Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/dev/md/md.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /*- 2 * ---------------------------------------------------------------------------- 3 * "THE BEER-WARE LICENSE" (Revision 42): 4 * <phk@FreeBSD.ORG> wrote this file. As long as you retain this notice you 5 * can do whatever you want with this stuff. If we meet some day, and you think 6 * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp 7 * ---------------------------------------------------------------------------- 8 * 9 * $FreeBSD$ 10 * 11 */ 12 13 /*- 14 * The following functions are based in the vn(4) driver: mdstart_swap(), 15 * mdstart_vnode(), mdcreate_swap(), mdcreate_vnode() and mddestroy(), 16 * and as such under the following copyright: 17 * 18 * Copyright (c) 1988 University of Utah. 19 * Copyright (c) 1990, 1993 20 * The Regents of the University of California. All rights reserved. 21 * 22 * This code is derived from software contributed to Berkeley by 23 * the Systems Programming Group of the University of Utah Computer 24 * Science Department. 25 * 26 * Redistribution and use in source and binary forms, with or without 27 * modification, are permitted provided that the following conditions 28 * are met: 29 * 1. Redistributions of source code must retain the above copyright 30 * notice, this list of conditions and the following disclaimer. 31 * 2. Redistributions in binary form must reproduce the above copyright 32 * notice, this list of conditions and the following disclaimer in the 33 * documentation and/or other materials provided with the distribution. 34 * 4. Neither the name of the University nor the names of its contributors 35 * may be used to endorse or promote products derived from this software 36 * without specific prior written permission. 37 * 38 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 39 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 40 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 41 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 42 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 43 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 44 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 45 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 46 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 47 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 48 * SUCH DAMAGE. 49 * 50 * from: Utah Hdr: vn.c 1.13 94/04/02 51 * 52 * from: @(#)vn.c 8.6 (Berkeley) 4/1/94 53 * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03 54 */ 55 56 #include "opt_geom.h" 57 #include "opt_md.h" 58 59 #include <sys/param.h> 60 #include <sys/systm.h> 61 #include <sys/bio.h> 62 #include <sys/conf.h> 63 #include <sys/fcntl.h> 64 #include <sys/kernel.h> 65 #include <sys/kthread.h> 66 #include <sys/linker.h> 67 #include <sys/lock.h> 68 #include <sys/malloc.h> 69 #include <sys/mdioctl.h> 70 #include <sys/mount.h> 71 #include <sys/mutex.h> 72 #include <sys/sx.h> 73 #include <sys/namei.h> 74 #include <sys/proc.h> 75 #include <sys/queue.h> 76 #include <sys/sched.h> 77 #include <sys/sf_buf.h> 78 #include <sys/sysctl.h> 79 #include <sys/vnode.h> 80 81 #include <geom/geom.h> 82 83 #include <vm/vm.h> 84 #include <vm/vm_object.h> 85 #include <vm/vm_page.h> 86 #include <vm/vm_pager.h> 87 #include <vm/swap_pager.h> 88 #include <vm/uma.h> 89 90 #define MD_MODVER 1 91 92 #define MD_SHUTDOWN 0x10000 /* Tell worker thread to terminate. */ 93 #define MD_EXITING 0x20000 /* Worker thread is exiting. */ 94 95 #ifndef MD_NSECT 96 #define MD_NSECT (10000 * 2) 97 #endif 98 99 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk"); 100 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors"); 101 102 static int md_debug; 103 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, ""); 104 105 #if defined(MD_ROOT) && defined(MD_ROOT_SIZE) 106 /* 107 * Preloaded image gets put here. 108 * Applications that patch the object with the image can determine 109 * the size looking at the start and end markers (strings), 110 * so we want them contiguous. 111 */ 112 static struct { 113 u_char start[MD_ROOT_SIZE*1024]; 114 u_char end[128]; 115 } mfs_root = { 116 .start = "MFS Filesystem goes here", 117 .end = "MFS Filesystem had better STOP here", 118 }; 119 #endif 120 121 static g_init_t g_md_init; 122 static g_fini_t g_md_fini; 123 static g_start_t g_md_start; 124 static g_access_t g_md_access; 125 static void g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 126 struct g_consumer *cp __unused, struct g_provider *pp); 127 128 static int mdunits; 129 static struct cdev *status_dev = 0; 130 static struct sx md_sx; 131 132 static d_ioctl_t mdctlioctl; 133 134 static struct cdevsw mdctl_cdevsw = { 135 .d_version = D_VERSION, 136 .d_ioctl = mdctlioctl, 137 .d_name = MD_NAME, 138 }; 139 140 struct g_class g_md_class = { 141 .name = "MD", 142 .version = G_VERSION, 143 .init = g_md_init, 144 .fini = g_md_fini, 145 .start = g_md_start, 146 .access = g_md_access, 147 .dumpconf = g_md_dumpconf, 148 }; 149 150 DECLARE_GEOM_CLASS(g_md_class, g_md); 151 152 153 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(&md_softc_list); 154 155 #define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 156 #define NMASK (NINDIR-1) 157 static int nshift; 158 159 struct indir { 160 uintptr_t *array; 161 u_int total; 162 u_int used; 163 u_int shift; 164 }; 165 166 struct md_s { 167 int unit; 168 LIST_ENTRY(md_s) list; 169 struct bio_queue_head bio_queue; 170 struct mtx queue_mtx; 171 struct cdev *dev; 172 enum md_types type; 173 off_t mediasize; 174 unsigned sectorsize; 175 unsigned opencount; 176 unsigned fwheads; 177 unsigned fwsectors; 178 unsigned flags; 179 char name[20]; 180 struct proc *procp; 181 struct g_geom *gp; 182 struct g_provider *pp; 183 int (*start)(struct md_s *sc, struct bio *bp); 184 185 /* MD_MALLOC related fields */ 186 struct indir *indir; 187 uma_zone_t uma; 188 189 /* MD_PRELOAD related fields */ 190 u_char *pl_ptr; 191 size_t pl_len; 192 193 /* MD_VNODE related fields */ 194 struct vnode *vnode; 195 char file[PATH_MAX]; 196 struct ucred *cred; 197 198 /* MD_SWAP related fields */ 199 vm_object_t object; 200 }; 201 202 static struct indir * 203 new_indir(u_int shift) 204 { 205 struct indir *ip; 206 207 ip = malloc(sizeof *ip, M_MD, M_NOWAIT | M_ZERO); 208 if (ip == NULL) 209 return (NULL); 210 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 211 M_MDSECT, M_NOWAIT | M_ZERO); 212 if (ip->array == NULL) { 213 free(ip, M_MD); 214 return (NULL); 215 } 216 ip->total = NINDIR; 217 ip->shift = shift; 218 return (ip); 219 } 220 221 static void 222 del_indir(struct indir *ip) 223 { 224 225 free(ip->array, M_MDSECT); 226 free(ip, M_MD); 227 } 228 229 static void 230 destroy_indir(struct md_s *sc, struct indir *ip) 231 { 232 int i; 233 234 for (i = 0; i < NINDIR; i++) { 235 if (!ip->array[i]) 236 continue; 237 if (ip->shift) 238 destroy_indir(sc, (struct indir*)(ip->array[i])); 239 else if (ip->array[i] > 255) 240 uma_zfree(sc->uma, (void *)(ip->array[i])); 241 } 242 del_indir(ip); 243 } 244 245 /* 246 * This function does the math and allocates the top level "indir" structure 247 * for a device of "size" sectors. 248 */ 249 250 static struct indir * 251 dimension(off_t size) 252 { 253 off_t rcnt; 254 struct indir *ip; 255 int layer; 256 257 rcnt = size; 258 layer = 0; 259 while (rcnt > NINDIR) { 260 rcnt /= NINDIR; 261 layer++; 262 } 263 264 /* 265 * XXX: the top layer is probably not fully populated, so we allocate 266 * too much space for ip->array in here. 267 */ 268 ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO); 269 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 270 M_MDSECT, M_WAITOK | M_ZERO); 271 ip->total = NINDIR; 272 ip->shift = layer * nshift; 273 return (ip); 274 } 275 276 /* 277 * Read a given sector 278 */ 279 280 static uintptr_t 281 s_read(struct indir *ip, off_t offset) 282 { 283 struct indir *cip; 284 int idx; 285 uintptr_t up; 286 287 if (md_debug > 1) 288 printf("s_read(%jd)\n", (intmax_t)offset); 289 up = 0; 290 for (cip = ip; cip != NULL;) { 291 if (cip->shift) { 292 idx = (offset >> cip->shift) & NMASK; 293 up = cip->array[idx]; 294 cip = (struct indir *)up; 295 continue; 296 } 297 idx = offset & NMASK; 298 return (cip->array[idx]); 299 } 300 return (0); 301 } 302 303 /* 304 * Write a given sector, prune the tree if the value is 0 305 */ 306 307 static int 308 s_write(struct indir *ip, off_t offset, uintptr_t ptr) 309 { 310 struct indir *cip, *lip[10]; 311 int idx, li; 312 uintptr_t up; 313 314 if (md_debug > 1) 315 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 316 up = 0; 317 li = 0; 318 cip = ip; 319 for (;;) { 320 lip[li++] = cip; 321 if (cip->shift) { 322 idx = (offset >> cip->shift) & NMASK; 323 up = cip->array[idx]; 324 if (up != 0) { 325 cip = (struct indir *)up; 326 continue; 327 } 328 /* Allocate branch */ 329 cip->array[idx] = 330 (uintptr_t)new_indir(cip->shift - nshift); 331 if (cip->array[idx] == 0) 332 return (ENOSPC); 333 cip->used++; 334 up = cip->array[idx]; 335 cip = (struct indir *)up; 336 continue; 337 } 338 /* leafnode */ 339 idx = offset & NMASK; 340 up = cip->array[idx]; 341 if (up != 0) 342 cip->used--; 343 cip->array[idx] = ptr; 344 if (ptr != 0) 345 cip->used++; 346 break; 347 } 348 if (cip->used != 0 || li == 1) 349 return (0); 350 li--; 351 while (cip->used == 0 && cip != ip) { 352 li--; 353 idx = (offset >> lip[li]->shift) & NMASK; 354 up = lip[li]->array[idx]; 355 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 356 del_indir(cip); 357 lip[li]->array[idx] = 0; 358 lip[li]->used--; 359 cip = lip[li]; 360 } 361 return (0); 362 } 363 364 365 static int 366 g_md_access(struct g_provider *pp, int r, int w, int e) 367 { 368 struct md_s *sc; 369 370 sc = pp->geom->softc; 371 if (sc == NULL) 372 return (ENXIO); 373 r += pp->acr; 374 w += pp->acw; 375 e += pp->ace; 376 if ((sc->flags & MD_READONLY) != 0 && w > 0) 377 return (EROFS); 378 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 379 sc->opencount = 1; 380 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 381 sc->opencount = 0; 382 } 383 return (0); 384 } 385 386 static void 387 g_md_start(struct bio *bp) 388 { 389 struct md_s *sc; 390 391 sc = bp->bio_to->geom->softc; 392 mtx_lock(&sc->queue_mtx); 393 bioq_disksort(&sc->bio_queue, bp); 394 mtx_unlock(&sc->queue_mtx); 395 wakeup(sc); 396 } 397 398 static int 399 mdstart_malloc(struct md_s *sc, struct bio *bp) 400 { 401 int i, error; 402 u_char *dst; 403 off_t secno, nsec, uc; 404 uintptr_t sp, osp; 405 406 switch (bp->bio_cmd) { 407 case BIO_READ: 408 case BIO_WRITE: 409 case BIO_DELETE: 410 break; 411 default: 412 return (EOPNOTSUPP); 413 } 414 415 nsec = bp->bio_length / sc->sectorsize; 416 secno = bp->bio_offset / sc->sectorsize; 417 dst = bp->bio_data; 418 error = 0; 419 while (nsec--) { 420 osp = s_read(sc->indir, secno); 421 if (bp->bio_cmd == BIO_DELETE) { 422 if (osp != 0) 423 error = s_write(sc->indir, secno, 0); 424 } else if (bp->bio_cmd == BIO_READ) { 425 if (osp == 0) 426 bzero(dst, sc->sectorsize); 427 else if (osp <= 255) 428 for (i = 0; i < sc->sectorsize; i++) 429 dst[i] = osp; 430 else 431 bcopy((void *)osp, dst, sc->sectorsize); 432 osp = 0; 433 } else if (bp->bio_cmd == BIO_WRITE) { 434 if (sc->flags & MD_COMPRESS) { 435 uc = dst[0]; 436 for (i = 1; i < sc->sectorsize; i++) 437 if (dst[i] != uc) 438 break; 439 } else { 440 i = 0; 441 uc = 0; 442 } 443 if (i == sc->sectorsize) { 444 if (osp != uc) 445 error = s_write(sc->indir, secno, uc); 446 } else { 447 if (osp <= 255) { 448 sp = (uintptr_t)uma_zalloc(sc->uma, 449 M_NOWAIT); 450 if (sp == 0) { 451 error = ENOSPC; 452 break; 453 } 454 bcopy(dst, (void *)sp, sc->sectorsize); 455 error = s_write(sc->indir, secno, sp); 456 } else { 457 bcopy(dst, (void *)osp, sc->sectorsize); 458 osp = 0; 459 } 460 } 461 } else { 462 error = EOPNOTSUPP; 463 } 464 if (osp > 255) 465 uma_zfree(sc->uma, (void*)osp); 466 if (error != 0) 467 break; 468 secno++; 469 dst += sc->sectorsize; 470 } 471 bp->bio_resid = 0; 472 return (error); 473 } 474 475 static int 476 mdstart_preload(struct md_s *sc, struct bio *bp) 477 { 478 479 switch (bp->bio_cmd) { 480 case BIO_READ: 481 bcopy(sc->pl_ptr + bp->bio_offset, bp->bio_data, 482 bp->bio_length); 483 break; 484 case BIO_WRITE: 485 bcopy(bp->bio_data, sc->pl_ptr + bp->bio_offset, 486 bp->bio_length); 487 break; 488 } 489 bp->bio_resid = 0; 490 return (0); 491 } 492 493 static int 494 mdstart_vnode(struct md_s *sc, struct bio *bp) 495 { 496 int error, vfslocked; 497 struct uio auio; 498 struct iovec aiov; 499 struct mount *mp; 500 struct vnode *vp; 501 struct thread *td; 502 503 switch (bp->bio_cmd) { 504 case BIO_READ: 505 case BIO_WRITE: 506 case BIO_FLUSH: 507 break; 508 default: 509 return (EOPNOTSUPP); 510 } 511 512 td = curthread; 513 vp = sc->vnode; 514 515 /* 516 * VNODE I/O 517 * 518 * If an error occurs, we set BIO_ERROR but we do not set 519 * B_INVAL because (for a write anyway), the buffer is 520 * still valid. 521 */ 522 523 if (bp->bio_cmd == BIO_FLUSH) { 524 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 525 (void) vn_start_write(vp, &mp, V_WAIT); 526 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 527 error = VOP_FSYNC(vp, MNT_WAIT, td); 528 VOP_UNLOCK(vp, 0, td); 529 vn_finished_write(mp); 530 VFS_UNLOCK_GIANT(vfslocked); 531 return (error); 532 } 533 534 bzero(&auio, sizeof(auio)); 535 536 aiov.iov_base = bp->bio_data; 537 aiov.iov_len = bp->bio_length; 538 auio.uio_iov = &aiov; 539 auio.uio_iovcnt = 1; 540 auio.uio_offset = (vm_ooffset_t)bp->bio_offset; 541 auio.uio_segflg = UIO_SYSSPACE; 542 if (bp->bio_cmd == BIO_READ) 543 auio.uio_rw = UIO_READ; 544 else if (bp->bio_cmd == BIO_WRITE) 545 auio.uio_rw = UIO_WRITE; 546 else 547 panic("wrong BIO_OP in mdstart_vnode"); 548 auio.uio_resid = bp->bio_length; 549 auio.uio_td = td; 550 /* 551 * When reading set IO_DIRECT to try to avoid double-caching 552 * the data. When writing IO_DIRECT is not optimal. 553 */ 554 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 555 if (bp->bio_cmd == BIO_READ) { 556 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 557 error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred); 558 VOP_UNLOCK(vp, 0, td); 559 } else { 560 (void) vn_start_write(vp, &mp, V_WAIT); 561 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 562 error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC, 563 sc->cred); 564 VOP_UNLOCK(vp, 0, td); 565 vn_finished_write(mp); 566 } 567 VFS_UNLOCK_GIANT(vfslocked); 568 bp->bio_resid = auio.uio_resid; 569 return (error); 570 } 571 572 static int 573 mdstart_swap(struct md_s *sc, struct bio *bp) 574 { 575 struct sf_buf *sf; 576 int rv, offs, len, lastend; 577 vm_pindex_t i, lastp; 578 vm_page_t m; 579 u_char *p; 580 581 switch (bp->bio_cmd) { 582 case BIO_READ: 583 case BIO_WRITE: 584 case BIO_DELETE: 585 break; 586 default: 587 return (EOPNOTSUPP); 588 } 589 590 p = bp->bio_data; 591 592 /* 593 * offs is the offset at which to start operating on the 594 * next (ie, first) page. lastp is the last page on 595 * which we're going to operate. lastend is the ending 596 * position within that last page (ie, PAGE_SIZE if 597 * we're operating on complete aligned pages). 598 */ 599 offs = bp->bio_offset % PAGE_SIZE; 600 lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE; 601 lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1; 602 603 rv = VM_PAGER_OK; 604 VM_OBJECT_LOCK(sc->object); 605 vm_object_pip_add(sc->object, 1); 606 for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) { 607 len = ((i == lastp) ? lastend : PAGE_SIZE) - offs; 608 609 m = vm_page_grab(sc->object, i, 610 VM_ALLOC_NORMAL|VM_ALLOC_RETRY); 611 VM_OBJECT_UNLOCK(sc->object); 612 sched_pin(); 613 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 614 VM_OBJECT_LOCK(sc->object); 615 if (bp->bio_cmd == BIO_READ) { 616 if (m->valid != VM_PAGE_BITS_ALL) 617 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 618 if (rv == VM_PAGER_ERROR) { 619 sf_buf_free(sf); 620 sched_unpin(); 621 vm_page_lock_queues(); 622 vm_page_wakeup(m); 623 vm_page_unlock_queues(); 624 break; 625 } 626 bcopy((void *)(sf_buf_kva(sf) + offs), p, len); 627 } else if (bp->bio_cmd == BIO_WRITE) { 628 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 629 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 630 if (rv == VM_PAGER_ERROR) { 631 sf_buf_free(sf); 632 sched_unpin(); 633 vm_page_lock_queues(); 634 vm_page_wakeup(m); 635 vm_page_unlock_queues(); 636 break; 637 } 638 bcopy(p, (void *)(sf_buf_kva(sf) + offs), len); 639 m->valid = VM_PAGE_BITS_ALL; 640 #if 0 641 } else if (bp->bio_cmd == BIO_DELETE) { 642 if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL) 643 rv = vm_pager_get_pages(sc->object, &m, 1, 0); 644 if (rv == VM_PAGER_ERROR) { 645 sf_buf_free(sf); 646 sched_unpin(); 647 vm_page_lock_queues(); 648 vm_page_wakeup(m); 649 vm_page_unlock_queues(); 650 break; 651 } 652 bzero((void *)(sf_buf_kva(sf) + offs), len); 653 vm_page_dirty(m); 654 m->valid = VM_PAGE_BITS_ALL; 655 #endif 656 } 657 sf_buf_free(sf); 658 sched_unpin(); 659 vm_page_lock_queues(); 660 vm_page_wakeup(m); 661 vm_page_activate(m); 662 if (bp->bio_cmd == BIO_WRITE) 663 vm_page_dirty(m); 664 vm_page_unlock_queues(); 665 666 /* Actions on further pages start at offset 0 */ 667 p += PAGE_SIZE - offs; 668 offs = 0; 669 #if 0 670 if (bootverbose || bp->bio_offset / PAGE_SIZE < 17) 671 printf("wire_count %d busy %d flags %x hold_count %d act_count %d queue %d valid %d dirty %d @ %d\n", 672 m->wire_count, m->busy, 673 m->flags, m->hold_count, m->act_count, m->queue, m->valid, m->dirty, i); 674 #endif 675 } 676 vm_object_pip_subtract(sc->object, 1); 677 vm_object_set_writeable_dirty(sc->object); 678 VM_OBJECT_UNLOCK(sc->object); 679 return (rv != VM_PAGER_ERROR ? 0 : ENOSPC); 680 } 681 682 static void 683 md_kthread(void *arg) 684 { 685 struct md_s *sc; 686 struct bio *bp; 687 int error; 688 689 sc = arg; 690 thread_lock(curthread); 691 sched_prio(curthread, PRIBIO); 692 thread_unlock(curthread); 693 if (sc->type == MD_VNODE) 694 curthread->td_pflags |= TDP_NORUNNINGBUF; 695 696 for (;;) { 697 mtx_lock(&sc->queue_mtx); 698 if (sc->flags & MD_SHUTDOWN) { 699 sc->flags |= MD_EXITING; 700 mtx_unlock(&sc->queue_mtx); 701 kthread_exit(0); 702 } 703 bp = bioq_takefirst(&sc->bio_queue); 704 if (!bp) { 705 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 706 continue; 707 } 708 mtx_unlock(&sc->queue_mtx); 709 if (bp->bio_cmd == BIO_GETATTR) { 710 if (sc->fwsectors && sc->fwheads && 711 (g_handleattr_int(bp, "GEOM::fwsectors", 712 sc->fwsectors) || 713 g_handleattr_int(bp, "GEOM::fwheads", 714 sc->fwheads))) 715 error = -1; 716 else 717 error = EOPNOTSUPP; 718 } else { 719 error = sc->start(sc, bp); 720 } 721 722 if (error != -1) { 723 bp->bio_completed = bp->bio_length; 724 g_io_deliver(bp, error); 725 } 726 } 727 } 728 729 static struct md_s * 730 mdfind(int unit) 731 { 732 struct md_s *sc; 733 734 LIST_FOREACH(sc, &md_softc_list, list) { 735 if (sc->unit == unit) 736 break; 737 } 738 return (sc); 739 } 740 741 static struct md_s * 742 mdnew(int unit, int *errp, enum md_types type) 743 { 744 struct md_s *sc, *sc2; 745 int error, max = -1; 746 747 *errp = 0; 748 LIST_FOREACH(sc2, &md_softc_list, list) { 749 if (unit == sc2->unit) { 750 *errp = EBUSY; 751 return (NULL); 752 } 753 if (unit == -1 && sc2->unit > max) 754 max = sc2->unit; 755 } 756 if (unit == -1) 757 unit = max + 1; 758 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 759 sc->type = type; 760 bioq_init(&sc->bio_queue); 761 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 762 sc->unit = unit; 763 sprintf(sc->name, "md%d", unit); 764 LIST_INSERT_HEAD(&md_softc_list, sc, list); 765 error = kthread_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 766 if (error == 0) 767 return (sc); 768 LIST_REMOVE(sc, list); 769 mtx_destroy(&sc->queue_mtx); 770 free(sc, M_MD); 771 *errp = error; 772 return (NULL); 773 } 774 775 static void 776 mdinit(struct md_s *sc) 777 { 778 779 struct g_geom *gp; 780 struct g_provider *pp; 781 782 g_topology_lock(); 783 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 784 gp->softc = sc; 785 pp = g_new_providerf(gp, "md%d", sc->unit); 786 pp->mediasize = sc->mediasize; 787 pp->sectorsize = sc->sectorsize; 788 sc->gp = gp; 789 sc->pp = pp; 790 g_error_provider(pp, 0); 791 g_topology_unlock(); 792 } 793 794 /* 795 * XXX: we should check that the range they feed us is mapped. 796 * XXX: we should implement read-only. 797 */ 798 799 static int 800 mdcreate_preload(struct md_s *sc, struct md_ioctl *mdio) 801 { 802 803 if (mdio->md_options & ~(MD_AUTOUNIT | MD_FORCE)) 804 return (EINVAL); 805 sc->flags = mdio->md_options & MD_FORCE; 806 /* Cast to pointer size, then to pointer to avoid warning */ 807 sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base; 808 sc->pl_len = (size_t)sc->mediasize; 809 return (0); 810 } 811 812 813 static int 814 mdcreate_malloc(struct md_s *sc, struct md_ioctl *mdio) 815 { 816 uintptr_t sp; 817 int error; 818 off_t u; 819 820 error = 0; 821 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 822 return (EINVAL); 823 if (mdio->md_sectorsize != 0 && !powerof2(mdio->md_sectorsize)) 824 return (EINVAL); 825 /* Compression doesn't make sense if we have reserved space */ 826 if (mdio->md_options & MD_RESERVE) 827 mdio->md_options &= ~MD_COMPRESS; 828 if (mdio->md_fwsectors != 0) 829 sc->fwsectors = mdio->md_fwsectors; 830 if (mdio->md_fwheads != 0) 831 sc->fwheads = mdio->md_fwheads; 832 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 833 sc->indir = dimension(sc->mediasize / sc->sectorsize); 834 sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL, 835 0x1ff, 0); 836 if (mdio->md_options & MD_RESERVE) { 837 off_t nsectors; 838 839 nsectors = sc->mediasize / sc->sectorsize; 840 for (u = 0; u < nsectors; u++) { 841 sp = (uintptr_t)uma_zalloc(sc->uma, M_NOWAIT | M_ZERO); 842 if (sp != 0) 843 error = s_write(sc->indir, u, sp); 844 else 845 error = ENOMEM; 846 if (error != 0) 847 break; 848 } 849 } 850 return (error); 851 } 852 853 854 static int 855 mdsetcred(struct md_s *sc, struct ucred *cred) 856 { 857 char *tmpbuf; 858 int error = 0; 859 860 /* 861 * Set credits in our softc 862 */ 863 864 if (sc->cred) 865 crfree(sc->cred); 866 sc->cred = crhold(cred); 867 868 /* 869 * Horrible kludge to establish credentials for NFS XXX. 870 */ 871 872 if (sc->vnode) { 873 struct uio auio; 874 struct iovec aiov; 875 876 tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK); 877 bzero(&auio, sizeof(auio)); 878 879 aiov.iov_base = tmpbuf; 880 aiov.iov_len = sc->sectorsize; 881 auio.uio_iov = &aiov; 882 auio.uio_iovcnt = 1; 883 auio.uio_offset = 0; 884 auio.uio_rw = UIO_READ; 885 auio.uio_segflg = UIO_SYSSPACE; 886 auio.uio_resid = aiov.iov_len; 887 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 888 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 889 VOP_UNLOCK(sc->vnode, 0, curthread); 890 free(tmpbuf, M_TEMP); 891 } 892 return (error); 893 } 894 895 static int 896 mdcreate_vnode(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 897 { 898 struct vattr vattr; 899 struct nameidata nd; 900 int error, flags, vfslocked; 901 902 error = copyinstr(mdio->md_file, sc->file, sizeof(sc->file), NULL); 903 if (error != 0) 904 return (error); 905 flags = FREAD|FWRITE; 906 /* 907 * If the user specified that this is a read only device, unset the 908 * FWRITE mask before trying to open the backing store. 909 */ 910 if ((mdio->md_options & MD_READONLY) != 0) 911 flags &= ~FWRITE; 912 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, sc->file, td); 913 error = vn_open(&nd, &flags, 0, NULL); 914 if (error != 0) 915 return (error); 916 vfslocked = NDHASGIANT(&nd); 917 NDFREE(&nd, NDF_ONLY_PNBUF); 918 if (nd.ni_vp->v_type != VREG) { 919 error = EINVAL; 920 goto bad; 921 } 922 error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred, td); 923 if (error != 0) 924 goto bad; 925 if (VOP_ISLOCKED(nd.ni_vp, td) != LK_EXCLUSIVE) { 926 vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY, td); 927 if (nd.ni_vp->v_iflag & VI_DOOMED) { 928 /* Forced unmount. */ 929 error = EBADF; 930 goto bad; 931 } 932 } 933 nd.ni_vp->v_vflag |= VV_MD; 934 VOP_UNLOCK(nd.ni_vp, 0, td); 935 936 if (mdio->md_fwsectors != 0) 937 sc->fwsectors = mdio->md_fwsectors; 938 if (mdio->md_fwheads != 0) 939 sc->fwheads = mdio->md_fwheads; 940 sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC); 941 if (!(flags & FWRITE)) 942 sc->flags |= MD_READONLY; 943 sc->vnode = nd.ni_vp; 944 945 error = mdsetcred(sc, td->td_ucred); 946 if (error != 0) { 947 sc->vnode = NULL; 948 vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY, td); 949 nd.ni_vp->v_vflag &= ~VV_MD; 950 goto bad; 951 } 952 VFS_UNLOCK_GIANT(vfslocked); 953 return (0); 954 bad: 955 VOP_UNLOCK(nd.ni_vp, 0, td); 956 (void)vn_close(nd.ni_vp, flags, td->td_ucred, td); 957 VFS_UNLOCK_GIANT(vfslocked); 958 return (error); 959 } 960 961 static int 962 mddestroy(struct md_s *sc, struct thread *td) 963 { 964 int vfslocked; 965 966 if (sc->gp) { 967 sc->gp->softc = NULL; 968 g_topology_lock(); 969 g_wither_geom(sc->gp, ENXIO); 970 g_topology_unlock(); 971 sc->gp = NULL; 972 sc->pp = NULL; 973 } 974 mtx_lock(&sc->queue_mtx); 975 sc->flags |= MD_SHUTDOWN; 976 wakeup(sc); 977 while (!(sc->flags & MD_EXITING)) 978 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10); 979 mtx_unlock(&sc->queue_mtx); 980 mtx_destroy(&sc->queue_mtx); 981 if (sc->vnode != NULL) { 982 vfslocked = VFS_LOCK_GIANT(sc->vnode->v_mount); 983 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, td); 984 sc->vnode->v_vflag &= ~VV_MD; 985 VOP_UNLOCK(sc->vnode, 0, td); 986 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 987 FREAD : (FREAD|FWRITE), sc->cred, td); 988 VFS_UNLOCK_GIANT(vfslocked); 989 } 990 if (sc->cred != NULL) 991 crfree(sc->cred); 992 if (sc->object != NULL) 993 vm_object_deallocate(sc->object); 994 if (sc->indir) 995 destroy_indir(sc, sc->indir); 996 if (sc->uma) 997 uma_zdestroy(sc->uma); 998 999 LIST_REMOVE(sc, list); 1000 free(sc, M_MD); 1001 return (0); 1002 } 1003 1004 static int 1005 mdcreate_swap(struct md_s *sc, struct md_ioctl *mdio, struct thread *td) 1006 { 1007 vm_ooffset_t npage; 1008 int error; 1009 1010 /* 1011 * Range check. Disallow negative sizes or any size less then the 1012 * size of a page. Then round to a page. 1013 */ 1014 if (sc->mediasize == 0 || (sc->mediasize % PAGE_SIZE) != 0) 1015 return (EDOM); 1016 1017 /* 1018 * Allocate an OBJT_SWAP object. 1019 * 1020 * Note the truncation. 1021 */ 1022 1023 npage = mdio->md_mediasize / PAGE_SIZE; 1024 if (mdio->md_fwsectors != 0) 1025 sc->fwsectors = mdio->md_fwsectors; 1026 if (mdio->md_fwheads != 0) 1027 sc->fwheads = mdio->md_fwheads; 1028 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage, 1029 VM_PROT_DEFAULT, 0); 1030 if (sc->object == NULL) 1031 return (ENOMEM); 1032 sc->flags = mdio->md_options & MD_FORCE; 1033 if (mdio->md_options & MD_RESERVE) { 1034 if (swap_pager_reserve(sc->object, 0, npage) < 0) { 1035 vm_object_deallocate(sc->object); 1036 sc->object = NULL; 1037 return (EDOM); 1038 } 1039 } 1040 error = mdsetcred(sc, td->td_ucred); 1041 if (error != 0) { 1042 vm_object_deallocate(sc->object); 1043 sc->object = NULL; 1044 } 1045 return (error); 1046 } 1047 1048 1049 static int 1050 xmdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1051 { 1052 struct md_ioctl *mdio; 1053 struct md_s *sc; 1054 int error, i; 1055 1056 if (md_debug) 1057 printf("mdctlioctl(%s %lx %p %x %p)\n", 1058 devtoname(dev), cmd, addr, flags, td); 1059 1060 mdio = (struct md_ioctl *)addr; 1061 if (mdio->md_version != MDIOVERSION) 1062 return (EINVAL); 1063 1064 /* 1065 * We assert the version number in the individual ioctl 1066 * handlers instead of out here because (a) it is possible we 1067 * may add another ioctl in the future which doesn't read an 1068 * mdio, and (b) the correct return value for an unknown ioctl 1069 * is ENOIOCTL, not EINVAL. 1070 */ 1071 error = 0; 1072 switch (cmd) { 1073 case MDIOCATTACH: 1074 switch (mdio->md_type) { 1075 case MD_MALLOC: 1076 case MD_PRELOAD: 1077 case MD_VNODE: 1078 case MD_SWAP: 1079 break; 1080 default: 1081 return (EINVAL); 1082 } 1083 if (mdio->md_options & MD_AUTOUNIT) 1084 sc = mdnew(-1, &error, mdio->md_type); 1085 else 1086 sc = mdnew(mdio->md_unit, &error, mdio->md_type); 1087 if (sc == NULL) 1088 return (error); 1089 if (mdio->md_options & MD_AUTOUNIT) 1090 mdio->md_unit = sc->unit; 1091 sc->mediasize = mdio->md_mediasize; 1092 if (mdio->md_sectorsize == 0) 1093 sc->sectorsize = DEV_BSIZE; 1094 else 1095 sc->sectorsize = mdio->md_sectorsize; 1096 error = EDOOFUS; 1097 switch (sc->type) { 1098 case MD_MALLOC: 1099 sc->start = mdstart_malloc; 1100 error = mdcreate_malloc(sc, mdio); 1101 break; 1102 case MD_PRELOAD: 1103 sc->start = mdstart_preload; 1104 error = mdcreate_preload(sc, mdio); 1105 break; 1106 case MD_VNODE: 1107 sc->start = mdstart_vnode; 1108 error = mdcreate_vnode(sc, mdio, td); 1109 break; 1110 case MD_SWAP: 1111 sc->start = mdstart_swap; 1112 error = mdcreate_swap(sc, mdio, td); 1113 break; 1114 } 1115 if (error != 0) { 1116 mddestroy(sc, td); 1117 return (error); 1118 } 1119 1120 /* Prune off any residual fractional sector */ 1121 i = sc->mediasize % sc->sectorsize; 1122 sc->mediasize -= i; 1123 1124 mdinit(sc); 1125 return (0); 1126 case MDIOCDETACH: 1127 if (mdio->md_mediasize != 0 || mdio->md_options != 0) 1128 return (EINVAL); 1129 1130 sc = mdfind(mdio->md_unit); 1131 if (sc == NULL) 1132 return (ENOENT); 1133 if (sc->opencount != 0 && !(sc->flags & MD_FORCE)) 1134 return (EBUSY); 1135 return (mddestroy(sc, td)); 1136 case MDIOCQUERY: 1137 sc = mdfind(mdio->md_unit); 1138 if (sc == NULL) 1139 return (ENOENT); 1140 mdio->md_type = sc->type; 1141 mdio->md_options = sc->flags; 1142 mdio->md_mediasize = sc->mediasize; 1143 mdio->md_sectorsize = sc->sectorsize; 1144 if (sc->type == MD_VNODE) 1145 error = copyout(sc->file, mdio->md_file, 1146 strlen(sc->file) + 1); 1147 return (error); 1148 case MDIOCLIST: 1149 i = 1; 1150 LIST_FOREACH(sc, &md_softc_list, list) { 1151 if (i == MDNPAD - 1) 1152 mdio->md_pad[i] = -1; 1153 else 1154 mdio->md_pad[i++] = sc->unit; 1155 } 1156 mdio->md_pad[0] = i - 1; 1157 return (0); 1158 default: 1159 return (ENOIOCTL); 1160 }; 1161 } 1162 1163 static int 1164 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1165 { 1166 int error; 1167 1168 sx_xlock(&md_sx); 1169 error = xmdctlioctl(dev, cmd, addr, flags, td); 1170 sx_xunlock(&md_sx); 1171 return (error); 1172 } 1173 1174 static void 1175 md_preloaded(u_char *image, size_t length) 1176 { 1177 struct md_s *sc; 1178 int error; 1179 1180 sc = mdnew(-1, &error, MD_PRELOAD); 1181 if (sc == NULL) 1182 return; 1183 sc->mediasize = length; 1184 sc->sectorsize = DEV_BSIZE; 1185 sc->pl_ptr = image; 1186 sc->pl_len = length; 1187 sc->start = mdstart_preload; 1188 #ifdef MD_ROOT 1189 if (sc->unit == 0) 1190 rootdevnames[0] = "ufs:/dev/md0"; 1191 #endif 1192 mdinit(sc); 1193 } 1194 1195 static void 1196 g_md_init(struct g_class *mp __unused) 1197 { 1198 1199 caddr_t mod; 1200 caddr_t c; 1201 u_char *ptr, *name, *type; 1202 unsigned len; 1203 int i; 1204 1205 /* figure out log2(NINDIR) */ 1206 for (i = NINDIR, nshift = -1; i; nshift++) 1207 i >>= 1; 1208 1209 mod = NULL; 1210 sx_init(&md_sx, "MD config lock"); 1211 g_topology_unlock(); 1212 #ifdef MD_ROOT_SIZE 1213 sx_xlock(&md_sx); 1214 md_preloaded(mfs_root.start, sizeof(mfs_root.start)); 1215 sx_xunlock(&md_sx); 1216 #endif 1217 /* XXX: are preload_* static or do they need Giant ? */ 1218 while ((mod = preload_search_next_name(mod)) != NULL) { 1219 name = (char *)preload_search_info(mod, MODINFO_NAME); 1220 if (name == NULL) 1221 continue; 1222 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1223 if (type == NULL) 1224 continue; 1225 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1226 continue; 1227 c = preload_search_info(mod, MODINFO_ADDR); 1228 ptr = *(u_char **)c; 1229 c = preload_search_info(mod, MODINFO_SIZE); 1230 len = *(size_t *)c; 1231 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1232 MD_NAME, mdunits, name, len, ptr); 1233 sx_xlock(&md_sx); 1234 md_preloaded(ptr, len); 1235 sx_xunlock(&md_sx); 1236 } 1237 status_dev = make_dev(&mdctl_cdevsw, MAXMINOR, UID_ROOT, GID_WHEEL, 1238 0600, MDCTL_NAME); 1239 g_topology_lock(); 1240 } 1241 1242 static void 1243 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 1244 struct g_consumer *cp __unused, struct g_provider *pp) 1245 { 1246 struct md_s *mp; 1247 char *type; 1248 1249 mp = gp->softc; 1250 if (mp == NULL) 1251 return; 1252 1253 switch (mp->type) { 1254 case MD_MALLOC: 1255 type = "malloc"; 1256 break; 1257 case MD_PRELOAD: 1258 type = "preload"; 1259 break; 1260 case MD_VNODE: 1261 type = "vnode"; 1262 break; 1263 case MD_SWAP: 1264 type = "swap"; 1265 break; 1266 default: 1267 type = "unknown"; 1268 break; 1269 } 1270 1271 if (pp != NULL) { 1272 if (indent == NULL) { 1273 sbuf_printf(sb, " u %d", mp->unit); 1274 sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize); 1275 sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads); 1276 sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors); 1277 sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize); 1278 sbuf_printf(sb, " t %s", type); 1279 if (mp->type == MD_VNODE && mp->vnode != NULL) 1280 sbuf_printf(sb, " file %s", mp->file); 1281 } else { 1282 sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, 1283 mp->unit); 1284 sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n", 1285 indent, (uintmax_t) mp->sectorsize); 1286 sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n", 1287 indent, (uintmax_t) mp->fwheads); 1288 sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n", 1289 indent, (uintmax_t) mp->fwsectors); 1290 sbuf_printf(sb, "%s<length>%ju</length>\n", 1291 indent, (uintmax_t) mp->mediasize); 1292 sbuf_printf(sb, "%s<type>%s</type>\n", indent, 1293 type); 1294 if (mp->type == MD_VNODE && mp->vnode != NULL) 1295 sbuf_printf(sb, "%s<file>%s</file>\n", 1296 indent, mp->file); 1297 } 1298 } 1299 } 1300 1301 static void 1302 g_md_fini(struct g_class *mp __unused) 1303 { 1304 1305 sx_destroy(&md_sx); 1306 if (status_dev != NULL) 1307 destroy_dev(status_dev); 1308 }
Cache object: ba44c7c162c629f8c641bf365f4e4b91
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]