Commit | Line | Data |
---|---|---|
ca1f746a | 1 | /* kern_physio.c 4.36 82/11/02 */ |
663dbc72 BJ |
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" | |
973ecc4f | 13 | #include "../h/trace.h" |
d6d7360b | 14 | #include "../h/uio.h" |
663dbc72 | 15 | |
663dbc72 BJ |
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 | */ | |
4c05b581 BJ |
26 | struct buf *swbuf; |
27 | short *swsize; /* CAN WE JUST USE B_BCOUNT? */ | |
28 | int *swpf; | |
663dbc72 | 29 | |
663dbc72 BJ |
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; | |
39d536e6 | 46 | int nbytes, rdflg, flag; |
663dbc72 | 47 | dev_t dev; |
39d536e6 | 48 | u_int pfcent; |
663dbc72 BJ |
49 | { |
50 | register struct buf *bp; | |
e438ed8e | 51 | register u_int c; |
663dbc72 BJ |
52 | int p2dp; |
53 | register struct pte *dpte, *vpte; | |
530d0032 | 54 | int s; |
663dbc72 | 55 | |
530d0032 | 56 | s = spl6(); |
663dbc72 BJ |
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; | |
530d0032 | 63 | splx(s); |
663dbc72 BJ |
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) { | |
e438ed8e BJ |
85 | bp->b_bcount = nbytes; |
86 | minphys(bp); | |
87 | c = bp->b_bcount; | |
663dbc72 BJ |
88 | bp->b_blkno = dblkno; |
89 | bp->b_dev = dev; | |
d2f87136 BJ |
90 | if (flag & B_DIRTY) { |
91 | swpf[bp - swbuf] = pfcent; | |
92 | swsize[bp - swbuf] = nbytes; | |
93 | } | |
53f9ca20 BJ |
94 | #ifdef TRACE |
95 | trace(TR_SWAPIO, dev, bp->b_blkno); | |
96 | #endif | |
ca1f746a | 97 | physstrat(bp, bdevsw[major(dev)].d_strategy, PSWP); |
663dbc72 BJ |
98 | if (flag & B_DIRTY) { |
99 | if (c < nbytes) | |
100 | panic("big push"); | |
663dbc72 BJ |
101 | return; |
102 | } | |
663dbc72 BJ |
103 | bp->b_un.b_addr += c; |
104 | bp->b_flags &= ~B_DONE; | |
105 | if (bp->b_flags & B_ERROR) { | |
106 | if ((flag & (B_UAREA|B_PAGET)) || rdflg == B_WRITE) | |
107 | panic("hard IO err in swap"); | |
108 | swkill(p, (char *)0); | |
109 | } | |
110 | nbytes -= c; | |
e438ed8e | 111 | dblkno += c / DEV_BSIZE; |
663dbc72 | 112 | } |
530d0032 | 113 | s = spl6(); |
663dbc72 BJ |
114 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY); |
115 | bp->av_forw = bswlist.av_forw; | |
116 | bswlist.av_forw = bp; | |
117 | if (bswlist.b_flags & B_WANTED) { | |
118 | bswlist.b_flags &= ~B_WANTED; | |
119 | wakeup((caddr_t)&bswlist); | |
120 | wakeup((caddr_t)&proc[2]); | |
121 | } | |
530d0032 | 122 | splx(s); |
663dbc72 BJ |
123 | } |
124 | ||
125 | /* | |
126 | * If rout == 0 then killed on swap error, else | |
127 | * rout is the name of the routine where we ran out of | |
128 | * swap space. | |
129 | */ | |
130 | swkill(p, rout) | |
131 | struct proc *p; | |
132 | char *rout; | |
133 | { | |
444f631c | 134 | char *mesg; |
663dbc72 | 135 | |
444f631c | 136 | printf("pid %d: ", p->p_pid); |
663dbc72 | 137 | if (rout) |
444f631c | 138 | printf(mesg = "killed due to no swap space\n"); |
663dbc72 | 139 | else |
444f631c BJ |
140 | printf(mesg = "killed on swap error\n"); |
141 | uprintf("sorry, pid %d was %s", p->p_pid, mesg); | |
663dbc72 BJ |
142 | /* |
143 | * To be sure no looping (e.g. in vmsched trying to | |
144 | * swap out) mark process locked in core (as though | |
145 | * done by user) after killing it so noone will try | |
146 | * to swap it out. | |
147 | */ | |
a30d2e97 | 148 | psignal(p, SIGKILL); |
663dbc72 BJ |
149 | p->p_flag |= SULOCK; |
150 | } | |
151 | ||
663dbc72 BJ |
152 | /* |
153 | * Raw I/O. The arguments are | |
154 | * The strategy routine for the device | |
155 | * A buffer, which will always be a special buffer | |
156 | * header owned exclusively by the device for this purpose | |
157 | * The device number | |
158 | * Read/write flag | |
159 | * Essentially all the work is computing physical addresses and | |
160 | * validating them. | |
161 | * If the user has the proper access privilidges, the process is | |
162 | * marked 'delayed unlock' and the pages involved in the I/O are | |
163 | * faulted and locked. After the completion of the I/O, the above pages | |
164 | * are unlocked. | |
165 | */ | |
d6d7360b BJ |
166 | physio(strat, bp, dev, rw, mincnt, uio) |
167 | int (*strat)(); | |
168 | register struct buf *bp; | |
169 | dev_t dev; | |
170 | int rw; | |
171 | unsigned (*mincnt)(); | |
172 | struct uio *uio; | |
663dbc72 | 173 | { |
406ddcbe | 174 | register struct iovec *iov = uio->uio_iov; |
663dbc72 BJ |
175 | register int c; |
176 | char *a; | |
d6d7360b | 177 | int s, error = 0; |
663dbc72 | 178 | |
d6d7360b | 179 | nextiov: |
406ddcbe | 180 | if (uio->uio_iovcnt == 0) |
d6d7360b | 181 | return (0); |
406ddcbe | 182 | if (useracc(iov->iov_base,(u_int)iov->iov_len,rw==B_READ?B_WRITE:B_READ) == NULL) |
d6d7360b | 183 | return (EFAULT); |
530d0032 | 184 | s = spl6(); |
663dbc72 BJ |
185 | while (bp->b_flags&B_BUSY) { |
186 | bp->b_flags |= B_WANTED; | |
187 | sleep((caddr_t)bp, PRIBIO+1); | |
188 | } | |
ef3b3d5a | 189 | splx(s); |
663dbc72 BJ |
190 | bp->b_error = 0; |
191 | bp->b_proc = u.u_procp; | |
d6d7360b BJ |
192 | bp->b_un.b_addr = iov->iov_base; |
193 | while (iov->iov_len > 0) { | |
663dbc72 BJ |
194 | bp->b_flags = B_BUSY | B_PHYS | rw; |
195 | bp->b_dev = dev; | |
e438ed8e | 196 | bp->b_blkno = uio->uio_offset / DEV_BSIZE; |
d6d7360b | 197 | bp->b_bcount = iov->iov_len; |
663dbc72 BJ |
198 | (*mincnt)(bp); |
199 | c = bp->b_bcount; | |
200 | u.u_procp->p_flag |= SPHYSIO; | |
201 | vslock(a = bp->b_un.b_addr, c); | |
e438ed8e | 202 | physstrat(bp, strat, PRIBIO); |
81263dba | 203 | (void) spl6(); |
663dbc72 BJ |
204 | vsunlock(a, c, rw); |
205 | u.u_procp->p_flag &= ~SPHYSIO; | |
206 | if (bp->b_flags&B_WANTED) | |
207 | wakeup((caddr_t)bp); | |
530d0032 | 208 | splx(s); |
d6d7360b | 209 | c -= bp->b_resid; |
663dbc72 | 210 | bp->b_un.b_addr += c; |
d6d7360b BJ |
211 | iov->iov_len -= c; |
212 | uio->uio_resid -= c; | |
213 | uio->uio_offset += c; | |
52a593fa BJ |
214 | if (bp->b_flags&B_ERROR) |
215 | break; | |
663dbc72 BJ |
216 | } |
217 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS); | |
d6d7360b | 218 | error = geterror(bp); |
406ddcbe | 219 | if (error) |
d6d7360b | 220 | return (error); |
d6d7360b BJ |
221 | uio->uio_iov++; |
222 | uio->uio_iovcnt--; | |
223 | goto nextiov; | |
663dbc72 BJ |
224 | } |
225 | ||
663dbc72 BJ |
226 | unsigned |
227 | minphys(bp) | |
d6d7360b | 228 | struct buf *bp; |
663dbc72 BJ |
229 | { |
230 | ||
2ec65c94 BJ |
231 | if (bp->b_bcount > 63 * 1024) |
232 | bp->b_bcount = 63 * 1024; | |
663dbc72 BJ |
233 | } |
234 |