sys/vfs/ufs/ufs_readwrite.c

   1 /*-
   2  * Copyright (c) 1993
   3  *      The Regents of the University of California.  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  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)ufs_readwrite.c     8.11 (Berkeley) 5/8/95
  30  * $FreeBSD: src/sys/ufs/ufs/ufs_readwrite.c,v 1.65.2.14 2003/04/04 22:21:29 tegge Exp $
  31  */
  32
  33 #define BLKSIZE(a, b, c)        blksize(a, b, c)
  34 #define FS                      struct fs
  35 #define I_FS                    i_fs
  36
  37 #include <vm/vm.h>
  38 #include <vm/vm_object.h>
  39 #include <vm/vm_pager.h>
  40 #include <vm/vm_map.h>
  41 #include <vm/vnode_pager.h>
  42 #include <sys/event.h>
  43 #include <sys/vmmeter.h>
  44 #include <sys/sysctl.h>
  45 #include <vm/vm_page2.h>
  46
  47 #define VN_KNOTE(vp, b) KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, (b))
  48
  49 SYSCTL_DECL(_vfs_ffs);
  50
  51 /*
  52  * Vnode op for reading.
  53  *
  54  * ffs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
  55  *          struct ucred *a_cred)
  56  */
  57 /* ARGSUSED */
  58 int
  59 ffs_read(struct vop_read_args *ap)
  60 {
  61         struct vnode *vp;
  62         struct inode *ip;
  63         struct uio *uio;
  64         FS *fs;
  65         struct buf *bp;
  66         off_t bytesinfile;
  67         int xfersize, blkoffset;
  68         int error, orig_resid;
  69         int seqcount;
  70         int ioflag;
  71
  72         vp = ap->a_vp;
  73         seqcount = ap->a_ioflag >> IO_SEQSHIFT;
  74         ip = VTOI(vp);
  75         uio = ap->a_uio;
  76         ioflag = ap->a_ioflag;
  77 #ifdef DIAGNOSTIC
  78         if (uio->uio_rw != UIO_READ)
  79                 panic("ffs_read: mode");
  80
  81         if (vp->v_type == VLNK) {
  82                 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
  83                         panic("ffs_read: short symlink");
  84         } else if (vp->v_type != VREG && vp->v_type != VDIR)
  85                 panic("ffs_read: type %d", vp->v_type);
  86 #endif
  87         fs = ip->I_FS;
  88         if ((uint64_t)uio->uio_offset > fs->fs_maxfilesize)
  89                 return (EFBIG);
  90
  91         orig_resid = uio->uio_resid;
  92         if (orig_resid <= 0)
  93                 return (0);
  94
  95         bytesinfile = ip->i_size - uio->uio_offset;
  96         if (bytesinfile <= 0) {
  97                 if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
  98                         ip->i_flag |= IN_ACCESS;
  99                 return 0;
 100         }
 101
 102         /*
 103          * Ok so we couldn't do it all in one vm trick...
 104          * so cycle around trying smaller bites..
 105          */
 106         for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
 107                 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
 108                         break;
 109
 110                 error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL,
 111                                         &bp, seqcount);
 112                 if (error)
 113                         break;
 114
 115                 /*
 116                  * If IO_DIRECT then set B_DIRECT for the buffer.  This
 117                  * will cause us to attempt to release the buffer later on
 118                  * and will cause the buffer cache to attempt to free the
 119                  * underlying pages.
 120                  */
 121                 if (ioflag & IO_DIRECT)
 122                         bp->b_flags |= B_DIRECT;
 123
 124                 /*
 125                  * We should only get non-zero b_resid when an I/O error
 126                  * has occurred, which should cause us to break above.
 127                  * However, if the short read did not cause an error,
 128                  * then we want to ensure that we do not uiomove bad
 129                  * or uninitialized data.
 130                  *
 131                  * XXX b_resid is only valid when an actual I/O has occured
 132                  * and may be incorrect if the buffer is B_CACHE or if the
 133                  * last op on the buffer was a failed write.  This KASSERT
 134                  * is a precursor to removing it from the UFS code.
 135                  */
 136                 KASSERT(bp->b_resid == 0, ("bp->b_resid != 0"));
 137
 138                 /*
 139                  * Calculate how much data we can copy
 140                  */
 141                 blkoffset = blkoff(fs, uio->uio_offset);
 142                 xfersize = bp->b_bufsize - blkoffset;
 143                 if (xfersize > uio->uio_resid)
 144                         xfersize = uio->uio_resid;
 145                 if (xfersize > bytesinfile)
 146                         xfersize = bytesinfile;
 147                 if (xfersize <= 0) {
 148                         panic("ufs_readwrite: impossible xfersize: %d",
 149                               xfersize);
 150                 }
 151
 152                 /*
 153                  * otherwise use the general form
 154                  */
 155                 error = uiomovebp(bp, bp->b_data + blkoffset, xfersize, uio);
 156
 157                 if (error)
 158                         break;
 159
 160                 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
 161                     (LIST_FIRST(&bp->b_dep) == NULL)) {
 162                         /*
 163                          * If there are no dependencies, and it's VMIO,
 164                          * then we don't need the buf, mark it available
 165                          * for freeing. The VM has the data.
 166                          */
 167                         bp->b_flags |= B_RELBUF;
 168                         brelse(bp);
 169                 } else {
 170                         /*
 171                          * Otherwise let whoever
 172                          * made the request take care of
 173                          * freeing it. We just queue
 174                          * it onto another list.
 175                          */
 176                         bqrelse(bp);
 177                 }
 178         }
 179
 180         /*
 181          * This can only happen in the case of an error
 182          * because the loop above resets bp to NULL on each iteration
 183          * and on normal completion has not set a new value into it.
 184          * so it must have come from a 'break' statement
 185          */
 186         if (bp != NULL) {
 187                 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
 188                     (LIST_FIRST(&bp->b_dep) == NULL)) {
 189                         bp->b_flags |= B_RELBUF;
 190                         brelse(bp);
 191                 } else {
 192                         bqrelse(bp);
 193                 }
 194         }
 195
 196         if ((error == 0 || uio->uio_resid != orig_resid) &&
 197             (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
 198                 ip->i_flag |= IN_ACCESS;
 199         return (error);
 200 }
 201
 202 /*
 203  * Vnode op for writing.
 204  *
 205  * ffs_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
 206  *           struct ucred *a_cred)
 207  */
 208 int
 209 ffs_write(struct vop_write_args *ap)
 210 {
 211         struct vnode *vp;
 212         struct uio *uio;
 213         struct inode *ip;
 214         FS *fs;
 215         struct buf *bp;
 216         ufs_daddr_t lbn;
 217         off_t osize;
 218         off_t nsize;
 219         int seqcount;
 220         int blkoffset, error, extended, flags, ioflag, resid, size, xfersize;
 221         struct thread *td;
 222
 223         extended = 0;
 224         seqcount = ap->a_ioflag >> IO_SEQSHIFT;
 225         ioflag = ap->a_ioflag;
 226         uio = ap->a_uio;
 227         vp = ap->a_vp;
 228         ip = VTOI(vp);
 229
 230 #ifdef DIAGNOSTIC
 231         if (uio->uio_rw != UIO_WRITE)
 232                 panic("ffs_write: mode");
 233 #endif
 234
 235         switch (vp->v_type) {
 236         case VREG:
 237                 if (ioflag & IO_APPEND)
 238                         uio->uio_offset = ip->i_size;
 239                 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
 240                         return (EPERM);
 241                 /* FALLTHROUGH */
 242         case VLNK:
 243                 break;
 244         case VDIR:
 245                 panic("ffs_write: dir write");
 246                 break;
 247         default:
 248                 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
 249                         (int)uio->uio_offset,
 250                         (int)uio->uio_resid
 251                 );
 252         }
 253
 254         fs = ip->I_FS;
 255         if (uio->uio_offset < 0 ||
 256             (uint64_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize) {
 257                 return (EFBIG);
 258         }
 259         /*
 260          * Maybe this should be above the vnode op call, but so long as
 261          * file servers have no limits, I don't think it matters.
 262          */
 263         td = uio->uio_td;
 264         if (vp->v_type == VREG && td && td->td_proc &&
 265             uio->uio_offset + uio->uio_resid >
 266             td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
 267                 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
 268                 return (EFBIG);
 269         }
 270
 271         resid = uio->uio_resid;
 272         osize = ip->i_size;
 273
 274         /*
 275          * NOTE! These B_ flags are actually balloc-only flags, not buffer
 276          * flags.  They are similar to the BA_ flags in fbsd.
 277          */
 278         if (seqcount > B_SEQMAX)
 279                 flags = B_SEQMAX << B_SEQSHIFT;
 280         else
 281                 flags = seqcount << B_SEQSHIFT;
 282         if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
 283                 flags |= B_SYNC;
 284
 285         if (uio->uio_segflg == UIO_NOCOPY)
 286                 ip->i_flag |= IN_NOCOPYWRITE;
 287         else
 288                 vclrflags(vp, VLASTWRITETS);
 289
 290         for (error = 0; uio->uio_resid > 0;) {
 291                 lbn = lblkno(fs, uio->uio_offset);
 292                 blkoffset = blkoff(fs, uio->uio_offset);
 293                 xfersize = fs->fs_bsize - blkoffset;
 294                 if (uio->uio_resid < xfersize)
 295                         xfersize = uio->uio_resid;
 296
 297                 if (uio->uio_offset + xfersize > ip->i_size) {
 298                         nsize = uio->uio_offset + xfersize;
 299                         nvnode_pager_setsize(vp, nsize,
 300                                 blkoffresize(fs, nsize), blkoff(fs, nsize));
 301                 }
 302
 303 #if 0
 304                 /*
 305                  * If doing a dummy write to flush the buffer for a
 306                  * putpages we must perform a read-before-write to
 307                  * fill in any missing spots and clear any invalid
 308                  * areas.  Otherwise a multi-page buffer may not properly
 309                  * flush.
 310                  *
 311                  * We must clear any invalid areas
 312                  */
 313                 if (uio->uio_segflg == UIO_NOCOPY) {
 314                         error = ffs_blkatoff(vp, uio->uio_offset, NULL, &bp);
 315                         if (error)
 316                                 break;
 317                         bqrelse(bp);
 318                 }
 319 #endif
 320
 321                 /*
 322                  * We must clear invalid areas.
 323                  */
 324                 if (xfersize < fs->fs_bsize || uio->uio_segflg == UIO_NOCOPY)
 325                         flags |= B_CLRBUF;
 326                 else
 327                         flags &= ~B_CLRBUF;
 328 /* XXX is uio->uio_offset the right thing here? */
 329                 error = VOP_BALLOC(vp, uio->uio_offset, xfersize,
 330                                    ap->a_cred, flags, &bp);
 331                 if (error != 0)
 332                         break;
 333                 /*
 334                  * If the buffer is not valid and we did not clear garbage
 335                  * out above, we have to do so here even though the write
 336                  * covers the entire buffer in order to avoid a mmap()/write
 337                  * race where another process may see the garbage prior to
 338                  * the uiomove() for a write replacing it.
 339                  */
 340                 if ((bp->b_flags & B_CACHE) == 0 && (flags & B_CLRBUF) == 0)
 341                         vfs_bio_clrbuf(bp);
 342                 if (ioflag & IO_DIRECT)
 343                         bp->b_flags |= B_DIRECT;
 344                 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
 345                         bp->b_flags |= B_NOCACHE;
 346
 347                 if (uio->uio_offset + xfersize > ip->i_size) {
 348                         ip->i_size = uio->uio_offset + xfersize;
 349                         extended = 1;
 350                 }
 351
 352                 size = BLKSIZE(fs, ip, lbn) - bp->b_resid;
 353                 if (size < xfersize)
 354                         xfersize = size;
 355
 356                 error = uiomovebp(bp, bp->b_data + blkoffset, xfersize, uio);
 357                 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
 358                     (LIST_FIRST(&bp->b_dep) == NULL)) {
 359                         bp->b_flags |= B_RELBUF;
 360                 }
 361
 362                 /*
 363                  * If IO_SYNC each buffer is written synchronously.  Otherwise
 364                  * if we have a severe page deficiency write the buffer
 365                  * asynchronously.  Otherwise try to cluster, and if that
 366                  * doesn't do it then either do an async write (if O_DIRECT),
 367                  * or a delayed write (if not).
 368                  */
 369
 370                 if (ioflag & IO_SYNC) {
 371                         (void)bwrite(bp);
 372                 } else if (vm_paging_severe() ||
 373                             buf_dirty_count_severe() ||
 374                             (ioflag & IO_ASYNC)) {
 375                         bp->b_flags |= B_CLUSTEROK;
 376                         bawrite(bp);
 377                 } else if (xfersize + blkoffset == fs->fs_bsize) {
 378                         if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
 379                                 bp->b_flags |= B_CLUSTEROK;
 380                                 cluster_write(bp, (off_t)ip->i_size, fs->fs_bsize, seqcount);
 381                         } else {
 382                                 bawrite(bp);
 383                         }
 384                 } else if (ioflag & IO_DIRECT) {
 385                         bp->b_flags |= B_CLUSTEROK;
 386                         bawrite(bp);
 387                 } else {
 388                         bp->b_flags |= B_CLUSTEROK;
 389                         bdwrite(bp);
 390                 }
 391                 if (error || xfersize == 0)
 392                         break;
 393                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 394         }
 395         /*
 396          * If we successfully wrote any data, and we are not the superuser
 397          * we clear the setuid and setgid bits as a precaution against
 398          * tampering.
 399          */
 400         if (resid > uio->uio_resid && ap->a_cred && ap->a_cred->cr_uid != 0)
 401                 ip->i_mode &= ~(ISUID | ISGID);
 402         if (resid > uio->uio_resid)
 403                 VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0));
 404         if (error) {
 405                 if (ioflag & IO_UNIT) {
 406                         (void)ffs_truncate(vp, osize, ioflag & IO_SYNC,
 407                                            ap->a_cred);
 408                         uio->uio_offset -= resid - uio->uio_resid;
 409                         uio->uio_resid = resid;
 410                 }
 411         } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) {
 412                 error = ffs_update(vp, 1);
 413         } else {
 414                 ufs_itimes(vp);
 415         }
 416         if (uio->uio_segflg == UIO_NOCOPY)
 417                 ip->i_flag &= ~IN_NOCOPYWRITE;
 418
 419         return (error);
 420 }
 421