kern_physio.c revision 1.15
1/* $OpenBSD: kern_physio.c,v 1.15 2001/11/09 15:33:02 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 65struct buf *getphysbuf __P((void)); 66void 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 */ 74int 75physio(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 /* 91 * [check user read/write access to the data buffer] 92 * 93 * Check each iov one by one. Note that we know if we're reading or 94 * writing, so we ignore the uio's rw parameter. Also note that if 95 * we're doing a read, that's a *write* to user-space. 96 */ 97 if (uio->uio_segflg == UIO_USERSPACE) 98 for (i = 0; i < uio->uio_iovcnt; i++) 99 /* XXX - obsolete now that vslock can error? */ 100 if (!uvm_useracc(uio->uio_iov[i].iov_base, 101 uio->uio_iov[i].iov_len, 102 (flags == B_READ) ? B_WRITE : B_READ)) 103 return (EFAULT); 104 105 /* Make sure we have a buffer, creating one if necessary. */ 106 if ((nobuf = (bp == NULL)) != 0) 107 bp = getphysbuf(); 108 109 /* [raise the processor priority level to splbio;] */ 110 s = splbio(); 111 112 /* [while the buffer is marked busy] */ 113 while (bp->b_flags & B_BUSY) { 114 /* [mark the buffer wanted] */ 115 bp->b_flags |= B_WANTED; 116 /* [wait until the buffer is available] */ 117 tsleep((caddr_t)bp, PRIBIO+1, "physbuf", 0); 118 } 119 120 /* Mark it busy, so nobody else will use it. */ 121 bp->b_flags |= B_BUSY; 122 123 /* [lower the priority level] */ 124 splx(s); 125 126 /* [set up the fixed part of the buffer for a transfer] */ 127 bp->b_dev = dev; 128 bp->b_error = 0; 129 bp->b_proc = p; 130 LIST_INIT(&bp->b_dep); 131 132 /* 133 * [while there are data to transfer and no I/O error] 134 * Note that I/O errors are handled with a 'goto' at the bottom 135 * of the 'while' loop. 136 */ 137 for (i = 0; i < uio->uio_iovcnt; i++) { 138 iovp = &uio->uio_iov[i]; 139 while (iovp->iov_len > 0) { 140 /* 141 * [mark the buffer busy for physical I/O] 142 * (i.e. set B_PHYS (because it's an I/O to user 143 * memory, and B_RAW, because B_RAW is to be 144 * "Set by physio for raw transfers.", in addition 145 * to the "busy" and read/write flag.) 146 */ 147 bp->b_flags = B_BUSY | B_PHYS | B_RAW | flags; 148 149 /* [set up the buffer for a maximum-sized transfer] */ 150 bp->b_blkno = btodb(uio->uio_offset); 151 bp->b_bcount = iovp->iov_len; 152 bp->b_data = iovp->iov_base; 153 154 /* 155 * [call minphys to bound the tranfer size] 156 * and remember the amount of data to transfer, 157 * for later comparison. 158 */ 159 (*minphys)(bp); 160 todo = bp->b_bcount; 161#ifdef DIAGNOSTIC 162 if (todo < 0) 163 panic("todo < 0; minphys broken"); 164 if (todo > MAXPHYS) 165 panic("todo > MAXPHYS; minphys broken"); 166#endif 167 168 /* 169 * [lock the part of the user address space involved 170 * in the transfer] 171 * Beware vmapbuf(); it clobbers b_data and 172 * saves it in b_saveaddr. However, vunmapbuf() 173 * restores it. 174 */ 175 PHOLD(p); 176 if (uvm_vslock(p, bp->b_data, todo, (flags & B_READ) ? 177 VM_PROT_READ | VM_PROT_WRITE : VM_PROT_READ) != 178 KERN_SUCCESS) { 179 bp->b_flags |= B_ERROR; 180 bp->b_error = EFAULT; 181 goto after_unlock; 182 } 183 vmapbuf(bp, todo); 184 185 /* [call strategy to start the transfer] */ 186 (*strategy)(bp); 187 188 /* 189 * Note that the raise/wait/lower/get error 190 * steps below would be done by biowait(), but 191 * we want to unlock the address space before 192 * we lower the priority. 193 * 194 * [raise the priority level to splbio] 195 */ 196 s = splbio(); 197 198 /* [wait for the transfer to complete] */ 199 while ((bp->b_flags & B_DONE) == 0) 200 tsleep((caddr_t) bp, PRIBIO + 1, "physio", 0); 201 202 /* Mark it busy again, so nobody else will use it. */ 203 bp->b_flags |= B_BUSY; 204 205 /* [lower the priority level] */ 206 splx(s); 207 208 /* 209 * [unlock the part of the address space previously 210 * locked] 211 */ 212 vunmapbuf(bp, todo); 213 uvm_vsunlock(p, bp->b_data, todo); 214after_unlock: 215 PRELE(p); 216 217 /* remember error value (save a splbio/splx pair) */ 218 if (bp->b_flags & B_ERROR) 219 error = (bp->b_error ? bp->b_error : EIO); 220 221 /* 222 * [deduct the transfer size from the total number 223 * of data to transfer] 224 */ 225 done = bp->b_bcount - bp->b_resid; 226#ifdef DIAGNOSTIC 227 if (done < 0) 228 panic("done < 0; strategy broken"); 229 if (done > todo) 230 panic("done > todo; strategy broken"); 231#endif 232 iovp->iov_len -= done; 233 iovp->iov_base = (caddr_t)iovp->iov_base + done; 234 uio->uio_offset += done; 235 uio->uio_resid -= done; 236 237 /* 238 * Now, check for an error. 239 * Also, handle weird end-of-disk semantics. 240 */ 241 if (error || done < todo) 242 goto done; 243 } 244 } 245 246done: 247 /* 248 * [clean up the state of the buffer] 249 * Remember if somebody wants it, so we can wake them up below. 250 * Also, if we had to steal it, give it back. 251 */ 252 s = splbio(); 253 bp->b_flags &= ~(B_BUSY | B_PHYS | B_RAW); 254 if (nobuf) 255 putphysbuf(bp); 256 else { 257 /* 258 * [if another process is waiting for the raw I/O buffer, 259 * wake up processes waiting to do physical I/O; 260 */ 261 if (bp->b_flags & B_WANTED) { 262 bp->b_flags &= ~B_WANTED; 263 wakeup(bp); 264 } 265 } 266 splx(s); 267 268 return (error); 269} 270 271/* 272 * Get a swap buffer structure, for use in physical I/O. 273 * Mostly taken from /sys/vm/swap_pager.c, except that it no longer 274 * records buffer list-empty conditions, and sleeps at PRIBIO + 1, 275 * rather than PSWP + 1 (and on a different wchan). 276 */ 277struct buf * 278getphysbuf() 279{ 280 struct buf *bp; 281 282 bp = pool_get(&bufpool, PR_WAITOK); 283 bzero(bp, sizeof(*bp)); 284 285 /* XXXCDC: are the following two lines necessary? */ 286 bp->b_rcred = bp->b_wcred = NOCRED; 287 bp->b_vnbufs.le_next = NOLIST; 288 289 return (bp); 290} 291 292/* 293 * Get rid of a swap buffer structure which has been used in physical I/O. 294 * Mostly taken from /sys/vm/swap_pager.c, except that it now uses 295 * wakeup() rather than the VM-internal thread_wakeup(), and that the caller 296 * must mask disk interrupts, rather than putphysbuf() itself. 297 */ 298void 299putphysbuf(bp) 300 struct buf *bp; 301{ 302 /* XXXCDC: is this necesary? */ 303 if (bp->b_vp) 304 brelvp(bp); 305 306#ifdef DIAGNOSTIC 307 if (bp->b_flags & B_WANTED) 308 panic("putphysbuf: private buf B_WANTED"); 309#endif 310 pool_put(&bufpool, bp); 311} 312 313/* 314 * Leffler, et al., says on p. 231: 315 * "The minphys() routine is called by physio() to adjust the 316 * size of each I/O transfer before the latter is passed to 317 * the strategy routine..." 318 * 319 * so, just adjust the buffer's count accounting to MAXPHYS here, 320 * and return the new count; 321 */ 322void 323minphys(bp) 324 struct buf *bp; 325{ 326 327 if (bp->b_bcount > MAXPHYS) 328 bp->b_bcount = MAXPHYS; 329} 330