bsd/vm/vnode_pager.c

   1 /*
   2  * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
   7  *
   8  * This file contains Original Code and/or Modifications of Original Code
   9  * as defined in and that are subject to the Apple Public Source License
  10  * Version 2.0 (the 'License'). You may not use this file except in
  11  * compliance with the License. Please obtain a copy of the License at
  12  * http://www.opensource.apple.com/apsl/ and read it before using this
  13  * file.
  14  *
  15  * The Original Code and all software distributed under the License are
  16  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  17  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  18  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  20  * Please see the License for the specific language governing rights and
  21  * limitations under the License.
  22  *
  23  * @APPLE_LICENSE_HEADER_END@
  24  */
  25 /*
  26  * Mach Operating System
  27  * Copyright (c) 1987 Carnegie-Mellon University
  28  * All rights reserved.  The CMU software License Agreement specifies
  29  * the terms and conditions for use and redistribution.
  30  */
  31 /*
  32  *      File:   vnode_pager.c
  33  *
  34  *      "Swap" pager that pages to/from vnodes.  Also
  35  *      handles demand paging from files.
  36  *
  37  */
  38
  39 #include <mach/boolean.h>
  40 #include <sys/param.h>
  41 #include <sys/systm.h>
  42 #include <sys/proc.h>
  43 #include <sys/buf.h>
  44 #include <sys/uio.h>
  45 #include <sys/vnode.h>
  46 #include <sys/namei.h>
  47 #include <sys/mount.h>
  48 #include <sys/ubc.h>
  49 #include <sys/lock.h>
  50
  51 #include <mach/mach_types.h>
  52 #include <mach/memory_object_types.h>
  53
  54 #include <vm/vm_map.h>
  55 #include <vm/vm_kern.h>
  56 #include <kern/zalloc.h>
  57 #include <kern/kalloc.h>
  58 #include <libkern/libkern.h>
  59
  60 #include <vm/vnode_pager.h>
  61 #include <vm/vm_pageout.h>
  62
  63 #include <kern/assert.h>
  64 #include <sys/kdebug.h>
  65
  66 unsigned int vp_pagein=0;
  67 unsigned int vp_pgodirty=0;
  68 unsigned int vp_pgoclean=0;
  69 unsigned int dp_pgouts=0;       /* Default pager pageouts */
  70 unsigned int dp_pgins=0;        /* Default pager pageins */
  71
  72 vm_object_offset_t
  73 vnode_pager_get_filesize(struct vnode *vp)
  74 {
  75         if (UBCINVALID(vp)) {
  76                 return (vm_object_offset_t) 0;
  77         }
  78
  79         return (vm_object_offset_t) ubc_getsize(vp);
  80
  81 }
  82
  83 pager_return_t
  84 vnode_pageout(struct vnode *vp,
  85         upl_t                   upl,
  86         vm_offset_t             upl_offset,
  87         vm_object_offset_t      f_offset,
  88         vm_size_t               size,
  89         int                     flags,
  90         int                     *errorp)
  91 {
  92         int             result = PAGER_SUCCESS;
  93         struct proc     *p = current_proc();
  94         int             error = 0;
  95         int blkno=0, s;
  96         int cnt, isize;
  97         int pg_index;
  98         int offset;
  99         struct buf *bp;
 100         boolean_t       funnel_state;
 101         upl_page_info_t *pl;
 102         upl_t vpupl = NULL;
 103
 104         funnel_state = thread_funnel_set(kernel_flock, TRUE);
 105
 106         isize = (int)size;
 107
 108         if (isize <= 0) {
 109                 result = error = PAGER_ERROR;
 110                 goto out;
 111         }
 112         UBCINFOCHECK("vnode_pageout", vp);
 113
 114         if (UBCINVALID(vp)) {
 115                 result = error = PAGER_ERROR;
 116
 117                 if (upl && !(flags & UPL_NOCOMMIT))
 118                         ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY);
 119                 goto out;
 120         }
 121         if (upl) {
 122                 /*
 123                  * This is a pageout from the Default pager,
 124                  * just go ahead and call VOP_PAGEOUT
 125                  */
 126                 dp_pgouts++;
 127
 128                 KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START,
 129                                       size, 1, 0, 0, 0);
 130
 131                 if (error = VOP_PAGEOUT(vp, upl, upl_offset, (off_t)f_offset,
 132                                         (size_t)size, p->p_ucred, flags))
 133                         result = error = PAGER_ERROR;
 134
 135                 KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END,
 136                                       size, 1, 0, 0, 0);
 137
 138                 goto out;
 139         }
 140         ubc_create_upl(vp, f_offset, isize, &vpupl, &pl, UPL_FOR_PAGEOUT | UPL_COPYOUT_FROM);
 141
 142         if (vpupl == (upl_t) 0) {
 143                 result = error = PAGER_ABSENT;
 144                 goto out;
 145         }
 146         /*
 147          * if we get here, we've created the upl and
 148          * are responsible for commiting/aborting it
 149          * regardless of what the caller has passed in
 150          */
 151         flags &= ~UPL_NOCOMMIT;
 152
 153         if (ubc_getsize(vp) == 0) {
 154                 for (offset = 0; isize; isize -= PAGE_SIZE,
 155                                         offset += PAGE_SIZE) {
 156                         blkno = ubc_offtoblk(vp, (off_t)f_offset);
 157                         f_offset += PAGE_SIZE;
 158                         if ((bp = incore(vp, blkno)) &&
 159                             ISSET(bp->b_flags, B_BUSY)) {
 160                                 ubc_upl_abort_range(vpupl, offset, PAGE_SIZE,
 161                                                     UPL_ABORT_FREE_ON_EMPTY);
 162                                 result = error = PAGER_ERROR;
 163                                 continue;
 164                         } else if (bp) {
 165                                 bremfree(bp);
 166                                 SET(bp->b_flags, B_BUSY | B_INVAL);
 167                                 brelse(bp);
 168                         }
 169                         ubc_upl_commit_range(vpupl, offset, PAGE_SIZE,
 170                                              UPL_COMMIT_FREE_ON_EMPTY);
 171                 }
 172                 goto out;
 173         }
 174         pg_index = 0;
 175         offset   = 0;
 176
 177         while (isize) {
 178                 int  xsize;
 179                 int  num_of_pages;
 180
 181                 if ( !upl_valid_page(pl, pg_index)) {
 182                         ubc_upl_abort_range(vpupl, offset, PAGE_SIZE,
 183                                             UPL_ABORT_FREE_ON_EMPTY);
 184                         offset += PAGE_SIZE;
 185                         isize  -= PAGE_SIZE;
 186                         pg_index++;
 187
 188                         continue;
 189                 }
 190                 if ( !upl_dirty_page(pl, pg_index)) {
 191                         /*
 192                          * if the page is not dirty and reached here it is
 193                          * marked precious or it is due to invalidation in
 194                          * memory_object_lock request as part of truncation
 195                          * We also get here from vm_object_terminate()
 196                          * So all you need to do in these
 197                          * cases is to invalidate incore buffer if it is there
 198                          * Note we must not sleep here if B_BUSY - that is
 199                          * a lock inversion which causes deadlock.
 200                          */
 201                         blkno = ubc_offtoblk(vp, (off_t)(f_offset + offset));
 202                         s = splbio();
 203                         vp_pgoclean++;
 204                         if ((bp = incore(vp, blkno)) &&
 205                             ISSET(bp->b_flags, B_BUSY | B_NEEDCOMMIT)) {
 206                                 splx(s);
 207                                 ubc_upl_abort_range(vpupl, offset, PAGE_SIZE,
 208                                                     UPL_ABORT_FREE_ON_EMPTY);
 209                                 result = error = PAGER_ERROR;
 210                                 offset += PAGE_SIZE;
 211                                 isize -= PAGE_SIZE;
 212                                 pg_index++;
 213                                 continue;
 214                         } else if (bp) {
 215                                 bremfree(bp);
 216                                 SET(bp->b_flags, B_BUSY | B_INVAL );
 217                                 splx(s);
 218                                 brelse(bp);
 219                         } else
 220                                 splx(s);
 221
 222                         ubc_upl_commit_range(vpupl, offset, PAGE_SIZE,
 223                                              UPL_COMMIT_FREE_ON_EMPTY);
 224                         offset += PAGE_SIZE;
 225                         isize  -= PAGE_SIZE;
 226                         pg_index++;
 227
 228                         continue;
 229                 }
 230                 vp_pgodirty++;
 231
 232                 num_of_pages = 1;
 233                 xsize = isize - PAGE_SIZE;
 234
 235                 while (xsize) {
 236                         if ( !upl_valid_page(pl, pg_index + num_of_pages))
 237                                 break;
 238                         if ( !upl_dirty_page(pl, pg_index + num_of_pages))
 239                                 break;
 240                         num_of_pages++;
 241                         xsize -= PAGE_SIZE;
 242                 }
 243                 xsize = num_of_pages * PAGE_SIZE;
 244
 245                 KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START,
 246                                       xsize, 0, 0, 0, 0);
 247
 248                 if (error = VOP_PAGEOUT(vp, vpupl, (vm_offset_t)offset,
 249                                         (off_t)(f_offset + offset), xsize,
 250                                         p->p_ucred, flags))
 251                         result = error = PAGER_ERROR;
 252
 253                 KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END,
 254                                       xsize, 0, 0, 0, 0);
 255
 256                 offset += xsize;
 257                 isize  -= xsize;
 258                 pg_index += num_of_pages;
 259         }
 260 out:
 261         if (errorp)
 262                 *errorp = result;
 263
 264         thread_funnel_set(kernel_flock, funnel_state);
 265
 266         return (error);
 267 }
 268
 269
 270 pager_return_t
 271 vnode_pagein(
 272         struct vnode            *vp,
 273         upl_t                   upl,
 274         vm_offset_t             upl_offset,
 275         vm_object_offset_t      f_offset,
 276         vm_size_t               size,
 277         int                     flags,
 278         int                     *errorp)
 279 {
 280         struct proc     *p = current_proc();
 281         upl_page_info_t *pl;
 282         int             result = PAGER_SUCCESS;
 283         int             error = 0;
 284         int             xfer_size;
 285         int             pages_in_upl;
 286         int             start_pg;
 287         int             last_pg;
 288         int             first_pg;
 289         int             xsize;
 290         int             abort_needed = 1;
 291         boolean_t       funnel_state;
 292
 293
 294         funnel_state = thread_funnel_set(kernel_flock, TRUE);
 295
 296         UBCINFOCHECK("vnode_pagein", vp);
 297
 298         if (UBCINVALID(vp)) {
 299                 result = PAGER_ERROR;
 300                 error  = PAGER_ERROR;
 301                 if (upl && !(flags & UPL_NOCOMMIT)) {
 302                         ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR);
 303                 }
 304                 goto out;
 305         }
 306         if (upl == (upl_t)NULL) {
 307                 if (size > (MAX_UPL_TRANSFER * PAGE_SIZE)) {
 308                         result = PAGER_ERROR;
 309                         error  = PAGER_ERROR;
 310                         goto out;
 311                 }
 312                 ubc_create_upl(vp, f_offset, size, &upl, &pl, UPL_RET_ONLY_ABSENT);
 313
 314                 if (upl == (upl_t)NULL) {
 315                         result =  PAGER_ABSENT;
 316                         error = PAGER_ABSENT;
 317                         goto out;
 318                 }
 319                 upl_offset = 0;
 320                 /*
 321                  * if we get here, we've created the upl and
 322                  * are responsible for commiting/aborting it
 323                  * regardless of what the caller has passed in
 324                  */
 325                 flags &= ~UPL_NOCOMMIT;
 326
 327                 vp_pagein++;
 328         } else {
 329                 pl = ubc_upl_pageinfo(upl);
 330
 331                 dp_pgins++;
 332         }
 333         pages_in_upl = size / PAGE_SIZE;
 334         first_pg     = upl_offset / PAGE_SIZE;
 335
 336         /*
 337          * before we start marching forward, we must make sure we end on
 338          * a present page, otherwise we will be working with a freed
 339          * upl
 340          */
 341         for (last_pg = pages_in_upl - 1; last_pg >= first_pg; last_pg--) {
 342                 if (upl_page_present(pl, last_pg))
 343                         break;
 344         }
 345         pages_in_upl = last_pg + 1;
 346
 347         for (last_pg = first_pg; last_pg < pages_in_upl;) {
 348                 /*
 349                  * scan the upl looking for the next
 350                  * page that is present.... if all of the
 351                  * pages are absent, we're done
 352                  */
 353                 for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) {
 354                         if (upl_page_present(pl, last_pg))
 355                                 break;
 356                 }
 357                 if (last_pg == pages_in_upl)
 358                         break;
 359
 360                 /*
 361                  * if we get here, we've sitting on a page
 362                  * that is present... we want to skip over
 363                  * any range of 'valid' pages... if this takes
 364                  * us to the end of the request, than we're done
 365                  */
 366                 for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) {
 367                         if (!upl_valid_page(pl, last_pg) || !upl_page_present(pl, last_pg))
 368                                 break;
 369                 }
 370                 if (last_pg > start_pg) {
 371                         /*
 372                          * we've found a range of valid pages
 373                          * if we've got COMMIT responsibility
 374                          * commit this range of pages back to the
 375                          * cache unchanged
 376                          */
 377                         xsize = (last_pg - start_pg) * PAGE_SIZE;
 378
 379                         if (!(flags & UPL_NOCOMMIT))
 380                                 ubc_upl_abort_range(upl, start_pg * PAGE_SIZE, xsize, UPL_ABORT_FREE_ON_EMPTY);
 381
 382                         abort_needed = 0;
 383                 }
 384                 if (last_pg == pages_in_upl)
 385                         break;
 386
 387                 if (!upl_page_present(pl, last_pg))
 388                         /*
 389                          * if we found a range of valid pages
 390                          * terminated by a non-present page
 391                          * than start over
 392                          */
 393                         continue;
 394
 395                 /*
 396                  * scan from the found invalid page looking for a valid
 397                  * or non-present page before the end of the upl is reached, if we
 398                  * find one, then it will be the last page of the request to
 399                  * 'cluster_io'
 400                  */
 401                 for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) {
 402                         if (upl_valid_page(pl, last_pg) || !upl_page_present(pl, last_pg))
 403                                 break;
 404                 }
 405                 if (last_pg > start_pg) {
 406                         int xoff;
 407
 408                         xsize = (last_pg - start_pg) * PAGE_SIZE;
 409                         xoff  = start_pg * PAGE_SIZE;
 410
 411                         if (error = VOP_PAGEIN(vp, upl, (vm_offset_t) xoff,
 412                                                (off_t)f_offset + xoff,
 413                                                xsize, p->p_ucred,
 414                                                flags)) {
 415                                 result = PAGER_ERROR;
 416                                 error  = PAGER_ERROR;
 417
 418                         }
 419                         abort_needed = 0;
 420                 }
 421         }
 422         if (!(flags & UPL_NOCOMMIT) && abort_needed)
 423                 ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY);
 424 out:
 425         if (errorp)
 426                 *errorp = result;
 427         thread_funnel_set(kernel_flock, funnel_state);
 428
 429         return (error);
 430 }
 431
 432 void
 433 vnode_pager_shutdown()
 434 {
 435         int i;
 436         extern struct bs_map  bs_port_table[];
 437         struct vnode *vp;
 438
 439         for(i = 0; i < MAX_BACKING_STORE; i++) {
 440                 vp = (struct vnode *)(bs_port_table[i]).vp;
 441                 if (vp) {
 442                         (bs_port_table[i]).vp = 0;
 443                         ubc_rele(vp);
 444                         /* get rid of macx_swapon() namei() reference */
 445                         vrele(vp);
 446
 447                         /* get rid of macx_swapon() "extra" reference */
 448                         vrele(vp);
 449                 }
 450         }
 451 }
 452
 453
 454 void *
 455 upl_get_internal_page_list(upl_t upl)
 456 {
 457   return(UPL_GET_INTERNAL_PAGE_LIST(upl));
 458
 459 }