xref: /openbsd-src/sys/kern/kern_physio.c (revision db3296cf5c1dd9058ceecc3a29fe4aaa0bd26000)
1 /*	$OpenBSD: kern_physio.c,v 1.21 2003/07/21 22:44:50 tedu Exp $	*/
2 /*	$NetBSD: kern_physio.c,v 1.28 1997/05/19 10:43:28 pk Exp $	*/
3 
4 /*-
5  * Copyright (c) 1994 Christopher G. Demetriou
6  * Copyright (c) 1982, 1986, 1990, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_physio.c	8.1 (Berkeley) 6/10/93
39  */
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/buf.h>
44 #include <sys/conf.h>
45 #include <sys/proc.h>
46 #include <sys/pool.h>
47 
48 #include <uvm/uvm_extern.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(void);
62 void putphysbuf(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 	void (*strategy)(struct buf *);
73 	struct buf *bp;
74 	dev_t dev;
75 	int flags;
76 	void (*minphys)(struct buf *);
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 	/* Make sure we have a buffer, creating one if necessary. */
87 	if ((nobuf = (bp == NULL)) != 0)
88 		bp = getphysbuf();
89 
90 	/* [raise the processor priority level to splbio;] */
91 	s = splbio();
92 
93 	/* [while the buffer is marked busy] */
94 	while (bp->b_flags & B_BUSY) {
95 		/* [mark the buffer wanted] */
96 		bp->b_flags |= B_WANTED;
97 		/* [wait until the buffer is available] */
98 		tsleep(bp, PRIBIO+1, "physbuf", 0);
99 	}
100 
101 	/* Mark it busy, so nobody else will use it. */
102 	bp->b_flags |= B_BUSY;
103 
104 	/* [lower the priority level] */
105 	splx(s);
106 
107 	/* [set up the fixed part of the buffer for a transfer] */
108 	bp->b_dev = dev;
109 	bp->b_error = 0;
110 	bp->b_proc = p;
111 	LIST_INIT(&bp->b_dep);
112 
113 	/*
114 	 * [while there are data to transfer and no I/O error]
115 	 * Note that I/O errors are handled with a 'goto' at the bottom
116 	 * of the 'while' loop.
117 	 */
118 	for (i = 0; i < uio->uio_iovcnt; i++) {
119 		iovp = &uio->uio_iov[i];
120 		while (iovp->iov_len > 0) {
121 			/*
122 			 * [mark the buffer busy for physical I/O]
123 			 * (i.e. set B_PHYS (because it's an I/O to user
124 			 * memory, and B_RAW, because B_RAW is to be
125 			 * "Set by physio for raw transfers.", in addition
126 			 * to the "busy" and read/write flag.)
127 			 */
128 			bp->b_flags = B_BUSY | B_PHYS | B_RAW | flags;
129 
130 			/* [set up the buffer for a maximum-sized transfer] */
131 			bp->b_blkno = btodb(uio->uio_offset);
132 			bp->b_bcount = iovp->iov_len;
133 			bp->b_data = iovp->iov_base;
134 
135 			/*
136 			 * [call minphys to bound the tranfer size]
137 			 * and remember the amount of data to transfer,
138 			 * for later comparison.
139 			 */
140 			(*minphys)(bp);
141 			todo = bp->b_bcount;
142 #ifdef DIAGNOSTIC
143 			if (todo < 0)
144 				panic("todo < 0; minphys broken");
145 			if (todo > MAXPHYS)
146 				panic("todo > MAXPHYS; minphys broken");
147 #endif
148 
149 			/*
150 			 * [lock the part of the user address space involved
151 			 *    in the transfer]
152 			 * Beware vmapbuf(); it clobbers b_data and
153 			 * saves it in b_saveaddr.  However, vunmapbuf()
154 			 * restores it.
155 			 */
156 			PHOLD(p);
157 			error = uvm_vslock(p, bp->b_data, todo,
158 			    (flags & B_READ) ?
159 			    VM_PROT_READ | VM_PROT_WRITE : VM_PROT_READ);
160 			if (error) {
161 				bp->b_flags |= B_ERROR;
162 				bp->b_error = error;
163 				goto after_unlock;
164 			}
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(bp, PRIBIO + 1, "physio", 0);
183 
184 			/* Mark it busy again, so nobody else will use it. */
185 			bp->b_flags |= B_BUSY;
186 
187 			/* [lower the priority level] */
188 			splx(s);
189 
190 			/*
191 			 * [unlock the part of the address space previously
192 			 *    locked]
193 			 */
194 			vunmapbuf(bp, todo);
195 			uvm_vsunlock(p, bp->b_data, todo);
196 after_unlock:
197 			PRELE(p);
198 
199 			/* remember error value (save a splbio/splx pair) */
200 			if (bp->b_flags & B_ERROR)
201 				error = (bp->b_error ? bp->b_error : EIO);
202 
203 			/*
204 			 * [deduct the transfer size from the total number
205 			 *    of data to transfer]
206 			 */
207 			done = bp->b_bcount - bp->b_resid;
208 #ifdef DIAGNOSTIC
209 			if (done < 0)
210 				panic("done < 0; strategy broken");
211 			if (done > todo)
212 				panic("done > todo; strategy broken");
213 #endif
214 			iovp->iov_len -= done;
215 			iovp->iov_base = (caddr_t)iovp->iov_base + done;
216 			uio->uio_offset += done;
217 			uio->uio_resid -= done;
218 
219 			/*
220 			 * Now, check for an error.
221 			 * Also, handle weird end-of-disk semantics.
222 			 */
223 			if (error || done < todo)
224 				goto done;
225 		}
226 	}
227 
228 done:
229 	/*
230 	 * [clean up the state of the buffer]
231 	 * Remember if somebody wants it, so we can wake them up below.
232 	 * Also, if we had to steal it, give it back.
233 	 */
234 	s = splbio();
235 	bp->b_flags &= ~(B_BUSY | B_PHYS | B_RAW);
236 	if (nobuf)
237 		putphysbuf(bp);
238 	else {
239 		/*
240 		 * [if another process is waiting for the raw I/O buffer,
241 		 *    wake up processes waiting to do physical I/O;
242 		 */
243 		if (bp->b_flags & B_WANTED) {
244 			bp->b_flags &= ~B_WANTED;
245 			wakeup(bp);
246 		}
247 	}
248 	splx(s);
249 
250 	return (error);
251 }
252 
253 /*
254  * Get a swap buffer structure, for use in physical I/O.
255  * Mostly taken from /sys/vm/swap_pager.c, except that it no longer
256  * records buffer list-empty conditions, and sleeps at PRIBIO + 1,
257  * rather than PSWP + 1 (and on a different wchan).
258  */
259 struct buf *
260 getphysbuf()
261 {
262 	struct buf *bp;
263 
264 	bp = pool_get(&bufpool, PR_WAITOK);
265 	bzero(bp, sizeof(*bp));
266 
267 	/* XXXCDC: are the following two lines necessary? */
268 	bp->b_vnbufs.le_next = NOLIST;
269 
270 	return (bp);
271 }
272 
273 /*
274  * Get rid of a swap buffer structure which has been used in physical I/O.
275  * Mostly taken from /sys/vm/swap_pager.c, except that it now uses
276  * wakeup() rather than the VM-internal thread_wakeup(), and that the caller
277  * must mask disk interrupts, rather than putphysbuf() itself.
278  */
279 void
280 putphysbuf(bp)
281 	struct buf *bp;
282 {
283 	/* XXXCDC: is this necesary? */
284 	if (bp->b_vp)
285 		brelvp(bp);
286 
287 #ifdef DIAGNOSTIC
288 	if (bp->b_flags & B_WANTED)
289 		panic("putphysbuf: private buf B_WANTED");
290 #endif
291 	pool_put(&bufpool, bp);
292 }
293 
294 /*
295  * Leffler, et al., says on p. 231:
296  * "The minphys() routine is called by physio() to adjust the
297  * size of each I/O transfer before the latter is passed to
298  * the strategy routine..."
299  *
300  * so, just adjust the buffer's count accounting to MAXPHYS here,
301  * and return the new count;
302  */
303 void
304 minphys(bp)
305 	struct buf *bp;
306 {
307 
308 	if (bp->b_bcount > MAXPHYS)
309 		bp->b_bcount = MAXPHYS;
310 }
311