xref: /openbsd-src/sys/kern/kern_physio.c (revision f6aab3d83b51b91c24247ad2c2573574de475a82)
1 /*	$OpenBSD: kern_physio.c,v 1.47 2020/02/20 16:26:01 krw 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/pool.h>
45 
46 #include <uvm/uvm_extern.h>
47 
48 /*
49  * The routines implemented in this file are described in:
50  *	Leffler, et al.: The Design and Implementation of the 4.3BSD
51  *	    UNIX Operating System (Addison Welley, 1989)
52  * on pages 231-233.
53  */
54 
55 /*
56  * Do "physical I/O" on behalf of a user.  "Physical I/O" is I/O directly
57  * from the raw device to user buffers, and bypasses the buffer cache.
58  *
59  * Comments in brackets are from Leffler, et al.'s pseudo-code implementation.
60  */
61 int
62 physio(void (*strategy)(struct buf *), dev_t dev, int flags,
63     void (*minphys)(struct buf *), struct uio *uio)
64 {
65 	struct iovec *iovp;
66 	struct proc *p = curproc;
67 	long done, todo;
68 	int error, i, s;
69 	struct buf *bp;
70 
71 	if ((uio->uio_offset % DEV_BSIZE) != 0)
72 		return (EINVAL);
73 
74 	error = 0;
75 	flags &= B_READ | B_WRITE;
76 
77 	/* Create a buffer. */
78 	s = splbio();
79 	bp = pool_get(&bufpool, PR_WAITOK | PR_ZERO);
80 
81 	/* [set up the fixed part of the buffer for a transfer] */
82 	bp->b_vnbufs.le_next = NOLIST;
83 	bp->b_dev = dev;
84 	bp->b_error = 0;
85 	bp->b_proc = p;
86 	bp->b_flags = B_BUSY;
87 	LIST_INIT(&bp->b_dep);
88 	splx(s);
89 
90 	/*
91 	 * [while there are data to transfer and no I/O error]
92 	 * Note that I/O errors are handled with a 'goto' at the bottom
93 	 * of the 'while' loop.
94 	 */
95 	for (i = 0; i < uio->uio_iovcnt; i++) {
96 		iovp = &uio->uio_iov[i];
97 		while (iovp->iov_len > 0) {
98 			void *map = NULL;
99 
100 			/*
101 			 * [mark the buffer busy for physical I/O]
102 			 * (i.e. set B_PHYS (because it's an I/O to user
103 			 * memory), and B_RAW, because B_RAW is to be
104 			 * "Set by physio for raw transfers.", in addition
105 			 * to the "busy" and read/write flag.)
106 			 */
107 			CLR(bp->b_flags, B_DONE | B_ERROR);
108 			bp->b_flags |= (B_BUSY | B_PHYS | B_RAW | flags);
109 
110 			/* [set up the buffer for a maximum-sized transfer] */
111 			bp->b_blkno = btodb(uio->uio_offset);
112 
113 			/*
114 			 * Because iov_len is size_t (unsigned) but b_bcount is
115 			 * long (signed), an overflow is possible. Therefore
116 			 * limit b_bcount to LONG_MAX before calling the provided
117 			 * minphys.
118 			 */
119 			if (iovp->iov_len > LONG_MAX)
120 				bp->b_bcount = LONG_MAX;
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 			KASSERTMSG(todo >= 0, "minphys broken");
132 
133 			/*
134 			 * [lock the part of the user address space involved
135 			 *    in the transfer]
136 			 * Beware vmapbuf(); it clobbers b_data and
137 			 * saves it in b_saveaddr.  However, vunmapbuf()
138 			 * restores it.
139 			 */
140 			error = uvm_vslock_device(p, iovp->iov_base, todo,
141 			    (flags & B_READ) ?
142 			    PROT_READ | PROT_WRITE : PROT_READ, &map);
143 			if (error)
144 				goto done;
145 			if (map) {
146 				bp->b_data = map;
147 			} else {
148 				bp->b_data = iovp->iov_base;
149 				vmapbuf(bp, todo);
150 			}
151 
152 			/* [call strategy to start the transfer] */
153 			(*strategy)(bp);
154 
155 			/*
156 			 * Note that the raise/wait/lower/get error
157 			 * steps below would be done by biowait(), but
158 			 * we want to unlock the address space before
159 			 * we lower the priority.
160 			 *
161 			 * [raise the priority level to splbio]
162 			 */
163 			s = splbio();
164 
165 			/* [wait for the transfer to complete] */
166 			while ((bp->b_flags & B_DONE) == 0)
167 				tsleep_nsec(bp, PRIBIO + 1, "physio", INFSLP);
168 
169 			/* Mark it busy again, so nobody else will use it. */
170 			bp->b_flags |= B_BUSY;
171 
172 			/* [lower the priority level] */
173 			splx(s);
174 
175 			/*
176 			 * [unlock the part of the address space previously
177 			 *    locked]
178 			 */
179 			if (!map)
180 				vunmapbuf(bp, todo);
181 			uvm_vsunlock_device(p, iovp->iov_base, todo, map);
182 
183 			/* remember error value (save a splbio/splx pair) */
184 			if (bp->b_flags & B_ERROR)
185 				error = (bp->b_error ? bp->b_error : EIO);
186 
187 			/*
188 			 * [deduct the transfer size from the total number
189 			 *    of data to transfer]
190 			 */
191 			KASSERTMSG(bp->b_resid <= LONG_MAX, "strategy broken");
192 			done = bp->b_bcount - bp->b_resid;
193 			KASSERTMSG(done >= 0, "strategy broken");
194 			KASSERTMSG(done <= todo, "strategy broken");
195 			iovp->iov_len -= done;
196 			iovp->iov_base = (caddr_t)iovp->iov_base + done;
197 			uio->uio_offset += done;
198 			uio->uio_resid -= done;
199 
200 			/*
201 			 * Now, check for an error.
202 			 * Also, handle weird end-of-disk semantics.
203 			 */
204 			if (error || done < todo)
205 				goto done;
206 		}
207 	}
208 
209 done:
210 	/*
211 	 * [clean up the state of the buffer]
212 	 */
213 	s = splbio();
214 	/* XXXCDC: is this necessary? */
215 	if (bp->b_vp)
216 		brelvp(bp);
217 	splx(s);
218 	pool_put(&bufpool, bp);
219 
220 	return (error);
221 }
222 
223 /*
224  * Leffler, et al., says on p. 231:
225  * "The minphys() routine is called by physio() to adjust the
226  * size of each I/O transfer before the latter is passed to
227  * the strategy routine..."
228  *
229  * so, just adjust the buffer's count accounting to MAXPHYS here,
230  * and return the new count;
231  */
232 void
233 minphys(struct buf *bp)
234 {
235 
236 	if (bp->b_bcount > MAXPHYS)
237 		bp->b_bcount = MAXPHYS;
238 }
239