md.c revision 217583
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 217583 2011-01-19 16:48:07Z kib $ 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/sched.h> 79#include <sys/sf_buf.h> 80#include <sys/sysctl.h> 81#include <sys/vnode.h> 82 83#include <geom/geom.h> 84 85#include <vm/vm.h> 86#include <vm/vm_object.h> 87#include <vm/vm_page.h> 88#include <vm/vm_pager.h> 89#include <vm/swap_pager.h> 90#include <vm/uma.h> 91 92#define MD_MODVER 1 93 94#define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */ 95#define MD_EXITING 0x20000 /* Worker thread is exiting. */ 96 97#ifndef MD_NSECT 98#define MD_NSECT (10000 * 2) 99#endif 100 101static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk"); 102static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors"); 103 104static int md_debug; 105SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, ""); 106static int md_malloc_wait; 107SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0, ""); 108 109#if defined(MD_ROOT) && defined(MD_ROOT_SIZE) 110/* 111 * Preloaded image gets put here. 112 * Applications that patch the object with the image can determine 113 * the size looking at the start and end markers (strings), 114 * so we want them contiguous. 115 */ 116static struct { 117 u_char start[MD_ROOT_SIZE*1024]; 118 u_char end[128]; 119} mfs_root = { 120 .start = "MFS Filesystem goes here", 121 .end = "MFS Filesystem had better STOP here", 122}; 123#endif 124 125static g_init_t g_md_init; 126static g_fini_t g_md_fini; 127static g_start_t g_md_start; 128static g_access_t g_md_access; 129static void g_md_dumpconf(struct sbuf *sb, const char *indent, 130 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp); 131 132static int mdunits; 133static struct cdev *status_dev = 0; 134static struct sx md_sx; 135static struct unrhdr *md_uh; 136 137static d_ioctl_t mdctlioctl; 138 139static struct cdevsw mdctl_cdevsw = { 140 .d_version = D_VERSION, 141 .d_ioctl = mdctlioctl, 142 .d_name = MD_NAME, 143}; 144 145struct g_class g_md_class = { 146 .name = "MD", 147 .version = G_VERSION, 148 .init = g_md_init, 149 .fini = g_md_fini, 150 .start = g_md_start, 151 .access = g_md_access, 152 .dumpconf = g_md_dumpconf, 153}; 154 155DECLARE_GEOM_CLASS(g_md_class, g_md); 156 157 158static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list); 159 160#define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 161#define NMASK (NINDIR-1) 162static int nshift; 163 164struct indir { 165 uintptr_t *array; 166 u_int total; 167 u_int used; 168 u_int shift; 169}; 170 171struct md_s { 172 int unit; 173 LIST_ENTRY(md_s) list; 174 struct bio_queue_head bio_queue; 175 struct mtx queue_mtx; 176 struct cdev *dev; 177 enum md_types type; 178 off_t mediasize; 179 unsigned sectorsize; 180 unsigned opencount; 181 unsigned fwheads; 182 unsigned fwsectors; 183 unsigned flags; 184 char name[20]; 185 struct proc *procp; 186 struct g_geom *gp; 187 struct g_provider *pp; 188 int (*start)(struct md_s *sc, struct bio *bp); 189 struct devstat *devstat; 190 191 /* MD_MALLOC related fields */ 192 struct indir *indir; 193 uma_zone_t uma; 194 195 /* MD_PRELOAD related fields */ 196 u_char *pl_ptr; 197 size_t pl_len; 198 199 /* MD_VNODE related fields */ 200 struct vnode *vnode; 201 char file[PATH_MAX]; 202 struct ucred *cred; 203 204 /* MD_SWAP related fields */ 205 vm_object_t object; 206}; 207 208static struct indir * 209new_indir(u_int shift) 210{ 211 struct indir *ip; 212 213 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT) 214 | M_ZERO); 215 if (ip == NULL) 216 return (NULL); 217 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 218 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO); 219 if (ip->array == NULL) { 220 free(ip, M_MD); 221 return (NULL); 222 } 223 ip->total = NINDIR; 224 ip->shift = shift; 225 return (ip); 226} 227 228static void 229del_indir(struct indir *ip) 230{ 231 232 free(ip->array, M_MDSECT); 233 free(ip, M_MD); 234} 235 236static void 237destroy_indir(struct md_s *sc, struct indir *ip) 238{ 239 int i; 240 241 for (i = 0; i < NINDIR; i++) { 242 if (!ip->array[i]) 243 continue; 244 if (ip->shift) 245 destroy_indir(sc, (struct indir*)(ip->array[i])); 246 else if (ip->array[i] > 255) 247 uma_zfree(sc->uma, (void *)(ip->array[i])); 248 } 249 del_indir(ip); 250} 251 252/* 253 * This function does the math and allocates the top level "indir" structure 254 * for a device of "size" sectors. 255 */ 256 257static struct indir * 258dimension(off_t size) 259{ 260 off_t rcnt; 261 struct indir *ip; 262 int layer; 263 264 rcnt = size; 265 layer = 0; 266 while (rcnt > NINDIR) { 267 rcnt /= NINDIR; 268 layer++; 269 } 270 271 /* 272 * XXX: the top layer is probably not fully populated, so we allocate 273 * too much space for ip->array in here. 274 */ 275 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 276 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 277 M_MDSECT, M_WAITOK | M_ZERO); 278 ip->total = NINDIR; 279 ip->shift = layer * nshift; 280 return (ip); 281} 282 283/* 284 * Read a given sector 285 */ 286 287static uintptr_t 288s_read(struct indir *ip, off_t offset) 289{ 290 struct indir *cip; 291 int idx; 292 uintptr_t up; 293 294 if (md_debug > 1) 295 printf("s_read(%jd)\n", (intmax_t)offset); 296 up = 0; 297 for (cip = ip; cip != NULL;) { 298 if (cip->shift) { 299 idx = (offset >> cip->shift) & NMASK; 300 up = cip->array[idx]; 301 cip = (struct indir *)up; 302 continue; 303 } 304 idx = offset & NMASK; 305 return (cip->array[idx]); 306 } 307 return (0); 308} 309 310/* 311 * Write a given sector, prune the tree if the value is 0 312 */ 313 314static int 315s_write(struct indir *ip, off_t offset, uintptr_t ptr) 316{ 317 struct indir *cip, *lip[10]; 318 int idx, li; 319 uintptr_t up; 320 321 if (md_debug > 1) 322 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 323 up = 0; 324 li = 0; 325 cip = ip; 326 for (;;) { 327 lip[li++] = cip; 328 if (cip->shift) { 329 idx = (offset >> cip->shift) & NMASK; 330 up = cip->array[idx]; 331 if (up != 0) { 332 cip = (struct indir *)up; 333 continue; 334 } 335 /* Allocate branch */ 336 cip->array[idx] = 337 (uintptr_t)new_indir(cip->shift - nshift); 338 if (cip->array[idx] == 0) 339 return (ENOSPC); 340 cip->used++; 341 up = cip->array[idx]; 342 cip = (struct indir *)up; 343 continue; 344 } 345 /* leafnode */ 346 idx = offset & NMASK; 347 up = cip->array[idx]; 348 if (up != 0) 349 cip->used--; 350 cip->array[idx] = ptr; 351 if (ptr != 0) 352 cip->used++; 353 break; 354 } 355 if (cip->used != 0 || li == 1) 356 return (0); 357 li--; 358 while (cip->used == 0 && cip != ip) { 359 li--; 360 idx = (offset >> lip[li]->shift) & NMASK; 361 up = lip[li]->array[idx]; 362 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 363 del_indir(cip); 364 lip[li]->array[idx] = 0; 365 lip[li]->used--; 366 cip = lip[li]; 367 } 368 return (0); 369} 370 371 372static int 373g_md_access(struct g_provider *pp, int r, int w, int e) 374{ 375 struct md_s *sc; 376 377 sc = pp->geom->softc; 378 if (sc == NULL) { 379 if (r <= 0 && w <= 0 && e <= 0) 380 return (0); 381 return (ENXIO); 382 } 383 r += pp->acr; 384 w += pp->acw; 385 e += pp->ace; 386 if ((sc->flags & MD_READONLY) != 0 && w > 0) 387 return (EROFS); 388 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 389 sc->opencount = 1; 390 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 391 sc->opencount = 0; 392 } 393 return (0); 394} 395 396static void 397g_md_start(struct bio *bp) 398{ 399 struct md_s *sc; 400 401 sc = bp->bio_to->geom->softc; 402 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) 403 devstat_start_transaction_bio(sc->devstat, bp); 404 mtx_lock(&sc->queue_mtx); 405 bioq_disksort(&sc->bio_queue, bp); 406 mtx_unlock(&sc->queue_mtx); 407 wakeup(sc); 408} 409 410static int 411mdstart_malloc(struct md_s *sc, struct bio *bp) 412{ 413 int i, error; 414 u_char *dst; 415 off_t secno, nsec, uc; 416 uintptr_t sp, osp; 417 418 switch (bp->bio_cmd) { 419 case BIO_READ: 420 case BIO_WRITE: 421 case BIO_DELETE: 422 break; 423 default: 424 return (EOPNOTSUPP); 425 } 426 427 nsec = bp->bio_length / sc->sectorsize; 428 secno = bp->bio_offset / sc->sectorsize; 429 dst = bp->bio_data; 430 error = 0; 431 while (nsec--) { 432 osp = s_read(sc->indir, secno); 433 if (bp->bio_cmd == BIO_DELETE) { 434 if (osp != 0) 435 error = s_write(sc->indir, secno, 0); 436 } else if (bp->bio_cmd == BIO_READ) { 437 if (osp == 0) 438 bzero(dst, sc->sectorsize); 439 else if (osp <= 255) 440 memset(dst, osp, sc->sectorsize); 441 else { 442 bcopy((void *)osp, dst, sc->sectorsize); 443 cpu_flush_dcache(dst, sc->sectorsize); 444 } 445 osp = 0; 446 } else if (bp->bio_cmd == BIO_WRITE) { 447 if (sc->flags & MD_COMPRESS) { 448 uc = dst[0]; 449 for (i = 1; i < sc->sectorsize; i++) 450 if (dst[i] != uc) 451 break; 452 } else { 453 i = 0; 454 uc = 0; 455 } 456 if (i == sc->sectorsize) { 457 if (osp != uc) 458 error = s_write(sc->indir, secno, uc); 459 } else { 460 if (osp <= 255) { 461 sp = (uintptr_t)uma_zalloc(sc->uma, 462 md_malloc_wait ? M_WAITOK : 463 M_NOWAIT); 464 if (sp == 0) { 465 error = ENOSPC; 466 break; 467 } 468 bcopy(dst, (void *)sp, sc->sectorsize); 469 error = s_write(sc->indir, secno, sp); 470 } else { 471 bcopy(dst, (void *)osp, sc->sectorsize); 472 osp = 0; 473 } 474 } 475 } else { 476 error = EOPNOTSUPP; 477 } 478 if (osp > 255) 479 uma_zfree(sc->uma, (void*)osp); 480 if (error != 0) 481 break; 482 secno++; 483 dst += sc->sectorsize; 484 } 485 bp->bio_resid = 0; 486 return (error); 487} 488 489static int 490mdstart_preload(struct md_s *sc, struct bio *bp) 491{ 492 493 switch (bp->bio_cmd) { 494 case BIO_READ: 495 bcopy(sc->pl_ptr + bp->bio_offset, bp->bio_data, 496 bp->bio_length); 497 cpu_flush_dcache(bp->bio_data, bp->bio_length); 498 break; 499 case BIO_WRITE: 500 bcopy(bp->bio_data, sc->pl_ptr + bp->bio_offset, 501 bp->bio_length); 502 break; 503 } 504 bp->bio_resid = 0; 505 return (0); 506} 507 508static int 509mdstart_vnode(struct md_s *sc, struct bio *bp) 510{ 511 int error, vfslocked; 512 struct uio auio; 513 struct iovec aiov; 514 struct mount *mp; 515 struct vnode *vp; 516 struct thread *td; 517 518 switch (bp->bio_cmd) { 519 case BIO_READ: 520 case BIO_WRITE: 521 case BIO_FLUSH: 522 break; 523 default: 524 return (EOPNOTSUPP); 525 } 526 527 td = curthread; 528 vp = sc->vnode; 529 530 /* 531 * VNODE I/O 532 * 533 * If an error occurs, we set BIO_ERROR but we do not set 534 * B_INVAL because (for a write anyway), the buffer is 535 * still valid. 536 */ 537 538 if (bp->bio_cmd == BIO_FLUSH) { 539 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 540 (void) vn_start_write(vp, &mp, V_WAIT); 541 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 542 error = VOP_FSYNC(vp, MNT_WAIT, td); 543 VOP_UNLOCK(vp, 0); 544 vn_finished_write(mp); 545 VFS_UNLOCK_GIANT(vfslocked); 546 return (error); 547 } 548 549 bzero(&auio, sizeof(auio)); 550 551 aiov.iov_base = bp->bio_data; 552 aiov.iov_len = bp->bio_length; 553 auio.uio_iov = &aiov; 554 auio.uio_iovcnt = 1; 555 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 556 auio.uio_segflg = UIO_SYSSPACE; 557 if (bp->bio_cmd == BIO_READ) 558 auio.uio_rw = UIO_READ; 559 else if (bp->bio_cmd == BIO_WRITE) 560 auio.uio_rw = UIO_WRITE; 561 else 562 panic("wrong BIO_OP in mdstart_vnode"); 563 auio.uio_resid = bp->bio_length; 564 auio.uio_td = td; 565 /* 566 * When reading set IO_DIRECT to try to avoid double-caching 567 * the data. When writing IO_DIRECT is not optimal. 568 */ 569 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 570 if (bp->bio_cmd == BIO_READ) { 571 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 572 error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred); 573 VOP_UNLOCK(vp, 0); 574 } else { 575 (void) vn_start_write(vp, &mp, V_WAIT); 576 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 577 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC, 578 sc->cred); 579 VOP_UNLOCK(vp, 0); 580 vn_finished_write(mp); 581 } 582 VFS_UNLOCK_GIANT(vfslocked); 583 bp->bio_resid = auio.uio_resid; 584 return (error); 585} 586 587static int 588mdstart_swap(struct md_s *sc, struct bio *bp) 589{ 590 struct sf_buf *sf; 591 int rv, offs, len, lastend; 592 vm_pindex_t i, lastp; 593 vm_page_t m; 594 u_char *p; 595 596 switch (bp->bio_cmd) { 597 case BIO_READ: 598 case BIO_WRITE: 599 case BIO_DELETE: 600 break; 601 default: 602 return (EOPNOTSUPP); 603 } 604 605 p = bp->bio_data; 606 607 /* 608 * offs is the offset at which to start operating on the 609 * next (ie, first) page. lastp is the last page on 610 * which we're going to operate. lastend is the ending 611 * position within that last page (ie, PAGE_SIZE if 612 * we're operating on complete aligned pages). 613 */ 614 offs = bp->bio_offset % PAGE_SIZE; 615 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE; 616 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1; 617 618 rv = VM_PAGER_OK; 619 VM_OBJECT_LOCK(sc->object); 620 vm_object_pip_add(sc->object, 1); 621 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) { 622 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs; 623 624 m = vm_page_grab(sc->object, i, 625 VM_ALLOC_NORMAL|VM_ALLOC_RETRY); 626 VM_OBJECT_UNLOCK(sc->object); 627 sched_pin(); 628 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 629 VM_OBJECT_LOCK(sc->object); 630 if (bp->bio_cmd == BIO_READ) { 631 if (m->valid != VM_PAGE_BITS_ALL) 632 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 633 if (rv == VM_PAGER_ERROR) { 634 sf_buf_free(sf); 635 sched_unpin(); 636 vm_page_wakeup(m); 637 break; 638 } 639 bcopy((void *)(sf_buf_kva(sf) + offs), p, len); 640 cpu_flush_dcache(p, len); 641 } else if (bp->bio_cmd == BIO_WRITE) { 642 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 643 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 644 if (rv == VM_PAGER_ERROR) { 645 sf_buf_free(sf); 646 sched_unpin(); 647 vm_page_wakeup(m); 648 break; 649 } 650 bcopy(p, (void *)(sf_buf_kva(sf) + offs), len); 651 m->valid = VM_PAGE_BITS_ALL; 652#if 0 653 } else if (bp->bio_cmd == BIO_DELETE) { 654 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 655 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 656 if (rv == VM_PAGER_ERROR) { 657 sf_buf_free(sf); 658 sched_unpin(); 659 vm_page_wakeup(m); 660 break; 661 } 662 bzero((void *)(sf_buf_kva(sf) + offs), len); 663 vm_page_dirty(m); 664 m->valid = VM_PAGE_BITS_ALL; 665#endif 666 } 667 sf_buf_free(sf); 668 sched_unpin(); 669 vm_page_wakeup(m); 670 vm_page_lock(m); 671 vm_page_activate(m); 672 vm_page_unlock(m); 673 if (bp->bio_cmd == BIO_WRITE) 674 vm_page_dirty(m); 675 676 /* Actions on further pages start at offset 0 */ 677 p += PAGE_SIZE - offs; 678 offs = 0; 679#if 0 680if (bootverbose || bp->bio_offset / PAGE_SIZE < 17) 681printf("wire_count %d busy %d flags %x hold_count %d act_count %d queue %d valid %d dirty %d @ %d\n", 682 m->wire_count, m->busy, 683 m->flags, m->hold_count, m->act_count, m->queue, m->valid, m->dirty, i); 684#endif 685 } 686 vm_object_pip_subtract(sc->object, 1); 687 VM_OBJECT_UNLOCK(sc->object); 688 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC); 689} 690 691static void 692md_kthread(void *arg) 693{ 694 struct md_s *sc; 695 struct bio *bp; 696 int error; 697 698 sc = arg; 699 thread_lock(curthread); 700 sched_prio(curthread, PRIBIO); 701 thread_unlock(curthread); 702 if (sc->type == MD_VNODE) 703 curthread->td_pflags |= TDP_NORUNNINGBUF; 704 705 for (;;) { 706 mtx_lock(&sc->queue_mtx); 707 if (sc->flags & MD_SHUTDOWN) { 708 sc->flags |= MD_EXITING; 709 mtx_unlock(&sc->queue_mtx); 710 kproc_exit(0); 711 } 712 bp = bioq_takefirst(&sc->bio_queue); 713 if (!bp) { 714 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 715 continue; 716 } 717 mtx_unlock(&sc->queue_mtx); 718 if (bp->bio_cmd == BIO_GETATTR) { 719 if ((sc->fwsectors && sc->fwheads && 720 (g_handleattr_int(bp, "GEOM::fwsectors", 721 sc->fwsectors) || 722 g_handleattr_int(bp, "GEOM::fwheads", 723 sc->fwheads))) || 724 g_handleattr_int(bp, "GEOM::candelete", 1)) 725 error = -1; 726 else 727 error = EOPNOTSUPP; 728 } else { 729 error = sc->start(sc, bp); 730 } 731 732 if (error != -1) { 733 bp->bio_completed = bp->bio_length; 734 g_io_deliver(bp, error); 735 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) 736 devstat_end_transaction_bio(sc->devstat, bp); 737 } 738 } 739} 740 741static struct md_s * 742mdfind(int unit) 743{ 744 struct md_s *sc; 745 746 LIST_FOREACH(sc, &md_softc_list, list) { 747 if (sc->unit == unit) 748 break; 749 } 750 return (sc); 751} 752 753static struct md_s * 754mdnew(int unit, int *errp, enum md_types type) 755{ 756 struct md_s *sc; 757 int error; 758 759 *errp = 0; 760 if (unit == -1) 761 unit = alloc_unr(md_uh); 762 else 763 unit = alloc_unr_specific(md_uh, unit); 764 765 if (unit == -1) { 766 *errp = EBUSY; 767 return (NULL); 768 } 769 770 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 771 sc->type = type; 772 bioq_init(&sc->bio_queue); 773 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 774 sc->unit = unit; 775 sprintf(sc->name, "md%d", unit); 776 LIST_INSERT_HEAD(&md_softc_list, sc, list); 777 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 778 if (error == 0) 779 return (sc); 780 LIST_REMOVE(sc, list); 781 mtx_destroy(&sc->queue_mtx); 782 free_unr(md_uh, sc->unit); 783 free(sc, M_MD); 784 *errp = error; 785 return (NULL); 786} 787 788static void 789mdinit(struct md_s *sc) 790{ 791 struct g_geom *gp; 792 struct g_provider *pp; 793 794 g_topology_lock(); 795 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 796 gp->softc = sc; 797 pp = g_new_providerf(gp, "md%d", sc->unit); 798 pp->mediasize = sc->mediasize; 799 pp->sectorsize = sc->sectorsize; 800 sc->gp = gp; 801 sc->pp = pp; 802 g_error_provider(pp, 0); 803 g_topology_unlock(); 804 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize, 805 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX); 806} 807 808/* 809 * XXX: we should check that the range they feed us is mapped. 810 * XXX: we should implement read-only. 811 */ 812 813static int 814mdcreate_preload(struct md_s *sc, struct md_ioctl *mdio) 815{ 816 817 if (mdio->md_options & ~(MD_AUTOUNIT | MD_FORCE)) 818 return (EINVAL); 819 if (mdio->md_base == 0) 820 return (EINVAL); 821 sc->flags = mdio->md_options & MD_FORCE; 822 /* Cast to pointer size, then to pointer to avoid warning */ 823 sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base; 824 sc->pl_len = (size_t)sc->mediasize; 825 return (0); 826} 827 828 829static int 830mdcreate_malloc(struct md_s *sc, struct md_ioctl *mdio) 831{ 832 uintptr_t sp; 833 int error; 834 off_t u; 835 836 error = 0; 837 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 838 return (EINVAL); 839 if (mdio->md_sectorsize != 0 && !powerof2(mdio->md_sectorsize)) 840 return (EINVAL); 841 /* Compression doesn't make sense if we have reserved space */ 842 if (mdio->md_options & MD_RESERVE) 843 mdio->md_options &= ~MD_COMPRESS; 844 if (mdio->md_fwsectors != 0) 845 sc->fwsectors = mdio->md_fwsectors; 846 if (mdio->md_fwheads != 0) 847 sc->fwheads = mdio->md_fwheads; 848 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 849 sc->indir = dimension(sc->mediasize / sc->sectorsize); 850 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL, 851 0x1ff, 0); 852 if (mdio->md_options & MD_RESERVE) { 853 off_t nsectors; 854 855 nsectors = sc->mediasize / sc->sectorsize; 856 for (u = 0; u < nsectors; u++) { 857 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ? 858 M_WAITOK : M_NOWAIT) | M_ZERO); 859 if (sp != 0) 860 error = s_write(sc->indir, u, sp); 861 else 862 error = ENOMEM; 863 if (error != 0) 864 break; 865 } 866 } 867 return (error); 868} 869 870 871static int 872mdsetcred(struct md_s *sc, struct ucred *cred) 873{ 874 char *tmpbuf; 875 int error = 0; 876 877 /* 878 * Set credits in our softc 879 */ 880 881 if (sc->cred) 882 crfree(sc->cred); 883 sc->cred = crhold(cred); 884 885 /* 886 * Horrible kludge to establish credentials for NFS XXX. 887 */ 888 889 if (sc->vnode) { 890 struct uio auio; 891 struct iovec aiov; 892 893 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK); 894 bzero(&auio, sizeof(auio)); 895 896 aiov.iov_base = tmpbuf; 897 aiov.iov_len = sc->sectorsize; 898 auio.uio_iov = &aiov; 899 auio.uio_iovcnt = 1; 900 auio.uio_offset = 0; 901 auio.uio_rw = UIO_READ; 902 auio.uio_segflg = UIO_SYSSPACE; 903 auio.uio_resid = aiov.iov_len; 904 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 905 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 906 VOP_UNLOCK(sc->vnode, 0); 907 free(tmpbuf, M_TEMP); 908 } 909 return (error); 910} 911 912static int 913mdcreate_vnode(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 914{ 915 struct vattr vattr; 916 struct nameidata nd; 917 char *fname; 918 int error, flags, vfslocked; 919 920 /* 921 * Kernel-originated requests must have the filename appended 922 * to the mdio structure to protect against malicious software. 923 */ 924 fname = mdio->md_file; 925 if ((void *)fname != (void *)(mdio + 1)) { 926 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL); 927 if (error != 0) 928 return (error); 929 } else 930 strlcpy(sc->file, fname, sizeof(sc->file)); 931 932 /* 933 * If the user specified that this is a read only device, don't 934 * set the FWRITE mask before trying to open the backing store. 935 */ 936 flags = FREAD | ((mdio->md_options & MD_READONLY) ? 0 : FWRITE); 937 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, sc->file, td); 938 error = vn_open(&nd, &flags, 0, NULL); 939 if (error != 0) 940 return (error); 941 vfslocked = NDHASGIANT(&nd); 942 NDFREE(&nd, NDF_ONLY_PNBUF); 943 if (nd.ni_vp->v_type != VREG) { 944 error = EINVAL; 945 goto bad; 946 } 947 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred); 948 if (error != 0) 949 goto bad; 950 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) { 951 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY); 952 if (nd.ni_vp->v_iflag & VI_DOOMED) { 953 /* Forced unmount. */ 954 error = EBADF; 955 goto bad; 956 } 957 } 958 nd.ni_vp->v_vflag |= VV_MD; 959 VOP_UNLOCK(nd.ni_vp, 0); 960 961 if (mdio->md_fwsectors != 0) 962 sc->fwsectors = mdio->md_fwsectors; 963 if (mdio->md_fwheads != 0) 964 sc->fwheads = mdio->md_fwheads; 965 sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC); 966 if (!(flags & FWRITE)) 967 sc->flags |= MD_READONLY; 968 sc->vnode = nd.ni_vp; 969 970 error = mdsetcred(sc, td->td_ucred); 971 if (error != 0) { 972 sc->vnode = NULL; 973 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY); 974 nd.ni_vp->v_vflag &= ~VV_MD; 975 goto bad; 976 } 977 VFS_UNLOCK_GIANT(vfslocked); 978 return (0); 979bad: 980 VOP_UNLOCK(nd.ni_vp, 0); 981 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td); 982 VFS_UNLOCK_GIANT(vfslocked); 983 return (error); 984} 985 986static int 987mddestroy(struct md_s *sc, struct thread *td) 988{ 989 int vfslocked; 990 991 if (sc->gp) { 992 sc->gp->softc = NULL; 993 g_topology_lock(); 994 g_wither_geom(sc->gp, ENXIO); 995 g_topology_unlock(); 996 sc->gp = NULL; 997 sc->pp = NULL; 998 } 999 if (sc->devstat) { 1000 devstat_remove_entry(sc->devstat); 1001 sc->devstat = NULL; 1002 } 1003 mtx_lock(&sc->queue_mtx); 1004 sc->flags |= MD_SHUTDOWN; 1005 wakeup(sc); 1006 while (!(sc->flags & MD_EXITING)) 1007 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10); 1008 mtx_unlock(&sc->queue_mtx); 1009 mtx_destroy(&sc->queue_mtx); 1010 if (sc->vnode != NULL) { 1011 vfslocked = VFS_LOCK_GIANT(sc->vnode->v_mount); 1012 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 1013 sc->vnode->v_vflag &= ~VV_MD; 1014 VOP_UNLOCK(sc->vnode, 0); 1015 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 1016 FREAD : (FREAD|FWRITE), sc->cred, td); 1017 VFS_UNLOCK_GIANT(vfslocked); 1018 } 1019 if (sc->cred != NULL) 1020 crfree(sc->cred); 1021 if (sc->object != NULL) 1022 vm_object_deallocate(sc->object); 1023 if (sc->indir) 1024 destroy_indir(sc, sc->indir); 1025 if (sc->uma) 1026 uma_zdestroy(sc->uma); 1027 1028 LIST_REMOVE(sc, list); 1029 free_unr(md_uh, sc->unit); 1030 free(sc, M_MD); 1031 return (0); 1032} 1033 1034static int 1035mdcreate_swap(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 1036{ 1037 vm_ooffset_t npage; 1038 int error; 1039 1040 /* 1041 * Range check. Disallow negative sizes or any size less then the 1042 * size of a page. Then round to a page. 1043 */ 1044 if (sc->mediasize == 0 || (sc->mediasize % PAGE_SIZE) != 0) 1045 return (EDOM); 1046 1047 /* 1048 * Allocate an OBJT_SWAP object. 1049 * 1050 * Note the truncation. 1051 */ 1052 1053 npage = mdio->md_mediasize / PAGE_SIZE; 1054 if (mdio->md_fwsectors != 0) 1055 sc->fwsectors = mdio->md_fwsectors; 1056 if (mdio->md_fwheads != 0) 1057 sc->fwheads = mdio->md_fwheads; 1058 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage, 1059 VM_PROT_DEFAULT, 0, td->td_ucred); 1060 if (sc->object == NULL) 1061 return (ENOMEM); 1062 sc->flags = mdio->md_options & MD_FORCE; 1063 if (mdio->md_options & MD_RESERVE) { 1064 if (swap_pager_reserve(sc->object, 0, npage) < 0) { 1065 error = EDOM; 1066 goto finish; 1067 } 1068 } 1069 error = mdsetcred(sc, td->td_ucred); 1070 finish: 1071 if (error != 0) { 1072 vm_object_deallocate(sc->object); 1073 sc->object = NULL; 1074 } 1075 return (error); 1076} 1077 1078 1079static int 1080xmdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1081{ 1082 struct md_ioctl *mdio; 1083 struct md_s *sc; 1084 int error, i; 1085 1086 if (md_debug) 1087 printf("mdctlioctl(%s %lx %p %x %p)\n", 1088 devtoname(dev), cmd, addr, flags, td); 1089 1090 mdio = (struct md_ioctl *)addr; 1091 if (mdio->md_version != MDIOVERSION) 1092 return (EINVAL); 1093 1094 /* 1095 * We assert the version number in the individual ioctl 1096 * handlers instead of out here because (a) it is possible we 1097 * may add another ioctl in the future which doesn't read an 1098 * mdio, and (b) the correct return value for an unknown ioctl 1099 * is ENOIOCTL, not EINVAL. 1100 */ 1101 error = 0; 1102 switch (cmd) { 1103 case MDIOCATTACH: 1104 switch (mdio->md_type) { 1105 case MD_MALLOC: 1106 case MD_PRELOAD: 1107 case MD_VNODE: 1108 case MD_SWAP: 1109 break; 1110 default: 1111 return (EINVAL); 1112 } 1113 if (mdio->md_options & MD_AUTOUNIT) 1114 sc = mdnew(-1, &error, mdio->md_type); 1115 else { 1116 if (mdio->md_unit > INT_MAX) 1117 return (EINVAL); 1118 sc = mdnew(mdio->md_unit, &error, mdio->md_type); 1119 } 1120 if (sc == NULL) 1121 return (error); 1122 if (mdio->md_options & MD_AUTOUNIT) 1123 mdio->md_unit = sc->unit; 1124 sc->mediasize = mdio->md_mediasize; 1125 if (mdio->md_sectorsize == 0) 1126 sc->sectorsize = DEV_BSIZE; 1127 else 1128 sc->sectorsize = mdio->md_sectorsize; 1129 error = EDOOFUS; 1130 switch (sc->type) { 1131 case MD_MALLOC: 1132 sc->start = mdstart_malloc; 1133 error = mdcreate_malloc(sc, mdio); 1134 break; 1135 case MD_PRELOAD: 1136 sc->start = mdstart_preload; 1137 error = mdcreate_preload(sc, mdio); 1138 break; 1139 case MD_VNODE: 1140 sc->start = mdstart_vnode; 1141 error = mdcreate_vnode(sc, mdio, td); 1142 break; 1143 case MD_SWAP: 1144 sc->start = mdstart_swap; 1145 error = mdcreate_swap(sc, mdio, td); 1146 break; 1147 } 1148 if (error != 0) { 1149 mddestroy(sc, td); 1150 return (error); 1151 } 1152 1153 /* Prune off any residual fractional sector */ 1154 i = sc->mediasize % sc->sectorsize; 1155 sc->mediasize -= i; 1156 1157 mdinit(sc); 1158 return (0); 1159 case MDIOCDETACH: 1160 if (mdio->md_mediasize != 0 || 1161 (mdio->md_options & ~MD_FORCE) != 0) 1162 return (EINVAL); 1163 1164 sc = mdfind(mdio->md_unit); 1165 if (sc == NULL) 1166 return (ENOENT); 1167 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) && 1168 !(mdio->md_options & MD_FORCE)) 1169 return (EBUSY); 1170 return (mddestroy(sc, td)); 1171 case MDIOCQUERY: 1172 sc = mdfind(mdio->md_unit); 1173 if (sc == NULL) 1174 return (ENOENT); 1175 mdio->md_type = sc->type; 1176 mdio->md_options = sc->flags; 1177 mdio->md_mediasize = sc->mediasize; 1178 mdio->md_sectorsize = sc->sectorsize; 1179 if (sc->type == MD_VNODE) 1180 error = copyout(sc->file, mdio->md_file, 1181 strlen(sc->file) + 1); 1182 return (error); 1183 case MDIOCLIST: 1184 i = 1; 1185 LIST_FOREACH(sc, &md_softc_list, list) { 1186 if (i == MDNPAD - 1) 1187 mdio->md_pad[i] = -1; 1188 else 1189 mdio->md_pad[i++] = sc->unit; 1190 } 1191 mdio->md_pad[0] = i - 1; 1192 return (0); 1193 default: 1194 return (ENOIOCTL); 1195 }; 1196} 1197 1198static int 1199mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1200{ 1201 int error; 1202 1203 sx_xlock(&md_sx); 1204 error = xmdctlioctl(dev, cmd, addr, flags, td); 1205 sx_xunlock(&md_sx); 1206 return (error); 1207} 1208 1209static void 1210md_preloaded(u_char *image, size_t length) 1211{ 1212 struct md_s *sc; 1213 int error; 1214 1215 sc = mdnew(-1, &error, MD_PRELOAD); 1216 if (sc == NULL) 1217 return; 1218 sc->mediasize = length; 1219 sc->sectorsize = DEV_BSIZE; 1220 sc->pl_ptr = image; 1221 sc->pl_len = length; 1222 sc->start = mdstart_preload; 1223#ifdef MD_ROOT 1224 if (sc->unit == 0) 1225 rootdevnames[0] = "ufs:/dev/md0"; 1226#endif 1227 mdinit(sc); 1228} 1229 1230static void 1231g_md_init(struct g_class *mp __unused) 1232{ 1233 caddr_t mod; 1234 caddr_t c; 1235 u_char *ptr, *name, *type; 1236 unsigned len; 1237 int i; 1238 1239 /* figure out log2(NINDIR) */ 1240 for (i = NINDIR, nshift = -1; i; nshift++) 1241 i >>= 1; 1242 1243 mod = NULL; 1244 sx_init(&md_sx, "MD config lock"); 1245 g_topology_unlock(); 1246 md_uh = new_unrhdr(0, INT_MAX, NULL); 1247#ifdef MD_ROOT_SIZE 1248 sx_xlock(&md_sx); 1249 md_preloaded(mfs_root.start, sizeof(mfs_root.start)); 1250 sx_xunlock(&md_sx); 1251#endif 1252 /* XXX: are preload_* static or do they need Giant ? */ 1253 while ((mod = preload_search_next_name(mod)) != NULL) { 1254 name = (char *)preload_search_info(mod, MODINFO_NAME); 1255 if (name == NULL) 1256 continue; 1257 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1258 if (type == NULL) 1259 continue; 1260 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1261 continue; 1262 c = preload_search_info(mod, MODINFO_ADDR); 1263 ptr = *(u_char **)c; 1264 c = preload_search_info(mod, MODINFO_SIZE); 1265 len = *(size_t *)c; 1266 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1267 MD_NAME, mdunits, name, len, ptr); 1268 sx_xlock(&md_sx); 1269 md_preloaded(ptr, len); 1270 sx_xunlock(&md_sx); 1271 } 1272 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL, 1273 0600, MDCTL_NAME); 1274 g_topology_lock(); 1275} 1276 1277static void 1278g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 1279 struct g_consumer *cp __unused, struct g_provider *pp) 1280{ 1281 struct md_s *mp; 1282 char *type; 1283 1284 mp = gp->softc; 1285 if (mp == NULL) 1286 return; 1287 1288 switch (mp->type) { 1289 case MD_MALLOC: 1290 type = "malloc"; 1291 break; 1292 case MD_PRELOAD: 1293 type = "preload"; 1294 break; 1295 case MD_VNODE: 1296 type = "vnode"; 1297 break; 1298 case MD_SWAP: 1299 type = "swap"; 1300 break; 1301 default: 1302 type = "unknown"; 1303 break; 1304 } 1305 1306 if (pp != NULL) { 1307 if (indent == NULL) { 1308 sbuf_printf(sb, " u %d", mp->unit); 1309 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize); 1310 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads); 1311 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors); 1312 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize); 1313 sbuf_printf(sb, " t %s", type); 1314 if (mp->type == MD_VNODE && mp->vnode != NULL) 1315 sbuf_printf(sb, " file %s", mp->file); 1316 } else { 1317 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, 1318 mp->unit); 1319 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n", 1320 indent, (uintmax_t) mp->sectorsize); 1321 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n", 1322 indent, (uintmax_t) mp->fwheads); 1323 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n", 1324 indent, (uintmax_t) mp->fwsectors); 1325 sbuf_printf(sb, "%s<length>%ju</length>\n", 1326 indent, (uintmax_t) mp->mediasize); 1327 sbuf_printf(sb, "%s<type>%s</type>\n", indent, 1328 type); 1329 if (mp->type == MD_VNODE && mp->vnode != NULL) 1330 sbuf_printf(sb, "%s<file>%s</file>\n", 1331 indent, mp->file); 1332 } 1333 } 1334} 1335 1336static void 1337g_md_fini(struct g_class *mp __unused) 1338{ 1339 1340 sx_destroy(&md_sx); 1341 if (status_dev != NULL) 1342 destroy_dev(status_dev); 1343 delete_unrhdr(md_uh); 1344} 1345