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 #ifdef MAP_32BIT 112 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31) 113 #endif 114 115 _Static_assert(MAXPAGESIZES <= 4, "MINCORE_SUPER too narrow"); 116 117 #ifndef _SYS_SYSPROTO_H_ 118 struct sbrk_args { 119 int incr; 120 }; 121 #endif 122 123 int 124 sys_sbrk(struct thread *td, struct sbrk_args *uap) 125 { 126 /* Not yet implemented */ 127 return (EOPNOTSUPP); 128 } 129 130 #ifndef _SYS_SYSPROTO_H_ 131 struct sstk_args { 132 int incr; 133 }; 134 #endif 135 136 int 137 sys_sstk(struct thread *td, struct sstk_args *uap) 138 { 139 /* Not yet implemented */ 140 return (EOPNOTSUPP); 141 } 142 143 #if defined(COMPAT_43) 144 int 145 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap) 146 { 147 148 td->td_retval[0] = PAGE_SIZE; 149 return (0); 150 } 151 #endif /* COMPAT_43 */ 152 153 /* 154 * Memory Map (mmap) system call. Note that the file offset 155 * and address are allowed to be NOT page aligned, though if 156 * the MAP_FIXED flag it set, both must have the same remainder 157 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not 158 * page-aligned, the actual mapping starts at trunc_page(addr) 159 * and the return value is adjusted up by the page offset. 160 * 161 * Generally speaking, only character devices which are themselves 162 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise 163 * there would be no cache coherency between a descriptor and a VM mapping 164 * both to the same character device. 165 */ 166 #ifndef _SYS_SYSPROTO_H_ 167 struct mmap_args { 168 void *addr; 169 size_t len; 170 int prot; 171 int flags; 172 int fd; 173 long pad; 174 off_t pos; 175 }; 176 #endif 177 178 int 179 sys_mmap(struct thread *td, struct mmap_args *uap) 180 { 181 182 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot, 183 uap->flags, uap->fd, uap->pos)); 184 } 185 186 int 187 kern_mmap_maxprot(struct proc *p, int prot) 188 { 189 190 if ((p->p_flag2 & P2_PROTMAX_DISABLE) != 0 || 191 (p->p_fctl0 & NT_FREEBSD_FCTL_PROTMAX_DISABLE) != 0) 192 return (_PROT_ALL); 193 if (((p->p_flag2 & P2_PROTMAX_ENABLE) != 0 || imply_prot_max) && 194 prot != PROT_NONE) 195 return (prot); 196 return (_PROT_ALL); 197 } 198 199 int 200 kern_mmap(struct thread *td, uintptr_t addr0, size_t len, int prot, int flags, 201 int fd, off_t pos) 202 { 203 struct mmap_req mr = { 204 .mr_hint = addr0, 205 .mr_len = len, 206 .mr_prot = prot, 207 .mr_flags = flags, 208 .mr_fd = fd, 209 .mr_pos = pos 210 }; 211 212 return (kern_mmap_req(td, &mr)); 213 } 214 215 int 216 kern_mmap_req(struct thread *td, const struct mmap_req *mrp) 217 { 218 struct vmspace *vms; 219 struct file *fp; 220 struct proc *p; 221 off_t pos; 222 vm_offset_t addr, orig_addr; 223 vm_size_t len, pageoff, size; 224 vm_prot_t cap_maxprot; 225 int align, error, fd, flags, max_prot, prot; 226 cap_rights_t rights; 227 mmap_check_fp_fn check_fp_fn; 228 229 orig_addr = addr = mrp->mr_hint; 230 len = mrp->mr_len; 231 prot = mrp->mr_prot; 232 flags = mrp->mr_flags; 233 fd = mrp->mr_fd; 234 pos = mrp->mr_pos; 235 check_fp_fn = mrp->mr_check_fp_fn; 236 237 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0) 238 return (EINVAL); 239 max_prot = PROT_MAX_EXTRACT(prot); 240 prot = PROT_EXTRACT(prot); 241 if (max_prot != 0 && (max_prot & prot) != prot) 242 return (ENOTSUP); 243 244 p = td->td_proc; 245 246 /* 247 * Always honor PROT_MAX if set. If not, default to all 248 * permissions unless we're implying maximum permissions. 249 */ 250 if (max_prot == 0) 251 max_prot = kern_mmap_maxprot(p, prot); 252 253 vms = p->p_vmspace; 254 fp = NULL; 255 AUDIT_ARG_FD(fd); 256 257 /* 258 * Ignore old flags that used to be defined but did not do anything. 259 */ 260 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040); 261 262 /* 263 * Enforce the constraints. 264 * Mapping of length 0 is only allowed for old binaries. 265 * Anonymous mapping shall specify -1 as filedescriptor and 266 * zero position for new code. Be nice to ancient a.out 267 * binaries and correct pos for anonymous mapping, since old 268 * ld.so sometimes issues anonymous map requests with non-zero 269 * pos. 270 */ 271 if (!SV_CURPROC_FLAG(SV_AOUT)) { 272 if ((len == 0 && p->p_osrel >= P_OSREL_MAP_ANON) || 273 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0))) 274 return (EINVAL); 275 } else { 276 if ((flags & MAP_ANON) != 0) 277 pos = 0; 278 } 279 280 if (flags & MAP_STACK) { 281 if ((fd != -1) || 282 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) 283 return (EINVAL); 284 flags |= MAP_ANON; 285 pos = 0; 286 } 287 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE | 288 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE | 289 MAP_PREFAULT_READ | MAP_GUARD | 290 #ifdef MAP_32BIT 291 MAP_32BIT | 292 #endif 293 MAP_ALIGNMENT_MASK)) != 0) 294 return (EINVAL); 295 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL) 296 return (EINVAL); 297 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE)) 298 return (EINVAL); 299 if (prot != PROT_NONE && 300 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0) 301 return (EINVAL); 302 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 || 303 pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL | 304 #ifdef MAP_32BIT 305 MAP_32BIT | 306 #endif 307 MAP_ALIGNMENT_MASK)) != 0)) 308 return (EINVAL); 309 310 /* 311 * Align the file position to a page boundary, 312 * and save its page offset component. 313 */ 314 pageoff = (pos & PAGE_MASK); 315 pos -= pageoff; 316 317 /* Compute size from len by rounding (on both ends). */ 318 size = len + pageoff; /* low end... */ 319 size = round_page(size); /* hi end */ 320 /* Check for rounding up to zero. */ 321 if (len > size) 322 return (ENOMEM); 323 324 /* Ensure alignment is at least a page and fits in a pointer. */ 325 align = flags & MAP_ALIGNMENT_MASK; 326 if (align != 0 && align != MAP_ALIGNED_SUPER && 327 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY || 328 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT)) 329 return (EINVAL); 330 331 /* 332 * Check for illegal addresses. Watch out for address wrap... Note 333 * that VM_*_ADDRESS are not constants due to casts (argh). 334 */ 335 if (flags & MAP_FIXED) { 336 /* 337 * The specified address must have the same remainder 338 * as the file offset taken modulo PAGE_SIZE, so it 339 * should be aligned after adjustment by pageoff. 340 */ 341 addr -= pageoff; 342 if (addr & PAGE_MASK) 343 return (EINVAL); 344 345 /* Address range must be all in user VM space. */ 346 if (!vm_map_range_valid(&vms->vm_map, addr, addr + size)) 347 return (EINVAL); 348 #ifdef MAP_32BIT 349 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR) 350 return (EINVAL); 351 } else if (flags & MAP_32BIT) { 352 /* 353 * For MAP_32BIT, override the hint if it is too high and 354 * do not bother moving the mapping past the heap (since 355 * the heap is usually above 2GB). 356 */ 357 if (addr + size > MAP_32BIT_MAX_ADDR) 358 addr = 0; 359 #endif 360 } else { 361 /* 362 * XXX for non-fixed mappings where no hint is provided or 363 * the hint would fall in the potential heap space, 364 * place it after the end of the largest possible heap. 365 * 366 * There should really be a pmap call to determine a reasonable 367 * location. 368 */ 369 if (addr == 0 || 370 (addr >= round_page((vm_offset_t)vms->vm_taddr) && 371 addr < round_page((vm_offset_t)vms->vm_daddr + 372 lim_max(td, RLIMIT_DATA)))) 373 addr = round_page((vm_offset_t)vms->vm_daddr + 374 lim_max(td, RLIMIT_DATA)); 375 } 376 if (len == 0) { 377 /* 378 * Return success without mapping anything for old 379 * binaries that request a page-aligned mapping of 380 * length 0. For modern binaries, this function 381 * returns an error earlier. 382 */ 383 error = 0; 384 } else if ((flags & MAP_GUARD) != 0) { 385 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE, 386 VM_PROT_NONE, flags, NULL, pos, FALSE, td); 387 } else if ((flags & MAP_ANON) != 0) { 388 /* 389 * Mapping blank space is trivial. 390 * 391 * This relies on VM_PROT_* matching PROT_*. 392 */ 393 error = vm_mmap_object(&vms->vm_map, &addr, size, prot, 394 max_prot, flags, NULL, pos, FALSE, td); 395 } else { 396 /* 397 * Mapping file, get fp for validation and don't let the 398 * descriptor disappear on us if we block. Check capability 399 * rights, but also return the maximum rights to be combined 400 * with maxprot later. 401 */ 402 cap_rights_init_one(&rights, CAP_MMAP); 403 if (prot & PROT_READ) 404 cap_rights_set_one(&rights, CAP_MMAP_R); 405 if ((flags & MAP_SHARED) != 0) { 406 if (prot & PROT_WRITE) 407 cap_rights_set_one(&rights, CAP_MMAP_W); 408 } 409 if (prot & PROT_EXEC) 410 cap_rights_set_one(&rights, CAP_MMAP_X); 411 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp); 412 if (error != 0) 413 goto done; 414 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 && 415 p->p_osrel >= P_OSREL_MAP_FSTRICT) { 416 error = EINVAL; 417 goto done; 418 } 419 if (check_fp_fn != NULL) { 420 error = check_fp_fn(fp, prot, max_prot & cap_maxprot, 421 flags); 422 if (error != 0) 423 goto done; 424 } 425 if (fp->f_ops == &shm_ops && shm_largepage(fp->f_data)) 426 addr = orig_addr; 427 /* This relies on VM_PROT_* matching PROT_*. */ 428 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot, 429 max_prot & cap_maxprot, flags, pos, td); 430 } 431 432 if (error == 0) 433 td->td_retval[0] = (register_t) (addr + pageoff); 434 done: 435 if (fp) 436 fdrop(fp, td); 437 438 return (error); 439 } 440 441 #if defined(COMPAT_FREEBSD6) 442 int 443 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap) 444 { 445 446 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot, 447 uap->flags, uap->fd, uap->pos)); 448 } 449 #endif 450 451 #ifdef COMPAT_43 452 #ifndef _SYS_SYSPROTO_H_ 453 struct ommap_args { 454 caddr_t addr; 455 int len; 456 int prot; 457 int flags; 458 int fd; 459 long pos; 460 }; 461 #endif 462 int 463 ommap(struct thread *td, struct ommap_args *uap) 464 { 465 static const char cvtbsdprot[8] = { 466 0, 467 PROT_EXEC, 468 PROT_WRITE, 469 PROT_EXEC | PROT_WRITE, 470 PROT_READ, 471 PROT_EXEC | PROT_READ, 472 PROT_WRITE | PROT_READ, 473 PROT_EXEC | PROT_WRITE | PROT_READ, 474 }; 475 int flags, prot; 476 477 #define OMAP_ANON 0x0002 478 #define OMAP_COPY 0x0020 479 #define OMAP_SHARED 0x0010 480 #define OMAP_FIXED 0x0100 481 482 prot = cvtbsdprot[uap->prot & 0x7]; 483 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__) 484 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) && 485 prot != 0) 486 prot |= PROT_EXEC; 487 #endif 488 flags = 0; 489 if (uap->flags & OMAP_ANON) 490 flags |= MAP_ANON; 491 if (uap->flags & OMAP_COPY) 492 flags |= MAP_COPY; 493 if (uap->flags & OMAP_SHARED) 494 flags |= MAP_SHARED; 495 else 496 flags |= MAP_PRIVATE; 497 if (uap->flags & OMAP_FIXED) 498 flags |= MAP_FIXED; 499 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags, 500 uap->fd, uap->pos)); 501 } 502 #endif /* COMPAT_43 */ 503 504 #ifndef _SYS_SYSPROTO_H_ 505 struct msync_args { 506 void *addr; 507 size_t len; 508 int flags; 509 }; 510 #endif 511 int 512 sys_msync(struct thread *td, struct msync_args *uap) 513 { 514 515 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags)); 516 } 517 518 int 519 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags) 520 { 521 vm_offset_t addr; 522 vm_size_t pageoff; 523 vm_map_t map; 524 int rv; 525 526 addr = addr0; 527 pageoff = (addr & PAGE_MASK); 528 addr -= pageoff; 529 size += pageoff; 530 size = (vm_size_t) round_page(size); 531 if (addr + size < addr) 532 return (EINVAL); 533 534 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) 535 return (EINVAL); 536 537 map = &td->td_proc->p_vmspace->vm_map; 538 539 /* 540 * Clean the pages and interpret the return value. 541 */ 542 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0, 543 (flags & MS_INVALIDATE) != 0); 544 switch (rv) { 545 case KERN_SUCCESS: 546 return (0); 547 case KERN_INVALID_ADDRESS: 548 return (ENOMEM); 549 case KERN_INVALID_ARGUMENT: 550 return (EBUSY); 551 case KERN_FAILURE: 552 return (EIO); 553 default: 554 return (EINVAL); 555 } 556 } 557 558 #ifndef _SYS_SYSPROTO_H_ 559 struct munmap_args { 560 void *addr; 561 size_t len; 562 }; 563 #endif 564 int 565 sys_munmap(struct thread *td, struct munmap_args *uap) 566 { 567 568 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len)); 569 } 570 571 int 572 kern_munmap(struct thread *td, uintptr_t addr0, size_t size) 573 { 574 #ifdef HWPMC_HOOKS 575 struct pmckern_map_out pkm; 576 vm_map_entry_t entry; 577 bool pmc_handled; 578 #endif 579 vm_offset_t addr, end; 580 vm_size_t pageoff; 581 vm_map_t map; 582 int rv; 583 584 if (size == 0) 585 return (EINVAL); 586 587 addr = addr0; 588 pageoff = (addr & PAGE_MASK); 589 addr -= pageoff; 590 size += pageoff; 591 size = (vm_size_t) round_page(size); 592 end = addr + size; 593 map = &td->td_proc->p_vmspace->vm_map; 594 if (!vm_map_range_valid(map, addr, end)) 595 return (EINVAL); 596 597 vm_map_lock(map); 598 #ifdef HWPMC_HOOKS 599 pmc_handled = false; 600 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) { 601 pmc_handled = true; 602 /* 603 * Inform hwpmc if the address range being unmapped contains 604 * an executable region. 605 */ 606 pkm.pm_address = (uintptr_t) NULL; 607 if (vm_map_lookup_entry(map, addr, &entry)) { 608 for (; entry->start < end; 609 entry = vm_map_entry_succ(entry)) { 610 if (vm_map_check_protection(map, entry->start, 611 entry->end, VM_PROT_EXECUTE) == TRUE) { 612 pkm.pm_address = (uintptr_t) addr; 613 pkm.pm_size = (size_t) size; 614 break; 615 } 616 } 617 } 618 } 619 #endif 620 rv = vm_map_delete(map, addr, end); 621 622 #ifdef HWPMC_HOOKS 623 if (rv == KERN_SUCCESS && __predict_false(pmc_handled)) { 624 /* downgrade the lock to prevent a LOR with the pmc-sx lock */ 625 vm_map_lock_downgrade(map); 626 if (pkm.pm_address != (uintptr_t) NULL) 627 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm); 628 vm_map_unlock_read(map); 629 } else 630 #endif 631 vm_map_unlock(map); 632 633 return (vm_mmap_to_errno(rv)); 634 } 635 636 #ifndef _SYS_SYSPROTO_H_ 637 struct mprotect_args { 638 const void *addr; 639 size_t len; 640 int prot; 641 }; 642 #endif 643 int 644 sys_mprotect(struct thread *td, struct mprotect_args *uap) 645 { 646 647 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot)); 648 } 649 650 int 651 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot) 652 { 653 vm_offset_t addr; 654 vm_size_t pageoff; 655 int vm_error, max_prot; 656 int flags; 657 658 addr = addr0; 659 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0) 660 return (EINVAL); 661 max_prot = PROT_MAX_EXTRACT(prot); 662 prot = PROT_EXTRACT(prot); 663 pageoff = (addr & PAGE_MASK); 664 addr -= pageoff; 665 size += pageoff; 666 size = (vm_size_t) round_page(size); 667 #ifdef COMPAT_FREEBSD32 668 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 669 if (((addr + size) & 0xffffffff) < addr) 670 return (EINVAL); 671 } else 672 #endif 673 if (addr + size < addr) 674 return (EINVAL); 675 676 flags = VM_MAP_PROTECT_SET_PROT; 677 if (max_prot != 0) 678 flags |= VM_MAP_PROTECT_SET_MAXPROT; 679 vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map, 680 addr, addr + size, prot, max_prot, flags); 681 682 switch (vm_error) { 683 case KERN_SUCCESS: 684 return (0); 685 case KERN_PROTECTION_FAILURE: 686 return (EACCES); 687 case KERN_RESOURCE_SHORTAGE: 688 return (ENOMEM); 689 case KERN_OUT_OF_BOUNDS: 690 return (ENOTSUP); 691 } 692 return (EINVAL); 693 } 694 695 #ifndef _SYS_SYSPROTO_H_ 696 struct minherit_args { 697 void *addr; 698 size_t len; 699 int inherit; 700 }; 701 #endif 702 int 703 sys_minherit(struct thread *td, struct minherit_args *uap) 704 { 705 706 return (kern_minherit(td, (uintptr_t)uap->addr, uap->len, 707 uap->inherit)); 708 } 709 710 int 711 kern_minherit(struct thread *td, uintptr_t addr0, size_t len, int inherit0) 712 { 713 vm_offset_t addr; 714 vm_size_t size, pageoff; 715 vm_inherit_t inherit; 716 717 addr = (vm_offset_t)addr0; 718 size = len; 719 inherit = inherit0; 720 721 pageoff = (addr & PAGE_MASK); 722 addr -= pageoff; 723 size += pageoff; 724 size = (vm_size_t) round_page(size); 725 if (addr + size < addr) 726 return (EINVAL); 727 728 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr, 729 addr + size, inherit)) { 730 case KERN_SUCCESS: 731 return (0); 732 case KERN_PROTECTION_FAILURE: 733 return (EACCES); 734 } 735 return (EINVAL); 736 } 737 738 #ifndef _SYS_SYSPROTO_H_ 739 struct madvise_args { 740 void *addr; 741 size_t len; 742 int behav; 743 }; 744 #endif 745 746 int 747 sys_madvise(struct thread *td, struct madvise_args *uap) 748 { 749 750 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav)); 751 } 752 753 int 754 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav) 755 { 756 vm_map_t map; 757 vm_offset_t addr, end, start; 758 int flags; 759 760 /* 761 * Check for our special case, advising the swap pager we are 762 * "immortal." 763 */ 764 if (behav == MADV_PROTECT) { 765 flags = PPROT_SET; 766 return (kern_procctl(td, P_PID, td->td_proc->p_pid, 767 PROC_SPROTECT, &flags)); 768 } 769 770 /* 771 * Check for illegal addresses. Watch out for address wrap... Note 772 * that VM_*_ADDRESS are not constants due to casts (argh). 773 */ 774 map = &td->td_proc->p_vmspace->vm_map; 775 addr = addr0; 776 if (!vm_map_range_valid(map, addr, addr + len)) 777 return (EINVAL); 778 779 /* 780 * Since this routine is only advisory, we default to conservative 781 * behavior. 782 */ 783 start = trunc_page(addr); 784 end = round_page(addr + len); 785 786 /* 787 * vm_map_madvise() checks for illegal values of behav. 788 */ 789 return (vm_map_madvise(map, start, end, behav)); 790 } 791 792 #ifndef _SYS_SYSPROTO_H_ 793 struct mincore_args { 794 const void *addr; 795 size_t len; 796 char *vec; 797 }; 798 #endif 799 800 int 801 sys_mincore(struct thread *td, struct mincore_args *uap) 802 { 803 804 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec)); 805 } 806 807 int 808 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec) 809 { 810 pmap_t pmap; 811 vm_map_t map; 812 vm_map_entry_t current, entry; 813 vm_object_t object; 814 vm_offset_t addr, cend, end, first_addr; 815 vm_paddr_t pa; 816 vm_page_t m; 817 vm_pindex_t pindex; 818 int error, lastvecindex, mincoreinfo, vecindex; 819 unsigned int timestamp; 820 821 /* 822 * Make sure that the addresses presented are valid for user 823 * mode. 824 */ 825 first_addr = addr = trunc_page(addr0); 826 end = round_page(addr0 + len); 827 map = &td->td_proc->p_vmspace->vm_map; 828 if (end > vm_map_max(map) || end < addr) 829 return (ENOMEM); 830 831 pmap = vmspace_pmap(td->td_proc->p_vmspace); 832 833 vm_map_lock_read(map); 834 RestartScan: 835 timestamp = map->timestamp; 836 837 if (!vm_map_lookup_entry(map, addr, &entry)) { 838 vm_map_unlock_read(map); 839 return (ENOMEM); 840 } 841 842 /* 843 * Do this on a map entry basis so that if the pages are not 844 * in the current processes address space, we can easily look 845 * up the pages elsewhere. 846 */ 847 lastvecindex = -1; 848 while (entry->start < end) { 849 /* 850 * check for contiguity 851 */ 852 current = entry; 853 entry = vm_map_entry_succ(current); 854 if (current->end < end && 855 entry->start > current->end) { 856 vm_map_unlock_read(map); 857 return (ENOMEM); 858 } 859 860 /* 861 * ignore submaps (for now) or null objects 862 */ 863 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) || 864 current->object.vm_object == NULL) 865 continue; 866 867 /* 868 * limit this scan to the current map entry and the 869 * limits for the mincore call 870 */ 871 if (addr < current->start) 872 addr = current->start; 873 cend = current->end; 874 if (cend > end) 875 cend = end; 876 877 for (; addr < cend; addr += PAGE_SIZE) { 878 /* 879 * Check pmap first, it is likely faster, also 880 * it can provide info as to whether we are the 881 * one referencing or modifying the page. 882 */ 883 m = NULL; 884 object = NULL; 885 retry: 886 pa = 0; 887 mincoreinfo = pmap_mincore(pmap, addr, &pa); 888 if (mincore_mapped) { 889 /* 890 * We only care about this pmap's 891 * mapping of the page, if any. 892 */ 893 ; 894 } else if (pa != 0) { 895 /* 896 * The page is mapped by this process but not 897 * both accessed and modified. It is also 898 * managed. Acquire the object lock so that 899 * other mappings might be examined. The page's 900 * identity may change at any point before its 901 * object lock is acquired, so re-validate if 902 * necessary. 903 */ 904 m = PHYS_TO_VM_PAGE(pa); 905 while (object == NULL || m->object != object) { 906 if (object != NULL) 907 VM_OBJECT_WUNLOCK(object); 908 object = atomic_load_ptr(&m->object); 909 if (object == NULL) 910 goto retry; 911 VM_OBJECT_WLOCK(object); 912 } 913 if (pa != pmap_extract(pmap, addr)) 914 goto retry; 915 KASSERT(vm_page_all_valid(m), 916 ("mincore: page %p is mapped but invalid", 917 m)); 918 } else if (mincoreinfo == 0) { 919 /* 920 * The page is not mapped by this process. If 921 * the object implements managed pages, then 922 * determine if the page is resident so that 923 * the mappings might be examined. 924 */ 925 if (current->object.vm_object != object) { 926 if (object != NULL) 927 VM_OBJECT_WUNLOCK(object); 928 object = current->object.vm_object; 929 VM_OBJECT_WLOCK(object); 930 } 931 if (object->type == OBJT_DEFAULT || 932 object->type == OBJT_SWAP || 933 object->type == OBJT_VNODE) { 934 pindex = OFF_TO_IDX(current->offset + 935 (addr - current->start)); 936 m = vm_page_lookup(object, pindex); 937 if (m != NULL && vm_page_none_valid(m)) 938 m = NULL; 939 if (m != NULL) 940 mincoreinfo = MINCORE_INCORE; 941 } 942 } 943 if (m != NULL) { 944 VM_OBJECT_ASSERT_WLOCKED(m->object); 945 946 /* Examine other mappings of the page. */ 947 if (m->dirty == 0 && pmap_is_modified(m)) 948 vm_page_dirty(m); 949 if (m->dirty != 0) 950 mincoreinfo |= MINCORE_MODIFIED_OTHER; 951 952 /* 953 * The first test for PGA_REFERENCED is an 954 * optimization. The second test is 955 * required because a concurrent pmap 956 * operation could clear the last reference 957 * and set PGA_REFERENCED before the call to 958 * pmap_is_referenced(). 959 */ 960 if ((m->a.flags & PGA_REFERENCED) != 0 || 961 pmap_is_referenced(m) || 962 (m->a.flags & PGA_REFERENCED) != 0) 963 mincoreinfo |= MINCORE_REFERENCED_OTHER; 964 } 965 if (object != NULL) 966 VM_OBJECT_WUNLOCK(object); 967 968 /* 969 * subyte may page fault. In case it needs to modify 970 * the map, we release the lock. 971 */ 972 vm_map_unlock_read(map); 973 974 /* 975 * calculate index into user supplied byte vector 976 */ 977 vecindex = atop(addr - first_addr); 978 979 /* 980 * If we have skipped map entries, we need to make sure that 981 * the byte vector is zeroed for those skipped entries. 982 */ 983 while ((lastvecindex + 1) < vecindex) { 984 ++lastvecindex; 985 error = subyte(vec + lastvecindex, 0); 986 if (error) { 987 error = EFAULT; 988 goto done2; 989 } 990 } 991 992 /* 993 * Pass the page information to the user 994 */ 995 error = subyte(vec + vecindex, mincoreinfo); 996 if (error) { 997 error = EFAULT; 998 goto done2; 999 } 1000 1001 /* 1002 * If the map has changed, due to the subyte, the previous 1003 * output may be invalid. 1004 */ 1005 vm_map_lock_read(map); 1006 if (timestamp != map->timestamp) 1007 goto RestartScan; 1008 1009 lastvecindex = vecindex; 1010 } 1011 } 1012 1013 /* 1014 * subyte may page fault. In case it needs to modify 1015 * the map, we release the lock. 1016 */ 1017 vm_map_unlock_read(map); 1018 1019 /* 1020 * Zero the last entries in the byte vector. 1021 */ 1022 vecindex = atop(end - first_addr); 1023 while ((lastvecindex + 1) < vecindex) { 1024 ++lastvecindex; 1025 error = subyte(vec + lastvecindex, 0); 1026 if (error) { 1027 error = EFAULT; 1028 goto done2; 1029 } 1030 } 1031 1032 /* 1033 * If the map has changed, due to the subyte, the previous 1034 * output may be invalid. 1035 */ 1036 vm_map_lock_read(map); 1037 if (timestamp != map->timestamp) 1038 goto RestartScan; 1039 vm_map_unlock_read(map); 1040 done2: 1041 return (error); 1042 } 1043 1044 #ifndef _SYS_SYSPROTO_H_ 1045 struct mlock_args { 1046 const void *addr; 1047 size_t len; 1048 }; 1049 #endif 1050 int 1051 sys_mlock(struct thread *td, struct mlock_args *uap) 1052 { 1053 1054 return (kern_mlock(td->td_proc, td->td_ucred, 1055 __DECONST(uintptr_t, uap->addr), uap->len)); 1056 } 1057 1058 int 1059 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len) 1060 { 1061 vm_offset_t addr, end, last, start; 1062 vm_size_t npages, size; 1063 vm_map_t map; 1064 unsigned long nsize; 1065 int error; 1066 1067 error = priv_check_cred(cred, PRIV_VM_MLOCK); 1068 if (error) 1069 return (error); 1070 addr = addr0; 1071 size = len; 1072 last = addr + size; 1073 start = trunc_page(addr); 1074 end = round_page(last); 1075 if (last < addr || end < addr) 1076 return (EINVAL); 1077 npages = atop(end - start); 1078 if (npages > vm_page_max_user_wired) 1079 return (ENOMEM); 1080 map = &proc->p_vmspace->vm_map; 1081 PROC_LOCK(proc); 1082 nsize = ptoa(npages + pmap_wired_count(map->pmap)); 1083 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) { 1084 PROC_UNLOCK(proc); 1085 return (ENOMEM); 1086 } 1087 PROC_UNLOCK(proc); 1088 #ifdef RACCT 1089 if (racct_enable) { 1090 PROC_LOCK(proc); 1091 error = racct_set(proc, RACCT_MEMLOCK, nsize); 1092 PROC_UNLOCK(proc); 1093 if (error != 0) 1094 return (ENOMEM); 1095 } 1096 #endif 1097 error = vm_map_wire(map, start, end, 1098 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 1099 #ifdef RACCT 1100 if (racct_enable && error != KERN_SUCCESS) { 1101 PROC_LOCK(proc); 1102 racct_set(proc, RACCT_MEMLOCK, 1103 ptoa(pmap_wired_count(map->pmap))); 1104 PROC_UNLOCK(proc); 1105 } 1106 #endif 1107 switch (error) { 1108 case KERN_SUCCESS: 1109 return (0); 1110 case KERN_INVALID_ARGUMENT: 1111 return (EINVAL); 1112 default: 1113 return (ENOMEM); 1114 } 1115 } 1116 1117 #ifndef _SYS_SYSPROTO_H_ 1118 struct mlockall_args { 1119 int how; 1120 }; 1121 #endif 1122 1123 int 1124 sys_mlockall(struct thread *td, struct mlockall_args *uap) 1125 { 1126 vm_map_t map; 1127 int error; 1128 1129 map = &td->td_proc->p_vmspace->vm_map; 1130 error = priv_check(td, PRIV_VM_MLOCK); 1131 if (error) 1132 return (error); 1133 1134 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0)) 1135 return (EINVAL); 1136 1137 /* 1138 * If wiring all pages in the process would cause it to exceed 1139 * a hard resource limit, return ENOMEM. 1140 */ 1141 if (!old_mlock && uap->how & MCL_CURRENT) { 1142 if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) 1143 return (ENOMEM); 1144 } 1145 #ifdef RACCT 1146 if (racct_enable) { 1147 PROC_LOCK(td->td_proc); 1148 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size); 1149 PROC_UNLOCK(td->td_proc); 1150 if (error != 0) 1151 return (ENOMEM); 1152 } 1153 #endif 1154 1155 if (uap->how & MCL_FUTURE) { 1156 vm_map_lock(map); 1157 vm_map_modflags(map, MAP_WIREFUTURE, 0); 1158 vm_map_unlock(map); 1159 error = 0; 1160 } 1161 1162 if (uap->how & MCL_CURRENT) { 1163 /* 1164 * P1003.1-2001 mandates that all currently mapped pages 1165 * will be memory resident and locked (wired) upon return 1166 * from mlockall(). vm_map_wire() will wire pages, by 1167 * calling vm_fault_wire() for each page in the region. 1168 */ 1169 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map), 1170 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); 1171 if (error == KERN_SUCCESS) 1172 error = 0; 1173 else if (error == KERN_RESOURCE_SHORTAGE) 1174 error = ENOMEM; 1175 else 1176 error = EAGAIN; 1177 } 1178 #ifdef RACCT 1179 if (racct_enable && error != KERN_SUCCESS) { 1180 PROC_LOCK(td->td_proc); 1181 racct_set(td->td_proc, RACCT_MEMLOCK, 1182 ptoa(pmap_wired_count(map->pmap))); 1183 PROC_UNLOCK(td->td_proc); 1184 } 1185 #endif 1186 1187 return (error); 1188 } 1189 1190 #ifndef _SYS_SYSPROTO_H_ 1191 struct munlockall_args { 1192 register_t dummy; 1193 }; 1194 #endif 1195 1196 int 1197 sys_munlockall(struct thread *td, struct munlockall_args *uap) 1198 { 1199 vm_map_t map; 1200 int error; 1201 1202 map = &td->td_proc->p_vmspace->vm_map; 1203 error = priv_check(td, PRIV_VM_MUNLOCK); 1204 if (error) 1205 return (error); 1206 1207 /* Clear the MAP_WIREFUTURE flag from this vm_map. */ 1208 vm_map_lock(map); 1209 vm_map_modflags(map, 0, MAP_WIREFUTURE); 1210 vm_map_unlock(map); 1211 1212 /* Forcibly unwire all pages. */ 1213 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map), 1214 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); 1215 #ifdef RACCT 1216 if (racct_enable && error == KERN_SUCCESS) { 1217 PROC_LOCK(td->td_proc); 1218 racct_set(td->td_proc, RACCT_MEMLOCK, 0); 1219 PROC_UNLOCK(td->td_proc); 1220 } 1221 #endif 1222 1223 return (error); 1224 } 1225 1226 #ifndef _SYS_SYSPROTO_H_ 1227 struct munlock_args { 1228 const void *addr; 1229 size_t len; 1230 }; 1231 #endif 1232 int 1233 sys_munlock(struct thread *td, struct munlock_args *uap) 1234 { 1235 1236 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len)); 1237 } 1238 1239 int 1240 kern_munlock(struct thread *td, uintptr_t addr0, size_t size) 1241 { 1242 vm_offset_t addr, end, last, start; 1243 #ifdef RACCT 1244 vm_map_t map; 1245 #endif 1246 int error; 1247 1248 error = priv_check(td, PRIV_VM_MUNLOCK); 1249 if (error) 1250 return (error); 1251 addr = addr0; 1252 last = addr + size; 1253 start = trunc_page(addr); 1254 end = round_page(last); 1255 if (last < addr || end < addr) 1256 return (EINVAL); 1257 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end, 1258 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 1259 #ifdef RACCT 1260 if (racct_enable && error == KERN_SUCCESS) { 1261 PROC_LOCK(td->td_proc); 1262 map = &td->td_proc->p_vmspace->vm_map; 1263 racct_set(td->td_proc, RACCT_MEMLOCK, 1264 ptoa(pmap_wired_count(map->pmap))); 1265 PROC_UNLOCK(td->td_proc); 1266 } 1267 #endif 1268 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1269 } 1270 1271 /* 1272 * vm_mmap_vnode() 1273 * 1274 * Helper function for vm_mmap. Perform sanity check specific for mmap 1275 * operations on vnodes. 1276 */ 1277 int 1278 vm_mmap_vnode(struct thread *td, vm_size_t objsize, 1279 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp, 1280 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp, 1281 boolean_t *writecounted) 1282 { 1283 struct vattr va; 1284 vm_object_t obj; 1285 vm_ooffset_t foff; 1286 struct ucred *cred; 1287 int error, flags; 1288 bool writex; 1289 1290 cred = td->td_ucred; 1291 writex = (*maxprotp & VM_PROT_WRITE) != 0 && 1292 (*flagsp & MAP_SHARED) != 0; 1293 if ((error = vget(vp, LK_SHARED)) != 0) 1294 return (error); 1295 AUDIT_ARG_VNODE1(vp); 1296 foff = *foffp; 1297 flags = *flagsp; 1298 obj = vp->v_object; 1299 if (vp->v_type == VREG) { 1300 /* 1301 * Get the proper underlying object 1302 */ 1303 if (obj == NULL) { 1304 error = EINVAL; 1305 goto done; 1306 } 1307 if (obj->type == OBJT_VNODE && obj->handle != vp) { 1308 vput(vp); 1309 vp = (struct vnode *)obj->handle; 1310 /* 1311 * Bypass filesystems obey the mpsafety of the 1312 * underlying fs. Tmpfs never bypasses. 1313 */ 1314 error = vget(vp, LK_SHARED); 1315 if (error != 0) 1316 return (error); 1317 } 1318 if (writex) { 1319 *writecounted = TRUE; 1320 vm_pager_update_writecount(obj, 0, objsize); 1321 } 1322 } else { 1323 error = EINVAL; 1324 goto done; 1325 } 1326 if ((error = VOP_GETATTR(vp, &va, cred))) 1327 goto done; 1328 #ifdef MAC 1329 /* This relies on VM_PROT_* matching PROT_*. */ 1330 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags); 1331 if (error != 0) 1332 goto done; 1333 #endif 1334 if ((flags & MAP_SHARED) != 0) { 1335 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) { 1336 if (prot & VM_PROT_WRITE) { 1337 error = EPERM; 1338 goto done; 1339 } 1340 *maxprotp &= ~VM_PROT_WRITE; 1341 } 1342 } 1343 /* 1344 * If it is a regular file without any references 1345 * we do not need to sync it. 1346 * Adjust object size to be the size of actual file. 1347 */ 1348 objsize = round_page(va.va_size); 1349 if (va.va_nlink == 0) 1350 flags |= MAP_NOSYNC; 1351 if (obj->type == OBJT_VNODE) { 1352 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, 1353 cred); 1354 if (obj == NULL) { 1355 error = ENOMEM; 1356 goto done; 1357 } 1358 } else { 1359 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP, 1360 ("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 /* 1446 * vm_mmap() 1447 * 1448 * Internal version of mmap used by exec, sys5 shared memory, and 1449 * various device drivers. Handle is either a vnode pointer, a 1450 * character device, or NULL for MAP_ANON. 1451 */ 1452 int 1453 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1454 vm_prot_t maxprot, int flags, 1455 objtype_t handle_type, void *handle, 1456 vm_ooffset_t foff) 1457 { 1458 vm_object_t object; 1459 struct thread *td = curthread; 1460 int error; 1461 boolean_t writecounted; 1462 1463 if (size == 0) 1464 return (EINVAL); 1465 1466 size = round_page(size); 1467 object = NULL; 1468 writecounted = FALSE; 1469 1470 /* 1471 * Lookup/allocate object. 1472 */ 1473 switch (handle_type) { 1474 case OBJT_DEVICE: { 1475 struct cdevsw *dsw; 1476 struct cdev *cdev; 1477 int ref; 1478 1479 cdev = handle; 1480 dsw = dev_refthread(cdev, &ref); 1481 if (dsw == NULL) 1482 return (ENXIO); 1483 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev, 1484 dsw, &foff, &object); 1485 dev_relthread(cdev, ref); 1486 break; 1487 } 1488 case OBJT_VNODE: 1489 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, 1490 handle, &foff, &object, &writecounted); 1491 break; 1492 case OBJT_DEFAULT: 1493 if (handle == NULL) { 1494 error = 0; 1495 break; 1496 } 1497 /* FALLTHROUGH */ 1498 default: 1499 error = EINVAL; 1500 break; 1501 } 1502 if (error) 1503 return (error); 1504 1505 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1506 foff, writecounted, td); 1507 if (error != 0 && object != NULL) { 1508 /* 1509 * If this mapping was accounted for in the vnode's 1510 * writecount, then undo that now. 1511 */ 1512 if (writecounted) 1513 vm_pager_release_writecount(object, 0, size); 1514 vm_object_deallocate(object); 1515 } 1516 return (error); 1517 } 1518 1519 int 1520 kern_mmap_racct_check(struct thread *td, vm_map_t map, vm_size_t size) 1521 { 1522 int error; 1523 1524 RACCT_PROC_LOCK(td->td_proc); 1525 if (map->size + size > lim_cur(td, RLIMIT_VMEM)) { 1526 RACCT_PROC_UNLOCK(td->td_proc); 1527 return (ENOMEM); 1528 } 1529 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) { 1530 RACCT_PROC_UNLOCK(td->td_proc); 1531 return (ENOMEM); 1532 } 1533 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 1534 if (ptoa(pmap_wired_count(map->pmap)) + size > 1535 lim_cur(td, RLIMIT_MEMLOCK)) { 1536 racct_set_force(td->td_proc, RACCT_VMEM, map->size); 1537 RACCT_PROC_UNLOCK(td->td_proc); 1538 return (ENOMEM); 1539 } 1540 error = racct_set(td->td_proc, RACCT_MEMLOCK, 1541 ptoa(pmap_wired_count(map->pmap)) + size); 1542 if (error != 0) { 1543 racct_set_force(td->td_proc, RACCT_VMEM, map->size); 1544 RACCT_PROC_UNLOCK(td->td_proc); 1545 return (error); 1546 } 1547 } 1548 RACCT_PROC_UNLOCK(td->td_proc); 1549 return (0); 1550 } 1551 1552 /* 1553 * Internal version of mmap that maps a specific VM object into an 1554 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap. 1555 */ 1556 int 1557 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1558 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff, 1559 boolean_t writecounted, struct thread *td) 1560 { 1561 vm_offset_t max_addr; 1562 int docow, error, findspace, rv; 1563 bool curmap, fitit; 1564 1565 curmap = map == &td->td_proc->p_vmspace->vm_map; 1566 if (curmap) { 1567 error = kern_mmap_racct_check(td, map, size); 1568 if (error != 0) 1569 return (error); 1570 } 1571 1572 /* 1573 * We currently can only deal with page aligned file offsets. 1574 * The mmap() system call already enforces this by subtracting 1575 * the page offset from the file offset, but checking here 1576 * catches errors in device drivers (e.g. d_single_mmap() 1577 * callbacks) and other internal mapping requests (such as in 1578 * exec). 1579 */ 1580 if (foff & PAGE_MASK) 1581 return (EINVAL); 1582 1583 if ((flags & MAP_FIXED) == 0) { 1584 fitit = TRUE; 1585 *addr = round_page(*addr); 1586 } else { 1587 if (*addr != trunc_page(*addr)) 1588 return (EINVAL); 1589 fitit = FALSE; 1590 } 1591 1592 if (flags & MAP_ANON) { 1593 if (object != NULL || foff != 0) 1594 return (EINVAL); 1595 docow = 0; 1596 } else if (flags & MAP_PREFAULT_READ) 1597 docow = MAP_PREFAULT; 1598 else 1599 docow = MAP_PREFAULT_PARTIAL; 1600 1601 if ((flags & (MAP_ANON|MAP_SHARED)) == 0) 1602 docow |= MAP_COPY_ON_WRITE; 1603 if (flags & MAP_NOSYNC) 1604 docow |= MAP_DISABLE_SYNCER; 1605 if (flags & MAP_NOCORE) 1606 docow |= MAP_DISABLE_COREDUMP; 1607 /* Shared memory is also shared with children. */ 1608 if (flags & MAP_SHARED) 1609 docow |= MAP_INHERIT_SHARE; 1610 if (writecounted) 1611 docow |= MAP_WRITECOUNT; 1612 if (flags & MAP_STACK) { 1613 if (object != NULL) 1614 return (EINVAL); 1615 docow |= MAP_STACK_GROWS_DOWN; 1616 } 1617 if ((flags & MAP_EXCL) != 0) 1618 docow |= MAP_CHECK_EXCL; 1619 if ((flags & MAP_GUARD) != 0) 1620 docow |= MAP_CREATE_GUARD; 1621 1622 if (fitit) { 1623 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER) 1624 findspace = VMFS_SUPER_SPACE; 1625 else if ((flags & MAP_ALIGNMENT_MASK) != 0) 1626 findspace = VMFS_ALIGNED_SPACE(flags >> 1627 MAP_ALIGNMENT_SHIFT); 1628 else 1629 findspace = VMFS_OPTIMAL_SPACE; 1630 max_addr = 0; 1631 #ifdef MAP_32BIT 1632 if ((flags & MAP_32BIT) != 0) 1633 max_addr = MAP_32BIT_MAX_ADDR; 1634 #endif 1635 if (curmap) { 1636 rv = vm_map_find_min(map, object, foff, addr, size, 1637 round_page((vm_offset_t)td->td_proc->p_vmspace-> 1638 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr, 1639 findspace, prot, maxprot, docow); 1640 } else { 1641 rv = vm_map_find(map, object, foff, addr, size, 1642 max_addr, findspace, prot, maxprot, docow); 1643 } 1644 } else { 1645 rv = vm_map_fixed(map, object, foff, *addr, size, 1646 prot, maxprot, docow); 1647 } 1648 1649 if (rv == KERN_SUCCESS) { 1650 /* 1651 * If the process has requested that all future mappings 1652 * be wired, then heed this. 1653 */ 1654 if ((map->flags & MAP_WIREFUTURE) != 0) { 1655 vm_map_lock(map); 1656 if ((map->flags & MAP_WIREFUTURE) != 0) 1657 (void)vm_map_wire_locked(map, *addr, 1658 *addr + size, VM_MAP_WIRE_USER | 1659 ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK : 1660 VM_MAP_WIRE_NOHOLES)); 1661 vm_map_unlock(map); 1662 } 1663 } 1664 return (vm_mmap_to_errno(rv)); 1665 } 1666 1667 /* 1668 * Translate a Mach VM return code to zero on success or the appropriate errno 1669 * on failure. 1670 */ 1671 int 1672 vm_mmap_to_errno(int rv) 1673 { 1674 1675 switch (rv) { 1676 case KERN_SUCCESS: 1677 return (0); 1678 case KERN_INVALID_ADDRESS: 1679 case KERN_NO_SPACE: 1680 return (ENOMEM); 1681 case KERN_PROTECTION_FAILURE: 1682 return (EACCES); 1683 default: 1684 return (EINVAL); 1685 } 1686 } 1687