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 1.28 2008/06/19 23:27:39 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 buf *bp);
  56 static int ffs_rawread_main(struct vnode *vp,
  57                             struct uio *uio);
  58
  59 static int ffs_rawread_sync(struct vnode *vp);
  60
  61 int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
  62
  63 void ffs_rawread_setup(void);
  64
  65 SYSCTL_DECL(_vfs_ffs);
  66
  67 static int ffsrawbufcnt = 4;
  68 SYSCTL_INT(_vfs_ffs, OID_AUTO, ffsrawbufcnt, CTLFLAG_RD, &ffsrawbufcnt, 0,
  69            "Buffers available for raw reads");
  70
  71 static int allowrawread = 1;
  72 SYSCTL_INT(_vfs_ffs, OID_AUTO, allowrawread, CTLFLAG_RW, &allowrawread, 0,
  73            "Flag to enable raw reads");
  74
  75 static int rawreadahead = 1;
  76 SYSCTL_INT(_vfs_ffs, OID_AUTO, rawreadahead, CTLFLAG_RW, &rawreadahead, 0,
  77            "Flag to enable readahead for long raw reads");
  78
  79
  80 void
  81 ffs_rawread_setup(void)
  82 {
  83         ffsrawbufcnt = (nswbuf > 100 ) ? (nswbuf - (nswbuf >> 4)) : nswbuf - 8;
  84 }
  85
  86
  87 static int
  88 ffs_rawread_sync(struct vnode *vp)
  89 {
  90         int error;
  91         int upgraded;
  92
  93         /*
  94          * Check for dirty mmap, pending writes and dirty buffers
  95          */
  96         lwkt_gettoken(&vp->v_token);
  97         if (bio_track_active(&vp->v_track_write) ||
  98             !RB_EMPTY(&vp->v_rbdirty_tree) ||
  99             (vp->v_flag & VOBJDIRTY) != 0) {
 100                 if (vn_islocked(vp) != LK_EXCLUSIVE) {
 101                         upgraded = 1;
 102                         /* Upgrade to exclusive lock, this might block */
 103                         vn_lock(vp, LK_UPGRADE);
 104                 } else
 105                         upgraded = 0;
 106
 107                 /* Attempt to msync mmap() regions to clean dirty mmap */
 108                 if ((vp->v_flag & VOBJDIRTY) != 0) {
 109                         struct vm_object *obj;
 110                         if ((obj = vp->v_object) != NULL)
 111                                 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
 112                 }
 113
 114                 /* Wait for pending writes to complete */
 115                 error = bio_track_wait(&vp->v_track_write, 0, 0);
 116                 if (error != 0) {
 117                         if (upgraded != 0)
 118                                 vn_lock(vp, LK_DOWNGRADE);
 119                         goto done;
 120                 }
 121                 /* Flush dirty buffers */
 122                 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
 123                         if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0) {
 124                                 if (upgraded != 0)
 125                                         vn_lock(vp, LK_DOWNGRADE);
 126                                 goto done;
 127                         }
 128                         if (bio_track_active(&vp->v_track_write) ||
 129                             !RB_EMPTY(&vp->v_rbdirty_tree))
 130                                 panic("ffs_rawread_sync: dirty bufs");
 131                 }
 132                 if (upgraded != 0)
 133                         vn_lock(vp, LK_DOWNGRADE);
 134         } else {
 135                 error = 0;
 136         }
 137 done:
 138         lwkt_reltoken(&vp->v_token);
 139         return error;
 140 }
 141
 142
 143 static int
 144 ffs_rawread_readahead(struct vnode *vp, caddr_t udata, off_t loffset,
 145                       size_t len, struct buf *bp)
 146 {
 147         int error;
 148         int iolen;
 149         int blockoff;
 150         int bsize;
 151         struct vnode *dp;
 152         int bforwards;
 153
 154         bsize = vp->v_mount->mnt_stat.f_iosize;
 155
 156         /*
 157          * Make sure it fits into the pbuf
 158          */
 159         iolen = (int)(intptr_t)udata & PAGE_MASK;
 160         if (len + iolen > bp->b_kvasize) {
 161                 len = bp->b_kvasize;
 162                 if (iolen != 0)
 163                         len -= PAGE_SIZE;
 164         }
 165
 166         /*
 167          * Raw disk address is in bio2, but we wait for it to
 168          * chain to bio1.
 169          */
 170         bp->b_flags &= ~B_ERROR;
 171         bp->b_loffset = loffset;
 172         bp->b_bio2.bio_offset = NOOFFSET;
 173         bp->b_bio1.bio_done = biodone_sync;
 174         bp->b_bio1.bio_flags |= BIO_SYNC;
 175
 176         blockoff = (loffset % bsize) / DEV_BSIZE;
 177
 178         error = VOP_BMAP(vp, bp->b_loffset, &bp->b_bio2.bio_offset,
 179                          &bforwards, NULL, BUF_CMD_READ);
 180         if (error != 0)
 181                 return error;
 182         dp = VTOI(vp)->i_devvp;
 183         if (bp->b_bio2.bio_offset == NOOFFSET) {
 184                 /*
 185                  * Fill holes with NULs to preserve semantics
 186                  */
 187                 if (len + blockoff * DEV_BSIZE > bsize)
 188                         len = bsize - blockoff * DEV_BSIZE;
 189
 190                 if (vmapbuf(bp, udata, len) < 0)
 191                         return EFAULT;
 192
 193                 lwkt_user_yield();
 194                 bzero(bp->b_data, bp->b_bcount);
 195
 196                 /* Mark operation completed (similar to bufdone()) */
 197
 198                 bp->b_resid = 0;
 199                 return 0;
 200         }
 201
 202         if (len + blockoff * DEV_BSIZE > bforwards)
 203                 len = bforwards - blockoff * DEV_BSIZE;
 204         bp->b_bio2.bio_offset += blockoff * DEV_BSIZE;
 205
 206         if (vmapbuf(bp, udata, len) < 0)
 207                 return EFAULT;
 208
 209         /*
 210          * Access the block device layer using the device vnode (dp) and
 211          * the translated block number (bio2) instead of the logical block
 212          * number (bio1).
 213          *
 214          * Even though we are bypassing the vnode layer, we still
 215          * want the vnode state to indicate that an I/O on its behalf
 216          * is in progress.
 217          */
 218         bp->b_cmd = BUF_CMD_READ;
 219         bio_start_transaction(&bp->b_bio1, &vp->v_track_read);
 220         vn_strategy(dp, &bp->b_bio2);
 221         return 0;
 222 }
 223
 224 static int
 225 ffs_rawread_main(struct vnode *vp, struct uio *uio)
 226 {
 227         int error, nerror;
 228         struct buf *bp, *nbp, *tbp;
 229         int iolen;
 230         caddr_t udata;
 231         int resid;
 232         off_t offset;
 233
 234         udata = uio->uio_iov->iov_base;
 235         resid = uio->uio_resid;
 236         offset = uio->uio_offset;
 237
 238         error = 0;
 239         nerror = 0;
 240
 241         bp = NULL;
 242         nbp = NULL;
 243
 244         while (resid > 0) {
 245
 246                 if (bp == NULL) { /* Setup first read */
 247                         /* XXX: Leave some bufs for swap */
 248                         bp = getpbuf_kva(&ffsrawbufcnt);
 249                         error = ffs_rawread_readahead(vp, udata, offset,
 250                                                       resid, bp);
 251                         if (error != 0)
 252                                 break;
 253
 254                         if (resid > bp->b_bufsize) { /* Setup fist readahead */
 255                                 /* XXX: Leave bufs for swap */
 256                                 if (rawreadahead != 0)
 257                                         nbp = trypbuf_kva(&ffsrawbufcnt);
 258                                 else
 259                                         nbp = NULL;
 260                                 if (nbp != NULL) {
 261                                         nerror = ffs_rawread_readahead(
 262                                                         vp,
 263                                                         udata + bp->b_bufsize,
 264                                                         offset + bp->b_bufsize,
 265                                                         resid - bp->b_bufsize,
 266                                                         nbp);
 267                                         if (nerror) {
 268                                                 relpbuf(nbp, &ffsrawbufcnt);
 269                                                 nbp = NULL;
 270                                         }
 271                                 }
 272                         }
 273                 }
 274
 275                 biowait(&bp->b_bio1, "rawrd");
 276
 277                 vunmapbuf(bp);
 278
 279                 iolen = bp->b_bcount - bp->b_resid;
 280                 if (iolen == 0 && (bp->b_flags & B_ERROR) == 0) {
 281                         nerror = 0;     /* Ignore possible beyond EOF error */
 282                         break; /* EOF */
 283                 }
 284
 285                 if ((bp->b_flags & B_ERROR) != 0) {
 286                         error = bp->b_error;
 287                         break;
 288                 }
 289                 clearbiocache(&bp->b_bio2);
 290                 resid -= iolen;
 291                 udata += iolen;
 292                 offset += iolen;
 293                 if (iolen < bp->b_bufsize) {
 294                         /* Incomplete read.  Try to read remaining part */
 295                         error = ffs_rawread_readahead(
 296                                     vp, udata, offset,
 297                                     bp->b_bufsize - iolen, bp);
 298                         if (error != 0)
 299                                 break;
 300                 } else if (nbp != NULL) { /* Complete read with readahead */
 301
 302                         tbp = bp;
 303                         bp = nbp;
 304                         nbp = tbp;
 305
 306                         clearbiocache(&nbp->b_bio2);
 307
 308                         if (resid <= bp->b_bufsize) { /* No more readaheads */
 309                                 relpbuf(nbp, &ffsrawbufcnt);
 310                                 nbp = NULL;
 311                         } else { /* Setup next readahead */
 312                                 nerror = ffs_rawread_readahead(
 313                                                 vp, udata + bp->b_bufsize,
 314                                                 offset + bp->b_bufsize,
 315                                                 resid - bp->b_bufsize,
 316                                                 nbp);
 317                                 if (nerror != 0) {
 318                                         relpbuf(nbp, &ffsrawbufcnt);
 319                                         nbp = NULL;
 320                                 }
 321                         }
 322                 } else if (nerror != 0) {/* Deferred Readahead error */
 323                         break;
 324                 }  else if (resid > 0) { /* More to read, no readahead */
 325                         error = ffs_rawread_readahead(vp, udata, offset,
 326                                                       resid, bp);
 327                         if (error != 0)
 328                                 break;
 329                 }
 330         }
 331
 332         if (bp != NULL)
 333                 relpbuf(bp, &ffsrawbufcnt);
 334         if (nbp != NULL) {                      /* Run down readahead buffer */
 335                 biowait(&nbp->b_bio1, "rawrd");
 336                 vunmapbuf(nbp);
 337                 relpbuf(nbp, &ffsrawbufcnt);
 338         }
 339
 340         if (error == 0)
 341                 error = nerror;
 342         uio->uio_iov->iov_base = udata;
 343         uio->uio_resid = resid;
 344         uio->uio_offset = offset;
 345         return error;
 346 }
 347
 348
 349 int
 350 ffs_rawread(struct vnode *vp,
 351             struct uio *uio,
 352             int *workdone)
 353 {
 354         if (allowrawread != 0 &&
 355             uio->uio_iovcnt == 1 &&
 356             uio->uio_segflg == UIO_USERSPACE &&
 357             uio->uio_resid == uio->uio_iov->iov_len &&
 358             (curthread->td_flags & TDF_DEADLKTREAT) == 0) {
 359                 int secsize;            /* Media sector size */
 360                 off_t filebytes;        /* Bytes left of file */
 361                 int blockbytes;         /* Bytes left of file in full blocks */
 362                 int partialbytes;       /* Bytes in last partial block */
 363                 int skipbytes;          /* Bytes not to read in ffs_rawread */
 364                 struct inode *ip;
 365                 int error;
 366
 367
 368                 /* Only handle sector aligned reads */
 369                 ip = VTOI(vp);
 370                 secsize = ip->i_devvp->v_rdev->si_bsize_phys;
 371                 if ((uio->uio_offset & (secsize - 1)) == 0 &&
 372                     (uio->uio_resid & (secsize - 1)) == 0) {
 373
 374                         /* Sync dirty pages and buffers if needed */
 375                         error = ffs_rawread_sync(vp);
 376                         if (error != 0)
 377                                 return error;
 378
 379                         /* Check for end of file */
 380                         if (ip->i_size > uio->uio_offset) {
 381                                 filebytes = ip->i_size - uio->uio_offset;
 382
 383                                 /* No special eof handling needed ? */
 384                                 if (uio->uio_resid <= filebytes) {
 385                                         *workdone = 1;
 386                                         return ffs_rawread_main(vp, uio);
 387                                 }
 388
 389                                 partialbytes = ((unsigned int) ip->i_size) %
 390                                         ip->i_fs->fs_bsize;
 391                                 blockbytes = (int) filebytes - partialbytes;
 392                                 if (blockbytes > 0) {
 393                                         skipbytes = uio->uio_resid -
 394                                                 blockbytes;
 395                                         uio->uio_resid = blockbytes;
 396                                         error = ffs_rawread_main(vp, uio);
 397                                         uio->uio_resid += skipbytes;
 398                                         if (error != 0)
 399                                                 return error;
 400                                         /* Read remaining part using buffer */
 401                                 }
 402                         }
 403                 }
 404         }
 405         *workdone = 0;
 406         return 0;
 407 }
 408