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. All advertising materials mentioning features or use of this software
16 * must display the following acknowledgement:
17 * This product includes software developed by the University of
18 * California, Berkeley and its contributors.
19 * 4. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * @(#)vm_swap.c 8.5 (Berkeley) 2/17/94
36 * $FreeBSD: src/sys/vm/vm_swap.c,v 1.96.2.2 2001/10/14 18:46:47 iedowse Exp $
37 * $DragonFly: src/sys/vm/vm_swap.c,v 1.36 2007/07/20 17:21:54 dillon Exp $
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/sysproto.h>
48 #include <sys/nlookup.h>
49 #include <sys/sysctl.h>
50 #include <sys/dmap.h> /* XXX */
51 #include <sys/vnode.h>
52 #include <sys/fcntl.h>
53 #include <sys/blist.h>
54 #include <sys/kernel.h>
60 #include <vm/vm_extern.h>
61 #include <vm/swap_pager.h>
62 #include <vm/vm_zone.h>
63 #include <vm/vm_param.h>
65 #include <sys/thread2.h>
66 #include <sys/mplock2.h>
67 #include <sys/mutex2.h>
68 #include <sys/spinlock2.h>
71 * Indirect driver for multi-controller paging.
77 static struct swdevt should_be_malloced[NSWAPDEV];
78 struct swdevt *swdevt = should_be_malloced; /* exported to pstat/systat */
79 static swblk_t nswap; /* first block after the interleaved devs */
80 static struct mtx swap_mtx = MTX_INITIALIZER;
81 int nswdev = NSWAPDEV; /* exported to pstat/systat */
85 static int swapoff_one(int index);
86 struct vnode *swapdev_vp;
89 * (struct vnode *a_vp, struct bio *b_bio)
91 * vn_strategy() for swapdev_vp. Perform swap strategy interleave device
97 swapdev_strategy(struct vop_strategy_args *ap)
99 struct bio *bio = ap->a_bio;
101 struct buf *bp = bio->bio_buf;
102 int sz, off, seg, index, blkno, nblkno;
107 sz = howmany(bp->b_bcount, PAGE_SIZE);
108 blkno = (int)(bio->bio_offset >> PAGE_SHIFT);
111 * Convert interleaved swap into per-device swap. Note that
112 * the block size is left in PAGE_SIZE'd chunks (for the newswap)
115 nbio = push_bio(bio);
118 if (off + sz > dmmax) {
119 bp->b_error = EINVAL;
120 bp->b_flags |= B_ERROR;
125 index = seg % nswdev;
127 nbio->bio_offset = (off_t)(seg * dmmax + off) << PAGE_SHIFT;
130 nbio->bio_offset = bio->bio_offset;
132 nblkno = (int)(nbio->bio_offset >> PAGE_SHIFT);
134 if (nblkno + sz > sp->sw_nblks) {
135 bp->b_error = EINVAL;
136 bp->b_flags |= B_ERROR;
137 /* I/O was never started on nbio, must biodone(bio) */
141 if (sp->sw_vp == NULL) {
142 bp->b_error = ENODEV;
143 bp->b_flags |= B_ERROR;
144 /* I/O was never started on nbio, must biodone(bio) */
150 * Issue a strategy call on the appropriate swap vnode. Note that
151 * bp->b_vp is not modified. Strategy code is always supposed to
154 * We have to use vn_strategy() here even if we know we have a
155 * device in order to properly break up requests which exceed the
156 * device's DMA limits.
158 vn_strategy(sp->sw_vp, nbio);
163 swapdev_inactive(struct vop_inactive_args *ap)
170 swapdev_reclaim(struct vop_reclaim_args *ap)
176 * Create a special vnode op vector for swapdev_vp - we only use
177 * vn_strategy(), everything else returns an error.
179 static struct vop_ops swapdev_vnode_vops = {
180 .vop_default = vop_defaultop,
181 .vop_strategy = swapdev_strategy,
182 .vop_inactive = swapdev_inactive,
183 .vop_reclaim = swapdev_reclaim
185 static struct vop_ops *swapdev_vnode_vops_p = &swapdev_vnode_vops;
187 VNODEOP_SET(swapdev_vnode_vops);
190 * swapon_args(char *name)
192 * System call swapon(name) enables swapping on device name,
193 * which must be in the swdevsw. Return EBUSY
194 * if already swapping on this device.
199 sys_swapon(struct swapon_args *uap)
201 struct thread *td = curthread;
204 struct nlookupdata nd;
210 error = priv_check(td, PRIV_ROOT);
217 error = nlookup_init(&nd, uap->name, UIO_USERSPACE, NLC_FOLLOW);
219 error = nlookup(&nd);
221 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
225 mtx_unlock(&swap_mtx);
229 if (vn_isdisk(vp, &error)) {
230 error = swaponvp(td, vp, 0);
231 } else if (vp->v_type == VREG && vp->v_tag == VT_NFS &&
232 (error = VOP_GETATTR(vp, &attr)) == 0) {
234 * Allow direct swapping to NFS regular files in the same
235 * way that nfs_mountroot() sets up diskless swapping.
237 error = swaponvp(td, vp, attr.va_size / DEV_BSIZE);
242 mtx_unlock(&swap_mtx);
248 * Swfree(index) frees the index'th portion of the swap map.
249 * Each of the nswdev devices provides 1/nswdev'th of the swap
250 * space, which is laid out with blocks of dmmax pages circularly
253 * The new swap code uses page-sized blocks. The old swap code used
254 * DEV_BSIZE'd chunks.
256 * XXX locking when multiple swapon's run in parallel
259 swaponvp(struct thread *td, struct vnode *vp, u_quad_t nblks)
261 swblk_t aligned_nblks;
277 error = getspecialvnode(VT_NON, NULL, &swapdev_vnode_vops_p,
280 panic("Cannot get vnode for swapdev");
281 swapdev_vp->v_type = VNON; /* Untyped */
282 vx_unlock(swapdev_vp);
285 for (sp = swdevt, index = 0 ; index < nswdev; index++, sp++) {
286 if (sp->sw_vp == vp) {
287 mtx_unlock(&swap_mtx);
294 mtx_unlock(&swap_mtx);
297 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
298 error = VOP_OPEN(vp, FREAD | FWRITE, cred, NULL);
301 mtx_unlock(&swap_mtx);
306 * v_rdev is not valid until after the VOP_OPEN() call. dev_psize()
307 * must be supported if a character device has been specified.
309 if (vp->v_type == VCHR)
314 if (nblks == 0 && dev != NULL) {
315 dpsize = dev_dpsize(dev);
317 VOP_CLOSE(vp, FREAD | FWRITE);
318 mtx_unlock(&swap_mtx);
321 nblks = (u_quad_t)dpsize;
324 VOP_CLOSE(vp, FREAD | FWRITE);
325 mtx_unlock(&swap_mtx);
330 * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks.
331 * First chop nblks off to page-align it, then convert.
333 * sw->sw_nblks is in page-sized chunks now too.
335 nblks &= ~(u_quad_t)(ctodb(1) - 1);
336 nblks = dbtoc(nblks);
339 * Post-conversion nblks must not be >= BLIST_MAXBLKS, and
340 * we impose a 4-swap-device limit so we have to divide it out
341 * further. Going beyond this will result in overflows in the
344 * Post-conversion nblks must fit within a (swblk_t), which
345 * this test also ensures.
347 if (nblks > BLIST_MAXBLKS / nswdev) {
348 kprintf("exceeded maximum of %d blocks per swap unit\n",
349 (int)BLIST_MAXBLKS / nswdev);
350 VOP_CLOSE(vp, FREAD | FWRITE);
351 mtx_unlock(&swap_mtx);
356 sp->sw_dev = dev2udev(dev);
358 sp->sw_flags = SW_FREED;
362 * nblks, nswap, and dmmax are PAGE_SIZE'd parameters now, not
363 * DEV_BSIZE'd. aligned_nblks is used to calculate the
364 * size of the swap bitmap, taking into account the stripe size.
366 aligned_nblks = (swblk_t)((nblks + (dmmax - 1)) & ~(u_long)(dmmax - 1));
367 sp->sw_nblks = aligned_nblks;
369 if (aligned_nblks * nswdev > nswap)
370 nswap = aligned_nblks * nswdev;
372 if (swapblist == NULL)
373 swapblist = blist_create(nswap);
375 blist_resize(&swapblist, nswap, 0);
377 for (dvbase = dmmax; dvbase < aligned_nblks; dvbase += dmmax) {
378 blk = min(aligned_nblks - dvbase, dmmax);
379 vsbase = index * dmmax + dvbase * nswdev;
380 blist_free(swapblist, vsbase, blk);
384 swap_pager_newswap();
386 mtx_unlock(&swap_mtx);
391 * swapoff_args(char *name)
393 * System call swapoff(name) disables swapping on device name,
394 * which must be an active swap device. Return ENOMEM
395 * if there is not enough memory to page in the contents of
401 sys_swapoff(struct swapoff_args *uap)
404 struct nlookupdata nd;
408 error = priv_check(curthread, PRIV_ROOT);
415 error = nlookup_init(&nd, uap->name, UIO_USERSPACE, NLC_FOLLOW);
417 error = nlookup(&nd);
419 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
424 for (sp = swdevt, index = 0; index < nswdev; index++, sp++) {
431 error = swapoff_one(index);
435 mtx_unlock(&swap_mtx);
440 swapoff_one(int index)
442 swblk_t blk, aligned_nblks;
443 swblk_t dvbase, vsbase;
444 u_int pq_active_clean, pq_inactive_clean;
446 struct vm_page marker;
453 aligned_nblks = sp->sw_nblks;
454 pq_active_clean = pq_inactive_clean = 0;
457 * We can turn off this swap device safely only if the
458 * available virtual memory in the system will fit the amount
459 * of data we will have to page back in, plus an epsilon so
460 * the system doesn't become critically low on swap space.
462 for (q = 0; q < PQ_L2_SIZE; ++q) {
463 bzero(&marker, sizeof(marker));
464 marker.flags = PG_BUSY | PG_FICTITIOUS | PG_MARKER;
465 marker.queue = PQ_ACTIVE + q;
467 marker.wire_count = 1;
469 vm_page_queues_spin_lock(marker.queue);
470 TAILQ_INSERT_HEAD(&vm_page_queues[marker.queue].pl,
473 while ((m = TAILQ_NEXT(&marker, pageq)) != NULL) {
474 TAILQ_REMOVE(&vm_page_queues[marker.queue].pl,
476 TAILQ_INSERT_AFTER(&vm_page_queues[marker.queue].pl, m,
478 if (m->flags & (PG_MARKER | PG_FICTITIOUS))
481 if (vm_page_busy_try(m, FALSE) == 0) {
482 vm_page_queues_spin_unlock(marker.queue);
484 vm_page_test_dirty(m);
489 vm_page_queues_spin_lock(marker.queue);
492 TAILQ_REMOVE(&vm_page_queues[marker.queue].pl, &marker, pageq);
493 vm_page_queues_spin_unlock(marker.queue);
495 marker.queue = PQ_INACTIVE + q;
497 vm_page_queues_spin_lock(marker.queue);
498 TAILQ_INSERT_HEAD(&vm_page_queues[marker.queue].pl,
501 while ((m = TAILQ_NEXT(&marker, pageq)) != NULL) {
503 &vm_page_queues[marker.queue].pl,
506 &vm_page_queues[marker.queue].pl,
508 if (m->flags & (PG_MARKER | PG_FICTITIOUS))
511 if (vm_page_busy_try(m, FALSE) == 0) {
512 vm_page_queues_spin_unlock(marker.queue);
514 vm_page_test_dirty(m);
519 vm_page_queues_spin_lock(marker.queue);
522 TAILQ_REMOVE(&vm_page_queues[marker.queue].pl,
524 vm_page_queues_spin_unlock(marker.queue);
527 if (vmstats.v_free_count + vmstats.v_cache_count + pq_active_clean +
528 pq_inactive_clean + vm_swap_size < aligned_nblks + nswap_lowat) {
529 mtx_unlock(&swap_mtx);
534 * Prevent further allocations on this device
536 sp->sw_flags |= SW_CLOSING;
537 for (dvbase = dmmax; dvbase < aligned_nblks; dvbase += dmmax) {
538 blk = min(aligned_nblks - dvbase, dmmax);
539 vsbase = index * dmmax + dvbase * nswdev;
540 vm_swap_size -= blist_fill(swapblist, vsbase, blk);
545 * Page in the contents of the device and close it.
547 if (swap_pager_swapoff(index)) {
548 mtx_unlock(&swap_mtx);
552 VOP_CLOSE(sp->sw_vp, FREAD | FWRITE);
554 bzero(swdevt + index, sizeof(struct swdevt));
557 * Resize the bitmap based on the nem largest swap device,
558 * or free the bitmap if there are no more devices.
560 for (sp = swdevt, aligned_nblks = 0; sp < swdevt + nswdev; sp++) {
562 aligned_nblks = max(aligned_nblks, sp->sw_nblks);
565 nswap = aligned_nblks * nswdev;
568 blist_destroy(swapblist);
573 blist_resize(&swapblist, nswap, 0);
576 mtx_unlock(&swap_mtx);
581 * Account for swap space in individual swdevt's. The caller ensures
582 * that the provided range falls into a single swdevt.
585 * -count space allocated
588 swapacctspace(swblk_t base, swblk_t count)
593 vm_swap_size += count;
595 index = seg % nswdev;
596 swdevt[index].sw_nused -= count;
603 sysctl_vm_swap_info(SYSCTL_HANDLER_ARGS)
611 for (n = 0; n < nswdev; ++n) {
614 xs.xsw_size = sizeof(xs);
615 xs.xsw_version = XSWDEV_VERSION;
616 xs.xsw_blksize = PAGE_SIZE;
617 xs.xsw_dev = sp->sw_dev;
618 xs.xsw_flags = sp->sw_flags;
619 xs.xsw_nblks = sp->sw_nblks;
620 xs.xsw_used = sp->sw_nused;
622 error = SYSCTL_OUT(req, &xs, sizeof(xs));
629 SYSCTL_INT(_vm, OID_AUTO, nswapdev, CTLFLAG_RD, &nswdev, 0,
630 "Number of swap devices");
631 SYSCTL_NODE(_vm, OID_AUTO, swap_info_array, CTLFLAG_RD, sysctl_vm_swap_info,
632 "Swap statistics by device");