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 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from: @(#)vm_pager.c 8.6 (Berkeley) 1/12/94 33 * 34 * 35 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 36 * All rights reserved. 37 * 38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 39 * 40 * Permission to use, copy, modify and distribute this software and 41 * its documentation is hereby granted, provided that both the copyright 42 * notice and this permission notice appear in all copies of the 43 * software, derivative works or modified versions, and any portions 44 * thereof, and that both notices appear in supporting documentation. 45 * 46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 49 * 50 * Carnegie Mellon requests users of this software to return to 51 * 52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 53 * School of Computer Science 54 * Carnegie Mellon University 55 * Pittsburgh PA 15213-3890 56 * 57 * any improvements or extensions that they make and grant Carnegie the 58 * rights to redistribute these changes. 59 */ 60 61 /* 62 * Paging space routine stubs. Emulates a matchmaker-like interface 63 * for builtin pagers. 64 */ 65 66 #include <sys/cdefs.h> 67 __FBSDID("$FreeBSD$"); 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/kernel.h> 72 #include <sys/vnode.h> 73 #include <sys/bio.h> 74 #include <sys/buf.h> 75 #include <sys/ucred.h> 76 #include <sys/malloc.h> 77 78 #include <vm/vm.h> 79 #include <vm/vm_param.h> 80 #include <vm/vm_object.h> 81 #include <vm/vm_page.h> 82 #include <vm/vm_pager.h> 83 #include <vm/vm_extern.h> 84 85 MALLOC_DEFINE(M_VMPGDATA, "vm_pgdata", "XXX: VM pager private data"); 86 87 int cluster_pbuf_freecnt = -1; /* unlimited to begin with */ 88 89 static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int); 90 static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t, 91 vm_ooffset_t); 92 static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *); 93 static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *); 94 static void dead_pager_dealloc(vm_object_t); 95 96 static int 97 dead_pager_getpages(obj, ma, count, req) 98 vm_object_t obj; 99 vm_page_t *ma; 100 int count; 101 int req; 102 { 103 return VM_PAGER_FAIL; 104 } 105 106 static vm_object_t 107 dead_pager_alloc(handle, size, prot, off) 108 void *handle; 109 vm_ooffset_t size; 110 vm_prot_t prot; 111 vm_ooffset_t off; 112 { 113 return NULL; 114 } 115 116 static void 117 dead_pager_putpages(object, m, count, flags, rtvals) 118 vm_object_t object; 119 vm_page_t *m; 120 int count; 121 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(object, pindex, prev, next) 133 vm_object_t object; 134 vm_pindex_t pindex; 135 int *prev; 136 int *next; 137 { 138 if (prev) 139 *prev = 0; 140 if (next) 141 *next = 0; 142 return FALSE; 143 } 144 145 static void 146 dead_pager_dealloc(object) 147 vm_object_t object; 148 { 149 return; 150 } 151 152 static struct pagerops deadpagerops = { 153 .pgo_alloc = dead_pager_alloc, 154 .pgo_dealloc = dead_pager_dealloc, 155 .pgo_getpages = dead_pager_getpages, 156 .pgo_putpages = dead_pager_putpages, 157 .pgo_haspage = dead_pager_haspage, 158 }; 159 160 struct pagerops *pagertab[] = { 161 &defaultpagerops, /* OBJT_DEFAULT */ 162 &swappagerops, /* OBJT_SWAP */ 163 &vnodepagerops, /* OBJT_VNODE */ 164 &devicepagerops, /* OBJT_DEVICE */ 165 &physpagerops, /* OBJT_PHYS */ 166 &deadpagerops /* OBJT_DEAD */ 167 }; 168 169 static const int npagers = sizeof(pagertab) / sizeof(pagertab[0]); 170 171 /* 172 * Kernel address space for mapping pages. 173 * Used by pagers where KVAs are needed for IO. 174 * 175 * XXX needs to be large enough to support the number of pending async 176 * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size 177 * (MAXPHYS == 64k) if you want to get the most efficiency. 178 */ 179 vm_map_t pager_map; 180 static int bswneeded; 181 static vm_offset_t swapbkva; /* swap buffers kva */ 182 struct mtx pbuf_mtx; 183 static TAILQ_HEAD(swqueue, buf) bswlist; 184 185 void 186 vm_pager_init() 187 { 188 struct pagerops **pgops; 189 190 TAILQ_INIT(&bswlist); 191 /* 192 * Initialize known pagers 193 */ 194 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++) 195 if (pgops && ((*pgops)->pgo_init != NULL)) 196 (*(*pgops)->pgo_init) (); 197 } 198 199 void 200 vm_pager_bufferinit() 201 { 202 struct buf *bp; 203 int i; 204 205 mtx_init(&pbuf_mtx, "pbuf mutex", NULL, MTX_DEF); 206 bp = swbuf; 207 /* 208 * Now set up swap and physical I/O buffer headers. 209 */ 210 for (i = 0; i < nswbuf; i++, bp++) { 211 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 212 BUF_LOCKINIT(bp); 213 LIST_INIT(&bp->b_dep); 214 bp->b_rcred = bp->b_wcred = NOCRED; 215 bp->b_xflags = 0; 216 } 217 218 cluster_pbuf_freecnt = nswbuf / 2; 219 vnode_pbuf_freecnt = nswbuf / 2 + 1; 220 221 swapbkva = kmem_alloc_nofault(pager_map, nswbuf * MAXPHYS); 222 if (!swapbkva) 223 panic("Not enough pager_map VM space for physical buffers"); 224 } 225 226 /* 227 * Allocate an instance of a pager of the given type. 228 * Size, protection and offset parameters are passed in for pagers that 229 * need to perform page-level validation (e.g. the device pager). 230 */ 231 vm_object_t 232 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, 233 vm_prot_t prot, vm_ooffset_t off) 234 { 235 vm_object_t ret; 236 struct pagerops *ops; 237 238 ops = pagertab[type]; 239 if (ops) 240 ret = (*ops->pgo_alloc) (handle, size, prot, off); 241 else 242 ret = NULL; 243 return (ret); 244 } 245 246 /* 247 * The object must be locked. 248 */ 249 void 250 vm_pager_deallocate(object) 251 vm_object_t object; 252 { 253 254 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 255 (*pagertab[object->type]->pgo_dealloc) (object); 256 } 257 258 /* 259 * vm_pager_get_pages() - inline, see vm/vm_pager.h 260 * vm_pager_put_pages() - inline, see vm/vm_pager.h 261 * vm_pager_has_page() - inline, see vm/vm_pager.h 262 */ 263 264 vm_object_t 265 vm_pager_object_lookup(pg_list, handle) 266 struct pagerlst *pg_list; 267 void *handle; 268 { 269 vm_object_t object; 270 271 TAILQ_FOREACH(object, pg_list, pager_object_list) 272 if (object->handle == handle) 273 return (object); 274 return (NULL); 275 } 276 277 /* 278 * initialize a physical buffer 279 */ 280 281 /* 282 * XXX This probably belongs in vfs_bio.c 283 */ 284 static void 285 initpbuf(struct buf *bp) 286 { 287 KASSERT(bp->b_bufobj == NULL, ("initpbuf with bufobj")); 288 KASSERT(bp->b_vp == NULL, ("initpbuf with vp")); 289 bp->b_rcred = NOCRED; 290 bp->b_wcred = NOCRED; 291 bp->b_qindex = 0; /* On no queue (QUEUE_NONE) */ 292 bp->b_saveaddr = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva; 293 bp->b_data = bp->b_saveaddr; 294 bp->b_kvabase = bp->b_saveaddr; 295 bp->b_kvasize = MAXPHYS; 296 bp->b_xflags = 0; 297 bp->b_flags = 0; 298 bp->b_ioflags = 0; 299 bp->b_iodone = NULL; 300 bp->b_error = 0; 301 BUF_LOCK(bp, LK_EXCLUSIVE, NULL); 302 } 303 304 /* 305 * allocate a physical buffer 306 * 307 * There are a limited number (nswbuf) of physical buffers. We need 308 * to make sure that no single subsystem is able to hog all of them, 309 * so each subsystem implements a counter which is typically initialized 310 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and 311 * increments it on release, and blocks if the counter hits zero. A 312 * subsystem may initialize the counter to -1 to disable the feature, 313 * but it must still be sure to match up all uses of getpbuf() with 314 * relpbuf() using the same variable. 315 * 316 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 317 * relatively soon when the rest of the subsystems get smart about it. XXX 318 */ 319 struct buf * 320 getpbuf(int *pfreecnt) 321 { 322 struct buf *bp; 323 324 mtx_lock(&pbuf_mtx); 325 326 for (;;) { 327 if (pfreecnt) { 328 while (*pfreecnt == 0) { 329 msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0); 330 } 331 } 332 333 /* get a bp from the swap buffer header pool */ 334 if ((bp = TAILQ_FIRST(&bswlist)) != NULL) 335 break; 336 337 bswneeded = 1; 338 msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0); 339 /* loop in case someone else grabbed one */ 340 } 341 TAILQ_REMOVE(&bswlist, bp, b_freelist); 342 if (pfreecnt) 343 --*pfreecnt; 344 mtx_unlock(&pbuf_mtx); 345 346 initpbuf(bp); 347 return bp; 348 } 349 350 /* 351 * allocate a physical buffer, if one is available. 352 * 353 * Note that there is no NULL hack here - all subsystems using this 354 * call understand how to use pfreecnt. 355 */ 356 struct buf * 357 trypbuf(int *pfreecnt) 358 { 359 struct buf *bp; 360 361 mtx_lock(&pbuf_mtx); 362 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) { 363 mtx_unlock(&pbuf_mtx); 364 return NULL; 365 } 366 TAILQ_REMOVE(&bswlist, bp, b_freelist); 367 368 --*pfreecnt; 369 370 mtx_unlock(&pbuf_mtx); 371 372 initpbuf(bp); 373 374 return bp; 375 } 376 377 /* 378 * release a physical buffer 379 * 380 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 381 * relatively soon when the rest of the subsystems get smart about it. XXX 382 */ 383 void 384 relpbuf(struct buf *bp, int *pfreecnt) 385 { 386 387 if (bp->b_rcred != NOCRED) { 388 crfree(bp->b_rcred); 389 bp->b_rcred = NOCRED; 390 } 391 if (bp->b_wcred != NOCRED) { 392 crfree(bp->b_wcred); 393 bp->b_wcred = NOCRED; 394 } 395 396 KASSERT(bp->b_vp == NULL, ("relpbuf with vp")); 397 KASSERT(bp->b_bufobj == NULL, ("relpbuf with bufobj")); 398 399 BUF_UNLOCK(bp); 400 401 mtx_lock(&pbuf_mtx); 402 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 403 404 if (bswneeded) { 405 bswneeded = 0; 406 wakeup(&bswneeded); 407 } 408 if (pfreecnt) { 409 if (++*pfreecnt == 1) 410 wakeup(pfreecnt); 411 } 412 mtx_unlock(&pbuf_mtx); 413 } 414 415 /* 416 * Associate a p-buffer with a vnode. 417 * 418 * Also sets B_PAGING flag to indicate that vnode is not fully associated 419 * with the buffer. i.e. the bp has not been linked into the vnode or 420 * ref-counted. 421 */ 422 void 423 pbgetvp(struct vnode *vp, struct buf *bp) 424 { 425 426 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 427 KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)")); 428 429 bp->b_vp = vp; 430 bp->b_flags |= B_PAGING; 431 bp->b_bufobj = &vp->v_bufobj; 432 } 433 434 /* 435 * Associate a p-buffer with a vnode. 436 * 437 * Also sets B_PAGING flag to indicate that vnode is not fully associated 438 * with the buffer. i.e. the bp has not been linked into the vnode or 439 * ref-counted. 440 */ 441 void 442 pbgetbo(struct bufobj *bo, struct buf *bp) 443 { 444 445 KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)")); 446 KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)")); 447 448 bp->b_flags |= B_PAGING; 449 bp->b_bufobj = bo; 450 } 451 452 /* 453 * Disassociate a p-buffer from a vnode. 454 */ 455 void 456 pbrelvp(struct buf *bp) 457 { 458 459 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 460 KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj")); 461 462 /* XXX REMOVE ME */ 463 BO_LOCK(bp->b_bufobj); 464 if (TAILQ_NEXT(bp, b_bobufs) != NULL) { 465 panic( 466 "relpbuf(): b_vp was probably reassignbuf()d %p %x", 467 bp, 468 (int)bp->b_flags 469 ); 470 } 471 BO_UNLOCK(bp->b_bufobj); 472 bp->b_vp = NULL; 473 bp->b_bufobj = NULL; 474 bp->b_flags &= ~B_PAGING; 475 } 476 477 /* 478 * Disassociate a p-buffer from a bufobj. 479 */ 480 void 481 pbrelbo(struct buf *bp) 482 { 483 484 KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode")); 485 KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj")); 486 487 /* XXX REMOVE ME */ 488 BO_LOCK(bp->b_bufobj); 489 if (TAILQ_NEXT(bp, b_bobufs) != NULL) { 490 panic( 491 "relpbuf(): b_vp was probably reassignbuf()d %p %x", 492 bp, 493 (int)bp->b_flags 494 ); 495 } 496 BO_UNLOCK(bp->b_bufobj); 497 bp->b_bufobj = NULL; 498 bp->b_flags &= ~B_PAGING; 499 } 500