1 /*- 2 * SPDX-License-Identifier: (Beerware AND BSD-3-Clause) 3 * 4 * ---------------------------------------------------------------------------- 5 * "THE BEER-WARE LICENSE" (Revision 42): 6 * <[email protected]> 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 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root)) 191 #endif 192 #endif 193 194 static g_init_t g_md_init; 195 static g_fini_t g_md_fini; 196 static g_start_t g_md_start; 197 static g_access_t g_md_access; 198 static void g_md_dumpconf(struct sbuf *sb, const char *indent, 199 struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp); 200 static g_provgone_t g_md_providergone; 201 202 static struct cdev *status_dev = NULL; 203 static struct sx md_sx; 204 static struct unrhdr *md_uh; 205 206 static d_ioctl_t mdctlioctl; 207 208 static struct cdevsw mdctl_cdevsw = { 209 .d_version = D_VERSION, 210 .d_ioctl = mdctlioctl, 211 .d_name = MD_NAME, 212 }; 213 214 struct g_class g_md_class = { 215 .name = "MD", 216 .version = G_VERSION, 217 .init = g_md_init, 218 .fini = g_md_fini, 219 .start = g_md_start, 220 .access = g_md_access, 221 .dumpconf = g_md_dumpconf, 222 .providergone = g_md_providergone, 223 }; 224 225 DECLARE_GEOM_CLASS(g_md_class, g_md); 226 MODULE_VERSION(geom_md, 0); 227 228 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list); 229 230 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 231 #define NMASK (NINDIR-1) 232 static int nshift; 233 234 static uma_zone_t md_pbuf_zone; 235 236 struct indir { 237 uintptr_t *array; 238 u_int total; 239 u_int used; 240 u_int shift; 241 }; 242 243 struct md_s { 244 int unit; 245 LIST_ENTRY(md_s) list; 246 struct bio_queue_head bio_queue; 247 struct mtx queue_mtx; 248 struct cdev *dev; 249 enum md_types type; 250 off_t mediasize; 251 unsigned sectorsize; 252 unsigned opencount; 253 unsigned fwheads; 254 unsigned fwsectors; 255 char ident[32]; 256 unsigned flags; 257 char name[20]; 258 struct proc *procp; 259 struct g_geom *gp; 260 struct g_provider *pp; 261 int (*start)(struct md_s *sc, struct bio *bp); 262 struct devstat *devstat; 263 264 /* MD_MALLOC related fields */ 265 struct indir *indir; 266 uma_zone_t uma; 267 268 /* MD_PRELOAD related fields */ 269 u_char *pl_ptr; 270 size_t pl_len; 271 272 /* MD_VNODE related fields */ 273 struct vnode *vnode; 274 char file[PATH_MAX]; 275 char label[PATH_MAX]; 276 struct ucred *cred; 277 278 /* MD_SWAP related fields */ 279 vm_object_t object; 280 }; 281 282 static struct indir * 283 new_indir(u_int shift) 284 { 285 struct indir *ip; 286 287 ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT) 288 | M_ZERO); 289 if (ip == NULL) 290 return (NULL); 291 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 292 M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO); 293 if (ip->array == NULL) { 294 free(ip, M_MD); 295 return (NULL); 296 } 297 ip->total = NINDIR; 298 ip->shift = shift; 299 return (ip); 300 } 301 302 static void 303 del_indir(struct indir *ip) 304 { 305 306 free(ip->array, M_MDSECT); 307 free(ip, M_MD); 308 } 309 310 static void 311 destroy_indir(struct md_s *sc, struct indir *ip) 312 { 313 int i; 314 315 for (i = 0; i < NINDIR; i++) { 316 if (!ip->array[i]) 317 continue; 318 if (ip->shift) 319 destroy_indir(sc, (struct indir*)(ip->array[i])); 320 else if (ip->array[i] > 255) 321 uma_zfree(sc->uma, (void *)(ip->array[i])); 322 } 323 del_indir(ip); 324 } 325 326 /* 327 * This function does the math and allocates the top level "indir" structure 328 * for a device of "size" sectors. 329 */ 330 331 static struct indir * 332 dimension(off_t size) 333 { 334 off_t rcnt; 335 struct indir *ip; 336 int layer; 337 338 rcnt = size; 339 layer = 0; 340 while (rcnt > NINDIR) { 341 rcnt /= NINDIR; 342 layer++; 343 } 344 345 /* 346 * XXX: the top layer is probably not fully populated, so we allocate 347 * too much space for ip->array in here. 348 */ 349 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 350 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 351 M_MDSECT, M_WAITOK | M_ZERO); 352 ip->total = NINDIR; 353 ip->shift = layer * nshift; 354 return (ip); 355 } 356 357 /* 358 * Read a given sector 359 */ 360 361 static uintptr_t 362 s_read(struct indir *ip, off_t offset) 363 { 364 struct indir *cip; 365 int idx; 366 uintptr_t up; 367 368 if (md_debug > 1) 369 printf("s_read(%jd)\n", (intmax_t)offset); 370 up = 0; 371 for (cip = ip; cip != NULL;) { 372 if (cip->shift) { 373 idx = (offset >> cip->shift) & NMASK; 374 up = cip->array[idx]; 375 cip = (struct indir *)up; 376 continue; 377 } 378 idx = offset & NMASK; 379 return (cip->array[idx]); 380 } 381 return (0); 382 } 383 384 /* 385 * Write a given sector, prune the tree if the value is 0 386 */ 387 388 static int 389 s_write(struct indir *ip, off_t offset, uintptr_t ptr) 390 { 391 struct indir *cip, *lip[10]; 392 int idx, li; 393 uintptr_t up; 394 395 if (md_debug > 1) 396 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 397 up = 0; 398 li = 0; 399 cip = ip; 400 for (;;) { 401 lip[li++] = cip; 402 if (cip->shift) { 403 idx = (offset >> cip->shift) & NMASK; 404 up = cip->array[idx]; 405 if (up != 0) { 406 cip = (struct indir *)up; 407 continue; 408 } 409 /* Allocate branch */ 410 cip->array[idx] = 411 (uintptr_t)new_indir(cip->shift - nshift); 412 if (cip->array[idx] == 0) 413 return (ENOSPC); 414 cip->used++; 415 up = cip->array[idx]; 416 cip = (struct indir *)up; 417 continue; 418 } 419 /* leafnode */ 420 idx = offset & NMASK; 421 up = cip->array[idx]; 422 if (up != 0) 423 cip->used--; 424 cip->array[idx] = ptr; 425 if (ptr != 0) 426 cip->used++; 427 break; 428 } 429 if (cip->used != 0 || li == 1) 430 return (0); 431 li--; 432 while (cip->used == 0 && cip != ip) { 433 li--; 434 idx = (offset >> lip[li]->shift) & NMASK; 435 up = lip[li]->array[idx]; 436 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 437 del_indir(cip); 438 lip[li]->array[idx] = 0; 439 lip[li]->used--; 440 cip = lip[li]; 441 } 442 return (0); 443 } 444 445 static int 446 g_md_access(struct g_provider *pp, int r, int w, int e) 447 { 448 struct md_s *sc; 449 450 sc = pp->geom->softc; 451 if (sc == NULL) { 452 if (r <= 0 && w <= 0 && e <= 0) 453 return (0); 454 return (ENXIO); 455 } 456 r += pp->acr; 457 w += pp->acw; 458 e += pp->ace; 459 if ((sc->flags & MD_READONLY) != 0 && w > 0) 460 return (EROFS); 461 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 462 sc->opencount = 1; 463 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 464 sc->opencount = 0; 465 } 466 return (0); 467 } 468 469 static void 470 g_md_start(struct bio *bp) 471 { 472 struct md_s *sc; 473 474 sc = bp->bio_to->geom->softc; 475 if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) { 476 devstat_start_transaction_bio(sc->devstat, bp); 477 } 478 mtx_lock(&sc->queue_mtx); 479 bioq_disksort(&sc->bio_queue, bp); 480 wakeup(sc); 481 mtx_unlock(&sc->queue_mtx); 482 } 483 484 #define MD_MALLOC_MOVE_ZERO 1 485 #define MD_MALLOC_MOVE_FILL 2 486 #define MD_MALLOC_MOVE_READ 3 487 #define MD_MALLOC_MOVE_WRITE 4 488 #define MD_MALLOC_MOVE_CMP 5 489 490 static int 491 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize, 492 void *ptr, u_char fill, int op) 493 { 494 struct sf_buf *sf; 495 vm_page_t m, *mp1; 496 char *p, first; 497 off_t *uc; 498 unsigned n; 499 int error, i, ma_offs1, sz, first_read; 500 501 m = NULL; 502 error = 0; 503 sf = NULL; 504 /* if (op == MD_MALLOC_MOVE_CMP) { gcc */ 505 first = 0; 506 first_read = 0; 507 uc = ptr; 508 mp1 = *mp; 509 ma_offs1 = *ma_offs; 510 /* } */ 511 sched_pin(); 512 for (n = sectorsize; n != 0; n -= sz) { 513 sz = imin(PAGE_SIZE - *ma_offs, n); 514 if (m != **mp) { 515 if (sf != NULL) 516 sf_buf_free(sf); 517 m = **mp; 518 sf = sf_buf_alloc(m, SFB_CPUPRIVATE | 519 (md_malloc_wait ? 0 : SFB_NOWAIT)); 520 if (sf == NULL) { 521 error = ENOMEM; 522 break; 523 } 524 } 525 p = (char *)sf_buf_kva(sf) + *ma_offs; 526 switch (op) { 527 case MD_MALLOC_MOVE_ZERO: 528 bzero(p, sz); 529 break; 530 case MD_MALLOC_MOVE_FILL: 531 memset(p, fill, sz); 532 break; 533 case MD_MALLOC_MOVE_READ: 534 bcopy(ptr, p, sz); 535 cpu_flush_dcache(p, sz); 536 break; 537 case MD_MALLOC_MOVE_WRITE: 538 bcopy(p, ptr, sz); 539 break; 540 case MD_MALLOC_MOVE_CMP: 541 for (i = 0; i < sz; i++, p++) { 542 if (!first_read) { 543 *uc = (u_char)*p; 544 first = *p; 545 first_read = 1; 546 } else if (*p != first) { 547 error = EDOOFUS; 548 break; 549 } 550 } 551 break; 552 default: 553 KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op)); 554 break; 555 } 556 if (error != 0) 557 break; 558 *ma_offs += sz; 559 *ma_offs %= PAGE_SIZE; 560 if (*ma_offs == 0) 561 (*mp)++; 562 ptr = (char *)ptr + sz; 563 } 564 565 if (sf != NULL) 566 sf_buf_free(sf); 567 sched_unpin(); 568 if (op == MD_MALLOC_MOVE_CMP && error != 0) { 569 *mp = mp1; 570 *ma_offs = ma_offs1; 571 } 572 return (error); 573 } 574 575 static int 576 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs, 577 unsigned len, void *ptr, u_char fill, int op) 578 { 579 bus_dma_segment_t *vlist; 580 uint8_t *p, *end, first; 581 off_t *uc; 582 int ma_offs, seg_len; 583 584 vlist = *pvlist; 585 ma_offs = *pma_offs; 586 uc = ptr; 587 588 for (; len != 0; len -= seg_len) { 589 seg_len = imin(vlist->ds_len - ma_offs, len); 590 p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs; 591 switch (op) { 592 case MD_MALLOC_MOVE_ZERO: 593 bzero(p, seg_len); 594 break; 595 case MD_MALLOC_MOVE_FILL: 596 memset(p, fill, seg_len); 597 break; 598 case MD_MALLOC_MOVE_READ: 599 bcopy(ptr, p, seg_len); 600 cpu_flush_dcache(p, seg_len); 601 break; 602 case MD_MALLOC_MOVE_WRITE: 603 bcopy(p, ptr, seg_len); 604 break; 605 case MD_MALLOC_MOVE_CMP: 606 end = p + seg_len; 607 first = *uc = *p; 608 /* Confirm all following bytes match the first */ 609 while (++p < end) { 610 if (*p != first) 611 return (EDOOFUS); 612 } 613 break; 614 default: 615 KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op)); 616 break; 617 } 618 619 ma_offs += seg_len; 620 if (ma_offs == vlist->ds_len) { 621 ma_offs = 0; 622 vlist++; 623 } 624 ptr = (uint8_t *)ptr + seg_len; 625 } 626 *pvlist = vlist; 627 *pma_offs = ma_offs; 628 629 return (0); 630 } 631 632 static int 633 mdstart_malloc(struct md_s *sc, struct bio *bp) 634 { 635 u_char *dst; 636 vm_page_t *m; 637 bus_dma_segment_t *vlist; 638 int i, error, error1, ma_offs, notmapped; 639 off_t secno, nsec, uc; 640 uintptr_t sp, osp; 641 642 switch (bp->bio_cmd) { 643 case BIO_READ: 644 case BIO_WRITE: 645 case BIO_DELETE: 646 break; 647 default: 648 return (EOPNOTSUPP); 649 } 650 651 notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0; 652 vlist = (bp->bio_flags & BIO_VLIST) != 0 ? 653 (bus_dma_segment_t *)bp->bio_data : NULL; 654 if (notmapped) { 655 m = bp->bio_ma; 656 ma_offs = bp->bio_ma_offset; 657 dst = NULL; 658 KASSERT(vlist == NULL, ("vlists cannot be unmapped")); 659 } else if (vlist != NULL) { 660 ma_offs = bp->bio_ma_offset; 661 dst = NULL; 662 } else { 663 dst = bp->bio_data; 664 } 665 666 nsec = bp->bio_length / sc->sectorsize; 667 secno = bp->bio_offset / sc->sectorsize; 668 error = 0; 669 while (nsec--) { 670 osp = s_read(sc->indir, secno); 671 if (bp->bio_cmd == BIO_DELETE) { 672 if (osp != 0) 673 error = s_write(sc->indir, secno, 0); 674 } else if (bp->bio_cmd == BIO_READ) { 675 if (osp == 0) { 676 if (notmapped) { 677 error = md_malloc_move_ma(&m, &ma_offs, 678 sc->sectorsize, NULL, 0, 679 MD_MALLOC_MOVE_ZERO); 680 } else if (vlist != NULL) { 681 error = md_malloc_move_vlist(&vlist, 682 &ma_offs, sc->sectorsize, NULL, 0, 683 MD_MALLOC_MOVE_ZERO); 684 } else 685 bzero(dst, sc->sectorsize); 686 } else if (osp <= 255) { 687 if (notmapped) { 688 error = md_malloc_move_ma(&m, &ma_offs, 689 sc->sectorsize, NULL, osp, 690 MD_MALLOC_MOVE_FILL); 691 } else if (vlist != NULL) { 692 error = md_malloc_move_vlist(&vlist, 693 &ma_offs, sc->sectorsize, NULL, osp, 694 MD_MALLOC_MOVE_FILL); 695 } else 696 memset(dst, osp, sc->sectorsize); 697 } else { 698 if (notmapped) { 699 error = md_malloc_move_ma(&m, &ma_offs, 700 sc->sectorsize, (void *)osp, 0, 701 MD_MALLOC_MOVE_READ); 702 } else if (vlist != NULL) { 703 error = md_malloc_move_vlist(&vlist, 704 &ma_offs, sc->sectorsize, 705 (void *)osp, 0, 706 MD_MALLOC_MOVE_READ); 707 } else { 708 bcopy((void *)osp, dst, sc->sectorsize); 709 cpu_flush_dcache(dst, sc->sectorsize); 710 } 711 } 712 osp = 0; 713 } else if (bp->bio_cmd == BIO_WRITE) { 714 if (sc->flags & MD_COMPRESS) { 715 if (notmapped) { 716 error1 = md_malloc_move_ma(&m, &ma_offs, 717 sc->sectorsize, &uc, 0, 718 MD_MALLOC_MOVE_CMP); 719 i = error1 == 0 ? sc->sectorsize : 0; 720 } else if (vlist != NULL) { 721 error1 = md_malloc_move_vlist(&vlist, 722 &ma_offs, sc->sectorsize, &uc, 0, 723 MD_MALLOC_MOVE_CMP); 724 i = error1 == 0 ? sc->sectorsize : 0; 725 } else { 726 uc = dst[0]; 727 for (i = 1; i < sc->sectorsize; i++) { 728 if (dst[i] != uc) 729 break; 730 } 731 } 732 } else { 733 i = 0; 734 uc = 0; 735 } 736 if (i == sc->sectorsize) { 737 if (osp != uc) 738 error = s_write(sc->indir, secno, uc); 739 } else { 740 if (osp <= 255) { 741 sp = (uintptr_t)uma_zalloc(sc->uma, 742 md_malloc_wait ? M_WAITOK : 743 M_NOWAIT); 744 if (sp == 0) { 745 error = ENOSPC; 746 break; 747 } 748 if (notmapped) { 749 error = md_malloc_move_ma(&m, 750 &ma_offs, sc->sectorsize, 751 (void *)sp, 0, 752 MD_MALLOC_MOVE_WRITE); 753 } else if (vlist != NULL) { 754 error = md_malloc_move_vlist( 755 &vlist, &ma_offs, 756 sc->sectorsize, (void *)sp, 757 0, MD_MALLOC_MOVE_WRITE); 758 } else { 759 bcopy(dst, (void *)sp, 760 sc->sectorsize); 761 } 762 error = s_write(sc->indir, secno, sp); 763 } else { 764 if (notmapped) { 765 error = md_malloc_move_ma(&m, 766 &ma_offs, sc->sectorsize, 767 (void *)osp, 0, 768 MD_MALLOC_MOVE_WRITE); 769 } else if (vlist != NULL) { 770 error = md_malloc_move_vlist( 771 &vlist, &ma_offs, 772 sc->sectorsize, (void *)osp, 773 0, MD_MALLOC_MOVE_WRITE); 774 } else { 775 bcopy(dst, (void *)osp, 776 sc->sectorsize); 777 } 778 osp = 0; 779 } 780 } 781 } else { 782 error = EOPNOTSUPP; 783 } 784 if (osp > 255) 785 uma_zfree(sc->uma, (void*)osp); 786 if (error != 0) 787 break; 788 secno++; 789 if (!notmapped && vlist == NULL) 790 dst += sc->sectorsize; 791 } 792 bp->bio_resid = 0; 793 return (error); 794 } 795 796 static void 797 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len) 798 { 799 off_t seg_len; 800 801 while (offset >= vlist->ds_len) { 802 offset -= vlist->ds_len; 803 vlist++; 804 } 805 806 while (len != 0) { 807 seg_len = omin(len, vlist->ds_len - offset); 808 bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset), 809 seg_len); 810 offset = 0; 811 src = (uint8_t *)src + seg_len; 812 len -= seg_len; 813 vlist++; 814 } 815 } 816 817 static void 818 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len) 819 { 820 off_t seg_len; 821 822 while (offset >= vlist->ds_len) { 823 offset -= vlist->ds_len; 824 vlist++; 825 } 826 827 while (len != 0) { 828 seg_len = omin(len, vlist->ds_len - offset); 829 bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst, 830 seg_len); 831 offset = 0; 832 dst = (uint8_t *)dst + seg_len; 833 len -= seg_len; 834 vlist++; 835 } 836 } 837 838 static int 839 mdstart_preload(struct md_s *sc, struct bio *bp) 840 { 841 uint8_t *p; 842 843 p = sc->pl_ptr + bp->bio_offset; 844 switch (bp->bio_cmd) { 845 case BIO_READ: 846 if ((bp->bio_flags & BIO_VLIST) != 0) { 847 mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data, 848 bp->bio_ma_offset, bp->bio_length); 849 } else { 850 bcopy(p, bp->bio_data, bp->bio_length); 851 } 852 cpu_flush_dcache(bp->bio_data, bp->bio_length); 853 break; 854 case BIO_WRITE: 855 if ((bp->bio_flags & BIO_VLIST) != 0) { 856 mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data, 857 bp->bio_ma_offset, p, bp->bio_length); 858 } else { 859 bcopy(bp->bio_data, p, bp->bio_length); 860 } 861 break; 862 } 863 bp->bio_resid = 0; 864 return (0); 865 } 866 867 static int 868 mdstart_vnode(struct md_s *sc, struct bio *bp) 869 { 870 int error; 871 struct uio auio; 872 struct iovec aiov; 873 struct iovec *piov; 874 struct mount *mp; 875 struct vnode *vp; 876 struct buf *pb; 877 bus_dma_segment_t *vlist; 878 struct thread *td; 879 off_t iolen, iostart, len, zerosize; 880 int ma_offs, npages; 881 882 switch (bp->bio_cmd) { 883 case BIO_READ: 884 auio.uio_rw = UIO_READ; 885 break; 886 case BIO_WRITE: 887 case BIO_DELETE: 888 auio.uio_rw = UIO_WRITE; 889 break; 890 case BIO_FLUSH: 891 break; 892 default: 893 return (EOPNOTSUPP); 894 } 895 896 td = curthread; 897 vp = sc->vnode; 898 pb = NULL; 899 piov = NULL; 900 ma_offs = bp->bio_ma_offset; 901 len = bp->bio_length; 902 903 /* 904 * VNODE I/O 905 * 906 * If an error occurs, we set BIO_ERROR but we do not set 907 * B_INVAL because (for a write anyway), the buffer is 908 * still valid. 909 */ 910 911 if (bp->bio_cmd == BIO_FLUSH) { 912 (void) vn_start_write(vp, &mp, V_WAIT); 913 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 914 error = VOP_FSYNC(vp, MNT_WAIT, td); 915 VOP_UNLOCK(vp); 916 vn_finished_write(mp); 917 return (error); 918 } 919 920 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 921 auio.uio_resid = bp->bio_length; 922 auio.uio_segflg = UIO_SYSSPACE; 923 auio.uio_td = td; 924 925 if (bp->bio_cmd == BIO_DELETE) { 926 /* 927 * Emulate BIO_DELETE by writing zeros. 928 */ 929 zerosize = ZERO_REGION_SIZE - 930 (ZERO_REGION_SIZE % sc->sectorsize); 931 auio.uio_iovcnt = howmany(bp->bio_length, zerosize); 932 piov = malloc(sizeof(*piov) * auio.uio_iovcnt, M_MD, M_WAITOK); 933 auio.uio_iov = piov; 934 while (len > 0) { 935 piov->iov_base = __DECONST(void *, zero_region); 936 piov->iov_len = len; 937 if (len > zerosize) 938 piov->iov_len = zerosize; 939 len -= piov->iov_len; 940 piov++; 941 } 942 piov = auio.uio_iov; 943 } else if ((bp->bio_flags & BIO_VLIST) != 0) { 944 piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK); 945 auio.uio_iov = piov; 946 vlist = (bus_dma_segment_t *)bp->bio_data; 947 while (len > 0) { 948 piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr + 949 ma_offs); 950 piov->iov_len = vlist->ds_len - ma_offs; 951 if (piov->iov_len > len) 952 piov->iov_len = len; 953 len -= piov->iov_len; 954 ma_offs = 0; 955 vlist++; 956 piov++; 957 } 958 auio.uio_iovcnt = piov - auio.uio_iov; 959 piov = auio.uio_iov; 960 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 961 pb = uma_zalloc(md_pbuf_zone, M_WAITOK); 962 MPASS((pb->b_flags & B_MAXPHYS) != 0); 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 } else { 1015 bp->bio_resid = auio.uio_resid; 1016 } 1017 1018 free(piov, M_MD); 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 /* 1149 * The page may be deactivated prior to setting 1150 * PGA_REFERENCED, but in this case it will be 1151 * reactivated by the page daemon. 1152 */ 1153 if (vm_page_active(m)) 1154 vm_page_reference(m); 1155 else 1156 vm_page_activate(m); 1157 vm_page_xunbusy(m); 1158 } 1159 1160 /* Actions on further pages start at offset 0 */ 1161 p += PAGE_SIZE - offs; 1162 offs = 0; 1163 ma_offs += len; 1164 } 1165 vm_object_pip_wakeup(sc->object); 1166 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC); 1167 } 1168 1169 static int 1170 mdstart_null(struct md_s *sc, struct bio *bp) 1171 { 1172 1173 switch (bp->bio_cmd) { 1174 case BIO_READ: 1175 bzero(bp->bio_data, bp->bio_length); 1176 cpu_flush_dcache(bp->bio_data, bp->bio_length); 1177 break; 1178 case BIO_WRITE: 1179 break; 1180 } 1181 bp->bio_resid = 0; 1182 return (0); 1183 } 1184 1185 static void 1186 md_handleattr(struct md_s *sc, struct bio *bp) 1187 { 1188 if (sc->fwsectors && sc->fwheads && 1189 (g_handleattr_int(bp, "GEOM::fwsectors", sc->fwsectors) != 0 || 1190 g_handleattr_int(bp, "GEOM::fwheads", sc->fwheads) != 0)) 1191 return; 1192 if (g_handleattr_int(bp, "GEOM::candelete", 1) != 0) 1193 return; 1194 if (sc->ident[0] != '\0' && 1195 g_handleattr_str(bp, "GEOM::ident", sc->ident) != 0) 1196 return; 1197 if (g_handleattr_int(bp, "MNT::verified", (sc->flags & MD_VERIFY) != 0)) 1198 return; 1199 g_io_deliver(bp, EOPNOTSUPP); 1200 } 1201 1202 static void 1203 md_kthread(void *arg) 1204 { 1205 struct md_s *sc; 1206 struct bio *bp; 1207 int error; 1208 1209 sc = arg; 1210 thread_lock(curthread); 1211 sched_prio(curthread, PRIBIO); 1212 thread_unlock(curthread); 1213 if (sc->type == MD_VNODE) 1214 curthread->td_pflags |= TDP_NORUNNINGBUF; 1215 1216 for (;;) { 1217 mtx_lock(&sc->queue_mtx); 1218 if (sc->flags & MD_SHUTDOWN) { 1219 sc->flags |= MD_EXITING; 1220 mtx_unlock(&sc->queue_mtx); 1221 kproc_exit(0); 1222 } 1223 bp = bioq_takefirst(&sc->bio_queue); 1224 if (!bp) { 1225 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 1226 continue; 1227 } 1228 mtx_unlock(&sc->queue_mtx); 1229 if (bp->bio_cmd == BIO_GETATTR) { 1230 md_handleattr(sc, bp); 1231 } else { 1232 error = sc->start(sc, bp); 1233 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) { 1234 /* 1235 * Devstat uses (bio_bcount, bio_resid) for 1236 * determining the length of the completed part 1237 * of the i/o. g_io_deliver() will translate 1238 * from bio_completed to that, but it also 1239 * destroys the bio so we must do our own 1240 * translation. 1241 */ 1242 bp->bio_bcount = bp->bio_length; 1243 devstat_end_transaction_bio(sc->devstat, bp); 1244 } 1245 bp->bio_completed = bp->bio_length - bp->bio_resid; 1246 g_io_deliver(bp, error); 1247 } 1248 } 1249 } 1250 1251 static struct md_s * 1252 mdfind(int unit) 1253 { 1254 struct md_s *sc; 1255 1256 LIST_FOREACH(sc, &md_softc_list, list) { 1257 if (sc->unit == unit) 1258 break; 1259 } 1260 return (sc); 1261 } 1262 1263 static struct md_s * 1264 mdnew(int unit, int *errp, enum md_types type) 1265 { 1266 struct md_s *sc; 1267 int error; 1268 1269 *errp = 0; 1270 if (unit == -1) 1271 unit = alloc_unr(md_uh); 1272 else 1273 unit = alloc_unr_specific(md_uh, unit); 1274 1275 if (unit == -1) { 1276 *errp = EBUSY; 1277 return (NULL); 1278 } 1279 1280 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 1281 sc->type = type; 1282 bioq_init(&sc->bio_queue); 1283 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 1284 sc->unit = unit; 1285 sprintf(sc->name, "md%d", unit); 1286 LIST_INSERT_HEAD(&md_softc_list, sc, list); 1287 error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 1288 if (error == 0) 1289 return (sc); 1290 LIST_REMOVE(sc, list); 1291 mtx_destroy(&sc->queue_mtx); 1292 free_unr(md_uh, sc->unit); 1293 free(sc, M_MD); 1294 *errp = error; 1295 return (NULL); 1296 } 1297 1298 static void 1299 mdinit(struct md_s *sc) 1300 { 1301 struct g_geom *gp; 1302 struct g_provider *pp; 1303 1304 g_topology_lock(); 1305 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 1306 gp->softc = sc; 1307 pp = g_new_providerf(gp, "md%d", sc->unit); 1308 devstat_remove_entry(pp->stat); 1309 pp->stat = NULL; 1310 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE; 1311 pp->mediasize = sc->mediasize; 1312 pp->sectorsize = sc->sectorsize; 1313 switch (sc->type) { 1314 case MD_MALLOC: 1315 case MD_VNODE: 1316 case MD_SWAP: 1317 pp->flags |= G_PF_ACCEPT_UNMAPPED; 1318 break; 1319 case MD_PRELOAD: 1320 case MD_NULL: 1321 break; 1322 } 1323 sc->gp = gp; 1324 sc->pp = pp; 1325 sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize, 1326 DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX); 1327 sc->devstat->id = pp; 1328 g_error_provider(pp, 0); 1329 g_topology_unlock(); 1330 } 1331 1332 static int 1333 mdcreate_malloc(struct md_s *sc, struct md_req *mdr) 1334 { 1335 uintptr_t sp; 1336 int error; 1337 off_t u; 1338 1339 error = 0; 1340 if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 1341 return (EINVAL); 1342 if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize)) 1343 return (EINVAL); 1344 /* Compression doesn't make sense if we have reserved space */ 1345 if (mdr->md_options & MD_RESERVE) 1346 mdr->md_options &= ~MD_COMPRESS; 1347 if (mdr->md_fwsectors != 0) 1348 sc->fwsectors = mdr->md_fwsectors; 1349 if (mdr->md_fwheads != 0) 1350 sc->fwheads = mdr->md_fwheads; 1351 sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE); 1352 sc->indir = dimension(sc->mediasize / sc->sectorsize); 1353 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL, 1354 0x1ff, 0); 1355 if (mdr->md_options & MD_RESERVE) { 1356 off_t nsectors; 1357 1358 nsectors = sc->mediasize / sc->sectorsize; 1359 for (u = 0; u < nsectors; u++) { 1360 sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ? 1361 M_WAITOK : M_NOWAIT) | M_ZERO); 1362 if (sp != 0) 1363 error = s_write(sc->indir, u, sp); 1364 else 1365 error = ENOMEM; 1366 if (error != 0) 1367 break; 1368 } 1369 } 1370 return (error); 1371 } 1372 1373 static int 1374 mdsetcred(struct md_s *sc, struct ucred *cred) 1375 { 1376 char *tmpbuf; 1377 int error = 0; 1378 1379 /* 1380 * Set credits in our softc 1381 */ 1382 1383 if (sc->cred) 1384 crfree(sc->cred); 1385 sc->cred = crhold(cred); 1386 1387 /* 1388 * Horrible kludge to establish credentials for NFS XXX. 1389 */ 1390 1391 if (sc->vnode) { 1392 struct uio auio; 1393 struct iovec aiov; 1394 1395 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK); 1396 bzero(&auio, sizeof(auio)); 1397 1398 aiov.iov_base = tmpbuf; 1399 aiov.iov_len = sc->sectorsize; 1400 auio.uio_iov = &aiov; 1401 auio.uio_iovcnt = 1; 1402 auio.uio_offset = 0; 1403 auio.uio_rw = UIO_READ; 1404 auio.uio_segflg = UIO_SYSSPACE; 1405 auio.uio_resid = aiov.iov_len; 1406 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 1407 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 1408 VOP_UNLOCK(sc->vnode); 1409 free(tmpbuf, M_TEMP); 1410 } 1411 return (error); 1412 } 1413 1414 static int 1415 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td) 1416 { 1417 struct vattr vattr; 1418 struct nameidata nd; 1419 char *fname; 1420 int error, flags; 1421 1422 fname = mdr->md_file; 1423 if (mdr->md_file_seg == UIO_USERSPACE) { 1424 error = copyinstr(fname, sc->file, sizeof(sc->file), NULL); 1425 if (error != 0) 1426 return (error); 1427 } else if (mdr->md_file_seg == UIO_SYSSPACE) 1428 strlcpy(sc->file, fname, sizeof(sc->file)); 1429 else 1430 return (EDOOFUS); 1431 1432 /* 1433 * If the user specified that this is a read only device, don't 1434 * set the FWRITE mask before trying to open the backing store. 1435 */ 1436 flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \ 1437 | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0); 1438 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td); 1439 error = vn_open(&nd, &flags, 0, NULL); 1440 if (error != 0) 1441 return (error); 1442 NDFREE(&nd, NDF_ONLY_PNBUF); 1443 if (nd.ni_vp->v_type != VREG) { 1444 error = EINVAL; 1445 goto bad; 1446 } 1447 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred); 1448 if (error != 0) 1449 goto bad; 1450 if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) { 1451 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY); 1452 if (VN_IS_DOOMED(nd.ni_vp)) { 1453 /* Forced unmount. */ 1454 error = EBADF; 1455 goto bad; 1456 } 1457 } 1458 nd.ni_vp->v_vflag |= VV_MD; 1459 VOP_UNLOCK(nd.ni_vp); 1460 1461 if (mdr->md_fwsectors != 0) 1462 sc->fwsectors = mdr->md_fwsectors; 1463 if (mdr->md_fwheads != 0) 1464 sc->fwheads = mdr->md_fwheads; 1465 snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju", 1466 (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid); 1467 sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE | 1468 MD_VERIFY); 1469 if (!(flags & FWRITE)) 1470 sc->flags |= MD_READONLY; 1471 sc->vnode = nd.ni_vp; 1472 1473 error = mdsetcred(sc, td->td_ucred); 1474 if (error != 0) { 1475 sc->vnode = NULL; 1476 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY); 1477 nd.ni_vp->v_vflag &= ~VV_MD; 1478 goto bad; 1479 } 1480 return (0); 1481 bad: 1482 VOP_UNLOCK(nd.ni_vp); 1483 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td); 1484 return (error); 1485 } 1486 1487 static void 1488 g_md_providergone(struct g_provider *pp) 1489 { 1490 struct md_s *sc = pp->geom->softc; 1491 1492 mtx_lock(&sc->queue_mtx); 1493 sc->flags |= MD_PROVIDERGONE; 1494 wakeup(&sc->flags); 1495 mtx_unlock(&sc->queue_mtx); 1496 } 1497 1498 static int 1499 mddestroy(struct md_s *sc, struct thread *td) 1500 { 1501 1502 if (sc->gp) { 1503 g_topology_lock(); 1504 g_wither_geom(sc->gp, ENXIO); 1505 g_topology_unlock(); 1506 1507 mtx_lock(&sc->queue_mtx); 1508 while (!(sc->flags & MD_PROVIDERGONE)) 1509 msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0); 1510 mtx_unlock(&sc->queue_mtx); 1511 } 1512 if (sc->devstat) { 1513 devstat_remove_entry(sc->devstat); 1514 sc->devstat = NULL; 1515 } 1516 mtx_lock(&sc->queue_mtx); 1517 sc->flags |= MD_SHUTDOWN; 1518 wakeup(sc); 1519 while (!(sc->flags & MD_EXITING)) 1520 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10); 1521 mtx_unlock(&sc->queue_mtx); 1522 mtx_destroy(&sc->queue_mtx); 1523 if (sc->vnode != NULL) { 1524 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY); 1525 sc->vnode->v_vflag &= ~VV_MD; 1526 VOP_UNLOCK(sc->vnode); 1527 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 1528 FREAD : (FREAD|FWRITE), sc->cred, td); 1529 } 1530 if (sc->cred != NULL) 1531 crfree(sc->cred); 1532 if (sc->object != NULL) 1533 vm_object_deallocate(sc->object); 1534 if (sc->indir) 1535 destroy_indir(sc, sc->indir); 1536 if (sc->uma) 1537 uma_zdestroy(sc->uma); 1538 1539 LIST_REMOVE(sc, list); 1540 free_unr(md_uh, sc->unit); 1541 free(sc, M_MD); 1542 return (0); 1543 } 1544 1545 static int 1546 mdresize(struct md_s *sc, struct md_req *mdr) 1547 { 1548 int error, res; 1549 vm_pindex_t oldpages, newpages; 1550 1551 switch (sc->type) { 1552 case MD_VNODE: 1553 case MD_NULL: 1554 break; 1555 case MD_SWAP: 1556 if (mdr->md_mediasize <= 0 || 1557 (mdr->md_mediasize % PAGE_SIZE) != 0) 1558 return (EDOM); 1559 oldpages = OFF_TO_IDX(sc->mediasize); 1560 newpages = OFF_TO_IDX(mdr->md_mediasize); 1561 if (newpages < oldpages) { 1562 VM_OBJECT_WLOCK(sc->object); 1563 vm_object_page_remove(sc->object, newpages, 0, 0); 1564 swap_release_by_cred(IDX_TO_OFF(oldpages - 1565 newpages), sc->cred); 1566 sc->object->charge = IDX_TO_OFF(newpages); 1567 sc->object->size = newpages; 1568 VM_OBJECT_WUNLOCK(sc->object); 1569 } else if (newpages > oldpages) { 1570 res = swap_reserve_by_cred(IDX_TO_OFF(newpages - 1571 oldpages), sc->cred); 1572 if (!res) 1573 return (ENOMEM); 1574 if ((mdr->md_options & MD_RESERVE) || 1575 (sc->flags & MD_RESERVE)) { 1576 error = swap_pager_reserve(sc->object, 1577 oldpages, newpages - oldpages); 1578 if (error < 0) { 1579 swap_release_by_cred( 1580 IDX_TO_OFF(newpages - oldpages), 1581 sc->cred); 1582 return (EDOM); 1583 } 1584 } 1585 VM_OBJECT_WLOCK(sc->object); 1586 sc->object->charge = IDX_TO_OFF(newpages); 1587 sc->object->size = newpages; 1588 VM_OBJECT_WUNLOCK(sc->object); 1589 } 1590 break; 1591 default: 1592 return (EOPNOTSUPP); 1593 } 1594 1595 sc->mediasize = mdr->md_mediasize; 1596 1597 g_topology_lock(); 1598 g_resize_provider(sc->pp, sc->mediasize); 1599 g_topology_unlock(); 1600 return (0); 1601 } 1602 1603 static int 1604 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td) 1605 { 1606 vm_ooffset_t npage; 1607 int error; 1608 1609 /* 1610 * Range check. Disallow negative sizes and sizes not being 1611 * multiple of page size. 1612 */ 1613 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0) 1614 return (EDOM); 1615 1616 /* 1617 * Allocate an OBJT_SWAP object. 1618 * 1619 * Note the truncation. 1620 */ 1621 1622 if ((mdr->md_options & MD_VERIFY) != 0) 1623 return (EINVAL); 1624 npage = mdr->md_mediasize / PAGE_SIZE; 1625 if (mdr->md_fwsectors != 0) 1626 sc->fwsectors = mdr->md_fwsectors; 1627 if (mdr->md_fwheads != 0) 1628 sc->fwheads = mdr->md_fwheads; 1629 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage, 1630 VM_PROT_DEFAULT, 0, td->td_ucred); 1631 if (sc->object == NULL) 1632 return (ENOMEM); 1633 sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE); 1634 if (mdr->md_options & MD_RESERVE) { 1635 if (swap_pager_reserve(sc->object, 0, npage) < 0) { 1636 error = EDOM; 1637 goto finish; 1638 } 1639 } 1640 error = mdsetcred(sc, td->td_ucred); 1641 finish: 1642 if (error != 0) { 1643 vm_object_deallocate(sc->object); 1644 sc->object = NULL; 1645 } 1646 return (error); 1647 } 1648 1649 static int 1650 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td) 1651 { 1652 1653 /* 1654 * Range check. Disallow negative sizes and sizes not being 1655 * multiple of page size. 1656 */ 1657 if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0) 1658 return (EDOM); 1659 1660 return (0); 1661 } 1662 1663 static int 1664 kern_mdattach_locked(struct thread *td, struct md_req *mdr) 1665 { 1666 struct md_s *sc; 1667 unsigned sectsize; 1668 int error, i; 1669 1670 sx_assert(&md_sx, SA_XLOCKED); 1671 1672 switch (mdr->md_type) { 1673 case MD_MALLOC: 1674 case MD_PRELOAD: 1675 case MD_VNODE: 1676 case MD_SWAP: 1677 case MD_NULL: 1678 break; 1679 default: 1680 return (EINVAL); 1681 } 1682 if (mdr->md_sectorsize == 0) 1683 sectsize = DEV_BSIZE; 1684 else 1685 sectsize = mdr->md_sectorsize; 1686 if (sectsize > maxphys || mdr->md_mediasize < sectsize) 1687 return (EINVAL); 1688 if (mdr->md_options & MD_AUTOUNIT) 1689 sc = mdnew(-1, &error, mdr->md_type); 1690 else { 1691 if (mdr->md_unit > INT_MAX) 1692 return (EINVAL); 1693 sc = mdnew(mdr->md_unit, &error, mdr->md_type); 1694 } 1695 if (sc == NULL) 1696 return (error); 1697 if (mdr->md_label != NULL) 1698 error = copyinstr(mdr->md_label, sc->label, 1699 sizeof(sc->label), NULL); 1700 if (error != 0) 1701 goto err_after_new; 1702 if (mdr->md_options & MD_AUTOUNIT) 1703 mdr->md_unit = sc->unit; 1704 sc->mediasize = mdr->md_mediasize; 1705 sc->sectorsize = sectsize; 1706 error = EDOOFUS; 1707 switch (sc->type) { 1708 case MD_MALLOC: 1709 sc->start = mdstart_malloc; 1710 error = mdcreate_malloc(sc, mdr); 1711 break; 1712 case MD_PRELOAD: 1713 /* 1714 * We disallow attaching preloaded memory disks via 1715 * ioctl. Preloaded memory disks are automatically 1716 * attached in g_md_init(). 1717 */ 1718 error = EOPNOTSUPP; 1719 break; 1720 case MD_VNODE: 1721 sc->start = mdstart_vnode; 1722 error = mdcreate_vnode(sc, mdr, td); 1723 break; 1724 case MD_SWAP: 1725 sc->start = mdstart_swap; 1726 error = mdcreate_swap(sc, mdr, td); 1727 break; 1728 case MD_NULL: 1729 sc->start = mdstart_null; 1730 error = mdcreate_null(sc, mdr, td); 1731 break; 1732 } 1733 err_after_new: 1734 if (error != 0) { 1735 mddestroy(sc, td); 1736 return (error); 1737 } 1738 1739 /* Prune off any residual fractional sector */ 1740 i = sc->mediasize % sc->sectorsize; 1741 sc->mediasize -= i; 1742 1743 mdinit(sc); 1744 return (0); 1745 } 1746 1747 static int 1748 kern_mdattach(struct thread *td, struct md_req *mdr) 1749 { 1750 int error; 1751 1752 sx_xlock(&md_sx); 1753 error = kern_mdattach_locked(td, mdr); 1754 sx_xunlock(&md_sx); 1755 return (error); 1756 } 1757 1758 static int 1759 kern_mddetach_locked(struct thread *td, struct md_req *mdr) 1760 { 1761 struct md_s *sc; 1762 1763 sx_assert(&md_sx, SA_XLOCKED); 1764 1765 if (mdr->md_mediasize != 0 || 1766 (mdr->md_options & ~MD_FORCE) != 0) 1767 return (EINVAL); 1768 1769 sc = mdfind(mdr->md_unit); 1770 if (sc == NULL) 1771 return (ENOENT); 1772 if (sc->opencount != 0 && !(sc->flags & MD_FORCE) && 1773 !(mdr->md_options & MD_FORCE)) 1774 return (EBUSY); 1775 return (mddestroy(sc, td)); 1776 } 1777 1778 static int 1779 kern_mddetach(struct thread *td, struct md_req *mdr) 1780 { 1781 int error; 1782 1783 sx_xlock(&md_sx); 1784 error = kern_mddetach_locked(td, mdr); 1785 sx_xunlock(&md_sx); 1786 return (error); 1787 } 1788 1789 static int 1790 kern_mdresize_locked(struct md_req *mdr) 1791 { 1792 struct md_s *sc; 1793 1794 sx_assert(&md_sx, SA_XLOCKED); 1795 1796 if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0) 1797 return (EINVAL); 1798 1799 sc = mdfind(mdr->md_unit); 1800 if (sc == NULL) 1801 return (ENOENT); 1802 if (mdr->md_mediasize < sc->sectorsize) 1803 return (EINVAL); 1804 mdr->md_mediasize -= mdr->md_mediasize % sc->sectorsize; 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