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 /*
  35  * Kernel XIO interface.  An initialized XIO is basically a collection of
  36  * appropriately held vm_page_t's.  XIO buffers are vmspace agnostic and
  37  * can represent userspace or kernelspace buffers, and can be passed to
  38  * foreign threads outside of the originating vmspace.  XIO buffers are
  39  * not mapped into KVM and thus can be manipulated and passed around with
  40  * very low overheads.
  41  *
  42  * The intent is for XIO to be used in the I/O path, VFS, CAPS, and other
  43  * places that need to pass (possibly userspace) data between threads.
  44  *
  45  * TODO: check for busy page when modifying, check writeable.
  46  */
  47
  48 #include <sys/param.h>
  49 #include <sys/systm.h>
  50 #include <sys/uio.h>
  51 #include <sys/malloc.h>
  52 #include <sys/proc.h>
  53 #include <sys/vmmeter.h>
  54 #include <sys/vnode.h>
  55 #include <sys/xio.h>
  56
  57 #include <cpu/lwbuf.h>
  58
  59 #include <vm/vm.h>
  60 #include <vm/vm_param.h>
  61 #include <sys/lock.h>
  62 #include <vm/vm_kern.h>
  63 #include <vm/pmap.h>
  64 #include <vm/vm_map.h>
  65 #include <vm/vm_object.h>
  66 #include <vm/vm_page.h>
  67 #include <vm/vm_pageout.h>
  68 #include <vm/vm_pager.h>
  69 #include <vm/vm_extern.h>
  70 #include <vm/vm_page2.h>
  71
  72 /*
  73  * Just do basic initialization of an empty XIO
  74  */
  75 void
  76 xio_init(xio_t xio)
  77 {
  78     xio->xio_flags = 0;
  79     xio->xio_bytes = 0;
  80     xio->xio_error = 0;
  81     xio->xio_offset = 0;
  82     xio->xio_npages = 0;
  83     xio->xio_pages = xio->xio_internal_pages;
  84 }
  85
  86 /*
  87  * Initialize an XIO given a kernelspace buffer.  0 is returned on success,
  88  * an error code on failure.  The actual number of bytes that could be
  89  * accomodated in the XIO will be stored in xio_bytes and the page offset
  90  * will be stored in xio_offset.
  91  *
  92  * WARNING! We cannot map user memory directly into an xio unless we also
  93  *          make the mapping use managed pages, otherwise modifications to
  94  *          the memory will race against pageouts and flushes.
  95  */
  96 int
  97 xio_init_kbuf(xio_t xio, void *kbase, size_t kbytes)
  98 {
  99     vm_offset_t addr;
 100     vm_paddr_t paddr;
 101     vm_page_t m;
 102     int i;
 103     int n;
 104
 105     addr = trunc_page((vm_offset_t)kbase);
 106     xio->xio_flags = 0;
 107     xio->xio_offset = (vm_offset_t)kbase & PAGE_MASK;
 108     xio->xio_bytes = 0;
 109     xio->xio_pages = xio->xio_internal_pages;
 110     xio->xio_error = 0;
 111     if ((n = PAGE_SIZE - xio->xio_offset) > kbytes)
 112         n = kbytes;
 113     for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
 114         if ((paddr = pmap_kextract(addr)) == 0)
 115             break;
 116         m = PHYS_TO_VM_PAGE(paddr);
 117         vm_page_hold(m);
 118         xio->xio_pages[i] = m;
 119         kbytes -= n;
 120         xio->xio_bytes += n;
 121         if ((n = kbytes) > PAGE_SIZE)
 122             n = PAGE_SIZE;
 123         addr += PAGE_SIZE;
 124     }
 125     xio->xio_npages = i;
 126
 127     /*
 128      * If a failure occured clean out what we loaded and return EFAULT.
 129      * Return 0 on success.
 130      */
 131     if (i < XIO_INTERNAL_PAGES && n) {
 132         xio_release(xio);
 133         xio->xio_error = EFAULT;
 134     }
 135     return(xio->xio_error);
 136 }
 137
 138 /*
 139  * Initialize an XIO given an array of vm_page pointers.  The caller is
 140  * responsible for any modified state changes for the pages.
 141  */
 142 int
 143 xio_init_pages(xio_t xio, struct vm_page **mbase, int npages, int xflags)
 144 {
 145     int i;
 146
 147     KKASSERT(npages <= XIO_INTERNAL_PAGES);
 148
 149     xio->xio_flags = xflags;
 150     xio->xio_offset = 0;
 151     xio->xio_bytes = npages * PAGE_SIZE;
 152     xio->xio_pages = xio->xio_internal_pages;
 153     xio->xio_npages = npages;
 154     xio->xio_error = 0;
 155     for (i = 0; i < npages; ++i) {
 156         vm_page_hold(mbase[i]);
 157         xio->xio_pages[i] = mbase[i];
 158     }
 159     return(0);
 160 }
 161
 162 /*
 163  * Cleanup an XIO so it can be destroyed.  The pages associated with the
 164  * XIO are released.
 165  */
 166 void
 167 xio_release(xio_t xio)
 168 {
 169     int i;
 170     vm_page_t m;
 171
 172     for (i = 0; i < xio->xio_npages; ++i) {
 173         m = xio->xio_pages[i];
 174         if (xio->xio_flags & XIOF_WRITE)
 175                 vm_page_dirty(m);
 176         vm_page_unhold(m);
 177     }
 178     xio->xio_offset = 0;
 179     xio->xio_npages = 0;
 180     xio->xio_bytes = 0;
 181     xio->xio_error = ENOBUFS;
 182 }
 183
 184 /*
 185  * Copy data between an XIO and a UIO.  If the UIO represents userspace it
 186  * must be relative to the current context.
 187  *
 188  * uoffset is the abstracted starting offset in the XIO, not the actual
 189  * offset, and usually starts at 0.
 190  *
 191  * The XIO is not modified.  The UIO is updated to reflect the copy.
 192  *
 193  * UIO_READ     xio -> uio
 194  * UIO_WRITE    uio -> xio
 195  */
 196 int
 197 xio_uio_copy(xio_t xio, int uoffset, struct uio *uio, size_t *sizep)
 198 {
 199     size_t bytes;
 200     int error;
 201
 202     bytes = xio->xio_bytes - uoffset;
 203     if (bytes > uio->uio_resid)
 204         bytes = uio->uio_resid;
 205     KKASSERT(bytes >= 0);
 206     error = uiomove_fromphys(xio->xio_pages, xio->xio_offset + uoffset,
 207                                 bytes, uio);
 208     if (error == 0)
 209         *sizep = bytes;
 210     else
 211         *sizep = 0;
 212     return(error);
 213 }
 214
 215 /*
 216  * Copy the specified number of bytes from the xio to a userland
 217  * buffer.  Return an error code or 0 on success.
 218  *
 219  * uoffset is the abstracted starting offset in the XIO, not the actual
 220  * offset, and usually starts at 0.
 221  *
 222  * The XIO is not modified.
 223  */
 224 int
 225 xio_copy_xtou(xio_t xio, int uoffset, void *uptr, int bytes)
 226 {
 227     int i;
 228     int n;
 229     int error;
 230     int offset;
 231     vm_page_t m;
 232     struct lwbuf *lwb;
 233     struct lwbuf lwb_cache;
 234
 235     if (uoffset + bytes > xio->xio_bytes)
 236         return(EFAULT);
 237
 238     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 239     if ((n = PAGE_SIZE - offset) > bytes)
 240         n = bytes;
 241
 242     error = 0;
 243     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 244          i < xio->xio_npages;
 245          ++i
 246     ) {
 247         m = xio->xio_pages[i];
 248         lwb = lwbuf_alloc(m, &lwb_cache);
 249         error = copyout((char *)lwbuf_kva(lwb) + offset, uptr, n);
 250         lwbuf_free(lwb);
 251         if (error)
 252             break;
 253         bytes -= n;
 254         uptr = (char *)uptr + n;
 255         if (bytes == 0)
 256             break;
 257         if ((n = bytes) > PAGE_SIZE)
 258             n = PAGE_SIZE;
 259         offset = 0;
 260     }
 261     return(error);
 262 }
 263
 264 /*
 265  * Copy the specified number of bytes from the xio to a kernel
 266  * buffer.  Return an error code or 0 on success.
 267  *
 268  * uoffset is the abstracted starting offset in the XIO, not the actual
 269  * offset, and usually starts at 0.
 270  *
 271  * The XIO is not modified.
 272  */
 273 int
 274 xio_copy_xtok(xio_t xio, int uoffset, void *kptr, int bytes)
 275 {
 276     int i;
 277     int n;
 278     int error;
 279     int offset;
 280     vm_page_t m;
 281     struct lwbuf *lwb;
 282     struct lwbuf lwb_cache;
 283
 284     if (bytes + uoffset > xio->xio_bytes)
 285         return(EFAULT);
 286
 287     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 288     if ((n = PAGE_SIZE - offset) > bytes)
 289         n = bytes;
 290
 291     error = 0;
 292     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 293          i < xio->xio_npages;
 294          ++i
 295     ) {
 296         m = xio->xio_pages[i];
 297         lwb = lwbuf_alloc(m, &lwb_cache);
 298         bcopy((char *)lwbuf_kva(lwb) + offset, kptr, n);
 299         lwbuf_free(lwb);
 300         bytes -= n;
 301         kptr = (char *)kptr + n;
 302         if (bytes == 0)
 303             break;
 304         if ((n = bytes) > PAGE_SIZE)
 305             n = PAGE_SIZE;
 306         offset = 0;
 307     }
 308     return(error);
 309 }
 310
 311 /*
 312  * Copy the specified number of bytes from userland to the xio.
 313  * Return an error code or 0 on success.
 314  *
 315  * uoffset is the abstracted starting offset in the XIO, not the actual
 316  * offset, and usually starts at 0.
 317  *
 318  * Data in pages backing the XIO will be modified.
 319  */
 320 int
 321 xio_copy_utox(xio_t xio, int uoffset, const void *uptr, int bytes)
 322 {
 323     int i;
 324     int n;
 325     int error;
 326     int offset;
 327     vm_page_t m;
 328     struct lwbuf *lwb;
 329     struct lwbuf lwb_cache;
 330
 331     if (uoffset + bytes > xio->xio_bytes)
 332         return(EFAULT);
 333
 334     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 335     if ((n = PAGE_SIZE - offset) > bytes)
 336         n = bytes;
 337
 338     error = 0;
 339     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 340          i < xio->xio_npages;
 341          ++i
 342     ) {
 343         m = xio->xio_pages[i];
 344         lwb = lwbuf_alloc(m, &lwb_cache);
 345         error = copyin(uptr, (char *)lwbuf_kva(lwb) + offset, n);
 346         lwbuf_free(lwb);
 347         if (error)
 348             break;
 349         bytes -= n;
 350         uptr = (const char *)uptr + n;
 351         if (bytes == 0)
 352             break;
 353         if ((n = bytes) > PAGE_SIZE)
 354             n = PAGE_SIZE;
 355         offset = 0;
 356     }
 357     return(error);
 358 }
 359
 360 /*
 361  * Copy the specified number of bytes from the kernel to the xio.
 362  * Return an error code or 0 on success.
 363  *
 364  * uoffset is the abstracted starting offset in the XIO, not the actual
 365  * offset, and usually starts at 0.
 366  *
 367  * Data in pages backing the XIO will be modified.
 368  */
 369 int
 370 xio_copy_ktox(xio_t xio, int uoffset, const void *kptr, int bytes)
 371 {
 372     int i;
 373     int n;
 374     int error;
 375     int offset;
 376     vm_page_t m;
 377     struct lwbuf *lwb;
 378     struct lwbuf lwb_cache;
 379
 380     if (uoffset + bytes > xio->xio_bytes)
 381         return(EFAULT);
 382
 383     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 384     if ((n = PAGE_SIZE - offset) > bytes)
 385         n = bytes;
 386
 387     error = 0;
 388     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 389          i < xio->xio_npages;
 390          ++i
 391     ) {
 392         m = xio->xio_pages[i];
 393         lwb = lwbuf_alloc(m, &lwb_cache);
 394         bcopy(kptr, (char *)lwbuf_kva(lwb) + offset, n);
 395         lwbuf_free(lwb);
 396         bytes -= n;
 397         kptr = (const char *)kptr + n;
 398         if (bytes == 0)
 399             break;
 400         if ((n = bytes) > PAGE_SIZE)
 401             n = PAGE_SIZE;
 402         offset = 0;
 403     }
 404     return(error);
 405 }