xref: /csrg-svn/sys/kern/kern_physio.c (revision 18319)
1 /*	kern_physio.c	6.5	85/03/12	*/
2 
3 #include "../machine/pte.h"
4 
5 #include "param.h"
6 #include "systm.h"
7 #include "dir.h"
8 #include "user.h"
9 #include "buf.h"
10 #include "conf.h"
11 #include "proc.h"
12 #include "seg.h"
13 #include "vm.h"
14 #include "trace.h"
15 #include "map.h"
16 #include "uio.h"
17 
18 /*
19  * Swap IO headers -
20  * They contain the necessary information for the swap I/O.
21  * At any given time, a swap header can be in three
22  * different lists. When free it is in the free list,
23  * when allocated and the I/O queued, it is on the swap
24  * device list, and finally, if the operation was a dirty
25  * page push, when the I/O completes, it is inserted
26  * in a list of cleaned pages to be processed by the pageout daemon.
27  */
28 struct	buf *swbuf;
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 nbytes, rdflg, flag;
47 	dev_t dev;
48 	u_int pfcent;
49 {
50 	register struct buf *bp;
51 	register u_int c;
52 	int p2dp;
53 	register struct pte *dpte, *vpte;
54 	int s;
55 	extern swdone();
56 
57 	s = spl6();
58 	while (bswlist.av_forw == NULL) {
59 		bswlist.b_flags |= B_WANTED;
60 		sleep((caddr_t)&bswlist, PSWP+1);
61 	}
62 	bp = bswlist.av_forw;
63 	bswlist.av_forw = bp->av_forw;
64 	splx(s);
65 
66 	bp->b_flags = B_BUSY | B_PHYS | rdflg | flag;
67 	if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0)
68 		if (rdflg == B_READ)
69 			sum.v_pswpin += btoc(nbytes);
70 		else
71 			sum.v_pswpout += btoc(nbytes);
72 	bp->b_proc = p;
73 	if (flag & B_DIRTY) {
74 		p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
75 		dpte = dptopte(&proc[2], p2dp);
76 		vpte = vtopte(p, btop(addr));
77 		for (c = 0; c < nbytes; c += NBPG) {
78 			if (vpte->pg_pfnum == 0 || vpte->pg_fod)
79 				panic("swap bad pte");
80 			*dpte++ = *vpte++;
81 		}
82 		bp->b_un.b_addr = (caddr_t)ctob(dptov(&proc[2], p2dp));
83 		bp->b_flags |= B_CALL;
84 		bp->b_iodone = swdone;
85 		bp->b_pfcent = pfcent;
86 	} else
87 		bp->b_un.b_addr = addr;
88 	while (nbytes > 0) {
89 		bp->b_bcount = nbytes;
90 		minphys(bp);
91 		c = bp->b_bcount;
92 		bp->b_blkno = dblkno;
93 		bp->b_dev = dev;
94 #ifdef TRACE
95 		trace(TR_SWAPIO, dev, bp->b_blkno);
96 #endif
97 		physstrat(bp, bdevsw[major(dev)].d_strategy, PSWP);
98 		if (flag & B_DIRTY) {
99 			if (c < nbytes)
100 				panic("big push");
101 			return;
102 		}
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, "swap: read error from swap device");
109 		}
110 		nbytes -= c;
111 		dblkno += btodb(c);
112 	}
113 	s = spl6();
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 	}
122 	splx(s);
123 }
124 
125 /*
126  * Put a buffer on the clean list after I/O is done.
127  * Called from biodone.
128  */
129 swdone(bp)
130 	register struct buf *bp;
131 {
132 	register int s;
133 
134 	if (bp->b_flags & B_ERROR)
135 		panic("IO err in push");
136 	s = spl6();
137 	bp->av_forw = bclnlist;
138 	cnt.v_pgout++;
139 	cnt.v_pgpgout += bp->b_bcount / NBPG;
140 	bclnlist = bp;
141 	if (bswlist.b_flags & B_WANTED)
142 		wakeup((caddr_t)&proc[2]);
143 	splx(s);
144 }
145 
146 /*
147  * If rout == 0 then killed on swap error, else
148  * rout is the name of the routine where we ran out of
149  * swap space.
150  */
151 swkill(p, rout)
152 	struct proc *p;
153 	char *rout;
154 {
155 
156 	printf("pid %d: %s", p->p_pid, rout);
157 	uprintf("sorry, pid %d was killed in %s", p->p_pid, rout);
158 	/*
159 	 * To be sure no looping (e.g. in vmsched trying to
160 	 * swap out) mark process locked in core (as though
161 	 * done by user) after killing it so noone will try
162 	 * to swap it out.
163 	 */
164 	psignal(p, SIGKILL);
165 	p->p_flag |= SULOCK;
166 }
167 
168 /*
169  * Raw I/O. The arguments are
170  *	The strategy routine for the device
171  *	A buffer, which will always be a special buffer
172  *	  header owned exclusively by the device for this purpose
173  *	The device number
174  *	Read/write flag
175  * Essentially all the work is computing physical addresses and
176  * validating them.
177  * If the user has the proper access privilidges, the process is
178  * marked 'delayed unlock' and the pages involved in the I/O are
179  * faulted and locked. After the completion of the I/O, the above pages
180  * are unlocked.
181  */
182 physio(strat, bp, dev, rw, mincnt, uio)
183 	int (*strat)();
184 	register struct buf *bp;
185 	dev_t dev;
186 	int rw;
187 	unsigned (*mincnt)();
188 	struct uio *uio;
189 {
190 	register struct iovec *iov;
191 	register int c;
192 	char *a;
193 	int s, error = 0;
194 
195 nextiov:
196 	if (uio->uio_iovcnt == 0)
197 		return (0);
198 	iov = uio->uio_iov;
199 	if (useracc(iov->iov_base,(u_int)iov->iov_len,rw==B_READ?B_WRITE:B_READ) == NULL)
200 		return (EFAULT);
201 	s = spl6();
202 	while (bp->b_flags&B_BUSY) {
203 		bp->b_flags |= B_WANTED;
204 		sleep((caddr_t)bp, PRIBIO+1);
205 	}
206 	splx(s);
207 	bp->b_error = 0;
208 	bp->b_proc = u.u_procp;
209 	bp->b_un.b_addr = iov->iov_base;
210 	while (iov->iov_len > 0) {
211 		bp->b_flags = B_BUSY | B_PHYS | rw;
212 		bp->b_dev = dev;
213 		bp->b_blkno = btodb(uio->uio_offset);
214 		bp->b_bcount = iov->iov_len;
215 		(*mincnt)(bp);
216 		c = bp->b_bcount;
217 		u.u_procp->p_flag |= SPHYSIO;
218 		vslock(a = bp->b_un.b_addr, c);
219 		physstrat(bp, strat, PRIBIO);
220 		(void) spl6();
221 		vsunlock(a, c, rw);
222 		u.u_procp->p_flag &= ~SPHYSIO;
223 		if (bp->b_flags&B_WANTED)
224 			wakeup((caddr_t)bp);
225 		splx(s);
226 		c -= bp->b_resid;
227 		bp->b_un.b_addr += c;
228 		iov->iov_len -= c;
229 		uio->uio_resid -= c;
230 		uio->uio_offset += c;
231 		/* temp kludge for tape drives */
232 		if (bp->b_resid || (bp->b_flags&B_ERROR))
233 			break;
234 	}
235 	bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS);
236 	error = geterror(bp);
237 	/* temp kludge for tape drives */
238 	if (bp->b_resid || error)
239 		return (error);
240 	uio->uio_iov++;
241 	uio->uio_iovcnt--;
242 	goto nextiov;
243 }
244 
245 #define	MAXPHYS	(63 * 1024)
246 
247 unsigned
248 minphys(bp)
249 	struct buf *bp;
250 {
251 
252 	if (bp->b_bcount > MAXPHYS)
253 		bp->b_bcount = MAXPHYS;
254 }
255