1 /* 2 * Copyright (c) 1982, 1986 Regents of the University of California. 3 * All rights reserved. The Berkeley software License Agreement 4 * specifies the terms and conditions for redistribution. 5 * 6 * @(#)kern_physio.c 7.9 (Berkeley) 08/30/89 7 */ 8 9 #include "param.h" 10 #include "systm.h" 11 #include "user.h" 12 #include "buf.h" 13 #include "conf.h" 14 #include "proc.h" 15 #include "seg.h" 16 #include "vm.h" 17 #include "trace.h" 18 #include "map.h" 19 #include "vnode.h" 20 21 #include "machine/pte.h" 22 23 /* 24 * Swap IO headers - 25 * They contain the necessary information for the swap I/O. 26 * At any given time, a swap header can be in three 27 * different lists. When free it is in the free list, 28 * when allocated and the I/O queued, it is on the swap 29 * device list, and finally, if the operation was a dirty 30 * page push, when the I/O completes, it is inserted 31 * in a list of cleaned pages to be processed by the pageout daemon. 32 */ 33 struct buf *swbuf; 34 35 /* 36 * swap I/O - 37 * 38 * If the flag indicates a dirty page push initiated 39 * by the pageout daemon, we map the page into the i th 40 * virtual page of process 2 (the daemon itself) where i is 41 * the index of the swap header that has been allocated. 42 * We simply initialize the header and queue the I/O but 43 * do not wait for completion. When the I/O completes, 44 * biodone() will link the header to a list of cleaned 45 * pages to be processed by the pageout daemon. 46 */ 47 swap(p, dblkno, addr, nbytes, rdflg, flag, vp, pfcent) 48 struct proc *p; 49 swblk_t dblkno; 50 caddr_t addr; 51 int nbytes, rdflg, flag; 52 struct vnode *vp; 53 u_int pfcent; 54 { 55 register struct buf *bp; 56 register struct pte *dpte, *vpte; 57 register u_int c; 58 int p2dp, s, error = 0; 59 struct buf *getswbuf(); 60 int swdone(); 61 62 bp = getswbuf(PSWP+1); 63 bp->b_flags = B_BUSY | B_PHYS | rdflg | flag; 64 if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0) 65 if (rdflg == B_READ) 66 sum.v_pswpin += btoc(nbytes); 67 else 68 sum.v_pswpout += btoc(nbytes); 69 bp->b_proc = p; 70 if (flag & B_DIRTY) { 71 p2dp = ((bp - swbuf) * CLSIZE) * KLMAX; 72 dpte = dptopte(&proc[2], p2dp); 73 vpte = vtopte(p, btop(addr)); 74 for (c = 0; c < nbytes; c += NBPG) { 75 if (vpte->pg_pfnum == 0 || vpte->pg_fod) 76 panic("swap bad pte"); 77 *dpte++ = *vpte++; 78 } 79 bp->b_un.b_addr = (caddr_t)ctob(dptov(&proc[2], p2dp)); 80 bp->b_flags |= B_CALL; 81 bp->b_iodone = swdone; 82 bp->b_pfcent = pfcent; 83 } else 84 bp->b_un.b_addr = addr; 85 if (bp->b_vp) 86 panic("swap: active vp"); 87 VREF(vp); 88 bp->b_vp = vp; 89 bp->b_dev = vp->v_rdev; 90 while (nbytes > 0) { 91 bp->b_blkno = dblkno; 92 bp->b_bcount = nbytes; 93 minphys(bp); 94 c = bp->b_bcount; 95 #ifdef TRACE 96 trace(TR_SWAPIO, vp, bp->b_blkno); 97 #endif 98 VOP_STRATEGY(bp); 99 /* pageout daemon doesn't wait for pushed pages */ 100 if (flag & B_DIRTY) { 101 if (c < nbytes) 102 panic("big push"); 103 return (0); 104 } else { 105 s = splbio(); 106 while ((bp->b_flags & B_DONE) == 0) 107 sleep((caddr_t)bp, PSWP); 108 splx(s); 109 } 110 bp->b_un.b_addr += c; 111 bp->b_flags &= ~B_DONE; 112 if (bp->b_flags & B_ERROR) { 113 if ((flag & (B_UAREA|B_PAGET)) || rdflg == B_WRITE) 114 panic("hard IO err in swap"); 115 swkill(p, "swap: read error from swap device"); 116 error = EIO; 117 } 118 nbytes -= c; 119 dblkno += btodb(c); 120 } 121 bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY); 122 brelvp(bp); 123 freeswbuf(bp); 124 return (error); 125 } 126 127 /* 128 * Put a buffer on the clean list after I/O is done. 129 * Called from biodone. 130 */ 131 swdone(bp) 132 register struct buf *bp; 133 { 134 register int s; 135 136 if (bp->b_flags & B_ERROR) 137 panic("IO err in push"); 138 s = splbio(); 139 bp->av_forw = bclnlist; 140 cnt.v_pgout++; 141 cnt.v_pgpgout += bp->b_bcount / NBPG; 142 bclnlist = bp; 143 if (bswlist.b_flags & B_WANTED) 144 wakeup((caddr_t)&proc[2]); 145 splx(s); 146 } 147 148 /* 149 * If rout == 0 then killed on swap error, else 150 * rout is the name of the routine where we ran out of 151 * swap space. 152 */ 153 swkill(p, rout) 154 struct proc *p; 155 char *rout; 156 { 157 158 printf("pid %d: %s\n", p->p_pid, rout); 159 uprintf("sorry, pid %d was killed in %s\n", p->p_pid, rout); 160 /* 161 * To be sure no looping (e.g. in vmsched trying to 162 * swap out) mark process locked in core (as though 163 * done by user) after killing it so noone will try 164 * to swap it out. 165 */ 166 psignal(p, SIGKILL); 167 p->p_flag |= SULOCK; 168 } 169 170 /* 171 * Raw I/O. The arguments are 172 * The strategy routine for the device 173 * A buffer, which will either be a special buffer header owned 174 * exclusively by the device for this purpose, or NULL, 175 * indicating that we should use a swap buffer 176 * The device number 177 * Read/write flag 178 * Essentially all the work is computing physical addresses and 179 * validating them. 180 * If the user has the proper access privilidges, the process is 181 * marked 'delayed unlock' and the pages involved in the I/O are 182 * faulted and locked. After the completion of the I/O, the above pages 183 * are unlocked. 184 */ 185 physio(strat, bp, dev, rw, mincnt, uio) 186 int (*strat)(); 187 register struct buf *bp; 188 dev_t dev; 189 int rw; 190 u_int (*mincnt)(); 191 struct uio *uio; 192 { 193 register struct iovec *iov; 194 register int requested, done; 195 char *a; 196 int s, allocbuf = 0, error = 0; 197 struct buf *getswbuf(); 198 199 if (bp == NULL) { 200 allocbuf = 1; 201 bp = getswbuf(PRIBIO+1); 202 } 203 for (; uio->uio_iovcnt; uio->uio_iov++, uio->uio_iovcnt--) { 204 iov = uio->uio_iov; 205 if (!useracc(iov->iov_base, (u_int)iov->iov_len, 206 rw == B_READ ? B_WRITE : B_READ)) { 207 error = EFAULT; 208 break; 209 } 210 if (!allocbuf) { /* only if sharing caller's buffer */ 211 s = splbio(); 212 while (bp->b_flags&B_BUSY) { 213 bp->b_flags |= B_WANTED; 214 sleep((caddr_t)bp, PRIBIO+1); 215 } 216 splx(s); 217 } 218 bp->b_error = 0; 219 bp->b_proc = u.u_procp; 220 bp->b_un.b_addr = iov->iov_base; 221 while (iov->iov_len > 0) { 222 bp->b_flags = B_BUSY | B_PHYS | B_RAW | rw; 223 bp->b_dev = dev; 224 bp->b_blkno = btodb(uio->uio_offset); 225 bp->b_bcount = iov->iov_len; 226 (*mincnt)(bp); 227 requested = bp->b_bcount; 228 u.u_procp->p_flag |= SPHYSIO; 229 vslock(a = bp->b_un.b_addr, requested); 230 (*strat)(bp); 231 s = splbio(); 232 while ((bp->b_flags & B_DONE) == 0) 233 sleep((caddr_t)bp, PRIBIO); 234 vsunlock(a, requested, rw); 235 u.u_procp->p_flag &= ~SPHYSIO; 236 if (bp->b_flags&B_WANTED) /* rare */ 237 wakeup((caddr_t)bp); 238 splx(s); 239 done = bp->b_bcount - bp->b_resid; 240 bp->b_un.b_addr += done; 241 iov->iov_len -= done; 242 uio->uio_resid -= done; 243 uio->uio_offset += done; 244 /* temp kludge for disk drives */ 245 if (done < requested || bp->b_flags & B_ERROR) 246 break; 247 } 248 bp->b_flags &= ~(B_BUSY | B_WANTED | B_PHYS | B_RAW); 249 error = biowait(bp); 250 /* temp kludge for disk drives */ 251 if (done < requested || bp->b_flags & B_ERROR) 252 break; 253 } 254 if (allocbuf) 255 freeswbuf(bp); 256 return (error); 257 } 258 259 u_int 260 minphys(bp) 261 struct buf *bp; 262 { 263 if (bp->b_bcount > MAXPHYS) 264 bp->b_bcount = MAXPHYS; 265 } 266 267 static 268 struct buf * 269 getswbuf(prio) 270 int prio; 271 { 272 int s; 273 struct buf *bp; 274 275 s = splbio(); 276 while (bswlist.av_forw == NULL) { 277 bswlist.b_flags |= B_WANTED; 278 sleep((caddr_t)&bswlist, prio); 279 } 280 bp = bswlist.av_forw; 281 bswlist.av_forw = bp->av_forw; 282 splx(s); 283 return (bp); 284 } 285 286 static 287 freeswbuf(bp) 288 struct buf *bp; 289 { 290 int s; 291 292 s = splbio(); 293 bp->av_forw = bswlist.av_forw; 294 bswlist.av_forw = bp; 295 if (bswlist.b_flags & B_WANTED) { 296 bswlist.b_flags &= ~B_WANTED; 297 wakeup((caddr_t)&bswlist); 298 wakeup((caddr_t)&proc[2]); 299 } 300 splx(s); 301 } 302 303 rawread(dev, uio) 304 dev_t dev; 305 struct uio *uio; 306 { 307 return (physio(cdevsw[major(dev)].d_strategy, (struct buf *)NULL, 308 dev, B_READ, minphys, uio)); 309 } 310 311 rawwrite(dev, uio) 312 dev_t dev; 313 struct uio *uio; 314 { 315 return (physio(cdevsw[major(dev)].d_strategy, (struct buf *)NULL, 316 dev, B_WRITE, minphys, uio)); 317 } 318