sys/kern/kern_xio.c

   1 /*
   2  * Copyright (c) 2003,2004 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   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  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, 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
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/kern/kern_xio.c,v 1.16 2008/05/09 07:24:45 dillon Exp $
  35  */
  36 /*
  37  * Kernel XIO interface.  An initialized XIO is basically a collection of
  38  * appropriately held vm_page_t's.  XIO buffers are vmspace agnostic and
  39  * can represent userspace or kernelspace buffers, and can be passed to
  40  * foreign threads outside of the originating vmspace.  XIO buffers are
  41  * not mapped into KVM and thus can be manipulated and passed around with
  42  * very low overheads.
  43  *
  44  * The intent is for XIO to be used in the I/O path, VFS, CAPS, and other
  45  * places that need to pass (possibly userspace) data between threads.
  46  *
  47  * TODO: check for busy page when modifying, check writeable.
  48  */
  49
  50 #include <sys/param.h>
  51 #include <sys/systm.h>
  52 #include <sys/malloc.h>
  53 #include <sys/proc.h>
  54 #include <sys/vmmeter.h>
  55 #include <sys/vnode.h>
  56 #include <sys/xio.h>
  57
  58 #include <cpu/lwbuf.h>
  59
  60 #include <vm/vm.h>
  61 #include <vm/vm_param.h>
  62 #include <sys/lock.h>
  63 #include <vm/vm_kern.h>
  64 #include <vm/pmap.h>
  65 #include <vm/vm_map.h>
  66 #include <vm/vm_object.h>
  67 #include <vm/vm_page.h>
  68 #include <vm/vm_pageout.h>
  69 #include <vm/vm_pager.h>
  70 #include <vm/vm_extern.h>
  71 #include <vm/vm_page2.h>
  72
  73 /*
  74  * Just do basic initialization of an empty XIO
  75  */
  76 void
  77 xio_init(xio_t xio)
  78 {
  79     xio->xio_flags = 0;
  80     xio->xio_bytes = 0;
  81     xio->xio_error = 0;
  82     xio->xio_offset = 0;
  83     xio->xio_npages = 0;
  84     xio->xio_pages = xio->xio_internal_pages;
  85 }
  86
  87 /*
  88  * Initialize an XIO given a kernelspace buffer.  0 is returned on success,
  89  * an error code on failure.  The actual number of bytes that could be
  90  * accomodated in the XIO will be stored in xio_bytes and the page offset
  91  * will be stored in xio_offset.
  92  *
  93  * WARNING! We cannot map user memory directly into an xio unless we also
  94  *          make the mapping use managed pages, otherwise modifications to
  95  *          the memory will race against pageouts and flushes.
  96  */
  97 int
  98 xio_init_kbuf(xio_t xio, void *kbase, size_t kbytes)
  99 {
 100     vm_offset_t addr;
 101     vm_paddr_t paddr;
 102     vm_page_t m;
 103     int i;
 104     int n;
 105
 106     addr = trunc_page((vm_offset_t)kbase);
 107     xio->xio_flags = 0;
 108     xio->xio_offset = (vm_offset_t)kbase & PAGE_MASK;
 109     xio->xio_bytes = 0;
 110     xio->xio_pages = xio->xio_internal_pages;
 111     xio->xio_error = 0;
 112     if ((n = PAGE_SIZE - xio->xio_offset) > kbytes)
 113         n = kbytes;
 114     for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
 115         if ((paddr = pmap_kextract(addr)) == 0)
 116             break;
 117         m = PHYS_TO_VM_PAGE(paddr);
 118         vm_page_hold(m);
 119         xio->xio_pages[i] = m;
 120         kbytes -= n;
 121         xio->xio_bytes += n;
 122         if ((n = kbytes) > 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     return(xio->xio_error);
 137 }
 138
 139 /*
 140  * Initialize an XIO given an array of vm_page pointers.  The caller is
 141  * responsible for any modified state changes for the pages.
 142  */
 143 int
 144 xio_init_pages(xio_t xio, struct vm_page **mbase, int npages, int xflags)
 145 {
 146     int i;
 147
 148     KKASSERT(npages <= XIO_INTERNAL_PAGES);
 149
 150     xio->xio_flags = xflags;
 151     xio->xio_offset = 0;
 152     xio->xio_bytes = npages * PAGE_SIZE;
 153     xio->xio_pages = xio->xio_internal_pages;
 154     xio->xio_npages = npages;
 155     xio->xio_error = 0;
 156     for (i = 0; i < npages; ++i) {
 157         vm_page_hold(mbase[i]);
 158         xio->xio_pages[i] = mbase[i];
 159     }
 160     return(0);
 161 }
 162
 163 /*
 164  * Cleanup an XIO so it can be destroyed.  The pages associated with the
 165  * XIO are released.
 166  */
 167 void
 168 xio_release(xio_t xio)
 169 {
 170     int i;
 171     vm_page_t m;
 172
 173     for (i = 0; i < xio->xio_npages; ++i) {
 174         m = xio->xio_pages[i];
 175         if (xio->xio_flags & XIOF_WRITE)
 176                 vm_page_dirty(m);
 177         vm_page_unhold(m);
 178     }
 179     xio->xio_offset = 0;
 180     xio->xio_npages = 0;
 181     xio->xio_bytes = 0;
 182     xio->xio_error = ENOBUFS;
 183 }
 184
 185 /*
 186  * Copy data between an XIO and a UIO.  If the UIO represents userspace it
 187  * must be relative to the current context.
 188  *
 189  * uoffset is the abstracted starting offset in the XIO, not the actual
 190  * offset, and usually starts at 0.
 191  *
 192  * The XIO is not modified.  The UIO is updated to reflect the copy.
 193  *
 194  * UIO_READ     xio -> uio
 195  * UIO_WRITE    uio -> xio
 196  */
 197 int
 198 xio_uio_copy(xio_t xio, int uoffset, struct uio *uio, size_t *sizep)
 199 {
 200     size_t bytes;
 201     int error;
 202
 203     bytes = xio->xio_bytes - uoffset;
 204     if (bytes > uio->uio_resid)
 205         bytes = uio->uio_resid;
 206     KKASSERT(bytes >= 0);
 207     error = uiomove_fromphys(xio->xio_pages, xio->xio_offset + uoffset,
 208                                 bytes, uio);
 209     if (error == 0)
 210         *sizep = bytes;
 211     else
 212         *sizep = 0;
 213     return(error);
 214 }
 215
 216 /*
 217  * Copy the specified number of bytes from the xio to a userland
 218  * buffer.  Return an error code or 0 on success.
 219  *
 220  * uoffset is the abstracted starting offset in the XIO, not the actual
 221  * offset, and usually starts at 0.
 222  *
 223  * The XIO is not modified.
 224  */
 225 int
 226 xio_copy_xtou(xio_t xio, int uoffset, void *uptr, int bytes)
 227 {
 228     int i;
 229     int n;
 230     int error;
 231     int offset;
 232     vm_page_t m;
 233     struct lwbuf *lwb;
 234     struct lwbuf lwb_cache;
 235
 236     if (uoffset + bytes > xio->xio_bytes)
 237         return(EFAULT);
 238
 239     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 240     if ((n = PAGE_SIZE - offset) > bytes)
 241         n = bytes;
 242
 243     error = 0;
 244     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 245          i < xio->xio_npages;
 246          ++i
 247     ) {
 248         m = xio->xio_pages[i];
 249         lwb = lwbuf_alloc(m, &lwb_cache);
 250         error = copyout((char *)lwbuf_kva(lwb) + offset, uptr, n);
 251         lwbuf_free(lwb);
 252         if (error)
 253             break;
 254         bytes -= n;
 255         uptr = (char *)uptr + n;
 256         if (bytes == 0)
 257             break;
 258         if ((n = bytes) > PAGE_SIZE)
 259             n = PAGE_SIZE;
 260         offset = 0;
 261     }
 262     return(error);
 263 }
 264
 265 /*
 266  * Copy the specified number of bytes from the xio to a kernel
 267  * buffer.  Return an error code or 0 on success.
 268  *
 269  * uoffset is the abstracted starting offset in the XIO, not the actual
 270  * offset, and usually starts at 0.
 271  *
 272  * The XIO is not modified.
 273  */
 274 int
 275 xio_copy_xtok(xio_t xio, int uoffset, void *kptr, int bytes)
 276 {
 277     int i;
 278     int n;
 279     int error;
 280     int offset;
 281     vm_page_t m;
 282     struct lwbuf *lwb;
 283     struct lwbuf lwb_cache;
 284
 285     if (bytes + uoffset > xio->xio_bytes)
 286         return(EFAULT);
 287
 288     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 289     if ((n = PAGE_SIZE - offset) > bytes)
 290         n = bytes;
 291
 292     error = 0;
 293     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 294          i < xio->xio_npages;
 295          ++i
 296     ) {
 297         m = xio->xio_pages[i];
 298         lwb = lwbuf_alloc(m, &lwb_cache);
 299         bcopy((char *)lwbuf_kva(lwb) + offset, kptr, n);
 300         lwbuf_free(lwb);
 301         bytes -= n;
 302         kptr = (char *)kptr + n;
 303         if (bytes == 0)
 304             break;
 305         if ((n = bytes) > PAGE_SIZE)
 306             n = PAGE_SIZE;
 307         offset = 0;
 308     }
 309     return(error);
 310 }
 311
 312 /*
 313  * Copy the specified number of bytes from userland to the xio.
 314  * Return an error code or 0 on success.
 315  *
 316  * uoffset is the abstracted starting offset in the XIO, not the actual
 317  * offset, and usually starts at 0.
 318  *
 319  * Data in pages backing the XIO will be modified.
 320  */
 321 int
 322 xio_copy_utox(xio_t xio, int uoffset, const void *uptr, int bytes)
 323 {
 324     int i;
 325     int n;
 326     int error;
 327     int offset;
 328     vm_page_t m;
 329     struct lwbuf *lwb;
 330     struct lwbuf lwb_cache;
 331
 332     if (uoffset + bytes > xio->xio_bytes)
 333         return(EFAULT);
 334
 335     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 336     if ((n = PAGE_SIZE - offset) > bytes)
 337         n = bytes;
 338
 339     error = 0;
 340     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 341          i < xio->xio_npages;
 342          ++i
 343     ) {
 344         m = xio->xio_pages[i];
 345         lwb = lwbuf_alloc(m, &lwb_cache);
 346         error = copyin(uptr, (char *)lwbuf_kva(lwb) + offset, n);
 347         lwbuf_free(lwb);
 348         if (error)
 349             break;
 350         bytes -= n;
 351         uptr = (const char *)uptr + n;
 352         if (bytes == 0)
 353             break;
 354         if ((n = bytes) > PAGE_SIZE)
 355             n = PAGE_SIZE;
 356         offset = 0;
 357     }
 358     return(error);
 359 }
 360
 361 /*
 362  * Copy the specified number of bytes from the kernel to the xio.
 363  * Return an error code or 0 on success.
 364  *
 365  * uoffset is the abstracted starting offset in the XIO, not the actual
 366  * offset, and usually starts at 0.
 367  *
 368  * Data in pages backing the XIO will be modified.
 369  */
 370 int
 371 xio_copy_ktox(xio_t xio, int uoffset, const void *kptr, int bytes)
 372 {
 373     int i;
 374     int n;
 375     int error;
 376     int offset;
 377     vm_page_t m;
 378     struct lwbuf *lwb;
 379     struct lwbuf lwb_cache;
 380
 381     if (uoffset + bytes > xio->xio_bytes)
 382         return(EFAULT);
 383
 384     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 385     if ((n = PAGE_SIZE - offset) > bytes)
 386         n = bytes;
 387
 388     error = 0;
 389     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 390          i < xio->xio_npages;
 391          ++i
 392     ) {
 393         m = xio->xio_pages[i];
 394         lwb = lwbuf_alloc(m, &lwb_cache);
 395         bcopy(kptr, (char *)lwbuf_kva(lwb) + offset, n);
 396         lwbuf_free(lwb);
 397         bytes -= n;
 398         kptr = (const char *)kptr + n;
 399         if (bytes == 0)
 400             break;
 401         if ((n = bytes) > PAGE_SIZE)
 402             n = PAGE_SIZE;
 403         offset = 0;
 404     }
 405     return(error);
 406 }