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 layer; 259 260 rcnt = size; 261 layer = 0; 262 while (rcnt > NINDIR) { 263 rcnt /= NINDIR; 264 layer++; 265 } 266 267 /* 268 * XXX: the top layer is probably not fully populated, so we allocate 269 * too much space for ip->array in here. 270 */ 271 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 272 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 273 M_MDSECT, M_WAITOK | M_ZERO); 274 ip->total = NINDIR; 275 ip->shift = layer * nshift; 276 return (ip); 277 } 278 279 /* 280 * Read a given sector 281 */ 282 283 static uintptr_t 284 s_read(struct indir *ip, off_t offset) 285 { 286 struct indir *cip; 287 int idx; 288 uintptr_t up; 289 290 if (md_debug > 1) 291 printf("s_read(%jd)\n", (intmax_t)offset); 292 up = 0; 293 for (cip = ip; cip != NULL;) { 294 if (cip->shift) { 295 idx = (offset >> cip->shift) & NMASK; 296 up = cip->array[idx]; 297 cip = (struct indir *)up; 298 continue; 299 } 300 idx = offset & NMASK; 301 return (cip->array[idx]); 302 } 303 return (0); 304 } 305 306 /* 307 * Write a given sector, prune the tree if the value is 0 308 */ 309 310 static int 311 s_write(struct indir *ip, off_t offset, uintptr_t ptr) 312 { 313 struct indir *cip, *lip[10]; 314 int idx, li; 315 uintptr_t up; 316 317 if (md_debug > 1) 318 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 319 up = 0; 320 li = 0; 321 cip = ip; 322 for (;;) { 323 lip[li++] = cip; 324 if (cip->shift) { 325 idx = (offset >> cip->shift) & NMASK; 326 up = cip->array[idx]; 327 if (up != 0) { 328 cip = (struct indir *)up; 329 continue; 330 } 331 /* Allocate branch */ 332 cip->array[idx] = 333 (uintptr_t)new_indir(cip->shift - nshift); 334 if (cip->array[idx] == 0) 335 return (ENOSPC); 336 cip->used++; 337 up = cip->array[idx]; 338 cip = (struct indir *)up; 339 continue; 340 } 341 /* leafnode */ 342 idx = offset & NMASK; 343 up = cip->array[idx]; 344 if (up != 0) 345 cip->used--; 346 cip->array[idx] = ptr; 347 if (ptr != 0) 348 cip->used++; 349 break; 350 } 351 if (cip->used != 0 || li == 1) 352 return (0); 353 li--; 354 while (cip->used == 0 && cip != ip) { 355 li--; 356 idx = (offset >> lip[li]->shift) & NMASK; 357 up = lip[li]->array[idx]; 358 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 359 del_indir(cip); 360 lip[li]->array[idx] = 0; 361 lip[li]->used--; 362 cip = lip[li]; 363 } 364 return (0); 365 } 366 367 368 static int 369 g_md_access(struct g_provider *pp, int r, int w, int e) 370 { 371 struct md_s *sc; 372 373 sc = pp->geom->softc; 374 if (sc == NULL) { 375 if (r <= 0 && w <= 0 && e <= 0) 376 return (0); 377 return (ENXIO); 378 } 379 r += pp->acr; 380 w += pp->acw; 381 e += pp->ace; 382 if ((sc->flags & MD_READONLY) != 0 && w > 0) 383 return (EROFS); 384 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 385 sc->opencount = 1; 386 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 387 sc->opencount = 0; 388 } 389 return (0); 390 } 391 392 static void 393 g_md_start(struct bio *bp) 394 { 395 struct md_s *sc; 396 397 sc = bp->bio_to->geom->softc; 398 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) 399 devstat_start_transaction_bio(sc->devstat, bp); 400 mtx_lock(&sc->queue_mtx); 401 bioq_disksort(&sc->bio_queue, bp); 402 mtx_unlock(&sc->queue_mtx); 403 wakeup(sc); 404 } 405 406 static int 407 mdstart_malloc(struct md_s *sc, struct bio *bp) 408 { 409 int i, error; 410 u_char *dst; 411 off_t secno, nsec, uc; 412 uintptr_t sp, osp; 413 414 switch (bp->bio_cmd) { 415 case BIO_READ: 416 case BIO_WRITE: 417 case BIO_DELETE: 418 break; 419 default: 420 return (EOPNOTSUPP); 421 } 422 423 nsec = bp->bio_length / sc->sectorsize; 424 secno = bp->bio_offset / sc->sectorsize; 425 dst = bp->bio_data; 426 error = 0; 427 while (nsec--) { 428 osp = s_read(sc->indir, secno); 429 if (bp->bio_cmd == BIO_DELETE) { 430 if (osp != 0) 431 error = s_write(sc->indir, secno, 0); 432 } else if (bp->bio_cmd == BIO_READ) { 433 if (osp == 0) 434 bzero(dst, sc->sectorsize); 435 else if (osp <= 255) 436 memset(dst, osp, sc->sectorsize); 437 else { 438 bcopy((void *)osp, dst, sc->sectorsize); 439 cpu_flush_dcache(dst, sc->sectorsize); 440 } 441 osp = 0; 442 } else if (bp->bio_cmd == BIO_WRITE) { 443 if (sc->flags & MD_COMPRESS) { 444 uc = dst[0]; 445 for (i = 1; i < sc->sectorsize; i++) 446 if (dst[i] != uc) 447 break; 448 } else { 449 i = 0; 450 uc = 0; 451 } 452 if (i == sc->sectorsize) { 453 if (osp != uc) 454 error = s_write(sc->indir, secno, uc); 455 } else { 456 if (osp <= 255) { 457 sp = (uintptr_t)uma_zalloc(sc->uma, 458 M_NOWAIT); 459 if (sp == 0) { 460 error = ENOSPC; 461 break; 462 } 463 bcopy(dst, (void *)sp, sc->sectorsize); 464 error = s_write(sc->indir, secno, sp); 465 } else { 466 bcopy(dst, (void *)osp, sc->sectorsize); 467 osp = 0; 468 } 469 } 470 } else { 471 error = EOPNOTSUPP; 472 } 473 if (osp > 255) 474 uma_zfree(sc->uma, (void*)osp); 475 if (error != 0) 476 break; 477 secno++; 478 dst += sc->sectorsize; 479 } 480 bp->bio_resid = 0; 481 return (error); 482 } 483 484 static int 485 mdstart_preload(struct md_s *sc, struct bio *bp) 486 { 487 488 switch (bp->bio_cmd) { 489 case BIO_READ: 490 bcopy(sc->pl_ptr + bp->bio_offset, bp->bio_data, 491 bp->bio_length); 492 cpu_flush_dcache(bp->bio_data, bp->bio_length); 493 break; 494 case BIO_WRITE: 495 bcopy(bp->bio_data, sc->pl_ptr + bp->bio_offset, 496 bp->bio_length); 497 break; 498 } 499 bp->bio_resid = 0; 500 return (0); 501 } 502 503 static int 504 mdstart_vnode(struct md_s *sc, struct bio *bp) 505 { 506 int error, vfslocked; 507 struct uio auio; 508 struct iovec aiov; 509 struct mount *mp; 510 struct vnode *vp; 511 struct thread *td; 512 513 switch (bp->bio_cmd) { 514 case BIO_READ: 515 case BIO_WRITE: 516 case BIO_FLUSH: 517 break; 518 default: 519 return (EOPNOTSUPP); 520 } 521 522 td = curthread; 523 vp = sc->vnode; 524 525 /* 526 * VNODE I/O 527 * 528 * If an error occurs, we set BIO_ERROR but we do not set 529 * B_INVAL because (for a write anyway), the buffer is 530 * still valid. 531 */ 532 533 if (bp->bio_cmd == BIO_FLUSH) { 534 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 535 (void) vn_start_write(vp, &mp, V_WAIT); 536 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 537 error = VOP_FSYNC(vp, MNT_WAIT, td); 538 VOP_UNLOCK(vp, 0); 539 vn_finished_write(mp); 540 VFS_UNLOCK_GIANT(vfslocked); 541 return (error); 542 } 543 544 bzero(&auio, sizeof(auio)); 545 546 aiov.iov_base = bp->bio_data; 547 aiov.iov_len = bp->bio_length; 548 auio.uio_iov = &aiov; 549 auio.uio_iovcnt = 1; 550 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 551 auio.uio_segflg = UIO_SYSSPACE; 552 if (bp->bio_cmd == BIO_READ) 553 auio.uio_rw = UIO_READ; 554 else if (bp->bio_cmd == BIO_WRITE) 555 auio.uio_rw = UIO_WRITE; 556 else 557 panic("wrong BIO_OP in mdstart_vnode"); 558 auio.uio_resid = bp->bio_length; 559 auio.uio_td = td; 560 /* 561 * When reading set IO_DIRECT to try to avoid double-caching 562 * the data. When writing IO_DIRECT is not optimal. 563 */ 564 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 565 if (bp->bio_cmd == BIO_READ) { 566 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 567 error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred); 568 VOP_UNLOCK(vp, 0); 569 } else { 570 (void) vn_start_write(vp, &mp, V_WAIT); 571 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 572 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC, 573 sc->cred); 574 VOP_UNLOCK(vp, 0); 575 vn_finished_write(mp); 576 } 577 VFS_UNLOCK_GIANT(vfslocked); 578 bp->bio_resid = auio.uio_resid; 579 return (error); 580 } 581 582 static int 583 mdstart_swap(struct md_s *sc, struct bio *bp) 584 { 585 struct sf_buf *sf; 586 int rv, offs, len, lastend; 587 vm_pindex_t i, lastp; 588 vm_page_t m; 589 u_char *p; 590 591 switch (bp->bio_cmd) { 592 case BIO_READ: 593 case BIO_WRITE: 594 case BIO_DELETE: 595 break; 596 default: 597 return (EOPNOTSUPP); 598 } 599 600 p = bp->bio_data; 601 602 /* 603 * offs is the offset at which to start operating on the 604 * next (ie, first) page. lastp is the last page on 605 * which we're going to operate. lastend is the ending 606 * position within that last page (ie, PAGE_SIZE if 607 * we're operating on complete aligned pages). 608 */ 609 offs = bp->bio_offset % PAGE_SIZE; 610 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE; 611 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1; 612 613 rv = VM_PAGER_OK; 614 VM_OBJECT_LOCK(sc->object); 615 vm_object_pip_add(sc->object, 1); 616 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) { 617 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs; 618 619 m = vm_page_grab(sc->object, i, 620 VM_ALLOC_NORMAL|VM_ALLOC_RETRY); 621 VM_OBJECT_UNLOCK(sc->object); 622 sched_pin(); 623 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 624 VM_OBJECT_LOCK(sc->object); 625 if (bp->bio_cmd == BIO_READ) { 626 if (m->valid != VM_PAGE_BITS_ALL) 627 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 628 if (rv == VM_PAGER_ERROR) { 629 sf_buf_free(sf); 630 sched_unpin(); 631 vm_page_wakeup(m); 632 break; 633 } 634 bcopy((void *)(sf_buf_kva(sf) + offs), p, len); 635 cpu_flush_dcache(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(m); 666 vm_page_activate(m); 667 vm_page_unlock(m); 668 if (bp->bio_cmd == BIO_WRITE) 669 vm_page_dirty(m); 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 if (mdio->md_base == 0) 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 } 822 823 824 static int 825 mdcreate_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 } 863 864 865 static int 866 mdsetcred(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 } 905 906 static int 907 mdcreate_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 = EINVAL; 931 goto bad; 932 } 933 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred); 934 if (error != 0) 935 goto bad; 936 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) { 937 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY); 938 if (nd.ni_vp->v_iflag & VI_DOOMED) { 939 /* Forced unmount. */ 940 error = EBADF; 941 goto bad; 942 } 943 } 944 nd.ni_vp->v_vflag |= VV_MD; 945 VOP_UNLOCK(nd.ni_vp, 0); 946 947 if (mdio->md_fwsectors != 0) 948 sc->fwsectors = mdio->md_fwsectors; 949 if (mdio->md_fwheads != 0) 950 sc->fwheads = mdio->md_fwheads; 951 sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC); 952 if (!(flags & FWRITE)) 953 sc->flags |= MD_READONLY; 954 sc->vnode = nd.ni_vp; 955 956 error = mdsetcred(sc, td->td_ucred); 957 if (error != 0) { 958 sc->vnode = NULL; 959 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY); 960 nd.ni_vp->v_vflag &= ~VV_MD; 961 goto bad; 962 } 963 VFS_UNLOCK_GIANT(vfslocked); 964 return (0); 965 bad: 966 VOP_UNLOCK(nd.ni_vp, 0); 967 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td); 968 VFS_UNLOCK_GIANT(vfslocked); 969 return (error); 970 } 971 972 static int 973 mddestroy(struct md_s *sc, struct thread *td) 974 { 975 int vfslocked; 976 977 if (sc->gp) { 978 sc->gp->softc = NULL; 979 g_topology_lock(); 980 g_wither_geom(sc->gp, ENXIO); 981 g_topology_unlock(); 982 sc->gp = NULL; 983 sc->pp = NULL; 984 } 985 if (sc->devstat) { 986 devstat_remove_entry(sc->devstat); 987 sc->devstat = NULL; 988 } 989 mtx_lock(&sc->queue_mtx); 990 sc->flags |= MD_SHUTDOWN; 991 wakeup(sc); 992 while (!(sc->flags & MD_EXITING)) 993 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10); 994 mtx_unlock(&sc->queue_mtx); 995 mtx_destroy(&sc->queue_mtx); 996 if (sc->vnode != NULL) { 997 vfslocked = VFS_LOCK_GIANT(sc->vnode->v_mount); 998 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 999 sc->vnode->v_vflag &= ~VV_MD; 1000 VOP_UNLOCK(sc->vnode, 0); 1001 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 1002 FREAD : (FREAD|FWRITE), sc->cred, td); 1003 VFS_UNLOCK_GIANT(vfslocked); 1004 } 1005 if (sc->cred != NULL) 1006 crfree(sc->cred); 1007 if (sc->object != NULL) 1008 vm_object_deallocate(sc->object); 1009 if (sc->indir) 1010 destroy_indir(sc, sc->indir); 1011 if (sc->uma) 1012 uma_zdestroy(sc->uma); 1013 1014 LIST_REMOVE(sc, list); 1015 free(sc, M_MD); 1016 return (0); 1017 } 1018 1019 static int 1020 mdcreate_swap(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 1021 { 1022 vm_ooffset_t npage; 1023 int error; 1024 1025 /* 1026 * Range check. Disallow negative sizes or any size less then the 1027 * size of a page. Then round to a page. 1028 */ 1029 if (sc->mediasize == 0 || (sc->mediasize % PAGE_SIZE) != 0) 1030 return (EDOM); 1031 1032 /* 1033 * Allocate an OBJT_SWAP object. 1034 * 1035 * Note the truncation. 1036 */ 1037 1038 npage = mdio->md_mediasize / PAGE_SIZE; 1039 if (mdio->md_fwsectors != 0) 1040 sc->fwsectors = mdio->md_fwsectors; 1041 if (mdio->md_fwheads != 0) 1042 sc->fwheads = mdio->md_fwheads; 1043 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage, 1044 VM_PROT_DEFAULT, 0, td->td_ucred); 1045 if (sc->object == NULL) 1046 return (ENOMEM); 1047 sc->flags = mdio->md_options & MD_FORCE; 1048 if (mdio->md_options & MD_RESERVE) { 1049 if (swap_pager_reserve(sc->object, 0, npage) < 0) { 1050 error = EDOM; 1051 goto finish; 1052 } 1053 } 1054 error = mdsetcred(sc, td->td_ucred); 1055 finish: 1056 if (error != 0) { 1057 vm_object_deallocate(sc->object); 1058 sc->object = NULL; 1059 } 1060 return (error); 1061 } 1062 1063 1064 static int 1065 xmdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1066 { 1067 struct md_ioctl *mdio; 1068 struct md_s *sc; 1069 int error, i; 1070 1071 if (md_debug) 1072 printf("mdctlioctl(%s %lx %p %x %p)\n", 1073 devtoname(dev), cmd, addr, flags, td); 1074 1075 mdio = (struct md_ioctl *)addr; 1076 if (mdio->md_version != MDIOVERSION) 1077 return (EINVAL); 1078 1079 /* 1080 * We assert the version number in the individual ioctl 1081 * handlers instead of out here because (a) it is possible we 1082 * may add another ioctl in the future which doesn't read an 1083 * mdio, and (b) the correct return value for an unknown ioctl 1084 * is ENOIOCTL, not EINVAL. 1085 */ 1086 error = 0; 1087 switch (cmd) { 1088 case MDIOCATTACH: 1089 switch (mdio->md_type) { 1090 case MD_MALLOC: 1091 case MD_PRELOAD: 1092 case MD_VNODE: 1093 case MD_SWAP: 1094 break; 1095 default: 1096 return (EINVAL); 1097 } 1098 if (mdio->md_options & MD_AUTOUNIT) 1099 sc = mdnew(-1, &error, mdio->md_type); 1100 else 1101 sc = mdnew(mdio->md_unit, &error, mdio->md_type); 1102 if (sc == NULL) 1103 return (error); 1104 if (mdio->md_options & MD_AUTOUNIT) 1105 mdio->md_unit = sc->unit; 1106 sc->mediasize = mdio->md_mediasize; 1107 if (mdio->md_sectorsize == 0) 1108 sc->sectorsize = DEV_BSIZE; 1109 else 1110 sc->sectorsize = mdio->md_sectorsize; 1111 error = EDOOFUS; 1112 switch (sc->type) { 1113 case MD_MALLOC: 1114 sc->start = mdstart_malloc; 1115 error = mdcreate_malloc(sc, mdio); 1116 break; 1117 case MD_PRELOAD: 1118 sc->start = mdstart_preload; 1119 error = mdcreate_preload(sc, mdio); 1120 break; 1121 case MD_VNODE: 1122 sc->start = mdstart_vnode; 1123 error = mdcreate_vnode(sc, mdio, td); 1124 break; 1125 case MD_SWAP: 1126 sc->start = mdstart_swap; 1127 error = mdcreate_swap(sc, mdio, td); 1128 break; 1129 } 1130 if (error != 0) { 1131 mddestroy(sc, td); 1132 return (error); 1133 } 1134 1135 /* Prune off any residual fractional sector */ 1136 i = sc->mediasize % sc->sectorsize; 1137 sc->mediasize -= i; 1138 1139 mdinit(sc); 1140 return (0); 1141 case MDIOCDETACH: 1142 if (mdio->md_mediasize != 0 || 1143 (mdio->md_options & ~MD_FORCE) != 0) 1144 return (EINVAL); 1145 1146 sc = mdfind(mdio->md_unit); 1147 if (sc == NULL) 1148 return (ENOENT); 1149 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) && 1150 !(mdio->md_options & MD_FORCE)) 1151 return (EBUSY); 1152 return (mddestroy(sc, td)); 1153 case MDIOCQUERY: 1154 sc = mdfind(mdio->md_unit); 1155 if (sc == NULL) 1156 return (ENOENT); 1157 mdio->md_type = sc->type; 1158 mdio->md_options = sc->flags; 1159 mdio->md_mediasize = sc->mediasize; 1160 mdio->md_sectorsize = sc->sectorsize; 1161 if (sc->type == MD_VNODE) 1162 error = copyout(sc->file, mdio->md_file, 1163 strlen(sc->file) + 1); 1164 return (error); 1165 case MDIOCLIST: 1166 i = 1; 1167 LIST_FOREACH(sc, &md_softc_list, list) { 1168 if (i == MDNPAD - 1) 1169 mdio->md_pad[i] = -1; 1170 else 1171 mdio->md_pad[i++] = sc->unit; 1172 } 1173 mdio->md_pad[0] = i - 1; 1174 return (0); 1175 default: 1176 return (ENOIOCTL); 1177 }; 1178 } 1179 1180 static int 1181 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1182 { 1183 int error; 1184 1185 sx_xlock(&md_sx); 1186 error = xmdctlioctl(dev, cmd, addr, flags, td); 1187 sx_xunlock(&md_sx); 1188 return (error); 1189 } 1190 1191 static void 1192 md_preloaded(u_char *image, size_t length) 1193 { 1194 struct md_s *sc; 1195 int error; 1196 1197 sc = mdnew(-1, &error, MD_PRELOAD); 1198 if (sc == NULL) 1199 return; 1200 sc->mediasize = length; 1201 sc->sectorsize = DEV_BSIZE; 1202 sc->pl_ptr = image; 1203 sc->pl_len = length; 1204 sc->start = mdstart_preload; 1205 #ifdef MD_ROOT 1206 if (sc->unit == 0) 1207 rootdevnames[0] = "ufs:/dev/md0"; 1208 #endif 1209 mdinit(sc); 1210 } 1211 1212 static void 1213 g_md_init(struct g_class *mp __unused) 1214 { 1215 1216 caddr_t mod; 1217 caddr_t c; 1218 u_char *ptr, *name, *type; 1219 unsigned len; 1220 int i; 1221 1222 /* figure out log2(NINDIR) */ 1223 for (i = NINDIR, nshift = -1; i; nshift++) 1224 i >>= 1; 1225 1226 mod = NULL; 1227 sx_init(&md_sx, "MD config lock"); 1228 g_topology_unlock(); 1229 #ifdef MD_ROOT_SIZE 1230 sx_xlock(&md_sx); 1231 md_preloaded(mfs_root.start, sizeof(mfs_root.start)); 1232 sx_xunlock(&md_sx); 1233 #endif 1234 /* XXX: are preload_* static or do they need Giant ? */ 1235 while ((mod = preload_search_next_name(mod)) != NULL) { 1236 name = (char *)preload_search_info(mod, MODINFO_NAME); 1237 if (name == NULL) 1238 continue; 1239 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1240 if (type == NULL) 1241 continue; 1242 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1243 continue; 1244 c = preload_search_info(mod, MODINFO_ADDR); 1245 ptr = *(u_char **)c; 1246 c = preload_search_info(mod, MODINFO_SIZE); 1247 len = *(size_t *)c; 1248 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1249 MD_NAME, mdunits, name, len, ptr); 1250 sx_xlock(&md_sx); 1251 md_preloaded(ptr, len); 1252 sx_xunlock(&md_sx); 1253 } 1254 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL, 1255 0600, MDCTL_NAME); 1256 g_topology_lock(); 1257 } 1258 1259 static void 1260 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 1261 struct g_consumer *cp __unused, struct g_provider *pp) 1262 { 1263 struct md_s *mp; 1264 char *type; 1265 1266 mp = gp->softc; 1267 if (mp == NULL) 1268 return; 1269 1270 switch (mp->type) { 1271 case MD_MALLOC: 1272 type = "malloc"; 1273 break; 1274 case MD_PRELOAD: 1275 type = "preload"; 1276 break; 1277 case MD_VNODE: 1278 type = "vnode"; 1279 break; 1280 case MD_SWAP: 1281 type = "swap"; 1282 break; 1283 default: 1284 type = "unknown"; 1285 break; 1286 } 1287 1288 if (pp != NULL) { 1289 if (indent == NULL) { 1290 sbuf_printf(sb, " u %d", mp->unit); 1291 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize); 1292 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads); 1293 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors); 1294 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize); 1295 sbuf_printf(sb, " t %s", type); 1296 if (mp->type == MD_VNODE && mp->vnode != NULL) 1297 sbuf_printf(sb, " file %s", mp->file); 1298 } else { 1299 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, 1300 mp->unit); 1301 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n", 1302 indent, (uintmax_t) mp->sectorsize); 1303 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n", 1304 indent, (uintmax_t) mp->fwheads); 1305 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n", 1306 indent, (uintmax_t) mp->fwsectors); 1307 sbuf_printf(sb, "%s<length>%ju</length>\n", 1308 indent, (uintmax_t) mp->mediasize); 1309 sbuf_printf(sb, "%s<type>%s</type>\n", indent, 1310 type); 1311 if (mp->type == MD_VNODE && mp->vnode != NULL) 1312 sbuf_printf(sb, "%s<file>%s</file>\n", 1313 indent, mp->file); 1314 } 1315 } 1316 } 1317 1318 static void 1319 g_md_fini(struct g_class *mp __unused) 1320 { 1321 1322 sx_destroy(&md_sx); 1323 if (status_dev != NULL) 1324 destroy_dev(status_dev); 1325 } 1326