bsd/vm/vnode_pager.c

   1 /*
   2  * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * The contents of this file constitute Original Code as defined in and
   7  * are subject to the Apple Public Source License Version 1.1 (the
   8  * "License").  You may not use this file except in compliance with the
   9  * License.  Please obtain a copy of the License at
  10  * http://www.apple.com/publicsource and read it before using this file.
  11  *
  12  * This Original Code and all software distributed under the License are
  13  * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  14  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  15  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
  17  * License for the specific language governing rights and limitations
  18  * under the License.
  19  *
  20  * @APPLE_LICENSE_HEADER_END@
  21  */
  22 /*
  23  * Mach Operating System
  24  * Copyright (c) 1987 Carnegie-Mellon University
  25  * All rights reserved.  The CMU software License Agreement specifies
  26  * the terms and conditions for use and redistribution.
  27  */
  28 /*
  29  *      File:   vnode_pager.c
  30  *
  31  *      "Swap" pager that pages to/from vnodes.  Also
  32  *      handles demand paging from files.
  33  *
  34  */
  35
  36 #include <mach/boolean.h>
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/user.h>
  40 #include <sys/proc.h>
  41 #include <sys/kauth.h>
  42 #include <sys/buf.h>
  43 #include <sys/uio.h>
  44 #include <sys/vnode_internal.h>
  45 #include <sys/namei.h>
  46 #include <sys/mount_internal.h> /* needs internal due to fhandle_t */
  47 #include <sys/ubc_internal.h>
  48 #include <sys/lock.h>
  49
  50 #include <mach/mach_types.h>
  51 #include <mach/memory_object_types.h>
  52
  53 #include <vm/vm_map.h>
  54 #include <vm/vm_kern.h>
  55 #include <kern/zalloc.h>
  56 #include <kern/kalloc.h>
  57 #include <libkern/libkern.h>
  58
  59 #include <vm/vnode_pager.h>
  60 #include <vm/vm_pageout.h>
  61
  62 #include <kern/assert.h>
  63 #include <sys/kdebug.h>
  64 #include <machine/spl.h>
  65
  66 #include <nfs/rpcv2.h>
  67 #include <nfs/nfsproto.h>
  68 #include <nfs/nfs.h>
  69
  70 #include <vm/vm_protos.h>
  71
  72 unsigned int vp_pagein=0;
  73 unsigned int vp_pgodirty=0;
  74 unsigned int vp_pgoclean=0;
  75 unsigned int dp_pgouts=0;       /* Default pager pageouts */
  76 unsigned int dp_pgins=0;        /* Default pager pageins */
  77
  78 vm_object_offset_t
  79 vnode_pager_get_filesize(struct vnode *vp)
  80 {
  81
  82         return (vm_object_offset_t) ubc_getsize(vp);
  83 }
  84
  85 pager_return_t
  86 vnode_pageout(struct vnode *vp,
  87         upl_t                   upl,
  88         vm_offset_t             upl_offset,
  89         vm_object_offset_t      f_offset,
  90         vm_size_t               size,
  91         int                     flags,
  92         int                     *errorp)
  93 {
  94         struct proc     *p = current_proc();
  95         int             result = PAGER_SUCCESS;
  96         int             error = 0;
  97         int             error_ret = 0;
  98         daddr64_t blkno;
  99         int isize;
 100         int pg_index;
 101         int base_index;
 102         int offset;
 103         upl_page_info_t *pl;
 104         struct vfs_context context;
 105
 106         context.vc_proc = p;
 107         context.vc_ucred = kauth_cred_get();
 108
 109         isize = (int)size;
 110
 111         if (isize <= 0) {
 112                 result    = PAGER_ERROR;
 113                 error_ret = EINVAL;
 114                 goto out;
 115         }
 116         UBCINFOCHECK("vnode_pageout", vp);
 117
 118         if (UBCINVALID(vp)) {
 119                 result    = PAGER_ERROR;
 120                 error_ret = EINVAL;
 121
 122                 if (upl && !(flags & UPL_NOCOMMIT))
 123                         ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY);
 124                 goto out;
 125         }
 126         if ( !(flags & UPL_VNODE_PAGER)) {
 127                 /*
 128                  * This is a pageout from the default pager,
 129                  * just go ahead and call vnop_pageout since
 130                  * it has already sorted out the dirty ranges
 131                  */
 132                 dp_pgouts++;
 133
 134                 KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START,
 135                                       size, 1, 0, 0, 0);
 136
 137                 if ( (error_ret = VNOP_PAGEOUT(vp, upl, upl_offset, (off_t)f_offset,
 138                                                (size_t)size, flags, &context)) )
 139                         result = PAGER_ERROR;
 140
 141                 KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END,
 142                                       size, 1, 0, 0, 0);
 143
 144                 goto out;
 145         }
 146         /*
 147          * we come here for pageouts to 'real' files and
 148          * for msyncs...  the upl may not contain any
 149          * dirty pages.. it's our responsibility to sort
 150          * through it and find the 'runs' of dirty pages
 151          * to call VNOP_PAGEOUT on...
 152          */
 153         pl = ubc_upl_pageinfo(upl);
 154
 155         if (ubc_getsize(vp) == 0) {
 156                 /*
 157                  * if the file has been effectively deleted, then
 158                  * we need to go through the UPL and invalidate any
 159                  * buffer headers we might have that reference any
 160                  * of it's pages
 161                  */
 162                 for (offset = upl_offset; isize; isize -= PAGE_SIZE, offset += PAGE_SIZE) {
 163 #if NFSCLIENT
 164                         if (vp->v_tag == VT_NFS)
 165                                 /* check with nfs if page is OK to drop */
 166                                 error = nfs_buf_page_inval(vp, (off_t)f_offset);
 167                         else
 168 #endif
 169                         {
 170                                 blkno = ubc_offtoblk(vp, (off_t)f_offset);
 171                                 error = buf_invalblkno(vp, blkno, 0);
 172                         }
 173                         if (error) {
 174                                 if ( !(flags & UPL_NOCOMMIT))
 175                                         ubc_upl_abort_range(upl, offset, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY);
 176                                 if (error_ret == 0)
 177                                         error_ret = error;
 178                                 result = PAGER_ERROR;
 179
 180                         } else if ( !(flags & UPL_NOCOMMIT)) {
 181                                 ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY);
 182                         }
 183                         f_offset += PAGE_SIZE;
 184                 }
 185                 goto out;
 186         }
 187         /*
 188          * Ignore any non-present pages at the end of the
 189          * UPL so that we aren't looking at a upl that
 190          * may already have been freed by the preceeding
 191          * aborts/completions.
 192          */
 193         base_index = upl_offset / PAGE_SIZE;
 194
 195         for (pg_index = (upl_offset + isize) / PAGE_SIZE; pg_index > base_index;) {
 196                 if (upl_page_present(pl, --pg_index))
 197                         break;
 198                 if (pg_index == base_index) {
 199                         /*
 200                          * no pages were returned, so release
 201                          * our hold on the upl and leave
 202                          */
 203                         if ( !(flags & UPL_NOCOMMIT))
 204                                 ubc_upl_abort_range(upl, upl_offset, isize, UPL_ABORT_FREE_ON_EMPTY);
 205
 206                         goto out;
 207                 }
 208         }
 209         isize = (pg_index + 1) * PAGE_SIZE;
 210
 211         offset = upl_offset;
 212         pg_index = base_index;
 213
 214         while (isize) {
 215                 int  xsize;
 216                 int  num_of_pages;
 217
 218                 if ( !upl_page_present(pl, pg_index)) {
 219                         /*
 220                          * we asked for RET_ONLY_DIRTY, so it's possible
 221                          * to get back empty slots in the UPL
 222                          * just skip over them
 223                          */
 224                         offset += PAGE_SIZE;
 225                         isize  -= PAGE_SIZE;
 226                         pg_index++;
 227
 228                         continue;
 229                 }
 230                 if ( !upl_dirty_page(pl, pg_index)) {
 231                         /*
 232                          * if the page is not dirty and reached here it is
 233                          * marked precious or it is due to invalidation in
 234                          * memory_object_lock request as part of truncation
 235                          * We also get here from vm_object_terminate()
 236                          * So all you need to do in these
 237                          * cases is to invalidate incore buffer if it is there
 238                          * Note we must not sleep here if the buffer is busy - that is
 239                          * a lock inversion which causes deadlock.
 240                          */
 241                         vp_pgoclean++;
 242
 243 #if NFSCLIENT
 244                         if (vp->v_tag == VT_NFS)
 245                                 /* check with nfs if page is OK to drop */
 246                                 error = nfs_buf_page_inval(vp, (off_t)(f_offset + offset));
 247                         else
 248 #endif
 249                         {
 250                                 blkno = ubc_offtoblk(vp, (off_t)(f_offset + offset));
 251                                 error = buf_invalblkno(vp, blkno, 0);
 252                         }
 253                         if (error) {
 254                                 if ( !(flags & UPL_NOCOMMIT))
 255                                         ubc_upl_abort_range(upl, offset, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY);
 256                                 if (error_ret == 0)
 257                                         error_ret = error;
 258                                 result = PAGER_ERROR;
 259
 260                         } else if ( !(flags & UPL_NOCOMMIT)) {
 261                                 ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY);
 262                         }
 263                         offset += PAGE_SIZE;
 264                         isize  -= PAGE_SIZE;
 265                         pg_index++;
 266
 267                         continue;
 268                 }
 269                 vp_pgodirty++;
 270
 271                 num_of_pages = 1;
 272                 xsize = isize - PAGE_SIZE;
 273
 274                 while (xsize) {
 275                         if ( !upl_dirty_page(pl, pg_index + num_of_pages))
 276                                 break;
 277                         num_of_pages++;
 278                         xsize -= PAGE_SIZE;
 279                 }
 280                 xsize = num_of_pages * PAGE_SIZE;
 281
 282                 KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START,
 283                                       xsize, (int)(f_offset + offset), 0, 0, 0);
 284
 285                 if ( (error = VNOP_PAGEOUT(vp, upl, (vm_offset_t)offset,
 286                                         (off_t)(f_offset + offset), xsize,
 287                                            flags, &context)) ) {
 288                         if (error_ret == 0)
 289                                 error_ret = error;
 290                         result = PAGER_ERROR;
 291                 }
 292                 KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END,
 293                                       xsize, 0, 0, 0, 0);
 294
 295                 offset += xsize;
 296                 isize  -= xsize;
 297                 pg_index += num_of_pages;
 298         }
 299 out:
 300         if (errorp)
 301                 *errorp = error_ret;
 302
 303         return (result);
 304 }
 305
 306
 307 void IOSleep(int);
 308
 309 pager_return_t
 310 vnode_pagein(
 311         struct vnode            *vp,
 312         upl_t                   upl,
 313         vm_offset_t             upl_offset,
 314         vm_object_offset_t      f_offset,
 315         vm_size_t               size,
 316         int                     flags,
 317         int                     *errorp)
 318 {
 319         struct proc     *p = current_proc();
 320         struct uthread  *ut;
 321         upl_page_info_t *pl;
 322         int             result = PAGER_SUCCESS;
 323         int             error = 0;
 324         int             pages_in_upl;
 325         int             start_pg;
 326         int             last_pg;
 327         int             first_pg;
 328         int             xsize;
 329         int             abort_needed = 1;
 330
 331
 332         UBCINFOCHECK("vnode_pagein", vp);
 333
 334         if (UBCINVALID(vp)) {
 335                 result = PAGER_ERROR;
 336                 error  = PAGER_ERROR;
 337                 if (upl && !(flags & UPL_NOCOMMIT)) {
 338                         ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR);
 339                 }
 340                 goto out;
 341         }
 342         if (upl == (upl_t)NULL) {
 343                 if (size > (MAX_UPL_TRANSFER * PAGE_SIZE)) {
 344                         result = PAGER_ERROR;
 345                         error  = PAGER_ERROR;
 346                         goto out;
 347                 }
 348                 ubc_create_upl(vp, f_offset, size, &upl, &pl, UPL_RET_ONLY_ABSENT | UPL_SET_LITE);
 349
 350                 if (upl == (upl_t)NULL) {
 351                         result =  PAGER_ABSENT;
 352                         error = PAGER_ABSENT;
 353                         goto out;
 354                 }
 355                 upl_offset = 0;
 356                 /*
 357                  * if we get here, we've created the upl and
 358                  * are responsible for commiting/aborting it
 359                  * regardless of what the caller has passed in
 360                  */
 361                 flags &= ~UPL_NOCOMMIT;
 362
 363                 vp_pagein++;
 364         } else {
 365                 pl = ubc_upl_pageinfo(upl);
 366
 367                 dp_pgins++;
 368         }
 369         pages_in_upl = size / PAGE_SIZE;
 370         first_pg     = upl_offset / PAGE_SIZE;
 371
 372         /*
 373          * before we start marching forward, we must make sure we end on
 374          * a present page, otherwise we will be working with a freed
 375          * upl
 376          */
 377         for (last_pg = pages_in_upl - 1; last_pg >= first_pg; last_pg--) {
 378                 if (upl_page_present(pl, last_pg))
 379                         break;
 380         }
 381         pages_in_upl = last_pg + 1;
 382
 383         for (last_pg = first_pg; last_pg < pages_in_upl;) {
 384                 /*
 385                  * scan the upl looking for the next
 386                  * page that is present.... if all of the
 387                  * pages are absent, we're done
 388                  */
 389                 for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) {
 390                         if (upl_page_present(pl, last_pg))
 391                                 break;
 392                 }
 393                 if (last_pg == pages_in_upl)
 394                         break;
 395
 396                 /*
 397                  * if we get here, we've sitting on a page
 398                  * that is present... we want to skip over
 399                  * any range of 'valid' pages... if this takes
 400                  * us to the end of the request, than we're done
 401                  */
 402                 for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) {
 403                         if (!upl_valid_page(pl, last_pg) || !upl_page_present(pl, last_pg))
 404                                 break;
 405                 }
 406                 if (last_pg > start_pg) {
 407                         /*
 408                          * we've found a range of valid pages
 409                          * if we've got COMMIT responsibility
 410                          * commit this range of pages back to the
 411                          * cache unchanged
 412                          */
 413                         xsize = (last_pg - start_pg) * PAGE_SIZE;
 414
 415                         if (!(flags & UPL_NOCOMMIT))
 416                                 ubc_upl_abort_range(upl, start_pg * PAGE_SIZE, xsize, UPL_ABORT_FREE_ON_EMPTY);
 417
 418                         abort_needed = 0;
 419                 }
 420                 if (last_pg == pages_in_upl)
 421                         break;
 422
 423                 if (!upl_page_present(pl, last_pg))
 424                         /*
 425                          * if we found a range of valid pages
 426                          * terminated by a non-present page
 427                          * than start over
 428                          */
 429                         continue;
 430
 431                 /*
 432                  * scan from the found invalid page looking for a valid
 433                  * or non-present page before the end of the upl is reached, if we
 434                  * find one, then it will be the last page of the request to
 435                  * 'cluster_io'
 436                  */
 437                 for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) {
 438                         if (upl_valid_page(pl, last_pg) || !upl_page_present(pl, last_pg))
 439                                 break;
 440                 }
 441                 if (last_pg > start_pg) {
 442                         int xoff;
 443                         struct vfs_context context;
 444
 445                         context.vc_proc = p;
 446                         context.vc_ucred = kauth_cred_get();
 447                         xsize = (last_pg - start_pg) * PAGE_SIZE;
 448                         xoff  = start_pg * PAGE_SIZE;
 449
 450                         if ( (error = VNOP_PAGEIN(vp, upl, (vm_offset_t) xoff,
 451                                                (off_t)f_offset + xoff,
 452                                                xsize, flags, &context)) ) {
 453                                 result = PAGER_ERROR;
 454                                 error  = PAGER_ERROR;
 455
 456                         }
 457                         abort_needed = 0;
 458                 }
 459         }
 460         if (!(flags & UPL_NOCOMMIT) && abort_needed)
 461                 ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY);
 462 out:
 463         if (errorp)
 464                 *errorp = result;
 465
 466         ut = get_bsdthread_info(current_thread());
 467
 468         if (ut->uu_lowpri_delay) {
 469                 /*
 470                  * task is marked as a low priority I/O type
 471                  * and the I/O we issued while in this system call
 472                  * collided with normal I/O operations... we'll
 473                  * delay in order to mitigate the impact of this
 474                  * task on the normal operation of the system
 475                  */
 476                 IOSleep(ut->uu_lowpri_delay);
 477                 ut->uu_lowpri_delay = 0;
 478         }
 479         return (error);
 480 }
 481
 482 void
 483 vnode_pager_shutdown(void)
 484 {
 485         int i;
 486         vnode_t vp;
 487
 488         for(i = 0; i < MAX_BACKING_STORE; i++) {
 489                 vp = (vnode_t)(bs_port_table[i]).vp;
 490                 if (vp) {
 491                         (bs_port_table[i]).vp = 0;
 492
 493                         /* get rid of macx_swapon() reference */
 494                         vnode_rele(vp);
 495                 }
 496         }
 497 }
 498
 499
 500 void *
 501 upl_get_internal_page_list(upl_t upl)
 502 {
 503   return(UPL_GET_INTERNAL_PAGE_LIST(upl));
 504
 505 }