sys/vfs/specfs/spec_vnops.c

   1 /*
   2  * Copyright (c) 1989, 1993, 1995
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. All advertising materials mentioning features or use of this software
  14  *    must display the following acknowledgement:
  15  *      This product includes software developed by the University of
  16  *      California, Berkeley and its contributors.
  17  * 4. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  *      @(#)spec_vnops.c        8.14 (Berkeley) 5/21/95
  34  * $FreeBSD: src/sys/miscfs/specfs/spec_vnops.c,v 1.131.2.4 2001/02/26 04:23:20 jlemon Exp $
  35  * $DragonFly: src/sys/vfs/specfs/spec_vnops.c,v 1.58 2008/05/06 00:14:12 dillon Exp $
  36  */
  37
  38 #include <sys/param.h>
  39 #include <sys/proc.h>
  40 #include <sys/systm.h>
  41 #include <sys/kernel.h>
  42 #include <sys/conf.h>
  43 #include <sys/buf.h>
  44 #include <sys/device.h>
  45 #include <sys/mount.h>
  46 #include <sys/vnode.h>
  47 #include <sys/stat.h>
  48 #include <sys/fcntl.h>
  49 #include <sys/vmmeter.h>
  50 #include <sys/bus.h>
  51 #include <sys/tty.h>
  52
  53 #include <vm/vm.h>
  54 #include <vm/vm_object.h>
  55 #include <vm/vm_page.h>
  56 #include <vm/vm_pager.h>
  57
  58 #include <machine/limits.h>
  59
  60 #include <sys/buf2.h>
  61
  62 #include <sys/thread2.h>
  63
  64 /*
  65  * Specfs chained debugging (bitmask)
  66  *
  67  * 0 - disable debugging
  68  * 1 - report chained I/Os
  69  * 2 - force 4K chained I/Os
  70  */
  71 #define SPEC_CHAIN_DEBUG        0
  72
  73 static int      spec_advlock (struct vop_advlock_args *);
  74 static int      spec_bmap (struct vop_bmap_args *);
  75 static int      spec_close (struct vop_close_args *);
  76 static int      spec_freeblks (struct vop_freeblks_args *);
  77 static int      spec_fsync (struct  vop_fsync_args *);
  78 static int      spec_getpages (struct vop_getpages_args *);
  79 static int      spec_inactive (struct  vop_inactive_args *);
  80 static int      spec_ioctl (struct vop_ioctl_args *);
  81 static int      spec_open (struct vop_open_args *);
  82 static int      spec_poll (struct vop_poll_args *);
  83 static int      spec_kqfilter (struct vop_kqfilter_args *);
  84 static int      spec_print (struct vop_print_args *);
  85 static int      spec_read (struct vop_read_args *);
  86 static int      spec_strategy (struct vop_strategy_args *);
  87 static int      spec_write (struct vop_write_args *);
  88 static void     spec_strategy_done(struct bio *nbio);
  89
  90 struct vop_ops spec_vnode_vops = {
  91         .vop_default =          vop_defaultop,
  92         .vop_access =           (void *)vop_ebadf,
  93         .vop_advlock =          spec_advlock,
  94         .vop_bmap =             spec_bmap,
  95         .vop_close =            spec_close,
  96         .vop_old_create =       (void *)vop_panic,
  97         .vop_freeblks =         spec_freeblks,
  98         .vop_fsync =            spec_fsync,
  99         .vop_getpages =         spec_getpages,
 100         .vop_inactive =         spec_inactive,
 101         .vop_ioctl =            spec_ioctl,
 102         .vop_old_link =         (void *)vop_panic,
 103         .vop_old_mkdir =        (void *)vop_panic,
 104         .vop_old_mknod =        (void *)vop_panic,
 105         .vop_open =             spec_open,
 106         .vop_pathconf =         vop_stdpathconf,
 107         .vop_poll =             spec_poll,
 108         .vop_kqfilter =         spec_kqfilter,
 109         .vop_print =            spec_print,
 110         .vop_read =             spec_read,
 111         .vop_readdir =          (void *)vop_panic,
 112         .vop_readlink =         (void *)vop_panic,
 113         .vop_reallocblks =      (void *)vop_panic,
 114         .vop_reclaim =          (void *)vop_null,
 115         .vop_old_remove =       (void *)vop_panic,
 116         .vop_old_rename =       (void *)vop_panic,
 117         .vop_old_rmdir =        (void *)vop_panic,
 118         .vop_setattr =          (void *)vop_ebadf,
 119         .vop_strategy =         spec_strategy,
 120         .vop_old_symlink =      (void *)vop_panic,
 121         .vop_write =            spec_write
 122 };
 123
 124 struct vop_ops *spec_vnode_vops_p = &spec_vnode_vops;
 125
 126 VNODEOP_SET(spec_vnode_vops);
 127
 128 extern int dev_ref_debug;
 129
 130 /*
 131  * spec_vnoperate()
 132  */
 133 int
 134 spec_vnoperate(struct vop_generic_args *ap)
 135 {
 136         return (VOCALL(&spec_vnode_vops, ap));
 137 }
 138
 139 static void spec_getpages_iodone (struct bio *bio);
 140
 141 /*
 142  * Open a special file.
 143  *
 144  * spec_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
 145  *           struct file *a_fp)
 146  */
 147 /* ARGSUSED */
 148 static int
 149 spec_open(struct vop_open_args *ap)
 150 {
 151         struct vnode *vp = ap->a_vp;
 152         cdev_t dev;
 153         int error;
 154         const char *cp;
 155
 156         /*
 157          * Don't allow open if fs is mounted -nodev.
 158          */
 159         if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
 160                 return (ENXIO);
 161         if (vp->v_type == VBLK)
 162                 return (ENXIO);
 163
 164         /*
 165          * Resolve the device.  If the vnode is already open v_rdev may
 166          * already be resolved.  However, if the device changes out from
 167          * under us we report it (and, for now, we allow it).  Since
 168          * v_release_rdev() zero's v_opencount, we have to save and restore
 169          * it when replacing the rdev reference.
 170          */
 171         if (vp->v_rdev != NULL) {
 172                 dev = get_dev(vp->v_umajor, vp->v_uminor);
 173                 if (dev != vp->v_rdev) {
 174                         int oc = vp->v_opencount;
 175                         kprintf(
 176                             "Warning: spec_open: dev %s was lost",
 177                             vp->v_rdev->si_name);
 178                         v_release_rdev(vp);
 179                         error = v_associate_rdev(vp,
 180                                         get_dev(vp->v_umajor, vp->v_uminor));
 181                         if (error) {
 182                                 kprintf(", reacquisition failed\n");
 183                         } else {
 184                                 vp->v_opencount = oc;
 185                                 kprintf(", reacquisition successful\n");
 186                         }
 187                 } else {
 188                         error = 0;
 189                 }
 190         } else {
 191                 error = v_associate_rdev(vp, get_dev(vp->v_umajor, vp->v_uminor));
 192         }
 193         if (error)
 194                 return(error);
 195
 196         /*
 197          * Prevent degenerate open/close sequences from nulling out rdev.
 198          */
 199         dev = vp->v_rdev;
 200         KKASSERT(dev != NULL);
 201
 202         /*
 203          * Make this field valid before any I/O in ->d_open.  XXX the
 204          * device itself should probably be required to initialize
 205          * this field in d_open.
 206          */
 207         if (!dev->si_iosize_max)
 208                 dev->si_iosize_max = DFLTPHYS;
 209
 210         /*
 211          * XXX: Disks get special billing here, but it is mostly wrong.
 212          * XXX: diskpartitions can overlap and the real checks should
 213          * XXX: take this into account, and consequently they need to
 214          * XXX: live in the diskslicing code.  Some checks do.
 215          */
 216         if (vn_isdisk(vp, NULL) && ap->a_cred != FSCRED &&
 217             (ap->a_mode & FWRITE)) {
 218                 /*
 219                  * Never allow opens for write if the device is mounted R/W
 220                  */
 221                 if (vp->v_rdev && vp->v_rdev->si_mountpoint &&
 222                     !(vp->v_rdev->si_mountpoint->mnt_flag & MNT_RDONLY)) {
 223                                 error = EBUSY;
 224                                 goto done;
 225                 }
 226
 227                 /*
 228                  * When running in secure mode, do not allow opens
 229                  * for writing if the device is mounted
 230                  */
 231                 if (securelevel >= 1 && vfs_mountedon(vp)) {
 232                         error = EPERM;
 233                         goto done;
 234                 }
 235
 236                 /*
 237                  * When running in very secure mode, do not allow
 238                  * opens for writing of any devices.
 239                  */
 240                 if (securelevel >= 2) {
 241                         error = EPERM;
 242                         goto done;
 243                 }
 244         }
 245
 246         /* XXX: Special casing of ttys for deadfs.  Probably redundant */
 247         if (dev_dflags(dev) & D_TTY)
 248                 vp->v_flag |= VISTTY;
 249
 250         /*
 251          * dev_dopen() is always called for each open.  dev_dclose() is
 252          * only called for the last close unless D_TRACKCLOSE is set.
 253          */
 254         vn_unlock(vp);
 255         error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred);
 256         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 257
 258         if (error)
 259                 goto done;
 260
 261         if (dev_dflags(dev) & D_TTY) {
 262                 if (dev->si_tty) {
 263                         struct tty *tp;
 264                         tp = dev->si_tty;
 265                         if (!tp->t_stop) {
 266                                 kprintf("Warning:%s: no t_stop, using nottystop\n", devtoname(dev));
 267                                 tp->t_stop = nottystop;
 268                         }
 269                 }
 270         }
 271
 272         /*
 273          * If this is 'disk' or disk-like device, associate a VM object
 274          * with it.
 275          */
 276         if (vn_isdisk(vp, NULL)) {
 277                 if (!dev->si_bsize_phys)
 278                         dev->si_bsize_phys = DEV_BSIZE;
 279                 vinitvmio(vp, IDX_TO_OFF(INT_MAX));
 280         }
 281         if ((dev_dflags(dev) & D_DISK) == 0) {
 282                 cp = devtoname(dev);
 283                 if (*cp == '#') {
 284                         kprintf("WARNING: driver %s should register devices with make_dev() (cdev_t = \"%s\")\n",
 285                             dev_dname(dev), cp);
 286                 }
 287         }
 288
 289         /*
 290          * If we were handed a file pointer we may be able to install a
 291          * shortcut which issues device read and write operations directly
 292          * from the fileops rather then having to go through spec_read()
 293          * and spec_write().
 294          */
 295         if (ap->a_fp)
 296                 vn_setspecops(ap->a_fp);
 297
 298         if (dev_ref_debug)
 299                 kprintf("spec_open: %s %d\n", dev->si_name, vp->v_opencount);
 300 done:
 301         if (error) {
 302                 if (vp->v_opencount == 0)
 303                         v_release_rdev(vp);
 304         } else {
 305                 vop_stdopen(ap);
 306         }
 307         return (error);
 308 }
 309
 310 /*
 311  * Vnode op for read
 312  *
 313  * spec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
 314  *           struct ucred *a_cred)
 315  */
 316 /* ARGSUSED */
 317 static int
 318 spec_read(struct vop_read_args *ap)
 319 {
 320         struct vnode *vp;
 321         struct thread *td;
 322         struct uio *uio;
 323         cdev_t dev;
 324         int error;
 325
 326         vp = ap->a_vp;
 327         dev = vp->v_rdev;
 328         uio = ap->a_uio;
 329         td = uio->uio_td;
 330
 331         if (dev == NULL)                /* device was revoked */
 332                 return (EBADF);
 333         if (uio->uio_resid == 0)
 334                 return (0);
 335
 336         vn_unlock(vp);
 337         error = dev_dread(dev, uio, ap->a_ioflag);
 338         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 339         return (error);
 340 }
 341
 342 /*
 343  * Vnode op for write
 344  *
 345  * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
 346  *            struct ucred *a_cred)
 347  */
 348 /* ARGSUSED */
 349 static int
 350 spec_write(struct vop_write_args *ap)
 351 {
 352         struct vnode *vp;
 353         struct thread *td;
 354         struct uio *uio;
 355         cdev_t dev;
 356         int error;
 357
 358         vp = ap->a_vp;
 359         dev = vp->v_rdev;
 360         uio = ap->a_uio;
 361         td = uio->uio_td;
 362
 363         KKASSERT(uio->uio_segflg != UIO_NOCOPY);
 364
 365         if (dev == NULL)                /* device was revoked */
 366                 return (EBADF);
 367
 368         vn_unlock(vp);
 369         error = dev_dwrite(dev, uio, ap->a_ioflag);
 370         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 371         return (error);
 372 }
 373
 374 /*
 375  * Device ioctl operation.
 376  *
 377  * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
 378  *            int a_fflag, struct ucred *a_cred)
 379  */
 380 /* ARGSUSED */
 381 static int
 382 spec_ioctl(struct vop_ioctl_args *ap)
 383 {
 384         cdev_t dev;
 385
 386         if ((dev = ap->a_vp->v_rdev) == NULL)
 387                 return (EBADF);         /* device was revoked */
 388
 389         return (dev_dioctl(dev, ap->a_command, ap->a_data,
 390                     ap->a_fflag, ap->a_cred));
 391 }
 392
 393 /*
 394  * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred)
 395  */
 396 /* ARGSUSED */
 397 static int
 398 spec_poll(struct vop_poll_args *ap)
 399 {
 400         cdev_t dev;
 401
 402         if ((dev = ap->a_vp->v_rdev) == NULL)
 403                 return (EBADF);         /* device was revoked */
 404         return (dev_dpoll(dev, ap->a_events));
 405 }
 406
 407 /*
 408  * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
 409  */
 410 /* ARGSUSED */
 411 static int
 412 spec_kqfilter(struct vop_kqfilter_args *ap)
 413 {
 414         cdev_t dev;
 415
 416         if ((dev = ap->a_vp->v_rdev) == NULL)
 417                 return (EBADF);         /* device was revoked */
 418         return (dev_dkqfilter(dev, ap->a_kn));
 419 }
 420
 421 /*
 422  * Synch buffers associated with a block device
 423  *
 424  * spec_fsync(struct vnode *a_vp, int a_waitfor)
 425  */
 426 /* ARGSUSED */
 427 static int
 428 spec_fsync(struct vop_fsync_args *ap)
 429 {
 430         struct vnode *vp = ap->a_vp;
 431         int error;
 432
 433         if (!vn_isdisk(vp, NULL))
 434                 return (0);
 435
 436         /*
 437          * Flush all dirty buffers associated with a block device.
 438          */
 439         error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
 440         return (error);
 441 }
 442
 443 /*
 444  * spec_inactive(struct vnode *a_vp)
 445  */
 446 static int
 447 spec_inactive(struct vop_inactive_args *ap)
 448 {
 449         return (0);
 450 }
 451
 452 /*
 453  * Convert a vnode strategy call into a device strategy call.  Vnode strategy
 454  * calls are not limited to device DMA limits so we have to deal with the
 455  * case.
 456  *
 457  * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
 458  */
 459 static int
 460 spec_strategy(struct vop_strategy_args *ap)
 461 {
 462         struct bio *bio = ap->a_bio;
 463         struct buf *bp = bio->bio_buf;
 464         struct buf *nbp;
 465         struct vnode *vp;
 466         struct mount *mp;
 467         int chunksize;
 468         int maxiosize;
 469
 470         if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
 471                 buf_start(bp);
 472
 473         /*
 474          * Collect statistics on synchronous and asynchronous read
 475          * and write counts for disks that have associated filesystems.
 476          */
 477         vp = ap->a_vp;
 478         KKASSERT(vp->v_rdev != NULL);   /* XXX */
 479         if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
 480                 if (bp->b_cmd == BUF_CMD_READ) {
 481                         if (bp->b_flags & B_ASYNC)
 482                                 mp->mnt_stat.f_asyncreads++;
 483                         else
 484                                 mp->mnt_stat.f_syncreads++;
 485                 } else {
 486                         if (bp->b_flags & B_ASYNC)
 487                                 mp->mnt_stat.f_asyncwrites++;
 488                         else
 489                                 mp->mnt_stat.f_syncwrites++;
 490                 }
 491         }
 492
 493         /*
 494          * Device iosize limitations only apply to read and write.  Shortcut
 495          * the I/O if it fits.
 496          */
 497         if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
 498                 kprintf("%s: si_iosize_max not set!\n", dev_dname(vp->v_rdev));
 499                 maxiosize = MAXPHYS;
 500         }
 501 #if SPEC_CHAIN_DEBUG & 2
 502         maxiosize = 4096;
 503 #endif
 504         if (bp->b_bcount <= maxiosize ||
 505             (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
 506                 dev_dstrategy_chain(vp->v_rdev, bio);
 507                 return (0);
 508         }
 509
 510         /*
 511          * Clone the buffer and set up an I/O chain to chunk up the I/O.
 512          */
 513         nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
 514         initbufbio(nbp);
 515         buf_dep_init(nbp);
 516         BUF_LOCKINIT(nbp);
 517         BUF_LOCK(nbp, LK_EXCLUSIVE);
 518         BUF_KERNPROC(nbp);
 519         nbp->b_vp = vp;
 520         nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
 521         nbp->b_data = bp->b_data;
 522         nbp->b_bio1.bio_done = spec_strategy_done;
 523         nbp->b_bio1.bio_offset = bio->bio_offset;
 524         nbp->b_bio1.bio_caller_info1.ptr = bio;
 525
 526         /*
 527          * Start the first transfer
 528          */
 529         if (vn_isdisk(vp, NULL))
 530                 chunksize = vp->v_rdev->si_bsize_phys;
 531         else
 532                 chunksize = DEV_BSIZE;
 533         chunksize = maxiosize / chunksize * chunksize;
 534 #if SPEC_CHAIN_DEBUG & 1
 535         kprintf("spec_strategy chained I/O chunksize=%d\n", chunksize);
 536 #endif
 537         nbp->b_cmd = bp->b_cmd;
 538         nbp->b_bcount = chunksize;
 539         nbp->b_bufsize = chunksize;     /* used to detect a short I/O */
 540         nbp->b_bio1.bio_caller_info2.index = chunksize;
 541
 542 #if SPEC_CHAIN_DEBUG & 1
 543         kprintf("spec_strategy: chain %p offset %d/%d bcount %d\n",
 544                 bp, 0, bp->b_bcount, nbp->b_bcount);
 545 #endif
 546
 547         dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
 548         return (0);
 549 }
 550
 551 /*
 552  * Chunked up transfer completion routine - chain transfers until done
 553  */
 554 static
 555 void
 556 spec_strategy_done(struct bio *nbio)
 557 {
 558         struct buf *nbp = nbio->bio_buf;
 559         struct bio *bio = nbio->bio_caller_info1.ptr;   /* original bio */
 560         struct buf *bp = bio->bio_buf;                  /* original bp */
 561         int chunksize = nbio->bio_caller_info2.index;   /* chunking */
 562         int boffset = nbp->b_data - bp->b_data;
 563
 564         if (nbp->b_flags & B_ERROR) {
 565                 /*
 566                  * An error terminates the chain, propogate the error back
 567                  * to the original bp
 568                  */
 569                 bp->b_flags |= B_ERROR;
 570                 bp->b_error = nbp->b_error;
 571                 bp->b_resid = bp->b_bcount - boffset +
 572                               (nbp->b_bcount - nbp->b_resid);
 573 #if SPEC_CHAIN_DEBUG & 1
 574                 kprintf("spec_strategy: chain %p error %d bcount %d/%d\n",
 575                         bp, bp->b_error, bp->b_bcount,
 576                         bp->b_bcount - bp->b_resid);
 577 #endif
 578                 kfree(nbp, M_DEVBUF);
 579                 biodone(bio);
 580         } else if (nbp->b_resid) {
 581                 /*
 582                  * A short read or write terminates the chain
 583                  */
 584                 bp->b_error = nbp->b_error;
 585                 bp->b_resid = bp->b_bcount - boffset +
 586                               (nbp->b_bcount - nbp->b_resid);
 587 #if SPEC_CHAIN_DEBUG & 1
 588                 kprintf("spec_strategy: chain %p short read(1) bcount %d/%d\n",
 589                         bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
 590 #endif
 591                 kfree(nbp, M_DEVBUF);
 592                 biodone(bio);
 593         } else if (nbp->b_bcount != nbp->b_bufsize) {
 594                 /*
 595                  * A short read or write can also occur by truncating b_bcount
 596                  */
 597 #if SPEC_CHAIN_DEBUG & 1
 598                 kprintf("spec_strategy: chain %p short read(2) bcount %d/%d\n",
 599                         bp, nbp->b_bcount + boffset, bp->b_bcount);
 600 #endif
 601                 bp->b_error = 0;
 602                 bp->b_bcount = nbp->b_bcount + boffset;
 603                 bp->b_resid = nbp->b_resid;
 604                 kfree(nbp, M_DEVBUF);
 605                 biodone(bio);
 606         } else if (nbp->b_bcount + boffset == bp->b_bcount) {
 607                 /*
 608                  * No more data terminates the chain
 609                  */
 610 #if SPEC_CHAIN_DEBUG & 1
 611                 kprintf("spec_strategy: chain %p finished bcount %d\n",
 612                         bp, bp->b_bcount);
 613 #endif
 614                 bp->b_error = 0;
 615                 bp->b_resid = 0;
 616                 kfree(nbp, M_DEVBUF);
 617                 biodone(bio);
 618         } else {
 619                 /*
 620                  * Continue the chain
 621                  */
 622                 boffset += nbp->b_bcount;
 623                 nbp->b_data = bp->b_data + boffset;
 624                 nbp->b_bcount = bp->b_bcount - boffset;
 625                 if (nbp->b_bcount > chunksize)
 626                         nbp->b_bcount = chunksize;
 627                 nbp->b_bio1.bio_done = spec_strategy_done;
 628                 nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
 629
 630 #if SPEC_CHAIN_DEBUG & 1
 631                 kprintf("spec_strategy: chain %p offset %d/%d bcount %d\n",
 632                         bp, boffset, bp->b_bcount, nbp->b_bcount);
 633 #endif
 634
 635                 dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
 636         }
 637 }
 638
 639 /*
 640  * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
 641  */
 642 static int
 643 spec_freeblks(struct vop_freeblks_args *ap)
 644 {
 645         struct buf *bp;
 646
 647         /*
 648          * XXX: This assumes that strategy does the deed right away.
 649          * XXX: this may not be TRTTD.
 650          */
 651         KKASSERT(ap->a_vp->v_rdev != NULL);
 652         if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0)
 653                 return (0);
 654         bp = geteblk(ap->a_length);
 655         bp->b_cmd = BUF_CMD_FREEBLKS;
 656         bp->b_bio1.bio_offset = ap->a_offset;
 657         bp->b_bcount = ap->a_length;
 658         dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
 659         return (0);
 660 }
 661
 662 /*
 663  * Implement degenerate case where the block requested is the block
 664  * returned, and assume that the entire device is contiguous in regards
 665  * to the contiguous block range (runp and runb).
 666  *
 667  * spec_bmap(struct vnode *a_vp, off_t a_loffset,
 668  *           off_t *a_doffsetp, int *a_runp, int *a_runb)
 669  */
 670 static int
 671 spec_bmap(struct vop_bmap_args *ap)
 672 {
 673         if (ap->a_doffsetp != NULL)
 674                 *ap->a_doffsetp = ap->a_loffset;
 675         if (ap->a_runp != NULL)
 676                 *ap->a_runp = MAXBSIZE;
 677         if (ap->a_runb != NULL) {
 678                 if (ap->a_loffset < MAXBSIZE)
 679                         *ap->a_runb = (int)ap->a_loffset;
 680                 else
 681                         *ap->a_runb = MAXBSIZE;
 682         }
 683         return (0);
 684 }
 685
 686 /*
 687  * Device close routine
 688  *
 689  * spec_close(struct vnode *a_vp, int a_fflag)
 690  *
 691  * NOTE: the vnode may or may not be locked on call.
 692  */
 693 /* ARGSUSED */
 694 static int
 695 spec_close(struct vop_close_args *ap)
 696 {
 697         struct proc *p = curproc;
 698         struct vnode *vp = ap->a_vp;
 699         cdev_t dev = vp->v_rdev;
 700         int error;
 701         int needrelock;
 702
 703         /*
 704          * Hack: a tty device that is a controlling terminal
 705          * has a reference from the session structure.
 706          * We cannot easily tell that a character device is
 707          * a controlling terminal, unless it is the closing
 708          * process' controlling terminal.  In that case,
 709          * if the reference count is 2 (this last descriptor
 710          * plus the session), release the reference from the session.
 711          *
 712          * It is possible for v_opencount to be 0 or 1 in this case, 0
 713          * because the tty might have been revoked.
 714          */
 715         if (dev)
 716                 reference_dev(dev);
 717         if (vcount(vp) == 2 && vp->v_opencount <= 1 &&
 718             p && vp == p->p_session->s_ttyvp) {
 719                 p->p_session->s_ttyvp = NULL;
 720                 vrele(vp);
 721         }
 722
 723         /*
 724          * Vnodes can be opened and close multiple times.  Do not really
 725          * close the device unless (1) it is being closed forcibly,
 726          * (2) the device wants to track closes, or (3) this is the last
 727          * vnode doing its last close on the device.
 728          *
 729          * XXX the VXLOCK (force close) case can leave vnodes referencing
 730          * a closed device.
 731          */
 732         if (dev && ((vp->v_flag & VRECLAIMED) ||
 733             (dev_dflags(dev) & D_TRACKCLOSE) ||
 734             (vcount(vp) <= 1 && vp->v_opencount == 1))) {
 735                 needrelock = 0;
 736                 if (vn_islocked(vp)) {
 737                         needrelock = 1;
 738                         vn_unlock(vp);
 739                 }
 740                 error = dev_dclose(dev, ap->a_fflag, S_IFCHR);
 741                 if (needrelock)
 742                         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 743         } else {
 744                 error = 0;
 745         }
 746
 747         /*
 748          * Track the actual opens and closes on the vnode.  The last close
 749          * disassociates the rdev.  If the rdev is already disassociated
 750          * the vnode might have been revoked and no further opencount
 751          * tracking occurs.
 752          */
 753         if (dev) {
 754                 /*KKASSERT(vp->v_opencount > 0);*/
 755                 if (dev_ref_debug) {
 756                         kprintf("spec_close: %s %d\n",
 757                                 dev->si_name, vp->v_opencount - 1);
 758                 }
 759                 if (vp->v_opencount == 1)
 760                         v_release_rdev(vp);
 761                 release_dev(dev);
 762         }
 763         vop_stdclose(ap);
 764         return(error);
 765 }
 766
 767 /*
 768  * Print out the contents of a special device vnode.
 769  *
 770  * spec_print(struct vnode *a_vp)
 771  */
 772 static int
 773 spec_print(struct vop_print_args *ap)
 774 {
 775         kprintf("tag VT_NON, dev %s\n", devtoname(ap->a_vp->v_rdev));
 776         return (0);
 777 }
 778
 779 /*
 780  * Special device advisory byte-level locks.
 781  *
 782  * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
 783  *              struct flock *a_fl, int a_flags)
 784  */
 785 /* ARGSUSED */
 786 static int
 787 spec_advlock(struct vop_advlock_args *ap)
 788 {
 789         return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
 790 }
 791
 792 static void
 793 spec_getpages_iodone(struct bio *bio)
 794 {
 795         bio->bio_buf->b_cmd = BUF_CMD_DONE;
 796         wakeup(bio->bio_buf);
 797 }
 798
 799 /*
 800  * spec_getpages() - get pages associated with device vnode.
 801  *
 802  * Note that spec_read and spec_write do not use the buffer cache, so we
 803  * must fully implement getpages here.
 804  */
 805 static int
 806 spec_getpages(struct vop_getpages_args *ap)
 807 {
 808         vm_offset_t kva;
 809         int error;
 810         int i, pcount, size;
 811         struct buf *bp;
 812         vm_page_t m;
 813         vm_ooffset_t offset;
 814         int toff, nextoff, nread;
 815         struct vnode *vp = ap->a_vp;
 816         int blksiz;
 817         int gotreqpage;
 818
 819         error = 0;
 820         pcount = round_page(ap->a_count) / PAGE_SIZE;
 821
 822         /*
 823          * Calculate the offset of the transfer and do sanity check.
 824          */
 825         offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
 826
 827         /*
 828          * Round up physical size for real devices.  We cannot round using
 829          * v_mount's block size data because v_mount has nothing to do with
 830          * the device.  i.e. it's usually '/dev'.  We need the physical block
 831          * size for the device itself.
 832          *
 833          * We can't use v_rdev->si_mountpoint because it only exists when the
 834          * block device is mounted.  However, we can use v_rdev.
 835          */
 836
 837         if (vn_isdisk(vp, NULL))
 838                 blksiz = vp->v_rdev->si_bsize_phys;
 839         else
 840                 blksiz = DEV_BSIZE;
 841
 842         size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
 843
 844         bp = getpbuf(NULL);
 845         kva = (vm_offset_t)bp->b_data;
 846
 847         /*
 848          * Map the pages to be read into the kva.
 849          */
 850         pmap_qenter(kva, ap->a_m, pcount);
 851
 852         /* Build a minimal buffer header. */
 853         bp->b_cmd = BUF_CMD_READ;
 854         bp->b_bcount = size;
 855         bp->b_resid = 0;
 856         bp->b_runningbufspace = size;
 857         if (size) {
 858                 runningbufspace += bp->b_runningbufspace;
 859                 ++runningbufcount;
 860         }
 861
 862         bp->b_bio1.bio_offset = offset;
 863         bp->b_bio1.bio_done = spec_getpages_iodone;
 864
 865         mycpu->gd_cnt.v_vnodein++;
 866         mycpu->gd_cnt.v_vnodepgsin += pcount;
 867
 868         /* Do the input. */
 869         vn_strategy(ap->a_vp, &bp->b_bio1);
 870
 871         crit_enter();
 872
 873         /* We definitely need to be at splbio here. */
 874         while (bp->b_cmd != BUF_CMD_DONE)
 875                 tsleep(bp, 0, "spread", 0);
 876
 877         crit_exit();
 878
 879         if (bp->b_flags & B_ERROR) {
 880                 if (bp->b_error)
 881                         error = bp->b_error;
 882                 else
 883                         error = EIO;
 884         }
 885
 886         /*
 887          * If EOF is encountered we must zero-extend the result in order
 888          * to ensure that the page does not contain garabge.  When no
 889          * error occurs, an early EOF is indicated if b_bcount got truncated.
 890          * b_resid is relative to b_bcount and should be 0, but some devices
 891          * might indicate an EOF with b_resid instead of truncating b_bcount.
 892          */
 893         nread = bp->b_bcount - bp->b_resid;
 894         if (nread < ap->a_count)
 895                 bzero((caddr_t)kva + nread, ap->a_count - nread);
 896         pmap_qremove(kva, pcount);
 897
 898         gotreqpage = 0;
 899         for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
 900                 nextoff = toff + PAGE_SIZE;
 901                 m = ap->a_m[i];
 902
 903                 m->flags &= ~PG_ZERO;
 904
 905                 if (nextoff <= nread) {
 906                         m->valid = VM_PAGE_BITS_ALL;
 907                         vm_page_undirty(m);
 908                 } else if (toff < nread) {
 909                         /*
 910                          * Since this is a VM request, we have to supply the
 911                          * unaligned offset to allow vm_page_set_validclean()
 912                          * to zero sub-DEV_BSIZE'd portions of the page.
 913                          */
 914                         vm_page_set_validclean(m, 0, nread - toff);
 915                 } else {
 916                         m->valid = 0;
 917                         vm_page_undirty(m);
 918                 }
 919
 920                 if (i != ap->a_reqpage) {
 921                         /*
 922                          * Just in case someone was asking for this page we
 923                          * now tell them that it is ok to use.
 924                          */
 925                         if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
 926                                 if (m->valid) {
 927                                         if (m->flags & PG_WANTED) {
 928                                                 vm_page_activate(m);
 929                                         } else {
 930                                                 vm_page_deactivate(m);
 931                                         }
 932                                         vm_page_wakeup(m);
 933                                 } else {
 934                                         vm_page_free(m);
 935                                 }
 936                         } else {
 937                                 vm_page_free(m);
 938                         }
 939                 } else if (m->valid) {
 940                         gotreqpage = 1;
 941                         /*
 942                          * Since this is a VM request, we need to make the
 943                          * entire page presentable by zeroing invalid sections.
 944                          */
 945                         if (m->valid != VM_PAGE_BITS_ALL)
 946                             vm_page_zero_invalid(m, FALSE);
 947                 }
 948         }
 949         if (!gotreqpage) {
 950                 m = ap->a_m[ap->a_reqpage];
 951                 kprintf(
 952             "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
 953                         devtoname(vp->v_rdev), error, bp, bp->b_vp);
 954                 kprintf(
 955             "               size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
 956                     size, bp->b_resid, ap->a_count, m->valid);
 957                 kprintf(
 958             "               nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
 959                     nread, ap->a_reqpage, (u_long)m->pindex, pcount);
 960                 /*
 961                  * Free the buffer header back to the swap buffer pool.
 962                  */
 963                 relpbuf(bp, NULL);
 964                 return VM_PAGER_ERROR;
 965         }
 966         /*
 967          * Free the buffer header back to the swap buffer pool.
 968          */
 969         relpbuf(bp, NULL);
 970         return VM_PAGER_OK;
 971 }