11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/mm/nommu.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Replacement code for mm functions to support CPU's that don't 51da177e4SLinus Torvalds * have any form of memory management unit (thus no virtual memory). 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * See Documentation/nommu-mmap.txt 81da177e4SLinus Torvalds * 98feae131SDavid Howells * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com> 101da177e4SLinus Torvalds * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com> 111da177e4SLinus Torvalds * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org> 121da177e4SLinus Torvalds * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com> 13eb6434d9SPaul Mundt * Copyright (c) 2007-2009 Paul Mundt <lethal@linux-sh.org> 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 16f2b8544fSDavid Howells #include <linux/module.h> 171da177e4SLinus Torvalds #include <linux/mm.h> 181da177e4SLinus Torvalds #include <linux/mman.h> 191da177e4SLinus Torvalds #include <linux/swap.h> 201da177e4SLinus Torvalds #include <linux/file.h> 211da177e4SLinus Torvalds #include <linux/highmem.h> 221da177e4SLinus Torvalds #include <linux/pagemap.h> 231da177e4SLinus Torvalds #include <linux/slab.h> 241da177e4SLinus Torvalds #include <linux/vmalloc.h> 25fa8e26ccSRoland McGrath #include <linux/tracehook.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 281da177e4SLinus Torvalds #include <linux/mount.h> 291da177e4SLinus Torvalds #include <linux/personality.h> 301da177e4SLinus Torvalds #include <linux/security.h> 311da177e4SLinus Torvalds #include <linux/syscalls.h> 321da177e4SLinus Torvalds 331da177e4SLinus Torvalds #include <asm/uaccess.h> 341da177e4SLinus Torvalds #include <asm/tlb.h> 351da177e4SLinus Torvalds #include <asm/tlbflush.h> 36eb8cdec4SBernd Schmidt #include <asm/mmu_context.h> 378feae131SDavid Howells #include "internal.h" 388feae131SDavid Howells 398feae131SDavid Howells static inline __attribute__((format(printf, 1, 2))) 408feae131SDavid Howells void no_printk(const char *fmt, ...) 418feae131SDavid Howells { 428feae131SDavid Howells } 438feae131SDavid Howells 448feae131SDavid Howells #if 0 458feae131SDavid Howells #define kenter(FMT, ...) \ 468feae131SDavid Howells printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 478feae131SDavid Howells #define kleave(FMT, ...) \ 488feae131SDavid Howells printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 498feae131SDavid Howells #define kdebug(FMT, ...) \ 508feae131SDavid Howells printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) 518feae131SDavid Howells #else 528feae131SDavid Howells #define kenter(FMT, ...) \ 538feae131SDavid Howells no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 548feae131SDavid Howells #define kleave(FMT, ...) \ 558feae131SDavid Howells no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 568feae131SDavid Howells #define kdebug(FMT, ...) \ 578feae131SDavid Howells no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) 588feae131SDavid Howells #endif 591da177e4SLinus Torvalds 601da177e4SLinus Torvalds void *high_memory; 611da177e4SLinus Torvalds struct page *mem_map; 621da177e4SLinus Torvalds unsigned long max_mapnr; 631da177e4SLinus Torvalds unsigned long num_physpages; 644266c97aSHugh Dickins unsigned long highest_memmap_pfn; 6500a62ce9SKOSAKI Motohiro struct percpu_counter vm_committed_as; 661da177e4SLinus Torvalds int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 671da177e4SLinus Torvalds int sysctl_overcommit_ratio = 50; /* default is 50% */ 681da177e4SLinus Torvalds int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; 69fc4d5c29SDavid Howells int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; 701da177e4SLinus Torvalds int heap_stack_gap = 0; 711da177e4SLinus Torvalds 7233e5d769SDavid Howells atomic_long_t mmap_pages_allocated; 738feae131SDavid Howells 741da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 756a04de6dSWu, Bryan EXPORT_SYMBOL(num_physpages); 761da177e4SLinus Torvalds 778feae131SDavid Howells /* list of mapped, potentially shareable regions */ 788feae131SDavid Howells static struct kmem_cache *vm_region_jar; 798feae131SDavid Howells struct rb_root nommu_region_tree = RB_ROOT; 808feae131SDavid Howells DECLARE_RWSEM(nommu_region_sem); 811da177e4SLinus Torvalds 82f0f37e2fSAlexey Dobriyan const struct vm_operations_struct generic_file_vm_ops = { 831da177e4SLinus Torvalds }; 841da177e4SLinus Torvalds 851da177e4SLinus Torvalds /* 861da177e4SLinus Torvalds * Return the total memory allocated for this pointer, not 871da177e4SLinus Torvalds * just what the caller asked for. 881da177e4SLinus Torvalds * 891da177e4SLinus Torvalds * Doesn't have to be accurate, i.e. may have races. 901da177e4SLinus Torvalds */ 911da177e4SLinus Torvalds unsigned int kobjsize(const void *objp) 921da177e4SLinus Torvalds { 931da177e4SLinus Torvalds struct page *page; 941da177e4SLinus Torvalds 954016a139SMichael Hennerich /* 964016a139SMichael Hennerich * If the object we have should not have ksize performed on it, 974016a139SMichael Hennerich * return size of 0 984016a139SMichael Hennerich */ 995a1603beSPaul Mundt if (!objp || !virt_addr_valid(objp)) 1006cfd53fcSPaul Mundt return 0; 1016cfd53fcSPaul Mundt 1026cfd53fcSPaul Mundt page = virt_to_head_page(objp); 1036cfd53fcSPaul Mundt 1046cfd53fcSPaul Mundt /* 1056cfd53fcSPaul Mundt * If the allocator sets PageSlab, we know the pointer came from 1066cfd53fcSPaul Mundt * kmalloc(). 1076cfd53fcSPaul Mundt */ 1081da177e4SLinus Torvalds if (PageSlab(page)) 1091da177e4SLinus Torvalds return ksize(objp); 1101da177e4SLinus Torvalds 1116cfd53fcSPaul Mundt /* 112ab2e83eaSPaul Mundt * If it's not a compound page, see if we have a matching VMA 113ab2e83eaSPaul Mundt * region. This test is intentionally done in reverse order, 114ab2e83eaSPaul Mundt * so if there's no VMA, we still fall through and hand back 115ab2e83eaSPaul Mundt * PAGE_SIZE for 0-order pages. 116ab2e83eaSPaul Mundt */ 117ab2e83eaSPaul Mundt if (!PageCompound(page)) { 118ab2e83eaSPaul Mundt struct vm_area_struct *vma; 119ab2e83eaSPaul Mundt 120ab2e83eaSPaul Mundt vma = find_vma(current->mm, (unsigned long)objp); 121ab2e83eaSPaul Mundt if (vma) 122ab2e83eaSPaul Mundt return vma->vm_end - vma->vm_start; 123ab2e83eaSPaul Mundt } 124ab2e83eaSPaul Mundt 125ab2e83eaSPaul Mundt /* 1266cfd53fcSPaul Mundt * The ksize() function is only guaranteed to work for pointers 1275a1603beSPaul Mundt * returned by kmalloc(). So handle arbitrary pointers here. 1286cfd53fcSPaul Mundt */ 1295a1603beSPaul Mundt return PAGE_SIZE << compound_order(page); 1301da177e4SLinus Torvalds } 1311da177e4SLinus Torvalds 132b291f000SNick Piggin int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 1334266c97aSHugh Dickins unsigned long start, int nr_pages, unsigned int foll_flags, 1341da177e4SLinus Torvalds struct page **pages, struct vm_area_struct **vmas) 1351da177e4SLinus Torvalds { 136910e46daSSonic Zhang struct vm_area_struct *vma; 1377b4d5b8bSDavid Howells unsigned long vm_flags; 1387b4d5b8bSDavid Howells int i; 1397b4d5b8bSDavid Howells 1407b4d5b8bSDavid Howells /* calculate required read or write permissions. 14158fa879eSHugh Dickins * If FOLL_FORCE is set, we only require the "MAY" flags. 1427b4d5b8bSDavid Howells */ 14358fa879eSHugh Dickins vm_flags = (foll_flags & FOLL_WRITE) ? 14458fa879eSHugh Dickins (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 14558fa879eSHugh Dickins vm_flags &= (foll_flags & FOLL_FORCE) ? 14658fa879eSHugh Dickins (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 1471da177e4SLinus Torvalds 1489d73777eSPeter Zijlstra for (i = 0; i < nr_pages; i++) { 149910e46daSSonic Zhang vma = find_vma(mm, start); 150910e46daSSonic Zhang if (!vma) 1517b4d5b8bSDavid Howells goto finish_or_fault; 1527b4d5b8bSDavid Howells 1537b4d5b8bSDavid Howells /* protect what we can, including chardevs */ 1541c3aff1cSHugh Dickins if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || 1551c3aff1cSHugh Dickins !(vm_flags & vma->vm_flags)) 1567b4d5b8bSDavid Howells goto finish_or_fault; 157910e46daSSonic Zhang 1581da177e4SLinus Torvalds if (pages) { 1591da177e4SLinus Torvalds pages[i] = virt_to_page(start); 1601da177e4SLinus Torvalds if (pages[i]) 1611da177e4SLinus Torvalds page_cache_get(pages[i]); 1621da177e4SLinus Torvalds } 1631da177e4SLinus Torvalds if (vmas) 164910e46daSSonic Zhang vmas[i] = vma; 1651da177e4SLinus Torvalds start += PAGE_SIZE; 1661da177e4SLinus Torvalds } 1677b4d5b8bSDavid Howells 1687b4d5b8bSDavid Howells return i; 1697b4d5b8bSDavid Howells 1707b4d5b8bSDavid Howells finish_or_fault: 1717b4d5b8bSDavid Howells return i ? : -EFAULT; 1721da177e4SLinus Torvalds } 173b291f000SNick Piggin 174b291f000SNick Piggin /* 175b291f000SNick Piggin * get a list of pages in an address range belonging to the specified process 176b291f000SNick Piggin * and indicate the VMA that covers each page 177b291f000SNick Piggin * - this is potentially dodgy as we may end incrementing the page count of a 178b291f000SNick Piggin * slab page or a secondary page from a compound page 179b291f000SNick Piggin * - don't permit access to VMAs that don't support it, such as I/O mappings 180b291f000SNick Piggin */ 181b291f000SNick Piggin int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 1829d73777eSPeter Zijlstra unsigned long start, int nr_pages, int write, int force, 183b291f000SNick Piggin struct page **pages, struct vm_area_struct **vmas) 184b291f000SNick Piggin { 185b291f000SNick Piggin int flags = 0; 186b291f000SNick Piggin 187b291f000SNick Piggin if (write) 18858fa879eSHugh Dickins flags |= FOLL_WRITE; 189b291f000SNick Piggin if (force) 19058fa879eSHugh Dickins flags |= FOLL_FORCE; 191b291f000SNick Piggin 1929d73777eSPeter Zijlstra return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas); 193b291f000SNick Piggin } 19466aa2b4bSGreg Ungerer EXPORT_SYMBOL(get_user_pages); 19566aa2b4bSGreg Ungerer 196dfc2f91aSPaul Mundt /** 197dfc2f91aSPaul Mundt * follow_pfn - look up PFN at a user virtual address 198dfc2f91aSPaul Mundt * @vma: memory mapping 199dfc2f91aSPaul Mundt * @address: user virtual address 200dfc2f91aSPaul Mundt * @pfn: location to store found PFN 201dfc2f91aSPaul Mundt * 202dfc2f91aSPaul Mundt * Only IO mappings and raw PFN mappings are allowed. 203dfc2f91aSPaul Mundt * 204dfc2f91aSPaul Mundt * Returns zero and the pfn at @pfn on success, -ve otherwise. 205dfc2f91aSPaul Mundt */ 206dfc2f91aSPaul Mundt int follow_pfn(struct vm_area_struct *vma, unsigned long address, 207dfc2f91aSPaul Mundt unsigned long *pfn) 208dfc2f91aSPaul Mundt { 209dfc2f91aSPaul Mundt if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 210dfc2f91aSPaul Mundt return -EINVAL; 211dfc2f91aSPaul Mundt 212dfc2f91aSPaul Mundt *pfn = address >> PAGE_SHIFT; 213dfc2f91aSPaul Mundt return 0; 214dfc2f91aSPaul Mundt } 215dfc2f91aSPaul Mundt EXPORT_SYMBOL(follow_pfn); 216dfc2f91aSPaul Mundt 2171da177e4SLinus Torvalds DEFINE_RWLOCK(vmlist_lock); 2181da177e4SLinus Torvalds struct vm_struct *vmlist; 2191da177e4SLinus Torvalds 220b3bdda02SChristoph Lameter void vfree(const void *addr) 2211da177e4SLinus Torvalds { 2221da177e4SLinus Torvalds kfree(addr); 2231da177e4SLinus Torvalds } 224b5073173SPaul Mundt EXPORT_SYMBOL(vfree); 2251da177e4SLinus Torvalds 226dd0fc66fSAl Viro void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 2271da177e4SLinus Torvalds { 2281da177e4SLinus Torvalds /* 2298518609dSRobert P. J. Day * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 2308518609dSRobert P. J. Day * returns only a logical address. 2311da177e4SLinus Torvalds */ 23284097518SNick Piggin return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 2331da177e4SLinus Torvalds } 234b5073173SPaul Mundt EXPORT_SYMBOL(__vmalloc); 2351da177e4SLinus Torvalds 236f905bc44SPaul Mundt void *vmalloc_user(unsigned long size) 237f905bc44SPaul Mundt { 238f905bc44SPaul Mundt void *ret; 239f905bc44SPaul Mundt 240f905bc44SPaul Mundt ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 241f905bc44SPaul Mundt PAGE_KERNEL); 242f905bc44SPaul Mundt if (ret) { 243f905bc44SPaul Mundt struct vm_area_struct *vma; 244f905bc44SPaul Mundt 245f905bc44SPaul Mundt down_write(¤t->mm->mmap_sem); 246f905bc44SPaul Mundt vma = find_vma(current->mm, (unsigned long)ret); 247f905bc44SPaul Mundt if (vma) 248f905bc44SPaul Mundt vma->vm_flags |= VM_USERMAP; 249f905bc44SPaul Mundt up_write(¤t->mm->mmap_sem); 250f905bc44SPaul Mundt } 251f905bc44SPaul Mundt 252f905bc44SPaul Mundt return ret; 253f905bc44SPaul Mundt } 254f905bc44SPaul Mundt EXPORT_SYMBOL(vmalloc_user); 255f905bc44SPaul Mundt 256b3bdda02SChristoph Lameter struct page *vmalloc_to_page(const void *addr) 2571da177e4SLinus Torvalds { 2581da177e4SLinus Torvalds return virt_to_page(addr); 2591da177e4SLinus Torvalds } 260b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_to_page); 2611da177e4SLinus Torvalds 262b3bdda02SChristoph Lameter unsigned long vmalloc_to_pfn(const void *addr) 2631da177e4SLinus Torvalds { 2641da177e4SLinus Torvalds return page_to_pfn(virt_to_page(addr)); 2651da177e4SLinus Torvalds } 266b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_to_pfn); 2671da177e4SLinus Torvalds 2681da177e4SLinus Torvalds long vread(char *buf, char *addr, unsigned long count) 2691da177e4SLinus Torvalds { 2701da177e4SLinus Torvalds memcpy(buf, addr, count); 2711da177e4SLinus Torvalds return count; 2721da177e4SLinus Torvalds } 2731da177e4SLinus Torvalds 2741da177e4SLinus Torvalds long vwrite(char *buf, char *addr, unsigned long count) 2751da177e4SLinus Torvalds { 2761da177e4SLinus Torvalds /* Don't allow overflow */ 2771da177e4SLinus Torvalds if ((unsigned long) addr + count < count) 2781da177e4SLinus Torvalds count = -(unsigned long) addr; 2791da177e4SLinus Torvalds 2801da177e4SLinus Torvalds memcpy(addr, buf, count); 2811da177e4SLinus Torvalds return(count); 2821da177e4SLinus Torvalds } 2831da177e4SLinus Torvalds 2841da177e4SLinus Torvalds /* 2851da177e4SLinus Torvalds * vmalloc - allocate virtually continguos memory 2861da177e4SLinus Torvalds * 2871da177e4SLinus Torvalds * @size: allocation size 2881da177e4SLinus Torvalds * 2891da177e4SLinus Torvalds * Allocate enough pages to cover @size from the page level 2901da177e4SLinus Torvalds * allocator and map them into continguos kernel virtual space. 2911da177e4SLinus Torvalds * 292c1c8897fSMichael Opdenacker * For tight control over page level allocator and protection flags 2931da177e4SLinus Torvalds * use __vmalloc() instead. 2941da177e4SLinus Torvalds */ 2951da177e4SLinus Torvalds void *vmalloc(unsigned long size) 2961da177e4SLinus Torvalds { 2971da177e4SLinus Torvalds return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 2981da177e4SLinus Torvalds } 299f6138882SAndrew Morton EXPORT_SYMBOL(vmalloc); 300f6138882SAndrew Morton 301f6138882SAndrew Morton void *vmalloc_node(unsigned long size, int node) 302f6138882SAndrew Morton { 303f6138882SAndrew Morton return vmalloc(size); 304f6138882SAndrew Morton } 305f6138882SAndrew Morton EXPORT_SYMBOL(vmalloc_node); 3061da177e4SLinus Torvalds 3071af446edSPaul Mundt #ifndef PAGE_KERNEL_EXEC 3081af446edSPaul Mundt # define PAGE_KERNEL_EXEC PAGE_KERNEL 3091af446edSPaul Mundt #endif 3101af446edSPaul Mundt 3111af446edSPaul Mundt /** 3121af446edSPaul Mundt * vmalloc_exec - allocate virtually contiguous, executable memory 3131af446edSPaul Mundt * @size: allocation size 3141af446edSPaul Mundt * 3151af446edSPaul Mundt * Kernel-internal function to allocate enough pages to cover @size 3161af446edSPaul Mundt * the page level allocator and map them into contiguous and 3171af446edSPaul Mundt * executable kernel virtual space. 3181af446edSPaul Mundt * 3191af446edSPaul Mundt * For tight control over page level allocator and protection flags 3201af446edSPaul Mundt * use __vmalloc() instead. 3211af446edSPaul Mundt */ 3221af446edSPaul Mundt 3231af446edSPaul Mundt void *vmalloc_exec(unsigned long size) 3241af446edSPaul Mundt { 3251af446edSPaul Mundt return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); 3261af446edSPaul Mundt } 3271af446edSPaul Mundt 328b5073173SPaul Mundt /** 329b5073173SPaul Mundt * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 3301da177e4SLinus Torvalds * @size: allocation size 3311da177e4SLinus Torvalds * 3321da177e4SLinus Torvalds * Allocate enough 32bit PA addressable pages to cover @size from the 3331da177e4SLinus Torvalds * page level allocator and map them into continguos kernel virtual space. 3341da177e4SLinus Torvalds */ 3351da177e4SLinus Torvalds void *vmalloc_32(unsigned long size) 3361da177e4SLinus Torvalds { 3371da177e4SLinus Torvalds return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 3381da177e4SLinus Torvalds } 339b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_32); 340b5073173SPaul Mundt 341b5073173SPaul Mundt /** 342b5073173SPaul Mundt * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 343b5073173SPaul Mundt * @size: allocation size 344b5073173SPaul Mundt * 345b5073173SPaul Mundt * The resulting memory area is 32bit addressable and zeroed so it can be 346b5073173SPaul Mundt * mapped to userspace without leaking data. 347f905bc44SPaul Mundt * 348f905bc44SPaul Mundt * VM_USERMAP is set on the corresponding VMA so that subsequent calls to 349f905bc44SPaul Mundt * remap_vmalloc_range() are permissible. 350b5073173SPaul Mundt */ 351b5073173SPaul Mundt void *vmalloc_32_user(unsigned long size) 352b5073173SPaul Mundt { 353f905bc44SPaul Mundt /* 354f905bc44SPaul Mundt * We'll have to sort out the ZONE_DMA bits for 64-bit, 355f905bc44SPaul Mundt * but for now this can simply use vmalloc_user() directly. 356f905bc44SPaul Mundt */ 357f905bc44SPaul Mundt return vmalloc_user(size); 358b5073173SPaul Mundt } 359b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_32_user); 3601da177e4SLinus Torvalds 3611da177e4SLinus Torvalds void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 3621da177e4SLinus Torvalds { 3631da177e4SLinus Torvalds BUG(); 3641da177e4SLinus Torvalds return NULL; 3651da177e4SLinus Torvalds } 366b5073173SPaul Mundt EXPORT_SYMBOL(vmap); 3671da177e4SLinus Torvalds 368b3bdda02SChristoph Lameter void vunmap(const void *addr) 3691da177e4SLinus Torvalds { 3701da177e4SLinus Torvalds BUG(); 3711da177e4SLinus Torvalds } 372b5073173SPaul Mundt EXPORT_SYMBOL(vunmap); 3731da177e4SLinus Torvalds 374eb6434d9SPaul Mundt void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) 375eb6434d9SPaul Mundt { 376eb6434d9SPaul Mundt BUG(); 377eb6434d9SPaul Mundt return NULL; 378eb6434d9SPaul Mundt } 379eb6434d9SPaul Mundt EXPORT_SYMBOL(vm_map_ram); 380eb6434d9SPaul Mundt 381eb6434d9SPaul Mundt void vm_unmap_ram(const void *mem, unsigned int count) 382eb6434d9SPaul Mundt { 383eb6434d9SPaul Mundt BUG(); 384eb6434d9SPaul Mundt } 385eb6434d9SPaul Mundt EXPORT_SYMBOL(vm_unmap_ram); 386eb6434d9SPaul Mundt 387eb6434d9SPaul Mundt void vm_unmap_aliases(void) 388eb6434d9SPaul Mundt { 389eb6434d9SPaul Mundt } 390eb6434d9SPaul Mundt EXPORT_SYMBOL_GPL(vm_unmap_aliases); 391eb6434d9SPaul Mundt 3921da177e4SLinus Torvalds /* 3931eeb66a1SChristoph Hellwig * Implement a stub for vmalloc_sync_all() if the architecture chose not to 3941eeb66a1SChristoph Hellwig * have one. 3951eeb66a1SChristoph Hellwig */ 3961eeb66a1SChristoph Hellwig void __attribute__((weak)) vmalloc_sync_all(void) 3971eeb66a1SChristoph Hellwig { 3981eeb66a1SChristoph Hellwig } 3991eeb66a1SChristoph Hellwig 400b5073173SPaul Mundt int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 401b5073173SPaul Mundt struct page *page) 402b5073173SPaul Mundt { 403b5073173SPaul Mundt return -EINVAL; 404b5073173SPaul Mundt } 405b5073173SPaul Mundt EXPORT_SYMBOL(vm_insert_page); 406b5073173SPaul Mundt 4071eeb66a1SChristoph Hellwig /* 4081da177e4SLinus Torvalds * sys_brk() for the most part doesn't need the global kernel 4091da177e4SLinus Torvalds * lock, except when an application is doing something nasty 4101da177e4SLinus Torvalds * like trying to un-brk an area that has already been mapped 4111da177e4SLinus Torvalds * to a regular file. in this case, the unmapping will need 4121da177e4SLinus Torvalds * to invoke file system routines that need the global lock. 4131da177e4SLinus Torvalds */ 4146a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 4151da177e4SLinus Torvalds { 4161da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 4171da177e4SLinus Torvalds 4181da177e4SLinus Torvalds if (brk < mm->start_brk || brk > mm->context.end_brk) 4191da177e4SLinus Torvalds return mm->brk; 4201da177e4SLinus Torvalds 4211da177e4SLinus Torvalds if (mm->brk == brk) 4221da177e4SLinus Torvalds return mm->brk; 4231da177e4SLinus Torvalds 4241da177e4SLinus Torvalds /* 4251da177e4SLinus Torvalds * Always allow shrinking brk 4261da177e4SLinus Torvalds */ 4271da177e4SLinus Torvalds if (brk <= mm->brk) { 4281da177e4SLinus Torvalds mm->brk = brk; 4291da177e4SLinus Torvalds return brk; 4301da177e4SLinus Torvalds } 4311da177e4SLinus Torvalds 4321da177e4SLinus Torvalds /* 4331da177e4SLinus Torvalds * Ok, looks good - let it rip. 4341da177e4SLinus Torvalds */ 435cfe79c00SMike Frysinger flush_icache_range(mm->brk, brk); 4361da177e4SLinus Torvalds return mm->brk = brk; 4371da177e4SLinus Torvalds } 4381da177e4SLinus Torvalds 4398feae131SDavid Howells /* 4408feae131SDavid Howells * initialise the VMA and region record slabs 4418feae131SDavid Howells */ 4428feae131SDavid Howells void __init mmap_init(void) 4431da177e4SLinus Torvalds { 44400a62ce9SKOSAKI Motohiro int ret; 44500a62ce9SKOSAKI Motohiro 44600a62ce9SKOSAKI Motohiro ret = percpu_counter_init(&vm_committed_as, 0); 44700a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 44833e5d769SDavid Howells vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); 4498feae131SDavid Howells } 4501da177e4SLinus Torvalds 4518feae131SDavid Howells /* 4528feae131SDavid Howells * validate the region tree 4538feae131SDavid Howells * - the caller must hold the region lock 4548feae131SDavid Howells */ 4558feae131SDavid Howells #ifdef CONFIG_DEBUG_NOMMU_REGIONS 4568feae131SDavid Howells static noinline void validate_nommu_regions(void) 4578feae131SDavid Howells { 4588feae131SDavid Howells struct vm_region *region, *last; 4598feae131SDavid Howells struct rb_node *p, *lastp; 4601da177e4SLinus Torvalds 4618feae131SDavid Howells lastp = rb_first(&nommu_region_tree); 4628feae131SDavid Howells if (!lastp) 4638feae131SDavid Howells return; 4648feae131SDavid Howells 4658feae131SDavid Howells last = rb_entry(lastp, struct vm_region, vm_rb); 46633e5d769SDavid Howells BUG_ON(unlikely(last->vm_end <= last->vm_start)); 46733e5d769SDavid Howells BUG_ON(unlikely(last->vm_top < last->vm_end)); 4688feae131SDavid Howells 4698feae131SDavid Howells while ((p = rb_next(lastp))) { 4708feae131SDavid Howells region = rb_entry(p, struct vm_region, vm_rb); 4718feae131SDavid Howells last = rb_entry(lastp, struct vm_region, vm_rb); 4728feae131SDavid Howells 47333e5d769SDavid Howells BUG_ON(unlikely(region->vm_end <= region->vm_start)); 47433e5d769SDavid Howells BUG_ON(unlikely(region->vm_top < region->vm_end)); 47533e5d769SDavid Howells BUG_ON(unlikely(region->vm_start < last->vm_top)); 4768feae131SDavid Howells 4778feae131SDavid Howells lastp = p; 4781da177e4SLinus Torvalds } 4791da177e4SLinus Torvalds } 4808feae131SDavid Howells #else 48133e5d769SDavid Howells static void validate_nommu_regions(void) 48233e5d769SDavid Howells { 48333e5d769SDavid Howells } 4848feae131SDavid Howells #endif 4858feae131SDavid Howells 4868feae131SDavid Howells /* 4878feae131SDavid Howells * add a region into the global tree 4888feae131SDavid Howells */ 4898feae131SDavid Howells static void add_nommu_region(struct vm_region *region) 4908feae131SDavid Howells { 4918feae131SDavid Howells struct vm_region *pregion; 4928feae131SDavid Howells struct rb_node **p, *parent; 4938feae131SDavid Howells 4948feae131SDavid Howells validate_nommu_regions(); 4958feae131SDavid Howells 4968feae131SDavid Howells parent = NULL; 4978feae131SDavid Howells p = &nommu_region_tree.rb_node; 4988feae131SDavid Howells while (*p) { 4998feae131SDavid Howells parent = *p; 5008feae131SDavid Howells pregion = rb_entry(parent, struct vm_region, vm_rb); 5018feae131SDavid Howells if (region->vm_start < pregion->vm_start) 5028feae131SDavid Howells p = &(*p)->rb_left; 5038feae131SDavid Howells else if (region->vm_start > pregion->vm_start) 5048feae131SDavid Howells p = &(*p)->rb_right; 5058feae131SDavid Howells else if (pregion == region) 5068feae131SDavid Howells return; 5078feae131SDavid Howells else 5088feae131SDavid Howells BUG(); 5098feae131SDavid Howells } 5108feae131SDavid Howells 5118feae131SDavid Howells rb_link_node(®ion->vm_rb, parent, p); 5128feae131SDavid Howells rb_insert_color(®ion->vm_rb, &nommu_region_tree); 5138feae131SDavid Howells 5148feae131SDavid Howells validate_nommu_regions(); 5158feae131SDavid Howells } 5168feae131SDavid Howells 5178feae131SDavid Howells /* 5188feae131SDavid Howells * delete a region from the global tree 5198feae131SDavid Howells */ 5208feae131SDavid Howells static void delete_nommu_region(struct vm_region *region) 5218feae131SDavid Howells { 5228feae131SDavid Howells BUG_ON(!nommu_region_tree.rb_node); 5238feae131SDavid Howells 5248feae131SDavid Howells validate_nommu_regions(); 5258feae131SDavid Howells rb_erase(®ion->vm_rb, &nommu_region_tree); 5268feae131SDavid Howells validate_nommu_regions(); 5278feae131SDavid Howells } 5288feae131SDavid Howells 5298feae131SDavid Howells /* 5308feae131SDavid Howells * free a contiguous series of pages 5318feae131SDavid Howells */ 5328feae131SDavid Howells static void free_page_series(unsigned long from, unsigned long to) 5338feae131SDavid Howells { 5348feae131SDavid Howells for (; from < to; from += PAGE_SIZE) { 5358feae131SDavid Howells struct page *page = virt_to_page(from); 5368feae131SDavid Howells 5378feae131SDavid Howells kdebug("- free %lx", from); 53833e5d769SDavid Howells atomic_long_dec(&mmap_pages_allocated); 5398feae131SDavid Howells if (page_count(page) != 1) 54033e5d769SDavid Howells kdebug("free page %p: refcount not one: %d", 54133e5d769SDavid Howells page, page_count(page)); 5428feae131SDavid Howells put_page(page); 5438feae131SDavid Howells } 5448feae131SDavid Howells } 5458feae131SDavid Howells 5468feae131SDavid Howells /* 5478feae131SDavid Howells * release a reference to a region 54833e5d769SDavid Howells * - the caller must hold the region semaphore for writing, which this releases 549dd8632a1SPaul Mundt * - the region may not have been added to the tree yet, in which case vm_top 5508feae131SDavid Howells * will equal vm_start 5518feae131SDavid Howells */ 5528feae131SDavid Howells static void __put_nommu_region(struct vm_region *region) 5538feae131SDavid Howells __releases(nommu_region_sem) 5548feae131SDavid Howells { 555*1e2ae599SDavid Howells kenter("%p{%d}", region, region->vm_usage); 5568feae131SDavid Howells 5578feae131SDavid Howells BUG_ON(!nommu_region_tree.rb_node); 5588feae131SDavid Howells 559*1e2ae599SDavid Howells if (--region->vm_usage == 0) { 560dd8632a1SPaul Mundt if (region->vm_top > region->vm_start) 5618feae131SDavid Howells delete_nommu_region(region); 5628feae131SDavid Howells up_write(&nommu_region_sem); 5638feae131SDavid Howells 5648feae131SDavid Howells if (region->vm_file) 5658feae131SDavid Howells fput(region->vm_file); 5668feae131SDavid Howells 5678feae131SDavid Howells /* IO memory and memory shared directly out of the pagecache 5688feae131SDavid Howells * from ramfs/tmpfs mustn't be released here */ 5698feae131SDavid Howells if (region->vm_flags & VM_MAPPED_COPY) { 5708feae131SDavid Howells kdebug("free series"); 571dd8632a1SPaul Mundt free_page_series(region->vm_start, region->vm_top); 5728feae131SDavid Howells } 5738feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 5748feae131SDavid Howells } else { 5758feae131SDavid Howells up_write(&nommu_region_sem); 5768feae131SDavid Howells } 5778feae131SDavid Howells } 5788feae131SDavid Howells 5798feae131SDavid Howells /* 5808feae131SDavid Howells * release a reference to a region 5818feae131SDavid Howells */ 5828feae131SDavid Howells static void put_nommu_region(struct vm_region *region) 5838feae131SDavid Howells { 5848feae131SDavid Howells down_write(&nommu_region_sem); 5858feae131SDavid Howells __put_nommu_region(region); 5868feae131SDavid Howells } 5871da177e4SLinus Torvalds 5883034097aSDavid Howells /* 589eb8cdec4SBernd Schmidt * update protection on a vma 590eb8cdec4SBernd Schmidt */ 591eb8cdec4SBernd Schmidt static void protect_vma(struct vm_area_struct *vma, unsigned long flags) 592eb8cdec4SBernd Schmidt { 593eb8cdec4SBernd Schmidt #ifdef CONFIG_MPU 594eb8cdec4SBernd Schmidt struct mm_struct *mm = vma->vm_mm; 595eb8cdec4SBernd Schmidt long start = vma->vm_start & PAGE_MASK; 596eb8cdec4SBernd Schmidt while (start < vma->vm_end) { 597eb8cdec4SBernd Schmidt protect_page(mm, start, flags); 598eb8cdec4SBernd Schmidt start += PAGE_SIZE; 599eb8cdec4SBernd Schmidt } 600eb8cdec4SBernd Schmidt update_protections(mm); 601eb8cdec4SBernd Schmidt #endif 602eb8cdec4SBernd Schmidt } 603eb8cdec4SBernd Schmidt 604eb8cdec4SBernd Schmidt /* 6053034097aSDavid Howells * add a VMA into a process's mm_struct in the appropriate place in the list 6068feae131SDavid Howells * and tree and add to the address space's page tree also if not an anonymous 6078feae131SDavid Howells * page 6083034097aSDavid Howells * - should be called with mm->mmap_sem held writelocked 6093034097aSDavid Howells */ 6108feae131SDavid Howells static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) 6113034097aSDavid Howells { 6128feae131SDavid Howells struct vm_area_struct *pvma, **pp; 6138feae131SDavid Howells struct address_space *mapping; 6148feae131SDavid Howells struct rb_node **p, *parent; 6153034097aSDavid Howells 6168feae131SDavid Howells kenter(",%p", vma); 6178feae131SDavid Howells 6188feae131SDavid Howells BUG_ON(!vma->vm_region); 6198feae131SDavid Howells 6208feae131SDavid Howells mm->map_count++; 6218feae131SDavid Howells vma->vm_mm = mm; 6228feae131SDavid Howells 623eb8cdec4SBernd Schmidt protect_vma(vma, vma->vm_flags); 624eb8cdec4SBernd Schmidt 6258feae131SDavid Howells /* add the VMA to the mapping */ 6268feae131SDavid Howells if (vma->vm_file) { 6278feae131SDavid Howells mapping = vma->vm_file->f_mapping; 6288feae131SDavid Howells 6298feae131SDavid Howells flush_dcache_mmap_lock(mapping); 6308feae131SDavid Howells vma_prio_tree_insert(vma, &mapping->i_mmap); 6318feae131SDavid Howells flush_dcache_mmap_unlock(mapping); 6328feae131SDavid Howells } 6338feae131SDavid Howells 6348feae131SDavid Howells /* add the VMA to the tree */ 6358feae131SDavid Howells parent = NULL; 6368feae131SDavid Howells p = &mm->mm_rb.rb_node; 6378feae131SDavid Howells while (*p) { 6388feae131SDavid Howells parent = *p; 6398feae131SDavid Howells pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 6408feae131SDavid Howells 6418feae131SDavid Howells /* sort by: start addr, end addr, VMA struct addr in that order 6428feae131SDavid Howells * (the latter is necessary as we may get identical VMAs) */ 6438feae131SDavid Howells if (vma->vm_start < pvma->vm_start) 6448feae131SDavid Howells p = &(*p)->rb_left; 6458feae131SDavid Howells else if (vma->vm_start > pvma->vm_start) 6468feae131SDavid Howells p = &(*p)->rb_right; 6478feae131SDavid Howells else if (vma->vm_end < pvma->vm_end) 6488feae131SDavid Howells p = &(*p)->rb_left; 6498feae131SDavid Howells else if (vma->vm_end > pvma->vm_end) 6508feae131SDavid Howells p = &(*p)->rb_right; 6518feae131SDavid Howells else if (vma < pvma) 6528feae131SDavid Howells p = &(*p)->rb_left; 6538feae131SDavid Howells else if (vma > pvma) 6548feae131SDavid Howells p = &(*p)->rb_right; 6558feae131SDavid Howells else 6568feae131SDavid Howells BUG(); 6578feae131SDavid Howells } 6588feae131SDavid Howells 6598feae131SDavid Howells rb_link_node(&vma->vm_rb, parent, p); 6608feae131SDavid Howells rb_insert_color(&vma->vm_rb, &mm->mm_rb); 6618feae131SDavid Howells 6628feae131SDavid Howells /* add VMA to the VMA list also */ 6638feae131SDavid Howells for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) { 6648feae131SDavid Howells if (pvma->vm_start > vma->vm_start) 6653034097aSDavid Howells break; 6668feae131SDavid Howells if (pvma->vm_start < vma->vm_start) 6678feae131SDavid Howells continue; 6688feae131SDavid Howells if (pvma->vm_end < vma->vm_end) 6698feae131SDavid Howells break; 6708feae131SDavid Howells } 6713034097aSDavid Howells 6728feae131SDavid Howells vma->vm_next = *pp; 6738feae131SDavid Howells *pp = vma; 6748feae131SDavid Howells } 6758feae131SDavid Howells 6768feae131SDavid Howells /* 6778feae131SDavid Howells * delete a VMA from its owning mm_struct and address space 6788feae131SDavid Howells */ 6798feae131SDavid Howells static void delete_vma_from_mm(struct vm_area_struct *vma) 6808feae131SDavid Howells { 6818feae131SDavid Howells struct vm_area_struct **pp; 6828feae131SDavid Howells struct address_space *mapping; 6838feae131SDavid Howells struct mm_struct *mm = vma->vm_mm; 6848feae131SDavid Howells 6858feae131SDavid Howells kenter("%p", vma); 6868feae131SDavid Howells 687eb8cdec4SBernd Schmidt protect_vma(vma, 0); 688eb8cdec4SBernd Schmidt 6898feae131SDavid Howells mm->map_count--; 6908feae131SDavid Howells if (mm->mmap_cache == vma) 6918feae131SDavid Howells mm->mmap_cache = NULL; 6928feae131SDavid Howells 6938feae131SDavid Howells /* remove the VMA from the mapping */ 6948feae131SDavid Howells if (vma->vm_file) { 6958feae131SDavid Howells mapping = vma->vm_file->f_mapping; 6968feae131SDavid Howells 6978feae131SDavid Howells flush_dcache_mmap_lock(mapping); 6988feae131SDavid Howells vma_prio_tree_remove(vma, &mapping->i_mmap); 6998feae131SDavid Howells flush_dcache_mmap_unlock(mapping); 7008feae131SDavid Howells } 7018feae131SDavid Howells 7028feae131SDavid Howells /* remove from the MM's tree and list */ 7038feae131SDavid Howells rb_erase(&vma->vm_rb, &mm->mm_rb); 7048feae131SDavid Howells for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) { 7058feae131SDavid Howells if (*pp == vma) { 7068feae131SDavid Howells *pp = vma->vm_next; 7078feae131SDavid Howells break; 7088feae131SDavid Howells } 7098feae131SDavid Howells } 7108feae131SDavid Howells 7118feae131SDavid Howells vma->vm_mm = NULL; 7128feae131SDavid Howells } 7138feae131SDavid Howells 7148feae131SDavid Howells /* 7158feae131SDavid Howells * destroy a VMA record 7168feae131SDavid Howells */ 7178feae131SDavid Howells static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) 7188feae131SDavid Howells { 7198feae131SDavid Howells kenter("%p", vma); 7208feae131SDavid Howells if (vma->vm_ops && vma->vm_ops->close) 7218feae131SDavid Howells vma->vm_ops->close(vma); 7228feae131SDavid Howells if (vma->vm_file) { 7238feae131SDavid Howells fput(vma->vm_file); 7248feae131SDavid Howells if (vma->vm_flags & VM_EXECUTABLE) 7258feae131SDavid Howells removed_exe_file_vma(mm); 7268feae131SDavid Howells } 7278feae131SDavid Howells put_nommu_region(vma->vm_region); 7288feae131SDavid Howells kmem_cache_free(vm_area_cachep, vma); 7293034097aSDavid Howells } 7303034097aSDavid Howells 7313034097aSDavid Howells /* 7323034097aSDavid Howells * look up the first VMA in which addr resides, NULL if none 7333034097aSDavid Howells * - should be called with mm->mmap_sem at least held readlocked 7343034097aSDavid Howells */ 7353034097aSDavid Howells struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 7363034097aSDavid Howells { 7378feae131SDavid Howells struct vm_area_struct *vma; 7388feae131SDavid Howells struct rb_node *n = mm->mm_rb.rb_node; 7393034097aSDavid Howells 7408feae131SDavid Howells /* check the cache first */ 7418feae131SDavid Howells vma = mm->mmap_cache; 7428feae131SDavid Howells if (vma && vma->vm_start <= addr && vma->vm_end > addr) 7438feae131SDavid Howells return vma; 7448feae131SDavid Howells 7458feae131SDavid Howells /* trawl the tree (there may be multiple mappings in which addr 7468feae131SDavid Howells * resides) */ 7478feae131SDavid Howells for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { 7488feae131SDavid Howells vma = rb_entry(n, struct vm_area_struct, vm_rb); 7498feae131SDavid Howells if (vma->vm_start > addr) 7508feae131SDavid Howells return NULL; 7518feae131SDavid Howells if (vma->vm_end > addr) { 7528feae131SDavid Howells mm->mmap_cache = vma; 7538feae131SDavid Howells return vma; 7543034097aSDavid Howells } 7558feae131SDavid Howells } 7563034097aSDavid Howells 7573034097aSDavid Howells return NULL; 7583034097aSDavid Howells } 7593034097aSDavid Howells EXPORT_SYMBOL(find_vma); 7603034097aSDavid Howells 7613034097aSDavid Howells /* 762930e652aSDavid Howells * find a VMA 763930e652aSDavid Howells * - we don't extend stack VMAs under NOMMU conditions 764930e652aSDavid Howells */ 765930e652aSDavid Howells struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) 766930e652aSDavid Howells { 767930e652aSDavid Howells return find_vma(mm, addr); 768930e652aSDavid Howells } 769930e652aSDavid Howells 7708feae131SDavid Howells /* 7718feae131SDavid Howells * expand a stack to a given address 7728feae131SDavid Howells * - not supported under NOMMU conditions 7738feae131SDavid Howells */ 77457c8f63eSGreg Ungerer int expand_stack(struct vm_area_struct *vma, unsigned long address) 77557c8f63eSGreg Ungerer { 77657c8f63eSGreg Ungerer return -ENOMEM; 77757c8f63eSGreg Ungerer } 77857c8f63eSGreg Ungerer 779930e652aSDavid Howells /* 7806fa5f80bSDavid Howells * look up the first VMA exactly that exactly matches addr 7816fa5f80bSDavid Howells * - should be called with mm->mmap_sem at least held readlocked 7826fa5f80bSDavid Howells */ 7838feae131SDavid Howells static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, 7848feae131SDavid Howells unsigned long addr, 7858feae131SDavid Howells unsigned long len) 7861da177e4SLinus Torvalds { 7871da177e4SLinus Torvalds struct vm_area_struct *vma; 7888feae131SDavid Howells struct rb_node *n = mm->mm_rb.rb_node; 7898feae131SDavid Howells unsigned long end = addr + len; 7901da177e4SLinus Torvalds 7918feae131SDavid Howells /* check the cache first */ 7928feae131SDavid Howells vma = mm->mmap_cache; 7938feae131SDavid Howells if (vma && vma->vm_start == addr && vma->vm_end == end) 7941da177e4SLinus Torvalds return vma; 7958feae131SDavid Howells 7968feae131SDavid Howells /* trawl the tree (there may be multiple mappings in which addr 7978feae131SDavid Howells * resides) */ 7988feae131SDavid Howells for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { 7998feae131SDavid Howells vma = rb_entry(n, struct vm_area_struct, vm_rb); 8008feae131SDavid Howells if (vma->vm_start < addr) 8018feae131SDavid Howells continue; 8028feae131SDavid Howells if (vma->vm_start > addr) 8038feae131SDavid Howells return NULL; 8048feae131SDavid Howells if (vma->vm_end == end) { 8058feae131SDavid Howells mm->mmap_cache = vma; 8068feae131SDavid Howells return vma; 8078feae131SDavid Howells } 8081da177e4SLinus Torvalds } 8091da177e4SLinus Torvalds 8101da177e4SLinus Torvalds return NULL; 8111da177e4SLinus Torvalds } 8121da177e4SLinus Torvalds 8133034097aSDavid Howells /* 8141da177e4SLinus Torvalds * determine whether a mapping should be permitted and, if so, what sort of 8151da177e4SLinus Torvalds * mapping we're capable of supporting 8161da177e4SLinus Torvalds */ 8171da177e4SLinus Torvalds static int validate_mmap_request(struct file *file, 8181da177e4SLinus Torvalds unsigned long addr, 8191da177e4SLinus Torvalds unsigned long len, 8201da177e4SLinus Torvalds unsigned long prot, 8211da177e4SLinus Torvalds unsigned long flags, 8221da177e4SLinus Torvalds unsigned long pgoff, 8231da177e4SLinus Torvalds unsigned long *_capabilities) 8241da177e4SLinus Torvalds { 8258feae131SDavid Howells unsigned long capabilities, rlen; 8261da177e4SLinus Torvalds unsigned long reqprot = prot; 8271da177e4SLinus Torvalds int ret; 8281da177e4SLinus Torvalds 8291da177e4SLinus Torvalds /* do the simple checks first */ 83006aab5a3SDavid Howells if (flags & MAP_FIXED) { 8311da177e4SLinus Torvalds printk(KERN_DEBUG 8321da177e4SLinus Torvalds "%d: Can't do fixed-address/overlay mmap of RAM\n", 8331da177e4SLinus Torvalds current->pid); 8341da177e4SLinus Torvalds return -EINVAL; 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds 8371da177e4SLinus Torvalds if ((flags & MAP_TYPE) != MAP_PRIVATE && 8381da177e4SLinus Torvalds (flags & MAP_TYPE) != MAP_SHARED) 8391da177e4SLinus Torvalds return -EINVAL; 8401da177e4SLinus Torvalds 841f81cff0dSMike Frysinger if (!len) 8421da177e4SLinus Torvalds return -EINVAL; 8431da177e4SLinus Torvalds 844f81cff0dSMike Frysinger /* Careful about overflows.. */ 8458feae131SDavid Howells rlen = PAGE_ALIGN(len); 8468feae131SDavid Howells if (!rlen || rlen > TASK_SIZE) 847f81cff0dSMike Frysinger return -ENOMEM; 848f81cff0dSMike Frysinger 8491da177e4SLinus Torvalds /* offset overflow? */ 8508feae131SDavid Howells if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) 851f81cff0dSMike Frysinger return -EOVERFLOW; 8521da177e4SLinus Torvalds 8531da177e4SLinus Torvalds if (file) { 8541da177e4SLinus Torvalds /* validate file mapping requests */ 8551da177e4SLinus Torvalds struct address_space *mapping; 8561da177e4SLinus Torvalds 8571da177e4SLinus Torvalds /* files must support mmap */ 8581da177e4SLinus Torvalds if (!file->f_op || !file->f_op->mmap) 8591da177e4SLinus Torvalds return -ENODEV; 8601da177e4SLinus Torvalds 8611da177e4SLinus Torvalds /* work out if what we've got could possibly be shared 8621da177e4SLinus Torvalds * - we support chardevs that provide their own "memory" 8631da177e4SLinus Torvalds * - we support files/blockdevs that are memory backed 8641da177e4SLinus Torvalds */ 8651da177e4SLinus Torvalds mapping = file->f_mapping; 8661da177e4SLinus Torvalds if (!mapping) 867e9536ae7SJosef Sipek mapping = file->f_path.dentry->d_inode->i_mapping; 8681da177e4SLinus Torvalds 8691da177e4SLinus Torvalds capabilities = 0; 8701da177e4SLinus Torvalds if (mapping && mapping->backing_dev_info) 8711da177e4SLinus Torvalds capabilities = mapping->backing_dev_info->capabilities; 8721da177e4SLinus Torvalds 8731da177e4SLinus Torvalds if (!capabilities) { 8741da177e4SLinus Torvalds /* no explicit capabilities set, so assume some 8751da177e4SLinus Torvalds * defaults */ 876e9536ae7SJosef Sipek switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) { 8771da177e4SLinus Torvalds case S_IFREG: 8781da177e4SLinus Torvalds case S_IFBLK: 8791da177e4SLinus Torvalds capabilities = BDI_CAP_MAP_COPY; 8801da177e4SLinus Torvalds break; 8811da177e4SLinus Torvalds 8821da177e4SLinus Torvalds case S_IFCHR: 8831da177e4SLinus Torvalds capabilities = 8841da177e4SLinus Torvalds BDI_CAP_MAP_DIRECT | 8851da177e4SLinus Torvalds BDI_CAP_READ_MAP | 8861da177e4SLinus Torvalds BDI_CAP_WRITE_MAP; 8871da177e4SLinus Torvalds break; 8881da177e4SLinus Torvalds 8891da177e4SLinus Torvalds default: 8901da177e4SLinus Torvalds return -EINVAL; 8911da177e4SLinus Torvalds } 8921da177e4SLinus Torvalds } 8931da177e4SLinus Torvalds 8941da177e4SLinus Torvalds /* eliminate any capabilities that we can't support on this 8951da177e4SLinus Torvalds * device */ 8961da177e4SLinus Torvalds if (!file->f_op->get_unmapped_area) 8971da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 8981da177e4SLinus Torvalds if (!file->f_op->read) 8991da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_COPY; 9001da177e4SLinus Torvalds 90128d7a6aeSGraff Yang /* The file shall have been opened with read permission. */ 90228d7a6aeSGraff Yang if (!(file->f_mode & FMODE_READ)) 90328d7a6aeSGraff Yang return -EACCES; 90428d7a6aeSGraff Yang 9051da177e4SLinus Torvalds if (flags & MAP_SHARED) { 9061da177e4SLinus Torvalds /* do checks for writing, appending and locking */ 9071da177e4SLinus Torvalds if ((prot & PROT_WRITE) && 9081da177e4SLinus Torvalds !(file->f_mode & FMODE_WRITE)) 9091da177e4SLinus Torvalds return -EACCES; 9101da177e4SLinus Torvalds 911e9536ae7SJosef Sipek if (IS_APPEND(file->f_path.dentry->d_inode) && 9121da177e4SLinus Torvalds (file->f_mode & FMODE_WRITE)) 9131da177e4SLinus Torvalds return -EACCES; 9141da177e4SLinus Torvalds 915e9536ae7SJosef Sipek if (locks_verify_locked(file->f_path.dentry->d_inode)) 9161da177e4SLinus Torvalds return -EAGAIN; 9171da177e4SLinus Torvalds 9181da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_DIRECT)) 9191da177e4SLinus Torvalds return -ENODEV; 9201da177e4SLinus Torvalds 9211da177e4SLinus Torvalds if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || 9221da177e4SLinus Torvalds ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || 9231da177e4SLinus Torvalds ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) 9241da177e4SLinus Torvalds ) { 9251da177e4SLinus Torvalds printk("MAP_SHARED not completely supported on !MMU\n"); 9261da177e4SLinus Torvalds return -EINVAL; 9271da177e4SLinus Torvalds } 9281da177e4SLinus Torvalds 9291da177e4SLinus Torvalds /* we mustn't privatise shared mappings */ 9301da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_COPY; 9311da177e4SLinus Torvalds } 9321da177e4SLinus Torvalds else { 9331da177e4SLinus Torvalds /* we're going to read the file into private memory we 9341da177e4SLinus Torvalds * allocate */ 9351da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_COPY)) 9361da177e4SLinus Torvalds return -ENODEV; 9371da177e4SLinus Torvalds 9381da177e4SLinus Torvalds /* we don't permit a private writable mapping to be 9391da177e4SLinus Torvalds * shared with the backing device */ 9401da177e4SLinus Torvalds if (prot & PROT_WRITE) 9411da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 9421da177e4SLinus Torvalds } 9431da177e4SLinus Torvalds 9441da177e4SLinus Torvalds /* handle executable mappings and implied executable 9451da177e4SLinus Torvalds * mappings */ 946e9536ae7SJosef Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 9471da177e4SLinus Torvalds if (prot & PROT_EXEC) 9481da177e4SLinus Torvalds return -EPERM; 9491da177e4SLinus Torvalds } 9501da177e4SLinus Torvalds else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { 9511da177e4SLinus Torvalds /* handle implication of PROT_EXEC by PROT_READ */ 9521da177e4SLinus Torvalds if (current->personality & READ_IMPLIES_EXEC) { 9531da177e4SLinus Torvalds if (capabilities & BDI_CAP_EXEC_MAP) 9541da177e4SLinus Torvalds prot |= PROT_EXEC; 9551da177e4SLinus Torvalds } 9561da177e4SLinus Torvalds } 9571da177e4SLinus Torvalds else if ((prot & PROT_READ) && 9581da177e4SLinus Torvalds (prot & PROT_EXEC) && 9591da177e4SLinus Torvalds !(capabilities & BDI_CAP_EXEC_MAP) 9601da177e4SLinus Torvalds ) { 9611da177e4SLinus Torvalds /* backing file is not executable, try to copy */ 9621da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 9631da177e4SLinus Torvalds } 9641da177e4SLinus Torvalds } 9651da177e4SLinus Torvalds else { 9661da177e4SLinus Torvalds /* anonymous mappings are always memory backed and can be 9671da177e4SLinus Torvalds * privately mapped 9681da177e4SLinus Torvalds */ 9691da177e4SLinus Torvalds capabilities = BDI_CAP_MAP_COPY; 9701da177e4SLinus Torvalds 9711da177e4SLinus Torvalds /* handle PROT_EXEC implication by PROT_READ */ 9721da177e4SLinus Torvalds if ((prot & PROT_READ) && 9731da177e4SLinus Torvalds (current->personality & READ_IMPLIES_EXEC)) 9741da177e4SLinus Torvalds prot |= PROT_EXEC; 9751da177e4SLinus Torvalds } 9761da177e4SLinus Torvalds 9771da177e4SLinus Torvalds /* allow the security API to have its say */ 978ed032189SEric Paris ret = security_file_mmap(file, reqprot, prot, flags, addr, 0); 9791da177e4SLinus Torvalds if (ret < 0) 9801da177e4SLinus Torvalds return ret; 9811da177e4SLinus Torvalds 9821da177e4SLinus Torvalds /* looks okay */ 9831da177e4SLinus Torvalds *_capabilities = capabilities; 9841da177e4SLinus Torvalds return 0; 9851da177e4SLinus Torvalds } 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds /* 9881da177e4SLinus Torvalds * we've determined that we can make the mapping, now translate what we 9891da177e4SLinus Torvalds * now know into VMA flags 9901da177e4SLinus Torvalds */ 9911da177e4SLinus Torvalds static unsigned long determine_vm_flags(struct file *file, 9921da177e4SLinus Torvalds unsigned long prot, 9931da177e4SLinus Torvalds unsigned long flags, 9941da177e4SLinus Torvalds unsigned long capabilities) 9951da177e4SLinus Torvalds { 9961da177e4SLinus Torvalds unsigned long vm_flags; 9971da177e4SLinus Torvalds 9981da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); 9991da177e4SLinus Torvalds vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 10001da177e4SLinus Torvalds /* vm_flags |= mm->def_flags; */ 10011da177e4SLinus Torvalds 10021da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_DIRECT)) { 10031da177e4SLinus Torvalds /* attempt to share read-only copies of mapped file chunks */ 10041da177e4SLinus Torvalds if (file && !(prot & PROT_WRITE)) 10051da177e4SLinus Torvalds vm_flags |= VM_MAYSHARE; 10061da177e4SLinus Torvalds } 10071da177e4SLinus Torvalds else { 10081da177e4SLinus Torvalds /* overlay a shareable mapping on the backing device or inode 10091da177e4SLinus Torvalds * if possible - used for chardevs, ramfs/tmpfs/shmfs and 10101da177e4SLinus Torvalds * romfs/cramfs */ 10111da177e4SLinus Torvalds if (flags & MAP_SHARED) 10121da177e4SLinus Torvalds vm_flags |= VM_MAYSHARE | VM_SHARED; 10131da177e4SLinus Torvalds else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0) 10141da177e4SLinus Torvalds vm_flags |= VM_MAYSHARE; 10151da177e4SLinus Torvalds } 10161da177e4SLinus Torvalds 10171da177e4SLinus Torvalds /* refuse to let anyone share private mappings with this process if 10181da177e4SLinus Torvalds * it's being traced - otherwise breakpoints set in it may interfere 10191da177e4SLinus Torvalds * with another untraced process 10201da177e4SLinus Torvalds */ 1021fa8e26ccSRoland McGrath if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current)) 10221da177e4SLinus Torvalds vm_flags &= ~VM_MAYSHARE; 10231da177e4SLinus Torvalds 10241da177e4SLinus Torvalds return vm_flags; 10251da177e4SLinus Torvalds } 10261da177e4SLinus Torvalds 10271da177e4SLinus Torvalds /* 10288feae131SDavid Howells * set up a shared mapping on a file (the driver or filesystem provides and 10298feae131SDavid Howells * pins the storage) 10301da177e4SLinus Torvalds */ 10318feae131SDavid Howells static int do_mmap_shared_file(struct vm_area_struct *vma) 10321da177e4SLinus Torvalds { 10331da177e4SLinus Torvalds int ret; 10341da177e4SLinus Torvalds 10351da177e4SLinus Torvalds ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1036dd8632a1SPaul Mundt if (ret == 0) { 1037dd8632a1SPaul Mundt vma->vm_region->vm_top = vma->vm_region->vm_end; 1038645d83c5SDavid Howells return 0; 1039dd8632a1SPaul Mundt } 10401da177e4SLinus Torvalds if (ret != -ENOSYS) 10411da177e4SLinus Torvalds return ret; 10421da177e4SLinus Torvalds 10431da177e4SLinus Torvalds /* getting an ENOSYS error indicates that direct mmap isn't 10441da177e4SLinus Torvalds * possible (as opposed to tried but failed) so we'll fall 10451da177e4SLinus Torvalds * through to making a private copy of the data and mapping 10461da177e4SLinus Torvalds * that if we can */ 10471da177e4SLinus Torvalds return -ENODEV; 10481da177e4SLinus Torvalds } 10491da177e4SLinus Torvalds 10501da177e4SLinus Torvalds /* 10511da177e4SLinus Torvalds * set up a private mapping or an anonymous shared mapping 10521da177e4SLinus Torvalds */ 10538feae131SDavid Howells static int do_mmap_private(struct vm_area_struct *vma, 10548feae131SDavid Howells struct vm_region *region, 1055645d83c5SDavid Howells unsigned long len, 1056645d83c5SDavid Howells unsigned long capabilities) 10571da177e4SLinus Torvalds { 10588feae131SDavid Howells struct page *pages; 10598feae131SDavid Howells unsigned long total, point, n, rlen; 10601da177e4SLinus Torvalds void *base; 10618feae131SDavid Howells int ret, order; 10621da177e4SLinus Torvalds 10631da177e4SLinus Torvalds /* invoke the file's mapping function so that it can keep track of 10641da177e4SLinus Torvalds * shared mappings on devices or memory 10651da177e4SLinus Torvalds * - VM_MAYSHARE will be set if it may attempt to share 10661da177e4SLinus Torvalds */ 1067645d83c5SDavid Howells if (capabilities & BDI_CAP_MAP_DIRECT) { 10681da177e4SLinus Torvalds ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1069dd8632a1SPaul Mundt if (ret == 0) { 10701da177e4SLinus Torvalds /* shouldn't return success if we're not sharing */ 1071dd8632a1SPaul Mundt BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); 1072dd8632a1SPaul Mundt vma->vm_region->vm_top = vma->vm_region->vm_end; 1073645d83c5SDavid Howells return 0; 10741da177e4SLinus Torvalds } 1075dd8632a1SPaul Mundt if (ret != -ENOSYS) 1076dd8632a1SPaul Mundt return ret; 10771da177e4SLinus Torvalds 10781da177e4SLinus Torvalds /* getting an ENOSYS error indicates that direct mmap isn't 10791da177e4SLinus Torvalds * possible (as opposed to tried but failed) so we'll try to 10801da177e4SLinus Torvalds * make a private copy of the data and map that instead */ 10811da177e4SLinus Torvalds } 10821da177e4SLinus Torvalds 10838feae131SDavid Howells rlen = PAGE_ALIGN(len); 10848feae131SDavid Howells 10851da177e4SLinus Torvalds /* allocate some memory to hold the mapping 10861da177e4SLinus Torvalds * - note that this may not return a page-aligned address if the object 10871da177e4SLinus Torvalds * we're allocating is smaller than a page 10881da177e4SLinus Torvalds */ 10898feae131SDavid Howells order = get_order(rlen); 10908feae131SDavid Howells kdebug("alloc order %d for %lx", order, len); 10918feae131SDavid Howells 10928feae131SDavid Howells pages = alloc_pages(GFP_KERNEL, order); 10938feae131SDavid Howells if (!pages) 10941da177e4SLinus Torvalds goto enomem; 10951da177e4SLinus Torvalds 10968feae131SDavid Howells total = 1 << order; 109733e5d769SDavid Howells atomic_long_add(total, &mmap_pages_allocated); 10981da177e4SLinus Torvalds 10998feae131SDavid Howells point = rlen >> PAGE_SHIFT; 1100dd8632a1SPaul Mundt 1101dd8632a1SPaul Mundt /* we allocated a power-of-2 sized page set, so we may want to trim off 1102dd8632a1SPaul Mundt * the excess */ 1103dd8632a1SPaul Mundt if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { 11048feae131SDavid Howells while (total > point) { 11058feae131SDavid Howells order = ilog2(total - point); 11068feae131SDavid Howells n = 1 << order; 11078feae131SDavid Howells kdebug("shave %lu/%lu @%lu", n, total - point, total); 110833e5d769SDavid Howells atomic_long_sub(n, &mmap_pages_allocated); 11098feae131SDavid Howells total -= n; 11108feae131SDavid Howells set_page_refcounted(pages + total); 11118feae131SDavid Howells __free_pages(pages + total, order); 11128feae131SDavid Howells } 1113dd8632a1SPaul Mundt } 11148feae131SDavid Howells 11158feae131SDavid Howells for (point = 1; point < total; point++) 11168feae131SDavid Howells set_page_refcounted(&pages[point]); 11178feae131SDavid Howells 11188feae131SDavid Howells base = page_address(pages); 11198feae131SDavid Howells region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; 11208feae131SDavid Howells region->vm_start = (unsigned long) base; 11218feae131SDavid Howells region->vm_end = region->vm_start + rlen; 1122dd8632a1SPaul Mundt region->vm_top = region->vm_start + (total << PAGE_SHIFT); 11238feae131SDavid Howells 11248feae131SDavid Howells vma->vm_start = region->vm_start; 11258feae131SDavid Howells vma->vm_end = region->vm_start + len; 11261da177e4SLinus Torvalds 11271da177e4SLinus Torvalds if (vma->vm_file) { 11281da177e4SLinus Torvalds /* read the contents of a file into the copy */ 11291da177e4SLinus Torvalds mm_segment_t old_fs; 11301da177e4SLinus Torvalds loff_t fpos; 11311da177e4SLinus Torvalds 11321da177e4SLinus Torvalds fpos = vma->vm_pgoff; 11331da177e4SLinus Torvalds fpos <<= PAGE_SHIFT; 11341da177e4SLinus Torvalds 11351da177e4SLinus Torvalds old_fs = get_fs(); 11361da177e4SLinus Torvalds set_fs(KERNEL_DS); 11378feae131SDavid Howells ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos); 11381da177e4SLinus Torvalds set_fs(old_fs); 11391da177e4SLinus Torvalds 11401da177e4SLinus Torvalds if (ret < 0) 11411da177e4SLinus Torvalds goto error_free; 11421da177e4SLinus Torvalds 11431da177e4SLinus Torvalds /* clear the last little bit */ 11448feae131SDavid Howells if (ret < rlen) 11458feae131SDavid Howells memset(base + ret, 0, rlen - ret); 11461da177e4SLinus Torvalds 11471da177e4SLinus Torvalds } 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds return 0; 11501da177e4SLinus Torvalds 11511da177e4SLinus Torvalds error_free: 11528feae131SDavid Howells free_page_series(region->vm_start, region->vm_end); 11538feae131SDavid Howells region->vm_start = vma->vm_start = 0; 11548feae131SDavid Howells region->vm_end = vma->vm_end = 0; 1155dd8632a1SPaul Mundt region->vm_top = 0; 11561da177e4SLinus Torvalds return ret; 11571da177e4SLinus Torvalds 11581da177e4SLinus Torvalds enomem: 115905ae6fa3SGreg Ungerer printk("Allocation of length %lu from process %d (%s) failed\n", 116005ae6fa3SGreg Ungerer len, current->pid, current->comm); 11611da177e4SLinus Torvalds show_free_areas(); 11621da177e4SLinus Torvalds return -ENOMEM; 11631da177e4SLinus Torvalds } 11641da177e4SLinus Torvalds 11651da177e4SLinus Torvalds /* 11661da177e4SLinus Torvalds * handle mapping creation for uClinux 11671da177e4SLinus Torvalds */ 11681da177e4SLinus Torvalds unsigned long do_mmap_pgoff(struct file *file, 11691da177e4SLinus Torvalds unsigned long addr, 11701da177e4SLinus Torvalds unsigned long len, 11711da177e4SLinus Torvalds unsigned long prot, 11721da177e4SLinus Torvalds unsigned long flags, 11731da177e4SLinus Torvalds unsigned long pgoff) 11741da177e4SLinus Torvalds { 11758feae131SDavid Howells struct vm_area_struct *vma; 11768feae131SDavid Howells struct vm_region *region; 11771da177e4SLinus Torvalds struct rb_node *rb; 11788feae131SDavid Howells unsigned long capabilities, vm_flags, result; 11791da177e4SLinus Torvalds int ret; 11801da177e4SLinus Torvalds 11818feae131SDavid Howells kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); 11828feae131SDavid Howells 11831da177e4SLinus Torvalds /* decide whether we should attempt the mapping, and if so what sort of 11841da177e4SLinus Torvalds * mapping */ 11851da177e4SLinus Torvalds ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 11861da177e4SLinus Torvalds &capabilities); 11878feae131SDavid Howells if (ret < 0) { 11888feae131SDavid Howells kleave(" = %d [val]", ret); 11891da177e4SLinus Torvalds return ret; 11908feae131SDavid Howells } 11911da177e4SLinus Torvalds 119206aab5a3SDavid Howells /* we ignore the address hint */ 119306aab5a3SDavid Howells addr = 0; 119406aab5a3SDavid Howells 11951da177e4SLinus Torvalds /* we've determined that we can make the mapping, now translate what we 11961da177e4SLinus Torvalds * now know into VMA flags */ 11971da177e4SLinus Torvalds vm_flags = determine_vm_flags(file, prot, flags, capabilities); 11981da177e4SLinus Torvalds 11998feae131SDavid Howells /* we're going to need to record the mapping */ 12008feae131SDavid Howells region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); 12018feae131SDavid Howells if (!region) 12028feae131SDavid Howells goto error_getting_region; 12031da177e4SLinus Torvalds 12048feae131SDavid Howells vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 12058feae131SDavid Howells if (!vma) 12068feae131SDavid Howells goto error_getting_vma; 12071da177e4SLinus Torvalds 1208*1e2ae599SDavid Howells region->vm_usage = 1; 12098feae131SDavid Howells region->vm_flags = vm_flags; 12108feae131SDavid Howells region->vm_pgoff = pgoff; 12118feae131SDavid Howells 12128feae131SDavid Howells INIT_LIST_HEAD(&vma->anon_vma_node); 12138feae131SDavid Howells vma->vm_flags = vm_flags; 12148feae131SDavid Howells vma->vm_pgoff = pgoff; 12158feae131SDavid Howells 12168feae131SDavid Howells if (file) { 12178feae131SDavid Howells region->vm_file = file; 12188feae131SDavid Howells get_file(file); 12198feae131SDavid Howells vma->vm_file = file; 12208feae131SDavid Howells get_file(file); 12218feae131SDavid Howells if (vm_flags & VM_EXECUTABLE) { 12228feae131SDavid Howells added_exe_file_vma(current->mm); 12238feae131SDavid Howells vma->vm_mm = current->mm; 12248feae131SDavid Howells } 12258feae131SDavid Howells } 12268feae131SDavid Howells 12278feae131SDavid Howells down_write(&nommu_region_sem); 12288feae131SDavid Howells 12298feae131SDavid Howells /* if we want to share, we need to check for regions created by other 12301da177e4SLinus Torvalds * mmap() calls that overlap with our proposed mapping 12318feae131SDavid Howells * - we can only share with a superset match on most regular files 12321da177e4SLinus Torvalds * - shared mappings on character devices and memory backed files are 12331da177e4SLinus Torvalds * permitted to overlap inexactly as far as we are concerned for in 12341da177e4SLinus Torvalds * these cases, sharing is handled in the driver or filesystem rather 12351da177e4SLinus Torvalds * than here 12361da177e4SLinus Torvalds */ 12371da177e4SLinus Torvalds if (vm_flags & VM_MAYSHARE) { 12388feae131SDavid Howells struct vm_region *pregion; 12398feae131SDavid Howells unsigned long pglen, rpglen, pgend, rpgend, start; 12401da177e4SLinus Torvalds 12418feae131SDavid Howells pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 12428feae131SDavid Howells pgend = pgoff + pglen; 1243165b2392SDavid Howells 12448feae131SDavid Howells for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { 12458feae131SDavid Howells pregion = rb_entry(rb, struct vm_region, vm_rb); 12461da177e4SLinus Torvalds 12478feae131SDavid Howells if (!(pregion->vm_flags & VM_MAYSHARE)) 12481da177e4SLinus Torvalds continue; 12491da177e4SLinus Torvalds 12501da177e4SLinus Torvalds /* search for overlapping mappings on the same file */ 12518feae131SDavid Howells if (pregion->vm_file->f_path.dentry->d_inode != 12528feae131SDavid Howells file->f_path.dentry->d_inode) 12531da177e4SLinus Torvalds continue; 12541da177e4SLinus Torvalds 12558feae131SDavid Howells if (pregion->vm_pgoff >= pgend) 12561da177e4SLinus Torvalds continue; 12571da177e4SLinus Torvalds 12588feae131SDavid Howells rpglen = pregion->vm_end - pregion->vm_start; 12598feae131SDavid Howells rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; 12608feae131SDavid Howells rpgend = pregion->vm_pgoff + rpglen; 12618feae131SDavid Howells if (pgoff >= rpgend) 12621da177e4SLinus Torvalds continue; 12631da177e4SLinus Torvalds 12648feae131SDavid Howells /* handle inexactly overlapping matches between 12658feae131SDavid Howells * mappings */ 12668feae131SDavid Howells if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && 12678feae131SDavid Howells !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { 12688feae131SDavid Howells /* new mapping is not a subset of the region */ 12691da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_DIRECT)) 12701da177e4SLinus Torvalds goto sharing_violation; 12711da177e4SLinus Torvalds continue; 12721da177e4SLinus Torvalds } 12731da177e4SLinus Torvalds 12748feae131SDavid Howells /* we've found a region we can share */ 1275*1e2ae599SDavid Howells pregion->vm_usage++; 12768feae131SDavid Howells vma->vm_region = pregion; 12778feae131SDavid Howells start = pregion->vm_start; 12788feae131SDavid Howells start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; 12798feae131SDavid Howells vma->vm_start = start; 12808feae131SDavid Howells vma->vm_end = start + len; 12811da177e4SLinus Torvalds 12828feae131SDavid Howells if (pregion->vm_flags & VM_MAPPED_COPY) { 12838feae131SDavid Howells kdebug("share copy"); 12848feae131SDavid Howells vma->vm_flags |= VM_MAPPED_COPY; 12858feae131SDavid Howells } else { 12868feae131SDavid Howells kdebug("share mmap"); 12878feae131SDavid Howells ret = do_mmap_shared_file(vma); 12888feae131SDavid Howells if (ret < 0) { 12898feae131SDavid Howells vma->vm_region = NULL; 12908feae131SDavid Howells vma->vm_start = 0; 12918feae131SDavid Howells vma->vm_end = 0; 1292*1e2ae599SDavid Howells pregion->vm_usage--; 12938feae131SDavid Howells pregion = NULL; 12948feae131SDavid Howells goto error_just_free; 12951da177e4SLinus Torvalds } 12968feae131SDavid Howells } 12978feae131SDavid Howells fput(region->vm_file); 12988feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 12998feae131SDavid Howells region = pregion; 13008feae131SDavid Howells result = start; 13018feae131SDavid Howells goto share; 13028feae131SDavid Howells } 13031da177e4SLinus Torvalds 13041da177e4SLinus Torvalds /* obtain the address at which to make a shared mapping 13051da177e4SLinus Torvalds * - this is the hook for quasi-memory character devices to 13061da177e4SLinus Torvalds * tell us the location of a shared mapping 13071da177e4SLinus Torvalds */ 1308645d83c5SDavid Howells if (capabilities & BDI_CAP_MAP_DIRECT) { 13091da177e4SLinus Torvalds addr = file->f_op->get_unmapped_area(file, addr, len, 13101da177e4SLinus Torvalds pgoff, flags); 13111da177e4SLinus Torvalds if (IS_ERR((void *) addr)) { 13121da177e4SLinus Torvalds ret = addr; 13131da177e4SLinus Torvalds if (ret != (unsigned long) -ENOSYS) 13148feae131SDavid Howells goto error_just_free; 13151da177e4SLinus Torvalds 13161da177e4SLinus Torvalds /* the driver refused to tell us where to site 13171da177e4SLinus Torvalds * the mapping so we'll have to attempt to copy 13181da177e4SLinus Torvalds * it */ 13191da177e4SLinus Torvalds ret = (unsigned long) -ENODEV; 13201da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_COPY)) 13218feae131SDavid Howells goto error_just_free; 13221da177e4SLinus Torvalds 13231da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 13248feae131SDavid Howells } else { 13258feae131SDavid Howells vma->vm_start = region->vm_start = addr; 13268feae131SDavid Howells vma->vm_end = region->vm_end = addr + len; 13271da177e4SLinus Torvalds } 13281da177e4SLinus Torvalds } 13291da177e4SLinus Torvalds } 13301da177e4SLinus Torvalds 13318feae131SDavid Howells vma->vm_region = region; 13321da177e4SLinus Torvalds 1333645d83c5SDavid Howells /* set up the mapping 1334645d83c5SDavid Howells * - the region is filled in if BDI_CAP_MAP_DIRECT is still set 1335645d83c5SDavid Howells */ 13361da177e4SLinus Torvalds if (file && vma->vm_flags & VM_SHARED) 13378feae131SDavid Howells ret = do_mmap_shared_file(vma); 13381da177e4SLinus Torvalds else 1339645d83c5SDavid Howells ret = do_mmap_private(vma, region, len, capabilities); 13401da177e4SLinus Torvalds if (ret < 0) 1341645d83c5SDavid Howells goto error_just_free; 1342645d83c5SDavid Howells add_nommu_region(region); 13438feae131SDavid Howells 1344ea637639SJie Zhang /* clear anonymous mappings that don't ask for uninitialized data */ 1345ea637639SJie Zhang if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) 1346ea637639SJie Zhang memset((void *)region->vm_start, 0, 1347ea637639SJie Zhang region->vm_end - region->vm_start); 1348ea637639SJie Zhang 13491da177e4SLinus Torvalds /* okay... we have a mapping; now we have to register it */ 13508feae131SDavid Howells result = vma->vm_start; 13511da177e4SLinus Torvalds 13521da177e4SLinus Torvalds current->mm->total_vm += len >> PAGE_SHIFT; 13531da177e4SLinus Torvalds 13548feae131SDavid Howells share: 13558feae131SDavid Howells add_vma_to_mm(current->mm, vma); 13561da177e4SLinus Torvalds 1357cfe79c00SMike Frysinger /* we flush the region from the icache only when the first executable 1358cfe79c00SMike Frysinger * mapping of it is made */ 1359cfe79c00SMike Frysinger if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { 1360cfe79c00SMike Frysinger flush_icache_range(region->vm_start, region->vm_end); 1361cfe79c00SMike Frysinger region->vm_icache_flushed = true; 1362cfe79c00SMike Frysinger } 13631da177e4SLinus Torvalds 1364cfe79c00SMike Frysinger up_write(&nommu_region_sem); 13651da177e4SLinus Torvalds 13668feae131SDavid Howells kleave(" = %lx", result); 13678feae131SDavid Howells return result; 13681da177e4SLinus Torvalds 13698feae131SDavid Howells error_just_free: 13708feae131SDavid Howells up_write(&nommu_region_sem); 13718feae131SDavid Howells error: 137289a86402SDavid Howells if (region->vm_file) 13738feae131SDavid Howells fput(region->vm_file); 13748feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 137589a86402SDavid Howells if (vma->vm_file) 13768feae131SDavid Howells fput(vma->vm_file); 13778feae131SDavid Howells if (vma->vm_flags & VM_EXECUTABLE) 13788feae131SDavid Howells removed_exe_file_vma(vma->vm_mm); 13798feae131SDavid Howells kmem_cache_free(vm_area_cachep, vma); 13808feae131SDavid Howells kleave(" = %d", ret); 13811da177e4SLinus Torvalds return ret; 13821da177e4SLinus Torvalds 13831da177e4SLinus Torvalds sharing_violation: 13848feae131SDavid Howells up_write(&nommu_region_sem); 13858feae131SDavid Howells printk(KERN_WARNING "Attempt to share mismatched mappings\n"); 13868feae131SDavid Howells ret = -EINVAL; 13878feae131SDavid Howells goto error; 13881da177e4SLinus Torvalds 13891da177e4SLinus Torvalds error_getting_vma: 13908feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 13918feae131SDavid Howells printk(KERN_WARNING "Allocation of vma for %lu byte allocation" 13928feae131SDavid Howells " from process %d failed\n", 13931da177e4SLinus Torvalds len, current->pid); 13941da177e4SLinus Torvalds show_free_areas(); 13951da177e4SLinus Torvalds return -ENOMEM; 13961da177e4SLinus Torvalds 13978feae131SDavid Howells error_getting_region: 13988feae131SDavid Howells printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" 13998feae131SDavid Howells " from process %d failed\n", 14001da177e4SLinus Torvalds len, current->pid); 14011da177e4SLinus Torvalds show_free_areas(); 14021da177e4SLinus Torvalds return -ENOMEM; 14031da177e4SLinus Torvalds } 1404b5073173SPaul Mundt EXPORT_SYMBOL(do_mmap_pgoff); 14051da177e4SLinus Torvalds 140666f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 140766f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 140866f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 140966f0dc48SHugh Dickins { 141066f0dc48SHugh Dickins struct file *file = NULL; 141166f0dc48SHugh Dickins unsigned long retval = -EBADF; 141266f0dc48SHugh Dickins 141366f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 141466f0dc48SHugh Dickins file = fget(fd); 141566f0dc48SHugh Dickins if (!file) 141666f0dc48SHugh Dickins goto out; 141766f0dc48SHugh Dickins } 141866f0dc48SHugh Dickins 141966f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 142066f0dc48SHugh Dickins 142166f0dc48SHugh Dickins down_write(¤t->mm->mmap_sem); 142266f0dc48SHugh Dickins retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff); 142366f0dc48SHugh Dickins up_write(¤t->mm->mmap_sem); 142466f0dc48SHugh Dickins 142566f0dc48SHugh Dickins if (file) 142666f0dc48SHugh Dickins fput(file); 142766f0dc48SHugh Dickins out: 142866f0dc48SHugh Dickins return retval; 142966f0dc48SHugh Dickins } 143066f0dc48SHugh Dickins 14311da177e4SLinus Torvalds /* 14328feae131SDavid Howells * split a vma into two pieces at address 'addr', a new vma is allocated either 14338feae131SDavid Howells * for the first part or the tail. 14341da177e4SLinus Torvalds */ 14358feae131SDavid Howells int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 14368feae131SDavid Howells unsigned long addr, int new_below) 14371da177e4SLinus Torvalds { 14388feae131SDavid Howells struct vm_area_struct *new; 14398feae131SDavid Howells struct vm_region *region; 14408feae131SDavid Howells unsigned long npages; 14411da177e4SLinus Torvalds 14428feae131SDavid Howells kenter(""); 14431da177e4SLinus Torvalds 14448feae131SDavid Howells /* we're only permitted to split anonymous regions that have a single 14458feae131SDavid Howells * owner */ 14468feae131SDavid Howells if (vma->vm_file || 1447*1e2ae599SDavid Howells vma->vm_region->vm_usage != 1) 14488feae131SDavid Howells return -ENOMEM; 14491da177e4SLinus Torvalds 14508feae131SDavid Howells if (mm->map_count >= sysctl_max_map_count) 14518feae131SDavid Howells return -ENOMEM; 14521da177e4SLinus Torvalds 14538feae131SDavid Howells region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); 14548feae131SDavid Howells if (!region) 14558feae131SDavid Howells return -ENOMEM; 14568feae131SDavid Howells 14578feae131SDavid Howells new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 14588feae131SDavid Howells if (!new) { 14598feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 14608feae131SDavid Howells return -ENOMEM; 14611da177e4SLinus Torvalds } 14621da177e4SLinus Torvalds 14638feae131SDavid Howells /* most fields are the same, copy all, and then fixup */ 14648feae131SDavid Howells *new = *vma; 14658feae131SDavid Howells *region = *vma->vm_region; 14668feae131SDavid Howells new->vm_region = region; 14678feae131SDavid Howells 14688feae131SDavid Howells npages = (addr - vma->vm_start) >> PAGE_SHIFT; 14698feae131SDavid Howells 14708feae131SDavid Howells if (new_below) { 1471dd8632a1SPaul Mundt region->vm_top = region->vm_end = new->vm_end = addr; 14728feae131SDavid Howells } else { 14738feae131SDavid Howells region->vm_start = new->vm_start = addr; 14748feae131SDavid Howells region->vm_pgoff = new->vm_pgoff += npages; 14751da177e4SLinus Torvalds } 14768feae131SDavid Howells 14778feae131SDavid Howells if (new->vm_ops && new->vm_ops->open) 14788feae131SDavid Howells new->vm_ops->open(new); 14798feae131SDavid Howells 14808feae131SDavid Howells delete_vma_from_mm(vma); 14818feae131SDavid Howells down_write(&nommu_region_sem); 14828feae131SDavid Howells delete_nommu_region(vma->vm_region); 14838feae131SDavid Howells if (new_below) { 14848feae131SDavid Howells vma->vm_region->vm_start = vma->vm_start = addr; 14858feae131SDavid Howells vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; 14868feae131SDavid Howells } else { 14878feae131SDavid Howells vma->vm_region->vm_end = vma->vm_end = addr; 1488dd8632a1SPaul Mundt vma->vm_region->vm_top = addr; 14898feae131SDavid Howells } 14908feae131SDavid Howells add_nommu_region(vma->vm_region); 14918feae131SDavid Howells add_nommu_region(new->vm_region); 14928feae131SDavid Howells up_write(&nommu_region_sem); 14938feae131SDavid Howells add_vma_to_mm(mm, vma); 14948feae131SDavid Howells add_vma_to_mm(mm, new); 14958feae131SDavid Howells return 0; 14968feae131SDavid Howells } 14978feae131SDavid Howells 14988feae131SDavid Howells /* 14998feae131SDavid Howells * shrink a VMA by removing the specified chunk from either the beginning or 15008feae131SDavid Howells * the end 15018feae131SDavid Howells */ 15028feae131SDavid Howells static int shrink_vma(struct mm_struct *mm, 15038feae131SDavid Howells struct vm_area_struct *vma, 15048feae131SDavid Howells unsigned long from, unsigned long to) 15058feae131SDavid Howells { 15068feae131SDavid Howells struct vm_region *region; 15078feae131SDavid Howells 15088feae131SDavid Howells kenter(""); 15098feae131SDavid Howells 15108feae131SDavid Howells /* adjust the VMA's pointers, which may reposition it in the MM's tree 15118feae131SDavid Howells * and list */ 15128feae131SDavid Howells delete_vma_from_mm(vma); 15138feae131SDavid Howells if (from > vma->vm_start) 15148feae131SDavid Howells vma->vm_end = from; 15158feae131SDavid Howells else 15168feae131SDavid Howells vma->vm_start = to; 15178feae131SDavid Howells add_vma_to_mm(mm, vma); 15188feae131SDavid Howells 15198feae131SDavid Howells /* cut the backing region down to size */ 15208feae131SDavid Howells region = vma->vm_region; 1521*1e2ae599SDavid Howells BUG_ON(region->vm_usage != 1); 15228feae131SDavid Howells 15238feae131SDavid Howells down_write(&nommu_region_sem); 15248feae131SDavid Howells delete_nommu_region(region); 1525dd8632a1SPaul Mundt if (from > region->vm_start) { 1526dd8632a1SPaul Mundt to = region->vm_top; 1527dd8632a1SPaul Mundt region->vm_top = region->vm_end = from; 1528dd8632a1SPaul Mundt } else { 15298feae131SDavid Howells region->vm_start = to; 1530dd8632a1SPaul Mundt } 15318feae131SDavid Howells add_nommu_region(region); 15328feae131SDavid Howells up_write(&nommu_region_sem); 15338feae131SDavid Howells 15348feae131SDavid Howells free_page_series(from, to); 15358feae131SDavid Howells return 0; 15361da177e4SLinus Torvalds } 15371da177e4SLinus Torvalds 15383034097aSDavid Howells /* 15393034097aSDavid Howells * release a mapping 15408feae131SDavid Howells * - under NOMMU conditions the chunk to be unmapped must be backed by a single 15418feae131SDavid Howells * VMA, though it need not cover the whole VMA 15423034097aSDavid Howells */ 15438feae131SDavid Howells int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 15441da177e4SLinus Torvalds { 15458feae131SDavid Howells struct vm_area_struct *vma; 15468feae131SDavid Howells struct rb_node *rb; 15478feae131SDavid Howells unsigned long end = start + len; 15488feae131SDavid Howells int ret; 15491da177e4SLinus Torvalds 15508feae131SDavid Howells kenter(",%lx,%zx", start, len); 15511da177e4SLinus Torvalds 15528feae131SDavid Howells if (len == 0) 15531da177e4SLinus Torvalds return -EINVAL; 15541da177e4SLinus Torvalds 15558feae131SDavid Howells /* find the first potentially overlapping VMA */ 15568feae131SDavid Howells vma = find_vma(mm, start); 15578feae131SDavid Howells if (!vma) { 155833e5d769SDavid Howells static int limit = 0; 155933e5d769SDavid Howells if (limit < 5) { 15608feae131SDavid Howells printk(KERN_WARNING 156133e5d769SDavid Howells "munmap of memory not mmapped by process %d" 156233e5d769SDavid Howells " (%s): 0x%lx-0x%lx\n", 156333e5d769SDavid Howells current->pid, current->comm, 156433e5d769SDavid Howells start, start + len - 1); 156533e5d769SDavid Howells limit++; 156633e5d769SDavid Howells } 15678feae131SDavid Howells return -EINVAL; 15688feae131SDavid Howells } 15691da177e4SLinus Torvalds 15708feae131SDavid Howells /* we're allowed to split an anonymous VMA but not a file-backed one */ 15718feae131SDavid Howells if (vma->vm_file) { 15728feae131SDavid Howells do { 15738feae131SDavid Howells if (start > vma->vm_start) { 15748feae131SDavid Howells kleave(" = -EINVAL [miss]"); 15758feae131SDavid Howells return -EINVAL; 15768feae131SDavid Howells } 15778feae131SDavid Howells if (end == vma->vm_end) 15788feae131SDavid Howells goto erase_whole_vma; 15798feae131SDavid Howells rb = rb_next(&vma->vm_rb); 15808feae131SDavid Howells vma = rb_entry(rb, struct vm_area_struct, vm_rb); 15818feae131SDavid Howells } while (rb); 15828feae131SDavid Howells kleave(" = -EINVAL [split file]"); 15838feae131SDavid Howells return -EINVAL; 15848feae131SDavid Howells } else { 15858feae131SDavid Howells /* the chunk must be a subset of the VMA found */ 15868feae131SDavid Howells if (start == vma->vm_start && end == vma->vm_end) 15878feae131SDavid Howells goto erase_whole_vma; 15888feae131SDavid Howells if (start < vma->vm_start || end > vma->vm_end) { 15898feae131SDavid Howells kleave(" = -EINVAL [superset]"); 15908feae131SDavid Howells return -EINVAL; 15918feae131SDavid Howells } 15928feae131SDavid Howells if (start & ~PAGE_MASK) { 15938feae131SDavid Howells kleave(" = -EINVAL [unaligned start]"); 15948feae131SDavid Howells return -EINVAL; 15958feae131SDavid Howells } 15968feae131SDavid Howells if (end != vma->vm_end && end & ~PAGE_MASK) { 15978feae131SDavid Howells kleave(" = -EINVAL [unaligned split]"); 15988feae131SDavid Howells return -EINVAL; 15998feae131SDavid Howells } 16008feae131SDavid Howells if (start != vma->vm_start && end != vma->vm_end) { 16018feae131SDavid Howells ret = split_vma(mm, vma, start, 1); 16028feae131SDavid Howells if (ret < 0) { 16038feae131SDavid Howells kleave(" = %d [split]", ret); 16048feae131SDavid Howells return ret; 16058feae131SDavid Howells } 16068feae131SDavid Howells } 16078feae131SDavid Howells return shrink_vma(mm, vma, start, end); 16088feae131SDavid Howells } 16091da177e4SLinus Torvalds 16108feae131SDavid Howells erase_whole_vma: 16118feae131SDavid Howells delete_vma_from_mm(vma); 16128feae131SDavid Howells delete_vma(mm, vma); 16138feae131SDavid Howells kleave(" = 0"); 16141da177e4SLinus Torvalds return 0; 16151da177e4SLinus Torvalds } 1616b5073173SPaul Mundt EXPORT_SYMBOL(do_munmap); 16171da177e4SLinus Torvalds 16186a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 16193034097aSDavid Howells { 16203034097aSDavid Howells int ret; 16213034097aSDavid Howells struct mm_struct *mm = current->mm; 16223034097aSDavid Howells 16233034097aSDavid Howells down_write(&mm->mmap_sem); 16243034097aSDavid Howells ret = do_munmap(mm, addr, len); 16253034097aSDavid Howells up_write(&mm->mmap_sem); 16263034097aSDavid Howells return ret; 16273034097aSDavid Howells } 16283034097aSDavid Howells 16293034097aSDavid Howells /* 16308feae131SDavid Howells * release all the mappings made in a process's VM space 16313034097aSDavid Howells */ 16321da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 16331da177e4SLinus Torvalds { 16348feae131SDavid Howells struct vm_area_struct *vma; 16351da177e4SLinus Torvalds 16368feae131SDavid Howells if (!mm) 16378feae131SDavid Howells return; 16388feae131SDavid Howells 16398feae131SDavid Howells kenter(""); 16401da177e4SLinus Torvalds 16411da177e4SLinus Torvalds mm->total_vm = 0; 16421da177e4SLinus Torvalds 16438feae131SDavid Howells while ((vma = mm->mmap)) { 16448feae131SDavid Howells mm->mmap = vma->vm_next; 16458feae131SDavid Howells delete_vma_from_mm(vma); 16468feae131SDavid Howells delete_vma(mm, vma); 16471da177e4SLinus Torvalds } 16481da177e4SLinus Torvalds 16498feae131SDavid Howells kleave(""); 16501da177e4SLinus Torvalds } 16511da177e4SLinus Torvalds 16521da177e4SLinus Torvalds unsigned long do_brk(unsigned long addr, unsigned long len) 16531da177e4SLinus Torvalds { 16541da177e4SLinus Torvalds return -ENOMEM; 16551da177e4SLinus Torvalds } 16561da177e4SLinus Torvalds 16571da177e4SLinus Torvalds /* 16586fa5f80bSDavid Howells * expand (or shrink) an existing mapping, potentially moving it at the same 16596fa5f80bSDavid Howells * time (controlled by the MREMAP_MAYMOVE flag and available VM space) 16601da177e4SLinus Torvalds * 16616fa5f80bSDavid Howells * under NOMMU conditions, we only permit changing a mapping's size, and only 16628feae131SDavid Howells * as long as it stays within the region allocated by do_mmap_private() and the 16638feae131SDavid Howells * block is not shareable 16641da177e4SLinus Torvalds * 16656fa5f80bSDavid Howells * MREMAP_FIXED is not supported under NOMMU conditions 16661da177e4SLinus Torvalds */ 16671da177e4SLinus Torvalds unsigned long do_mremap(unsigned long addr, 16681da177e4SLinus Torvalds unsigned long old_len, unsigned long new_len, 16691da177e4SLinus Torvalds unsigned long flags, unsigned long new_addr) 16701da177e4SLinus Torvalds { 16716fa5f80bSDavid Howells struct vm_area_struct *vma; 16721da177e4SLinus Torvalds 16731da177e4SLinus Torvalds /* insanity checks first */ 16748feae131SDavid Howells if (old_len == 0 || new_len == 0) 16751da177e4SLinus Torvalds return (unsigned long) -EINVAL; 16761da177e4SLinus Torvalds 16778feae131SDavid Howells if (addr & ~PAGE_MASK) 16788feae131SDavid Howells return -EINVAL; 16798feae131SDavid Howells 16801da177e4SLinus Torvalds if (flags & MREMAP_FIXED && new_addr != addr) 16811da177e4SLinus Torvalds return (unsigned long) -EINVAL; 16821da177e4SLinus Torvalds 16838feae131SDavid Howells vma = find_vma_exact(current->mm, addr, old_len); 16846fa5f80bSDavid Howells if (!vma) 16851da177e4SLinus Torvalds return (unsigned long) -EINVAL; 16861da177e4SLinus Torvalds 16876fa5f80bSDavid Howells if (vma->vm_end != vma->vm_start + old_len) 16881da177e4SLinus Torvalds return (unsigned long) -EFAULT; 16891da177e4SLinus Torvalds 16906fa5f80bSDavid Howells if (vma->vm_flags & VM_MAYSHARE) 16911da177e4SLinus Torvalds return (unsigned long) -EPERM; 16921da177e4SLinus Torvalds 16938feae131SDavid Howells if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) 16941da177e4SLinus Torvalds return (unsigned long) -ENOMEM; 16951da177e4SLinus Torvalds 16961da177e4SLinus Torvalds /* all checks complete - do it */ 16976fa5f80bSDavid Howells vma->vm_end = vma->vm_start + new_len; 16986fa5f80bSDavid Howells return vma->vm_start; 16996fa5f80bSDavid Howells } 1700b5073173SPaul Mundt EXPORT_SYMBOL(do_mremap); 17016fa5f80bSDavid Howells 17026a6160a7SHeiko Carstens SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, 17036a6160a7SHeiko Carstens unsigned long, new_len, unsigned long, flags, 17046a6160a7SHeiko Carstens unsigned long, new_addr) 17056fa5f80bSDavid Howells { 17066fa5f80bSDavid Howells unsigned long ret; 17076fa5f80bSDavid Howells 17086fa5f80bSDavid Howells down_write(¤t->mm->mmap_sem); 17096fa5f80bSDavid Howells ret = do_mremap(addr, old_len, new_len, flags, new_addr); 17106fa5f80bSDavid Howells up_write(¤t->mm->mmap_sem); 17116fa5f80bSDavid Howells return ret; 17121da177e4SLinus Torvalds } 17131da177e4SLinus Torvalds 17146aab341eSLinus Torvalds struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 1715deceb6cdSHugh Dickins unsigned int foll_flags) 17161da177e4SLinus Torvalds { 17171da177e4SLinus Torvalds return NULL; 17181da177e4SLinus Torvalds } 17191da177e4SLinus Torvalds 17201da177e4SLinus Torvalds int remap_pfn_range(struct vm_area_struct *vma, unsigned long from, 17211da177e4SLinus Torvalds unsigned long to, unsigned long size, pgprot_t prot) 17221da177e4SLinus Torvalds { 172366aa2b4bSGreg Ungerer vma->vm_start = vma->vm_pgoff << PAGE_SHIFT; 172466aa2b4bSGreg Ungerer return 0; 17251da177e4SLinus Torvalds } 172622c4af40SLuke Yang EXPORT_SYMBOL(remap_pfn_range); 17271da177e4SLinus Torvalds 1728f905bc44SPaul Mundt int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, 1729f905bc44SPaul Mundt unsigned long pgoff) 1730f905bc44SPaul Mundt { 1731f905bc44SPaul Mundt unsigned int size = vma->vm_end - vma->vm_start; 1732f905bc44SPaul Mundt 1733f905bc44SPaul Mundt if (!(vma->vm_flags & VM_USERMAP)) 1734f905bc44SPaul Mundt return -EINVAL; 1735f905bc44SPaul Mundt 1736f905bc44SPaul Mundt vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); 1737f905bc44SPaul Mundt vma->vm_end = vma->vm_start + size; 1738f905bc44SPaul Mundt 1739f905bc44SPaul Mundt return 0; 1740f905bc44SPaul Mundt } 1741f905bc44SPaul Mundt EXPORT_SYMBOL(remap_vmalloc_range); 1742f905bc44SPaul Mundt 17431da177e4SLinus Torvalds void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) 17441da177e4SLinus Torvalds { 17451da177e4SLinus Torvalds } 17461da177e4SLinus Torvalds 17471da177e4SLinus Torvalds unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, 17481da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 17491da177e4SLinus Torvalds { 17501da177e4SLinus Torvalds return -ENOMEM; 17511da177e4SLinus Torvalds } 17521da177e4SLinus Torvalds 17531363c3cdSWolfgang Wander void arch_unmap_area(struct mm_struct *mm, unsigned long addr) 17541da177e4SLinus Torvalds { 17551da177e4SLinus Torvalds } 17561da177e4SLinus Torvalds 17571da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 17581da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, 17591da177e4SLinus Torvalds int even_cows) 17601da177e4SLinus Torvalds { 17611da177e4SLinus Torvalds } 176222c4af40SLuke Yang EXPORT_SYMBOL(unmap_mapping_range); 17631da177e4SLinus Torvalds 17641da177e4SLinus Torvalds /* 1765d56e03cdSDavid Howells * ask for an unmapped area at which to create a mapping on a file 1766d56e03cdSDavid Howells */ 1767d56e03cdSDavid Howells unsigned long get_unmapped_area(struct file *file, unsigned long addr, 1768d56e03cdSDavid Howells unsigned long len, unsigned long pgoff, 1769d56e03cdSDavid Howells unsigned long flags) 1770d56e03cdSDavid Howells { 1771d56e03cdSDavid Howells unsigned long (*get_area)(struct file *, unsigned long, unsigned long, 1772d56e03cdSDavid Howells unsigned long, unsigned long); 1773d56e03cdSDavid Howells 1774d56e03cdSDavid Howells get_area = current->mm->get_unmapped_area; 1775d56e03cdSDavid Howells if (file && file->f_op && file->f_op->get_unmapped_area) 1776d56e03cdSDavid Howells get_area = file->f_op->get_unmapped_area; 1777d56e03cdSDavid Howells 1778d56e03cdSDavid Howells if (!get_area) 1779d56e03cdSDavid Howells return -ENOSYS; 1780d56e03cdSDavid Howells 1781d56e03cdSDavid Howells return get_area(file, addr, len, pgoff, flags); 1782d56e03cdSDavid Howells } 1783d56e03cdSDavid Howells EXPORT_SYMBOL(get_unmapped_area); 1784d56e03cdSDavid Howells 1785d56e03cdSDavid Howells /* 17861da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 17871da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 17881da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 17891da177e4SLinus Torvalds * 17901da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 17911da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 17921da177e4SLinus Torvalds * 17931da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 17941da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 17951da177e4SLinus Torvalds * 17961da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 17971da177e4SLinus Torvalds * 17981da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 17991da177e4SLinus Torvalds * wish to use this logic. 18001da177e4SLinus Torvalds */ 180134b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 18021da177e4SLinus Torvalds { 18031da177e4SLinus Torvalds unsigned long free, allowed; 18041da177e4SLinus Torvalds 18051da177e4SLinus Torvalds vm_acct_memory(pages); 18061da177e4SLinus Torvalds 18071da177e4SLinus Torvalds /* 18081da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 18091da177e4SLinus Torvalds */ 18101da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 18111da177e4SLinus Torvalds return 0; 18121da177e4SLinus Torvalds 18131da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 18141da177e4SLinus Torvalds unsigned long n; 18151da177e4SLinus Torvalds 1816347ce434SChristoph Lameter free = global_page_state(NR_FILE_PAGES); 18171da177e4SLinus Torvalds free += nr_swap_pages; 18181da177e4SLinus Torvalds 18191da177e4SLinus Torvalds /* 18201da177e4SLinus Torvalds * Any slabs which are created with the 18211da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 18221da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 18231da177e4SLinus Torvalds * cache and most inode caches should fall into this 18241da177e4SLinus Torvalds */ 1825972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 18261da177e4SLinus Torvalds 18271da177e4SLinus Torvalds /* 18281da177e4SLinus Torvalds * Leave the last 3% for root 18291da177e4SLinus Torvalds */ 18301da177e4SLinus Torvalds if (!cap_sys_admin) 18311da177e4SLinus Torvalds free -= free / 32; 18321da177e4SLinus Torvalds 18331da177e4SLinus Torvalds if (free > pages) 18341da177e4SLinus Torvalds return 0; 18351da177e4SLinus Torvalds 18361da177e4SLinus Torvalds /* 18371da177e4SLinus Torvalds * nr_free_pages() is very expensive on large systems, 18381da177e4SLinus Torvalds * only call if we're about to fail. 18391da177e4SLinus Torvalds */ 18401da177e4SLinus Torvalds n = nr_free_pages(); 1841d5ddc79bSHideo AOKI 1842d5ddc79bSHideo AOKI /* 1843d5ddc79bSHideo AOKI * Leave reserved pages. The pages are not for anonymous pages. 1844d5ddc79bSHideo AOKI */ 1845d5ddc79bSHideo AOKI if (n <= totalreserve_pages) 1846d5ddc79bSHideo AOKI goto error; 1847d5ddc79bSHideo AOKI else 1848d5ddc79bSHideo AOKI n -= totalreserve_pages; 1849d5ddc79bSHideo AOKI 1850d5ddc79bSHideo AOKI /* 1851d5ddc79bSHideo AOKI * Leave the last 3% for root 1852d5ddc79bSHideo AOKI */ 18531da177e4SLinus Torvalds if (!cap_sys_admin) 18541da177e4SLinus Torvalds n -= n / 32; 18551da177e4SLinus Torvalds free += n; 18561da177e4SLinus Torvalds 18571da177e4SLinus Torvalds if (free > pages) 18581da177e4SLinus Torvalds return 0; 1859d5ddc79bSHideo AOKI 1860d5ddc79bSHideo AOKI goto error; 18611da177e4SLinus Torvalds } 18621da177e4SLinus Torvalds 18631da177e4SLinus Torvalds allowed = totalram_pages * sysctl_overcommit_ratio / 100; 18641da177e4SLinus Torvalds /* 18651da177e4SLinus Torvalds * Leave the last 3% for root 18661da177e4SLinus Torvalds */ 18671da177e4SLinus Torvalds if (!cap_sys_admin) 18681da177e4SLinus Torvalds allowed -= allowed / 32; 18691da177e4SLinus Torvalds allowed += total_swap_pages; 18701da177e4SLinus Torvalds 18711da177e4SLinus Torvalds /* Don't let a single process grow too big: 18721da177e4SLinus Torvalds leave 3% of the size of this process for other processes */ 1873731572d3SAlan Cox if (mm) 1874731572d3SAlan Cox allowed -= mm->total_vm / 32; 18751da177e4SLinus Torvalds 187600a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 18771da177e4SLinus Torvalds return 0; 187800a62ce9SKOSAKI Motohiro 1879d5ddc79bSHideo AOKI error: 18801da177e4SLinus Torvalds vm_unacct_memory(pages); 18811da177e4SLinus Torvalds 18821da177e4SLinus Torvalds return -ENOMEM; 18831da177e4SLinus Torvalds } 18841da177e4SLinus Torvalds 18851da177e4SLinus Torvalds int in_gate_area_no_task(unsigned long addr) 18861da177e4SLinus Torvalds { 18871da177e4SLinus Torvalds return 0; 18881da177e4SLinus Torvalds } 1889b0e15190SDavid Howells 1890d0217ac0SNick Piggin int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1891b0e15190SDavid Howells { 1892b0e15190SDavid Howells BUG(); 1893d0217ac0SNick Piggin return 0; 1894b0e15190SDavid Howells } 1895b5073173SPaul Mundt EXPORT_SYMBOL(filemap_fault); 18960ec76a11SDavid Howells 18970ec76a11SDavid Howells /* 18980ec76a11SDavid Howells * Access another process' address space. 18990ec76a11SDavid Howells * - source/target buffer must be kernel space 19000ec76a11SDavid Howells */ 19010ec76a11SDavid Howells int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) 19020ec76a11SDavid Howells { 19030ec76a11SDavid Howells struct vm_area_struct *vma; 19040ec76a11SDavid Howells struct mm_struct *mm; 19050ec76a11SDavid Howells 19060ec76a11SDavid Howells if (addr + len < addr) 19070ec76a11SDavid Howells return 0; 19080ec76a11SDavid Howells 19090ec76a11SDavid Howells mm = get_task_mm(tsk); 19100ec76a11SDavid Howells if (!mm) 19110ec76a11SDavid Howells return 0; 19120ec76a11SDavid Howells 19130ec76a11SDavid Howells down_read(&mm->mmap_sem); 19140ec76a11SDavid Howells 19150ec76a11SDavid Howells /* the access must start within one of the target process's mappings */ 19160159b141SDavid Howells vma = find_vma(mm, addr); 19170159b141SDavid Howells if (vma) { 19180ec76a11SDavid Howells /* don't overrun this mapping */ 19190ec76a11SDavid Howells if (addr + len >= vma->vm_end) 19200ec76a11SDavid Howells len = vma->vm_end - addr; 19210ec76a11SDavid Howells 19220ec76a11SDavid Howells /* only read or write mappings where it is permitted */ 1923d00c7b99SDavid Howells if (write && vma->vm_flags & VM_MAYWRITE) 19247959722bSJie Zhang copy_to_user_page(vma, NULL, addr, 19257959722bSJie Zhang (void *) addr, buf, len); 1926d00c7b99SDavid Howells else if (!write && vma->vm_flags & VM_MAYREAD) 19277959722bSJie Zhang copy_from_user_page(vma, NULL, addr, 19287959722bSJie Zhang buf, (void *) addr, len); 19290ec76a11SDavid Howells else 19300ec76a11SDavid Howells len = 0; 19310ec76a11SDavid Howells } else { 19320ec76a11SDavid Howells len = 0; 19330ec76a11SDavid Howells } 19340ec76a11SDavid Howells 19350ec76a11SDavid Howells up_read(&mm->mmap_sem); 19360ec76a11SDavid Howells mmput(mm); 19370ec76a11SDavid Howells return len; 19380ec76a11SDavid Howells } 1939