xref: /freebsd/sys/kern/kern_physio.c (revision eb69d1f144a6fcc765d1b9d44a5ae8082353e70b)
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/uio.h>
34 #include <geom/geom.h>
35 
36 #include <vm/vm.h>
37 #include <vm/vm_page.h>
38 #include <vm/vm_extern.h>
39 #include <vm/vm_map.h>
40 
41 int
42 physio(struct cdev *dev, struct uio *uio, int ioflag)
43 {
44 	struct cdevsw *csw;
45 	struct buf *pbuf;
46 	struct bio *bp;
47 	struct vm_page **pages;
48 	caddr_t sa;
49 	u_int iolen, poff;
50 	int error, i, npages, maxpages;
51 	vm_prot_t prot;
52 
53 	csw = dev->si_devsw;
54 	/* check if character device is being destroyed */
55 	if (csw == NULL)
56 		return (ENXIO);
57 
58 	/* XXX: sanity check */
59 	if(dev->si_iosize_max < PAGE_SIZE) {
60 		printf("WARNING: %s si_iosize_max=%d, using DFLTPHYS.\n",
61 		    devtoname(dev), dev->si_iosize_max);
62 		dev->si_iosize_max = DFLTPHYS;
63 	}
64 
65 	/*
66 	 * If the driver does not want I/O to be split, that means that we
67 	 * need to reject any requests that will not fit into one buffer.
68 	 */
69 	if (dev->si_flags & SI_NOSPLIT &&
70 	    (uio->uio_resid > dev->si_iosize_max || uio->uio_resid > MAXPHYS ||
71 	    uio->uio_iovcnt > 1)) {
72 		/*
73 		 * Tell the user why his I/O was rejected.
74 		 */
75 		if (uio->uio_resid > dev->si_iosize_max)
76 			uprintf("%s: request size=%zd > si_iosize_max=%d; "
77 			    "cannot split request\n", devtoname(dev),
78 			    uio->uio_resid, dev->si_iosize_max);
79 		if (uio->uio_resid > MAXPHYS)
80 			uprintf("%s: request size=%zd > MAXPHYS=%d; "
81 			    "cannot split request\n", devtoname(dev),
82 			    uio->uio_resid, MAXPHYS);
83 		if (uio->uio_iovcnt > 1)
84 			uprintf("%s: request vectors=%d > 1; "
85 			    "cannot split request\n", devtoname(dev),
86 			    uio->uio_iovcnt);
87 		return (EFBIG);
88 	}
89 
90 	/*
91 	 * Keep the process UPAGES from being swapped.  Processes swapped
92 	 * out while holding pbufs, used by swapper, may lead to deadlock.
93 	 */
94 	PHOLD(curproc);
95 
96 	bp = g_alloc_bio();
97 	if (uio->uio_segflg != UIO_USERSPACE) {
98 		pbuf = NULL;
99 		pages = NULL;
100 	} else if ((dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed) {
101 		pbuf = NULL;
102 		maxpages = btoc(MIN(uio->uio_resid, MAXPHYS)) + 1;
103 		pages = malloc(sizeof(*pages) * maxpages, M_DEVBUF, M_WAITOK);
104 	} else {
105 		pbuf = getpbuf(NULL);
106 		sa = pbuf->b_data;
107 		maxpages = btoc(MAXPHYS);
108 		pages = pbuf->b_pages;
109 	}
110 	prot = VM_PROT_READ;
111 	if (uio->uio_rw == UIO_READ)
112 		prot |= VM_PROT_WRITE;	/* Less backwards than it looks */
113 	error = 0;
114 	for (i = 0; i < uio->uio_iovcnt; i++) {
115 #ifdef RACCT
116 		if (racct_enable) {
117 			PROC_LOCK(curproc);
118 			if (uio->uio_rw == UIO_READ) {
119 				racct_add_force(curproc, RACCT_READBPS,
120 				    uio->uio_iov[i].iov_len);
121 				racct_add_force(curproc, RACCT_READIOPS, 1);
122 			} else {
123 				racct_add_force(curproc, RACCT_WRITEBPS,
124 				    uio->uio_iov[i].iov_len);
125 				racct_add_force(curproc, RACCT_WRITEIOPS, 1);
126 			}
127 			PROC_UNLOCK(curproc);
128 		}
129 #endif /* RACCT */
130 
131 		while (uio->uio_iov[i].iov_len) {
132 			g_reset_bio(bp);
133 			if (uio->uio_rw == UIO_READ) {
134 				bp->bio_cmd = BIO_READ;
135 				curthread->td_ru.ru_inblock++;
136 			} else {
137 				bp->bio_cmd = BIO_WRITE;
138 				curthread->td_ru.ru_oublock++;
139 			}
140 			bp->bio_offset = uio->uio_offset;
141 			bp->bio_data = uio->uio_iov[i].iov_base;
142 			bp->bio_length = uio->uio_iov[i].iov_len;
143 			if (bp->bio_length > dev->si_iosize_max)
144 				bp->bio_length = dev->si_iosize_max;
145 			if (bp->bio_length > MAXPHYS)
146 				bp->bio_length = MAXPHYS;
147 
148 			/*
149 			 * Make sure the pbuf can map the request.
150 			 * The pbuf has kvasize = MAXPHYS, so a request
151 			 * larger than MAXPHYS - PAGE_SIZE must be
152 			 * page aligned or it will be fragmented.
153 			 */
154 			poff = (vm_offset_t)bp->bio_data & PAGE_MASK;
155 			if (pbuf && bp->bio_length + poff > pbuf->b_kvasize) {
156 				if (dev->si_flags & SI_NOSPLIT) {
157 					uprintf("%s: request ptr %p is not "
158 					    "on a page boundary; cannot split "
159 					    "request\n", devtoname(dev),
160 					    bp->bio_data);
161 					error = EFBIG;
162 					goto doerror;
163 				}
164 				bp->bio_length = pbuf->b_kvasize;
165 				if (poff != 0)
166 					bp->bio_length -= PAGE_SIZE;
167 			}
168 
169 			bp->bio_bcount = bp->bio_length;
170 			bp->bio_dev = dev;
171 
172 			if (pages) {
173 				if ((npages = vm_fault_quick_hold_pages(
174 				    &curproc->p_vmspace->vm_map,
175 				    (vm_offset_t)bp->bio_data, bp->bio_length,
176 				    prot, pages, maxpages)) < 0) {
177 					error = EFAULT;
178 					goto doerror;
179 				}
180 				if (pbuf) {
181 					pmap_qenter((vm_offset_t)sa,
182 					    pages, npages);
183 					bp->bio_data = sa + poff;
184 				} else {
185 					bp->bio_ma = pages;
186 					bp->bio_ma_n = npages;
187 					bp->bio_ma_offset = poff;
188 					bp->bio_data = unmapped_buf;
189 					bp->bio_flags |= BIO_UNMAPPED;
190 				}
191 			}
192 
193 			csw->d_strategy(bp);
194 			if (uio->uio_rw == UIO_READ)
195 				biowait(bp, "physrd");
196 			else
197 				biowait(bp, "physwr");
198 
199 			if (pages) {
200 				if (pbuf)
201 					pmap_qremove((vm_offset_t)sa, npages);
202 				vm_page_unhold_pages(pages, npages);
203 			}
204 
205 			iolen = bp->bio_length - bp->bio_resid;
206 			if (iolen == 0 && !(bp->bio_flags & BIO_ERROR))
207 				goto doerror;	/* EOF */
208 			uio->uio_iov[i].iov_len -= iolen;
209 			uio->uio_iov[i].iov_base =
210 			    (char *)uio->uio_iov[i].iov_base + iolen;
211 			uio->uio_resid -= iolen;
212 			uio->uio_offset += iolen;
213 			if (bp->bio_flags & BIO_ERROR) {
214 				error = bp->bio_error;
215 				goto doerror;
216 			}
217 		}
218 	}
219 doerror:
220 	if (pbuf)
221 		relpbuf(pbuf, NULL);
222 	else if (pages)
223 		free(pages, M_DEVBUF);
224 	g_destroy_bio(bp);
225 	PRELE(curproc);
226 	return (error);
227 }
228