md.c revision 175294
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 175294 2008-01-13 14:44:15Z attilio $ 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/linker.h> 68#include <sys/lock.h> 69#include <sys/malloc.h> 70#include <sys/mdioctl.h> 71#include <sys/mount.h> 72#include <sys/mutex.h> 73#include <sys/sx.h> 74#include <sys/namei.h> 75#include <sys/proc.h> 76#include <sys/queue.h> 77#include <sys/sched.h> 78#include <sys/sf_buf.h> 79#include <sys/sysctl.h> 80#include <sys/vnode.h> 81 82#include <geom/geom.h> 83 84#include <vm/vm.h> 85#include <vm/vm_object.h> 86#include <vm/vm_page.h> 87#include <vm/vm_pager.h> 88#include <vm/swap_pager.h> 89#include <vm/uma.h> 90 91#define MD_MODVER 1 92 93#define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */ 94#define MD_EXITING 0x20000 /* Worker thread is exiting. */ 95 96#ifndef MD_NSECT 97#define MD_NSECT (10000 * 2) 98#endif 99 100static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk"); 101static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors"); 102 103static int md_debug; 104SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, ""); 105 106#if defined(MD_ROOT) && defined(MD_ROOT_SIZE) 107/* 108 * Preloaded image gets put here. 109 * Applications that patch the object with the image can determine 110 * the size looking at the start and end markers (strings), 111 * so we want them contiguous. 112 */ 113static struct { 114 u_char start[MD_ROOT_SIZE*1024]; 115 u_char end[128]; 116} mfs_root = { 117 .start = "MFS Filesystem goes here", 118 .end = "MFS Filesystem had better STOP here", 119}; 120#endif 121 122static g_init_t g_md_init; 123static g_fini_t g_md_fini; 124static g_start_t g_md_start; 125static g_access_t g_md_access; 126static void g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 127 struct g_consumer *cp __unused, struct g_provider *pp); 128 129static int mdunits; 130static struct cdev *status_dev = 0; 131static struct sx md_sx; 132 133static d_ioctl_t mdctlioctl; 134 135static struct cdevsw mdctl_cdevsw = { 136 .d_version = D_VERSION, 137 .d_ioctl = mdctlioctl, 138 .d_name = MD_NAME, 139}; 140 141struct g_class g_md_class = { 142 .name = "MD", 143 .version = G_VERSION, 144 .init = g_md_init, 145 .fini = g_md_fini, 146 .start = g_md_start, 147 .access = g_md_access, 148 .dumpconf = g_md_dumpconf, 149}; 150 151DECLARE_GEOM_CLASS(g_md_class, g_md); 152 153 154static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(&md_softc_list); 155 156#define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 157#define NMASK (NINDIR-1) 158static int nshift; 159 160struct indir { 161 uintptr_t *array; 162 u_int total; 163 u_int used; 164 u_int shift; 165}; 166 167struct md_s { 168 int unit; 169 LIST_ENTRY(md_s) list; 170 struct bio_queue_head bio_queue; 171 struct mtx queue_mtx; 172 struct cdev *dev; 173 enum md_types type; 174 off_t mediasize; 175 unsigned sectorsize; 176 unsigned opencount; 177 unsigned fwheads; 178 unsigned fwsectors; 179 unsigned flags; 180 char name[20]; 181 struct proc *procp; 182 struct g_geom *gp; 183 struct g_provider *pp; 184 int (*start)(struct md_s *sc, struct bio *bp); 185 struct devstat *devstat; 186 187 /* MD_MALLOC related fields */ 188 struct indir *indir; 189 uma_zone_t uma; 190 191 /* MD_PRELOAD related fields */ 192 u_char *pl_ptr; 193 size_t pl_len; 194 195 /* MD_VNODE related fields */ 196 struct vnode *vnode; 197 char file[PATH_MAX]; 198 struct ucred *cred; 199 200 /* MD_SWAP related fields */ 201 vm_object_t object; 202}; 203 204static struct indir * 205new_indir(u_int shift) 206{ 207 struct indir *ip; 208 209 ip = malloc(sizeof *ip, M_MD, M_NOWAIT | M_ZERO); 210 if (ip == NULL) 211 return (NULL); 212 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 213 M_MDSECT, M_NOWAIT | M_ZERO); 214 if (ip->array == NULL) { 215 free(ip, M_MD); 216 return (NULL); 217 } 218 ip->total = NINDIR; 219 ip->shift = shift; 220 return (ip); 221} 222 223static void 224del_indir(struct indir *ip) 225{ 226 227 free(ip->array, M_MDSECT); 228 free(ip, M_MD); 229} 230 231static void 232destroy_indir(struct md_s *sc, struct indir *ip) 233{ 234 int i; 235 236 for (i = 0; i < NINDIR; i++) { 237 if (!ip->array[i]) 238 continue; 239 if (ip->shift) 240 destroy_indir(sc, (struct indir*)(ip->array[i])); 241 else if (ip->array[i] > 255) 242 uma_zfree(sc->uma, (void *)(ip->array[i])); 243 } 244 del_indir(ip); 245} 246 247/* 248 * This function does the math and allocates the top level "indir" structure 249 * for a device of "size" sectors. 250 */ 251 252static struct indir * 253dimension(off_t size) 254{ 255 off_t rcnt; 256 struct indir *ip; 257 int i, layer; 258 259 rcnt = size; 260 layer = 0; 261 while (rcnt > NINDIR) { 262 rcnt /= NINDIR; 263 layer++; 264 } 265 /* figure out log2(NINDIR) */ 266 for (i = NINDIR, nshift = -1; i; nshift++) 267 i >>= 1; 268 269 /* 270 * XXX: the top layer is probably not fully populated, so we allocate 271 * too much space for ip->array in here. 272 */ 273 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 274 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 275 M_MDSECT, M_WAITOK | M_ZERO); 276 ip->total = NINDIR; 277 ip->shift = layer * nshift; 278 return (ip); 279} 280 281/* 282 * Read a given sector 283 */ 284 285static uintptr_t 286s_read(struct indir *ip, off_t offset) 287{ 288 struct indir *cip; 289 int idx; 290 uintptr_t up; 291 292 if (md_debug > 1) 293 printf("s_read(%jd)\n", (intmax_t)offset); 294 up = 0; 295 for (cip = ip; cip != NULL;) { 296 if (cip->shift) { 297 idx = (offset >> cip->shift) & NMASK; 298 up = cip->array[idx]; 299 cip = (struct indir *)up; 300 continue; 301 } 302 idx = offset & NMASK; 303 return (cip->array[idx]); 304 } 305 return (0); 306} 307 308/* 309 * Write a given sector, prune the tree if the value is 0 310 */ 311 312static int 313s_write(struct indir *ip, off_t offset, uintptr_t ptr) 314{ 315 struct indir *cip, *lip[10]; 316 int idx, li; 317 uintptr_t up; 318 319 if (md_debug > 1) 320 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 321 up = 0; 322 li = 0; 323 cip = ip; 324 for (;;) { 325 lip[li++] = cip; 326 if (cip->shift) { 327 idx = (offset >> cip->shift) & NMASK; 328 up = cip->array[idx]; 329 if (up != 0) { 330 cip = (struct indir *)up; 331 continue; 332 } 333 /* Allocate branch */ 334 cip->array[idx] = 335 (uintptr_t)new_indir(cip->shift - nshift); 336 if (cip->array[idx] == 0) 337 return (ENOSPC); 338 cip->used++; 339 up = cip->array[idx]; 340 cip = (struct indir *)up; 341 continue; 342 } 343 /* leafnode */ 344 idx = offset & NMASK; 345 up = cip->array[idx]; 346 if (up != 0) 347 cip->used--; 348 cip->array[idx] = ptr; 349 if (ptr != 0) 350 cip->used++; 351 break; 352 } 353 if (cip->used != 0 || li == 1) 354 return (0); 355 li--; 356 while (cip->used == 0 && cip != ip) { 357 li--; 358 idx = (offset >> lip[li]->shift) & NMASK; 359 up = lip[li]->array[idx]; 360 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 361 del_indir(cip); 362 lip[li]->array[idx] = 0; 363 lip[li]->used--; 364 cip = lip[li]; 365 } 366 return (0); 367} 368 369 370static int 371g_md_access(struct g_provider *pp, int r, int w, int e) 372{ 373 struct md_s *sc; 374 375 sc = pp->geom->softc; 376 if (sc == NULL) 377 return (ENXIO); 378 r += pp->acr; 379 w += pp->acw; 380 e += pp->ace; 381 if ((sc->flags & MD_READONLY) != 0 && w > 0) 382 return (EROFS); 383 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 384 sc->opencount = 1; 385 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 386 sc->opencount = 0; 387 } 388 return (0); 389} 390 391static void 392g_md_start(struct bio *bp) 393{ 394 struct md_s *sc; 395 396 sc = bp->bio_to->geom->softc; 397 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) 398 devstat_start_transaction_bio(sc->devstat, bp); 399 mtx_lock(&sc->queue_mtx); 400 bioq_disksort(&sc->bio_queue, bp); 401 mtx_unlock(&sc->queue_mtx); 402 wakeup(sc); 403} 404 405static int 406mdstart_malloc(struct md_s *sc, struct bio *bp) 407{ 408 int i, error; 409 u_char *dst; 410 off_t secno, nsec, uc; 411 uintptr_t sp, osp; 412 413 switch (bp->bio_cmd) { 414 case BIO_READ: 415 case BIO_WRITE: 416 case BIO_DELETE: 417 break; 418 default: 419 return (EOPNOTSUPP); 420 } 421 422 nsec = bp->bio_length / sc->sectorsize; 423 secno = bp->bio_offset / sc->sectorsize; 424 dst = bp->bio_data; 425 error = 0; 426 while (nsec--) { 427 osp = s_read(sc->indir, secno); 428 if (bp->bio_cmd == BIO_DELETE) { 429 if (osp != 0) 430 error = s_write(sc->indir, secno, 0); 431 } else if (bp->bio_cmd == BIO_READ) { 432 if (osp == 0) 433 bzero(dst, sc->sectorsize); 434 else if (osp <= 255) 435 for (i = 0; i < sc->sectorsize; i++) 436 dst[i] = osp; 437 else 438 bcopy((void *)osp, dst, sc->sectorsize); 439 osp = 0; 440 } else if (bp->bio_cmd == BIO_WRITE) { 441 if (sc->flags & MD_COMPRESS) { 442 uc = dst[0]; 443 for (i = 1; i < sc->sectorsize; i++) 444 if (dst[i] != uc) 445 break; 446 } else { 447 i = 0; 448 uc = 0; 449 } 450 if (i == sc->sectorsize) { 451 if (osp != uc) 452 error = s_write(sc->indir, secno, uc); 453 } else { 454 if (osp <= 255) { 455 sp = (uintptr_t)uma_zalloc(sc->uma, 456 M_NOWAIT); 457 if (sp == 0) { 458 error = ENOSPC; 459 break; 460 } 461 bcopy(dst, (void *)sp, sc->sectorsize); 462 error = s_write(sc->indir, secno, sp); 463 } else { 464 bcopy(dst, (void *)osp, sc->sectorsize); 465 osp = 0; 466 } 467 } 468 } else { 469 error = EOPNOTSUPP; 470 } 471 if (osp > 255) 472 uma_zfree(sc->uma, (void*)osp); 473 if (error != 0) 474 break; 475 secno++; 476 dst += sc->sectorsize; 477 } 478 bp->bio_resid = 0; 479 return (error); 480} 481 482static int 483mdstart_preload(struct md_s *sc, struct bio *bp) 484{ 485 486 switch (bp->bio_cmd) { 487 case BIO_READ: 488 bcopy(sc->pl_ptr + bp->bio_offset, bp->bio_data, 489 bp->bio_length); 490 break; 491 case BIO_WRITE: 492 bcopy(bp->bio_data, sc->pl_ptr + bp->bio_offset, 493 bp->bio_length); 494 break; 495 } 496 bp->bio_resid = 0; 497 return (0); 498} 499 500static int 501mdstart_vnode(struct md_s *sc, struct bio *bp) 502{ 503 int error, vfslocked; 504 struct uio auio; 505 struct iovec aiov; 506 struct mount *mp; 507 struct vnode *vp; 508 struct thread *td; 509 510 switch (bp->bio_cmd) { 511 case BIO_READ: 512 case BIO_WRITE: 513 case BIO_FLUSH: 514 break; 515 default: 516 return (EOPNOTSUPP); 517 } 518 519 td = curthread; 520 vp = sc->vnode; 521 522 /* 523 * VNODE I/O 524 * 525 * If an error occurs, we set BIO_ERROR but we do not set 526 * B_INVAL because (for a write anyway), the buffer is 527 * still valid. 528 */ 529 530 if (bp->bio_cmd == BIO_FLUSH) { 531 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 532 (void) vn_start_write(vp, &mp, V_WAIT); 533 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 534 error = VOP_FSYNC(vp, MNT_WAIT, td); 535 VOP_UNLOCK(vp, 0); 536 vn_finished_write(mp); 537 VFS_UNLOCK_GIANT(vfslocked); 538 return (error); 539 } 540 541 bzero(&auio, sizeof(auio)); 542 543 aiov.iov_base = bp->bio_data; 544 aiov.iov_len = bp->bio_length; 545 auio.uio_iov = &aiov; 546 auio.uio_iovcnt = 1; 547 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 548 auio.uio_segflg = UIO_SYSSPACE; 549 if (bp->bio_cmd == BIO_READ) 550 auio.uio_rw = UIO_READ; 551 else if (bp->bio_cmd == BIO_WRITE) 552 auio.uio_rw = UIO_WRITE; 553 else 554 panic("wrong BIO_OP in mdstart_vnode"); 555 auio.uio_resid = bp->bio_length; 556 auio.uio_td = td; 557 /* 558 * When reading set IO_DIRECT to try to avoid double-caching 559 * the data. When writing IO_DIRECT is not optimal. 560 */ 561 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 562 if (bp->bio_cmd == BIO_READ) { 563 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 564 error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred); 565 VOP_UNLOCK(vp, 0); 566 } else { 567 (void) vn_start_write(vp, &mp, V_WAIT); 568 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 569 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC, 570 sc->cred); 571 VOP_UNLOCK(vp, 0); 572 vn_finished_write(mp); 573 } 574 VFS_UNLOCK_GIANT(vfslocked); 575 bp->bio_resid = auio.uio_resid; 576 return (error); 577} 578 579static int 580mdstart_swap(struct md_s *sc, struct bio *bp) 581{ 582 struct sf_buf *sf; 583 int rv, offs, len, lastend; 584 vm_pindex_t i, lastp; 585 vm_page_t m; 586 u_char *p; 587 588 switch (bp->bio_cmd) { 589 case BIO_READ: 590 case BIO_WRITE: 591 case BIO_DELETE: 592 break; 593 default: 594 return (EOPNOTSUPP); 595 } 596 597 p = bp->bio_data; 598 599 /* 600 * offs is the offset at which to start operating on the 601 * next (ie, first) page. lastp is the last page on 602 * which we're going to operate. lastend is the ending 603 * position within that last page (ie, PAGE_SIZE if 604 * we're operating on complete aligned pages). 605 */ 606 offs = bp->bio_offset % PAGE_SIZE; 607 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE; 608 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1; 609 610 rv = VM_PAGER_OK; 611 VM_OBJECT_LOCK(sc->object); 612 vm_object_pip_add(sc->object, 1); 613 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) { 614 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs; 615 616 m = vm_page_grab(sc->object, i, 617 VM_ALLOC_NORMAL|VM_ALLOC_RETRY); 618 VM_OBJECT_UNLOCK(sc->object); 619 sched_pin(); 620 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 621 VM_OBJECT_LOCK(sc->object); 622 if (bp->bio_cmd == BIO_READ) { 623 if (m->valid != VM_PAGE_BITS_ALL) 624 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 625 if (rv == VM_PAGER_ERROR) { 626 sf_buf_free(sf); 627 sched_unpin(); 628 vm_page_lock_queues(); 629 vm_page_wakeup(m); 630 vm_page_unlock_queues(); 631 break; 632 } 633 bcopy((void *)(sf_buf_kva(sf) + offs), p, len); 634 } else if (bp->bio_cmd == BIO_WRITE) { 635 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 636 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 637 if (rv == VM_PAGER_ERROR) { 638 sf_buf_free(sf); 639 sched_unpin(); 640 vm_page_lock_queues(); 641 vm_page_wakeup(m); 642 vm_page_unlock_queues(); 643 break; 644 } 645 bcopy(p, (void *)(sf_buf_kva(sf) + offs), len); 646 m->valid = VM_PAGE_BITS_ALL; 647#if 0 648 } else if (bp->bio_cmd == BIO_DELETE) { 649 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 650 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 651 if (rv == VM_PAGER_ERROR) { 652 sf_buf_free(sf); 653 sched_unpin(); 654 vm_page_lock_queues(); 655 vm_page_wakeup(m); 656 vm_page_unlock_queues(); 657 break; 658 } 659 bzero((void *)(sf_buf_kva(sf) + offs), len); 660 vm_page_dirty(m); 661 m->valid = VM_PAGE_BITS_ALL; 662#endif 663 } 664 sf_buf_free(sf); 665 sched_unpin(); 666 vm_page_lock_queues(); 667 vm_page_wakeup(m); 668 vm_page_activate(m); 669 if (bp->bio_cmd == BIO_WRITE) 670 vm_page_dirty(m); 671 vm_page_unlock_queues(); 672 673 /* Actions on further pages start at offset 0 */ 674 p += PAGE_SIZE - offs; 675 offs = 0; 676#if 0 677if (bootverbose || bp->bio_offset / PAGE_SIZE < 17) 678printf("wire_count %d busy %d flags %x hold_count %d act_count %d queue %d valid %d dirty %d @ %d\n", 679 m->wire_count, m->busy, 680 m->flags, m->hold_count, m->act_count, m->queue, m->valid, m->dirty, i); 681#endif 682 } 683 vm_object_pip_subtract(sc->object, 1); 684 vm_object_set_writeable_dirty(sc->object); 685 VM_OBJECT_UNLOCK(sc->object); 686 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC); 687} 688 689static void 690md_kthread(void *arg) 691{ 692 struct md_s *sc; 693 struct bio *bp; 694 int error; 695 696 sc = arg; 697 thread_lock(curthread); 698 sched_prio(curthread, PRIBIO); 699 thread_unlock(curthread); 700 if (sc->type == MD_VNODE) 701 curthread->td_pflags |= TDP_NORUNNINGBUF; 702 703 for (;;) { 704 mtx_lock(&sc->queue_mtx); 705 if (sc->flags & MD_SHUTDOWN) { 706 sc->flags |= MD_EXITING; 707 mtx_unlock(&sc->queue_mtx); 708 kproc_exit(0); 709 } 710 bp = bioq_takefirst(&sc->bio_queue); 711 if (!bp) { 712 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 713 continue; 714 } 715 mtx_unlock(&sc->queue_mtx); 716 if (bp->bio_cmd == BIO_GETATTR) { 717 if (sc->fwsectors && sc->fwheads && 718 (g_handleattr_int(bp, "GEOM::fwsectors", 719 sc->fwsectors) || 720 g_handleattr_int(bp, "GEOM::fwheads", 721 sc->fwheads))) 722 error = -1; 723 else 724 error = EOPNOTSUPP; 725 } else { 726 error = sc->start(sc, bp); 727 } 728 729 if (error != -1) { 730 bp->bio_completed = bp->bio_length; 731 g_io_deliver(bp, error); 732 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) 733 devstat_end_transaction_bio(sc->devstat, bp); 734 } 735 } 736} 737 738static struct md_s * 739mdfind(int unit) 740{ 741 struct md_s *sc; 742 743 LIST_FOREACH(sc, &md_softc_list, list) { 744 if (sc->unit == unit) 745 break; 746 } 747 return (sc); 748} 749 750static struct md_s * 751mdnew(int unit, int *errp, enum md_types type) 752{ 753 struct md_s *sc, *sc2; 754 int error, max = -1; 755 756 *errp = 0; 757 LIST_FOREACH(sc2, &md_softc_list, list) { 758 if (unit == sc2->unit) { 759 *errp = EBUSY; 760 return (NULL); 761 } 762 if (unit == -1 && sc2->unit > max) 763 max = sc2->unit; 764 } 765 if (unit == -1) 766 unit = max + 1; 767 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 768 sc->type = type; 769 bioq_init(&sc->bio_queue); 770 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 771 sc->unit = unit; 772 sprintf(sc->name, "md%d", unit); 773 LIST_INSERT_HEAD(&md_softc_list, sc, list); 774 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 775 if (error == 0) 776 return (sc); 777 LIST_REMOVE(sc, list); 778 mtx_destroy(&sc->queue_mtx); 779 free(sc, M_MD); 780 *errp = error; 781 return (NULL); 782} 783 784static void 785mdinit(struct md_s *sc) 786{ 787 788 struct g_geom *gp; 789 struct g_provider *pp; 790 791 g_topology_lock(); 792 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 793 gp->softc = sc; 794 pp = g_new_providerf(gp, "md%d", sc->unit); 795 pp->mediasize = sc->mediasize; 796 pp->sectorsize = sc->sectorsize; 797 sc->gp = gp; 798 sc->pp = pp; 799 g_error_provider(pp, 0); 800 g_topology_unlock(); 801 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize, 802 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX); 803} 804 805/* 806 * XXX: we should check that the range they feed us is mapped. 807 * XXX: we should implement read-only. 808 */ 809 810static int 811mdcreate_preload(struct md_s *sc, struct md_ioctl *mdio) 812{ 813 814 if (mdio->md_options & ~(MD_AUTOUNIT | MD_FORCE)) 815 return (EINVAL); 816 sc->flags = mdio->md_options & MD_FORCE; 817 /* Cast to pointer size, then to pointer to avoid warning */ 818 sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base; 819 sc->pl_len = (size_t)sc->mediasize; 820 return (0); 821} 822 823 824static int 825mdcreate_malloc(struct md_s *sc, struct md_ioctl *mdio) 826{ 827 uintptr_t sp; 828 int error; 829 off_t u; 830 831 error = 0; 832 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 833 return (EINVAL); 834 if (mdio->md_sectorsize != 0 && !powerof2(mdio->md_sectorsize)) 835 return (EINVAL); 836 /* Compression doesn't make sense if we have reserved space */ 837 if (mdio->md_options & MD_RESERVE) 838 mdio->md_options &= ~MD_COMPRESS; 839 if (mdio->md_fwsectors != 0) 840 sc->fwsectors = mdio->md_fwsectors; 841 if (mdio->md_fwheads != 0) 842 sc->fwheads = mdio->md_fwheads; 843 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 844 sc->indir = dimension(sc->mediasize / sc->sectorsize); 845 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL, 846 0x1ff, 0); 847 if (mdio->md_options & MD_RESERVE) { 848 off_t nsectors; 849 850 nsectors = sc->mediasize / sc->sectorsize; 851 for (u = 0; u < nsectors; u++) { 852 sp = (uintptr_t)uma_zalloc(sc->uma, M_NOWAIT | M_ZERO); 853 if (sp != 0) 854 error = s_write(sc->indir, u, sp); 855 else 856 error = ENOMEM; 857 if (error != 0) 858 break; 859 } 860 } 861 return (error); 862} 863 864 865static int 866mdsetcred(struct md_s *sc, struct ucred *cred) 867{ 868 char *tmpbuf; 869 int error = 0; 870 871 /* 872 * Set credits in our softc 873 */ 874 875 if (sc->cred) 876 crfree(sc->cred); 877 sc->cred = crhold(cred); 878 879 /* 880 * Horrible kludge to establish credentials for NFS XXX. 881 */ 882 883 if (sc->vnode) { 884 struct uio auio; 885 struct iovec aiov; 886 887 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK); 888 bzero(&auio, sizeof(auio)); 889 890 aiov.iov_base = tmpbuf; 891 aiov.iov_len = sc->sectorsize; 892 auio.uio_iov = &aiov; 893 auio.uio_iovcnt = 1; 894 auio.uio_offset = 0; 895 auio.uio_rw = UIO_READ; 896 auio.uio_segflg = UIO_SYSSPACE; 897 auio.uio_resid = aiov.iov_len; 898 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 899 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 900 VOP_UNLOCK(sc->vnode, 0); 901 free(tmpbuf, M_TEMP); 902 } 903 return (error); 904} 905 906static int 907mdcreate_vnode(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 908{ 909 struct vattr vattr; 910 struct nameidata nd; 911 int error, flags, vfslocked; 912 913 error = copyinstr(mdio->md_file, sc->file, sizeof(sc->file), NULL); 914 if (error != 0) 915 return (error); 916 flags = FREAD|FWRITE; 917 /* 918 * If the user specified that this is a read only device, unset the 919 * FWRITE mask before trying to open the backing store. 920 */ 921 if ((mdio->md_options & MD_READONLY) != 0) 922 flags &= ~FWRITE; 923 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, sc->file, td); 924 error = vn_open(&nd, &flags, 0, NULL); 925 if (error != 0) 926 return (error); 927 vfslocked = NDHASGIANT(&nd); 928 NDFREE(&nd, NDF_ONLY_PNBUF); 929 if (nd.ni_vp->v_type != VREG || 930 (error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred, td))) { 931 VOP_UNLOCK(nd.ni_vp, 0); 932 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td); 933 VFS_UNLOCK_GIANT(vfslocked); 934 return (error ? error : EINVAL); 935 } 936 nd.ni_vp->v_vflag |= VV_MD; 937 VOP_UNLOCK(nd.ni_vp, 0); 938 939 if (mdio->md_fwsectors != 0) 940 sc->fwsectors = mdio->md_fwsectors; 941 if (mdio->md_fwheads != 0) 942 sc->fwheads = mdio->md_fwheads; 943 sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC); 944 if (!(flags & FWRITE)) 945 sc->flags |= MD_READONLY; 946 sc->vnode = nd.ni_vp; 947 948 error = mdsetcred(sc, td->td_ucred); 949 if (error != 0) { 950 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY); 951 nd.ni_vp->v_vflag &= ~VV_MD; 952 VOP_UNLOCK(nd.ni_vp, 0); 953 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td); 954 VFS_UNLOCK_GIANT(vfslocked); 955 return (error); 956 } 957 VFS_UNLOCK_GIANT(vfslocked); 958 return (0); 959} 960 961static int 962mddestroy(struct md_s *sc, struct thread *td) 963{ 964 int vfslocked; 965 966 if (sc->gp) { 967 sc->gp->softc = NULL; 968 g_topology_lock(); 969 g_wither_geom(sc->gp, ENXIO); 970 g_topology_unlock(); 971 sc->gp = NULL; 972 sc->pp = NULL; 973 } 974 if (sc->devstat) { 975 devstat_remove_entry(sc->devstat); 976 sc->devstat = NULL; 977 } 978 mtx_lock(&sc->queue_mtx); 979 sc->flags |= MD_SHUTDOWN; 980 wakeup(sc); 981 while (!(sc->flags & MD_EXITING)) 982 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10); 983 mtx_unlock(&sc->queue_mtx); 984 mtx_destroy(&sc->queue_mtx); 985 if (sc->vnode != NULL) { 986 vfslocked = VFS_LOCK_GIANT(sc->vnode->v_mount); 987 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 988 sc->vnode->v_vflag &= ~VV_MD; 989 VOP_UNLOCK(sc->vnode, 0); 990 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 991 FREAD : (FREAD|FWRITE), sc->cred, td); 992 VFS_UNLOCK_GIANT(vfslocked); 993 } 994 if (sc->cred != NULL) 995 crfree(sc->cred); 996 if (sc->object != NULL) 997 vm_object_deallocate(sc->object); 998 if (sc->indir) 999 destroy_indir(sc, sc->indir); 1000 if (sc->uma) 1001 uma_zdestroy(sc->uma); 1002 1003 LIST_REMOVE(sc, list); 1004 free(sc, M_MD); 1005 return (0); 1006} 1007 1008static int 1009mdcreate_swap(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 1010{ 1011 vm_ooffset_t npage; 1012 int error; 1013 1014 /* 1015 * Range check. Disallow negative sizes or any size less then the 1016 * size of a page. Then round to a page. 1017 */ 1018 if (sc->mediasize == 0 || (sc->mediasize % PAGE_SIZE) != 0) 1019 return (EDOM); 1020 1021 /* 1022 * Allocate an OBJT_SWAP object. 1023 * 1024 * Note the truncation. 1025 */ 1026 1027 npage = mdio->md_mediasize / PAGE_SIZE; 1028 if (mdio->md_fwsectors != 0) 1029 sc->fwsectors = mdio->md_fwsectors; 1030 if (mdio->md_fwheads != 0) 1031 sc->fwheads = mdio->md_fwheads; 1032 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage, 1033 VM_PROT_DEFAULT, 0); 1034 if (sc->object == NULL) 1035 return (ENOMEM); 1036 sc->flags = mdio->md_options & MD_FORCE; 1037 if (mdio->md_options & MD_RESERVE) { 1038 if (swap_pager_reserve(sc->object, 0, npage) < 0) { 1039 vm_object_deallocate(sc->object); 1040 sc->object = NULL; 1041 return (EDOM); 1042 } 1043 } 1044 error = mdsetcred(sc, td->td_ucred); 1045 if (error != 0) { 1046 vm_object_deallocate(sc->object); 1047 sc->object = NULL; 1048 } 1049 return (error); 1050} 1051 1052 1053static int 1054xmdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1055{ 1056 struct md_ioctl *mdio; 1057 struct md_s *sc; 1058 int error, i; 1059 1060 if (md_debug) 1061 printf("mdctlioctl(%s %lx %p %x %p)\n", 1062 devtoname(dev), cmd, addr, flags, td); 1063 1064 mdio = (struct md_ioctl *)addr; 1065 if (mdio->md_version != MDIOVERSION) 1066 return (EINVAL); 1067 1068 /* 1069 * We assert the version number in the individual ioctl 1070 * handlers instead of out here because (a) it is possible we 1071 * may add another ioctl in the future which doesn't read an 1072 * mdio, and (b) the correct return value for an unknown ioctl 1073 * is ENOIOCTL, not EINVAL. 1074 */ 1075 error = 0; 1076 switch (cmd) { 1077 case MDIOCATTACH: 1078 switch (mdio->md_type) { 1079 case MD_MALLOC: 1080 case MD_PRELOAD: 1081 case MD_VNODE: 1082 case MD_SWAP: 1083 break; 1084 default: 1085 return (EINVAL); 1086 } 1087 if (mdio->md_options & MD_AUTOUNIT) 1088 sc = mdnew(-1, &error, mdio->md_type); 1089 else 1090 sc = mdnew(mdio->md_unit, &error, mdio->md_type); 1091 if (sc == NULL) 1092 return (error); 1093 if (mdio->md_options & MD_AUTOUNIT) 1094 mdio->md_unit = sc->unit; 1095 sc->mediasize = mdio->md_mediasize; 1096 if (mdio->md_sectorsize == 0) 1097 sc->sectorsize = DEV_BSIZE; 1098 else 1099 sc->sectorsize = mdio->md_sectorsize; 1100 error = EDOOFUS; 1101 switch (sc->type) { 1102 case MD_MALLOC: 1103 sc->start = mdstart_malloc; 1104 error = mdcreate_malloc(sc, mdio); 1105 break; 1106 case MD_PRELOAD: 1107 sc->start = mdstart_preload; 1108 error = mdcreate_preload(sc, mdio); 1109 break; 1110 case MD_VNODE: 1111 sc->start = mdstart_vnode; 1112 error = mdcreate_vnode(sc, mdio, td); 1113 break; 1114 case MD_SWAP: 1115 sc->start = mdstart_swap; 1116 error = mdcreate_swap(sc, mdio, td); 1117 break; 1118 } 1119 if (error != 0) { 1120 mddestroy(sc, td); 1121 return (error); 1122 } 1123 1124 /* Prune off any residual fractional sector */ 1125 i = sc->mediasize % sc->sectorsize; 1126 sc->mediasize -= i; 1127 1128 mdinit(sc); 1129 return (0); 1130 case MDIOCDETACH: 1131 if (mdio->md_mediasize != 0 || mdio->md_options != 0) 1132 return (EINVAL); 1133 1134 sc = mdfind(mdio->md_unit); 1135 if (sc == NULL) 1136 return (ENOENT); 1137 if (sc->opencount != 0 && !(sc->flags & MD_FORCE)) 1138 return (EBUSY); 1139 return (mddestroy(sc, td)); 1140 case MDIOCQUERY: 1141 sc = mdfind(mdio->md_unit); 1142 if (sc == NULL) 1143 return (ENOENT); 1144 mdio->md_type = sc->type; 1145 mdio->md_options = sc->flags; 1146 mdio->md_mediasize = sc->mediasize; 1147 mdio->md_sectorsize = sc->sectorsize; 1148 if (sc->type == MD_VNODE) 1149 error = copyout(sc->file, mdio->md_file, 1150 strlen(sc->file) + 1); 1151 return (error); 1152 case MDIOCLIST: 1153 i = 1; 1154 LIST_FOREACH(sc, &md_softc_list, list) { 1155 if (i == MDNPAD - 1) 1156 mdio->md_pad[i] = -1; 1157 else 1158 mdio->md_pad[i++] = sc->unit; 1159 } 1160 mdio->md_pad[0] = i - 1; 1161 return (0); 1162 default: 1163 return (ENOIOCTL); 1164 }; 1165} 1166 1167static int 1168mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1169{ 1170 int error; 1171 1172 sx_xlock(&md_sx); 1173 error = xmdctlioctl(dev, cmd, addr, flags, td); 1174 sx_xunlock(&md_sx); 1175 return (error); 1176} 1177 1178static void 1179md_preloaded(u_char *image, size_t length) 1180{ 1181 struct md_s *sc; 1182 int error; 1183 1184 sc = mdnew(-1, &error, MD_PRELOAD); 1185 if (sc == NULL) 1186 return; 1187 sc->mediasize = length; 1188 sc->sectorsize = DEV_BSIZE; 1189 sc->pl_ptr = image; 1190 sc->pl_len = length; 1191 sc->start = mdstart_preload; 1192#ifdef MD_ROOT 1193 if (sc->unit == 0) 1194 rootdevnames[0] = "ufs:/dev/md0"; 1195#endif 1196 mdinit(sc); 1197} 1198 1199static void 1200g_md_init(struct g_class *mp __unused) 1201{ 1202 1203 caddr_t mod; 1204 caddr_t c; 1205 u_char *ptr, *name, *type; 1206 unsigned len; 1207 1208 mod = NULL; 1209 sx_init(&md_sx, "MD config lock"); 1210 g_topology_unlock(); 1211#ifdef MD_ROOT_SIZE 1212 sx_xlock(&md_sx); 1213 md_preloaded(mfs_root.start, sizeof(mfs_root.start)); 1214 sx_xunlock(&md_sx); 1215#endif 1216 /* XXX: are preload_* static or do they need Giant ? */ 1217 while ((mod = preload_search_next_name(mod)) != NULL) { 1218 name = (char *)preload_search_info(mod, MODINFO_NAME); 1219 if (name == NULL) 1220 continue; 1221 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1222 if (type == NULL) 1223 continue; 1224 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1225 continue; 1226 c = preload_search_info(mod, MODINFO_ADDR); 1227 ptr = *(u_char **)c; 1228 c = preload_search_info(mod, MODINFO_SIZE); 1229 len = *(size_t *)c; 1230 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1231 MD_NAME, mdunits, name, len, ptr); 1232 sx_xlock(&md_sx); 1233 md_preloaded(ptr, len); 1234 sx_xunlock(&md_sx); 1235 } 1236 status_dev = make_dev(&mdctl_cdevsw, MAXMINOR, UID_ROOT, GID_WHEEL, 1237 0600, MDCTL_NAME); 1238 g_topology_lock(); 1239} 1240 1241static void 1242g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 1243 struct g_consumer *cp __unused, struct g_provider *pp) 1244{ 1245 struct md_s *mp; 1246 char *type; 1247 1248 mp = gp->softc; 1249 if (mp == NULL) 1250 return; 1251 1252 switch (mp->type) { 1253 case MD_MALLOC: 1254 type = "malloc"; 1255 break; 1256 case MD_PRELOAD: 1257 type = "preload"; 1258 break; 1259 case MD_VNODE: 1260 type = "vnode"; 1261 break; 1262 case MD_SWAP: 1263 type = "swap"; 1264 break; 1265 default: 1266 type = "unknown"; 1267 break; 1268 } 1269 1270 if (pp != NULL) { 1271 if (indent == NULL) { 1272 sbuf_printf(sb, " u %d", mp->unit); 1273 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize); 1274 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads); 1275 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors); 1276 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize); 1277 sbuf_printf(sb, " t %s", type); 1278 if (mp->type == MD_VNODE && mp->vnode != NULL) 1279 sbuf_printf(sb, " file %s", mp->file); 1280 } else { 1281 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, 1282 mp->unit); 1283 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n", 1284 indent, (uintmax_t) mp->sectorsize); 1285 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n", 1286 indent, (uintmax_t) mp->fwheads); 1287 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n", 1288 indent, (uintmax_t) mp->fwsectors); 1289 sbuf_printf(sb, "%s<length>%ju</length>\n", 1290 indent, (uintmax_t) mp->mediasize); 1291 sbuf_printf(sb, "%s<type>%s</type>\n", indent, 1292 type); 1293 if (mp->type == MD_VNODE && mp->vnode != NULL) 1294 sbuf_printf(sb, "%s<file>%s</file>\n", 1295 indent, mp->file); 1296 } 1297 } 1298} 1299 1300static void 1301g_md_fini(struct g_class *mp __unused) 1302{ 1303 1304 sx_destroy(&md_sx); 1305 if (status_dev != NULL) 1306 destroy_dev(status_dev); 1307} 1308