1 /* 2 * Copyright (c) 2004 Matthew Dillon <dillon@backplane.com> 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, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $DragonFly: src/sys/kern/kern_xio.c,v 1.6 2004/06/05 19:57:35 dillon Exp $ 27 */ 28 /* 29 * Kernel XIO interface. An initialized XIO is basically a collection of 30 * appropriately held vm_page_t's. XIO buffers are vmspace agnostic and 31 * can represent userspace or kernelspace buffers, and can be passed to 32 * foreign threads outside of the originating vmspace. XIO buffers are 33 * not mapped into KVM and thus can be manipulated and passed around with 34 * very low overheads. 35 * 36 * The intent is for XIO to be used in the I/O path, VFS, CAPS, and other 37 * places that need to pass (possibly userspace) data between threads. 38 * 39 * TODO: check for busy page when modifying, check writeable. 40 */ 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/malloc.h> 45 #include <sys/proc.h> 46 #include <sys/vmmeter.h> 47 #include <sys/vnode.h> 48 #include <sys/xio.h> 49 #include <sys/sfbuf.h> 50 51 #include <vm/vm.h> 52 #include <vm/vm_param.h> 53 #include <sys/lock.h> 54 #include <vm/vm_kern.h> 55 #include <vm/pmap.h> 56 #include <vm/vm_map.h> 57 #include <vm/vm_object.h> 58 #include <vm/vm_page.h> 59 #include <vm/vm_pageout.h> 60 #include <vm/vm_pager.h> 61 #include <vm/vm_extern.h> 62 #include <vm/vm_page2.h> 63 64 /* 65 * Just do basic initialization of an empty XIO 66 */ 67 void 68 xio_init(xio_t xio) 69 { 70 xio->xio_flags = 0; 71 xio->xio_bytes = 0; 72 xio->xio_error = 0; 73 xio->xio_offset = 0; 74 xio->xio_npages = 0; 75 xio->xio_pages = xio->xio_internal_pages; 76 } 77 78 /* 79 * Initialize an XIO given a userspace buffer. 0 is returned on success, 80 * an error code on failure. The actual number of bytes that could be 81 * accomodated in the XIO will be stored in xio_bytes. 82 * 83 * Note that you cannot legally accessed a previously cached linmap with 84 * a newly initialized xio until after calling xio_linmap(). 85 */ 86 int 87 xio_init_ubuf(xio_t xio, void *ubase, size_t ubytes, int flags) 88 { 89 vm_offset_t addr; 90 vm_paddr_t paddr; 91 vm_page_t m; 92 int i; 93 int n; 94 int s; 95 int vmprot; 96 97 addr = trunc_page((vm_offset_t)ubase); 98 xio->xio_flags = flags; 99 xio->xio_bytes = 0; 100 xio->xio_error = 0; 101 if (ubytes == 0) { 102 xio->xio_offset = 0; 103 xio->xio_npages = 0; 104 } else { 105 vmprot = (flags & XIOF_WRITE) ? VM_PROT_WRITE : VM_PROT_READ; 106 xio->xio_offset = (vm_offset_t)ubase & PAGE_MASK; 107 xio->xio_pages = xio->xio_internal_pages; 108 if ((n = PAGE_SIZE - xio->xio_offset) > ubytes) 109 n = ubytes; 110 for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) { 111 if (vm_fault_quick((caddr_t)addr, vmprot) < 0) 112 break; 113 if ((paddr = pmap_kextract(addr)) == 0) 114 break; 115 s = splvm(); 116 m = PHYS_TO_VM_PAGE(paddr); 117 vm_page_hold(m); 118 splx(s); 119 xio->xio_pages[i] = m; 120 ubytes -= n; 121 xio->xio_bytes += n; 122 if ((n = ubytes) > PAGE_SIZE) 123 n = PAGE_SIZE; 124 addr += PAGE_SIZE; 125 } 126 xio->xio_npages = i; 127 128 /* 129 * If a failure occured clean out what we loaded and return EFAULT. 130 * Return 0 on success. 131 */ 132 if (i < XIO_INTERNAL_PAGES && n) { 133 xio_release(xio); 134 xio->xio_error = EFAULT; 135 } 136 } 137 return(xio->xio_error); 138 } 139 140 /* 141 * Initialize an XIO given a kernelspace buffer. 0 is returned on success, 142 * an error code on failure. The actual number of bytes that could be 143 * accomodated in the XIO will be stored in xio_bytes. 144 * 145 * vmprot is usually either VM_PROT_READ or VM_PROT_WRITE. 146 * 147 * Note that you cannot legally accessed a previously cached linmap with 148 * a newly initialized xio until after calling xio_linmap(). 149 */ 150 int 151 xio_init_kbuf(xio_t xio, void *kbase, size_t kbytes) 152 { 153 vm_offset_t addr; 154 vm_paddr_t paddr; 155 vm_page_t m; 156 int i; 157 int n; 158 int s; 159 160 addr = trunc_page((vm_offset_t)kbase); 161 xio->xio_flags = 0; 162 xio->xio_offset = (vm_offset_t)kbase & PAGE_MASK; 163 xio->xio_bytes = 0; 164 xio->xio_pages = xio->xio_internal_pages; 165 xio->xio_error = 0; 166 if ((n = PAGE_SIZE - xio->xio_offset) > kbytes) 167 n = kbytes; 168 for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) { 169 if ((paddr = pmap_kextract(addr)) == 0) 170 break; 171 s = splvm(); 172 m = PHYS_TO_VM_PAGE(paddr); 173 vm_page_hold(m); 174 splx(s); 175 xio->xio_pages[i] = m; 176 kbytes -= n; 177 xio->xio_bytes += n; 178 if ((n = kbytes) > PAGE_SIZE) 179 n = PAGE_SIZE; 180 addr += PAGE_SIZE; 181 } 182 xio->xio_npages = i; 183 184 /* 185 * If a failure occured clean out what we loaded and return EFAULT. 186 * Return 0 on success. 187 */ 188 if (i < XIO_INTERNAL_PAGES && n) { 189 xio_release(xio); 190 xio->xio_error = EFAULT; 191 } 192 return(xio->xio_error); 193 } 194 195 /* 196 * Cleanup an XIO so it can be destroyed. The pages associated with the 197 * XIO are released. If a linear mapping buffer is active, it will be 198 * unlocked but the mappings will be left intact for optimal reconstitution 199 * in a later xio_linmap() call. 200 * 201 * Note that you cannot legally accessed the linmap on a released XIO. 202 */ 203 void 204 xio_release(xio_t xio) 205 { 206 int i; 207 int s; 208 vm_page_t m; 209 210 s = splvm(); 211 for (i = 0; i < xio->xio_npages; ++i) { 212 m = xio->xio_pages[i]; 213 vm_page_unhold(m); 214 } 215 splx(s); 216 if (xio->xio_flags & XIOF_LINMAP) { 217 xio->xio_flags &= ~XIOF_LINMAP; 218 /* XXX */ 219 } 220 xio->xio_offset = 0; 221 xio->xio_npages = 0; 222 xio->xio_bytes = 0; 223 xio->xio_error = ENOBUFS; 224 } 225 226 /* 227 * Copy data between an XIO and a UIO. If the UIO represents userspace it 228 * must be relative to the current context. Both the UIO and the XIO are 229 * modified, but the XIO's pages are not released when exhausted. 230 * 231 * UIO_READ xio -> uio 232 * UIO_WRITE uio -> xio 233 */ 234 int 235 xio_uio_copy(xio_t xio, struct uio *uio, int *sizep) 236 { 237 int error; 238 int bytes; 239 240 if ((bytes = xio->xio_bytes) > uio->uio_resid) 241 bytes = uio->uio_resid; 242 error = uiomove_fromphys(xio->xio_pages, xio->xio_offset, bytes, uio); 243 if (error == 0) { 244 xio->xio_bytes -= bytes; 245 xio->xio_offset += bytes; 246 *sizep = bytes; 247 } else { 248 *sizep = 0; 249 } 250 return(error); 251 } 252 253 /* 254 * Copy the specified number of bytes from the xio to a userland 255 * buffer. Return an error code or 0 on success. 256 * 257 * The XIO is modified, but the XIO's pages are not released when exhausted. 258 */ 259 int 260 xio_copy_xtou(xio_t xio, void *uptr, int bytes) 261 { 262 int i; 263 int n; 264 int error; 265 int offset; 266 vm_page_t m; 267 struct sf_buf *sf; 268 269 if (bytes > xio->xio_bytes) 270 return(EFAULT); 271 272 offset = xio->xio_offset & PAGE_MASK; 273 if ((n = PAGE_SIZE - offset) > bytes) 274 n = bytes; 275 276 error = 0; 277 for (i = xio->xio_offset >> PAGE_SHIFT; i < xio->xio_npages; ++i) { 278 m = xio->xio_pages[i]; 279 sf = sf_buf_alloc(m, SFBA_QUICK); 280 error = copyout((char *)sf_buf_kva(sf) + offset, uptr, n); 281 sf_buf_free(sf); 282 if (error) 283 break; 284 bytes -= n; 285 xio->xio_bytes -= n; 286 xio->xio_offset += n; 287 uptr = (char *)uptr + n; 288 if (bytes == 0) 289 break; 290 if ((n = bytes) > PAGE_SIZE) 291 n = PAGE_SIZE; 292 offset = 0; 293 } 294 return(error); 295 } 296 297 /* 298 * Copy the specified number of bytes from the xio to a kernel 299 * buffer. Return an error code or 0 on success. 300 * 301 * The XIO is modified, but the XIO's pages are not released when exhausted. 302 */ 303 int 304 xio_copy_xtok(xio_t xio, void *kptr, int bytes) 305 { 306 int i; 307 int n; 308 int error; 309 int offset; 310 vm_page_t m; 311 struct sf_buf *sf; 312 313 if (bytes > xio->xio_bytes) 314 return(EFAULT); 315 316 offset = xio->xio_offset & PAGE_MASK; 317 if ((n = PAGE_SIZE - offset) > bytes) 318 n = bytes; 319 320 error = 0; 321 for (i = xio->xio_offset >> PAGE_SHIFT; i < xio->xio_npages; ++i) { 322 m = xio->xio_pages[i]; 323 sf = sf_buf_alloc(m, SFBA_QUICK); 324 bcopy((char *)sf_buf_kva(sf) + offset, kptr, n); 325 sf_buf_free(sf); 326 bytes -= n; 327 xio->xio_bytes -= n; 328 xio->xio_offset += n; 329 kptr = (char *)kptr + n; 330 if (bytes == 0) 331 break; 332 if ((n = bytes) > PAGE_SIZE) 333 n = PAGE_SIZE; 334 offset = 0; 335 } 336 return(error); 337 } 338 339