xref: /freebsd/sys/kern/kern_physio.c (revision 2008043f386721d58158e37e0d7e50df8095942d)
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 #include <sys/param.h>
22 #include <sys/systm.h>
23 #include <sys/bio.h>
24 #include <sys/buf.h>
25 #include <sys/conf.h>
26 #include <sys/malloc.h>
27 #include <sys/proc.h>
28 #include <sys/racct.h>
29 #include <sys/rwlock.h>
30 #include <sys/uio.h>
31 #include <geom/geom.h>
32 
33 #include <vm/vm.h>
34 #include <vm/vm_object.h>
35 #include <vm/vm_page.h>
36 #include <vm/vm_pager.h>
37 #include <vm/vm_extern.h>
38 #include <vm/vm_map.h>
39 
40 int
41 physio(struct cdev *dev, struct uio *uio, int ioflag)
42 {
43 	struct cdevsw *csw;
44 	struct buf *pbuf;
45 	struct bio *bp;
46 	struct vm_page **pages;
47 	char *base, *sa;
48 	u_int iolen, poff;
49 	int error, i, npages, maxpages;
50 	vm_prot_t prot;
51 
52 	csw = dev->si_devsw;
53 	npages = 0;
54 	sa = NULL;
55 	/* check if character device is being destroyed */
56 	if (csw == NULL)
57 		return (ENXIO);
58 
59 	/* XXX: sanity check */
60 	if (dev->si_iosize_max < PAGE_SIZE) {
61 		printf("WARNING: %s si_iosize_max=%d, using DFLTPHYS.\n",
62 		    devtoname(dev), dev->si_iosize_max);
63 		dev->si_iosize_max = DFLTPHYS;
64 	}
65 
66 	/*
67 	 * If the driver does not want I/O to be split, that means that we
68 	 * need to reject any requests that will not fit into one buffer.
69 	 */
70 	if (dev->si_flags & SI_NOSPLIT &&
71 	    (uio->uio_resid > dev->si_iosize_max || uio->uio_resid > maxphys ||
72 	    uio->uio_iovcnt > 1)) {
73 		/*
74 		 * Tell the user why his I/O was rejected.
75 		 */
76 		if (uio->uio_resid > dev->si_iosize_max)
77 			uprintf("%s: request size=%zd > si_iosize_max=%d; "
78 			    "cannot split request\n", devtoname(dev),
79 			    uio->uio_resid, dev->si_iosize_max);
80 		if (uio->uio_resid > maxphys)
81 			uprintf("%s: request size=%zd > maxphys=%lu; "
82 			    "cannot split request\n", devtoname(dev),
83 			    uio->uio_resid, maxphys);
84 		if (uio->uio_iovcnt > 1)
85 			uprintf("%s: request vectors=%d > 1; "
86 			    "cannot split request\n", devtoname(dev),
87 			    uio->uio_iovcnt);
88 		return (EFBIG);
89 	}
90 
91 	/*
92 	 * Keep the process UPAGES from being swapped.  Processes swapped
93 	 * out while holding pbufs, used by swapper, may lead to deadlock.
94 	 */
95 	PHOLD(curproc);
96 
97 	bp = g_alloc_bio();
98 	if (uio->uio_segflg != UIO_USERSPACE) {
99 		pbuf = NULL;
100 		pages = NULL;
101 	} else if ((dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed) {
102 		pbuf = NULL;
103 		maxpages = btoc(MIN(uio->uio_resid, maxphys)) + 1;
104 		pages = malloc(sizeof(*pages) * maxpages, M_DEVBUF, M_WAITOK);
105 	} else {
106 		pbuf = uma_zalloc(pbuf_zone, M_WAITOK);
107 		MPASS((pbuf->b_flags & B_MAXPHYS) != 0);
108 		sa = pbuf->b_data;
109 		maxpages = PBUF_PAGES;
110 		pages = pbuf->b_pages;
111 	}
112 	prot = VM_PROT_READ;
113 	if (uio->uio_rw == UIO_READ)
114 		prot |= VM_PROT_WRITE;	/* Less backwards than it looks */
115 	error = 0;
116 	for (i = 0; i < uio->uio_iovcnt; i++) {
117 #ifdef RACCT
118 		if (racct_enable) {
119 			PROC_LOCK(curproc);
120 			if (uio->uio_rw == UIO_READ) {
121 				racct_add_force(curproc, RACCT_READBPS,
122 				    uio->uio_iov[i].iov_len);
123 				racct_add_force(curproc, RACCT_READIOPS, 1);
124 			} else {
125 				racct_add_force(curproc, RACCT_WRITEBPS,
126 				    uio->uio_iov[i].iov_len);
127 				racct_add_force(curproc, RACCT_WRITEIOPS, 1);
128 			}
129 			PROC_UNLOCK(curproc);
130 		}
131 #endif /* RACCT */
132 
133 		while (uio->uio_iov[i].iov_len) {
134 			g_reset_bio(bp);
135 			if (uio->uio_rw == UIO_READ) {
136 				bp->bio_cmd = BIO_READ;
137 				curthread->td_ru.ru_inblock++;
138 			} else {
139 				bp->bio_cmd = BIO_WRITE;
140 				curthread->td_ru.ru_oublock++;
141 			}
142 			bp->bio_offset = uio->uio_offset;
143 			base = uio->uio_iov[i].iov_base;
144 			bp->bio_length = uio->uio_iov[i].iov_len;
145 			if (bp->bio_length > dev->si_iosize_max)
146 				bp->bio_length = dev->si_iosize_max;
147 			if (bp->bio_length > maxphys)
148 				bp->bio_length = maxphys;
149 			bp->bio_bcount = bp->bio_length;
150 			bp->bio_dev = dev;
151 
152 			if (pages) {
153 				if ((npages = vm_fault_quick_hold_pages(
154 				    &curproc->p_vmspace->vm_map,
155 				    (vm_offset_t)base, bp->bio_length,
156 				    prot, pages, maxpages)) < 0) {
157 					error = EFAULT;
158 					goto doerror;
159 				}
160 				poff = (vm_offset_t)base & PAGE_MASK;
161 				if (pbuf && sa) {
162 					pmap_qenter((vm_offset_t)sa,
163 					    pages, npages);
164 					bp->bio_data = sa + poff;
165 				} else {
166 					bp->bio_ma = pages;
167 					bp->bio_ma_n = npages;
168 					bp->bio_ma_offset = poff;
169 					bp->bio_data = unmapped_buf;
170 					bp->bio_flags |= BIO_UNMAPPED;
171 				}
172 			} else
173 				bp->bio_data = base;
174 
175 			csw->d_strategy(bp);
176 			if (uio->uio_rw == UIO_READ)
177 				biowait(bp, "physrd");
178 			else
179 				biowait(bp, "physwr");
180 
181 			if (pages) {
182 				if (pbuf)
183 					pmap_qremove((vm_offset_t)sa, npages);
184 				vm_page_unhold_pages(pages, npages);
185 			}
186 
187 			iolen = bp->bio_length - bp->bio_resid;
188 			if (iolen == 0 && !(bp->bio_flags & BIO_ERROR))
189 				goto doerror;	/* EOF */
190 			uio->uio_iov[i].iov_len -= iolen;
191 			uio->uio_iov[i].iov_base =
192 			    (char *)uio->uio_iov[i].iov_base + iolen;
193 			uio->uio_resid -= iolen;
194 			uio->uio_offset += iolen;
195 			if (bp->bio_flags & BIO_ERROR) {
196 				error = bp->bio_error;
197 				goto doerror;
198 			}
199 		}
200 	}
201 doerror:
202 	if (pbuf)
203 		uma_zfree(pbuf_zone, pbuf);
204 	else if (pages)
205 		free(pages, M_DEVBUF);
206 	g_destroy_bio(bp);
207 	PRELE(curproc);
208 	return (error);
209 }
210