md.c revision 238991
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 238991 2012-08-02 15:05:34Z jh $ 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 * 4. Neither the name of the University nor the names of its contributors 35 * may be used to endorse or promote products derived from this software 36 * without specific prior written permission. 37 * 38 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 39 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 40 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 41 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 42 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 43 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 44 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 45 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 46 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 47 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 48 * SUCH DAMAGE. 49 * 50 * from: Utah Hdr: vn.c 1.13 94/04/02 51 * 52 * from: @(#)vn.c 8.6 (Berkeley) 4/1/94 53 * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03 54 */ 55 56#include "opt_geom.h" 57#include "opt_md.h" 58 59#include <sys/param.h> 60#include <sys/systm.h> 61#include <sys/bio.h> 62#include <sys/conf.h> 63#include <sys/devicestat.h> 64#include <sys/fcntl.h> 65#include <sys/kernel.h> 66#include <sys/kthread.h> 67#include <sys/limits.h> 68#include <sys/linker.h> 69#include <sys/lock.h> 70#include <sys/malloc.h> 71#include <sys/mdioctl.h> 72#include <sys/mount.h> 73#include <sys/mutex.h> 74#include <sys/sx.h> 75#include <sys/namei.h> 76#include <sys/proc.h> 77#include <sys/queue.h> 78#include <sys/sbuf.h> 79#include <sys/sched.h> 80#include <sys/sf_buf.h> 81#include <sys/sysctl.h> 82#include <sys/vnode.h> 83 84#include <geom/geom.h> 85 86#include <vm/vm.h> 87#include <vm/vm_object.h> 88#include <vm/vm_page.h> 89#include <vm/vm_pager.h> 90#include <vm/swap_pager.h> 91#include <vm/uma.h> 92 93#include <machine/vmparam.h> 94 95#define MD_MODVER 1 96 97#define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */ 98#define MD_EXITING 0x20000 /* Worker thread is exiting. */ 99 100#ifndef MD_NSECT 101#define MD_NSECT (10000 * 2) 102#endif 103 104static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk"); 105static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors"); 106 107static int md_debug; 108SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, 109 "Enable md(4) debug messages"); 110static int md_malloc_wait; 111SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0, 112 "Allow malloc to wait for memory allocations"); 113 114#if defined(MD_ROOT) && defined(MD_ROOT_SIZE) 115/* 116 * Preloaded image gets put here. 117 * Applications that patch the object with the image can determine 118 * the size looking at the start and end markers (strings), 119 * so we want them contiguous. 120 */ 121static struct { 122 u_char start[MD_ROOT_SIZE*1024]; 123 u_char end[128]; 124} mfs_root = { 125 .start = "MFS Filesystem goes here", 126 .end = "MFS Filesystem had better STOP here", 127}; 128#endif 129 130static g_init_t g_md_init; 131static g_fini_t g_md_fini; 132static g_start_t g_md_start; 133static g_access_t g_md_access; 134static void g_md_dumpconf(struct sbuf *sb, const char *indent, 135 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp); 136 137static int mdunits; 138static struct cdev *status_dev = 0; 139static struct sx md_sx; 140static struct unrhdr *md_uh; 141 142static d_ioctl_t mdctlioctl; 143 144static struct cdevsw mdctl_cdevsw = { 145 .d_version = D_VERSION, 146 .d_ioctl = mdctlioctl, 147 .d_name = MD_NAME, 148}; 149 150struct g_class g_md_class = { 151 .name = "MD", 152 .version = G_VERSION, 153 .init = g_md_init, 154 .fini = g_md_fini, 155 .start = g_md_start, 156 .access = g_md_access, 157 .dumpconf = g_md_dumpconf, 158}; 159 160DECLARE_GEOM_CLASS(g_md_class, g_md); 161 162 163static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list); 164 165#define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 166#define NMASK (NINDIR-1) 167static int nshift; 168 169struct indir { 170 uintptr_t *array; 171 u_int total; 172 u_int used; 173 u_int shift; 174}; 175 176struct md_s { 177 int unit; 178 LIST_ENTRY(md_s) list; 179 struct bio_queue_head bio_queue; 180 struct mtx queue_mtx; 181 struct cdev *dev; 182 enum md_types type; 183 off_t mediasize; 184 unsigned sectorsize; 185 unsigned opencount; 186 unsigned fwheads; 187 unsigned fwsectors; 188 unsigned flags; 189 char name[20]; 190 struct proc *procp; 191 struct g_geom *gp; 192 struct g_provider *pp; 193 int (*start)(struct md_s *sc, struct bio *bp); 194 struct devstat *devstat; 195 196 /* MD_MALLOC related fields */ 197 struct indir *indir; 198 uma_zone_t uma; 199 200 /* MD_PRELOAD related fields */ 201 u_char *pl_ptr; 202 size_t pl_len; 203 204 /* MD_VNODE related fields */ 205 struct vnode *vnode; 206 char file[PATH_MAX]; 207 struct ucred *cred; 208 209 /* MD_SWAP related fields */ 210 vm_object_t object; 211}; 212 213static struct indir * 214new_indir(u_int shift) 215{ 216 struct indir *ip; 217 218 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT) 219 | M_ZERO); 220 if (ip == NULL) 221 return (NULL); 222 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 223 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO); 224 if (ip->array == NULL) { 225 free(ip, M_MD); 226 return (NULL); 227 } 228 ip->total = NINDIR; 229 ip->shift = shift; 230 return (ip); 231} 232 233static void 234del_indir(struct indir *ip) 235{ 236 237 free(ip->array, M_MDSECT); 238 free(ip, M_MD); 239} 240 241static void 242destroy_indir(struct md_s *sc, struct indir *ip) 243{ 244 int i; 245 246 for (i = 0; i < NINDIR; i++) { 247 if (!ip->array[i]) 248 continue; 249 if (ip->shift) 250 destroy_indir(sc, (struct indir*)(ip->array[i])); 251 else if (ip->array[i] > 255) 252 uma_zfree(sc->uma, (void *)(ip->array[i])); 253 } 254 del_indir(ip); 255} 256 257/* 258 * This function does the math and allocates the top level "indir" structure 259 * for a device of "size" sectors. 260 */ 261 262static struct indir * 263dimension(off_t size) 264{ 265 off_t rcnt; 266 struct indir *ip; 267 int layer; 268 269 rcnt = size; 270 layer = 0; 271 while (rcnt > NINDIR) { 272 rcnt /= NINDIR; 273 layer++; 274 } 275 276 /* 277 * XXX: the top layer is probably not fully populated, so we allocate 278 * too much space for ip->array in here. 279 */ 280 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 281 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 282 M_MDSECT, M_WAITOK | M_ZERO); 283 ip->total = NINDIR; 284 ip->shift = layer * nshift; 285 return (ip); 286} 287 288/* 289 * Read a given sector 290 */ 291 292static uintptr_t 293s_read(struct indir *ip, off_t offset) 294{ 295 struct indir *cip; 296 int idx; 297 uintptr_t up; 298 299 if (md_debug > 1) 300 printf("s_read(%jd)\n", (intmax_t)offset); 301 up = 0; 302 for (cip = ip; cip != NULL;) { 303 if (cip->shift) { 304 idx = (offset >> cip->shift) & NMASK; 305 up = cip->array[idx]; 306 cip = (struct indir *)up; 307 continue; 308 } 309 idx = offset & NMASK; 310 return (cip->array[idx]); 311 } 312 return (0); 313} 314 315/* 316 * Write a given sector, prune the tree if the value is 0 317 */ 318 319static int 320s_write(struct indir *ip, off_t offset, uintptr_t ptr) 321{ 322 struct indir *cip, *lip[10]; 323 int idx, li; 324 uintptr_t up; 325 326 if (md_debug > 1) 327 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 328 up = 0; 329 li = 0; 330 cip = ip; 331 for (;;) { 332 lip[li++] = cip; 333 if (cip->shift) { 334 idx = (offset >> cip->shift) & NMASK; 335 up = cip->array[idx]; 336 if (up != 0) { 337 cip = (struct indir *)up; 338 continue; 339 } 340 /* Allocate branch */ 341 cip->array[idx] = 342 (uintptr_t)new_indir(cip->shift - nshift); 343 if (cip->array[idx] == 0) 344 return (ENOSPC); 345 cip->used++; 346 up = cip->array[idx]; 347 cip = (struct indir *)up; 348 continue; 349 } 350 /* leafnode */ 351 idx = offset & NMASK; 352 up = cip->array[idx]; 353 if (up != 0) 354 cip->used--; 355 cip->array[idx] = ptr; 356 if (ptr != 0) 357 cip->used++; 358 break; 359 } 360 if (cip->used != 0 || li == 1) 361 return (0); 362 li--; 363 while (cip->used == 0 && cip != ip) { 364 li--; 365 idx = (offset >> lip[li]->shift) & NMASK; 366 up = lip[li]->array[idx]; 367 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 368 del_indir(cip); 369 lip[li]->array[idx] = 0; 370 lip[li]->used--; 371 cip = lip[li]; 372 } 373 return (0); 374} 375 376 377static int 378g_md_access(struct g_provider *pp, int r, int w, int e) 379{ 380 struct md_s *sc; 381 382 sc = pp->geom->softc; 383 if (sc == NULL) { 384 if (r <= 0 && w <= 0 && e <= 0) 385 return (0); 386 return (ENXIO); 387 } 388 r += pp->acr; 389 w += pp->acw; 390 e += pp->ace; 391 if ((sc->flags & MD_READONLY) != 0 && w > 0) 392 return (EROFS); 393 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 394 sc->opencount = 1; 395 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 396 sc->opencount = 0; 397 } 398 return (0); 399} 400 401static void 402g_md_start(struct bio *bp) 403{ 404 struct md_s *sc; 405 406 sc = bp->bio_to->geom->softc; 407 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) 408 devstat_start_transaction_bio(sc->devstat, bp); 409 mtx_lock(&sc->queue_mtx); 410 bioq_disksort(&sc->bio_queue, bp); 411 mtx_unlock(&sc->queue_mtx); 412 wakeup(sc); 413} 414 415static int 416mdstart_malloc(struct md_s *sc, struct bio *bp) 417{ 418 int i, error; 419 u_char *dst; 420 off_t secno, nsec, uc; 421 uintptr_t sp, osp; 422 423 switch (bp->bio_cmd) { 424 case BIO_READ: 425 case BIO_WRITE: 426 case BIO_DELETE: 427 break; 428 default: 429 return (EOPNOTSUPP); 430 } 431 432 nsec = bp->bio_length / sc->sectorsize; 433 secno = bp->bio_offset / sc->sectorsize; 434 dst = bp->bio_data; 435 error = 0; 436 while (nsec--) { 437 osp = s_read(sc->indir, secno); 438 if (bp->bio_cmd == BIO_DELETE) { 439 if (osp != 0) 440 error = s_write(sc->indir, secno, 0); 441 } else if (bp->bio_cmd == BIO_READ) { 442 if (osp == 0) 443 bzero(dst, sc->sectorsize); 444 else if (osp <= 255) 445 memset(dst, osp, sc->sectorsize); 446 else { 447 bcopy((void *)osp, dst, sc->sectorsize); 448 cpu_flush_dcache(dst, sc->sectorsize); 449 } 450 osp = 0; 451 } else if (bp->bio_cmd == BIO_WRITE) { 452 if (sc->flags & MD_COMPRESS) { 453 uc = dst[0]; 454 for (i = 1; i < sc->sectorsize; i++) 455 if (dst[i] != uc) 456 break; 457 } else { 458 i = 0; 459 uc = 0; 460 } 461 if (i == sc->sectorsize) { 462 if (osp != uc) 463 error = s_write(sc->indir, secno, uc); 464 } else { 465 if (osp <= 255) { 466 sp = (uintptr_t)uma_zalloc(sc->uma, 467 md_malloc_wait ? M_WAITOK : 468 M_NOWAIT); 469 if (sp == 0) { 470 error = ENOSPC; 471 break; 472 } 473 bcopy(dst, (void *)sp, sc->sectorsize); 474 error = s_write(sc->indir, secno, sp); 475 } else { 476 bcopy(dst, (void *)osp, sc->sectorsize); 477 osp = 0; 478 } 479 } 480 } else { 481 error = EOPNOTSUPP; 482 } 483 if (osp > 255) 484 uma_zfree(sc->uma, (void*)osp); 485 if (error != 0) 486 break; 487 secno++; 488 dst += sc->sectorsize; 489 } 490 bp->bio_resid = 0; 491 return (error); 492} 493 494static int 495mdstart_preload(struct md_s *sc, struct bio *bp) 496{ 497 498 switch (bp->bio_cmd) { 499 case BIO_READ: 500 bcopy(sc->pl_ptr + bp->bio_offset, bp->bio_data, 501 bp->bio_length); 502 cpu_flush_dcache(bp->bio_data, bp->bio_length); 503 break; 504 case BIO_WRITE: 505 bcopy(bp->bio_data, sc->pl_ptr + bp->bio_offset, 506 bp->bio_length); 507 break; 508 } 509 bp->bio_resid = 0; 510 return (0); 511} 512 513static int 514mdstart_vnode(struct md_s *sc, struct bio *bp) 515{ 516 int error, vfslocked; 517 struct uio auio; 518 struct iovec aiov; 519 struct mount *mp; 520 struct vnode *vp; 521 struct thread *td; 522 off_t end, zerosize; 523 524 switch (bp->bio_cmd) { 525 case BIO_READ: 526 case BIO_WRITE: 527 case BIO_DELETE: 528 case BIO_FLUSH: 529 break; 530 default: 531 return (EOPNOTSUPP); 532 } 533 534 td = curthread; 535 vp = sc->vnode; 536 537 /* 538 * VNODE I/O 539 * 540 * If an error occurs, we set BIO_ERROR but we do not set 541 * B_INVAL because (for a write anyway), the buffer is 542 * still valid. 543 */ 544 545 if (bp->bio_cmd == BIO_FLUSH) { 546 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 547 (void) vn_start_write(vp, &mp, V_WAIT); 548 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 549 error = VOP_FSYNC(vp, MNT_WAIT, td); 550 VOP_UNLOCK(vp, 0); 551 vn_finished_write(mp); 552 VFS_UNLOCK_GIANT(vfslocked); 553 return (error); 554 } 555 556 bzero(&auio, sizeof(auio)); 557 558 /* 559 * Special case for BIO_DELETE. On the surface, this is very 560 * similar to BIO_WRITE, except that we write from our own 561 * fixed-length buffer, so we have to loop. The net result is 562 * that the two cases end up having very little in common. 563 */ 564 if (bp->bio_cmd == BIO_DELETE) { 565 zerosize = ZERO_REGION_SIZE - 566 (ZERO_REGION_SIZE % sc->sectorsize); 567 auio.uio_iov = &aiov; 568 auio.uio_iovcnt = 1; 569 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 570 auio.uio_segflg = UIO_SYSSPACE; 571 auio.uio_rw = UIO_WRITE; 572 auio.uio_td = td; 573 end = bp->bio_offset + bp->bio_length; 574 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 575 (void) vn_start_write(vp, &mp, V_WAIT); 576 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 577 error = 0; 578 while (auio.uio_offset < end) { 579 aiov.iov_base = __DECONST(void *, zero_region); 580 aiov.iov_len = end - auio.uio_offset; 581 if (aiov.iov_len > zerosize) 582 aiov.iov_len = zerosize; 583 auio.uio_resid = aiov.iov_len; 584 error = VOP_WRITE(vp, &auio, 585 sc->flags & MD_ASYNC ? 0 : IO_SYNC, sc->cred); 586 if (error != 0) 587 break; 588 } 589 VOP_UNLOCK(vp, 0); 590 vn_finished_write(mp); 591 bp->bio_resid = end - auio.uio_offset; 592 VFS_UNLOCK_GIANT(vfslocked); 593 return (error); 594 } 595 596 aiov.iov_base = bp->bio_data; 597 aiov.iov_len = bp->bio_length; 598 auio.uio_iov = &aiov; 599 auio.uio_iovcnt = 1; 600 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 601 auio.uio_segflg = UIO_SYSSPACE; 602 if (bp->bio_cmd == BIO_READ) 603 auio.uio_rw = UIO_READ; 604 else if (bp->bio_cmd == BIO_WRITE) 605 auio.uio_rw = UIO_WRITE; 606 else 607 panic("wrong BIO_OP in mdstart_vnode"); 608 auio.uio_resid = bp->bio_length; 609 auio.uio_td = td; 610 /* 611 * When reading set IO_DIRECT to try to avoid double-caching 612 * the data. When writing IO_DIRECT is not optimal. 613 */ 614 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 615 if (bp->bio_cmd == BIO_READ) { 616 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 617 error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred); 618 VOP_UNLOCK(vp, 0); 619 } else { 620 (void) vn_start_write(vp, &mp, V_WAIT); 621 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 622 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC, 623 sc->cred); 624 VOP_UNLOCK(vp, 0); 625 vn_finished_write(mp); 626 } 627 VFS_UNLOCK_GIANT(vfslocked); 628 bp->bio_resid = auio.uio_resid; 629 return (error); 630} 631 632static int 633mdstart_swap(struct md_s *sc, struct bio *bp) 634{ 635 struct sf_buf *sf; 636 int rv, offs, len, lastend; 637 vm_pindex_t i, lastp; 638 vm_page_t m; 639 u_char *p; 640 641 switch (bp->bio_cmd) { 642 case BIO_READ: 643 case BIO_WRITE: 644 case BIO_DELETE: 645 break; 646 default: 647 return (EOPNOTSUPP); 648 } 649 650 p = bp->bio_data; 651 652 /* 653 * offs is the offset at which to start operating on the 654 * next (ie, first) page. lastp is the last page on 655 * which we're going to operate. lastend is the ending 656 * position within that last page (ie, PAGE_SIZE if 657 * we're operating on complete aligned pages). 658 */ 659 offs = bp->bio_offset % PAGE_SIZE; 660 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE; 661 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1; 662 663 rv = VM_PAGER_OK; 664 VM_OBJECT_LOCK(sc->object); 665 vm_object_pip_add(sc->object, 1); 666 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) { 667 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs; 668 669 m = vm_page_grab(sc->object, i, 670 VM_ALLOC_NORMAL|VM_ALLOC_RETRY); 671 VM_OBJECT_UNLOCK(sc->object); 672 sched_pin(); 673 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 674 VM_OBJECT_LOCK(sc->object); 675 if (bp->bio_cmd == BIO_READ) { 676 if (m->valid != VM_PAGE_BITS_ALL) 677 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 678 if (rv == VM_PAGER_ERROR) { 679 sf_buf_free(sf); 680 sched_unpin(); 681 vm_page_wakeup(m); 682 break; 683 } 684 bcopy((void *)(sf_buf_kva(sf) + offs), p, len); 685 cpu_flush_dcache(p, len); 686 } else if (bp->bio_cmd == BIO_WRITE) { 687 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 688 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 689 if (rv == VM_PAGER_ERROR) { 690 sf_buf_free(sf); 691 sched_unpin(); 692 vm_page_wakeup(m); 693 break; 694 } 695 bcopy(p, (void *)(sf_buf_kva(sf) + offs), len); 696 m->valid = VM_PAGE_BITS_ALL; 697 } else if (bp->bio_cmd == BIO_DELETE) { 698 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 699 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 700 if (rv == VM_PAGER_ERROR) { 701 sf_buf_free(sf); 702 sched_unpin(); 703 vm_page_wakeup(m); 704 break; 705 } 706 if (len != PAGE_SIZE) { 707 bzero((void *)(sf_buf_kva(sf) + offs), len); 708 vm_page_clear_dirty(m, offs, len); 709 m->valid = VM_PAGE_BITS_ALL; 710 } else 711 vm_pager_page_unswapped(m); 712 } 713 sf_buf_free(sf); 714 sched_unpin(); 715 vm_page_wakeup(m); 716 vm_page_lock(m); 717 if (bp->bio_cmd == BIO_DELETE && len == PAGE_SIZE) 718 vm_page_free(m); 719 else 720 vm_page_activate(m); 721 vm_page_unlock(m); 722 if (bp->bio_cmd == BIO_WRITE) 723 vm_page_dirty(m); 724 725 /* Actions on further pages start at offset 0 */ 726 p += PAGE_SIZE - offs; 727 offs = 0; 728#if 0 729if (bootverbose || bp->bio_offset / PAGE_SIZE < 17) 730printf("wire_count %d busy %d flags %x hold_count %d act_count %d queue %d valid %d dirty %d @ %d\n", 731 m->wire_count, m->busy, 732 m->flags, m->hold_count, m->act_count, m->queue, m->valid, m->dirty, i); 733#endif 734 } 735 vm_object_pip_subtract(sc->object, 1); 736 VM_OBJECT_UNLOCK(sc->object); 737 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC); 738} 739 740static void 741md_kthread(void *arg) 742{ 743 struct md_s *sc; 744 struct bio *bp; 745 int error; 746 747 sc = arg; 748 thread_lock(curthread); 749 sched_prio(curthread, PRIBIO); 750 thread_unlock(curthread); 751 if (sc->type == MD_VNODE) 752 curthread->td_pflags |= TDP_NORUNNINGBUF; 753 754 for (;;) { 755 mtx_lock(&sc->queue_mtx); 756 if (sc->flags & MD_SHUTDOWN) { 757 sc->flags |= MD_EXITING; 758 mtx_unlock(&sc->queue_mtx); 759 kproc_exit(0); 760 } 761 bp = bioq_takefirst(&sc->bio_queue); 762 if (!bp) { 763 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 764 continue; 765 } 766 mtx_unlock(&sc->queue_mtx); 767 if (bp->bio_cmd == BIO_GETATTR) { 768 if ((sc->fwsectors && sc->fwheads && 769 (g_handleattr_int(bp, "GEOM::fwsectors", 770 sc->fwsectors) || 771 g_handleattr_int(bp, "GEOM::fwheads", 772 sc->fwheads))) || 773 g_handleattr_int(bp, "GEOM::candelete", 1)) 774 error = -1; 775 else 776 error = EOPNOTSUPP; 777 } else { 778 error = sc->start(sc, bp); 779 } 780 781 if (error != -1) { 782 bp->bio_completed = bp->bio_length; 783 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) 784 devstat_end_transaction_bio(sc->devstat, bp); 785 g_io_deliver(bp, error); 786 } 787 } 788} 789 790static struct md_s * 791mdfind(int unit) 792{ 793 struct md_s *sc; 794 795 LIST_FOREACH(sc, &md_softc_list, list) { 796 if (sc->unit == unit) 797 break; 798 } 799 return (sc); 800} 801 802static struct md_s * 803mdnew(int unit, int *errp, enum md_types type) 804{ 805 struct md_s *sc; 806 int error; 807 808 *errp = 0; 809 if (unit == -1) 810 unit = alloc_unr(md_uh); 811 else 812 unit = alloc_unr_specific(md_uh, unit); 813 814 if (unit == -1) { 815 *errp = EBUSY; 816 return (NULL); 817 } 818 819 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 820 sc->type = type; 821 bioq_init(&sc->bio_queue); 822 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 823 sc->unit = unit; 824 sprintf(sc->name, "md%d", unit); 825 LIST_INSERT_HEAD(&md_softc_list, sc, list); 826 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 827 if (error == 0) 828 return (sc); 829 LIST_REMOVE(sc, list); 830 mtx_destroy(&sc->queue_mtx); 831 free_unr(md_uh, sc->unit); 832 free(sc, M_MD); 833 *errp = error; 834 return (NULL); 835} 836 837static void 838mdinit(struct md_s *sc) 839{ 840 struct g_geom *gp; 841 struct g_provider *pp; 842 843 g_topology_lock(); 844 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 845 gp->softc = sc; 846 pp = g_new_providerf(gp, "md%d", sc->unit); 847 pp->mediasize = sc->mediasize; 848 pp->sectorsize = sc->sectorsize; 849 sc->gp = gp; 850 sc->pp = pp; 851 g_error_provider(pp, 0); 852 g_topology_unlock(); 853 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize, 854 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX); 855} 856 857/* 858 * XXX: we should check that the range they feed us is mapped. 859 * XXX: we should implement read-only. 860 */ 861 862static int 863mdcreate_preload(struct md_s *sc, struct md_ioctl *mdio) 864{ 865 866 if (mdio->md_options & ~(MD_AUTOUNIT | MD_FORCE)) 867 return (EINVAL); 868 if (mdio->md_base == 0) 869 return (EINVAL); 870 sc->flags = mdio->md_options & MD_FORCE; 871 /* Cast to pointer size, then to pointer to avoid warning */ 872 sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base; 873 sc->pl_len = (size_t)sc->mediasize; 874 return (0); 875} 876 877 878static int 879mdcreate_malloc(struct md_s *sc, struct md_ioctl *mdio) 880{ 881 uintptr_t sp; 882 int error; 883 off_t u; 884 885 error = 0; 886 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 887 return (EINVAL); 888 if (mdio->md_sectorsize != 0 && !powerof2(mdio->md_sectorsize)) 889 return (EINVAL); 890 /* Compression doesn't make sense if we have reserved space */ 891 if (mdio->md_options & MD_RESERVE) 892 mdio->md_options &= ~MD_COMPRESS; 893 if (mdio->md_fwsectors != 0) 894 sc->fwsectors = mdio->md_fwsectors; 895 if (mdio->md_fwheads != 0) 896 sc->fwheads = mdio->md_fwheads; 897 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 898 sc->indir = dimension(sc->mediasize / sc->sectorsize); 899 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL, 900 0x1ff, 0); 901 if (mdio->md_options & MD_RESERVE) { 902 off_t nsectors; 903 904 nsectors = sc->mediasize / sc->sectorsize; 905 for (u = 0; u < nsectors; u++) { 906 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ? 907 M_WAITOK : M_NOWAIT) | M_ZERO); 908 if (sp != 0) 909 error = s_write(sc->indir, u, sp); 910 else 911 error = ENOMEM; 912 if (error != 0) 913 break; 914 } 915 } 916 return (error); 917} 918 919 920static int 921mdsetcred(struct md_s *sc, struct ucred *cred) 922{ 923 char *tmpbuf; 924 int error = 0; 925 926 /* 927 * Set credits in our softc 928 */ 929 930 if (sc->cred) 931 crfree(sc->cred); 932 sc->cred = crhold(cred); 933 934 /* 935 * Horrible kludge to establish credentials for NFS XXX. 936 */ 937 938 if (sc->vnode) { 939 struct uio auio; 940 struct iovec aiov; 941 942 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK); 943 bzero(&auio, sizeof(auio)); 944 945 aiov.iov_base = tmpbuf; 946 aiov.iov_len = sc->sectorsize; 947 auio.uio_iov = &aiov; 948 auio.uio_iovcnt = 1; 949 auio.uio_offset = 0; 950 auio.uio_rw = UIO_READ; 951 auio.uio_segflg = UIO_SYSSPACE; 952 auio.uio_resid = aiov.iov_len; 953 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 954 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 955 VOP_UNLOCK(sc->vnode, 0); 956 free(tmpbuf, M_TEMP); 957 } 958 return (error); 959} 960 961static int 962mdcreate_vnode(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 963{ 964 struct vattr vattr; 965 struct nameidata nd; 966 char *fname; 967 int error, flags, vfslocked; 968 969 /* 970 * Kernel-originated requests must have the filename appended 971 * to the mdio structure to protect against malicious software. 972 */ 973 fname = mdio->md_file; 974 if ((void *)fname != (void *)(mdio + 1)) { 975 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL); 976 if (error != 0) 977 return (error); 978 } else 979 strlcpy(sc->file, fname, sizeof(sc->file)); 980 981 /* 982 * If the user specified that this is a read only device, don't 983 * set the FWRITE mask before trying to open the backing store. 984 */ 985 flags = FREAD | ((mdio->md_options & MD_READONLY) ? 0 : FWRITE); 986 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, sc->file, td); 987 error = vn_open(&nd, &flags, 0, NULL); 988 if (error != 0) 989 return (error); 990 vfslocked = NDHASGIANT(&nd); 991 NDFREE(&nd, NDF_ONLY_PNBUF); 992 if (nd.ni_vp->v_type != VREG) { 993 error = EINVAL; 994 goto bad; 995 } 996 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred); 997 if (error != 0) 998 goto bad; 999 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) { 1000 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY); 1001 if (nd.ni_vp->v_iflag & VI_DOOMED) { 1002 /* Forced unmount. */ 1003 error = EBADF; 1004 goto bad; 1005 } 1006 } 1007 nd.ni_vp->v_vflag |= VV_MD; 1008 VOP_UNLOCK(nd.ni_vp, 0); 1009 1010 if (mdio->md_fwsectors != 0) 1011 sc->fwsectors = mdio->md_fwsectors; 1012 if (mdio->md_fwheads != 0) 1013 sc->fwheads = mdio->md_fwheads; 1014 sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC); 1015 if (!(flags & FWRITE)) 1016 sc->flags |= MD_READONLY; 1017 sc->vnode = nd.ni_vp; 1018 1019 error = mdsetcred(sc, td->td_ucred); 1020 if (error != 0) { 1021 sc->vnode = NULL; 1022 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY); 1023 nd.ni_vp->v_vflag &= ~VV_MD; 1024 goto bad; 1025 } 1026 VFS_UNLOCK_GIANT(vfslocked); 1027 return (0); 1028bad: 1029 VOP_UNLOCK(nd.ni_vp, 0); 1030 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td); 1031 VFS_UNLOCK_GIANT(vfslocked); 1032 return (error); 1033} 1034 1035static int 1036mddestroy(struct md_s *sc, struct thread *td) 1037{ 1038 int vfslocked; 1039 1040 if (sc->gp) { 1041 sc->gp->softc = NULL; 1042 g_topology_lock(); 1043 g_wither_geom(sc->gp, ENXIO); 1044 g_topology_unlock(); 1045 sc->gp = NULL; 1046 sc->pp = NULL; 1047 } 1048 if (sc->devstat) { 1049 devstat_remove_entry(sc->devstat); 1050 sc->devstat = NULL; 1051 } 1052 mtx_lock(&sc->queue_mtx); 1053 sc->flags |= MD_SHUTDOWN; 1054 wakeup(sc); 1055 while (!(sc->flags & MD_EXITING)) 1056 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10); 1057 mtx_unlock(&sc->queue_mtx); 1058 mtx_destroy(&sc->queue_mtx); 1059 if (sc->vnode != NULL) { 1060 vfslocked = VFS_LOCK_GIANT(sc->vnode->v_mount); 1061 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 1062 sc->vnode->v_vflag &= ~VV_MD; 1063 VOP_UNLOCK(sc->vnode, 0); 1064 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 1065 FREAD : (FREAD|FWRITE), sc->cred, td); 1066 VFS_UNLOCK_GIANT(vfslocked); 1067 } 1068 if (sc->cred != NULL) 1069 crfree(sc->cred); 1070 if (sc->object != NULL) 1071 vm_object_deallocate(sc->object); 1072 if (sc->indir) 1073 destroy_indir(sc, sc->indir); 1074 if (sc->uma) 1075 uma_zdestroy(sc->uma); 1076 1077 LIST_REMOVE(sc, list); 1078 free_unr(md_uh, sc->unit); 1079 free(sc, M_MD); 1080 return (0); 1081} 1082 1083static int 1084mdresize(struct md_s *sc, struct md_ioctl *mdio) 1085{ 1086 int error, res; 1087 vm_pindex_t oldpages, newpages; 1088 1089 switch (sc->type) { 1090 case MD_VNODE: 1091 break; 1092 case MD_SWAP: 1093 if (mdio->md_mediasize <= 0 || 1094 (mdio->md_mediasize % PAGE_SIZE) != 0) 1095 return (EDOM); 1096 oldpages = OFF_TO_IDX(round_page(sc->mediasize)); 1097 newpages = OFF_TO_IDX(round_page(mdio->md_mediasize)); 1098 if (newpages < oldpages) { 1099 VM_OBJECT_LOCK(sc->object); 1100 vm_object_page_remove(sc->object, newpages, 0, 0); 1101 swap_pager_freespace(sc->object, newpages, 1102 oldpages - newpages); 1103 swap_release_by_cred(IDX_TO_OFF(oldpages - 1104 newpages), sc->cred); 1105 sc->object->charge = IDX_TO_OFF(newpages); 1106 sc->object->size = newpages; 1107 VM_OBJECT_UNLOCK(sc->object); 1108 } else if (newpages > oldpages) { 1109 res = swap_reserve_by_cred(IDX_TO_OFF(newpages - 1110 oldpages), sc->cred); 1111 if (!res) 1112 return (ENOMEM); 1113 if ((mdio->md_options & MD_RESERVE) || 1114 (sc->flags & MD_RESERVE)) { 1115 error = swap_pager_reserve(sc->object, 1116 oldpages, newpages - oldpages); 1117 if (error < 0) { 1118 swap_release_by_cred( 1119 IDX_TO_OFF(newpages - oldpages), 1120 sc->cred); 1121 return (EDOM); 1122 } 1123 } 1124 VM_OBJECT_LOCK(sc->object); 1125 sc->object->charge = IDX_TO_OFF(newpages); 1126 sc->object->size = newpages; 1127 VM_OBJECT_UNLOCK(sc->object); 1128 } 1129 break; 1130 default: 1131 return (EOPNOTSUPP); 1132 } 1133 1134 sc->mediasize = mdio->md_mediasize; 1135 g_topology_lock(); 1136 g_resize_provider(sc->pp, sc->mediasize); 1137 g_topology_unlock(); 1138 return (0); 1139} 1140 1141static int 1142mdcreate_swap(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 1143{ 1144 vm_ooffset_t npage; 1145 int error; 1146 1147 /* 1148 * Range check. Disallow negative sizes or any size less then the 1149 * size of a page. Then round to a page. 1150 */ 1151 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0) 1152 return (EDOM); 1153 1154 /* 1155 * Allocate an OBJT_SWAP object. 1156 * 1157 * Note the truncation. 1158 */ 1159 1160 npage = mdio->md_mediasize / PAGE_SIZE; 1161 if (mdio->md_fwsectors != 0) 1162 sc->fwsectors = mdio->md_fwsectors; 1163 if (mdio->md_fwheads != 0) 1164 sc->fwheads = mdio->md_fwheads; 1165 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage, 1166 VM_PROT_DEFAULT, 0, td->td_ucred); 1167 if (sc->object == NULL) 1168 return (ENOMEM); 1169 sc->flags = mdio->md_options & (MD_FORCE | MD_RESERVE); 1170 if (mdio->md_options & MD_RESERVE) { 1171 if (swap_pager_reserve(sc->object, 0, npage) < 0) { 1172 error = EDOM; 1173 goto finish; 1174 } 1175 } 1176 error = mdsetcred(sc, td->td_ucred); 1177 finish: 1178 if (error != 0) { 1179 vm_object_deallocate(sc->object); 1180 sc->object = NULL; 1181 } 1182 return (error); 1183} 1184 1185 1186static int 1187xmdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1188{ 1189 struct md_ioctl *mdio; 1190 struct md_s *sc; 1191 int error, i; 1192 unsigned sectsize; 1193 1194 if (md_debug) 1195 printf("mdctlioctl(%s %lx %p %x %p)\n", 1196 devtoname(dev), cmd, addr, flags, td); 1197 1198 mdio = (struct md_ioctl *)addr; 1199 if (mdio->md_version != MDIOVERSION) 1200 return (EINVAL); 1201 1202 /* 1203 * We assert the version number in the individual ioctl 1204 * handlers instead of out here because (a) it is possible we 1205 * may add another ioctl in the future which doesn't read an 1206 * mdio, and (b) the correct return value for an unknown ioctl 1207 * is ENOIOCTL, not EINVAL. 1208 */ 1209 error = 0; 1210 switch (cmd) { 1211 case MDIOCATTACH: 1212 switch (mdio->md_type) { 1213 case MD_MALLOC: 1214 case MD_PRELOAD: 1215 case MD_VNODE: 1216 case MD_SWAP: 1217 break; 1218 default: 1219 return (EINVAL); 1220 } 1221 if (mdio->md_sectorsize == 0) 1222 sectsize = DEV_BSIZE; 1223 else 1224 sectsize = mdio->md_sectorsize; 1225 if (sectsize > MAXPHYS || mdio->md_mediasize < sectsize) 1226 return (EINVAL); 1227 if (mdio->md_options & MD_AUTOUNIT) 1228 sc = mdnew(-1, &error, mdio->md_type); 1229 else { 1230 if (mdio->md_unit > INT_MAX) 1231 return (EINVAL); 1232 sc = mdnew(mdio->md_unit, &error, mdio->md_type); 1233 } 1234 if (sc == NULL) 1235 return (error); 1236 if (mdio->md_options & MD_AUTOUNIT) 1237 mdio->md_unit = sc->unit; 1238 sc->mediasize = mdio->md_mediasize; 1239 sc->sectorsize = sectsize; 1240 error = EDOOFUS; 1241 switch (sc->type) { 1242 case MD_MALLOC: 1243 sc->start = mdstart_malloc; 1244 error = mdcreate_malloc(sc, mdio); 1245 break; 1246 case MD_PRELOAD: 1247 sc->start = mdstart_preload; 1248 error = mdcreate_preload(sc, mdio); 1249 break; 1250 case MD_VNODE: 1251 sc->start = mdstart_vnode; 1252 error = mdcreate_vnode(sc, mdio, td); 1253 break; 1254 case MD_SWAP: 1255 sc->start = mdstart_swap; 1256 error = mdcreate_swap(sc, mdio, td); 1257 break; 1258 } 1259 if (error != 0) { 1260 mddestroy(sc, td); 1261 return (error); 1262 } 1263 1264 /* Prune off any residual fractional sector */ 1265 i = sc->mediasize % sc->sectorsize; 1266 sc->mediasize -= i; 1267 1268 mdinit(sc); 1269 return (0); 1270 case MDIOCDETACH: 1271 if (mdio->md_mediasize != 0 || 1272 (mdio->md_options & ~MD_FORCE) != 0) 1273 return (EINVAL); 1274 1275 sc = mdfind(mdio->md_unit); 1276 if (sc == NULL) 1277 return (ENOENT); 1278 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) && 1279 !(mdio->md_options & MD_FORCE)) 1280 return (EBUSY); 1281 return (mddestroy(sc, td)); 1282 case MDIOCRESIZE: 1283 if ((mdio->md_options & ~(MD_FORCE | MD_RESERVE)) != 0) 1284 return (EINVAL); 1285 1286 sc = mdfind(mdio->md_unit); 1287 if (sc == NULL) 1288 return (ENOENT); 1289 if (mdio->md_mediasize < sc->sectorsize) 1290 return (EINVAL); 1291 if (mdio->md_mediasize < sc->mediasize && 1292 !(sc->flags & MD_FORCE) && 1293 !(mdio->md_options & MD_FORCE)) 1294 return (EBUSY); 1295 return (mdresize(sc, mdio)); 1296 case MDIOCQUERY: 1297 sc = mdfind(mdio->md_unit); 1298 if (sc == NULL) 1299 return (ENOENT); 1300 mdio->md_type = sc->type; 1301 mdio->md_options = sc->flags; 1302 mdio->md_mediasize = sc->mediasize; 1303 mdio->md_sectorsize = sc->sectorsize; 1304 if (sc->type == MD_VNODE) 1305 error = copyout(sc->file, mdio->md_file, 1306 strlen(sc->file) + 1); 1307 return (error); 1308 case MDIOCLIST: 1309 i = 1; 1310 LIST_FOREACH(sc, &md_softc_list, list) { 1311 if (i == MDNPAD - 1) 1312 mdio->md_pad[i] = -1; 1313 else 1314 mdio->md_pad[i++] = sc->unit; 1315 } 1316 mdio->md_pad[0] = i - 1; 1317 return (0); 1318 default: 1319 return (ENOIOCTL); 1320 }; 1321} 1322 1323static int 1324mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1325{ 1326 int error; 1327 1328 sx_xlock(&md_sx); 1329 error = xmdctlioctl(dev, cmd, addr, flags, td); 1330 sx_xunlock(&md_sx); 1331 return (error); 1332} 1333 1334static void 1335md_preloaded(u_char *image, size_t length) 1336{ 1337 struct md_s *sc; 1338 int error; 1339 1340 sc = mdnew(-1, &error, MD_PRELOAD); 1341 if (sc == NULL) 1342 return; 1343 sc->mediasize = length; 1344 sc->sectorsize = DEV_BSIZE; 1345 sc->pl_ptr = image; 1346 sc->pl_len = length; 1347 sc->start = mdstart_preload; 1348#ifdef MD_ROOT 1349 if (sc->unit == 0) 1350 rootdevnames[0] = "ufs:/dev/md0"; 1351#endif 1352 mdinit(sc); 1353} 1354 1355static void 1356g_md_init(struct g_class *mp __unused) 1357{ 1358 caddr_t mod; 1359 u_char *ptr, *name, *type; 1360 unsigned len; 1361 int i; 1362 1363 /* figure out log2(NINDIR) */ 1364 for (i = NINDIR, nshift = -1; i; nshift++) 1365 i >>= 1; 1366 1367 mod = NULL; 1368 sx_init(&md_sx, "MD config lock"); 1369 g_topology_unlock(); 1370 md_uh = new_unrhdr(0, INT_MAX, NULL); 1371#ifdef MD_ROOT_SIZE 1372 sx_xlock(&md_sx); 1373 md_preloaded(mfs_root.start, sizeof(mfs_root.start)); 1374 sx_xunlock(&md_sx); 1375#endif 1376 /* XXX: are preload_* static or do they need Giant ? */ 1377 while ((mod = preload_search_next_name(mod)) != NULL) { 1378 name = (char *)preload_search_info(mod, MODINFO_NAME); 1379 if (name == NULL) 1380 continue; 1381 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1382 if (type == NULL) 1383 continue; 1384 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1385 continue; 1386 ptr = preload_fetch_addr(mod); 1387 len = preload_fetch_size(mod); 1388 if (ptr != NULL && len != 0) { 1389 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1390 MD_NAME, mdunits, name, len, ptr); 1391 sx_xlock(&md_sx); 1392 md_preloaded(ptr, len); 1393 sx_xunlock(&md_sx); 1394 } 1395 } 1396 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL, 1397 0600, MDCTL_NAME); 1398 g_topology_lock(); 1399} 1400 1401static void 1402g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 1403 struct g_consumer *cp __unused, struct g_provider *pp) 1404{ 1405 struct md_s *mp; 1406 char *type; 1407 1408 mp = gp->softc; 1409 if (mp == NULL) 1410 return; 1411 1412 switch (mp->type) { 1413 case MD_MALLOC: 1414 type = "malloc"; 1415 break; 1416 case MD_PRELOAD: 1417 type = "preload"; 1418 break; 1419 case MD_VNODE: 1420 type = "vnode"; 1421 break; 1422 case MD_SWAP: 1423 type = "swap"; 1424 break; 1425 default: 1426 type = "unknown"; 1427 break; 1428 } 1429 1430 if (pp != NULL) { 1431 if (indent == NULL) { 1432 sbuf_printf(sb, " u %d", mp->unit); 1433 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize); 1434 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads); 1435 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors); 1436 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize); 1437 sbuf_printf(sb, " t %s", type); 1438 if (mp->type == MD_VNODE && mp->vnode != NULL) 1439 sbuf_printf(sb, " file %s", mp->file); 1440 } else { 1441 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, 1442 mp->unit); 1443 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n", 1444 indent, (uintmax_t) mp->sectorsize); 1445 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n", 1446 indent, (uintmax_t) mp->fwheads); 1447 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n", 1448 indent, (uintmax_t) mp->fwsectors); 1449 sbuf_printf(sb, "%s<length>%ju</length>\n", 1450 indent, (uintmax_t) mp->mediasize); 1451 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent, 1452 (mp->flags & MD_COMPRESS) == 0 ? "off": "on"); 1453 sbuf_printf(sb, "%s<access>%s</access>\n", indent, 1454 (mp->flags & MD_READONLY) == 0 ? "read-write": 1455 "read-only"); 1456 sbuf_printf(sb, "%s<type>%s</type>\n", indent, 1457 type); 1458 if (mp->type == MD_VNODE && mp->vnode != NULL) 1459 sbuf_printf(sb, "%s<file>%s</file>\n", 1460 indent, mp->file); 1461 } 1462 } 1463} 1464 1465static void 1466g_md_fini(struct g_class *mp __unused) 1467{ 1468 1469 sx_destroy(&md_sx); 1470 if (status_dev != NULL) 1471 destroy_dev(status_dev); 1472 delete_unrhdr(md_uh); 1473} 1474