xref: /freebsd/sys/kern/kern_physio.c (revision 3416500aef140042c64bc149cb1ec6620483bc44)
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/racct.h>
31 #include <sys/uio.h>
32 #include <geom/geom.h>
33 
34 #include <vm/vm.h>
35 #include <vm/vm_page.h>
36 #include <vm/vm_extern.h>
37 #include <vm/vm_map.h>
38 
39 int
40 physio(struct cdev *dev, struct uio *uio, int ioflag)
41 {
42 	struct cdevsw *csw;
43 	struct buf *pbuf;
44 	struct bio *bp;
45 	struct vm_page **pages;
46 	caddr_t sa;
47 	u_int iolen, poff;
48 	int error, i, npages, maxpages;
49 	vm_prot_t prot;
50 
51 	csw = dev->si_devsw;
52 	/* check if character device is being destroyed */
53 	if (csw == NULL)
54 		return (ENXIO);
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 	/*
64 	 * If the driver does not want I/O to be split, that means that we
65 	 * need to reject any requests that will not fit into one buffer.
66 	 */
67 	if (dev->si_flags & SI_NOSPLIT &&
68 	    (uio->uio_resid > dev->si_iosize_max || uio->uio_resid > MAXPHYS ||
69 	    uio->uio_iovcnt > 1)) {
70 		/*
71 		 * Tell the user why his I/O was rejected.
72 		 */
73 		if (uio->uio_resid > dev->si_iosize_max)
74 			uprintf("%s: request size=%zd > si_iosize_max=%d; "
75 			    "cannot split request\n", devtoname(dev),
76 			    uio->uio_resid, dev->si_iosize_max);
77 		if (uio->uio_resid > MAXPHYS)
78 			uprintf("%s: request size=%zd > MAXPHYS=%d; "
79 			    "cannot split request\n", devtoname(dev),
80 			    uio->uio_resid, MAXPHYS);
81 		if (uio->uio_iovcnt > 1)
82 			uprintf("%s: request vectors=%d > 1; "
83 			    "cannot split request\n", devtoname(dev),
84 			    uio->uio_iovcnt);
85 		return (EFBIG);
86 	}
87 
88 	/*
89 	 * Keep the process UPAGES from being swapped.  Processes swapped
90 	 * out while holding pbufs, used by swapper, may lead to deadlock.
91 	 */
92 	PHOLD(curproc);
93 
94 	bp = g_alloc_bio();
95 	if (uio->uio_segflg != UIO_USERSPACE) {
96 		pbuf = NULL;
97 		pages = NULL;
98 	} else if ((dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed) {
99 		pbuf = NULL;
100 		maxpages = btoc(MIN(uio->uio_resid, MAXPHYS)) + 1;
101 		pages = malloc(sizeof(*pages) * maxpages, M_DEVBUF, M_WAITOK);
102 	} else {
103 		pbuf = getpbuf(NULL);
104 		sa = pbuf->b_data;
105 		maxpages = btoc(MAXPHYS);
106 		pages = pbuf->b_pages;
107 	}
108 	prot = VM_PROT_READ;
109 	if (uio->uio_rw == UIO_READ)
110 		prot |= VM_PROT_WRITE;	/* Less backwards than it looks */
111 	error = 0;
112 	for (i = 0; i < uio->uio_iovcnt; i++) {
113 #ifdef RACCT
114 		if (racct_enable) {
115 			PROC_LOCK(curproc);
116 			if (uio->uio_rw == UIO_READ) {
117 				racct_add_force(curproc, RACCT_READBPS,
118 				    uio->uio_iov[i].iov_len);
119 				racct_add_force(curproc, RACCT_READIOPS, 1);
120 			} else {
121 				racct_add_force(curproc, RACCT_WRITEBPS,
122 				    uio->uio_iov[i].iov_len);
123 				racct_add_force(curproc, RACCT_WRITEIOPS, 1);
124 			}
125 			PROC_UNLOCK(curproc);
126 		}
127 #endif /* RACCT */
128 
129 		while (uio->uio_iov[i].iov_len) {
130 			g_reset_bio(bp);
131 			if (uio->uio_rw == UIO_READ) {
132 				bp->bio_cmd = BIO_READ;
133 				curthread->td_ru.ru_inblock++;
134 			} else {
135 				bp->bio_cmd = BIO_WRITE;
136 				curthread->td_ru.ru_oublock++;
137 			}
138 			bp->bio_offset = uio->uio_offset;
139 			bp->bio_data = uio->uio_iov[i].iov_base;
140 			bp->bio_length = uio->uio_iov[i].iov_len;
141 			if (bp->bio_length > dev->si_iosize_max)
142 				bp->bio_length = dev->si_iosize_max;
143 			if (bp->bio_length > MAXPHYS)
144 				bp->bio_length = MAXPHYS;
145 
146 			/*
147 			 * Make sure the pbuf can map the request.
148 			 * The pbuf has kvasize = MAXPHYS, so a request
149 			 * larger than MAXPHYS - PAGE_SIZE must be
150 			 * page aligned or it will be fragmented.
151 			 */
152 			poff = (vm_offset_t)bp->bio_data & PAGE_MASK;
153 			if (pbuf && bp->bio_length + poff > pbuf->b_kvasize) {
154 				if (dev->si_flags & SI_NOSPLIT) {
155 					uprintf("%s: request ptr %p is not "
156 					    "on a page boundary; cannot split "
157 					    "request\n", devtoname(dev),
158 					    bp->bio_data);
159 					error = EFBIG;
160 					goto doerror;
161 				}
162 				bp->bio_length = pbuf->b_kvasize;
163 				if (poff != 0)
164 					bp->bio_length -= PAGE_SIZE;
165 			}
166 
167 			bp->bio_bcount = bp->bio_length;
168 			bp->bio_dev = dev;
169 
170 			if (pages) {
171 				if ((npages = vm_fault_quick_hold_pages(
172 				    &curproc->p_vmspace->vm_map,
173 				    (vm_offset_t)bp->bio_data, bp->bio_length,
174 				    prot, pages, maxpages)) < 0) {
175 					error = EFAULT;
176 					goto doerror;
177 				}
178 				if (pbuf) {
179 					pmap_qenter((vm_offset_t)sa,
180 					    pages, npages);
181 					bp->bio_data = sa + poff;
182 				} else {
183 					bp->bio_ma = pages;
184 					bp->bio_ma_n = npages;
185 					bp->bio_ma_offset = poff;
186 					bp->bio_data = unmapped_buf;
187 					bp->bio_flags |= BIO_UNMAPPED;
188 				}
189 			}
190 
191 			csw->d_strategy(bp);
192 			if (uio->uio_rw == UIO_READ)
193 				biowait(bp, "physrd");
194 			else
195 				biowait(bp, "physwr");
196 
197 			if (pages) {
198 				if (pbuf)
199 					pmap_qremove((vm_offset_t)sa, npages);
200 				vm_page_unhold_pages(pages, npages);
201 			}
202 
203 			iolen = bp->bio_length - bp->bio_resid;
204 			if (iolen == 0 && !(bp->bio_flags & BIO_ERROR))
205 				goto doerror;	/* EOF */
206 			uio->uio_iov[i].iov_len -= iolen;
207 			uio->uio_iov[i].iov_base =
208 			    (char *)uio->uio_iov[i].iov_base + iolen;
209 			uio->uio_resid -= iolen;
210 			uio->uio_offset += iolen;
211 			if (bp->bio_flags & BIO_ERROR) {
212 				error = bp->bio_error;
213 				goto doerror;
214 			}
215 		}
216 	}
217 doerror:
218 	if (pbuf)
219 		relpbuf(pbuf, NULL);
220 	else if (pages)
221 		free(pages, M_DEVBUF);
222 	g_destroy_bio(bp);
223 	PRELE(curproc);
224 	return (error);
225 }
226