xref: /csrg-svn/sys/kern/kern_physio.c (revision 24448)
1 /*
2  * Copyright (c) 1982 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	6.7 (Berkeley) 08/28/85
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 
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  * iodone() 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, dev, pfcent)
49 	struct proc *p;
50 	swblk_t dblkno;
51 	caddr_t addr;
52 	int nbytes, rdflg, flag;
53 	dev_t dev;
54 	u_int pfcent;
55 {
56 	register struct buf *bp;
57 	register u_int c;
58 	int p2dp;
59 	register struct pte *dpte, *vpte;
60 	int s;
61 	extern swdone();
62 
63 	s = spl6();
64 	while (bswlist.av_forw == NULL) {
65 		bswlist.b_flags |= B_WANTED;
66 		sleep((caddr_t)&bswlist, PSWP+1);
67 	}
68 	bp = bswlist.av_forw;
69 	bswlist.av_forw = bp->av_forw;
70 	splx(s);
71 
72 	bp->b_flags = B_BUSY | B_PHYS | rdflg | flag;
73 	if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0)
74 		if (rdflg == B_READ)
75 			sum.v_pswpin += btoc(nbytes);
76 		else
77 			sum.v_pswpout += btoc(nbytes);
78 	bp->b_proc = p;
79 	if (flag & B_DIRTY) {
80 		p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
81 		dpte = dptopte(&proc[2], p2dp);
82 		vpte = vtopte(p, btop(addr));
83 		for (c = 0; c < nbytes; c += NBPG) {
84 			if (vpte->pg_pfnum == 0 || vpte->pg_fod)
85 				panic("swap bad pte");
86 			*dpte++ = *vpte++;
87 		}
88 		bp->b_un.b_addr = (caddr_t)ctob(dptov(&proc[2], p2dp));
89 		bp->b_flags |= B_CALL;
90 		bp->b_iodone = swdone;
91 		bp->b_pfcent = pfcent;
92 	} else
93 		bp->b_un.b_addr = addr;
94 	while (nbytes > 0) {
95 		bp->b_bcount = nbytes;
96 		minphys(bp);
97 		c = bp->b_bcount;
98 		bp->b_blkno = dblkno;
99 		bp->b_dev = dev;
100 #ifdef TRACE
101 		trace(TR_SWAPIO, dev, bp->b_blkno);
102 #endif
103 		physstrat(bp, bdevsw[major(dev)].d_strategy, PSWP);
104 		if (flag & B_DIRTY) {
105 			if (c < nbytes)
106 				panic("big push");
107 			return;
108 		}
109 		bp->b_un.b_addr += c;
110 		bp->b_flags &= ~B_DONE;
111 		if (bp->b_flags & B_ERROR) {
112 			if ((flag & (B_UAREA|B_PAGET)) || rdflg == B_WRITE)
113 				panic("hard IO err in swap");
114 			swkill(p, "swap: read error from swap device");
115 		}
116 		nbytes -= c;
117 		dblkno += btodb(c);
118 	}
119 	s = spl6();
120 	bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY);
121 	bp->av_forw = bswlist.av_forw;
122 	bswlist.av_forw = bp;
123 	if (bswlist.b_flags & B_WANTED) {
124 		bswlist.b_flags &= ~B_WANTED;
125 		wakeup((caddr_t)&bswlist);
126 		wakeup((caddr_t)&proc[2]);
127 	}
128 	splx(s);
129 }
130 
131 /*
132  * Put a buffer on the clean list after I/O is done.
133  * Called from biodone.
134  */
135 swdone(bp)
136 	register struct buf *bp;
137 {
138 	register int s;
139 
140 	if (bp->b_flags & B_ERROR)
141 		panic("IO err in push");
142 	s = spl6();
143 	bp->av_forw = bclnlist;
144 	cnt.v_pgout++;
145 	cnt.v_pgpgout += bp->b_bcount / NBPG;
146 	bclnlist = bp;
147 	if (bswlist.b_flags & B_WANTED)
148 		wakeup((caddr_t)&proc[2]);
149 	splx(s);
150 }
151 
152 /*
153  * If rout == 0 then killed on swap error, else
154  * rout is the name of the routine where we ran out of
155  * swap space.
156  */
157 swkill(p, rout)
158 	struct proc *p;
159 	char *rout;
160 {
161 
162 	printf("pid %d: %s\n", p->p_pid, rout);
163 	uprintf("sorry, pid %d was killed in %s\n", p->p_pid, rout);
164 	/*
165 	 * To be sure no looping (e.g. in vmsched trying to
166 	 * swap out) mark process locked in core (as though
167 	 * done by user) after killing it so noone will try
168 	 * to swap it out.
169 	 */
170 	psignal(p, SIGKILL);
171 	p->p_flag |= SULOCK;
172 }
173 
174 /*
175  * Raw I/O. The arguments are
176  *	The strategy routine for the device
177  *	A buffer, which will always be a special buffer
178  *	  header owned exclusively by the device for this purpose
179  *	The device number
180  *	Read/write flag
181  * Essentially all the work is computing physical addresses and
182  * validating them.
183  * If the user has the proper access privilidges, the process is
184  * marked 'delayed unlock' and the pages involved in the I/O are
185  * faulted and locked. After the completion of the I/O, the above pages
186  * are unlocked.
187  */
188 physio(strat, bp, dev, rw, mincnt, uio)
189 	int (*strat)();
190 	register struct buf *bp;
191 	dev_t dev;
192 	int rw;
193 	unsigned (*mincnt)();
194 	struct uio *uio;
195 {
196 	register struct iovec *iov;
197 	register int c;
198 	char *a;
199 	int s, error = 0;
200 
201 nextiov:
202 	if (uio->uio_iovcnt == 0)
203 		return (0);
204 	iov = uio->uio_iov;
205 	if (useracc(iov->iov_base,(u_int)iov->iov_len,rw==B_READ?B_WRITE:B_READ) == NULL)
206 		return (EFAULT);
207 	s = spl6();
208 	while (bp->b_flags&B_BUSY) {
209 		bp->b_flags |= B_WANTED;
210 		sleep((caddr_t)bp, PRIBIO+1);
211 	}
212 	splx(s);
213 	bp->b_error = 0;
214 	bp->b_proc = u.u_procp;
215 	bp->b_un.b_addr = iov->iov_base;
216 	while (iov->iov_len > 0) {
217 		bp->b_flags = B_BUSY | B_PHYS | rw;
218 		bp->b_dev = dev;
219 		bp->b_blkno = btodb(uio->uio_offset);
220 		bp->b_bcount = iov->iov_len;
221 		(*mincnt)(bp);
222 		c = bp->b_bcount;
223 		u.u_procp->p_flag |= SPHYSIO;
224 		vslock(a = bp->b_un.b_addr, c);
225 		physstrat(bp, strat, PRIBIO);
226 		(void) spl6();
227 		vsunlock(a, c, rw);
228 		u.u_procp->p_flag &= ~SPHYSIO;
229 		if (bp->b_flags&B_WANTED)
230 			wakeup((caddr_t)bp);
231 		splx(s);
232 		c -= bp->b_resid;
233 		bp->b_un.b_addr += c;
234 		iov->iov_len -= c;
235 		uio->uio_resid -= c;
236 		uio->uio_offset += c;
237 		/* temp kludge for tape drives */
238 		if (bp->b_resid || (bp->b_flags&B_ERROR))
239 			break;
240 	}
241 	bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS);
242 	error = geterror(bp);
243 	/* temp kludge for tape drives */
244 	if (bp->b_resid || error)
245 		return (error);
246 	uio->uio_iov++;
247 	uio->uio_iovcnt--;
248 	goto nextiov;
249 }
250 
251 #define	MAXPHYS	(63 * 1024)
252 
253 unsigned
254 minphys(bp)
255 	struct buf *bp;
256 {
257 
258 	if (bp->b_bcount > MAXPHYS)
259 		bp->b_bcount = MAXPHYS;
260 }
261