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. All advertising materials mentioning features or use of this software
  14  *    must display the following acknowledgement:
  15  *      This product includes software developed by the University of
  16  *      California, Berkeley and its contributors.
  17  * 4. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  *      @(#)ufs_readwrite.c     8.11 (Berkeley) 5/8/95
  34  * $FreeBSD: src/sys/ufs/ufs/ufs_readwrite.c,v 1.65.2.14 2003/04/04 22:21:29 tegge Exp $
  35  * $DragonFly: src/sys/vfs/ufs/ufs_readwrite.c,v 1.23 2007/08/21 17:26:48 dillon Exp $
  36  */
  37
  38 #define BLKSIZE(a, b, c)        blksize(a, b, c)
  39 #define FS                      struct fs
  40 #define I_FS                    i_fs
  41
  42 #include <vm/vm.h>
  43 #include <vm/vm_object.h>
  44 #include <vm/vm_pager.h>
  45 #include <vm/vm_map.h>
  46 #include <vm/vnode_pager.h>
  47 #include <sys/event.h>
  48 #include <sys/vmmeter.h>
  49 #include <vm/vm_page2.h>
  50
  51 #include "opt_directio.h"
  52
  53 #define VN_KNOTE(vp, b) \
  54         KNOTE((struct klist *)&vp->v_pollinfo.vpi_selinfo.si_note, (b))
  55
  56 #ifdef DIRECTIO
  57 extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
  58 #endif
  59
  60 /*
  61  * Vnode op for reading.
  62  *
  63  * ffs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
  64  *          struct ucred *a_cred)
  65  */
  66 /* ARGSUSED */
  67 int
  68 ffs_read(struct vop_read_args *ap)
  69 {
  70         struct vnode *vp;
  71         struct inode *ip;
  72         struct uio *uio;
  73         FS *fs;
  74         struct buf *bp;
  75         off_t bytesinfile;
  76         int xfersize, blkoffset;
  77         int error, orig_resid;
  78         u_short mode;
  79         int seqcount;
  80         int ioflag;
  81
  82         vp = ap->a_vp;
  83         seqcount = ap->a_ioflag >> 16;
  84         ip = VTOI(vp);
  85         mode = ip->i_mode;
  86         uio = ap->a_uio;
  87         ioflag = ap->a_ioflag;
  88 #ifdef DIRECTIO
  89         if ((ioflag & IO_DIRECT) != 0) {
  90                 int workdone;
  91
  92                 error = ffs_rawread(vp, uio, &workdone);
  93                 if (error || workdone)
  94                         return error;
  95         }
  96 #endif
  97
  98 #ifdef DIAGNOSTIC
  99         if (uio->uio_rw != UIO_READ)
 100                 panic("ffs_read: mode");
 101
 102         if (vp->v_type == VLNK) {
 103                 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
 104                         panic("ffs_read: short symlink");
 105         } else if (vp->v_type != VREG && vp->v_type != VDIR)
 106                 panic("ffs_read: type %d", vp->v_type);
 107 #endif
 108         fs = ip->I_FS;
 109         if ((uint64_t)uio->uio_offset > fs->fs_maxfilesize)
 110                 return (EFBIG);
 111
 112         orig_resid = uio->uio_resid;
 113         if (orig_resid <= 0)
 114                 return (0);
 115
 116         bytesinfile = ip->i_size - uio->uio_offset;
 117         if (bytesinfile <= 0) {
 118                 if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
 119                         ip->i_flag |= IN_ACCESS;
 120                 return 0;
 121         }
 122
 123         /*
 124          * Ok so we couldn't do it all in one vm trick...
 125          * so cycle around trying smaller bites..
 126          */
 127         for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
 128                 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
 129                         break;
 130
 131                 error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL,
 132                                         &bp, seqcount);
 133                 if (error)
 134                         break;
 135
 136                 /*
 137                  * If IO_DIRECT then set B_DIRECT for the buffer.  This
 138                  * will cause us to attempt to release the buffer later on
 139                  * and will cause the buffer cache to attempt to free the
 140                  * underlying pages.
 141                  */
 142                 if (ioflag & IO_DIRECT)
 143                         bp->b_flags |= B_DIRECT;
 144
 145                 /*
 146                  * We should only get non-zero b_resid when an I/O error
 147                  * has occurred, which should cause us to break above.
 148                  * However, if the short read did not cause an error,
 149                  * then we want to ensure that we do not uiomove bad
 150                  * or uninitialized data.
 151                  *
 152                  * XXX b_resid is only valid when an actual I/O has occured
 153                  * and may be incorrect if the buffer is B_CACHE or if the
 154                  * last op on the buffer was a failed write.  This KASSERT
 155                  * is a precursor to removing it from the UFS code.
 156                  */
 157                 KASSERT(bp->b_resid == 0, ("bp->b_resid != 0"));
 158
 159                 /*
 160                  * Calculate how much data we can copy
 161                  */
 162                 blkoffset = blkoff(fs, uio->uio_offset);
 163                 xfersize = bp->b_bufsize - blkoffset;
 164                 if (xfersize > uio->uio_resid)
 165                         xfersize = uio->uio_resid;
 166                 if (xfersize > bytesinfile)
 167                         xfersize = bytesinfile;
 168                 if (xfersize <= 0) {
 169                         panic("ufs_readwrite: impossible xfersize: %d",
 170                               xfersize);
 171                 }
 172
 173                 /*
 174                  * otherwise use the general form
 175                  */
 176                 error = uiomove((char *)bp->b_data + blkoffset,
 177                                 (int)xfersize, uio);
 178
 179                 if (error)
 180                         break;
 181
 182                 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
 183                     (LIST_FIRST(&bp->b_dep) == NULL)) {
 184                         /*
 185                          * If there are no dependencies, and it's VMIO,
 186                          * then we don't need the buf, mark it available
 187                          * for freeing. The VM has the data.
 188                          */
 189                         bp->b_flags |= B_RELBUF;
 190                         brelse(bp);
 191                 } else {
 192                         /*
 193                          * Otherwise let whoever
 194                          * made the request take care of
 195                          * freeing it. We just queue
 196                          * it onto another list.
 197                          */
 198                         bqrelse(bp);
 199                 }
 200         }
 201
 202         /*
 203          * This can only happen in the case of an error
 204          * because the loop above resets bp to NULL on each iteration
 205          * and on normal completion has not set a new value into it.
 206          * so it must have come from a 'break' statement
 207          */
 208         if (bp != NULL) {
 209                 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
 210                     (LIST_FIRST(&bp->b_dep) == NULL)) {
 211                         bp->b_flags |= B_RELBUF;
 212                         brelse(bp);
 213                 } else {
 214                         bqrelse(bp);
 215                 }
 216         }
 217
 218         if ((error == 0 || uio->uio_resid != orig_resid) &&
 219             (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
 220                 ip->i_flag |= IN_ACCESS;
 221         return (error);
 222 }
 223
 224 /*
 225  * Vnode op for writing.
 226  *
 227  * ffs_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
 228  *           struct ucred *a_cred)
 229  */
 230 int
 231 ffs_write(struct vop_write_args *ap)
 232 {
 233         struct vnode *vp;
 234         struct uio *uio;
 235         struct inode *ip;
 236         FS *fs;
 237         struct buf *bp;
 238         ufs_daddr_t lbn;
 239         off_t osize;
 240         int seqcount;
 241         int blkoffset, error, extended, flags, ioflag, resid, size, xfersize;
 242         struct thread *td;
 243
 244         extended = 0;
 245         seqcount = ap->a_ioflag >> 16;
 246         ioflag = ap->a_ioflag;
 247         uio = ap->a_uio;
 248         vp = ap->a_vp;
 249         ip = VTOI(vp);
 250
 251 #ifdef DIAGNOSTIC
 252         if (uio->uio_rw != UIO_WRITE)
 253                 panic("ffs_write: mode");
 254 #endif
 255
 256         switch (vp->v_type) {
 257         case VREG:
 258                 if (ioflag & IO_APPEND)
 259                         uio->uio_offset = ip->i_size;
 260                 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
 261                         return (EPERM);
 262                 /* FALLTHROUGH */
 263         case VLNK:
 264                 break;
 265         case VDIR:
 266                 panic("ffs_write: dir write");
 267                 break;
 268         default:
 269                 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
 270                         (int)uio->uio_offset,
 271                         (int)uio->uio_resid
 272                 );
 273         }
 274
 275         fs = ip->I_FS;
 276         if (uio->uio_offset < 0 ||
 277             (uint64_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize) {
 278                 return (EFBIG);
 279         }
 280         /*
 281          * Maybe this should be above the vnode op call, but so long as
 282          * file servers have no limits, I don't think it matters.
 283          */
 284         td = uio->uio_td;
 285         if (vp->v_type == VREG && td && td->td_proc &&
 286             uio->uio_offset + uio->uio_resid >
 287             td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
 288                 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
 289                 return (EFBIG);
 290         }
 291
 292         resid = uio->uio_resid;
 293         osize = ip->i_size;
 294
 295         /*
 296          * NOTE! These B_ flags are actually balloc-only flags, not buffer
 297          * flags.  They are similar to the BA_ flags in fbsd.
 298          */
 299         if (seqcount > B_SEQMAX)
 300                 flags = B_SEQMAX << B_SEQSHIFT;
 301         else
 302                 flags = seqcount << B_SEQSHIFT;
 303         if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
 304                 flags |= B_SYNC;
 305
 306         for (error = 0; uio->uio_resid > 0;) {
 307                 lbn = lblkno(fs, uio->uio_offset);
 308                 blkoffset = blkoff(fs, uio->uio_offset);
 309                 xfersize = fs->fs_bsize - blkoffset;
 310                 if (uio->uio_resid < xfersize)
 311                         xfersize = uio->uio_resid;
 312
 313                 if (uio->uio_offset + xfersize > ip->i_size)
 314                         vnode_pager_setsize(vp, uio->uio_offset + xfersize);
 315
 316                 /*
 317                  * We must perform a read-before-write if the transfer
 318                  * size does not cover the entire buffer, or if doing
 319                  * a dummy write to flush the buffer.
 320                  */
 321                 if (xfersize < fs->fs_bsize || uio->uio_segflg == UIO_NOCOPY)
 322                         flags |= B_CLRBUF;
 323                 else
 324                         flags &= ~B_CLRBUF;
 325 /* XXX is uio->uio_offset the right thing here? */
 326                 error = VOP_BALLOC(vp, uio->uio_offset, xfersize,
 327                     ap->a_cred, flags, &bp);
 328                 if (error != 0)
 329                         break;
 330                 /*
 331                  * If the buffer is not valid and we did not clear garbage
 332                  * out above, we have to do so here even though the write
 333                  * covers the entire buffer in order to avoid a mmap()/write
 334                  * race where another process may see the garbage prior to
 335                  * the uiomove() for a write replacing it.
 336                  */
 337                 if ((bp->b_flags & B_CACHE) == 0 && (flags & B_CLRBUF) == 0)
 338                         vfs_bio_clrbuf(bp);
 339                 if (ioflag & IO_DIRECT)
 340                         bp->b_flags |= B_DIRECT;
 341                 if (ioflag & IO_NOWDRAIN)
 342                         bp->b_flags |= B_NOWDRAIN;
 343                 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
 344                         bp->b_flags |= B_NOCACHE;
 345
 346                 if (uio->uio_offset + xfersize > ip->i_size) {
 347                         ip->i_size = uio->uio_offset + xfersize;
 348                         extended = 1;
 349                 }
 350
 351                 size = BLKSIZE(fs, ip, lbn) - bp->b_resid;
 352                 if (size < xfersize)
 353                         xfersize = size;
 354
 355                 error =
 356                     uiomove((char *)bp->b_data + blkoffset, (int)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_page_count_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, 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         }
 414
 415         return (error);
 416 }
 417
 418
 419 /*
 420  * get page routine
 421  */
 422 int
 423 ffs_getpages(struct vop_getpages_args *ap)
 424 {
 425         off_t foff, physoffset;
 426         int i, size, bsize;
 427         struct vnode *dp, *vp;
 428         vm_object_t obj;
 429         vm_pindex_t pindex, firstindex;
 430         vm_page_t mreq;
 431         int bbackwards, bforwards;
 432         int pbackwards, pforwards;
 433         int firstpage;
 434         off_t reqoffset;
 435         off_t doffset;
 436         int poff;
 437         int pcount;
 438         int rtval;
 439         int pagesperblock;
 440
 441
 442         pcount = round_page(ap->a_count) / PAGE_SIZE;
 443         mreq = ap->a_m[ap->a_reqpage];
 444         firstindex = ap->a_m[0]->pindex;
 445
 446         /*
 447          * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
 448          * then the entire page is valid.  Since the page may be mapped,
 449          * user programs might reference data beyond the actual end of file
 450          * occuring within the page.  We have to zero that data.
 451          */
 452         if (mreq->valid) {
 453                 if (mreq->valid != VM_PAGE_BITS_ALL)
 454                         vm_page_zero_invalid(mreq, TRUE);
 455                 for (i = 0; i < pcount; i++) {
 456                         if (i != ap->a_reqpage) {
 457                                 vm_page_free(ap->a_m[i]);
 458                         }
 459                 }
 460                 return VM_PAGER_OK;
 461         }
 462
 463         vp = ap->a_vp;
 464         obj = vp->v_object;
 465         bsize = vp->v_mount->mnt_stat.f_iosize;
 466         pindex = mreq->pindex;
 467         foff = IDX_TO_OFF(pindex) /* + ap->a_offset should be zero */;
 468
 469         if (bsize < PAGE_SIZE)
 470                 return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
 471                                                     ap->a_count,
 472                                                     ap->a_reqpage);
 473
 474         /*
 475          * foff is the file offset of the required page
 476          * reqlblkno is the logical block that contains the page
 477          * poff is the bytes offset of the page in the logical block
 478          */
 479         poff = (int)(foff % bsize);
 480         reqoffset = foff - poff;
 481
 482         if (VOP_BMAP(vp, reqoffset, &doffset, &bforwards, &bbackwards) ||
 483             doffset == NOOFFSET
 484         ) {
 485                 for (i = 0; i < pcount; i++) {
 486                         if (i != ap->a_reqpage)
 487                                 vm_page_free(ap->a_m[i]);
 488                 }
 489                 if (doffset == NOOFFSET) {
 490                         if ((mreq->flags & PG_ZERO) == 0)
 491                                 vm_page_zero_fill(mreq);
 492                         vm_page_undirty(mreq);
 493                         mreq->valid = VM_PAGE_BITS_ALL;
 494                         return VM_PAGER_OK;
 495                 } else {
 496                         return VM_PAGER_ERROR;
 497                 }
 498         }
 499
 500         physoffset = doffset + poff;
 501         pagesperblock = bsize / PAGE_SIZE;
 502
 503         /*
 504          * find the first page that is contiguous.
 505          *
 506          * bforwards and bbackwards are the number of contiguous bytes
 507          * available before and after the block offset.  poff is the page
 508          * offset, in bytes, relative to the block offset.
 509          *
 510          * pforwards and pbackwards are the number of contiguous pages
 511          * relative to the requested page, non-inclusive of the requested
 512          * page (so a pbackwards and  pforwards of 0 indicates just the
 513          * requested page).
 514          */
 515         firstpage = 0;
 516         if (ap->a_count) {
 517                 /*
 518                  * Calculate pbackwards and clean up any requested
 519                  * pages that are too far back.
 520                  */
 521                 pbackwards = (poff + bbackwards) >> PAGE_SHIFT;
 522                 if (ap->a_reqpage > pbackwards) {
 523                         firstpage = ap->a_reqpage - pbackwards;
 524                         for (i = 0; i < firstpage; i++)
 525                                 vm_page_free(ap->a_m[i]);
 526                 }
 527
 528                 /*
 529                  * Calculate pforwards
 530                  */
 531                 pforwards = (bforwards - poff - PAGE_SIZE) >> PAGE_SHIFT;
 532                 if (pforwards < 0)
 533                         pforwards = 0;
 534                 if (pforwards < (pcount - (ap->a_reqpage + 1))) {
 535                         for(i = ap->a_reqpage + pforwards + 1; i < pcount; i++)
 536                                 vm_page_free(ap->a_m[i]);
 537                         pcount = ap->a_reqpage + pforwards + 1;
 538                 }
 539
 540                 /*
 541                  * Adjust pcount to be relative to firstpage.  All pages prior
 542                  * to firstpage in the array have been cleaned up.
 543                  */
 544                 pcount -= firstpage;
 545         }
 546
 547         /*
 548          * calculate the size of the transfer
 549          */
 550         size = pcount * PAGE_SIZE;
 551
 552         if ((IDX_TO_OFF(ap->a_m[firstpage]->pindex) + size) > vp->v_filesize) {
 553                 size = vp->v_filesize - IDX_TO_OFF(ap->a_m[firstpage]->pindex);
 554         }
 555
 556         physoffset -= foff;
 557         dp = VTOI(ap->a_vp)->i_devvp;
 558         rtval = VOP_GETPAGES(dp, &ap->a_m[firstpage], size,
 559                              (ap->a_reqpage - firstpage), physoffset);
 560
 561         return (rtval);
 562 }
 563
 564 /*
 565  * put page routine
 566  *
 567  * XXX By default, wimp out... note that a_offset is ignored (and always
 568  * XXX has been).
 569  */
 570 int
 571 ffs_putpages(struct vop_putpages_args *ap)
 572 {
 573         return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count,
 574                 ap->a_sync, ap->a_rtvals);
 575 }