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  */
  28
  29 #include <sys/param.h>
  30 #include <sys/systm.h>
  31 #include <sys/fcntl.h>
  32 #include <sys/file.h>
  33 #include <sys/stat.h>
  34 #include <sys/proc.h>
  35 #include <sys/mount.h>
  36 #include <sys/namei.h>
  37 #include <sys/vnode.h>
  38 #include <sys/conf.h>
  39 #include <sys/filio.h>
  40 #include <sys/ttycom.h>
  41 #include <sys/buf.h>
  42 #include <ufs/ufs/quota.h>
  43 #include <ufs/ufs/inode.h>
  44 #include <ufs/ffs/fs.h>
  45
  46 #include <machine/limits.h>
  47 #include <vm/vm.h>
  48 #include <vm/vm_extern.h>
  49 #include <vm/vm_object.h>
  50 #include <sys/kernel.h>
  51 #include <sys/sysctl.h>
  52
  53 static int ffs_rawread_readahead(struct vnode *vp,
  54                                  caddr_t udata,
  55                                  off_t offset,
  56                                  size_t len,
  57                                  struct proc *p,
  58                                  struct buf *bp,
  59                                  caddr_t sa);
  60 static int ffs_rawread_main(struct vnode *vp,
  61                             struct uio *uio);
  62
  63 static int ffs_rawread_sync(struct vnode *vp, struct proc *p);
  64
  65 int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
  66
  67 void ffs_rawread_setup(void);
  68
  69 static void ffs_rawreadwakeup(struct buf *bp);
  70
  71
  72 SYSCTL_DECL(_vfs_ffs);
  73
  74 static int ffsrawbufcnt = 4;
  75 SYSCTL_INT(_vfs_ffs, OID_AUTO, ffsrawbufcnt, CTLFLAG_RD, &ffsrawbufcnt, 0,
  76            "Buffers available for raw reads");
  77
  78 static int allowrawread = 1;
  79 SYSCTL_INT(_vfs_ffs, OID_AUTO, allowrawread, CTLFLAG_RW, &allowrawread, 0,
  80            "Flag to enable raw reads");
  81
  82 static int rawreadahead = 1;
  83 SYSCTL_INT(_vfs_ffs, OID_AUTO, rawreadahead, CTLFLAG_RW, &rawreadahead, 0,
  84            "Flag to enable readahead for long raw reads");
  85
  86
  87 void
  88 ffs_rawread_setup(void)
  89 {
  90         ffsrawbufcnt = (nswbuf > 100 ) ? (nswbuf - (nswbuf >> 4)) : nswbuf - 8;
  91 }
  92
  93
  94 static int
  95 ffs_rawread_sync(struct vnode *vp, struct proc *p)
  96 {
  97         int spl;
  98         int error;
  99         int upgraded;
 100
 101         /* Check for dirty mmap, pending writes and dirty buffers */
 102         spl = splbio();
 103         if (vp->v_numoutput > 0 ||
 104             !TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
 105             (vp->v_flag & VOBJDIRTY) != 0) {
 106                 splx(spl);
 107
 108                 if (VOP_ISLOCKED(vp, p) != LK_EXCLUSIVE) {
 109                         upgraded = 1;
 110                         /* Upgrade to exclusive lock, this might block */
 111                         VOP_LOCK(vp, LK_UPGRADE | LK_NOPAUSE, p);
 112                 } else
 113                         upgraded = 0;
 114
 115                 /* Attempt to msync mmap() regions to clean dirty mmap */
 116                 if ((vp->v_flag & VOBJDIRTY) != 0) {
 117                         struct vm_object *obj;
 118                         if (VOP_GETVOBJECT(vp, &obj) == 0)
 119                                 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
 120                 }
 121
 122                 /* Wait for pending writes to complete */
 123                 spl = splbio();
 124                 while (vp->v_numoutput) {
 125                         vp->v_flag |= VBWAIT;
 126                         error = tsleep((caddr_t)&vp->v_numoutput,
 127                                        PRIBIO + 1,
 128                                        "rawrdfls", 0);
 129                         if (error != 0) {
 130                                 splx(spl);
 131                                 if (upgraded != 0)
 132                                         VOP_LOCK(vp, LK_DOWNGRADE, p);
 133                                 return (error);
 134                         }
 135                 }
 136                 /* Flush dirty buffers */
 137                 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
 138                         splx(spl);
 139                         if ((error = VOP_FSYNC(vp, NOCRED, MNT_WAIT, p)) != 0) {
 140                                 if (upgraded != 0)
 141                                         VOP_LOCK(vp, LK_DOWNGRADE, p);
 142                                 return (error);
 143                         }
 144                         spl = splbio();
 145                         if (vp->v_numoutput > 0 ||
 146                             !TAILQ_EMPTY(&vp->v_dirtyblkhd))
 147                                 panic("ffs_rawread_sync: dirty bufs");
 148                 }
 149                 splx(spl);
 150                 if (upgraded != 0)
 151                         VOP_LOCK(vp, LK_DOWNGRADE, p);
 152         } else {
 153                 splx(spl);
 154         }
 155         return 0;
 156 }
 157
 158
 159 static int
 160 ffs_rawread_readahead(struct vnode *vp,
 161                       caddr_t udata,
 162                       off_t offset,
 163                       size_t len,
 164                       struct proc *p,
 165                       struct buf *bp,
 166                       caddr_t sa)
 167 {
 168         int error;
 169         u_int iolen;
 170         off_t blockno;
 171         int blockoff;
 172         int bsize;
 173         struct vnode *dp;
 174         int bforwards;
 175
 176         bsize = vp->v_mount->mnt_stat.f_iosize;
 177
 178         iolen = ((vm_offset_t) udata) & PAGE_MASK;
 179         bp->b_bcount = len;
 180         if (bp->b_bcount + iolen > bp->b_kvasize) {
 181                 bp->b_bcount = bp->b_kvasize;
 182                 if (iolen != 0)
 183                         bp->b_bcount -= PAGE_SIZE;
 184         }
 185         bp->b_flags = B_PHYS | B_CALL | B_READ;
 186         bp->b_iodone = ffs_rawreadwakeup;
 187         bp->b_data = udata;
 188         bp->b_saveaddr = sa;
 189         bp->b_offset = offset;
 190         blockno = bp->b_offset / bsize;
 191         blockoff = (bp->b_offset % bsize) / DEV_BSIZE;
 192         if ((daddr_t) blockno != blockno) {
 193                 return EINVAL; /* blockno overflow */
 194         }
 195
 196         bp->b_lblkno = bp->b_blkno = blockno;
 197
 198         error = VOP_BMAP(vp, bp->b_lblkno, &dp, &bp->b_blkno, &bforwards,
 199                          NULL);
 200         if (error != 0) {
 201                 return error;
 202         }
 203         if (bp->b_blkno == -1) {
 204
 205                 /* Fill holes with NULs to preserve semantics */
 206
 207                 if (bp->b_bcount + blockoff * DEV_BSIZE > bsize)
 208                         bp->b_bcount = bsize - blockoff * DEV_BSIZE;
 209                 bp->b_bufsize = bp->b_bcount;
 210
 211                 if (vmapbuf(bp) < 0)
 212                         return EFAULT;
 213
 214                 if (ticks - switchticks >= hogticks)
 215                         uio_yield();
 216                 bzero(bp->b_data, bp->b_bufsize);
 217
 218                 /* Mark operation completed (similar to bufdone()) */
 219
 220                 bp->b_resid = 0;
 221                 bp->b_flags |= B_DONE;
 222                 return 0;
 223         }
 224
 225         if (bp->b_bcount + blockoff * DEV_BSIZE > bsize * (1 + bforwards))
 226                 bp->b_bcount = bsize * (1 + bforwards) - blockoff * DEV_BSIZE;
 227         bp->b_bufsize = bp->b_bcount;
 228         bp->b_blkno += blockoff;
 229         bp->b_dev = dp->v_rdev;
 230
 231         if (vmapbuf(bp) < 0)
 232                 return EFAULT;
 233
 234         (void) VOP_STRATEGY(dp, bp);
 235         return 0;
 236 }
 237
 238
 239 static int
 240 ffs_rawread_main(struct vnode *vp,
 241                  struct uio *uio)
 242 {
 243         int error, nerror;
 244         struct buf *bp, *nbp, *tbp;
 245         caddr_t sa, nsa, tsa;
 246         u_int iolen;
 247         int spl;
 248         caddr_t udata;
 249         long resid;
 250         off_t offset;
 251         struct proc *p;
 252
 253         p = uio->uio_procp ? uio->uio_procp : curproc;
 254         udata = uio->uio_iov->iov_base;
 255         resid = uio->uio_resid;
 256         offset = uio->uio_offset;
 257
 258         /*
 259          * keep the process from being swapped
 260          */
 261         PHOLD(p);
 262
 263         error = 0;
 264         nerror = 0;
 265
 266         bp = NULL;
 267         nbp = NULL;
 268         sa = NULL;
 269         nsa = NULL;
 270
 271         while (resid > 0) {
 272
 273                 if (bp == NULL) { /* Setup first read */
 274                         /* XXX: Leave some bufs for swap */
 275                         bp = getpbuf(&ffsrawbufcnt);
 276                         sa = bp->b_data;
 277                         bp->b_vp = vp;
 278                         error = ffs_rawread_readahead(vp, udata, offset,
 279                                                      resid, p, bp, sa);
 280                         if (error != 0)
 281                                 break;
 282
 283                         if (resid > bp->b_bufsize) { /* Setup fist readahead */
 284                                 /* XXX: Leave bufs for swap */
 285                                 if (rawreadahead != 0)
 286                                         nbp = trypbuf(&ffsrawbufcnt);
 287                                 else
 288                                         nbp = NULL;
 289                                 if (nbp != NULL) {
 290                                         nsa = nbp->b_data;
 291                                         nbp->b_vp = vp;
 292
 293                                         nerror = ffs_rawread_readahead(vp,
 294                                                                        udata +
 295                                                                        bp->b_bufsize,
 296                                                                        offset +
 297                                                                        bp->b_bufsize,
 298                                                                        resid -
 299                                                                        bp->b_bufsize,
 300                                                                        p,
 301                                                                        nbp,
 302                                                                        nsa);
 303                                         if (nerror) {
 304                                                 relpbuf(nbp, &ffsrawbufcnt);
 305                                                 nbp = NULL;
 306                                         }
 307                                 }
 308                         }
 309                 }
 310
 311                 spl = splbio();
 312                 while ((bp->b_flags & B_DONE) == 0) {
 313                         tsleep((caddr_t)bp, PRIBIO, "rawrd", 0);
 314                 }
 315                 splx(spl);
 316
 317                 vunmapbuf(bp);
 318
 319                 iolen = bp->b_bcount - bp->b_resid;
 320                 if (iolen == 0 && (bp->b_flags & B_ERROR) == 0) {
 321                         nerror = 0;     /* Ignore possible beyond EOF error */
 322                         break; /* EOF */
 323                 }
 324
 325                 if ((bp->b_flags & B_ERROR) != 0) {
 326                         error = bp->b_error;
 327                         break;
 328                 }
 329                 resid -= iolen;
 330                 udata += iolen;
 331                 offset += iolen;
 332                 if (iolen < bp->b_bufsize) {
 333                         /* Incomplete read.  Try to read remaining part */
 334                         error = ffs_rawread_readahead(vp,
 335                                                       udata,
 336                                                       offset,
 337                                                       bp->b_bufsize - iolen,
 338                                                       p,
 339                                                       bp,
 340                                                       sa);
 341                         if (error != 0)
 342                                 break;
 343                 } else if (nbp != NULL) { /* Complete read with readahead */
 344
 345                         tbp = bp;
 346                         bp = nbp;
 347                         nbp = tbp;
 348
 349                         tsa = sa;
 350                         sa = nsa;
 351                         nsa = tsa;
 352
 353                         if (resid <= bp->b_bufsize) { /* No more readaheads */
 354                                 relpbuf(nbp, &ffsrawbufcnt);
 355                                 nbp = NULL;
 356                         } else { /* Setup next readahead */
 357                                 nerror = ffs_rawread_readahead(vp,
 358                                                                udata +
 359                                                                bp->b_bufsize,
 360                                                                offset +
 361                                                                bp->b_bufsize,
 362                                                                resid -
 363                                                                bp->b_bufsize,
 364                                                                p,
 365                                                                nbp,
 366                                                                nsa);
 367                                 if (nerror != 0) {
 368                                         relpbuf(nbp, &ffsrawbufcnt);
 369                                         nbp = NULL;
 370                                 }
 371                         }
 372                 } else if (nerror != 0) {/* Deferred Readahead error */
 373                         break;
 374                 }  else if (resid > 0) { /* More to read, no readahead */
 375                         error = ffs_rawread_readahead(vp, udata, offset,
 376                                                       resid, p, bp, sa);
 377                         if (error != 0)
 378                                 break;
 379                 }
 380         }
 381
 382         if (bp != NULL)
 383                 relpbuf(bp, &ffsrawbufcnt);
 384         if (nbp != NULL) {                      /* Run down readahead buffer */
 385                 spl = splbio();
 386                 while ((nbp->b_flags & B_DONE) == 0) {
 387                         tsleep((caddr_t)nbp, PRIBIO, "rawrd", 0);
 388                 }
 389                 splx(spl);
 390                 vunmapbuf(nbp);
 391                 relpbuf(nbp, &ffsrawbufcnt);
 392         }
 393
 394         if (error == 0)
 395                 error = nerror;
 396         PRELE(p);
 397         uio->uio_iov->iov_base = udata;
 398         uio->uio_resid = resid;
 399         uio->uio_offset = offset;
 400         return error;
 401 }
 402
 403
 404 int
 405 ffs_rawread(struct vnode *vp,
 406             struct uio *uio,
 407             int *workdone)
 408 {
 409         if (allowrawread != 0 &&
 410             uio->uio_iovcnt == 1 &&
 411             uio->uio_segflg == UIO_USERSPACE &&
 412             uio->uio_resid == uio->uio_iov->iov_len &&
 413             (((uio->uio_procp != NULL) ? uio->uio_procp : curproc)->p_flag &
 414              P_DEADLKTREAT) == 0) {
 415                 int secsize;            /* Media sector size */
 416                 off_t filebytes;        /* Bytes left of file */
 417                 int blockbytes;         /* Bytes left of file in full blocks */
 418                 int partialbytes;       /* Bytes in last partial block */
 419                 int skipbytes;          /* Bytes not to read in ffs_rawread */
 420                 struct inode *ip;
 421                 int error;
 422
 423
 424                 /* Only handle sector aligned reads */
 425                 ip = VTOI(vp);
 426                 secsize = ip->i_devvp->v_rdev->si_bsize_phys;
 427                 if ((uio->uio_offset & (secsize - 1)) == 0 &&
 428                     (uio->uio_resid & (secsize - 1)) == 0) {
 429
 430                         /* Sync dirty pages and buffers if needed */
 431                         error = ffs_rawread_sync(vp,
 432                                                  (uio->uio_procp != NULL) ?
 433                                                  uio->uio_procp : curproc);
 434                         if (error != 0)
 435                                 return error;
 436
 437                         /* Check for end of file */
 438                         if (ip->i_size > uio->uio_offset) {
 439                                 filebytes = ip->i_size - uio->uio_offset;
 440
 441                                 /* No special eof handling needed ? */
 442                                 if (uio->uio_resid <= filebytes) {
 443                                         *workdone = 1;
 444                                         return ffs_rawread_main(vp, uio);
 445                                 }
 446
 447                                 partialbytes = ((unsigned int) ip->i_size) %
 448                                         ip->i_fs->fs_bsize;
 449                                 blockbytes = (int) filebytes - partialbytes;
 450                                 if (blockbytes > 0) {
 451                                         skipbytes = uio->uio_resid -
 452                                                 blockbytes;
 453                                         uio->uio_resid = blockbytes;
 454                                         error = ffs_rawread_main(vp, uio);
 455                                         uio->uio_resid += skipbytes;
 456                                         if (error != 0)
 457                                                 return error;
 458                                         /* Read remaining part using buffer */
 459                                 }
 460                         }
 461                 }
 462         }
 463         *workdone = 0;
 464         return 0;
 465 }
 466
 467
 468 static void
 469 ffs_rawreadwakeup(struct buf *bp)
 470 {
 471         wakeup((caddr_t) bp);
 472 }