xref: /openbsd-src/sys/kern/kern_physio.c (revision 4c1e55dc91edd6e69ccc60ce855900fbc12cf34f)
1 /*	$OpenBSD: kern_physio.c,v 1.39 2011/07/18 02:49:20 matthew 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 
57 /*
58  * Do "physical I/O" on behalf of a user.  "Physical I/O" is I/O directly
59  * from the raw device to user buffers, and bypasses the buffer cache.
60  *
61  * Comments in brackets are from Leffler, et al.'s pseudo-code implementation.
62  */
63 int
64 physio(void (*strategy)(struct buf *), dev_t dev, int flags,
65     void (*minphys)(struct buf *), struct uio *uio)
66 {
67 	struct iovec *iovp;
68 	struct proc *p = curproc;
69 	int error, done, i, s, todo;
70 	struct buf *bp;
71 
72 	if ((uio->uio_offset % DEV_BSIZE) != 0)
73 		return (EINVAL);
74 
75 	error = 0;
76 	flags &= B_READ | B_WRITE;
77 
78 	/* Create a buffer. */
79 	s = splbio();
80 	bp = pool_get(&bufpool, PR_WAITOK | PR_ZERO);
81 
82 	/* [set up the fixed part of the buffer for a transfer] */
83 	bp->b_vnbufs.le_next = NOLIST;
84 	bp->b_dev = dev;
85 	bp->b_error = 0;
86 	bp->b_proc = p;
87 	bp->b_flags = B_BUSY;
88 	LIST_INIT(&bp->b_dep);
89 	splx(s);
90 
91 	/*
92 	 * [while there are data to transfer and no I/O error]
93 	 * Note that I/O errors are handled with a 'goto' at the bottom
94 	 * of the 'while' loop.
95 	 */
96 	for (i = 0; i < uio->uio_iovcnt; i++) {
97 		iovp = &uio->uio_iov[i];
98 		while (iovp->iov_len > 0) {
99 			void *map = NULL;
100 
101 			/*
102 			 * [mark the buffer busy for physical I/O]
103 			 * (i.e. set B_PHYS (because it's an I/O to user
104 			 * memory), and B_RAW, because B_RAW is to be
105 			 * "Set by physio for raw transfers.", in addition
106 			 * to the "busy" and read/write flag.)
107 			 */
108 			CLR(bp->b_flags, B_DONE | B_ERROR);
109 			bp->b_flags |= (B_BUSY | B_PHYS | B_RAW | flags);
110 
111 			/* [set up the buffer for a maximum-sized transfer] */
112 			bp->b_blkno = btodb(uio->uio_offset);
113 
114 			/*
115 			 * Because iov_len is unsigned but b_bcount is signed,
116 			 * an overflow is possible. Therefore bound to MAXPHYS
117 			 * before calling minphys.
118 			 */
119 			if (iovp->iov_len > MAXPHYS)
120 				bp->b_bcount = MAXPHYS;
121 			else
122 				bp->b_bcount = iovp->iov_len;
123 
124 			/*
125 			 * [call minphys to bound the transfer size]
126 			 * and remember the amount of data to transfer,
127 			 * for later comparison.
128 			 */
129 			(*minphys)(bp);
130 			todo = bp->b_bcount;
131 #ifdef DIAGNOSTIC
132 			if (todo < 0)
133 				panic("todo < 0; minphys broken");
134 			if (todo > MAXPHYS)
135 				panic("todo > MAXPHYS; minphys broken");
136 #endif
137 
138 			/*
139 			 * [lock the part of the user address space involved
140 			 *    in the transfer]
141 			 * Beware vmapbuf(); it clobbers b_data and
142 			 * saves it in b_saveaddr.  However, vunmapbuf()
143 			 * restores it.
144 			 */
145 			error = uvm_vslock_device(p, iovp->iov_base, todo,
146 			    (flags & B_READ) ?
147 			    VM_PROT_READ | VM_PROT_WRITE : VM_PROT_READ, &map);
148 			if (error)
149 				goto done;
150 			if (map) {
151 				bp->b_data = map;
152 			} else {
153 				bp->b_data = iovp->iov_base;
154 				vmapbuf(bp, todo);
155 			}
156 
157 			/* [call strategy to start the transfer] */
158 			(*strategy)(bp);
159 
160 			/*
161 			 * Note that the raise/wait/lower/get error
162 			 * steps below would be done by biowait(), but
163 			 * we want to unlock the address space before
164 			 * we lower the priority.
165 			 *
166 			 * [raise the priority level to splbio]
167 			 */
168 			s = splbio();
169 
170 			/* [wait for the transfer to complete] */
171 			while ((bp->b_flags & B_DONE) == 0)
172 				tsleep(bp, PRIBIO + 1, "physio", 0);
173 
174 			/* Mark it busy again, so nobody else will use it. */
175 			bp->b_flags |= B_BUSY;
176 
177 			/* [lower the priority level] */
178 			splx(s);
179 
180 			/*
181 			 * [unlock the part of the address space previously
182 			 *    locked]
183 			 */
184 			if (!map)
185 				vunmapbuf(bp, todo);
186 			uvm_vsunlock_device(p, iovp->iov_base, todo, map);
187 
188 			/* remember error value (save a splbio/splx pair) */
189 			if (bp->b_flags & B_ERROR)
190 				error = (bp->b_error ? bp->b_error : EIO);
191 
192 			/*
193 			 * [deduct the transfer size from the total number
194 			 *    of data to transfer]
195 			 */
196 			done = bp->b_bcount - bp->b_resid;
197 #ifdef DIAGNOSTIC
198 			if (done < 0)
199 				panic("done < 0; strategy broken");
200 			if (done > todo)
201 				panic("done > todo; strategy broken");
202 #endif
203 			iovp->iov_len -= done;
204 			iovp->iov_base = (caddr_t)iovp->iov_base + done;
205 			uio->uio_offset += done;
206 			uio->uio_resid -= done;
207 
208 			/*
209 			 * Now, check for an error.
210 			 * Also, handle weird end-of-disk semantics.
211 			 */
212 			if (error || done < todo)
213 				goto done;
214 		}
215 	}
216 
217 done:
218 	/*
219 	 * [clean up the state of the buffer]
220 	 */
221 	s = splbio();
222 	/* XXXCDC: is this necessary? */
223 	if (bp->b_vp)
224 		brelvp(bp);
225 	splx(s);
226 	pool_put(&bufpool, bp);
227 
228 	return (error);
229 }
230 
231 /*
232  * Leffler, et al., says on p. 231:
233  * "The minphys() routine is called by physio() to adjust the
234  * size of each I/O transfer before the latter is passed to
235  * the strategy routine..."
236  *
237  * so, just adjust the buffer's count accounting to MAXPHYS here,
238  * and return the new count;
239  */
240 void
241 minphys(struct buf *bp)
242 {
243 
244 	if (bp->b_bcount > MAXPHYS)
245 		bp->b_bcount = MAXPHYS;
246 }
247