132}; 133 134 135static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(&md_softc_list); 136 137#define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 138#define NMASK (NINDIR-1) 139static int nshift; 140 141struct indir { 142 uintptr_t *array; 143 uint total; 144 uint used; 145 uint shift; 146}; 147 148struct md_s { 149 int unit; 150 LIST_ENTRY(md_s) list; 151 struct devstat stats; 152 struct bio_queue_head bio_queue; 153 struct mtx queue_mtx; 154 struct disk disk; 155 dev_t dev; 156 enum md_types type; 157 unsigned nsect; 158 unsigned opencount; 159 unsigned secsize; 160 unsigned flags; 161 char name[20]; 162 struct proc *procp; 163 struct g_geom *gp; 164 struct g_provider *pp; 165 166 /* MD_MALLOC related fields */ 167 struct indir *indir; 168 uma_zone_t uma; 169 170 /* MD_PRELOAD related fields */ 171 u_char *pl_ptr; 172 unsigned pl_len; 173 174 /* MD_VNODE related fields */ 175 struct vnode *vnode; 176 struct ucred *cred; 177 178 /* MD_SWAP related fields */ 179 vm_object_t object; 180}; 181 182static int mddestroy(struct md_s *sc, struct thread *td); 183 184static struct indir * 185new_indir(uint shift) 186{ 187 struct indir *ip; 188 189 ip = malloc(sizeof *ip, M_MD, M_NOWAIT | M_ZERO); 190 if (ip == NULL) 191 return (NULL); 192 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 193 M_MDSECT, M_NOWAIT | M_ZERO); 194 if (ip->array == NULL) { 195 free(ip, M_MD); 196 return (NULL); 197 } 198 ip->total = NINDIR; 199 ip->shift = shift; 200 return (ip); 201} 202 203static void 204del_indir(struct indir *ip) 205{ 206 207 free(ip->array, M_MDSECT); 208 free(ip, M_MD); 209} 210 211static void 212destroy_indir(struct md_s *sc, struct indir *ip) 213{ 214 int i; 215 216 for (i = 0; i < NINDIR; i++) { 217 if (!ip->array[i]) 218 continue; 219 if (ip->shift) 220 destroy_indir(sc, (struct indir*)(ip->array[i])); 221 else if (ip->array[i] > 255) 222 uma_zfree(sc->uma, (void *)(ip->array[i])); 223 } 224 del_indir(ip); 225} 226 227/* 228 * This function does the math and alloctes the top level "indir" structure 229 * for a device of "size" sectors. 230 */ 231 232static struct indir * 233dimension(off_t size) 234{ 235 off_t rcnt; 236 struct indir *ip; 237 int i, layer; 238 239 rcnt = size; 240 layer = 0; 241 while (rcnt > NINDIR) { 242 rcnt /= NINDIR; 243 layer++; 244 } 245 /* figure out log2(NINDIR) */ 246 for (i = NINDIR, nshift = -1; i; nshift++) 247 i >>= 1; 248 249 /* 250 * XXX: the top layer is probably not fully populated, so we allocate 251 * too much space for ip->array in new_indir() here. 252 */ 253 ip = new_indir(layer * nshift); 254 return (ip); 255} 256 257/* 258 * Read a given sector 259 */ 260 261static uintptr_t 262s_read(struct indir *ip, off_t offset) 263{ 264 struct indir *cip; 265 int idx; 266 uintptr_t up; 267 268 if (md_debug > 1) 269 printf("s_read(%jd)\n", (intmax_t)offset); 270 up = 0; 271 for (cip = ip; cip != NULL;) { 272 if (cip->shift) { 273 idx = (offset >> cip->shift) & NMASK; 274 up = cip->array[idx]; 275 cip = (struct indir *)up; 276 continue; 277 } 278 idx = offset & NMASK; 279 return (cip->array[idx]); 280 } 281 return (0); 282} 283 284/* 285 * Write a given sector, prune the tree if the value is 0 286 */ 287 288static int 289s_write(struct indir *ip, off_t offset, uintptr_t ptr) 290{ 291 struct indir *cip, *lip[10]; 292 int idx, li; 293 uintptr_t up; 294 295 if (md_debug > 1) 296 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 297 up = 0; 298 li = 0; 299 cip = ip; 300 for (;;) { 301 lip[li++] = cip; 302 if (cip->shift) { 303 idx = (offset >> cip->shift) & NMASK; 304 up = cip->array[idx]; 305 if (up != 0) { 306 cip = (struct indir *)up; 307 continue; 308 } 309 /* Allocate branch */ 310 cip->array[idx] = 311 (uintptr_t)new_indir(cip->shift - nshift); 312 if (cip->array[idx] == 0) 313 return (ENOSPC); 314 cip->used++; 315 up = cip->array[idx]; 316 cip = (struct indir *)up; 317 continue; 318 } 319 /* leafnode */ 320 idx = offset & NMASK; 321 up = cip->array[idx]; 322 if (up != 0) 323 cip->used--; 324 cip->array[idx] = ptr; 325 if (ptr != 0) 326 cip->used++; 327 break; 328 } 329 if (cip->used != 0 || li == 1) 330 return (0); 331 li--; 332 while (cip->used == 0 && cip != ip) { 333 li--; 334 idx = (offset >> lip[li]->shift) & NMASK; 335 up = lip[li]->array[idx]; 336 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 337 del_indir(cip); 338 lip[li]->array[idx] = 0; 339 lip[li]->used--; 340 cip = lip[li]; 341 } 342 return (0); 343} 344 345 346struct g_class g_md_class = { 347 "MD", 348 NULL, 349 NULL, 350 G_CLASS_INITIALIZER 351 352}; 353 354static int 355g_md_access(struct g_provider *pp, int r, int w, int e) 356{ 357 struct md_s *sc; 358 359 sc = pp->geom->softc; 360 r += pp->acr; 361 w += pp->acw; 362 e += pp->ace; 363 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 364 sc->opencount = 1; 365 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 366 sc->opencount = 0; 367 } 368 return (0); 369} 370 371static void 372g_md_start(struct bio *bp) 373{ 374 struct md_s *sc; 375 376 sc = bp->bio_to->geom->softc; 377 378 switch(bp->bio_cmd) { 379 case BIO_GETATTR: 380 case BIO_SETATTR: 381 g_io_deliver(bp, EOPNOTSUPP); 382 return; 383 } 384 bp->bio_blkno = bp->bio_offset >> DEV_BSHIFT; 385 bp->bio_pblkno = bp->bio_offset / sc->secsize; 386 bp->bio_bcount = bp->bio_length; 387 mtx_lock(&sc->queue_mtx); 388 bioqdisksort(&sc->bio_queue, bp); 389 mtx_unlock(&sc->queue_mtx); 390 391 wakeup(sc); 392} 393 394DECLARE_GEOM_CLASS(g_md_class, g_md); 395 396 397static int 398mdstart_malloc(struct md_s *sc, struct bio *bp) 399{ 400 int i, error; 401 u_char *dst; 402 unsigned secno, nsec, uc; 403 uintptr_t sp, osp; 404 405 nsec = bp->bio_bcount / sc->secsize; 406 secno = bp->bio_pblkno; 407 dst = bp->bio_data; 408 error = 0; 409 while (nsec--) { 410 osp = s_read(sc->indir, secno); 411 if (bp->bio_cmd == BIO_DELETE) { 412 if (osp != 0) 413 error = s_write(sc->indir, secno, 0); 414 } else if (bp->bio_cmd == BIO_READ) { 415 if (osp == 0) 416 bzero(dst, sc->secsize); 417 else if (osp <= 255) 418 for (i = 0; i < sc->secsize; i++) 419 dst[i] = osp; 420 else 421 bcopy((void *)osp, dst, sc->secsize); 422 osp = 0; 423 } else if (bp->bio_cmd == BIO_WRITE) { 424 if (sc->flags & MD_COMPRESS) { 425 uc = dst[0]; 426 for (i = 1; i < sc->secsize; i++) 427 if (dst[i] != uc) 428 break; 429 } else { 430 i = 0; 431 uc = 0; 432 } 433 if (i == sc->secsize) { 434 if (osp != uc) 435 error = s_write(sc->indir, secno, uc); 436 } else { 437 if (osp <= 255) { 438 sp = (uintptr_t) uma_zalloc( 439 sc->uma, M_NOWAIT); 440 if (sp == 0) { 441 error = ENOSPC; 442 break; 443 } 444 bcopy(dst, (void *)sp, sc->secsize); 445 error = s_write(sc->indir, secno, sp); 446 } else { 447 bcopy(dst, (void *)osp, sc->secsize); 448 osp = 0; 449 } 450 } 451 } else { 452 error = EOPNOTSUPP; 453 } 454 if (osp > 255) 455 uma_zfree(sc->uma, (void*)osp); 456 if (error) 457 break; 458 secno++; 459 dst += sc->secsize; 460 } 461 bp->bio_resid = 0; 462 return (error); 463} 464 465static int 466mdstart_preload(struct md_s *sc, struct bio *bp) 467{ 468 469 if (bp->bio_cmd == BIO_DELETE) { 470 } else if (bp->bio_cmd == BIO_READ) { 471 bcopy(sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_data, bp->bio_bcount); 472 } else { 473 bcopy(bp->bio_data, sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_bcount); 474 } 475 bp->bio_resid = 0; 476 return (0); 477} 478 479static int 480mdstart_vnode(struct md_s *sc, struct bio *bp) 481{ 482 int error; 483 struct uio auio; 484 struct iovec aiov; 485 struct mount *mp; 486 487 /* 488 * VNODE I/O 489 * 490 * If an error occurs, we set BIO_ERROR but we do not set 491 * B_INVAL because (for a write anyway), the buffer is 492 * still valid. 493 */ 494 495 bzero(&auio, sizeof(auio)); 496 497 aiov.iov_base = bp->bio_data; 498 aiov.iov_len = bp->bio_bcount; 499 auio.uio_iov = &aiov; 500 auio.uio_iovcnt = 1; 501 auio.uio_offset = (vm_ooffset_t)bp->bio_pblkno * sc->secsize; 502 auio.uio_segflg = UIO_SYSSPACE; 503 if(bp->bio_cmd == BIO_READ) 504 auio.uio_rw = UIO_READ; 505 else 506 auio.uio_rw = UIO_WRITE; 507 auio.uio_resid = bp->bio_bcount; 508 auio.uio_td = curthread; 509 /* 510 * When reading set IO_DIRECT to try to avoid double-caching 511 * the data. When writing IO_DIRECT is not optimal, but we 512 * must set IO_NOWDRAIN to avoid a wdrain deadlock. 513 */ 514 if (bp->bio_cmd == BIO_READ) { 515 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 516 error = VOP_READ(sc->vnode, &auio, IO_DIRECT, sc->cred); 517 } else { 518 (void) vn_start_write(sc->vnode, &mp, V_WAIT); 519 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 520 error = VOP_WRITE(sc->vnode, &auio, IO_NOWDRAIN, sc->cred); 521 vn_finished_write(mp); 522 } 523 VOP_UNLOCK(sc->vnode, 0, curthread); 524 bp->bio_resid = auio.uio_resid; 525 return (error); 526} 527 528static void 529mddone_swap(struct bio *bp) 530{ 531 532 bp->bio_completed = bp->bio_length - bp->bio_resid; 533 g_std_done(bp); 534} 535 536static int 537mdstart_swap(struct md_s *sc, struct bio *bp) 538{ 539 { 540 struct bio *bp2; 541 542 bp2 = g_clone_bio(bp); 543 bp2->bio_done = mddone_swap; 544 bp2->bio_blkno = bp2->bio_offset >> DEV_BSHIFT; 545 bp2->bio_pblkno = bp2->bio_offset / sc->secsize; 546 bp2->bio_bcount = bp2->bio_length; 547 bp = bp2; 548 } 549 550 bp->bio_resid = 0; 551 if ((bp->bio_cmd == BIO_DELETE) && (sc->flags & MD_RESERVE)) 552 biodone(bp); 553 else 554 vm_pager_strategy(sc->object, bp); 555 return (-1); 556} 557 558static void 559md_kthread(void *arg) 560{ 561 struct md_s *sc; 562 struct bio *bp; 563 int error, hasgiant; 564 565 sc = arg; 566 curthread->td_base_pri = PRIBIO; 567 568 switch (sc->type) { 569 case MD_SWAP: 570 case MD_VNODE: 571 mtx_lock(&Giant); 572 hasgiant = 1; 573 break; 574 case MD_MALLOC: 575 case MD_PRELOAD: 576 default: 577 hasgiant = 0; 578 break; 579 } 580 581 for (;;) { 582 mtx_lock(&sc->queue_mtx); 583 bp = bioq_first(&sc->bio_queue); 584 if (bp) 585 bioq_remove(&sc->bio_queue, bp); 586 if (!bp) { 587 if (sc->flags & MD_SHUTDOWN) { 588 mtx_unlock(&sc->queue_mtx); 589 sc->procp = NULL; 590 wakeup(&sc->procp); 591 if (!hasgiant) 592 mtx_lock(&Giant); 593 kthread_exit(0); 594 } 595 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 596 continue; 597 } 598 mtx_unlock(&sc->queue_mtx); 599 600 switch (sc->type) { 601 case MD_MALLOC: 602 devstat_start_transaction(&sc->stats); 603 error = mdstart_malloc(sc, bp); 604 break; 605 case MD_PRELOAD: 606 devstat_start_transaction(&sc->stats); 607 error = mdstart_preload(sc, bp); 608 break; 609 case MD_VNODE: 610 devstat_start_transaction(&sc->stats); 611 error = mdstart_vnode(sc, bp); 612 break; 613 case MD_SWAP: 614 error = mdstart_swap(sc, bp); 615 break; 616 default: 617 panic("Impossible md(type)"); 618 break; 619 } 620 621 if (error != -1) { 622 bp->bio_completed = bp->bio_length; 623 g_io_deliver(bp, error); 624 } 625 } 626} 627 628static struct md_s * 629mdfind(int unit) 630{ 631 struct md_s *sc; 632 633 /* XXX: LOCK(unique unit numbers) */ 634 LIST_FOREACH(sc, &md_softc_list, list) { 635 if (sc->unit == unit) 636 break; 637 } 638 /* XXX: UNLOCK(unique unit numbers) */ 639 return (sc); 640} 641 642static struct md_s * 643mdnew(int unit) 644{ 645 struct md_s *sc; 646 int error, max = -1; 647 648 /* XXX: LOCK(unique unit numbers) */ 649 LIST_FOREACH(sc, &md_softc_list, list) { 650 if (sc->unit == unit) { 651 /* XXX: UNLOCK(unique unit numbers) */ 652 return (NULL); 653 } 654 if (sc->unit > max) 655 max = sc->unit; 656 } 657 if (unit == -1) 658 unit = max + 1; 659 if (unit > 255) 660 return (NULL); 661 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 662 sc->unit = unit; 663 bioq_init(&sc->bio_queue); 664 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 665 sprintf(sc->name, "md%d", unit); 666 error = kthread_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 667 if (error) { 668 free(sc, M_MD); 669 return (NULL); 670 } 671 LIST_INSERT_HEAD(&md_softc_list, sc, list); 672 /* XXX: UNLOCK(unique unit numbers) */ 673 return (sc); 674} 675 676static void 677mdinit(struct md_s *sc) 678{ 679 680 devstat_add_entry(&sc->stats, MD_NAME, sc->unit, sc->secsize, 681 DEVSTAT_NO_ORDERED_TAGS, 682 DEVSTAT_TYPE_DIRECT | DEVSTAT_TYPE_IF_OTHER, 683 DEVSTAT_PRIORITY_OTHER); 684 { 685 struct g_geom *gp; 686 struct g_provider *pp; 687 688 DROP_GIANT(); 689 g_topology_lock(); 690 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 691 gp->start = g_md_start; 692 gp->access = g_md_access; 693 gp->softc = sc; 694 pp = g_new_providerf(gp, "md%d", sc->unit); 695 pp->mediasize = (off_t)sc->nsect * sc->secsize; 696 pp->sectorsize = sc->secsize; 697 sc->gp = gp; 698 sc->pp = pp; 699 g_error_provider(pp, 0); 700 g_topology_unlock(); 701 PICKUP_GIANT(); 702 } 703} 704 705/* 706 * XXX: we should check that the range they feed us is mapped. 707 * XXX: we should implement read-only. 708 */ 709 710static int 711mdcreate_preload(struct md_ioctl *mdio) 712{ 713 struct md_s *sc; 714 715 if (mdio->md_size == 0) 716 return (EINVAL); 717 if (mdio->md_options & ~(MD_AUTOUNIT)) 718 return (EINVAL); 719 if (mdio->md_options & MD_AUTOUNIT) { 720 sc = mdnew(-1); 721 if (sc == NULL) 722 return (ENOMEM); 723 mdio->md_unit = sc->unit; 724 } else { 725 sc = mdnew(mdio->md_unit); 726 if (sc == NULL) 727 return (EBUSY); 728 } 729 sc->type = MD_PRELOAD; 730 sc->secsize = DEV_BSIZE; 731 sc->nsect = mdio->md_size; 732 sc->flags = mdio->md_options & MD_FORCE; 733 /* Cast to pointer size, then to pointer to avoid warning */ 734 sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base; 735 sc->pl_len = (mdio->md_size << DEV_BSHIFT); 736 mdinit(sc); 737 return (0); 738} 739 740 741static int 742mdcreate_malloc(struct md_ioctl *mdio) 743{ 744 struct md_s *sc; 745 off_t u; 746 uintptr_t sp; 747 int error; 748 749 error = 0; 750 if (mdio->md_size == 0) 751 return (EINVAL); 752 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 753 return (EINVAL); 754 /* Compression doesn't make sense if we have reserved space */ 755 if (mdio->md_options & MD_RESERVE) 756 mdio->md_options &= ~MD_COMPRESS; 757 if (mdio->md_options & MD_AUTOUNIT) { 758 sc = mdnew(-1); 759 if (sc == NULL) 760 return (ENOMEM); 761 mdio->md_unit = sc->unit; 762 } else { 763 sc = mdnew(mdio->md_unit); 764 if (sc == NULL) 765 return (EBUSY); 766 } 767 sc->type = MD_MALLOC; 768 sc->secsize = DEV_BSIZE; 769 sc->nsect = mdio->md_size; 770 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 771 sc->indir = dimension(sc->nsect); 772 sc->uma = uma_zcreate(sc->name, sc->secsize, 773 NULL, NULL, NULL, NULL, 0x1ff, 0); 774 if (mdio->md_options & MD_RESERVE) { 775 for (u = 0; u < sc->nsect; u++) { 776 sp = (uintptr_t) uma_zalloc(sc->uma, M_NOWAIT | M_ZERO); 777 if (sp != 0) 778 error = s_write(sc->indir, u, sp); 779 else 780 error = ENOMEM; 781 if (error) 782 break; 783 } 784 } 785 if (error) { 786 mddestroy(sc, NULL); 787 return (error); 788 } 789 mdinit(sc); 790 if (!(mdio->md_options & MD_RESERVE)) 791 sc->pp->flags |= G_PF_CANDELETE; 792 return (0); 793} 794 795 796static int 797mdsetcred(struct md_s *sc, struct ucred *cred) 798{ 799 char *tmpbuf; 800 int error = 0; 801 802 /* 803 * Set credits in our softc 804 */ 805 806 if (sc->cred) 807 crfree(sc->cred); 808 sc->cred = crhold(cred); 809 810 /* 811 * Horrible kludge to establish credentials for NFS XXX. 812 */ 813 814 if (sc->vnode) { 815 struct uio auio; 816 struct iovec aiov; 817 818 tmpbuf = malloc(sc->secsize, M_TEMP, M_WAITOK); 819 bzero(&auio, sizeof(auio)); 820 821 aiov.iov_base = tmpbuf; 822 aiov.iov_len = sc->secsize; 823 auio.uio_iov = &aiov; 824 auio.uio_iovcnt = 1; 825 auio.uio_offset = 0; 826 auio.uio_rw = UIO_READ; 827 auio.uio_segflg = UIO_SYSSPACE; 828 auio.uio_resid = aiov.iov_len; 829 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 830 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 831 VOP_UNLOCK(sc->vnode, 0, curthread); 832 free(tmpbuf, M_TEMP); 833 } 834 return (error); 835} 836 837static int 838mdcreate_vnode(struct md_ioctl *mdio, struct thread *td) 839{ 840 struct md_s *sc; 841 struct vattr vattr; 842 struct nameidata nd; 843 int error, flags; 844 845 flags = FREAD|FWRITE; 846 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 847 error = vn_open(&nd, &flags, 0); 848 if (error) { 849 if (error != EACCES && error != EPERM && error != EROFS) 850 return (error); 851 flags &= ~FWRITE; 852 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 853 error = vn_open(&nd, &flags, 0); 854 if (error) 855 return (error); 856 } 857 NDFREE(&nd, NDF_ONLY_PNBUF); 858 if (nd.ni_vp->v_type != VREG || 859 (error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred, td))) { 860 VOP_UNLOCK(nd.ni_vp, 0, td); 861 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 862 return (error ? error : EINVAL); 863 } 864 VOP_UNLOCK(nd.ni_vp, 0, td); 865 866 if (mdio->md_options & MD_AUTOUNIT) { 867 sc = mdnew(-1); 868 mdio->md_unit = sc->unit; 869 } else { 870 sc = mdnew(mdio->md_unit); 871 } 872 if (sc == NULL) { 873 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 874 return (EBUSY); 875 } 876 877 sc->type = MD_VNODE; 878 sc->flags = mdio->md_options & MD_FORCE; 879 if (!(flags & FWRITE)) 880 sc->flags |= MD_READONLY; 881 sc->secsize = DEV_BSIZE; 882 sc->vnode = nd.ni_vp; 883 884 /* 885 * If the size is specified, override the file attributes. 886 */ 887 if (mdio->md_size) 888 sc->nsect = mdio->md_size; 889 else 890 sc->nsect = vattr.va_size / sc->secsize; /* XXX: round up ? */ 891 if (sc->nsect == 0) { 892 mddestroy(sc, td); 893 return (EINVAL); 894 } 895 error = mdsetcred(sc, td->td_ucred); 896 if (error) { 897 mddestroy(sc, td); 898 return (error); 899 } 900 mdinit(sc); 901 return (0); 902} 903 904static int 905mddestroy(struct md_s *sc, struct thread *td) 906{ 907 908 GIANT_REQUIRED; 909 910 mtx_destroy(&sc->queue_mtx); 911 devstat_remove_entry(&sc->stats); 912 { 913 if (sc->gp) { 914 sc->gp->flags |= G_GEOM_WITHER; 915 sc->gp->softc = NULL; 916 } 917 if (sc->pp) 918 g_orphan_provider(sc->pp, ENXIO); 919 } 920 sc->flags |= MD_SHUTDOWN; 921 wakeup(sc); 922 while (sc->procp != NULL) 923 tsleep(&sc->procp, PRIBIO, "mddestroy", hz / 10); 924 if (sc->vnode != NULL) 925 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 926 FREAD : (FREAD|FWRITE), sc->cred, td); 927 if (sc->cred != NULL) 928 crfree(sc->cred); 929 if (sc->object != NULL) { 930 vm_pager_deallocate(sc->object); 931 } 932 if (sc->indir) 933 destroy_indir(sc, sc->indir); 934 if (sc->uma) 935 uma_zdestroy(sc->uma); 936 937 /* XXX: LOCK(unique unit numbers) */ 938 LIST_REMOVE(sc, list); 939 /* XXX: UNLOCK(unique unit numbers) */ 940 free(sc, M_MD); 941 return (0); 942} 943 944static int 945mdcreate_swap(struct md_ioctl *mdio, struct thread *td) 946{ 947 int error; 948 struct md_s *sc; 949 950 GIANT_REQUIRED; 951 952 if (mdio->md_options & MD_AUTOUNIT) { 953 sc = mdnew(-1); 954 mdio->md_unit = sc->unit; 955 } else { 956 sc = mdnew(mdio->md_unit); 957 } 958 if (sc == NULL) 959 return (EBUSY); 960 961 sc->type = MD_SWAP; 962 963 /* 964 * Range check. Disallow negative sizes or any size less then the 965 * size of a page. Then round to a page. 966 */ 967 968 if (mdio->md_size == 0) { 969 mddestroy(sc, td); 970 return (EDOM); 971 } 972 973 /* 974 * Allocate an OBJT_SWAP object. 975 * 976 * sc_secsize is PAGE_SIZE'd 977 * 978 * mdio->size is in DEV_BSIZE'd chunks. 979 * Note the truncation. 980 */ 981 982 sc->secsize = PAGE_SIZE; 983 sc->nsect = mdio->md_size / (PAGE_SIZE / DEV_BSIZE); 984 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, sc->secsize * (vm_offset_t)sc->nsect, VM_PROT_DEFAULT, 0); 985 sc->flags = mdio->md_options & MD_FORCE; 986 if (mdio->md_options & MD_RESERVE) { 987 if (swap_pager_reserve(sc->object, 0, sc->nsect) < 0) { 988 vm_pager_deallocate(sc->object); 989 sc->object = NULL; 990 mddestroy(sc, td); 991 return (EDOM); 992 } 993 } 994 error = mdsetcred(sc, td->td_ucred); 995 if (error) { 996 mddestroy(sc, td); 997 return (error); 998 } 999 mdinit(sc); 1000 if (!(mdio->md_options & MD_RESERVE)) 1001 sc->pp->flags |= G_PF_CANDELETE; 1002 return (0); 1003} 1004 1005static int 1006mddetach(int unit, struct thread *td) 1007{ 1008 struct md_s *sc; 1009 1010 sc = mdfind(unit); 1011 if (sc == NULL) 1012 return (ENOENT); 1013 if (sc->opencount != 0 && !(sc->flags & MD_FORCE)) 1014 return (EBUSY); 1015 switch(sc->type) { 1016 case MD_VNODE: 1017 case MD_SWAP: 1018 case MD_MALLOC: 1019 case MD_PRELOAD: 1020 return (mddestroy(sc, td)); 1021 default: 1022 return (EOPNOTSUPP); 1023 } 1024} 1025 1026static int 1027mdctlioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1028{ 1029 struct md_ioctl *mdio; 1030 struct md_s *sc; 1031 int i; 1032 1033 if (md_debug) 1034 printf("mdctlioctl(%s %lx %p %x %p)\n", 1035 devtoname(dev), cmd, addr, flags, td); 1036 1037 /* 1038 * We assert the version number in the individual ioctl 1039 * handlers instead of out here because (a) it is possible we 1040 * may add another ioctl in the future which doesn't read an 1041 * mdio, and (b) the correct return value for an unknown ioctl 1042 * is ENOIOCTL, not EINVAL. 1043 */ 1044 mdio = (struct md_ioctl *)addr; 1045 switch (cmd) { 1046 case MDIOCATTACH: 1047 if (mdio->md_version != MDIOVERSION) 1048 return (EINVAL); 1049 switch (mdio->md_type) { 1050 case MD_MALLOC: 1051 return (mdcreate_malloc(mdio)); 1052 case MD_PRELOAD: 1053 return (mdcreate_preload(mdio)); 1054 case MD_VNODE: 1055 return (mdcreate_vnode(mdio, td)); 1056 case MD_SWAP: 1057 return (mdcreate_swap(mdio, td)); 1058 default: 1059 return (EINVAL); 1060 } 1061 case MDIOCDETACH: 1062 if (mdio->md_version != MDIOVERSION) 1063 return (EINVAL); 1064 if (mdio->md_file != NULL || mdio->md_size != 0 || 1065 mdio->md_options != 0) 1066 return (EINVAL); 1067 return (mddetach(mdio->md_unit, td)); 1068 case MDIOCQUERY: 1069 if (mdio->md_version != MDIOVERSION) 1070 return (EINVAL); 1071 sc = mdfind(mdio->md_unit); 1072 if (sc == NULL) 1073 return (ENOENT); 1074 mdio->md_type = sc->type; 1075 mdio->md_options = sc->flags; 1076 switch (sc->type) { 1077 case MD_MALLOC: 1078 mdio->md_size = sc->nsect; 1079 break; 1080 case MD_PRELOAD: 1081 mdio->md_size = sc->nsect; 1082 mdio->md_base = (uint64_t)(intptr_t)sc->pl_ptr; 1083 break; 1084 case MD_SWAP: 1085 mdio->md_size = sc->nsect * (PAGE_SIZE / DEV_BSIZE); 1086 break; 1087 case MD_VNODE: 1088 mdio->md_size = sc->nsect; 1089 /* XXX fill this in */ 1090 mdio->md_file = NULL; 1091 break; 1092 } 1093 return (0); 1094 case MDIOCLIST: 1095 i = 1; 1096 LIST_FOREACH(sc, &md_softc_list, list) { 1097 if (i == MDNPAD - 1) 1098 mdio->md_pad[i] = -1; 1099 else 1100 mdio->md_pad[i++] = sc->unit; 1101 } 1102 mdio->md_pad[0] = i - 1; 1103 return (0); 1104 default: 1105 return (ENOIOCTL); 1106 }; 1107 return (ENOIOCTL); 1108} 1109 1110static void 1111md_preloaded(u_char *image, unsigned length) 1112{ 1113 struct md_s *sc; 1114 1115 sc = mdnew(-1); 1116 if (sc == NULL) 1117 return; 1118 sc->type = MD_PRELOAD; 1119 sc->secsize = DEV_BSIZE; 1120 sc->nsect = length / DEV_BSIZE; 1121 sc->pl_ptr = image; 1122 sc->pl_len = length; 1123 if (sc->unit == 0) 1124 mdrootready = 1; 1125 mdinit(sc); 1126} 1127 1128static void 1129md_drvinit(void *unused) 1130{ 1131 1132 caddr_t mod; 1133 caddr_t c; 1134 u_char *ptr, *name, *type; 1135 unsigned len; 1136 1137#ifdef MD_ROOT_SIZE 1138 md_preloaded(mfs_root, MD_ROOT_SIZE*1024); 1139#endif 1140 mod = NULL; 1141 while ((mod = preload_search_next_name(mod)) != NULL) { 1142 name = (char *)preload_search_info(mod, MODINFO_NAME); 1143 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1144 if (name == NULL) 1145 continue; 1146 if (type == NULL) 1147 continue; 1148 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1149 continue; 1150 c = preload_search_info(mod, MODINFO_ADDR); 1151 ptr = *(u_char **)c; 1152 c = preload_search_info(mod, MODINFO_SIZE); 1153 len = *(size_t *)c; 1154 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1155 MD_NAME, mdunits, name, len, ptr); 1156 md_preloaded(ptr, len); 1157 } 1158 status_dev = make_dev(&mdctl_cdevsw, 0xffff00ff, UID_ROOT, GID_WHEEL, 1159 0600, MDCTL_NAME); 1160} 1161 1162static int 1163md_modevent(module_t mod, int type, void *data) 1164{ 1165 int error; 1166 struct md_s *sc; 1167 1168 switch (type) { 1169 case MOD_LOAD: 1170 md_drvinit(NULL); 1171 break; 1172 case MOD_UNLOAD: 1173 LIST_FOREACH(sc, &md_softc_list, list) { 1174 error = mddetach(sc->unit, curthread); 1175 if (error != 0) 1176 return (error); 1177 } 1178 if (status_dev) 1179 destroy_dev(status_dev); 1180 status_dev = 0; 1181 break; 1182 default: 1183 break; 1184 } 1185 return (0); 1186} 1187 1188static moduledata_t md_mod = { 1189 MD_NAME, 1190 md_modevent, 1191 NULL 1192}; 1193DECLARE_MODULE(md, md_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE+CDEV_MAJOR); 1194MODULE_VERSION(md, MD_MODVER); 1195 1196 1197#ifdef MD_ROOT 1198static void 1199md_takeroot(void *junk) 1200{ 1201 if (mdrootready) 1202 rootdevnames[0] = "ufs:/dev/md0"; 1203} 1204 1205SYSINIT(md_root, SI_SUB_MOUNT_ROOT, SI_ORDER_FIRST, md_takeroot, NULL); 1206#endif /* MD_ROOT */
| 127}; 128 129 130static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(&md_softc_list); 131 132#define NINDIR (PAGE_SIZE / sizeof(uintptr_t)) 133#define NMASK (NINDIR-1) 134static int nshift; 135 136struct indir { 137 uintptr_t *array; 138 uint total; 139 uint used; 140 uint shift; 141}; 142 143struct md_s { 144 int unit; 145 LIST_ENTRY(md_s) list; 146 struct devstat stats; 147 struct bio_queue_head bio_queue; 148 struct mtx queue_mtx; 149 struct disk disk; 150 dev_t dev; 151 enum md_types type; 152 unsigned nsect; 153 unsigned opencount; 154 unsigned secsize; 155 unsigned flags; 156 char name[20]; 157 struct proc *procp; 158 struct g_geom *gp; 159 struct g_provider *pp; 160 161 /* MD_MALLOC related fields */ 162 struct indir *indir; 163 uma_zone_t uma; 164 165 /* MD_PRELOAD related fields */ 166 u_char *pl_ptr; 167 unsigned pl_len; 168 169 /* MD_VNODE related fields */ 170 struct vnode *vnode; 171 struct ucred *cred; 172 173 /* MD_SWAP related fields */ 174 vm_object_t object; 175}; 176 177static int mddestroy(struct md_s *sc, struct thread *td); 178 179static struct indir * 180new_indir(uint shift) 181{ 182 struct indir *ip; 183 184 ip = malloc(sizeof *ip, M_MD, M_NOWAIT | M_ZERO); 185 if (ip == NULL) 186 return (NULL); 187 ip->array = malloc(sizeof(uintptr_t) * NINDIR, 188 M_MDSECT, M_NOWAIT | M_ZERO); 189 if (ip->array == NULL) { 190 free(ip, M_MD); 191 return (NULL); 192 } 193 ip->total = NINDIR; 194 ip->shift = shift; 195 return (ip); 196} 197 198static void 199del_indir(struct indir *ip) 200{ 201 202 free(ip->array, M_MDSECT); 203 free(ip, M_MD); 204} 205 206static void 207destroy_indir(struct md_s *sc, struct indir *ip) 208{ 209 int i; 210 211 for (i = 0; i < NINDIR; i++) { 212 if (!ip->array[i]) 213 continue; 214 if (ip->shift) 215 destroy_indir(sc, (struct indir*)(ip->array[i])); 216 else if (ip->array[i] > 255) 217 uma_zfree(sc->uma, (void *)(ip->array[i])); 218 } 219 del_indir(ip); 220} 221 222/* 223 * This function does the math and alloctes the top level "indir" structure 224 * for a device of "size" sectors. 225 */ 226 227static struct indir * 228dimension(off_t size) 229{ 230 off_t rcnt; 231 struct indir *ip; 232 int i, layer; 233 234 rcnt = size; 235 layer = 0; 236 while (rcnt > NINDIR) { 237 rcnt /= NINDIR; 238 layer++; 239 } 240 /* figure out log2(NINDIR) */ 241 for (i = NINDIR, nshift = -1; i; nshift++) 242 i >>= 1; 243 244 /* 245 * XXX: the top layer is probably not fully populated, so we allocate 246 * too much space for ip->array in new_indir() here. 247 */ 248 ip = new_indir(layer * nshift); 249 return (ip); 250} 251 252/* 253 * Read a given sector 254 */ 255 256static uintptr_t 257s_read(struct indir *ip, off_t offset) 258{ 259 struct indir *cip; 260 int idx; 261 uintptr_t up; 262 263 if (md_debug > 1) 264 printf("s_read(%jd)\n", (intmax_t)offset); 265 up = 0; 266 for (cip = ip; cip != NULL;) { 267 if (cip->shift) { 268 idx = (offset >> cip->shift) & NMASK; 269 up = cip->array[idx]; 270 cip = (struct indir *)up; 271 continue; 272 } 273 idx = offset & NMASK; 274 return (cip->array[idx]); 275 } 276 return (0); 277} 278 279/* 280 * Write a given sector, prune the tree if the value is 0 281 */ 282 283static int 284s_write(struct indir *ip, off_t offset, uintptr_t ptr) 285{ 286 struct indir *cip, *lip[10]; 287 int idx, li; 288 uintptr_t up; 289 290 if (md_debug > 1) 291 printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr); 292 up = 0; 293 li = 0; 294 cip = ip; 295 for (;;) { 296 lip[li++] = cip; 297 if (cip->shift) { 298 idx = (offset >> cip->shift) & NMASK; 299 up = cip->array[idx]; 300 if (up != 0) { 301 cip = (struct indir *)up; 302 continue; 303 } 304 /* Allocate branch */ 305 cip->array[idx] = 306 (uintptr_t)new_indir(cip->shift - nshift); 307 if (cip->array[idx] == 0) 308 return (ENOSPC); 309 cip->used++; 310 up = cip->array[idx]; 311 cip = (struct indir *)up; 312 continue; 313 } 314 /* leafnode */ 315 idx = offset & NMASK; 316 up = cip->array[idx]; 317 if (up != 0) 318 cip->used--; 319 cip->array[idx] = ptr; 320 if (ptr != 0) 321 cip->used++; 322 break; 323 } 324 if (cip->used != 0 || li == 1) 325 return (0); 326 li--; 327 while (cip->used == 0 && cip != ip) { 328 li--; 329 idx = (offset >> lip[li]->shift) & NMASK; 330 up = lip[li]->array[idx]; 331 KASSERT(up == (uintptr_t)cip, ("md screwed up")); 332 del_indir(cip); 333 lip[li]->array[idx] = 0; 334 lip[li]->used--; 335 cip = lip[li]; 336 } 337 return (0); 338} 339 340 341struct g_class g_md_class = { 342 "MD", 343 NULL, 344 NULL, 345 G_CLASS_INITIALIZER 346 347}; 348 349static int 350g_md_access(struct g_provider *pp, int r, int w, int e) 351{ 352 struct md_s *sc; 353 354 sc = pp->geom->softc; 355 r += pp->acr; 356 w += pp->acw; 357 e += pp->ace; 358 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 359 sc->opencount = 1; 360 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 361 sc->opencount = 0; 362 } 363 return (0); 364} 365 366static void 367g_md_start(struct bio *bp) 368{ 369 struct md_s *sc; 370 371 sc = bp->bio_to->geom->softc; 372 373 switch(bp->bio_cmd) { 374 case BIO_GETATTR: 375 case BIO_SETATTR: 376 g_io_deliver(bp, EOPNOTSUPP); 377 return; 378 } 379 bp->bio_blkno = bp->bio_offset >> DEV_BSHIFT; 380 bp->bio_pblkno = bp->bio_offset / sc->secsize; 381 bp->bio_bcount = bp->bio_length; 382 mtx_lock(&sc->queue_mtx); 383 bioqdisksort(&sc->bio_queue, bp); 384 mtx_unlock(&sc->queue_mtx); 385 386 wakeup(sc); 387} 388 389DECLARE_GEOM_CLASS(g_md_class, g_md); 390 391 392static int 393mdstart_malloc(struct md_s *sc, struct bio *bp) 394{ 395 int i, error; 396 u_char *dst; 397 unsigned secno, nsec, uc; 398 uintptr_t sp, osp; 399 400 nsec = bp->bio_bcount / sc->secsize; 401 secno = bp->bio_pblkno; 402 dst = bp->bio_data; 403 error = 0; 404 while (nsec--) { 405 osp = s_read(sc->indir, secno); 406 if (bp->bio_cmd == BIO_DELETE) { 407 if (osp != 0) 408 error = s_write(sc->indir, secno, 0); 409 } else if (bp->bio_cmd == BIO_READ) { 410 if (osp == 0) 411 bzero(dst, sc->secsize); 412 else if (osp <= 255) 413 for (i = 0; i < sc->secsize; i++) 414 dst[i] = osp; 415 else 416 bcopy((void *)osp, dst, sc->secsize); 417 osp = 0; 418 } else if (bp->bio_cmd == BIO_WRITE) { 419 if (sc->flags & MD_COMPRESS) { 420 uc = dst[0]; 421 for (i = 1; i < sc->secsize; i++) 422 if (dst[i] != uc) 423 break; 424 } else { 425 i = 0; 426 uc = 0; 427 } 428 if (i == sc->secsize) { 429 if (osp != uc) 430 error = s_write(sc->indir, secno, uc); 431 } else { 432 if (osp <= 255) { 433 sp = (uintptr_t) uma_zalloc( 434 sc->uma, M_NOWAIT); 435 if (sp == 0) { 436 error = ENOSPC; 437 break; 438 } 439 bcopy(dst, (void *)sp, sc->secsize); 440 error = s_write(sc->indir, secno, sp); 441 } else { 442 bcopy(dst, (void *)osp, sc->secsize); 443 osp = 0; 444 } 445 } 446 } else { 447 error = EOPNOTSUPP; 448 } 449 if (osp > 255) 450 uma_zfree(sc->uma, (void*)osp); 451 if (error) 452 break; 453 secno++; 454 dst += sc->secsize; 455 } 456 bp->bio_resid = 0; 457 return (error); 458} 459 460static int 461mdstart_preload(struct md_s *sc, struct bio *bp) 462{ 463 464 if (bp->bio_cmd == BIO_DELETE) { 465 } else if (bp->bio_cmd == BIO_READ) { 466 bcopy(sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_data, bp->bio_bcount); 467 } else { 468 bcopy(bp->bio_data, sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_bcount); 469 } 470 bp->bio_resid = 0; 471 return (0); 472} 473 474static int 475mdstart_vnode(struct md_s *sc, struct bio *bp) 476{ 477 int error; 478 struct uio auio; 479 struct iovec aiov; 480 struct mount *mp; 481 482 /* 483 * VNODE I/O 484 * 485 * If an error occurs, we set BIO_ERROR but we do not set 486 * B_INVAL because (for a write anyway), the buffer is 487 * still valid. 488 */ 489 490 bzero(&auio, sizeof(auio)); 491 492 aiov.iov_base = bp->bio_data; 493 aiov.iov_len = bp->bio_bcount; 494 auio.uio_iov = &aiov; 495 auio.uio_iovcnt = 1; 496 auio.uio_offset = (vm_ooffset_t)bp->bio_pblkno * sc->secsize; 497 auio.uio_segflg = UIO_SYSSPACE; 498 if(bp->bio_cmd == BIO_READ) 499 auio.uio_rw = UIO_READ; 500 else 501 auio.uio_rw = UIO_WRITE; 502 auio.uio_resid = bp->bio_bcount; 503 auio.uio_td = curthread; 504 /* 505 * When reading set IO_DIRECT to try to avoid double-caching 506 * the data. When writing IO_DIRECT is not optimal, but we 507 * must set IO_NOWDRAIN to avoid a wdrain deadlock. 508 */ 509 if (bp->bio_cmd == BIO_READ) { 510 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 511 error = VOP_READ(sc->vnode, &auio, IO_DIRECT, sc->cred); 512 } else { 513 (void) vn_start_write(sc->vnode, &mp, V_WAIT); 514 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 515 error = VOP_WRITE(sc->vnode, &auio, IO_NOWDRAIN, sc->cred); 516 vn_finished_write(mp); 517 } 518 VOP_UNLOCK(sc->vnode, 0, curthread); 519 bp->bio_resid = auio.uio_resid; 520 return (error); 521} 522 523static void 524mddone_swap(struct bio *bp) 525{ 526 527 bp->bio_completed = bp->bio_length - bp->bio_resid; 528 g_std_done(bp); 529} 530 531static int 532mdstart_swap(struct md_s *sc, struct bio *bp) 533{ 534 { 535 struct bio *bp2; 536 537 bp2 = g_clone_bio(bp); 538 bp2->bio_done = mddone_swap; 539 bp2->bio_blkno = bp2->bio_offset >> DEV_BSHIFT; 540 bp2->bio_pblkno = bp2->bio_offset / sc->secsize; 541 bp2->bio_bcount = bp2->bio_length; 542 bp = bp2; 543 } 544 545 bp->bio_resid = 0; 546 if ((bp->bio_cmd == BIO_DELETE) && (sc->flags & MD_RESERVE)) 547 biodone(bp); 548 else 549 vm_pager_strategy(sc->object, bp); 550 return (-1); 551} 552 553static void 554md_kthread(void *arg) 555{ 556 struct md_s *sc; 557 struct bio *bp; 558 int error, hasgiant; 559 560 sc = arg; 561 curthread->td_base_pri = PRIBIO; 562 563 switch (sc->type) { 564 case MD_SWAP: 565 case MD_VNODE: 566 mtx_lock(&Giant); 567 hasgiant = 1; 568 break; 569 case MD_MALLOC: 570 case MD_PRELOAD: 571 default: 572 hasgiant = 0; 573 break; 574 } 575 576 for (;;) { 577 mtx_lock(&sc->queue_mtx); 578 bp = bioq_first(&sc->bio_queue); 579 if (bp) 580 bioq_remove(&sc->bio_queue, bp); 581 if (!bp) { 582 if (sc->flags & MD_SHUTDOWN) { 583 mtx_unlock(&sc->queue_mtx); 584 sc->procp = NULL; 585 wakeup(&sc->procp); 586 if (!hasgiant) 587 mtx_lock(&Giant); 588 kthread_exit(0); 589 } 590 msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0); 591 continue; 592 } 593 mtx_unlock(&sc->queue_mtx); 594 595 switch (sc->type) { 596 case MD_MALLOC: 597 devstat_start_transaction(&sc->stats); 598 error = mdstart_malloc(sc, bp); 599 break; 600 case MD_PRELOAD: 601 devstat_start_transaction(&sc->stats); 602 error = mdstart_preload(sc, bp); 603 break; 604 case MD_VNODE: 605 devstat_start_transaction(&sc->stats); 606 error = mdstart_vnode(sc, bp); 607 break; 608 case MD_SWAP: 609 error = mdstart_swap(sc, bp); 610 break; 611 default: 612 panic("Impossible md(type)"); 613 break; 614 } 615 616 if (error != -1) { 617 bp->bio_completed = bp->bio_length; 618 g_io_deliver(bp, error); 619 } 620 } 621} 622 623static struct md_s * 624mdfind(int unit) 625{ 626 struct md_s *sc; 627 628 /* XXX: LOCK(unique unit numbers) */ 629 LIST_FOREACH(sc, &md_softc_list, list) { 630 if (sc->unit == unit) 631 break; 632 } 633 /* XXX: UNLOCK(unique unit numbers) */ 634 return (sc); 635} 636 637static struct md_s * 638mdnew(int unit) 639{ 640 struct md_s *sc; 641 int error, max = -1; 642 643 /* XXX: LOCK(unique unit numbers) */ 644 LIST_FOREACH(sc, &md_softc_list, list) { 645 if (sc->unit == unit) { 646 /* XXX: UNLOCK(unique unit numbers) */ 647 return (NULL); 648 } 649 if (sc->unit > max) 650 max = sc->unit; 651 } 652 if (unit == -1) 653 unit = max + 1; 654 if (unit > 255) 655 return (NULL); 656 sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO); 657 sc->unit = unit; 658 bioq_init(&sc->bio_queue); 659 mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF); 660 sprintf(sc->name, "md%d", unit); 661 error = kthread_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name); 662 if (error) { 663 free(sc, M_MD); 664 return (NULL); 665 } 666 LIST_INSERT_HEAD(&md_softc_list, sc, list); 667 /* XXX: UNLOCK(unique unit numbers) */ 668 return (sc); 669} 670 671static void 672mdinit(struct md_s *sc) 673{ 674 675 devstat_add_entry(&sc->stats, MD_NAME, sc->unit, sc->secsize, 676 DEVSTAT_NO_ORDERED_TAGS, 677 DEVSTAT_TYPE_DIRECT | DEVSTAT_TYPE_IF_OTHER, 678 DEVSTAT_PRIORITY_OTHER); 679 { 680 struct g_geom *gp; 681 struct g_provider *pp; 682 683 DROP_GIANT(); 684 g_topology_lock(); 685 gp = g_new_geomf(&g_md_class, "md%d", sc->unit); 686 gp->start = g_md_start; 687 gp->access = g_md_access; 688 gp->softc = sc; 689 pp = g_new_providerf(gp, "md%d", sc->unit); 690 pp->mediasize = (off_t)sc->nsect * sc->secsize; 691 pp->sectorsize = sc->secsize; 692 sc->gp = gp; 693 sc->pp = pp; 694 g_error_provider(pp, 0); 695 g_topology_unlock(); 696 PICKUP_GIANT(); 697 } 698} 699 700/* 701 * XXX: we should check that the range they feed us is mapped. 702 * XXX: we should implement read-only. 703 */ 704 705static int 706mdcreate_preload(struct md_ioctl *mdio) 707{ 708 struct md_s *sc; 709 710 if (mdio->md_size == 0) 711 return (EINVAL); 712 if (mdio->md_options & ~(MD_AUTOUNIT)) 713 return (EINVAL); 714 if (mdio->md_options & MD_AUTOUNIT) { 715 sc = mdnew(-1); 716 if (sc == NULL) 717 return (ENOMEM); 718 mdio->md_unit = sc->unit; 719 } else { 720 sc = mdnew(mdio->md_unit); 721 if (sc == NULL) 722 return (EBUSY); 723 } 724 sc->type = MD_PRELOAD; 725 sc->secsize = DEV_BSIZE; 726 sc->nsect = mdio->md_size; 727 sc->flags = mdio->md_options & MD_FORCE; 728 /* Cast to pointer size, then to pointer to avoid warning */ 729 sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base; 730 sc->pl_len = (mdio->md_size << DEV_BSHIFT); 731 mdinit(sc); 732 return (0); 733} 734 735 736static int 737mdcreate_malloc(struct md_ioctl *mdio) 738{ 739 struct md_s *sc; 740 off_t u; 741 uintptr_t sp; 742 int error; 743 744 error = 0; 745 if (mdio->md_size == 0) 746 return (EINVAL); 747 if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE)) 748 return (EINVAL); 749 /* Compression doesn't make sense if we have reserved space */ 750 if (mdio->md_options & MD_RESERVE) 751 mdio->md_options &= ~MD_COMPRESS; 752 if (mdio->md_options & MD_AUTOUNIT) { 753 sc = mdnew(-1); 754 if (sc == NULL) 755 return (ENOMEM); 756 mdio->md_unit = sc->unit; 757 } else { 758 sc = mdnew(mdio->md_unit); 759 if (sc == NULL) 760 return (EBUSY); 761 } 762 sc->type = MD_MALLOC; 763 sc->secsize = DEV_BSIZE; 764 sc->nsect = mdio->md_size; 765 sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE); 766 sc->indir = dimension(sc->nsect); 767 sc->uma = uma_zcreate(sc->name, sc->secsize, 768 NULL, NULL, NULL, NULL, 0x1ff, 0); 769 if (mdio->md_options & MD_RESERVE) { 770 for (u = 0; u < sc->nsect; u++) { 771 sp = (uintptr_t) uma_zalloc(sc->uma, M_NOWAIT | M_ZERO); 772 if (sp != 0) 773 error = s_write(sc->indir, u, sp); 774 else 775 error = ENOMEM; 776 if (error) 777 break; 778 } 779 } 780 if (error) { 781 mddestroy(sc, NULL); 782 return (error); 783 } 784 mdinit(sc); 785 if (!(mdio->md_options & MD_RESERVE)) 786 sc->pp->flags |= G_PF_CANDELETE; 787 return (0); 788} 789 790 791static int 792mdsetcred(struct md_s *sc, struct ucred *cred) 793{ 794 char *tmpbuf; 795 int error = 0; 796 797 /* 798 * Set credits in our softc 799 */ 800 801 if (sc->cred) 802 crfree(sc->cred); 803 sc->cred = crhold(cred); 804 805 /* 806 * Horrible kludge to establish credentials for NFS XXX. 807 */ 808 809 if (sc->vnode) { 810 struct uio auio; 811 struct iovec aiov; 812 813 tmpbuf = malloc(sc->secsize, M_TEMP, M_WAITOK); 814 bzero(&auio, sizeof(auio)); 815 816 aiov.iov_base = tmpbuf; 817 aiov.iov_len = sc->secsize; 818 auio.uio_iov = &aiov; 819 auio.uio_iovcnt = 1; 820 auio.uio_offset = 0; 821 auio.uio_rw = UIO_READ; 822 auio.uio_segflg = UIO_SYSSPACE; 823 auio.uio_resid = aiov.iov_len; 824 vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread); 825 error = VOP_READ(sc->vnode, &auio, 0, sc->cred); 826 VOP_UNLOCK(sc->vnode, 0, curthread); 827 free(tmpbuf, M_TEMP); 828 } 829 return (error); 830} 831 832static int 833mdcreate_vnode(struct md_ioctl *mdio, struct thread *td) 834{ 835 struct md_s *sc; 836 struct vattr vattr; 837 struct nameidata nd; 838 int error, flags; 839 840 flags = FREAD|FWRITE; 841 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 842 error = vn_open(&nd, &flags, 0); 843 if (error) { 844 if (error != EACCES && error != EPERM && error != EROFS) 845 return (error); 846 flags &= ~FWRITE; 847 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td); 848 error = vn_open(&nd, &flags, 0); 849 if (error) 850 return (error); 851 } 852 NDFREE(&nd, NDF_ONLY_PNBUF); 853 if (nd.ni_vp->v_type != VREG || 854 (error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred, td))) { 855 VOP_UNLOCK(nd.ni_vp, 0, td); 856 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 857 return (error ? error : EINVAL); 858 } 859 VOP_UNLOCK(nd.ni_vp, 0, td); 860 861 if (mdio->md_options & MD_AUTOUNIT) { 862 sc = mdnew(-1); 863 mdio->md_unit = sc->unit; 864 } else { 865 sc = mdnew(mdio->md_unit); 866 } 867 if (sc == NULL) { 868 (void) vn_close(nd.ni_vp, flags, td->td_ucred, td); 869 return (EBUSY); 870 } 871 872 sc->type = MD_VNODE; 873 sc->flags = mdio->md_options & MD_FORCE; 874 if (!(flags & FWRITE)) 875 sc->flags |= MD_READONLY; 876 sc->secsize = DEV_BSIZE; 877 sc->vnode = nd.ni_vp; 878 879 /* 880 * If the size is specified, override the file attributes. 881 */ 882 if (mdio->md_size) 883 sc->nsect = mdio->md_size; 884 else 885 sc->nsect = vattr.va_size / sc->secsize; /* XXX: round up ? */ 886 if (sc->nsect == 0) { 887 mddestroy(sc, td); 888 return (EINVAL); 889 } 890 error = mdsetcred(sc, td->td_ucred); 891 if (error) { 892 mddestroy(sc, td); 893 return (error); 894 } 895 mdinit(sc); 896 return (0); 897} 898 899static int 900mddestroy(struct md_s *sc, struct thread *td) 901{ 902 903 GIANT_REQUIRED; 904 905 mtx_destroy(&sc->queue_mtx); 906 devstat_remove_entry(&sc->stats); 907 { 908 if (sc->gp) { 909 sc->gp->flags |= G_GEOM_WITHER; 910 sc->gp->softc = NULL; 911 } 912 if (sc->pp) 913 g_orphan_provider(sc->pp, ENXIO); 914 } 915 sc->flags |= MD_SHUTDOWN; 916 wakeup(sc); 917 while (sc->procp != NULL) 918 tsleep(&sc->procp, PRIBIO, "mddestroy", hz / 10); 919 if (sc->vnode != NULL) 920 (void)vn_close(sc->vnode, sc->flags & MD_READONLY ? 921 FREAD : (FREAD|FWRITE), sc->cred, td); 922 if (sc->cred != NULL) 923 crfree(sc->cred); 924 if (sc->object != NULL) { 925 vm_pager_deallocate(sc->object); 926 } 927 if (sc->indir) 928 destroy_indir(sc, sc->indir); 929 if (sc->uma) 930 uma_zdestroy(sc->uma); 931 932 /* XXX: LOCK(unique unit numbers) */ 933 LIST_REMOVE(sc, list); 934 /* XXX: UNLOCK(unique unit numbers) */ 935 free(sc, M_MD); 936 return (0); 937} 938 939static int 940mdcreate_swap(struct md_ioctl *mdio, struct thread *td) 941{ 942 int error; 943 struct md_s *sc; 944 945 GIANT_REQUIRED; 946 947 if (mdio->md_options & MD_AUTOUNIT) { 948 sc = mdnew(-1); 949 mdio->md_unit = sc->unit; 950 } else { 951 sc = mdnew(mdio->md_unit); 952 } 953 if (sc == NULL) 954 return (EBUSY); 955 956 sc->type = MD_SWAP; 957 958 /* 959 * Range check. Disallow negative sizes or any size less then the 960 * size of a page. Then round to a page. 961 */ 962 963 if (mdio->md_size == 0) { 964 mddestroy(sc, td); 965 return (EDOM); 966 } 967 968 /* 969 * Allocate an OBJT_SWAP object. 970 * 971 * sc_secsize is PAGE_SIZE'd 972 * 973 * mdio->size is in DEV_BSIZE'd chunks. 974 * Note the truncation. 975 */ 976 977 sc->secsize = PAGE_SIZE; 978 sc->nsect = mdio->md_size / (PAGE_SIZE / DEV_BSIZE); 979 sc->object = vm_pager_allocate(OBJT_SWAP, NULL, sc->secsize * (vm_offset_t)sc->nsect, VM_PROT_DEFAULT, 0); 980 sc->flags = mdio->md_options & MD_FORCE; 981 if (mdio->md_options & MD_RESERVE) { 982 if (swap_pager_reserve(sc->object, 0, sc->nsect) < 0) { 983 vm_pager_deallocate(sc->object); 984 sc->object = NULL; 985 mddestroy(sc, td); 986 return (EDOM); 987 } 988 } 989 error = mdsetcred(sc, td->td_ucred); 990 if (error) { 991 mddestroy(sc, td); 992 return (error); 993 } 994 mdinit(sc); 995 if (!(mdio->md_options & MD_RESERVE)) 996 sc->pp->flags |= G_PF_CANDELETE; 997 return (0); 998} 999 1000static int 1001mddetach(int unit, struct thread *td) 1002{ 1003 struct md_s *sc; 1004 1005 sc = mdfind(unit); 1006 if (sc == NULL) 1007 return (ENOENT); 1008 if (sc->opencount != 0 && !(sc->flags & MD_FORCE)) 1009 return (EBUSY); 1010 switch(sc->type) { 1011 case MD_VNODE: 1012 case MD_SWAP: 1013 case MD_MALLOC: 1014 case MD_PRELOAD: 1015 return (mddestroy(sc, td)); 1016 default: 1017 return (EOPNOTSUPP); 1018 } 1019} 1020 1021static int 1022mdctlioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 1023{ 1024 struct md_ioctl *mdio; 1025 struct md_s *sc; 1026 int i; 1027 1028 if (md_debug) 1029 printf("mdctlioctl(%s %lx %p %x %p)\n", 1030 devtoname(dev), cmd, addr, flags, td); 1031 1032 /* 1033 * We assert the version number in the individual ioctl 1034 * handlers instead of out here because (a) it is possible we 1035 * may add another ioctl in the future which doesn't read an 1036 * mdio, and (b) the correct return value for an unknown ioctl 1037 * is ENOIOCTL, not EINVAL. 1038 */ 1039 mdio = (struct md_ioctl *)addr; 1040 switch (cmd) { 1041 case MDIOCATTACH: 1042 if (mdio->md_version != MDIOVERSION) 1043 return (EINVAL); 1044 switch (mdio->md_type) { 1045 case MD_MALLOC: 1046 return (mdcreate_malloc(mdio)); 1047 case MD_PRELOAD: 1048 return (mdcreate_preload(mdio)); 1049 case MD_VNODE: 1050 return (mdcreate_vnode(mdio, td)); 1051 case MD_SWAP: 1052 return (mdcreate_swap(mdio, td)); 1053 default: 1054 return (EINVAL); 1055 } 1056 case MDIOCDETACH: 1057 if (mdio->md_version != MDIOVERSION) 1058 return (EINVAL); 1059 if (mdio->md_file != NULL || mdio->md_size != 0 || 1060 mdio->md_options != 0) 1061 return (EINVAL); 1062 return (mddetach(mdio->md_unit, td)); 1063 case MDIOCQUERY: 1064 if (mdio->md_version != MDIOVERSION) 1065 return (EINVAL); 1066 sc = mdfind(mdio->md_unit); 1067 if (sc == NULL) 1068 return (ENOENT); 1069 mdio->md_type = sc->type; 1070 mdio->md_options = sc->flags; 1071 switch (sc->type) { 1072 case MD_MALLOC: 1073 mdio->md_size = sc->nsect; 1074 break; 1075 case MD_PRELOAD: 1076 mdio->md_size = sc->nsect; 1077 mdio->md_base = (uint64_t)(intptr_t)sc->pl_ptr; 1078 break; 1079 case MD_SWAP: 1080 mdio->md_size = sc->nsect * (PAGE_SIZE / DEV_BSIZE); 1081 break; 1082 case MD_VNODE: 1083 mdio->md_size = sc->nsect; 1084 /* XXX fill this in */ 1085 mdio->md_file = NULL; 1086 break; 1087 } 1088 return (0); 1089 case MDIOCLIST: 1090 i = 1; 1091 LIST_FOREACH(sc, &md_softc_list, list) { 1092 if (i == MDNPAD - 1) 1093 mdio->md_pad[i] = -1; 1094 else 1095 mdio->md_pad[i++] = sc->unit; 1096 } 1097 mdio->md_pad[0] = i - 1; 1098 return (0); 1099 default: 1100 return (ENOIOCTL); 1101 }; 1102 return (ENOIOCTL); 1103} 1104 1105static void 1106md_preloaded(u_char *image, unsigned length) 1107{ 1108 struct md_s *sc; 1109 1110 sc = mdnew(-1); 1111 if (sc == NULL) 1112 return; 1113 sc->type = MD_PRELOAD; 1114 sc->secsize = DEV_BSIZE; 1115 sc->nsect = length / DEV_BSIZE; 1116 sc->pl_ptr = image; 1117 sc->pl_len = length; 1118 if (sc->unit == 0) 1119 mdrootready = 1; 1120 mdinit(sc); 1121} 1122 1123static void 1124md_drvinit(void *unused) 1125{ 1126 1127 caddr_t mod; 1128 caddr_t c; 1129 u_char *ptr, *name, *type; 1130 unsigned len; 1131 1132#ifdef MD_ROOT_SIZE 1133 md_preloaded(mfs_root, MD_ROOT_SIZE*1024); 1134#endif 1135 mod = NULL; 1136 while ((mod = preload_search_next_name(mod)) != NULL) { 1137 name = (char *)preload_search_info(mod, MODINFO_NAME); 1138 type = (char *)preload_search_info(mod, MODINFO_TYPE); 1139 if (name == NULL) 1140 continue; 1141 if (type == NULL) 1142 continue; 1143 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 1144 continue; 1145 c = preload_search_info(mod, MODINFO_ADDR); 1146 ptr = *(u_char **)c; 1147 c = preload_search_info(mod, MODINFO_SIZE); 1148 len = *(size_t *)c; 1149 printf("%s%d: Preloaded image <%s> %d bytes at %p\n", 1150 MD_NAME, mdunits, name, len, ptr); 1151 md_preloaded(ptr, len); 1152 } 1153 status_dev = make_dev(&mdctl_cdevsw, 0xffff00ff, UID_ROOT, GID_WHEEL, 1154 0600, MDCTL_NAME); 1155} 1156 1157static int 1158md_modevent(module_t mod, int type, void *data) 1159{ 1160 int error; 1161 struct md_s *sc; 1162 1163 switch (type) { 1164 case MOD_LOAD: 1165 md_drvinit(NULL); 1166 break; 1167 case MOD_UNLOAD: 1168 LIST_FOREACH(sc, &md_softc_list, list) { 1169 error = mddetach(sc->unit, curthread); 1170 if (error != 0) 1171 return (error); 1172 } 1173 if (status_dev) 1174 destroy_dev(status_dev); 1175 status_dev = 0; 1176 break; 1177 default: 1178 break; 1179 } 1180 return (0); 1181} 1182 1183static moduledata_t md_mod = { 1184 MD_NAME, 1185 md_modevent, 1186 NULL 1187}; 1188DECLARE_MODULE(md, md_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE+CDEV_MAJOR); 1189MODULE_VERSION(md, MD_MODVER); 1190 1191 1192#ifdef MD_ROOT 1193static void 1194md_takeroot(void *junk) 1195{ 1196 if (mdrootready) 1197 rootdevnames[0] = "ufs:/dev/md0"; 1198} 1199 1200SYSINIT(md_root, SI_SUB_MOUNT_ROOT, SI_ORDER_FIRST, md_takeroot, NULL); 1201#endif /* MD_ROOT */
|