| 1 | /* |
| 2 | * Copyright (c) 1988 University of Utah. |
| 3 | * Copyright (c) 1991, 1993 |
| 4 | * The Regents of the University of California. 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. |
| 9 | * |
| 10 | * Redistribution and use in source and binary forms, with or without |
| 11 | * modification, are permitted provided that the following conditions |
| 12 | * are met: |
| 13 | * 1. Redistributions of source code must retain the above copyright |
| 14 | * notice, this list of conditions and the following disclaimer. |
| 15 | * 2. Redistributions in binary form must reproduce the above copyright |
| 16 | * notice, this list of conditions and the following disclaimer in the |
| 17 | * documentation and/or other materials provided with the distribution. |
| 18 | * 3. All advertising materials mentioning features or use of this software |
| 19 | * must display the following acknowledgement: |
| 20 | * This product includes software developed by the University of |
| 21 | * California, Berkeley and its contributors. |
| 22 | * 4. Neither the name of the University nor the names of its contributors |
| 23 | * may be used to endorse or promote products derived from this software |
| 24 | * without specific prior written permission. |
| 25 | * |
| 26 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 27 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 28 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 29 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 30 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 31 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 32 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 33 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 34 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 35 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 36 | * SUCH DAMAGE. |
| 37 | * |
| 38 | * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$ |
| 39 | * |
| 40 | * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94 |
| 41 | * $FreeBSD: src/sys/vm/vm_mmap.c,v 1.108.2.6 2002/07/02 20:06:19 dillon Exp $ |
| 42 | * $DragonFly: src/sys/vm/vm_mmap.c,v 1.30 2006/06/05 07:26:11 dillon Exp $ |
| 43 | */ |
| 44 | |
| 45 | /* |
| 46 | * Mapped file (mmap) interface to VM |
| 47 | */ |
| 48 | |
| 49 | #include <sys/param.h> |
| 50 | #include <sys/kernel.h> |
| 51 | #include <sys/systm.h> |
| 52 | #include <sys/sysproto.h> |
| 53 | #include <sys/filedesc.h> |
| 54 | #include <sys/kern_syscall.h> |
| 55 | #include <sys/proc.h> |
| 56 | #include <sys/resource.h> |
| 57 | #include <sys/resourcevar.h> |
| 58 | #include <sys/vnode.h> |
| 59 | #include <sys/fcntl.h> |
| 60 | #include <sys/file.h> |
| 61 | #include <sys/mman.h> |
| 62 | #include <sys/conf.h> |
| 63 | #include <sys/stat.h> |
| 64 | #include <sys/vmmeter.h> |
| 65 | #include <sys/sysctl.h> |
| 66 | |
| 67 | #include <vm/vm.h> |
| 68 | #include <vm/vm_param.h> |
| 69 | #include <sys/lock.h> |
| 70 | #include <vm/pmap.h> |
| 71 | #include <vm/vm_map.h> |
| 72 | #include <vm/vm_object.h> |
| 73 | #include <vm/vm_page.h> |
| 74 | #include <vm/vm_pager.h> |
| 75 | #include <vm/vm_pageout.h> |
| 76 | #include <vm/vm_extern.h> |
| 77 | #include <vm/vm_page.h> |
| 78 | #include <vm/vm_kern.h> |
| 79 | |
| 80 | #include <sys/file2.h> |
| 81 | #include <sys/thread2.h> |
| 82 | |
| 83 | static int max_proc_mmap; |
| 84 | SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, ""); |
| 85 | |
| 86 | /* |
| 87 | * Set the maximum number of vm_map_entry structures per process. Roughly |
| 88 | * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100 |
| 89 | * of our KVM malloc space still results in generous limits. We want a |
| 90 | * default that is good enough to prevent the kernel running out of resources |
| 91 | * if attacked from compromised user account but generous enough such that |
| 92 | * multi-threaded processes are not unduly inconvenienced. |
| 93 | */ |
| 94 | |
| 95 | static void vmmapentry_rsrc_init (void *); |
| 96 | SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL) |
| 97 | |
| 98 | static void |
| 99 | vmmapentry_rsrc_init(void *dummy) |
| 100 | { |
| 101 | max_proc_mmap = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / |
| 102 | sizeof(struct vm_map_entry); |
| 103 | max_proc_mmap /= 100; |
| 104 | } |
| 105 | |
| 106 | /* ARGSUSED */ |
| 107 | int |
| 108 | sys_sbrk(struct sbrk_args *uap) |
| 109 | { |
| 110 | /* Not yet implemented */ |
| 111 | return (EOPNOTSUPP); |
| 112 | } |
| 113 | |
| 114 | /* |
| 115 | * sstk_args(int incr) |
| 116 | */ |
| 117 | /* ARGSUSED */ |
| 118 | int |
| 119 | sys_sstk(struct sstk_args *uap) |
| 120 | { |
| 121 | /* Not yet implemented */ |
| 122 | return (EOPNOTSUPP); |
| 123 | } |
| 124 | |
| 125 | /* |
| 126 | * mmap_args(void *addr, size_t len, int prot, int flags, int fd, |
| 127 | * long pad, off_t pos) |
| 128 | * |
| 129 | * Memory Map (mmap) system call. Note that the file offset |
| 130 | * and address are allowed to be NOT page aligned, though if |
| 131 | * the MAP_FIXED flag it set, both must have the same remainder |
| 132 | * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not |
| 133 | * page-aligned, the actual mapping starts at trunc_page(addr) |
| 134 | * and the return value is adjusted up by the page offset. |
| 135 | * |
| 136 | * Generally speaking, only character devices which are themselves |
| 137 | * memory-based, such as a video framebuffer, can be mmap'd. Otherwise |
| 138 | * there would be no cache coherency between a descriptor and a VM mapping |
| 139 | * both to the same character device. |
| 140 | * |
| 141 | * Block devices can be mmap'd no matter what they represent. Cache coherency |
| 142 | * is maintained as long as you do not write directly to the underlying |
| 143 | * character device. |
| 144 | */ |
| 145 | |
| 146 | int |
| 147 | kern_mmap(caddr_t uaddr, size_t ulen, int uprot, int uflags, int fd, |
| 148 | off_t upos, void **res) |
| 149 | { |
| 150 | struct thread *td = curthread; |
| 151 | struct proc *p = td->td_proc; |
| 152 | struct file *fp = NULL; |
| 153 | struct vnode *vp; |
| 154 | vm_offset_t addr; |
| 155 | vm_size_t size, pageoff; |
| 156 | vm_prot_t prot, maxprot; |
| 157 | void *handle; |
| 158 | int flags, error; |
| 159 | int disablexworkaround; |
| 160 | off_t pos; |
| 161 | struct vmspace *vms = p->p_vmspace; |
| 162 | vm_object_t obj; |
| 163 | |
| 164 | KKASSERT(p); |
| 165 | |
| 166 | addr = (vm_offset_t) uaddr; |
| 167 | size = ulen; |
| 168 | prot = uprot & VM_PROT_ALL; |
| 169 | flags = uflags; |
| 170 | pos = upos; |
| 171 | |
| 172 | /* make sure mapping fits into numeric range etc */ |
| 173 | if ((ssize_t) ulen < 0 || |
| 174 | ((flags & MAP_ANON) && fd != -1)) |
| 175 | return (EINVAL); |
| 176 | |
| 177 | if (flags & MAP_STACK) { |
| 178 | if ((fd != -1) || |
| 179 | ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) |
| 180 | return (EINVAL); |
| 181 | flags |= MAP_ANON; |
| 182 | pos = 0; |
| 183 | } |
| 184 | |
| 185 | /* |
| 186 | * Align the file position to a page boundary, |
| 187 | * and save its page offset component. |
| 188 | */ |
| 189 | pageoff = (pos & PAGE_MASK); |
| 190 | pos -= pageoff; |
| 191 | |
| 192 | /* Adjust size for rounding (on both ends). */ |
| 193 | size += pageoff; /* low end... */ |
| 194 | size = (vm_size_t) round_page(size); /* hi end */ |
| 195 | |
| 196 | /* |
| 197 | * Check for illegal addresses. Watch out for address wrap... Note |
| 198 | * that VM_*_ADDRESS are not constants due to casts (argh). |
| 199 | */ |
| 200 | if (flags & MAP_FIXED) { |
| 201 | /* |
| 202 | * The specified address must have the same remainder |
| 203 | * as the file offset taken modulo PAGE_SIZE, so it |
| 204 | * should be aligned after adjustment by pageoff. |
| 205 | */ |
| 206 | addr -= pageoff; |
| 207 | if (addr & PAGE_MASK) |
| 208 | return (EINVAL); |
| 209 | /* Address range must be all in user VM space. */ |
| 210 | if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS) |
| 211 | return (EINVAL); |
| 212 | #ifndef i386 |
| 213 | if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS) |
| 214 | return (EINVAL); |
| 215 | #endif |
| 216 | if (addr + size < addr) |
| 217 | return (EINVAL); |
| 218 | } |
| 219 | /* |
| 220 | * XXX for non-fixed mappings where no hint is provided or |
| 221 | * the hint would fall in the potential heap space, |
| 222 | * place it after the end of the largest possible heap. |
| 223 | * |
| 224 | * There should really be a pmap call to determine a reasonable |
| 225 | * location. |
| 226 | */ |
| 227 | else if (addr == 0 || |
| 228 | (addr >= round_page((vm_offset_t)vms->vm_taddr) && |
| 229 | addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz))) |
| 230 | addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz); |
| 231 | |
| 232 | if (flags & MAP_ANON) { |
| 233 | /* |
| 234 | * Mapping blank space is trivial. |
| 235 | */ |
| 236 | handle = NULL; |
| 237 | maxprot = VM_PROT_ALL; |
| 238 | pos = 0; |
| 239 | } else { |
| 240 | /* |
| 241 | * Mapping file, get fp for validation. Obtain vnode and make |
| 242 | * sure it is of appropriate type. |
| 243 | */ |
| 244 | fp = holdfp(p->p_fd, fd, -1); |
| 245 | if (fp == NULL) |
| 246 | return (EBADF); |
| 247 | if (fp->f_type != DTYPE_VNODE) { |
| 248 | error = EINVAL; |
| 249 | goto done; |
| 250 | } |
| 251 | /* |
| 252 | * POSIX shared-memory objects are defined to have |
| 253 | * kernel persistence, and are not defined to support |
| 254 | * read(2)/write(2) -- or even open(2). Thus, we can |
| 255 | * use MAP_ASYNC to trade on-disk coherence for speed. |
| 256 | * The shm_open(3) library routine turns on the FPOSIXSHM |
| 257 | * flag to request this behavior. |
| 258 | */ |
| 259 | if (fp->f_flag & FPOSIXSHM) |
| 260 | flags |= MAP_NOSYNC; |
| 261 | vp = (struct vnode *) fp->f_data; |
| 262 | if (vp->v_type != VREG && vp->v_type != VCHR) { |
| 263 | error = EINVAL; |
| 264 | goto done; |
| 265 | } |
| 266 | if (vp->v_type == VREG) { |
| 267 | /* |
| 268 | * Get the proper underlying object |
| 269 | */ |
| 270 | if ((obj = vp->v_object) == NULL) { |
| 271 | error = EINVAL; |
| 272 | goto done; |
| 273 | } |
| 274 | KKASSERT(vp == (struct vnode *)obj->handle); |
| 275 | } |
| 276 | |
| 277 | /* |
| 278 | * XXX hack to handle use of /dev/zero to map anon memory (ala |
| 279 | * SunOS). |
| 280 | */ |
| 281 | if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) { |
| 282 | handle = NULL; |
| 283 | maxprot = VM_PROT_ALL; |
| 284 | flags |= MAP_ANON; |
| 285 | pos = 0; |
| 286 | } else { |
| 287 | /* |
| 288 | * cdevs does not provide private mappings of any kind. |
| 289 | */ |
| 290 | /* |
| 291 | * However, for XIG X server to continue to work, |
| 292 | * we should allow the superuser to do it anyway. |
| 293 | * We only allow it at securelevel < 1. |
| 294 | * (Because the XIG X server writes directly to video |
| 295 | * memory via /dev/mem, it should never work at any |
| 296 | * other securelevel. |
| 297 | * XXX this will have to go |
| 298 | */ |
| 299 | if (securelevel >= 1) |
| 300 | disablexworkaround = 1; |
| 301 | else |
| 302 | disablexworkaround = suser(td); |
| 303 | if (vp->v_type == VCHR && disablexworkaround && |
| 304 | (flags & (MAP_PRIVATE|MAP_COPY))) { |
| 305 | error = EINVAL; |
| 306 | goto done; |
| 307 | } |
| 308 | /* |
| 309 | * Ensure that file and memory protections are |
| 310 | * compatible. Note that we only worry about |
| 311 | * writability if mapping is shared; in this case, |
| 312 | * current and max prot are dictated by the open file. |
| 313 | * XXX use the vnode instead? Problem is: what |
| 314 | * credentials do we use for determination? What if |
| 315 | * proc does a setuid? |
| 316 | */ |
| 317 | maxprot = VM_PROT_EXECUTE; /* ??? */ |
| 318 | if (fp->f_flag & FREAD) { |
| 319 | maxprot |= VM_PROT_READ; |
| 320 | } else if (prot & PROT_READ) { |
| 321 | error = EACCES; |
| 322 | goto done; |
| 323 | } |
| 324 | /* |
| 325 | * If we are sharing potential changes (either via |
| 326 | * MAP_SHARED or via the implicit sharing of character |
| 327 | * device mappings), and we are trying to get write |
| 328 | * permission although we opened it without asking |
| 329 | * for it, bail out. Check for superuser, only if |
| 330 | * we're at securelevel < 1, to allow the XIG X server |
| 331 | * to continue to work. |
| 332 | */ |
| 333 | |
| 334 | if ((flags & MAP_SHARED) != 0 || |
| 335 | (vp->v_type == VCHR && disablexworkaround)) { |
| 336 | if ((fp->f_flag & FWRITE) != 0) { |
| 337 | struct vattr va; |
| 338 | if ((error = VOP_GETATTR(vp, &va))) { |
| 339 | goto done; |
| 340 | } |
| 341 | if ((va.va_flags & |
| 342 | (IMMUTABLE|APPEND)) == 0) { |
| 343 | maxprot |= VM_PROT_WRITE; |
| 344 | } else if (prot & PROT_WRITE) { |
| 345 | error = EPERM; |
| 346 | goto done; |
| 347 | } |
| 348 | } else if ((prot & PROT_WRITE) != 0) { |
| 349 | error = EACCES; |
| 350 | goto done; |
| 351 | } |
| 352 | } else { |
| 353 | maxprot |= VM_PROT_WRITE; |
| 354 | } |
| 355 | handle = (void *)vp; |
| 356 | } |
| 357 | } |
| 358 | |
| 359 | /* |
| 360 | * Do not allow more then a certain number of vm_map_entry structures |
| 361 | * per process. Scale with the number of rforks sharing the map |
| 362 | * to make the limit reasonable for threads. |
| 363 | */ |
| 364 | if (max_proc_mmap && |
| 365 | vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) { |
| 366 | error = ENOMEM; |
| 367 | goto done; |
| 368 | } |
| 369 | |
| 370 | error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot, |
| 371 | flags, handle, pos); |
| 372 | if (error == 0) |
| 373 | *res = (void *)(addr + pageoff); |
| 374 | done: |
| 375 | if (fp) |
| 376 | fdrop(fp); |
| 377 | return (error); |
| 378 | } |
| 379 | |
| 380 | int |
| 381 | sys_mmap(struct mmap_args *uap) |
| 382 | { |
| 383 | int error; |
| 384 | |
| 385 | error = kern_mmap(uap->addr, uap->len, uap->prot, uap->flags, |
| 386 | uap->fd, uap->pos, &uap->sysmsg_resultp); |
| 387 | |
| 388 | return (error); |
| 389 | } |
| 390 | |
| 391 | /* |
| 392 | * msync_args(void *addr, int len, int flags) |
| 393 | */ |
| 394 | int |
| 395 | sys_msync(struct msync_args *uap) |
| 396 | { |
| 397 | struct proc *p = curproc; |
| 398 | vm_offset_t addr; |
| 399 | vm_size_t size, pageoff; |
| 400 | int flags; |
| 401 | vm_map_t map; |
| 402 | int rv; |
| 403 | |
| 404 | addr = (vm_offset_t) uap->addr; |
| 405 | size = uap->len; |
| 406 | flags = uap->flags; |
| 407 | |
| 408 | pageoff = (addr & PAGE_MASK); |
| 409 | addr -= pageoff; |
| 410 | size += pageoff; |
| 411 | size = (vm_size_t) round_page(size); |
| 412 | if (addr + size < addr) |
| 413 | return(EINVAL); |
| 414 | |
| 415 | if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) |
| 416 | return (EINVAL); |
| 417 | |
| 418 | map = &p->p_vmspace->vm_map; |
| 419 | |
| 420 | /* |
| 421 | * XXX Gak! If size is zero we are supposed to sync "all modified |
| 422 | * pages with the region containing addr". Unfortunately, we don't |
| 423 | * really keep track of individual mmaps so we approximate by flushing |
| 424 | * the range of the map entry containing addr. This can be incorrect |
| 425 | * if the region splits or is coalesced with a neighbor. |
| 426 | */ |
| 427 | if (size == 0) { |
| 428 | vm_map_entry_t entry; |
| 429 | |
| 430 | vm_map_lock_read(map); |
| 431 | rv = vm_map_lookup_entry(map, addr, &entry); |
| 432 | vm_map_unlock_read(map); |
| 433 | if (rv == FALSE) |
| 434 | return (EINVAL); |
| 435 | addr = entry->start; |
| 436 | size = entry->end - entry->start; |
| 437 | } |
| 438 | |
| 439 | /* |
| 440 | * Clean the pages and interpret the return value. |
| 441 | */ |
| 442 | rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0, |
| 443 | (flags & MS_INVALIDATE) != 0); |
| 444 | |
| 445 | switch (rv) { |
| 446 | case KERN_SUCCESS: |
| 447 | break; |
| 448 | case KERN_INVALID_ADDRESS: |
| 449 | return (EINVAL); /* Sun returns ENOMEM? */ |
| 450 | case KERN_FAILURE: |
| 451 | return (EIO); |
| 452 | default: |
| 453 | return (EINVAL); |
| 454 | } |
| 455 | |
| 456 | return (0); |
| 457 | } |
| 458 | |
| 459 | /* |
| 460 | * munmap_args(void *addr, size_t len) |
| 461 | */ |
| 462 | int |
| 463 | sys_munmap(struct munmap_args *uap) |
| 464 | { |
| 465 | struct proc *p = curproc; |
| 466 | vm_offset_t addr; |
| 467 | vm_size_t size, pageoff; |
| 468 | vm_map_t map; |
| 469 | |
| 470 | addr = (vm_offset_t) uap->addr; |
| 471 | size = uap->len; |
| 472 | |
| 473 | pageoff = (addr & PAGE_MASK); |
| 474 | addr -= pageoff; |
| 475 | size += pageoff; |
| 476 | size = (vm_size_t) round_page(size); |
| 477 | if (addr + size < addr) |
| 478 | return(EINVAL); |
| 479 | |
| 480 | if (size == 0) |
| 481 | return (0); |
| 482 | |
| 483 | /* |
| 484 | * Check for illegal addresses. Watch out for address wrap... Note |
| 485 | * that VM_*_ADDRESS are not constants due to casts (argh). |
| 486 | */ |
| 487 | if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS) |
| 488 | return (EINVAL); |
| 489 | #ifndef i386 |
| 490 | if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS) |
| 491 | return (EINVAL); |
| 492 | #endif |
| 493 | map = &p->p_vmspace->vm_map; |
| 494 | /* |
| 495 | * Make sure entire range is allocated. |
| 496 | */ |
| 497 | if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE)) |
| 498 | return (EINVAL); |
| 499 | /* returns nothing but KERN_SUCCESS anyway */ |
| 500 | vm_map_remove(map, addr, addr + size); |
| 501 | return (0); |
| 502 | } |
| 503 | |
| 504 | /* |
| 505 | * mprotect_args(const void *addr, size_t len, int prot) |
| 506 | */ |
| 507 | int |
| 508 | sys_mprotect(struct mprotect_args *uap) |
| 509 | { |
| 510 | struct proc *p = curproc; |
| 511 | vm_offset_t addr; |
| 512 | vm_size_t size, pageoff; |
| 513 | vm_prot_t prot; |
| 514 | |
| 515 | addr = (vm_offset_t) uap->addr; |
| 516 | size = uap->len; |
| 517 | prot = uap->prot & VM_PROT_ALL; |
| 518 | #if defined(VM_PROT_READ_IS_EXEC) |
| 519 | if (prot & VM_PROT_READ) |
| 520 | prot |= VM_PROT_EXECUTE; |
| 521 | #endif |
| 522 | |
| 523 | pageoff = (addr & PAGE_MASK); |
| 524 | addr -= pageoff; |
| 525 | size += pageoff; |
| 526 | size = (vm_size_t) round_page(size); |
| 527 | if (addr + size < addr) |
| 528 | return(EINVAL); |
| 529 | |
| 530 | switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot, |
| 531 | FALSE)) { |
| 532 | case KERN_SUCCESS: |
| 533 | return (0); |
| 534 | case KERN_PROTECTION_FAILURE: |
| 535 | return (EACCES); |
| 536 | } |
| 537 | return (EINVAL); |
| 538 | } |
| 539 | |
| 540 | /* |
| 541 | * minherit_args(void *addr, size_t len, int inherit) |
| 542 | */ |
| 543 | int |
| 544 | sys_minherit(struct minherit_args *uap) |
| 545 | { |
| 546 | struct proc *p = curproc; |
| 547 | vm_offset_t addr; |
| 548 | vm_size_t size, pageoff; |
| 549 | vm_inherit_t inherit; |
| 550 | |
| 551 | addr = (vm_offset_t)uap->addr; |
| 552 | size = uap->len; |
| 553 | inherit = uap->inherit; |
| 554 | |
| 555 | pageoff = (addr & PAGE_MASK); |
| 556 | addr -= pageoff; |
| 557 | size += pageoff; |
| 558 | size = (vm_size_t) round_page(size); |
| 559 | if (addr + size < addr) |
| 560 | return(EINVAL); |
| 561 | |
| 562 | switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size, |
| 563 | inherit)) { |
| 564 | case KERN_SUCCESS: |
| 565 | return (0); |
| 566 | case KERN_PROTECTION_FAILURE: |
| 567 | return (EACCES); |
| 568 | } |
| 569 | return (EINVAL); |
| 570 | } |
| 571 | |
| 572 | /* |
| 573 | * madvise_args(void *addr, size_t len, int behav) |
| 574 | */ |
| 575 | /* ARGSUSED */ |
| 576 | int |
| 577 | sys_madvise(struct madvise_args *uap) |
| 578 | { |
| 579 | struct proc *p = curproc; |
| 580 | vm_offset_t start, end; |
| 581 | |
| 582 | /* |
| 583 | * Check for illegal behavior |
| 584 | */ |
| 585 | if (uap->behav < 0 || uap->behav > MADV_CORE) |
| 586 | return (EINVAL); |
| 587 | /* |
| 588 | * Check for illegal addresses. Watch out for address wrap... Note |
| 589 | * that VM_*_ADDRESS are not constants due to casts (argh). |
| 590 | */ |
| 591 | if (VM_MAXUSER_ADDRESS > 0 && |
| 592 | ((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS) |
| 593 | return (EINVAL); |
| 594 | #ifndef i386 |
| 595 | if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS) |
| 596 | return (EINVAL); |
| 597 | #endif |
| 598 | if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr) |
| 599 | return (EINVAL); |
| 600 | |
| 601 | /* |
| 602 | * Since this routine is only advisory, we default to conservative |
| 603 | * behavior. |
| 604 | */ |
| 605 | start = trunc_page((vm_offset_t) uap->addr); |
| 606 | end = round_page((vm_offset_t) uap->addr + uap->len); |
| 607 | |
| 608 | if (vm_map_madvise(&p->p_vmspace->vm_map, start, end, uap->behav)) |
| 609 | return (EINVAL); |
| 610 | return (0); |
| 611 | } |
| 612 | |
| 613 | /* |
| 614 | * mincore_args(const void *addr, size_t len, char *vec) |
| 615 | */ |
| 616 | /* ARGSUSED */ |
| 617 | int |
| 618 | sys_mincore(struct mincore_args *uap) |
| 619 | { |
| 620 | struct proc *p = curproc; |
| 621 | vm_offset_t addr, first_addr; |
| 622 | vm_offset_t end, cend; |
| 623 | pmap_t pmap; |
| 624 | vm_map_t map; |
| 625 | char *vec; |
| 626 | int error; |
| 627 | int vecindex, lastvecindex; |
| 628 | vm_map_entry_t current; |
| 629 | vm_map_entry_t entry; |
| 630 | int mincoreinfo; |
| 631 | unsigned int timestamp; |
| 632 | |
| 633 | /* |
| 634 | * Make sure that the addresses presented are valid for user |
| 635 | * mode. |
| 636 | */ |
| 637 | first_addr = addr = trunc_page((vm_offset_t) uap->addr); |
| 638 | end = addr + (vm_size_t)round_page(uap->len); |
| 639 | if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS) |
| 640 | return (EINVAL); |
| 641 | if (end < addr) |
| 642 | return (EINVAL); |
| 643 | |
| 644 | /* |
| 645 | * Address of byte vector |
| 646 | */ |
| 647 | vec = uap->vec; |
| 648 | |
| 649 | map = &p->p_vmspace->vm_map; |
| 650 | pmap = vmspace_pmap(p->p_vmspace); |
| 651 | |
| 652 | vm_map_lock_read(map); |
| 653 | RestartScan: |
| 654 | timestamp = map->timestamp; |
| 655 | |
| 656 | if (!vm_map_lookup_entry(map, addr, &entry)) |
| 657 | entry = entry->next; |
| 658 | |
| 659 | /* |
| 660 | * Do this on a map entry basis so that if the pages are not |
| 661 | * in the current processes address space, we can easily look |
| 662 | * up the pages elsewhere. |
| 663 | */ |
| 664 | lastvecindex = -1; |
| 665 | for(current = entry; |
| 666 | (current != &map->header) && (current->start < end); |
| 667 | current = current->next) { |
| 668 | |
| 669 | /* |
| 670 | * ignore submaps (for now) or null objects |
| 671 | */ |
| 672 | if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) || |
| 673 | current->object.vm_object == NULL) |
| 674 | continue; |
| 675 | |
| 676 | /* |
| 677 | * limit this scan to the current map entry and the |
| 678 | * limits for the mincore call |
| 679 | */ |
| 680 | if (addr < current->start) |
| 681 | addr = current->start; |
| 682 | cend = current->end; |
| 683 | if (cend > end) |
| 684 | cend = end; |
| 685 | |
| 686 | /* |
| 687 | * scan this entry one page at a time |
| 688 | */ |
| 689 | while (addr < cend) { |
| 690 | /* |
| 691 | * Check pmap first, it is likely faster, also |
| 692 | * it can provide info as to whether we are the |
| 693 | * one referencing or modifying the page. |
| 694 | */ |
| 695 | mincoreinfo = pmap_mincore(pmap, addr); |
| 696 | if (!mincoreinfo) { |
| 697 | vm_pindex_t pindex; |
| 698 | vm_ooffset_t offset; |
| 699 | vm_page_t m; |
| 700 | |
| 701 | /* |
| 702 | * calculate the page index into the object |
| 703 | */ |
| 704 | offset = current->offset + (addr - current->start); |
| 705 | pindex = OFF_TO_IDX(offset); |
| 706 | |
| 707 | /* |
| 708 | * if the page is resident, then gather |
| 709 | * information about it. spl protection is |
| 710 | * required to maintain the object |
| 711 | * association. And XXX what if the page is |
| 712 | * busy? What's the deal with that? |
| 713 | */ |
| 714 | crit_enter(); |
| 715 | m = vm_page_lookup(current->object.vm_object, |
| 716 | pindex); |
| 717 | if (m && m->valid) { |
| 718 | mincoreinfo = MINCORE_INCORE; |
| 719 | if (m->dirty || |
| 720 | pmap_is_modified(m)) |
| 721 | mincoreinfo |= MINCORE_MODIFIED_OTHER; |
| 722 | if ((m->flags & PG_REFERENCED) || |
| 723 | pmap_ts_referenced(m)) { |
| 724 | vm_page_flag_set(m, PG_REFERENCED); |
| 725 | mincoreinfo |= MINCORE_REFERENCED_OTHER; |
| 726 | } |
| 727 | } |
| 728 | crit_exit(); |
| 729 | } |
| 730 | |
| 731 | /* |
| 732 | * subyte may page fault. In case it needs to modify |
| 733 | * the map, we release the lock. |
| 734 | */ |
| 735 | vm_map_unlock_read(map); |
| 736 | |
| 737 | /* |
| 738 | * calculate index into user supplied byte vector |
| 739 | */ |
| 740 | vecindex = OFF_TO_IDX(addr - first_addr); |
| 741 | |
| 742 | /* |
| 743 | * If we have skipped map entries, we need to make sure that |
| 744 | * the byte vector is zeroed for those skipped entries. |
| 745 | */ |
| 746 | while((lastvecindex + 1) < vecindex) { |
| 747 | error = subyte( vec + lastvecindex, 0); |
| 748 | if (error) { |
| 749 | return (EFAULT); |
| 750 | } |
| 751 | ++lastvecindex; |
| 752 | } |
| 753 | |
| 754 | /* |
| 755 | * Pass the page information to the user |
| 756 | */ |
| 757 | error = subyte( vec + vecindex, mincoreinfo); |
| 758 | if (error) { |
| 759 | return (EFAULT); |
| 760 | } |
| 761 | |
| 762 | /* |
| 763 | * If the map has changed, due to the subyte, the previous |
| 764 | * output may be invalid. |
| 765 | */ |
| 766 | vm_map_lock_read(map); |
| 767 | if (timestamp != map->timestamp) |
| 768 | goto RestartScan; |
| 769 | |
| 770 | lastvecindex = vecindex; |
| 771 | addr += PAGE_SIZE; |
| 772 | } |
| 773 | } |
| 774 | |
| 775 | /* |
| 776 | * subyte may page fault. In case it needs to modify |
| 777 | * the map, we release the lock. |
| 778 | */ |
| 779 | vm_map_unlock_read(map); |
| 780 | |
| 781 | /* |
| 782 | * Zero the last entries in the byte vector. |
| 783 | */ |
| 784 | vecindex = OFF_TO_IDX(end - first_addr); |
| 785 | while((lastvecindex + 1) < vecindex) { |
| 786 | error = subyte( vec + lastvecindex, 0); |
| 787 | if (error) { |
| 788 | return (EFAULT); |
| 789 | } |
| 790 | ++lastvecindex; |
| 791 | } |
| 792 | |
| 793 | /* |
| 794 | * If the map has changed, due to the subyte, the previous |
| 795 | * output may be invalid. |
| 796 | */ |
| 797 | vm_map_lock_read(map); |
| 798 | if (timestamp != map->timestamp) |
| 799 | goto RestartScan; |
| 800 | vm_map_unlock_read(map); |
| 801 | |
| 802 | return (0); |
| 803 | } |
| 804 | |
| 805 | /* |
| 806 | * mlock_args(const void *addr, size_t len) |
| 807 | */ |
| 808 | int |
| 809 | sys_mlock(struct mlock_args *uap) |
| 810 | { |
| 811 | vm_offset_t addr; |
| 812 | vm_size_t size, pageoff; |
| 813 | int error; |
| 814 | struct proc *p = curproc; |
| 815 | |
| 816 | addr = (vm_offset_t) uap->addr; |
| 817 | size = uap->len; |
| 818 | |
| 819 | pageoff = (addr & PAGE_MASK); |
| 820 | addr -= pageoff; |
| 821 | size += pageoff; |
| 822 | size = (vm_size_t) round_page(size); |
| 823 | |
| 824 | /* disable wrap around */ |
| 825 | if (addr + size < addr) |
| 826 | return (EINVAL); |
| 827 | |
| 828 | if (atop(size) + vmstats.v_wire_count > vm_page_max_wired) |
| 829 | return (EAGAIN); |
| 830 | |
| 831 | #ifdef pmap_wired_count |
| 832 | if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) > |
| 833 | p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) |
| 834 | return (ENOMEM); |
| 835 | #else |
| 836 | error = suser_cred(p->p_ucred, 0); |
| 837 | if (error) |
| 838 | return (error); |
| 839 | #endif |
| 840 | |
| 841 | error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE); |
| 842 | return (error == KERN_SUCCESS ? 0 : ENOMEM); |
| 843 | } |
| 844 | |
| 845 | /* |
| 846 | * mlockall_args(int how) |
| 847 | */ |
| 848 | int |
| 849 | sys_mlockall(struct mlockall_args *uap) |
| 850 | { |
| 851 | return 0; |
| 852 | } |
| 853 | |
| 854 | /* |
| 855 | * munlockall_args(void) |
| 856 | */ |
| 857 | int |
| 858 | sys_munlockall(struct munlockall_args *uap) |
| 859 | { |
| 860 | return 0; |
| 861 | } |
| 862 | |
| 863 | /* |
| 864 | * munlock_args(const void *addr, size_t len) |
| 865 | */ |
| 866 | int |
| 867 | sys_munlock(struct munlock_args *uap) |
| 868 | { |
| 869 | struct thread *td = curthread; |
| 870 | struct proc *p = td->td_proc; |
| 871 | vm_offset_t addr; |
| 872 | vm_size_t size, pageoff; |
| 873 | int error; |
| 874 | |
| 875 | addr = (vm_offset_t) uap->addr; |
| 876 | size = uap->len; |
| 877 | |
| 878 | pageoff = (addr & PAGE_MASK); |
| 879 | addr -= pageoff; |
| 880 | size += pageoff; |
| 881 | size = (vm_size_t) round_page(size); |
| 882 | |
| 883 | /* disable wrap around */ |
| 884 | if (addr + size < addr) |
| 885 | return (EINVAL); |
| 886 | |
| 887 | #ifndef pmap_wired_count |
| 888 | error = suser(td); |
| 889 | if (error) |
| 890 | return (error); |
| 891 | #endif |
| 892 | |
| 893 | error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE); |
| 894 | return (error == KERN_SUCCESS ? 0 : ENOMEM); |
| 895 | } |
| 896 | |
| 897 | /* |
| 898 | * Internal version of mmap. |
| 899 | * Currently used by mmap, exec, and sys5 shared memory. |
| 900 | * Handle is either a vnode pointer or NULL for MAP_ANON. |
| 901 | */ |
| 902 | int |
| 903 | vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, |
| 904 | vm_prot_t maxprot, int flags, |
| 905 | void *handle, |
| 906 | vm_ooffset_t foff) |
| 907 | { |
| 908 | boolean_t fitit; |
| 909 | vm_object_t object; |
| 910 | struct vnode *vp = NULL; |
| 911 | objtype_t type; |
| 912 | int rv = KERN_SUCCESS; |
| 913 | off_t objsize; |
| 914 | int docow; |
| 915 | struct thread *td = curthread; /* XXX */ |
| 916 | struct proc *p = td->td_proc; |
| 917 | |
| 918 | KKASSERT(p); |
| 919 | |
| 920 | if (size == 0) |
| 921 | return (0); |
| 922 | |
| 923 | objsize = size = round_page(size); |
| 924 | |
| 925 | if (p->p_vmspace->vm_map.size + size > |
| 926 | p->p_rlimit[RLIMIT_VMEM].rlim_cur) { |
| 927 | return(ENOMEM); |
| 928 | } |
| 929 | |
| 930 | /* |
| 931 | * We currently can only deal with page aligned file offsets. |
| 932 | * The check is here rather than in the syscall because the |
| 933 | * kernel calls this function internally for other mmaping |
| 934 | * operations (such as in exec) and non-aligned offsets will |
| 935 | * cause pmap inconsistencies...so we want to be sure to |
| 936 | * disallow this in all cases. |
| 937 | */ |
| 938 | if (foff & PAGE_MASK) |
| 939 | return (EINVAL); |
| 940 | |
| 941 | if ((flags & MAP_FIXED) == 0) { |
| 942 | fitit = TRUE; |
| 943 | *addr = round_page(*addr); |
| 944 | } else { |
| 945 | if (*addr != trunc_page(*addr)) |
| 946 | return (EINVAL); |
| 947 | fitit = FALSE; |
| 948 | vm_map_remove(map, *addr, *addr + size); |
| 949 | } |
| 950 | |
| 951 | /* |
| 952 | * Lookup/allocate object. |
| 953 | */ |
| 954 | if (flags & MAP_ANON) { |
| 955 | type = OBJT_DEFAULT; |
| 956 | /* |
| 957 | * Unnamed anonymous regions always start at 0. |
| 958 | */ |
| 959 | if (handle == 0) |
| 960 | foff = 0; |
| 961 | } else { |
| 962 | vp = (struct vnode *) handle; |
| 963 | if (vp->v_type == VCHR) { |
| 964 | type = OBJT_DEVICE; |
| 965 | handle = (void *)(intptr_t)vp->v_rdev; |
| 966 | } else { |
| 967 | struct vattr vat; |
| 968 | int error; |
| 969 | |
| 970 | error = VOP_GETATTR(vp, &vat); |
| 971 | if (error) |
| 972 | return (error); |
| 973 | objsize = vat.va_size; |
| 974 | type = OBJT_VNODE; |
| 975 | /* |
| 976 | * if it is a regular file without any references |
| 977 | * we do not need to sync it. |
| 978 | */ |
| 979 | if (vp->v_type == VREG && vat.va_nlink == 0) { |
| 980 | flags |= MAP_NOSYNC; |
| 981 | } |
| 982 | } |
| 983 | } |
| 984 | |
| 985 | if (handle == NULL) { |
| 986 | object = NULL; |
| 987 | docow = 0; |
| 988 | } else { |
| 989 | object = vm_pager_allocate(type, handle, objsize, prot, foff); |
| 990 | if (object == NULL) |
| 991 | return (type == OBJT_DEVICE ? EINVAL : ENOMEM); |
| 992 | docow = MAP_PREFAULT_PARTIAL; |
| 993 | } |
| 994 | |
| 995 | /* |
| 996 | * Force device mappings to be shared. |
| 997 | */ |
| 998 | if (type == OBJT_DEVICE || type == OBJT_PHYS) { |
| 999 | flags &= ~(MAP_PRIVATE|MAP_COPY); |
| 1000 | flags |= MAP_SHARED; |
| 1001 | } |
| 1002 | |
| 1003 | if ((flags & (MAP_ANON|MAP_SHARED)) == 0) |
| 1004 | docow |= MAP_COPY_ON_WRITE; |
| 1005 | if (flags & MAP_NOSYNC) |
| 1006 | docow |= MAP_DISABLE_SYNCER; |
| 1007 | if (flags & MAP_NOCORE) |
| 1008 | docow |= MAP_DISABLE_COREDUMP; |
| 1009 | |
| 1010 | #if defined(VM_PROT_READ_IS_EXEC) |
| 1011 | if (prot & VM_PROT_READ) |
| 1012 | prot |= VM_PROT_EXECUTE; |
| 1013 | |
| 1014 | if (maxprot & VM_PROT_READ) |
| 1015 | maxprot |= VM_PROT_EXECUTE; |
| 1016 | #endif |
| 1017 | |
| 1018 | if (fitit) { |
| 1019 | *addr = pmap_addr_hint(object, *addr, size); |
| 1020 | } |
| 1021 | |
| 1022 | if (flags & MAP_STACK) |
| 1023 | rv = vm_map_stack (map, *addr, size, prot, |
| 1024 | maxprot, docow); |
| 1025 | else |
| 1026 | rv = vm_map_find(map, object, foff, addr, size, fitit, |
| 1027 | prot, maxprot, docow); |
| 1028 | |
| 1029 | if (rv != KERN_SUCCESS) { |
| 1030 | /* |
| 1031 | * Lose the object reference. Will destroy the |
| 1032 | * object if it's an unnamed anonymous mapping |
| 1033 | * or named anonymous without other references. |
| 1034 | */ |
| 1035 | vm_object_deallocate(object); |
| 1036 | goto out; |
| 1037 | } |
| 1038 | |
| 1039 | /* |
| 1040 | * Shared memory is also shared with children. |
| 1041 | */ |
| 1042 | if (flags & (MAP_SHARED|MAP_INHERIT)) { |
| 1043 | rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE); |
| 1044 | if (rv != KERN_SUCCESS) { |
| 1045 | vm_map_remove(map, *addr, *addr + size); |
| 1046 | goto out; |
| 1047 | } |
| 1048 | } |
| 1049 | out: |
| 1050 | switch (rv) { |
| 1051 | case KERN_SUCCESS: |
| 1052 | return (0); |
| 1053 | case KERN_INVALID_ADDRESS: |
| 1054 | case KERN_NO_SPACE: |
| 1055 | return (ENOMEM); |
| 1056 | case KERN_PROTECTION_FAILURE: |
| 1057 | return (EACCES); |
| 1058 | default: |
| 1059 | return (EINVAL); |
| 1060 | } |
| 1061 | } |