xref: /openbsd-src/sys/kern/kern_physio.c (revision 81fb472f136e03be4db5acdf90d9ac9fd7ec035c)
1 /*	$OpenBSD: kern_physio.c,v 1.49 2024/02/03 18:51:58 beck 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
physio(void (* strategy)(struct buf *),dev_t dev,int flags,void (* minphys)(struct buf *),struct uio * uio)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 	splx(s);
88 
89 	/*
90 	 * [while there are data to transfer and no I/O error]
91 	 * Note that I/O errors are handled with a 'goto' at the bottom
92 	 * of the 'while' loop.
93 	 */
94 	for (i = 0; i < uio->uio_iovcnt; i++) {
95 		iovp = &uio->uio_iov[i];
96 		while (iovp->iov_len > 0) {
97 			void *map = NULL;
98 
99 			/*
100 			 * [mark the buffer busy for physical I/O]
101 			 * (i.e. set B_PHYS (because it's an I/O to user
102 			 * memory), and B_RAW, because B_RAW is to be
103 			 * "Set by physio for raw transfers.", in addition
104 			 * to the "busy" and read/write flag.)
105 			 */
106 			CLR(bp->b_flags, B_DONE | B_ERROR);
107 			bp->b_flags |= (B_BUSY | B_PHYS | B_RAW | flags);
108 
109 			/* [set up the buffer for a maximum-sized transfer] */
110 			bp->b_blkno = btodb(uio->uio_offset);
111 
112 			/*
113 			 * Because iov_len is size_t (unsigned) but b_bcount is
114 			 * long (signed), an overflow is possible. Therefore
115 			 * limit b_bcount to LONG_MAX before calling the
116 			 * provided minphys.
117 			 */
118 			if (iovp->iov_len > LONG_MAX)
119 				bp->b_bcount = LONG_MAX;
120 			else
121 				bp->b_bcount = iovp->iov_len;
122 
123 			/*
124 			 * [call minphys to bound the transfer size]
125 			 * and remember the amount of data to transfer,
126 			 * for later comparison.
127 			 */
128 			(*minphys)(bp);
129 			todo = bp->b_bcount;
130 			KASSERTMSG(todo >= 0, "minphys broken");
131 
132 			/*
133 			 * [lock the part of the user address space involved
134 			 *    in the transfer]
135 			 * Beware vmapbuf(); it clobbers b_data and
136 			 * saves it in b_saveaddr.  However, vunmapbuf()
137 			 * restores it.
138 			 */
139 			error = uvm_vslock_device(p, iovp->iov_base, todo,
140 			    (flags & B_READ) ?
141 			    PROT_READ | PROT_WRITE : PROT_READ, &map);
142 			if (error)
143 				goto done;
144 			if (map) {
145 				bp->b_data = map;
146 			} else {
147 				bp->b_data = iovp->iov_base;
148 				vmapbuf(bp, todo);
149 			}
150 
151 			/* [call strategy to start the transfer] */
152 			(*strategy)(bp);
153 
154 			/*
155 			 * Note that the raise/wait/lower/get error
156 			 * steps below would be done by biowait(), but
157 			 * we want to unlock the address space before
158 			 * we lower the priority.
159 			 *
160 			 * [raise the priority level to splbio]
161 			 */
162 			s = splbio();
163 
164 			/* [wait for the transfer to complete] */
165 			while ((bp->b_flags & B_DONE) == 0)
166 				tsleep_nsec(bp, PRIBIO + 1, "physio", INFSLP);
167 
168 			/* Mark it busy again, so nobody else will use it. */
169 			bp->b_flags |= B_BUSY;
170 
171 			/* [lower the priority level] */
172 			splx(s);
173 
174 			/*
175 			 * [unlock the part of the address space previously
176 			 *    locked]
177 			 */
178 			if (!map)
179 				vunmapbuf(bp, todo);
180 			uvm_vsunlock_device(p, iovp->iov_base, todo, map);
181 
182 			/* remember error value (save a splbio/splx pair) */
183 			if (bp->b_flags & B_ERROR)
184 				error = (bp->b_error ? bp->b_error : EIO);
185 
186 			/*
187 			 * [deduct the transfer size from the total number
188 			 *    of data to transfer]
189 			 */
190 			KASSERTMSG(bp->b_resid <= LONG_MAX, "strategy broken");
191 			done = bp->b_bcount - bp->b_resid;
192 			KASSERTMSG(done >= 0, "strategy broken");
193 			KASSERTMSG(done <= todo, "strategy broken");
194 			iovp->iov_len -= done;
195 			iovp->iov_base = (caddr_t)iovp->iov_base + done;
196 			uio->uio_offset += done;
197 			uio->uio_resid -= done;
198 
199 			/*
200 			 * Now, check for an error.
201 			 * Also, handle weird end-of-disk semantics.
202 			 */
203 			if (error || done < todo)
204 				goto done;
205 		}
206 	}
207 
208 done:
209 	/*
210 	 * [clean up the state of the buffer]
211 	 */
212 	s = splbio();
213 	/* XXXCDC: is this necessary? */
214 	if (bp->b_vp)
215 		brelvp(bp);
216 	splx(s);
217 	pool_put(&bufpool, bp);
218 
219 	return (error);
220 }
221 
222 /*
223  * Leffler, et al., says on p. 231:
224  * "The minphys() routine is called by physio() to adjust the
225  * size of each I/O transfer before the latter is passed to
226  * the strategy routine..."
227  *
228  * so, just adjust the buffer's count accounting to MAXPHYS here,
229  * and return the new count;
230  */
231 void
minphys(struct buf * bp)232 minphys(struct buf *bp)
233 {
234 
235 	if (bp->b_bcount > MAXPHYS)
236 		bp->b_bcount = MAXPHYS;
237 }
238