The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/vm/vnode_pager.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*
    2  * Copyright (c) 1990 University of Utah.
    3  * Copyright (c) 1991 The Regents of the University of California.
    4  * All rights reserved.
    5  * Copyright (c) 1993, 1994 John S. Dyson
    6  * Copyright (c) 1995, David Greenman
    7  *
    8  * This code is derived from software contributed to Berkeley by
    9  * the Systems Programming Group of the University of Utah Computer
   10  * Science Department.
   11  *
   12  * Redistribution and use in source and binary forms, with or without
   13  * modification, are permitted provided that the following conditions
   14  * are met:
   15  * 1. Redistributions of source code must retain the above copyright
   16  *    notice, this list of conditions and the following disclaimer.
   17  * 2. Redistributions in binary form must reproduce the above copyright
   18  *    notice, this list of conditions and the following disclaimer in the
   19  *    documentation and/or other materials provided with the distribution.
   20  * 3. All advertising materials mentioning features or use of this software
   21  *    must display the following acknowledgement:
   22  *      This product includes software developed by the University of
   23  *      California, Berkeley and its contributors.
   24  * 4. Neither the name of the University nor the names of its contributors
   25  *    may be used to endorse or promote products derived from this software
   26  *    without specific prior written permission.
   27  *
   28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   38  * SUCH DAMAGE.
   39  *
   40  *      from: @(#)vnode_pager.c 7.5 (Berkeley) 4/20/91
   41  * $FreeBSD$
   42  */
   43 
   44 /*
   45  * Page to/from files (vnodes).
   46  */
   47 
   48 /*
   49  * TODO:
   50  *      Implement VOP_GETPAGES/PUTPAGES interface for filesystems. Will
   51  *      greatly re-simplify the vnode_pager.
   52  */
   53 
   54 #include <sys/param.h>
   55 #include <sys/systm.h>
   56 #include <sys/proc.h>
   57 #include <sys/vnode.h>
   58 #include <sys/mount.h>
   59 #include <sys/buf.h>
   60 #include <sys/vmmeter.h>
   61 #include <sys/conf.h>
   62 
   63 #include <vm/vm.h>
   64 #include <vm/vm_object.h>
   65 #include <vm/vm_page.h>
   66 #include <vm/vm_pager.h>
   67 #include <vm/vm_map.h>
   68 #include <vm/vnode_pager.h>
   69 #include <vm/vm_extern.h>
   70 
   71 static vm_offset_t vnode_pager_addr __P((struct vnode *vp, vm_ooffset_t address,
   72                                          int *run));
   73 static void vnode_pager_iodone __P((struct buf *bp));
   74 static int vnode_pager_input_smlfs __P((vm_object_t object, vm_page_t m));
   75 static int vnode_pager_input_old __P((vm_object_t object, vm_page_t m));
   76 static void vnode_pager_dealloc __P((vm_object_t));
   77 static int vnode_pager_getpages __P((vm_object_t, vm_page_t *, int, int));
   78 static void vnode_pager_putpages __P((vm_object_t, vm_page_t *, int, boolean_t, int *));
   79 static boolean_t vnode_pager_haspage __P((vm_object_t, vm_pindex_t, int *, int *));
   80 
   81 struct pagerops vnodepagerops = {
   82         NULL,
   83         vnode_pager_alloc,
   84         vnode_pager_dealloc,
   85         vnode_pager_getpages,
   86         vnode_pager_putpages,
   87         vnode_pager_haspage,
   88         NULL
   89 };
   90 
   91 int vnode_pbuf_freecnt = -1;    /* start out unlimited */
   92 
   93 /*
   94  * Allocate (or lookup) pager for a vnode.
   95  * Handle is a vnode pointer.
   96  */
   97 vm_object_t
   98 vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
   99                   vm_ooffset_t offset)
  100 {
  101         vm_object_t object;
  102         struct vnode *vp;
  103 
  104         /*
  105          * Pageout to vnode, no can do yet.
  106          */
  107         if (handle == NULL)
  108                 return (NULL);
  109 
  110         /*
  111          * XXX hack - This initialization should be put somewhere else.
  112          */
  113         if (vnode_pbuf_freecnt < 0) {
  114             vnode_pbuf_freecnt = nswbuf / 2 + 1;
  115         }
  116 
  117         vp = (struct vnode *) handle;
  118 
  119         /*
  120          * Prevent race condition when allocating the object. This
  121          * can happen with NFS vnodes since the nfsnode isn't locked.
  122          */
  123         while (vp->v_flag & VOLOCK) {
  124                 vp->v_flag |= VOWANT;
  125                 tsleep(vp, PVM, "vnpobj", 0);
  126         }
  127         vp->v_flag |= VOLOCK;
  128 
  129         /*
  130          * If the object is being terminated, wait for it to
  131          * go away.
  132          */
  133         while (((object = vp->v_object) != NULL) &&
  134                 (object->flags & OBJ_DEAD)) {
  135                 tsleep(object, PVM, "vadead", 0);
  136         }
  137 
  138         if (vp->v_usecount == 0)
  139                 panic("vnode_pager_alloc: no vnode reference");
  140 
  141         if (object == NULL) {
  142                 /*
  143                  * And an object of the appropriate size
  144                  */
  145                 object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size)));
  146                 object->flags = 0;
  147 
  148                 object->un_pager.vnp.vnp_size = size;
  149 
  150                 object->handle = handle;
  151                 vp->v_object = object;
  152                 vp->v_usecount++;
  153         } else {
  154                 object->ref_count++;
  155                 vp->v_usecount++;
  156         }
  157 
  158         vp->v_flag &= ~VOLOCK;
  159         if (vp->v_flag & VOWANT) {
  160                 vp->v_flag &= ~VOWANT;
  161                 wakeup(vp);
  162         }
  163         return (object);
  164 }
  165 
  166 static void
  167 vnode_pager_dealloc(object)
  168         vm_object_t object;
  169 {
  170         register struct vnode *vp = object->handle;
  171 
  172         if (vp == NULL)
  173                 panic("vnode_pager_dealloc: pager already dealloced");
  174 
  175         vm_object_pip_wait(object, "vnpdea");
  176 
  177         object->handle = NULL;
  178         object->type = OBJT_DEAD;
  179         vp->v_object = NULL;
  180         vp->v_flag &= ~(VTEXT | VOBJBUF);
  181 }
  182 
  183 static boolean_t
  184 vnode_pager_haspage(object, pindex, before, after)
  185         vm_object_t object;
  186         vm_pindex_t pindex;
  187         int *before;
  188         int *after;
  189 {
  190         struct vnode *vp = object->handle;
  191         daddr_t bn;
  192         int err;
  193         daddr_t reqblock;
  194         int poff;
  195         int bsize;
  196         int pagesperblock, blocksperpage;
  197 
  198         /*
  199          * If no vp or vp is doomed or marked transparent to VM, we do not
  200          * have the page.
  201          */
  202         if ((vp == NULL) || (vp->v_flag & VDOOMED))
  203                 return FALSE;
  204 
  205         /*
  206          * If filesystem no longer mounted or offset beyond end of file we do
  207          * not have the page.
  208          */
  209         if ((vp->v_mount == NULL) ||
  210                 (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size))
  211                 return FALSE;
  212 
  213         bsize = vp->v_mount->mnt_stat.f_iosize;
  214         pagesperblock = bsize / PAGE_SIZE;
  215         blocksperpage = 0;
  216         if (pagesperblock > 0) {
  217                 reqblock = pindex / pagesperblock;
  218         } else {
  219                 blocksperpage = (PAGE_SIZE / bsize);
  220                 reqblock = pindex * blocksperpage;
  221         }
  222         err = VOP_BMAP(vp, reqblock, (struct vnode **) 0, &bn,
  223                 after, before);
  224         if (err)
  225                 return TRUE;
  226         if ( bn == -1)
  227                 return FALSE;
  228         if (pagesperblock > 0) {
  229                 poff = pindex - (reqblock * pagesperblock);
  230                 if (before) {
  231                         *before *= pagesperblock;
  232                         *before += poff;
  233                 }
  234                 if (after) {
  235                         int numafter;
  236                         *after *= pagesperblock;
  237                         numafter = pagesperblock - (poff + 1);
  238                         if (IDX_TO_OFF(pindex + numafter) > object->un_pager.vnp.vnp_size) {
  239                                 numafter = OFF_TO_IDX((object->un_pager.vnp.vnp_size - IDX_TO_OFF(pindex)));
  240                         }
  241                         *after += numafter;
  242                 }
  243         } else {
  244                 if (before) {
  245                         *before /= blocksperpage;
  246                 }
  247 
  248                 if (after) {
  249                         *after /= blocksperpage;
  250                 }
  251         }
  252         return TRUE;
  253 }
  254 
  255 /*
  256  * Lets the VM system know about a change in size for a file.
  257  * We adjust our own internal size and flush any cached pages in
  258  * the associated object that are affected by the size change.
  259  *
  260  * Note: this routine may be invoked as a result of a pager put
  261  * operation (possibly at object termination time), so we must be careful.
  262  */
  263 void
  264 vnode_pager_setsize(vp, nsize)
  265         struct vnode *vp;
  266         vm_ooffset_t nsize;
  267 {
  268         vm_pindex_t nobjsize;
  269         vm_object_t object = vp->v_object;
  270 
  271         if (object == NULL)
  272                 return;
  273 
  274         /*
  275          * Hasn't changed size
  276          */
  277         if (nsize == object->un_pager.vnp.vnp_size)
  278                 return;
  279 
  280         nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
  281 
  282         /*
  283          * File has shrunk. Toss any cached pages beyond the new EOF.
  284          */
  285         if (nsize < object->un_pager.vnp.vnp_size) {
  286                 vm_freeze_copyopts(object, OFF_TO_IDX(nsize), object->size);
  287                 if (nobjsize < object->size) {
  288                         vm_object_page_remove(object, nobjsize, object->size,
  289                                 FALSE);
  290                 }
  291                 /*
  292                  * this gets rid of garbage at the end of a page that is now
  293                  * only partially backed by the vnode.
  294                  *
  295                  * XXX for some reason (I don't know yet), if we take a
  296                  * completely invalid page and mark it partially valid
  297                  * it can screw up NFS reads, so we don't allow the case.
  298                  */
  299                 if (nsize & PAGE_MASK) {
  300                         vm_offset_t kva;
  301                         vm_page_t m;
  302 
  303                         m = vm_page_lookup(object, OFF_TO_IDX(nsize));
  304                         if (m && m->valid) {
  305                                 int base = (int)nsize & PAGE_MASK;
  306                                 int size = PAGE_SIZE - base;
  307 
  308                                 /*
  309                                  * Clear out partial-page garbage in case
  310                                  * the page has been mapped.
  311                                  */
  312                                 kva = vm_pager_map_page(m);
  313                                 bzero((caddr_t)kva + base, size);
  314                                 vm_pager_unmap_page(kva);
  315 
  316                                 /*
  317                                  * XXX work around SMP data integrity race
  318                                  * by unmapping the page from user processes.
  319                                  * The garbage we just cleared may be mapped
  320                                  * to a user process running on another cpu
  321                                  * and this code is not running through normal
  322                                  * I/O channels which handle SMP issues for
  323                                  * us, so unmap page to synchronize all cpus.
  324                                  *
  325                                  * XXX should vm_pager_unmap_page() have
  326                                  * dealt with this?
  327                                  */
  328                                 vm_page_protect(m, VM_PROT_NONE);
  329 
  330                                 /*
  331                                  * Clear out partial-page dirty bits.  This
  332                                  * has the side effect of setting the valid
  333                                  * bits, but that is ok.  There are a bunch
  334                                  * of places in the VM system where we expected
  335                                  * m->dirty == VM_PAGE_BITS_ALL.  The file EOF
  336                                  * case is one of them.  If the page is still
  337                                  * partially dirty, make it fully dirty.
  338                                  *
  339                                  * note that we do not clear out the valid
  340                                  * bits.  This would prevent bogus_page
  341                                  * replacement from working properly.
  342                                  */
  343                                 vm_page_set_validclean(m, base, size);
  344                                 if (m->dirty != 0)
  345                                         m->dirty = VM_PAGE_BITS_ALL;
  346                         }
  347                 }
  348         }
  349         object->un_pager.vnp.vnp_size = nsize;
  350         object->size = nobjsize;
  351 }
  352 
  353 void
  354 vnode_pager_freepage(m)
  355         vm_page_t m;
  356 {
  357         vm_page_free(m);
  358 }
  359 
  360 /*
  361  * calculate the linear (byte) disk address of specified virtual
  362  * file address
  363  */
  364 static vm_offset_t
  365 vnode_pager_addr(vp, address, run)
  366         struct vnode *vp;
  367         vm_ooffset_t address;
  368         int *run;
  369 {
  370         int rtaddress;
  371         int bsize;
  372         daddr_t block;
  373         struct vnode *rtvp;
  374         int err;
  375         daddr_t vblock;
  376         int voffset;
  377 
  378         if ((int) address < 0)
  379                 return -1;
  380 
  381         if (vp->v_mount == NULL)
  382                 return -1;
  383 
  384         bsize = vp->v_mount->mnt_stat.f_iosize;
  385         vblock = address / bsize;
  386         voffset = address % bsize;
  387 
  388         err = VOP_BMAP(vp, vblock, &rtvp, &block, run, NULL);
  389 
  390         if (err || (block == -1))
  391                 rtaddress = -1;
  392         else {
  393                 rtaddress = block + voffset / DEV_BSIZE;
  394                 if( run) {
  395                         *run += 1;
  396                         *run *= bsize/PAGE_SIZE;
  397                         *run -= voffset/PAGE_SIZE;
  398                 }
  399         }
  400 
  401         return rtaddress;
  402 }
  403 
  404 /*
  405  * interrupt routine for I/O completion
  406  */
  407 static void
  408 vnode_pager_iodone(bp)
  409         struct buf *bp;
  410 {
  411         bp->b_flags |= B_DONE;
  412         wakeup(bp);
  413 }
  414 
  415 /*
  416  * small block file system vnode pager input
  417  */
  418 static int
  419 vnode_pager_input_smlfs(object, m)
  420         vm_object_t object;
  421         vm_page_t m;
  422 {
  423         int i;
  424         int s;
  425         struct vnode *dp, *vp;
  426         struct buf *bp;
  427         vm_offset_t kva;
  428         int fileaddr;
  429         vm_offset_t bsize;
  430         int error = 0;
  431 
  432         vp = object->handle;
  433         if (vp->v_mount == NULL)
  434                 return VM_PAGER_BAD;
  435 
  436         bsize = vp->v_mount->mnt_stat.f_iosize;
  437 
  438 
  439         VOP_BMAP(vp, 0, &dp, 0, NULL, NULL);
  440 
  441         kva = vm_pager_map_page(m);
  442 
  443         for (i = 0; i < PAGE_SIZE / bsize; i++) {
  444                 vm_ooffset_t address;
  445 
  446                 if (vm_page_bits(i * bsize, bsize) & m->valid)
  447                         continue;
  448 
  449                 address = IDX_TO_OFF(m->pindex) + i * bsize;
  450                 if (address >= object->un_pager.vnp.vnp_size) {
  451                         fileaddr = -1;
  452                 } else {
  453                         fileaddr = vnode_pager_addr(vp, address, NULL);
  454                 }
  455                 if (fileaddr != -1) {
  456                         bp = getpbuf(&vnode_pbuf_freecnt);
  457 
  458                         /* build a minimal buffer header */
  459                         bp->b_flags = B_READ | B_CALL;
  460                         bp->b_iodone = vnode_pager_iodone;
  461                         bp->b_rcred = bp->b_wcred = curproc->p_ucred;
  462                         if (bp->b_rcred != NOCRED)
  463                                 crhold(bp->b_rcred);
  464                         if (bp->b_wcred != NOCRED)
  465                                 crhold(bp->b_wcred);
  466                         bp->b_data = (caddr_t) kva + i * bsize;
  467                         bp->b_blkno = fileaddr;
  468                         pbgetvp(dp, bp);
  469                         bp->b_bcount = bsize;
  470                         bp->b_bufsize = bsize;
  471                         bp->b_runningbufspace = bp->b_bufsize;
  472                         runningbufspace += bp->b_runningbufspace;
  473 
  474                         /* do the input */
  475                         VOP_STRATEGY(bp->b_vp, bp);
  476 
  477                         /* we definitely need to be at splvm here */
  478 
  479                         s = splvm();
  480                         while ((bp->b_flags & B_DONE) == 0) {
  481                                 tsleep(bp, PVM, "vnsrd", 0);
  482                         }
  483                         splx(s);
  484                         if ((bp->b_flags & B_ERROR) != 0)
  485                                 error = EIO;
  486 
  487                         /*
  488                          * free the buffer header back to the swap buffer pool
  489                          */
  490                         relpbuf(bp, &vnode_pbuf_freecnt);
  491                         if (error)
  492                                 break;
  493 
  494                         vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize);
  495                 } else {
  496                         vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize);
  497                         bzero((caddr_t) kva + i * bsize, bsize);
  498                 }
  499         }
  500         vm_pager_unmap_page(kva);
  501         pmap_clear_modify(m);
  502         vm_page_flag_clear(m, PG_ZERO);
  503         if (error) {
  504                 return VM_PAGER_ERROR;
  505         }
  506         return VM_PAGER_OK;
  507 
  508 }
  509 
  510 
  511 /*
  512  * old style vnode pager output routine
  513  */
  514 static int
  515 vnode_pager_input_old(object, m)
  516         vm_object_t object;
  517         vm_page_t m;
  518 {
  519         struct uio auio;
  520         struct iovec aiov;
  521         int error;
  522         int size;
  523         vm_offset_t kva;
  524 
  525         error = 0;
  526 
  527         /*
  528          * Return failure if beyond current EOF
  529          */
  530         if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
  531                 return VM_PAGER_BAD;
  532         } else {
  533                 size = PAGE_SIZE;
  534                 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
  535                         size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
  536 
  537                 /*
  538                  * Allocate a kernel virtual address and initialize so that
  539                  * we can use VOP_READ/WRITE routines.
  540                  */
  541                 kva = vm_pager_map_page(m);
  542 
  543                 aiov.iov_base = (caddr_t) kva;
  544                 aiov.iov_len = size;
  545                 auio.uio_iov = &aiov;
  546                 auio.uio_iovcnt = 1;
  547                 auio.uio_offset = IDX_TO_OFF(m->pindex);
  548                 auio.uio_segflg = UIO_SYSSPACE;
  549                 auio.uio_rw = UIO_READ;
  550                 auio.uio_resid = size;
  551                 auio.uio_procp = curproc;
  552 
  553                 error = VOP_READ(object->handle, &auio, 0, curproc->p_ucred);
  554                 if (!error) {
  555                         register int count = size - auio.uio_resid;
  556 
  557                         if (count == 0)
  558                                 error = EINVAL;
  559                         else if (count != PAGE_SIZE)
  560                                 bzero((caddr_t) kva + count, PAGE_SIZE - count);
  561                 }
  562                 vm_pager_unmap_page(kva);
  563         }
  564         pmap_clear_modify(m);
  565         vm_page_undirty(m);
  566         vm_page_flag_clear(m, PG_ZERO);
  567         if (!error)
  568                 m->valid = VM_PAGE_BITS_ALL;
  569         return error ? VM_PAGER_ERROR : VM_PAGER_OK;
  570 }
  571 
  572 /*
  573  * generic vnode pager input routine
  574  */
  575 
  576 /*
  577  * EOPNOTSUPP is no longer legal.  For local media VFS's that do not
  578  * implement their own VOP_GETPAGES, their VOP_GETPAGES should call to
  579  * vnode_pager_generic_getpages() to implement the previous behaviour.
  580  *
  581  * All other FS's should use the bypass to get to the local media
  582  * backing vp's VOP_GETPAGES.
  583  */
  584 static int
  585 vnode_pager_getpages(object, m, count, reqpage)
  586         vm_object_t object;
  587         vm_page_t *m;
  588         int count;
  589         int reqpage;
  590 {
  591         int rtval;
  592         struct vnode *vp;
  593         int bytes = count * PAGE_SIZE;
  594 
  595         vp = object->handle;
  596         /* 
  597          * XXX temporary diagnostic message to help track stale FS code,
  598          * Returning EOPNOTSUPP from here may make things unhappy.
  599          */
  600         rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0);
  601         if (rtval == EOPNOTSUPP) {
  602             printf("vnode_pager: *** WARNING *** stale FS getpages\n");
  603             rtval = vnode_pager_generic_getpages( vp, m, bytes, reqpage);
  604         }
  605         return rtval;
  606 }
  607 
  608 
  609 /*
  610  * This is now called from local media FS's to operate against their
  611  * own vnodes if they fail to implement VOP_GETPAGES.
  612  */
  613 int
  614 vnode_pager_generic_getpages(vp, m, bytecount, reqpage)
  615         struct vnode *vp;
  616         vm_page_t *m;
  617         int bytecount;
  618         int reqpage;
  619 {
  620         vm_object_t object;
  621         vm_offset_t kva;
  622         off_t foff, tfoff, nextoff;
  623         int i, size, bsize, first, firstaddr;
  624         struct vnode *dp;
  625         int runpg;
  626         int runend;
  627         struct buf *bp;
  628         int s;
  629         int count;
  630         int error = 0;
  631 
  632         object = vp->v_object;
  633         count = bytecount / PAGE_SIZE;
  634 
  635         if (vp->v_mount == NULL)
  636                 return VM_PAGER_BAD;
  637 
  638         bsize = vp->v_mount->mnt_stat.f_iosize;
  639 
  640         /* get the UNDERLYING device for the file with VOP_BMAP() */
  641 
  642         /*
  643          * originally, we did not check for an error return value -- assuming
  644          * an fs always has a bmap entry point -- that assumption is wrong!!!
  645          */
  646         foff = IDX_TO_OFF(m[reqpage]->pindex);
  647 
  648         /*
  649          * if we can't bmap, use old VOP code
  650          */
  651         if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) {
  652                 for (i = 0; i < count; i++) {
  653                         if (i != reqpage) {
  654                                 vnode_pager_freepage(m[i]);
  655                         }
  656                 }
  657                 cnt.v_vnodein++;
  658                 cnt.v_vnodepgsin++;
  659                 return vnode_pager_input_old(object, m[reqpage]);
  660 
  661                 /*
  662                  * if the blocksize is smaller than a page size, then use
  663                  * special small filesystem code.  NFS sometimes has a small
  664                  * blocksize, but it can handle large reads itself.
  665                  */
  666         } else if ((PAGE_SIZE / bsize) > 1 &&
  667             (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) {
  668                 for (i = 0; i < count; i++) {
  669                         if (i != reqpage) {
  670                                 vnode_pager_freepage(m[i]);
  671                         }
  672                 }
  673                 cnt.v_vnodein++;
  674                 cnt.v_vnodepgsin++;
  675                 return vnode_pager_input_smlfs(object, m[reqpage]);
  676         }
  677 
  678         /*
  679          * If we have a completely valid page available to us, we can
  680          * clean up and return.  Otherwise we have to re-read the
  681          * media.
  682          */
  683 
  684         if (m[reqpage]->valid == VM_PAGE_BITS_ALL) {
  685                 for (i = 0; i < count; i++) {
  686                         if (i != reqpage)
  687                                 vnode_pager_freepage(m[i]);
  688                 }
  689                 return VM_PAGER_OK;
  690         }
  691         m[reqpage]->valid = 0;
  692 
  693         /*
  694          * here on direct device I/O
  695          */
  696 
  697         firstaddr = -1;
  698         /*
  699          * calculate the run that includes the required page
  700          */
  701         for(first = 0, i = 0; i < count; i = runend) {
  702                 firstaddr = vnode_pager_addr(vp,
  703                         IDX_TO_OFF(m[i]->pindex), &runpg);
  704                 if (firstaddr == -1) {
  705                         if (i == reqpage && foff < object->un_pager.vnp.vnp_size) {
  706                                 /* XXX no %qd in kernel. */
  707                                 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %d, foff: 0x%lx%08lx, vnp_size: 0x%lx%08lx",
  708                                  firstaddr, (u_long)(foff >> 32),
  709                                  (u_long)(u_int32_t)foff,
  710                                  (u_long)(u_int32_t)
  711                                  (object->un_pager.vnp.vnp_size >> 32),
  712                                  (u_long)(u_int32_t)
  713                                  object->un_pager.vnp.vnp_size);
  714                         }
  715                         vnode_pager_freepage(m[i]);
  716                         runend = i + 1;
  717                         first = runend;
  718                         continue;
  719                 }
  720                 runend = i + runpg;
  721                 if (runend <= reqpage) {
  722                         int j;
  723                         for (j = i; j < runend; j++) {
  724                                 vnode_pager_freepage(m[j]);
  725                         }
  726                 } else {
  727                         if (runpg < (count - first)) {
  728                                 for (i = first + runpg; i < count; i++)
  729                                         vnode_pager_freepage(m[i]);
  730                                 count = first + runpg;
  731                         }
  732                         break;
  733                 }
  734                 first = runend;
  735         }
  736 
  737         /*
  738          * the first and last page have been calculated now, move input pages
  739          * to be zero based...
  740          */
  741         if (first != 0) {
  742                 for (i = first; i < count; i++) {
  743                         m[i - first] = m[i];
  744                 }
  745                 count -= first;
  746                 reqpage -= first;
  747         }
  748 
  749         /*
  750          * calculate the file virtual address for the transfer
  751          */
  752         foff = IDX_TO_OFF(m[0]->pindex);
  753 
  754         /*
  755          * calculate the size of the transfer
  756          */
  757         size = count * PAGE_SIZE;
  758         if ((foff + size) > object->un_pager.vnp.vnp_size)
  759                 size = object->un_pager.vnp.vnp_size - foff;
  760 
  761         /*
  762          * round up physical size for real devices.
  763          */
  764         if (dp->v_type == VBLK || dp->v_type == VCHR) {
  765                 int secmask = dp->v_rdev->si_bsize_phys - 1;
  766                 KASSERT(secmask < PAGE_SIZE, ("vnode_pager_generic_getpages: sector size %d too large\n", secmask + 1));
  767                 size = (size + secmask) & ~secmask;
  768         }
  769 
  770         bp = getpbuf(&vnode_pbuf_freecnt);
  771         kva = (vm_offset_t) bp->b_data;
  772 
  773         /*
  774          * and map the pages to be read into the kva
  775          */
  776         pmap_qenter(kva, m, count);
  777 
  778         /* build a minimal buffer header */
  779         bp->b_flags = B_READ | B_CALL;
  780         bp->b_iodone = vnode_pager_iodone;
  781         /* B_PHYS is not set, but it is nice to fill this in */
  782         bp->b_rcred = bp->b_wcred = curproc->p_ucred;
  783         if (bp->b_rcred != NOCRED)
  784                 crhold(bp->b_rcred);
  785         if (bp->b_wcred != NOCRED)
  786                 crhold(bp->b_wcred);
  787         bp->b_blkno = firstaddr;
  788         pbgetvp(dp, bp);
  789         bp->b_bcount = size;
  790         bp->b_bufsize = size;
  791         bp->b_runningbufspace = bp->b_bufsize;
  792         runningbufspace += bp->b_runningbufspace;
  793 
  794         cnt.v_vnodein++;
  795         cnt.v_vnodepgsin += count;
  796 
  797         /* do the input */
  798         VOP_STRATEGY(bp->b_vp, bp);
  799 
  800         s = splvm();
  801         /* we definitely need to be at splvm here */
  802 
  803         while ((bp->b_flags & B_DONE) == 0) {
  804                 tsleep(bp, PVM, "vnread", 0);
  805         }
  806         splx(s);
  807         if ((bp->b_flags & B_ERROR) != 0)
  808                 error = EIO;
  809 
  810         if (!error) {
  811                 if (size != count * PAGE_SIZE)
  812                         bzero((caddr_t) kva + size, PAGE_SIZE * count - size);
  813         }
  814         pmap_qremove(kva, count);
  815 
  816         /*
  817          * free the buffer header back to the swap buffer pool
  818          */
  819         relpbuf(bp, &vnode_pbuf_freecnt);
  820 
  821         for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) {
  822                 vm_page_t mt;
  823 
  824                 nextoff = tfoff + PAGE_SIZE;
  825                 mt = m[i];
  826 
  827                 if (nextoff <= object->un_pager.vnp.vnp_size) {
  828                         /*
  829                          * Read filled up entire page.
  830                          */
  831                         mt->valid = VM_PAGE_BITS_ALL;
  832                         vm_page_undirty(mt);    /* should be an assert? XXX */
  833                         pmap_clear_modify(mt);
  834                 } else {
  835                         /*
  836                          * Read did not fill up entire page.  Since this
  837                          * is getpages, the page may be mapped, so we have
  838                          * to zero the invalid portions of the page even
  839                          * though we aren't setting them valid.
  840                          *
  841                          * Currently we do not set the entire page valid,
  842                          * we just try to clear the piece that we couldn't
  843                          * read.
  844                          */
  845                         vm_page_set_validclean(mt, 0,
  846                             object->un_pager.vnp.vnp_size - tfoff);
  847                         /* handled by vm_fault now */
  848                         /* vm_page_zero_invalid(mt, FALSE); */
  849                 }
  850                 
  851                 vm_page_flag_clear(mt, PG_ZERO);
  852                 if (i != reqpage) {
  853 
  854                         /*
  855                          * whether or not to leave the page activated is up in
  856                          * the air, but we should put the page on a page queue
  857                          * somewhere. (it already is in the object). Result:
  858                          * It appears that empirical results show that
  859                          * deactivating pages is best.
  860                          */
  861 
  862                         /*
  863                          * just in case someone was asking for this page we
  864                          * now tell them that it is ok to use
  865                          */
  866                         if (!error) {
  867                                 if (mt->flags & PG_WANTED)
  868                                         vm_page_activate(mt);
  869                                 else
  870                                         vm_page_deactivate(mt);
  871                                 vm_page_wakeup(mt);
  872                         } else {
  873                                 vnode_pager_freepage(mt);
  874                         }
  875                 }
  876         }
  877         if (error) {
  878                 printf("vnode_pager_getpages: I/O read error\n");
  879         }
  880         return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
  881 }
  882 
  883 /*
  884  * EOPNOTSUPP is no longer legal.  For local media VFS's that do not
  885  * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
  886  * vnode_pager_generic_putpages() to implement the previous behaviour.
  887  *
  888  * All other FS's should use the bypass to get to the local media
  889  * backing vp's VOP_PUTPAGES.
  890  */
  891 static void
  892 vnode_pager_putpages(object, m, count, sync, rtvals)
  893         vm_object_t object;
  894         vm_page_t *m;
  895         int count;
  896         boolean_t sync;
  897         int *rtvals;
  898 {
  899         int rtval;
  900         struct vnode *vp;
  901         int bytes = count * PAGE_SIZE;
  902 
  903         /*
  904          * Force synchronous operation if we are extremely low on memory
  905          * to prevent a low-memory deadlock.  VOP operations often need to
  906          * allocate more memory to initiate the I/O ( i.e. do a BMAP 
  907          * operation ).  The swapper handles the case by limiting the amount
  908          * of asynchronous I/O, but that sort of solution doesn't scale well
  909          * for the vnode pager without a lot of work.
  910          *
  911          * Also, the backing vnode's iodone routine may not wake the pageout
  912          * daemon up.  This should be probably be addressed XXX.
  913          */
  914 
  915         if ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min)
  916                 sync |= OBJPC_SYNC;
  917 
  918         /*
  919          * Call device-specific putpages function
  920          */
  921 
  922         vp = object->handle;
  923         rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0);
  924         if (rtval == EOPNOTSUPP) {
  925             printf("vnode_pager: *** WARNING *** stale FS putpages\n");
  926             rtval = vnode_pager_generic_putpages( vp, m, bytes, sync, rtvals);
  927         }
  928 }
  929 
  930 
  931 /*
  932  * This is now called from local media FS's to operate against their
  933  * own vnodes if they fail to implement VOP_PUTPAGES.
  934  *
  935  * This is typically called indirectly via the pageout daemon and
  936  * clustering has already typically occured, so in general we ask the
  937  * underlying filesystem to write the data out asynchronously rather
  938  * then delayed.
  939  */
  940 int
  941 vnode_pager_generic_putpages(vp, m, bytecount, flags, rtvals)
  942         struct vnode *vp;
  943         vm_page_t *m;
  944         int bytecount;
  945         int flags;
  946         int *rtvals;
  947 {
  948         int i;
  949         vm_object_t object;
  950         int count;
  951 
  952         int maxsize, ncount;
  953         vm_ooffset_t poffset;
  954         struct uio auio;
  955         struct iovec aiov;
  956         int error;
  957         int ioflags;
  958 
  959         object = vp->v_object;
  960         count = bytecount / PAGE_SIZE;
  961 
  962         for (i = 0; i < count; i++)
  963                 rtvals[i] = VM_PAGER_AGAIN;
  964 
  965         if ((int) m[0]->pindex < 0) {
  966                 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%x)\n",
  967                         (long)m[0]->pindex, m[0]->dirty);
  968                 rtvals[0] = VM_PAGER_BAD;
  969                 return VM_PAGER_BAD;
  970         }
  971 
  972         maxsize = count * PAGE_SIZE;
  973         ncount = count;
  974 
  975         poffset = IDX_TO_OFF(m[0]->pindex);
  976 
  977         /*
  978          * If the page-aligned write is larger then the actual file we
  979          * have to invalidate pages occuring beyond the file EOF.  However,
  980          * there is an edge case where a file may not be page-aligned where
  981          * the last page is partially invalid.  In this case the filesystem
  982          * may not properly clear the dirty bits for the entire page (which
  983          * could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
  984          * With the page locked we are free to fix-up the dirty bits here.
  985          *
  986          * We do not under any circumstances truncate the valid bits, as
  987          * this will screw up bogus page replacement.
  988          */
  989         if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
  990                 if (object->un_pager.vnp.vnp_size > poffset) {
  991                         int pgoff;
  992 
  993                         maxsize = object->un_pager.vnp.vnp_size - poffset;
  994                         ncount = btoc(maxsize);
  995                         if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
  996                                 vm_page_clear_dirty(m[ncount - 1], pgoff,
  997                                         PAGE_SIZE - pgoff);
  998                         }
  999                 } else {
 1000                         maxsize = 0;
 1001                         ncount = 0;
 1002                 }
 1003                 if (ncount < count) {
 1004                         for (i = ncount; i < count; i++) {
 1005                                 rtvals[i] = VM_PAGER_BAD;
 1006                         }
 1007                 }
 1008         }
 1009 
 1010         /*
 1011          * pageouts are already clustered, use IO_ASYNC to force a bawrite()
 1012          * rather then a bdwrite() to prevent paging I/O from saturating
 1013          * the buffer cache.  Dummy-up the sequential heuristic to cause
 1014          * large ranges to cluster.  If neither IO_SYNC or IO_ASYNC is set,
 1015          * the system decides how to cluster.
 1016          */
 1017         ioflags = IO_VMIO;
 1018         if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
 1019                 ioflags |= IO_SYNC;
 1020         else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
 1021                 ioflags |= IO_ASYNC;
 1022         ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
 1023         ioflags |= IO_SEQMAX << IO_SEQSHIFT;
 1024 
 1025         aiov.iov_base = (caddr_t) 0;
 1026         aiov.iov_len = maxsize;
 1027         auio.uio_iov = &aiov;
 1028         auio.uio_iovcnt = 1;
 1029         auio.uio_offset = poffset;
 1030         auio.uio_segflg = UIO_NOCOPY;
 1031         auio.uio_rw = UIO_WRITE;
 1032         auio.uio_resid = maxsize;
 1033         auio.uio_procp = (struct proc *) 0;
 1034         error = VOP_WRITE(vp, &auio, ioflags, curproc->p_ucred);
 1035         cnt.v_vnodeout++;
 1036         cnt.v_vnodepgsout += ncount;
 1037 
 1038         if (error) {
 1039                 printf("vnode_pager_putpages: I/O error %d\n", error);
 1040         }
 1041         if (auio.uio_resid) {
 1042                 printf("vnode_pager_putpages: residual I/O %d at %lu\n",
 1043                     auio.uio_resid, (u_long)m[0]->pindex);
 1044         }
 1045         for (i = 0; i < ncount; i++) {
 1046                 rtvals[i] = VM_PAGER_OK;
 1047         }
 1048         return rtvals[0];
 1049 }
 1050 
 1051 struct vnode *
 1052 vnode_pager_lock(object)
 1053         vm_object_t object;
 1054 {
 1055         struct proc *p = curproc;       /* XXX */
 1056 
 1057         for (; object != NULL; object = object->backing_object) {
 1058                 if (object->type != OBJT_VNODE)
 1059                         continue;
 1060                 if (object->flags & OBJ_DEAD)
 1061                         return NULL;
 1062 
 1063                 while (vget(object->handle,
 1064                         LK_NOPAUSE | LK_SHARED | LK_RETRY | LK_CANRECURSE, p)) {
 1065                         if ((object->flags & OBJ_DEAD) || (object->type != OBJT_VNODE))
 1066                                 return NULL;
 1067                         printf("vnode_pager_lock: retrying\n");
 1068                 }
 1069                 return object->handle;
 1070         }
 1071         return NULL;
 1072 }

Cache object: 60447f12f3e8b3e001e95ca891b43709


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.