xref: /netbsd-src/sys/kern/kern_physio.c (revision 3b01aba77a7a698587faaae455bbfe740923c1f5)
1 /*	$NetBSD: kern_physio.c,v 1.49 2001/07/24 16:43:03 wiz 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/malloc.h>
48 #include <sys/proc.h>
49 
50 #include <uvm/uvm_extern.h>
51 
52 /*
53  * The routines implemented in this file are described in:
54  *	Leffler, et al.: The Design and Implementation of the 4.3BSD
55  *	    UNIX Operating System (Addison Welley, 1989)
56  * on pages 231-233.
57  *
58  * The routines "getphysbuf" and "putphysbuf" steal and return a swap
59  * buffer.  Leffler, et al., says that swap buffers are used to do the
60  * I/O, so raw I/O requests don't have to be single-threaded.
61  */
62 
63 struct buf *getphysbuf __P((void));
64 void putphysbuf __P((struct buf *bp));
65 
66 /*
67  * Do "physical I/O" on behalf of a user.  "Physical I/O" is I/O directly
68  * from the raw device to user buffers, and bypasses the buffer cache.
69  *
70  * Comments in brackets are from Leffler, et al.'s pseudo-code implementation.
71  */
72 int
73 physio(strategy, bp, dev, flags, minphys, uio)
74 	void (*strategy) __P((struct buf *));
75 	struct buf *bp;
76 	dev_t dev;
77 	int flags;
78 	void (*minphys) __P((struct buf *));
79 	struct uio *uio;
80 {
81 	struct iovec *iovp;
82 	struct proc *p = curproc;
83 	int error, done, i, nobuf, s;
84 	long todo;
85 
86 	error = 0;
87 	flags &= B_READ | B_WRITE | B_ORDERED;
88 
89 	/* Make sure we have a buffer, creating one if necessary. */
90 	if ((nobuf = (bp == NULL)) != 0) {
91 
92 		bp = getphysbuf();
93 		/* bp was just malloc'd so can't already be busy */
94 		bp->b_flags |= B_BUSY;
95 
96 	} else {
97 
98 		/* [raise the processor priority level to splbio;] */
99 		s = splbio();
100 
101 		/* [while the buffer is marked busy] */
102 		while (bp->b_flags & B_BUSY) {
103 			/* [mark the buffer wanted] */
104 			bp->b_flags |= B_WANTED;
105 			/* [wait until the buffer is available] */
106 			tsleep((caddr_t)bp, PRIBIO+1, "physbuf", 0);
107 		}
108 
109 		/* Mark it busy, so nobody else will use it. */
110 		bp->b_flags |= B_BUSY;
111 
112 		/* [lower the priority level] */
113 		splx(s);
114 	}
115 
116 	/* [set up the fixed part of the buffer for a transfer] */
117 	bp->b_dev = dev;
118 	bp->b_error = 0;
119 	bp->b_proc = p;
120 	LIST_INIT(&bp->b_dep);
121 
122 	/*
123 	 * [while there are data to transfer and no I/O error]
124 	 * Note that I/O errors are handled with a 'goto' at the bottom
125 	 * of the 'while' loop.
126 	 */
127 	for (i = 0; i < uio->uio_iovcnt; i++) {
128 		iovp = &uio->uio_iov[i];
129 		while (iovp->iov_len > 0) {
130 
131 			/*
132 			 * [mark the buffer busy for physical I/O]
133 			 * (i.e. set B_PHYS (because it's an I/O to user
134 			 * memory, and B_RAW, because B_RAW is to be
135 			 * "Set by physio for raw transfers.", in addition
136 			 * to the "busy" and read/write flag.)
137 			 */
138 			bp->b_flags = B_BUSY | B_PHYS | B_RAW | flags;
139 
140 			/* [set up the buffer for a maximum-sized transfer] */
141 			bp->b_blkno = btodb(uio->uio_offset);
142 			bp->b_bcount = iovp->iov_len;
143 			bp->b_data = iovp->iov_base;
144 
145 			/*
146 			 * [call minphys to bound the transfer size]
147 			 * and remember the amount of data to transfer,
148 			 * for later comparison.
149 			 */
150 			(*minphys)(bp);
151 			todo = bp->b_bcount;
152 #ifdef DIAGNOSTIC
153 			if (todo <= 0)
154 				panic("todo(%ld) <= 0; minphys broken", todo);
155 			if (todo > MAXPHYS)
156 				panic("todo(%ld) > MAXPHYS; minphys broken",
157 				      todo);
158 #endif
159 
160 			/*
161 			 * [lock the part of the user address space involved
162 			 *    in the transfer]
163 			 * Beware vmapbuf(); it clobbers b_data and
164 			 * saves it in b_saveaddr.  However, vunmapbuf()
165 			 * restores it.
166 			 */
167 			PHOLD(p);
168 			error = uvm_vslock(p, bp->b_data, todo,
169 					   (flags & B_READ) ?
170 					   VM_PROT_READ | VM_PROT_WRITE :
171 					   VM_PROT_READ);
172 			if (error) {
173 				bp->b_flags |= B_ERROR;
174 				bp->b_error = error;
175 				goto after_vsunlock;
176 			}
177 			vmapbuf(bp, todo);
178 
179 			/* [call strategy to start the transfer] */
180 			(*strategy)(bp);
181 
182 			/*
183 			 * Note that the raise/wait/lower/get error
184 			 * steps below would be done by biowait(), but
185 			 * we want to unlock the address space before
186 			 * we lower the priority.
187 			 *
188 			 * [raise the priority level to splbio]
189 			 */
190 			s = splbio();
191 
192 			/* [wait for the transfer to complete] */
193 			while ((bp->b_flags & B_DONE) == 0)
194 				tsleep((caddr_t) bp, PRIBIO + 1, "physio", 0);
195 
196 			/* Mark it busy again, so nobody else will use it. */
197 			bp->b_flags |= B_BUSY;
198 
199 			/* [lower the priority level] */
200 			splx(s);
201 
202 			/*
203 			 * [unlock the part of the address space previously
204 			 *    locked]
205 			 */
206 			vunmapbuf(bp, todo);
207 			uvm_vsunlock(p, bp->b_data, todo);
208  after_vsunlock:
209 			PRELE(p);
210 
211 			/* remember error value (save a splbio/splx pair) */
212 			if (bp->b_flags & B_ERROR)
213 				error = (bp->b_error ? bp->b_error : EIO);
214 
215 			/*
216 			 * [deduct the transfer size from the total number
217 			 *    of data to transfer]
218 			 */
219 			done = bp->b_bcount - bp->b_resid;
220 			KASSERT(done >= 0);
221 			KASSERT(done <= todo);
222 
223 			iovp->iov_len -= done;
224 			iovp->iov_base = (caddr_t)iovp->iov_base + done;
225 			uio->uio_offset += done;
226 			uio->uio_resid -= done;
227 
228 			/*
229 			 * Now, check for an error.
230 			 * Also, handle weird end-of-disk semantics.
231 			 */
232 			if (error || done < todo)
233 				goto done;
234 		}
235 	}
236 
237 done:
238 	/*
239 	 * [clean up the state of the buffer]
240 	 * Remember if somebody wants it, so we can wake them up below.
241 	 * Also, if we had to steal it, give it back.
242 	 */
243 	s = splbio();
244 	bp->b_flags &= ~(B_BUSY | B_PHYS | B_RAW);
245 	if (nobuf)
246 		putphysbuf(bp);
247 	else {
248 		/*
249 		 * [if another process is waiting for the raw I/O buffer,
250 		 *    wake up processes waiting to do physical I/O;
251 		 */
252 		if (bp->b_flags & B_WANTED) {
253 			bp->b_flags &= ~B_WANTED;
254 			wakeup(bp);
255 		}
256 	}
257 	splx(s);
258 
259 	return (error);
260 }
261 
262 /*
263  * allocate a buffer structure for use in physical I/O.
264  */
265 struct buf *
266 getphysbuf()
267 {
268 	struct buf *bp;
269 	int s;
270 
271 	s = splbio();
272 	bp = pool_get(&bufpool, PR_WAITOK);
273 	splx(s);
274 	memset(bp, 0, sizeof(*bp));
275 
276 	/* XXXCDC: is the following line necessary? */
277 	bp->b_vnbufs.le_next = NOLIST;
278 
279 	return(bp);
280 }
281 
282 /*
283  * get rid of a swap buffer structure which has been used in physical I/O.
284  */
285 void
286 putphysbuf(bp)
287         struct buf *bp;
288 {
289 	int s;
290 
291 	/* XXXCDC: is this necessary? */
292 	if (bp->b_vp)
293 		brelvp(bp);
294 
295 	if (__predict_false(bp->b_flags & B_WANTED))
296 		panic("putphysbuf: private buf B_WANTED");
297 	s = splbio();
298 	pool_put(&bufpool, bp);
299 	splx(s);
300 }
301 
302 /*
303  * Leffler, et al., says on p. 231:
304  * "The minphys() routine is called by physio() to adjust the
305  * size of each I/O transfer before the latter is passed to
306  * the strategy routine..."
307  *
308  * so, just adjust the buffer's count accounting to MAXPHYS here,
309  * and return the new count;
310  */
311 void
312 minphys(bp)
313 	struct buf *bp;
314 {
315 
316 	if (bp->b_bcount > MAXPHYS)
317 		bp->b_bcount = MAXPHYS;
318 }
319