1 /*- 2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) 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 * 3. 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("$FreeBSD$"); 70 71 #include "opt_param.h" 72 73 #include <sys/param.h> 74 #include <sys/systm.h> 75 #include <sys/kernel.h> 76 #include <sys/vnode.h> 77 #include <sys/bio.h> 78 #include <sys/buf.h> 79 #include <sys/ucred.h> 80 #include <sys/malloc.h> 81 #include <sys/rwlock.h> 82 #include <sys/user.h> 83 84 #include <vm/vm.h> 85 #include <vm/vm_param.h> 86 #include <vm/vm_kern.h> 87 #include <vm/vm_object.h> 88 #include <vm/vm_page.h> 89 #include <vm/vm_pager.h> 90 #include <vm/vm_extern.h> 91 #include <vm/uma.h> 92 93 uma_zone_t pbuf_zone; 94 static int pbuf_init(void *, int, int); 95 static int pbuf_ctor(void *, int, void *, int); 96 static void pbuf_dtor(void *, int, void *); 97 98 static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *); 99 static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t, 100 vm_ooffset_t, struct ucred *); 101 static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *); 102 static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *); 103 static void dead_pager_dealloc(vm_object_t); 104 static void dead_pager_getvp(vm_object_t, struct vnode **, bool *); 105 106 static int 107 dead_pager_getpages(vm_object_t obj, vm_page_t *ma, int count, int *rbehind, 108 int *rahead) 109 { 110 111 return (VM_PAGER_FAIL); 112 } 113 114 static vm_object_t 115 dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, 116 vm_ooffset_t off, struct ucred *cred) 117 { 118 119 return (NULL); 120 } 121 122 static void 123 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, 124 int flags, int *rtvals) 125 { 126 int i; 127 128 for (i = 0; i < count; i++) 129 rtvals[i] = VM_PAGER_AGAIN; 130 } 131 132 static int 133 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *prev, int *next) 134 { 135 136 if (prev != NULL) 137 *prev = 0; 138 if (next != NULL) 139 *next = 0; 140 return (FALSE); 141 } 142 143 static void 144 dead_pager_dealloc(vm_object_t object) 145 { 146 147 } 148 149 static void 150 dead_pager_getvp(vm_object_t object, struct vnode **vpp, bool *vp_heldp) 151 { 152 /* 153 * For OBJT_DEAD objects, v_writecount was handled in 154 * vnode_pager_dealloc(). 155 */ 156 } 157 158 static const struct pagerops deadpagerops = { 159 .pgo_kvme_type = KVME_TYPE_DEAD, 160 .pgo_alloc = dead_pager_alloc, 161 .pgo_dealloc = dead_pager_dealloc, 162 .pgo_getpages = dead_pager_getpages, 163 .pgo_putpages = dead_pager_putpages, 164 .pgo_haspage = dead_pager_haspage, 165 .pgo_getvp = dead_pager_getvp, 166 }; 167 168 const struct pagerops *pagertab[16] __read_mostly = { 169 [OBJT_DEFAULT] = &defaultpagerops, 170 [OBJT_SWAP] = &swappagerops, 171 [OBJT_VNODE] = &vnodepagerops, 172 [OBJT_DEVICE] = &devicepagerops, 173 [OBJT_PHYS] = &physpagerops, 174 [OBJT_DEAD] = &deadpagerops, 175 [OBJT_SG] = &sgpagerops, 176 [OBJT_MGTDEVICE] = &mgtdevicepagerops, 177 }; 178 static struct mtx pagertab_lock; 179 180 void 181 vm_pager_init(void) 182 { 183 const struct pagerops **pgops; 184 int i; 185 186 mtx_init(&pagertab_lock, "dynpag", NULL, MTX_DEF); 187 188 /* 189 * Initialize known pagers 190 */ 191 for (i = 0; i < OBJT_FIRST_DYN; i++) { 192 pgops = &pagertab[i]; 193 if ((*pgops)->pgo_init != NULL) 194 (*(*pgops)->pgo_init)(); 195 } 196 } 197 198 static int nswbuf_max; 199 200 void 201 vm_pager_bufferinit(void) 202 { 203 204 /* Main zone for paging bufs. */ 205 pbuf_zone = uma_zcreate("pbuf", 206 sizeof(struct buf) + PBUF_PAGES * sizeof(vm_page_t), 207 pbuf_ctor, pbuf_dtor, pbuf_init, NULL, UMA_ALIGN_CACHE, 208 UMA_ZONE_NOFREE); 209 /* Few systems may still use this zone directly, so it needs a limit. */ 210 nswbuf_max += uma_zone_set_max(pbuf_zone, NSWBUF_MIN); 211 } 212 213 uma_zone_t 214 pbuf_zsecond_create(const char *name, int max) 215 { 216 uma_zone_t zone; 217 218 zone = uma_zsecond_create(name, pbuf_ctor, pbuf_dtor, NULL, NULL, 219 pbuf_zone); 220 221 #ifdef KMSAN 222 /* 223 * Shrink the size of the pbuf pools if KMSAN is enabled, otherwise the 224 * shadows of the large KVA allocations eat up too much memory. 225 */ 226 max /= 3; 227 #endif 228 229 /* 230 * uma_prealloc() rounds up to items per slab. If we would prealloc 231 * immediately on every pbuf_zsecond_create(), we may accumulate too 232 * much of difference between hard limit and prealloced items, which 233 * means wasted memory. 234 */ 235 if (nswbuf_max > 0) 236 nswbuf_max += uma_zone_set_max(zone, max); 237 else 238 uma_prealloc(pbuf_zone, uma_zone_set_max(zone, max)); 239 240 return (zone); 241 } 242 243 static void 244 pbuf_prealloc(void *arg __unused) 245 { 246 247 uma_prealloc(pbuf_zone, nswbuf_max); 248 nswbuf_max = -1; 249 } 250 251 SYSINIT(pbuf, SI_SUB_KTHREAD_BUF, SI_ORDER_ANY, pbuf_prealloc, NULL); 252 253 /* 254 * Allocate an instance of a pager of the given type. 255 * Size, protection and offset parameters are passed in for pagers that 256 * need to perform page-level validation (e.g. the device pager). 257 */ 258 vm_object_t 259 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, 260 vm_prot_t prot, vm_ooffset_t off, struct ucred *cred) 261 { 262 MPASS(type < nitems(pagertab)); 263 264 return ((*pagertab[type]->pgo_alloc)(handle, size, prot, off, cred)); 265 } 266 267 /* 268 * The object must be locked. 269 */ 270 void 271 vm_pager_deallocate(vm_object_t object) 272 { 273 274 VM_OBJECT_ASSERT_WLOCKED(object); 275 MPASS(object->type < nitems(pagertab)); 276 (*pagertab[object->type]->pgo_dealloc) (object); 277 } 278 279 static void 280 vm_pager_assert_in(vm_object_t object, vm_page_t *m, int count) 281 { 282 #ifdef INVARIANTS 283 284 /* 285 * All pages must be consecutive, busied, not mapped, not fully valid, 286 * not dirty and belong to the proper object. Some pages may be the 287 * bogus page, but the first and last pages must be a real ones. 288 */ 289 290 VM_OBJECT_ASSERT_UNLOCKED(object); 291 VM_OBJECT_ASSERT_PAGING(object); 292 KASSERT(count > 0, ("%s: 0 count", __func__)); 293 for (int i = 0 ; i < count; i++) { 294 if (m[i] == bogus_page) { 295 KASSERT(i != 0 && i != count - 1, 296 ("%s: page %d is the bogus page", __func__, i)); 297 continue; 298 } 299 vm_page_assert_xbusied(m[i]); 300 KASSERT(!pmap_page_is_mapped(m[i]), 301 ("%s: page %p is mapped", __func__, m[i])); 302 KASSERT(m[i]->valid != VM_PAGE_BITS_ALL, 303 ("%s: request for a valid page %p", __func__, m[i])); 304 KASSERT(m[i]->dirty == 0, 305 ("%s: page %p is dirty", __func__, m[i])); 306 KASSERT(m[i]->object == object, 307 ("%s: wrong object %p/%p", __func__, object, m[i]->object)); 308 KASSERT(m[i]->pindex == m[0]->pindex + i, 309 ("%s: page %p isn't consecutive", __func__, m[i])); 310 } 311 #endif 312 } 313 314 /* 315 * Page in the pages for the object using its associated pager. 316 * The requested page must be fully valid on successful return. 317 */ 318 int 319 vm_pager_get_pages(vm_object_t object, vm_page_t *m, int count, int *rbehind, 320 int *rahead) 321 { 322 #ifdef INVARIANTS 323 vm_pindex_t pindex = m[0]->pindex; 324 #endif 325 int r; 326 327 MPASS(object->type < nitems(pagertab)); 328 vm_pager_assert_in(object, m, count); 329 330 r = (*pagertab[object->type]->pgo_getpages)(object, m, count, rbehind, 331 rahead); 332 if (r != VM_PAGER_OK) 333 return (r); 334 335 for (int i = 0; i < count; i++) { 336 /* 337 * If pager has replaced a page, assert that it had 338 * updated the array. 339 */ 340 #ifdef INVARIANTS 341 KASSERT(m[i] == vm_page_relookup(object, pindex++), 342 ("%s: mismatch page %p pindex %ju", __func__, 343 m[i], (uintmax_t )pindex - 1)); 344 #endif 345 346 /* 347 * Zero out partially filled data. 348 */ 349 if (m[i]->valid != VM_PAGE_BITS_ALL) 350 vm_page_zero_invalid(m[i], TRUE); 351 } 352 return (VM_PAGER_OK); 353 } 354 355 int 356 vm_pager_get_pages_async(vm_object_t object, vm_page_t *m, int count, 357 int *rbehind, int *rahead, pgo_getpages_iodone_t iodone, void *arg) 358 { 359 360 MPASS(object->type < nitems(pagertab)); 361 vm_pager_assert_in(object, m, count); 362 363 return ((*pagertab[object->type]->pgo_getpages_async)(object, m, 364 count, rbehind, rahead, iodone, arg)); 365 } 366 367 /* 368 * vm_pager_put_pages() - inline, see vm/vm_pager.h 369 * vm_pager_has_page() - inline, see vm/vm_pager.h 370 */ 371 372 /* 373 * Search the specified pager object list for an object with the 374 * specified handle. If an object with the specified handle is found, 375 * increase its reference count and return it. Otherwise, return NULL. 376 * 377 * The pager object list must be locked. 378 */ 379 vm_object_t 380 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle) 381 { 382 vm_object_t object; 383 384 TAILQ_FOREACH(object, pg_list, pager_object_list) { 385 if (object->handle == handle) { 386 VM_OBJECT_WLOCK(object); 387 if ((object->flags & OBJ_DEAD) == 0) { 388 vm_object_reference_locked(object); 389 VM_OBJECT_WUNLOCK(object); 390 break; 391 } 392 VM_OBJECT_WUNLOCK(object); 393 } 394 } 395 return (object); 396 } 397 398 int 399 vm_pager_alloc_dyn_type(struct pagerops *ops, int base_type) 400 { 401 int res; 402 403 mtx_lock(&pagertab_lock); 404 MPASS(base_type == -1 || 405 (base_type >= OBJT_DEFAULT && base_type < nitems(pagertab))); 406 for (res = OBJT_FIRST_DYN; res < nitems(pagertab); res++) { 407 if (pagertab[res] == NULL) 408 break; 409 } 410 if (res == nitems(pagertab)) { 411 mtx_unlock(&pagertab_lock); 412 return (-1); 413 } 414 if (base_type != -1) { 415 MPASS(pagertab[base_type] != NULL); 416 #define FIX(n) \ 417 if (ops->pgo_##n == NULL) \ 418 ops->pgo_##n = pagertab[base_type]->pgo_##n 419 FIX(init); 420 FIX(alloc); 421 FIX(dealloc); 422 FIX(getpages); 423 FIX(getpages_async); 424 FIX(putpages); 425 FIX(haspage); 426 FIX(populate); 427 FIX(pageunswapped); 428 FIX(update_writecount); 429 FIX(release_writecount); 430 FIX(set_writeable_dirty); 431 FIX(mightbedirty); 432 FIX(getvp); 433 FIX(freespace); 434 #undef FIX 435 } 436 pagertab[res] = ops; /* XXXKIB should be rel, but acq is too much */ 437 mtx_unlock(&pagertab_lock); 438 return (res); 439 } 440 441 void 442 vm_pager_free_dyn_type(objtype_t type) 443 { 444 MPASS(type >= OBJT_FIRST_DYN && type < nitems(pagertab)); 445 446 mtx_lock(&pagertab_lock); 447 MPASS(pagertab[type] != NULL); 448 pagertab[type] = NULL; 449 mtx_unlock(&pagertab_lock); 450 } 451 452 static int 453 pbuf_ctor(void *mem, int size, void *arg, int flags) 454 { 455 struct buf *bp = mem; 456 457 bp->b_vp = NULL; 458 bp->b_bufobj = NULL; 459 460 /* copied from initpbuf() */ 461 bp->b_rcred = NOCRED; 462 bp->b_wcred = NOCRED; 463 bp->b_qindex = 0; /* On no queue (QUEUE_NONE) */ 464 bp->b_data = bp->b_kvabase; 465 bp->b_xflags = 0; 466 bp->b_flags = B_MAXPHYS; 467 bp->b_ioflags = 0; 468 bp->b_iodone = NULL; 469 bp->b_error = 0; 470 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL); 471 472 return (0); 473 } 474 475 static void 476 pbuf_dtor(void *mem, int size, void *arg) 477 { 478 struct buf *bp = mem; 479 480 if (bp->b_rcred != NOCRED) { 481 crfree(bp->b_rcred); 482 bp->b_rcred = NOCRED; 483 } 484 if (bp->b_wcred != NOCRED) { 485 crfree(bp->b_wcred); 486 bp->b_wcred = NOCRED; 487 } 488 489 BUF_UNLOCK(bp); 490 } 491 492 static const char pbuf_wmesg[] = "pbufwait"; 493 494 static int 495 pbuf_init(void *mem, int size, int flags) 496 { 497 struct buf *bp = mem; 498 499 bp->b_kvabase = (void *)kva_alloc(ptoa(PBUF_PAGES)); 500 if (bp->b_kvabase == NULL) 501 return (ENOMEM); 502 bp->b_kvasize = ptoa(PBUF_PAGES); 503 BUF_LOCKINIT(bp, pbuf_wmesg); 504 LIST_INIT(&bp->b_dep); 505 bp->b_rcred = bp->b_wcred = NOCRED; 506 bp->b_xflags = 0; 507 508 return (0); 509 } 510 511 /* 512 * Associate a p-buffer with a vnode. 513 * 514 * Also sets B_PAGING flag to indicate that vnode is not fully associated 515 * with the buffer. i.e. the bp has not been linked into the vnode or 516 * ref-counted. 517 */ 518 void 519 pbgetvp(struct vnode *vp, struct buf *bp) 520 { 521 522 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 523 KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)")); 524 525 bp->b_vp = vp; 526 bp->b_flags |= B_PAGING; 527 bp->b_bufobj = &vp->v_bufobj; 528 } 529 530 /* 531 * Associate a p-buffer with a vnode. 532 * 533 * Also sets B_PAGING flag to indicate that vnode is not fully associated 534 * with the buffer. i.e. the bp has not been linked into the vnode or 535 * ref-counted. 536 */ 537 void 538 pbgetbo(struct bufobj *bo, struct buf *bp) 539 { 540 541 KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)")); 542 KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)")); 543 544 bp->b_flags |= B_PAGING; 545 bp->b_bufobj = bo; 546 } 547 548 /* 549 * Disassociate a p-buffer from a vnode. 550 */ 551 void 552 pbrelvp(struct buf *bp) 553 { 554 555 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 556 KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj")); 557 KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0, 558 ("pbrelvp: pager buf on vnode list.")); 559 560 bp->b_vp = NULL; 561 bp->b_bufobj = NULL; 562 bp->b_flags &= ~B_PAGING; 563 } 564 565 /* 566 * Disassociate a p-buffer from a bufobj. 567 */ 568 void 569 pbrelbo(struct buf *bp) 570 { 571 572 KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode")); 573 KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj")); 574 KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0, 575 ("pbrelbo: pager buf on vnode list.")); 576 577 bp->b_bufobj = NULL; 578 bp->b_flags &= ~B_PAGING; 579 } 580 581 void 582 vm_object_set_writeable_dirty(vm_object_t object) 583 { 584 pgo_set_writeable_dirty_t *method; 585 586 MPASS(object->type < nitems(pagertab)); 587 588 method = pagertab[object->type]->pgo_set_writeable_dirty; 589 if (method != NULL) 590 method(object); 591 } 592 593 bool 594 vm_object_mightbedirty(vm_object_t object) 595 { 596 pgo_mightbedirty_t *method; 597 598 MPASS(object->type < nitems(pagertab)); 599 600 method = pagertab[object->type]->pgo_mightbedirty; 601 if (method == NULL) 602 return (false); 603 return (method(object)); 604 } 605 606 /* 607 * Return the kvme type of the given object. 608 * If vpp is not NULL, set it to the object's vm_object_vnode() or NULL. 609 */ 610 int 611 vm_object_kvme_type(vm_object_t object, struct vnode **vpp) 612 { 613 VM_OBJECT_ASSERT_LOCKED(object); 614 MPASS(object->type < nitems(pagertab)); 615 616 if (vpp != NULL) 617 *vpp = vm_object_vnode(object); 618 return (pagertab[object->type]->pgo_kvme_type); 619 } 620