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 1.15 2005/08/08 16:53:12 hmp 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 *, vm_ooffset_t, vm_prot_t, 104 vm_ooffset_t); 105 static void dead_pager_putpages (vm_object_t, vm_page_t *, int, int, int *); 106 static boolean_t dead_pager_haspage (vm_object_t, vm_pindex_t, int *, int *); 107 static void dead_pager_dealloc (vm_object_t); 108 109 static int 110 dead_pager_getpages(vm_object_t obj, vm_page_t *ma, int count, int req) 111 { 112 return VM_PAGER_FAIL; 113 } 114 115 static vm_object_t 116 dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, 117 vm_ooffset_t off) 118 { 119 return NULL; 120 } 121 122 static void 123 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags, 124 int *rtvals) 125 { 126 int i; 127 128 for (i = 0; i < count; i++) { 129 rtvals[i] = VM_PAGER_AGAIN; 130 } 131 } 132 133 static int 134 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *prev, int *next) 135 { 136 if (prev) 137 *prev = 0; 138 if (next) 139 *next = 0; 140 return FALSE; 141 } 142 143 static void 144 dead_pager_dealloc(vm_object_t object) 145 { 146 return; 147 } 148 149 static struct pagerops deadpagerops = { 150 NULL, 151 dead_pager_alloc, 152 dead_pager_dealloc, 153 dead_pager_getpages, 154 dead_pager_putpages, 155 dead_pager_haspage, 156 NULL 157 }; 158 159 struct pagerops *pagertab[] = { 160 &defaultpagerops, /* OBJT_DEFAULT */ 161 &swappagerops, /* OBJT_SWAP */ 162 &vnodepagerops, /* OBJT_VNODE */ 163 &devicepagerops, /* OBJT_DEVICE */ 164 &physpagerops, /* OBJT_PHYS */ 165 &deadpagerops /* OBJT_DEAD */ 166 }; 167 168 int npagers = sizeof(pagertab) / sizeof(pagertab[0]); 169 170 /* 171 * Kernel address space for mapping pages. 172 * Used by pagers where KVAs are needed for IO. 173 * 174 * XXX needs to be large enough to support the number of pending async 175 * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size 176 * (MAXPHYS == 64k) if you want to get the most efficiency. 177 */ 178 #define PAGER_MAP_SIZE (8 * 1024 * 1024) 179 180 int pager_map_size = PAGER_MAP_SIZE; 181 vm_map_t pager_map; 182 static int bswneeded; 183 static vm_offset_t swapbkva; /* swap buffers kva */ 184 static TAILQ_HEAD(swqueue, buf) bswlist; 185 186 void 187 vm_pager_init(void) 188 { 189 struct pagerops **pgops; 190 191 /* 192 * Initialize the swap buffer list. 193 */ 194 TAILQ_INIT(&bswlist); 195 196 /* 197 * Initialize known pagers 198 */ 199 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++) 200 if (pgops && ((*pgops)->pgo_init != NULL)) 201 (*(*pgops)->pgo_init) (); 202 } 203 204 void 205 vm_pager_bufferinit(void) 206 { 207 struct buf *bp; 208 int i; 209 210 bp = swbuf; 211 /* 212 * Now set up swap and physical I/O buffer headers. 213 */ 214 for (i = 0; i < nswbuf; i++, bp++) { 215 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 216 BUF_LOCKINIT(bp); 217 LIST_INIT(&bp->b_dep); 218 bp->b_xflags = 0; 219 } 220 221 cluster_pbuf_freecnt = nswbuf / 2; 222 223 swapbkva = kmem_alloc_pageable(pager_map, nswbuf * MAXPHYS); 224 if (!swapbkva) 225 panic("Not enough pager_map VM space for physical buffers"); 226 } 227 228 /* 229 * Allocate an instance of a pager of the given type. 230 * Size, protection and offset parameters are passed in for pagers that 231 * need to perform page-level validation (e.g. the device pager). 232 */ 233 vm_object_t 234 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, vm_prot_t prot, 235 vm_ooffset_t off) 236 { 237 struct pagerops *ops; 238 239 ops = pagertab[type]; 240 if (ops) 241 return ((*ops->pgo_alloc) (handle, size, prot, off)); 242 return (NULL); 243 } 244 245 void 246 vm_pager_deallocate(vm_object_t object) 247 { 248 (*pagertab[object->type]->pgo_dealloc) (object); 249 } 250 251 /* 252 * vm_pager_strategy: 253 * 254 * called with no specific spl 255 * Execute strategy routine directly to pager. 256 */ 257 258 void 259 vm_pager_strategy(vm_object_t object, struct buf *bp) 260 { 261 if (pagertab[object->type]->pgo_strategy) { 262 (*pagertab[object->type]->pgo_strategy)(object, bp); 263 } else { 264 bp->b_flags |= B_ERROR; 265 bp->b_error = ENXIO; 266 biodone(bp); 267 } 268 } 269 270 /* 271 * vm_pager_get_pages() - inline, see vm/vm_pager.h 272 * vm_pager_put_pages() - inline, see vm/vm_pager.h 273 * vm_pager_has_page() - inline, see vm/vm_pager.h 274 * vm_pager_page_inserted() - inline, see vm/vm_pager.h 275 * vm_pager_page_removed() - inline, see vm/vm_pager.h 276 */ 277 278 #if 0 279 /* 280 * vm_pager_sync: 281 * 282 * Called by pageout daemon before going back to sleep. 283 * Gives pagers a chance to clean up any completed async pageing 284 * operations. 285 */ 286 void 287 vm_pager_sync(void) 288 { 289 struct pagerops **pgops; 290 291 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++) 292 if (pgops && ((*pgops)->pgo_sync != NULL)) 293 (*(*pgops)->pgo_sync) (); 294 } 295 296 #endif 297 298 vm_object_t 299 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle) 300 { 301 vm_object_t object; 302 303 for (object = TAILQ_FIRST(pg_list); object != NULL; object = TAILQ_NEXT(object,pager_object_list)) 304 if (object->handle == handle) 305 return (object); 306 return (NULL); 307 } 308 309 /* 310 * initialize a physical buffer 311 */ 312 313 static void 314 initpbuf(struct buf *bp) 315 { 316 bp->b_qindex = 0; /* BQUEUE_NONE */ 317 bp->b_data = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva; 318 bp->b_kvabase = bp->b_data; 319 bp->b_kvasize = MAXPHYS; 320 bp->b_xflags = 0; 321 bp->b_flags = 0; 322 bp->b_error = 0; 323 xio_init(&bp->b_xio); 324 BUF_LOCK(bp, LK_EXCLUSIVE); 325 } 326 327 /* 328 * allocate a physical buffer 329 * 330 * There are a limited number (nswbuf) of physical buffers. We need 331 * to make sure that no single subsystem is able to hog all of them, 332 * so each subsystem implements a counter which is typically initialized 333 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and 334 * increments it on release, and blocks if the counter hits zero. A 335 * subsystem may initialize the counter to -1 to disable the feature, 336 * but it must still be sure to match up all uses of getpbuf() with 337 * relpbuf() using the same variable. 338 * 339 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 340 * relatively soon when the rest of the subsystems get smart about it. XXX 341 */ 342 struct buf * 343 getpbuf(int *pfreecnt) 344 { 345 struct buf *bp; 346 347 crit_enter(); 348 349 for (;;) { 350 if (pfreecnt) { 351 while (*pfreecnt == 0) { 352 tsleep(pfreecnt, 0, "wswbuf0", 0); 353 } 354 } 355 356 /* get a bp from the swap buffer header pool */ 357 if ((bp = TAILQ_FIRST(&bswlist)) != NULL) 358 break; 359 360 bswneeded = 1; 361 tsleep(&bswneeded, 0, "wswbuf1", 0); 362 /* loop in case someone else grabbed one */ 363 } 364 TAILQ_REMOVE(&bswlist, bp, b_freelist); 365 if (pfreecnt) 366 --*pfreecnt; 367 crit_exit(); 368 369 initpbuf(bp); 370 return bp; 371 } 372 373 /* 374 * allocate a physical buffer, if one is available. 375 * 376 * Note that there is no NULL hack here - all subsystems using this 377 * call understand how to use pfreecnt. 378 */ 379 struct buf * 380 trypbuf(int *pfreecnt) 381 { 382 struct buf *bp; 383 384 crit_enter(); 385 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) { 386 crit_exit(); 387 return NULL; 388 } 389 TAILQ_REMOVE(&bswlist, bp, b_freelist); 390 391 --*pfreecnt; 392 393 crit_exit(); 394 395 initpbuf(bp); 396 397 return bp; 398 } 399 400 /* 401 * release a physical buffer 402 * 403 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 404 * relatively soon when the rest of the subsystems get smart about it. XXX 405 */ 406 void 407 relpbuf(struct buf *bp, int *pfreecnt) 408 { 409 crit_enter(); 410 411 if (bp->b_vp) 412 pbrelvp(bp); 413 414 BUF_UNLOCK(bp); 415 416 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 417 418 if (bswneeded) { 419 bswneeded = 0; 420 wakeup(&bswneeded); 421 } 422 if (pfreecnt) { 423 if (++*pfreecnt == 1) 424 wakeup(pfreecnt); 425 } 426 crit_exit(); 427 } 428 429 /******************************************************** 430 * CHAINING FUNCTIONS * 431 ******************************************************** 432 * 433 * These functions support recursion of I/O operations 434 * on bp's, typically by chaining one or more 'child' bp's 435 * to the parent. Synchronous, asynchronous, and semi-synchronous 436 * chaining is possible. 437 */ 438 439 /* 440 * vm_pager_chain_iodone: 441 * 442 * io completion routine for child bp. Currently we fudge a bit 443 * on dealing with b_resid. Since users of these routines may issue 444 * multiple children simultaniously, sequencing of the error can be lost. 445 */ 446 447 static void 448 vm_pager_chain_iodone(struct buf *nbp) 449 { 450 struct buf *bp; 451 452 if ((bp = nbp->b_chain.parent) != NULL) { 453 if (nbp->b_flags & B_ERROR) { 454 bp->b_flags |= B_ERROR; 455 bp->b_error = nbp->b_error; 456 } else if (nbp->b_resid != 0) { 457 bp->b_flags |= B_ERROR; 458 bp->b_error = EINVAL; 459 } else { 460 bp->b_resid -= nbp->b_bcount; 461 } 462 nbp->b_chain.parent = NULL; 463 --bp->b_chain.count; 464 if (bp->b_flags & B_WANT) { 465 bp->b_flags &= ~B_WANT; 466 wakeup(bp); 467 } 468 if (!bp->b_chain.count && (bp->b_xflags & BX_AUTOCHAINDONE)) { 469 bp->b_xflags &= ~BX_AUTOCHAINDONE; 470 if (bp->b_resid != 0 && !(bp->b_flags & B_ERROR)) { 471 bp->b_flags |= B_ERROR; 472 bp->b_error = EINVAL; 473 } 474 biodone(bp); 475 } 476 } 477 nbp->b_flags |= B_DONE; 478 nbp->b_flags &= ~B_ASYNC; 479 relpbuf(nbp, NULL); 480 } 481 482 /* 483 * getchainbuf: 484 * 485 * Obtain a physical buffer and chain it to its parent buffer. When 486 * I/O completes, the parent buffer will be B_SIGNAL'd. Errors are 487 * automatically propogated to the parent 488 * 489 * Since these are brand new buffers, we do not have to clear B_INVAL 490 * and B_ERROR because they are already clear. 491 */ 492 493 struct buf * 494 getchainbuf(struct buf *bp, struct vnode *vp, int flags) 495 { 496 struct buf *nbp = getpbuf(NULL); 497 498 nbp->b_chain.parent = bp; 499 ++bp->b_chain.count; 500 501 if (bp->b_chain.count > 4) 502 waitchainbuf(bp, 4, 0); 503 504 nbp->b_flags = (bp->b_flags & B_ORDERED) | flags; 505 nbp->b_iodone = vm_pager_chain_iodone; 506 507 if (vp) 508 pbgetvp(vp, nbp); 509 return(nbp); 510 } 511 512 void 513 flushchainbuf(struct buf *nbp) 514 { 515 if (nbp->b_bcount) { 516 nbp->b_bufsize = nbp->b_bcount; 517 if ((nbp->b_flags & B_READ) == 0) 518 nbp->b_dirtyend = nbp->b_bcount; 519 BUF_KERNPROC(nbp); 520 VOP_STRATEGY(nbp->b_vp, nbp); 521 } else { 522 biodone(nbp); 523 } 524 } 525 526 void 527 waitchainbuf(struct buf *bp, int count, int done) 528 { 529 crit_enter(); 530 while (bp->b_chain.count > count) { 531 bp->b_flags |= B_WANT; 532 tsleep(bp, 0, "bpchain", 0); 533 } 534 if (done) { 535 if (bp->b_resid != 0 && !(bp->b_flags & B_ERROR)) { 536 bp->b_flags |= B_ERROR; 537 bp->b_error = EINVAL; 538 } 539 biodone(bp); 540 } 541 crit_exit(); 542 } 543 544 void 545 autochaindone(struct buf *bp) 546 { 547 crit_enter(); 548 if (bp->b_chain.count == 0) 549 biodone(bp); 550 else 551 bp->b_xflags |= BX_AUTOCHAINDONE; 552 crit_exit(); 553 } 554