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/filedesc.h> 59 #include <sys/priv.h> 60 #include <sys/proc.h> 61 #include <sys/procctl.h> 62 #include <sys/racct.h> 63 #include <sys/resource.h> 64 #include <sys/resourcevar.h> 65 #include <sys/rwlock.h> 66 #include <sys/sysctl.h> 67 #include <sys/vnode.h> 68 #include <sys/fcntl.h> 69 #include <sys/file.h> 70 #include <sys/mman.h> 71 #include <sys/mount.h> 72 #include <sys/conf.h> 73 #include <sys/stat.h> 74 #include <sys/syscallsubr.h> 75 #include <sys/sysent.h> 76 #include <sys/vmmeter.h> 77 78 #include <security/audit/audit.h> 79 #include <security/mac/mac_framework.h> 80 81 #include <vm/vm.h> 82 #include <vm/vm_param.h> 83 #include <vm/pmap.h> 84 #include <vm/vm_map.h> 85 #include <vm/vm_object.h> 86 #include <vm/vm_page.h> 87 #include <vm/vm_pager.h> 88 #include <vm/vm_pageout.h> 89 #include <vm/vm_extern.h> 90 #include <vm/vm_page.h> 91 #include <vm/vnode_pager.h> 92 93 #ifdef HWPMC_HOOKS 94 #include <sys/pmckern.h> 95 #endif 96 97 int old_mlock = 0; 98 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0, 99 "Do not apply RLIMIT_MEMLOCK on mlockall"); 100 101 #ifdef MAP_32BIT 102 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31) 103 #endif 104 105 #ifndef _SYS_SYSPROTO_H_ 106 struct sbrk_args { 107 int incr; 108 }; 109 #endif 110 111 int 112 sys_sbrk(struct thread *td, struct sbrk_args *uap) 113 { 114 /* Not yet implemented */ 115 return (EOPNOTSUPP); 116 } 117 118 #ifndef _SYS_SYSPROTO_H_ 119 struct sstk_args { 120 int incr; 121 }; 122 #endif 123 124 int 125 sys_sstk(struct thread *td, struct sstk_args *uap) 126 { 127 /* Not yet implemented */ 128 return (EOPNOTSUPP); 129 } 130 131 #if defined(COMPAT_43) 132 #ifndef _SYS_SYSPROTO_H_ 133 struct getpagesize_args { 134 int dummy; 135 }; 136 #endif 137 138 int 139 ogetpagesize(struct thread *td, struct getpagesize_args *uap) 140 { 141 142 td->td_retval[0] = PAGE_SIZE; 143 return (0); 144 } 145 #endif /* COMPAT_43 */ 146 147 148 /* 149 * Memory Map (mmap) system call. Note that the file offset 150 * and address are allowed to be NOT page aligned, though if 151 * the MAP_FIXED flag it set, both must have the same remainder 152 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not 153 * page-aligned, the actual mapping starts at trunc_page(addr) 154 * and the return value is adjusted up by the page offset. 155 * 156 * Generally speaking, only character devices which are themselves 157 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise 158 * there would be no cache coherency between a descriptor and a VM mapping 159 * both to the same character device. 160 */ 161 #ifndef _SYS_SYSPROTO_H_ 162 struct mmap_args { 163 void *addr; 164 size_t len; 165 int prot; 166 int flags; 167 int fd; 168 long pad; 169 off_t pos; 170 }; 171 #endif 172 173 int 174 sys_mmap(struct thread *td, struct mmap_args *uap) 175 { 176 177 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot, 178 uap->flags, uap->fd, uap->pos)); 179 } 180 181 int 182 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags, 183 int fd, off_t pos) 184 { 185 struct vmspace *vms; 186 struct file *fp; 187 vm_offset_t addr; 188 vm_size_t pageoff; 189 vm_prot_t cap_maxprot; 190 int align, error; 191 cap_rights_t rights; 192 193 vms = td->td_proc->p_vmspace; 194 fp = NULL; 195 AUDIT_ARG_FD(fd); 196 addr = addr0; 197 198 /* 199 * Ignore old flags that used to be defined but did not do anything. 200 */ 201 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040); 202 203 /* 204 * Enforce the constraints. 205 * Mapping of length 0 is only allowed for old binaries. 206 * Anonymous mapping shall specify -1 as filedescriptor and 207 * zero position for new code. Be nice to ancient a.out 208 * binaries and correct pos for anonymous mapping, since old 209 * ld.so sometimes issues anonymous map requests with non-zero 210 * pos. 211 */ 212 if (!SV_CURPROC_FLAG(SV_AOUT)) { 213 if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) || 214 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0))) 215 return (EINVAL); 216 } else { 217 if ((flags & MAP_ANON) != 0) 218 pos = 0; 219 } 220 221 if (flags & MAP_STACK) { 222 if ((fd != -1) || 223 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) 224 return (EINVAL); 225 flags |= MAP_ANON; 226 pos = 0; 227 } 228 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE | 229 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE | 230 MAP_PREFAULT_READ | MAP_GUARD | 231 #ifdef MAP_32BIT 232 MAP_32BIT | 233 #endif 234 MAP_ALIGNMENT_MASK)) != 0) 235 return (EINVAL); 236 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL) 237 return (EINVAL); 238 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE)) 239 return (EINVAL); 240 if (prot != PROT_NONE && 241 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0) 242 return (EINVAL); 243 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 || 244 pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL | 245 #ifdef MAP_32BIT 246 MAP_32BIT | 247 #endif 248 MAP_ALIGNMENT_MASK)) != 0)) 249 return (EINVAL); 250 251 /* 252 * Align the file position to a page boundary, 253 * and save its page offset component. 254 */ 255 pageoff = (pos & PAGE_MASK); 256 pos -= pageoff; 257 258 /* Adjust size for rounding (on both ends). */ 259 size += pageoff; /* low end... */ 260 size = (vm_size_t) round_page(size); /* hi end */ 261 262 /* Ensure alignment is at least a page and fits in a pointer. */ 263 align = flags & MAP_ALIGNMENT_MASK; 264 if (align != 0 && align != MAP_ALIGNED_SUPER && 265 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY || 266 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT)) 267 return (EINVAL); 268 269 /* 270 * Check for illegal addresses. Watch out for address wrap... Note 271 * that VM_*_ADDRESS are not constants due to casts (argh). 272 */ 273 if (flags & MAP_FIXED) { 274 /* 275 * The specified address must have the same remainder 276 * as the file offset taken modulo PAGE_SIZE, so it 277 * should be aligned after adjustment by pageoff. 278 */ 279 addr -= pageoff; 280 if (addr & PAGE_MASK) 281 return (EINVAL); 282 283 /* Address range must be all in user VM space. */ 284 if (addr < vm_map_min(&vms->vm_map) || 285 addr + size > vm_map_max(&vms->vm_map)) 286 return (EINVAL); 287 if (addr + size < addr) 288 return (EINVAL); 289 #ifdef MAP_32BIT 290 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR) 291 return (EINVAL); 292 } else if (flags & MAP_32BIT) { 293 /* 294 * For MAP_32BIT, override the hint if it is too high and 295 * do not bother moving the mapping past the heap (since 296 * the heap is usually above 2GB). 297 */ 298 if (addr + size > MAP_32BIT_MAX_ADDR) 299 addr = 0; 300 #endif 301 } else { 302 /* 303 * XXX for non-fixed mappings where no hint is provided or 304 * the hint would fall in the potential heap space, 305 * place it after the end of the largest possible heap. 306 * 307 * There should really be a pmap call to determine a reasonable 308 * location. 309 */ 310 if (addr == 0 || 311 (addr >= round_page((vm_offset_t)vms->vm_taddr) && 312 addr < round_page((vm_offset_t)vms->vm_daddr + 313 lim_max(td, RLIMIT_DATA)))) 314 addr = round_page((vm_offset_t)vms->vm_daddr + 315 lim_max(td, RLIMIT_DATA)); 316 } 317 if (size == 0) { 318 /* 319 * Return success without mapping anything for old 320 * binaries that request a page-aligned mapping of 321 * length 0. For modern binaries, this function 322 * returns an error earlier. 323 */ 324 error = 0; 325 } else if ((flags & MAP_GUARD) != 0) { 326 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE, 327 VM_PROT_NONE, flags, NULL, pos, FALSE, td); 328 } else if ((flags & MAP_ANON) != 0) { 329 /* 330 * Mapping blank space is trivial. 331 * 332 * This relies on VM_PROT_* matching PROT_*. 333 */ 334 error = vm_mmap_object(&vms->vm_map, &addr, size, prot, 335 VM_PROT_ALL, flags, NULL, pos, FALSE, td); 336 } else { 337 /* 338 * Mapping file, get fp for validation and don't let the 339 * descriptor disappear on us if we block. Check capability 340 * rights, but also return the maximum rights to be combined 341 * with maxprot later. 342 */ 343 cap_rights_init(&rights, CAP_MMAP); 344 if (prot & PROT_READ) 345 cap_rights_set(&rights, CAP_MMAP_R); 346 if ((flags & MAP_SHARED) != 0) { 347 if (prot & PROT_WRITE) 348 cap_rights_set(&rights, CAP_MMAP_W); 349 } 350 if (prot & PROT_EXEC) 351 cap_rights_set(&rights, CAP_MMAP_X); 352 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp); 353 if (error != 0) 354 goto done; 355 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 && 356 td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) { 357 error = EINVAL; 358 goto done; 359 } 360 361 /* This relies on VM_PROT_* matching PROT_*. */ 362 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot, 363 cap_maxprot, flags, pos, td); 364 } 365 366 if (error == 0) 367 td->td_retval[0] = (register_t) (addr + pageoff); 368 done: 369 if (fp) 370 fdrop(fp, td); 371 372 return (error); 373 } 374 375 #if defined(COMPAT_FREEBSD6) 376 int 377 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap) 378 { 379 380 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot, 381 uap->flags, uap->fd, uap->pos)); 382 } 383 #endif 384 385 #ifdef COMPAT_43 386 #ifndef _SYS_SYSPROTO_H_ 387 struct ommap_args { 388 caddr_t addr; 389 int len; 390 int prot; 391 int flags; 392 int fd; 393 long pos; 394 }; 395 #endif 396 int 397 ommap(struct thread *td, struct ommap_args *uap) 398 { 399 static const char cvtbsdprot[8] = { 400 0, 401 PROT_EXEC, 402 PROT_WRITE, 403 PROT_EXEC | PROT_WRITE, 404 PROT_READ, 405 PROT_EXEC | PROT_READ, 406 PROT_WRITE | PROT_READ, 407 PROT_EXEC | PROT_WRITE | PROT_READ, 408 }; 409 int flags, prot; 410 411 #define OMAP_ANON 0x0002 412 #define OMAP_COPY 0x0020 413 #define OMAP_SHARED 0x0010 414 #define OMAP_FIXED 0x0100 415 416 prot = cvtbsdprot[uap->prot & 0x7]; 417 #ifdef COMPAT_FREEBSD32 418 #if defined(__amd64__) 419 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) && 420 prot != 0) 421 prot |= PROT_EXEC; 422 #endif 423 #endif 424 flags = 0; 425 if (uap->flags & OMAP_ANON) 426 flags |= MAP_ANON; 427 if (uap->flags & OMAP_COPY) 428 flags |= MAP_COPY; 429 if (uap->flags & OMAP_SHARED) 430 flags |= MAP_SHARED; 431 else 432 flags |= MAP_PRIVATE; 433 if (uap->flags & OMAP_FIXED) 434 flags |= MAP_FIXED; 435 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags, 436 uap->fd, uap->pos)); 437 } 438 #endif /* COMPAT_43 */ 439 440 441 #ifndef _SYS_SYSPROTO_H_ 442 struct msync_args { 443 void *addr; 444 size_t len; 445 int flags; 446 }; 447 #endif 448 int 449 sys_msync(struct thread *td, struct msync_args *uap) 450 { 451 452 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags)); 453 } 454 455 int 456 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags) 457 { 458 vm_offset_t addr; 459 vm_size_t pageoff; 460 vm_map_t map; 461 int rv; 462 463 addr = addr0; 464 pageoff = (addr & PAGE_MASK); 465 addr -= pageoff; 466 size += pageoff; 467 size = (vm_size_t) round_page(size); 468 if (addr + size < addr) 469 return (EINVAL); 470 471 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) 472 return (EINVAL); 473 474 map = &td->td_proc->p_vmspace->vm_map; 475 476 /* 477 * Clean the pages and interpret the return value. 478 */ 479 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0, 480 (flags & MS_INVALIDATE) != 0); 481 switch (rv) { 482 case KERN_SUCCESS: 483 return (0); 484 case KERN_INVALID_ADDRESS: 485 return (ENOMEM); 486 case KERN_INVALID_ARGUMENT: 487 return (EBUSY); 488 case KERN_FAILURE: 489 return (EIO); 490 default: 491 return (EINVAL); 492 } 493 } 494 495 #ifndef _SYS_SYSPROTO_H_ 496 struct munmap_args { 497 void *addr; 498 size_t len; 499 }; 500 #endif 501 int 502 sys_munmap(struct thread *td, struct munmap_args *uap) 503 { 504 505 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len)); 506 } 507 508 int 509 kern_munmap(struct thread *td, uintptr_t addr0, size_t size) 510 { 511 #ifdef HWPMC_HOOKS 512 struct pmckern_map_out pkm; 513 vm_map_entry_t entry; 514 bool pmc_handled; 515 #endif 516 vm_offset_t addr; 517 vm_size_t pageoff; 518 vm_map_t map; 519 520 if (size == 0) 521 return (EINVAL); 522 523 addr = addr0; 524 pageoff = (addr & PAGE_MASK); 525 addr -= pageoff; 526 size += pageoff; 527 size = (vm_size_t) round_page(size); 528 if (addr + size < addr) 529 return (EINVAL); 530 531 /* 532 * Check for illegal addresses. Watch out for address wrap... 533 */ 534 map = &td->td_proc->p_vmspace->vm_map; 535 if (addr < vm_map_min(map) || addr + size > vm_map_max(map)) 536 return (EINVAL); 537 vm_map_lock(map); 538 #ifdef HWPMC_HOOKS 539 pmc_handled = false; 540 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) { 541 pmc_handled = true; 542 /* 543 * Inform hwpmc if the address range being unmapped contains 544 * an executable region. 545 */ 546 pkm.pm_address = (uintptr_t) NULL; 547 if (vm_map_lookup_entry(map, addr, &entry)) { 548 for (; entry->start < addr + size; 549 entry = entry->next) { 550 if (vm_map_check_protection(map, entry->start, 551 entry->end, VM_PROT_EXECUTE) == TRUE) { 552 pkm.pm_address = (uintptr_t) addr; 553 pkm.pm_size = (size_t) size; 554 break; 555 } 556 } 557 } 558 } 559 #endif 560 vm_map_delete(map, addr, addr + size); 561 562 #ifdef HWPMC_HOOKS 563 if (__predict_false(pmc_handled)) { 564 /* downgrade the lock to prevent a LOR with the pmc-sx lock */ 565 vm_map_lock_downgrade(map); 566 if (pkm.pm_address != (uintptr_t) NULL) 567 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm); 568 vm_map_unlock_read(map); 569 } else 570 #endif 571 vm_map_unlock(map); 572 573 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */ 574 return (0); 575 } 576 577 #ifndef _SYS_SYSPROTO_H_ 578 struct mprotect_args { 579 const void *addr; 580 size_t len; 581 int prot; 582 }; 583 #endif 584 int 585 sys_mprotect(struct thread *td, struct mprotect_args *uap) 586 { 587 588 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot)); 589 } 590 591 int 592 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot) 593 { 594 vm_offset_t addr; 595 vm_size_t pageoff; 596 597 addr = addr0; 598 prot = (prot & VM_PROT_ALL); 599 pageoff = (addr & PAGE_MASK); 600 addr -= pageoff; 601 size += pageoff; 602 size = (vm_size_t) round_page(size); 603 #ifdef COMPAT_FREEBSD32 604 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 605 if (((addr + size) & 0xffffffff) < addr) 606 return (EINVAL); 607 } else 608 #endif 609 if (addr + size < addr) 610 return (EINVAL); 611 612 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr, 613 addr + size, prot, FALSE)) { 614 case KERN_SUCCESS: 615 return (0); 616 case KERN_PROTECTION_FAILURE: 617 return (EACCES); 618 case KERN_RESOURCE_SHORTAGE: 619 return (ENOMEM); 620 } 621 return (EINVAL); 622 } 623 624 #ifndef _SYS_SYSPROTO_H_ 625 struct minherit_args { 626 void *addr; 627 size_t len; 628 int inherit; 629 }; 630 #endif 631 int 632 sys_minherit(struct thread *td, struct minherit_args *uap) 633 { 634 vm_offset_t addr; 635 vm_size_t size, pageoff; 636 vm_inherit_t inherit; 637 638 addr = (vm_offset_t)uap->addr; 639 size = uap->len; 640 inherit = uap->inherit; 641 642 pageoff = (addr & PAGE_MASK); 643 addr -= pageoff; 644 size += pageoff; 645 size = (vm_size_t) round_page(size); 646 if (addr + size < addr) 647 return (EINVAL); 648 649 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr, 650 addr + size, inherit)) { 651 case KERN_SUCCESS: 652 return (0); 653 case KERN_PROTECTION_FAILURE: 654 return (EACCES); 655 } 656 return (EINVAL); 657 } 658 659 #ifndef _SYS_SYSPROTO_H_ 660 struct madvise_args { 661 void *addr; 662 size_t len; 663 int behav; 664 }; 665 #endif 666 667 int 668 sys_madvise(struct thread *td, struct madvise_args *uap) 669 { 670 671 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav)); 672 } 673 674 int 675 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav) 676 { 677 vm_map_t map; 678 vm_offset_t addr, end, start; 679 int flags; 680 681 /* 682 * Check for our special case, advising the swap pager we are 683 * "immortal." 684 */ 685 if (behav == MADV_PROTECT) { 686 flags = PPROT_SET; 687 return (kern_procctl(td, P_PID, td->td_proc->p_pid, 688 PROC_SPROTECT, &flags)); 689 } 690 691 /* 692 * Check for illegal addresses. Watch out for address wrap... Note 693 * that VM_*_ADDRESS are not constants due to casts (argh). 694 */ 695 map = &td->td_proc->p_vmspace->vm_map; 696 addr = addr0; 697 if (addr < vm_map_min(map) || addr + len > vm_map_max(map)) 698 return (EINVAL); 699 if ((addr + len) < addr) 700 return (EINVAL); 701 702 /* 703 * Since this routine is only advisory, we default to conservative 704 * behavior. 705 */ 706 start = trunc_page(addr); 707 end = round_page(addr + len); 708 709 /* 710 * vm_map_madvise() checks for illegal values of behav. 711 */ 712 return (vm_map_madvise(map, start, end, behav)); 713 } 714 715 #ifndef _SYS_SYSPROTO_H_ 716 struct mincore_args { 717 const void *addr; 718 size_t len; 719 char *vec; 720 }; 721 #endif 722 723 int 724 sys_mincore(struct thread *td, struct mincore_args *uap) 725 { 726 727 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec)); 728 } 729 730 int 731 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec) 732 { 733 vm_offset_t addr, first_addr; 734 vm_offset_t end, cend; 735 pmap_t pmap; 736 vm_map_t map; 737 int error = 0; 738 int vecindex, lastvecindex; 739 vm_map_entry_t current; 740 vm_map_entry_t entry; 741 vm_object_t object; 742 vm_paddr_t locked_pa; 743 vm_page_t m; 744 vm_pindex_t pindex; 745 int mincoreinfo; 746 unsigned int timestamp; 747 boolean_t locked; 748 749 /* 750 * Make sure that the addresses presented are valid for user 751 * mode. 752 */ 753 first_addr = addr = trunc_page(addr0); 754 end = addr + (vm_size_t)round_page(len); 755 map = &td->td_proc->p_vmspace->vm_map; 756 if (end > vm_map_max(map) || end < addr) 757 return (ENOMEM); 758 759 pmap = vmspace_pmap(td->td_proc->p_vmspace); 760 761 vm_map_lock_read(map); 762 RestartScan: 763 timestamp = map->timestamp; 764 765 if (!vm_map_lookup_entry(map, addr, &entry)) { 766 vm_map_unlock_read(map); 767 return (ENOMEM); 768 } 769 770 /* 771 * Do this on a map entry basis so that if the pages are not 772 * in the current processes address space, we can easily look 773 * up the pages elsewhere. 774 */ 775 lastvecindex = -1; 776 for (current = entry; current->start < end; current = current->next) { 777 778 /* 779 * check for contiguity 780 */ 781 if (current->end < end && current->next->start > current->end) { 782 vm_map_unlock_read(map); 783 return (ENOMEM); 784 } 785 786 /* 787 * ignore submaps (for now) or null objects 788 */ 789 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) || 790 current->object.vm_object == NULL) 791 continue; 792 793 /* 794 * limit this scan to the current map entry and the 795 * limits for the mincore call 796 */ 797 if (addr < current->start) 798 addr = current->start; 799 cend = current->end; 800 if (cend > end) 801 cend = end; 802 803 /* 804 * scan this entry one page at a time 805 */ 806 while (addr < cend) { 807 /* 808 * Check pmap first, it is likely faster, also 809 * it can provide info as to whether we are the 810 * one referencing or modifying the page. 811 */ 812 object = NULL; 813 locked_pa = 0; 814 retry: 815 m = NULL; 816 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa); 817 if (locked_pa != 0) { 818 /* 819 * The page is mapped by this process but not 820 * both accessed and modified. It is also 821 * managed. Acquire the object lock so that 822 * other mappings might be examined. 823 */ 824 m = PHYS_TO_VM_PAGE(locked_pa); 825 if (m->object != object) { 826 if (object != NULL) 827 VM_OBJECT_WUNLOCK(object); 828 object = m->object; 829 locked = VM_OBJECT_TRYWLOCK(object); 830 vm_page_unlock(m); 831 if (!locked) { 832 VM_OBJECT_WLOCK(object); 833 vm_page_lock(m); 834 goto retry; 835 } 836 } else 837 vm_page_unlock(m); 838 KASSERT(m->valid == VM_PAGE_BITS_ALL, 839 ("mincore: page %p is mapped but invalid", 840 m)); 841 } else if (mincoreinfo == 0) { 842 /* 843 * The page is not mapped by this process. If 844 * the object implements managed pages, then 845 * determine if the page is resident so that 846 * the mappings might be examined. 847 */ 848 if (current->object.vm_object != object) { 849 if (object != NULL) 850 VM_OBJECT_WUNLOCK(object); 851 object = current->object.vm_object; 852 VM_OBJECT_WLOCK(object); 853 } 854 if (object->type == OBJT_DEFAULT || 855 object->type == OBJT_SWAP || 856 object->type == OBJT_VNODE) { 857 pindex = OFF_TO_IDX(current->offset + 858 (addr - current->start)); 859 m = vm_page_lookup(object, pindex); 860 if (m != NULL && m->valid == 0) 861 m = NULL; 862 if (m != NULL) 863 mincoreinfo = MINCORE_INCORE; 864 } 865 } 866 if (m != NULL) { 867 /* Examine other mappings to the page. */ 868 if (m->dirty == 0 && pmap_is_modified(m)) 869 vm_page_dirty(m); 870 if (m->dirty != 0) 871 mincoreinfo |= MINCORE_MODIFIED_OTHER; 872 /* 873 * The first test for PGA_REFERENCED is an 874 * optimization. The second test is 875 * required because a concurrent pmap 876 * operation could clear the last reference 877 * and set PGA_REFERENCED before the call to 878 * pmap_is_referenced(). 879 */ 880 if ((m->aflags & PGA_REFERENCED) != 0 || 881 pmap_is_referenced(m) || 882 (m->aflags & PGA_REFERENCED) != 0) 883 mincoreinfo |= MINCORE_REFERENCED_OTHER; 884 } 885 if (object != NULL) 886 VM_OBJECT_WUNLOCK(object); 887 888 /* 889 * subyte may page fault. In case it needs to modify 890 * the map, we release the lock. 891 */ 892 vm_map_unlock_read(map); 893 894 /* 895 * calculate index into user supplied byte vector 896 */ 897 vecindex = atop(addr - first_addr); 898 899 /* 900 * If we have skipped map entries, we need to make sure that 901 * the byte vector is zeroed for those skipped entries. 902 */ 903 while ((lastvecindex + 1) < vecindex) { 904 ++lastvecindex; 905 error = subyte(vec + lastvecindex, 0); 906 if (error) { 907 error = EFAULT; 908 goto done2; 909 } 910 } 911 912 /* 913 * Pass the page information to the user 914 */ 915 error = subyte(vec + vecindex, mincoreinfo); 916 if (error) { 917 error = EFAULT; 918 goto done2; 919 } 920 921 /* 922 * If the map has changed, due to the subyte, the previous 923 * output may be invalid. 924 */ 925 vm_map_lock_read(map); 926 if (timestamp != map->timestamp) 927 goto RestartScan; 928 929 lastvecindex = vecindex; 930 addr += PAGE_SIZE; 931 } 932 } 933 934 /* 935 * subyte may page fault. In case it needs to modify 936 * the map, we release the lock. 937 */ 938 vm_map_unlock_read(map); 939 940 /* 941 * Zero the last entries in the byte vector. 942 */ 943 vecindex = atop(end - first_addr); 944 while ((lastvecindex + 1) < vecindex) { 945 ++lastvecindex; 946 error = subyte(vec + lastvecindex, 0); 947 if (error) { 948 error = EFAULT; 949 goto done2; 950 } 951 } 952 953 /* 954 * If the map has changed, due to the subyte, the previous 955 * output may be invalid. 956 */ 957 vm_map_lock_read(map); 958 if (timestamp != map->timestamp) 959 goto RestartScan; 960 vm_map_unlock_read(map); 961 done2: 962 return (error); 963 } 964 965 #ifndef _SYS_SYSPROTO_H_ 966 struct mlock_args { 967 const void *addr; 968 size_t len; 969 }; 970 #endif 971 int 972 sys_mlock(struct thread *td, struct mlock_args *uap) 973 { 974 975 return (kern_mlock(td->td_proc, td->td_ucred, 976 __DECONST(uintptr_t, uap->addr), uap->len)); 977 } 978 979 int 980 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len) 981 { 982 vm_offset_t addr, end, last, start; 983 vm_size_t npages, size; 984 vm_map_t map; 985 unsigned long nsize; 986 int error; 987 988 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0); 989 if (error) 990 return (error); 991 addr = addr0; 992 size = len; 993 last = addr + size; 994 start = trunc_page(addr); 995 end = round_page(last); 996 if (last < addr || end < addr) 997 return (EINVAL); 998 npages = atop(end - start); 999 if (npages > vm_page_max_wired) 1000 return (ENOMEM); 1001 map = &proc->p_vmspace->vm_map; 1002 PROC_LOCK(proc); 1003 nsize = ptoa(npages + pmap_wired_count(map->pmap)); 1004 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) { 1005 PROC_UNLOCK(proc); 1006 return (ENOMEM); 1007 } 1008 PROC_UNLOCK(proc); 1009 if (npages + vm_wire_count() > vm_page_max_wired) 1010 return (EAGAIN); 1011 #ifdef RACCT 1012 if (racct_enable) { 1013 PROC_LOCK(proc); 1014 error = racct_set(proc, RACCT_MEMLOCK, nsize); 1015 PROC_UNLOCK(proc); 1016 if (error != 0) 1017 return (ENOMEM); 1018 } 1019 #endif 1020 error = vm_map_wire(map, start, end, 1021 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 1022 #ifdef RACCT 1023 if (racct_enable && error != KERN_SUCCESS) { 1024 PROC_LOCK(proc); 1025 racct_set(proc, RACCT_MEMLOCK, 1026 ptoa(pmap_wired_count(map->pmap))); 1027 PROC_UNLOCK(proc); 1028 } 1029 #endif 1030 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1031 } 1032 1033 #ifndef _SYS_SYSPROTO_H_ 1034 struct mlockall_args { 1035 int how; 1036 }; 1037 #endif 1038 1039 int 1040 sys_mlockall(struct thread *td, struct mlockall_args *uap) 1041 { 1042 vm_map_t map; 1043 int error; 1044 1045 map = &td->td_proc->p_vmspace->vm_map; 1046 error = priv_check(td, PRIV_VM_MLOCK); 1047 if (error) 1048 return (error); 1049 1050 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0)) 1051 return (EINVAL); 1052 1053 /* 1054 * If wiring all pages in the process would cause it to exceed 1055 * a hard resource limit, return ENOMEM. 1056 */ 1057 if (!old_mlock && uap->how & MCL_CURRENT) { 1058 if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) 1059 return (ENOMEM); 1060 } 1061 #ifdef RACCT 1062 if (racct_enable) { 1063 PROC_LOCK(td->td_proc); 1064 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size); 1065 PROC_UNLOCK(td->td_proc); 1066 if (error != 0) 1067 return (ENOMEM); 1068 } 1069 #endif 1070 1071 if (uap->how & MCL_FUTURE) { 1072 vm_map_lock(map); 1073 vm_map_modflags(map, MAP_WIREFUTURE, 0); 1074 vm_map_unlock(map); 1075 error = 0; 1076 } 1077 1078 if (uap->how & MCL_CURRENT) { 1079 /* 1080 * P1003.1-2001 mandates that all currently mapped pages 1081 * will be memory resident and locked (wired) upon return 1082 * from mlockall(). vm_map_wire() will wire pages, by 1083 * calling vm_fault_wire() for each page in the region. 1084 */ 1085 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map), 1086 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); 1087 error = (error == KERN_SUCCESS ? 0 : EAGAIN); 1088 } 1089 #ifdef RACCT 1090 if (racct_enable && error != KERN_SUCCESS) { 1091 PROC_LOCK(td->td_proc); 1092 racct_set(td->td_proc, RACCT_MEMLOCK, 1093 ptoa(pmap_wired_count(map->pmap))); 1094 PROC_UNLOCK(td->td_proc); 1095 } 1096 #endif 1097 1098 return (error); 1099 } 1100 1101 #ifndef _SYS_SYSPROTO_H_ 1102 struct munlockall_args { 1103 register_t dummy; 1104 }; 1105 #endif 1106 1107 int 1108 sys_munlockall(struct thread *td, struct munlockall_args *uap) 1109 { 1110 vm_map_t map; 1111 int error; 1112 1113 map = &td->td_proc->p_vmspace->vm_map; 1114 error = priv_check(td, PRIV_VM_MUNLOCK); 1115 if (error) 1116 return (error); 1117 1118 /* Clear the MAP_WIREFUTURE flag from this vm_map. */ 1119 vm_map_lock(map); 1120 vm_map_modflags(map, 0, MAP_WIREFUTURE); 1121 vm_map_unlock(map); 1122 1123 /* Forcibly unwire all pages. */ 1124 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map), 1125 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); 1126 #ifdef RACCT 1127 if (racct_enable && error == KERN_SUCCESS) { 1128 PROC_LOCK(td->td_proc); 1129 racct_set(td->td_proc, RACCT_MEMLOCK, 0); 1130 PROC_UNLOCK(td->td_proc); 1131 } 1132 #endif 1133 1134 return (error); 1135 } 1136 1137 #ifndef _SYS_SYSPROTO_H_ 1138 struct munlock_args { 1139 const void *addr; 1140 size_t len; 1141 }; 1142 #endif 1143 int 1144 sys_munlock(struct thread *td, struct munlock_args *uap) 1145 { 1146 1147 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len)); 1148 } 1149 1150 int 1151 kern_munlock(struct thread *td, uintptr_t addr0, size_t size) 1152 { 1153 vm_offset_t addr, end, last, start; 1154 #ifdef RACCT 1155 vm_map_t map; 1156 #endif 1157 int error; 1158 1159 error = priv_check(td, PRIV_VM_MUNLOCK); 1160 if (error) 1161 return (error); 1162 addr = addr0; 1163 last = addr + size; 1164 start = trunc_page(addr); 1165 end = round_page(last); 1166 if (last < addr || end < addr) 1167 return (EINVAL); 1168 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end, 1169 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 1170 #ifdef RACCT 1171 if (racct_enable && error == KERN_SUCCESS) { 1172 PROC_LOCK(td->td_proc); 1173 map = &td->td_proc->p_vmspace->vm_map; 1174 racct_set(td->td_proc, RACCT_MEMLOCK, 1175 ptoa(pmap_wired_count(map->pmap))); 1176 PROC_UNLOCK(td->td_proc); 1177 } 1178 #endif 1179 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1180 } 1181 1182 /* 1183 * vm_mmap_vnode() 1184 * 1185 * Helper function for vm_mmap. Perform sanity check specific for mmap 1186 * operations on vnodes. 1187 */ 1188 int 1189 vm_mmap_vnode(struct thread *td, vm_size_t objsize, 1190 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp, 1191 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp, 1192 boolean_t *writecounted) 1193 { 1194 struct vattr va; 1195 vm_object_t obj; 1196 vm_ooffset_t foff; 1197 struct ucred *cred; 1198 int error, flags, locktype; 1199 1200 cred = td->td_ucred; 1201 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED)) 1202 locktype = LK_EXCLUSIVE; 1203 else 1204 locktype = LK_SHARED; 1205 if ((error = vget(vp, locktype, td)) != 0) 1206 return (error); 1207 AUDIT_ARG_VNODE1(vp); 1208 foff = *foffp; 1209 flags = *flagsp; 1210 obj = vp->v_object; 1211 if (vp->v_type == VREG) { 1212 /* 1213 * Get the proper underlying object 1214 */ 1215 if (obj == NULL) { 1216 error = EINVAL; 1217 goto done; 1218 } 1219 if (obj->type == OBJT_VNODE && obj->handle != vp) { 1220 vput(vp); 1221 vp = (struct vnode *)obj->handle; 1222 /* 1223 * Bypass filesystems obey the mpsafety of the 1224 * underlying fs. Tmpfs never bypasses. 1225 */ 1226 error = vget(vp, locktype, td); 1227 if (error != 0) 1228 return (error); 1229 } 1230 if (locktype == LK_EXCLUSIVE) { 1231 *writecounted = TRUE; 1232 vnode_pager_update_writecount(obj, 0, objsize); 1233 } 1234 } else { 1235 error = EINVAL; 1236 goto done; 1237 } 1238 if ((error = VOP_GETATTR(vp, &va, cred))) 1239 goto done; 1240 #ifdef MAC 1241 /* This relies on VM_PROT_* matching PROT_*. */ 1242 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags); 1243 if (error != 0) 1244 goto done; 1245 #endif 1246 if ((flags & MAP_SHARED) != 0) { 1247 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) { 1248 if (prot & VM_PROT_WRITE) { 1249 error = EPERM; 1250 goto done; 1251 } 1252 *maxprotp &= ~VM_PROT_WRITE; 1253 } 1254 } 1255 /* 1256 * If it is a regular file without any references 1257 * we do not need to sync it. 1258 * Adjust object size to be the size of actual file. 1259 */ 1260 objsize = round_page(va.va_size); 1261 if (va.va_nlink == 0) 1262 flags |= MAP_NOSYNC; 1263 if (obj->type == OBJT_VNODE) { 1264 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, 1265 cred); 1266 if (obj == NULL) { 1267 error = ENOMEM; 1268 goto done; 1269 } 1270 } else { 1271 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP, 1272 ("wrong object type")); 1273 VM_OBJECT_WLOCK(obj); 1274 vm_object_reference_locked(obj); 1275 #if VM_NRESERVLEVEL > 0 1276 vm_object_color(obj, 0); 1277 #endif 1278 VM_OBJECT_WUNLOCK(obj); 1279 } 1280 *objp = obj; 1281 *flagsp = flags; 1282 1283 vfs_mark_atime(vp, cred); 1284 1285 done: 1286 if (error != 0 && *writecounted) { 1287 *writecounted = FALSE; 1288 vnode_pager_update_writecount(obj, objsize, 0); 1289 } 1290 vput(vp); 1291 return (error); 1292 } 1293 1294 /* 1295 * vm_mmap_cdev() 1296 * 1297 * Helper function for vm_mmap. Perform sanity check specific for mmap 1298 * operations on cdevs. 1299 */ 1300 int 1301 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot, 1302 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw, 1303 vm_ooffset_t *foff, vm_object_t *objp) 1304 { 1305 vm_object_t obj; 1306 int error, flags; 1307 1308 flags = *flagsp; 1309 1310 if (dsw->d_flags & D_MMAP_ANON) { 1311 *objp = NULL; 1312 *foff = 0; 1313 *maxprotp = VM_PROT_ALL; 1314 *flagsp |= MAP_ANON; 1315 return (0); 1316 } 1317 /* 1318 * cdevs do not provide private mappings of any kind. 1319 */ 1320 if ((*maxprotp & VM_PROT_WRITE) == 0 && 1321 (prot & VM_PROT_WRITE) != 0) 1322 return (EACCES); 1323 if (flags & (MAP_PRIVATE|MAP_COPY)) 1324 return (EINVAL); 1325 /* 1326 * Force device mappings to be shared. 1327 */ 1328 flags |= MAP_SHARED; 1329 #ifdef MAC_XXX 1330 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot); 1331 if (error != 0) 1332 return (error); 1333 #endif 1334 /* 1335 * First, try d_mmap_single(). If that is not implemented 1336 * (returns ENODEV), fall back to using the device pager. 1337 * Note that d_mmap_single() must return a reference to the 1338 * object (it needs to bump the reference count of the object 1339 * it returns somehow). 1340 * 1341 * XXX assumes VM_PROT_* == PROT_* 1342 */ 1343 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot); 1344 if (error != ENODEV) 1345 return (error); 1346 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff, 1347 td->td_ucred); 1348 if (obj == NULL) 1349 return (EINVAL); 1350 *objp = obj; 1351 *flagsp = flags; 1352 return (0); 1353 } 1354 1355 /* 1356 * vm_mmap() 1357 * 1358 * Internal version of mmap used by exec, sys5 shared memory, and 1359 * various device drivers. Handle is either a vnode pointer, a 1360 * character device, or NULL for MAP_ANON. 1361 */ 1362 int 1363 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1364 vm_prot_t maxprot, int flags, 1365 objtype_t handle_type, void *handle, 1366 vm_ooffset_t foff) 1367 { 1368 vm_object_t object; 1369 struct thread *td = curthread; 1370 int error; 1371 boolean_t writecounted; 1372 1373 if (size == 0) 1374 return (EINVAL); 1375 1376 size = round_page(size); 1377 object = NULL; 1378 writecounted = FALSE; 1379 1380 /* 1381 * Lookup/allocate object. 1382 */ 1383 switch (handle_type) { 1384 case OBJT_DEVICE: { 1385 struct cdevsw *dsw; 1386 struct cdev *cdev; 1387 int ref; 1388 1389 cdev = handle; 1390 dsw = dev_refthread(cdev, &ref); 1391 if (dsw == NULL) 1392 return (ENXIO); 1393 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev, 1394 dsw, &foff, &object); 1395 dev_relthread(cdev, ref); 1396 break; 1397 } 1398 case OBJT_VNODE: 1399 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, 1400 handle, &foff, &object, &writecounted); 1401 break; 1402 case OBJT_DEFAULT: 1403 if (handle == NULL) { 1404 error = 0; 1405 break; 1406 } 1407 /* FALLTHROUGH */ 1408 default: 1409 error = EINVAL; 1410 break; 1411 } 1412 if (error) 1413 return (error); 1414 1415 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1416 foff, writecounted, td); 1417 if (error != 0 && object != NULL) { 1418 /* 1419 * If this mapping was accounted for in the vnode's 1420 * writecount, then undo that now. 1421 */ 1422 if (writecounted) 1423 vnode_pager_release_writecount(object, 0, size); 1424 vm_object_deallocate(object); 1425 } 1426 return (error); 1427 } 1428 1429 /* 1430 * Internal version of mmap that maps a specific VM object into an 1431 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap. 1432 */ 1433 int 1434 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1435 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff, 1436 boolean_t writecounted, struct thread *td) 1437 { 1438 boolean_t curmap, fitit; 1439 vm_offset_t max_addr; 1440 int docow, error, findspace, rv; 1441 1442 curmap = map == &td->td_proc->p_vmspace->vm_map; 1443 if (curmap) { 1444 RACCT_PROC_LOCK(td->td_proc); 1445 if (map->size + size > lim_cur(td, RLIMIT_VMEM)) { 1446 RACCT_PROC_UNLOCK(td->td_proc); 1447 return (ENOMEM); 1448 } 1449 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) { 1450 RACCT_PROC_UNLOCK(td->td_proc); 1451 return (ENOMEM); 1452 } 1453 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 1454 if (ptoa(pmap_wired_count(map->pmap)) + size > 1455 lim_cur(td, RLIMIT_MEMLOCK)) { 1456 racct_set_force(td->td_proc, RACCT_VMEM, 1457 map->size); 1458 RACCT_PROC_UNLOCK(td->td_proc); 1459 return (ENOMEM); 1460 } 1461 error = racct_set(td->td_proc, RACCT_MEMLOCK, 1462 ptoa(pmap_wired_count(map->pmap)) + size); 1463 if (error != 0) { 1464 racct_set_force(td->td_proc, RACCT_VMEM, 1465 map->size); 1466 RACCT_PROC_UNLOCK(td->td_proc); 1467 return (error); 1468 } 1469 } 1470 RACCT_PROC_UNLOCK(td->td_proc); 1471 } 1472 1473 /* 1474 * We currently can only deal with page aligned file offsets. 1475 * The mmap() system call already enforces this by subtracting 1476 * the page offset from the file offset, but checking here 1477 * catches errors in device drivers (e.g. d_single_mmap() 1478 * callbacks) and other internal mapping requests (such as in 1479 * exec). 1480 */ 1481 if (foff & PAGE_MASK) 1482 return (EINVAL); 1483 1484 if ((flags & MAP_FIXED) == 0) { 1485 fitit = TRUE; 1486 *addr = round_page(*addr); 1487 } else { 1488 if (*addr != trunc_page(*addr)) 1489 return (EINVAL); 1490 fitit = FALSE; 1491 } 1492 1493 if (flags & MAP_ANON) { 1494 if (object != NULL || foff != 0) 1495 return (EINVAL); 1496 docow = 0; 1497 } else if (flags & MAP_PREFAULT_READ) 1498 docow = MAP_PREFAULT; 1499 else 1500 docow = MAP_PREFAULT_PARTIAL; 1501 1502 if ((flags & (MAP_ANON|MAP_SHARED)) == 0) 1503 docow |= MAP_COPY_ON_WRITE; 1504 if (flags & MAP_NOSYNC) 1505 docow |= MAP_DISABLE_SYNCER; 1506 if (flags & MAP_NOCORE) 1507 docow |= MAP_DISABLE_COREDUMP; 1508 /* Shared memory is also shared with children. */ 1509 if (flags & MAP_SHARED) 1510 docow |= MAP_INHERIT_SHARE; 1511 if (writecounted) 1512 docow |= MAP_VN_WRITECOUNT; 1513 if (flags & MAP_STACK) { 1514 if (object != NULL) 1515 return (EINVAL); 1516 docow |= MAP_STACK_GROWS_DOWN; 1517 } 1518 if ((flags & MAP_EXCL) != 0) 1519 docow |= MAP_CHECK_EXCL; 1520 if ((flags & MAP_GUARD) != 0) 1521 docow |= MAP_CREATE_GUARD; 1522 1523 if (fitit) { 1524 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER) 1525 findspace = VMFS_SUPER_SPACE; 1526 else if ((flags & MAP_ALIGNMENT_MASK) != 0) 1527 findspace = VMFS_ALIGNED_SPACE(flags >> 1528 MAP_ALIGNMENT_SHIFT); 1529 else 1530 findspace = VMFS_OPTIMAL_SPACE; 1531 max_addr = 0; 1532 #ifdef MAP_32BIT 1533 if ((flags & MAP_32BIT) != 0) 1534 max_addr = MAP_32BIT_MAX_ADDR; 1535 #endif 1536 if (curmap) { 1537 rv = vm_map_find_min(map, object, foff, addr, size, 1538 round_page((vm_offset_t)td->td_proc->p_vmspace-> 1539 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr, 1540 findspace, prot, maxprot, docow); 1541 } else { 1542 rv = vm_map_find(map, object, foff, addr, size, 1543 max_addr, findspace, prot, maxprot, docow); 1544 } 1545 } else { 1546 rv = vm_map_fixed(map, object, foff, *addr, size, 1547 prot, maxprot, docow); 1548 } 1549 1550 if (rv == KERN_SUCCESS) { 1551 /* 1552 * If the process has requested that all future mappings 1553 * be wired, then heed this. 1554 */ 1555 if (map->flags & MAP_WIREFUTURE) { 1556 vm_map_wire(map, *addr, *addr + size, 1557 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ? 1558 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES)); 1559 } 1560 } 1561 return (vm_mmap_to_errno(rv)); 1562 } 1563 1564 /* 1565 * Translate a Mach VM return code to zero on success or the appropriate errno 1566 * on failure. 1567 */ 1568 int 1569 vm_mmap_to_errno(int rv) 1570 { 1571 1572 switch (rv) { 1573 case KERN_SUCCESS: 1574 return (0); 1575 case KERN_INVALID_ADDRESS: 1576 case KERN_NO_SPACE: 1577 return (ENOMEM); 1578 case KERN_PROTECTION_FAILURE: 1579 return (EACCES); 1580 default: 1581 return (EINVAL); 1582 } 1583 } 1584