xref: /freebsd/sys/vm/vm_pager.c (revision 410556f1f10fd35b350102725fd8504c3cb0afc8)
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",
187 	    sizeof(struct buf) + PBUF_PAGES * sizeof(vm_page_t),
188 	    pbuf_ctor, pbuf_dtor, pbuf_init, NULL, UMA_ALIGN_CACHE,
189 	    UMA_ZONE_NOFREE);
190 	/* Few systems may still use this zone directly, so it needs a limit. */
191 	nswbuf_max += uma_zone_set_max(pbuf_zone, NSWBUF_MIN);
192 }
193 
194 uma_zone_t
195 pbuf_zsecond_create(const char *name, int max)
196 {
197 	uma_zone_t zone;
198 
199 	zone = uma_zsecond_create(name, pbuf_ctor, pbuf_dtor, NULL, NULL,
200 	    pbuf_zone);
201 	/*
202 	 * uma_prealloc() rounds up to items per slab. If we would prealloc
203 	 * immediately on every pbuf_zsecond_create(), we may accumulate too
204 	 * much of difference between hard limit and prealloced items, which
205 	 * means wasted memory.
206 	 */
207 	if (nswbuf_max > 0)
208 		nswbuf_max += uma_zone_set_max(zone, max);
209 	else
210 		uma_prealloc(pbuf_zone, uma_zone_set_max(zone, max));
211 
212 	return (zone);
213 }
214 
215 static void
216 pbuf_prealloc(void *arg __unused)
217 {
218 
219 	uma_prealloc(pbuf_zone, nswbuf_max);
220 	nswbuf_max = -1;
221 }
222 
223 SYSINIT(pbuf, SI_SUB_KTHREAD_BUF, SI_ORDER_ANY, pbuf_prealloc, NULL);
224 
225 /*
226  * Allocate an instance of a pager of the given type.
227  * Size, protection and offset parameters are passed in for pagers that
228  * need to perform page-level validation (e.g. the device pager).
229  */
230 vm_object_t
231 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
232     vm_prot_t prot, vm_ooffset_t off, struct ucred *cred)
233 {
234 	vm_object_t ret;
235 	struct pagerops *ops;
236 
237 	ops = pagertab[type];
238 	if (ops)
239 		ret = (*ops->pgo_alloc)(handle, size, prot, off, cred);
240 	else
241 		ret = NULL;
242 	return (ret);
243 }
244 
245 /*
246  *	The object must be locked.
247  */
248 void
249 vm_pager_deallocate(vm_object_t object)
250 {
251 
252 	VM_OBJECT_ASSERT_WLOCKED(object);
253 	(*pagertab[object->type]->pgo_dealloc) (object);
254 }
255 
256 static void
257 vm_pager_assert_in(vm_object_t object, vm_page_t *m, int count)
258 {
259 #ifdef INVARIANTS
260 
261 	/*
262 	 * All pages must be consecutive, busied, not mapped, not fully valid,
263 	 * not dirty and belong to the proper object.  Some pages may be the
264 	 * bogus page, but the first and last pages must be a real ones.
265 	 */
266 
267 	VM_OBJECT_ASSERT_UNLOCKED(object);
268 	VM_OBJECT_ASSERT_PAGING(object);
269 	KASSERT(count > 0, ("%s: 0 count", __func__));
270 	for (int i = 0 ; i < count; i++) {
271 		if (m[i] == bogus_page) {
272 			KASSERT(i != 0 && i != count - 1,
273 			    ("%s: page %d is the bogus page", __func__, i));
274 			continue;
275 		}
276 		vm_page_assert_xbusied(m[i]);
277 		KASSERT(!pmap_page_is_mapped(m[i]),
278 		    ("%s: page %p is mapped", __func__, m[i]));
279 		KASSERT(m[i]->valid != VM_PAGE_BITS_ALL,
280 		    ("%s: request for a valid page %p", __func__, m[i]));
281 		KASSERT(m[i]->dirty == 0,
282 		    ("%s: page %p is dirty", __func__, m[i]));
283 		KASSERT(m[i]->object == object,
284 		    ("%s: wrong object %p/%p", __func__, object, m[i]->object));
285 		KASSERT(m[i]->pindex == m[0]->pindex + i,
286 		    ("%s: page %p isn't consecutive", __func__, m[i]));
287 	}
288 #endif
289 }
290 
291 /*
292  * Page in the pages for the object using its associated pager.
293  * The requested page must be fully valid on successful return.
294  */
295 int
296 vm_pager_get_pages(vm_object_t object, vm_page_t *m, int count, int *rbehind,
297     int *rahead)
298 {
299 #ifdef INVARIANTS
300 	vm_pindex_t pindex = m[0]->pindex;
301 #endif
302 	int r;
303 
304 	vm_pager_assert_in(object, m, count);
305 
306 	r = (*pagertab[object->type]->pgo_getpages)(object, m, count, rbehind,
307 	    rahead);
308 	if (r != VM_PAGER_OK)
309 		return (r);
310 
311 	for (int i = 0; i < count; i++) {
312 		/*
313 		 * If pager has replaced a page, assert that it had
314 		 * updated the array.
315 		 */
316 #ifdef INVARIANTS
317 		VM_OBJECT_RLOCK(object);
318 		KASSERT(m[i] == vm_page_lookup(object, pindex++),
319 		    ("%s: mismatch page %p pindex %ju", __func__,
320 		    m[i], (uintmax_t )pindex - 1));
321 		VM_OBJECT_RUNLOCK(object);
322 #endif
323 		/*
324 		 * Zero out partially filled data.
325 		 */
326 		if (m[i]->valid != VM_PAGE_BITS_ALL)
327 			vm_page_zero_invalid(m[i], TRUE);
328 	}
329 	return (VM_PAGER_OK);
330 }
331 
332 int
333 vm_pager_get_pages_async(vm_object_t object, vm_page_t *m, int count,
334     int *rbehind, int *rahead, pgo_getpages_iodone_t iodone, void *arg)
335 {
336 
337 	vm_pager_assert_in(object, m, count);
338 
339 	return ((*pagertab[object->type]->pgo_getpages_async)(object, m,
340 	    count, rbehind, rahead, iodone, arg));
341 }
342 
343 /*
344  * vm_pager_put_pages() - inline, see vm/vm_pager.h
345  * vm_pager_has_page() - inline, see vm/vm_pager.h
346  */
347 
348 /*
349  * Search the specified pager object list for an object with the
350  * specified handle.  If an object with the specified handle is found,
351  * increase its reference count and return it.  Otherwise, return NULL.
352  *
353  * The pager object list must be locked.
354  */
355 vm_object_t
356 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
357 {
358 	vm_object_t object;
359 
360 	TAILQ_FOREACH(object, pg_list, pager_object_list) {
361 		if (object->handle == handle) {
362 			VM_OBJECT_WLOCK(object);
363 			if ((object->flags & OBJ_DEAD) == 0) {
364 				vm_object_reference_locked(object);
365 				VM_OBJECT_WUNLOCK(object);
366 				break;
367 			}
368 			VM_OBJECT_WUNLOCK(object);
369 		}
370 	}
371 	return (object);
372 }
373 
374 static int
375 pbuf_ctor(void *mem, int size, void *arg, int flags)
376 {
377 	struct buf *bp = mem;
378 
379 	bp->b_vp = NULL;
380 	bp->b_bufobj = NULL;
381 
382 	/* copied from initpbuf() */
383 	bp->b_rcred = NOCRED;
384 	bp->b_wcred = NOCRED;
385 	bp->b_qindex = 0;       /* On no queue (QUEUE_NONE) */
386 	bp->b_data = bp->b_kvabase;
387 	bp->b_xflags = 0;
388 	bp->b_flags = B_MAXPHYS;
389 	bp->b_ioflags = 0;
390 	bp->b_iodone = NULL;
391 	bp->b_error = 0;
392 	BUF_LOCK(bp, LK_EXCLUSIVE, NULL);
393 
394 	return (0);
395 }
396 
397 static void
398 pbuf_dtor(void *mem, int size, void *arg)
399 {
400 	struct buf *bp = mem;
401 
402 	if (bp->b_rcred != NOCRED) {
403 		crfree(bp->b_rcred);
404 		bp->b_rcred = NOCRED;
405 	}
406 	if (bp->b_wcred != NOCRED) {
407 		crfree(bp->b_wcred);
408 		bp->b_wcred = NOCRED;
409 	}
410 
411 	BUF_UNLOCK(bp);
412 }
413 
414 static int
415 pbuf_init(void *mem, int size, int flags)
416 {
417 	struct buf *bp = mem;
418 
419 	bp->b_kvabase = (void *)kva_alloc(ptoa(PBUF_PAGES));
420 	if (bp->b_kvabase == NULL)
421 		return (ENOMEM);
422 	bp->b_kvasize = ptoa(PBUF_PAGES);
423 	BUF_LOCKINIT(bp);
424 	LIST_INIT(&bp->b_dep);
425 	bp->b_rcred = bp->b_wcred = NOCRED;
426 	bp->b_xflags = 0;
427 
428 	return (0);
429 }
430 
431 /*
432  * Associate a p-buffer with a vnode.
433  *
434  * Also sets B_PAGING flag to indicate that vnode is not fully associated
435  * with the buffer.  i.e. the bp has not been linked into the vnode or
436  * ref-counted.
437  */
438 void
439 pbgetvp(struct vnode *vp, struct buf *bp)
440 {
441 
442 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
443 	KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)"));
444 
445 	bp->b_vp = vp;
446 	bp->b_flags |= B_PAGING;
447 	bp->b_bufobj = &vp->v_bufobj;
448 }
449 
450 /*
451  * Associate a p-buffer with a vnode.
452  *
453  * Also sets B_PAGING flag to indicate that vnode is not fully associated
454  * with the buffer.  i.e. the bp has not been linked into the vnode or
455  * ref-counted.
456  */
457 void
458 pbgetbo(struct bufobj *bo, struct buf *bp)
459 {
460 
461 	KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)"));
462 	KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)"));
463 
464 	bp->b_flags |= B_PAGING;
465 	bp->b_bufobj = bo;
466 }
467 
468 /*
469  * Disassociate a p-buffer from a vnode.
470  */
471 void
472 pbrelvp(struct buf *bp)
473 {
474 
475 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
476 	KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj"));
477 	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
478 	    ("pbrelvp: pager buf on vnode list."));
479 
480 	bp->b_vp = NULL;
481 	bp->b_bufobj = NULL;
482 	bp->b_flags &= ~B_PAGING;
483 }
484 
485 /*
486  * Disassociate a p-buffer from a bufobj.
487  */
488 void
489 pbrelbo(struct buf *bp)
490 {
491 
492 	KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode"));
493 	KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj"));
494 	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
495 	    ("pbrelbo: pager buf on vnode list."));
496 
497 	bp->b_bufobj = NULL;
498 	bp->b_flags &= ~B_PAGING;
499 }
500