proc->thread stage 4: rework the VFS and DEVICE subsystems to take thread
[dragonfly.git] / sys / vm / vm_swap.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)vm_swap.c 8.5 (Berkeley) 2/17/94
34 * $FreeBSD: src/sys/vm/vm_swap.c,v 1.96.2.2 2001/10/14 18:46:47 iedowse Exp $
dadab5e9 35 * $DragonFly: src/sys/vm/vm_swap.c,v 1.4 2003/06/25 03:56:13 dillon Exp $
984263bc
MD
36 */
37
38#include "opt_swap.h"
39
40#include <sys/param.h>
41#include <sys/systm.h>
42#include <sys/sysproto.h>
43#include <sys/buf.h>
44#include <sys/proc.h>
45#include <sys/namei.h>
46#include <sys/dmap.h> /* XXX */
47#include <sys/vnode.h>
48#include <sys/fcntl.h>
49#include <sys/blist.h>
50#include <sys/kernel.h>
51#include <sys/lock.h>
52#include <sys/conf.h>
53#include <sys/stat.h>
54#include <vm/vm.h>
55#include <vm/vm_extern.h>
56#include <vm/swap_pager.h>
57#include <vm/vm_zone.h>
58
59/*
60 * Indirect driver for multi-controller paging.
61 */
62
63#ifndef NSWAPDEV
64#define NSWAPDEV 4
65#endif
66static struct swdevt should_be_malloced[NSWAPDEV];
67static struct swdevt *swdevt = should_be_malloced;
68static int nswap; /* first block after the interleaved devs */
69static int nswdev = NSWAPDEV;
70int vm_swap_size;
71
72static int swapdev_strategy __P((struct vop_strategy_args *ap));
73struct vnode *swapdev_vp;
74
75/*
76 * swapdev_strategy:
77 *
78 * VOP_STRATEGY() for swapdev_vp.
79 * Perform swap strategy interleave device selection.
80 *
81 * The bp is expected to be locked and *not* B_DONE on call.
82 */
83
84static int
85swapdev_strategy(ap)
86 struct vop_strategy_args /* {
87 struct vnode *a_vp;
88 struct buf *a_bp;
89 } */ *ap;
90{
91 int s, sz, off, seg, index;
92 register struct swdevt *sp;
93 struct vnode *vp;
94 struct buf *bp;
95
96 bp = ap->a_bp;
97 sz = howmany(bp->b_bcount, PAGE_SIZE);
98
99 /*
100 * Convert interleaved swap into per-device swap. Note that
101 * the block size is left in PAGE_SIZE'd chunks (for the newswap)
102 * here.
103 */
104 if (nswdev > 1) {
105 off = bp->b_blkno % dmmax;
106 if (off + sz > dmmax) {
107 bp->b_error = EINVAL;
108 bp->b_flags |= B_ERROR;
109 biodone(bp);
110 return 0;
111 }
112 seg = bp->b_blkno / dmmax;
113 index = seg % nswdev;
114 seg /= nswdev;
115 bp->b_blkno = seg * dmmax + off;
116 } else {
117 index = 0;
118 }
119 sp = &swdevt[index];
120 if (bp->b_blkno + sz > sp->sw_nblks) {
121 bp->b_error = EINVAL;
122 bp->b_flags |= B_ERROR;
123 biodone(bp);
124 return 0;
125 }
126 bp->b_dev = sp->sw_device;
127 if (sp->sw_vp == NULL) {
128 bp->b_error = ENODEV;
129 bp->b_flags |= B_ERROR;
130 biodone(bp);
131 return 0;
132 }
133
134 /*
135 * Convert from PAGE_SIZE'd to DEV_BSIZE'd chunks for the actual I/O
136 */
137 bp->b_blkno = ctodb(bp->b_blkno);
138
139 vhold(sp->sw_vp);
140 s = splvm();
141 if ((bp->b_flags & B_READ) == 0) {
142 vp = bp->b_vp;
143 if (vp) {
144 vp->v_numoutput--;
145 if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
146 vp->v_flag &= ~VBWAIT;
147 wakeup(&vp->v_numoutput);
148 }
149 }
150 sp->sw_vp->v_numoutput++;
151 }
152 pbreassignbuf(bp, sp->sw_vp);
153 splx(s);
154 VOP_STRATEGY(bp->b_vp, bp);
155 return 0;
156}
157
158/*
159 * Create a special vnode op vector for swapdev_vp - we only use
160 * VOP_STRATEGY(), everything else returns an error.
161 */
162vop_t **swapdev_vnodeop_p;
163static struct vnodeopv_entry_desc swapdev_vnodeop_entries[] = {
164 { &vop_default_desc, (vop_t *) vop_defaultop },
165 { &vop_strategy_desc, (vop_t *) swapdev_strategy },
166 { NULL, NULL }
167};
168static struct vnodeopv_desc swapdev_vnodeop_opv_desc =
169 { &swapdev_vnodeop_p, swapdev_vnodeop_entries };
170
171VNODEOP_SET(swapdev_vnodeop_opv_desc);
172
173/*
41c20dac
MD
174 * swapon_args(char *name)
175 *
984263bc
MD
176 * System call swapon(name) enables swapping on device name,
177 * which must be in the swdevsw. Return EBUSY
178 * if already swapping on this device.
179 */
984263bc
MD
180/* ARGSUSED */
181int
41c20dac 182swapon(struct swapon_args *uap)
984263bc 183{
dadab5e9 184 struct thread *td = curthread;
984263bc
MD
185 struct vattr attr;
186 register struct vnode *vp;
187 struct nameidata nd;
188 int error;
dadab5e9 189 struct ucred *cred;
984263bc 190
dadab5e9
MD
191 KKASSERT(td->td_proc);
192 cred = td->td_proc->p_ucred;
193
194 error = suser(td);
984263bc
MD
195 if (error)
196 return (error);
197
dadab5e9 198 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, uap->name, td);
984263bc
MD
199 error = namei(&nd);
200 if (error)
201 return (error);
202
203 NDFREE(&nd, NDF_ONLY_PNBUF);
204 vp = nd.ni_vp;
205
206 if (vn_isdisk(vp, &error))
dadab5e9 207 error = swaponvp(td, vp, vp->v_rdev, 0);
984263bc 208 else if (vp->v_type == VREG && vp->v_tag == VT_NFS &&
dadab5e9 209 (error = VOP_GETATTR(vp, &attr, cred, td)) == 0) {
984263bc
MD
210 /*
211 * Allow direct swapping to NFS regular files in the same
212 * way that nfs_mountroot() sets up diskless swapping.
213 */
dadab5e9 214 error = swaponvp(td, vp, NODEV, attr.va_size / DEV_BSIZE);
984263bc
MD
215 }
216
217 if (error)
218 vrele(vp);
219
220 return (error);
221}
222
223/*
224 * Swfree(index) frees the index'th portion of the swap map.
225 * Each of the nswdev devices provides 1/nswdev'th of the swap
226 * space, which is laid out with blocks of dmmax pages circularly
227 * among the devices.
228 *
229 * The new swap code uses page-sized blocks. The old swap code used
230 * DEV_BSIZE'd chunks.
231 *
232 * XXX locking when multiple swapon's run in parallel
233 */
234int
dadab5e9 235swaponvp(struct thread *td, struct vnode *vp, dev_t dev, u_long nblks)
984263bc
MD
236{
237 int index;
238 register struct swdevt *sp;
239 register swblk_t vsbase;
240 register long blk;
241 swblk_t dvbase;
242 int error;
243 u_long aligned_nblks;
dadab5e9
MD
244 struct ucred *cred;
245
246 KKASSERT(td->td_proc);
247 cred = td->td_proc->p_ucred;
984263bc
MD
248
249 if (!swapdev_vp) {
250 error = getnewvnode(VT_NON, NULL, swapdev_vnodeop_p,
251 &swapdev_vp);
252 if (error)
253 panic("Cannot get vnode for swapdev");
254 swapdev_vp->v_type = VNON; /* Untyped */
255 }
256
257 ASSERT_VOP_UNLOCKED(vp, "swaponvp");
258 for (sp = swdevt, index = 0 ; index < nswdev; index++, sp++) {
259 if (sp->sw_vp == vp)
260 return EBUSY;
261 if (!sp->sw_vp)
262 goto found;
263
264 }
265 return EINVAL;
266 found:
dadab5e9
MD
267 (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
268 error = VOP_OPEN(vp, FREAD | FWRITE, cred, td);
269 (void) VOP_UNLOCK(vp, 0, td);
984263bc
MD
270 if (error)
271 return (error);
272
273 if (nblks == 0 && dev != NODEV && (devsw(dev)->d_psize == 0 ||
274 (nblks = (*devsw(dev)->d_psize) (dev)) == -1)) {
dadab5e9 275 (void) VOP_CLOSE(vp, FREAD | FWRITE, cred, td);
984263bc
MD
276 return (ENXIO);
277 }
278 if (nblks == 0) {
dadab5e9 279 (void) VOP_CLOSE(vp, FREAD | FWRITE, cred, td);
984263bc
MD
280 return (ENXIO);
281 }
282
283 /*
284 * If we go beyond this, we get overflows in the radix
285 * tree bitmap code.
286 */
287 if (nblks > 0x40000000 / BLIST_META_RADIX / nswdev) {
288 printf("exceeded maximum of %d blocks per swap unit\n",
289 0x40000000 / BLIST_META_RADIX / nswdev);
dadab5e9 290 (void) VOP_CLOSE(vp, FREAD | FWRITE, cred, td);
984263bc
MD
291 return (ENXIO);
292 }
293 /*
294 * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks.
295 * First chop nblks off to page-align it, then convert.
296 *
297 * sw->sw_nblks is in page-sized chunks now too.
298 */
299 nblks &= ~(ctodb(1) - 1);
300 nblks = dbtoc(nblks);
301
302 sp->sw_vp = vp;
303 sp->sw_dev = dev2udev(dev);
304 sp->sw_device = dev;
305 sp->sw_flags |= SW_FREED;
306 sp->sw_nblks = nblks;
307
308 /*
309 * nblks, nswap, and dmmax are PAGE_SIZE'd parameters now, not
310 * DEV_BSIZE'd. aligned_nblks is used to calculate the
311 * size of the swap bitmap, taking into account the stripe size.
312 */
313 aligned_nblks = (nblks + (dmmax - 1)) & ~(u_long)(dmmax - 1);
314
315 if (aligned_nblks * nswdev > nswap)
316 nswap = aligned_nblks * nswdev;
317
318 if (swapblist == NULL)
319 swapblist = blist_create(nswap);
320 else
321 blist_resize(&swapblist, nswap, 0);
322
323 for (dvbase = dmmax; dvbase < nblks; dvbase += dmmax) {
324 blk = min(nblks - dvbase, dmmax);
325 vsbase = index * dmmax + dvbase * nswdev;
326 blist_free(swapblist, vsbase, blk);
327 vm_swap_size += blk;
328 }
329
330 return (0);
331}