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 7.7 (Berkeley) 06/27/89 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 while (nbytes > 0) { 86 bp->b_blkno = dblkno; 87 bp->b_dev = vp->v_rdev; 88 if (bp->b_vp) 89 brelvp(bp); 90 VREF(vp); 91 bp->b_vp = vp; 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 freeswbuf(bp); 123 return (error); 124 } 125 126 /* 127 * Put a buffer on the clean list after I/O is done. 128 * Called from biodone. 129 */ 130 swdone(bp) 131 register struct buf *bp; 132 { 133 register int s; 134 135 if (bp->b_flags & B_ERROR) 136 panic("IO err in push"); 137 s = splbio(); 138 bp->av_forw = bclnlist; 139 cnt.v_pgout++; 140 cnt.v_pgpgout += bp->b_bcount / NBPG; 141 bclnlist = bp; 142 if (bswlist.b_flags & B_WANTED) 143 wakeup((caddr_t)&proc[2]); 144 splx(s); 145 } 146 147 /* 148 * If rout == 0 then killed on swap error, else 149 * rout is the name of the routine where we ran out of 150 * swap space. 151 */ 152 swkill(p, rout) 153 struct proc *p; 154 char *rout; 155 { 156 157 printf("pid %d: %s\n", p->p_pid, rout); 158 uprintf("sorry, pid %d was killed in %s\n", p->p_pid, rout); 159 /* 160 * To be sure no looping (e.g. in vmsched trying to 161 * swap out) mark process locked in core (as though 162 * done by user) after killing it so noone will try 163 * to swap it out. 164 */ 165 psignal(p, SIGKILL); 166 p->p_flag |= SULOCK; 167 } 168 169 /* 170 * Raw I/O. The arguments are 171 * The strategy routine for the device 172 * A buffer, which will either be a special buffer header owned 173 * exclusively by the device for this purpose, or NULL, 174 * indicating that we should use a swap buffer 175 * The device number 176 * Read/write flag 177 * Essentially all the work is computing physical addresses and 178 * validating them. 179 * If the user has the proper access privilidges, the process is 180 * marked 'delayed unlock' and the pages involved in the I/O are 181 * faulted and locked. After the completion of the I/O, the above pages 182 * are unlocked. 183 */ 184 physio(strat, bp, dev, rw, mincnt, uio) 185 int (*strat)(); 186 register struct buf *bp; 187 dev_t dev; 188 int rw; 189 u_int (*mincnt)(); 190 struct uio *uio; 191 { 192 register struct iovec *iov; 193 register int c; 194 char *a; 195 int s, allocbuf = 0, error = 0; 196 struct buf *getswbuf(); 197 198 if (bp == NULL) { 199 allocbuf = 1; 200 bp = getswbuf(PRIBIO+1); 201 } 202 for (; uio->uio_iovcnt; uio->uio_iov++, uio->uio_iovcnt--) { 203 iov = uio->uio_iov; 204 if (!useracc(iov->iov_base, (u_int)iov->iov_len, 205 rw == B_READ ? B_WRITE : B_READ)) { 206 error = EFAULT; 207 break; 208 } 209 if (!allocbuf) { /* only if sharing caller's buffer */ 210 s = splbio(); 211 while (bp->b_flags&B_BUSY) { 212 bp->b_flags |= B_WANTED; 213 sleep((caddr_t)bp, PRIBIO+1); 214 } 215 splx(s); 216 } 217 bp->b_error = 0; 218 bp->b_proc = u.u_procp; 219 bp->b_un.b_addr = iov->iov_base; 220 while (iov->iov_len > 0) { 221 bp->b_flags = B_BUSY | B_PHYS | B_RAW | rw; 222 bp->b_dev = dev; 223 bp->b_blkno = btodb(uio->uio_offset); 224 bp->b_bcount = iov->iov_len; 225 (*mincnt)(bp); 226 c = bp->b_bcount; 227 u.u_procp->p_flag |= SPHYSIO; 228 vslock(a = bp->b_un.b_addr, c); 229 (*strat)(bp); 230 s = splbio(); 231 while ((bp->b_flags & B_DONE) == 0) 232 sleep((caddr_t)bp, PRIBIO); 233 vsunlock(a, c, rw); 234 u.u_procp->p_flag &= ~SPHYSIO; 235 if (bp->b_flags&B_WANTED) /* rare */ 236 wakeup((caddr_t)bp); 237 splx(s); 238 c -= bp->b_resid; 239 bp->b_un.b_addr += c; 240 iov->iov_len -= c; 241 uio->uio_resid -= c; 242 uio->uio_offset += c; 243 /* temp kludge for tape drives */ 244 if (bp->b_resid || (bp->b_flags&B_ERROR)) 245 break; 246 } 247 bp->b_flags &= ~(B_BUSY | B_WANTED | B_PHYS | B_RAW); 248 error = biowait(bp); 249 /* temp kludge for tape drives */ 250 if (bp->b_resid || error) 251 break; 252 } 253 if (allocbuf) 254 freeswbuf(bp); 255 return (error); 256 } 257 258 u_int 259 minphys(bp) 260 struct buf *bp; 261 { 262 if (bp->b_bcount > MAXPHYS) 263 bp->b_bcount = MAXPHYS; 264 } 265 266 static 267 struct buf * 268 getswbuf(prio) 269 int prio; 270 { 271 int s; 272 struct buf *bp; 273 274 s = splbio(); 275 while (bswlist.av_forw == NULL) { 276 bswlist.b_flags |= B_WANTED; 277 sleep((caddr_t)&bswlist, prio); 278 } 279 bp = bswlist.av_forw; 280 bswlist.av_forw = bp->av_forw; 281 splx(s); 282 return (bp); 283 } 284 285 static 286 freeswbuf(bp) 287 struct buf *bp; 288 { 289 int s; 290 291 s = splbio(); 292 bp->av_forw = bswlist.av_forw; 293 bswlist.av_forw = bp; 294 if (bswlist.b_flags & B_WANTED) { 295 bswlist.b_flags &= ~B_WANTED; 296 wakeup((caddr_t)&bswlist); 297 wakeup((caddr_t)&proc[2]); 298 } 299 splx(s); 300 } 301 302 rawread(dev, uio) 303 dev_t dev; 304 struct uio *uio; 305 { 306 return (physio(cdevsw[major(dev)].d_strategy, (struct buf *)NULL, 307 dev, B_READ, minphys, uio)); 308 } 309 310 rawwrite(dev, uio) 311 dev_t dev; 312 struct uio *uio; 313 { 314 return (physio(cdevsw[major(dev)].d_strategy, (struct buf *)NULL, 315 dev, B_WRITE, minphys, uio)); 316 } 317