xref: /freebsd/sys/vm/vm_pager.c (revision 756a5412798b7de1709bb1de2db5ba2a5908cba3)
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 /*
167  * Kernel address space for mapping pages.
168  * Used by pagers where KVAs are needed for IO.
169  *
170  * XXX needs to be large enough to support the number of pending async
171  * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size
172  * (MAXPHYS == 64k) if you want to get the most efficiency.
173  */
174 vm_offset_t swapbkva;		/* swap buffers kva */
175 
176 void
177 vm_pager_init(void)
178 {
179 	struct pagerops **pgops;
180 
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 static int nswbuf_max;
190 
191 void
192 vm_pager_bufferinit(void)
193 {
194 
195 	/* Main zone for paging bufs. */
196 	pbuf_zone = uma_zcreate("pbuf", sizeof(struct buf),
197 	    pbuf_ctor, pbuf_dtor, pbuf_init, NULL, UMA_ALIGN_CACHE,
198 	    UMA_ZONE_VM | UMA_ZONE_NOFREE);
199 	/* Few systems may still use this zone directly, so it needs a limit. */
200 	nswbuf_max += uma_zone_set_max(pbuf_zone, NSWBUF_MIN);
201 }
202 
203 uma_zone_t
204 pbuf_zsecond_create(char *name, int max)
205 {
206 	uma_zone_t zone;
207 
208 	zone = uma_zsecond_create(name, pbuf_ctor, pbuf_dtor, NULL, NULL,
209 	    pbuf_zone);
210 	/*
211 	 * uma_prealloc() rounds up to items per slab. If we would prealloc
212 	 * immediately on every pbuf_zsecond_create(), we may accumulate too
213 	 * much of difference between hard limit and prealloced items, which
214 	 * means wasted memory.
215 	 */
216 	if (nswbuf_max > 0)
217 		nswbuf_max += uma_zone_set_max(zone, max);
218 	else
219 		uma_prealloc(pbuf_zone, uma_zone_set_max(zone, max));
220 
221 	return (zone);
222 }
223 
224 static void
225 pbuf_prealloc(void *arg __unused)
226 {
227 
228 	uma_prealloc(pbuf_zone, nswbuf_max);
229 	nswbuf_max = -1;
230 }
231 
232 SYSINIT(pbuf, SI_SUB_KTHREAD_BUF, SI_ORDER_ANY, pbuf_prealloc, NULL);
233 
234 /*
235  * Allocate an instance of a pager of the given type.
236  * Size, protection and offset parameters are passed in for pagers that
237  * need to perform page-level validation (e.g. the device pager).
238  */
239 vm_object_t
240 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
241     vm_prot_t prot, vm_ooffset_t off, struct ucred *cred)
242 {
243 	vm_object_t ret;
244 	struct pagerops *ops;
245 
246 	ops = pagertab[type];
247 	if (ops)
248 		ret = (*ops->pgo_alloc)(handle, size, prot, off, cred);
249 	else
250 		ret = NULL;
251 	return (ret);
252 }
253 
254 /*
255  *	The object must be locked.
256  */
257 void
258 vm_pager_deallocate(vm_object_t object)
259 {
260 
261 	VM_OBJECT_ASSERT_WLOCKED(object);
262 	(*pagertab[object->type]->pgo_dealloc) (object);
263 }
264 
265 static void
266 vm_pager_assert_in(vm_object_t object, vm_page_t *m, int count)
267 {
268 #ifdef INVARIANTS
269 
270 	VM_OBJECT_ASSERT_WLOCKED(object);
271 	KASSERT(count > 0, ("%s: 0 count", __func__));
272 	/*
273 	 * All pages must be busied, not mapped, not fully valid,
274 	 * not dirty and belong to the proper object.
275 	 */
276 	for (int i = 0 ; i < count; i++) {
277 		if (m[i] == bogus_page)
278 			continue;
279 		vm_page_assert_xbusied(m[i]);
280 		KASSERT(!pmap_page_is_mapped(m[i]),
281 		    ("%s: page %p is mapped", __func__, m[i]));
282 		KASSERT(m[i]->valid != VM_PAGE_BITS_ALL,
283 		    ("%s: request for a valid page %p", __func__, m[i]));
284 		KASSERT(m[i]->dirty == 0,
285 		    ("%s: page %p is dirty", __func__, m[i]));
286 		KASSERT(m[i]->object == object,
287 		    ("%s: wrong object %p/%p", __func__, object, m[i]->object));
288 	}
289 #endif
290 }
291 
292 /*
293  * Page in the pages for the object using its associated pager.
294  * The requested page must be fully valid on successful return.
295  */
296 int
297 vm_pager_get_pages(vm_object_t object, vm_page_t *m, int count, int *rbehind,
298     int *rahead)
299 {
300 #ifdef INVARIANTS
301 	vm_pindex_t pindex = m[0]->pindex;
302 #endif
303 	int r;
304 
305 	vm_pager_assert_in(object, m, count);
306 
307 	r = (*pagertab[object->type]->pgo_getpages)(object, m, count, rbehind,
308 	    rahead);
309 	if (r != VM_PAGER_OK)
310 		return (r);
311 
312 	for (int i = 0; i < count; i++) {
313 		/*
314 		 * If pager has replaced a page, assert that it had
315 		 * updated the array.
316 		 */
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 		/*
321 		 * Zero out partially filled data.
322 		 */
323 		if (m[i]->valid != VM_PAGE_BITS_ALL)
324 			vm_page_zero_invalid(m[i], TRUE);
325 	}
326 	return (VM_PAGER_OK);
327 }
328 
329 int
330 vm_pager_get_pages_async(vm_object_t object, vm_page_t *m, int count,
331     int *rbehind, int *rahead, pgo_getpages_iodone_t iodone, void *arg)
332 {
333 
334 	vm_pager_assert_in(object, m, count);
335 
336 	return ((*pagertab[object->type]->pgo_getpages_async)(object, m,
337 	    count, rbehind, rahead, iodone, arg));
338 }
339 
340 /*
341  * vm_pager_put_pages() - inline, see vm/vm_pager.h
342  * vm_pager_has_page() - inline, see vm/vm_pager.h
343  */
344 
345 /*
346  * Search the specified pager object list for an object with the
347  * specified handle.  If an object with the specified handle is found,
348  * increase its reference count and return it.  Otherwise, return NULL.
349  *
350  * The pager object list must be locked.
351  */
352 vm_object_t
353 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
354 {
355 	vm_object_t object;
356 
357 	TAILQ_FOREACH(object, pg_list, pager_object_list) {
358 		if (object->handle == handle) {
359 			VM_OBJECT_WLOCK(object);
360 			if ((object->flags & OBJ_DEAD) == 0) {
361 				vm_object_reference_locked(object);
362 				VM_OBJECT_WUNLOCK(object);
363 				break;
364 			}
365 			VM_OBJECT_WUNLOCK(object);
366 		}
367 	}
368 	return (object);
369 }
370 
371 static int
372 pbuf_ctor(void *mem, int size, void *arg, int flags)
373 {
374 	struct buf *bp = mem;
375 
376 	bp->b_vp = NULL;
377 	bp->b_bufobj = NULL;
378 
379 	/* copied from initpbuf() */
380 	bp->b_rcred = NOCRED;
381 	bp->b_wcred = NOCRED;
382 	bp->b_qindex = 0;       /* On no queue (QUEUE_NONE) */
383 	bp->b_data = bp->b_kvabase;
384 	bp->b_xflags = 0;
385 	bp->b_flags = 0;
386 	bp->b_ioflags = 0;
387 	bp->b_iodone = NULL;
388 	bp->b_error = 0;
389 	BUF_LOCK(bp, LK_EXCLUSIVE, NULL);
390 
391 	return (0);
392 }
393 
394 static void
395 pbuf_dtor(void *mem, int size, void *arg)
396 {
397 	struct buf *bp = mem;
398 
399 	if (bp->b_rcred != NOCRED) {
400 		crfree(bp->b_rcred);
401 		bp->b_rcred = NOCRED;
402 	}
403 	if (bp->b_wcred != NOCRED) {
404 		crfree(bp->b_wcred);
405 		bp->b_wcred = NOCRED;
406 	}
407 
408 	BUF_UNLOCK(bp);
409 }
410 
411 static int
412 pbuf_init(void *mem, int size, int flags)
413 {
414 	struct buf *bp = mem;
415 
416 	bp->b_kvabase = (void *)kva_alloc(MAXPHYS);
417 	if (bp->b_kvabase == NULL)
418 		return (ENOMEM);
419 	bp->b_kvasize = MAXPHYS;
420 	BUF_LOCKINIT(bp);
421 	LIST_INIT(&bp->b_dep);
422 	bp->b_rcred = bp->b_wcred = NOCRED;
423 	bp->b_xflags = 0;
424 
425 	return (0);
426 }
427 
428 /*
429  * Associate a p-buffer with a vnode.
430  *
431  * Also sets B_PAGING flag to indicate that vnode is not fully associated
432  * with the buffer.  i.e. the bp has not been linked into the vnode or
433  * ref-counted.
434  */
435 void
436 pbgetvp(struct vnode *vp, struct buf *bp)
437 {
438 
439 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
440 	KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)"));
441 
442 	bp->b_vp = vp;
443 	bp->b_flags |= B_PAGING;
444 	bp->b_bufobj = &vp->v_bufobj;
445 }
446 
447 /*
448  * Associate a p-buffer with a vnode.
449  *
450  * Also sets B_PAGING flag to indicate that vnode is not fully associated
451  * with the buffer.  i.e. the bp has not been linked into the vnode or
452  * ref-counted.
453  */
454 void
455 pbgetbo(struct bufobj *bo, struct buf *bp)
456 {
457 
458 	KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)"));
459 	KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)"));
460 
461 	bp->b_flags |= B_PAGING;
462 	bp->b_bufobj = bo;
463 }
464 
465 /*
466  * Disassociate a p-buffer from a vnode.
467  */
468 void
469 pbrelvp(struct buf *bp)
470 {
471 
472 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
473 	KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj"));
474 	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
475 	    ("pbrelvp: pager buf on vnode list."));
476 
477 	bp->b_vp = NULL;
478 	bp->b_bufobj = NULL;
479 	bp->b_flags &= ~B_PAGING;
480 }
481 
482 /*
483  * Disassociate a p-buffer from a bufobj.
484  */
485 void
486 pbrelbo(struct buf *bp)
487 {
488 
489 	KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode"));
490 	KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj"));
491 	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
492 	    ("pbrelbo: pager buf on vnode list."));
493 
494 	bp->b_bufobj = NULL;
495 	bp->b_flags &= ~B_PAGING;
496 }
497