vm_pager.c revision 330897
1/*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * The Mach Operating System project at Carnegie-Mellon University. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from: @(#)vm_pager.c 8.6 (Berkeley) 1/12/94 35 * 36 * 37 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 38 * All rights reserved. 39 * 40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 41 * 42 * Permission to use, copy, modify and distribute this software and 43 * its documentation is hereby granted, provided that both the copyright 44 * notice and this permission notice appear in all copies of the 45 * software, derivative works or modified versions, and any portions 46 * thereof, and that both notices appear in supporting documentation. 47 * 48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 51 * 52 * Carnegie Mellon requests users of this software to return to 53 * 54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 55 * School of Computer Science 56 * Carnegie Mellon University 57 * Pittsburgh PA 15213-3890 58 * 59 * any improvements or extensions that they make and grant Carnegie the 60 * rights to redistribute these changes. 61 */ 62 63/* 64 * Paging space routine stubs. Emulates a matchmaker-like interface 65 * for builtin pagers. 66 */ 67 68#include <sys/cdefs.h> 69__FBSDID("$FreeBSD: stable/11/sys/vm/vm_pager.c 330897 2018-03-14 03:19:51Z eadler $"); 70 71#include <sys/param.h> 72#include <sys/systm.h> 73#include <sys/kernel.h> 74#include <sys/vnode.h> 75#include <sys/bio.h> 76#include <sys/buf.h> 77#include <sys/ucred.h> 78#include <sys/malloc.h> 79#include <sys/rwlock.h> 80 81#include <vm/vm.h> 82#include <vm/vm_param.h> 83#include <vm/vm_kern.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 89int cluster_pbuf_freecnt = -1; /* unlimited to begin with */ 90 91struct buf *swbuf; 92 93static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *); 94static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t, 95 vm_ooffset_t, struct ucred *); 96static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *); 97static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *); 98static void dead_pager_dealloc(vm_object_t); 99 100static int 101dead_pager_getpages(vm_object_t obj, vm_page_t *ma, int count, int *rbehind, 102 int *rahead) 103{ 104 105 return (VM_PAGER_FAIL); 106} 107 108static vm_object_t 109dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, 110 vm_ooffset_t off, struct ucred *cred) 111{ 112 113 return (NULL); 114} 115 116static void 117dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, 118 int flags, int *rtvals) 119{ 120 int i; 121 122 for (i = 0; i < count; i++) 123 rtvals[i] = VM_PAGER_AGAIN; 124} 125 126static int 127dead_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *prev, int *next) 128{ 129 130 if (prev != NULL) 131 *prev = 0; 132 if (next != NULL) 133 *next = 0; 134 return (FALSE); 135} 136 137static void 138dead_pager_dealloc(vm_object_t object) 139{ 140 141} 142 143static struct pagerops deadpagerops = { 144 .pgo_alloc = dead_pager_alloc, 145 .pgo_dealloc = dead_pager_dealloc, 146 .pgo_getpages = dead_pager_getpages, 147 .pgo_putpages = dead_pager_putpages, 148 .pgo_haspage = dead_pager_haspage, 149}; 150 151struct pagerops *pagertab[] = { 152 &defaultpagerops, /* OBJT_DEFAULT */ 153 &swappagerops, /* OBJT_SWAP */ 154 &vnodepagerops, /* OBJT_VNODE */ 155 &devicepagerops, /* OBJT_DEVICE */ 156 &physpagerops, /* OBJT_PHYS */ 157 &deadpagerops, /* OBJT_DEAD */ 158 &sgpagerops, /* OBJT_SG */ 159 &mgtdevicepagerops, /* OBJT_MGTDEVICE */ 160}; 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 __exclusive_cache_line 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[nitems(pagertab)]; 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 vnode_async_pbuf_freecnt = nswbuf / 2; 211} 212 213/* 214 * Allocate an instance of a pager of the given type. 215 * Size, protection and offset parameters are passed in for pagers that 216 * need to perform page-level validation (e.g. the device pager). 217 */ 218vm_object_t 219vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, 220 vm_prot_t prot, vm_ooffset_t off, struct ucred *cred) 221{ 222 vm_object_t ret; 223 struct pagerops *ops; 224 225 ops = pagertab[type]; 226 if (ops) 227 ret = (*ops->pgo_alloc)(handle, size, prot, off, cred); 228 else 229 ret = NULL; 230 return (ret); 231} 232 233/* 234 * The object must be locked. 235 */ 236void 237vm_pager_deallocate(vm_object_t object) 238{ 239 240 VM_OBJECT_ASSERT_WLOCKED(object); 241 (*pagertab[object->type]->pgo_dealloc) (object); 242} 243 244static void 245vm_pager_assert_in(vm_object_t object, vm_page_t *m, int count) 246{ 247#ifdef INVARIANTS 248 249 VM_OBJECT_ASSERT_WLOCKED(object); 250 KASSERT(count > 0, ("%s: 0 count", __func__)); 251 /* 252 * All pages must be busied, not mapped, not fully valid, 253 * not dirty and belong to the proper object. 254 */ 255 for (int i = 0 ; i < count; i++) { 256 vm_page_assert_xbusied(m[i]); 257 KASSERT(!pmap_page_is_mapped(m[i]), 258 ("%s: page %p is mapped", __func__, m[i])); 259 KASSERT(m[i]->valid != VM_PAGE_BITS_ALL, 260 ("%s: request for a valid page %p", __func__, m[i])); 261 KASSERT(m[i]->dirty == 0, 262 ("%s: page %p is dirty", __func__, m[i])); 263 KASSERT(m[i]->object == object, 264 ("%s: wrong object %p/%p", __func__, object, m[i]->object)); 265 } 266#endif 267} 268 269/* 270 * Page in the pages for the object using its associated pager. 271 * The requested page must be fully valid on successful return. 272 */ 273int 274vm_pager_get_pages(vm_object_t object, vm_page_t *m, int count, int *rbehind, 275 int *rahead) 276{ 277#ifdef INVARIANTS 278 vm_pindex_t pindex = m[0]->pindex; 279#endif 280 int r; 281 282 vm_pager_assert_in(object, m, count); 283 284 r = (*pagertab[object->type]->pgo_getpages)(object, m, count, rbehind, 285 rahead); 286 if (r != VM_PAGER_OK) 287 return (r); 288 289 for (int i = 0; i < count; i++) { 290 /* 291 * If pager has replaced a page, assert that it had 292 * updated the array. 293 */ 294 KASSERT(m[i] == vm_page_lookup(object, pindex++), 295 ("%s: mismatch page %p pindex %ju", __func__, 296 m[i], (uintmax_t )pindex - 1)); 297 /* 298 * Zero out partially filled data. 299 */ 300 if (m[i]->valid != VM_PAGE_BITS_ALL) 301 vm_page_zero_invalid(m[i], TRUE); 302 } 303 return (VM_PAGER_OK); 304} 305 306int 307vm_pager_get_pages_async(vm_object_t object, vm_page_t *m, int count, 308 int *rbehind, int *rahead, pgo_getpages_iodone_t iodone, void *arg) 309{ 310 311 vm_pager_assert_in(object, m, count); 312 313 return ((*pagertab[object->type]->pgo_getpages_async)(object, m, 314 count, rbehind, rahead, iodone, arg)); 315} 316 317/* 318 * vm_pager_put_pages() - inline, see vm/vm_pager.h 319 * vm_pager_has_page() - inline, see vm/vm_pager.h 320 */ 321 322/* 323 * Search the specified pager object list for an object with the 324 * specified handle. If an object with the specified handle is found, 325 * increase its reference count and return it. Otherwise, return NULL. 326 * 327 * The pager object list must be locked. 328 */ 329vm_object_t 330vm_pager_object_lookup(struct pagerlst *pg_list, void *handle) 331{ 332 vm_object_t object; 333 334 TAILQ_FOREACH(object, pg_list, pager_object_list) { 335 if (object->handle == handle) { 336 VM_OBJECT_WLOCK(object); 337 if ((object->flags & OBJ_DEAD) == 0) { 338 vm_object_reference_locked(object); 339 VM_OBJECT_WUNLOCK(object); 340 break; 341 } 342 VM_OBJECT_WUNLOCK(object); 343 } 344 } 345 return (object); 346} 347 348/* 349 * initialize a physical buffer 350 */ 351 352/* 353 * XXX This probably belongs in vfs_bio.c 354 */ 355static void 356initpbuf(struct buf *bp) 357{ 358 359 KASSERT(bp->b_bufobj == NULL, ("initpbuf with bufobj")); 360 KASSERT(bp->b_vp == NULL, ("initpbuf with vp")); 361 bp->b_rcred = NOCRED; 362 bp->b_wcred = NOCRED; 363 bp->b_qindex = 0; /* On no queue (QUEUE_NONE) */ 364 bp->b_kvabase = (caddr_t)(MAXPHYS * (bp - swbuf)) + swapbkva; 365 bp->b_data = bp->b_kvabase; 366 bp->b_kvasize = MAXPHYS; 367 bp->b_flags = 0; 368 bp->b_xflags = 0; 369 bp->b_ioflags = 0; 370 bp->b_iodone = NULL; 371 bp->b_error = 0; 372 BUF_LOCK(bp, LK_EXCLUSIVE, NULL); 373} 374 375/* 376 * allocate a physical buffer 377 * 378 * There are a limited number (nswbuf) of physical buffers. We need 379 * to make sure that no single subsystem is able to hog all of them, 380 * so each subsystem implements a counter which is typically initialized 381 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and 382 * increments it on release, and blocks if the counter hits zero. A 383 * subsystem may initialize the counter to -1 to disable the feature, 384 * but it must still be sure to match up all uses of getpbuf() with 385 * relpbuf() using the same variable. 386 * 387 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 388 * relatively soon when the rest of the subsystems get smart about it. XXX 389 */ 390struct buf * 391getpbuf(int *pfreecnt) 392{ 393 struct buf *bp; 394 395 mtx_lock(&pbuf_mtx); 396 for (;;) { 397 if (pfreecnt != NULL) { 398 while (*pfreecnt == 0) { 399 msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0); 400 } 401 } 402 403 /* get a bp from the swap buffer header pool */ 404 if ((bp = TAILQ_FIRST(&bswlist)) != NULL) 405 break; 406 407 bswneeded = 1; 408 msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0); 409 /* loop in case someone else grabbed one */ 410 } 411 TAILQ_REMOVE(&bswlist, bp, b_freelist); 412 if (pfreecnt) 413 --*pfreecnt; 414 mtx_unlock(&pbuf_mtx); 415 initpbuf(bp); 416 return (bp); 417} 418 419/* 420 * allocate a physical buffer, if one is available. 421 * 422 * Note that there is no NULL hack here - all subsystems using this 423 * call understand how to use pfreecnt. 424 */ 425struct buf * 426trypbuf(int *pfreecnt) 427{ 428 struct buf *bp; 429 430 mtx_lock(&pbuf_mtx); 431 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) { 432 mtx_unlock(&pbuf_mtx); 433 return NULL; 434 } 435 TAILQ_REMOVE(&bswlist, bp, b_freelist); 436 --*pfreecnt; 437 mtx_unlock(&pbuf_mtx); 438 initpbuf(bp); 439 return (bp); 440} 441 442/* 443 * release a physical buffer 444 * 445 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 446 * relatively soon when the rest of the subsystems get smart about it. XXX 447 */ 448void 449relpbuf(struct buf *bp, int *pfreecnt) 450{ 451 452 if (bp->b_rcred != NOCRED) { 453 crfree(bp->b_rcred); 454 bp->b_rcred = NOCRED; 455 } 456 if (bp->b_wcred != NOCRED) { 457 crfree(bp->b_wcred); 458 bp->b_wcred = NOCRED; 459 } 460 461 KASSERT(bp->b_vp == NULL, ("relpbuf with vp")); 462 KASSERT(bp->b_bufobj == NULL, ("relpbuf with bufobj")); 463 464 BUF_UNLOCK(bp); 465 466 mtx_lock(&pbuf_mtx); 467 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 468 469 if (bswneeded) { 470 bswneeded = 0; 471 wakeup(&bswneeded); 472 } 473 if (pfreecnt) { 474 if (++*pfreecnt == 1) 475 wakeup(pfreecnt); 476 } 477 mtx_unlock(&pbuf_mtx); 478} 479 480/* 481 * Associate a p-buffer with a vnode. 482 * 483 * Also sets B_PAGING flag to indicate that vnode is not fully associated 484 * with the buffer. i.e. the bp has not been linked into the vnode or 485 * ref-counted. 486 */ 487void 488pbgetvp(struct vnode *vp, struct buf *bp) 489{ 490 491 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 492 KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)")); 493 494 bp->b_vp = vp; 495 bp->b_flags |= B_PAGING; 496 bp->b_bufobj = &vp->v_bufobj; 497} 498 499/* 500 * Associate a p-buffer with a vnode. 501 * 502 * Also sets B_PAGING flag to indicate that vnode is not fully associated 503 * with the buffer. i.e. the bp has not been linked into the vnode or 504 * ref-counted. 505 */ 506void 507pbgetbo(struct bufobj *bo, struct buf *bp) 508{ 509 510 KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)")); 511 KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)")); 512 513 bp->b_flags |= B_PAGING; 514 bp->b_bufobj = bo; 515} 516 517/* 518 * Disassociate a p-buffer from a vnode. 519 */ 520void 521pbrelvp(struct buf *bp) 522{ 523 524 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 525 KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj")); 526 KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0, 527 ("pbrelvp: pager buf on vnode list.")); 528 529 bp->b_vp = NULL; 530 bp->b_bufobj = NULL; 531 bp->b_flags &= ~B_PAGING; 532} 533 534/* 535 * Disassociate a p-buffer from a bufobj. 536 */ 537void 538pbrelbo(struct buf *bp) 539{ 540 541 KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode")); 542 KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj")); 543 KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0, 544 ("pbrelbo: pager buf on vnode list.")); 545 546 bp->b_bufobj = NULL; 547 bp->b_flags &= ~B_PAGING; 548} 549