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 userspace 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 int
  94 xio_init_ubuf(xio_t xio, void *ubase, size_t ubytes, int flags)
  95 {
  96     vm_offset_t addr;
  97     vm_page_t m;
  98     vm_page_t m0;
  99     int error;
 100     int i;
 101     int n;
 102     int vmprot;
 103
 104     addr = trunc_page((vm_offset_t)ubase);
 105     xio->xio_flags = flags;
 106     xio->xio_bytes = 0;
 107     xio->xio_error = 0;
 108     if (ubytes == 0) {
 109         xio->xio_offset = 0;
 110         xio->xio_npages = 0;
 111     } else {
 112         vmprot = (flags & XIOF_WRITE) ? VM_PROT_WRITE : VM_PROT_READ;
 113         xio->xio_offset = (vm_offset_t)ubase & PAGE_MASK;
 114         xio->xio_pages = xio->xio_internal_pages;
 115         if ((n = PAGE_SIZE - xio->xio_offset) > ubytes)
 116             n = ubytes;
 117         m0 = NULL;
 118         for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
 119             m = vm_fault_page_quick(addr, vmprot, &error);
 120             if (m == NULL)
 121                 break;
 122             xio->xio_pages[i] = m;
 123             ubytes -= n;
 124             xio->xio_bytes += n;
 125             if ((n = ubytes) > PAGE_SIZE)
 126                 n = PAGE_SIZE;
 127             addr += PAGE_SIZE;
 128
 129             /*
 130              * Check linearity, used by syslink to memory map DMA buffers.
 131              */
 132             if (flags & XIOF_VMLINEAR) {
 133                 if (i == 0) {
 134                     m0 = m;
 135                 } else
 136                 if (m->object != m0->object || m->pindex != m0->pindex + i) {
 137                     error = EINVAL;
 138                     break;
 139                 }
 140             }
 141         }
 142         xio->xio_npages = i;
 143
 144         /*
 145          * If a failure occured clean out what we loaded and return EFAULT.
 146          * Return 0 on success.  Do not dirty the pages.
 147          */
 148         if (i < XIO_INTERNAL_PAGES && n) {
 149             xio->xio_flags &= ~XIOF_WRITE;
 150             xio_release(xio);
 151             xio->xio_error = EFAULT;
 152         }
 153     }
 154     return(xio->xio_error);
 155 }
 156
 157 /*
 158  * Initialize an XIO given a kernelspace buffer.  0 is returned on success,
 159  * an error code on failure.  The actual number of bytes that could be
 160  * accomodated in the XIO will be stored in xio_bytes and the page offset
 161  * will be stored in xio_offset.
 162  */
 163 int
 164 xio_init_kbuf(xio_t xio, void *kbase, size_t kbytes)
 165 {
 166     vm_offset_t addr;
 167     vm_paddr_t paddr;
 168     vm_page_t m;
 169     int i;
 170     int n;
 171
 172     addr = trunc_page((vm_offset_t)kbase);
 173     xio->xio_flags = 0;
 174     xio->xio_offset = (vm_offset_t)kbase & PAGE_MASK;
 175     xio->xio_bytes = 0;
 176     xio->xio_pages = xio->xio_internal_pages;
 177     xio->xio_error = 0;
 178     if ((n = PAGE_SIZE - xio->xio_offset) > kbytes)
 179         n = kbytes;
 180     for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
 181         if ((paddr = pmap_kextract(addr)) == 0)
 182             break;
 183         m = PHYS_TO_VM_PAGE(paddr);
 184         vm_page_hold(m);
 185         xio->xio_pages[i] = m;
 186         kbytes -= n;
 187         xio->xio_bytes += n;
 188         if ((n = kbytes) > PAGE_SIZE)
 189             n = PAGE_SIZE;
 190         addr += PAGE_SIZE;
 191     }
 192     xio->xio_npages = i;
 193
 194     /*
 195      * If a failure occured clean out what we loaded and return EFAULT.
 196      * Return 0 on success.
 197      */
 198     if (i < XIO_INTERNAL_PAGES && n) {
 199         xio_release(xio);
 200         xio->xio_error = EFAULT;
 201     }
 202     return(xio->xio_error);
 203 }
 204
 205 /*
 206  * Initialize an XIO given an array of vm_page pointers.  The caller is
 207  * responsible for any modified state changes for the pages.
 208  */
 209 int
 210 xio_init_pages(xio_t xio, struct vm_page **mbase, int npages, int xflags)
 211 {
 212     int i;
 213
 214     KKASSERT(npages <= XIO_INTERNAL_PAGES);
 215
 216     xio->xio_flags = xflags;
 217     xio->xio_offset = 0;
 218     xio->xio_bytes = npages * PAGE_SIZE;
 219     xio->xio_pages = xio->xio_internal_pages;
 220     xio->xio_npages = npages;
 221     xio->xio_error = 0;
 222     for (i = 0; i < npages; ++i) {
 223         vm_page_hold(mbase[i]);
 224         xio->xio_pages[i] = mbase[i];
 225     }
 226     return(0);
 227 }
 228
 229 /*
 230  * Cleanup an XIO so it can be destroyed.  The pages associated with the
 231  * XIO are released.
 232  */
 233 void
 234 xio_release(xio_t xio)
 235 {
 236     int i;
 237     vm_page_t m;
 238
 239     for (i = 0; i < xio->xio_npages; ++i) {
 240         m = xio->xio_pages[i];
 241         if (xio->xio_flags & XIOF_WRITE)
 242                 vm_page_dirty(m);
 243         vm_page_unhold(m);
 244     }
 245     xio->xio_offset = 0;
 246     xio->xio_npages = 0;
 247     xio->xio_bytes = 0;
 248     xio->xio_error = ENOBUFS;
 249 }
 250
 251 /*
 252  * Copy data between an XIO and a UIO.  If the UIO represents userspace it
 253  * must be relative to the current context.
 254  *
 255  * uoffset is the abstracted starting offset in the XIO, not the actual
 256  * offset, and usually starts at 0.
 257  *
 258  * The XIO is not modified.  The UIO is updated to reflect the copy.
 259  *
 260  * UIO_READ     xio -> uio
 261  * UIO_WRITE    uio -> xio
 262  */
 263 int
 264 xio_uio_copy(xio_t xio, int uoffset, struct uio *uio, size_t *sizep)
 265 {
 266     size_t bytes;
 267     int error;
 268
 269     bytes = xio->xio_bytes - uoffset;
 270     if (bytes > uio->uio_resid)
 271         bytes = uio->uio_resid;
 272     KKASSERT(bytes >= 0);
 273     error = uiomove_fromphys(xio->xio_pages, xio->xio_offset + uoffset,
 274                                 bytes, uio);
 275     if (error == 0)
 276         *sizep = bytes;
 277     else
 278         *sizep = 0;
 279     return(error);
 280 }
 281
 282 /*
 283  * Copy the specified number of bytes from the xio to a userland
 284  * buffer.  Return an error code or 0 on success.
 285  *
 286  * uoffset is the abstracted starting offset in the XIO, not the actual
 287  * offset, and usually starts at 0.
 288  *
 289  * The XIO is not modified.
 290  */
 291 int
 292 xio_copy_xtou(xio_t xio, int uoffset, void *uptr, int bytes)
 293 {
 294     int i;
 295     int n;
 296     int error;
 297     int offset;
 298     vm_page_t m;
 299     struct lwbuf *lwb;
 300     struct lwbuf lwb_cache;
 301
 302     if (uoffset + bytes > xio->xio_bytes)
 303         return(EFAULT);
 304
 305     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 306     if ((n = PAGE_SIZE - offset) > bytes)
 307         n = bytes;
 308
 309     error = 0;
 310     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 311          i < xio->xio_npages;
 312          ++i
 313     ) {
 314         m = xio->xio_pages[i];
 315         lwb = lwbuf_alloc(m, &lwb_cache);
 316         error = copyout((char *)lwbuf_kva(lwb) + offset, uptr, n);
 317         lwbuf_free(lwb);
 318         if (error)
 319             break;
 320         bytes -= n;
 321         uptr = (char *)uptr + n;
 322         if (bytes == 0)
 323             break;
 324         if ((n = bytes) > PAGE_SIZE)
 325             n = PAGE_SIZE;
 326         offset = 0;
 327     }
 328     return(error);
 329 }
 330
 331 /*
 332  * Copy the specified number of bytes from the xio to a kernel
 333  * buffer.  Return an error code or 0 on success.
 334  *
 335  * uoffset is the abstracted starting offset in the XIO, not the actual
 336  * offset, and usually starts at 0.
 337  *
 338  * The XIO is not modified.
 339  */
 340 int
 341 xio_copy_xtok(xio_t xio, int uoffset, void *kptr, int bytes)
 342 {
 343     int i;
 344     int n;
 345     int error;
 346     int offset;
 347     vm_page_t m;
 348     struct lwbuf *lwb;
 349     struct lwbuf lwb_cache;
 350
 351     if (bytes + uoffset > xio->xio_bytes)
 352         return(EFAULT);
 353
 354     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 355     if ((n = PAGE_SIZE - offset) > bytes)
 356         n = bytes;
 357
 358     error = 0;
 359     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 360          i < xio->xio_npages;
 361          ++i
 362     ) {
 363         m = xio->xio_pages[i];
 364         lwb = lwbuf_alloc(m, &lwb_cache);
 365         bcopy((char *)lwbuf_kva(lwb) + offset, kptr, n);
 366         lwbuf_free(lwb);
 367         bytes -= n;
 368         kptr = (char *)kptr + n;
 369         if (bytes == 0)
 370             break;
 371         if ((n = bytes) > PAGE_SIZE)
 372             n = PAGE_SIZE;
 373         offset = 0;
 374     }
 375     return(error);
 376 }
 377
 378 /*
 379  * Copy the specified number of bytes from userland to the xio.
 380  * Return an error code or 0 on success.
 381  *
 382  * uoffset is the abstracted starting offset in the XIO, not the actual
 383  * offset, and usually starts at 0.
 384  *
 385  * Data in pages backing the XIO will be modified.
 386  */
 387 int
 388 xio_copy_utox(xio_t xio, int uoffset, const void *uptr, int bytes)
 389 {
 390     int i;
 391     int n;
 392     int error;
 393     int offset;
 394     vm_page_t m;
 395     struct lwbuf *lwb;
 396     struct lwbuf lwb_cache;
 397
 398     if (uoffset + bytes > xio->xio_bytes)
 399         return(EFAULT);
 400
 401     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 402     if ((n = PAGE_SIZE - offset) > bytes)
 403         n = bytes;
 404
 405     error = 0;
 406     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 407          i < xio->xio_npages;
 408          ++i
 409     ) {
 410         m = xio->xio_pages[i];
 411         lwb = lwbuf_alloc(m, &lwb_cache);
 412         error = copyin(uptr, (char *)lwbuf_kva(lwb) + offset, n);
 413         lwbuf_free(lwb);
 414         if (error)
 415             break;
 416         bytes -= n;
 417         uptr = (const char *)uptr + n;
 418         if (bytes == 0)
 419             break;
 420         if ((n = bytes) > PAGE_SIZE)
 421             n = PAGE_SIZE;
 422         offset = 0;
 423     }
 424     return(error);
 425 }
 426
 427 /*
 428  * Copy the specified number of bytes from the kernel to the xio.
 429  * Return an error code or 0 on success.
 430  *
 431  * uoffset is the abstracted starting offset in the XIO, not the actual
 432  * offset, and usually starts at 0.
 433  *
 434  * Data in pages backing the XIO will be modified.
 435  */
 436 int
 437 xio_copy_ktox(xio_t xio, int uoffset, const void *kptr, int bytes)
 438 {
 439     int i;
 440     int n;
 441     int error;
 442     int offset;
 443     vm_page_t m;
 444     struct lwbuf *lwb;
 445     struct lwbuf lwb_cache;
 446
 447     if (uoffset + bytes > xio->xio_bytes)
 448         return(EFAULT);
 449
 450     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 451     if ((n = PAGE_SIZE - offset) > bytes)
 452         n = bytes;
 453
 454     error = 0;
 455     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 456          i < xio->xio_npages;
 457          ++i
 458     ) {
 459         m = xio->xio_pages[i];
 460         lwb = lwbuf_alloc(m, &lwb_cache);
 461         bcopy(kptr, (char *)lwbuf_kva(lwb) + offset, n);
 462         lwbuf_free(lwb);
 463         bytes -= n;
 464         kptr = (const char *)kptr + n;
 465         if (bytes == 0)
 466             break;
 467         if ((n = bytes) > PAGE_SIZE)
 468             n = PAGE_SIZE;
 469         offset = 0;
 470     }
 471     return(error);
 472 }