sys/vfs/ufs/ffs_rawread.c

   1 /*-
   2  * Copyright (c) 2000-2003 Tor Egge
   3  * 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  *
  14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  24  * SUCH DAMAGE.
  25  *
  26  * $FreeBSD: src/sys/ufs/ffs/ffs_rawread.c,v 1.3.2.2 2003/05/29 06:15:35 alc Exp $
  27  * $DragonFly: src/sys/vfs/ufs/ffs_rawread.c,v 1.15 2006/03/24 18:35:34 dillon Exp $
  28  */
  29
  30 #include <sys/param.h>
  31 #include <sys/systm.h>
  32 #include <sys/fcntl.h>
  33 #include <sys/file.h>
  34 #include <sys/stat.h>
  35 #include <sys/proc.h>
  36 #include <sys/mount.h>
  37 #include <sys/namei.h>
  38 #include <sys/vnode.h>
  39 #include <sys/conf.h>
  40 #include <sys/filio.h>
  41 #include <sys/ttycom.h>
  42 #include <sys/buf.h>
  43 #include "quota.h"
  44 #include "inode.h"
  45 #include "fs.h"
  46
  47 #include <machine/limits.h>
  48 #include <vm/vm.h>
  49 #include <vm/vm_extern.h>
  50 #include <vm/vm_object.h>
  51 #include <sys/kernel.h>
  52 #include <sys/sysctl.h>
  53
  54 static int ffs_rawread_readahead(struct vnode *vp, caddr_t udata, off_t offset,
  55                                  size_t len, struct thread *td, struct buf *bp,
  56                                  caddr_t sa, int *baseticks);
  57 static int ffs_rawread_main(struct vnode *vp,
  58                             struct uio *uio);
  59
  60 static int ffs_rawread_sync(struct vnode *vp, struct thread *td);
  61
  62 int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
  63
  64 void ffs_rawread_setup(void);
  65
  66 static void ffs_rawreadwakeup(struct bio *bio);
  67
  68
  69 SYSCTL_DECL(_vfs_ffs);
  70
  71 static int ffsrawbufcnt = 4;
  72 SYSCTL_INT(_vfs_ffs, OID_AUTO, ffsrawbufcnt, CTLFLAG_RD, &ffsrawbufcnt, 0,
  73            "Buffers available for raw reads");
  74
  75 static int allowrawread = 1;
  76 SYSCTL_INT(_vfs_ffs, OID_AUTO, allowrawread, CTLFLAG_RW, &allowrawread, 0,
  77            "Flag to enable raw reads");
  78
  79 static int rawreadahead = 1;
  80 SYSCTL_INT(_vfs_ffs, OID_AUTO, rawreadahead, CTLFLAG_RW, &rawreadahead, 0,
  81            "Flag to enable readahead for long raw reads");
  82
  83
  84 void
  85 ffs_rawread_setup(void)
  86 {
  87         ffsrawbufcnt = (nswbuf > 100 ) ? (nswbuf - (nswbuf >> 4)) : nswbuf - 8;
  88 }
  89
  90
  91 static int
  92 ffs_rawread_sync(struct vnode *vp, struct thread *td)
  93 {
  94         int error;
  95         int upgraded;
  96
  97         /* Check for dirty mmap, pending writes and dirty buffers */
  98         crit_enter();
  99         if (vp->v_track_write.bk_active > 0 ||
 100             !RB_EMPTY(&vp->v_rbdirty_tree) ||
 101             (vp->v_flag & VOBJDIRTY) != 0) {
 102                 crit_exit();
 103
 104                 if (VOP_ISLOCKED(vp, td) != LK_EXCLUSIVE) {
 105                         upgraded = 1;
 106                         /* Upgrade to exclusive lock, this might block */
 107                         VOP_LOCK(vp, LK_UPGRADE | LK_NOPAUSE, td);
 108                 } else
 109                         upgraded = 0;
 110
 111                 /* Attempt to msync mmap() regions to clean dirty mmap */
 112                 if ((vp->v_flag & VOBJDIRTY) != 0) {
 113                         struct vm_object *obj;
 114                         if (VOP_GETVOBJECT(vp, &obj) == 0)
 115                                 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
 116                 }
 117
 118                 /* Wait for pending writes to complete */
 119                 crit_enter();
 120                 while (vp->v_track_write.bk_active) {
 121                         vp->v_track_write.bk_waitflag = 1;
 122                         error = tsleep(&vp->v_track_write, 0, "rawrdfls", 0);
 123                         if (error != 0) {
 124                                 crit_exit();
 125                                 if (upgraded != 0)
 126                                         VOP_LOCK(vp, LK_DOWNGRADE, td);
 127                                 return (error);
 128                         }
 129                 }
 130                 /* Flush dirty buffers */
 131                 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
 132                         crit_exit();
 133                         if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0) {
 134                                 if (upgraded != 0)
 135                                         VOP_LOCK(vp, LK_DOWNGRADE, td);
 136                                 return (error);
 137                         }
 138                         crit_enter();
 139                         if (vp->v_track_write.bk_active > 0 ||
 140                             !RB_EMPTY(&vp->v_rbdirty_tree))
 141                                 panic("ffs_rawread_sync: dirty bufs");
 142                 }
 143                 crit_exit();
 144                 if (upgraded != 0)
 145                         VOP_LOCK(vp, LK_DOWNGRADE, td);
 146         } else {
 147                 crit_exit();
 148         }
 149         return 0;
 150 }
 151
 152
 153 static int
 154 ffs_rawread_readahead(struct vnode *vp, caddr_t udata, off_t loffset,
 155                       size_t len, struct thread *td, struct buf *bp,
 156                       caddr_t sa, int *baseticks)
 157 {
 158         int error;
 159         uint iolen;
 160         off_t blockno;
 161         int blockoff;
 162         int bsize;
 163         struct vnode *dp;
 164         int bforwards;
 165
 166         bsize = vp->v_mount->mnt_stat.f_iosize;
 167
 168         iolen = ((vm_offset_t) udata) & PAGE_MASK;
 169         bp->b_bcount = len;
 170         if (bp->b_bcount + iolen > bp->b_kvasize) {
 171                 bp->b_bcount = bp->b_kvasize;
 172                 if (iolen != 0)
 173                         bp->b_bcount -= PAGE_SIZE;
 174         }
 175         bp->b_flags = B_PHYS | B_READ;
 176         bp->b_data = udata;
 177         bp->b_saveaddr = sa;
 178         bp->b_loffset = loffset;
 179         bp->b_bio2.bio_offset = NOOFFSET;
 180         bp->b_bio2.bio_done = ffs_rawreadwakeup;
 181
 182         error = VOP_BMAP(vp, bp->b_loffset, &dp, &bp->b_bio2.bio_offset,
 183                          &bforwards, NULL);
 184         if (error != 0) {
 185                 return error;
 186         }
 187         if (bp->b_bio2.bio_offset == NOOFFSET) {
 188                 /*
 189                  * Fill holes with NULs to preserve semantics
 190                  */
 191                 if (bp->b_bcount + blockoff * DEV_BSIZE > bsize)
 192                         bp->b_bcount = bsize - blockoff * DEV_BSIZE;
 193                 bp->b_bufsize = bp->b_bcount;
 194
 195                 if (vmapbuf(bp) < 0)
 196                         return EFAULT;
 197
 198                 if (ticks - *baseticks >= hogticks) {
 199                         *baseticks = ticks;
 200                         uio_yield();
 201                 }
 202                 bzero(bp->b_data, bp->b_bufsize);
 203
 204                 /* Mark operation completed (similar to bufdone()) */
 205
 206                 bp->b_resid = 0;
 207                 bp->b_flags |= B_DONE;
 208                 return 0;
 209         }
 210
 211         if (bp->b_bcount + blockoff * DEV_BSIZE > bforwards)
 212                 bp->b_bcount = bforwards - blockoff * DEV_BSIZE;
 213         bp->b_bufsize = bp->b_bcount;
 214         bp->b_bio2.bio_offset += blockoff * DEV_BSIZE;
 215
 216         if (vmapbuf(bp) < 0)
 217                 return EFAULT;
 218
 219         /*
 220          * Access the block device layer using the device vnode (dp) and
 221          * the translated block number (bio2) instead of the logical block
 222          * number (bio1).
 223          *
 224          * Even though we are bypassing the vnode layer, we still
 225          * want the vnode state to indicate that an I/O on its behalf
 226          * is in progress.
 227          */
 228         bio_start_transaction(&bp->b_bio1, &vp->v_track_read);
 229         vn_strategy(dp, &bp->b_bio2);
 230         return 0;
 231 }
 232
 233
 234 static int
 235 ffs_rawread_main(struct vnode *vp, struct uio *uio)
 236 {
 237         int error, nerror;
 238         struct buf *bp, *nbp, *tbp;
 239         caddr_t sa, nsa, tsa;
 240         uint iolen;
 241         int baseticks = ticks;
 242         caddr_t udata;
 243         int resid;
 244         off_t offset;
 245         struct thread *td;
 246
 247         td = uio->uio_td ? uio->uio_td : curthread;
 248         udata = uio->uio_iov->iov_base;
 249         resid = uio->uio_resid;
 250         offset = uio->uio_offset;
 251
 252         error = 0;
 253         nerror = 0;
 254
 255         bp = NULL;
 256         nbp = NULL;
 257         sa = NULL;
 258         nsa = NULL;
 259
 260         while (resid > 0) {
 261
 262                 if (bp == NULL) { /* Setup first read */
 263                         /* XXX: Leave some bufs for swap */
 264                         bp = getpbuf(&ffsrawbufcnt);
 265                         sa = bp->b_data;
 266                         bp->b_vp = vp;
 267                         error = ffs_rawread_readahead(vp, udata, offset, resid,
 268                                     td, bp, sa, &baseticks);
 269                         if (error != 0)
 270                                 break;
 271
 272                         if (resid > bp->b_bufsize) { /* Setup fist readahead */
 273                                 /* XXX: Leave bufs for swap */
 274                                 if (rawreadahead != 0)
 275                                         nbp = trypbuf(&ffsrawbufcnt);
 276                                 else
 277                                         nbp = NULL;
 278                                 if (nbp != NULL) {
 279                                         nsa = nbp->b_data;
 280                                         nbp->b_vp = vp;
 281
 282                                         nerror = ffs_rawread_readahead(
 283                                                         vp,
 284                                                         udata + bp->b_bufsize,
 285                                                         offset + bp->b_bufsize,
 286                                                         resid - bp->b_bufsize,
 287                                                         td, nbp, nsa,
 288                                                         &baseticks);
 289                                         if (nerror) {
 290                                                 relpbuf(nbp, &ffsrawbufcnt);
 291                                                 nbp = NULL;
 292                                         }
 293                                 }
 294                         }
 295                 }
 296
 297                 crit_enter();
 298                 while ((bp->b_flags & B_DONE) == 0) {
 299                         tsleep((caddr_t)&bp->b_bio2, 0, "rawrd", 0);
 300                 }
 301                 crit_exit();
 302
 303                 vunmapbuf(bp);
 304
 305                 iolen = bp->b_bcount - bp->b_resid;
 306                 if (iolen == 0 && (bp->b_flags & B_ERROR) == 0) {
 307                         nerror = 0;     /* Ignore possible beyond EOF error */
 308                         break; /* EOF */
 309                 }
 310
 311                 if ((bp->b_flags & B_ERROR) != 0) {
 312                         error = bp->b_error;
 313                         break;
 314                 }
 315                 clearbiocache(&bp->b_bio2);
 316                 resid -= iolen;
 317                 udata += iolen;
 318                 offset += iolen;
 319                 if (iolen < bp->b_bufsize) {
 320                         /* Incomplete read.  Try to read remaining part */
 321                         error = ffs_rawread_readahead(
 322                                     vp, udata, offset,
 323                                     bp->b_bufsize - iolen, td, bp,
 324                                     sa, &baseticks);
 325                         if (error != 0)
 326                                 break;
 327                 } else if (nbp != NULL) { /* Complete read with readahead */
 328
 329                         tbp = bp;
 330                         bp = nbp;
 331                         nbp = tbp;
 332
 333                         tsa = sa;
 334                         sa = nsa;
 335                         nsa = tsa;
 336
 337                         clearbiocache(&nbp->b_bio2);
 338
 339                         if (resid <= bp->b_bufsize) { /* No more readaheads */
 340                                 relpbuf(nbp, &ffsrawbufcnt);
 341                                 nbp = NULL;
 342                         } else { /* Setup next readahead */
 343                                 nerror = ffs_rawread_readahead(
 344                                                 vp, udata + bp->b_bufsize,
 345                                                 offset + bp->b_bufsize,
 346                                                 resid - bp->b_bufsize,
 347                                                 td, nbp, nsa, &baseticks);
 348                                 if (nerror != 0) {
 349                                         relpbuf(nbp, &ffsrawbufcnt);
 350                                         nbp = NULL;
 351                                 }
 352                         }
 353                 } else if (nerror != 0) {/* Deferred Readahead error */
 354                         break;
 355                 }  else if (resid > 0) { /* More to read, no readahead */
 356                         error = ffs_rawread_readahead(vp, udata, offset,
 357                                                       resid, td, bp, sa,
 358                                                       &baseticks);
 359                         if (error != 0)
 360                                 break;
 361                 }
 362         }
 363
 364         if (bp != NULL)
 365                 relpbuf(bp, &ffsrawbufcnt);
 366         if (nbp != NULL) {                      /* Run down readahead buffer */
 367                 crit_enter();
 368                 while ((nbp->b_flags & B_DONE) == 0) {
 369                         tsleep(&nbp->b_bio2, 0, "rawrd", 0);
 370                 }
 371                 crit_exit();
 372                 vunmapbuf(nbp);
 373                 relpbuf(nbp, &ffsrawbufcnt);
 374         }
 375
 376         if (error == 0)
 377                 error = nerror;
 378         uio->uio_iov->iov_base = udata;
 379         uio->uio_resid = resid;
 380         uio->uio_offset = offset;
 381         return error;
 382 }
 383
 384
 385 int
 386 ffs_rawread(struct vnode *vp,
 387             struct uio *uio,
 388             int *workdone)
 389 {
 390         if (allowrawread != 0 &&
 391             uio->uio_iovcnt == 1 &&
 392             uio->uio_segflg == UIO_USERSPACE &&
 393             uio->uio_resid == uio->uio_iov->iov_len &&
 394             (((uio->uio_td != NULL) ? uio->uio_td : curthread)->td_flags &
 395              TDF_DEADLKTREAT) == 0) {
 396                 int secsize;            /* Media sector size */
 397                 off_t filebytes;        /* Bytes left of file */
 398                 int blockbytes;         /* Bytes left of file in full blocks */
 399                 int partialbytes;       /* Bytes in last partial block */
 400                 int skipbytes;          /* Bytes not to read in ffs_rawread */
 401                 struct inode *ip;
 402                 int error;
 403
 404
 405                 /* Only handle sector aligned reads */
 406                 ip = VTOI(vp);
 407                 secsize = ip->i_devvp->v_rdev->si_bsize_phys;
 408                 if ((uio->uio_offset & (secsize - 1)) == 0 &&
 409                     (uio->uio_resid & (secsize - 1)) == 0) {
 410
 411                         /* Sync dirty pages and buffers if needed */
 412                         error = ffs_rawread_sync(vp,
 413                                                  (uio->uio_td != NULL) ?
 414                                                  uio->uio_td : curthread);
 415                         if (error != 0)
 416                                 return error;
 417
 418                         /* Check for end of file */
 419                         if (ip->i_size > uio->uio_offset) {
 420                                 filebytes = ip->i_size - uio->uio_offset;
 421
 422                                 /* No special eof handling needed ? */
 423                                 if (uio->uio_resid <= filebytes) {
 424                                         *workdone = 1;
 425                                         return ffs_rawread_main(vp, uio);
 426                                 }
 427
 428                                 partialbytes = ((unsigned int) ip->i_size) %
 429                                         ip->i_fs->fs_bsize;
 430                                 blockbytes = (int) filebytes - partialbytes;
 431                                 if (blockbytes > 0) {
 432                                         skipbytes = uio->uio_resid -
 433                                                 blockbytes;
 434                                         uio->uio_resid = blockbytes;
 435                                         error = ffs_rawread_main(vp, uio);
 436                                         uio->uio_resid += skipbytes;
 437                                         if (error != 0)
 438                                                 return error;
 439                                         /* Read remaining part using buffer */
 440                                 }
 441                         }
 442                 }
 443         }
 444         *workdone = 0;
 445         return 0;
 446 }
 447
 448
 449 static void
 450 ffs_rawreadwakeup(struct bio *bio)
 451 {
 452         wakeup(bio);
 453 }