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. 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_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 87 int cluster_pbuf_freecnt = -1; /* unlimited to begin with */ 88 89 struct buf *swbuf; 90 91 static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *); 92 static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t, 93 vm_ooffset_t, struct ucred *); 94 static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *); 95 static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *); 96 static void dead_pager_dealloc(vm_object_t); 97 98 static int 99 dead_pager_getpages(vm_object_t obj, vm_page_t *ma, int count, int *rbehind, 100 int *rahead) 101 { 102 103 return (VM_PAGER_FAIL); 104 } 105 106 static vm_object_t 107 dead_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 } 113 114 static void 115 dead_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 } 123 124 static int 125 dead_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 } 134 135 static void 136 dead_pager_dealloc(vm_object_t object) 137 { 138 139 } 140 141 static 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 }; 148 149 struct 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 }; 159 160 /* 161 * Kernel address space for mapping pages. 162 * Used by pagers where KVAs are needed for IO. 163 * 164 * XXX needs to be large enough to support the number of pending async 165 * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size 166 * (MAXPHYS == 64k) if you want to get the most efficiency. 167 */ 168 struct mtx_padalign __exclusive_cache_line pbuf_mtx; 169 static TAILQ_HEAD(swqueue, buf) bswlist; 170 static int bswneeded; 171 vm_offset_t swapbkva; /* swap buffers kva */ 172 173 void 174 vm_pager_init(void) 175 { 176 struct pagerops **pgops; 177 178 TAILQ_INIT(&bswlist); 179 /* 180 * Initialize known pagers 181 */ 182 for (pgops = pagertab; pgops < &pagertab[nitems(pagertab)]; pgops++) 183 if ((*pgops)->pgo_init != NULL) 184 (*(*pgops)->pgo_init)(); 185 } 186 187 void 188 vm_pager_bufferinit(void) 189 { 190 struct buf *bp; 191 int i; 192 193 mtx_init(&pbuf_mtx, "pbuf mutex", NULL, MTX_DEF); 194 bp = swbuf; 195 /* 196 * Now set up swap and physical I/O buffer headers. 197 */ 198 for (i = 0; i < nswbuf; i++, bp++) { 199 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 200 BUF_LOCKINIT(bp); 201 LIST_INIT(&bp->b_dep); 202 bp->b_rcred = bp->b_wcred = NOCRED; 203 bp->b_xflags = 0; 204 } 205 206 cluster_pbuf_freecnt = nswbuf / 2; 207 vnode_pbuf_freecnt = nswbuf / 2 + 1; 208 vnode_async_pbuf_freecnt = nswbuf / 2; 209 } 210 211 /* 212 * Allocate an instance of a pager of the given type. 213 * Size, protection and offset parameters are passed in for pagers that 214 * need to perform page-level validation (e.g. the device pager). 215 */ 216 vm_object_t 217 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, 218 vm_prot_t prot, vm_ooffset_t off, struct ucred *cred) 219 { 220 vm_object_t ret; 221 struct pagerops *ops; 222 223 ops = pagertab[type]; 224 if (ops) 225 ret = (*ops->pgo_alloc)(handle, size, prot, off, cred); 226 else 227 ret = NULL; 228 return (ret); 229 } 230 231 /* 232 * The object must be locked. 233 */ 234 void 235 vm_pager_deallocate(vm_object_t object) 236 { 237 238 VM_OBJECT_ASSERT_WLOCKED(object); 239 (*pagertab[object->type]->pgo_dealloc) (object); 240 } 241 242 static void 243 vm_pager_assert_in(vm_object_t object, vm_page_t *m, int count) 244 { 245 #ifdef INVARIANTS 246 247 VM_OBJECT_ASSERT_WLOCKED(object); 248 KASSERT(count > 0, ("%s: 0 count", __func__)); 249 /* 250 * All pages must be busied, not mapped, not fully valid, 251 * not dirty and belong to the proper object. 252 */ 253 for (int i = 0 ; i < count; i++) { 254 if (m[i] == bogus_page) 255 continue; 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 */ 273 int 274 vm_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 } 305 306 int 307 vm_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 */ 329 vm_object_t 330 vm_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 */ 355 static void 356 initpbuf(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 buf_track(bp, __func__); 374 } 375 376 /* 377 * allocate a physical buffer 378 * 379 * There are a limited number (nswbuf) of physical buffers. We need 380 * to make sure that no single subsystem is able to hog all of them, 381 * so each subsystem implements a counter which is typically initialized 382 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and 383 * increments it on release, and blocks if the counter hits zero. A 384 * subsystem may initialize the counter to -1 to disable the feature, 385 * but it must still be sure to match up all uses of getpbuf() with 386 * relpbuf() using the same variable. 387 * 388 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 389 * relatively soon when the rest of the subsystems get smart about it. XXX 390 */ 391 struct buf * 392 getpbuf(int *pfreecnt) 393 { 394 struct buf *bp; 395 396 mtx_lock(&pbuf_mtx); 397 for (;;) { 398 if (pfreecnt != NULL) { 399 while (*pfreecnt == 0) { 400 msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0); 401 } 402 } 403 404 /* get a bp from the swap buffer header pool */ 405 if ((bp = TAILQ_FIRST(&bswlist)) != NULL) 406 break; 407 408 bswneeded = 1; 409 msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0); 410 /* loop in case someone else grabbed one */ 411 } 412 TAILQ_REMOVE(&bswlist, bp, b_freelist); 413 if (pfreecnt) 414 --*pfreecnt; 415 mtx_unlock(&pbuf_mtx); 416 initpbuf(bp); 417 return (bp); 418 } 419 420 /* 421 * allocate a physical buffer, if one is available. 422 * 423 * Note that there is no NULL hack here - all subsystems using this 424 * call understand how to use pfreecnt. 425 */ 426 struct buf * 427 trypbuf(int *pfreecnt) 428 { 429 struct buf *bp; 430 431 mtx_lock(&pbuf_mtx); 432 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) { 433 mtx_unlock(&pbuf_mtx); 434 return NULL; 435 } 436 TAILQ_REMOVE(&bswlist, bp, b_freelist); 437 --*pfreecnt; 438 mtx_unlock(&pbuf_mtx); 439 initpbuf(bp); 440 return (bp); 441 } 442 443 /* 444 * release a physical buffer 445 * 446 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 447 * relatively soon when the rest of the subsystems get smart about it. XXX 448 */ 449 void 450 relpbuf(struct buf *bp, int *pfreecnt) 451 { 452 453 if (bp->b_rcred != NOCRED) { 454 crfree(bp->b_rcred); 455 bp->b_rcred = NOCRED; 456 } 457 if (bp->b_wcred != NOCRED) { 458 crfree(bp->b_wcred); 459 bp->b_wcred = NOCRED; 460 } 461 462 KASSERT(bp->b_vp == NULL, ("relpbuf with vp")); 463 KASSERT(bp->b_bufobj == NULL, ("relpbuf with bufobj")); 464 465 buf_track(bp, __func__); 466 BUF_UNLOCK(bp); 467 468 mtx_lock(&pbuf_mtx); 469 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); 470 471 if (bswneeded) { 472 bswneeded = 0; 473 wakeup(&bswneeded); 474 } 475 if (pfreecnt) { 476 if (++*pfreecnt == 1) 477 wakeup(pfreecnt); 478 } 479 mtx_unlock(&pbuf_mtx); 480 } 481 482 /* 483 * Associate a p-buffer with a vnode. 484 * 485 * Also sets B_PAGING flag to indicate that vnode is not fully associated 486 * with the buffer. i.e. the bp has not been linked into the vnode or 487 * ref-counted. 488 */ 489 void 490 pbgetvp(struct vnode *vp, struct buf *bp) 491 { 492 493 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 494 KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)")); 495 496 bp->b_vp = vp; 497 bp->b_flags |= B_PAGING; 498 bp->b_bufobj = &vp->v_bufobj; 499 } 500 501 /* 502 * Associate a p-buffer with a vnode. 503 * 504 * Also sets B_PAGING flag to indicate that vnode is not fully associated 505 * with the buffer. i.e. the bp has not been linked into the vnode or 506 * ref-counted. 507 */ 508 void 509 pbgetbo(struct bufobj *bo, struct buf *bp) 510 { 511 512 KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)")); 513 KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)")); 514 515 bp->b_flags |= B_PAGING; 516 bp->b_bufobj = bo; 517 } 518 519 /* 520 * Disassociate a p-buffer from a vnode. 521 */ 522 void 523 pbrelvp(struct buf *bp) 524 { 525 526 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 527 KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj")); 528 KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0, 529 ("pbrelvp: pager buf on vnode list.")); 530 531 bp->b_vp = NULL; 532 bp->b_bufobj = NULL; 533 bp->b_flags &= ~B_PAGING; 534 } 535 536 /* 537 * Disassociate a p-buffer from a bufobj. 538 */ 539 void 540 pbrelbo(struct buf *bp) 541 { 542 543 KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode")); 544 KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj")); 545 KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0, 546 ("pbrelbo: pager buf on vnode list.")); 547 548 bp->b_bufobj = NULL; 549 bp->b_flags &= ~B_PAGING; 550 } 551