xref: /freebsd/sys/kern/kern_physio.c (revision 8ddb146abcdf061be9f2c0db7e391697dafad85c)
1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause
3  *
4  * Copyright (c) 1994 John S. Dyson
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice immediately at the beginning of the file, without modification,
12  *    this list of conditions, and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Absolutely no warranty of function or purpose is made by the author
17  *    John S. Dyson.
18  * 4. Modifications may be freely made to this file if the above conditions
19  *    are met.
20  */
21 
22 #include <sys/cdefs.h>
23 __FBSDID("$FreeBSD$");
24 
25 #include <sys/param.h>
26 #include <sys/systm.h>
27 #include <sys/bio.h>
28 #include <sys/buf.h>
29 #include <sys/conf.h>
30 #include <sys/malloc.h>
31 #include <sys/proc.h>
32 #include <sys/racct.h>
33 #include <sys/rwlock.h>
34 #include <sys/uio.h>
35 #include <geom/geom.h>
36 
37 #include <vm/vm.h>
38 #include <vm/vm_object.h>
39 #include <vm/vm_page.h>
40 #include <vm/vm_pager.h>
41 #include <vm/vm_extern.h>
42 #include <vm/vm_map.h>
43 
44 int
45 physio(struct cdev *dev, struct uio *uio, int ioflag)
46 {
47 	struct cdevsw *csw;
48 	struct buf *pbuf;
49 	struct bio *bp;
50 	struct vm_page **pages;
51 	char *base, *sa;
52 	u_int iolen, poff;
53 	int error, i, npages, maxpages;
54 	vm_prot_t prot;
55 
56 	csw = dev->si_devsw;
57 	npages = 0;
58 	sa = NULL;
59 	/* check if character device is being destroyed */
60 	if (csw == NULL)
61 		return (ENXIO);
62 
63 	/* XXX: sanity check */
64 	if (dev->si_iosize_max < PAGE_SIZE) {
65 		printf("WARNING: %s si_iosize_max=%d, using DFLTPHYS.\n",
66 		    devtoname(dev), dev->si_iosize_max);
67 		dev->si_iosize_max = DFLTPHYS;
68 	}
69 
70 	/*
71 	 * If the driver does not want I/O to be split, that means that we
72 	 * need to reject any requests that will not fit into one buffer.
73 	 */
74 	if (dev->si_flags & SI_NOSPLIT &&
75 	    (uio->uio_resid > dev->si_iosize_max || uio->uio_resid > maxphys ||
76 	    uio->uio_iovcnt > 1)) {
77 		/*
78 		 * Tell the user why his I/O was rejected.
79 		 */
80 		if (uio->uio_resid > dev->si_iosize_max)
81 			uprintf("%s: request size=%zd > si_iosize_max=%d; "
82 			    "cannot split request\n", devtoname(dev),
83 			    uio->uio_resid, dev->si_iosize_max);
84 		if (uio->uio_resid > maxphys)
85 			uprintf("%s: request size=%zd > maxphys=%lu; "
86 			    "cannot split request\n", devtoname(dev),
87 			    uio->uio_resid, maxphys);
88 		if (uio->uio_iovcnt > 1)
89 			uprintf("%s: request vectors=%d > 1; "
90 			    "cannot split request\n", devtoname(dev),
91 			    uio->uio_iovcnt);
92 		return (EFBIG);
93 	}
94 
95 	/*
96 	 * Keep the process UPAGES from being swapped.  Processes swapped
97 	 * out while holding pbufs, used by swapper, may lead to deadlock.
98 	 */
99 	PHOLD(curproc);
100 
101 	bp = g_alloc_bio();
102 	if (uio->uio_segflg != UIO_USERSPACE) {
103 		pbuf = NULL;
104 		pages = NULL;
105 	} else if ((dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed) {
106 		pbuf = NULL;
107 		maxpages = btoc(MIN(uio->uio_resid, maxphys)) + 1;
108 		pages = malloc(sizeof(*pages) * maxpages, M_DEVBUF, M_WAITOK);
109 	} else {
110 		pbuf = uma_zalloc(pbuf_zone, M_WAITOK);
111 		MPASS((pbuf->b_flags & B_MAXPHYS) != 0);
112 		sa = pbuf->b_data;
113 		maxpages = PBUF_PAGES;
114 		pages = pbuf->b_pages;
115 	}
116 	prot = VM_PROT_READ;
117 	if (uio->uio_rw == UIO_READ)
118 		prot |= VM_PROT_WRITE;	/* Less backwards than it looks */
119 	error = 0;
120 	for (i = 0; i < uio->uio_iovcnt; i++) {
121 #ifdef RACCT
122 		if (racct_enable) {
123 			PROC_LOCK(curproc);
124 			if (uio->uio_rw == UIO_READ) {
125 				racct_add_force(curproc, RACCT_READBPS,
126 				    uio->uio_iov[i].iov_len);
127 				racct_add_force(curproc, RACCT_READIOPS, 1);
128 			} else {
129 				racct_add_force(curproc, RACCT_WRITEBPS,
130 				    uio->uio_iov[i].iov_len);
131 				racct_add_force(curproc, RACCT_WRITEIOPS, 1);
132 			}
133 			PROC_UNLOCK(curproc);
134 		}
135 #endif /* RACCT */
136 
137 		while (uio->uio_iov[i].iov_len) {
138 			g_reset_bio(bp);
139 			if (uio->uio_rw == UIO_READ) {
140 				bp->bio_cmd = BIO_READ;
141 				curthread->td_ru.ru_inblock++;
142 			} else {
143 				bp->bio_cmd = BIO_WRITE;
144 				curthread->td_ru.ru_oublock++;
145 			}
146 			bp->bio_offset = uio->uio_offset;
147 			base = uio->uio_iov[i].iov_base;
148 			bp->bio_length = uio->uio_iov[i].iov_len;
149 			if (bp->bio_length > dev->si_iosize_max)
150 				bp->bio_length = dev->si_iosize_max;
151 			if (bp->bio_length > maxphys)
152 				bp->bio_length = maxphys;
153 			bp->bio_bcount = bp->bio_length;
154 			bp->bio_dev = dev;
155 
156 			if (pages) {
157 				if ((npages = vm_fault_quick_hold_pages(
158 				    &curproc->p_vmspace->vm_map,
159 				    (vm_offset_t)base, bp->bio_length,
160 				    prot, pages, maxpages)) < 0) {
161 					error = EFAULT;
162 					goto doerror;
163 				}
164 				poff = (vm_offset_t)base & PAGE_MASK;
165 				if (pbuf && sa) {
166 					pmap_qenter((vm_offset_t)sa,
167 					    pages, npages);
168 					bp->bio_data = sa + poff;
169 				} else {
170 					bp->bio_ma = pages;
171 					bp->bio_ma_n = npages;
172 					bp->bio_ma_offset = poff;
173 					bp->bio_data = unmapped_buf;
174 					bp->bio_flags |= BIO_UNMAPPED;
175 				}
176 			} else
177 				bp->bio_data = base;
178 
179 			csw->d_strategy(bp);
180 			if (uio->uio_rw == UIO_READ)
181 				biowait(bp, "physrd");
182 			else
183 				biowait(bp, "physwr");
184 
185 			if (pages) {
186 				if (pbuf)
187 					pmap_qremove((vm_offset_t)sa, npages);
188 				vm_page_unhold_pages(pages, npages);
189 			}
190 
191 			iolen = bp->bio_length - bp->bio_resid;
192 			if (iolen == 0 && !(bp->bio_flags & BIO_ERROR))
193 				goto doerror;	/* EOF */
194 			uio->uio_iov[i].iov_len -= iolen;
195 			uio->uio_iov[i].iov_base =
196 			    (char *)uio->uio_iov[i].iov_base + iolen;
197 			uio->uio_resid -= iolen;
198 			uio->uio_offset += iolen;
199 			if (bp->bio_flags & BIO_ERROR) {
200 				error = bp->bio_error;
201 				goto doerror;
202 			}
203 		}
204 	}
205 doerror:
206 	if (pbuf)
207 		uma_zfree(pbuf_zone, pbuf);
208 	else if (pages)
209 		free(pages, M_DEVBUF);
210 	g_destroy_bio(bp);
211 	PRELE(curproc);
212 	return (error);
213 }
214