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> 32cf4aebc2SClark Williams #include <linux/sched/sysctl.h> 331da177e4SLinus Torvalds 341da177e4SLinus Torvalds #include <asm/uaccess.h> 351da177e4SLinus Torvalds #include <asm/tlb.h> 361da177e4SLinus Torvalds #include <asm/tlbflush.h> 37eb8cdec4SBernd Schmidt #include <asm/mmu_context.h> 388feae131SDavid Howells #include "internal.h" 398feae131SDavid Howells 408feae131SDavid Howells #if 0 418feae131SDavid Howells #define kenter(FMT, ...) \ 428feae131SDavid Howells printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 438feae131SDavid Howells #define kleave(FMT, ...) \ 448feae131SDavid Howells printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 458feae131SDavid Howells #define kdebug(FMT, ...) \ 468feae131SDavid Howells printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) 478feae131SDavid Howells #else 488feae131SDavid Howells #define kenter(FMT, ...) \ 498feae131SDavid Howells no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 508feae131SDavid Howells #define kleave(FMT, ...) \ 518feae131SDavid Howells no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 528feae131SDavid Howells #define kdebug(FMT, ...) \ 538feae131SDavid Howells no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) 548feae131SDavid Howells #endif 551da177e4SLinus Torvalds 561da177e4SLinus Torvalds void *high_memory; 571da177e4SLinus Torvalds struct page *mem_map; 581da177e4SLinus Torvalds unsigned long max_mapnr; 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; 65c9b1d098SAndrew Shewmaker unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ 664eeab4f5SAndrew Shewmaker unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ 671da177e4SLinus Torvalds int heap_stack_gap = 0; 681da177e4SLinus Torvalds 6933e5d769SDavid Howells atomic_long_t mmap_pages_allocated; 708feae131SDavid Howells 71997071bcSK. Y. Srinivasan /* 72997071bcSK. Y. Srinivasan * The global memory commitment made in the system can be a metric 73997071bcSK. Y. Srinivasan * that can be used to drive ballooning decisions when Linux is hosted 74997071bcSK. Y. Srinivasan * as a guest. On Hyper-V, the host implements a policy engine for dynamically 75997071bcSK. Y. Srinivasan * balancing memory across competing virtual machines that are hosted. 76997071bcSK. Y. Srinivasan * Several metrics drive this policy engine including the guest reported 77997071bcSK. Y. Srinivasan * memory commitment. 78997071bcSK. Y. Srinivasan */ 79997071bcSK. Y. Srinivasan unsigned long vm_memory_committed(void) 80997071bcSK. Y. Srinivasan { 81997071bcSK. Y. Srinivasan return percpu_counter_read_positive(&vm_committed_as); 82997071bcSK. Y. Srinivasan } 83997071bcSK. Y. Srinivasan 84997071bcSK. Y. Srinivasan EXPORT_SYMBOL_GPL(vm_memory_committed); 85997071bcSK. Y. Srinivasan 861da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 871da177e4SLinus Torvalds 888feae131SDavid Howells /* list of mapped, potentially shareable regions */ 898feae131SDavid Howells static struct kmem_cache *vm_region_jar; 908feae131SDavid Howells struct rb_root nommu_region_tree = RB_ROOT; 918feae131SDavid Howells DECLARE_RWSEM(nommu_region_sem); 921da177e4SLinus Torvalds 93f0f37e2fSAlexey Dobriyan const struct vm_operations_struct generic_file_vm_ops = { 941da177e4SLinus Torvalds }; 951da177e4SLinus Torvalds 961da177e4SLinus Torvalds /* 971da177e4SLinus Torvalds * Return the total memory allocated for this pointer, not 981da177e4SLinus Torvalds * just what the caller asked for. 991da177e4SLinus Torvalds * 1001da177e4SLinus Torvalds * Doesn't have to be accurate, i.e. may have races. 1011da177e4SLinus Torvalds */ 1021da177e4SLinus Torvalds unsigned int kobjsize(const void *objp) 1031da177e4SLinus Torvalds { 1041da177e4SLinus Torvalds struct page *page; 1051da177e4SLinus Torvalds 1064016a139SMichael Hennerich /* 1074016a139SMichael Hennerich * If the object we have should not have ksize performed on it, 1084016a139SMichael Hennerich * return size of 0 1094016a139SMichael Hennerich */ 1105a1603beSPaul Mundt if (!objp || !virt_addr_valid(objp)) 1116cfd53fcSPaul Mundt return 0; 1126cfd53fcSPaul Mundt 1136cfd53fcSPaul Mundt page = virt_to_head_page(objp); 1146cfd53fcSPaul Mundt 1156cfd53fcSPaul Mundt /* 1166cfd53fcSPaul Mundt * If the allocator sets PageSlab, we know the pointer came from 1176cfd53fcSPaul Mundt * kmalloc(). 1186cfd53fcSPaul Mundt */ 1191da177e4SLinus Torvalds if (PageSlab(page)) 1201da177e4SLinus Torvalds return ksize(objp); 1211da177e4SLinus Torvalds 1226cfd53fcSPaul Mundt /* 123ab2e83eaSPaul Mundt * If it's not a compound page, see if we have a matching VMA 124ab2e83eaSPaul Mundt * region. This test is intentionally done in reverse order, 125ab2e83eaSPaul Mundt * so if there's no VMA, we still fall through and hand back 126ab2e83eaSPaul Mundt * PAGE_SIZE for 0-order pages. 127ab2e83eaSPaul Mundt */ 128ab2e83eaSPaul Mundt if (!PageCompound(page)) { 129ab2e83eaSPaul Mundt struct vm_area_struct *vma; 130ab2e83eaSPaul Mundt 131ab2e83eaSPaul Mundt vma = find_vma(current->mm, (unsigned long)objp); 132ab2e83eaSPaul Mundt if (vma) 133ab2e83eaSPaul Mundt return vma->vm_end - vma->vm_start; 134ab2e83eaSPaul Mundt } 135ab2e83eaSPaul Mundt 136ab2e83eaSPaul Mundt /* 1376cfd53fcSPaul Mundt * The ksize() function is only guaranteed to work for pointers 1385a1603beSPaul Mundt * returned by kmalloc(). So handle arbitrary pointers here. 1396cfd53fcSPaul Mundt */ 1405a1603beSPaul Mundt return PAGE_SIZE << compound_order(page); 1411da177e4SLinus Torvalds } 1421da177e4SLinus Torvalds 14328a35716SMichel Lespinasse long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 14428a35716SMichel Lespinasse unsigned long start, unsigned long nr_pages, 14528a35716SMichel Lespinasse unsigned int foll_flags, struct page **pages, 14628a35716SMichel Lespinasse struct vm_area_struct **vmas, int *nonblocking) 1471da177e4SLinus Torvalds { 148910e46daSSonic Zhang struct vm_area_struct *vma; 1497b4d5b8bSDavid Howells unsigned long vm_flags; 1507b4d5b8bSDavid Howells int i; 1517b4d5b8bSDavid Howells 1527b4d5b8bSDavid Howells /* calculate required read or write permissions. 15358fa879eSHugh Dickins * If FOLL_FORCE is set, we only require the "MAY" flags. 1547b4d5b8bSDavid Howells */ 15558fa879eSHugh Dickins vm_flags = (foll_flags & FOLL_WRITE) ? 15658fa879eSHugh Dickins (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 15758fa879eSHugh Dickins vm_flags &= (foll_flags & FOLL_FORCE) ? 15858fa879eSHugh Dickins (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 1591da177e4SLinus Torvalds 1609d73777eSPeter Zijlstra for (i = 0; i < nr_pages; i++) { 1617561e8caSDavid Howells vma = find_vma(mm, start); 162910e46daSSonic Zhang if (!vma) 1637b4d5b8bSDavid Howells goto finish_or_fault; 1647b4d5b8bSDavid Howells 1657b4d5b8bSDavid Howells /* protect what we can, including chardevs */ 1661c3aff1cSHugh Dickins if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || 1671c3aff1cSHugh Dickins !(vm_flags & vma->vm_flags)) 1687b4d5b8bSDavid Howells goto finish_or_fault; 169910e46daSSonic Zhang 1701da177e4SLinus Torvalds if (pages) { 1711da177e4SLinus Torvalds pages[i] = virt_to_page(start); 1721da177e4SLinus Torvalds if (pages[i]) 1731da177e4SLinus Torvalds page_cache_get(pages[i]); 1741da177e4SLinus Torvalds } 1751da177e4SLinus Torvalds if (vmas) 176910e46daSSonic Zhang vmas[i] = vma; 177e1ee65d8SDavid Howells start = (start + PAGE_SIZE) & PAGE_MASK; 1781da177e4SLinus Torvalds } 1797b4d5b8bSDavid Howells 1807b4d5b8bSDavid Howells return i; 1817b4d5b8bSDavid Howells 1827b4d5b8bSDavid Howells finish_or_fault: 1837b4d5b8bSDavid Howells return i ? : -EFAULT; 1841da177e4SLinus Torvalds } 185b291f000SNick Piggin 186b291f000SNick Piggin /* 187b291f000SNick Piggin * get a list of pages in an address range belonging to the specified process 188b291f000SNick Piggin * and indicate the VMA that covers each page 189b291f000SNick Piggin * - this is potentially dodgy as we may end incrementing the page count of a 190b291f000SNick Piggin * slab page or a secondary page from a compound page 191b291f000SNick Piggin * - don't permit access to VMAs that don't support it, such as I/O mappings 192b291f000SNick Piggin */ 19328a35716SMichel Lespinasse long get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 19428a35716SMichel Lespinasse unsigned long start, unsigned long nr_pages, 19528a35716SMichel Lespinasse int write, int force, struct page **pages, 19628a35716SMichel Lespinasse struct vm_area_struct **vmas) 197b291f000SNick Piggin { 198b291f000SNick Piggin int flags = 0; 199b291f000SNick Piggin 200b291f000SNick Piggin if (write) 20158fa879eSHugh Dickins flags |= FOLL_WRITE; 202b291f000SNick Piggin if (force) 20358fa879eSHugh Dickins flags |= FOLL_FORCE; 204b291f000SNick Piggin 20553a7706dSMichel Lespinasse return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, 20653a7706dSMichel Lespinasse NULL); 207b291f000SNick Piggin } 20866aa2b4bSGreg Ungerer EXPORT_SYMBOL(get_user_pages); 20966aa2b4bSGreg Ungerer 210dfc2f91aSPaul Mundt /** 211dfc2f91aSPaul Mundt * follow_pfn - look up PFN at a user virtual address 212dfc2f91aSPaul Mundt * @vma: memory mapping 213dfc2f91aSPaul Mundt * @address: user virtual address 214dfc2f91aSPaul Mundt * @pfn: location to store found PFN 215dfc2f91aSPaul Mundt * 216dfc2f91aSPaul Mundt * Only IO mappings and raw PFN mappings are allowed. 217dfc2f91aSPaul Mundt * 218dfc2f91aSPaul Mundt * Returns zero and the pfn at @pfn on success, -ve otherwise. 219dfc2f91aSPaul Mundt */ 220dfc2f91aSPaul Mundt int follow_pfn(struct vm_area_struct *vma, unsigned long address, 221dfc2f91aSPaul Mundt unsigned long *pfn) 222dfc2f91aSPaul Mundt { 223dfc2f91aSPaul Mundt if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 224dfc2f91aSPaul Mundt return -EINVAL; 225dfc2f91aSPaul Mundt 226dfc2f91aSPaul Mundt *pfn = address >> PAGE_SHIFT; 227dfc2f91aSPaul Mundt return 0; 228dfc2f91aSPaul Mundt } 229dfc2f91aSPaul Mundt EXPORT_SYMBOL(follow_pfn); 230dfc2f91aSPaul Mundt 231f1c4069eSJoonsoo Kim LIST_HEAD(vmap_area_list); 2321da177e4SLinus Torvalds 233b3bdda02SChristoph Lameter void vfree(const void *addr) 2341da177e4SLinus Torvalds { 2351da177e4SLinus Torvalds kfree(addr); 2361da177e4SLinus Torvalds } 237b5073173SPaul Mundt EXPORT_SYMBOL(vfree); 2381da177e4SLinus Torvalds 239dd0fc66fSAl Viro void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 2401da177e4SLinus Torvalds { 2411da177e4SLinus Torvalds /* 2428518609dSRobert P. J. Day * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 2438518609dSRobert P. J. Day * returns only a logical address. 2441da177e4SLinus Torvalds */ 24584097518SNick Piggin return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 2461da177e4SLinus Torvalds } 247b5073173SPaul Mundt EXPORT_SYMBOL(__vmalloc); 2481da177e4SLinus Torvalds 249f905bc44SPaul Mundt void *vmalloc_user(unsigned long size) 250f905bc44SPaul Mundt { 251f905bc44SPaul Mundt void *ret; 252f905bc44SPaul Mundt 253f905bc44SPaul Mundt ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 254f905bc44SPaul Mundt PAGE_KERNEL); 255f905bc44SPaul Mundt if (ret) { 256f905bc44SPaul Mundt struct vm_area_struct *vma; 257f905bc44SPaul Mundt 258f905bc44SPaul Mundt down_write(¤t->mm->mmap_sem); 259f905bc44SPaul Mundt vma = find_vma(current->mm, (unsigned long)ret); 260f905bc44SPaul Mundt if (vma) 261f905bc44SPaul Mundt vma->vm_flags |= VM_USERMAP; 262f905bc44SPaul Mundt up_write(¤t->mm->mmap_sem); 263f905bc44SPaul Mundt } 264f905bc44SPaul Mundt 265f905bc44SPaul Mundt return ret; 266f905bc44SPaul Mundt } 267f905bc44SPaul Mundt EXPORT_SYMBOL(vmalloc_user); 268f905bc44SPaul Mundt 269b3bdda02SChristoph Lameter struct page *vmalloc_to_page(const void *addr) 2701da177e4SLinus Torvalds { 2711da177e4SLinus Torvalds return virt_to_page(addr); 2721da177e4SLinus Torvalds } 273b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_to_page); 2741da177e4SLinus Torvalds 275b3bdda02SChristoph Lameter unsigned long vmalloc_to_pfn(const void *addr) 2761da177e4SLinus Torvalds { 2771da177e4SLinus Torvalds return page_to_pfn(virt_to_page(addr)); 2781da177e4SLinus Torvalds } 279b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_to_pfn); 2801da177e4SLinus Torvalds 2811da177e4SLinus Torvalds long vread(char *buf, char *addr, unsigned long count) 2821da177e4SLinus Torvalds { 2839bde916bSChen Gang /* Don't allow overflow */ 2849bde916bSChen Gang if ((unsigned long) buf + count < count) 2859bde916bSChen Gang count = -(unsigned long) buf; 2869bde916bSChen Gang 2871da177e4SLinus Torvalds memcpy(buf, addr, count); 2881da177e4SLinus Torvalds return count; 2891da177e4SLinus Torvalds } 2901da177e4SLinus Torvalds 2911da177e4SLinus Torvalds long vwrite(char *buf, char *addr, unsigned long count) 2921da177e4SLinus Torvalds { 2931da177e4SLinus Torvalds /* Don't allow overflow */ 2941da177e4SLinus Torvalds if ((unsigned long) addr + count < count) 2951da177e4SLinus Torvalds count = -(unsigned long) addr; 2961da177e4SLinus Torvalds 2971da177e4SLinus Torvalds memcpy(addr, buf, count); 2981da177e4SLinus Torvalds return(count); 2991da177e4SLinus Torvalds } 3001da177e4SLinus Torvalds 3011da177e4SLinus Torvalds /* 3021da177e4SLinus Torvalds * vmalloc - allocate virtually continguos memory 3031da177e4SLinus Torvalds * 3041da177e4SLinus Torvalds * @size: allocation size 3051da177e4SLinus Torvalds * 3061da177e4SLinus Torvalds * Allocate enough pages to cover @size from the page level 3071da177e4SLinus Torvalds * allocator and map them into continguos kernel virtual space. 3081da177e4SLinus Torvalds * 309c1c8897fSMichael Opdenacker * For tight control over page level allocator and protection flags 3101da177e4SLinus Torvalds * use __vmalloc() instead. 3111da177e4SLinus Torvalds */ 3121da177e4SLinus Torvalds void *vmalloc(unsigned long size) 3131da177e4SLinus Torvalds { 3141da177e4SLinus Torvalds return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 3151da177e4SLinus Torvalds } 316f6138882SAndrew Morton EXPORT_SYMBOL(vmalloc); 317f6138882SAndrew Morton 318e1ca7788SDave Young /* 319e1ca7788SDave Young * vzalloc - allocate virtually continguos memory with zero fill 320e1ca7788SDave Young * 321e1ca7788SDave Young * @size: allocation size 322e1ca7788SDave Young * 323e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 324e1ca7788SDave Young * allocator and map them into continguos kernel virtual space. 325e1ca7788SDave Young * The memory allocated is set to zero. 326e1ca7788SDave Young * 327e1ca7788SDave Young * For tight control over page level allocator and protection flags 328e1ca7788SDave Young * use __vmalloc() instead. 329e1ca7788SDave Young */ 330e1ca7788SDave Young void *vzalloc(unsigned long size) 331e1ca7788SDave Young { 332e1ca7788SDave Young return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 333e1ca7788SDave Young PAGE_KERNEL); 334e1ca7788SDave Young } 335e1ca7788SDave Young EXPORT_SYMBOL(vzalloc); 336e1ca7788SDave Young 337e1ca7788SDave Young /** 338e1ca7788SDave Young * vmalloc_node - allocate memory on a specific node 339e1ca7788SDave Young * @size: allocation size 340e1ca7788SDave Young * @node: numa node 341e1ca7788SDave Young * 342e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 343e1ca7788SDave Young * allocator and map them into contiguous kernel virtual space. 344e1ca7788SDave Young * 345e1ca7788SDave Young * For tight control over page level allocator and protection flags 346e1ca7788SDave Young * use __vmalloc() instead. 347e1ca7788SDave Young */ 348f6138882SAndrew Morton void *vmalloc_node(unsigned long size, int node) 349f6138882SAndrew Morton { 350f6138882SAndrew Morton return vmalloc(size); 351f6138882SAndrew Morton } 3529a14f653SPaul Mundt EXPORT_SYMBOL(vmalloc_node); 353e1ca7788SDave Young 354e1ca7788SDave Young /** 355e1ca7788SDave Young * vzalloc_node - allocate memory on a specific node with zero fill 356e1ca7788SDave Young * @size: allocation size 357e1ca7788SDave Young * @node: numa node 358e1ca7788SDave Young * 359e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 360e1ca7788SDave Young * allocator and map them into contiguous kernel virtual space. 361e1ca7788SDave Young * The memory allocated is set to zero. 362e1ca7788SDave Young * 363e1ca7788SDave Young * For tight control over page level allocator and protection flags 364e1ca7788SDave Young * use __vmalloc() instead. 365e1ca7788SDave Young */ 366e1ca7788SDave Young void *vzalloc_node(unsigned long size, int node) 367e1ca7788SDave Young { 368e1ca7788SDave Young return vzalloc(size); 369e1ca7788SDave Young } 370e1ca7788SDave Young EXPORT_SYMBOL(vzalloc_node); 3711da177e4SLinus Torvalds 3721af446edSPaul Mundt #ifndef PAGE_KERNEL_EXEC 3731af446edSPaul Mundt # define PAGE_KERNEL_EXEC PAGE_KERNEL 3741af446edSPaul Mundt #endif 3751af446edSPaul Mundt 3761af446edSPaul Mundt /** 3771af446edSPaul Mundt * vmalloc_exec - allocate virtually contiguous, executable memory 3781af446edSPaul Mundt * @size: allocation size 3791af446edSPaul Mundt * 3801af446edSPaul Mundt * Kernel-internal function to allocate enough pages to cover @size 3811af446edSPaul Mundt * the page level allocator and map them into contiguous and 3821af446edSPaul Mundt * executable kernel virtual space. 3831af446edSPaul Mundt * 3841af446edSPaul Mundt * For tight control over page level allocator and protection flags 3851af446edSPaul Mundt * use __vmalloc() instead. 3861af446edSPaul Mundt */ 3871af446edSPaul Mundt 3881af446edSPaul Mundt void *vmalloc_exec(unsigned long size) 3891af446edSPaul Mundt { 3901af446edSPaul Mundt return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); 3911af446edSPaul Mundt } 3921af446edSPaul Mundt 393b5073173SPaul Mundt /** 394b5073173SPaul Mundt * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 3951da177e4SLinus Torvalds * @size: allocation size 3961da177e4SLinus Torvalds * 3971da177e4SLinus Torvalds * Allocate enough 32bit PA addressable pages to cover @size from the 3981da177e4SLinus Torvalds * page level allocator and map them into continguos kernel virtual space. 3991da177e4SLinus Torvalds */ 4001da177e4SLinus Torvalds void *vmalloc_32(unsigned long size) 4011da177e4SLinus Torvalds { 4021da177e4SLinus Torvalds return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 4031da177e4SLinus Torvalds } 404b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_32); 405b5073173SPaul Mundt 406b5073173SPaul Mundt /** 407b5073173SPaul Mundt * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 408b5073173SPaul Mundt * @size: allocation size 409b5073173SPaul Mundt * 410b5073173SPaul Mundt * The resulting memory area is 32bit addressable and zeroed so it can be 411b5073173SPaul Mundt * mapped to userspace without leaking data. 412f905bc44SPaul Mundt * 413f905bc44SPaul Mundt * VM_USERMAP is set on the corresponding VMA so that subsequent calls to 414f905bc44SPaul Mundt * remap_vmalloc_range() are permissible. 415b5073173SPaul Mundt */ 416b5073173SPaul Mundt void *vmalloc_32_user(unsigned long size) 417b5073173SPaul Mundt { 418f905bc44SPaul Mundt /* 419f905bc44SPaul Mundt * We'll have to sort out the ZONE_DMA bits for 64-bit, 420f905bc44SPaul Mundt * but for now this can simply use vmalloc_user() directly. 421f905bc44SPaul Mundt */ 422f905bc44SPaul Mundt return vmalloc_user(size); 423b5073173SPaul Mundt } 424b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_32_user); 4251da177e4SLinus Torvalds 4261da177e4SLinus Torvalds void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 4271da177e4SLinus Torvalds { 4281da177e4SLinus Torvalds BUG(); 4291da177e4SLinus Torvalds return NULL; 4301da177e4SLinus Torvalds } 431b5073173SPaul Mundt EXPORT_SYMBOL(vmap); 4321da177e4SLinus Torvalds 433b3bdda02SChristoph Lameter void vunmap(const void *addr) 4341da177e4SLinus Torvalds { 4351da177e4SLinus Torvalds BUG(); 4361da177e4SLinus Torvalds } 437b5073173SPaul Mundt EXPORT_SYMBOL(vunmap); 4381da177e4SLinus Torvalds 439eb6434d9SPaul Mundt void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) 440eb6434d9SPaul Mundt { 441eb6434d9SPaul Mundt BUG(); 442eb6434d9SPaul Mundt return NULL; 443eb6434d9SPaul Mundt } 444eb6434d9SPaul Mundt EXPORT_SYMBOL(vm_map_ram); 445eb6434d9SPaul Mundt 446eb6434d9SPaul Mundt void vm_unmap_ram(const void *mem, unsigned int count) 447eb6434d9SPaul Mundt { 448eb6434d9SPaul Mundt BUG(); 449eb6434d9SPaul Mundt } 450eb6434d9SPaul Mundt EXPORT_SYMBOL(vm_unmap_ram); 451eb6434d9SPaul Mundt 452eb6434d9SPaul Mundt void vm_unmap_aliases(void) 453eb6434d9SPaul Mundt { 454eb6434d9SPaul Mundt } 455eb6434d9SPaul Mundt EXPORT_SYMBOL_GPL(vm_unmap_aliases); 456eb6434d9SPaul Mundt 4571da177e4SLinus Torvalds /* 4581eeb66a1SChristoph Hellwig * Implement a stub for vmalloc_sync_all() if the architecture chose not to 4591eeb66a1SChristoph Hellwig * have one. 4601eeb66a1SChristoph Hellwig */ 4611eeb66a1SChristoph Hellwig void __attribute__((weak)) vmalloc_sync_all(void) 4621eeb66a1SChristoph Hellwig { 4631eeb66a1SChristoph Hellwig } 4641eeb66a1SChristoph Hellwig 46529c185e5SPaul Mundt /** 46629c185e5SPaul Mundt * alloc_vm_area - allocate a range of kernel address space 46729c185e5SPaul Mundt * @size: size of the area 46829c185e5SPaul Mundt * 46929c185e5SPaul Mundt * Returns: NULL on failure, vm_struct on success 47029c185e5SPaul Mundt * 47129c185e5SPaul Mundt * This function reserves a range of kernel address space, and 47229c185e5SPaul Mundt * allocates pagetables to map that range. No actual mappings 47329c185e5SPaul Mundt * are created. If the kernel address space is not shared 47429c185e5SPaul Mundt * between processes, it syncs the pagetable across all 47529c185e5SPaul Mundt * processes. 47629c185e5SPaul Mundt */ 477cd12909cSDavid Vrabel struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) 47829c185e5SPaul Mundt { 47929c185e5SPaul Mundt BUG(); 48029c185e5SPaul Mundt return NULL; 48129c185e5SPaul Mundt } 48229c185e5SPaul Mundt EXPORT_SYMBOL_GPL(alloc_vm_area); 48329c185e5SPaul Mundt 48429c185e5SPaul Mundt void free_vm_area(struct vm_struct *area) 48529c185e5SPaul Mundt { 48629c185e5SPaul Mundt BUG(); 48729c185e5SPaul Mundt } 48829c185e5SPaul Mundt EXPORT_SYMBOL_GPL(free_vm_area); 48929c185e5SPaul Mundt 490b5073173SPaul Mundt int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 491b5073173SPaul Mundt struct page *page) 492b5073173SPaul Mundt { 493b5073173SPaul Mundt return -EINVAL; 494b5073173SPaul Mundt } 495b5073173SPaul Mundt EXPORT_SYMBOL(vm_insert_page); 496b5073173SPaul Mundt 4971eeb66a1SChristoph Hellwig /* 4981da177e4SLinus Torvalds * sys_brk() for the most part doesn't need the global kernel 4991da177e4SLinus Torvalds * lock, except when an application is doing something nasty 5001da177e4SLinus Torvalds * like trying to un-brk an area that has already been mapped 5011da177e4SLinus Torvalds * to a regular file. in this case, the unmapping will need 5021da177e4SLinus Torvalds * to invoke file system routines that need the global lock. 5031da177e4SLinus Torvalds */ 5046a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 5051da177e4SLinus Torvalds { 5061da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 5071da177e4SLinus Torvalds 5081da177e4SLinus Torvalds if (brk < mm->start_brk || brk > mm->context.end_brk) 5091da177e4SLinus Torvalds return mm->brk; 5101da177e4SLinus Torvalds 5111da177e4SLinus Torvalds if (mm->brk == brk) 5121da177e4SLinus Torvalds return mm->brk; 5131da177e4SLinus Torvalds 5141da177e4SLinus Torvalds /* 5151da177e4SLinus Torvalds * Always allow shrinking brk 5161da177e4SLinus Torvalds */ 5171da177e4SLinus Torvalds if (brk <= mm->brk) { 5181da177e4SLinus Torvalds mm->brk = brk; 5191da177e4SLinus Torvalds return brk; 5201da177e4SLinus Torvalds } 5211da177e4SLinus Torvalds 5221da177e4SLinus Torvalds /* 5231da177e4SLinus Torvalds * Ok, looks good - let it rip. 5241da177e4SLinus Torvalds */ 525cfe79c00SMike Frysinger flush_icache_range(mm->brk, brk); 5261da177e4SLinus Torvalds return mm->brk = brk; 5271da177e4SLinus Torvalds } 5281da177e4SLinus Torvalds 5298feae131SDavid Howells /* 5308feae131SDavid Howells * initialise the VMA and region record slabs 5318feae131SDavid Howells */ 5328feae131SDavid Howells void __init mmap_init(void) 5331da177e4SLinus Torvalds { 53400a62ce9SKOSAKI Motohiro int ret; 53500a62ce9SKOSAKI Motohiro 53600a62ce9SKOSAKI Motohiro ret = percpu_counter_init(&vm_committed_as, 0); 53700a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 53833e5d769SDavid Howells vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); 5398feae131SDavid Howells } 5401da177e4SLinus Torvalds 5418feae131SDavid Howells /* 5428feae131SDavid Howells * validate the region tree 5438feae131SDavid Howells * - the caller must hold the region lock 5448feae131SDavid Howells */ 5458feae131SDavid Howells #ifdef CONFIG_DEBUG_NOMMU_REGIONS 5468feae131SDavid Howells static noinline void validate_nommu_regions(void) 5478feae131SDavid Howells { 5488feae131SDavid Howells struct vm_region *region, *last; 5498feae131SDavid Howells struct rb_node *p, *lastp; 5501da177e4SLinus Torvalds 5518feae131SDavid Howells lastp = rb_first(&nommu_region_tree); 5528feae131SDavid Howells if (!lastp) 5538feae131SDavid Howells return; 5548feae131SDavid Howells 5558feae131SDavid Howells last = rb_entry(lastp, struct vm_region, vm_rb); 55633e5d769SDavid Howells BUG_ON(unlikely(last->vm_end <= last->vm_start)); 55733e5d769SDavid Howells BUG_ON(unlikely(last->vm_top < last->vm_end)); 5588feae131SDavid Howells 5598feae131SDavid Howells while ((p = rb_next(lastp))) { 5608feae131SDavid Howells region = rb_entry(p, struct vm_region, vm_rb); 5618feae131SDavid Howells last = rb_entry(lastp, struct vm_region, vm_rb); 5628feae131SDavid Howells 56333e5d769SDavid Howells BUG_ON(unlikely(region->vm_end <= region->vm_start)); 56433e5d769SDavid Howells BUG_ON(unlikely(region->vm_top < region->vm_end)); 56533e5d769SDavid Howells BUG_ON(unlikely(region->vm_start < last->vm_top)); 5668feae131SDavid Howells 5678feae131SDavid Howells lastp = p; 5681da177e4SLinus Torvalds } 5691da177e4SLinus Torvalds } 5708feae131SDavid Howells #else 57133e5d769SDavid Howells static void validate_nommu_regions(void) 57233e5d769SDavid Howells { 57333e5d769SDavid Howells } 5748feae131SDavid Howells #endif 5758feae131SDavid Howells 5768feae131SDavid Howells /* 5778feae131SDavid Howells * add a region into the global tree 5788feae131SDavid Howells */ 5798feae131SDavid Howells static void add_nommu_region(struct vm_region *region) 5808feae131SDavid Howells { 5818feae131SDavid Howells struct vm_region *pregion; 5828feae131SDavid Howells struct rb_node **p, *parent; 5838feae131SDavid Howells 5848feae131SDavid Howells validate_nommu_regions(); 5858feae131SDavid Howells 5868feae131SDavid Howells parent = NULL; 5878feae131SDavid Howells p = &nommu_region_tree.rb_node; 5888feae131SDavid Howells while (*p) { 5898feae131SDavid Howells parent = *p; 5908feae131SDavid Howells pregion = rb_entry(parent, struct vm_region, vm_rb); 5918feae131SDavid Howells if (region->vm_start < pregion->vm_start) 5928feae131SDavid Howells p = &(*p)->rb_left; 5938feae131SDavid Howells else if (region->vm_start > pregion->vm_start) 5948feae131SDavid Howells p = &(*p)->rb_right; 5958feae131SDavid Howells else if (pregion == region) 5968feae131SDavid Howells return; 5978feae131SDavid Howells else 5988feae131SDavid Howells BUG(); 5998feae131SDavid Howells } 6008feae131SDavid Howells 6018feae131SDavid Howells rb_link_node(®ion->vm_rb, parent, p); 6028feae131SDavid Howells rb_insert_color(®ion->vm_rb, &nommu_region_tree); 6038feae131SDavid Howells 6048feae131SDavid Howells validate_nommu_regions(); 6058feae131SDavid Howells } 6068feae131SDavid Howells 6078feae131SDavid Howells /* 6088feae131SDavid Howells * delete a region from the global tree 6098feae131SDavid Howells */ 6108feae131SDavid Howells static void delete_nommu_region(struct vm_region *region) 6118feae131SDavid Howells { 6128feae131SDavid Howells BUG_ON(!nommu_region_tree.rb_node); 6138feae131SDavid Howells 6148feae131SDavid Howells validate_nommu_regions(); 6158feae131SDavid Howells rb_erase(®ion->vm_rb, &nommu_region_tree); 6168feae131SDavid Howells validate_nommu_regions(); 6178feae131SDavid Howells } 6188feae131SDavid Howells 6198feae131SDavid Howells /* 6208feae131SDavid Howells * free a contiguous series of pages 6218feae131SDavid Howells */ 6228feae131SDavid Howells static void free_page_series(unsigned long from, unsigned long to) 6238feae131SDavid Howells { 6248feae131SDavid Howells for (; from < to; from += PAGE_SIZE) { 6258feae131SDavid Howells struct page *page = virt_to_page(from); 6268feae131SDavid Howells 6278feae131SDavid Howells kdebug("- free %lx", from); 62833e5d769SDavid Howells atomic_long_dec(&mmap_pages_allocated); 6298feae131SDavid Howells if (page_count(page) != 1) 63033e5d769SDavid Howells kdebug("free page %p: refcount not one: %d", 63133e5d769SDavid Howells page, page_count(page)); 6328feae131SDavid Howells put_page(page); 6338feae131SDavid Howells } 6348feae131SDavid Howells } 6358feae131SDavid Howells 6368feae131SDavid Howells /* 6378feae131SDavid Howells * release a reference to a region 63833e5d769SDavid Howells * - the caller must hold the region semaphore for writing, which this releases 639dd8632a1SPaul Mundt * - the region may not have been added to the tree yet, in which case vm_top 6408feae131SDavid Howells * will equal vm_start 6418feae131SDavid Howells */ 6428feae131SDavid Howells static void __put_nommu_region(struct vm_region *region) 6438feae131SDavid Howells __releases(nommu_region_sem) 6448feae131SDavid Howells { 6451e2ae599SDavid Howells kenter("%p{%d}", region, region->vm_usage); 6468feae131SDavid Howells 6478feae131SDavid Howells BUG_ON(!nommu_region_tree.rb_node); 6488feae131SDavid Howells 6491e2ae599SDavid Howells if (--region->vm_usage == 0) { 650dd8632a1SPaul Mundt if (region->vm_top > region->vm_start) 6518feae131SDavid Howells delete_nommu_region(region); 6528feae131SDavid Howells up_write(&nommu_region_sem); 6538feae131SDavid Howells 6548feae131SDavid Howells if (region->vm_file) 6558feae131SDavid Howells fput(region->vm_file); 6568feae131SDavid Howells 6578feae131SDavid Howells /* IO memory and memory shared directly out of the pagecache 6588feae131SDavid Howells * from ramfs/tmpfs mustn't be released here */ 6598feae131SDavid Howells if (region->vm_flags & VM_MAPPED_COPY) { 6608feae131SDavid Howells kdebug("free series"); 661dd8632a1SPaul Mundt free_page_series(region->vm_start, region->vm_top); 6628feae131SDavid Howells } 6638feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 6648feae131SDavid Howells } else { 6658feae131SDavid Howells up_write(&nommu_region_sem); 6668feae131SDavid Howells } 6678feae131SDavid Howells } 6688feae131SDavid Howells 6698feae131SDavid Howells /* 6708feae131SDavid Howells * release a reference to a region 6718feae131SDavid Howells */ 6728feae131SDavid Howells static void put_nommu_region(struct vm_region *region) 6738feae131SDavid Howells { 6748feae131SDavid Howells down_write(&nommu_region_sem); 6758feae131SDavid Howells __put_nommu_region(region); 6768feae131SDavid Howells } 6771da177e4SLinus Torvalds 6783034097aSDavid Howells /* 679eb8cdec4SBernd Schmidt * update protection on a vma 680eb8cdec4SBernd Schmidt */ 681eb8cdec4SBernd Schmidt static void protect_vma(struct vm_area_struct *vma, unsigned long flags) 682eb8cdec4SBernd Schmidt { 683eb8cdec4SBernd Schmidt #ifdef CONFIG_MPU 684eb8cdec4SBernd Schmidt struct mm_struct *mm = vma->vm_mm; 685eb8cdec4SBernd Schmidt long start = vma->vm_start & PAGE_MASK; 686eb8cdec4SBernd Schmidt while (start < vma->vm_end) { 687eb8cdec4SBernd Schmidt protect_page(mm, start, flags); 688eb8cdec4SBernd Schmidt start += PAGE_SIZE; 689eb8cdec4SBernd Schmidt } 690eb8cdec4SBernd Schmidt update_protections(mm); 691eb8cdec4SBernd Schmidt #endif 692eb8cdec4SBernd Schmidt } 693eb8cdec4SBernd Schmidt 694eb8cdec4SBernd Schmidt /* 6953034097aSDavid Howells * add a VMA into a process's mm_struct in the appropriate place in the list 6968feae131SDavid Howells * and tree and add to the address space's page tree also if not an anonymous 6978feae131SDavid Howells * page 6983034097aSDavid Howells * - should be called with mm->mmap_sem held writelocked 6993034097aSDavid Howells */ 7008feae131SDavid Howells static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) 7013034097aSDavid Howells { 7026038def0SNamhyung Kim struct vm_area_struct *pvma, *prev; 7038feae131SDavid Howells struct address_space *mapping; 7046038def0SNamhyung Kim struct rb_node **p, *parent, *rb_prev; 7053034097aSDavid Howells 7068feae131SDavid Howells kenter(",%p", vma); 7078feae131SDavid Howells 7088feae131SDavid Howells BUG_ON(!vma->vm_region); 7098feae131SDavid Howells 7108feae131SDavid Howells mm->map_count++; 7118feae131SDavid Howells vma->vm_mm = mm; 7128feae131SDavid Howells 713eb8cdec4SBernd Schmidt protect_vma(vma, vma->vm_flags); 714eb8cdec4SBernd Schmidt 7158feae131SDavid Howells /* add the VMA to the mapping */ 7168feae131SDavid Howells if (vma->vm_file) { 7178feae131SDavid Howells mapping = vma->vm_file->f_mapping; 7188feae131SDavid Howells 719918e556eSDavid Howells mutex_lock(&mapping->i_mmap_mutex); 7208feae131SDavid Howells flush_dcache_mmap_lock(mapping); 7216b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 7228feae131SDavid Howells flush_dcache_mmap_unlock(mapping); 723918e556eSDavid Howells mutex_unlock(&mapping->i_mmap_mutex); 7248feae131SDavid Howells } 7258feae131SDavid Howells 7268feae131SDavid Howells /* add the VMA to the tree */ 7276038def0SNamhyung Kim parent = rb_prev = NULL; 7288feae131SDavid Howells p = &mm->mm_rb.rb_node; 7298feae131SDavid Howells while (*p) { 7308feae131SDavid Howells parent = *p; 7318feae131SDavid Howells pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 7328feae131SDavid Howells 7338feae131SDavid Howells /* sort by: start addr, end addr, VMA struct addr in that order 7348feae131SDavid Howells * (the latter is necessary as we may get identical VMAs) */ 7358feae131SDavid Howells if (vma->vm_start < pvma->vm_start) 7368feae131SDavid Howells p = &(*p)->rb_left; 7376038def0SNamhyung Kim else if (vma->vm_start > pvma->vm_start) { 7386038def0SNamhyung Kim rb_prev = parent; 7398feae131SDavid Howells p = &(*p)->rb_right; 7406038def0SNamhyung Kim } else if (vma->vm_end < pvma->vm_end) 7418feae131SDavid Howells p = &(*p)->rb_left; 7426038def0SNamhyung Kim else if (vma->vm_end > pvma->vm_end) { 7436038def0SNamhyung Kim rb_prev = parent; 7448feae131SDavid Howells p = &(*p)->rb_right; 7456038def0SNamhyung Kim } else if (vma < pvma) 7468feae131SDavid Howells p = &(*p)->rb_left; 7476038def0SNamhyung Kim else if (vma > pvma) { 7486038def0SNamhyung Kim rb_prev = parent; 7498feae131SDavid Howells p = &(*p)->rb_right; 7506038def0SNamhyung Kim } else 7518feae131SDavid Howells BUG(); 7528feae131SDavid Howells } 7538feae131SDavid Howells 7548feae131SDavid Howells rb_link_node(&vma->vm_rb, parent, p); 7558feae131SDavid Howells rb_insert_color(&vma->vm_rb, &mm->mm_rb); 7568feae131SDavid Howells 7578feae131SDavid Howells /* add VMA to the VMA list also */ 7586038def0SNamhyung Kim prev = NULL; 7596038def0SNamhyung Kim if (rb_prev) 7606038def0SNamhyung Kim prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 7613034097aSDavid Howells 7626038def0SNamhyung Kim __vma_link_list(mm, vma, prev, parent); 7638feae131SDavid Howells } 7648feae131SDavid Howells 7658feae131SDavid Howells /* 7668feae131SDavid Howells * delete a VMA from its owning mm_struct and address space 7678feae131SDavid Howells */ 7688feae131SDavid Howells static void delete_vma_from_mm(struct vm_area_struct *vma) 7698feae131SDavid Howells { 7708feae131SDavid Howells struct address_space *mapping; 7718feae131SDavid Howells struct mm_struct *mm = vma->vm_mm; 7728feae131SDavid Howells 7738feae131SDavid Howells kenter("%p", vma); 7748feae131SDavid Howells 775eb8cdec4SBernd Schmidt protect_vma(vma, 0); 776eb8cdec4SBernd Schmidt 7778feae131SDavid Howells mm->map_count--; 7788feae131SDavid Howells if (mm->mmap_cache == vma) 7798feae131SDavid Howells mm->mmap_cache = NULL; 7808feae131SDavid Howells 7818feae131SDavid Howells /* remove the VMA from the mapping */ 7828feae131SDavid Howells if (vma->vm_file) { 7838feae131SDavid Howells mapping = vma->vm_file->f_mapping; 7848feae131SDavid Howells 785918e556eSDavid Howells mutex_lock(&mapping->i_mmap_mutex); 7868feae131SDavid Howells flush_dcache_mmap_lock(mapping); 7876b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 7888feae131SDavid Howells flush_dcache_mmap_unlock(mapping); 789918e556eSDavid Howells mutex_unlock(&mapping->i_mmap_mutex); 7908feae131SDavid Howells } 7918feae131SDavid Howells 7928feae131SDavid Howells /* remove from the MM's tree and list */ 7938feae131SDavid Howells rb_erase(&vma->vm_rb, &mm->mm_rb); 794b951bf2cSNamhyung Kim 795b951bf2cSNamhyung Kim if (vma->vm_prev) 796b951bf2cSNamhyung Kim vma->vm_prev->vm_next = vma->vm_next; 797b951bf2cSNamhyung Kim else 798b951bf2cSNamhyung Kim mm->mmap = vma->vm_next; 799b951bf2cSNamhyung Kim 800b951bf2cSNamhyung Kim if (vma->vm_next) 801b951bf2cSNamhyung Kim vma->vm_next->vm_prev = vma->vm_prev; 8028feae131SDavid Howells } 8038feae131SDavid Howells 8048feae131SDavid Howells /* 8058feae131SDavid Howells * destroy a VMA record 8068feae131SDavid Howells */ 8078feae131SDavid Howells static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) 8088feae131SDavid Howells { 8098feae131SDavid Howells kenter("%p", vma); 8108feae131SDavid Howells if (vma->vm_ops && vma->vm_ops->close) 8118feae131SDavid Howells vma->vm_ops->close(vma); 812e9714acfSKonstantin Khlebnikov if (vma->vm_file) 8138feae131SDavid Howells fput(vma->vm_file); 8148feae131SDavid Howells put_nommu_region(vma->vm_region); 8158feae131SDavid Howells kmem_cache_free(vm_area_cachep, vma); 8163034097aSDavid Howells } 8173034097aSDavid Howells 8183034097aSDavid Howells /* 8193034097aSDavid Howells * look up the first VMA in which addr resides, NULL if none 8203034097aSDavid Howells * - should be called with mm->mmap_sem at least held readlocked 8213034097aSDavid Howells */ 8223034097aSDavid Howells struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 8233034097aSDavid Howells { 8248feae131SDavid Howells struct vm_area_struct *vma; 8253034097aSDavid Howells 8268feae131SDavid Howells /* check the cache first */ 827b6a9b7f6SJan Stancek vma = ACCESS_ONCE(mm->mmap_cache); 8288feae131SDavid Howells if (vma && vma->vm_start <= addr && vma->vm_end > addr) 8298feae131SDavid Howells return vma; 8308feae131SDavid Howells 831e922c4c5SNamhyung Kim /* trawl the list (there may be multiple mappings in which addr 8328feae131SDavid Howells * resides) */ 833e922c4c5SNamhyung Kim for (vma = mm->mmap; vma; vma = vma->vm_next) { 8348feae131SDavid Howells if (vma->vm_start > addr) 8358feae131SDavid Howells return NULL; 8368feae131SDavid Howells if (vma->vm_end > addr) { 8378feae131SDavid Howells mm->mmap_cache = vma; 8388feae131SDavid Howells return vma; 8393034097aSDavid Howells } 8408feae131SDavid Howells } 8413034097aSDavid Howells 8423034097aSDavid Howells return NULL; 8433034097aSDavid Howells } 8443034097aSDavid Howells EXPORT_SYMBOL(find_vma); 8453034097aSDavid Howells 8463034097aSDavid Howells /* 847930e652aSDavid Howells * find a VMA 848930e652aSDavid Howells * - we don't extend stack VMAs under NOMMU conditions 849930e652aSDavid Howells */ 850930e652aSDavid Howells struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) 851930e652aSDavid Howells { 8527561e8caSDavid Howells return find_vma(mm, addr); 853930e652aSDavid Howells } 854930e652aSDavid Howells 8558feae131SDavid Howells /* 8568feae131SDavid Howells * expand a stack to a given address 8578feae131SDavid Howells * - not supported under NOMMU conditions 8588feae131SDavid Howells */ 85957c8f63eSGreg Ungerer int expand_stack(struct vm_area_struct *vma, unsigned long address) 86057c8f63eSGreg Ungerer { 86157c8f63eSGreg Ungerer return -ENOMEM; 86257c8f63eSGreg Ungerer } 86357c8f63eSGreg Ungerer 864930e652aSDavid Howells /* 8656fa5f80bSDavid Howells * look up the first VMA exactly that exactly matches addr 8666fa5f80bSDavid Howells * - should be called with mm->mmap_sem at least held readlocked 8676fa5f80bSDavid Howells */ 8688feae131SDavid Howells static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, 8698feae131SDavid Howells unsigned long addr, 8708feae131SDavid Howells unsigned long len) 8711da177e4SLinus Torvalds { 8721da177e4SLinus Torvalds struct vm_area_struct *vma; 8738feae131SDavid Howells unsigned long end = addr + len; 8741da177e4SLinus Torvalds 8758feae131SDavid Howells /* check the cache first */ 8768feae131SDavid Howells vma = mm->mmap_cache; 8778feae131SDavid Howells if (vma && vma->vm_start == addr && vma->vm_end == end) 8781da177e4SLinus Torvalds return vma; 8798feae131SDavid Howells 880e922c4c5SNamhyung Kim /* trawl the list (there may be multiple mappings in which addr 8818feae131SDavid Howells * resides) */ 882e922c4c5SNamhyung Kim for (vma = mm->mmap; vma; vma = vma->vm_next) { 8838feae131SDavid Howells if (vma->vm_start < addr) 8848feae131SDavid Howells continue; 8858feae131SDavid Howells if (vma->vm_start > addr) 8868feae131SDavid Howells return NULL; 8878feae131SDavid Howells if (vma->vm_end == end) { 8888feae131SDavid Howells mm->mmap_cache = vma; 8898feae131SDavid Howells return vma; 8908feae131SDavid Howells } 8911da177e4SLinus Torvalds } 8921da177e4SLinus Torvalds 8931da177e4SLinus Torvalds return NULL; 8941da177e4SLinus Torvalds } 8951da177e4SLinus Torvalds 8963034097aSDavid Howells /* 8971da177e4SLinus Torvalds * determine whether a mapping should be permitted and, if so, what sort of 8981da177e4SLinus Torvalds * mapping we're capable of supporting 8991da177e4SLinus Torvalds */ 9001da177e4SLinus Torvalds static int validate_mmap_request(struct file *file, 9011da177e4SLinus Torvalds unsigned long addr, 9021da177e4SLinus Torvalds unsigned long len, 9031da177e4SLinus Torvalds unsigned long prot, 9041da177e4SLinus Torvalds unsigned long flags, 9051da177e4SLinus Torvalds unsigned long pgoff, 9061da177e4SLinus Torvalds unsigned long *_capabilities) 9071da177e4SLinus Torvalds { 9088feae131SDavid Howells unsigned long capabilities, rlen; 9091da177e4SLinus Torvalds int ret; 9101da177e4SLinus Torvalds 9111da177e4SLinus Torvalds /* do the simple checks first */ 91206aab5a3SDavid Howells if (flags & MAP_FIXED) { 9131da177e4SLinus Torvalds printk(KERN_DEBUG 9141da177e4SLinus Torvalds "%d: Can't do fixed-address/overlay mmap of RAM\n", 9151da177e4SLinus Torvalds current->pid); 9161da177e4SLinus Torvalds return -EINVAL; 9171da177e4SLinus Torvalds } 9181da177e4SLinus Torvalds 9191da177e4SLinus Torvalds if ((flags & MAP_TYPE) != MAP_PRIVATE && 9201da177e4SLinus Torvalds (flags & MAP_TYPE) != MAP_SHARED) 9211da177e4SLinus Torvalds return -EINVAL; 9221da177e4SLinus Torvalds 923f81cff0dSMike Frysinger if (!len) 9241da177e4SLinus Torvalds return -EINVAL; 9251da177e4SLinus Torvalds 926f81cff0dSMike Frysinger /* Careful about overflows.. */ 9278feae131SDavid Howells rlen = PAGE_ALIGN(len); 9288feae131SDavid Howells if (!rlen || rlen > TASK_SIZE) 929f81cff0dSMike Frysinger return -ENOMEM; 930f81cff0dSMike Frysinger 9311da177e4SLinus Torvalds /* offset overflow? */ 9328feae131SDavid Howells if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) 933f81cff0dSMike Frysinger return -EOVERFLOW; 9341da177e4SLinus Torvalds 9351da177e4SLinus Torvalds if (file) { 9361da177e4SLinus Torvalds /* validate file mapping requests */ 9371da177e4SLinus Torvalds struct address_space *mapping; 9381da177e4SLinus Torvalds 9391da177e4SLinus Torvalds /* files must support mmap */ 940*72c2d531SAl Viro if (!file->f_op->mmap) 9411da177e4SLinus Torvalds return -ENODEV; 9421da177e4SLinus Torvalds 9431da177e4SLinus Torvalds /* work out if what we've got could possibly be shared 9441da177e4SLinus Torvalds * - we support chardevs that provide their own "memory" 9451da177e4SLinus Torvalds * - we support files/blockdevs that are memory backed 9461da177e4SLinus Torvalds */ 9471da177e4SLinus Torvalds mapping = file->f_mapping; 9481da177e4SLinus Torvalds if (!mapping) 949496ad9aaSAl Viro mapping = file_inode(file)->i_mapping; 9501da177e4SLinus Torvalds 9511da177e4SLinus Torvalds capabilities = 0; 9521da177e4SLinus Torvalds if (mapping && mapping->backing_dev_info) 9531da177e4SLinus Torvalds capabilities = mapping->backing_dev_info->capabilities; 9541da177e4SLinus Torvalds 9551da177e4SLinus Torvalds if (!capabilities) { 9561da177e4SLinus Torvalds /* no explicit capabilities set, so assume some 9571da177e4SLinus Torvalds * defaults */ 958496ad9aaSAl Viro switch (file_inode(file)->i_mode & S_IFMT) { 9591da177e4SLinus Torvalds case S_IFREG: 9601da177e4SLinus Torvalds case S_IFBLK: 9611da177e4SLinus Torvalds capabilities = BDI_CAP_MAP_COPY; 9621da177e4SLinus Torvalds break; 9631da177e4SLinus Torvalds 9641da177e4SLinus Torvalds case S_IFCHR: 9651da177e4SLinus Torvalds capabilities = 9661da177e4SLinus Torvalds BDI_CAP_MAP_DIRECT | 9671da177e4SLinus Torvalds BDI_CAP_READ_MAP | 9681da177e4SLinus Torvalds BDI_CAP_WRITE_MAP; 9691da177e4SLinus Torvalds break; 9701da177e4SLinus Torvalds 9711da177e4SLinus Torvalds default: 9721da177e4SLinus Torvalds return -EINVAL; 9731da177e4SLinus Torvalds } 9741da177e4SLinus Torvalds } 9751da177e4SLinus Torvalds 9761da177e4SLinus Torvalds /* eliminate any capabilities that we can't support on this 9771da177e4SLinus Torvalds * device */ 9781da177e4SLinus Torvalds if (!file->f_op->get_unmapped_area) 9791da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 9801da177e4SLinus Torvalds if (!file->f_op->read) 9811da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_COPY; 9821da177e4SLinus Torvalds 98328d7a6aeSGraff Yang /* The file shall have been opened with read permission. */ 98428d7a6aeSGraff Yang if (!(file->f_mode & FMODE_READ)) 98528d7a6aeSGraff Yang return -EACCES; 98628d7a6aeSGraff Yang 9871da177e4SLinus Torvalds if (flags & MAP_SHARED) { 9881da177e4SLinus Torvalds /* do checks for writing, appending and locking */ 9891da177e4SLinus Torvalds if ((prot & PROT_WRITE) && 9901da177e4SLinus Torvalds !(file->f_mode & FMODE_WRITE)) 9911da177e4SLinus Torvalds return -EACCES; 9921da177e4SLinus Torvalds 993496ad9aaSAl Viro if (IS_APPEND(file_inode(file)) && 9941da177e4SLinus Torvalds (file->f_mode & FMODE_WRITE)) 9951da177e4SLinus Torvalds return -EACCES; 9961da177e4SLinus Torvalds 997496ad9aaSAl Viro if (locks_verify_locked(file_inode(file))) 9981da177e4SLinus Torvalds return -EAGAIN; 9991da177e4SLinus Torvalds 10001da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_DIRECT)) 10011da177e4SLinus Torvalds return -ENODEV; 10021da177e4SLinus Torvalds 10031da177e4SLinus Torvalds /* we mustn't privatise shared mappings */ 10041da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_COPY; 10051da177e4SLinus Torvalds } 10061da177e4SLinus Torvalds else { 10071da177e4SLinus Torvalds /* we're going to read the file into private memory we 10081da177e4SLinus Torvalds * allocate */ 10091da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_COPY)) 10101da177e4SLinus Torvalds return -ENODEV; 10111da177e4SLinus Torvalds 10121da177e4SLinus Torvalds /* we don't permit a private writable mapping to be 10131da177e4SLinus Torvalds * shared with the backing device */ 10141da177e4SLinus Torvalds if (prot & PROT_WRITE) 10151da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 10161da177e4SLinus Torvalds } 10171da177e4SLinus Torvalds 10183c7b2045SBernd Schmidt if (capabilities & BDI_CAP_MAP_DIRECT) { 10193c7b2045SBernd Schmidt if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || 10203c7b2045SBernd Schmidt ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || 10213c7b2045SBernd Schmidt ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) 10223c7b2045SBernd Schmidt ) { 10233c7b2045SBernd Schmidt capabilities &= ~BDI_CAP_MAP_DIRECT; 10243c7b2045SBernd Schmidt if (flags & MAP_SHARED) { 10253c7b2045SBernd Schmidt printk(KERN_WARNING 10263c7b2045SBernd Schmidt "MAP_SHARED not completely supported on !MMU\n"); 10273c7b2045SBernd Schmidt return -EINVAL; 10283c7b2045SBernd Schmidt } 10293c7b2045SBernd Schmidt } 10303c7b2045SBernd Schmidt } 10313c7b2045SBernd Schmidt 10321da177e4SLinus Torvalds /* handle executable mappings and implied executable 10331da177e4SLinus Torvalds * mappings */ 1034e9536ae7SJosef Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 10351da177e4SLinus Torvalds if (prot & PROT_EXEC) 10361da177e4SLinus Torvalds return -EPERM; 10371da177e4SLinus Torvalds } 10381da177e4SLinus Torvalds else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { 10391da177e4SLinus Torvalds /* handle implication of PROT_EXEC by PROT_READ */ 10401da177e4SLinus Torvalds if (current->personality & READ_IMPLIES_EXEC) { 10411da177e4SLinus Torvalds if (capabilities & BDI_CAP_EXEC_MAP) 10421da177e4SLinus Torvalds prot |= PROT_EXEC; 10431da177e4SLinus Torvalds } 10441da177e4SLinus Torvalds } 10451da177e4SLinus Torvalds else if ((prot & PROT_READ) && 10461da177e4SLinus Torvalds (prot & PROT_EXEC) && 10471da177e4SLinus Torvalds !(capabilities & BDI_CAP_EXEC_MAP) 10481da177e4SLinus Torvalds ) { 10491da177e4SLinus Torvalds /* backing file is not executable, try to copy */ 10501da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 10511da177e4SLinus Torvalds } 10521da177e4SLinus Torvalds } 10531da177e4SLinus Torvalds else { 10541da177e4SLinus Torvalds /* anonymous mappings are always memory backed and can be 10551da177e4SLinus Torvalds * privately mapped 10561da177e4SLinus Torvalds */ 10571da177e4SLinus Torvalds capabilities = BDI_CAP_MAP_COPY; 10581da177e4SLinus Torvalds 10591da177e4SLinus Torvalds /* handle PROT_EXEC implication by PROT_READ */ 10601da177e4SLinus Torvalds if ((prot & PROT_READ) && 10611da177e4SLinus Torvalds (current->personality & READ_IMPLIES_EXEC)) 10621da177e4SLinus Torvalds prot |= PROT_EXEC; 10631da177e4SLinus Torvalds } 10641da177e4SLinus Torvalds 10651da177e4SLinus Torvalds /* allow the security API to have its say */ 1066e5467859SAl Viro ret = security_mmap_addr(addr); 1067e5467859SAl Viro if (ret < 0) 1068e5467859SAl Viro return ret; 10691da177e4SLinus Torvalds 10701da177e4SLinus Torvalds /* looks okay */ 10711da177e4SLinus Torvalds *_capabilities = capabilities; 10721da177e4SLinus Torvalds return 0; 10731da177e4SLinus Torvalds } 10741da177e4SLinus Torvalds 10751da177e4SLinus Torvalds /* 10761da177e4SLinus Torvalds * we've determined that we can make the mapping, now translate what we 10771da177e4SLinus Torvalds * now know into VMA flags 10781da177e4SLinus Torvalds */ 10791da177e4SLinus Torvalds static unsigned long determine_vm_flags(struct file *file, 10801da177e4SLinus Torvalds unsigned long prot, 10811da177e4SLinus Torvalds unsigned long flags, 10821da177e4SLinus Torvalds unsigned long capabilities) 10831da177e4SLinus Torvalds { 10841da177e4SLinus Torvalds unsigned long vm_flags; 10851da177e4SLinus Torvalds 10861da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); 10871da177e4SLinus Torvalds /* vm_flags |= mm->def_flags; */ 10881da177e4SLinus Torvalds 10891da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_DIRECT)) { 10901da177e4SLinus Torvalds /* attempt to share read-only copies of mapped file chunks */ 10913c7b2045SBernd Schmidt vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 10921da177e4SLinus Torvalds if (file && !(prot & PROT_WRITE)) 10931da177e4SLinus Torvalds vm_flags |= VM_MAYSHARE; 10943c7b2045SBernd Schmidt } else { 10951da177e4SLinus Torvalds /* overlay a shareable mapping on the backing device or inode 10961da177e4SLinus Torvalds * if possible - used for chardevs, ramfs/tmpfs/shmfs and 10971da177e4SLinus Torvalds * romfs/cramfs */ 10983c7b2045SBernd Schmidt vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS); 10991da177e4SLinus Torvalds if (flags & MAP_SHARED) 11003c7b2045SBernd Schmidt vm_flags |= VM_SHARED; 11011da177e4SLinus Torvalds } 11021da177e4SLinus Torvalds 11031da177e4SLinus Torvalds /* refuse to let anyone share private mappings with this process if 11041da177e4SLinus Torvalds * it's being traced - otherwise breakpoints set in it may interfere 11051da177e4SLinus Torvalds * with another untraced process 11061da177e4SLinus Torvalds */ 1107a288eeccSTejun Heo if ((flags & MAP_PRIVATE) && current->ptrace) 11081da177e4SLinus Torvalds vm_flags &= ~VM_MAYSHARE; 11091da177e4SLinus Torvalds 11101da177e4SLinus Torvalds return vm_flags; 11111da177e4SLinus Torvalds } 11121da177e4SLinus Torvalds 11131da177e4SLinus Torvalds /* 11148feae131SDavid Howells * set up a shared mapping on a file (the driver or filesystem provides and 11158feae131SDavid Howells * pins the storage) 11161da177e4SLinus Torvalds */ 11178feae131SDavid Howells static int do_mmap_shared_file(struct vm_area_struct *vma) 11181da177e4SLinus Torvalds { 11191da177e4SLinus Torvalds int ret; 11201da177e4SLinus Torvalds 11211da177e4SLinus Torvalds ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1122dd8632a1SPaul Mundt if (ret == 0) { 1123dd8632a1SPaul Mundt vma->vm_region->vm_top = vma->vm_region->vm_end; 1124645d83c5SDavid Howells return 0; 1125dd8632a1SPaul Mundt } 11261da177e4SLinus Torvalds if (ret != -ENOSYS) 11271da177e4SLinus Torvalds return ret; 11281da177e4SLinus Torvalds 11293fa30460SDavid Howells /* getting -ENOSYS indicates that direct mmap isn't possible (as 11303fa30460SDavid Howells * opposed to tried but failed) so we can only give a suitable error as 11313fa30460SDavid Howells * it's not possible to make a private copy if MAP_SHARED was given */ 11321da177e4SLinus Torvalds return -ENODEV; 11331da177e4SLinus Torvalds } 11341da177e4SLinus Torvalds 11351da177e4SLinus Torvalds /* 11361da177e4SLinus Torvalds * set up a private mapping or an anonymous shared mapping 11371da177e4SLinus Torvalds */ 11388feae131SDavid Howells static int do_mmap_private(struct vm_area_struct *vma, 11398feae131SDavid Howells struct vm_region *region, 1140645d83c5SDavid Howells unsigned long len, 1141645d83c5SDavid Howells unsigned long capabilities) 11421da177e4SLinus Torvalds { 11438feae131SDavid Howells struct page *pages; 1144f67d9b15SBob Liu unsigned long total, point, n; 11451da177e4SLinus Torvalds void *base; 11468feae131SDavid Howells int ret, order; 11471da177e4SLinus Torvalds 11481da177e4SLinus Torvalds /* invoke the file's mapping function so that it can keep track of 11491da177e4SLinus Torvalds * shared mappings on devices or memory 11501da177e4SLinus Torvalds * - VM_MAYSHARE will be set if it may attempt to share 11511da177e4SLinus Torvalds */ 1152645d83c5SDavid Howells if (capabilities & BDI_CAP_MAP_DIRECT) { 11531da177e4SLinus Torvalds ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1154dd8632a1SPaul Mundt if (ret == 0) { 11551da177e4SLinus Torvalds /* shouldn't return success if we're not sharing */ 1156dd8632a1SPaul Mundt BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); 1157dd8632a1SPaul Mundt vma->vm_region->vm_top = vma->vm_region->vm_end; 1158645d83c5SDavid Howells return 0; 11591da177e4SLinus Torvalds } 1160dd8632a1SPaul Mundt if (ret != -ENOSYS) 1161dd8632a1SPaul Mundt return ret; 11621da177e4SLinus Torvalds 11631da177e4SLinus Torvalds /* getting an ENOSYS error indicates that direct mmap isn't 11641da177e4SLinus Torvalds * possible (as opposed to tried but failed) so we'll try to 11651da177e4SLinus Torvalds * make a private copy of the data and map that instead */ 11661da177e4SLinus Torvalds } 11671da177e4SLinus Torvalds 11688feae131SDavid Howells 11691da177e4SLinus Torvalds /* allocate some memory to hold the mapping 11701da177e4SLinus Torvalds * - note that this may not return a page-aligned address if the object 11711da177e4SLinus Torvalds * we're allocating is smaller than a page 11721da177e4SLinus Torvalds */ 1173f67d9b15SBob Liu order = get_order(len); 11748feae131SDavid Howells kdebug("alloc order %d for %lx", order, len); 11758feae131SDavid Howells 11768feae131SDavid Howells pages = alloc_pages(GFP_KERNEL, order); 11778feae131SDavid Howells if (!pages) 11781da177e4SLinus Torvalds goto enomem; 11791da177e4SLinus Torvalds 11808feae131SDavid Howells total = 1 << order; 118133e5d769SDavid Howells atomic_long_add(total, &mmap_pages_allocated); 11821da177e4SLinus Torvalds 1183f67d9b15SBob Liu point = len >> PAGE_SHIFT; 1184dd8632a1SPaul Mundt 1185dd8632a1SPaul Mundt /* we allocated a power-of-2 sized page set, so we may want to trim off 1186dd8632a1SPaul Mundt * the excess */ 1187dd8632a1SPaul Mundt if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { 11888feae131SDavid Howells while (total > point) { 11898feae131SDavid Howells order = ilog2(total - point); 11908feae131SDavid Howells n = 1 << order; 11918feae131SDavid Howells kdebug("shave %lu/%lu @%lu", n, total - point, total); 119233e5d769SDavid Howells atomic_long_sub(n, &mmap_pages_allocated); 11938feae131SDavid Howells total -= n; 11948feae131SDavid Howells set_page_refcounted(pages + total); 11958feae131SDavid Howells __free_pages(pages + total, order); 11968feae131SDavid Howells } 1197dd8632a1SPaul Mundt } 11988feae131SDavid Howells 11998feae131SDavid Howells for (point = 1; point < total; point++) 12008feae131SDavid Howells set_page_refcounted(&pages[point]); 12018feae131SDavid Howells 12028feae131SDavid Howells base = page_address(pages); 12038feae131SDavid Howells region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; 12048feae131SDavid Howells region->vm_start = (unsigned long) base; 1205f67d9b15SBob Liu region->vm_end = region->vm_start + len; 1206dd8632a1SPaul Mundt region->vm_top = region->vm_start + (total << PAGE_SHIFT); 12078feae131SDavid Howells 12088feae131SDavid Howells vma->vm_start = region->vm_start; 12098feae131SDavid Howells vma->vm_end = region->vm_start + len; 12101da177e4SLinus Torvalds 12111da177e4SLinus Torvalds if (vma->vm_file) { 12121da177e4SLinus Torvalds /* read the contents of a file into the copy */ 12131da177e4SLinus Torvalds mm_segment_t old_fs; 12141da177e4SLinus Torvalds loff_t fpos; 12151da177e4SLinus Torvalds 12161da177e4SLinus Torvalds fpos = vma->vm_pgoff; 12171da177e4SLinus Torvalds fpos <<= PAGE_SHIFT; 12181da177e4SLinus Torvalds 12191da177e4SLinus Torvalds old_fs = get_fs(); 12201da177e4SLinus Torvalds set_fs(KERNEL_DS); 1221f67d9b15SBob Liu ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos); 12221da177e4SLinus Torvalds set_fs(old_fs); 12231da177e4SLinus Torvalds 12241da177e4SLinus Torvalds if (ret < 0) 12251da177e4SLinus Torvalds goto error_free; 12261da177e4SLinus Torvalds 12271da177e4SLinus Torvalds /* clear the last little bit */ 1228f67d9b15SBob Liu if (ret < len) 1229f67d9b15SBob Liu memset(base + ret, 0, len - ret); 12301da177e4SLinus Torvalds 12311da177e4SLinus Torvalds } 12321da177e4SLinus Torvalds 12331da177e4SLinus Torvalds return 0; 12341da177e4SLinus Torvalds 12351da177e4SLinus Torvalds error_free: 12367223bb4aSNamhyung Kim free_page_series(region->vm_start, region->vm_top); 12378feae131SDavid Howells region->vm_start = vma->vm_start = 0; 12388feae131SDavid Howells region->vm_end = vma->vm_end = 0; 1239dd8632a1SPaul Mundt region->vm_top = 0; 12401da177e4SLinus Torvalds return ret; 12411da177e4SLinus Torvalds 12421da177e4SLinus Torvalds enomem: 124305ae6fa3SGreg Ungerer printk("Allocation of length %lu from process %d (%s) failed\n", 124405ae6fa3SGreg Ungerer len, current->pid, current->comm); 12457bf02ea2SDavid Rientjes show_free_areas(0); 12461da177e4SLinus Torvalds return -ENOMEM; 12471da177e4SLinus Torvalds } 12481da177e4SLinus Torvalds 12491da177e4SLinus Torvalds /* 12501da177e4SLinus Torvalds * handle mapping creation for uClinux 12511da177e4SLinus Torvalds */ 1252e3fc629dSAl Viro unsigned long do_mmap_pgoff(struct file *file, 12531da177e4SLinus Torvalds unsigned long addr, 12541da177e4SLinus Torvalds unsigned long len, 12551da177e4SLinus Torvalds unsigned long prot, 12561da177e4SLinus Torvalds unsigned long flags, 1257bebeb3d6SMichel Lespinasse unsigned long pgoff, 125841badc15SMichel Lespinasse unsigned long *populate) 12591da177e4SLinus Torvalds { 12608feae131SDavid Howells struct vm_area_struct *vma; 12618feae131SDavid Howells struct vm_region *region; 12621da177e4SLinus Torvalds struct rb_node *rb; 12638feae131SDavid Howells unsigned long capabilities, vm_flags, result; 12641da177e4SLinus Torvalds int ret; 12651da177e4SLinus Torvalds 12668feae131SDavid Howells kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); 12678feae131SDavid Howells 126841badc15SMichel Lespinasse *populate = 0; 1269bebeb3d6SMichel Lespinasse 12701da177e4SLinus Torvalds /* decide whether we should attempt the mapping, and if so what sort of 12711da177e4SLinus Torvalds * mapping */ 12721da177e4SLinus Torvalds ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 12731da177e4SLinus Torvalds &capabilities); 12748feae131SDavid Howells if (ret < 0) { 12758feae131SDavid Howells kleave(" = %d [val]", ret); 12761da177e4SLinus Torvalds return ret; 12778feae131SDavid Howells } 12781da177e4SLinus Torvalds 127906aab5a3SDavid Howells /* we ignore the address hint */ 128006aab5a3SDavid Howells addr = 0; 1281f67d9b15SBob Liu len = PAGE_ALIGN(len); 128206aab5a3SDavid Howells 12831da177e4SLinus Torvalds /* we've determined that we can make the mapping, now translate what we 12841da177e4SLinus Torvalds * now know into VMA flags */ 12851da177e4SLinus Torvalds vm_flags = determine_vm_flags(file, prot, flags, capabilities); 12861da177e4SLinus Torvalds 12878feae131SDavid Howells /* we're going to need to record the mapping */ 12888feae131SDavid Howells region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); 12898feae131SDavid Howells if (!region) 12908feae131SDavid Howells goto error_getting_region; 12911da177e4SLinus Torvalds 12928feae131SDavid Howells vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 12938feae131SDavid Howells if (!vma) 12948feae131SDavid Howells goto error_getting_vma; 12951da177e4SLinus Torvalds 12961e2ae599SDavid Howells region->vm_usage = 1; 12978feae131SDavid Howells region->vm_flags = vm_flags; 12988feae131SDavid Howells region->vm_pgoff = pgoff; 12998feae131SDavid Howells 13005beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 13018feae131SDavid Howells vma->vm_flags = vm_flags; 13028feae131SDavid Howells vma->vm_pgoff = pgoff; 13038feae131SDavid Howells 13048feae131SDavid Howells if (file) { 1305cb0942b8SAl Viro region->vm_file = get_file(file); 1306cb0942b8SAl Viro vma->vm_file = get_file(file); 13078feae131SDavid Howells } 13088feae131SDavid Howells 13098feae131SDavid Howells down_write(&nommu_region_sem); 13108feae131SDavid Howells 13118feae131SDavid Howells /* if we want to share, we need to check for regions created by other 13121da177e4SLinus Torvalds * mmap() calls that overlap with our proposed mapping 13138feae131SDavid Howells * - we can only share with a superset match on most regular files 13141da177e4SLinus Torvalds * - shared mappings on character devices and memory backed files are 13151da177e4SLinus Torvalds * permitted to overlap inexactly as far as we are concerned for in 13161da177e4SLinus Torvalds * these cases, sharing is handled in the driver or filesystem rather 13171da177e4SLinus Torvalds * than here 13181da177e4SLinus Torvalds */ 13191da177e4SLinus Torvalds if (vm_flags & VM_MAYSHARE) { 13208feae131SDavid Howells struct vm_region *pregion; 13218feae131SDavid Howells unsigned long pglen, rpglen, pgend, rpgend, start; 13221da177e4SLinus Torvalds 13238feae131SDavid Howells pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 13248feae131SDavid Howells pgend = pgoff + pglen; 1325165b2392SDavid Howells 13268feae131SDavid Howells for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { 13278feae131SDavid Howells pregion = rb_entry(rb, struct vm_region, vm_rb); 13281da177e4SLinus Torvalds 13298feae131SDavid Howells if (!(pregion->vm_flags & VM_MAYSHARE)) 13301da177e4SLinus Torvalds continue; 13311da177e4SLinus Torvalds 13321da177e4SLinus Torvalds /* search for overlapping mappings on the same file */ 1333496ad9aaSAl Viro if (file_inode(pregion->vm_file) != 1334496ad9aaSAl Viro file_inode(file)) 13351da177e4SLinus Torvalds continue; 13361da177e4SLinus Torvalds 13378feae131SDavid Howells if (pregion->vm_pgoff >= pgend) 13381da177e4SLinus Torvalds continue; 13391da177e4SLinus Torvalds 13408feae131SDavid Howells rpglen = pregion->vm_end - pregion->vm_start; 13418feae131SDavid Howells rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; 13428feae131SDavid Howells rpgend = pregion->vm_pgoff + rpglen; 13438feae131SDavid Howells if (pgoff >= rpgend) 13441da177e4SLinus Torvalds continue; 13451da177e4SLinus Torvalds 13468feae131SDavid Howells /* handle inexactly overlapping matches between 13478feae131SDavid Howells * mappings */ 13488feae131SDavid Howells if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && 13498feae131SDavid Howells !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { 13508feae131SDavid Howells /* new mapping is not a subset of the region */ 13511da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_DIRECT)) 13521da177e4SLinus Torvalds goto sharing_violation; 13531da177e4SLinus Torvalds continue; 13541da177e4SLinus Torvalds } 13551da177e4SLinus Torvalds 13568feae131SDavid Howells /* we've found a region we can share */ 13571e2ae599SDavid Howells pregion->vm_usage++; 13588feae131SDavid Howells vma->vm_region = pregion; 13598feae131SDavid Howells start = pregion->vm_start; 13608feae131SDavid Howells start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; 13618feae131SDavid Howells vma->vm_start = start; 13628feae131SDavid Howells vma->vm_end = start + len; 13631da177e4SLinus Torvalds 13648feae131SDavid Howells if (pregion->vm_flags & VM_MAPPED_COPY) { 13658feae131SDavid Howells kdebug("share copy"); 13668feae131SDavid Howells vma->vm_flags |= VM_MAPPED_COPY; 13678feae131SDavid Howells } else { 13688feae131SDavid Howells kdebug("share mmap"); 13698feae131SDavid Howells ret = do_mmap_shared_file(vma); 13708feae131SDavid Howells if (ret < 0) { 13718feae131SDavid Howells vma->vm_region = NULL; 13728feae131SDavid Howells vma->vm_start = 0; 13738feae131SDavid Howells vma->vm_end = 0; 13741e2ae599SDavid Howells pregion->vm_usage--; 13758feae131SDavid Howells pregion = NULL; 13768feae131SDavid Howells goto error_just_free; 13771da177e4SLinus Torvalds } 13788feae131SDavid Howells } 13798feae131SDavid Howells fput(region->vm_file); 13808feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 13818feae131SDavid Howells region = pregion; 13828feae131SDavid Howells result = start; 13838feae131SDavid Howells goto share; 13848feae131SDavid Howells } 13851da177e4SLinus Torvalds 13861da177e4SLinus Torvalds /* obtain the address at which to make a shared mapping 13871da177e4SLinus Torvalds * - this is the hook for quasi-memory character devices to 13881da177e4SLinus Torvalds * tell us the location of a shared mapping 13891da177e4SLinus Torvalds */ 1390645d83c5SDavid Howells if (capabilities & BDI_CAP_MAP_DIRECT) { 13911da177e4SLinus Torvalds addr = file->f_op->get_unmapped_area(file, addr, len, 13921da177e4SLinus Torvalds pgoff, flags); 1393bb005a59SNamhyung Kim if (IS_ERR_VALUE(addr)) { 13941da177e4SLinus Torvalds ret = addr; 1395bb005a59SNamhyung Kim if (ret != -ENOSYS) 13968feae131SDavid Howells goto error_just_free; 13971da177e4SLinus Torvalds 13981da177e4SLinus Torvalds /* the driver refused to tell us where to site 13991da177e4SLinus Torvalds * the mapping so we'll have to attempt to copy 14001da177e4SLinus Torvalds * it */ 1401bb005a59SNamhyung Kim ret = -ENODEV; 14021da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_COPY)) 14038feae131SDavid Howells goto error_just_free; 14041da177e4SLinus Torvalds 14051da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 14068feae131SDavid Howells } else { 14078feae131SDavid Howells vma->vm_start = region->vm_start = addr; 14088feae131SDavid Howells vma->vm_end = region->vm_end = addr + len; 14091da177e4SLinus Torvalds } 14101da177e4SLinus Torvalds } 14111da177e4SLinus Torvalds } 14121da177e4SLinus Torvalds 14138feae131SDavid Howells vma->vm_region = region; 14141da177e4SLinus Torvalds 1415645d83c5SDavid Howells /* set up the mapping 1416645d83c5SDavid Howells * - the region is filled in if BDI_CAP_MAP_DIRECT is still set 1417645d83c5SDavid Howells */ 14181da177e4SLinus Torvalds if (file && vma->vm_flags & VM_SHARED) 14198feae131SDavid Howells ret = do_mmap_shared_file(vma); 14201da177e4SLinus Torvalds else 1421645d83c5SDavid Howells ret = do_mmap_private(vma, region, len, capabilities); 14221da177e4SLinus Torvalds if (ret < 0) 1423645d83c5SDavid Howells goto error_just_free; 1424645d83c5SDavid Howells add_nommu_region(region); 14258feae131SDavid Howells 1426ea637639SJie Zhang /* clear anonymous mappings that don't ask for uninitialized data */ 1427ea637639SJie Zhang if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) 1428ea637639SJie Zhang memset((void *)region->vm_start, 0, 1429ea637639SJie Zhang region->vm_end - region->vm_start); 1430ea637639SJie Zhang 14311da177e4SLinus Torvalds /* okay... we have a mapping; now we have to register it */ 14328feae131SDavid Howells result = vma->vm_start; 14331da177e4SLinus Torvalds 14341da177e4SLinus Torvalds current->mm->total_vm += len >> PAGE_SHIFT; 14351da177e4SLinus Torvalds 14368feae131SDavid Howells share: 14378feae131SDavid Howells add_vma_to_mm(current->mm, vma); 14381da177e4SLinus Torvalds 1439cfe79c00SMike Frysinger /* we flush the region from the icache only when the first executable 1440cfe79c00SMike Frysinger * mapping of it is made */ 1441cfe79c00SMike Frysinger if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { 1442cfe79c00SMike Frysinger flush_icache_range(region->vm_start, region->vm_end); 1443cfe79c00SMike Frysinger region->vm_icache_flushed = true; 1444cfe79c00SMike Frysinger } 14451da177e4SLinus Torvalds 1446cfe79c00SMike Frysinger up_write(&nommu_region_sem); 14471da177e4SLinus Torvalds 14488feae131SDavid Howells kleave(" = %lx", result); 14498feae131SDavid Howells return result; 14501da177e4SLinus Torvalds 14518feae131SDavid Howells error_just_free: 14528feae131SDavid Howells up_write(&nommu_region_sem); 14538feae131SDavid Howells error: 145489a86402SDavid Howells if (region->vm_file) 14558feae131SDavid Howells fput(region->vm_file); 14568feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 145789a86402SDavid Howells if (vma->vm_file) 14588feae131SDavid Howells fput(vma->vm_file); 14598feae131SDavid Howells kmem_cache_free(vm_area_cachep, vma); 14608feae131SDavid Howells kleave(" = %d", ret); 14611da177e4SLinus Torvalds return ret; 14621da177e4SLinus Torvalds 14631da177e4SLinus Torvalds sharing_violation: 14648feae131SDavid Howells up_write(&nommu_region_sem); 14658feae131SDavid Howells printk(KERN_WARNING "Attempt to share mismatched mappings\n"); 14668feae131SDavid Howells ret = -EINVAL; 14678feae131SDavid Howells goto error; 14681da177e4SLinus Torvalds 14691da177e4SLinus Torvalds error_getting_vma: 14708feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 14718feae131SDavid Howells printk(KERN_WARNING "Allocation of vma for %lu byte allocation" 14728feae131SDavid Howells " from process %d failed\n", 14731da177e4SLinus Torvalds len, current->pid); 14747bf02ea2SDavid Rientjes show_free_areas(0); 14751da177e4SLinus Torvalds return -ENOMEM; 14761da177e4SLinus Torvalds 14778feae131SDavid Howells error_getting_region: 14788feae131SDavid Howells printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" 14798feae131SDavid Howells " from process %d failed\n", 14801da177e4SLinus Torvalds len, current->pid); 14817bf02ea2SDavid Rientjes show_free_areas(0); 14821da177e4SLinus Torvalds return -ENOMEM; 14831da177e4SLinus Torvalds } 14846be5ceb0SLinus Torvalds 148566f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 148666f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 148766f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 148866f0dc48SHugh Dickins { 148966f0dc48SHugh Dickins struct file *file = NULL; 149066f0dc48SHugh Dickins unsigned long retval = -EBADF; 149166f0dc48SHugh Dickins 1492120a795dSAl Viro audit_mmap_fd(fd, flags); 149366f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 149466f0dc48SHugh Dickins file = fget(fd); 149566f0dc48SHugh Dickins if (!file) 149666f0dc48SHugh Dickins goto out; 149766f0dc48SHugh Dickins } 149866f0dc48SHugh Dickins 149966f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 150066f0dc48SHugh Dickins 1501ad1ed293SGreg Ungerer retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 150266f0dc48SHugh Dickins 150366f0dc48SHugh Dickins if (file) 150466f0dc48SHugh Dickins fput(file); 150566f0dc48SHugh Dickins out: 150666f0dc48SHugh Dickins return retval; 150766f0dc48SHugh Dickins } 150866f0dc48SHugh Dickins 1509a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1510a4679373SChristoph Hellwig struct mmap_arg_struct { 1511a4679373SChristoph Hellwig unsigned long addr; 1512a4679373SChristoph Hellwig unsigned long len; 1513a4679373SChristoph Hellwig unsigned long prot; 1514a4679373SChristoph Hellwig unsigned long flags; 1515a4679373SChristoph Hellwig unsigned long fd; 1516a4679373SChristoph Hellwig unsigned long offset; 1517a4679373SChristoph Hellwig }; 1518a4679373SChristoph Hellwig 1519a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1520a4679373SChristoph Hellwig { 1521a4679373SChristoph Hellwig struct mmap_arg_struct a; 1522a4679373SChristoph Hellwig 1523a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1524a4679373SChristoph Hellwig return -EFAULT; 1525a4679373SChristoph Hellwig if (a.offset & ~PAGE_MASK) 1526a4679373SChristoph Hellwig return -EINVAL; 1527a4679373SChristoph Hellwig 1528a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1529a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1530a4679373SChristoph Hellwig } 1531a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1532a4679373SChristoph Hellwig 15331da177e4SLinus Torvalds /* 15348feae131SDavid Howells * split a vma into two pieces at address 'addr', a new vma is allocated either 15358feae131SDavid Howells * for the first part or the tail. 15361da177e4SLinus Torvalds */ 15378feae131SDavid Howells int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 15388feae131SDavid Howells unsigned long addr, int new_below) 15391da177e4SLinus Torvalds { 15408feae131SDavid Howells struct vm_area_struct *new; 15418feae131SDavid Howells struct vm_region *region; 15428feae131SDavid Howells unsigned long npages; 15431da177e4SLinus Torvalds 15448feae131SDavid Howells kenter(""); 15451da177e4SLinus Torvalds 1546779c1023SDavid Howells /* we're only permitted to split anonymous regions (these should have 1547779c1023SDavid Howells * only a single usage on the region) */ 1548779c1023SDavid Howells if (vma->vm_file) 15498feae131SDavid Howells return -ENOMEM; 15501da177e4SLinus Torvalds 15518feae131SDavid Howells if (mm->map_count >= sysctl_max_map_count) 15528feae131SDavid Howells return -ENOMEM; 15531da177e4SLinus Torvalds 15548feae131SDavid Howells region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); 15558feae131SDavid Howells if (!region) 15568feae131SDavid Howells return -ENOMEM; 15578feae131SDavid Howells 15588feae131SDavid Howells new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 15598feae131SDavid Howells if (!new) { 15608feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 15618feae131SDavid Howells return -ENOMEM; 15621da177e4SLinus Torvalds } 15631da177e4SLinus Torvalds 15648feae131SDavid Howells /* most fields are the same, copy all, and then fixup */ 15658feae131SDavid Howells *new = *vma; 15668feae131SDavid Howells *region = *vma->vm_region; 15678feae131SDavid Howells new->vm_region = region; 15688feae131SDavid Howells 15698feae131SDavid Howells npages = (addr - vma->vm_start) >> PAGE_SHIFT; 15708feae131SDavid Howells 15718feae131SDavid Howells if (new_below) { 1572dd8632a1SPaul Mundt region->vm_top = region->vm_end = new->vm_end = addr; 15738feae131SDavid Howells } else { 15748feae131SDavid Howells region->vm_start = new->vm_start = addr; 15758feae131SDavid Howells region->vm_pgoff = new->vm_pgoff += npages; 15761da177e4SLinus Torvalds } 15778feae131SDavid Howells 15788feae131SDavid Howells if (new->vm_ops && new->vm_ops->open) 15798feae131SDavid Howells new->vm_ops->open(new); 15808feae131SDavid Howells 15818feae131SDavid Howells delete_vma_from_mm(vma); 15828feae131SDavid Howells down_write(&nommu_region_sem); 15838feae131SDavid Howells delete_nommu_region(vma->vm_region); 15848feae131SDavid Howells if (new_below) { 15858feae131SDavid Howells vma->vm_region->vm_start = vma->vm_start = addr; 15868feae131SDavid Howells vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; 15878feae131SDavid Howells } else { 15888feae131SDavid Howells vma->vm_region->vm_end = vma->vm_end = addr; 1589dd8632a1SPaul Mundt vma->vm_region->vm_top = addr; 15908feae131SDavid Howells } 15918feae131SDavid Howells add_nommu_region(vma->vm_region); 15928feae131SDavid Howells add_nommu_region(new->vm_region); 15938feae131SDavid Howells up_write(&nommu_region_sem); 15948feae131SDavid Howells add_vma_to_mm(mm, vma); 15958feae131SDavid Howells add_vma_to_mm(mm, new); 15968feae131SDavid Howells return 0; 15978feae131SDavid Howells } 15988feae131SDavid Howells 15998feae131SDavid Howells /* 16008feae131SDavid Howells * shrink a VMA by removing the specified chunk from either the beginning or 16018feae131SDavid Howells * the end 16028feae131SDavid Howells */ 16038feae131SDavid Howells static int shrink_vma(struct mm_struct *mm, 16048feae131SDavid Howells struct vm_area_struct *vma, 16058feae131SDavid Howells unsigned long from, unsigned long to) 16068feae131SDavid Howells { 16078feae131SDavid Howells struct vm_region *region; 16088feae131SDavid Howells 16098feae131SDavid Howells kenter(""); 16108feae131SDavid Howells 16118feae131SDavid Howells /* adjust the VMA's pointers, which may reposition it in the MM's tree 16128feae131SDavid Howells * and list */ 16138feae131SDavid Howells delete_vma_from_mm(vma); 16148feae131SDavid Howells if (from > vma->vm_start) 16158feae131SDavid Howells vma->vm_end = from; 16168feae131SDavid Howells else 16178feae131SDavid Howells vma->vm_start = to; 16188feae131SDavid Howells add_vma_to_mm(mm, vma); 16198feae131SDavid Howells 16208feae131SDavid Howells /* cut the backing region down to size */ 16218feae131SDavid Howells region = vma->vm_region; 16221e2ae599SDavid Howells BUG_ON(region->vm_usage != 1); 16238feae131SDavid Howells 16248feae131SDavid Howells down_write(&nommu_region_sem); 16258feae131SDavid Howells delete_nommu_region(region); 1626dd8632a1SPaul Mundt if (from > region->vm_start) { 1627dd8632a1SPaul Mundt to = region->vm_top; 1628dd8632a1SPaul Mundt region->vm_top = region->vm_end = from; 1629dd8632a1SPaul Mundt } else { 16308feae131SDavid Howells region->vm_start = to; 1631dd8632a1SPaul Mundt } 16328feae131SDavid Howells add_nommu_region(region); 16338feae131SDavid Howells up_write(&nommu_region_sem); 16348feae131SDavid Howells 16358feae131SDavid Howells free_page_series(from, to); 16368feae131SDavid Howells return 0; 16371da177e4SLinus Torvalds } 16381da177e4SLinus Torvalds 16393034097aSDavid Howells /* 16403034097aSDavid Howells * release a mapping 16418feae131SDavid Howells * - under NOMMU conditions the chunk to be unmapped must be backed by a single 16428feae131SDavid Howells * VMA, though it need not cover the whole VMA 16433034097aSDavid Howells */ 16448feae131SDavid Howells int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 16451da177e4SLinus Torvalds { 16468feae131SDavid Howells struct vm_area_struct *vma; 1647f67d9b15SBob Liu unsigned long end; 16488feae131SDavid Howells int ret; 16491da177e4SLinus Torvalds 16508feae131SDavid Howells kenter(",%lx,%zx", start, len); 16511da177e4SLinus Torvalds 1652f67d9b15SBob Liu len = PAGE_ALIGN(len); 16538feae131SDavid Howells if (len == 0) 16541da177e4SLinus Torvalds return -EINVAL; 16551da177e4SLinus Torvalds 1656f67d9b15SBob Liu end = start + len; 1657f67d9b15SBob Liu 16588feae131SDavid Howells /* find the first potentially overlapping VMA */ 16598feae131SDavid Howells vma = find_vma(mm, start); 16608feae131SDavid Howells if (!vma) { 166133e5d769SDavid Howells static int limit = 0; 166233e5d769SDavid Howells if (limit < 5) { 16638feae131SDavid Howells printk(KERN_WARNING 166433e5d769SDavid Howells "munmap of memory not mmapped by process %d" 166533e5d769SDavid Howells " (%s): 0x%lx-0x%lx\n", 166633e5d769SDavid Howells current->pid, current->comm, 166733e5d769SDavid Howells start, start + len - 1); 166833e5d769SDavid Howells limit++; 166933e5d769SDavid Howells } 16708feae131SDavid Howells return -EINVAL; 16718feae131SDavid Howells } 16721da177e4SLinus Torvalds 16738feae131SDavid Howells /* we're allowed to split an anonymous VMA but not a file-backed one */ 16748feae131SDavid Howells if (vma->vm_file) { 16758feae131SDavid Howells do { 16768feae131SDavid Howells if (start > vma->vm_start) { 16778feae131SDavid Howells kleave(" = -EINVAL [miss]"); 16788feae131SDavid Howells return -EINVAL; 16798feae131SDavid Howells } 16808feae131SDavid Howells if (end == vma->vm_end) 16818feae131SDavid Howells goto erase_whole_vma; 1682d75a310cSNamhyung Kim vma = vma->vm_next; 1683d75a310cSNamhyung Kim } while (vma); 16848feae131SDavid Howells kleave(" = -EINVAL [split file]"); 16858feae131SDavid Howells return -EINVAL; 16868feae131SDavid Howells } else { 16878feae131SDavid Howells /* the chunk must be a subset of the VMA found */ 16888feae131SDavid Howells if (start == vma->vm_start && end == vma->vm_end) 16898feae131SDavid Howells goto erase_whole_vma; 16908feae131SDavid Howells if (start < vma->vm_start || end > vma->vm_end) { 16918feae131SDavid Howells kleave(" = -EINVAL [superset]"); 16928feae131SDavid Howells return -EINVAL; 16938feae131SDavid Howells } 16948feae131SDavid Howells if (start & ~PAGE_MASK) { 16958feae131SDavid Howells kleave(" = -EINVAL [unaligned start]"); 16968feae131SDavid Howells return -EINVAL; 16978feae131SDavid Howells } 16988feae131SDavid Howells if (end != vma->vm_end && end & ~PAGE_MASK) { 16998feae131SDavid Howells kleave(" = -EINVAL [unaligned split]"); 17008feae131SDavid Howells return -EINVAL; 17018feae131SDavid Howells } 17028feae131SDavid Howells if (start != vma->vm_start && end != vma->vm_end) { 17038feae131SDavid Howells ret = split_vma(mm, vma, start, 1); 17048feae131SDavid Howells if (ret < 0) { 17058feae131SDavid Howells kleave(" = %d [split]", ret); 17068feae131SDavid Howells return ret; 17078feae131SDavid Howells } 17088feae131SDavid Howells } 17098feae131SDavid Howells return shrink_vma(mm, vma, start, end); 17108feae131SDavid Howells } 17111da177e4SLinus Torvalds 17128feae131SDavid Howells erase_whole_vma: 17138feae131SDavid Howells delete_vma_from_mm(vma); 17148feae131SDavid Howells delete_vma(mm, vma); 17158feae131SDavid Howells kleave(" = 0"); 17161da177e4SLinus Torvalds return 0; 17171da177e4SLinus Torvalds } 1718b5073173SPaul Mundt EXPORT_SYMBOL(do_munmap); 17191da177e4SLinus Torvalds 1720bfce281cSAl Viro int vm_munmap(unsigned long addr, size_t len) 17213034097aSDavid Howells { 1722bfce281cSAl Viro struct mm_struct *mm = current->mm; 17233034097aSDavid Howells int ret; 17243034097aSDavid Howells 17253034097aSDavid Howells down_write(&mm->mmap_sem); 17263034097aSDavid Howells ret = do_munmap(mm, addr, len); 17273034097aSDavid Howells up_write(&mm->mmap_sem); 17283034097aSDavid Howells return ret; 17293034097aSDavid Howells } 1730a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 1731a46ef99dSLinus Torvalds 1732a46ef99dSLinus Torvalds SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 1733a46ef99dSLinus Torvalds { 1734bfce281cSAl Viro return vm_munmap(addr, len); 1735a46ef99dSLinus Torvalds } 17363034097aSDavid Howells 17373034097aSDavid Howells /* 17388feae131SDavid Howells * release all the mappings made in a process's VM space 17393034097aSDavid Howells */ 17401da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 17411da177e4SLinus Torvalds { 17428feae131SDavid Howells struct vm_area_struct *vma; 17431da177e4SLinus Torvalds 17448feae131SDavid Howells if (!mm) 17458feae131SDavid Howells return; 17468feae131SDavid Howells 17478feae131SDavid Howells kenter(""); 17481da177e4SLinus Torvalds 17491da177e4SLinus Torvalds mm->total_vm = 0; 17501da177e4SLinus Torvalds 17518feae131SDavid Howells while ((vma = mm->mmap)) { 17528feae131SDavid Howells mm->mmap = vma->vm_next; 17538feae131SDavid Howells delete_vma_from_mm(vma); 17548feae131SDavid Howells delete_vma(mm, vma); 175504c34961SSteven J. Magnani cond_resched(); 17561da177e4SLinus Torvalds } 17571da177e4SLinus Torvalds 17588feae131SDavid Howells kleave(""); 17591da177e4SLinus Torvalds } 17601da177e4SLinus Torvalds 1761e4eb1ff6SLinus Torvalds unsigned long vm_brk(unsigned long addr, unsigned long len) 17621da177e4SLinus Torvalds { 17631da177e4SLinus Torvalds return -ENOMEM; 17641da177e4SLinus Torvalds } 17651da177e4SLinus Torvalds 17661da177e4SLinus Torvalds /* 17676fa5f80bSDavid Howells * expand (or shrink) an existing mapping, potentially moving it at the same 17686fa5f80bSDavid Howells * time (controlled by the MREMAP_MAYMOVE flag and available VM space) 17691da177e4SLinus Torvalds * 17706fa5f80bSDavid Howells * under NOMMU conditions, we only permit changing a mapping's size, and only 17718feae131SDavid Howells * as long as it stays within the region allocated by do_mmap_private() and the 17728feae131SDavid Howells * block is not shareable 17731da177e4SLinus Torvalds * 17746fa5f80bSDavid Howells * MREMAP_FIXED is not supported under NOMMU conditions 17751da177e4SLinus Torvalds */ 17764b377babSAl Viro static unsigned long do_mremap(unsigned long addr, 17771da177e4SLinus Torvalds unsigned long old_len, unsigned long new_len, 17781da177e4SLinus Torvalds unsigned long flags, unsigned long new_addr) 17791da177e4SLinus Torvalds { 17806fa5f80bSDavid Howells struct vm_area_struct *vma; 17811da177e4SLinus Torvalds 17821da177e4SLinus Torvalds /* insanity checks first */ 1783f67d9b15SBob Liu old_len = PAGE_ALIGN(old_len); 1784f67d9b15SBob Liu new_len = PAGE_ALIGN(new_len); 17858feae131SDavid Howells if (old_len == 0 || new_len == 0) 17861da177e4SLinus Torvalds return (unsigned long) -EINVAL; 17871da177e4SLinus Torvalds 17888feae131SDavid Howells if (addr & ~PAGE_MASK) 17898feae131SDavid Howells return -EINVAL; 17908feae131SDavid Howells 17911da177e4SLinus Torvalds if (flags & MREMAP_FIXED && new_addr != addr) 17921da177e4SLinus Torvalds return (unsigned long) -EINVAL; 17931da177e4SLinus Torvalds 17948feae131SDavid Howells vma = find_vma_exact(current->mm, addr, old_len); 17956fa5f80bSDavid Howells if (!vma) 17961da177e4SLinus Torvalds return (unsigned long) -EINVAL; 17971da177e4SLinus Torvalds 17986fa5f80bSDavid Howells if (vma->vm_end != vma->vm_start + old_len) 17991da177e4SLinus Torvalds return (unsigned long) -EFAULT; 18001da177e4SLinus Torvalds 18016fa5f80bSDavid Howells if (vma->vm_flags & VM_MAYSHARE) 18021da177e4SLinus Torvalds return (unsigned long) -EPERM; 18031da177e4SLinus Torvalds 18048feae131SDavid Howells if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) 18051da177e4SLinus Torvalds return (unsigned long) -ENOMEM; 18061da177e4SLinus Torvalds 18071da177e4SLinus Torvalds /* all checks complete - do it */ 18086fa5f80bSDavid Howells vma->vm_end = vma->vm_start + new_len; 18096fa5f80bSDavid Howells return vma->vm_start; 18106fa5f80bSDavid Howells } 18116fa5f80bSDavid Howells 18126a6160a7SHeiko Carstens SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, 18136a6160a7SHeiko Carstens unsigned long, new_len, unsigned long, flags, 18146a6160a7SHeiko Carstens unsigned long, new_addr) 18156fa5f80bSDavid Howells { 18166fa5f80bSDavid Howells unsigned long ret; 18176fa5f80bSDavid Howells 18186fa5f80bSDavid Howells down_write(¤t->mm->mmap_sem); 18196fa5f80bSDavid Howells ret = do_mremap(addr, old_len, new_len, flags, new_addr); 18206fa5f80bSDavid Howells up_write(¤t->mm->mmap_sem); 18216fa5f80bSDavid Howells return ret; 18221da177e4SLinus Torvalds } 18231da177e4SLinus Torvalds 1824240aadeeSMichel Lespinasse struct page *follow_page_mask(struct vm_area_struct *vma, 1825240aadeeSMichel Lespinasse unsigned long address, unsigned int flags, 1826240aadeeSMichel Lespinasse unsigned int *page_mask) 18271da177e4SLinus Torvalds { 1828240aadeeSMichel Lespinasse *page_mask = 0; 18291da177e4SLinus Torvalds return NULL; 18301da177e4SLinus Torvalds } 18311da177e4SLinus Torvalds 18328f3b1327SBob Liu int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 18338f3b1327SBob Liu unsigned long pfn, unsigned long size, pgprot_t prot) 18341da177e4SLinus Torvalds { 18358f3b1327SBob Liu if (addr != (pfn << PAGE_SHIFT)) 18368f3b1327SBob Liu return -EINVAL; 18378f3b1327SBob Liu 1838314e51b9SKonstantin Khlebnikov vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 183966aa2b4bSGreg Ungerer return 0; 18401da177e4SLinus Torvalds } 184122c4af40SLuke Yang EXPORT_SYMBOL(remap_pfn_range); 18421da177e4SLinus Torvalds 18433c0b9de6SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 18443c0b9de6SLinus Torvalds { 18453c0b9de6SLinus Torvalds unsigned long pfn = start >> PAGE_SHIFT; 18463c0b9de6SLinus Torvalds unsigned long vm_len = vma->vm_end - vma->vm_start; 18473c0b9de6SLinus Torvalds 18483c0b9de6SLinus Torvalds pfn += vma->vm_pgoff; 18493c0b9de6SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 18503c0b9de6SLinus Torvalds } 18513c0b9de6SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 18523c0b9de6SLinus Torvalds 1853f905bc44SPaul Mundt int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, 1854f905bc44SPaul Mundt unsigned long pgoff) 1855f905bc44SPaul Mundt { 1856f905bc44SPaul Mundt unsigned int size = vma->vm_end - vma->vm_start; 1857f905bc44SPaul Mundt 1858f905bc44SPaul Mundt if (!(vma->vm_flags & VM_USERMAP)) 1859f905bc44SPaul Mundt return -EINVAL; 1860f905bc44SPaul Mundt 1861f905bc44SPaul Mundt vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); 1862f905bc44SPaul Mundt vma->vm_end = vma->vm_start + size; 1863f905bc44SPaul Mundt 1864f905bc44SPaul Mundt return 0; 1865f905bc44SPaul Mundt } 1866f905bc44SPaul Mundt EXPORT_SYMBOL(remap_vmalloc_range); 1867f905bc44SPaul Mundt 18681da177e4SLinus Torvalds unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, 18691da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 18701da177e4SLinus Torvalds { 18711da177e4SLinus Torvalds return -ENOMEM; 18721da177e4SLinus Torvalds } 18731da177e4SLinus Torvalds 18741da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 18751da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, 18761da177e4SLinus Torvalds int even_cows) 18771da177e4SLinus Torvalds { 18781da177e4SLinus Torvalds } 187922c4af40SLuke Yang EXPORT_SYMBOL(unmap_mapping_range); 18801da177e4SLinus Torvalds 18811da177e4SLinus Torvalds /* 18821da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 18831da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 18841da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 18851da177e4SLinus Torvalds * 18861da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 18871da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 18881da177e4SLinus Torvalds * 18891da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 18901da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 18911da177e4SLinus Torvalds * 18921da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 18931da177e4SLinus Torvalds * 18941da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 18951da177e4SLinus Torvalds * wish to use this logic. 18961da177e4SLinus Torvalds */ 189734b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 18981da177e4SLinus Torvalds { 1899c9b1d098SAndrew Shewmaker unsigned long free, allowed, reserve; 19001da177e4SLinus Torvalds 19011da177e4SLinus Torvalds vm_acct_memory(pages); 19021da177e4SLinus Torvalds 19031da177e4SLinus Torvalds /* 19041da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 19051da177e4SLinus Torvalds */ 19061da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 19071da177e4SLinus Torvalds return 0; 19081da177e4SLinus Torvalds 19091da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1910c15bef30SDmitry Fink free = global_page_state(NR_FREE_PAGES); 1911c15bef30SDmitry Fink free += global_page_state(NR_FILE_PAGES); 19121da177e4SLinus Torvalds 1913c15bef30SDmitry Fink /* 1914c15bef30SDmitry Fink * shmem pages shouldn't be counted as free in this 1915c15bef30SDmitry Fink * case, they can't be purged, only swapped out, and 1916c15bef30SDmitry Fink * that won't affect the overall amount of available 1917c15bef30SDmitry Fink * memory in the system. 1918c15bef30SDmitry Fink */ 1919c15bef30SDmitry Fink free -= global_page_state(NR_SHMEM); 1920c15bef30SDmitry Fink 1921ec8acf20SShaohua Li free += get_nr_swap_pages(); 19221da177e4SLinus Torvalds 19231da177e4SLinus Torvalds /* 19241da177e4SLinus Torvalds * Any slabs which are created with the 19251da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 19261da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 19271da177e4SLinus Torvalds * cache and most inode caches should fall into this 19281da177e4SLinus Torvalds */ 1929972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 19301da177e4SLinus Torvalds 19311da177e4SLinus Torvalds /* 1932c15bef30SDmitry Fink * Leave reserved pages. The pages are not for anonymous pages. 1933c15bef30SDmitry Fink */ 1934c15bef30SDmitry Fink if (free <= totalreserve_pages) 1935c15bef30SDmitry Fink goto error; 1936c15bef30SDmitry Fink else 1937c15bef30SDmitry Fink free -= totalreserve_pages; 1938c15bef30SDmitry Fink 1939c15bef30SDmitry Fink /* 19404eeab4f5SAndrew Shewmaker * Reserve some for root 19411da177e4SLinus Torvalds */ 19421da177e4SLinus Torvalds if (!cap_sys_admin) 19434eeab4f5SAndrew Shewmaker free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 19441da177e4SLinus Torvalds 19451da177e4SLinus Torvalds if (free > pages) 19461da177e4SLinus Torvalds return 0; 19471da177e4SLinus Torvalds 1948d5ddc79bSHideo AOKI goto error; 19491da177e4SLinus Torvalds } 19501da177e4SLinus Torvalds 19511da177e4SLinus Torvalds allowed = totalram_pages * sysctl_overcommit_ratio / 100; 19521da177e4SLinus Torvalds /* 19534eeab4f5SAndrew Shewmaker * Reserve some 3% for root 19541da177e4SLinus Torvalds */ 19551da177e4SLinus Torvalds if (!cap_sys_admin) 19564eeab4f5SAndrew Shewmaker allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 19571da177e4SLinus Torvalds allowed += total_swap_pages; 19581da177e4SLinus Torvalds 1959c9b1d098SAndrew Shewmaker /* 1960c9b1d098SAndrew Shewmaker * Don't let a single process grow so big a user can't recover 1961c9b1d098SAndrew Shewmaker */ 1962c9b1d098SAndrew Shewmaker if (mm) { 1963c9b1d098SAndrew Shewmaker reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); 1964c9b1d098SAndrew Shewmaker allowed -= min(mm->total_vm / 32, reserve); 1965c9b1d098SAndrew Shewmaker } 19661da177e4SLinus Torvalds 196700a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 19681da177e4SLinus Torvalds return 0; 196900a62ce9SKOSAKI Motohiro 1970d5ddc79bSHideo AOKI error: 19711da177e4SLinus Torvalds vm_unacct_memory(pages); 19721da177e4SLinus Torvalds 19731da177e4SLinus Torvalds return -ENOMEM; 19741da177e4SLinus Torvalds } 19751da177e4SLinus Torvalds 1976cae5d390SStephen Wilson int in_gate_area_no_mm(unsigned long addr) 19771da177e4SLinus Torvalds { 19781da177e4SLinus Torvalds return 0; 19791da177e4SLinus Torvalds } 1980b0e15190SDavid Howells 1981d0217ac0SNick Piggin int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1982b0e15190SDavid Howells { 1983b0e15190SDavid Howells BUG(); 1984d0217ac0SNick Piggin return 0; 1985b0e15190SDavid Howells } 1986b5073173SPaul Mundt EXPORT_SYMBOL(filemap_fault); 19870ec76a11SDavid Howells 19880b173bc4SKonstantin Khlebnikov int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr, 19890b173bc4SKonstantin Khlebnikov unsigned long size, pgoff_t pgoff) 19900b173bc4SKonstantin Khlebnikov { 19910b173bc4SKonstantin Khlebnikov BUG(); 19920b173bc4SKonstantin Khlebnikov return 0; 19930b173bc4SKonstantin Khlebnikov } 19940b173bc4SKonstantin Khlebnikov EXPORT_SYMBOL(generic_file_remap_pages); 19950b173bc4SKonstantin Khlebnikov 1996f55f199bSMike Frysinger static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 1997f55f199bSMike Frysinger unsigned long addr, void *buf, int len, int write) 19980ec76a11SDavid Howells { 19990ec76a11SDavid Howells struct vm_area_struct *vma; 20000ec76a11SDavid Howells 20010ec76a11SDavid Howells down_read(&mm->mmap_sem); 20020ec76a11SDavid Howells 20030ec76a11SDavid Howells /* the access must start within one of the target process's mappings */ 20040159b141SDavid Howells vma = find_vma(mm, addr); 20050159b141SDavid Howells if (vma) { 20060ec76a11SDavid Howells /* don't overrun this mapping */ 20070ec76a11SDavid Howells if (addr + len >= vma->vm_end) 20080ec76a11SDavid Howells len = vma->vm_end - addr; 20090ec76a11SDavid Howells 20100ec76a11SDavid Howells /* only read or write mappings where it is permitted */ 2011d00c7b99SDavid Howells if (write && vma->vm_flags & VM_MAYWRITE) 20127959722bSJie Zhang copy_to_user_page(vma, NULL, addr, 20137959722bSJie Zhang (void *) addr, buf, len); 2014d00c7b99SDavid Howells else if (!write && vma->vm_flags & VM_MAYREAD) 20157959722bSJie Zhang copy_from_user_page(vma, NULL, addr, 20167959722bSJie Zhang buf, (void *) addr, len); 20170ec76a11SDavid Howells else 20180ec76a11SDavid Howells len = 0; 20190ec76a11SDavid Howells } else { 20200ec76a11SDavid Howells len = 0; 20210ec76a11SDavid Howells } 20220ec76a11SDavid Howells 20230ec76a11SDavid Howells up_read(&mm->mmap_sem); 2024f55f199bSMike Frysinger 2025f55f199bSMike Frysinger return len; 2026f55f199bSMike Frysinger } 2027f55f199bSMike Frysinger 2028f55f199bSMike Frysinger /** 2029f55f199bSMike Frysinger * @access_remote_vm - access another process' address space 2030f55f199bSMike Frysinger * @mm: the mm_struct of the target address space 2031f55f199bSMike Frysinger * @addr: start address to access 2032f55f199bSMike Frysinger * @buf: source or destination buffer 2033f55f199bSMike Frysinger * @len: number of bytes to transfer 2034f55f199bSMike Frysinger * @write: whether the access is a write 2035f55f199bSMike Frysinger * 2036f55f199bSMike Frysinger * The caller must hold a reference on @mm. 2037f55f199bSMike Frysinger */ 2038f55f199bSMike Frysinger int access_remote_vm(struct mm_struct *mm, unsigned long addr, 2039f55f199bSMike Frysinger void *buf, int len, int write) 2040f55f199bSMike Frysinger { 2041f55f199bSMike Frysinger return __access_remote_vm(NULL, mm, addr, buf, len, write); 2042f55f199bSMike Frysinger } 2043f55f199bSMike Frysinger 2044f55f199bSMike Frysinger /* 2045f55f199bSMike Frysinger * Access another process' address space. 2046f55f199bSMike Frysinger * - source/target buffer must be kernel space 2047f55f199bSMike Frysinger */ 2048f55f199bSMike Frysinger int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) 2049f55f199bSMike Frysinger { 2050f55f199bSMike Frysinger struct mm_struct *mm; 2051f55f199bSMike Frysinger 2052f55f199bSMike Frysinger if (addr + len < addr) 2053f55f199bSMike Frysinger return 0; 2054f55f199bSMike Frysinger 2055f55f199bSMike Frysinger mm = get_task_mm(tsk); 2056f55f199bSMike Frysinger if (!mm) 2057f55f199bSMike Frysinger return 0; 2058f55f199bSMike Frysinger 2059f55f199bSMike Frysinger len = __access_remote_vm(tsk, mm, addr, buf, len, write); 2060f55f199bSMike Frysinger 20610ec76a11SDavid Howells mmput(mm); 20620ec76a11SDavid Howells return len; 20630ec76a11SDavid Howells } 20647e660872SDavid Howells 20657e660872SDavid Howells /** 20667e660872SDavid Howells * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode 20677e660872SDavid Howells * @inode: The inode to check 20687e660872SDavid Howells * @size: The current filesize of the inode 20697e660872SDavid Howells * @newsize: The proposed filesize of the inode 20707e660872SDavid Howells * 20717e660872SDavid Howells * Check the shared mappings on an inode on behalf of a shrinking truncate to 20727e660872SDavid Howells * make sure that that any outstanding VMAs aren't broken and then shrink the 20737e660872SDavid Howells * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't 20747e660872SDavid Howells * automatically grant mappings that are too large. 20757e660872SDavid Howells */ 20767e660872SDavid Howells int nommu_shrink_inode_mappings(struct inode *inode, size_t size, 20777e660872SDavid Howells size_t newsize) 20787e660872SDavid Howells { 20797e660872SDavid Howells struct vm_area_struct *vma; 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 */ 20916b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) { 20927e660872SDavid Howells /* found one - only interested if it's shared out of the page 20937e660872SDavid Howells * cache */ 20947e660872SDavid Howells if (vma->vm_flags & VM_SHARED) { 2095918e556eSDavid Howells mutex_unlock(&inode->i_mapping->i_mmap_mutex); 20967e660872SDavid Howells up_write(&nommu_region_sem); 20977e660872SDavid Howells return -ETXTBSY; /* not quite true, but near enough */ 20987e660872SDavid Howells } 20997e660872SDavid Howells } 21007e660872SDavid Howells 21017e660872SDavid Howells /* reduce any regions that overlap the dead zone - if in existence, 21027e660872SDavid Howells * these will be pointed to by VMAs that don't overlap the dead zone 21037e660872SDavid Howells * 21047e660872SDavid Howells * we don't check for any regions that start beyond the EOF as there 21057e660872SDavid Howells * shouldn't be any 21067e660872SDavid Howells */ 21076b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 21087e660872SDavid Howells 0, ULONG_MAX) { 21097e660872SDavid Howells if (!(vma->vm_flags & VM_SHARED)) 21107e660872SDavid Howells continue; 21117e660872SDavid Howells 21127e660872SDavid Howells region = vma->vm_region; 21137e660872SDavid Howells r_size = region->vm_top - region->vm_start; 21147e660872SDavid Howells r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; 21157e660872SDavid Howells 21167e660872SDavid Howells if (r_top > newsize) { 21177e660872SDavid Howells region->vm_top -= r_top - newsize; 21187e660872SDavid Howells if (region->vm_end > region->vm_top) 21197e660872SDavid Howells region->vm_end = region->vm_top; 21207e660872SDavid Howells } 21217e660872SDavid Howells } 21227e660872SDavid Howells 2123918e556eSDavid Howells mutex_unlock(&inode->i_mapping->i_mmap_mutex); 21247e660872SDavid Howells up_write(&nommu_region_sem); 21257e660872SDavid Howells return 0; 21267e660872SDavid Howells } 2127c9b1d098SAndrew Shewmaker 2128c9b1d098SAndrew Shewmaker /* 2129c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 2130c9b1d098SAndrew Shewmaker * 2131c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 2132c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 2133c9b1d098SAndrew Shewmaker * mode. 2134c9b1d098SAndrew Shewmaker * 2135c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 2136c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 2137c9b1d098SAndrew Shewmaker */ 2138c9b1d098SAndrew Shewmaker static int __meminit init_user_reserve(void) 2139c9b1d098SAndrew Shewmaker { 2140c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 2141c9b1d098SAndrew Shewmaker 2142c9b1d098SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 2143c9b1d098SAndrew Shewmaker 2144c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 2145c9b1d098SAndrew Shewmaker return 0; 2146c9b1d098SAndrew Shewmaker } 2147c9b1d098SAndrew Shewmaker module_init(init_user_reserve) 21484eeab4f5SAndrew Shewmaker 21494eeab4f5SAndrew Shewmaker /* 21504eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 21514eeab4f5SAndrew Shewmaker * 21524eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 21534eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 21544eeab4f5SAndrew Shewmaker * 21554eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 21564eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 21574eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 21584eeab4f5SAndrew Shewmaker */ 21594eeab4f5SAndrew Shewmaker static int __meminit init_admin_reserve(void) 21604eeab4f5SAndrew Shewmaker { 21614eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 21624eeab4f5SAndrew Shewmaker 21634eeab4f5SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 21644eeab4f5SAndrew Shewmaker 21654eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 21664eeab4f5SAndrew Shewmaker return 0; 21674eeab4f5SAndrew Shewmaker } 21684eeab4f5SAndrew Shewmaker module_init(init_admin_reserve) 2169