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     lwkt_gettoken(&vm_token);
 181     crit_enter();
 182     for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
 183         if ((paddr = pmap_kextract(addr)) == 0)
 184             break;
 185         m = PHYS_TO_VM_PAGE(paddr);
 186         vm_page_hold(m);
 187         xio->xio_pages[i] = m;
 188         kbytes -= n;
 189         xio->xio_bytes += n;
 190         if ((n = kbytes) > PAGE_SIZE)
 191             n = PAGE_SIZE;
 192         addr += PAGE_SIZE;
 193     }
 194     crit_exit();
 195     lwkt_reltoken(&vm_token);
 196     xio->xio_npages = i;
 197
 198     /*
 199      * If a failure occured clean out what we loaded and return EFAULT.
 200      * Return 0 on success.
 201      */
 202     if (i < XIO_INTERNAL_PAGES && n) {
 203         xio_release(xio);
 204         xio->xio_error = EFAULT;
 205     }
 206     return(xio->xio_error);
 207 }
 208
 209 /*
 210  * Initialize an XIO given an array of vm_page pointers.  The caller is
 211  * responsible for any modified state changes for the pages.
 212  */
 213 int
 214 xio_init_pages(xio_t xio, struct vm_page **mbase, int npages, int xflags)
 215 {
 216     int i;
 217
 218     KKASSERT(npages <= XIO_INTERNAL_PAGES);
 219
 220     xio->xio_flags = xflags;
 221     xio->xio_offset = 0;
 222     xio->xio_bytes = npages * PAGE_SIZE;
 223     xio->xio_pages = xio->xio_internal_pages;
 224     xio->xio_npages = npages;
 225     xio->xio_error = 0;
 226     lwkt_gettoken(&vm_token);
 227     crit_enter();
 228     for (i = 0; i < npages; ++i) {
 229         vm_page_hold(mbase[i]);
 230         xio->xio_pages[i] = mbase[i];
 231     }
 232     crit_exit();
 233     lwkt_reltoken(&vm_token);
 234     return(0);
 235 }
 236
 237 /*
 238  * Cleanup an XIO so it can be destroyed.  The pages associated with the
 239  * XIO are released.
 240  */
 241 void
 242 xio_release(xio_t xio)
 243 {
 244     int i;
 245     vm_page_t m;
 246
 247     lwkt_gettoken(&vm_token);
 248     crit_enter();
 249     for (i = 0; i < xio->xio_npages; ++i) {
 250         m = xio->xio_pages[i];
 251         if (xio->xio_flags & XIOF_WRITE)
 252                 vm_page_dirty(m);
 253         vm_page_unhold(m);
 254     }
 255     crit_exit();
 256     lwkt_reltoken(&vm_token);
 257     xio->xio_offset = 0;
 258     xio->xio_npages = 0;
 259     xio->xio_bytes = 0;
 260     xio->xio_error = ENOBUFS;
 261 }
 262
 263 /*
 264  * Copy data between an XIO and a UIO.  If the UIO represents userspace it
 265  * must be relative to the current context.
 266  *
 267  * uoffset is the abstracted starting offset in the XIO, not the actual
 268  * offset, and usually starts at 0.
 269  *
 270  * The XIO is not modified.  The UIO is updated to reflect the copy.
 271  *
 272  * UIO_READ     xio -> uio
 273  * UIO_WRITE    uio -> xio
 274  */
 275 int
 276 xio_uio_copy(xio_t xio, int uoffset, struct uio *uio, size_t *sizep)
 277 {
 278     size_t bytes;
 279     int error;
 280
 281     bytes = xio->xio_bytes - uoffset;
 282     if (bytes > uio->uio_resid)
 283         bytes = uio->uio_resid;
 284     KKASSERT(bytes >= 0);
 285     error = uiomove_fromphys(xio->xio_pages, xio->xio_offset + uoffset,
 286                                 bytes, uio);
 287     if (error == 0)
 288         *sizep = bytes;
 289     else
 290         *sizep = 0;
 291     return(error);
 292 }
 293
 294 /*
 295  * Copy the specified number of bytes from the xio to a userland
 296  * buffer.  Return an error code or 0 on success.
 297  *
 298  * uoffset is the abstracted starting offset in the XIO, not the actual
 299  * offset, and usually starts at 0.
 300  *
 301  * The XIO is not modified.
 302  */
 303 int
 304 xio_copy_xtou(xio_t xio, int uoffset, void *uptr, int bytes)
 305 {
 306     int i;
 307     int n;
 308     int error;
 309     int offset;
 310     vm_page_t m;
 311     struct lwbuf *lwb;
 312
 313     if (uoffset + bytes > xio->xio_bytes)
 314         return(EFAULT);
 315
 316     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 317     if ((n = PAGE_SIZE - offset) > bytes)
 318         n = bytes;
 319
 320     error = 0;
 321     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 322          i < xio->xio_npages;
 323          ++i
 324     ) {
 325         m = xio->xio_pages[i];
 326         lwb = lwbuf_alloc(m);
 327         error = copyout((char *)lwbuf_kva(lwb) + offset, uptr, n);
 328         lwbuf_free(lwb);
 329         if (error)
 330             break;
 331         bytes -= n;
 332         uptr = (char *)uptr + n;
 333         if (bytes == 0)
 334             break;
 335         if ((n = bytes) > PAGE_SIZE)
 336             n = PAGE_SIZE;
 337         offset = 0;
 338     }
 339     return(error);
 340 }
 341
 342 /*
 343  * Copy the specified number of bytes from the xio to a kernel
 344  * buffer.  Return an error code or 0 on success.
 345  *
 346  * uoffset is the abstracted starting offset in the XIO, not the actual
 347  * offset, and usually starts at 0.
 348  *
 349  * The XIO is not modified.
 350  */
 351 int
 352 xio_copy_xtok(xio_t xio, int uoffset, void *kptr, int bytes)
 353 {
 354     int i;
 355     int n;
 356     int error;
 357     int offset;
 358     vm_page_t m;
 359     struct lwbuf *lwb;
 360
 361     if (bytes + uoffset > xio->xio_bytes)
 362         return(EFAULT);
 363
 364     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 365     if ((n = PAGE_SIZE - offset) > bytes)
 366         n = bytes;
 367
 368     error = 0;
 369     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 370          i < xio->xio_npages;
 371          ++i
 372     ) {
 373         m = xio->xio_pages[i];
 374         lwb = lwbuf_alloc(m);
 375         bcopy((char *)lwbuf_kva(lwb) + offset, kptr, n);
 376         lwbuf_free(lwb);
 377         bytes -= n;
 378         kptr = (char *)kptr + n;
 379         if (bytes == 0)
 380             break;
 381         if ((n = bytes) > PAGE_SIZE)
 382             n = PAGE_SIZE;
 383         offset = 0;
 384     }
 385     return(error);
 386 }
 387
 388 /*
 389  * Copy the specified number of bytes from userland to the xio.
 390  * Return an error code or 0 on success.
 391  *
 392  * uoffset is the abstracted starting offset in the XIO, not the actual
 393  * offset, and usually starts at 0.
 394  *
 395  * Data in pages backing the XIO will be modified.
 396  */
 397 int
 398 xio_copy_utox(xio_t xio, int uoffset, const void *uptr, int bytes)
 399 {
 400     int i;
 401     int n;
 402     int error;
 403     int offset;
 404     vm_page_t m;
 405     struct lwbuf *lwb;
 406
 407     if (uoffset + bytes > xio->xio_bytes)
 408         return(EFAULT);
 409
 410     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 411     if ((n = PAGE_SIZE - offset) > bytes)
 412         n = bytes;
 413
 414     error = 0;
 415     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 416          i < xio->xio_npages;
 417          ++i
 418     ) {
 419         m = xio->xio_pages[i];
 420         lwb = lwbuf_alloc(m);
 421         error = copyin(uptr, (char *)lwbuf_kva(lwb) + offset, n);
 422         lwbuf_free(lwb);
 423         if (error)
 424             break;
 425         bytes -= n;
 426         uptr = (const char *)uptr + n;
 427         if (bytes == 0)
 428             break;
 429         if ((n = bytes) > PAGE_SIZE)
 430             n = PAGE_SIZE;
 431         offset = 0;
 432     }
 433     return(error);
 434 }
 435
 436 /*
 437  * Copy the specified number of bytes from the kernel to the xio.
 438  * Return an error code or 0 on success.
 439  *
 440  * uoffset is the abstracted starting offset in the XIO, not the actual
 441  * offset, and usually starts at 0.
 442  *
 443  * Data in pages backing the XIO will be modified.
 444  */
 445 int
 446 xio_copy_ktox(xio_t xio, int uoffset, const void *kptr, int bytes)
 447 {
 448     int i;
 449     int n;
 450     int error;
 451     int offset;
 452     vm_page_t m;
 453     struct lwbuf *lwb;
 454
 455     if (uoffset + bytes > xio->xio_bytes)
 456         return(EFAULT);
 457
 458     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
 459     if ((n = PAGE_SIZE - offset) > bytes)
 460         n = bytes;
 461
 462     error = 0;
 463     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
 464          i < xio->xio_npages;
 465          ++i
 466     ) {
 467         m = xio->xio_pages[i];
 468         lwb = lwbuf_alloc(m);
 469         bcopy(kptr, (char *)lwbuf_kva(lwb) + offset, n);
 470         lwbuf_free(lwb);
 471         bytes -= n;
 472         kptr = (const char *)kptr + n;
 473         if (bytes == 0)
 474             break;
 475         if ((n = bytes) > PAGE_SIZE)
 476             n = PAGE_SIZE;
 477         offset = 0;
 478     }
 479     return(error);
 480 }