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 int vm_swap_max;
  77
  78 static int swapdev_strategy (struct vop_strategy_args *ap);
  79 struct vnode *swapdev_vp;
  80
  81 /*
  82  *      swapdev_strategy:
  83  *
  84  *      vn_strategy() for swapdev_vp.
  85  *      Perform swap strategy interleave device selection.
  86  *
  87  *      The bp is expected to be locked and on call.
  88  *
  89  *      (struct vnode *a_vp, struct bio *b_bio)
  90  */
  91
  92 static int
  93 swapdev_strategy(struct vop_strategy_args *ap)
  94 {
  95         struct bio *bio = ap->a_bio;
  96         struct bio *nbio;
  97         struct buf *bp = bio->bio_buf;
  98         int sz, off, seg, index, blkno, nblkno;
  99         struct swdevt *sp;
 100         struct vnode *vp;
 101
 102         vp = ap->a_vp;
 103         sz = howmany(bp->b_bcount, PAGE_SIZE);
 104         blkno = (int)(bio->bio_offset >> PAGE_SHIFT);
 105
 106         /*
 107          * Convert interleaved swap into per-device swap.  Note that
 108          * the block size is left in PAGE_SIZE'd chunks (for the newswap)
 109          * here.
 110          */
 111         nbio = push_bio(bio);
 112         if (nswdev > 1) {
 113                 off = blkno % dmmax;
 114                 if (off + sz > dmmax) {
 115                         bp->b_error = EINVAL;
 116                         bp->b_flags |= B_ERROR;
 117                         biodone(bio);
 118                         return 0;
 119                 }
 120                 seg = blkno / dmmax;
 121                 index = seg % nswdev;
 122                 seg /= nswdev;
 123                 nbio->bio_offset = (off_t)(seg * dmmax + off) << PAGE_SHIFT;
 124         } else {
 125                 index = 0;
 126                 nbio->bio_offset = bio->bio_offset;
 127         }
 128         nblkno = (int)(nbio->bio_offset >> PAGE_SHIFT);
 129         sp = &swdevt[index];
 130         if (nblkno + sz > sp->sw_nblks) {
 131                 bp->b_error = EINVAL;
 132                 bp->b_flags |= B_ERROR;
 133                 /* I/O was never started on nbio, must biodone(bio) */
 134                 biodone(bio);
 135                 return 0;
 136         }
 137         if (sp->sw_vp == NULL) {
 138                 bp->b_error = ENODEV;
 139                 bp->b_flags |= B_ERROR;
 140                 /* I/O was never started on nbio, must biodone(bio) */
 141                 biodone(bio);
 142                 return 0;
 143         }
 144
 145         /*
 146          * Issue a strategy call on the appropriate swap vnode.  Note that
 147          * bp->b_vp is not modified.  Strategy code is always supposed to
 148          * use the passed vp.
 149          *
 150          * We have to use vn_strategy() here even if we know we have a
 151          * device in order to properly break up requests which exceed the
 152          * device's DMA limits.
 153          */
 154         vn_strategy(sp->sw_vp, nbio);
 155         return 0;
 156 }
 157
 158 /*
 159  * Create a special vnode op vector for swapdev_vp - we only use
 160  * vn_strategy(), everything else returns an error.
 161  */
 162 static struct vop_ops swapdev_vnode_vops = {
 163         .vop_default =          vop_defaultop,
 164         .vop_strategy =         swapdev_strategy
 165 };
 166 static struct vop_ops *swapdev_vnode_vops_p = &swapdev_vnode_vops;
 167
 168 VNODEOP_SET(swapdev_vnode_vops);
 169
 170 /*
 171  * swapon_args(char *name)
 172  *
 173  * System call swapon(name) enables swapping on device name,
 174  * which must be in the swdevsw.  Return EBUSY
 175  * if already swapping on this device.
 176  *
 177  * MPALMOSTSAFE
 178  */
 179 int
 180 sys_swapon(struct swapon_args *uap)
 181 {
 182         struct thread *td = curthread;
 183         struct vattr attr;
 184         struct vnode *vp;
 185         struct nlookupdata nd;
 186         int error;
 187         struct ucred *cred;
 188
 189         cred = td->td_ucred;
 190
 191         error = priv_check(td, PRIV_ROOT);
 192         if (error)
 193                 return (error);
 194
 195         get_mplock();
 196         vp = NULL;
 197         error = nlookup_init(&nd, uap->name, UIO_USERSPACE, NLC_FOLLOW);
 198         if (error == 0)
 199                 error = nlookup(&nd);
 200         if (error == 0)
 201                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
 202         nlookup_done(&nd);
 203         if (error) {
 204                 rel_mplock();
 205                 return (error);
 206         }
 207
 208         if (vn_isdisk(vp, &error)) {
 209                 error = swaponvp(td, vp, 0);
 210         } else if (vp->v_type == VREG && vp->v_tag == VT_NFS &&
 211                    (error = VOP_GETATTR(vp, &attr)) == 0) {
 212                 /*
 213                  * Allow direct swapping to NFS regular files in the same
 214                  * way that nfs_mountroot() sets up diskless swapping.
 215                  */
 216                 error = swaponvp(td, vp, attr.va_size / DEV_BSIZE);
 217         }
 218         if (error)
 219                 vrele(vp);
 220         rel_mplock();
 221
 222         return (error);
 223 }
 224
 225 /*
 226  * Swfree(index) frees the index'th portion of the swap map.
 227  * Each of the nswdev devices provides 1/nswdev'th of the swap
 228  * space, which is laid out with blocks of dmmax pages circularly
 229  * among the devices.
 230  *
 231  * The new swap code uses page-sized blocks.  The old swap code used
 232  * DEV_BSIZE'd chunks.
 233  *
 234  * XXX locking when multiple swapon's run in parallel
 235  */
 236 int
 237 swaponvp(struct thread *td, struct vnode *vp, u_quad_t nblks)
 238 {
 239         swblk_t aligned_nblks;
 240         int64_t dpsize;
 241         struct ucred *cred;
 242         struct swdevt *sp;
 243         swblk_t vsbase;
 244         swblk_t dvbase;
 245         cdev_t dev;
 246         int index;
 247         int error;
 248         long blk;
 249
 250         cred = td->td_ucred;
 251
 252         if (!swapdev_vp) {
 253                 error = getspecialvnode(VT_NON, NULL, &swapdev_vnode_vops_p,
 254                                     &swapdev_vp, 0, 0);
 255                 if (error)
 256                         panic("Cannot get vnode for swapdev");
 257                 swapdev_vp->v_type = VNON;      /* Untyped */
 258                 vx_unlock(swapdev_vp);
 259         }
 260
 261         for (sp = swdevt, index = 0 ; index < nswdev; index++, sp++) {
 262                 if (sp->sw_vp == vp)
 263                         return EBUSY;
 264                 if (!sp->sw_vp)
 265                         goto found;
 266
 267         }
 268         return EINVAL;
 269     found:
 270         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 271         error = VOP_OPEN(vp, FREAD | FWRITE, cred, NULL);
 272         vn_unlock(vp);
 273         if (error)
 274                 return (error);
 275
 276         /*
 277          * v_rdev is not valid until after the VOP_OPEN() call.  dev_psize()
 278          * must be supported if a character device has been specified.
 279          */
 280         if (vp->v_type == VCHR)
 281                 dev = vp->v_rdev;
 282         else
 283                 dev = NULL;
 284
 285         if (nblks == 0 && dev != NULL) {
 286                 dpsize = dev_dpsize(dev);
 287                 if (dpsize == -1) {
 288                         VOP_CLOSE(vp, FREAD | FWRITE);
 289                         return (ENXIO);
 290                 }
 291                 nblks = (u_quad_t)dpsize;
 292         }
 293         if (nblks == 0) {
 294                 VOP_CLOSE(vp, FREAD | FWRITE);
 295                 return (ENXIO);
 296         }
 297
 298         /*
 299          * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks.
 300          * First chop nblks off to page-align it, then convert.
 301          *
 302          * sw->sw_nblks is in page-sized chunks now too.
 303          */
 304         nblks &= ~(u_quad_t)(ctodb(1) - 1);
 305         nblks = dbtoc(nblks);
 306
 307         /*
 308          * Post-conversion nblks must not be >= BLIST_MAXBLKS, and
 309          * we impose a 4-swap-device limit so we have to divide it out
 310          * further.  Going beyond this will result in overflows in the
 311          * blist code.
 312          *
 313          * Post-conversion nblks must fit within a (swblk_t), which
 314          * this test also ensures.
 315          */
 316         if (nblks > BLIST_MAXBLKS / nswdev) {
 317                 kprintf("exceeded maximum of %d blocks per swap unit\n",
 318                         (int)BLIST_MAXBLKS / nswdev);
 319                 VOP_CLOSE(vp, FREAD | FWRITE);
 320                 return (ENXIO);
 321         }
 322
 323         sp->sw_vp = vp;
 324         sp->sw_dev = dev2udev(dev);
 325         sp->sw_device = dev;
 326         sp->sw_flags |= SW_FREED;
 327         sp->sw_nblks = (swblk_t)nblks;
 328
 329         /*
 330          * nblks, nswap, and dmmax are PAGE_SIZE'd parameters now, not
 331          * DEV_BSIZE'd.   aligned_nblks is used to calculate the
 332          * size of the swap bitmap, taking into account the stripe size.
 333          */
 334         aligned_nblks = (swblk_t)((nblks + (dmmax - 1)) & ~(u_long)(dmmax - 1));
 335
 336         if (aligned_nblks * nswdev > nswap)
 337                 nswap = aligned_nblks * nswdev;
 338
 339         if (swapblist == NULL)
 340                 swapblist = blist_create(nswap);
 341         else
 342                 blist_resize(&swapblist, nswap, 0);
 343
 344         for (dvbase = dmmax; dvbase < nblks; dvbase += dmmax) {
 345                 blk = min(nblks - dvbase, dmmax);
 346                 vsbase = index * dmmax + dvbase * nswdev;
 347                 blist_free(swapblist, vsbase, blk);
 348                 vm_swap_size += blk;
 349                 vm_swap_max += blk;
 350         }
 351         swap_pager_newswap();
 352
 353         return (0);
 354 }