1 /*- 2 * SPDX-License-Identifier: (Beerware AND BSD-3-Clause) 3 * 4 * ---------------------------------------------------------------------------- 5 * "THE BEER-WARE LICENSE" (Revision 42): 6 * <phk@FreeBSD.ORG> wrote this file. As long as you retain this notice you 7 * can do whatever you want with this stuff. If we meet some day, and you think 8 * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp 9 * ---------------------------------------------------------------------------- 10 * 11 * $FreeBSD$ 12 * 13 */ 14 15 /*- 16 * The following functions are based on the vn(4) driver: mdstart_swap(), 17 * mdstart_vnode(), mdcreate_swap(), mdcreate_vnode() and mddestroy(), 18 * and as such under the following copyright: 19 * 20 * Copyright (c) 1988 University of Utah. 21 * Copyright (c) 1990, 1993 22 * The Regents of the University of California. All rights reserved. 23 * Copyright (c) 2013 The FreeBSD Foundation 24 * All rights reserved. 25 * 26 * This code is derived from software contributed to Berkeley by 27 * the Systems Programming Group of the University of Utah Computer 28 * Science Department. 29 * 30 * Portions of this software were developed by Konstantin Belousov 31 * under sponsorship from the FreeBSD Foundation. 32 * 33 * Redistribution and use in source and binary forms, with or without 34 * modification, are permitted provided that the following conditions 35 * are met: 36 * 1. Redistributions of source code must retain the above copyright 37 * notice, this list of conditions and the following disclaimer. 38 * 2. Redistributions in binary form must reproduce the above copyright 39 * notice, this list of conditions and the following disclaimer in the 40 * documentation and/or other materials provided with the distribution. 41 * 3. Neither the name of the University nor the names of its contributors 42 * may be used to endorse or promote products derived from this software 43 * without specific prior written permission. 44 * 45 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 48 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 55 * SUCH DAMAGE. 56 * 57 * from: Utah Hdr: vn.c 1.13 94/04/02 58 * 59 * from: @(#)vn.c 8.6 (Berkeley) 4/1/94 60 * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03 61 */ 62 63 #include "opt_rootdevname.h" 64 #include "opt_geom.h" 65 #include "opt_md.h" 66 67 #include <sys/param.h> 68 #include <sys/systm.h> 69 #include <sys/bio.h> 70 #include <sys/buf.h> 71 #include <sys/conf.h> 72 #include <sys/devicestat.h> 73 #include <sys/fcntl.h> 74 #include <sys/kernel.h> 75 #include <sys/kthread.h> 76 #include <sys/limits.h> 77 #include <sys/linker.h> 78 #include <sys/lock.h> 79 #include <sys/malloc.h> 80 #include <sys/mdioctl.h> 81 #include <sys/mount.h> 82 #include <sys/mutex.h> 83 #include <sys/sx.h> 84 #include <sys/namei.h> 85 #include <sys/proc.h> 86 #include <sys/queue.h> 87 #include <sys/rwlock.h> 88 #include <sys/sbuf.h> 89 #include <sys/sched.h> 90 #include <sys/sf_buf.h> 91 #include <sys/sysctl.h> 92 #include <sys/uio.h> 93 #include <sys/vnode.h> 94 #include <sys/disk.h> 95 96 #include <geom/geom.h> 97 #include <geom/geom_int.h> 98 99 #include <vm/vm.h> 100 #include <vm/vm_param.h> 101 #include <vm/vm_object.h> 102 #include <vm/vm_page.h> 103 #include <vm/vm_pager.h> 104 #include <vm/swap_pager.h> 105 #include <vm/uma.h> 106 107 #include <machine/bus.h> 108 109 #define MD_MODVER 1 110 111 #define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */ 112 #define MD_EXITING 0x20000 /* Worker thread is exiting. */ 113 #define MD_PROVIDERGONE 0x40000 /* Safe to free the softc */ 114 115 #ifndef MD_NSECT 116 #define MD_NSECT (10000 * 2) 117 #endif 118 119 struct md_req { 120 unsigned md_unit; /* unit number */ 121 enum md_types md_type; /* type of disk */ 122 off_t md_mediasize; /* size of disk in bytes */ 123 unsigned md_sectorsize; /* sectorsize */ 124 unsigned md_options; /* options */ 125 int md_fwheads; /* firmware heads */ 126 int md_fwsectors; /* firmware sectors */ 127 char *md_file; /* pathname of file to mount */ 128 enum uio_seg md_file_seg; /* location of md_file */ 129 char *md_label; /* label of the device (userspace) */ 130 int *md_units; /* pointer to units array (kernel) */ 131 size_t md_units_nitems; /* items in md_units array */ 132 }; 133 134 #ifdef COMPAT_FREEBSD32 135 struct md_ioctl32 { 136 unsigned md_version; 137 unsigned md_unit; 138 enum md_types md_type; 139 uint32_t md_file; 140 off_t md_mediasize; 141 unsigned md_sectorsize; 142 unsigned md_options; 143 uint64_t md_base; 144 int md_fwheads; 145 int md_fwsectors; 146 uint32_t md_label; 147 int md_pad[MDNPAD]; 148 } __attribute__((__packed__)); 149 CTASSERT((sizeof(struct md_ioctl32)) == 436); 150 151 #define MDIOCATTACH_32 _IOC_NEWTYPE(MDIOCATTACH, struct md_ioctl32) 152 #define MDIOCDETACH_32 _IOC_NEWTYPE(MDIOCDETACH, struct md_ioctl32) 153 #define MDIOCQUERY_32 _IOC_NEWTYPE(MDIOCQUERY, struct md_ioctl32) 154 #define MDIOCRESIZE_32 _IOC_NEWTYPE(MDIOCRESIZE, struct md_ioctl32) 155 #endif /* COMPAT_FREEBSD32 */ 156 157 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk"); 158 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors"); 159 160 static int md_debug; 161 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, 162 "Enable md(4) debug messages"); 163 static int md_malloc_wait; 164 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0, 165 "Allow malloc to wait for memory allocations"); 166 167 #if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE) 168 #define MD_ROOT_FSTYPE "ufs" 169 #endif 170 171 #if defined(MD_ROOT) 172 /* 173 * Preloaded image gets put here. 174 */ 175 #if defined(MD_ROOT_SIZE) 176 /* 177 * We put the mfs_root symbol into the oldmfs section of the kernel object file. 178 * Applications that patch the object with the image can determine 179 * the size looking at the oldmfs section size within the kernel. 180 */ 181 u_char mfs_root[MD_ROOT_SIZE*1024] __attribute__ ((section ("oldmfs"))); 182 const int mfs_root_size = sizeof(mfs_root); 183 #elif defined(MD_ROOT_MEM) 184 /* MD region already mapped in the memory */ 185 u_char *mfs_root; 186 int mfs_root_size; 187 #else 188 extern volatile u_char __weak_symbol mfs_root; 189 extern volatile u_char __weak_symbol mfs_root_end; 190 __GLOBL(mfs_root); 191 __GLOBL(mfs_root_end); 192 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root)) 193 #endif 194 #endif 195 196 static g_init_t g_md_init; 197 static g_fini_t g_md_fini; 198 static g_start_t g_md_start; 199 static g_access_t g_md_access; 200 static void g_md_dumpconf(struct sbuf *sb, const char *indent, 201 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp); 202 static g_provgone_t g_md_providergone; 203 204 static struct cdev *status_dev = NULL; 205 static struct sx md_sx; 206 static struct unrhdr *md_uh; 207 208 static d_ioctl_t mdctlioctl; 209 210 static struct cdevsw mdctl_cdevsw = { 211 .d_version = D_VERSION, 212 .d_ioctl = mdctlioctl, 213 .d_name = MD_NAME, 214 }; 215 216 struct g_class g_md_class = { 217 .name = "MD", 218 .version = G_VERSION, 219 .init = g_md_init, 220 .fini = g_md_fini, 221 .start = g_md_start, 222 .access = g_md_access, 223 .dumpconf = g_md_dumpconf, 224 .providergone = g_md_providergone, 225 }; 226 227 DECLARE_GEOM_CLASS(g_md_class, g_md); 228 229 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list); 230 231 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 232 #define NMASK (NINDIR-1) 233 static int nshift; 234 235 static uma_zone_t md_pbuf_zone; 236 237 struct indir { 238 uintptr_t *array; 239 u_int total; 240 u_int used; 241 u_int shift; 242 }; 243 244 struct md_s { 245 int unit; 246 LIST_ENTRY(md_s) list; 247 struct bio_queue_head bio_queue; 248 struct mtx queue_mtx; 249 struct cdev *dev; 250 enum md_types type; 251 off_t mediasize; 252 unsigned sectorsize; 253 unsigned opencount; 254 unsigned fwheads; 255 unsigned fwsectors; 256 char ident[32]; 257 unsigned flags; 258 char name[20]; 259 struct proc *procp; 260 struct g_geom *gp; 261 struct g_provider *pp; 262 int (*start)(struct md_s *sc, struct bio *bp); 263 struct devstat *devstat; 264 265 /* MD_MALLOC related fields */ 266 struct indir *indir; 267 uma_zone_t uma; 268 269 /* MD_PRELOAD related fields */ 270 u_char *pl_ptr; 271 size_t pl_len; 272 273 /* MD_VNODE related fields */ 274 struct vnode *vnode; 275 char file[PATH_MAX]; 276 char label[PATH_MAX]; 277 struct ucred *cred; 278 279 /* MD_SWAP related fields */ 280 vm_object_t object; 281 }; 282 283 static struct indir * 284 new_indir(u_int shift) 285 { 286 struct indir *ip; 287 288 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT) 289 | M_ZERO); 290 if (ip == NULL) 291 return (NULL); 292 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 293 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO); 294 if (ip->array == NULL) { 295 free(ip, M_MD); 296 return (NULL); 297 } 298 ip->total = NINDIR; 299 ip->shift = shift; 300 return (ip); 301 } 302 303 static void 304 del_indir(struct indir *ip) 305 { 306 307 free(ip->array, M_MDSECT); 308 free(ip, M_MD); 309 } 310 311 static void 312 destroy_indir(struct md_s *sc, struct indir *ip) 313 { 314 int i; 315 316 for (i = 0; i < NINDIR; i++) { 317 if (!ip->array[i]) 318 continue; 319 if (ip->shift) 320 destroy_indir(sc, (struct indir*)(ip->array[i])); 321 else if (ip->array[i] > 255) 322 uma_zfree(sc->uma, (void *)(ip->array[i])); 323 } 324 del_indir(ip); 325 } 326 327 /* 328 * This function does the math and allocates the top level "indir" structure 329 * for a device of "size" sectors. 330 */ 331 332 static struct indir * 333 dimension(off_t size) 334 { 335 off_t rcnt; 336 struct indir *ip; 337 int layer; 338 339 rcnt = size; 340 layer = 0; 341 while (rcnt > NINDIR) { 342 rcnt /= NINDIR; 343 layer++; 344 } 345 346 /* 347 * XXX: the top layer is probably not fully populated, so we allocate 348 * too much space for ip->array in here. 349 */ 350 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 351 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 352 M_MDSECT, M_WAITOK | M_ZERO); 353 ip->total = NINDIR; 354 ip->shift = layer * nshift; 355 return (ip); 356 } 357 358 /* 359 * Read a given sector 360 */ 361 362 static uintptr_t 363 s_read(struct indir *ip, off_t offset) 364 { 365 struct indir *cip; 366 int idx; 367 uintptr_t up; 368 369 if (md_debug > 1) 370 printf("s_read(%jd)\n", (intmax_t)offset); 371 up = 0; 372 for (cip = ip; cip != NULL;) { 373 if (cip->shift) { 374 idx = (offset >> cip->shift) & NMASK; 375 up = cip->array[idx]; 376 cip = (struct indir *)up; 377 continue; 378 } 379 idx = offset & NMASK; 380 return (cip->array[idx]); 381 } 382 return (0); 383 } 384 385 /* 386 * Write a given sector, prune the tree if the value is 0 387 */ 388 389 static int 390 s_write(struct indir *ip, off_t offset, uintptr_t ptr) 391 { 392 struct indir *cip, *lip[10]; 393 int idx, li; 394 uintptr_t up; 395 396 if (md_debug > 1) 397 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 398 up = 0; 399 li = 0; 400 cip = ip; 401 for (;;) { 402 lip[li++] = cip; 403 if (cip->shift) { 404 idx = (offset >> cip->shift) & NMASK; 405 up = cip->array[idx]; 406 if (up != 0) { 407 cip = (struct indir *)up; 408 continue; 409 } 410 /* Allocate branch */ 411 cip->array[idx] = 412 (uintptr_t)new_indir(cip->shift - nshift); 413 if (cip->array[idx] == 0) 414 return (ENOSPC); 415 cip->used++; 416 up = cip->array[idx]; 417 cip = (struct indir *)up; 418 continue; 419 } 420 /* leafnode */ 421 idx = offset & NMASK; 422 up = cip->array[idx]; 423 if (up != 0) 424 cip->used--; 425 cip->array[idx] = ptr; 426 if (ptr != 0) 427 cip->used++; 428 break; 429 } 430 if (cip->used != 0 || li == 1) 431 return (0); 432 li--; 433 while (cip->used == 0 && cip != ip) { 434 li--; 435 idx = (offset >> lip[li]->shift) & NMASK; 436 up = lip[li]->array[idx]; 437 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 438 del_indir(cip); 439 lip[li]->array[idx] = 0; 440 lip[li]->used--; 441 cip = lip[li]; 442 } 443 return (0); 444 } 445 446 static int 447 g_md_access(struct g_provider *pp, int r, int w, int e) 448 { 449 struct md_s *sc; 450 451 sc = pp->geom->softc; 452 if (sc == NULL) { 453 if (r <= 0 && w <= 0 && e <= 0) 454 return (0); 455 return (ENXIO); 456 } 457 r += pp->acr; 458 w += pp->acw; 459 e += pp->ace; 460 if ((sc->flags & MD_READONLY) != 0 && w > 0) 461 return (EROFS); 462 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 463 sc->opencount = 1; 464 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 465 sc->opencount = 0; 466 } 467 return (0); 468 } 469 470 static void 471 g_md_start(struct bio *bp) 472 { 473 struct md_s *sc; 474 475 sc = bp->bio_to->geom->softc; 476 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) { 477 devstat_start_transaction_bio(sc->devstat, bp); 478 } 479 mtx_lock(&sc->queue_mtx); 480 bioq_disksort(&sc->bio_queue, bp); 481 wakeup(sc); 482 mtx_unlock(&sc->queue_mtx); 483 } 484 485 #define MD_MALLOC_MOVE_ZERO 1 486 #define MD_MALLOC_MOVE_FILL 2 487 #define MD_MALLOC_MOVE_READ 3 488 #define MD_MALLOC_MOVE_WRITE 4 489 #define MD_MALLOC_MOVE_CMP 5 490 491 static int 492 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize, 493 void *ptr, u_char fill, int op) 494 { 495 struct sf_buf *sf; 496 vm_page_t m, *mp1; 497 char *p, first; 498 off_t *uc; 499 unsigned n; 500 int error, i, ma_offs1, sz, first_read; 501 502 m = NULL; 503 error = 0; 504 sf = NULL; 505 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */ 506 first = 0; 507 first_read = 0; 508 uc = ptr; 509 mp1 = *mp; 510 ma_offs1 = *ma_offs; 511 /* } */ 512 sched_pin(); 513 for (n = sectorsize; n != 0; n -= sz) { 514 sz = imin(PAGE_SIZE - *ma_offs, n); 515 if (m != **mp) { 516 if (sf != NULL) 517 sf_buf_free(sf); 518 m = **mp; 519 sf = sf_buf_alloc(m, SFB_CPUPRIVATE | 520 (md_malloc_wait ? 0 : SFB_NOWAIT)); 521 if (sf == NULL) { 522 error = ENOMEM; 523 break; 524 } 525 } 526 p = (char *)sf_buf_kva(sf) + *ma_offs; 527 switch (op) { 528 case MD_MALLOC_MOVE_ZERO: 529 bzero(p, sz); 530 break; 531 case MD_MALLOC_MOVE_FILL: 532 memset(p, fill, sz); 533 break; 534 case MD_MALLOC_MOVE_READ: 535 bcopy(ptr, p, sz); 536 cpu_flush_dcache(p, sz); 537 break; 538 case MD_MALLOC_MOVE_WRITE: 539 bcopy(p, ptr, sz); 540 break; 541 case MD_MALLOC_MOVE_CMP: 542 for (i = 0; i < sz; i++, p++) { 543 if (!first_read) { 544 *uc = (u_char)*p; 545 first = *p; 546 first_read = 1; 547 } else if (*p != first) { 548 error = EDOOFUS; 549 break; 550 } 551 } 552 break; 553 default: 554 KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op)); 555 break; 556 } 557 if (error != 0) 558 break; 559 *ma_offs += sz; 560 *ma_offs %= PAGE_SIZE; 561 if (*ma_offs == 0) 562 (*mp)++; 563 ptr = (char *)ptr + sz; 564 } 565 566 if (sf != NULL) 567 sf_buf_free(sf); 568 sched_unpin(); 569 if (op == MD_MALLOC_MOVE_CMP && error != 0) { 570 *mp = mp1; 571 *ma_offs = ma_offs1; 572 } 573 return (error); 574 } 575 576 static int 577 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs, 578 unsigned len, void *ptr, u_char fill, int op) 579 { 580 bus_dma_segment_t *vlist; 581 uint8_t *p, *end, first; 582 off_t *uc; 583 int ma_offs, seg_len; 584 585 vlist = *pvlist; 586 ma_offs = *pma_offs; 587 uc = ptr; 588 589 for (; len != 0; len -= seg_len) { 590 seg_len = imin(vlist->ds_len - ma_offs, len); 591 p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs; 592 switch (op) { 593 case MD_MALLOC_MOVE_ZERO: 594 bzero(p, seg_len); 595 break; 596 case MD_MALLOC_MOVE_FILL: 597 memset(p, fill, seg_len); 598 break; 599 case MD_MALLOC_MOVE_READ: 600 bcopy(ptr, p, seg_len); 601 cpu_flush_dcache(p, seg_len); 602 break; 603 case MD_MALLOC_MOVE_WRITE: 604 bcopy(p, ptr, seg_len); 605 break; 606 case MD_MALLOC_MOVE_CMP: 607 end = p + seg_len; 608 first = *uc = *p; 609 /* Confirm all following bytes match the first */ 610 while (++p < end) { 611 if (*p != first) 612 return (EDOOFUS); 613 } 614 break; 615 default: 616 KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op)); 617 break; 618 } 619 620 ma_offs += seg_len; 621 if (ma_offs == vlist->ds_len) { 622 ma_offs = 0; 623 vlist++; 624 } 625 ptr = (uint8_t *)ptr + seg_len; 626 } 627 *pvlist = vlist; 628 *pma_offs = ma_offs; 629 630 return (0); 631 } 632 633 static int 634 mdstart_malloc(struct md_s *sc, struct bio *bp) 635 { 636 u_char *dst; 637 vm_page_t *m; 638 bus_dma_segment_t *vlist; 639 int i, error, error1, ma_offs, notmapped; 640 off_t secno, nsec, uc; 641 uintptr_t sp, osp; 642 643 switch (bp->bio_cmd) { 644 case BIO_READ: 645 case BIO_WRITE: 646 case BIO_DELETE: 647 break; 648 default: 649 return (EOPNOTSUPP); 650 } 651 652 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0; 653 vlist = (bp->bio_flags & BIO_VLIST) != 0 ? 654 (bus_dma_segment_t *)bp->bio_data : NULL; 655 if (notmapped) { 656 m = bp->bio_ma; 657 ma_offs = bp->bio_ma_offset; 658 dst = NULL; 659 KASSERT(vlist == NULL, ("vlists cannot be unmapped")); 660 } else if (vlist != NULL) { 661 ma_offs = bp->bio_ma_offset; 662 dst = NULL; 663 } else { 664 dst = bp->bio_data; 665 } 666 667 nsec = bp->bio_length / sc->sectorsize; 668 secno = bp->bio_offset / sc->sectorsize; 669 error = 0; 670 while (nsec--) { 671 osp = s_read(sc->indir, secno); 672 if (bp->bio_cmd == BIO_DELETE) { 673 if (osp != 0) 674 error = s_write(sc->indir, secno, 0); 675 } else if (bp->bio_cmd == BIO_READ) { 676 if (osp == 0) { 677 if (notmapped) { 678 error = md_malloc_move_ma(&m, &ma_offs, 679 sc->sectorsize, NULL, 0, 680 MD_MALLOC_MOVE_ZERO); 681 } else if (vlist != NULL) { 682 error = md_malloc_move_vlist(&vlist, 683 &ma_offs, sc->sectorsize, NULL, 0, 684 MD_MALLOC_MOVE_ZERO); 685 } else 686 bzero(dst, sc->sectorsize); 687 } else if (osp <= 255) { 688 if (notmapped) { 689 error = md_malloc_move_ma(&m, &ma_offs, 690 sc->sectorsize, NULL, osp, 691 MD_MALLOC_MOVE_FILL); 692 } else if (vlist != NULL) { 693 error = md_malloc_move_vlist(&vlist, 694 &ma_offs, sc->sectorsize, NULL, osp, 695 MD_MALLOC_MOVE_FILL); 696 } else 697 memset(dst, osp, sc->sectorsize); 698 } else { 699 if (notmapped) { 700 error = md_malloc_move_ma(&m, &ma_offs, 701 sc->sectorsize, (void *)osp, 0, 702 MD_MALLOC_MOVE_READ); 703 } else if (vlist != NULL) { 704 error = md_malloc_move_vlist(&vlist, 705 &ma_offs, sc->sectorsize, 706 (void *)osp, 0, 707 MD_MALLOC_MOVE_READ); 708 } else { 709 bcopy((void *)osp, dst, sc->sectorsize); 710 cpu_flush_dcache(dst, sc->sectorsize); 711 } 712 } 713 osp = 0; 714 } else if (bp->bio_cmd == BIO_WRITE) { 715 if (sc->flags & MD_COMPRESS) { 716 if (notmapped) { 717 error1 = md_malloc_move_ma(&m, &ma_offs, 718 sc->sectorsize, &uc, 0, 719 MD_MALLOC_MOVE_CMP); 720 i = error1 == 0 ? sc->sectorsize : 0; 721 } else if (vlist != NULL) { 722 error1 = md_malloc_move_vlist(&vlist, 723 &ma_offs, sc->sectorsize, &uc, 0, 724 MD_MALLOC_MOVE_CMP); 725 i = error1 == 0 ? sc->sectorsize : 0; 726 } else { 727 uc = dst[0]; 728 for (i = 1; i < sc->sectorsize; i++) { 729 if (dst[i] != uc) 730 break; 731 } 732 } 733 } else { 734 i = 0; 735 uc = 0; 736 } 737 if (i == sc->sectorsize) { 738 if (osp != uc) 739 error = s_write(sc->indir, secno, uc); 740 } else { 741 if (osp <= 255) { 742 sp = (uintptr_t)uma_zalloc(sc->uma, 743 md_malloc_wait ? M_WAITOK : 744 M_NOWAIT); 745 if (sp == 0) { 746 error = ENOSPC; 747 break; 748 } 749 if (notmapped) { 750 error = md_malloc_move_ma(&m, 751 &ma_offs, sc->sectorsize, 752 (void *)sp, 0, 753 MD_MALLOC_MOVE_WRITE); 754 } else if (vlist != NULL) { 755 error = md_malloc_move_vlist( 756 &vlist, &ma_offs, 757 sc->sectorsize, (void *)sp, 758 0, MD_MALLOC_MOVE_WRITE); 759 } else { 760 bcopy(dst, (void *)sp, 761 sc->sectorsize); 762 } 763 error = s_write(sc->indir, secno, sp); 764 } else { 765 if (notmapped) { 766 error = md_malloc_move_ma(&m, 767 &ma_offs, sc->sectorsize, 768 (void *)osp, 0, 769 MD_MALLOC_MOVE_WRITE); 770 } else if (vlist != NULL) { 771 error = md_malloc_move_vlist( 772 &vlist, &ma_offs, 773 sc->sectorsize, (void *)osp, 774 0, MD_MALLOC_MOVE_WRITE); 775 } else { 776 bcopy(dst, (void *)osp, 777 sc->sectorsize); 778 } 779 osp = 0; 780 } 781 } 782 } else { 783 error = EOPNOTSUPP; 784 } 785 if (osp > 255) 786 uma_zfree(sc->uma, (void*)osp); 787 if (error != 0) 788 break; 789 secno++; 790 if (!notmapped && vlist == NULL) 791 dst += sc->sectorsize; 792 } 793 bp->bio_resid = 0; 794 return (error); 795 } 796 797 static void 798 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len) 799 { 800 off_t seg_len; 801 802 while (offset >= vlist->ds_len) { 803 offset -= vlist->ds_len; 804 vlist++; 805 } 806 807 while (len != 0) { 808 seg_len = omin(len, vlist->ds_len - offset); 809 bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset), 810 seg_len); 811 offset = 0; 812 src = (uint8_t *)src + seg_len; 813 len -= seg_len; 814 vlist++; 815 } 816 } 817 818 static void 819 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len) 820 { 821 off_t seg_len; 822 823 while (offset >= vlist->ds_len) { 824 offset -= vlist->ds_len; 825 vlist++; 826 } 827 828 while (len != 0) { 829 seg_len = omin(len, vlist->ds_len - offset); 830 bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst, 831 seg_len); 832 offset = 0; 833 dst = (uint8_t *)dst + seg_len; 834 len -= seg_len; 835 vlist++; 836 } 837 } 838 839 static int 840 mdstart_preload(struct md_s *sc, struct bio *bp) 841 { 842 uint8_t *p; 843 844 p = sc->pl_ptr + bp->bio_offset; 845 switch (bp->bio_cmd) { 846 case BIO_READ: 847 if ((bp->bio_flags & BIO_VLIST) != 0) { 848 mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data, 849 bp->bio_ma_offset, bp->bio_length); 850 } else { 851 bcopy(p, bp->bio_data, bp->bio_length); 852 } 853 cpu_flush_dcache(bp->bio_data, bp->bio_length); 854 break; 855 case BIO_WRITE: 856 if ((bp->bio_flags & BIO_VLIST) != 0) { 857 mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data, 858 bp->bio_ma_offset, p, bp->bio_length); 859 } else { 860 bcopy(bp->bio_data, p, bp->bio_length); 861 } 862 break; 863 } 864 bp->bio_resid = 0; 865 return (0); 866 } 867 868 static int 869 mdstart_vnode(struct md_s *sc, struct bio *bp) 870 { 871 int error; 872 struct uio auio; 873 struct iovec aiov; 874 struct iovec *piov; 875 struct mount *mp; 876 struct vnode *vp; 877 struct buf *pb; 878 bus_dma_segment_t *vlist; 879 struct thread *td; 880 off_t iolen, iostart, len, zerosize; 881 int ma_offs, npages; 882 883 switch (bp->bio_cmd) { 884 case BIO_READ: 885 auio.uio_rw = UIO_READ; 886 break; 887 case BIO_WRITE: 888 case BIO_DELETE: 889 auio.uio_rw = UIO_WRITE; 890 break; 891 case BIO_FLUSH: 892 break; 893 default: 894 return (EOPNOTSUPP); 895 } 896 897 td = curthread; 898 vp = sc->vnode; 899 pb = NULL; 900 piov = NULL; 901 ma_offs = bp->bio_ma_offset; 902 len = bp->bio_length; 903 904 /* 905 * VNODE I/O 906 * 907 * If an error occurs, we set BIO_ERROR but we do not set 908 * B_INVAL because (for a write anyway), the buffer is 909 * still valid. 910 */ 911 912 if (bp->bio_cmd == BIO_FLUSH) { 913 (void) vn_start_write(vp, &mp, V_WAIT); 914 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 915 error = VOP_FSYNC(vp, MNT_WAIT, td); 916 VOP_UNLOCK(vp); 917 vn_finished_write(mp); 918 return (error); 919 } 920 921 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 922 auio.uio_resid = bp->bio_length; 923 auio.uio_segflg = UIO_SYSSPACE; 924 auio.uio_td = td; 925 926 if (bp->bio_cmd == BIO_DELETE) { 927 /* 928 * Emulate BIO_DELETE by writing zeros. 929 */ 930 zerosize = ZERO_REGION_SIZE - 931 (ZERO_REGION_SIZE % sc->sectorsize); 932 auio.uio_iovcnt = howmany(bp->bio_length, zerosize); 933 piov = malloc(sizeof(*piov) * auio.uio_iovcnt, M_MD, M_WAITOK); 934 auio.uio_iov = piov; 935 while (len > 0) { 936 piov->iov_base = __DECONST(void *, zero_region); 937 piov->iov_len = len; 938 if (len > zerosize) 939 piov->iov_len = zerosize; 940 len -= piov->iov_len; 941 piov++; 942 } 943 piov = auio.uio_iov; 944 } else if ((bp->bio_flags & BIO_VLIST) != 0) { 945 piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK); 946 auio.uio_iov = piov; 947 vlist = (bus_dma_segment_t *)bp->bio_data; 948 while (len > 0) { 949 piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr + 950 ma_offs); 951 piov->iov_len = vlist->ds_len - ma_offs; 952 if (piov->iov_len > len) 953 piov->iov_len = len; 954 len -= piov->iov_len; 955 ma_offs = 0; 956 vlist++; 957 piov++; 958 } 959 auio.uio_iovcnt = piov - auio.uio_iov; 960 piov = auio.uio_iov; 961 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 962 pb = uma_zalloc(md_pbuf_zone, M_WAITOK); 963 bp->bio_resid = len; 964 unmapped_step: 965 npages = atop(min(MAXPHYS, round_page(len + (ma_offs & 966 PAGE_MASK)))); 967 iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len); 968 KASSERT(iolen > 0, ("zero iolen")); 969 pmap_qenter((vm_offset_t)pb->b_data, 970 &bp->bio_ma[atop(ma_offs)], npages); 971 aiov.iov_base = (void *)((vm_offset_t)pb->b_data + 972 (ma_offs & PAGE_MASK)); 973 aiov.iov_len = iolen; 974 auio.uio_iov = &aiov; 975 auio.uio_iovcnt = 1; 976 auio.uio_resid = iolen; 977 } else { 978 aiov.iov_base = bp->bio_data; 979 aiov.iov_len = bp->bio_length; 980 auio.uio_iov = &aiov; 981 auio.uio_iovcnt = 1; 982 } 983 iostart = auio.uio_offset; 984 if (auio.uio_rw == UIO_READ) { 985 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 986 error = VOP_READ(vp, &auio, 0, sc->cred); 987 VOP_UNLOCK(vp); 988 } else { 989 (void) vn_start_write(vp, &mp, V_WAIT); 990 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 991 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC, 992 sc->cred); 993 VOP_UNLOCK(vp); 994 vn_finished_write(mp); 995 if (error == 0) 996 sc->flags &= ~MD_VERIFY; 997 } 998 999 /* When MD_CACHE is set, try to avoid double-caching the data. */ 1000 if (error == 0 && (sc->flags & MD_CACHE) == 0) 1001 VOP_ADVISE(vp, iostart, auio.uio_offset - 1, 1002 POSIX_FADV_DONTNEED); 1003 1004 if (pb != NULL) { 1005 pmap_qremove((vm_offset_t)pb->b_data, npages); 1006 if (error == 0) { 1007 len -= iolen; 1008 bp->bio_resid -= iolen; 1009 ma_offs += iolen; 1010 if (len > 0) 1011 goto unmapped_step; 1012 } 1013 uma_zfree(md_pbuf_zone, pb); 1014 } 1015 1016 free(piov, M_MD); 1017 if (pb == NULL) 1018 bp->bio_resid = auio.uio_resid; 1019 return (error); 1020 } 1021 1022 static int 1023 mdstart_swap(struct md_s *sc, struct bio *bp) 1024 { 1025 vm_page_t m; 1026 u_char *p; 1027 vm_pindex_t i, lastp; 1028 bus_dma_segment_t *vlist; 1029 int rv, ma_offs, offs, len, lastend; 1030 1031 switch (bp->bio_cmd) { 1032 case BIO_READ: 1033 case BIO_WRITE: 1034 case BIO_DELETE: 1035 break; 1036 default: 1037 return (EOPNOTSUPP); 1038 } 1039 1040 p = bp->bio_data; 1041 ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ? 1042 bp->bio_ma_offset : 0; 1043 vlist = (bp->bio_flags & BIO_VLIST) != 0 ? 1044 (bus_dma_segment_t *)bp->bio_data : NULL; 1045 1046 /* 1047 * offs is the offset at which to start operating on the 1048 * next (ie, first) page. lastp is the last page on 1049 * which we're going to operate. lastend is the ending 1050 * position within that last page (ie, PAGE_SIZE if 1051 * we're operating on complete aligned pages). 1052 */ 1053 offs = bp->bio_offset % PAGE_SIZE; 1054 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE; 1055 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1; 1056 1057 rv = VM_PAGER_OK; 1058 vm_object_pip_add(sc->object, 1); 1059 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) { 1060 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs; 1061 m = vm_page_grab_unlocked(sc->object, i, VM_ALLOC_SYSTEM); 1062 if (bp->bio_cmd == BIO_READ) { 1063 if (vm_page_all_valid(m)) 1064 rv = VM_PAGER_OK; 1065 else 1066 rv = vm_pager_get_pages(sc->object, &m, 1, 1067 NULL, NULL); 1068 if (rv == VM_PAGER_ERROR) { 1069 VM_OBJECT_WLOCK(sc->object); 1070 vm_page_free(m); 1071 VM_OBJECT_WUNLOCK(sc->object); 1072 break; 1073 } else if (rv == VM_PAGER_FAIL) { 1074 /* 1075 * Pager does not have the page. Zero 1076 * the allocated page, and mark it as 1077 * valid. Do not set dirty, the page 1078 * can be recreated if thrown out. 1079 */ 1080 pmap_zero_page(m); 1081 vm_page_valid(m); 1082 } 1083 if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 1084 pmap_copy_pages(&m, offs, bp->bio_ma, 1085 ma_offs, len); 1086 } else if ((bp->bio_flags & BIO_VLIST) != 0) { 1087 physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs, 1088 vlist, ma_offs, len); 1089 cpu_flush_dcache(p, len); 1090 } else { 1091 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len); 1092 cpu_flush_dcache(p, len); 1093 } 1094 } else if (bp->bio_cmd == BIO_WRITE) { 1095 if (len == PAGE_SIZE || vm_page_all_valid(m)) 1096 rv = VM_PAGER_OK; 1097 else 1098 rv = vm_pager_get_pages(sc->object, &m, 1, 1099 NULL, NULL); 1100 if (rv == VM_PAGER_ERROR) { 1101 VM_OBJECT_WLOCK(sc->object); 1102 vm_page_free(m); 1103 VM_OBJECT_WUNLOCK(sc->object); 1104 break; 1105 } else if (rv == VM_PAGER_FAIL) 1106 pmap_zero_page(m); 1107 1108 if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 1109 pmap_copy_pages(bp->bio_ma, ma_offs, &m, 1110 offs, len); 1111 } else if ((bp->bio_flags & BIO_VLIST) != 0) { 1112 physcopyin_vlist(vlist, ma_offs, 1113 VM_PAGE_TO_PHYS(m) + offs, len); 1114 } else { 1115 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len); 1116 } 1117 1118 vm_page_valid(m); 1119 vm_page_set_dirty(m); 1120 } else if (bp->bio_cmd == BIO_DELETE) { 1121 if (len == PAGE_SIZE || vm_page_all_valid(m)) 1122 rv = VM_PAGER_OK; 1123 else 1124 rv = vm_pager_get_pages(sc->object, &m, 1, 1125 NULL, NULL); 1126 VM_OBJECT_WLOCK(sc->object); 1127 if (rv == VM_PAGER_ERROR) { 1128 vm_page_free(m); 1129 VM_OBJECT_WUNLOCK(sc->object); 1130 break; 1131 } else if (rv == VM_PAGER_FAIL) { 1132 vm_page_free(m); 1133 m = NULL; 1134 } else { 1135 /* Page is valid. */ 1136 if (len != PAGE_SIZE) { 1137 pmap_zero_page_area(m, offs, len); 1138 vm_page_set_dirty(m); 1139 } else { 1140 vm_pager_page_unswapped(m); 1141 vm_page_free(m); 1142 m = NULL; 1143 } 1144 } 1145 VM_OBJECT_WUNLOCK(sc->object); 1146 } 1147 if (m != NULL) { 1148 vm_page_xunbusy(m); 1149 1150 /* 1151 * The page may be deactivated prior to setting 1152 * PGA_REFERENCED, but in this case it will be 1153 * reactivated by the page daemon. 1154 */ 1155 if (vm_page_active(m)) 1156 vm_page_reference(m); 1157 else 1158 vm_page_activate(m); 1159 } 1160 1161 /* Actions on further pages start at offset 0 */ 1162 p += PAGE_SIZE - offs; 1163 offs = 0; 1164 ma_offs += len; 1165 } 1166 vm_object_pip_wakeup(sc->object); 1167 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC); 1168 } 1169 1170 static int 1171 mdstart_null(struct md_s *sc, struct bio *bp) 1172 { 1173 1174 switch (bp->bio_cmd) { 1175 case BIO_READ: 1176 bzero(bp->bio_data, bp->bio_length); 1177 cpu_flush_dcache(bp->bio_data, bp->bio_length); 1178 break; 1179 case BIO_WRITE: 1180 break; 1181 } 1182 bp->bio_resid = 0; 1183 return (0); 1184 } 1185 1186 static void 1187 md_kthread(void *arg) 1188 { 1189 struct md_s *sc; 1190 struct bio *bp; 1191 int error; 1192 1193 sc = arg; 1194 thread_lock(curthread); 1195 sched_prio(curthread, PRIBIO); 1196 thread_unlock(curthread); 1197 if (sc->type == MD_VNODE) 1198 curthread->td_pflags |= TDP_NORUNNINGBUF; 1199 1200 for (;;) { 1201 mtx_lock(&sc->queue_mtx); 1202 if (sc->flags & MD_SHUTDOWN) { 1203 sc->flags |= MD_EXITING; 1204 mtx_unlock(&sc->queue_mtx); 1205 kproc_exit(0); 1206 } 1207 bp = bioq_takefirst(&sc->bio_queue); 1208 if (!bp) { 1209 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 1210 continue; 1211 } 1212 mtx_unlock(&sc->queue_mtx); 1213 if (bp->bio_cmd == BIO_GETATTR) { 1214 int isv = ((sc->flags & MD_VERIFY) != 0); 1215 1216 if ((sc->fwsectors && sc->fwheads && 1217 (g_handleattr_int(bp, "GEOM::fwsectors", 1218 sc->fwsectors) || 1219 g_handleattr_int(bp, "GEOM::fwheads", 1220 sc->fwheads))) || 1221 g_handleattr_int(bp, "GEOM::candelete", 1)) 1222 error = -1; 1223 else if (sc->ident[0] != '\0' && 1224 g_handleattr_str(bp, "GEOM::ident", sc->ident)) 1225 error = -1; 1226 else if (g_handleattr_int(bp, "MNT::verified", isv)) 1227 error = -1; 1228 else 1229 error = EOPNOTSUPP; 1230 } else { 1231 error = sc->start(sc, bp); 1232 } 1233 1234 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) { 1235 /* 1236 * Devstat uses (bio_bcount, bio_resid) for 1237 * determining the length of the completed part of 1238 * the i/o. g_io_deliver() will translate from 1239 * bio_completed to that, but it also destroys the 1240 * bio so we must do our own translation. 1241 */ 1242 bp->bio_bcount = bp->bio_length; 1243 bp->bio_resid = (error == -1 ? bp->bio_bcount : 0); 1244 devstat_end_transaction_bio(sc->devstat, bp); 1245 } 1246 if (error != -1) { 1247 bp->bio_completed = bp->bio_length; 1248 g_io_deliver(bp, error); 1249 } 1250 } 1251 } 1252 1253 static struct md_s * 1254 mdfind(int unit) 1255 { 1256 struct md_s *sc; 1257 1258 LIST_FOREACH(sc, &md_softc_list, list) { 1259 if (sc->unit == unit) 1260 break; 1261 } 1262 return (sc); 1263 } 1264 1265 static struct md_s * 1266 mdnew(int unit, int *errp, enum md_types type) 1267 { 1268 struct md_s *sc; 1269 int error; 1270 1271 *errp = 0; 1272 if (unit == -1) 1273 unit = alloc_unr(md_uh); 1274 else 1275 unit = alloc_unr_specific(md_uh, unit); 1276 1277 if (unit == -1) { 1278 *errp = EBUSY; 1279 return (NULL); 1280 } 1281 1282 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 1283 sc->type = type; 1284 bioq_init(&sc->bio_queue); 1285 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 1286 sc->unit = unit; 1287 sprintf(sc->name, "md%d", unit); 1288 LIST_INSERT_HEAD(&md_softc_list, sc, list); 1289 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 1290 if (error == 0) 1291 return (sc); 1292 LIST_REMOVE(sc, list); 1293 mtx_destroy(&sc->queue_mtx); 1294 free_unr(md_uh, sc->unit); 1295 free(sc, M_MD); 1296 *errp = error; 1297 return (NULL); 1298 } 1299 1300 static void 1301 mdinit(struct md_s *sc) 1302 { 1303 struct g_geom *gp; 1304 struct g_provider *pp; 1305 1306 g_topology_lock(); 1307 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 1308 gp->softc = sc; 1309 pp = g_new_providerf(gp, "md%d", sc->unit); 1310 devstat_remove_entry(pp->stat); 1311 pp->stat = NULL; 1312 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE; 1313 pp->mediasize = sc->mediasize; 1314 pp->sectorsize = sc->sectorsize; 1315 switch (sc->type) { 1316 case MD_MALLOC: 1317 case MD_VNODE: 1318 case MD_SWAP: 1319 pp->flags |= G_PF_ACCEPT_UNMAPPED; 1320 break; 1321 case MD_PRELOAD: 1322 case MD_NULL: 1323 break; 1324 } 1325 sc->gp = gp; 1326 sc->pp = pp; 1327 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize, 1328 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX); 1329 sc->devstat->id = pp; 1330 g_error_provider(pp, 0); 1331 g_topology_unlock(); 1332 } 1333 1334 static int 1335 mdcreate_malloc(struct md_s *sc, struct md_req *mdr) 1336 { 1337 uintptr_t sp; 1338 int error; 1339 off_t u; 1340 1341 error = 0; 1342 if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 1343 return (EINVAL); 1344 if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize)) 1345 return (EINVAL); 1346 /* Compression doesn't make sense if we have reserved space */ 1347 if (mdr->md_options & MD_RESERVE) 1348 mdr->md_options &= ~MD_COMPRESS; 1349 if (mdr->md_fwsectors != 0) 1350 sc->fwsectors = mdr->md_fwsectors; 1351 if (mdr->md_fwheads != 0) 1352 sc->fwheads = mdr->md_fwheads; 1353 sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE); 1354 sc->indir = dimension(sc->mediasize / sc->sectorsize); 1355 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL, 1356 0x1ff, 0); 1357 if (mdr->md_options & MD_RESERVE) { 1358 off_t nsectors; 1359 1360 nsectors = sc->mediasize / sc->sectorsize; 1361 for (u = 0; u < nsectors; u++) { 1362 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ? 1363 M_WAITOK : M_NOWAIT) | M_ZERO); 1364 if (sp != 0) 1365 error = s_write(sc->indir, u, sp); 1366 else 1367 error = ENOMEM; 1368 if (error != 0) 1369 break; 1370 } 1371 } 1372 return (error); 1373 } 1374 1375 static int 1376 mdsetcred(struct md_s *sc, struct ucred *cred) 1377 { 1378 char *tmpbuf; 1379 int error = 0; 1380 1381 /* 1382 * Set credits in our softc 1383 */ 1384 1385 if (sc->cred) 1386 crfree(sc->cred); 1387 sc->cred = crhold(cred); 1388 1389 /* 1390 * Horrible kludge to establish credentials for NFS XXX. 1391 */ 1392 1393 if (sc->vnode) { 1394 struct uio auio; 1395 struct iovec aiov; 1396 1397 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK); 1398 bzero(&auio, sizeof(auio)); 1399 1400 aiov.iov_base = tmpbuf; 1401 aiov.iov_len = sc->sectorsize; 1402 auio.uio_iov = &aiov; 1403 auio.uio_iovcnt = 1; 1404 auio.uio_offset = 0; 1405 auio.uio_rw = UIO_READ; 1406 auio.uio_segflg = UIO_SYSSPACE; 1407 auio.uio_resid = aiov.iov_len; 1408 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 1409 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 1410 VOP_UNLOCK(sc->vnode); 1411 free(tmpbuf, M_TEMP); 1412 } 1413 return (error); 1414 } 1415 1416 static int 1417 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td) 1418 { 1419 struct vattr vattr; 1420 struct nameidata nd; 1421 char *fname; 1422 int error, flags; 1423 1424 fname = mdr->md_file; 1425 if (mdr->md_file_seg == UIO_USERSPACE) { 1426 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL); 1427 if (error != 0) 1428 return (error); 1429 } else if (mdr->md_file_seg == UIO_SYSSPACE) 1430 strlcpy(sc->file, fname, sizeof(sc->file)); 1431 else 1432 return (EDOOFUS); 1433 1434 /* 1435 * If the user specified that this is a read only device, don't 1436 * set the FWRITE mask before trying to open the backing store. 1437 */ 1438 flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \ 1439 | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0); 1440 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td); 1441 error = vn_open(&nd, &flags, 0, NULL); 1442 if (error != 0) 1443 return (error); 1444 NDFREE(&nd, NDF_ONLY_PNBUF); 1445 if (nd.ni_vp->v_type != VREG) { 1446 error = EINVAL; 1447 goto bad; 1448 } 1449 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred); 1450 if (error != 0) 1451 goto bad; 1452 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) { 1453 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY); 1454 if (VN_IS_DOOMED(nd.ni_vp)) { 1455 /* Forced unmount. */ 1456 error = EBADF; 1457 goto bad; 1458 } 1459 } 1460 nd.ni_vp->v_vflag |= VV_MD; 1461 VOP_UNLOCK(nd.ni_vp); 1462 1463 if (mdr->md_fwsectors != 0) 1464 sc->fwsectors = mdr->md_fwsectors; 1465 if (mdr->md_fwheads != 0) 1466 sc->fwheads = mdr->md_fwheads; 1467 snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju", 1468 (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid); 1469 sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE | 1470 MD_VERIFY); 1471 if (!(flags & FWRITE)) 1472 sc->flags |= MD_READONLY; 1473 sc->vnode = nd.ni_vp; 1474 1475 error = mdsetcred(sc, td->td_ucred); 1476 if (error != 0) { 1477 sc->vnode = NULL; 1478 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY); 1479 nd.ni_vp->v_vflag &= ~VV_MD; 1480 goto bad; 1481 } 1482 return (0); 1483 bad: 1484 VOP_UNLOCK(nd.ni_vp); 1485 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td); 1486 return (error); 1487 } 1488 1489 static void 1490 g_md_providergone(struct g_provider *pp) 1491 { 1492 struct md_s *sc = pp->geom->softc; 1493 1494 mtx_lock(&sc->queue_mtx); 1495 sc->flags |= MD_PROVIDERGONE; 1496 wakeup(&sc->flags); 1497 mtx_unlock(&sc->queue_mtx); 1498 } 1499 1500 static int 1501 mddestroy(struct md_s *sc, struct thread *td) 1502 { 1503 1504 if (sc->gp) { 1505 g_topology_lock(); 1506 g_wither_geom(sc->gp, ENXIO); 1507 g_topology_unlock(); 1508 1509 mtx_lock(&sc->queue_mtx); 1510 while (!(sc->flags & MD_PROVIDERGONE)) 1511 msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0); 1512 mtx_unlock(&sc->queue_mtx); 1513 } 1514 if (sc->devstat) { 1515 devstat_remove_entry(sc->devstat); 1516 sc->devstat = NULL; 1517 } 1518 mtx_lock(&sc->queue_mtx); 1519 sc->flags |= MD_SHUTDOWN; 1520 wakeup(sc); 1521 while (!(sc->flags & MD_EXITING)) 1522 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10); 1523 mtx_unlock(&sc->queue_mtx); 1524 mtx_destroy(&sc->queue_mtx); 1525 if (sc->vnode != NULL) { 1526 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 1527 sc->vnode->v_vflag &= ~VV_MD; 1528 VOP_UNLOCK(sc->vnode); 1529 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 1530 FREAD : (FREAD|FWRITE), sc->cred, td); 1531 } 1532 if (sc->cred != NULL) 1533 crfree(sc->cred); 1534 if (sc->object != NULL) 1535 vm_object_deallocate(sc->object); 1536 if (sc->indir) 1537 destroy_indir(sc, sc->indir); 1538 if (sc->uma) 1539 uma_zdestroy(sc->uma); 1540 1541 LIST_REMOVE(sc, list); 1542 free_unr(md_uh, sc->unit); 1543 free(sc, M_MD); 1544 return (0); 1545 } 1546 1547 static int 1548 mdresize(struct md_s *sc, struct md_req *mdr) 1549 { 1550 int error, res; 1551 vm_pindex_t oldpages, newpages; 1552 1553 switch (sc->type) { 1554 case MD_VNODE: 1555 case MD_NULL: 1556 break; 1557 case MD_SWAP: 1558 if (mdr->md_mediasize <= 0 || 1559 (mdr->md_mediasize % PAGE_SIZE) != 0) 1560 return (EDOM); 1561 oldpages = OFF_TO_IDX(sc->mediasize); 1562 newpages = OFF_TO_IDX(mdr->md_mediasize); 1563 if (newpages < oldpages) { 1564 VM_OBJECT_WLOCK(sc->object); 1565 vm_object_page_remove(sc->object, newpages, 0, 0); 1566 swap_release_by_cred(IDX_TO_OFF(oldpages - 1567 newpages), sc->cred); 1568 sc->object->charge = IDX_TO_OFF(newpages); 1569 sc->object->size = newpages; 1570 VM_OBJECT_WUNLOCK(sc->object); 1571 } else if (newpages > oldpages) { 1572 res = swap_reserve_by_cred(IDX_TO_OFF(newpages - 1573 oldpages), sc->cred); 1574 if (!res) 1575 return (ENOMEM); 1576 if ((mdr->md_options & MD_RESERVE) || 1577 (sc->flags & MD_RESERVE)) { 1578 error = swap_pager_reserve(sc->object, 1579 oldpages, newpages - oldpages); 1580 if (error < 0) { 1581 swap_release_by_cred( 1582 IDX_TO_OFF(newpages - oldpages), 1583 sc->cred); 1584 return (EDOM); 1585 } 1586 } 1587 VM_OBJECT_WLOCK(sc->object); 1588 sc->object->charge = IDX_TO_OFF(newpages); 1589 sc->object->size = newpages; 1590 VM_OBJECT_WUNLOCK(sc->object); 1591 } 1592 break; 1593 default: 1594 return (EOPNOTSUPP); 1595 } 1596 1597 sc->mediasize = mdr->md_mediasize; 1598 g_topology_lock(); 1599 g_resize_provider(sc->pp, sc->mediasize); 1600 g_topology_unlock(); 1601 return (0); 1602 } 1603 1604 static int 1605 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td) 1606 { 1607 vm_ooffset_t npage; 1608 int error; 1609 1610 /* 1611 * Range check. Disallow negative sizes and sizes not being 1612 * multiple of page size. 1613 */ 1614 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0) 1615 return (EDOM); 1616 1617 /* 1618 * Allocate an OBJT_SWAP object. 1619 * 1620 * Note the truncation. 1621 */ 1622 1623 if ((mdr->md_options & MD_VERIFY) != 0) 1624 return (EINVAL); 1625 npage = mdr->md_mediasize / PAGE_SIZE; 1626 if (mdr->md_fwsectors != 0) 1627 sc->fwsectors = mdr->md_fwsectors; 1628 if (mdr->md_fwheads != 0) 1629 sc->fwheads = mdr->md_fwheads; 1630 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage, 1631 VM_PROT_DEFAULT, 0, td->td_ucred); 1632 if (sc->object == NULL) 1633 return (ENOMEM); 1634 sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE); 1635 if (mdr->md_options & MD_RESERVE) { 1636 if (swap_pager_reserve(sc->object, 0, npage) < 0) { 1637 error = EDOM; 1638 goto finish; 1639 } 1640 } 1641 error = mdsetcred(sc, td->td_ucred); 1642 finish: 1643 if (error != 0) { 1644 vm_object_deallocate(sc->object); 1645 sc->object = NULL; 1646 } 1647 return (error); 1648 } 1649 1650 static int 1651 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td) 1652 { 1653 1654 /* 1655 * Range check. Disallow negative sizes and sizes not being 1656 * multiple of page size. 1657 */ 1658 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0) 1659 return (EDOM); 1660 1661 return (0); 1662 } 1663 1664 static int 1665 kern_mdattach_locked(struct thread *td, struct md_req *mdr) 1666 { 1667 struct md_s *sc; 1668 unsigned sectsize; 1669 int error, i; 1670 1671 sx_assert(&md_sx, SA_XLOCKED); 1672 1673 switch (mdr->md_type) { 1674 case MD_MALLOC: 1675 case MD_PRELOAD: 1676 case MD_VNODE: 1677 case MD_SWAP: 1678 case MD_NULL: 1679 break; 1680 default: 1681 return (EINVAL); 1682 } 1683 if (mdr->md_sectorsize == 0) 1684 sectsize = DEV_BSIZE; 1685 else 1686 sectsize = mdr->md_sectorsize; 1687 if (sectsize > MAXPHYS || mdr->md_mediasize < sectsize) 1688 return (EINVAL); 1689 if (mdr->md_options & MD_AUTOUNIT) 1690 sc = mdnew(-1, &error, mdr->md_type); 1691 else { 1692 if (mdr->md_unit > INT_MAX) 1693 return (EINVAL); 1694 sc = mdnew(mdr->md_unit, &error, mdr->md_type); 1695 } 1696 if (sc == NULL) 1697 return (error); 1698 if (mdr->md_label != NULL) 1699 error = copyinstr(mdr->md_label, sc->label, 1700 sizeof(sc->label), NULL); 1701 if (error != 0) 1702 goto err_after_new; 1703 if (mdr->md_options & MD_AUTOUNIT) 1704 mdr->md_unit = sc->unit; 1705 sc->mediasize = mdr->md_mediasize; 1706 sc->sectorsize = sectsize; 1707 error = EDOOFUS; 1708 switch (sc->type) { 1709 case MD_MALLOC: 1710 sc->start = mdstart_malloc; 1711 error = mdcreate_malloc(sc, mdr); 1712 break; 1713 case MD_PRELOAD: 1714 /* 1715 * We disallow attaching preloaded memory disks via 1716 * ioctl. Preloaded memory disks are automatically 1717 * attached in g_md_init(). 1718 */ 1719 error = EOPNOTSUPP; 1720 break; 1721 case MD_VNODE: 1722 sc->start = mdstart_vnode; 1723 error = mdcreate_vnode(sc, mdr, td); 1724 break; 1725 case MD_SWAP: 1726 sc->start = mdstart_swap; 1727 error = mdcreate_swap(sc, mdr, td); 1728 break; 1729 case MD_NULL: 1730 sc->start = mdstart_null; 1731 error = mdcreate_null(sc, mdr, td); 1732 break; 1733 } 1734 err_after_new: 1735 if (error != 0) { 1736 mddestroy(sc, td); 1737 return (error); 1738 } 1739 1740 /* Prune off any residual fractional sector */ 1741 i = sc->mediasize % sc->sectorsize; 1742 sc->mediasize -= i; 1743 1744 mdinit(sc); 1745 return (0); 1746 } 1747 1748 static int 1749 kern_mdattach(struct thread *td, struct md_req *mdr) 1750 { 1751 int error; 1752 1753 sx_xlock(&md_sx); 1754 error = kern_mdattach_locked(td, mdr); 1755 sx_xunlock(&md_sx); 1756 return (error); 1757 } 1758 1759 static int 1760 kern_mddetach_locked(struct thread *td, struct md_req *mdr) 1761 { 1762 struct md_s *sc; 1763 1764 sx_assert(&md_sx, SA_XLOCKED); 1765 1766 if (mdr->md_mediasize != 0 || 1767 (mdr->md_options & ~MD_FORCE) != 0) 1768 return (EINVAL); 1769 1770 sc = mdfind(mdr->md_unit); 1771 if (sc == NULL) 1772 return (ENOENT); 1773 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) && 1774 !(mdr->md_options & MD_FORCE)) 1775 return (EBUSY); 1776 return (mddestroy(sc, td)); 1777 } 1778 1779 static int 1780 kern_mddetach(struct thread *td, struct md_req *mdr) 1781 { 1782 int error; 1783 1784 sx_xlock(&md_sx); 1785 error = kern_mddetach_locked(td, mdr); 1786 sx_xunlock(&md_sx); 1787 return (error); 1788 } 1789 1790 static int 1791 kern_mdresize_locked(struct md_req *mdr) 1792 { 1793 struct md_s *sc; 1794 1795 sx_assert(&md_sx, SA_XLOCKED); 1796 1797 if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0) 1798 return (EINVAL); 1799 1800 sc = mdfind(mdr->md_unit); 1801 if (sc == NULL) 1802 return (ENOENT); 1803 if (mdr->md_mediasize < sc->sectorsize) 1804 return (EINVAL); 1805 if (mdr->md_mediasize < sc->mediasize && 1806 !(sc->flags & MD_FORCE) && 1807 !(mdr->md_options & MD_FORCE)) 1808 return (EBUSY); 1809 return (mdresize(sc, mdr)); 1810 } 1811 1812 static int 1813 kern_mdresize(struct md_req *mdr) 1814 { 1815 int error; 1816 1817 sx_xlock(&md_sx); 1818 error = kern_mdresize_locked(mdr); 1819 sx_xunlock(&md_sx); 1820 return (error); 1821 } 1822 1823 static int 1824 kern_mdquery_locked(struct md_req *mdr) 1825 { 1826 struct md_s *sc; 1827 int error; 1828 1829 sx_assert(&md_sx, SA_XLOCKED); 1830 1831 sc = mdfind(mdr->md_unit); 1832 if (sc == NULL) 1833 return (ENOENT); 1834 mdr->md_type = sc->type; 1835 mdr->md_options = sc->flags; 1836 mdr->md_mediasize = sc->mediasize; 1837 mdr->md_sectorsize = sc->sectorsize; 1838 error = 0; 1839 if (mdr->md_label != NULL) { 1840 error = copyout(sc->label, mdr->md_label, 1841 strlen(sc->label) + 1); 1842 if (error != 0) 1843 return (error); 1844 } 1845 if (sc->type == MD_VNODE || 1846 (sc->type == MD_PRELOAD && mdr->md_file != NULL)) 1847 error = copyout(sc->file, mdr->md_file, 1848 strlen(sc->file) + 1); 1849 return (error); 1850 } 1851 1852 static int 1853 kern_mdquery(struct md_req *mdr) 1854 { 1855 int error; 1856 1857 sx_xlock(&md_sx); 1858 error = kern_mdquery_locked(mdr); 1859 sx_xunlock(&md_sx); 1860 return (error); 1861 } 1862 1863 /* Copy members that are not userspace pointers. */ 1864 #define MD_IOCTL2REQ(mdio, mdr) do { \ 1865 (mdr)->md_unit = (mdio)->md_unit; \ 1866 (mdr)->md_type = (mdio)->md_type; \ 1867 (mdr)->md_mediasize = (mdio)->md_mediasize; \ 1868 (mdr)->md_sectorsize = (mdio)->md_sectorsize; \ 1869 (mdr)->md_options = (mdio)->md_options; \ 1870 (mdr)->md_fwheads = (mdio)->md_fwheads; \ 1871 (mdr)->md_fwsectors = (mdio)->md_fwsectors; \ 1872 (mdr)->md_units = &(mdio)->md_pad[0]; \ 1873 (mdr)->md_units_nitems = nitems((mdio)->md_pad); \ 1874 } while(0) 1875 1876 /* Copy members that might have been updated */ 1877 #define MD_REQ2IOCTL(mdr, mdio) do { \ 1878 (mdio)->md_unit = (mdr)->md_unit; \ 1879 (mdio)->md_type = (mdr)->md_type; \ 1880 (mdio)->md_mediasize = (mdr)->md_mediasize; \ 1881 (mdio)->md_sectorsize = (mdr)->md_sectorsize; \ 1882 (mdio)->md_options = (mdr)->md_options; \ 1883 (mdio)->md_fwheads = (mdr)->md_fwheads; \ 1884 (mdio)->md_fwsectors = (mdr)->md_fwsectors; \ 1885 } while(0) 1886 1887 static int 1888 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, 1889 struct thread *td) 1890 { 1891 struct md_req mdr; 1892 int error; 1893 1894 if (md_debug) 1895 printf("mdctlioctl(%s %lx %p %x %p)\n", 1896 devtoname(dev), cmd, addr, flags, td); 1897 1898 bzero(&mdr, sizeof(mdr)); 1899 switch (cmd) { 1900 case MDIOCATTACH: 1901 case MDIOCDETACH: 1902 case MDIOCRESIZE: 1903 case MDIOCQUERY: { 1904 struct md_ioctl *mdio = (struct md_ioctl *)addr; 1905 if (mdio->md_version != MDIOVERSION) 1906 return (EINVAL); 1907 MD_IOCTL2REQ(mdio, &mdr); 1908 mdr.md_file = mdio->md_file; 1909 mdr.md_file_seg = UIO_USERSPACE; 1910 /* If the file is adjacent to the md_ioctl it's in kernel. */ 1911 if ((void *)mdio->md_file == (void *)(mdio + 1)) 1912 mdr.md_file_seg = UIO_SYSSPACE; 1913 mdr.md_label = mdio->md_label; 1914 break; 1915 } 1916 #ifdef COMPAT_FREEBSD32 1917 case MDIOCATTACH_32: 1918 case MDIOCDETACH_32: 1919 case MDIOCRESIZE_32: 1920 case MDIOCQUERY_32: { 1921 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr; 1922 if (mdio->md_version != MDIOVERSION) 1923 return (EINVAL); 1924 MD_IOCTL2REQ(mdio, &mdr); 1925 mdr.md_file = (void *)(uintptr_t)mdio->md_file; 1926 mdr.md_file_seg = UIO_USERSPACE; 1927 mdr.md_label = (void *)(uintptr_t)mdio->md_label; 1928 break; 1929 } 1930 #endif 1931 default: 1932 /* Fall through to handler switch. */ 1933 break; 1934 } 1935 1936 error = 0; 1937 switch (cmd) { 1938 case MDIOCATTACH: 1939 #ifdef COMPAT_FREEBSD32 1940 case MDIOCATTACH_32: 1941 #endif 1942 error = kern_mdattach(td, &mdr); 1943 break; 1944 case MDIOCDETACH: 1945 #ifdef COMPAT_FREEBSD32 1946 case MDIOCDETACH_32: 1947 #endif 1948 error = kern_mddetach(td, &mdr); 1949 break; 1950 case MDIOCRESIZE: 1951 #ifdef COMPAT_FREEBSD32 1952 case MDIOCRESIZE_32: 1953 #endif 1954 error = kern_mdresize(&mdr); 1955 break; 1956 case MDIOCQUERY: 1957 #ifdef COMPAT_FREEBSD32 1958 case MDIOCQUERY_32: 1959 #endif 1960 error = kern_mdquery(&mdr); 1961 break; 1962 default: 1963 error = ENOIOCTL; 1964 } 1965 1966 switch (cmd) { 1967 case MDIOCATTACH: 1968 case MDIOCQUERY: { 1969 struct md_ioctl *mdio = (struct md_ioctl *)addr; 1970 MD_REQ2IOCTL(&mdr, mdio); 1971 break; 1972 } 1973 #ifdef COMPAT_FREEBSD32 1974 case MDIOCATTACH_32: 1975 case MDIOCQUERY_32: { 1976 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr; 1977 MD_REQ2IOCTL(&mdr, mdio); 1978 break; 1979 } 1980 #endif 1981 default: 1982 /* Other commands to not alter mdr. */ 1983 break; 1984 } 1985 1986 return (error); 1987 } 1988 1989 static void 1990 md_preloaded(u_char *image, size_t length, const char *name) 1991 { 1992 struct md_s *sc; 1993 int error; 1994 1995 sc = mdnew(-1, &error, MD_PRELOAD); 1996 if (sc == NULL) 1997 return; 1998 sc->mediasize = length; 1999 sc->sectorsize = DEV_BSIZE; 2000 sc->pl_ptr = image; 2001 sc->pl_len = length; 2002 sc->start = mdstart_preload; 2003 if (name != NULL) 2004 strlcpy(sc->file, name, sizeof(sc->file)); 2005 #ifdef MD_ROOT 2006 if (sc->unit == 0) { 2007 #ifndef ROOTDEVNAME 2008 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0"; 2009 #endif 2010 #ifdef MD_ROOT_READONLY 2011 sc->flags |= MD_READONLY; 2012 #endif 2013 } 2014 #endif 2015 mdinit(sc); 2016 if (name != NULL) { 2017 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n", 2018 MD_NAME, sc->unit, name, length, image); 2019 } else { 2020 printf("%s%d: Embedded image %zd bytes at %p\n", 2021 MD_NAME, sc->unit, length, image); 2022 } 2023 } 2024 2025 static void 2026 g_md_init(struct g_class *mp __unused) 2027 { 2028 caddr_t mod; 2029 u_char *ptr, *name, *type; 2030 unsigned len; 2031 int i; 2032 2033 /* figure out log2(NINDIR) */ 2034 for (i = NINDIR, nshift = -1; i; nshift++) 2035 i >>= 1; 2036 2037 mod = NULL; 2038 sx_init(&md_sx, "MD config lock"); 2039 g_topology_unlock(); 2040 md_uh = new_unrhdr(0, INT_MAX, NULL); 2041 #ifdef MD_ROOT 2042 if (mfs_root_size != 0) { 2043 sx_xlock(&md_sx); 2044 #ifdef MD_ROOT_MEM 2045 md_preloaded(mfs_root, mfs_root_size, NULL); 2046 #else 2047 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size, 2048 NULL); 2049 #endif 2050 sx_xunlock(&md_sx); 2051 } 2052 #endif 2053 /* XXX: are preload_* static or do they need Giant ? */ 2054 while ((mod = preload_search_next_name(mod)) != NULL) { 2055 name = (char *)preload_search_info(mod, MODINFO_NAME); 2056 if (name == NULL) 2057 continue; 2058 type = (char *)preload_search_info(mod, MODINFO_TYPE); 2059 if (type == NULL) 2060 continue; 2061 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 2062 continue; 2063 ptr = preload_fetch_addr(mod); 2064 len = preload_fetch_size(mod); 2065 if (ptr != NULL && len != 0) { 2066 sx_xlock(&md_sx); 2067 md_preloaded(ptr, len, name); 2068 sx_xunlock(&md_sx); 2069 } 2070 } 2071 md_pbuf_zone = pbuf_zsecond_create("mdpbuf", nswbuf / 10); 2072 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL, 2073 0600, MDCTL_NAME); 2074 g_topology_lock(); 2075 } 2076 2077 static void 2078 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2079 struct g_consumer *cp __unused, struct g_provider *pp) 2080 { 2081 struct md_s *mp; 2082 char *type; 2083 2084 mp = gp->softc; 2085 if (mp == NULL) 2086 return; 2087 2088 switch (mp->type) { 2089 case MD_MALLOC: 2090 type = "malloc"; 2091 break; 2092 case MD_PRELOAD: 2093 type = "preload"; 2094 break; 2095 case MD_VNODE: 2096 type = "vnode"; 2097 break; 2098 case MD_SWAP: 2099 type = "swap"; 2100 break; 2101 case MD_NULL: 2102 type = "null"; 2103 break; 2104 default: 2105 type = "unknown"; 2106 break; 2107 } 2108 2109 if (pp != NULL) { 2110 if (indent == NULL) { 2111 sbuf_printf(sb, " u %d", mp->unit); 2112 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize); 2113 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads); 2114 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors); 2115 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize); 2116 sbuf_printf(sb, " t %s", type); 2117 if ((mp->type == MD_VNODE && mp->vnode != NULL) || 2118 (mp->type == MD_PRELOAD && mp->file[0] != '\0')) 2119 sbuf_printf(sb, " file %s", mp->file); 2120 sbuf_printf(sb, " label %s", mp->label); 2121 } else { 2122 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, 2123 mp->unit); 2124 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n", 2125 indent, (uintmax_t) mp->sectorsize); 2126 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n", 2127 indent, (uintmax_t) mp->fwheads); 2128 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n", 2129 indent, (uintmax_t) mp->fwsectors); 2130 if (mp->ident[0] != '\0') { 2131 sbuf_printf(sb, "%s<ident>", indent); 2132 g_conf_printf_escaped(sb, "%s", mp->ident); 2133 sbuf_printf(sb, "</ident>\n"); 2134 } 2135 sbuf_printf(sb, "%s<length>%ju</length>\n", 2136 indent, (uintmax_t) mp->mediasize); 2137 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent, 2138 (mp->flags & MD_COMPRESS) == 0 ? "off": "on"); 2139 sbuf_printf(sb, "%s<access>%s</access>\n", indent, 2140 (mp->flags & MD_READONLY) == 0 ? "read-write": 2141 "read-only"); 2142 sbuf_printf(sb, "%s<type>%s</type>\n", indent, 2143 type); 2144 if ((mp->type == MD_VNODE && mp->vnode != NULL) || 2145 (mp->type == MD_PRELOAD && mp->file[0] != '\0')) { 2146 sbuf_printf(sb, "%s<file>", indent); 2147 g_conf_printf_escaped(sb, "%s", mp->file); 2148 sbuf_printf(sb, "</file>\n"); 2149 } 2150 if (mp->type == MD_VNODE) 2151 sbuf_printf(sb, "%s<cache>%s</cache>\n", indent, 2152 (mp->flags & MD_CACHE) == 0 ? "off": "on"); 2153 sbuf_printf(sb, "%s<label>", indent); 2154 g_conf_printf_escaped(sb, "%s", mp->label); 2155 sbuf_printf(sb, "</label>\n"); 2156 } 2157 } 2158 } 2159 2160 static void 2161 g_md_fini(struct g_class *mp __unused) 2162 { 2163 2164 sx_destroy(&md_sx); 2165 if (status_dev != NULL) 2166 destroy_dev(status_dev); 2167 uma_zdestroy(md_pbuf_zone); 2168 delete_unrhdr(md_uh); 2169 } 2170