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 #include <sys/rwlock.h> 78 79 #include <vm/vm.h> 80 #include <vm/vm_param.h> 81 #include <vm/vm_object.h> 82 #include <vm/vm_page.h> 83 #include <vm/vm_pager.h> 84 #include <vm/vm_extern.h> 85 86 int cluster_pbuf_freecnt = -1; /* unlimited to begin with */ 87 88 static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int); 89 static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t, 90 vm_ooffset_t, struct ucred *); 91 static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *); 92 static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *); 93 static void dead_pager_dealloc(vm_object_t); 94 95 static int 96 dead_pager_getpages(obj, ma, count, req) 97 vm_object_t obj; 98 vm_page_t *ma; 99 int count; 100 int req; 101 { 102 return VM_PAGER_FAIL; 103 } 104 105 static vm_object_t 106 dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, 107 vm_ooffset_t off, struct ucred *cred) 108 { 109 return NULL; 110 } 111 112 static void 113 dead_pager_putpages(object, m, count, flags, rtvals) 114 vm_object_t object; 115 vm_page_t *m; 116 int count; 117 int flags; 118 int *rtvals; 119 { 120 int i; 121 122 for (i = 0; i < count; i++) { 123 rtvals[i] = VM_PAGER_AGAIN; 124 } 125 } 126 127 static int 128 dead_pager_haspage(object, pindex, prev, next) 129 vm_object_t object; 130 vm_pindex_t pindex; 131 int *prev; 132 int *next; 133 { 134 if (prev) 135 *prev = 0; 136 if (next) 137 *next = 0; 138 return FALSE; 139 } 140 141 static void 142 dead_pager_dealloc(object) 143 vm_object_t object; 144 { 145 return; 146 } 147 148 static struct pagerops deadpagerops = { 149 .pgo_alloc = dead_pager_alloc, 150 .pgo_dealloc = dead_pager_dealloc, 151 .pgo_getpages = dead_pager_getpages, 152 .pgo_putpages = dead_pager_putpages, 153 .pgo_haspage = dead_pager_haspage, 154 }; 155 156 struct pagerops *pagertab[] = { 157 &defaultpagerops, /* OBJT_DEFAULT */ 158 &swappagerops, /* OBJT_SWAP */ 159 &vnodepagerops, /* OBJT_VNODE */ 160 &devicepagerops, /* OBJT_DEVICE */ 161 &physpagerops, /* OBJT_PHYS */ 162 &deadpagerops, /* OBJT_DEAD */ 163 &sgpagerops, /* OBJT_SG */ 164 &mgtdevicepagerops, /* OBJT_MGTDEVICE */ 165 }; 166 167 static const int npagers = sizeof(pagertab) / sizeof(pagertab[0]); 168 169 /* 170 * Kernel address space for mapping pages. 171 * Used by pagers where KVAs are needed for IO. 172 * 173 * XXX needs to be large enough to support the number of pending async 174 * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size 175 * (MAXPHYS == 64k) if you want to get the most efficiency. 176 */ 177 vm_map_t pager_map; 178 static int bswneeded; 179 static vm_offset_t swapbkva; /* swap buffers kva */ 180 struct mtx pbuf_mtx; 181 static TAILQ_HEAD(swqueue, buf) bswlist; 182 183 void 184 vm_pager_init() 185 { 186 struct pagerops **pgops; 187 188 TAILQ_INIT(&bswlist); 189 /* 190 * Initialize known pagers 191 */ 192 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++) 193 if ((*pgops)->pgo_init != NULL) 194 (*(*pgops)->pgo_init) (); 195 } 196 197 void 198 vm_pager_bufferinit() 199 { 200 struct buf *bp; 201 int i; 202 203 mtx_init(&pbuf_mtx, "pbuf mutex", NULL, MTX_DEF); 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_rcred = bp->b_wcred = NOCRED; 213 bp->b_xflags = 0; 214 } 215 216 cluster_pbuf_freecnt = nswbuf / 2; 217 vnode_pbuf_freecnt = nswbuf / 2 + 1; 218 219 swapbkva = kmem_alloc_nofault(pager_map, nswbuf * MAXPHYS); 220 if (!swapbkva) 221 panic("Not enough pager_map VM space for physical buffers"); 222 } 223 224 /* 225 * Allocate an instance of a pager of the given type. 226 * Size, protection and offset parameters are passed in for pagers that 227 * need to perform page-level validation (e.g. the device pager). 228 */ 229 vm_object_t 230 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, 231 vm_prot_t prot, vm_ooffset_t off, struct ucred *cred) 232 { 233 vm_object_t ret; 234 struct pagerops *ops; 235 236 ops = pagertab[type]; 237 if (ops) 238 ret = (*ops->pgo_alloc) (handle, size, prot, off, cred); 239 else 240 ret = NULL; 241 return (ret); 242 } 243 244 /* 245 * The object must be locked. 246 */ 247 void 248 vm_pager_deallocate(object) 249 vm_object_t object; 250 { 251 252 VM_OBJECT_ASSERT_WLOCKED(object); 253 (*pagertab[object->type]->pgo_dealloc) (object); 254 } 255 256 /* 257 * vm_pager_get_pages() - inline, see vm/vm_pager.h 258 * vm_pager_put_pages() - inline, see vm/vm_pager.h 259 * vm_pager_has_page() - inline, see vm/vm_pager.h 260 */ 261 262 /* 263 * Search the specified pager object list for an object with the 264 * specified handle. If an object with the specified handle is found, 265 * increase its reference count and return it. Otherwise, return NULL. 266 * 267 * The pager object list must be locked. 268 */ 269 vm_object_t 270 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle) 271 { 272 vm_object_t object; 273 274 TAILQ_FOREACH(object, pg_list, pager_object_list) { 275 if (object->handle == handle) { 276 VM_OBJECT_WLOCK(object); 277 if ((object->flags & OBJ_DEAD) == 0) { 278 vm_object_reference_locked(object); 279 VM_OBJECT_WUNLOCK(object); 280 break; 281 } 282 VM_OBJECT_WUNLOCK(object); 283 } 284 } 285 return (object); 286 } 287 288 /* 289 * initialize a physical buffer 290 */ 291 292 /* 293 * XXX This probably belongs in vfs_bio.c 294 */ 295 static void 296 initpbuf(struct buf *bp) 297 { 298 KASSERT(bp->b_bufobj == NULL, ("initpbuf with bufobj")); 299 KASSERT(bp->b_vp == NULL, ("initpbuf with vp")); 300 bp->b_rcred = NOCRED; 301 bp->b_wcred = NOCRED; 302 bp->b_qindex = 0; /* On no queue (QUEUE_NONE) */ 303 bp->b_saveaddr = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva; 304 bp->b_data = bp->b_saveaddr; 305 bp->b_kvabase = bp->b_saveaddr; 306 bp->b_kvasize = MAXPHYS; 307 bp->b_xflags = 0; 308 bp->b_flags = 0; 309 bp->b_ioflags = 0; 310 bp->b_iodone = NULL; 311 bp->b_error = 0; 312 BUF_LOCK(bp, LK_EXCLUSIVE, NULL); 313 } 314 315 /* 316 * allocate a physical buffer 317 * 318 * There are a limited number (nswbuf) of physical buffers. We need 319 * to make sure that no single subsystem is able to hog all of them, 320 * so each subsystem implements a counter which is typically initialized 321 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and 322 * increments it on release, and blocks if the counter hits zero. A 323 * subsystem may initialize the counter to -1 to disable the feature, 324 * but it must still be sure to match up all uses of getpbuf() with 325 * relpbuf() using the same variable. 326 * 327 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 328 * relatively soon when the rest of the subsystems get smart about it. XXX 329 */ 330 struct buf * 331 getpbuf(int *pfreecnt) 332 { 333 struct buf *bp; 334 335 mtx_lock(&pbuf_mtx); 336 337 for (;;) { 338 if (pfreecnt) { 339 while (*pfreecnt == 0) { 340 msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0); 341 } 342 } 343 344 /* get a bp from the swap buffer header pool */ 345 if ((bp = TAILQ_FIRST(&bswlist)) != NULL) 346 break; 347 348 bswneeded = 1; 349 msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0); 350 /* loop in case someone else grabbed one */ 351 } 352 TAILQ_REMOVE(&bswlist, bp, b_freelist); 353 if (pfreecnt) 354 --*pfreecnt; 355 mtx_unlock(&pbuf_mtx); 356 357 initpbuf(bp); 358 return bp; 359 } 360 361 /* 362 * allocate a physical buffer, if one is available. 363 * 364 * Note that there is no NULL hack here - all subsystems using this 365 * call understand how to use pfreecnt. 366 */ 367 struct buf * 368 trypbuf(int *pfreecnt) 369 { 370 struct buf *bp; 371 372 mtx_lock(&pbuf_mtx); 373 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) { 374 mtx_unlock(&pbuf_mtx); 375 return NULL; 376 } 377 TAILQ_REMOVE(&bswlist, bp, b_freelist); 378 379 --*pfreecnt; 380 381 mtx_unlock(&pbuf_mtx); 382 383 initpbuf(bp); 384 385 return bp; 386 } 387 388 /* 389 * release a physical buffer 390 * 391 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 392 * relatively soon when the rest of the subsystems get smart about it. XXX 393 */ 394 void 395 relpbuf(struct buf *bp, int *pfreecnt) 396 { 397 398 if (bp->b_rcred != NOCRED) { 399 crfree(bp->b_rcred); 400 bp->b_rcred = NOCRED; 401 } 402 if (bp->b_wcred != NOCRED) { 403 crfree(bp->b_wcred); 404 bp->b_wcred = NOCRED; 405 } 406 407 KASSERT(bp->b_vp == NULL, ("relpbuf with vp")); 408 KASSERT(bp->b_bufobj == NULL, ("relpbuf with bufobj")); 409 410 BUF_UNLOCK(bp); 411 412 mtx_lock(&pbuf_mtx); 413 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 414 415 if (bswneeded) { 416 bswneeded = 0; 417 wakeup(&bswneeded); 418 } 419 if (pfreecnt) { 420 if (++*pfreecnt == 1) 421 wakeup(pfreecnt); 422 } 423 mtx_unlock(&pbuf_mtx); 424 } 425 426 /* 427 * Associate a p-buffer with a vnode. 428 * 429 * Also sets B_PAGING flag to indicate that vnode is not fully associated 430 * with the buffer. i.e. the bp has not been linked into the vnode or 431 * ref-counted. 432 */ 433 void 434 pbgetvp(struct vnode *vp, struct buf *bp) 435 { 436 437 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 438 KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)")); 439 440 bp->b_vp = vp; 441 bp->b_flags |= B_PAGING; 442 bp->b_bufobj = &vp->v_bufobj; 443 } 444 445 /* 446 * Associate a p-buffer with a vnode. 447 * 448 * Also sets B_PAGING flag to indicate that vnode is not fully associated 449 * with the buffer. i.e. the bp has not been linked into the vnode or 450 * ref-counted. 451 */ 452 void 453 pbgetbo(struct bufobj *bo, struct buf *bp) 454 { 455 456 KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)")); 457 KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)")); 458 459 bp->b_flags |= B_PAGING; 460 bp->b_bufobj = bo; 461 } 462 463 /* 464 * Disassociate a p-buffer from a vnode. 465 */ 466 void 467 pbrelvp(struct buf *bp) 468 { 469 470 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 471 KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj")); 472 KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0, 473 ("pbrelvp: pager buf on vnode list.")); 474 475 bp->b_vp = NULL; 476 bp->b_bufobj = NULL; 477 bp->b_flags &= ~B_PAGING; 478 } 479 480 /* 481 * Disassociate a p-buffer from a bufobj. 482 */ 483 void 484 pbrelbo(struct buf *bp) 485 { 486 487 KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode")); 488 KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj")); 489 KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0, 490 ("pbrelbo: pager buf on vnode list.")); 491 492 bp->b_bufobj = NULL; 493 bp->b_flags &= ~B_PAGING; 494 } 495