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 * $FreeBSD$ 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 100 101 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk"); 102 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors"); 103 104 static int md_debug; 105 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, ""); 106 107 #if defined(MD_ROOT) && defined(MD_ROOT_SIZE) 108 /* 109 * Preloaded image gets put here. 110 * Applications that patch the object with the image can determine 111 * the size looking at the start and end markers (strings), 112 * so we want them contiguous. 113 */ 114 static struct { 115 u_char start[MD_ROOT_SIZE*1024]; 116 u_char end[128]; 117 } mfs_root = { 118 .start = "MFS Filesystem goes here", 119 .end = "MFS Filesystem had better STOP here", 120 }; 121 #endif 122 123 static g_init_t g_md_init; 124 static g_fini_t g_md_fini; 125 static g_start_t g_md_start; 126 static g_access_t g_md_access; 127 static void g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 128 struct g_consumer *cp __unused, struct g_provider *pp); 129 130 static int mdunits; 131 static struct cdev *status_dev = 0; 132 static struct sx md_sx; 133 134 static d_ioctl_t mdctlioctl; 135 136 static struct cdevsw mdctl_cdevsw = { 137 .d_version = D_VERSION, 138 .d_ioctl = mdctlioctl, 139 .d_name = MD_NAME, 140 }; 141 142 struct g_class g_md_class = { 143 .name = "MD", 144 .version = G_VERSION, 145 .init = g_md_init, 146 .fini = g_md_fini, 147 .start = g_md_start, 148 .access = g_md_access, 149 .dumpconf = g_md_dumpconf, 150 }; 151 152 DECLARE_GEOM_CLASS(g_md_class, g_md); 153 154 155 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(&md_softc_list); 156 157 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 158 #define NMASK (NINDIR-1) 159 static int nshift; 160 161 struct indir { 162 uintptr_t *array; 163 u_int total; 164 u_int used; 165 u_int shift; 166 }; 167 168 struct md_s { 169 int unit; 170 LIST_ENTRY(md_s) list; 171 struct bio_queue_head bio_queue; 172 struct mtx queue_mtx; 173 struct cdev *dev; 174 enum md_types type; 175 off_t mediasize; 176 unsigned sectorsize; 177 unsigned opencount; 178 unsigned fwheads; 179 unsigned fwsectors; 180 unsigned flags; 181 char name[20]; 182 struct proc *procp; 183 struct g_geom *gp; 184 struct g_provider *pp; 185 int (*start)(struct md_s *sc, struct bio *bp); 186 struct devstat *devstat; 187 188 /* MD_MALLOC related fields */ 189 struct indir *indir; 190 uma_zone_t uma; 191 192 /* MD_PRELOAD related fields */ 193 u_char *pl_ptr; 194 size_t pl_len; 195 196 /* MD_VNODE related fields */ 197 struct vnode *vnode; 198 char file[PATH_MAX]; 199 struct ucred *cred; 200 201 /* MD_SWAP related fields */ 202 vm_object_t object; 203 }; 204 205 static struct indir * 206 new_indir(u_int shift) 207 { 208 struct indir *ip; 209 210 ip = malloc(sizeof *ip, M_MD, M_NOWAIT | M_ZERO); 211 if (ip == NULL) 212 return (NULL); 213 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 214 M_MDSECT, M_NOWAIT | M_ZERO); 215 if (ip->array == NULL) { 216 free(ip, M_MD); 217 return (NULL); 218 } 219 ip->total = NINDIR; 220 ip->shift = shift; 221 return (ip); 222 } 223 224 static void 225 del_indir(struct indir *ip) 226 { 227 228 free(ip->array, M_MDSECT); 229 free(ip, M_MD); 230 } 231 232 static void 233 destroy_indir(struct md_s *sc, struct indir *ip) 234 { 235 int i; 236 237 for (i = 0; i < NINDIR; i++) { 238 if (!ip->array[i]) 239 continue; 240 if (ip->shift) 241 destroy_indir(sc, (struct indir*)(ip->array[i])); 242 else if (ip->array[i] > 255) 243 uma_zfree(sc->uma, (void *)(ip->array[i])); 244 } 245 del_indir(ip); 246 } 247 248 /* 249 * This function does the math and allocates the top level "indir" structure 250 * for a device of "size" sectors. 251 */ 252 253 static struct indir * 254 dimension(off_t size) 255 { 256 off_t rcnt; 257 struct indir *ip; 258 int i, layer; 259 260 rcnt = size; 261 layer = 0; 262 while (rcnt > NINDIR) { 263 rcnt /= NINDIR; 264 layer++; 265 } 266 /* figure out log2(NINDIR) */ 267 for (i = NINDIR, nshift = -1; i; nshift++) 268 i >>= 1; 269 270 /* 271 * XXX: the top layer is probably not fully populated, so we allocate 272 * too much space for ip->array in here. 273 */ 274 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 275 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 276 M_MDSECT, M_WAITOK | M_ZERO); 277 ip->total = NINDIR; 278 ip->shift = layer * nshift; 279 return (ip); 280 } 281 282 /* 283 * Read a given sector 284 */ 285 286 static uintptr_t 287 s_read(struct indir *ip, off_t offset) 288 { 289 struct indir *cip; 290 int idx; 291 uintptr_t up; 292 293 if (md_debug > 1) 294 printf("s_read(%jd)\n", (intmax_t)offset); 295 up = 0; 296 for (cip = ip; cip != NULL;) { 297 if (cip->shift) { 298 idx = (offset >> cip->shift) & NMASK; 299 up = cip->array[idx]; 300 cip = (struct indir *)up; 301 continue; 302 } 303 idx = offset & NMASK; 304 return (cip->array[idx]); 305 } 306 return (0); 307 } 308 309 /* 310 * Write a given sector, prune the tree if the value is 0 311 */ 312 313 static int 314 s_write(struct indir *ip, off_t offset, uintptr_t ptr) 315 { 316 struct indir *cip, *lip[10]; 317 int idx, li; 318 uintptr_t up; 319 320 if (md_debug > 1) 321 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 322 up = 0; 323 li = 0; 324 cip = ip; 325 for (;;) { 326 lip[li++] = cip; 327 if (cip->shift) { 328 idx = (offset >> cip->shift) & NMASK; 329 up = cip->array[idx]; 330 if (up != 0) { 331 cip = (struct indir *)up; 332 continue; 333 } 334 /* Allocate branch */ 335 cip->array[idx] = 336 (uintptr_t)new_indir(cip->shift - nshift); 337 if (cip->array[idx] == 0) 338 return (ENOSPC); 339 cip->used++; 340 up = cip->array[idx]; 341 cip = (struct indir *)up; 342 continue; 343 } 344 /* leafnode */ 345 idx = offset & NMASK; 346 up = cip->array[idx]; 347 if (up != 0) 348 cip->used--; 349 cip->array[idx] = ptr; 350 if (ptr != 0) 351 cip->used++; 352 break; 353 } 354 if (cip->used != 0 || li == 1) 355 return (0); 356 li--; 357 while (cip->used == 0 && cip != ip) { 358 li--; 359 idx = (offset >> lip[li]->shift) & NMASK; 360 up = lip[li]->array[idx]; 361 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 362 del_indir(cip); 363 lip[li]->array[idx] = 0; 364 lip[li]->used--; 365 cip = lip[li]; 366 } 367 return (0); 368 } 369 370 371 static int 372 g_md_access(struct g_provider *pp, int r, int w, int e) 373 { 374 struct md_s *sc; 375 376 sc = pp->geom->softc; 377 if (sc == NULL) { 378 if (r <= 0 && w <= 0 && e <= 0) 379 return (0); 380 return (ENXIO); 381 } 382 r += pp->acr; 383 w += pp->acw; 384 e += pp->ace; 385 if ((sc->flags & MD_READONLY) != 0 && w > 0) 386 return (EROFS); 387 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 388 sc->opencount = 1; 389 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 390 sc->opencount = 0; 391 } 392 return (0); 393 } 394 395 static void 396 g_md_start(struct bio *bp) 397 { 398 struct md_s *sc; 399 400 sc = bp->bio_to->geom->softc; 401 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) 402 devstat_start_transaction_bio(sc->devstat, bp); 403 mtx_lock(&sc->queue_mtx); 404 bioq_disksort(&sc->bio_queue, bp); 405 mtx_unlock(&sc->queue_mtx); 406 wakeup(sc); 407 } 408 409 static int 410 mdstart_malloc(struct md_s *sc, struct bio *bp) 411 { 412 int i, error; 413 u_char *dst; 414 off_t secno, nsec, uc; 415 uintptr_t sp, osp; 416 417 switch (bp->bio_cmd) { 418 case BIO_READ: 419 case BIO_WRITE: 420 case BIO_DELETE: 421 break; 422 default: 423 return (EOPNOTSUPP); 424 } 425 426 nsec = bp->bio_length / sc->sectorsize; 427 secno = bp->bio_offset / sc->sectorsize; 428 dst = bp->bio_data; 429 error = 0; 430 while (nsec--) { 431 osp = s_read(sc->indir, secno); 432 if (bp->bio_cmd == BIO_DELETE) { 433 if (osp != 0) 434 error = s_write(sc->indir, secno, 0); 435 } else if (bp->bio_cmd == BIO_READ) { 436 if (osp == 0) 437 bzero(dst, sc->sectorsize); 438 else if (osp <= 255) 439 for (i = 0; i < sc->sectorsize; i++) 440 dst[i] = osp; 441 else 442 bcopy((void *)osp, dst, sc->sectorsize); 443 osp = 0; 444 } else if (bp->bio_cmd == BIO_WRITE) { 445 if (sc->flags & MD_COMPRESS) { 446 uc = dst[0]; 447 for (i = 1; i < sc->sectorsize; i++) 448 if (dst[i] != uc) 449 break; 450 } else { 451 i = 0; 452 uc = 0; 453 } 454 if (i == sc->sectorsize) { 455 if (osp != uc) 456 error = s_write(sc->indir, secno, uc); 457 } else { 458 if (osp <= 255) { 459 sp = (uintptr_t)uma_zalloc(sc->uma, 460 M_NOWAIT); 461 if (sp == 0) { 462 error = ENOSPC; 463 break; 464 } 465 bcopy(dst, (void *)sp, sc->sectorsize); 466 error = s_write(sc->indir, secno, sp); 467 } else { 468 bcopy(dst, (void *)osp, sc->sectorsize); 469 osp = 0; 470 } 471 } 472 } else { 473 error = EOPNOTSUPP; 474 } 475 if (osp > 255) 476 uma_zfree(sc->uma, (void*)osp); 477 if (error != 0) 478 break; 479 secno++; 480 dst += sc->sectorsize; 481 } 482 bp->bio_resid = 0; 483 return (error); 484 } 485 486 static int 487 mdstart_preload(struct md_s *sc, struct bio *bp) 488 { 489 490 switch (bp->bio_cmd) { 491 case BIO_READ: 492 bcopy(sc->pl_ptr + bp->bio_offset, bp->bio_data, 493 bp->bio_length); 494 break; 495 case BIO_WRITE: 496 bcopy(bp->bio_data, sc->pl_ptr + bp->bio_offset, 497 bp->bio_length); 498 break; 499 } 500 bp->bio_resid = 0; 501 return (0); 502 } 503 504 static int 505 mdstart_vnode(struct md_s *sc, struct bio *bp) 506 { 507 int error, vfslocked; 508 struct uio auio; 509 struct iovec aiov; 510 struct mount *mp; 511 struct vnode *vp; 512 struct thread *td; 513 514 switch (bp->bio_cmd) { 515 case BIO_READ: 516 case BIO_WRITE: 517 case BIO_FLUSH: 518 break; 519 default: 520 return (EOPNOTSUPP); 521 } 522 523 td = curthread; 524 vp = sc->vnode; 525 526 /* 527 * VNODE I/O 528 * 529 * If an error occurs, we set BIO_ERROR but we do not set 530 * B_INVAL because (for a write anyway), the buffer is 531 * still valid. 532 */ 533 534 if (bp->bio_cmd == BIO_FLUSH) { 535 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 536 (void) vn_start_write(vp, &mp, V_WAIT); 537 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 538 error = VOP_FSYNC(vp, MNT_WAIT, td); 539 VOP_UNLOCK(vp, 0); 540 vn_finished_write(mp); 541 VFS_UNLOCK_GIANT(vfslocked); 542 return (error); 543 } 544 545 bzero(&auio, sizeof(auio)); 546 547 aiov.iov_base = bp->bio_data; 548 aiov.iov_len = bp->bio_length; 549 auio.uio_iov = &aiov; 550 auio.uio_iovcnt = 1; 551 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 552 auio.uio_segflg = UIO_SYSSPACE; 553 if (bp->bio_cmd == BIO_READ) 554 auio.uio_rw = UIO_READ; 555 else if (bp->bio_cmd == BIO_WRITE) 556 auio.uio_rw = UIO_WRITE; 557 else 558 panic("wrong BIO_OP in mdstart_vnode"); 559 auio.uio_resid = bp->bio_length; 560 auio.uio_td = td; 561 /* 562 * When reading set IO_DIRECT to try to avoid double-caching 563 * the data. When writing IO_DIRECT is not optimal. 564 */ 565 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 566 if (bp->bio_cmd == BIO_READ) { 567 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 568 error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred); 569 VOP_UNLOCK(vp, 0); 570 } else { 571 (void) vn_start_write(vp, &mp, V_WAIT); 572 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 573 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC, 574 sc->cred); 575 VOP_UNLOCK(vp, 0); 576 vn_finished_write(mp); 577 } 578 VFS_UNLOCK_GIANT(vfslocked); 579 bp->bio_resid = auio.uio_resid; 580 return (error); 581 } 582 583 static int 584 mdstart_swap(struct md_s *sc, struct bio *bp) 585 { 586 struct sf_buf *sf; 587 int rv, offs, len, lastend; 588 vm_pindex_t i, lastp; 589 vm_page_t m; 590 u_char *p; 591 592 switch (bp->bio_cmd) { 593 case BIO_READ: 594 case BIO_WRITE: 595 case BIO_DELETE: 596 break; 597 default: 598 return (EOPNOTSUPP); 599 } 600 601 p = bp->bio_data; 602 603 /* 604 * offs is the offset at which to start operating on the 605 * next (ie, first) page. lastp is the last page on 606 * which we're going to operate. lastend is the ending 607 * position within that last page (ie, PAGE_SIZE if 608 * we're operating on complete aligned pages). 609 */ 610 offs = bp->bio_offset % PAGE_SIZE; 611 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE; 612 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1; 613 614 rv = VM_PAGER_OK; 615 VM_OBJECT_LOCK(sc->object); 616 vm_object_pip_add(sc->object, 1); 617 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) { 618 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs; 619 620 m = vm_page_grab(sc->object, i, 621 VM_ALLOC_NORMAL|VM_ALLOC_RETRY); 622 VM_OBJECT_UNLOCK(sc->object); 623 sched_pin(); 624 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 625 VM_OBJECT_LOCK(sc->object); 626 if (bp->bio_cmd == BIO_READ) { 627 if (m->valid != VM_PAGE_BITS_ALL) 628 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 629 if (rv == VM_PAGER_ERROR) { 630 sf_buf_free(sf); 631 sched_unpin(); 632 vm_page_wakeup(m); 633 break; 634 } 635 bcopy((void *)(sf_buf_kva(sf) + offs), p, len); 636 } else if (bp->bio_cmd == BIO_WRITE) { 637 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 638 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 639 if (rv == VM_PAGER_ERROR) { 640 sf_buf_free(sf); 641 sched_unpin(); 642 vm_page_wakeup(m); 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_wakeup(m); 655 break; 656 } 657 bzero((void *)(sf_buf_kva(sf) + offs), len); 658 vm_page_dirty(m); 659 m->valid = VM_PAGE_BITS_ALL; 660 #endif 661 } 662 sf_buf_free(sf); 663 sched_unpin(); 664 vm_page_wakeup(m); 665 vm_page_lock_queues(); 666 vm_page_activate(m); 667 if (bp->bio_cmd == BIO_WRITE) 668 vm_page_dirty(m); 669 vm_page_unlock_queues(); 670 671 /* Actions on further pages start at offset 0 */ 672 p += PAGE_SIZE - offs; 673 offs = 0; 674 #if 0 675 if (bootverbose || bp->bio_offset / PAGE_SIZE < 17) 676 printf("wire_count %d busy %d flags %x hold_count %d act_count %d queue %d valid %d dirty %d @ %d\n", 677 m->wire_count, m->busy, 678 m->flags, m->hold_count, m->act_count, m->queue, m->valid, m->dirty, i); 679 #endif 680 } 681 vm_object_pip_subtract(sc->object, 1); 682 vm_object_set_writeable_dirty(sc->object); 683 VM_OBJECT_UNLOCK(sc->object); 684 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC); 685 } 686 687 static void 688 md_kthread(void *arg) 689 { 690 struct md_s *sc; 691 struct bio *bp; 692 int error; 693 694 sc = arg; 695 thread_lock(curthread); 696 sched_prio(curthread, PRIBIO); 697 thread_unlock(curthread); 698 if (sc->type == MD_VNODE) 699 curthread->td_pflags |= TDP_NORUNNINGBUF; 700 701 for (;;) { 702 mtx_lock(&sc->queue_mtx); 703 if (sc->flags & MD_SHUTDOWN) { 704 sc->flags |= MD_EXITING; 705 mtx_unlock(&sc->queue_mtx); 706 kproc_exit(0); 707 } 708 bp = bioq_takefirst(&sc->bio_queue); 709 if (!bp) { 710 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 711 continue; 712 } 713 mtx_unlock(&sc->queue_mtx); 714 if (bp->bio_cmd == BIO_GETATTR) { 715 if (sc->fwsectors && sc->fwheads && 716 (g_handleattr_int(bp, "GEOM::fwsectors", 717 sc->fwsectors) || 718 g_handleattr_int(bp, "GEOM::fwheads", 719 sc->fwheads))) 720 error = -1; 721 else 722 error = EOPNOTSUPP; 723 } else { 724 error = sc->start(sc, bp); 725 } 726 727 if (error != -1) { 728 bp->bio_completed = bp->bio_length; 729 g_io_deliver(bp, error); 730 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) 731 devstat_end_transaction_bio(sc->devstat, bp); 732 } 733 } 734 } 735 736 static struct md_s * 737 mdfind(int unit) 738 { 739 struct md_s *sc; 740 741 LIST_FOREACH(sc, &md_softc_list, list) { 742 if (sc->unit == unit) 743 break; 744 } 745 return (sc); 746 } 747 748 static struct md_s * 749 mdnew(int unit, int *errp, enum md_types type) 750 { 751 struct md_s *sc, *sc2; 752 int error, max = -1; 753 754 *errp = 0; 755 LIST_FOREACH(sc2, &md_softc_list, list) { 756 if (unit == sc2->unit) { 757 *errp = EBUSY; 758 return (NULL); 759 } 760 if (unit == -1 && sc2->unit > max) 761 max = sc2->unit; 762 } 763 if (unit == -1) 764 unit = max + 1; 765 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 766 sc->type = type; 767 bioq_init(&sc->bio_queue); 768 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 769 sc->unit = unit; 770 sprintf(sc->name, "md%d", unit); 771 LIST_INSERT_HEAD(&md_softc_list, sc, list); 772 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 773 if (error == 0) 774 return (sc); 775 LIST_REMOVE(sc, list); 776 mtx_destroy(&sc->queue_mtx); 777 free(sc, M_MD); 778 *errp = error; 779 return (NULL); 780 } 781 782 static void 783 mdinit(struct md_s *sc) 784 { 785 786 struct g_geom *gp; 787 struct g_provider *pp; 788 789 g_topology_lock(); 790 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 791 gp->softc = sc; 792 pp = g_new_providerf(gp, "md%d", sc->unit); 793 pp->mediasize = sc->mediasize; 794 pp->sectorsize = sc->sectorsize; 795 sc->gp = gp; 796 sc->pp = pp; 797 g_error_provider(pp, 0); 798 g_topology_unlock(); 799 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize, 800 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX); 801 } 802 803 /* 804 * XXX: we should check that the range they feed us is mapped. 805 * XXX: we should implement read-only. 806 */ 807 808 static int 809 mdcreate_preload(struct md_s *sc, struct md_ioctl *mdio) 810 { 811 812 if (mdio->md_options & ~(MD_AUTOUNIT | MD_FORCE)) 813 return (EINVAL); 814 sc->flags = mdio->md_options & MD_FORCE; 815 /* Cast to pointer size, then to pointer to avoid warning */ 816 sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base; 817 sc->pl_len = (size_t)sc->mediasize; 818 return (0); 819 } 820 821 822 static int 823 mdcreate_malloc(struct md_s *sc, struct md_ioctl *mdio) 824 { 825 uintptr_t sp; 826 int error; 827 off_t u; 828 829 error = 0; 830 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 831 return (EINVAL); 832 if (mdio->md_sectorsize != 0 && !powerof2(mdio->md_sectorsize)) 833 return (EINVAL); 834 /* Compression doesn't make sense if we have reserved space */ 835 if (mdio->md_options & MD_RESERVE) 836 mdio->md_options &= ~MD_COMPRESS; 837 if (mdio->md_fwsectors != 0) 838 sc->fwsectors = mdio->md_fwsectors; 839 if (mdio->md_fwheads != 0) 840 sc->fwheads = mdio->md_fwheads; 841 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 842 sc->indir = dimension(sc->mediasize / sc->sectorsize); 843 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL, 844 0x1ff, 0); 845 if (mdio->md_options & MD_RESERVE) { 846 off_t nsectors; 847 848 nsectors = sc->mediasize / sc->sectorsize; 849 for (u = 0; u < nsectors; u++) { 850 sp = (uintptr_t)uma_zalloc(sc->uma, M_NOWAIT | M_ZERO); 851 if (sp != 0) 852 error = s_write(sc->indir, u, sp); 853 else 854 error = ENOMEM; 855 if (error != 0) 856 break; 857 } 858 } 859 return (error); 860 } 861 862 863 static int 864 mdsetcred(struct md_s *sc, struct ucred *cred) 865 { 866 char *tmpbuf; 867 int error = 0; 868 869 /* 870 * Set credits in our softc 871 */ 872 873 if (sc->cred) 874 crfree(sc->cred); 875 sc->cred = crhold(cred); 876 877 /* 878 * Horrible kludge to establish credentials for NFS XXX. 879 */ 880 881 if (sc->vnode) { 882 struct uio auio; 883 struct iovec aiov; 884 885 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK); 886 bzero(&auio, sizeof(auio)); 887 888 aiov.iov_base = tmpbuf; 889 aiov.iov_len = sc->sectorsize; 890 auio.uio_iov = &aiov; 891 auio.uio_iovcnt = 1; 892 auio.uio_offset = 0; 893 auio.uio_rw = UIO_READ; 894 auio.uio_segflg = UIO_SYSSPACE; 895 auio.uio_resid = aiov.iov_len; 896 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 897 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 898 VOP_UNLOCK(sc->vnode, 0); 899 free(tmpbuf, M_TEMP); 900 } 901 return (error); 902 } 903 904 static int 905 mdcreate_vnode(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 906 { 907 struct vattr vattr; 908 struct nameidata nd; 909 int error, flags, vfslocked; 910 911 error = copyinstr(mdio->md_file, sc->file, sizeof(sc->file), NULL); 912 if (error != 0) 913 return (error); 914 flags = FREAD|FWRITE; 915 /* 916 * If the user specified that this is a read only device, unset the 917 * FWRITE mask before trying to open the backing store. 918 */ 919 if ((mdio->md_options & MD_READONLY) != 0) 920 flags &= ~FWRITE; 921 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, sc->file, td); 922 error = vn_open(&nd, &flags, 0, NULL); 923 if (error != 0) 924 return (error); 925 vfslocked = NDHASGIANT(&nd); 926 NDFREE(&nd, NDF_ONLY_PNBUF); 927 if (nd.ni_vp->v_type != VREG) { 928 error = EINVAL; 929 goto bad; 930 } 931 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred); 932 if (error != 0) 933 goto bad; 934 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) { 935 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY); 936 if (nd.ni_vp->v_iflag & VI_DOOMED) { 937 /* Forced unmount. */ 938 error = EBADF; 939 goto bad; 940 } 941 } 942 nd.ni_vp->v_vflag |= VV_MD; 943 VOP_UNLOCK(nd.ni_vp, 0); 944 945 if (mdio->md_fwsectors != 0) 946 sc->fwsectors = mdio->md_fwsectors; 947 if (mdio->md_fwheads != 0) 948 sc->fwheads = mdio->md_fwheads; 949 sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC); 950 if (!(flags & FWRITE)) 951 sc->flags |= MD_READONLY; 952 sc->vnode = nd.ni_vp; 953 954 error = mdsetcred(sc, td->td_ucred); 955 if (error != 0) { 956 sc->vnode = NULL; 957 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY); 958 nd.ni_vp->v_vflag &= ~VV_MD; 959 goto bad; 960 } 961 VFS_UNLOCK_GIANT(vfslocked); 962 return (0); 963 bad: 964 VOP_UNLOCK(nd.ni_vp, 0); 965 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td); 966 VFS_UNLOCK_GIANT(vfslocked); 967 return (error); 968 } 969 970 static int 971 mddestroy(struct md_s *sc, struct thread *td) 972 { 973 int vfslocked; 974 975 if (sc->gp) { 976 sc->gp->softc = NULL; 977 g_topology_lock(); 978 g_wither_geom(sc->gp, ENXIO); 979 g_topology_unlock(); 980 sc->gp = NULL; 981 sc->pp = NULL; 982 } 983 if (sc->devstat) { 984 devstat_remove_entry(sc->devstat); 985 sc->devstat = NULL; 986 } 987 mtx_lock(&sc->queue_mtx); 988 sc->flags |= MD_SHUTDOWN; 989 wakeup(sc); 990 while (!(sc->flags & MD_EXITING)) 991 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10); 992 mtx_unlock(&sc->queue_mtx); 993 mtx_destroy(&sc->queue_mtx); 994 if (sc->vnode != NULL) { 995 vfslocked = VFS_LOCK_GIANT(sc->vnode->v_mount); 996 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 997 sc->vnode->v_vflag &= ~VV_MD; 998 VOP_UNLOCK(sc->vnode, 0); 999 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 1000 FREAD : (FREAD|FWRITE), sc->cred, td); 1001 VFS_UNLOCK_GIANT(vfslocked); 1002 } 1003 if (sc->cred != NULL) 1004 crfree(sc->cred); 1005 if (sc->object != NULL) 1006 vm_object_deallocate(sc->object); 1007 if (sc->indir) 1008 destroy_indir(sc, sc->indir); 1009 if (sc->uma) 1010 uma_zdestroy(sc->uma); 1011 1012 LIST_REMOVE(sc, list); 1013 free(sc, M_MD); 1014 return (0); 1015 } 1016 1017 static int 1018 mdcreate_swap(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 1019 { 1020 vm_ooffset_t npage; 1021 int error; 1022 1023 /* 1024 * Range check. Disallow negative sizes or any size less then the 1025 * size of a page. Then round to a page. 1026 */ 1027 if (sc->mediasize == 0 || (sc->mediasize % PAGE_SIZE) != 0) 1028 return (EDOM); 1029 1030 /* 1031 * Allocate an OBJT_SWAP object. 1032 * 1033 * Note the truncation. 1034 */ 1035 1036 npage = mdio->md_mediasize / PAGE_SIZE; 1037 if (mdio->md_fwsectors != 0) 1038 sc->fwsectors = mdio->md_fwsectors; 1039 if (mdio->md_fwheads != 0) 1040 sc->fwheads = mdio->md_fwheads; 1041 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage, 1042 VM_PROT_DEFAULT, 0); 1043 if (sc->object == NULL) 1044 return (ENOMEM); 1045 sc->flags = mdio->md_options & MD_FORCE; 1046 if (mdio->md_options & MD_RESERVE) { 1047 if (swap_pager_reserve(sc->object, 0, npage) < 0) { 1048 vm_object_deallocate(sc->object); 1049 sc->object = NULL; 1050 return (EDOM); 1051 } 1052 } 1053 error = mdsetcred(sc, td->td_ucred); 1054 if (error != 0) { 1055 vm_object_deallocate(sc->object); 1056 sc->object = NULL; 1057 } 1058 return (error); 1059 } 1060 1061 1062 static int 1063 xmdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1064 { 1065 struct md_ioctl *mdio; 1066 struct md_s *sc; 1067 int error, i; 1068 1069 if (md_debug) 1070 printf("mdctlioctl(%s %lx %p %x %p)\n", 1071 devtoname(dev), cmd, addr, flags, td); 1072 1073 mdio = (struct md_ioctl *)addr; 1074 if (mdio->md_version != MDIOVERSION) 1075 return (EINVAL); 1076 1077 /* 1078 * We assert the version number in the individual ioctl 1079 * handlers instead of out here because (a) it is possible we 1080 * may add another ioctl in the future which doesn't read an 1081 * mdio, and (b) the correct return value for an unknown ioctl 1082 * is ENOIOCTL, not EINVAL. 1083 */ 1084 error = 0; 1085 switch (cmd) { 1086 case MDIOCATTACH: 1087 switch (mdio->md_type) { 1088 case MD_MALLOC: 1089 case MD_PRELOAD: 1090 case MD_VNODE: 1091 case MD_SWAP: 1092 break; 1093 default: 1094 return (EINVAL); 1095 } 1096 if (mdio->md_options & MD_AUTOUNIT) 1097 sc = mdnew(-1, &error, mdio->md_type); 1098 else 1099 sc = mdnew(mdio->md_unit, &error, mdio->md_type); 1100 if (sc == NULL) 1101 return (error); 1102 if (mdio->md_options & MD_AUTOUNIT) 1103 mdio->md_unit = sc->unit; 1104 sc->mediasize = mdio->md_mediasize; 1105 if (mdio->md_sectorsize == 0) 1106 sc->sectorsize = DEV_BSIZE; 1107 else 1108 sc->sectorsize = mdio->md_sectorsize; 1109 error = EDOOFUS; 1110 switch (sc->type) { 1111 case MD_MALLOC: 1112 sc->start = mdstart_malloc; 1113 error = mdcreate_malloc(sc, mdio); 1114 break; 1115 case MD_PRELOAD: 1116 sc->start = mdstart_preload; 1117 error = mdcreate_preload(sc, mdio); 1118 break; 1119 case MD_VNODE: 1120 sc->start = mdstart_vnode; 1121 error = mdcreate_vnode(sc, mdio, td); 1122 break; 1123 case MD_SWAP: 1124 sc->start = mdstart_swap; 1125 error = mdcreate_swap(sc, mdio, td); 1126 break; 1127 } 1128 if (error != 0) { 1129 mddestroy(sc, td); 1130 return (error); 1131 } 1132 1133 /* Prune off any residual fractional sector */ 1134 i = sc->mediasize % sc->sectorsize; 1135 sc->mediasize -= i; 1136 1137 mdinit(sc); 1138 return (0); 1139 case MDIOCDETACH: 1140 if (mdio->md_mediasize != 0 || 1141 (mdio->md_options & ~MD_FORCE) != 0) 1142 return (EINVAL); 1143 1144 sc = mdfind(mdio->md_unit); 1145 if (sc == NULL) 1146 return (ENOENT); 1147 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) && 1148 !(mdio->md_options & MD_FORCE)) 1149 return (EBUSY); 1150 return (mddestroy(sc, td)); 1151 case MDIOCQUERY: 1152 sc = mdfind(mdio->md_unit); 1153 if (sc == NULL) 1154 return (ENOENT); 1155 mdio->md_type = sc->type; 1156 mdio->md_options = sc->flags; 1157 mdio->md_mediasize = sc->mediasize; 1158 mdio->md_sectorsize = sc->sectorsize; 1159 if (sc->type == MD_VNODE) 1160 error = copyout(sc->file, mdio->md_file, 1161 strlen(sc->file) + 1); 1162 return (error); 1163 case MDIOCLIST: 1164 i = 1; 1165 LIST_FOREACH(sc, &md_softc_list, list) { 1166 if (i == MDNPAD - 1) 1167 mdio->md_pad[i] = -1; 1168 else 1169 mdio->md_pad[i++] = sc->unit; 1170 } 1171 mdio->md_pad[0] = i - 1; 1172 return (0); 1173 default: 1174 return (ENOIOCTL); 1175 }; 1176 } 1177 1178 static int 1179 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1180 { 1181 int error; 1182 1183 sx_xlock(&md_sx); 1184 error = xmdctlioctl(dev, cmd, addr, flags, td); 1185 sx_xunlock(&md_sx); 1186 return (error); 1187 } 1188 1189 static void 1190 md_preloaded(u_char *image, size_t length) 1191 { 1192 struct md_s *sc; 1193 int error; 1194 1195 sc = mdnew(-1, &error, MD_PRELOAD); 1196 if (sc == NULL) 1197 return; 1198 sc->mediasize = length; 1199 sc->sectorsize = DEV_BSIZE; 1200 sc->pl_ptr = image; 1201 sc->pl_len = length; 1202 sc->start = mdstart_preload; 1203 #ifdef MD_ROOT 1204 if (sc->unit == 0) 1205 rootdevnames[0] = "ufs:/dev/md0"; 1206 #endif 1207 mdinit(sc); 1208 } 1209 1210 static void 1211 g_md_init(struct g_class *mp __unused) 1212 { 1213 1214 caddr_t mod; 1215 caddr_t c; 1216 u_char *ptr, *name, *type; 1217 unsigned len; 1218 1219 mod = NULL; 1220 sx_init(&md_sx, "MD config lock"); 1221 g_topology_unlock(); 1222 #ifdef MD_ROOT_SIZE 1223 sx_xlock(&md_sx); 1224 md_preloaded(mfs_root.start, sizeof(mfs_root.start)); 1225 sx_xunlock(&md_sx); 1226 #endif 1227 /* XXX: are preload_* static or do they need Giant ? */ 1228 while ((mod = preload_search_next_name(mod)) != NULL) { 1229 name = (char *)preload_search_info(mod, MODINFO_NAME); 1230 if (name == NULL) 1231 continue; 1232 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1233 if (type == NULL) 1234 continue; 1235 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1236 continue; 1237 c = preload_search_info(mod, MODINFO_ADDR); 1238 ptr = *(u_char **)c; 1239 c = preload_search_info(mod, MODINFO_SIZE); 1240 len = *(size_t *)c; 1241 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1242 MD_NAME, mdunits, name, len, ptr); 1243 sx_xlock(&md_sx); 1244 md_preloaded(ptr, len); 1245 sx_xunlock(&md_sx); 1246 } 1247 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL, 1248 0600, MDCTL_NAME); 1249 g_topology_lock(); 1250 } 1251 1252 static void 1253 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 1254 struct g_consumer *cp __unused, struct g_provider *pp) 1255 { 1256 struct md_s *mp; 1257 char *type; 1258 1259 mp = gp->softc; 1260 if (mp == NULL) 1261 return; 1262 1263 switch (mp->type) { 1264 case MD_MALLOC: 1265 type = "malloc"; 1266 break; 1267 case MD_PRELOAD: 1268 type = "preload"; 1269 break; 1270 case MD_VNODE: 1271 type = "vnode"; 1272 break; 1273 case MD_SWAP: 1274 type = "swap"; 1275 break; 1276 default: 1277 type = "unknown"; 1278 break; 1279 } 1280 1281 if (pp != NULL) { 1282 if (indent == NULL) { 1283 sbuf_printf(sb, " u %d", mp->unit); 1284 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize); 1285 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads); 1286 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors); 1287 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize); 1288 sbuf_printf(sb, " t %s", type); 1289 if (mp->type == MD_VNODE && mp->vnode != NULL) 1290 sbuf_printf(sb, " file %s", mp->file); 1291 } else { 1292 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, 1293 mp->unit); 1294 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n", 1295 indent, (uintmax_t) mp->sectorsize); 1296 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n", 1297 indent, (uintmax_t) mp->fwheads); 1298 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n", 1299 indent, (uintmax_t) mp->fwsectors); 1300 sbuf_printf(sb, "%s<length>%ju</length>\n", 1301 indent, (uintmax_t) mp->mediasize); 1302 sbuf_printf(sb, "%s<type>%s</type>\n", indent, 1303 type); 1304 if (mp->type == MD_VNODE && mp->vnode != NULL) 1305 sbuf_printf(sb, "%s<file>%s</file>\n", 1306 indent, mp->file); 1307 } 1308 } 1309 } 1310 1311 static void 1312 g_md_fini(struct g_class *mp __unused) 1313 { 1314 1315 sx_destroy(&md_sx); 1316 if (status_dev != NULL) 1317 destroy_dev(status_dev); 1318 } 1319