sys/vm/vm_swap.c

   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 $
  35  * $DragonFly: src/sys/vm/vm_swap.c,v 1.35 2007/05/15 22:44:21 dillon Exp $
  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/nlookup.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 <sys/thread2.h>
  55 #include <vm/vm.h>
  56 #include <vm/vm_extern.h>
  57 #include <vm/swap_pager.h>
  58 #include <vm/vm_zone.h>
  59
  60 /*
  61  * Indirect driver for multi-controller paging.
  62  */
  63
  64 #ifndef NSWAPDEV
  65 #define NSWAPDEV        4
  66 #endif
  67 static struct swdevt should_be_malloced[NSWAPDEV];
  68 struct swdevt *swdevt = should_be_malloced;     /* exported to pstat/systat */
  69 static int nswap;               /* first block after the interleaved devs */
  70 int nswdev = NSWAPDEV;                          /* exported to pstat/systat */
  71 int vm_swap_size;
  72
  73 static int swapdev_strategy (struct vop_strategy_args *ap);
  74 struct vnode *swapdev_vp;
  75
  76 /*
  77  *      swapdev_strategy:
  78  *
  79  *      vn_strategy() for swapdev_vp.
  80  *      Perform swap strategy interleave device selection.
  81  *
  82  *      The bp is expected to be locked and on call.
  83  *
  84  *      (struct vnode *a_vp, struct bio *b_bio)
  85  */
  86
  87 static int
  88 swapdev_strategy(struct vop_strategy_args *ap)
  89 {
  90         struct bio *bio = ap->a_bio;
  91         struct bio *nbio;
  92         struct buf *bp = bio->bio_buf;
  93         int sz, off, seg, index, blkno, nblkno;
  94         struct swdevt *sp;
  95         struct vnode *vp;
  96
  97         vp = ap->a_vp;
  98         sz = howmany(bp->b_bcount, PAGE_SIZE);
  99         blkno = (int)(bio->bio_offset >> PAGE_SHIFT);
 100
 101         /*
 102          * Convert interleaved swap into per-device swap.  Note that
 103          * the block size is left in PAGE_SIZE'd chunks (for the newswap)
 104          * here.
 105          */
 106         nbio = push_bio(bio);
 107         if (nswdev > 1) {
 108                 off = blkno % dmmax;
 109                 if (off + sz > dmmax) {
 110                         bp->b_error = EINVAL;
 111                         bp->b_flags |= B_ERROR;
 112                         biodone(bio);
 113                         return 0;
 114                 }
 115                 seg = blkno / dmmax;
 116                 index = seg % nswdev;
 117                 seg /= nswdev;
 118                 nbio->bio_offset = (off_t)(seg * dmmax + off) << PAGE_SHIFT;
 119         } else {
 120                 index = 0;
 121                 nbio->bio_offset = bio->bio_offset;
 122         }
 123         nblkno = (int)(nbio->bio_offset >> PAGE_SHIFT);
 124         sp = &swdevt[index];
 125         if (nblkno + sz > sp->sw_nblks) {
 126                 bp->b_error = EINVAL;
 127                 bp->b_flags |= B_ERROR;
 128                 /* I/O was never started on nbio, must biodone(bio) */
 129                 biodone(bio);
 130                 return 0;
 131         }
 132         if (sp->sw_vp == NULL) {
 133                 bp->b_error = ENODEV;
 134                 bp->b_flags |= B_ERROR;
 135                 /* I/O was never started on nbio, must biodone(bio) */
 136                 biodone(bio);
 137                 return 0;
 138         }
 139
 140         /*
 141          * Issue a strategy call on the appropriate swap vnode.  Note that
 142          * bp->b_vp is not modified.  Strategy code is always supposed to
 143          * use the passed vp.
 144          *
 145          * XXX do a dev_dstrategy() call on sp->sw_device instead of on
 146          * sp->sw_vp ?
 147          */
 148         vn_strategy(sp->sw_vp, nbio);
 149         return 0;
 150 }
 151
 152 /*
 153  * Create a special vnode op vector for swapdev_vp - we only use
 154  * vn_strategy(), everything else returns an error.
 155  */
 156 static struct vop_ops swapdev_vnode_vops = {
 157         .vop_default =          vop_defaultop,
 158         .vop_strategy =         swapdev_strategy
 159 };
 160 static struct vop_ops *swapdev_vnode_vops_p = &swapdev_vnode_vops;
 161
 162 VNODEOP_SET(swapdev_vnode_vops);
 163
 164 /*
 165  * swapon_args(char *name)
 166  *
 167  * System call swapon(name) enables swapping on device name,
 168  * which must be in the swdevsw.  Return EBUSY
 169  * if already swapping on this device.
 170  */
 171 /* ARGSUSED */
 172 int
 173 sys_swapon(struct swapon_args *uap)
 174 {
 175         struct thread *td = curthread;
 176         struct vattr attr;
 177         struct vnode *vp;
 178         struct nlookupdata nd;
 179         int error;
 180         struct ucred *cred;
 181
 182         KKASSERT(td->td_proc);
 183         cred = td->td_proc->p_ucred;
 184
 185         error = suser(td);
 186         if (error)
 187                 return (error);
 188
 189         vp = NULL;
 190         error = nlookup_init(&nd, uap->name, UIO_USERSPACE, NLC_FOLLOW);
 191         if (error == 0)
 192                 error = nlookup(&nd);
 193         if (error == 0)
 194                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
 195         nlookup_done(&nd);
 196         if (error)
 197                 return (error);
 198
 199         if (vn_isdisk(vp, &error))
 200                 error = swaponvp(td, vp, 0);
 201         else if (vp->v_type == VREG && vp->v_tag == VT_NFS &&
 202             (error = VOP_GETATTR(vp, &attr)) == 0) {
 203                 /*
 204                  * Allow direct swapping to NFS regular files in the same
 205                  * way that nfs_mountroot() sets up diskless swapping.
 206                  */
 207                 error = swaponvp(td, vp, attr.va_size / DEV_BSIZE);
 208         }
 209
 210         if (error)
 211                 vrele(vp);
 212
 213         return (error);
 214 }
 215
 216 /*
 217  * Swfree(index) frees the index'th portion of the swap map.
 218  * Each of the nswdev devices provides 1/nswdev'th of the swap
 219  * space, which is laid out with blocks of dmmax pages circularly
 220  * among the devices.
 221  *
 222  * The new swap code uses page-sized blocks.  The old swap code used
 223  * DEV_BSIZE'd chunks.
 224  *
 225  * XXX locking when multiple swapon's run in parallel
 226  */
 227 int
 228 swaponvp(struct thread *td, struct vnode *vp, u_long nblks)
 229 {
 230         u_long aligned_nblks;
 231         int64_t dpsize;
 232         struct ucred *cred;
 233         struct swdevt *sp;
 234         swblk_t vsbase;
 235         swblk_t dvbase;
 236         cdev_t dev;
 237         int index;
 238         int error;
 239         long blk;
 240
 241         KKASSERT(td->td_proc);
 242         cred = td->td_proc->p_ucred;
 243
 244         if (!swapdev_vp) {
 245                 error = getspecialvnode(VT_NON, NULL, &swapdev_vnode_vops_p,
 246                                     &swapdev_vp, 0, 0);
 247                 if (error)
 248                         panic("Cannot get vnode for swapdev");
 249                 swapdev_vp->v_type = VNON;      /* Untyped */
 250                 vx_unlock(swapdev_vp);
 251         }
 252
 253         for (sp = swdevt, index = 0 ; index < nswdev; index++, sp++) {
 254                 if (sp->sw_vp == vp)
 255                         return EBUSY;
 256                 if (!sp->sw_vp)
 257                         goto found;
 258
 259         }
 260         return EINVAL;
 261     found:
 262         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 263         error = VOP_OPEN(vp, FREAD | FWRITE, cred, NULL);
 264         vn_unlock(vp);
 265         if (error)
 266                 return (error);
 267
 268         /*
 269          * v_rdev is not valid until after the VOP_OPEN() call.  dev_psize()
 270          * must be supported if a character device has been specified.
 271          */
 272         if (vp->v_type == VCHR)
 273                 dev = vp->v_rdev;
 274         else
 275                 dev = NULL;
 276
 277         if (nblks == 0 && dev != NULL) {
 278                 dpsize = dev_dpsize(dev);
 279                 if (dpsize == -1) {
 280                         VOP_CLOSE(vp, FREAD | FWRITE);
 281                         return (ENXIO);
 282                 }
 283                 if ((u_int64_t)dpsize < 0x100000000ULL)
 284                         nblks = (u_long)dpsize;
 285                 else
 286                         nblks = 0xffffffffU;
 287         }
 288         if (nblks == 0) {
 289                 VOP_CLOSE(vp, FREAD | FWRITE);
 290                 return (ENXIO);
 291         }
 292
 293         /*
 294          * If we go beyond this, we get overflows in the radix
 295          * tree bitmap code.
 296          */
 297         if (nblks > 0x40000000 / BLIST_META_RADIX / nswdev) {
 298                 kprintf("exceeded maximum of %d blocks per swap unit\n",
 299                         0x40000000 / BLIST_META_RADIX / nswdev);
 300                 VOP_CLOSE(vp, FREAD | FWRITE);
 301                 return (ENXIO);
 302         }
 303         /*
 304          * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks.
 305          * First chop nblks off to page-align it, then convert.
 306          *
 307          * sw->sw_nblks is in page-sized chunks now too.
 308          */
 309         nblks &= ~(ctodb(1) - 1);
 310         nblks = dbtoc(nblks);
 311
 312         sp->sw_vp = vp;
 313         sp->sw_dev = dev2udev(dev);
 314         sp->sw_device = dev;
 315         sp->sw_flags |= SW_FREED;
 316         sp->sw_nblks = nblks;
 317
 318         /*
 319          * nblks, nswap, and dmmax are PAGE_SIZE'd parameters now, not
 320          * DEV_BSIZE'd.   aligned_nblks is used to calculate the
 321          * size of the swap bitmap, taking into account the stripe size.
 322          */
 323         aligned_nblks = (nblks + (dmmax - 1)) & ~(u_long)(dmmax - 1);
 324
 325         if (aligned_nblks * nswdev > nswap)
 326                 nswap = aligned_nblks * nswdev;
 327
 328         if (swapblist == NULL)
 329                 swapblist = blist_create(nswap);
 330         else
 331                 blist_resize(&swapblist, nswap, 0);
 332
 333         for (dvbase = dmmax; dvbase < nblks; dvbase += dmmax) {
 334                 blk = min(nblks - dvbase, dmmax);
 335                 vsbase = index * dmmax + dvbase * nswdev;
 336                 blist_free(swapblist, vsbase, blk);
 337                 vm_swap_size += blk;
 338         }
 339
 340         return (0);
 341 }