1 /* 2 * Copyright (c) 1982, 1986, 1989 Regents of the University of California. 3 * All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)vm_swap.c 7.28 (Berkeley) 04/19/93 8 */ 9 10 #include <sys/param.h> 11 #include <sys/systm.h> 12 #include <sys/buf.h> 13 #include <sys/conf.h> 14 #include <sys/proc.h> 15 #include <sys/namei.h> 16 #include <sys/dmap.h> /* XXX */ 17 #include <sys/vnode.h> 18 #include <sys/map.h> 19 #include <sys/file.h> 20 21 #include <miscfs/specfs/specdev.h> 22 23 /* 24 * Indirect driver for multi-controller paging. 25 */ 26 27 int nswap, nswdev; 28 29 /* 30 * Set up swap devices. 31 * Initialize linked list of free swap 32 * headers. These do not actually point 33 * to buffers, but rather to pages that 34 * are being swapped in and out. 35 */ 36 void 37 swapinit() 38 { 39 register int i; 40 register struct buf *sp = swbuf; 41 register struct proc *p = &proc0; /* XXX */ 42 struct swdevt *swp; 43 int error; 44 45 /* 46 * Count swap devices, and adjust total swap space available. 47 * Some of this space will not be available until a swapon() 48 * system is issued, usually when the system goes multi-user. 49 * 50 * If using NFS for swap, swdevt[0] will already be bdevvp'd. XXX 51 */ 52 nswdev = 0; 53 nswap = 0; 54 for (swp = swdevt; swp->sw_dev != NODEV || swp->sw_vp != NULL; swp++) { 55 nswdev++; 56 if (swp->sw_nblks > nswap) 57 nswap = swp->sw_nblks; 58 } 59 if (nswdev == 0) 60 panic("swapinit"); 61 if (nswdev > 1) 62 nswap = ((nswap + dmmax - 1) / dmmax) * dmmax; 63 nswap *= nswdev; 64 if (swdevt[0].sw_vp == NULL && 65 bdevvp(swdevt[0].sw_dev, &swdevt[0].sw_vp)) 66 panic("swapvp"); 67 if (error = swfree(p, 0)) { 68 printf("swfree errno %d\n", error); /* XXX */ 69 panic("swapinit swfree 0"); 70 } 71 72 /* 73 * Now set up swap buffer headers. 74 */ 75 bswlist.b_actf = sp; 76 for (i = 0; i < nswbuf - 1; i++, sp++) { 77 sp->b_actf = sp + 1; 78 sp->b_rcred = sp->b_wcred = p->p_ucred; 79 sp->b_vnbufs.qe_next = NOLIST; 80 } 81 sp->b_rcred = sp->b_wcred = p->p_ucred; 82 sp->b_vnbufs.qe_next = NOLIST; 83 sp->b_actf = NULL; 84 } 85 86 void 87 swstrategy(bp) 88 register struct buf *bp; 89 { 90 int sz, off, seg, index; 91 register struct swdevt *sp; 92 struct vnode *vp; 93 94 #ifdef GENERIC 95 /* 96 * A mini-root gets copied into the front of the swap 97 * and we run over top of the swap area just long 98 * enough for us to do a mkfs and restor of the real 99 * root (sure beats rewriting standalone restor). 100 */ 101 #define MINIROOTSIZE 4096 102 if (rootdev == dumpdev) 103 bp->b_blkno += MINIROOTSIZE; 104 #endif 105 sz = howmany(bp->b_bcount, DEV_BSIZE); 106 if (bp->b_blkno + sz > nswap) { 107 bp->b_flags |= B_ERROR; 108 biodone(bp); 109 return; 110 } 111 if (nswdev > 1) { 112 off = bp->b_blkno % dmmax; 113 if (off+sz > dmmax) { 114 bp->b_flags |= B_ERROR; 115 biodone(bp); 116 return; 117 } 118 seg = bp->b_blkno / dmmax; 119 index = seg % nswdev; 120 seg /= nswdev; 121 bp->b_blkno = seg*dmmax + off; 122 } else 123 index = 0; 124 sp = &swdevt[index]; 125 if ((bp->b_dev = sp->sw_dev) == NODEV) 126 panic("swstrategy"); 127 if (sp->sw_vp == NULL) { 128 bp->b_error |= B_ERROR; 129 biodone(bp); 130 return; 131 } 132 VHOLD(sp->sw_vp); 133 if ((bp->b_flags & B_READ) == 0) { 134 if (vp = bp->b_vp) { 135 vp->v_numoutput--; 136 if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) { 137 vp->v_flag &= ~VBWAIT; 138 wakeup((caddr_t)&vp->v_numoutput); 139 } 140 } 141 sp->sw_vp->v_numoutput++; 142 } 143 if (bp->b_vp != NULL) 144 brelvp(bp); 145 bp->b_vp = sp->sw_vp; 146 VOP_STRATEGY(bp); 147 } 148 149 /* 150 * System call swapon(name) enables swapping on device name, 151 * which must be in the swdevsw. Return EBUSY 152 * if already swapping on this device. 153 */ 154 struct swapon_args { 155 char *name; 156 }; 157 /* ARGSUSED */ 158 int 159 swapon(p, uap, retval) 160 struct proc *p; 161 struct swapon_args *uap; 162 int *retval; 163 { 164 register struct vnode *vp; 165 register struct swdevt *sp; 166 dev_t dev; 167 int error; 168 struct nameidata nd; 169 170 if (error = suser(p->p_ucred, &p->p_acflag)) 171 return (error); 172 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, uap->name, p); 173 if (error = namei(&nd)) 174 return (error); 175 vp = nd.ni_vp; 176 if (vp->v_type != VBLK) { 177 vrele(vp); 178 return (ENOTBLK); 179 } 180 dev = (dev_t)vp->v_rdev; 181 if (major(dev) >= nblkdev) { 182 vrele(vp); 183 return (ENXIO); 184 } 185 for (sp = &swdevt[0]; sp->sw_dev != NODEV; sp++) 186 if (sp->sw_dev == dev) { 187 if (sp->sw_freed) { 188 vrele(vp); 189 return (EBUSY); 190 } 191 sp->sw_vp = vp; 192 if (error = swfree(p, sp - swdevt)) { 193 vrele(vp); 194 return (error); 195 } 196 return (0); 197 } 198 vrele(vp); 199 return (EINVAL); 200 } 201 202 /* 203 * Swfree(index) frees the index'th portion of the swap map. 204 * Each of the nswdev devices provides 1/nswdev'th of the swap 205 * space, which is laid out with blocks of dmmax pages circularly 206 * among the devices. 207 */ 208 int 209 swfree(p, index) 210 struct proc *p; 211 int index; 212 { 213 register struct swdevt *sp; 214 register swblk_t vsbase; 215 register long blk; 216 struct vnode *vp; 217 register swblk_t dvbase; 218 register int nblks; 219 int error; 220 221 sp = &swdevt[index]; 222 vp = sp->sw_vp; 223 if (error = VOP_OPEN(vp, FREAD|FWRITE, p->p_ucred, p)) 224 return (error); 225 sp->sw_freed = 1; 226 nblks = sp->sw_nblks; 227 for (dvbase = 0; dvbase < nblks; dvbase += dmmax) { 228 blk = nblks - dvbase; 229 if ((vsbase = index*dmmax + dvbase*nswdev) >= nswap) 230 panic("swfree"); 231 if (blk > dmmax) 232 blk = dmmax; 233 if (vsbase == 0) { 234 /* 235 * First of all chunks... initialize the swapmap 236 * the second half of the hunk. 237 */ 238 rminit(swapmap, (long)(blk/2), (long)(blk/2), 239 "swap", nswapmap); 240 } else if (dvbase == 0) { 241 /* 242 * Don't use the first cluster of the device 243 * in case it starts with a label or boot block. 244 */ 245 rmfree(swapmap, blk - ctod(CLSIZE), 246 vsbase + ctod(CLSIZE)); 247 } else 248 rmfree(swapmap, blk, vsbase); 249 } 250 return (0); 251 } 252