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 1.12 2005/06/02 20:57:21 swildner 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 */ 184 185 void 186 vm_pager_init(void) 187 { 188 struct pagerops **pgops; 189 190 /* 191 * Initialize known pagers 192 */ 193 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++) 194 if (pgops && ((*pgops)->pgo_init != NULL)) 195 (*(*pgops)->pgo_init) (); 196 } 197 198 void 199 vm_pager_bufferinit(void) 200 { 201 struct buf *bp; 202 int i; 203 204 bp = swbuf; 205 /* 206 * Now set up swap and physical I/O buffer headers. 207 */ 208 for (i = 0; i < nswbuf; i++, bp++) { 209 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 210 BUF_LOCKINIT(bp); 211 LIST_INIT(&bp->b_dep); 212 bp->b_xflags = 0; 213 } 214 215 cluster_pbuf_freecnt = nswbuf / 2; 216 217 swapbkva = kmem_alloc_pageable(pager_map, nswbuf * MAXPHYS); 218 if (!swapbkva) 219 panic("Not enough pager_map VM space for physical buffers"); 220 } 221 222 /* 223 * Allocate an instance of a pager of the given type. 224 * Size, protection and offset parameters are passed in for pagers that 225 * need to perform page-level validation (e.g. the device pager). 226 */ 227 vm_object_t 228 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, vm_prot_t prot, 229 vm_ooffset_t off) 230 { 231 struct pagerops *ops; 232 233 ops = pagertab[type]; 234 if (ops) 235 return ((*ops->pgo_alloc) (handle, size, prot, off)); 236 return (NULL); 237 } 238 239 void 240 vm_pager_deallocate(vm_object_t object) 241 { 242 (*pagertab[object->type]->pgo_dealloc) (object); 243 } 244 245 /* 246 * vm_pager_strategy: 247 * 248 * called with no specific spl 249 * Execute strategy routine directly to pager. 250 */ 251 252 void 253 vm_pager_strategy(vm_object_t object, struct buf *bp) 254 { 255 if (pagertab[object->type]->pgo_strategy) { 256 (*pagertab[object->type]->pgo_strategy)(object, bp); 257 } else { 258 bp->b_flags |= B_ERROR; 259 bp->b_error = ENXIO; 260 biodone(bp); 261 } 262 } 263 264 /* 265 * vm_pager_get_pages() - inline, see vm/vm_pager.h 266 * vm_pager_put_pages() - inline, see vm/vm_pager.h 267 * vm_pager_has_page() - inline, see vm/vm_pager.h 268 * vm_pager_page_inserted() - inline, see vm/vm_pager.h 269 * vm_pager_page_removed() - inline, see vm/vm_pager.h 270 */ 271 272 #if 0 273 /* 274 * vm_pager_sync: 275 * 276 * Called by pageout daemon before going back to sleep. 277 * Gives pagers a chance to clean up any completed async pageing 278 * operations. 279 */ 280 void 281 vm_pager_sync(void) 282 { 283 struct pagerops **pgops; 284 285 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++) 286 if (pgops && ((*pgops)->pgo_sync != NULL)) 287 (*(*pgops)->pgo_sync) (); 288 } 289 290 #endif 291 292 vm_object_t 293 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle) 294 { 295 vm_object_t object; 296 297 for (object = TAILQ_FIRST(pg_list); object != NULL; object = TAILQ_NEXT(object,pager_object_list)) 298 if (object->handle == handle) 299 return (object); 300 return (NULL); 301 } 302 303 /* 304 * initialize a physical buffer 305 */ 306 307 static void 308 initpbuf(struct buf *bp) 309 { 310 bp->b_qindex = QUEUE_NONE; 311 bp->b_data = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva; 312 bp->b_kvabase = bp->b_data; 313 bp->b_kvasize = MAXPHYS; 314 bp->b_xflags = 0; 315 bp->b_flags = 0; 316 bp->b_error = 0; 317 xio_init(&bp->b_xio); 318 BUF_LOCK(bp, LK_EXCLUSIVE); 319 } 320 321 /* 322 * allocate a physical buffer 323 * 324 * There are a limited number (nswbuf) of physical buffers. We need 325 * to make sure that no single subsystem is able to hog all of them, 326 * so each subsystem implements a counter which is typically initialized 327 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and 328 * increments it on release, and blocks if the counter hits zero. A 329 * subsystem may initialize the counter to -1 to disable the feature, 330 * but it must still be sure to match up all uses of getpbuf() with 331 * relpbuf() using the same variable. 332 * 333 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 334 * relatively soon when the rest of the subsystems get smart about it. XXX 335 */ 336 struct buf * 337 getpbuf(int *pfreecnt) 338 { 339 struct buf *bp; 340 341 crit_enter(); 342 343 for (;;) { 344 if (pfreecnt) { 345 while (*pfreecnt == 0) { 346 tsleep(pfreecnt, 0, "wswbuf0", 0); 347 } 348 } 349 350 /* get a bp from the swap buffer header pool */ 351 if ((bp = TAILQ_FIRST(&bswlist)) != NULL) 352 break; 353 354 bswneeded = 1; 355 tsleep(&bswneeded, 0, "wswbuf1", 0); 356 /* loop in case someone else grabbed one */ 357 } 358 TAILQ_REMOVE(&bswlist, bp, b_freelist); 359 if (pfreecnt) 360 --*pfreecnt; 361 crit_exit(); 362 363 initpbuf(bp); 364 return bp; 365 } 366 367 /* 368 * allocate a physical buffer, if one is available. 369 * 370 * Note that there is no NULL hack here - all subsystems using this 371 * call understand how to use pfreecnt. 372 */ 373 struct buf * 374 trypbuf(int *pfreecnt) 375 { 376 struct buf *bp; 377 378 crit_enter(); 379 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) { 380 crit_exit(); 381 return NULL; 382 } 383 TAILQ_REMOVE(&bswlist, bp, b_freelist); 384 385 --*pfreecnt; 386 387 crit_exit(); 388 389 initpbuf(bp); 390 391 return bp; 392 } 393 394 /* 395 * release a physical buffer 396 * 397 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 398 * relatively soon when the rest of the subsystems get smart about it. XXX 399 */ 400 void 401 relpbuf(struct buf *bp, int *pfreecnt) 402 { 403 crit_enter(); 404 405 if (bp->b_vp) 406 pbrelvp(bp); 407 408 BUF_UNLOCK(bp); 409 410 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 411 412 if (bswneeded) { 413 bswneeded = 0; 414 wakeup(&bswneeded); 415 } 416 if (pfreecnt) { 417 if (++*pfreecnt == 1) 418 wakeup(pfreecnt); 419 } 420 crit_exit(); 421 } 422 423 /******************************************************** 424 * CHAINING FUNCTIONS * 425 ******************************************************** 426 * 427 * These functions support recursion of I/O operations 428 * on bp's, typically by chaining one or more 'child' bp's 429 * to the parent. Synchronous, asynchronous, and semi-synchronous 430 * chaining is possible. 431 */ 432 433 /* 434 * vm_pager_chain_iodone: 435 * 436 * io completion routine for child bp. Currently we fudge a bit 437 * on dealing with b_resid. Since users of these routines may issue 438 * multiple children simultaniously, sequencing of the error can be lost. 439 */ 440 441 static void 442 vm_pager_chain_iodone(struct buf *nbp) 443 { 444 struct buf *bp; 445 446 if ((bp = nbp->b_chain.parent) != NULL) { 447 if (nbp->b_flags & B_ERROR) { 448 bp->b_flags |= B_ERROR; 449 bp->b_error = nbp->b_error; 450 } else if (nbp->b_resid != 0) { 451 bp->b_flags |= B_ERROR; 452 bp->b_error = EINVAL; 453 } else { 454 bp->b_resid -= nbp->b_bcount; 455 } 456 nbp->b_chain.parent = NULL; 457 --bp->b_chain.count; 458 if (bp->b_flags & B_WANT) { 459 bp->b_flags &= ~B_WANT; 460 wakeup(bp); 461 } 462 if (!bp->b_chain.count && (bp->b_xflags & BX_AUTOCHAINDONE)) { 463 bp->b_xflags &= ~BX_AUTOCHAINDONE; 464 if (bp->b_resid != 0 && !(bp->b_flags & B_ERROR)) { 465 bp->b_flags |= B_ERROR; 466 bp->b_error = EINVAL; 467 } 468 biodone(bp); 469 } 470 } 471 nbp->b_flags |= B_DONE; 472 nbp->b_flags &= ~B_ASYNC; 473 relpbuf(nbp, NULL); 474 } 475 476 /* 477 * getchainbuf: 478 * 479 * Obtain a physical buffer and chain it to its parent buffer. When 480 * I/O completes, the parent buffer will be B_SIGNAL'd. Errors are 481 * automatically propogated to the parent 482 * 483 * Since these are brand new buffers, we do not have to clear B_INVAL 484 * and B_ERROR because they are already clear. 485 */ 486 487 struct buf * 488 getchainbuf(struct buf *bp, struct vnode *vp, int flags) 489 { 490 struct buf *nbp = getpbuf(NULL); 491 492 nbp->b_chain.parent = bp; 493 ++bp->b_chain.count; 494 495 if (bp->b_chain.count > 4) 496 waitchainbuf(bp, 4, 0); 497 498 nbp->b_flags = B_CALL | (bp->b_flags & B_ORDERED) | flags; 499 nbp->b_iodone = vm_pager_chain_iodone; 500 501 if (vp) 502 pbgetvp(vp, nbp); 503 return(nbp); 504 } 505 506 void 507 flushchainbuf(struct buf *nbp) 508 { 509 if (nbp->b_bcount) { 510 nbp->b_bufsize = nbp->b_bcount; 511 if ((nbp->b_flags & B_READ) == 0) 512 nbp->b_dirtyend = nbp->b_bcount; 513 BUF_KERNPROC(nbp); 514 VOP_STRATEGY(nbp->b_vp, nbp); 515 } else { 516 biodone(nbp); 517 } 518 } 519 520 void 521 waitchainbuf(struct buf *bp, int count, int done) 522 { 523 crit_enter(); 524 while (bp->b_chain.count > count) { 525 bp->b_flags |= B_WANT; 526 tsleep(bp, 0, "bpchain", 0); 527 } 528 if (done) { 529 if (bp->b_resid != 0 && !(bp->b_flags & B_ERROR)) { 530 bp->b_flags |= B_ERROR; 531 bp->b_error = EINVAL; 532 } 533 biodone(bp); 534 } 535 crit_exit(); 536 } 537 538 void 539 autochaindone(struct buf *bp) 540 { 541 crit_enter(); 542 if (bp->b_chain.count == 0) 543 biodone(bp); 544 else 545 bp->b_xflags |= BX_AUTOCHAINDONE; 546 crit_exit(); 547 } 548