1 /* 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)vm_swap.c 8.4 (Berkeley) 01/06/94 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 #ifdef SEQSWAP 29 int niswdev; /* number of interleaved swap devices */ 30 int niswap; /* size of interleaved swap area */ 31 #endif 32 33 /* 34 * Set up swap devices. 35 * Initialize linked list of free swap 36 * headers. These do not actually point 37 * to buffers, but rather to pages that 38 * are being swapped in and out. 39 */ 40 void 41 swapinit() 42 { 43 register int i; 44 register struct buf *sp = swbuf; 45 register struct proc *p = &proc0; /* XXX */ 46 struct swdevt *swp; 47 int error; 48 49 /* 50 * Count swap devices, and adjust total swap space available. 51 * Some of the space will not be countable until later (dynamically 52 * configurable devices) and some of the counted space will not be 53 * available until a swapon() system call is issued, both usually 54 * happen when the system goes multi-user. 55 * 56 * If using NFS for swap, swdevt[0] will already be bdevvp'd. XXX 57 */ 58 #ifdef SEQSWAP 59 nswdev = niswdev = 0; 60 nswap = niswap = 0; 61 /* 62 * All interleaved devices must come first 63 */ 64 for (swp = swdevt; swp->sw_dev != NODEV || swp->sw_vp != NULL; swp++) { 65 if (swp->sw_flags & SW_SEQUENTIAL) 66 break; 67 niswdev++; 68 if (swp->sw_nblks > niswap) 69 niswap = swp->sw_nblks; 70 } 71 niswap = roundup(niswap, dmmax); 72 niswap *= niswdev; 73 if (swdevt[0].sw_vp == NULL && 74 bdevvp(swdevt[0].sw_dev, &swdevt[0].sw_vp)) 75 panic("swapvp"); 76 /* 77 * The remainder must be sequential 78 */ 79 for ( ; swp->sw_dev != NODEV; swp++) { 80 if ((swp->sw_flags & SW_SEQUENTIAL) == 0) 81 panic("binit: mis-ordered swap devices"); 82 nswdev++; 83 if (swp->sw_nblks > 0) { 84 if (swp->sw_nblks % dmmax) 85 swp->sw_nblks -= (swp->sw_nblks % dmmax); 86 nswap += swp->sw_nblks; 87 } 88 } 89 nswdev += niswdev; 90 if (nswdev == 0) 91 panic("swapinit"); 92 nswap += niswap; 93 #else 94 nswdev = 0; 95 nswap = 0; 96 for (swp = swdevt; swp->sw_dev != NODEV || swp->sw_vp != NULL; swp++) { 97 nswdev++; 98 if (swp->sw_nblks > nswap) 99 nswap = swp->sw_nblks; 100 } 101 if (nswdev == 0) 102 panic("swapinit"); 103 if (nswdev > 1) 104 nswap = ((nswap + dmmax - 1) / dmmax) * dmmax; 105 nswap *= nswdev; 106 if (swdevt[0].sw_vp == NULL && 107 bdevvp(swdevt[0].sw_dev, &swdevt[0].sw_vp)) 108 panic("swapvp"); 109 #endif 110 if (nswap == 0) 111 printf("WARNING: no swap space found\n"); 112 else if (error = swfree(p, 0)) { 113 printf("swfree errno %d\n", error); /* XXX */ 114 panic("swapinit swfree 0"); 115 } 116 117 /* 118 * Now set up swap buffer headers. 119 */ 120 bswlist.b_actf = sp; 121 for (i = 0; i < nswbuf - 1; i++, sp++) { 122 sp->b_actf = sp + 1; 123 sp->b_rcred = sp->b_wcred = p->p_ucred; 124 sp->b_vnbufs.le_next = NOLIST; 125 } 126 sp->b_rcred = sp->b_wcred = p->p_ucred; 127 sp->b_vnbufs.le_next = NOLIST; 128 sp->b_actf = NULL; 129 } 130 131 void 132 swstrategy(bp) 133 register struct buf *bp; 134 { 135 int sz, off, seg, index; 136 register struct swdevt *sp; 137 struct vnode *vp; 138 139 #ifdef GENERIC 140 /* 141 * A mini-root gets copied into the front of the swap 142 * and we run over top of the swap area just long 143 * enough for us to do a mkfs and restor of the real 144 * root (sure beats rewriting standalone restor). 145 */ 146 #define MINIROOTSIZE 4096 147 if (rootdev == dumpdev) 148 bp->b_blkno += MINIROOTSIZE; 149 #endif 150 sz = howmany(bp->b_bcount, DEV_BSIZE); 151 if (bp->b_blkno + sz > nswap) { 152 bp->b_flags |= B_ERROR; 153 biodone(bp); 154 return; 155 } 156 if (nswdev > 1) { 157 #ifdef SEQSWAP 158 if (bp->b_blkno < niswap) { 159 if (niswdev > 1) { 160 off = bp->b_blkno % dmmax; 161 if (off+sz > dmmax) { 162 bp->b_flags |= B_ERROR; 163 biodone(bp); 164 return; 165 } 166 seg = bp->b_blkno / dmmax; 167 index = seg % niswdev; 168 seg /= niswdev; 169 bp->b_blkno = seg*dmmax + off; 170 } else 171 index = 0; 172 } else { 173 register struct swdevt *swp; 174 175 bp->b_blkno -= niswap; 176 for (index = niswdev, swp = &swdevt[niswdev]; 177 swp->sw_dev != NODEV; 178 swp++, index++) { 179 if (bp->b_blkno < swp->sw_nblks) 180 break; 181 bp->b_blkno -= swp->sw_nblks; 182 } 183 if (swp->sw_dev == NODEV || 184 bp->b_blkno+sz > swp->sw_nblks) { 185 bp->b_flags |= B_ERROR; 186 biodone(bp); 187 return; 188 } 189 } 190 #else 191 off = bp->b_blkno % dmmax; 192 if (off+sz > dmmax) { 193 bp->b_flags |= B_ERROR; 194 biodone(bp); 195 return; 196 } 197 seg = bp->b_blkno / dmmax; 198 index = seg % nswdev; 199 seg /= nswdev; 200 bp->b_blkno = seg*dmmax + off; 201 #endif 202 } else 203 index = 0; 204 sp = &swdevt[index]; 205 if ((bp->b_dev = sp->sw_dev) == NODEV) 206 panic("swstrategy"); 207 if (sp->sw_vp == NULL) { 208 bp->b_error |= B_ERROR; 209 biodone(bp); 210 return; 211 } 212 VHOLD(sp->sw_vp); 213 if ((bp->b_flags & B_READ) == 0) { 214 if (vp = bp->b_vp) { 215 vp->v_numoutput--; 216 if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) { 217 vp->v_flag &= ~VBWAIT; 218 wakeup((caddr_t)&vp->v_numoutput); 219 } 220 } 221 sp->sw_vp->v_numoutput++; 222 } 223 if (bp->b_vp != NULL) 224 brelvp(bp); 225 bp->b_vp = sp->sw_vp; 226 VOP_STRATEGY(bp); 227 } 228 229 /* 230 * System call swapon(name) enables swapping on device name, 231 * which must be in the swdevsw. Return EBUSY 232 * if already swapping on this device. 233 */ 234 struct swapon_args { 235 char *name; 236 }; 237 /* ARGSUSED */ 238 int 239 swapon(p, uap, retval) 240 struct proc *p; 241 struct swapon_args *uap; 242 int *retval; 243 { 244 register struct vnode *vp; 245 register struct swdevt *sp; 246 dev_t dev; 247 int error; 248 struct nameidata nd; 249 250 if (error = suser(p->p_ucred, &p->p_acflag)) 251 return (error); 252 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, uap->name, p); 253 if (error = namei(&nd)) 254 return (error); 255 vp = nd.ni_vp; 256 if (vp->v_type != VBLK) { 257 vrele(vp); 258 return (ENOTBLK); 259 } 260 dev = (dev_t)vp->v_rdev; 261 if (major(dev) >= nblkdev) { 262 vrele(vp); 263 return (ENXIO); 264 } 265 for (sp = &swdevt[0]; sp->sw_dev != NODEV; sp++) { 266 if (sp->sw_dev == dev) { 267 if (sp->sw_flags & SW_FREED) { 268 vrele(vp); 269 return (EBUSY); 270 } 271 sp->sw_vp = vp; 272 if (error = swfree(p, sp - swdevt)) { 273 vrele(vp); 274 return (error); 275 } 276 return (0); 277 } 278 #ifdef SEQSWAP 279 /* 280 * If we have reached a non-freed sequential device without 281 * finding what we are looking for, it is an error. 282 * That is because all interleaved devices must come first 283 * and sequential devices must be freed in order. 284 */ 285 if ((sp->sw_flags & (SW_SEQUENTIAL|SW_FREED)) == SW_SEQUENTIAL) 286 break; 287 #endif 288 } 289 vrele(vp); 290 return (EINVAL); 291 } 292 293 /* 294 * Swfree(index) frees the index'th portion of the swap map. 295 * Each of the nswdev devices provides 1/nswdev'th of the swap 296 * space, which is laid out with blocks of dmmax pages circularly 297 * among the devices. 298 */ 299 int 300 swfree(p, index) 301 struct proc *p; 302 int index; 303 { 304 register struct swdevt *sp; 305 register swblk_t vsbase; 306 register long blk; 307 struct vnode *vp; 308 register swblk_t dvbase; 309 register int nblks; 310 int error; 311 312 sp = &swdevt[index]; 313 vp = sp->sw_vp; 314 if (error = VOP_OPEN(vp, FREAD|FWRITE, p->p_ucred, p)) 315 return (error); 316 sp->sw_flags |= SW_FREED; 317 nblks = sp->sw_nblks; 318 /* 319 * Some devices may not exist til after boot time. 320 * If so, their nblk count will be 0. 321 */ 322 if (nblks <= 0) { 323 int perdev; 324 dev_t dev = sp->sw_dev; 325 326 if (bdevsw[major(dev)].d_psize == 0 || 327 (nblks = (*bdevsw[major(dev)].d_psize)(dev)) == -1) { 328 (void) VOP_CLOSE(vp, FREAD|FWRITE, p->p_ucred, p); 329 sp->sw_flags &= ~SW_FREED; 330 return (ENXIO); 331 } 332 #ifdef SEQSWAP 333 if (index < niswdev) { 334 perdev = niswap / niswdev; 335 if (nblks > perdev) 336 nblks = perdev; 337 } else { 338 if (nblks % dmmax) 339 nblks -= (nblks % dmmax); 340 nswap += nblks; 341 } 342 #else 343 perdev = nswap / nswdev; 344 if (nblks > perdev) 345 nblks = perdev; 346 #endif 347 sp->sw_nblks = nblks; 348 } 349 if (nblks == 0) { 350 (void) VOP_CLOSE(vp, FREAD|FWRITE, p->p_ucred, p); 351 sp->sw_flags &= ~SW_FREED; 352 return (0); /* XXX error? */ 353 } 354 #ifdef SEQSWAP 355 if (sp->sw_flags & SW_SEQUENTIAL) { 356 register struct swdevt *swp; 357 358 blk = niswap; 359 for (swp = &swdevt[niswdev]; swp != sp; swp++) 360 blk += swp->sw_nblks; 361 rmfree(swapmap, nblks, blk); 362 return (0); 363 } 364 #endif 365 for (dvbase = 0; dvbase < nblks; dvbase += dmmax) { 366 blk = nblks - dvbase; 367 #ifdef SEQSWAP 368 if ((vsbase = index*dmmax + dvbase*niswdev) >= niswap) 369 panic("swfree"); 370 #else 371 if ((vsbase = index*dmmax + dvbase*nswdev) >= nswap) 372 panic("swfree"); 373 #endif 374 if (blk > dmmax) 375 blk = dmmax; 376 if (vsbase == 0) { 377 /* 378 * First of all chunks... initialize the swapmap. 379 * Don't use the first cluster of the device 380 * in case it starts with a label or boot block. 381 */ 382 rminit(swapmap, blk - ctod(CLSIZE), 383 vsbase + ctod(CLSIZE), "swap", nswapmap); 384 } else if (dvbase == 0) { 385 /* 386 * Don't use the first cluster of the device 387 * in case it starts with a label or boot block. 388 */ 389 rmfree(swapmap, blk - ctod(CLSIZE), 390 vsbase + ctod(CLSIZE)); 391 } else 392 rmfree(swapmap, blk, vsbase); 393 } 394 return (0); 395 } 396