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: stable/10/sys/vm/vm_pager.c 311645 2017-01-07 12:04:30Z kib $"); 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_kern.h> 82#include <vm/vm_object.h> 83#include <vm/vm_page.h> 84#include <vm/vm_pager.h> 85#include <vm/vm_extern.h> 86 87int cluster_pbuf_freecnt = -1; /* unlimited to begin with */ 88 89static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int); 90static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t, 91 vm_ooffset_t, struct ucred *); 92static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *); 93static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *); 94static void dead_pager_dealloc(vm_object_t); 95 96static int 97dead_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 106static vm_object_t 107dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, 108 vm_ooffset_t off, struct ucred *cred) 109{ 110 111 return (NULL); 112} 113 114static void 115dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, 116 int flags, int *rtvals) 117{ 118 int i; 119 120 for (i = 0; i < count; i++) 121 rtvals[i] = VM_PAGER_AGAIN; 122} 123 124static int 125dead_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *prev, int *next) 126{ 127 128 if (prev != NULL) 129 *prev = 0; 130 if (next != NULL) 131 *next = 0; 132 return (FALSE); 133} 134 135static void 136dead_pager_dealloc(vm_object_t object) 137{ 138 139} 140 141static struct pagerops deadpagerops = { 142 .pgo_alloc = dead_pager_alloc, 143 .pgo_dealloc = dead_pager_dealloc, 144 .pgo_getpages = dead_pager_getpages, 145 .pgo_putpages = dead_pager_putpages, 146 .pgo_haspage = dead_pager_haspage, 147}; 148 149struct pagerops *pagertab[] = { 150 &defaultpagerops, /* OBJT_DEFAULT */ 151 &swappagerops, /* OBJT_SWAP */ 152 &vnodepagerops, /* OBJT_VNODE */ 153 &devicepagerops, /* OBJT_DEVICE */ 154 &physpagerops, /* OBJT_PHYS */ 155 &deadpagerops, /* OBJT_DEAD */ 156 &sgpagerops, /* OBJT_SG */ 157 &mgtdevicepagerops, /* OBJT_MGTDEVICE */ 158}; 159 160static const int npagers = sizeof(pagertab) / sizeof(pagertab[0]); 161 162/* 163 * Kernel address space for mapping pages. 164 * Used by pagers where KVAs are needed for IO. 165 * 166 * XXX needs to be large enough to support the number of pending async 167 * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size 168 * (MAXPHYS == 64k) if you want to get the most efficiency. 169 */ 170struct mtx_padalign pbuf_mtx; 171static TAILQ_HEAD(swqueue, buf) bswlist; 172static int bswneeded; 173vm_offset_t swapbkva; /* swap buffers kva */ 174 175void 176vm_pager_init(void) 177{ 178 struct pagerops **pgops; 179 180 TAILQ_INIT(&bswlist); 181 /* 182 * Initialize known pagers 183 */ 184 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++) 185 if ((*pgops)->pgo_init != NULL) 186 (*(*pgops)->pgo_init)(); 187} 188 189void 190vm_pager_bufferinit(void) 191{ 192 struct buf *bp; 193 int i; 194 195 mtx_init(&pbuf_mtx, "pbuf mutex", NULL, MTX_DEF); 196 bp = swbuf; 197 /* 198 * Now set up swap and physical I/O buffer headers. 199 */ 200 for (i = 0; i < nswbuf; i++, bp++) { 201 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 202 BUF_LOCKINIT(bp); 203 LIST_INIT(&bp->b_dep); 204 bp->b_rcred = bp->b_wcred = NOCRED; 205 bp->b_xflags = 0; 206 } 207 208 cluster_pbuf_freecnt = nswbuf / 2; 209 vnode_pbuf_freecnt = nswbuf / 2 + 1; 210} 211 212/* 213 * Allocate an instance of a pager of the given type. 214 * Size, protection and offset parameters are passed in for pagers that 215 * need to perform page-level validation (e.g. the device pager). 216 */ 217vm_object_t 218vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, 219 vm_prot_t prot, vm_ooffset_t off, struct ucred *cred) 220{ 221 vm_object_t ret; 222 struct pagerops *ops; 223 224 ops = pagertab[type]; 225 if (ops) 226 ret = (*ops->pgo_alloc)(handle, size, prot, off, cred); 227 else 228 ret = NULL; 229 return (ret); 230} 231 232/* 233 * The object must be locked. 234 */ 235void 236vm_pager_deallocate(vm_object_t object) 237{ 238 239 VM_OBJECT_ASSERT_WLOCKED(object); 240 (*pagertab[object->type]->pgo_dealloc) (object); 241} 242 243/* 244 * vm_pager_get_pages() - inline, see vm/vm_pager.h 245 * vm_pager_put_pages() - inline, see vm/vm_pager.h 246 * vm_pager_has_page() - inline, see vm/vm_pager.h 247 */ 248 249/* 250 * Search the specified pager object list for an object with the 251 * specified handle. If an object with the specified handle is found, 252 * increase its reference count and return it. Otherwise, return NULL. 253 * 254 * The pager object list must be locked. 255 */ 256vm_object_t 257vm_pager_object_lookup(struct pagerlst *pg_list, void *handle) 258{ 259 vm_object_t object; 260 261 TAILQ_FOREACH(object, pg_list, pager_object_list) { 262 if (object->handle == handle) { 263 VM_OBJECT_WLOCK(object); 264 if ((object->flags & OBJ_DEAD) == 0) { 265 vm_object_reference_locked(object); 266 VM_OBJECT_WUNLOCK(object); 267 break; 268 } 269 VM_OBJECT_WUNLOCK(object); 270 } 271 } 272 return (object); 273} 274 275/* 276 * initialize a physical buffer 277 */ 278 279/* 280 * XXX This probably belongs in vfs_bio.c 281 */ 282static void 283initpbuf(struct buf *bp) 284{ 285 286 KASSERT(bp->b_bufobj == NULL, ("initpbuf with bufobj")); 287 KASSERT(bp->b_vp == NULL, ("initpbuf with vp")); 288 bp->b_rcred = NOCRED; 289 bp->b_wcred = NOCRED; 290 bp->b_qindex = 0; /* On no queue (QUEUE_NONE) */ 291 bp->b_saveaddr = (caddr_t)(MAXPHYS * (bp - swbuf)) + swapbkva; 292 bp->b_data = bp->b_saveaddr; 293 bp->b_kvabase = bp->b_saveaddr; 294 bp->b_kvasize = MAXPHYS; 295 bp->b_xflags = 0; 296 bp->b_flags = 0; 297 bp->b_ioflags = 0; 298 bp->b_iodone = NULL; 299 bp->b_error = 0; 300 BUF_LOCK(bp, LK_EXCLUSIVE, NULL); 301} 302 303/* 304 * allocate a physical buffer 305 * 306 * There are a limited number (nswbuf) of physical buffers. We need 307 * to make sure that no single subsystem is able to hog all of them, 308 * so each subsystem implements a counter which is typically initialized 309 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and 310 * increments it on release, and blocks if the counter hits zero. A 311 * subsystem may initialize the counter to -1 to disable the feature, 312 * but it must still be sure to match up all uses of getpbuf() with 313 * relpbuf() using the same variable. 314 * 315 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 316 * relatively soon when the rest of the subsystems get smart about it. XXX 317 */ 318struct buf * 319getpbuf(int *pfreecnt) 320{ 321 struct buf *bp; 322 323 mtx_lock(&pbuf_mtx); 324 for (;;) { 325 if (pfreecnt != NULL) { 326 while (*pfreecnt == 0) { 327 msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0); 328 } 329 } 330 331 /* get a bp from the swap buffer header pool */ 332 if ((bp = TAILQ_FIRST(&bswlist)) != NULL) 333 break; 334 335 bswneeded = 1; 336 msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0); 337 /* loop in case someone else grabbed one */ 338 } 339 TAILQ_REMOVE(&bswlist, bp, b_freelist); 340 if (pfreecnt) 341 --*pfreecnt; 342 mtx_unlock(&pbuf_mtx); 343 initpbuf(bp); 344 return (bp); 345} 346 347/* 348 * allocate a physical buffer, if one is available. 349 * 350 * Note that there is no NULL hack here - all subsystems using this 351 * call understand how to use pfreecnt. 352 */ 353struct buf * 354trypbuf(int *pfreecnt) 355{ 356 struct buf *bp; 357 358 mtx_lock(&pbuf_mtx); 359 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) { 360 mtx_unlock(&pbuf_mtx); 361 return NULL; 362 } 363 TAILQ_REMOVE(&bswlist, bp, b_freelist); 364 --*pfreecnt; 365 mtx_unlock(&pbuf_mtx); 366 initpbuf(bp); 367 return (bp); 368} 369 370/* 371 * release a physical buffer 372 * 373 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 374 * relatively soon when the rest of the subsystems get smart about it. XXX 375 */ 376void 377relpbuf(struct buf *bp, int *pfreecnt) 378{ 379 380 if (bp->b_rcred != NOCRED) { 381 crfree(bp->b_rcred); 382 bp->b_rcred = NOCRED; 383 } 384 if (bp->b_wcred != NOCRED) { 385 crfree(bp->b_wcred); 386 bp->b_wcred = NOCRED; 387 } 388 389 KASSERT(bp->b_vp == NULL, ("relpbuf with vp")); 390 KASSERT(bp->b_bufobj == NULL, ("relpbuf with bufobj")); 391 392 BUF_UNLOCK(bp); 393 394 mtx_lock(&pbuf_mtx); 395 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 396 397 if (bswneeded) { 398 bswneeded = 0; 399 wakeup(&bswneeded); 400 } 401 if (pfreecnt) { 402 if (++*pfreecnt == 1) 403 wakeup(pfreecnt); 404 } 405 mtx_unlock(&pbuf_mtx); 406} 407 408/* 409 * Associate a p-buffer with a vnode. 410 * 411 * Also sets B_PAGING flag to indicate that vnode is not fully associated 412 * with the buffer. i.e. the bp has not been linked into the vnode or 413 * ref-counted. 414 */ 415void 416pbgetvp(struct vnode *vp, struct buf *bp) 417{ 418 419 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 420 KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)")); 421 422 bp->b_vp = vp; 423 bp->b_flags |= B_PAGING; 424 bp->b_bufobj = &vp->v_bufobj; 425} 426 427/* 428 * Associate a p-buffer with a vnode. 429 * 430 * Also sets B_PAGING flag to indicate that vnode is not fully associated 431 * with the buffer. i.e. the bp has not been linked into the vnode or 432 * ref-counted. 433 */ 434void 435pbgetbo(struct bufobj *bo, struct buf *bp) 436{ 437 438 KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)")); 439 KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)")); 440 441 bp->b_flags |= B_PAGING; 442 bp->b_bufobj = bo; 443} 444 445/* 446 * Disassociate a p-buffer from a vnode. 447 */ 448void 449pbrelvp(struct buf *bp) 450{ 451 452 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 453 KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj")); 454 KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0, 455 ("pbrelvp: pager buf on vnode list.")); 456 457 bp->b_vp = NULL; 458 bp->b_bufobj = NULL; 459 bp->b_flags &= ~B_PAGING; 460} 461 462/* 463 * Disassociate a p-buffer from a bufobj. 464 */ 465void 466pbrelbo(struct buf *bp) 467{ 468 469 KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode")); 470 KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj")); 471 KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0, 472 ("pbrelbo: pager buf on vnode list.")); 473 474 bp->b_bufobj = NULL; 475 bp->b_flags &= ~B_PAGING; 476} 477