xref: /original-bsd/sys/kern/kern_physio.c (revision 90bde559)
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	7.14 (Berkeley) 05/15/90
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 #include "specdev.h"
21 
22 #include "machine/pte.h"
23 
24 /*
25  * Swap IO headers -
26  * They contain the necessary information for the swap I/O.
27  * At any given time, a swap header can be in three
28  * different lists. When free it is in the free list,
29  * when allocated and the I/O queued, it is on the swap
30  * device list, and finally, if the operation was a dirty
31  * page push, when the I/O completes, it is inserted
32  * in a list of cleaned pages to be processed by the pageout daemon.
33  */
34 struct	buf *swbuf;
35 
36 /*
37  * swap I/O -
38  *
39  * If the flag indicates a dirty page push initiated
40  * by the pageout daemon, we map the page into the i th
41  * virtual page of process 2 (the daemon itself) where i is
42  * the index of the swap header that has been allocated.
43  * We simply initialize the header and queue the I/O but
44  * do not wait for completion. When the I/O completes,
45  * biodone() will link the header to a list of cleaned
46  * pages to be processed by the pageout daemon.
47  */
48 swap(p, dblkno, addr, nbytes, rdflg, flag, vp, pfcent)
49 	struct proc *p;
50 	swblk_t dblkno;
51 	caddr_t addr;
52 	int nbytes, rdflg, flag;
53 	struct vnode *vp;
54 	u_int pfcent;
55 {
56 	register struct buf *bp;
57 	register struct pte *dpte, *vpte;
58 	register u_int c;
59 	int p2dp, s, error = 0;
60 	struct buf *getswbuf();
61 	int swdone();
62 
63 	bp = getswbuf(PSWP+1);
64 	bp->b_flags = B_BUSY | B_PHYS | rdflg | flag;
65 	if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0)
66 		if (rdflg == B_READ)
67 			sum.v_pswpin += btoc(nbytes);
68 		else
69 			sum.v_pswpout += btoc(nbytes);
70 	bp->b_proc = p;
71 	if (flag & B_DIRTY) {
72 		p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
73 		dpte = dptopte(&proc[2], p2dp);
74 		vpte = vtopte(p, btop(addr));
75 		for (c = 0; c < nbytes; c += NBPG) {
76 			if (vpte->pg_pfnum == 0 || vpte->pg_fod)
77 				panic("swap bad pte");
78 			*dpte++ = *vpte++;
79 		}
80 		bp->b_un.b_addr = (caddr_t)ctob(dptov(&proc[2], p2dp));
81 		bp->b_flags |= B_CALL;
82 		bp->b_iodone = swdone;
83 		bp->b_pfcent = pfcent;
84 	} else
85 		bp->b_un.b_addr = addr;
86 	while (nbytes > 0) {
87 		bp->b_blkno = dblkno;
88 		if (bp->b_vp)
89 			brelvp(bp);
90 		VHOLD(vp);
91 		bp->b_vp = vp;
92 		bp->b_dev = vp->v_rdev;
93 		bp->b_bcount = nbytes;
94 		if ((bp->b_flags & B_READ) == 0)
95 			vp->v_numoutput++;
96 		minphys(bp);
97 		c = bp->b_bcount;
98 #ifdef TRACE
99 		trace(TR_SWAPIO, vp, bp->b_blkno);
100 #endif
101 #if defined(hp300)
102 		vmapbuf(bp);
103 #endif
104 		VOP_STRATEGY(bp);
105 		/* pageout daemon doesn't wait for pushed pages */
106 		if (flag & B_DIRTY) {
107 			if (c < nbytes)
108 				panic("big push");
109 			return (0);
110 		} else {
111 			s = splbio();
112 			while ((bp->b_flags & B_DONE) == 0)
113 				sleep((caddr_t)bp, PSWP);
114 			splx(s);
115 		}
116 #if defined(hp300)
117 		vunmapbuf(bp);
118 #endif
119 		bp->b_un.b_addr += c;
120 		bp->b_flags &= ~B_DONE;
121 		if (bp->b_flags & B_ERROR) {
122 			if ((flag & (B_UAREA|B_PAGET)) || rdflg == B_WRITE)
123 				panic("hard IO err in swap");
124 			swkill(p, "swap: read error from swap device");
125 			error = EIO;
126 		}
127 		nbytes -= c;
128 		dblkno += btodb(c);
129 	}
130 	bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY);
131 	freeswbuf(bp);
132 	return (error);
133 }
134 
135 /*
136  * Put a buffer on the clean list after I/O is done.
137  * Called from biodone.
138  */
139 swdone(bp)
140 	register struct buf *bp;
141 {
142 	register int s;
143 
144 	if (bp->b_flags & B_ERROR)
145 		panic("IO err in push");
146 	s = splbio();
147 	bp->av_forw = bclnlist;
148 	cnt.v_pgout++;
149 	cnt.v_pgpgout += bp->b_bcount / NBPG;
150 	bclnlist = bp;
151 	if (bswlist.b_flags & B_WANTED)
152 		wakeup((caddr_t)&proc[2]);
153 #if defined(hp300)
154 	vunmapbuf(bp);
155 #endif
156 	splx(s);
157 }
158 
159 /*
160  * If rout == 0 then killed on swap error, else
161  * rout is the name of the routine where we ran out of
162  * swap space.
163  */
164 swkill(p, rout)
165 	struct proc *p;
166 	char *rout;
167 {
168 
169 	printf("pid %d: %s\n", p->p_pid, rout);
170 	uprintf("sorry, pid %d was killed in %s\n", p->p_pid, rout);
171 	/*
172 	 * To be sure no looping (e.g. in vmsched trying to
173 	 * swap out) mark process locked in core (as though
174 	 * done by user) after killing it so noone will try
175 	 * to swap it out.
176 	 */
177 	psignal(p, SIGKILL);
178 	p->p_flag |= SULOCK;
179 }
180 
181 /*
182  * Raw I/O. The arguments are
183  *	The strategy routine for the device
184  *	A buffer, which will either be a special buffer header owned
185  *	    exclusively by the device for this purpose, or NULL,
186  *	    indicating that we should use a swap buffer
187  *	The device number
188  *	Read/write flag
189  * Essentially all the work is computing physical addresses and
190  * validating them.
191  * If the user has the proper access privilidges, the process is
192  * marked 'delayed unlock' and the pages involved in the I/O are
193  * faulted and locked. After the completion of the I/O, the above pages
194  * are unlocked.
195  */
196 physio(strat, bp, dev, rw, mincnt, uio)
197 	int (*strat)();
198 	register struct buf *bp;
199 	dev_t dev;
200 	int rw;
201 	u_int (*mincnt)();
202 	struct uio *uio;
203 {
204 	register struct iovec *iov;
205 	register int requested, done;
206 	char *a;
207 	int s, allocbuf = 0, error = 0;
208 	struct buf *getswbuf();
209 
210 	if (bp == NULL) {
211 		allocbuf = 1;
212 		bp = getswbuf(PRIBIO+1);
213 	}
214 	for (; uio->uio_iovcnt; uio->uio_iov++, uio->uio_iovcnt--) {
215 		iov = uio->uio_iov;
216 		if (!useracc(iov->iov_base, (u_int)iov->iov_len,
217 		    rw == B_READ ? B_WRITE : B_READ)) {
218 			error = EFAULT;
219 			break;
220 		}
221 		if (!allocbuf) {	/* only if sharing caller's buffer */
222 			s = splbio();
223 			while (bp->b_flags&B_BUSY) {
224 				bp->b_flags |= B_WANTED;
225 				sleep((caddr_t)bp, PRIBIO+1);
226 			}
227 			splx(s);
228 		}
229 		bp->b_error = 0;
230 		bp->b_proc = u.u_procp;
231 #ifdef HPUXCOMPAT
232 		if (ISHPMMADDR(iov->iov_base))
233 			bp->b_un.b_addr = (caddr_t)HPMMBASEADDR(iov->iov_base);
234 		else
235 #endif
236 		bp->b_un.b_addr = iov->iov_base;
237 		while (iov->iov_len > 0) {
238 			bp->b_flags = B_BUSY | B_PHYS | B_RAW | rw;
239 			bp->b_dev = dev;
240 			bp->b_blkno = btodb(uio->uio_offset);
241 			bp->b_bcount = iov->iov_len;
242 			(*mincnt)(bp);
243 			requested = bp->b_bcount;
244 			u.u_procp->p_flag |= SPHYSIO;
245 			vslock(a = bp->b_un.b_addr, requested);
246 #if defined(hp300)
247 			vmapbuf(bp);
248 #endif
249 			(*strat)(bp);
250 			s = splbio();
251 			while ((bp->b_flags & B_DONE) == 0)
252 				sleep((caddr_t)bp, PRIBIO);
253 #if defined(hp300)
254 			vunmapbuf(bp);
255 #endif
256 			vsunlock(a, requested, rw);
257 			u.u_procp->p_flag &= ~SPHYSIO;
258 			if (bp->b_flags&B_WANTED)	/* rare */
259 				wakeup((caddr_t)bp);
260 			splx(s);
261 			done = bp->b_bcount - bp->b_resid;
262 			bp->b_un.b_addr += done;
263 			iov->iov_len -= done;
264 			uio->uio_resid -= done;
265 			uio->uio_offset += done;
266 			/* temp kludge for disk drives */
267 			if (done < requested || bp->b_flags & B_ERROR)
268 				break;
269 		}
270 		bp->b_flags &= ~(B_BUSY | B_WANTED | B_PHYS | B_RAW);
271 		error = biowait(bp);
272 		/* temp kludge for disk drives */
273 		if (done < requested || bp->b_flags & B_ERROR)
274 			break;
275 	}
276 #if defined(hp300)
277 	DCIU();
278 #endif
279 	if (allocbuf)
280 		freeswbuf(bp);
281 	return (error);
282 }
283 
284 u_int
285 minphys(bp)
286 	struct buf *bp;
287 {
288 	if (bp->b_bcount > MAXPHYS)
289 		bp->b_bcount = MAXPHYS;
290 }
291 
292 static
293 struct buf *
294 getswbuf(prio)
295 	int prio;
296 {
297 	int s;
298 	struct buf *bp;
299 
300 	s = splbio();
301 	while (bswlist.av_forw == NULL) {
302 		bswlist.b_flags |= B_WANTED;
303 		sleep((caddr_t)&bswlist, prio);
304 	}
305 	bp = bswlist.av_forw;
306 	bswlist.av_forw = bp->av_forw;
307 	splx(s);
308 	return (bp);
309 }
310 
311 static
312 freeswbuf(bp)
313 	struct buf *bp;
314 {
315 	int s;
316 
317 	s = splbio();
318 	bp->av_forw = bswlist.av_forw;
319 	bswlist.av_forw = bp;
320 	if (bp->b_vp)
321 		brelvp(bp);
322 	if (bswlist.b_flags & B_WANTED) {
323 		bswlist.b_flags &= ~B_WANTED;
324 		wakeup((caddr_t)&bswlist);
325 		wakeup((caddr_t)&proc[2]);
326 	}
327 	splx(s);
328 }
329 
330 rawread(dev, uio)
331 	dev_t dev;
332 	struct uio *uio;
333 {
334 	return (physio(cdevsw[major(dev)].d_strategy, (struct buf *)NULL,
335 	    dev, B_READ, minphys, uio));
336 }
337 
338 rawwrite(dev, uio)
339 	dev_t dev;
340 	struct uio *uio;
341 {
342 	return (physio(cdevsw[major(dev)].d_strategy, (struct buf *)NULL,
343 	    dev, B_WRITE, minphys, uio));
344 }
345