xref: /netbsd-src/sys/kern/kern_physio.c (revision 1f2744e6e4915c9da2a3f980279398c4cf7d5e6d)
1 /*	$NetBSD: kern_physio.c,v 1.20 1994/06/29 06:32:34 cgd Exp $	*/
2 
3 /*-
4  * Copyright (c) 1994 Christopher G. Demetriou
5  * Copyright (c) 1982, 1986, 1990, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *	This product includes software developed by the University of
24  *	California, Berkeley and its contributors.
25  * 4. Neither the name of the University nor the names of its contributors
26  *    may be used to endorse or promote products derived from this software
27  *    without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39  * SUCH DAMAGE.
40  *
41  *	@(#)kern_physio.c	8.1 (Berkeley) 6/10/93
42  */
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/buf.h>
47 #include <sys/conf.h>
48 #include <sys/proc.h>
49 
50 /*
51  * The routines implemented in this file are described in:
52  *	Leffler, et al.: The Design and Implementation of the 4.3BSD
53  *	    UNIX Operating System (Addison Welley, 1989)
54  * on pages 231-233.
55  *
56  * The routines "getphysbuf" and "putphysbuf" steal and return a swap
57  * buffer.  Leffler, et al., says that swap buffers are used to do the
58  * I/O, so raw I/O requests don't have to be single-threaded.
59  */
60 
61 struct buf *getphysbuf __P((void));
62 void putphysbuf __P((struct buf *bp));
63 
64 /*
65  * Do "physical I/O" on behalf of a user.  "Physical I/O" is I/O directly
66  * from the raw device to user buffers, and bypasses the buffer cache.
67  *
68  * Comments in brackets are from Leffler, et al.'s pseudo-code implementation.
69  */
70 int
71 physio(strategy, bp, dev, flags, minphys, uio)
72 	int (*strategy)();
73 	struct buf *bp;
74 	dev_t dev;
75 	int flags;
76 	u_int (*minphys)();
77 	struct uio *uio;
78 {
79 	struct iovec *iovp;
80 	struct proc *p = curproc;
81 	int error, done, i, nobuf, s, todo;
82 
83 	error = 0;
84 	flags &= B_READ | B_WRITE;
85 
86 	/*
87 	 * [check user read/write access to the data buffer]
88 	 *
89 	 * Check each iov one by one.  Note that we know if we're reading or
90 	 * writing, so we ignore the uio's rw parameter.  Also note that if
91 	 * we're doing a read, that's a *write* to user-space.
92 	 */
93 	for (i = 0; i < uio->uio_iovcnt; i++)
94 		if (!useracc(uio->uio_iov[i].iov_base, uio->uio_iov[i].iov_len,
95 		    (flags == B_READ) ? B_WRITE : B_READ))
96 			return (EFAULT);
97 
98 	/* Make sure we have a buffer, creating one if necessary. */
99 	if (nobuf = (bp == NULL))
100 		bp = getphysbuf();
101 
102 	/* [raise the processor priority level to splbio;] */
103 	s = splbio();
104 
105 	/* [while the buffer is marked busy] */
106 	while (bp->b_flags & B_BUSY) {
107 		/* [mark the buffer wanted] */
108 		bp->b_flags |= B_WANTED;
109 		/* [wait until the buffer is available] */
110 		tsleep((caddr_t)bp, PRIBIO+1, "physbuf", 0);
111 	}
112 
113 	/* Mark it busy, so nobody else will use it. */
114 	bp->b_flags |= B_BUSY;
115 
116 	/* [lower the priority level] */
117 	splx(s);
118 
119 	/* [set up the fixed part of the buffer for a transfer] */
120 	bp->b_dev = dev;
121 	bp->b_error = 0;
122 	bp->b_proc = p;
123 
124 	/*
125 	 * [while there are data to transfer and no I/O error]
126 	 * Note that I/O errors are handled with a 'goto' at the bottom
127 	 * of the 'while' loop.
128 	 */
129 	for (i = 0; i < uio->uio_iovcnt; i++) {
130 		iovp = &uio->uio_iov[i];
131 		while (iovp->iov_len > 0) {
132 			/*
133 			 * [mark the buffer busy for physical I/O]
134 			 * (i.e. set B_PHYS (because it's an I/O to user
135 			 * memory, and B_RAW, because B_RAW is to be
136 			 * "Set by physio for raw transfers.", in addition
137 			 * to the "busy" and read/write flag.)
138 			 */
139 			s = splbio();
140 			bp->b_flags = B_BUSY | B_PHYS | B_RAW | flags;
141 			splx(s);
142 
143 			/* [set up the buffer for a maximum-sized transfer] */
144 			bp->b_blkno = btodb(uio->uio_offset);
145 			bp->b_bcount = iovp->iov_len;
146 			bp->b_data = iovp->iov_base;
147 
148 			/*
149 			 * [call minphys to bound the tranfer size]
150 			 * and remember the amount of data to transfer,
151 			 * for later comparison.
152 			 */
153 			(*minphys)(bp);
154 			todo = bp->b_bcount;
155 
156 			/*
157 			 * [lock the part of the user address space involved
158 			 *    in the transfer]
159 			 * Beware vmapbuf(); it clobbers b_data and
160 			 * saves it in b_saveaddr.  However, vunmapbuf()
161 			 * restores it.
162 			 */
163 			p->p_holdcnt++;
164 			vslock(bp->b_data, todo);
165 			vmapbuf(bp, todo);
166 
167 			/* [call strategy to start the transfer] */
168 			(*strategy)(bp);
169 
170 			/*
171 			 * Note that the raise/wait/lower/get error
172 			 * steps below would be done by biowait(), but
173 			 * we want to unlock the address space before
174 			 * we lower the priority.
175 			 *
176 			 * [raise the priority level to splbio]
177 			 */
178 			s = splbio();
179 
180 			/* [wait for the transfer to complete] */
181 			while ((bp->b_flags & B_DONE) == 0)
182 				tsleep((caddr_t) bp, PRIBIO + 1, "physio", 0);
183 
184 			/*
185 			 * [unlock the part of the address space previously
186 			 *    locked]
187 			 */
188 			vunmapbuf(bp, todo);
189 			vsunlock(bp->b_data, todo);
190 			p->p_holdcnt--;
191 
192 			/* remember error value (save a splbio/splx pair) */
193 			if (bp->b_flags & B_ERROR)
194 				error = (bp->b_error ? bp->b_error : EIO);
195 
196 			/* [lower the priority level] */
197 			splx(s);
198 
199 			/*
200 			 * [deduct the transfer size from the total number
201 			 *    of data to transfer]
202 			 */
203 			done = bp->b_bcount - bp->b_resid;
204 			iovp->iov_len -= done;
205                         iovp->iov_base += done;
206                         uio->uio_offset += done;
207                         uio->uio_resid -= done;
208 
209 			/*
210 			 * Now, check for an error.
211 			 * Also, handle weird end-of-disk semantics.
212 			 */
213 			if (error || done < todo)
214 				goto done;
215 		}
216 	}
217 
218 done:
219 	/*
220 	 * [clean up the state of the buffer]
221 	 * Remember if somebody wants it, so we can wake them up below.
222 	 * Also, if we had to steal it, give it back.
223 	 */
224 	s = splbio();
225 	bp->b_flags &= ~(B_BUSY | B_PHYS | B_RAW);
226 	if (nobuf)
227 		putphysbuf(bp);
228 	else {
229 		/*
230 		 * [if another process is waiting for the raw I/O buffer,
231 		 *    wake up processes waiting to do physical I/O;
232 		 */
233 		if (bp->b_flags & B_WANTED) {
234 			bp->b_flags &= ~B_WANTED;
235 			wakeup(bp);
236 		}
237 	}
238 	splx(s);
239 
240 	return (error);
241 }
242 
243 /*
244  * Get a swap buffer structure, for use in physical I/O.
245  * Mostly taken from /sys/vm/swap_pager.c, except that it no longer
246  * records buffer list-empty conditions, and sleeps at PRIBIO + 1,
247  * rather than PSWP + 1 (and on a different wchan).
248  */
249 struct buf *
250 getphysbuf()
251 {
252 	struct buf *bp;
253 	int s;
254 
255 	s = splbio();
256         while (bswlist.b_actf == NULL) {
257                 bswlist.b_flags |= B_WANTED;
258                 tsleep((caddr_t)&bswlist, PRIBIO + 1, "getphys", 0);
259         }
260         bp = bswlist.b_actf;
261         bswlist.b_actf = bp->b_actf;
262         splx(s);
263 	return (bp);
264 }
265 
266 /*
267  * Get rid of a swap buffer structure which has been used in physical I/O.
268  * Mostly taken from /sys/vm/swap_pager.c, except that it now uses
269  * wakeup() rather than the VM-internal thread_wakeup(), and that the caller
270  * must mask disk interrupts, rather than putphysbuf() itself.
271  */
272 void
273 putphysbuf(bp)
274 	struct buf *bp;
275 {
276 
277         bp->b_actf = bswlist.b_actf;
278         bswlist.b_actf = bp;
279         if (bp->b_vp)
280                 brelvp(bp);
281         if (bswlist.b_flags & B_WANTED) {
282                 bswlist.b_flags &= ~B_WANTED;
283                 wakeup(&bswlist);
284         }
285 }
286 
287 /*
288  * Leffler, et al., says on p. 231:
289  * "The minphys() routine is called by physio() to adjust the
290  * size of each I/O transfer before the latter is passed to
291  * the strategy routine..."
292  *
293  * so, just adjust the buffer's count accounting to MAXPHYS here,
294  * and return the new count;
295  */
296 u_int
297 minphys(bp)
298 	struct buf *bp;
299 {
300 
301 	bp->b_bcount = min(MAXPHYS, bp->b_bcount);
302         return bp->b_bcount;
303 }
304 
305 /*
306  * Do a read on a device for a user process.
307  */
308 rawread(dev, uio)
309 	dev_t dev;
310 	struct uio *uio;
311 {
312 	return (physio(cdevsw[major(dev)].d_strategy, (struct buf *)NULL,
313 	    dev, B_READ, minphys, uio));
314 }
315 
316 /*
317  * Do a write on a device for a user process.
318  */
319 rawwrite(dev, uio)
320 	dev_t dev;
321 	struct uio *uio;
322 {
323 	return (physio(cdevsw[major(dev)].d_strategy, (struct buf *)NULL,
324 	    dev, B_WRITE, minphys, uio));
325 }
326