sys/uvm/uvm_vnode.c

   1 /*      $NetBSD: uvm_vnode.c,v 1.89 2007/12/01 10:40:28 yamt Exp $      */
   2
   3 /*
   4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
   5  * Copyright (c) 1991, 1993
   6  *      The Regents of the University of California.
   7  * Copyright (c) 1990 University of Utah.
   8  *
   9  * All rights reserved.
  10  *
  11  * This code is derived from software contributed to Berkeley by
  12  * the Systems Programming Group of the University of Utah Computer
  13  * Science Department.
  14  *
  15  * Redistribution and use in source and binary forms, with or without
  16  * modification, are permitted provided that the following conditions
  17  * are met:
  18  * 1. Redistributions of source code must retain the above copyright
  19  *    notice, this list of conditions and the following disclaimer.
  20  * 2. Redistributions in binary form must reproduce the above copyright
  21  *    notice, this list of conditions and the following disclaimer in the
  22  *    documentation and/or other materials provided with the distribution.
  23  * 3. All advertising materials mentioning features or use of this software
  24  *    must display the following acknowledgement:
  25  *      This product includes software developed by Charles D. Cranor,
  26  *      Washington University, the University of California, Berkeley and
  27  *      its contributors.
  28  * 4. Neither the name of the University nor the names of its contributors
  29  *    may be used to endorse or promote products derived from this software
  30  *    without specific prior written permission.
  31  *
  32  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  33  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  34  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  35  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  36  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  37  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  38  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  39  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  40  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  41  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  42  * SUCH DAMAGE.
  43  *
  44  *      @(#)vnode_pager.c       8.8 (Berkeley) 2/13/94
  45  * from: Id: uvm_vnode.c,v 1.1.2.26 1998/02/02 20:38:07 chuck Exp
  46  */
  47
  48 /*
  49  * uvm_vnode.c: the vnode pager.
  50  */
  51
  52 #include <sys/cdefs.h>
  53 __KERNEL_RCSID(0, "$NetBSD: uvm_vnode.c,v 1.89 2007/12/01 10:40:28 yamt Exp $");
  54
  55 #include "fs_nfs.h"
  56 #include "opt_uvmhist.h"
  57 #include "opt_ddb.h"
  58
  59 #include <sys/param.h>
  60 #include <sys/systm.h>
  61 #include <sys/kernel.h>
  62 #include <sys/proc.h>
  63 #include <sys/malloc.h>
  64 #include <sys/vnode.h>
  65 #include <sys/disklabel.h>
  66 #include <sys/ioctl.h>
  67 #include <sys/fcntl.h>
  68 #include <sys/conf.h>
  69 #include <sys/pool.h>
  70 #include <sys/mount.h>
  71
  72 #include <miscfs/specfs/specdev.h>
  73
  74 #include <uvm/uvm.h>
  75 #include <uvm/uvm_readahead.h>
  76
  77 /*
  78  * functions
  79  */
  80
  81 static void     uvn_detach(struct uvm_object *);
  82 static int      uvn_get(struct uvm_object *, voff_t, struct vm_page **, int *,
  83                         int, vm_prot_t, int, int);
  84 static int      uvn_put(struct uvm_object *, voff_t, voff_t, int);
  85 static void     uvn_reference(struct uvm_object *);
  86
  87 static int      uvn_findpage(struct uvm_object *, voff_t, struct vm_page **,
  88                              int);
  89
  90 /*
  91  * master pager structure
  92  */
  93
  94 const struct uvm_pagerops uvm_vnodeops = {
  95         .pgo_reference = uvn_reference,
  96         .pgo_detach = uvn_detach,
  97         .pgo_get = uvn_get,
  98         .pgo_put = uvn_put,
  99 };
 100
 101 /*
 102  * the ops!
 103  */
 104
 105 /*
 106  * uvn_reference
 107  *
 108  * duplicate a reference to a VM object.  Note that the reference
 109  * count must already be at least one (the passed in reference) so
 110  * there is no chance of the uvn being killed or locked out here.
 111  *
 112  * => caller must call with object unlocked.
 113  * => caller must be using the same accessprot as was used at attach time
 114  */
 115
 116 static void
 117 uvn_reference(struct uvm_object *uobj)
 118 {
 119         VREF((struct vnode *)uobj);
 120 }
 121
 122
 123 /*
 124  * uvn_detach
 125  *
 126  * remove a reference to a VM object.
 127  *
 128  * => caller must call with object unlocked and map locked.
 129  */
 130
 131 static void
 132 uvn_detach(struct uvm_object *uobj)
 133 {
 134         vrele((struct vnode *)uobj);
 135 }
 136
 137 /*
 138  * uvn_put: flush page data to backing store.
 139  *
 140  * => object must be locked on entry!   VOP_PUTPAGES must unlock it.
 141  * => flags: PGO_SYNCIO -- use sync. I/O
 142  * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed)
 143  */
 144
 145 static int
 146 uvn_put(struct uvm_object *uobj, voff_t offlo, voff_t offhi, int flags)
 147 {
 148         struct vnode *vp = (struct vnode *)uobj;
 149         int error;
 150
 151         KASSERT(mutex_owned(&vp->v_interlock));
 152         error = VOP_PUTPAGES(vp, offlo, offhi, flags);
 153
 154         return error;
 155 }
 156
 157
 158 /*
 159  * uvn_get: get pages (synchronously) from backing store
 160  *
 161  * => prefer map unlocked (not required)
 162  * => object must be locked!  we will _unlock_ it before starting any I/O.
 163  * => flags: PGO_ALLPAGES: get all of the pages
 164  *           PGO_LOCKED: fault data structures are locked
 165  * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx]
 166  * => NOTE: caller must check for released pages!!
 167  */
 168
 169 static int
 170 uvn_get(struct uvm_object *uobj, voff_t offset,
 171     struct vm_page **pps /* IN/OUT */,
 172     int *npagesp /* IN (OUT if PGO_LOCKED)*/,
 173     int centeridx, vm_prot_t access_type, int advice, int flags)
 174 {
 175         struct vnode *vp = (struct vnode *)uobj;
 176         int error;
 177
 178         UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist);
 179
 180         UVMHIST_LOG(ubchist, "vp %p off 0x%x", vp, (int)offset, 0,0);
 181
 182         if ((access_type & VM_PROT_WRITE) == 0 && (flags & PGO_LOCKED) == 0) {
 183                 vn_ra_allocctx(vp);
 184                 uvm_ra_request(vp->v_ractx, advice, uobj, offset,
 185                     *npagesp << PAGE_SHIFT);
 186         }
 187
 188         error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx,
 189                              access_type, advice, flags);
 190
 191         KASSERT(((flags & PGO_LOCKED) != 0 && mutex_owned(&vp->v_interlock)) ||
 192             (flags & PGO_LOCKED) == 0);
 193         return error;
 194 }
 195
 196
 197 /*
 198  * uvn_findpages:
 199  * return the page for the uobj and offset requested, allocating if needed.
 200  * => uobj must be locked.
 201  * => returned pages will be BUSY.
 202  */
 203
 204 int
 205 uvn_findpages(struct uvm_object *uobj, voff_t offset, int *npagesp,
 206     struct vm_page **pgs, int flags)
 207 {
 208         int i, count, found, npages, rv;
 209
 210         count = found = 0;
 211         npages = *npagesp;
 212         if (flags & UFP_BACKWARD) {
 213                 for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) {
 214                         rv = uvn_findpage(uobj, offset, &pgs[i], flags);
 215                         if (rv == 0) {
 216                                 if (flags & UFP_DIRTYONLY)
 217                                         break;
 218                         } else
 219                                 found++;
 220                         count++;
 221                 }
 222         } else {
 223                 for (i = 0; i < npages; i++, offset += PAGE_SIZE) {
 224                         rv = uvn_findpage(uobj, offset, &pgs[i], flags);
 225                         if (rv == 0) {
 226                                 if (flags & UFP_DIRTYONLY)
 227                                         break;
 228                         } else
 229                                 found++;
 230                         count++;
 231                 }
 232         }
 233         *npagesp = count;
 234         return (found);
 235 }
 236
 237 static int
 238 uvn_findpage(struct uvm_object *uobj, voff_t offset, struct vm_page **pgp,
 239     int flags)
 240 {
 241         struct vm_page *pg;
 242         bool dirty;
 243         UVMHIST_FUNC("uvn_findpage"); UVMHIST_CALLED(ubchist);
 244         UVMHIST_LOG(ubchist, "vp %p off 0x%lx", uobj, offset,0,0);
 245
 246         if (*pgp != NULL) {
 247                 UVMHIST_LOG(ubchist, "dontcare", 0,0,0,0);
 248                 return 0;
 249         }
 250         for (;;) {
 251                 /* look for an existing page */
 252                 pg = uvm_pagelookup(uobj, offset);
 253
 254                 /* nope?  allocate one now */
 255                 if (pg == NULL) {
 256                         if (flags & UFP_NOALLOC) {
 257                                 UVMHIST_LOG(ubchist, "noalloc", 0,0,0,0);
 258                                 return 0;
 259                         }
 260                         pg = uvm_pagealloc(uobj, offset, NULL, 0);
 261                         if (pg == NULL) {
 262                                 if (flags & UFP_NOWAIT) {
 263                                         UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
 264                                         return 0;
 265                                 }
 266                                 mutex_exit(&uobj->vmobjlock);
 267                                 uvm_wait("uvn_fp1");
 268                                 mutex_enter(&uobj->vmobjlock);
 269                                 continue;
 270                         }
 271                         UVMHIST_LOG(ubchist, "alloced %p", pg,0,0,0);
 272                         break;
 273                 } else if (flags & UFP_NOCACHE) {
 274                         UVMHIST_LOG(ubchist, "nocache",0,0,0,0);
 275                         return 0;
 276                 }
 277
 278                 /* page is there, see if we need to wait on it */
 279                 if ((pg->flags & PG_BUSY) != 0) {
 280                         if (flags & UFP_NOWAIT) {
 281                                 UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
 282                                 return 0;
 283                         }
 284                         pg->flags |= PG_WANTED;
 285                         UVMHIST_LOG(ubchist, "wait %p", pg,0,0,0);
 286                         UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
 287                                             "uvn_fp2", 0);
 288                         mutex_enter(&uobj->vmobjlock);
 289                         continue;
 290                 }
 291
 292                 /* skip PG_RDONLY pages if requested */
 293                 if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) {
 294                         UVMHIST_LOG(ubchist, "nordonly",0,0,0,0);
 295                         return 0;
 296                 }
 297
 298                 /* stop on clean pages if requested */
 299                 if (flags & UFP_DIRTYONLY) {
 300                         dirty = pmap_clear_modify(pg) ||
 301                                 (pg->flags & PG_CLEAN) == 0;
 302                         pg->flags |= PG_CLEAN;
 303                         if (!dirty) {
 304                                 UVMHIST_LOG(ubchist, "dirtonly", 0,0,0,0);
 305                                 return 0;
 306                         }
 307                 }
 308
 309                 /* mark the page BUSY and we're done. */
 310                 pg->flags |= PG_BUSY;
 311                 UVM_PAGE_OWN(pg, "uvn_findpage");
 312                 UVMHIST_LOG(ubchist, "found %p", pg,0,0,0);
 313                 break;
 314         }
 315         *pgp = pg;
 316         return 1;
 317 }
 318
 319 /*
 320  * uvm_vnp_setsize: grow or shrink a vnode uobj
 321  *
 322  * grow   => just update size value
 323  * shrink => toss un-needed pages
 324  *
 325  * => we assume that the caller has a reference of some sort to the
 326  *      vnode in question so that it will not be yanked out from under
 327  *      us.
 328  */
 329
 330 void
 331 uvm_vnp_setsize(struct vnode *vp, voff_t newsize)
 332 {
 333         struct uvm_object *uobj = &vp->v_uobj;
 334         voff_t pgend = round_page(newsize);
 335         voff_t oldsize;
 336         UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist);
 337
 338         mutex_enter(&uobj->vmobjlock);
 339         UVMHIST_LOG(ubchist, "vp %p old 0x%x new 0x%x",
 340             vp, vp->v_size, newsize, 0);
 341
 342         /*
 343          * now check if the size has changed: if we shrink we had better
 344          * toss some pages...
 345          */
 346
 347         KASSERT(newsize != VSIZENOTSET);
 348         KASSERT(vp->v_size <= vp->v_writesize);
 349         KASSERT(vp->v_size == vp->v_writesize ||
 350             newsize == vp->v_writesize || newsize <= vp->v_size);
 351
 352         oldsize = vp->v_writesize;
 353         KASSERT(oldsize != VSIZENOTSET || pgend > oldsize);
 354
 355         if (oldsize > pgend) {
 356                 (void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO);
 357                 mutex_enter(&uobj->vmobjlock);
 358         }
 359         vp->v_size = vp->v_writesize = newsize;
 360         mutex_exit(&uobj->vmobjlock);
 361 }
 362
 363 void
 364 uvm_vnp_setwritesize(struct vnode *vp, voff_t newsize)
 365 {
 366
 367         mutex_enter(&vp->v_interlock);
 368         KASSERT(newsize != VSIZENOTSET);
 369         KASSERT(vp->v_size != VSIZENOTSET);
 370         KASSERT(vp->v_writesize != VSIZENOTSET);
 371         KASSERT(vp->v_size <= vp->v_writesize);
 372         KASSERT(vp->v_size <= newsize);
 373         vp->v_writesize = newsize;
 374         mutex_exit(&vp->v_interlock);
 375 }
 376
 377 /*
 378  * uvm_vnp_zerorange:  set a range of bytes in a file to zero.
 379  */
 380
 381 void
 382 uvm_vnp_zerorange(struct vnode *vp, off_t off, size_t len)
 383 {
 384         void *win;
 385         int flags;
 386
 387         /*
 388          * XXXUBC invent kzero() and use it
 389          */
 390
 391         while (len) {
 392                 vsize_t bytelen = len;
 393
 394                 win = ubc_alloc(&vp->v_uobj, off, &bytelen, UVM_ADV_NORMAL,
 395                     UBC_WRITE);
 396                 memset(win, 0, bytelen);
 397                 flags = UBC_WANT_UNMAP(vp) ? UBC_UNMAP : 0;
 398                 ubc_release(win, flags);
 399
 400                 off += bytelen;
 401                 len -= bytelen;
 402         }
 403 }
 404
 405 bool
 406 uvn_text_p(struct uvm_object *uobj)
 407 {
 408         struct vnode *vp = (struct vnode *)uobj;
 409
 410         return (vp->v_iflag & VI_EXECMAP) != 0;
 411 }
 412
 413 bool
 414 uvn_clean_p(struct uvm_object *uobj)
 415 {
 416         struct vnode *vp = (struct vnode *)uobj;
 417
 418         return (vp->v_iflag & VI_ONWORKLST) == 0;
 419 }
 420
 421 bool
 422 uvn_needs_writefault_p(struct uvm_object *uobj)
 423 {
 424         struct vnode *vp = (struct vnode *)uobj;
 425
 426         return uvn_clean_p(uobj) ||
 427             (vp->v_iflag & (VI_WRMAP|VI_WRMAPDIRTY)) == VI_WRMAP;
 428 }