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 | D_MPSAFE_READ | D_MPSAFE_WRITE },
  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 static cdev_t   zerodev = NULL;
  94
  95 MALLOC_DEFINE(M_MEMDESC, "memdesc", "memory range descriptors");
  96 static int mem_ioctl (cdev_t, u_long, caddr_t, int, struct ucred *);
  97 static int random_ioctl (cdev_t, u_long, caddr_t, int, struct ucred *);
  98
  99 struct mem_range_softc mem_range_softc;
 100
 101
 102 static int
 103 mmopen(struct dev_open_args *ap)
 104 {
 105         cdev_t dev = ap->a_head.a_dev;
 106         int error;
 107
 108         switch (minor(dev)) {
 109         case 0:
 110         case 1:
 111                 if (ap->a_oflags & FWRITE) {
 112                         if (securelevel > 0 || kernel_mem_readonly)
 113                                 return (EPERM);
 114                 }
 115                 error = 0;
 116                 break;
 117         case 14:
 118                 error = priv_check_cred(ap->a_cred, PRIV_ROOT, 0);
 119                 if (error != 0)
 120                         break;
 121                 if (securelevel > 0 || kernel_mem_readonly) {
 122                         error = EPERM;
 123                         break;
 124                 }
 125                 error = cpu_set_iopl();
 126                 break;
 127         default:
 128                 error = 0;
 129                 break;
 130         }
 131         return (error);
 132 }
 133
 134 static int
 135 mmclose(struct dev_close_args *ap)
 136 {
 137         cdev_t dev = ap->a_head.a_dev;
 138         int error;
 139
 140         switch (minor(dev)) {
 141         case 14:
 142                 error = cpu_clr_iopl();
 143                 break;
 144         default:
 145                 error = 0;
 146                 break;
 147         }
 148         return (error);
 149 }
 150
 151
 152 static int
 153 mmrw(cdev_t dev, struct uio *uio, int flags)
 154 {
 155         int o;
 156         u_int c, v;
 157         u_int poolsize;
 158         struct iovec *iov;
 159         int error = 0;
 160         caddr_t buf = NULL;
 161
 162         while (uio->uio_resid > 0 && error == 0) {
 163                 iov = uio->uio_iov;
 164                 if (iov->iov_len == 0) {
 165                         uio->uio_iov++;
 166                         uio->uio_iovcnt--;
 167                         if (uio->uio_iovcnt < 0)
 168                                 panic("mmrw");
 169                         continue;
 170                 }
 171                 switch (minor(dev)) {
 172                 case 0:
 173                         /*
 174                          * minor device 0 is physical memory, /dev/mem
 175                          */
 176                         v = uio->uio_offset;
 177                         v &= ~PAGE_MASK;
 178                         pmap_kenter((vm_offset_t)ptvmmap, v);
 179                         o = (int)uio->uio_offset & PAGE_MASK;
 180                         c = (u_int)(PAGE_SIZE - ((uintptr_t)iov->iov_base & PAGE_MASK));
 181                         c = min(c, (u_int)(PAGE_SIZE - o));
 182                         c = min(c, (u_int)iov->iov_len);
 183                         error = uiomove((caddr_t)&ptvmmap[o], (int)c, uio);
 184                         pmap_kremove((vm_offset_t)ptvmmap);
 185                         continue;
 186
 187                 case 1: {
 188                         /*
 189                          * minor device 1 is kernel memory, /dev/kmem
 190                          */
 191                         vm_offset_t saddr, eaddr;
 192                         int prot;
 193
 194                         c = iov->iov_len;
 195
 196                         /*
 197                          * Make sure that all of the pages are currently
 198                          * resident so that we don't create any zero-fill
 199                          * pages.
 200                          */
 201                         saddr = trunc_page(uio->uio_offset);
 202                         eaddr = round_page(uio->uio_offset + c);
 203                         if (saddr > eaddr)
 204                                 return EFAULT;
 205
 206                         /*
 207                          * Make sure the kernel addresses are mapped.
 208                          * platform_direct_mapped() can be used to bypass
 209                          * default mapping via the page table (virtual kernels
 210                          * contain a lot of out-of-band data).
 211                          */
 212                         prot = VM_PROT_READ;
 213                         if (uio->uio_rw != UIO_READ)
 214                                 prot |= VM_PROT_WRITE;
 215                         error = kvm_access_check(saddr, eaddr, prot);
 216                         if (error)
 217                                 return (error);
 218                         error = uiomove((caddr_t)(vm_offset_t)uio->uio_offset,
 219                                         (int)c, uio);
 220                         continue;
 221                 }
 222                 case 2:
 223                         /*
 224                          * minor device 2 is EOF/RATHOLE
 225                          */
 226                         if (uio->uio_rw == UIO_READ)
 227                                 return (0);
 228                         c = iov->iov_len;
 229                         break;
 230                 case 3:
 231                         /*
 232                          * minor device 3 (/dev/random) is source of filth
 233                          * on read, seeder on write
 234                          */
 235                         if (buf == NULL)
 236                                 buf = kmalloc(PAGE_SIZE, M_TEMP, M_WAITOK);
 237                         c = min(iov->iov_len, PAGE_SIZE);
 238                         if (uio->uio_rw == UIO_WRITE) {
 239                                 error = uiomove(buf, (int)c, uio);
 240                                 if (error == 0)
 241                                         error = add_buffer_randomness(buf, c);
 242                         } else {
 243                                 poolsize = read_random(buf, c);
 244                                 if (poolsize == 0) {
 245                                         if (buf)
 246                                                 kfree(buf, M_TEMP);
 247                                         if ((flags & IO_NDELAY) != 0)
 248                                                 return (EWOULDBLOCK);
 249                                         return (0);
 250                                 }
 251                                 c = min(c, poolsize);
 252                                 error = uiomove(buf, (int)c, uio);
 253                         }
 254                         continue;
 255                 case 4:
 256                         /*
 257                          * minor device 4 (/dev/urandom) is source of muck
 258                          * on read, writes are disallowed.
 259                          */
 260                         c = min(iov->iov_len, PAGE_SIZE);
 261                         if (uio->uio_rw == UIO_WRITE) {
 262                                 error = EPERM;
 263                                 break;
 264                         }
 265                         if (CURSIG(curthread->td_lwp) != 0) {
 266                                 /*
 267                                  * Use tsleep() to get the error code right.
 268                                  * It should return immediately.
 269                                  */
 270                                 error = tsleep(&rand_bolt, PCATCH, "urand", 1);
 271                                 if (error != 0 && error != EWOULDBLOCK)
 272                                         continue;
 273                         }
 274                         if (buf == NULL)
 275                                 buf = kmalloc(PAGE_SIZE, M_TEMP, M_WAITOK);
 276                         poolsize = read_random_unlimited(buf, c);
 277                         c = min(c, poolsize);
 278                         error = uiomove(buf, (int)c, uio);
 279                         continue;
 280                 case 12:
 281                         /*
 282                          * minor device 12 (/dev/zero) is source of nulls
 283                          * on read, write are disallowed.
 284                          */
 285                         if (uio->uio_rw == UIO_WRITE) {
 286                                 c = iov->iov_len;
 287                                 break;
 288                         }
 289                         if (zbuf == NULL) {
 290                                 zbuf = (caddr_t)kmalloc(PAGE_SIZE, M_TEMP,
 291                                     M_WAITOK | M_ZERO);
 292                         }
 293                         c = min(iov->iov_len, PAGE_SIZE);
 294                         error = uiomove(zbuf, (int)c, uio);
 295                         continue;
 296                 default:
 297                         return (ENODEV);
 298                 }
 299                 if (error)
 300                         break;
 301                 iov->iov_base = (char *)iov->iov_base + c;
 302                 iov->iov_len -= c;
 303                 uio->uio_offset += c;
 304                 uio->uio_resid -= c;
 305         }
 306         if (buf)
 307                 kfree(buf, M_TEMP);
 308         return (error);
 309 }
 310
 311 static int
 312 mmread(struct dev_read_args *ap)
 313 {
 314         return(mmrw(ap->a_head.a_dev, ap->a_uio, ap->a_ioflag));
 315 }
 316
 317 static int
 318 mmwrite(struct dev_write_args *ap)
 319 {
 320         return(mmrw(ap->a_head.a_dev, ap->a_uio, ap->a_ioflag));
 321 }
 322
 323
 324
 325
 326
 327 /*******************************************************\
 328 * allow user processes to MMAP some memory sections     *
 329 * instead of going through read/write                   *
 330 \*******************************************************/
 331
 332 static int
 333 memmmap(struct dev_mmap_args *ap)
 334 {
 335         cdev_t dev = ap->a_head.a_dev;
 336
 337         switch (minor(dev)) {
 338         case 0:
 339                 /*
 340                  * minor device 0 is physical memory
 341                  */
 342 #if defined(__i386__)
 343                 ap->a_result = i386_btop(ap->a_offset);
 344 #elif defined(__x86_64__)
 345                 ap->a_result = x86_64_btop(ap->a_offset);
 346 #endif
 347                 return 0;
 348         case 1:
 349                 /*
 350                  * minor device 1 is kernel memory
 351                  */
 352 #if defined(__i386__)
 353                 ap->a_result = i386_btop(vtophys(ap->a_offset));
 354 #elif defined(__x86_64__)
 355                 ap->a_result = x86_64_btop(vtophys(ap->a_offset));
 356 #endif
 357                 return 0;
 358
 359         default:
 360                 return EINVAL;
 361         }
 362 }
 363
 364 static int
 365 mmioctl(struct dev_ioctl_args *ap)
 366 {
 367         cdev_t dev = ap->a_head.a_dev;
 368
 369         switch (minor(dev)) {
 370         case 0:
 371                 return mem_ioctl(dev, ap->a_cmd, ap->a_data,
 372                                  ap->a_fflag, ap->a_cred);
 373         case 3:
 374         case 4:
 375                 return random_ioctl(dev, ap->a_cmd, ap->a_data,
 376                                     ap->a_fflag, ap->a_cred);
 377         }
 378         return (ENODEV);
 379 }
 380
 381 /*
 382  * Operations for changing memory attributes.
 383  *
 384  * This is basically just an ioctl shim for mem_range_attr_get
 385  * and mem_range_attr_set.
 386  */
 387 static int
 388 mem_ioctl(cdev_t dev, u_long cmd, caddr_t data, int flags, struct ucred *cred)
 389 {
 390         int nd, error = 0;
 391         struct mem_range_op *mo = (struct mem_range_op *)data;
 392         struct mem_range_desc *md;
 393
 394         /* is this for us? */
 395         if ((cmd != MEMRANGE_GET) &&
 396             (cmd != MEMRANGE_SET))
 397                 return (ENOTTY);
 398
 399         /* any chance we can handle this? */
 400         if (mem_range_softc.mr_op == NULL)
 401                 return (EOPNOTSUPP);
 402
 403         /* do we have any descriptors? */
 404         if (mem_range_softc.mr_ndesc == 0)
 405                 return (ENXIO);
 406
 407         switch (cmd) {
 408         case MEMRANGE_GET:
 409                 nd = imin(mo->mo_arg[0], mem_range_softc.mr_ndesc);
 410                 if (nd > 0) {
 411                         md = (struct mem_range_desc *)
 412                                 kmalloc(nd * sizeof(struct mem_range_desc),
 413                                        M_MEMDESC, M_WAITOK);
 414                         error = mem_range_attr_get(md, &nd);
 415                         if (!error)
 416                                 error = copyout(md, mo->mo_desc,
 417                                         nd * sizeof(struct mem_range_desc));
 418                         kfree(md, M_MEMDESC);
 419                 } else {
 420                         nd = mem_range_softc.mr_ndesc;
 421                 }
 422                 mo->mo_arg[0] = nd;
 423                 break;
 424
 425         case MEMRANGE_SET:
 426                 md = (struct mem_range_desc *)kmalloc(sizeof(struct mem_range_desc),
 427                                                     M_MEMDESC, M_WAITOK);
 428                 error = copyin(mo->mo_desc, md, sizeof(struct mem_range_desc));
 429                 /* clamp description string */
 430                 md->mr_owner[sizeof(md->mr_owner) - 1] = 0;
 431                 if (error == 0)
 432                         error = mem_range_attr_set(md, &mo->mo_arg[0]);
 433                 kfree(md, M_MEMDESC);
 434                 break;
 435         }
 436         return (error);
 437 }
 438
 439 /*
 440  * Implementation-neutral, kernel-callable functions for manipulating
 441  * memory range attributes.
 442  */
 443 int
 444 mem_range_attr_get(struct mem_range_desc *mrd, int *arg)
 445 {
 446         /* can we handle this? */
 447         if (mem_range_softc.mr_op == NULL)
 448                 return (EOPNOTSUPP);
 449
 450         if (*arg == 0) {
 451                 *arg = mem_range_softc.mr_ndesc;
 452         } else {
 453                 bcopy(mem_range_softc.mr_desc, mrd, (*arg) * sizeof(struct mem_range_desc));
 454         }
 455         return (0);
 456 }
 457
 458 int
 459 mem_range_attr_set(struct mem_range_desc *mrd, int *arg)
 460 {
 461         /* can we handle this? */
 462         if (mem_range_softc.mr_op == NULL)
 463                 return (EOPNOTSUPP);
 464
 465         return (mem_range_softc.mr_op->set(&mem_range_softc, mrd, arg));
 466 }
 467
 468 #ifdef SMP
 469 void
 470 mem_range_AP_init(void)
 471 {
 472         if (mem_range_softc.mr_op && mem_range_softc.mr_op->initAP)
 473                 return (mem_range_softc.mr_op->initAP(&mem_range_softc));
 474 }
 475 #endif
 476
 477 static int
 478 random_ioctl(cdev_t dev, u_long cmd, caddr_t data, int flags, struct ucred *cred)
 479 {
 480         int error;
 481         int intr;
 482
 483         /*
 484          * Even inspecting the state is privileged, since it gives a hint
 485          * about how easily the randomness might be guessed.
 486          */
 487         error = 0;
 488
 489         switch (cmd) {
 490         /* Really handled in upper layer */
 491         case FIOASYNC:
 492                 break;
 493         case MEM_SETIRQ:
 494                 intr = *(int16_t *)data;
 495                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
 496                         break;
 497                 if (intr < 0 || intr >= MAX_INTS)
 498                         return (EINVAL);
 499                 register_randintr(intr);
 500                 break;
 501         case MEM_CLEARIRQ:
 502                 intr = *(int16_t *)data;
 503                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
 504                         break;
 505                 if (intr < 0 || intr >= MAX_INTS)
 506                         return (EINVAL);
 507                 unregister_randintr(intr);
 508                 break;
 509         case MEM_RETURNIRQ:
 510                 error = ENOTSUP;
 511                 break;
 512         case MEM_FINDIRQ:
 513                 intr = *(int16_t *)data;
 514                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
 515                         break;
 516                 if (intr < 0 || intr >= MAX_INTS)
 517                         return (EINVAL);
 518                 intr = next_registered_randintr(intr);
 519                 if (intr == MAX_INTS)
 520                         return (ENOENT);
 521                 *(u_int16_t *)data = intr;
 522                 break;
 523         default:
 524                 error = ENOTSUP;
 525                 break;
 526         }
 527         return (error);
 528 }
 529
 530 int
 531 mmpoll(struct dev_poll_args *ap)
 532 {
 533         cdev_t dev = ap->a_head.a_dev;
 534         int revents;
 535
 536         switch (minor(dev)) {
 537         case 3:         /* /dev/random */
 538                 revents = random_poll(dev, ap->a_events);
 539                 break;
 540         case 4:         /* /dev/urandom */
 541         default:
 542                 revents = seltrue(dev, ap->a_events);
 543                 break;
 544         }
 545         ap->a_events = revents;
 546         return (0);
 547 }
 548
 549 int
 550 iszerodev(cdev_t dev)
 551 {
 552         return (zerodev == dev);
 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         make_dev(&mem_ops, 0, UID_ROOT, GID_KMEM, 0640, "mem");
 564         make_dev(&mem_ops, 1, UID_ROOT, GID_KMEM, 0640, "kmem");
 565         make_dev(&mem_ops, 2, UID_ROOT, GID_WHEEL, 0666, "null");
 566         make_dev(&mem_ops, 3, UID_ROOT, GID_WHEEL, 0644, "random");
 567         make_dev(&mem_ops, 4, UID_ROOT, GID_WHEEL, 0644, "urandom");
 568         zerodev = make_dev(&mem_ops, 12, UID_ROOT, GID_WHEEL, 0666, "zero");
 569         make_dev(&mem_ops, 14, UID_ROOT, GID_WHEEL, 0600, "io");
 570 }
 571
 572 SYSINIT(memdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,mem_drvinit,NULL)
 573