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