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