sys/kern/kern_memio.c

   1 /*-
   2  * Copyright (c) 1988 University of Utah.
   3  * Copyright (c) 1982, 1986, 1990 The Regents of the University of California.
   4  * All rights reserved.
   5  *
   6  * This code is derived from software contributed to Berkeley by
   7  * the Systems Programming Group of the University of Utah Computer
   8  * Science Department, and code derived from software contributed to
   9  * Berkeley by William Jolitz.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 3. All advertising materials mentioning features or use of this software
  20  *    must display the following acknowledgement:
  21  *      This product includes software developed by the University of
  22  *      California, Berkeley and its contributors.
  23  * 4. Neither the name of the University nor the names of its contributors
  24  *    may be used to endorse or promote products derived from this software
  25  *    without specific prior written permission.
  26  *
  27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  37  * SUCH DAMAGE.
  38  *
  39  *      from: Utah $Hdr: mem.c 1.13 89/10/08$
  40  *      from: @(#)mem.c 7.2 (Berkeley) 5/9/91
  41  * $FreeBSD: src/sys/i386/i386/mem.c,v 1.79.2.9 2003/01/04 22:58:01 njl Exp $
  42  * $DragonFly: src/sys/kern/kern_memio.c,v 1.32 2008/07/23 16:39:28 dillon Exp $
  43  */
  44
  45 /*
  46  * Memory special file
  47  */
  48
  49 #include <sys/param.h>
  50 #include <sys/systm.h>
  51 #include <sys/buf.h>
  52 #include <sys/conf.h>
  53 #include <sys/fcntl.h>
  54 #include <sys/filio.h>
  55 #include <sys/kernel.h>
  56 #include <sys/malloc.h>
  57 #include <sys/memrange.h>
  58 #include <sys/proc.h>
  59 #include <sys/priv.h>
  60 #include <sys/random.h>
  61 #include <sys/signalvar.h>
  62 #include <sys/signal2.h>
  63 #include <sys/uio.h>
  64 #include <sys/vnode.h>
  65
  66 #include <vm/vm.h>
  67 #include <vm/pmap.h>
  68 #include <vm/vm_extern.h>
  69
  70
  71 static  d_open_t        mmopen;
  72 static  d_close_t       mmclose;
  73 static  d_read_t        mmread;
  74 static  d_write_t       mmwrite;
  75 static  d_ioctl_t       mmioctl;
  76 static  d_mmap_t        memmmap;
  77 static  d_poll_t        mmpoll;
  78
  79 #define CDEV_MAJOR 2
  80 static struct dev_ops mem_ops = {
  81         { "mem", CDEV_MAJOR, D_MEM },
  82         .d_open =       mmopen,
  83         .d_close =      mmclose,
  84         .d_read =       mmread,
  85         .d_write =      mmwrite,
  86         .d_ioctl =      mmioctl,
  87         .d_poll =       mmpoll,
  88         .d_mmap =       memmmap,
  89 };
  90
  91 static int rand_bolt;
  92 static caddr_t  zbuf;
  93
  94 MALLOC_DEFINE(M_MEMDESC, "memdesc", "memory range descriptors");
  95 static int mem_ioctl (cdev_t, u_long, caddr_t, int, struct ucred *);
  96 static int random_ioctl (cdev_t, u_long, caddr_t, int, struct ucred *);
  97
  98 struct mem_range_softc mem_range_softc;
  99
 100
 101 static int
 102 mmopen(struct dev_open_args *ap)
 103 {
 104         cdev_t dev = ap->a_head.a_dev;
 105         int error;
 106
 107         switch (minor(dev)) {
 108         case 0:
 109         case 1:
 110                 if (ap->a_oflags & FWRITE) {
 111                         if (securelevel > 0 || kernel_mem_readonly)
 112                                 return (EPERM);
 113                 }
 114                 error = 0;
 115                 break;
 116         case 14:
 117                 error = priv_check_cred(ap->a_cred, PRIV_ROOT, 0);
 118                 if (error != 0)
 119                         break;
 120                 if (securelevel > 0 || kernel_mem_readonly) {
 121                         error = EPERM;
 122                         break;
 123                 }
 124                 error = cpu_set_iopl();
 125                 break;
 126         default:
 127                 error = 0;
 128                 break;
 129         }
 130         return (error);
 131 }
 132
 133 static int
 134 mmclose(struct dev_close_args *ap)
 135 {
 136         cdev_t dev = ap->a_head.a_dev;
 137         int error;
 138
 139         switch (minor(dev)) {
 140         case 14:
 141                 error = cpu_clr_iopl();
 142                 break;
 143         default:
 144                 error = 0;
 145                 break;
 146         }
 147         return (error);
 148 }
 149
 150
 151 static int
 152 mmrw(cdev_t dev, struct uio *uio, int flags)
 153 {
 154         int o;
 155         u_int c, v;
 156         u_int poolsize;
 157         struct iovec *iov;
 158         int error = 0;
 159         caddr_t buf = NULL;
 160
 161         while (uio->uio_resid > 0 && error == 0) {
 162                 iov = uio->uio_iov;
 163                 if (iov->iov_len == 0) {
 164                         uio->uio_iov++;
 165                         uio->uio_iovcnt--;
 166                         if (uio->uio_iovcnt < 0)
 167                                 panic("mmrw");
 168                         continue;
 169                 }
 170                 switch (minor(dev)) {
 171                 case 0:
 172                         /*
 173                          * minor device 0 is physical memory, /dev/mem
 174                          */
 175                         v = uio->uio_offset;
 176                         v &= ~PAGE_MASK;
 177                         pmap_kenter((vm_offset_t)ptvmmap, v);
 178                         o = (int)uio->uio_offset & PAGE_MASK;
 179                         c = (u_int)(PAGE_SIZE - ((uintptr_t)iov->iov_base & PAGE_MASK));
 180                         c = min(c, (u_int)(PAGE_SIZE - o));
 181                         c = min(c, (u_int)iov->iov_len);
 182                         error = uiomove((caddr_t)&ptvmmap[o], (int)c, uio);
 183                         pmap_kremove((vm_offset_t)ptvmmap);
 184                         continue;
 185
 186                 case 1: {
 187                         /*
 188                          * minor device 1 is kernel memory, /dev/kmem
 189                          */
 190                         vm_offset_t saddr, eaddr;
 191                         int prot;
 192
 193                         c = iov->iov_len;
 194
 195                         /*
 196                          * Make sure that all of the pages are currently
 197                          * resident so that we don't create any zero-fill
 198                          * pages.
 199                          */
 200                         saddr = trunc_page(uio->uio_offset);
 201                         eaddr = round_page(uio->uio_offset + c);
 202                         if (saddr > eaddr)
 203                                 return EFAULT;
 204
 205                         /*
 206                          * Make sure the kernel addresses are mapped.
 207                          * platform_direct_mapped() can be used to bypass
 208                          * default mapping via the page table (virtual kernels
 209                          * contain a lot of out-of-band data).
 210                          */
 211                         prot = VM_PROT_READ;
 212                         if (uio->uio_rw != UIO_READ)
 213                                 prot |= VM_PROT_WRITE;
 214                         error = kvm_access_check(saddr, eaddr, prot);
 215                         if (error)
 216                                 return (error);
 217                         error = uiomove((caddr_t)(vm_offset_t)uio->uio_offset,
 218                                         (int)c, uio);
 219                         continue;
 220                 }
 221                 case 2:
 222                         /*
 223                          * minor device 2 is EOF/RATHOLE
 224                          */
 225                         if (uio->uio_rw == UIO_READ)
 226                                 return (0);
 227                         c = iov->iov_len;
 228                         break;
 229                 case 3:
 230                         /*
 231                          * minor device 3 (/dev/random) is source of filth
 232                          * on read, seeder on write
 233                          */
 234                         if (buf == NULL)
 235                                 buf = kmalloc(PAGE_SIZE, M_TEMP, M_WAITOK);
 236                         c = min(iov->iov_len, PAGE_SIZE);
 237                         if (uio->uio_rw == UIO_WRITE) {
 238                                 error = uiomove(buf, (int)c, uio);
 239                                 if (error == 0)
 240                                         error = add_buffer_randomness(buf, c);
 241                         } else {
 242                                 poolsize = read_random(buf, c);
 243                                 if (poolsize == 0) {
 244                                         if (buf)
 245                                                 kfree(buf, M_TEMP);
 246                                         if ((flags & IO_NDELAY) != 0)
 247                                                 return (EWOULDBLOCK);
 248                                         return (0);
 249                                 }
 250                                 c = min(c, poolsize);
 251                                 error = uiomove(buf, (int)c, uio);
 252                         }
 253                         continue;
 254                 case 4:
 255                         /*
 256                          * minor device 4 (/dev/urandom) is source of muck
 257                          * on read, writes are disallowed.
 258                          */
 259                         c = min(iov->iov_len, PAGE_SIZE);
 260                         if (uio->uio_rw == UIO_WRITE) {
 261                                 error = EPERM;
 262                                 break;
 263                         }
 264                         if (CURSIG(curthread->td_lwp) != 0) {
 265                                 /*
 266                                  * Use tsleep() to get the error code right.
 267                                  * It should return immediately.
 268                                  */
 269                                 error = tsleep(&rand_bolt, PCATCH, "urand", 1);
 270                                 if (error != 0 && error != EWOULDBLOCK)
 271                                         continue;
 272                         }
 273                         if (buf == NULL)
 274                                 buf = kmalloc(PAGE_SIZE, M_TEMP, M_WAITOK);
 275                         poolsize = read_random_unlimited(buf, c);
 276                         c = min(c, poolsize);
 277                         error = uiomove(buf, (int)c, uio);
 278                         continue;
 279                 case 12:
 280                         /*
 281                          * minor device 12 (/dev/zero) is source of nulls
 282                          * on read, write are disallowed.
 283                          */
 284                         if (uio->uio_rw == UIO_WRITE) {
 285                                 c = iov->iov_len;
 286                                 break;
 287                         }
 288                         if (zbuf == NULL) {
 289                                 zbuf = (caddr_t)kmalloc(PAGE_SIZE, M_TEMP,
 290                                     M_WAITOK | M_ZERO);
 291                         }
 292                         c = min(iov->iov_len, PAGE_SIZE);
 293                         error = uiomove(zbuf, (int)c, uio);
 294                         continue;
 295                 default:
 296                         return (ENODEV);
 297                 }
 298                 if (error)
 299                         break;
 300                 iov->iov_base = (char *)iov->iov_base + c;
 301                 iov->iov_len -= c;
 302                 uio->uio_offset += c;
 303                 uio->uio_resid -= c;
 304         }
 305         if (buf)
 306                 kfree(buf, M_TEMP);
 307         return (error);
 308 }
 309
 310 static int
 311 mmread(struct dev_read_args *ap)
 312 {
 313         return(mmrw(ap->a_head.a_dev, ap->a_uio, ap->a_ioflag));
 314 }
 315
 316 static int
 317 mmwrite(struct dev_write_args *ap)
 318 {
 319         return(mmrw(ap->a_head.a_dev, ap->a_uio, ap->a_ioflag));
 320 }
 321
 322
 323
 324
 325
 326 /*******************************************************\
 327 * allow user processes to MMAP some memory sections     *
 328 * instead of going through read/write                   *
 329 \*******************************************************/
 330
 331 static int
 332 memmmap(struct dev_mmap_args *ap)
 333 {
 334         cdev_t dev = ap->a_head.a_dev;
 335
 336         switch (minor(dev)) {
 337         case 0:
 338                 /*
 339                  * minor device 0 is physical memory
 340                  */
 341 #if defined(__i386__)
 342                 ap->a_result = i386_btop(ap->a_offset);
 343 #elif defined(__amd64__)
 344                 ap->a_result = amd64_btop(ap->a_offset);
 345 #endif
 346                 return 0;
 347         case 1:
 348                 /*
 349                  * minor device 1 is kernel memory
 350                  */
 351 #if defined(__i386__)
 352                 ap->a_result = i386_btop(vtophys(ap->a_offset));
 353 #elif defined(__amd64__)
 354                 ap->a_result = amd64_btop(vtophys(ap->a_offset));
 355 #endif
 356                 return 0;
 357
 358         default:
 359                 return EINVAL;
 360         }
 361 }
 362
 363 static int
 364 mmioctl(struct dev_ioctl_args *ap)
 365 {
 366         cdev_t dev = ap->a_head.a_dev;
 367
 368         switch (minor(dev)) {
 369         case 0:
 370                 return mem_ioctl(dev, ap->a_cmd, ap->a_data,
 371                                  ap->a_fflag, ap->a_cred);
 372         case 3:
 373         case 4:
 374                 return random_ioctl(dev, ap->a_cmd, ap->a_data,
 375                                     ap->a_fflag, ap->a_cred);
 376         }
 377         return (ENODEV);
 378 }
 379
 380 /*
 381  * Operations for changing memory attributes.
 382  *
 383  * This is basically just an ioctl shim for mem_range_attr_get
 384  * and mem_range_attr_set.
 385  */
 386 static int
 387 mem_ioctl(cdev_t dev, u_long cmd, caddr_t data, int flags, struct ucred *cred)
 388 {
 389         int nd, error = 0;
 390         struct mem_range_op *mo = (struct mem_range_op *)data;
 391         struct mem_range_desc *md;
 392
 393         /* is this for us? */
 394         if ((cmd != MEMRANGE_GET) &&
 395             (cmd != MEMRANGE_SET))
 396                 return (ENOTTY);
 397
 398         /* any chance we can handle this? */
 399         if (mem_range_softc.mr_op == NULL)
 400                 return (EOPNOTSUPP);
 401
 402         /* do we have any descriptors? */
 403         if (mem_range_softc.mr_ndesc == 0)
 404                 return (ENXIO);
 405
 406         switch (cmd) {
 407         case MEMRANGE_GET:
 408                 nd = imin(mo->mo_arg[0], mem_range_softc.mr_ndesc);
 409                 if (nd > 0) {
 410                         md = (struct mem_range_desc *)
 411                                 kmalloc(nd * sizeof(struct mem_range_desc),
 412                                        M_MEMDESC, M_WAITOK);
 413                         error = mem_range_attr_get(md, &nd);
 414                         if (!error)
 415                                 error = copyout(md, mo->mo_desc,
 416                                         nd * sizeof(struct mem_range_desc));
 417                         kfree(md, M_MEMDESC);
 418                 } else {
 419                         nd = mem_range_softc.mr_ndesc;
 420                 }
 421                 mo->mo_arg[0] = nd;
 422                 break;
 423
 424         case MEMRANGE_SET:
 425                 md = (struct mem_range_desc *)kmalloc(sizeof(struct mem_range_desc),
 426                                                     M_MEMDESC, M_WAITOK);
 427                 error = copyin(mo->mo_desc, md, sizeof(struct mem_range_desc));
 428                 /* clamp description string */
 429                 md->mr_owner[sizeof(md->mr_owner) - 1] = 0;
 430                 if (error == 0)
 431                         error = mem_range_attr_set(md, &mo->mo_arg[0]);
 432                 kfree(md, M_MEMDESC);
 433                 break;
 434         }
 435         return (error);
 436 }
 437
 438 /*
 439  * Implementation-neutral, kernel-callable functions for manipulating
 440  * memory range attributes.
 441  */
 442 int
 443 mem_range_attr_get(struct mem_range_desc *mrd, int *arg)
 444 {
 445         /* can we handle this? */
 446         if (mem_range_softc.mr_op == NULL)
 447                 return (EOPNOTSUPP);
 448
 449         if (*arg == 0) {
 450                 *arg = mem_range_softc.mr_ndesc;
 451         } else {
 452                 bcopy(mem_range_softc.mr_desc, mrd, (*arg) * sizeof(struct mem_range_desc));
 453         }
 454         return (0);
 455 }
 456
 457 int
 458 mem_range_attr_set(struct mem_range_desc *mrd, int *arg)
 459 {
 460         /* can we handle this? */
 461         if (mem_range_softc.mr_op == NULL)
 462                 return (EOPNOTSUPP);
 463
 464         return (mem_range_softc.mr_op->set(&mem_range_softc, mrd, arg));
 465 }
 466
 467 #ifdef SMP
 468 void
 469 mem_range_AP_init(void)
 470 {
 471         if (mem_range_softc.mr_op && mem_range_softc.mr_op->initAP)
 472                 return (mem_range_softc.mr_op->initAP(&mem_range_softc));
 473 }
 474 #endif
 475
 476 static int
 477 random_ioctl(cdev_t dev, u_long cmd, caddr_t data, int flags, struct ucred *cred)
 478 {
 479         int error;
 480         int intr;
 481
 482         /*
 483          * Even inspecting the state is privileged, since it gives a hint
 484          * about how easily the randomness might be guessed.
 485          */
 486         error = 0;
 487
 488         switch (cmd) {
 489         /* Really handled in upper layer */
 490         case FIOASYNC:
 491                 break;
 492         case MEM_SETIRQ:
 493                 intr = *(int16_t *)data;
 494                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
 495                         break;
 496                 if (intr < 0 || intr >= MAX_INTS)
 497                         return (EINVAL);
 498                 register_randintr(intr);
 499                 break;
 500         case MEM_CLEARIRQ:
 501                 intr = *(int16_t *)data;
 502                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
 503                         break;
 504                 if (intr < 0 || intr >= MAX_INTS)
 505                         return (EINVAL);
 506                 unregister_randintr(intr);
 507                 break;
 508         case MEM_RETURNIRQ:
 509                 error = ENOTSUP;
 510                 break;
 511         case MEM_FINDIRQ:
 512                 intr = *(int16_t *)data;
 513                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
 514                         break;
 515                 if (intr < 0 || intr >= MAX_INTS)
 516                         return (EINVAL);
 517                 intr = next_registered_randintr(intr);
 518                 if (intr == MAX_INTS)
 519                         return (ENOENT);
 520                 *(u_int16_t *)data = intr;
 521                 break;
 522         default:
 523                 error = ENOTSUP;
 524                 break;
 525         }
 526         return (error);
 527 }
 528
 529 int
 530 mmpoll(struct dev_poll_args *ap)
 531 {
 532         cdev_t dev = ap->a_head.a_dev;
 533         int revents;
 534
 535         switch (minor(dev)) {
 536         case 3:         /* /dev/random */
 537                 revents = random_poll(dev, ap->a_events);
 538                 break;
 539         case 4:         /* /dev/urandom */
 540         default:
 541                 revents = seltrue(dev, ap->a_events);
 542                 break;
 543         }
 544         ap->a_events = revents;
 545         return (0);
 546 }
 547
 548 int
 549 iszerodev(cdev_t dev)
 550 {
 551         return ((major(dev) == mem_ops.head.maj)
 552           && minor(dev) == 12);
 553 }
 554
 555 static void
 556 mem_drvinit(void *unused)
 557 {
 558
 559         /* Initialise memory range handling */
 560         if (mem_range_softc.mr_op != NULL)
 561                 mem_range_softc.mr_op->init(&mem_range_softc);
 562
 563         dev_ops_add(&mem_ops, 0xf0, 0);
 564         make_dev(&mem_ops, 0, UID_ROOT, GID_KMEM, 0640, "mem");
 565         make_dev(&mem_ops, 1, UID_ROOT, GID_KMEM, 0640, "kmem");
 566         make_dev(&mem_ops, 2, UID_ROOT, GID_WHEEL, 0666, "null");
 567         make_dev(&mem_ops, 3, UID_ROOT, GID_WHEEL, 0644, "random");
 568         make_dev(&mem_ops, 4, UID_ROOT, GID_WHEEL, 0644, "urandom");
 569         make_dev(&mem_ops, 12, UID_ROOT, GID_WHEEL, 0666, "zero");
 570         make_dev(&mem_ops, 14, UID_ROOT, GID_WHEEL, 0600, "io");
 571 }
 572
 573 SYSINIT(memdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,mem_drvinit,NULL)
 574