xref: /netbsd-src/sys/kern/kern_physio.c (revision df0caa2637da0538ecdf6b878c4d08e684b43d8f)
1 /*	$NetBSD: kern_physio.c,v 1.61 2005/06/23 23:15:12 thorpej 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 1.61 2005/06/23 23:15:12 thorpej 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.  Of course,
93  * NetBSD doesn't use "swap buffers" -- we have our own memory pool for
94  * buffer descriptors.
95  */
96 
97 /*
98  * allocate a buffer structure for use in physical I/O.
99  */
100 static struct buf *
101 getphysbuf(void)
102 {
103 	struct buf *bp;
104 	int s;
105 
106 	s = splbio();
107 	bp = pool_get(&bufpool, PR_WAITOK);
108 	splx(s);
109 	memset(bp, 0, sizeof(*bp));
110 	BUF_INIT(bp);
111 	return(bp);
112 }
113 
114 /*
115  * get rid of a swap buffer structure which has been used in physical I/O.
116  */
117 static void
118 putphysbuf(struct buf *bp)
119 {
120 	int s;
121 
122 	if (__predict_false(bp->b_flags & B_WANTED))
123 		panic("putphysbuf: private buf B_WANTED");
124 	s = splbio();
125 	pool_put(&bufpool, bp);
126 	splx(s);
127 }
128 
129 /*
130  * Do "physical I/O" on behalf of a user.  "Physical I/O" is I/O directly
131  * from the raw device to user buffers, and bypasses the buffer cache.
132  *
133  * Comments in brackets are from Leffler, et al.'s pseudo-code implementation.
134  */
135 int
136 physio(void (*strategy)(struct buf *), struct buf *bp, dev_t dev, int flags,
137     void (*min_phys)(struct buf *), struct uio *uio)
138 {
139 	struct iovec *iovp;
140 	struct lwp *l = curlwp;
141 	struct proc *p = l->l_proc;
142 	int error, done, i, nobuf, s;
143 	long todo;
144 
145 	error = 0;
146 	flags &= B_READ | B_WRITE;
147 
148 	/* Make sure we have a buffer, creating one if necessary. */
149 	if ((nobuf = (bp == NULL)) != 0) {
150 
151 		bp = getphysbuf();
152 		/* bp was just malloc'd so can't already be busy */
153 		bp->b_flags |= B_BUSY;
154 
155 	} else {
156 
157 		/* [raise the processor priority level to splbio;] */
158 		s = splbio();
159 
160 		/* [while the buffer is marked busy] */
161 		while (bp->b_flags & B_BUSY) {
162 			/* [mark the buffer wanted] */
163 			bp->b_flags |= B_WANTED;
164 			/* [wait until the buffer is available] */
165 			tsleep((caddr_t)bp, PRIBIO+1, "physbuf", 0);
166 		}
167 
168 		/* Mark it busy, so nobody else will use it. */
169 		bp->b_flags |= B_BUSY;
170 
171 		/* [lower the priority level] */
172 		splx(s);
173 	}
174 
175 	/* [set up the fixed part of the buffer for a transfer] */
176 	bp->b_dev = dev;
177 	bp->b_error = 0;
178 	bp->b_proc = p;
179 	LIST_INIT(&bp->b_dep);
180 
181 	/*
182 	 * [while there are data to transfer and no I/O error]
183 	 * Note that I/O errors are handled with a 'goto' at the bottom
184 	 * of the 'while' loop.
185 	 */
186 	for (i = 0; i < uio->uio_iovcnt; i++) {
187 		iovp = &uio->uio_iov[i];
188 		while (iovp->iov_len > 0) {
189 
190 			/*
191 			 * [mark the buffer busy for physical I/O]
192 			 * (i.e. set B_PHYS (because it's an I/O to user
193 			 * memory, and B_RAW, because B_RAW is to be
194 			 * "Set by physio for raw transfers.", in addition
195 			 * to the "busy" and read/write flag.)
196 			 */
197 			bp->b_flags = B_BUSY | B_PHYS | B_RAW | flags;
198 
199 			/* [set up the buffer for a maximum-sized transfer] */
200 			bp->b_blkno = btodb(uio->uio_offset);
201 			bp->b_bcount = iovp->iov_len;
202 			bp->b_data = iovp->iov_base;
203 
204 			/*
205 			 * [call minphys to bound the transfer size]
206 			 * and remember the amount of data to transfer,
207 			 * for later comparison.
208 			 */
209 			(*min_phys)(bp);
210 			todo = bp->b_bcount;
211 #ifdef DIAGNOSTIC
212 			if (todo <= 0)
213 				panic("todo(%ld) <= 0; minphys broken", todo);
214 			if (todo > MAXPHYS)
215 				panic("todo(%ld) > MAXPHYS; minphys broken",
216 				      todo);
217 #endif
218 
219 			/*
220 			 * [lock the part of the user address space involved
221 			 *    in the transfer]
222 			 * Beware vmapbuf(); it clobbers b_data and
223 			 * saves it in b_saveaddr.  However, vunmapbuf()
224 			 * restores it.
225 			 */
226 			PHOLD(l);
227 			error = uvm_vslock(p, bp->b_data, todo,
228 					   (flags & B_READ) ?
229 					   VM_PROT_WRITE : VM_PROT_READ);
230 			if (error) {
231 				bp->b_flags |= B_ERROR;
232 				bp->b_error = error;
233 				goto after_vsunlock;
234 			}
235 			vmapbuf(bp, todo);
236 
237 			BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
238 
239 			/* [call strategy to start the transfer] */
240 			(*strategy)(bp);
241 
242 			/*
243 			 * Note that the raise/wait/lower/get error
244 			 * steps below would be done by biowait(), but
245 			 * we want to unlock the address space before
246 			 * we lower the priority.
247 			 *
248 			 * [raise the priority level to splbio]
249 			 */
250 			s = splbio();
251 
252 			/* [wait for the transfer to complete] */
253 			while ((bp->b_flags & B_DONE) == 0)
254 				tsleep((caddr_t) bp, PRIBIO + 1, "physio", 0);
255 
256 			/* Mark it busy again, so nobody else will use it. */
257 			bp->b_flags |= B_BUSY;
258 
259 			/* [lower the priority level] */
260 			splx(s);
261 
262 			/*
263 			 * [unlock the part of the address space previously
264 			 *    locked]
265 			 */
266 			vunmapbuf(bp, todo);
267 			uvm_vsunlock(p, bp->b_data, todo);
268  after_vsunlock:
269 			PRELE(l);
270 
271 			/* remember error value (save a splbio/splx pair) */
272 			if (bp->b_flags & B_ERROR)
273 				error = (bp->b_error ? bp->b_error : EIO);
274 
275 			/*
276 			 * [deduct the transfer size from the total number
277 			 *    of data to transfer]
278 			 */
279 			done = bp->b_bcount - bp->b_resid;
280 			KASSERT(done >= 0);
281 			KASSERT(done <= todo);
282 
283 			iovp->iov_len -= done;
284 			iovp->iov_base = (caddr_t)iovp->iov_base + done;
285 			uio->uio_offset += done;
286 			uio->uio_resid -= done;
287 
288 			/*
289 			 * Now, check for an error.
290 			 * Also, handle weird end-of-disk semantics.
291 			 */
292 			if (error || done < todo)
293 				goto done;
294 		}
295 	}
296 
297 done:
298 	/*
299 	 * [clean up the state of the buffer]
300 	 * Remember if somebody wants it, so we can wake them up below.
301 	 * Also, if we had to steal it, give it back.
302 	 */
303 	s = splbio();
304 	bp->b_flags &= ~(B_BUSY | B_PHYS | B_RAW);
305 	if (nobuf)
306 		putphysbuf(bp);
307 	else {
308 		/*
309 		 * [if another process is waiting for the raw I/O buffer,
310 		 *    wake up processes waiting to do physical I/O;
311 		 */
312 		if (bp->b_flags & B_WANTED) {
313 			bp->b_flags &= ~B_WANTED;
314 			wakeup(bp);
315 		}
316 	}
317 	splx(s);
318 
319 	return (error);
320 }
321 
322 /*
323  * Leffler, et al., says on p. 231:
324  * "The minphys() routine is called by physio() to adjust the
325  * size of each I/O transfer before the latter is passed to
326  * the strategy routine..."
327  *
328  * so, just adjust the buffer's count accounting to MAXPHYS here,
329  * and return the new count;
330  */
331 void
332 minphys(struct buf *bp)
333 {
334 
335 	if (bp->b_bcount > MAXPHYS)
336 		bp->b_bcount = MAXPHYS;
337 }
338