4 * Copyright (c) 1982, 1986, 1989, 1993
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)vm_swap.c 8.5 (Berkeley) 2/17/94
32 * $FreeBSD: src/sys/vm/vm_swap.c,v 1.96.2.2 2001/10/14 18:46:47 iedowse Exp $
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/sysproto.h>
43 #include <sys/nlookup.h>
44 #include <sys/sysctl.h>
45 #include <sys/dmap.h> /* XXX */
46 #include <sys/vnode.h>
47 #include <sys/fcntl.h>
48 #include <sys/blist.h>
49 #include <sys/kernel.h>
55 #include <vm/vm_extern.h>
56 #include <vm/swap_pager.h>
57 #include <vm/vm_zone.h>
58 #include <vm/vm_param.h>
60 #include <sys/thread2.h>
61 #include <sys/mplock2.h>
62 #include <sys/mutex2.h>
63 #include <sys/spinlock2.h>
66 * Indirect driver for multi-controller paging.
72 static struct swdevt should_be_malloced
[NSWAPDEV
];
73 struct swdevt
*swdevt
= should_be_malloced
; /* exported to pstat/systat */
74 static swblk_t nswap
; /* first block after the interleaved devs */
75 static struct mtx swap_mtx
= MTX_INITIALIZER
;
76 int nswdev
= NSWAPDEV
; /* exported to pstat/systat */
80 static int swapoff_one(int index
);
81 struct vnode
*swapdev_vp
;
84 * (struct vnode *a_vp, struct bio *b_bio)
86 * vn_strategy() for swapdev_vp. Perform swap strategy interleave device
92 swapdev_strategy(struct vop_strategy_args
*ap
)
94 struct bio
*bio
= ap
->a_bio
;
96 struct buf
*bp
= bio
->bio_buf
;
97 int sz
, off
, seg
, index
, blkno
, nblkno
;
99 sz
= howmany(bp
->b_bcount
, PAGE_SIZE
);
100 blkno
= (int)(bio
->bio_offset
>> PAGE_SHIFT
);
103 * Convert interleaved swap into per-device swap. Note that
104 * the block size is left in PAGE_SIZE'd chunks (for the newswap)
107 nbio
= push_bio(bio
);
110 if (off
+ sz
> dmmax
) {
111 bp
->b_error
= EINVAL
;
112 bp
->b_flags
|= B_ERROR
;
117 index
= seg
% nswdev
;
119 nbio
->bio_offset
= (off_t
)(seg
* dmmax
+ off
) << PAGE_SHIFT
;
122 nbio
->bio_offset
= bio
->bio_offset
;
124 nblkno
= (int)(nbio
->bio_offset
>> PAGE_SHIFT
);
126 if (nblkno
+ sz
> sp
->sw_nblks
) {
127 bp
->b_error
= EINVAL
;
128 bp
->b_flags
|= B_ERROR
;
129 /* I/O was never started on nbio, must biodone(bio) */
133 if (sp
->sw_vp
== NULL
) {
134 bp
->b_error
= ENODEV
;
135 bp
->b_flags
|= B_ERROR
;
136 /* I/O was never started on nbio, must biodone(bio) */
142 * Issue a strategy call on the appropriate swap vnode. Note that
143 * bp->b_vp is not modified. Strategy code is always supposed to
146 * We have to use vn_strategy() here even if we know we have a
147 * device in order to properly break up requests which exceed the
148 * device's DMA limits.
150 vn_strategy(sp
->sw_vp
, nbio
);
155 swapdev_inactive(struct vop_inactive_args
*ap
)
162 swapdev_reclaim(struct vop_reclaim_args
*ap
)
168 * Create a special vnode op vector for swapdev_vp - we only use
169 * vn_strategy(), everything else returns an error.
171 static struct vop_ops swapdev_vnode_vops
= {
172 .vop_default
= vop_defaultop
,
173 .vop_strategy
= swapdev_strategy
,
174 .vop_inactive
= swapdev_inactive
,
175 .vop_reclaim
= swapdev_reclaim
177 static struct vop_ops
*swapdev_vnode_vops_p
= &swapdev_vnode_vops
;
179 VNODEOP_SET(swapdev_vnode_vops
);
182 * swapon_args(char *name)
184 * System call swapon(name) enables swapping on device name,
185 * which must be in the swdevsw. Return EBUSY
186 * if already swapping on this device.
191 sys_swapon(struct swapon_args
*uap
)
193 struct thread
*td
= curthread
;
196 struct nlookupdata nd
;
199 error
= priv_check(td
, PRIV_ROOT
);
206 error
= nlookup_init(&nd
, uap
->name
, UIO_USERSPACE
, NLC_FOLLOW
);
208 error
= nlookup(&nd
);
210 error
= cache_vref(&nd
.nl_nch
, nd
.nl_cred
, &vp
);
214 mtx_unlock(&swap_mtx
);
218 if (vn_isdisk(vp
, &error
)) {
219 error
= swaponvp(td
, vp
, 0);
220 } else if (vp
->v_type
== VREG
&& vp
->v_tag
== VT_NFS
&&
221 (error
= VOP_GETATTR(vp
, &attr
)) == 0) {
223 * Allow direct swapping to NFS regular files in the same
224 * way that nfs_mountroot() sets up diskless swapping.
226 error
= swaponvp(td
, vp
, attr
.va_size
/ DEV_BSIZE
);
231 mtx_unlock(&swap_mtx
);
237 * Swfree(index) frees the index'th portion of the swap map.
238 * Each of the nswdev devices provides 1/nswdev'th of the swap
239 * space, which is laid out with blocks of dmmax pages circularly
242 * The new swap code uses page-sized blocks. The old swap code used
243 * DEV_BSIZE'd chunks.
245 * XXX locking when multiple swapon's run in parallel
248 swaponvp(struct thread
*td
, struct vnode
*vp
, u_quad_t nblks
)
250 swblk_t aligned_nblks
;
263 lwkt_gettoken(&vm_token
); /* needed for vm_swap_size and blist */
267 error
= getspecialvnode(VT_NON
, NULL
, &swapdev_vnode_vops_p
,
270 panic("Cannot get vnode for swapdev");
271 swapdev_vp
->v_type
= VNON
; /* Untyped */
272 vx_unlock(swapdev_vp
);
275 for (sp
= swdevt
, index
= 0 ; index
< nswdev
; index
++, sp
++) {
276 if (sp
->sw_vp
== vp
) {
287 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
288 error
= VOP_OPEN(vp
, FREAD
| FWRITE
, cred
, NULL
);
294 * v_rdev is not valid until after the VOP_OPEN() call. dev_psize()
295 * must be supported if a character device has been specified.
297 if (vp
->v_type
== VCHR
)
302 if (nblks
== 0 && dev
!= NULL
) {
303 dpsize
= dev_dpsize(dev
);
305 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
306 VOP_CLOSE(vp
, FREAD
| FWRITE
, NULL
);
311 nblks
= (u_quad_t
)dpsize
;
314 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
315 VOP_CLOSE(vp
, FREAD
| FWRITE
, NULL
);
322 * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks.
323 * First chop nblks off to page-align it, then convert.
325 * sw->sw_nblks is in page-sized chunks now too.
327 nblks
&= ~(u_quad_t
)(ctodb(1) - 1);
328 nblks
= dbtoc(nblks
);
331 * Post-conversion nblks must not be >= BLIST_MAXBLKS, and
332 * we impose a 4-swap-device limit so we have to divide it out
333 * further. Going beyond this will result in overflows in the
336 * Post-conversion nblks must fit within a (swblk_t), which
337 * this test also ensures.
339 if (nblks
> BLIST_MAXBLKS
/ nswdev
) {
340 kprintf("exceeded maximum of %d blocks per swap unit\n",
341 (int)BLIST_MAXBLKS
/ nswdev
);
342 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
343 VOP_CLOSE(vp
, FREAD
| FWRITE
, NULL
);
350 sp
->sw_dev
= dev2udev(dev
);
352 sp
->sw_flags
= SW_FREED
;
356 * nblks, nswap, and dmmax are PAGE_SIZE'd parameters now, not
357 * DEV_BSIZE'd. aligned_nblks is used to calculate the
358 * size of the swap bitmap, taking into account the stripe size.
360 aligned_nblks
= (swblk_t
)((nblks
+ (dmmax
- 1)) & ~(u_long
)(dmmax
- 1));
361 sp
->sw_nblks
= aligned_nblks
;
363 if (aligned_nblks
* nswdev
> nswap
)
364 nswap
= aligned_nblks
* nswdev
;
366 if (swapblist
== NULL
)
367 swapblist
= blist_create(nswap
);
369 blist_resize(&swapblist
, nswap
, 0);
371 for (dvbase
= dmmax
; dvbase
< aligned_nblks
; dvbase
+= dmmax
) {
372 blk
= min(aligned_nblks
- dvbase
, dmmax
);
373 vsbase
= index
* dmmax
+ dvbase
* nswdev
;
374 blist_free(swapblist
, vsbase
, blk
);
378 swap_pager_newswap();
381 mtx_unlock(&swap_mtx
);
382 lwkt_reltoken(&vm_token
);
387 * swapoff_args(char *name)
389 * System call swapoff(name) disables swapping on device name,
390 * which must be an active swap device. Return ENOMEM
391 * if there is not enough memory to page in the contents of
397 sys_swapoff(struct swapoff_args
*uap
)
400 struct nlookupdata nd
;
404 error
= priv_check(curthread
, PRIV_ROOT
);
411 error
= nlookup_init(&nd
, uap
->name
, UIO_USERSPACE
, NLC_FOLLOW
);
413 error
= nlookup(&nd
);
415 error
= cache_vref(&nd
.nl_nch
, nd
.nl_cred
, &vp
);
420 for (sp
= swdevt
, index
= 0; index
< nswdev
; index
++, sp
++) {
427 error
= swapoff_one(index
);
431 mtx_unlock(&swap_mtx
);
436 swapoff_one(int index
)
438 swblk_t blk
, aligned_nblks
;
439 swblk_t dvbase
, vsbase
;
440 u_int pq_active_clean
, pq_inactive_clean
;
442 struct vm_page marker
;
449 aligned_nblks
= sp
->sw_nblks
;
450 pq_active_clean
= pq_inactive_clean
= 0;
453 * We can turn off this swap device safely only if the
454 * available virtual memory in the system will fit the amount
455 * of data we will have to page back in, plus an epsilon so
456 * the system doesn't become critically low on swap space.
458 for (q
= 0; q
< PQ_L2_SIZE
; ++q
) {
459 bzero(&marker
, sizeof(marker
));
460 marker
.flags
= PG_BUSY
| PG_FICTITIOUS
| PG_MARKER
;
461 marker
.queue
= PQ_ACTIVE
+ q
;
463 marker
.wire_count
= 1;
465 vm_page_queues_spin_lock(marker
.queue
);
466 TAILQ_INSERT_HEAD(&vm_page_queues
[marker
.queue
].pl
,
469 while ((m
= TAILQ_NEXT(&marker
, pageq
)) != NULL
) {
470 TAILQ_REMOVE(&vm_page_queues
[marker
.queue
].pl
,
472 TAILQ_INSERT_AFTER(&vm_page_queues
[marker
.queue
].pl
, m
,
474 if (m
->flags
& (PG_MARKER
| PG_FICTITIOUS
))
477 if (vm_page_busy_try(m
, FALSE
) == 0) {
478 vm_page_queues_spin_unlock(marker
.queue
);
480 vm_page_test_dirty(m
);
485 vm_page_queues_spin_lock(marker
.queue
);
488 TAILQ_REMOVE(&vm_page_queues
[marker
.queue
].pl
, &marker
, pageq
);
489 vm_page_queues_spin_unlock(marker
.queue
);
491 marker
.queue
= PQ_INACTIVE
+ q
;
493 vm_page_queues_spin_lock(marker
.queue
);
494 TAILQ_INSERT_HEAD(&vm_page_queues
[marker
.queue
].pl
,
497 while ((m
= TAILQ_NEXT(&marker
, pageq
)) != NULL
) {
499 &vm_page_queues
[marker
.queue
].pl
,
502 &vm_page_queues
[marker
.queue
].pl
,
504 if (m
->flags
& (PG_MARKER
| PG_FICTITIOUS
))
507 if (vm_page_busy_try(m
, FALSE
) == 0) {
508 vm_page_queues_spin_unlock(marker
.queue
);
510 vm_page_test_dirty(m
);
515 vm_page_queues_spin_lock(marker
.queue
);
518 TAILQ_REMOVE(&vm_page_queues
[marker
.queue
].pl
,
520 vm_page_queues_spin_unlock(marker
.queue
);
523 if (vmstats
.v_free_count
+ vmstats
.v_cache_count
+ pq_active_clean
+
524 pq_inactive_clean
+ vm_swap_size
< aligned_nblks
+ nswap_lowat
) {
525 mtx_unlock(&swap_mtx
);
530 * Prevent further allocations on this device
532 sp
->sw_flags
|= SW_CLOSING
;
533 for (dvbase
= dmmax
; dvbase
< aligned_nblks
; dvbase
+= dmmax
) {
534 blk
= min(aligned_nblks
- dvbase
, dmmax
);
535 vsbase
= index
* dmmax
+ dvbase
* nswdev
;
536 vm_swap_size
-= blist_fill(swapblist
, vsbase
, blk
);
541 * Page in the contents of the device and close it.
543 if (swap_pager_swapoff(index
) && swap_pager_swapoff(index
)) {
544 mtx_unlock(&swap_mtx
);
548 vn_lock(sp
->sw_vp
, LK_EXCLUSIVE
| LK_RETRY
);
549 VOP_CLOSE(sp
->sw_vp
, FREAD
| FWRITE
, NULL
);
550 vn_unlock(sp
->sw_vp
);
552 bzero(swdevt
+ index
, sizeof(struct swdevt
));
555 * Resize the bitmap based on the nem largest swap device,
556 * or free the bitmap if there are no more devices.
558 for (sp
= swdevt
, aligned_nblks
= 0; sp
< swdevt
+ nswdev
; sp
++) {
560 aligned_nblks
= max(aligned_nblks
, sp
->sw_nblks
);
563 nswap
= aligned_nblks
* nswdev
;
566 blist_destroy(swapblist
);
571 blist_resize(&swapblist
, nswap
, 0);
574 mtx_unlock(&swap_mtx
);
579 * Account for swap space in individual swdevt's. The caller ensures
580 * that the provided range falls into a single swdevt.
583 * -count space allocated
586 swapacctspace(swblk_t base
, swblk_t count
)
591 vm_swap_size
+= count
;
593 index
= seg
% nswdev
;
594 swdevt
[index
].sw_nused
-= count
;
601 sysctl_vm_swap_info(SYSCTL_HANDLER_ARGS
)
609 for (n
= 0; n
< nswdev
; ++n
) {
612 xs
.xsw_size
= sizeof(xs
);
613 xs
.xsw_version
= XSWDEV_VERSION
;
614 xs
.xsw_blksize
= PAGE_SIZE
;
615 xs
.xsw_dev
= sp
->sw_dev
;
616 xs
.xsw_flags
= sp
->sw_flags
;
617 xs
.xsw_nblks
= sp
->sw_nblks
;
618 xs
.xsw_used
= sp
->sw_nused
;
620 error
= SYSCTL_OUT(req
, &xs
, sizeof(xs
));
627 SYSCTL_INT(_vm
, OID_AUTO
, nswapdev
, CTLFLAG_RD
, &nswdev
, 0,
628 "Number of swap devices");
629 SYSCTL_NODE(_vm
, OID_AUTO
, swap_info_array
, CTLFLAG_RD
, sysctl_vm_swap_info
,
630 "Swap statistics by device");