1 /* 2 * linux/mm/nommu.c 3 * 4 * Replacement code for mm functions to support CPU's that don't 5 * have any form of memory management unit (thus no virtual memory). 6 * 7 * See Documentation/nommu-mmap.txt 8 * 9 * Copyright (c) 2004-2005 David Howells <dhowells@redhat.com> 10 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com> 11 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org> 12 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com> 13 * Copyright (c) 2007 Paul Mundt <lethal@linux-sh.org> 14 */ 15 16 #include <linux/module.h> 17 #include <linux/mm.h> 18 #include <linux/mman.h> 19 #include <linux/swap.h> 20 #include <linux/file.h> 21 #include <linux/highmem.h> 22 #include <linux/pagemap.h> 23 #include <linux/slab.h> 24 #include <linux/vmalloc.h> 25 #include <linux/tracehook.h> 26 #include <linux/blkdev.h> 27 #include <linux/backing-dev.h> 28 #include <linux/mount.h> 29 #include <linux/personality.h> 30 #include <linux/security.h> 31 #include <linux/syscalls.h> 32 33 #include <asm/uaccess.h> 34 #include <asm/tlb.h> 35 #include <asm/tlbflush.h> 36 37 void *high_memory; 38 struct page *mem_map; 39 unsigned long max_mapnr; 40 unsigned long num_physpages; 41 unsigned long askedalloc, realalloc; 42 atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0); 43 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 44 int sysctl_overcommit_ratio = 50; /* default is 50% */ 45 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; 46 int heap_stack_gap = 0; 47 48 EXPORT_SYMBOL(mem_map); 49 EXPORT_SYMBOL(num_physpages); 50 51 /* list of shareable VMAs */ 52 struct rb_root nommu_vma_tree = RB_ROOT; 53 DECLARE_RWSEM(nommu_vma_sem); 54 55 struct vm_operations_struct generic_file_vm_ops = { 56 }; 57 58 /* 59 * Handle all mappings that got truncated by a "truncate()" 60 * system call. 61 * 62 * NOTE! We have to be ready to update the memory sharing 63 * between the file and the memory map for a potential last 64 * incomplete page. Ugly, but necessary. 65 */ 66 int vmtruncate(struct inode *inode, loff_t offset) 67 { 68 struct address_space *mapping = inode->i_mapping; 69 unsigned long limit; 70 71 if (inode->i_size < offset) 72 goto do_expand; 73 i_size_write(inode, offset); 74 75 truncate_inode_pages(mapping, offset); 76 goto out_truncate; 77 78 do_expand: 79 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; 80 if (limit != RLIM_INFINITY && offset > limit) 81 goto out_sig; 82 if (offset > inode->i_sb->s_maxbytes) 83 goto out; 84 i_size_write(inode, offset); 85 86 out_truncate: 87 if (inode->i_op && inode->i_op->truncate) 88 inode->i_op->truncate(inode); 89 return 0; 90 out_sig: 91 send_sig(SIGXFSZ, current, 0); 92 out: 93 return -EFBIG; 94 } 95 96 EXPORT_SYMBOL(vmtruncate); 97 98 /* 99 * Return the total memory allocated for this pointer, not 100 * just what the caller asked for. 101 * 102 * Doesn't have to be accurate, i.e. may have races. 103 */ 104 unsigned int kobjsize(const void *objp) 105 { 106 struct page *page; 107 108 /* 109 * If the object we have should not have ksize performed on it, 110 * return size of 0 111 */ 112 if (!objp || !virt_addr_valid(objp)) 113 return 0; 114 115 page = virt_to_head_page(objp); 116 117 /* 118 * If the allocator sets PageSlab, we know the pointer came from 119 * kmalloc(). 120 */ 121 if (PageSlab(page)) 122 return ksize(objp); 123 124 /* 125 * The ksize() function is only guaranteed to work for pointers 126 * returned by kmalloc(). So handle arbitrary pointers here. 127 */ 128 return PAGE_SIZE << compound_order(page); 129 } 130 131 /* 132 * get a list of pages in an address range belonging to the specified process 133 * and indicate the VMA that covers each page 134 * - this is potentially dodgy as we may end incrementing the page count of a 135 * slab page or a secondary page from a compound page 136 * - don't permit access to VMAs that don't support it, such as I/O mappings 137 */ 138 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 139 unsigned long start, int len, int write, int force, 140 struct page **pages, struct vm_area_struct **vmas) 141 { 142 struct vm_area_struct *vma; 143 unsigned long vm_flags; 144 int i; 145 146 /* calculate required read or write permissions. 147 * - if 'force' is set, we only require the "MAY" flags. 148 */ 149 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 150 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 151 152 for (i = 0; i < len; i++) { 153 vma = find_vma(mm, start); 154 if (!vma) 155 goto finish_or_fault; 156 157 /* protect what we can, including chardevs */ 158 if (vma->vm_flags & (VM_IO | VM_PFNMAP) || 159 !(vm_flags & vma->vm_flags)) 160 goto finish_or_fault; 161 162 if (pages) { 163 pages[i] = virt_to_page(start); 164 if (pages[i]) 165 page_cache_get(pages[i]); 166 } 167 if (vmas) 168 vmas[i] = vma; 169 start += PAGE_SIZE; 170 } 171 172 return i; 173 174 finish_or_fault: 175 return i ? : -EFAULT; 176 } 177 EXPORT_SYMBOL(get_user_pages); 178 179 DEFINE_RWLOCK(vmlist_lock); 180 struct vm_struct *vmlist; 181 182 void vfree(const void *addr) 183 { 184 kfree(addr); 185 } 186 EXPORT_SYMBOL(vfree); 187 188 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 189 { 190 /* 191 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 192 * returns only a logical address. 193 */ 194 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 195 } 196 EXPORT_SYMBOL(__vmalloc); 197 198 void *vmalloc_user(unsigned long size) 199 { 200 void *ret; 201 202 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 203 PAGE_KERNEL); 204 if (ret) { 205 struct vm_area_struct *vma; 206 207 down_write(¤t->mm->mmap_sem); 208 vma = find_vma(current->mm, (unsigned long)ret); 209 if (vma) 210 vma->vm_flags |= VM_USERMAP; 211 up_write(¤t->mm->mmap_sem); 212 } 213 214 return ret; 215 } 216 EXPORT_SYMBOL(vmalloc_user); 217 218 struct page *vmalloc_to_page(const void *addr) 219 { 220 return virt_to_page(addr); 221 } 222 EXPORT_SYMBOL(vmalloc_to_page); 223 224 unsigned long vmalloc_to_pfn(const void *addr) 225 { 226 return page_to_pfn(virt_to_page(addr)); 227 } 228 EXPORT_SYMBOL(vmalloc_to_pfn); 229 230 long vread(char *buf, char *addr, unsigned long count) 231 { 232 memcpy(buf, addr, count); 233 return count; 234 } 235 236 long vwrite(char *buf, char *addr, unsigned long count) 237 { 238 /* Don't allow overflow */ 239 if ((unsigned long) addr + count < count) 240 count = -(unsigned long) addr; 241 242 memcpy(addr, buf, count); 243 return(count); 244 } 245 246 /* 247 * vmalloc - allocate virtually continguos memory 248 * 249 * @size: allocation size 250 * 251 * Allocate enough pages to cover @size from the page level 252 * allocator and map them into continguos kernel virtual space. 253 * 254 * For tight control over page level allocator and protection flags 255 * use __vmalloc() instead. 256 */ 257 void *vmalloc(unsigned long size) 258 { 259 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 260 } 261 EXPORT_SYMBOL(vmalloc); 262 263 void *vmalloc_node(unsigned long size, int node) 264 { 265 return vmalloc(size); 266 } 267 EXPORT_SYMBOL(vmalloc_node); 268 269 #ifndef PAGE_KERNEL_EXEC 270 # define PAGE_KERNEL_EXEC PAGE_KERNEL 271 #endif 272 273 /** 274 * vmalloc_exec - allocate virtually contiguous, executable memory 275 * @size: allocation size 276 * 277 * Kernel-internal function to allocate enough pages to cover @size 278 * the page level allocator and map them into contiguous and 279 * executable kernel virtual space. 280 * 281 * For tight control over page level allocator and protection flags 282 * use __vmalloc() instead. 283 */ 284 285 void *vmalloc_exec(unsigned long size) 286 { 287 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); 288 } 289 290 /** 291 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 292 * @size: allocation size 293 * 294 * Allocate enough 32bit PA addressable pages to cover @size from the 295 * page level allocator and map them into continguos kernel virtual space. 296 */ 297 void *vmalloc_32(unsigned long size) 298 { 299 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 300 } 301 EXPORT_SYMBOL(vmalloc_32); 302 303 /** 304 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 305 * @size: allocation size 306 * 307 * The resulting memory area is 32bit addressable and zeroed so it can be 308 * mapped to userspace without leaking data. 309 * 310 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to 311 * remap_vmalloc_range() are permissible. 312 */ 313 void *vmalloc_32_user(unsigned long size) 314 { 315 /* 316 * We'll have to sort out the ZONE_DMA bits for 64-bit, 317 * but for now this can simply use vmalloc_user() directly. 318 */ 319 return vmalloc_user(size); 320 } 321 EXPORT_SYMBOL(vmalloc_32_user); 322 323 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 324 { 325 BUG(); 326 return NULL; 327 } 328 EXPORT_SYMBOL(vmap); 329 330 void vunmap(const void *addr) 331 { 332 BUG(); 333 } 334 EXPORT_SYMBOL(vunmap); 335 336 /* 337 * Implement a stub for vmalloc_sync_all() if the architecture chose not to 338 * have one. 339 */ 340 void __attribute__((weak)) vmalloc_sync_all(void) 341 { 342 } 343 344 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 345 struct page *page) 346 { 347 return -EINVAL; 348 } 349 EXPORT_SYMBOL(vm_insert_page); 350 351 /* 352 * sys_brk() for the most part doesn't need the global kernel 353 * lock, except when an application is doing something nasty 354 * like trying to un-brk an area that has already been mapped 355 * to a regular file. in this case, the unmapping will need 356 * to invoke file system routines that need the global lock. 357 */ 358 asmlinkage unsigned long sys_brk(unsigned long brk) 359 { 360 struct mm_struct *mm = current->mm; 361 362 if (brk < mm->start_brk || brk > mm->context.end_brk) 363 return mm->brk; 364 365 if (mm->brk == brk) 366 return mm->brk; 367 368 /* 369 * Always allow shrinking brk 370 */ 371 if (brk <= mm->brk) { 372 mm->brk = brk; 373 return brk; 374 } 375 376 /* 377 * Ok, looks good - let it rip. 378 */ 379 return mm->brk = brk; 380 } 381 382 #ifdef DEBUG 383 static void show_process_blocks(void) 384 { 385 struct vm_list_struct *vml; 386 387 printk("Process blocks %d:", current->pid); 388 389 for (vml = ¤t->mm->context.vmlist; vml; vml = vml->next) { 390 printk(" %p: %p", vml, vml->vma); 391 if (vml->vma) 392 printk(" (%d @%lx #%d)", 393 kobjsize((void *) vml->vma->vm_start), 394 vml->vma->vm_start, 395 atomic_read(&vml->vma->vm_usage)); 396 printk(vml->next ? " ->" : ".\n"); 397 } 398 } 399 #endif /* DEBUG */ 400 401 /* 402 * add a VMA into a process's mm_struct in the appropriate place in the list 403 * - should be called with mm->mmap_sem held writelocked 404 */ 405 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml) 406 { 407 struct vm_list_struct **ppv; 408 409 for (ppv = ¤t->mm->context.vmlist; *ppv; ppv = &(*ppv)->next) 410 if ((*ppv)->vma->vm_start > vml->vma->vm_start) 411 break; 412 413 vml->next = *ppv; 414 *ppv = vml; 415 } 416 417 /* 418 * look up the first VMA in which addr resides, NULL if none 419 * - should be called with mm->mmap_sem at least held readlocked 420 */ 421 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 422 { 423 struct vm_list_struct *loop, *vml; 424 425 /* search the vm_start ordered list */ 426 vml = NULL; 427 for (loop = mm->context.vmlist; loop; loop = loop->next) { 428 if (loop->vma->vm_start > addr) 429 break; 430 vml = loop; 431 } 432 433 if (vml && vml->vma->vm_end > addr) 434 return vml->vma; 435 436 return NULL; 437 } 438 EXPORT_SYMBOL(find_vma); 439 440 /* 441 * find a VMA 442 * - we don't extend stack VMAs under NOMMU conditions 443 */ 444 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) 445 { 446 return find_vma(mm, addr); 447 } 448 449 int expand_stack(struct vm_area_struct *vma, unsigned long address) 450 { 451 return -ENOMEM; 452 } 453 454 /* 455 * look up the first VMA exactly that exactly matches addr 456 * - should be called with mm->mmap_sem at least held readlocked 457 */ 458 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm, 459 unsigned long addr) 460 { 461 struct vm_list_struct *vml; 462 463 /* search the vm_start ordered list */ 464 for (vml = mm->context.vmlist; vml; vml = vml->next) { 465 if (vml->vma->vm_start == addr) 466 return vml->vma; 467 if (vml->vma->vm_start > addr) 468 break; 469 } 470 471 return NULL; 472 } 473 474 /* 475 * find a VMA in the global tree 476 */ 477 static inline struct vm_area_struct *find_nommu_vma(unsigned long start) 478 { 479 struct vm_area_struct *vma; 480 struct rb_node *n = nommu_vma_tree.rb_node; 481 482 while (n) { 483 vma = rb_entry(n, struct vm_area_struct, vm_rb); 484 485 if (start < vma->vm_start) 486 n = n->rb_left; 487 else if (start > vma->vm_start) 488 n = n->rb_right; 489 else 490 return vma; 491 } 492 493 return NULL; 494 } 495 496 /* 497 * add a VMA in the global tree 498 */ 499 static void add_nommu_vma(struct vm_area_struct *vma) 500 { 501 struct vm_area_struct *pvma; 502 struct address_space *mapping; 503 struct rb_node **p = &nommu_vma_tree.rb_node; 504 struct rb_node *parent = NULL; 505 506 /* add the VMA to the mapping */ 507 if (vma->vm_file) { 508 mapping = vma->vm_file->f_mapping; 509 510 flush_dcache_mmap_lock(mapping); 511 vma_prio_tree_insert(vma, &mapping->i_mmap); 512 flush_dcache_mmap_unlock(mapping); 513 } 514 515 /* add the VMA to the master list */ 516 while (*p) { 517 parent = *p; 518 pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 519 520 if (vma->vm_start < pvma->vm_start) { 521 p = &(*p)->rb_left; 522 } 523 else if (vma->vm_start > pvma->vm_start) { 524 p = &(*p)->rb_right; 525 } 526 else { 527 /* mappings are at the same address - this can only 528 * happen for shared-mem chardevs and shared file 529 * mappings backed by ramfs/tmpfs */ 530 BUG_ON(!(pvma->vm_flags & VM_SHARED)); 531 532 if (vma < pvma) 533 p = &(*p)->rb_left; 534 else if (vma > pvma) 535 p = &(*p)->rb_right; 536 else 537 BUG(); 538 } 539 } 540 541 rb_link_node(&vma->vm_rb, parent, p); 542 rb_insert_color(&vma->vm_rb, &nommu_vma_tree); 543 } 544 545 /* 546 * delete a VMA from the global list 547 */ 548 static void delete_nommu_vma(struct vm_area_struct *vma) 549 { 550 struct address_space *mapping; 551 552 /* remove the VMA from the mapping */ 553 if (vma->vm_file) { 554 mapping = vma->vm_file->f_mapping; 555 556 flush_dcache_mmap_lock(mapping); 557 vma_prio_tree_remove(vma, &mapping->i_mmap); 558 flush_dcache_mmap_unlock(mapping); 559 } 560 561 /* remove from the master list */ 562 rb_erase(&vma->vm_rb, &nommu_vma_tree); 563 } 564 565 /* 566 * determine whether a mapping should be permitted and, if so, what sort of 567 * mapping we're capable of supporting 568 */ 569 static int validate_mmap_request(struct file *file, 570 unsigned long addr, 571 unsigned long len, 572 unsigned long prot, 573 unsigned long flags, 574 unsigned long pgoff, 575 unsigned long *_capabilities) 576 { 577 unsigned long capabilities; 578 unsigned long reqprot = prot; 579 int ret; 580 581 /* do the simple checks first */ 582 if (flags & MAP_FIXED || addr) { 583 printk(KERN_DEBUG 584 "%d: Can't do fixed-address/overlay mmap of RAM\n", 585 current->pid); 586 return -EINVAL; 587 } 588 589 if ((flags & MAP_TYPE) != MAP_PRIVATE && 590 (flags & MAP_TYPE) != MAP_SHARED) 591 return -EINVAL; 592 593 if (!len) 594 return -EINVAL; 595 596 /* Careful about overflows.. */ 597 len = PAGE_ALIGN(len); 598 if (!len || len > TASK_SIZE) 599 return -ENOMEM; 600 601 /* offset overflow? */ 602 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 603 return -EOVERFLOW; 604 605 if (file) { 606 /* validate file mapping requests */ 607 struct address_space *mapping; 608 609 /* files must support mmap */ 610 if (!file->f_op || !file->f_op->mmap) 611 return -ENODEV; 612 613 /* work out if what we've got could possibly be shared 614 * - we support chardevs that provide their own "memory" 615 * - we support files/blockdevs that are memory backed 616 */ 617 mapping = file->f_mapping; 618 if (!mapping) 619 mapping = file->f_path.dentry->d_inode->i_mapping; 620 621 capabilities = 0; 622 if (mapping && mapping->backing_dev_info) 623 capabilities = mapping->backing_dev_info->capabilities; 624 625 if (!capabilities) { 626 /* no explicit capabilities set, so assume some 627 * defaults */ 628 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) { 629 case S_IFREG: 630 case S_IFBLK: 631 capabilities = BDI_CAP_MAP_COPY; 632 break; 633 634 case S_IFCHR: 635 capabilities = 636 BDI_CAP_MAP_DIRECT | 637 BDI_CAP_READ_MAP | 638 BDI_CAP_WRITE_MAP; 639 break; 640 641 default: 642 return -EINVAL; 643 } 644 } 645 646 /* eliminate any capabilities that we can't support on this 647 * device */ 648 if (!file->f_op->get_unmapped_area) 649 capabilities &= ~BDI_CAP_MAP_DIRECT; 650 if (!file->f_op->read) 651 capabilities &= ~BDI_CAP_MAP_COPY; 652 653 if (flags & MAP_SHARED) { 654 /* do checks for writing, appending and locking */ 655 if ((prot & PROT_WRITE) && 656 !(file->f_mode & FMODE_WRITE)) 657 return -EACCES; 658 659 if (IS_APPEND(file->f_path.dentry->d_inode) && 660 (file->f_mode & FMODE_WRITE)) 661 return -EACCES; 662 663 if (locks_verify_locked(file->f_path.dentry->d_inode)) 664 return -EAGAIN; 665 666 if (!(capabilities & BDI_CAP_MAP_DIRECT)) 667 return -ENODEV; 668 669 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || 670 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || 671 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) 672 ) { 673 printk("MAP_SHARED not completely supported on !MMU\n"); 674 return -EINVAL; 675 } 676 677 /* we mustn't privatise shared mappings */ 678 capabilities &= ~BDI_CAP_MAP_COPY; 679 } 680 else { 681 /* we're going to read the file into private memory we 682 * allocate */ 683 if (!(capabilities & BDI_CAP_MAP_COPY)) 684 return -ENODEV; 685 686 /* we don't permit a private writable mapping to be 687 * shared with the backing device */ 688 if (prot & PROT_WRITE) 689 capabilities &= ~BDI_CAP_MAP_DIRECT; 690 } 691 692 /* handle executable mappings and implied executable 693 * mappings */ 694 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 695 if (prot & PROT_EXEC) 696 return -EPERM; 697 } 698 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { 699 /* handle implication of PROT_EXEC by PROT_READ */ 700 if (current->personality & READ_IMPLIES_EXEC) { 701 if (capabilities & BDI_CAP_EXEC_MAP) 702 prot |= PROT_EXEC; 703 } 704 } 705 else if ((prot & PROT_READ) && 706 (prot & PROT_EXEC) && 707 !(capabilities & BDI_CAP_EXEC_MAP) 708 ) { 709 /* backing file is not executable, try to copy */ 710 capabilities &= ~BDI_CAP_MAP_DIRECT; 711 } 712 } 713 else { 714 /* anonymous mappings are always memory backed and can be 715 * privately mapped 716 */ 717 capabilities = BDI_CAP_MAP_COPY; 718 719 /* handle PROT_EXEC implication by PROT_READ */ 720 if ((prot & PROT_READ) && 721 (current->personality & READ_IMPLIES_EXEC)) 722 prot |= PROT_EXEC; 723 } 724 725 /* allow the security API to have its say */ 726 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0); 727 if (ret < 0) 728 return ret; 729 730 /* looks okay */ 731 *_capabilities = capabilities; 732 return 0; 733 } 734 735 /* 736 * we've determined that we can make the mapping, now translate what we 737 * now know into VMA flags 738 */ 739 static unsigned long determine_vm_flags(struct file *file, 740 unsigned long prot, 741 unsigned long flags, 742 unsigned long capabilities) 743 { 744 unsigned long vm_flags; 745 746 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); 747 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 748 /* vm_flags |= mm->def_flags; */ 749 750 if (!(capabilities & BDI_CAP_MAP_DIRECT)) { 751 /* attempt to share read-only copies of mapped file chunks */ 752 if (file && !(prot & PROT_WRITE)) 753 vm_flags |= VM_MAYSHARE; 754 } 755 else { 756 /* overlay a shareable mapping on the backing device or inode 757 * if possible - used for chardevs, ramfs/tmpfs/shmfs and 758 * romfs/cramfs */ 759 if (flags & MAP_SHARED) 760 vm_flags |= VM_MAYSHARE | VM_SHARED; 761 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0) 762 vm_flags |= VM_MAYSHARE; 763 } 764 765 /* refuse to let anyone share private mappings with this process if 766 * it's being traced - otherwise breakpoints set in it may interfere 767 * with another untraced process 768 */ 769 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current)) 770 vm_flags &= ~VM_MAYSHARE; 771 772 return vm_flags; 773 } 774 775 /* 776 * set up a shared mapping on a file 777 */ 778 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len) 779 { 780 int ret; 781 782 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 783 if (ret != -ENOSYS) 784 return ret; 785 786 /* getting an ENOSYS error indicates that direct mmap isn't 787 * possible (as opposed to tried but failed) so we'll fall 788 * through to making a private copy of the data and mapping 789 * that if we can */ 790 return -ENODEV; 791 } 792 793 /* 794 * set up a private mapping or an anonymous shared mapping 795 */ 796 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len) 797 { 798 void *base; 799 int ret; 800 801 /* invoke the file's mapping function so that it can keep track of 802 * shared mappings on devices or memory 803 * - VM_MAYSHARE will be set if it may attempt to share 804 */ 805 if (vma->vm_file) { 806 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 807 if (ret != -ENOSYS) { 808 /* shouldn't return success if we're not sharing */ 809 BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE)); 810 return ret; /* success or a real error */ 811 } 812 813 /* getting an ENOSYS error indicates that direct mmap isn't 814 * possible (as opposed to tried but failed) so we'll try to 815 * make a private copy of the data and map that instead */ 816 } 817 818 /* allocate some memory to hold the mapping 819 * - note that this may not return a page-aligned address if the object 820 * we're allocating is smaller than a page 821 */ 822 base = kmalloc(len, GFP_KERNEL|__GFP_COMP); 823 if (!base) 824 goto enomem; 825 826 vma->vm_start = (unsigned long) base; 827 vma->vm_end = vma->vm_start + len; 828 vma->vm_flags |= VM_MAPPED_COPY; 829 830 #ifdef WARN_ON_SLACK 831 if (len + WARN_ON_SLACK <= kobjsize(result)) 832 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n", 833 len, current->pid, kobjsize(result) - len); 834 #endif 835 836 if (vma->vm_file) { 837 /* read the contents of a file into the copy */ 838 mm_segment_t old_fs; 839 loff_t fpos; 840 841 fpos = vma->vm_pgoff; 842 fpos <<= PAGE_SHIFT; 843 844 old_fs = get_fs(); 845 set_fs(KERNEL_DS); 846 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos); 847 set_fs(old_fs); 848 849 if (ret < 0) 850 goto error_free; 851 852 /* clear the last little bit */ 853 if (ret < len) 854 memset(base + ret, 0, len - ret); 855 856 } else { 857 /* if it's an anonymous mapping, then just clear it */ 858 memset(base, 0, len); 859 } 860 861 return 0; 862 863 error_free: 864 kfree(base); 865 vma->vm_start = 0; 866 return ret; 867 868 enomem: 869 printk("Allocation of length %lu from process %d failed\n", 870 len, current->pid); 871 show_free_areas(); 872 return -ENOMEM; 873 } 874 875 /* 876 * handle mapping creation for uClinux 877 */ 878 unsigned long do_mmap_pgoff(struct file *file, 879 unsigned long addr, 880 unsigned long len, 881 unsigned long prot, 882 unsigned long flags, 883 unsigned long pgoff) 884 { 885 struct vm_list_struct *vml = NULL; 886 struct vm_area_struct *vma = NULL; 887 struct rb_node *rb; 888 unsigned long capabilities, vm_flags; 889 void *result; 890 int ret; 891 892 if (!(flags & MAP_FIXED)) 893 addr = round_hint_to_min(addr); 894 895 /* decide whether we should attempt the mapping, and if so what sort of 896 * mapping */ 897 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 898 &capabilities); 899 if (ret < 0) 900 return ret; 901 902 /* we've determined that we can make the mapping, now translate what we 903 * now know into VMA flags */ 904 vm_flags = determine_vm_flags(file, prot, flags, capabilities); 905 906 /* we're going to need to record the mapping if it works */ 907 vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL); 908 if (!vml) 909 goto error_getting_vml; 910 911 down_write(&nommu_vma_sem); 912 913 /* if we want to share, we need to check for VMAs created by other 914 * mmap() calls that overlap with our proposed mapping 915 * - we can only share with an exact match on most regular files 916 * - shared mappings on character devices and memory backed files are 917 * permitted to overlap inexactly as far as we are concerned for in 918 * these cases, sharing is handled in the driver or filesystem rather 919 * than here 920 */ 921 if (vm_flags & VM_MAYSHARE) { 922 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 923 unsigned long vmpglen; 924 925 /* suppress VMA sharing for shared regions */ 926 if (vm_flags & VM_SHARED && 927 capabilities & BDI_CAP_MAP_DIRECT) 928 goto dont_share_VMAs; 929 930 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) { 931 vma = rb_entry(rb, struct vm_area_struct, vm_rb); 932 933 if (!(vma->vm_flags & VM_MAYSHARE)) 934 continue; 935 936 /* search for overlapping mappings on the same file */ 937 if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode) 938 continue; 939 940 if (vma->vm_pgoff >= pgoff + pglen) 941 continue; 942 943 vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1; 944 vmpglen >>= PAGE_SHIFT; 945 if (pgoff >= vma->vm_pgoff + vmpglen) 946 continue; 947 948 /* handle inexactly overlapping matches between mappings */ 949 if (vma->vm_pgoff != pgoff || vmpglen != pglen) { 950 if (!(capabilities & BDI_CAP_MAP_DIRECT)) 951 goto sharing_violation; 952 continue; 953 } 954 955 /* we've found a VMA we can share */ 956 atomic_inc(&vma->vm_usage); 957 958 vml->vma = vma; 959 result = (void *) vma->vm_start; 960 goto shared; 961 } 962 963 dont_share_VMAs: 964 vma = NULL; 965 966 /* obtain the address at which to make a shared mapping 967 * - this is the hook for quasi-memory character devices to 968 * tell us the location of a shared mapping 969 */ 970 if (file && file->f_op->get_unmapped_area) { 971 addr = file->f_op->get_unmapped_area(file, addr, len, 972 pgoff, flags); 973 if (IS_ERR((void *) addr)) { 974 ret = addr; 975 if (ret != (unsigned long) -ENOSYS) 976 goto error; 977 978 /* the driver refused to tell us where to site 979 * the mapping so we'll have to attempt to copy 980 * it */ 981 ret = (unsigned long) -ENODEV; 982 if (!(capabilities & BDI_CAP_MAP_COPY)) 983 goto error; 984 985 capabilities &= ~BDI_CAP_MAP_DIRECT; 986 } 987 } 988 } 989 990 /* we're going to need a VMA struct as well */ 991 vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL); 992 if (!vma) 993 goto error_getting_vma; 994 995 INIT_LIST_HEAD(&vma->anon_vma_node); 996 atomic_set(&vma->vm_usage, 1); 997 if (file) { 998 get_file(file); 999 if (vm_flags & VM_EXECUTABLE) { 1000 added_exe_file_vma(current->mm); 1001 vma->vm_mm = current->mm; 1002 } 1003 } 1004 vma->vm_file = file; 1005 vma->vm_flags = vm_flags; 1006 vma->vm_start = addr; 1007 vma->vm_end = addr + len; 1008 vma->vm_pgoff = pgoff; 1009 1010 vml->vma = vma; 1011 1012 /* set up the mapping */ 1013 if (file && vma->vm_flags & VM_SHARED) 1014 ret = do_mmap_shared_file(vma, len); 1015 else 1016 ret = do_mmap_private(vma, len); 1017 if (ret < 0) 1018 goto error; 1019 1020 /* okay... we have a mapping; now we have to register it */ 1021 result = (void *) vma->vm_start; 1022 1023 if (vma->vm_flags & VM_MAPPED_COPY) { 1024 realalloc += kobjsize(result); 1025 askedalloc += len; 1026 } 1027 1028 realalloc += kobjsize(vma); 1029 askedalloc += sizeof(*vma); 1030 1031 current->mm->total_vm += len >> PAGE_SHIFT; 1032 1033 add_nommu_vma(vma); 1034 1035 shared: 1036 realalloc += kobjsize(vml); 1037 askedalloc += sizeof(*vml); 1038 1039 add_vma_to_mm(current->mm, vml); 1040 1041 up_write(&nommu_vma_sem); 1042 1043 if (prot & PROT_EXEC) 1044 flush_icache_range((unsigned long) result, 1045 (unsigned long) result + len); 1046 1047 #ifdef DEBUG 1048 printk("do_mmap:\n"); 1049 show_process_blocks(); 1050 #endif 1051 1052 return (unsigned long) result; 1053 1054 error: 1055 up_write(&nommu_vma_sem); 1056 kfree(vml); 1057 if (vma) { 1058 if (vma->vm_file) { 1059 fput(vma->vm_file); 1060 if (vma->vm_flags & VM_EXECUTABLE) 1061 removed_exe_file_vma(vma->vm_mm); 1062 } 1063 kfree(vma); 1064 } 1065 return ret; 1066 1067 sharing_violation: 1068 up_write(&nommu_vma_sem); 1069 printk("Attempt to share mismatched mappings\n"); 1070 kfree(vml); 1071 return -EINVAL; 1072 1073 error_getting_vma: 1074 up_write(&nommu_vma_sem); 1075 kfree(vml); 1076 printk("Allocation of vma for %lu byte allocation from process %d failed\n", 1077 len, current->pid); 1078 show_free_areas(); 1079 return -ENOMEM; 1080 1081 error_getting_vml: 1082 printk("Allocation of vml for %lu byte allocation from process %d failed\n", 1083 len, current->pid); 1084 show_free_areas(); 1085 return -ENOMEM; 1086 } 1087 EXPORT_SYMBOL(do_mmap_pgoff); 1088 1089 /* 1090 * handle mapping disposal for uClinux 1091 */ 1092 static void put_vma(struct mm_struct *mm, struct vm_area_struct *vma) 1093 { 1094 if (vma) { 1095 down_write(&nommu_vma_sem); 1096 1097 if (atomic_dec_and_test(&vma->vm_usage)) { 1098 delete_nommu_vma(vma); 1099 1100 if (vma->vm_ops && vma->vm_ops->close) 1101 vma->vm_ops->close(vma); 1102 1103 /* IO memory and memory shared directly out of the pagecache from 1104 * ramfs/tmpfs mustn't be released here */ 1105 if (vma->vm_flags & VM_MAPPED_COPY) { 1106 realalloc -= kobjsize((void *) vma->vm_start); 1107 askedalloc -= vma->vm_end - vma->vm_start; 1108 kfree((void *) vma->vm_start); 1109 } 1110 1111 realalloc -= kobjsize(vma); 1112 askedalloc -= sizeof(*vma); 1113 1114 if (vma->vm_file) { 1115 fput(vma->vm_file); 1116 if (vma->vm_flags & VM_EXECUTABLE) 1117 removed_exe_file_vma(mm); 1118 } 1119 kfree(vma); 1120 } 1121 1122 up_write(&nommu_vma_sem); 1123 } 1124 } 1125 1126 /* 1127 * release a mapping 1128 * - under NOMMU conditions the parameters must match exactly to the mapping to 1129 * be removed 1130 */ 1131 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len) 1132 { 1133 struct vm_list_struct *vml, **parent; 1134 unsigned long end = addr + len; 1135 1136 #ifdef DEBUG 1137 printk("do_munmap:\n"); 1138 #endif 1139 1140 for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) { 1141 if ((*parent)->vma->vm_start > addr) 1142 break; 1143 if ((*parent)->vma->vm_start == addr && 1144 ((len == 0) || ((*parent)->vma->vm_end == end))) 1145 goto found; 1146 } 1147 1148 printk("munmap of non-mmaped memory by process %d (%s): %p\n", 1149 current->pid, current->comm, (void *) addr); 1150 return -EINVAL; 1151 1152 found: 1153 vml = *parent; 1154 1155 put_vma(mm, vml->vma); 1156 1157 *parent = vml->next; 1158 realalloc -= kobjsize(vml); 1159 askedalloc -= sizeof(*vml); 1160 kfree(vml); 1161 1162 update_hiwater_vm(mm); 1163 mm->total_vm -= len >> PAGE_SHIFT; 1164 1165 #ifdef DEBUG 1166 show_process_blocks(); 1167 #endif 1168 1169 return 0; 1170 } 1171 EXPORT_SYMBOL(do_munmap); 1172 1173 asmlinkage long sys_munmap(unsigned long addr, size_t len) 1174 { 1175 int ret; 1176 struct mm_struct *mm = current->mm; 1177 1178 down_write(&mm->mmap_sem); 1179 ret = do_munmap(mm, addr, len); 1180 up_write(&mm->mmap_sem); 1181 return ret; 1182 } 1183 1184 /* 1185 * Release all mappings 1186 */ 1187 void exit_mmap(struct mm_struct * mm) 1188 { 1189 struct vm_list_struct *tmp; 1190 1191 if (mm) { 1192 #ifdef DEBUG 1193 printk("Exit_mmap:\n"); 1194 #endif 1195 1196 mm->total_vm = 0; 1197 1198 while ((tmp = mm->context.vmlist)) { 1199 mm->context.vmlist = tmp->next; 1200 put_vma(mm, tmp->vma); 1201 1202 realalloc -= kobjsize(tmp); 1203 askedalloc -= sizeof(*tmp); 1204 kfree(tmp); 1205 } 1206 1207 #ifdef DEBUG 1208 show_process_blocks(); 1209 #endif 1210 } 1211 } 1212 1213 unsigned long do_brk(unsigned long addr, unsigned long len) 1214 { 1215 return -ENOMEM; 1216 } 1217 1218 /* 1219 * expand (or shrink) an existing mapping, potentially moving it at the same 1220 * time (controlled by the MREMAP_MAYMOVE flag and available VM space) 1221 * 1222 * under NOMMU conditions, we only permit changing a mapping's size, and only 1223 * as long as it stays within the hole allocated by the kmalloc() call in 1224 * do_mmap_pgoff() and the block is not shareable 1225 * 1226 * MREMAP_FIXED is not supported under NOMMU conditions 1227 */ 1228 unsigned long do_mremap(unsigned long addr, 1229 unsigned long old_len, unsigned long new_len, 1230 unsigned long flags, unsigned long new_addr) 1231 { 1232 struct vm_area_struct *vma; 1233 1234 /* insanity checks first */ 1235 if (new_len == 0) 1236 return (unsigned long) -EINVAL; 1237 1238 if (flags & MREMAP_FIXED && new_addr != addr) 1239 return (unsigned long) -EINVAL; 1240 1241 vma = find_vma_exact(current->mm, addr); 1242 if (!vma) 1243 return (unsigned long) -EINVAL; 1244 1245 if (vma->vm_end != vma->vm_start + old_len) 1246 return (unsigned long) -EFAULT; 1247 1248 if (vma->vm_flags & VM_MAYSHARE) 1249 return (unsigned long) -EPERM; 1250 1251 if (new_len > kobjsize((void *) addr)) 1252 return (unsigned long) -ENOMEM; 1253 1254 /* all checks complete - do it */ 1255 vma->vm_end = vma->vm_start + new_len; 1256 1257 askedalloc -= old_len; 1258 askedalloc += new_len; 1259 1260 return vma->vm_start; 1261 } 1262 EXPORT_SYMBOL(do_mremap); 1263 1264 asmlinkage unsigned long sys_mremap(unsigned long addr, 1265 unsigned long old_len, unsigned long new_len, 1266 unsigned long flags, unsigned long new_addr) 1267 { 1268 unsigned long ret; 1269 1270 down_write(¤t->mm->mmap_sem); 1271 ret = do_mremap(addr, old_len, new_len, flags, new_addr); 1272 up_write(¤t->mm->mmap_sem); 1273 return ret; 1274 } 1275 1276 struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 1277 unsigned int foll_flags) 1278 { 1279 return NULL; 1280 } 1281 1282 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from, 1283 unsigned long to, unsigned long size, pgprot_t prot) 1284 { 1285 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT; 1286 return 0; 1287 } 1288 EXPORT_SYMBOL(remap_pfn_range); 1289 1290 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, 1291 unsigned long pgoff) 1292 { 1293 unsigned int size = vma->vm_end - vma->vm_start; 1294 1295 if (!(vma->vm_flags & VM_USERMAP)) 1296 return -EINVAL; 1297 1298 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); 1299 vma->vm_end = vma->vm_start + size; 1300 1301 return 0; 1302 } 1303 EXPORT_SYMBOL(remap_vmalloc_range); 1304 1305 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) 1306 { 1307 } 1308 1309 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, 1310 unsigned long len, unsigned long pgoff, unsigned long flags) 1311 { 1312 return -ENOMEM; 1313 } 1314 1315 void arch_unmap_area(struct mm_struct *mm, unsigned long addr) 1316 { 1317 } 1318 1319 void unmap_mapping_range(struct address_space *mapping, 1320 loff_t const holebegin, loff_t const holelen, 1321 int even_cows) 1322 { 1323 } 1324 EXPORT_SYMBOL(unmap_mapping_range); 1325 1326 /* 1327 * ask for an unmapped area at which to create a mapping on a file 1328 */ 1329 unsigned long get_unmapped_area(struct file *file, unsigned long addr, 1330 unsigned long len, unsigned long pgoff, 1331 unsigned long flags) 1332 { 1333 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, 1334 unsigned long, unsigned long); 1335 1336 get_area = current->mm->get_unmapped_area; 1337 if (file && file->f_op && file->f_op->get_unmapped_area) 1338 get_area = file->f_op->get_unmapped_area; 1339 1340 if (!get_area) 1341 return -ENOSYS; 1342 1343 return get_area(file, addr, len, pgoff, flags); 1344 } 1345 EXPORT_SYMBOL(get_unmapped_area); 1346 1347 /* 1348 * Check that a process has enough memory to allocate a new virtual 1349 * mapping. 0 means there is enough memory for the allocation to 1350 * succeed and -ENOMEM implies there is not. 1351 * 1352 * We currently support three overcommit policies, which are set via the 1353 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1354 * 1355 * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1356 * Additional code 2002 Jul 20 by Robert Love. 1357 * 1358 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1359 * 1360 * Note this is a helper function intended to be used by LSMs which 1361 * wish to use this logic. 1362 */ 1363 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1364 { 1365 unsigned long free, allowed; 1366 1367 vm_acct_memory(pages); 1368 1369 /* 1370 * Sometimes we want to use more memory than we have 1371 */ 1372 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1373 return 0; 1374 1375 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1376 unsigned long n; 1377 1378 free = global_page_state(NR_FILE_PAGES); 1379 free += nr_swap_pages; 1380 1381 /* 1382 * Any slabs which are created with the 1383 * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1384 * which are reclaimable, under pressure. The dentry 1385 * cache and most inode caches should fall into this 1386 */ 1387 free += global_page_state(NR_SLAB_RECLAIMABLE); 1388 1389 /* 1390 * Leave the last 3% for root 1391 */ 1392 if (!cap_sys_admin) 1393 free -= free / 32; 1394 1395 if (free > pages) 1396 return 0; 1397 1398 /* 1399 * nr_free_pages() is very expensive on large systems, 1400 * only call if we're about to fail. 1401 */ 1402 n = nr_free_pages(); 1403 1404 /* 1405 * Leave reserved pages. The pages are not for anonymous pages. 1406 */ 1407 if (n <= totalreserve_pages) 1408 goto error; 1409 else 1410 n -= totalreserve_pages; 1411 1412 /* 1413 * Leave the last 3% for root 1414 */ 1415 if (!cap_sys_admin) 1416 n -= n / 32; 1417 free += n; 1418 1419 if (free > pages) 1420 return 0; 1421 1422 goto error; 1423 } 1424 1425 allowed = totalram_pages * sysctl_overcommit_ratio / 100; 1426 /* 1427 * Leave the last 3% for root 1428 */ 1429 if (!cap_sys_admin) 1430 allowed -= allowed / 32; 1431 allowed += total_swap_pages; 1432 1433 /* Don't let a single process grow too big: 1434 leave 3% of the size of this process for other processes */ 1435 allowed -= current->mm->total_vm / 32; 1436 1437 /* 1438 * cast `allowed' as a signed long because vm_committed_space 1439 * sometimes has a negative value 1440 */ 1441 if (atomic_long_read(&vm_committed_space) < (long)allowed) 1442 return 0; 1443 error: 1444 vm_unacct_memory(pages); 1445 1446 return -ENOMEM; 1447 } 1448 1449 int in_gate_area_no_task(unsigned long addr) 1450 { 1451 return 0; 1452 } 1453 1454 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1455 { 1456 BUG(); 1457 return 0; 1458 } 1459 EXPORT_SYMBOL(filemap_fault); 1460 1461 /* 1462 * Access another process' address space. 1463 * - source/target buffer must be kernel space 1464 */ 1465 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) 1466 { 1467 struct vm_area_struct *vma; 1468 struct mm_struct *mm; 1469 1470 if (addr + len < addr) 1471 return 0; 1472 1473 mm = get_task_mm(tsk); 1474 if (!mm) 1475 return 0; 1476 1477 down_read(&mm->mmap_sem); 1478 1479 /* the access must start within one of the target process's mappings */ 1480 vma = find_vma(mm, addr); 1481 if (vma) { 1482 /* don't overrun this mapping */ 1483 if (addr + len >= vma->vm_end) 1484 len = vma->vm_end - addr; 1485 1486 /* only read or write mappings where it is permitted */ 1487 if (write && vma->vm_flags & VM_MAYWRITE) 1488 len -= copy_to_user((void *) addr, buf, len); 1489 else if (!write && vma->vm_flags & VM_MAYREAD) 1490 len -= copy_from_user(buf, (void *) addr, len); 1491 else 1492 len = 0; 1493 } else { 1494 len = 0; 1495 } 1496 1497 up_read(&mm->mmap_sem); 1498 mmput(mm); 1499 return len; 1500 } 1501