xref: /freebsd/sys/kern/kern_physio.c (revision 6ae1554a5d9b318f8ad53ccc39fa5a961403da73)
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 
20 #include <sys/cdefs.h>
21 __FBSDID("$FreeBSD$");
22 
23 #include <sys/param.h>
24 #include <sys/systm.h>
25 #include <sys/bio.h>
26 #include <sys/buf.h>
27 #include <sys/conf.h>
28 #include <sys/malloc.h>
29 #include <sys/proc.h>
30 #include <sys/uio.h>
31 #include <geom/geom.h>
32 
33 #include <vm/vm.h>
34 #include <vm/vm_page.h>
35 #include <vm/vm_extern.h>
36 #include <vm/vm_map.h>
37 
38 int
39 physio(struct cdev *dev, struct uio *uio, int ioflag)
40 {
41 	struct buf *pbuf;
42 	struct bio *bp;
43 	struct vm_page **pages;
44 	caddr_t sa;
45 	u_int iolen, poff;
46 	int error, i, npages, maxpages;
47 	vm_prot_t prot;
48 
49 	/* XXX: sanity check */
50 	if(dev->si_iosize_max < PAGE_SIZE) {
51 		printf("WARNING: %s si_iosize_max=%d, using DFLTPHYS.\n",
52 		    devtoname(dev), dev->si_iosize_max);
53 		dev->si_iosize_max = DFLTPHYS;
54 	}
55 
56 	/*
57 	 * If the driver does not want I/O to be split, that means that we
58 	 * need to reject any requests that will not fit into one buffer.
59 	 */
60 	if (dev->si_flags & SI_NOSPLIT &&
61 	    (uio->uio_resid > dev->si_iosize_max || uio->uio_resid > MAXPHYS ||
62 	    uio->uio_iovcnt > 1)) {
63 		/*
64 		 * Tell the user why his I/O was rejected.
65 		 */
66 		if (uio->uio_resid > dev->si_iosize_max)
67 			uprintf("%s: request size=%zd > si_iosize_max=%d; "
68 			    "cannot split request\n", devtoname(dev),
69 			    uio->uio_resid, dev->si_iosize_max);
70 		if (uio->uio_resid > MAXPHYS)
71 			uprintf("%s: request size=%zd > MAXPHYS=%d; "
72 			    "cannot split request\n", devtoname(dev),
73 			    uio->uio_resid, MAXPHYS);
74 		if (uio->uio_iovcnt > 1)
75 			uprintf("%s: request vectors=%d > 1; "
76 			    "cannot split request\n", devtoname(dev),
77 			    uio->uio_iovcnt);
78 		return (EFBIG);
79 	}
80 
81 	/*
82 	 * Keep the process UPAGES from being swapped.  Processes swapped
83 	 * out while holding pbufs, used by swapper, may lead to deadlock.
84 	 */
85 	PHOLD(curproc);
86 
87 	bp = g_alloc_bio();
88 	if (uio->uio_segflg != UIO_USERSPACE) {
89 		pbuf = NULL;
90 		pages = NULL;
91 	} else if ((dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed) {
92 		pbuf = NULL;
93 		maxpages = btoc(MIN(uio->uio_resid, MAXPHYS)) + 1;
94 		pages = malloc(sizeof(*pages) * maxpages, M_DEVBUF, M_WAITOK);
95 	} else {
96 		pbuf = getpbuf(NULL);
97 		sa = pbuf->b_data;
98 		maxpages = btoc(MAXPHYS);
99 		pages = pbuf->b_pages;
100 	}
101 	prot = VM_PROT_READ;
102 	if (uio->uio_rw == UIO_READ)
103 		prot |= VM_PROT_WRITE;	/* Less backwards than it looks */
104 	error = 0;
105 	for (i = 0; i < uio->uio_iovcnt; i++) {
106 		while (uio->uio_iov[i].iov_len) {
107 			bzero(bp, sizeof(*bp));
108 			if (uio->uio_rw == UIO_READ) {
109 				bp->bio_cmd = BIO_READ;
110 				curthread->td_ru.ru_inblock++;
111 			} else {
112 				bp->bio_cmd = BIO_WRITE;
113 				curthread->td_ru.ru_oublock++;
114 			}
115 			bp->bio_offset = uio->uio_offset;
116 			bp->bio_data = uio->uio_iov[i].iov_base;
117 			bp->bio_length = uio->uio_iov[i].iov_len;
118 			if (bp->bio_length > dev->si_iosize_max)
119 				bp->bio_length = dev->si_iosize_max;
120 			if (bp->bio_length > MAXPHYS)
121 				bp->bio_length = MAXPHYS;
122 
123 			/*
124 			 * Make sure the pbuf can map the request.
125 			 * The pbuf has kvasize = MAXPHYS, so a request
126 			 * larger than MAXPHYS - PAGE_SIZE must be
127 			 * page aligned or it will be fragmented.
128 			 */
129 			poff = (vm_offset_t)bp->bio_data & PAGE_MASK;
130 			if (pbuf && bp->bio_length + poff > pbuf->b_kvasize) {
131 				if (dev->si_flags & SI_NOSPLIT) {
132 					uprintf("%s: request ptr %p is not "
133 					    "on a page boundary; cannot split "
134 					    "request\n", devtoname(dev),
135 					    bp->bio_data);
136 					error = EFBIG;
137 					goto doerror;
138 				}
139 				bp->bio_length = pbuf->b_kvasize;
140 				if (poff != 0)
141 					bp->bio_length -= PAGE_SIZE;
142 			}
143 
144 			bp->bio_bcount = bp->bio_length;
145 			bp->bio_dev = dev;
146 
147 			if (pages) {
148 				if ((npages = vm_fault_quick_hold_pages(
149 				    &curproc->p_vmspace->vm_map,
150 				    (vm_offset_t)bp->bio_data, bp->bio_length,
151 				    prot, pages, maxpages)) < 0) {
152 					error = EFAULT;
153 					goto doerror;
154 				}
155 				if (pbuf) {
156 					pmap_qenter((vm_offset_t)sa,
157 					    pages, npages);
158 					bp->bio_data = sa + poff;
159 				} else {
160 					bp->bio_ma = pages;
161 					bp->bio_ma_n = npages;
162 					bp->bio_ma_offset = poff;
163 					bp->bio_data = unmapped_buf;
164 					bp->bio_flags |= BIO_UNMAPPED;
165 				}
166 			}
167 
168 			dev->si_devsw->d_strategy(bp);
169 			if (uio->uio_rw == UIO_READ)
170 				biowait(bp, "physrd");
171 			else
172 				biowait(bp, "physwr");
173 
174 			if (pages) {
175 				if (pbuf)
176 					pmap_qremove((vm_offset_t)sa, npages);
177 				vm_page_unhold_pages(pages, npages);
178 			}
179 
180 			iolen = bp->bio_length - bp->bio_resid;
181 			if (iolen == 0 && !(bp->bio_flags & BIO_ERROR))
182 				goto doerror;	/* EOF */
183 			uio->uio_iov[i].iov_len -= iolen;
184 			uio->uio_iov[i].iov_base =
185 			    (char *)uio->uio_iov[i].iov_base + iolen;
186 			uio->uio_resid -= iolen;
187 			uio->uio_offset += iolen;
188 			if (bp->bio_flags & BIO_ERROR) {
189 				error = bp->bio_error;
190 				goto doerror;
191 			}
192 		}
193 	}
194 doerror:
195 	if (pbuf)
196 		relpbuf(pbuf, NULL);
197 	else if (pages)
198 		free(pages, M_DEVBUF);
199 	g_destroy_bio(bp);
200 	PRELE(curproc);
201 	return (error);
202 }
203