xref: /csrg-svn/sys/kern/kern_physio.c (revision 30750)
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	7.2.1.1 (Berkeley) 04/02/87
7  */
8 
9 #include "../machine/pte.h"
10 
11 #include "param.h"
12 #include "systm.h"
13 #include "dir.h"
14 #include "user.h"
15 #include "buf.h"
16 #include "conf.h"
17 #include "proc.h"
18 #include "seg.h"
19 #include "vm.h"
20 #include "trace.h"
21 #include "map.h"
22 #include "uio.h"
23 #ifdef SECSIZE
24 #include "file.h"
25 #include "ioctl.h"
26 #include "disklabel.h"
27 #endif SECSIZE
28 
29 /*
30  * Swap IO headers -
31  * They contain the necessary information for the swap I/O.
32  * At any given time, a swap header can be in three
33  * different lists. When free it is in the free list,
34  * when allocated and the I/O queued, it is on the swap
35  * device list, and finally, if the operation was a dirty
36  * page push, when the I/O completes, it is inserted
37  * in a list of cleaned pages to be processed by the pageout daemon.
38  */
39 struct	buf *swbuf;
40 
41 /*
42  * swap I/O -
43  *
44  * If the flag indicates a dirty page push initiated
45  * by the pageout daemon, we map the page into the i th
46  * virtual page of process 2 (the daemon itself) where i is
47  * the index of the swap header that has been allocated.
48  * We simply initialize the header and queue the I/O but
49  * do not wait for completion. When the I/O completes,
50  * biodone() will link the header to a list of cleaned
51  * pages to be processed by the pageout daemon.
52  */
53 swap(p, dblkno, addr, nbytes, rdflg, flag, dev, pfcent)
54 	struct proc *p;
55 	swblk_t dblkno;
56 	caddr_t addr;
57 	int nbytes, rdflg, flag;
58 	dev_t dev;
59 	u_int pfcent;
60 {
61 	register struct buf *bp;
62 	register u_int c;
63 	int p2dp;
64 	register struct pte *dpte, *vpte;
65 	int s;
66 	extern swdone();
67 	int error = 0;
68 
69 	s = splbio();
70 	while (bswlist.av_forw == NULL) {
71 		bswlist.b_flags |= B_WANTED;
72 		sleep((caddr_t)&bswlist, PSWP+1);
73 	}
74 	bp = bswlist.av_forw;
75 	bswlist.av_forw = bp->av_forw;
76 	splx(s);
77 
78 	bp->b_flags = B_BUSY | B_PHYS | rdflg | flag;
79 #ifdef SECSIZE
80 	bp->b_blksize = DEV_BSIZE;
81 #endif SECSIZE
82 	if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0)
83 		if (rdflg == B_READ)
84 			sum.v_pswpin += btoc(nbytes);
85 		else
86 			sum.v_pswpout += btoc(nbytes);
87 	bp->b_proc = p;
88 	if (flag & B_DIRTY) {
89 		p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
90 		dpte = dptopte(&proc[2], p2dp);
91 		vpte = vtopte(p, btop(addr));
92 		for (c = 0; c < nbytes; c += NBPG) {
93 			if (vpte->pg_pfnum == 0 || vpte->pg_fod)
94 				panic("swap bad pte");
95 			*dpte++ = *vpte++;
96 		}
97 		bp->b_un.b_addr = (caddr_t)ctob(dptov(&proc[2], p2dp));
98 		bp->b_flags |= B_CALL;
99 		bp->b_iodone = swdone;
100 		bp->b_pfcent = pfcent;
101 	} else
102 		bp->b_un.b_addr = addr;
103 	while (nbytes > 0) {
104 		bp->b_bcount = nbytes;
105 		minphys(bp);
106 		c = bp->b_bcount;
107 		bp->b_blkno = dblkno;
108 		bp->b_dev = dev;
109 #ifdef TRACE
110 		trace(TR_SWAPIO, dev, bp->b_blkno);
111 #endif
112 		physstrat(bp, bdevsw[major(dev)].d_strategy, PSWP);
113 		if (flag & B_DIRTY) {
114 			if (c < nbytes)
115 				panic("big push");
116 			return (0);
117 		}
118 		bp->b_un.b_addr += c;
119 		bp->b_flags &= ~B_DONE;
120 		if (bp->b_flags & B_ERROR) {
121 			if ((flag & (B_UAREA|B_PAGET)) || rdflg == B_WRITE)
122 				panic("hard IO err in swap");
123 			swkill(p, "swap: read error from swap device");
124 			error = EIO;
125 		}
126 		nbytes -= c;
127 #ifdef SECSIZE
128 		if (flag & B_PGIN && nbytes > 0)
129 			panic("big pgin");
130 #endif SECSIZE
131 		dblkno += btodb(c);
132 	}
133 	s = splbio();
134 	bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY);
135 	bp->av_forw = bswlist.av_forw;
136 	bswlist.av_forw = bp;
137 	if (bswlist.b_flags & B_WANTED) {
138 		bswlist.b_flags &= ~B_WANTED;
139 		wakeup((caddr_t)&bswlist);
140 		wakeup((caddr_t)&proc[2]);
141 	}
142 	splx(s);
143 	return (error);
144 }
145 
146 /*
147  * Put a buffer on the clean list after I/O is done.
148  * Called from biodone.
149  */
150 swdone(bp)
151 	register struct buf *bp;
152 {
153 	register int s;
154 
155 	if (bp->b_flags & B_ERROR)
156 		panic("IO err in push");
157 	s = splbio();
158 	bp->av_forw = bclnlist;
159 	cnt.v_pgout++;
160 	cnt.v_pgpgout += bp->b_bcount / NBPG;
161 	bclnlist = bp;
162 	if (bswlist.b_flags & B_WANTED)
163 		wakeup((caddr_t)&proc[2]);
164 	splx(s);
165 }
166 
167 /*
168  * If rout == 0 then killed on swap error, else
169  * rout is the name of the routine where we ran out of
170  * swap space.
171  */
172 swkill(p, rout)
173 	struct proc *p;
174 	char *rout;
175 {
176 
177 	printf("pid %d: %s\n", p->p_pid, rout);
178 	uprintf("sorry, pid %d was killed in %s\n", p->p_pid, rout);
179 	/*
180 	 * To be sure no looping (e.g. in vmsched trying to
181 	 * swap out) mark process locked in core (as though
182 	 * done by user) after killing it so noone will try
183 	 * to swap it out.
184 	 */
185 	psignal(p, SIGKILL);
186 	p->p_flag |= SULOCK;
187 }
188 
189 /*
190  * Raw I/O. The arguments are
191  *	The strategy routine for the device
192  *	A buffer, which will always be a special buffer
193  *	  header owned exclusively by the device for this purpose
194  *	The device number
195  *	Read/write flag
196  * Essentially all the work is computing physical addresses and
197  * validating them.
198  * If the user has the proper access privilidges, the process is
199  * marked 'delayed unlock' and the pages involved in the I/O are
200  * faulted and locked. After the completion of the I/O, the above pages
201  * are unlocked.
202  */
203 physio(strat, bp, dev, rw, mincnt, uio)
204 	int (*strat)();
205 	register struct buf *bp;
206 	dev_t dev;
207 	int rw;
208 	unsigned (*mincnt)();
209 	struct uio *uio;
210 {
211 	register struct iovec *iov;
212 	register int c;
213 	char *a;
214 	int s, error = 0;
215 #ifdef SECSIZE
216 	int bsize;
217 	struct partinfo dpart;
218 #endif SECSIZE
219 
220 #ifdef SECSIZE
221 	if ((unsigned)major(dev) < nchrdev &&
222 	    (*cdevsw[major(dev)].d_ioctl)(dev, DIOCGPART, (caddr_t)&dpart,
223 	    FREAD) == 0)
224 		bsize = dpart.disklab->d_secsize;
225 	else
226 		bsize = DEV_BSIZE;
227 #endif SECSIZE
228 	for (;;) {
229 		if (uio->uio_iovcnt == 0)
230 			return (0);
231 		iov = uio->uio_iov;
232 		if (useracc(iov->iov_base, (u_int)iov->iov_len,
233 		    rw==B_READ? B_WRITE : B_READ) == NULL)
234 			return (EFAULT);
235 		s = splbio();
236 		while (bp->b_flags&B_BUSY) {
237 			bp->b_flags |= B_WANTED;
238 			sleep((caddr_t)bp, PRIBIO+1);
239 		}
240 		splx(s);
241 		bp->b_error = 0;
242 		bp->b_proc = u.u_procp;
243 #ifdef SECSIZE
244 		bp->b_blksize = bsize;
245 #endif SECSIZE
246 		bp->b_un.b_addr = iov->iov_base;
247 		while (iov->iov_len > 0) {
248 			bp->b_flags = B_BUSY | B_PHYS | rw;
249 			bp->b_dev = dev;
250 #ifdef SECSIZE
251 			bp->b_blkno = uio->uio_offset / bsize;
252 #else SECSIZE
253 			bp->b_blkno = btodb(uio->uio_offset);
254 #endif SECSIZE
255 			bp->b_bcount = iov->iov_len;
256 			(*mincnt)(bp);
257 			c = bp->b_bcount;
258 			u.u_procp->p_flag |= SPHYSIO;
259 			vslock(a = bp->b_un.b_addr, c);
260 			physstrat(bp, strat, PRIBIO);
261 			(void) splbio();
262 			vsunlock(a, c, rw);
263 			u.u_procp->p_flag &= ~SPHYSIO;
264 			if (bp->b_flags&B_WANTED)
265 				wakeup((caddr_t)bp);
266 			splx(s);
267 			c -= bp->b_resid;
268 			bp->b_un.b_addr += c;
269 			iov->iov_len -= c;
270 			uio->uio_resid -= c;
271 			uio->uio_offset += c;
272 			/* temp kludge for tape drives */
273 			if (bp->b_resid || (bp->b_flags&B_ERROR))
274 				break;
275 		}
276 		bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS);
277 		error = geterror(bp);
278 		/* temp kludge for tape drives */
279 		if (bp->b_resid || error)
280 			return (error);
281 		uio->uio_iov++;
282 		uio->uio_iovcnt--;
283 	}
284 }
285 
286 unsigned
287 minphys(bp)
288 	struct buf *bp;
289 {
290 
291 	if (bp->b_bcount > MAXPHYS)
292 		bp->b_bcount = MAXPHYS;
293 }
294