sys/vm/vm_pager.c

   1 /*
   2  * Copyright (c) 1991, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * The Mach Operating System project at Carnegie-Mellon University.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  *
  36  *      from: @(#)vm_pager.c    8.6 (Berkeley) 1/12/94
  37  *
  38  *
  39  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
  40  * All rights reserved.
  41  *
  42  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
  43  *
  44  * Permission to use, copy, modify and distribute this software and
  45  * its documentation is hereby granted, provided that both the copyright
  46  * notice and this permission notice appear in all copies of the
  47  * software, derivative works or modified versions, and any portions
  48  * thereof, and that both notices appear in supporting documentation.
  49  *
  50  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  51  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
  52  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
  53  *
  54  * Carnegie Mellon requests users of this software to return to
  55  *
  56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  57  *  School of Computer Science
  58  *  Carnegie Mellon University
  59  *  Pittsburgh PA 15213-3890
  60  *
  61  * any improvements or extensions that they make and grant Carnegie the
  62  * rights to redistribute these changes.
  63  *
  64  * $FreeBSD: src/sys/vm/vm_pager.c,v 1.54.2.2 2001/11/18 07:11:00 dillon Exp $
  65  * $DragonFly: src/sys/vm/vm_pager.c,v 1.24 2007/11/06 03:50:01 dillon Exp $
  66  */
  67
  68 /*
  69  *      Paging space routine stubs.  Emulates a matchmaker-like interface
  70  *      for builtin pagers.
  71  */
  72
  73 #include <sys/param.h>
  74 #include <sys/systm.h>
  75 #include <sys/kernel.h>
  76 #include <sys/vnode.h>
  77 #include <sys/buf.h>
  78 #include <sys/ucred.h>
  79 #include <sys/malloc.h>
  80 #include <sys/proc.h>
  81 #include <sys/thread2.h>
  82
  83 #include <vm/vm.h>
  84 #include <vm/vm_param.h>
  85 #include <vm/vm_object.h>
  86 #include <vm/vm_page.h>
  87 #include <vm/vm_pager.h>
  88 #include <vm/vm_extern.h>
  89
  90 #include <sys/buf2.h>
  91
  92 MALLOC_DEFINE(M_VMPGDATA, "VM pgdata", "XXX: VM pager private data");
  93
  94 extern struct pagerops defaultpagerops;
  95 extern struct pagerops swappagerops;
  96 extern struct pagerops vnodepagerops;
  97 extern struct pagerops devicepagerops;
  98 extern struct pagerops physpagerops;
  99
 100 int cluster_pbuf_freecnt = -1;  /* unlimited to begin with */
 101
 102 static int dead_pager_getpages (vm_object_t, vm_page_t *, int, int);
 103 static vm_object_t dead_pager_alloc (void *, off_t, vm_prot_t, off_t);
 104 static void dead_pager_putpages (vm_object_t, vm_page_t *, int, int, int *);
 105 static boolean_t dead_pager_haspage (vm_object_t, vm_pindex_t, int *, int *);
 106 static void dead_pager_dealloc (vm_object_t);
 107
 108 static int
 109 dead_pager_getpages(vm_object_t obj, vm_page_t *ma, int count, int req)
 110 {
 111         return VM_PAGER_FAIL;
 112 }
 113
 114 static vm_object_t
 115 dead_pager_alloc(void *handle, off_t size, vm_prot_t prot, off_t off)
 116 {
 117         return NULL;
 118 }
 119
 120 static void
 121 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
 122     int *rtvals)
 123 {
 124         int i;
 125
 126         for (i = 0; i < count; i++) {
 127                 rtvals[i] = VM_PAGER_AGAIN;
 128         }
 129 }
 130
 131 static int
 132 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *prev, int *next)
 133 {
 134         if (prev)
 135                 *prev = 0;
 136         if (next)
 137                 *next = 0;
 138         return FALSE;
 139 }
 140
 141 static void
 142 dead_pager_dealloc(vm_object_t object)
 143 {
 144         return;
 145 }
 146
 147 static struct pagerops deadpagerops = {
 148         NULL,
 149         dead_pager_alloc,
 150         dead_pager_dealloc,
 151         dead_pager_getpages,
 152         dead_pager_putpages,
 153         dead_pager_haspage,
 154         NULL
 155 };
 156
 157 struct pagerops *pagertab[] = {
 158         &defaultpagerops,       /* OBJT_DEFAULT */
 159         &swappagerops,          /* OBJT_SWAP */
 160         &vnodepagerops,         /* OBJT_VNODE */
 161         &devicepagerops,        /* OBJT_DEVICE */
 162         &physpagerops,          /* OBJT_PHYS */
 163         &deadpagerops           /* OBJT_DEAD */
 164 };
 165
 166 int npagers = sizeof(pagertab) / sizeof(pagertab[0]);
 167
 168 /*
 169  * Kernel address space for mapping pages.
 170  * Used by pagers where KVAs are needed for IO.
 171  *
 172  * XXX needs to be large enough to support the number of pending async
 173  * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size
 174  * (MAXPHYS == 64k) if you want to get the most efficiency.
 175  */
 176 #define PAGER_MAP_SIZE  (8 * 1024 * 1024)
 177
 178 int pager_map_size = PAGER_MAP_SIZE;
 179 struct vm_map pager_map;
 180
 181 static int bswneeded;
 182 static vm_offset_t swapbkva;            /* swap buffers kva */
 183 static TAILQ_HEAD(swqueue, buf) bswlist;
 184
 185 void
 186 vm_pager_init(void)
 187 {
 188         struct pagerops **pgops;
 189
 190         /*
 191          * Initialize the swap buffer list.
 192          */
 193         TAILQ_INIT(&bswlist);
 194
 195         /*
 196          * Initialize known pagers
 197          */
 198         for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
 199                 if (pgops && ((*pgops)->pgo_init != NULL))
 200                         (*(*pgops)->pgo_init) ();
 201 }
 202
 203 void
 204 vm_pager_bufferinit(void)
 205 {
 206         struct buf *bp;
 207         int i;
 208
 209         /*
 210          * Reserve KVM space for pbuf data.
 211          */
 212         swapbkva = kmem_alloc_pageable(&pager_map, nswbuf * MAXPHYS);
 213         if (!swapbkva)
 214                 panic("Not enough pager_map VM space for physical buffers");
 215
 216         /*
 217          * Initial pbuf setup.
 218          */
 219         bp = swbuf;
 220         for (i = 0; i < nswbuf; ++i, ++bp) {
 221                 bp->b_kvabase = (caddr_t)(i * MAXPHYS) + swapbkva;
 222                 bp->b_kvasize = MAXPHYS;
 223                 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
 224                 BUF_LOCKINIT(bp);
 225                 buf_dep_init(bp);
 226         }
 227
 228         /*
 229          * Allow the clustering code to use half of our pbufs.
 230          */
 231         cluster_pbuf_freecnt = nswbuf / 2;
 232 }
 233
 234 /*
 235  * Allocate an instance of a pager of the given type.
 236  * Size, protection and offset parameters are passed in for pagers that
 237  * need to perform page-level validation (e.g. the device pager).
 238  */
 239 vm_object_t
 240 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
 241                   vm_prot_t prot, off_t off)
 242 {
 243         struct pagerops *ops;
 244
 245         ops = pagertab[type];
 246         if (ops)
 247                 return ((*ops->pgo_alloc) (handle, size, prot, off));
 248         return (NULL);
 249 }
 250
 251 void
 252 vm_pager_deallocate(vm_object_t object)
 253 {
 254         (*pagertab[object->type]->pgo_dealloc) (object);
 255 }
 256
 257 /*
 258  *      vm_pager_strategy:
 259  *
 260  *      called with no specific spl
 261  *      Execute strategy routine directly to pager.
 262  */
 263
 264 void
 265 vm_pager_strategy(vm_object_t object, struct bio *bio)
 266 {
 267         struct buf *bp;
 268
 269         if (pagertab[object->type]->pgo_strategy) {
 270             (*pagertab[object->type]->pgo_strategy)(object, bio);
 271         } else {
 272                 bp = bio->bio_buf;
 273                 bp->b_flags |= B_ERROR;
 274                 bp->b_error = ENXIO;
 275                 biodone(bio);
 276         }
 277 }
 278
 279 /*
 280  * vm_pager_get_pages() - inline, see vm/vm_pager.h
 281  * vm_pager_put_pages() - inline, see vm/vm_pager.h
 282  * vm_pager_has_page() - inline, see vm/vm_pager.h
 283  * vm_pager_page_inserted() - inline, see vm/vm_pager.h
 284  * vm_pager_page_removed() - inline, see vm/vm_pager.h
 285  */
 286
 287 #if 0
 288 /*
 289  *      vm_pager_sync:
 290  *
 291  *      Called by pageout daemon before going back to sleep.
 292  *      Gives pagers a chance to clean up any completed async pageing
 293  *      operations.
 294  */
 295 void
 296 vm_pager_sync(void)
 297 {
 298         struct pagerops **pgops;
 299
 300         for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
 301                 if (pgops && ((*pgops)->pgo_sync != NULL))
 302                         (*(*pgops)->pgo_sync) ();
 303 }
 304
 305 #endif
 306
 307 vm_object_t
 308 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
 309 {
 310         vm_object_t object;
 311
 312         for (object = TAILQ_FIRST(pg_list); object != NULL; object = TAILQ_NEXT(object,pager_object_list))
 313                 if (object->handle == handle)
 314                         return (object);
 315         return (NULL);
 316 }
 317
 318 /*
 319  * Initialize a physical buffer.
 320  */
 321 static void
 322 initpbuf(struct buf *bp)
 323 {
 324         bp->b_qindex = 0; /* BQUEUE_NONE */
 325         bp->b_data = bp->b_kvabase;
 326         bp->b_flags = B_PAGING;
 327         bp->b_cmd = BUF_CMD_DONE;
 328         bp->b_error = 0;
 329         bp->b_bcount = 0;
 330         bp->b_bufsize = MAXPHYS;
 331         initbufbio(bp);
 332         xio_init(&bp->b_xio);
 333         BUF_LOCK(bp, LK_EXCLUSIVE);
 334 }
 335
 336 /*
 337  * allocate a physical buffer
 338  *
 339  *      There are a limited number (nswbuf) of physical buffers.  We need
 340  *      to make sure that no single subsystem is able to hog all of them,
 341  *      so each subsystem implements a counter which is typically initialized
 342  *      to 1/2 nswbuf.  getpbuf() decrements this counter in allocation and
 343  *      increments it on release, and blocks if the counter hits zero.  A
 344  *      subsystem may initialize the counter to -1 to disable the feature,
 345  *      but it must still be sure to match up all uses of getpbuf() with
 346  *      relpbuf() using the same variable.
 347  *
 348  *      NOTE: pfreecnt can be NULL, but this 'feature' will be removed
 349  *      relatively soon when the rest of the subsystems get smart about it. XXX
 350  */
 351 struct buf *
 352 getpbuf(int *pfreecnt)
 353 {
 354         struct buf *bp;
 355
 356         crit_enter();
 357
 358         for (;;) {
 359                 if (pfreecnt) {
 360                         while (*pfreecnt == 0) {
 361                                 tsleep(pfreecnt, 0, "wswbuf0", 0);
 362                         }
 363                 }
 364
 365                 /* get a bp from the swap buffer header pool */
 366                 if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
 367                         break;
 368
 369                 bswneeded = 1;
 370                 tsleep(&bswneeded, 0, "wswbuf1", 0);
 371                 /* loop in case someone else grabbed one */
 372         }
 373         TAILQ_REMOVE(&bswlist, bp, b_freelist);
 374         if (pfreecnt)
 375                 --*pfreecnt;
 376         crit_exit();
 377
 378         initpbuf(bp);
 379         return bp;
 380 }
 381
 382 /*
 383  * allocate a physical buffer, if one is available.
 384  *
 385  *      Note that there is no NULL hack here - all subsystems using this
 386  *      call understand how to use pfreecnt.
 387  */
 388 struct buf *
 389 trypbuf(int *pfreecnt)
 390 {
 391         struct buf *bp;
 392
 393         crit_enter();
 394         if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
 395                 crit_exit();
 396                 return NULL;
 397         }
 398         TAILQ_REMOVE(&bswlist, bp, b_freelist);
 399
 400         --*pfreecnt;
 401
 402         crit_exit();
 403
 404         initpbuf(bp);
 405
 406         return bp;
 407 }
 408
 409 /*
 410  * release a physical buffer
 411  *
 412  *      NOTE: pfreecnt can be NULL, but this 'feature' will be removed
 413  *      relatively soon when the rest of the subsystems get smart about it. XXX
 414  */
 415 void
 416 relpbuf(struct buf *bp, int *pfreecnt)
 417 {
 418         crit_enter();
 419
 420         KKASSERT(bp->b_flags & B_PAGING);
 421         BUF_UNLOCK(bp);
 422
 423         TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
 424
 425         if (bswneeded) {
 426                 bswneeded = 0;
 427                 wakeup(&bswneeded);
 428         }
 429         if (pfreecnt) {
 430                 if (++*pfreecnt == 1)
 431                         wakeup(pfreecnt);
 432         }
 433         crit_exit();
 434 }
 435