1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1988 University of Utah. 5 * Copyright (c) 1991, 1993 6 * The Regents of the University of California. All rights reserved. 7 * 8 * This code is derived from software contributed to Berkeley by 9 * the Systems Programming Group of the University of Utah Computer 10 * Science Department. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$ 37 * 38 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94 39 */ 40 41 /* 42 * Mapped file (mmap) interface to VM 43 */ 44 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 47 48 #include "opt_hwpmc_hooks.h" 49 #include "opt_vm.h" 50 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/capsicum.h> 54 #include <sys/kernel.h> 55 #include <sys/lock.h> 56 #include <sys/mutex.h> 57 #include <sys/sysproto.h> 58 #include <sys/elf.h> 59 #include <sys/filedesc.h> 60 #include <sys/priv.h> 61 #include <sys/proc.h> 62 #include <sys/procctl.h> 63 #include <sys/racct.h> 64 #include <sys/resource.h> 65 #include <sys/resourcevar.h> 66 #include <sys/rwlock.h> 67 #include <sys/sysctl.h> 68 #include <sys/vnode.h> 69 #include <sys/fcntl.h> 70 #include <sys/file.h> 71 #include <sys/mman.h> 72 #include <sys/mount.h> 73 #include <sys/conf.h> 74 #include <sys/stat.h> 75 #include <sys/syscallsubr.h> 76 #include <sys/sysent.h> 77 #include <sys/vmmeter.h> 78 #if defined(__amd64__) || defined(__i386__) /* for i386_read_exec */ 79 #include <machine/md_var.h> 80 #endif 81 82 #include <security/audit/audit.h> 83 #include <security/mac/mac_framework.h> 84 85 #include <vm/vm.h> 86 #include <vm/vm_param.h> 87 #include <vm/pmap.h> 88 #include <vm/vm_map.h> 89 #include <vm/vm_object.h> 90 #include <vm/vm_page.h> 91 #include <vm/vm_pager.h> 92 #include <vm/vm_pageout.h> 93 #include <vm/vm_extern.h> 94 #include <vm/vm_page.h> 95 #include <vm/vnode_pager.h> 96 97 #ifdef HWPMC_HOOKS 98 #include <sys/pmckern.h> 99 #endif 100 101 int old_mlock = 0; 102 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0, 103 "Do not apply RLIMIT_MEMLOCK on mlockall"); 104 static int mincore_mapped = 1; 105 SYSCTL_INT(_vm, OID_AUTO, mincore_mapped, CTLFLAG_RWTUN, &mincore_mapped, 0, 106 "mincore reports mappings, not residency"); 107 static int imply_prot_max = 0; 108 SYSCTL_INT(_vm, OID_AUTO, imply_prot_max, CTLFLAG_RWTUN, &imply_prot_max, 0, 109 "Imply maximum page protections in mmap() when none are specified"); 110 111 _Static_assert(MAXPAGESIZES <= 4, "MINCORE_SUPER too narrow"); 112 113 #ifndef _SYS_SYSPROTO_H_ 114 struct sbrk_args { 115 int incr; 116 }; 117 #endif 118 119 int 120 sys_sbrk(struct thread *td, struct sbrk_args *uap) 121 { 122 /* Not yet implemented */ 123 return (EOPNOTSUPP); 124 } 125 126 #ifndef _SYS_SYSPROTO_H_ 127 struct sstk_args { 128 int incr; 129 }; 130 #endif 131 132 int 133 sys_sstk(struct thread *td, struct sstk_args *uap) 134 { 135 /* Not yet implemented */ 136 return (EOPNOTSUPP); 137 } 138 139 #if defined(COMPAT_43) 140 int 141 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap) 142 { 143 144 td->td_retval[0] = PAGE_SIZE; 145 return (0); 146 } 147 #endif /* COMPAT_43 */ 148 149 /* 150 * Memory Map (mmap) system call. Note that the file offset 151 * and address are allowed to be NOT page aligned, though if 152 * the MAP_FIXED flag it set, both must have the same remainder 153 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not 154 * page-aligned, the actual mapping starts at trunc_page(addr) 155 * and the return value is adjusted up by the page offset. 156 * 157 * Generally speaking, only character devices which are themselves 158 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise 159 * there would be no cache coherency between a descriptor and a VM mapping 160 * both to the same character device. 161 */ 162 #ifndef _SYS_SYSPROTO_H_ 163 struct mmap_args { 164 void *addr; 165 size_t len; 166 int prot; 167 int flags; 168 int fd; 169 long pad; 170 off_t pos; 171 }; 172 #endif 173 174 int 175 sys_mmap(struct thread *td, struct mmap_args *uap) 176 { 177 178 return (kern_mmap(td, &(struct mmap_req){ 179 .mr_hint = (uintptr_t)uap->addr, 180 .mr_len = uap->len, 181 .mr_prot = uap->prot, 182 .mr_flags = uap->flags, 183 .mr_fd = uap->fd, 184 .mr_pos = uap->pos, 185 })); 186 } 187 188 int 189 kern_mmap_maxprot(struct proc *p, int prot) 190 { 191 192 if ((p->p_flag2 & P2_PROTMAX_DISABLE) != 0 || 193 (p->p_fctl0 & NT_FREEBSD_FCTL_PROTMAX_DISABLE) != 0) 194 return (_PROT_ALL); 195 if (((p->p_flag2 & P2_PROTMAX_ENABLE) != 0 || imply_prot_max) && 196 prot != PROT_NONE) 197 return (prot); 198 return (_PROT_ALL); 199 } 200 201 int 202 kern_mmap(struct thread *td, const struct mmap_req *mrp) 203 { 204 struct vmspace *vms; 205 struct file *fp; 206 struct proc *p; 207 off_t pos; 208 vm_offset_t addr, orig_addr; 209 vm_size_t len, pageoff, size; 210 vm_prot_t cap_maxprot; 211 int align, error, fd, flags, max_prot, prot; 212 cap_rights_t rights; 213 mmap_check_fp_fn check_fp_fn; 214 215 orig_addr = addr = mrp->mr_hint; 216 len = mrp->mr_len; 217 prot = mrp->mr_prot; 218 flags = mrp->mr_flags; 219 fd = mrp->mr_fd; 220 pos = mrp->mr_pos; 221 check_fp_fn = mrp->mr_check_fp_fn; 222 223 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0) 224 return (EINVAL); 225 max_prot = PROT_MAX_EXTRACT(prot); 226 prot = PROT_EXTRACT(prot); 227 if (max_prot != 0 && (max_prot & prot) != prot) 228 return (ENOTSUP); 229 230 p = td->td_proc; 231 232 /* 233 * Always honor PROT_MAX if set. If not, default to all 234 * permissions unless we're implying maximum permissions. 235 */ 236 if (max_prot == 0) 237 max_prot = kern_mmap_maxprot(p, prot); 238 239 vms = p->p_vmspace; 240 fp = NULL; 241 AUDIT_ARG_FD(fd); 242 243 /* 244 * Ignore old flags that used to be defined but did not do anything. 245 */ 246 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040); 247 248 /* 249 * Enforce the constraints. 250 * Mapping of length 0 is only allowed for old binaries. 251 * Anonymous mapping shall specify -1 as filedescriptor and 252 * zero position for new code. Be nice to ancient a.out 253 * binaries and correct pos for anonymous mapping, since old 254 * ld.so sometimes issues anonymous map requests with non-zero 255 * pos. 256 */ 257 if (!SV_CURPROC_FLAG(SV_AOUT)) { 258 if ((len == 0 && p->p_osrel >= P_OSREL_MAP_ANON) || 259 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0))) 260 return (EINVAL); 261 } else { 262 if ((flags & MAP_ANON) != 0) 263 pos = 0; 264 } 265 266 if (flags & MAP_STACK) { 267 if ((fd != -1) || 268 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) 269 return (EINVAL); 270 flags |= MAP_ANON; 271 pos = 0; 272 } 273 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE | 274 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE | 275 MAP_PREFAULT_READ | MAP_GUARD | MAP_32BIT | MAP_ALIGNMENT_MASK)) != 0) 276 return (EINVAL); 277 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL) 278 return (EINVAL); 279 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE)) 280 return (EINVAL); 281 if (prot != PROT_NONE && 282 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0) 283 return (EINVAL); 284 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 || 285 pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL | 286 MAP_32BIT | MAP_ALIGNMENT_MASK)) != 0)) 287 return (EINVAL); 288 289 /* 290 * Align the file position to a page boundary, 291 * and save its page offset component. 292 */ 293 pageoff = (pos & PAGE_MASK); 294 pos -= pageoff; 295 296 /* Compute size from len by rounding (on both ends). */ 297 size = len + pageoff; /* low end... */ 298 size = round_page(size); /* hi end */ 299 /* Check for rounding up to zero. */ 300 if (len > size) 301 return (ENOMEM); 302 303 /* Ensure alignment is at least a page and fits in a pointer. */ 304 align = flags & MAP_ALIGNMENT_MASK; 305 if (align != 0 && align != MAP_ALIGNED_SUPER && 306 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY || 307 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT)) 308 return (EINVAL); 309 310 /* 311 * Check for illegal addresses. Watch out for address wrap... Note 312 * that VM_*_ADDRESS are not constants due to casts (argh). 313 */ 314 if (flags & MAP_FIXED) { 315 /* 316 * The specified address must have the same remainder 317 * as the file offset taken modulo PAGE_SIZE, so it 318 * should be aligned after adjustment by pageoff. 319 */ 320 addr -= pageoff; 321 if (addr & PAGE_MASK) 322 return (EINVAL); 323 324 /* Address range must be all in user VM space. */ 325 if (!vm_map_range_valid(&vms->vm_map, addr, addr + size)) 326 return (EINVAL); 327 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR) 328 return (EINVAL); 329 } else if (flags & MAP_32BIT) { 330 /* 331 * For MAP_32BIT, override the hint if it is too high and 332 * do not bother moving the mapping past the heap (since 333 * the heap is usually above 2GB). 334 */ 335 if (addr + size > MAP_32BIT_MAX_ADDR) 336 addr = 0; 337 } else { 338 /* 339 * XXX for non-fixed mappings where no hint is provided or 340 * the hint would fall in the potential heap space, 341 * place it after the end of the largest possible heap. 342 * 343 * For anonymous mappings within the address space of the 344 * calling process, the absence of a hint is handled at a 345 * lower level in order to implement different clustering 346 * strategies for ASLR. 347 */ 348 if (((flags & MAP_ANON) == 0 && addr == 0) || 349 (addr >= round_page((vm_offset_t)vms->vm_taddr) && 350 addr < round_page((vm_offset_t)vms->vm_daddr + 351 lim_max(td, RLIMIT_DATA)))) 352 addr = round_page((vm_offset_t)vms->vm_daddr + 353 lim_max(td, RLIMIT_DATA)); 354 } 355 if (len == 0) { 356 /* 357 * Return success without mapping anything for old 358 * binaries that request a page-aligned mapping of 359 * length 0. For modern binaries, this function 360 * returns an error earlier. 361 */ 362 error = 0; 363 } else if ((flags & MAP_GUARD) != 0) { 364 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE, 365 VM_PROT_NONE, flags, NULL, pos, FALSE, td); 366 } else if ((flags & MAP_ANON) != 0) { 367 /* 368 * Mapping blank space is trivial. 369 * 370 * This relies on VM_PROT_* matching PROT_*. 371 */ 372 error = vm_mmap_object(&vms->vm_map, &addr, size, prot, 373 max_prot, flags, NULL, pos, FALSE, td); 374 } else { 375 /* 376 * Mapping file, get fp for validation and don't let the 377 * descriptor disappear on us if we block. Check capability 378 * rights, but also return the maximum rights to be combined 379 * with maxprot later. 380 */ 381 cap_rights_init_one(&rights, CAP_MMAP); 382 if (prot & PROT_READ) 383 cap_rights_set_one(&rights, CAP_MMAP_R); 384 if ((flags & MAP_SHARED) != 0) { 385 if (prot & PROT_WRITE) 386 cap_rights_set_one(&rights, CAP_MMAP_W); 387 } 388 if (prot & PROT_EXEC) 389 cap_rights_set_one(&rights, CAP_MMAP_X); 390 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp); 391 if (error != 0) 392 goto done; 393 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 && 394 p->p_osrel >= P_OSREL_MAP_FSTRICT) { 395 error = EINVAL; 396 goto done; 397 } 398 if (check_fp_fn != NULL) { 399 error = check_fp_fn(fp, prot, max_prot & cap_maxprot, 400 flags); 401 if (error != 0) 402 goto done; 403 } 404 if (fp->f_ops == &shm_ops && shm_largepage(fp->f_data)) 405 addr = orig_addr; 406 /* This relies on VM_PROT_* matching PROT_*. */ 407 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot, 408 max_prot & cap_maxprot, flags, pos, td); 409 } 410 411 if (error == 0) 412 td->td_retval[0] = addr + pageoff; 413 done: 414 if (fp) 415 fdrop(fp, td); 416 417 return (error); 418 } 419 420 #if defined(COMPAT_FREEBSD6) 421 int 422 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap) 423 { 424 return (kern_mmap(td, &(struct mmap_req){ 425 .mr_hint = (uintptr_t)uap->addr, 426 .mr_len = uap->len, 427 .mr_prot = uap->prot, 428 .mr_flags = uap->flags, 429 .mr_fd = uap->fd, 430 .mr_pos = uap->pos, 431 })); 432 } 433 #endif 434 435 #ifdef COMPAT_43 436 #ifndef _SYS_SYSPROTO_H_ 437 struct ommap_args { 438 caddr_t addr; 439 int len; 440 int prot; 441 int flags; 442 int fd; 443 long pos; 444 }; 445 #endif 446 int 447 ommap(struct thread *td, struct ommap_args *uap) 448 { 449 return (kern_ommap(td, (uintptr_t)uap->addr, uap->len, uap->prot, 450 uap->flags, uap->fd, uap->pos)); 451 } 452 453 int 454 kern_ommap(struct thread *td, uintptr_t hint, int len, int oprot, 455 int oflags, int fd, long pos) 456 { 457 static const char cvtbsdprot[8] = { 458 0, 459 PROT_EXEC, 460 PROT_WRITE, 461 PROT_EXEC | PROT_WRITE, 462 PROT_READ, 463 PROT_EXEC | PROT_READ, 464 PROT_WRITE | PROT_READ, 465 PROT_EXEC | PROT_WRITE | PROT_READ, 466 }; 467 int flags, prot; 468 469 if (len < 0) 470 return (EINVAL); 471 472 #define OMAP_ANON 0x0002 473 #define OMAP_COPY 0x0020 474 #define OMAP_SHARED 0x0010 475 #define OMAP_FIXED 0x0100 476 477 prot = cvtbsdprot[oprot & 0x7]; 478 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__) 479 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) && 480 prot != 0) 481 prot |= PROT_EXEC; 482 #endif 483 flags = 0; 484 if (oflags & OMAP_ANON) 485 flags |= MAP_ANON; 486 if (oflags & OMAP_COPY) 487 flags |= MAP_COPY; 488 if (oflags & OMAP_SHARED) 489 flags |= MAP_SHARED; 490 else 491 flags |= MAP_PRIVATE; 492 if (oflags & OMAP_FIXED) 493 flags |= MAP_FIXED; 494 return (kern_mmap(td, &(struct mmap_req){ 495 .mr_hint = hint, 496 .mr_len = len, 497 .mr_prot = prot, 498 .mr_flags = flags, 499 .mr_fd = fd, 500 .mr_pos = pos, 501 })); 502 } 503 #endif /* COMPAT_43 */ 504 505 #ifndef _SYS_SYSPROTO_H_ 506 struct msync_args { 507 void *addr; 508 size_t len; 509 int flags; 510 }; 511 #endif 512 int 513 sys_msync(struct thread *td, struct msync_args *uap) 514 { 515 516 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags)); 517 } 518 519 int 520 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags) 521 { 522 vm_offset_t addr; 523 vm_size_t pageoff; 524 vm_map_t map; 525 int rv; 526 527 addr = addr0; 528 pageoff = (addr & PAGE_MASK); 529 addr -= pageoff; 530 size += pageoff; 531 size = (vm_size_t) round_page(size); 532 if (addr + size < addr) 533 return (EINVAL); 534 535 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) 536 return (EINVAL); 537 538 map = &td->td_proc->p_vmspace->vm_map; 539 540 /* 541 * Clean the pages and interpret the return value. 542 */ 543 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0, 544 (flags & MS_INVALIDATE) != 0); 545 switch (rv) { 546 case KERN_SUCCESS: 547 return (0); 548 case KERN_INVALID_ADDRESS: 549 return (ENOMEM); 550 case KERN_INVALID_ARGUMENT: 551 return (EBUSY); 552 case KERN_FAILURE: 553 return (EIO); 554 default: 555 return (EINVAL); 556 } 557 } 558 559 #ifndef _SYS_SYSPROTO_H_ 560 struct munmap_args { 561 void *addr; 562 size_t len; 563 }; 564 #endif 565 int 566 sys_munmap(struct thread *td, struct munmap_args *uap) 567 { 568 569 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len)); 570 } 571 572 int 573 kern_munmap(struct thread *td, uintptr_t addr0, size_t size) 574 { 575 #ifdef HWPMC_HOOKS 576 struct pmckern_map_out pkm; 577 vm_map_entry_t entry; 578 bool pmc_handled; 579 #endif 580 vm_offset_t addr, end; 581 vm_size_t pageoff; 582 vm_map_t map; 583 int rv; 584 585 if (size == 0) 586 return (EINVAL); 587 588 addr = addr0; 589 pageoff = (addr & PAGE_MASK); 590 addr -= pageoff; 591 size += pageoff; 592 size = (vm_size_t) round_page(size); 593 end = addr + size; 594 map = &td->td_proc->p_vmspace->vm_map; 595 if (!vm_map_range_valid(map, addr, end)) 596 return (EINVAL); 597 598 vm_map_lock(map); 599 #ifdef HWPMC_HOOKS 600 pmc_handled = false; 601 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) { 602 pmc_handled = true; 603 /* 604 * Inform hwpmc if the address range being unmapped contains 605 * an executable region. 606 */ 607 pkm.pm_address = (uintptr_t) NULL; 608 if (vm_map_lookup_entry(map, addr, &entry)) { 609 for (; entry->start < end; 610 entry = vm_map_entry_succ(entry)) { 611 if (vm_map_check_protection(map, entry->start, 612 entry->end, VM_PROT_EXECUTE) == TRUE) { 613 pkm.pm_address = (uintptr_t) addr; 614 pkm.pm_size = (size_t) size; 615 break; 616 } 617 } 618 } 619 } 620 #endif 621 rv = vm_map_delete(map, addr, end); 622 623 #ifdef HWPMC_HOOKS 624 if (rv == KERN_SUCCESS && __predict_false(pmc_handled)) { 625 /* downgrade the lock to prevent a LOR with the pmc-sx lock */ 626 vm_map_lock_downgrade(map); 627 if (pkm.pm_address != (uintptr_t) NULL) 628 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm); 629 vm_map_unlock_read(map); 630 } else 631 #endif 632 vm_map_unlock(map); 633 634 return (vm_mmap_to_errno(rv)); 635 } 636 637 #ifndef _SYS_SYSPROTO_H_ 638 struct mprotect_args { 639 const void *addr; 640 size_t len; 641 int prot; 642 }; 643 #endif 644 int 645 sys_mprotect(struct thread *td, struct mprotect_args *uap) 646 { 647 648 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, 649 uap->prot, 0)); 650 } 651 652 int 653 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot, 654 int flags) 655 { 656 vm_offset_t addr; 657 vm_size_t pageoff; 658 int vm_error, max_prot; 659 660 addr = addr0; 661 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0) 662 return (EINVAL); 663 max_prot = PROT_MAX_EXTRACT(prot); 664 prot = PROT_EXTRACT(prot); 665 pageoff = (addr & PAGE_MASK); 666 addr -= pageoff; 667 size += pageoff; 668 size = (vm_size_t) round_page(size); 669 #ifdef COMPAT_FREEBSD32 670 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 671 if (((addr + size) & 0xffffffff) < addr) 672 return (EINVAL); 673 } else 674 #endif 675 if (addr + size < addr) 676 return (EINVAL); 677 678 flags |= VM_MAP_PROTECT_SET_PROT; 679 if (max_prot != 0) 680 flags |= VM_MAP_PROTECT_SET_MAXPROT; 681 vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map, 682 addr, addr + size, prot, max_prot, flags); 683 684 switch (vm_error) { 685 case KERN_SUCCESS: 686 return (0); 687 case KERN_PROTECTION_FAILURE: 688 return (EACCES); 689 case KERN_RESOURCE_SHORTAGE: 690 return (ENOMEM); 691 case KERN_OUT_OF_BOUNDS: 692 return (ENOTSUP); 693 } 694 return (EINVAL); 695 } 696 697 #ifndef _SYS_SYSPROTO_H_ 698 struct minherit_args { 699 void *addr; 700 size_t len; 701 int inherit; 702 }; 703 #endif 704 int 705 sys_minherit(struct thread *td, struct minherit_args *uap) 706 { 707 708 return (kern_minherit(td, (uintptr_t)uap->addr, uap->len, 709 uap->inherit)); 710 } 711 712 int 713 kern_minherit(struct thread *td, uintptr_t addr0, size_t len, int inherit0) 714 { 715 vm_offset_t addr; 716 vm_size_t size, pageoff; 717 vm_inherit_t inherit; 718 719 addr = (vm_offset_t)addr0; 720 size = len; 721 inherit = inherit0; 722 723 pageoff = (addr & PAGE_MASK); 724 addr -= pageoff; 725 size += pageoff; 726 size = (vm_size_t) round_page(size); 727 if (addr + size < addr) 728 return (EINVAL); 729 730 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr, 731 addr + size, inherit)) { 732 case KERN_SUCCESS: 733 return (0); 734 case KERN_PROTECTION_FAILURE: 735 return (EACCES); 736 } 737 return (EINVAL); 738 } 739 740 #ifndef _SYS_SYSPROTO_H_ 741 struct madvise_args { 742 void *addr; 743 size_t len; 744 int behav; 745 }; 746 #endif 747 748 int 749 sys_madvise(struct thread *td, struct madvise_args *uap) 750 { 751 752 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav)); 753 } 754 755 int 756 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav) 757 { 758 vm_map_t map; 759 vm_offset_t addr, end, start; 760 int flags; 761 762 /* 763 * Check for our special case, advising the swap pager we are 764 * "immortal." 765 */ 766 if (behav == MADV_PROTECT) { 767 flags = PPROT_SET; 768 return (kern_procctl(td, P_PID, td->td_proc->p_pid, 769 PROC_SPROTECT, &flags)); 770 } 771 772 /* 773 * Check for illegal addresses. Watch out for address wrap... Note 774 * that VM_*_ADDRESS are not constants due to casts (argh). 775 */ 776 map = &td->td_proc->p_vmspace->vm_map; 777 addr = addr0; 778 if (!vm_map_range_valid(map, addr, addr + len)) 779 return (EINVAL); 780 781 /* 782 * Since this routine is only advisory, we default to conservative 783 * behavior. 784 */ 785 start = trunc_page(addr); 786 end = round_page(addr + len); 787 788 /* 789 * vm_map_madvise() checks for illegal values of behav. 790 */ 791 return (vm_map_madvise(map, start, end, behav)); 792 } 793 794 #ifndef _SYS_SYSPROTO_H_ 795 struct mincore_args { 796 const void *addr; 797 size_t len; 798 char *vec; 799 }; 800 #endif 801 802 int 803 sys_mincore(struct thread *td, struct mincore_args *uap) 804 { 805 806 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec)); 807 } 808 809 int 810 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec) 811 { 812 pmap_t pmap; 813 vm_map_t map; 814 vm_map_entry_t current, entry; 815 vm_object_t object; 816 vm_offset_t addr, cend, end, first_addr; 817 vm_paddr_t pa; 818 vm_page_t m; 819 vm_pindex_t pindex; 820 int error, lastvecindex, mincoreinfo, vecindex; 821 unsigned int timestamp; 822 823 /* 824 * Make sure that the addresses presented are valid for user 825 * mode. 826 */ 827 first_addr = addr = trunc_page(addr0); 828 end = round_page(addr0 + len); 829 map = &td->td_proc->p_vmspace->vm_map; 830 if (end > vm_map_max(map) || end < addr) 831 return (ENOMEM); 832 833 pmap = vmspace_pmap(td->td_proc->p_vmspace); 834 835 vm_map_lock_read(map); 836 RestartScan: 837 timestamp = map->timestamp; 838 839 if (!vm_map_lookup_entry(map, addr, &entry)) { 840 vm_map_unlock_read(map); 841 return (ENOMEM); 842 } 843 844 /* 845 * Do this on a map entry basis so that if the pages are not 846 * in the current processes address space, we can easily look 847 * up the pages elsewhere. 848 */ 849 lastvecindex = -1; 850 while (entry->start < end) { 851 /* 852 * check for contiguity 853 */ 854 current = entry; 855 entry = vm_map_entry_succ(current); 856 if (current->end < end && 857 entry->start > current->end) { 858 vm_map_unlock_read(map); 859 return (ENOMEM); 860 } 861 862 /* 863 * ignore submaps (for now) or null objects 864 */ 865 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) || 866 current->object.vm_object == NULL) 867 continue; 868 869 /* 870 * limit this scan to the current map entry and the 871 * limits for the mincore call 872 */ 873 if (addr < current->start) 874 addr = current->start; 875 cend = current->end; 876 if (cend > end) 877 cend = end; 878 879 for (; addr < cend; addr += PAGE_SIZE) { 880 /* 881 * Check pmap first, it is likely faster, also 882 * it can provide info as to whether we are the 883 * one referencing or modifying the page. 884 */ 885 m = NULL; 886 object = NULL; 887 retry: 888 pa = 0; 889 mincoreinfo = pmap_mincore(pmap, addr, &pa); 890 if (mincore_mapped) { 891 /* 892 * We only care about this pmap's 893 * mapping of the page, if any. 894 */ 895 ; 896 } else if (pa != 0) { 897 /* 898 * The page is mapped by this process but not 899 * both accessed and modified. It is also 900 * managed. Acquire the object lock so that 901 * other mappings might be examined. The page's 902 * identity may change at any point before its 903 * object lock is acquired, so re-validate if 904 * necessary. 905 */ 906 m = PHYS_TO_VM_PAGE(pa); 907 while (object == NULL || m->object != object) { 908 if (object != NULL) 909 VM_OBJECT_WUNLOCK(object); 910 object = atomic_load_ptr(&m->object); 911 if (object == NULL) 912 goto retry; 913 VM_OBJECT_WLOCK(object); 914 } 915 if (pa != pmap_extract(pmap, addr)) 916 goto retry; 917 KASSERT(vm_page_all_valid(m), 918 ("mincore: page %p is mapped but invalid", 919 m)); 920 } else if (mincoreinfo == 0) { 921 /* 922 * The page is not mapped by this process. If 923 * the object implements managed pages, then 924 * determine if the page is resident so that 925 * the mappings might be examined. 926 */ 927 if (current->object.vm_object != object) { 928 if (object != NULL) 929 VM_OBJECT_WUNLOCK(object); 930 object = current->object.vm_object; 931 VM_OBJECT_WLOCK(object); 932 } 933 if ((object->flags & OBJ_SWAP) != 0 || 934 object->type == OBJT_VNODE) { 935 pindex = OFF_TO_IDX(current->offset + 936 (addr - current->start)); 937 m = vm_page_lookup(object, pindex); 938 if (m != NULL && vm_page_none_valid(m)) 939 m = NULL; 940 if (m != NULL) 941 mincoreinfo = MINCORE_INCORE; 942 } 943 } 944 if (m != NULL) { 945 VM_OBJECT_ASSERT_WLOCKED(m->object); 946 947 /* Examine other mappings of the page. */ 948 if (m->dirty == 0 && pmap_is_modified(m)) 949 vm_page_dirty(m); 950 if (m->dirty != 0) 951 mincoreinfo |= MINCORE_MODIFIED_OTHER; 952 953 /* 954 * The first test for PGA_REFERENCED is an 955 * optimization. The second test is 956 * required because a concurrent pmap 957 * operation could clear the last reference 958 * and set PGA_REFERENCED before the call to 959 * pmap_is_referenced(). 960 */ 961 if ((m->a.flags & PGA_REFERENCED) != 0 || 962 pmap_is_referenced(m) || 963 (m->a.flags & PGA_REFERENCED) != 0) 964 mincoreinfo |= MINCORE_REFERENCED_OTHER; 965 } 966 if (object != NULL) 967 VM_OBJECT_WUNLOCK(object); 968 969 /* 970 * subyte may page fault. In case it needs to modify 971 * the map, we release the lock. 972 */ 973 vm_map_unlock_read(map); 974 975 /* 976 * calculate index into user supplied byte vector 977 */ 978 vecindex = atop(addr - first_addr); 979 980 /* 981 * If we have skipped map entries, we need to make sure that 982 * the byte vector is zeroed for those skipped entries. 983 */ 984 while ((lastvecindex + 1) < vecindex) { 985 ++lastvecindex; 986 error = subyte(vec + lastvecindex, 0); 987 if (error) { 988 error = EFAULT; 989 goto done2; 990 } 991 } 992 993 /* 994 * Pass the page information to the user 995 */ 996 error = subyte(vec + vecindex, mincoreinfo); 997 if (error) { 998 error = EFAULT; 999 goto done2; 1000 } 1001 1002 /* 1003 * If the map has changed, due to the subyte, the previous 1004 * output may be invalid. 1005 */ 1006 vm_map_lock_read(map); 1007 if (timestamp != map->timestamp) 1008 goto RestartScan; 1009 1010 lastvecindex = vecindex; 1011 } 1012 } 1013 1014 /* 1015 * subyte may page fault. In case it needs to modify 1016 * the map, we release the lock. 1017 */ 1018 vm_map_unlock_read(map); 1019 1020 /* 1021 * Zero the last entries in the byte vector. 1022 */ 1023 vecindex = atop(end - first_addr); 1024 while ((lastvecindex + 1) < vecindex) { 1025 ++lastvecindex; 1026 error = subyte(vec + lastvecindex, 0); 1027 if (error) { 1028 error = EFAULT; 1029 goto done2; 1030 } 1031 } 1032 1033 /* 1034 * If the map has changed, due to the subyte, the previous 1035 * output may be invalid. 1036 */ 1037 vm_map_lock_read(map); 1038 if (timestamp != map->timestamp) 1039 goto RestartScan; 1040 vm_map_unlock_read(map); 1041 done2: 1042 return (error); 1043 } 1044 1045 #ifndef _SYS_SYSPROTO_H_ 1046 struct mlock_args { 1047 const void *addr; 1048 size_t len; 1049 }; 1050 #endif 1051 int 1052 sys_mlock(struct thread *td, struct mlock_args *uap) 1053 { 1054 1055 return (kern_mlock(td->td_proc, td->td_ucred, 1056 __DECONST(uintptr_t, uap->addr), uap->len)); 1057 } 1058 1059 int 1060 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len) 1061 { 1062 vm_offset_t addr, end, last, start; 1063 vm_size_t npages, size; 1064 vm_map_t map; 1065 unsigned long nsize; 1066 int error; 1067 1068 error = priv_check_cred(cred, PRIV_VM_MLOCK); 1069 if (error) 1070 return (error); 1071 addr = addr0; 1072 size = len; 1073 last = addr + size; 1074 start = trunc_page(addr); 1075 end = round_page(last); 1076 if (last < addr || end < addr) 1077 return (EINVAL); 1078 npages = atop(end - start); 1079 if (npages > vm_page_max_user_wired) 1080 return (ENOMEM); 1081 map = &proc->p_vmspace->vm_map; 1082 PROC_LOCK(proc); 1083 nsize = ptoa(npages + pmap_wired_count(map->pmap)); 1084 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) { 1085 PROC_UNLOCK(proc); 1086 return (ENOMEM); 1087 } 1088 PROC_UNLOCK(proc); 1089 #ifdef RACCT 1090 if (racct_enable) { 1091 PROC_LOCK(proc); 1092 error = racct_set(proc, RACCT_MEMLOCK, nsize); 1093 PROC_UNLOCK(proc); 1094 if (error != 0) 1095 return (ENOMEM); 1096 } 1097 #endif 1098 error = vm_map_wire(map, start, end, 1099 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 1100 #ifdef RACCT 1101 if (racct_enable && error != KERN_SUCCESS) { 1102 PROC_LOCK(proc); 1103 racct_set(proc, RACCT_MEMLOCK, 1104 ptoa(pmap_wired_count(map->pmap))); 1105 PROC_UNLOCK(proc); 1106 } 1107 #endif 1108 switch (error) { 1109 case KERN_SUCCESS: 1110 return (0); 1111 case KERN_INVALID_ARGUMENT: 1112 return (EINVAL); 1113 default: 1114 return (ENOMEM); 1115 } 1116 } 1117 1118 #ifndef _SYS_SYSPROTO_H_ 1119 struct mlockall_args { 1120 int how; 1121 }; 1122 #endif 1123 1124 int 1125 sys_mlockall(struct thread *td, struct mlockall_args *uap) 1126 { 1127 vm_map_t map; 1128 int error; 1129 1130 map = &td->td_proc->p_vmspace->vm_map; 1131 error = priv_check(td, PRIV_VM_MLOCK); 1132 if (error) 1133 return (error); 1134 1135 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0)) 1136 return (EINVAL); 1137 1138 /* 1139 * If wiring all pages in the process would cause it to exceed 1140 * a hard resource limit, return ENOMEM. 1141 */ 1142 if (!old_mlock && uap->how & MCL_CURRENT) { 1143 if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) 1144 return (ENOMEM); 1145 } 1146 #ifdef RACCT 1147 if (racct_enable) { 1148 PROC_LOCK(td->td_proc); 1149 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size); 1150 PROC_UNLOCK(td->td_proc); 1151 if (error != 0) 1152 return (ENOMEM); 1153 } 1154 #endif 1155 1156 if (uap->how & MCL_FUTURE) { 1157 vm_map_lock(map); 1158 vm_map_modflags(map, MAP_WIREFUTURE, 0); 1159 vm_map_unlock(map); 1160 error = 0; 1161 } 1162 1163 if (uap->how & MCL_CURRENT) { 1164 /* 1165 * P1003.1-2001 mandates that all currently mapped pages 1166 * will be memory resident and locked (wired) upon return 1167 * from mlockall(). vm_map_wire() will wire pages, by 1168 * calling vm_fault_wire() for each page in the region. 1169 */ 1170 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map), 1171 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); 1172 if (error == KERN_SUCCESS) 1173 error = 0; 1174 else if (error == KERN_RESOURCE_SHORTAGE) 1175 error = ENOMEM; 1176 else 1177 error = EAGAIN; 1178 } 1179 #ifdef RACCT 1180 if (racct_enable && error != KERN_SUCCESS) { 1181 PROC_LOCK(td->td_proc); 1182 racct_set(td->td_proc, RACCT_MEMLOCK, 1183 ptoa(pmap_wired_count(map->pmap))); 1184 PROC_UNLOCK(td->td_proc); 1185 } 1186 #endif 1187 1188 return (error); 1189 } 1190 1191 #ifndef _SYS_SYSPROTO_H_ 1192 struct munlockall_args { 1193 register_t dummy; 1194 }; 1195 #endif 1196 1197 int 1198 sys_munlockall(struct thread *td, struct munlockall_args *uap) 1199 { 1200 vm_map_t map; 1201 int error; 1202 1203 map = &td->td_proc->p_vmspace->vm_map; 1204 error = priv_check(td, PRIV_VM_MUNLOCK); 1205 if (error) 1206 return (error); 1207 1208 /* Clear the MAP_WIREFUTURE flag from this vm_map. */ 1209 vm_map_lock(map); 1210 vm_map_modflags(map, 0, MAP_WIREFUTURE); 1211 vm_map_unlock(map); 1212 1213 /* Forcibly unwire all pages. */ 1214 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map), 1215 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); 1216 #ifdef RACCT 1217 if (racct_enable && error == KERN_SUCCESS) { 1218 PROC_LOCK(td->td_proc); 1219 racct_set(td->td_proc, RACCT_MEMLOCK, 0); 1220 PROC_UNLOCK(td->td_proc); 1221 } 1222 #endif 1223 1224 return (error); 1225 } 1226 1227 #ifndef _SYS_SYSPROTO_H_ 1228 struct munlock_args { 1229 const void *addr; 1230 size_t len; 1231 }; 1232 #endif 1233 int 1234 sys_munlock(struct thread *td, struct munlock_args *uap) 1235 { 1236 1237 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len)); 1238 } 1239 1240 int 1241 kern_munlock(struct thread *td, uintptr_t addr0, size_t size) 1242 { 1243 vm_offset_t addr, end, last, start; 1244 #ifdef RACCT 1245 vm_map_t map; 1246 #endif 1247 int error; 1248 1249 error = priv_check(td, PRIV_VM_MUNLOCK); 1250 if (error) 1251 return (error); 1252 addr = addr0; 1253 last = addr + size; 1254 start = trunc_page(addr); 1255 end = round_page(last); 1256 if (last < addr || end < addr) 1257 return (EINVAL); 1258 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end, 1259 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 1260 #ifdef RACCT 1261 if (racct_enable && error == KERN_SUCCESS) { 1262 PROC_LOCK(td->td_proc); 1263 map = &td->td_proc->p_vmspace->vm_map; 1264 racct_set(td->td_proc, RACCT_MEMLOCK, 1265 ptoa(pmap_wired_count(map->pmap))); 1266 PROC_UNLOCK(td->td_proc); 1267 } 1268 #endif 1269 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1270 } 1271 1272 /* 1273 * vm_mmap_vnode() 1274 * 1275 * Helper function for vm_mmap. Perform sanity check specific for mmap 1276 * operations on vnodes. 1277 */ 1278 int 1279 vm_mmap_vnode(struct thread *td, vm_size_t objsize, 1280 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp, 1281 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp, 1282 boolean_t *writecounted) 1283 { 1284 struct vattr va; 1285 vm_object_t obj; 1286 vm_ooffset_t foff; 1287 struct ucred *cred; 1288 int error, flags; 1289 bool writex; 1290 1291 cred = td->td_ucred; 1292 writex = (*maxprotp & VM_PROT_WRITE) != 0 && 1293 (*flagsp & MAP_SHARED) != 0; 1294 if ((error = vget(vp, LK_SHARED)) != 0) 1295 return (error); 1296 AUDIT_ARG_VNODE1(vp); 1297 foff = *foffp; 1298 flags = *flagsp; 1299 obj = vp->v_object; 1300 if (vp->v_type == VREG) { 1301 /* 1302 * Get the proper underlying object 1303 */ 1304 if (obj == NULL) { 1305 error = EINVAL; 1306 goto done; 1307 } 1308 if (obj->type == OBJT_VNODE && obj->handle != vp) { 1309 vput(vp); 1310 vp = (struct vnode *)obj->handle; 1311 /* 1312 * Bypass filesystems obey the mpsafety of the 1313 * underlying fs. Tmpfs never bypasses. 1314 */ 1315 error = vget(vp, LK_SHARED); 1316 if (error != 0) 1317 return (error); 1318 } 1319 if (writex) { 1320 *writecounted = TRUE; 1321 vm_pager_update_writecount(obj, 0, objsize); 1322 } 1323 } else { 1324 error = EINVAL; 1325 goto done; 1326 } 1327 if ((error = VOP_GETATTR(vp, &va, cred))) 1328 goto done; 1329 #ifdef MAC 1330 /* This relies on VM_PROT_* matching PROT_*. */ 1331 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags); 1332 if (error != 0) 1333 goto done; 1334 #endif 1335 if ((flags & MAP_SHARED) != 0) { 1336 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) { 1337 if (prot & VM_PROT_WRITE) { 1338 error = EPERM; 1339 goto done; 1340 } 1341 *maxprotp &= ~VM_PROT_WRITE; 1342 } 1343 } 1344 /* 1345 * If it is a regular file without any references 1346 * we do not need to sync it. 1347 * Adjust object size to be the size of actual file. 1348 */ 1349 objsize = round_page(va.va_size); 1350 if (va.va_nlink == 0) 1351 flags |= MAP_NOSYNC; 1352 if (obj->type == OBJT_VNODE) { 1353 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, 1354 cred); 1355 if (obj == NULL) { 1356 error = ENOMEM; 1357 goto done; 1358 } 1359 } else { 1360 KASSERT((obj->flags & OBJ_SWAP) != 0, ("wrong object type")); 1361 vm_object_reference(obj); 1362 #if VM_NRESERVLEVEL > 0 1363 if ((obj->flags & OBJ_COLORED) == 0) { 1364 VM_OBJECT_WLOCK(obj); 1365 vm_object_color(obj, 0); 1366 VM_OBJECT_WUNLOCK(obj); 1367 } 1368 #endif 1369 } 1370 *objp = obj; 1371 *flagsp = flags; 1372 1373 VOP_MMAPPED(vp); 1374 1375 done: 1376 if (error != 0 && *writecounted) { 1377 *writecounted = FALSE; 1378 vm_pager_update_writecount(obj, objsize, 0); 1379 } 1380 vput(vp); 1381 return (error); 1382 } 1383 1384 /* 1385 * vm_mmap_cdev() 1386 * 1387 * Helper function for vm_mmap. Perform sanity check specific for mmap 1388 * operations on cdevs. 1389 */ 1390 int 1391 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot, 1392 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw, 1393 vm_ooffset_t *foff, vm_object_t *objp) 1394 { 1395 vm_object_t obj; 1396 int error, flags; 1397 1398 flags = *flagsp; 1399 1400 if (dsw->d_flags & D_MMAP_ANON) { 1401 *objp = NULL; 1402 *foff = 0; 1403 *maxprotp = VM_PROT_ALL; 1404 *flagsp |= MAP_ANON; 1405 return (0); 1406 } 1407 /* 1408 * cdevs do not provide private mappings of any kind. 1409 */ 1410 if ((*maxprotp & VM_PROT_WRITE) == 0 && 1411 (prot & VM_PROT_WRITE) != 0) 1412 return (EACCES); 1413 if (flags & (MAP_PRIVATE|MAP_COPY)) 1414 return (EINVAL); 1415 /* 1416 * Force device mappings to be shared. 1417 */ 1418 flags |= MAP_SHARED; 1419 #ifdef MAC_XXX 1420 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot); 1421 if (error != 0) 1422 return (error); 1423 #endif 1424 /* 1425 * First, try d_mmap_single(). If that is not implemented 1426 * (returns ENODEV), fall back to using the device pager. 1427 * Note that d_mmap_single() must return a reference to the 1428 * object (it needs to bump the reference count of the object 1429 * it returns somehow). 1430 * 1431 * XXX assumes VM_PROT_* == PROT_* 1432 */ 1433 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot); 1434 if (error != ENODEV) 1435 return (error); 1436 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff, 1437 td->td_ucred); 1438 if (obj == NULL) 1439 return (EINVAL); 1440 *objp = obj; 1441 *flagsp = flags; 1442 return (0); 1443 } 1444 1445 int 1446 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1447 vm_prot_t maxprot, int flags, 1448 objtype_t handle_type, void *handle, 1449 vm_ooffset_t foff) 1450 { 1451 vm_object_t object; 1452 struct thread *td = curthread; 1453 int error; 1454 boolean_t writecounted; 1455 1456 if (size == 0) 1457 return (EINVAL); 1458 1459 size = round_page(size); 1460 object = NULL; 1461 writecounted = FALSE; 1462 1463 switch (handle_type) { 1464 case OBJT_DEVICE: { 1465 struct cdevsw *dsw; 1466 struct cdev *cdev; 1467 int ref; 1468 1469 cdev = handle; 1470 dsw = dev_refthread(cdev, &ref); 1471 if (dsw == NULL) 1472 return (ENXIO); 1473 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev, 1474 dsw, &foff, &object); 1475 dev_relthread(cdev, ref); 1476 break; 1477 } 1478 case OBJT_VNODE: 1479 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, 1480 handle, &foff, &object, &writecounted); 1481 break; 1482 default: 1483 error = EINVAL; 1484 break; 1485 } 1486 if (error) 1487 return (error); 1488 1489 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1490 foff, writecounted, td); 1491 if (error != 0 && object != NULL) { 1492 /* 1493 * If this mapping was accounted for in the vnode's 1494 * writecount, then undo that now. 1495 */ 1496 if (writecounted) 1497 vm_pager_release_writecount(object, 0, size); 1498 vm_object_deallocate(object); 1499 } 1500 return (error); 1501 } 1502 1503 int 1504 kern_mmap_racct_check(struct thread *td, vm_map_t map, vm_size_t size) 1505 { 1506 int error; 1507 1508 RACCT_PROC_LOCK(td->td_proc); 1509 if (map->size + size > lim_cur(td, RLIMIT_VMEM)) { 1510 RACCT_PROC_UNLOCK(td->td_proc); 1511 return (ENOMEM); 1512 } 1513 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) { 1514 RACCT_PROC_UNLOCK(td->td_proc); 1515 return (ENOMEM); 1516 } 1517 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 1518 if (ptoa(pmap_wired_count(map->pmap)) + size > 1519 lim_cur(td, RLIMIT_MEMLOCK)) { 1520 racct_set_force(td->td_proc, RACCT_VMEM, map->size); 1521 RACCT_PROC_UNLOCK(td->td_proc); 1522 return (ENOMEM); 1523 } 1524 error = racct_set(td->td_proc, RACCT_MEMLOCK, 1525 ptoa(pmap_wired_count(map->pmap)) + size); 1526 if (error != 0) { 1527 racct_set_force(td->td_proc, RACCT_VMEM, map->size); 1528 RACCT_PROC_UNLOCK(td->td_proc); 1529 return (error); 1530 } 1531 } 1532 RACCT_PROC_UNLOCK(td->td_proc); 1533 return (0); 1534 } 1535 1536 /* 1537 * Internal version of mmap that maps a specific VM object into an 1538 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap. 1539 */ 1540 int 1541 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1542 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff, 1543 boolean_t writecounted, struct thread *td) 1544 { 1545 vm_offset_t default_addr, max_addr; 1546 int docow, error, findspace, rv; 1547 bool curmap, fitit; 1548 1549 curmap = map == &td->td_proc->p_vmspace->vm_map; 1550 if (curmap) { 1551 error = kern_mmap_racct_check(td, map, size); 1552 if (error != 0) 1553 return (error); 1554 } 1555 1556 /* 1557 * We currently can only deal with page aligned file offsets. 1558 * The mmap() system call already enforces this by subtracting 1559 * the page offset from the file offset, but checking here 1560 * catches errors in device drivers (e.g. d_single_mmap() 1561 * callbacks) and other internal mapping requests (such as in 1562 * exec). 1563 */ 1564 if (foff & PAGE_MASK) 1565 return (EINVAL); 1566 1567 if ((flags & MAP_FIXED) == 0) { 1568 fitit = true; 1569 *addr = round_page(*addr); 1570 } else { 1571 if (*addr != trunc_page(*addr)) 1572 return (EINVAL); 1573 fitit = false; 1574 } 1575 1576 if (flags & MAP_ANON) { 1577 if (object != NULL || foff != 0) 1578 return (EINVAL); 1579 docow = 0; 1580 } else if (flags & MAP_PREFAULT_READ) 1581 docow = MAP_PREFAULT; 1582 else 1583 docow = MAP_PREFAULT_PARTIAL; 1584 1585 if ((flags & (MAP_ANON|MAP_SHARED)) == 0) 1586 docow |= MAP_COPY_ON_WRITE; 1587 if (flags & MAP_NOSYNC) 1588 docow |= MAP_DISABLE_SYNCER; 1589 if (flags & MAP_NOCORE) 1590 docow |= MAP_DISABLE_COREDUMP; 1591 /* Shared memory is also shared with children. */ 1592 if (flags & MAP_SHARED) 1593 docow |= MAP_INHERIT_SHARE; 1594 if (writecounted) 1595 docow |= MAP_WRITECOUNT; 1596 if (flags & MAP_STACK) { 1597 if (object != NULL) 1598 return (EINVAL); 1599 docow |= MAP_STACK_GROWS_DOWN; 1600 } 1601 if ((flags & MAP_EXCL) != 0) 1602 docow |= MAP_CHECK_EXCL; 1603 if ((flags & MAP_GUARD) != 0) 1604 docow |= MAP_CREATE_GUARD; 1605 1606 if (fitit) { 1607 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER) 1608 findspace = VMFS_SUPER_SPACE; 1609 else if ((flags & MAP_ALIGNMENT_MASK) != 0) 1610 findspace = VMFS_ALIGNED_SPACE(flags >> 1611 MAP_ALIGNMENT_SHIFT); 1612 else 1613 findspace = VMFS_OPTIMAL_SPACE; 1614 max_addr = 0; 1615 if ((flags & MAP_32BIT) != 0) 1616 max_addr = MAP_32BIT_MAX_ADDR; 1617 if (curmap) { 1618 default_addr = 1619 round_page((vm_offset_t)td->td_proc->p_vmspace-> 1620 vm_daddr + lim_max(td, RLIMIT_DATA)); 1621 if ((flags & MAP_32BIT) != 0) 1622 default_addr = 0; 1623 rv = vm_map_find_min(map, object, foff, addr, size, 1624 default_addr, max_addr, findspace, prot, maxprot, 1625 docow); 1626 } else { 1627 rv = vm_map_find(map, object, foff, addr, size, 1628 max_addr, findspace, prot, maxprot, docow); 1629 } 1630 } else { 1631 rv = vm_map_fixed(map, object, foff, *addr, size, 1632 prot, maxprot, docow); 1633 } 1634 1635 if (rv == KERN_SUCCESS) { 1636 /* 1637 * If the process has requested that all future mappings 1638 * be wired, then heed this. 1639 */ 1640 if ((map->flags & MAP_WIREFUTURE) != 0) { 1641 vm_map_lock(map); 1642 if ((map->flags & MAP_WIREFUTURE) != 0) 1643 (void)vm_map_wire_locked(map, *addr, 1644 *addr + size, VM_MAP_WIRE_USER | 1645 ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK : 1646 VM_MAP_WIRE_NOHOLES)); 1647 vm_map_unlock(map); 1648 } 1649 } 1650 return (vm_mmap_to_errno(rv)); 1651 } 1652 1653 /* 1654 * Translate a Mach VM return code to zero on success or the appropriate errno 1655 * on failure. 1656 */ 1657 int 1658 vm_mmap_to_errno(int rv) 1659 { 1660 1661 switch (rv) { 1662 case KERN_SUCCESS: 1663 return (0); 1664 case KERN_INVALID_ADDRESS: 1665 case KERN_NO_SPACE: 1666 return (ENOMEM); 1667 case KERN_PROTECTION_FAILURE: 1668 return (EACCES); 1669 default: 1670 return (EINVAL); 1671 } 1672 } 1673