xref: /freebsd/sys/vm/vm_pager.c (revision c6989859ae9388eeb46a24fe88f9b8d07101c710)
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 
83 #include <vm/vm.h>
84 #include <vm/vm_param.h>
85 #include <vm/vm_kern.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pager.h>
89 #include <vm/vm_extern.h>
90 #include <vm/uma.h>
91 
92 uma_zone_t pbuf_zone;
93 static int	pbuf_init(void *, int, int);
94 static int	pbuf_ctor(void *, int, void *, int);
95 static void	pbuf_dtor(void *, int, void *);
96 
97 static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *);
98 static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
99     vm_ooffset_t, struct ucred *);
100 static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
101 static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
102 static void dead_pager_dealloc(vm_object_t);
103 
104 static int
105 dead_pager_getpages(vm_object_t obj, vm_page_t *ma, int count, int *rbehind,
106     int *rahead)
107 {
108 
109 	return (VM_PAGER_FAIL);
110 }
111 
112 static vm_object_t
113 dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
114     vm_ooffset_t off, struct ucred *cred)
115 {
116 
117 	return (NULL);
118 }
119 
120 static void
121 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count,
122     int flags, int *rtvals)
123 {
124 	int i;
125 
126 	for (i = 0; i < count; i++)
127 		rtvals[i] = VM_PAGER_AGAIN;
128 }
129 
130 static int
131 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *prev, int *next)
132 {
133 
134 	if (prev != NULL)
135 		*prev = 0;
136 	if (next != NULL)
137 		*next = 0;
138 	return (FALSE);
139 }
140 
141 static void
142 dead_pager_dealloc(vm_object_t object)
143 {
144 
145 }
146 
147 static struct pagerops deadpagerops = {
148 	.pgo_alloc = 	dead_pager_alloc,
149 	.pgo_dealloc =	dead_pager_dealloc,
150 	.pgo_getpages =	dead_pager_getpages,
151 	.pgo_putpages =	dead_pager_putpages,
152 	.pgo_haspage =	dead_pager_haspage,
153 };
154 
155 struct pagerops *pagertab[] = {
156 	&defaultpagerops,	/* OBJT_DEFAULT */
157 	&swappagerops,		/* OBJT_SWAP */
158 	&vnodepagerops,		/* OBJT_VNODE */
159 	&devicepagerops,	/* OBJT_DEVICE */
160 	&physpagerops,		/* OBJT_PHYS */
161 	&deadpagerops,		/* OBJT_DEAD */
162 	&sgpagerops,		/* OBJT_SG */
163 	&mgtdevicepagerops,	/* OBJT_MGTDEVICE */
164 };
165 
166 void
167 vm_pager_init(void)
168 {
169 	struct pagerops **pgops;
170 
171 	/*
172 	 * Initialize known pagers
173 	 */
174 	for (pgops = pagertab; pgops < &pagertab[nitems(pagertab)]; pgops++)
175 		if ((*pgops)->pgo_init != NULL)
176 			(*(*pgops)->pgo_init)();
177 }
178 
179 static int nswbuf_max;
180 
181 void
182 vm_pager_bufferinit(void)
183 {
184 
185 	/* Main zone for paging bufs. */
186 	pbuf_zone = uma_zcreate("pbuf", sizeof(struct buf),
187 	    pbuf_ctor, pbuf_dtor, pbuf_init, NULL, UMA_ALIGN_CACHE,
188 	    UMA_ZONE_NOFREE);
189 	/* Few systems may still use this zone directly, so it needs a limit. */
190 	nswbuf_max += uma_zone_set_max(pbuf_zone, NSWBUF_MIN);
191 }
192 
193 uma_zone_t
194 pbuf_zsecond_create(const char *name, int max)
195 {
196 	uma_zone_t zone;
197 
198 	zone = uma_zsecond_create(name, pbuf_ctor, pbuf_dtor, NULL, NULL,
199 	    pbuf_zone);
200 	/*
201 	 * uma_prealloc() rounds up to items per slab. If we would prealloc
202 	 * immediately on every pbuf_zsecond_create(), we may accumulate too
203 	 * much of difference between hard limit and prealloced items, which
204 	 * means wasted memory.
205 	 */
206 	if (nswbuf_max > 0)
207 		nswbuf_max += uma_zone_set_max(zone, max);
208 	else
209 		uma_prealloc(pbuf_zone, uma_zone_set_max(zone, max));
210 
211 	return (zone);
212 }
213 
214 static void
215 pbuf_prealloc(void *arg __unused)
216 {
217 
218 	uma_prealloc(pbuf_zone, nswbuf_max);
219 	nswbuf_max = -1;
220 }
221 
222 SYSINIT(pbuf, SI_SUB_KTHREAD_BUF, SI_ORDER_ANY, pbuf_prealloc, NULL);
223 
224 /*
225  * Allocate an instance of a pager of the given type.
226  * Size, protection and offset parameters are passed in for pagers that
227  * need to perform page-level validation (e.g. the device pager).
228  */
229 vm_object_t
230 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
231     vm_prot_t prot, vm_ooffset_t off, struct ucred *cred)
232 {
233 	vm_object_t ret;
234 	struct pagerops *ops;
235 
236 	ops = pagertab[type];
237 	if (ops)
238 		ret = (*ops->pgo_alloc)(handle, size, prot, off, cred);
239 	else
240 		ret = NULL;
241 	return (ret);
242 }
243 
244 /*
245  *	The object must be locked.
246  */
247 void
248 vm_pager_deallocate(vm_object_t object)
249 {
250 
251 	VM_OBJECT_ASSERT_WLOCKED(object);
252 	(*pagertab[object->type]->pgo_dealloc) (object);
253 }
254 
255 static void
256 vm_pager_assert_in(vm_object_t object, vm_page_t *m, int count)
257 {
258 #ifdef INVARIANTS
259 
260 	/*
261 	 * All pages must be consecutive, busied, not mapped, not fully valid,
262 	 * not dirty and belong to the proper object.  Some pages may be the
263 	 * bogus page, but the first and last pages must be a real ones.
264 	 */
265 
266 	VM_OBJECT_ASSERT_UNLOCKED(object);
267 	VM_OBJECT_ASSERT_PAGING(object);
268 	KASSERT(count > 0, ("%s: 0 count", __func__));
269 	for (int i = 0 ; i < count; i++) {
270 		if (m[i] == bogus_page) {
271 			KASSERT(i != 0 && i != count - 1,
272 			    ("%s: page %d is the bogus page", __func__, i));
273 			continue;
274 		}
275 		vm_page_assert_xbusied(m[i]);
276 		KASSERT(!pmap_page_is_mapped(m[i]),
277 		    ("%s: page %p is mapped", __func__, m[i]));
278 		KASSERT(m[i]->valid != VM_PAGE_BITS_ALL,
279 		    ("%s: request for a valid page %p", __func__, m[i]));
280 		KASSERT(m[i]->dirty == 0,
281 		    ("%s: page %p is dirty", __func__, m[i]));
282 		KASSERT(m[i]->object == object,
283 		    ("%s: wrong object %p/%p", __func__, object, m[i]->object));
284 		KASSERT(m[i]->pindex == m[0]->pindex + i,
285 		    ("%s: page %p isn't consecutive", __func__, m[i]));
286 	}
287 #endif
288 }
289 
290 /*
291  * Page in the pages for the object using its associated pager.
292  * The requested page must be fully valid on successful return.
293  */
294 int
295 vm_pager_get_pages(vm_object_t object, vm_page_t *m, int count, int *rbehind,
296     int *rahead)
297 {
298 #ifdef INVARIANTS
299 	vm_pindex_t pindex = m[0]->pindex;
300 #endif
301 	int r;
302 
303 	vm_pager_assert_in(object, m, count);
304 
305 	r = (*pagertab[object->type]->pgo_getpages)(object, m, count, rbehind,
306 	    rahead);
307 	if (r != VM_PAGER_OK)
308 		return (r);
309 
310 	for (int i = 0; i < count; i++) {
311 		/*
312 		 * If pager has replaced a page, assert that it had
313 		 * updated the array.
314 		 */
315 #ifdef INVARIANTS
316 		VM_OBJECT_RLOCK(object);
317 		KASSERT(m[i] == vm_page_lookup(object, pindex++),
318 		    ("%s: mismatch page %p pindex %ju", __func__,
319 		    m[i], (uintmax_t )pindex - 1));
320 		VM_OBJECT_RUNLOCK(object);
321 #endif
322 		/*
323 		 * Zero out partially filled data.
324 		 */
325 		if (m[i]->valid != VM_PAGE_BITS_ALL)
326 			vm_page_zero_invalid(m[i], TRUE);
327 	}
328 	return (VM_PAGER_OK);
329 }
330 
331 int
332 vm_pager_get_pages_async(vm_object_t object, vm_page_t *m, int count,
333     int *rbehind, int *rahead, pgo_getpages_iodone_t iodone, void *arg)
334 {
335 
336 	vm_pager_assert_in(object, m, count);
337 
338 	return ((*pagertab[object->type]->pgo_getpages_async)(object, m,
339 	    count, rbehind, rahead, iodone, arg));
340 }
341 
342 /*
343  * vm_pager_put_pages() - inline, see vm/vm_pager.h
344  * vm_pager_has_page() - inline, see vm/vm_pager.h
345  */
346 
347 /*
348  * Search the specified pager object list for an object with the
349  * specified handle.  If an object with the specified handle is found,
350  * increase its reference count and return it.  Otherwise, return NULL.
351  *
352  * The pager object list must be locked.
353  */
354 vm_object_t
355 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
356 {
357 	vm_object_t object;
358 
359 	TAILQ_FOREACH(object, pg_list, pager_object_list) {
360 		if (object->handle == handle) {
361 			VM_OBJECT_WLOCK(object);
362 			if ((object->flags & OBJ_DEAD) == 0) {
363 				vm_object_reference_locked(object);
364 				VM_OBJECT_WUNLOCK(object);
365 				break;
366 			}
367 			VM_OBJECT_WUNLOCK(object);
368 		}
369 	}
370 	return (object);
371 }
372 
373 static int
374 pbuf_ctor(void *mem, int size, void *arg, int flags)
375 {
376 	struct buf *bp = mem;
377 
378 	bp->b_vp = NULL;
379 	bp->b_bufobj = NULL;
380 
381 	/* copied from initpbuf() */
382 	bp->b_rcred = NOCRED;
383 	bp->b_wcred = NOCRED;
384 	bp->b_qindex = 0;       /* On no queue (QUEUE_NONE) */
385 	bp->b_data = bp->b_kvabase;
386 	bp->b_xflags = 0;
387 	bp->b_flags = 0;
388 	bp->b_ioflags = 0;
389 	bp->b_iodone = NULL;
390 	bp->b_error = 0;
391 	BUF_LOCK(bp, LK_EXCLUSIVE, NULL);
392 
393 	return (0);
394 }
395 
396 static void
397 pbuf_dtor(void *mem, int size, void *arg)
398 {
399 	struct buf *bp = mem;
400 
401 	if (bp->b_rcred != NOCRED) {
402 		crfree(bp->b_rcred);
403 		bp->b_rcred = NOCRED;
404 	}
405 	if (bp->b_wcred != NOCRED) {
406 		crfree(bp->b_wcred);
407 		bp->b_wcred = NOCRED;
408 	}
409 
410 	BUF_UNLOCK(bp);
411 }
412 
413 static int
414 pbuf_init(void *mem, int size, int flags)
415 {
416 	struct buf *bp = mem;
417 
418 	bp->b_kvabase = (void *)kva_alloc(MAXPHYS);
419 	if (bp->b_kvabase == NULL)
420 		return (ENOMEM);
421 	bp->b_kvasize = MAXPHYS;
422 	BUF_LOCKINIT(bp);
423 	LIST_INIT(&bp->b_dep);
424 	bp->b_rcred = bp->b_wcred = NOCRED;
425 	bp->b_xflags = 0;
426 
427 	return (0);
428 }
429 
430 /*
431  * Associate a p-buffer with a vnode.
432  *
433  * Also sets B_PAGING flag to indicate that vnode is not fully associated
434  * with the buffer.  i.e. the bp has not been linked into the vnode or
435  * ref-counted.
436  */
437 void
438 pbgetvp(struct vnode *vp, struct buf *bp)
439 {
440 
441 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
442 	KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)"));
443 
444 	bp->b_vp = vp;
445 	bp->b_flags |= B_PAGING;
446 	bp->b_bufobj = &vp->v_bufobj;
447 }
448 
449 /*
450  * Associate a p-buffer with a vnode.
451  *
452  * Also sets B_PAGING flag to indicate that vnode is not fully associated
453  * with the buffer.  i.e. the bp has not been linked into the vnode or
454  * ref-counted.
455  */
456 void
457 pbgetbo(struct bufobj *bo, struct buf *bp)
458 {
459 
460 	KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)"));
461 	KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)"));
462 
463 	bp->b_flags |= B_PAGING;
464 	bp->b_bufobj = bo;
465 }
466 
467 /*
468  * Disassociate a p-buffer from a vnode.
469  */
470 void
471 pbrelvp(struct buf *bp)
472 {
473 
474 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
475 	KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj"));
476 	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
477 	    ("pbrelvp: pager buf on vnode list."));
478 
479 	bp->b_vp = NULL;
480 	bp->b_bufobj = NULL;
481 	bp->b_flags &= ~B_PAGING;
482 }
483 
484 /*
485  * Disassociate a p-buffer from a bufobj.
486  */
487 void
488 pbrelbo(struct buf *bp)
489 {
490 
491 	KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode"));
492 	KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj"));
493 	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
494 	    ("pbrelbo: pager buf on vnode list."));
495 
496 	bp->b_bufobj = NULL;
497 	bp->b_flags &= ~B_PAGING;
498 }
499