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> 1329c185e5SPaul Mundt * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org> 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 16b95f1b31SPaul Gortmaker #include <linux/export.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> 251da177e4SLinus Torvalds #include <linux/blkdev.h> 261da177e4SLinus Torvalds #include <linux/backing-dev.h> 271da177e4SLinus Torvalds #include <linux/mount.h> 281da177e4SLinus Torvalds #include <linux/personality.h> 291da177e4SLinus Torvalds #include <linux/security.h> 301da177e4SLinus Torvalds #include <linux/syscalls.h> 31120a795dSAl Viro #include <linux/audit.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 #if 0 408feae131SDavid Howells #define kenter(FMT, ...) \ 418feae131SDavid Howells printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 428feae131SDavid Howells #define kleave(FMT, ...) \ 438feae131SDavid Howells printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 448feae131SDavid Howells #define kdebug(FMT, ...) \ 458feae131SDavid Howells printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) 468feae131SDavid Howells #else 478feae131SDavid Howells #define kenter(FMT, ...) \ 488feae131SDavid Howells no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 498feae131SDavid Howells #define kleave(FMT, ...) \ 508feae131SDavid Howells no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 518feae131SDavid Howells #define kdebug(FMT, ...) \ 528feae131SDavid Howells no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) 538feae131SDavid Howells #endif 541da177e4SLinus Torvalds 551da177e4SLinus Torvalds void *high_memory; 561da177e4SLinus Torvalds struct page *mem_map; 571da177e4SLinus Torvalds unsigned long max_mapnr; 581da177e4SLinus Torvalds unsigned long num_physpages; 594266c97aSHugh Dickins unsigned long highest_memmap_pfn; 6000a62ce9SKOSAKI Motohiro struct percpu_counter vm_committed_as; 611da177e4SLinus Torvalds int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 621da177e4SLinus Torvalds int sysctl_overcommit_ratio = 50; /* default is 50% */ 631da177e4SLinus Torvalds int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; 64fc4d5c29SDavid Howells int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; 651da177e4SLinus Torvalds int heap_stack_gap = 0; 661da177e4SLinus Torvalds 6733e5d769SDavid Howells atomic_long_t mmap_pages_allocated; 688feae131SDavid Howells 691da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 706a04de6dSWu, Bryan EXPORT_SYMBOL(num_physpages); 711da177e4SLinus Torvalds 728feae131SDavid Howells /* list of mapped, potentially shareable regions */ 738feae131SDavid Howells static struct kmem_cache *vm_region_jar; 748feae131SDavid Howells struct rb_root nommu_region_tree = RB_ROOT; 758feae131SDavid Howells DECLARE_RWSEM(nommu_region_sem); 761da177e4SLinus Torvalds 77f0f37e2fSAlexey Dobriyan const struct vm_operations_struct generic_file_vm_ops = { 781da177e4SLinus Torvalds }; 791da177e4SLinus Torvalds 801da177e4SLinus Torvalds /* 811da177e4SLinus Torvalds * Return the total memory allocated for this pointer, not 821da177e4SLinus Torvalds * just what the caller asked for. 831da177e4SLinus Torvalds * 841da177e4SLinus Torvalds * Doesn't have to be accurate, i.e. may have races. 851da177e4SLinus Torvalds */ 861da177e4SLinus Torvalds unsigned int kobjsize(const void *objp) 871da177e4SLinus Torvalds { 881da177e4SLinus Torvalds struct page *page; 891da177e4SLinus Torvalds 904016a139SMichael Hennerich /* 914016a139SMichael Hennerich * If the object we have should not have ksize performed on it, 924016a139SMichael Hennerich * return size of 0 934016a139SMichael Hennerich */ 945a1603beSPaul Mundt if (!objp || !virt_addr_valid(objp)) 956cfd53fcSPaul Mundt return 0; 966cfd53fcSPaul Mundt 976cfd53fcSPaul Mundt page = virt_to_head_page(objp); 986cfd53fcSPaul Mundt 996cfd53fcSPaul Mundt /* 1006cfd53fcSPaul Mundt * If the allocator sets PageSlab, we know the pointer came from 1016cfd53fcSPaul Mundt * kmalloc(). 1026cfd53fcSPaul Mundt */ 1031da177e4SLinus Torvalds if (PageSlab(page)) 1041da177e4SLinus Torvalds return ksize(objp); 1051da177e4SLinus Torvalds 1066cfd53fcSPaul Mundt /* 107ab2e83eaSPaul Mundt * If it's not a compound page, see if we have a matching VMA 108ab2e83eaSPaul Mundt * region. This test is intentionally done in reverse order, 109ab2e83eaSPaul Mundt * so if there's no VMA, we still fall through and hand back 110ab2e83eaSPaul Mundt * PAGE_SIZE for 0-order pages. 111ab2e83eaSPaul Mundt */ 112ab2e83eaSPaul Mundt if (!PageCompound(page)) { 113ab2e83eaSPaul Mundt struct vm_area_struct *vma; 114ab2e83eaSPaul Mundt 115ab2e83eaSPaul Mundt vma = find_vma(current->mm, (unsigned long)objp); 116ab2e83eaSPaul Mundt if (vma) 117ab2e83eaSPaul Mundt return vma->vm_end - vma->vm_start; 118ab2e83eaSPaul Mundt } 119ab2e83eaSPaul Mundt 120ab2e83eaSPaul Mundt /* 1216cfd53fcSPaul Mundt * The ksize() function is only guaranteed to work for pointers 1225a1603beSPaul Mundt * returned by kmalloc(). So handle arbitrary pointers here. 1236cfd53fcSPaul Mundt */ 1245a1603beSPaul Mundt return PAGE_SIZE << compound_order(page); 1251da177e4SLinus Torvalds } 1261da177e4SLinus Torvalds 127b291f000SNick Piggin int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 1284266c97aSHugh Dickins unsigned long start, int nr_pages, unsigned int foll_flags, 12953a7706dSMichel Lespinasse struct page **pages, struct vm_area_struct **vmas, 13053a7706dSMichel Lespinasse int *retry) 1311da177e4SLinus Torvalds { 132910e46daSSonic Zhang struct vm_area_struct *vma; 1337b4d5b8bSDavid Howells unsigned long vm_flags; 1347b4d5b8bSDavid Howells int i; 1357b4d5b8bSDavid Howells 1367b4d5b8bSDavid Howells /* calculate required read or write permissions. 13758fa879eSHugh Dickins * If FOLL_FORCE is set, we only require the "MAY" flags. 1387b4d5b8bSDavid Howells */ 13958fa879eSHugh Dickins vm_flags = (foll_flags & FOLL_WRITE) ? 14058fa879eSHugh Dickins (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 14158fa879eSHugh Dickins vm_flags &= (foll_flags & FOLL_FORCE) ? 14258fa879eSHugh Dickins (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 1431da177e4SLinus Torvalds 1449d73777eSPeter Zijlstra for (i = 0; i < nr_pages; i++) { 1457561e8caSDavid Howells vma = find_vma(mm, start); 146910e46daSSonic Zhang if (!vma) 1477b4d5b8bSDavid Howells goto finish_or_fault; 1487b4d5b8bSDavid Howells 1497b4d5b8bSDavid Howells /* protect what we can, including chardevs */ 1501c3aff1cSHugh Dickins if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || 1511c3aff1cSHugh Dickins !(vm_flags & vma->vm_flags)) 1527b4d5b8bSDavid Howells goto finish_or_fault; 153910e46daSSonic Zhang 1541da177e4SLinus Torvalds if (pages) { 1551da177e4SLinus Torvalds pages[i] = virt_to_page(start); 1561da177e4SLinus Torvalds if (pages[i]) 1571da177e4SLinus Torvalds page_cache_get(pages[i]); 1581da177e4SLinus Torvalds } 1591da177e4SLinus Torvalds if (vmas) 160910e46daSSonic Zhang vmas[i] = vma; 161e1ee65d8SDavid Howells start = (start + PAGE_SIZE) & PAGE_MASK; 1621da177e4SLinus Torvalds } 1637b4d5b8bSDavid Howells 1647b4d5b8bSDavid Howells return i; 1657b4d5b8bSDavid Howells 1667b4d5b8bSDavid Howells finish_or_fault: 1677b4d5b8bSDavid Howells return i ? : -EFAULT; 1681da177e4SLinus Torvalds } 169b291f000SNick Piggin 170b291f000SNick Piggin /* 171b291f000SNick Piggin * get a list of pages in an address range belonging to the specified process 172b291f000SNick Piggin * and indicate the VMA that covers each page 173b291f000SNick Piggin * - this is potentially dodgy as we may end incrementing the page count of a 174b291f000SNick Piggin * slab page or a secondary page from a compound page 175b291f000SNick Piggin * - don't permit access to VMAs that don't support it, such as I/O mappings 176b291f000SNick Piggin */ 177b291f000SNick Piggin int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 1789d73777eSPeter Zijlstra unsigned long start, int nr_pages, int write, int force, 179b291f000SNick Piggin struct page **pages, struct vm_area_struct **vmas) 180b291f000SNick Piggin { 181b291f000SNick Piggin int flags = 0; 182b291f000SNick Piggin 183b291f000SNick Piggin if (write) 18458fa879eSHugh Dickins flags |= FOLL_WRITE; 185b291f000SNick Piggin if (force) 18658fa879eSHugh Dickins flags |= FOLL_FORCE; 187b291f000SNick Piggin 18853a7706dSMichel Lespinasse return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, 18953a7706dSMichel Lespinasse NULL); 190b291f000SNick Piggin } 19166aa2b4bSGreg Ungerer EXPORT_SYMBOL(get_user_pages); 19266aa2b4bSGreg Ungerer 193dfc2f91aSPaul Mundt /** 194dfc2f91aSPaul Mundt * follow_pfn - look up PFN at a user virtual address 195dfc2f91aSPaul Mundt * @vma: memory mapping 196dfc2f91aSPaul Mundt * @address: user virtual address 197dfc2f91aSPaul Mundt * @pfn: location to store found PFN 198dfc2f91aSPaul Mundt * 199dfc2f91aSPaul Mundt * Only IO mappings and raw PFN mappings are allowed. 200dfc2f91aSPaul Mundt * 201dfc2f91aSPaul Mundt * Returns zero and the pfn at @pfn on success, -ve otherwise. 202dfc2f91aSPaul Mundt */ 203dfc2f91aSPaul Mundt int follow_pfn(struct vm_area_struct *vma, unsigned long address, 204dfc2f91aSPaul Mundt unsigned long *pfn) 205dfc2f91aSPaul Mundt { 206dfc2f91aSPaul Mundt if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 207dfc2f91aSPaul Mundt return -EINVAL; 208dfc2f91aSPaul Mundt 209dfc2f91aSPaul Mundt *pfn = address >> PAGE_SHIFT; 210dfc2f91aSPaul Mundt return 0; 211dfc2f91aSPaul Mundt } 212dfc2f91aSPaul Mundt EXPORT_SYMBOL(follow_pfn); 213dfc2f91aSPaul Mundt 2141da177e4SLinus Torvalds DEFINE_RWLOCK(vmlist_lock); 2151da177e4SLinus Torvalds struct vm_struct *vmlist; 2161da177e4SLinus Torvalds 217b3bdda02SChristoph Lameter void vfree(const void *addr) 2181da177e4SLinus Torvalds { 2191da177e4SLinus Torvalds kfree(addr); 2201da177e4SLinus Torvalds } 221b5073173SPaul Mundt EXPORT_SYMBOL(vfree); 2221da177e4SLinus Torvalds 223dd0fc66fSAl Viro void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 2241da177e4SLinus Torvalds { 2251da177e4SLinus Torvalds /* 2268518609dSRobert P. J. Day * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 2278518609dSRobert P. J. Day * returns only a logical address. 2281da177e4SLinus Torvalds */ 22984097518SNick Piggin return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 2301da177e4SLinus Torvalds } 231b5073173SPaul Mundt EXPORT_SYMBOL(__vmalloc); 2321da177e4SLinus Torvalds 233f905bc44SPaul Mundt void *vmalloc_user(unsigned long size) 234f905bc44SPaul Mundt { 235f905bc44SPaul Mundt void *ret; 236f905bc44SPaul Mundt 237f905bc44SPaul Mundt ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 238f905bc44SPaul Mundt PAGE_KERNEL); 239f905bc44SPaul Mundt if (ret) { 240f905bc44SPaul Mundt struct vm_area_struct *vma; 241f905bc44SPaul Mundt 242f905bc44SPaul Mundt down_write(¤t->mm->mmap_sem); 243f905bc44SPaul Mundt vma = find_vma(current->mm, (unsigned long)ret); 244f905bc44SPaul Mundt if (vma) 245f905bc44SPaul Mundt vma->vm_flags |= VM_USERMAP; 246f905bc44SPaul Mundt up_write(¤t->mm->mmap_sem); 247f905bc44SPaul Mundt } 248f905bc44SPaul Mundt 249f905bc44SPaul Mundt return ret; 250f905bc44SPaul Mundt } 251f905bc44SPaul Mundt EXPORT_SYMBOL(vmalloc_user); 252f905bc44SPaul Mundt 253b3bdda02SChristoph Lameter struct page *vmalloc_to_page(const void *addr) 2541da177e4SLinus Torvalds { 2551da177e4SLinus Torvalds return virt_to_page(addr); 2561da177e4SLinus Torvalds } 257b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_to_page); 2581da177e4SLinus Torvalds 259b3bdda02SChristoph Lameter unsigned long vmalloc_to_pfn(const void *addr) 2601da177e4SLinus Torvalds { 2611da177e4SLinus Torvalds return page_to_pfn(virt_to_page(addr)); 2621da177e4SLinus Torvalds } 263b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_to_pfn); 2641da177e4SLinus Torvalds 2651da177e4SLinus Torvalds long vread(char *buf, char *addr, unsigned long count) 2661da177e4SLinus Torvalds { 2671da177e4SLinus Torvalds memcpy(buf, addr, count); 2681da177e4SLinus Torvalds return count; 2691da177e4SLinus Torvalds } 2701da177e4SLinus Torvalds 2711da177e4SLinus Torvalds long vwrite(char *buf, char *addr, unsigned long count) 2721da177e4SLinus Torvalds { 2731da177e4SLinus Torvalds /* Don't allow overflow */ 2741da177e4SLinus Torvalds if ((unsigned long) addr + count < count) 2751da177e4SLinus Torvalds count = -(unsigned long) addr; 2761da177e4SLinus Torvalds 2771da177e4SLinus Torvalds memcpy(addr, buf, count); 2781da177e4SLinus Torvalds return(count); 2791da177e4SLinus Torvalds } 2801da177e4SLinus Torvalds 2811da177e4SLinus Torvalds /* 2821da177e4SLinus Torvalds * vmalloc - allocate virtually continguos memory 2831da177e4SLinus Torvalds * 2841da177e4SLinus Torvalds * @size: allocation size 2851da177e4SLinus Torvalds * 2861da177e4SLinus Torvalds * Allocate enough pages to cover @size from the page level 2871da177e4SLinus Torvalds * allocator and map them into continguos kernel virtual space. 2881da177e4SLinus Torvalds * 289c1c8897fSMichael Opdenacker * For tight control over page level allocator and protection flags 2901da177e4SLinus Torvalds * use __vmalloc() instead. 2911da177e4SLinus Torvalds */ 2921da177e4SLinus Torvalds void *vmalloc(unsigned long size) 2931da177e4SLinus Torvalds { 2941da177e4SLinus Torvalds return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 2951da177e4SLinus Torvalds } 296f6138882SAndrew Morton EXPORT_SYMBOL(vmalloc); 297f6138882SAndrew Morton 298e1ca7788SDave Young /* 299e1ca7788SDave Young * vzalloc - allocate virtually continguos memory with zero fill 300e1ca7788SDave Young * 301e1ca7788SDave Young * @size: allocation size 302e1ca7788SDave Young * 303e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 304e1ca7788SDave Young * allocator and map them into continguos kernel virtual space. 305e1ca7788SDave Young * The memory allocated is set to zero. 306e1ca7788SDave Young * 307e1ca7788SDave Young * For tight control over page level allocator and protection flags 308e1ca7788SDave Young * use __vmalloc() instead. 309e1ca7788SDave Young */ 310e1ca7788SDave Young void *vzalloc(unsigned long size) 311e1ca7788SDave Young { 312e1ca7788SDave Young return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 313e1ca7788SDave Young PAGE_KERNEL); 314e1ca7788SDave Young } 315e1ca7788SDave Young EXPORT_SYMBOL(vzalloc); 316e1ca7788SDave Young 317e1ca7788SDave Young /** 318e1ca7788SDave Young * vmalloc_node - allocate memory on a specific node 319e1ca7788SDave Young * @size: allocation size 320e1ca7788SDave Young * @node: numa node 321e1ca7788SDave Young * 322e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 323e1ca7788SDave Young * allocator and map them into contiguous kernel virtual space. 324e1ca7788SDave Young * 325e1ca7788SDave Young * For tight control over page level allocator and protection flags 326e1ca7788SDave Young * use __vmalloc() instead. 327e1ca7788SDave Young */ 328f6138882SAndrew Morton void *vmalloc_node(unsigned long size, int node) 329f6138882SAndrew Morton { 330f6138882SAndrew Morton return vmalloc(size); 331f6138882SAndrew Morton } 3329a14f653SPaul Mundt EXPORT_SYMBOL(vmalloc_node); 333e1ca7788SDave Young 334e1ca7788SDave Young /** 335e1ca7788SDave Young * vzalloc_node - allocate memory on a specific node with zero fill 336e1ca7788SDave Young * @size: allocation size 337e1ca7788SDave Young * @node: numa node 338e1ca7788SDave Young * 339e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 340e1ca7788SDave Young * allocator and map them into contiguous kernel virtual space. 341e1ca7788SDave Young * The memory allocated is set to zero. 342e1ca7788SDave Young * 343e1ca7788SDave Young * For tight control over page level allocator and protection flags 344e1ca7788SDave Young * use __vmalloc() instead. 345e1ca7788SDave Young */ 346e1ca7788SDave Young void *vzalloc_node(unsigned long size, int node) 347e1ca7788SDave Young { 348e1ca7788SDave Young return vzalloc(size); 349e1ca7788SDave Young } 350e1ca7788SDave Young EXPORT_SYMBOL(vzalloc_node); 3511da177e4SLinus Torvalds 3521af446edSPaul Mundt #ifndef PAGE_KERNEL_EXEC 3531af446edSPaul Mundt # define PAGE_KERNEL_EXEC PAGE_KERNEL 3541af446edSPaul Mundt #endif 3551af446edSPaul Mundt 3561af446edSPaul Mundt /** 3571af446edSPaul Mundt * vmalloc_exec - allocate virtually contiguous, executable memory 3581af446edSPaul Mundt * @size: allocation size 3591af446edSPaul Mundt * 3601af446edSPaul Mundt * Kernel-internal function to allocate enough pages to cover @size 3611af446edSPaul Mundt * the page level allocator and map them into contiguous and 3621af446edSPaul Mundt * executable kernel virtual space. 3631af446edSPaul Mundt * 3641af446edSPaul Mundt * For tight control over page level allocator and protection flags 3651af446edSPaul Mundt * use __vmalloc() instead. 3661af446edSPaul Mundt */ 3671af446edSPaul Mundt 3681af446edSPaul Mundt void *vmalloc_exec(unsigned long size) 3691af446edSPaul Mundt { 3701af446edSPaul Mundt return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); 3711af446edSPaul Mundt } 3721af446edSPaul Mundt 373b5073173SPaul Mundt /** 374b5073173SPaul Mundt * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 3751da177e4SLinus Torvalds * @size: allocation size 3761da177e4SLinus Torvalds * 3771da177e4SLinus Torvalds * Allocate enough 32bit PA addressable pages to cover @size from the 3781da177e4SLinus Torvalds * page level allocator and map them into continguos kernel virtual space. 3791da177e4SLinus Torvalds */ 3801da177e4SLinus Torvalds void *vmalloc_32(unsigned long size) 3811da177e4SLinus Torvalds { 3821da177e4SLinus Torvalds return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 3831da177e4SLinus Torvalds } 384b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_32); 385b5073173SPaul Mundt 386b5073173SPaul Mundt /** 387b5073173SPaul Mundt * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 388b5073173SPaul Mundt * @size: allocation size 389b5073173SPaul Mundt * 390b5073173SPaul Mundt * The resulting memory area is 32bit addressable and zeroed so it can be 391b5073173SPaul Mundt * mapped to userspace without leaking data. 392f905bc44SPaul Mundt * 393f905bc44SPaul Mundt * VM_USERMAP is set on the corresponding VMA so that subsequent calls to 394f905bc44SPaul Mundt * remap_vmalloc_range() are permissible. 395b5073173SPaul Mundt */ 396b5073173SPaul Mundt void *vmalloc_32_user(unsigned long size) 397b5073173SPaul Mundt { 398f905bc44SPaul Mundt /* 399f905bc44SPaul Mundt * We'll have to sort out the ZONE_DMA bits for 64-bit, 400f905bc44SPaul Mundt * but for now this can simply use vmalloc_user() directly. 401f905bc44SPaul Mundt */ 402f905bc44SPaul Mundt return vmalloc_user(size); 403b5073173SPaul Mundt } 404b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_32_user); 4051da177e4SLinus Torvalds 4061da177e4SLinus Torvalds void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 4071da177e4SLinus Torvalds { 4081da177e4SLinus Torvalds BUG(); 4091da177e4SLinus Torvalds return NULL; 4101da177e4SLinus Torvalds } 411b5073173SPaul Mundt EXPORT_SYMBOL(vmap); 4121da177e4SLinus Torvalds 413b3bdda02SChristoph Lameter void vunmap(const void *addr) 4141da177e4SLinus Torvalds { 4151da177e4SLinus Torvalds BUG(); 4161da177e4SLinus Torvalds } 417b5073173SPaul Mundt EXPORT_SYMBOL(vunmap); 4181da177e4SLinus Torvalds 419eb6434d9SPaul Mundt void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) 420eb6434d9SPaul Mundt { 421eb6434d9SPaul Mundt BUG(); 422eb6434d9SPaul Mundt return NULL; 423eb6434d9SPaul Mundt } 424eb6434d9SPaul Mundt EXPORT_SYMBOL(vm_map_ram); 425eb6434d9SPaul Mundt 426eb6434d9SPaul Mundt void vm_unmap_ram(const void *mem, unsigned int count) 427eb6434d9SPaul Mundt { 428eb6434d9SPaul Mundt BUG(); 429eb6434d9SPaul Mundt } 430eb6434d9SPaul Mundt EXPORT_SYMBOL(vm_unmap_ram); 431eb6434d9SPaul Mundt 432eb6434d9SPaul Mundt void vm_unmap_aliases(void) 433eb6434d9SPaul Mundt { 434eb6434d9SPaul Mundt } 435eb6434d9SPaul Mundt EXPORT_SYMBOL_GPL(vm_unmap_aliases); 436eb6434d9SPaul Mundt 4371da177e4SLinus Torvalds /* 4381eeb66a1SChristoph Hellwig * Implement a stub for vmalloc_sync_all() if the architecture chose not to 4391eeb66a1SChristoph Hellwig * have one. 4401eeb66a1SChristoph Hellwig */ 4411eeb66a1SChristoph Hellwig void __attribute__((weak)) vmalloc_sync_all(void) 4421eeb66a1SChristoph Hellwig { 4431eeb66a1SChristoph Hellwig } 4441eeb66a1SChristoph Hellwig 44529c185e5SPaul Mundt /** 44629c185e5SPaul Mundt * alloc_vm_area - allocate a range of kernel address space 44729c185e5SPaul Mundt * @size: size of the area 44829c185e5SPaul Mundt * 44929c185e5SPaul Mundt * Returns: NULL on failure, vm_struct on success 45029c185e5SPaul Mundt * 45129c185e5SPaul Mundt * This function reserves a range of kernel address space, and 45229c185e5SPaul Mundt * allocates pagetables to map that range. No actual mappings 45329c185e5SPaul Mundt * are created. If the kernel address space is not shared 45429c185e5SPaul Mundt * between processes, it syncs the pagetable across all 45529c185e5SPaul Mundt * processes. 45629c185e5SPaul Mundt */ 457cd12909cSDavid Vrabel struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) 45829c185e5SPaul Mundt { 45929c185e5SPaul Mundt BUG(); 46029c185e5SPaul Mundt return NULL; 46129c185e5SPaul Mundt } 46229c185e5SPaul Mundt EXPORT_SYMBOL_GPL(alloc_vm_area); 46329c185e5SPaul Mundt 46429c185e5SPaul Mundt void free_vm_area(struct vm_struct *area) 46529c185e5SPaul Mundt { 46629c185e5SPaul Mundt BUG(); 46729c185e5SPaul Mundt } 46829c185e5SPaul Mundt EXPORT_SYMBOL_GPL(free_vm_area); 46929c185e5SPaul Mundt 470b5073173SPaul Mundt int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 471b5073173SPaul Mundt struct page *page) 472b5073173SPaul Mundt { 473b5073173SPaul Mundt return -EINVAL; 474b5073173SPaul Mundt } 475b5073173SPaul Mundt EXPORT_SYMBOL(vm_insert_page); 476b5073173SPaul Mundt 4771eeb66a1SChristoph Hellwig /* 4781da177e4SLinus Torvalds * sys_brk() for the most part doesn't need the global kernel 4791da177e4SLinus Torvalds * lock, except when an application is doing something nasty 4801da177e4SLinus Torvalds * like trying to un-brk an area that has already been mapped 4811da177e4SLinus Torvalds * to a regular file. in this case, the unmapping will need 4821da177e4SLinus Torvalds * to invoke file system routines that need the global lock. 4831da177e4SLinus Torvalds */ 4846a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 4851da177e4SLinus Torvalds { 4861da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 4871da177e4SLinus Torvalds 4881da177e4SLinus Torvalds if (brk < mm->start_brk || brk > mm->context.end_brk) 4891da177e4SLinus Torvalds return mm->brk; 4901da177e4SLinus Torvalds 4911da177e4SLinus Torvalds if (mm->brk == brk) 4921da177e4SLinus Torvalds return mm->brk; 4931da177e4SLinus Torvalds 4941da177e4SLinus Torvalds /* 4951da177e4SLinus Torvalds * Always allow shrinking brk 4961da177e4SLinus Torvalds */ 4971da177e4SLinus Torvalds if (brk <= mm->brk) { 4981da177e4SLinus Torvalds mm->brk = brk; 4991da177e4SLinus Torvalds return brk; 5001da177e4SLinus Torvalds } 5011da177e4SLinus Torvalds 5021da177e4SLinus Torvalds /* 5031da177e4SLinus Torvalds * Ok, looks good - let it rip. 5041da177e4SLinus Torvalds */ 505cfe79c00SMike Frysinger flush_icache_range(mm->brk, brk); 5061da177e4SLinus Torvalds return mm->brk = brk; 5071da177e4SLinus Torvalds } 5081da177e4SLinus Torvalds 5098feae131SDavid Howells /* 5108feae131SDavid Howells * initialise the VMA and region record slabs 5118feae131SDavid Howells */ 5128feae131SDavid Howells void __init mmap_init(void) 5131da177e4SLinus Torvalds { 51400a62ce9SKOSAKI Motohiro int ret; 51500a62ce9SKOSAKI Motohiro 51600a62ce9SKOSAKI Motohiro ret = percpu_counter_init(&vm_committed_as, 0); 51700a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 51833e5d769SDavid Howells vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); 5198feae131SDavid Howells } 5201da177e4SLinus Torvalds 5218feae131SDavid Howells /* 5228feae131SDavid Howells * validate the region tree 5238feae131SDavid Howells * - the caller must hold the region lock 5248feae131SDavid Howells */ 5258feae131SDavid Howells #ifdef CONFIG_DEBUG_NOMMU_REGIONS 5268feae131SDavid Howells static noinline void validate_nommu_regions(void) 5278feae131SDavid Howells { 5288feae131SDavid Howells struct vm_region *region, *last; 5298feae131SDavid Howells struct rb_node *p, *lastp; 5301da177e4SLinus Torvalds 5318feae131SDavid Howells lastp = rb_first(&nommu_region_tree); 5328feae131SDavid Howells if (!lastp) 5338feae131SDavid Howells return; 5348feae131SDavid Howells 5358feae131SDavid Howells last = rb_entry(lastp, struct vm_region, vm_rb); 53633e5d769SDavid Howells BUG_ON(unlikely(last->vm_end <= last->vm_start)); 53733e5d769SDavid Howells BUG_ON(unlikely(last->vm_top < last->vm_end)); 5388feae131SDavid Howells 5398feae131SDavid Howells while ((p = rb_next(lastp))) { 5408feae131SDavid Howells region = rb_entry(p, struct vm_region, vm_rb); 5418feae131SDavid Howells last = rb_entry(lastp, struct vm_region, vm_rb); 5428feae131SDavid Howells 54333e5d769SDavid Howells BUG_ON(unlikely(region->vm_end <= region->vm_start)); 54433e5d769SDavid Howells BUG_ON(unlikely(region->vm_top < region->vm_end)); 54533e5d769SDavid Howells BUG_ON(unlikely(region->vm_start < last->vm_top)); 5468feae131SDavid Howells 5478feae131SDavid Howells lastp = p; 5481da177e4SLinus Torvalds } 5491da177e4SLinus Torvalds } 5508feae131SDavid Howells #else 55133e5d769SDavid Howells static void validate_nommu_regions(void) 55233e5d769SDavid Howells { 55333e5d769SDavid Howells } 5548feae131SDavid Howells #endif 5558feae131SDavid Howells 5568feae131SDavid Howells /* 5578feae131SDavid Howells * add a region into the global tree 5588feae131SDavid Howells */ 5598feae131SDavid Howells static void add_nommu_region(struct vm_region *region) 5608feae131SDavid Howells { 5618feae131SDavid Howells struct vm_region *pregion; 5628feae131SDavid Howells struct rb_node **p, *parent; 5638feae131SDavid Howells 5648feae131SDavid Howells validate_nommu_regions(); 5658feae131SDavid Howells 5668feae131SDavid Howells parent = NULL; 5678feae131SDavid Howells p = &nommu_region_tree.rb_node; 5688feae131SDavid Howells while (*p) { 5698feae131SDavid Howells parent = *p; 5708feae131SDavid Howells pregion = rb_entry(parent, struct vm_region, vm_rb); 5718feae131SDavid Howells if (region->vm_start < pregion->vm_start) 5728feae131SDavid Howells p = &(*p)->rb_left; 5738feae131SDavid Howells else if (region->vm_start > pregion->vm_start) 5748feae131SDavid Howells p = &(*p)->rb_right; 5758feae131SDavid Howells else if (pregion == region) 5768feae131SDavid Howells return; 5778feae131SDavid Howells else 5788feae131SDavid Howells BUG(); 5798feae131SDavid Howells } 5808feae131SDavid Howells 5818feae131SDavid Howells rb_link_node(®ion->vm_rb, parent, p); 5828feae131SDavid Howells rb_insert_color(®ion->vm_rb, &nommu_region_tree); 5838feae131SDavid Howells 5848feae131SDavid Howells validate_nommu_regions(); 5858feae131SDavid Howells } 5868feae131SDavid Howells 5878feae131SDavid Howells /* 5888feae131SDavid Howells * delete a region from the global tree 5898feae131SDavid Howells */ 5908feae131SDavid Howells static void delete_nommu_region(struct vm_region *region) 5918feae131SDavid Howells { 5928feae131SDavid Howells BUG_ON(!nommu_region_tree.rb_node); 5938feae131SDavid Howells 5948feae131SDavid Howells validate_nommu_regions(); 5958feae131SDavid Howells rb_erase(®ion->vm_rb, &nommu_region_tree); 5968feae131SDavid Howells validate_nommu_regions(); 5978feae131SDavid Howells } 5988feae131SDavid Howells 5998feae131SDavid Howells /* 6008feae131SDavid Howells * free a contiguous series of pages 6018feae131SDavid Howells */ 6028feae131SDavid Howells static void free_page_series(unsigned long from, unsigned long to) 6038feae131SDavid Howells { 6048feae131SDavid Howells for (; from < to; from += PAGE_SIZE) { 6058feae131SDavid Howells struct page *page = virt_to_page(from); 6068feae131SDavid Howells 6078feae131SDavid Howells kdebug("- free %lx", from); 60833e5d769SDavid Howells atomic_long_dec(&mmap_pages_allocated); 6098feae131SDavid Howells if (page_count(page) != 1) 61033e5d769SDavid Howells kdebug("free page %p: refcount not one: %d", 61133e5d769SDavid Howells page, page_count(page)); 6128feae131SDavid Howells put_page(page); 6138feae131SDavid Howells } 6148feae131SDavid Howells } 6158feae131SDavid Howells 6168feae131SDavid Howells /* 6178feae131SDavid Howells * release a reference to a region 61833e5d769SDavid Howells * - the caller must hold the region semaphore for writing, which this releases 619dd8632a1SPaul Mundt * - the region may not have been added to the tree yet, in which case vm_top 6208feae131SDavid Howells * will equal vm_start 6218feae131SDavid Howells */ 6228feae131SDavid Howells static void __put_nommu_region(struct vm_region *region) 6238feae131SDavid Howells __releases(nommu_region_sem) 6248feae131SDavid Howells { 6251e2ae599SDavid Howells kenter("%p{%d}", region, region->vm_usage); 6268feae131SDavid Howells 6278feae131SDavid Howells BUG_ON(!nommu_region_tree.rb_node); 6288feae131SDavid Howells 6291e2ae599SDavid Howells if (--region->vm_usage == 0) { 630dd8632a1SPaul Mundt if (region->vm_top > region->vm_start) 6318feae131SDavid Howells delete_nommu_region(region); 6328feae131SDavid Howells up_write(&nommu_region_sem); 6338feae131SDavid Howells 6348feae131SDavid Howells if (region->vm_file) 6358feae131SDavid Howells fput(region->vm_file); 6368feae131SDavid Howells 6378feae131SDavid Howells /* IO memory and memory shared directly out of the pagecache 6388feae131SDavid Howells * from ramfs/tmpfs mustn't be released here */ 6398feae131SDavid Howells if (region->vm_flags & VM_MAPPED_COPY) { 6408feae131SDavid Howells kdebug("free series"); 641dd8632a1SPaul Mundt free_page_series(region->vm_start, region->vm_top); 6428feae131SDavid Howells } 6438feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 6448feae131SDavid Howells } else { 6458feae131SDavid Howells up_write(&nommu_region_sem); 6468feae131SDavid Howells } 6478feae131SDavid Howells } 6488feae131SDavid Howells 6498feae131SDavid Howells /* 6508feae131SDavid Howells * release a reference to a region 6518feae131SDavid Howells */ 6528feae131SDavid Howells static void put_nommu_region(struct vm_region *region) 6538feae131SDavid Howells { 6548feae131SDavid Howells down_write(&nommu_region_sem); 6558feae131SDavid Howells __put_nommu_region(region); 6568feae131SDavid Howells } 6571da177e4SLinus Torvalds 6583034097aSDavid Howells /* 659eb8cdec4SBernd Schmidt * update protection on a vma 660eb8cdec4SBernd Schmidt */ 661eb8cdec4SBernd Schmidt static void protect_vma(struct vm_area_struct *vma, unsigned long flags) 662eb8cdec4SBernd Schmidt { 663eb8cdec4SBernd Schmidt #ifdef CONFIG_MPU 664eb8cdec4SBernd Schmidt struct mm_struct *mm = vma->vm_mm; 665eb8cdec4SBernd Schmidt long start = vma->vm_start & PAGE_MASK; 666eb8cdec4SBernd Schmidt while (start < vma->vm_end) { 667eb8cdec4SBernd Schmidt protect_page(mm, start, flags); 668eb8cdec4SBernd Schmidt start += PAGE_SIZE; 669eb8cdec4SBernd Schmidt } 670eb8cdec4SBernd Schmidt update_protections(mm); 671eb8cdec4SBernd Schmidt #endif 672eb8cdec4SBernd Schmidt } 673eb8cdec4SBernd Schmidt 674eb8cdec4SBernd Schmidt /* 6753034097aSDavid Howells * add a VMA into a process's mm_struct in the appropriate place in the list 6768feae131SDavid Howells * and tree and add to the address space's page tree also if not an anonymous 6778feae131SDavid Howells * page 6783034097aSDavid Howells * - should be called with mm->mmap_sem held writelocked 6793034097aSDavid Howells */ 6808feae131SDavid Howells static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) 6813034097aSDavid Howells { 6826038def0SNamhyung Kim struct vm_area_struct *pvma, *prev; 6838feae131SDavid Howells struct address_space *mapping; 6846038def0SNamhyung Kim struct rb_node **p, *parent, *rb_prev; 6853034097aSDavid Howells 6868feae131SDavid Howells kenter(",%p", vma); 6878feae131SDavid Howells 6888feae131SDavid Howells BUG_ON(!vma->vm_region); 6898feae131SDavid Howells 6908feae131SDavid Howells mm->map_count++; 6918feae131SDavid Howells vma->vm_mm = mm; 6928feae131SDavid Howells 693eb8cdec4SBernd Schmidt protect_vma(vma, vma->vm_flags); 694eb8cdec4SBernd Schmidt 6958feae131SDavid Howells /* add the VMA to the mapping */ 6968feae131SDavid Howells if (vma->vm_file) { 6978feae131SDavid Howells mapping = vma->vm_file->f_mapping; 6988feae131SDavid Howells 699918e556eSDavid Howells mutex_lock(&mapping->i_mmap_mutex); 7008feae131SDavid Howells flush_dcache_mmap_lock(mapping); 7018feae131SDavid Howells vma_prio_tree_insert(vma, &mapping->i_mmap); 7028feae131SDavid Howells flush_dcache_mmap_unlock(mapping); 703918e556eSDavid Howells mutex_unlock(&mapping->i_mmap_mutex); 7048feae131SDavid Howells } 7058feae131SDavid Howells 7068feae131SDavid Howells /* add the VMA to the tree */ 7076038def0SNamhyung Kim parent = rb_prev = NULL; 7088feae131SDavid Howells p = &mm->mm_rb.rb_node; 7098feae131SDavid Howells while (*p) { 7108feae131SDavid Howells parent = *p; 7118feae131SDavid Howells pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 7128feae131SDavid Howells 7138feae131SDavid Howells /* sort by: start addr, end addr, VMA struct addr in that order 7148feae131SDavid Howells * (the latter is necessary as we may get identical VMAs) */ 7158feae131SDavid Howells if (vma->vm_start < pvma->vm_start) 7168feae131SDavid Howells p = &(*p)->rb_left; 7176038def0SNamhyung Kim else if (vma->vm_start > pvma->vm_start) { 7186038def0SNamhyung Kim rb_prev = parent; 7198feae131SDavid Howells p = &(*p)->rb_right; 7206038def0SNamhyung Kim } else if (vma->vm_end < pvma->vm_end) 7218feae131SDavid Howells p = &(*p)->rb_left; 7226038def0SNamhyung Kim else if (vma->vm_end > pvma->vm_end) { 7236038def0SNamhyung Kim rb_prev = parent; 7248feae131SDavid Howells p = &(*p)->rb_right; 7256038def0SNamhyung Kim } else if (vma < pvma) 7268feae131SDavid Howells p = &(*p)->rb_left; 7276038def0SNamhyung Kim else if (vma > pvma) { 7286038def0SNamhyung Kim rb_prev = parent; 7298feae131SDavid Howells p = &(*p)->rb_right; 7306038def0SNamhyung Kim } else 7318feae131SDavid Howells BUG(); 7328feae131SDavid Howells } 7338feae131SDavid Howells 7348feae131SDavid Howells rb_link_node(&vma->vm_rb, parent, p); 7358feae131SDavid Howells rb_insert_color(&vma->vm_rb, &mm->mm_rb); 7368feae131SDavid Howells 7378feae131SDavid Howells /* add VMA to the VMA list also */ 7386038def0SNamhyung Kim prev = NULL; 7396038def0SNamhyung Kim if (rb_prev) 7406038def0SNamhyung Kim prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 7413034097aSDavid Howells 7426038def0SNamhyung Kim __vma_link_list(mm, vma, prev, parent); 7438feae131SDavid Howells } 7448feae131SDavid Howells 7458feae131SDavid Howells /* 7468feae131SDavid Howells * delete a VMA from its owning mm_struct and address space 7478feae131SDavid Howells */ 7488feae131SDavid Howells static void delete_vma_from_mm(struct vm_area_struct *vma) 7498feae131SDavid Howells { 7508feae131SDavid Howells struct address_space *mapping; 7518feae131SDavid Howells struct mm_struct *mm = vma->vm_mm; 7528feae131SDavid Howells 7538feae131SDavid Howells kenter("%p", vma); 7548feae131SDavid Howells 755eb8cdec4SBernd Schmidt protect_vma(vma, 0); 756eb8cdec4SBernd Schmidt 7578feae131SDavid Howells mm->map_count--; 7588feae131SDavid Howells if (mm->mmap_cache == vma) 7598feae131SDavid Howells mm->mmap_cache = NULL; 7608feae131SDavid Howells 7618feae131SDavid Howells /* remove the VMA from the mapping */ 7628feae131SDavid Howells if (vma->vm_file) { 7638feae131SDavid Howells mapping = vma->vm_file->f_mapping; 7648feae131SDavid Howells 765918e556eSDavid Howells mutex_lock(&mapping->i_mmap_mutex); 7668feae131SDavid Howells flush_dcache_mmap_lock(mapping); 7678feae131SDavid Howells vma_prio_tree_remove(vma, &mapping->i_mmap); 7688feae131SDavid Howells flush_dcache_mmap_unlock(mapping); 769918e556eSDavid Howells mutex_unlock(&mapping->i_mmap_mutex); 7708feae131SDavid Howells } 7718feae131SDavid Howells 7728feae131SDavid Howells /* remove from the MM's tree and list */ 7738feae131SDavid Howells rb_erase(&vma->vm_rb, &mm->mm_rb); 774b951bf2cSNamhyung Kim 775b951bf2cSNamhyung Kim if (vma->vm_prev) 776b951bf2cSNamhyung Kim vma->vm_prev->vm_next = vma->vm_next; 777b951bf2cSNamhyung Kim else 778b951bf2cSNamhyung Kim mm->mmap = vma->vm_next; 779b951bf2cSNamhyung Kim 780b951bf2cSNamhyung Kim if (vma->vm_next) 781b951bf2cSNamhyung Kim vma->vm_next->vm_prev = vma->vm_prev; 7828feae131SDavid Howells } 7838feae131SDavid Howells 7848feae131SDavid Howells /* 7858feae131SDavid Howells * destroy a VMA record 7868feae131SDavid Howells */ 7878feae131SDavid Howells static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) 7888feae131SDavid Howells { 7898feae131SDavid Howells kenter("%p", vma); 7908feae131SDavid Howells if (vma->vm_ops && vma->vm_ops->close) 7918feae131SDavid Howells vma->vm_ops->close(vma); 7928feae131SDavid Howells if (vma->vm_file) { 7938feae131SDavid Howells fput(vma->vm_file); 7948feae131SDavid Howells if (vma->vm_flags & VM_EXECUTABLE) 7958feae131SDavid Howells removed_exe_file_vma(mm); 7968feae131SDavid Howells } 7978feae131SDavid Howells put_nommu_region(vma->vm_region); 7988feae131SDavid Howells kmem_cache_free(vm_area_cachep, vma); 7993034097aSDavid Howells } 8003034097aSDavid Howells 8013034097aSDavid Howells /* 8023034097aSDavid Howells * look up the first VMA in which addr resides, NULL if none 8033034097aSDavid Howells * - should be called with mm->mmap_sem at least held readlocked 8043034097aSDavid Howells */ 8053034097aSDavid Howells struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 8063034097aSDavid Howells { 8078feae131SDavid Howells struct vm_area_struct *vma; 8083034097aSDavid Howells 8098feae131SDavid Howells /* check the cache first */ 8108feae131SDavid Howells vma = mm->mmap_cache; 8118feae131SDavid Howells if (vma && vma->vm_start <= addr && vma->vm_end > addr) 8128feae131SDavid Howells return vma; 8138feae131SDavid Howells 814e922c4c5SNamhyung Kim /* trawl the list (there may be multiple mappings in which addr 8158feae131SDavid Howells * resides) */ 816e922c4c5SNamhyung Kim for (vma = mm->mmap; vma; vma = vma->vm_next) { 8178feae131SDavid Howells if (vma->vm_start > addr) 8188feae131SDavid Howells return NULL; 8198feae131SDavid Howells if (vma->vm_end > addr) { 8208feae131SDavid Howells mm->mmap_cache = vma; 8218feae131SDavid Howells return vma; 8223034097aSDavid Howells } 8238feae131SDavid Howells } 8243034097aSDavid Howells 8253034097aSDavid Howells return NULL; 8263034097aSDavid Howells } 8273034097aSDavid Howells EXPORT_SYMBOL(find_vma); 8283034097aSDavid Howells 8293034097aSDavid Howells /* 830930e652aSDavid Howells * find a VMA 831930e652aSDavid Howells * - we don't extend stack VMAs under NOMMU conditions 832930e652aSDavid Howells */ 833930e652aSDavid Howells struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) 834930e652aSDavid Howells { 8357561e8caSDavid Howells return find_vma(mm, addr); 836930e652aSDavid Howells } 837930e652aSDavid Howells 8388feae131SDavid Howells /* 8398feae131SDavid Howells * expand a stack to a given address 8408feae131SDavid Howells * - not supported under NOMMU conditions 8418feae131SDavid Howells */ 84257c8f63eSGreg Ungerer int expand_stack(struct vm_area_struct *vma, unsigned long address) 84357c8f63eSGreg Ungerer { 84457c8f63eSGreg Ungerer return -ENOMEM; 84557c8f63eSGreg Ungerer } 84657c8f63eSGreg Ungerer 847930e652aSDavid Howells /* 8486fa5f80bSDavid Howells * look up the first VMA exactly that exactly matches addr 8496fa5f80bSDavid Howells * - should be called with mm->mmap_sem at least held readlocked 8506fa5f80bSDavid Howells */ 8518feae131SDavid Howells static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, 8528feae131SDavid Howells unsigned long addr, 8538feae131SDavid Howells unsigned long len) 8541da177e4SLinus Torvalds { 8551da177e4SLinus Torvalds struct vm_area_struct *vma; 8568feae131SDavid Howells unsigned long end = addr + len; 8571da177e4SLinus Torvalds 8588feae131SDavid Howells /* check the cache first */ 8598feae131SDavid Howells vma = mm->mmap_cache; 8608feae131SDavid Howells if (vma && vma->vm_start == addr && vma->vm_end == end) 8611da177e4SLinus Torvalds return vma; 8628feae131SDavid Howells 863e922c4c5SNamhyung Kim /* trawl the list (there may be multiple mappings in which addr 8648feae131SDavid Howells * resides) */ 865e922c4c5SNamhyung Kim for (vma = mm->mmap; vma; vma = vma->vm_next) { 8668feae131SDavid Howells if (vma->vm_start < addr) 8678feae131SDavid Howells continue; 8688feae131SDavid Howells if (vma->vm_start > addr) 8698feae131SDavid Howells return NULL; 8708feae131SDavid Howells if (vma->vm_end == end) { 8718feae131SDavid Howells mm->mmap_cache = vma; 8728feae131SDavid Howells return vma; 8738feae131SDavid Howells } 8741da177e4SLinus Torvalds } 8751da177e4SLinus Torvalds 8761da177e4SLinus Torvalds return NULL; 8771da177e4SLinus Torvalds } 8781da177e4SLinus Torvalds 8793034097aSDavid Howells /* 8801da177e4SLinus Torvalds * determine whether a mapping should be permitted and, if so, what sort of 8811da177e4SLinus Torvalds * mapping we're capable of supporting 8821da177e4SLinus Torvalds */ 8831da177e4SLinus Torvalds static int validate_mmap_request(struct file *file, 8841da177e4SLinus Torvalds unsigned long addr, 8851da177e4SLinus Torvalds unsigned long len, 8861da177e4SLinus Torvalds unsigned long prot, 8871da177e4SLinus Torvalds unsigned long flags, 8881da177e4SLinus Torvalds unsigned long pgoff, 8891da177e4SLinus Torvalds unsigned long *_capabilities) 8901da177e4SLinus Torvalds { 8918feae131SDavid Howells unsigned long capabilities, rlen; 8921da177e4SLinus Torvalds unsigned long reqprot = prot; 8931da177e4SLinus Torvalds int ret; 8941da177e4SLinus Torvalds 8951da177e4SLinus Torvalds /* do the simple checks first */ 89606aab5a3SDavid Howells if (flags & MAP_FIXED) { 8971da177e4SLinus Torvalds printk(KERN_DEBUG 8981da177e4SLinus Torvalds "%d: Can't do fixed-address/overlay mmap of RAM\n", 8991da177e4SLinus Torvalds current->pid); 9001da177e4SLinus Torvalds return -EINVAL; 9011da177e4SLinus Torvalds } 9021da177e4SLinus Torvalds 9031da177e4SLinus Torvalds if ((flags & MAP_TYPE) != MAP_PRIVATE && 9041da177e4SLinus Torvalds (flags & MAP_TYPE) != MAP_SHARED) 9051da177e4SLinus Torvalds return -EINVAL; 9061da177e4SLinus Torvalds 907f81cff0dSMike Frysinger if (!len) 9081da177e4SLinus Torvalds return -EINVAL; 9091da177e4SLinus Torvalds 910f81cff0dSMike Frysinger /* Careful about overflows.. */ 9118feae131SDavid Howells rlen = PAGE_ALIGN(len); 9128feae131SDavid Howells if (!rlen || rlen > TASK_SIZE) 913f81cff0dSMike Frysinger return -ENOMEM; 914f81cff0dSMike Frysinger 9151da177e4SLinus Torvalds /* offset overflow? */ 9168feae131SDavid Howells if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) 917f81cff0dSMike Frysinger return -EOVERFLOW; 9181da177e4SLinus Torvalds 9191da177e4SLinus Torvalds if (file) { 9201da177e4SLinus Torvalds /* validate file mapping requests */ 9211da177e4SLinus Torvalds struct address_space *mapping; 9221da177e4SLinus Torvalds 9231da177e4SLinus Torvalds /* files must support mmap */ 9241da177e4SLinus Torvalds if (!file->f_op || !file->f_op->mmap) 9251da177e4SLinus Torvalds return -ENODEV; 9261da177e4SLinus Torvalds 9271da177e4SLinus Torvalds /* work out if what we've got could possibly be shared 9281da177e4SLinus Torvalds * - we support chardevs that provide their own "memory" 9291da177e4SLinus Torvalds * - we support files/blockdevs that are memory backed 9301da177e4SLinus Torvalds */ 9311da177e4SLinus Torvalds mapping = file->f_mapping; 9321da177e4SLinus Torvalds if (!mapping) 933e9536ae7SJosef Sipek mapping = file->f_path.dentry->d_inode->i_mapping; 9341da177e4SLinus Torvalds 9351da177e4SLinus Torvalds capabilities = 0; 9361da177e4SLinus Torvalds if (mapping && mapping->backing_dev_info) 9371da177e4SLinus Torvalds capabilities = mapping->backing_dev_info->capabilities; 9381da177e4SLinus Torvalds 9391da177e4SLinus Torvalds if (!capabilities) { 9401da177e4SLinus Torvalds /* no explicit capabilities set, so assume some 9411da177e4SLinus Torvalds * defaults */ 942e9536ae7SJosef Sipek switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) { 9431da177e4SLinus Torvalds case S_IFREG: 9441da177e4SLinus Torvalds case S_IFBLK: 9451da177e4SLinus Torvalds capabilities = BDI_CAP_MAP_COPY; 9461da177e4SLinus Torvalds break; 9471da177e4SLinus Torvalds 9481da177e4SLinus Torvalds case S_IFCHR: 9491da177e4SLinus Torvalds capabilities = 9501da177e4SLinus Torvalds BDI_CAP_MAP_DIRECT | 9511da177e4SLinus Torvalds BDI_CAP_READ_MAP | 9521da177e4SLinus Torvalds BDI_CAP_WRITE_MAP; 9531da177e4SLinus Torvalds break; 9541da177e4SLinus Torvalds 9551da177e4SLinus Torvalds default: 9561da177e4SLinus Torvalds return -EINVAL; 9571da177e4SLinus Torvalds } 9581da177e4SLinus Torvalds } 9591da177e4SLinus Torvalds 9601da177e4SLinus Torvalds /* eliminate any capabilities that we can't support on this 9611da177e4SLinus Torvalds * device */ 9621da177e4SLinus Torvalds if (!file->f_op->get_unmapped_area) 9631da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 9641da177e4SLinus Torvalds if (!file->f_op->read) 9651da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_COPY; 9661da177e4SLinus Torvalds 96728d7a6aeSGraff Yang /* The file shall have been opened with read permission. */ 96828d7a6aeSGraff Yang if (!(file->f_mode & FMODE_READ)) 96928d7a6aeSGraff Yang return -EACCES; 97028d7a6aeSGraff Yang 9711da177e4SLinus Torvalds if (flags & MAP_SHARED) { 9721da177e4SLinus Torvalds /* do checks for writing, appending and locking */ 9731da177e4SLinus Torvalds if ((prot & PROT_WRITE) && 9741da177e4SLinus Torvalds !(file->f_mode & FMODE_WRITE)) 9751da177e4SLinus Torvalds return -EACCES; 9761da177e4SLinus Torvalds 977e9536ae7SJosef Sipek if (IS_APPEND(file->f_path.dentry->d_inode) && 9781da177e4SLinus Torvalds (file->f_mode & FMODE_WRITE)) 9791da177e4SLinus Torvalds return -EACCES; 9801da177e4SLinus Torvalds 981e9536ae7SJosef Sipek if (locks_verify_locked(file->f_path.dentry->d_inode)) 9821da177e4SLinus Torvalds return -EAGAIN; 9831da177e4SLinus Torvalds 9841da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_DIRECT)) 9851da177e4SLinus Torvalds return -ENODEV; 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds /* we mustn't privatise shared mappings */ 9881da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_COPY; 9891da177e4SLinus Torvalds } 9901da177e4SLinus Torvalds else { 9911da177e4SLinus Torvalds /* we're going to read the file into private memory we 9921da177e4SLinus Torvalds * allocate */ 9931da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_COPY)) 9941da177e4SLinus Torvalds return -ENODEV; 9951da177e4SLinus Torvalds 9961da177e4SLinus Torvalds /* we don't permit a private writable mapping to be 9971da177e4SLinus Torvalds * shared with the backing device */ 9981da177e4SLinus Torvalds if (prot & PROT_WRITE) 9991da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 10001da177e4SLinus Torvalds } 10011da177e4SLinus Torvalds 10023c7b2045SBernd Schmidt if (capabilities & BDI_CAP_MAP_DIRECT) { 10033c7b2045SBernd Schmidt if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || 10043c7b2045SBernd Schmidt ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || 10053c7b2045SBernd Schmidt ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) 10063c7b2045SBernd Schmidt ) { 10073c7b2045SBernd Schmidt capabilities &= ~BDI_CAP_MAP_DIRECT; 10083c7b2045SBernd Schmidt if (flags & MAP_SHARED) { 10093c7b2045SBernd Schmidt printk(KERN_WARNING 10103c7b2045SBernd Schmidt "MAP_SHARED not completely supported on !MMU\n"); 10113c7b2045SBernd Schmidt return -EINVAL; 10123c7b2045SBernd Schmidt } 10133c7b2045SBernd Schmidt } 10143c7b2045SBernd Schmidt } 10153c7b2045SBernd Schmidt 10161da177e4SLinus Torvalds /* handle executable mappings and implied executable 10171da177e4SLinus Torvalds * mappings */ 1018e9536ae7SJosef Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 10191da177e4SLinus Torvalds if (prot & PROT_EXEC) 10201da177e4SLinus Torvalds return -EPERM; 10211da177e4SLinus Torvalds } 10221da177e4SLinus Torvalds else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { 10231da177e4SLinus Torvalds /* handle implication of PROT_EXEC by PROT_READ */ 10241da177e4SLinus Torvalds if (current->personality & READ_IMPLIES_EXEC) { 10251da177e4SLinus Torvalds if (capabilities & BDI_CAP_EXEC_MAP) 10261da177e4SLinus Torvalds prot |= PROT_EXEC; 10271da177e4SLinus Torvalds } 10281da177e4SLinus Torvalds } 10291da177e4SLinus Torvalds else if ((prot & PROT_READ) && 10301da177e4SLinus Torvalds (prot & PROT_EXEC) && 10311da177e4SLinus Torvalds !(capabilities & BDI_CAP_EXEC_MAP) 10321da177e4SLinus Torvalds ) { 10331da177e4SLinus Torvalds /* backing file is not executable, try to copy */ 10341da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 10351da177e4SLinus Torvalds } 10361da177e4SLinus Torvalds } 10371da177e4SLinus Torvalds else { 10381da177e4SLinus Torvalds /* anonymous mappings are always memory backed and can be 10391da177e4SLinus Torvalds * privately mapped 10401da177e4SLinus Torvalds */ 10411da177e4SLinus Torvalds capabilities = BDI_CAP_MAP_COPY; 10421da177e4SLinus Torvalds 10431da177e4SLinus Torvalds /* handle PROT_EXEC implication by PROT_READ */ 10441da177e4SLinus Torvalds if ((prot & PROT_READ) && 10451da177e4SLinus Torvalds (current->personality & READ_IMPLIES_EXEC)) 10461da177e4SLinus Torvalds prot |= PROT_EXEC; 10471da177e4SLinus Torvalds } 10481da177e4SLinus Torvalds 10491da177e4SLinus Torvalds /* allow the security API to have its say */ 1050ed032189SEric Paris ret = security_file_mmap(file, reqprot, prot, flags, addr, 0); 10511da177e4SLinus Torvalds if (ret < 0) 10521da177e4SLinus Torvalds return ret; 10531da177e4SLinus Torvalds 10541da177e4SLinus Torvalds /* looks okay */ 10551da177e4SLinus Torvalds *_capabilities = capabilities; 10561da177e4SLinus Torvalds return 0; 10571da177e4SLinus Torvalds } 10581da177e4SLinus Torvalds 10591da177e4SLinus Torvalds /* 10601da177e4SLinus Torvalds * we've determined that we can make the mapping, now translate what we 10611da177e4SLinus Torvalds * now know into VMA flags 10621da177e4SLinus Torvalds */ 10631da177e4SLinus Torvalds static unsigned long determine_vm_flags(struct file *file, 10641da177e4SLinus Torvalds unsigned long prot, 10651da177e4SLinus Torvalds unsigned long flags, 10661da177e4SLinus Torvalds unsigned long capabilities) 10671da177e4SLinus Torvalds { 10681da177e4SLinus Torvalds unsigned long vm_flags; 10691da177e4SLinus Torvalds 10701da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); 10711da177e4SLinus Torvalds /* vm_flags |= mm->def_flags; */ 10721da177e4SLinus Torvalds 10731da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_DIRECT)) { 10741da177e4SLinus Torvalds /* attempt to share read-only copies of mapped file chunks */ 10753c7b2045SBernd Schmidt vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 10761da177e4SLinus Torvalds if (file && !(prot & PROT_WRITE)) 10771da177e4SLinus Torvalds vm_flags |= VM_MAYSHARE; 10783c7b2045SBernd Schmidt } else { 10791da177e4SLinus Torvalds /* overlay a shareable mapping on the backing device or inode 10801da177e4SLinus Torvalds * if possible - used for chardevs, ramfs/tmpfs/shmfs and 10811da177e4SLinus Torvalds * romfs/cramfs */ 10823c7b2045SBernd Schmidt vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS); 10831da177e4SLinus Torvalds if (flags & MAP_SHARED) 10843c7b2045SBernd Schmidt vm_flags |= VM_SHARED; 10851da177e4SLinus Torvalds } 10861da177e4SLinus Torvalds 10871da177e4SLinus Torvalds /* refuse to let anyone share private mappings with this process if 10881da177e4SLinus Torvalds * it's being traced - otherwise breakpoints set in it may interfere 10891da177e4SLinus Torvalds * with another untraced process 10901da177e4SLinus Torvalds */ 1091a288eeccSTejun Heo if ((flags & MAP_PRIVATE) && current->ptrace) 10921da177e4SLinus Torvalds vm_flags &= ~VM_MAYSHARE; 10931da177e4SLinus Torvalds 10941da177e4SLinus Torvalds return vm_flags; 10951da177e4SLinus Torvalds } 10961da177e4SLinus Torvalds 10971da177e4SLinus Torvalds /* 10988feae131SDavid Howells * set up a shared mapping on a file (the driver or filesystem provides and 10998feae131SDavid Howells * pins the storage) 11001da177e4SLinus Torvalds */ 11018feae131SDavid Howells static int do_mmap_shared_file(struct vm_area_struct *vma) 11021da177e4SLinus Torvalds { 11031da177e4SLinus Torvalds int ret; 11041da177e4SLinus Torvalds 11051da177e4SLinus Torvalds ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1106dd8632a1SPaul Mundt if (ret == 0) { 1107dd8632a1SPaul Mundt vma->vm_region->vm_top = vma->vm_region->vm_end; 1108645d83c5SDavid Howells return 0; 1109dd8632a1SPaul Mundt } 11101da177e4SLinus Torvalds if (ret != -ENOSYS) 11111da177e4SLinus Torvalds return ret; 11121da177e4SLinus Torvalds 11133fa30460SDavid Howells /* getting -ENOSYS indicates that direct mmap isn't possible (as 11143fa30460SDavid Howells * opposed to tried but failed) so we can only give a suitable error as 11153fa30460SDavid Howells * it's not possible to make a private copy if MAP_SHARED was given */ 11161da177e4SLinus Torvalds return -ENODEV; 11171da177e4SLinus Torvalds } 11181da177e4SLinus Torvalds 11191da177e4SLinus Torvalds /* 11201da177e4SLinus Torvalds * set up a private mapping or an anonymous shared mapping 11211da177e4SLinus Torvalds */ 11228feae131SDavid Howells static int do_mmap_private(struct vm_area_struct *vma, 11238feae131SDavid Howells struct vm_region *region, 1124645d83c5SDavid Howells unsigned long len, 1125645d83c5SDavid Howells unsigned long capabilities) 11261da177e4SLinus Torvalds { 11278feae131SDavid Howells struct page *pages; 1128f67d9b15SBob Liu unsigned long total, point, n; 11291da177e4SLinus Torvalds void *base; 11308feae131SDavid Howells int ret, order; 11311da177e4SLinus Torvalds 11321da177e4SLinus Torvalds /* invoke the file's mapping function so that it can keep track of 11331da177e4SLinus Torvalds * shared mappings on devices or memory 11341da177e4SLinus Torvalds * - VM_MAYSHARE will be set if it may attempt to share 11351da177e4SLinus Torvalds */ 1136645d83c5SDavid Howells if (capabilities & BDI_CAP_MAP_DIRECT) { 11371da177e4SLinus Torvalds ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1138dd8632a1SPaul Mundt if (ret == 0) { 11391da177e4SLinus Torvalds /* shouldn't return success if we're not sharing */ 1140dd8632a1SPaul Mundt BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); 1141dd8632a1SPaul Mundt vma->vm_region->vm_top = vma->vm_region->vm_end; 1142645d83c5SDavid Howells return 0; 11431da177e4SLinus Torvalds } 1144dd8632a1SPaul Mundt if (ret != -ENOSYS) 1145dd8632a1SPaul Mundt return ret; 11461da177e4SLinus Torvalds 11471da177e4SLinus Torvalds /* getting an ENOSYS error indicates that direct mmap isn't 11481da177e4SLinus Torvalds * possible (as opposed to tried but failed) so we'll try to 11491da177e4SLinus Torvalds * make a private copy of the data and map that instead */ 11501da177e4SLinus Torvalds } 11511da177e4SLinus Torvalds 11528feae131SDavid Howells 11531da177e4SLinus Torvalds /* allocate some memory to hold the mapping 11541da177e4SLinus Torvalds * - note that this may not return a page-aligned address if the object 11551da177e4SLinus Torvalds * we're allocating is smaller than a page 11561da177e4SLinus Torvalds */ 1157f67d9b15SBob Liu order = get_order(len); 11588feae131SDavid Howells kdebug("alloc order %d for %lx", order, len); 11598feae131SDavid Howells 11608feae131SDavid Howells pages = alloc_pages(GFP_KERNEL, order); 11618feae131SDavid Howells if (!pages) 11621da177e4SLinus Torvalds goto enomem; 11631da177e4SLinus Torvalds 11648feae131SDavid Howells total = 1 << order; 116533e5d769SDavid Howells atomic_long_add(total, &mmap_pages_allocated); 11661da177e4SLinus Torvalds 1167f67d9b15SBob Liu point = len >> PAGE_SHIFT; 1168dd8632a1SPaul Mundt 1169dd8632a1SPaul Mundt /* we allocated a power-of-2 sized page set, so we may want to trim off 1170dd8632a1SPaul Mundt * the excess */ 1171dd8632a1SPaul Mundt if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { 11728feae131SDavid Howells while (total > point) { 11738feae131SDavid Howells order = ilog2(total - point); 11748feae131SDavid Howells n = 1 << order; 11758feae131SDavid Howells kdebug("shave %lu/%lu @%lu", n, total - point, total); 117633e5d769SDavid Howells atomic_long_sub(n, &mmap_pages_allocated); 11778feae131SDavid Howells total -= n; 11788feae131SDavid Howells set_page_refcounted(pages + total); 11798feae131SDavid Howells __free_pages(pages + total, order); 11808feae131SDavid Howells } 1181dd8632a1SPaul Mundt } 11828feae131SDavid Howells 11838feae131SDavid Howells for (point = 1; point < total; point++) 11848feae131SDavid Howells set_page_refcounted(&pages[point]); 11858feae131SDavid Howells 11868feae131SDavid Howells base = page_address(pages); 11878feae131SDavid Howells region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; 11888feae131SDavid Howells region->vm_start = (unsigned long) base; 1189f67d9b15SBob Liu region->vm_end = region->vm_start + len; 1190dd8632a1SPaul Mundt region->vm_top = region->vm_start + (total << PAGE_SHIFT); 11918feae131SDavid Howells 11928feae131SDavid Howells vma->vm_start = region->vm_start; 11938feae131SDavid Howells vma->vm_end = region->vm_start + len; 11941da177e4SLinus Torvalds 11951da177e4SLinus Torvalds if (vma->vm_file) { 11961da177e4SLinus Torvalds /* read the contents of a file into the copy */ 11971da177e4SLinus Torvalds mm_segment_t old_fs; 11981da177e4SLinus Torvalds loff_t fpos; 11991da177e4SLinus Torvalds 12001da177e4SLinus Torvalds fpos = vma->vm_pgoff; 12011da177e4SLinus Torvalds fpos <<= PAGE_SHIFT; 12021da177e4SLinus Torvalds 12031da177e4SLinus Torvalds old_fs = get_fs(); 12041da177e4SLinus Torvalds set_fs(KERNEL_DS); 1205f67d9b15SBob Liu ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos); 12061da177e4SLinus Torvalds set_fs(old_fs); 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds if (ret < 0) 12091da177e4SLinus Torvalds goto error_free; 12101da177e4SLinus Torvalds 12111da177e4SLinus Torvalds /* clear the last little bit */ 1212f67d9b15SBob Liu if (ret < len) 1213f67d9b15SBob Liu memset(base + ret, 0, len - ret); 12141da177e4SLinus Torvalds 12151da177e4SLinus Torvalds } 12161da177e4SLinus Torvalds 12171da177e4SLinus Torvalds return 0; 12181da177e4SLinus Torvalds 12191da177e4SLinus Torvalds error_free: 12207223bb4aSNamhyung Kim free_page_series(region->vm_start, region->vm_top); 12218feae131SDavid Howells region->vm_start = vma->vm_start = 0; 12228feae131SDavid Howells region->vm_end = vma->vm_end = 0; 1223dd8632a1SPaul Mundt region->vm_top = 0; 12241da177e4SLinus Torvalds return ret; 12251da177e4SLinus Torvalds 12261da177e4SLinus Torvalds enomem: 122705ae6fa3SGreg Ungerer printk("Allocation of length %lu from process %d (%s) failed\n", 122805ae6fa3SGreg Ungerer len, current->pid, current->comm); 12297bf02ea2SDavid Rientjes show_free_areas(0); 12301da177e4SLinus Torvalds return -ENOMEM; 12311da177e4SLinus Torvalds } 12321da177e4SLinus Torvalds 12331da177e4SLinus Torvalds /* 12341da177e4SLinus Torvalds * handle mapping creation for uClinux 12351da177e4SLinus Torvalds */ 12366be5ceb0SLinus Torvalds static unsigned long do_mmap_pgoff(struct file *file, 12371da177e4SLinus Torvalds unsigned long addr, 12381da177e4SLinus Torvalds unsigned long len, 12391da177e4SLinus Torvalds unsigned long prot, 12401da177e4SLinus Torvalds unsigned long flags, 12411da177e4SLinus Torvalds unsigned long pgoff) 12421da177e4SLinus Torvalds { 12438feae131SDavid Howells struct vm_area_struct *vma; 12448feae131SDavid Howells struct vm_region *region; 12451da177e4SLinus Torvalds struct rb_node *rb; 12468feae131SDavid Howells unsigned long capabilities, vm_flags, result; 12471da177e4SLinus Torvalds int ret; 12481da177e4SLinus Torvalds 12498feae131SDavid Howells kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); 12508feae131SDavid Howells 12511da177e4SLinus Torvalds /* decide whether we should attempt the mapping, and if so what sort of 12521da177e4SLinus Torvalds * mapping */ 12531da177e4SLinus Torvalds ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 12541da177e4SLinus Torvalds &capabilities); 12558feae131SDavid Howells if (ret < 0) { 12568feae131SDavid Howells kleave(" = %d [val]", ret); 12571da177e4SLinus Torvalds return ret; 12588feae131SDavid Howells } 12591da177e4SLinus Torvalds 126006aab5a3SDavid Howells /* we ignore the address hint */ 126106aab5a3SDavid Howells addr = 0; 1262f67d9b15SBob Liu len = PAGE_ALIGN(len); 126306aab5a3SDavid Howells 12641da177e4SLinus Torvalds /* we've determined that we can make the mapping, now translate what we 12651da177e4SLinus Torvalds * now know into VMA flags */ 12661da177e4SLinus Torvalds vm_flags = determine_vm_flags(file, prot, flags, capabilities); 12671da177e4SLinus Torvalds 12688feae131SDavid Howells /* we're going to need to record the mapping */ 12698feae131SDavid Howells region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); 12708feae131SDavid Howells if (!region) 12718feae131SDavid Howells goto error_getting_region; 12721da177e4SLinus Torvalds 12738feae131SDavid Howells vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 12748feae131SDavid Howells if (!vma) 12758feae131SDavid Howells goto error_getting_vma; 12761da177e4SLinus Torvalds 12771e2ae599SDavid Howells region->vm_usage = 1; 12788feae131SDavid Howells region->vm_flags = vm_flags; 12798feae131SDavid Howells region->vm_pgoff = pgoff; 12808feae131SDavid Howells 12815beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 12828feae131SDavid Howells vma->vm_flags = vm_flags; 12838feae131SDavid Howells vma->vm_pgoff = pgoff; 12848feae131SDavid Howells 12858feae131SDavid Howells if (file) { 12868feae131SDavid Howells region->vm_file = file; 12878feae131SDavid Howells get_file(file); 12888feae131SDavid Howells vma->vm_file = file; 12898feae131SDavid Howells get_file(file); 12908feae131SDavid Howells if (vm_flags & VM_EXECUTABLE) { 12918feae131SDavid Howells added_exe_file_vma(current->mm); 12928feae131SDavid Howells vma->vm_mm = current->mm; 12938feae131SDavid Howells } 12948feae131SDavid Howells } 12958feae131SDavid Howells 12968feae131SDavid Howells down_write(&nommu_region_sem); 12978feae131SDavid Howells 12988feae131SDavid Howells /* if we want to share, we need to check for regions created by other 12991da177e4SLinus Torvalds * mmap() calls that overlap with our proposed mapping 13008feae131SDavid Howells * - we can only share with a superset match on most regular files 13011da177e4SLinus Torvalds * - shared mappings on character devices and memory backed files are 13021da177e4SLinus Torvalds * permitted to overlap inexactly as far as we are concerned for in 13031da177e4SLinus Torvalds * these cases, sharing is handled in the driver or filesystem rather 13041da177e4SLinus Torvalds * than here 13051da177e4SLinus Torvalds */ 13061da177e4SLinus Torvalds if (vm_flags & VM_MAYSHARE) { 13078feae131SDavid Howells struct vm_region *pregion; 13088feae131SDavid Howells unsigned long pglen, rpglen, pgend, rpgend, start; 13091da177e4SLinus Torvalds 13108feae131SDavid Howells pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 13118feae131SDavid Howells pgend = pgoff + pglen; 1312165b2392SDavid Howells 13138feae131SDavid Howells for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { 13148feae131SDavid Howells pregion = rb_entry(rb, struct vm_region, vm_rb); 13151da177e4SLinus Torvalds 13168feae131SDavid Howells if (!(pregion->vm_flags & VM_MAYSHARE)) 13171da177e4SLinus Torvalds continue; 13181da177e4SLinus Torvalds 13191da177e4SLinus Torvalds /* search for overlapping mappings on the same file */ 13208feae131SDavid Howells if (pregion->vm_file->f_path.dentry->d_inode != 13218feae131SDavid Howells file->f_path.dentry->d_inode) 13221da177e4SLinus Torvalds continue; 13231da177e4SLinus Torvalds 13248feae131SDavid Howells if (pregion->vm_pgoff >= pgend) 13251da177e4SLinus Torvalds continue; 13261da177e4SLinus Torvalds 13278feae131SDavid Howells rpglen = pregion->vm_end - pregion->vm_start; 13288feae131SDavid Howells rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; 13298feae131SDavid Howells rpgend = pregion->vm_pgoff + rpglen; 13308feae131SDavid Howells if (pgoff >= rpgend) 13311da177e4SLinus Torvalds continue; 13321da177e4SLinus Torvalds 13338feae131SDavid Howells /* handle inexactly overlapping matches between 13348feae131SDavid Howells * mappings */ 13358feae131SDavid Howells if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && 13368feae131SDavid Howells !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { 13378feae131SDavid Howells /* new mapping is not a subset of the region */ 13381da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_DIRECT)) 13391da177e4SLinus Torvalds goto sharing_violation; 13401da177e4SLinus Torvalds continue; 13411da177e4SLinus Torvalds } 13421da177e4SLinus Torvalds 13438feae131SDavid Howells /* we've found a region we can share */ 13441e2ae599SDavid Howells pregion->vm_usage++; 13458feae131SDavid Howells vma->vm_region = pregion; 13468feae131SDavid Howells start = pregion->vm_start; 13478feae131SDavid Howells start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; 13488feae131SDavid Howells vma->vm_start = start; 13498feae131SDavid Howells vma->vm_end = start + len; 13501da177e4SLinus Torvalds 13518feae131SDavid Howells if (pregion->vm_flags & VM_MAPPED_COPY) { 13528feae131SDavid Howells kdebug("share copy"); 13538feae131SDavid Howells vma->vm_flags |= VM_MAPPED_COPY; 13548feae131SDavid Howells } else { 13558feae131SDavid Howells kdebug("share mmap"); 13568feae131SDavid Howells ret = do_mmap_shared_file(vma); 13578feae131SDavid Howells if (ret < 0) { 13588feae131SDavid Howells vma->vm_region = NULL; 13598feae131SDavid Howells vma->vm_start = 0; 13608feae131SDavid Howells vma->vm_end = 0; 13611e2ae599SDavid Howells pregion->vm_usage--; 13628feae131SDavid Howells pregion = NULL; 13638feae131SDavid Howells goto error_just_free; 13641da177e4SLinus Torvalds } 13658feae131SDavid Howells } 13668feae131SDavid Howells fput(region->vm_file); 13678feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 13688feae131SDavid Howells region = pregion; 13698feae131SDavid Howells result = start; 13708feae131SDavid Howells goto share; 13718feae131SDavid Howells } 13721da177e4SLinus Torvalds 13731da177e4SLinus Torvalds /* obtain the address at which to make a shared mapping 13741da177e4SLinus Torvalds * - this is the hook for quasi-memory character devices to 13751da177e4SLinus Torvalds * tell us the location of a shared mapping 13761da177e4SLinus Torvalds */ 1377645d83c5SDavid Howells if (capabilities & BDI_CAP_MAP_DIRECT) { 13781da177e4SLinus Torvalds addr = file->f_op->get_unmapped_area(file, addr, len, 13791da177e4SLinus Torvalds pgoff, flags); 1380bb005a59SNamhyung Kim if (IS_ERR_VALUE(addr)) { 13811da177e4SLinus Torvalds ret = addr; 1382bb005a59SNamhyung Kim if (ret != -ENOSYS) 13838feae131SDavid Howells goto error_just_free; 13841da177e4SLinus Torvalds 13851da177e4SLinus Torvalds /* the driver refused to tell us where to site 13861da177e4SLinus Torvalds * the mapping so we'll have to attempt to copy 13871da177e4SLinus Torvalds * it */ 1388bb005a59SNamhyung Kim ret = -ENODEV; 13891da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_COPY)) 13908feae131SDavid Howells goto error_just_free; 13911da177e4SLinus Torvalds 13921da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 13938feae131SDavid Howells } else { 13948feae131SDavid Howells vma->vm_start = region->vm_start = addr; 13958feae131SDavid Howells vma->vm_end = region->vm_end = addr + len; 13961da177e4SLinus Torvalds } 13971da177e4SLinus Torvalds } 13981da177e4SLinus Torvalds } 13991da177e4SLinus Torvalds 14008feae131SDavid Howells vma->vm_region = region; 14011da177e4SLinus Torvalds 1402645d83c5SDavid Howells /* set up the mapping 1403645d83c5SDavid Howells * - the region is filled in if BDI_CAP_MAP_DIRECT is still set 1404645d83c5SDavid Howells */ 14051da177e4SLinus Torvalds if (file && vma->vm_flags & VM_SHARED) 14068feae131SDavid Howells ret = do_mmap_shared_file(vma); 14071da177e4SLinus Torvalds else 1408645d83c5SDavid Howells ret = do_mmap_private(vma, region, len, capabilities); 14091da177e4SLinus Torvalds if (ret < 0) 1410645d83c5SDavid Howells goto error_just_free; 1411645d83c5SDavid Howells add_nommu_region(region); 14128feae131SDavid Howells 1413ea637639SJie Zhang /* clear anonymous mappings that don't ask for uninitialized data */ 1414ea637639SJie Zhang if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) 1415ea637639SJie Zhang memset((void *)region->vm_start, 0, 1416ea637639SJie Zhang region->vm_end - region->vm_start); 1417ea637639SJie Zhang 14181da177e4SLinus Torvalds /* okay... we have a mapping; now we have to register it */ 14198feae131SDavid Howells result = vma->vm_start; 14201da177e4SLinus Torvalds 14211da177e4SLinus Torvalds current->mm->total_vm += len >> PAGE_SHIFT; 14221da177e4SLinus Torvalds 14238feae131SDavid Howells share: 14248feae131SDavid Howells add_vma_to_mm(current->mm, vma); 14251da177e4SLinus Torvalds 1426cfe79c00SMike Frysinger /* we flush the region from the icache only when the first executable 1427cfe79c00SMike Frysinger * mapping of it is made */ 1428cfe79c00SMike Frysinger if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { 1429cfe79c00SMike Frysinger flush_icache_range(region->vm_start, region->vm_end); 1430cfe79c00SMike Frysinger region->vm_icache_flushed = true; 1431cfe79c00SMike Frysinger } 14321da177e4SLinus Torvalds 1433cfe79c00SMike Frysinger up_write(&nommu_region_sem); 14341da177e4SLinus Torvalds 14358feae131SDavid Howells kleave(" = %lx", result); 14368feae131SDavid Howells return result; 14371da177e4SLinus Torvalds 14388feae131SDavid Howells error_just_free: 14398feae131SDavid Howells up_write(&nommu_region_sem); 14408feae131SDavid Howells error: 144189a86402SDavid Howells if (region->vm_file) 14428feae131SDavid Howells fput(region->vm_file); 14438feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 144489a86402SDavid Howells if (vma->vm_file) 14458feae131SDavid Howells fput(vma->vm_file); 14468feae131SDavid Howells if (vma->vm_flags & VM_EXECUTABLE) 14478feae131SDavid Howells removed_exe_file_vma(vma->vm_mm); 14488feae131SDavid Howells kmem_cache_free(vm_area_cachep, vma); 14498feae131SDavid Howells kleave(" = %d", ret); 14501da177e4SLinus Torvalds return ret; 14511da177e4SLinus Torvalds 14521da177e4SLinus Torvalds sharing_violation: 14538feae131SDavid Howells up_write(&nommu_region_sem); 14548feae131SDavid Howells printk(KERN_WARNING "Attempt to share mismatched mappings\n"); 14558feae131SDavid Howells ret = -EINVAL; 14568feae131SDavid Howells goto error; 14571da177e4SLinus Torvalds 14581da177e4SLinus Torvalds error_getting_vma: 14598feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 14608feae131SDavid Howells printk(KERN_WARNING "Allocation of vma for %lu byte allocation" 14618feae131SDavid Howells " from process %d failed\n", 14621da177e4SLinus Torvalds len, current->pid); 14637bf02ea2SDavid Rientjes show_free_areas(0); 14641da177e4SLinus Torvalds return -ENOMEM; 14651da177e4SLinus Torvalds 14668feae131SDavid Howells error_getting_region: 14678feae131SDavid Howells printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" 14688feae131SDavid Howells " from process %d failed\n", 14691da177e4SLinus Torvalds len, current->pid); 14707bf02ea2SDavid Rientjes show_free_areas(0); 14711da177e4SLinus Torvalds return -ENOMEM; 14721da177e4SLinus Torvalds } 14736be5ceb0SLinus Torvalds 14746be5ceb0SLinus Torvalds unsigned long do_mmap(struct file *file, unsigned long addr, 14756be5ceb0SLinus Torvalds unsigned long len, unsigned long prot, 14766be5ceb0SLinus Torvalds unsigned long flag, unsigned long offset) 14776be5ceb0SLinus Torvalds { 14786be5ceb0SLinus Torvalds if (unlikely(offset + PAGE_ALIGN(len) < offset)) 14796be5ceb0SLinus Torvalds return -EINVAL; 14806be5ceb0SLinus Torvalds if (unlikely(offset & ~PAGE_MASK)) 14816be5ceb0SLinus Torvalds return -EINVAL; 14826be5ceb0SLinus Torvalds return do_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT); 14836be5ceb0SLinus Torvalds } 14846be5ceb0SLinus Torvalds EXPORT_SYMBOL(do_mmap); 14856be5ceb0SLinus Torvalds 14866be5ceb0SLinus Torvalds unsigned long vm_mmap(struct file *file, unsigned long addr, 14876be5ceb0SLinus Torvalds unsigned long len, unsigned long prot, 14886be5ceb0SLinus Torvalds unsigned long flag, unsigned long offset) 14896be5ceb0SLinus Torvalds { 14906be5ceb0SLinus Torvalds unsigned long ret; 14916be5ceb0SLinus Torvalds struct mm_struct *mm = current->mm; 14926be5ceb0SLinus Torvalds 14936be5ceb0SLinus Torvalds down_write(&mm->mmap_sem); 14946be5ceb0SLinus Torvalds ret = do_mmap(file, addr, len, prot, flag, offset); 14956be5ceb0SLinus Torvalds up_write(&mm->mmap_sem); 14966be5ceb0SLinus Torvalds return ret; 14976be5ceb0SLinus Torvalds } 14986be5ceb0SLinus Torvalds EXPORT_SYMBOL(vm_mmap); 14991da177e4SLinus Torvalds 150066f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 150166f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 150266f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 150366f0dc48SHugh Dickins { 150466f0dc48SHugh Dickins struct file *file = NULL; 150566f0dc48SHugh Dickins unsigned long retval = -EBADF; 150666f0dc48SHugh Dickins 1507120a795dSAl Viro audit_mmap_fd(fd, flags); 150866f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 150966f0dc48SHugh Dickins file = fget(fd); 151066f0dc48SHugh Dickins if (!file) 151166f0dc48SHugh Dickins goto out; 151266f0dc48SHugh Dickins } 151366f0dc48SHugh Dickins 151466f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 151566f0dc48SHugh Dickins 151666f0dc48SHugh Dickins down_write(¤t->mm->mmap_sem); 151766f0dc48SHugh Dickins retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff); 151866f0dc48SHugh Dickins up_write(¤t->mm->mmap_sem); 151966f0dc48SHugh Dickins 152066f0dc48SHugh Dickins if (file) 152166f0dc48SHugh Dickins fput(file); 152266f0dc48SHugh Dickins out: 152366f0dc48SHugh Dickins return retval; 152466f0dc48SHugh Dickins } 152566f0dc48SHugh Dickins 1526a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1527a4679373SChristoph Hellwig struct mmap_arg_struct { 1528a4679373SChristoph Hellwig unsigned long addr; 1529a4679373SChristoph Hellwig unsigned long len; 1530a4679373SChristoph Hellwig unsigned long prot; 1531a4679373SChristoph Hellwig unsigned long flags; 1532a4679373SChristoph Hellwig unsigned long fd; 1533a4679373SChristoph Hellwig unsigned long offset; 1534a4679373SChristoph Hellwig }; 1535a4679373SChristoph Hellwig 1536a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1537a4679373SChristoph Hellwig { 1538a4679373SChristoph Hellwig struct mmap_arg_struct a; 1539a4679373SChristoph Hellwig 1540a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1541a4679373SChristoph Hellwig return -EFAULT; 1542a4679373SChristoph Hellwig if (a.offset & ~PAGE_MASK) 1543a4679373SChristoph Hellwig return -EINVAL; 1544a4679373SChristoph Hellwig 1545a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1546a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1547a4679373SChristoph Hellwig } 1548a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1549a4679373SChristoph Hellwig 15501da177e4SLinus Torvalds /* 15518feae131SDavid Howells * split a vma into two pieces at address 'addr', a new vma is allocated either 15528feae131SDavid Howells * for the first part or the tail. 15531da177e4SLinus Torvalds */ 15548feae131SDavid Howells int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 15558feae131SDavid Howells unsigned long addr, int new_below) 15561da177e4SLinus Torvalds { 15578feae131SDavid Howells struct vm_area_struct *new; 15588feae131SDavid Howells struct vm_region *region; 15598feae131SDavid Howells unsigned long npages; 15601da177e4SLinus Torvalds 15618feae131SDavid Howells kenter(""); 15621da177e4SLinus Torvalds 1563779c1023SDavid Howells /* we're only permitted to split anonymous regions (these should have 1564779c1023SDavid Howells * only a single usage on the region) */ 1565779c1023SDavid Howells if (vma->vm_file) 15668feae131SDavid Howells return -ENOMEM; 15671da177e4SLinus Torvalds 15688feae131SDavid Howells if (mm->map_count >= sysctl_max_map_count) 15698feae131SDavid Howells return -ENOMEM; 15701da177e4SLinus Torvalds 15718feae131SDavid Howells region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); 15728feae131SDavid Howells if (!region) 15738feae131SDavid Howells return -ENOMEM; 15748feae131SDavid Howells 15758feae131SDavid Howells new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 15768feae131SDavid Howells if (!new) { 15778feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 15788feae131SDavid Howells return -ENOMEM; 15791da177e4SLinus Torvalds } 15801da177e4SLinus Torvalds 15818feae131SDavid Howells /* most fields are the same, copy all, and then fixup */ 15828feae131SDavid Howells *new = *vma; 15838feae131SDavid Howells *region = *vma->vm_region; 15848feae131SDavid Howells new->vm_region = region; 15858feae131SDavid Howells 15868feae131SDavid Howells npages = (addr - vma->vm_start) >> PAGE_SHIFT; 15878feae131SDavid Howells 15888feae131SDavid Howells if (new_below) { 1589dd8632a1SPaul Mundt region->vm_top = region->vm_end = new->vm_end = addr; 15908feae131SDavid Howells } else { 15918feae131SDavid Howells region->vm_start = new->vm_start = addr; 15928feae131SDavid Howells region->vm_pgoff = new->vm_pgoff += npages; 15931da177e4SLinus Torvalds } 15948feae131SDavid Howells 15958feae131SDavid Howells if (new->vm_ops && new->vm_ops->open) 15968feae131SDavid Howells new->vm_ops->open(new); 15978feae131SDavid Howells 15988feae131SDavid Howells delete_vma_from_mm(vma); 15998feae131SDavid Howells down_write(&nommu_region_sem); 16008feae131SDavid Howells delete_nommu_region(vma->vm_region); 16018feae131SDavid Howells if (new_below) { 16028feae131SDavid Howells vma->vm_region->vm_start = vma->vm_start = addr; 16038feae131SDavid Howells vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; 16048feae131SDavid Howells } else { 16058feae131SDavid Howells vma->vm_region->vm_end = vma->vm_end = addr; 1606dd8632a1SPaul Mundt vma->vm_region->vm_top = addr; 16078feae131SDavid Howells } 16088feae131SDavid Howells add_nommu_region(vma->vm_region); 16098feae131SDavid Howells add_nommu_region(new->vm_region); 16108feae131SDavid Howells up_write(&nommu_region_sem); 16118feae131SDavid Howells add_vma_to_mm(mm, vma); 16128feae131SDavid Howells add_vma_to_mm(mm, new); 16138feae131SDavid Howells return 0; 16148feae131SDavid Howells } 16158feae131SDavid Howells 16168feae131SDavid Howells /* 16178feae131SDavid Howells * shrink a VMA by removing the specified chunk from either the beginning or 16188feae131SDavid Howells * the end 16198feae131SDavid Howells */ 16208feae131SDavid Howells static int shrink_vma(struct mm_struct *mm, 16218feae131SDavid Howells struct vm_area_struct *vma, 16228feae131SDavid Howells unsigned long from, unsigned long to) 16238feae131SDavid Howells { 16248feae131SDavid Howells struct vm_region *region; 16258feae131SDavid Howells 16268feae131SDavid Howells kenter(""); 16278feae131SDavid Howells 16288feae131SDavid Howells /* adjust the VMA's pointers, which may reposition it in the MM's tree 16298feae131SDavid Howells * and list */ 16308feae131SDavid Howells delete_vma_from_mm(vma); 16318feae131SDavid Howells if (from > vma->vm_start) 16328feae131SDavid Howells vma->vm_end = from; 16338feae131SDavid Howells else 16348feae131SDavid Howells vma->vm_start = to; 16358feae131SDavid Howells add_vma_to_mm(mm, vma); 16368feae131SDavid Howells 16378feae131SDavid Howells /* cut the backing region down to size */ 16388feae131SDavid Howells region = vma->vm_region; 16391e2ae599SDavid Howells BUG_ON(region->vm_usage != 1); 16408feae131SDavid Howells 16418feae131SDavid Howells down_write(&nommu_region_sem); 16428feae131SDavid Howells delete_nommu_region(region); 1643dd8632a1SPaul Mundt if (from > region->vm_start) { 1644dd8632a1SPaul Mundt to = region->vm_top; 1645dd8632a1SPaul Mundt region->vm_top = region->vm_end = from; 1646dd8632a1SPaul Mundt } else { 16478feae131SDavid Howells region->vm_start = to; 1648dd8632a1SPaul Mundt } 16498feae131SDavid Howells add_nommu_region(region); 16508feae131SDavid Howells up_write(&nommu_region_sem); 16518feae131SDavid Howells 16528feae131SDavid Howells free_page_series(from, to); 16538feae131SDavid Howells return 0; 16541da177e4SLinus Torvalds } 16551da177e4SLinus Torvalds 16563034097aSDavid Howells /* 16573034097aSDavid Howells * release a mapping 16588feae131SDavid Howells * - under NOMMU conditions the chunk to be unmapped must be backed by a single 16598feae131SDavid Howells * VMA, though it need not cover the whole VMA 16603034097aSDavid Howells */ 16618feae131SDavid Howells int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 16621da177e4SLinus Torvalds { 16638feae131SDavid Howells struct vm_area_struct *vma; 1664f67d9b15SBob Liu unsigned long end; 16658feae131SDavid Howells int ret; 16661da177e4SLinus Torvalds 16678feae131SDavid Howells kenter(",%lx,%zx", start, len); 16681da177e4SLinus Torvalds 1669f67d9b15SBob Liu len = PAGE_ALIGN(len); 16708feae131SDavid Howells if (len == 0) 16711da177e4SLinus Torvalds return -EINVAL; 16721da177e4SLinus Torvalds 1673f67d9b15SBob Liu end = start + len; 1674f67d9b15SBob Liu 16758feae131SDavid Howells /* find the first potentially overlapping VMA */ 16768feae131SDavid Howells vma = find_vma(mm, start); 16778feae131SDavid Howells if (!vma) { 167833e5d769SDavid Howells static int limit = 0; 167933e5d769SDavid Howells if (limit < 5) { 16808feae131SDavid Howells printk(KERN_WARNING 168133e5d769SDavid Howells "munmap of memory not mmapped by process %d" 168233e5d769SDavid Howells " (%s): 0x%lx-0x%lx\n", 168333e5d769SDavid Howells current->pid, current->comm, 168433e5d769SDavid Howells start, start + len - 1); 168533e5d769SDavid Howells limit++; 168633e5d769SDavid Howells } 16878feae131SDavid Howells return -EINVAL; 16888feae131SDavid Howells } 16891da177e4SLinus Torvalds 16908feae131SDavid Howells /* we're allowed to split an anonymous VMA but not a file-backed one */ 16918feae131SDavid Howells if (vma->vm_file) { 16928feae131SDavid Howells do { 16938feae131SDavid Howells if (start > vma->vm_start) { 16948feae131SDavid Howells kleave(" = -EINVAL [miss]"); 16958feae131SDavid Howells return -EINVAL; 16968feae131SDavid Howells } 16978feae131SDavid Howells if (end == vma->vm_end) 16988feae131SDavid Howells goto erase_whole_vma; 1699d75a310cSNamhyung Kim vma = vma->vm_next; 1700d75a310cSNamhyung Kim } while (vma); 17018feae131SDavid Howells kleave(" = -EINVAL [split file]"); 17028feae131SDavid Howells return -EINVAL; 17038feae131SDavid Howells } else { 17048feae131SDavid Howells /* the chunk must be a subset of the VMA found */ 17058feae131SDavid Howells if (start == vma->vm_start && end == vma->vm_end) 17068feae131SDavid Howells goto erase_whole_vma; 17078feae131SDavid Howells if (start < vma->vm_start || end > vma->vm_end) { 17088feae131SDavid Howells kleave(" = -EINVAL [superset]"); 17098feae131SDavid Howells return -EINVAL; 17108feae131SDavid Howells } 17118feae131SDavid Howells if (start & ~PAGE_MASK) { 17128feae131SDavid Howells kleave(" = -EINVAL [unaligned start]"); 17138feae131SDavid Howells return -EINVAL; 17148feae131SDavid Howells } 17158feae131SDavid Howells if (end != vma->vm_end && end & ~PAGE_MASK) { 17168feae131SDavid Howells kleave(" = -EINVAL [unaligned split]"); 17178feae131SDavid Howells return -EINVAL; 17188feae131SDavid Howells } 17198feae131SDavid Howells if (start != vma->vm_start && end != vma->vm_end) { 17208feae131SDavid Howells ret = split_vma(mm, vma, start, 1); 17218feae131SDavid Howells if (ret < 0) { 17228feae131SDavid Howells kleave(" = %d [split]", ret); 17238feae131SDavid Howells return ret; 17248feae131SDavid Howells } 17258feae131SDavid Howells } 17268feae131SDavid Howells return shrink_vma(mm, vma, start, end); 17278feae131SDavid Howells } 17281da177e4SLinus Torvalds 17298feae131SDavid Howells erase_whole_vma: 17308feae131SDavid Howells delete_vma_from_mm(vma); 17318feae131SDavid Howells delete_vma(mm, vma); 17328feae131SDavid Howells kleave(" = 0"); 17331da177e4SLinus Torvalds return 0; 17341da177e4SLinus Torvalds } 1735b5073173SPaul Mundt EXPORT_SYMBOL(do_munmap); 17361da177e4SLinus Torvalds 1737*bfce281cSAl Viro int vm_munmap(unsigned long addr, size_t len) 17383034097aSDavid Howells { 1739*bfce281cSAl Viro struct mm_struct *mm = current->mm; 17403034097aSDavid Howells int ret; 17413034097aSDavid Howells 17423034097aSDavid Howells down_write(&mm->mmap_sem); 17433034097aSDavid Howells ret = do_munmap(mm, addr, len); 17443034097aSDavid Howells up_write(&mm->mmap_sem); 17453034097aSDavid Howells return ret; 17463034097aSDavid Howells } 1747a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 1748a46ef99dSLinus Torvalds 1749a46ef99dSLinus Torvalds SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 1750a46ef99dSLinus Torvalds { 1751*bfce281cSAl Viro return vm_munmap(addr, len); 1752a46ef99dSLinus Torvalds } 17533034097aSDavid Howells 17543034097aSDavid Howells /* 17558feae131SDavid Howells * release all the mappings made in a process's VM space 17563034097aSDavid Howells */ 17571da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 17581da177e4SLinus Torvalds { 17598feae131SDavid Howells struct vm_area_struct *vma; 17601da177e4SLinus Torvalds 17618feae131SDavid Howells if (!mm) 17628feae131SDavid Howells return; 17638feae131SDavid Howells 17648feae131SDavid Howells kenter(""); 17651da177e4SLinus Torvalds 17661da177e4SLinus Torvalds mm->total_vm = 0; 17671da177e4SLinus Torvalds 17688feae131SDavid Howells while ((vma = mm->mmap)) { 17698feae131SDavid Howells mm->mmap = vma->vm_next; 17708feae131SDavid Howells delete_vma_from_mm(vma); 17718feae131SDavid Howells delete_vma(mm, vma); 177204c34961SSteven J. Magnani cond_resched(); 17731da177e4SLinus Torvalds } 17741da177e4SLinus Torvalds 17758feae131SDavid Howells kleave(""); 17761da177e4SLinus Torvalds } 17771da177e4SLinus Torvalds 1778e4eb1ff6SLinus Torvalds unsigned long vm_brk(unsigned long addr, unsigned long len) 17791da177e4SLinus Torvalds { 17801da177e4SLinus Torvalds return -ENOMEM; 17811da177e4SLinus Torvalds } 17821da177e4SLinus Torvalds 17831da177e4SLinus Torvalds /* 17846fa5f80bSDavid Howells * expand (or shrink) an existing mapping, potentially moving it at the same 17856fa5f80bSDavid Howells * time (controlled by the MREMAP_MAYMOVE flag and available VM space) 17861da177e4SLinus Torvalds * 17876fa5f80bSDavid Howells * under NOMMU conditions, we only permit changing a mapping's size, and only 17888feae131SDavid Howells * as long as it stays within the region allocated by do_mmap_private() and the 17898feae131SDavid Howells * block is not shareable 17901da177e4SLinus Torvalds * 17916fa5f80bSDavid Howells * MREMAP_FIXED is not supported under NOMMU conditions 17921da177e4SLinus Torvalds */ 17931da177e4SLinus Torvalds unsigned long do_mremap(unsigned long addr, 17941da177e4SLinus Torvalds unsigned long old_len, unsigned long new_len, 17951da177e4SLinus Torvalds unsigned long flags, unsigned long new_addr) 17961da177e4SLinus Torvalds { 17976fa5f80bSDavid Howells struct vm_area_struct *vma; 17981da177e4SLinus Torvalds 17991da177e4SLinus Torvalds /* insanity checks first */ 1800f67d9b15SBob Liu old_len = PAGE_ALIGN(old_len); 1801f67d9b15SBob Liu new_len = PAGE_ALIGN(new_len); 18028feae131SDavid Howells if (old_len == 0 || new_len == 0) 18031da177e4SLinus Torvalds return (unsigned long) -EINVAL; 18041da177e4SLinus Torvalds 18058feae131SDavid Howells if (addr & ~PAGE_MASK) 18068feae131SDavid Howells return -EINVAL; 18078feae131SDavid Howells 18081da177e4SLinus Torvalds if (flags & MREMAP_FIXED && new_addr != addr) 18091da177e4SLinus Torvalds return (unsigned long) -EINVAL; 18101da177e4SLinus Torvalds 18118feae131SDavid Howells vma = find_vma_exact(current->mm, addr, old_len); 18126fa5f80bSDavid Howells if (!vma) 18131da177e4SLinus Torvalds return (unsigned long) -EINVAL; 18141da177e4SLinus Torvalds 18156fa5f80bSDavid Howells if (vma->vm_end != vma->vm_start + old_len) 18161da177e4SLinus Torvalds return (unsigned long) -EFAULT; 18171da177e4SLinus Torvalds 18186fa5f80bSDavid Howells if (vma->vm_flags & VM_MAYSHARE) 18191da177e4SLinus Torvalds return (unsigned long) -EPERM; 18201da177e4SLinus Torvalds 18218feae131SDavid Howells if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) 18221da177e4SLinus Torvalds return (unsigned long) -ENOMEM; 18231da177e4SLinus Torvalds 18241da177e4SLinus Torvalds /* all checks complete - do it */ 18256fa5f80bSDavid Howells vma->vm_end = vma->vm_start + new_len; 18266fa5f80bSDavid Howells return vma->vm_start; 18276fa5f80bSDavid Howells } 1828b5073173SPaul Mundt EXPORT_SYMBOL(do_mremap); 18296fa5f80bSDavid Howells 18306a6160a7SHeiko Carstens SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, 18316a6160a7SHeiko Carstens unsigned long, new_len, unsigned long, flags, 18326a6160a7SHeiko Carstens unsigned long, new_addr) 18336fa5f80bSDavid Howells { 18346fa5f80bSDavid Howells unsigned long ret; 18356fa5f80bSDavid Howells 18366fa5f80bSDavid Howells down_write(¤t->mm->mmap_sem); 18376fa5f80bSDavid Howells ret = do_mremap(addr, old_len, new_len, flags, new_addr); 18386fa5f80bSDavid Howells up_write(¤t->mm->mmap_sem); 18396fa5f80bSDavid Howells return ret; 18401da177e4SLinus Torvalds } 18411da177e4SLinus Torvalds 18426aab341eSLinus Torvalds struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 1843deceb6cdSHugh Dickins unsigned int foll_flags) 18441da177e4SLinus Torvalds { 18451da177e4SLinus Torvalds return NULL; 18461da177e4SLinus Torvalds } 18471da177e4SLinus Torvalds 18488f3b1327SBob Liu int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 18498f3b1327SBob Liu unsigned long pfn, unsigned long size, pgprot_t prot) 18501da177e4SLinus Torvalds { 18518f3b1327SBob Liu if (addr != (pfn << PAGE_SHIFT)) 18528f3b1327SBob Liu return -EINVAL; 18538f3b1327SBob Liu 18548f3b1327SBob Liu vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; 185566aa2b4bSGreg Ungerer return 0; 18561da177e4SLinus Torvalds } 185722c4af40SLuke Yang EXPORT_SYMBOL(remap_pfn_range); 18581da177e4SLinus Torvalds 1859f905bc44SPaul Mundt int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, 1860f905bc44SPaul Mundt unsigned long pgoff) 1861f905bc44SPaul Mundt { 1862f905bc44SPaul Mundt unsigned int size = vma->vm_end - vma->vm_start; 1863f905bc44SPaul Mundt 1864f905bc44SPaul Mundt if (!(vma->vm_flags & VM_USERMAP)) 1865f905bc44SPaul Mundt return -EINVAL; 1866f905bc44SPaul Mundt 1867f905bc44SPaul Mundt vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); 1868f905bc44SPaul Mundt vma->vm_end = vma->vm_start + size; 1869f905bc44SPaul Mundt 1870f905bc44SPaul Mundt return 0; 1871f905bc44SPaul Mundt } 1872f905bc44SPaul Mundt EXPORT_SYMBOL(remap_vmalloc_range); 1873f905bc44SPaul Mundt 18741da177e4SLinus Torvalds unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, 18751da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 18761da177e4SLinus Torvalds { 18771da177e4SLinus Torvalds return -ENOMEM; 18781da177e4SLinus Torvalds } 18791da177e4SLinus Torvalds 18801363c3cdSWolfgang Wander void arch_unmap_area(struct mm_struct *mm, unsigned long addr) 18811da177e4SLinus Torvalds { 18821da177e4SLinus Torvalds } 18831da177e4SLinus Torvalds 18841da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 18851da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, 18861da177e4SLinus Torvalds int even_cows) 18871da177e4SLinus Torvalds { 18881da177e4SLinus Torvalds } 188922c4af40SLuke Yang EXPORT_SYMBOL(unmap_mapping_range); 18901da177e4SLinus Torvalds 18911da177e4SLinus Torvalds /* 18921da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 18931da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 18941da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 18951da177e4SLinus Torvalds * 18961da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 18971da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 18981da177e4SLinus Torvalds * 18991da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 19001da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 19011da177e4SLinus Torvalds * 19021da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 19031da177e4SLinus Torvalds * 19041da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 19051da177e4SLinus Torvalds * wish to use this logic. 19061da177e4SLinus Torvalds */ 190734b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 19081da177e4SLinus Torvalds { 19091da177e4SLinus Torvalds unsigned long free, allowed; 19101da177e4SLinus Torvalds 19111da177e4SLinus Torvalds vm_acct_memory(pages); 19121da177e4SLinus Torvalds 19131da177e4SLinus Torvalds /* 19141da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 19151da177e4SLinus Torvalds */ 19161da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 19171da177e4SLinus Torvalds return 0; 19181da177e4SLinus Torvalds 19191da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1920c15bef30SDmitry Fink free = global_page_state(NR_FREE_PAGES); 1921c15bef30SDmitry Fink free += global_page_state(NR_FILE_PAGES); 19221da177e4SLinus Torvalds 1923c15bef30SDmitry Fink /* 1924c15bef30SDmitry Fink * shmem pages shouldn't be counted as free in this 1925c15bef30SDmitry Fink * case, they can't be purged, only swapped out, and 1926c15bef30SDmitry Fink * that won't affect the overall amount of available 1927c15bef30SDmitry Fink * memory in the system. 1928c15bef30SDmitry Fink */ 1929c15bef30SDmitry Fink free -= global_page_state(NR_SHMEM); 1930c15bef30SDmitry Fink 19311da177e4SLinus Torvalds free += nr_swap_pages; 19321da177e4SLinus Torvalds 19331da177e4SLinus Torvalds /* 19341da177e4SLinus Torvalds * Any slabs which are created with the 19351da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 19361da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 19371da177e4SLinus Torvalds * cache and most inode caches should fall into this 19381da177e4SLinus Torvalds */ 1939972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 19401da177e4SLinus Torvalds 19411da177e4SLinus Torvalds /* 1942c15bef30SDmitry Fink * Leave reserved pages. The pages are not for anonymous pages. 1943c15bef30SDmitry Fink */ 1944c15bef30SDmitry Fink if (free <= totalreserve_pages) 1945c15bef30SDmitry Fink goto error; 1946c15bef30SDmitry Fink else 1947c15bef30SDmitry Fink free -= totalreserve_pages; 1948c15bef30SDmitry Fink 1949c15bef30SDmitry Fink /* 19501da177e4SLinus Torvalds * Leave the last 3% for root 19511da177e4SLinus Torvalds */ 19521da177e4SLinus Torvalds if (!cap_sys_admin) 19531da177e4SLinus Torvalds free -= free / 32; 19541da177e4SLinus Torvalds 19551da177e4SLinus Torvalds if (free > pages) 19561da177e4SLinus Torvalds return 0; 19571da177e4SLinus Torvalds 1958d5ddc79bSHideo AOKI goto error; 19591da177e4SLinus Torvalds } 19601da177e4SLinus Torvalds 19611da177e4SLinus Torvalds allowed = totalram_pages * sysctl_overcommit_ratio / 100; 19621da177e4SLinus Torvalds /* 19631da177e4SLinus Torvalds * Leave the last 3% for root 19641da177e4SLinus Torvalds */ 19651da177e4SLinus Torvalds if (!cap_sys_admin) 19661da177e4SLinus Torvalds allowed -= allowed / 32; 19671da177e4SLinus Torvalds allowed += total_swap_pages; 19681da177e4SLinus Torvalds 19691da177e4SLinus Torvalds /* Don't let a single process grow too big: 19701da177e4SLinus Torvalds leave 3% of the size of this process for other processes */ 1971731572d3SAlan Cox if (mm) 1972731572d3SAlan Cox allowed -= mm->total_vm / 32; 19731da177e4SLinus Torvalds 197400a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 19751da177e4SLinus Torvalds return 0; 197600a62ce9SKOSAKI Motohiro 1977d5ddc79bSHideo AOKI error: 19781da177e4SLinus Torvalds vm_unacct_memory(pages); 19791da177e4SLinus Torvalds 19801da177e4SLinus Torvalds return -ENOMEM; 19811da177e4SLinus Torvalds } 19821da177e4SLinus Torvalds 1983cae5d390SStephen Wilson int in_gate_area_no_mm(unsigned long addr) 19841da177e4SLinus Torvalds { 19851da177e4SLinus Torvalds return 0; 19861da177e4SLinus Torvalds } 1987b0e15190SDavid Howells 1988d0217ac0SNick Piggin int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1989b0e15190SDavid Howells { 1990b0e15190SDavid Howells BUG(); 1991d0217ac0SNick Piggin return 0; 1992b0e15190SDavid Howells } 1993b5073173SPaul Mundt EXPORT_SYMBOL(filemap_fault); 19940ec76a11SDavid Howells 1995f55f199bSMike Frysinger static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 1996f55f199bSMike Frysinger unsigned long addr, void *buf, int len, int write) 19970ec76a11SDavid Howells { 19980ec76a11SDavid Howells struct vm_area_struct *vma; 19990ec76a11SDavid Howells 20000ec76a11SDavid Howells down_read(&mm->mmap_sem); 20010ec76a11SDavid Howells 20020ec76a11SDavid Howells /* the access must start within one of the target process's mappings */ 20030159b141SDavid Howells vma = find_vma(mm, addr); 20040159b141SDavid Howells if (vma) { 20050ec76a11SDavid Howells /* don't overrun this mapping */ 20060ec76a11SDavid Howells if (addr + len >= vma->vm_end) 20070ec76a11SDavid Howells len = vma->vm_end - addr; 20080ec76a11SDavid Howells 20090ec76a11SDavid Howells /* only read or write mappings where it is permitted */ 2010d00c7b99SDavid Howells if (write && vma->vm_flags & VM_MAYWRITE) 20117959722bSJie Zhang copy_to_user_page(vma, NULL, addr, 20127959722bSJie Zhang (void *) addr, buf, len); 2013d00c7b99SDavid Howells else if (!write && vma->vm_flags & VM_MAYREAD) 20147959722bSJie Zhang copy_from_user_page(vma, NULL, addr, 20157959722bSJie Zhang buf, (void *) addr, len); 20160ec76a11SDavid Howells else 20170ec76a11SDavid Howells len = 0; 20180ec76a11SDavid Howells } else { 20190ec76a11SDavid Howells len = 0; 20200ec76a11SDavid Howells } 20210ec76a11SDavid Howells 20220ec76a11SDavid Howells up_read(&mm->mmap_sem); 2023f55f199bSMike Frysinger 2024f55f199bSMike Frysinger return len; 2025f55f199bSMike Frysinger } 2026f55f199bSMike Frysinger 2027f55f199bSMike Frysinger /** 2028f55f199bSMike Frysinger * @access_remote_vm - access another process' address space 2029f55f199bSMike Frysinger * @mm: the mm_struct of the target address space 2030f55f199bSMike Frysinger * @addr: start address to access 2031f55f199bSMike Frysinger * @buf: source or destination buffer 2032f55f199bSMike Frysinger * @len: number of bytes to transfer 2033f55f199bSMike Frysinger * @write: whether the access is a write 2034f55f199bSMike Frysinger * 2035f55f199bSMike Frysinger * The caller must hold a reference on @mm. 2036f55f199bSMike Frysinger */ 2037f55f199bSMike Frysinger int access_remote_vm(struct mm_struct *mm, unsigned long addr, 2038f55f199bSMike Frysinger void *buf, int len, int write) 2039f55f199bSMike Frysinger { 2040f55f199bSMike Frysinger return __access_remote_vm(NULL, mm, addr, buf, len, write); 2041f55f199bSMike Frysinger } 2042f55f199bSMike Frysinger 2043f55f199bSMike Frysinger /* 2044f55f199bSMike Frysinger * Access another process' address space. 2045f55f199bSMike Frysinger * - source/target buffer must be kernel space 2046f55f199bSMike Frysinger */ 2047f55f199bSMike Frysinger int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) 2048f55f199bSMike Frysinger { 2049f55f199bSMike Frysinger struct mm_struct *mm; 2050f55f199bSMike Frysinger 2051f55f199bSMike Frysinger if (addr + len < addr) 2052f55f199bSMike Frysinger return 0; 2053f55f199bSMike Frysinger 2054f55f199bSMike Frysinger mm = get_task_mm(tsk); 2055f55f199bSMike Frysinger if (!mm) 2056f55f199bSMike Frysinger return 0; 2057f55f199bSMike Frysinger 2058f55f199bSMike Frysinger len = __access_remote_vm(tsk, mm, addr, buf, len, write); 2059f55f199bSMike Frysinger 20600ec76a11SDavid Howells mmput(mm); 20610ec76a11SDavid Howells return len; 20620ec76a11SDavid Howells } 20637e660872SDavid Howells 20647e660872SDavid Howells /** 20657e660872SDavid Howells * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode 20667e660872SDavid Howells * @inode: The inode to check 20677e660872SDavid Howells * @size: The current filesize of the inode 20687e660872SDavid Howells * @newsize: The proposed filesize of the inode 20697e660872SDavid Howells * 20707e660872SDavid Howells * Check the shared mappings on an inode on behalf of a shrinking truncate to 20717e660872SDavid Howells * make sure that that any outstanding VMAs aren't broken and then shrink the 20727e660872SDavid Howells * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't 20737e660872SDavid Howells * automatically grant mappings that are too large. 20747e660872SDavid Howells */ 20757e660872SDavid Howells int nommu_shrink_inode_mappings(struct inode *inode, size_t size, 20767e660872SDavid Howells size_t newsize) 20777e660872SDavid Howells { 20787e660872SDavid Howells struct vm_area_struct *vma; 20797e660872SDavid Howells struct prio_tree_iter iter; 20807e660872SDavid Howells struct vm_region *region; 20817e660872SDavid Howells pgoff_t low, high; 20827e660872SDavid Howells size_t r_size, r_top; 20837e660872SDavid Howells 20847e660872SDavid Howells low = newsize >> PAGE_SHIFT; 20857e660872SDavid Howells high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; 20867e660872SDavid Howells 20877e660872SDavid Howells down_write(&nommu_region_sem); 2088918e556eSDavid Howells mutex_lock(&inode->i_mapping->i_mmap_mutex); 20897e660872SDavid Howells 20907e660872SDavid Howells /* search for VMAs that fall within the dead zone */ 20917e660872SDavid Howells vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap, 20927e660872SDavid Howells low, high) { 20937e660872SDavid Howells /* found one - only interested if it's shared out of the page 20947e660872SDavid Howells * cache */ 20957e660872SDavid Howells if (vma->vm_flags & VM_SHARED) { 2096918e556eSDavid Howells mutex_unlock(&inode->i_mapping->i_mmap_mutex); 20977e660872SDavid Howells up_write(&nommu_region_sem); 20987e660872SDavid Howells return -ETXTBSY; /* not quite true, but near enough */ 20997e660872SDavid Howells } 21007e660872SDavid Howells } 21017e660872SDavid Howells 21027e660872SDavid Howells /* reduce any regions that overlap the dead zone - if in existence, 21037e660872SDavid Howells * these will be pointed to by VMAs that don't overlap the dead zone 21047e660872SDavid Howells * 21057e660872SDavid Howells * we don't check for any regions that start beyond the EOF as there 21067e660872SDavid Howells * shouldn't be any 21077e660872SDavid Howells */ 21087e660872SDavid Howells vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap, 21097e660872SDavid Howells 0, ULONG_MAX) { 21107e660872SDavid Howells if (!(vma->vm_flags & VM_SHARED)) 21117e660872SDavid Howells continue; 21127e660872SDavid Howells 21137e660872SDavid Howells region = vma->vm_region; 21147e660872SDavid Howells r_size = region->vm_top - region->vm_start; 21157e660872SDavid Howells r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; 21167e660872SDavid Howells 21177e660872SDavid Howells if (r_top > newsize) { 21187e660872SDavid Howells region->vm_top -= r_top - newsize; 21197e660872SDavid Howells if (region->vm_end > region->vm_top) 21207e660872SDavid Howells region->vm_end = region->vm_top; 21217e660872SDavid Howells } 21227e660872SDavid Howells } 21237e660872SDavid Howells 2124918e556eSDavid Howells mutex_unlock(&inode->i_mapping->i_mmap_mutex); 21257e660872SDavid Howells up_write(&nommu_region_sem); 21267e660872SDavid Howells return 0; 21277e660872SDavid Howells } 2128