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 */ 43 44 /* 45 * Memory special file 46 */ 47 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/buf.h> 51 #include <sys/conf.h> 52 #include <sys/fcntl.h> 53 #include <sys/filio.h> 54 #include <sys/kernel.h> 55 #include <sys/malloc.h> 56 #include <sys/memrange.h> 57 #include <sys/proc.h> 58 #include <sys/priv.h> 59 #include <sys/random.h> 60 #include <sys/signalvar.h> 61 #include <sys/uio.h> 62 #include <sys/vnode.h> 63 64 #include <sys/signal2.h> 65 #include <sys/mplock2.h> 66 67 #include <vm/vm.h> 68 #include <vm/pmap.h> 69 #include <vm/vm_extern.h> 70 71 72 static d_open_t mmopen; 73 static d_close_t mmclose; 74 static d_read_t mmread; 75 static d_write_t mmwrite; 76 static d_ioctl_t mmioctl; 77 static d_mmap_t memmmap; 78 static d_kqfilter_t mmkqfilter; 79 80 #define CDEV_MAJOR 2 81 static struct dev_ops mem_ops = { 82 { "mem", 0, D_MPSAFE }, 83 .d_open = mmopen, 84 .d_close = mmclose, 85 .d_read = mmread, 86 .d_write = mmwrite, 87 .d_ioctl = mmioctl, 88 .d_kqfilter = mmkqfilter, 89 .d_mmap = memmmap, 90 }; 91 92 static int rand_bolt; 93 static caddr_t zbuf; 94 static cdev_t zerodev = NULL; 95 96 MALLOC_DEFINE(M_MEMDESC, "memdesc", "memory range descriptors"); 97 static int mem_ioctl (cdev_t, u_long, caddr_t, int, struct ucred *); 98 static int random_ioctl (cdev_t, u_long, caddr_t, int, struct ucred *); 99 100 struct mem_range_softc mem_range_softc; 101 102 103 static int 104 mmopen(struct dev_open_args *ap) 105 { 106 cdev_t dev = ap->a_head.a_dev; 107 int error; 108 109 switch (minor(dev)) { 110 case 0: 111 case 1: 112 if (ap->a_oflags & FWRITE) { 113 if (securelevel > 0 || kernel_mem_readonly) 114 return (EPERM); 115 } 116 error = 0; 117 break; 118 case 14: 119 error = priv_check_cred(ap->a_cred, PRIV_ROOT, 0); 120 if (error != 0) 121 break; 122 if (securelevel > 0 || kernel_mem_readonly) { 123 error = EPERM; 124 break; 125 } 126 error = cpu_set_iopl(); 127 break; 128 default: 129 error = 0; 130 break; 131 } 132 return (error); 133 } 134 135 static int 136 mmclose(struct dev_close_args *ap) 137 { 138 cdev_t dev = ap->a_head.a_dev; 139 int error; 140 141 switch (minor(dev)) { 142 case 14: 143 error = cpu_clr_iopl(); 144 break; 145 default: 146 error = 0; 147 break; 148 } 149 return (error); 150 } 151 152 153 static int 154 mmrw(cdev_t dev, struct uio *uio, int flags) 155 { 156 int o; 157 u_int c, v; 158 u_int poolsize; 159 struct iovec *iov; 160 int error = 0; 161 caddr_t buf = NULL; 162 163 while (uio->uio_resid > 0 && error == 0) { 164 iov = uio->uio_iov; 165 if (iov->iov_len == 0) { 166 uio->uio_iov++; 167 uio->uio_iovcnt--; 168 if (uio->uio_iovcnt < 0) 169 panic("mmrw"); 170 continue; 171 } 172 switch (minor(dev)) { 173 case 0: 174 /* 175 * minor device 0 is physical memory, /dev/mem 176 */ 177 v = uio->uio_offset; 178 v &= ~PAGE_MASK; 179 pmap_kenter((vm_offset_t)ptvmmap, v); 180 o = (int)uio->uio_offset & PAGE_MASK; 181 c = (u_int)(PAGE_SIZE - ((uintptr_t)iov->iov_base & PAGE_MASK)); 182 c = min(c, (u_int)(PAGE_SIZE - o)); 183 c = min(c, (u_int)iov->iov_len); 184 error = uiomove((caddr_t)&ptvmmap[o], (int)c, uio); 185 pmap_kremove((vm_offset_t)ptvmmap); 186 continue; 187 188 case 1: { 189 /* 190 * minor device 1 is kernel memory, /dev/kmem 191 */ 192 vm_offset_t saddr, eaddr; 193 int prot; 194 195 c = iov->iov_len; 196 197 /* 198 * Make sure that all of the pages are currently 199 * resident so that we don't create any zero-fill 200 * pages. 201 */ 202 saddr = trunc_page(uio->uio_offset); 203 eaddr = round_page(uio->uio_offset + c); 204 if (saddr > eaddr) 205 return EFAULT; 206 207 /* 208 * Make sure the kernel addresses are mapped. 209 * platform_direct_mapped() can be used to bypass 210 * default mapping via the page table (virtual kernels 211 * contain a lot of out-of-band data). 212 */ 213 prot = VM_PROT_READ; 214 if (uio->uio_rw != UIO_READ) 215 prot |= VM_PROT_WRITE; 216 error = kvm_access_check(saddr, eaddr, prot); 217 if (error) 218 return (error); 219 error = uiomove((caddr_t)(vm_offset_t)uio->uio_offset, 220 (int)c, uio); 221 continue; 222 } 223 case 2: 224 /* 225 * minor device 2 (/dev/null) is EOF/RATHOLE 226 */ 227 if (uio->uio_rw == UIO_READ) 228 return (0); 229 c = iov->iov_len; 230 break; 231 case 3: 232 /* 233 * minor device 3 (/dev/random) is source of filth 234 * on read, seeder on write 235 */ 236 if (buf == NULL) 237 buf = kmalloc(PAGE_SIZE, M_TEMP, M_WAITOK); 238 c = min(iov->iov_len, PAGE_SIZE); 239 if (uio->uio_rw == UIO_WRITE) { 240 error = uiomove(buf, (int)c, uio); 241 if (error == 0) 242 error = add_buffer_randomness(buf, c); 243 } else { 244 poolsize = read_random(buf, c); 245 if (poolsize == 0) { 246 if (buf) 247 kfree(buf, M_TEMP); 248 if ((flags & IO_NDELAY) != 0) 249 return (EWOULDBLOCK); 250 return (0); 251 } 252 c = min(c, poolsize); 253 error = uiomove(buf, (int)c, uio); 254 } 255 continue; 256 case 4: 257 /* 258 * minor device 4 (/dev/urandom) is source of muck 259 * on read, writes are disallowed. 260 */ 261 c = min(iov->iov_len, PAGE_SIZE); 262 if (uio->uio_rw == UIO_WRITE) { 263 error = EPERM; 264 break; 265 } 266 if (CURSIG(curthread->td_lwp) != 0) { 267 /* 268 * Use tsleep() to get the error code right. 269 * It should return immediately. 270 */ 271 error = tsleep(&rand_bolt, PCATCH, "urand", 1); 272 if (error != 0 && error != EWOULDBLOCK) 273 continue; 274 } 275 if (buf == NULL) 276 buf = kmalloc(PAGE_SIZE, M_TEMP, M_WAITOK); 277 poolsize = read_random_unlimited(buf, c); 278 c = min(c, poolsize); 279 error = uiomove(buf, (int)c, uio); 280 continue; 281 case 12: 282 /* 283 * minor device 12 (/dev/zero) is source of nulls 284 * on read, write are disallowed. 285 */ 286 if (uio->uio_rw == UIO_WRITE) { 287 c = iov->iov_len; 288 break; 289 } 290 if (zbuf == NULL) { 291 zbuf = (caddr_t)kmalloc(PAGE_SIZE, M_TEMP, 292 M_WAITOK | M_ZERO); 293 } 294 c = min(iov->iov_len, PAGE_SIZE); 295 error = uiomove(zbuf, (int)c, uio); 296 continue; 297 default: 298 return (ENODEV); 299 } 300 if (error) 301 break; 302 iov->iov_base = (char *)iov->iov_base + c; 303 iov->iov_len -= c; 304 uio->uio_offset += c; 305 uio->uio_resid -= c; 306 } 307 if (buf) 308 kfree(buf, M_TEMP); 309 return (error); 310 } 311 312 static int 313 mmread(struct dev_read_args *ap) 314 { 315 return(mmrw(ap->a_head.a_dev, ap->a_uio, ap->a_ioflag)); 316 } 317 318 static int 319 mmwrite(struct dev_write_args *ap) 320 { 321 return(mmrw(ap->a_head.a_dev, ap->a_uio, ap->a_ioflag)); 322 } 323 324 325 326 327 328 /*******************************************************\ 329 * allow user processes to MMAP some memory sections * 330 * instead of going through read/write * 331 \*******************************************************/ 332 333 static int 334 memmmap(struct dev_mmap_args *ap) 335 { 336 cdev_t dev = ap->a_head.a_dev; 337 338 switch (minor(dev)) { 339 case 0: 340 /* 341 * minor device 0 is physical memory 342 */ 343 #if defined(__i386__) 344 ap->a_result = i386_btop(ap->a_offset); 345 #elif defined(__x86_64__) 346 ap->a_result = x86_64_btop(ap->a_offset); 347 #endif 348 return 0; 349 case 1: 350 /* 351 * minor device 1 is kernel memory 352 */ 353 #if defined(__i386__) 354 ap->a_result = i386_btop(vtophys(ap->a_offset)); 355 #elif defined(__x86_64__) 356 ap->a_result = x86_64_btop(vtophys(ap->a_offset)); 357 #endif 358 return 0; 359 360 default: 361 return EINVAL; 362 } 363 } 364 365 static int 366 mmioctl(struct dev_ioctl_args *ap) 367 { 368 cdev_t dev = ap->a_head.a_dev; 369 int error; 370 371 get_mplock(); 372 373 switch (minor(dev)) { 374 case 0: 375 error = mem_ioctl(dev, ap->a_cmd, ap->a_data, 376 ap->a_fflag, ap->a_cred); 377 break; 378 case 3: 379 case 4: 380 error = random_ioctl(dev, ap->a_cmd, ap->a_data, 381 ap->a_fflag, ap->a_cred); 382 break; 383 default: 384 error = ENODEV; 385 break; 386 } 387 388 rel_mplock(); 389 return (error); 390 } 391 392 /* 393 * Operations for changing memory attributes. 394 * 395 * This is basically just an ioctl shim for mem_range_attr_get 396 * and mem_range_attr_set. 397 */ 398 static int 399 mem_ioctl(cdev_t dev, u_long cmd, caddr_t data, int flags, struct ucred *cred) 400 { 401 int nd, error = 0; 402 struct mem_range_op *mo = (struct mem_range_op *)data; 403 struct mem_range_desc *md; 404 405 /* is this for us? */ 406 if ((cmd != MEMRANGE_GET) && 407 (cmd != MEMRANGE_SET)) 408 return (ENOTTY); 409 410 /* any chance we can handle this? */ 411 if (mem_range_softc.mr_op == NULL) 412 return (EOPNOTSUPP); 413 414 /* do we have any descriptors? */ 415 if (mem_range_softc.mr_ndesc == 0) 416 return (ENXIO); 417 418 switch (cmd) { 419 case MEMRANGE_GET: 420 nd = imin(mo->mo_arg[0], mem_range_softc.mr_ndesc); 421 if (nd > 0) { 422 md = (struct mem_range_desc *) 423 kmalloc(nd * sizeof(struct mem_range_desc), 424 M_MEMDESC, M_WAITOK); 425 error = mem_range_attr_get(md, &nd); 426 if (!error) 427 error = copyout(md, mo->mo_desc, 428 nd * sizeof(struct mem_range_desc)); 429 kfree(md, M_MEMDESC); 430 } else { 431 nd = mem_range_softc.mr_ndesc; 432 } 433 mo->mo_arg[0] = nd; 434 break; 435 436 case MEMRANGE_SET: 437 md = (struct mem_range_desc *)kmalloc(sizeof(struct mem_range_desc), 438 M_MEMDESC, M_WAITOK); 439 error = copyin(mo->mo_desc, md, sizeof(struct mem_range_desc)); 440 /* clamp description string */ 441 md->mr_owner[sizeof(md->mr_owner) - 1] = 0; 442 if (error == 0) 443 error = mem_range_attr_set(md, &mo->mo_arg[0]); 444 kfree(md, M_MEMDESC); 445 break; 446 } 447 return (error); 448 } 449 450 /* 451 * Implementation-neutral, kernel-callable functions for manipulating 452 * memory range attributes. 453 */ 454 int 455 mem_range_attr_get(struct mem_range_desc *mrd, int *arg) 456 { 457 /* can we handle this? */ 458 if (mem_range_softc.mr_op == NULL) 459 return (EOPNOTSUPP); 460 461 if (*arg == 0) { 462 *arg = mem_range_softc.mr_ndesc; 463 } else { 464 bcopy(mem_range_softc.mr_desc, mrd, (*arg) * sizeof(struct mem_range_desc)); 465 } 466 return (0); 467 } 468 469 int 470 mem_range_attr_set(struct mem_range_desc *mrd, int *arg) 471 { 472 /* can we handle this? */ 473 if (mem_range_softc.mr_op == NULL) 474 return (EOPNOTSUPP); 475 476 return (mem_range_softc.mr_op->set(&mem_range_softc, mrd, arg)); 477 } 478 479 #ifdef SMP 480 void 481 mem_range_AP_init(void) 482 { 483 if (mem_range_softc.mr_op && mem_range_softc.mr_op->initAP) 484 return (mem_range_softc.mr_op->initAP(&mem_range_softc)); 485 } 486 #endif 487 488 static int 489 random_ioctl(cdev_t dev, u_long cmd, caddr_t data, int flags, struct ucred *cred) 490 { 491 int error; 492 int intr; 493 494 /* 495 * Even inspecting the state is privileged, since it gives a hint 496 * about how easily the randomness might be guessed. 497 */ 498 error = 0; 499 500 switch (cmd) { 501 /* Really handled in upper layer */ 502 case FIOASYNC: 503 break; 504 case MEM_SETIRQ: 505 intr = *(int16_t *)data; 506 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0) 507 break; 508 if (intr < 0 || intr >= MAX_INTS) 509 return (EINVAL); 510 register_randintr(intr); 511 break; 512 case MEM_CLEARIRQ: 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 unregister_randintr(intr); 519 break; 520 case MEM_RETURNIRQ: 521 error = ENOTSUP; 522 break; 523 case MEM_FINDIRQ: 524 intr = *(int16_t *)data; 525 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0) 526 break; 527 if (intr < 0 || intr >= MAX_INTS) 528 return (EINVAL); 529 intr = next_registered_randintr(intr); 530 if (intr == MAX_INTS) 531 return (ENOENT); 532 *(u_int16_t *)data = intr; 533 break; 534 default: 535 error = ENOTSUP; 536 break; 537 } 538 return (error); 539 } 540 541 static int 542 mm_filter_read(struct knote *kn, long hint) 543 { 544 return (1); 545 } 546 547 static int 548 mm_filter_write(struct knote *kn, long hint) 549 { 550 return (1); 551 } 552 553 static void 554 dummy_filter_detach(struct knote *kn) {} 555 556 static struct filterops random_read_filtops = 557 { FILTEROP_ISFD, NULL, dummy_filter_detach, random_filter_read }; 558 559 static struct filterops mm_read_filtops = 560 { FILTEROP_ISFD, NULL, dummy_filter_detach, mm_filter_read }; 561 562 static struct filterops mm_write_filtops = 563 { FILTEROP_ISFD, NULL, dummy_filter_detach, mm_filter_write }; 564 565 int 566 mmkqfilter(struct dev_kqfilter_args *ap) 567 { 568 struct knote *kn = ap->a_kn; 569 cdev_t dev = ap->a_head.a_dev; 570 571 ap->a_result = 0; 572 switch (kn->kn_filter) { 573 case EVFILT_READ: 574 switch (minor(dev)) { 575 case 3: 576 kn->kn_fop = &random_read_filtops; 577 break; 578 default: 579 kn->kn_fop = &mm_read_filtops; 580 break; 581 } 582 break; 583 case EVFILT_WRITE: 584 kn->kn_fop = &mm_write_filtops; 585 break; 586 default: 587 ap->a_result = EOPNOTSUPP; 588 return (0); 589 } 590 591 return (0); 592 } 593 594 int 595 iszerodev(cdev_t dev) 596 { 597 return (zerodev == dev); 598 } 599 600 static void 601 mem_drvinit(void *unused) 602 { 603 604 /* Initialise memory range handling */ 605 if (mem_range_softc.mr_op != NULL) 606 mem_range_softc.mr_op->init(&mem_range_softc); 607 608 make_dev(&mem_ops, 0, UID_ROOT, GID_KMEM, 0640, "mem"); 609 make_dev(&mem_ops, 1, UID_ROOT, GID_KMEM, 0640, "kmem"); 610 make_dev(&mem_ops, 2, UID_ROOT, GID_WHEEL, 0666, "null"); 611 make_dev(&mem_ops, 3, UID_ROOT, GID_WHEEL, 0644, "random"); 612 make_dev(&mem_ops, 4, UID_ROOT, GID_WHEEL, 0644, "urandom"); 613 zerodev = make_dev(&mem_ops, 12, UID_ROOT, GID_WHEEL, 0666, "zero"); 614 make_dev(&mem_ops, 14, UID_ROOT, GID_WHEEL, 0600, "io"); 615 } 616 617 SYSINIT(memdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,mem_drvinit,NULL) 618 619