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