vm_pager.c revision 48225
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 * $Id: vm_pager.c,v 1.47 1999/05/06 20:00:34 phk Exp $ 65 */ 66 67/* 68 * Paging space routine stubs. Emulates a matchmaker-like interface 69 * for builtin pagers. 70 */ 71 72#include <sys/param.h> 73#include <sys/systm.h> 74#include <sys/kernel.h> 75#include <sys/vnode.h> 76#include <sys/buf.h> 77#include <sys/ucred.h> 78#include <sys/malloc.h> 79#include <sys/proc.h> 80 81#include <vm/vm.h> 82#include <vm/vm_param.h> 83#include <vm/vm_prot.h> 84#include <vm/vm_object.h> 85#include <vm/vm_page.h> 86#include <vm/vm_pager.h> 87#include <vm/vm_extern.h> 88 89MALLOC_DEFINE(M_VMPGDATA, "VM pgdata", "XXX: VM pager private data"); 90 91extern struct pagerops defaultpagerops; 92extern struct pagerops swappagerops; 93extern struct pagerops vnodepagerops; 94extern struct pagerops devicepagerops; 95 96int cluster_pbuf_freecnt = -1; /* unlimited to begin with */ 97 98static int dead_pager_getpages __P((vm_object_t, vm_page_t *, int, int)); 99static vm_object_t dead_pager_alloc __P((void *, vm_ooffset_t, vm_prot_t, 100 vm_ooffset_t)); 101static void dead_pager_putpages __P((vm_object_t, vm_page_t *, int, int, int *)); 102static boolean_t dead_pager_haspage __P((vm_object_t, vm_pindex_t, int *, int *)); 103static void dead_pager_dealloc __P((vm_object_t)); 104 105static int 106dead_pager_getpages(obj, ma, count, req) 107 vm_object_t obj; 108 vm_page_t *ma; 109 int count; 110 int req; 111{ 112 return VM_PAGER_FAIL; 113} 114 115static vm_object_t 116dead_pager_alloc(handle, size, prot, off) 117 void *handle; 118 vm_ooffset_t size; 119 vm_prot_t prot; 120 vm_ooffset_t off; 121{ 122 return NULL; 123} 124 125static void 126dead_pager_putpages(object, m, count, flags, rtvals) 127 vm_object_t object; 128 vm_page_t *m; 129 int count; 130 int flags; 131 int *rtvals; 132{ 133 int i; 134 135 for (i = 0; i < count; i++) { 136 rtvals[i] = VM_PAGER_AGAIN; 137 } 138} 139 140static int 141dead_pager_haspage(object, pindex, prev, next) 142 vm_object_t object; 143 vm_pindex_t pindex; 144 int *prev; 145 int *next; 146{ 147 if (prev) 148 *prev = 0; 149 if (next) 150 *next = 0; 151 return FALSE; 152} 153 154static void 155dead_pager_dealloc(object) 156 vm_object_t object; 157{ 158 return; 159} 160 161static struct pagerops deadpagerops = { 162 NULL, 163 dead_pager_alloc, 164 dead_pager_dealloc, 165 dead_pager_getpages, 166 dead_pager_putpages, 167 dead_pager_haspage, 168 NULL 169}; 170 171struct pagerops *pagertab[] = { 172 &defaultpagerops, /* OBJT_DEFAULT */ 173 &swappagerops, /* OBJT_SWAP */ 174 &vnodepagerops, /* OBJT_VNODE */ 175 &devicepagerops, /* OBJT_DEVICE */ 176 &deadpagerops /* OBJT_DEAD */ 177}; 178 179int npagers = sizeof(pagertab) / sizeof(pagertab[0]); 180 181/* 182 * Kernel address space for mapping pages. 183 * Used by pagers where KVAs are needed for IO. 184 * 185 * XXX needs to be large enough to support the number of pending async 186 * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size 187 * (MAXPHYS == 64k) if you want to get the most efficiency. 188 */ 189#define PAGER_MAP_SIZE (8 * 1024 * 1024) 190 191int pager_map_size = PAGER_MAP_SIZE; 192vm_map_t pager_map; 193static int bswneeded; 194static vm_offset_t swapbkva; /* swap buffers kva */ 195 196void 197vm_pager_init() 198{ 199 struct pagerops **pgops; 200 201 /* 202 * Initialize known pagers 203 */ 204 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++) 205 if (pgops && ((*pgops)->pgo_init != NULL)) 206 (*(*pgops)->pgo_init) (); 207} 208 209void 210vm_pager_bufferinit() 211{ 212 struct buf *bp; 213 int i; 214 215 bp = swbuf; 216 /* 217 * Now set up swap and physical I/O buffer headers. 218 */ 219 for (i = 0; i < nswbuf; i++, bp++) { 220 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 221 BUF_LOCKINIT(bp); 222 LIST_INIT(&bp->b_dep); 223 bp->b_rcred = bp->b_wcred = NOCRED; 224 bp->b_xflags = 0; 225 } 226 227 cluster_pbuf_freecnt = nswbuf / 2; 228 229 swapbkva = kmem_alloc_pageable(pager_map, nswbuf * MAXPHYS); 230 if (!swapbkva) 231 panic("Not enough pager_map VM space for physical buffers"); 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 */ 239vm_object_t 240vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, vm_prot_t prot, 241 vm_ooffset_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 251void 252vm_pager_deallocate(object) 253 vm_object_t object; 254{ 255 (*pagertab[object->type]->pgo_dealloc) (object); 256} 257 258/* 259 * vm_pager_strategy: 260 * 261 * called with no specific spl 262 * Execute strategy routine directly to pager. 263 */ 264 265void 266vm_pager_strategy(vm_object_t object, struct buf *bp) 267{ 268 if (pagertab[object->type]->pgo_strategy) { 269 (*pagertab[object->type]->pgo_strategy)(object, bp); 270 } else { 271 bp->b_flags |= B_ERROR; 272 bp->b_error = ENXIO; 273 biodone(bp); 274 } 275} 276 277/* 278 * vm_pager_get_pages() - inline, see vm/vm_pager.h 279 * vm_pager_put_pages() - inline, see vm/vm_pager.h 280 * vm_pager_has_page() - inline, see vm/vm_pager.h 281 * vm_pager_page_inserted() - inline, see vm/vm_pager.h 282 * vm_pager_page_removed() - inline, see vm/vm_pager.h 283 */ 284 285#if 0 286/* 287 * vm_pager_sync: 288 * 289 * Called by pageout daemon before going back to sleep. 290 * Gives pagers a chance to clean up any completed async pageing 291 * operations. 292 */ 293void 294vm_pager_sync() 295{ 296 struct pagerops **pgops; 297 298 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++) 299 if (pgops && ((*pgops)->pgo_sync != NULL)) 300 (*(*pgops)->pgo_sync) (); 301} 302 303#endif 304 305vm_offset_t 306vm_pager_map_page(m) 307 vm_page_t m; 308{ 309 vm_offset_t kva; 310 311 kva = kmem_alloc_wait(pager_map, PAGE_SIZE); 312 pmap_kenter(kva, VM_PAGE_TO_PHYS(m)); 313 return (kva); 314} 315 316void 317vm_pager_unmap_page(kva) 318 vm_offset_t kva; 319{ 320 pmap_kremove(kva); 321 kmem_free_wakeup(pager_map, kva, PAGE_SIZE); 322} 323 324vm_object_t 325vm_pager_object_lookup(pg_list, handle) 326 register struct pagerlst *pg_list; 327 void *handle; 328{ 329 register vm_object_t object; 330 331 for (object = TAILQ_FIRST(pg_list); object != NULL; object = TAILQ_NEXT(object,pager_object_list)) 332 if (object->handle == handle) 333 return (object); 334 return (NULL); 335} 336 337/* 338 * initialize a physical buffer 339 */ 340 341static void 342initpbuf(struct buf *bp) { 343 bp->b_rcred = NOCRED; 344 bp->b_wcred = NOCRED; 345 bp->b_qindex = QUEUE_NONE; 346 bp->b_data = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva; 347 bp->b_kvabase = bp->b_data; 348 bp->b_kvasize = MAXPHYS; 349 bp->b_xflags = 0; 350 bp->b_flags = 0; 351 bp->b_error = 0; 352 BUF_LOCK(bp, LK_EXCLUSIVE); 353} 354 355/* 356 * allocate a physical buffer 357 * 358 * There are a limited number (nswbuf) of physical buffers. We need 359 * to make sure that no single subsystem is able to hog all of them, 360 * so each subsystem implements a counter which is typically initialized 361 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and 362 * increments it on release, and blocks if the counter hits zero. A 363 * subsystem may initialize the counter to -1 to disable the feature, 364 * but it must still be sure to match up all uses of getpbuf() with 365 * relpbuf() using the same variable. 366 * 367 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 368 * relatively soon when the rest of the subsystems get smart about it. XXX 369 */ 370struct buf * 371getpbuf(pfreecnt) 372 int *pfreecnt; 373{ 374 int s; 375 struct buf *bp; 376 377 s = splvm(); 378 379 if (pfreecnt) { 380 while (*pfreecnt == 0) { 381 tsleep(pfreecnt, PVM, "wswbuf0", 0); 382 } 383 } 384 385 /* get a bp from the swap buffer header pool */ 386 while ((bp = TAILQ_FIRST(&bswlist)) == NULL) { 387 bswneeded = 1; 388 tsleep(&bswneeded, PVM, "wswbuf1", 0); 389 } 390 TAILQ_REMOVE(&bswlist, bp, b_freelist); 391 if (pfreecnt) 392 --*pfreecnt; 393 splx(s); 394 395 initpbuf(bp); 396 return bp; 397} 398 399/* 400 * allocate a physical buffer, if one is available. 401 * 402 * Note that there is no NULL hack here - all subsystems using this 403 * call understand how to use pfreecnt. 404 */ 405struct buf * 406trypbuf(pfreecnt) 407 int *pfreecnt; 408{ 409 int s; 410 struct buf *bp; 411 412 s = splvm(); 413 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) { 414 splx(s); 415 return NULL; 416 } 417 TAILQ_REMOVE(&bswlist, bp, b_freelist); 418 419 --*pfreecnt; 420 421 splx(s); 422 423 initpbuf(bp); 424 425 return bp; 426} 427 428/* 429 * release a physical buffer 430 * 431 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 432 * relatively soon when the rest of the subsystems get smart about it. XXX 433 */ 434void 435relpbuf(bp, pfreecnt) 436 struct buf *bp; 437 int *pfreecnt; 438{ 439 int s; 440 441 s = splvm(); 442 443 if (bp->b_rcred != NOCRED) { 444 crfree(bp->b_rcred); 445 bp->b_rcred = NOCRED; 446 } 447 if (bp->b_wcred != NOCRED) { 448 crfree(bp->b_wcred); 449 bp->b_wcred = NOCRED; 450 } 451 452 if (bp->b_vp) 453 pbrelvp(bp); 454 455 BUF_UNLOCK(bp); 456 457 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 458 459 if (bswneeded) { 460 bswneeded = 0; 461 wakeup(&bswneeded); 462 } 463 if (pfreecnt) { 464 if (++*pfreecnt == 1) 465 wakeup(pfreecnt); 466 } 467 splx(s); 468} 469 470/******************************************************** 471 * CHAINING FUNCTIONS * 472 ******************************************************** 473 * 474 * These functions support recursion of I/O operations 475 * on bp's, typically by chaining one or more 'child' bp's 476 * to the parent. Synchronous, asynchronous, and semi-synchronous 477 * chaining is possible. 478 */ 479 480/* 481 * vm_pager_chain_iodone: 482 * 483 * io completion routine for child bp. Currently we fudge a bit 484 * on dealing with b_resid. Since users of these routines may issue 485 * multiple children simultaniously, sequencing of the error can be lost. 486 */ 487 488static void 489vm_pager_chain_iodone(struct buf *nbp) 490{ 491 struct buf *bp; 492 493 if ((bp = nbp->b_chain.parent) != NULL) { 494 if (nbp->b_flags & B_ERROR) { 495 bp->b_flags |= B_ERROR; 496 bp->b_error = nbp->b_error; 497 } else if (nbp->b_resid != 0) { 498 bp->b_flags |= B_ERROR; 499 bp->b_error = EINVAL; 500 } else { 501 bp->b_resid -= nbp->b_bcount; 502 } 503 nbp->b_chain.parent = NULL; 504 --bp->b_chain.count; 505 if (bp->b_flags & B_WANT) { 506 bp->b_flags &= ~B_WANT; 507 wakeup(bp); 508 } 509 if (!bp->b_chain.count && (bp->b_flags & B_AUTOCHAINDONE)) { 510 bp->b_flags &= ~B_AUTOCHAINDONE; 511 if (bp->b_resid != 0 && !(bp->b_flags & B_ERROR)) { 512 bp->b_flags |= B_ERROR; 513 bp->b_error = EINVAL; 514 } 515 biodone(bp); 516 } 517 } 518 nbp->b_flags |= B_DONE; 519 nbp->b_flags &= ~B_ASYNC; 520 relpbuf(nbp, NULL); 521} 522 523/* 524 * getchainbuf: 525 * 526 * Obtain a physical buffer and chain it to its parent buffer. When 527 * I/O completes, the parent buffer will be B_SIGNAL'd. Errors are 528 * automatically propogated to the parent 529 * 530 * Since these are brand new buffers, we do not have to clear B_INVAL 531 * and B_ERROR because they are already clear. 532 */ 533 534struct buf * 535getchainbuf(struct buf *bp, struct vnode *vp, int flags) 536{ 537 struct buf *nbp = getpbuf(NULL); 538 539 nbp->b_chain.parent = bp; 540 ++bp->b_chain.count; 541 542 if (bp->b_chain.count > 4) 543 waitchainbuf(bp, 4, 0); 544 545 nbp->b_flags = B_CALL | (bp->b_flags & B_ORDERED) | flags; 546 nbp->b_rcred = nbp->b_wcred = proc0.p_ucred; 547 nbp->b_iodone = vm_pager_chain_iodone; 548 549 crhold(nbp->b_rcred); 550 crhold(nbp->b_wcred); 551 552 if (vp) 553 pbgetvp(vp, nbp); 554 return(nbp); 555} 556 557void 558flushchainbuf(struct buf *nbp) 559{ 560 if (nbp->b_bcount) { 561 nbp->b_bufsize = nbp->b_bcount; 562 if ((nbp->b_flags & B_READ) == 0) 563 nbp->b_dirtyend = nbp->b_bcount; 564 VOP_STRATEGY(nbp->b_vp, nbp); 565 } else { 566 biodone(nbp); 567 } 568} 569 570void 571waitchainbuf(struct buf *bp, int count, int done) 572{ 573 int s; 574 575 s = splbio(); 576 while (bp->b_chain.count > count) { 577 bp->b_flags |= B_WANT; 578 tsleep(bp, PRIBIO + 4, "bpchain", 0); 579 } 580 if (done) { 581 if (bp->b_resid != 0 && !(bp->b_flags & B_ERROR)) { 582 bp->b_flags |= B_ERROR; 583 bp->b_error = EINVAL; 584 } 585 biodone(bp); 586 } 587 splx(s); 588} 589 590void 591autochaindone(struct buf *bp) 592{ 593 int s; 594 595 s = splbio(); 596 if (bp->b_chain.count == 0) 597 biodone(bp); 598 else 599 bp->b_flags |= B_AUTOCHAINDONE; 600 splx(s); 601} 602 603