1 /* $OpenBSD: kern_physio.c,v 1.28 2007/05/26 20:26:51 pedro 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(void (*strategy)(struct buf *), struct buf *bp, dev_t dev, int flags,
72 void (*minphys)(struct buf *), struct uio *uio)
73 {
74 struct iovec *iovp;
75 struct proc *p = curproc;
76 int error, done, i, nobuf, s, todo;
77
78 error = 0;
79 flags &= B_READ | B_WRITE;
80
81 /* Make sure we have a buffer, creating one if necessary. */
82 if ((nobuf = (bp == NULL)) != 0)
83 bp = getphysbuf();
84
85 /* [raise the processor priority level to splbio;] */
86 s = splbio();
87
88 /* [while the buffer is marked busy] */
89 while (bp->b_flags & B_BUSY) {
90 /* [mark the buffer wanted] */
91 bp->b_flags |= B_WANTED;
92 /* [wait until the buffer is available] */
93 tsleep(bp, PRIBIO+1, "physbuf", 0);
94 }
95
96 /* Mark it busy, so nobody else will use it. */
97 bp->b_flags |= B_BUSY;
98
99 /* [lower the priority level] */
100 splx(s);
101
102 /* [set up the fixed part of the buffer for a transfer] */
103 bp->b_dev = dev;
104 bp->b_error = 0;
105 bp->b_proc = p;
106 LIST_INIT(&bp->b_dep);
107
108 /*
109 * [while there are data to transfer and no I/O error]
110 * Note that I/O errors are handled with a 'goto' at the bottom
111 * of the 'while' loop.
112 */
113 for (i = 0; i < uio->uio_iovcnt; i++) {
114 iovp = &uio->uio_iov[i];
115 while (iovp->iov_len > 0) {
116 /*
117 * [mark the buffer busy for physical I/O]
118 * (i.e. set B_PHYS (because it's an I/O to user
119 * memory), and B_RAW, because B_RAW is to be
120 * "Set by physio for raw transfers.", in addition
121 * to the "busy" and read/write flag.)
122 */
123 bp->b_flags = B_BUSY | B_PHYS | B_RAW | flags;
124
125 /* [set up the buffer for a maximum-sized transfer] */
126 bp->b_blkno = btodb(uio->uio_offset);
127 bp->b_data = iovp->iov_base;
128
129 /*
130 * Because iov_len is unsigned but b_bcount is signed,
131 * an overflow is possible. Therefore bound to MAXPHYS
132 * before calling minphys.
133 */
134 if (iovp->iov_len > MAXPHYS)
135 bp->b_bcount = MAXPHYS;
136 else
137 bp->b_bcount = iovp->iov_len;
138
139 /*
140 * [call minphys to bound the transfer 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 error = uvm_vslock(p, bp->b_data, todo,
161 (flags & B_READ) ?
162 VM_PROT_READ | VM_PROT_WRITE : VM_PROT_READ);
163 if (error) {
164 bp->b_flags |= B_ERROR;
165 bp->b_error = error;
166 goto after_unlock;
167 }
168 vmapbuf(bp, todo);
169
170 /* [call strategy to start the transfer] */
171 (*strategy)(bp);
172
173 /*
174 * Note that the raise/wait/lower/get error
175 * steps below would be done by biowait(), but
176 * we want to unlock the address space before
177 * we lower the priority.
178 *
179 * [raise the priority level to splbio]
180 */
181 s = splbio();
182
183 /* [wait for the transfer to complete] */
184 while ((bp->b_flags & B_DONE) == 0)
185 tsleep(bp, PRIBIO + 1, "physio", 0);
186
187 /* Mark it busy again, so nobody else will use it. */
188 bp->b_flags |= B_BUSY;
189
190 /* [lower the priority level] */
191 splx(s);
192
193 /*
194 * [unlock the part of the address space previously
195 * locked]
196 */
197 vunmapbuf(bp, todo);
198 uvm_vsunlock(p, bp->b_data, todo);
199 after_unlock:
200
201 /* remember error value (save a splbio/splx pair) */
202 if (bp->b_flags & B_ERROR)
203 error = (bp->b_error ? bp->b_error : EIO);
204
205 /*
206 * [deduct the transfer size from the total number
207 * of data to transfer]
208 */
209 done = bp->b_bcount - bp->b_resid;
210 #ifdef DIAGNOSTIC
211 if (done < 0)
212 panic("done < 0; strategy broken");
213 if (done > todo)
214 panic("done > todo; strategy broken");
215 #endif
216 iovp->iov_len -= done;
217 iovp->iov_base = (caddr_t)iovp->iov_base + done;
218 uio->uio_offset += done;
219 uio->uio_resid -= done;
220
221 /*
222 * Now, check for an error.
223 * Also, handle weird end-of-disk semantics.
224 */
225 if (error || done < todo)
226 goto done;
227 }
228 }
229
230 done:
231 /*
232 * [clean up the state of the buffer]
233 * Remember if somebody wants it, so we can wake them up below.
234 * Also, if we had to steal it, give it back.
235 */
236 s = splbio();
237 bp->b_flags &= ~(B_BUSY | B_PHYS | B_RAW);
238 if (nobuf)
239 putphysbuf(bp);
240 else {
241 /*
242 * [if another process is waiting for the raw I/O buffer,
243 * wake up processes waiting to do physical I/O]
244 */
245 if (bp->b_flags & B_WANTED) {
246 bp->b_flags &= ~B_WANTED;
247 wakeup(bp);
248 }
249 }
250 splx(s);
251
252 return (error);
253 }
254
255 /*
256 * Get a swap buffer structure, for use in physical I/O.
257 * Mostly taken from /sys/vm/swap_pager.c, except that it no longer
258 * records buffer list-empty conditions, and sleeps at PRIBIO + 1,
259 * rather than PSWP + 1 (and on a different wchan).
260 */
261 struct buf *
262 getphysbuf(void)
263 {
264 struct buf *bp;
265 int s;
266
267 s = splbio();
268 bp = pool_get(&bufpool, PR_WAITOK);
269 splx(s);
270
271 bzero(bp, sizeof(*bp));
272
273 /* XXXCDC: are the following two lines necessary? */
274 bp->b_vnbufs.le_next = NOLIST;
275
276 return (bp);
277 }
278
279 /*
280 * Get rid of a swap buffer structure which has been used in physical I/O.
281 * Mostly taken from /sys/vm/swap_pager.c, except that it now uses
282 * wakeup() rather than the VM-internal thread_wakeup(), and that the caller
283 * must mask disk interrupts, rather than putphysbuf() itself.
284 */
285 void
286 putphysbuf(struct buf *bp)
287 {
288 splassert(IPL_BIO);
289
290 /* XXXCDC: is this necessary? */
291 if (bp->b_vp)
292 brelvp(bp);
293
294 #ifdef DIAGNOSTIC
295 if (bp->b_flags & B_WANTED)
296 panic("putphysbuf: private buf B_WANTED");
297 #endif
298 pool_put(&bufpool, bp);
299 }
300
301 /*
302 * Leffler, et al., says on p. 231:
303 * "The minphys() routine is called by physio() to adjust the
304 * size of each I/O transfer before the latter is passed to
305 * the strategy routine..."
306 *
307 * so, just adjust the buffer's count accounting to MAXPHYS here,
308 * and return the new count;
309 */
310 void
311 minphys(struct buf *bp)
312 {
313
314 if (bp->b_bcount > MAXPHYS)
315 bp->b_bcount = MAXPHYS;
316 }