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) 2004-2008 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) 2007-2010 Paul Mundt <lethal@linux-sh.org> 14 */ 15 16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 17 18 #include <linux/export.h> 19 #include <linux/mm.h> 20 #include <linux/vmacache.h> 21 #include <linux/mman.h> 22 #include <linux/swap.h> 23 #include <linux/file.h> 24 #include <linux/highmem.h> 25 #include <linux/pagemap.h> 26 #include <linux/slab.h> 27 #include <linux/vmalloc.h> 28 #include <linux/blkdev.h> 29 #include <linux/backing-dev.h> 30 #include <linux/compiler.h> 31 #include <linux/mount.h> 32 #include <linux/personality.h> 33 #include <linux/security.h> 34 #include <linux/syscalls.h> 35 #include <linux/audit.h> 36 #include <linux/sched/sysctl.h> 37 #include <linux/printk.h> 38 39 #include <asm/uaccess.h> 40 #include <asm/tlb.h> 41 #include <asm/tlbflush.h> 42 #include <asm/mmu_context.h> 43 #include "internal.h" 44 45 #if 0 46 #define kenter(FMT, ...) \ 47 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 48 #define kleave(FMT, ...) \ 49 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 50 #define kdebug(FMT, ...) \ 51 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) 52 #else 53 #define kenter(FMT, ...) \ 54 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 55 #define kleave(FMT, ...) \ 56 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 57 #define kdebug(FMT, ...) \ 58 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) 59 #endif 60 61 void *high_memory; 62 EXPORT_SYMBOL(high_memory); 63 struct page *mem_map; 64 unsigned long max_mapnr; 65 unsigned long highest_memmap_pfn; 66 struct percpu_counter vm_committed_as; 67 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 68 int sysctl_overcommit_ratio = 50; /* default is 50% */ 69 unsigned long sysctl_overcommit_kbytes __read_mostly; 70 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; 71 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; 72 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ 73 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ 74 int heap_stack_gap = 0; 75 76 atomic_long_t mmap_pages_allocated; 77 78 /* 79 * The global memory commitment made in the system can be a metric 80 * that can be used to drive ballooning decisions when Linux is hosted 81 * as a guest. On Hyper-V, the host implements a policy engine for dynamically 82 * balancing memory across competing virtual machines that are hosted. 83 * Several metrics drive this policy engine including the guest reported 84 * memory commitment. 85 */ 86 unsigned long vm_memory_committed(void) 87 { 88 return percpu_counter_read_positive(&vm_committed_as); 89 } 90 91 EXPORT_SYMBOL_GPL(vm_memory_committed); 92 93 EXPORT_SYMBOL(mem_map); 94 95 /* list of mapped, potentially shareable regions */ 96 static struct kmem_cache *vm_region_jar; 97 struct rb_root nommu_region_tree = RB_ROOT; 98 DECLARE_RWSEM(nommu_region_sem); 99 100 const struct vm_operations_struct generic_file_vm_ops = { 101 }; 102 103 /* 104 * Return the total memory allocated for this pointer, not 105 * just what the caller asked for. 106 * 107 * Doesn't have to be accurate, i.e. may have races. 108 */ 109 unsigned int kobjsize(const void *objp) 110 { 111 struct page *page; 112 113 /* 114 * If the object we have should not have ksize performed on it, 115 * return size of 0 116 */ 117 if (!objp || !virt_addr_valid(objp)) 118 return 0; 119 120 page = virt_to_head_page(objp); 121 122 /* 123 * If the allocator sets PageSlab, we know the pointer came from 124 * kmalloc(). 125 */ 126 if (PageSlab(page)) 127 return ksize(objp); 128 129 /* 130 * If it's not a compound page, see if we have a matching VMA 131 * region. This test is intentionally done in reverse order, 132 * so if there's no VMA, we still fall through and hand back 133 * PAGE_SIZE for 0-order pages. 134 */ 135 if (!PageCompound(page)) { 136 struct vm_area_struct *vma; 137 138 vma = find_vma(current->mm, (unsigned long)objp); 139 if (vma) 140 return vma->vm_end - vma->vm_start; 141 } 142 143 /* 144 * The ksize() function is only guaranteed to work for pointers 145 * returned by kmalloc(). So handle arbitrary pointers here. 146 */ 147 return PAGE_SIZE << compound_order(page); 148 } 149 150 long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 151 unsigned long start, unsigned long nr_pages, 152 unsigned int foll_flags, struct page **pages, 153 struct vm_area_struct **vmas, int *nonblocking) 154 { 155 struct vm_area_struct *vma; 156 unsigned long vm_flags; 157 int i; 158 159 /* calculate required read or write permissions. 160 * If FOLL_FORCE is set, we only require the "MAY" flags. 161 */ 162 vm_flags = (foll_flags & FOLL_WRITE) ? 163 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 164 vm_flags &= (foll_flags & FOLL_FORCE) ? 165 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 166 167 for (i = 0; i < nr_pages; i++) { 168 vma = find_vma(mm, start); 169 if (!vma) 170 goto finish_or_fault; 171 172 /* protect what we can, including chardevs */ 173 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || 174 !(vm_flags & vma->vm_flags)) 175 goto finish_or_fault; 176 177 if (pages) { 178 pages[i] = virt_to_page(start); 179 if (pages[i]) 180 page_cache_get(pages[i]); 181 } 182 if (vmas) 183 vmas[i] = vma; 184 start = (start + PAGE_SIZE) & PAGE_MASK; 185 } 186 187 return i; 188 189 finish_or_fault: 190 return i ? : -EFAULT; 191 } 192 193 /* 194 * get a list of pages in an address range belonging to the specified process 195 * and indicate the VMA that covers each page 196 * - this is potentially dodgy as we may end incrementing the page count of a 197 * slab page or a secondary page from a compound page 198 * - don't permit access to VMAs that don't support it, such as I/O mappings 199 */ 200 long get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 201 unsigned long start, unsigned long nr_pages, 202 int write, int force, struct page **pages, 203 struct vm_area_struct **vmas) 204 { 205 int flags = 0; 206 207 if (write) 208 flags |= FOLL_WRITE; 209 if (force) 210 flags |= FOLL_FORCE; 211 212 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, 213 NULL); 214 } 215 EXPORT_SYMBOL(get_user_pages); 216 217 long get_user_pages_locked(struct task_struct *tsk, struct mm_struct *mm, 218 unsigned long start, unsigned long nr_pages, 219 int write, int force, struct page **pages, 220 int *locked) 221 { 222 return get_user_pages(tsk, mm, start, nr_pages, write, force, 223 pages, NULL); 224 } 225 EXPORT_SYMBOL(get_user_pages_locked); 226 227 long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm, 228 unsigned long start, unsigned long nr_pages, 229 int write, int force, struct page **pages, 230 unsigned int gup_flags) 231 { 232 long ret; 233 down_read(&mm->mmap_sem); 234 ret = get_user_pages(tsk, mm, start, nr_pages, write, force, 235 pages, NULL); 236 up_read(&mm->mmap_sem); 237 return ret; 238 } 239 EXPORT_SYMBOL(__get_user_pages_unlocked); 240 241 long get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm, 242 unsigned long start, unsigned long nr_pages, 243 int write, int force, struct page **pages) 244 { 245 return __get_user_pages_unlocked(tsk, mm, start, nr_pages, write, 246 force, pages, 0); 247 } 248 EXPORT_SYMBOL(get_user_pages_unlocked); 249 250 /** 251 * follow_pfn - look up PFN at a user virtual address 252 * @vma: memory mapping 253 * @address: user virtual address 254 * @pfn: location to store found PFN 255 * 256 * Only IO mappings and raw PFN mappings are allowed. 257 * 258 * Returns zero and the pfn at @pfn on success, -ve otherwise. 259 */ 260 int follow_pfn(struct vm_area_struct *vma, unsigned long address, 261 unsigned long *pfn) 262 { 263 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 264 return -EINVAL; 265 266 *pfn = address >> PAGE_SHIFT; 267 return 0; 268 } 269 EXPORT_SYMBOL(follow_pfn); 270 271 LIST_HEAD(vmap_area_list); 272 273 void vfree(const void *addr) 274 { 275 kfree(addr); 276 } 277 EXPORT_SYMBOL(vfree); 278 279 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 280 { 281 /* 282 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 283 * returns only a logical address. 284 */ 285 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 286 } 287 EXPORT_SYMBOL(__vmalloc); 288 289 void *vmalloc_user(unsigned long size) 290 { 291 void *ret; 292 293 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 294 PAGE_KERNEL); 295 if (ret) { 296 struct vm_area_struct *vma; 297 298 down_write(¤t->mm->mmap_sem); 299 vma = find_vma(current->mm, (unsigned long)ret); 300 if (vma) 301 vma->vm_flags |= VM_USERMAP; 302 up_write(¤t->mm->mmap_sem); 303 } 304 305 return ret; 306 } 307 EXPORT_SYMBOL(vmalloc_user); 308 309 struct page *vmalloc_to_page(const void *addr) 310 { 311 return virt_to_page(addr); 312 } 313 EXPORT_SYMBOL(vmalloc_to_page); 314 315 unsigned long vmalloc_to_pfn(const void *addr) 316 { 317 return page_to_pfn(virt_to_page(addr)); 318 } 319 EXPORT_SYMBOL(vmalloc_to_pfn); 320 321 long vread(char *buf, char *addr, unsigned long count) 322 { 323 /* Don't allow overflow */ 324 if ((unsigned long) buf + count < count) 325 count = -(unsigned long) buf; 326 327 memcpy(buf, addr, count); 328 return count; 329 } 330 331 long vwrite(char *buf, char *addr, unsigned long count) 332 { 333 /* Don't allow overflow */ 334 if ((unsigned long) addr + count < count) 335 count = -(unsigned long) addr; 336 337 memcpy(addr, buf, count); 338 return count; 339 } 340 341 /* 342 * vmalloc - allocate virtually continguos memory 343 * 344 * @size: allocation size 345 * 346 * Allocate enough pages to cover @size from the page level 347 * allocator and map them into continguos kernel virtual space. 348 * 349 * For tight control over page level allocator and protection flags 350 * use __vmalloc() instead. 351 */ 352 void *vmalloc(unsigned long size) 353 { 354 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 355 } 356 EXPORT_SYMBOL(vmalloc); 357 358 /* 359 * vzalloc - allocate virtually continguos memory with zero fill 360 * 361 * @size: allocation size 362 * 363 * Allocate enough pages to cover @size from the page level 364 * allocator and map them into continguos kernel virtual space. 365 * The memory allocated is set to zero. 366 * 367 * For tight control over page level allocator and protection flags 368 * use __vmalloc() instead. 369 */ 370 void *vzalloc(unsigned long size) 371 { 372 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 373 PAGE_KERNEL); 374 } 375 EXPORT_SYMBOL(vzalloc); 376 377 /** 378 * vmalloc_node - allocate memory on a specific node 379 * @size: allocation size 380 * @node: numa node 381 * 382 * Allocate enough pages to cover @size from the page level 383 * allocator and map them into contiguous kernel virtual space. 384 * 385 * For tight control over page level allocator and protection flags 386 * use __vmalloc() instead. 387 */ 388 void *vmalloc_node(unsigned long size, int node) 389 { 390 return vmalloc(size); 391 } 392 EXPORT_SYMBOL(vmalloc_node); 393 394 /** 395 * vzalloc_node - allocate memory on a specific node with zero fill 396 * @size: allocation size 397 * @node: numa node 398 * 399 * Allocate enough pages to cover @size from the page level 400 * allocator and map them into contiguous kernel virtual space. 401 * The memory allocated is set to zero. 402 * 403 * For tight control over page level allocator and protection flags 404 * use __vmalloc() instead. 405 */ 406 void *vzalloc_node(unsigned long size, int node) 407 { 408 return vzalloc(size); 409 } 410 EXPORT_SYMBOL(vzalloc_node); 411 412 #ifndef PAGE_KERNEL_EXEC 413 # define PAGE_KERNEL_EXEC PAGE_KERNEL 414 #endif 415 416 /** 417 * vmalloc_exec - allocate virtually contiguous, executable memory 418 * @size: allocation size 419 * 420 * Kernel-internal function to allocate enough pages to cover @size 421 * the page level allocator and map them into contiguous and 422 * executable kernel virtual space. 423 * 424 * For tight control over page level allocator and protection flags 425 * use __vmalloc() instead. 426 */ 427 428 void *vmalloc_exec(unsigned long size) 429 { 430 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); 431 } 432 433 /** 434 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 435 * @size: allocation size 436 * 437 * Allocate enough 32bit PA addressable pages to cover @size from the 438 * page level allocator and map them into continguos kernel virtual space. 439 */ 440 void *vmalloc_32(unsigned long size) 441 { 442 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 443 } 444 EXPORT_SYMBOL(vmalloc_32); 445 446 /** 447 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 448 * @size: allocation size 449 * 450 * The resulting memory area is 32bit addressable and zeroed so it can be 451 * mapped to userspace without leaking data. 452 * 453 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to 454 * remap_vmalloc_range() are permissible. 455 */ 456 void *vmalloc_32_user(unsigned long size) 457 { 458 /* 459 * We'll have to sort out the ZONE_DMA bits for 64-bit, 460 * but for now this can simply use vmalloc_user() directly. 461 */ 462 return vmalloc_user(size); 463 } 464 EXPORT_SYMBOL(vmalloc_32_user); 465 466 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 467 { 468 BUG(); 469 return NULL; 470 } 471 EXPORT_SYMBOL(vmap); 472 473 void vunmap(const void *addr) 474 { 475 BUG(); 476 } 477 EXPORT_SYMBOL(vunmap); 478 479 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) 480 { 481 BUG(); 482 return NULL; 483 } 484 EXPORT_SYMBOL(vm_map_ram); 485 486 void vm_unmap_ram(const void *mem, unsigned int count) 487 { 488 BUG(); 489 } 490 EXPORT_SYMBOL(vm_unmap_ram); 491 492 void vm_unmap_aliases(void) 493 { 494 } 495 EXPORT_SYMBOL_GPL(vm_unmap_aliases); 496 497 /* 498 * Implement a stub for vmalloc_sync_all() if the architecture chose not to 499 * have one. 500 */ 501 void __weak vmalloc_sync_all(void) 502 { 503 } 504 505 /** 506 * alloc_vm_area - allocate a range of kernel address space 507 * @size: size of the area 508 * 509 * Returns: NULL on failure, vm_struct on success 510 * 511 * This function reserves a range of kernel address space, and 512 * allocates pagetables to map that range. No actual mappings 513 * are created. If the kernel address space is not shared 514 * between processes, it syncs the pagetable across all 515 * processes. 516 */ 517 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) 518 { 519 BUG(); 520 return NULL; 521 } 522 EXPORT_SYMBOL_GPL(alloc_vm_area); 523 524 void free_vm_area(struct vm_struct *area) 525 { 526 BUG(); 527 } 528 EXPORT_SYMBOL_GPL(free_vm_area); 529 530 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 531 struct page *page) 532 { 533 return -EINVAL; 534 } 535 EXPORT_SYMBOL(vm_insert_page); 536 537 /* 538 * sys_brk() for the most part doesn't need the global kernel 539 * lock, except when an application is doing something nasty 540 * like trying to un-brk an area that has already been mapped 541 * to a regular file. in this case, the unmapping will need 542 * to invoke file system routines that need the global lock. 543 */ 544 SYSCALL_DEFINE1(brk, unsigned long, brk) 545 { 546 struct mm_struct *mm = current->mm; 547 548 if (brk < mm->start_brk || brk > mm->context.end_brk) 549 return mm->brk; 550 551 if (mm->brk == brk) 552 return mm->brk; 553 554 /* 555 * Always allow shrinking brk 556 */ 557 if (brk <= mm->brk) { 558 mm->brk = brk; 559 return brk; 560 } 561 562 /* 563 * Ok, looks good - let it rip. 564 */ 565 flush_icache_range(mm->brk, brk); 566 return mm->brk = brk; 567 } 568 569 /* 570 * initialise the VMA and region record slabs 571 */ 572 void __init mmap_init(void) 573 { 574 int ret; 575 576 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 577 VM_BUG_ON(ret); 578 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); 579 } 580 581 /* 582 * validate the region tree 583 * - the caller must hold the region lock 584 */ 585 #ifdef CONFIG_DEBUG_NOMMU_REGIONS 586 static noinline void validate_nommu_regions(void) 587 { 588 struct vm_region *region, *last; 589 struct rb_node *p, *lastp; 590 591 lastp = rb_first(&nommu_region_tree); 592 if (!lastp) 593 return; 594 595 last = rb_entry(lastp, struct vm_region, vm_rb); 596 BUG_ON(unlikely(last->vm_end <= last->vm_start)); 597 BUG_ON(unlikely(last->vm_top < last->vm_end)); 598 599 while ((p = rb_next(lastp))) { 600 region = rb_entry(p, struct vm_region, vm_rb); 601 last = rb_entry(lastp, struct vm_region, vm_rb); 602 603 BUG_ON(unlikely(region->vm_end <= region->vm_start)); 604 BUG_ON(unlikely(region->vm_top < region->vm_end)); 605 BUG_ON(unlikely(region->vm_start < last->vm_top)); 606 607 lastp = p; 608 } 609 } 610 #else 611 static void validate_nommu_regions(void) 612 { 613 } 614 #endif 615 616 /* 617 * add a region into the global tree 618 */ 619 static void add_nommu_region(struct vm_region *region) 620 { 621 struct vm_region *pregion; 622 struct rb_node **p, *parent; 623 624 validate_nommu_regions(); 625 626 parent = NULL; 627 p = &nommu_region_tree.rb_node; 628 while (*p) { 629 parent = *p; 630 pregion = rb_entry(parent, struct vm_region, vm_rb); 631 if (region->vm_start < pregion->vm_start) 632 p = &(*p)->rb_left; 633 else if (region->vm_start > pregion->vm_start) 634 p = &(*p)->rb_right; 635 else if (pregion == region) 636 return; 637 else 638 BUG(); 639 } 640 641 rb_link_node(®ion->vm_rb, parent, p); 642 rb_insert_color(®ion->vm_rb, &nommu_region_tree); 643 644 validate_nommu_regions(); 645 } 646 647 /* 648 * delete a region from the global tree 649 */ 650 static void delete_nommu_region(struct vm_region *region) 651 { 652 BUG_ON(!nommu_region_tree.rb_node); 653 654 validate_nommu_regions(); 655 rb_erase(®ion->vm_rb, &nommu_region_tree); 656 validate_nommu_regions(); 657 } 658 659 /* 660 * free a contiguous series of pages 661 */ 662 static void free_page_series(unsigned long from, unsigned long to) 663 { 664 for (; from < to; from += PAGE_SIZE) { 665 struct page *page = virt_to_page(from); 666 667 kdebug("- free %lx", from); 668 atomic_long_dec(&mmap_pages_allocated); 669 if (page_count(page) != 1) 670 kdebug("free page %p: refcount not one: %d", 671 page, page_count(page)); 672 put_page(page); 673 } 674 } 675 676 /* 677 * release a reference to a region 678 * - the caller must hold the region semaphore for writing, which this releases 679 * - the region may not have been added to the tree yet, in which case vm_top 680 * will equal vm_start 681 */ 682 static void __put_nommu_region(struct vm_region *region) 683 __releases(nommu_region_sem) 684 { 685 kenter("%p{%d}", region, region->vm_usage); 686 687 BUG_ON(!nommu_region_tree.rb_node); 688 689 if (--region->vm_usage == 0) { 690 if (region->vm_top > region->vm_start) 691 delete_nommu_region(region); 692 up_write(&nommu_region_sem); 693 694 if (region->vm_file) 695 fput(region->vm_file); 696 697 /* IO memory and memory shared directly out of the pagecache 698 * from ramfs/tmpfs mustn't be released here */ 699 if (region->vm_flags & VM_MAPPED_COPY) { 700 kdebug("free series"); 701 free_page_series(region->vm_start, region->vm_top); 702 } 703 kmem_cache_free(vm_region_jar, region); 704 } else { 705 up_write(&nommu_region_sem); 706 } 707 } 708 709 /* 710 * release a reference to a region 711 */ 712 static void put_nommu_region(struct vm_region *region) 713 { 714 down_write(&nommu_region_sem); 715 __put_nommu_region(region); 716 } 717 718 /* 719 * update protection on a vma 720 */ 721 static void protect_vma(struct vm_area_struct *vma, unsigned long flags) 722 { 723 #ifdef CONFIG_MPU 724 struct mm_struct *mm = vma->vm_mm; 725 long start = vma->vm_start & PAGE_MASK; 726 while (start < vma->vm_end) { 727 protect_page(mm, start, flags); 728 start += PAGE_SIZE; 729 } 730 update_protections(mm); 731 #endif 732 } 733 734 /* 735 * add a VMA into a process's mm_struct in the appropriate place in the list 736 * and tree and add to the address space's page tree also if not an anonymous 737 * page 738 * - should be called with mm->mmap_sem held writelocked 739 */ 740 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) 741 { 742 struct vm_area_struct *pvma, *prev; 743 struct address_space *mapping; 744 struct rb_node **p, *parent, *rb_prev; 745 746 kenter(",%p", vma); 747 748 BUG_ON(!vma->vm_region); 749 750 mm->map_count++; 751 vma->vm_mm = mm; 752 753 protect_vma(vma, vma->vm_flags); 754 755 /* add the VMA to the mapping */ 756 if (vma->vm_file) { 757 mapping = vma->vm_file->f_mapping; 758 759 i_mmap_lock_write(mapping); 760 flush_dcache_mmap_lock(mapping); 761 vma_interval_tree_insert(vma, &mapping->i_mmap); 762 flush_dcache_mmap_unlock(mapping); 763 i_mmap_unlock_write(mapping); 764 } 765 766 /* add the VMA to the tree */ 767 parent = rb_prev = NULL; 768 p = &mm->mm_rb.rb_node; 769 while (*p) { 770 parent = *p; 771 pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 772 773 /* sort by: start addr, end addr, VMA struct addr in that order 774 * (the latter is necessary as we may get identical VMAs) */ 775 if (vma->vm_start < pvma->vm_start) 776 p = &(*p)->rb_left; 777 else if (vma->vm_start > pvma->vm_start) { 778 rb_prev = parent; 779 p = &(*p)->rb_right; 780 } else if (vma->vm_end < pvma->vm_end) 781 p = &(*p)->rb_left; 782 else if (vma->vm_end > pvma->vm_end) { 783 rb_prev = parent; 784 p = &(*p)->rb_right; 785 } else if (vma < pvma) 786 p = &(*p)->rb_left; 787 else if (vma > pvma) { 788 rb_prev = parent; 789 p = &(*p)->rb_right; 790 } else 791 BUG(); 792 } 793 794 rb_link_node(&vma->vm_rb, parent, p); 795 rb_insert_color(&vma->vm_rb, &mm->mm_rb); 796 797 /* add VMA to the VMA list also */ 798 prev = NULL; 799 if (rb_prev) 800 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 801 802 __vma_link_list(mm, vma, prev, parent); 803 } 804 805 /* 806 * delete a VMA from its owning mm_struct and address space 807 */ 808 static void delete_vma_from_mm(struct vm_area_struct *vma) 809 { 810 int i; 811 struct address_space *mapping; 812 struct mm_struct *mm = vma->vm_mm; 813 struct task_struct *curr = current; 814 815 kenter("%p", vma); 816 817 protect_vma(vma, 0); 818 819 mm->map_count--; 820 for (i = 0; i < VMACACHE_SIZE; i++) { 821 /* if the vma is cached, invalidate the entire cache */ 822 if (curr->vmacache[i] == vma) { 823 vmacache_invalidate(mm); 824 break; 825 } 826 } 827 828 /* remove the VMA from the mapping */ 829 if (vma->vm_file) { 830 mapping = vma->vm_file->f_mapping; 831 832 i_mmap_lock_write(mapping); 833 flush_dcache_mmap_lock(mapping); 834 vma_interval_tree_remove(vma, &mapping->i_mmap); 835 flush_dcache_mmap_unlock(mapping); 836 i_mmap_unlock_write(mapping); 837 } 838 839 /* remove from the MM's tree and list */ 840 rb_erase(&vma->vm_rb, &mm->mm_rb); 841 842 if (vma->vm_prev) 843 vma->vm_prev->vm_next = vma->vm_next; 844 else 845 mm->mmap = vma->vm_next; 846 847 if (vma->vm_next) 848 vma->vm_next->vm_prev = vma->vm_prev; 849 } 850 851 /* 852 * destroy a VMA record 853 */ 854 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) 855 { 856 kenter("%p", vma); 857 if (vma->vm_ops && vma->vm_ops->close) 858 vma->vm_ops->close(vma); 859 if (vma->vm_file) 860 fput(vma->vm_file); 861 put_nommu_region(vma->vm_region); 862 kmem_cache_free(vm_area_cachep, vma); 863 } 864 865 /* 866 * look up the first VMA in which addr resides, NULL if none 867 * - should be called with mm->mmap_sem at least held readlocked 868 */ 869 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 870 { 871 struct vm_area_struct *vma; 872 873 /* check the cache first */ 874 vma = vmacache_find(mm, addr); 875 if (likely(vma)) 876 return vma; 877 878 /* trawl the list (there may be multiple mappings in which addr 879 * resides) */ 880 for (vma = mm->mmap; vma; vma = vma->vm_next) { 881 if (vma->vm_start > addr) 882 return NULL; 883 if (vma->vm_end > addr) { 884 vmacache_update(addr, vma); 885 return vma; 886 } 887 } 888 889 return NULL; 890 } 891 EXPORT_SYMBOL(find_vma); 892 893 /* 894 * find a VMA 895 * - we don't extend stack VMAs under NOMMU conditions 896 */ 897 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) 898 { 899 return find_vma(mm, addr); 900 } 901 902 /* 903 * expand a stack to a given address 904 * - not supported under NOMMU conditions 905 */ 906 int expand_stack(struct vm_area_struct *vma, unsigned long address) 907 { 908 return -ENOMEM; 909 } 910 911 /* 912 * look up the first VMA exactly that exactly matches addr 913 * - should be called with mm->mmap_sem at least held readlocked 914 */ 915 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, 916 unsigned long addr, 917 unsigned long len) 918 { 919 struct vm_area_struct *vma; 920 unsigned long end = addr + len; 921 922 /* check the cache first */ 923 vma = vmacache_find_exact(mm, addr, end); 924 if (vma) 925 return vma; 926 927 /* trawl the list (there may be multiple mappings in which addr 928 * resides) */ 929 for (vma = mm->mmap; vma; vma = vma->vm_next) { 930 if (vma->vm_start < addr) 931 continue; 932 if (vma->vm_start > addr) 933 return NULL; 934 if (vma->vm_end == end) { 935 vmacache_update(addr, vma); 936 return vma; 937 } 938 } 939 940 return NULL; 941 } 942 943 /* 944 * determine whether a mapping should be permitted and, if so, what sort of 945 * mapping we're capable of supporting 946 */ 947 static int validate_mmap_request(struct file *file, 948 unsigned long addr, 949 unsigned long len, 950 unsigned long prot, 951 unsigned long flags, 952 unsigned long pgoff, 953 unsigned long *_capabilities) 954 { 955 unsigned long capabilities, rlen; 956 int ret; 957 958 /* do the simple checks first */ 959 if (flags & MAP_FIXED) { 960 printk(KERN_DEBUG 961 "%d: Can't do fixed-address/overlay mmap of RAM\n", 962 current->pid); 963 return -EINVAL; 964 } 965 966 if ((flags & MAP_TYPE) != MAP_PRIVATE && 967 (flags & MAP_TYPE) != MAP_SHARED) 968 return -EINVAL; 969 970 if (!len) 971 return -EINVAL; 972 973 /* Careful about overflows.. */ 974 rlen = PAGE_ALIGN(len); 975 if (!rlen || rlen > TASK_SIZE) 976 return -ENOMEM; 977 978 /* offset overflow? */ 979 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) 980 return -EOVERFLOW; 981 982 if (file) { 983 /* files must support mmap */ 984 if (!file->f_op->mmap) 985 return -ENODEV; 986 987 /* work out if what we've got could possibly be shared 988 * - we support chardevs that provide their own "memory" 989 * - we support files/blockdevs that are memory backed 990 */ 991 if (file->f_op->mmap_capabilities) { 992 capabilities = file->f_op->mmap_capabilities(file); 993 } else { 994 /* no explicit capabilities set, so assume some 995 * defaults */ 996 switch (file_inode(file)->i_mode & S_IFMT) { 997 case S_IFREG: 998 case S_IFBLK: 999 capabilities = NOMMU_MAP_COPY; 1000 break; 1001 1002 case S_IFCHR: 1003 capabilities = 1004 NOMMU_MAP_DIRECT | 1005 NOMMU_MAP_READ | 1006 NOMMU_MAP_WRITE; 1007 break; 1008 1009 default: 1010 return -EINVAL; 1011 } 1012 } 1013 1014 /* eliminate any capabilities that we can't support on this 1015 * device */ 1016 if (!file->f_op->get_unmapped_area) 1017 capabilities &= ~NOMMU_MAP_DIRECT; 1018 if (!file->f_op->read) 1019 capabilities &= ~NOMMU_MAP_COPY; 1020 1021 /* The file shall have been opened with read permission. */ 1022 if (!(file->f_mode & FMODE_READ)) 1023 return -EACCES; 1024 1025 if (flags & MAP_SHARED) { 1026 /* do checks for writing, appending and locking */ 1027 if ((prot & PROT_WRITE) && 1028 !(file->f_mode & FMODE_WRITE)) 1029 return -EACCES; 1030 1031 if (IS_APPEND(file_inode(file)) && 1032 (file->f_mode & FMODE_WRITE)) 1033 return -EACCES; 1034 1035 if (locks_verify_locked(file)) 1036 return -EAGAIN; 1037 1038 if (!(capabilities & NOMMU_MAP_DIRECT)) 1039 return -ENODEV; 1040 1041 /* we mustn't privatise shared mappings */ 1042 capabilities &= ~NOMMU_MAP_COPY; 1043 } else { 1044 /* we're going to read the file into private memory we 1045 * allocate */ 1046 if (!(capabilities & NOMMU_MAP_COPY)) 1047 return -ENODEV; 1048 1049 /* we don't permit a private writable mapping to be 1050 * shared with the backing device */ 1051 if (prot & PROT_WRITE) 1052 capabilities &= ~NOMMU_MAP_DIRECT; 1053 } 1054 1055 if (capabilities & NOMMU_MAP_DIRECT) { 1056 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) || 1057 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) || 1058 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC)) 1059 ) { 1060 capabilities &= ~NOMMU_MAP_DIRECT; 1061 if (flags & MAP_SHARED) { 1062 printk(KERN_WARNING 1063 "MAP_SHARED not completely supported on !MMU\n"); 1064 return -EINVAL; 1065 } 1066 } 1067 } 1068 1069 /* handle executable mappings and implied executable 1070 * mappings */ 1071 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 1072 if (prot & PROT_EXEC) 1073 return -EPERM; 1074 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { 1075 /* handle implication of PROT_EXEC by PROT_READ */ 1076 if (current->personality & READ_IMPLIES_EXEC) { 1077 if (capabilities & NOMMU_MAP_EXEC) 1078 prot |= PROT_EXEC; 1079 } 1080 } else if ((prot & PROT_READ) && 1081 (prot & PROT_EXEC) && 1082 !(capabilities & NOMMU_MAP_EXEC) 1083 ) { 1084 /* backing file is not executable, try to copy */ 1085 capabilities &= ~NOMMU_MAP_DIRECT; 1086 } 1087 } else { 1088 /* anonymous mappings are always memory backed and can be 1089 * privately mapped 1090 */ 1091 capabilities = NOMMU_MAP_COPY; 1092 1093 /* handle PROT_EXEC implication by PROT_READ */ 1094 if ((prot & PROT_READ) && 1095 (current->personality & READ_IMPLIES_EXEC)) 1096 prot |= PROT_EXEC; 1097 } 1098 1099 /* allow the security API to have its say */ 1100 ret = security_mmap_addr(addr); 1101 if (ret < 0) 1102 return ret; 1103 1104 /* looks okay */ 1105 *_capabilities = capabilities; 1106 return 0; 1107 } 1108 1109 /* 1110 * we've determined that we can make the mapping, now translate what we 1111 * now know into VMA flags 1112 */ 1113 static unsigned long determine_vm_flags(struct file *file, 1114 unsigned long prot, 1115 unsigned long flags, 1116 unsigned long capabilities) 1117 { 1118 unsigned long vm_flags; 1119 1120 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); 1121 /* vm_flags |= mm->def_flags; */ 1122 1123 if (!(capabilities & NOMMU_MAP_DIRECT)) { 1124 /* attempt to share read-only copies of mapped file chunks */ 1125 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 1126 if (file && !(prot & PROT_WRITE)) 1127 vm_flags |= VM_MAYSHARE; 1128 } else { 1129 /* overlay a shareable mapping on the backing device or inode 1130 * if possible - used for chardevs, ramfs/tmpfs/shmfs and 1131 * romfs/cramfs */ 1132 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS); 1133 if (flags & MAP_SHARED) 1134 vm_flags |= VM_SHARED; 1135 } 1136 1137 /* refuse to let anyone share private mappings with this process if 1138 * it's being traced - otherwise breakpoints set in it may interfere 1139 * with another untraced process 1140 */ 1141 if ((flags & MAP_PRIVATE) && current->ptrace) 1142 vm_flags &= ~VM_MAYSHARE; 1143 1144 return vm_flags; 1145 } 1146 1147 /* 1148 * set up a shared mapping on a file (the driver or filesystem provides and 1149 * pins the storage) 1150 */ 1151 static int do_mmap_shared_file(struct vm_area_struct *vma) 1152 { 1153 int ret; 1154 1155 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1156 if (ret == 0) { 1157 vma->vm_region->vm_top = vma->vm_region->vm_end; 1158 return 0; 1159 } 1160 if (ret != -ENOSYS) 1161 return ret; 1162 1163 /* getting -ENOSYS indicates that direct mmap isn't possible (as 1164 * opposed to tried but failed) so we can only give a suitable error as 1165 * it's not possible to make a private copy if MAP_SHARED was given */ 1166 return -ENODEV; 1167 } 1168 1169 /* 1170 * set up a private mapping or an anonymous shared mapping 1171 */ 1172 static int do_mmap_private(struct vm_area_struct *vma, 1173 struct vm_region *region, 1174 unsigned long len, 1175 unsigned long capabilities) 1176 { 1177 unsigned long total, point; 1178 void *base; 1179 int ret, order; 1180 1181 /* invoke the file's mapping function so that it can keep track of 1182 * shared mappings on devices or memory 1183 * - VM_MAYSHARE will be set if it may attempt to share 1184 */ 1185 if (capabilities & NOMMU_MAP_DIRECT) { 1186 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1187 if (ret == 0) { 1188 /* shouldn't return success if we're not sharing */ 1189 BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); 1190 vma->vm_region->vm_top = vma->vm_region->vm_end; 1191 return 0; 1192 } 1193 if (ret != -ENOSYS) 1194 return ret; 1195 1196 /* getting an ENOSYS error indicates that direct mmap isn't 1197 * possible (as opposed to tried but failed) so we'll try to 1198 * make a private copy of the data and map that instead */ 1199 } 1200 1201 1202 /* allocate some memory to hold the mapping 1203 * - note that this may not return a page-aligned address if the object 1204 * we're allocating is smaller than a page 1205 */ 1206 order = get_order(len); 1207 kdebug("alloc order %d for %lx", order, len); 1208 1209 total = 1 << order; 1210 point = len >> PAGE_SHIFT; 1211 1212 /* we don't want to allocate a power-of-2 sized page set */ 1213 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { 1214 total = point; 1215 kdebug("try to alloc exact %lu pages", total); 1216 } 1217 1218 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL); 1219 if (!base) 1220 goto enomem; 1221 1222 atomic_long_add(total, &mmap_pages_allocated); 1223 1224 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; 1225 region->vm_start = (unsigned long) base; 1226 region->vm_end = region->vm_start + len; 1227 region->vm_top = region->vm_start + (total << PAGE_SHIFT); 1228 1229 vma->vm_start = region->vm_start; 1230 vma->vm_end = region->vm_start + len; 1231 1232 if (vma->vm_file) { 1233 /* read the contents of a file into the copy */ 1234 mm_segment_t old_fs; 1235 loff_t fpos; 1236 1237 fpos = vma->vm_pgoff; 1238 fpos <<= PAGE_SHIFT; 1239 1240 old_fs = get_fs(); 1241 set_fs(KERNEL_DS); 1242 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos); 1243 set_fs(old_fs); 1244 1245 if (ret < 0) 1246 goto error_free; 1247 1248 /* clear the last little bit */ 1249 if (ret < len) 1250 memset(base + ret, 0, len - ret); 1251 1252 } 1253 1254 return 0; 1255 1256 error_free: 1257 free_page_series(region->vm_start, region->vm_top); 1258 region->vm_start = vma->vm_start = 0; 1259 region->vm_end = vma->vm_end = 0; 1260 region->vm_top = 0; 1261 return ret; 1262 1263 enomem: 1264 pr_err("Allocation of length %lu from process %d (%s) failed\n", 1265 len, current->pid, current->comm); 1266 show_free_areas(0); 1267 return -ENOMEM; 1268 } 1269 1270 /* 1271 * handle mapping creation for uClinux 1272 */ 1273 unsigned long do_mmap_pgoff(struct file *file, 1274 unsigned long addr, 1275 unsigned long len, 1276 unsigned long prot, 1277 unsigned long flags, 1278 unsigned long pgoff, 1279 unsigned long *populate) 1280 { 1281 struct vm_area_struct *vma; 1282 struct vm_region *region; 1283 struct rb_node *rb; 1284 unsigned long capabilities, vm_flags, result; 1285 int ret; 1286 1287 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); 1288 1289 *populate = 0; 1290 1291 /* decide whether we should attempt the mapping, and if so what sort of 1292 * mapping */ 1293 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 1294 &capabilities); 1295 if (ret < 0) { 1296 kleave(" = %d [val]", ret); 1297 return ret; 1298 } 1299 1300 /* we ignore the address hint */ 1301 addr = 0; 1302 len = PAGE_ALIGN(len); 1303 1304 /* we've determined that we can make the mapping, now translate what we 1305 * now know into VMA flags */ 1306 vm_flags = determine_vm_flags(file, prot, flags, capabilities); 1307 1308 /* we're going to need to record the mapping */ 1309 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); 1310 if (!region) 1311 goto error_getting_region; 1312 1313 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 1314 if (!vma) 1315 goto error_getting_vma; 1316 1317 region->vm_usage = 1; 1318 region->vm_flags = vm_flags; 1319 region->vm_pgoff = pgoff; 1320 1321 INIT_LIST_HEAD(&vma->anon_vma_chain); 1322 vma->vm_flags = vm_flags; 1323 vma->vm_pgoff = pgoff; 1324 1325 if (file) { 1326 region->vm_file = get_file(file); 1327 vma->vm_file = get_file(file); 1328 } 1329 1330 down_write(&nommu_region_sem); 1331 1332 /* if we want to share, we need to check for regions created by other 1333 * mmap() calls that overlap with our proposed mapping 1334 * - we can only share with a superset match on most regular files 1335 * - shared mappings on character devices and memory backed files are 1336 * permitted to overlap inexactly as far as we are concerned for in 1337 * these cases, sharing is handled in the driver or filesystem rather 1338 * than here 1339 */ 1340 if (vm_flags & VM_MAYSHARE) { 1341 struct vm_region *pregion; 1342 unsigned long pglen, rpglen, pgend, rpgend, start; 1343 1344 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1345 pgend = pgoff + pglen; 1346 1347 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { 1348 pregion = rb_entry(rb, struct vm_region, vm_rb); 1349 1350 if (!(pregion->vm_flags & VM_MAYSHARE)) 1351 continue; 1352 1353 /* search for overlapping mappings on the same file */ 1354 if (file_inode(pregion->vm_file) != 1355 file_inode(file)) 1356 continue; 1357 1358 if (pregion->vm_pgoff >= pgend) 1359 continue; 1360 1361 rpglen = pregion->vm_end - pregion->vm_start; 1362 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; 1363 rpgend = pregion->vm_pgoff + rpglen; 1364 if (pgoff >= rpgend) 1365 continue; 1366 1367 /* handle inexactly overlapping matches between 1368 * mappings */ 1369 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && 1370 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { 1371 /* new mapping is not a subset of the region */ 1372 if (!(capabilities & NOMMU_MAP_DIRECT)) 1373 goto sharing_violation; 1374 continue; 1375 } 1376 1377 /* we've found a region we can share */ 1378 pregion->vm_usage++; 1379 vma->vm_region = pregion; 1380 start = pregion->vm_start; 1381 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; 1382 vma->vm_start = start; 1383 vma->vm_end = start + len; 1384 1385 if (pregion->vm_flags & VM_MAPPED_COPY) { 1386 kdebug("share copy"); 1387 vma->vm_flags |= VM_MAPPED_COPY; 1388 } else { 1389 kdebug("share mmap"); 1390 ret = do_mmap_shared_file(vma); 1391 if (ret < 0) { 1392 vma->vm_region = NULL; 1393 vma->vm_start = 0; 1394 vma->vm_end = 0; 1395 pregion->vm_usage--; 1396 pregion = NULL; 1397 goto error_just_free; 1398 } 1399 } 1400 fput(region->vm_file); 1401 kmem_cache_free(vm_region_jar, region); 1402 region = pregion; 1403 result = start; 1404 goto share; 1405 } 1406 1407 /* obtain the address at which to make a shared mapping 1408 * - this is the hook for quasi-memory character devices to 1409 * tell us the location of a shared mapping 1410 */ 1411 if (capabilities & NOMMU_MAP_DIRECT) { 1412 addr = file->f_op->get_unmapped_area(file, addr, len, 1413 pgoff, flags); 1414 if (IS_ERR_VALUE(addr)) { 1415 ret = addr; 1416 if (ret != -ENOSYS) 1417 goto error_just_free; 1418 1419 /* the driver refused to tell us where to site 1420 * the mapping so we'll have to attempt to copy 1421 * it */ 1422 ret = -ENODEV; 1423 if (!(capabilities & NOMMU_MAP_COPY)) 1424 goto error_just_free; 1425 1426 capabilities &= ~NOMMU_MAP_DIRECT; 1427 } else { 1428 vma->vm_start = region->vm_start = addr; 1429 vma->vm_end = region->vm_end = addr + len; 1430 } 1431 } 1432 } 1433 1434 vma->vm_region = region; 1435 1436 /* set up the mapping 1437 * - the region is filled in if NOMMU_MAP_DIRECT is still set 1438 */ 1439 if (file && vma->vm_flags & VM_SHARED) 1440 ret = do_mmap_shared_file(vma); 1441 else 1442 ret = do_mmap_private(vma, region, len, capabilities); 1443 if (ret < 0) 1444 goto error_just_free; 1445 add_nommu_region(region); 1446 1447 /* clear anonymous mappings that don't ask for uninitialized data */ 1448 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) 1449 memset((void *)region->vm_start, 0, 1450 region->vm_end - region->vm_start); 1451 1452 /* okay... we have a mapping; now we have to register it */ 1453 result = vma->vm_start; 1454 1455 current->mm->total_vm += len >> PAGE_SHIFT; 1456 1457 share: 1458 add_vma_to_mm(current->mm, vma); 1459 1460 /* we flush the region from the icache only when the first executable 1461 * mapping of it is made */ 1462 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { 1463 flush_icache_range(region->vm_start, region->vm_end); 1464 region->vm_icache_flushed = true; 1465 } 1466 1467 up_write(&nommu_region_sem); 1468 1469 kleave(" = %lx", result); 1470 return result; 1471 1472 error_just_free: 1473 up_write(&nommu_region_sem); 1474 error: 1475 if (region->vm_file) 1476 fput(region->vm_file); 1477 kmem_cache_free(vm_region_jar, region); 1478 if (vma->vm_file) 1479 fput(vma->vm_file); 1480 kmem_cache_free(vm_area_cachep, vma); 1481 kleave(" = %d", ret); 1482 return ret; 1483 1484 sharing_violation: 1485 up_write(&nommu_region_sem); 1486 printk(KERN_WARNING "Attempt to share mismatched mappings\n"); 1487 ret = -EINVAL; 1488 goto error; 1489 1490 error_getting_vma: 1491 kmem_cache_free(vm_region_jar, region); 1492 printk(KERN_WARNING "Allocation of vma for %lu byte allocation" 1493 " from process %d failed\n", 1494 len, current->pid); 1495 show_free_areas(0); 1496 return -ENOMEM; 1497 1498 error_getting_region: 1499 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" 1500 " from process %d failed\n", 1501 len, current->pid); 1502 show_free_areas(0); 1503 return -ENOMEM; 1504 } 1505 1506 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1507 unsigned long, prot, unsigned long, flags, 1508 unsigned long, fd, unsigned long, pgoff) 1509 { 1510 struct file *file = NULL; 1511 unsigned long retval = -EBADF; 1512 1513 audit_mmap_fd(fd, flags); 1514 if (!(flags & MAP_ANONYMOUS)) { 1515 file = fget(fd); 1516 if (!file) 1517 goto out; 1518 } 1519 1520 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 1521 1522 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1523 1524 if (file) 1525 fput(file); 1526 out: 1527 return retval; 1528 } 1529 1530 #ifdef __ARCH_WANT_SYS_OLD_MMAP 1531 struct mmap_arg_struct { 1532 unsigned long addr; 1533 unsigned long len; 1534 unsigned long prot; 1535 unsigned long flags; 1536 unsigned long fd; 1537 unsigned long offset; 1538 }; 1539 1540 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1541 { 1542 struct mmap_arg_struct a; 1543 1544 if (copy_from_user(&a, arg, sizeof(a))) 1545 return -EFAULT; 1546 if (a.offset & ~PAGE_MASK) 1547 return -EINVAL; 1548 1549 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1550 a.offset >> PAGE_SHIFT); 1551 } 1552 #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1553 1554 /* 1555 * split a vma into two pieces at address 'addr', a new vma is allocated either 1556 * for the first part or the tail. 1557 */ 1558 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 1559 unsigned long addr, int new_below) 1560 { 1561 struct vm_area_struct *new; 1562 struct vm_region *region; 1563 unsigned long npages; 1564 1565 kenter(""); 1566 1567 /* we're only permitted to split anonymous regions (these should have 1568 * only a single usage on the region) */ 1569 if (vma->vm_file) 1570 return -ENOMEM; 1571 1572 if (mm->map_count >= sysctl_max_map_count) 1573 return -ENOMEM; 1574 1575 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); 1576 if (!region) 1577 return -ENOMEM; 1578 1579 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 1580 if (!new) { 1581 kmem_cache_free(vm_region_jar, region); 1582 return -ENOMEM; 1583 } 1584 1585 /* most fields are the same, copy all, and then fixup */ 1586 *new = *vma; 1587 *region = *vma->vm_region; 1588 new->vm_region = region; 1589 1590 npages = (addr - vma->vm_start) >> PAGE_SHIFT; 1591 1592 if (new_below) { 1593 region->vm_top = region->vm_end = new->vm_end = addr; 1594 } else { 1595 region->vm_start = new->vm_start = addr; 1596 region->vm_pgoff = new->vm_pgoff += npages; 1597 } 1598 1599 if (new->vm_ops && new->vm_ops->open) 1600 new->vm_ops->open(new); 1601 1602 delete_vma_from_mm(vma); 1603 down_write(&nommu_region_sem); 1604 delete_nommu_region(vma->vm_region); 1605 if (new_below) { 1606 vma->vm_region->vm_start = vma->vm_start = addr; 1607 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; 1608 } else { 1609 vma->vm_region->vm_end = vma->vm_end = addr; 1610 vma->vm_region->vm_top = addr; 1611 } 1612 add_nommu_region(vma->vm_region); 1613 add_nommu_region(new->vm_region); 1614 up_write(&nommu_region_sem); 1615 add_vma_to_mm(mm, vma); 1616 add_vma_to_mm(mm, new); 1617 return 0; 1618 } 1619 1620 /* 1621 * shrink a VMA by removing the specified chunk from either the beginning or 1622 * the end 1623 */ 1624 static int shrink_vma(struct mm_struct *mm, 1625 struct vm_area_struct *vma, 1626 unsigned long from, unsigned long to) 1627 { 1628 struct vm_region *region; 1629 1630 kenter(""); 1631 1632 /* adjust the VMA's pointers, which may reposition it in the MM's tree 1633 * and list */ 1634 delete_vma_from_mm(vma); 1635 if (from > vma->vm_start) 1636 vma->vm_end = from; 1637 else 1638 vma->vm_start = to; 1639 add_vma_to_mm(mm, vma); 1640 1641 /* cut the backing region down to size */ 1642 region = vma->vm_region; 1643 BUG_ON(region->vm_usage != 1); 1644 1645 down_write(&nommu_region_sem); 1646 delete_nommu_region(region); 1647 if (from > region->vm_start) { 1648 to = region->vm_top; 1649 region->vm_top = region->vm_end = from; 1650 } else { 1651 region->vm_start = to; 1652 } 1653 add_nommu_region(region); 1654 up_write(&nommu_region_sem); 1655 1656 free_page_series(from, to); 1657 return 0; 1658 } 1659 1660 /* 1661 * release a mapping 1662 * - under NOMMU conditions the chunk to be unmapped must be backed by a single 1663 * VMA, though it need not cover the whole VMA 1664 */ 1665 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 1666 { 1667 struct vm_area_struct *vma; 1668 unsigned long end; 1669 int ret; 1670 1671 kenter(",%lx,%zx", start, len); 1672 1673 len = PAGE_ALIGN(len); 1674 if (len == 0) 1675 return -EINVAL; 1676 1677 end = start + len; 1678 1679 /* find the first potentially overlapping VMA */ 1680 vma = find_vma(mm, start); 1681 if (!vma) { 1682 static int limit; 1683 if (limit < 5) { 1684 printk(KERN_WARNING 1685 "munmap of memory not mmapped by process %d" 1686 " (%s): 0x%lx-0x%lx\n", 1687 current->pid, current->comm, 1688 start, start + len - 1); 1689 limit++; 1690 } 1691 return -EINVAL; 1692 } 1693 1694 /* we're allowed to split an anonymous VMA but not a file-backed one */ 1695 if (vma->vm_file) { 1696 do { 1697 if (start > vma->vm_start) { 1698 kleave(" = -EINVAL [miss]"); 1699 return -EINVAL; 1700 } 1701 if (end == vma->vm_end) 1702 goto erase_whole_vma; 1703 vma = vma->vm_next; 1704 } while (vma); 1705 kleave(" = -EINVAL [split file]"); 1706 return -EINVAL; 1707 } else { 1708 /* the chunk must be a subset of the VMA found */ 1709 if (start == vma->vm_start && end == vma->vm_end) 1710 goto erase_whole_vma; 1711 if (start < vma->vm_start || end > vma->vm_end) { 1712 kleave(" = -EINVAL [superset]"); 1713 return -EINVAL; 1714 } 1715 if (start & ~PAGE_MASK) { 1716 kleave(" = -EINVAL [unaligned start]"); 1717 return -EINVAL; 1718 } 1719 if (end != vma->vm_end && end & ~PAGE_MASK) { 1720 kleave(" = -EINVAL [unaligned split]"); 1721 return -EINVAL; 1722 } 1723 if (start != vma->vm_start && end != vma->vm_end) { 1724 ret = split_vma(mm, vma, start, 1); 1725 if (ret < 0) { 1726 kleave(" = %d [split]", ret); 1727 return ret; 1728 } 1729 } 1730 return shrink_vma(mm, vma, start, end); 1731 } 1732 1733 erase_whole_vma: 1734 delete_vma_from_mm(vma); 1735 delete_vma(mm, vma); 1736 kleave(" = 0"); 1737 return 0; 1738 } 1739 EXPORT_SYMBOL(do_munmap); 1740 1741 int vm_munmap(unsigned long addr, size_t len) 1742 { 1743 struct mm_struct *mm = current->mm; 1744 int ret; 1745 1746 down_write(&mm->mmap_sem); 1747 ret = do_munmap(mm, addr, len); 1748 up_write(&mm->mmap_sem); 1749 return ret; 1750 } 1751 EXPORT_SYMBOL(vm_munmap); 1752 1753 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 1754 { 1755 return vm_munmap(addr, len); 1756 } 1757 1758 /* 1759 * release all the mappings made in a process's VM space 1760 */ 1761 void exit_mmap(struct mm_struct *mm) 1762 { 1763 struct vm_area_struct *vma; 1764 1765 if (!mm) 1766 return; 1767 1768 kenter(""); 1769 1770 mm->total_vm = 0; 1771 1772 while ((vma = mm->mmap)) { 1773 mm->mmap = vma->vm_next; 1774 delete_vma_from_mm(vma); 1775 delete_vma(mm, vma); 1776 cond_resched(); 1777 } 1778 1779 kleave(""); 1780 } 1781 1782 unsigned long vm_brk(unsigned long addr, unsigned long len) 1783 { 1784 return -ENOMEM; 1785 } 1786 1787 /* 1788 * expand (or shrink) an existing mapping, potentially moving it at the same 1789 * time (controlled by the MREMAP_MAYMOVE flag and available VM space) 1790 * 1791 * under NOMMU conditions, we only permit changing a mapping's size, and only 1792 * as long as it stays within the region allocated by do_mmap_private() and the 1793 * block is not shareable 1794 * 1795 * MREMAP_FIXED is not supported under NOMMU conditions 1796 */ 1797 static unsigned long do_mremap(unsigned long addr, 1798 unsigned long old_len, unsigned long new_len, 1799 unsigned long flags, unsigned long new_addr) 1800 { 1801 struct vm_area_struct *vma; 1802 1803 /* insanity checks first */ 1804 old_len = PAGE_ALIGN(old_len); 1805 new_len = PAGE_ALIGN(new_len); 1806 if (old_len == 0 || new_len == 0) 1807 return (unsigned long) -EINVAL; 1808 1809 if (addr & ~PAGE_MASK) 1810 return -EINVAL; 1811 1812 if (flags & MREMAP_FIXED && new_addr != addr) 1813 return (unsigned long) -EINVAL; 1814 1815 vma = find_vma_exact(current->mm, addr, old_len); 1816 if (!vma) 1817 return (unsigned long) -EINVAL; 1818 1819 if (vma->vm_end != vma->vm_start + old_len) 1820 return (unsigned long) -EFAULT; 1821 1822 if (vma->vm_flags & VM_MAYSHARE) 1823 return (unsigned long) -EPERM; 1824 1825 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) 1826 return (unsigned long) -ENOMEM; 1827 1828 /* all checks complete - do it */ 1829 vma->vm_end = vma->vm_start + new_len; 1830 return vma->vm_start; 1831 } 1832 1833 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, 1834 unsigned long, new_len, unsigned long, flags, 1835 unsigned long, new_addr) 1836 { 1837 unsigned long ret; 1838 1839 down_write(¤t->mm->mmap_sem); 1840 ret = do_mremap(addr, old_len, new_len, flags, new_addr); 1841 up_write(¤t->mm->mmap_sem); 1842 return ret; 1843 } 1844 1845 struct page *follow_page_mask(struct vm_area_struct *vma, 1846 unsigned long address, unsigned int flags, 1847 unsigned int *page_mask) 1848 { 1849 *page_mask = 0; 1850 return NULL; 1851 } 1852 1853 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 1854 unsigned long pfn, unsigned long size, pgprot_t prot) 1855 { 1856 if (addr != (pfn << PAGE_SHIFT)) 1857 return -EINVAL; 1858 1859 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 1860 return 0; 1861 } 1862 EXPORT_SYMBOL(remap_pfn_range); 1863 1864 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 1865 { 1866 unsigned long pfn = start >> PAGE_SHIFT; 1867 unsigned long vm_len = vma->vm_end - vma->vm_start; 1868 1869 pfn += vma->vm_pgoff; 1870 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 1871 } 1872 EXPORT_SYMBOL(vm_iomap_memory); 1873 1874 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, 1875 unsigned long pgoff) 1876 { 1877 unsigned int size = vma->vm_end - vma->vm_start; 1878 1879 if (!(vma->vm_flags & VM_USERMAP)) 1880 return -EINVAL; 1881 1882 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); 1883 vma->vm_end = vma->vm_start + size; 1884 1885 return 0; 1886 } 1887 EXPORT_SYMBOL(remap_vmalloc_range); 1888 1889 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, 1890 unsigned long len, unsigned long pgoff, unsigned long flags) 1891 { 1892 return -ENOMEM; 1893 } 1894 1895 void unmap_mapping_range(struct address_space *mapping, 1896 loff_t const holebegin, loff_t const holelen, 1897 int even_cows) 1898 { 1899 } 1900 EXPORT_SYMBOL(unmap_mapping_range); 1901 1902 /* 1903 * Check that a process has enough memory to allocate a new virtual 1904 * mapping. 0 means there is enough memory for the allocation to 1905 * succeed and -ENOMEM implies there is not. 1906 * 1907 * We currently support three overcommit policies, which are set via the 1908 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1909 * 1910 * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1911 * Additional code 2002 Jul 20 by Robert Love. 1912 * 1913 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1914 * 1915 * Note this is a helper function intended to be used by LSMs which 1916 * wish to use this logic. 1917 */ 1918 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1919 { 1920 long free, allowed, reserve; 1921 1922 vm_acct_memory(pages); 1923 1924 /* 1925 * Sometimes we want to use more memory than we have 1926 */ 1927 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1928 return 0; 1929 1930 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1931 free = global_page_state(NR_FREE_PAGES); 1932 free += global_page_state(NR_FILE_PAGES); 1933 1934 /* 1935 * shmem pages shouldn't be counted as free in this 1936 * case, they can't be purged, only swapped out, and 1937 * that won't affect the overall amount of available 1938 * memory in the system. 1939 */ 1940 free -= global_page_state(NR_SHMEM); 1941 1942 free += get_nr_swap_pages(); 1943 1944 /* 1945 * Any slabs which are created with the 1946 * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1947 * which are reclaimable, under pressure. The dentry 1948 * cache and most inode caches should fall into this 1949 */ 1950 free += global_page_state(NR_SLAB_RECLAIMABLE); 1951 1952 /* 1953 * Leave reserved pages. The pages are not for anonymous pages. 1954 */ 1955 if (free <= totalreserve_pages) 1956 goto error; 1957 else 1958 free -= totalreserve_pages; 1959 1960 /* 1961 * Reserve some for root 1962 */ 1963 if (!cap_sys_admin) 1964 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 1965 1966 if (free > pages) 1967 return 0; 1968 1969 goto error; 1970 } 1971 1972 allowed = vm_commit_limit(); 1973 /* 1974 * Reserve some 3% for root 1975 */ 1976 if (!cap_sys_admin) 1977 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 1978 1979 /* 1980 * Don't let a single process grow so big a user can't recover 1981 */ 1982 if (mm) { 1983 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); 1984 allowed -= min_t(long, mm->total_vm / 32, reserve); 1985 } 1986 1987 if (percpu_counter_read_positive(&vm_committed_as) < allowed) 1988 return 0; 1989 1990 error: 1991 vm_unacct_memory(pages); 1992 1993 return -ENOMEM; 1994 } 1995 1996 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1997 { 1998 BUG(); 1999 return 0; 2000 } 2001 EXPORT_SYMBOL(filemap_fault); 2002 2003 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) 2004 { 2005 BUG(); 2006 } 2007 EXPORT_SYMBOL(filemap_map_pages); 2008 2009 static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 2010 unsigned long addr, void *buf, int len, int write) 2011 { 2012 struct vm_area_struct *vma; 2013 2014 down_read(&mm->mmap_sem); 2015 2016 /* the access must start within one of the target process's mappings */ 2017 vma = find_vma(mm, addr); 2018 if (vma) { 2019 /* don't overrun this mapping */ 2020 if (addr + len >= vma->vm_end) 2021 len = vma->vm_end - addr; 2022 2023 /* only read or write mappings where it is permitted */ 2024 if (write && vma->vm_flags & VM_MAYWRITE) 2025 copy_to_user_page(vma, NULL, addr, 2026 (void *) addr, buf, len); 2027 else if (!write && vma->vm_flags & VM_MAYREAD) 2028 copy_from_user_page(vma, NULL, addr, 2029 buf, (void *) addr, len); 2030 else 2031 len = 0; 2032 } else { 2033 len = 0; 2034 } 2035 2036 up_read(&mm->mmap_sem); 2037 2038 return len; 2039 } 2040 2041 /** 2042 * @access_remote_vm - access another process' address space 2043 * @mm: the mm_struct of the target address space 2044 * @addr: start address to access 2045 * @buf: source or destination buffer 2046 * @len: number of bytes to transfer 2047 * @write: whether the access is a write 2048 * 2049 * The caller must hold a reference on @mm. 2050 */ 2051 int access_remote_vm(struct mm_struct *mm, unsigned long addr, 2052 void *buf, int len, int write) 2053 { 2054 return __access_remote_vm(NULL, mm, addr, buf, len, write); 2055 } 2056 2057 /* 2058 * Access another process' address space. 2059 * - source/target buffer must be kernel space 2060 */ 2061 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) 2062 { 2063 struct mm_struct *mm; 2064 2065 if (addr + len < addr) 2066 return 0; 2067 2068 mm = get_task_mm(tsk); 2069 if (!mm) 2070 return 0; 2071 2072 len = __access_remote_vm(tsk, mm, addr, buf, len, write); 2073 2074 mmput(mm); 2075 return len; 2076 } 2077 2078 /** 2079 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode 2080 * @inode: The inode to check 2081 * @size: The current filesize of the inode 2082 * @newsize: The proposed filesize of the inode 2083 * 2084 * Check the shared mappings on an inode on behalf of a shrinking truncate to 2085 * make sure that that any outstanding VMAs aren't broken and then shrink the 2086 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't 2087 * automatically grant mappings that are too large. 2088 */ 2089 int nommu_shrink_inode_mappings(struct inode *inode, size_t size, 2090 size_t newsize) 2091 { 2092 struct vm_area_struct *vma; 2093 struct vm_region *region; 2094 pgoff_t low, high; 2095 size_t r_size, r_top; 2096 2097 low = newsize >> PAGE_SHIFT; 2098 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; 2099 2100 down_write(&nommu_region_sem); 2101 i_mmap_lock_read(inode->i_mapping); 2102 2103 /* search for VMAs that fall within the dead zone */ 2104 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) { 2105 /* found one - only interested if it's shared out of the page 2106 * cache */ 2107 if (vma->vm_flags & VM_SHARED) { 2108 i_mmap_unlock_read(inode->i_mapping); 2109 up_write(&nommu_region_sem); 2110 return -ETXTBSY; /* not quite true, but near enough */ 2111 } 2112 } 2113 2114 /* reduce any regions that overlap the dead zone - if in existence, 2115 * these will be pointed to by VMAs that don't overlap the dead zone 2116 * 2117 * we don't check for any regions that start beyond the EOF as there 2118 * shouldn't be any 2119 */ 2120 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) { 2121 if (!(vma->vm_flags & VM_SHARED)) 2122 continue; 2123 2124 region = vma->vm_region; 2125 r_size = region->vm_top - region->vm_start; 2126 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; 2127 2128 if (r_top > newsize) { 2129 region->vm_top -= r_top - newsize; 2130 if (region->vm_end > region->vm_top) 2131 region->vm_end = region->vm_top; 2132 } 2133 } 2134 2135 i_mmap_unlock_read(inode->i_mapping); 2136 up_write(&nommu_region_sem); 2137 return 0; 2138 } 2139 2140 /* 2141 * Initialise sysctl_user_reserve_kbytes. 2142 * 2143 * This is intended to prevent a user from starting a single memory hogging 2144 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 2145 * mode. 2146 * 2147 * The default value is min(3% of free memory, 128MB) 2148 * 128MB is enough to recover with sshd/login, bash, and top/kill. 2149 */ 2150 static int __meminit init_user_reserve(void) 2151 { 2152 unsigned long free_kbytes; 2153 2154 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 2155 2156 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 2157 return 0; 2158 } 2159 module_init(init_user_reserve) 2160 2161 /* 2162 * Initialise sysctl_admin_reserve_kbytes. 2163 * 2164 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 2165 * to log in and kill a memory hogging process. 2166 * 2167 * Systems with more than 256MB will reserve 8MB, enough to recover 2168 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 2169 * only reserve 3% of free pages by default. 2170 */ 2171 static int __meminit init_admin_reserve(void) 2172 { 2173 unsigned long free_kbytes; 2174 2175 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 2176 2177 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 2178 return 0; 2179 } 2180 module_init(init_admin_reserve) 2181