sys/ufs/ffs/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
  27 #include <sys/cdefs.h>
  28 __FBSDID("$FreeBSD$");
  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/limits.h>
  37 #include <sys/mount.h>
  38 #include <sys/namei.h>
  39 #include <sys/vnode.h>
  40 #include <sys/conf.h>
  41 #include <sys/filio.h>
  42 #include <sys/ttycom.h>
  43 #include <sys/bio.h>
  44 #include <sys/buf.h>
  45 #include <sys/rwlock.h>
  46 #include <ufs/ufs/extattr.h>
  47 #include <ufs/ufs/quota.h>
  48 #include <ufs/ufs/inode.h>
  49 #include <ufs/ufs/ufsmount.h>
  50 #include <ufs/ufs/ufs_extern.h>
  51 #include <ufs/ffs/fs.h>
  52 #include <ufs/ffs/ffs_extern.h>
  53
  54 #include <vm/vm.h>
  55 #include <vm/vm_extern.h>
  56 #include <vm/vm_object.h>
  57 #include <sys/kernel.h>
  58 #include <sys/sysctl.h>
  59
  60 static int ffs_rawread_readahead(struct vnode *vp,
  61                                  caddr_t udata,
  62                                  off_t offset,
  63                                  size_t len,
  64                                  struct thread *td,
  65                                  struct buf *bp,
  66                                  caddr_t sa);
  67 static int ffs_rawread_main(struct vnode *vp,
  68                             struct uio *uio);
  69
  70 static int ffs_rawread_sync(struct vnode *vp);
  71
  72 int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
  73
  74 void ffs_rawread_setup(void);
  75
  76 SYSCTL_DECL(_vfs_ffs);
  77
  78 static int ffsrawbufcnt = 4;
  79 SYSCTL_INT(_vfs_ffs, OID_AUTO, ffsrawbufcnt, CTLFLAG_RD, &ffsrawbufcnt, 0,
  80            "Buffers available for raw reads");
  81
  82 static int allowrawread = 1;
  83 SYSCTL_INT(_vfs_ffs, OID_AUTO, allowrawread, CTLFLAG_RW, &allowrawread, 0,
  84            "Flag to enable raw reads");
  85
  86 static int rawreadahead = 1;
  87 SYSCTL_INT(_vfs_ffs, OID_AUTO, rawreadahead, CTLFLAG_RW, &rawreadahead, 0,
  88            "Flag to enable readahead for long raw reads");
  89
  90
  91 void
  92 ffs_rawread_setup(void)
  93 {
  94         ffsrawbufcnt = (nswbuf > 100 ) ? (nswbuf - (nswbuf >> 4)) : nswbuf - 8;
  95 }
  96
  97
  98 static int
  99 ffs_rawread_sync(struct vnode *vp)
 100 {
 101         int error;
 102         int upgraded;
 103         struct bufobj *bo;
 104         struct mount *mp;
 105         vm_object_t obj;
 106
 107         /* Check for dirty mmap, pending writes and dirty buffers */
 108         bo = &vp->v_bufobj;
 109         BO_LOCK(bo);
 110         VI_LOCK(vp);
 111         if (bo->bo_numoutput > 0 ||
 112             bo->bo_dirty.bv_cnt > 0 ||
 113             ((obj = vp->v_object) != NULL &&
 114              (obj->flags & OBJ_MIGHTBEDIRTY) != 0)) {
 115                 VI_UNLOCK(vp);
 116                 BO_UNLOCK(bo);
 117
 118                 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
 119                         if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
 120                                 upgraded = 1;
 121                         else
 122                                 upgraded = 0;
 123                         VOP_UNLOCK(vp, 0);
 124                         (void) vn_start_write(vp, &mp, V_WAIT);
 125                         VOP_LOCK(vp, LK_EXCLUSIVE);
 126                 } else if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 127                         upgraded = 1;
 128                         /* Upgrade to exclusive lock, this might block */
 129                         VOP_LOCK(vp, LK_UPGRADE);
 130                 } else
 131                         upgraded = 0;
 132
 133
 134                 VI_LOCK(vp);
 135                 /* Check if vnode was reclaimed while unlocked. */
 136                 if ((vp->v_iflag & VI_DOOMED) != 0) {
 137                         VI_UNLOCK(vp);
 138                         if (upgraded != 0)
 139                                 VOP_LOCK(vp, LK_DOWNGRADE);
 140                         vn_finished_write(mp);
 141                         return (EIO);
 142                 }
 143                 /* Attempt to msync mmap() regions to clean dirty mmap */
 144                 if ((obj = vp->v_object) != NULL &&
 145                     (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
 146                         VI_UNLOCK(vp);
 147                         VM_OBJECT_WLOCK(obj);
 148                         vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
 149                         VM_OBJECT_WUNLOCK(obj);
 150                 } else
 151                         VI_UNLOCK(vp);
 152
 153                 /* Wait for pending writes to complete */
 154                 BO_LOCK(bo);
 155                 error = bufobj_wwait(&vp->v_bufobj, 0, 0);
 156                 if (error != 0) {
 157                         /* XXX: can't happen with a zero timeout ??? */
 158                         BO_UNLOCK(bo);
 159                         if (upgraded != 0)
 160                                 VOP_LOCK(vp, LK_DOWNGRADE);
 161                         vn_finished_write(mp);
 162                         return (error);
 163                 }
 164                 /* Flush dirty buffers */
 165                 if (bo->bo_dirty.bv_cnt > 0) {
 166                         BO_UNLOCK(bo);
 167                         if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) {
 168                                 if (upgraded != 0)
 169                                         VOP_LOCK(vp, LK_DOWNGRADE);
 170                                 vn_finished_write(mp);
 171                                 return (error);
 172                         }
 173                         BO_LOCK(bo);
 174                         if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
 175                                 panic("ffs_rawread_sync: dirty bufs");
 176                 }
 177                 BO_UNLOCK(bo);
 178                 if (upgraded != 0)
 179                         VOP_LOCK(vp, LK_DOWNGRADE);
 180                 vn_finished_write(mp);
 181         } else {
 182                 VI_UNLOCK(vp);
 183                 BO_UNLOCK(bo);
 184         }
 185         return 0;
 186 }
 187
 188
 189 static int
 190 ffs_rawread_readahead(struct vnode *vp,
 191                       caddr_t udata,
 192                       off_t offset,
 193                       size_t len,
 194                       struct thread *td,
 195                       struct buf *bp,
 196                       caddr_t sa)
 197 {
 198         int error;
 199         u_int iolen;
 200         off_t blockno;
 201         int blockoff;
 202         int bsize;
 203         struct vnode *dp;
 204         int bforwards;
 205         struct inode *ip;
 206         ufs2_daddr_t blkno;
 207
 208         bsize = vp->v_mount->mnt_stat.f_iosize;
 209
 210         ip = VTOI(vp);
 211         dp = ip->i_devvp;
 212
 213         iolen = ((vm_offset_t) udata) & PAGE_MASK;
 214         bp->b_bcount = len;
 215         if (bp->b_bcount + iolen > bp->b_kvasize) {
 216                 bp->b_bcount = bp->b_kvasize;
 217                 if (iolen != 0)
 218                         bp->b_bcount -= PAGE_SIZE;
 219         }
 220         bp->b_flags = 0;        /* XXX necessary ? */
 221         bp->b_iocmd = BIO_READ;
 222         bp->b_iodone = bdone;
 223         bp->b_data = udata;
 224         bp->b_saveaddr = sa;
 225         blockno = offset / bsize;
 226         blockoff = (offset % bsize) / DEV_BSIZE;
 227         if ((daddr_t) blockno != blockno) {
 228                 return EINVAL; /* blockno overflow */
 229         }
 230
 231         bp->b_lblkno = bp->b_blkno = blockno;
 232
 233         error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, NULL, &bforwards, NULL);
 234         if (error != 0)
 235                 return error;
 236         if (blkno == -1) {
 237
 238                 /* Fill holes with NULs to preserve semantics */
 239
 240                 if (bp->b_bcount + blockoff * DEV_BSIZE > bsize)
 241                         bp->b_bcount = bsize - blockoff * DEV_BSIZE;
 242                 bp->b_bufsize = bp->b_bcount;
 243
 244                 if (vmapbuf(bp, 1) < 0)
 245                         return EFAULT;
 246
 247                 maybe_yield();
 248                 bzero(bp->b_data, bp->b_bufsize);
 249
 250                 /* Mark operation completed (similar to bufdone()) */
 251
 252                 bp->b_resid = 0;
 253                 bp->b_flags |= B_DONE;
 254                 return 0;
 255         }
 256         bp->b_blkno = blkno + blockoff;
 257         bp->b_offset = bp->b_iooffset = (blkno + blockoff) * DEV_BSIZE;
 258
 259         if (bp->b_bcount + blockoff * DEV_BSIZE > bsize * (1 + bforwards))
 260                 bp->b_bcount = bsize * (1 + bforwards) - blockoff * DEV_BSIZE;
 261         bp->b_bufsize = bp->b_bcount;
 262
 263         if (vmapbuf(bp, 1) < 0)
 264                 return EFAULT;
 265
 266         BO_STRATEGY(&dp->v_bufobj, bp);
 267         return 0;
 268 }
 269
 270
 271 static int
 272 ffs_rawread_main(struct vnode *vp,
 273                  struct uio *uio)
 274 {
 275         int error, nerror;
 276         struct buf *bp, *nbp, *tbp;
 277         caddr_t sa, nsa, tsa;
 278         u_int iolen;
 279         int spl;
 280         caddr_t udata;
 281         long resid;
 282         off_t offset;
 283         struct thread *td;
 284
 285         td = uio->uio_td ? uio->uio_td : curthread;
 286         udata = uio->uio_iov->iov_base;
 287         resid = uio->uio_resid;
 288         offset = uio->uio_offset;
 289
 290         /*
 291          * keep the process from being swapped
 292          */
 293         PHOLD(td->td_proc);
 294
 295         error = 0;
 296         nerror = 0;
 297
 298         bp = NULL;
 299         nbp = NULL;
 300         sa = NULL;
 301         nsa = NULL;
 302
 303         while (resid > 0) {
 304
 305                 if (bp == NULL) { /* Setup first read */
 306                         /* XXX: Leave some bufs for swap */
 307                         bp = getpbuf(&ffsrawbufcnt);
 308                         sa = bp->b_data;
 309                         pbgetvp(vp, bp);
 310                         error = ffs_rawread_readahead(vp, udata, offset,
 311                                                      resid, td, bp, sa);
 312                         if (error != 0)
 313                                 break;
 314
 315                         if (resid > bp->b_bufsize) { /* Setup fist readahead */
 316                                 /* XXX: Leave bufs for swap */
 317                                 if (rawreadahead != 0)
 318                                         nbp = trypbuf(&ffsrawbufcnt);
 319                                 else
 320                                         nbp = NULL;
 321                                 if (nbp != NULL) {
 322                                         nsa = nbp->b_data;
 323                                         pbgetvp(vp, nbp);
 324
 325                                         nerror = ffs_rawread_readahead(vp,
 326                                                                        udata +
 327                                                                        bp->b_bufsize,
 328                                                                        offset +
 329                                                                        bp->b_bufsize,
 330                                                                        resid -
 331                                                                        bp->b_bufsize,
 332                                                                        td,
 333                                                                        nbp,
 334                                                                        nsa);
 335                                         if (nerror) {
 336                                                 pbrelvp(nbp);
 337                                                 relpbuf(nbp, &ffsrawbufcnt);
 338                                                 nbp = NULL;
 339                                         }
 340                                 }
 341                         }
 342                 }
 343
 344                 spl = splbio();
 345                 bwait(bp, PRIBIO, "rawrd");
 346                 splx(spl);
 347
 348                 vunmapbuf(bp);
 349
 350                 iolen = bp->b_bcount - bp->b_resid;
 351                 if (iolen == 0 && (bp->b_ioflags & BIO_ERROR) == 0) {
 352                         nerror = 0;     /* Ignore possible beyond EOF error */
 353                         break; /* EOF */
 354                 }
 355
 356                 if ((bp->b_ioflags & BIO_ERROR) != 0) {
 357                         error = bp->b_error;
 358                         break;
 359                 }
 360                 resid -= iolen;
 361                 udata += iolen;
 362                 offset += iolen;
 363                 if (iolen < bp->b_bufsize) {
 364                         /* Incomplete read.  Try to read remaining part */
 365                         error = ffs_rawread_readahead(vp,
 366                                                       udata,
 367                                                       offset,
 368                                                       bp->b_bufsize - iolen,
 369                                                       td,
 370                                                       bp,
 371                                                       sa);
 372                         if (error != 0)
 373                                 break;
 374                 } else if (nbp != NULL) { /* Complete read with readahead */
 375
 376                         tbp = bp;
 377                         bp = nbp;
 378                         nbp = tbp;
 379
 380                         tsa = sa;
 381                         sa = nsa;
 382                         nsa = tsa;
 383
 384                         if (resid <= bp->b_bufsize) { /* No more readaheads */
 385                                 pbrelvp(nbp);
 386                                 relpbuf(nbp, &ffsrawbufcnt);
 387                                 nbp = NULL;
 388                         } else { /* Setup next readahead */
 389                                 nerror = ffs_rawread_readahead(vp,
 390                                                                udata +
 391                                                                bp->b_bufsize,
 392                                                                offset +
 393                                                                bp->b_bufsize,
 394                                                                resid -
 395                                                                bp->b_bufsize,
 396                                                                td,
 397                                                                nbp,
 398                                                                nsa);
 399                                 if (nerror != 0) {
 400                                         pbrelvp(nbp);
 401                                         relpbuf(nbp, &ffsrawbufcnt);
 402                                         nbp = NULL;
 403                                 }
 404                         }
 405                 } else if (nerror != 0) {/* Deferred Readahead error */
 406                         break;
 407                 }  else if (resid > 0) { /* More to read, no readahead */
 408                         error = ffs_rawread_readahead(vp, udata, offset,
 409                                                       resid, td, bp, sa);
 410                         if (error != 0)
 411                                 break;
 412                 }
 413         }
 414
 415         if (bp != NULL) {
 416                 pbrelvp(bp);
 417                 relpbuf(bp, &ffsrawbufcnt);
 418         }
 419         if (nbp != NULL) {                      /* Run down readahead buffer */
 420                 spl = splbio();
 421                 bwait(nbp, PRIBIO, "rawrd");
 422                 splx(spl);
 423                 vunmapbuf(nbp);
 424                 pbrelvp(nbp);
 425                 relpbuf(nbp, &ffsrawbufcnt);
 426         }
 427
 428         if (error == 0)
 429                 error = nerror;
 430         PRELE(td->td_proc);
 431         uio->uio_iov->iov_base = udata;
 432         uio->uio_resid = resid;
 433         uio->uio_offset = offset;
 434         return error;
 435 }
 436
 437
 438 int
 439 ffs_rawread(struct vnode *vp,
 440             struct uio *uio,
 441             int *workdone)
 442 {
 443         if (allowrawread != 0 &&
 444             uio->uio_iovcnt == 1 &&
 445             uio->uio_segflg == UIO_USERSPACE &&
 446             uio->uio_resid == uio->uio_iov->iov_len &&
 447             (((uio->uio_td != NULL) ? uio->uio_td : curthread)->td_pflags &
 448              TDP_DEADLKTREAT) == 0) {
 449                 int secsize;            /* Media sector size */
 450                 off_t filebytes;        /* Bytes left of file */
 451                 int blockbytes;         /* Bytes left of file in full blocks */
 452                 int partialbytes;       /* Bytes in last partial block */
 453                 int skipbytes;          /* Bytes not to read in ffs_rawread */
 454                 struct inode *ip;
 455                 int error;
 456
 457
 458                 /* Only handle sector aligned reads */
 459                 ip = VTOI(vp);
 460                 secsize = ip->i_devvp->v_bufobj.bo_bsize;
 461                 if ((uio->uio_offset & (secsize - 1)) == 0 &&
 462                     (uio->uio_resid & (secsize - 1)) == 0) {
 463
 464                         /* Sync dirty pages and buffers if needed */
 465                         error = ffs_rawread_sync(vp);
 466                         if (error != 0)
 467                                 return error;
 468
 469                         /* Check for end of file */
 470                         if (ip->i_size > uio->uio_offset) {
 471                                 filebytes = ip->i_size - uio->uio_offset;
 472
 473                                 /* No special eof handling needed ? */
 474                                 if (uio->uio_resid <= filebytes) {
 475                                         *workdone = 1;
 476                                         return ffs_rawread_main(vp, uio);
 477                                 }
 478
 479                                 partialbytes = ((unsigned int) ip->i_size) %
 480                                         ip->i_fs->fs_bsize;
 481                                 blockbytes = (int) filebytes - partialbytes;
 482                                 if (blockbytes > 0) {
 483                                         skipbytes = uio->uio_resid -
 484                                                 blockbytes;
 485                                         uio->uio_resid = blockbytes;
 486                                         error = ffs_rawread_main(vp, uio);
 487                                         uio->uio_resid += skipbytes;
 488                                         if (error != 0)
 489                                                 return error;
 490                                         /* Read remaining part using buffer */
 491                                 }
 492                         }
 493                 }
 494         }
 495         *workdone = 0;
 496         return 0;
 497 }