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 1.14 2006/02/17 19:18:08 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         blockno = loffset / bsize;
 180         blockoff = (loffset % bsize) / DEV_BSIZE;
 181         if ((daddr_t)blockno != blockno) {
 182                 return EINVAL; /* blockno overflow */
 183         }
 184         bp->b_lblkno = blockno;
 185         bp->b_bio2.bio_offset = NOOFFSET;
 186         bp->b_bio2.bio_blkno = (daddr_t)-1;
 187         bp->b_bio2.bio_done = ffs_rawreadwakeup;
 188
 189         error = VOP_BMAP(vp, bp->b_lblkno, &dp, &bp->b_bio2.bio_blkno,
 190                          &bforwards, NULL);
 191         if (error != 0) {
 192                 return error;
 193         }
 194         if (bp->b_bio2.bio_blkno == (daddr_t)-1) {
 195                 /*
 196                  * Fill holes with NULs to preserve semantics
 197                  */
 198                 if (bp->b_bcount + blockoff * DEV_BSIZE > bsize)
 199                         bp->b_bcount = bsize - blockoff * DEV_BSIZE;
 200                 bp->b_bufsize = bp->b_bcount;
 201
 202                 if (vmapbuf(bp) < 0)
 203                         return EFAULT;
 204
 205                 if (ticks - *baseticks >= hogticks) {
 206                         *baseticks = ticks;
 207                         uio_yield();
 208                 }
 209                 bzero(bp->b_data, bp->b_bufsize);
 210
 211                 /* Mark operation completed (similar to bufdone()) */
 212
 213                 bp->b_resid = 0;
 214                 bp->b_flags |= B_DONE;
 215                 return 0;
 216         }
 217
 218         if (bp->b_bcount + blockoff * DEV_BSIZE > bsize * (1 + bforwards))
 219                 bp->b_bcount = bsize * (1 + bforwards) - blockoff * DEV_BSIZE;
 220         bp->b_bufsize = bp->b_bcount;
 221         bp->b_bio2.bio_blkno += blockoff;
 222
 223         if (vmapbuf(bp) < 0)
 224                 return EFAULT;
 225
 226         /*
 227          * Access the block device layer using the device vnode (dp) and
 228          * the translated block number (bio2) instead of the logical block
 229          * number (bio1).
 230          *
 231          * Even though we are bypassing the vnode layer, we still
 232          * want the vnode state to indicate that an I/O on its behalf
 233          * is in progress.
 234          */
 235         bio_start_transaction(&bp->b_bio1, &vp->v_track_read);
 236         vn_strategy(dp, &bp->b_bio2);
 237         return 0;
 238 }
 239
 240
 241 static int
 242 ffs_rawread_main(struct vnode *vp, struct uio *uio)
 243 {
 244         int error, nerror;
 245         struct buf *bp, *nbp, *tbp;
 246         caddr_t sa, nsa, tsa;
 247         uint iolen;
 248         int baseticks = ticks;
 249         caddr_t udata;
 250         long resid;
 251         off_t offset;
 252         struct thread *td;
 253
 254         td = uio->uio_td ? uio->uio_td : curthread;
 255         udata = uio->uio_iov->iov_base;
 256         resid = uio->uio_resid;
 257         offset = uio->uio_offset;
 258
 259         error = 0;
 260         nerror = 0;
 261
 262         bp = NULL;
 263         nbp = NULL;
 264         sa = NULL;
 265         nsa = NULL;
 266
 267         while (resid > 0) {
 268
 269                 if (bp == NULL) { /* Setup first read */
 270                         /* XXX: Leave some bufs for swap */
 271                         bp = getpbuf(&ffsrawbufcnt);
 272                         sa = bp->b_data;
 273                         bp->b_vp = vp;
 274                         error = ffs_rawread_readahead(vp, udata, offset, resid,
 275                                     td, bp, sa, &baseticks);
 276                         if (error != 0)
 277                                 break;
 278
 279                         if (resid > bp->b_bufsize) { /* Setup fist readahead */
 280                                 /* XXX: Leave bufs for swap */
 281                                 if (rawreadahead != 0)
 282                                         nbp = trypbuf(&ffsrawbufcnt);
 283                                 else
 284                                         nbp = NULL;
 285                                 if (nbp != NULL) {
 286                                         nsa = nbp->b_data;
 287                                         nbp->b_vp = vp;
 288
 289                                         nerror = ffs_rawread_readahead(
 290                                                         vp,
 291                                                         udata + bp->b_bufsize,
 292                                                         offset + bp->b_bufsize,
 293                                                         resid - bp->b_bufsize,
 294                                                         td, nbp, nsa,
 295                                                         &baseticks);
 296                                         if (nerror) {
 297                                                 relpbuf(nbp, &ffsrawbufcnt);
 298                                                 nbp = NULL;
 299                                         }
 300                                 }
 301                         }
 302                 }
 303
 304                 crit_enter();
 305                 while ((bp->b_flags & B_DONE) == 0) {
 306                         tsleep((caddr_t)&bp->b_bio2, 0, "rawrd", 0);
 307                 }
 308                 crit_exit();
 309
 310                 vunmapbuf(bp);
 311
 312                 iolen = bp->b_bcount - bp->b_resid;
 313                 if (iolen == 0 && (bp->b_flags & B_ERROR) == 0) {
 314                         nerror = 0;     /* Ignore possible beyond EOF error */
 315                         break; /* EOF */
 316                 }
 317
 318                 if ((bp->b_flags & B_ERROR) != 0) {
 319                         error = bp->b_error;
 320                         break;
 321                 }
 322                 clearbiocache(&bp->b_bio2);
 323                 resid -= iolen;
 324                 udata += iolen;
 325                 offset += iolen;
 326                 if (iolen < bp->b_bufsize) {
 327                         /* Incomplete read.  Try to read remaining part */
 328                         error = ffs_rawread_readahead(
 329                                     vp, udata, offset,
 330                                     bp->b_bufsize - iolen, td, bp,
 331                                     sa, &baseticks);
 332                         if (error != 0)
 333                                 break;
 334                 } else if (nbp != NULL) { /* Complete read with readahead */
 335
 336                         tbp = bp;
 337                         bp = nbp;
 338                         nbp = tbp;
 339
 340                         tsa = sa;
 341                         sa = nsa;
 342                         nsa = tsa;
 343
 344                         clearbiocache(&nbp->b_bio2);
 345
 346                         if (resid <= bp->b_bufsize) { /* No more readaheads */
 347                                 relpbuf(nbp, &ffsrawbufcnt);
 348                                 nbp = NULL;
 349                         } else { /* Setup next readahead */
 350                                 nerror = ffs_rawread_readahead(
 351                                                 vp, udata + bp->b_bufsize,
 352                                                 offset + bp->b_bufsize,
 353                                                 resid - bp->b_bufsize,
 354                                                 td, nbp, nsa, &baseticks);
 355                                 if (nerror != 0) {
 356                                         relpbuf(nbp, &ffsrawbufcnt);
 357                                         nbp = NULL;
 358                                 }
 359                         }
 360                 } else if (nerror != 0) {/* Deferred Readahead error */
 361                         break;
 362                 }  else if (resid > 0) { /* More to read, no readahead */
 363                         error = ffs_rawread_readahead(vp, udata, offset,
 364                                                       resid, td, bp, sa,
 365                                                       &baseticks);
 366                         if (error != 0)
 367                                 break;
 368                 }
 369         }
 370
 371         if (bp != NULL)
 372                 relpbuf(bp, &ffsrawbufcnt);
 373         if (nbp != NULL) {                      /* Run down readahead buffer */
 374                 crit_enter();
 375                 while ((nbp->b_flags & B_DONE) == 0) {
 376                         tsleep(&nbp->b_bio2, 0, "rawrd", 0);
 377                 }
 378                 crit_exit();
 379                 vunmapbuf(nbp);
 380                 relpbuf(nbp, &ffsrawbufcnt);
 381         }
 382
 383         if (error == 0)
 384                 error = nerror;
 385         uio->uio_iov->iov_base = udata;
 386         uio->uio_resid = resid;
 387         uio->uio_offset = offset;
 388         return error;
 389 }
 390
 391
 392 int
 393 ffs_rawread(struct vnode *vp,
 394             struct uio *uio,
 395             int *workdone)
 396 {
 397         if (allowrawread != 0 &&
 398             uio->uio_iovcnt == 1 &&
 399             uio->uio_segflg == UIO_USERSPACE &&
 400             uio->uio_resid == uio->uio_iov->iov_len &&
 401             (((uio->uio_td != NULL) ? uio->uio_td : curthread)->td_flags &
 402              TDF_DEADLKTREAT) == 0) {
 403                 int secsize;            /* Media sector size */
 404                 off_t filebytes;        /* Bytes left of file */
 405                 int blockbytes;         /* Bytes left of file in full blocks */
 406                 int partialbytes;       /* Bytes in last partial block */
 407                 int skipbytes;          /* Bytes not to read in ffs_rawread */
 408                 struct inode *ip;
 409                 int error;
 410
 411
 412                 /* Only handle sector aligned reads */
 413                 ip = VTOI(vp);
 414                 secsize = ip->i_devvp->v_rdev->si_bsize_phys;
 415                 if ((uio->uio_offset & (secsize - 1)) == 0 &&
 416                     (uio->uio_resid & (secsize - 1)) == 0) {
 417
 418                         /* Sync dirty pages and buffers if needed */
 419                         error = ffs_rawread_sync(vp,
 420                                                  (uio->uio_td != NULL) ?
 421                                                  uio->uio_td : curthread);
 422                         if (error != 0)
 423                                 return error;
 424
 425                         /* Check for end of file */
 426                         if (ip->i_size > uio->uio_offset) {
 427                                 filebytes = ip->i_size - uio->uio_offset;
 428
 429                                 /* No special eof handling needed ? */
 430                                 if (uio->uio_resid <= filebytes) {
 431                                         *workdone = 1;
 432                                         return ffs_rawread_main(vp, uio);
 433                                 }
 434
 435                                 partialbytes = ((unsigned int) ip->i_size) %
 436                                         ip->i_fs->fs_bsize;
 437                                 blockbytes = (int) filebytes - partialbytes;
 438                                 if (blockbytes > 0) {
 439                                         skipbytes = uio->uio_resid -
 440                                                 blockbytes;
 441                                         uio->uio_resid = blockbytes;
 442                                         error = ffs_rawread_main(vp, uio);
 443                                         uio->uio_resid += skipbytes;
 444                                         if (error != 0)
 445                                                 return error;
 446                                         /* Read remaining part using buffer */
 447                                 }
 448                         }
 449                 }
 450         }
 451         *workdone = 0;
 452         return 0;
 453 }
 454
 455
 456 static void
 457 ffs_rawreadwakeup(struct bio *bio)
 458 {
 459         wakeup(bio);
 460 }