kern_physio.c revision 109572 
1/*
2 * Copyright (c) 1994 John S. Dyson
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice immediately at the beginning of the file, without modification,
10 *    this list of conditions, and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 * 3. Absolutely no warranty of function or purpose is made by the author
15 *    John S. Dyson.
16 * 4. Modifications may be freely made to this file if the above conditions
17 *    are met.
18 *
19 * $FreeBSD: head/sys/kern/kern_physio.c 109572 2003-01-20 17:46:48Z dillon $
20 */
21
22#include <sys/param.h>
23#include <sys/systm.h>
24#include <sys/bio.h>
25#include <sys/buf.h>
26#include <sys/conf.h>
27#include <sys/proc.h>
28#include <sys/uio.h>
29
30#include <vm/vm.h>
31#include <vm/vm_extern.h>
32
33static void
34physwakeup(struct buf *bp)
35{
36	wakeup(bp);
37}
38
39int
40physio(dev_t dev, struct uio *uio, int ioflag)
41{
42	int i;
43	int error;
44	int spl;
45	caddr_t sa;
46	u_int iolen;
47	struct buf *bp;
48
49	/* Keep the process UPAGES from being swapped. XXX: why ? */
50	PHOLD(curproc);
51
52	bp = getpbuf(NULL);
53	sa = bp->b_data;
54	error = bp->b_error = 0;
55
56	/* XXX: sanity check */
57	if(dev->si_iosize_max < PAGE_SIZE) {
58		printf("WARNING: %s si_iosize_max=%d, using DFLTPHYS.\n",
59		    devtoname(dev), dev->si_iosize_max);
60		dev->si_iosize_max = DFLTPHYS;
61	}
62
63	for (i = 0; i < uio->uio_iovcnt; i++) {
64		while (uio->uio_iov[i].iov_len) {
65			bp->b_flags = B_PHYS;
66			if (uio->uio_rw == UIO_READ)
67				bp->b_iocmd = BIO_READ;
68			else
69				bp->b_iocmd = BIO_WRITE;
70			bp->b_dev = dev;
71			bp->b_iodone = physwakeup;
72			bp->b_data = uio->uio_iov[i].iov_base;
73			bp->b_bcount = uio->uio_iov[i].iov_len;
74			bp->b_offset = uio->uio_offset;
75			bp->b_saveaddr = sa;
76
77			/* Don't exceed drivers iosize limit */
78			if (bp->b_bcount > dev->si_iosize_max)
79				bp->b_bcount = dev->si_iosize_max;
80
81			/*
82			 * Make sure the pbuf can map the request
83			 * XXX: The pbuf has kvasize = MAXPHYS so a request
84			 * XXX: larger than MAXPHYS - PAGE_SIZE must be
85			 * XXX: page aligned or it will be fragmented.
86			 */
87			iolen = ((vm_offset_t) bp->b_data) & PAGE_MASK;
88			if ((bp->b_bcount + iolen) > bp->b_kvasize) {
89				bp->b_bcount = bp->b_kvasize;
90				if (iolen != 0)
91					bp->b_bcount -= PAGE_SIZE;
92			}
93			bp->b_bufsize = bp->b_bcount;
94
95			bp->b_blkno = btodb(bp->b_offset);
96
97			if (uio->uio_segflg == UIO_USERSPACE) {
98				/*
99				 * Note that useracc() alone is not a
100				 * sufficient test.  vmapbuf() can still fail
101				 * due to a smaller file mapped into a larger
102				 * area of VM, or if userland races against
103				 * vmapbuf() after the useracc() check.
104				 */
105				if (!useracc(bp->b_data, bp->b_bufsize,
106				    bp->b_iocmd == BIO_READ ?
107				    VM_PROT_WRITE : VM_PROT_READ)) {
108					error = EFAULT;
109					goto doerror;
110				}
111				if (vmapbuf(bp) < 0) {
112					error = EFAULT;
113					goto doerror;
114				}
115			}
116
117			DEV_STRATEGY(bp);
118			spl = splbio();
119			while ((bp->b_flags & B_DONE) == 0)
120				tsleep(bp, PRIBIO, "physstr", 0);
121			splx(spl);
122
123			if (uio->uio_segflg == UIO_USERSPACE)
124				vunmapbuf(bp);
125			iolen = bp->b_bcount - bp->b_resid;
126			if (iolen == 0 && !(bp->b_ioflags & BIO_ERROR))
127				goto doerror;	/* EOF */
128			uio->uio_iov[i].iov_len -= iolen;
129			uio->uio_iov[i].iov_base =
130			    (char *)uio->uio_iov[i].iov_base + iolen;
131			uio->uio_resid -= iolen;
132			uio->uio_offset += iolen;
133			if( bp->b_ioflags & BIO_ERROR) {
134				error = bp->b_error;
135				goto doerror;
136			}
137		}
138	}
139doerror:
140	relpbuf(bp, NULL);
141	PRELE(curproc);
142	return (error);
143}
144