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 1.2 2003/06/17 04:29:00 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 82 #include <vm/vm.h> 83 #include <vm/vm_param.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 89 MALLOC_DEFINE(M_VMPGDATA, "VM pgdata", "XXX: VM pager private data"); 90 91 extern struct pagerops defaultpagerops; 92 extern struct pagerops swappagerops; 93 extern struct pagerops vnodepagerops; 94 extern struct pagerops devicepagerops; 95 extern struct pagerops physpagerops; 96 97 int cluster_pbuf_freecnt = -1; /* unlimited to begin with */ 98 99 static int dead_pager_getpages __P((vm_object_t, vm_page_t *, int, int)); 100 static vm_object_t dead_pager_alloc __P((void *, vm_ooffset_t, vm_prot_t, 101 vm_ooffset_t)); 102 static void dead_pager_putpages __P((vm_object_t, vm_page_t *, int, int, int *)); 103 static boolean_t dead_pager_haspage __P((vm_object_t, vm_pindex_t, int *, int *)); 104 static void dead_pager_dealloc __P((vm_object_t)); 105 106 static int 107 dead_pager_getpages(obj, ma, count, req) 108 vm_object_t obj; 109 vm_page_t *ma; 110 int count; 111 int req; 112 { 113 return VM_PAGER_FAIL; 114 } 115 116 static vm_object_t 117 dead_pager_alloc(handle, size, prot, off) 118 void *handle; 119 vm_ooffset_t size; 120 vm_prot_t prot; 121 vm_ooffset_t off; 122 { 123 return NULL; 124 } 125 126 static void 127 dead_pager_putpages(object, m, count, flags, rtvals) 128 vm_object_t object; 129 vm_page_t *m; 130 int count; 131 int flags; 132 int *rtvals; 133 { 134 int i; 135 136 for (i = 0; i < count; i++) { 137 rtvals[i] = VM_PAGER_AGAIN; 138 } 139 } 140 141 static int 142 dead_pager_haspage(object, pindex, prev, next) 143 vm_object_t object; 144 vm_pindex_t pindex; 145 int *prev; 146 int *next; 147 { 148 if (prev) 149 *prev = 0; 150 if (next) 151 *next = 0; 152 return FALSE; 153 } 154 155 static void 156 dead_pager_dealloc(object) 157 vm_object_t object; 158 { 159 return; 160 } 161 162 static struct pagerops deadpagerops = { 163 NULL, 164 dead_pager_alloc, 165 dead_pager_dealloc, 166 dead_pager_getpages, 167 dead_pager_putpages, 168 dead_pager_haspage, 169 NULL 170 }; 171 172 struct pagerops *pagertab[] = { 173 &defaultpagerops, /* OBJT_DEFAULT */ 174 &swappagerops, /* OBJT_SWAP */ 175 &vnodepagerops, /* OBJT_VNODE */ 176 &devicepagerops, /* OBJT_DEVICE */ 177 &physpagerops, /* OBJT_PHYS */ 178 &deadpagerops /* OBJT_DEAD */ 179 }; 180 181 int npagers = sizeof(pagertab) / sizeof(pagertab[0]); 182 183 /* 184 * Kernel address space for mapping pages. 185 * Used by pagers where KVAs are needed for IO. 186 * 187 * XXX needs to be large enough to support the number of pending async 188 * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size 189 * (MAXPHYS == 64k) if you want to get the most efficiency. 190 */ 191 #define PAGER_MAP_SIZE (8 * 1024 * 1024) 192 193 int pager_map_size = PAGER_MAP_SIZE; 194 vm_map_t pager_map; 195 static int bswneeded; 196 static vm_offset_t swapbkva; /* swap buffers kva */ 197 198 void 199 vm_pager_init() 200 { 201 struct pagerops **pgops; 202 203 /* 204 * Initialize known pagers 205 */ 206 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++) 207 if (pgops && ((*pgops)->pgo_init != NULL)) 208 (*(*pgops)->pgo_init) (); 209 } 210 211 void 212 vm_pager_bufferinit() 213 { 214 struct buf *bp; 215 int i; 216 217 bp = swbuf; 218 /* 219 * Now set up swap and physical I/O buffer headers. 220 */ 221 for (i = 0; i < nswbuf; i++, bp++) { 222 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 223 BUF_LOCKINIT(bp); 224 LIST_INIT(&bp->b_dep); 225 bp->b_rcred = bp->b_wcred = NOCRED; 226 bp->b_xflags = 0; 227 } 228 229 cluster_pbuf_freecnt = nswbuf / 2; 230 231 swapbkva = kmem_alloc_pageable(pager_map, nswbuf * MAXPHYS); 232 if (!swapbkva) 233 panic("Not enough pager_map VM space for physical buffers"); 234 } 235 236 /* 237 * Allocate an instance of a pager of the given type. 238 * Size, protection and offset parameters are passed in for pagers that 239 * need to perform page-level validation (e.g. the device pager). 240 */ 241 vm_object_t 242 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, vm_prot_t prot, 243 vm_ooffset_t off) 244 { 245 struct pagerops *ops; 246 247 ops = pagertab[type]; 248 if (ops) 249 return ((*ops->pgo_alloc) (handle, size, prot, off)); 250 return (NULL); 251 } 252 253 void 254 vm_pager_deallocate(object) 255 vm_object_t object; 256 { 257 (*pagertab[object->type]->pgo_dealloc) (object); 258 } 259 260 /* 261 * vm_pager_strategy: 262 * 263 * called with no specific spl 264 * Execute strategy routine directly to pager. 265 */ 266 267 void 268 vm_pager_strategy(vm_object_t object, struct buf *bp) 269 { 270 if (pagertab[object->type]->pgo_strategy) { 271 (*pagertab[object->type]->pgo_strategy)(object, bp); 272 } else { 273 bp->b_flags |= B_ERROR; 274 bp->b_error = ENXIO; 275 biodone(bp); 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() 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_offset_t 308 vm_pager_map_page(m) 309 vm_page_t m; 310 { 311 vm_offset_t kva; 312 313 kva = kmem_alloc_wait(pager_map, PAGE_SIZE); 314 pmap_kenter(kva, VM_PAGE_TO_PHYS(m)); 315 return (kva); 316 } 317 318 void 319 vm_pager_unmap_page(kva) 320 vm_offset_t kva; 321 { 322 pmap_kremove(kva); 323 kmem_free_wakeup(pager_map, kva, PAGE_SIZE); 324 } 325 326 vm_object_t 327 vm_pager_object_lookup(pg_list, handle) 328 register struct pagerlst *pg_list; 329 void *handle; 330 { 331 register vm_object_t object; 332 333 for (object = TAILQ_FIRST(pg_list); object != NULL; object = TAILQ_NEXT(object,pager_object_list)) 334 if (object->handle == handle) 335 return (object); 336 return (NULL); 337 } 338 339 /* 340 * initialize a physical buffer 341 */ 342 343 static void 344 initpbuf(struct buf *bp) 345 { 346 bp->b_rcred = NOCRED; 347 bp->b_wcred = NOCRED; 348 bp->b_qindex = QUEUE_NONE; 349 bp->b_data = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva; 350 bp->b_kvabase = bp->b_data; 351 bp->b_kvasize = MAXPHYS; 352 bp->b_xflags = 0; 353 bp->b_flags = 0; 354 bp->b_error = 0; 355 BUF_LOCK(bp, LK_EXCLUSIVE); 356 } 357 358 /* 359 * allocate a physical buffer 360 * 361 * There are a limited number (nswbuf) of physical buffers. We need 362 * to make sure that no single subsystem is able to hog all of them, 363 * so each subsystem implements a counter which is typically initialized 364 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and 365 * increments it on release, and blocks if the counter hits zero. A 366 * subsystem may initialize the counter to -1 to disable the feature, 367 * but it must still be sure to match up all uses of getpbuf() with 368 * relpbuf() using the same variable. 369 * 370 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 371 * relatively soon when the rest of the subsystems get smart about it. XXX 372 */ 373 struct buf * 374 getpbuf(pfreecnt) 375 int *pfreecnt; 376 { 377 int s; 378 struct buf *bp; 379 380 s = splvm(); 381 382 for (;;) { 383 if (pfreecnt) { 384 while (*pfreecnt == 0) { 385 tsleep(pfreecnt, PVM, "wswbuf0", 0); 386 } 387 } 388 389 /* get a bp from the swap buffer header pool */ 390 if ((bp = TAILQ_FIRST(&bswlist)) != NULL) 391 break; 392 393 bswneeded = 1; 394 tsleep(&bswneeded, PVM, "wswbuf1", 0); 395 /* loop in case someone else grabbed one */ 396 } 397 TAILQ_REMOVE(&bswlist, bp, b_freelist); 398 if (pfreecnt) 399 --*pfreecnt; 400 splx(s); 401 402 initpbuf(bp); 403 return bp; 404 } 405 406 /* 407 * allocate a physical buffer, if one is available. 408 * 409 * Note that there is no NULL hack here - all subsystems using this 410 * call understand how to use pfreecnt. 411 */ 412 struct buf * 413 trypbuf(pfreecnt) 414 int *pfreecnt; 415 { 416 int s; 417 struct buf *bp; 418 419 s = splvm(); 420 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) { 421 splx(s); 422 return NULL; 423 } 424 TAILQ_REMOVE(&bswlist, bp, b_freelist); 425 426 --*pfreecnt; 427 428 splx(s); 429 430 initpbuf(bp); 431 432 return bp; 433 } 434 435 /* 436 * release a physical buffer 437 * 438 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 439 * relatively soon when the rest of the subsystems get smart about it. XXX 440 */ 441 void 442 relpbuf(bp, pfreecnt) 443 struct buf *bp; 444 int *pfreecnt; 445 { 446 int s; 447 448 s = splvm(); 449 450 if (bp->b_rcred != NOCRED) { 451 crfree(bp->b_rcred); 452 bp->b_rcred = NOCRED; 453 } 454 if (bp->b_wcred != NOCRED) { 455 crfree(bp->b_wcred); 456 bp->b_wcred = NOCRED; 457 } 458 459 if (bp->b_vp) 460 pbrelvp(bp); 461 462 BUF_UNLOCK(bp); 463 464 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 465 466 if (bswneeded) { 467 bswneeded = 0; 468 wakeup(&bswneeded); 469 } 470 if (pfreecnt) { 471 if (++*pfreecnt == 1) 472 wakeup(pfreecnt); 473 } 474 splx(s); 475 } 476 477 /******************************************************** 478 * CHAINING FUNCTIONS * 479 ******************************************************** 480 * 481 * These functions support recursion of I/O operations 482 * on bp's, typically by chaining one or more 'child' bp's 483 * to the parent. Synchronous, asynchronous, and semi-synchronous 484 * chaining is possible. 485 */ 486 487 /* 488 * vm_pager_chain_iodone: 489 * 490 * io completion routine for child bp. Currently we fudge a bit 491 * on dealing with b_resid. Since users of these routines may issue 492 * multiple children simultaniously, sequencing of the error can be lost. 493 */ 494 495 static void 496 vm_pager_chain_iodone(struct buf *nbp) 497 { 498 struct buf *bp; 499 500 if ((bp = nbp->b_chain.parent) != NULL) { 501 if (nbp->b_flags & B_ERROR) { 502 bp->b_flags |= B_ERROR; 503 bp->b_error = nbp->b_error; 504 } else if (nbp->b_resid != 0) { 505 bp->b_flags |= B_ERROR; 506 bp->b_error = EINVAL; 507 } else { 508 bp->b_resid -= nbp->b_bcount; 509 } 510 nbp->b_chain.parent = NULL; 511 --bp->b_chain.count; 512 if (bp->b_flags & B_WANT) { 513 bp->b_flags &= ~B_WANT; 514 wakeup(bp); 515 } 516 if (!bp->b_chain.count && (bp->b_xflags & BX_AUTOCHAINDONE)) { 517 bp->b_xflags &= ~BX_AUTOCHAINDONE; 518 if (bp->b_resid != 0 && !(bp->b_flags & B_ERROR)) { 519 bp->b_flags |= B_ERROR; 520 bp->b_error = EINVAL; 521 } 522 biodone(bp); 523 } 524 } 525 nbp->b_flags |= B_DONE; 526 nbp->b_flags &= ~B_ASYNC; 527 relpbuf(nbp, NULL); 528 } 529 530 /* 531 * getchainbuf: 532 * 533 * Obtain a physical buffer and chain it to its parent buffer. When 534 * I/O completes, the parent buffer will be B_SIGNAL'd. Errors are 535 * automatically propogated to the parent 536 * 537 * Since these are brand new buffers, we do not have to clear B_INVAL 538 * and B_ERROR because they are already clear. 539 */ 540 541 struct buf * 542 getchainbuf(struct buf *bp, struct vnode *vp, int flags) 543 { 544 struct buf *nbp = getpbuf(NULL); 545 546 nbp->b_chain.parent = bp; 547 ++bp->b_chain.count; 548 549 if (bp->b_chain.count > 4) 550 waitchainbuf(bp, 4, 0); 551 552 nbp->b_flags = B_CALL | (bp->b_flags & B_ORDERED) | flags; 553 nbp->b_rcred = nbp->b_wcred = proc0.p_ucred; 554 nbp->b_iodone = vm_pager_chain_iodone; 555 556 crhold(nbp->b_rcred); 557 crhold(nbp->b_wcred); 558 559 if (vp) 560 pbgetvp(vp, nbp); 561 return(nbp); 562 } 563 564 void 565 flushchainbuf(struct buf *nbp) 566 { 567 if (nbp->b_bcount) { 568 nbp->b_bufsize = nbp->b_bcount; 569 if ((nbp->b_flags & B_READ) == 0) 570 nbp->b_dirtyend = nbp->b_bcount; 571 BUF_KERNPROC(nbp); 572 VOP_STRATEGY(nbp->b_vp, nbp); 573 } else { 574 biodone(nbp); 575 } 576 } 577 578 void 579 waitchainbuf(struct buf *bp, int count, int done) 580 { 581 int s; 582 583 s = splbio(); 584 while (bp->b_chain.count > count) { 585 bp->b_flags |= B_WANT; 586 tsleep(bp, PRIBIO + 4, "bpchain", 0); 587 } 588 if (done) { 589 if (bp->b_resid != 0 && !(bp->b_flags & B_ERROR)) { 590 bp->b_flags |= B_ERROR; 591 bp->b_error = EINVAL; 592 } 593 biodone(bp); 594 } 595 splx(s); 596 } 597 598 void 599 autochaindone(struct buf *bp) 600 { 601 int s; 602 603 s = splbio(); 604 if (bp->b_chain.count == 0) 605 biodone(bp); 606 else 607 bp->b_xflags |= BX_AUTOCHAINDONE; 608 splx(s); 609 } 610 611