xref: /netbsd-src/sys/kern/kern_physio.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: kern_physio.c,v 1.60 2004/03/23 13:22:33 junyoung Exp $	*/
2 
3 /*-
4  * Copyright (c) 1982, 1986, 1990, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_physio.c	8.1 (Berkeley) 6/10/93
37  */
38 
39 /*-
40  * Copyright (c) 1994 Christopher G. Demetriou
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. All advertising materials mentioning features or use of this software
51  *    must display the following acknowledgement:
52  *	This product includes software developed by the University of
53  *	California, Berkeley and its contributors.
54  * 4. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  *	@(#)kern_physio.c	8.1 (Berkeley) 6/10/93
71  */
72 
73 #include <sys/cdefs.h>
74 __KERNEL_RCSID(0, "$NetBSD: kern_physio.c,v 1.60 2004/03/23 13:22:33 junyoung Exp $");
75 
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/buf.h>
79 #include <sys/malloc.h>
80 #include <sys/proc.h>
81 
82 #include <uvm/uvm_extern.h>
83 
84 /*
85  * The routines implemented in this file are described in:
86  *	Leffler, et al.: The Design and Implementation of the 4.3BSD
87  *	    UNIX Operating System (Addison Welley, 1989)
88  * on pages 231-233.
89  *
90  * The routines "getphysbuf" and "putphysbuf" steal and return a swap
91  * buffer.  Leffler, et al., says that swap buffers are used to do the
92  * I/O, so raw I/O requests don't have to be single-threaded.
93  */
94 
95 struct buf *getphysbuf(void);
96 void putphysbuf(struct buf *bp);
97 
98 /*
99  * Do "physical I/O" on behalf of a user.  "Physical I/O" is I/O directly
100  * from the raw device to user buffers, and bypasses the buffer cache.
101  *
102  * Comments in brackets are from Leffler, et al.'s pseudo-code implementation.
103  */
104 int
105 physio(strategy, bp, dev, flags, min_phys, uio)
106 	void (*strategy)(struct buf *);
107 	struct buf *bp;
108 	dev_t dev;
109 	int flags;
110 	void (*min_phys)(struct buf *);
111 	struct uio *uio;
112 {
113 	struct iovec *iovp;
114 	struct lwp *l = curlwp;
115 	struct proc *p = l->l_proc;
116 	int error, done, i, nobuf, s;
117 	long todo;
118 
119 	error = 0;
120 	flags &= B_READ | B_WRITE;
121 
122 	/* Make sure we have a buffer, creating one if necessary. */
123 	if ((nobuf = (bp == NULL)) != 0) {
124 
125 		bp = getphysbuf();
126 		/* bp was just malloc'd so can't already be busy */
127 		bp->b_flags |= B_BUSY;
128 
129 	} else {
130 
131 		/* [raise the processor priority level to splbio;] */
132 		s = splbio();
133 
134 		/* [while the buffer is marked busy] */
135 		while (bp->b_flags & B_BUSY) {
136 			/* [mark the buffer wanted] */
137 			bp->b_flags |= B_WANTED;
138 			/* [wait until the buffer is available] */
139 			tsleep((caddr_t)bp, PRIBIO+1, "physbuf", 0);
140 		}
141 
142 		/* Mark it busy, so nobody else will use it. */
143 		bp->b_flags |= B_BUSY;
144 
145 		/* [lower the priority level] */
146 		splx(s);
147 	}
148 
149 	/* [set up the fixed part of the buffer for a transfer] */
150 	bp->b_dev = dev;
151 	bp->b_error = 0;
152 	bp->b_proc = p;
153 	LIST_INIT(&bp->b_dep);
154 
155 	/*
156 	 * [while there are data to transfer and no I/O error]
157 	 * Note that I/O errors are handled with a 'goto' at the bottom
158 	 * of the 'while' loop.
159 	 */
160 	for (i = 0; i < uio->uio_iovcnt; i++) {
161 		iovp = &uio->uio_iov[i];
162 		while (iovp->iov_len > 0) {
163 
164 			/*
165 			 * [mark the buffer busy for physical I/O]
166 			 * (i.e. set B_PHYS (because it's an I/O to user
167 			 * memory, and B_RAW, because B_RAW is to be
168 			 * "Set by physio for raw transfers.", in addition
169 			 * to the "busy" and read/write flag.)
170 			 */
171 			bp->b_flags = B_BUSY | B_PHYS | B_RAW | flags;
172 
173 			/* [set up the buffer for a maximum-sized transfer] */
174 			bp->b_blkno = btodb(uio->uio_offset);
175 			bp->b_bcount = iovp->iov_len;
176 			bp->b_data = iovp->iov_base;
177 
178 			/*
179 			 * [call minphys to bound the transfer size]
180 			 * and remember the amount of data to transfer,
181 			 * for later comparison.
182 			 */
183 			(*min_phys)(bp);
184 			todo = bp->b_bcount;
185 #ifdef DIAGNOSTIC
186 			if (todo <= 0)
187 				panic("todo(%ld) <= 0; minphys broken", todo);
188 			if (todo > MAXPHYS)
189 				panic("todo(%ld) > MAXPHYS; minphys broken",
190 				      todo);
191 #endif
192 
193 			/*
194 			 * [lock the part of the user address space involved
195 			 *    in the transfer]
196 			 * Beware vmapbuf(); it clobbers b_data and
197 			 * saves it in b_saveaddr.  However, vunmapbuf()
198 			 * restores it.
199 			 */
200 			PHOLD(l);
201 			error = uvm_vslock(p, bp->b_data, todo,
202 					   (flags & B_READ) ?
203 					   VM_PROT_WRITE : VM_PROT_READ);
204 			if (error) {
205 				bp->b_flags |= B_ERROR;
206 				bp->b_error = error;
207 				goto after_vsunlock;
208 			}
209 			vmapbuf(bp, todo);
210 
211 			BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
212 
213 			/* [call strategy to start the transfer] */
214 			(*strategy)(bp);
215 
216 			/*
217 			 * Note that the raise/wait/lower/get error
218 			 * steps below would be done by biowait(), but
219 			 * we want to unlock the address space before
220 			 * we lower the priority.
221 			 *
222 			 * [raise the priority level to splbio]
223 			 */
224 			s = splbio();
225 
226 			/* [wait for the transfer to complete] */
227 			while ((bp->b_flags & B_DONE) == 0)
228 				tsleep((caddr_t) bp, PRIBIO + 1, "physio", 0);
229 
230 			/* Mark it busy again, so nobody else will use it. */
231 			bp->b_flags |= B_BUSY;
232 
233 			/* [lower the priority level] */
234 			splx(s);
235 
236 			/*
237 			 * [unlock the part of the address space previously
238 			 *    locked]
239 			 */
240 			vunmapbuf(bp, todo);
241 			uvm_vsunlock(p, bp->b_data, todo);
242  after_vsunlock:
243 			PRELE(l);
244 
245 			/* remember error value (save a splbio/splx pair) */
246 			if (bp->b_flags & B_ERROR)
247 				error = (bp->b_error ? bp->b_error : EIO);
248 
249 			/*
250 			 * [deduct the transfer size from the total number
251 			 *    of data to transfer]
252 			 */
253 			done = bp->b_bcount - bp->b_resid;
254 			KASSERT(done >= 0);
255 			KASSERT(done <= todo);
256 
257 			iovp->iov_len -= done;
258 			iovp->iov_base = (caddr_t)iovp->iov_base + done;
259 			uio->uio_offset += done;
260 			uio->uio_resid -= done;
261 
262 			/*
263 			 * Now, check for an error.
264 			 * Also, handle weird end-of-disk semantics.
265 			 */
266 			if (error || done < todo)
267 				goto done;
268 		}
269 	}
270 
271 done:
272 	/*
273 	 * [clean up the state of the buffer]
274 	 * Remember if somebody wants it, so we can wake them up below.
275 	 * Also, if we had to steal it, give it back.
276 	 */
277 	s = splbio();
278 	bp->b_flags &= ~(B_BUSY | B_PHYS | B_RAW);
279 	if (nobuf)
280 		putphysbuf(bp);
281 	else {
282 		/*
283 		 * [if another process is waiting for the raw I/O buffer,
284 		 *    wake up processes waiting to do physical I/O;
285 		 */
286 		if (bp->b_flags & B_WANTED) {
287 			bp->b_flags &= ~B_WANTED;
288 			wakeup(bp);
289 		}
290 	}
291 	splx(s);
292 
293 	return (error);
294 }
295 
296 /*
297  * allocate a buffer structure for use in physical I/O.
298  */
299 struct buf *
300 getphysbuf()
301 {
302 	struct buf *bp;
303 	int s;
304 
305 	s = splbio();
306 	bp = pool_get(&bufpool, PR_WAITOK);
307 	splx(s);
308 	memset(bp, 0, sizeof(*bp));
309 	BUF_INIT(bp);
310 	return(bp);
311 }
312 
313 /*
314  * get rid of a swap buffer structure which has been used in physical I/O.
315  */
316 void
317 putphysbuf(bp)
318         struct buf *bp;
319 {
320 	int s;
321 
322 	if (__predict_false(bp->b_flags & B_WANTED))
323 		panic("putphysbuf: private buf B_WANTED");
324 	s = splbio();
325 	pool_put(&bufpool, bp);
326 	splx(s);
327 }
328 
329 /*
330  * Leffler, et al., says on p. 231:
331  * "The minphys() routine is called by physio() to adjust the
332  * size of each I/O transfer before the latter is passed to
333  * the strategy routine..."
334  *
335  * so, just adjust the buffer's count accounting to MAXPHYS here,
336  * and return the new count;
337  */
338 void
339 minphys(bp)
340 	struct buf *bp;
341 {
342 
343 	if (bp->b_bcount > MAXPHYS)
344 		bp->b_bcount = MAXPHYS;
345 }
346