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 char *base, *sa; 49 u_int iolen, poff; 50 int error, i, npages, maxpages; 51 vm_prot_t prot; 52 53 csw = dev->si_devsw; 54 npages = 0; 55 sa = NULL; 56 /* check if character device is being destroyed */ 57 if (csw == NULL) 58 return (ENXIO); 59 60 /* XXX: sanity check */ 61 if(dev->si_iosize_max < PAGE_SIZE) { 62 printf("WARNING: %s si_iosize_max=%d, using DFLTPHYS.\n", 63 devtoname(dev), dev->si_iosize_max); 64 dev->si_iosize_max = DFLTPHYS; 65 } 66 67 /* 68 * If the driver does not want I/O to be split, that means that we 69 * need to reject any requests that will not fit into one buffer. 70 */ 71 if (dev->si_flags & SI_NOSPLIT && 72 (uio->uio_resid > dev->si_iosize_max || uio->uio_resid > MAXPHYS || 73 uio->uio_iovcnt > 1)) { 74 /* 75 * Tell the user why his I/O was rejected. 76 */ 77 if (uio->uio_resid > dev->si_iosize_max) 78 uprintf("%s: request size=%zd > si_iosize_max=%d; " 79 "cannot split request\n", devtoname(dev), 80 uio->uio_resid, dev->si_iosize_max); 81 if (uio->uio_resid > MAXPHYS) 82 uprintf("%s: request size=%zd > MAXPHYS=%d; " 83 "cannot split request\n", devtoname(dev), 84 uio->uio_resid, MAXPHYS); 85 if (uio->uio_iovcnt > 1) 86 uprintf("%s: request vectors=%d > 1; " 87 "cannot split request\n", devtoname(dev), 88 uio->uio_iovcnt); 89 return (EFBIG); 90 } 91 92 /* 93 * Keep the process UPAGES from being swapped. Processes swapped 94 * out while holding pbufs, used by swapper, may lead to deadlock. 95 */ 96 PHOLD(curproc); 97 98 bp = g_alloc_bio(); 99 if (uio->uio_segflg != UIO_USERSPACE) { 100 pbuf = NULL; 101 pages = NULL; 102 } else if ((dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed) { 103 pbuf = NULL; 104 maxpages = btoc(MIN(uio->uio_resid, MAXPHYS)) + 1; 105 pages = malloc(sizeof(*pages) * maxpages, M_DEVBUF, M_WAITOK); 106 } else { 107 pbuf = uma_zalloc(pbuf_zone, M_WAITOK); 108 sa = pbuf->b_data; 109 maxpages = btoc(MAXPHYS); 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; 149 150 /* 151 * Make sure the pbuf can map the request. 152 * The pbuf has kvasize = MAXPHYS, so a request 153 * larger than MAXPHYS - PAGE_SIZE must be 154 * page aligned or it will be fragmented. 155 */ 156 poff = (vm_offset_t)base & PAGE_MASK; 157 if (pbuf && bp->bio_length + poff > pbuf->b_kvasize) { 158 if (dev->si_flags & SI_NOSPLIT) { 159 uprintf("%s: request ptr %p is not " 160 "on a page boundary; cannot split " 161 "request\n", devtoname(dev), 162 base); 163 error = EFBIG; 164 goto doerror; 165 } 166 bp->bio_length = pbuf->b_kvasize; 167 if (poff != 0) 168 bp->bio_length -= PAGE_SIZE; 169 } 170 171 bp->bio_bcount = bp->bio_length; 172 bp->bio_dev = dev; 173 174 if (pages) { 175 if ((npages = vm_fault_quick_hold_pages( 176 &curproc->p_vmspace->vm_map, 177 (vm_offset_t)base, bp->bio_length, 178 prot, pages, maxpages)) < 0) { 179 error = EFAULT; 180 goto doerror; 181 } 182 if (pbuf && sa) { 183 pmap_qenter((vm_offset_t)sa, 184 pages, npages); 185 bp->bio_data = sa + poff; 186 } else { 187 bp->bio_ma = pages; 188 bp->bio_ma_n = npages; 189 bp->bio_ma_offset = poff; 190 bp->bio_data = unmapped_buf; 191 bp->bio_flags |= BIO_UNMAPPED; 192 } 193 } else 194 bp->bio_data = base; 195 196 csw->d_strategy(bp); 197 if (uio->uio_rw == UIO_READ) 198 biowait(bp, "physrd"); 199 else 200 biowait(bp, "physwr"); 201 202 if (pages) { 203 if (pbuf) 204 pmap_qremove((vm_offset_t)sa, npages); 205 vm_page_unhold_pages(pages, npages); 206 } 207 208 iolen = bp->bio_length - bp->bio_resid; 209 if (iolen == 0 && !(bp->bio_flags & BIO_ERROR)) 210 goto doerror; /* EOF */ 211 uio->uio_iov[i].iov_len -= iolen; 212 uio->uio_iov[i].iov_base = 213 (char *)uio->uio_iov[i].iov_base + iolen; 214 uio->uio_resid -= iolen; 215 uio->uio_offset += iolen; 216 if (bp->bio_flags & BIO_ERROR) { 217 error = bp->bio_error; 218 goto doerror; 219 } 220 } 221 } 222 doerror: 223 if (pbuf) 224 uma_zfree(pbuf_zone, pbuf); 225 else if (pages) 226 free(pages, M_DEVBUF); 227 g_destroy_bio(bp); 228 PRELE(curproc); 229 return (error); 230 } 231