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 1.36 2007/07/20 17:21:54 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/priv.h>
  46 #include <sys/nlookup.h>
  47 #include <sys/dmap.h>           /* XXX */
  48 #include <sys/vnode.h>
  49 #include <sys/fcntl.h>
  50 #include <sys/blist.h>
  51 #include <sys/kernel.h>
  52 #include <sys/lock.h>
  53 #include <sys/conf.h>
  54 #include <sys/stat.h>
  55 #include <sys/thread2.h>
  56 #include <vm/vm.h>
  57 #include <vm/vm_extern.h>
  58 #include <vm/swap_pager.h>
  59 #include <vm/vm_zone.h>
  60
  61 /*
  62  * Indirect driver for multi-controller paging.
  63  */
  64
  65 #ifndef NSWAPDEV
  66 #define NSWAPDEV        4
  67 #endif
  68 static struct swdevt should_be_malloced[NSWAPDEV];
  69 struct swdevt *swdevt = should_be_malloced;     /* exported to pstat/systat */
  70 static swblk_t nswap;           /* first block after the interleaved devs */
  71 int nswdev = NSWAPDEV;                          /* exported to pstat/systat */
  72 int vm_swap_size;
  73
  74 static int swapdev_strategy (struct vop_strategy_args *ap);
  75 struct vnode *swapdev_vp;
  76
  77 /*
  78  *      swapdev_strategy:
  79  *
  80  *      vn_strategy() for swapdev_vp.
  81  *      Perform swap strategy interleave device selection.
  82  *
  83  *      The bp is expected to be locked and on call.
  84  *
  85  *      (struct vnode *a_vp, struct bio *b_bio)
  86  */
  87
  88 static int
  89 swapdev_strategy(struct vop_strategy_args *ap)
  90 {
  91         struct bio *bio = ap->a_bio;
  92         struct bio *nbio;
  93         struct buf *bp = bio->bio_buf;
  94         int sz, off, seg, index, blkno, nblkno;
  95         struct swdevt *sp;
  96         struct vnode *vp;
  97
  98         vp = ap->a_vp;
  99         sz = howmany(bp->b_bcount, PAGE_SIZE);
 100         blkno = (int)(bio->bio_offset >> PAGE_SHIFT);
 101
 102         /*
 103          * Convert interleaved swap into per-device swap.  Note that
 104          * the block size is left in PAGE_SIZE'd chunks (for the newswap)
 105          * here.
 106          */
 107         nbio = push_bio(bio);
 108         if (nswdev > 1) {
 109                 off = blkno % dmmax;
 110                 if (off + sz > dmmax) {
 111                         bp->b_error = EINVAL;
 112                         bp->b_flags |= B_ERROR;
 113                         biodone(bio);
 114                         return 0;
 115                 }
 116                 seg = blkno / dmmax;
 117                 index = seg % nswdev;
 118                 seg /= nswdev;
 119                 nbio->bio_offset = (off_t)(seg * dmmax + off) << PAGE_SHIFT;
 120         } else {
 121                 index = 0;
 122                 nbio->bio_offset = bio->bio_offset;
 123         }
 124         nblkno = (int)(nbio->bio_offset >> PAGE_SHIFT);
 125         sp = &swdevt[index];
 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) */
 130                 biodone(bio);
 131                 return 0;
 132         }
 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) */
 137                 biodone(bio);
 138                 return 0;
 139         }
 140
 141         /*
 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
 144          * use the passed vp.
 145          *
 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.
 149          */
 150         vn_strategy(sp->sw_vp, nbio);
 151         return 0;
 152 }
 153
 154 /*
 155  * Create a special vnode op vector for swapdev_vp - we only use
 156  * vn_strategy(), everything else returns an error.
 157  */
 158 static struct vop_ops swapdev_vnode_vops = {
 159         .vop_default =          vop_defaultop,
 160         .vop_strategy =         swapdev_strategy
 161 };
 162 static struct vop_ops *swapdev_vnode_vops_p = &swapdev_vnode_vops;
 163
 164 VNODEOP_SET(swapdev_vnode_vops);
 165
 166 /*
 167  * swapon_args(char *name)
 168  *
 169  * System call swapon(name) enables swapping on device name,
 170  * which must be in the swdevsw.  Return EBUSY
 171  * if already swapping on this device.
 172  */
 173 /* ARGSUSED */
 174 int
 175 sys_swapon(struct swapon_args *uap)
 176 {
 177         struct thread *td = curthread;
 178         struct vattr attr;
 179         struct vnode *vp;
 180         struct nlookupdata nd;
 181         int error;
 182         struct ucred *cred;
 183
 184         KKASSERT(td->td_proc);
 185         cred = td->td_proc->p_ucred;
 186
 187         error = priv_check(td, PRIV_ROOT);
 188         if (error)
 189                 return (error);
 190
 191         vp = NULL;
 192         error = nlookup_init(&nd, uap->name, UIO_USERSPACE, NLC_FOLLOW);
 193         if (error == 0)
 194                 error = nlookup(&nd);
 195         if (error == 0)
 196                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
 197         nlookup_done(&nd);
 198         if (error)
 199                 return (error);
 200
 201         if (vn_isdisk(vp, &error))
 202                 error = swaponvp(td, vp, 0);
 203         else if (vp->v_type == VREG && vp->v_tag == VT_NFS &&
 204             (error = VOP_GETATTR(vp, &attr)) == 0) {
 205                 /*
 206                  * Allow direct swapping to NFS regular files in the same
 207                  * way that nfs_mountroot() sets up diskless swapping.
 208                  */
 209                 error = swaponvp(td, vp, attr.va_size / DEV_BSIZE);
 210         }
 211
 212         if (error)
 213                 vrele(vp);
 214
 215         return (error);
 216 }
 217
 218 /*
 219  * Swfree(index) frees the index'th portion of the swap map.
 220  * Each of the nswdev devices provides 1/nswdev'th of the swap
 221  * space, which is laid out with blocks of dmmax pages circularly
 222  * among the devices.
 223  *
 224  * The new swap code uses page-sized blocks.  The old swap code used
 225  * DEV_BSIZE'd chunks.
 226  *
 227  * XXX locking when multiple swapon's run in parallel
 228  */
 229 int
 230 swaponvp(struct thread *td, struct vnode *vp, u_quad_t nblks)
 231 {
 232         swblk_t aligned_nblks;
 233         int64_t dpsize;
 234         struct ucred *cred;
 235         struct swdevt *sp;
 236         swblk_t vsbase;
 237         swblk_t dvbase;
 238         cdev_t dev;
 239         int index;
 240         int error;
 241         long blk;
 242
 243         KKASSERT(td->td_proc);
 244         cred = td->td_proc->p_ucred;
 245
 246         if (!swapdev_vp) {
 247                 error = getspecialvnode(VT_NON, NULL, &swapdev_vnode_vops_p,
 248                                     &swapdev_vp, 0, 0);
 249                 if (error)
 250                         panic("Cannot get vnode for swapdev");
 251                 swapdev_vp->v_type = VNON;      /* Untyped */
 252                 vx_unlock(swapdev_vp);
 253         }
 254
 255         for (sp = swdevt, index = 0 ; index < nswdev; index++, sp++) {
 256                 if (sp->sw_vp == vp)
 257                         return EBUSY;
 258                 if (!sp->sw_vp)
 259                         goto found;
 260
 261         }
 262         return EINVAL;
 263     found:
 264         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 265         error = VOP_OPEN(vp, FREAD | FWRITE, cred, NULL);
 266         vn_unlock(vp);
 267         if (error)
 268                 return (error);
 269
 270         /*
 271          * v_rdev is not valid until after the VOP_OPEN() call.  dev_psize()
 272          * must be supported if a character device has been specified.
 273          */
 274         if (vp->v_type == VCHR)
 275                 dev = vp->v_rdev;
 276         else
 277                 dev = NULL;
 278
 279         if (nblks == 0 && dev != NULL) {
 280                 dpsize = dev_dpsize(dev);
 281                 if (dpsize == -1) {
 282                         VOP_CLOSE(vp, FREAD | FWRITE);
 283                         return (ENXIO);
 284                 }
 285                 nblks = (u_quad_t)dpsize;
 286         }
 287         if (nblks == 0) {
 288                 VOP_CLOSE(vp, FREAD | FWRITE);
 289                 return (ENXIO);
 290         }
 291
 292         /*
 293          * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks.
 294          * First chop nblks off to page-align it, then convert.
 295          *
 296          * sw->sw_nblks is in page-sized chunks now too.
 297          */
 298         nblks &= ~(u_quad_t)(ctodb(1) - 1);
 299         nblks = dbtoc(nblks);
 300
 301         /*
 302          * Post-conversion nblks must not be >= BLIST_MAXBLKS, and
 303          * we impose a 4-swap-device limit so we have to divide it out
 304          * further.  Going beyond this will result in overflows in the
 305          * blist code.
 306          *
 307          * Post-conversion nblks must fit within a (swblk_t), which
 308          * this test also ensures.
 309          */
 310         if (nblks > BLIST_MAXBLKS / nswdev) {
 311                 kprintf("exceeded maximum of %d blocks per swap unit\n",
 312                         (int)BLIST_MAXBLKS / nswdev);
 313                 VOP_CLOSE(vp, FREAD | FWRITE);
 314                 return (ENXIO);
 315         }
 316
 317         sp->sw_vp = vp;
 318         sp->sw_dev = dev2udev(dev);
 319         sp->sw_device = dev;
 320         sp->sw_flags |= SW_FREED;
 321         sp->sw_nblks = (swblk_t)nblks;
 322
 323         /*
 324          * nblks, nswap, and dmmax are PAGE_SIZE'd parameters now, not
 325          * DEV_BSIZE'd.   aligned_nblks is used to calculate the
 326          * size of the swap bitmap, taking into account the stripe size.
 327          */
 328         aligned_nblks = (swblk_t)((nblks + (dmmax - 1)) & ~(u_long)(dmmax - 1));
 329
 330         if (aligned_nblks * nswdev > nswap)
 331                 nswap = aligned_nblks * nswdev;
 332
 333         if (swapblist == NULL)
 334                 swapblist = blist_create(nswap);
 335         else
 336                 blist_resize(&swapblist, nswap, 0);
 337
 338         for (dvbase = dmmax; dvbase < nblks; dvbase += dmmax) {
 339                 blk = min(nblks - dvbase, dmmax);
 340                 vsbase = index * dmmax + dvbase * nswdev;
 341                 blist_free(swapblist, vsbase, blk);
 342                 vm_swap_size += blk;
 343         }
 344         swap_pager_newswap();
 345
 346         return (0);
 347 }