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 1.26 2008/06/19 23:27:39 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 <sys/sysctl.h>
  50 #include <vm/vm_page2.h>
  51
  52 #include "opt_directio.h"
  53
  54 #define VN_KNOTE(vp, b) \
  55         KNOTE((struct klist *)&vp->v_pollinfo.vpi_selinfo.si_note, (b))
  56
  57 #ifdef DIRECTIO
  58 extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
  59 #endif
  60
  61 SYSCTL_DECL(_vfs_ffs);
  62 static int getpages_uses_bufcache = 0;
  63 SYSCTL_INT(_vfs_ffs, OID_AUTO, getpages_uses_bufcache, CTLFLAG_RW, &getpages_uses_bufcache, 0, "");
  64
  65 /*
  66  * Vnode op for reading.
  67  *
  68  * ffs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
  69  *          struct ucred *a_cred)
  70  */
  71 /* ARGSUSED */
  72 int
  73 ffs_read(struct vop_read_args *ap)
  74 {
  75         struct vnode *vp;
  76         struct inode *ip;
  77         struct uio *uio;
  78         FS *fs;
  79         struct buf *bp;
  80         off_t bytesinfile;
  81         int xfersize, blkoffset;
  82         int error, orig_resid;
  83         u_short mode;
  84         int seqcount;
  85         int ioflag;
  86
  87         vp = ap->a_vp;
  88         seqcount = ap->a_ioflag >> 16;
  89         ip = VTOI(vp);
  90         mode = ip->i_mode;
  91         uio = ap->a_uio;
  92         ioflag = ap->a_ioflag;
  93 #ifdef DIRECTIO
  94         if ((ioflag & IO_DIRECT) != 0) {
  95                 int workdone;
  96
  97                 error = ffs_rawread(vp, uio, &workdone);
  98                 if (error || workdone)
  99                         return error;
 100         }
 101 #endif
 102
 103 #ifdef DIAGNOSTIC
 104         if (uio->uio_rw != UIO_READ)
 105                 panic("ffs_read: mode");
 106
 107         if (vp->v_type == VLNK) {
 108                 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
 109                         panic("ffs_read: short symlink");
 110         } else if (vp->v_type != VREG && vp->v_type != VDIR)
 111                 panic("ffs_read: type %d", vp->v_type);
 112 #endif
 113         fs = ip->I_FS;
 114         if ((uint64_t)uio->uio_offset > fs->fs_maxfilesize)
 115                 return (EFBIG);
 116
 117         orig_resid = uio->uio_resid;
 118         if (orig_resid <= 0)
 119                 return (0);
 120
 121         bytesinfile = ip->i_size - uio->uio_offset;
 122         if (bytesinfile <= 0) {
 123                 if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
 124                         ip->i_flag |= IN_ACCESS;
 125                 return 0;
 126         }
 127
 128         /*
 129          * Ok so we couldn't do it all in one vm trick...
 130          * so cycle around trying smaller bites..
 131          */
 132         for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
 133                 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
 134                         break;
 135
 136                 error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL,
 137                                         &bp, seqcount);
 138                 if (error)
 139                         break;
 140
 141                 /*
 142                  * If IO_DIRECT then set B_DIRECT for the buffer.  This
 143                  * will cause us to attempt to release the buffer later on
 144                  * and will cause the buffer cache to attempt to free the
 145                  * underlying pages.
 146                  */
 147                 if (ioflag & IO_DIRECT)
 148                         bp->b_flags |= B_DIRECT;
 149
 150                 /*
 151                  * We should only get non-zero b_resid when an I/O error
 152                  * has occurred, which should cause us to break above.
 153                  * However, if the short read did not cause an error,
 154                  * then we want to ensure that we do not uiomove bad
 155                  * or uninitialized data.
 156                  *
 157                  * XXX b_resid is only valid when an actual I/O has occured
 158                  * and may be incorrect if the buffer is B_CACHE or if the
 159                  * last op on the buffer was a failed write.  This KASSERT
 160                  * is a precursor to removing it from the UFS code.
 161                  */
 162                 KASSERT(bp->b_resid == 0, ("bp->b_resid != 0"));
 163
 164                 /*
 165                  * Calculate how much data we can copy
 166                  */
 167                 blkoffset = blkoff(fs, uio->uio_offset);
 168                 xfersize = bp->b_bufsize - blkoffset;
 169                 if (xfersize > uio->uio_resid)
 170                         xfersize = uio->uio_resid;
 171                 if (xfersize > bytesinfile)
 172                         xfersize = bytesinfile;
 173                 if (xfersize <= 0) {
 174                         panic("ufs_readwrite: impossible xfersize: %d",
 175                               xfersize);
 176                 }
 177
 178                 /*
 179                  * otherwise use the general form
 180                  */
 181                 error = uiomove((char *)bp->b_data + blkoffset,
 182                                 (int)xfersize, uio);
 183
 184                 if (error)
 185                         break;
 186
 187                 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
 188                     (LIST_FIRST(&bp->b_dep) == NULL)) {
 189                         /*
 190                          * If there are no dependencies, and it's VMIO,
 191                          * then we don't need the buf, mark it available
 192                          * for freeing. The VM has the data.
 193                          */
 194                         bp->b_flags |= B_RELBUF;
 195                         brelse(bp);
 196                 } else {
 197                         /*
 198                          * Otherwise let whoever
 199                          * made the request take care of
 200                          * freeing it. We just queue
 201                          * it onto another list.
 202                          */
 203                         bqrelse(bp);
 204                 }
 205         }
 206
 207         /*
 208          * This can only happen in the case of an error
 209          * because the loop above resets bp to NULL on each iteration
 210          * and on normal completion has not set a new value into it.
 211          * so it must have come from a 'break' statement
 212          */
 213         if (bp != NULL) {
 214                 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
 215                     (LIST_FIRST(&bp->b_dep) == NULL)) {
 216                         bp->b_flags |= B_RELBUF;
 217                         brelse(bp);
 218                 } else {
 219                         bqrelse(bp);
 220                 }
 221         }
 222
 223         if ((error == 0 || uio->uio_resid != orig_resid) &&
 224             (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
 225                 ip->i_flag |= IN_ACCESS;
 226         return (error);
 227 }
 228
 229 /*
 230  * Vnode op for writing.
 231  *
 232  * ffs_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
 233  *           struct ucred *a_cred)
 234  */
 235 int
 236 ffs_write(struct vop_write_args *ap)
 237 {
 238         struct vnode *vp;
 239         struct uio *uio;
 240         struct inode *ip;
 241         FS *fs;
 242         struct buf *bp;
 243         ufs_daddr_t lbn;
 244         off_t osize;
 245         int seqcount;
 246         int blkoffset, error, extended, flags, ioflag, resid, size, xfersize;
 247         struct thread *td;
 248
 249         extended = 0;
 250         seqcount = ap->a_ioflag >> 16;
 251         ioflag = ap->a_ioflag;
 252         uio = ap->a_uio;
 253         vp = ap->a_vp;
 254         ip = VTOI(vp);
 255
 256 #ifdef DIAGNOSTIC
 257         if (uio->uio_rw != UIO_WRITE)
 258                 panic("ffs_write: mode");
 259 #endif
 260
 261         switch (vp->v_type) {
 262         case VREG:
 263                 if (ioflag & IO_APPEND)
 264                         uio->uio_offset = ip->i_size;
 265                 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
 266                         return (EPERM);
 267                 /* FALLTHROUGH */
 268         case VLNK:
 269                 break;
 270         case VDIR:
 271                 panic("ffs_write: dir write");
 272                 break;
 273         default:
 274                 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
 275                         (int)uio->uio_offset,
 276                         (int)uio->uio_resid
 277                 );
 278         }
 279
 280         fs = ip->I_FS;
 281         if (uio->uio_offset < 0 ||
 282             (uint64_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize) {
 283                 return (EFBIG);
 284         }
 285         /*
 286          * Maybe this should be above the vnode op call, but so long as
 287          * file servers have no limits, I don't think it matters.
 288          */
 289         td = uio->uio_td;
 290         if (vp->v_type == VREG && td && td->td_proc &&
 291             uio->uio_offset + uio->uio_resid >
 292             td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
 293                 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
 294                 return (EFBIG);
 295         }
 296
 297         resid = uio->uio_resid;
 298         osize = ip->i_size;
 299
 300         /*
 301          * NOTE! These B_ flags are actually balloc-only flags, not buffer
 302          * flags.  They are similar to the BA_ flags in fbsd.
 303          */
 304         if (seqcount > B_SEQMAX)
 305                 flags = B_SEQMAX << B_SEQSHIFT;
 306         else
 307                 flags = seqcount << B_SEQSHIFT;
 308         if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
 309                 flags |= B_SYNC;
 310
 311         for (error = 0; uio->uio_resid > 0;) {
 312                 lbn = lblkno(fs, uio->uio_offset);
 313                 blkoffset = blkoff(fs, uio->uio_offset);
 314                 xfersize = fs->fs_bsize - blkoffset;
 315                 if (uio->uio_resid < xfersize)
 316                         xfersize = uio->uio_resid;
 317
 318                 if (uio->uio_offset + xfersize > ip->i_size)
 319                         vnode_pager_setsize(vp, uio->uio_offset + xfersize);
 320
 321                 /*
 322                  * We must perform a read-before-write if the transfer
 323                  * size does not cover the entire buffer, or if doing
 324                  * a dummy write to flush the buffer.
 325                  */
 326                 if (xfersize < fs->fs_bsize || uio->uio_segflg == UIO_NOCOPY)
 327                         flags |= B_CLRBUF;
 328                 else
 329                         flags &= ~B_CLRBUF;
 330 /* XXX is uio->uio_offset the right thing here? */
 331                 error = VOP_BALLOC(vp, uio->uio_offset, xfersize,
 332                     ap->a_cred, flags, &bp);
 333                 if (error != 0)
 334                         break;
 335                 /*
 336                  * If the buffer is not valid and we did not clear garbage
 337                  * out above, we have to do so here even though the write
 338                  * covers the entire buffer in order to avoid a mmap()/write
 339                  * race where another process may see the garbage prior to
 340                  * the uiomove() for a write replacing it.
 341                  */
 342                 if ((bp->b_flags & B_CACHE) == 0 && (flags & B_CLRBUF) == 0)
 343                         vfs_bio_clrbuf(bp);
 344                 if (ioflag & IO_DIRECT)
 345                         bp->b_flags |= B_DIRECT;
 346                 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
 347                         bp->b_flags |= B_NOCACHE;
 348
 349                 if (uio->uio_offset + xfersize > ip->i_size) {
 350                         ip->i_size = uio->uio_offset + xfersize;
 351                         extended = 1;
 352                 }
 353
 354                 size = BLKSIZE(fs, ip, lbn) - bp->b_resid;
 355                 if (size < xfersize)
 356                         xfersize = size;
 357
 358                 error =
 359                     uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
 360                 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
 361                     (LIST_FIRST(&bp->b_dep) == NULL)) {
 362                         bp->b_flags |= B_RELBUF;
 363                 }
 364
 365                 /*
 366                  * If IO_SYNC each buffer is written synchronously.  Otherwise
 367                  * if we have a severe page deficiency write the buffer
 368                  * asynchronously.  Otherwise try to cluster, and if that
 369                  * doesn't do it then either do an async write (if O_DIRECT),
 370                  * or a delayed write (if not).
 371                  */
 372
 373                 if (ioflag & IO_SYNC) {
 374                         (void)bwrite(bp);
 375                 } else if (vm_page_count_severe() ||
 376                             buf_dirty_count_severe() ||
 377                             (ioflag & IO_ASYNC)) {
 378                         bp->b_flags |= B_CLUSTEROK;
 379                         bawrite(bp);
 380                 } else if (xfersize + blkoffset == fs->fs_bsize) {
 381                         if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
 382                                 bp->b_flags |= B_CLUSTEROK;
 383                                 cluster_write(bp, (off_t)ip->i_size, vp->v_mount->mnt_stat.f_iosize, seqcount);
 384                         } else {
 385                                 bawrite(bp);
 386                         }
 387                 } else if (ioflag & IO_DIRECT) {
 388                         bp->b_flags |= B_CLUSTEROK;
 389                         bawrite(bp);
 390                 } else {
 391                         bp->b_flags |= B_CLUSTEROK;
 392                         bdwrite(bp);
 393                 }
 394                 if (error || xfersize == 0)
 395                         break;
 396                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 397         }
 398         /*
 399          * If we successfully wrote any data, and we are not the superuser
 400          * we clear the setuid and setgid bits as a precaution against
 401          * tampering.
 402          */
 403         if (resid > uio->uio_resid && ap->a_cred && ap->a_cred->cr_uid != 0)
 404                 ip->i_mode &= ~(ISUID | ISGID);
 405         if (resid > uio->uio_resid)
 406                 VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0));
 407         if (error) {
 408                 if (ioflag & IO_UNIT) {
 409                         (void)ffs_truncate(vp, osize, ioflag & IO_SYNC,
 410                                            ap->a_cred);
 411                         uio->uio_offset -= resid - uio->uio_resid;
 412                         uio->uio_resid = resid;
 413                 }
 414         } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) {
 415                 error = ffs_update(vp, 1);
 416         }
 417
 418         return (error);
 419 }
 420
 421
 422 /*
 423  * get page routine
 424  */
 425 int
 426 ffs_getpages(struct vop_getpages_args *ap)
 427 {
 428         off_t foff, physoffset;
 429         int i, size, bsize;
 430         struct vnode *dp, *vp;
 431         vm_object_t obj;
 432         vm_pindex_t pindex, firstindex;
 433         vm_page_t mreq;
 434         int bbackwards, bforwards;
 435         int pbackwards, pforwards;
 436         int firstpage;
 437         off_t reqoffset;
 438         off_t doffset;
 439         int poff;
 440         int pcount;
 441         int rtval;
 442         int pagesperblock;
 443
 444         /*
 445          * If set just use the system standard getpages which issues a
 446          * UIO_NOCOPY VOP_READ.
 447          */
 448         if (getpages_uses_bufcache) {
 449                 return vop_stdgetpages(ap);
 450         }
 451
 452         pcount = round_page(ap->a_count) / PAGE_SIZE;
 453         mreq = ap->a_m[ap->a_reqpage];
 454         firstindex = ap->a_m[0]->pindex;
 455
 456         /*
 457          * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
 458          * then the entire page is valid.  Since the page may be mapped,
 459          * user programs might reference data beyond the actual end of file
 460          * occuring within the page.  We have to zero that data.
 461          */
 462         if (mreq->valid) {
 463                 if (mreq->valid != VM_PAGE_BITS_ALL)
 464                         vm_page_zero_invalid(mreq, TRUE);
 465                 for (i = 0; i < pcount; i++) {
 466                         if (i != ap->a_reqpage) {
 467                                 vm_page_free(ap->a_m[i]);
 468                         }
 469                 }
 470                 return VM_PAGER_OK;
 471         }
 472
 473         vp = ap->a_vp;
 474         obj = vp->v_object;
 475         bsize = vp->v_mount->mnt_stat.f_iosize;
 476         pindex = mreq->pindex;
 477         foff = IDX_TO_OFF(pindex) /* + ap->a_offset should be zero */;
 478
 479         if (bsize < PAGE_SIZE)
 480                 return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
 481                                                     ap->a_count,
 482                                                     ap->a_reqpage);
 483
 484         /*
 485          * foff is the file offset of the required page
 486          * reqlblkno is the logical block that contains the page
 487          * poff is the bytes offset of the page in the logical block
 488          */
 489         poff = (int)(foff % bsize);
 490         reqoffset = foff - poff;
 491
 492         if (VOP_BMAP(vp, reqoffset, &doffset, &bforwards, &bbackwards, BUF_CMD_READ) ||
 493             doffset == NOOFFSET
 494         ) {
 495                 for (i = 0; i < pcount; i++) {
 496                         if (i != ap->a_reqpage)
 497                                 vm_page_free(ap->a_m[i]);
 498                 }
 499                 if (doffset == NOOFFSET) {
 500                         if ((mreq->flags & PG_ZERO) == 0)
 501                                 vm_page_zero_fill(mreq);
 502                         vm_page_undirty(mreq);
 503                         mreq->valid = VM_PAGE_BITS_ALL;
 504                         return VM_PAGER_OK;
 505                 } else {
 506                         return VM_PAGER_ERROR;
 507                 }
 508         }
 509
 510         physoffset = doffset + poff;
 511         pagesperblock = bsize / PAGE_SIZE;
 512
 513         /*
 514          * find the first page that is contiguous.
 515          *
 516          * bforwards and bbackwards are the number of contiguous bytes
 517          * available before and after the block offset.  poff is the page
 518          * offset, in bytes, relative to the block offset.
 519          *
 520          * pforwards and pbackwards are the number of contiguous pages
 521          * relative to the requested page, non-inclusive of the requested
 522          * page (so a pbackwards and  pforwards of 0 indicates just the
 523          * requested page).
 524          */
 525         firstpage = 0;
 526         if (ap->a_count) {
 527                 /*
 528                  * Calculate pbackwards and clean up any requested
 529                  * pages that are too far back.
 530                  */
 531                 pbackwards = (poff + bbackwards) >> PAGE_SHIFT;
 532                 if (ap->a_reqpage > pbackwards) {
 533                         firstpage = ap->a_reqpage - pbackwards;
 534                         for (i = 0; i < firstpage; i++)
 535                                 vm_page_free(ap->a_m[i]);
 536                 }
 537
 538                 /*
 539                  * Calculate pforwards
 540                  */
 541                 pforwards = (bforwards - poff - PAGE_SIZE) >> PAGE_SHIFT;
 542                 if (pforwards < 0)
 543                         pforwards = 0;
 544                 if (pforwards < (pcount - (ap->a_reqpage + 1))) {
 545                         for(i = ap->a_reqpage + pforwards + 1; i < pcount; i++)
 546                                 vm_page_free(ap->a_m[i]);
 547                         pcount = ap->a_reqpage + pforwards + 1;
 548                 }
 549
 550                 /*
 551                  * Adjust pcount to be relative to firstpage.  All pages prior
 552                  * to firstpage in the array have been cleaned up.
 553                  */
 554                 pcount -= firstpage;
 555         }
 556
 557         /*
 558          * calculate the size of the transfer
 559          */
 560         size = pcount * PAGE_SIZE;
 561
 562         if ((IDX_TO_OFF(ap->a_m[firstpage]->pindex) + size) > vp->v_filesize) {
 563                 size = vp->v_filesize - IDX_TO_OFF(ap->a_m[firstpage]->pindex);
 564         }
 565
 566         physoffset -= foff;
 567         dp = VTOI(ap->a_vp)->i_devvp;
 568         rtval = VOP_GETPAGES(dp, &ap->a_m[firstpage], size,
 569                              (ap->a_reqpage - firstpage), physoffset);
 570
 571         return (rtval);
 572 }
 573