xref: /freebsd/sys/kern/kern_physio.c (revision 2b8331622f0b212cf3bb4fc4914a501e5321d506)
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/rwlock.h>
32 #include <sys/uio.h>
33 #include <geom/geom.h>
34 
35 #include <vm/vm.h>
36 #include <vm/vm_object.h>
37 #include <vm/vm_page.h>
38 #include <vm/vm_pager.h>
39 #include <vm/vm_extern.h>
40 #include <vm/vm_map.h>
41 
42 int
43 physio(struct cdev *dev, struct uio *uio, int ioflag)
44 {
45 	struct cdevsw *csw;
46 	struct buf *pbuf;
47 	struct bio *bp;
48 	struct vm_page **pages;
49 	char *base, *sa;
50 	u_int iolen, poff;
51 	int error, i, npages, maxpages;
52 	vm_prot_t prot;
53 
54 	csw = dev->si_devsw;
55 	npages = 0;
56 	sa = NULL;
57 	/* check if character device is being destroyed */
58 	if (csw == NULL)
59 		return (ENXIO);
60 
61 	/* XXX: sanity check */
62 	if (dev->si_iosize_max < PAGE_SIZE) {
63 		printf("WARNING: %s si_iosize_max=%d, using DFLTPHYS.\n",
64 		    devtoname(dev), dev->si_iosize_max);
65 		dev->si_iosize_max = DFLTPHYS;
66 	}
67 
68 	/*
69 	 * If the driver does not want I/O to be split, that means that we
70 	 * need to reject any requests that will not fit into one buffer.
71 	 */
72 	if (dev->si_flags & SI_NOSPLIT &&
73 	    (uio->uio_resid > dev->si_iosize_max || uio->uio_resid > maxphys ||
74 	    uio->uio_iovcnt > 1)) {
75 		/*
76 		 * Tell the user why his I/O was rejected.
77 		 */
78 		if (uio->uio_resid > dev->si_iosize_max)
79 			uprintf("%s: request size=%zd > si_iosize_max=%d; "
80 			    "cannot split request\n", devtoname(dev),
81 			    uio->uio_resid, dev->si_iosize_max);
82 		if (uio->uio_resid > maxphys)
83 			uprintf("%s: request size=%zd > maxphys=%lu; "
84 			    "cannot split request\n", devtoname(dev),
85 			    uio->uio_resid, maxphys);
86 		if (uio->uio_iovcnt > 1)
87 			uprintf("%s: request vectors=%d > 1; "
88 			    "cannot split request\n", devtoname(dev),
89 			    uio->uio_iovcnt);
90 		return (EFBIG);
91 	}
92 
93 	/*
94 	 * Keep the process UPAGES from being swapped.  Processes swapped
95 	 * out while holding pbufs, used by swapper, may lead to deadlock.
96 	 */
97 	PHOLD(curproc);
98 
99 	bp = g_alloc_bio();
100 	if (uio->uio_segflg != UIO_USERSPACE) {
101 		pbuf = NULL;
102 		pages = NULL;
103 	} else if ((dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed) {
104 		pbuf = NULL;
105 		maxpages = btoc(MIN(uio->uio_resid, maxphys)) + 1;
106 		pages = malloc(sizeof(*pages) * maxpages, M_DEVBUF, M_WAITOK);
107 	} else {
108 		pbuf = uma_zalloc(pbuf_zone, M_WAITOK);
109 		MPASS((pbuf->b_flags & B_MAXPHYS) != 0);
110 		sa = pbuf->b_data;
111 		maxpages = PBUF_PAGES;
112 		pages = pbuf->b_pages;
113 	}
114 	prot = VM_PROT_READ;
115 	if (uio->uio_rw == UIO_READ)
116 		prot |= VM_PROT_WRITE;	/* Less backwards than it looks */
117 	error = 0;
118 	for (i = 0; i < uio->uio_iovcnt; i++) {
119 #ifdef RACCT
120 		if (racct_enable) {
121 			PROC_LOCK(curproc);
122 			if (uio->uio_rw == UIO_READ) {
123 				racct_add_force(curproc, RACCT_READBPS,
124 				    uio->uio_iov[i].iov_len);
125 				racct_add_force(curproc, RACCT_READIOPS, 1);
126 			} else {
127 				racct_add_force(curproc, RACCT_WRITEBPS,
128 				    uio->uio_iov[i].iov_len);
129 				racct_add_force(curproc, RACCT_WRITEIOPS, 1);
130 			}
131 			PROC_UNLOCK(curproc);
132 		}
133 #endif /* RACCT */
134 
135 		while (uio->uio_iov[i].iov_len) {
136 			g_reset_bio(bp);
137 			if (uio->uio_rw == UIO_READ) {
138 				bp->bio_cmd = BIO_READ;
139 				curthread->td_ru.ru_inblock++;
140 			} else {
141 				bp->bio_cmd = BIO_WRITE;
142 				curthread->td_ru.ru_oublock++;
143 			}
144 			bp->bio_offset = uio->uio_offset;
145 			base = uio->uio_iov[i].iov_base;
146 			bp->bio_length = uio->uio_iov[i].iov_len;
147 			if (bp->bio_length > dev->si_iosize_max)
148 				bp->bio_length = dev->si_iosize_max;
149 			if (bp->bio_length > maxphys)
150 				bp->bio_length = maxphys;
151 			bp->bio_bcount = bp->bio_length;
152 			bp->bio_dev = dev;
153 
154 			if (pages) {
155 				if ((npages = vm_fault_quick_hold_pages(
156 				    &curproc->p_vmspace->vm_map,
157 				    (vm_offset_t)base, bp->bio_length,
158 				    prot, pages, maxpages)) < 0) {
159 					error = EFAULT;
160 					goto doerror;
161 				}
162 				poff = (vm_offset_t)base & PAGE_MASK;
163 				if (pbuf && sa) {
164 					pmap_qenter((vm_offset_t)sa,
165 					    pages, npages);
166 					bp->bio_data = sa + poff;
167 				} else {
168 					bp->bio_ma = pages;
169 					bp->bio_ma_n = npages;
170 					bp->bio_ma_offset = poff;
171 					bp->bio_data = unmapped_buf;
172 					bp->bio_flags |= BIO_UNMAPPED;
173 				}
174 			} else
175 				bp->bio_data = base;
176 
177 			csw->d_strategy(bp);
178 			if (uio->uio_rw == UIO_READ)
179 				biowait(bp, "physrd");
180 			else
181 				biowait(bp, "physwr");
182 
183 			if (pages) {
184 				if (pbuf)
185 					pmap_qremove((vm_offset_t)sa, npages);
186 				vm_page_unhold_pages(pages, npages);
187 			}
188 
189 			iolen = bp->bio_length - bp->bio_resid;
190 			if (iolen == 0 && !(bp->bio_flags & BIO_ERROR))
191 				goto doerror;	/* EOF */
192 			uio->uio_iov[i].iov_len -= iolen;
193 			uio->uio_iov[i].iov_base =
194 			    (char *)uio->uio_iov[i].iov_base + iolen;
195 			uio->uio_resid -= iolen;
196 			uio->uio_offset += iolen;
197 			if (bp->bio_flags & BIO_ERROR) {
198 				error = bp->bio_error;
199 				goto doerror;
200 			}
201 		}
202 	}
203 doerror:
204 	if (pbuf)
205 		uma_zfree(pbuf_zone, pbuf);
206 	else if (pages)
207 		free(pages, M_DEVBUF);
208 	g_destroy_bio(bp);
209 	PRELE(curproc);
210 	return (error);
211 }
212