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
  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 #include <sys/thread2.h>
  62 #include <sys/mplock2.h>
  63
  64 /*
  65  * Indirect driver for multi-controller paging.
  66  */
  67
  68 #ifndef NSWAPDEV
  69 #define NSWAPDEV        4
  70 #endif
  71 static struct swdevt should_be_malloced[NSWAPDEV];
  72 struct swdevt *swdevt = should_be_malloced;     /* exported to pstat/systat */
  73 static swblk_t nswap;           /* first block after the interleaved devs */
  74 int nswdev = NSWAPDEV;                          /* exported to pstat/systat */
  75 int vm_swap_size;
  76
  77 static int swapdev_strategy (struct vop_strategy_args *ap);
  78 struct vnode *swapdev_vp;
  79
  80 /*
  81  *      swapdev_strategy:
  82  *
  83  *      vn_strategy() for swapdev_vp.
  84  *      Perform swap strategy interleave device selection.
  85  *
  86  *      The bp is expected to be locked and on call.
  87  *
  88  *      (struct vnode *a_vp, struct bio *b_bio)
  89  */
  90
  91 static int
  92 swapdev_strategy(struct vop_strategy_args *ap)
  93 {
  94         struct bio *bio = ap->a_bio;
  95         struct bio *nbio;
  96         struct buf *bp = bio->bio_buf;
  97         int sz, off, seg, index, blkno, nblkno;
  98         struct swdevt *sp;
  99         struct vnode *vp;
 100
 101         vp = ap->a_vp;
 102         sz = howmany(bp->b_bcount, PAGE_SIZE);
 103         blkno = (int)(bio->bio_offset >> PAGE_SHIFT);
 104
 105         /*
 106          * Convert interleaved swap into per-device swap.  Note that
 107          * the block size is left in PAGE_SIZE'd chunks (for the newswap)
 108          * here.
 109          */
 110         nbio = push_bio(bio);
 111         if (nswdev > 1) {
 112                 off = blkno % dmmax;
 113                 if (off + sz > dmmax) {
 114                         bp->b_error = EINVAL;
 115                         bp->b_flags |= B_ERROR;
 116                         biodone(bio);
 117                         return 0;
 118                 }
 119                 seg = blkno / dmmax;
 120                 index = seg % nswdev;
 121                 seg /= nswdev;
 122                 nbio->bio_offset = (off_t)(seg * dmmax + off) << PAGE_SHIFT;
 123         } else {
 124                 index = 0;
 125                 nbio->bio_offset = bio->bio_offset;
 126         }
 127         nblkno = (int)(nbio->bio_offset >> PAGE_SHIFT);
 128         sp = &swdevt[index];
 129         if (nblkno + sz > sp->sw_nblks) {
 130                 bp->b_error = EINVAL;
 131                 bp->b_flags |= B_ERROR;
 132                 /* I/O was never started on nbio, must biodone(bio) */
 133                 biodone(bio);
 134                 return 0;
 135         }
 136         if (sp->sw_vp == NULL) {
 137                 bp->b_error = ENODEV;
 138                 bp->b_flags |= B_ERROR;
 139                 /* I/O was never started on nbio, must biodone(bio) */
 140                 biodone(bio);
 141                 return 0;
 142         }
 143
 144         /*
 145          * Issue a strategy call on the appropriate swap vnode.  Note that
 146          * bp->b_vp is not modified.  Strategy code is always supposed to
 147          * use the passed vp.
 148          *
 149          * We have to use vn_strategy() here even if we know we have a
 150          * device in order to properly break up requests which exceed the
 151          * device's DMA limits.
 152          */
 153         vn_strategy(sp->sw_vp, nbio);
 154         return 0;
 155 }
 156
 157 /*
 158  * Create a special vnode op vector for swapdev_vp - we only use
 159  * vn_strategy(), everything else returns an error.
 160  */
 161 static struct vop_ops swapdev_vnode_vops = {
 162         .vop_default =          vop_defaultop,
 163         .vop_strategy =         swapdev_strategy
 164 };
 165 static struct vop_ops *swapdev_vnode_vops_p = &swapdev_vnode_vops;
 166
 167 VNODEOP_SET(swapdev_vnode_vops);
 168
 169 /*
 170  * swapon_args(char *name)
 171  *
 172  * System call swapon(name) enables swapping on device name,
 173  * which must be in the swdevsw.  Return EBUSY
 174  * if already swapping on this device.
 175  *
 176  * MPALMOSTSAFE
 177  */
 178 int
 179 sys_swapon(struct swapon_args *uap)
 180 {
 181         struct thread *td = curthread;
 182         struct vattr attr;
 183         struct vnode *vp;
 184         struct nlookupdata nd;
 185         int error;
 186         struct ucred *cred;
 187
 188         cred = td->td_ucred;
 189
 190         error = priv_check(td, PRIV_ROOT);
 191         if (error)
 192                 return (error);
 193
 194         get_mplock();
 195         vp = NULL;
 196         error = nlookup_init(&nd, uap->name, UIO_USERSPACE, NLC_FOLLOW);
 197         if (error == 0)
 198                 error = nlookup(&nd);
 199         if (error == 0)
 200                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
 201         nlookup_done(&nd);
 202         if (error) {
 203                 rel_mplock();
 204                 return (error);
 205         }
 206
 207         if (vn_isdisk(vp, &error)) {
 208                 error = swaponvp(td, vp, 0);
 209         } else if (vp->v_type == VREG && vp->v_tag == VT_NFS &&
 210                    (error = VOP_GETATTR(vp, &attr)) == 0) {
 211                 /*
 212                  * Allow direct swapping to NFS regular files in the same
 213                  * way that nfs_mountroot() sets up diskless swapping.
 214                  */
 215                 error = swaponvp(td, vp, attr.va_size / DEV_BSIZE);
 216         }
 217         if (error)
 218                 vrele(vp);
 219         rel_mplock();
 220
 221         return (error);
 222 }
 223
 224 /*
 225  * Swfree(index) frees the index'th portion of the swap map.
 226  * Each of the nswdev devices provides 1/nswdev'th of the swap
 227  * space, which is laid out with blocks of dmmax pages circularly
 228  * among the devices.
 229  *
 230  * The new swap code uses page-sized blocks.  The old swap code used
 231  * DEV_BSIZE'd chunks.
 232  *
 233  * XXX locking when multiple swapon's run in parallel
 234  */
 235 int
 236 swaponvp(struct thread *td, struct vnode *vp, u_quad_t nblks)
 237 {
 238         swblk_t aligned_nblks;
 239         int64_t dpsize;
 240         struct ucred *cred;
 241         struct swdevt *sp;
 242         swblk_t vsbase;
 243         swblk_t dvbase;
 244         cdev_t dev;
 245         int index;
 246         int error;
 247         long blk;
 248
 249         cred = td->td_ucred;
 250
 251         if (!swapdev_vp) {
 252                 error = getspecialvnode(VT_NON, NULL, &swapdev_vnode_vops_p,
 253                                     &swapdev_vp, 0, 0);
 254                 if (error)
 255                         panic("Cannot get vnode for swapdev");
 256                 swapdev_vp->v_type = VNON;      /* Untyped */
 257                 vx_unlock(swapdev_vp);
 258         }
 259
 260         for (sp = swdevt, index = 0 ; index < nswdev; index++, sp++) {
 261                 if (sp->sw_vp == vp)
 262                         return EBUSY;
 263                 if (!sp->sw_vp)
 264                         goto found;
 265
 266         }
 267         return EINVAL;
 268     found:
 269         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 270         error = VOP_OPEN(vp, FREAD | FWRITE, cred, NULL);
 271         vn_unlock(vp);
 272         if (error)
 273                 return (error);
 274
 275         /*
 276          * v_rdev is not valid until after the VOP_OPEN() call.  dev_psize()
 277          * must be supported if a character device has been specified.
 278          */
 279         if (vp->v_type == VCHR)
 280                 dev = vp->v_rdev;
 281         else
 282                 dev = NULL;
 283
 284         if (nblks == 0 && dev != NULL) {
 285                 dpsize = dev_dpsize(dev);
 286                 if (dpsize == -1) {
 287                         VOP_CLOSE(vp, FREAD | FWRITE);
 288                         return (ENXIO);
 289                 }
 290                 nblks = (u_quad_t)dpsize;
 291         }
 292         if (nblks == 0) {
 293                 VOP_CLOSE(vp, FREAD | FWRITE);
 294                 return (ENXIO);
 295         }
 296
 297         /*
 298          * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks.
 299          * First chop nblks off to page-align it, then convert.
 300          *
 301          * sw->sw_nblks is in page-sized chunks now too.
 302          */
 303         nblks &= ~(u_quad_t)(ctodb(1) - 1);
 304         nblks = dbtoc(nblks);
 305
 306         /*
 307          * Post-conversion nblks must not be >= BLIST_MAXBLKS, and
 308          * we impose a 4-swap-device limit so we have to divide it out
 309          * further.  Going beyond this will result in overflows in the
 310          * blist code.
 311          *
 312          * Post-conversion nblks must fit within a (swblk_t), which
 313          * this test also ensures.
 314          */
 315         if (nblks > BLIST_MAXBLKS / nswdev) {
 316                 kprintf("exceeded maximum of %d blocks per swap unit\n",
 317                         (int)BLIST_MAXBLKS / nswdev);
 318                 VOP_CLOSE(vp, FREAD | FWRITE);
 319                 return (ENXIO);
 320         }
 321
 322         sp->sw_vp = vp;
 323         sp->sw_dev = dev2udev(dev);
 324         sp->sw_device = dev;
 325         sp->sw_flags |= SW_FREED;
 326         sp->sw_nblks = (swblk_t)nblks;
 327
 328         /*
 329          * nblks, nswap, and dmmax are PAGE_SIZE'd parameters now, not
 330          * DEV_BSIZE'd.   aligned_nblks is used to calculate the
 331          * size of the swap bitmap, taking into account the stripe size.
 332          */
 333         aligned_nblks = (swblk_t)((nblks + (dmmax - 1)) & ~(u_long)(dmmax - 1));
 334
 335         if (aligned_nblks * nswdev > nswap)
 336                 nswap = aligned_nblks * nswdev;
 337
 338         if (swapblist == NULL)
 339                 swapblist = blist_create(nswap);
 340         else
 341                 blist_resize(&swapblist, nswap, 0);
 342
 343         for (dvbase = dmmax; dvbase < nblks; dvbase += dmmax) {
 344                 blk = min(nblks - dvbase, dmmax);
 345                 vsbase = index * dmmax + dvbase * nswdev;
 346                 blist_free(swapblist, vsbase, blk);
 347                 vm_swap_size += blk;
 348         }
 349         swap_pager_newswap();
 350
 351         return (0);
 352 }