| 1 | /* kern_physio.c 4.31 82/08/13 */ |
| 2 | |
| 3 | #include "../h/param.h" |
| 4 | #include "../h/systm.h" |
| 5 | #include "../h/dir.h" |
| 6 | #include "../h/user.h" |
| 7 | #include "../h/buf.h" |
| 8 | #include "../h/conf.h" |
| 9 | #include "../h/proc.h" |
| 10 | #include "../h/seg.h" |
| 11 | #include "../h/pte.h" |
| 12 | #include "../h/vm.h" |
| 13 | #include "../h/trace.h" |
| 14 | #include "../h/uio.h" |
| 15 | |
| 16 | /* |
| 17 | * Swap IO headers - |
| 18 | * They contain the necessary information for the swap I/O. |
| 19 | * At any given time, a swap header can be in three |
| 20 | * different lists. When free it is in the free list, |
| 21 | * when allocated and the I/O queued, it is on the swap |
| 22 | * device list, and finally, if the operation was a dirty |
| 23 | * page push, when the I/O completes, it is inserted |
| 24 | * in a list of cleaned pages to be processed by the pageout daemon. |
| 25 | */ |
| 26 | struct buf *swbuf; |
| 27 | short *swsize; /* CAN WE JUST USE B_BCOUNT? */ |
| 28 | int *swpf; |
| 29 | |
| 30 | /* |
| 31 | * swap I/O - |
| 32 | * |
| 33 | * If the flag indicates a dirty page push initiated |
| 34 | * by the pageout daemon, we map the page into the i th |
| 35 | * virtual page of process 2 (the daemon itself) where i is |
| 36 | * the index of the swap header that has been allocated. |
| 37 | * We simply initialize the header and queue the I/O but |
| 38 | * do not wait for completion. When the I/O completes, |
| 39 | * iodone() will link the header to a list of cleaned |
| 40 | * pages to be processed by the pageout daemon. |
| 41 | */ |
| 42 | swap(p, dblkno, addr, nbytes, rdflg, flag, dev, pfcent) |
| 43 | struct proc *p; |
| 44 | swblk_t dblkno; |
| 45 | caddr_t addr; |
| 46 | int flag, nbytes; |
| 47 | dev_t dev; |
| 48 | unsigned pfcent; |
| 49 | { |
| 50 | register struct buf *bp; |
| 51 | register int c; |
| 52 | int p2dp; |
| 53 | register struct pte *dpte, *vpte; |
| 54 | int s; |
| 55 | |
| 56 | s = spl6(); |
| 57 | while (bswlist.av_forw == NULL) { |
| 58 | bswlist.b_flags |= B_WANTED; |
| 59 | sleep((caddr_t)&bswlist, PSWP+1); |
| 60 | } |
| 61 | bp = bswlist.av_forw; |
| 62 | bswlist.av_forw = bp->av_forw; |
| 63 | splx(s); |
| 64 | |
| 65 | bp->b_flags = B_BUSY | B_PHYS | rdflg | flag; |
| 66 | if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0) |
| 67 | if (rdflg == B_READ) |
| 68 | sum.v_pswpin += btoc(nbytes); |
| 69 | else |
| 70 | sum.v_pswpout += btoc(nbytes); |
| 71 | bp->b_proc = p; |
| 72 | if (flag & B_DIRTY) { |
| 73 | p2dp = ((bp - swbuf) * CLSIZE) * KLMAX; |
| 74 | dpte = dptopte(&proc[2], p2dp); |
| 75 | vpte = vtopte(p, btop(addr)); |
| 76 | for (c = 0; c < nbytes; c += NBPG) { |
| 77 | if (vpte->pg_pfnum == 0 || vpte->pg_fod) |
| 78 | panic("swap bad pte"); |
| 79 | *dpte++ = *vpte++; |
| 80 | } |
| 81 | bp->b_un.b_addr = (caddr_t)ctob(p2dp); |
| 82 | } else |
| 83 | bp->b_un.b_addr = addr; |
| 84 | while (nbytes > 0) { |
| 85 | c = imin(ctob(120), nbytes); |
| 86 | bp->b_bcount = c; |
| 87 | bp->b_blkno = dblkno; |
| 88 | bp->b_dev = dev; |
| 89 | if (flag & B_DIRTY) { |
| 90 | swpf[bp - swbuf] = pfcent; |
| 91 | swsize[bp - swbuf] = nbytes; |
| 92 | } |
| 93 | #ifdef TRACE |
| 94 | trace(TR_SWAPIO, dev, bp->b_blkno); |
| 95 | #endif |
| 96 | (*bdevsw[major(dev)].d_strategy)(bp); |
| 97 | if (flag & B_DIRTY) { |
| 98 | if (c < nbytes) |
| 99 | panic("big push"); |
| 100 | return; |
| 101 | } |
| 102 | s = spl6(); |
| 103 | while((bp->b_flags&B_DONE)==0) |
| 104 | sleep((caddr_t)bp, PSWP); |
| 105 | splx(s); |
| 106 | bp->b_un.b_addr += c; |
| 107 | bp->b_flags &= ~B_DONE; |
| 108 | if (bp->b_flags & B_ERROR) { |
| 109 | if ((flag & (B_UAREA|B_PAGET)) || rdflg == B_WRITE) |
| 110 | panic("hard IO err in swap"); |
| 111 | swkill(p, (char *)0); |
| 112 | } |
| 113 | nbytes -= c; |
| 114 | dblkno += btoc(c); |
| 115 | } |
| 116 | s = spl6(); |
| 117 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY); |
| 118 | bp->av_forw = bswlist.av_forw; |
| 119 | bswlist.av_forw = bp; |
| 120 | if (bswlist.b_flags & B_WANTED) { |
| 121 | bswlist.b_flags &= ~B_WANTED; |
| 122 | wakeup((caddr_t)&bswlist); |
| 123 | wakeup((caddr_t)&proc[2]); |
| 124 | } |
| 125 | splx(s); |
| 126 | } |
| 127 | |
| 128 | /* |
| 129 | * If rout == 0 then killed on swap error, else |
| 130 | * rout is the name of the routine where we ran out of |
| 131 | * swap space. |
| 132 | */ |
| 133 | swkill(p, rout) |
| 134 | struct proc *p; |
| 135 | char *rout; |
| 136 | { |
| 137 | char *mesg; |
| 138 | |
| 139 | printf("pid %d: ", p->p_pid); |
| 140 | if (rout) |
| 141 | printf(mesg = "killed due to no swap space\n"); |
| 142 | else |
| 143 | printf(mesg = "killed on swap error\n"); |
| 144 | uprintf("sorry, pid %d was %s", p->p_pid, mesg); |
| 145 | /* |
| 146 | * To be sure no looping (e.g. in vmsched trying to |
| 147 | * swap out) mark process locked in core (as though |
| 148 | * done by user) after killing it so noone will try |
| 149 | * to swap it out. |
| 150 | */ |
| 151 | psignal(p, SIGKILL); |
| 152 | p->p_flag |= SULOCK; |
| 153 | } |
| 154 | |
| 155 | /* |
| 156 | * Raw I/O. The arguments are |
| 157 | * The strategy routine for the device |
| 158 | * A buffer, which will always be a special buffer |
| 159 | * header owned exclusively by the device for this purpose |
| 160 | * The device number |
| 161 | * Read/write flag |
| 162 | * Essentially all the work is computing physical addresses and |
| 163 | * validating them. |
| 164 | * If the user has the proper access privilidges, the process is |
| 165 | * marked 'delayed unlock' and the pages involved in the I/O are |
| 166 | * faulted and locked. After the completion of the I/O, the above pages |
| 167 | * are unlocked. |
| 168 | */ |
| 169 | physio(strat, bp, dev, rw, mincnt, uio) |
| 170 | int (*strat)(); |
| 171 | register struct buf *bp; |
| 172 | dev_t dev; |
| 173 | int rw; |
| 174 | unsigned (*mincnt)(); |
| 175 | struct uio *uio; |
| 176 | { |
| 177 | register int c; |
| 178 | struct uio auio; |
| 179 | register struct iovec *iov; |
| 180 | struct iovec aiov; |
| 181 | char *a; |
| 182 | int s, error = 0; |
| 183 | |
| 184 | if (uio == 0) { |
| 185 | uio = &auio; |
| 186 | uio->uio_iov = &aiov; |
| 187 | uio->uio_iovcnt = 1; |
| 188 | uio->uio_offset = u.u_offset; |
| 189 | uio->uio_segflg = u.u_segflg; |
| 190 | iov = &aiov; |
| 191 | iov->iov_base = u.u_base; |
| 192 | iov->iov_len = u.u_count; |
| 193 | uio->uio_resid = iov->iov_len; |
| 194 | } else |
| 195 | iov = uio->uio_iov; |
| 196 | nextiov: |
| 197 | if (uio->uio_iovcnt == 0) { |
| 198 | u.u_count = uio->uio_resid; |
| 199 | return (0); |
| 200 | } |
| 201 | if (useracc(iov->iov_base,iov->iov_len,rw==B_READ?B_WRITE:B_READ) == NULL) { |
| 202 | u.u_count = uio->uio_resid; |
| 203 | u.u_error = EFAULT; |
| 204 | return (EFAULT); |
| 205 | } |
| 206 | s = spl6(); |
| 207 | while (bp->b_flags&B_BUSY) { |
| 208 | bp->b_flags |= B_WANTED; |
| 209 | sleep((caddr_t)bp, PRIBIO+1); |
| 210 | } |
| 211 | splx(s); |
| 212 | bp->b_error = 0; |
| 213 | bp->b_proc = u.u_procp; |
| 214 | bp->b_un.b_addr = iov->iov_base; |
| 215 | while (iov->iov_len > 0) { |
| 216 | bp->b_flags = B_BUSY | B_PHYS | rw; |
| 217 | bp->b_dev = dev; |
| 218 | bp->b_blkno = uio->uio_offset >> PGSHIFT; |
| 219 | bp->b_bcount = iov->iov_len; |
| 220 | (*mincnt)(bp); |
| 221 | c = bp->b_bcount; |
| 222 | u.u_procp->p_flag |= SPHYSIO; |
| 223 | vslock(a = bp->b_un.b_addr, c); |
| 224 | (*strat)(bp); |
| 225 | (void) spl6(); |
| 226 | while ((bp->b_flags&B_DONE) == 0) |
| 227 | sleep((caddr_t)bp, PRIBIO); |
| 228 | vsunlock(a, c, rw); |
| 229 | u.u_procp->p_flag &= ~SPHYSIO; |
| 230 | if (bp->b_flags&B_WANTED) |
| 231 | wakeup((caddr_t)bp); |
| 232 | splx(s); |
| 233 | c -= bp->b_resid; |
| 234 | bp->b_un.b_addr += c; |
| 235 | iov->iov_len -= c; |
| 236 | uio->uio_resid -= c; |
| 237 | uio->uio_offset += c; |
| 238 | if (bp->b_flags&B_ERROR) |
| 239 | break; |
| 240 | } |
| 241 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS); |
| 242 | error = geterror(bp); |
| 243 | if (error) { |
| 244 | u.u_error = error; |
| 245 | u.u_count = uio->uio_resid; |
| 246 | return (error); |
| 247 | } |
| 248 | uio->uio_iov++; |
| 249 | uio->uio_iovcnt--; |
| 250 | goto nextiov; |
| 251 | } |
| 252 | |
| 253 | /*ARGSUSED*/ |
| 254 | unsigned |
| 255 | minphys(bp) |
| 256 | struct buf *bp; |
| 257 | { |
| 258 | |
| 259 | if (bp->b_bcount > 63 * 1024) |
| 260 | bp->b_bcount = 63 * 1024; |
| 261 | } |
| 262 | |