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 1.7 2004/03/01 06:33:23 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 buf *bp);
  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 spl;
  95         int error;
  96         int upgraded;
  97
  98         /* Check for dirty mmap, pending writes and dirty buffers */
  99         spl = splbio();
 100         if (vp->v_numoutput > 0 ||
 101             !TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
 102             (vp->v_flag & VOBJDIRTY) != 0) {
 103                 splx(spl);
 104
 105                 if (VOP_ISLOCKED(vp, td) != LK_EXCLUSIVE) {
 106                         upgraded = 1;
 107                         /* Upgrade to exclusive lock, this might block */
 108                         VOP_LOCK(vp, NULL, LK_UPGRADE | LK_NOPAUSE, td);
 109                 } else
 110                         upgraded = 0;
 111
 112                 /* Attempt to msync mmap() regions to clean dirty mmap */
 113                 if ((vp->v_flag & VOBJDIRTY) != 0) {
 114                         struct vm_object *obj;
 115                         if (VOP_GETVOBJECT(vp, &obj) == 0)
 116                                 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
 117                 }
 118
 119                 /* Wait for pending writes to complete */
 120                 spl = splbio();
 121                 while (vp->v_numoutput) {
 122                         vp->v_flag |= VBWAIT;
 123                         error = tsleep((caddr_t)&vp->v_numoutput, 0,
 124                                        "rawrdfls", 0);
 125                         if (error != 0) {
 126                                 splx(spl);
 127                                 if (upgraded != 0)
 128                                         VOP_LOCK(vp, NULL, LK_DOWNGRADE, td);
 129                                 return (error);
 130                         }
 131                 }
 132                 /* Flush dirty buffers */
 133                 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
 134                         splx(spl);
 135                         if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0) {
 136                                 if (upgraded != 0)
 137                                         VOP_LOCK(vp, NULL, LK_DOWNGRADE, td);
 138                                 return (error);
 139                         }
 140                         spl = splbio();
 141                         if (vp->v_numoutput > 0 ||
 142                             !TAILQ_EMPTY(&vp->v_dirtyblkhd))
 143                                 panic("ffs_rawread_sync: dirty bufs");
 144                 }
 145                 splx(spl);
 146                 if (upgraded != 0)
 147                         VOP_LOCK(vp, NULL, LK_DOWNGRADE, td);
 148         } else {
 149                 splx(spl);
 150         }
 151         return 0;
 152 }
 153
 154
 155 static int
 156 ffs_rawread_readahead(struct vnode *vp, caddr_t udata, off_t offset,
 157                       size_t len, struct thread *td, struct buf *bp,
 158                       caddr_t sa, int *baseticks)
 159 {
 160         int error;
 161         u_int iolen;
 162         off_t blockno;
 163         int blockoff;
 164         int bsize;
 165         struct vnode *dp;
 166         int bforwards;
 167
 168         bsize = vp->v_mount->mnt_stat.f_iosize;
 169
 170         iolen = ((vm_offset_t) udata) & PAGE_MASK;
 171         bp->b_bcount = len;
 172         if (bp->b_bcount + iolen > bp->b_kvasize) {
 173                 bp->b_bcount = bp->b_kvasize;
 174                 if (iolen != 0)
 175                         bp->b_bcount -= PAGE_SIZE;
 176         }
 177         bp->b_flags = B_PHYS | B_CALL | B_READ;
 178         bp->b_iodone = ffs_rawreadwakeup;
 179         bp->b_data = udata;
 180         bp->b_saveaddr = sa;
 181         bp->b_offset = offset;
 182         blockno = bp->b_offset / bsize;
 183         blockoff = (bp->b_offset % bsize) / DEV_BSIZE;
 184         if ((daddr_t) blockno != blockno) {
 185                 return EINVAL; /* blockno overflow */
 186         }
 187
 188         bp->b_lblkno = bp->b_blkno = blockno;
 189
 190         error = VOP_BMAP(vp, bp->b_lblkno, &dp, &bp->b_blkno, &bforwards,
 191                          NULL);
 192         if (error != 0) {
 193                 return error;
 194         }
 195         if (bp->b_blkno == -1) {
 196
 197                 /* Fill holes with NULs to preserve semantics */
 198
 199                 if (bp->b_bcount + blockoff * DEV_BSIZE > bsize)
 200                         bp->b_bcount = bsize - blockoff * DEV_BSIZE;
 201                 bp->b_bufsize = bp->b_bcount;
 202
 203                 if (vmapbuf(bp) < 0)
 204                         return EFAULT;
 205
 206                 if (ticks - *baseticks >= hogticks) {
 207                         *baseticks = ticks;
 208                         uio_yield();
 209                 }
 210                 bzero(bp->b_data, bp->b_bufsize);
 211
 212                 /* Mark operation completed (similar to bufdone()) */
 213
 214                 bp->b_resid = 0;
 215                 bp->b_flags |= B_DONE;
 216                 return 0;
 217         }
 218
 219         if (bp->b_bcount + blockoff * DEV_BSIZE > bsize * (1 + bforwards))
 220                 bp->b_bcount = bsize * (1 + bforwards) - blockoff * DEV_BSIZE;
 221         bp->b_bufsize = bp->b_bcount;
 222         bp->b_blkno += blockoff;
 223         bp->b_dev = dp->v_rdev;
 224
 225         if (vmapbuf(bp) < 0)
 226                 return EFAULT;
 227
 228         (void) VOP_STRATEGY(dp, bp);
 229         return 0;
 230 }
 231
 232
 233 static int
 234 ffs_rawread_main(struct vnode *vp, struct uio *uio)
 235 {
 236         int error, nerror;
 237         struct buf *bp, *nbp, *tbp;
 238         caddr_t sa, nsa, tsa;
 239         u_int iolen;
 240         int spl;
 241         int baseticks = ticks;
 242         caddr_t udata;
 243         long 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         /*
 253          * keep the process from being swapped
 254          */
 255         if (uio->uio_td && uio->uio_td->td_proc)
 256                 PHOLD(uio->uio_td->td_proc);
 257
 258         error = 0;
 259         nerror = 0;
 260
 261         bp = NULL;
 262         nbp = NULL;
 263         sa = NULL;
 264         nsa = NULL;
 265
 266         while (resid > 0) {
 267
 268                 if (bp == NULL) { /* Setup first read */
 269                         /* XXX: Leave some bufs for swap */
 270                         bp = getpbuf(&ffsrawbufcnt);
 271                         sa = bp->b_data;
 272                         bp->b_vp = vp;
 273                         error = ffs_rawread_readahead(vp, udata, offset, resid,
 274                                     td, bp, sa, &baseticks);
 275                         if (error != 0)
 276                                 break;
 277
 278                         if (resid > bp->b_bufsize) { /* Setup fist readahead */
 279                                 /* XXX: Leave bufs for swap */
 280                                 if (rawreadahead != 0)
 281                                         nbp = trypbuf(&ffsrawbufcnt);
 282                                 else
 283                                         nbp = NULL;
 284                                 if (nbp != NULL) {
 285                                         nsa = nbp->b_data;
 286                                         nbp->b_vp = vp;
 287
 288                                         nerror = ffs_rawread_readahead(
 289                                                         vp,
 290                                                         udata + bp->b_bufsize,
 291                                                         offset + bp->b_bufsize,
 292                                                         resid - bp->b_bufsize,
 293                                                         td, nbp, nsa,
 294                                                         &baseticks);
 295                                         if (nerror) {
 296                                                 relpbuf(nbp, &ffsrawbufcnt);
 297                                                 nbp = NULL;
 298                                         }
 299                                 }
 300                         }
 301                 }
 302
 303                 spl = splbio();
 304                 while ((bp->b_flags & B_DONE) == 0) {
 305                         tsleep((caddr_t)bp, 0, "rawrd", 0);
 306                 }
 307                 splx(spl);
 308
 309                 vunmapbuf(bp);
 310
 311                 iolen = bp->b_bcount - bp->b_resid;
 312                 if (iolen == 0 && (bp->b_flags & B_ERROR) == 0) {
 313                         nerror = 0;     /* Ignore possible beyond EOF error */
 314                         break; /* EOF */
 315                 }
 316
 317                 if ((bp->b_flags & B_ERROR) != 0) {
 318                         error = bp->b_error;
 319                         break;
 320                 }
 321                 resid -= iolen;
 322                 udata += iolen;
 323                 offset += iolen;
 324                 if (iolen < bp->b_bufsize) {
 325                         /* Incomplete read.  Try to read remaining part */
 326                         error = ffs_rawread_readahead(
 327                                     vp, udata, offset,
 328                                     bp->b_bufsize - iolen, td, bp,
 329                                     sa, &baseticks);
 330                         if (error != 0)
 331                                 break;
 332                 } else if (nbp != NULL) { /* Complete read with readahead */
 333
 334                         tbp = bp;
 335                         bp = nbp;
 336                         nbp = tbp;
 337
 338                         tsa = sa;
 339                         sa = nsa;
 340                         nsa = tsa;
 341
 342                         if (resid <= bp->b_bufsize) { /* No more readaheads */
 343                                 relpbuf(nbp, &ffsrawbufcnt);
 344                                 nbp = NULL;
 345                         } else { /* Setup next readahead */
 346                                 nerror = ffs_rawread_readahead(
 347                                                 vp, udata + bp->b_bufsize,
 348                                                 offset + bp->b_bufsize,
 349                                                 resid - bp->b_bufsize,
 350                                                 td, nbp, nsa, &baseticks);
 351                                 if (nerror != 0) {
 352                                         relpbuf(nbp, &ffsrawbufcnt);
 353                                         nbp = NULL;
 354                                 }
 355                         }
 356                 } else if (nerror != 0) {/* Deferred Readahead error */
 357                         break;
 358                 }  else if (resid > 0) { /* More to read, no readahead */
 359                         error = ffs_rawread_readahead(vp, udata, offset,
 360                                                       resid, td, bp, sa,
 361                                                       &baseticks);
 362                         if (error != 0)
 363                                 break;
 364                 }
 365         }
 366
 367         if (bp != NULL)
 368                 relpbuf(bp, &ffsrawbufcnt);
 369         if (nbp != NULL) {                      /* Run down readahead buffer */
 370                 spl = splbio();
 371                 while ((nbp->b_flags & B_DONE) == 0) {
 372                         tsleep((caddr_t)nbp, 0, "rawrd", 0);
 373                 }
 374                 splx(spl);
 375                 vunmapbuf(nbp);
 376                 relpbuf(nbp, &ffsrawbufcnt);
 377         }
 378
 379         if (error == 0)
 380                 error = nerror;
 381         if (uio->uio_td && uio->uio_td->td_proc)
 382                 PRELE(uio->uio_td->td_proc);
 383         uio->uio_iov->iov_base = udata;
 384         uio->uio_resid = resid;
 385         uio->uio_offset = offset;
 386         return error;
 387 }
 388
 389
 390 int
 391 ffs_rawread(struct vnode *vp,
 392             struct uio *uio,
 393             int *workdone)
 394 {
 395         if (allowrawread != 0 &&
 396             uio->uio_iovcnt == 1 &&
 397             uio->uio_segflg == UIO_USERSPACE &&
 398             uio->uio_resid == uio->uio_iov->iov_len &&
 399             (((uio->uio_td != NULL) ? uio->uio_td : curthread)->td_flags &
 400              TDF_DEADLKTREAT) == 0) {
 401                 int secsize;            /* Media sector size */
 402                 off_t filebytes;        /* Bytes left of file */
 403                 int blockbytes;         /* Bytes left of file in full blocks */
 404                 int partialbytes;       /* Bytes in last partial block */
 405                 int skipbytes;          /* Bytes not to read in ffs_rawread */
 406                 struct inode *ip;
 407                 int error;
 408
 409
 410                 /* Only handle sector aligned reads */
 411                 ip = VTOI(vp);
 412                 secsize = ip->i_devvp->v_rdev->si_bsize_phys;
 413                 if ((uio->uio_offset & (secsize - 1)) == 0 &&
 414                     (uio->uio_resid & (secsize - 1)) == 0) {
 415
 416                         /* Sync dirty pages and buffers if needed */
 417                         error = ffs_rawread_sync(vp,
 418                                                  (uio->uio_td != NULL) ?
 419                                                  uio->uio_td : curthread);
 420                         if (error != 0)
 421                                 return error;
 422
 423                         /* Check for end of file */
 424                         if (ip->i_size > uio->uio_offset) {
 425                                 filebytes = ip->i_size - uio->uio_offset;
 426
 427                                 /* No special eof handling needed ? */
 428                                 if (uio->uio_resid <= filebytes) {
 429                                         *workdone = 1;
 430                                         return ffs_rawread_main(vp, uio);
 431                                 }
 432
 433                                 partialbytes = ((unsigned int) ip->i_size) %
 434                                         ip->i_fs->fs_bsize;
 435                                 blockbytes = (int) filebytes - partialbytes;
 436                                 if (blockbytes > 0) {
 437                                         skipbytes = uio->uio_resid -
 438                                                 blockbytes;
 439                                         uio->uio_resid = blockbytes;
 440                                         error = ffs_rawread_main(vp, uio);
 441                                         uio->uio_resid += skipbytes;
 442                                         if (error != 0)
 443                                                 return error;
 444                                         /* Read remaining part using buffer */
 445                                 }
 446                         }
 447                 }
 448         }
 449         *workdone = 0;
 450         return 0;
 451 }
 452
 453
 454 static void
 455 ffs_rawreadwakeup(struct buf *bp)
 456 {
 457         wakeup((caddr_t) bp);
 458 }