xref: /openbsd-src/sys/kern/kern_physio.c (revision 3a3fbb3f2e2521ab7c4a56b7ff7462ebd9095ec5)
1 /*	$OpenBSD: kern_physio.c,v 1.18 2001/12/10 17:37:51 art 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. All advertising materials mentioning features or use of this software
23  *    must display the following acknowledgement:
24  *	This product includes software developed by the University of
25  *	California, Berkeley and its contributors.
26  * 4. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	@(#)kern_physio.c	8.1 (Berkeley) 6/10/93
43  */
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/buf.h>
48 #include <sys/conf.h>
49 #include <sys/proc.h>
50 #include <sys/pool.h>
51 
52 #include <uvm/uvm_extern.h>
53 
54 /*
55  * The routines implemented in this file are described in:
56  *	Leffler, et al.: The Design and Implementation of the 4.3BSD
57  *	    UNIX Operating System (Addison Welley, 1989)
58  * on pages 231-233.
59  *
60  * The routines "getphysbuf" and "putphysbuf" steal and return a swap
61  * buffer.  Leffler, et al., says that swap buffers are used to do the
62  * I/O, so raw I/O requests don't have to be single-threaded.
63  */
64 
65 struct buf *getphysbuf __P((void));
66 void putphysbuf __P((struct buf *bp));
67 
68 /*
69  * Do "physical I/O" on behalf of a user.  "Physical I/O" is I/O directly
70  * from the raw device to user buffers, and bypasses the buffer cache.
71  *
72  * Comments in brackets are from Leffler, et al.'s pseudo-code implementation.
73  */
74 int
75 physio(strategy, bp, dev, flags, minphys, uio)
76 	void (*strategy) __P((struct buf *));
77 	struct buf *bp;
78 	dev_t dev;
79 	int flags;
80 	void (*minphys) __P((struct buf *));
81 	struct uio *uio;
82 {
83 	struct iovec *iovp;
84 	struct proc *p = curproc;
85 	int error, done, i, nobuf, s, todo;
86 
87 	error = 0;
88 	flags &= B_READ | B_WRITE;
89 
90 	/* Make sure we have a buffer, creating one if necessary. */
91 	if ((nobuf = (bp == NULL)) != 0)
92 		bp = getphysbuf();
93 
94 	/* [raise the processor priority level to splbio;] */
95 	s = splbio();
96 
97 	/* [while the buffer is marked busy] */
98 	while (bp->b_flags & B_BUSY) {
99 		/* [mark the buffer wanted] */
100 		bp->b_flags |= B_WANTED;
101 		/* [wait until the buffer is available] */
102 		tsleep((caddr_t)bp, PRIBIO+1, "physbuf", 0);
103 	}
104 
105 	/* Mark it busy, so nobody else will use it. */
106 	bp->b_flags |= B_BUSY;
107 
108 	/* [lower the priority level] */
109 	splx(s);
110 
111 	/* [set up the fixed part of the buffer for a transfer] */
112 	bp->b_dev = dev;
113 	bp->b_error = 0;
114 	bp->b_proc = p;
115 	LIST_INIT(&bp->b_dep);
116 
117 	/*
118 	 * [while there are data to transfer and no I/O error]
119 	 * Note that I/O errors are handled with a 'goto' at the bottom
120 	 * of the 'while' loop.
121 	 */
122 	for (i = 0; i < uio->uio_iovcnt; i++) {
123 		iovp = &uio->uio_iov[i];
124 		while (iovp->iov_len > 0) {
125 			/*
126 			 * [mark the buffer busy for physical I/O]
127 			 * (i.e. set B_PHYS (because it's an I/O to user
128 			 * memory, and B_RAW, because B_RAW is to be
129 			 * "Set by physio for raw transfers.", in addition
130 			 * to the "busy" and read/write flag.)
131 			 */
132 			bp->b_flags = B_BUSY | B_PHYS | B_RAW | flags;
133 
134 			/* [set up the buffer for a maximum-sized transfer] */
135 			bp->b_blkno = btodb(uio->uio_offset);
136 			bp->b_bcount = iovp->iov_len;
137 			bp->b_data = iovp->iov_base;
138 
139 			/*
140 			 * [call minphys to bound the tranfer size]
141 			 * and remember the amount of data to transfer,
142 			 * for later comparison.
143 			 */
144 			(*minphys)(bp);
145 			todo = bp->b_bcount;
146 #ifdef DIAGNOSTIC
147 			if (todo < 0)
148 				panic("todo < 0; minphys broken");
149 			if (todo > MAXPHYS)
150 				panic("todo > MAXPHYS; minphys broken");
151 #endif
152 
153 			/*
154 			 * [lock the part of the user address space involved
155 			 *    in the transfer]
156 			 * Beware vmapbuf(); it clobbers b_data and
157 			 * saves it in b_saveaddr.  However, vunmapbuf()
158 			 * restores it.
159 			 */
160 			PHOLD(p);
161 			error = uvm_vslock(p, bp->b_data, todo,
162 			    (flags & B_READ) ?
163 			    VM_PROT_READ | VM_PROT_WRITE : VM_PROT_READ);
164 			if (error) {
165 				bp->b_flags |= B_ERROR;
166 				bp->b_error = error;
167 				goto after_unlock;
168 			}
169 			vmapbuf(bp, todo);
170 
171 			/* [call strategy to start the transfer] */
172 			(*strategy)(bp);
173 
174 			/*
175 			 * Note that the raise/wait/lower/get error
176 			 * steps below would be done by biowait(), but
177 			 * we want to unlock the address space before
178 			 * we lower the priority.
179 			 *
180 			 * [raise the priority level to splbio]
181 			 */
182 			s = splbio();
183 
184 			/* [wait for the transfer to complete] */
185 			while ((bp->b_flags & B_DONE) == 0)
186 				tsleep((caddr_t) bp, PRIBIO + 1, "physio", 0);
187 
188 			/* Mark it busy again, so nobody else will use it. */
189 			bp->b_flags |= B_BUSY;
190 
191 			/* [lower the priority level] */
192 			splx(s);
193 
194 			/*
195 			 * [unlock the part of the address space previously
196 			 *    locked]
197 			 */
198 			vunmapbuf(bp, todo);
199 			uvm_vsunlock(p, bp->b_data, todo);
200 after_unlock:
201 			PRELE(p);
202 
203 			/* remember error value (save a splbio/splx pair) */
204 			if (bp->b_flags & B_ERROR)
205 				error = (bp->b_error ? bp->b_error : EIO);
206 
207 			/*
208 			 * [deduct the transfer size from the total number
209 			 *    of data to transfer]
210 			 */
211 			done = bp->b_bcount - bp->b_resid;
212 #ifdef DIAGNOSTIC
213 			if (done < 0)
214 				panic("done < 0; strategy broken");
215 			if (done > todo)
216 				panic("done > todo; strategy broken");
217 #endif
218 			iovp->iov_len -= done;
219 			iovp->iov_base = (caddr_t)iovp->iov_base + done;
220 			uio->uio_offset += done;
221 			uio->uio_resid -= done;
222 
223 			/*
224 			 * Now, check for an error.
225 			 * Also, handle weird end-of-disk semantics.
226 			 */
227 			if (error || done < todo)
228 				goto done;
229 		}
230 	}
231 
232 done:
233 	/*
234 	 * [clean up the state of the buffer]
235 	 * Remember if somebody wants it, so we can wake them up below.
236 	 * Also, if we had to steal it, give it back.
237 	 */
238 	s = splbio();
239 	bp->b_flags &= ~(B_BUSY | B_PHYS | B_RAW);
240 	if (nobuf)
241 		putphysbuf(bp);
242 	else {
243 		/*
244 		 * [if another process is waiting for the raw I/O buffer,
245 		 *    wake up processes waiting to do physical I/O;
246 		 */
247 		if (bp->b_flags & B_WANTED) {
248 			bp->b_flags &= ~B_WANTED;
249 			wakeup(bp);
250 		}
251 	}
252 	splx(s);
253 
254 	return (error);
255 }
256 
257 /*
258  * Get a swap buffer structure, for use in physical I/O.
259  * Mostly taken from /sys/vm/swap_pager.c, except that it no longer
260  * records buffer list-empty conditions, and sleeps at PRIBIO + 1,
261  * rather than PSWP + 1 (and on a different wchan).
262  */
263 struct buf *
264 getphysbuf()
265 {
266 	struct buf *bp;
267 
268 	bp = pool_get(&bufpool, PR_WAITOK);
269 	bzero(bp, sizeof(*bp));
270 
271 	/* XXXCDC: are the following two lines necessary? */
272 	bp->b_vnbufs.le_next = NOLIST;
273 
274 	return (bp);
275 }
276 
277 /*
278  * Get rid of a swap buffer structure which has been used in physical I/O.
279  * Mostly taken from /sys/vm/swap_pager.c, except that it now uses
280  * wakeup() rather than the VM-internal thread_wakeup(), and that the caller
281  * must mask disk interrupts, rather than putphysbuf() itself.
282  */
283 void
284 putphysbuf(bp)
285 	struct buf *bp;
286 {
287 	/* XXXCDC: is this necesary? */
288 	if (bp->b_vp)
289 		brelvp(bp);
290 
291 #ifdef DIAGNOSTIC
292 	if (bp->b_flags & B_WANTED)
293 		panic("putphysbuf: private buf B_WANTED");
294 #endif
295 	pool_put(&bufpool, bp);
296 }
297 
298 /*
299  * Leffler, et al., says on p. 231:
300  * "The minphys() routine is called by physio() to adjust the
301  * size of each I/O transfer before the latter is passed to
302  * the strategy routine..."
303  *
304  * so, just adjust the buffer's count accounting to MAXPHYS here,
305  * and return the new count;
306  */
307 void
308 minphys(bp)
309 	struct buf *bp;
310 {
311 
312 	if (bp->b_bcount > MAXPHYS)
313 		bp->b_bcount = MAXPHYS;
314 }
315