xref: /freebsd/sys/vm/vm_pager.c (revision 47dd1d1b619cc035b82b49a91a25544309ff95ae)
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 <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 
89 int cluster_pbuf_freecnt = -1;	/* unlimited to begin with */
90 
91 struct buf *swbuf;
92 
93 static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *);
94 static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
95     vm_ooffset_t, struct ucred *);
96 static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
97 static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
98 static void dead_pager_dealloc(vm_object_t);
99 
100 static int
101 dead_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 
108 static vm_object_t
109 dead_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 
116 static void
117 dead_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 
126 static int
127 dead_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 
137 static void
138 dead_pager_dealloc(vm_object_t object)
139 {
140 
141 }
142 
143 static 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 
151 struct 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  */
170 struct mtx_padalign __exclusive_cache_line pbuf_mtx;
171 static TAILQ_HEAD(swqueue, buf) bswlist;
172 static int bswneeded;
173 vm_offset_t swapbkva;		/* swap buffers kva */
174 
175 void
176 vm_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 
189 void
190 vm_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  */
218 vm_object_t
219 vm_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  */
236 void
237 vm_pager_deallocate(vm_object_t object)
238 {
239 
240 	VM_OBJECT_ASSERT_WLOCKED(object);
241 	(*pagertab[object->type]->pgo_dealloc) (object);
242 }
243 
244 static void
245 vm_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 		if (m[i] == bogus_page)
257 			continue;
258 		vm_page_assert_xbusied(m[i]);
259 		KASSERT(!pmap_page_is_mapped(m[i]),
260 		    ("%s: page %p is mapped", __func__, m[i]));
261 		KASSERT(m[i]->valid != VM_PAGE_BITS_ALL,
262 		    ("%s: request for a valid page %p", __func__, m[i]));
263 		KASSERT(m[i]->dirty == 0,
264 		    ("%s: page %p is dirty", __func__, m[i]));
265 		KASSERT(m[i]->object == object,
266 		    ("%s: wrong object %p/%p", __func__, object, m[i]->object));
267 	}
268 #endif
269 }
270 
271 /*
272  * Page in the pages for the object using its associated pager.
273  * The requested page must be fully valid on successful return.
274  */
275 int
276 vm_pager_get_pages(vm_object_t object, vm_page_t *m, int count, int *rbehind,
277     int *rahead)
278 {
279 #ifdef INVARIANTS
280 	vm_pindex_t pindex = m[0]->pindex;
281 #endif
282 	int r;
283 
284 	vm_pager_assert_in(object, m, count);
285 
286 	r = (*pagertab[object->type]->pgo_getpages)(object, m, count, rbehind,
287 	    rahead);
288 	if (r != VM_PAGER_OK)
289 		return (r);
290 
291 	for (int i = 0; i < count; i++) {
292 		/*
293 		 * If pager has replaced a page, assert that it had
294 		 * updated the array.
295 		 */
296 		KASSERT(m[i] == vm_page_lookup(object, pindex++),
297 		    ("%s: mismatch page %p pindex %ju", __func__,
298 		    m[i], (uintmax_t )pindex - 1));
299 		/*
300 		 * Zero out partially filled data.
301 		 */
302 		if (m[i]->valid != VM_PAGE_BITS_ALL)
303 			vm_page_zero_invalid(m[i], TRUE);
304 	}
305 	return (VM_PAGER_OK);
306 }
307 
308 int
309 vm_pager_get_pages_async(vm_object_t object, vm_page_t *m, int count,
310     int *rbehind, int *rahead, pgo_getpages_iodone_t iodone, void *arg)
311 {
312 
313 	vm_pager_assert_in(object, m, count);
314 
315 	return ((*pagertab[object->type]->pgo_getpages_async)(object, m,
316 	    count, rbehind, rahead, iodone, arg));
317 }
318 
319 /*
320  * vm_pager_put_pages() - inline, see vm/vm_pager.h
321  * vm_pager_has_page() - inline, see vm/vm_pager.h
322  */
323 
324 /*
325  * Search the specified pager object list for an object with the
326  * specified handle.  If an object with the specified handle is found,
327  * increase its reference count and return it.  Otherwise, return NULL.
328  *
329  * The pager object list must be locked.
330  */
331 vm_object_t
332 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
333 {
334 	vm_object_t object;
335 
336 	TAILQ_FOREACH(object, pg_list, pager_object_list) {
337 		if (object->handle == handle) {
338 			VM_OBJECT_WLOCK(object);
339 			if ((object->flags & OBJ_DEAD) == 0) {
340 				vm_object_reference_locked(object);
341 				VM_OBJECT_WUNLOCK(object);
342 				break;
343 			}
344 			VM_OBJECT_WUNLOCK(object);
345 		}
346 	}
347 	return (object);
348 }
349 
350 /*
351  * initialize a physical buffer
352  */
353 
354 /*
355  * XXX This probably belongs in vfs_bio.c
356  */
357 static void
358 initpbuf(struct buf *bp)
359 {
360 
361 	KASSERT(bp->b_bufobj == NULL, ("initpbuf with bufobj"));
362 	KASSERT(bp->b_vp == NULL, ("initpbuf with vp"));
363 	bp->b_rcred = NOCRED;
364 	bp->b_wcred = NOCRED;
365 	bp->b_qindex = 0;	/* On no queue (QUEUE_NONE) */
366 	bp->b_kvabase = (caddr_t)(MAXPHYS * (bp - swbuf)) + swapbkva;
367 	bp->b_data = bp->b_kvabase;
368 	bp->b_kvasize = MAXPHYS;
369 	bp->b_flags = 0;
370 	bp->b_xflags = 0;
371 	bp->b_ioflags = 0;
372 	bp->b_iodone = NULL;
373 	bp->b_error = 0;
374 	BUF_LOCK(bp, LK_EXCLUSIVE, NULL);
375 	buf_track(bp, __func__);
376 }
377 
378 /*
379  * allocate a physical buffer
380  *
381  *	There are a limited number (nswbuf) of physical buffers.  We need
382  *	to make sure that no single subsystem is able to hog all of them,
383  *	so each subsystem implements a counter which is typically initialized
384  *	to 1/2 nswbuf.  getpbuf() decrements this counter in allocation and
385  *	increments it on release, and blocks if the counter hits zero.  A
386  *	subsystem may initialize the counter to -1 to disable the feature,
387  *	but it must still be sure to match up all uses of getpbuf() with
388  *	relpbuf() using the same variable.
389  *
390  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
391  *	relatively soon when the rest of the subsystems get smart about it. XXX
392  */
393 struct buf *
394 getpbuf(int *pfreecnt)
395 {
396 	struct buf *bp;
397 
398 	mtx_lock(&pbuf_mtx);
399 	for (;;) {
400 		if (pfreecnt != NULL) {
401 			while (*pfreecnt == 0) {
402 				msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0);
403 			}
404 		}
405 
406 		/* get a bp from the swap buffer header pool */
407 		if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
408 			break;
409 
410 		bswneeded = 1;
411 		msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0);
412 		/* loop in case someone else grabbed one */
413 	}
414 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
415 	if (pfreecnt)
416 		--*pfreecnt;
417 	mtx_unlock(&pbuf_mtx);
418 	initpbuf(bp);
419 	return (bp);
420 }
421 
422 /*
423  * allocate a physical buffer, if one is available.
424  *
425  *	Note that there is no NULL hack here - all subsystems using this
426  *	call understand how to use pfreecnt.
427  */
428 struct buf *
429 trypbuf(int *pfreecnt)
430 {
431 	struct buf *bp;
432 
433 	mtx_lock(&pbuf_mtx);
434 	if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
435 		mtx_unlock(&pbuf_mtx);
436 		return NULL;
437 	}
438 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
439 	--*pfreecnt;
440 	mtx_unlock(&pbuf_mtx);
441 	initpbuf(bp);
442 	return (bp);
443 }
444 
445 /*
446  * release a physical buffer
447  *
448  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
449  *	relatively soon when the rest of the subsystems get smart about it. XXX
450  */
451 void
452 relpbuf(struct buf *bp, int *pfreecnt)
453 {
454 
455 	if (bp->b_rcred != NOCRED) {
456 		crfree(bp->b_rcred);
457 		bp->b_rcred = NOCRED;
458 	}
459 	if (bp->b_wcred != NOCRED) {
460 		crfree(bp->b_wcred);
461 		bp->b_wcred = NOCRED;
462 	}
463 
464 	KASSERT(bp->b_vp == NULL, ("relpbuf with vp"));
465 	KASSERT(bp->b_bufobj == NULL, ("relpbuf with bufobj"));
466 
467 	buf_track(bp, __func__);
468 	BUF_UNLOCK(bp);
469 
470 	mtx_lock(&pbuf_mtx);
471 	TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
472 
473 	if (bswneeded) {
474 		bswneeded = 0;
475 		wakeup(&bswneeded);
476 	}
477 	if (pfreecnt) {
478 		if (++*pfreecnt == 1)
479 			wakeup(pfreecnt);
480 	}
481 	mtx_unlock(&pbuf_mtx);
482 }
483 
484 /*
485  * Associate a p-buffer with a vnode.
486  *
487  * Also sets B_PAGING flag to indicate that vnode is not fully associated
488  * with the buffer.  i.e. the bp has not been linked into the vnode or
489  * ref-counted.
490  */
491 void
492 pbgetvp(struct vnode *vp, struct buf *bp)
493 {
494 
495 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
496 	KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)"));
497 
498 	bp->b_vp = vp;
499 	bp->b_flags |= B_PAGING;
500 	bp->b_bufobj = &vp->v_bufobj;
501 }
502 
503 /*
504  * Associate a p-buffer with a vnode.
505  *
506  * Also sets B_PAGING flag to indicate that vnode is not fully associated
507  * with the buffer.  i.e. the bp has not been linked into the vnode or
508  * ref-counted.
509  */
510 void
511 pbgetbo(struct bufobj *bo, struct buf *bp)
512 {
513 
514 	KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)"));
515 	KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)"));
516 
517 	bp->b_flags |= B_PAGING;
518 	bp->b_bufobj = bo;
519 }
520 
521 /*
522  * Disassociate a p-buffer from a vnode.
523  */
524 void
525 pbrelvp(struct buf *bp)
526 {
527 
528 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
529 	KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj"));
530 	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
531 	    ("pbrelvp: pager buf on vnode list."));
532 
533 	bp->b_vp = NULL;
534 	bp->b_bufobj = NULL;
535 	bp->b_flags &= ~B_PAGING;
536 }
537 
538 /*
539  * Disassociate a p-buffer from a bufobj.
540  */
541 void
542 pbrelbo(struct buf *bp)
543 {
544 
545 	KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode"));
546 	KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj"));
547 	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
548 	    ("pbrelbo: pager buf on vnode list."));
549 
550 	bp->b_bufobj = NULL;
551 	bp->b_flags &= ~B_PAGING;
552 }
553