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 in 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); 228 229 230 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list); 231 232 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 233 #define NMASK (NINDIR-1) 234 static int nshift; 235 236 static uma_zone_t md_pbuf_zone; 237 238 struct indir { 239 uintptr_t *array; 240 u_int total; 241 u_int used; 242 u_int shift; 243 }; 244 245 struct md_s { 246 int unit; 247 LIST_ENTRY(md_s) list; 248 struct bio_queue_head bio_queue; 249 struct mtx queue_mtx; 250 struct cdev *dev; 251 enum md_types type; 252 off_t mediasize; 253 unsigned sectorsize; 254 unsigned opencount; 255 unsigned fwheads; 256 unsigned fwsectors; 257 char ident[32]; 258 unsigned flags; 259 char name[20]; 260 struct proc *procp; 261 struct g_geom *gp; 262 struct g_provider *pp; 263 int (*start)(struct md_s *sc, struct bio *bp); 264 struct devstat *devstat; 265 266 /* MD_MALLOC related fields */ 267 struct indir *indir; 268 uma_zone_t uma; 269 270 /* MD_PRELOAD related fields */ 271 u_char *pl_ptr; 272 size_t pl_len; 273 274 /* MD_VNODE related fields */ 275 struct vnode *vnode; 276 char file[PATH_MAX]; 277 char label[PATH_MAX]; 278 struct ucred *cred; 279 280 /* MD_SWAP related fields */ 281 vm_object_t object; 282 }; 283 284 static struct indir * 285 new_indir(u_int shift) 286 { 287 struct indir *ip; 288 289 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT) 290 | M_ZERO); 291 if (ip == NULL) 292 return (NULL); 293 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 294 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO); 295 if (ip->array == NULL) { 296 free(ip, M_MD); 297 return (NULL); 298 } 299 ip->total = NINDIR; 300 ip->shift = shift; 301 return (ip); 302 } 303 304 static void 305 del_indir(struct indir *ip) 306 { 307 308 free(ip->array, M_MDSECT); 309 free(ip, M_MD); 310 } 311 312 static void 313 destroy_indir(struct md_s *sc, struct indir *ip) 314 { 315 int i; 316 317 for (i = 0; i < NINDIR; i++) { 318 if (!ip->array[i]) 319 continue; 320 if (ip->shift) 321 destroy_indir(sc, (struct indir*)(ip->array[i])); 322 else if (ip->array[i] > 255) 323 uma_zfree(sc->uma, (void *)(ip->array[i])); 324 } 325 del_indir(ip); 326 } 327 328 /* 329 * This function does the math and allocates the top level "indir" structure 330 * for a device of "size" sectors. 331 */ 332 333 static struct indir * 334 dimension(off_t size) 335 { 336 off_t rcnt; 337 struct indir *ip; 338 int layer; 339 340 rcnt = size; 341 layer = 0; 342 while (rcnt > NINDIR) { 343 rcnt /= NINDIR; 344 layer++; 345 } 346 347 /* 348 * XXX: the top layer is probably not fully populated, so we allocate 349 * too much space for ip->array in here. 350 */ 351 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 352 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 353 M_MDSECT, M_WAITOK | M_ZERO); 354 ip->total = NINDIR; 355 ip->shift = layer * nshift; 356 return (ip); 357 } 358 359 /* 360 * Read a given sector 361 */ 362 363 static uintptr_t 364 s_read(struct indir *ip, off_t offset) 365 { 366 struct indir *cip; 367 int idx; 368 uintptr_t up; 369 370 if (md_debug > 1) 371 printf("s_read(%jd)\n", (intmax_t)offset); 372 up = 0; 373 for (cip = ip; cip != NULL;) { 374 if (cip->shift) { 375 idx = (offset >> cip->shift) & NMASK; 376 up = cip->array[idx]; 377 cip = (struct indir *)up; 378 continue; 379 } 380 idx = offset & NMASK; 381 return (cip->array[idx]); 382 } 383 return (0); 384 } 385 386 /* 387 * Write a given sector, prune the tree if the value is 0 388 */ 389 390 static int 391 s_write(struct indir *ip, off_t offset, uintptr_t ptr) 392 { 393 struct indir *cip, *lip[10]; 394 int idx, li; 395 uintptr_t up; 396 397 if (md_debug > 1) 398 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 399 up = 0; 400 li = 0; 401 cip = ip; 402 for (;;) { 403 lip[li++] = cip; 404 if (cip->shift) { 405 idx = (offset >> cip->shift) & NMASK; 406 up = cip->array[idx]; 407 if (up != 0) { 408 cip = (struct indir *)up; 409 continue; 410 } 411 /* Allocate branch */ 412 cip->array[idx] = 413 (uintptr_t)new_indir(cip->shift - nshift); 414 if (cip->array[idx] == 0) 415 return (ENOSPC); 416 cip->used++; 417 up = cip->array[idx]; 418 cip = (struct indir *)up; 419 continue; 420 } 421 /* leafnode */ 422 idx = offset & NMASK; 423 up = cip->array[idx]; 424 if (up != 0) 425 cip->used--; 426 cip->array[idx] = ptr; 427 if (ptr != 0) 428 cip->used++; 429 break; 430 } 431 if (cip->used != 0 || li == 1) 432 return (0); 433 li--; 434 while (cip->used == 0 && cip != ip) { 435 li--; 436 idx = (offset >> lip[li]->shift) & NMASK; 437 up = lip[li]->array[idx]; 438 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 439 del_indir(cip); 440 lip[li]->array[idx] = 0; 441 lip[li]->used--; 442 cip = lip[li]; 443 } 444 return (0); 445 } 446 447 448 static int 449 g_md_access(struct g_provider *pp, int r, int w, int e) 450 { 451 struct md_s *sc; 452 453 sc = pp->geom->softc; 454 if (sc == NULL) { 455 if (r <= 0 && w <= 0 && e <= 0) 456 return (0); 457 return (ENXIO); 458 } 459 r += pp->acr; 460 w += pp->acw; 461 e += pp->ace; 462 if ((sc->flags & MD_READONLY) != 0 && w > 0) 463 return (EROFS); 464 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 465 sc->opencount = 1; 466 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 467 sc->opencount = 0; 468 } 469 return (0); 470 } 471 472 static void 473 g_md_start(struct bio *bp) 474 { 475 struct md_s *sc; 476 477 sc = bp->bio_to->geom->softc; 478 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) { 479 devstat_start_transaction_bio(sc->devstat, bp); 480 } 481 mtx_lock(&sc->queue_mtx); 482 bioq_disksort(&sc->bio_queue, bp); 483 wakeup(sc); 484 mtx_unlock(&sc->queue_mtx); 485 } 486 487 #define MD_MALLOC_MOVE_ZERO 1 488 #define MD_MALLOC_MOVE_FILL 2 489 #define MD_MALLOC_MOVE_READ 3 490 #define MD_MALLOC_MOVE_WRITE 4 491 #define MD_MALLOC_MOVE_CMP 5 492 493 static int 494 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize, 495 void *ptr, u_char fill, int op) 496 { 497 struct sf_buf *sf; 498 vm_page_t m, *mp1; 499 char *p, first; 500 off_t *uc; 501 unsigned n; 502 int error, i, ma_offs1, sz, first_read; 503 504 m = NULL; 505 error = 0; 506 sf = NULL; 507 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */ 508 first = 0; 509 first_read = 0; 510 uc = ptr; 511 mp1 = *mp; 512 ma_offs1 = *ma_offs; 513 /* } */ 514 sched_pin(); 515 for (n = sectorsize; n != 0; n -= sz) { 516 sz = imin(PAGE_SIZE - *ma_offs, n); 517 if (m != **mp) { 518 if (sf != NULL) 519 sf_buf_free(sf); 520 m = **mp; 521 sf = sf_buf_alloc(m, SFB_CPUPRIVATE | 522 (md_malloc_wait ? 0 : SFB_NOWAIT)); 523 if (sf == NULL) { 524 error = ENOMEM; 525 break; 526 } 527 } 528 p = (char *)sf_buf_kva(sf) + *ma_offs; 529 switch (op) { 530 case MD_MALLOC_MOVE_ZERO: 531 bzero(p, sz); 532 break; 533 case MD_MALLOC_MOVE_FILL: 534 memset(p, fill, sz); 535 break; 536 case MD_MALLOC_MOVE_READ: 537 bcopy(ptr, p, sz); 538 cpu_flush_dcache(p, sz); 539 break; 540 case MD_MALLOC_MOVE_WRITE: 541 bcopy(p, ptr, sz); 542 break; 543 case MD_MALLOC_MOVE_CMP: 544 for (i = 0; i < sz; i++, p++) { 545 if (!first_read) { 546 *uc = (u_char)*p; 547 first = *p; 548 first_read = 1; 549 } else if (*p != first) { 550 error = EDOOFUS; 551 break; 552 } 553 } 554 break; 555 default: 556 KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op)); 557 break; 558 } 559 if (error != 0) 560 break; 561 *ma_offs += sz; 562 *ma_offs %= PAGE_SIZE; 563 if (*ma_offs == 0) 564 (*mp)++; 565 ptr = (char *)ptr + sz; 566 } 567 568 if (sf != NULL) 569 sf_buf_free(sf); 570 sched_unpin(); 571 if (op == MD_MALLOC_MOVE_CMP && error != 0) { 572 *mp = mp1; 573 *ma_offs = ma_offs1; 574 } 575 return (error); 576 } 577 578 static int 579 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs, 580 unsigned len, void *ptr, u_char fill, int op) 581 { 582 bus_dma_segment_t *vlist; 583 uint8_t *p, *end, first; 584 off_t *uc; 585 int ma_offs, seg_len; 586 587 vlist = *pvlist; 588 ma_offs = *pma_offs; 589 uc = ptr; 590 591 for (; len != 0; len -= seg_len) { 592 seg_len = imin(vlist->ds_len - ma_offs, len); 593 p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs; 594 switch (op) { 595 case MD_MALLOC_MOVE_ZERO: 596 bzero(p, seg_len); 597 break; 598 case MD_MALLOC_MOVE_FILL: 599 memset(p, fill, seg_len); 600 break; 601 case MD_MALLOC_MOVE_READ: 602 bcopy(ptr, p, seg_len); 603 cpu_flush_dcache(p, seg_len); 604 break; 605 case MD_MALLOC_MOVE_WRITE: 606 bcopy(p, ptr, seg_len); 607 break; 608 case MD_MALLOC_MOVE_CMP: 609 end = p + seg_len; 610 first = *uc = *p; 611 /* Confirm all following bytes match the first */ 612 while (++p < end) { 613 if (*p != first) 614 return (EDOOFUS); 615 } 616 break; 617 default: 618 KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op)); 619 break; 620 } 621 622 ma_offs += seg_len; 623 if (ma_offs == vlist->ds_len) { 624 ma_offs = 0; 625 vlist++; 626 } 627 ptr = (uint8_t *)ptr + seg_len; 628 } 629 *pvlist = vlist; 630 *pma_offs = ma_offs; 631 632 return (0); 633 } 634 635 static int 636 mdstart_malloc(struct md_s *sc, struct bio *bp) 637 { 638 u_char *dst; 639 vm_page_t *m; 640 bus_dma_segment_t *vlist; 641 int i, error, error1, ma_offs, notmapped; 642 off_t secno, nsec, uc; 643 uintptr_t sp, osp; 644 645 switch (bp->bio_cmd) { 646 case BIO_READ: 647 case BIO_WRITE: 648 case BIO_DELETE: 649 break; 650 default: 651 return (EOPNOTSUPP); 652 } 653 654 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0; 655 vlist = (bp->bio_flags & BIO_VLIST) != 0 ? 656 (bus_dma_segment_t *)bp->bio_data : NULL; 657 if (notmapped) { 658 m = bp->bio_ma; 659 ma_offs = bp->bio_ma_offset; 660 dst = NULL; 661 KASSERT(vlist == NULL, ("vlists cannot be unmapped")); 662 } else if (vlist != NULL) { 663 ma_offs = bp->bio_ma_offset; 664 dst = NULL; 665 } else { 666 dst = bp->bio_data; 667 } 668 669 nsec = bp->bio_length / sc->sectorsize; 670 secno = bp->bio_offset / sc->sectorsize; 671 error = 0; 672 while (nsec--) { 673 osp = s_read(sc->indir, secno); 674 if (bp->bio_cmd == BIO_DELETE) { 675 if (osp != 0) 676 error = s_write(sc->indir, secno, 0); 677 } else if (bp->bio_cmd == BIO_READ) { 678 if (osp == 0) { 679 if (notmapped) { 680 error = md_malloc_move_ma(&m, &ma_offs, 681 sc->sectorsize, NULL, 0, 682 MD_MALLOC_MOVE_ZERO); 683 } else if (vlist != NULL) { 684 error = md_malloc_move_vlist(&vlist, 685 &ma_offs, sc->sectorsize, NULL, 0, 686 MD_MALLOC_MOVE_ZERO); 687 } else 688 bzero(dst, sc->sectorsize); 689 } else if (osp <= 255) { 690 if (notmapped) { 691 error = md_malloc_move_ma(&m, &ma_offs, 692 sc->sectorsize, NULL, osp, 693 MD_MALLOC_MOVE_FILL); 694 } else if (vlist != NULL) { 695 error = md_malloc_move_vlist(&vlist, 696 &ma_offs, sc->sectorsize, NULL, osp, 697 MD_MALLOC_MOVE_FILL); 698 } else 699 memset(dst, osp, sc->sectorsize); 700 } else { 701 if (notmapped) { 702 error = md_malloc_move_ma(&m, &ma_offs, 703 sc->sectorsize, (void *)osp, 0, 704 MD_MALLOC_MOVE_READ); 705 } else if (vlist != NULL) { 706 error = md_malloc_move_vlist(&vlist, 707 &ma_offs, sc->sectorsize, 708 (void *)osp, 0, 709 MD_MALLOC_MOVE_READ); 710 } else { 711 bcopy((void *)osp, dst, sc->sectorsize); 712 cpu_flush_dcache(dst, sc->sectorsize); 713 } 714 } 715 osp = 0; 716 } else if (bp->bio_cmd == BIO_WRITE) { 717 if (sc->flags & MD_COMPRESS) { 718 if (notmapped) { 719 error1 = md_malloc_move_ma(&m, &ma_offs, 720 sc->sectorsize, &uc, 0, 721 MD_MALLOC_MOVE_CMP); 722 i = error1 == 0 ? sc->sectorsize : 0; 723 } else if (vlist != NULL) { 724 error1 = md_malloc_move_vlist(&vlist, 725 &ma_offs, sc->sectorsize, &uc, 0, 726 MD_MALLOC_MOVE_CMP); 727 i = error1 == 0 ? sc->sectorsize : 0; 728 } else { 729 uc = dst[0]; 730 for (i = 1; i < sc->sectorsize; i++) { 731 if (dst[i] != uc) 732 break; 733 } 734 } 735 } else { 736 i = 0; 737 uc = 0; 738 } 739 if (i == sc->sectorsize) { 740 if (osp != uc) 741 error = s_write(sc->indir, secno, uc); 742 } else { 743 if (osp <= 255) { 744 sp = (uintptr_t)uma_zalloc(sc->uma, 745 md_malloc_wait ? M_WAITOK : 746 M_NOWAIT); 747 if (sp == 0) { 748 error = ENOSPC; 749 break; 750 } 751 if (notmapped) { 752 error = md_malloc_move_ma(&m, 753 &ma_offs, sc->sectorsize, 754 (void *)sp, 0, 755 MD_MALLOC_MOVE_WRITE); 756 } else if (vlist != NULL) { 757 error = md_malloc_move_vlist( 758 &vlist, &ma_offs, 759 sc->sectorsize, (void *)sp, 760 0, MD_MALLOC_MOVE_WRITE); 761 } else { 762 bcopy(dst, (void *)sp, 763 sc->sectorsize); 764 } 765 error = s_write(sc->indir, secno, sp); 766 } else { 767 if (notmapped) { 768 error = md_malloc_move_ma(&m, 769 &ma_offs, sc->sectorsize, 770 (void *)osp, 0, 771 MD_MALLOC_MOVE_WRITE); 772 } else if (vlist != NULL) { 773 error = md_malloc_move_vlist( 774 &vlist, &ma_offs, 775 sc->sectorsize, (void *)osp, 776 0, MD_MALLOC_MOVE_WRITE); 777 } else { 778 bcopy(dst, (void *)osp, 779 sc->sectorsize); 780 } 781 osp = 0; 782 } 783 } 784 } else { 785 error = EOPNOTSUPP; 786 } 787 if (osp > 255) 788 uma_zfree(sc->uma, (void*)osp); 789 if (error != 0) 790 break; 791 secno++; 792 if (!notmapped && vlist == NULL) 793 dst += sc->sectorsize; 794 } 795 bp->bio_resid = 0; 796 return (error); 797 } 798 799 static void 800 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len) 801 { 802 off_t seg_len; 803 804 while (offset >= vlist->ds_len) { 805 offset -= vlist->ds_len; 806 vlist++; 807 } 808 809 while (len != 0) { 810 seg_len = omin(len, vlist->ds_len - offset); 811 bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset), 812 seg_len); 813 offset = 0; 814 src = (uint8_t *)src + seg_len; 815 len -= seg_len; 816 vlist++; 817 } 818 } 819 820 static void 821 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len) 822 { 823 off_t seg_len; 824 825 while (offset >= vlist->ds_len) { 826 offset -= vlist->ds_len; 827 vlist++; 828 } 829 830 while (len != 0) { 831 seg_len = omin(len, vlist->ds_len - offset); 832 bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst, 833 seg_len); 834 offset = 0; 835 dst = (uint8_t *)dst + seg_len; 836 len -= seg_len; 837 vlist++; 838 } 839 } 840 841 static int 842 mdstart_preload(struct md_s *sc, struct bio *bp) 843 { 844 uint8_t *p; 845 846 p = sc->pl_ptr + bp->bio_offset; 847 switch (bp->bio_cmd) { 848 case BIO_READ: 849 if ((bp->bio_flags & BIO_VLIST) != 0) { 850 mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data, 851 bp->bio_ma_offset, bp->bio_length); 852 } else { 853 bcopy(p, bp->bio_data, bp->bio_length); 854 } 855 cpu_flush_dcache(bp->bio_data, bp->bio_length); 856 break; 857 case BIO_WRITE: 858 if ((bp->bio_flags & BIO_VLIST) != 0) { 859 mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data, 860 bp->bio_ma_offset, p, bp->bio_length); 861 } else { 862 bcopy(bp->bio_data, p, bp->bio_length); 863 } 864 break; 865 } 866 bp->bio_resid = 0; 867 return (0); 868 } 869 870 static int 871 mdstart_vnode(struct md_s *sc, struct bio *bp) 872 { 873 int error; 874 struct uio auio; 875 struct iovec aiov; 876 struct iovec *piov; 877 struct mount *mp; 878 struct vnode *vp; 879 struct buf *pb; 880 bus_dma_segment_t *vlist; 881 struct thread *td; 882 off_t iolen, iostart, len, zerosize; 883 int ma_offs, npages; 884 885 switch (bp->bio_cmd) { 886 case BIO_READ: 887 auio.uio_rw = UIO_READ; 888 break; 889 case BIO_WRITE: 890 case BIO_DELETE: 891 auio.uio_rw = UIO_WRITE; 892 break; 893 case BIO_FLUSH: 894 break; 895 default: 896 return (EOPNOTSUPP); 897 } 898 899 td = curthread; 900 vp = sc->vnode; 901 pb = NULL; 902 piov = NULL; 903 ma_offs = bp->bio_ma_offset; 904 len = bp->bio_length; 905 906 /* 907 * VNODE I/O 908 * 909 * If an error occurs, we set BIO_ERROR but we do not set 910 * B_INVAL because (for a write anyway), the buffer is 911 * still valid. 912 */ 913 914 if (bp->bio_cmd == BIO_FLUSH) { 915 (void) vn_start_write(vp, &mp, V_WAIT); 916 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 917 error = VOP_FSYNC(vp, MNT_WAIT, td); 918 VOP_UNLOCK(vp); 919 vn_finished_write(mp); 920 return (error); 921 } 922 923 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 924 auio.uio_resid = bp->bio_length; 925 auio.uio_segflg = UIO_SYSSPACE; 926 auio.uio_td = td; 927 928 if (bp->bio_cmd == BIO_DELETE) { 929 /* 930 * Emulate BIO_DELETE by writing zeros. 931 */ 932 zerosize = ZERO_REGION_SIZE - 933 (ZERO_REGION_SIZE % sc->sectorsize); 934 auio.uio_iovcnt = howmany(bp->bio_length, zerosize); 935 piov = malloc(sizeof(*piov) * auio.uio_iovcnt, M_MD, M_WAITOK); 936 auio.uio_iov = piov; 937 while (len > 0) { 938 piov->iov_base = __DECONST(void *, zero_region); 939 piov->iov_len = len; 940 if (len > zerosize) 941 piov->iov_len = zerosize; 942 len -= piov->iov_len; 943 piov++; 944 } 945 piov = auio.uio_iov; 946 } else if ((bp->bio_flags & BIO_VLIST) != 0) { 947 piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK); 948 auio.uio_iov = piov; 949 vlist = (bus_dma_segment_t *)bp->bio_data; 950 while (len > 0) { 951 piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr + 952 ma_offs); 953 piov->iov_len = vlist->ds_len - ma_offs; 954 if (piov->iov_len > len) 955 piov->iov_len = len; 956 len -= piov->iov_len; 957 ma_offs = 0; 958 vlist++; 959 piov++; 960 } 961 auio.uio_iovcnt = piov - auio.uio_iov; 962 piov = auio.uio_iov; 963 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 964 pb = uma_zalloc(md_pbuf_zone, M_WAITOK); 965 bp->bio_resid = len; 966 unmapped_step: 967 npages = atop(min(MAXPHYS, round_page(len + (ma_offs & 968 PAGE_MASK)))); 969 iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len); 970 KASSERT(iolen > 0, ("zero iolen")); 971 pmap_qenter((vm_offset_t)pb->b_data, 972 &bp->bio_ma[atop(ma_offs)], npages); 973 aiov.iov_base = (void *)((vm_offset_t)pb->b_data + 974 (ma_offs & PAGE_MASK)); 975 aiov.iov_len = iolen; 976 auio.uio_iov = &aiov; 977 auio.uio_iovcnt = 1; 978 auio.uio_resid = iolen; 979 } else { 980 aiov.iov_base = bp->bio_data; 981 aiov.iov_len = bp->bio_length; 982 auio.uio_iov = &aiov; 983 auio.uio_iovcnt = 1; 984 } 985 iostart = auio.uio_offset; 986 if (auio.uio_rw == UIO_READ) { 987 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 988 error = VOP_READ(vp, &auio, 0, sc->cred); 989 VOP_UNLOCK(vp); 990 } else { 991 (void) vn_start_write(vp, &mp, V_WAIT); 992 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 993 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC, 994 sc->cred); 995 VOP_UNLOCK(vp); 996 vn_finished_write(mp); 997 if (error == 0) 998 sc->flags &= ~MD_VERIFY; 999 } 1000 1001 /* When MD_CACHE is set, try to avoid double-caching the data. */ 1002 if (error == 0 && (sc->flags & MD_CACHE) == 0) 1003 VOP_ADVISE(vp, iostart, auio.uio_offset - 1, 1004 POSIX_FADV_DONTNEED); 1005 1006 if (pb != NULL) { 1007 pmap_qremove((vm_offset_t)pb->b_data, npages); 1008 if (error == 0) { 1009 len -= iolen; 1010 bp->bio_resid -= iolen; 1011 ma_offs += iolen; 1012 if (len > 0) 1013 goto unmapped_step; 1014 } 1015 uma_zfree(md_pbuf_zone, pb); 1016 } 1017 1018 free(piov, M_MD); 1019 if (pb == NULL) 1020 bp->bio_resid = auio.uio_resid; 1021 return (error); 1022 } 1023 1024 static int 1025 mdstart_swap(struct md_s *sc, struct bio *bp) 1026 { 1027 vm_page_t m; 1028 u_char *p; 1029 vm_pindex_t i, lastp; 1030 bus_dma_segment_t *vlist; 1031 int rv, ma_offs, offs, len, lastend; 1032 1033 switch (bp->bio_cmd) { 1034 case BIO_READ: 1035 case BIO_WRITE: 1036 case BIO_DELETE: 1037 break; 1038 default: 1039 return (EOPNOTSUPP); 1040 } 1041 1042 p = bp->bio_data; 1043 ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ? 1044 bp->bio_ma_offset : 0; 1045 vlist = (bp->bio_flags & BIO_VLIST) != 0 ? 1046 (bus_dma_segment_t *)bp->bio_data : NULL; 1047 1048 /* 1049 * offs is the offset at which to start operating on the 1050 * next (ie, first) page. lastp is the last page on 1051 * which we're going to operate. lastend is the ending 1052 * position within that last page (ie, PAGE_SIZE if 1053 * we're operating on complete aligned pages). 1054 */ 1055 offs = bp->bio_offset % PAGE_SIZE; 1056 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE; 1057 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1; 1058 1059 rv = VM_PAGER_OK; 1060 vm_object_pip_add(sc->object, 1); 1061 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) { 1062 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs; 1063 m = vm_page_grab_unlocked(sc->object, i, VM_ALLOC_SYSTEM); 1064 if (bp->bio_cmd == BIO_READ) { 1065 if (vm_page_all_valid(m)) 1066 rv = VM_PAGER_OK; 1067 else 1068 rv = vm_pager_get_pages(sc->object, &m, 1, 1069 NULL, NULL); 1070 if (rv == VM_PAGER_ERROR) { 1071 VM_OBJECT_WLOCK(sc->object); 1072 vm_page_free(m); 1073 VM_OBJECT_WUNLOCK(sc->object); 1074 break; 1075 } else if (rv == VM_PAGER_FAIL) { 1076 /* 1077 * Pager does not have the page. Zero 1078 * the allocated page, and mark it as 1079 * valid. Do not set dirty, the page 1080 * can be recreated if thrown out. 1081 */ 1082 pmap_zero_page(m); 1083 vm_page_valid(m); 1084 } 1085 if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 1086 pmap_copy_pages(&m, offs, bp->bio_ma, 1087 ma_offs, len); 1088 } else if ((bp->bio_flags & BIO_VLIST) != 0) { 1089 physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs, 1090 vlist, ma_offs, len); 1091 cpu_flush_dcache(p, len); 1092 } else { 1093 physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len); 1094 cpu_flush_dcache(p, len); 1095 } 1096 } else if (bp->bio_cmd == BIO_WRITE) { 1097 if (len == PAGE_SIZE || vm_page_all_valid(m)) 1098 rv = VM_PAGER_OK; 1099 else 1100 rv = vm_pager_get_pages(sc->object, &m, 1, 1101 NULL, NULL); 1102 if (rv == VM_PAGER_ERROR) { 1103 VM_OBJECT_WLOCK(sc->object); 1104 vm_page_free(m); 1105 VM_OBJECT_WUNLOCK(sc->object); 1106 break; 1107 } else if (rv == VM_PAGER_FAIL) 1108 pmap_zero_page(m); 1109 1110 if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 1111 pmap_copy_pages(bp->bio_ma, ma_offs, &m, 1112 offs, len); 1113 } else if ((bp->bio_flags & BIO_VLIST) != 0) { 1114 physcopyin_vlist(vlist, ma_offs, 1115 VM_PAGE_TO_PHYS(m) + offs, len); 1116 } else { 1117 physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len); 1118 } 1119 1120 vm_page_valid(m); 1121 vm_page_set_dirty(m); 1122 } else if (bp->bio_cmd == BIO_DELETE) { 1123 if (len == PAGE_SIZE || vm_page_all_valid(m)) 1124 rv = VM_PAGER_OK; 1125 else 1126 rv = vm_pager_get_pages(sc->object, &m, 1, 1127 NULL, NULL); 1128 VM_OBJECT_WLOCK(sc->object); 1129 if (rv == VM_PAGER_ERROR) { 1130 vm_page_free(m); 1131 VM_OBJECT_WUNLOCK(sc->object); 1132 break; 1133 } else if (rv == VM_PAGER_FAIL) { 1134 vm_page_free(m); 1135 m = NULL; 1136 } else { 1137 /* Page is valid. */ 1138 if (len != PAGE_SIZE) { 1139 pmap_zero_page_area(m, offs, len); 1140 vm_page_set_dirty(m); 1141 } else { 1142 vm_pager_page_unswapped(m); 1143 vm_page_free(m); 1144 m = NULL; 1145 } 1146 } 1147 VM_OBJECT_WUNLOCK(sc->object); 1148 } 1149 if (m != NULL) { 1150 vm_page_xunbusy(m); 1151 1152 /* 1153 * The page may be deactivated prior to setting 1154 * PGA_REFERENCED, but in this case it will be 1155 * reactivated by the page daemon. 1156 */ 1157 if (vm_page_active(m)) 1158 vm_page_reference(m); 1159 else 1160 vm_page_activate(m); 1161 } 1162 1163 /* Actions on further pages start at offset 0 */ 1164 p += PAGE_SIZE - offs; 1165 offs = 0; 1166 ma_offs += len; 1167 } 1168 vm_object_pip_wakeup(sc->object); 1169 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC); 1170 } 1171 1172 static int 1173 mdstart_null(struct md_s *sc, struct bio *bp) 1174 { 1175 1176 switch (bp->bio_cmd) { 1177 case BIO_READ: 1178 bzero(bp->bio_data, bp->bio_length); 1179 cpu_flush_dcache(bp->bio_data, bp->bio_length); 1180 break; 1181 case BIO_WRITE: 1182 break; 1183 } 1184 bp->bio_resid = 0; 1185 return (0); 1186 } 1187 1188 static void 1189 md_kthread(void *arg) 1190 { 1191 struct md_s *sc; 1192 struct bio *bp; 1193 int error; 1194 1195 sc = arg; 1196 thread_lock(curthread); 1197 sched_prio(curthread, PRIBIO); 1198 thread_unlock(curthread); 1199 if (sc->type == MD_VNODE) 1200 curthread->td_pflags |= TDP_NORUNNINGBUF; 1201 1202 for (;;) { 1203 mtx_lock(&sc->queue_mtx); 1204 if (sc->flags & MD_SHUTDOWN) { 1205 sc->flags |= MD_EXITING; 1206 mtx_unlock(&sc->queue_mtx); 1207 kproc_exit(0); 1208 } 1209 bp = bioq_takefirst(&sc->bio_queue); 1210 if (!bp) { 1211 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 1212 continue; 1213 } 1214 mtx_unlock(&sc->queue_mtx); 1215 if (bp->bio_cmd == BIO_GETATTR) { 1216 int isv = ((sc->flags & MD_VERIFY) != 0); 1217 1218 if ((sc->fwsectors && sc->fwheads && 1219 (g_handleattr_int(bp, "GEOM::fwsectors", 1220 sc->fwsectors) || 1221 g_handleattr_int(bp, "GEOM::fwheads", 1222 sc->fwheads))) || 1223 g_handleattr_int(bp, "GEOM::candelete", 1)) 1224 error = -1; 1225 else if (sc->ident[0] != '\0' && 1226 g_handleattr_str(bp, "GEOM::ident", sc->ident)) 1227 error = -1; 1228 else if (g_handleattr_int(bp, "MNT::verified", isv)) 1229 error = -1; 1230 else 1231 error = EOPNOTSUPP; 1232 } else { 1233 error = sc->start(sc, bp); 1234 } 1235 1236 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) { 1237 /* 1238 * Devstat uses (bio_bcount, bio_resid) for 1239 * determining the length of the completed part of 1240 * the i/o. g_io_deliver() will translate from 1241 * bio_completed to that, but it also destroys the 1242 * bio so we must do our own translation. 1243 */ 1244 bp->bio_bcount = bp->bio_length; 1245 bp->bio_resid = (error == -1 ? bp->bio_bcount : 0); 1246 devstat_end_transaction_bio(sc->devstat, bp); 1247 } 1248 if (error != -1) { 1249 bp->bio_completed = bp->bio_length; 1250 g_io_deliver(bp, error); 1251 } 1252 } 1253 } 1254 1255 static struct md_s * 1256 mdfind(int unit) 1257 { 1258 struct md_s *sc; 1259 1260 LIST_FOREACH(sc, &md_softc_list, list) { 1261 if (sc->unit == unit) 1262 break; 1263 } 1264 return (sc); 1265 } 1266 1267 static struct md_s * 1268 mdnew(int unit, int *errp, enum md_types type) 1269 { 1270 struct md_s *sc; 1271 int error; 1272 1273 *errp = 0; 1274 if (unit == -1) 1275 unit = alloc_unr(md_uh); 1276 else 1277 unit = alloc_unr_specific(md_uh, unit); 1278 1279 if (unit == -1) { 1280 *errp = EBUSY; 1281 return (NULL); 1282 } 1283 1284 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 1285 sc->type = type; 1286 bioq_init(&sc->bio_queue); 1287 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 1288 sc->unit = unit; 1289 sprintf(sc->name, "md%d", unit); 1290 LIST_INSERT_HEAD(&md_softc_list, sc, list); 1291 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 1292 if (error == 0) 1293 return (sc); 1294 LIST_REMOVE(sc, list); 1295 mtx_destroy(&sc->queue_mtx); 1296 free_unr(md_uh, sc->unit); 1297 free(sc, M_MD); 1298 *errp = error; 1299 return (NULL); 1300 } 1301 1302 static void 1303 mdinit(struct md_s *sc) 1304 { 1305 struct g_geom *gp; 1306 struct g_provider *pp; 1307 1308 g_topology_lock(); 1309 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 1310 gp->softc = sc; 1311 pp = g_new_providerf(gp, "md%d", sc->unit); 1312 devstat_remove_entry(pp->stat); 1313 pp->stat = NULL; 1314 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE; 1315 pp->mediasize = sc->mediasize; 1316 pp->sectorsize = sc->sectorsize; 1317 switch (sc->type) { 1318 case MD_MALLOC: 1319 case MD_VNODE: 1320 case MD_SWAP: 1321 pp->flags |= G_PF_ACCEPT_UNMAPPED; 1322 break; 1323 case MD_PRELOAD: 1324 case MD_NULL: 1325 break; 1326 } 1327 sc->gp = gp; 1328 sc->pp = pp; 1329 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize, 1330 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX); 1331 sc->devstat->id = pp; 1332 g_error_provider(pp, 0); 1333 g_topology_unlock(); 1334 } 1335 1336 static int 1337 mdcreate_malloc(struct md_s *sc, struct md_req *mdr) 1338 { 1339 uintptr_t sp; 1340 int error; 1341 off_t u; 1342 1343 error = 0; 1344 if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 1345 return (EINVAL); 1346 if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize)) 1347 return (EINVAL); 1348 /* Compression doesn't make sense if we have reserved space */ 1349 if (mdr->md_options & MD_RESERVE) 1350 mdr->md_options &= ~MD_COMPRESS; 1351 if (mdr->md_fwsectors != 0) 1352 sc->fwsectors = mdr->md_fwsectors; 1353 if (mdr->md_fwheads != 0) 1354 sc->fwheads = mdr->md_fwheads; 1355 sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE); 1356 sc->indir = dimension(sc->mediasize / sc->sectorsize); 1357 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL, 1358 0x1ff, 0); 1359 if (mdr->md_options & MD_RESERVE) { 1360 off_t nsectors; 1361 1362 nsectors = sc->mediasize / sc->sectorsize; 1363 for (u = 0; u < nsectors; u++) { 1364 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ? 1365 M_WAITOK : M_NOWAIT) | M_ZERO); 1366 if (sp != 0) 1367 error = s_write(sc->indir, u, sp); 1368 else 1369 error = ENOMEM; 1370 if (error != 0) 1371 break; 1372 } 1373 } 1374 return (error); 1375 } 1376 1377 1378 static int 1379 mdsetcred(struct md_s *sc, struct ucred *cred) 1380 { 1381 char *tmpbuf; 1382 int error = 0; 1383 1384 /* 1385 * Set credits in our softc 1386 */ 1387 1388 if (sc->cred) 1389 crfree(sc->cred); 1390 sc->cred = crhold(cred); 1391 1392 /* 1393 * Horrible kludge to establish credentials for NFS XXX. 1394 */ 1395 1396 if (sc->vnode) { 1397 struct uio auio; 1398 struct iovec aiov; 1399 1400 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK); 1401 bzero(&auio, sizeof(auio)); 1402 1403 aiov.iov_base = tmpbuf; 1404 aiov.iov_len = sc->sectorsize; 1405 auio.uio_iov = &aiov; 1406 auio.uio_iovcnt = 1; 1407 auio.uio_offset = 0; 1408 auio.uio_rw = UIO_READ; 1409 auio.uio_segflg = UIO_SYSSPACE; 1410 auio.uio_resid = aiov.iov_len; 1411 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 1412 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 1413 VOP_UNLOCK(sc->vnode); 1414 free(tmpbuf, M_TEMP); 1415 } 1416 return (error); 1417 } 1418 1419 static int 1420 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td) 1421 { 1422 struct vattr vattr; 1423 struct nameidata nd; 1424 char *fname; 1425 int error, flags; 1426 1427 fname = mdr->md_file; 1428 if (mdr->md_file_seg == UIO_USERSPACE) { 1429 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL); 1430 if (error != 0) 1431 return (error); 1432 } else if (mdr->md_file_seg == UIO_SYSSPACE) 1433 strlcpy(sc->file, fname, sizeof(sc->file)); 1434 else 1435 return (EDOOFUS); 1436 1437 /* 1438 * If the user specified that this is a read only device, don't 1439 * set the FWRITE mask before trying to open the backing store. 1440 */ 1441 flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \ 1442 | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0); 1443 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td); 1444 error = vn_open(&nd, &flags, 0, NULL); 1445 if (error != 0) 1446 return (error); 1447 NDFREE(&nd, NDF_ONLY_PNBUF); 1448 if (nd.ni_vp->v_type != VREG) { 1449 error = EINVAL; 1450 goto bad; 1451 } 1452 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred); 1453 if (error != 0) 1454 goto bad; 1455 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) { 1456 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY); 1457 if (VN_IS_DOOMED(nd.ni_vp)) { 1458 /* Forced unmount. */ 1459 error = EBADF; 1460 goto bad; 1461 } 1462 } 1463 nd.ni_vp->v_vflag |= VV_MD; 1464 VOP_UNLOCK(nd.ni_vp); 1465 1466 if (mdr->md_fwsectors != 0) 1467 sc->fwsectors = mdr->md_fwsectors; 1468 if (mdr->md_fwheads != 0) 1469 sc->fwheads = mdr->md_fwheads; 1470 snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju", 1471 (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid); 1472 sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE | 1473 MD_VERIFY); 1474 if (!(flags & FWRITE)) 1475 sc->flags |= MD_READONLY; 1476 sc->vnode = nd.ni_vp; 1477 1478 error = mdsetcred(sc, td->td_ucred); 1479 if (error != 0) { 1480 sc->vnode = NULL; 1481 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY); 1482 nd.ni_vp->v_vflag &= ~VV_MD; 1483 goto bad; 1484 } 1485 return (0); 1486 bad: 1487 VOP_UNLOCK(nd.ni_vp); 1488 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td); 1489 return (error); 1490 } 1491 1492 static void 1493 g_md_providergone(struct g_provider *pp) 1494 { 1495 struct md_s *sc = pp->geom->softc; 1496 1497 mtx_lock(&sc->queue_mtx); 1498 sc->flags |= MD_PROVIDERGONE; 1499 wakeup(&sc->flags); 1500 mtx_unlock(&sc->queue_mtx); 1501 } 1502 1503 static int 1504 mddestroy(struct md_s *sc, struct thread *td) 1505 { 1506 1507 if (sc->gp) { 1508 g_topology_lock(); 1509 g_wither_geom(sc->gp, ENXIO); 1510 g_topology_unlock(); 1511 1512 mtx_lock(&sc->queue_mtx); 1513 while (!(sc->flags & MD_PROVIDERGONE)) 1514 msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0); 1515 mtx_unlock(&sc->queue_mtx); 1516 } 1517 if (sc->devstat) { 1518 devstat_remove_entry(sc->devstat); 1519 sc->devstat = NULL; 1520 } 1521 mtx_lock(&sc->queue_mtx); 1522 sc->flags |= MD_SHUTDOWN; 1523 wakeup(sc); 1524 while (!(sc->flags & MD_EXITING)) 1525 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10); 1526 mtx_unlock(&sc->queue_mtx); 1527 mtx_destroy(&sc->queue_mtx); 1528 if (sc->vnode != NULL) { 1529 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 1530 sc->vnode->v_vflag &= ~VV_MD; 1531 VOP_UNLOCK(sc->vnode); 1532 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 1533 FREAD : (FREAD|FWRITE), sc->cred, td); 1534 } 1535 if (sc->cred != NULL) 1536 crfree(sc->cred); 1537 if (sc->object != NULL) 1538 vm_object_deallocate(sc->object); 1539 if (sc->indir) 1540 destroy_indir(sc, sc->indir); 1541 if (sc->uma) 1542 uma_zdestroy(sc->uma); 1543 1544 LIST_REMOVE(sc, list); 1545 free_unr(md_uh, sc->unit); 1546 free(sc, M_MD); 1547 return (0); 1548 } 1549 1550 static int 1551 mdresize(struct md_s *sc, struct md_req *mdr) 1552 { 1553 int error, res; 1554 vm_pindex_t oldpages, newpages; 1555 1556 switch (sc->type) { 1557 case MD_VNODE: 1558 case MD_NULL: 1559 break; 1560 case MD_SWAP: 1561 if (mdr->md_mediasize <= 0 || 1562 (mdr->md_mediasize % PAGE_SIZE) != 0) 1563 return (EDOM); 1564 oldpages = OFF_TO_IDX(sc->mediasize); 1565 newpages = OFF_TO_IDX(mdr->md_mediasize); 1566 if (newpages < oldpages) { 1567 VM_OBJECT_WLOCK(sc->object); 1568 vm_object_page_remove(sc->object, newpages, 0, 0); 1569 swap_release_by_cred(IDX_TO_OFF(oldpages - 1570 newpages), sc->cred); 1571 sc->object->charge = IDX_TO_OFF(newpages); 1572 sc->object->size = newpages; 1573 VM_OBJECT_WUNLOCK(sc->object); 1574 } else if (newpages > oldpages) { 1575 res = swap_reserve_by_cred(IDX_TO_OFF(newpages - 1576 oldpages), sc->cred); 1577 if (!res) 1578 return (ENOMEM); 1579 if ((mdr->md_options & MD_RESERVE) || 1580 (sc->flags & MD_RESERVE)) { 1581 error = swap_pager_reserve(sc->object, 1582 oldpages, newpages - oldpages); 1583 if (error < 0) { 1584 swap_release_by_cred( 1585 IDX_TO_OFF(newpages - oldpages), 1586 sc->cred); 1587 return (EDOM); 1588 } 1589 } 1590 VM_OBJECT_WLOCK(sc->object); 1591 sc->object->charge = IDX_TO_OFF(newpages); 1592 sc->object->size = newpages; 1593 VM_OBJECT_WUNLOCK(sc->object); 1594 } 1595 break; 1596 default: 1597 return (EOPNOTSUPP); 1598 } 1599 1600 sc->mediasize = mdr->md_mediasize; 1601 g_topology_lock(); 1602 g_resize_provider(sc->pp, sc->mediasize); 1603 g_topology_unlock(); 1604 return (0); 1605 } 1606 1607 static int 1608 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td) 1609 { 1610 vm_ooffset_t npage; 1611 int error; 1612 1613 /* 1614 * Range check. Disallow negative sizes and sizes not being 1615 * multiple of page size. 1616 */ 1617 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0) 1618 return (EDOM); 1619 1620 /* 1621 * Allocate an OBJT_SWAP object. 1622 * 1623 * Note the truncation. 1624 */ 1625 1626 if ((mdr->md_options & MD_VERIFY) != 0) 1627 return (EINVAL); 1628 npage = mdr->md_mediasize / PAGE_SIZE; 1629 if (mdr->md_fwsectors != 0) 1630 sc->fwsectors = mdr->md_fwsectors; 1631 if (mdr->md_fwheads != 0) 1632 sc->fwheads = mdr->md_fwheads; 1633 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage, 1634 VM_PROT_DEFAULT, 0, td->td_ucred); 1635 if (sc->object == NULL) 1636 return (ENOMEM); 1637 sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE); 1638 if (mdr->md_options & MD_RESERVE) { 1639 if (swap_pager_reserve(sc->object, 0, npage) < 0) { 1640 error = EDOM; 1641 goto finish; 1642 } 1643 } 1644 error = mdsetcred(sc, td->td_ucred); 1645 finish: 1646 if (error != 0) { 1647 vm_object_deallocate(sc->object); 1648 sc->object = NULL; 1649 } 1650 return (error); 1651 } 1652 1653 static int 1654 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td) 1655 { 1656 1657 /* 1658 * Range check. Disallow negative sizes and sizes not being 1659 * multiple of page size. 1660 */ 1661 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0) 1662 return (EDOM); 1663 1664 return (0); 1665 } 1666 1667 static int 1668 kern_mdattach_locked(struct thread *td, struct md_req *mdr) 1669 { 1670 struct md_s *sc; 1671 unsigned sectsize; 1672 int error, i; 1673 1674 sx_assert(&md_sx, SA_XLOCKED); 1675 1676 switch (mdr->md_type) { 1677 case MD_MALLOC: 1678 case MD_PRELOAD: 1679 case MD_VNODE: 1680 case MD_SWAP: 1681 case MD_NULL: 1682 break; 1683 default: 1684 return (EINVAL); 1685 } 1686 if (mdr->md_sectorsize == 0) 1687 sectsize = DEV_BSIZE; 1688 else 1689 sectsize = mdr->md_sectorsize; 1690 if (sectsize > MAXPHYS || mdr->md_mediasize < sectsize) 1691 return (EINVAL); 1692 if (mdr->md_options & MD_AUTOUNIT) 1693 sc = mdnew(-1, &error, mdr->md_type); 1694 else { 1695 if (mdr->md_unit > INT_MAX) 1696 return (EINVAL); 1697 sc = mdnew(mdr->md_unit, &error, mdr->md_type); 1698 } 1699 if (sc == NULL) 1700 return (error); 1701 if (mdr->md_label != NULL) 1702 error = copyinstr(mdr->md_label, sc->label, 1703 sizeof(sc->label), NULL); 1704 if (error != 0) 1705 goto err_after_new; 1706 if (mdr->md_options & MD_AUTOUNIT) 1707 mdr->md_unit = sc->unit; 1708 sc->mediasize = mdr->md_mediasize; 1709 sc->sectorsize = sectsize; 1710 error = EDOOFUS; 1711 switch (sc->type) { 1712 case MD_MALLOC: 1713 sc->start = mdstart_malloc; 1714 error = mdcreate_malloc(sc, mdr); 1715 break; 1716 case MD_PRELOAD: 1717 /* 1718 * We disallow attaching preloaded memory disks via 1719 * ioctl. Preloaded memory disks are automatically 1720 * attached in g_md_init(). 1721 */ 1722 error = EOPNOTSUPP; 1723 break; 1724 case MD_VNODE: 1725 sc->start = mdstart_vnode; 1726 error = mdcreate_vnode(sc, mdr, td); 1727 break; 1728 case MD_SWAP: 1729 sc->start = mdstart_swap; 1730 error = mdcreate_swap(sc, mdr, td); 1731 break; 1732 case MD_NULL: 1733 sc->start = mdstart_null; 1734 error = mdcreate_null(sc, mdr, td); 1735 break; 1736 } 1737 err_after_new: 1738 if (error != 0) { 1739 mddestroy(sc, td); 1740 return (error); 1741 } 1742 1743 /* Prune off any residual fractional sector */ 1744 i = sc->mediasize % sc->sectorsize; 1745 sc->mediasize -= i; 1746 1747 mdinit(sc); 1748 return (0); 1749 } 1750 1751 static int 1752 kern_mdattach(struct thread *td, struct md_req *mdr) 1753 { 1754 int error; 1755 1756 sx_xlock(&md_sx); 1757 error = kern_mdattach_locked(td, mdr); 1758 sx_xunlock(&md_sx); 1759 return (error); 1760 } 1761 1762 static int 1763 kern_mddetach_locked(struct thread *td, struct md_req *mdr) 1764 { 1765 struct md_s *sc; 1766 1767 sx_assert(&md_sx, SA_XLOCKED); 1768 1769 if (mdr->md_mediasize != 0 || 1770 (mdr->md_options & ~MD_FORCE) != 0) 1771 return (EINVAL); 1772 1773 sc = mdfind(mdr->md_unit); 1774 if (sc == NULL) 1775 return (ENOENT); 1776 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) && 1777 !(mdr->md_options & MD_FORCE)) 1778 return (EBUSY); 1779 return (mddestroy(sc, td)); 1780 } 1781 1782 static int 1783 kern_mddetach(struct thread *td, struct md_req *mdr) 1784 { 1785 int error; 1786 1787 sx_xlock(&md_sx); 1788 error = kern_mddetach_locked(td, mdr); 1789 sx_xunlock(&md_sx); 1790 return (error); 1791 } 1792 1793 static int 1794 kern_mdresize_locked(struct md_req *mdr) 1795 { 1796 struct md_s *sc; 1797 1798 sx_assert(&md_sx, SA_XLOCKED); 1799 1800 if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0) 1801 return (EINVAL); 1802 1803 sc = mdfind(mdr->md_unit); 1804 if (sc == NULL) 1805 return (ENOENT); 1806 if (mdr->md_mediasize < sc->sectorsize) 1807 return (EINVAL); 1808 if (mdr->md_mediasize < sc->mediasize && 1809 !(sc->flags & MD_FORCE) && 1810 !(mdr->md_options & MD_FORCE)) 1811 return (EBUSY); 1812 return (mdresize(sc, mdr)); 1813 } 1814 1815 static int 1816 kern_mdresize(struct md_req *mdr) 1817 { 1818 int error; 1819 1820 sx_xlock(&md_sx); 1821 error = kern_mdresize_locked(mdr); 1822 sx_xunlock(&md_sx); 1823 return (error); 1824 } 1825 1826 static int 1827 kern_mdquery_locked(struct md_req *mdr) 1828 { 1829 struct md_s *sc; 1830 int error; 1831 1832 sx_assert(&md_sx, SA_XLOCKED); 1833 1834 sc = mdfind(mdr->md_unit); 1835 if (sc == NULL) 1836 return (ENOENT); 1837 mdr->md_type = sc->type; 1838 mdr->md_options = sc->flags; 1839 mdr->md_mediasize = sc->mediasize; 1840 mdr->md_sectorsize = sc->sectorsize; 1841 error = 0; 1842 if (mdr->md_label != NULL) { 1843 error = copyout(sc->label, mdr->md_label, 1844 strlen(sc->label) + 1); 1845 if (error != 0) 1846 return (error); 1847 } 1848 if (sc->type == MD_VNODE || 1849 (sc->type == MD_PRELOAD && mdr->md_file != NULL)) 1850 error = copyout(sc->file, mdr->md_file, 1851 strlen(sc->file) + 1); 1852 return (error); 1853 } 1854 1855 static int 1856 kern_mdquery(struct md_req *mdr) 1857 { 1858 int error; 1859 1860 sx_xlock(&md_sx); 1861 error = kern_mdquery_locked(mdr); 1862 sx_xunlock(&md_sx); 1863 return (error); 1864 } 1865 1866 /* Copy members that are not userspace pointers. */ 1867 #define MD_IOCTL2REQ(mdio, mdr) do { \ 1868 (mdr)->md_unit = (mdio)->md_unit; \ 1869 (mdr)->md_type = (mdio)->md_type; \ 1870 (mdr)->md_mediasize = (mdio)->md_mediasize; \ 1871 (mdr)->md_sectorsize = (mdio)->md_sectorsize; \ 1872 (mdr)->md_options = (mdio)->md_options; \ 1873 (mdr)->md_fwheads = (mdio)->md_fwheads; \ 1874 (mdr)->md_fwsectors = (mdio)->md_fwsectors; \ 1875 (mdr)->md_units = &(mdio)->md_pad[0]; \ 1876 (mdr)->md_units_nitems = nitems((mdio)->md_pad); \ 1877 } while(0) 1878 1879 /* Copy members that might have been updated */ 1880 #define MD_REQ2IOCTL(mdr, mdio) do { \ 1881 (mdio)->md_unit = (mdr)->md_unit; \ 1882 (mdio)->md_type = (mdr)->md_type; \ 1883 (mdio)->md_mediasize = (mdr)->md_mediasize; \ 1884 (mdio)->md_sectorsize = (mdr)->md_sectorsize; \ 1885 (mdio)->md_options = (mdr)->md_options; \ 1886 (mdio)->md_fwheads = (mdr)->md_fwheads; \ 1887 (mdio)->md_fwsectors = (mdr)->md_fwsectors; \ 1888 } while(0) 1889 1890 static int 1891 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, 1892 struct thread *td) 1893 { 1894 struct md_req mdr; 1895 int error; 1896 1897 if (md_debug) 1898 printf("mdctlioctl(%s %lx %p %x %p)\n", 1899 devtoname(dev), cmd, addr, flags, td); 1900 1901 bzero(&mdr, sizeof(mdr)); 1902 switch (cmd) { 1903 case MDIOCATTACH: 1904 case MDIOCDETACH: 1905 case MDIOCRESIZE: 1906 case MDIOCQUERY: { 1907 struct md_ioctl *mdio = (struct md_ioctl *)addr; 1908 if (mdio->md_version != MDIOVERSION) 1909 return (EINVAL); 1910 MD_IOCTL2REQ(mdio, &mdr); 1911 mdr.md_file = mdio->md_file; 1912 mdr.md_file_seg = UIO_USERSPACE; 1913 /* If the file is adjacent to the md_ioctl it's in kernel. */ 1914 if ((void *)mdio->md_file == (void *)(mdio + 1)) 1915 mdr.md_file_seg = UIO_SYSSPACE; 1916 mdr.md_label = mdio->md_label; 1917 break; 1918 } 1919 #ifdef COMPAT_FREEBSD32 1920 case MDIOCATTACH_32: 1921 case MDIOCDETACH_32: 1922 case MDIOCRESIZE_32: 1923 case MDIOCQUERY_32: { 1924 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr; 1925 if (mdio->md_version != MDIOVERSION) 1926 return (EINVAL); 1927 MD_IOCTL2REQ(mdio, &mdr); 1928 mdr.md_file = (void *)(uintptr_t)mdio->md_file; 1929 mdr.md_file_seg = UIO_USERSPACE; 1930 mdr.md_label = (void *)(uintptr_t)mdio->md_label; 1931 break; 1932 } 1933 #endif 1934 default: 1935 /* Fall through to handler switch. */ 1936 break; 1937 } 1938 1939 error = 0; 1940 switch (cmd) { 1941 case MDIOCATTACH: 1942 #ifdef COMPAT_FREEBSD32 1943 case MDIOCATTACH_32: 1944 #endif 1945 error = kern_mdattach(td, &mdr); 1946 break; 1947 case MDIOCDETACH: 1948 #ifdef COMPAT_FREEBSD32 1949 case MDIOCDETACH_32: 1950 #endif 1951 error = kern_mddetach(td, &mdr); 1952 break; 1953 case MDIOCRESIZE: 1954 #ifdef COMPAT_FREEBSD32 1955 case MDIOCRESIZE_32: 1956 #endif 1957 error = kern_mdresize(&mdr); 1958 break; 1959 case MDIOCQUERY: 1960 #ifdef COMPAT_FREEBSD32 1961 case MDIOCQUERY_32: 1962 #endif 1963 error = kern_mdquery(&mdr); 1964 break; 1965 default: 1966 error = ENOIOCTL; 1967 } 1968 1969 switch (cmd) { 1970 case MDIOCATTACH: 1971 case MDIOCQUERY: { 1972 struct md_ioctl *mdio = (struct md_ioctl *)addr; 1973 MD_REQ2IOCTL(&mdr, mdio); 1974 break; 1975 } 1976 #ifdef COMPAT_FREEBSD32 1977 case MDIOCATTACH_32: 1978 case MDIOCQUERY_32: { 1979 struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr; 1980 MD_REQ2IOCTL(&mdr, mdio); 1981 break; 1982 } 1983 #endif 1984 default: 1985 /* Other commands to not alter mdr. */ 1986 break; 1987 } 1988 1989 return (error); 1990 } 1991 1992 static void 1993 md_preloaded(u_char *image, size_t length, const char *name) 1994 { 1995 struct md_s *sc; 1996 int error; 1997 1998 sc = mdnew(-1, &error, MD_PRELOAD); 1999 if (sc == NULL) 2000 return; 2001 sc->mediasize = length; 2002 sc->sectorsize = DEV_BSIZE; 2003 sc->pl_ptr = image; 2004 sc->pl_len = length; 2005 sc->start = mdstart_preload; 2006 if (name != NULL) 2007 strlcpy(sc->file, name, sizeof(sc->file)); 2008 #ifdef MD_ROOT 2009 if (sc->unit == 0) { 2010 #ifndef ROOTDEVNAME 2011 rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0"; 2012 #endif 2013 #ifdef MD_ROOT_READONLY 2014 sc->flags |= MD_READONLY; 2015 #endif 2016 } 2017 #endif 2018 mdinit(sc); 2019 if (name != NULL) { 2020 printf("%s%d: Preloaded image <%s> %zd bytes at %p\n", 2021 MD_NAME, sc->unit, name, length, image); 2022 } else { 2023 printf("%s%d: Embedded image %zd bytes at %p\n", 2024 MD_NAME, sc->unit, length, image); 2025 } 2026 } 2027 2028 static void 2029 g_md_init(struct g_class *mp __unused) 2030 { 2031 caddr_t mod; 2032 u_char *ptr, *name, *type; 2033 unsigned len; 2034 int i; 2035 2036 /* figure out log2(NINDIR) */ 2037 for (i = NINDIR, nshift = -1; i; nshift++) 2038 i >>= 1; 2039 2040 mod = NULL; 2041 sx_init(&md_sx, "MD config lock"); 2042 g_topology_unlock(); 2043 md_uh = new_unrhdr(0, INT_MAX, NULL); 2044 #ifdef MD_ROOT 2045 if (mfs_root_size != 0) { 2046 sx_xlock(&md_sx); 2047 #ifdef MD_ROOT_MEM 2048 md_preloaded(mfs_root, mfs_root_size, NULL); 2049 #else 2050 md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size, 2051 NULL); 2052 #endif 2053 sx_xunlock(&md_sx); 2054 } 2055 #endif 2056 /* XXX: are preload_* static or do they need Giant ? */ 2057 while ((mod = preload_search_next_name(mod)) != NULL) { 2058 name = (char *)preload_search_info(mod, MODINFO_NAME); 2059 if (name == NULL) 2060 continue; 2061 type = (char *)preload_search_info(mod, MODINFO_TYPE); 2062 if (type == NULL) 2063 continue; 2064 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 2065 continue; 2066 ptr = preload_fetch_addr(mod); 2067 len = preload_fetch_size(mod); 2068 if (ptr != NULL && len != 0) { 2069 sx_xlock(&md_sx); 2070 md_preloaded(ptr, len, name); 2071 sx_xunlock(&md_sx); 2072 } 2073 } 2074 md_pbuf_zone = pbuf_zsecond_create("mdpbuf", nswbuf / 10); 2075 status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL, 2076 0600, MDCTL_NAME); 2077 g_topology_lock(); 2078 } 2079 2080 static void 2081 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2082 struct g_consumer *cp __unused, struct g_provider *pp) 2083 { 2084 struct md_s *mp; 2085 char *type; 2086 2087 mp = gp->softc; 2088 if (mp == NULL) 2089 return; 2090 2091 switch (mp->type) { 2092 case MD_MALLOC: 2093 type = "malloc"; 2094 break; 2095 case MD_PRELOAD: 2096 type = "preload"; 2097 break; 2098 case MD_VNODE: 2099 type = "vnode"; 2100 break; 2101 case MD_SWAP: 2102 type = "swap"; 2103 break; 2104 case MD_NULL: 2105 type = "null"; 2106 break; 2107 default: 2108 type = "unknown"; 2109 break; 2110 } 2111 2112 if (pp != NULL) { 2113 if (indent == NULL) { 2114 sbuf_printf(sb, " u %d", mp->unit); 2115 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize); 2116 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads); 2117 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors); 2118 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize); 2119 sbuf_printf(sb, " t %s", type); 2120 if ((mp->type == MD_VNODE && mp->vnode != NULL) || 2121 (mp->type == MD_PRELOAD && mp->file[0] != '\0')) 2122 sbuf_printf(sb, " file %s", mp->file); 2123 sbuf_printf(sb, " label %s", mp->label); 2124 } else { 2125 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, 2126 mp->unit); 2127 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n", 2128 indent, (uintmax_t) mp->sectorsize); 2129 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n", 2130 indent, (uintmax_t) mp->fwheads); 2131 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n", 2132 indent, (uintmax_t) mp->fwsectors); 2133 if (mp->ident[0] != '\0') { 2134 sbuf_printf(sb, "%s<ident>", indent); 2135 g_conf_printf_escaped(sb, "%s", mp->ident); 2136 sbuf_printf(sb, "</ident>\n"); 2137 } 2138 sbuf_printf(sb, "%s<length>%ju</length>\n", 2139 indent, (uintmax_t) mp->mediasize); 2140 sbuf_printf(sb, "%s<compression>%s</compression>\n", indent, 2141 (mp->flags & MD_COMPRESS) == 0 ? "off": "on"); 2142 sbuf_printf(sb, "%s<access>%s</access>\n", indent, 2143 (mp->flags & MD_READONLY) == 0 ? "read-write": 2144 "read-only"); 2145 sbuf_printf(sb, "%s<type>%s</type>\n", indent, 2146 type); 2147 if ((mp->type == MD_VNODE && mp->vnode != NULL) || 2148 (mp->type == MD_PRELOAD && mp->file[0] != '\0')) { 2149 sbuf_printf(sb, "%s<file>", indent); 2150 g_conf_printf_escaped(sb, "%s", mp->file); 2151 sbuf_printf(sb, "</file>\n"); 2152 } 2153 if (mp->type == MD_VNODE) 2154 sbuf_printf(sb, "%s<cache>%s</cache>\n", indent, 2155 (mp->flags & MD_CACHE) == 0 ? "off": "on"); 2156 sbuf_printf(sb, "%s<label>", indent); 2157 g_conf_printf_escaped(sb, "%s", mp->label); 2158 sbuf_printf(sb, "</label>\n"); 2159 } 2160 } 2161 } 2162 2163 static void 2164 g_md_fini(struct g_class *mp __unused) 2165 { 2166 2167 sx_destroy(&md_sx); 2168 if (status_dev != NULL) 2169 destroy_dev(status_dev); 2170 uma_zdestroy(md_pbuf_zone); 2171 delete_unrhdr(md_uh); 2172 } 2173