nommu.c revision f2b8544f5f50073fcc705e16b45a6821d50eb080
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 */ 14 15#include <linux/module.h> 16#include <linux/mm.h> 17#include <linux/mman.h> 18#include <linux/swap.h> 19#include <linux/file.h> 20#include <linux/highmem.h> 21#include <linux/pagemap.h> 22#include <linux/slab.h> 23#include <linux/vmalloc.h> 24#include <linux/ptrace.h> 25#include <linux/blkdev.h> 26#include <linux/backing-dev.h> 27#include <linux/mount.h> 28#include <linux/personality.h> 29#include <linux/security.h> 30#include <linux/syscalls.h> 31 32#include <asm/uaccess.h> 33#include <asm/tlb.h> 34#include <asm/tlbflush.h> 35 36void *high_memory; 37struct page *mem_map; 38unsigned long max_mapnr; 39unsigned long num_physpages; 40unsigned long askedalloc, realalloc; 41atomic_t vm_committed_space = ATOMIC_INIT(0); 42int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 43int sysctl_overcommit_ratio = 50; /* default is 50% */ 44int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; 45int heap_stack_gap = 0; 46 47EXPORT_SYMBOL(mem_map); 48EXPORT_SYMBOL(num_physpages); 49 50/* list of shareable VMAs */ 51struct rb_root nommu_vma_tree = RB_ROOT; 52DECLARE_RWSEM(nommu_vma_sem); 53 54struct vm_operations_struct generic_file_vm_ops = { 55}; 56 57/* 58 * Handle all mappings that got truncated by a "truncate()" 59 * system call. 60 * 61 * NOTE! We have to be ready to update the memory sharing 62 * between the file and the memory map for a potential last 63 * incomplete page. Ugly, but necessary. 64 */ 65int vmtruncate(struct inode *inode, loff_t offset) 66{ 67 struct address_space *mapping = inode->i_mapping; 68 unsigned long limit; 69 70 if (inode->i_size < offset) 71 goto do_expand; 72 i_size_write(inode, offset); 73 74 truncate_inode_pages(mapping, offset); 75 goto out_truncate; 76 77do_expand: 78 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; 79 if (limit != RLIM_INFINITY && offset > limit) 80 goto out_sig; 81 if (offset > inode->i_sb->s_maxbytes) 82 goto out; 83 i_size_write(inode, offset); 84 85out_truncate: 86 if (inode->i_op && inode->i_op->truncate) 87 inode->i_op->truncate(inode); 88 return 0; 89out_sig: 90 send_sig(SIGXFSZ, current, 0); 91out: 92 return -EFBIG; 93} 94 95EXPORT_SYMBOL(vmtruncate); 96 97/* 98 * Return the total memory allocated for this pointer, not 99 * just what the caller asked for. 100 * 101 * Doesn't have to be accurate, i.e. may have races. 102 */ 103unsigned int kobjsize(const void *objp) 104{ 105 struct page *page; 106 107 if (!objp || !((page = virt_to_page(objp)))) 108 return 0; 109 110 if (PageSlab(page)) 111 return ksize(objp); 112 113 BUG_ON(page->index < 0); 114 BUG_ON(page->index >= MAX_ORDER); 115 116 return (PAGE_SIZE << page->index); 117} 118 119/* 120 * get a list of pages in an address range belonging to the specified process 121 * and indicate the VMA that covers each page 122 * - this is potentially dodgy as we may end incrementing the page count of a 123 * slab page or a secondary page from a compound page 124 * - don't permit access to VMAs that don't support it, such as I/O mappings 125 */ 126int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 127 unsigned long start, int len, int write, int force, 128 struct page **pages, struct vm_area_struct **vmas) 129{ 130 struct vm_area_struct *vma; 131 unsigned long vm_flags; 132 int i; 133 134 /* calculate required read or write permissions. 135 * - if 'force' is set, we only require the "MAY" flags. 136 */ 137 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 138 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 139 140 for (i = 0; i < len; i++) { 141 vma = find_vma(mm, start); 142 if (!vma) 143 goto finish_or_fault; 144 145 /* protect what we can, including chardevs */ 146 if (vma->vm_flags & (VM_IO | VM_PFNMAP) || 147 !(vm_flags & vma->vm_flags)) 148 goto finish_or_fault; 149 150 if (pages) { 151 pages[i] = virt_to_page(start); 152 if (pages[i]) 153 page_cache_get(pages[i]); 154 } 155 if (vmas) 156 vmas[i] = vma; 157 start += PAGE_SIZE; 158 } 159 160 return i; 161 162finish_or_fault: 163 return i ? : -EFAULT; 164} 165EXPORT_SYMBOL(get_user_pages); 166 167DEFINE_RWLOCK(vmlist_lock); 168struct vm_struct *vmlist; 169 170void vfree(void *addr) 171{ 172 kfree(addr); 173} 174EXPORT_SYMBOL(vfree); 175 176void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 177{ 178 /* 179 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 180 * returns only a logical address. 181 */ 182 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 183} 184EXPORT_SYMBOL(__vmalloc); 185 186struct page * vmalloc_to_page(void *addr) 187{ 188 return virt_to_page(addr); 189} 190EXPORT_SYMBOL(vmalloc_to_page); 191 192unsigned long vmalloc_to_pfn(void *addr) 193{ 194 return page_to_pfn(virt_to_page(addr)); 195} 196EXPORT_SYMBOL(vmalloc_to_pfn); 197 198long vread(char *buf, char *addr, unsigned long count) 199{ 200 memcpy(buf, addr, count); 201 return count; 202} 203 204long vwrite(char *buf, char *addr, unsigned long count) 205{ 206 /* Don't allow overflow */ 207 if ((unsigned long) addr + count < count) 208 count = -(unsigned long) addr; 209 210 memcpy(addr, buf, count); 211 return(count); 212} 213 214/* 215 * vmalloc - allocate virtually continguos memory 216 * 217 * @size: allocation size 218 * 219 * Allocate enough pages to cover @size from the page level 220 * allocator and map them into continguos kernel virtual space. 221 * 222 * For tight control over page level allocator and protection flags 223 * use __vmalloc() instead. 224 */ 225void *vmalloc(unsigned long size) 226{ 227 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 228} 229EXPORT_SYMBOL(vmalloc); 230 231void *vmalloc_node(unsigned long size, int node) 232{ 233 return vmalloc(size); 234} 235EXPORT_SYMBOL(vmalloc_node); 236 237/** 238 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 239 * @size: allocation size 240 * 241 * Allocate enough 32bit PA addressable pages to cover @size from the 242 * page level allocator and map them into continguos kernel virtual space. 243 */ 244void *vmalloc_32(unsigned long size) 245{ 246 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 247} 248EXPORT_SYMBOL(vmalloc_32); 249 250/** 251 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 252 * @size: allocation size 253 * 254 * The resulting memory area is 32bit addressable and zeroed so it can be 255 * mapped to userspace without leaking data. 256 */ 257void *vmalloc_32_user(unsigned long size) 258{ 259 return __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL); 260} 261EXPORT_SYMBOL(vmalloc_32_user); 262 263void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 264{ 265 BUG(); 266 return NULL; 267} 268EXPORT_SYMBOL(vmap); 269 270void vunmap(void *addr) 271{ 272 BUG(); 273} 274EXPORT_SYMBOL(vunmap); 275 276/* 277 * Implement a stub for vmalloc_sync_all() if the architecture chose not to 278 * have one. 279 */ 280void __attribute__((weak)) vmalloc_sync_all(void) 281{ 282} 283 284int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 285 struct page *page) 286{ 287 return -EINVAL; 288} 289EXPORT_SYMBOL(vm_insert_page); 290 291/* 292 * sys_brk() for the most part doesn't need the global kernel 293 * lock, except when an application is doing something nasty 294 * like trying to un-brk an area that has already been mapped 295 * to a regular file. in this case, the unmapping will need 296 * to invoke file system routines that need the global lock. 297 */ 298asmlinkage unsigned long sys_brk(unsigned long brk) 299{ 300 struct mm_struct *mm = current->mm; 301 302 if (brk < mm->start_brk || brk > mm->context.end_brk) 303 return mm->brk; 304 305 if (mm->brk == brk) 306 return mm->brk; 307 308 /* 309 * Always allow shrinking brk 310 */ 311 if (brk <= mm->brk) { 312 mm->brk = brk; 313 return brk; 314 } 315 316 /* 317 * Ok, looks good - let it rip. 318 */ 319 return mm->brk = brk; 320} 321 322#ifdef DEBUG 323static void show_process_blocks(void) 324{ 325 struct vm_list_struct *vml; 326 327 printk("Process blocks %d:", current->pid); 328 329 for (vml = ¤t->mm->context.vmlist; vml; vml = vml->next) { 330 printk(" %p: %p", vml, vml->vma); 331 if (vml->vma) 332 printk(" (%d @%lx #%d)", 333 kobjsize((void *) vml->vma->vm_start), 334 vml->vma->vm_start, 335 atomic_read(&vml->vma->vm_usage)); 336 printk(vml->next ? " ->" : ".\n"); 337 } 338} 339#endif /* DEBUG */ 340 341/* 342 * add a VMA into a process's mm_struct in the appropriate place in the list 343 * - should be called with mm->mmap_sem held writelocked 344 */ 345static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml) 346{ 347 struct vm_list_struct **ppv; 348 349 for (ppv = ¤t->mm->context.vmlist; *ppv; ppv = &(*ppv)->next) 350 if ((*ppv)->vma->vm_start > vml->vma->vm_start) 351 break; 352 353 vml->next = *ppv; 354 *ppv = vml; 355} 356 357/* 358 * look up the first VMA in which addr resides, NULL if none 359 * - should be called with mm->mmap_sem at least held readlocked 360 */ 361struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 362{ 363 struct vm_list_struct *loop, *vml; 364 365 /* search the vm_start ordered list */ 366 vml = NULL; 367 for (loop = mm->context.vmlist; loop; loop = loop->next) { 368 if (loop->vma->vm_start > addr) 369 break; 370 vml = loop; 371 } 372 373 if (vml && vml->vma->vm_end > addr) 374 return vml->vma; 375 376 return NULL; 377} 378EXPORT_SYMBOL(find_vma); 379 380/* 381 * find a VMA 382 * - we don't extend stack VMAs under NOMMU conditions 383 */ 384struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) 385{ 386 return find_vma(mm, addr); 387} 388 389int expand_stack(struct vm_area_struct *vma, unsigned long address) 390{ 391 return -ENOMEM; 392} 393 394/* 395 * look up the first VMA exactly that exactly matches addr 396 * - should be called with mm->mmap_sem at least held readlocked 397 */ 398static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm, 399 unsigned long addr) 400{ 401 struct vm_list_struct *vml; 402 403 /* search the vm_start ordered list */ 404 for (vml = mm->context.vmlist; vml; vml = vml->next) { 405 if (vml->vma->vm_start == addr) 406 return vml->vma; 407 if (vml->vma->vm_start > addr) 408 break; 409 } 410 411 return NULL; 412} 413 414/* 415 * find a VMA in the global tree 416 */ 417static inline struct vm_area_struct *find_nommu_vma(unsigned long start) 418{ 419 struct vm_area_struct *vma; 420 struct rb_node *n = nommu_vma_tree.rb_node; 421 422 while (n) { 423 vma = rb_entry(n, struct vm_area_struct, vm_rb); 424 425 if (start < vma->vm_start) 426 n = n->rb_left; 427 else if (start > vma->vm_start) 428 n = n->rb_right; 429 else 430 return vma; 431 } 432 433 return NULL; 434} 435 436/* 437 * add a VMA in the global tree 438 */ 439static void add_nommu_vma(struct vm_area_struct *vma) 440{ 441 struct vm_area_struct *pvma; 442 struct address_space *mapping; 443 struct rb_node **p = &nommu_vma_tree.rb_node; 444 struct rb_node *parent = NULL; 445 446 /* add the VMA to the mapping */ 447 if (vma->vm_file) { 448 mapping = vma->vm_file->f_mapping; 449 450 flush_dcache_mmap_lock(mapping); 451 vma_prio_tree_insert(vma, &mapping->i_mmap); 452 flush_dcache_mmap_unlock(mapping); 453 } 454 455 /* add the VMA to the master list */ 456 while (*p) { 457 parent = *p; 458 pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 459 460 if (vma->vm_start < pvma->vm_start) { 461 p = &(*p)->rb_left; 462 } 463 else if (vma->vm_start > pvma->vm_start) { 464 p = &(*p)->rb_right; 465 } 466 else { 467 /* mappings are at the same address - this can only 468 * happen for shared-mem chardevs and shared file 469 * mappings backed by ramfs/tmpfs */ 470 BUG_ON(!(pvma->vm_flags & VM_SHARED)); 471 472 if (vma < pvma) 473 p = &(*p)->rb_left; 474 else if (vma > pvma) 475 p = &(*p)->rb_right; 476 else 477 BUG(); 478 } 479 } 480 481 rb_link_node(&vma->vm_rb, parent, p); 482 rb_insert_color(&vma->vm_rb, &nommu_vma_tree); 483} 484 485/* 486 * delete a VMA from the global list 487 */ 488static void delete_nommu_vma(struct vm_area_struct *vma) 489{ 490 struct address_space *mapping; 491 492 /* remove the VMA from the mapping */ 493 if (vma->vm_file) { 494 mapping = vma->vm_file->f_mapping; 495 496 flush_dcache_mmap_lock(mapping); 497 vma_prio_tree_remove(vma, &mapping->i_mmap); 498 flush_dcache_mmap_unlock(mapping); 499 } 500 501 /* remove from the master list */ 502 rb_erase(&vma->vm_rb, &nommu_vma_tree); 503} 504 505/* 506 * determine whether a mapping should be permitted and, if so, what sort of 507 * mapping we're capable of supporting 508 */ 509static int validate_mmap_request(struct file *file, 510 unsigned long addr, 511 unsigned long len, 512 unsigned long prot, 513 unsigned long flags, 514 unsigned long pgoff, 515 unsigned long *_capabilities) 516{ 517 unsigned long capabilities; 518 unsigned long reqprot = prot; 519 int ret; 520 521 /* do the simple checks first */ 522 if (flags & MAP_FIXED || addr) { 523 printk(KERN_DEBUG 524 "%d: Can't do fixed-address/overlay mmap of RAM\n", 525 current->pid); 526 return -EINVAL; 527 } 528 529 if ((flags & MAP_TYPE) != MAP_PRIVATE && 530 (flags & MAP_TYPE) != MAP_SHARED) 531 return -EINVAL; 532 533 if (!len) 534 return -EINVAL; 535 536 /* Careful about overflows.. */ 537 len = PAGE_ALIGN(len); 538 if (!len || len > TASK_SIZE) 539 return -ENOMEM; 540 541 /* offset overflow? */ 542 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 543 return -EOVERFLOW; 544 545 if (file) { 546 /* validate file mapping requests */ 547 struct address_space *mapping; 548 549 /* files must support mmap */ 550 if (!file->f_op || !file->f_op->mmap) 551 return -ENODEV; 552 553 /* work out if what we've got could possibly be shared 554 * - we support chardevs that provide their own "memory" 555 * - we support files/blockdevs that are memory backed 556 */ 557 mapping = file->f_mapping; 558 if (!mapping) 559 mapping = file->f_path.dentry->d_inode->i_mapping; 560 561 capabilities = 0; 562 if (mapping && mapping->backing_dev_info) 563 capabilities = mapping->backing_dev_info->capabilities; 564 565 if (!capabilities) { 566 /* no explicit capabilities set, so assume some 567 * defaults */ 568 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) { 569 case S_IFREG: 570 case S_IFBLK: 571 capabilities = BDI_CAP_MAP_COPY; 572 break; 573 574 case S_IFCHR: 575 capabilities = 576 BDI_CAP_MAP_DIRECT | 577 BDI_CAP_READ_MAP | 578 BDI_CAP_WRITE_MAP; 579 break; 580 581 default: 582 return -EINVAL; 583 } 584 } 585 586 /* eliminate any capabilities that we can't support on this 587 * device */ 588 if (!file->f_op->get_unmapped_area) 589 capabilities &= ~BDI_CAP_MAP_DIRECT; 590 if (!file->f_op->read) 591 capabilities &= ~BDI_CAP_MAP_COPY; 592 593 if (flags & MAP_SHARED) { 594 /* do checks for writing, appending and locking */ 595 if ((prot & PROT_WRITE) && 596 !(file->f_mode & FMODE_WRITE)) 597 return -EACCES; 598 599 if (IS_APPEND(file->f_path.dentry->d_inode) && 600 (file->f_mode & FMODE_WRITE)) 601 return -EACCES; 602 603 if (locks_verify_locked(file->f_path.dentry->d_inode)) 604 return -EAGAIN; 605 606 if (!(capabilities & BDI_CAP_MAP_DIRECT)) 607 return -ENODEV; 608 609 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || 610 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || 611 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) 612 ) { 613 printk("MAP_SHARED not completely supported on !MMU\n"); 614 return -EINVAL; 615 } 616 617 /* we mustn't privatise shared mappings */ 618 capabilities &= ~BDI_CAP_MAP_COPY; 619 } 620 else { 621 /* we're going to read the file into private memory we 622 * allocate */ 623 if (!(capabilities & BDI_CAP_MAP_COPY)) 624 return -ENODEV; 625 626 /* we don't permit a private writable mapping to be 627 * shared with the backing device */ 628 if (prot & PROT_WRITE) 629 capabilities &= ~BDI_CAP_MAP_DIRECT; 630 } 631 632 /* handle executable mappings and implied executable 633 * mappings */ 634 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 635 if (prot & PROT_EXEC) 636 return -EPERM; 637 } 638 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { 639 /* handle implication of PROT_EXEC by PROT_READ */ 640 if (current->personality & READ_IMPLIES_EXEC) { 641 if (capabilities & BDI_CAP_EXEC_MAP) 642 prot |= PROT_EXEC; 643 } 644 } 645 else if ((prot & PROT_READ) && 646 (prot & PROT_EXEC) && 647 !(capabilities & BDI_CAP_EXEC_MAP) 648 ) { 649 /* backing file is not executable, try to copy */ 650 capabilities &= ~BDI_CAP_MAP_DIRECT; 651 } 652 } 653 else { 654 /* anonymous mappings are always memory backed and can be 655 * privately mapped 656 */ 657 capabilities = BDI_CAP_MAP_COPY; 658 659 /* handle PROT_EXEC implication by PROT_READ */ 660 if ((prot & PROT_READ) && 661 (current->personality & READ_IMPLIES_EXEC)) 662 prot |= PROT_EXEC; 663 } 664 665 /* allow the security API to have its say */ 666 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0); 667 if (ret < 0) 668 return ret; 669 670 /* looks okay */ 671 *_capabilities = capabilities; 672 return 0; 673} 674 675/* 676 * we've determined that we can make the mapping, now translate what we 677 * now know into VMA flags 678 */ 679static unsigned long determine_vm_flags(struct file *file, 680 unsigned long prot, 681 unsigned long flags, 682 unsigned long capabilities) 683{ 684 unsigned long vm_flags; 685 686 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); 687 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 688 /* vm_flags |= mm->def_flags; */ 689 690 if (!(capabilities & BDI_CAP_MAP_DIRECT)) { 691 /* attempt to share read-only copies of mapped file chunks */ 692 if (file && !(prot & PROT_WRITE)) 693 vm_flags |= VM_MAYSHARE; 694 } 695 else { 696 /* overlay a shareable mapping on the backing device or inode 697 * if possible - used for chardevs, ramfs/tmpfs/shmfs and 698 * romfs/cramfs */ 699 if (flags & MAP_SHARED) 700 vm_flags |= VM_MAYSHARE | VM_SHARED; 701 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0) 702 vm_flags |= VM_MAYSHARE; 703 } 704 705 /* refuse to let anyone share private mappings with this process if 706 * it's being traced - otherwise breakpoints set in it may interfere 707 * with another untraced process 708 */ 709 if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED)) 710 vm_flags &= ~VM_MAYSHARE; 711 712 return vm_flags; 713} 714 715/* 716 * set up a shared mapping on a file 717 */ 718static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len) 719{ 720 int ret; 721 722 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 723 if (ret != -ENOSYS) 724 return ret; 725 726 /* getting an ENOSYS error indicates that direct mmap isn't 727 * possible (as opposed to tried but failed) so we'll fall 728 * through to making a private copy of the data and mapping 729 * that if we can */ 730 return -ENODEV; 731} 732 733/* 734 * set up a private mapping or an anonymous shared mapping 735 */ 736static int do_mmap_private(struct vm_area_struct *vma, unsigned long len) 737{ 738 void *base; 739 int ret; 740 741 /* invoke the file's mapping function so that it can keep track of 742 * shared mappings on devices or memory 743 * - VM_MAYSHARE will be set if it may attempt to share 744 */ 745 if (vma->vm_file) { 746 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 747 if (ret != -ENOSYS) { 748 /* shouldn't return success if we're not sharing */ 749 BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE)); 750 return ret; /* success or a real error */ 751 } 752 753 /* getting an ENOSYS error indicates that direct mmap isn't 754 * possible (as opposed to tried but failed) so we'll try to 755 * make a private copy of the data and map that instead */ 756 } 757 758 /* allocate some memory to hold the mapping 759 * - note that this may not return a page-aligned address if the object 760 * we're allocating is smaller than a page 761 */ 762 base = kmalloc(len, GFP_KERNEL|__GFP_COMP); 763 if (!base) 764 goto enomem; 765 766 vma->vm_start = (unsigned long) base; 767 vma->vm_end = vma->vm_start + len; 768 vma->vm_flags |= VM_MAPPED_COPY; 769 770#ifdef WARN_ON_SLACK 771 if (len + WARN_ON_SLACK <= kobjsize(result)) 772 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n", 773 len, current->pid, kobjsize(result) - len); 774#endif 775 776 if (vma->vm_file) { 777 /* read the contents of a file into the copy */ 778 mm_segment_t old_fs; 779 loff_t fpos; 780 781 fpos = vma->vm_pgoff; 782 fpos <<= PAGE_SHIFT; 783 784 old_fs = get_fs(); 785 set_fs(KERNEL_DS); 786 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos); 787 set_fs(old_fs); 788 789 if (ret < 0) 790 goto error_free; 791 792 /* clear the last little bit */ 793 if (ret < len) 794 memset(base + ret, 0, len - ret); 795 796 } else { 797 /* if it's an anonymous mapping, then just clear it */ 798 memset(base, 0, len); 799 } 800 801 return 0; 802 803error_free: 804 kfree(base); 805 vma->vm_start = 0; 806 return ret; 807 808enomem: 809 printk("Allocation of length %lu from process %d failed\n", 810 len, current->pid); 811 show_free_areas(); 812 return -ENOMEM; 813} 814 815/* 816 * handle mapping creation for uClinux 817 */ 818unsigned long do_mmap_pgoff(struct file *file, 819 unsigned long addr, 820 unsigned long len, 821 unsigned long prot, 822 unsigned long flags, 823 unsigned long pgoff) 824{ 825 struct vm_list_struct *vml = NULL; 826 struct vm_area_struct *vma = NULL; 827 struct rb_node *rb; 828 unsigned long capabilities, vm_flags; 829 void *result; 830 int ret; 831 832 /* decide whether we should attempt the mapping, and if so what sort of 833 * mapping */ 834 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 835 &capabilities); 836 if (ret < 0) 837 return ret; 838 839 /* we've determined that we can make the mapping, now translate what we 840 * now know into VMA flags */ 841 vm_flags = determine_vm_flags(file, prot, flags, capabilities); 842 843 /* we're going to need to record the mapping if it works */ 844 vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL); 845 if (!vml) 846 goto error_getting_vml; 847 848 down_write(&nommu_vma_sem); 849 850 /* if we want to share, we need to check for VMAs created by other 851 * mmap() calls that overlap with our proposed mapping 852 * - we can only share with an exact match on most regular files 853 * - shared mappings on character devices and memory backed files are 854 * permitted to overlap inexactly as far as we are concerned for in 855 * these cases, sharing is handled in the driver or filesystem rather 856 * than here 857 */ 858 if (vm_flags & VM_MAYSHARE) { 859 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 860 unsigned long vmpglen; 861 862 /* suppress VMA sharing for shared regions */ 863 if (vm_flags & VM_SHARED && 864 capabilities & BDI_CAP_MAP_DIRECT) 865 goto dont_share_VMAs; 866 867 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) { 868 vma = rb_entry(rb, struct vm_area_struct, vm_rb); 869 870 if (!(vma->vm_flags & VM_MAYSHARE)) 871 continue; 872 873 /* search for overlapping mappings on the same file */ 874 if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode) 875 continue; 876 877 if (vma->vm_pgoff >= pgoff + pglen) 878 continue; 879 880 vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1; 881 vmpglen >>= PAGE_SHIFT; 882 if (pgoff >= vma->vm_pgoff + vmpglen) 883 continue; 884 885 /* handle inexactly overlapping matches between mappings */ 886 if (vma->vm_pgoff != pgoff || vmpglen != pglen) { 887 if (!(capabilities & BDI_CAP_MAP_DIRECT)) 888 goto sharing_violation; 889 continue; 890 } 891 892 /* we've found a VMA we can share */ 893 atomic_inc(&vma->vm_usage); 894 895 vml->vma = vma; 896 result = (void *) vma->vm_start; 897 goto shared; 898 } 899 900 dont_share_VMAs: 901 vma = NULL; 902 903 /* obtain the address at which to make a shared mapping 904 * - this is the hook for quasi-memory character devices to 905 * tell us the location of a shared mapping 906 */ 907 if (file && file->f_op->get_unmapped_area) { 908 addr = file->f_op->get_unmapped_area(file, addr, len, 909 pgoff, flags); 910 if (IS_ERR((void *) addr)) { 911 ret = addr; 912 if (ret != (unsigned long) -ENOSYS) 913 goto error; 914 915 /* the driver refused to tell us where to site 916 * the mapping so we'll have to attempt to copy 917 * it */ 918 ret = (unsigned long) -ENODEV; 919 if (!(capabilities & BDI_CAP_MAP_COPY)) 920 goto error; 921 922 capabilities &= ~BDI_CAP_MAP_DIRECT; 923 } 924 } 925 } 926 927 /* we're going to need a VMA struct as well */ 928 vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL); 929 if (!vma) 930 goto error_getting_vma; 931 932 INIT_LIST_HEAD(&vma->anon_vma_node); 933 atomic_set(&vma->vm_usage, 1); 934 if (file) 935 get_file(file); 936 vma->vm_file = file; 937 vma->vm_flags = vm_flags; 938 vma->vm_start = addr; 939 vma->vm_end = addr + len; 940 vma->vm_pgoff = pgoff; 941 942 vml->vma = vma; 943 944 /* set up the mapping */ 945 if (file && vma->vm_flags & VM_SHARED) 946 ret = do_mmap_shared_file(vma, len); 947 else 948 ret = do_mmap_private(vma, len); 949 if (ret < 0) 950 goto error; 951 952 /* okay... we have a mapping; now we have to register it */ 953 result = (void *) vma->vm_start; 954 955 if (vma->vm_flags & VM_MAPPED_COPY) { 956 realalloc += kobjsize(result); 957 askedalloc += len; 958 } 959 960 realalloc += kobjsize(vma); 961 askedalloc += sizeof(*vma); 962 963 current->mm->total_vm += len >> PAGE_SHIFT; 964 965 add_nommu_vma(vma); 966 967 shared: 968 realalloc += kobjsize(vml); 969 askedalloc += sizeof(*vml); 970 971 add_vma_to_mm(current->mm, vml); 972 973 up_write(&nommu_vma_sem); 974 975 if (prot & PROT_EXEC) 976 flush_icache_range((unsigned long) result, 977 (unsigned long) result + len); 978 979#ifdef DEBUG 980 printk("do_mmap:\n"); 981 show_process_blocks(); 982#endif 983 984 return (unsigned long) result; 985 986 error: 987 up_write(&nommu_vma_sem); 988 kfree(vml); 989 if (vma) { 990 if (vma->vm_file) 991 fput(vma->vm_file); 992 kfree(vma); 993 } 994 return ret; 995 996 sharing_violation: 997 up_write(&nommu_vma_sem); 998 printk("Attempt to share mismatched mappings\n"); 999 kfree(vml); 1000 return -EINVAL; 1001 1002 error_getting_vma: 1003 up_write(&nommu_vma_sem); 1004 kfree(vml); 1005 printk("Allocation of vma for %lu byte allocation from process %d failed\n", 1006 len, current->pid); 1007 show_free_areas(); 1008 return -ENOMEM; 1009 1010 error_getting_vml: 1011 printk("Allocation of vml for %lu byte allocation from process %d failed\n", 1012 len, current->pid); 1013 show_free_areas(); 1014 return -ENOMEM; 1015} 1016EXPORT_SYMBOL(do_mmap_pgoff); 1017 1018/* 1019 * handle mapping disposal for uClinux 1020 */ 1021static void put_vma(struct vm_area_struct *vma) 1022{ 1023 if (vma) { 1024 down_write(&nommu_vma_sem); 1025 1026 if (atomic_dec_and_test(&vma->vm_usage)) { 1027 delete_nommu_vma(vma); 1028 1029 if (vma->vm_ops && vma->vm_ops->close) 1030 vma->vm_ops->close(vma); 1031 1032 /* IO memory and memory shared directly out of the pagecache from 1033 * ramfs/tmpfs mustn't be released here */ 1034 if (vma->vm_flags & VM_MAPPED_COPY) { 1035 realalloc -= kobjsize((void *) vma->vm_start); 1036 askedalloc -= vma->vm_end - vma->vm_start; 1037 kfree((void *) vma->vm_start); 1038 } 1039 1040 realalloc -= kobjsize(vma); 1041 askedalloc -= sizeof(*vma); 1042 1043 if (vma->vm_file) 1044 fput(vma->vm_file); 1045 kfree(vma); 1046 } 1047 1048 up_write(&nommu_vma_sem); 1049 } 1050} 1051 1052/* 1053 * release a mapping 1054 * - under NOMMU conditions the parameters must match exactly to the mapping to 1055 * be removed 1056 */ 1057int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len) 1058{ 1059 struct vm_list_struct *vml, **parent; 1060 unsigned long end = addr + len; 1061 1062#ifdef DEBUG 1063 printk("do_munmap:\n"); 1064#endif 1065 1066 for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) { 1067 if ((*parent)->vma->vm_start > addr) 1068 break; 1069 if ((*parent)->vma->vm_start == addr && 1070 ((len == 0) || ((*parent)->vma->vm_end == end))) 1071 goto found; 1072 } 1073 1074 printk("munmap of non-mmaped memory by process %d (%s): %p\n", 1075 current->pid, current->comm, (void *) addr); 1076 return -EINVAL; 1077 1078 found: 1079 vml = *parent; 1080 1081 put_vma(vml->vma); 1082 1083 *parent = vml->next; 1084 realalloc -= kobjsize(vml); 1085 askedalloc -= sizeof(*vml); 1086 kfree(vml); 1087 1088 update_hiwater_vm(mm); 1089 mm->total_vm -= len >> PAGE_SHIFT; 1090 1091#ifdef DEBUG 1092 show_process_blocks(); 1093#endif 1094 1095 return 0; 1096} 1097EXPORT_SYMBOL(do_munmap); 1098 1099asmlinkage long sys_munmap(unsigned long addr, size_t len) 1100{ 1101 int ret; 1102 struct mm_struct *mm = current->mm; 1103 1104 down_write(&mm->mmap_sem); 1105 ret = do_munmap(mm, addr, len); 1106 up_write(&mm->mmap_sem); 1107 return ret; 1108} 1109 1110/* 1111 * Release all mappings 1112 */ 1113void exit_mmap(struct mm_struct * mm) 1114{ 1115 struct vm_list_struct *tmp; 1116 1117 if (mm) { 1118#ifdef DEBUG 1119 printk("Exit_mmap:\n"); 1120#endif 1121 1122 mm->total_vm = 0; 1123 1124 while ((tmp = mm->context.vmlist)) { 1125 mm->context.vmlist = tmp->next; 1126 put_vma(tmp->vma); 1127 1128 realalloc -= kobjsize(tmp); 1129 askedalloc -= sizeof(*tmp); 1130 kfree(tmp); 1131 } 1132 1133#ifdef DEBUG 1134 show_process_blocks(); 1135#endif 1136 } 1137} 1138 1139unsigned long do_brk(unsigned long addr, unsigned long len) 1140{ 1141 return -ENOMEM; 1142} 1143 1144/* 1145 * expand (or shrink) an existing mapping, potentially moving it at the same 1146 * time (controlled by the MREMAP_MAYMOVE flag and available VM space) 1147 * 1148 * under NOMMU conditions, we only permit changing a mapping's size, and only 1149 * as long as it stays within the hole allocated by the kmalloc() call in 1150 * do_mmap_pgoff() and the block is not shareable 1151 * 1152 * MREMAP_FIXED is not supported under NOMMU conditions 1153 */ 1154unsigned long do_mremap(unsigned long addr, 1155 unsigned long old_len, unsigned long new_len, 1156 unsigned long flags, unsigned long new_addr) 1157{ 1158 struct vm_area_struct *vma; 1159 1160 /* insanity checks first */ 1161 if (new_len == 0) 1162 return (unsigned long) -EINVAL; 1163 1164 if (flags & MREMAP_FIXED && new_addr != addr) 1165 return (unsigned long) -EINVAL; 1166 1167 vma = find_vma_exact(current->mm, addr); 1168 if (!vma) 1169 return (unsigned long) -EINVAL; 1170 1171 if (vma->vm_end != vma->vm_start + old_len) 1172 return (unsigned long) -EFAULT; 1173 1174 if (vma->vm_flags & VM_MAYSHARE) 1175 return (unsigned long) -EPERM; 1176 1177 if (new_len > kobjsize((void *) addr)) 1178 return (unsigned long) -ENOMEM; 1179 1180 /* all checks complete - do it */ 1181 vma->vm_end = vma->vm_start + new_len; 1182 1183 askedalloc -= old_len; 1184 askedalloc += new_len; 1185 1186 return vma->vm_start; 1187} 1188EXPORT_SYMBOL(do_mremap); 1189 1190asmlinkage unsigned long sys_mremap(unsigned long addr, 1191 unsigned long old_len, unsigned long new_len, 1192 unsigned long flags, unsigned long new_addr) 1193{ 1194 unsigned long ret; 1195 1196 down_write(¤t->mm->mmap_sem); 1197 ret = do_mremap(addr, old_len, new_len, flags, new_addr); 1198 up_write(¤t->mm->mmap_sem); 1199 return ret; 1200} 1201 1202struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 1203 unsigned int foll_flags) 1204{ 1205 return NULL; 1206} 1207 1208int remap_pfn_range(struct vm_area_struct *vma, unsigned long from, 1209 unsigned long to, unsigned long size, pgprot_t prot) 1210{ 1211 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT; 1212 return 0; 1213} 1214EXPORT_SYMBOL(remap_pfn_range); 1215 1216void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) 1217{ 1218} 1219 1220unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, 1221 unsigned long len, unsigned long pgoff, unsigned long flags) 1222{ 1223 return -ENOMEM; 1224} 1225 1226void arch_unmap_area(struct mm_struct *mm, unsigned long addr) 1227{ 1228} 1229 1230void unmap_mapping_range(struct address_space *mapping, 1231 loff_t const holebegin, loff_t const holelen, 1232 int even_cows) 1233{ 1234} 1235EXPORT_SYMBOL(unmap_mapping_range); 1236 1237/* 1238 * ask for an unmapped area at which to create a mapping on a file 1239 */ 1240unsigned long get_unmapped_area(struct file *file, unsigned long addr, 1241 unsigned long len, unsigned long pgoff, 1242 unsigned long flags) 1243{ 1244 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, 1245 unsigned long, unsigned long); 1246 1247 get_area = current->mm->get_unmapped_area; 1248 if (file && file->f_op && file->f_op->get_unmapped_area) 1249 get_area = file->f_op->get_unmapped_area; 1250 1251 if (!get_area) 1252 return -ENOSYS; 1253 1254 return get_area(file, addr, len, pgoff, flags); 1255} 1256EXPORT_SYMBOL(get_unmapped_area); 1257 1258/* 1259 * Check that a process has enough memory to allocate a new virtual 1260 * mapping. 0 means there is enough memory for the allocation to 1261 * succeed and -ENOMEM implies there is not. 1262 * 1263 * We currently support three overcommit policies, which are set via the 1264 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1265 * 1266 * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1267 * Additional code 2002 Jul 20 by Robert Love. 1268 * 1269 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1270 * 1271 * Note this is a helper function intended to be used by LSMs which 1272 * wish to use this logic. 1273 */ 1274int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1275{ 1276 unsigned long free, allowed; 1277 1278 vm_acct_memory(pages); 1279 1280 /* 1281 * Sometimes we want to use more memory than we have 1282 */ 1283 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1284 return 0; 1285 1286 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1287 unsigned long n; 1288 1289 free = global_page_state(NR_FILE_PAGES); 1290 free += nr_swap_pages; 1291 1292 /* 1293 * Any slabs which are created with the 1294 * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1295 * which are reclaimable, under pressure. The dentry 1296 * cache and most inode caches should fall into this 1297 */ 1298 free += global_page_state(NR_SLAB_RECLAIMABLE); 1299 1300 /* 1301 * Leave the last 3% for root 1302 */ 1303 if (!cap_sys_admin) 1304 free -= free / 32; 1305 1306 if (free > pages) 1307 return 0; 1308 1309 /* 1310 * nr_free_pages() is very expensive on large systems, 1311 * only call if we're about to fail. 1312 */ 1313 n = nr_free_pages(); 1314 1315 /* 1316 * Leave reserved pages. The pages are not for anonymous pages. 1317 */ 1318 if (n <= totalreserve_pages) 1319 goto error; 1320 else 1321 n -= totalreserve_pages; 1322 1323 /* 1324 * Leave the last 3% for root 1325 */ 1326 if (!cap_sys_admin) 1327 n -= n / 32; 1328 free += n; 1329 1330 if (free > pages) 1331 return 0; 1332 1333 goto error; 1334 } 1335 1336 allowed = totalram_pages * sysctl_overcommit_ratio / 100; 1337 /* 1338 * Leave the last 3% for root 1339 */ 1340 if (!cap_sys_admin) 1341 allowed -= allowed / 32; 1342 allowed += total_swap_pages; 1343 1344 /* Don't let a single process grow too big: 1345 leave 3% of the size of this process for other processes */ 1346 allowed -= current->mm->total_vm / 32; 1347 1348 /* 1349 * cast `allowed' as a signed long because vm_committed_space 1350 * sometimes has a negative value 1351 */ 1352 if (atomic_read(&vm_committed_space) < (long)allowed) 1353 return 0; 1354error: 1355 vm_unacct_memory(pages); 1356 1357 return -ENOMEM; 1358} 1359 1360int in_gate_area_no_task(unsigned long addr) 1361{ 1362 return 0; 1363} 1364 1365int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1366{ 1367 BUG(); 1368 return 0; 1369} 1370EXPORT_SYMBOL(filemap_fault); 1371 1372/* 1373 * Access another process' address space. 1374 * - source/target buffer must be kernel space 1375 */ 1376int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) 1377{ 1378 struct vm_area_struct *vma; 1379 struct mm_struct *mm; 1380 1381 if (addr + len < addr) 1382 return 0; 1383 1384 mm = get_task_mm(tsk); 1385 if (!mm) 1386 return 0; 1387 1388 down_read(&mm->mmap_sem); 1389 1390 /* the access must start within one of the target process's mappings */ 1391 vma = find_vma(mm, addr); 1392 if (vma) { 1393 /* don't overrun this mapping */ 1394 if (addr + len >= vma->vm_end) 1395 len = vma->vm_end - addr; 1396 1397 /* only read or write mappings where it is permitted */ 1398 if (write && vma->vm_flags & VM_MAYWRITE) 1399 len -= copy_to_user((void *) addr, buf, len); 1400 else if (!write && vma->vm_flags & VM_MAYREAD) 1401 len -= copy_from_user(buf, (void *) addr, len); 1402 else 1403 len = 0; 1404 } else { 1405 len = 0; 1406 } 1407 1408 up_read(&mm->mmap_sem); 1409 mmput(mm); 1410 return len; 1411} 1412