > 0) 416 return (EROFS); 417 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 418 sc->opencount = 1; 419 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 420 sc->opencount = 0; 421 } 422 return (0); 423} 424 425static void 426g_md_start(struct bio *bp) 427{ 428 struct md_s *sc; 429 430 sc = bp->bio_to->geom->softc; 431 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) { 432 mtx_lock(&sc->stat_mtx); 433 devstat_start_transaction_bio(sc->devstat, bp); 434 mtx_unlock(&sc->stat_mtx); 435 } 436 mtx_lock(&sc->queue_mtx); 437 bioq_disksort(&sc->bio_queue, bp); 438 mtx_unlock(&sc->queue_mtx); 439 wakeup(sc); 440} 441 442#define MD_MALLOC_MOVE_ZERO 1 443#define MD_MALLOC_MOVE_FILL 2 444#define MD_MALLOC_MOVE_READ 3 445#define MD_MALLOC_MOVE_WRITE 4 446#define MD_MALLOC_MOVE_CMP 5 447 448static int 449md_malloc_move(vm_page_t **mp, int *ma_offs, unsigned sectorsize, 450 void *ptr, u_char fill, int op) 451{ 452 struct sf_buf *sf; 453 vm_page_t m, *mp1; 454 char *p, first; 455 off_t *uc; 456 unsigned n; 457 int error, i, ma_offs1, sz, first_read; 458 459 m = NULL; 460 error = 0; 461 sf = NULL; 462 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */ 463 first = 0; 464 first_read = 0; 465 uc = ptr; 466 mp1 = *mp; 467 ma_offs1 = *ma_offs; 468 /* } */ 469 sched_pin(); 470 for (n = sectorsize; n != 0; n -= sz) { 471 sz = imin(PAGE_SIZE - *ma_offs, n); 472 if (m != **mp) { 473 if (sf != NULL) 474 sf_buf_free(sf); 475 m = **mp; 476 sf = sf_buf_alloc(m, SFB_CPUPRIVATE | 477 (md_malloc_wait ? 0 : SFB_NOWAIT)); 478 if (sf == NULL) { 479 error = ENOMEM; 480 break; 481 } 482 } 483 p = (char *)sf_buf_kva(sf) + *ma_offs; 484 switch (op) { 485 case MD_MALLOC_MOVE_ZERO: 486 bzero(p, sz); 487 break; 488 case MD_MALLOC_MOVE_FILL: 489 memset(p, fill, sz); 490 break; 491 case MD_MALLOC_MOVE_READ: 492 bcopy(ptr, p, sz); 493 cpu_flush_dcache(p, sz); 494 break; 495 case MD_MALLOC_MOVE_WRITE: 496 bcopy(p, ptr, sz); 497 break; 498 case MD_MALLOC_MOVE_CMP: 499 for (i = 0; i < sz; i++, p++) { 500 if (!first_read) { 501 *uc = (u_char)*p; 502 first = *p; 503 first_read = 1; 504 } else if (*p != first) { 505 error = EDOOFUS; 506 break; 507 } 508 } 509 break; 510 default: 511 KASSERT(0, ("md_malloc_move unknown op %d\n", op)); 512 break; 513 } 514 if (error != 0) 515 break; 516 *ma_offs += sz; 517 *ma_offs %= PAGE_SIZE; 518 if (*ma_offs == 0) 519 (*mp)++; 520 ptr = (char *)ptr + sz; 521 } 522 523 if (sf != NULL) 524 sf_buf_free(sf); 525 sched_unpin(); 526 if (op == MD_MALLOC_MOVE_CMP && error != 0) { 527 *mp = mp1; 528 *ma_offs = ma_offs1; 529 } 530 return (error); 531} 532 533static int 534mdstart_malloc(struct md_s *sc, struct bio *bp) 535{ 536 u_char *dst; 537 vm_page_t *m; 538 int i, error, error1, ma_offs, notmapped; 539 off_t secno, nsec, uc; 540 uintptr_t sp, osp; 541 542 switch (bp->bio_cmd) { 543 case BIO_READ: 544 case BIO_WRITE: 545 case BIO_DELETE: 546 break; 547 default: 548 return (EOPNOTSUPP); 549 } 550 551 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0; 552 if (notmapped) { 553 m = bp->bio_ma; 554 ma_offs = bp->bio_ma_offset; 555 dst = NULL; 556 } else { 557 dst = bp->bio_data; 558 } 559 560 nsec = bp->bio_length / sc->sectorsize; 561 secno = bp->bio_offset / sc->sectorsize; 562 error = 0; 563 while (nsec--) { 564 osp = s_read(sc->indir, secno); 565 if (bp->bio_cmd == BIO_DELETE) { 566 if (osp != 0) 567 error = s_write(sc->indir, secno, 0); 568 } else if (bp->bio_cmd == BIO_READ) { 569 if (osp == 0) { 570 if (notmapped) { 571 error = md_malloc_move(&m, &ma_offs, 572 sc->sectorsize, NULL, 0, 573 MD_MALLOC_MOVE_ZERO); 574 } else 575 bzero(dst, sc->sectorsize); 576 } else if (osp <= 255) { 577 if (notmapped) { 578 error = md_malloc_move(&m, &ma_offs, 579 sc->sectorsize, NULL, osp, 580 MD_MALLOC_MOVE_FILL); 581 } else 582 memset(dst, osp, sc->sectorsize); 583 } else { 584 if (notmapped) { 585 error = md_malloc_move(&m, &ma_offs, 586 sc->sectorsize, (void *)osp, 0, 587 MD_MALLOC_MOVE_READ); 588 } else { 589 bcopy((void *)osp, dst, sc->sectorsize); 590 cpu_flush_dcache(dst, sc->sectorsize); 591 } 592 } 593 osp = 0; 594 } else if (bp->bio_cmd == BIO_WRITE) { 595 if (sc->flags & MD_COMPRESS) { 596 if (notmapped) { 597 error1 = md_malloc_move(&m, &ma_offs, 598 sc->sectorsize, &uc, 0, 599 MD_MALLOC_MOVE_CMP); 600 i = error1 == 0 ? sc->sectorsize : 0; 601 } else { 602 uc = dst[0]; 603 for (i = 1; i < sc->sectorsize; i++) { 604 if (dst[i] != uc) 605 break; 606 } 607 } 608 } else { 609 i = 0; 610 uc = 0; 611 } 612 if (i == sc->sectorsize) { 613 if (osp != uc) 614 error = s_write(sc->indir, secno, uc); 615 } else { 616 if (osp <= 255) { 617 sp = (uintptr_t)uma_zalloc(sc->uma, 618 md_malloc_wait ? M_WAITOK : 619 M_NOWAIT); 620 if (sp == 0) { 621 error = ENOSPC; 622 break; 623 } 624 if (notmapped) { 625 error = md_malloc_move(&m, 626 &ma_offs, sc->sectorsize, 627 (void *)sp, 0, 628 MD_MALLOC_MOVE_WRITE); 629 } else { 630 bcopy(dst, (void *)sp, 631 sc->sectorsize); 632 } 633 error = s_write(sc->indir, secno, sp); 634 } else { 635 if (notmapped) { 636 error = md_malloc_move(&m, 637 &ma_offs, sc->sectorsize, 638 (void *)osp, 0, 639 MD_MALLOC_MOVE_WRITE); 640 } else { 641 bcopy(dst, (void *)osp, 642 sc->sectorsize); 643 } 644 osp = 0; 645 } 646 } 647 } else { 648 error = EOPNOTSUPP; 649 } 650 if (osp > 255) 651 uma_zfree(sc->uma, (void*)osp); 652 if (error != 0) 653 break; 654 secno++; 655 if (!notmapped) 656 dst += sc->sectorsize; 657 } 658 bp->bio_resid = 0; 659 return (error); 660} 661 662static int 663mdstart_preload(struct md_s *sc, struct bio *bp) 664{ 665 666 switch (bp->bio_cmd) { 667 case BIO_READ: 668 bcopy(sc->pl_ptr + bp->bio_offset, bp->bio_data, 669 bp->bio_length); 670 cpu_flush_dcache(bp->bio_data, bp->bio_length); 671 break; 672 case BIO_WRITE: 673 bcopy(bp->bio_data, sc->pl_ptr + bp->bio_offset, 674 bp->bio_length); 675 break; 676 } 677 bp->bio_resid = 0; 678 return (0); 679} 680 681static int 682mdstart_vnode(struct md_s *sc, struct bio *bp) 683{ 684 int error; 685 struct uio auio; 686 struct iovec aiov; 687 struct mount *mp; 688 struct vnode *vp; 689 struct buf *pb; 690 struct thread *td; 691 off_t end, zerosize; 692 693 switch (bp->bio_cmd) { 694 case BIO_READ: 695 case BIO_WRITE: 696 case BIO_DELETE: 697 case BIO_FLUSH: 698 break; 699 default: 700 return (EOPNOTSUPP); 701 } 702 703 td = curthread; 704 vp = sc->vnode; 705 706 /* 707 * VNODE I/O 708 * 709 * If an error occurs, we set BIO_ERROR but we do not set 710 * B_INVAL because (for a write anyway), the buffer is 711 * still valid. 712 */ 713 714 if (bp->bio_cmd == BIO_FLUSH) { 715 (void) vn_start_write(vp, &mp, V_WAIT); 716 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 717 error = VOP_FSYNC(vp, MNT_WAIT, td); 718 VOP_UNLOCK(vp, 0); 719 vn_finished_write(mp); 720 return (error); 721 } 722 723 bzero(&auio, sizeof(auio)); 724 725 /* 726 * Special case for BIO_DELETE. On the surface, this is very 727 * similar to BIO_WRITE, except that we write from our own 728 * fixed-length buffer, so we have to loop. The net result is 729 * that the two cases end up having very little in common. 730 */ 731 if (bp->bio_cmd == BIO_DELETE) { 732 zerosize = ZERO_REGION_SIZE - 733 (ZERO_REGION_SIZE % sc->sectorsize); 734 auio.uio_iov = &aiov; 735 auio.uio_iovcnt = 1; 736 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 737 auio.uio_segflg = UIO_SYSSPACE; 738 auio.uio_rw = UIO_WRITE; 739 auio.uio_td = td; 740 end = bp->bio_offset + bp->bio_length; 741 (void) vn_start_write(vp, &mp, V_WAIT); 742 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 743 error = 0; 744 while (auio.uio_offset < end) { 745 aiov.iov_base = __DECONST(void *, zero_region); 746 aiov.iov_len = end - auio.uio_offset; 747 if (aiov.iov_len > zerosize) 748 aiov.iov_len = zerosize; 749 auio.uio_resid = aiov.iov_len; 750 error = VOP_WRITE(vp, &auio, 751 sc->flags & MD_ASYNC ? 0 : IO_SYNC, sc->cred); 752 if (error != 0) 753 break; 754 } 755 VOP_UNLOCK(vp, 0); 756 vn_finished_write(mp); 757 bp->bio_resid = end - auio.uio_offset; 758 return (error); 759 } 760 761 if ((bp->bio_flags & BIO_UNMAPPED) == 0) { 762 pb = NULL; 763 aiov.iov_base = bp->bio_data; 764 } else { 765 KASSERT(bp->bio_length <= MAXPHYS, ("bio_length %jd", 766 (uintmax_t)bp->bio_length)); 767 pb = getpbuf(&md_vnode_pbuf_freecnt); 768 pmap_qenter((vm_offset_t)pb->b_data, bp->bio_ma, bp->bio_ma_n); 769 aiov.iov_base = (void *)((vm_offset_t)pb->b_data + 770 bp->bio_ma_offset); 771 } 772 aiov.iov_len = bp->bio_length; 773 auio.uio_iov = &aiov; 774 auio.uio_iovcnt = 1; 775 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 776 auio.uio_segflg = UIO_SYSSPACE; 777 if (bp->bio_cmd == BIO_READ) 778 auio.uio_rw = UIO_READ; 779 else if (bp->bio_cmd == BIO_WRITE) 780 auio.uio_rw = UIO_WRITE; 781 else 782 panic("wrong BIO_OP in mdstart_vnode"); 783 auio.uio_resid = bp->bio_length; 784 auio.uio_td = td; 785 /* 786 * When reading set IO_DIRECT to try to avoid double-caching 787 * the data. When writing IO_DIRECT is not optimal. 788 */ 789 if (bp->bio_cmd == BIO_READ) { 790 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 791 error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred); 792 VOP_UNLOCK(vp, 0); 793 } else { 794 (void) vn_start_write(vp, &mp, V_WAIT); 795 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 796 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC, 797 sc->cred); 798 VOP_UNLOCK(vp, 0); 799 vn_finished_write(mp); 800 } 801 if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 802 pmap_qremove((vm_offset_t)pb->b_data, bp->bio_ma_n); 803 relpbuf(pb, &md_vnode_pbuf_freecnt); 804 } 805 bp->bio_resid = auio.uio_resid; 806 return (error); 807} 808 809static int 810mdstart_swap(struct md_s *sc, struct bio *bp) 811{ 812 vm_page_t m; 813 u_char *p; 814 vm_pindex_t i, lastp; 815 int rv, ma_offs, offs, len, lastend; 816 817 switch (bp->bio_cmd) { 818 case BIO_READ: 819 case BIO_WRITE: 820 case BIO_DELETE: 821 break; 822 default: 823 return (EOPNOTSUPP); 824 } 825 826 p = bp->bio_data; 827 ma_offs = (bp->bio_flags & BIO_UNMAPPED) == 0 ? 0 : bp->bio_ma_offset; 828 829 /* 830 * offs is the offset at which to start operating on the 831 * next (ie, first) page. lastp is the last page on 832 * which we're going to operate. lastend is the ending 833 * position within that last page (ie, PAGE_SIZE if 834 * we're operating on complete aligned pages). 835 */ 836 offs = bp->bio_offset % PAGE_SIZE; 837 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE; 838 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1; 839 840 rv = VM_PAGER_OK; 841 VM_OBJECT_WLOCK(sc->object); 842 vm_object_pip_add(sc->object, 1); 843 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) { 844 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs; 845 m = vm_page_grab(sc->object, i, VM_ALLOC_SYSTEM); 846 if (bp->bio_cmd == BIO_READ) { 847 if (m->valid == VM_PAGE_BITS_ALL) 848 rv = VM_PAGER_OK; 849 else 850 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 851 if (rv == VM_PAGER_ERROR) { 852 vm_page_xunbusy(m); 853 break; 854 } else if (rv == VM_PAGER_FAIL) { 855 /* 856 * Pager does not have the page. Zero 857 * the allocated page, and mark it as 858 * valid. Do not set dirty, the page 859 * can be recreated if thrown out. 860 */ 861 pmap_zero_page(m); 862 m->valid = VM_PAGE_BITS_ALL; 863 } 864 if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 865 pmap_copy_pages(&m, offs, bp->bio_ma, 866 ma_offs, len); 867 } else { 868 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len); 869 cpu_flush_dcache(p, len); 870 } 871 } else if (bp->bio_cmd == BIO_WRITE) { 872 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 873 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 874 else 875 rv = VM_PAGER_OK; 876 if (rv == VM_PAGER_ERROR) { 877 vm_page_xunbusy(m); 878 break; 879 } 880 if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 881 pmap_copy_pages(bp->bio_ma, ma_offs, &m, 882 offs, len); 883 } else { 884 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len); 885 } 886 m->valid = VM_PAGE_BITS_ALL; 887 } else if (bp->bio_cmd == BIO_DELETE) { 888 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 889 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 890 else 891 rv = VM_PAGER_OK; 892 if (rv == VM_PAGER_ERROR) { 893 vm_page_xunbusy(m); 894 break; 895 } 896 if (len != PAGE_SIZE) { 897 pmap_zero_page_area(m, offs, len); 898 vm_page_clear_dirty(m, offs, len); 899 m->valid = VM_PAGE_BITS_ALL; 900 } else 901 vm_pager_page_unswapped(m); 902 } 903 vm_page_xunbusy(m); 904 vm_page_lock(m); 905 if (bp->bio_cmd == BIO_DELETE && len == PAGE_SIZE) 906 vm_page_free(m); 907 else 908 vm_page_activate(m); 909 vm_page_unlock(m); 910 if (bp->bio_cmd == BIO_WRITE) { 911 vm_page_dirty(m); 912 vm_pager_page_unswapped(m); 913 } 914 915 /* Actions on further pages start at offset 0 */ 916 p += PAGE_SIZE - offs; 917 offs = 0; 918 ma_offs += len; 919 } 920 vm_object_pip_wakeup(sc->object); 921 VM_OBJECT_WUNLOCK(sc->object); 922 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC); 923} 924 925static int 926mdstart_null(struct md_s *sc, struct bio *bp) 927{ 928 929 switch (bp->bio_cmd) { 930 case BIO_READ: 931 bzero(bp->bio_data, bp->bio_length); 932 cpu_flush_dcache(bp->bio_data, bp->bio_length); 933 break; 934 case BIO_WRITE: 935 break; 936 } 937 bp->bio_resid = 0; 938 return (0); 939} 940 941static void 942md_kthread(void *arg) 943{ 944 struct md_s *sc; 945 struct bio *bp; 946 int error; 947 948 sc = arg; 949 thread_lock(curthread); 950 sched_prio(curthread, PRIBIO); 951 thread_unlock(curthread); 952 if (sc->type == MD_VNODE) 953 curthread->td_pflags |= TDP_NORUNNINGBUF; 954 955 for (;;) { 956 mtx_lock(&sc->queue_mtx); 957 if (sc->flags & MD_SHUTDOWN) { 958 sc->flags |= MD_EXITING; 959 mtx_unlock(&sc->queue_mtx); 960 kproc_exit(0); 961 } 962 bp = bioq_takefirst(&sc->bio_queue); 963 if (!bp) { 964 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 965 continue; 966 } 967 mtx_unlock(&sc->queue_mtx); 968 if (bp->bio_cmd == BIO_GETATTR) { 969 if ((sc->fwsectors && sc->fwheads && 970 (g_handleattr_int(bp, "GEOM::fwsectors", 971 sc->fwsectors) || 972 g_handleattr_int(bp, "GEOM::fwheads", 973 sc->fwheads))) || 974 g_handleattr_int(bp, "GEOM::candelete", 1)) 975 error = -1; 976 else 977 error = EOPNOTSUPP; 978 } else { 979 error = sc->start(sc, bp); 980 } 981 982 if (error != -1) { 983 bp->bio_completed = bp->bio_length; 984 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) 985 devstat_end_transaction_bio(sc->devstat, bp); 986 g_io_deliver(bp, error); 987 } 988 } 989} 990 991static struct md_s * 992mdfind(int unit) 993{ 994 struct md_s *sc; 995 996 LIST_FOREACH(sc, &md_softc_list, list) { 997 if (sc->unit == unit) 998 break; 999 } 1000 return (sc); 1001} 1002 1003static struct md_s * 1004mdnew(int unit, int *errp, enum md_types type) 1005{ 1006 struct md_s *sc; 1007 int error; 1008 1009 *errp = 0; 1010 if (unit == -1) 1011 unit = alloc_unr(md_uh); 1012 else 1013 unit = alloc_unr_specific(md_uh, unit); 1014 1015 if (unit == -1) { 1016 *errp = EBUSY; 1017 return (NULL); 1018 } 1019 1020 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 1021 sc->type = type; 1022 bioq_init(&sc->bio_queue); 1023 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 1024 mtx_init(&sc->stat_mtx, "md stat", NULL, MTX_DEF); 1025 sc->unit = unit; 1026 sprintf(sc->name, "md%d", unit); 1027 LIST_INSERT_HEAD(&md_softc_list, sc, list); 1028 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 1029 if (error == 0) 1030 return (sc); 1031 LIST_REMOVE(sc, list); 1032 mtx_destroy(&sc->stat_mtx); 1033 mtx_destroy(&sc->queue_mtx); 1034 free_unr(md_uh, sc->unit); 1035 free(sc, M_MD); 1036 *errp = error; 1037 return (NULL); 1038} 1039 1040static void 1041mdinit(struct md_s *sc) 1042{ 1043 struct g_geom *gp; 1044 struct g_provider *pp; 1045 1046 g_topology_lock(); 1047 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 1048 gp->softc = sc; 1049 pp = g_new_providerf(gp, "md%d", sc->unit); 1050 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE; 1051 pp->mediasize = sc->mediasize; 1052 pp->sectorsize = sc->sectorsize; 1053 switch (sc->type) { 1054 case MD_MALLOC: 1055 case MD_VNODE: 1056 case MD_SWAP: 1057 pp->flags |= G_PF_ACCEPT_UNMAPPED; 1058 break; 1059 case MD_PRELOAD: 1060 case MD_NULL: 1061 break; 1062 } 1063 sc->gp = gp; 1064 sc->pp = pp; 1065 g_error_provider(pp, 0); 1066 g_topology_unlock(); 1067 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize, 1068 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX); 1069} 1070 1071static int 1072mdcreate_malloc(struct md_s *sc, struct md_ioctl *mdio) 1073{ 1074 uintptr_t sp; 1075 int error; 1076 off_t u; 1077 1078 error = 0; 1079 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 1080 return (EINVAL); 1081 if (mdio->md_sectorsize != 0 && !powerof2(mdio->md_sectorsize)) 1082 return (EINVAL); 1083 /* Compression doesn't make sense if we have reserved space */ 1084 if (mdio->md_options & MD_RESERVE) 1085 mdio->md_options &= ~MD_COMPRESS; 1086 if (mdio->md_fwsectors != 0) 1087 sc->fwsectors = mdio->md_fwsectors; 1088 if (mdio->md_fwheads != 0) 1089 sc->fwheads = mdio->md_fwheads; 1090 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 1091 sc->indir = dimension(sc->mediasize / sc->sectorsize); 1092 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL, 1093 0x1ff, 0); 1094 if (mdio->md_options & MD_RESERVE) { 1095 off_t nsectors; 1096 1097 nsectors = sc->mediasize / sc->sectorsize; 1098 for (u = 0; u < nsectors; u++) { 1099 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ? 1100 M_WAITOK : M_NOWAIT) | M_ZERO); 1101 if (sp != 0) 1102 error = s_write(sc->indir, u, sp); 1103 else 1104 error = ENOMEM; 1105 if (error != 0) 1106 break; 1107 } 1108 } 1109 return (error); 1110} 1111 1112 1113static int 1114mdsetcred(struct md_s *sc, struct ucred *cred) 1115{ 1116 char *tmpbuf; 1117 int error = 0; 1118 1119 /* 1120 * Set credits in our softc 1121 */ 1122 1123 if (sc->cred) 1124 crfree(sc->cred); 1125 sc->cred = crhold(cred); 1126 1127 /* 1128 * Horrible kludge to establish credentials for NFS XXX. 1129 */ 1130 1131 if (sc->vnode) { 1132 struct uio auio; 1133 struct iovec aiov; 1134 1135 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK); 1136 bzero(&auio, sizeof(auio)); 1137 1138 aiov.iov_base = tmpbuf; 1139 aiov.iov_len = sc->sectorsize; 1140 auio.uio_iov = &aiov; 1141 auio.uio_iovcnt = 1; 1142 auio.uio_offset = 0; 1143 auio.uio_rw = UIO_READ; 1144 auio.uio_segflg = UIO_SYSSPACE; 1145 auio.uio_resid = aiov.iov_len; 1146 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 1147 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 1148 VOP_UNLOCK(sc->vnode, 0); 1149 free(tmpbuf, M_TEMP); 1150 } 1151 return (error); 1152} 1153 1154static int 1155mdcreate_vnode(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 1156{ 1157 struct vattr vattr; 1158 struct nameidata nd; 1159 char *fname; 1160 int error, flags; 1161 1162 /* 1163 * Kernel-originated requests must have the filename appended 1164 * to the mdio structure to protect against malicious software. 1165 */ 1166 fname = mdio->md_file; 1167 if ((void *)fname != (void *)(mdio + 1)) { 1168 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL); 1169 if (error != 0) 1170 return (error); 1171 } else 1172 strlcpy(sc->file, fname, sizeof(sc->file)); 1173 1174 /* 1175 * If the user specified that this is a read only device, don't 1176 * set the FWRITE mask before trying to open the backing store. 1177 */ 1178 flags = FREAD | ((mdio->md_options & MD_READONLY) ? 0 : FWRITE); 1179 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td); 1180 error = vn_open(&nd, &flags, 0, NULL); 1181 if (error != 0) 1182 return (error); 1183 NDFREE(&nd, NDF_ONLY_PNBUF); 1184 if (nd.ni_vp->v_type != VREG) { 1185 error = EINVAL; 1186 goto bad; 1187 } 1188 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred); 1189 if (error != 0) 1190 goto bad; 1191 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) { 1192 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY); 1193 if (nd.ni_vp->v_iflag & VI_DOOMED) { 1194 /* Forced unmount. */ 1195 error = EBADF; 1196 goto bad; 1197 } 1198 } 1199 nd.ni_vp->v_vflag |= VV_MD; 1200 VOP_UNLOCK(nd.ni_vp, 0); 1201 1202 if (mdio->md_fwsectors != 0) 1203 sc->fwsectors = mdio->md_fwsectors; 1204 if (mdio->md_fwheads != 0) 1205 sc->fwheads = mdio->md_fwheads; 1206 sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC); 1207 if (!(flags & FWRITE)) 1208 sc->flags |= MD_READONLY; 1209 sc->vnode = nd.ni_vp; 1210 1211 error = mdsetcred(sc, td->td_ucred); 1212 if (error != 0) { 1213 sc->vnode = NULL; 1214 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY); 1215 nd.ni_vp->v_vflag &= ~VV_MD; 1216 goto bad; 1217 } 1218 return (0); 1219bad: 1220 VOP_UNLOCK(nd.ni_vp, 0); 1221 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td); 1222 return (error); 1223} 1224 1225static int 1226mddestroy(struct md_s *sc, struct thread *td) 1227{ 1228 1229 if (sc->gp) { 1230 sc->gp->softc = NULL; 1231 g_topology_lock(); 1232 g_wither_geom(sc->gp, ENXIO); 1233 g_topology_unlock(); 1234 sc->gp = NULL; 1235 sc->pp = NULL; 1236 } 1237 if (sc->devstat) { 1238 devstat_remove_entry(sc->devstat); 1239 sc->devstat = NULL; 1240 } 1241 mtx_lock(&sc->queue_mtx); 1242 sc->flags |= MD_SHUTDOWN; 1243 wakeup(sc); 1244 while (!(sc->flags & MD_EXITING)) 1245 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10); 1246 mtx_unlock(&sc->queue_mtx); 1247 mtx_destroy(&sc->stat_mtx); 1248 mtx_destroy(&sc->queue_mtx); 1249 if (sc->vnode != NULL) { 1250 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 1251 sc->vnode->v_vflag &= ~VV_MD; 1252 VOP_UNLOCK(sc->vnode, 0); 1253 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 1254 FREAD : (FREAD|FWRITE), sc->cred, td); 1255 } 1256 if (sc->cred != NULL) 1257 crfree(sc->cred); 1258 if (sc->object != NULL) 1259 vm_object_deallocate(sc->object); 1260 if (sc->indir) 1261 destroy_indir(sc, sc->indir); 1262 if (sc->uma) 1263 uma_zdestroy(sc->uma); 1264 1265 LIST_REMOVE(sc, list); 1266 free_unr(md_uh, sc->unit); 1267 free(sc, M_MD); 1268 return (0); 1269} 1270 1271static int 1272mdresize(struct md_s *sc, struct md_ioctl *mdio) 1273{ 1274 int error, res; 1275 vm_pindex_t oldpages, newpages; 1276 1277 switch (sc->type) { 1278 case MD_VNODE: 1279 case MD_NULL: 1280 break; 1281 case MD_SWAP: 1282 if (mdio->md_mediasize <= 0 || 1283 (mdio->md_mediasize % PAGE_SIZE) != 0) 1284 return (EDOM); 1285 oldpages = OFF_TO_IDX(round_page(sc->mediasize)); 1286 newpages = OFF_TO_IDX(round_page(mdio->md_mediasize)); 1287 if (newpages < oldpages) { 1288 VM_OBJECT_WLOCK(sc->object); 1289 vm_object_page_remove(sc->object, newpages, 0, 0); 1290 swap_pager_freespace(sc->object, newpages, 1291 oldpages - newpages); 1292 swap_release_by_cred(IDX_TO_OFF(oldpages - 1293 newpages), sc->cred); 1294 sc->object->charge = IDX_TO_OFF(newpages); 1295 sc->object->size = newpages; 1296 VM_OBJECT_WUNLOCK(sc->object); 1297 } else if (newpages > oldpages) { 1298 res = swap_reserve_by_cred(IDX_TO_OFF(newpages - 1299 oldpages), sc->cred); 1300 if (!res) 1301 return (ENOMEM); 1302 if ((mdio->md_options & MD_RESERVE) || 1303 (sc->flags & MD_RESERVE)) { 1304 error = swap_pager_reserve(sc->object, 1305 oldpages, newpages - oldpages); 1306 if (error < 0) { 1307 swap_release_by_cred( 1308 IDX_TO_OFF(newpages - oldpages), 1309 sc->cred); 1310 return (EDOM); 1311 } 1312 } 1313 VM_OBJECT_WLOCK(sc->object); 1314 sc->object->charge = IDX_TO_OFF(newpages); 1315 sc->object->size = newpages; 1316 VM_OBJECT_WUNLOCK(sc->object); 1317 } 1318 break; 1319 default: 1320 return (EOPNOTSUPP); 1321 } 1322 1323 sc->mediasize = mdio->md_mediasize; 1324 g_topology_lock(); 1325 g_resize_provider(sc->pp, sc->mediasize); 1326 g_topology_unlock(); 1327 return (0); 1328} 1329 1330static int 1331mdcreate_swap(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 1332{ 1333 vm_ooffset_t npage; 1334 int error; 1335 1336 /* 1337 * Range check. Disallow negative sizes and sizes not being 1338 * multiple of page size. 1339 */ 1340 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0) 1341 return (EDOM); 1342 1343 /* 1344 * Allocate an OBJT_SWAP object. 1345 * 1346 * Note the truncation. 1347 */ 1348 1349 npage = mdio->md_mediasize / PAGE_SIZE; 1350 if (mdio->md_fwsectors != 0) 1351 sc->fwsectors = mdio->md_fwsectors; 1352 if (mdio->md_fwheads != 0) 1353 sc->fwheads = mdio->md_fwheads; 1354 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage, 1355 VM_PROT_DEFAULT, 0, td->td_ucred); 1356 if (sc->object == NULL) 1357 return (ENOMEM); 1358 sc->flags = mdio->md_options & (MD_FORCE | MD_RESERVE); 1359 if (mdio->md_options & MD_RESERVE) { 1360 if (swap_pager_reserve(sc->object, 0, npage) < 0) { 1361 error = EDOM; 1362 goto finish; 1363 } 1364 } 1365 error = mdsetcred(sc, td->td_ucred); 1366 finish: 1367 if (error != 0) { 1368 vm_object_deallocate(sc->object); 1369 sc->object = NULL; 1370 } 1371 return (error); 1372} 1373 1374static int 1375mdcreate_null(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 1376{ 1377 1378 /* 1379 * Range check. Disallow negative sizes and sizes not being 1380 * multiple of page size. 1381 */ 1382 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0) 1383 return (EDOM); 1384 1385 return (0); 1386} 1387 1388static int 1389xmdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1390{ 1391 struct md_ioctl *mdio; 1392 struct md_s *sc; 1393 int error, i; 1394 unsigned sectsize; 1395 1396 if (md_debug) 1397 printf("mdctlioctl(%s %lx %p %x %p)\n", 1398 devtoname(dev), cmd, addr, flags, td); 1399 1400 mdio = (struct md_ioctl *)addr; 1401 if (mdio->md_version != MDIOVERSION) 1402 return (EINVAL); 1403 1404 /* 1405 * We assert the version number in the individual ioctl 1406 * handlers instead of out here because (a) it is possible we 1407 * may add another ioctl in the future which doesn't read an 1408 * mdio, and (b) the correct return value for an unknown ioctl 1409 * is ENOIOCTL, not EINVAL. 1410 */ 1411 error = 0; 1412 switch (cmd) { 1413 case MDIOCATTACH: 1414 switch (mdio->md_type) { 1415 case MD_MALLOC: 1416 case MD_PRELOAD: 1417 case MD_VNODE: 1418 case MD_SWAP: 1419 case MD_NULL: 1420 break; 1421 default: 1422 return (EINVAL); 1423 } 1424 if (mdio->md_sectorsize == 0) 1425 sectsize = DEV_BSIZE; 1426 else 1427 sectsize = mdio->md_sectorsize; 1428 if (sectsize > MAXPHYS || mdio->md_mediasize < sectsize) 1429 return (EINVAL); 1430 if (mdio->md_options & MD_AUTOUNIT) 1431 sc = mdnew(-1, &error, mdio->md_type); 1432 else { 1433 if (mdio->md_unit > INT_MAX) 1434 return (EINVAL); 1435 sc = mdnew(mdio->md_unit, &error, mdio->md_type); 1436 } 1437 if (sc == NULL) 1438 return (error); 1439 if (mdio->md_options & MD_AUTOUNIT) 1440 mdio->md_unit = sc->unit; 1441 sc->mediasize = mdio->md_mediasize; 1442 sc->sectorsize = sectsize; 1443 error = EDOOFUS; 1444 switch (sc->type) { 1445 case MD_MALLOC: 1446 sc->start = mdstart_malloc; 1447 error = mdcreate_malloc(sc, mdio); 1448 break; 1449 case MD_PRELOAD: 1450 /* 1451 * We disallow attaching preloaded memory disks via 1452 * ioctl. Preloaded memory disks are automatically 1453 * attached in g_md_init(). 1454 */ 1455 error = EOPNOTSUPP; 1456 break; 1457 case MD_VNODE: 1458 sc->start = mdstart_vnode; 1459 error = mdcreate_vnode(sc, mdio, td); 1460 break; 1461 case MD_SWAP: 1462 sc->start = mdstart_swap; 1463 error = mdcreate_swap(sc, mdio, td); 1464 break; 1465 case MD_NULL: 1466 sc->start = mdstart_null; 1467 error = mdcreate_null(sc, mdio, td); 1468 break; 1469 } 1470 if (error != 0) { 1471 mddestroy(sc, td); 1472 return (error); 1473 } 1474 1475 /* Prune off any residual fractional sector */ 1476 i = sc->mediasize % sc->sectorsize; 1477 sc->mediasize -= i; 1478 1479 mdinit(sc); 1480 return (0); 1481 case MDIOCDETACH: 1482 if (mdio->md_mediasize != 0 || 1483 (mdio->md_options & ~MD_FORCE) != 0) 1484 return (EINVAL); 1485 1486 sc = mdfind(mdio->md_unit); 1487 if (sc == NULL) 1488 return (ENOENT); 1489 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) && 1490 !(mdio->md_options & MD_FORCE)) 1491 return (EBUSY); 1492 return (mddestroy(sc, td)); 1493 case MDIOCRESIZE: 1494 if ((mdio->md_options & ~(MD_FORCE | MD_RESERVE)) != 0) 1495 return (EINVAL); 1496 1497 sc = mdfind(mdio->md_unit); 1498 if (sc == NULL) 1499 return (ENOENT); 1500 if (mdio->md_mediasize < sc->sectorsize) 1501 return (EINVAL); 1502 if (mdio->md_mediasize < sc->mediasize && 1503 !(sc->flags & MD_FORCE) && 1504 !(mdio->md_options & MD_FORCE)) 1505 return (EBUSY); 1506 return (mdresize(sc, mdio)); 1507 case MDIOCQUERY: 1508 sc = mdfind(mdio->md_unit); 1509 if (sc == NULL) 1510 return (ENOENT); 1511 mdio->md_type = sc->type; 1512 mdio->md_options = sc->flags; 1513 mdio->md_mediasize = sc->mediasize; 1514 mdio->md_sectorsize = sc->sectorsize; 1515 if (sc->type == MD_VNODE) 1516 error = copyout(sc->file, mdio->md_file, 1517 strlen(sc->file) + 1); 1518 return (error); 1519 case MDIOCLIST: 1520 i = 1; 1521 LIST_FOREACH(sc, &md_softc_list, list) { 1522 if (i == MDNPAD - 1) 1523 mdio->md_pad[i] = -1; 1524 else 1525 mdio->md_pad[i++] = sc->unit; 1526 } 1527 mdio->md_pad[0] = i - 1; 1528 return (0); 1529 default: 1530 return (ENOIOCTL); 1531 }; 1532} 1533 1534static int 1535mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1536{ 1537 int error; 1538 1539 sx_xlock(&md_sx); 1540 error = xmdctlioctl(dev, cmd, addr, flags, td); 1541 sx_xunlock(&md_sx); 1542 return (error); 1543} 1544 1545static void 1546md_preloaded(u_char *image, size_t length, const char *name) 1547{ 1548 struct md_s *sc; 1549 int error; 1550 1551 sc = mdnew(-1, &error, MD_PRELOAD); 1552 if (sc == NULL) 1553 return; 1554 sc->mediasize = length; 1555 sc->sectorsize = DEV_BSIZE; 1556 sc->pl_ptr = image; 1557 sc->pl_len = length; 1558 sc->start = mdstart_preload; 1559#ifdef MD_ROOT 1560 if (sc->unit == 0) 1561 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0"; 1562#endif 1563 mdinit(sc); 1564 if (name != NULL) { 1565 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n", 1566 MD_NAME, sc->unit, name, length, image); 1567 } else { 1568 printf("%s%d: Embedded image %zd bytes as %p\n", 1569 MD_NAME, sc->unit, length, image); 1570 } 1571} 1572 1573static void 1574g_md_init(struct g_class *mp __unused) 1575{ 1576 caddr_t mod; 1577 u_char *ptr, *name, *type; 1578 unsigned len; 1579 int i; 1580 1581 /* figure out log2(NINDIR) */ 1582 for (i = NINDIR, nshift = -1; i; nshift++) 1583 i >>= 1; 1584 1585 mod = NULL; 1586 sx_init(&md_sx, "MD config lock"); 1587 g_topology_unlock(); 1588 md_uh = new_unrhdr(0, INT_MAX, NULL); 1589#ifdef MD_ROOT 1590 if (mfs_root_size != 0) { 1591 sx_xlock(&md_sx); 1592 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size, 1593 NULL); 1594 sx_xunlock(&md_sx); 1595 } 1596#endif 1597 /* XXX: are preload_* static or do they need Giant ? */ 1598 while ((mod = preload_search_next_name(mod)) != NULL) { 1599 name = (char *)preload_search_info(mod, MODINFO_NAME); 1600 if (name == NULL) 1601 continue; 1602 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1603 if (type == NULL) 1604 continue; 1605 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1606 continue; 1607 ptr = preload_fetch_addr(mod); 1608 len = preload_fetch_size(mod); 1609 if (ptr != NULL && len != 0) { 1610 sx_xlock(&md_sx); 1611 md_preloaded(ptr, len, name); 1612 sx_xunlock(&md_sx); 1613 } 1614 } 1615 md_vnode_pbuf_freecnt = nswbuf / 10; 1616 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL, 1617 0600, MDCTL_NAME); 1618 g_topology_lock(); 1619} 1620 1621static void 1622g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 1623 struct g_consumer *cp __unused, struct g_provider *pp) 1624{ 1625 struct md_s *mp; 1626 char *type; 1627 1628 mp = gp->softc; 1629 if (mp == NULL) 1630 return; 1631 1632 switch (mp->type) { 1633 case MD_MALLOC: 1634 type = "malloc"; 1635 break; 1636 case MD_PRELOAD: 1637 type = "preload"; 1638 break; 1639 case MD_VNODE: 1640 type = "vnode"; 1641 break; 1642 case MD_SWAP: 1643 type = "swap"; 1644 break; 1645 case MD_NULL: 1646 type = "null"; 1647 break; 1648 default: 1649 type = "unknown"; 1650 break; 1651 } 1652 1653 if (pp != NULL) { 1654 if (indent == NULL) { 1655 sbuf_printf(sb, " u %d", mp->unit); 1656 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize); 1657 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads); 1658 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors); 1659 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize); 1660 sbuf_printf(sb, " t %s", type); 1661 if (mp->type == MD_VNODE && mp->vnode != NULL) 1662 sbuf_printf(sb, " file %s", mp->file); 1663 } else { 1664 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, 1665 mp->unit); 1666 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n", 1667 indent, (uintmax_t) mp->sectorsize); 1668 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n", 1669 indent, (uintmax_t) mp->fwheads); 1670 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n", 1671 indent, (uintmax_t) mp->fwsectors); 1672 sbuf_printf(sb, "%s<length>%ju</length>\n", 1673 indent, (uintmax_t) mp->mediasize); 1674 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent, 1675 (mp->flags & MD_COMPRESS) == 0 ? "off": "on"); 1676 sbuf_printf(sb, "%s<access>%s</access>\n", indent, 1677 (mp->flags & MD_READONLY) == 0 ? "read-write": 1678 "read-only"); 1679 sbuf_printf(sb, "%s<type>%s</type>\n", indent, 1680 type); 1681 if (mp->type == MD_VNODE && mp->vnode != NULL) { 1682 sbuf_printf(sb, "%s<file>", indent); 1683 g_conf_printf_escaped(sb, "%s", mp->file); 1684 sbuf_printf(sb, "</file>\n"); 1685 } 1686 } 1687 } 1688} 1689 1690static void 1691g_md_fini(struct g_class *mp __unused) 1692{ 1693 1694 sx_destroy(&md_sx); 1695 if (status_dev != NULL) 1696 destroy_dev(status_dev); 1697 delete_unrhdr(md_uh); 1698} 1699