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, int *baseticks);
  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 static void ffs_rawreadwakeup(struct bio *bio);
  66
  67
  68 SYSCTL_DECL(_vfs_ffs);
  69
  70 static int ffsrawbufcnt = 4;
  71 SYSCTL_INT(_vfs_ffs, OID_AUTO, ffsrawbufcnt, CTLFLAG_RD, &ffsrawbufcnt, 0,
  72            "Buffers available for raw reads");
  73
  74 static int allowrawread = 1;
  75 SYSCTL_INT(_vfs_ffs, OID_AUTO, allowrawread, CTLFLAG_RW, &allowrawread, 0,
  76            "Flag to enable raw reads");
  77
  78 static int rawreadahead = 1;
  79 SYSCTL_INT(_vfs_ffs, OID_AUTO, rawreadahead, CTLFLAG_RW, &rawreadahead, 0,
  80            "Flag to enable readahead for long raw reads");
  81
  82
  83 void
  84 ffs_rawread_setup(void)
  85 {
  86         ffsrawbufcnt = (nswbuf > 100 ) ? (nswbuf - (nswbuf >> 4)) : nswbuf - 8;
  87 }
  88
  89
  90 static int
  91 ffs_rawread_sync(struct vnode *vp)
  92 {
  93         int error;
  94         int upgraded;
  95
  96         /* Check for dirty mmap, pending writes and dirty buffers */
  97         crit_enter();
  98         if (vp->v_track_write.bk_active > 0 ||
  99             !RB_EMPTY(&vp->v_rbdirty_tree) ||
 100             (vp->v_flag & VOBJDIRTY) != 0) {
 101                 crit_exit();
 102
 103                 if (vn_islocked(vp) != LK_EXCLUSIVE) {
 104                         upgraded = 1;
 105                         /* Upgrade to exclusive lock, this might block */
 106                         vn_lock(vp, LK_UPGRADE);
 107                 } else
 108                         upgraded = 0;
 109
 110                 /* Attempt to msync mmap() regions to clean dirty mmap */
 111                 if ((vp->v_flag & VOBJDIRTY) != 0) {
 112                         struct vm_object *obj;
 113                         if ((obj = vp->v_object) != NULL)
 114                                 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
 115                 }
 116
 117                 /* Wait for pending writes to complete */
 118                 crit_enter();
 119                 while (vp->v_track_write.bk_active) {
 120                         vp->v_track_write.bk_waitflag = 1;
 121                         error = tsleep(&vp->v_track_write, 0, "rawrdfls", 0);
 122                         if (error != 0) {
 123                                 crit_exit();
 124                                 if (upgraded != 0)
 125                                         vn_lock(vp, LK_DOWNGRADE);
 126                                 return (error);
 127                         }
 128                 }
 129                 /* Flush dirty buffers */
 130                 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
 131                         crit_exit();
 132                         if ((error = VOP_FSYNC(vp, MNT_WAIT)) != 0) {
 133                                 if (upgraded != 0)
 134                                         vn_lock(vp, LK_DOWNGRADE);
 135                                 return (error);
 136                         }
 137                         crit_enter();
 138                         if (vp->v_track_write.bk_active > 0 ||
 139                             !RB_EMPTY(&vp->v_rbdirty_tree))
 140                                 panic("ffs_rawread_sync: dirty bufs");
 141                 }
 142                 crit_exit();
 143                 if (upgraded != 0)
 144                         vn_lock(vp, LK_DOWNGRADE);
 145         } else {
 146                 crit_exit();
 147         }
 148         return 0;
 149 }
 150
 151
 152 static int
 153 ffs_rawread_readahead(struct vnode *vp, caddr_t udata, off_t loffset,
 154                       size_t len, struct buf *bp, int *baseticks)
 155 {
 156         int error;
 157         int iolen;
 158         int blockoff;
 159         int bsize;
 160         struct vnode *dp;
 161         int bforwards;
 162
 163         bsize = vp->v_mount->mnt_stat.f_iosize;
 164
 165         /*
 166          * Make sure it fits into the pbuf
 167          */
 168         iolen = (int)(intptr_t)udata & PAGE_MASK;
 169         if (len + iolen > bp->b_kvasize) {
 170                 len = bp->b_kvasize;
 171                 if (iolen != 0)
 172                         len -= PAGE_SIZE;
 173         }
 174         bp->b_flags &= ~B_ERROR;
 175         bp->b_loffset = loffset;
 176         bp->b_bio2.bio_offset = NOOFFSET;
 177         bp->b_bio2.bio_done = ffs_rawreadwakeup;
 178
 179         blockoff = (loffset % bsize) / DEV_BSIZE;
 180
 181         error = VOP_BMAP(vp, bp->b_loffset, &bp->b_bio2.bio_offset,
 182                          &bforwards, NULL, BUF_CMD_READ);
 183         if (error != 0)
 184                 return error;
 185         dp = VTOI(vp)->i_devvp;
 186         if (bp->b_bio2.bio_offset == NOOFFSET) {
 187                 /*
 188                  * Fill holes with NULs to preserve semantics
 189                  */
 190                 if (len + blockoff * DEV_BSIZE > bsize)
 191                         len = bsize - blockoff * DEV_BSIZE;
 192
 193                 if (vmapbuf(bp, udata, len) < 0)
 194                         return EFAULT;
 195
 196                 if (ticks - *baseticks >= hogticks) {
 197                         *baseticks = ticks;
 198                         uio_yield();
 199                 }
 200                 bzero(bp->b_data, bp->b_bcount);
 201
 202                 /* Mark operation completed (similar to bufdone()) */
 203
 204                 bp->b_resid = 0;
 205                 return 0;
 206         }
 207
 208         if (len + blockoff * DEV_BSIZE > bforwards)
 209                 len = bforwards - blockoff * DEV_BSIZE;
 210         bp->b_bio2.bio_offset += blockoff * DEV_BSIZE;
 211
 212         if (vmapbuf(bp, udata, len) < 0)
 213                 return EFAULT;
 214
 215         /*
 216          * Access the block device layer using the device vnode (dp) and
 217          * the translated block number (bio2) instead of the logical block
 218          * number (bio1).
 219          *
 220          * Even though we are bypassing the vnode layer, we still
 221          * want the vnode state to indicate that an I/O on its behalf
 222          * is in progress.
 223          */
 224         bp->b_cmd = BUF_CMD_READ;
 225         bio_start_transaction(&bp->b_bio1, &vp->v_track_read);
 226         vn_strategy(dp, &bp->b_bio2);
 227         return 0;
 228 }
 229
 230 static int
 231 ffs_rawread_main(struct vnode *vp, struct uio *uio)
 232 {
 233         int error, nerror;
 234         struct buf *bp, *nbp, *tbp;
 235         int iolen;
 236         int baseticks = ticks;
 237         caddr_t udata;
 238         int resid;
 239         off_t offset;
 240
 241         udata = uio->uio_iov->iov_base;
 242         resid = uio->uio_resid;
 243         offset = uio->uio_offset;
 244
 245         error = 0;
 246         nerror = 0;
 247
 248         bp = NULL;
 249         nbp = NULL;
 250
 251         while (resid > 0) {
 252
 253                 if (bp == NULL) { /* Setup first read */
 254                         /* XXX: Leave some bufs for swap */
 255                         bp = getpbuf(&ffsrawbufcnt);
 256                         error = ffs_rawread_readahead(vp, udata, offset, resid,
 257                                                       bp, &baseticks);
 258                         if (error != 0)
 259                                 break;
 260
 261                         if (resid > bp->b_bufsize) { /* Setup fist readahead */
 262                                 /* XXX: Leave bufs for swap */
 263                                 if (rawreadahead != 0)
 264                                         nbp = trypbuf(&ffsrawbufcnt);
 265                                 else
 266                                         nbp = NULL;
 267                                 if (nbp != NULL) {
 268                                         nerror = ffs_rawread_readahead(
 269                                                         vp,
 270                                                         udata + bp->b_bufsize,
 271                                                         offset + bp->b_bufsize,
 272                                                         resid - bp->b_bufsize,
 273                                                         nbp, &baseticks);
 274                                         if (nerror) {
 275                                                 relpbuf(nbp, &ffsrawbufcnt);
 276                                                 nbp = NULL;
 277                                         }
 278                                 }
 279                         }
 280                 }
 281
 282                 crit_enter();
 283                 while (bp->b_cmd != BUF_CMD_DONE)
 284                         tsleep((caddr_t)&bp->b_bio2, 0, "rawrd", 0);
 285                 crit_exit();
 286
 287                 vunmapbuf(bp);
 288
 289                 iolen = bp->b_bcount - bp->b_resid;
 290                 if (iolen == 0 && (bp->b_flags & B_ERROR) == 0) {
 291                         nerror = 0;     /* Ignore possible beyond EOF error */
 292                         break; /* EOF */
 293                 }
 294
 295                 if ((bp->b_flags & B_ERROR) != 0) {
 296                         error = bp->b_error;
 297                         break;
 298                 }
 299                 clearbiocache(&bp->b_bio2);
 300                 resid -= iolen;
 301                 udata += iolen;
 302                 offset += iolen;
 303                 if (iolen < bp->b_bufsize) {
 304                         /* Incomplete read.  Try to read remaining part */
 305                         error = ffs_rawread_readahead(
 306                                     vp, udata, offset,
 307                                     bp->b_bufsize - iolen, bp, &baseticks);
 308                         if (error != 0)
 309                                 break;
 310                 } else if (nbp != NULL) { /* Complete read with readahead */
 311
 312                         tbp = bp;
 313                         bp = nbp;
 314                         nbp = tbp;
 315
 316                         clearbiocache(&nbp->b_bio2);
 317
 318                         if (resid <= bp->b_bufsize) { /* No more readaheads */
 319                                 relpbuf(nbp, &ffsrawbufcnt);
 320                                 nbp = NULL;
 321                         } else { /* Setup next readahead */
 322                                 nerror = ffs_rawread_readahead(
 323                                                 vp, udata + bp->b_bufsize,
 324                                                 offset + bp->b_bufsize,
 325                                                 resid - bp->b_bufsize,
 326                                                 nbp, &baseticks);
 327                                 if (nerror != 0) {
 328                                         relpbuf(nbp, &ffsrawbufcnt);
 329                                         nbp = NULL;
 330                                 }
 331                         }
 332                 } else if (nerror != 0) {/* Deferred Readahead error */
 333                         break;
 334                 }  else if (resid > 0) { /* More to read, no readahead */
 335                         error = ffs_rawread_readahead(vp, udata, offset,
 336                                                       resid, bp,
 337                                                       &baseticks);
 338                         if (error != 0)
 339                                 break;
 340                 }
 341         }
 342
 343         if (bp != NULL)
 344                 relpbuf(bp, &ffsrawbufcnt);
 345         if (nbp != NULL) {                      /* Run down readahead buffer */
 346                 crit_enter();
 347                 while (nbp->b_cmd != BUF_CMD_DONE)
 348                         tsleep(&nbp->b_bio2, 0, "rawrd", 0);
 349                 crit_exit();
 350                 vunmapbuf(nbp);
 351                 relpbuf(nbp, &ffsrawbufcnt);
 352         }
 353
 354         if (error == 0)
 355                 error = nerror;
 356         uio->uio_iov->iov_base = udata;
 357         uio->uio_resid = resid;
 358         uio->uio_offset = offset;
 359         return error;
 360 }
 361
 362
 363 int
 364 ffs_rawread(struct vnode *vp,
 365             struct uio *uio,
 366             int *workdone)
 367 {
 368         if (allowrawread != 0 &&
 369             uio->uio_iovcnt == 1 &&
 370             uio->uio_segflg == UIO_USERSPACE &&
 371             uio->uio_resid == uio->uio_iov->iov_len &&
 372             (curthread->td_flags & TDF_DEADLKTREAT) == 0) {
 373                 int secsize;            /* Media sector size */
 374                 off_t filebytes;        /* Bytes left of file */
 375                 int blockbytes;         /* Bytes left of file in full blocks */
 376                 int partialbytes;       /* Bytes in last partial block */
 377                 int skipbytes;          /* Bytes not to read in ffs_rawread */
 378                 struct inode *ip;
 379                 int error;
 380
 381
 382                 /* Only handle sector aligned reads */
 383                 ip = VTOI(vp);
 384                 secsize = ip->i_devvp->v_rdev->si_bsize_phys;
 385                 if ((uio->uio_offset & (secsize - 1)) == 0 &&
 386                     (uio->uio_resid & (secsize - 1)) == 0) {
 387
 388                         /* Sync dirty pages and buffers if needed */
 389                         error = ffs_rawread_sync(vp);
 390                         if (error != 0)
 391                                 return error;
 392
 393                         /* Check for end of file */
 394                         if (ip->i_size > uio->uio_offset) {
 395                                 filebytes = ip->i_size - uio->uio_offset;
 396
 397                                 /* No special eof handling needed ? */
 398                                 if (uio->uio_resid <= filebytes) {
 399                                         *workdone = 1;
 400                                         return ffs_rawread_main(vp, uio);
 401                                 }
 402
 403                                 partialbytes = ((unsigned int) ip->i_size) %
 404                                         ip->i_fs->fs_bsize;
 405                                 blockbytes = (int) filebytes - partialbytes;
 406                                 if (blockbytes > 0) {
 407                                         skipbytes = uio->uio_resid -
 408                                                 blockbytes;
 409                                         uio->uio_resid = blockbytes;
 410                                         error = ffs_rawread_main(vp, uio);
 411                                         uio->uio_resid += skipbytes;
 412                                         if (error != 0)
 413                                                 return error;
 414                                         /* Read remaining part using buffer */
 415                                 }
 416                         }
 417                 }
 418         }
 419         *workdone = 0;
 420         return 0;
 421 }
 422
 423
 424 static void
 425 ffs_rawreadwakeup(struct bio *bio)
 426 {
 427         bio->bio_buf->b_cmd = BUF_CMD_DONE;
 428         wakeup(bio);
 429 }